Abstract and Applied Analysis

Stability and Bifurcation

Analysis of Differential Equations
and Its Applications
Guest Editors: Yongli Song, Junling Ma, Yonghui Xia, Sanling Yuan,
and Tonghua Zhang
Stability and Bifurcation Analysis
of Differential Equations and Its Applications
Abstract and Applied Analysis

Stability and Bifurcation Analysis

of Differential Equations and Its Applications

Guest Editors: Yongli Song, Junling Ma, Yonghui Xia,

Sanling Yuan, and Tonghua Zhang
Copyright © 2015 Hindawi Publishing Corporation. All rights reserved.

This is a special issue published in “Abstract and Applied Analysis.” All articles are open access articles distributed under the Creative
Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original
work is properly cited.
Editorial Board
Ravi P. Agarwal, USA Juan-Carlos Cortés, Spain Chaitan P. Gupta, USA
Bashir Ahmad, Saudi Arabia Graziano Crasta, Italy Uno Hämarik, Estonia
M. O. Ahmedou, Germany Zhihua Cui, China Maoan Han, China
Nicholas D. Alikakos, Greece Bernard Dacorogna, Switzerland Ferenc Hartung, Hungary
Debora Amadori, Italy Vladimir Danilov, Russia Jiaxin Hu, China
Douglas R. Anderson, USA Mohammad T. Darvishi, Iran Chengming Huang, China
Jan Andres, Czech Republic L. F. Pinheiro de Castro, Portugal Zhongyi Huang, China
Giovanni Anello, Italy Toka Diagana, USA Gennaro Infante, Italy
Stanislav Antontsev, Portugal Jesús I. Dı́az, Spain Ivan Ivanov, Bulgaria
Mohamed K. Aouf, Egypt Josef Diblı́k, Czech Republic Hossein Jafari, South Africa
Narcisa C. Apreutesei, Romania Fasma Diele, Italy Jaan Janno, Estonia
Natig M. Atakishiyev, Mexico Tomas Dominguez, Spain Aref Jeribi, Tunisia
Ferhan M. Atici, USA Alexander Domoshnitsky, Israel Uncig Ji, Korea
Ivan Avramidi, USA Marco Donatelli, Italy Zhongxiao Jia, China
Soohyun Bae, Korea BoQing Dong, China Lucas Jódar, Spain
Chuanzhi Bai, China Wei-Shih Du, Taiwan Jong S. Jung, Republic of Korea
Zhanbing Bai, China Luiz Duarte, Brazil Henrik Kalisch, Norway
Dumitru Baleanu, Turkey Roman Dwilewicz, USA Chaudry Masood Khalique, South Africa
Józef Banaś, Poland Paul W. Eloe, USA Satyanad Kichenassamy, France
Martino Bardi, Italy Luca Esposito, Italy Tero Kilpeläinen, Finland
Roberto Barrio, Spain Khalil Ezzinbi, Morocco Sung G. Kim, Republic of Korea
Feyzi Başar, Turkey Julian F. Bonder, Argentina Ljubisa Kocinac, Serbia
Abdelghani Bellouquid, Morocco Dashan Fan, USA Andrei Korobeinikov, Spain
Daniele Bertaccini, Italy Angelo Favini, Italy Pekka Koskela, Finland
Lucio Boccardo, Italy Márcia Federson, Brazil Victor Kovtunenko, Austria
Igor Boglaev, New Zealand Stathis Filippas, Equatorial Guinea Ren-Jieh Kuo, Taiwan
Martin J. Bohner, USA Alberto Fiorenza, Italy Pavel Kurasov, Sweden
Geraldo Botelho, Brazil Xianlong Fu, China Milton C. L. Filho, Brazil
Elena Braverman, Canada Massimo Furi, Italy Miroslaw Lachowicz, Poland
Romeo Brunetti, Italy Jesús G. Falset, Spain Kunquan Lan, Canada
Janusz Brzdek, Poland Giovanni P. Galdi, USA Ruediger Landes, USA
Detlev Buchholz, Germany Isaac Garcia, Spain Irena Lasiecka, USA
Sun-Sig Byun, Korea J. A. G.a-Rodrı́guez, Spain Matti Lassas, Finland
Fabio M. Camilli, Italy Leszek Gasinski, Poland Chun-Kong Law, Taiwan
Jinde Cao, China György Gát, Hungary Ming-Yi Lee, Taiwan
Anna Capietto, Italy Vladimir Georgiev, Italy Gongbao Li, China
Jianqing Chen, China Lorenzo Giacomelli, Italy Elena Litsyn, Israel
Wing-Sum Cheung, Hong Kong Jaume Giné, Spain Yansheng Liu, China
Michel Chipot, Switzerland Valery Y. Glizer, Israel Shengqiang Liu, China
Changbum Chun, Korea Jean P. Gossez, Belgium Carlos Lizama, Chile
Soon-Yeong Chung, Korea Jose L. Gracia, Spain Guozhen Lu, USA
Jaeyoung Chung, Korea Maurizio Grasselli, Italy Jinhu Lü, China
Silvia Cingolani, Italy Luca Guerrini, Italy Grzegorz Lukaszewicz, Poland
Jean M. Combes, France Yuxia Guo, China Wanbiao Ma, China
Monica Conti, Italy Qian Guo, China Nazim I. Mahmudov, Turkey
Eberhard Malkowsky, Turkey Simeon Reich, Israel Csaba Varga, Romania
Salvatore A. Marano, Italy Abdelaziz Rhandi, Italy Carlos Vazquez, Spain
Cristina Marcelli, Italy Hassan Riahi, Malaysia Jesus Vigo-Aguiar, Spain
Paolo Marcellini, Italy Juan P. Rincón-Zapatero, Spain Qing-Wen Wang, China
J. Marı́n-Solano, Spain Luigi Rodino, Italy Yushun Wang, China
Mieczysław Mastyło, Poland Yuriy Rogovchenko, Norway Shawn X. Wang, Canada
Ming Mei, Canada Julio D. Rossi, Argentina Youyu Wang, China
Taras Mel’nyk, Ukraine Wolfgang Ruess, Germany Jing P. Wang, UK
Anna Mercaldo, Italy Bernhard Ruf, Italy Peixuan Weng, China
Stanislaw Migorski, Poland Satit Saejung, Thailand Noemi Wolanski, Argentina
Mihai Mihǎilescu, Romania Stefan G. Samko, Portugal Ngai-Ching Wong, Taiwan
Feliz Minhós, Portugal Martin Schechter, USA Patricia J. Y. Wong, Singapore
Dumitru Motreanu, France Javier Segura, Spain Yonghong Wu, Australia
Gaston Mandata N’guérékata, USA Valery Serov, Finland Zili Wu, China
Maria Grazia Naso, Italy Naseer Shahzad, Saudi Arabia Shi-Liang Wu, China
Micah Osilike, Nigeria Andrey Shishkov, Ukraine Shanhe Wu, China
Mitsuharu Ôtani, Japan Stefan Siegmund, Germany Tiecheng Xia, China
Turgut Öziş, Turkey Abdel-Maksoud A. Soliman, Egypt Xu Xian, China
Nikolaos S. Papageorgiou, Greece Pierpaolo Soravia, Italy Yanni Xiao, China
Sehie Park, Korea Marco Squassina, Italy Fuding Xie, China
Kailash C. Patidar, South Africa Hari M. Srivastava, Canada Gongnan Xie, China
Kevin R. Payne, Italy Svatoslav Staněk, Czech Republic Daoyi Xu, China
Ademir F. Pazoto, Brazil Antonio Suárez, Spain Zhenya Yan, China
Shuangjie Peng, China Wenchang Sun, China Xiaodong Yan, USA
Antonio M. Peralta, Spain Wenyu Sun, China Norio Yoshida, Japan
Sergei V. Pereverzyev, Austria Robert Szalai, UK Beong In Yun, Korea
Allan Peterson, USA Sanyi Tang, China Agacik Zafer, Kuwait
Andrew Pickering, Spain Chun-Lei Tang, China Jianming Zhan, China
Cristina Pignotti, Italy Gabriella Tarantello, Italy Weinian Zhang, China
Somyot Plubtieng, Thailand Nasser-Eddine Tatar, Saudi Arabia Chengjian Zhang, China
Milan Pokorny, Czech Republic Gerd Teschke, Germany Zengqin Zhao, China
Sergio Polidoro, Italy Sergey Tikhonov, Spain Sining Zheng, China
Ziemowit Popowicz, Poland Claudia Timofte, Romania Yong Zhou, China
Maria M. Porzio, Italy Thanh Tran, Australia Tianshou Zhou, China
Enrico Priola, Italy Juan J. Trujillo, Spain Qiji J. Zhu, USA
Vladimir S. Rabinovich, Mexico Gabriel Turinici, France Chun-Gang Zhu, China
Irena Rachu̇nková, Czech Republic Milan Tvrdy, Czech Republic Malisa R. Zizovic, Serbia
Maria A. Ragusa, Italy Mehmet Ünal, Turkey Wenming Zou, China
Contents
Stability and Bifurcation Analysis of Differential Equations and Its Applications, Yongli Song,
Junling Ma, Yonghui Xia, Sanling Yuan, and Tonghua Zhang
Volume 2015, Article ID 343528, 1 pages

Bifurcation of Traveling Wave Solutions for (2+1)-Dimensional Nonlinear Models Generated by the
Jaulent-Miodek Hierarchy, Yanping Ran, Jing Li, Xin Li, and Zheng Tian
Volume 2015, Article ID 820916, 14 pages
Hopf Bifurcation, Cascade of Period-Doubling, Chaos, and the Possibility of Cure in a 3D Cancer
Model, Marluci Cristina Galindo, Cristiane Nespoli, and Marcelo Messias
Volume 2015, Article ID 354918, 11 pages

Synchronization Transition and Traffic Congestion in One-Dimensional Traffic Model, Zhi Xin
and Jian Xu
Volume 2015, Article ID 167430, 10 pages

The Dynamics of an Impulsive Predator-Prey System with Stage Structure and Holling Type III
Functional Response, Zhixiang Ju, Yuanfu Shao, Xiaolan Xie, Xiangmin Ma, and Xianjia Fang
Volume 2015, Article ID 183526, 8 pages

Skew Circulant Type Matrices Involving the Sum of Fibonacci and Lucas Numbers, Zhaolin Jiang
and Yunlan Wei
Volume 2015, Article ID 951340, 9 pages

A Note on the Existence of a Smale Horseshoe in the Planar Circular Restricted Three-Body Problem,
Xuhua Cheng and Zhikun She
Volume 2015, Article ID 965829, 8 pages

Effect of Leakage Delay on Stability of Neutral-Type Genetic Regulatory Networks, Li Li, Yongqing Yang,
and Chuanzhi Bai
Volume 2015, Article ID 826020, 8 pages

Dynamic Analysis of a Delayed Reaction-Diffusion Predator-Prey System with Modified

Holling-Tanner Functional Response, Xinhong Pan, Min Zhao, Chuanjun Dai, and Yapei Wang
Volume 2015, Article ID 620891, 12 pages

The Stochastic Θ-Method for Nonlinear Stochastic Volterra Integro-Differential Equations, Peng Hu
and Chengming Huang
Volume 2014, Article ID 583930, 13 pages
On the Number of Limit Cycles of a Piecewise Quadratic Near-Hamiltonian System, Jing Tian
Volume 2014, Article ID 385103, 10 pages

Normal Form for High-Dimensional Nonlinear System and Its Application to a Viscoelastic Moving
Belt, S. P. Chen and Y. H. Qian
Volume 2014, Article ID 879564, 11 pages

Periodic Solutions of Multispecies Mutualism System with Infinite Delays, Wenbo Zhao, Caocuan Ma,
Taotao Zheng, and Xiao-Ke Sun
Volume 2014, Article ID 127876, 5 pages
Global and Blow-Up Solutions for a Class of Nonlinear Parabolic Problems under Robin Boundary
Condition, Lingling Zhang and Hui Wang
Volume 2014, Article ID 241650, 9 pages

Bifurcation Analysis of a Lotka-Volterra Mutualistic System with Multiple Delays, Xin-You Meng
and Hai-Feng Huo
Volume 2014, Article ID 958140, 18 pages
Stability and Bifurcation Analysis on an Ecoepidemiological Model with Stage Structure and Time
Delay, Lingshu Wang and Guanghui Feng
Volume 2014, Article ID 727818, 10 pages

Permanence of Diffusive Models for Three Competing Species in Heterogeneous Environments,

Benlong Xu and Zhenzhang Ni
Volume 2014, Article ID 376919, 10 pages

Global Dynamics of an HTLV-1 Model with Cell-to-Cell Infection and Mitosis, Sumei Li and Yicang Zhou
Volume 2014, Article ID 132781, 12 pages

Bifurcation of Traveling Wave Solutions of the Dual Ito Equation, Xinghua Fan and Shasha Li
Volume 2014, Article ID 153139, 9 pages

Improved Stability Criteria for Markovian Jump Systems with Time-Varying Delays, Yu-cai Ding,
Hui Liu, and Baodan Tian
Volume 2014, Article ID 248061, 9 pages

Dynamics of an Almost Periodic Food Chain System with Impulsive Effects, Yaqin Li, Wenquan Wu,
and Tianwei Zhang
Volume 2014, Article ID 324912, 10 pages

Limit Cycles in a Cubic Kolmogorov System with Harvest and Two Positive Equilibrium Points,
Qi-Ming Zhang, Feng Li, and Yulin Zhao
Volume 2014, Article ID 786962, 6 pages

Limit Cycle Bifurcations by Perturbing a Compound Loop with a Cusp and a Nilpotent Saddle,
Huanhuan Tian and Maoan Han
Volume 2014, Article ID 819798, 14 pages

On Regularity Criteria for the Two-Dimensional Generalized Liquid Crystal Model, Yanan Wang
and Zaihong Jiang
Volume 2014, Article ID 319585, 8 pages

Stability Switches and Hopf Bifurcation in a Coupled FitzHugh-Nagumo Neural System with Multiple
Delays, Shengwei Yao and Huonian Tu
Volume 2014, Article ID 874701, 13 pages

Nonlinear Instability for a Volume-Filling Chemotaxis Model with Logistic Growth, Haiyan Gao
and Shengmao Fu
Volume 2014, Article ID 248657, 11 pages
A Note on Gronwall’s Inequality on Time Scales, Xueru Lin
Volume 2014, Article ID 623726, 4 pages
Stabilization and Synchronization of Unified Chaotic System via Impulsive Control, Cheng Hu
and Haijun Jiang
Volume 2014, Article ID 369842, 8 pages

Global Stability of a Computer Virus Propagation Model with Two Kinds of Generic Nonlinear
Probabilities, Chenquan Gan, Xiaofan Yang, and Qingyi Zhu
Volume 2014, Article ID 735327, 7 pages
Traveling Wave Solutions of the Kadomtsev-Petviashvili-Benjamin-Bona-Mahony Equation,
Zhengyong Ouyang
Volume 2014, Article ID 943167, 9 pages

Dynamical Behaviors of Rumor Spreading Model with Control Measures, Xia-Xia Zhao
and Jian-Zhong Wang
Volume 2014, Article ID 247359, 11 pages
Stability Analysis of a Class of Higher Order Difference Equations, Yuanyuan Liu and Fanwei Meng
Volume 2014, Article ID 434621, 7 pages

Existence of Positive Solutions for a Kind of Fractional Boundary Value Problems, Hongjie Liu, Xiao Fu,
and Liangping Qi
Volume 2014, Article ID 602604, 8 pages
Travelling Waves of an n-Species Food Chain Model with Spatial Diffusion and Time Delays, Fei Hu,
Yuyin Xu, Z. Wang, and Wei Ding
Volume 2014, Article ID 613648, 7 pages

Quasiperiodic Solutions of Completely Resonant Wave Equations with Quasiperiodic Forced Terms,
Yixian Gao, Weipeng Zhang, and Jing Chang
Volume 2014, Article ID 649270, 15 pages

Periodic Solutions of a Stage-Structured Plant-Hare Model with Toxin-Determined Functional

Responses, Huicheng Wang, Patricia J. Y. Wong, and Yonghui Xia
Volume 2014, Article ID 357534, 9 pages
Stability Conditions of Second Order Integrodifferential Equations with Variable Delay, Dingheng Pi
Volume 2014, Article ID 371639, 11 pages

Asymptotically Almost Periodic Solutions for a Class of Stochastic Functional Differential Equations,
Aimin Liu, Yongjian Liu, and Qun Liu
Volume 2014, Article ID 934534, 11 pages

A Stochastic Weakly Damped, Forced KdV-BO Equation, Guolian Wang

Volume 2014, Article ID 576087, 8 pages
Global Analysis of a Virus Dynamics Model with General Incidence Function and Cure Rate, Yu Yang
Volume 2014, Article ID 726349, 7 pages
The Problem of Bicenter and Isochronicity for a Class of Quasi Symmetric Planar Systems,
Du Chaoxiong
Volume 2014, Article ID 482450, 11 pages

Hopf Bifurcation and Stability Analysis of a Congestion Control Model with Delay in Wireless Access
Network, Wen-bo Zhao, Xiao-ke Sun, and Huicheng Wang
Volume 2014, Article ID 632564, 12 pages
Dynamics of an Information Spreading Model with Isolation, Xia-Xia Zhao and Jian-Zhong Wang
Volume 2014, Article ID 484630, 6 pages

Modelling the Influence of Awareness Programs by Media on the Drinking Dynamics, Hai-Feng Huo
and Qian Wang
Volume 2014, Article ID 938080, 8 pages

Existence and Stability of Periodic Solution to Delayed Nonlinear Differential Equations, Xiang Gu,
Huicheng Wang, P. J. Y. Wong, and Yonghui Xia
Volume 2014, Article ID 156948, 12 pages

Stability and Hopf Bifurcation in a Delayed Predator-Prey System with Herd Behavior, Chaoqun Xu
and Sanling Yuan
Volume 2014, Article ID 568943, 8 pages

Existence of Wave Front Solutions of an Integral Differential Equation in Nonlinear Nonlocal Neuronal
Network, Lijun Zhang, Linghai Zhang, Jie Yuan, and C. M. Khalique
Volume 2014, Article ID 753614, 9 pages

Travelling Wave Solutions of Nonlinear Dynamical Equations in a Double-Chain Model of DNA,

Zheng-yong Ouyang and Shan Zheng
Volume 2014, Article ID 317543, 5 pages

Global Attractivity of an Integrodifferential Model of Mutualism, Xiangdong Xie, Fengde Chen,

Kun Yang, and Yalong Xue
Volume 2014, Article ID 928726, 6 pages
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2015, Article ID 343528, 1 page
http://dx.doi.org/10.1155/2015/343528

Editorial
Stability and Bifurcation Analysis of Differential Equations and
Its Applications

Yongli Song,1 Junling Ma,2 Yonghui Xia,3 Sanling Yuan,4 and Tonghua Zhang5
1
Department of Mathematics, Tongji University, Shanghai 200092, China
2
Department of Mathematics and Statistics, University of Victoria, Victoria, BC, Canada V8W 3R4
3
Department of Mathematics, Zhejiang Normal University, Jinhua 321004, China
4
College of Sciences, University of Shanghai for Science and Technology, Shanghai 20093, China
5
Department of Mathematics, Swinburne University of Technology, Melbourne, VIC 3122, Australia

Correspondence should be addressed to Yongli Song; 05143@tongji.edu.cn

Received 9 December 2014; Accepted 9 December 2014

Copyright © 2015 Yongli Song et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Starting from Poincaré’s qualitative theory and Lyapunov’s a single special issue. But we believe that the papers to be
stability theory of a dynamical system, stability and bifurca- published in this special issue can at least partially reflect
tion theory has undergone a prodigious development. Stabil- some new advances and ideas in the field and do hope this
ity and bifurcation theory of differential equations is relatively special issue can influence the research field of bifurcation
a mature research area, yet it has seen rapid developments in theory of differential equations in future.
recent years. These advances have led to broad applications
in many fields, such as physics, engineering, biology, neuro- Acknowledgments
science, economics, and even life and social sciences.
It is well known that delay is typically a primary source of The guest editors of this special issue would like to take this
oscillatory behaviour in delay differential equations and dif- opportunity to thank all contributors for submitting their
fusion often causes Turing instability and becomes a primary excellent work to this issue and all reviewers for their hard
source of spatial dynamics in reaction-diffusion equations. work and academic support to this special issue. They would
Therefore, we have targeted these topics in this special issue. also like to thank the editorial board members of this journal
The special issue received tremendous response from the for their technical support and help during the whole period.
researchers in this research field. So far, we have received 124
papers, which contribute to the research field with the infu- Yongli Song
sion of new ideas and methods. All papers submitted to this Junling Ma
special issue went through a rigorous peer-review process. Yonghui Xia
Based on the reviewers’ reports, we have carefully selected Sanling Yuan
48 original research papers for publication, which contain Tonghua Zhang
the delay-induced instability, stability switches, and Hopf
bifurcations in delay differential equations; nonlinear insta-
bility, bifurcations, and blow-up solutions and travelling wave
solutions in the reaction-diffusion equations; and almost
periodic solutions in the stochastic differential equations.
It is impossible to collect all recently important advances
in the field of bifurcation theory of differential equations by
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2015, Article ID 820916, 14 pages
http://dx.doi.org/10.1155/2015/820916

Research Article
Bifurcation of Traveling Wave Solutions for (2+1)-Dimensional
Nonlinear Models Generated by the Jaulent-Miodek Hierarchy

Yanping Ran,1,2 Jing Li,1 Xin Li,1 and Zheng Tian1

1
College of Applied Science, Beijing University of Technology, Beijing 100124, China
2
School of Mathematics and Statistics, Tianshui Normal University, Tianshui, Gansu 741001, China

Correspondence should be addressed to Jing Li; leejing@bjut.edu.cn

Received 27 June 2014; Accepted 15 July 2014

Academic Editor: Yonghui Xia

Copyright © 2015 Yanping Ran et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Four (2+1)-dimensional nonlinear evolution equations, generated by the Jaulent-Miodek hierarchy, are investigated by the
bifurcation method of planar dynamical systems. The bifurcation regions in different subsets of the parameters space are obtained.
According to the different phase portraits in different regions, we obtain kink (antikink) wave solutions, solitary wave solutions,
and periodic wave solutions for the third of these models by dynamical system method. Furthermore, the explicit exact expressions
of these bounded traveling waves are obtained. All these wave solutions obtained are characterized by distinct physical structures.

1. Introduction Four (2+1)-dimensional nonlinear models generated by

the Jaulent-Miodek hierarchy [3] are given by
In [1–4], four (2+1)-dimensional nonlinear models gen-
3
erated by the Jaulent-Miodek hierarchy were developed. 𝑤𝑡 = − (𝑤𝑥𝑥 − 2𝑤3 )𝑥 − (𝑤 𝜕−1 𝑤 + 𝑤𝑤𝑦 ) ,
These nonlinear models are completely integrable evolution 2 𝑥 𝑥 𝑦
equations. There are many approaches to investigate nonlin- 1 3 1
ear evolution equation, for example, the inverse scattering 𝑤𝑡 = (𝑤 − 2𝑤3 )𝑥 − (− 𝜕𝑥−1 𝑤𝑦𝑦 + 𝑤𝑤𝑦 ) ,
2 𝑥𝑥 2 4
method, the Bäcklund transformation method, the Darboux
transformation method, the Hirota bilinear method [1–3, 5– 1 3 1
𝑤𝑡 = (𝑤 − 2𝑤3 )𝑥 − ( 𝜕𝑥−1 𝑤𝑦𝑦 + 𝑤𝑥 𝜕𝑥−1 𝑤𝑦 ) ,
8], and the dynamical systems method [9–11]. The Hirota 4 𝑥𝑥 4 4
bilinear method [3] is used to formally derive the multiple 3 −1
kink solutions and multiple singular kink solutions of these 𝑤𝑡 = 2 (𝑤𝑥𝑥 − 2𝑤3 )𝑥 − (𝜕 𝑤 − 2𝑤𝑥 𝜕𝑥−1 𝑤𝑦 − 6𝑤𝑤𝑦 ) ,
models. By applying the direct symmetry method [4], group 4 𝑥 𝑦𝑦
(1)
invariant solutions and some new exact solutions of the
(2+1)-dimensional Jaulent-Miodek equation are obtained. where 𝜕𝑥−1 is the inverse of 𝜕𝑥 with 𝜕𝑥 𝜕𝑥−1 = 𝜕𝑥−1 𝜕𝑥 = 1 and
Dynamical systems method is a very effective method to 𝑥
research qualitative behavior for traveling wave solutions of 𝜕𝑥−1 = ∫ 𝑓 (𝑡) 𝑑𝑡. (2)
these completely integrable evolution equations. In [11], only −∞
considering bifurcation parametric 𝑐, some exact traveling We will study the third model given by
wave solutions are given by applying the method of dynamical
systems for these models. In this paper, all wave solutions 1 3 1
𝑤𝑡 = (𝑤𝑥𝑥 − 2𝑤3 )𝑥 − ( 𝜕𝑥−1 𝑤𝑦𝑦 + 𝑤𝑥 𝜕𝑥−1 𝑤𝑦 ) . (3)
are given by the method of dynamical systems under more 4 4 4
general parametric conditions. Some computer symbolic By introducing the potential
systems such as Maple and Mathmatic allow us to perform
complicated and tedious calculations. 𝑤 (𝑥, 𝑦, 𝑡) = 𝑢𝑥 (𝑥, 𝑦, 𝑡) , (4)
2 Abstract and Applied Analysis

to remove the integral term in the system (3), we obtain the Obviously, the above system (11) is a Hamiltonian system with
following equation Hamiltonian function
1 2 3 3
𝑢𝑥𝑡 + 𝑢𝑥𝑥𝑥𝑥 − 𝑢𝑥2 𝑢𝑥𝑥 + 𝑢𝑦𝑦 + 𝑢𝑥𝑥 𝑢𝑦 = 0. (5) 2
𝑦2 (3𝑟 − 16𝑘𝑐) 2 𝑟 3 𝜑4
4 3 16 4 𝐻 (𝜑, 𝑦) = + 𝜑 + 𝜑 − . (12)
2 8𝑘4 6𝑘2 2
We are interested in the wave solutions of the system (3) in
this paper. Motivated by [9], we obtain dynamical properties
of (11) and different wave solutions of the system (3) in In order to research the system (11), let 𝑎 = −(3𝑟2 −
detail. This paper is organized as follows. In Section 2, we 16𝑘𝑐)/4𝑘4 , 𝑏 = −3𝑟/2𝑘2 ; the system (11) becomes
establish the traveling wave equation (3) for the third model
of (1). Furthermore, we obtain the first integral of dynamical 𝑑𝜑
governing equation of the system (11). Then, we analyze the = 𝑦,
𝑑𝜉
bifurcation behaviors of the system (11). Phase portraits in (13)
the different subsets of parameter space will be presented in 𝑑𝑦
= 𝑎𝜑 + 𝑏𝜑2 + 2𝜑3 .
Section 3. In Section 4, using the information of the phase 𝑑𝜉
portraits in Section 3, we analyze all the possible traveling
wave solutions of the system (11). Some explicit parametric The Hamiltonian function of (13) is
representations of traveling wave solutions of (3) and the
system (11) are also obtained. The final section includes brief 1 𝑎 1 1
summary, future plans, and potential fields of applications. 𝐻 (𝜑, 𝑦) = 𝑦2 − 𝜑2 − 𝑏𝜑3 − 𝜑4 . (14)
2 2 3 2

2. Traveling Wave Equation for the System (3) 3. The Bifurcation Analysis of the System (11)
We assume that the traveling wave transform of the system
In this section, our aim is to study the traveling wave solu-
(3) is in the form
tions of the system (11) by applying bifurcation method and
𝑢 (𝑥, 𝑦, 𝑡) = Ψ (𝜉) , qualitative theory of dynamical systems [9, 10]. Through
(6) some special phase orbits, we obtain smooth periodic wave
𝜉 = 𝑘𝑥 + 𝑟𝑦 − 𝑐𝑡, solutions, solitary wave solutions, kink and antikink wave
solutions, and so on. Fixing 𝑐, we discuss the phase portrait
where 𝑐 is propagating wave velocity. Let 𝑘 = 𝑟, 𝑐 = 1, the of the system (11) along with the changes of parameters 𝑟
traveling wave transform of (6) is equivalent to 𝜉 = 𝑥 + and 𝑘 so as to study traveling wave solutions of the system
𝑦 − 𝑐𝑡 [11]. So, our traveling wave transform is more general. (11). Further more, through the traveling wave solutions of
According to physical meaning of traveling wave solutions of the system (11) and the potential relation (4), traveling wave
the system (3), we always assume that 𝑐 > 0, 𝑘 ≠ 0, and solutions of the system (3) will be obtained.
𝑟 ≠0. Now, substituting (6) into (5), we have the traveling
wave equation
3.1. Phase Portraits and Qualitative Analysis of the System (11).
𝑘4 3 3 3 In order to investigate the phase portrait of the system (11),
−𝑘𝑐𝑢𝜉𝜉 + 𝑢𝜉𝜉𝜉𝜉 − 𝑘4 𝑢𝜉2 𝑢𝜉𝜉 + 𝑟2 𝑢𝜉𝜉 + 𝑘2 𝑟𝑢𝜉𝜉 𝑢𝜉 = 0.
4 2 16 4 we set
(7)

Integrating (7) with respect to 𝜉 once, we have 𝑓 (𝜑) = 𝑎𝜑 + 𝑏𝜑2 + 2𝜑3 = 𝜑 (𝑎 + 𝑏𝜑 + 2𝜑2 ) . (15)

−16𝑘𝑐𝑢𝜉 + 4𝑘4 𝑢𝜉𝜉𝜉 − 8𝑘4 𝑢𝜉3 + 3𝑟2 𝑢𝜉 + 6𝑘2 𝑟𝑢𝜉2 = 0. (8) Let Δ = (33𝑟2 − 128𝑘𝑐)/4𝑘4 . Obviously, 𝑓(𝜑) has at
Setting 𝑢𝜉 = 𝜑, (8) becomes least one zero point (𝜑0 , 𝑓(𝜑0 )) = (0, 0). The number of the
singular points of the system (11) may be decided by the sign
−16𝑘𝑐𝜑 + 4𝑘4 𝜑𝜉𝜉 − 8𝑘4 𝜑3 + 3𝑟2 𝜑 + 6𝑘2 𝑟𝜑2 = 0. (9) of Δ. Obviously, the system (11) has only one trivial singular
point (0, 0). Thus the other singular points of the system (11)
Furthermore, (8) can be rewritten as are given as follows. (1) When Δ < 0, the system (11) has
only one trivial singular point (0, 0); (2) when Δ > 0, the
4𝑘4 𝜑𝜉𝜉 + (3𝑟2 − 16𝑘𝑐) 𝜑 + 6𝑘2 𝑟𝜑2 − 8𝑘4 𝜑3 = 0. (10) system (11) has two singular points (𝜑1,2 , 0), where 𝜑1 = (3𝑟 +
2𝑘2 √Δ)/8𝑘2 > 𝜑2 = (3𝑟 − 2𝑘2 √Δ)/8𝑘2 ; (3) when Δ = 0,
Letting 𝜑󸀠 = 𝑦, then we have the following planar system the system has a second-order singular point (𝜑3 , 0), where
𝑑𝜑 𝜑3 = 𝜑1 = 𝜑2 .
= 𝑦, We notice that the Jacobian of linearized system of the
𝑑𝜉
(11) system (11) at the singular points is given by
𝑑𝑦 (3𝑟2 − 16𝑘𝑐) 3𝑟
=− 𝜑 − 2 𝜑2 + 2𝜑3 .
𝑑𝜉 4𝑘4 2𝑘 𝐽 (𝜑𝑖 , 0) = −𝑓󸀠 (𝜑𝑖 ) , (𝑖 = 0, 1, 2, 3) . (16)
Abstract and Applied Analysis 3

Thus, the characteristic values of linearized system of the Proof. When 𝑟 = ±2√𝑘𝑐, we have 𝜑0,2 = −√𝑘𝑐/𝑘2 < 0 and
system (11) at (𝜑𝑖 , 0) are 𝜆 = ±√𝑓󸀠 (𝜑𝑖 ). From the qualitative 𝑓󸀠 (−√𝑘𝑐/𝑘2 ) < 0. According to the qualitative theory of
theory of dynamical system, we know that dynamical system, (𝜑0,2 , 0) are saddle points. Furthermore,
when 𝑟 = ±2√𝑘𝑐, −𝑓(𝜑1∗ ) = 𝑓(𝜑2∗ ) holds. Similarly, if 𝑟 ∈
(i) if 𝑓󸀠 (𝜑𝑖 ) > 0, (𝜑𝑖 , 0) is a saddle point;
(−4√3𝑘𝑐/3, −2√𝑘𝑐) ∪ (2√𝑘𝑐, 4√3𝑘𝑐/3), we have that (𝜑1,2 , 0)
(ii) if 𝑓󸀠 (𝜑𝑖 ) < 0, (𝜑𝑖 , 0) is a center point; is the saddle point and −𝑓(𝜑1∗ ) ≠ 𝑓(𝜑2∗ ) holds. Applying
(iii) if 𝑓󸀠 (𝜑𝑖 ) = 0, (𝜑𝑖 , 0) is a degenerate saddle point. Theorems 1 and 2 [12], Theorem 1 is proved.
Let In order to give the details of the bifurcation, if 𝑐 > 0,
𝑘 > 0, we can obtain the following six bifurcation boundaries:
𝐻 (𝜑, 𝑦) = ℎ, (17)
4√3𝑘𝑐
where ℎ is Hamiltonian value. When 𝑟2 ≥ 4𝑘𝑐, 𝐿1 : 𝑟 = − ,
3
𝑎 2 𝑏 3 1 4
𝐻 (𝜑, 0) = 𝜑 + 𝜑 + 𝜑 =0 (18) 𝐿 2 : 𝑟 = −2√𝑘𝑐,
2 3 2
has four real roots. 8√66𝑘𝑐
It is well known that the planar Hamiltonian system is 𝐿3 : 𝑟 = − ,
33
determined by its potential energy level curve and its singular (23)
point in the form of (𝜑∗ , 0). So, we are interested in looking 8√66𝑘𝑐
𝐿4 : 𝑟 = ,
for the possible zeros of (15) and determining whether there 33
are heteroclinic orbits, homoclinic orbits, periodic orbits at
different singular points. 𝐿 5 : 𝑟 = 2√𝑘𝑐,
In order to find the heteroclinic orbits and the homoclinic
orbits of the system (11), let 4√3𝑘𝑐
𝐿6 : 𝑟 = .
3
𝑓󸀠 (𝜑) = 𝑎 + 2𝑏𝜑 + 6𝜑2 = 0. (19)
All these bifurcation boundaries divide the parameter
From (19), we can get the following expressions of its roots: space into seven regions (see Figure 1(a)) in which different
phase portraits exist. All the corresponding phase portraits
3𝑟 + √27𝑟2 − 96𝑘𝑐 will be shown in Figure 2.
𝜑1∗ = ,
12𝑘2 If 𝑐 > 0, 𝑘 < 0, there is one bifurcation boundary:
(20)
3𝑟 − √27𝑟2 − 96𝑘𝑐 𝐿 7 : 𝑟 = 0. (24)
𝜑2∗ = .
12𝑘2
In this case, the corresponding phase portraits in two
Substituting (20) into (15), we can get bifurcation regions 𝐹7 and 𝐹8 (see Figure 1(b)) will be shown
in Figure 3.
𝑓 (𝜑1∗ ) = − ((3𝑟 + √27𝑟2 − 96𝑘𝑐) Assuming that the following conditions hold:
−1 𝑘 > 0, 𝑐 > 0, Δ ≥ 0. (𝐼)
× (15𝑟2 − 64𝑘𝑐 + 𝑟√27𝑟2 − 96𝑘𝑐)) (288𝑘6 ) ,
Therefore, we can obtain the phase portraits of the system (11)
𝑓 (𝜑2∗ ) = − ((−3𝑟 + √27𝑟2 − 96𝑘𝑐) in Figure 2.
Set
−1
× (−15𝑟2 − 64𝑘𝑐 + 𝑟√27𝑟2 − 96𝑘𝑐)) (288𝑘6 ) , ℎ𝑖 = 𝐻 (𝜑𝑖 , 𝑦) . (25)
(21)
According to Figure 2, we obtain Case 1 as follows.
𝑟 (−𝑟2 + 4𝑘𝑐)
−𝑓 (𝜑1∗ ) − 𝑓 (𝜑2∗ ) = − . (22) Case 1. Suppose that 𝑘 > 0, 𝑐 > 0, and Δ ≥ 0, in addition
2𝑘6 to one of conditions (1)–(12), we can obtain the sign of 𝑓󸀠 (𝜑𝑖 )
Theorem 1. When 𝑘 > 0, 𝑐 > 0, from (22), one has the and the relation among ℎ(𝜑𝑖 , 0) by choosing suitable 𝑟, 𝑘, and
following. 𝑐, respectively.
(i) When 𝑟 = ±2√𝑘𝑐, there are two heteroclinic orbits (1) When 𝑟 < −4√3𝑘𝑐/3, the fact is that ℎ0 < ℎ1 < ℎ2
formed by the saddle points (±2√𝑘𝑐, 0). exists and the system (11) has two saddles at (𝜑1,2 , 0) and
(ii) When 𝑟 ∈ (−4√3𝑘𝑐/3, −2√𝑘𝑐) ∪ (2√𝑘𝑐, 4√3𝑘𝑐/3), a center at (0, 0) determined by (16). When ℎ ∈ (ℎ0 , ℎ1 ),
there are no heteroclinic orbits, while there are homo- the system (11) has a family of periodic orbits in which the
clinic orbits formed by other saddle points except for periodic orbit Γ1 is included (see Figure 2(a)). When ℎ ∈
two saddle points (±2√𝑘𝑐, 0). (−∞, ℎ0 ) ∪ (ℎ2 , ∞), periodic orbits become open curves.
4 Abstract and Applied Analysis

15
−1 −0.5 0 0.5 1
10
−1

−2
0
1 2 3 4 5
k −3
−5

−10 −4

−15
−5
L2 L6 L7
L5 L3
L1 L4

(a) 𝑘 > 0, 𝑐 > 0, Δ ≥ 0 (b) 𝑘 < 0, 𝑐 > 0

Figure 1: Transition boundaries on (𝑘 − 𝑟) plane of system (10).

(2) When 𝑟 = −4√3𝑘𝑐/3, the coefficient of 𝜑 vanishes. in the system (11) (see Figure 2(g)). The system (11) has a
Both singular points (𝜑0,1 , 0) are degenerated to (𝜑0 , 0) and family of open curves.
(𝜑2 , 0) becoming a saddle point in the system (11) (see (8) When 8√66𝑘𝑐/33 < 𝑟 < 2√𝑘𝑐, the system (11) has two
Figure 2(b)). In the system (11), all the level curves are saddles at (𝜑0,1 , 0) and a center at (𝜑2 , 0). The system (11) has
open. a family of open curves (see Figure 2(h)).
(3) When −4√3𝑘𝑐/3 < 𝑟 < −2√𝑘𝑐, the fact is that ℎ1 < (9) When 𝑟 = 2√𝑘𝑐, the system (11) has two saddles at
ℎ0 < ℎ2 exists and the system (11) has two saddles at (𝜑0,2 , 0)
(𝜑0,2 , 0) and a center at (𝜑1 , 0), where 𝜑2 = 2𝜑1 = −√𝑘𝑐/𝑘2 ,
and a center at (𝜑1 , 0). When ℎ = ℎ0 , the system (11) has
homoclinics orbit Γ4 and a special orbit Γ5 (see Figure 2(c)). ℎ2 = −9𝑐2 /4𝑘4 < ℎ0 = ℎ1 . When ℎ = ℎ0 , the system (11) has
When ℎ ∈ (ℎ1 , ℎ0 ), the system (11) has a family of periodic heteroclinic orbits consisting of Γ13 and Γ14 , which connects
orbits in which the periodic orbit Γ6 is included. When ℎ → two saddles (𝜑0,2 , 0). When ℎ ∈ (ℎ1 , ℎ0 ), the system (11) has
ℎ0− , periodic orbits become the homoclinic orbit Γ4 . When a family of periodic orbits in which the periodic orbit Γ15 is
ℎ ∈ (−∞, ℎ1 ), periodic orbits become open curves. included (see Figure 2(i)). But when ℎ → ℎ0+ , periodic orbits
become the heteroclinic orbit Γ13 and the orbit Γ14 .
(4) When 𝑟 = −2√𝑘𝑐, the system (11) has two saddles at
(𝜑0,2 , 0) and a center at (𝜑1 , 0), where 𝜑2 = 2𝜑1 = −√𝑘𝑐/𝑘2 , (10) When 2√𝑘𝑐 < 𝑟 < 4√3𝑘𝑐/3, the system (11) has two
saddles at (𝜑0,1 , 0) and a center at (𝜑2 , 0) and ℎ2 < ℎ0 < ℎ1
ℎ1 = −9𝑐2 /4𝑘4 < ℎ0 = ℎ2 . When ℎ = ℎ0 , the system (11) has exists. When ℎ = ℎ0 , the system (11) has homoclinics orbit
heteroclinic orbits consisting of Γ10 and Γ11 , which connects Γ7 and a special orbit Γ8 . When ℎ ∈ (ℎ0 , ℎ1 ), the system (11)
two saddles (𝜑0,2 , 0) (see Figure 2(d)). When ℎ ∈ (ℎ1 , ℎ0 ), has a family of periodic orbits in which the periodic orbit Γ9
the system (11) has a family of periodic orbits in which the is included (see Figure 2(j)). When ℎ → ℎ0+ , periodic orbits
periodic orbit Γ12 is included. But when ℎ → ℎ0− , periodic become the homoclinics orbit Γ7 . When ℎ ∈ (ℎ1 , ∞), periodic
orbits become the heteroclinic orbit Γ10 and the orbit Γ11 . orbits become open curves.
(5) When −2√𝑘𝑐 < 𝑟 < −8√66𝑘𝑐/33, we can obtain ℎ2 < (11) When 𝑟 = 4√3𝑘𝑐/3, the coefficient of 𝜑 vanishes.
ℎ1 < ℎ0 and the system (11) has two saddles at (𝜑0,2 , 0) and a Both singular points (𝜑0,2 , 0) are degenerated to (𝜑0 , 0) and
center at (𝜑1 , 0) (see Figure 2(e)). The system (11) has a family (𝜑1 , 0) becoming a saddle in the system (11) (see Figure 2(k)).
of open curves. In the system (11), all the level curves are open.
(6) When 𝑟 = −8√66𝑘𝑐/33, both singular points (𝜑1,2 , 0) (12) When 𝑟 > 4√3𝑘𝑐/3, the system (11) has two saddles
are degenerated to (−2√66𝑘𝑐/11, 0); (𝜑0 , 0) becomes a saddle at (𝜑1,2 , 0) and a center at (0, 0) and ℎ0 < ℎ2 < ℎ1 exists.
in the system (11) (see Figure 2(f)). The system (11) has a When ℎ ∈ (ℎ0 , ℎ2 ), the system (11) has a family of periodic
family of open curves. orbits in which the periodic orbit Γ16 is included. When ℎ ∈
(7) When 𝑟 = 8√66𝑘𝑐/33, both singular points (𝜑1,2 , 0) (−∞, ℎ0 ) ∪ (ℎ1 , ∞), periodic orbits become open curves (see
are degenerated to (2√66𝑘𝑐/11, 0), (𝜑0 , 0) becomes a saddle Figure 2(l)).
Abstract and Applied Analysis 5

2 8
6 4
y 1 y 4
y
2
2
0 0
0
−6 −5 −4 −3 −2 −1 0 1 −5 −4 −3 −2 −1 0 1 2
−2 −4 −3 −2 −1 0 1
𝜑 𝜑
𝜑 −2
−1 −4
−6
−4
−2 −8
(a) 𝑟 < −4√3𝑘𝑐/3 (b) 𝑟 = −4√3𝑘𝑐/3 (c) −4√3𝑘𝑐/3 < 𝑟 < −2√𝑘𝑐

2 2 2

y 1 y 1 y 1

0 0 0
−2.5 −2 −1.5 −1 −0.5 0 0.5 −2.5 −2 −1.5 −1 −0.5 0 0.5 −2.5 −2 −1.5 −1 −0.5 0 0.5
𝜑 𝜑 𝜑
−1 −1 −1

−2 −2 −2
(d) 𝑟 = −2√𝑘𝑐 (e) −2√𝑘𝑐 < 𝑟 < −8√66𝑘𝑐/33 (f) 𝑟 = −8√66𝑘𝑐/33

2 2 2

y 1 y 1 y 1

0 0 0
−0.5 0 0.5 1 1.5 2 2.5 −0.5 0 0.5 1 1.5 2 2.5 −0.5 0 0.5 1 1.5 2 2.5
𝜑 𝜑 𝜑
−1 −1 −1

−2 −2 −2
(g) 𝑟 = 8√66𝑘𝑐/33 (h) 8√66𝑘𝑐/33 < 𝑟 < 2√𝑘𝑐 (i) 𝑟 = 2√𝑘𝑐

4 6 4
y 4
y y
2 2
2
0
0 0
−2 −1 0 1 2 3 4 5
−1 0 1 2 3 4 −2 −1 0 1 2 3 4 5 6
𝜑
−2 𝜑 −2 𝜑
−4
−6
−4 −4
−8
(j) 2√𝑘𝑐 < 𝑟 < 4√3𝑘𝑐/3 (k) 𝑟 = 4√3𝑘𝑐/3 (l) 𝑟 > 4√3𝑘𝑐/3

Figure 2: The bifurcation phase portraits in different regions of Figure 1(a) for the system (11).
6 Abstract and Applied Analysis

10 10

y
y 5 5

0
−5 −4 −3 −2 −1 0 1 2 3 4 −4 −3 −2 −1 0 1 2 3 4 5
𝜑 𝜑

−5 −5

−10 −10
(a) 𝑟 < 0 (b) 𝑟 > 0

Figure 3: The bifurcation phase portraits in different regions of Figure 1(b) for the system (11).

0.2 0.2

0.1 0.1

0 0
−4 −3 −2 −1 0 1 2 3 4 −4 −3 −2 −1 0 1 2 3 4
𝜒 𝜉
−0.1 −0.1

−0.2 −0.2

(a) 𝜑1 (𝜉) (b) 𝑤1 (𝜉) (𝑐1 = 0)

Figure 4: The periodic wave solutions of the system (11) and the system (3) when 𝑟 < −4√3𝑘𝑐/3, 𝑘 > 0, 𝑐 > 0, and Δ ≥ 0.

Assuming that the following conditions (14) When 𝑟 > 0, the system (11) has two saddles at
(𝜑1,2 , 0) and a center at (0, 0) and ℎ0 < ℎ2 < ℎ1 exists. When
𝑘 < 0, 𝑐>0 (𝐼𝐼) ℎ ∈ (ℎ0 , ℎ2 ) ∪ (ℎ2 , ℎ1 ), the system (11) has a family of periodic
orbits in which the periodic orbit Γ18 is included; under other
hold, according to Figure 3, we obtain Case 2 as follows. cases, periodic orbits become open curves (see Figure 4(b)).
Case 2. Suppose that 𝑘 < 0, 𝑐 > 0, similarly, we have the
following. 4. Smooth Solitary Wave Solutions, Periodic
Wave Solutions, and Kink Wave Solutions
(13) When 𝑟 < 0, the system (11) has two saddles at (𝜑1,2 , 0) for the System (11) and the System (3)
and a center at (0, 0) and ℎ0 < ℎ1 < ℎ2 exists. When ℎ ∈
(ℎ0 , ℎ1 )∪(ℎ1 , ℎ2 ), the system (11) has a family of periodic orbits In this section, we will seek all traveling wave solutions which
in which the periodic orbit Γ17 is included; under other cases, correspond to the special bounded phase orbits of the system
periodic orbits become open curves (see Figure 4(a)). (11) in Section 3. The explicit expressions of the system (3) are
Abstract and Applied Analysis 7

also obtained by all traveling wave solutions of the system (11) Proposition 3. Under conditions (𝐼), one has following results.
and the relation (4).
(i) When −4√3𝑘𝑐/3 < 𝑟 < −2√𝑘𝑐, the system (11) has
4.1. Smooth Solitary Wave Solutions, Periodic Wave Solutions, a smooth solitary wave solution of valley type, which
and Kink and Antikink Wave Solutions of the System (11). corresponds to the orbit Γ4 of 𝐻(𝜑, 𝑦) = ℎ0 .
From the qualitative theory of dynamical system, we know (ii) When 2√k𝑐 < 𝑟 < 4√3𝑘𝑐/3, the system (11) has
that a smooth solitary wave solution of a partial differ- a smooth solitary wave solution of peak type, which
ential system corresponds to a smooth homoclinic orbit corresponds to the orbit Γ19 of 𝐻(𝜑, 𝑦) = ℎ0 .
of a traveling wave equation. A periodic orbit of traveling
wave equation corresponds to a periodic traveling wave Finally, we mention the conditions of existence for kink
solution of a partial differential system. Similarly, a smooth wave solutions of the system (11).
heteroclinic orbit of traveling wave equation corresponds to a
smooth kink (antikink) wave solution of a partial differential Proposition 4. When conditions (𝐼) hold, the system (11) has
system. smooth kink (antkink) under one of the following conditions:
According to the above analysis, in this section, we con- (1) 𝑟 = −2√kc, the system (11) has smooth kink which
sider the existence and the explicit exact expressions of corresponds to the orbits Γ9 and Γ10 of 𝐻(𝜑, 𝑦) = ℎ0
smooth periodic wave solutions, smooth solitary wave solu- (see Figure 2(d));
tions, and smooth kink (antikink) wave solutions for the
system (11) and the system (3) under the parameter conditions (2) 𝑟 = 2√kc, the system (11) has smooth kink which
(𝐼) and (𝐼𝐼). corresponds to the orbits Γ14 and Γ15 of 𝐻(𝜑, 𝑦) = ℎ0
Firstly, we consider the existence of smooth periodic wave (see Figure 2(i)).
solutions under parameter conditions (𝐼) and (𝐼𝐼).
4.2. Exact Traveling Wave Solutions of the System (11) and the
Proposition 2. (i) When 𝑘 > 0, 𝑐 > 0, and Δ ≥ 0, the System (3). Firstly, we will obtain some explicit expressions of
system (11) has a family of smooth periodic wave solutions (see traveling wave solutions for the system (11) when conditions
Figure 2), which correspond to 𝐻(𝜑, 𝑦) = ℎ, ℎ ∈ 𝐼, where 𝐼 is (𝐼) and (𝐼𝐼) hold. Furthermore, using potential (4) for the
one of intervals in (1), (3), (4), (9), (10), and (12) of Case 1. system (3), its exact traveling wave solutions are given as
follows.
(1) When 𝑟 < −4√3𝑘𝑐/3, ℎ ∈ (ℎ0 , ℎ1 ), where ℎ0 < ℎ1 < ℎ2 We only choose one of the periodic orbits to calculate
in Case 1(1) (see Figure 2(a)). periodic wave solutions.
(1) Periodic wave solutions for the system (11) and the
(2) When −4√3𝑘𝑐/3 < 𝑟 < −2√𝑘𝑐, ℎ ∈ (ℎ1 , ℎ0 ), where
system (3).
ℎ1 < ℎ0 < ℎ2 in Case 1(3) (see Figure 2(c)).
There are periodic orbits such as Γ1 , Γ6 , Γ9 , Γ12 , Γ15 , Γ16 ,
(3) When 𝑟 = −2√𝑘𝑐, ℎ ∈ (ℎ1 , ℎ0 ), where ℎ1 < ℎ0 = ℎ2 in Γ17 , and Γ18 (see Figures 2(a), 2(c), 2(d), 2(i), 2(j), 2(l), 3(a),
Case 1(4) (see Figure 2(d)). and 3(b)), which correspond to periodic wave solutions for
the system (11). We only choose one of the periodic orbits
(4) When 𝑟 = 2√𝑘𝑐, ℎ ∈ (ℎ2 , ℎ0 ), where ℎ2 < ℎ0 = ℎ1 in (see Figure 2(a)) to calculate periodic wave solutions. This
Case 1(9) (see Figure 2(i)). method can be used for other periodic orbits.
(5) When 2√𝑘𝑐 < 𝑟 < 4√3𝑘𝑐/3, ℎ ∈ (ℎ0 , ℎ1 ), where ℎ2 < When 𝑟 < −4√3𝑘𝑐/3 (see Figure 2(a)), we notice that
ℎ0 < ℎ1 in Case 1(10) (see Figure 2(j)). there are periodic orbit Γ1 and two special orbits Γ2 , Γ3 passing
the points (𝜑4 , 0), (𝜑5 , 0), (𝜑6 , 0), and (𝜑7 , 0). In the (𝜑, 𝑦)-
(6) When 𝑟 > 4√3𝑘𝑐/3, ℎ ∈ (ℎ0 , ℎ2 ), where ℎ0 < ℎ1 < ℎ2 plane the expressions of the orbits are given as
in Case 1(12) (see Figure 2(l)).

(ii) When 𝑘 < 0, 𝑐 > 0, the system (11) has a family (3𝑟2 − 16𝑘𝑐) 27𝑟 3
𝑦 = ±√ − 𝜑2 − 𝜑 + 𝜑4 + ℎ
of smooth periodic wave solutions (see Figure 2), which corre- 8𝑘4 𝑘2 (26)
spond to 𝐻(𝜑, 𝑦) = ℎ, ℎ ∈ 𝐼, where 𝐼 is one of the intervals in
(13) and (14) of Case 2.
= ±√(𝜑 − 𝜑4 ) (𝜑 − 𝜑5 ) (𝜑 − 𝜑6 ) (𝜑 − 𝜑7 ),
(1) When 𝑟 < 0, ℎ ∈ (ℎ0 , ℎ1 )∪(ℎ1 , ℎ2 ), where ℎ0 < ℎ1 < ℎ2
in Case 2(13) (see Figure 3(a)). where 𝜑4 < 𝜑5 < 0 < 𝜑6 < 𝜑7 .
Substituting (26) into 𝑑𝜑/𝑑𝜉 = 𝑦 and integrating them
(2) When 𝑟 < 0, ℎ ∈ (ℎ0 , ℎ2 )∪(ℎ2 , ℎ1 ), where ℎ0 < ℎ2 < ℎ1 along Γ1 , Γ2 , and Γ3 , it follows that
in Case 2(14) (see Figure 3(b)).
𝜑 𝜉
1
Secondly, we discuss the existence of solitary wave solu- ±∫ 𝑑𝑠 = ∫ 𝑑𝑠.
tions under group (I). We can summarize the results for the
𝜑5 √(𝜑 − 𝜑4 ) (𝜑 − 𝜑5 ) (𝜑 − 𝜑6 ) (𝜑 − 𝜑7 ) 0

system (11) from Figures 2(c) and 2(j). (27)

8 Abstract and Applied Analysis

Completing the above integral, we obtain one of the periodic Completing the above integral, we obtain the following soli-
traveling wave solutions (see Figure 4(a)) of (26): tary wave solution (see Figure 5(a)) of the system (11):
𝜑1 (𝜉) 2𝜑8 𝜑9
𝜑2 (𝜉) = .
(𝜑8 − 𝜑9 ) cosh ((√2/2) √𝜑8 𝜑9 𝜉) + 𝜑8 + 𝜑9
(𝜑5 − 𝜑4 ) 𝑠𝑛 (𝜉, 𝜑4 , 𝜑5 , 𝜑6 , 𝜑7 )
= 𝜑4 + , (33)
1 − ((𝜑6 − 𝜑5 ) / (𝜑6 − 𝜑4 )) 𝑠𝑛 (𝜉, 𝜑4 , 𝜑5 , 𝜑6 , 𝜑7 )
(28) Noting (6), we obtain the following exact solitary wave
solutions (see Figure 5(a)) of the system (11) from (33):
where 𝑠𝑛(𝜉, 𝜑4 , 𝜑5 , 𝜑6 , 𝜑7 ) = 𝑠𝑛2 (√2(𝜑5 − 𝜑7 )(𝜑4 − 𝜑6 )𝜉/2,
(𝜑6 − 𝜑5 )(𝜑7 − 𝜑4 )/(𝜑7 − 𝜑5 )(𝜑6 − 𝜑4 )). 𝑞2 (𝑥, 𝑦, 𝑡)
Noting (6), we obtain the the following exact periodic 2𝑎𝑘2
wave solutions of the system (11) from (28): =− ,
2√𝑟2 − 4𝑘𝑐 cosh ((√2𝑎/2) (𝑘𝑥 + 𝑟𝑦 − 𝑐𝑡)) − 𝑟
𝑞1 (𝑥, 𝑦, 𝑡) (34)
(𝜑5 − 𝜑4 ) 𝑠𝑛 (𝑥, 𝑦, 𝑡, 𝜑4 , 𝜑5 , 𝜑6 , 𝜑7 ) where 𝑞2 (𝑥, 𝑦, 𝑡) is a solitary wave solution of the system (11).
= 𝜑4 + ,
1 − ((𝜑6 − 𝜑5 ) / (𝜑6 − 𝜑4 )) 𝑠𝑛 (𝑥, 𝑦, 𝑡, 𝜑4 , 𝜑5 , 𝜑6 , 𝜑7 ) Since 𝑢(𝜉) = ∫ 𝜑(𝜉)𝑑𝜉, integrating (33) about 𝜉, we can
(29) obtain
𝑢2 (𝜉)
where 𝑠𝑛(𝑥, 𝑦, 𝑡, 𝜑4, 𝜑5, 𝜑6, 𝜑7) = 𝑠𝑛2 (2√(𝜑5 − 𝜑7 )(𝜑4 − 𝜑6)(𝑘𝑥+
𝑟𝑦 − 𝑐𝑡)/2, (𝜑6 − 𝜑5 )(𝜑7 − 𝜑4 )/(𝜑7 − 𝜑5 )(𝜑6 − 𝜑4 )). 𝑞1 (𝑥, 𝑦, 𝑡) is (𝑟 + 2√𝑟2 − 4𝑘𝑐) tanh (√2𝑎𝜉/4)
one of the smooth periodic wave solutions of the system (11). = 2√2tanh−1 ( ).
2𝑘2 √𝑎
Since 𝑢1 (𝜉) = ∫ 𝜑1 (𝜉)𝑑𝜉, integrating (28) about 𝜉, by (6), (35)
we can obtain one of smooth wave solutions 𝑢1 (𝑥, 𝑦, 𝑡) of sys-
tem (5). Applying the potential (4), the periodic wave solution According to (6), wave solutions of traveling wave equation
for the system (3) is obtained as follows: (7) from (5) are able to obtain

𝑤1 (𝑥, 𝑦, 𝑡) = 𝑘𝑞1 (𝑥, 𝑦, 𝑡) + 𝑐1 , (30) 𝑢2 (𝑥, 𝑦, 𝑡)

where 𝑐1 is a constant. √2𝑎 (𝑘𝑥 + 𝑟𝑦 − 𝑐𝑡)

= 2√2tanh−1 ((𝑟 + 2√𝑟2 − 4𝑘𝑐) ( )
Noting (6), the periodic traveling wave solution 𝑤1 (𝜉) for 4
the system (3) is obtained (see Figure 4(b)).
√2𝑎 (𝑘𝑥 + 𝑟𝑦 − 𝑐𝑡)
× tanh ( )
Remark 5. In [11], some periodic wave solutions of the system 4
(11) are obtained, but the periodic wave solutions of the
system (3) are not given. The periodic wave solutions of the −1
system (3) cannot be derived by the method [3]. In this paper, × (2𝑘2 √𝑎) ) ,
we obtain all periodic wave solutions of the system (11) and (36)
the system (3).
where 𝑢2 (𝑥, 𝑦, 𝑡) is one of the smooth wave solutions of (5).
(2) Solitary wave solutions for the system (11) and the We substitute (36) into the potential 𝑤(𝑥, 𝑦, 𝑡) = 𝑢𝑥 (𝑥, 𝑦, 𝑡)
system (3). as defined in (4) to obtain
When −4√3𝑘𝑐/3 < 𝑟 < −2√𝑘𝑐, from the phase portrait
(see Figure 2(c)), we notice that there are a homoclinic orbit 𝑤2 (𝑥, 𝑦, 𝑡) = − (2𝑘3 𝑎 (𝑟 + 2√𝑟2 − 4𝑘𝑐))
Γ4 and a special orbit Γ5 passing the points (𝜑8 , 0), (𝜑9 , 0), and
(0, 0). In (𝜑, 𝑦)-plane, the expressions of the orbits are given × ((5𝑟2 + 4𝑟√𝑟2 − 4𝑘𝑐 − 4𝑘4 𝑎 − 16𝑘𝑐)
as
2
𝑦 = ±√𝜑2 (𝜑 − 𝜑8 ) (𝜑 − 𝜑9 ), (31) √2𝑎 (𝑘𝑥 + 𝑟𝑦 − 𝑐𝑡)
× cosh ( )
4
when 𝐻(𝜑, 𝑦) = 0, where 𝜑8 = (𝑟 − 2√𝑟2 − 4𝑘𝑐)/2𝑘2 < 𝜑9 = −1
(𝑟 + 2√𝑟2 − 4𝑘𝑐)/2𝑘2 < 0. −4𝑟√𝑟2 − 4𝑘𝑐 + 5𝑟2 + 16𝑘𝑐) ,
Substituting (31) into 𝑑𝜑/𝑑𝜉 = 𝑦 and integrating them
along Γ4 and Γ5 , we have (37)
𝜑
1 𝜉 where 𝑤2 (𝑥, 𝑦, 𝑡) is a solitary wave solution of the system (3).
±∫ 𝑑𝑠 = ∫ 𝑑𝑠. (32) Using traveling wave transform (6), the solitary wave solution
𝜑9 √𝜑2 (𝜑 − 𝜑8 ) (𝜑 − 𝜑9 ) 0
𝑤2 (𝜉) of the system (3) is obtained (see Figure 5(b)).
Abstract and Applied Analysis 9

−4 −3 −2 −1 0 1 2 3 4 −4 −3 −2 −1 0 1 2 3 4
𝜉 𝜉
−0.1 −0.1

−0.2 −0.2

−0.3 −0.3
y y
−0.4 −0.4

−0.5 −0.5

−0.6 −0.6

−0.7 −0.7
(a) 𝜑2 (𝜉) (b) 𝑤2 (𝜉)

Figure 5: The solitary wave of the system (11) and the system (3) when −4√3𝑘𝑐/3 < 𝑟 < −2√𝑘𝑐, 𝑘 > 0, 𝑐 > 0, and Δ ≥ 0.

When 2√𝑘𝑐 < 𝑟 < 4√3𝑘𝑐/3 (see Figure 2(j)), the expres- According to (6), one of the smooth wave solutions of (5) is
sions of the homoclinic orbit Γ7 and the special orbit Γ8 able to obtain
passing the points (𝜑11 , 0), (0, 0), and (𝜑12 , 0) are given as in
(𝜑, 𝑦)-plane: 𝑢3 (𝑥, 𝑦, 𝑡)

𝑦 = ±√𝜑2 (𝜑 − 𝜑11 ) (𝜑 − 𝜑12 ), (38) = −2√2tanh−1 (((−𝑟 + 2√𝑟2 − 4𝑘𝑐)

when 𝐻(𝜑, 𝑦) = 0, where 0 < 𝜑11 = (𝑟 − 2√𝑟2 − 4𝑘𝑐)/2𝑘2 <
𝜑12 = (𝑟 + 2√𝑟2 − 4𝑘𝑐)/2𝑘2 . √2𝑎 (𝑘𝑥 + 𝑟𝑦 − 𝑐𝑡)
× tanh ( ))
Substituting (38) into 𝑑𝜑/𝑑𝜉 = 𝑦 and integrating them 4
along Γ19 and Γ20 , we have
𝜑11 𝜉 −1
±∫
1
𝑑𝑠 = ∫ 𝑑𝑠. × (2𝑘2 √𝑎) ) .
(39)
𝜑 √𝜑2 (𝜑 − 𝜑11 ) (𝜑 − 𝜑12 ) 0
(43)
Completing the above integral, we obtain the following soli-
tary wave solution (see Figure 6(a)) of the system (11): Substitute (42) into the potential 𝑤(𝑥, 𝑦, 𝑡) = 𝑢𝑥 (𝑥, 𝑦, 𝑡)
as defined in (4) to obtain
2𝜑8 𝜑9
𝜑3 (𝜉) = .
(𝜑9 − 𝜑8 ) cosh ((√2/2) √𝜑8 𝜑9 𝜉) + 𝜑8 + 𝜑9 𝑤3 (𝑥, 𝑦, 𝑡) = (2𝑘3 𝑎 (𝑟 − 2√𝑟2 − 4𝑘𝑐))
(40)
Noting (6), we obtain the following exact wave solutions
of the system (11) from (40) × ((−5𝑟2 + 4𝑟√𝑟2 − 4𝑘𝑐 + 4𝑘2 𝑎 + 16𝑘𝑐)

𝑞3 (𝑥, 𝑦, 𝑡)
2
√2𝑎 (𝑘𝑥 + 𝑟𝑦 − 𝑐𝑡)
2𝑘 𝑎 2 (41) × (cosh ( ))
= , 4
√ 2 √
2 𝑟 − 4𝑘𝑐 cosh (( 2𝑎/2) (𝑘𝑥 + 𝑟𝑦 − 𝑐𝑡)) + 𝑟
−1
where 𝑞3 (𝑥, 𝑦, 𝑡) is a solitary wave solution of the system (11).
−4𝑟√𝑟2 − 4𝑘𝑐 + 5𝑟2 − 16𝑘𝑐) ,
Since 𝑢(𝜉) = ∫ 𝜑(𝜉)𝑑𝜉, integrating (40) about 𝜉, we can
obtain (44)
(−𝑟 + 2√𝑟2 − 4𝑘𝑐) tanh (√2𝑎𝜉/4)
𝑢3 (𝜉) = −2√2tanh−1 ( ). where 𝑤3 (𝑥, 𝑦, 𝑡) is a solitary wave solution of the system (3).
2𝑘2 √𝑎 Applying (6), the solitary wave solution 𝑤3 (𝜉) is obtained for
(42) the system (3) (see Figure 6(b)).
10 Abstract and Applied Analysis

0.7 0.7

0.6 0.6

0.5 0.5

0.4 0.4
y y
0.3 0.3

0.2 0.2

0.1 0.1

−4 −3 −2 −1 0 1 2 3 4 −4 −3 −2 −1 0 1 2 3 4
𝜉 𝜉
(a) 𝜑3 (𝜉) (b) 𝑤3 (𝜉)

Figure 6: The solitary wave of the system (11) and the system (3) when 2√𝑘𝑐 < 𝑟 < 4√3𝑘𝑐/3, 𝑘 > 0, 𝑐 > 0, and Δ ≥ 0.

Remark 6. In [11], only one solitary wave solution of peak Noting (6), we obtain the exact wave solution of the system
type wave solutions of the system (3) is obtained. However, (11) from (47). Consider
the solitary wave solutions of system (3) cannot be found
by Hirota’s bilinear method [3]. Fortunately, we obtain all √𝑘𝑐 (1 − tanh (√2𝑘𝑐 (𝑘𝑥 + 𝑟𝑦 − 𝑐𝑡) /2𝑘2 ))
𝑞4 (𝑥, 𝑦, 𝑡) = − ,
solitary wave solutions of the system (11) and the system (3). 2𝑘2
(48)
(3) Kink (antikink) wave solutions for the system (11) and
the system (3). where 𝑞4 (𝑥, 𝑦, 𝑡) is a kink wave solution of the system (11).
When 𝑟 = −2√𝑘𝑐 (see Figure 2(d)), in (𝜑, 𝑦)-plane, the Since 𝑢(𝜉) = ∫ 𝜑(𝜉)𝑑𝜉, integrating (47) about 𝜉, we can
expressions of the heteroclinic orbits Γ10 and Γ11 passing the obtain
points (𝜑13 , 0), (0, 0) are given as 𝑢4 (𝜉)

1 √ √2
𝑐𝜑2 √𝑘𝑐 1 = −( 𝑘𝑐𝜉 +
𝑦 = ±√ 3 + 2 2 𝜑3 + 𝜑4 , (45) 2𝑘 2 4
𝑘 𝑘 2
√2𝑘𝑐𝜉 (49)
when 𝐻(𝜑, 𝑦) = 0, where 𝜑13 = −√𝑘𝑐/𝑘2 < 0. × ( ln (1 + tanh ( ))
2𝑘2
Substituting (45) into 𝑑𝜑/𝑑𝜉 = 𝑦 and integrating them
along Γ10 and Γ11 , we have √2 √2𝑘𝑐𝜉
+ ln (1 − tanh ( )))) .
4 2𝑘2
𝜑
1 √2 𝜉
∫ 𝑑𝑠 = ∫ 𝑑𝑠, According to (6), kink wave solutions of (5) are able to obtain
𝜑13 2 2 0 (46a)
√ 𝜑2 (𝜑 + √𝑘𝑐/𝑘2 )
𝑢4 (𝑥, 𝑦, 𝑡)
𝜑
1 √2 𝜉 1 √
−∫ 𝑑𝑠 = ∫ 𝑑𝑠. = −( 𝑘𝑐 (𝑘𝑥 + 𝑟𝑦 − 𝑐𝑡)
𝜑13 2 2 0 (46b) 2𝑘2
√ 𝜑2 (𝜑 + √𝑘𝑐/𝑘2 )
√2 √2𝑘𝑐 (𝑘𝑥 + 𝑟𝑦 − 𝑐𝑡)
+ ( ln (− tanh ( ) − 1)
Completing the above integral (46a), we obtain the following 4 2𝑘2
kink wave solution of the system (11) (see Figure 7(a)):
√2 √2𝑘𝑐 (𝑘𝑥 + 𝑟𝑦 − 𝑐𝑡)
+ ln (1 − tanh ( )))) .
√𝑘𝑐 (1 − tanh (√2𝑘𝑐𝜉/2𝑘2 )) 4 2𝑘2
𝜑4 (𝜉) = − . (47)
2𝑘2 (50)
Abstract and Applied Analysis 11

−4 −2 0 2 4 −4 −2 0 2 4
𝜉 𝜉

−0.5 −0.5

y −1 y −1

−1.5 −1.5

−2 −2
2 2
(a) 𝜑4 (𝜉) = −(√𝑘𝑐(1 − tanh(√2𝑘𝑐𝜉/2𝑘 )))/2𝑘 (b) 𝑤4 (𝜉) = √𝑘𝑐(−1 + tanh(√2𝑘𝑐𝜉/2𝑘2 ))/2𝑘

−4 −2 0 2 4 −4 −2 0 2 4
𝜉 𝜉

−0.5 −0.5

y −1 y −1

−1.5 −1.5

−2 −2
(c) 𝜑4󸀠 (𝜉) = √𝑘𝑐(1 + tanh(√2𝑘𝑐𝜉/2𝑘2 ))/ − 2𝑘2 (d) 𝑤4󸀠 (𝜉) = √𝑘𝑐(1 + tanh(√2𝑘𝑐𝜉/2𝑘2 ))/ − 2𝑘

Figure 7: The kink (antikink) wave solutions of the system (11) and the system (3) when 𝑟 = −2√𝑘𝑐, 𝑘 > 0, 𝑐 > 0, and Δ ≥ 0.

Using the potential (4), the kink wave solution for the √𝑘𝑐 (1 + tanh (√2𝑘𝑐 (𝑘𝑥 + 𝑟𝑦 − 𝑐𝑡) /2𝑘2 ))
system (3) are obtained as follows: 𝑤4󸀠 (𝑥, 𝑦, 𝑡) = − ,
2𝑘
(53)
√𝑘𝑐 (1 − tanh (√2𝑘𝑐 (𝑘𝑥 + 𝑟𝑦 − 𝑐𝑡) /2𝑘2 ))
𝑤4 (𝑥, 𝑦, 𝑡) = − ,
2𝑘 for the system (11) and the system (3), respectively (see
(51) Figures 7(c) and 7(d)).
When 𝑟 = 2√𝑘𝑐 (see Figure 2(i)), in (𝜑, 𝑦)-plane, the
where 𝑤4 (𝑥, 𝑦, 𝑡) is the smooth kink wave solution of the expressions of the heteroclinic orbits Γ13 and Γ14 passing the
system (3). The smooth kink wave solution 𝑤4 (𝜉) of the sys- points (0, 0), (𝜑14 , 0) are given as
tem (3) is obtained from (51) (see Figure 7(b)). Analogously,
completing the above integral (46b), we have the following
𝑐𝜑2 √𝑘𝑐 1
antikink wave solution 𝑦 = ±√ 3
− 2 2 𝜑3 + 𝜑4 , (54)
𝑘 𝑘 2
√𝑘𝑐 (1 + tanh (√2𝑘𝑐𝜉/2𝑘2 ))
𝜑4󸀠 (𝜉) = − , (52) when 𝐻(𝜑, 𝑦) = 0, 𝜑14 = √𝑘𝑐/𝑘2 > 0.
2𝑘2
12 Abstract and Applied Analysis

Substituting (54) into 𝑑𝜑/𝑑𝜉 = 𝑦 and integrating them Applying the potential (4), the kink wave solution of the
along Γ13 and Γ14 , we have system (3) is obtained as follows:
𝑤5 (𝑥, 𝑦, 𝑡)
𝜑14 √2 𝜉
∫ 𝑑𝑠 = ∫ 𝑑𝑠, √𝑘𝑐 (1 + tanh (√𝑘𝑐 (𝑘𝑥 + 𝑟𝑦 − 𝑐𝑡) /2𝑘)) (60)
𝜑 2 0 (55a)
√ 𝜑2 (𝜑 − √𝑘𝑐/𝑘2 ) = ,
2𝑘
where 𝑤5 (𝑥, 𝑦, 𝑡) is the smooth kink wave solution of the
𝜑14 √2 𝜉
system (3).
−∫ 𝑑𝑠 = ∫ 𝑑𝑠.
𝜑 2 0 (55b) Noting (6), we can obtain the smooth kink wave solution
√ 𝜑2 (𝜑 − √𝑘𝑐/𝑘2 )
𝑤5 (𝜉) of the system (3) (see Figure 8(b)). Analogously, com-
pleting the above integral (55b), we obtain the exact antikink
wave solution:
Completing the above integral (55a), we obtain the exact kink
wave solution of the system (11) (see Figure 8(a)): √𝑘𝑐 (1 − tanh (√2𝑘𝑐𝜉/2𝑘2 ))
𝜑5󸀠 (𝜉) = ,
2𝑘2
√𝑘𝑐 (1 + tanh (√2𝑘𝑐𝜉/2𝑘2 )) (61)
(56) √𝑘𝑐 (1 − tanh (√𝑘𝑐 (𝜉) /2𝑘))
𝜑5 (𝜉) = .
2𝑘2 𝑤5󸀠 (𝜉) = ,
2𝑘

Noting (6), we obtain the following exact wave solution of the for the system (11) and the system (3), respectively (see
system (11) from (56): Figures 8(c) and 8(d)).

Remark 7. In [3], applying the necessary condition for the

𝑞5 (𝑥, 𝑦, 𝑡) kink waves to exist, multiple kink solutions and multiple
singular kink solutions of the system (3) are formally derived.
√𝑘𝑐 (1 + tanh (√𝑘𝑐 (𝑘𝑥 + 𝑟𝑦 − 𝑐𝑡) /2𝑘)) (57) By the special traveling wave transform in [11], no kink
= , (antikink) solutions of the system (11) and system (3) are
2𝑘2
obtained. In this paper, not considering the necessary con-
dition for the kink waves to exist [3], we obtain all kink and
where 𝑞5 (𝑥, 𝑦, 𝑡) is a kink wave solution of the system antikink wave solutions of the system (11) and the system (3)
(11). by the bifurcation method of dynamical systems.
Since 𝑢(𝜉) = ∫ 𝜑(𝜉)𝑑𝜉, integrating (56) about 𝜉, we can From (1) to (12), we can obtain three theorems about
obtain the exact periodic wave solutions and smooth solitary wave
solutions and kink (antikink) wave solutions for the system
√𝑘𝑐 √2 (11) and the system (3).
𝑢5 (𝜉) = ( 𝜉−
2𝑘2 4
Theorem 8. When conditions (𝐼) or (𝐼𝐼) hold, one can obtain
√𝑘𝑐𝜉 some representative smooth exact periodic wave solutions of the
× ( ln (tanh ( ) − 1) (58)
2𝑘 system (11) from the periodic obits (see Figures 2(a), 2(c), 2(i),
2(j), and 2(l)) and Figures 3(a) and 3(b)), which correspond
√2 √2𝑘𝑐𝜉 to the representative smooth exact periodic wave solutions of
− ln (1 + tanh ( )))) . the system (3), such as the periodic wave solution 𝑞1 (𝑥, 𝑦, 𝑡)
4 2𝑘2
of the system (11) corresponding to the periodic wave solution
𝑤1 (𝑥, 𝑦, 𝑡) of the system (3).
According to (6), the traveling wave solution of (5) is able to
obtain Theorem 9. Under conditions (𝐼), the following results hold.
(1) When −4√3𝑘𝑐/3 < 𝑟 < −2√𝑘𝑐, the system (11)
𝑢5 (𝑥, 𝑦, 𝑡)
has an exact solitary wave solution 𝑞2 (𝑥, 𝑦, 𝑡), which
√𝑘𝑐 corresponds to the solitary wave solutions 𝑤2 (𝑥, 𝑦, 𝑡) of
=( (𝑘𝑥 + 𝑟𝑦 − 𝑐𝑡) the system (3).
2𝑘2
(2) When 2√𝑘𝑐 < 𝑟 < 4√3𝑘𝑐/3, the system (11) has exact
√2 √𝑘𝑐 (𝑘𝑥 + 𝑟𝑦 − 𝑐𝑡) solitary wave solutions 𝑞3 (𝑥, 𝑦, 𝑡), which correspond to
− ( ln (tanh ( ) − 1) the solitary wave solutions 𝑤3 (𝑥, 𝑦, 𝑡) of the system (3).
4 2𝑘

√2 √2𝑘𝑐 (𝑘𝑥 + 𝑟𝑦 − 𝑐𝑡) Theorem 10. Under conditions (𝐼), the following results hold.
− ln (1 + tanh ( )))).
4 2𝑘2 (1) When 𝑟 = −2√𝑘𝑐, the system (11) has an exact smooth
(59) kink wave solution 𝑞4 (𝑥, 𝑦, 𝑡), which corresponds to
Abstract and Applied Analysis 13

2 2

1.5 1.5

y 1 y 1

0.5 0.5

−4 −2 0 2 4 −4 −2 0 2 4
𝜉 𝜉
(a) 𝜑5 (𝜉) = √𝑘𝑐(1 + tanh(√2𝑘𝑐𝜉/2𝑘2 ))/2𝑘2 (b) 𝑤5 (𝜉) = √𝑘𝑐(1 + tanh(√𝑘𝑐𝜉/2𝑘))/2𝑘

2 2

1.5 1.5

y 1 y 1

0.5 0.5

−4 −2 0 2 4 −4 −2 0 2 4
𝜉 𝜉
(c) 𝜑5󸀠 (𝜉) = √𝑘𝑐(1 − tanh(√2𝑘𝑐𝜉/2𝑘2 ))/2𝑘2 (d) 𝑤5󸀠 (𝜉) = √𝑘𝑐(1 − tanh(√𝑘𝑐𝜉/2𝑘))/2𝑘

Figure 8: The kink (antikink) wave solutions of system (11) and (3) when 𝑟 = 2√𝑘𝑐, 𝑘 > 0, 𝑐 > 0, and Δ ≥ 0.

the kink wave solution 𝑤4 (𝑥, 𝑦, 𝑡) of the system (3), integrability of these models in [3], multiple kink solutions
respectively. and multiple singular kink solutions were formally derived
for the third model. Compared to the method in [3], the
(2) When 𝑟 = 2√𝑘𝑐, the system (11) has an exact smooth necessary condition which is among the coefficients of the
kink wave solution 𝑞5 (𝑥, 𝑦, 𝑡), which corresponds to the spatial variables is not necessary in our method. In [11], only
kink wave solution 𝑤5 (𝑥, 𝑦, 𝑡) of the system (3). solitary wave solutions are obtained by a special traveling
wave transform. To our knowledge, we completely obtain all
5. Discussion solitary wave solutions and kink (antikink) wave solutions
for these models by the bifurcation method of dynamical
In this work we obtain all wave solutions of the complete system. The idea of applying the method of dynamical system
integrability of the (2+1)-dimensional nonlinear evolution to research the complete integrability can be used for other
equation, the third model, by dynamical systems method. models. This will be examined in forthcoming works. For the
This method can be used for the remaining three models. remaining three models, we can follow the same approach to
By determining the necessary condition for the complete derive all wave solutions for them.
14 Abstract and Applied Analysis

Conflict of Interests [12] Y. Zhou and Q. Liu, “Bifurcation of travelling wave solutions for
a (2 + 1)-dimensional nonlinear dispersive long wave equation,”
The authors declare that there is no conflict of interests Applied Mathematics and Computation, vol. 189, no. 1, pp. 970–
regarding the publication of this paper. 979, 2007.

Acknowledgments
The authors gratefully acknowledge the support of the
National Natural Science Foundation of China through Grant
nos. 11072007, 11372014, and 11290152, the Natural Science
Foundation of Beijing through Grant no. 1122001, the Inter-
national Science and Technology Cooperation Program of
China through Grant no. 2014DFR61080, and the Research
Fund for the Doctoral Program of Higher Education of China
through Grant no. 20131103120027. All authors wish to thank
professor Li Jibin for many valuable suggestions leading to the
improvement of this paper.

References
[1] X. Geng, C. Cao, and H. H. Dai, “Quasi-periodic solutions for
some (2 + 1)-dimensional integrable models generated by the
Jaulent-Miodek hierarchy,” Journal of Physics A: Mathematical
and General, vol. 34, no. 5, pp. 989–1004, 2001.
[2] X. Geng and Y. Ma, “N-soliton solution and its Wronskian form
of a (3+1)-dimensional nonlinear evolution equation,” Physics
Letters A, vol. 369, no. 4, pp. 285–289, 2007.
[3] A. Wazwaz, “Multiple kink solutions and multiple singular kink
solutions for (2+1)-dimensional nonlinear models generated by
the Jaulent-Miodek hierarchy,” Physics Letters A, vol. 373, no. 21,
pp. 1844–1846, 2009.
[4] Y. Y. Zhang, X. Q. Liu, and G. W. Wang, “Symmetry reductions
and exact solutions of the (2 + 1)-dimensional Jaulent-Miodek
equation,” Applied Mathematics and Computation, vol. 219, no.
3, pp. 911–916, 2012.
[5] R. Hirota, The Direct Method in Soliton Theory, vol. 155,
Cambridge University Press, Cambridge, UK, 2004.
[6] W. Hereman and A. Nuseir, “Symbolic methods to construct
exact solutions of nonlinear partial differential equations,”
Mathematics and Computers in Simulation, vol. 43, no. 1, pp. 13–
27, 1997.
[7] A. M. Wazwaz, Partial Differential Equations and Solitary Waves
Theorem, Springer and HEP, Berlin, Germany, 2009.
[8] A. Wazwaz, “Multiple soliton solutions for some (3 + 1)-dimen-
sional nonlinear models generated by the Jaulent-Miodek hier-
archy,” Applied Mathematics Letters, vol. 25, no. 11, pp. 1936–
1940, 2012.
[9] J. B. Li and G. R. Chen, “Exact traveling wave solutions and
their bifurcations for the Kudryashov-Sinelshchikov equation,”
International Journal of Bifurcation and Chaos, vol. 22, no. 5,
Article ID 1250118, pp. 1–19, 2012.
[10] J. Li, Y. Zhang, and X. Zhao, “On a class of singular nonlinear
traveling wave equations (II): an example of Gckdv equations,”
International Journal of Bifurcation and Chaos in Applied Sci-
ences and Engineering, vol. 19, no. 6, pp. 1995–2007, 2009.
[11] H. Liu and F. Yan, “The bifurcation and exact travelling wave
solutions for (2 + 1)-dimensional nonlinear models generated
by the Jaulent-Miodek hierarchy,” International Journal of Non-
linear Science, vol. 11, no. 2, pp. 200–205, 2011.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2015, Article ID 354918, 11 pages
http://dx.doi.org/10.1155/2015/354918

Research Article
Hopf Bifurcation, Cascade of Period-Doubling, Chaos,
and the Possibility of Cure in a 3D Cancer Model

Marluci Cristina Galindo, Cristiane Nespoli, and Marcelo Messias

Departamento de Matemática e Computação, Faculdade de Ciências e Tecnologia, Universidade Estadual Paulista (UNESP),
19060-900 Presidente Prudente, SP, Brazil
Correspondence should be addressed to Marcelo Messias; marcelo@fct.unesp.br

Received 27 June 2014; Accepted 13 October 2014

Academic Editor: Yongli Song

Copyright © 2015 Marluci Cristina Galindo et al. This is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is
properly cited.

We study a cancer model given by a three-dimensional system of ordinary differential equations, depending on eight parameters,
which describe the interaction among healthy cells, tumour cells, and effector cells of immune system. The model was previously
studied in the literature and was shown to have a chaotic attractor. In this paper we study how such a chaotic attractor is formed.
More precisely, by varying one of the parameters, we prove that a supercritical Hopf bifurcation occurs, leading to the creation of a
stable limit cycle. Then studying the continuation of this limit cycle we numerically found a cascade of period-doubling bifurcations
which leads to the formation of the mentioned chaotic attractor. Moreover, analyzing the model dynamics from a biological point
of view, we notice the possibility of both the tumour cells and the immune system cells to vanish and only the healthy cells survive,
suggesting the possibility of cure, since the interactions with the immune system can eliminate tumour cells.

1. Introduction destruction by NK cells and phagocytosis by macrophages.

The study of different antibodies found in patients with
Mathematical models describing cancer tumour growth have cancer tumour has been useful for determining tumour-
been extensively studied in the literature in order to under- associated antigens and pointed out some information for
stand the mechanism of disease and to predict its future the development of therapeutic antibodies with clinical
behaviour; see [1] and references therein as a brief review on application. More specifically, several researches seeking the
the subject. One class of these cancer models, well known as elimination of tumour cells through the immune cells have
Lotka-Volterra systems, is based on nonlinear mathematical been developed [3–5]. Despite the increasing number of
systems of interacting species. researches in this area, in [1] the authors pointed out the
The competition for nutrients, the effects of growth existence of a gap in the understanding of tumour growth
factors, and the actions of the immune system stand out and invasion, requiring an intense and closer collaboration in
among possible interactions. investigations by different scientists, such as mathematicians,
The immune system has the function of recognizing biologists, and experimentalists.
internal and external threats to the organism, reacting to Aiming to contribute to the understanding of these types
eliminate, neutralize, or tolerate such threats. The recognition of model we perform a bifurcation analysis of a cancer
of tumour cells by the immune system can happen in distinct tumour growth model of three competing cell populations
and complementary ways. According to Chammas et al. [2], (tumour, healthy, and immune cells), which was proposed in
the lymphocytes of type B participate in the immune response [6]. The populations of these cells in a given instant of time
against tumours by producing specific antibodies against 𝑡 will be denoted here by 𝑥(𝑡), 𝑦(𝑡), and 𝑧(𝑡), respectively.
tumour antigens. These antibodies bind with specific anti- Analogous to what is done in competition models in pop-
gens on tumour cells which facilitate their recognition and ulation ecology, the mathematical model for the interaction
2 Abstract and Applied Analysis

of these cells is given by the following three-dimensional point, we numerically found the occurrence of a cascade of
adimensional system of ordinary differential equations (for period-doubling bifurcations which leads to the formation
details on the derivation of the system, see the nice paper of the chaotic attractor shown to exist in [6]. Trying to
[6]), analyse this dynamical behaviour from the biological point
of view, we observed that, after the occurrence of chaotic
𝑑𝑥 dynamics, both the tumour cells and the immune system
= 𝑥 (1 − 𝑥) − 𝑎𝑥𝑦 − 𝑏𝑥𝑧, cells (represented by the coordinates 𝑥(𝑡) and 𝑧(𝑡)) vanish
𝑑𝑡
and only the healthy cells remain positive and their amount
𝑑𝑦 tends to the carrying capacity 𝑦(𝑡) = 1. The chaotic dynamics
= 𝑐𝑦 (1 − 𝑦) − 𝑑𝑥𝑦, (1)
𝑑𝑡 occurs when the parameter 𝑎 varies from the critical value 𝑎 ≅
𝑑𝑧 𝑒𝑥𝑧 0.717455 to 𝑎 = 1.0 and the dynamics is not affected by the
= − 𝑔𝑥𝑧 − ℎ𝑧, variation of parameter 𝑏, which must only be taken different
𝑑𝑡 𝑥+𝑓
from zero, as will be shown in the calculations ahead. In
this way, the bifurcation analysis presented here is performed
in which the parameters 𝑎, 𝑏, 𝑐, 𝑑, 𝑒, 𝑓, 𝑔, ℎ are all positive. focusing on the variation of parameter 𝑎.
The equations of system (1) describe the variation rate of The parameter 𝑎 considered in the mentioned bifurcation
tumour cells, healthy cells, and effector cells with the time analysis represents the interaction between the cancer cells
𝑡. In the first equation, a logistic growth of tumour cells is and the healthy cells. However, this parameter can also be
considered, the term 𝑎𝑥𝑦 represents the negative effects that related to the action of the immune system cells against
the healthy cells have on tumour cells, and 𝑏𝑥𝑧 represents the cancer cells. In fact, we can rewrite the first equation of system
negative effects that the effector cells of the immune system (1) as
exert on tumour cells. In the second equation, the healthy
cells also grow logistically, with growth rate 𝑐. The term 𝑑𝑥𝑦
𝑑𝑥 𝑏
represents the negative effects that tumour cells exert on = 𝑥 (1 − 𝑥) − 𝑎𝑥 (𝑦 + 𝑧) . (2)
healthy cells. In the last equation, 𝑒𝑥𝑧/(𝑥 + 𝑓) describes the 𝑑𝑡 𝑎
stimulation of the immune system by tumour cells through
specific antigens. Therefore, the stimulus of the immune In this way, for 𝑎 smaller than 1.0 and 𝑏 ≠ 0 fixed, the
system depends directly on the number of tumour cells. negative effect of the immune system cells onto the tumour
The term 𝑔𝑥𝑧 represents the negative effects that tumour cells is more significant and it decreases as 𝑎 approaches the
cells exert on the effector cells of the immune system and, critical value 𝑎 = 1.0, when the chaotic dynamics occurs.
finally, ℎ refers to the natural death rate of the effector cells. Although being a very complex matter, this and another
It is important to notice that system (1) is a slight modi- tentative analysis relating the dynamical behaviour and the
fication of a model of cancer without treatment, proposed biological meaning of the variables of system (1) will be
in [7], where a first phase space analysis of the model was further considered later on, in Sections 2, 4, and 5.
presented. We believe that the bifurcation analysis and the biological
In [6], by using the Shilnikov’s theorem [8] and by the considerations presented here complement the results and
calculation of Lyapunov exponents and Lyapunov dimension, the analysis of system (1) performed in [6, 7, 9].
the authors have shown that system (1) presents a chaotic The paper is organized as follows. In this introductory
attractor for the following fixed set of parameters: 𝑎 = 1.0, section, we present the studied model, describing the vari-
𝑏 = 2.5, 𝑐 = 0.6, 𝑑 = 1.5, 𝑒 = 4.5, 𝑓 = 1.0, 𝑔 = 0.2, ables and parameters involved, and state the main results
and ℎ = 0.5. In a second paper concerning the dynamical obtained. In Section 2, we make a linear analysis of system
analysis of system (1) [9], the authors described in detail the (1): calculate the equilibrium points, check whether these
topological structure of this chaotic attractor. However, in equilibria are biologically admissible (i.e., if all their coor-
none of these papers the authors show how this attractor is dinates are nonnegative), analyse their local stability, and
formed nor do they relate the chaotic behaviour of system (1) give an outline of the Hopf bifurcation theorem. In Section 3
with biological aspects of the cancer evolution. we prove the main result of the paper, the occurrence of
In this paper we analyse the local stability of the equilibria a supercritical Hopf bifurcation in one of the equilibrium
of system (1); then, varying two of the parameters involved, points of system (1), which gives rise to a small stable
namely, the parameters 𝑎 and 𝑏, and taking for the other limit cycle; we also present some numerical simulations
parameters the same values considered in [6], we perform which corroborate with the occurrence of this bifurcation.
a bifurcation analysis of system (1). Through this analysis In Section 4 we numerically show that a cascade of period-
we prove that, for 𝑏 > 0 fixed and varying the parameter doubling bifurcations occurs with the limit cycle created at
𝑎, the system presents a supercritical Hopf bifurcation at the Hopf bifurcation, leading to the creation of a chaotic
an equilibrium point for the critical value 𝑎 ≅ 0.717455, attractor (exactly that one shown to exist in [6]). Finally,
which leads to the creation of a small stable limit cycle. Section 5 presents some concluding remarks, including a
Furthermore, numerically studying the continuation of this possible interpretation of the dynamical properties of system
stable limit cycle by increasing the parameter 𝑎 from the Hopf (1) from a biological point of view.
Abstract and Applied Analysis 3

2. Linear Analysis the values of parameters extracted from [6], that is, 𝑐 = 0.6,
𝑑 = 1.5, 𝑒 = 4.5, 𝑓 = 1, 𝑔 = 0.2, and ℎ = 0.5, and letting
Considering 𝑥̇ = 𝑦̇= 𝑧̇ = 0 in system (1), we obtain the the parameters 𝑎 and 𝑏 vary. For these parameter values we
following equilibrium points (for details on the calculation obtain
and on the local analysis of the equilibrium points of system
(1), see [6]): 0.867497
𝐸 = (0.132503, 0, ),
𝑏
𝐴 = (0, 0, 0) , 𝐵 = (0, 1, 0) , 17.867497
𝐸1 = (18.8675, 0, − ),
𝐶 = (1, 0, 0) , 𝐸 = (𝑥, 0, 𝑧) , 𝑏
(3) 2 (𝑎 − 1) 3
𝑐 (𝑎 − 1) 𝑑 − 𝑐 𝐹=( , , 0) ,
𝐸1 = (𝑥1 , 0, 𝑧1 ) , 𝐹=( , , 0) , 5𝑎 − 2 5𝑎 − 2
𝑎𝑑 − 𝑐 𝑎𝑑 − 𝑐
0.1 (8.675 − 6.687𝑎)
𝐺 = (𝑥2 , 𝑦2 , 𝑧2 ) , 𝐺1 = (𝑥3 , 𝑦3 , 𝑧3 ) . 𝐺 = (0.132503, 0.668742, ),
𝑏
These equilibria are called biologically admissible if their 0.1 (−178.675 + 461.687𝑎)
𝐺1 = (18.8675, −46.1687, ).
three coordinates are greater than or equal to zero. The 𝑏
equilibrium points 𝐴, 𝐵, and 𝐶 satisfy such condition. (4)
For the equilibrium point 𝐹 we have the following
possibilities: Note that the equilibria 𝐴, 𝐵, 𝐶 remain the same and 𝐸1
and 𝐺1 are not biologically admissible; therefore it will not
(i) if 𝑎 = 1, the equilibrium 𝐹 coincides with the be considered here. Let us analyse the conditions on the
equilibrium 𝐵; equilibria 𝐸, 𝐹, and 𝐺.

(ii) if 𝑐 = 𝑑, 𝐹 coincides with the equilibrium 𝐶; (i) 𝐸 is biologically admissible, since we are considering
𝑏 > 0;
(iii) if 𝑎 > 1, 𝑎𝑑 > 𝑐, and 𝑑 > 𝑐, 𝐹 is biologically admissible; (ii) the equilibrium 𝐹 exists if 𝑎 ≠0.4 and coincides with
(iv) if 𝑎 < 1, 𝑎𝑑 < 𝑐, and 𝑑 < 𝑐, 𝐹 is also admissible; the equilibrium 𝐵 for 𝑎 = 1, as seen previously. If 𝑎 >
1, 𝐹 is biologically admissible, which does not occur
(v) if none of the previous inequalities are satisfied, 𝐹 is if 𝑎 < 1;
not biologically admissible.
(iii) 𝐺 is biologically admissible if 𝑎 < 1.2972, since 𝑏 > 0.
As the equilibria 𝐸, 𝐸1 and 𝐺, 𝐺1 may present negative
coordinates, aiming to simplify the analysis of system (1), we 2.1. Linear Stability of the Equilibrium Points. The Jacobian
will present a study of the feasibility of these equilibria using matrix of system (1) at a point (𝑥, 𝑦, 𝑧) is given by

1 − 2𝑥 − 𝑎𝑦 − 𝑏𝑧 −𝑎𝑥 −𝑏𝑥
−1.5𝑦 0.6 − 1.2𝑦 − 1.5𝑥 0
𝐽=( 2
). (5)
0.1𝑧 (43 − 2𝑥 − 4𝑥) 0.1 (38𝑥 − 2𝑥2 − 5)
0
(𝑥 + 1)2 𝑥+1

It will be used bellow to study the local stability of the Thus, the stability of 𝐵 depends on the value of 𝜆 1 ,
admissible equilibria. whereas the other two eigenvalues are negative. If
𝜆 1 > 0, that is, if 𝑎 < 1, then the equilibrium 𝐵
(i) Equilibrium 𝐴 = (0, 0, 0): the Jacobian matrix 𝐽
is a saddle point, therefore unstable; if 𝑎 > 1, 𝐵 is
applied at 𝐴 has the eigenvalues
asymptotically stable (a stable node); and if 𝑎 = 1, 𝐵
𝜆 1 = 1, 𝜆 2 = 0.6, 𝜆 3 = −0.5. (6) is nonhyperbolic, since 𝜆 1 = 0. The equilibrium point
𝐵 itself represents the absence of cancer and immune
As 𝜆 1 > 0 and 𝜆 2 > 0, 𝐴 is (unstable) saddle point, system cells and the permanence of the healthy cells,
which is coherent from the biological point of view, which tend to its carrying capacity 𝑦(𝑡) = 1. In this
since this point represents the simultaneous extinc- way, from the biological point of view the instability of
tion of normal cells, immune cells, and tumour cells. 𝐵 (case 𝑎 < 1) represents the existence of cancer cells
(ii) Equilibrium 𝐵 = (0, 1, 0): the eigenvalues of the and, consequently, the existence of immune system
matrix 𝐽 at this point are cells that is the persistence of the disease; on the
other hand, the stability of 𝐵 represents the absence
𝜆 1 = 1 − 𝑎, 𝜆 2 = −0.6, 𝜆 3 = −0.5. (7) of disease, which could be possible through a given
4 Abstract and Applied Analysis

treatment or by the action of the immune system. As (vi) Equilibrium 𝐺 = (0.132503, 0.668742, 0.1(8.675 −
the model considered does not admit treatment and 6.687𝑎)/𝑏): the Jacobian matrix 𝐽 calculated at the
includes the action of immune system cells, we expect point 𝐺 has the following characteristic polynomial
that these cells somehow combat the tumour cells,
thus enabling the cure, which justifies the possible 𝜆3 + 𝑎1 𝜆2 + 𝑎2 𝜆 + 𝑎3 = 0, (13)
stability of the equilibrium point 𝐵 (case 𝑎 > 1).
(iii) Equilibrium 𝐶 = (1, 0, 0): the Jacobian matrix 𝐽 at 𝐶 where 𝑎1 = 0.533748, 𝑎2 = 0.433476 − 0.426092𝑎, and
has the eigenvalues 𝑎3 = 0.152597 − 0.117636𝑎. Hence the linear stability
of 𝐺 depends on the value of parameter 𝑎.
𝜆 1 = −1, 𝜆 2 = −0.9, 𝜆 3 = 1.55. (8)
We observe that the equilibrium point 𝐺 plays an impor-
As 𝜆 3 > 0, 𝐶 is a saddle point; therefore it is unstable, tant role in our analysis, since it represents the coexistence
which is expected from the biological standpoint, of the three types of cells considered in the model. Thus a
since this equilibrium point represents the extinction detailed analysis of its stability will be presented here. The
of healthy and immune cells, remaining only tumour following result will be used in such analysis (for a proof, see
cells. [10]).
(iv) Equilibrium 𝐸 = (0.132503, 0, (0.867497/𝑏)): the Lemma 1. Consider the cubic polynomial
Jacobian matrix 𝐽 calculated at 𝐸 has the eigenvalues
𝑝 (𝑥) = 𝑥3 + 𝑎𝑥2 + 𝑏𝑥 + 𝑐, (14)
𝜆 1 = 0.401245, 𝜆 2 = −0.6625 + 0.613𝑖,
(9)
𝜆 3 = −0.6625 − 0.613𝑖, with real coefficients.

from which follows that 𝐸 is a saddle-focus. As 𝜆 1 > (a) Then (Routh-Hurwitz Criterion) 𝑝 is stable (i.e., all its
0, the equilibrium 𝐸 is unstable, which could be roots have negative real part) if and only if
possible from the biological point of view, since this
equilibrium represents the extinction of healthy cells 𝑎, 𝑏, 𝑐 > 0, 𝑎𝑏 − 𝑐 > 0. (15)
and the persistence of tumour cells and immune
system cells. (b) If 𝑐 > 0 and 𝑎𝑏 − 𝑐 < 0, then 𝑝 has two roots with
(v) Equilibrium 𝐹 = (2(𝑎 − 1)/(5𝑎 − 2), 3/(5𝑎 − 2), 0): the positive real part and one negative real root.
Jacobian matrix 𝐽 at this point has two eigenvalues of (c) If 𝑎, 𝑏, 𝑐 > 0 and 𝑎𝑏 − 𝑐 = 0, then 𝑝 has two complex
the form conjugate roots with zero real part and one negative real
root.
𝛼 + √𝛽
𝜆1 = ,
(7𝑎 − 3) (5𝑎 − 2) Applying the previous lemma to the polynomial (13),
(10) we can obtain the equilibrium 𝐺 as a candidate for the
𝛼 − √𝛽
𝜆2 = occurrence of a Hopf bifurcation in system (1). We will prove
(7𝑎 3) (5𝑎 − 2)
− that this bifurcation indeed occurs at the point 𝐺 in the next
section. Before doing this, let us remember some results about
with the Hopf bifurcation theory.
𝛼 = −7.5𝑎2 + 4.5𝑎 − 0.6,
2.2. Outline of the Hopf Bifurcation Theory and the Projection
𝛽 = 5525𝑎4 − 11850𝑎3 + 9025𝑎2 − 2880𝑎 + 324. Method. This section is a brief review of the projection
(11) method described in [8] for the calculation of the first
Lyapunov coefficient associated with the well-known Hopf
The third eigenvalue is given by bifurcation, denoted by 𝑙1 .
Consider the differential equation
0.1 (124 − 416𝑎 + 247𝑎2 )
𝜆3 = . (12) x󸀠 = 𝑓 (x, 𝜁) , (16)
(8 − 34𝑎 + 35𝑎2 )
where x ∈ R3 and 𝜁 ∈ R3 are, respectively, vectors represent-
Recalling that 𝐹 is biologically admissible if 𝑎 > 1, we
ing phase variables and control parameters. Assume that 𝑓 is
have 𝜆 1 > 0. Hence 𝐹 is a saddle point. The instability
of 𝐹 represents the biological fact that, for the set of of class 𝐶∞ in R3 × R3 . Suppose that (16) has an equilibrium
parameters adopted, tumour cells and healthy cells point x = x0 at 𝜁 = 𝜁0 and, denoting the variable x − x0 also
cannot coexist without the presence of immune cells; by x, write
that is, the tumour cells actually activate the immune
system cells. 𝐹 (x) = 𝑓 (x, 𝜁0 ) (17)
Abstract and Applied Analysis 5

as The first Lyapunov coefficient 𝑙1 is defined by

1 1
𝐹 (x) = 𝐴x + 𝐵 (x, x) + 𝐶 (x, x, x)
2 6 1
1 1 𝑙1 = Re 𝐺21 , (24)
+ 𝐷 (x, x, x, x) + 𝐸 (x, x, x, x, x) 2
24 120
(18)
1
+ 𝐾 (x, x, x, x, x, x) where 𝐺21 = ⟨𝑝, H21 ⟩ and H21 = 𝐶(𝑞, 𝑞, 𝑞) + 𝐵(𝑞, ℎ20 ) +
720 2𝐵(𝑞, ℎ11 ).
1 A Hopf point (x0 , 𝜁0 ) of system (16) is an equilibrium point
+ 𝐿 (x, x, x, x, x, x, x) + 𝑂 (‖x‖8 ) ,
5040 where the Jacobian matrix 𝐴 has a pair of purely imaginary
where 𝐴 = 𝑓x (0, 𝜁0 ) and, for 𝑖 = 1, 2, 3, eigenvalues 𝜆 2,3 = ±𝑖𝜔0 , 𝜔0 > 0 and the other eigenvalue
󵄨󵄨 𝜆 1 ≠ 0. From the Center Manifold Theorem, at a Hopf
3
𝜕2 𝐹𝑖 (𝜉) 󵄨󵄨󵄨
𝐵𝑖 (x, y) = ∑ 󵄨󵄨 𝑥𝑗 𝑦𝑘 , point a two-dimensional center manifold is well-defined, and
𝜕𝜉𝑗 𝜕𝜉𝑘 󵄨󵄨󵄨 it is invariant under the flow generated by (16) and can
𝑗,𝑘=1 󵄨𝜉=0
󵄨󵄨 (19) be continued with arbitrary high class of differentiability to
3
𝜕3 𝐹𝑖 (𝜉) 󵄨󵄨󵄨 nearby parameter values (see [8]).
𝐶𝑖 (x, y, z) = ∑ 󵄨󵄨 𝑥𝑗 𝑦𝑘 𝑧𝑙 ,
𝜕𝜉 𝜕𝜉 𝜕𝜉 󵄨󵄨
𝑗,𝑘,𝑙=1 𝑗 𝑘 𝑙 󵄨󵄨𝜉=0
A Hopf point is called transversal if the parameter
dependent complex eigenvalues cross the imaginary axis with
and so on for 𝐷𝑖 , 𝐸𝑖 , 𝐾𝑖 , and 𝐿 𝑖 . nonzero derivative. In a neighborhood of a transversal Hopf
Suppose that (x0 , 𝜁0 ) is an equilibrium point of (16) point with 𝑙1 ≠0 the dynamic behaviour of the system (16),
where the Jacobian matrix 𝐴 has a pair of purely imaginary reduced to the family of parameter-dependent continuations
eigenvalues 𝜆 2,3 = ±𝑖𝜔0 , 𝜔0 > 0, and admits no other of the center manifold, is orbitally topologically equivalent to
eigenvalue with zero real part. Let 𝑇𝑐 be the generalized the following complex normal form:
eigenspace of 𝐴 corresponding to 𝜆 2,3 . By this it is meant the
largest subspace invariant by 𝐴 on which the eigenvalues are
𝜆 2,3 . 𝑤󸀠 = (𝜂 + 𝑖𝜔) 𝑤 + 𝑙1 𝑤 |𝑤|2 , (25)
Let 𝑝, 𝑞 ∈ C3 be vectors such that
3
where 𝑤 ∈ C and 𝜂, 𝜔, and 𝑙1 are real functions having
𝐴𝑞 = 𝑖𝜔0 𝑞, 𝐴⊤ 𝑝 = −𝑖𝜔0 𝑝, ⟨𝑝, 𝑞⟩ = ∑𝑝𝑖 𝑞𝑖 = 1,
derivatives of arbitrary higher order, which are continuations
𝑖=1
(20) of 0, 𝜔0 , and the first Lyapunov coefficient at the Hopf point.
See [8] for details. When 𝑙1 < 0 (𝑙1 > 0) one family of
where 𝐴⊤ is the transpose of the matrix 𝐴. Any vector 𝑦 ∈ 𝑇𝑐 stable (unstable) periodic orbits can be found on this family
can be represented as 𝑦 = 𝑤𝑞 + 𝑤 𝑞, where 𝑤 = ⟨𝑝, 𝑦⟩ ∈ of manifolds, shrinking to an equilibrium point at the Hopf
C. The two dimensional center manifold associated with the point.
eigenvalues 𝜆 2,3 = ±𝑖𝜔0 can be parameterized by the variables Based on these results, a numerical algorithm for calcu-
𝑤 and 𝑤 by means of an immersion of the form x = 𝐻(𝑤, 𝑤), lating the first Lyapunov coefficient of a system 𝑥̇= 𝑓(𝑥, 𝜇)
where 𝐻 : C2 → R3 has a Taylor expansion of the form with 𝑥 ∈ R3 and 𝜇 ∈ R can be obtained. This algorithm has
1 been used to make the calculations of the next section.
𝐻 (𝑤, 𝑤) = 𝑤𝑞 + 𝑤 𝑞 + ∑ ℎ𝑗𝑘 𝑤𝑗 𝑤𝑘 + 𝑂 (|𝑤|8 ) ,
2≤𝑗+𝑘≤7
𝑗!𝑘!
(21) 3. Hopf Bifurcation at the Point 𝐺
3
with ℎ𝑗𝑘 ∈ C and ℎ𝑗𝑘 = ℎ𝑘𝑗 . Substituting this expression into
By using the Hopf bifurcation theorem stated in [8] and
(16) we obtain the following differential equation:
the projection method presented at the same reference and
𝐻𝑤 𝑤󸀠 + 𝐻𝑤 𝑤󸀠 = 𝐹 (𝐻 (𝑤, 𝑤)) , (22) synthesized in the previous section we prove the following
result.
where 𝐹 is given by (17). The complex vectors ℎ𝑖𝑗 are
obtained solving the system of linear equations defined by the Theorem 2. Consider system (1) with the following parameter
coefficients of (22), taking into account the coefficients of 𝐹,
values: 𝑏 = 2, 𝑐 = 0.6, 𝑑 = 1.5, 𝑒 = 4.5, 𝑓 = 1, 𝑔 = 0.2, and ℎ =
so that system (22), on the chart 𝑤 for a central manifold, is
0.5. If 𝑎 < 0.717455, then the equilibrium point 𝐺 is a stable
written as follows:
focus of system (1). If 𝑎 ≅ 0.717455, 𝐺 is a weak stable focus
1 1
𝑤󸀠 = 𝑖𝜔0 𝑤 + 𝐺21 𝑤 |𝑤|2 + 𝐺32 𝑤 |𝑤|4 (or a vague attractor). On the other hand, for 𝑎 > 0.717455,
2 12 and close to this value, the equilibrium 𝐺 becomes an unstable
(23)
1 focus and a small stable limit cycle arises around 𝐺. In short,
+ 𝐺43 𝑤 |𝑤|6 + 𝑂 (|𝑤|8 ) ,
144 a supercritical Hopf bifurcation occurs at the equilibrium point
with 𝐺𝑗𝑘 ∈ C. 𝐺 for the critical value 𝑎 = 0.717455.
6 Abstract and Applied Analysis

Proof. Considering the characteristic polynomial of the Jaco- 4. Period-Doubling Bifurcations

bian matrix 𝐽 calculated at the point 𝐺, given by (13), we have and Chaotic Dynamics
(1) 𝑎1 > 0;
For 𝑎 close to and greater than the critical value 𝑎 = 0.717455,
(2) 𝑎2 > 0 if 𝑎 < 1.017; for which the Hopf bifurcation at the equilibrium 𝐺 occurs,
(3) 𝑎3 > 0 if 𝑎 < 1.297; we have a stable limit cycle of period 𝑇 (see Figure 3);
(4) 𝑎1 𝑎2 − 𝑎3 > 0 if 𝑎 < 0.717455. increasing the value of 𝑎, we numerically found that this limit
cycle splits into a period 2𝑇 periodic orbit (see Figure 4).
The necessary condition for the point 𝐺 being biologically When the value of 𝑎 increases a bit more, a 4𝑇 periodic orbit
admissible is 𝑎 < 1.297; therefore 𝑎3 is always positive. If 𝑎 < is observed (see Figure 5); increasing further the value of 𝑎,
0.717455, then 𝑎2 > 0 and 𝑎1 𝑎2 − 𝑎3 > 0; then, according a series of 𝑛 duplications and a limit cycle of period 2𝑛 𝑇
to the Routh-Hurwitz Criterion, 𝐺 is a stable focus. On the are observed, until a chaotic attractor is formed for 𝑎 ≅ 1
other hand, if 𝑎 > 0.717455, then 𝑎1 𝑎2 − 𝑎3 < 0 and 𝑎3 > 0. (see Figure 6(a)). This type of bifurcation is called a period-
Hence, from item (b) of Lemma 1 it follows that (13) has two doubling or “flip bifurcation.” System (1) presents therefore an
roots with positive real part and a negative real root. Thus 𝐺 infinite cascade of flip bifurcations with a finite accumulation
is an unstable focus, remaining unstable for 𝑎2 > 0 or 𝑎2 < 0. point; the associated dynamics at this limit point is chaotic;
If 𝑎 = 0.717455, then 𝑎1 , 𝑎2 , 𝑎3 > 0 and 𝑎1 𝑎2 − 𝑎3 = that is, the cascade of period-doubling bifurcations is a
0; hence from item (c) of Lemma 1, the polynomial (13) continuous transition from periodic solutions to chaos (see
has two complex conjugate roots with zero real part and a Figures 3–6).
negative real root, which is one of the conditions for the Figure 6(b) shows the sensitivity to initial conditions of
occurrence of the Hopf bifurcation. In order to guarantee that two solutions of system (1) with very close initial conditions,
the Hopf bifurcation actually occurs at the equilibrium 𝐺 and confirming the chaotic behaviour of the system, for these
determine the stability of the limit cycle originated from it, parameter values.
we have to calculate the first Lyapunov coefficient of system A bifurcation diagram confirming the occurrence of the
(1) associated with this point. Using the projection method period-doubling cascade mentioned above, culminating in
presented in the previous section, we obtained the value for chaos, is shown in Figure 7.
the first Lyapunov coefficient of system (1) at the point 𝐺:
𝑙1 = −1.854776782 ⋅ 108 . (26) 4.1. The Elimination of Tumour and Immune System Cells after
the Chaotic Dynamics. The chaotic dynamics occurs when
To finish the proof, the transversality condition remains the value of 𝑎 is approximately 1. In the adimensional system
to be checked. Using software MAPLE, we calculate the (1) the parameter 𝑎, which represents the interaction between
derivative of the Jacobian matrix 𝐴(𝑎) with respect to the tumour cells and healthy cells, can be obtained in terms of the
parameter 𝑎 applying the critical value 𝑎 = 0.717455 from parameters in the original model (see [6]), given by
which we find
−0.668742 −0.132503 0 𝑑𝑇 𝑇
𝑑𝐴 = 𝑟1 𝑇 (1 − ) − 𝑎̃𝑇𝐻 − ̃𝑏𝑇𝐸,
(0.717455) = ( 0 0 0) . (27) 𝑑̃𝑡 𝑘1
𝑑𝑎 0 0 0
𝑑𝐻 𝐻 ̃
Solving the equation = 𝑐̃𝐻 (1 − ) − 𝑑𝑇𝐻, (30)
𝑑̃𝑡 𝑘2
𝑑𝛾 𝑑𝐴
(0.717455) = Re ⟨𝑝, (0.717455) 𝑞⟩ , (28) 𝑑𝐸 𝑒̃𝑇𝐸
𝑑𝑎 𝑑𝑎 = ̃ − ̃ℎ𝐸,
− 𝑔𝑇𝐸
𝑑̃𝑡 𝑇 + 𝑓̃
we obtain
𝑑𝛾 in the following way:
(0.717455) = 9.18 ⋅ 107 ≠0. (29)
𝑑𝑎
𝑎̃𝑘2
This ends the proof of Theorem 2. 𝑎= , (31)
𝑟1
The numerical simulations presented in Figures 1 and 2
illustrate the results stated in Theorem 2. from which we have
The existence of a stable limit cycle implies the occurrence 𝑎̃𝑘2
of a periodic behaviour of system (1); that is, from the 𝑟1 = . (32)
𝑎
biological point of view, healthy cells, tumour cells, and
immune system cells can coexist. Thus, for small values of parameter 𝑎 we have large values of
In the next section, by studying the continuation of this 𝑟1 , since 𝑘2 is the largest parameter of the original system,
limit cycle when the parameter 𝑎 moves away from the critical because it represents the carrying capacity of healthy cells.
value 𝑎 = 0.717455, we numerically find a cascade of period- Then the value of 𝑟1 , which represents the growth rate of
doubling bifurcations initiated with the limit cycle created in tumour cells, is large when 𝑎 ≤ 1. This explains the fact
the Hopf bifurcation and leading to the creation of a chaotic that the tumour-free equilibrium point 𝐵 = (0, 1, 0) is
attractor, when 𝑎 tends to the limit value 𝑎 = 1. unstable, since, as the growth rate of the tumour cells is
Abstract and Applied Analysis 7

0.26

0.25

0.24

z(t)
0.23

0.22

0.21

0.2
0.634
0.117
0.127

0.654
0.137

0.674
0.147

0.694
0.1305
0.133 0.674
0.1355
x(t) y(t) x(t) y(t)
(a) (b)

Figure 1: Phase portrait of system (1) near the equilibrium point 𝐺, with initial conditions: (0.1325, 0.66, 0.19). (a) 𝐺 is a stable focus for
𝑎 = 0.6. Integration time: [0, 300]. (b) 𝐺 is a weakly stable focus for 𝑎 = 0.71. Integration time: [0, 3000].

0.54
0 54
0.5 0.54

0.44
0.4
.444 0.44

0.344
00.3 0.34
()
zz(t)
(t

z(t)

00.244
0.2 0.24

0.14
0.1
.1
.14 0.14

0.04
0..004
0.0
0. 0.04
04
0.4
00..33
0.3

0.4
0.3
00..4
0.4

0.4
0.14
0.114

.55
0.5

0.14

0.5
.6
0.6

0.6
0.77
0.24
24

0.7
0.24
.

0.88

0.8
0.344

0.34

y(t) y(t)
x(t) x(t)
(a) (b)

Figure 2: Phase portrait of system (1) near the equilibrium 𝐺, with initial conditions: (0.1325, 0.66, 0.19). (a) 𝐺 is an unstable focus for 𝑎 = 0.8.
Integration time: [50, 500]. (b) Limit cycle created with the Hopf bifurcation that occurs when 𝑎 passes through the critical value 𝑎 = 0.717455.
Integration time: [450, 500].

large, the tumour cells are increasing when 𝑎 ≤ 1. On the a decreasing in the number of effector cells of the immune
other hand, the effector cells of the immune system also system, since its variation rate is directly proportional to the
grow because their growth is proportional to the growth of number of tumour cells (see the last equation in (30)). In
tumour cells; then the number of effector cells is also large this way, it is possible that the interactions with the immune
when 𝑎 ≤ 1. Thus, we assume that this growth of tumour system eliminate the tumour cells. Thus, if there are no more
cells and effector cells, along with the interactions of healthy tumour cells, there will be no immune cells, leaving only
cells, results in a dispute whose apex occurs for 𝑎 = 1. This healthy cells, which makes the tumour-free equilibrium point
analysis suggests that the apex of the battle between cancer stable, allowing a possible cure of the disease for the set of
cells, healthy cells, and immune system cells is represented parameters considered in this analysis.
by the chaotic behaviour of the system. In Figure 8 we present some numerical simulations of the
In turn, for 𝑎 > 1, the value of 𝑟1 becomes smaller, solutions of system (1) with 𝑎 > 1, showing the stability of
so the number of tumour cells decreases, which implies in the equilibrium point 𝐵 = (0, 1, 0), illustrating the fact that,
8 Abstract and Applied Analysis

0.64
0.5

0.44
z(t) 0.4

0.3

x(t)
0.24

0.2
0.04

0.1
0.04
0.16
0.14

0.36
0.24

0.56
0.34

0.76
0.44
0.54

y(t) 200 300 400 500

x(t) t
(a) (b)
Figure 3: Periodic orbit of period 𝑇 created by the Hopf bifurcation that occurs at the equilibrium point 𝐺. (a) Limit cycle for system (1), for
𝑎 = 0.9. Integration time: [1500, 1600]. Initial conditions: (0.1325, 0.6687, 0.19), (0.13, 0.6, 0.2). (b) Periodic behaviour of the coordinate 𝑥(𝑡)
of the solution in (a). Integration time: [1500, 1600]. Initial conditions: (0.1325, 0.6687, 0.2).

0.84
0.6

0.64 0.5
z(t)

0.44 0.4
x(t)

0.3
0.24

0.2
0.04

0.1
0.04
0.08

0.28
0.24

0.48

0.68
0.44

0.88
0.64

400 500 600 700 800

y(t)
x(t) t
(a) (b)
Figure 4: Period-doubling bifurcation occurred with the periodic solution created in the Hopf bifurcation at the equilibrium 𝐺. (a) Periodic
orbit of period 2𝑇, for 𝑎 = 0.93. Integration time: [1500, 1600]. Initial conditions: (0.1325, 0.6687, 0.19), (0.13, 0.6, 0.2). (b) Periodic behaviour
of the coordinate 𝑥(𝑡) of the solution in (a). Integration time: [1500, 1600]. Initial conditions: (0.1325, 0.6687, 0.2).

for the set of parameters considered here, only healthy cells a three-dimensional cancer model, describing the interac-
survive after the battle between tumour cells and immune tions among healthy cells, tumour cells, and immune system
cells. cells. The analyzed model was proposed in [6], where the
existence of chaotic dynamics for a fixed set of parameter
5. Concluding Remarks values was proved. Here we performed a bifurcation analysis
In this work, using the concepts of qualitative theory, by varying one of the parameters involved in the system,
bifurcations, and chaos in dynamical systems, we analyze showing that a Hopf bifurcation occurs, giving rise to a stable
Abstract and Applied Analysis 9

0.7

0.8
0.6

0.5
0.55
z(t)

0.4

x(t)
0.3
0.3

0.2
0.05

0.1
0.04
0.08

0.28
0.24

0.48
0.44

0.68

0.88
0.64

200 300 400 500 600 700 800

x(t) y(t) t
(a) (b)

Figure 5: (a) Periodic orbit of period 4𝑇, for 𝑎 = 0.948. Integration time: [1500, 1600]. Initial conditions: (0.1325, 0.6687, 0.19), (0.13, 0.6, 0.2).
(b) Periodic behaviour of the coordinate 𝑥(𝑡) of the solution in (a). Integration time: [1500, 1600]. Initial conditions: (0.1325, 0.6687, 0.2).

0.84
0.9

0.8
0.64
0.7

0.6
0.44
z(t)

0.5
x(t)

0.24 0.4

0.3

0.04 0.2
0.04
0.04

0.1
0.24
0.24

0.44
0.44

0.64
0.64

0.84
0.84

800 900 1000 1100 1200 1300 1400 1500

y(t)
x(t) t
(a) (b)

Figure 6: Chaotic attractor created through the period-doubling bifurcations. Initial conditions: (0.1325, 0.6687, 0.19), (0.13, 0.6, 0.2). (a)
Chaotic attractor of system (1), with 𝑎 = 0.99. Integration time: [0, 1850]. (b) Sensitive dependence on initial conditions. Two solutions very
close diverge exponentially in time, for 𝑎 = 0.99. Integration time: [800, 1550].

limit cycle, which represents the possibility of coexistence of In a first glance one could argue that, from the biological
tumour cells, immune cells, and healthy cells. Analyzing the point of view, the chaotic dynamics could be related to
continuation of this limit cycle numerically, we find a cascade the process of uncontrolled growth of tumour cells. However,
of period-doubling bifurcations, leading to the creation of a we surprisingly observed that, for the set of parameter values
chaotic attractor, which is exactly the one shown to exist in considered the period-doubling bifurcations which occur to
[6]. the cancer cells lead to the growth of effector cells, since
10 Abstract and Applied Analysis

0.95 system cells vanish and the healthy cells tend to their carrying
capacity. It indicates that in the studied model the chaotic
0.9 dynamics could be considered as a kind of mechanism related
to a possible cure of the cancer, through the elimination
of tumour cells, at least for the set of parameter values
0.85
considered. This situation is obviously very optimistic but it is
mathematically possible in the studied model. Of course, the
0.8 model could be studied with other parameter values, in order
to detect other dynamical behaviour, which could represent
Max (y(t))

0.75 biological aspects different from the one presented here.

Indeed, after the acceptance of this paper for publication, we
knew about the paper by Lettelier et al. [11] concerning the
0.7
cancer model studied in this note, which can be of interest of
the readers.
0.65

Conflict of Interests
0.6
The authors declare that there is no conflict of interests
0.55 regarding the publication of this paper.
0.9 0.92 0.94 0.96 0.98
a
Acknowledgments
Figure 7: Bifurcation diagram for system (1). From left to right
we see bifurcations of period-2𝑘 : as the value of the parameter 𝑎
This work was partially supported by PROPG/UNESP. The
increases, successive bifurcations occur, resulting in chaos. third author is supported by CNPq-Brazil, under the Project
308315/2012-0 and by FAPESP-Brazil under the Process
2012/18413-7. The authors would like to thank the referees for
their valuable comments and suggestions, which helped them
to improve the presentation of the paper.
1.0

References
0.75
[1] R. P. Araujo and D. L. McElwain, “A history of the study of solid
z(t)

tumour growth: the contribution of mathematical modelling,”

00.55 Bulletin of Mathematical Biology, vol. 66, no. 5, pp. 1039–1091,
2004.
0.25
0.2
25 [2] R. Chammas, D. Silva, A. Wainstein, and K. Abdallah,
“Imunologia clinica das neoplasias,” in Imunologia Clı́nica na
Prática Médica, pp. 447–460, Atheneu, São Paulo, Brazil, 2009.
00..00
0.0 [3] W. Chang, L. Crowl, E. Malm, K. Todd-Brown, L. Thomas,
and M. Vrable, Analyzing Immunotherapy and Chemotherapy
0.0
0.05
0.25

0.3
0.5

0.55
0.75

0.8
1.0
1.25

1.05
1.3
1.5

of Tumors Through Mathematical Modeling, Department of

x(t) Mathematics, Harvey-Mudd University, Claremont, Calif, USA,
y(t)
2003.
Figure 8: Phase portrait of system (1) for 𝑎 = 1.2 showing [4] D. Kirschner and J. C. Panetta, “Modeling immunotherapy
the asymptotic stability of the point 𝐵 = (0, 1, 0). Integra- of the tumor—immune interaction,” Journal of Mathematical
tion time: [0, 500]. Initial conditions: (0.01, 0.01, 0.01), (0.01, 0.9, Biology, vol. 37, no. 3, pp. 235–252, 1998.
0.01), (1.5, 0.01, 0.01), (0.01, 0.2, 0.01), (0.01, 1.5, 0.01), (0.9, 0.01,
[5] M. Galach, “Dynamics of the tumor-immune system
0.01), (0.01, 0.1, 0.01), (0.01, 0.01, 0.1).
competition—the effect of time delay,” International Journal of
Applied Mathematics and Computer Science, vol. 13, no. 3, pp.
395–406, 2003.
their growth is proportional to the growth of tumour cells. [6] M. Itik and S. P. Banks, “Chaos in a three-dimensional cancer
In this way, it seems that, for the particular values of the model,” International Journal of Bifurcation and Chaos, vol. 20,
parameter considered here, the growth of both types of cells no. 1, pp. 71–79, 2010.
occurs simultaneously and is represented mathematically by [7] L. G. de Pillis and A. Radunskaya, “The dynamics of an opti-
the period-doubling cascade, which leads to the creation of a mally controlled tumor model: a case study,” Mathematical and
chaotic attractor. Furthermore, supposing the increasing val- Computer Modelling, vol. 37, no. 11, pp. 1221–1244, 2003.
ues of the parameter 𝑎, then we observe that after the chaotic [8] Y. A. Kuznetsov, Elements of Applied Bifurcation Theory,
dynamics (i.e., for 𝑎 > 1) both tumour cells and immune Springer, New York, NY, USA, 2004.
Abstract and Applied Analysis 11

[9] M. Itik and S. P. Banks, “On the structure of periodic orbits on a

simple branched manifold,” International Journal of Bifurcation
and Chaos, vol. 20, no. 11, pp. 3517–3528, 2010.
[10] L. S. Pontryagin, Ordinary Differential Equations, Addison-
Wesley, New York, NY, USA, 1962.
[11] C. Letellier, F. Denis, and L. A. Aguirre, “What can be learned
from a chaotic cancer model?” Journal of Theoretical Biology,
vol. 322, pp. 7–16, 2013.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2015, Article ID 167430, 10 pages
http://dx.doi.org/10.1155/2015/167430

Research Article
Synchronization Transition and Traffic Congestion in
One-Dimensional Traffic Model

Zhi Xin1,2 and Jian Xu1

1
School of Aerospace Engineering and Applied Mechanics, Tongji University, Shanghai 200092, China
2
Mathematics Science College, Inner Mongolia Normal University, Hohhot 010022, China

Correspondence should be addressed to Jian Xu; xujian@tongji.edu.cn

Received 23 June 2014; Accepted 28 August 2014

Academic Editor: Yongli Song

Copyright © 2015 Z. Xin and J. Xu. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

A nonlinear car-following model with driver’s reaction time is studied from the synchronization transition viewpoint. We investigate
the traffic congestion from the view of chaos system synchronization transition. Our result shows that the uniform flow corresponds
to the complete synchronization and the stop-and-go congested state corresponds to the lag synchronization of the vehicles. An
analytical criterion for synchronization manifolds stability is obtained; the analytical result and the numerical result are consistent.
The synchronization transition is also trigged by the driver’s reaction time. We analyze the car-following model by the use of the
nonlinear analysis method and derive the modified KdV equation describing the kink density wave.

1. Introduction uniform flow as well as stop-and-go waves; however, the

transition [15–28] between the two qualitatively different
Traffic jams are very annoying in our life and have been solutions is still not clarified.
studied by many physicists [1]; however, the precise mech- According to the car-following model theory, for each
anism for generation and propagation of traffic jams is individual vehicle, an equation of motion is the analogue of
still not clearly understood. Some traffic models have been Newton’s equation for each individual particle in a system of
established to study the dynamics of traffic flow, such as interacting classical particles. In Newtonian mechanics, the
car-following models, cellular automaton models, gas kinetic acceleration may be regarded as the response of the particle
models, and hydrodynamic models [2]. to the stimulus, and it receives in the form of force which
Car-following models or microscopic models describe includes both the external force and those arising from its
the behaviors of individual vehicles, which are described by interaction with the other particles in the system. In traffic
ordinary differential equation or delay differential equation. system, each driver can respond to the surrounding traffic
Recently, one more widely used car-following model is the conditions only by accelerating or decelerating the vehicle;
optimal velocity model [3, 4]; by this kind of model, the effects so, the basic philosophy of the car-following theories can be
of fluctuations of traffic jams are analyzed, the jamming summarized by the equation [Response]𝑘 − [Stimulus]𝑘 for
transitions and density waves have been invested [5–9], and the 𝑘th vehicle (𝑘 = 1, 2, . . .).
the bifurcation phenomena of the oscillating solution are also We know, for a chaotic system, the appearances and
explored by using numerical continuations techniques [10– robustness of chaotic synchronization states have been estab-
12]. Another popular car-following model is the intelligent lished by means of different coupling schemes [29], one of
driver model that was introduced by Treiber et al. [13, 14]; which is Pecora and Carroll method (unidirectional coupling
in this model, all parameters have a clear physical meaning; or drive-response coupling).
congested traffic states have been empirically observed and In this paper, we regard the car-following system as a
microscopically simulated. Many models are able to explain drive-response coupling chaotic system; we make use of
2 Abstract and Applied Analysis

chaotic systems synchronization transition [30–34] method We introduce the rescaled variables ̂ℎ = ℎ/ℎstop , ̂𝑡 =
to study the synchronization transition of microscopic move- V0 (𝑡/ℎstop ); the OV function becomes
ment of the vehicles and further reveal the relationship
between synchronization transition and traffic congestion. 0, ̂ℎ ≤ 1,
{
{
First, we use the long wave expansion method to give an { ̂ 3
𝑉 (̂ℎ) = { (ℎ − 1) (4)
analytical criterion for synchronization manifolds stability. In {
{ , ̂ℎ > 1.
order to verify the analytical result, we use DDE-BIFTOOL to 1 + (̂ℎ − 1)3
{
perform a two-parameter bifurcation analysis of the model;
due to the demands on the CPU time and the memory, the Model (1) and model (2) are transformed into
investigation was restricted to the setting of 𝑁 = 15 vehicles,
̇
and the analytical result is consistent with the numerical ℎ̂𝑘 (̂𝑡) = ̂V𝑘+1 (̂𝑡) − ̂V𝑘 (̂𝑡) ,
result. Second, we find different transition region in two- (5)
parameter plane and the vehicles display different motion. ̂ (̂ℎ𝑘 (̂𝑡 − 𝜏̂)) − ̂V𝑘 (̂𝑡)] .
̂V̇𝑘 (̂𝑡) = 𝛼̂ [𝑉
Third, we consider how the driver’s reaction time impacts on
the synchronization transition. Finally, we analyze the car- Here 𝛼̂ = 𝛼ℎstop /V0 , 𝜏̂ = 𝜏V0 /ℎstop ; in the remainder of
following model by the use of the nonlinear analysis method this paper we take the rescaled OV function and model. In
and we derive the modified KdV equation describing the kink order to express simplicity, the OV function and model are
density wave. expressed:
The layout of this paper is organized as follows. We
introduce the car-following model in Section 2 and derive
analytical criteria for synchronization manifolds stability in {0,
{ ℎ ≤ 1,
𝑉 (ℎ) = { (ℎ − 1)3 (6)
Section 3. Numerical simulation is in Section 4. Nonlinear { 3
, ℎ > 1,
analysis is in Section 5. In Section 6, we present the main { 1 + (ℎ − 1)
results.
ℎ̇𝑘 (𝑡) = V𝑘+1 (𝑡) − V𝑘 (𝑡) , (7)

2. Car-Following Model V̇𝑘 (𝑡) = 𝛼 [𝑉 (ℎ𝑘 (𝑡 − 𝜏)) − V𝑘 (𝑡)] . (8)

Here, we consider a single lane of traffic with identical 3. Synchronization Transition

vehicles; displacements and velocities are denoted as 𝑥𝑘 (𝑡)
and V𝑘 (𝑡), respectively; the spacing of adjacent vehicle is called The synchronization manifold of the system (7) and (8) is
headway (ℎ𝑘 (𝑡)). For the sake of simplicity, we suppose that 𝑁
vehicles are placed on a circular road of length (∑𝑁 𝑘=1 ℎ𝑘 = 𝐿). 𝑋1 (ℎ1 , V1 ) = 𝑋2 (ℎ2 , V2 ) = ⋅ ⋅ ⋅ = 𝑋𝑁 (ℎ𝑁, V𝑁) . (9)
The headway consists of the condition
The system ((7), (8)) possesses uniform flow equilibrium
ℎ̇𝑘 (𝑡) = V𝑘+1 (𝑡) − V𝑘 (𝑡) . (1) 𝐿
ℎ1 = ℎ2 = ⋅ ⋅ ⋅ = ℎ𝑁 = ℎ∗ = ,
𝑁 (10)
The acceleration of the 𝑘th vehicle is given by
∗ ∗
V1 = V2 = ⋅ ⋅ ⋅ = V𝑁 = V = 𝑉 (ℎ ) .
V̇𝑘 (𝑡) = 𝛼 [𝑉 (ℎ𝑘 (𝑡 − 𝜏)) − V𝑘 (𝑡)] . (2)
All vehicles reach complete synchronization. The stability
of the synchronization manifolds will change when the
The 𝛼 is called the sensitivity of the vehicles and the 𝜏 is
parameter ℎ∗ is varied. To see whether the synchronization
driver’s reaction time. The function 𝑉(ℎ) is optimal velocity
manifold of the system ((7), (8)) is stable or not, we add a
function; it has the following properties.
small perturbation
(a) 𝑉(ℎ) is a nonnegative, continuous, and monotone
ℎ𝑘 (𝑡) = ℎ∗ + ̃ℎ𝑘 (𝑡) , V𝑘 (𝑡) = V∗ + ̃V𝑘 (𝑡) . (11)
increasing function.
(b) 𝑉(ℎ) → V0 as ℎ → ∞.
According to (7) and (8), we can calculate a linearized
(c) There exists a jam headway ℎstop such that 𝑉(ℎ) = 0 equation with respect to the uniform flow (10). The linearized
for ℎ ∈ [0, ℎstop ]. equation is
̇
The dimensional parameter OV function [12] is ℎ̃𝑘 (𝑡) = ̃V𝑘+1 (𝑡) − ̃V𝑘 (𝑡) ,
(12)
{ 0, ℎ ≤ ℎstop , ̃V̇𝑘 (𝑡) = 𝛼𝑉󸀠 (ℎ∗ ) ̃ℎ𝑘 (𝑡 − 𝜏) − 𝛼̃V𝑘 (𝑡) .
{
{ 3
𝑉 (ℎ) = { (ℎ − ℎstop ) (3)
{ V0
{ 3
, ℎ > ℎstop . Indeed, we have ̃ℎ𝑘 (𝑡) = 𝑥̃𝑘+1 (𝑡) − 𝑥̃𝑘 (𝑡), 𝑥
̃̇𝑘+1 (𝑡) = ̃V𝑘 (𝑡),
3
{ ℎstop + (ℎ − ℎstop ) ̇
and ℎ̃ (𝑡) = ̃V (𝑡) − ̃V (𝑡).
𝑘+1 𝑘+1 𝑘
Abstract and Applied Analysis 3

5 5

4 4

3 3
𝛼 𝛼
2 2

1 1

Unstable Unstable
0 0
0 1 2 3 4 5 0 1 2 3 4 5
h∗ h∗
Analytical stability curve Analytical stability curve
Numerical continuation Hopf (j = 1) curve Numerical continuation Hopf (j = 1) curve
(a) (b)

Figure 1: Stability diagram in the (ℎ∗ , 𝛼) plane for (a) 𝜏 = 0.2 and for (b) 𝜏 = 1.

Orosz et al. have used the dynamical system approach If the number of the vehicles is large, to obtain the stable
and numerical continuation technique to give out the sta- boundary curve by the numerical continuations method
ble curves, but the numerical continuation method (DDE- requires a lot of time. In order to compare with numerical
BIFTOOL) needs a lot of CPU time and the memory; below, method, we now focus on the case of 𝑁 = 15 vehicles. For
we can find a simple method to analyze the linear stability of 𝜏 > 0, the stable boundary curve is shown in Figure 1. We use
the synchronization manifold [10]. the numerical continuation techniques (DDEBIFTOOL) to
get the blue curve and we get the analytical curve (red curve)
Physical Approach. Let 𝑥̃𝑘 (𝑡) = 𝐴𝑒𝑖𝑘𝜃+𝜆𝑡 , 𝜆 ∈ 𝐶, 𝜃 ∈ (0, 2𝜋); according to (16). We can find that the blue curve and the red
the characteristic equation is given by curve completely overlap, and this shows that our analytical
approach is correct. In Section 4 numerical simulations also
𝑁 𝑁
[𝜆2 + 𝛼𝜆 + 𝛼𝑉󸀠 (ℎ∗ ) 𝑒−𝜆𝜏 ] − [𝛼𝑉󸀠 (ℎ∗ ) 𝑒𝑖𝜃𝑘−𝜆𝜏 ] = 0. verify our conclusion.
(13) When the stability of the synchronization manifold is lost,
we can get 𝑥̃𝑘 (𝑡) = 𝐴 sin(𝜔𝑡+(2𝑗𝜋/𝑁)𝑘), ̃V𝑘 (𝑡) = 𝐴𝜔 cos(𝜔𝑡+
We know that the leading term of 𝜆 is of order 𝑖𝜃. When 𝑖𝜃 → (2𝑗𝜋/𝑁)𝑘), 2𝑗𝜋/𝑁 = 𝜃, and ̃V𝑘 (𝑡) = ̃V𝑘+1 (𝑡 − (𝑇per /𝑁)), and
0, 𝜆 → 0. We can derive the long wave expansion of 𝜆, which 𝑇per = 2𝑗𝜋/𝜔 is the period. For two adjacent vehicles, the
is determined order by order around 𝑖𝜃 ≈ 0. By expanding speed of the two vehicles has a certain time delay; the vehicles
will get lag synchronization.
𝜆 = 𝜆 1 (𝑖𝜃) + 𝜆 2 (𝑖𝜃)2 + 𝜆 3 (𝑖𝜃)3 + ⋅ ⋅ ⋅ , (14) Figure 1(a) shows that the synchronization manifold
is stable above the sable boundary curve and the vehi-
the first- and second-order terms of 𝜆 are obtained: cles can achieve complete synchronization; the synchro-
nization manifold is unstable below the stable boundary
𝜆 1 = 𝑉󸀠 (ℎ∗ ) , curve and the vehicles can achieve lag synchronization.
(15) The synchronization transition of the vehicles arises when
𝑉󸀠 (ℎ∗ ) 𝑉󸀠 (ℎ∗ )
𝜆2 = (1 − 2𝜏𝑉󸀠 (ℎ∗ ) − 2 ). the sensitivity (𝛼) is fixed and the headway (ℎ∗ ) varies
2 𝛼 from the right to the left through the critical curve.
Figure 1(b) shows that the unstable region is unbounded for
If 𝜆 2 is a positive value, the synchronization manifold is 𝜏 > 0.595.
stable; if 𝜆 2 is a negative value, the synchronization manifold
is unstable. Below we focus on the synchronization transition of the
By 𝜆 2 > 0, we get vehicles. If 0 ≤ 𝜏 ≤ 0.595, according to Figure 2, by adding a
vertical line across the top point ((1.79, 2.53)) of the critical
𝑉󸀠 (ℎ∗ ) curve, we can divide the whole region into three regions,
1 − 2𝜏𝑉󸀠 (ℎ∗ ) − 2 > 0. (16) region I, region II, and region III. Region I and region II are
𝛼
stable; region III is unstable. If 𝜏 > 0.595, the region is divided
Equation (16) is the analytical criterion for synchronization analogous to the situation when 0 ≤ 𝜏 ≤ 0.595, as is shown in
manifold stability. Figure 1(b).
4 Abstract and Applied Analysis

5 5. Nonlinear Analysis
We analyze the car-following model by the use of the
4 nonlinear analysis method. We derive the modified KdV
equation describing the kink density wave. We now consider
long-wavelength models in the traffic flow on coarse-grained
3 scales. The simplest way to describe the long-wavelength
𝛼 (1.79, 2.53) models is the long wave expansion. We consider the slowly
varying behavior at long wavelengths near the critical point
2
(ℎ𝑐∗ , 𝛼𝑐 ). We assume that the value of V𝑘 (𝑡) is determined
I II III
adiabatically by ℎ𝑘 . This statement is expressed by the relation
1 V𝑘 (𝑡) = V (ℎ𝑘 , ℎ𝑘+1 , ℎ𝑘−1 , . . .) . (17)
Unstable We define the headway as
0
0 1 2 3 4 5 ℎ𝑘 (𝑡) = ℎ𝑐∗ + 𝜀̃ℎ (𝑍, 𝑇) , (18)
∗
h
and 𝑍 and 𝑇 are a scaled position and time defined by
Figure 2: The synchronization transition area.
𝑍 = 𝜀 (𝑘 + 𝑏𝑡) , 𝑇 = 𝜀3 𝑝𝑡. (19)
Then, from (18), V𝑘 (𝑡) is expressed as
V𝑘 (𝑡) = V (𝜀̃ℎ (𝑍, 𝑇) , 𝜀̃ℎ (𝑍 + 2𝜀, 𝑇) , 𝜀̃ℎ (𝑍 − 2𝜀, 𝑇) , . . .) .
4. Numerical Simulation (20)

In Figure 3, we show the space-time evolutions of the vehicles’ Since 𝜀 is a small parameter, a Taylor expansion can be applied
velocity (V𝑖 ) and location (𝑥𝑖 ) plotted against time (𝑡). We to terms like ̃ℎ(𝑍 + 2𝜀, 𝑇) in (20). This leads to the expression
consider the velocity of the first and the second vehicles, the
V𝑘 (𝑡) = ̃V (𝜀̃ℎ (𝑍, 𝑇) , 𝜀2 𝜕𝑍̃ℎ (𝑍, 𝑇) , 𝜀3 𝜕𝑍2 ̃ℎ (𝑍, 𝑇) , . . .) . (21)
location of fifteen vehicles. First, we take a point in region III
where 𝜏 = 0.2, 𝛼 = 0.5, ℎ∗ = 4, V = 0.97, and 𝑁 = 15. The Here, ̃V remains unknown; we determine it by assuming the
corresponding velocity-time plot and the space-time plot are form
shown in Figures 3(a) and 3(b). In region III, the headway ∞
is larger and the length of the road is longer; all vehicles ̃V = ∑ 𝜀𝑛 ̃V(𝑛) (̃ℎ (𝑍, 𝑇) , 𝜕𝑍̃ℎ (𝑍, 𝑇) , 𝜕𝑍2 ̃ℎ (𝑍, 𝑇) , . . .) . (22)
travel with the same high speed. This is a freely moving phase. 𝑛=0
Second, we take a point in region II where 𝜏 = 0.2, 𝛼 = 0.5,
From the form of (22), we find the terms ̃V(𝑛) are expressed as
ℎ∗ = 2, V = 0.5, and 𝑁 = 15. The corresponding velocity-
time history plot and the space-time plot are shown in Figures ̃V(0) = ̃V0
3(c) and 3(d). The vehicles travel with stop-and-go state and
get lag synchronization; the velocity of the adjacent vehicles ̃V(1) = ̃V1 ̃ℎ
has a certain time delay. Last, we take a point in region I where
𝜏 = 0.2, 𝛼 = 0.5, ℎ∗ = 1.2, V = 0.008, and 𝑁 = 15. Figures 3(e) ̃V(2) = ̃V20 ̃ℎ2 + ̃V21 𝜕𝑍̃ℎ
and 3(f) show the velocity-time history plot and the space- (23)
time relation; the vehicles travel with a low speed close to ̃V(3) = ̃V30 ̃ℎ3 + ̃V31 𝜕𝑍̃ℎ2 + ̃V32 𝜕𝑍2 ̃ℎ
zero and get complete synchronization, that is, a uniformly
congested phase. ̃V(4) = ̃V40 ̃ℎ4 + ̃V41 𝜕𝑍̃ℎ3 + ̃V42 𝜕𝑍2 ̃ℎ2 + ̃V43 𝜕𝑍3 ̃ℎ
In Figure 4 the stability under different time delay influ-
ence is compared. We know that the unstable region is ..
.
bounded for 0 ≤ 𝜏 < 3/4√3 2 ≈ 0.595; when 𝜏 > 0.595,
the unstable region is unbounded. We find that the unstable where ̃V0 , ̃V1 , ̃V20 , ̃V21 , ̃V30 , ̃V31 , ̃V32 , ̃V40 , ̃V41 , ̃V42 , ̃V43 , and so forth
region becomes larger as the time delay increases. are constants, which are calculated in the following way. First,
In Figure 5 we consider the time delay effect on the we substitute (22) (using (23)) into (8), where V𝑘̇is expressed
synchronization transition. For 𝜏 = 0, the vehicles travel as a function of ̃ℎ,
with the same speed and they get complete synchronization.
V𝑘̇ = 𝜀2 ̃V1 𝑏𝜕𝑍̃ℎ + 𝜀3 (𝑏̃V20 ̃ℎ2 + 𝑏̃V21 𝜕𝑍2 ̃ℎ)
For 𝜏 = 0.4, the vehicles get lag synchronization and travel
with moderate amplitude. If the time delay is increased to
1, the vehicles travel with large amplitude. If the time delay + 𝜀4 (𝑝̃V1 𝜕𝑇̃ℎ + 𝑏̃V30 ̃ℎ3 + 𝑏̃V31 𝜕𝑍2 ̃ℎ2 + 𝑏̃V32 𝜕𝑍3 ̃ℎ)
continues to increase, the traffic is completely congested. (24)
According to the analytical stability condition (23), the + 𝜀5 (𝑝̃V20 𝜕𝑇̃ℎ2 + 𝑝̃V21 𝜕𝑍,𝑇̃ℎ + 𝑏̃V40 ̃ℎ4
stability of the synchronization manifold is strictly dependent
on the three parameters (𝛼, 𝜏, ℎ∗ ). +𝑏̃V41 𝜕𝑍2 ̃ℎ3 + 𝑏̃V42 𝜕𝑍3 ̃ℎ2 + 𝑏̃V43 𝜕𝑍4 ̃ℎ) + ⋅ ⋅ ⋅ .
Abstract and Applied Analysis 5

60

1.0
50

0.8
40

0.6 30

xi
1 , 2

0.4 20

0.2 10

0.0 0
0 50 100 150 200 250 0 50 100 150 200 250
t t
(a) (b)
30

25

1.0 20

0.8
15
xi

0.6
1 , 2

10
0.4
5
0.2

0.0 0
300 350 400 450 500 300 350 400 450 500 550 600
t t
(c) (d)

15

0.10

0.08 10
xi

0.06
1 , 2

0.04 5

0.02

0.00 0
0 200 400 600 800 1000 0 200 400 600 800 1000
t t
(e) (f)

Figure 3: Plot of the velocity and the location of vehicles against time in different regions. We consider the speeds of the first two cars and
the location of fifteen cars. The blue curve represents the speed of the first car and the black curve represents the speed of the second car. (a)
and (b): region III, 𝜏 = 0.2, 𝛼 = 0.5, ℎ∗ = 4, V = 0.97. (c) and (d): region II, 𝜏 = 0.2, 𝛼 = 0.5, ℎ∗ = 2, V = 0.5. (e) and (f): region I, 𝜏 = 0.2,
𝛼 = 0.5, ℎ∗ = 1.2, V = 0.008.
6 Abstract and Applied Analysis

5 5

4 4

3 3

𝛼 𝛼

2 2

1 1

Unstable Unstable
0 0
0 1 2 3 4 5 0 1 2 3 4 5
h∗ h∗

𝜏=0 𝜏 = 0.2 𝜏=1 𝜏=3

𝜏 = 0.1 𝜏 = 0.3 𝜏=2 𝜏=5
(a) (b)

Figure 4: Stability diagram on the (ℎ∗ , 𝛼) plane on delay impact. (a) 𝜏 = 0, 0.1, 0.2, 0.3. (b) 𝜏 = 1, 2, 3, 5.

According to (18), we get the expression about ℎ𝑘 (𝑡 − 𝜏) We expand the optimal velocity function at the turning point:

𝑉 (ℎ𝑘 (𝑡 − 𝜏))
ℎ𝑘 (𝑡 − 𝜏) = ℎ𝑐∗ + 𝜀̃ℎ (𝑍 (𝑡 − 𝜏) , 𝑇 (𝑡 − 𝜏)) . (25)
𝑉󸀠󸀠 (ℎ𝑐∗ ) 2
= 𝑉 (ℎ𝑐∗ ) + 𝑉󸀠 (ℎ𝑐∗ ) 𝑞 + 𝑞
2!
By expanding (25) to the fifth order of 𝜀, we get
󸀠󸀠󸀠
𝑉 (ℎ𝑐∗ ) 3 𝑉(4) (ℎ𝑐∗ ) 4 𝑉(5) (ℎ𝑐∗ ) 5
+ 𝑞 + 𝑞 + 𝑞,
3! 4! 5!
2
(𝑏𝜏) 2 ̃
ℎ𝑘 (𝑡 − 𝜏) = ℎ𝑐∗ + 𝜀̃ℎ − 𝜀2 𝑏𝜏𝜕𝑍̃ℎ + 𝜀3 𝜕 ℎ 𝑉 (ℎ𝑘 (𝑡 − 𝜏))
2 𝑍
(𝑏𝜏)3 3 ̃
− 𝜀4 𝑝𝜏𝜕𝑇̃ℎ − 𝜀4 𝜕 ℎ (26) = 𝑉 (ℎ𝑐∗ ) + 𝜀̃ℎ𝑉󸀠 (ℎ𝑐∗ )
6 𝑍
2 ̃ (𝑏𝜏)4 4 ̃ 𝑉󸀠󸀠 (ℎ𝑐∗ ) ̃ 2
+ 𝜀5 𝑏𝑝𝜏2 𝜕𝑍,𝑇 ℎ + 𝜀5 𝜕 ℎ. + 𝜀2 ( ℎ − 𝑏𝜏𝑉󸀠 (ℎ𝑐∗ ) 𝜕𝑍̃ℎ)
24 𝑍 2
󸀠󸀠󸀠
𝑉 (ℎ𝑐∗ ) ̃ 3
Let +𝜀 [3
ℎ − 𝑉󸀠󸀠 (ℎ𝑐∗ ) 𝑏𝜏̃ℎ𝜕𝑍̃ℎ
3!

(𝑏𝜏)2 󸀠 ∗ 2 ̃
ℎ𝑘 (𝑡 − 𝜏) = ℎ𝑐∗ + 𝑞, (27) + 𝑉 (ℎ𝑐 ) 𝜕𝑍ℎ]
2

(𝑏𝜏)3 󸀠 ∗ 3 ̃
+ 𝜀4 [− 𝑉 (ℎ𝑐 ) 𝜕𝑍ℎ − 𝑝𝜏𝑉󸀠 (ℎ𝑐∗ ) 𝜕𝑇̃ℎ
where 6

𝑏2 𝜏2 󸀠󸀠 ∗ 2 𝑏2 𝜏2 ̃ 󸀠󸀠 ∗ 2 ̃
+ 𝑉 (ℎ𝑐 ) (𝜕𝑍̃ℎ) + ℎ𝑉 (ℎ𝑐 ) 𝜕𝑍ℎ
(𝑏𝜏) 2 ̃ 2 2 2
𝑞 = 𝜀̃ℎ − 𝜀2 𝑏𝜏𝜕𝑍̃ℎ + 𝜀3 𝜕 ℎ − 𝜀4 𝑝𝜏𝜕𝑇̃ℎ
2 𝑍 1 2 󸀠󸀠󸀠 𝑉(4) (ℎ𝑐∗ ) ̃ 4
(28) − (̃ℎ) 𝑏𝜏𝑉 (ℎ𝑐∗ ) 𝜕𝑍̃ℎ + ℎ ] + ⋅⋅⋅ .
(𝑏𝜏)3 3 ̃ 2 ̃ (𝑏𝜏)4 4 ̃ 2 4!
− 𝜀4 𝜕𝑍ℎ + 𝜀5 𝑏𝑝𝜏2 𝜕𝑍,𝑇 ℎ + 𝜀5 𝜕 ℎ.
6 24 𝑍 (29)
Abstract and Applied Analysis 7

20

15
1.0

xi
0.8
10
0.6
1 , 2

0.4
5
0.2

0.0 0
0 100 200 300 400 500 0 100 200 300 400 500
t t
(a) (b)

20

1.0 15

0.8
xi

10
0.6
1 , 2

0.4
5
0.2

0.0 0
0 50 100 150 200 0 50 100 150 200
t t
(c) (d)

20

15
1.0
xi

0.8
10
0.6
1 , 2

0.4
5
0.2

0.0 0
0 50 100 150 200 0 50 100 150 200
t t
(e) (f)

Figure 5: Plot of the velocity and the location of vehicles against time in different regions. We consider the speeds of the first two cars and
the location of fifteen cars. The blue curve represents the speed of the first car and the black curve represents the speed of the second car. The
initial condition is set for 𝛼 = 2, ℎ∗ = 1.6, V = 0.18. (a) and (b): 𝜏 = 0.2. (c) and (d): 𝜏 = 0.4. (e) and (f): 𝜏 = 1.
8 Abstract and Applied Analysis

Then Substituting ̃V derived in this way into (7), we get a reduced

equation for ̃ℎ:
𝑉 (ℎ𝑘 (𝑡 − 𝜏)) − V𝑘 = 𝑉 (ℎ𝑐∗ ) − ̃V0 + 𝜀𝑝1 + 𝜀2 𝑝2 + 𝜀3 𝑝3 + 𝜀4 𝑝4 ,
(30) V𝑘̇ = 𝜀2 𝑏𝜕𝑍̃ℎ + 𝜀4 𝑝𝜕𝑇̃ℎ

where = V𝑘+1 − V𝑘

= ̃V (𝜀̃ℎ (𝑍 + 2𝜀, 𝑇) , 𝜀2 𝜕𝑍̃ℎ (𝑍 + 2𝜀, 𝑇) ,

𝑝1 = ̃ℎ𝑉󸀠 (ℎ𝑐∗ ) − ̃ℎ̃V1 ,

𝑉󸀠󸀠 (ℎ𝑐∗ ) ̃ 2 𝜀3 𝜕𝑍2 ̃ℎ (𝑍 + 2𝜀, 𝑇) , . . .)

𝑝2 = ℎ − 𝑏𝜏𝑉󸀠 (ℎ𝑐∗ ) 𝜕𝑍̃ℎ − ̃V20 ̃ℎ2 − ̃V21 𝜕𝑍̃ℎ,
2 − ̃V (𝜀̃ℎ (𝑍, 𝑇) , 𝜀2 𝜕𝑍̃ℎ (𝑍, 𝑇) , 𝜀3 𝜕𝑍2 ̃ℎ (𝑍, 𝑇) , . . .)
󸀠󸀠󸀠
𝑉 (ℎ𝑐∗ ) ̃ 3 (𝑏𝜏)2 󸀠 ∗ 2 ̃
𝑝3 = ℎ − 𝑉󸀠󸀠 (ℎ𝑐∗ ) 𝑏𝜏̃ℎ𝜕𝑍̃ℎ + 𝑉 (ℎ𝑐 ) 𝜕𝑍ℎ = 𝜀2 [̃V1 𝜕𝑍̃ℎ] + 𝜀3 [2 (̃V1 + ̃V21 ) 𝜕𝑍2 ̃ℎ]
3! 2
4
− ̃V30 ̃ℎ3 − ̃V31 𝜕𝑍̃ℎ2 − ̃V32 𝜕𝑍2 ̃ℎ, + 𝜀4 [ ̃V1 𝜕𝑍3 ̃ℎ + 2̃V21 𝜕𝑍3 ̃ℎ + 2̃V30 𝜕𝑍̃ℎ3
3
(𝑏𝜏)3 󸀠 ∗ 3 ̃
𝑝4 = − 𝑉 (ℎ𝑐 ) 𝜕𝑍ℎ − 𝑝𝜏𝑉󸀠 (ℎ𝑐∗ ) 𝜕𝑇̃ℎ 2
+4̃V31 (𝜕𝑍̃ℎ) + 2̃V32 𝜕𝑍3 ̃ℎ]
6
𝑏2 𝜏2 󸀠󸀠 ∗ 2 𝑏2 𝜏2 ̃ 󸀠󸀠 2 4 2
+ 𝑉 (ℎ𝑐 ) (𝜕𝑍̃ℎ) + ℎ𝑉 (ℎ𝑐 ) 𝜕𝑍2 ̃ℎ + 𝜀5 [ ̃V1 𝜕𝑍4 ̃ℎ + ̃V 𝜕4 ̃ℎ + ̃V30 (6̃ℎ2 𝜕𝑍2 ̃ℎ + 12̃ℎ(𝜕𝑍̃ℎ) )
2 2 3 3 21 𝑍
2
1 2 󸀠󸀠󸀠 𝑉(4) (ℎ𝑐∗ ) ̃ 4 + ̃V40 8̃ℎ3 𝜕𝑍̃ℎ + ̃V42 (4̃ℎ𝜕𝑍2 ̃ℎ + 4(𝜕𝑍̃ℎ) )
− (̃ℎ) 𝑏𝜏𝑉 (ℎ𝑐∗ ) 𝜕𝑍̃ℎ + ℎ − ̃V40 ̃ℎ4
2 4!
− ̃V41 𝜕𝑍̃ℎ3 − ̃V42 𝜕𝑍2 ̃ℎ2 − ̃V43 𝜕𝑍3 ̃ℎ. + 6̃V41 ̃ℎ2 𝜕𝑍2 ̃ℎ + 2̃V43 𝜕𝑍4 ̃ℎ] .
(31)
(33)
Then, collecting terms of equal order on both sides and
comparing coefficients of ̃ℎ, ̃ℎ2 , 𝜕𝑍̃ℎ, and so forth, we obtain By taking 𝑏 = 2̃V1 = 2𝑉󸀠 (ℎ𝑐∗ ), 𝜏 = 1/2𝑉󸀠 (ℎ𝑐∗ ) − 1/𝛼, the
̃V𝑛 : second-order term of 𝜀 and the third-order term of 𝜀 are
eliminated from (33); then we get
̃V0 = 𝑉 (ℎ𝑐∗ ) ,
1 4
󸀠 𝜕𝑇̃ℎ = [( ̃V1 + 2̃V21 + 2̃V32 ) 𝜕𝑍3 ̃ℎ + 2̃V30 𝜕𝑍̃ℎ3 ]
̃V1 = 𝑉 (ℎ𝑐∗ ) , 𝑝 3
(34)
𝑉󸀠󸀠 (ℎ𝑐∗ ) 𝜀
̃V20 = = 0, + [𝑐 𝜕4 ̃ℎ + 𝑐2 𝜕𝑍2 ̃ℎ3 + 𝑐3 𝜕𝑍̃ℎ4 + 𝑐4 𝜕𝑍̃ℎ𝜕𝑍̃ℎ2 ] ,
2 𝑝 1 𝑍

𝑏̃V1 where
̃V21 = −𝑏𝜏 − ,
𝛼
(2/3) ̃V1 + (4/3) ̃V21 + 2̃V32 + 2̃V43
󸀠󸀠󸀠 𝑐1 = ,
𝑉 (ℎ𝑐∗ ) 𝑝
̃V30 = ,
6 2 (̃V30 + ̃V41 ) 2̃V40
𝑐2 = , 𝑐3 = , (35)
̃V31 = 0, (32) 𝑝 𝑝
(𝑏𝜏)2 󸀠 ∗ 𝑏̃V 6 (̃V30 + ̃V41 )
̃V32 = 𝑉 (ℎ𝑐 ) − 21 , 𝑐4 = .
2 𝛼 𝑝
𝑉(4) (ℎ𝑐∗ ) Let
̃V40 = ,
4!
(4/3) ̃V1 + 2̃V21 + 2̃V32 2̃V30
1 󸀠󸀠󸀠 𝑏̃V 𝑔1 = , 𝑔2 = − ; (36)
̃V41 = − 𝑏𝜏𝑉 (ℎ𝑐∗ ) − 30 , 𝑝 𝑝
6 𝛼
̃V42 = 0, if we ignore the 𝑂(𝜀) terms in (34), then (34) is transformed
into a Kdv equation
1 𝑏̃V
̃V43 = − (𝑏𝜏)3 𝑉󸀠 (ℎ𝑐∗ ) − 32 . 𝜕𝑇̃ℎ = 𝑔1 𝜕𝑍3 ̃ℎ − 𝑔2 𝜕𝑍̃ℎ3 . (37)
6 𝛼
Abstract and Applied Analysis 9

In order to derive the regularized equation, we make the 6. Conclusion and Discussion
following transformation for (37):
In this paper, we study a car-following model with driver’s
reaction time. First, we investigate the traffic congestion
𝑇󸀠 = 𝑔1 𝑇, ̃ℎ = √ 𝑔1 ̃ℎ󸀠 ; (38) from the view of chaos system synchronization transition.
𝑔2 Our result shows that the uniform flow corresponds to the
complete synchronization and the stop-and-go congested
then (37) is transformed into a regularized Kdv equation state corresponds to the lag synchronization of the vehicles.
Second, we derive analytical criteria for synchronization
3 manifolds stability; the analytical result and the numerical
𝜕𝑇󸀠 ̃ℎ󸀠 = 𝜕𝑍3 ̃ℎ󸀠 − 𝜕𝑍 (̃ℎ󸀠 ) . (39)
result are consistent. Third, the synchronization manifold
stability regions are further classified; we find that the vehicles
If we ignore the 𝑂(𝜀) terms in (39), this is just the modified exhibit different states of motion in different regions, and the
Kdv equation with a kink solution as the desired solution synchronization transition can also be trigged by the driver’s
reaction time. Last, we analyze the car-following model by
̃ℎ󸀠 (𝑍, 𝑇󸀠 ) = √𝑐 tanh √ 𝑐 (𝑍 − 𝑐𝑇󸀠 ) . (40) the use of the nonlinear analysis method and we derive the
0
2 modified KdV equation describing the kink density wave.
Our result also shows that it is crucial to pay attention to
The selected value of propagation velocity 𝑐 for the kink the behavior of individual drivers in order to understand
solution is determined from the 𝑂(𝜀) term. the emergent behavior of traffic. The driver’s reaction time
Next, assuming that ̃ℎ󸀠 (𝑍, 𝑇󸀠 ) = ̃ℎ0󸀠 (𝑍, 𝑇󸀠 )+𝜀̃ℎ1󸀠 (𝑍, 𝑇󸀠 ), we is influenced by individual differences, personality, environ-
take into account the 𝑂(𝜀) correction: ment, and other factors. In further work, we need to study
how to better characterize the reaction time of the drivers.
1
𝑂 (𝜀) = [𝑐 𝜕4 ̃ℎ + 𝑐2 𝜕𝑍2 ̃ℎ3 + 𝑐3 𝜕𝑍̃ℎ4 + 𝑐4 𝜕𝑍̃ℎ𝜕𝑍̃ℎ2 ] ,
𝑝 1 𝑍 Conflict of Interests
3
𝑂 (𝜀) = 𝑐1󸀠 𝜕𝑍4 ̃ℎ󸀠 + 𝑐2󸀠 𝜕𝑍2 (̃ℎ󸀠 ) ̃󸀠 4 ̃󸀠 2
+ 𝑐3󸀠 𝜕𝑍(ℎ ) + 𝑐4󸀠 𝜕𝑍̃ℎ󸀠 𝜕𝑍 (ℎ ) , The authors declare that there is no conflict of interests
(41) regarding the publication of this paper.

where Acknowledgments
3
𝑔1 𝑔1 This research was supported by the National Natural Science
𝑐1󸀠 = 𝑐1 √ , 𝑐2󸀠 = 𝑐2 (√ ), Foundation of China (Grant no. 11032009) and National
𝑔2 𝑔2
Natural Science Foundation of China (Grant no. 11272236).
(42)
3
𝑔2 𝑔1
𝑐3󸀠 = 𝑐3 12 , 𝑐4󸀠 = 𝑐4 (√ ). References
𝑔2 𝑔2
[1] D. Helbing, “Traffic and related self-driven many-particle sys-
In order to determine the selected value of the propagation tems,” Reviews of Modern Physics, vol. 73, no. 4, pp. 1067–1141,
velocity 𝑐 for the kink solution (37), it is necessary to satisfy 2001.
the solvability condition [2] M. Gerwinski and J. Krug, “Analytic approach to the critical
density in cellular automata for traffic flow,” Physical Review E,
vol. 60, no. 1, pp. 188–196, 1999.
(̃ℎ0󸀠 (𝑍, 𝑇󸀠 ) , 𝑀 (̃ℎ0󸀠 (𝑍, 𝑇󸀠 ))) [3] M. Bando, K. Hasebe, A. Nakayama, A. Shibata, and Y.
+∞ (43) Sugiyama, “Dynamical model of traffic congestion and numer-
≡∫ 𝑑𝑍̃ℎ0󸀠 (𝑍, 𝑇󸀠 ) 𝑀 [̃ℎ0󸀠 (𝑍, 𝑇󸀠 )] = 0, ical simulation,” Physical Review E, vol. 51, no. 2, pp. 1035–1042,
−∞ 1995.
[4] M. Bando, K. Hasebe, K. Nakanishi, and A. Nakayama, “Anal-
where ysis of optimal velocity model with explicit delay,” Physical
Review E: Statistical Physics, Plasmas, Fluids, and Related Inter-
󸀠 󸀠
𝑀 [̃ℎ0󸀠 (𝑍, 𝑇󸀠 )] = 𝑐1󸀠 𝜕𝑍4 ̃ℎ󸀠 + 𝑐2󸀠 𝜕𝑍2 (̃ℎ3 ) + 𝑐3󸀠 𝜕𝑍 (̃ℎ4 ) disciplinary Topics, vol. 58, no. 5, pp. 5429–5435, 1998.
(44) [5] T. Nagatani, “Bunching of cars in asymmetric exclusion models
󸀠 for freeway traffic,” Physical Review E, vol. 51, no. 2, pp. 922–928,
+ 𝑐4󸀠 𝜕𝑍̃ℎ󸀠 𝜕𝑍(̃ℎ2 ) . 1995.
[6] T. Nagatani and K. Nakanishi, “Delay effect on phase transitions
Then, we can get in traffic dynamics,” Physical Review E, vol. 57, no. 6, pp. 6415–
6421, 1998.
̃ℎ (𝑍, 𝑇) = √ 𝑔2 𝑐 tanh √ 𝑐 (𝑍 − 𝑐𝑔 𝑇) .
[7] T. Nagatani, “Thermodynamic theory for the jamming transi-
1 (45) tion in traffic flow,” Physical Review E, vol. 58, no. 4, pp. 4271–
𝑔1 2
4276, 1998.
10 Abstract and Applied Analysis

[8] T. Nagatani, “Chaotic jam and phase transition in traffic flow [27] R. Jiang and Q.-S. Wu, “Toward an improvement over Kerner-
with passing,” Physical Review E, vol. 60, no. 2, pp. 1535–1541, Klenov-Wolf three-phase cellular automaton model,” Physical
1999. Review E, vol. 72, no. 6, Article ID 067103, 2005.
[9] T. Nagatani, “Traffic behavior in a mixture of different vehicles,” [28] H. Y. Lee, H.-W. Lee, and D. Kim, “Origin of synchronized traffic
Physica A: Statistical Mechanics and its Applications, vol. 284, no. flow on highways and its dynamic phase transitions,” Physical
1, pp. 405–420, 2000. Review Letters, vol. 81, no. 5, pp. 1130–1133, 1998.
[10] G. Orosz, R. E. Wilson, and B. Krauskopf, “Global bifurcation [29] S. Boccaletti, J. Kurths, G. Osipov, D. L. Valladares, and C. S.
investigation of an optimal velocity traffic model with driver Zhou, “The synchronization of chaotic systems,” Physics Reports
reaction time,” Physical Review E: Statistical, Nonlinear, and Soft A, vol. 366, no. 1-2, pp. 1–101, 2002.
Matter Physics, vol. 70, no. 2, Article ID 026207, 10 pages, 2004. [30] A. Balanov, N. Janson, D. Postnov, and O. Sosnovtseva, Synchro-
[11] G. Orosz, R. E. Wilson, R. Szalai, and G. Stépán, “Exciting traffic nization, Springer, Berlin, Germany, 2009.
jams: nonlinear phenomena behind traffic jam formation on [31] A. Pikovsky, M. Rosenblum, and J. Kurths, Synchronization—A
highways,” Physical Review E, vol. 80, no. 4, Article ID 046205, Universal Concept in Nonlinear Science, Cambridge University
2009. Press, Cambridge, UK, 2001.
[12] G. Orosz, R. E. Wilson, and G. Stépán, “Traffic jams: dynamics
[32] Y. A. Kuznetsov, Elements of Applied Bifurcation Theory,
and control,” Philosophical Transactions of the Royal Society of
Springer, Berlin, Germany, 2nd edition, 1997.
London A, vol. 368, no. 1928, pp. 4455–4479, 2010.
[13] M. Treiber, A. Hennecke, and D. Helbing, “Congested traffic [33] J. Hale, Theory of Functional Differential Equations, World
states in empirical observations and microscopic simulations,” Publishing Corporation, Beijing, China, 2003.
Physical Review E, vol. 62, no. 2, pp. 1805–1824, 2000. [34] P. Hövel, Control of Complex Nonlinear Systems with Delay,
[14] D. Helbing, A. Hennecke, and M. Treiber, “Phase diagram and Springer, Berlin, Germany, 2010.
traffic state in the presence of inhomogeneities,” Physical Review
Letters, vol. 82, pp. 4360–4363, 1999.
[15] B. S. Kerner, The Physics of Traffic, Springer, Berlin, Germany,
2004.
[16] B. S. Kerner and S. L. Klenov, “Deterministic microscopic three-
phase traffic flow models,” Journal of Physics A: Mathematical
and General, vol. 39, no. 8, pp. 1775–1809, 2006.
[17] B. S. Kerner and S. L. Klenov, “Microscopic theory of spatial-
temporal congested traffic patterns at highway bottlenecks,”
Physical Review E, vol. 68, no. 3, Article ID 036130, 2003.
[18] B. S. Kerner, S. L. Klenov, A. Hiller, and H. Rehborn, “Micro-
scopic features of moving traffic jams,” Physical Review E:
Statistical, Nonlinear, and Soft Matter Physics, vol. 73, no. 4,
Article ID 046107, 2006.
[19] B. S. Kerner and S. L. Klenov, “Phase transitions in traffic flow
on multilane roads,” Physical Review E, vol. 80, no. 5, Article ID
056101, 2009.
[20] D. Helbing and M. Treiber, “Gas-kinetic-based traffic model
explaining observed hysteretic phase transition,” Physical
Review Letters, vol. 81, no. 14, pp. 3042–3045, 1998.
[21] M. Treiber and D. Helbing, “Memory effects in microscopic
traffic models and wide scattering in flow-density data,” Physical
Review E: Statistical, Nonlinear, and Soft Matter Physics, vol. 68,
no. 4, Article ID 046119, pp. 461191–461198, 2003.
[22] H. K. Lee, R. Barlovic, M. Schreckenberg, and D. Kim,
“Mechanical restriction versus human overreaction triggering
congested traffic states,” Physical Review Letters, vol. 92, Article
ID 238702, 2004.
[23] L. C. Davis, “Comment on ‘analysis of optimal velocity model
with explicit delay’,” Physical Review E: Statistical, Nonlinear, and
Soft Matter Physics, vol. 66, no. 3, Article ID 038101, 2002.
[24] I. Gasser, G. Sirito, and B. Werner, “Bifurcation analysis of a
class of ‘car following’ traffic models,” Physica D: Nonlinear
Phenomena, vol. 197, no. 3-4, pp. 222–241, 2004.
[25] M. Treiber, A. Kesting, and D. Helbing, “Three-phase traffic
theory and two-phase models with a fundamental diagram in
the light of empirical stylized facts,” Transportation Research B,
vol. 44, no. 8-9, pp. 983–1000, 2010.
[26] R. Jiang, Q. Wu, and Z. Zhu, “Full velocity difference model for
a car-following theory,” Physical Review E, vol. 64, no. 1, Article
ID 017101, 2001.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2015, Article ID 183526, 8 pages
http://dx.doi.org/10.1155/2015/183526

Research Article
The Dynamics of an Impulsive Predator-Prey System with
Stage Structure and Holling Type III Functional Response

Zhixiang Ju,1 Yuanfu Shao,1 Xiaolan Xie,2 Xiangmin Ma,1 and Xianjia Fang1
1
School of Science, Guilin University of Technology, Guilin, Guangxi 541004, China
2
Institute of Information Science and Technology, Guilin University of Technology, Guilin, Guangxi 541004, China

Correspondence should be addressed to Yuanfu Shao; shaoyuanfu@163.com

Received 7 May 2014; Revised 12 September 2014; Accepted 14 September 2014

Academic Editor: Yonghui Xia

Copyright © 2015 Zhixiang Ju et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Based on the biological resource management of natural resources, a stage-structured predator-prey model with Holling type III
functional response, birth pulse, and impulsive harvesting at different moments is proposed in this paper. By applying comparison
theorem and some analysis techniques, the global attractivity of predator-extinction periodic solution and the permanence of this
system are studied. At last, examples and numerical simulations are given to verify the validity of the main results.

1. Introduction refer to [13]. Authors obtained the sufficient conditions of the

persistence for system (1).
In recent decades, with the increase of population and the However, it is well known that many evolution processes
development of science and technology, human acceler- are characterized by the fact that at certain moments their
ated the exploitation of natural resources. Many harvesting stage changes abruptly. For example, for IPM strategy on
predator-prey models have been studied. The dynamic rela- ecosystem, the predators are released periodically every time
tionship between predators and their preys has long been 𝑇, and periodic catching or spraying pesticide is also applied.
and will continue to be one of the dominant themes in Hence, the predator and prey experience a change of state
both ecology and mathematical ecology due to its universal abruptly. Consequently, it is natural to assume that these
existence and importance; see [1–6] and the references cited processes act in the form of impulse. Impulsive methods have
therein. On the other hand, in the real world, many species been applied in almost every field of applied sciences. For
usually go through two or even more life stages as they example, many population models assume that the popula-
proceed from birth to death. Thus, it is practical to introduce tions are born throughout the year, whereas it is often the case
the stage structure into predator-prey models; see [7–12]. that many species give birth only during a single period of the
For example, Wei and Wang [13] considered the following year; that is, births occur in regular pulses. Hence, the authors
predator-prey system with stage structure: Z. Xiang, D. Long, and X. Y. Song gave a single population
logistic model with birth pulse and impulsive harvesting at
𝑥̇(𝑡) = 𝑥 (𝑡) (𝑟1 − 𝑎11 𝑥 (𝑡) − 𝑎12 𝑦2 (𝑡)) , different moments as follows:
𝑦1̇(𝑡) = 𝛼𝑥 (𝑡) 𝑦2 (𝑡) − 𝛾𝑦1 (𝑡) − 𝛼𝑒−𝛾𝜏 𝑥 (𝑡 − 𝜏) 𝑦2 (𝑡 − 𝜏) , 𝑥̇(𝑡) = − 𝜇𝑥 (𝑡) , 𝑡 ≠(𝑛 + 𝑙) 𝑇, 𝑡 ≠(𝑛 + 1) 𝑇,

𝑦2̇(𝑡) = 𝛼𝑒−𝛾𝜏 𝑥 (𝑡 − 𝜏) 𝑦2 (𝑡 − 𝜏) − 𝑟2 𝑦2 (𝑡) − 𝑎22 𝑦22 (𝑡) , 𝑥 (𝑡)

Δ𝑥 (𝑡) = 𝑟𝑥 (𝑡) (1 − ), 𝑡 = (𝑛 + 𝑙) 𝑇, (2)
(1) 𝐾
Δ𝑥 (𝑡) = − 𝑝𝑥 (𝑡) , 𝑡 = (𝑛 + 1) 𝑇, 𝑛 ∈ 𝑍+ ,
where 𝑥(𝑡), 𝑦1 (𝑡), and 𝑦2 (𝑡) denote the densities of prey
population and immature and mature individual predators where 𝑥(𝑡) represents the density of the resource population
at time 𝑡, respectively. The meanings of all parameters may at time 𝑡. Parameter 𝑟 is the intrinsic growth rate, the positive
2 Abstract and Applied Analysis

constant 𝐾 is referred to as the environmental carrying (𝜙1 (𝑠) , 𝜙2 (𝑠) , 𝜙3 (𝑠)) ∈ 𝐶 ([−𝜏, 0] , 𝑅+3 ) ,
capacity, and parameter 𝜇 is the death rate of resource
population. Parameter 𝑝 denotes the harvest rate of resource 𝜙𝑖 (0) > 0, 𝑖 = 1, 2, 3,
population. For more details of the biological meaning of (4)
system (2), we can refer to [14–17].
In addition, in the description of dynamical interactions where 𝑥(𝑡) denotes the density of the prey and 𝑦1 (𝑡), 𝑦2 (𝑡)
on predators and their preys, a crucial element of all models represent the immature and mature predator densities,
is the classic definition of a predator’s functional response. A respectively. Parameters 𝑟, 𝐾, 𝛽, 𝜆, 𝑑1 , 𝑑2 , 𝑑3 , and 𝑑4 are
functional response of the predator to the prey density refers positive constants, where 𝑟 is the intrinsic growth rate of
to the change in the density of prey attached per unit time per the prey, 𝐾 denotes the capacity rate, concerned with the
predator as the prey density changes. For example, Zhang et maintaining of the evolution of the population, 𝛽 represents
al. [18] suggested a function the predation rate of predator, 𝜆 is the conversion rate
that translated into predator population increase, 𝑑1 , 𝑑2 ,
𝑥2 and 𝑑3 denote the death rate of prey, immature predator,
𝜑 (𝑥) = , (3)
𝑐 + 𝑥2 and mature predator, respectively, and 𝑑4 is the intrinsic-
called Holling type III. It is monotonic in the first quadrant, specific competition rate of the mature predator. Parameter
that is, if the prey population increases, then the consumption 𝜏 represents a constant time from immaturity to maturity.
rate of prey per predator will increase too. And √𝑐 is the half- Parameters 𝑝1 , 𝑝2 denote the harvesting rates of prey and
saturation constant. The field of research on the dynamics mature predator at 𝑡 = 𝑛𝑇, 𝑛 ∈ 𝑍+ , respectively, 𝑍+ =
of impulsive predator-prey model with functional response {1, 2, . . .}, and 𝑇 is the period of the impulsive effect.
seems to be a new increasingly interesting area, which draws It is well known that, in the sustainable development of
many scholars’ attention. natural resources, it is very important to study the sustainable
Moreover, by the picture of ocean’s food-chain, we know survival of species. So, in this paper, we aim to investigate the
that small fish can prey on fish larvae as a predator; also it can global attractivity of predator-extinction periodic solution
be eaten by the higher predator as a prey (see [19]). Hence, and the permanence of system (4). From the biological point
according to the nature of biological resource management, it of view, we only consider (4) in the biological meaning region
is interesting to investigate impulsive harvesting on prey and 𝐷 = {(𝑥 (𝑡) , 𝑦1 (𝑡) , 𝑦2 (𝑡)) : 𝑥 (𝑡) ≥ 0, 𝑦1 (𝑡) ≥ 0, 𝑦2 (𝑡) ≥ 0} .
mature predator (e.g., the small fish and higher predator in (5)
the ocean’s food-chain) simultaneously at some fixed time.
Based on the above discussion, we consider the stage- This paper is organized as follows. Firstly, some prelim-
structured predator-prey model with Holling type III func- inaries are given in Section 2. In Section 3, the sufficient
tional response, birth pulse, and impulsive harvesting at conditions for the global attractivity of predator-extinction
different moments as follows: periodic solution are obtained. The permanence of system
(4) is investigated in Section 4. In Section 5, we present some
𝑥2 (𝑡)
𝑥̇(𝑡) = −𝛽 𝑦 (𝑡) − 𝑑1 𝑥 (𝑡) , examples and simulations to illustrate our results. At last, a
𝑐 + 𝑥2 (𝑡) 2 brief conclusion is given in Section 6.
𝑥2 (𝑡)
𝑦1̇(𝑡) = 𝜆𝛽 𝑦 (𝑡) 2. Preliminaries
𝑐 + 𝑥2 (𝑡) 2
𝑥2 (𝑡 − 𝜏) In this section, some definitions and lemmas are introduced.
− 𝜆𝛽𝑒−𝑑2 𝜏 𝑦 (𝑡 − 𝜏) − 𝑑2 𝑦1 (𝑡) ,
𝑐 + 𝑥2 (𝑡 − 𝜏) 2 Let 𝑅+ = [0, ∞) and 𝑅+3 = {𝑥 ∈ 𝑅3 , 𝑥 ≥ 0}. Denote𝑓 =
𝑥2 (𝑡 − 𝜏) (𝑓1 , 𝑓2 , 𝑓3 )𝑇 to be the map defined by the right hand of system
𝑦2̇(𝑡) = 𝜆𝛽𝑒−𝑑2 𝜏 𝑦 (𝑡 − 𝜏) (4). Let 𝑉 : 𝑅+ × 𝑅+3 → 𝑅+ ; then 𝑉 is said to belong to class
𝑐 + 𝑥2 (𝑡 − 𝜏) 2
𝑉0 , if
− 𝑑3 𝑦2 (𝑡) − 𝑑4 𝑦22 (𝑡) ,
(i) 𝑉 is continuous in ((𝑛 − 1)𝑇, (𝑛 + 𝑙 − 1)𝑇] × 𝑅+3 and
𝑡 ≠(𝑛 + 𝑙 − 1) 𝑇, 𝑡 ≠𝑛𝑇,
((𝑛 + 𝑙 − 1)𝑇, 𝑛𝑇] × 𝑅+3 for each 𝑥 ∈ 𝑅+3 , 𝑛 ∈ 𝑁,
𝑥 (𝑡) lim(𝑡,𝑦) → ((𝑛+𝑙−1)𝑇+ ,𝑥) 𝑉(𝑡, 𝑦) = 𝑉((𝑛 + 𝑙 − 1)𝑇+ , 𝑥) and
𝑥 (𝑡+ ) = 𝑥 (𝑡) + 𝑟𝑥 (𝑡) (1 − ),
𝐾 lim(𝑡,𝑦) → (𝑛𝑇+ ,𝑥) 𝑉(𝑡, 𝑦) = 𝑉(𝑛𝑇+ , 𝑥) exists,
𝑦1 (𝑡+ ) = 𝑦1 (𝑡) , (ii) 𝑉 is locally Lipschitzian in 𝑥.
𝑦2 (𝑡+ ) = 𝑦2 (𝑡) ,
Definition 1. Let 𝑉 ∈ 𝑉0 , and (𝑡, V) ∈ ((𝑛−1)𝑇, (𝑛+𝑙−1)𝑇]×𝑅+3
𝑡 = (𝑛 + 𝑙 − 1) 𝑇,
and ((𝑛+𝑙−1)𝑇, 𝑛𝑇]×𝑅+3 , the upper right derivative of 𝑉(𝑡, V)
𝑥 (𝑡+ ) = (1 − 𝑝1 ) 𝑥 (𝑡) , with respect to the impulsive differential system (4) is defined
as
𝑦1 (𝑡+ ) = 𝑦1 (𝑡) ,
1
𝑦2 (𝑡+ ) = (1 − 𝑝2 ) 𝑦2 (𝑡) , 𝐷+ 𝑉 (𝑡, V) = lim+ sup [𝑉 (𝑡 + ℎ, V + ℎ𝑓 (𝑡, V)) − 𝑉 (𝑡, V)] .
ℎ→0 ℎ
𝑡 = 𝑛𝑇, (6)
Abstract and Applied Analysis 3

The solution of (4), denote by V = (𝑥(𝑡), 𝑦1 (𝑡), 𝑦2 (𝑡))𝑇 , is If 𝑡 = (𝑛 + 𝑙 − 1)𝑇, we have

a piecewise continuous function V : 𝑅+ → 𝑅+3 , V(𝑡) is 𝑥 (𝑡)
continuous on ((𝑛 − 1)𝑇, (𝑛 + 𝑙 − 1)𝑇] × 𝑅+3 and 𝑉 ((𝑛 + 𝑙 − 1) 𝑇+ ) = 𝑉 ((𝑛 + 𝑙 − 1) 𝑇) + 𝜆𝑟𝑥 (𝑡) (1 − )
𝐾
((𝑛 + 𝑙 − 1) 𝑇, 𝑛𝑇] × 𝑅+3 (𝑛 ∈ 𝑁, 0 ≤ 𝑙 ≤ 1) . (7) 𝜆𝑟𝐾
≤ 𝑉 ((𝑛 + 𝑙 − 1) 𝑇) + ,
4
Obviously the existence and uniqueness of solutions of (4) is (12)
guaranteed by the smoothness properties of 𝑓, which denotes
and if 𝑡 = 𝑛𝑇, it is clear that 𝑉(𝑛𝑇+ ) ≤ 𝑉(𝑛𝑇).
the map defined by the right hand of system (4).
Then, by Lemma 2.2 of [20], for all 𝑡 ≥ 0, we have
The following lemmas are useful for the proof of the main 𝜆𝑟𝐾 𝑒−𝑑(𝑡−𝑇)
results. 𝑉 (𝑡) ≤ 𝑉 (0+ ) 𝑒−𝑑1 𝑡 +
4 1 − 𝑒−𝑑𝑇
(13)
Lemma 2 (see [5, 7]). Consider the following differential 𝜆𝑟𝐾 𝑒𝑑𝑇 𝜆𝑟𝐾 𝑒𝑑𝑇
equation: + 󳨀→ , 𝑡 󳨀→ ∞.
4 𝑒𝑑𝑇 − 1 4 𝑒𝑑𝑇 − 1
𝑥̇(𝑡) = 𝑎𝑥 (𝑡 − 𝜏) − 𝑏𝑥 (𝑡) − 𝑐𝑥2 (𝑡) , (8) Therefore, there exists a positive constant

where 𝑎, 𝑏, 𝑐, and 𝜏 are positive constants and 𝑥(𝑡) > 0 for 𝜆𝑟𝐾 𝑒𝑑𝑇
𝑀= (14)
𝑡 ∈ [−𝜏, 0]. We have the following: 4 𝑒𝑑𝑇 − 1
such that 𝑥(𝑡) ≤ 𝑀/𝜆, 𝑦1 (𝑡) ≤ 𝑀, and 𝑦2 (𝑡) ≤ 𝑀 for 𝑡 large
(i) if 𝑎 < 𝑏, then lim𝑡 → +∞ 𝑥(𝑡) = 0; enough. This completes the proof.
(ii) if 𝑎 > 𝑏, then lim𝑡 → +∞ 𝑥(𝑡) = (𝑎 − 𝑏)/𝑐.
3. Global Attractivity of a Predator-Extinction
Lemma 3 (see [14]). Consider the following system:
Periodic Solution
𝑢̇(𝑡) = − 𝑤𝑢 (𝑡) , 𝑡 ≠(𝑛 + 𝑙 − 1) 𝑇, 𝑡 ≠𝑛𝑇, In this section, we will demonstrate the existence and global
𝑢 (𝑡) attractivity of a predator-extinction periodic solution, in
𝑢 (𝑡+ ) = 𝑢 (𝑡) + 𝑟𝑢 (𝑡) (1 − ), 𝑡 = (𝑛 + 𝑙 − 1) 𝑇, (9) which the predator individuals are entirely absent from the
𝐾 population permanently; that is, 𝑦1 (𝑡) = 0 and 𝑦2 (𝑡) = 0 for
𝑢 (𝑡+ ) = (1 − 𝑝) 𝑢 (𝑡) , 𝑡 = 𝑛𝑇. all 𝑡 ≥ 0.
Firstly, by Lemma 3, we can easily obtain the existence of
Then, system (9) has a positive periodic solution 𝑢∗ (𝑡) with predator-extinction periodic solution for system (4).
period 𝑇, which is globally stable, where
Theorem 5. System (4) has a predator-extinction periodic
𝑤𝑇 solution (𝑥∗ (𝑡), 0, 0) which is globally stable; that is, for 𝑡 ∈
{ 1+𝑟 𝑒
{
{ 𝑒𝑤𝑙𝑇 ( 𝐾− 𝐾) 𝑒−𝑤(𝑡−(𝑛−1)𝑇) , ((𝑛 − 1)𝑇, 𝑛𝑇] and any solution (𝑥(𝑡), 𝑦1 (𝑡), 𝑦2 (𝑡)) of system
{
{ 𝑟 (1 − 𝑝) 𝑟 (4), we have 𝑥(𝑡) → 𝑥∗ (𝑡) as 𝑡 → ∞, where
{
{
{
{
{ 𝑡 ∈ ((𝑛 − 1) 𝑇, (𝑛 + 𝑙 − 1) 𝑇] ,
∗
𝑢 (𝑡) = { 𝑤𝑇 { 1+𝑟 𝑒𝑑1 𝑇
{ 𝑒 1+𝑟 𝑒𝑤𝑇 {
{ 𝑒𝑑1 𝑙𝑇 ( 𝐾− 𝐾) 𝑒−𝑑1 (𝑡−(𝑛−1)𝑇) ,
{
{ {
{ 𝑟 (1 − 𝑝 ) 𝑟
{
{ ( 𝐾 − 𝐾) 𝑒−𝑤(𝑡−(𝑛+𝑙−1)𝑇) , {
{ 1
{
{ 1 − 𝑝 𝑟 (1 − 𝑝) 𝑟 {
{
{
{ { 𝑡 ∈ ((𝑛 − 1) 𝑇, (𝑛 + 𝑙 − 1) 𝑇] ,
{ 𝑡 ∈ ((𝑛 + 𝑙 − 1) 𝑇, 𝑛𝑇] , 𝑥∗ (𝑡) = { 𝑑 𝑇 𝑑 𝑇
{ 𝑒
{ 1
1+𝑟 𝑒 1
{
{ ( 𝐾− 𝐾) 𝑒−𝑑1 (𝑡−(𝑛+𝑙−1)𝑇) ,
(10) {
{ 1 − 𝑝 𝑟 (1 − 𝑝1 ) 𝑟
{
{ 1
{
if 𝑝 < 1 − (𝑒𝑤𝑇 /(1 + 𝑟)). { 𝑡 ∈ ((𝑛 + 𝑙 − 1) 𝑇, 𝑛𝑇] ,
(15)
Lemma 4. There is a positive constant M such that 𝑥(𝑡) ≤
𝑀/𝜆, 𝑦1 (𝑡) ≤ 𝑀, and 𝑦2 (𝑡) ≤ 𝑀 for every solution (𝑥(𝑡), 𝑦1 (𝑡), and 𝑥∗ (0+ ) = 𝑒𝑑1 𝑙𝑇 (((1 + 𝑟)/𝑟)𝐾 − (𝑒𝑑1 𝑇 /(1 − 𝑝1 )𝑟)𝐾).
𝑦2 (𝑡)) of system (4) with all 𝑡 sufficiently large, where 𝜆 is a Next, we give the conditions on the global attractivity of the
positive constant defined in system (4). predator-extinction periodic solution (𝑥∗ (𝑡), 0, 0) of the system
(4).
Proof. Define 𝑉(𝑡) = 𝜆𝑥(𝑡) + 𝑦1 (𝑡) + 𝑦2 (𝑡).
If 𝑡 ≠(𝑛 + 𝑙 − 1)𝑇, 𝑡 ≠𝑛𝑇, we let 𝑑 = min{𝑑1 , 𝑑2 , 𝑑3 }; then Theorem 6. The predator-extinction periodic solution (𝑥∗ (𝑡),
0, 0) of system (4) is globally attractive, if
𝐷+ 𝑉 (𝑡) + 𝑑𝑉 (𝑡) = −𝜆 (𝑑1 − 𝑑) 𝑥 (𝑡) − (𝑑2 − 𝑑) 𝑦1 (𝑡) (A1 ) 1 − 𝑝1 − (𝑒𝑑1 𝑇 /(1 + 𝑟)) > 0,
(11)
− (𝑑3 − 𝑑) 𝑦2 (𝑡) − 𝑑4 𝑦22 (𝑡) ≤ 0, (A2 ) 𝜆𝛽𝑒−𝑑2 𝜏 (𝜌2 /(𝑐 + 𝜌2 )) < 𝑑3 .
4 Abstract and Applied Analysis

Proof. Let (𝑥(𝑡), 𝑦1 (𝑡), 𝑦2 (𝑡)) be any solution of system (4); Consider the auxiliary system of (23) as follows:
from the first, the fourth, and the seventh equations of system
(4), we have 𝜌2
𝑧̇
3 (𝑡) = 𝜆𝛽 𝜀 − 𝑑2 𝑧3 (𝑡) . (24)
𝑐 + 𝜌2 2
𝑥̇(𝑡) ≤ − 𝑑1 𝑥 (𝑡) , 𝑡 ≠(𝑛 + 𝑙 − 1) 𝑇, 𝑡 ≠𝑛𝑇,
𝑥 (𝑡) Integrating (24), one can easily get 𝑧3 (𝑡) = 𝜆𝛽(𝜌2 /𝑑2 (𝑐 +
𝑥 (𝑡+ ) = 𝑥 (𝑡) + 𝑟𝑥 (𝑡) (1 − ), 𝑡 = (𝑛 + 𝑙 − 1) 𝑇, (16) 𝜌2 ))𝜀2 as 𝑡 → ∞. Then, for any small constant 𝜀3 > 0, there
𝐾
exists 𝑡2 > 0 such that
𝑥 (𝑡+ ) = (1 − 𝑝1 ) 𝑥 (𝑡) , 𝑡 = 𝑛𝑇.
Consider the auxiliary system of (16) as follows: 𝜌2
𝑦1 (𝑡) ≤ 𝑧3 (𝑡) ≤ 𝜆𝛽 𝜀 +𝜀 , 𝑡 > 𝑡2 . (25)
𝑑2 (𝑐 + 𝜌2 ) 2 3
𝑧̇
1 (𝑡) = − 𝑑1 𝑧1 (𝑡) , 𝑡 ≠(𝑛 + 𝑙 − 1) 𝑇, 𝑡 ≠𝑛𝑇,
𝑧1 (𝑡) According to the first, the fourth, and the seventh equations
𝑧1 (𝑡+ ) = 𝑧1 (𝑡) + 𝑟𝑧1 (𝑡) (1 − ), 𝑡 = (𝑛 + 𝑙 − 1) 𝑇, of (4), we have
𝐾
𝑥̇(𝑡) ≥ − (𝛽𝜉𝜀2 + 𝑑1 ) 𝑥 (𝑡) , 𝑡 ≠(𝑛 + 𝑙 − 1) 𝑇, 𝑡 ≠𝑛𝑇,
𝑧1 (𝑡+ ) = (1 − 𝑝1 ) 𝑧1 (𝑡) , 𝑡 = 𝑛𝑇.
(17) 𝑥 (𝑡)
𝑥 (𝑡+ ) = 𝑥 (𝑡) + 𝑟𝑥 (𝑡) (1 − ), 𝑡 = (𝑛 + 𝑙 − 1) 𝑇,
By Lemma 3 and condition (A1 ), we have 𝐾

𝑑1 𝑙𝑇 1 + 𝑟 𝑒𝑑1 𝑇 𝑥 (𝑡+ ) = (1 − 𝑝1 ) 𝑥 (𝑡) , 𝑡 = 𝑛𝑇,

{
{ 𝑒 ( 𝐾 − 𝐾) 𝑒−𝑑1 (𝑡−(𝑛−1)𝑇) ,
{
{ 𝑟 (1 − 𝑝1 ) 𝑟 (26)
{
{
{
{
∗ { 𝑡 ∈ ((𝑛 − 1) 𝑇, (𝑛 + 𝑙 − 1) 𝑇] , where 𝜉 = max{√𝑐/2𝑐, 𝜌/(𝑐 + 𝜌2 )}. Again, consider the
𝑧1 (𝑡) = { 𝑑 𝑇
{
{ 𝑒1
{ 1+𝑟 𝑒𝑑1 𝑇 auxiliary system of (26):
{
{ ( 𝐾 − 𝐾) 𝑒−𝑑1 (𝑡−(𝑛+𝑙−1)𝑇) ,
{ 1 − 𝑝1
{ 𝑟 (1 − 𝑝1 ) 𝑟 𝑧̇
{ 4 (𝑡) = − 𝑑1 𝑧4 (𝑡) , 𝑡 ≠(𝑛 + 𝑙 − 1) 𝑇, 𝑡 ≠𝑛𝑇,
{ 𝑡 ∈ ((𝑛 + 𝑙 − 1) 𝑇, 𝑛𝑇] ,
𝑧4 (𝑡)
(18) 𝑧4 (𝑡+ ) = 𝑧4 (𝑡) + 𝑟𝑧4 (𝑡) (1 − ), 𝑡 = (𝑛 + 𝑙 − 1) 𝑇,
𝐾
which is unique and positive periodic solution of system (17)
and is globally attractive. Applying comparison theorem of 𝑧4 (𝑡+ ) = (1 − 𝑝1 ) 𝑧4 (𝑡) , 𝑡 = 𝑛𝑇.
impulsive differential equation [21], there exist 𝑛1 ∈ 𝑍+ and a (27)
sufficiently small constant 𝜀1 > 0 such that
In view of Lemma 3, system (27) has a unique positive
𝑥 (𝑡) ≤ 𝑧1 (𝑡) ≤ 𝑧1∗ (𝑡) + 𝜀1 periodic solution as follows:
𝑒𝑑1 𝑇 𝑒𝑑1 𝑇 (19)
1+𝑟 Δ {
{ (𝛽𝜉𝜀2 +𝑑1 )𝑙𝑇 1 + 𝑟 𝑒(𝛽𝜉𝜀2 +𝑑1 )𝑇
≤ ( 𝐾− 𝐾) + 𝜀1 = 𝜌, { 𝑒 ( 𝐾 − 𝐾)
1 − 𝑝1 𝑟 (1 − 𝑝1 ) 𝑟 {
{ 𝑟 (1 − 𝑝1 ) 𝑟
{
{
{
{ −(𝛽𝜉𝜀2 +𝑑1 )(𝑡−(𝑛−1)𝑇)
for 𝑡 ≥ (𝑛1 − 1)𝑇. {
{ ×𝑒
{
{
,
From inequality (19), we can obtain {
{
{ 𝑡 ∈ ((𝑛 − 1) 𝑇, (𝑛 + 𝑙 − 1) 𝑇] ,
𝜌2 𝑧4∗ (𝑡) = { (𝛽𝜉𝜀 +𝑑 )𝑇 (28)
𝑦2̇(𝑡) ≤ 𝜆𝛽𝑒−𝑑2 𝜏 𝑦 (𝑡 − 𝜏) − 𝑑3 𝑦2 (𝑡) − 𝑑4 𝑦22 (𝑡) , {
{ 𝑒 2 1
1+𝑟 𝑒(𝛽𝜉𝜀2 +𝑑1 )𝑇
{
{ ( 𝐾− 𝐾)
𝑐 + 𝜌2 2 {
{
{ 1 − 𝑝1
{ 𝑟 (1 − 𝑝1 ) 𝑟
{
{
𝑡 ≠𝑛𝑇, {
{ ×𝑒−(𝛽𝜉𝜀2 +𝑑1 )(𝑡−(𝑛+𝑙−1)𝑇) ,
{
{
{
𝑦2 (𝑡+ ) = (1 − 𝑝2 ) 𝑦2 (𝑡) ≤ 𝑦2 (𝑡) , 𝑡 = 𝑛𝑇. { 𝑡 ∈ ((𝑛 + 𝑙 − 1) 𝑇, 𝑛𝑇] .
(20) Similarly, for any arbitrary small constant 𝜀1 > 0, there exists
Δ
Consider the auxiliary system of (20): 𝑡3 > 0 such that 𝑥(𝑡) ≥ 𝑧4∗ (𝑡) − 𝜀1 = 𝛿. Let 𝜀2 → 0; then,
𝜌2 𝑧4∗ (𝑡) → 𝑥∗ (𝑡), and
−𝑑2 𝜏
𝑧̇
2 (𝑡) = 𝜆𝛽𝑒 𝑧 (𝑡 − 𝜏) − 𝑑3 𝑧2 (𝑡) − 𝑑4 𝑧22 (𝑡) . (21)
𝑐 + 𝜌2 2 𝑥 (𝑡) ≥ 𝑥∗ (𝑡) − 𝜀1 . (29)
Applying Lemma 2, (A2 ), and comparison theorem, we have It follows from (29) and (19) that 𝑥(𝑡) → 𝑥∗ (𝑡) as 𝑡 → ∞.
𝑧2 (𝑡) → 0 as 𝑡 → ∞. Then, for any small constant 𝜀2 > 0, Let 𝜀2 , 𝜀3 → 0; then, we have 𝑦1 (𝑡) → 0, 𝑦2 (𝑡) → 0 as 𝑡
there exists 𝑡1 > 0, such that sufficiently large enough. The proof is complete.
𝑦2 (𝑡) ≤ 𝑧2 (𝑡) ≤ 𝜀2 , 𝑡 > 𝑡1 . (22)
By the second equation of system (4) and (22), we have 4. Permanence of System (4)
2
𝜌 In the real world, from the principle of ecosystem balance and
𝑦1̇(𝑡) ≤ 𝜆𝛽 𝜀 − 𝑑2 𝑦1 (𝑡) . (23)
𝑐 + 𝜌2 2 saving resources, we only need to control the predator under
Abstract and Applied Analysis 5

the economic threshold level and not to eradicate the predator for all 𝑡 > 𝑡4 . Consider the auxiliary system of (35) as follows:
totally. Thus, we focus on the permanence of system (4). 𝑀 ∗
First, we give the definition of permanence. 𝑧̇
5 (𝑡) = − (𝛽 𝑚 + 𝑑1 ) 𝑧5 (𝑡) , 𝑡 ≠(𝑛 + 𝑙 − 1) 𝑇, 𝑡 ≠𝑛𝑇,
𝜆𝑐 2
𝑧 (𝑡)
Definition 7. System (4) is said to be persistent if there 𝑧5 (𝑡+ ) = 𝑧5 (𝑡) + 𝑟𝑧5 (𝑡) (1 − 5 ) , 𝑡 = (𝑛 + 𝑙 − 1) 𝑇,
exist positive constants 𝑚 and 𝑀 such that every positive 𝐾
solution (𝑥(𝑡), 𝑦1 (𝑡), 𝑦2 (𝑡)) of system (4) satisfies 𝑚 ≤ 𝑧5 (𝑡+ ) = (1 − 𝑝1 ) 𝑧5 (𝑡) , 𝑡 = 𝑛𝑇.
𝑥(𝑡), 𝑦1 (𝑡), 𝑦2 (𝑡) ≤ 𝑀 for 𝑡 sufficiently large enough. (36)
Theorem 8. System (4) is permanent, if the following condi- By hypothesis (A4 ), we can obtain from Lemma 3 that
∗
tions hold:
{ ∗ 1+𝑟 𝑒(𝛽(𝑀/𝑐𝜆)𝑚2 +𝑑1 )𝑇
{
{ 𝑒(𝛽(𝑀/𝑐𝜆)𝑚2 +𝑑1 )𝑙𝑇 ( 𝐾− 𝐾)
(A3 ) 𝜆𝛽𝑒−𝑑2 𝜏 (𝑞2 /(𝑐 + 𝑞2 )) − 𝑑3 − 𝑀𝑑4 > 0, {
{ 𝑟 (1 − 𝑝1 ) 𝑟
{
{
{
{
∗
×𝑒−(𝛽(𝑀/𝑐𝜆)𝑚2 +𝑑1 )(𝑡−(𝑛−1)𝑇) ,
∗
(A4 ) 1 − 𝑝1 > 𝑒((𝛽𝑀/𝜆𝑐)𝑚2 +𝑑1 )𝑇 /(1 + 𝑟), {
{
{
{
{
{ 𝑡 ∈ ((𝑛 − 1) 𝑇, (𝑛 + 𝑙 − 1) 𝑇] ,
where 𝑀, 𝑚2∗ , and 𝑞 are defined in (14), (34), and (39), 𝑧5∗ (𝑡) = { (𝛽(𝑀/𝑐𝜆)𝑚∗ +𝑑 )𝑇 ∗
respectively. {𝑒
{ 2 1
1+𝑟 𝑒(𝛽(𝑀/𝑐𝜆)𝑚2 +𝑑1 )𝑇
{
{ ( 𝐾− 𝐾)
{
{ 1 − 𝑝1 𝑟 (1 − 𝑝1 ) 𝑟
{
{
Proof. It is obvious that the third equation of system (4) can {
{ ∗
{
{ ×𝑒−(𝛽(𝑀/𝑐𝜆)𝑚2 +𝑑1 )(𝑡−(𝑛+𝑙−1)𝑇) ,
be rewritten as follows: {
{
𝑥2 (𝑡) { 𝑡 ∈ ((𝑛 + 𝑙 − 1) 𝑇, 𝑛𝑇] ,
𝑦2̇(𝑡) = (𝜆𝛽𝑒−𝑑2 𝜏 − 𝑑3 − 𝑑4 𝑦2 (𝑡)) 𝑦2 (𝑡) (37)
𝑐 + 𝑥2 (𝑡)
(30) which is the unique positive periodic solution of system
−𝑑2 𝜏 𝑑 𝑡 𝑥2 (𝜃) (36) and is globally asymptotically stable. By the comparison
− 𝜆𝛽𝑒 ∫ 𝑦 (𝜃) 𝑑𝜃.
𝑑𝑡 𝑡−𝜏 𝑐 + 𝑥2 (𝜃) 2 theorem of impulsive differential equation, we know there
𝑡 exists 𝑡5 (> 𝑡4 + 𝜏) such that 𝑥(𝑡) ≥ 𝑧5∗ (𝑡) − 𝜀4 .
Define 𝑉(𝑡) = 𝑦2 (𝑡) + 𝜆𝛽𝑒−𝑑2 𝜏 ∫𝑡−𝜏 (𝑥2 (𝜃)/(𝑐 + 𝑥2 (𝜃)))𝑦2 (𝜃)𝑑𝜃. On the other hand, for all 𝑡 ≥ 𝑡5 , we have
By computation, we have ∗
{
{ 1+𝑟 𝑒(𝛽(𝑀/𝑐𝜆)𝑚2 +𝑑1 )𝑇 }
}
{
{ 𝐾 − 𝐾 }
}
𝑥2 (𝑡) { 𝑟 (1 − 𝑝1 ) 𝑟 }
𝑉̇(𝑡) = (𝜆𝛽𝑒−𝑑2 𝜏 − 𝑑3 − 𝑑4 𝑦2 (𝑡)) 𝑦2 (𝑡) . (31) ∗
𝑧5 (𝑡) ≥ {
𝑐 + 𝑥2 (𝑡) { }
𝑒(𝛽(𝑀/𝑐𝜆)𝑚2 +𝑑1 )𝑙𝑇 }
∗ ∗
{
{ 1+𝑟 𝑒(𝛽(𝑀/𝑐𝜆)𝑚2 +𝑑1 )𝑇 }
}
{( 𝐾− 𝐾) }
Applying Lemma 4, we have { 𝑟 (1 − 𝑝 1 ) 𝑟 (1 − 𝑝 1 ) }
(𝛽(𝑀/𝑐𝜆)𝑚2∗ +𝑑1 )𝑇
𝑥2 (𝑡) 1+𝑟 𝑒 Δ
𝑉̇(𝑡) > (𝜆𝛽𝑒−𝑑2 𝜏 − 𝑑3 − 𝑀𝑑4 ) 𝑦2 (𝑡) . (32) ≥ 𝐾− 𝐾 = 𝜎.
𝑐 + 𝑥2 (𝑡) 𝑟 (1 − 𝑝1 ) 𝑟
By hypothesis (A3 ), for the arbitrary small constant 𝜀4 > 0,
we have (38)
2 Thus,
(𝑞 + 𝜀4 )
𝜆𝛽𝑒−𝑑2 𝜏 2
> 𝑑3 + 𝑀𝑑4 . (33) Δ
𝑐 + (𝑞 + 𝜀4 ) 𝑥 (𝑡) ≥ 𝜎 − 𝜀4 = 𝑞, ∀𝑡 ≥ 𝑡5 . (39)

Let 𝑚2∗ be determined by the following equation: By (33), we have

𝜎2
1+𝑟 𝑒(𝛽(𝑀/𝑐𝜆)𝑚2 +𝑑1 )𝑇
∗
𝑐 (𝑑3 + 𝑀𝑑4 ) 𝜆𝛽𝑒−𝑑2 𝜏 > 𝑑3 + 𝑀𝑑4 . (40)
𝐾− 𝐾=√ . 𝑐 + 𝜎2
𝑟 (1 − 𝑝1 ) 𝑟 𝜆𝛽𝑒−𝑑2 𝜏 − 𝑑3 − 𝑀𝑑4 Hence, by (32) and (40), we have
(34)
𝜎2
Then, for any 𝑡4 > 0, it is impossible that 𝑦2 (𝑡) < 𝑚2∗
for all 𝑉̇(𝑡) ≥ (𝜆𝛽𝑒−𝑑2 𝜏 − 𝑑3 − 𝑀𝑑4 ) 𝑦2 (𝑡) , 𝑡 ≥ 𝑡5 . (41)
𝑐 + 𝜎2
𝑡 > 𝑡4 . Suppose that the claim is invalid; then, there exists
𝑡4 > 0 such that 𝑦2 (𝑡) < 𝑚2∗ for all 𝑡 > 𝑡4 . It follows from the Let 𝑦2𝑚 = min𝑡∈[𝑡1 ,𝑡1 +𝜏] 𝑦2 (𝑡). We will show that 𝑦2 (𝑡) ≥ 𝑦2𝑚 for
first equation of system (4) that all 𝑡 ≥ 𝑡5 . Otherwise, there exists a constant 𝛼 > 0 such that
𝑦2 (𝑡) ≥ 𝑦2𝑚 for 𝑡5 ≤ 𝑡 ≤ 𝑡5 + 𝜏 + 𝛼, 𝑦2 (𝑡5 + 𝜏 + 𝛼) ≥ 𝑦2𝑚 , and
𝑀 ∗ 𝑦2̇(𝑡5 + 𝜏 + 𝛼) < 0. However, by the third equation of system
𝑥̇(𝑡) ≥ − (𝛽 𝑚 + 𝑑1 ) 𝑥 (𝑡) , 𝑡 ≠(𝑛 + 𝑙 − 1) 𝑇, 𝑡 ≠𝑛𝑇,
𝑐𝜆 2 (4) and (40), we have
𝑥 (𝑡) 𝜎2
𝑥 (𝑡+ ) = 𝑥 (𝑡) + 𝑟𝑥 (𝑡) (1 − ), 𝑡 = (𝑛 + 𝑙 − 1) 𝑇, 𝑦2̇(𝑡5 + 𝜏 + 𝛼) ≥ (𝜆𝛽𝑒−𝑑2 𝜏 − 𝑑3 − 𝑀𝑑4 ) 𝑦2𝑚 > 0,
𝐾 𝑐 + 𝜎2
𝑥 (𝑡+ ) = (1 − 𝑝1 ) 𝑥 (𝑡) , 𝑡 = 𝑛𝑇, 𝑡 ≥ 𝑡5 ,
(35) (42)
6 Abstract and Applied Analysis

1.8 0.5

1.6 0.45
0.4
1.4
0.35
1.2
0.3
1
x 0.25

y1
0.8
0.2
0.6 0.15
0.4 0.1
0.2 0.05

0 0
0 200 400 600 800 1000 0 50 100 150 200 250 300
t t
(a) (b)
0.5
0.45
0.4
0.35
0.3
0.25
y2

0.2
0.15
0.1
0.05

0
0 50 100 150 200 250 300
t
(c)

Figure 1: Dynamical behavior of system (4) with 𝑟 = 2, 𝐾 = 5, 𝛽 = 1, 𝑐 = 10, 𝑑1 = 0.5, 𝜆 = 1, 𝑑2 = 0.4, 𝑑3 = 0.3, 𝑑4 = 0.1, 𝑝1 = 0.4,
𝑝2 = 0.1, 𝜏 = 0, 𝑇 = 1. (a) Time series of the prey population; (b) time series of the immature predator population; (c) time series of the
mature predator population.

which is a contradiction. Thus, we have 𝑦2 (𝑡) ≥ 𝑦2𝑚 for all We prove that 𝑦2 (𝑡) ≥ 𝑚2 for all 𝑡 sufficiently large. Suppose
𝑡 ≥ 𝑡5 . that there exist positive constants 𝜇, 𝜂 such that 𝑦2 (𝜇) =
By (40) and (41), therefore, 𝑦2 (𝜇 + 𝜂) = 𝑚2∗ and 𝑦2 (𝑡) < 𝑚2∗ for all 𝜇 < 𝑡 < 𝜇 + 𝜂,
and inequality (40) holds true for 𝜇 < 𝑡 < 𝜇 + 𝜂, where
𝜎2 𝜇 is sufficiently large enough. Since 𝑦2 (𝑡) is continuous and
𝑉̇(𝑡) > (𝜆𝛽𝑒−𝑑2 𝜏 − 𝑑3 − 𝑀𝑑4 ) 𝑦2𝑚 > 0, 𝑡 ≥ 𝑡5 ,
𝑐 + 𝜎2 bounded and is not affected by impulses, we conclude that
(43) 𝑦2 (𝑡) is uniformly continuous. Hence, there exists a constant
𝑇0 (0 < 𝑇0 < 𝜏 and 𝑇0 is independent of the choice of 𝜇) such
which means that 𝑉(𝑡) → ∞ as 𝑡 → ∞. It is a contradiction that 𝑦2 (𝜇) > 𝑚2∗ /2 for 𝜇 ≤ 𝑡 ≤ 𝜇 + 𝑇0 . If 𝜂 ≤ 𝑇0 , our aim
with 𝑉(𝑡) ≤ (𝑀/𝜆)(1 + 𝑟𝜏𝑒−𝑑1 𝜏 ). Therefore, for any 𝑡4 > 0, the is obtained. If 𝑇0 < 𝜂 ≤ 𝜏, from the third equation of (4),
inequality 𝑦2 (𝑡) < 𝑚2∗ cannot hold for all 𝑡 ≥ 𝑡4 . So, there exist we have that, for 𝜇 < 𝑡 < 𝜇 + 𝜂, 𝑦2̇(𝑡) ≥ −𝑑3 𝑦2 (𝑡) − 𝑑4 𝑦22 (𝑡).
the following two cases: According to the assumptions 𝑦2 (𝜇) = 𝑚2∗ and 𝑦2 (𝜇) < 𝑚2∗
(i) if 𝑦2 (𝑡) ≥ 𝑚2∗ holds for all 𝑡 large enough, then our for 𝜇 < 𝑡 < 𝜇 + 𝜂, we have 𝑦2̇(𝑡) ≥ −(𝑑3 + 𝑑4 𝑚2∗ )𝑦2 (𝑡) for 𝜇 <
∗
goal is obtained; 𝑡 ≤ 𝜇 + 𝜂 ≤ 𝜇 + 𝜏. Then, we derive that 𝑦2 (𝑡) ≥ 𝑚2∗ 𝑒−(𝑑3 +𝑑4 𝑚2 )𝜏 .
(ii) if 𝑦2 (𝑡) is oscillatory about 𝑚2∗ , let It is clear that 𝑦2 (𝑡) ≥ 𝑚2 for 𝜇 ≤ 𝑡 ≤ 𝜇 + 𝜂. If 𝜂 ≥ 𝜏, then we
have 𝑦2 (𝑡) ≥ 𝑚2 for 𝜇 < 𝑡 ≤ 𝜇 + 𝜏. The same arguments can
𝑚2∗ ∗ −(𝑑3 +𝑑4 𝑚2∗ )𝜏 be continued. We can obtain 𝑦2 (𝑡) ≥ 𝑚2 for 𝜇 + 𝜏 ≤ 𝑡 ≤ 𝜇 + 𝜂.
𝑚2 = min { , 𝑚2 𝑒 }. (44)
Since the interval [𝜇, 𝜇 + 𝜂] is arbitrarily chosen, we get that
2
Abstract and Applied Analysis 7

1.5
0.5
1.4
1.3 0.48

1.2
0.46
1.1
x 1 0.44

y1
0.9
0.8 0.42

0.7
0.4
0.6
0.5
0 500 1000 1500 2000 2500 3000 0.38
0 500 1000 1500 2000
t t
(a) (b)
0.56

0.54

0.52

0.5

0.48
y2

0.46

0.44

0.42

0.4

0.38
0 500 1000 1500 2000 2500
t
(c)
Figure 2: Dynamical behavior of system (4) with 𝑟 = 1, 𝐾 = 3, 𝛽 = 2, 𝑐 = 1, 𝑑1 = 0.5, 𝜆 = 1, 𝑑2 = 0.3, 𝑑3 = 0.2, 𝑑4 = 0.1, 𝑝1 = 0.3, 𝑝2 = 0.1,
𝜏 = 1, 𝑇 = 0.1. (a) Time series of the prey population; (b) time series of the immature predator population; (c) time series of the mature
predator population.

𝑦2 (𝑡) ≥ 𝑚2 for 𝑡 sufficiently large. In view of our arguments 5. Examples and Numerical Simulations
above, the choice of 𝑚2 is independent of the positive solution In this section, we give some examples and numerical
of (4) which satisfies that 𝑦2 (𝑡) ≥ 𝑚2 for 𝑡 large enough. simulations to show the effectiveness of the main results.
Next, from the second equation of system (4), we have In system (4), we let 𝑟 = 2, 𝐾 = 5, 𝛽 = 1, 𝑐 = 10, 𝑑1 = 0.5,
𝜆 = 1, 𝑑2 = 0.4, 𝑑3 = 0.3, 𝑑4 = 0.1, 𝑝1 = 0.4, 𝑝2 = 0.1, 𝜏 = 0,
𝑦1̇(𝑡) ≥ 𝜆𝛽𝑚2 (1 − 𝑒−𝑑2 𝜏 ) − 𝑑2 𝑦1 (𝑡) . (45) 𝑇 = 1. By computation, one can obtain that the conditions of
Theorem 6 are satisfied, so the predator-extinction periodic
Integrating (45) and using comparison theorem, we can easily solution is globally attractive, which can be shown by Figure 1.
get If 𝑟 = 1, 𝐾 = 3, 𝛽 = 2, 𝑐 = 1, 𝑑1 = 0.5, 𝜆 = 1, 𝑑2 = 0.3,
𝑑3 = 0.2, 𝑑4 = 0.1, 𝑝1 = 0.3, 𝑝2 = 0.1, 𝜏 = 1, 𝑇 = 0.1, then by
𝜆𝛽𝑚2 (1 − 𝑒−𝑑2 𝜏 ) Δ
computation, the conditions of Theorem 8 are also satisfied;
𝑦1 (𝑡) ≥ min {𝑦1 (0+ ) , } = 𝑚1 . (46) hence, by Theorem 8, system (4) is permanent; see Figure 2.
𝑑2

Δ 6. Conclusion
On the other hand, by (29), we have 𝑥(𝑡) ≥ 𝑥∗ (𝑡) − 𝜀1 = 𝑚0 .
Let 𝑚 = min{𝑚0 , 𝑚1 , 𝑚2 }; then, we have 𝑥(𝑡) ≥ 𝑚, 𝑦1 (𝑡) ≥ 𝑚, In this paper, a stage-structured predator-prey model with
𝑦2 (𝑡) ≥ 𝑚. By Lemma 4 and above discussion, system (4) is Holling type III functional response, birth pulse, and impul-
permanent. This completes the proof. sive harvesting at different moments is proposed. By using
8 Abstract and Applied Analysis

comparison theorem of impulsive differential equations and [8] J. J. Jiao, X. Z. Meng, and L. S. Chen, “A stage-structured Holling
some analysis techniques, the sufficient conditions ensuring mass defence predator-prey model with impulsive perturba-
the predator-extinction periodic solution and the perma- tions on predators,” Applied Mathematics and Computation, vol.
nence of system (4) are obtained. Theorem 6 implies that 189, no. 2, pp. 1448–1458, 2007.
reducing 𝑝1 and 𝑇 appropriately and (A2 ) can be propitious [9] Z. J. Liu and R. H. Tan, “Impulsive harvesting and stocking in
to the global attractivity of the predator-extinction periodic a Monod-Haldane functional response predator-prey system,”
solution (𝑥∗ (𝑡), 0, 0). That is, if the mature predator is caught Chaos, Solitons and Fractals, vol. 34, no. 2, pp. 454–464, 2007.
excessively or the immature predator is stocked too few, it will [10] Y. Shao and B. Dai, “The dynamics of an impulsive delay
lose the merits of exploitative mature predator population. predator-prey model with stage structure and Beddington-type
Similarly, we can see from Theorem 8 that reducing 𝑝1 and functional response,” Nonlinear Analysis: Real World Applica-
tions, vol. 11, no. 5, pp. 3567–3576, 2010.
𝑇 appropriately and (A4 ) are in favor of the permanence of
system (4). It implies that reasonable harvesting on mature [11] X. Song, M. Hao, and X. Meng, “A stage-structured predator-
prey model with disturbing pulse and time delays,” Applied
predator and appropriate protecting on immature predator
Mathematical Modelling, vol. 33, no. 1, pp. 211–223, 2009.
play an important role in the permanence of system (4).
[12] Y. F. Shao and Y. Li, “Dynamical analysis of a stage structured
Moreover, by Theorems 6 and 8, we believe that there exists
predator-prey system with impulsive diffusion and generic
a sharp threshold, which is beneficial for people to make functional response,” Applied Mathematics and Computation,
the best use of biological resources but will not break the vol. 220, pp. 472–481, 2013.
biological balance. It will be interesting for us to continue to
[13] F. Y. Wei and K. Wang, “Persistence of some stage structured
study the optimal harvesting policy of system (4) in the near ecosystems with finite and infinite delay,” Nonlinear Analysis:
future. Real World Application, vol. 12, pp. 1401–1409, 2011.
[14] Z. Y. Xiang, D. Long, and X. Y. Song, “A delayed Lotka-Volterra
Conflict of Interests model with birth pulse and impulsive effect at different moment
on the prey,” Applied Mathematics and Computation, vol. 219, no.
The authors declare that there is no conflict of interests 20, pp. 10263–10270, 2013.
regarding the publication of this paper. [15] J. Jiao, S. Cai, and L. Chen, “Analysis of a stage-structured
predatory-prey system with birth pulse and impulsive har-
vesting at different moments,” Nonlinear Analysis. Real World
Acknowledgments Applications, vol. 12, no. 4, pp. 2232–2244, 2011.
[16] Z. Ma, J. Yang, and G. Jiang, “Impulsive control in a stage struc-
This paper is supported by National Natural Science Foun- ture population model with birth pulses,” Applied Mathematics
dation of China (11161015, 11361012) and Natural Science and Computation, vol. 217, no. 7, pp. 3453–3460, 2010.
Foundation of Guangxi (2013GXNSFAA019003) and partially [17] J. Liu, “Analysis of an epidemic model with density-dependent
supported by the National High Technology Research and birth rate, birth pulses,” Communications in Nonlinear Science
Development Program 863 under Grant no. 2013AA12A402. and Numerical Simulation, vol. 15, no. 11, pp. 3568–3576, 2010.
[18] C. Zhang, N.-J. Huang, and D. O’Regan, “Almost periodic
References solutions for a Volterra model with mutual interference and
Holling type III functional response,” Applied Mathematics and
[1] W. X. Li and K. Wang, “Optimal harvesting policy for general Computation, vol. 225, pp. 503–511, 2013.
stochastic logistic population model,” Journal of Mathematical [19] Y. F. Lv, R. Yuan, and Y. Z. Pei, “A prey-predator model
Analysis and Applications, vol. 368, no. 2, pp. 420–428, 2010. with harvesting for fishery resource with reserve area,” Applied
[2] T. Das, R. N. Mukherjee, and K. S. Chaudhuri, “Harvesting Mathematical Modelling, vol. 37, no. 5, pp. 3048–3062, 2013.
of a prey-predator fishery in the presence of toxicity,” Applied [20] B. Shulgin, L. Stone, and Z. Agur, “Pulse vaccination strategy in
Mathematical Modelling, vol. 33, no. 5, pp. 2282–2292, 2009. the SIR epidemic model,” Bulletin of Mathematical Biology, vol.
60, no. 6, pp. 1123–1148, 1998.
[3] R. Zhang, J. F. Sun, and H. X. Yang, “Analysis of a prey-predator
fishery model with prey reserve,” Applied Mathematical Sci- [21] V. Lakshmikantham, D. D. Bainov, and P. S. Simeonov, Theory
ences, vol. 1, no. 50, pp. 2481–2492, 2007. of Impulsive Differential Equations, World Scientific, Singapore,
1989.
[4] Z. J. Zeng, “Asymptotically periodic solution and optimal
harvesting policy for gompertz system,” Nonlinear Analysis:
Real World Applications, vol. 12, no. 3, pp. 1401–1409, 2011.
[5] X. Y. Song and L. S. Chen, “Optimal harvesting and stability
for a two-species competitive system with stage structure,”
Mathematical Biosciences, vol. 170, no. 2, pp. 173–186, 2001.
[6] T. K. Kar, “Selective harvesting in a prey-predator fishery with
time delay,” Mathematical and Computer Modelling, vol. 38, no.
3-4, pp. 449–458, 2003.
[7] X. W. Jiang, Q. Song, and M. Y. Hao, “Dynamics behaviors of
a delayed stage-structured predator-prey model with impulsive
effect,” Applied Mathematics and Computation, vol. 215, no. 12,
pp. 4221–4229, 2010.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2015, Article ID 951340, 9 pages
http://dx.doi.org/10.1155/2015/951340

Research Article
Skew Circulant Type Matrices Involving the Sum of
Fibonacci and Lucas Numbers

Zhaolin Jiang and Yunlan Wei

School of Science, Linyi University, Shuangling Road, Linyi, Shandong 276000, China

Correspondence should be addressed to Zhaolin Jiang; jzh1208@sina.com

Received 27 June 2014; Revised 8 August 2014; Accepted 11 August 2014

Academic Editor: Yongli Song

Copyright © 2015 Z. Jiang and Y. Wei. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Skew circulant and circulant matrices have been an ideal research area and hot issue for solving various differential equations. In
this paper, the skew circulant type matrices with the sum of Fibonacci and Lucas numbers are discussed. The invertibility of the
skew circulant type matrices is considered. The determinant and the inverse matrices are presented. Furthermore, the maximum
column sum matrix norm, the spectral norm, the Euclidean (or Frobenius) norm, the maximum row sum matrix norm, and bounds
for the spread of these matrices are given, respectively.

1. Introduction to perform filtering procedure in approximately half of com-

putational cost for real signals. In [9], a new fast algorithm for
As is well-known, skew circulant and circulant matrices optimal design of block digital filters (BDFs) was proposed
play a crucial role for solving various differential equations. based on skew circulant matrix. Spectral decompositions of
Authors in [1] presented the skew circulant matrices as skew circulant and skew left circulant matrices were discussed
preconditioners for linear multistep formulae (LMF-)based in [10]. Li et al. [11] gave the style spectral decomposition of
ordinary differential equations (ODEs) codes. Claeyssen skew circulant matrix firstly and then dealt with the optimal
et al. [2] discussed factor block circulant and periodic solu- backward perturbation analysis for the linear system with
tions of undamped matrix differential equations. Using cir- skew circulant coefficient matrix.
culant matrix, Karasözen and Şimşek [3] considered periodic Some scholars have given various algorithms for the
boundary conditions such that no additional boundary terms determinants and inverses of nonsingular circulant matrices
will appear after semidiscretization. Meyer and Rjasanow [4] [12, 13]. Unfortunately, the computational complexity of
have presented an effective direct solution method for certain these algorithms is very amazing with the order of matrix
boundary element equations in 3D. Guo et al. concerned on increasing. However, some authors gave the explicit determi-
generic Dn-Hopf bifurcation to a delayed Hopfield-Cohen- nants and inverse of circulant and skew circulant involving
Grossberg model of neural networks (5.17) in [5], where some famous numbers. For example, Shen et al. considered
𝑇 denoted an interconnection matrix. In particular, they circulant matrices with Fibonacci and Lucas numbers and
assumed that 𝑇 is a symmetric circulant matrix. In [6], presented their explicit determinants and inverses by con-
Jin et al. proposed the GMRES method with the Strang-type structing the transformation matrices in [14]. Gao et al. [15]
block-circulant preconditioner for solving singular pertur- gave explicit determinants and inverses of skew circulant
bation delay differential equations. In [7], two new normal- and skew left circulant matrices with Fibonacci and Lucas
form realizations are presented which utilize circulant and numbers. In [16], Jiang et al. discussed the nonsingularity
skew circulant matrices as their state transition matrices. The of the skew circulant type matrices and presented explicit
well known second-order coupled form is a special case of determinants and inverse matrices of these special matrices.
the skew circulant form. Compared with cyclic convolution Furthermore, four kinds of norms and bounds for the spread
algorithm, the skew cyclic convolution algorithm [8] is able of these matrices are given separately. In [17], Jiang and
2 Abstract and Applied Analysis

Hong give exact determinants of some special circulant Lemma 1. Let {L𝑛 } be the sum of Fibonacci and lucas numbers;
matrices involving four kinds of famous numbers. Authors then,
[18] discussed the nonsingularity of the circulant type matrix 𝑛
and presented the explicit determinant and inverse matrices. (i) ∑L𝑖 = L𝑛+2 − 4, (6)
There are several papers on the norms of some special 𝑖=1
matrices. Solak [19] established the lower and upper bounds
𝑛
for the spectral norms of circulant matrices with classical
Fibonacci and Lucas numbers entries. İpek [20] investigated (ii) ∑L2𝑖 = L𝑛 L𝑛+1 − 4, (7)
𝑖=1
an improved estimation for spectral norms of these matrices.
Beginning with Mirsky [21] several authors [22–24] have 𝑛
obtained bounds for the spread of a matrix. (iii) ∑𝑖L𝑖 = (𝑛 − 2) L𝑛+2 + L𝑛 + 6. (8)
Additionally, skew circulant type matrices include skew 𝑖=1
circulant and skew left circulant matrices. The norm and Proof. According to
spread of skew circulant type matrices have not been studied.
It is hoped that this paper will help in changing this. More L𝑛+1 = L𝑛 + L𝑛−1 where L0 = 2, L1 = 2, (9)
work continuing the present paper is forthcoming.
The sum of Fibonacci and Lucas sequence is defined by we have
the following recurrence relations: 𝑛
∑L 𝑖 = L 1 + L 2 + ⋅ ⋅ ⋅ + L 𝑛
L𝑛+2 = L𝑛+1 + L𝑛 where L0 = 2, L1 = 2, (1) 𝑖=1

for 𝑛 ≥ 0. The first few values of the sequence are given by the = (L3 − L2 ) + (L4 − L3 )
following table:
+ ⋅ ⋅ ⋅ + (L𝑛+2 − L𝑛+1 )
𝑛 0 1 2 3 4 5 6 7 8
. (2)
L𝑛 2 2 4 6 10 16 26 42 68 = L𝑛+2 − L2 ,
The {L𝑛 } is given by the formula 𝑛
∑L2𝑖 = L21 + L22 + ⋅ ⋅ ⋅ + L2𝑛
𝛼𝑛−1 − 𝛽𝑛−1 𝛼𝑛 − 𝛽𝑛 𝛼𝑛+1 − 𝛽𝑛+1 𝑖=1
L𝑛 = + + , (3)
𝛼−𝛽 𝛼−𝛽 𝛼−𝛽 = L1 (L2 − L0 ) + L2 (L3 − L1 )
2
where 𝛼 and 𝛽 are the roots of the characteristic equation 𝑥 − (10)
+ ⋅ ⋅ ⋅ + L𝑛 (L𝑛+1 − L𝑛−1 )
𝑥 − 1 = 0.
In this paper, we consider skew circulant type matrices,
= L𝑛 L𝑛+1 − L0 L1 ,
including the skew circulant and skew left circulant matrices.
We define a sum of Fibonacci and Lucas skew circulant 𝑛
matrix which is an 𝑛 × 𝑛 complex matrix with following form: ∑𝑖L𝑖 = L1 + 2L2 + ⋅ ⋅ ⋅ + 𝑛L𝑛
𝑖=1
SCirc (L1 , L2 , . . . , L𝑛 )
= (L2 − L0 ) + 2 (L2 − L0 )
L1 L2 ⋅ ⋅ ⋅ L𝑛−1 L𝑛
−L𝑛 L1 ⋅ ⋅ ⋅ L𝑛−2 L𝑛−1 + ⋅ ⋅ ⋅ + 𝑛 (L𝑛+1 − L𝑛−1 )
(4)
= ( ... ..
. d ..
. .. ) ,
. 𝑛
−L3 −L4 ⋅ ⋅ ⋅ L1 L2 = − 2∑L𝑖 + (𝑛 + 1) L𝑛 + 𝑛L𝑛+1 − L0
−L2 −L3 ⋅ ⋅ ⋅ −L𝑛 L1 𝑖=1

= (𝑛 − 2) L𝑛+2 + L𝑛 + 2L2 − L0 ,
where each row is a cyclic shift of the row above the right.
Besides, a sum of Fibonacci and Lucas skew left circulant and hence
matrix is given by 𝑛
∑L𝑖 = L𝑛+2 − 4,
SLCirc (L1 , L2 , . . . , L𝑛 ) 𝑖=1

L1 L2 ⋅ ⋅ ⋅ L𝑛−1 L𝑛 𝑛
L2 L3 ⋅ ⋅ ⋅ L𝑛 −L1 ∑L2𝑖 = L𝑛 L𝑛+1 − 4, (11)
(5)
= ( ... .. .. .. ) , 𝑖=1
. d . . 𝑛
L𝑛−1 L𝑛 ⋅ ⋅ ⋅ −L𝑛−3 −L𝑛−2 ∑𝑖L𝑖 = (𝑛 − 2) L𝑛+2 + L𝑛 + 6.
L𝑛 −L1 ⋅ ⋅ ⋅ −L𝑛−2 −L𝑛−1 𝑖=1

where each row is a cyclic shift of the row above the left. This completes the proof.
Abstract and Applied Analysis 3

Lemma 2 (see [10]). Let 𝐴 = SLCirc(𝑎1 , 𝑎2 , . . . , 𝑎𝑛 ) be a skew Proof. Obviously, det 𝐴 3 = 304 satisfies the equation. In the
left circulant matrix and let 𝑛 be odd; then, case 𝑛 > 3, let
1
𝑡 1
󵄨󵄨 𝑛 󵄨󵄨 ( 1 1 −1)
󵄨󵄨 (𝑗−(1/2))(𝑘−1) 󵄨󵄨󵄨 𝑛−1 (0 0 1 −1 −1)
󵄨
𝜆 𝑗 = ± 󵄨󵄨 ∑ 𝑎𝑘 𝜔 󵄨󵄨 (𝑗 = 1, 2, . . . , ), ( )
󵄨󵄨𝑘=1 󵄨󵄨 2 ℶ = ( .. ),
󵄨 󵄨 (. c c c )
(12) ( )
𝑛 0 1 c c
󵄨 󵄨
𝜆 (𝑛+1)/2 = ∑ 󵄨󵄨󵄨󵄨𝑎𝑘 (−1)𝑘−1 󵄨󵄨󵄨󵄨 , 0 1 −1 c 0
𝑘=1 ( 0 1 −1 −1 ) (15)

1 0 0 ⋅⋅⋅ 0 0
0 𝑥𝑛−2 0 ⋅⋅⋅ 0 1
𝑛−3
where 𝜆 𝑗 (𝑗 = 1, 2, . . . , (𝑛 − 1)/2, (𝑛 + 1)/2) are the eigenvalues (0 𝑥 0 ⋅ ⋅ ⋅ 1 0)
of 𝐴. Ω1 = ( . . .. . .)
.. .. . d .. ..
0 𝑥 1 ⋅⋅⋅ 0 0
(0 1 0 ⋅ ⋅ ⋅ 0 0)
2. Determinant and Inverse of Skew
be two 𝑛 × 𝑛 matrices; then, we have
Circulant Matrix with the Sum of Fibonacci
and Lucas Numbers 2 𝑙𝑛󸀠 𝑐13 ⋅ ⋅ ⋅ 𝑐1,𝑛−1 𝑐1𝑛
0 𝑙𝑛 𝑐23 ⋅ ⋅ ⋅ 𝑐2,𝑛−1 𝑐2𝑛
In this section, let 𝐴 𝑛 = SCirc (L1 , . . . , L𝑛 ) be a skew circulant
(0 0 𝑐 )
matrix. First of all, a determinant explicit formula for the ℶ𝐴 𝑛 Ω1 = (0 0 𝑑 d ), (16)
matrix 𝐴 𝑛 is given. After that, we prove that 𝐴 𝑛 is an invertible .. ..
matrix for any positive interger 𝑛, and then we find the inverse . . d 𝑐
of the matrix 𝐴 𝑛 . In the following, let (0 0 𝑑 𝑐 )

where
𝑐1𝑗 = L𝑛+2−𝑗 (𝑗 = 3, 4, . . . , 𝑛) ,
2 + L𝑛 (17)
𝑥=− , 𝑡 = 2, 𝑐 = 2 + L𝑛+1 , 𝑐2𝑗 = 𝑡L𝑛+2−𝑗 − L𝑛+3−𝑗 (𝑗 = 3, 4, . . . , 𝑛) .
2 + L𝑛+1

𝑛−1 So it holds that

𝑑 = 2 + L𝑛 , 𝑙𝑛󸀠 = ∑ L𝑘+1 ⋅ 𝑥 𝑛−(𝑘+1)
(𝑛 ≥ 2) , (13) 𝑛−2
𝑘=1 det ℶ det 𝐴 𝑛 det Ω1 = 2 [2 + 𝑡L𝑛 + ∑ (𝑡L𝑘+1 − L𝑘+2 )
𝑘=1
𝑛−2 (18)
𝑛−(𝑘+1)
𝑙𝑛 = 2 + 𝑡L𝑛 + ∑ (𝑡L𝑘+1 − L𝑘+2 ) ⋅ 𝑥 (𝑛 ≥ 3) .
𝑛−(𝑘+1) 𝑛−2
𝑘=1 ×𝑥 ] ⋅ (2 + L𝑛+1 ) .

While taking det ℶ = det Ω1 = (−1)(𝑛−1)(𝑛−2)/2 , we have

Theorem 3. Let 𝐴 𝑛 = SCirc (L1 , . . . , L𝑛 ) be a skew circulant 𝑛−2
matrix; then det 𝐴 𝑛 = 2 [2 + 𝑡L𝑛 + ∑ (𝑡L𝑘+1 − L𝑘+2 )
𝑘=1
(19)
𝑛−2
𝑛−2 × 𝑥𝑛−(𝑘+1) ] ⋅ (2 + L𝑛+1 ) .
𝑛−(𝑖+1) 𝑛−2
det 𝐴 𝑛 = 2 [2 + 𝑡L𝑛 + ∑ (𝑡L𝑖+1 − L𝑖+2 ) 𝑥 ]⋅𝑐
𝑖=1
This completes the proof.
(𝑛 ≥ 3) , Theorem 4. Let 𝐴 𝑛 = SCirc (L1 , . . . , L𝑛 ) be a skew circulant
(14) matrix; then 𝐴 𝑛 is an invertible matrix for any positive interger
𝑛.

Proof. Taking 𝑛 = 1 in Theorem 3, we have det 𝐴 1 ≠

where L𝑛 is the 𝑛th sum of Fibonacci andLucas numbers. 0. Hence 𝐴 1 is invertible. In the case 𝑛 > 1, since
4 Abstract and Applied Analysis

󸀠 𝑛−2
L𝑛 = (𝛼𝑛−1 − 𝛽𝑛−1 )/(𝛼 − 𝛽) + (𝛼𝑛 − 𝛽𝑛 )/(𝛼 − 𝛽) + (𝛼𝑛+1 − Then the inverse H−1 = [ℎ𝑖,𝑗 ]𝑖,𝑗=1 of the matrix H is equal to
𝛽𝑛+1 )/(𝛼 − 𝛽), where 𝛼 + 𝛽 = 1, 𝛼𝛽 = −1, we have
𝑖−𝑗
𝑛
𝑗−1 { (−𝑑) , 𝑖 ≥ 𝑗,
𝑓 (𝜔𝑘 𝜂) = ∑ L𝑗 (𝜔𝑘 𝜂) ℎ𝑖𝑗󸀠 = { 𝑐𝑖−𝑗+1 (24)
𝑗=1 {0, 𝑖 < 𝑗.
𝑛
𝛼𝑗−1 − 𝛽𝑗−1 𝛼𝑗 − 𝛽𝑗 Proof. Let 𝑒𝑖𝑗 = ∑𝑛−2 󸀠
𝑘=1 ℎ𝑖𝑘 ℎ𝑘𝑗 . Obviously, 𝑒𝑖𝑗 = 0 for 𝑖 < 𝑗. In the
= ∑( +
𝑗=1 𝛼−𝛽 𝛼−𝛽 case 𝑖 = 𝑗, we obtain 𝑒𝑖𝑖 = ℎ𝑖𝑖 ℎ𝑖𝑖󸀠 = (2 + L𝑛+1 ) ⋅ 1/(2 + L𝑛+1 ) = 1.
For 𝑖 ≥ 𝑗 + 1, we obtain
𝛼𝑗+1 − 𝛽𝑗+1 𝑗−1
+ ) (𝜔𝑘 𝜂) 𝑛−2
𝛼−𝛽 󸀠
𝑒𝑖𝑗 = ∑ ℎ𝑖𝑘 ℎ𝑘𝑗 󸀠
= ℎ𝑖,𝑖−1 ℎ𝑖−1,𝑗 + ℎ𝑖𝑖 ℎ𝑖𝑗󸀠
𝑛 𝑘=1
𝛼𝑗−1 + 𝛼𝑗 + 𝛼𝑗+1 𝑗−1 (25)
= ∑( ) (𝜔𝑘 𝜂)
𝛼−𝛽 (−𝑑)𝑖−𝑗−1 (−𝑑)𝑖−𝑗
𝑗=1 =𝑑⋅ + 𝑐 ⋅ = 0.
𝑐𝑖−𝑗 𝑐𝑖−𝑗+1
𝑛
𝛽𝑗−1 + 𝛽𝑗 + 𝛽𝑗+1 𝑗−1
(20)
−∑( ) (𝜔𝑘 𝜂) Hence, we get HH−1 = 𝐼𝑛−2 ; here 𝐼𝑛−2 is (𝑛 − 2) × (𝑛 − 2)
𝑗=1 𝛼−𝛽 identity matrix. Similarly, we can verify H−1 H = 𝐼𝑛−2 . Thus,
the proof is completed.
1 + 𝛼 + 𝛼2 1 + 𝛼𝑛
= ⋅
𝛼−𝛽 1 − 𝛼𝜔𝑘 𝜂 Theorem 6. Let 𝐴 𝑛 = SCirc(L1 , . . . , L𝑛 ) be a skew circulant
matrix; then
1 + 𝛽 + 𝛽2 1 + 𝛽𝑛
− ⋅ −1 1
𝛼−𝛽 1 − 𝛽𝜔𝑘 𝜂 (𝐴 𝑛 ) = ⋅ SCirc (𝑦1󸀠 , 𝑦2󸀠 , . . . , 𝑦𝑛󸀠 ) (𝑛 ≥ 4) , (26)
𝑙𝑛
2 + L𝑛+1 + (2 + L𝑛 ) 𝜔𝑘 𝜂 where
=
1 − 𝜔𝑘 𝜂 − 𝜔2𝑘 𝜂2
(−𝑑)𝑛−3
(𝑘 = 1, 2, . . . , 𝑛 − 1) , 𝑦1󸀠 = 1 − [(6 − 4𝑡) ⋅
𝑐𝑛−2
where 𝜔 = exp(2𝜋𝑖/𝑛), 𝜂 = exp(𝜋𝑖/𝑛). If there exists 𝜔𝑙 𝜂 (𝑙 = 𝑛−3
(−𝑑)𝑖−1
1, 2, . . . , 𝑛 − 1) such that 𝑓(𝜔𝑙 𝜂) = 0, we have 2 + L𝑛+1 + (2 + + ∑ (L𝑛+2−𝑖 − 𝑡L𝑛+1−𝑖 ) ⋅ ]
𝑐𝑖
L𝑛 )𝜔𝑙 𝜂 = 0 for 1 − 𝜔𝑙 𝜂 − 𝜔2𝑙 𝜂2 ≠0, and hence it follows that 𝑖=1

𝜔𝑙 𝜂 = −(2 + L𝑛+1 )/(2 + L𝑛 ) is a real number. Since (𝑛 ≥ 4) ,

(2𝑙 + 1) 𝜋𝑖
𝜔𝑙 𝜂 = exp ( ) 𝑛−2
(−𝑑)𝑖−1
𝑛 𝑦2󸀠 = −𝑡 − ∑ (L𝑛+1−𝑖 − 𝑡L𝑛−𝑖 ) ⋅ ,
(21) 𝑖=1 𝑐𝑖
(2𝑙 + 1) 𝜋 (2𝑙 + 1) 𝜋 (27)
= cos + 𝑖 sin , 1
𝑛 𝑛 𝑦3󸀠 = − (6 − 4𝑡) ⋅ ,
𝑐
it yields that sin((2𝑙 + 1)𝜋)/𝑛 = 0, so we have 𝜔𝑙 𝜂 = −1 for 0 <
(2𝑙 + 1)𝜋/𝑛 < 2𝜋. Since 𝑥 = −1 is not the root of the equation 2
(−𝑑)𝑖−1
2 + L𝑛+1 + (2 + L𝑛 )𝑥 = 0 (𝑛 ≥ 1). We obtain 𝑓(𝜔𝑘 𝜂) ≠0 for 𝑦4󸀠 = −∑ (L1+𝑖 − 𝑡L𝑖 ) ⋅ ,
𝑖=1 𝑐𝑖
any 𝜔𝑘 𝜂 (𝑘 = 1, 2, . . . , 𝑛 − 1), while
2
𝑛 (−𝑑)𝑘−5+𝑖
𝑓 (𝜂) = ∑L𝑗 𝜂 𝑗−1 𝑦𝑘󸀠 = −∑ (L1+𝑖 − 𝑡L𝑟+𝑖 ) ⋅
𝑖=1 𝑐𝑘−4+𝑖
𝑗=1
(22)
(𝑘 = 5, 6, . . . , 𝑛) .
2 + L𝑛+1 + (2 + L𝑛 ) 𝜂
= ≠0.
1 − 𝜂 − 𝜂2 Proof. Let
It follows from Lemma 1 in [25] that the conclusion holds. 𝑙𝑛󸀠
1 − 𝜔 𝜔 ⋅ ⋅ ⋅ 𝜔1𝑛
2 13 14
Lemma 5. Let the matrix H = [ℎ𝑖𝑗 ]𝑛−2
𝑖,𝑗=1 be of the form
(0 1 𝜔23 𝜔24 ⋅ ⋅ ⋅ 𝜔2𝑛 )
( ⋅⋅⋅ 0 )
Ω2 = ( 0 0 1 0 ), (28)
2 + L𝑛+1 , 𝑖 = 𝑗, (0 0 0 1 ⋅⋅⋅ 0 )
{
{ .. .. .. .. .
ℎ𝑖𝑗 = {2 + L𝑛 ,
{
𝑖 = 𝑗 + 1, (23) . . . . d ..
{0, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒. (0 0 0 0 ⋅⋅⋅ 1 )
Abstract and Applied Analysis 5

where − 𝑦𝑘 = 𝑇2,𝑛−𝑘+3 − 𝑇2,𝑛−𝑘+4 − 𝑇2,𝑛−𝑘+5

󸀠
1 𝑙𝑛 1 2 (−𝑑)𝑘−5+𝑖
𝜔1𝑗 = [ (𝑡L𝑛+2−𝑗 − L𝑛+3−𝑗 ) − L𝑛+2−𝑗 ] = ∑ (L1+𝑖 − 𝑡L𝑖 ) ⋅ 𝑘−4+𝑖 ,
2 𝑙𝑛 𝑙𝑛 𝑖=1 𝑐
(𝑗 = 3, 4, . . . , 𝑛) , (29)
..
.
1
𝜔2𝑗 = ⋅ (L𝑛+3−𝑗 − 𝑡L𝑛+2−𝑗 ) (𝑗 = 3, 4, . . . , 𝑛) .
𝑙𝑛 − 𝑦𝑛 = 𝑇23 − 𝑇24 − 𝑇25

𝑛−2
Then we have (−𝑑)𝑖−1 𝑛−3 (−𝑑)𝑖−1
= ∑ 𝜔2,2+𝑖 ⋅ − ∑ 𝜔2,3+𝑖 ⋅
𝑖=1 𝑐𝑖 𝑖=1 𝑐𝑖
2 0 00 0 ⋅⋅⋅
0 𝑙𝑛 00 0 ⋅⋅⋅ 𝑛−4
0 0 𝑐0) 0 ⋅⋅⋅ (−𝑑)𝑖−1
( − ∑ 𝜔2,4+𝑖 ⋅
ℶ𝐴 𝑛 Ω1 Ω2 = (0 0 𝑑0) , 𝑐 ⋅⋅⋅ (30) 𝑖=1 𝑐𝑖
.. .. ..
. ..
. d ..
. . .
1 2 (−𝑑)𝑛−5+𝑖
(0 0 0 0 ⋅ ⋅ ⋅ 𝑐 ) = ∑ (L1+𝑖 − 𝑡L𝑖 ) ⋅ 𝑛−4+𝑖 ,
𝑙𝑛 𝑖=1 𝑐
so ℶ𝐴 𝑛 Ω1 Ω2 = D ⊕ H, and here 𝐷 = diag(2, 𝑙𝑛 ) is a 1
diagonal matrix, and D ⊕ H is the direct sum of D and H. 𝑦1 = − 𝑇23 − 𝑇24
𝑙𝑛
If we denote Ω = Ω1 Ω2 , then we obtain 𝐴−1 𝑛 = Ω(D
−1
⊕
−1
H )Σ. Since the last row elements of the matrix Ω are
(0, 1, 𝜔23 , 𝜔24 , . . . , 𝜔2,𝑛−1 , 𝜔2𝑛 ), the last row elements of the 1 1 (−𝑑)𝑛−3
= − [(6 − 4𝑡) ⋅ 𝑛−2
matrix Ω(D−1 ⊕ H−1 ) are (0, 1/𝑙𝑛 , 𝑇23 , 𝑇24 , . . . , 𝑇2𝑛 ), where 𝑙𝑛 𝑙𝑛 𝑐

𝑛−3
𝑛−2 (−𝑑)𝑖−1
(−𝑑)𝑖−1 + ∑ (L𝑛+2−𝑖 − 𝑡L𝑛+1−𝑖 ) ⋅ ]
𝑇23 = ∑ 𝜔2,2+𝑖 ⋅ (𝑛 ≥ 3) , 𝑖=1 𝑐𝑖
𝑖=1 𝑐𝑖
(31)
𝑛+1−𝑘 (𝑛 ≥ 4) .
(−𝑑)𝑖−1
𝑇2𝑘 = ∑ 𝜔2,𝑘−1+𝑖 ⋅ (𝑘 = 3, 4, . . . , 𝑛) . (32)
𝑖=1 𝑐𝑖

Hence it follows from Lemma 5 that by letting 𝐴−1 𝑛 =

SCirc (𝑦1 , 𝑦2 , . . . , 𝑦𝑛 ) then its last row elements are (−𝑦2 , Hence, we obtain
−𝑦3 , . . . , −𝑦𝑛 , 𝑦1 ) which are given by the following equations:

𝑡 1 1 (−𝑑)𝑛−3
− 𝑦2 = + 𝑇23 𝑦1 = − [(6 − 4𝑡) ⋅ 𝑛−2
𝑙𝑛 𝑙𝑛 𝑙𝑛 𝑐

𝑡 1 𝑛−2 (−𝑑)𝑖−1 𝑛−3

(−𝑑)𝑖−1
= + ∑ (L𝑛+1−𝑖 − 𝑡L𝑛−𝑖 ) ⋅ , + ∑ (L𝑛+2−𝑖 − 𝑡L𝑛+1−𝑖 ) ],
𝑙𝑛 𝑙𝑛 𝑖=1 𝑐𝑖 𝑐𝑖
𝑖=1

1 1
− 𝑦3 = 𝑇2,𝑛 = (6 − 4𝑡) ⋅ , 𝑡 1 𝑛−2 (−𝑑)𝑖−1
𝑙𝑛 𝑐 𝑦2 = − − ∑ (L𝑛+1−𝑖 − 𝑡L𝑛−𝑖 ) ⋅ ,
𝑙𝑛 𝑙𝑛 𝑖=1 𝑐𝑖
− 𝑦4 = 𝑇2,𝑛−1 − 𝑇2𝑛
1 1
2 𝑖−1
𝑦3 = − (6 − 4𝑡) ⋅ ,
1 (−𝑑) 𝑙𝑛 𝑐
= ∑ (L1+𝑖 − 𝑡L𝑖 ) ⋅ ,
𝑙𝑛 𝑖=1 𝑐𝑖
1 2 (−𝑑)𝑖−1
𝑦4 = − ∑ (L1+𝑖 − 𝑡L𝑖 ) ⋅ ,
− 𝑦5 = 𝑇2,𝑛−2 − 𝑇2𝑛−1 − 𝑇2𝑛 𝑙𝑛 𝑖=1 𝑐𝑖

1 2 (−𝑑)𝑖 1 2 (−𝑑)𝑖
= ∑ (L1+𝑖 − 𝑡L𝑖 ) ⋅ 𝑖+1 , 𝑦5 = − ∑ (L1+𝑖 − 𝑡L𝑖 ) ⋅ 𝑖+1 ,
𝑙𝑛 𝑖=1 𝑐 𝑙𝑛 𝑖=1 𝑐
6 Abstract and Applied Analysis

𝑛 𝑛
1 2 (−𝑑)𝑘−5+𝑖 󵄩 󵄩 2 󵄨 󵄨2
𝑦𝑘 = − ∑ (L1+𝑖 − 𝑡L𝑖 ) ⋅ 𝑘−4+𝑖 , (󵄩󵄩󵄩𝐴 𝑛 󵄩󵄩󵄩𝐹 ) = ∑∑ 󵄨󵄨󵄨󵄨𝑎𝑖𝑗 󵄨󵄨󵄨󵄨
𝑙𝑛 𝑖=1 𝑐 𝑖=1 𝑗=1

.. 𝑛
. = 𝑛∑L2𝑖
𝑖=1
1 2 (−𝑑)𝑛−5+𝑖
𝑦𝑛 = ∑ (L1+𝑖 − 𝑡L𝑖 ) ⋅ 𝑛−4+𝑖 , = 𝑛 (L𝑛 L𝑛+1 − 4) .
𝑙𝑛 𝑖=1 𝑐
(37)
1 Thus
𝐴−1
𝑛 = ⋅ SCirc (𝑦1󸀠 , 𝑦2󸀠 , . . . , 𝑦𝑛󸀠 ) ,
𝑙𝑛 󵄩󵄩 󵄩󵄩
󵄩󵄩𝐴 𝑛 󵄩󵄩𝐹 = √𝑛 (L𝑛 L𝑛+1 − 4). (38)
(33)

where
Theorem 8. Let 𝐴󸀠𝑛 = SCirc(L1 , −L2 , . . . , −L𝑛−1 , L𝑛 ) be an
odd-order alternative skew circulant matrix and let 𝑛 be odd.
(−𝑑)𝑛−3
𝑦1󸀠 = 1 − [(6 − 4𝑡) ⋅ Then
𝑐𝑛−2 𝑛
󵄩󵄩 󸀠 󵄩󵄩
󵄩󵄩𝐴 𝑛 󵄩󵄩 = ∑L𝑖 = L𝑛+2 − 4. (39)
𝑛−3 󵄩 󵄩2
(−𝑑)𝑖−1 𝑖=1
+ ∑ (L𝑛+2−𝑖 − 𝑡L𝑛+1−𝑖 ) ⋅ ]
𝑖=1 𝑐𝑖 Proof. By Lemma 1 in [25], we have
𝑛
𝑖−1
(𝑛 ≥ 4) , 𝜆 𝑗 (𝐴󸀠𝑛 ) = ∑(−1)𝑖−1 L𝑖 (𝜔𝑗 𝜂) . (40)
𝑖=1
𝑛−2
(−𝑑)𝑖−1
𝑦2󸀠 = −𝑡 − ∑ (L𝑛+1−𝑖 − 𝑡L𝑛−𝑖 ) ⋅ , Therefore
𝑖=1 𝑐𝑖 (34) 󵄨󵄨 󵄨 𝑛 󵄨 󵄨 󵄨 𝑖−1 󵄨
󵄨󵄨𝜆 𝑗 (𝐴󸀠𝑛 )󵄨󵄨󵄨 ≤ ∑ 󵄨󵄨󵄨(−1)𝑖−1 L𝑖 󵄨󵄨󵄨 ⋅ 󵄨󵄨󵄨󵄨(𝜔𝑗 𝜂) 󵄨󵄨󵄨󵄨
󵄨 󵄨 󵄨 󵄨 󵄨 󵄨
1 𝑖=1
𝑦3󸀠 = − (6 − 4𝑡) ⋅ , (41)
𝑐 𝑛
= ∑L 𝑖 (𝑗 = 0, 1, . . . , 𝑛 − 1) .
2
(−𝑑)𝑖−1 𝑖=1
𝑦4󸀠 = −∑ (L1+𝑖 − 𝑡L𝑖 ) ⋅ ,
𝑖=1 𝑐𝑖 Since 𝑛 is odd, ∑𝑛𝑖=1 L𝑖 is an eigenvalue of 𝐴󸀠𝑛 , which is

2 1
(−𝑑)𝑘−5+𝑖 L1 −L2 ⋅ ⋅ ⋅ L𝑛
−1
𝑦𝑘󸀠 = −∑ (L1+𝑖 − 𝑡L𝑖 ) ⋅ , (𝑘 = 5, 6, . . . , 𝑛) . −L𝑛 L1 ⋅ ⋅ ⋅ −L𝑛−1
𝑐𝑘−4+𝑖
𝑖=1 (1)
( 𝑛−1 −L𝑛 ⋅ ⋅ ⋅ L𝑛−2 ) (−1)
L
.. .. .. ..
This completes the proof. . . . . ..
.
L2 −L3 ⋅ ⋅ ⋅ L1
1
( )
3. Norm and Spread of Skew Circulant (42)
1
Matrix with the Sum of Fibonacci and −1
Lucas Numbers
(1)
𝑛
= ∑L𝑖 ⋅ (−1) .
Theorem 7. Let 𝐴 𝑛 = SCirc(L1 , . . . , L𝑛 ) be a skew circulant 𝑖=1 ..
matrix. The three kinds norms of 𝐴 𝑛 are given by .
1
( )
󵄩󵄩 󵄩󵄩 󵄩 󵄩
󵄩󵄩𝐴 𝑛 󵄩󵄩1 = 󵄩󵄩󵄩𝐴 𝑛 󵄩󵄩󵄩∞ = L𝑛+2 − 4, (35) To sum up, we have
󵄩󵄩 󵄩󵄩 󵄨 󵄨 𝑛
󵄩󵄩𝐴 𝑛 󵄩󵄩𝐹 = √𝑛 (L𝑛 L𝑛+1 − 4). (36) max 󵄨󵄨󵄨󵄨𝜆 𝑗 (𝐴󸀠𝑛 )󵄨󵄨󵄨󵄨 = ∑L𝑖 . (43)
0≤𝑗≤𝑛−1
𝑖=1

Proof. By Definition 4 in [26], (6), and (7), we have Since all skew circulant matrices are normal, by Lemma 7 in
[26], (6), and (43), we have
󵄩󵄩 󵄩󵄩 󵄩 󵄩
󵄩󵄩𝐴 𝑛 󵄩󵄩1 = 󵄩󵄩󵄩𝐴 𝑛 󵄩󵄩󵄩∞ 󵄩󵄩 󸀠 󵄩󵄩 𝑛
󵄩󵄩𝐴 𝑛 󵄩󵄩 = ∑L𝑖 = L𝑛+2 − 4, (44)
𝑛 󵄩 󵄩2
𝑖=1
= ∑L𝑖 = L𝑛+2 − 4,
𝑖=1 which completes the proof.
Abstract and Applied Analysis 7

Theorem 9. Let 𝐴 𝑛 = SCirc(L1 , . . . , L𝑛 ) be a skew circulant Theorem 11. Let 𝐴󸀠󸀠𝑛 = SLCirc(L1 , . . . , L𝑛 ) be a skew left
matrix; then the bounds for the spread of 𝐴 𝑛 are circulant matrix for any positive interger 𝑛; then 𝐴󸀠󸀠𝑛 is an
invertible matrix.
𝑠 (𝐴 𝑛 ) ⩽ √2𝑛 (L𝑛 L𝑛+1 − 8),
(45) Theorem 12. Let 𝐴󸀠󸀠𝑛 = SLCirc(L1 , . . . , L𝑛 ) be a skew left
1 󵄨󵄨 󵄨 circulant matrix; then
𝑠 (𝐴 𝑛 ) ≥ 󵄨2L − 𝑛L𝑛+2 − 6𝑛 − 20󵄨󵄨󵄨 .
𝑛 − 1 󵄨 𝑛+4
−1 1
Proof. The trace of 𝐴 𝑛 is denoted by tr 𝐴 𝑛 = 𝑛L1 . By (𝐴󸀠󸀠𝑛 ) = SLCirc (𝑦1󸀠󸀠 , 𝑦2󸀠󸀠 , . . . , 𝑦𝑛󸀠󸀠 ) (𝑛 ≥ 4) , (51)
𝑙𝑛
Definition 5 in [26] and (36), we have
where
𝑠 (𝐴 𝑛 ) ⩽ √2𝑛 (L𝑛 L𝑛+1 − 8). (46)
(−𝑑)𝑛−3
Since 𝑦1󸀠󸀠 = 1 − [(6 − 4𝑡)
𝑐𝑛−2
𝑛 𝑛
∑𝑎𝑖𝑗 = ∑ [𝑛 − (𝑘 − 1)] L𝑘 − ∑ (𝑘 − 1) L𝑘 𝑛−3
(−𝑑)𝑖−1
𝑖=𝑗̸ 𝑘=2 𝑘=2 + ∑ (L𝑛+2−𝑖 − 𝑡L𝑛+1−𝑖 ) ⋅ ],
(47) 𝑖=1 𝑐𝑖
𝑛 𝑛
= (𝑛 + 2) ∑ L𝑘 − 2 ∑ 𝑘L𝑘 , 𝑦𝑘󸀠󸀠 = −𝑦𝑛−𝑘+2
󸀠
𝑘=2 𝑘=2
2
by (6) and (8), (−𝑑)𝑛−𝑘−3+𝑖
= ∑ (L1+𝑖 − 𝑡L𝑖 ) ⋅ ,
𝑖=1 𝑐𝑛−𝑘−2+𝑖 (52)
∑𝑎𝑖𝑗 = (𝑛 + 2) (L𝑛+2 − 6) − 2 (𝑛L𝑛+2 − L𝑛+3 − 6 + L4 )
𝑖=𝑗̸ (𝑘 = 2, 3, . . . , 𝑛 − 2) .
= 2L𝑛+4 − 𝑛L𝑛+2 − 6𝑛 − 20. 󸀠󸀠 1
𝑦𝑛−1 = −𝑦3󸀠 = (6 − 4𝑡) ⋅ ,
(48) 𝑐
By Lemma 6 in [26], we have 𝑦𝑛󸀠󸀠 = −𝑦2󸀠
1 󵄨󵄨 󵄨 𝑛−2
(−𝑑)𝑖−1
𝑆 (𝐴 𝑛 ) ≥ 󵄨2L − 𝑛L𝑛+2 − 6𝑛 − 20󵄨󵄨󵄨 . (49)
𝑛 − 1 󵄨 𝑛+4 = 𝑡 + ∑ (L𝑛+1−𝑖 − 𝑡L𝑛−𝑖 ) ⋅
𝑐𝑖
.
𝑖=1

5. Norm and Spread of Skew Left

4. Determinant and Inverse of Skew Left Circulant Matrix with the Sum of Fibonacci
Circulant Matrix with the Sum of Fibonacci and Lucas Numbers
and Lucas Numbers
Theorem 13. Let 𝐴󸀠󸀠𝑛 = SLCirc(L1 , . . . , L𝑛 ) be a skew left
In this section, let 𝐴󸀠󸀠𝑛
= SLCirc(L1 , . . . , L𝑛 ) be a skew circulant matrix. Then three kinds norms of 𝐴󸀠󸀠𝑛 are given by
left circulant matrix. By using the obtained conclusions in
󵄩󵄩 󸀠󸀠 󵄩󵄩 󵄩 󵄩
Section 2, we give a determinant explicit formula for the 󵄩󵄩𝐴 𝑛 󵄩󵄩 = 󵄩󵄩󵄩𝐴󸀠󸀠𝑛 󵄩󵄩󵄩 = L𝑛+2 − 4,
matrix 𝐴󸀠󸀠𝑛 . And then, we prove that 𝐴󸀠󸀠𝑛 is an invertible matrix 󵄩 󵄩1 󵄩 󵄩∞
(53)
for any positive interger 𝑛. The inverse of the matrix 𝐴󸀠󸀠𝑛 is also 󵄩󵄩 󸀠󸀠 󵄩󵄩
󵄩󵄩𝐴 𝑛 󵄩󵄩 = √𝑛 (L𝑛 L𝑛+1 − 4).
presented. 󵄩 󵄩𝐹
According to Lemmas 4 and 5 in [15] and Theorems 3, 4,
and 6, we can obtain the following theorems. Proof. Using the similar method in Theorem 7, the conclu-
sion is obtained.
Theorem 10. Let 𝐴󸀠󸀠𝑛 = SLCirc(L1 , . . . , L𝑛 ) be a skew left
circulant matrix; then Theorem 14. Let 𝐴󸀠󸀠󸀠𝑛 = SLCirc(L1 , −L2 , . . . , −L𝑛−1 , L𝑛 ) be
an odd-order alternative skew left circulant matrix; then
det 𝐴󸀠󸀠𝑛 = (−1)𝑛(𝑛−1)/2
𝑛
󵄩󵄩 󸀠󸀠󸀠 󵄩󵄩
𝑛−2 󵄩󵄩𝐴 𝑛 󵄩󵄩 = ∑L𝑖 = L𝑛+2 − 4. (54)
× 2 [2 + 𝑡L𝑛 + ∑ (𝑡L1+𝑖 − L2+𝑖 ) 𝑥𝑛−1−𝑖 ] ⋅ 𝑐𝑛−2 󵄩 󵄩2
𝑖=1
𝑘=1
Proof. According to Lemma 2,
(𝑛 ≥ 3) ,
(50) 󵄨󵄨 𝑛 󵄨󵄨
󵄨󵄨 (𝑗−(1/2))(𝑘−1) 󵄨󵄨󵄨
𝜆 𝑗 (𝐴󸀠󸀠󸀠 󵄨 𝑖−1
𝑛 ) = ± 󵄨󵄨∑(−1) L𝑖 𝜔
󵄨󵄨𝑖=1
󵄨󵄨 ,
󵄨󵄨
(55)
where L𝑛 is the 𝑛th sum of Fibonacci and Lucas number. 󵄨 󵄨
8 Abstract and Applied Analysis

for 𝑗 = 1, 2, . . . , (𝑛 − 1)/2, and 6. Conclusion

𝑛 We discuss the invertibility of the skew circulant type
𝜆 (𝑛+1)/2 (𝐴󸀠󸀠󸀠
𝑛 ) = ∑L 𝑖 . (56) matrices with the sum of Fibonacci and Lucas numbers
𝑖=1 and present the determinant and the inverse matrices by
constructing the transformation matrices. The four kinds of
So
norms and bounds for the spread of these matrices are given,
󵄨󵄨 󵄨󵄨 𝑛 󵄨󵄨 󵄨 respectively.
󵄨󵄨𝜆 𝑗 (𝐴󸀠󸀠󸀠 ) 󵄨󵄨 ≤ ∑ 󵄨󵄨(−1)𝑖−1 L𝑖 (−1)𝑖−1 󵄨󵄨󵄨
󵄨 𝑛 󵄨 󵄨 󵄨
𝑖=1
(57) Conflict of Interests
𝑛
𝑛−1
= ∑L𝑖 , (𝑗 = 1, 2, . . . , ). The authors declare that there is no conflict of interests
𝑖=1 2
regarding the publication of this paper.
By (56) and (57), we have
Acknowledgments
󵄨 󵄨 𝑛
max 󵄨󵄨󵄨󵄨𝜆 𝑖 (𝐴󸀠󸀠󸀠
𝑛 )󵄨󵄨󵄨󵄨 = ∑L𝑖 . (58) The research is supported by the Development Project of
0≤𝑖≤𝑛+1
𝑖=1 Science & Technology of Shandong Province (Grant no.
2012GGX10115) and the AMEP of Linyi University, China.
Since all skew left circulant matrices are symmetrical, by
Lemma 7 in [26], (6), and (58), we obtain
󵄩󵄩 󸀠󸀠󸀠 󵄩󵄩
References
󵄩󵄩𝐴 𝑛 󵄩󵄩 = L𝑛+2 − 4. (59)
󵄩 󵄩2 [1] D. Bertaccini and M. K. Ng, “Skew circulant preconditioners
for systems of LMF-based,” in Numerical Analysis and Its
Applications, Lecture Notes in Computer Science, pp. 93–101,
2001.
Theorem 15. Let 𝐴󸀠󸀠𝑛 = SLCirc(L1 , . . . , L𝑛 ) be a skew left
[2] J. R. Claeyssen, M. Davila, and T. Tsukazan, “Factor circulant
circulant matrix; then the bounds for the spread of 𝐴󸀠󸀠𝑛 are
block matrices and even order undamped matrix differential
equations,” Matemática Aplicada e Computacional, vol. 3, no. 1,
{ 2 2
{ √ pp. 81–96, 1983.
󸀠󸀠 { 𝑀−
{ 𝑁 , if 𝑛 is odd,
𝑛 1 [3] B. Karasözen and G. Şimşek, “Energy preserving integration
2L𝑛 ≤ 𝑠 (𝐴 𝑛 ) ≤ { (60)
{
{ 2 2 of bi-Hamiltonian partial differential equations,” Applied and
{√ 𝑀 − 𝑁 , if 𝑛 is even, Engineering Mathematics, vol. 26, no. 12, pp. 1125–1133, 2013.
{ 𝑛 2
[4] A. Meyer and S. Rjasanow, “An effective direct solution method
where 𝑀 = 2𝑛(L𝑛 L𝑛+1 − 4), 𝑁1 = L𝑛−1 − L−1 , 𝑁2 = 0. for certain boundary element equations in 3D,” Mathematical
Methods in the Applied Sciences, vol. 13, no. 1, pp. 43–53, 1990.
Proof. Since 𝐴󸀠󸀠𝑛 is a symmetric matrix, by Lemma 6 in [26], [5] S. J. Guo, Y. M. Chen, and J. H. Wu, “Equivariant normal forms
we get 𝑠(𝐴󸀠󸀠𝑛 ) ≥ 2max𝑖=𝑗̸|𝑎𝑖𝑗 | = 2L𝑛 . for parameterized delay differential equations with applications
to bifurcation theory,” Acta Mathematica Sinica, vol. 28, no. 4,
The trace of 𝐴󸀠󸀠𝑛 is, if 𝑛 is odd, then tr(𝐴󸀠󸀠𝑛 ) = L1 − L2 + pp. 825–856, 2012.
L3 − ⋅ ⋅ ⋅ + L𝑛 = L1 + L1 + L3 + ⋅ ⋅ ⋅ + L𝑛−2 = 4 + ∑𝑛−3 𝑖=1 L𝑖 , by [6] X. Q. Jin, S. L. Lei, and Y. Wei, “Circulant preconditioners
(6), we have for solving singular perturbation delay differential equations,”
Numerical Linear Algebra with Applications, vol. 12, no. 2-3, pp.
tr (𝐴󸀠󸀠𝑛 ) = L𝑛−1 − L−1 = 𝑁1 . (61) 327–336, 2005.
[7] V. C. Liu and P. P. Vaidyanathan, “Circulant and skew-circulant
Let 𝑀 = 2𝑛(L𝑛 L𝑛+1 −4); then by Definition 5 in [26], (53), matrices as new normal-form realization of IIR digital filters,”
and (61), we obtain IEEE Transactions on Circuits and Systems, vol. 35, no. 6, pp.
625–635, 1988.
2 [8] M. J. Narasimha, “Linear convolution using skew-cyclic convo-
𝑠 (𝐴󸀠󸀠𝑛 ) ⩽ √ 𝑀 − 𝑁12 . (62)
𝑛 lutions,” IEEE Signal Processing Letters, vol. 14, no. 3, pp. 173–176,
2007.
If 𝑛 is even, then [9] D. Q. Fu, Z. L. Jiang, Y. F. Cui, and S. T. Jhang, “New fast
algorithm for optimal design of block digital filters by skew-
tr (𝐴󸀠󸀠𝑛 ) = 𝑁2 = L1 − L1 + L3 − L3 + ⋅ ⋅ ⋅ − L𝑛−1 = 0, (63) cyclic convolution,” IET Signal Processing, vol. 8, no. 6, pp. 633–
638, 2014.
hence [10] H. Karner, J. Schneid, and C. W. Ueberhuber, “Spectral decom-
position of real circulant matrices,” Linear Algebra and Its
tr (𝐴󸀠󸀠𝑛 ) = 𝑁2 = 0. (64) Applications, vol. 367, pp. 301–311, 2003.
[11] J. Li, Z. Jiang, N. Shen, and J. Zhou, “On optimal backward
So the result follows. perturbation analysis for the linear system with skew circulant
Abstract and Applied Analysis 9

coefficient matrix,” Computational and Mathematical Methods

in Medicine, vol. 2013, Article ID 707381, 7 pages, 2013.
[12] P. J. Davis, Circulant Matrices, John Wiley & Sons, New York,
NY, USA, 1979.
[13] Z. L. Jiang and Z. X. Zhou, Circulant Matrices, Chengdu
Technology University, Chengdu, China, 1999.
[14] S. Q. Shen, J. M. Cen, and Y. Hao, “On the determinants
and inverses of circulant matrices with Fibonacci and Lucas
numbers,” Applied Mathematics and Computation, vol. 217, no.
23, pp. 9790–9797, 2011.
[15] Y. Gao, Z. L. Jiang, and Y. P. Gong, “On the determinants
and inverses of skew circulant and skew left circulant matrices
with Fibonacci and Lucas numbers,” WSEAS Transactions on
Mathematics, vol. 12, no. 4, pp. 472–481, 2013.
[16] Z. L. Jiang, J. J. Yao, and F. L. Lu, “On skew circulant type matri-
ces involving any continuous Fibonacci numbers,” Abstract and
Applied Analysis, vol. 2014, Article ID 483021, 10 pages, 2014.
[17] X. Y. Jiang and K. Hong, “Exact determinants of some special
circulant matrices involving four kinds of famous numbers,”
Abstract and Applied Analysis, vol. 2014, Article ID 273680, 12
pages, 2014.
[18] Z. Jiang and D. Li, “The invertibility, explicit determinants, and
inverses of circulant and left circulant and g-circulant matri-
ces involving any continuous Fibonacci and Lucas numbers,”
Abstract and Applied Analysis, vol. 2014, Article ID 931451, 14
pages, 2014.
[19] S. Solak, “On the norms of circulant matrices with the Fibonacci
and Lucas numbers,” Applied Mathematics and Computation,
vol. 160, no. 1, pp. 125–132, 2005.
[20] A. İpek, “On the spectral norms of circulant matrices with
classical Fibonacci and Lucas numbers entries,” Applied Mathe-
matics and Computation, vol. 217, no. 12, pp. 6011–6012, 2011.
[21] L. Mirsky, “The spread of a matrix,” Mathematika, vol. 3, pp. 127–
130, 1956.
[22] R. Sharma and R. Kumar, “Remark on upper bounds for the
spread of a matrix,” Linear Algebra and its Applications, vol. 438,
no. 11, pp. 4359–4362, 2013.
[23] J. Wu, P. Zhang, and W. Liao, “Upper bounds for the spread of a
matrix,” Linear Algebra and Its Applications, vol. 437, no. 11, pp.
2813–2822, 2012.
[24] C. R. Johnson, R. Kumar, and H. Wolkowicz, “Lower bounds for
the spread of a matrix,” Linear Algebra and its Applications, vol.
71, pp. 161–173, 1985.
[25] Z. Jiang, Y. Gong, and Y. Gao, “Invertibility and explicit inverses
of circulant-type matrices with $k$-Fibonacci and $k$-Lucas
numbers,” Abstract and Applied Analysis, vol. 2014, Article ID
238953, 9 pages, 2014.
[26] J. Li, Z. L. Jiang, and F. L. Lu, “Determinants, norms, and the
spread of circulant matrices with Tribonacci and generalized
Lucas numbers,” Abstract and Applied Analysis, vol. 2014, Article
ID 381829, 9 pages, 2014.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2015, Article ID 965829, 8 pages
http://dx.doi.org/10.1155/2015/965829

Research Article
A Note on the Existence of a Smale Horseshoe in the Planar
Circular Restricted Three-Body Problem

Xuhua Cheng and Zhikun She

SKLSDE, LMIB and School of Mathematics and Systems Science, Beihang University, Beijing 100191, China

Correspondence should be addressed to Xuhua Cheng; chengxuhua88@163.com

Received 26 June 2014; Accepted 24 September 2014

Academic Editor: Tonghua Zhang

Copyright © 2015 X. Cheng and Z. She. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

It has been proved that, in the classical planar circular restricted three-body problem, the degenerate saddle point processes
transverse homoclinic orbits. Since the standard Smale-Birkhoff theorem cannot be directly applied to indicate the chaotic dynamics
of the Smale horseshoe type, we in this note alternatively apply the Conley-Moser conditions to analytically prove the existence of
a Smale horseshoe in this classical restricted three-body problem.

1. Introduction and Preliminaries 𝑑𝑝𝜌 𝑝𝜑2 (1 − 𝜇) (𝜌 − 𝜇 cos 𝜑)

= −
𝑑𝜏 𝜌3 (𝜌2 + 𝜇2 − 2𝜌𝜇 cos 𝜑)
3/2
Few bodies problems [1–7] have been studied for long time
in celestial mechanics, either as simplified models of more
complex planetary systems or as benchmark models where 𝜇 [𝜌 + (1 − 𝜇) cos 𝜑]
new mathematical theories can be tested. The three-body − 3/2
,
2
problem has been the source of inspiration and study in celes- [𝜌2 + (1 − 𝜇) + 2𝜌 (1 − 𝜇) cos 𝜑]
tial mechanics since Newton and Euler [8–14]. Especially,
the following classical planar circular restricted three-body
model has been extensively studied in the literature. Let two 𝑑𝜑 𝑝𝜑
= − 1,
particles 𝑃1 and 𝑃2 , of mass 1 − 𝜇 and 𝜇, move uniformly 𝑑𝜏 𝜌2
in a circular orbit about their common center of mass with
angular velocity 𝜔. The orbit is located in the 𝑂𝑥𝑦 plane of the
𝑑𝑝𝜑
inertial frame of reference and the common center of mass is = −𝜇 (1 − 𝜇)
in the origin. The particle 𝑃3 of infinitesimal mass 𝑚3 moves 𝑑𝜏
in the gravitational field generated by 𝑃1 and 𝑃2 . Note that
since the mass of 𝑃3 is so small, its effects on other three 1
× 𝜌 sin 𝜑 [ 3/2
particles can be ignored. Without loss of generality, assume 2 2
that, in the 𝑂𝑥𝑦 plane of the rotating frame of reference, the [ [𝜌 + 𝜇 − 2𝜌𝜇 cos 𝜑]
particles 𝑃1 and 𝑃2 rest at the points (𝜇, 0) and (𝜇 − 1, 0),
respectively. By denoting their polar coordinates by 𝜌 and 1 ],
− 3/2
𝜑 and using the polar angle 𝜏 = 𝜔𝑡 as a new independent [𝜌2
2
+ (1 − 𝜇) + 2𝜌 (1 − 𝜇) cos 𝜑]
variable, the equation of motion of the infinitesimal particle ]
(1)
𝑃3 can be written as follows:
𝑑𝜌 where 𝑝𝜌 and 𝑝𝜑 are momenta canonically conjugate to the
= 𝑝𝜌 , coordinates 𝜌 and 𝜑, respectively.
𝑑𝜏
2 Abstract and Applied Analysis

The Hamiltonian of the system (1) is Assumption 1. 0 ≤ 𝜇V 𝜇ℎ < 1 and, for each 𝑖 ∈ {1, 2, . . . , 𝑁},
𝑓 maps 𝐻𝑖 homeomorphically onto 𝑉𝑖 ; that is, 𝑓(𝐻𝑖 ) = 𝑉𝑖 .
𝑝𝜑2 Moreover, the horizontal boundaries of 𝐻𝑖 are mapped to the
1
𝐻= (𝑝𝜌2 + 2 − 2𝑝𝜑 ) horizontal boundaries of 𝑉𝑖 and the vertical boundaries of 𝐻𝑖
2 𝜌 are mapped to the vertical boundaries of 𝑉𝑖 .
1−𝜇
− 1/2 (2) Assumption 2. 𝐷𝑓(S𝑢H ) ⊂ S𝑢V and 𝐷𝑓−1 (S𝑠V ) ⊂ S𝑠H .
(𝜌2 + 𝜇2 − 2𝜌𝜇 cos 𝜑)
Moreover, there exists a positive number 𝜆 satisfying 0 < 𝜆 <
𝜇 1 − 𝜇V 𝜇ℎ such that
− 1/2
.
2
(𝜌2 + (1 − 𝜇) + 2𝜌 (1 − 𝜇) cos 𝜑)
(1) if (𝜉𝑧0 , 𝜂𝑧0 ) ∈ S𝑢𝑧0 and (𝜉𝑓(𝑧0 ) , 𝜂𝑓(𝑧0 ) ) ≐ 𝐷𝑓(𝑧0 )(𝜉𝑧0 ,
For the above classical model, Xia [4] has showed, by 𝜂𝑧0 ) ∈ S𝑢𝑓(𝑧0 ) , then |𝜂𝑓(𝑧0 ) | ≥ (1/𝜆)|𝜂𝑧0 |;
proper coordinate change for transforming the points at
infinity to the origin (i.e, the McGehee transformation [2]), (2) if (𝜉𝑧0 , 𝜂𝑧0 ) ∈ S𝑠𝑧0 and (𝜉𝑓−1 (𝑧0 ) , 𝜂𝑓−1 (𝑧0 ) ) ≐ 𝐷𝑓−1 (𝑧0 )
that there is a periodic solution at infinity. Moreover, from
(𝜉𝑧0 , 𝜂𝑧0 ) ∈ S𝑠𝑓−1 (𝑧0 ) , then |𝜉𝑓−1 (𝑧0 ) | ≥ (1/𝜆)|𝜉𝑧0 |.
[2, 4], we know that this periodic solution is a degenerate
saddle in the sense [2] that, for the Poincaré map of the
Based on Assumptions 1 and 2, we directly have the
periodic orbit introduced at infinity, its derivative (i.e., the
following.
Jacobian) at the origin is the identity.
Further, Xia [4] and Zhu and Xiang [12] both proved the
existence of transversal homoclinic orbits by the Melnikov
method to the periodic solution at infinity, which corre- Lemma 3 (see [16]). If the map 𝑓 satisfies Assumptions 1 and 2,
sponds to the origin under the coordinate change. However, then 𝑓 has an invariant Cantor set, on which it is topologically
since the origin is a degenerate fixed point, the standard conjugate to a full shift on 𝑁 symbols and has
Smale-Birkhoff theorem [15] cannot be directly applied to
indicate the existence of a Smale horseshoe. This problem (i) a countable infinity of periodic orbits of arbitrarily high
has also been pointed out by Dankowicz and Holmes [6] period;
and Llibre and Perez-Chavela [8]. Thus, in this present note,
we try to alternatively apply the Conley-Moser conditions (ii) an uncountable infinity of nonperiodic orbits;
to analytically prove the existence of a Smale horseshoe in
the above classical model. For this, we introduce the Conley-
Moser conditions [16] as follows. (iii) a dense orbit.
Let 𝑓 : 𝐷 󳨃 → R2 be an invertible map, where 𝐷 = {(𝑥, 𝑦) ∈
Remark 4 (see [16–18]). If 𝑓 satisfies Assumption 2, we call
R | 0 ≤ 𝑥 ≤ 1, 0 ≤ 𝑦 ≤ 1}, and 𝑓 is at least 𝐶1 . For two given
2
that 𝑓 satisfies the (𝜇ℎ , 𝜇V )-cone condition.
𝜇V > 0 and 𝜇ℎ > 0, let 𝐾 = {1, 2, . . . , 𝑁} (𝑁 ≥ 2) be an index
set, let 𝐻1 , . . . , 𝐻𝑁 be the 𝑁 disjoint 𝜇ℎ -horizontal strips, and
𝑉1 , . . . , 𝑉𝑁 be the 𝑁 disjoint 𝜇V -vertical strips. For each 𝑖, 𝑗 ∈ 2. Main Result
𝐾, denote 𝑓(𝐻𝑖 ) ⋂ 𝐻𝑗 as 𝑉𝑗𝑖 and 𝐻𝑖 ⋂ 𝑓−1 (𝐻𝑗 ) as 𝐻𝑖𝑗 . Clearly,
𝐻𝑖𝑗 = 𝑓−1 (𝑉𝑗𝑖 ). Define H = ⋃𝑖,𝑗∈𝐾 𝐻𝑖𝑗 and V = ⋃𝑖,𝑗∈𝐾 𝑉𝑗𝑖 . It In this section, we will analytically prove the existence of
is also obvious that 𝑓(H) = V. a Smale horseshoe in the classical planar circular restricted
For an arbitrary point 𝑧0 = (𝑥0 , 𝑦0 ) ∈ H ⋃ V, let three-body problem introduced in Section 1, arriving at the
(𝜉𝑧0 , 𝜂𝑧0 ) be a vector emanating from the point 𝑧0 in the following theorem.
tangent space of 𝑧0 . The stable sector at 𝑧0 is then defined as
S𝑠𝑧0 = {(𝜉𝑧0 , 𝜂𝑧0 ) ∈ R2 | |𝜂𝑧0 | ≤ 𝜇ℎ |𝜉𝑧0 |}. Similarly, the unstable
Theorem 5. For the classical planar circular restricted three-
sector at 𝑧0 is defined as S𝑢𝑧0 = {(𝜉𝑧0 , 𝜂𝑧0 ) ∈ R2 | |𝜉𝑧0 | ≤
body problem introduced in Section 1, when the mass ratio 𝜇 is
𝜇V |𝜂𝑧0 |}. By taking the union of the stable and unstable sectors sufficiently small, there exists a Smale horseshoe and thus the
over all points in H and V, we can define sector bundles as system (1) processes chaotic dynamics of the Smale horseshoe
follows: type.
S𝑠H = ⋃ S𝑠𝑧0 , S𝑠V = ⋃ S𝑠𝑧0 ;
𝑧0 ∈H 𝑧0 ∈V In order to prove Theorem 5, we will construct an invert-
(3) ible map 𝑓 and then verify that this 𝑓 satisfies the Conley-
S𝑢H = ⋃ S𝑢𝑧0 , S𝑢V = ⋃ S𝑢𝑧0 . Moser conditions.
𝑧0 ∈H 𝑧0 ∈V

Then, the Conley-Moser conditions for the map 𝑓 are 2.1. Construction of an Invertible Map 𝑓. According to
described by the following two assumptions. the McGehee transformation 𝜌 = 1/𝑥2 , 𝑝𝜌 = 𝑦 [2],
Abstract and Applied Analysis 3

the Hamiltonian of the system (1) can be reformulated as Moreover, the Poincaré map of the periodic orbit (𝑥, 𝑦) =
follows: (0, 0) has the form
1 2
𝐻= (𝑦 + 𝑥4 𝑝𝜑2 − 2𝑝𝜑 ) 𝑃0 : 𝑥 󳨀→ 𝑥 + 𝑘1 𝑥3 (𝑦 + 𝑟1 (𝑥, 𝑦))
2
(7)
(1 − 𝜇) 𝑥2 𝑦 󳨀→ 𝑦 + 𝑘2 𝑥3 (𝑥 + 𝑟2 (𝑥, 𝑦)) ,
− 1/2
(1 + 𝑥4 𝜇2 − 2𝑥2 𝜇 cos 𝜑)
where 𝑘1 = 𝜋, 𝑘2 = 2𝜋(1 − 2𝜇), and 𝑟1 , 𝑟2 are real analytic and
𝜇𝑥 2 contain terms of at least second order in (𝑥, 𝑦).
− 1/2
. Using polar coordinates (𝜌, 𝜃), the Poincaré map 𝑃0 can
2
[1 + 𝑥4 (1 − 𝜇) + 2𝑥2 (1 − 𝜇) cos 𝜑] be reformulated as
(4)
𝑃0 : 𝑟 󳨀→ 𝑟 − 𝑘1 𝑟4 cos4 𝜃 ((4𝜇 − 3) sin 𝜃 + 𝑜 (𝑟))
Thus, the system (1) can be reformulated as
𝜃 󳨀→ 𝜃 − 𝑘2 𝑟3 cos3 𝜃
𝑑𝑥 1 (8)
= − 𝑥3 𝑦, 1
𝑑𝜏 2 ×( sin2 𝜃 − cos2 𝜃 + 𝑜 (𝑟)) .
2 (1 − 2𝜇)
𝑑𝑦 2 6
(1 − 𝜇) (1 − 𝜇𝑥2 cos 𝜑) 𝑥4
= 𝑝𝜑 𝑥 − 3/2
𝑑𝜏 (1 + 𝜇2 𝑥4 − 2𝜇𝑥2 cos 𝜑) According to formula (8), by making the following linear
transformation:
𝜇 [1 + (1 − 𝜇) 𝑥2 cos 𝜑] 𝑥4
− 3/2
, 𝑥 = 𝑢 + V,
2
[1 + (1 − 𝜇) 𝑥4 + 2 (1 − 𝜇) 𝑥2 cos 𝜑] (9)
𝑦 = −√2 (1 − 2𝜇) (𝑢 − V) ,
𝑑𝜑
= 𝑝𝜑 𝑥4 − 1,
𝑑𝜏
the system (6) can be reformulated as follows:
𝑑𝑝𝜑 4
= 𝜇 (1 − 𝜇) 𝑥 𝑑𝑢
𝑑𝜏 = (𝑢 + V)3 𝑘3 𝑢,
𝑑𝜏
1
× sin 𝜑 [ 3/2
𝑑V
= −(𝑢 + V)3 (𝑘3 V + ℎ1 (𝑢, V, 𝜑, 𝜇)) , (10)
2 4 2
[ [1 + (1 − 𝑢) 𝑥 + 2 (1 − 𝜇) 𝑥 cos 𝜑] 𝑑𝜏
𝑑𝜑
1 = −1 + ℎ2 (𝑢, V, 𝜑, 𝜇) ,
− ]. 𝑑𝜏
3/2
(1 + 𝜇2 𝑥4 − 2𝜇𝑥2 cos 𝜑) ]
(5) where 𝑘3 = √2(1 − 2𝜇)/2. Due to the symmetry of the
problem, we subsequently restrict our discussion to the
For the energy surface 𝐻 = ℎ, where ℎ is a constant, positive quadrant.
there exists a 2𝜋-periodic solution with respect to 𝜑; that is, We neglect the higher order terms of (10) and then obtain
(𝑥, 𝑦, 𝑝𝜑 ) = (0, 0, −ℎ). Further, near this periodic solution, that 𝑑𝑢/𝑑V = −𝑢/V. It is clear that its solution remains on the
by solving the Jacobi integral for 𝑝𝜑 , we have 𝑝𝜑 = −ℎ + hyperbolae 𝑢V = 𝑐0 > 0, where 𝑐0 is a constant. We substitute
]1 (𝑥, 𝑦, 𝜑), where ]1 (𝑥, 𝑦, 𝜑) is second order in 𝑥 and 𝑦 and V = 𝑐0 /𝑢 into the first expression of (10) and neglect the higher
2𝜋-periodic with respect to 𝜑. order terms, arriving at 𝑑𝑢/𝑑𝜑 = −𝑘3 ((𝑢2 + 𝑐0 )3 /𝑢2 ).
Thus, the system (5) can be further reformulated as Let Σ be a plane transversal to the periodic orbit 𝛾𝜇 at the
origin (0, 0) and let 𝑈0 be a sufficiently small neighborhood
𝑑𝑥 1 of the origin (0, 0) in the plane Σ. For an arbitrary but fixed
= − 𝑥3 𝑦,
𝑑𝜏 2 point (𝑢0 , V0 ) ∈ 𝑈0 \ {(0, 0)}, we define 𝑇𝜑 (𝑢0 , V0 ) = (𝑢𝜑 , V𝜑 )
𝑑𝑦 with 𝑇0 (𝑢0 , V0 ) = (𝑢0 , V0 ).
= − (1 − 2𝜇) (𝑥4 + 𝑔1 (𝑥, 𝑦, 𝜑, 𝜇)) , (6) Assume that 𝑢𝜑 = √𝑐0 tan(𝜙𝜑 /4); then V𝜑 =
𝑑𝜏
√𝑐0 cot(𝜙𝜑 /4), where 𝜙𝜑 is an auxiliary variable. Substituting
𝑑𝜑 𝑢𝜑 = √𝑐0 tan(𝜙𝜑 /4) into 𝑑𝑢/𝑑𝜑 = −𝑘3 ((𝑢2 + 𝑐0 )3 /𝑢2 ), we can
= −1 + 𝑔2 (𝑥, 𝑦, 𝜑, 𝜇) ,
𝑑𝜏 obtain
where 𝑔1 and 𝑔2 are 2𝜋-periodic with respect to 𝜑, 𝑔1 is the
third order in (𝑥, 𝑦), and 𝑔2 is fourth order in (𝑥, 𝑦). 𝜙𝜑 − sin 𝜙𝜑 = 𝑘0 − 32𝑘𝑐03/2 𝜑, 𝑘0 = 𝜙0 − sin 𝜙0 , (11)
From [4, 12], the origin (0, 0) can be regarded as a periodic
orbit 𝛾𝜇 with period 2𝜋 with respect to 𝜑 in the system (6). where 𝑐0 = 𝑢0 V0 and 𝜙0 = 4 arctan √𝑢0 /V0 .
4 Abstract and Applied Analysis

Moreover, we can calculate

𝜕𝑢𝜑 𝜕𝑢𝜑
[ 𝜕𝑢 𝜕V0 ]
[ 0 ]
𝐷𝑇𝜑 = [ 𝜕V𝜑 𝜕V𝜑 ]
[ ]
𝜕𝑢 𝜕V0
[ 0 ] (12)
𝑢𝜑 3 𝑢0 𝑢𝜑 3
[ 2𝑢 (1 + Δ − 3𝑘3 (𝑢𝜑 + V𝜑 ) 𝜑) 2𝑐
(1 − Δ − 3𝑘3 (𝑢𝜑 + V𝜑 ) 𝜑)]
=[
[ 𝑐
0
3 𝑢0 3
],
]
(1 − Δ + 3𝑘3 (𝑢𝜑 + V𝜑 ) 𝜑) (1 + Δ + 3𝑘3 (𝑢𝜑 + V𝜑 ) 𝜑)
2𝑢
[ 0 𝜑𝑢 2𝑢 𝜑 ]

where Δ = ((𝑢𝜑 + V𝜑 )/(𝑢0 + V0 ))3 . Clearly, det 𝐷𝑇𝜑 = Δ ≠

0. there exists a sufficiently small 𝛿1 such that 𝐹𝐵̃V ⋂ 𝐵ℎ = {𝑝},
For the approximate system obtained by neglecting the (𝜕𝐹𝑉/𝜕V)|𝐵̃V ≠ 0, 𝐹𝐵̃V ⋂ 𝜕𝐵ℎ = 0, 𝐹𝜕𝐵̃V ⋂ 𝐵ℎ = 0. Moreover,
higher order terms in the system (10), we can describe the let 𝜕V 𝐵 = {(𝑢, V) ∈ 𝐵 | 𝑢 = 1 ± 𝛿1 } and 𝜕ℎ 𝐵̃ = {(𝑢, V) ∈
Poincaré map 𝑃 defined over the plane Σ by using the
𝐵̃ | V = 1 ± 𝛿1 }. We can further obtain that there exists a
truncated flow near the degenerate saddle as follows:
sufficiently small 𝛿2 such that (𝜕𝐹𝑉/𝜕V)|𝐵̃ ≠ 0, 𝐹𝐵̃ ⋂ 𝜕V 𝐵 = 0,
𝑃 : (𝑢0 , V0 ) 󳨃
󳨀→ (𝑢2𝜋 , V2𝜋 ) , where (𝑢0 , V0 ) ∈ 𝑈0 . (13) ̃
𝐹(𝜕ℎ 𝐵) ⋂ 𝐵 = 0.
Based on 𝑃 and 𝐹, we construct a successor map Δ 𝑘 =
Since the terms neglected in (10) are both 𝑜(𝑢4 , V4 ) and 𝑂(𝜇), 𝐹 ∘ 𝑃𝑘 : 𝐷𝑘 → 𝑈𝑝 . Further, we define another map 𝑓
we can use this Poincaré map 𝑃 to approximate 𝑃0 .
over the set ⋃𝑘 𝐷𝑘 such that 𝑓|𝐷𝑘 = Δ 𝑘 . Clearly, 𝑓 is also
Letting 𝑢0 = √𝑐0 tan(𝜙0 /4) and V0 = √𝑐0 cot(𝜙0 /4), then
a homeomorphism.
we can obtain
𝜙2𝑘𝜋 𝜙 2.2. Proofs of Some Propositions for 𝑓. In order to prove
𝑃𝑘 (𝑢0 , V0 ) ≗ (𝑢𝑘 , V𝑘 ) = (√𝑐0 tan , √𝑐0 cot 2𝑘𝜋 ) . (14)
4 4 that 𝑓 satisfies the Conley-Moser conditions, we need to
For the system (10), the coordinate axis V = 0 corresponds introduce one lemma and then prove four propositions in this
𝑠
to the local stable manifold 𝑊loc (𝛾𝜇 ) and 𝑢 = 0 corresponds to subsection.
𝑢
the unstable manifold 𝑊loc (𝛾𝜇 ), respectively. Moreover, from Lemma 6 (see [17, 18]). Consider two invertible linear opera-
[4, 12], when the mass ratio 𝜇 is sufficiently small, there exists tors of 𝑅1 × 𝑅1 into itself:
a transversal homoclinic orbit, denoted as 𝛾, of the periodic
orbit 𝛾𝜇 . Thus, there exist two points 𝑝 and 𝑞 such that 𝑝 ∈ 𝑎 𝑏 Λ 𝐸
𝐼=[ ], 𝐽=[ ], (15)
𝑠
𝑊loc 𝑢
(𝛾𝜇 ), 𝑞 ∈ 𝑊loc (𝛾𝜇 ), and 𝑝, 𝑞 ∈ Σ ⋂ 𝛾. For convenience, 𝑐 𝑑 𝐺 𝑀
by introducing a scale transformation, we can further assume where 𝑑𝑀 ≠ 0. Let 𝐿 > 0 be a constant such that the following
that 𝑝 = (1, 0) and 𝑞 = (0, 1). conditions hold:
We define 𝐵 = {(𝑢, V) | |𝑢 − 1| ≤ 𝛿1 , |V| ≤ 𝛿2 } and 󵄩󵄩 −1 󵄩󵄩
𝐵̃ = {(𝑢, V) | |𝑢| ≤ 𝛿2 , |V − 1| ≤ 𝛿1 } as the corresponding ‖𝐼‖ < 𝐿, 󵄩󵄩𝐼 󵄩󵄩 < 𝐿,
󵄩 󵄩
neighborhoods of 𝑝 and 𝑞, respectively. For sufficiently small 󵄨󵄨 −1 󵄨󵄨
󵄨󵄨𝑑 󵄨󵄨 < 𝐿, (16)
positive numbers 𝛿1 and 𝛿2 , 𝐵 and 𝐵̃ satisfy 𝑃𝐵 ⋂ 𝐵 = 0, 󵄨 󵄨
󵄨󵄨 󵄨
𝑃−1 𝐵̃ ⋂ 𝐵̃ = 0. Let 𝐷𝑘 = 𝑃−𝑘 𝐵̃ ⋂ 𝐵. When 𝑘 is sufficiently 󵄨󵄨𝐸𝑀−1 󵄨󵄨󵄨 < 𝐿.
󵄨 󵄨
large, 𝐷𝑘 ≠ 0. Moreover, we also can obtain 𝐷𝑘 ⋂ 𝐷𝑚 = 0 for
𝑘 ≠ 𝑚. Again let 𝐷 ̃𝑘 = 𝑃𝑘 𝐷𝑘 . When 𝑘 is sufficiently large, Then, for arbitrary 0 < 𝜇ℎ < 𝜇V−1 ≪ 1, there exists a positive
𝐷̃𝑘 ≠ 0. The relation between 𝐷𝑘 and 𝐷 ̃𝑘 can be seen from constant 𝛿0 , which is dependent on 𝐿, 𝜇ℎ and 𝜇V , such that if
Figure 1. the following conditions hold:
Since 𝑝, 𝑞 ∈ Σ ⋂ 𝛾, when 𝛿1 and 𝛿2 are sufficiently small, 󵄨󵄨󵄨𝑀−1 󵄨󵄨󵄨 < 𝛿 ,
󵄨󵄨 󵄨󵄨 0
every positive half-orbit of the system (10) that starts from 𝐵̃
intersects a neighborhood 𝑈𝑝 of the point 𝑝 at a point, where 󵄨󵄨 −1 󵄨 󵄨
󵄨󵄨󵄨Λ − 𝐸𝑀 𝐺󵄨󵄨󵄨 < 𝛿0 ,
𝑈𝑝 ⊂ Σ. This can be depicted by the map 𝐹 : 𝐵̃ → 𝑈𝑝 . It is 󵄨󵄨 󵄨
clear that 𝐹 is a 𝐶1 diffeomorphism. Let 𝐹(𝑢, V) = (𝐹𝑈, 𝐹𝑉). 󵄨󵄨Λ𝑀−1 󵄨󵄨󵄨 < 𝛿0 , (17)
󵄨 󵄨
Since the stable manifold and the unstable manifold of the 󵄨󵄨 󵄨
periodic orbit 𝛾𝜇 transversally intersect along 𝛾, we can obtain 󵄨󵄨𝐺𝑀−1 󵄨󵄨󵄨 < 𝛿0 ,
󵄨 󵄨
(𝜕𝐹𝑉/𝜕V)|𝑞 ≠ 0. 󵄨󵄨 −1 󵄨󵄨
Let 𝐵ℎ = 𝐵 ⋂{V = 0}, 𝐵̃V = 𝐵̃ ⋂{𝑢 = 0}, 𝜕𝐵ℎ = {𝑢 = 󵄨󵄨󵄨𝑐𝐸𝑀 󵄨󵄨󵄨 < 𝛿0 ,
1 ± 𝛿1 , V = 0}, and 𝜕𝐵̃V = {𝑢 = 0, V = 1 ± 𝛿1 }. Then, the linear map 𝐴 = 𝐼𝐽 satisfies the (𝜇ℎ , 𝜇V )-cone condition.
Abstract and Applied Analysis 5

Proposition 8. When 𝑘 is sufficiently large, 𝑃𝑘 satisfies the

(𝜇ℎ , 𝜇V )-cone condition.
̃k
D
Proof. Let 𝑁 ≥ 2 be an arbitrary but fixed integer. For
q sufficiently large 𝑘, let 𝐻𝑙 = 𝐷𝑙+𝑘−1 , 𝑉𝑙 = 𝑓(𝐷𝑙+𝑘−1 ), 𝑉𝑗𝑖 =
𝑃𝑘 𝐻𝑖 ⋂ 𝐻𝑗 , and 𝐻𝑖𝑗 = 𝐻𝑖 ⋂ 𝑃−𝑘 𝐻𝑗 , where 1 ≤ 𝑙, 𝑖, 𝑗 ≤ 𝑁.
̃
B
Moreover, let H = ⋃𝑖,𝑗 𝐻𝑖𝑗 and V = ⋃𝑖,𝑗 𝑉𝑖𝑗 .
pk For an arbitrary point 𝑧0 ∈ H ⋃ V, let (𝜉𝑧0 , 𝜂𝑧0 ) be a
vector emanating from the point 𝑧0 in the tangent space of 𝑧0 .
In addition, for given 𝜇ℎ and 𝜇V , let S𝑢𝑧0 = {(𝜉𝑧0 , 𝜂𝑧0 ) | |𝜉𝑧0 | <
𝜇V |𝜂𝑧0 |} be the unstable sector at 𝑧0 and let S𝑠𝑧0 = {(𝜉𝑧0 , 𝜂𝑧0 ) |
Dk |𝜂𝑧0 | < 𝜇ℎ |𝜉𝑧0 |} be the stable sector at 𝑧0 . Similar to Section 1,
we also have S𝑢H , S𝑢V , S𝑠H , and S𝑠V .
p In order to prove that 𝑃𝑘 satisfies the (𝜇ℎ , 𝜇V )-cone
condition, by Remark 4, we need to prove that 𝑃𝑘 satisfies
Assumption 2. That is, we need to prove the following:
B
(1) 𝐷𝑃𝑘 (S𝑢H ) ⊂ S𝑢V and 𝐷𝑃−𝑘 (S𝑠V ) ⊂ S𝑠H ;
̃𝑘 .
Figure 1: The relation between 𝐷𝑘 and 𝐷
(2) there exists a constant 𝜆 satisfying 0 < 𝜆 < 1 − 𝜇ℎ 𝜇V
such that |𝜂𝑃𝑘 (𝑧0 ) | ≥ 𝜆−1 |𝜂𝑧0 | if (𝜉𝑧0 , 𝜂𝑧0 ) ∈ S𝑢𝑧0 and
(𝜉𝑃𝑘 (𝑧0 ) , 𝜂𝑃𝑘 (𝑧0 ) ) ≐ 𝐷𝑃𝑘 (𝑧0 )(𝜉𝑧0 , 𝜂𝑧0 ) ∈ S𝑢𝑃𝑘 (𝑧 ) , where
0

By Lemma 6, we have the following proposition. (𝜉𝑃𝑘 (𝑧0 ) , 𝜂𝑃𝑘 (𝑧0 ) ) is a vector emanating from the point
𝑃𝑘 (𝑧0 ) in the tangent space of 𝑃𝑘 (𝑧0 ); |𝜉𝑃−𝑘 (𝑧0 ) | ≥
Proposition 7. For two arbitrary constants 𝜇ℎ and 𝜇V with 0 < 𝜆−1 |𝜉𝑧0 | if (𝜉𝑧0 , 𝜂𝑧0 ) ∈ S𝑠𝑧0 and (𝜉𝑃−𝑘 (𝑧0 ) , 𝜂𝑃−𝑘 (𝑧0 ) ) ≐
𝜇ℎ < 𝜇V−1 ≪ 1, when 𝑘 is sufficiently large, 𝑓|𝐷𝑘 satisfies the
𝐷𝑃−𝑘 (𝑧0 )(𝜉𝑧0 , 𝜂𝑧0 ) ∈ S𝑠𝑃−𝑘 (𝑧 ) , where (𝜉𝑃−𝑘 (𝑧0 ) , 𝜂𝑃−𝑘 (𝑧0 ) )
(𝜇ℎ , 𝜇V )-cone condition. 0

is a vector emanating from the point 𝑃−𝑘 (𝑧0 ) in the

Proof. Based on the chain rule on the derivative of a compos- tangent space of 𝑃−𝑘 (𝑧0 ).
ite function, we can obtain
First, we want to prove that 𝐷𝑃𝑘 (S𝑢H ) ⊂ S𝑢V . For this, it is
𝑘 sufficient to prove that, for an arbitrary 𝑧0 = (𝑢0 , V0 ) ∈ H with
𝐷𝑓|𝐷𝑘 = 𝐷𝐹 ⋅ 𝐷𝑃 = 𝐷𝐹 ⋅ 𝐷𝑇𝜑
(𝜉𝑧0 , 𝜂𝑧0 ) = (1, 𝜗) ∈ S𝑢𝑧0 , (𝜉𝑃𝑘 (𝑧0 ) , 𝜂𝑃𝑘 (𝑧0 ) ) = 𝐷𝑃𝑘 (𝑧0 )(𝜉𝑧0 , 𝜂𝑧0 ) ∈
𝜕𝐹𝑈 𝜕𝐹𝑈 𝜕𝑢𝜑 𝜕𝑢𝜑 S𝑢V .
[
[ 𝜕𝑢 𝜕V ] [ 𝜕𝑢0 𝜕V0 ] (18)
] Clearly, 𝑃𝑘 (𝑧0 ) ∈ V and 𝜗 is bounded. According
=[ ] 𝜕V
[ 𝜕𝐹𝑉 𝜕𝐹𝑉 ] . [ 𝜕V𝜑 ] ,
[ 𝜑 ] to the definitions of 𝑇𝜑 and 𝑃, 𝐷𝑃𝑘 (𝑧0 )(𝜉𝑧0 , 𝜂𝑧0 ) =
[ 𝜕𝑢 𝜕V ] [ 𝜕𝑢0 𝜕V0 𝐷𝑇2𝑘𝜋 (𝑧0 )(𝜉𝑧0 , 𝜂𝑧0 ) = ((𝜕𝑢2𝑘𝜋 /𝜕𝑢0 ) + (𝜕𝑢2𝑘𝜋 /𝜕V0 )𝜗,
]
(𝜕V2𝑘𝜋 /𝜕𝑢0 ) + (𝜕V2𝑘𝜋 /𝜕V0 )𝜗).
where 𝜑 = 2𝑘𝜋. Let 𝐿 = sup𝛽∈𝐵̃ {‖𝐷𝐹(𝛽)‖ , ‖𝐷𝐹(𝛽)−1 ‖, Since 𝛿1 and 𝛿2 for defining 𝐵 and 𝐵̃ are chosen to be
sufficiently small, 𝑢0 ≈ 1 and V0 ≤ 𝛿2 ≪ 1. Letting 𝑢0 = 𝛿
|𝜕𝐹𝑉/𝜕V|−1 }. Since 𝜕𝐹𝑉/𝜕V|𝐵̃ ≠ 0 and 𝐹 is 𝐶1 , we have 𝐿 < +∞. and 𝑐 = 𝑢0 V0 , then (𝑢2𝑘𝜋 , V2𝑘𝜋 ) ≈ (𝑐/𝛿, 𝛿). According to 𝐷𝑇𝜑
Let 𝑢0 = 𝛿 ≈ 1 and V0 = 𝑐0 /𝛿 ≪ 1. Then (𝑢𝑘 , V𝑘 ) ≈ (𝑐0 /𝛿, 𝛿). in Section 2.1, when 𝑘 is sufficiently large, we have
Let 𝐸 = 𝜕𝑢𝜑 /𝜕V0 , Λ = 𝜕𝑢𝜑 /𝜕𝑢0 , 𝐺 = 𝜕V𝜑 /𝜕𝑢0 , 𝑀 =
𝜕V𝜑 /𝜕V0 . Similar to the proof of Condition 1 in [19], after some 󵄨󵄨 (𝜕V /𝜕𝑢 ) + (𝜕V /𝜕V ) 𝜗 󵄨󵄨
󵄨󵄨 2𝑘𝜋 0 2𝑘𝜋 0 󵄨󵄨
simple calculations, we can obtain that lim𝑘 → +∞ |𝐸𝑀−1 | ≈ 󵄨󵄨 󵄨󵄨
󵄨󵄨󵄨 (𝜕𝑢2𝑘𝜋 /𝜕𝑢0 ) + (𝜕𝑢2𝑘𝜋 /𝜕V0 ) 𝜗 󵄨󵄨󵄨
𝑐0 /𝛿2 , lim𝑘 → +∞ |𝑀−1 | = 0, lim𝑘 → +∞ |Λ − 𝐸𝑀−1 𝐺| = 0,
lim𝑘 → +∞ |Λ𝑀−1 | ≈ 𝑐0 /𝛿4 , lim𝑘 → +∞ |𝐺𝑀−1 | ≈ 𝑐0 /𝛿2 , and 𝑐
= 2
lim𝑘 → +∞ |𝐸𝑀−1 | ≈ 𝑐0 /𝛿2 . Further, when 𝑐0 → 0, we can 𝑢2𝑘𝜋
obtain that 𝑐0 /𝛿2 → 0, 𝑐0 /𝛿4 → 0 and (𝜕𝐹𝑉/𝜕𝑢)(𝑐0 /𝛿2 ) → 󵄨󵄨
× 󵄨󵄨󵄨󵄨 (Δ (𝑢0 𝜗 − V0 ) + (1 + 6𝑘3 (𝑢2𝑘𝜋 + V2𝑘𝜋 ) 𝑘𝜋)
3
0.
󵄨
Thus, there exists a 𝛿 > 0 such that, for sufficiently large 𝑘,
inequalities (16) and (17) in Lemma 6 hold. Thus, according to × (V0 + 𝑢0 𝜗))
Lemma 6, we obtain that when 𝑘 is large enough, 𝑓|𝐷𝑘 satisfies
3
the (𝜇ℎ , 𝜇V )-cone condition. × (Δ (𝑢0 𝜗 − V0 ) − (1 − 6𝑘3 (𝑢2𝑘𝜋 + V2𝑘𝜋 ) 𝑘𝜋)
−1 󵄨󵄨
In fact, we can further prove the following. × (V0 + 𝑢0 𝜗)) 󵄨󵄨󵄨󵄨
󵄨
6 Abstract and Applied Analysis

󵄨󵄨
𝛿2 󵄨󵄨 2 3
Third, we want to prove 𝐷𝑃−𝑘 (S𝑠V ) ⊂ S𝑠H and |𝜉𝑃−𝑘 (𝑧0 ) | ≥
≈ 󵄨󵄨 (Δ (𝛿𝜗 − 𝑐 ) + (1 + 6𝑘 ( 𝛿 + 𝑐 ) 𝑘𝜋)
󵄨󵄨 3 𝜆 |𝜉𝑧0 |. For this, let 𝑇𝜑−1 (𝑢𝜑 , V𝜑 ) = (𝑢0 , V0 ) be the inverse map
−1
𝑐 󵄨󵄨 𝛿 𝛿
󵄨 of 𝑇𝜑 . Then,
𝑐
× (𝛿𝜗 + )) 𝜕𝑢𝜑 𝜕𝑢𝜑 −1
𝛿
−1 [ 𝜕𝑢 𝜕V ]
2 3 𝐷𝑇𝜑−1 = (𝐷𝑇𝜑 ) =[ 0 0]
[ 𝜕V𝜑 𝜕V𝜑 ]
𝑐 𝛿 +𝑐
× (Δ (𝛿𝜗 − ) − (1 − 6𝑘3 ( ) 𝑘𝜋)
𝛿 𝛿 [ 𝜕𝑢0 𝜕V0 ]
(21)
−1 󵄨󵄨 𝜕V𝜑 𝜕𝑢𝜑
𝑐 󵄨󵄨 −
󵄨 1 [ 𝜕V0 ]
× (𝛿𝜗 + )) 󵄨󵄨󵄨 = [ 𝜕V0 ]
𝛿 󵄨󵄨
󵄨󵄨 △ [ 𝜕V𝜑 𝜕𝑢𝜑 ] .
−
[ 𝜕𝑢0 𝜕𝑢0 ]
𝛿2 󵄨󵄨󵄨󵄨 1 󵄨󵄨
󵄨󵄨
≈ 󵄨󵄨1 + 󵄨.
𝑐 󵄨󵄨 3𝑘3 𝛿3 𝑘𝜋 󵄨󵄨󵄨 For any 𝑧0 = (𝑢0 , V0 ) ∈ V with (𝜉𝑧0 , 𝜂𝑧0 ) = (𝜛, 1) ∈ S𝑠𝑧0 ,
(19) (𝜉𝑃−𝑘 (𝑧0 ) , 𝜂𝑃−𝑘 (𝑧0 ) ) = 𝐷𝑃−𝑘 (𝜉𝑧0 , 𝜂𝑧0 ) = 𝐷𝑇2𝑘𝜋
−1
(𝜉𝑧0 , 𝜂𝑧0 ) =
(1/△)((𝜕V2𝑘𝜋 /𝜕V0 )𝜛 − (𝜕𝑢2𝑘𝜋 /𝜕V0 ), (−𝜕V2𝑘𝜋 /𝜕𝑢0 )𝜛 +
Clearly, lim𝑘 → +∞ (𝛿2 /𝑐)|1 + 1/3𝑘3 𝛿3 𝑘𝜋| = 𝛿2 /𝑐. More- (𝜕𝑢2𝑘𝜋 /𝜕𝑢0 )). Similarly, we can prove that |(𝜕V2𝑘𝜋 /𝜕V0 )𝜛 −
over, when 𝑐 → 0, 𝛿2 /𝑐 → +∞. So, for sufficiently (𝜕𝑢2𝑘𝜋 /𝜕V0 )/(−𝜕V2𝑘𝜋 /𝜕𝑢0 )𝜛 + (𝜕𝑢2𝑘𝜋 /𝜕𝑢0 )| > 𝜇ℎ
large 𝑘 and sufficiently small 𝛿1 and 𝛿2 , |((𝜕V2𝑘𝜋 /𝜕𝑢0 ) + and |(𝜕V2𝑘𝜋 /𝜕V0 )𝜛 − (𝜕𝑢2𝑘𝜋 /𝜕V0 )| > 𝜆−1 |𝜛|. Thus,
(𝜕V2𝑘𝜋 /𝜕V0 )𝜗)/((𝜕𝑢2𝑘𝜋 /𝜕𝑢0 ) + (𝜕𝑢2𝑘𝜋 /𝜕V0 )𝜗)| > 1/𝜇V . 𝐷𝑃−𝑘 (S𝑠V ) ⊂ S𝑠H and |𝜉𝑃−𝑘 (𝑧0 ) | ≥ 𝜆−1 |𝜉𝑧0 |.
Thus, 𝐷𝑃𝑘 (𝑧0 )(𝜉𝑧0 , 𝜂𝑧0 ) ∈ S𝑢V . This directly implies that Based on all above analysis and Remark 4, we can then
𝐷𝑃𝑘 (S𝑢H ) ⊂ S𝑢V . obtain that when 𝑘 is sufficiently large, 𝑃𝑘 satisfies the
Second, following the proof of 𝐷𝑃𝑘 (S𝑢H ) ⊂ S𝑢V , we want (𝜇ℎ , 𝜇V )-cone condition.
to prove that there exists a constant 𝜆 satisfying 0 < 𝜆 < 1 −
Based on Proposition 8, we can prove that 𝑓 satisfies the
𝜇ℎ 𝜇V such that |𝜂𝑃𝑘 (𝑧0 ) | ≥ 𝜆−1 |𝜂𝑧0 |.
boundary condition.
Similarly, for the above (𝑢0 , V0 ) and (𝑢2𝑘𝜋 , V2𝑘𝜋 ), when 𝑘 is
sufficiently large, Proposition 9. When 𝑖 and 𝑗 are sufficiently large,
󵄨󵄨 𝜕V 󵄨
󵄨󵄨 2𝑘𝜋 𝜕V2𝑘𝜋 󵄨󵄨󵄨 𝑓(𝜕ℎ 𝐷𝑖 ) ⋂ 𝐷𝑗 = 0 and 𝑓𝐷𝑖 ⋂ 𝜕V 𝐷𝑗 = 0, where 𝜕ℎ 𝐷𝑖 is
󵄨󵄨 + 𝜗󵄨󵄨 the horizontal boundary of 𝐷𝑖 and 𝜕V 𝐷𝑗 is the vertical
󵄨󵄨 𝜕𝑢0 𝜕V0 󵄨󵄨
boundary of 𝐷𝑗 .
1 󵄨󵄨 3
= 󵄨Δ (𝑢0 𝜗 − V0 ) + (1 + 6𝑘3 (𝑢2𝑘𝜋 + V2𝑘𝜋 ) 𝑘𝜋)
2𝑢2𝑘𝜋 󵄨 Proof. Due to Proposition 8, when 𝑘 is sufficiently large, 𝑃𝑘
satisfies the (𝜇ℎ , 𝜇V )-cone condition. This implies that 𝑃𝑘
󵄨
× (V0 + 𝑢0 𝜗)󵄨󵄨󵄨 contracts in the horizontal direction and expands in the
vertical direction. Moreover, 𝜇V -vertical curves are mapped to
󵄨 𝜇V -vertical curves under the map 𝑃𝑘 and 𝜇ℎ -horizontal curves
1 󵄨󵄨󵄨󵄨 2
≈ 󵄨Δ (𝛿 𝜗 − 𝑐) are mapped to 𝜇ℎ -horizontal curves under the map 𝑃−𝑘 . Thus,
2𝑐 󵄨󵄨󵄨󵄨
for sufficiently large 𝑖 and 𝑗, 𝐷𝑗 ⊂ 𝐵, 𝑃𝑖 𝐷𝑖 ⊂ 𝐵, ̃ 𝜕V 𝐷𝑗 ⊂ 𝜕V 𝐵
𝛿2 + 𝑐
3 𝑖 ̃
and 𝑃 (𝜕ℎ 𝐷𝑖 ) ⊂ 𝜕ℎ 𝐵. In addition, 𝑓(𝜕ℎ 𝐷𝑖 ) ⋂ 𝐷𝑗 = (𝐹 ∘
+ (1 + 6𝑘3 ( ) 𝑘𝜋)
𝛿 𝑃𝑖 (𝜕ℎ 𝐷𝑖 )) ⋂ 𝐷𝑗 = (𝐹 ∘ (𝑃𝑖 (𝜕ℎ 𝐷𝑖 ))) ⋂ 𝐷𝑗 . Thus, according to
the expression 𝐹(𝜕ℎ 𝐵)̃ ⋂ 𝐵 = 0 in Section 2.1, 𝑓(𝜕ℎ 𝐷𝑖 ) ⋂ 𝐷𝑗 =
󵄨󵄨
󵄨󵄨 0. Similarly, since 𝐹𝐵̃ ⋂ 𝜕V 𝐵 = 0, we can obtain that
× (𝑐 + 𝛿2 ) 󵄨󵄨󵄨 𝑓𝐷𝑖 ⋂ 𝜕V 𝐷𝑗 = 0.
󵄨󵄨
󵄨
Finally, we can prove that 𝑓 satisfies the intersection
𝛿2 󵄨󵄨 󵄨 condition as follows.
≈ 󵄨󵄨(1 + 3𝑘3 𝛿3 𝑘𝜋) 𝜗󵄨󵄨󵄨 .
𝑐 󵄨 󵄨
(20) Proposition 10. When 𝑖 and 𝑗 are sufficiently large,
𝑓𝐷𝑖 ⋂ 𝐷𝑗 ≠0.
For given 𝐵 and 𝐵, ̃ lim𝑘 → +∞ (𝛿2 /𝑐)|1 + 3𝑘3 𝛿3 𝑘𝜋| = ∞.
So, for any constant 𝜆 satisfying 0 < 𝜆 < 1 − 𝜇ℎ 𝜇V , when 𝑘 Proof. Let 𝐶V𝑖 (𝑢) = {𝑢} × (𝐷𝑖 )V be the family of vertical curves
is sufficiently large, (𝛿2 /𝑐)|1 + 3𝑘3 𝛿3 𝑘𝜋| > 𝜆−1 . Thus, we have in 𝐷𝑖 , where 𝑢 ∈ 𝐵ℎ and (𝐷𝑖 )V = {V | (𝑢, V) ∈ 𝐵, 𝑃𝑖 (𝑢, V) ∈
|(𝜕V2𝑘𝜋 /𝜕𝑢0 )+(𝜕V2𝑘𝜋 /𝜕V0 )𝜗| > 𝜆−1 |𝜗|, implying that |𝜂𝑃𝑘 (𝑧0 ) | ≥ ̃ From Proposition 8, 𝑃𝑖 with sufficiently large 𝑖 satisfies the
𝐵}.
𝜆−1 |𝜂𝑧0 |. (𝜇ℎ , 𝜇V )-cone condition. Thus, 𝑃𝑖 𝐶V𝑖 (𝑢) infinitely approaches
Abstract and Applied Analysis 7

𝐵̃V when 𝑖 → +∞. Similarly, letting 𝐶ℎ𝑖 (V) = 𝐵ℎ × {V} be the Acknowledgments
family of horizontal curves in 𝐷𝑖 , where V ∈ (𝐷𝑖 )V , we can
𝑗
obtain that 𝐶ℎ (V) infinitely approaches 𝐵ℎ when 𝑗 → +∞. This work was partly supported by NSFC-11422111, NSFC-
11290141, NSFC-11371047, and SKLSDE-2013ZX-10.
Since 𝐹 is 𝐶 , 𝐹(𝑃 𝐶V (𝑢)) infinitely approaches 𝐹(𝐵̃V )
1 𝑖 𝑖

when 𝑖 → +∞. By the expression 𝐹𝐵̃V ⋂ 𝐵ℎ = {𝑝} in

𝑗 References
Section 2.1, 𝐹(𝑃𝑖 𝐶V𝑖 (𝑢)) ⋂ 𝐶ℎ (V) ≠ 0. Thus, 𝑓𝐷𝑖 ⋂ 𝐷𝑗 ≠
0. [1] C. Chanu, L. Degiovanni, and G. Rastelli, “Superintegrable
three-body systems on the line,” Journal of Mathematical
Remark 11. In fact, we can prove that when 𝑖, 𝑗 → +∞, Physics, vol. 49, no. 11, Article ID 112901, 10 pages, 2008.
𝑗
𝑓𝐶V𝑖 (𝑢) and 𝐶ℎ (V) intersect at a unique point near 𝑝. [2] R. McGehee, “A stable manifold theorem for degenerate fixed
points with applications to celestial mechanics,” Journal of
Differential Equations, vol. 14, pp. 70–88, 1973.
2.3. Proof of Our Theorem 5. In order to prove our Theorem 5, [3] N. Haghighipour, “Resonance dynamics and partial averaging
similar to [19], we try to use Propositions 7, 9, and 10 to verify in a restricted three-body system,” Journal of Mathematical
that 𝑓 satisfies Assumptions 1 and 2. Then, from Lemma 3, we Physics, vol. 43, no. 7, pp. 3678–3694, 2002.
can obtain that 𝑓 is a horseshoe map as follows. [4] Z. Xia, “Melnikov method and transversal homoclinic point
in the restricted three-body problem,” Journal of Differential
Proof. From Proposition 7, when 𝑘 is sufficiently large, 𝑓|𝐷𝑘 Equations, vol. 96, pp. 170–184, 1992.
satisfies the (𝜇ℎ , 𝜇V )-cone condition. Thus, the map 𝑓 con- [5] P. T. Boyd and S. L. W. McMillan, “Chaotic scattering in the
tracts in the horizontal direction and expands in the vertical gravitational three-body problem,” Chaos, vol. 3, no. 4, pp. 507–
direction. Moreover, 𝜇V -vertical curves are mapped to 𝜇V - 524, 1993.
vertical curves under the map 𝑓 and 𝜇ℎ -horizontal curves are [6] H. Dankowicz and P. Holmes, “The existence of transverse
homoclinic points in the Sitnikov problem,” Journal of Differ-
mapped to 𝜇ℎ -horizontal curves under the map 𝑓−1 . There-
ential Equations, vol. 116, no. 2, pp. 468–483, 1995.
fore, from Propositions 9 and 10, for sufficiently large 𝑖 and
[7] Z. She, X. Cheng, and C. Li, “The existence of transversal
𝑗, 𝑓𝐷𝑖 ⋂ 𝐷𝑗 ≠ 0 is a (𝜇ℎ , 𝜇V )-curved rectangle and satisfies homoclinic orbits in a planar circular restricted four-body
𝜕ℎ (𝑓𝐷𝑖 ⋂ 𝐷𝑗 ) ⊂ 𝜕ℎ 𝐷𝑗 and 𝜕V (𝑓𝐷𝑖 ⋂ 𝐷𝑗 ) ⊂ 𝜕V (𝑓𝐷𝑖 ). Similarly, problem,” Celestial Mechanics and Dynamical Astronomy, vol.
for sufficiently large 𝑖 and 𝑗, 𝑓−1 (𝑓𝐷𝑖 ⋂ 𝐷𝑗 ) ≠ 0 is also a 115, no. 3, pp. 299–309, 2013.
(𝜇ℎ , 𝜇V )-curved rectangle and satisfies 𝜕V 𝑓−1 (𝑓𝐷𝑖 ⋂ 𝐷𝑗 ) ⊂ [8] J. Llibre and E. Perez-Chavela, “Transversal homoclinic orbits
in the collinear restricted three-body problem,” Journal of
𝜕V 𝐷𝑖 and 𝜕ℎ 𝑓−1 (𝑓𝐷𝑖 ⋂ 𝐷𝑗 ) ⊂ 𝜕ℎ (𝑓−1 𝐷𝑗 ). Nonlinear Science, vol. 15, no. 1, pp. 1–10, 2005.
Let 𝑁 ≥ 2 be an arbitrary but fixed positive integer. [9] M. Gidea and F. Deppe, “Chaotic orbits in a restricted three-
For sufficiently large 𝑘, 𝑖, and 𝑗, by letting 𝐻𝑙 = 𝐷𝑙+𝑘−1 , body problem: numerical experiments and heuristics,” Commu-
𝑉𝑙 = 𝑓(𝐷𝑙+𝑘−1 ), 𝑓𝐷𝑖 ⋂ 𝐷𝑗 = 𝑉(𝑗−𝑘+1)(𝑖−𝑘+1) , and 𝐷𝑖 ⋂ 𝑓−1 𝐷𝑗 = nications in Nonlinear Science and Numerical Simulation, vol. 11,
𝐻(𝑖−𝑘+1)(𝑗−𝑘+1) , where 1 ≤ 𝑙 ≤ 𝑁, we can obtain that 𝑓 satisfies no. 2, pp. 161–171, 2006.
Assumption 1. In addition, due to Remark 4, 𝑓 obviously [10] Z. Xia, “Arnold diffusion and oscillatory solutions in the planar
satisfies Assumption 2. three-body problem,” Journal of Differential Equations, vol. 110,
Thus, for an arbitrary but fixed 𝑁 ≥ 2, when 𝑘 is no. 2, pp. 289–321, 1994.
sufficiently large, the map 𝑓 over the set ⋃𝑁−1 𝑖=0 𝐷𝑘+𝑖 satisfies
[11] K. Zare and S. Chesley, “Order and chaos in the planar isosceles
Assumptions 1 and 2; that is, 𝑓 satisfies the Conley-Moser three-body problem,” Chaos, vol. 8, no. 2, pp. 475–494, 1998.
conditions. [12] R. Zhu and C. Xiang, “Studies of Melnikov method and
transversal homoclinic orbits in the circular planar restricted
By Lemma 3, when 𝑘 is sufficiently large, 𝑓 has an
three-body problem,” Applied Mathematics and Mechanics, vol.
invariant Cantor set, on which it is topologically conjugate 17, no. 12, pp. 1177–1187, 1996.
to a full shift on 𝑁 symbols. This directly implies that 𝑓 is a [13] D. Wilczak and P. Zgliczynski, “Heteroclinic connections
horseshoe map. between periodic orbits in planar restricted circular three-
body problem—a computer assisted proof,” Communications in
3. Conclusions Mathematical Physics, vol. 234, no. 1, pp. 37–75, 2003.
[14] D. Wilczak and P. Zgliczyński, “Heteroclinic connections
In this present note, we studied the existence of a Smale between periodic orbits in planar restricted circular three body
horseshoe in a planar circular restricted three-body problem problem. II,” Communications in Mathematical Physics, vol. 259,
by first defining an invertible map 𝑓 and then proving that no. 3, pp. 561–576, 2005.
this 𝑓 satisfies the Conley-Moser conditions. This implies that [15] J. Guckenheimer and P. Holmes, Nonlinear Osicillations,
the planar circular restricted three-body problem processes Dynamical Systems, and Bifurcations of Vector Fields, Springer,
New York, NY, USA, 1983.
chaotic dynamics of the Smale horseshoe type.
[16] S. Wiggins, Introduction to Applied Nonlinear Dynamical Sys-
tems and Chaos, vol. 2 of Texts in Applied Mathematics, Springer,
Conflict of Interests New York, NY, USA, 1990.
[17] Y. Ilyashenko and W. Li, Nonlocal Bifurcations, vol. 66 of
The authors declare that there is no conflict of interests Mathematical Surveys and Monographs, American Mathemat-
regarding the publication of this paper. ical Society, Providence, RI, USA, 1999.
8 Abstract and Applied Analysis

[18] Z. Zhang, C. Li, Z. Zheng, and W. Li, Bifurcations Theory of the

Vector Fields, Higher Education Press, Beijing, China, 1997.
[19] Z. She and X. Cheng, “The existence of a Smale horseshoe
in a planar circular restricted four-body problem,” Celestial
Mechanics & Dynamical Astronomy, vol. 118, no. 2, pp. 115–127,
2014.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2015, Article ID 826020, 8 pages
http://dx.doi.org/10.1155/2015/826020

Research Article
Effect of Leakage Delay on Stability of Neutral-Type Genetic
Regulatory Networks

Li Li,1,2 Yongqing Yang,1 and Chuanzhi Bai3

1
School of Science, Jiangnan University, Wuxi 214122, China
2
Key Laboratory of Advanced Process Control for Light Industry (Ministry of Education), Wuxi 214122, China
3
Department of Mathematics, Huaiyin Normal University, Huaian 223300, China

Correspondence should be addressed to Yongqing Yang; yongqingyang@163.com

Received 16 June 2014; Accepted 6 August 2014

Academic Editor: Yonghui Xia

Copyright © 2015 Li Li et al. This is an open access article distributed under the Creative Commons Attribution License, which
permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

The stability of neutral-type genetic regulatory networks with leakage delays is considered. Firstly, we describe the model of genetic
regulatory network with neutral delays and leakage delays. Then some sufficient conditions are derived to ensure the asymptotic
stability of the genetic regulatory network by the Lyapunov functional method. Further, the effect of leakage delay on stability is
discussed. Finally, a numerical example is given to show the effectiveness of the results.

1. Introduction and may cause destabilization, oscillation, bifurcation, and

chaos. So time delay should be taken into consideration when
Biology networks are important research tools of complex modelling the genetic networks. There are many types of
biological phenomena, which have been attracting attention time delays such as discrete delays [16], continuous delays
of scientists and engineers [1–4]. There are many kinds of bio- [17, 18], random delays [19], and mixed delays [20]. In recent
logical networks, such as metabolic network, signal network, years, the neutral delay has attracted attention because it can
genetic regulatory network (GRN), and protein interaction embody some important information about the derivation
network. GRN, which describes the complex interactions of the past state [21, 22]. In [23, 24], the authors considered
between genes (mRNA) and its products (proteins), is often asymptotic stability of BAM neural networks of neutral type.
modelled by a kind of dynamical system. The study of GRNs Lakshmanan and Balasubramaniam [25] obtained some new
can help scientists understand many important and complex results of robust stability analysis for neutral-type neural
phenomena of living cells. Up to now, several models of networks with time-varying delays and Markovian jumping
GRNs have been established, for example, Boolean network parameters. However, the stability analysis for genetic reg-
models [5], Bayesian network models [6], Petri network ulatory networks with neutral delay has been rarely investi-
models [7, 8], and the differential equation models. It is gated. In [26], Jung et al. discussed the stability for genetic
more convenient to analyze the dynamical behaviors by using regulatory network with neutral delay. Further, Jiao et al.
differential equation models and have been widely studied by [27] investigated the asymptotic stability of genetic regulatory
many experts [9–12]. network with neutral delay, which extended and improved
As we all know, stability is one of the most important the main results of Jung et al. [26]. On the other hand, leakage
issues of longtime quality behavior of dynamical systems. delay is another type of delay, which exists in the negative
Some important biological secrets have been revealed and feedback term of system [28, 29]. So far, few studies have
many significant results have been reported [13–15]. On been paid to the stability analysis for GRNs with leakage
the other hand, time delay is a common phenomenon that delay. In [30], the authors discussed the global asymptotic
describes the dynamical behaviors. The existence of time stability for genetic regulatory networks with leakage delay.
delay can make the dynamical behaviors more complicated In [31], stability was investigated for a class of uncertain
2 Abstract and Applied Analysis

impulsive stochastic genetic regulatory networks with time- In this paper, we will consider the GRNs with neutral
varying leakage delay. As far as we know, there is no result delays and leakage delays:
on the stability of genetic regulatory networks with neutral
delays and leakage delays. Motivated by the discussion above, 𝑥̇(𝑡) = − 𝐴𝑥 (𝑡 − 𝜌1 ) + 𝑊𝑓 (𝑦 (𝑡 − 𝜎 (𝑡))) ,
in this paper, we consider the stability of genetic regulatory
networks of neutral type with leakage delay. 𝑦̇(𝑡) = − 𝐶𝑦 (𝑡 − 𝜌2 ) + 𝐷1 𝑥 (𝑡 − 𝜏1 (𝑡)) + 𝐷2 𝑥̇(𝑡 − 𝜏2 (𝑡)) ,
The paper is organized as follows. In Section 2, the prob- (4)
lem is formulated and some preliminaries and assumption
are given. In Section 3, the asymptotic stability of neutral-
type genetic regulatory networks with leakage delay is inves- where 𝜌1 > 0 and 𝜌2 > 0 denote the leakage delays. 𝜏2 (𝑡) is
tigated. In Section 4, an illustrative example is given to show the neutral delay and satisfies 0 ≤ 𝜏2 (𝑡) ≤ 𝜏2 , 𝜏2̇(𝑡) ≤ 𝜏∗ .
the effectiveness of the theoretical results. Some conclusions The system (4) has the following equivalent form:
are proposed in Section 5.
𝑡
𝑑
𝑛 𝑛×𝑛 [𝑥 (𝑡) − 𝐴 ∫ 𝑥 (𝜇) 𝑑𝜇] = −𝐴𝑥 (𝑡) + 𝑊𝑓 (𝑦 (𝑡 − 𝜎 (𝑡))) ,
Notations. Throughout this paper, 𝑅 and 𝑅 denote the 𝑑𝑡 𝑡−𝜌1
𝑛-dimensional Euclidean space and the set of all 𝑛 × 𝑛
real matrices, respectively. 𝐴𝑇 denotes the transposition of 𝑑 𝑡

matrix 𝐴. Diag (⋅) denotes the diagonal matrix. Asterisk (∗) [𝑦 (𝑡) − 𝐶 ∫ 𝑦 (𝜇) 𝑑𝜇]
𝑑𝑡 𝑡−𝜌2
represents the symmetric block of the symmetric matrix.
= −𝐶𝑦 (𝑡) + 𝐷1 𝑥 (𝑡 − 𝜏1 (𝑡)) + 𝐷2 𝑥̇(𝑡 − 𝜏2 (𝑡)) .
2. Preliminaries (5)
GRNs with 𝑛 mRNA and 𝑛 proteins can be modelled by the
The initial conditions are given as follows:
following differential equations:
𝑛
𝑚̇𝑖 (𝑡) = − 𝑎𝑖 𝑚𝑖 (𝑡) + ∑𝜔𝑖𝑗 𝑔𝑗 (𝑝𝑗 (𝑡 − 𝜎 (𝑡))) + 𝐼𝑖 , 𝑥𝑖 (𝑠) = 𝜙𝑖 (𝑠) , 𝑠 ∈ (−𝜏, 0) ,
𝑗=1
(6)
(1)
𝑦𝑖 (𝑠) = 𝜓𝑖 (𝑠) , 𝑠 ∈ (−𝜎, 0) ,
𝑝𝑖̇(𝑡) = − 𝑐𝑖 𝑝𝑖 (𝑡) + 𝑑𝑖 𝑚𝑖 (𝑡 − 𝜏1 (𝑡)) ,
where 𝑚𝑖 (𝑡) and 𝑝𝑖 (𝑡) denote the concentrations of mRNA where 𝜏 = max{𝜌1 , 𝜎} and 𝜎 = max{𝜌2 , 𝜏1 , 𝜏2 }.
and protein of the 𝑖th node at time 𝑡, respectively. 𝑎𝑖 and 𝑐𝑖 are In this paper, we introduce the following assumption and
the degradation rates of the mRNA and protein, respectively. lemmas.
𝑑𝑖 is the translation rate, 𝜎(𝑡) is the feedback regulation delays,
Assumption 1. There exist constants 𝑘𝑖− , 𝑘𝑖+ such that the regu-
and 𝜏(𝑡) is the translation delays satisfying 0 ≤ 𝜎(𝑡) ≤
latory function 𝑓𝑖 (⋅) satisfies
𝜎, 0 ≤ 𝜏1 (𝑡) ≤ 𝜏1 , where 𝜎 and 𝜏 are constants. Consider
𝑔𝑗 (𝑝𝑗 (𝑠)) = (𝑝𝑗 (𝑠)/𝛽𝑗 )𝐻𝑗 /(1 + 𝑝𝑗 (𝑠)/𝛽𝑗 )𝐻𝑗 , where 𝐻𝑗 is the
Hill coefficient. 𝛽𝑗 is a positive constant which denotes the 𝑓𝑖 (𝑥) − 𝑓𝑖 (𝑦)
𝑘𝑖− ≤ ≤ 𝑘𝑖+ , (7)
feedback regulation of the protein on the transcription. The 𝑥−𝑦
coupling matrix 𝑊 = (𝜔𝑖𝑗 )𝑛×𝑛 ∈ 𝑅𝑛×𝑛 of the GRNs is defined
as follows: for all 𝑥, 𝑦 ∈ R, 𝑥 ≠ 𝑦 and 𝑖 = 1, 2, . . . , 𝑛.
𝜔𝑖𝑗
Remark 2. In the above assumption, 𝑘𝑖− , 𝑘𝑖+ are some real con-
stants, which may be positive, zero, or negative. Compared
{ 𝛼 , if transcription factor 𝑗 is an activator of gene 𝑖,
{ 𝑖𝑗 with the existing results in [32], this is less conservative and
= {0, if there is no link from 𝑗 to 𝑖, less restrictive.
{
{−𝛼𝑖𝑗 , if transcription factor 𝑗 is a repressor of gene 𝑖.
(2) The following lemmas will be used to derive our main
results.
𝑇
Let (𝑚∗ , 𝑝∗ ) be an equilibrium point of (1). The equi-
librium point can be shifted to the origin by transformation: Lemma 3 (see [33]). Given any real matrix of appropriate
𝑥𝑖 (𝑡) = 𝑚𝑖 (𝑡) − 𝑚𝑖∗ , 𝑦𝑖 (𝑡) = 𝑝𝑖 (𝑡) − 𝑝𝑖∗ . System (1) can be dimension 𝑀 = 𝑀𝑇 > 0 and a vector function 𝜔(⋅) : [𝑎, 𝑏] →
changed into the following compact matrix form: R𝑛 , such that the integrations concerned are well defined, then
𝑥̇(𝑡) = − 𝐴𝑥 (𝑡) + 𝑊𝑓 (𝑦 (𝑡 − 𝜎 (𝑡))) ,
𝑏 𝑇 𝑏
(3)
𝑦̇(𝑡) = − 𝐶𝑦 (𝑡) + 𝐷1 𝑥 (𝑡 − 𝜏1 (𝑡)) , [∫ 𝜔 (𝑠) 𝑑𝑠] 𝑀 [∫ 𝜔 (𝑠) 𝑑𝑠]
𝑎 𝑎
where 𝐴 = diag[𝑎1 , 𝑎2 , . . . , 𝑎𝑛 ], 𝐶 = diag[𝑐1 , 𝑐2 , . . . , 𝑐𝑛 ], 𝐷1 = (8)
diag[𝑑1 , 𝑑2 , . . . , 𝑑𝑛 ], and 𝑓(𝑦(𝑡)) = 𝑔(𝑦(𝑡) + 𝑝∗ ) − 𝑔(𝑝∗ ) with 𝑏

𝑓(0) = 0. ≤ (𝑏 − 𝑎) ∫ 𝜔𝑇 (𝑠) 𝑀𝜔 (𝑠) 𝑑𝑠.

𝑎
Abstract and Applied Analysis 3

Lemma 4 (see [34, 35]). For any diagonal matrices Λ > 0, if Π5,9 = 𝐷2𝑇 𝑄6 𝐶, Π5,13 = −𝐷2𝑇 𝑄𝐶,
Assumption 1 holds, then
Π6,6 = −𝑃2 , Π6,12 = −𝐴𝑃𝑊,
𝑇
𝑦 (𝑡) −Λ𝐿 Λ𝐾 𝑦 (𝑡)
[ ] [ ][ ] ≥ 0, (9) Π7,7 = −𝑄𝐶 − 𝐶𝑄 + 𝑄1 + 𝜌22 𝑄2 − Λ 1 𝐿,
𝑓 (𝑦 (𝑡)) ∗ −Λ 𝑓 (𝑦 (𝑡))
𝑇 𝑇
𝑇 Π7,10 = 𝑆13 , Π7,11 = Λ 1 𝐾, Π7,12 = 𝑆23 ,
𝑦 (𝑡 − 𝜎 (𝑡)) −Λ𝐿 Λ𝐾 𝑦 (𝑡 − 𝜎 (𝑡))
[ ] [ ][ ] ≥ 0,
𝑓 (𝑦 (𝑡 − 𝜎 (𝑡))) ∗ −Λ 𝑓 (𝑦 (𝑡 − 𝜎 (𝑡))) Π7,13 = 𝐶𝑄𝐶, Π8,8 = 𝜎𝑆33 − 𝑄5 − 𝑄5𝑇 ,
(10)
Π8,9 = −𝑄5 𝐶, Π9,9 = −𝑄1 ,
where 𝐾 = diag(𝑘1− 𝑘1+ , . . . , 𝑘𝑛− 𝑘𝑛+ ) and 𝐿 = diag((𝑘1− +
𝑘1+ )/2, . . . , (𝑘𝑛− + 𝑘𝑛+ )/2). Π10,10 = 𝜎𝑆11 − 𝑆13 − Λ 2 𝐿,
𝑇
Π10,12 = 𝜎𝑆12 − 𝑆23 + Λ 2 𝐾,
3. Main Results
Π11,11 = −Λ 1 , Π12,12 = 𝜎𝑆22 − Λ 2 ,
In this section, the asymptotic stability of system (5) is inves-
tigated. Some sufficient conditions are obtained to ensure Π13,13 = −𝑄2 , Π14,14 = −𝑅1 ,
the stability of system (5) based on the Lyapunov functional
approach. 𝑅3
Π15,15 = −𝑅2 , Π16,16 = − .
𝜏2
Theorem 5. Let Assumption 1 hold. Then the trivial solution (15)
of system (5) is asymptotically stable if there exist symmetric
positive definite matrices 𝑃, 𝑄, symmetric matrices 𝑃𝑖 , 𝑄𝑖 (𝑖 = Proof. Consider the following Lyapunov-Krasovskii func-
1, 2), diagonal positive definite matrix 𝑄3 , and positive definite tional:
matrices 𝑄4 , 𝑄5 , 𝑄6 such that the following LMIs hold:
𝑉 (𝑡) = 𝑉1 (𝑡) + 𝑉2 (𝑡) + 𝑉3 (𝑡) + 𝑉4 (𝑡) + 𝑉5 (𝑡) , (16)
∏ = (Π𝑖,𝑗 )16×16 < 0, (11)
where
𝑡 𝑇
𝑆11 𝑆12 𝑆13
[ ∗ 𝑆22 𝑆23 ] > 0, (12) 𝑉1 (𝑡) = [𝑥 (𝑡)− 𝐴 ∫ 𝑥 (𝜇) 𝑑𝜇]
𝑡−𝜌1
[ ∗ ∗ 𝑆33 ] 𝑡
× 𝑃 [𝑥 (𝑡)− 𝐴 ∫ 𝑥 (𝜇) 𝑑𝜇]
𝑇11 𝑇12 𝑡−𝜌1
[ ] > 0, (13)
∗ 𝑆22
𝑡 𝑇

𝑅11 𝑅12 + [𝑦 (𝑡)− 𝐶 ∫ 𝑦 (𝜇) 𝑑𝜇]

[ ] > 0, (14) 𝑡−𝜌2
∗ 𝑅22
𝑡
where × 𝑄 [𝑦 (𝑡)− 𝐶 ∫ 𝑦 (𝜇) 𝑑𝜇] ,
𝑡−𝜌2
Π1,1 = −𝑃𝐴 − 𝐴𝑃 + 𝑃1 + 𝜌12 𝑃2 + 𝑅1 + 𝜏2 𝑅3 , 𝑡
𝑇 𝑉2 (𝑡) = ∫ 𝑥𝑇 (𝑠) 𝑃1 𝑥 (𝑠) 𝑑𝑠
Π1,4 = 𝑇12 , Π1,6 = 𝐴𝑃𝐴, 𝑡−𝜌1

Π1,12 = 𝑃𝑊, Π1,14 = −𝑅12 , 𝑡 𝑡

+𝜌1 ∫ ∫ 𝑥𝑇 (𝜇) 𝑃2 𝑥 (𝜇) 𝑑𝜇 𝑑𝑠
𝑡−𝜌1 𝑠
Π2,2 = 𝜏1 𝑇22 + 𝑄3 − 𝑄4 − 𝑄4𝑇 + 𝑅1 + 𝜏2 𝑅22 ,
𝑡
Π2,3 = −𝑄4 𝐴, 𝑇
Π2,5 = 𝑅12 , Π2,12 = 𝑄4 𝑊, +∫ 𝑦𝑇 (𝑠) 𝑄1 𝑦 (𝑠) 𝑑𝑠
𝑡−𝜌2
𝑇
Π3,3 = −𝑃1 , Π4,4 = −𝑇12 − 𝑇12 + 𝜏1 𝑇11 , 𝑡 𝑡
+𝜌2 ∫ ∫ 𝑦𝑇 (𝜇) 𝑄2 𝑦 (𝜇) 𝑑𝜇 𝑑𝑠,
𝑡−𝜌2 𝑠
Π4,5 = −𝐷1𝑇 𝑄6𝑇 𝐷2 , Π4,7 = 𝐷1 𝑄𝑇 ,
𝑇
𝑦 (𝜇 − 𝜎 (𝜇))
Π4,8 = 𝐷1𝑇 𝑄5𝑇 , Π4,13 = −𝐷1𝑇 𝑄𝐶, 𝑡
[
𝜇
]
𝑉3 (𝑡) = ∫ ∫ [ (𝑦 (𝜇 − 𝜎 (𝜇)))]
𝑓
0 𝜇−𝜎(𝜇)
Π5,5 = (𝜏∗ − 1) 𝑄3 − 𝐷2𝑇 𝑄6 𝐷2 − 𝐷2𝑇 𝑄6𝑇 𝐷2 [ 𝑦̇(𝑠) ]
𝑇
+ 𝜏2 𝑅11 − 𝑅12 − 𝑅12 , 𝑆11 𝑆12 𝑆13
× [ ∗ 𝑆22 𝑆23 ]
Π5,7 = 𝐷2𝑇 𝑄𝑇 , Π5,8 = 𝐷2𝑇 𝑄5𝑇 + 𝐷2𝑇 𝑄6 , [ ∗ ∗ 𝑆33 ]
4 Abstract and Applied Analysis

𝑡
𝑦 (𝜇 − 𝜎 (𝜇))
+ 2𝑦𝑇 (𝑡) 𝐶𝑄𝐶 (∫ 𝑦 (𝜇) 𝑑𝜇)
× [𝑓 (𝑦 (𝜇 − 𝜎 (𝜇)))] 𝑑𝑠 𝑑𝜇 𝑡−𝜌2
[ 𝑦̇(𝑠) ]
𝑡 𝑇
𝑡 𝜇 𝑇
𝑥 (𝜇 − 𝜏1 (𝜇)) 𝑇 𝑇 − 2(∫ 𝑦 (𝜇) 𝑑𝜇) 𝐶𝑄𝐷1 𝑥 (𝑡 − 𝜏1 (𝑡))
+∫ ∫ [ ] [ 11 12 ] 𝑡−𝜌2
0 𝜇−𝜏1 (𝜇) 𝑥̇(𝑠) ∗ 𝑆22
𝑡
𝑥 (𝜇 − 𝜏1 (𝜇)) − 2𝑥̇(𝑡 − 𝜏2 (𝑡)) 𝐷2𝑇 𝑄𝐶 (∫ 𝑦 (𝜇) 𝑑𝜇) ,
×[ ] 𝑑𝑠 𝑑𝜇 𝑡−𝜌2
𝑥̇(𝑠)
𝑡 𝜇
𝑥̇(𝜇 − 𝜏2 (𝜇)) 𝑅 𝑅
𝑇 𝑉2̇(𝑡) = 𝑥𝑇 (𝑡) 𝑃1 𝑥 (𝑡) − 𝑥𝑇 (𝑡 − 𝜌1 ) 𝑃1 𝑥 (𝑡 − 𝜌1 )
+∫ ∫ [ ] [ 11 12 ]
0 𝜇−𝜏2 (𝜇) 𝑥̇(𝑠) ∗ 𝑅22 𝑡
− 𝜌1 ∫ 𝑥𝑇 (𝜇) 𝑃2 𝑥 (𝜇) 𝑑𝜇 + 𝜌12 𝑥𝑇 (𝑡) 𝑃2 𝑥 (𝑡)
𝑥̇(𝜇 − 𝜏2 (𝜇)) 𝑡−𝜌1
×[ ] 𝑑𝑠 𝑑𝜇,
𝑥̇(𝑠)
+ 𝑦𝑇 (𝑡) 𝑄1 𝑦 (𝑡) − 𝑦𝑇 (𝑡 − 𝜌2 ) 𝑄1 𝑦 (𝑡 − 𝜌2 )
0 𝑡
𝑉4 (𝑡) = ∫ ∫ 𝑦𝑇̇ (𝑠) 𝑆33 𝑦̇(𝑠) 𝑑𝑠 𝑑𝜇 𝑡
−𝜎 𝑡+𝜇 − 𝜌2 ∫ 𝑦𝑇 (𝜇) 𝑄2 𝑦 (𝜇) 𝑑𝜇 + 𝜌22 𝑦𝑇 (𝑡) 𝑄2 𝑦 (𝑡)
𝑡−𝜌2
0 𝑡
𝑇
+∫ ∫ 𝑥̇ (𝑠) 𝑇22 𝑥̇(𝑠) 𝑑𝑠 𝑑𝜇, ≤ 𝑥𝑇 (𝑡) 𝑃1 𝑥 (𝑡) − 𝑥𝑇 (𝑡 − 𝜌1 ) 𝑃1 𝑥 (𝑡 − 𝜌1 )
−𝜏1 𝑡+𝜇
𝑡 𝑇 𝑡
𝑡
𝑉5 (𝑡) = ∫ 𝑇
[𝑥 (𝑠) 𝑅1 𝑥 (𝑠) + 𝑥̇ (𝑠) 𝑅2 𝑥̇(𝑠)] 𝑑𝑠 𝑇 − [∫ 𝑥 (𝜇) 𝑑𝜇] 𝑃2 [∫ 𝑥 (𝜇) 𝑑𝜇]
𝑡−𝜏2 𝑡−𝜌1 𝑡−𝜌1

0 𝑡
𝑇
+ 𝜌12 𝑥𝑇 (𝑡) 𝑃2 𝑥 (𝑡) + 𝑦𝑇 (𝑡) 𝑄1 𝑦 (𝑡)
+∫ ∫ [𝑥 (𝑠) 𝑅3 𝑥 (𝑠)
−𝜏2 𝑡+𝜇
− 𝑦𝑇 (𝑡 − 𝜌2 ) 𝑄1 𝑦 (𝑡 − 𝜌2 )
𝑇
+𝑥̇ (𝑠) 𝑅22 𝑥̇(𝑠)] 𝑑𝑠 𝑑𝜇 𝑡 𝑇 𝑡
𝑡 − [∫ 𝑦 (𝜇) 𝑑𝜇] 𝑄2 [∫ 𝑦 (𝜇) 𝑑𝜇]
𝑇 𝑡−𝜌2 𝑡−𝜌2
+∫ 𝑥̇ (𝑠) 𝑄3 𝑥̇(𝑠) 𝑑𝑠.
𝑡−𝜏2 (𝑡)
+ 𝜌22 𝑦𝑇 (𝑡) 𝑄2 𝑦 (𝑡) ,
(17)
𝑇
𝑡 𝑦 (𝑡 − 𝜎 (𝑡)) 𝑆11 𝑆12 𝑆13
𝑉3̇(𝑡) = ∫ [𝑓 (𝑦 (𝑡 − 𝜎 (𝑡)))] [ ∗ 𝑆22 𝑆23 ]
Calculating the derivative of 𝑉(𝑡) along with the solution 𝜇−𝜎(𝑡)
of the system (5), we have [ 𝑦̇(𝑠) ] [ ∗ ∗ 𝑆33 ]
𝑦 (𝑡 − 𝜎 (𝑡))
𝑡 𝑇 × [𝑓 (𝑦 (𝑡 − 𝜎 (𝑡)))] 𝑑𝑠
𝑉1̇(𝑡) = 2[𝑥 (𝑡) − 𝐴 ∫ 𝑥 (𝜇) 𝑑𝜇] [ 𝑦̇(𝑠) ]
𝑡−𝜌1
𝑡 𝑇
× 𝑃 [−𝐴𝑥 (𝑡) + 𝑊𝑓 (𝑦 (𝑡 − 𝜎 (𝑡)))] 𝑥 (𝑡 − 𝜏1 (𝑡)) 𝑇 𝑇
+∫ [ ] [ 11 12 ]
𝑡−𝜏1 (𝑡) 𝑥̇(𝑠) ∗ 𝑆22
𝑡 𝑇
+ 2[𝑦 (𝑡) − 𝐶 ∫ 𝑦 (𝜇) 𝑑𝜇] 𝑥 (𝑡 − 𝜏1 (𝑡))
𝑡−𝜌2 ×[ ] 𝑑𝑠
𝑥̇(𝑠)
× [−𝐶𝑦 (𝑡) + 𝐷1 𝑥 (𝑡 − 𝜏1 (𝑡)) + 𝐷2 𝑥̇(𝑡 − 𝜏2 (𝑡))]
𝑡 𝑇
𝑥̇(𝑡 − 𝜏2 (𝑡)) 𝑅 𝑅
𝑇
= −2𝑥 (𝑡) 𝑃𝐴𝑥 (𝑡) + 2𝑥 (𝑡) 𝑃𝑊𝑓 (𝑦 (𝑡 − 𝜎 (𝑡))) 𝑇 +∫ [ ] [ 11 12 ]
𝑡−𝜏2 (𝑡) 𝑥̇(𝑠) ∗ 𝑅22
𝑡
+ 2𝑥𝑇 (𝑡) 𝐴𝑃𝐴 (∫ 𝑥 (𝜇) 𝑑𝜇) 𝑥̇(𝑡 − 𝜏2 (𝑡))
𝑡−𝜌1 ×[ ] 𝑑𝑠
𝑥̇(𝑠)
𝑡 𝑇
− 2(∫ 𝑥 (𝜇) 𝑑𝜇) 𝐴𝑃𝑊𝑓 (𝑦 (𝑡 − 𝜎 (𝑡))) ≤ 𝜏1 𝑥𝑇 (𝑡 − 𝜏1 (𝑡)) 𝑇11 𝑥 (𝑡 − 𝜏1 (𝑡))
𝑡−𝜌1

− 2𝑦𝑇 (𝑡) 𝑄𝐶𝑦 (𝑡) + 2𝑥𝑇 (𝑡 − 𝜏1 (𝑡)) 𝐷1 𝑄𝑇 𝑦 (𝑡) + 2𝑥𝑇 (𝑡) 𝑇12
𝑇
𝑥 (𝑡 − 𝜏1 (𝑡))

+ 2𝑥̇(𝑡 − 𝜏2 (𝑡)) 𝐷2𝑇 𝑄𝑇 𝑦 (𝑡) − 2𝑥𝑇 (𝑡 − 𝜏1 (𝑡)) 𝑇12

𝑇
𝑥 (𝑡 − 𝜏1 (𝑡))
Abstract and Applied Analysis 5

𝑡
+∫ 𝑇
𝑥̇ (𝑠) 𝑇22 𝑥̇(𝑠) 𝑑𝑠 + 𝜏2 𝑥𝑇 (𝑡) 𝑅3 𝑥 (𝑡) + 𝜏2 𝑥̇
𝑇
(𝑡) 𝑅22
𝑡−𝜏1
𝑡 𝑇
𝑡
1
𝑇
+ 𝜏2 𝑥̇ (𝑡 − 𝜏2 (𝑡)) 𝑅11 𝑥̇(𝑡 − 𝜏2 (𝑡)) − (∫ 𝑥 (𝑠) 𝑑𝑠) 𝑅3 (∫ 𝑥 (𝑠) 𝑑𝑠)
𝜏2 𝑡−𝜏2 𝑡−𝜏2
𝑇 𝑇
+ 2𝑥̇ (𝑡) 𝑅12 𝑥̇(𝑡 − 𝜏2 (𝑡)) 𝑡
𝑇 𝑇
−∫ 𝑥̇ (𝑠) 𝑅22 𝑥̇(𝑠) 𝑑𝑠 + 𝑥̇ (𝑡) 𝑄3 𝑥̇(𝑡)
𝑇 𝑇 𝑡−𝜏2
− 2𝑥̇ (𝑡 − 𝜏2 (𝑡)) 𝑅12 𝑥̇(𝑡 − 𝜏2 (𝑡))
𝑇
𝑡 + 𝑥̇ (𝑡 − 𝜏2 (𝑡)) 𝑄3 𝑥̇(𝑡 − 𝜏2 (𝑡)) (𝜏∗ − 1) .
𝑇
+∫ 𝑥̇ (𝑠) 𝑅22 𝑥̇(𝑠) 𝑑𝑠 (18)
𝑡−𝜏2

In addition, note that

+ 𝜎𝑦𝑇 (𝑡 − 𝜎 (𝑡)) 𝑆11 𝑦 (𝑡 − 𝜎 (𝑡))
𝑇
0 = 2𝑥̇ (𝑡) 𝑄4 [−𝑥̇(𝑡) + 𝑥̇(𝑡)]
+ 𝜎𝑓𝑇 (𝑦 (𝑡 − 𝜎 (𝑡))) 𝑆22 𝑓 (𝑦 (𝑡 − 𝜎 (𝑡)))
𝑇
𝑡
= 2𝑥̇ (𝑡) 𝑄4 [−𝑥̇(𝑡) − 𝐴𝑥 (𝑡 − 𝜌1 ) + 𝑊𝑓 (𝑦 (𝑡 − 𝜎 (𝑡)))]
𝑇
+∫ 𝑦̇ (𝑠) 𝑆33 𝑦̇(𝑠) 𝑑𝑠
𝑇 𝑇
𝑡−𝜎 = −2𝑥̇ (𝑡) 𝑄4 𝑥̇(𝑡) − 2𝑥̇ (𝑡) 𝑄4 𝐴𝑥 (𝑡 − 𝜌1 )
+ 𝜎𝑦𝑇 (𝑡 − 𝜎 (𝑡)) 𝑆12 𝑓 (𝑦 (𝑡 − 𝜎 (𝑡))) 𝑇
+ 2𝑥̇ (𝑡) 𝑄4 𝑊𝑓 (𝑦 (𝑡 − 𝜎 (𝑡))) ,
+ 𝑦𝑇 (𝑡 − 𝜎 (𝑡)) 𝑆13 [𝑦 (𝑡) − 𝑦 (𝑡 − 𝜎 (𝑡))]
(19)
+ 𝑓𝑇 (𝑦 (𝑡 − 𝜎 (𝑡))) 𝑆23 [𝑦 (𝑡) − 𝑦 (𝑡 − 𝜎 (𝑡))] ,
0
0 = 2𝑦𝑇̇ (𝑡) 𝑄5 [−𝑦̇(𝑡) + 𝑦̇(𝑡)]
𝑉4̇(𝑡) = 𝜎𝑦𝑇̇ (𝑡) 𝑆33 𝑦̇(𝑡) − ∫ 𝑦𝑇̇ (𝑡 + 𝑡) 𝑆33 𝑦̇(𝑡 + 𝑡) 𝑑𝑡
−𝜎 = 2𝑦𝑇̇ (𝑡) 𝑄5 [−𝑦̇(𝑡) − 𝐶𝑦 (𝑡 − 𝜌2 )
0
𝑇 𝑇
+ 𝜏1 𝑥̇ (𝑡) 𝑇22 𝑥̇(𝑡) − ∫ 𝑥̇ (𝑡 + 𝑡) 𝑇22 𝑥̇(𝑡 + 𝑡) 𝑑𝑡 + 𝐷1 𝑥 (𝑡 − 𝜏1 (𝑡)) + 𝐷2 𝑥̇(𝑡 − 𝜏2 (𝑡))]
−𝜏1

𝑡 = −2𝑦𝑇̇ (𝑡) 𝑄5 𝑦̇(𝑡) − 2𝑦𝑇̇ (𝑡) 𝑄5 𝐶𝑦 (𝑡 − 𝜌2 )

= 𝜎𝑦𝑇̇ (𝑡) 𝑆33 𝑦̇(𝑡) − ∫ 𝑦𝑇̇ (𝑠) 𝑆33 𝑦̇(𝑠) 𝑑𝑠
𝑡−𝜎
+ 2𝑥𝑇 (𝑡 − 𝜏1 (𝑡)) 𝐷1 𝑄5𝑇 𝑦̇(𝑡)
𝑡
𝑇 𝑇
+ 𝜏1 𝑥̇ (𝑡) 𝑇22 𝑥̇(𝑡) − ∫ 𝑥̇ (𝑠) 𝑇22 𝑥̇(𝑠) 𝑑𝑠, 𝑇
𝑡−𝜏1 + 2𝑥̇ (𝑡 − 𝜏2 (𝑡)) 𝐷2𝑇 𝑄5𝑇 𝑦̇(𝑡) ,
(20)
𝑉5̇(𝑡) = 𝑥𝑇 (𝑡) 𝑅1 𝑥 (𝑡) + 𝑥̇
𝑇
(𝑡) 𝑅1 𝑥̇(𝑡)
𝑇
0 = 2𝑥̇ (𝑡 − 𝜏2 (𝑡)) 𝐷2𝑇 𝑄6
− 𝑥𝑇 (𝑡 − 𝜏2 ) 𝑅1 𝑥 (𝑡 − 𝜏2 )
× [−𝐷2 𝑥̇(𝑡 − 𝜏2 (𝑡)) + 𝐷2 𝑥̇(𝑡 − 𝜏2 (𝑡))]
𝑇
− 𝑥̇ (𝑡 − 𝜏2 ) 𝑅2 𝑥̇(𝑡 − 𝜏2 )
𝑇
= 2𝑥̇ (𝑡 − 𝜏2 (𝑡)) 𝐷2𝑇 𝑄6
+ 𝜏2 𝑥𝑇 (𝑡) 𝑅3 𝑥 (𝑡) + 𝜏2 𝑥̇
𝑇
(𝑡) 𝑅22 𝑥̇(𝑡)
𝑡
× [−𝐷2 𝑥̇(𝑡 − 𝜏2 (𝑡)) + 𝑦̇(𝑡) + 𝐶𝑦 (𝑡 − 𝜌2 )
𝑇
−∫ 𝑥 (𝑠) 𝑅3 𝑥 (𝑠) 𝑑𝑠
𝑡−𝜏2 −𝐷1 𝑥 (𝑡 − 𝜏1 (𝑡))]
𝑡 𝑇
−∫ 𝑇
𝑥̇ (𝑠) 𝑅22 𝑥̇(𝑠) 𝑑𝑠 = −2𝑥̇ (𝑡 − 𝜏2 (𝑡)) 𝐷2𝑇 𝑄6 𝐷2 𝑥̇(𝑡 − 𝜏2 (𝑡))
𝑡−𝜏2
𝑇
𝑇 𝑇
+ 2𝑥̇ (𝑡 − 𝜏2 (𝑡)) 𝐷2𝑇 𝑄6 𝑦̇(𝑡)
+ 𝑥̇ (𝑡) 𝑄3 𝑥̇(𝑡) − 𝑥̇ (𝑡 − 𝜏2 (𝑡))
𝑇
+ 2𝑥̇ (𝑡 − 𝜏2 (𝑡)) 𝐷2𝑇 𝑄6 𝐶𝑦 (𝑡 − 𝜌2 )
× 𝑄3 𝑥̇(𝑡 − 𝜏2 (𝑡)) [1 − 𝜏2̇(𝑡)]
− 2𝑥𝑇 (𝑡 − 𝜏1 (𝑡)) 𝐷1𝑇 𝑄6𝑇 𝐷2 𝑥̇
𝑇
(𝑡 − 𝜏2 (𝑡)) .
≤ 𝑥𝑇 (𝑡) 𝑅1 𝑥 (𝑡) + 𝑥̇
𝑇
(𝑡) 𝑅1 𝑥̇(𝑡)
(21)
− 𝑥𝑇 (𝑡 − 𝜏2 ) 𝑅1 𝑥 (𝑡 − 𝜏2 ) From (9) to (21), we have
𝑇
− 𝑥̇ (𝑡 − 𝜏2 ) 𝑅2 𝑥̇(𝑡 − 𝜏2 ) 𝑉̇(𝑡) ≤ 𝜉𝑇 (𝑡) ∏ 𝜉 (𝑡) , (22)
6 Abstract and Applied Analysis

where ∏ is defined in (11) and Then we have

𝑘𝑖+ = 0.65, 𝑘𝑖− = 0,
(26)
𝜉𝑇 (𝑡) = [𝑥𝑇 (𝑡) , 𝑥̇
𝑇
(𝑡) , 𝑥𝑇 (𝑡 − 𝜌1 ) , 𝑥𝑇 (𝑡 − 𝜏1 (𝑡)) , 𝐾 = diag [0.325, 0.325] , 𝐿 = 0.

𝑡 𝑇 Time-varying delays and the leakage delays are chosen as

𝑥𝑇 (𝑡 − 𝜏2 (𝑡)) , (∫ 𝑥 (𝜇) 𝑑𝜇) ,
𝑡−𝜌1 𝜏1 = 0.995, 𝜏2 = 1, 𝜌1 = 𝜌2 = 0.01, 𝜎 = 0.01.
(27)
𝑦𝑇 (𝑡) , 𝑦𝑇̇ (𝑡) , 𝑦𝑇 (𝑡 − 𝜌2 ) , 𝑦𝑇 (𝑡 − 𝜎 (𝑡)) ,
(23) By using the MATLAB LMI Toolbox, we obtain
𝑓𝑇 (𝑦 (𝑡)) , 𝑓𝑇 (𝑦 (𝑡 − 𝜎 (𝑡))) ,
0.3039 0.0035 0.0011 0.0000
𝑡 𝑇 𝑃=[ ], 𝑄=[ ],
0.0035 0.1728 0.0000 0.0544
(∫ 𝑦 (𝜇) 𝑑𝜇) , 𝑥𝑇 (𝑡 − 𝜏2 ) , 𝑥̇
𝑇
(𝑡 − 𝜏2 ) ,
𝑡−𝜌2
1.4529 0.0114 64.7658 0.5258
𝑇 𝑃1 = [ ], 𝑃2 = [ ],
𝑡 0.0114 0.9774 0.5258 59.4535
(∫ 𝑥 (𝜇) 𝑑𝜇) ]
𝑡−𝜏2
0.0007 0.0001 4.2036 0.0059
𝑄1 = [ ], 𝑄2 = [ ],
0.0001 0.0627 0.0059 17.3389
which implies 𝑉(𝑡)̇ ≤ 0. It is easy to prove 𝑉(𝑡) ̇ ≤
𝑇 𝑇 𝑇
𝜆 min (∏)𝜉 (𝑡)𝜉(𝑡) ≤ 𝜆 min (∏)[𝑥 (𝑡)𝑥(𝑡) + 𝑦 (𝑡)𝑦(𝑡)], where 0.0100 0 0.0397 0.0007
𝜆 min (∏) < 0. According to the Lyapunov stability theory, 𝑄3 = [ ], 𝑄4 = [ ],
0 0.0100 0.0011 0.0178
the GRNs (5) are asymptotically stable. The proof is com-
plete. 0.0001 0.0000 0.0001 0.0000
𝑄5 = [ ], 𝑄6 = [ ],
0.0000 0.0100 0.0000 0.0101
Remark 6. If the leakage delay 𝜌1 = 𝜌2 = 0, the model is the
same as that in [27]. We can also deduce the sufficient
0.1108 0.0002 0.2207 0.0005
conditions for guaranteeing system stability. So the model 𝑇11 = [ ], 𝑇12 = [ ],
0.0002 0.0870 0.0005 0.1803
in [27] is a special case of the model in this paper. In
fact, the leakage delay has a large effect on the dynamic
0.0022 0.0001 0.0084 0.0005
behaviors of the model. It often leads to the instability. This 𝑇22 = [ ], 𝑆11 = [ ],
0.0001 0.0011 0.0005 0.7474
phenomenon can be observed in Figure 1. Considering the
impact of leakage delay, we analyze GRNs with neutral delays
−25.6827 −0.0276
and leakage delays simultaneously. 𝑆12 = [ ],
−1.7215 −21.2464

4. Simulation Results 0.0003 −0.0000

𝑆13 = [ ],
−0.0000 0.0344
In this section, an example is given to demonstrate the fea-
sibility and efficiency of the theoretical results above. The 10.4912 −1.0149
simulation is conducted on MATLAB, and the ordinary 𝑆22 = [ ],
−1.0149 9.3772
differential equation is solved by Runge-Kutta method.
0.0097 −0.0195
Example 7. Consider the following neutral-type genetic reg- 𝑆23 = [ ],
−0.0004 −0.1500
ulatory networks with leakage delays:
0.0018 0.0002 0.1961 0.0029
𝑆33 = [ ], 𝑅11 = [ ],
𝑥̇(𝑡) = − 𝐴𝑥 (𝑡 − 𝜌1 ) + 𝑊𝑓 (𝑦 (𝑡 − 𝜎 (𝑡))) , 0.0002 0.1414 0.0029 0.1490

𝑦̇(𝑡) = − 𝐶𝑦 (𝑡 − 𝜌2 ) + 𝐷1 𝑥 (𝑡 − 𝜏1 (𝑡)) + 𝐷2 𝑥̇(𝑡 − 𝜏2 (𝑡)) , 0.0188 0.0002 0.2200 0.0077

𝑅12 = [ ], 𝑅22 = [ ],
(24) 0.0004 0.0048 0.0077 0.1084

0.0044 0.0001 12.3518 0

𝑅1 = [ ], 𝑅2 = [ ],
where 0.0001 0.0014 0 12.3518

6 0 −1 0 2.5 0 3.4620 0.0037 0.0084 0

𝐴=[ ], 𝑊=[ ], 𝐶=[ ], 𝑅3 = [ ], Λ1 = [ ],
0 7 1 2 0 2.5 0.0037 3.1125 0 0.0084

0.2 0 8 0 𝑦2 0.7930 0
𝐷1 = [ ], 𝐷2 = [ ], 𝑓 (𝑦) = . Λ2 = [ ].
0 0.2 0 0.8 1 + 𝑦2 0 0.7930
(25) (28)
Abstract and Applied Analysis 7

200
200
100 100
x 0 y 0

−100 −100
−200
−200
0 20 40 60 80 100 0 20 40 60 80 100
t t
(a) (b)

Figure 1: State trajectory of the system with 𝜌1 = 𝜌2 = 0.24 for Example 7 with two nodes.

1.5
10
1
0.5 0
x y
0 −10
−0.5
−20
−1
0 10 20 30 40 50 0 10 20 30 40 50
t t
(a) (b)

Figure 2: State trajectory of the system with 𝜌1 = 𝜌2 = 0.13 for Example 7 with two nodes.

Table 1: Effect of leakage delay on the stability. together in the paper, which extended and improved the
previous results in [27, 30]. Finally, a numerical simulation
𝜌 (leakage delay) Condition Stability has been given to show the effectiveness of our theoretic
0 < 𝜌 ≤ 0.229 Satisfied Stable results.
0.229 < 𝜌 ≤ 0.239 Dissatisfied Stable
𝜌 ≥ 0.240 Dissatisfied Unstable
Conflict of Interests
The authors declare that there is no conflict of interests
regarding the publication of this paper.
This means that all conditions in Theorem 5 are satisfied.
Hence, the GRN model (24) is asymptotically stable. On the
other hand, we have the following simulation results shown Acknowledgments
in Figure 2, which also show the correctness of the theoretic
analysis. This work was jointly supported by the National Natural
Science Foundation of China under Grant 11226116, the Foun-
Remark 8. In order to reflect the effect of the leakage delays dation of Key Laboratory of Advanced Process Control for
on the stability of GRNs, we have given the simulation Light Industry (Jiangnan University), Ministry of Education,
results by using different leakage delays. Table 1 shows the China, and the Fundamental Research Funds for the Central
conclusions when the leakage delays are changed. When Universities (JUSRP51317B, JUDCF13042).
leakage delays specially 𝜌1 = 𝜌2 = 0.23, the conditions in
Theorem 5 are not satisfied. However, the system is still stable. References
Thus, there still exist rooms for improving the sufficient
conditions in Theorem 5. [1] U. Bower and H. Balouri, Computational Modelling of Genetic
and Biochemical Networks, MIT Press, Cambridge, Mass, USA,
2001.
5. Conclusion
[2] E. Davidson, Genomic Regulatory Systems, Academic Press, San
In this paper, we have investigated the stability of neural- Diego, Calif, USA, 2001.
type GRNs with leakage delays. Based on Lyapunov stability [3] E. Sontag, A. Kiyatkin, and B. N. Kholodenko, “Inferring dyna-
theory and linear matrix inequalities, some sufficient condi- mic architecture of cellular networks using time series of gene
tions have been obtained to guarantee stability for GRNs. It is expression, protein and metabolite data,” Bioinformatics, vol. 20,
noted that the leakage delay and neutral delay are considered no. 12, pp. 1877–1886, 2004.
8 Abstract and Applied Analysis

[4] P. Balasubramaniam, R. Rakkiyappan, and R. Krishnasamy, [21] J. Qiu and J. Cao, “Delay-dependent robust stability of neutral-
“Stochastic stability of Markovian jumping uncertain stochastic type neural networks with time delays,” Journal of Mathematical
genetic regulatory networks with interval time-varying delays,” Control Science and Applications, vol. 1, no. l, pp. 179–188, 2007.
Mathematical Biosciences, vol. 226, no. 2, pp. 97–108, 2010. [22] J. H. Park, O. M. Kwon, and S. M. Lee, “LMI optimization app-
[5] H. de Jong, “Modeling and simulation of genetic regulatory sys- roach on stability for delayed neural networks of neutral-type,”
tems: a literature review,” Journal of Computational Biology, vol. Applied Mathematics and Computation, vol. 196, no. 1, pp. 236–
9, no. 1, pp. 67–103, 2002. 244, 2008.
[6] N. Friedman, M. Linial, I. Nachman, and D. Pe’er, “Using [23] J. H. Park, C. H. Park, O. M. Kwon, and S. M. Lee, “A new
Bayesian networks to analyze expression data,” Journal of Com- stability criterion for bidirectional associative memory neural
putational Biology, vol. 7, no. 3-4, pp. 601–620, 2000. networks of neutral-type,” Applied Mathematics and Compu-
[7] C. Chaouiya, E. Remy, and D. Thieffry, “Petri net modelling of tation, vol. 199, no. 2, pp. 716–722, 2008.
biological regulatory networks,” Journal of Discrete Algorithms, [24] J. Liu and G. Zong, “New delay-dependent asymptotic stability
vol. 6, no. 2, pp. 165–177, 2008. conditions concerning BAM neural networks of neutral type,”
[8] S. Hardy and P. N. Robillard, “Modeling and simulation of Neurocomputing, vol. 72, no. 10–12, pp. 2549–2555, 2009.
molecular biology systems using Petri nets: modeling goals of [25] S. Lakshmanan and P. Balasubramaniam, “New results of robust
various approaches,” Journal of Bioinformatics and Computa- stability analysis for neutral-type neural networks with time-
tional Biology, vol. 2, no. 4, pp. 619–637, 2004. varying delays and Markovian jumping parameters,” Canadian
[9] Y. Tang, Z. Wang, and J. Fang, “Parameters identification of Journal of Physics, vol. 89, no. 8, pp. 827–840, 2011.
unknown delayed genetic regulatory networks by a switching [26] D. S. Jung, J. H. Koo, and S. C. Won, “Asymptotic stability of
particle swarm optimization algorithm,” Expert Systems with neutral functional differential equation model of genetic regu-
Applications, vol. 38, no. 3, pp. 2523–2535, 2011. latory networks with time-varying delays,” in Proceedings of
[10] C. Li, L. Chen, and K. Aihara, “Stability of genetic networks with the SICE Annual Conference, pp. 63–68, Taipei, Taiwan, August
SUM regulatory logic: lur’e system and LMI approach,” IEEE 2010.
Transactions on Circuits and Systems I: Regular Papers, vol. 53, [27] T. Jiao, G. Zong, and W. Zheng, “New stability conditions for
no. 11, pp. 2451–2458, 2006. GRNs with neutral delay,” Soft Computing, vol. 17, no. 4, pp. 703–
[11] G. He, J. Fang, and X. Wu, “Robust stability of Markovian jump- 712, 2013.
ing genetic regulatory networks with mode-dependent delays,” [28] K. Gopalsamy, “Leakage delays in BAM,” Journal of Mathemati-
Mathematical Problems in Engineering, vol. 2012, Article ID cal Analysis and Applications, vol. 325, no. 2, pp. 1117–1132, 2007.
504378, 18 pages, 2012.
[29] S. Peng, “Global attractive periodic solutions of BAM neural
[12] L. Yin and Y. Liu, “Exponential stability analysis for genetic networks with continuously distributed delays in the leakage
regulatory networks with both time-varying and continuous terms,” Nonlinear Analysis: Real World Applications, vol. 11, no.
distributed delays,” Abstract and Applied Analysis, vol. 2014, 3, pp. 2141–2151, 2010.
Article ID 897280, 10 pages, 2014.
[30] R. Rakkiyappan, P. Balasubramaniam, and K. Balachandran,
[13] H. Kitano, Foundations of Systems Biology, MIT Press, Cam- “Delay-dependent global asymptotic stability criteria for gene-
bridge, Mass, USA, 2001. tic regulatory networks with time delays in the leakage term,”
[14] M. Wu, F. Liu, P. Shi, Y. He, and R. Yokoyama, “Exponential Physica Scripta, vol. 84, no. 5, Article ID 055007, 2011.
stability analysis for neural networks with time-varying delay,” [31] J. Li, M. Hu, J. Cao, Y. Yang, and Y. Jin, “Stability of uncertain
IEEE Transactions on Systems, Man, and Cybernetics, vol. 38, no. impulsive stochastic genetic regulatory networks with time-
4, pp. 1152–1156, 2008. varying delay in the leakage term,” Abstract and Applied Analy-
[15] S. Mou, H. Gao, W. Qiang, and K. Chen, “New delay-dependent sis, vol. 2014, Article ID 706720, 15 pages, 2014.
exponential stability for neural networks with time delay,” IEEE [32] W. Pan, Z. Wang, H. Gao, and X. Liu, “Monostability and
Transactions on Systems, Man, and Cybernetics B: Cybernetics, multistability of genetic regulatory networks with different
vol. 38, no. 2, pp. 571–576, 2008. types of regulation functions,” Nonlinear Analysis: Real World
[16] J. Cao and F. Ren, “Exponential stability of discrete-time genetic Applications, vol. 11, no. 4, pp. 3170–3185, 2010.
regulatory networks with delays,” IEEE Transactions on Neural [33] K. Gu, “An integral inequality in the stability problem of
Networks, vol. 19, no. 3, pp. 520–523, 2008. time-delay systems,” in Proceedings of the 39th IEEE Confernce
[17] W. He and J. Cao, “Robust stability of genetic regulatory net- on Decision and Control, pp. 2805–2810, Sydney, Australia,
works with distributed delay,” Cognitive Neurodynamics, vol. 2, December 2000.
no. 4, pp. 355–361, 2008. [34] S. Lakshmanan, J. H. Park, R. Rakkiyappan, and H. Y.
[18] Y. Cao and C. Bai, “Finite-time stability of fractional-order Jung, “State estimator for neural networks with sampled data
BAM neural networks with distributed delay,” Abstract and App- using discontinuous Lyapunov functional approach,” Nonlinear
lied Analysis, vol. 2014, Article ID 634803, 8 pages, 2014. Dynamics, vol. 73, no. 1-2, pp. 509–520, 2013.
[19] X. Lou, Q. Ye, and B. Cui, “Exponential stability of genetic regu- [35] T. Li, S. Fei, and Q. Zhu, “Design of exponential state estimator
latory networks with random delays,” Neurocomputing, vol. 73, for neural networks with distributed delays,” Nonlinear Analysis:
no. 4-6, pp. 759–769, 2010. Real World Applications, vol. 10, no. 2, pp. 1229–1242, 2009.
[20] W. Zhang, J. Fang, and Y. Tang, “Stochastic stability of Marko-
vian jumping genetic regulatory networks with mixed time
delays,” Applied Mathematics and Computation, vol. 217, no. 17,
pp. 7210–7225, 2011.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2015, Article ID 620891, 12 pages
http://dx.doi.org/10.1155/2015/620891

Research Article
Dynamic Analysis of a Delayed Reaction-Diffusion
Predator-Prey System with Modified Holling-Tanner
Functional Response

Xinhong Pan,1,2 Min Zhao,2,3 Chuanjun Dai,2,3 and Yapei Wang1,2

1
School of Mathematics and Information Science, Wenzhou University, Wenzhou, Zhejiang 325035, China
2
Zhejiang Provincial Key Laboratory for Water Environment and Marine Biological Resource Protection,
Wenzhou University, Wenzhou, Zhejiang 325035, China
3
School of Life and Environmental Science, Wenzhou University, Wenzhou, Zhejiang 325035, China

Correspondence should be addressed to Min Zhao; zmcn@tom.com

Received 21 May 2014; Accepted 21 July 2014

Academic Editor: Sanling Yuan

Copyright © 2015 Xinhong Pan et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

A predator-prey model with modified Holling-Tanner functional response and time delays is considered. By regarding the delays
as bifurcation parameters, the local and global asymptotic stability of the positive equilibrium are investigated. The system has been
found to undergo a Hopf bifurcation at the positive equilibrium when the delays cross through a sequence of critical values. In
addition, the direction of the Hopf bifurcation and the stability of bifurcated periodic solutions are also studied, and an explicit
algorithm is obtained by applying normal form theory and the center manifold theorem. The main results are illustrated by
numerical simulations.

1. Introduction In recent years, models with time delay have been exten-
sively studied by many authors [5, 16–26]. The authors of [1]
The dynamic relationship between prey and predators has discussed model (1) with a discrete delay:
long been and will continue to be one of the dominant sub-
jects in mathematical ecology due to its universal existence 𝑑𝑢 𝑢 (𝑡 − 𝜏)
and importance [1–12]. In [13, 14], the author proposed the = 𝑟𝑢 (1 − ) − 𝑚𝑢V,
𝑑𝑡 𝐾
following predator-prey model based on the model in May (2)
[15]: 𝑑V ℎV
= V [𝑠 (1 − )]
𝑑𝑡 𝑢
𝑑𝑢 𝑢
= 𝑟𝑢 (1 − ) − 𝑝 (𝑢) V,
𝑑𝑡 𝐾 and obtained the stability of equilibria, the existence of
(1) Hopf bifurcation, and the direction of bifurcating periodic
𝑑V ℎV
= V [𝑠 (1 − )] , solutions. This paper focuses mainly on the effects of both
𝑑𝑡 𝑢
spatial diffusion and time delay on system (1). It is assumed
where 𝑢 and V denote the population of prey and predator, that the delay affects predation and consumption and that the
respectively, and 𝑟 and 𝑠 are the intrinsic growth rates of prey system has homogeneous Neumann boundary conditions:
and predator, respectively. The parameter 𝐾 represents the
carrying capacity of the prey and the ratio 𝑢/ℎ represents 𝜕𝑢 𝑢 𝑚𝑢2 V (𝑡 − 𝜏1 )
the carrying capacity of the predator. It has been assumed = 𝑑1 Δ𝑢 + 𝑟𝑢 (1 − ) − , 𝑥 ∈ Ω, 𝑡 > 0,
𝜕𝑡 𝐾 𝑎 + 𝑢2
that both prey and predator populations grow logistically and
that the predator consumes the prey according to a functional 𝜕V ℎV
= 𝑑2 ΔV + V [𝑠 (1 − )] , 𝑥 ∈ Ω, 𝑡 > 0,
𝑝(𝑢). 𝜕𝑡 𝑢 (𝑡 − 𝜏2 )
2 Abstract and Applied Analysis

𝜕𝑢 𝜕V where 𝑑Δ = (𝑑1 Δ, 𝑑2 Δ),

= = 0, 𝑥 = 𝜕Ω, 𝑡 > 0,
𝜕𝑛 𝜕𝑛
𝑢 (𝑥, 𝑡) = 𝑢0 (𝑥, 𝑡) , 𝑥 ∈ Ω, 𝑡 ∈ [−𝜏2 , 0] , dom (𝑑Δ) = {(𝑢, V)𝑇 : 𝑢, V ∈ 𝐶2 ([0, 𝑙𝜋] , 𝑅) ,
(8)
V (𝑥, 𝑡) = V0 (𝑥, 𝑡) , 𝑥 ∈ Ω, 𝑡 ∈ [−𝜏1 , 0] , 𝜕𝑢 𝜕V
(3) = = 0, 𝑥 = 0, 𝑙𝜋} ,
𝜕𝑥 𝜕𝑥

where 𝑑1 and 𝑑2 are the diffusion coefficients of prey and and 𝐿 : 𝐶([−𝜏, 0], 𝑋) → 𝑋 is defined as
predator, respectively, Δ = 𝜕2 /𝜕𝑥2 denotes the Laplacian 𝐿 (𝜙)
operator, and 𝑛 is the outward unit normal vector on 𝜕Ω. For
convenience, it is assumed that Ω = (0, 𝑙𝜋), 𝑙 > 0 and that all 2𝑟𝑢∗ 2𝑎𝑚𝑢∗ V∗ 𝑚𝑢∗
parameters are positive. [𝑟 − − ]𝜙1 (0) − 𝜙 (−𝜏1)
The rest of this paper is structured as follows. In Section 2, =( 𝐾 (𝑎 + 𝑢∗2 )
2 𝑎 + 𝑢∗2 2 )
𝑠
the local stability of equilibria is analyzed using the associated 𝜙1 (−𝜏2 ) − 𝑠𝜙2 (0)
characteristic equations, and the occurrence of the Hopf ℎ
(9)
bifurcation with time delays is presented. In Section 3, the
global asymptotical stability of the interior equilibrium for
for 𝜙 = (𝜙1 , 𝜙2 )𝑇 ∈ 𝐶([−𝜏, 0], 𝑋).
any 𝜏1 , 𝜏2 ≥ 0 is proved by means of the upper-lower
The characteristic equation of (7) is
solution method. In Section 4, using normal form theory and
the center manifold theorem, the stability and direction of
𝜆𝑦 − 𝑑Δ𝑦 − 𝐿 (𝑒𝜆 𝑦) = 0, 𝑦 ∈ dom (𝑑Δ) , 𝑦 ≠
0. (10)
bifurcating periodic orbits are investigated. Finally, numerical
simulations and a brief discussion are presented. Recall that −Δ under the Neumann boundary condition
has eigenvalues 0 = 𝜇0 < 𝜇1 ≤ 𝜇2 ≤ ⋅ ⋅ ⋅ ≤ 𝜇𝑛 ≤ 𝜇𝑛+1 ≤ ⋅ ⋅ ⋅
2. Local Stability and Hopf and lim𝑖 → ∞ 𝜇𝑖 = ∞, with the corresponding eigenfunctions
Bifurcation Analysis 𝜓𝑛 (𝑥). Substituting
∞
In this section, the local stability of the equilibria of system 𝑦
𝑦 = ∑ 𝜓𝑛 (𝑥) ( 1𝑛 ) (11)
(3) is analyzed. Denote 𝑛=0
𝑦2𝑛

into (10), the following expression results:

𝑋 = 𝐶 ([0, 𝑙𝜋] , 𝑅2 ) ,
(4)
2𝑟𝑢∗ 2𝑎𝑚𝑢∗ V∗ 𝑚𝑢∗ −𝜆𝜏1
⟨𝑢, V⟩ = ⟨𝑢1 , V1 ⟩ + ⟨𝑢2 , V2 ⟩ , 𝜏 = 𝜏1 + 𝜏2 , 𝑟− − 2
− 𝜇𝑛 𝑑1 − 𝑒
𝐾 (𝑎 + 𝑢∗2 ) 𝑎 + 𝑢∗2 𝑦
( ) ( 1𝑛 )
𝑠 −𝜆𝜏2 𝑦2𝑛
for 𝑢 = (𝑢1 , 𝑢2 ), V = (V1 , V2 ) ∈ 𝑋; then (𝑋, ⟨⋅, ⋅⟩) is a Hilbert 𝑒 −𝑠 − 𝜇𝑛 𝑑2
space. In the abstract space 𝐶([−𝜏, 0], 𝑋), system (3) can be ℎ
regarded as an abstract functional differential equation.
𝑦
System (3) has two nonnegative equilibria (𝐾, 0) and = 𝜆 ( 1𝑛 ) .
𝑦2𝑛
(𝑢∗ , V∗ ), where
(12)
1 ∗
V∗ = 𝑢 , (5) Hence, it can be concluded that the characteristic equation
ℎ (10) is equivalent to the equation

and 𝑢∗ is the positive root of the equation 𝜆2 + 𝐴 𝑛 𝜆 + 𝐵𝑛 + 𝐶𝑒−𝜆𝜏 = 0, 𝑛 = 0, 1, 2, . . . , (13)

𝑟ℎ𝑢3 + (𝑚𝐾 − 𝑟ℎ𝐾) 𝑢2 + 𝑎𝑟ℎ𝑢 − 𝑎𝑟ℎ𝐾 = 0. (6) where

2𝑟𝑢∗ 2𝑎𝑚𝑢∗ V∗
𝐴 𝑛 = 𝜇𝑛 (𝑑1 + 𝑑2 ) − 𝑟 + + + 𝑠,
Let 𝐴(𝑢) = 𝑟ℎ𝑢3 + (𝑚𝐾 − 𝑟ℎ𝐾)𝑢2 + 𝑎𝑟ℎ𝑢 − 𝑎𝑟ℎ𝐾; then 𝐴(0) = 𝐾 (𝑎 + 𝑢∗2 )
2

−𝑎𝑟ℎ𝐾 < 0, 𝐴(𝐾) = 𝑚𝐾3 > 0, which guarantee the existence

of 𝑢∗ ∈ (0, 𝐾). 2𝑟𝑢∗ 2𝑎𝑚𝑢∗ V∗
From analysis of the characteristic equation of (𝐾, 0), it 𝐵𝑛 = (𝑟 − − 2
− 𝜇𝑛 𝑑1 ) (−𝑠 − 𝜇𝑛 𝑑2 ) , (14)
𝐾 (𝑎 + 𝑢∗2 )
can easily be determined that it always has a saddle point. To
analyze the stability of the positive equilibrium (𝑢∗ , V∗ ), the 𝑠 𝑚𝑢∗
first step is to linearize system (3) at (𝑢∗ , V∗ ): 𝐶= , 𝜏 = 𝜏1 + 𝜏2 .
ℎ 𝑎 + 𝑢∗2

𝜕𝑈 (𝑡) The stability of the positive equilibrium (𝑢∗ , V∗ ) can be

= 𝑑Δ𝑈 (𝑡) + 𝐿 (𝑈𝑡 ) , (7) determined by the distribution of the roots of (13). It is locally
𝜕𝑡
Abstract and Applied Analysis 3

asymptotically stable if all the roots of (13) have negative real Proof. If 𝐵𝑛2 − 𝐶2 < 0, then there exists 𝑁0 ≥ 0 such that, for
parts for all 𝑛 = 0, 1, 2, . . .. Obviously, 0 is not a root of (13) 0 ≤ 𝑛 ≤ 𝑁0 , (20) has a unique positive real root:
for all 𝑛 = 0, 1, 2, . . .. When 𝜏 = 0 as well as 𝜏1 = 𝜏2 = 0, (13)
can be simplified as
2 2
√ 2𝐵𝑛 − 𝐴 𝑛 + √(𝐴 𝑛 − 2𝐵𝑛 ) − 4 (𝐵𝑛 − 𝐶 )
2 2 2
𝜆2 + 𝐴 𝑛 𝜆 + 𝐵𝑛 = 0. (23)
(15) 𝑤0 = .
2
It can be verified that
From (20), it follows that
2𝑟𝑢∗
𝐴 𝑛 > 0, 𝐵𝑛 > 0, if 𝑟 − < 0; (16)
𝐾
𝑤0 𝐴 𝑛 𝑤02 − 𝐵𝑛
that is, sin 𝑤0 𝜏 = , cos 𝑤0 𝜏 = . (24)
𝐶 𝐶
𝐾
𝑢∗ > , (17) Then (13) has a pair of pure imaginary roots ±𝑖𝑤0 when
2
which requires that
2𝑘𝜋
𝜏 = 𝜏𝑘 = 𝜏0 + , 𝑘 = 0, 1, 2 . . . , (25)
𝐾 𝑚 𝑟ℎ 1 𝑤0
𝐴 ( ) = 𝐾3 − 𝐾3 − 𝑎𝑟ℎ𝐾 < 0 (18)
2 4 8 2
or where

2𝑚 − 𝑟ℎ < 0. 2
(19) 1 𝑤 − 𝐵𝑛
𝜏0 = (arccos 0 ). (26)
𝑤0 𝐶
Therefore, the following theorem can be stated.

Theorem 1. If 2𝑚 − 𝑟ℎ < 0 holds, the interior equilibrium Substituting 𝜆(𝜏) into (13) and taking the derivatives with
(𝑢∗ , V∗ ) of system (3) with 𝜏1 = 𝜏2 = 0 is asymptotically stable. respect to 𝜏 lead to
When 𝜏 ≠ 0, assume that 𝜆 = 𝑖𝑤0 (𝑤0 > 0) is a root of
(13). Substituting it into (13) yields:
𝑑𝜆 (𝜏) 𝑑𝜆 (𝜏) 𝑑𝜆 (𝜏)
2𝜆 + 𝐴𝑛 + 𝐶𝑒−𝜆𝜏 (−𝜆 − 𝜏 ) = 0. (27)
𝑤02 − 𝐵𝑛 = 𝐶 cos 𝑤0 𝜏, 𝑑𝜏 𝑑𝜏 𝑑𝜏
(20)
𝑤0 𝐴 𝑛 = 𝐶 sin 𝑤0 𝜏;
For 𝜏 = 𝜏0 , considering 𝐶𝑒−𝑖𝑤0 𝜏 = 𝑤02 − 𝐵𝑛 − 𝑖𝑤0 𝐴 𝑛 , it follows
that
that is,

𝑤04 + (𝐴2𝑛 − 2𝐵𝑛 ) 𝑤02 + 𝐵𝑛2 − 𝐶2 = 0, (21) 𝑑 Re (𝜆 (𝜏)) 󵄨󵄨󵄨󵄨 𝑑𝜆 (𝜏) 󵄨󵄨󵄨󵄨
sign { }󵄨󵄨 = sign {Re ( )}󵄨󵄨
𝑑𝜏 󵄨󵄨𝜆=𝑖𝑤0 𝑑𝜏 󵄨󵄨𝜆=𝑖𝑤0
where
𝑖𝐶𝑤0 𝑒−𝑖𝑤0 𝜏0
2 = sign {Re ( )}
2𝑟𝑢∗ 2𝑎𝑚𝑢∗ V∗ 2𝑤0 𝑖 + 𝐴 𝑛 − 𝐶𝜏0 𝑒−𝑖𝑤0 𝜏0
𝐴2𝑛 − 2𝐵𝑛 = (−𝑟 + + 2
+ 𝜇𝑛 𝑑1 )
𝐾 (𝑎 + 𝑢∗2 ) 𝑖𝑤03 − 𝐵𝑛 𝑤0 𝑖 + 𝑤02 𝐴 𝑛
= sign {Re ( )}
+ (𝑠 + 𝜇𝑛 𝑑2 ) ,
2 2𝑤0 𝑖 + 𝐴 𝑛 − 𝜏0 (𝑤02 − 𝐵𝑛 − 𝑖𝑤0 𝐴 𝑛 )

2 (22) 𝑤02 𝐴 𝑛 + 𝑖 (𝑤03 − 𝐵𝑛 𝑤0 )

2𝑟𝑢∗ 2𝑎𝑚𝑢∗ V∗ = sign {Re ( )}
𝐵𝑛2 2
− 𝐶 = (−𝑟 + + 2
+ 𝜇𝑛 𝑑1 ) −𝜏0 (𝑤02 − 𝐵𝑛 ) + 𝐴 𝑛 + 𝑖 (𝜏0 𝑤0 𝐴 𝑛 + 2𝑤0 )
𝐾 (𝑎 + 𝑢∗2 )

2 𝑠 𝑚𝑢∗
2 = sign { (𝑤02 𝐴 𝑛 [−𝜏0 (𝑤02 − 𝐵𝑛 ) + 𝐴 𝑛 ]
× (𝑠 + 𝜇𝑛 𝑑2 ) − ( ) .
ℎ 𝑎 + 𝑢∗2
+ (𝑤03 − 𝐵𝑛 𝑤0 ) (𝜏0 𝑤0 𝐴 𝑛 + 2𝑤0 ))
This leads to the following theorem.
2
× ( [−𝜏0 (𝑤02 − 𝐵𝑛 ) + 𝐴 𝑛 ]
Theorem 2. If 2𝑚 − 𝑟ℎ < 0 and 𝐵𝑛2 2
− 𝐶 > 0 hold for 𝑛 =
0, 1, 2, . . ., then the interior equilibrium (𝑢∗ , V∗ ) of system (3) 2 −1
is asymptotically stable for all 𝜏1 , 𝜏2 ≥ 0. + [𝜏0 𝑤0 𝐴 𝑛 + 2𝑤0 ] ) }
4 Abstract and Applied Analysis

𝑤02 [𝐴2𝑛 + 2𝑤02 − 2𝐵𝑛 ] strictly positive for 𝑡 > 0. Because 𝑟 − (𝑚/2√𝑎ℎ) 𝐾 > 0, it
= sign { 2 2
} is possible to choose 𝜀0 satisfying
[−𝜏0 (𝑤02 − 𝐵𝑛 ) + 𝐴 𝑛 ] + [𝜏0 𝑤0 𝐴 𝑛 + 2𝑤0 ]
−1
𝑚 𝑚 𝑚 𝑟
{ 2
{ 2 2𝐵𝑛 − 𝐴2𝑛 + √(𝐴2𝑛 − 2𝐵𝑛 ) − 4 (𝐵𝑛2 − 𝐶2 ) 0 < 𝜀0 < (𝑟 − 𝐾) [ + + (1 + ℎ)] .
= sign {𝑤0 ⋅ 2√𝑎ℎ 2√𝑎ℎ 2√𝑎 𝐾
{ 2 (30)
{
2 Because
× ( [−𝜏0 (𝑤02 − 𝐵𝑛 ) + 𝐴 𝑛 ]
𝜕𝑢 𝑢 𝑚𝑢2 V (𝑡 − 𝜏1 ) 𝑢
− 𝑑1 Δ𝑢 = 𝑟𝑢 (1 − ) − ≤ 𝑟𝑢 (1 − )
} 𝜕𝑡 𝐾 𝑎 + 𝑢2 𝐾
2 −1 } (31)
+ [𝜏0 𝑤0 𝐴 𝑛 + 2𝑤0 ] ) } = 1.
}
} according to Lemma 4 and the comparison principle of
(28) parabolic equations, there exists 𝑡1 > 0 such that, for any
𝑡 > 𝑡1 , 𝑢(𝑥, 𝑡) ≤ 𝐾 + 𝜀0 ≜ 𝑢. This in turn implies that
In other words, (𝑑𝜆/𝑑𝜏)|𝜏=𝜏0 > 0.
𝜕V ℎV ℎV
− 𝑑1 ΔV = 𝑠V (1 − ) ≤ 𝑠V (1 − ) (32)
𝜕𝑡 𝑢 (𝑡 − 𝜏2 ) 𝐾 + 𝜀0
From the above discussion, the following theorem can be
stated. for 𝑡 > 𝑡1 + 𝜏2 .
Hence there exists 𝑡2 > 𝑡1 such that, for any 𝑡 > 𝑡2 ,
Theorem 3. If 𝐵𝑛2 − 𝐶2 < 0 holds, then the following statements
𝐾 + 𝜀0
are true. V (𝑥, 𝑡) ≤ + 𝜀0 ≜ V. (33)
ℎ
(1) If 0 ≤ 𝜏1 + 𝜏2 < 𝜏0 , then the interior equilibrium Consequently,
(𝑢∗ , V∗ ) of system (3) is asymptotically stable.
𝜕𝑢 𝑢 𝑚𝑢2 V (𝑡 − 𝜏1 )
(2) If 𝜏1 + 𝜏2 > 𝜏0 , then the interior equilibrium (𝑢∗ , V∗ ) of − 𝑑1 Δ𝑢 = 𝑟𝑢 (1 − ) −
system (3) is unstable. 𝜕𝑡 𝐾 𝑎 + 𝑢2

(3) System (3) undergoes a Hopf bifurcation at the interior 𝑢 𝑚𝑢2 V (𝑡 − 𝜏1 )

≥ 𝑟𝑢 (1 − )− (34)
equilibrium (𝑢∗ , V∗ ) for 𝜏1 + 𝜏2 = 𝜏𝑘 , 𝑘 = 0, 1, 2, . . .. 𝐾 2√𝑎𝑢
𝑢 𝑚 (𝐾 + (ℎ + 1) 𝜀0 )
≥ 𝑟𝑢 (1 − )− 𝑢
3. Global Stability 𝐾 2ℎ√𝑎
This section mainly proves that the interior equilibrium for 𝑡 > 𝑡2 + 𝜏1 .
(𝑢∗ , V∗ ) is globally asymptotically stable with the upper- Because 𝑟 − (𝑚/2√𝑎ℎ) 𝐾 > 0, it can be easily verified that
lower solution method in [27, 28]. For simplicity, let
(𝑢1 (𝑡, 𝑥), V1 (𝑡, 𝑥)) > (𝑢2 (𝑡, 𝑥), V2 (𝑡, 𝑥)) denote 𝑢1 (𝑡, 𝑥) > 𝑚 (𝐾 + (ℎ + 1) 𝜀0 )
𝑢2 (𝑡, 𝑥) and V1 (𝑡, 𝑥) > V2 (𝑡, 𝑥). 𝐾 (1 − ) > 0,
2𝑟ℎ√𝑎
(35)
Lemma 4 (see [29]). Assume that 𝑢(𝑡, 𝑥) is defined by 𝑚 (𝐾 + (ℎ + 1) 𝜀0 )
𝐾 (1 − ) − 𝜀0 > 0.
2𝑟ℎ√𝑎
𝑢
𝑢𝑡 − 𝑑1 Δ𝑢 = 𝑟𝑢 (1 − ), 𝑥 ∈ Ω, 𝑡 > 0,
𝐾 Hence, there exists 𝑡3 > 𝑡2 such that, for any 𝑡 > 𝑡3 ,
𝜕𝑢 (29)
= 0, 𝑥 ∈ 𝜕Ω, 𝑚 (𝐾 + (ℎ + 1) 𝜀0 )
𝜕𝑛 𝑢 (𝑥, 𝑡) ≥ 𝐾 (1 − ) − 𝜀0 ≜ 𝑢. (36)
2𝑟ℎ√𝑎
𝑢 (𝑥, 0) = 𝑢0 (𝑥) > 0, 𝑥 ∈ Ω.
This implies that
Then lim𝑡 → ∞ 𝑢(𝑥, 𝑡) = 𝐾.
𝜕V ℎV 𝑠ℎV2
− 𝑑1 ΔV = 𝑠V (1 − ) ≥ 𝑠V − (37)
Theorem 5. If 𝑟 − (𝑚/2√𝑎ℎ) 𝐾 > 0, then for system (3), the 𝜕𝑡 𝑢 (𝑡 − 𝜏2 ) 𝑢
positive equilibrium (𝑢∗ , V∗ ) is globally asymptotically stable.
for 𝑡 > 𝑡3 + 𝜏2 . Again it can be verified that
Proof. From the maximum principle of parabolic equations,
it is known that for any initial value (𝑢0 (𝑡, 𝑥), V0 (𝑡, 𝑥)) > (0, 0), 𝑢 𝐾 𝑚 (𝐾 + (ℎ + 1) 𝜀0 ) 𝜀
− 𝜀0 = (1 − ) − 0 − 𝜀0 > 0, (38)
the corresponding nonnegative solution (𝑢(𝑡, 𝑥), V(𝑡, 𝑥)) is ℎ ℎ 2𝑟ℎ√𝑎 ℎ
Abstract and Applied Analysis 5

and hence there exists 𝑡4 > 𝑡3 such that, for any 𝑡 > 𝑡4 , and there exist (̃𝑢, ̃V) > (0, 0) and (̂
𝑢, ̂V) > (0, 0) such that
lim𝑚 → ∞ 𝑢(𝑚) = 𝑢̃, lim𝑚 → ∞ V(𝑚) = ̃V, lim𝑚 → ∞ 𝑢(𝑚) = 𝑢̂,
𝐾 𝑚 (𝐾 + (ℎ + 1) 𝜀0 ) 𝜀 lim𝑚 → ∞ V(𝑚) = ̂V, and
V (𝑥, 𝑡) ≥ (1 − ) − 0 − 𝜀0 ≜ V. (39)
ℎ 2𝑟ℎ√𝑎 ℎ

Therefore, for 𝑡 > 𝑡4 , it is possible to obtain 𝑢̃ 𝑚̃𝑢̂V ℎ̃V

0 = 𝑢̃ (𝑟 (1 − )− ), 0 = ̃V (𝑠 (1 − )) ,
𝐾 𝑎 + 𝑢̃2 𝑢̃
𝑢 ≤ 𝑢 (𝑥, 𝑡) ≤ 𝑢 , V ≤ V (𝑥, 𝑡) ≤ V, (40)
𝑢̂ 𝑚̂𝑢̃V ℎ̂V
0 = 𝑢̂ (𝑟 (1 − )− ), 0 = ̂V (𝑠 (1 − )) .
and 𝑢, 𝑢, V, V satisfy 𝐾 𝑎 + 𝑢̂2 𝑢̂
(46)
𝑢 𝑚𝑢V ℎV
0 ≥ 𝑟 (1 − )− , 0 ≥ 𝑠 (1 − ),
𝐾 𝑎 + 𝑢2 𝑢
(41) Because (𝑢∗ , V∗ ) is the unique positive constant equilibrium
𝑢 𝑚𝑢V ℎV of system (3), it must hold that (̃ 𝑢, ̂V) = (𝑢∗ , V∗ ).
𝑢, ̃V) = (̂
0 ≤ 𝑟 (1 − ) − , 0 ≤ 𝑠 (1 − ) .
𝐾 𝑎 + 𝑢2 𝑢 According to the results of [27, 28], the solution
(𝑢(𝑥, 𝑡), V(𝑥, 𝑡)) of system (3) satisfies
This implies that (𝑢, V) and (𝑢, V) are a pair of coupled upper
and lower solutions of system (3), as in the definition in [29],
for the reason that (3) is a mixed quasimonotonic system. lim 𝑢 (𝑥, 𝑡) = 𝑢∗ , lim V (𝑥, 𝑡) = V∗ ,
𝑡→∞ 𝑡→∞
It is clear that there exists 𝐾 > 0 such that, for any (𝑢, V) ≤ (47)
(𝑢1 , V1 ), (𝑢2 , V2 ) ≤ (𝑢, V), uniformly for 𝑥 ∈ Ω.
󵄨󵄨 𝑚𝑢12 V1 𝑚𝑢22 V2 󵄨󵄨󵄨󵄨
󵄨󵄨 𝑢 𝑢2
󵄨󵄨𝑟𝑢1 (1 − 1 ) − − 𝑟𝑢 (1 − ) + 󵄨󵄨
󵄨󵄨 𝐾 𝑎 + 𝑢12 2
𝐾 𝑎 + 𝑢22 󵄨󵄨󵄨 Hence, the constant equilibrium (𝑢∗ , V∗ ) is globally asymp-
󵄨 (42)
󵄨 󵄨 󵄨 󵄨 totically stable.
≤ 𝐾 (󵄨󵄨󵄨𝑢1 − 𝑢2 󵄨󵄨󵄨 + 󵄨󵄨󵄨V1 − V2 󵄨󵄨󵄨) ,
󵄨󵄨 ℎV ℎV 󵄨󵄨󵄨
󵄨󵄨
󵄨󵄨𝑠V1 (1 − 1 ) − 𝑠V2 (1 − 2 )󵄨󵄨󵄨 4. Direction and Stability of Hopf Bifurcation
󵄨󵄨 𝑢1 𝑢2 󵄨󵄨 (43)
󵄨 󵄨 󵄨 󵄨 Part Two has already shown that system (3) undergoes Hopf
≤ 𝐾 (󵄨󵄨󵄨𝑢1 − 𝑢2 󵄨󵄨󵄨 + 󵄨󵄨󵄨V1 − V2 󵄨󵄨󵄨) . bifurcation at the interior equilibrium (𝑢∗ , V∗ ) for 𝜏1 + 𝜏2 =
𝜏𝑘 , 𝑘 = 0, 1, 2, . . .. In this section, the direction, stability, and
Define two iteration sequences (𝑢(𝑚) , V(𝑚) ) and (𝑢(𝑚) , V(𝑚) ) as period of the periodic solutions from the steady state will be
follows: for 𝑚 ≥ 1, studied by applying the method introduced by Hassard et al.
[30] and the center manifold theorem due to [31–35].
1 (𝑚−1)
𝑢(𝑚) = 𝑢(𝑚−1) + 𝑢 For convenience, let 𝑢󸀠 = 𝑢(𝑡 − 𝜏2 ), V󸀠 = V(𝑡), 𝜏 = 𝜏1 + 𝜏2 .
𝐾 Then system (3) is equal to a single-delay system:
𝑢(𝑚−1) 𝑚𝑢(𝑚−1) V(𝑚−1)
× (𝑟 (1 − )− 2
),
𝐾 𝑎 + 𝑢(𝑚−1) 𝜕𝑢󸀠 𝑢󸀠 𝑚𝑢󸀠2 V󸀠 (𝑡 − 𝜏)
= 𝑑1 Δ𝑢󸀠 + 𝑟𝑢 (1 − ) − ,
𝜕𝑡 𝐾 𝑎 + 𝑢󸀠2
1 (𝑚−1) ℎV(𝑚)
V(𝑚) = V(𝑚−1) + 𝑠V (1 − (𝑚) ) ,
𝐾 𝑢 𝑥 ∈ (0, 𝑙𝜋) , 𝑡 > 0,
(44)
1 (𝑚−1) 𝜕V󸀠 ℎV󸀠
𝑢(𝑚) = 𝑢(𝑚−1) + 𝑢 = 𝑑1 ΔV󸀠 + V󸀠 [𝑠 (1 − 󸀠 )] , (48)
𝐾 𝜕𝑡 𝑢
𝑢(𝑚−1) 𝑚𝑢(𝑚−1) V(𝑚−1)
× (𝑟 (1 − )− ), 𝑥 ∈ (0, 𝑙𝜋) , 𝑡 > 0,
𝐾 2
𝑎 + 𝑢(𝑚−1)
𝜕𝑢󸀠 𝜕V󸀠
1 (𝑚−1) ℎV (𝑚−1) = = 0, 𝑥 = 0, 𝑙𝜋, 𝑡 > 0.
V(𝑚) = V(𝑚−1) + 𝑠V (1 − (𝑚−1) ) , 𝜕𝑛 𝜕𝑛
𝐾 𝑢

where (𝑢(0) , V(0) ) = (𝑢, V), (𝑢(0) , V(0) ) = (𝑢, V). Let 𝑥(𝑡) = 𝑢󸀠 (𝜏𝑡) − 𝑢∗ , 𝑦(𝑡) = V󸀠 (𝜏𝑡) − V∗ , 𝜏 = 𝜏0 + 𝜇. Then
Then, for 𝑚 ≥ 1, 𝜇 = 0 is the Hopf bifurcation value of system (48), which can
be written as follows:
(𝑢, V) ≤ (𝑢(𝑚) , V(𝑚) ) ≤ (𝑢(𝑚+1) , V(𝑚+1) )
(45) 𝜕𝑈 (𝑡)
≤ (𝑢(𝑚+1) , V(𝑚+1) ) ≤ (𝑢(𝑚) , V(𝑚) ) ≤ (𝑢, V) , = 𝜏0 𝑑Δ𝑈 (𝑡) + 𝜏0 𝐿 (𝑈𝑡 ) + 𝐺 (𝑈𝑡 ) , (49)
𝜕𝑡
6 Abstract and Applied Analysis

where where 𝛿(𝜃) is a Dirac delta function satisfying

𝐿 (𝜙) 1, 𝜃 = 0,
𝛿 (𝜃) = { (55)
∗ ∗ ∗ ∗ 0, 𝜃 ≠0.
2𝑟𝑢 2𝑎𝑚𝑢 V 𝑚𝑢
[𝑟 − − ]𝜙1 (0) − 𝜙 (−1)
=( 𝐾 ∗2
(𝑎 + 𝑢 )
2 𝑎 + 𝑢∗2 2 ), For 𝜙 ∈ 𝐶1 ([−1, 0], 𝑅2 ), define 𝐴(0) as
𝑠
𝜙 (0) − 𝑠𝜙2 (0) 𝑑𝜙 (𝜃)
ℎ 1 {
{ , 𝜃 ∈ [−1, 0) ,
{
{ 𝑑𝜃
𝐺 (𝜙, 𝜇) = 𝜇𝑑Δ𝜙 (0) + 𝜇𝐿 (𝜙) + (𝜇 + 𝜏0 ) 𝐹 (𝜙) , 𝐴 (0) 𝜙 (𝜃) = { 0
{
{
{∫ 𝑑𝜂 (𝜃, 0) 𝜙 (𝜃) , 𝜃 = 0,
𝐹 (𝜙) (56)
{ −1
𝐹 (𝜙) = ( 1 ),
𝐹2 (𝜙)
0, 𝜃 ∈ [−1, 0) ,
(50) 𝑅 (0) 𝜙 (𝜃) = {
𝐹 (𝜙, 𝜇) , 𝜃 = 0.
with
For 𝜓 ∈ 𝐶1 ([−1, 0], (𝑅2 )∗ ), define
𝜙 (0) + 𝑢∗
𝐹1 (𝜙) = [𝑟 (𝜙1 (0) + 𝑢∗ ) (1 − 1 ) 𝑑𝜓 (𝑠)
𝐾 {
{ − , 𝑠 ∈ (0, 1] ,
{
{ 𝑑𝑠
2 𝐴∗ (𝜓 (𝑠)) = { 0 (57)
𝑚 (𝜙1 (0) + 𝑢∗ ) (𝜙2 (−1) + V∗ ) {
{
− ] 𝜙1 (0) {∫ 𝑑𝜂 (𝜃, 0) 𝜓 (−𝜉) , 𝑠 = 0.
𝑎 + 𝜙1 (0) + 𝑢∗2 { −1
𝑚𝑢∗ (51) Then 𝐴(0) and 𝐴∗ are adjoint operators under the bilinear
− 𝜙 (−1) , form:
𝑎 + 𝑢∗2 2
(𝜓(𝑠), 𝜙(𝜃))0 = 𝜓 (0) 𝜙 (0)
∗ℎ (𝜙2 (0) + V∗ )
𝐹2 (𝜙) = 𝑠 (𝜙2 (0) + V ) (1 − ) 0 𝜃 (58)
𝜙1 (0) + 𝑢∗
− ∫ ∫ 𝜓 (𝜉 − 𝜃) 𝑑𝜂 (0, 𝜃) 𝜙 (𝜉) 𝑑𝜉.
𝑠 −1 0
− 𝜙1 (0) + 𝑠𝜙2 (0)
ℎ Therefore, ±𝑖𝑤0 are eigenvalues of 𝐴(0) as well as 𝐴∗ .
for 𝜙 ∈ 𝐶([−1, 0], 𝑋). Next, the eigenvectors of 𝐴(0) and 𝐴∗ corresponding to the
From the previous discussion, it is known that ±𝑖𝑤0 is eigenvalues 𝑖𝑤0 and −𝑖𝑤0 can be calculated. Let
a pair of simple purely imaginary eigenvalues of the linear 1
system (7) and the following linear functional differential 𝑞 (𝜃) = ( ) 𝑒𝑖𝑤0 𝜏0 𝜃 . (59)
𝐶
equation:
Under the condition 𝐴(0)𝑞(0) = 𝑖𝑤0 𝜏0 𝑞(0), that is,
𝑑𝑧 (𝑡)
= 𝜏0 𝐿 (𝑧𝑡) . (52) 2𝑟𝑢∗ 2𝑎𝑚𝑢∗ V∗ 𝑚𝑢∗ −𝑖𝑤0 𝜏0
𝑑𝑡 𝑟− − − 𝑖𝑤 − 𝑒
2 0
By the Riesz representation theorem, there exists a 2 × 2 𝐾 (𝑎 + 𝑢∗2 ) 𝑎 + 𝑢∗2 1
𝜏0 ( )( )
matrix function 𝜂(𝜃, 𝜇) (𝜃 ∈ [−1, 0]), whose elements are 𝑠 𝐶
bounded variables, such that −𝑠 − 𝑖𝑤0
ℎ
0
(𝜏0 + 𝜇) 𝐿 (𝜙) = ∫ 𝑑𝜂 (𝜃, 𝜇) 𝜙 (𝜃) , 0
= ( ),
−1 (53) 0
2 (60)
for 𝜙 ∈ 𝐶 ([−1, 0] , 𝑅 ) .
it follows that
In fact, it is possible to choose 𝑠
𝐶= . (61)
𝜂 (𝜃, 𝜇) ℎ (𝑠 + 𝑖𝑤0 )

2𝑟𝑢∗ 2𝑎𝑚𝑢∗ V∗ Similarly, let 𝑞∗ (𝑠) = 𝐸(1 𝐷)𝑒𝑖𝑤0 𝜏0 𝑠 , and with 𝐴∗ 𝑞∗ (0) =
[𝑟 − − 2
] 0 −𝑖𝑤0 𝜏0 𝑞∗ (0), that is,
= (𝜏0 + 𝜇) (( 𝐾 (𝑎 + 𝑢∗2 ) ) 𝛿 (𝜃)
𝑠 2𝑟𝑢∗ 2𝑎𝑚𝑢∗ V∗ 𝑠
−𝑠 𝑟− − + 𝑖𝑤0
ℎ 𝐾 (𝑎 + 𝑢∗2 )
2 ℎ 1
𝜏0 ( )( )
∗ 𝐷
𝑚𝑢∗ 𝑚𝑢
− 𝑒−𝑖𝑤0 𝜏0 −𝑠 + 𝑖𝑤0 (62)
+ (0 − 𝑎 + 𝑢∗2 ) 𝛿 (𝜃 + 1)) , 𝑎 + 𝑢∗2
0 0
0
= ( ),
(54) 0
Abstract and Applied Analysis 7

it is possible to obtain for 𝜙 ∈ 𝐶([−1, 0], 𝑋). Then the center subspace of linear
2
system (7) is given by 𝑃𝐶𝑁C, where
−ℎ (𝑟 − 2𝑟𝑢∗ /𝐾 − 2𝑎𝑚𝑢∗ V∗ / (𝑎 + 𝑢∗2 ) + 𝑖𝑤0 )
𝐷= . (63) 𝑃𝐶𝑁𝜙 = Φ (Ψ, ⟨𝜙, 𝑓0 ⟩)0 ⋅ 𝑓0 , 𝜙 ∈ C,
𝑠 (71)
According to the conditions (𝑞∗ , 𝑞) = 1 and (𝑞∗ , 𝑞) = 0, 𝑃𝐶𝑁C = {(𝑞 (𝜃) 𝑧 + 𝑞 (𝜃) 𝑧) ⋅ 𝑓0 : 𝑧 ∈ 𝐶} ,

0 𝜃 and C = 𝑃𝐶𝑁C ⊕ 𝑃𝑆 C, where 𝑃𝑆 C is a stable subspace.

(𝑞∗ , 𝑞) = 𝑞∗ (0) 𝑞 (0) − ∫ ∫ 𝑞∗ (𝜉 − 𝜃) 𝑑𝜂 (0, 𝜃) 𝑞 (𝜉) 𝑑𝜉 According to [30], it is known that the infinitesimal
−1 𝜉=0
generator 𝐴 𝑈 of linear system (7) satisfies
0 𝜃
1
= 𝐸 (1 𝐷) ( ) − ∫ ∫ 𝐸 (1 𝐷) 𝑒−𝑖𝑤0 𝜏𝑘 (𝜉−𝜃) 𝑑𝜂 (𝜃) 𝐴 𝑈𝜓 = 𝜓̇
(𝜃) .
𝐶 −1 𝜉=0
(72)

1 Moreover, 𝜓 ∈ dom(𝐴 𝑈) if and only if

× ( ) 𝑒𝑖𝑤0 𝜏𝑘 𝜉 𝑑𝜉
𝐶
𝜓̇
(𝜃) ∈ C, 𝜓 (0) ∈ dom (Δ) ,
0
1 (73)
= 𝐸 {1 + 𝐷𝐶 − ∫ 𝜃 (1 𝐷) 𝑑𝜂 (0, 𝜃) ( ) 𝑒𝑖𝑤0 𝜏𝑘 𝜃 } 𝜓̇
−1 𝐶 (𝜃) (0) = 𝜏0 Δ𝜓 (0) + 𝜏0 𝐿 0 (𝜓) .

𝑚𝑢∗ Setting 𝜇 = 0 in (49), the center manifold

{
{ 0 −
= 𝐸 {1 + 𝐷𝐶 + (1 𝐷) 𝜏0 ( 𝑎 + 𝑢∗2 )
{ 𝑧2 𝑧2
{ 0 0 𝑊 (𝑧, 𝑧) = 𝑊20 (𝜃) + 𝑊11 (𝜃) 𝑧𝑧 + 𝑊02 (𝜃) + ⋅ ⋅ ⋅ (74)
2 2
1 } can be obtained in 𝑃𝑆 C. The flow of system (49) can be written
× ( ) 𝑒−𝑖𝑤0 𝜏𝑘 }
𝐶 as
}
𝑚𝑢∗ 𝑢𝑡 = Φ(𝑧 (𝑡) , 𝑧 (𝑡))𝑇 ⋅ 𝑓0 + 𝑊 (𝑧 (𝑡) , 𝑧 (𝑡)) , (75)
= 𝐸 {1 + 𝐷𝐶 + 𝜏0 𝐶 (− ) 𝑒−𝑖𝑤0 𝜏𝑘 } .
𝑎 + 𝑢∗2
(64) where

Therefore, (𝑡) = 𝑖𝑤0 𝜏0 𝑧 (𝑡) + 𝑞∗ (0)

𝑧̇

1 × ⟨𝐺 (Φ(𝑧 (𝑡) , 𝑧 (𝑡))𝑇 ⋅ 𝑓0 + 𝑊 (𝑧, 𝑧) , 0) , 𝑓0 ⟩

𝐸= . (65)
1 + 𝐷𝐶 + 𝜏0 𝐶 (−𝑚𝑢∗ / (𝑎 + 𝑢∗2 )) 𝑒𝑖𝑤0 𝜏𝑘
≜ 𝑖𝑤0 𝜏0 𝑧 (𝑡) + 𝑔 (𝑧, 𝑧) ,
Let Φ = (𝑞(𝜃), 𝑞(𝜃)), Ψ = (𝑞∗ (𝑠), 𝑞∗ (𝑠))𝑇 , 𝐼 = ( 10 01 ); then
(Ψ, Φ)0 = 𝐼. (66) (76)
Therefore, the center subspace of system (52) is 𝑃 = 𝑔 (𝑧, 𝑧) = 𝑞∗ (0)
span{𝑞(𝜃), 𝑞(𝜃)}, and the adjoint subspace is 𝑃∗ = span{𝑞∗ (𝑠),
𝑞∗ (𝑠)}. Let 𝑓0 = (𝑓01 , 𝑓02 ), where × ⟨𝐺 (Φ (𝑧 (𝑡) , 𝑧 (𝑡))𝑇 ⋅ 𝑓0 + 𝑊 (𝑧, 𝑧) , 0) , 𝑓0 ⟩

1 0 𝑧2 𝑧2
𝑓01 = ( ) , 𝑓02 = ( ) . (67) = 𝑔20 (𝜃) + 𝑔11 (𝜃) 𝑧𝑧 + 𝑔02 (𝜃)
0 1 2 2
Using the notion from [30], it is also possible to define 𝑧2 𝑧
+ 𝑔21 (𝜃) + ⋅⋅⋅ .
2
𝑐 ⋅ 𝑓0 = 𝑐1 𝑓01 + 𝑐2 𝑓02 (68) (77)
for 𝑐 = (𝑐1 , 𝑐2 )𝑇 ∈ 𝐶2 . From Taylor’s formula,
Define (𝜓 ⋅ 𝑓0 )(𝜃) = 𝜓(𝜃) ⋅ 𝑓0 for 𝜓(𝜃) ∈ [−1, 0] and
2
1 𝑙𝜋 1 𝑙𝜋 𝑚 (𝑢 + 𝑢∗ ) (V + V∗ )
⟨𝑢, V⟩ = ∫ 𝑢1 V1 𝑑𝑥 + ∫ 𝑢2 V2 𝑑𝑥 (69) 𝑓 (𝑢, V) =
𝑙𝜋 0 𝑙𝜋 0 𝑎 + (𝑢 + 𝑢∗ )2
2
for 𝑢 = (𝑢1 , 𝑢2 ), V = (V1 , V2 ) ∈ 𝐶([0, 𝑙𝜋], 𝑅2 ). 𝑚𝑢∗ V∗ 2𝑎𝑚𝑢∗ V∗ 𝑚𝑢∗2
= 2
+ 2
𝑢 + ∗2
V + 𝑏20 𝑢2
Hence, 𝑎+𝑢 ∗ ∗
(𝑎 + 𝑢 ) 𝑎 + 𝑢
𝑇
⟨𝜙, 𝑓0 ⟩ = (⟨𝜙, 𝑓01 ⟩ , ⟨𝜙, 𝑓02 ⟩) (70) + 𝑏11 𝑢V + 𝑏30 𝑢3 + 𝑏21 𝑢2 V + 𝑂 (4) ,
8 Abstract and Applied Analysis

𝑠ℎ (V + V∗ )
2 𝑟 𝑠
𝑔 (𝑢, V) = 𝑔21 = 2𝐸𝜏0 [(− − 𝑏20 ) − 𝐷 ∗ ]
𝑢 + 𝑢∗ 𝐾 ℎ𝑢

V∗ 1 2 1 1 2 1 𝑙𝜋 (1) (1)
= 𝑠ℎ ( − 𝑢 + V + 2 ∗ 𝑢2 + ∗ V2 − ∗ 𝑢V × ∫ (𝑊20 (0) + 2𝑊11 (0)) 𝑑𝑥
ℎ ℎ2 ℎ ℎ𝑢 𝑢 ℎ𝑢 𝑙𝜋 0

2 1 𝑠ℎ 1 𝑙𝜋 (1) (1)
+ 𝑢2 V − 2 ∗2 𝑢3 + 𝑂 (4) ) , − 2𝐸𝜏0 𝐷 ∫ (𝑊20 (0) 𝐶 + 2𝑊11 (0) 𝐶) 𝑑𝑥
ℎ2 𝑢∗2 ℎ𝑢 𝑢∗ 𝑙𝜋 0
(78) 1 𝑙𝜋 (1)
− 𝐸𝜏0 𝑏11 ∫ (𝑊20 (0) 𝐶𝑒𝑖𝑤0 𝜏0 + 2𝑊11
(1)
(0) 𝐶𝑒−𝑖𝑤0 𝜏0
where 𝑙𝜋 0
(2) (2)
2𝑎𝑚𝑢∗ + 𝑊20 (−1) + 2𝑊11 (−1)) 𝑑𝑥
𝑏11 = ,
(𝑎 + 𝑢∗ )2 2𝑠 1 𝑙𝜋 (1) (1)
+ 𝐸𝜏0 𝐷 ∫ (𝑊20 (0) 𝐶 + 2𝑊11 (0) 𝐶
∗
𝑎𝑚V (𝑎 + 2𝑎𝑢 2 ∗2 ∗4
+ 𝑢 ) − 4𝑎𝑚𝑢 V (𝑎 + 𝑢 ) ∗2 ∗ ∗2 𝑢∗ 𝑙𝜋 0
𝑏20 = 4
, (2) (2)
(𝑎 + 𝑢∗ 2 ) + 𝑊20 (0) + 2𝑊11 (0)) 𝑑𝑥

𝑎𝑚 (𝑎 + 𝑢∗2 ) − 4𝑎𝑚𝑢∗2 (𝑎 + 𝑢∗2 ) − 2𝐸𝜏0 𝑏21 (𝐶𝑒𝑖𝑤0 𝜏0 + 2𝐶𝑒−𝑖𝑤0 𝜏0 )

𝑏21 = 3
,
(𝑎 + 𝑢∗ 2 ) 2𝑠
− 2𝐸𝜏0 𝐷 (𝐶 + 2𝐶)
∗ ∗ ∗2 ∗ ℎ𝑢∗2
𝑏30 = [−8𝑎𝑚𝑢 V (𝑎 + 𝑢 ) − 8𝑎𝑚V
𝑠
− 6𝐸𝜏0 𝑏30 + 6𝐸𝜏0 𝐷 ∗2 .
× (3𝑢∗2 − 2𝑎𝑢∗ ) + 64𝑎𝑚𝑢∗3 V∗ ] ℎ𝑢
(81)
2 −1
× (𝑎 + 𝑢∗2 ) ((𝑎 + 𝑢∗ ) ) ,
In the last formula, 𝑊20 (0), 𝑊20 (−1), 𝑊11 (0), and 𝑊11 (−1)
4 are still unknown. Therefore, it is necessary to compute
𝑂 (4) = 𝑂 (‖(𝑢, V)‖ ) .
𝑊20 (𝜃) and 𝑊11 (𝜃), as described below.
(79) From (74), it is possible to obtain
𝑇
Let 𝐺(𝜙, 0) = 𝜏0 (𝐺1 , 𝐺2 ) , where 𝑊̇
= 𝑊20 𝑧𝑧̇ + 𝑊11 𝑧𝑧̇
+ 𝑊11 𝑧𝑧̇ + 𝑊02 𝑧𝑧̇
+ ⋅⋅⋅ ,
𝑟 𝑧2 𝑧2
(82)
𝐺1 = (− − 𝑏20 ) 𝜙12 (0) − 𝑏11 𝜙1 (0) 𝜙2 (−1) 𝐴 𝑈𝑊 = 𝐴 𝑈𝑊20 + 𝐴 𝑈𝑊11 𝑧𝑧 + 𝑊02 𝐴 𝑈 + ⋅ ⋅ ⋅ .
𝐾 2 2
− 𝑏30 𝜙13 (0) − 𝑏21 𝜙12 (0) 𝜙2 (−1) + 𝑂 (4) , In addition, 𝑊(𝑧, 𝑧) satisfies
𝑠 2 𝑠ℎ 2𝑠 (80)
𝐺2 = − 𝜙 (0) − ∗ 𝜙22 (0) + ∗ 𝜙1 (0) 𝜙2 (0) 𝑊̇
= 𝐴 𝑈𝑊 + 𝑋0 𝐺 (Φ(𝑧, 𝑧)𝑇 ⋅ 𝑓0 + 𝑊 (𝑧, 𝑧) , 0)
ℎ𝑢∗ 1 𝑢 𝑢
2𝑠 2 𝑠 − Φ (Ψ, ⟨𝑋0 𝐺 (Φ(𝑧, 𝑧)𝑇 ⋅ 𝑓0 + 𝑊 (𝑧, 𝑧) , 0) , 𝑓0 ⟩)0 ⋅ 𝑓0
− ∗2
𝜙1 (0) 𝜙2 (0) + ∗2 𝜙13 (0) + 𝑂 (4) .
ℎ𝑢 ℎ𝑢
𝑧2 𝑧2
From (77), ≜ 𝐴 𝑈𝑊 + 𝐻20 + 𝐻11 𝑧𝑧 + 𝐻02 + ⋅ ⋅ ⋅ .
2 2
𝑟 (83)
𝑔20 = 2𝐸𝜏0 [(− − 𝑏20 ) − 𝑏11 𝐶𝑒−𝑖𝑤0 𝜏0 ]
𝐾 Therefore,
𝑠 𝑠ℎ 2𝑠
+ 2𝐸𝜏0 𝐷 [− − 𝐶2 + ∗ 𝐶] , 2𝑖𝑤0 𝜏0 𝑊20 = 𝐴 𝑈𝑊20 + 𝐻20 ,
ℎ𝑢∗ 𝑢∗ 𝑢
−𝐴 𝑈𝑊11 = 𝐻11 , (84)
𝑟
𝑔11 = 𝐸𝜏0 [2 (− − 𝑏20 ) − 𝑏11 (𝐶𝑒−𝑖𝑤0 𝜏0 + 𝐶𝑒𝑖𝑤0 𝜏0 )]
𝐾 −2𝑖𝑤0 𝜏0 𝑊02 = 𝐴 𝑈𝑊02 + 𝐻02 .
2𝑠 2𝑠ℎ 2𝑠 For −1 ≤ 𝜃 < 0,
+ 𝐸𝜏0 𝐷 [− − ∗ 𝐶𝐶 + ∗ (𝐶 + 𝐶)] ,
ℎ𝑢∗ 𝑢 𝑢
𝑧2 𝑧2
𝑟 𝐻20 + 𝐻11 𝑧𝑧 + 𝐻02 + ⋅ ⋅ ⋅
𝑔02 = 2𝐸𝜏0 [(− − 𝑏20 ) − 𝑏11 𝐶𝑒𝑖𝑤0 𝜏0 ] 2 2
𝐾
𝑠 𝑠ℎ 2 2𝑠 = −Φ (Ψ, ⟨𝑋0 𝐺 (Φ(𝑧, 𝑧)𝑇 ⋅ 𝑓0 + 𝑊 (𝑧, 𝑧) , 0) , 𝑓0 ⟩)0 ⋅ 𝑓0 ,
+ 2𝐸𝜏0 𝐷 [− − 𝐶 + ∗ 𝐶] ,
ℎ𝑢∗ 𝑢∗ 𝑢 (85)
Abstract and Applied Analysis 9

and therefore It follows that

𝑟
−− 𝑏20 − 𝑏11 𝐶𝑒−𝑖𝑤0 𝜏0
𝐻20 (𝜃) = − [𝑔20 𝑞 (𝜃) + 𝑔02 𝑞 (𝜃)] ⋅ 𝑓0 , 𝐾
2𝜏0 ( 𝑠 𝑠ℎ 2𝑠 )
(86) − ∗ − ∗ 𝐶2 + ∗ 𝐶
𝐻11 (𝜃) = − [𝑔11 𝑞 (𝜃) + 𝑔11 𝑞 (𝜃)] ⋅ 𝑓0 . ℎ𝑢 𝑢 𝑢
= 𝑔20 𝑞 (0) + 𝑔02 𝑞 (0) + 2𝑖𝑤0 𝜏0 𝑊20 (0)
0
From (84) and (86), − ∫ 𝑑𝜂 (𝜃, 0) 𝑊20 (𝜃)
−1

𝑔02 𝑞 (0)
𝑊̇
20 (𝜃) = 2𝑖𝑤0 𝜏0 𝑊20 (𝜃) + [𝑔20 𝑞 (𝜃) + 𝑔02 𝑞 (𝜃)] ⋅ 𝑓0 . (87) = −𝑔20 𝑞 (0) + + 2𝑖𝑤0 𝜏0 𝐸1
3
0 𝑖𝑔20 𝑞 (𝜃) 𝑖𝑔02 𝑞 (𝜃)
Solving for 𝑊20 (𝜃), − ∫ 𝑑𝜂 (𝜃, 0) [ + + 𝐸1 𝑒2𝑖𝑤𝑝 𝜏0 𝜃 ]
−1 𝑤0 𝜏0 3𝑤0 𝜏0
0
𝑖𝑔 𝑞 (𝜃) 𝑖𝑔 𝑞 (𝜃) = (2𝑖𝑤0 𝜏0 𝐼 − ∫ 𝑑𝜂 (𝜃, 0) 𝑒2𝑖𝑤𝑝 𝜏0 𝜃 ) 𝐸1
𝑊20 (𝜃) = 20 ⋅ 𝑓0 + 02 ⋅ 𝑓0 + 𝐸1 𝑒2𝑖𝑤𝑝 𝜏0 𝜃 . (88) −1
𝑤0 𝜏0 3𝑤0 𝜏0
[ 1 0
= [2𝑖𝑤0 𝜏0 ( ) − 𝜏0
Similarly, 0 1
[
2𝑟𝑢∗ 2𝑎𝑚𝑢∗ V∗ 𝑚𝑢∗ −2𝑖𝑤0 𝜏0
−𝑖𝑔11 𝑞 (𝜃) 𝑖𝑔 𝑞 (𝜃) 𝑟− − − 𝑒 ]
𝑊11 (𝜃) = ⋅ 𝑓0 + 11 ⋅ 𝑓0 + 𝐸2 , (89) ×( 𝐾 (𝑎 + 𝑢∗2 )
2 𝑎 + 𝑢∗2 )]
𝑤0 𝜏0 𝑤0 𝜏0 𝑠 ] 𝐸1
−𝑠
ℎ ]
where 𝐸1 and 𝐸2 are both two-dimensional vectors and can 2𝑟𝑢∗ 2𝑎𝑚𝑢∗ V∗ 𝑚𝑢∗ −2𝑖𝑤0 𝜏0
2𝑖𝑤0 − 𝑟 + + 2 ∗2
𝑒
be determined as follows. = 𝜏0 ( 𝐾 (𝑎 + 𝑢∗2 ) 𝑎 + 𝑢 ) 𝐸1 .
For 𝜃 = 0, according to the definition of 𝐴 𝑈 and the first 𝑠
− 2𝑖𝑤0 + 𝑠
two equations of (84), ℎ
(93)

0 Therefore, it is possible to obtain

2𝑖𝑤0 𝜏0 𝑊20 (0) = ∫ 𝑑𝜂 (𝜃, 0) 𝑊20 (𝜃) + 𝐻20 (0)
−1 𝐸1 = 𝐸11 ⋅ 𝐸12 , (94)
(90)
0
− ∫ 𝑑𝜂 (𝜃, 0) 𝑊11 (𝜃) = 𝐻11 (0) . where
−1 −1
2𝑟𝑢∗ 2𝑎𝑚𝑢∗ V∗ 𝑚𝑢∗ −2𝑖𝑤0 𝜏0
2𝑖𝑤0 − 𝑟 + + 2 ∗2
𝑒
𝐸11 =( 𝐾 (𝑎 + 𝑢∗2 ) 𝑎 + 𝑢 ) ,
Then 𝑠
− 2𝑖𝑤0 + 𝑠
ℎ
𝐻20 (0) = − [𝑔20 𝑞 (0) + 𝑔02 𝑞 (0)] ⋅ 𝑓0 + 2𝜏0 𝑟
− − 𝑏20 − 𝑏11 𝐶𝑒−𝑖𝑤0 𝜏0
𝐸12 = 2 ( 𝑠 𝐾
𝑟 𝑠ℎ 2𝑠 ) .
− 𝑏20 − 𝑏11 𝐶𝑒−𝑖𝑤0 𝜏0
− (91) − ∗ − ∗ 𝐶2 + ∗ 𝐶
𝐾 ℎ𝑢 𝑢 𝑢
×( 𝑠 𝑠ℎ 2𝑠 ) . (95)
− ∗ − ∗ 𝐶2 + ∗ 𝐶
ℎ𝑢 𝑢 𝑢
Similarly,

Note that 𝐻11 (0) = − [𝑔11 𝑞 (0) + 𝑔11 𝑞 (0)] ⋅ 𝑓0 + 𝜏0

𝑟
2 (− − 𝑏20 ) − 𝑏11 (𝐶𝑒−𝑖𝑤0 𝜏0 + 𝐶𝑒𝑖𝑤0 𝜏0 )
0 ×( 𝐾 ).
∫ 𝑑𝜂 (𝜃, 0) 𝑒𝑖𝑤0 𝜏0 𝜃 𝑞 (0) = 𝑖𝑤0 𝜏0 𝑞 (0) , 2𝑠 2𝑠ℎ 2𝑠
− ∗ − ∗ 𝐶𝐶 + ∗ (𝐶 + 𝐶)
−1 ℎ𝑢 𝑢 𝑢
(92)
0
−𝑖𝑤0 𝜏0 𝜃
∫ 𝑑𝜂 (𝜃, 0) 𝑒 𝑞 (0) = −𝑖𝑤0 𝜏0 𝑞 (0) . (96)
−1
10 Abstract and Applied Analysis

1.5 1

0.8
1
0.6

0.4
0.5
0.2

0 0
3 3
2 100
1 100
80 2 80
60 60
1 40 1 40
20 20
0 0 0 0
(a) (b)

Figure 1: Convergence to positive equilibrium. Here the parameters are assumed to be 𝜏1 = 0, 𝜏2 = 1.5, and the initial values are 𝑢(𝑥, 𝑡) =
0.1 + 0.01𝑡 sin 𝑥, V(𝑥, 𝑡) = 0.5 + 0.01𝑡 cos 𝑥, and 𝑡 ∈ [−1.5, 0], 𝑥 ∈ [0, 𝜋].

Then Then 𝑔21 is expressed by the parameters. Therefore, the fol-

lowing values can be computed:
𝑟 󵄨 󵄨2
2 (− − 𝑏20 ) − 𝑏11 (𝐶𝑒−𝑖𝑤0 𝜏0 + 𝐶𝑒𝑖𝑤0 𝜏0 ) 𝑖 󵄨 󵄨2 󵄨󵄨𝑔 󵄨󵄨 𝑔
𝜏0 ( 𝐾
2𝑠 2𝑠ℎ 2𝑠 ) 𝑐1 (0) = (𝑔20 𝑔11 − 2 󵄨󵄨󵄨𝑔11 󵄨󵄨󵄨 − 󵄨 02 󵄨 ) + 21 ,
− ∗ − ∗ 𝐶𝐶 + ∗ (𝐶 + 𝐶) 2𝑤0 𝜏0 3 2
ℎ𝑢 𝑢 𝑢
0 Re (𝑐1 (0))
𝜇2 = − ,
= 𝑔11 𝑞 (0) + 𝑔11 𝑞 (0) − ∫ 𝑑𝜂 (𝜃, 0) 𝑊11 (𝜃) Re 𝜆󸀠 (𝜏0 ) (100)
−1

= 𝑔11 𝑞 (0) + 𝑔11 𝑞 (0) 𝛽2 = 2 Re (𝑐1 (0)) ,

0 −𝑖𝑔11 𝑞 (𝜃) 𝑖𝑔11 𝑞 (𝜃) (97) Im {𝑐1 (0)} + 𝜇2 Im {𝜆󸀠 (𝜏0 )}

− ∫ 𝑑𝜂 (𝜃, 0) [ + + 𝐸2 ] 𝑇2 = − .
−1 𝑤0 𝜏0 𝑤0 𝜏0 𝑤0 𝜏0
0 Theorem 6. If 𝜇2 , 𝛽2 , and 𝑇2 are defined as above, then the
= − ∫ 𝑑𝜂 (𝜃, 0) 𝐸2 following are true:
−1

2𝑟𝑢∗ 2𝑎𝑚𝑢∗ V∗ 𝑚𝑢∗ (i) if 𝜇2 > 0 (𝜇2 < 0), then the Hopf bifurcation is
−𝑟 + + 2 ∗2
supercritical (subcritical);
= 𝜏0 ( 𝐾 (𝑎 + 𝑢∗2 ) 𝑎 + 𝑢 ) 𝐸2 .
𝑠 (ii) if 𝛽2 < 0 (𝛽2 > 0), then the periodic solutions are stable
− 𝑠 (unstable);
ℎ
(iii) if 𝑇2 > 0 (𝑇2 < 0), then the period of the bifurcating
Therefore, it is possible to obtain periodic solutions of system (3) increases (decreases).

Remark 7. The direction and stability of the Hopf bifurcation

𝐸2 = 𝐸21 ⋅ 𝐸22 , (98) have already been computed when 𝜏 = 𝜏0 . The other
Hopf bifurcation value 𝜏𝑘 can be analyzed using the same
where procedure.

2𝑟𝑢∗ 2𝑎𝑚𝑢∗ V∗ 𝑚𝑢∗

−1 5. Simulations and Discussion
−𝑟 + + 2 ∗2
𝐸21 =( 𝐾 (𝑎 + 𝑢∗2 ) 𝑎 + 𝑢 ) , This paper has considered a delayed predator-prey model
𝑠 with modified Holling-Tanner functional response. The effect
− 𝑠
ℎ (99) of time delays on the positive equilibrium of system (3) has
𝑟 been investigated. The results obtained here show that the
2 (− − 𝑏20 ) − 𝑏11 (𝐶𝑒−𝑖𝑤0 𝜏0 + 𝐶𝑒𝑖𝑤0 𝜏0 ) time delay plays an important role in determining system
𝐸22 =( 𝐾 ).
2𝑠 2𝑠ℎ 2𝑠 stability. Theorem 3 shows that the time delay can cause a
− ∗ − ∗ 𝐶𝐶 + ∗ (𝐶 + 𝐶) stable equilibrium to become unstable or, in other words, that
ℎ𝑢 𝑢 𝑢
Abstract and Applied Analysis 11

2 2

1.5 1.5

1 1

0.5 0.5

0 0
3 3
2 150
1 1150
2
100 100
1 50 1 50
0 0 0 0
(a) (b)

Figure 2: Convergence to positive equilibrium. Here the parameters are assumed to be 𝜏1 = 0, 𝜏2 = 7, and the initial values 𝑢(𝑥, 𝑡) =
0.1 + 0.01𝑡 sin 𝑥, V(𝑥, 𝑡) = 0.5 + 0.01𝑡 cos 𝑥, and 𝑡 ∈ [−7, 0], 𝑥 ∈ [0, 𝜋].

there is a critical value 𝜏0 , such that, for 0 ≤ 𝜏1 + 𝜏2 < 𝜏0 , [2] G. Hu and W. Li, “Hopf bifurcation analysis for a delayed
equilibrium (𝑢∗ , V∗ ) is stable; when 𝜏1 + 𝜏2 approaches 𝜏0 , predator-prey system with diffusion effects,” Nonlinear Analysis:
equilibrium (𝑢∗ , V∗ ) bifurcates into periodic solutions; and Real World Applications, vol. 11, no. 2, pp. 819–826, 2010.
when 𝜏1 + 𝜏2 > 𝜏0 , the equilibrium becomes unstable. [3] H. Shi, W. Li, and G. Lin, “Positive steady states of a diffusive
Numerical simulations are performed to illustrate the predator-prey system with modified Holling-Tanner functional
dynamic behavior of system (3). Assume that 𝑟 = 1, 𝑎 = 1, response,” Nonlinear Analysis: Real World Applications, vol. 11,
ℎ = 1, 𝐾 = 2, 𝑠 = 1, 𝑚 = 4, 𝑑1 = 0.1, 𝑑2 = 0.2, and 𝑙 = 1. no. 5, pp. 3711–3721, 2010.
In this case, 𝜏0 = 2.201. As shown in Figure 1, the positive [4] S. Jana, M. Chakraborty, K. Chakraborty, and T. K. Kar, “Global
equilibrium is stable when 𝜏1 + 𝜏2 < 𝜏0 , and the predators and stability and bifurcation of time delayed prey-predator system
incorporating prey refuge,” Mathematics and Computers in
prey tend to a steady state. However, if 𝜏1 +𝜏2 > 𝜏0 , the positive
Simulation, vol. 85, pp. 57–77, 2012.
equilibrium loses its stability and Hopf bifurcation occurs,
[5] Y. Zhu, Y. Cai, S. Yan, and W. Wang, “Dynamical analysis of a
leading to oscillatory behavior. These results are identical to
delayed reaction-diffusion predator-prey system,” Abstract and
the numerical results illustrated in Figure 2. Applied Analysis, vol. 2012, Article ID 323186, 23 pages, 2012.
In both figures, the upper diagram represents the prey and
[6] M. Zhao and S. Lv, “Chaos in a three-species food chain model
the lower diagram represents the predator. with a Beddington-DeAngelis functional response,” Chaos,
Solitons and Fractals, vol. 40, no. 5, pp. 2305–2316, 2009.
[7] M. Zhao, H. Yu, and J. Zhu, “Effects of a population floor on
Conflict of Interests the persistence of chaos in a mutual interference host-parasitoid
The authors declare that there is no conflict of interests model,” Chaos, Solitons and Fractals, vol. 42, no. 2, pp. 1245–
1250, 2009.
regarding the publication of this paper.
[8] S. Lv and M. Zhao, “The dynamic complexity of a three species
food chain model,” Chaos, Solitons and Fractals, vol. 37, no. 5,
pp. 1469–1480, 2008.
Acknowledgments
[9] L. Zhu and M. Zhao, “Dynamic complexity of a host-parasitoid
This work was supported by the Key Program of Zhejiang ecological model with the Hassell growth function for the host,”
Provincial Natural Science Foundation of China (Grant Chaos, Solitons and Fractals, vol. 39, no. 3, pp. 1259–1269, 2009.
no. LZ12C03001), by the National Natural Science Founda- [10] L. Zhang and M. Zhao, “Dynamic complexities in a hyperpar-
tion of China (Grant no. 31370381), by the National Key asitic system with prolonged diapause for host,” Chaos, Solitons
Basic Research Program of China (973 Program, Grant and Fractals, vol. 42, no. 2, pp. 1136–1142, 2009.
no. 2012CB426510), and by the Zhejiang Provincial Natural [11] H. Yu, M. Zhao, S. Lv, and L. Zhu, “Dynamic complexities in
Science Foundation of China (Grant no. LY13A010010). a parasitoid-host-parasitoid ecological model,” Chaos, Solitons
and Fractals, vol. 39, no. 1, pp. 39–48, 2009.
[12] C. Dai, M. Zhao, and L. Chen, “Complex dynamic behavior
References of three-species ecological model with impulse perturbations
and seasonal disturbances,” Mathematics and Computers in
[1] S. Yuan and Y. Song, “Stability and Hopf bifurcations in a Simulation, vol. 84, pp. 83–97, 2012.
delayed Leslie-Gower predator-prey system,” Journal of Math- [13] D. J. Wollkind and J. A. Logan, “Temperature-dependent
ematical Analysis and Applications, vol. 355, no. 1, pp. 82–100, predator-prey mite ecosystem on apple tree foliage,” Journal of
2009. Mathematical Biology, vol. 6, no. 3, pp. 265–283, 1978.
12 Abstract and Applied Analysis

[14] D. J. Wollkind, J. B. Collings, and J. A. Logan, “Metastability in [32] X. D. Lin, J. W. So, and J. H. Wu, “Centre manifolds for partial
a temperature-dependent model system for predator-prey mite differential equations with delays,” Proceedings of the Royal
outbreak interactions on fruit trees,” Bulletin of Mathematical Society of Edinburgh A: Mathematics, vol. 122, no. 3-4, pp. 237–
Biology, vol. 50, no. 4, pp. 379–409, 1988. 254, 1992.
[15] R. M. May, Stability and Complexity in Model Ecosystems, [33] T. Faria, “Normal forms and Hopf bifurcation for partial
Princeton University Press, Princeton, NJ, USA, 1978. differential equations with delays,” Transactions of the American
[16] Y. Li and C. Li, “Stability and Hopf bifurcation analysis on Mathematical Society, vol. 352, no. 5, pp. 2217–2238, 2000.
a delayed Leslie-Gower predator-prey system incorporating a [34] T. Faria and L. T. Magalhães, “Normal forms for retarded
prey refuge,” Applied Mathematics and Computation, vol. 219, functional-differential equations with parameters and applica-
no. 9, pp. 4576–4589, 2013. tions to Hopf bifurcation,” Journal of Differential Equations, vol.
[17] C. Zhang, X. Yan, and G. Cui, “Hopf bifurcations in a predator- 122, no. 2, pp. 181–200, 1995.
prey system with a discrete delay and a distributed delay,” [35] T. Faria and L. T. Magalhaes, “Normal forms for retarded
Nonlinear Analysis: Real World Applications, vol. 11, no. 5, pp. functional-differential equations and applications to Bogdanov-
4141–4153, 2010. Takens singularity,” Journal of Differential Equations, vol. 122,
[18] F. Lian and Y. Xu, “Hopf bifurcation analysis of a predator- no. 2, pp. 201–224, 1995.
prey system with Holling type IV functional response and time
delay,” Applied Mathematics and Computation, vol. 215, no. 4,
pp. 1484–1495, 2009.
[19] Y. Chen and F. Zhang, “Dynamics of a delayed predator-
prey model with predator migration,” Applied Mathematical
Modelling, vol. 37, no. 3, pp. 1400–1412, 2013.
[20] W. Zuo and J. Wei, “Stability and Hopf bifurcation in a diffusive
predatory-prey system with delay effect,” Nonlinear Analysis:
Real World Applications, vol. 12, no. 4, pp. 1998–2011, 2011.
[21] X. Lian, S. Yan, and H. Wang, “Pattern formation in predator-
prey model with delay and cross diffusion,” Abstract and Applied
Analysis, Article ID 147232, 10 pages, 2013.
[22] M. Zhao, X. Wang, H. Yu, and J. Zhu, “Dynamics of an ecologi-
cal model with impulsive control strategy and distributed time
delay,” Mathematics and Computers in Simulation, vol. 82, no. 8,
pp. 1432–1444, 2012.
[23] H. Yu, M. Zhao, and R. P. Agarwal, “Stability and dynamics
analysis of time delayed eutrophication ecological model based
upon the Zeya reservoir,” Mathematics and Computers in Simu-
lation, vol. 97, pp. 53–67, 2014.
[24] F. Wei and Y. Cai, “Global asymptotic stability of stochastic
nonautonomous Lotka-Volterra models with infinite delay,”
Abstract and Applied Analysis, vol. 2013, Article ID 351676, 8
pages, 2013.
[25] Y. Zhai, H. Bai, Y. Xiong, and X. Ma, “Hopf bifurcation analysis
for the modified Rayleigh price model with time delay,” Abstract
and Applied Analysis, vol. 2013, Article ID 290497, 6 pages, 2013.
[26] R. Shi, J. Qi, and S. Tang, “Stability and bifurcation analysis
for a predator-prey model with discrete and distributed delay,”
Abstract and Applied Analysis, vol. 2013, Article ID 454097, 12
pages, 2013.
[27] C. V. Pao, “Dynamics of nonlinear parabolic systems with time
delays,” Journal of Mathematical Analysis and Applications, vol.
198, no. 3, pp. 751–779, 1996.
[28] C. V. Pao, “Convergence of solutions of reaction-diffusion
systems with time delays,” Nonlinear Analysis: Theory, Methods
& Applications, vol. 48, no. 3, pp. 349–362, 2002.
[29] Q. X. Ye and Z. Y. Li, Introduction to Reaction-Diffusion
Equations, Science Press, Beijing, China, 1990.
[30] B. D. Hassard, N. D. Kazarinoff, and Y. Wan, Theory and Applica-
tions of Hopf bifurcation, vol. 41 of London Mathematical Society
Lecture Note Series, Cambridge University Press, Cambridge,
UK, 1981.
[31] J. Wu, Theory and Applications of Partial Functional-Differential
Equations, vol. 119 of Applied Mathematical Sciences, Springer,
New York, NY, USA, 1996.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 583930, 13 pages
http://dx.doi.org/10.1155/2014/583930

Research Article
The Stochastic Θ-Method for Nonlinear Stochastic Volterra
Integro-Differential Equations

Peng Hu1 and Chengming Huang2

1
School of Mathematics and Physics, China University of Geosciences, Wuhan 430074, China
2
School of Mathematics and Statistics, Huazhong University of Science and Technology, Wuhan 430074, China

Correspondence should be addressed to Peng Hu; hupeng1118@126.com

Received 6 May 2014; Accepted 28 August 2014; Published 27 October 2014

Academic Editor: Sanling Yuan

Copyright © 2014 P. Hu and C. Huang. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

The stochastic Θ-method is extended to solve nonlinear stochastic Volterra integro-differential equations. The mean-square
convergence and asymptotic stability of the method are studied. First, we prove that the stochastic Θ-method is convergent of
order 1/2 in mean-square sense for such equations. Then, a sufficient condition for mean-square exponential stability of the true
solution is given. Under this condition, it is shown that the stochastic Θ-method is mean-square asymptotically stable for every
stepsize if 1/2 ≤ 𝜃 ≤ 1 and when 0 ≤ 𝜃 < 1/2, the stochastic Θ-method is mean-square asymptotically stable for some small
stepsizes. Finally, we validate our conclusions by numerical experiments.

1. Introduction In the last decades, VIDEs have received a great deal of atten-
tion. A well-known example of this type is the Volterra pop-
In this paper we study the numerical solution of the 𝑑- ulation equation
dimensional nonlinear stochastic Volterra integro-differ-
ential equation (SVIDE) with convolution kernels 𝑡
𝑥󸀠 (𝑡) = 𝑎𝑥 (𝑡) − 𝑏𝑥2 (𝑡) + 𝑥 (𝑡) ∫ 𝐺 (𝑡 − 𝑠) 𝑥 (𝑠) 𝑑𝑠. (4)
𝑡 0
𝑑𝑥 (𝑡) = 𝑓 (𝑥 (𝑡) , ∫ 𝐺 (𝑡 − 𝑠) 𝑥 (𝑠) 𝑑𝑠) 𝑑𝑡
0 For a general theory of VIDEs, we refer the reader to the
(1) classical book by Burton [1]. Also, many efficient numerical
𝑡
+ 𝑔 (𝑥 (𝑡) , ∫ 𝐻 (𝑡 − 𝑠) 𝑥 (𝑠) 𝑑𝑠) 𝑑𝑊 (𝑡) , methods such as linear multistep methods, Runge-Kutta
0 methods, and collocation methods have been constructed for
VIDEs (see [2–4] and the extensive bibliography therein).
with initial data 𝑥(0) = 𝑥0 . Here 𝑊(𝑡) is a scalar Brownian However, many real-world phenomena are subject to
motion, and 𝑓, 𝑔, 𝐺, and 𝐻 are given functions which map some random environmental effects. In recent years, the
study of stability for SVIDEs has attracted the attention
𝑓: 𝑅𝑑 × 𝑅𝑑 󳨀→ 𝑅𝑑 , 𝑔: 𝑅𝑑 × 𝑅𝑑 󳨀→ 𝑅𝑑 , of many authors. For example, in 2000, Mao [5] studied
(2) the stability of the stochastic integro-differential equation as
𝐺: 𝑅 󳨀→ 𝑅𝑑×𝑑 , 𝐻: 𝑅 󳨀→ 𝑅𝑑×𝑑 . follows:
𝑡
In particular, (1) can be regarded as a stochastically per- 𝑑𝑥 (𝑡) = 𝑓 (𝑥 (𝑡) , 𝑡) + 𝑔 (∫ 𝐺 (𝑡 − 𝑠) 𝑥 (𝑠) 𝑑𝑠, 𝑡) 𝑑𝑊 (𝑡) ,
turbed problem of the deterministic nonlinear Volterra inte- 0
gro-differential equation (VIDE) (5)
𝑡 and several criteria on mean-square exponential stability and
𝑥󸀠 (𝑡) = 𝑓 (𝑥 (𝑡) , ∫ 𝐺 (𝑡 − 𝑠) 𝑥 (𝑠)𝑑𝑠) . (3)
𝐿2 -stability have been obtained. Later, Mao and Riedle [6]
0
2 Abstract and Applied Analysis

extended these results to a more general type of equation. product in space 𝑅𝑑 and | ⋅ | be the corresponding norm. The
Appleby [7] studied the almost sure asymptotic stability and matrix norm is denoted by ‖ ⋅ ‖, and 𝐴𝑇 denotes the transpose
integrability of the zero solution of a finite dimensional Itô- of a matrix 𝐴. 𝐸(⋅) stands for the mathematical expectation
Volterra equation. operator. For any vectors 𝑥, 𝑦 ∈ 𝑅𝑑 , we denote ⟨𝑥, 𝑦⟩ = 𝑥 ⋅ 𝑦.
Since the majority of stochastic differential equations For the mean-square convergence analysis, we consider
cannot be solved analytically, numerical methods are adopted the autonomous 𝑑-dimensional nonlinear SVIDE on [0, 𝑇]
for obtaining approximate solutions. In recent years, a large where 𝑇 < ∞
number of numerical methods are constructed for stochastic 𝑡
ordinary differential equations (SODEs), such as stochastic 𝑑𝑥 (𝑡) = 𝑓 (𝑥 (𝑡) , ∫ 𝐺 (𝑡 − 𝑠) 𝑥 (𝑠) 𝑑𝑠) 𝑑𝑡
linear multistep methods, stochastic Runge-Kutta methods, 0
and stochastic Taylor methods. For more details, we refer (6)
𝑡
the reader to the papers [8–11], the Ph.D. thesis by Burrage + 𝑔 (𝑥 (𝑡) , ∫ 𝐻 (𝑡 − 𝑠) 𝑥 (𝑠) 𝑑𝑠) 𝑑𝑊 (𝑡) ,
[12], and the book by Kloeden and Platen [13]. But for 0
SVIDEs, up to now, there are only few results about the with initial data 𝑥(0) = 𝑥0 . We assume that the functions 𝑓
numerical approximation in the existing literatures. In [14, and 𝑔 are globally Lipschitz continuous in all variables and
15], Golec and Sathananthan studied the convergence of satisfy linear growth condition, that is, there exist positive
Euler-Maruyama scheme for SVIDEs under some different
constants 𝐿 and 𝐾, such that for all 𝑥, 𝑥, 𝑦, 𝑦 ∈ 𝑅𝑑 ,
assumptions. And in [16], the analytical asymptotic stability
and numerical asymptotic stability have been considered for 󵄨󵄨 󵄨2 󵄨 󵄨2
󵄨󵄨𝑓 (𝑥, 𝑦) − 𝑓 (𝑥, 𝑦)󵄨󵄨󵄨 ∨ 󵄨󵄨󵄨𝑔 (𝑥, 𝑦) − 𝑔 (𝑥, 𝑦)󵄨󵄨󵄨
a class of linear stochastic functional differential equation. (7)
󵄨 󵄨2
The areas of the regions of asymptotic stability for some ≤ 𝐿 (|𝑥 − 𝑥|2 + 󵄨󵄨󵄨𝑦 − 𝑦󵄨󵄨󵄨 ) ;
numerical methods are derived. However we have not found
any stability results for numerical methods for SVIDEs (1). 󵄨󵄨 󵄨2 󵄨 󵄨2 󵄨 󵄨2
󵄨󵄨𝑓 (𝑥, 𝑦)󵄨󵄨󵄨 ∨ 󵄨󵄨󵄨𝑔 (𝑥, 𝑦)󵄨󵄨󵄨 ≤ 𝐾 (1 + |𝑥| + 󵄨󵄨󵄨𝑦󵄨󵄨󵄨 ) ,
2
(8)
Stochastic Θ-method is a widely used method with strong
convergence order 1/2 for SODEs. If Θ = 0, it is the where 𝑎 ∨ 𝑏 denotes the maximum of 𝑎 and 𝑏. Also, we assume
classical Euler-Maruyama method, and it is the Backward that the functions 𝐺 and 𝐻 are continuous and satisfy
Euler method when Θ = 1. Similar to deterministic 𝐴-
stability, Higham [17, 18] introduced stochastic 𝐴-stability ̂ |𝑡 − 𝑠| ,
‖𝐺 (𝑡) − 𝐺 (𝑠)‖ ∨ ‖𝐻 (𝑡) − 𝐻 (𝑠)‖ ≤ 𝐿 (9)
which means that the problem stability implies the numerical
method stability for any stepsize, and it is shown that the for all 𝑡, 𝑠 ∈ [0, 𝑇], which implies
stochastic Θ-method is stochastic 𝐴-stable if 1/2 ≤ 𝜃 ≤ 1. ‖𝐺 (𝑡)‖ ∨ ‖𝐻 (𝑡)‖ ≤ 𝑀 < ∞, (10)
In this paper, we apply the stochastic Θ-method to SVIDE
(1). The mean-square convergence and stability of the method where 𝑀 and 𝐿 ̂ are positive constants. The above conditions
are considered. It is proved that the stochastic Θ-method (7)–(9) can guarantee that SVIDE (6) has a unique solution
for SVIDE has mean-square convergence order 1/2. For the 𝑥(𝑡).
numerical stability analysis, first we extend the stability result The following lemma will be used in our convergence
in [5] to the case of SVIDE (1), and a sufficient condition analysis.
for the mean-square exponential stability of the true solution
is obtained. Under this condition, we then show that the Lemma 1. Under conditions (8)-(9), the solution of SVIDE (6)
stochastic Θ-method is mean-square asymptotically stable has the property
for any stepsize if 1/2 ≤ 𝜃 ≤ 1, and when 0 ≤ 𝜃 < 1/2, the
stochastic Θ-method is mean-square asymptotically stable
𝐸 ( sup |𝑥 (𝑡)|2 ) ≤ 𝐶1 , (11)
for some small stepsizes. 0≤𝑡≤𝑇
This paper is organized as follows. In Section 2, we
study the mean-square convergence of the stochastic Θ- with
method for SVIDEs. In Section 3, we investigate the mean- 1 󵄨 󵄨2
square stability of the method. In Section 4, some numerical 𝐶1 := ( + 3𝐸󵄨󵄨󵄨𝑥0 󵄨󵄨󵄨 ) 𝑒(6𝑇+24)𝐾𝐾𝑇 , (12)
2
experiments are given to validate our conclusions.
where
𝐾 = max {1, 𝑇2 𝑀2 } . (13)
2. Mean-Square Convergence of the
Stochastic Θ-Method Moreover, for any 0 ≤ 𝑟 < 𝑡 ≤ 𝑇,
Throughout this paper, we use the following notations. Let 𝐸|𝑥 (𝑡) − 𝑥 (𝑟)|2 ≤ 𝐶2 (𝑡 − 𝑟) , (14)
(Ω, F, {F𝑡 }𝑡≥0 , 𝑃) be a complete probability space with a fil-
tration {F𝑡 }𝑡≥0 satisfying normal conditions, namely, {F𝑡 }𝑡≥0 where 𝐶2 = 2𝐾𝐾(𝑇 + 1)(1 + 2𝐶1 ).
is increasing and right continuous with F0 containing all 𝑃-
null sets. And let 𝑊(𝑡) be a scalar Brownian motion defined Proof. The proof of this lemma is similar to that of
on the probability space. Let ⟨⋅, ⋅⟩ be the Euclidean inner Lemma 5.5.2 in [19], and that of Theorem 2.3 in [20].
Abstract and Applied Analysis 3

𝑛−1 𝑡𝑗+1
Now we apply the stochastic Θ-method to SVIDE (6) and 󵄩󵄩 󵄩2
≤ 2𝑇 ∑ 𝐸 ∫ 󵄩󵄩𝐺 (𝑡𝑛 − 𝑠) − 𝐺 (𝑡𝑛 − 𝑡𝑗 )󵄩󵄩󵄩 |𝑥 (𝑠)|2 𝑑𝑠
obtain the following numerical scheme: 𝑡𝑗
󵄩 󵄩
𝑗=0

𝑥𝑛+1 𝑛−1 𝑡𝑗+1

󵄩󵄩 󵄩2 󵄨 󵄨2
+ 2𝑇 ∑ 𝐸 ∫ 󵄩󵄩𝐺 (𝑡𝑛 − 𝑡𝑗 )󵄩󵄩󵄩 󵄨󵄨󵄨𝑥 (𝑠) − 𝑥 (𝑡𝑗 )󵄨󵄨󵄨 𝑑𝑠.
= 𝑥𝑛 + ℎ [𝜃𝑓 (𝑥𝑛+1 , 𝑧𝑛+1 ) + (1 − 𝜃) 𝑓 (𝑥𝑛 , 𝑧𝑛 )] (15) 𝑡𝑗
󵄩 󵄩󵄨 󵄨
𝑗=0

+ 𝑔 (𝑥𝑛 , 𝑧𝑛 ) Δ𝑊𝑛 , (21)

where ℎ > 0 is a given stepsize, and 𝑥𝑛 is an approximation Therefore, by (9) and Lemma 1, we obtain
to 𝑥(𝑡𝑛 ) with 𝑡𝑛 = 𝑛ℎ, 𝑛 = 1, . . . , 𝑁 = 𝑇/ℎ. The increment 󵄨󵄨 𝑡𝑛
󵄨 󵄨󵄨󵄨2
Δ𝑊𝑛 = 𝑊(𝑡𝑛+1 ) − 𝑊(𝑡𝑛 ) is an 𝑁(0, ℎ)-distributed Gaussian 𝐸󵄨󵄨󵄨∫ 𝐺 (𝑡𝑛 − 𝑠) 𝑥 (𝑠) 𝑑𝑠 − 𝑧 (𝑡𝑛 )󵄨󵄨󵄨
random variable. And the arguments 𝑧𝑛 and 𝑧𝑛 denote 󵄨󵄨 0 󵄨󵄨
approximations to 𝑛−1 𝑡𝑗+1 2
𝑡𝑛 𝑡𝑛
̂∑ ∫
≤ 2𝑇𝐶1 𝐿 (𝑠 − 𝑡𝑗 ) 𝑑𝑠
∫ 𝐺 (𝑡𝑛 − 𝑠) 𝑥 (𝑠) 𝑑𝑠, ∫ 𝐻 (𝑡𝑛 − 𝑠) 𝑥 (𝑠) 𝑑𝑠. (16) 𝑗=0 𝑡𝑗
0 0 (22)
𝑛−1 𝑡𝑗+1
In this paper, we choose the composite left rectangular for- + 2𝑇𝑀2 𝐶2 ∑ ∫ (𝑠 − 𝑡𝑗 ) 𝑑𝑠
mula to discretize the integral terms and obtain the follow- 𝑗=0 𝑡𝑗
ing schemes
̂ 2
2𝑇2 𝐶1 𝐿
𝑛−1 𝑛−1 ≤ ℎ + 𝑇2 𝑀2 𝐶2 ℎ.
3
𝑧𝑛 = ℎ ∑ 𝐺 (𝑡𝑛 − 𝑡𝑗 ) 𝑥𝑗 , 𝑧𝑛 = ℎ ∑ 𝐻 (𝑡𝑛 − 𝑡𝑗 ) 𝑥𝑗 . (17)
𝑗=0 𝑗=0 Since 0 < ℎ < 1, thus, there exists a constant 𝐶3 which does
not depend on ℎ, such that (19) holds. Similar to the above
Moreover we denote that proof, (20) also holds.
𝑛−1
For the convergence analysis, the local truncation error is
𝑧 (𝑡𝑛 ) = ℎ ∑ 𝐺 (𝑡𝑛 − 𝑡𝑗 ) 𝑥 (𝑡𝑗 ) ,
defined by
𝑗=0
(18) 𝛿𝑛+1 = 𝑥 (𝑡𝑛+1 )
𝑛−1
𝑧 (𝑡𝑛 ) = ℎ ∑ 𝐻 (𝑡𝑛 − 𝑡𝑗 ) 𝑥 (𝑡𝑗 ) .
𝑗=0
− {𝑥 (𝑡𝑛 ) + ℎ [𝜃𝑓 (𝑥 (𝑡𝑛+1 ) , 𝑧 (𝑡𝑛+1 ))
(23)
In this section, without loss of generality, we assume 0 < ℎ < + (1 − 𝜃) 𝑓 (𝑥 (𝑡𝑛 ) , 𝑧 (𝑡𝑛 ))]
1.
+𝑔 (𝑥 (𝑡𝑛 ) , 𝑧 (𝑡𝑛 )) Δ𝑊𝑛 } ,
Lemma 2. Under conditions (8)-(9), there exist positive con- and the global error is defined by
stants 𝐶3 , 𝐶4 such that
𝜖𝑛 = 𝑥 (𝑡𝑛 ) − 𝑥𝑛 . (24)
󵄨󵄨 𝑡𝑛 󵄨󵄨2
󵄨 󵄨
𝐸󵄨󵄨󵄨∫ 𝐺 (𝑡𝑛 − 𝑠) 𝑥 (𝑠) 𝑑𝑠 − 𝑧 (𝑡𝑛 )󵄨󵄨󵄨 ≤ 𝐶3 ℎ, (19) It is obvious that 𝜖𝑛 is F𝑡𝑛 -measurable since both 𝑥(𝑡𝑛 ) and
󵄨󵄨 0 󵄨󵄨
𝑥𝑛 are F𝑡𝑛 -measurable random variables.
󵄨󵄨 𝑡𝑛 󵄨󵄨2
󵄨 󵄨
𝐸󵄨󵄨󵄨∫ 𝐻 (𝑡𝑛 − 𝑠) 𝑥 (𝑠) 𝑑𝑠 − 𝑧 (𝑡𝑛 )󵄨󵄨󵄨 ≤ 𝐶4 ℎ. (20) Definition 3. The stochastic Θ-method (15)–(17) is said to be
󵄨󵄨 0 󵄨󵄨 consistent with order 𝑝1 in the mean and with order 𝑝2 in
Proof. By (18), we have the mean-square sense if the following estimations hold with
𝑝2 ≥ 1/2 and 𝑝1 ≥ 𝑝2 + (1/2):
󵄨󵄨 𝑡𝑛 󵄨󵄨2
󵄨 󵄨
𝐸󵄨󵄨󵄨∫ 𝐺 (𝑡𝑛 − 𝑠) 𝑥 (𝑠) 𝑑𝑠 − 𝑧 (𝑡𝑛 )󵄨󵄨󵄨 󵄨 󵄨2 1/2
󵄨󵄨 0 󵄨󵄨 max (𝐸 (󵄨󵄨󵄨󵄨𝐸 (𝛿𝑛+1 | F𝑡𝑛 )󵄨󵄨󵄨󵄨 )) ≤ 𝐶ℎ𝑝1 , as ℎ 󳨀→ 0,
0≤𝑛≤𝑁−1
󵄨󵄨 𝑡 󵄨󵄨2
󵄨󵄨 𝑛 𝑛−1 󵄨󵄨 󵄨 󵄨2 1/2
= 𝐸󵄨󵄨󵄨∫ 𝐺 (𝑡𝑛 − 𝑠) 𝑥 (𝑠) 𝑑𝑠 − ℎ ∑ 𝐺 (𝑡𝑛 − 𝑡𝑗 ) 𝑥 (𝑡𝑗 )󵄨󵄨󵄨󵄨
󵄨 max (𝐸󵄨󵄨󵄨𝛿𝑛+1 󵄨󵄨󵄨 ) ≤ 𝐶ℎ𝑝2 , as ℎ 󳨀→ 0,
󵄨󵄨 0 󵄨󵄨 0≤𝑛≤𝑁−1
󵄨 𝑗=0 󵄨 (25)
󵄨󵄨𝑛−1 𝑡 󵄨󵄨2
󵄨󵄨 𝑗+1 󵄨󵄨 where the constant 𝐶 does not depend on ℎ but may depend
= 𝐸󵄨󵄨󵄨󵄨 ∑ ∫ [𝐺 (𝑡𝑛 − 𝑠) 𝑥 (𝑠) − 𝐺 (𝑡𝑛 − 𝑡𝑗 ) 𝑥 (𝑡𝑗 )] 𝑑𝑠 󵄨󵄨
󵄨󵄨 on 𝑇 and the initial data.
󵄨󵄨𝑗=0 𝑡𝑗 󵄨󵄨
󵄨 󵄨
𝑛−1 󵄨󵄨 𝑡𝑗+1 󵄨󵄨2 Lemma 4. Under conditions (7)–(9), the stochastic Θ-method
󵄨 󵄨󵄨
≤ 𝑛 ∑ 𝐸󵄨󵄨󵄨󵄨∫ [𝐺 (𝑡𝑛 − 𝑠) 𝑥 (𝑠) − 𝐺 (𝑡𝑛 − 𝑡𝑗 ) 𝑥 (𝑡𝑗 )] 𝑑𝑠 󵄨󵄨
󵄨󵄨
(15)–(17) is consistent with order 𝑝1 = 3/2 in the mean and
𝑗=0 󵄨󵄨 𝑡𝑗 󵄨 with order 𝑝2 = 1 in the mean-square sense.
4 Abstract and Applied Analysis

Proof. Note that By Hölder inequality and condition (7), we obtain

𝑥 (𝑡𝑛+1 ) − 𝑥 (𝑡𝑛 ) 𝑡𝑛 +(1−𝜃)ℎ 󵄨󵄨 𝑠

󵄨
𝐼1 ≤ (1 − 𝜃) ℎ ∫ 𝐸 󵄨󵄨󵄨𝑓 (𝑥 (𝑠) , ∫ 𝐺 (𝑠 − 𝑟) 𝑥 (𝑟) 𝑑𝑟)
𝑡𝑛+1 𝑠 𝑡𝑛 󵄨󵄨 0
=∫ 𝑓 (𝑥 (𝑠) , ∫ 𝐺 (𝑠 − 𝑟) 𝑥 (𝑟) 𝑑𝑟) 𝑑𝑠
𝑡𝑛 0 (26) 󵄨󵄨2
󵄨
−𝑓 (𝑥 (𝑡𝑛 ) , 𝑧 (𝑡𝑛 )) 󵄨󵄨󵄨 𝑑𝑠
𝑡𝑛+1 𝑠 󵄨󵄨
+∫ 𝑔 (𝑥 (𝑠) , ∫ 𝐻 (𝑠 − 𝑟) 𝑥 (𝑟) 𝑑𝑟) 𝑑𝑊 (𝑠) ,
𝑡𝑛 0 ≤ (1 − 𝜃) ℎ𝐿
then we have 𝑡𝑛 +(1−𝜃)ℎ
󵄨 󵄨2
×∫ (𝐸󵄨󵄨󵄨𝑥 (𝑠) − 𝑥 (𝑡𝑛 )󵄨󵄨󵄨
󵄨 󵄨2
𝐸 (󵄨󵄨󵄨󵄨𝐸 (𝛿𝑛+1 | F𝑡𝑛 )󵄨󵄨󵄨󵄨 ) 𝑡𝑛

󵄨󵄨 𝑠 󵄨󵄨2
󵄨󵄨 󵄨 󵄨
󵄨 𝑡𝑛+1 𝑠 + 𝐸󵄨󵄨󵄨∫ 𝐺 (𝑠 − 𝑟) 𝑥 (𝑟) 𝑑𝑟 − 𝑧 (𝑡𝑛 )󵄨󵄨󵄨 ) 𝑑𝑠.
= 𝐸 ( 󵄨󵄨󵄨󵄨𝐸 (∫ 𝑓 (𝑥 (𝑠) , ∫ 𝐺 (𝑠 − 𝑟) 𝑥 (𝑟) 𝑑𝑟) 𝑑𝑠 󵄨󵄨 0 󵄨󵄨
󵄨󵄨 𝑡𝑛 0
(29)
− ℎ (𝜃𝑓 (𝑥 (𝑡𝑛+1 ) , 𝑧 (𝑡𝑛+1 ))
By (9)–(11) and Lemma 2, we get
󵄨󵄨2
󵄨
+ (1 − 𝜃) 𝑓 (𝑥 (𝑡𝑛 ) , 𝑧 (𝑡𝑛 ))) | F𝑡𝑛 )󵄨󵄨󵄨 ) 󵄨󵄨 𝑠 󵄨󵄨2
󵄨󵄨 󵄨 󵄨
𝐸󵄨󵄨󵄨∫ 𝐺 (𝑠 − 𝑟) 𝑥 (𝑟) 𝑑𝑟 − 𝑧 (𝑡𝑛 )󵄨󵄨󵄨
󵄨󵄨 0 󵄨󵄨
󵄨󵄨 𝑡𝑛+1 𝑠
󵄨
≤ 𝐸 ( ( 󵄨󵄨󵄨󵄨∫ 𝑓 (𝑥 (𝑠) , ∫ 𝐺 (𝑠 − 𝑟) 𝑥 (𝑟) 𝑑𝑟) 𝑑𝑠 󵄨󵄨 𝑠
󵄨
󵄨󵄨 𝑡𝑛 0 ≤ 2𝐸 󵄨󵄨󵄨∫ (𝐺 (𝑠 − 𝑟)) 𝑥 (𝑟) 𝑑𝑟
󵄨󵄨 0
− ℎ (𝜃𝑓 (𝑥 (𝑡𝑛+1 ) , 𝑧 (𝑡𝑛+1 )) 𝑡𝑛 󵄨󵄨2
󵄨
− ∫ (𝐺 (𝑡𝑛 − 𝑟)) 𝑥 (𝑟) 𝑑𝑟󵄨󵄨󵄨
󵄨󵄨2 0 󵄨󵄨
󵄨
+ (1 − 𝜃) 𝑓 (𝑥 (𝑡𝑛 ) , 𝑧 (𝑡𝑛 ))) 󵄨󵄨󵄨 | F𝑡𝑛 ) )
󵄨󵄨 󵄨󵄨 𝑡𝑛 󵄨󵄨2
󵄨 󵄨
+ 2𝐸󵄨󵄨󵄨∫ (𝐺 (𝑡𝑛 − 𝑟)) 𝑥 (𝑟) 𝑑𝑟 − 𝑧 (𝑡𝑛 )󵄨󵄨󵄨 (30)
󵄨󵄨 𝑡𝑛+1
󵄨 𝑠 󵄨󵄨 0 󵄨󵄨
= 𝐸 ( 󵄨󵄨󵄨󵄨∫ 𝑓 (𝑥 (𝑠) , ∫ 𝐺 (𝑠 − 𝑟) 𝑥 (𝑟) 𝑑𝑟) 𝑑𝑠 󵄨󵄨 𝑡𝑛 󵄨󵄨2
󵄨󵄨 𝑡𝑛 0 󵄨 󵄨
≤ 4𝐸󵄨󵄨󵄨∫ (𝐺 (𝑠 − 𝑟) − 𝐺 (𝑡𝑛 − 𝑟)) 𝑥 (𝑟) 𝑑𝑟󵄨󵄨󵄨
󵄨󵄨 0 󵄨󵄨
− ℎ (𝜃𝑓 (𝑥 (𝑡𝑛+1 ) , 𝑧 (𝑡𝑛+1 ))
󵄨󵄨 𝑠 󵄨󵄨2
󵄨󵄨2 󵄨 󵄨
󵄨 + 4𝐸󵄨󵄨󵄨󵄨∫ (𝐺 (𝑠 − 𝑟)) 𝑥 (𝑟) 𝑑𝑟󵄨󵄨󵄨󵄨 + 2𝐶3 ℎ
+ (1 − 𝜃) 𝑓 (𝑥 (𝑡𝑛 ) , 𝑧 (𝑡𝑛 ))) 󵄨󵄨󵄨󵄨 ) 󵄨󵄨 𝑡𝑛 󵄨󵄨
󵄨󵄨
2
̂ 2 (𝑠 − 𝑡𝑛 ) + 4𝐶1 𝑀2 (𝑠 − 𝑡𝑛 ) + 2𝐶3 ℎ. 2
≤ 4𝑇2 𝐶1 𝐿
≤ 2𝐼1 + 2𝐼2 ,
(27) Therefore, by (14), we have
where
𝐼1 ≤ (1 − 𝜃) ℎ𝐿
𝐼1 𝑡𝑛 +(1−𝜃)ℎ
󵄨󵄨 𝑡𝑛 +(1−𝜃)ℎ 𝑠
×∫ (𝐶2 (𝑠 − 𝑡𝑛 )
󵄨 𝑡𝑛
= 𝐸 ( 󵄨󵄨󵄨󵄨∫ (𝑓 (𝑥 (𝑠) , ∫ 𝐺 (𝑠 − 𝑟) 𝑥 (𝑟) 𝑑𝑟)
󵄨󵄨 𝑡𝑛 0 2
̂ 2 + 𝑀2 ) (𝑠 − 𝑡𝑛 ) + 2𝐶3 ℎ) 𝑑𝑠
+4𝐶1 (𝑇2 𝐿
󵄨󵄨2
󵄨 1
−𝑓 (𝑥 (𝑡𝑛 ) , 𝑧 (𝑡𝑛 )) ) 𝑑𝑠󵄨󵄨󵄨󵄨 ) , = ( 𝐶2 (1 − 𝜃)3 + 2𝐶3 (1 − 𝜃)2 ) 𝐿ℎ3
󵄨󵄨 2
𝐼2 ̂ 2 + 𝑀2 ) 𝐿
4𝐶1 (𝑇2 𝐿
+ (1 − 𝜃)4 ℎ4 .
󵄨󵄨 𝑡𝑛+1 3
󵄨
= 𝐸 ( 󵄨󵄨󵄨󵄨 ∫ (𝑓 (𝑥 (𝑡𝑛+1 ) , 𝑧 (𝑡𝑛+1 )) (31)
󵄨󵄨 𝑡𝑛 +(1−𝜃)ℎ
Since 0 < ℎ < 1, there exists a constant 𝐶5 which does not
𝑠 󵄨󵄨2
󵄨 depend on ℎ, such that
−𝑓 (𝑥 (𝑠) , ∫ 𝐺 (𝑠 − 𝑟) 𝑥 (𝑟) 𝑑𝑟)) 𝑑𝑠󵄨󵄨󵄨 ) .
0 󵄨󵄨
(28) 𝐼1 ≤ 𝐶5 ℎ3 . (32)
Abstract and Applied Analysis 5

We can also prove that there exists a constant 𝐶6 which does where 𝐶8 –𝐶11 are independent of ℎ. Since 0 < ℎ < 1, we
not depend on ℎ, such that obtain

𝐼2 ≤ 𝐶6 ℎ3 , 󵄨 󵄨2
(33) 𝐸 (󵄨󵄨󵄨𝛿𝑛+1 󵄨󵄨󵄨 ) ≤ 𝐶12 ℎ2 , (38)
then we have
where 𝐶12 depends on 𝐶2 , 𝐶7 –𝐶11 . The proof of the lemma is
󵄨 󵄨2 1/2
𝐸(󵄨󵄨󵄨󵄨𝐸 (𝛿𝑛+1 | F𝑡𝑛 )󵄨󵄨󵄨󵄨 ) ≤ √2 (𝐶5 + 𝐶6 )ℎ3/2 . (34) completed.

Definition 5. The stochastic Θ-method (15)–(17) is said to be

Now we prove that the stochastic Θ-method is consistent convergent in mean-square sense with order 𝑝, if
with order 1 in the mean-square sense. Using the inequality
(𝑎 + 𝑏 + 𝑐)2 ≤ 3𝑎2 + 3𝑏2 + 3𝑐2 , Hölder’s inequality, and Doob’s 󵄨 󵄨2 1/2
martingale inequality, we have max (𝐸󵄨󵄨󵄨𝜖𝑛 󵄨󵄨󵄨 ) ≤ 𝐶ℎ𝑝 as ℎ 󳨀→ 0. (39)
0≤𝑛≤𝑁

󵄨 󵄨2
𝐸 (󵄨󵄨󵄨𝛿𝑛+1 󵄨󵄨󵄨 ) Theorem 6. Under conditions (7)–(9), the numerical solution
produced by the stochastic Θ-method (15)–(17) converges to the
≤ 3 (1 − 𝜃) ℎ exact solution of (6) with order 1/2 in the mean-square sense.
𝑡𝑛 +(1−𝜃)ℎ 󵄨󵄨 𝑠
󵄨
×∫ 𝐸 󵄨󵄨󵄨𝑓 (𝑥 (𝑠) , ∫ 𝐺 (𝑠 − 𝑟) 𝑥 (𝑟) 𝑑𝑟) Proof. Using the definition of the global error, we have
𝑡𝑛 󵄨󵄨 0
𝜖𝑛+1
󵄨󵄨2
󵄨
−𝑓 (𝑥 (𝑡𝑛 ) , 𝑧 (𝑡𝑛 )) 󵄨󵄨󵄨 𝑑𝑠 = 𝑥 (𝑡𝑛+1 ) − 𝑥𝑛+1
󵄨󵄨
𝑡𝑛+1 󵄨󵄨 = 𝑥 (𝑡𝑛+1 ) − 𝑥 (𝑡𝑛 )
󵄨
+ 3𝜃ℎ ∫ 𝐸 󵄨󵄨󵄨𝑓 (𝑥 (𝑡𝑛+1 ) , 𝑧 (𝑡𝑛+1 ))
𝑡𝑛 +(1−𝜃)ℎ 󵄨󵄨
− ℎ [𝜃𝑓 (𝑥 (𝑡𝑛+1 ) , 𝑧 (𝑡𝑛+1 )) + (1 − 𝜃) 𝑓 (𝑥 (𝑡𝑛 ) , 𝑧 (𝑡𝑛 ))]
𝑠 󵄨󵄨2
󵄨
−𝑓 (𝑥 (𝑠) , ∫ 𝐺 (𝑠 − 𝑟) 𝑥 (𝑟) 𝑑𝑟)󵄨󵄨󵄨 𝑑𝑠 − 𝑔 (𝑥 (𝑡𝑛 ) , 𝑧 (𝑡𝑛 )) Δ𝑊𝑛 + 𝑥 (𝑡𝑛 )
0 󵄨󵄨
𝑡𝑛+1 󵄨󵄨 𝑠 + ℎ [𝜃𝑓 (𝑥 (𝑡𝑛+1 ) , 𝑧 (𝑡𝑛+1 )) + (1 − 𝜃) 𝑓 (𝑥 (𝑡𝑛 ) , 𝑧 (𝑡𝑛 ))]
󵄨
+ 3∫ 𝐸 󵄨󵄨󵄨𝑔 (𝑥 (𝑠) , ∫ 𝐻 (𝑠 − 𝑟) 𝑥 (𝑟) 𝑑𝑟)
𝑡𝑛 󵄨󵄨 0 + 𝑔 (𝑥 (𝑡𝑛 ) , 𝑧 (𝑡𝑛 )) Δ𝑊𝑛 − 𝑥𝑛
󵄨󵄨2 − ℎ [𝜃𝑓 (𝑥𝑛+1 , 𝑧𝑛+1 ) + (1 − 𝜃) 𝑓 (𝑥𝑛 , 𝑧𝑛 )]
󵄨
−𝑔 (𝑥 (𝑡𝑛 ) , 𝑧 (𝑡𝑛 )) 󵄨󵄨󵄨 𝑑𝑠
󵄨󵄨
− 𝑔 (𝑥𝑛 , 𝑧𝑛 ) Δ𝑊𝑛
≜ 3 (1 − 𝜃) ℎ𝐼̃1 + 3𝜃ℎ𝐼̃2 + 3𝐼̃3 .
= 𝛿𝑛+1 + 𝜖𝑛 + 𝑢𝑛 ,
(35)
(40)
By (7), (14), and (30), we get
where
ℎ𝐼̃1
𝑢𝑛 = ℎ [𝜃𝑓 (𝑥 (𝑡𝑛+1 ) , 𝑧 (𝑡𝑛+1 ))
𝑡𝑛 +(1−𝜃)ℎ
󵄨 󵄨2
≤ 𝐿ℎ ∫ (𝐸󵄨󵄨󵄨𝑥 (𝑠) − 𝑥 (𝑡𝑛 )󵄨󵄨󵄨 + (1 − 𝜃) 𝑓 (𝑥 (𝑡𝑛 ) , 𝑧 (𝑡𝑛 ))]
𝑡𝑛

󵄨󵄨 𝑠 󵄨󵄨2 + 𝑔 (𝑥 (𝑡𝑛 ) , 𝑧 (𝑡𝑛 )) Δ𝑊𝑛

󵄨 󵄨
+𝐸󵄨󵄨󵄨∫ 𝐺 (𝑠 − 𝑟) 𝑥 (𝑟) 𝑑𝑟 − 𝑧 (𝑡𝑛 )󵄨󵄨󵄨 ) 𝑑𝑠
󵄨󵄨 0 󵄨󵄨 − ℎ [𝜃𝑓 (𝑥𝑛+1 , 𝑧𝑛+1 ) + (1 − 𝜃) 𝑓 (𝑥𝑛 , 𝑧𝑛 )]
𝑡𝑛 +(1−𝜃)ℎ
≤ 𝐿ℎ ∫ ̂ 2 (𝑠 − 𝑡𝑛 )2
(𝐶2 (𝑠 − 𝑡𝑛 ) + 4𝑇2 𝐶12 𝐿 − 𝑔 (𝑥𝑛 , 𝑧𝑛 ) Δ𝑊𝑛 .
𝑡𝑛 (41)
2
+4𝐶12 𝑀2 (𝑠 − 𝑡𝑛 ) + 2𝐶32 ℎ2 ) 𝑑𝑠 Squaring both sides of (40), taking expectation and absolute
values, yields
1
= 𝐿𝐶2 (1 − 𝜃)3 ℎ3 + 𝐶7 ℎ4 ,
2 󵄨 󵄨2
(36) 𝐸 (󵄨󵄨󵄨𝜖𝑛+1 󵄨󵄨󵄨 )
󵄨 󵄨2 󵄨 󵄨2 󵄨 󵄨2
where 𝐶7 is independent of ℎ. Similarly, we can obtain ≤ 2𝐸 (󵄨󵄨󵄨𝛿𝑛+1 󵄨󵄨󵄨 ) + 𝐸 (󵄨󵄨󵄨𝜖𝑛 󵄨󵄨󵄨 ) + 2𝐸 (󵄨󵄨󵄨𝑢𝑛 󵄨󵄨󵄨 ) (42)

ℎ𝐼̃2 ≤ 𝐶8 ℎ3 + 𝐶9 ℎ4 , 𝐼̃3 ≤ 𝐶10 ℎ2 + 𝐶11 ℎ3 , 󵄨 󵄨 󵄨 󵄨

(37) + 2 󵄨󵄨󵄨𝐸 (𝜖𝑛 ⋅ 𝛿𝑛+1 )󵄨󵄨󵄨 + 2 󵄨󵄨󵄨𝐸 (𝜖𝑛 ⋅ 𝑢𝑛 )󵄨󵄨󵄨 .
6 Abstract and Applied Analysis

We now estimate the separate terms in (42) individually and Due to properties of conditional expectation and the consis-
in sequence. Due to the consistency condition in the mean- tency condition in the mean sense,
square sense of the method, we have
󵄨 󵄨
2 󵄨󵄨󵄨𝐸 (𝜖𝑛 ⋅ 𝛿𝑛+1 )󵄨󵄨󵄨
󵄨 󵄨2
𝐸 (󵄨󵄨󵄨𝛿𝑛+1 󵄨󵄨󵄨 ) ≤ 𝐶
̃1 ℎ2 . (43) 󵄨 󵄨
= 2 󵄨󵄨󵄨󵄨𝐸 (𝜖𝑛 ⋅ 𝐸 (𝛿𝑛+1 | F𝑡𝑛 ))󵄨󵄨󵄨󵄨
(47)
By using the inequality (𝑎 + 𝑏 + 𝑐)2 ≤ 3𝑎2 + 3𝑏2 + 3𝑐2 and 󵄨 󵄨2 1/2 󵄨 󵄨2 1/2
≤ 2(𝐸 (󵄨󵄨󵄨𝜖𝑛 󵄨󵄨󵄨 )) ⋅ (𝐸 (󵄨󵄨󵄨󵄨𝐸 (𝛿𝑛+1 | F𝑡𝑛 )󵄨󵄨󵄨󵄨 ))
condition (7), we obtain
̃2 ℎ2 + ℎ𝐸 (󵄨󵄨󵄨𝜖𝑛 󵄨󵄨󵄨2 ) .
≤𝐶 󵄨 󵄨
󵄨 󵄨2 󵄨 󵄨2
𝐸 (󵄨󵄨󵄨𝑢𝑛 󵄨󵄨󵄨 ) ≤ 3ℎ2 𝜃2 𝐿𝐸 (󵄨󵄨󵄨𝜖𝑛+1 󵄨󵄨󵄨 )
󵄨 󵄨2 Using the Lipschtiz condition (7) and properties of condi-
+ 3ℎ2 𝜃2 𝐿𝐸 (󵄨󵄨󵄨𝑧 (𝑡𝑛+1 ) − 𝑧𝑛+1 󵄨󵄨󵄨 )
tional expectation, we obtain
󵄨 󵄨2
+ 3ℎ2 (1 − 𝜃)2 𝐿𝐸 (󵄨󵄨󵄨𝜖𝑛 󵄨󵄨󵄨 ) (44)
󵄨 󵄨
󵄨 󵄨2 2 󵄨󵄨󵄨𝐸 (𝜖𝑛 ⋅ 𝑢𝑛 )󵄨󵄨󵄨
+ 3ℎ (1 − 𝜃) 𝐿𝐸 (󵄨󵄨󵄨𝑧 (𝑡𝑛 ) − 𝑧𝑛 󵄨󵄨󵄨 )
2 2
󵄨 󵄨 󵄨 󵄨
≤ 2𝐸 (󵄨󵄨󵄨𝜖𝑛 󵄨󵄨󵄨 ⋅ 󵄨󵄨󵄨󵄨𝐸 (𝑢𝑛 | F𝑡𝑛 )󵄨󵄨󵄨󵄨)
󵄨 󵄨2 󵄨 󵄨2
+ 3ℎ𝐿𝐸 (󵄨󵄨󵄨𝜖𝑛 󵄨󵄨󵄨 ) + 3ℎ𝐿𝐸 (󵄨󵄨󵄨𝑧 (𝑡𝑛 ) − 𝑧𝑛 󵄨󵄨󵄨 ) .
󵄨 󵄨 󵄨
≤ 2ℎ𝜃𝐸 ( 󵄨󵄨󵄨𝜖𝑛 󵄨󵄨󵄨 ⋅ 󵄨󵄨󵄨󵄨𝐸 (𝑓 (𝑥 (𝑡𝑛+1 ) , 𝑧 (𝑡𝑛+1 ))
󵄨
Note that −𝑓 (𝑥𝑛+1 , 𝑧𝑛+1 ) | F𝑡𝑛 )󵄨󵄨󵄨󵄨)

󵄨 󵄨2 + 2ℎ (1 − 𝜃) 𝐸
𝐸 (󵄨󵄨󵄨𝑧 (𝑡𝑛+1 ) − 𝑧𝑛+1 󵄨󵄨󵄨 )
󵄨 󵄨 󵄨 󵄨
󵄨󵄨 𝑛 󵄨󵄨2 × (󵄨󵄨󵄨𝜖𝑛 󵄨󵄨󵄨 ⋅ 󵄨󵄨󵄨𝑓 (𝑥 (𝑡𝑛 ) , 𝑧 (𝑡𝑛 )) − 𝑓 (𝑥𝑛 , 𝑧𝑛 )󵄨󵄨󵄨)
󵄨󵄨 󵄨󵄨
= 𝐸 (󵄨󵄨󵄨ℎ∑ 𝐺 (𝑡𝑛+1 − 𝑡𝑗 ) 𝜖𝑗 󵄨󵄨󵄨󵄨 )
󵄨 󵄨 󵄨2 󵄨 󵄨2
≤ ℎ𝐸 (󵄨󵄨󵄨𝜖𝑛 󵄨󵄨󵄨 ) + 𝜃𝐿ℎ𝐸 (󵄨󵄨󵄨𝜖𝑛+1 󵄨󵄨󵄨 )
󵄨󵄨 𝑗=0 󵄨󵄨
󵄨 󵄨
󵄨 󵄨2
𝑛
󵄨 󵄨 𝑛 󵄨 󵄨 󵄨 󵄨2 + 𝜃𝐿ℎ𝐸 (󵄨󵄨󵄨𝑧 (𝑡𝑛+1 ) − 𝑧𝑛+1 󵄨󵄨󵄨 )
≤ ℎ ∑ 󵄨󵄨󵄨󵄨𝐺 (𝑡𝑛+1 − 𝑡𝑗 )󵄨󵄨󵄨󵄨 ⋅ ∑ 󵄨󵄨󵄨󵄨𝐺 (𝑡𝑛+1 − 𝑡𝑗 )󵄨󵄨󵄨󵄨 𝐸 (󵄨󵄨󵄨󵄨𝜖𝑗 󵄨󵄨󵄨󵄨 )
2 (48)
(45) 󵄨 󵄨2
𝑗=0 𝑗=0 + (1 − 𝜃) 𝐿ℎ𝐸 (󵄨󵄨󵄨𝜖𝑛 󵄨󵄨󵄨 )
𝑛
󵄨 󵄨2 󵄨 󵄨2
≤ ℎ𝑀2 𝑇 ∑𝐸 (󵄨󵄨󵄨󵄨𝜖𝑗 󵄨󵄨󵄨󵄨 ) , + (1 − 𝜃) 𝐿ℎ𝐸 (󵄨󵄨󵄨𝑧 (𝑡𝑛 ) − 𝑧𝑛 󵄨󵄨󵄨 )
𝑗=0
󵄨 󵄨2 󵄨 󵄨2
≤ ℎ𝐸 (󵄨󵄨󵄨𝜖𝑛 󵄨󵄨󵄨 ) + 𝜃𝐿ℎ𝐸 (󵄨󵄨󵄨𝜖𝑛+1 󵄨󵄨󵄨 )
𝑛−1
󵄨 󵄨2 󵄨 󵄨2
𝐸 (󵄨󵄨󵄨𝑧 (𝑡𝑛 ) − 𝑧𝑛 󵄨󵄨󵄨 ) ≤ ℎ𝑀2 𝑇 ∑ 𝐸 (󵄨󵄨󵄨󵄨𝜖𝑗 󵄨󵄨󵄨󵄨 ) , 𝑛
󵄨 󵄨2
𝑗=0 + 𝜃𝐿ℎ2 𝑀2 𝑇 ∑𝐸 (󵄨󵄨󵄨󵄨𝜖𝑗 󵄨󵄨󵄨󵄨 )
𝑗=0

󵄨 󵄨2
then we have + (1 − 𝜃) 𝐿ℎ𝐸 (󵄨󵄨󵄨𝜖𝑛 󵄨󵄨󵄨 )

󵄨 󵄨2 󵄨 󵄨2 𝑛−1
𝐸 (󵄨󵄨󵄨𝑢𝑛 󵄨󵄨󵄨 ) ≤ 3ℎ2 𝜃2 𝐿𝐸 (󵄨󵄨󵄨𝜖𝑛+1 󵄨󵄨󵄨 ) 󵄨 󵄨2
+ (1 − 𝜃) 𝐿ℎ2 𝑀2 𝑇 ∑ 𝐸 (󵄨󵄨󵄨󵄨𝜖𝑗 󵄨󵄨󵄨󵄨 ) .
𝑗=0
+ 3ℎ3 𝜃2 𝐿𝑀2 𝑇
𝑛
󵄨 󵄨2 󵄨 󵄨2 Substituting (43) and (46)–(48) into (42), we have
× ∑𝐸 (󵄨󵄨󵄨󵄨𝜖𝑗 󵄨󵄨󵄨󵄨 ) + 3ℎ2 (1 − 𝜃)2 𝐿𝐸 (󵄨󵄨󵄨𝜖𝑛 󵄨󵄨󵄨 )
𝑗=0
󵄨 󵄨2
𝑛−1 (1 − 6ℎ2 𝜃2 𝐿 − ℎ𝜃𝐿) 𝐸 (󵄨󵄨󵄨𝜖𝑛+1 󵄨󵄨󵄨 )
󵄨 󵄨2
+ 3ℎ (1 − 𝜃) 𝐿𝑀 𝑇 ∑ 𝐸 (󵄨󵄨󵄨󵄨𝜖𝑗 󵄨󵄨󵄨󵄨 )
3 2 2
𝑛
𝑗=0 󵄨 󵄨2 ̃4 ℎ2 ∑ 𝐸 (󵄨󵄨󵄨󵄨𝜖𝑗 󵄨󵄨󵄨󵄨2 )
̃3 ℎ𝐸 (󵄨󵄨󵄨𝜖𝑛 󵄨󵄨󵄨2 ) + 𝐶
≤ 𝐸 (󵄨󵄨󵄨𝜖𝑛 󵄨󵄨󵄨 ) + 𝐶 󵄨 󵄨 󵄨 󵄨
𝑛−1 𝑗=0 (49)
󵄨 󵄨2 󵄨 󵄨2
+ 3ℎ𝐿𝐸 (󵄨󵄨󵄨𝜖𝑛 󵄨󵄨󵄨 ) + 3ℎ2 𝐿𝑀2 𝑇 ∑ 𝐸 (󵄨󵄨󵄨󵄨𝜖𝑗 󵄨󵄨󵄨󵄨 ) . 𝑛
𝑗=0 ̃5 ℎ3 ∑ 𝐸 (󵄨󵄨󵄨󵄨𝜖𝑗 󵄨󵄨󵄨󵄨2 ) + (2𝐶
+𝐶 ̃1 + 𝐶
̃2 ) ℎ2 ,
(46) 󵄨 󵄨
𝑗=0
Abstract and Applied Analysis 7

where Definition 7. The solution 𝑥(𝑡) of SVIDE (1) is said to be

mean-square exponentially stable if there exists a pair of pos-
̃3 = 2 + 6𝐿 + (1 − 𝜃) 𝐿,
𝐶 ̃ such that
itive constants 𝜆 and 𝐶
̃4 = 6(1 − 𝜃)2 𝐿 + 7𝐿𝑀2 𝑇,
𝐶 󵄨 ̃ 󵄨󵄨󵄨󵄨 𝑥0 |2 𝑒−𝜆𝑡
(50) 𝐸 󵄨󵄨󵄨󵄨𝑥 (𝑡) |2 ≤ 𝐶𝐸 󵄨 on 𝑡 ≥ 0 (56)
̃5 = 6𝜃2 𝐿𝑀2 𝑇 + 6(1 − 𝜃)2 𝐿𝑀2 𝑇.
𝐶
for all initial data 𝑥0 ∈ 𝑅𝑑 .
Since 0 ≤ 𝜃 ≤ 1, for sufficiently small stepsize ℎ, there exists
a constant 𝜅 ∈ (0, 1), such that Definition 8. The solution 𝑥(𝑡) of SVIDE (1) is said to be
mean-square asymptotically stable if
1 − 6ℎ2 𝜃2 𝐿 − ℎ𝜃𝐿 ≥ 1 − 2ℎ𝜃𝐿 ≥ 𝜅,
(51) lim 𝐸|𝑥 (𝑡)|2 = 0, (57)
𝑡→∞
̃4 ℎ + 𝐶
𝐶 2 ̃5 ℎ ≤ 2𝐶
̃4 ℎ .
3 2

for all initial data 𝑥0 ∈ 𝑅𝑑 .

Let
󵄨 󵄨2 Obviously, (56) implies (57).
𝑅𝑛 = max (𝐸 (󵄨󵄨󵄨𝜖𝑖 󵄨󵄨󵄨 )) , (52)
1≤𝑖≤𝑛
Theorem 9. Assume that there exist six positive constants 𝜆 1 ,
then we have 𝜆 2 , 𝜆 3 , 𝜆 4 , 𝛽, and 𝛾 such that

1+𝐶 ̃3 ℎ 2𝐶̃4 𝑇ℎ ̃ +𝐶
2𝐶 ̃2 2 2𝑥𝑇 𝑓 (𝑥, 0) ≤ −𝜆 1 |𝑥|2 , (58)
𝑅𝑛+1 ≤ 𝑅𝑛 + 𝑅𝑛 + 1 ℎ
1 − 2ℎ𝜃𝐿 1 − 2ℎ𝜃𝐿 1 − 2ℎ𝜃𝐿 󵄨󵄨 󵄨 󵄨 󵄨
(53) 󵄨󵄨𝑓 (𝑥, 𝑦) − 𝑓 (𝑥, 0)󵄨󵄨󵄨 ≤ 𝜆 2 󵄨󵄨󵄨𝑦󵄨󵄨󵄨 , (59)
2𝜃𝐿 + 𝐶 ̃4 𝑇
̃3 + 2𝐶 ̃ +𝐶
2𝐶 ̃2 2
≤ [1 + ℎ] 𝑅𝑛 + 1 ℎ. 󵄨󵄨 󵄨2 󵄨 󵄨2
󵄨󵄨𝑔 (𝑥, 𝑦)󵄨󵄨󵄨 ≤ 𝜆 3 |𝑥| + 𝜆 4 󵄨󵄨󵄨𝑦󵄨󵄨󵄨 ,
2
𝜅 𝜅 (60)

Denote for all 𝑥, 𝑦 ∈ 𝑅𝑑 . And the convolution kernel functions 𝐺 and

𝐻 decay exponentially fast, that is
̃ ̃
̂ = 2𝜃𝐿 + 𝐶3 + 2𝐶4 𝑇 ,
𝐶 (54)
𝜅 ‖𝐺 (𝑡)‖ ∨ ‖𝐻 (𝑡)‖ ≤ 𝛽𝑒−𝛾𝑡 , 𝑡 ≥ 0. (61)
and note that 𝑅0 = 0, thus we obtain If
̃ ̃
̂ 𝑅𝑛 + 2𝐶1 + 𝐶2 ℎ2
𝑅𝑛+1 ≤ (1 + 𝐶ℎ) 𝛽2
𝜅 −𝜆 1 + 𝜆 2 + 𝜆 3 + (𝜆 2 + 𝜆 4 ) < 0, (62)
𝛾2
̃ ̃ 𝑛
̂ 𝑛+1 𝑅0 + 2𝐶1 + 𝐶2 ℎ2 ∑ (1 + 𝐶ℎ)
≤ (1 + 𝐶ℎ) ̂ 𝑗 then for any given initial data 𝑥0 , there exists a pair of positive
𝜅 𝑗=0 constants 𝜌 and 𝐶 such that the solution 𝑥(𝑡) of SVIDE (1) has
property that
̃1 + 𝐶
2𝐶 ̃2
= ̂ 𝑛+1 − 1) ℎ
((1 + 𝐶ℎ)
̂ 󵄨 󵄨2
𝐶 𝐸|𝑥 (𝑡)|2 ≤ 𝐶𝐸󵄨󵄨󵄨𝑥0 󵄨󵄨󵄨 𝑒−𝜌𝑡 , ∀𝑡 ≥ 0, (63)
̃1 + 𝐶
2𝐶 ̃2 𝐶ℎ(𝑛+1)
̂ 2𝐶 ̃2 𝐶𝑇
̃1 + 𝐶 ̂
≤ (𝑒 − 1) ℎ ≤ (𝑒 − 1) ℎ. which implies (57).
̂
𝐶𝜅 ̂
𝐶𝜅
(55) Proof. The proof of this theorem is similar to that of
Theorem 2.1 in [5] by choosing the Lyapunov function
From the above, the theorem is proved. 𝑉(𝑡, 𝑥) = 𝑒𝜆 1 𝑡 |𝑥|2 .

3. Stability Analysis Remark 10. If we apply Theorem 9 to problem (5), then the
corresponding mean-square stability condition is
In this section, we will discuss the analytical stability and
numerical stability of SVIDE (1). First, the analytical stability 𝜆 4 𝛽2 < 𝜆 1 𝛾2 . (64)
is derived by the technique in [5]. We extend the exponential
stability result to a more general type. And then the mean- This condition is in accordance with the assumption of
square asymptotic stability of the stochastic Θ-method is Theorem 2.1 in Mao’s papaer [5].
investigated.
Let us give the definition of the mean-square exponential In the following, we study the mean-square asymptotic
stability (see Mao [19]). stability of the stochastic Θ-method for SVIDE (1).
8 Abstract and Applied Analysis

Definition 11. The numerical solution {𝑥𝑛 }𝑛≥0 to SVIDE (1) is Substituting (71) into (70), we have
said to be mean-square asymptotically stable if the numerical
󵄨 󵄨2
solution has the property that 𝐸󵄨󵄨󵄨𝐹 (𝑥𝑛+1 )󵄨󵄨󵄨
󵄨 󵄨2 󵄨 󵄨2 󵄨 󵄨2
lim 𝐸󵄨󵄨󵄨𝑥𝑛 󵄨󵄨󵄨 = 0, (65) ≤ 𝐸󵄨󵄨󵄨𝐹 (𝑥𝑛 )󵄨󵄨󵄨 + ℎ (−𝜆 1 + 𝜆 2 + 𝜆 3 ) 𝐸󵄨󵄨󵄨𝑥𝑛 󵄨󵄨󵄨
𝑛→∞
(72)
for any initial data 𝑥0 . 2 𝑛−1
𝛽 󵄨 󵄨2
+ ℎ (𝜆 2 + 𝜆 4 ) (ℎ ∑ 𝑒−𝛾(𝑡𝑛 −𝑡𝑗 ) 𝐸󵄨󵄨󵄨󵄨𝑥𝑗 󵄨󵄨󵄨󵄨 ) ,
𝛾 𝑗=0
Theorem 12. Under conditions (58)–(62), the stochastic Θ-
method is mean-square asymptotically stable for any stepsize
ℎ > 0, if 1/2 ≤ 𝜃 ≤ 1. which implies
󵄨 󵄨2
Proof. By (15), we have 𝐸󵄨󵄨󵄨𝐹 (𝑥𝑛+1 )󵄨󵄨󵄨
𝑛
𝐹 (𝑥𝑛+1 ) = 𝐹 (𝑥𝑛 ) + ℎ𝑓 (𝑥𝑛 , 𝑧𝑛 ) + 𝑔 (𝑥𝑛 , 𝑧𝑛 ) Δ𝑊𝑛 , (66) 󵄨 󵄨2 󵄨 󵄨2
≤ 𝐸 󵄨󵄨󵄨𝐹 (𝑥0 )󵄨󵄨󵄨 + ℎ (−𝜆 1 + 𝜆 2 + 𝜆 3 ) ∑𝐸󵄨󵄨󵄨𝑥𝑖 󵄨󵄨󵄨
𝑖=0
where
𝛽2 𝑛 𝑖−1
󵄨 󵄨2
𝐹 (𝑥𝑛 ) = 𝑥𝑛 − 𝜃ℎ𝑓 (𝑥𝑛 , 𝑧𝑛 ) . (67) + ℎ (𝜆 2 + 𝜆 4 ) ∑ (ℎ∑ 𝑒−𝛾(𝑡𝑖 −𝑡𝑗 ) 𝐸󵄨󵄨󵄨󵄨𝑥𝑗 󵄨󵄨󵄨󵄨 )
𝛾 𝑖=0 𝑗=0
Consequently
𝑛
󵄨󵄨 󵄨2 󵄨 󵄨2 󵄨 󵄨2
󵄨󵄨𝐹 (𝑥𝑛+1 )󵄨󵄨󵄨 ≤ 𝐸󵄨󵄨󵄨𝐹 (𝑥0 )󵄨󵄨󵄨 + ℎ (−𝜆 1 + 𝜆 2 + 𝜆 3 ) ∑𝐸󵄨󵄨󵄨𝑥𝑖 󵄨󵄨󵄨
𝑖=0
󵄨 󵄨2 󵄨 󵄨2
= 󵄨󵄨󵄨𝐹 (𝑥𝑛 )󵄨󵄨󵄨 + (1 − 2𝜃) ℎ2 󵄨󵄨󵄨𝑓 (𝑥𝑛 , 𝑧𝑛 )󵄨󵄨󵄨
𝛽2 𝑛−1 𝑛
󵄨 󵄨2
󵄨 󵄨2
(68) + ℎ (𝜆 2 + 𝜆 4 ) ∑ (ℎ ∑ 𝑒−𝛾(𝑡𝑖 −𝑡𝑗 ) 𝐸󵄨󵄨󵄨󵄨𝑥𝑗 󵄨󵄨󵄨󵄨 ) (73)
+ 󵄨󵄨󵄨𝑔 (𝑥𝑛 , 𝑧𝑛 ) Δ𝑊𝑛 󵄨󵄨󵄨 + 2 ⟨𝑥𝑛 , ℎ𝑓 (𝑥𝑛 , 𝑧𝑛 )⟩ 𝛾 𝑗=0 𝑖=𝑗+1

+ 2 ⟨𝐹 (𝑥𝑛 ) + ℎ𝑓 (𝑥𝑛 , 𝑧𝑛 ) , 𝑔 (𝑥𝑛 , 𝑧𝑛 ) Δ𝑊𝑛 ⟩ . 󵄨 󵄨2

≤ 𝐸󵄨󵄨󵄨𝐹 (𝑥0 )󵄨󵄨󵄨 + ℎ (−𝜆 1 + 𝜆 2 + 𝜆 3 )
Taking expectation on both sides of (68) and by 1/2 ≤ 𝜃 ≤ 1, 𝑛
󵄨 󵄨2 𝛽2 𝑛−1 󵄨 󵄨2
it follows that × ∑𝐸󵄨󵄨󵄨𝑥𝑖 󵄨󵄨󵄨 + ℎ (𝜆 2 + 𝜆 4 ) 2 ∑ 𝐸󵄨󵄨󵄨󵄨𝑥𝑗 󵄨󵄨󵄨󵄨
𝑖=0 𝛾 𝑗=0
󵄨 󵄨2 󵄨 󵄨2 󵄨 󵄨2
𝐸󵄨󵄨󵄨𝐹 (𝑥𝑛+1 )󵄨󵄨󵄨 ≤ 𝐸󵄨󵄨󵄨𝐹 (𝑥𝑛 )󵄨󵄨󵄨 + ℎ𝐸󵄨󵄨󵄨𝑔 (𝑥𝑛 , 𝑧𝑛 )󵄨󵄨󵄨
2
(69) 󵄨 󵄨2 𝛽
+ 2𝐸 ⟨𝑥𝑛 , ℎ𝑓 (𝑥𝑛 , 𝑧𝑛 )⟩ . ≤ 𝐸󵄨󵄨󵄨𝐹 (𝑥0 )󵄨󵄨󵄨 + ℎ (−𝜆 1 + 𝜆 2 + 𝜆 3 + (𝜆 2 + 𝜆 4 ) 2 )
𝛾
By conditions (58)–(60), we have 𝑛
󵄨 󵄨2
× ∑𝐸󵄨󵄨󵄨𝑥𝑖 󵄨󵄨󵄨 .
󵄨 󵄨2
𝐸󵄨󵄨󵄨𝐹 (𝑥𝑛+1 )󵄨󵄨󵄨 𝑖=0

󵄨 󵄨2 󵄨 󵄨2
≤ 𝐸󵄨󵄨󵄨𝐹 (𝑥𝑛 )󵄨󵄨󵄨 + ℎ (−𝜆 1 + 𝜆 2 + 𝜆 3 ) 𝐸󵄨󵄨󵄨𝑥𝑛 󵄨󵄨󵄨 (70) Therefore by condition (62), we have

󵄨 󵄨2 󵄨 󵄨2 𝑛
+ ℎ𝜆 2 𝐸󵄨󵄨󵄨𝑧𝑛 󵄨󵄨󵄨 + ℎ𝜆 4 𝐸󵄨󵄨󵄨𝑧𝑛 󵄨󵄨󵄨 . 󵄨 󵄨2 󵄨 󵄨2
∑𝐸󵄨󵄨󵄨𝑥𝑖 󵄨󵄨󵄨 ≤ 𝐶𝐸󵄨󵄨󵄨𝐹 (𝑥0 )󵄨󵄨󵄨 (74)
𝑖=0
And by (61) and the Jensen inequality, we derive that
for any 𝑛 > 0, where 𝐶 = −1/ℎ(−𝜆 1 + 𝜆 2 + 𝜆 3 + (𝜆 2 +
󵄨󵄨 𝑛−1 󵄨󵄨2 2
󵄨󵄨 󵄨󵄨2 󵄨󵄨 󵄨󵄨 𝑛−1
−𝛾(𝑡𝑛 −𝑡𝑗 ) 󵄨󵄨 󵄨󵄨 𝜆 4 )(𝛽2 /𝛾2 )). Let 𝑛 → ∞, we find that the infinite series
󵄨 󵄨 2
󵄨󵄨𝑧𝑛 󵄨󵄨 = 󵄨󵄨󵄨ℎ ∑ 𝐺 (𝑡𝑛 − 𝑡𝑗 ) 𝑥𝑗 󵄨󵄨󵄨 ≤ 𝛽 (ℎ ∑ 𝑒 󵄨󵄨𝑥𝑗 󵄨󵄨)
󵄨󵄨 𝑗=0 󵄨󵄨 󵄨 󵄨 ∑∞ 2
𝑖=0 𝐸|𝑥𝑖 | is bounded, then we have
󵄨 󵄨 𝑗=0
󵄨 󵄨2
𝑛−1 𝑛−1 lim 𝐸󵄨󵄨󵄨𝑥𝑛 󵄨󵄨󵄨 = 0 (75)
−𝛾(𝑡𝑛 −𝑡𝑗 ) 󵄨󵄨 󵄨2 𝑛→∞
2
≤ 𝛽 (ℎ ∑ 𝑒 −𝛾(𝑡𝑛 −𝑡𝑗 )
) ⋅ (ℎ ∑ 𝑒 󵄨󵄨𝑥𝑗 󵄨󵄨󵄨 )
󵄨 󵄨
𝑗=0 𝑗=0 which shows that the stochastic Θ-method is mean-square
asymptotically stable. The proof of the theorem is com-
𝑛−1
𝛽2 󵄨 󵄨2
(ℎ ∑ 𝑒−𝛾(𝑡𝑛 −𝑡𝑗 ) 󵄨󵄨󵄨󵄨𝑥𝑗 󵄨󵄨󵄨󵄨 ) ,
pleted.
≤
𝛾 𝑗=0
Theorem 13. Assume that conditions (58)–(62) hold and that
2 𝑛−1 function 𝑓 satisfies the linear growth condition; namely, there
󵄨󵄨 󵄨󵄨2 𝛽 󵄨 󵄨2
󵄨󵄨𝑧𝑛 󵄨󵄨 ≤ (ℎ ∑ 𝑒−𝛾(𝑡𝑛 −𝑡𝑗 ) 󵄨󵄨󵄨󵄨𝑥𝑗 󵄨󵄨󵄨󵄨 ) . ̂ > 0 such that
exists a constant 𝐾
𝛾 𝑗=0
󵄨󵄨 󵄨2 ̂ 󵄨 󵄨2
(71) 󵄨󵄨𝑓 (𝑥, 𝑦)󵄨󵄨󵄨 ≤ 𝐾 (|𝑥|2 + 󵄨󵄨󵄨𝑦󵄨󵄨󵄨 ) . (76)
Abstract and Applied Analysis 9

If 0 ≤ 𝜃 < 1/2, then the stochastic Θ-method is mean-square 4. Numerical Experiments

asymptotically stable when the stepsize ℎ ∈ (0, ℎ0 ), where
In this section, some numerical examples are given to vali-
𝜆 1 − 𝜆 2 − 𝜆 3 − (𝜆 2 + 𝜆 4 ) (𝛽2 /𝛾2 ) date our conclusions. First we illustrate the theoretical con-
ℎ0 = . (77) vergence order of stochastic Θ-method. Here we consider the
(1 − 2𝜃) 𝐾 ̂ (1 + (𝛽2 /𝛾2 ))
following linear SVIDE:
Proof. By (68) and the linear growth condition (76), we have
𝑑𝑥 (𝑡)
󵄨 󵄨2
𝐸󵄨󵄨󵄨𝐹 (𝑥𝑛+1 )󵄨󵄨󵄨
𝑡
󵄨 󵄨2 = (𝑎𝑥 (𝑡) + 𝑏 ∫ 𝑒−3(𝑡−𝑠) 𝑥 (𝑠) 𝑑𝑠 ) 𝑑𝑡
≤ 𝐸󵄨󵄨󵄨𝐹 (𝑥𝑛 )󵄨󵄨󵄨 0
(78) (82)
󵄨 󵄨2 𝑡
+ [ℎ (−𝜆 1 + 𝜆 2 + 𝜆 3 ) + (1 − 2𝜃) ℎ 𝐾] 𝐸󵄨󵄨󵄨𝑥𝑛 󵄨󵄨󵄨
2̂
+ (𝑐𝑥 (𝑡) + 𝑑 ∫ 𝑒 −2(𝑡−𝑠)
𝑥 (𝑠) 𝑑𝑠 ) 𝑑𝑊 (𝑡) ,
0
̂ 𝐸󵄨󵄨󵄨𝑧𝑛 󵄨󵄨󵄨2 + ℎ𝜆 4 𝐸󵄨󵄨󵄨𝑧𝑛 󵄨󵄨󵄨2 ,
+ [ℎ𝜆 2 + (1 − 2𝜃) ℎ2 𝐾] 󵄨 󵄨 󵄨 󵄨 𝑡 ∈ [0, 𝑇] ,
which implies
with the initial data 𝑥(0) = 1. And we choose the parameters
󵄨 󵄨2
𝐸󵄨󵄨󵄨𝐹 (𝑥𝑛+1 )󵄨󵄨󵄨 𝑎 = −1, 𝑏 = 0.5, 𝑐 = 0.5, 𝑑 = 0.5, and 𝑇 = 1. Since the
true solution of (82) can not be obtained, thus we compute a
󵄨 󵄨2
≤ 𝐸󵄨󵄨󵄨𝐹 (𝑥0 )󵄨󵄨󵄨 reference solution with the trapezoidal rule (𝜃 = 0.5) under
a very small stepsize ℎ = 2−15 . The mean-square error is
𝑛
̂ ∑𝐸󵄨󵄨󵄨𝑥𝑖 󵄨󵄨󵄨2
+ [ℎ (−𝜆 1 + 𝜆 2 + 𝜆 3 ) + (1 − 2𝜃) ℎ2 𝐾]
denoted as follows:
󵄨 󵄨
𝑖=0 1/2
1 𝑀 󵄨󵄨 󵄨2
err := max ( ∑ 󵄨󵄨𝑥 (𝑡𝑛 , 𝜔𝑗 ) − 𝑥𝑛 (𝜔𝑗 )󵄨󵄨󵄨 )
̂ 𝛽2 𝑛−1 󵄨󵄨 󵄨󵄨2 𝑀 𝑗=1 󵄨 󵄨
+ [ℎ𝜆 2 + (1 − 2𝜃) ℎ2 𝐾] ∑ 𝐸󵄨󵄨𝑥 󵄨󵄨 1≤𝑛≤𝑁
𝛾2 𝑗=0 󵄨 𝑗 󵄨 (83)
󵄨 󵄨󵄨2 1/2
≈ max (𝐸󵄨󵄨󵄨𝜀𝑛 󵄨󵄨 ) ,
𝛽2 𝑛−1 󵄨 󵄨2 1≤𝑛≤𝑁
+ ℎ𝜆 4 2 ∑ 𝐸󵄨󵄨󵄨󵄨𝑥𝑗 󵄨󵄨󵄨󵄨
𝛾 𝑗=0 where 𝑁 denotes the number of steps and 𝑀 is the number
2
of sample paths. In our experiments, we choose 𝑀 = 500.
󵄨 󵄨2 𝛽 Here we choose four different numerical methods with 𝜃 = 0
≤ 𝐸󵄨󵄨󵄨𝐹 (𝑥0 )󵄨󵄨󵄨 + [ℎ (−𝜆 1 + 𝜆 2 + 𝜆 3 + (𝜆 2 + 𝜆 4 ) 2 )
𝛾 (classical Euler-Maruyama method), 𝜃 = 0.3, 𝜃 = 0.7, and
𝜃 = 1 (Backward Euler method), and the numerical results
𝑛
𝛽2 󵄨 󵄨2 are presented in Table 1. From the table, one can easily receive
̂ (1 +
+ ℎ2 (1 − 2𝜃) 𝐾 )] ∑𝐸󵄨󵄨󵄨𝑥𝑖 󵄨󵄨󵄨 .
𝛾 2
𝑖=0
that the mean-square convergence order of the stochastic Θ-
(79) method for SVIDEs is 1/2. Also, we illustrate the convergence
result by Figure 1, where the reference dotted curve has a slope
By condition (62), 0 ≤ 𝜃 < 1/2 and ℎ ∈ (0, ℎ0 ), we receive 1/2.
In the rest of this section, some numerical experiments
𝛽2
ℎ (−𝜆 1 + 𝜆 2 + 𝜆 3 + (𝜆 2 + 𝜆 4 ) ) are given to validate our stability results. Consider the fol-
𝛾2 lowing nonlinear SVIDE:
(80)
2
2 ̂ (1 + 𝛽 ) < 0.
+ ℎ (1 − 2𝜃) 𝐾 𝑑𝑥 (𝑡)
𝛾 2
𝑡
Therefore we conclude that = (𝑎𝑥 (𝑡) + 𝑏 sin (∫ 𝑒−3(𝑡−𝑠) 𝑥 (𝑠) 𝑑𝑠)) 𝑑𝑡
0
󵄨 󵄨2
lim 𝐸󵄨󵄨𝑥 󵄨󵄨 = 0. (81) (84)
𝑛 → ∞ 󵄨 𝑛󵄨 𝑡
−4(𝑡−𝑠)
+ (𝑐𝑥 (𝑡) + 𝑑 sin (∫ 𝑒 𝑥 (𝑠) 𝑑𝑠)) 𝑑𝑊 (𝑡) ,
The proof of the theorem is completed. 0

Remark 14. In recent years, there are so many valuable results 𝑡 ≥ 0,

on the stability analysis of numerical methods for stochastic
with the initial data 𝑥(0) = 1. We choose the parameters 𝑎 =
delay integro-differential equations. However, these results
−10, 𝑏 = 3, 𝑐 = 2, and 𝑑 = 1. By conditions (58)–(61), we
can not be applied to stochastic Volterra integro-differential
have
equations directly since they have essential differences. More-
over, for the deterministic cases, these two issues are studied 𝜆 1 = 20, 𝜆 2 = 3, 𝜆 3 = 6, 𝜆 4 = 3,
separately in the past few decades; please refer to the book [2] (85)
by Brunner. ̂ = 130,
𝐾 𝛽 = 1, 𝛾 = 3.
10 Abstract and Applied Analysis

Table 1: Mean-square errors of the four methods for (82).

𝜃=0 𝜃 = 0.3 𝜃 = 0.7 𝜃=1

ℎ = 2−4 0.03366846 0.03079402 0.02981565 0.03126486
ℎ = 2−5 0.02200355 0.02111309 0.02098438 0.02167162
ℎ = 2−6 0.01594512 0.01560370 0.01551390 0.01572048
ℎ = 2−7 0.01064680 0.01058695 0.01064387 0.01078690
ℎ = 2−8 0.00726183 0.00724734 0.00727678 0.00733495

100 100

10−1 10−1

log(err)
log(err)

10−2 10−2

10−3 10−3
10−3 10−2 10−1 10−3 10−2 10−1
log(h) log(h)

𝜃=0 𝜃 = 0.3
(a) (b)
0
10 100

10−1 10−1
log(err)

log(err)

10−2 10−2

10−3 10−3
10−3 10−2 10−1 10−3 10−2 10−1
log(h) log(h)

𝜃 = 0.7 𝜃=1
(c) (d)

Figure 1: Mean-square error plots of the four methods for (82).

Abstract and Applied Analysis 11

×1028 1

5.5 0.9
5 0.8
4.5
0.7
4
0.6
3.5
2

0.5
E(xn )

2
E(xn )
2.5 0.4
2
0.3
1.5
0.2
1
0.1
0.5

0 5 10 15 20 0 5 10 15 20
Time: t Time: t

𝜃 = 0, h = 1/4 𝜃 = 0, h = 1/16
(a) (b)

Figure 2: Numerical simulation with 𝜃 = 0 under different stepsizes ℎ = 1/4 and ℎ = 1/16.

1
2200
0.9
2000
0.8
1800
0.7
1600
1400 0.6
2
E(xn )
2
E(xn )

1200 0.5
1000 0.4
800
0.3
600
0.2
400
200 0.1

0 5 10 15 20 0 5 10 15 20
Time: t Time: t

𝜃 = 0.3, h = 1/2 𝜃 = 0.3, h = 1/8

(a) (b)

Figure 3: Numerical simulation with 𝜃 = 0.3 under different stepsizes ℎ = 1/2 and ℎ = 1/8.

Therefore, Θ-method with different parameters. Here we choose 𝜃 = 0,

𝛽 2 𝜃 = 0.3, 𝜃 = 0.8, and 𝜃 = 1 where 0, 0.3 ∈ [0, 1/2) and 0.8, 1 ∈
31
−𝜆 1 + 𝜆 2 + 𝜆 3 + (𝜆 2 + 𝜆 4 ) = − < 0, (86) [1/2, 1]. By Theorem 13, we have
𝛾2 3
which implies that the true solution of (84) is mean-square
asymptotically stable. (i) if 𝜃 = 0, ℎ0 = 93/1300. When the stepsize ℎ ∈
In the following numerical experiments, we will show the (0, 93/1300), the Euler-Maruyama method is mean-
numerical mean-square asymptotic stability of the stochastic square asymptotically stable;
12 Abstract and Applied Analysis

1 1

0.9 0.9

0.8 0.8

0.7 0.7

0.6 0.6
2
E(xn )

2
E(xn )
0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0 5 10 15 20 0 5 10 15 20
Time: t Time: t

𝜃 = 0.8, h = 2 𝜃 = 1, h = 4
(a) (b)

Figure 4: Numerical simulation with 𝜃 = 0.8 and 𝜃 = 1 under even large stepsizes ℎ = 2 and ℎ = 3.

(ii) if 𝜃 = 0.3, ℎ0 = 93/520. When the stepsize ℎ ∈ Conflict of Interests

(0, 93/520), the method is mean-square asymptoti-
cally stable. The authors declare that there is no conflict of interests
regarding the publication of this paper.
In Figures 2 and 3, one can find that when the stepsize ℎ ∈
(0, ℎ0 ), 𝐸|𝑥𝑛 |2 tends to zero, but it tends to infinity when
Acknowledgments
the stepsize is even large. Since the stability conditions are
sufficient, one may find that the restriction of the stepsize in The authors are indebted to the referees and the editors
Theorem 13 is not optimal. for their carefully reading of this paper and their valuable
Theorem 12 says that when 𝜃 ∈ [1/2, 1], the stochastic Θ- comments. This work was in part supported by NSF of China
method is mean-square asymptotically stable for any stepsize, (nos. 11326240, 11401545, 11371157, and 91130003). The first
and the following numerical experiments demonstrate this author was also supported by the Fundamental Research
for a selection stepsize (see Figure 4). Funds for the Central Universities, China University of
Geosciences (Wuhan) (no. CUG120828). The second author
5. Conclusions was supported by the Fundamental Research Funds for the
Central Universities (no. 2013TS137).
In this paper, we first studied the mean-square convergence
the stochastic Θ-method for stochastic Volterra integro-
differential equations and proved that the stochastic Θ-
References
method has mean-square convergence order 1/2 when the [1] T. A. Burton, Volterra Integral and Differential Equations, Elsev-
functions 𝑓 and 𝑔 in (1) are globally Lipschitz continuous in ier, 2005.
all variables and satisfy linear growth condition. And then the [2] H. Brunner, Collocation Methods for Volterra Integral and
mean-square stability of such method has been considered. Related Functional Differential Equations, Cambridge Univer-
It should be mentioned that the assumptions of the mean- sity Press, 2004.
square convergence in the paper are slightly strong. In the [3] A. Feldstein and J. R. Sopka, “Numerical methods for nonlinear
future, we will look for more relaxed conditions to establish Volterra integro-differential equations,” SIAM Journal on
new convergence results. In addition, as we mentioned in Numerical Analysis, vol. 11, pp. 826–846, 1974.
the introduction section, there are many efficient numerical [4] P. Linz, “Linear multistep methods for Volterra integro-dif-
methods to solve stochastic ordinary differential equations ferential equations,” Journal of the Association for Computing
in the existing literature, such as stochastic linear multistep Machinery, vol. 16, pp. 295–301, 1969.
method and stochastic Runge-Kutta methods. Therefore, the [5] X. R. Mao, “Stability of stochastic integro differential equations,”
adaptation of such methods for stochastic Volterra integro- Stochastic Analysis and Applications, vol. 18, no. 6, pp. 1005–1017,
differential equations is worth studying in the future. 2000.
Abstract and Applied Analysis 13

[6] X. R. Mao and M. Riedle, “Mean square stability of stochastic

Volterra integro-differential equations,” Systems and Control
Letters, vol. 55, no. 6, pp. 459–465, 2006.
[7] J. A. D. Appleby, “pth mean integrability and almost sure
asymptotic stability of solutions of It 𝑜̂-volterra equations,”
Journal of Integral Equations and Applications, vol. 15, no. 4, pp.
321–341, 2003.
[8] E. Buckwar and R. Winkler, “Multistep methods for SDEs and
their application to problems with small noise,” SIAM Journal
on Numerical Analysis, vol. 44, no. 2, pp. 779–803, 2006.
[9] E. Buckwar and R. Winkler, “Improved linear multi-step meth-
ods for stochastic ordinary differential equations,” Journal of
Computational and Applied Mathematics, vol. 205, no. 2, pp.
912–922, 2007.
[10] K. Burrage, P. M. Burrage, and T. Tian, “Numerical methods
for strong solutions of stochastic differential equations: an
overview,” Proceedings of The Royal Society of London A: Math-
ematical, Physical and Engineering Sciences, vol. 460, no. 2041,
pp. 373–402, 2004.
[11] T. H. Tian and K. Burrage, “Implicit Taylor methods for stiff
stochastic differential equations,” Applied Numerical Mathemat-
ics, vol. 38, no. 1-2, pp. 167–185, 2001.
[12] P. Burrage, Runge-Kutta methods for stochastic differential
equations [Ph.D. thesis], University of Queensland, Brisbane,
Australia, 1999.
[13] P. E. Kloeden and E. Platen, Numerical Solution of Stochastic
Differential Equations, Springer, Berlin, Germany, 1992.
[14] J. Golec and S. Sathananthan, “Sample path approximation
for stochastic integro-differential equations,” Stochastic Analysis
and Applications, vol. 17, no. 4, pp. 579–588, 1999.
[15] J. Golec and S. Sathananthan, “Strong approximations of sto-
chastic integro-differential equations,” Dynamics of Continuous,
Discrete & Impulsive Systems B: Applications & Algorithms, vol.
8, no. 1, pp. 139–151, 2001.
[16] L. E. Shaikhet and J. A. Roberts, “Reliability of difference ana-
logues to preserve stability properties of stochastic Volterra
integro-differential equations,” Advances in Difference Equa-
tions, vol. 2006, Article ID 73897, 22 pages, 2006.
[17] D. J. Higham, “Mean-square and asymptotic stability of the
stochastic theta method,” SIAM Journal on Numerical Analysis,
vol. 38, no. 3, pp. 753–769, 2000.
[18] D. J. Higham, “A-stability and stochastic mean-square stability,”
BIT Numerical Mathematics, vol. 40, no. 2, pp. 404–409, 2000.
[19] X. Mao, Stochastic Differential Equations and Applications,
Horwood, Chichester, UK, 2nd edition, 2008.
[20] C. T. H. Baker and E. Buckwar, “Continuous 𝜃-methods for
the stochastic pantograph equation,” Electronic Transactions on
Numerical Analysis, vol. 11, pp. 131–151, 2000.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 385103, 10 pages
http://dx.doi.org/10.1155/2014/385103

Research Article
On the Number of Limit Cycles of a Piecewise Quadratic
Near-Hamiltonian System

Jing Tian
Department of Mathematics, Shanghai Normal University, Shanghai 200234, China

Correspondence should be addressed to Jing Tian; tj0634@163.com

Received 24 June 2014; Accepted 23 July 2014; Published 14 October 2014

Academic Editor: Tonghua Zhang

Copyright © 2014 Jing Tian. This is an open access article distributed under the Creative Commons Attribution License, which
permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

This paper is concerned with the problem for the maximal number of limit cycles for a quadratic piecewise near-Hamiltonian
system. By using the method of the first order Melnikov function, we find that it can have 8 limit cycles.

1. Introduction and Main Results limit cycles, which implies that piecewise quadratic near-
Hamiltonian systems can have 5 limit cycles. The authors
Recently, piecewise smooth systems attracted many research- of [15] also obtained this result by perturbed quadratic
ers’ attention since many real processes and different modern polynomial differential systems containing an isochronous
devices can be modeled by them; see [1, 2] and references center with piecewise quadratic polynomials of degree 2.
therein. Due to their nonsmoothness, these systems can In this paper, we mainly study the problem for the
have richer dynamical phenomena than the smooth ones maximal number of a quadratic near-Hamiltonian system.
(see [3–5] and the references cited therein). For instance, That is to say, we consider a system of the form
nonsmooth system can have the sliding phenomena and in
[6, 7] Giannakopoulos and Pliete have studied the existence 𝑥̇= 𝑦 + 𝜀𝑝 (𝑥, 𝑦) , 𝑦̇= −𝑔 (𝑥) + 𝜀𝑞 (𝑥, 𝑦) , (1)
of sliding cycles and sliding homoclinic cycles for a planar where 𝜀 > 0 is a small parameter,
relay control feedback systems.
As we have seen, studying the existence and number of 𝑎2 𝑥2 + 𝑎1 𝑥 + 𝑎0 , 𝑥 < 0,
𝑔 (𝑥) = { 2 (2)
limit cycles is one of the main problems for piecewise smooth 𝑏2 𝑥 + 𝑏1 𝑥 + 𝑏0 , 𝑥 > 0
systems; see [8–16]. Limit cycles of piecewise smooth linear
with 𝑎2 𝑏2 ≠0, and
differential systems defined on two half-planes separated
by a straight line 𝑥 = 0 or 𝑦 = 0 have been studied 2
{
{
recently in [8–11], from which one can find that 3 limit {
{ ∑ 𝑎𝑖𝑗− 𝑥𝑖 𝑦𝑗 , 𝑥 < 0,
{
{𝑖+𝑗=0
cycles can appear for piecewise smooth linear systems. From 𝑝 (𝑥, 𝑦) = { 2
{
{
[12–14], we can know that piecewise linear and quadratic {
{ + 𝑖 𝑗
near-Hamiltonian systems can have 2 and 3 limit cycles, { ∑ 𝑎𝑖𝑗 𝑥 𝑦 , 𝑥 > 0,
respectively. In fact, these papers investigated the problem {𝑖+𝑗=0
(3)
for limit cycle bifurcations of piecewise linear Hamiltonian 2
{
{
systems under piecewise polynomial perturbations of degree {
{ ∑ 𝑏𝑖𝑗− 𝑥𝑖 𝑦𝑗 , 𝑥 < 0,
{
{𝑖+𝑗=0
𝑛, obtaining some new and interesting results. Furthermore, 𝑞 (𝑥, 𝑦) = { 2
the authors of [16] studied a piecewise quadratic Hamiltonian {
{
{
{ + 𝑖 𝑗
system (one side linear and another side quadratic) perturbed { ∑ 𝑏𝑖𝑗 𝑥 𝑦 , 𝑥 > 0.
inside the class of piecewise polynomial differential systems {𝑖+𝑗=0
of degree 𝑛 and achieved that it can have 2𝑛 + [(𝑛 + 1)/2] Then, our main results can be stated as follows.
2 Abstract and Applied Analysis

S C S
C

(a) 𝑎 < 0 (b) 𝑎 > 0

Figure 1: Level curves of (1/2)𝑦2 + (1/3)𝑎𝑥3 + (1/2)𝑏𝑥2 + 𝑐𝑥 = ℎ.

Theorem 1. There exists a system of form (1) which has 8 limit For systems (4a)|𝜀=0 and (4b)|𝜀=0 , we make the following
cycles. assumption:

In the next section, the proof of Theorem 1 is presented. (H) 𝑎12 > 4𝑎2 𝑎0 , 𝑏12 > 4𝑏2 𝑏0 .
Then, under the condition (H), for 𝜀 = 0, system (4a)
2. Proof of Theorem 1 (resp., (4b)) has a hyperbolic saddle 𝑆1 = (𝛼𝑆1 , 0) (resp.,
𝑆2 = (𝛼𝑆2 , 0)) and an elementary center 𝐶1 = (𝛼𝐶1 , 0) (resp.,
In this section, we divide the proof into two subsections: one 𝐶2 = (𝛼𝐶2 , 0)), where
section is to present some preliminary knowledge, which is
useful to verify our main result; another section is to give the −𝑎1 − √𝑎12 − 4𝑎2 𝑎0 −𝑏1 − √𝑏12 − 4𝑏2 𝑏0
process of the proof of Theorem 1. 𝛼𝑆1 = , 𝛼𝑆2 = ,
2𝑎2 2𝑏2
2.1. Preliminaries. Clearly, system (1) has the following two −𝑎1 + √𝑎12 − 4𝑎2 𝑎0 −𝑏1 + √𝑏12 − 4𝑏2 𝑏0
subsystems: 𝛼𝐶1 = , 𝛼𝐶2 = .
2𝑎2 2𝑏2
2
(6)
𝑥̇= 𝑦 + 𝜀 ∑ 𝑎𝑖𝑗− 𝑥𝑖 𝑦𝑗 ,
𝑖+𝑗=0 We will assume that
(4a)
2 𝑎2 < 0, (7)
𝑦̇= − (𝑎2 𝑥2 + 𝑎1 𝑥 + 𝑎0 ) + 𝜀 ∑ 𝑏𝑖𝑗− 𝑥𝑖 𝑦𝑗 ,
𝑖+𝑗=0 since, for 𝑎2 > 0, we can make a variable transformation 𝑥 →
−𝑥, 𝑦 → 𝑦, together with a time rescaling 𝑡 → −𝑡 such
2
that the resulting system satisfies (7). Further, note that there
𝑥̇= 𝑦 + 𝜀 ∑ 𝑎𝑖𝑗+ 𝑥𝑖 𝑦𝑗 ,
exist 2 possible cases for level curves of 𝐻(𝑥, 𝑦) = (1/2)𝑦2 +
𝑖+𝑗=0
(4b) (1/3)𝑎𝑥3 + (1/2)𝑏𝑥2 + 𝑐𝑥 = ℎ on the plane, where 𝑎 ≠0,
2 𝑏2 > 4𝑎𝑐; see Figure 1.
𝑦̇= − (𝑏2 𝑥2 + 𝑏1 𝑥 + 𝑏0 ) + 𝜀 ∑ 𝑏𝑖𝑗+ 𝑥𝑖 𝑦𝑗 , We remark that, in Figure 1, 𝐶 = ((−𝑏+ √𝑏2 − 4𝑎𝑐)/2𝑎, 0),
𝑖+𝑗=0
𝑆 = ((−𝑏 − √𝑏2 − 4𝑎𝑐)/2𝑎, 0). Then, one can obtain 9 possible
which are called left and right subsystems, respectively. phase portraits of system (1)|𝜀=0 under (H) and (7) when it
And systems (4a)|𝜀=0 and (4b)|𝜀=0 are Hamiltonian with has a generalized closed orbit and two hyperbolic saddles; see
Hamiltonian functions, respectively, Figure 2. For system (1)|𝜀=0 , a generalized closed orbit passes
1 1 1 through the 𝑦-axis two times.
𝐻− (𝑥, 𝑦) = 𝑦2 + 𝑎2 𝑥3 + 𝑎1 𝑥2 + 𝑎0 𝑥, Now, we study limit cycle bifurcations on system (1). For
2 3 2
(5) simplicity, we consider the case
1 1 1
𝐻 (𝑥, 𝑦) = 𝑦2 + 𝑏2 𝑥3 + 𝑏1 𝑥2 + 𝑏0 𝑥.
+
𝛼𝑆1 < 0 < 𝛼𝑆2 < 𝛼𝐶2 , 𝐻− (𝑆1 ) = 𝐻+ (𝑆2 ) ≜ ℎ1 ; (8)
2 3 2
Abstract and Applied Analysis 3

y y y

x x x
0 0 0

(a) 𝛼𝑆1 < 0 < 𝛼𝐶2 < 𝛼𝑆2 , 𝐻− (𝑆1 ) > (b) 𝛼𝑆1 < 0 < 𝛼𝐶2 < 𝛼𝑆2 , 𝐻− (𝑆1 ) = (c) 𝛼𝑆1 < 0 < 𝛼𝐶2 < 𝛼𝑆2 , 𝐻− (𝑆1 ) <
𝐻+ (𝑆2 ) 𝐻+ (𝑆2 ) 𝐻+ (𝑆2 )
y y y

0 x x x
0 0

(d) 𝛼𝑆1 < 0, 𝛼𝐶2 ≤ 0 < 𝛼𝑆2 , 𝐻− (𝑆1 ) > (e) 𝛼𝑆1 < 0, 𝛼𝐶2 ≤ 0 < 𝛼𝑆2 , 𝐻− (𝑆1 ) = (f) 𝛼𝑆1 < 0, 𝛼𝐶2 ≤ 0 < 𝛼𝑆2 , 𝐻− (𝑆1 ) <
𝐻+ (𝑆2 ) 𝐻+ (𝑆2 ) 𝐻+ (𝑆2 )
y y y

0 x x x
0 0

(g) 𝛼𝑆1 < 0 < 𝛼𝑆2 < 𝛼𝐶2 , 𝐻− (𝑆1 ) > (h) 𝛼𝑆1 < 0 < 𝛼𝑆2 < 𝛼𝐶2 , 𝐻− (𝑆1 ) = (i) 𝛼𝑆1 < 0 < 𝛼𝑆2 < 𝛼𝐶2 , 𝐻− (𝑆1 ) <
𝐻+ (𝑆2 ) 𝐻+ (𝑆2 ) 𝐻+ (𝑆2 )

Figure 2: Possible portraits of (1)|𝜀=0 with two saddles and a generalized closed orbit.

that is, 𝐿 3ℎ = 𝐿−3ℎ ∪ 𝐿+3ℎ = {(𝑥, 𝑦) | 𝐻− (𝑥, 𝑦) = ℎ, 𝛼𝑆1 < 𝑥 < 0}

𝑎12 > 4𝑎2 𝑎0 , 𝑏12 > 4𝑏2 𝑏0 , 𝑎2 < 0, 𝑏2 > 0, ∪ {(𝑥, 𝑦) | 𝐻+ (𝑥, 𝑦) = ℎ, 0 < 𝑥 < 𝛼𝑆2 } ,
𝑎1 < 0, 𝑏1 < 0, 𝑎0 < 0, 𝑏0 > 0,
0 < ℎ < ℎ1 ,
3/2
𝑎13 − 6𝑎0 𝑎1 𝑎2 + (𝑎12 − 4𝑎0 𝑎2 )
(9) 𝐿 ℎ = 𝐿−ℎ ∪ 𝐿+ℎ = {(𝑥, 𝑦) | 𝐻− (𝑥, 𝑦) = ℎ, 𝑥 < 0}
𝑎22
3/2
∪ {(𝑥, 𝑦) | 𝐻+ (𝑥, 𝑦) = ℎ, 𝑥 > 0} , ℎ > ℎ1 ,
𝑏13 − 6𝑏0 𝑏1 𝑏2 + (𝑏12 − 4𝑏0 𝑏2 ) (10)
= .
𝑏22
where 𝐻− (𝑥0− , 0) = ℎ1 , 𝐻+ (𝑥0+ , 0) = ℎ1 and 𝐻− (𝐶1 ) = ℎ0− ,
Then, the phase portrait of system (1)|𝜀=0 is Figure 2(h), and 𝐻+ (𝐶2 ) = ℎ0+ . Further, we have
there are four families of periodic orbits given by
lim 𝐿 1ℎ = 𝐿 1 , lim 𝐿 1ℎ = 𝐶1 , lim 𝐿 2ℎ = 𝐿 2 ,
ℎ → ℎ1 ℎ → ℎ0− ℎ → ℎ1
𝐿 1ℎ = {(𝑥, 𝑦) | 𝐻− (𝑥, 𝑦) = ℎ, 𝑥0− < 𝑥 < 𝛼𝑆1 } , ℎ0− < ℎ < ℎ1 ,
lim 𝐿 1ℎ = 𝐶2 , lim 𝐿−3ℎ = 𝐿− ,
𝐿 2ℎ = {(𝑥, 𝑦) | 𝐻+ (𝑥, 𝑦) = ℎ, 𝛼𝑆2 < 𝑥 < 𝑥0+ } , ℎ0+ < ℎ < ℎ1 , ℎ → ℎ0+ ℎ → ℎ1
4 Abstract and Applied Analysis

2
y
𝑀 (ℎ) = 𝑀− (ℎ) + 𝑀+ (ℎ) = ∫ ∑ 𝑏𝑖𝑗− 𝑥𝑖 𝑦𝑗 𝑑𝑥
𝐿−ℎ 𝑖+𝑗=0

2 2
− ∑ 𝑎𝑖𝑗− 𝑥𝑖 𝑦𝑗 𝑑𝑦 + ∫ ∑ 𝑏𝑖𝑗+ 𝑥𝑖 𝑦𝑗 𝑑𝑥
L−h L+h 𝐿+ℎ 𝑖+𝑗=0
𝑖+𝑗=0
+
−
L L
2
L1 L2 − ∑ 𝑎𝑖𝑗+ 𝑥𝑖 𝑦𝑗 𝑑𝑦, ℎ ∈ (ℎ1 , +∞) ,
L 1h L−3h L 2h 𝑖+𝑗=0
L+3h
x (12)
C1 0 C2
S1 S2
where 𝐿 1ℎ , 𝐿 2ℎ , 𝐿−3ℎ , 𝐿+3ℎ , 𝐿−ℎ , and 𝐿+ℎ are given in (10).
Now, we deduce the expansion of 𝑀𝑖 (ℎ), 𝑖 = 1, 2, 3, and
𝑀(ℎ) near ℎ = ℎ1 in the above. First, we have the following.

Lemma 2. The function 𝑀3− (ℎ) in (12) has the following

expression:

𝑀3− (ℎ) = 𝑐0 + 𝑐1 (ℎ − ℎ1 ) ln (ℎ1 − ℎ) + 𝑐2 (ℎ − ℎ1 )

Figure 3: Phase portrait of system (1)|𝜀=0 under (9).
2 2
+ 𝑐3 (ℎ − ℎ1 ) ln (ℎ1 − ℎ) + 𝑂 ((ℎ − ℎ1 ) ) , (13)

0 < ℎ1 − ℎ ≪ 1,
lim 𝐿+3ℎ = 𝐿+ , lim 𝐿−3ℎ = lim 𝐿+3ℎ = (0, 0) ,
ℎ → ℎ1 ℎ→0 ℎ→0 where
− − − −
lim 𝐿−ℎ = 𝐿 1 ∪ 𝐿− , lim 𝐿+ℎ = 𝐿 2 ∪ 𝐿+ ; 𝑐0 = ∫ [ (𝑎10 + 𝑏01 ) 𝑦 + (𝑏11 + 2𝑎20 ) 𝑥𝑦
ℎ → ℎ1 ℎ → ℎ1 𝐿−
(11) 1 − −
+ (𝑎 + 2𝑏02 ) 𝑦2 ] 𝑑𝑥 − 2𝑎00
−
√2ℎ1
2 11
see Figure 3. From (11), one can know that, as ℎ approaches 4 −
ℎ1 , the limit of 𝐿 ℎ is 𝐿 1 ∪ 𝐿− ∪ 𝐿 2 ∪ 𝐿+ denoted by 𝐿, which is − ℎ1 √2ℎ1 𝑎02 ,
3
a generalized polycycle with two hyperbolic saddles.
1 − − − −
Then, from Theorem 1.1 of [17], for system (1), we have 𝑐1 = − 1/4
[(𝑎10 + 𝑏01 ) + 𝛼𝑆1 (2𝑎20 + 𝑏11 )] ,
four first order Melnikov functions as follows: (𝑎12 − 4𝑎0 𝑎2 )

− − − − − −
2 2 𝑐2 |𝑐1 =0 = ∫ [(𝑎10 + 𝑏01 ) + (𝑏11 + 2𝑎20 ) 𝑥 + (𝑎11 + 2𝑏02 ) 𝑦] 𝑑𝑡
𝐿−
𝑀1 (ℎ) = ∮ ∑ 𝑏𝑖𝑗− 𝑥𝑖 𝑦𝑗 𝑑𝑥 − ∑ 𝑎𝑖𝑗− 𝑥𝑖 𝑦𝑗 𝑑𝑦,
𝐿 1ℎ 𝑖+𝑗=0 𝑖+𝑗=0 √2ℎ1 − −
− 𝑎 − 2√2ℎ1 𝑎02 ,
ℎ1 00
ℎ ∈ (ℎ0− , ℎ1 ) ,
𝑎2 − −
𝑐3 = 5/4
(2𝑎20 + 𝑏11 ).
2 2 2(𝑎12 − 4𝑎0 𝑎2 )
𝑀2 (ℎ) = ∮ ∑ 𝑏𝑖𝑗+ 𝑥𝑖 𝑦𝑗 𝑑𝑥 − ∑ 𝑎𝑖𝑗+ 𝑥𝑖 𝑦𝑗 𝑑𝑦, (14)
𝐿 2ℎ 𝑖+𝑗=0 𝑖+𝑗=0
Proof. Denote by 𝑈 a disk of diameter 𝜖0 > 0 with its center
ℎ ∈ (ℎ0+ , ℎ1 ) , at 𝑆1 . Then, for 𝜖0 > 0 small, rewrite 𝑀3− (ℎ) = 𝐼1 (ℎ) + 𝐼2 (ℎ),
where
2 2 2
𝑀3 (ℎ) = 𝑀3− (ℎ) + 𝑀3+ (ℎ) = ∫ ∑ 𝑏𝑖𝑗− 𝑥𝑖 𝑦𝑗 𝑑𝑥 𝐼1 (ℎ) = ∫ − ∑ 𝑏𝑖𝑗− 𝑥𝑖 𝑦𝑗 𝑑𝑥 − ∑ 𝑎𝑖𝑗− 𝑥𝑖 𝑦𝑗 𝑑𝑦,
𝐿−3ℎ 𝑖+𝑗=0 𝐿3ℎ 𝑖+𝑗=0 𝑖+𝑗=0
(15)
2 2 2 2
− ∑ 𝑎𝑖𝑗− 𝑥𝑖 𝑦𝑗 𝑑𝑦 +∫ ∑ 𝑏𝑖𝑗+ 𝑥𝑖 𝑦𝑗 𝑑𝑥 𝐼2 (ℎ) = ∫ ∑ 𝑏𝑖𝑗− 𝑥𝑖 𝑦𝑗 𝑑𝑥 − ∑ 𝑎𝑖𝑗− 𝑥𝑖 𝑦𝑗 𝑑𝑦
𝐿+3ℎ 𝑖+𝑗=0 ̃−
𝐿
𝑖+𝑗=0 3ℎ 𝑖+𝑗=0 𝑖+𝑗=0

2 with
− ∑ 𝑎𝑖𝑗+ 𝑥𝑖 𝑦𝑗 𝑑𝑦, ℎ ∈ (0, ℎ1 ) , −
̃ − = 𝐿− \ 𝐿− .
𝑖+𝑗=0
𝐿3ℎ = 𝐿−3ℎ ∩ 𝑈, 𝐿 3ℎ 3ℎ 3ℎ
(16)
Abstract and Applied Analysis 5

Note that 𝐼2 (ℎ) is a 𝐶∞ function for ℎ1 − ℎ > 0 small. Thus, it Thus, by the above discussion, one can obtain the expres-
suffices to investigate the expansion of 𝐼1 (ℎ) for 0 < ℎ1 − ℎ ≪ sion of (13) and the formulas for 𝑐1 and 𝑐3 . Now, we give the
1. To do this, we make a variable transformation expressions of 𝑐0 and 𝑐2 in (13).
On the curve 𝐿 3ℎ intersect 𝑦-axis with two points denoted
1/4
𝑢 = (𝑎12 − 4𝑎0 𝑎2 ) (𝑥 − 𝛼𝑆1 ) , V = 𝑦, by 𝐴 1 (ℎ) = (0, −√2ℎ) and 𝐴(ℎ) = (0, √2ℎ). Then, from (12),
(17) we have
1/4
𝜏= (𝑎12 − 4𝑎0 𝑎2 ) 𝑡. 2 2
𝑀3− (ℎ) = ∫ ∑ 𝑏𝑖𝑗− 𝑥𝑖 𝑦𝑗 𝑑𝑥 − ∑ 𝑎𝑖𝑗− 𝑥𝑖 𝑦𝑗 𝑑𝑦
𝐴 1̂
(ℎ)𝐴(ℎ) 𝑖+𝑗=0 𝑖+𝑗=0 (24)
Then system (4a) becomes

𝑢̇= V + 𝜀𝑝̃− (𝑢, V) , = 𝑁1 (ℎ) + 𝑁2 (ℎ) ,

where
𝑎2 𝑢2 (18)
V̇= 𝑢 − 3/4
𝑞− (𝑢, V) ,
+ 𝜀̃ 2
(𝑎12 − 4𝑎0 𝑎2 ) 𝑁1 (ℎ) = ∫ 󳨀󳨀󳨀󳨀󳨀󳨀󳨀󳨀󳨀→ ∑ 𝑏𝑖𝑗− 𝑥𝑖 𝑦𝑗 𝑑𝑥
𝐴 1̂
(ℎ)𝐴(ℎ)∪𝐴(ℎ)𝐴 1 (ℎ) 𝑖+𝑗=0
where
2
2 2 𝑖 − ∑ 𝑎𝑖𝑗− 𝑥𝑖 𝑦𝑗 𝑑𝑦, (25)
𝑢
𝑝̃− (𝑢, V) = ∑ 𝑎̃𝑖𝑗− 𝑢𝑖 V𝑗 = ∑ 𝑎𝑖𝑗− [ 1/4
+ 𝛼𝑆1 ] V , 𝑗 𝑖+𝑗=0
𝑖+𝑗=0 𝑖+𝑗=0 (𝑎12 − 4𝑎0 𝑎2 )
− − − 2
(19) 𝑁2 (ℎ) = ∫󳨀󳨀󳨀󳨀󳨀󳨀󳨀󳨀󳨀→ (𝑎00 + 𝑎01 𝑦 + 𝑎02 𝑦 ) 𝑑𝑦.
𝐴(ℎ)𝐴 1 (ℎ)
2
𝑞̃− (𝑢, V) = ∑ ̃𝑏𝑖𝑗− 𝑢𝑖 V𝑗 By Green’s formula two times to 𝑁1 (ℎ), we obtain
𝑖+𝑗=0
− − − −
𝑁1 (ℎ) = ∫ 󳨀󳨀󳨀󳨀󳨀󳨀󳨀󳨀󳨀→ [(𝑎10 + 𝑏01 ) + (𝑏11 + 2𝑎20 )𝑥
𝑖 Int.(𝐴 1̂
(ℎ)𝐴(ℎ)∪𝐴(ℎ)𝐴 1 (ℎ))
2 𝑏𝑖𝑗− 𝑢
= ∑ [ + 𝛼𝑆1 ] V𝑗 . − −
2
𝑖+𝑗=0 (𝑎1 − 4𝑎0 𝑎2 )
1/4
(𝑎12 − 4𝑎0 𝑎2 )
1/4 + (𝑎11 + 2𝑏02 ) 𝑦] 𝑑𝑥 𝑑𝑦
(20) − − − −
=∮ 󳨀󳨀󳨀󳨀󳨀󳨀󳨀󳨀󳨀→ [ (𝑎10 + 𝑏01 ) 𝑦 + (𝑏11 + 2𝑎20 ) 𝑥𝑦
𝐴 1̂
(ℎ)𝐴(ℎ)∪𝐴(ℎ)𝐴 1 (ℎ)
For 𝜀 = 0, the Hamiltonian function of (18) has the form
1 − −
𝑎2 + (𝑎11 + 2𝑏02 ) 𝑦2 ] 𝑑𝑥
̃ (𝑢, V) = 1 (V2 − 𝑢2 ) +
𝐻 𝑢3 + ℎ1 (21) 2
2 2 3/4
3(𝑎1 − 4𝑎0 𝑎2 )
− − − −
=∫ [ (𝑎10 + 𝑏01 ) 𝑦 + (𝑏11 + 2𝑎20 ) 𝑥𝑦
and the corresponding first order Melinikov function can 𝐿−3ℎ
̃
be denoted by 𝑀(ℎ). By Remark 3.1.4 of [18], we can know 1 − −
̃
that 𝑀3− (ℎ) = 𝑀(ℎ), and then, from Corollary 3.2.2 and the + (𝑎11 + 2𝑏02 ) 𝑦2 ] 𝑑𝑥.
2
Theorem 3.2.1 of [18], 𝐼1 can be written as (26)
2 Further, one has
𝐼1 (ℎ) = 𝑐1 (ℎ − ℎ1 ) ln (ℎ1 − ℎ) + 𝑐3 (ℎ − ℎ1 ) ln (ℎ1 − ℎ)
− − − 2
+ 𝑂 ((ℎ − ℎ1 ) ) ,
2
0 < ℎ1 − ℎ ≪ 1, 𝑁2 (ℎ) = ∫󳨀󳨀󳨀󳨀󳨀󳨀󳨀󳨀󳨀→ (𝑎00 + 𝑎01 𝑦 + 𝑎02 𝑦 ) 𝑑𝑦
𝐴(ℎ)𝐴 1 (ℎ)
(22)
−√2ℎ
− − − 2 (27)
where =∫ (𝑎00 + 𝑎01 𝑦 + 𝑎02 𝑦 ) 𝑑𝑦
√2ℎ

1 4
𝑐1 = − (̃−
𝑎10 + ̃𝑏01
−
)=− 1/4
−√
= −2𝑎00 2ℎ − ℎ√2ℎ𝑎02
−
.
(𝑎12 − 4𝑎0 𝑎2 ) 3
− − − − Thus, we have
× [(𝑎10 + 𝑏01 ) + 𝛼𝑆1 (2𝑎20 + 𝑏11 )] ,
3 𝑎2 1 𝑀3− (ℎ) = ∫ −
[ (𝑎10 −
+ 𝑏01 −
) 𝑦 + (𝑏11 −
+ 2𝑎20 ) 𝑥𝑦
𝑐3 = × × 𝑎20 + ̃𝑏11 )
(2̃ 𝐿−3ℎ
2 3(𝑎2 − 4𝑎 𝑎 )3/4 (𝑎2 − 4𝑎 𝑎 )2/4
1 0 2 1 0 2 1 − −
𝑎2 + (𝑎 + 2𝑏02 ) 𝑦2 ] 𝑑𝑥 (28)
= −
(2𝑎20 + −
𝑏11 ). 2 11
5/4
2(𝑎12 − 4𝑎0 𝑎2 ) −√ 4
(23) − 2𝑎00 2ℎ − ℎ√2ℎ𝑎02
−
.
3
6 Abstract and Applied Analysis

Then, let ℎ → ℎ1 . Then, by (11), we can obtain the expression By applying Theorem 3.2.3 of [18], together with the proof
of 𝑐0 in (13). Further, differentiating (13) with respect to ℎ, of Lemma 2, we have the following.
together with Lemma 3.1.2 of [18], we obtain
Lemma 4. The functions 𝑀𝑖 (ℎ), 𝑖 = 1, 2, given in (12) can be
𝑑 − expressed as
𝑀 (ℎ) = ∫ (𝑝𝑥 + 𝑞𝑦 ) |𝜀=0 𝑑𝑡
𝑑ℎ 3 𝐿−3ℎ
𝑀1 (ℎ) = 𝑐01 + 𝑐1 (ℎ − ℎ1 ) ln (ℎ1 − ℎ) + 𝑐21 (ℎ − ℎ1 )
− − − −
=∫ [(𝑎10 + 𝑏01 ) + (𝑏11 + 2𝑎20 )𝑥 2
𝐿−3ℎ + 𝑐3 (ℎ − ℎ1 ) ln (ℎ1 − ℎ)
2
√2ℎ − + 𝑂 ((ℎ − ℎ1 ) ) , 0 < ℎ1 − ℎ ≪ 1,
−
+ (𝑎11 −
+ 2𝑏02 ) 𝑦] 𝑑𝑡 − 𝑎 − 2√2ℎ𝑎02
−
ℎ 00 (33)
𝑀2 (ℎ) = 𝑐02 + 𝑐̃1 (ℎ − ℎ1 ) ln (ℎ1 − ℎ) + 𝑐22 (ℎ − ℎ1 )
= 𝑐1 ln (ℎ1 − ℎ) + 𝑐1 + 𝑐2
2
+ 𝑐̃3 (ℎ − ℎ1 ) ln (ℎ1 − ℎ)
+ 𝑂 ((ℎ − ℎ1 ) ln (ℎ1 − ℎ)) ,
2
(29) + 𝑂 ((ℎ − ℎ1 ) ) , 0 < ℎ1 − ℎ ≪ 1,
which, as 𝑐1 = 0, gives that where 𝑐1 , 𝑐3 , 𝑐̃1 , 𝑐̃3 are given in Lemmas 2 and 3, and
− − − − − − 2 2
∫ [(𝑎10 + 𝑏01 ) + (𝑏11 + 2𝑎20 ) 𝑥 + (𝑎11 + 2𝑏02 ) 𝑦] 𝑑𝑡
𝐿−3ℎ 𝑐01 = ∮ ∑ 𝑏𝑖𝑗− 𝑥𝑖 𝑦𝑗 𝑑𝑥 − ∑ 𝑎𝑖𝑗− 𝑥𝑖 𝑦𝑗 𝑑𝑦,
𝐿 1 𝑖+𝑗=0 𝑖+𝑗=0
√2ℎ − (30)
− 𝑎 − 2√2ℎ𝑎02
−
2 2
ℎ 00 𝑐02 = ∮ ∑ 𝑏𝑖𝑗+ 𝑥𝑖 𝑦𝑗 𝑑𝑥 − ∑ 𝑎𝑖𝑗+ 𝑥𝑖 𝑦𝑗 𝑑𝑦,
𝐿 2 𝑖+𝑗=0
= 𝑐2 + 𝑂 ((ℎ − ℎ1 ) ln (ℎ1 − ℎ)) . 𝑖+𝑗=0

Let ℎ → ℎ1 in the above. Then we can easily obtain the −

𝑐21 |𝑐1 =0 = ∮ [(𝑎10 −
+ 𝑏01 −
) + (2𝑎20 −
+ 𝑏11 )𝑥 (34)
formula for 𝑐2 in (13). This ends the proof. 𝐿1

In the same way, one can obtain the expressions of 𝑀3+ (ℎ). −
+ (𝑎11 −
+ 2𝑏02 ) 𝑦] 𝑑𝑡|𝑐1 =0 ,

Lemma 3. For 𝑀3− (ℎ) in (12), one has + + + +

𝑐22 |𝑐̃1 =0 = ∮ [(𝑎10 + 𝑏01 ) + (2𝑎20 + 𝑏11 )𝑥
𝐿2
𝑀3+ (ℎ) = 𝑐̃0 + 𝑐̃1 (ℎ − ℎ1 ) ln (ℎ1 − ℎ) + 𝑐̃2 (ℎ − ℎ1 )
+ +
2 2
+ (𝑎11 + 2𝑏02 ) 𝑦] 𝑑𝑡|𝑐̃1 =0 .
+ 𝑐̃3 (ℎ − ℎ1 ) ln (ℎ1 − ℎ) + 𝑂 ((ℎ − ℎ1 ) ) , (31)
By above three Lemmas 2, 3, and 4, together with (12), one
0 < ℎ1 − ℎ ≪ 1, can derive the following.
where Lemma 5. For 𝑀3 (ℎ) and 𝑀(ℎ), one has
+ + + +
𝑐̃0 = ∫ [ (𝑎10 + 𝑏01 )𝑦 + (𝑏11 + 2𝑎20 ) 𝑥𝑦 𝑀3 (ℎ) = (𝑐0 + 𝑐̃0 ) + (𝑐1 + 𝑐̃1 ) (ℎ − ℎ1 ) ln (ℎ1 − ℎ)
𝐿+
2
1 + + + (𝑐2 + 𝑐̃2 ) (ℎ − ℎ1 ) + (𝑐3 + 𝑐̃3 ) (ℎ − ℎ1 ) ln (ℎ1 − ℎ)
+ (𝑎11 + 2𝑏02 ) 𝑦2 ] 𝑑𝑥 + 2√2ℎ1 𝑎00
+
2 2
+ (𝑂(ℎ1 − ℎ) ) , 0 < ℎ1 − ℎ ≪ 1,
4 +
+ ℎ1 √2ℎ1 𝑎02 ,
3 𝑀 (ℎ) = (𝑐0 + 𝑐01 + 𝑐̃0 + 𝑐02 ) + 2 (𝑐1 + 𝑐̃1 )
1 + + + +
𝑐̃1 = − [(𝑎10 + 𝑏01 ) + 𝛼𝑆2 (2𝑎20 + 𝑏11 )] , × (ℎ − ℎ1 ) ln (ℎ − ℎ1 ) + (𝑐2 + 𝑐21 + 𝑐̃2 + 𝑐22 )
1/4
(𝑏12 − 4𝑏0 𝑏2 )
2
× (ℎ − ℎ1 ) + 2 (𝑐3 + 𝑐̃3 ) (ℎ − ℎ1 ) ln (ℎ − ℎ1 )
+ + + + + +
𝑐̃2 |𝑐̃1 =0 = ∫ [(𝑎10 + 𝑏01 ) + (𝑏11 + 2𝑎20 ) 𝑥 + (𝑎11 + 2𝑏02 ) 𝑦] 𝑑𝑡 2
𝐿+ + (𝑂(ℎ1 − ℎ) ) , 0 < ℎ − ℎ1 ≪ 1,
√2ℎ1 + +
+ 𝑎 + 2√2ℎ1 𝑎02 ,
ℎ1 00 (35)
𝑏2 + + where the corresponding coefficients are given in (14), (32), and
𝑐̃3 = 5/4
(2𝑎20 + 𝑏11 ). (34).
2(𝑏12 − 4𝑏0 𝑏2 )
(32) Then, using Lemmas 4 and 5, we can have the following.
Abstract and Applied Analysis 7

Theorem 6. Let (H) and (7) hold. Then if there exists 𝛿0 such Third, let 𝑐01 = 𝑐0 + 𝑐̃0 = 0, 0 < −𝑐02 ≪ −(𝑐1 + 𝑐̃1 ), implying
that
𝑀1 (ℎ) < 0, 𝑀2 (ℎ) < 0, 𝑀3 (ℎ) < 0,
𝑐01 (𝛿0 ) = 𝑐02 (𝛿0 ) = (𝑐0 + 𝑐̃0 ) (𝛿0 ) = (𝑐1 + 𝑐̃1 ) (𝛿0 )
for 0 < ℎ1 − ℎ ≪ 1, (40)
= (𝑐2 + 𝑐̃2 ) (𝛿0 ) = 0,
𝑀 (ℎ) < 0, for 0 < ℎ − ℎ1 ≪ 1.
(𝑐3 + 𝑐̃3 ) (𝛿0 ) < 0, 𝑐1 (𝛿0 ) < 0, 𝑐̃1 (𝛿0 ) > 0,
Then 𝑀2 (̃ℎ4 ) = 0, 𝑀(̃ℎ5 ) = 0 for some ̃ℎ4 < ℎ1 , ̃ℎ5 ∈ (ℎ1 , ̃ℎ3 ).
(𝑐1 + 𝑐̃1 ) (𝛿0 ) = 0, (𝑐21 + 𝑐̃22 ) (𝛿0 ) = 0, Fourth, let 𝑐0 + 𝑐̃0 = 0 and 0 < 𝑐01 ≪ −𝑐02 ; we have
𝜕 (𝑐01 , 𝑐02 , 𝑐0 + 𝑐̃0 , 𝑐1 + 𝑐̃1 , 𝑐2 + 𝑐̃2 )
rank = 5, 𝑀1 (ℎ) > 0, 𝑀2 (ℎ) < 0, 𝑀3 (ℎ) < 0,
𝜕𝛿
(36) for 0 < ℎ1 − ℎ ≪ 1, (41)
then system (1) can have 7 limit cycles near 𝐿. 𝑀 (ℎ) < 0, for 0 < ℎ − ℎ1 ≪ 1.

Proof. By the assumption and from the expansions in Lem- So there exists ̃ℎ6 < ℎ1 such that 𝑀1 (̃ℎ6 ) = 0.
mas 4 and 5 we have Finally, vary 0 < 𝑐0 + 𝑐̃0 ≪ 𝑐01 . In this case,
𝑀1 (ℎ) = 𝑐1 (𝛿0 ) (ℎ − ℎ1 ) ln (ℎ1 − ℎ) + 𝑂 (ℎ1 − ℎ) < 0,
𝑀1 (ℎ) > 0, 𝑀2 (ℎ) < 0, 𝑀3 (ℎ) > 0,
0 < ℎ1 − ℎ ≪ 1,
for 0 < ℎ1 − ℎ ≪ 1, (42)
𝑀2 (ℎ) = 𝑐̃1 (𝛿0 ) (ℎ − ℎ1 ) ln (ℎ1 − ℎ) + 𝑂 (ℎ − ℎ1 ) > 0,
𝑀 (ℎ) < 0, for 0 < ℎ − ℎ1 ≪ 1.
0 < ℎ1 − ℎ ≪ 1,
So there exists ̃ℎ7 ∈ (̃ℎ2 , ℎ1 ) such that 𝑀3 (̃ℎ7 ) = 0.
2
𝑀3 (ℎ) = (𝑐3 + 𝑐̃3 ) (𝛿0 ) (ℎ − ℎ1 ) ln (ℎ1 − ℎ) In a word, we have proved that if
2
+ (𝑂(ℎ1 − ℎ) ) > 0, 0 < ℎ1 − ℎ ≪ 1, 0 < 𝑐0 + 𝑐̃0 ≪ 𝑐01 ≪ −𝑐02 ≪ − (𝑐1 + 𝑐̃1 ) ≪ − (𝑐2 + 𝑐̃2 ) ≪ 1
(43)
2 2
𝑀 (ℎ) = 2 (𝑐3 + 𝑐̃3 ) (ℎ − ℎ1 ) ln (ℎ − ℎ1 ) + (𝑂(ℎ1 − ℎ) ) > 0,
then 𝑀1 , 𝑀2 , 𝑀3 , and 𝑀 have one zero, one zero, two zeros,
0 < ℎ − ℎ1 ≪ 1. and three zeros which implies that seven limit cycles can
(37) appear in the neighbourhood of 𝐿.

Note that 𝑐01 , 𝑐02 , 𝑐0 + 𝑐̃0 , 𝑐1 + 𝑐̃1 , and 𝑐2 + 𝑐̃2 can be taken as 2.2. Proofs. In this section, we will use Theorem 6 to prove
free parameters. Then we can vary them one by one to change our main result. For that purpose, let
the signs of 𝑀1 , 𝑀2 , 𝑀3 , and 𝑀.
First, let 𝑐01 = 𝑐02 = 𝑐0 +̃𝑐0 = 𝑐1 +̃𝑐1 = 0 and 0 < −(𝑐2 +̃𝑐2 ) ≪ 3 1
𝑎2 = −1, 𝑎1 = − , 𝑎0 = − ,
1. Then, in this case, we have 2 2
(44)
𝑀1 (ℎ) < 0, 𝑀2 (ℎ) > 0, 𝑀3 (ℎ) > 0, 11 41
𝑏2 = 1, 𝑏1 = − , 𝑏0 = .
3 48
for 0 < ℎ1 − ℎ ≪ 1, (38)
Then, the Hamiltonian functions in (5) can be rewritten as
𝑀 (ℎ) < 0, for 0 < ℎ − ℎ1 ≪ 1,
1 1 3 1
𝐻− (𝑥, 𝑦) = 𝑦2 − 𝑥3 − 𝑥2 − 𝑥,
which means that there exists a zero ̃ℎ1 with 0 < ̃ℎ1 − ℎ1 such 2 3 4 2
(45)
that 𝑀(̃ℎ1 ) = 0. 1 1 11 11
𝐻 (𝑥, 𝑦) = 𝑦2 + 𝑥3 − 𝑥2 + 𝑥.
+
Second, fix 𝑐01 = 𝑐02 = 𝑐0 +̃𝑐0 = 0 and vary 0 ≪ −(𝑐1 +̃𝑐1 ) ≪ 2 3 6 48
−(𝑐2 + 𝑐̃2 ), so that
And in this case
𝑀1 (ℎ) < 0, 𝑀2 (ℎ) > 0, 𝑀3 (ℎ) < 0, 1 1
𝑆1 = (− , 0) , 𝑆2 = ( , 0) , 𝐶1 = (−1, 0) ,
for 0 < ℎ1 − ℎ ≪ 1, (39) 2 4
(46)
41 5
𝑀 (ℎ) > 0, for 0 < ℎ − ℎ1 ≪ 1. 𝐶2 = ( , 0) , ℎ1 = .
12 48
Hence, there exist ℎ1 − ̃ℎ2 > 0, ̃ℎ3 ∈ (ℎ1 , ̃ℎ1 ) such that Under (44), it is not hard to obtain the expressions for 𝑐1 ,
𝑀3 (̃ℎ2 ) = 0, 𝑀(̃ℎ3 ) = 0. 𝑐̃1 , 𝑐3 , and 𝑐̃3 in Lemmas 2 and 3.
8 Abstract and Applied Analysis

Lemma 7. Under (44), for system (1), one has On the other hand, along the curve 𝐿 1 , we have (1/2)𝑦2 −
(1/3)𝑥3 − (3/4)𝑥2 − (1/2)𝑥 = 5/48, for −(5/4) ≤ 𝑥 ≤ −(1/2).
− − 1 − −
𝑐1 = −√2 [(𝑎10 + 𝑏01 )− (2𝑎20 + 𝑏11 )] , Then, (49) can be represented as
2
− − −1/2
𝑐3 = −√2 (2𝑎20 + 𝑏11 ), 𝑐01 = 2 ∫ −
[(𝑎10 −
+ 𝑏01 −
) + (𝑏11 −
+ 2𝑎20 ) 𝑥]
−5/4
6 1 (47) (50)
+ + + +
𝑐̃1 = −√ [(𝑎10 + 𝑏01 ) + (2𝑎20 + 𝑏11 )] , 5 2 3
19 4 × √ + 𝑥3 + 𝑥2 + 𝑥𝑑𝑥.
24 3 2
1 6 5/2 + +
𝑐̃3 = ( ) (2𝑎20 + 𝑏11 ).
2 19 By applying Maple, one can compute that
For the other expressions of coefficients in Lemmas 4 and −1/2
5 2 3 3√2
5, one has the following. ∫ √ + 𝑥3 + 𝑥2 + 𝑥𝑑𝑥 = ,
−5/4 24 3 2 40
Lemma 8. Under (44), for system (1), one has (51)
−1/2
5 2 3 39√2
∫ 𝑥√ + 𝑥3 + 𝑥2 + 𝑥𝑑𝑥 = − .
3√2 − − 39√2 − − −5/4 24 3 2 560
𝑐01 = (𝑎10 + 𝑏01 )− (2𝑎20 + 𝑏11 ),
20 280
− −
Inserting the above into (50) gives the expression of 𝑐01 .
𝑐21 = −3√2 (2𝑎20 + 𝑏11 ), Recall that 𝑐1 = 0 if and only if 𝑎10− −
+𝑏01 −
= (1/2)(2𝑎20 −
+𝑏11 ).
Thus, by (4a)|𝜀=0 , we rewrite 𝑐21 |𝑐1 =0 as
361√114 + + 29963√114 + +
𝑐02 = (𝑎10 + 𝑏01 )+ (2𝑎20 + 𝑏11 ),
180 5040 𝑐21 |𝑐1 =0
+ +
𝑐22 = √114 (2𝑎20 + 𝑏11 ), − − − − − −
[(1/2) (2𝑎20 + 𝑏11 ) + (2𝑎20 + 𝑏11 ) 𝑥 + (𝑎11 + 2𝑏02 ) 𝑦]
=∮ 𝑑𝑥
3√2 − − 39√2 5√30 − − 𝐿1 𝑦
𝑐0 = (𝑎10 + 𝑏01 )−( − ) (2𝑎20 + 𝑏11 )
20 280 168 1 1 + 2𝑥
− −
= (2𝑎20 + 𝑏11 )∮ 𝑑𝑥
√30 − 5√30 − 2 𝐿1 𝑦
− 𝑎 − 𝑎 ,
6 00 432 02 −1/2
1 + 2𝑥
− −
= (2𝑎20 + 𝑏11 )∫ 𝑑𝑥,
35 361 √ + + √(5/24) + (2/3) 𝑥3 + (3/2) 𝑥2 + 𝑥
𝑐̃0 = (− √30 + 114) (𝑎10 + 𝑏01 ) −5/4
9 180 (52)
29963 √ 730 √ + + √30 +
+( 114 − 30) (2𝑎20 + 𝑏11 )+ 𝑎 which, by Maple, yields the conclusion for 𝑐02 . Using a similar
5040 63 6 00 way, one can derive the formulas for 𝑐02 and 𝑐22 . Now, we
5√30 + deduce the formulas for 𝑐0 and 𝑐2 .
+ 𝑎 , Note that, on the curve 𝐿− : (1/2)𝑦2 − (1/3)𝑥3 − (3/4)𝑥2 −
432 02
(1/2)𝑥 = (5/48), −(1/2) < 𝑥 < 0, we have
− − 4√30 − √30 −
𝑐2 = (√30 − 3√2) (2𝑎20 + 𝑏11 )− 𝑎 − 𝑎 , 0
5 00 6 02 𝑐0 = 2 ∫ −
[(𝑎10 −
+ 𝑏01 −
) + (𝑏11 −
+ 2𝑎20 ) 𝑥]
−1/2
+ + 4√30 + √30 +
𝑐̃2 = (−2√30 + √114) (2𝑎20 + 𝑏11 )+ 𝑎 + 𝑎 .
5 00 6 02 5 2 3
(48) ×√ + 𝑥3 + 𝑥2 + 𝑥𝑑𝑥 (53)
24 3 2
Proof. Note that 𝑐01 = ∮𝐿 ∑2𝑖+𝑗=0 𝑏𝑖𝑗− 𝑥𝑖 𝑦𝑗 𝑑𝑥−∑2𝑖+𝑗=0 𝑎𝑖𝑗− 𝑥𝑖 𝑦𝑗 𝑑𝑦. √30 − 5√30 −
1 − 𝑎 − 𝑎 .
Then, by Green’s formula two times, we derive 6 00 432 02
− − − − By Maple again, we obtain that
𝑐01 = ∫ [(𝑎10 + 𝑏01 ) + (𝑏11 + 2𝑎20 )𝑥
Int.(𝐿 1 )

− −
0
5 2 3 3√2
+ (𝑎11 + 2𝑏02 ) 𝑦] 𝑑𝑥 𝑑𝑦 ∫ √ + 𝑥3 + 𝑥2 + 𝑥𝑑𝑥 = ,
−1/2 24 3 2 40
(49) (54)
− − − −
=∮ [ (𝑎10 + 𝑏01 )𝑦 + (𝑏11 + 2𝑎20 ) 𝑥𝑦 0
5 2 3 39√2 5√30
𝐿1 ∫ 𝑥√ + 𝑥3 + 𝑥2 + 𝑥𝑑𝑥 = − + .
−1/2 24 3 2 560 336
1 − −
+ (𝑎 + 2𝑏02 ) 𝑦2 ] 𝑑𝑥.
2 11 Thus, we can obtain the conclusion for 𝑐0 .
Abstract and Applied Analysis 9

− − − −
When 𝑐1 = 0, we have (𝑎10 + 𝑏01 ) = (1/2)(2𝑎20 + 𝑏11 ). Thus, Under (58), we have for 0 < (5/48) − ℎ ≪ 1
𝑐2 in (14) can be rewritten as
√114 5 5
𝑀1 (ℎ) = − 𝛿4 (ℎ − ) ln ( − ℎ)
1 1 + 2𝑥
−
2𝑎00 7 48 48
− −
𝑐2 = (2𝑎20 + 𝑏11 )∫ 𝑑𝑥 − 5
2 𝐿− 𝑦 √2ℎ1 + 𝑂 ((ℎ − )) < 0,
48
−
4𝑎02 󵄨󵄨 (61)
ℎ 󵄨
− (√2ℎ1 + 1 )󵄨󵄨󵄨󵄨 √114 5 5
3 √2ℎ1 󵄨󵄨ℎ1 =5/48 𝑀2 (ℎ) = 𝛿 (ℎ − ) ln ( − ℎ)
7 4 48 48
0
− − 1 + 2𝑥 5
= (2𝑎20 + 𝑏11 )∫ 𝑑𝑥 + 𝑂 ((ℎ − )) > 0.
−1/2 √(5/24) + (2/3) 𝑥3 + (3/2) 𝑥2 + 𝑥 48
Now, we derive the expansion of 𝑀1 (ℎ). To achieve this,
4 − √30 −
− √6𝑎00 − 𝑎 , make the transformation
5 6 02
(55) √2
𝑢= (𝑥 + 1) , V = 𝑦, (62)
2
which, by Maple, implies the conclusion. Further, we can
together with time scaling 𝜏 = (√2/2)𝑡. Then the system
prove 𝑐̃0 and 𝑐̃2 in the same way to 𝑐0 and 𝑐2 , respectively. This
becomes
finishes the proof.
−
𝑢̇= V + 𝜀𝑝 (𝑢, V) ,
(63)
Proof of Theorem 1. Let V̇= −𝑢 − 2√2𝑢2 + 𝜀𝑞− (𝑢, V) ,
− − − − + + where
𝛿1 = 𝑎10 + 𝑏01 , 𝛿2 = 2𝑎20 + 𝑏11 , 𝛿3 = 𝑎10 + 𝑏01 ,
2 2
+ + + − + − − 𝑖
𝛿4 = 2𝑎20 + 𝑏11 , 𝛿5 = 𝑎00 − 𝑎00 , 𝛿6 = 𝑎02 − 𝑎02 . 𝑝 (𝑢, V) = ∑ 𝑎−𝑖𝑗 𝑢𝑖 V𝑗 = ∑ 𝑎𝑖𝑗− (√2𝑢 − 1) V𝑗 ,
(56) 𝑖+𝑗=0 𝑖+𝑗=0
(64)
2 2
− − 𝑖 𝑗
Then, solve the equations 𝑞 (𝑢, V) = ∑ 𝑏𝑖𝑗 𝑢𝑖 V𝑗 = ∑ √2𝑏𝑖𝑗− (√2𝑢 − 1) V .
𝑖+𝑗=0 𝑖+𝑗=0
𝑐01 = 𝑐02 = 𝑐0 + 𝑐̃0 = 𝑐1 + 𝑐̃1 = 𝑐2 + 𝑐̃2 = 0, (57)
For 𝜀 = 0, the Hamiltonian function of (63) has the form
which, together with (56), gives that 1 2 2√2 3 1
𝐻 (𝑢, V) = (V + 𝑢2 ) + 𝑢 + . (65)
2 3 12
13√57 √57 83
𝛿1 = 𝛿, 𝛿2 = 𝛿, 𝛿3 = − 𝛿, Denote 𝑀1 (ℎ) by the first Melnikov function of system (63)
42 4 3 4 28 4
near 𝐶1 , and note that
5√57 5 18√57 108
𝛿5 = ( − ) 𝛿4 , 𝛿6 = (− + ) 𝛿4 , √2 − 𝑑𝑢 − −
42 7 7 7 𝑞− 𝑑𝑥 − 𝑝− 𝑑𝑦 = ( 𝑞 ) − 𝑝 𝑑V = 𝑞− 𝑑𝑢 − 𝑝 𝑑V.
2 (√2/2)
(58)
(66)
where 𝛿4 > 0. Then, under (58)
Then we have 𝑀1 (ℎ) = 𝑀1 (ℎ). From Theorem 2.4.1 of [18],
we can obtain
√114 √114
𝑐1 = − 𝛿 < 0, 𝑐̃1 = 𝛿 > 0, 1 1 2
7 4 7 4 𝑀1 (ℎ) = ̃𝑏0 (ℎ − ) + 𝑂 ((ℎ − ) ) < 0,
12 12
6805 √ (59) (67)
𝑐3 + 𝑐̃3 = − 114𝛿4 < 0, 𝑐1 + 𝑐̃1 = 0, 1
20577 0<ℎ− ≪ 1,
12
𝑐21 + 𝑐̃22 = 0.
where, under (58),
Further, ̃𝑏 = 2𝜋 (𝑝− + 𝑞− ) (0, 0)
0 𝑢 V

𝜕 (𝑐01 , 𝑐02 , 𝑐0 + 𝑐̃0 , 𝑐1 + 𝑐̃1 , 𝑐2 + 𝑐̃2 ) −

= 2√2𝜋 [(𝑎10 −
+ 𝑏01 −
) − (2𝑎20 −
+ 𝑏11 )] = 2√2𝜋 (𝛿1 − 𝛿2 )
rank = 5. (60)
𝜕 (𝛿1 , 𝛿2 , 𝛿3 , 𝛿5 , 𝛿6 )
√114
=− 𝜋𝛿4 .
Thus, by Theorem 6, there exist 7 limit cycles near 𝐿 for 𝜀 > 0 21
small. (68)
10 Abstract and Applied Analysis

Similarly, under (58) we have piecewise linear differential systems with a straight line of
separation,” International Journal of Bifurcation and Chaos, vol.
𝑀2 (ℎ) 23, no. 4, Article ID 1350066, 10 pages, 2013.
[10] J. Llibre and E. Ponce, “Three nested limit cycles in discon-
√114 1681 1681 2 tinuous piecewise linear differential systems with two zones,”
=− 𝛿4 (ℎ + ) + 𝑂 ((ℎ − ℎ + ) ) < 0,
228 324 324 Dynamics of Continuous, Discrete and Impulsive Systems B, vol.
19, no. 3, pp. 325–335, 2012.
1681
0<ℎ+ ≪ 1. [11] S. Huan and X. Yang, “On the number of limit cycles in general
324 planar piecewise linear systems of node-node types,” Journal of
(69) Mathematical Analysis and Applications, vol. 411, no. 1, pp. 340–
353, 2014.
This implies that there exists another zero between ℎ = [12] F. Liang, M. Han, and V. G. Romanovski, “Bifurcation of limit
−(1681/324) and ℎ = 5/48 for 𝑀2 (ℎ). This ends the cycles by perturbing a piecewise linear Hamiltonian system
proof. with a homoclinic loop,” Nonlinear Analysis: Theory, Methods
& Applications, vol. 75, no. 11, pp. 4355–4374, 2012.
Conflict of Interests [13] Y. Xiong and M. Han, “Bifurcation of limit cycles by perturbing
a piecewise linear Hamiltonian system,” Abstract and Applied
The author declares that there is no conflict of interests Analysis, vol. 2013, Article ID 575390, 19 pages, 2013.
regarding the publication of this paper. [14] Y. Liu and Y. Xiong, “Bifurcation of limit cycles by perturbing a
piecewise linear Hamiltonian system with one or two saddles,”
Chaos, Solitons & Fractals, vol. 66, pp. 86–95, 2014.
Acknowledgments [15] J. Llibre and A. C. Mereu, “Limit cycles for discontinuous
quadratic differential systems with two zones,” Journal of Math-
The author would like to give thanks to Professor Maoan
ematical Analysis and Applications, vol. 413, no. 2, pp. 763–775,
Han for his helpful discussions during the preparation of the 2014.
paper. The project was supported by National Natural Science
[16] Y. Xiong and M. Han, “Limit cycle bifurcations in a class of
Foundation of China (11271261), FP7-PEOPLE-2012-IRSES-
perturbed piecewise smooth systems,” Applied Mathematics and
316338, and Shanghai Municipal Natural Science Foundation Computation, vol. 242, pp. 47–64, 2014.
no. 12ZR1421600.
[17] X. Liu and M. Han, “Bifurcation of limit cycles by perturbing
piecewise Hamiltonian systems,” International Journal of Bifur-
References cation and Chaos in Applied Sciences and Engineering, vol. 20,
no. 5, pp. 1379–1390, 2010.
[1] M. di Bernardo, C. J. Budd, and A. R. Champneys, Piecewise- [18] M. Han, Bifurcation Theory of Limit Cycles, Science Press,
Smooth Dynamical Systems: Theory and Applications, vol. 163 of Beijing, China, 2013.
Applied Mathematical Sciences, Springer, London, UK, 2008.
[2] C. J. Budd, “Non-smooth dynamical systems and the grazing
bifurcation,” in Nonlinear Mathematics and Its Applications
(Guildford, 1995), pp. 219–235, Cambridge University Press,
Cambridge, UK, 1996.
[3] C. A. Buzzi, P. R. da Silva, and M. A. Teixeira, “A singular
approach to discontinuous vector fields on the plane,” Journal
of Differential Equations, vol. 231, no. 2, pp. 633–655, 2006.
[4] J. Llibre, E. Ponce, and F. Torres, “On the existence and
uniqueness of limit cycles in Liénard differential equations
allowing discontinuities,” Nonlinearity, vol. 21, no. 9, pp. 2121–
2142, 2008.
[5] M. Han and W. Zhang, “On Hopf bifurcation in non-smooth
planar systems,” Journal of Differential Equations, vol. 248, no.
9, pp. C2399–C2416, 2010.
[6] F. Giannakopoulos and K. Pliete, “Planar systems of piecewise
linear differential equations with a line of discontinuity,” Non-
linearity, vol. 14, no. 6, pp. 1611–1632, 2001.
[7] F. Giannakopoulos and K. Pliete, “Closed trajectories in planar
relay feedback systems,” Dynamical Systems, vol. 17, no. 4, pp.
343–358, 2002.
[8] J. C. Artés, J. Llibre, J. C. Medrado, and M. A. Teixeira,
“Piecewise linear differential systems with two real saddles,”
Mathematics and Computers in Simulation, vol. 95, pp. 13–22,
2014.
[9] J. Llibre, M. A. Teixeira, and J. Torregrosa, “Lower bounds
for the maximum number of limit cycles of discontinuous
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 879564, 11 pages
http://dx.doi.org/10.1155/2014/879564

Research Article
Normal Form for High-Dimensional Nonlinear System and Its
Application to a Viscoelastic Moving Belt

S. P. Chen1 and Y. H. Qian2

1
College of Mathematics, Xiamen University of Technology, Xiamen 361024, China
2
College of Mathematics, Physics and Information Engineering, Zhejiang Normal University, Jinhua, Zhejiang 321004, China

Correspondence should be addressed to S. P. Chen; shupingchen@126.com

Received 10 May 2014; Revised 24 July 2014; Accepted 11 August 2014; Published 27 August 2014

Academic Editor: Yongli Song

Copyright © 2014 S. P. Chen and Y. H. Qian. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.

This paper is concerned with the computation of the normal form and its application to a viscoelastic moving belt. First, a new
computation method is proposed for significantly refining the normal forms for high-dimensional nonlinear systems. The improved
method is described in detail by analyzing the four-dimensional nonlinear dynamical systems whose Jacobian matrices evaluated
at an equilibrium point contain three different cases, that are, (i) two pairs of pure imaginary eigenvalues, (ii) one nonsemisimple
double zero and a pair of pure imaginary eigenvalues, and (iii) two nonsemisimple double zero eigenvalues. Then, three explicit
formulae are derived, herein, which can be used to compute the coefficients of the normal form and the associated nonlinear
transformation. Finally, employing the present method, we study the nonlinear oscillation of the viscoelastic moving belt under
parametric excitations. The stability and bifurcation of the nonlinear vibration system are studied. Through the illustrative example,
the feasibility and merit of this novel method are also demonstrated and discussed.

1. Introduction forms of nilpotent systems, whose linear part about the

origin is a nilpotent matrix. Benderesky and Churchill [13,
Bifurcation and stability analysis of nonlinear differential 14] and Sanders [15] studied the spectral sequences for the
equations is one of the challenging problems of mathemati- normal forms of vectors. Such spectral sequences can provide
cians and engineers. Normal form theory is one of the most valuable information on the normal forms.
important tools for such analysis [1–3]. The normal form On the other hand, Kuznetsov [16] considered the normal
theory for differential equations can be dated back to the form theory in an application-oriented way for computa-
pioneer works of the renowned mathematician Poincaré [4]. tion of high-dimensional nonlinear systems. Zhang and his
He tried to use some change of variables to alter nonlinear coworkers [17, 18] employed the adjoint operator method to
systems into linear ones. The idea of the method is to simplify obtain the higher-order normal forms of high-dimensional
the system such that the topological behavior of the system in nonlinear dynamical systems and the associated nonlinear
the vicinity of a singularity point remains unchanged. transformations. Zhang and Leung [19] considered a general
Some recent developments of the theory of normal four-dimensional normal form of a double Hopf bifurcation.
form can be found in [5–15]. Chen et al. [5] presented Yu and his associates [20–22] developed efficient computing
the renormalization group theory, which was used for the methods for parametric normal forms. They also applied
search of normal forms for large classes of finite-dimensional the new method to consider controlling bifurcations of the
vector fields [6–8]. Stróżyna and Żołądek [9, 10] made use nonlinear dynamical systems. Chen and Dora [23] were
of the time rescaling to achieve the results on the orbital devoted to the development of effective methods for fur-
equivalence of vector fields. In addition, Dullin and Meiss ther reductions of the classical normal forms for complex
[11] and Murdock [12] dealt with the problems for the normal dynamical systems. From a practical point of view, a better
2 Abstract and Applied Analysis

understanding and knowledge of the normal forms of various where 𝐴𝑥 represents the linear part, 𝐴 is the Jacobian
complex nonlinear systems will further promote the potential matrix, and 𝑓𝑘 (𝑥) denotes the 𝑘th-order vector homogeneous
interest for the analysis of real engineering problems. polynomials of 𝑥.
The normal form theory plays an important role in Without loss of generality, 𝐴 is expressed in terms of
the study of bifurcation behavior of differential dynamical the standard Jordan canonical form. Note that system (1) is
systems. Itovich and Moiola [24] made use of the frequency assumed to have an equilibrium at the origin 𝑥 = 0.
domain and the normal form methodologies to analyze the We take the coordinate transformation as follows:
unfolding of a nonresonant double Hopf singularity. Zhang
𝑥 = 𝑦 + 𝜑𝑘 (𝑦) . (2)
et al. [25] employed the center manifold reduction and
normal form method to obtain the singular bifurcation of a Substituting (2) into (1) gives
ring of three coupled advertising oscillators with delay. Jiang
and Yuan [26] studied the classical Van Der Pol equation −1
= (𝐼 + 𝜕𝑦 𝜑𝑘 ) 𝐹 (𝑦 + 𝜑𝑘 (𝑦)) ,
𝑦̇ (3)
with Bogdanov-Takens singularity and bifurcation. Gattulli et
al. [27] analyzed the postcritical behavior of a single degree −1 󵄩 󵄩2𝑘−2
(𝐼 + 𝜕𝑦 𝜑𝑘 (𝑦)) = 𝐼 − 𝜕𝑦 𝜑𝑘 (𝑦) + 𝑂 (󵄩󵄩󵄩𝑦󵄩󵄩󵄩 ), (4)
of freedom system equipped with a Tuned Mass Damper
for double Hopf bifurcation in the neighbourhood of 1 : 1
resonance. Using the normal form method and the center where 𝜕𝑦 𝜑𝑘 is the Jacobian matrix of 𝜑𝑘 with respect to 𝑦.
manifold theory, Li and his associates [28] investigated the Then, (4) is substituted back into (3) to form
double Hopf bifurcation of the trivial equilibrium for delay-
= 𝐴𝑦 + 𝑓2 (𝑦) + ⋅ ⋅ ⋅ + 𝑓𝑘−1 (𝑦)
𝑦̇
coupled limit cycle oscillators. Buono and Belair [29] studied
the normal form of a vector field, which is generated by 󵄩 󵄩𝑘+1
+ {𝑓𝑘 (𝑦) − [𝜕𝑦 𝜑𝑘 (𝑦) 𝐴𝑦 − 𝐴𝜑𝑘 (𝑦)]} + 𝑂 (󵄩󵄩󵄩𝑦󵄩󵄩󵄩 ) .
a scalar delay-differential equations at nonresonant double
Hopf bifurcation points. (5)
Some of the above works use the Jordan canonical form of Employing the normal form theory, we introduce a linear
the leading matrix 𝐴. However, it is well known that handling operator as follows:
the eigenvalues and Jordan canonical forms is very difficult in
computer algebra system. In this paper, a new computation 𝐿𝑘𝐴 : 𝐻𝑘𝑛 󳨀→ 𝐻𝑘𝑛 ,
method by direct computation is developed to refine the (6)
normal forms for high-dimensional nonlinear systems. We (𝐿𝑘𝐴𝜑𝑘 ) (𝑦) = 𝜕𝑦 𝜑𝑘 (𝑦) 𝐴𝑦 − 𝐴𝜑𝑘 (𝑦) , 𝜑𝑘 ∈ 𝐻𝑘𝑛 .
do not need to compute the Jordan canonical form of 𝐴 nor its
eigenvalues. Our method is applicable in both the nilpotent Hence, we can write
and the nonnilpotent cases.
In this paper, we will develop an efficient method for 𝐻𝑘𝑛 = 𝑅𝑘 + 𝐶𝑘 , (7)
computing the normal forms directly for general four-
dimension systems and apply the method to consider con- where 𝑅𝑘 represents the range of 𝐿𝑘𝐴 and 𝐶𝑘 is the comple-
trolling bifurcations. The approach is efficient since it does mentary space to 𝑅𝑘 .
not require the computation of the Jordan canonical form of Hence, the 𝑘th-order terms can be simplified to
𝐴 or its eigenvalues. Besides, the proposed method is applied
to investigate the nonlinear oscillations of a viscoelastic 𝑔𝑘 (𝑦) ∈ 𝐶𝑘 . (8)
moving belt under parametric excitations. The rest of the The rationale for the classical normal form theory can be
paper is organized as follows. In Section 2, the essential idea explained by the following theorem (see [30]).
behind the method in [23] is briefly introduced. The new
computation method is described in detail by analyzing the Theorem 1. Let the notations be the same as above. Suppose
four-dimensional nonlinear dynamical systems in Section 3. that the decomposition (7) is given for 𝑘 = 2, . . . , 𝑁. Then, there
The applications to stability and bifurcation analysis on the exists a sequence of near identity changes of variables 𝑥 = 𝑦 +
viscoelastic moving belt are presented in Sections 4 and 5 to 𝜑𝑘 (𝑦), in which 𝜑𝑘 (𝑦) ∈ 𝐻𝑘𝑛 . Therefore, the dynamical system
show the efficiency of the method. Finally, conclusions are (1) is transformed into
drawn in Section 6.
󵄩 󵄩𝑁+1
= 𝐴𝑦 + 𝑔2 (𝑦) + ⋅ ⋅ ⋅ + 𝑔𝑁 (𝑦) + 𝑂 (󵄩󵄩󵄩𝑦󵄩󵄩󵄩 ) ,
𝑦̇ (9)
2. Normal Forms for Nonlinear System
where 𝑔𝑘 (𝑦) ∈ 𝐶𝑘 for 𝑘 = 2, . . . , 𝑁.
Consider a dynamical system described by the following
differential equation: By applying the Takens normal form theory [30], one
arrives at the 𝑘th-order normal form 𝑔𝑘 (𝑦), while those parts
∞
= 𝐹 (𝑥) = 𝐴𝑥 + ∑ 𝑓𝑘 (𝑥) ,
𝑥̇ 𝑥 ∈ 𝑅𝑛 , belonging to 𝑅𝑘 can be removed by properly choosing the
(1)
𝑘=2 coefficients of the nonlinear transformation 𝜑𝑘 (𝑦).
Abstract and Applied Analysis 3

3. Computation of Normal Forms and It is easy to see that the three-order polynomial solutions
Their Coefficients in four variables can be obtained from (12). To achieve this,
we write
Consider a four-dimensional generalized averaged system
with 𝑍2 ⊕ 𝑍2 -symmetry governed by ∑ 𝛼𝑞 𝑥𝑞 ∑ 𝛼𝑞󸀠 𝑦𝑞
|𝑞|=3 |𝑞|=3
3 4 ( ∑ 𝛽𝑞 𝑥𝑞 ) ( ∑ 𝛽𝑞󸀠 𝑦𝑞 )
𝑥̇
= 𝐹 (𝑥) = 𝐴𝑥 + 𝑓 (𝑥) , 𝑥∈𝑅 , (10) ( ) ( )
(|𝑞|=3 ) (|𝑞|=3 )
𝑓3 (𝑥) = (
(
),
) 𝑔3 (𝑦) = (
(
),
)
where 𝑓3 (𝑥) ∈ 𝐻43 ; that is ( ∑ 𝛾𝑞 𝑥𝑞 ) ( ∑ 𝛾𝑞󸀠 𝑦𝑞 )
( |𝑞|=3 ) ( |𝑞|=3 )
𝑇
𝑓3 (𝑥) = (𝑓13 (𝑥) , 𝑓23 (𝑥) , 𝑓33 (𝑥) , 𝑓43 (𝑥)) ∑ 𝜂𝑞 𝑥𝑞 ∑ 𝜂𝑞󸀠 𝑦𝑞
( |𝑞|=3 ) ( |𝑞|=3 )
𝑚 𝑚 𝑚 𝑚
= (∑ 𝑎𝑚1 𝑚2 𝑚3 𝑚4 𝑥1 1 𝑥2 2 𝑥3 3 𝑥4 4 ,
|𝑚|=3
∑ 𝑎𝑞 𝑦𝑞
𝑚 𝑚 𝑚 𝑚 |𝑞|=3
∑ 𝑏𝑚1 𝑚2 𝑚3 𝑚4 𝑥1 1 𝑥2 2 𝑥3 3 𝑥4 4 ,
|𝑚|=3 (11) ( ∑ 𝑏𝑞 𝑦𝑞 )
( )
𝑚 𝑚 𝑚 𝑚 ( |𝑞|=3 )
∑ 𝑐𝑚1 𝑚2 𝑚3 𝑚4 𝑥1 1 𝑥2 2 𝑥3 3 𝑥4 4 , (
𝜑 (𝑦) = ( ),
𝑞)
|𝑚|=3 ( ∑ 𝑐𝑞 𝑦 )
( |𝑞|=3 )
𝑇

∑
𝑚1 𝑚2 𝑚3 𝑚4
𝑑𝑚1 𝑚2 𝑚3 𝑚4 𝑥1 𝑥2 𝑥3 𝑥4 ) , ∑ 𝑑𝑞 𝑦𝑞
|𝑚|=3 (|𝑞|=3 )
(16)
and |𝑚| = 𝑚1 + 𝑚2 + 𝑚3 + 𝑚4 .
At the same time, (6) becomes with 𝑎𝑞 , 𝑏𝑞 , 𝑐𝑞 , 𝑑𝑞 , 𝛼𝑞󸀠 , 𝛽𝑞󸀠 , 𝛾𝑞󸀠 and 𝜂𝑞󸀠 to be determined later.
For the case (i), (12) can be expressed as
𝜕𝑦 𝜑3 (𝑦) 𝐴𝑦 − 𝐴𝜑3 (𝑦) = 𝑓3 − 𝑔3 , (12)
∑ [−𝑤1 (𝑞1 + 1) 𝑎𝑞+𝑒1 −𝑒2 + 𝑤1 (𝑞2 + 1) 𝑎𝑞−𝑒1 +𝑒2
3 |𝑞|=3
The problem at hand is to determine 𝜑, so that 𝑔 contains
the smallest possible number of monomials. −𝑤2 (𝑞3 + 1) 𝑎𝑞+𝑒3 −𝑒4 + 𝑤2 (𝑞4 + 1) 𝑎𝑞−𝑒3 +𝑒4 + 𝑤1 𝑏𝑞 ] 𝑦𝑞
Consider the following three cases of the Jordan matrix 𝐴
in four-dimensional nonlinear systems. = ∑ (𝛼𝑞 − 𝛼𝑞󸀠 ) 𝑦𝑞 ,
|𝑞|=3
(i) The Jordan matrix 𝐴 has two pairs of pure imaginary (17a)
eigenvalues.
∑ [−𝑤1 (𝑞1 + 1) 𝑏𝑞+𝑒1 −𝑒2 + 𝑤1 (𝑞2 + 1) 𝑏𝑞−𝑒1 +𝑒2
(ii) The Jordan matrix 𝐴 has one nonsemisimple double |𝑞|=3
zero and a pair of pure imaginary eigenvalues.
−𝑤2 (𝑞3 + 1) 𝑏𝑞+𝑒3 −𝑒4 + 𝑤2 (𝑞4 + 1) 𝑏𝑞−𝑒3 +𝑒4 − 𝑤1 𝑎𝑞 ] 𝑦𝑞
(iii) The Jordan matrix 𝐴 has two nonsemisimple double
zero eigenvalues. = ∑ (𝛽𝑞 − 𝛽𝑞󸀠 ) 𝑦𝑞 ,
|𝑞|=3
The forms of the Jordan matrix 𝐴 in these cases (i)–(iii) (17b)
can be, respectively, represented by (13)–(15) as follows:
∑ [−𝑤1 (𝑞1 + 1) 𝑐𝑞+𝑒1 −𝑒2 + 𝑤1 (𝑞2 + 1) 𝑐𝑞−𝑒1 +𝑒2
|𝑞|=3
0 −𝑤1 0 0
[𝑤1 0 0 0 ] 𝐽1 0 −𝑤2 (𝑞3 + 1) 𝑐𝑞+𝑒3 −𝑒4 + 𝑤2 (𝑞4 + 1) 𝑐𝑞−𝑒3 +𝑒4 + 𝑤2 𝑑𝑞 ] 𝑦𝑞
𝐴=[ ]
[ 0 0 0 −𝑤2 ] = [ 0 𝐽2 ] , (13)
= ∑ (𝛾𝑞 − 𝛾𝑞󸀠 ) 𝑦𝑞 ,
[ 0 0 𝑤2 0 ] |𝑞|=3
0 1 0 0 (17c)
[0 0 0 0 ]
[
𝐴=[ ] = [𝐽1 0 ] , (14) ∑ [−𝑤1 (𝑞1 + 1) 𝑑𝑞+𝑒1 −𝑒2 + 𝑤1 (𝑞2 + 1) 𝑑𝑞−𝑒1 +𝑒2
0 0 0 −𝑤] 0 𝐽2
|𝑞|=3
[0 0 𝑤 0 ]
−𝑤2 (𝑞3 + 1) 𝑑𝑞+𝑒3 −𝑒4 + 𝑤2 (𝑞4 + 1) 𝑑𝑞−𝑒3 +𝑒4 − 𝑤2 𝑐𝑞 ] 𝑦𝑞
0 1 0 0
[0 0 0 0] = ∑ (𝜂𝑞 − 𝜂𝑞󸀠 ) 𝑦𝑞 ,
𝐴=[ ] = [𝐽1 0 ] . (15)
[0 0 0 1] 0 𝐽2 |𝑞|=3

[0 0 0 0] (17d)
4 Abstract and Applied Analysis

where 𝑒1 = (1, 0, 0, 0), 𝑒2 = (0, 1, 0, 0), 𝑒3 = (0, 0, 1, 0), and and we need to solve the following equations:
𝑒4 = (0, 0, 0, 1).
Then, we solve the following equations:
(𝑞1 + 1) 𝑎𝑞+𝑒1 −𝑒2 − 𝑤 (𝑞3 + 1) 𝑎𝑞+𝑒3 −𝑒4
− 𝑤1 (𝑞1 + 1) 𝑎𝑞+𝑒1 −𝑒2 + 𝑤1 (𝑞2 + 1) 𝑎𝑞−𝑒1 +𝑒2 (20a)
+ 𝑤 (𝑞4 + 1) 𝑎𝑞−𝑒3 +𝑒4 − 𝑏𝑞 = 𝛼𝑞 − 𝛼𝑞󸀠 ,
− 𝑤2 (𝑞3 + 1) 𝑎𝑞+𝑒3 −𝑒4 (18a)

+ 𝑤2 (𝑞4 + 1) 𝑎𝑞−𝑒3 +𝑒4 + 𝑤1 𝑏𝑞 = 𝛼𝑞 − 𝛼𝑞󸀠 , (𝑞1 + 1) 𝑏𝑞+𝑒1 −𝑒2 − 𝑤 (𝑞3 + 1) 𝑏𝑞+𝑒3 −𝑒4
(20b)
− 𝑤1 (𝑞1 + 1) 𝑏𝑞+𝑒1 −𝑒2 + 𝑤1 (𝑞2 + 1) 𝑏𝑞−𝑒1 +𝑒2 + 𝑤 (𝑞4 + 1) 𝑏𝑞−𝑒3 +𝑒4 = 𝛽𝑞 − 𝛽𝑞󸀠 ,

− 𝑤2 (𝑞3 + 1) 𝑏𝑞+𝑒3 −𝑒4 (18b) (𝑞1 + 1) 𝑐𝑞+𝑒1 −𝑒2 − 𝑤 (𝑞3 + 1) 𝑐𝑞+𝑒3 −𝑒4
(20c)
+ 𝑤2 (𝑞4 + 1) 𝑏𝑞−𝑒3 +𝑒4 − 𝑤1 𝑎𝑞 = 𝛽𝑞 − 𝛽𝑞󸀠 ,
+ 𝑤 (𝑞4 + 1) 𝑐𝑞−𝑒3 +𝑒4 + 𝑤𝑑𝑞 = 𝛾𝑞 − 𝛾𝑞󸀠 ,
− 𝑤1 (𝑞1 + 1) 𝑐𝑞+𝑒1 −𝑒2 + 𝑤1 (𝑞2 + 1) 𝑐𝑞−𝑒1 +𝑒2
(𝑞1 + 1) 𝑑𝑞+𝑒1 −𝑒2 − 𝑤 (𝑞3 + 1) 𝑑𝑞+𝑒3 −𝑒4
− 𝑤2 (𝑞3 + 1) 𝑐𝑞+𝑒3 −𝑒4 (18c) (20d)
+ 𝑤 (𝑞4 + 1) 𝑑𝑞−𝑒3 +𝑒4 − 𝑤𝑐𝑞 = 𝜂𝑞 − 𝜂𝑞󸀠 .
+ 𝑤2 (𝑞4 + 1) 𝑐𝑞−𝑒3 +𝑒4 + 𝑤2 𝑑𝑞 = 𝛾𝑞 − 𝛾𝑞󸀠 ,
Finally, the case (iii) leads to
− 𝑤1 (𝑞1 + 1) 𝑑𝑞+𝑒1 −𝑒2 + 𝑤1 (𝑞2 + 1) 𝑑𝑞−𝑒1 +𝑒2

− 𝑤2 (𝑞3 + 1) 𝑑𝑞+𝑒3 −𝑒4 (18d)

∑ [(𝑞1 + 1) 𝑎𝑞+𝑒1 −𝑒2 + (𝑞3 + 1) 𝑎𝑞+𝑒3 −𝑒4 − 𝑏𝑞 ] 𝑦𝑞
+ 𝑤2 (𝑞4 + 1) 𝑑𝑞−𝑒3 +𝑒4 − 𝑤2 𝑐𝑞 = 𝜂𝑞 − 𝜂𝑞󸀠 . |𝑞|=3
(21a)
For the case (ii), we obtain = ∑ (𝛼𝑞 − 𝛼𝑞󸀠 ) 𝑦𝑞 ,
|𝑞|=3

∑ [(𝑞1 + 1) 𝑎𝑞+𝑒1 −𝑒2 − 𝑤 (𝑞3 + 1) 𝑎𝑞+𝑒3 −𝑒4 ∑ [(𝑞1 + 1) 𝑏𝑞+𝑒1 −𝑒2 + (𝑞3 + 1) 𝑏𝑞+𝑒3 −𝑒4 ] 𝑦𝑞
|𝑞|=3
|𝑞|=3
𝑞 (21b)
+𝑤 (𝑞4 + 1) 𝑎𝑞−𝑒3 +𝑒4 − 𝑏𝑞 ] 𝑦 (19a)
= ∑ (𝛽𝑞 − 𝛽𝑞󸀠 ) 𝑦𝑞 ,
|𝑞|=3
= ∑ (𝛼𝑞 − 𝛼𝑞󸀠 ) 𝑦𝑞 ,
|𝑞|=3
∑ [(𝑞1 + 1) 𝑐𝑞+𝑒1 −𝑒2 + (𝑞3 + 1) 𝑐𝑞+𝑒3 −𝑒4 − 𝑑𝑞 ] 𝑦𝑞
∑ [(𝑞1 + 1) 𝑏𝑞+𝑒1 −𝑒2 − 𝑤 (𝑞3 + 1) 𝑏𝑞+𝑒3 −𝑒4 |𝑞|=3
|𝑞|=3 (21c)
= ∑ (𝛾𝑞 − 𝛾𝑞󸀠 ) 𝑦𝑞 ,
+𝑤 (𝑞4 + 1) 𝑏𝑞−𝑒3 +𝑒4 ] 𝑦𝑞 (19b) |𝑞|=3

= ∑ (𝛽𝑞 − 𝛽𝑞󸀠 ) 𝑦𝑞 , ∑ [(𝑞1 + 1) 𝑑𝑞+𝑒1 −𝑒2 + (𝑞3 + 1) 𝑑𝑞+𝑒3 −𝑒4 ] 𝑦𝑞

|𝑞|=3 |𝑞|=3
(21d)
∑ [(𝑞1 + 1) 𝑐𝑞+𝑒1 −𝑒2 − 𝑤 (𝑞3 + 1) 𝑐𝑞+𝑒3 −𝑒4 = ∑ (𝜂𝑞 − 𝜂𝑞󸀠 ) 𝑦𝑞 .
|𝑞|=3
|𝑞|=3
𝑞
+𝑤 (𝑞4 + 1) 𝑐𝑞−𝑒3 +𝑒4 + 𝑤𝑑𝑞 ] 𝑦 (19c)
Similarly, the following equations must be solved:
= ∑ (𝛾𝑞 − 𝛾𝑞󸀠 ) 𝑦𝑞 ,
|𝑞|=3
(𝑞1 + 1) 𝑎𝑞+𝑒1 −𝑒2 + (𝑞3 + 1) 𝑎𝑞+𝑒3 −𝑒4 − 𝑏𝑞 = 𝛼𝑞 − 𝛼𝑞󸀠 , (22a)
∑ [(𝑞1 + 1) 𝑑𝑞+𝑒1 −𝑒2 − 𝑤 (𝑞3 + 1) 𝑑𝑞+𝑒3 −𝑒4
|𝑞|=3 (𝑞1 + 1) 𝑏𝑞+𝑒1 −𝑒2 + (𝑞3 + 1) 𝑏𝑞+𝑒3 −𝑒4 = 𝛽𝑞 − 𝛽𝑞󸀠 , (22b)
𝑞
+𝑤 (𝑞4 + 1) 𝑑𝑞−𝑒3 +𝑒4 − 𝑤𝑐𝑞 ] 𝑦 (19d)
(𝑞1 + 1) 𝑐𝑞+𝑒1 −𝑒2 + (𝑞3 + 1) 𝑐𝑞+𝑒3 −𝑒4 − 𝑑𝑞 = 𝛾𝑞 − 𝛾𝑞󸀠 , (22c)
= ∑ (𝜂𝑞 − 𝜂𝑞󸀠 ) 𝑦𝑞 ,
|𝑞|=3 (𝑞1 + 1) 𝑑𝑞+𝑒1 −𝑒2 + (𝑞3 + 1) 𝑎𝑞+𝑒3 −𝑒4 = 𝜂𝑞 − 𝜂𝑞󸀠 . (22d)
Abstract and Applied Analysis 5

Let Let 𝑍0,𝑗,𝑙 (0 ≤ 𝑗 ≤ 3, 0 ≤ 𝑙 ≤ 3) be given arbitrarily.

We can determine 𝑍𝑖+1,𝑗−1,𝑙 by the above formulae to
𝛼𝑞󸀠 achieve Λ󸀠𝑖,𝑗,𝑙 = 0 for 0 ≤ 𝑖 ≤ 𝑘 − 1, 1 ≤ 𝑗 ≤ 𝑘, and
𝑎𝑞 𝛼𝑞 0 ≤ 𝑙 ≤ 𝑘. In fact, we have
𝑏 𝛽𝑞 𝛽󸀠
( 𝑞)
𝑀𝑞 = ( 𝑞 ) , Λ𝑞 = ( ), Λ󸀠𝑞 = ( ). 𝑤1 (𝑖 + 1) 𝑍𝑖+1,𝑗−1,𝑙
𝑐𝑞 𝛾𝑞
𝛾𝑞󸀠
𝑑𝑞 𝜂𝑞
󸀠 = 𝑤1 (𝑗 + 1) 𝑍𝑖−1,𝑗+1,𝑙 − 𝑤2 (𝑙 + 1) 𝑍𝑖,𝑗,𝑙+1 (29)
( 𝜂𝑞 )
(23) + 𝑤2 (4 − 𝑖 − 𝑗 − 𝑙) 𝑍𝑖,𝑗,𝑙−1 − 𝐴𝑍𝑖,𝑗,𝑙 − Λ 𝑖,𝑗,𝑙 .

Substituting (23) into (18a)–(18d), (20a)–(20d), and (22a)– The remaining equation is (for 𝑗 = 0)
(22d), separately, the following three sets of 3-order nonlinear
algebraic equations are resulted. Λ󸀠𝑖,0,𝑙 = Λ 𝑖,0,𝑙 − 𝑤1 𝑍𝑖−1,1,𝑙 + 𝑤2 (𝑙 + 1) 𝑍𝑖,0,𝑙+1
(30)
(i) For the case of two pairs of pure imaginary eigenval- − 𝑤2 (4 − 𝑖 − 𝑙) 𝑍𝑖,0,𝑙−1 + 𝐴𝑍𝑖,0,𝑙 .
ues, we arrive at

− 𝑤1 (𝑞1 + 1) 𝑀𝑞+𝑒1 −𝑒2 + 𝑤1 (𝑞2 + 1) 𝑀𝑞−𝑒1 +𝑒2 (ii) For the case of one nonsemisimple double zero and a
pair of pure imaginary eigenvalues, (25) becomes
− 𝑤2 (𝑞3 + 1) 𝑀𝑞+𝑒3 −𝑒4 (24)
(𝑖 + 1) 𝑍𝑖+1,𝑗−1,𝑙 − 𝑤 (𝑙 + 1) 𝑍𝑖,𝑗,𝑙+1
+ 𝑤2 (𝑞4 + 1) 𝑀𝑞−𝑒3 +𝑒4 − 𝐴𝑀𝑞 = Λ 𝑞 − Λ󸀠𝑞 . + 𝑤 (4 − 𝑖 − 𝑗 − 𝑙) 𝑍𝑖,𝑗,𝑙−1 − 𝐴𝑍𝑖,𝑗,𝑙 (31)
= Λ 𝑖,𝑗,𝑙 − Λ󸀠𝑖,𝑗,𝑙 .
(ii) For the case of one nonsemisimple double zero and a
pair of pure imaginary eigenvalues, we get In view of (31), we have

(𝑞1 + 1) 𝑀𝑞+𝑒1 −𝑒2 − 𝑤 (𝑞3 + 1) 𝑀𝑞+𝑒3 −𝑒4 (𝑖 + 1) 𝑍𝑖+1,𝑗−1,𝑙

(25)
+ 𝑤 (𝑞4 + 1) 𝑀𝑞−𝑒3 +𝑒4 − 𝐴𝑀𝑞 = Λ 𝑞 − Λ󸀠𝑞 . = 𝑤 (𝑙 + 1) 𝑍𝑖,𝑗,𝑙+1 − 𝑤 (4 − 𝑖 − 𝑗 − 𝑙) 𝑍𝑖,𝑗,𝑙−1 (32)
+ 𝐴𝑍𝑖,𝑗,𝑙 + Λ 𝑖,𝑗,𝑙 .
(iii) For the case of two nonsemisimple double zero
Hence, the remaining equation is given by (for 𝑗 = 0)
eigenvalues, we derive

Λ󸀠𝑖,0,𝑙 = Λ 𝑖,0,𝑙 + 𝑤 (𝑙 + 1) 𝑍𝑖,0,𝑙+1

(𝑞1 + 1) 𝑀𝑞+𝑒1 −𝑒2 + (𝑞3 + 1) 𝑀𝑞+𝑒3 −𝑒4 − 𝐴𝑀𝑞 = Λ 𝑞 − Λ󸀠𝑞 . (33)
(26) − 𝑤 (4 − 𝑖 − 𝑙) 𝑍𝑖,0,𝑙−1 + 𝐴𝑍𝑖,0,𝑙 .

It should be clear that 𝑀𝑞 = 0 if one of the components of (iii) For the case of two nonsemisimple double zero
𝑞 is completely negative. We choose the lexicographical order eigenvalues, we deduce
on the set {𝑞 = (𝑞1 , 𝑞2 , 𝑞3 , 𝑞4 ) ∈ 𝑁4 : |𝑞| = 3}.
Let 𝑞1 = 𝑖, 𝑞2 = 𝑗, 𝑞3 = 𝑙, and 𝑞4 = 3 − 𝑖 − 𝑗 − 𝑙. Then, we (𝑖+1) 𝑍𝑖+1,𝑗−1,𝑙 + (𝑙+1) 𝑍𝑖,𝑗,𝑙+1 − 𝐴𝑍𝑖,𝑗,𝑙 = Λ 𝑖,𝑗,𝑙 − Λ󸀠𝑖,𝑗,𝑙 .
have (34)

𝑍𝑖,𝑗,𝑙 = 𝑀(𝑖,𝑗,𝑙,3−𝑖−𝑗−𝑙) , Λ 𝑖,𝑗,𝑙 = 𝑓3 , Λ󸀠𝑖,𝑗,𝑙 = 𝑔3 .

(27) Similarly, making use of (34) leads to

(𝑖 + 1) 𝑍𝑖+1,𝑗−1,𝑙 = Λ 𝑖,𝑗,𝑙 − (𝑙 + 1) 𝑍𝑖,𝑗,𝑙+1 + 𝐴𝑍𝑖,𝑗,𝑙 . (35)

Substituting (27) into (24)–(26) gives the following three
conclusions.
From (34) and (35), the remaining equation can be
(i) For the case of two pairs of pure imaginary eigenval- written as (for 𝑗 = 0)
ues, we have
Λ󸀠𝑖,0,𝑙 = Λ 𝑖,0,𝑙 − (𝑙 + 1) 𝑍𝑖,0,𝑙+1 + 𝐴𝑍𝑖,0,𝑙 . (36)
− 𝑤1 (𝑖 + 1) 𝑍𝑖+1,𝑗−1,𝑙 + 𝑤1 (𝑗 + 1) 𝑍𝑖−1,𝑗+1,𝑙 − 𝑤2 (𝑙 + 1) 𝑍𝑖,𝑗,𝑙+1
We determine all 𝑍𝑖,𝑗,𝑙 for 𝑖 ≥ 1 as functions of 𝑍0,𝑗,𝑙 ,
+ 𝑤2 (4 − 𝑖 − 𝑗 − 𝑙) 𝑍𝑖,𝑗,𝑙−1 − 𝐴𝑍𝑖,𝑗,𝑙 = Λ 𝑖,𝑗,𝑙 − Λ󸀠𝑖,𝑗,𝑙 . which may be solved in some cases by making some of the
(28) terms Λ󸀠𝑖,0,𝑙 equal zeroes.
6 Abstract and Applied Analysis

 y introduce the mass, gyroscopic, and linear stiffness operators

L as follows:
c
u
x 𝜕 𝜕2
z 𝑀 = 𝐼, 𝐺 = 2𝛾 , 𝐾 = (𝛾2 − 1) . (41)
w 𝜕𝑥 𝜕𝑥2
T T
Substituting (41) into (37) leads to the standard symbolic form
Figure 1: The model of a viscoelastic moving belt and the coordinate
system.
𝜕2 𝑤 𝜕𝑤
𝑀𝑤,𝑡𝑡 + 𝐺𝑤,𝑡 + 𝐾𝑤 = 𝑁 (𝑤) + 𝑎 cos 𝜔𝑡 2
− 2𝜇 .
4. Application to a Viscoelastic Moving 𝜕𝑥 𝜕𝑡
Belt Model (42)

In this section, we apply the proposed method in Section 3

to a parametrically excited viscoelastic moving belt with To obtain a system that is suitable for the application
the external damping. Consider the viscoelastic moving belt of the method of multiple scales, we introduce the scale
model [31] with cross-sectional area 𝐴, length 𝐿 between transformations
two end supports, axial velocity V, and viscous damping
coefficient 𝑐 as shown in Figure 1. A Cartesian coordinate
𝑎 󳨀→ 𝜀𝑎, 𝑁 󳨀→ 𝜀𝑁, 𝜇 󳨀→ 𝜀𝜇, (43)
system (Oxyz) is adopted, which is located in the plane of
the viscoelastic moving belt. Another coordinate system is
a moving coordinate fixed on the belt. The 𝑢 and 𝑤 denote where 𝜀 is the small perturbation parameter.
the displacements in the 𝑥 and 𝑦 directions, respectively. It Substituting (43) into (42), we obtain the following
is assumed that the tension 𝑇 is characterized as a small dimensionless nonlinear system under parametric excita-
periodic perturbation 𝑇1 cos Ω𝑡 on the steady-state tension tions
𝑇0 ; that is, 𝑇 = 𝑇0 +𝑇1 cos Ω𝑡. Since the belt tension is assumed
to dominate the transverse stiffness, the bending stiffness of
the viscoelastic moving belt is neglected. The equations of 𝜕2 𝑤 𝜕𝑤
motion for the transverse vibration of the belt are based on an 𝑀𝑤,𝑡𝑡 + 𝐺𝑤,𝑡 + 𝐾𝑤 = 𝜀𝑁 (𝑤) + 𝜀𝑎 cos 𝜔𝑡 − 2𝜀𝜇 .
𝜕𝑥2 𝜕𝑡
axially moving string model. The nondimensional nonlinear (44)
governing equation of motion for the viscoelastic moving belt
under parametric excitations can be written as follows [31–
33]: The method of multiple scales can now be applied to search
for the uniform solutions of (44) in the following form:
𝑤,𝑡𝑡 + 2𝛾𝑤,𝑡𝑥 + (𝛾2 − 1 −𝑎 cos 𝜔𝑡) 𝑤,𝑥𝑥 +2𝜇𝑤,𝑡 − 𝑁 (𝑤 ) = 0,
(37)
𝑤 (𝑥, 𝑡, 𝜀) = 𝑤0 (𝑥, 𝑇0 , 𝑇1 ) + 𝜀𝑤1 (𝑥, 𝑇0 , 𝑇1 ) + ⋅ ⋅ ⋅ , (45)
where the comma subscript denotes the partial differentia-
tion, and
where 𝑇0 = 𝑡, 𝑇1 = 𝜀𝑡.
𝑇1 𝜌𝐴 𝜌𝐴𝐿2 𝐸0 𝐴
𝑎= , 𝛾 = V√ , 𝜔 = Ω√ , 𝐸𝑒 = , The differential operators of the method of multiple scales
𝑇0 𝑇0 𝑇0 𝑇0 can be defined as

𝐴 1 𝐴
𝐸V = 𝜂√ , 𝜇 = 𝑐√ , 𝑑 𝜕 𝜕𝑇0 𝜕 𝜕𝑇1
𝜌𝑇0 𝐿2 2 𝑇0 𝜌 = + + ⋅ ⋅ ⋅ = 𝐷0 + 𝜀𝐷1 + ⋅ ⋅ ⋅ , (46a)
𝑑𝑡 𝜕𝑇0 𝜕𝑡 𝜕𝑇1 𝜕𝑡
(38)
3 𝜕 1 𝑑2 2
𝑁 (𝑤 ) =
2 2
𝐸 (𝑤 ) 𝑤,𝑥𝑥 + 𝐸V ( (𝑤,𝑥 ) ) 𝑤,𝑥𝑥 = (𝐷0 + 𝜀𝐷1 + ⋅ ⋅ ⋅ ) = 𝐷02 + 2𝜀𝐷0 𝐷1 + ⋅ ⋅ ⋅ , (46b)
2 𝑒 ,𝑥 𝜕𝑡 2 𝑑𝑡2
(39)
𝜕
+ 𝐸V 𝑤,𝑥 (𝑤,𝑥 𝑤,𝑥𝑥 ) . where 𝐷𝑛 = 𝜕/𝜕𝑇𝑛 , 𝑛 = 0, 1, . . ..
𝜕𝑡
In this paper, we investigate the case of primary paramet-
The boundary conditions are imposed by ric resonance for the nth and lth order modes of (44), and we
𝑤 (0, 𝑡) = 0, 𝑤 (𝐿, 𝑡) = 0. (40) introduce the following linear transformation:

In the subsequent analysis, we use the method of multiple

scales and Galerkin’s approach in the partial differential 𝑥1 󳨀→ 𝑥2 , 𝑥2 󳨀→ 𝑥1 , 𝑥3 󳨀→ 𝑥4 , 𝑥4 󳨀→ 𝑥3 .
governing equations of the viscoelastic moving belt. We (47)
Abstract and Applied Analysis 7

Then, the four-dimensional averaged equations in the Carte- 𝑎𝑒1V2 (2𝜎12 + 𝑓2 ) 𝑏𝑒2V1 𝜎2 𝑏 𝜎
sian form can be obtained as + 2
𝑥23 + 𝑥1 𝑥32 + 𝑒2V1 2 𝑥1 𝑥42
2(𝜎1 + 𝑓) 𝜎2 − 𝐷 𝜎2 + 𝐷

𝑥1̇= − 𝜇𝑥1 − (𝜎1 − 𝑓) 𝑥2 + 𝑎𝑒2V1 𝑥1 (𝑥12 + 𝑥22 ) 𝑏𝑒1V2 𝜎1 𝜎2 𝑏𝑒1V2 𝜎1 𝜎2

+ 𝑥2 𝑥32 + 𝑥 𝑥2 ,
(𝜎2 − 𝐷) (𝜎1 + 𝑓) (𝜎2 + 𝐷) (𝜎1 + 𝑓) 2 4
+ 𝑎𝑒1V2 𝑥2 (𝑥12 + 𝑥22 ) + 𝑏𝑒2V1 𝑥1 (𝑥32 + 𝑥42 )
(49a)
𝑎𝑒2V1 𝜎1
+ 𝑏𝑒1V2 𝑥2 (𝑥32 + 𝑥42 ) + 𝑐𝑒2V1 𝑥3 (𝑥12 − 𝑥22 ) (48a) 𝑥2̇= (𝜎1 + 𝑓) 𝑥1 − 𝜇𝑥2 + 𝑥𝑥 2
𝜎1 − 𝑓 2 1
− 𝑐𝑒1V2 𝑥4 (𝑥12 − 𝑥22 )
2 2
𝑎𝑒2V1 𝜎1 3 𝑎𝑒1V2 (2𝜎1 + 𝑓 ) 3
+ 2𝑐𝑒1V2 𝑥1 𝑥2 𝑥3 + 2𝑐𝑒2V1 𝑥1 𝑥2 𝑥4 , + 𝑥2 − 2
𝑥1
𝜎1 + 𝑓 2(𝜎1 − 𝑓)
𝑥2̇= (𝜎1 + 𝑓) 𝑥1 − 𝜇𝑥2 − 𝑎𝑒1V2 𝑥1 (𝑥12 + 𝑥22 )
𝑎𝑒1V2 (2𝜎12 + 𝑓2 ) 𝑏𝑒2V1 𝜎2 𝑏 𝜎
+ 𝑎𝑒2V1 𝑥2 (𝑥12 + 𝑥22 ) − 𝑏𝑒1V2 𝑥1 (𝑥32 + 𝑥42 ) − 𝑥1 𝑥22 + 𝑥 𝑥2 + 𝑒2V1 2 𝑥 𝑥2
2 (𝜎12 − 𝑓2 ) 𝜎2 − 𝐷 2 3 𝜎2 + 𝐷 2 4

+ 𝑏𝑒2V1 𝑥2 (𝑥32 + 𝑥42 ) − 𝑐𝑒1V2 𝑥3 (𝑥12 − 𝑥22 ) (48b)

𝑏𝑒1V2 𝜎1 𝜎2 𝑏𝑒1V2 𝜎1 𝜎2
− 𝑥1 𝑥32 − 𝑥 𝑥2 ,
(𝜎2 − 𝐷) (𝜎1 − 𝑓) (𝜎2 + 𝐷) (𝜎1 − 𝑓) 1 4
− 𝑐𝑒2V1 𝑥4 (𝑥12 − 𝑥22 )
(49b)
+ 2𝑐𝑒2V1 𝑥1 𝑥2 𝑥3 − 2𝑐𝑒1V2 𝑥1 𝑥2 𝑥4 ,
𝐴 𝑒2V1 𝜎2 2
𝑥3̇= − 𝜇𝑥3 + (𝐷 − 𝜎2 ) 𝑥4 + 𝑥
𝑥3̇= − 𝜇𝑥3 + (𝐷 − 𝜎2 ) 𝑥4 + 𝐴 𝑒2V1 𝑥3 (𝑥32 + 𝑥42 ) 𝜎2 − 𝐷 3

+ 𝐴 𝑒1V2 𝑥4 (𝑥32 + 𝑥42 ) + 𝐵𝑒2V1 𝑥3 (𝑥12 + 𝑥22 ) 𝐴 𝑒2V1 𝜎2 𝐴 𝑒1V2 (2𝜎22 + 𝐷2 )

+ 𝑥3 𝑥42 + 𝑥4 𝑥32
𝜎2 + 𝐷 2 (𝜎22 − 𝐷2 )
+ 𝐵𝑒1V2 𝑥4 (𝑥12 + 𝑥22 ) + 𝐶𝑒2V1 𝑥1 (𝑥32 − 𝑥42 )
(48c)
+ 𝐶𝑒2V1 𝑥1 (𝑥32 − 𝑥42 ) 𝐴 𝑒1V2 (2𝜎22 + 𝐷2 ) 𝐵𝑒2V1 𝜎1 𝐵 𝜎
+ 𝑥43 + 𝑥 𝑥2 + 𝑒2V1 1 𝑥 𝑥2
2(𝜎2 + 𝐷)
2 𝜎1 − 𝑓 3 1 𝜎1 + 𝑓 3 2
− 𝐶𝑒1V2 𝑥2 (𝑥32 − 𝑥42 ) + 2𝐶𝑒1V2 𝑥1 𝑥3 𝑥4
𝐵𝑒1V2 𝜎1 𝜎2 𝐵𝑒1V2 𝜎1 𝜎2
+ 2𝐶𝑒2V1 𝑥2 𝑥3 𝑥4 , + 𝑥 𝑥2 + 𝑥 𝑥2 ,
(𝜎1 − 𝑓) (𝜎2 + 𝐷) 4 1 (𝜎1 + 𝑓) (𝜎2 + 𝐷) 4 2
𝑥4̇= (𝜎2 + 𝐷) 𝑥3 − 𝜇𝑥4 − 𝐴 𝑒1V2 𝑥3 (𝑥32 + 𝑥42 ) (49c)

+ 𝐴 𝑒2V1 𝑥4 (𝑥32 + 𝑥42 ) − 𝐵𝑒1V2 𝑥3 (𝑥12 + 𝑥22 ) 𝐴 𝑒2V1 𝜎2

𝑥4̇= (𝜎2 + 𝐷) 𝑥3 − 𝜇𝑥4 + 𝑥 𝑥2
𝜎2 − 𝐷 4 3
+ 𝐵𝑒2V1 𝑥4 (𝑥12 + 𝑥22 ) − 𝐶𝑒1V2 𝑥1 (𝑥32 − 𝑥42 ) (48d)
2 2
𝐴 𝑒2V1 𝜎2 3 𝐴 𝑒1V2 (2𝜎2 + 𝐷 ) 3
− 𝐶𝑒2V1 𝑥2 (𝑥32 − 𝑥42 ) + 𝑥 − 𝑥3
𝜎2 + 𝐷 4 2(𝜎2 − 𝐷)
2

+ 2𝐶𝑒2V1 𝑥1 𝑥3 𝑥4 − 2𝐶𝑒1V2 𝑥2 𝑥3 𝑥4 ,
𝐴 𝑒1V2 (2𝜎22 + 𝐷2 ) 𝐵𝑒2V1 𝜎1 𝐵 𝜎
− 𝑥3 𝑥42 + 𝑥4 𝑥12 + 𝑒2V1 1 𝑥4 𝑥22
where the coefficients presented in (48a)–(48d) are listed in 2 (𝜎22 − 𝐷2 ) 𝜎1 − 𝑓 𝜎1 + 𝑓
the appendix.
The above algorithm applied to the system (48a)–(48d) 𝐵𝑒1V2 𝜎1 𝜎2 𝐵𝑒1V2 𝜎1 𝜎2
− 𝑥 𝑥2 − 𝑥 𝑥2 .
leads to the following normal form (𝜎1 − 𝑓) (𝜎2 − 𝐷) 3 1 (𝜎1 + 𝑓) (𝜎2 − 𝐷) 3 2
(49d)
𝑎𝑒2V1 𝜎1 3
𝑥1̇= − 𝜇𝑥1 − (𝜎1 − 𝑓) 𝑥2 + 𝑥
𝜎1 − 𝑓 1
Comparing the method developed here with other meth-
ods given in [17, 31], it is observed that normal form of the
𝑎𝑒2V1 𝜎1 𝑎𝑒1V2 (2𝜎12 + 𝑓2 ) averaged (49a)–(49d) is simpler than normal forms obtained
+ 𝑥1 𝑥22 + 𝑥2 𝑥12
𝜎1 + 𝑓 2 (𝜎12 − 𝑓2 ) in [17, 31].
8 Abstract and Applied Analysis

5. Stability and Bifurcation Analysis on the Let

Viscoelastic Moving Belt 𝑧1 = 𝑟1 𝑒𝑖𝜑1 , 𝑧2 = 𝑟2 𝑒𝑖𝜑2 . (56)
It is known that (49a)–(49d) has a trivial zero solution In polar coordinates (𝑟1 , 𝑟2 , 𝜑1 , 𝜑2 ), systems (55a) and (55b)
(𝑥1 , 𝑥2 , 𝑥3 , 𝑥4 ) = (0, 0, 0, 0) in which the Jacobi matrix can can be written as
be written as
4𝑎𝑒2V1 𝜎1 2 4𝑏𝑒2V1 𝜎2 2
−𝜇 − (𝜎1 − 𝑓) 0 0 𝑟1̇= 𝑟1 (−𝑢 + 𝑟 + 𝑟 ), (57a)
𝜎1 + 𝑓 1 𝜎2 + 𝐷 2
[𝜎1 + 𝑓 −𝜇 0 0 ]
[
𝐽 = 𝐷𝑥 𝑋 = [ ] . (50)
0 0 −𝜇 𝐷 − 𝜎2 ] 𝑟2̇= 𝑟2 (−𝑢 +
4𝐵𝑒2V1 𝜎1 2 4𝐴 𝑒2V1 𝜎2 2
𝑟 + 𝑟 ), (57b)
[ 0 0 𝜎2 + 𝐷 −𝜇 ] 𝜎1 + 𝑓 1 𝜎2 + 𝐷 2

The characteristic equation corresponding to the trivial 2𝑎𝑒1V2 (2𝜎12 + 𝑓2 )

zero solution is 𝜑1̇= √𝜎12 − 𝑓2 − 𝑟12
(𝜎1 + 𝑓) √𝜎12 − 𝑓2
(𝜆2 + 2𝜇𝜆 + 𝜇2 + 𝜎12 − 𝑓2 ) (𝜆2 + 2𝜇𝜆 + 𝜇2 + 𝜎22 − 𝐷2 ) = 0. (57c)
4𝑏𝑒1V2 𝜎1 𝜎2
(51) − 𝑟22 ,
(𝜎2 + 𝐷) √𝜎12 − 𝑓2
The eigenvalues of the above equations are
4𝐵𝑒1V2 𝜎1 𝜎2
𝜑2̇= √𝜎22 − 𝐷2 − 𝑟12
𝜆 1 = −𝜇 + 𝑖√𝜎12 − 𝑓2 , 𝜆 2 = −𝜇 − 𝑖√𝜎12 − 𝑓2 , (𝜎1 + 𝑓) √𝜎22 − 𝐷2
(52)
(57d)
𝜆 3 = −𝜇 + 𝑖√𝜎22 − 𝐷2 , 𝜆 4 = −𝜇 − 𝑖√𝜎22 − 𝐷2 . 2𝐴 𝑒1V2 (2𝜎22 + 𝐷2 )
− 𝑟22 .
Take a linear transformation of coordinate as follows: (𝜎2 + 𝐷) √𝜎22 − 𝐷2

{𝑋} = [𝐷]{𝑍} , (53) Since our analysis is local, we can truncate the higher order
and consider the system
where [𝐷] is the matrix of the eigenvectors of the linear part 4𝑎𝑒2V1 𝜎1 2 4𝑏𝑒2V1 𝜎2 2
in (49a)–(49d), 𝑟1̇= 𝑟1 (−𝑢 + 𝑟 + 𝑟 ), (58a)
𝜎1 + 𝑓 1 𝜎2 + 𝐷 2
𝑇 𝑇
{𝑋} = [𝑥1 , 𝑥2 , 𝑥3 , 𝑥4 ] , {𝑍} = [𝑧1 , 𝑧1 , 𝑧2 , 𝑧2 ] . (54) 4𝐵𝑒2V1 𝜎1 2 4𝐴 𝑒2V1 𝜎2 2
𝑟2̇= 𝑟2 (−𝑢 + 𝑟 + 𝑟 ), (58b)
After substituting the above transformation into (49a)–(49d), 𝜎1 + 𝑓 1 𝜎2 + 𝐷 2
one obtains the equations in new complex coordinateas
follows: 𝜑1̇= √𝜎12 − 𝑓2 , (58c)

1 = (−𝜇 + 𝑖√𝜎1 − 𝑓 ) 𝑧1
𝑧̇ 2 2 𝜑2̇= √𝜎22 − 𝐷2 . (58d)

The first two equations are independent of the last two.

4𝑎 𝜎 2𝑎𝑒1V2 (2𝜎12 + 𝑓2 ) 󵄨 󵄨2 The last two equations describe rotations in the planes 𝑟2 = 0
+ ( 𝑒2V1 1 − 𝑖 ) 𝑧1 󵄨󵄨󵄨𝑧1 󵄨󵄨󵄨
𝜎1 + 𝑓 2
(𝜎1 + 𝑓) √𝜎1 − 𝑓 2 (55a) and 𝑟1 = 0 with angular velocities √𝜎12 − 𝑓2 and √𝜎22 − 𝐷2 ,
respectively. This does not change the bifurcation diagrams.
So, it is enough to study the first two equations. Therefore, we
4𝑏𝑒2V1 𝜎2 4𝑏𝑒1V2 𝜎1 𝜎2 󵄨 󵄨2 can study the original four-dimensional system (49a)–(49d)
+( −𝑖 ) 𝑧1 󵄨󵄨󵄨𝑧2 󵄨󵄨󵄨 ,
𝜎2 + 𝐷 2
(𝜎2 + 𝐷) √𝜎1 − 𝑓2 by analyzing the planar system
4𝑎𝑒2V1 𝜎1 2 4𝑏𝑒2V1 𝜎2 2
𝑟1̇= 𝑟1 (−𝑢 + 𝑟 + 𝑟 ), (59a)
2 = (−𝑢 + 𝑖√𝜎2 − 𝐷 ) 𝑧2
𝑧̇ 2 2 𝜎1 + 𝑓 1 𝜎2 + 𝐷 2
4𝐵𝑒2V1 𝜎1 2 4𝐴 𝑒2V1 𝜎2 2
𝑟2̇= 𝑟2 (−𝑢 + 𝑟 + 𝑟 ). (59b)
4𝐵 𝜎 4𝐵𝑒1V2 𝜎1 𝜎2 󵄨 󵄨2 𝜎1 + 𝑓 1 𝜎2 + 𝐷 2
+ ( 𝑒2V1 1 − 𝑖 ) 𝑧2 󵄨󵄨󵄨𝑧1 󵄨󵄨󵄨
𝜎1 + 𝑓 2
(𝜎1 + 𝑓) √𝜎2 − 𝐷 2 This system is called the amplitude system. The trivial
equilibrium, 𝑟1 = 0, 𝑟2 = 0, corresponds to the trivial
equilibrium of the original system. The study of the amplitude
4𝐴 𝑒2V1 𝜎2 2𝐴 𝑒1V2 (2𝜎22 + 𝐷2 ) 󵄨 󵄨2 system is simplified if we use squares 𝜌1,2 of the amplitudes as
+( −𝑖 ) 𝑧2 󵄨󵄨󵄨𝑧2 󵄨󵄨󵄨 .
𝜎2 + 𝐷 2
(𝜎2 + 𝐷) √𝜎2 − 𝐷 2 follows:
(55b) 𝜌𝑘 = 𝑟𝑘2 , 𝑘 = 1, 2. (60)
Abstract and Applied Analysis 9

The equations for 𝜌1,2 read 0

4𝑎𝑒2V1 𝜎1 4𝑏 𝜎
𝜌1̇= 2𝜌1 (−𝑢 + 𝜌 + 𝑒2V1 2 𝜌 ) , (61a)
𝜎1 + 𝑓 1 𝜎2 + 𝐷 2
−2

x3
4𝐵 𝜎 4𝐴 𝑒2V1 𝜎2
𝜌2̇= 2𝜌2 (−𝑢 + 𝑒2V1 1 𝜌1 + 𝜌 ). (61b)
𝜎1 + 𝑓 𝜎2 + 𝐷 2
−4
The behavior of systems (61a) and (61b) depends on the 1
coefficients of 𝜌1 and 𝜌2 . We start our analysis with the case −1.5
when 4𝑎𝑒2V1 𝜎1 /(𝜎1 + 𝑓) < 0 and 4𝐴 𝑒2V1 𝜎2 /(𝜎2 + 𝐷) < 0. The x2
0 −2
case 4𝑎𝑒2V1 𝜎1 /(𝜎1 + 𝑓) > 0 and 4𝐴 𝑒2V1 𝜎2 /(𝜎2 + 𝐷) > 0 can −2.5
−1 −3 x1
be reduced to the previous one by time reversal. The choices
4𝑎𝑒2V1 𝜎1 /(𝜎1 + 𝑓) < 0 and 4𝐴 𝑒2V1 𝜎2 /(𝜎2 + 𝐷) < 0 imply that Figure 2: Periodic solution of the viscoelastic moving belt system.
both of the primary Hopf bifurcations are supercritical and
stable.
For the case of 16𝐴 𝑒2V1 𝑎𝑒2V1 𝜎1 𝜎2 /(𝜎1 + 𝑓)(𝜎2 + 𝐷) < 0,
first we can reduce the number of the coefficients in (61a) and form of the averaged equation for the viscoelastic moving belt
(61b) by rescaling. Let under parametric excitations. Based on the current studies,
it is found that the newly developed computation method
4𝑎𝑒2V1 𝜎1 4𝐴 𝑒2V1 𝜎2
𝜉1 = 𝜌, 𝜉2 = − 𝜌, 𝜏 = 2𝑡. (62) improves the classical normal form. Therefore, it is a further
𝜎1 + 𝑓 1 𝜎2 + 𝐷 2 reduction of the classical normal form. Meanwhile, the
normal form derived herein is used to explore the bifurcation
We obtain the system and stability analysis of axially viscoelastic moving belts
𝑏𝑒2V1 under parametric excitations.
𝜉1̇= 𝜉1 (−𝑢 + 𝜉1 − 𝜉 ), (63a) In contrast to earlier works, this article argues that the
𝐴 𝑒2V1 2 normal forms of high-dimensional nonlinear systems may
𝐵𝑒2V1 always be achieved without computing either the Jordan
𝜉2̇= 𝜉2 (−𝑢 + 𝜉 − 𝜉2 ) . (63b) canonical form of 𝐴 or its eigenvalues. This significantly
𝑎𝑒2V1 1
reduces the difficulties which other computing methods
The trivial and the nontrivial equilibria of (63a) and (63b) may face in obtaining normal forms. Therefore, it is more
have the representations convenient to utilize the approach developed here to compute
the normal forms of averaged equations for different resonant
𝐸1 = (𝑢, 0) , 𝐸2 = (0, 𝑢) , cases. It is also found that the normal form of the averaged
equation by using the new method is simpler than those
𝑎𝑒2V1 𝑢 (𝐴 𝑒2V1 − 𝑏𝑒2V1 ) 𝐴 𝑒2V1 𝑢 (𝐵𝑒2V1 − 𝑎𝑒2V1 ) obtained in [17, 31]. This is very useful for the complex behav-
𝐸3 = ( , ).
𝑎𝑒2V1 𝐴 𝑒2V1 − 𝑏𝑒2V1 𝐵𝑒2V1 𝑎𝑒2V1 𝐴 𝑒2V1 − 𝑏𝑒2V1 𝐵𝑒2V1 ior patterns analysis of the nonlinear dynamical systems in an
(64) abstract sense. Therefore, this has opened the area for more
research works.
The nontrivial equilibrium 𝐸3 can bifurcate. Suppose that
4𝑎𝑒2V1 𝜎1 /(𝜎1 + 𝑓) > 0 and 4𝐴 𝑒2V1 𝜎2 /(𝜎2 + 𝐷) < 0. The
opposite case can be treated similarly. The Hopf bifurcation Appendix
and consequent existence of cycles are only possible in The coefficients in the averaged system (48a)–(48d) are
following three cases: (1) 𝑏𝑒2V1 /𝐴 𝑒2V1 > 1, 𝐵𝑒2V1 /𝑎𝑒2V1 > 1; presented as follows,
(2) 𝑏𝑒2V1 /𝐴 𝑒2V1 > 1, 𝐵𝑒2V1 /𝑎𝑒2V1 < 1, 𝑏𝑒2V1 𝐵𝑒2V1 /𝐴 𝑒2V1 𝑎𝑒2V1 > 1;
(3) 𝑏𝑒2V1 /𝐴 𝑒2V1 < 0, 𝐵𝑒2V1 /𝑎𝑒2V1 < 0, 𝑏𝑒2V1 𝐵𝑒2V1 /𝐴 𝑒2V1 𝑎𝑒2V1 > 1.
3𝐸𝑒 𝑚 𝐸 𝑚
The numerical simulation result is given in Figure 2, 𝑎𝑒2V1 = Im ( 1𝑛 ) + V Re ( 1𝑛 ) ,
which is the trajectory from the 4-dimensional space 8𝜔𝑛 𝑚𝑛 4 𝑚𝑛
(𝑥1 , 𝑥2 , 𝑥3 , 𝑥4 ) into the 3-dimensional space (𝑥1 , 𝑥2 , 𝑥3 ).
3𝐸𝑒 𝑚 𝐸 𝑚
Figure 2 shows the trajectory of the systems (61a) and (61b) 𝑎𝑒1V2 = Re ( 1𝑛 ) − V Im ( 1𝑛 ) ,
for 𝜇 = 0.0001, 𝜎1 = 2, 𝑓 = −124, 𝑎𝑒2V1 = 56, 𝑎𝑒1V2 = 2.3, 8𝜔𝑛 𝑚𝑛 4 𝑚𝑛
𝑏𝑒2V1 = 15, 𝑏𝑒1V2 = 19.6, 𝜎2 = 4.5, 𝐴 𝑒2V1 = −127, 𝐴 𝑒1V2 = 12, 3𝐸𝑒 𝑚 𝐸 𝑚
𝐵𝑒2V1 = 1.82, 𝐵𝑒1V2 = 12, and 𝐷 = 190. 𝑏𝑒2V1 = Im ( 2𝑛 ) + V Re ( 2𝑛 ) ,
8𝜔𝑛 𝑚𝑛 4 𝑚𝑛

6. Conclusions 3𝐸𝑒 𝑚 𝐸 𝑚
𝑏𝑒1V2 = Re ( 2𝑛 ) − V Im ( 2𝑛 ) ,
8𝜔𝑛 𝑚𝑛 4 𝑚𝑛
An efficient method for computing the normal form of high-
dimensional nonlinear systems is presented in this paper. 3𝐸𝑒 𝑚 𝐸 𝜔𝐸 𝑚
𝑐𝑒2V1 = Im ( 3𝑛 ) + ( V − 𝑙 V ) Re ( 3𝑛 ) ,
This computation method is applied to obtain the normal 16𝜔𝑛 𝑚𝑛 4 8𝜔𝑛 𝑚𝑛
10 Abstract and Applied Analysis

3𝐸𝑒 𝑚 𝐸 𝜔𝐸 𝑚
𝑐𝑒1V2 = Re ( 3𝑛 ) − ( V − 𝑙 V ) Im ( 3𝑛 ) , [3] A. Rincon, F. Angulo, and G. Olivar, “Control of an anaerobic
16𝜔𝑛 𝑚𝑛 4 8𝜔𝑛 𝑚𝑛 digester through normal form of fold bifurcation,” Journal of
Process Control, vol. 19, no. 8, pp. 1355–1367, 2009.
𝑎𝑛2 𝜋2 (𝛾2 − 1) [4] H. Poincaré, Sur les propriétés des fonctions définies par des
𝑓= , equations aux différences partielles thèse inaugural, Gauthier-
4𝜔𝑛
Villars, Paris, France, 1879.
(A.1) [5] L. Y. Chen, N. Goldenfeld, and Y. Oono, “Renormalization
group theory for global asymptotic analysis,” Physical Review
3𝐸𝑒 𝑚 𝐸 𝑚
𝐴 𝑒2V1 = Im ( 1𝑙 ) + V Re ( 1𝑙 ) , Letters, vol. 73, no. 10, pp. 1311–1315, 1994.
8𝜔𝑙 𝑚𝑙 4 𝑚𝑙 [6] R. E. L. DeVille, A. Harkin, M. Holzer, K. Josic, and T. J. Kaper,
“Analysis of a renormalization group method and normal form
3𝐸𝑒 𝑚 𝐸 𝑚
𝐴 𝑒1V2 = Re ( 1𝑙 ) − V Im ( 1𝑙 ) , theory for perturbed ordinary differential equations,” Physica D:
8𝜔𝑙 𝑚𝑙 4 𝑚𝑙 Nonlinear Phenomena, vol. 237, no. 8, pp. 1029–1052, 2008.
[7] H. Chiba, “Simplified renormalization group method for ordi-
3𝐸𝑒 𝑚 𝐸 𝑚
𝐵𝑒2V1 = Im ( 2𝑙 ) + V Re ( 2𝑙 ) , nary differential equations,” Journal of Differential Equations,
8𝜔𝑙 𝑚𝑙 4 𝑚𝑙 vol. 246, no. 5, pp. 1991–2019, 2009.
[8] M. Holzer, Renormalization group methods for singularly per-
3𝐸𝑒 𝑚 𝐸 𝑚
𝐵𝑒1V2 = Re ( 2𝑙 ) − V Im ( 2𝑙 ) , turbed systems, normal forms and stability of traveling waves
8𝜔𝑙 𝑚𝑙 4 𝑚𝑙 in a reaction-diffusion-mechanics systems [Ph.D. thesis], Boston
University, 2010.
3𝐸𝑒 𝑚 𝐸 𝜔𝐸 𝑚
𝐶𝑒2V1 = Im ( 3𝑙 ) + ( V − 𝑛 V ) Re ( 3𝑙 ) , [9] E. Stróżyna and H. Żołądek, “The analytic and formal normal
16𝜔𝑙 𝑚𝑙 4 8𝜔𝑙 𝑚𝑙 form for the nilpotent singularity,” Journal of Differential Equa-
tions, vol. 179, no. 2, pp. 479–537, 2002.
3𝐸𝑒 𝑚 𝐸 𝜔𝐸 𝑚
𝐶𝑒1V2 = Re ( 3𝑙 ) − ( V − 𝑛 V ) Im ( 3𝑙 ) , [10] E. Strózyna and H. Zoladek, “Orbital formal normal forms for
16𝜔𝑙 𝑚𝑙 4 8𝜔𝑙 𝑚𝑙 general Bogdanov-Takens singularity,” Journal of Differential
Equations, vol. 193, no. 1, pp. 239–259, 2003.
𝑎𝑙2 𝜋2 (𝛾2 − 1)
𝐷= , 𝜔𝑘 = 𝑘𝜋 (1 − 𝛾2 ) , 𝑘 = 𝑛, 𝑙, [11] H. R. Dullin and J. D. Meiss, “Nilpotent normal form for
4𝜔𝑙 divergence-free vector fields and volume-preserving maps,”
(A.2) Physica D: Nonlinear Phenomena, vol. 237, no. 2, pp. 156–166,
2008.
where 𝑎, 𝐸𝑒 , 𝐸V , 𝜔𝑙 and 𝜔𝑛 are given in (38) and (48a)–(48d). [12] J. Murdock, “On the structure of nilpotent normal form mod-
The terms 𝑚𝑖𝑗 (𝑖 = 0, 2, 3; 𝑗 = 𝑙, 𝑛) are complex eigenfunc- ules,” Journal of Differential Equations, vol. 180, no. 1, pp. 198–
tions of the displacement field, the detailed derivations of 237, 2002.
𝑚𝑖𝑗 (𝑖 = 0, 2, 3; 𝑗 = 𝑙, 𝑛) in (A.1) and (A.2) can be referred [13] M. Benderesky and R. Churchill, “A spectral sequence approach
to [25, 27]. to normal forms,” in Recent Developments in Algebraic Topology,
vol. 407 of Contemporary Mathematics, pp. 27–81, 2006.
Conflict of Interests [14] M. Bendersky and R. C. Churchill, “Normal forms in a cyclically
graded Lie algebra,” Journal of Symbolic Computation, vol. 41,
The authors declare that there is no conflict of interests no. 6, pp. 633–662, 2006.
regarding the publication of this article. [15] J. A. Sanders, “Normal form theory and spectral sequences,”
Journal of Differential Equations, vol. 192, no. 2, pp. 536–552,
2003.
Acknowledgments [16] Y. A. Kuznetsov, “Practical computation of normal forms on
center manifolds at degenerate Bogdanov-Takens bifurcations,”
The authors gratefully acknowledge the support of the
International Journal of Bifurcation and Chaos in Applied Sci-
National Natural Science Foundations of China (NNSFC), ences and Engineering, vol. 15, no. 11, pp. 3535–3546, 2005.
through Grant nos. 11302184 and 11202189, the Scientific [17] W. Zhang, F. X. Wang, and J. W. Zu, “Computation of normal
Research Fund of Zhejiang Provincial Education Depart- forms for high dimensional non-linear systems and application
ment, through Grant no. Y201121157, and the Scientific to non-planar non-linear oscillations of a cantilever beam,”
Research Foundation of Xiamen University of Technology, Journal of Sound and Vibration, vol. 278, no. 4-5, pp. 949–974,
through Grant no. 90030631. We are also grateful for helpful 2004.
comments which were provided by Wei Zhang at the Beijing [18] W. Zhang, Y. Chen, and D. Cao, “Computation of normal
University of Technology. forms for eight-dimensional nonlinear dynamical system and
application to a viscoelastic moving belt,” International Journal
References of Nonlinear Sciences and Numerical Simulation, vol. 7, no. 1, pp.
35–58, 2006.
[1] J. Murdock, Normal Forms and Unfoldings for Local Dynamical [19] Q. C. Zhang and A. Y. T. Leung, “Normal form of double
Systems, Springer Monographs in Mathematics, Springer, New Hopf bifurcation in forced oscillators,” Journal of Sound and
York, NY, USA, 2003. Vibration, vol. 231, no. 4, pp. 1057–1069, 2000.
[2] Y. A. Kuznetsov, Elements of Applied Bifurcation Theory, vol. 112 [20] P. Yu and G. Chen, “The simplest parametrized normal forms of
of Applied Mathematical Sciences, Springer, New York, NY, USA, HOPf and generalized HOPf bifurcations,” Nonlinear Dynam-
2004. ics, vol. 50, no. 1-2, pp. 297–313, 2007.
Abstract and Applied Analysis 11

[21] P. Yu and A. Y. T. Leung, “Normal forms of vector fields with

perturbation parameters and their application,” Chaos, Solitons
& Fractals, vol. 34, no. 2, pp. 564–579, 2007.
[22] M. Gazor and P. Yu, “Infinite order parametric normal form of
Hopf singularity,” International Journal of Bifurcation and Chaos
in Applied Sciences and Engineering, vol. 18, no. 11, pp. 3393–
3408, 2008.
[23] G. Chen and J. D. Dora, “Further reductions of normal forms for
dynamical systems,” Journal of Differential Equations, vol. 166,
no. 1, pp. 79–106, 2000.
[24] G. R. Itovich and J. L. Moiola, “Non-resonant double Hopf
bifurcations: the complex case,” Journal of Sound and Vibration,
vol. 322, no. 1-2, pp. 358–380, 2009.
[25] C. Zhang, H. Yin, and H. Zheng, “Simple bifurcation of coupled
advertising oscillators with delay,” Applied Mathematics Letters,
vol. 24, no. 11, pp. 1840–1844, 2011.
[26] W. Jiang and Y. Yuan, “Bogdanov-Takens singularity in van der
Pol’s oscillator with delayed feedback,” Physica D: Nonlinear
Phenomena, vol. 227, no. 2, pp. 149–161, 2007.
[27] V. Gattulli, F. Di Fabio, and A. Luongo, “One to one resonant
double Hopf bifurcation in aeroelastic oscillators with tuned
mass dampers,” Journal of Sound and Vibration, vol. 262, no. 2,
pp. 201–217, 2003.
[28] Y. Li, W. Jiang, and H. Wang, “Double Hopf bifurcation and
quasi-periodic attractors in delay-coupled limit cycle oscilla-
tors,” Journal of Mathematical Analysis and Applications, vol.
387, no. 2, pp. 1114–1126, 2012.
[29] P. Buono and J. Belair, “Restrictions and unfolding of double
Hopf bifurcation in functional differential equations,” Journal
of Differential Equations, vol. 189, no. 1, pp. 234–266, 2003.
[30] F. Takens, “Singularities of vector fields,” Publications
Mathématiques de l’IHÉS, vol. 43, pp. 47–100, 1974.
[31] W. Zhang and M. H. Yao, “Multi-pulse orbits and chaotic
dynamics in motion of parametrically excited viscoelastic mov-
ing belt,” Chaos, Solitons and Fractals, vol. 28, no. 1, pp. 42–66,
2006.
[32] L. H. Chen, W. Zhang, and Y. Q. Liu, “Modeling of nonlin-
ear oscillations for viscoelastic moving belt using generalized
Hamilton’s principle,” Journal of Vibration and Acoustics, Trans-
actions of the ASME, vol. 129, no. 1, pp. 128–132, 2007.
[33] C. Z. Song, Studies on non-linear dynamics of a multi-degree-
of freedom viscoelastic drive belt system [M.S. thesis], Beijing
University of Technology, 2005.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 127876, 5 pages
http://dx.doi.org/10.1155/2014/127876

Research Article
Periodic Solutions of Multispecies Mutualism System with
Infinite Delays

Wenbo Zhao,1 Caocuan Ma,2 Taotao Zheng,3 and Xiao-Ke Sun2

1
School of Physics and Information, Tianshui Normal University, Tianshui, Gansu 741000, China
2
School of Mathematics and Statistics, Tianshui Normal University, Tianshui, Gansu 741000, China
3
Department of Mathematics, Zhejiang University, Hangzhou, Zhejiang 310000, China

Correspondence should be addressed to Wenbo Zhao; zhao wenbo@yahoo.com

Received 27 June 2014; Accepted 9 August 2014; Published 27 August 2014

Academic Editor: Yonghui Xia

Copyright © 2014 Wenbo Zhao et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

We studied the delayed periodic mutualism system with Gilpin-Ayala effect. Some new and interesting sufficient conditions are
obtained to guarantee the existence of periodic solution for the multispecies mutualism system with infinite delays. Our method is
based on Mawhin’s coincidence degree. To the best knowledge of the authors, there is no paper considering the existence of periodic
solutions for n-species mutualism system with infinite delays.

1. Introduction with discrete delays. Thus, in this paper, we considered the

following mutualism system with distributed delays:
Recently, there are many papers considering the existence
of periodic solutions for competitive Lotka-Volterra system
𝑛
based on Mawhin’s coincidence degree theory (see [1–4]). 𝛼
𝑦𝑖̇(𝑡) = 𝑦𝑖 (𝑡) [𝑟𝑖 (𝑡) − 𝑎𝑖𝑖 (𝑡) 𝑦𝑖 (𝑡)𝛼𝑖𝑗 (𝑡) + ∑ 𝑎𝑖𝑗 (𝑡) 𝑦𝑗 𝑖𝑗 (𝑡)
But there are few papers considering the periodicity of 𝑗=1,𝑗=𝑖̸
mutualism system; for example, one can refer to [5–7]. [
However, the references mentioned above only considered 𝑛
𝛽
two-dimensional mutualism system. To the best knowledge + ∑ 𝑏𝑖𝑗 (𝑡) 𝑦𝑗 𝑖𝑗 (𝑡 − 𝜏𝑖𝑗 )
of the authors, there is no paper considering the existence of 𝑗=1,𝑗=𝑖̸
periodic solutions for 𝑛-species mutualism system. It should
𝑛 ∞
be noted that the method used in [5–7] is difficult to be 𝛾
extended to the 𝑛-dimensional system. So, we employ the + ∑ 𝑐𝑖𝑗 (𝑡) ∫ 𝐻𝑖𝑗 (𝑠) 𝑦𝑗 𝑖𝑗 (𝑡 + 𝑠) 𝑑𝑠] ,
𝑗=1,𝑗=𝑖̸ 0
method used in [2–4]. However, the problem considered ]
in this paper is completely different from those mentioned 𝑖 = 1, 2, . . . 𝑛,
above. On the other hand, the above-mentioned works con-
(1)
sidered the models with constant discrete delays or without
delays. In practice, there will be a distribution of transmission
delays. In this case, the transmission of species is no longer where 𝑟𝑖 , 𝑎𝑖𝑗 , 𝑏𝑖𝑗 , and 𝑐𝑖𝑗 , 𝑖, 𝑗 = 1, 2, . . . , 𝑛, are 𝜔-periodic
instantaneous and cannot be modelled with discrete delays. functions, that is, 𝑟𝑖 (𝑡 + 𝜔) = 𝑟𝑖 (𝑡), 𝑎𝑖𝑗 (𝑡 + 𝜔) = 𝑎𝑖𝑗 (𝑡),
A more appropriate way is to incorporate distributed delays. 𝑏𝑖𝑗 (𝑡 + 𝜔) = 𝑏𝑖𝑗 (𝑡), and 𝑐𝑖𝑗 (𝑡 + 𝜔) = 𝑐𝑖𝑗 (𝑡), and 𝛼𝑖𝑗 , 𝛽𝑖𝑗 , and 𝛾𝑖𝑗 are
Therefore, the studies of the model with distributed delays constants, 𝑖, 𝑗 = 1, 2, . . . , 𝑛. From biological view, 𝑟𝑖 , 𝑎𝑖𝑗 , 𝑏𝑖𝑗 ,
∞
have more important significance than the ones of the model and 𝑐𝑖𝑗 , 𝑖, 𝑗 = 1, 2, . . . , 𝑛, are nonnegative, ∫0 𝐻𝑖𝑗 (𝑠)𝑑𝑠 = 1,
2 Abstract and Applied Analysis

and 𝑎𝑖𝑖 is positive. System (1) is associated with the IVP as (𝐻2 ) 𝛼𝑗𝑖 ≤ 𝛼𝑖𝑖 , 𝛽𝑗𝑖 ≤ 𝛼𝑖𝑖 , 𝛾𝑗𝑖 ≤ 𝛾𝑖𝑖 (𝑗 ≠𝑖), 𝑖, 𝑗 = 1, 2, . . . , 𝑛.
follows: (𝐻3 ) 𝜌(K) < 1, where K = (Γ𝑖𝑗 )𝑛×𝑛 and
𝑦𝑖 (𝑡0 ) = 𝑦𝑖0 , 𝑦𝑖0 > 0, 𝑖 = 1, 2, . . . 𝑛. (2)
{ 0, 𝑖 = 𝑗,
{
Γ𝑖𝑗 = { 𝑎𝑖𝑗 + 𝑏𝑖𝑗 + 𝑐𝑖𝑗 (6)
2. Existence of Periodic Solutions { , 𝑖 ≠𝑗.
{ 𝑎𝑗𝑗
For convenience, we introduce some notations, definitions,
and lemmas. If 𝑔(𝑡) is a continuous 𝜔-periodic function Then system (1) has at least one positive 𝜔-periodic solution.
defined on R, denote
󵄨 󵄨 󵄨 󵄨 Proof. Note that every solution 𝑦(𝑡) = (𝑦1 (𝑡), 𝑦2 (𝑡), ...𝑦𝑛 (𝑡))𝑇
𝑔 = min 󵄨󵄨󵄨𝑔 (𝑡)󵄨󵄨󵄨 , 𝑔 = max 󵄨󵄨󵄨𝑔 (𝑡)󵄨󵄨󵄨 , of system (1) with the initial value condition is positive. Make
𝑡∈[0,𝜔] 𝑡∈[0,𝜔]
(3) the change of variables
1 𝜔
𝑚 (𝑔) = ∫ 𝑔 (𝑡) 𝑑𝑡.
𝜔 0 𝑦𝑖 (𝑡) = 𝑒𝑥𝑖 (𝑡) , 𝑖 = 1, 2, . . . , 𝑛. (7)
We also denote the spectral radius of the matrix A by 𝜌(A). Then system (1) is the same as
Denote
𝑛

𝑋 = {𝑥 (𝑡) = (𝑥1 (𝑡), 𝑥2 (𝑡), . . . , 𝑥𝑛 (𝑡))𝑇 𝑥𝑖 (𝑡) = 𝑟𝑖 (𝑡) − 𝑎𝑖𝑖 (𝑡) 𝑒𝛼𝑖𝑗 (𝑡) + ∑ 𝑎𝑖𝑗 (𝑡) 𝑒𝛼𝑖𝑗 (𝑡)
𝑗=1,𝑗=𝑖̸
1 𝑛
∈ 𝐶 (R, R ) | 𝑥 (𝑡 + 𝜔) = 𝑥 (𝑡) ∀𝑡 ∈ R} , 𝑛
(4) + ∑ 𝑏𝑖𝑗 (𝑡) 𝑒𝛽𝑖𝑗 (𝑡 − 𝜏𝑖𝑗 )
𝑍 = {𝑥 (𝑡) = (𝑥1 (𝑡), 𝑥2 (𝑡), . . . , 𝑥𝑛 (𝑡))𝑇 𝑗=1,𝑗=𝑖̸ (8)
𝑛 ∞
∈ 𝐶 (R, R𝑛 ) | 𝑥 (𝑡 + 𝜔) = 𝑥 (𝑡) ∀𝑡 ∈ R} . + ∑ 𝑐𝑖𝑗 (𝑡) ∫ 𝐻𝑖𝑗 (𝑠) 𝑒𝛾𝑖𝑗 (𝑡 + 𝑠) 𝑑𝑠,
𝑗=1,𝑗=𝑖̸ 0
Lemma 1 (see [8]). Let Ω ⊂ 𝑋 be an open and bounded set. Let
𝐿 be a Fredholm mapping of index zero and 𝑁 be 𝐿-compact 𝑖 = 1, 2, . . . 𝑛.
on Ω (i.e., 𝑄𝑁(Ω) is bounded and 𝐾𝑃 (𝐼 − 𝑄)𝑁 : Ω → 𝑋 is
compact). Assume Obviously, if system (8) has at least one 𝜔-periodic
solution, then system (1) has at least one 𝜔-periodic solution.
(i) for each 𝜆 ∈ (0, 1), 𝑥 ∈ 𝜕Ω ∩ Dom 𝐿, 𝐿𝑥 ≠𝜆𝑁𝑥; To prove Theorem 4, we should find an appropriate open set
(ii) for each 𝑥 ∈ 𝜕Ω ∩ Ker 𝐿, 𝑄𝑁𝑥 ≠0 and deg{𝐽𝑄𝑁, Ω ∩ Ω satisfying Lemma 1. We divide the proof into three steps.
Ker 𝐿, 0} ≠0. Step 1. We verify that (i) of Lemma 1 is satisfied. For any 𝑥(𝑡) ∈
Then 𝐿𝑥 = 𝑁𝑥 has at least one solution in Ω ∩ Dom 𝐿. 𝑋, by periodicity, it is easy to check that
𝑛
Definition 2 (see [9, 10]). A real 𝑛 × 𝑛 matrix A = (𝑎𝑖𝑗 ) is said Δ 𝑖 (𝑥, 𝑡) = 𝑟𝑖 (𝑡) − 𝑎𝑖𝑖 (𝑡) 𝑒𝛼𝑖𝑗 (𝑡) + ∑ 𝑎𝑖𝑗 (𝑡) 𝑒𝛼𝑖𝑗 (𝑡)
to be an 𝑀-matrix if 𝑎𝑖𝑗 ≤ 0, 𝑖, 𝑗 = 1, 2, . . . , 𝑛, 𝑖 ≠𝑗, and 𝑗=1,𝑗=𝑖̸
A−1 ≥ 0. 𝑛
+ ∑ 𝑏𝑖𝑗 (𝑡) 𝑒𝛽𝑖𝑗 (𝑡 − 𝜏𝑖𝑗 ) (9)
Lemma 3 (see [9, 10]). Let A ≥ 0 be an 𝑛 × 𝑛 matrix and 𝑗=1,𝑗=𝑖̸
𝜌(A) < 1; then (𝐸𝑛 − A)−1 ≥ 0, where 𝐸𝑛 denotes the identity
matrix of size 𝑛. 𝑛 ∞
+ ∑ 𝑐𝑖𝑗 (𝑡) ∫ 𝐻𝑖𝑗 (𝑠) 𝑒𝛾𝑖𝑗 (𝑡 + 𝑠) 𝑑𝑠.
0
Theorem 4. Assume the following. 𝑗=1,𝑗=𝑖̸

(𝐻1 ) The algebraic equation And define 𝐿 : Dom 𝐿 ⊂ 𝑋 → 𝑍 and 𝑁 : 𝑋 → 𝑍 as follows:

𝑑𝑥 (𝑡)
𝛼
𝑛
𝛼 𝑋 ∋ 𝑥 (𝑡) 󳨀→ (𝐿𝑥) (𝑡) = ∈ 𝑍,
𝑓 (𝑢) := (𝑚 (𝑟𝑖 ) − 𝑚 (𝑎𝑖𝑖 ) 𝑢𝑖 𝑖𝑖 + ∑ 𝑚 (𝑎𝑖𝑗 ) 𝑢𝑗 𝑖𝑗 𝑑𝑡
𝑗=1
𝑋 ∋ 𝑥 (𝑡) 󳨀→
(5)
𝑛 𝑛
𝛽
+ ∑𝑚 (𝑏𝑖𝑗 ) 𝑢𝑗 𝑖𝑗 + ∑ 𝑚 (𝑐𝑖𝑗 ) 𝑢𝑗𝑖𝑗 )
𝛾
=0 (𝑁𝑥) (𝑡) = ((𝑁𝑥)1 (𝑡), (𝑁𝑥)2 (𝑡), . . . , (𝑁𝑥)𝑛 (𝑡))𝑇 ∈ 𝑍,
𝑗=1 𝑗=1,𝑗=𝑖̸ (10)
𝑛×1

where
has finite solutions (𝑢1∗ , 𝑢2∗ , . . . , 𝑢𝑛∗ )𝑇 ∈ R+𝑛 with 𝑢𝑖∗ > 0 and
∑𝑢∗ sgn 𝐽𝑓 (𝑢∗ ) ≠0. (𝑁𝑥)𝑖 (𝑡) = Δ 𝑖 (𝑥, 𝑡) , 𝑖 = 1, 2, . . . , 𝑛. (11)
Abstract and Applied Analysis 3

The projectors are defined by 𝑃 : 𝑋 → 𝑋 and 𝑄 : 𝑍 → 𝑍 𝑥𝑖 (𝑡𝑖 ) = max𝑡∈[0,𝜔] |𝑥𝑖 (𝑡)|, 𝑖 = 1, 2, . . . , 𝑛. Accordingly, 𝑥𝑖̇
(𝑡𝑖 ) =
by 0 and we arrive at
1 𝜔 1 𝜔 𝑛
𝑃𝑥 (𝑡) = ∫ 𝑥 (𝑡) 𝑑𝑡, 𝑄𝑧 (𝑡) = ∫ 𝑧 (𝑡) 𝑑𝑡, 𝑟𝑖 (𝑡𝑖 ) − 𝑎𝑖𝑖 (𝑡𝑖 ) 𝑒𝛼𝑖𝑖 𝑥𝑖 (𝑡𝑖 ) + ∑ 𝑎𝑖𝑗 (𝑡𝑖 ) 𝑒𝛼𝑖𝑗 𝑥𝑗 (𝑡𝑖 )
𝜔 0 𝜔 0 (12) 𝑗=1,𝑗=𝑖̸
𝑥 ∈ 𝑋, 𝑧 ∈ 𝑍. 𝑛
+ ∑ 𝑏𝑖𝑗 (𝑡𝑖 ) 𝑒𝛽𝑖𝑗 𝑥𝑗 (𝑡𝑖 −𝜏𝑖𝑗 )
It is easy to follow that 𝐿 is a Fredholm mapping of index zero. 𝑗=1,𝑗=𝑖̸ (17)
Furthermore, the generalized inverse (to 𝐿) 𝐾𝑃 : Im 𝐿 →
Dom 𝐿 ∩ Ker 𝑃 exists, which is given by 𝑛 ∞
+ ∑ 𝑐𝑖𝑗 (𝑡) ∫ 𝐻𝑖𝑗 (𝑠) 𝑒𝛾𝑖𝑗 𝑥𝑗 (𝑡𝑖 +𝑠) 𝑑𝑠 = 0,
𝑡
1 𝜔 𝑡 𝑗=1,𝑗=𝑖̸ 0
𝐾𝑃 (𝑦) = ∫ 𝑦 (𝑠) 𝑑𝑠 − ∫ ∫ 𝑦 (𝑠) 𝑑𝑠 𝑑𝑡. (13)
0 𝜔 0 0
𝑖 = 1, 2, . . . , 𝑛.
Then 𝑄𝑁 : 𝑋 → 𝑍 and 𝐾𝑃 (𝐼 − 𝑄)𝑁 : 𝑋 → 𝑋 are defined
by That is,

1 𝜔 1 𝜔 𝑎𝑖𝑖 (𝑡𝑖 ) 𝑒𝛼𝑖𝑖 𝑥𝑖 (𝑡𝑖 )

𝑄𝑁𝑥 = ( ∫ Δ 1 (𝑥, 𝑡) 𝑑𝑡, ∫ Δ 2 (𝑥, 𝑡) 𝑑𝑡,
𝜔 0 𝜔 0
𝑛 𝑛

1 𝜔 𝑇
(14) = 𝑟𝑖 (𝑡𝑖 ) + ∑ 𝑎𝑖𝑗 (𝑡𝑖 ) 𝑒𝑥𝑗 (𝑡𝑖 ) + ∑ 𝑏𝑖𝑗 (𝑡𝑖 ) 𝑒𝛽𝑖𝑗 𝑥𝑗 (𝑡𝑖 −𝜏𝑖𝑗 )
..., ∫ Δ (𝑥, 𝑡)𝑑𝑡) , 𝑗=1,𝑗=𝑖̸ 𝑗=1,𝑗=𝑖̸ (18)
𝜔 0 𝑛
𝑛 ∞
𝐾𝑃 (𝐼 − 𝑄) 𝑁𝑥 = (Ψ1 (𝑥, 𝑡), Ψ2 (𝑥, 𝑡), . . . , Ψ𝑛 (𝑥, 𝑡))𝑇 , + ∑ 𝑐𝑖𝑗 (𝑡) ∫ 𝐻𝑖𝑗 (𝑠) 𝑒𝛾𝑖𝑗 𝑥𝑗 (𝑡𝑖 +𝑠) 𝑑𝑠, 𝑖 = 1, 2, . . . 𝑛.
𝑗=1,𝑗=𝑖̸ 0
where
𝑡
1 𝜔 𝑡 Noticing that 𝑥𝑗 (𝑡𝑗 ) = max𝑡∈[0,𝜔] |𝑥𝑗 (𝑡)| implies
Ψ𝑘 (𝑥, 𝑡) = ∫ Δ 𝑘 (𝑥, 𝑠) 𝑑𝑠 − ∫ ∫ Δ (𝑥, 𝑠) 𝑑𝑠 𝑑𝑡
0 𝜔 0 0 𝑘
𝜔
𝑥𝑗 (𝑡𝑖 − 𝜏𝑖𝑗 ) ≤ 𝑥𝑗 (𝑡𝑗 ) . (19)
𝑡 1
− ( − ) ∫ Δ 𝑘 (𝑥, 𝑠) 𝑑𝑠, 𝑘 = 1, 2, . . . , 𝑛.
𝜔 2 0 It follows from (H 2 ) that

(15) 𝑎𝑖𝑖 𝑒𝛼𝑖𝑖 𝑥𝑖 (𝑡𝑖 )

Using similar arguments to Step 1 in [2], it is easy to show that 󵄨 󵄨
≤ 󵄨󵄨󵄨󵄨𝑎𝑖𝑖 (𝑡𝑖 ) 𝑒𝛼𝑖𝑖 𝑥𝑖 (𝑡𝑖 ) 󵄨󵄨󵄨󵄨
(𝐾𝑃 (𝐼−𝑄)𝑁𝑥)(Ω) is relatively compact in the space (𝑋, ‖⋅‖1 ).
󵄨󵄨
Step 2. In this step, we are in a position to search for an 󵄨󵄨 𝑛 𝑛
= 󵄨󵄨󵄨󵄨𝑟𝑖 (𝑡𝑖 ) + ∑ 𝑎𝑖𝑗 (𝑡𝑖 ) 𝑒𝛼𝑖𝑗 𝑥𝑗 (𝑡𝑖 ) + ∑ 𝑏𝑖𝑗 (𝑡𝑖 ) 𝑒𝛽𝑖𝑗 𝑥𝑗 (𝑡𝑖 −𝜏𝑖𝑗 )
appropriate open bounded subset Ω satisfying condition (i) of 󵄨󵄨
Lemma 1. Specifically, our aim is to search for an appropriate 󵄨 𝑗=1,𝑗=𝑖̸ 𝑗=1,𝑗=𝑖̸

ℎ𝑖 defined by Ω in Step 1 such that Ω satisfies condition (i) of 󵄨󵄨

𝑛 ∞
𝛾𝑖𝑗 𝑥𝑗 (𝑡𝑖 +𝑠) 󵄨󵄨
󵄨
Lemma 1. To this end, assume that 𝑥(𝑡) ∈ 𝑋 is a solution of + ∑ 𝑐𝑖𝑗 (𝑡) ∫ 𝐻𝑖𝑗 (𝑠) 𝑒 𝑑𝑠󵄨󵄨󵄨
the equation 𝐿𝑥 = 𝜆𝑁𝑥 for each 𝜆 ∈ (0, 1); that is, 0 󵄨󵄨
𝑗=1,𝑗=𝑖̸ 󵄨
𝑛 𝑛
𝑛
𝑥𝑖̇(𝑡) = 𝜆 [𝑟𝑖 (𝑡) − 𝑎𝑖𝑖 (𝑡) 𝑒𝛼𝑖𝑗 (𝑡) + ∑ 𝑎𝑖𝑗 (𝑡) 𝑒𝛼𝑖𝑗 (𝑡) ≤ 𝑟𝑖 + ∑ 𝑎𝑖𝑗 𝑒𝛼𝑖𝑗 𝑥𝑗 (𝑡𝑖 ) + ∑ 𝑏𝑖𝑗 𝑒𝛽𝑖𝑗 𝑥𝑗 (𝑡𝑖 −𝜏𝑖𝑗 )
𝑗=1,𝑗=𝑖̸ 𝑗=1,𝑗=𝑖̸
𝑗=1,𝑗=𝑖̸
[
𝑛 ∞
𝑛
𝛽𝑖𝑗 + ∑ 𝑐𝑖𝑗 (𝑡) 𝑒𝛾𝑖𝑗 𝑥𝑗 (𝑡𝑖 +𝑠) ∫ 𝐻𝑖𝑗 (𝑠) 𝑑𝑠
+ ∑ 𝑏𝑖𝑗 (𝑡) 𝑒 (𝑡 − 𝜏𝑖𝑗 ) 𝑗=1,𝑗=𝑖̸ 0
𝑗=1,𝑗=𝑖̸ (16)
𝑛 𝑛
𝑛 ∞ = 𝑟𝑖 + ∑ 𝑎𝑖𝑗 𝑒𝛼𝑖𝑗 𝑥𝑗 (𝑡𝑖 ) + ∑ 𝑏𝑖𝑗 𝑒𝛽𝑖𝑗 𝑥𝑗 (𝑡𝑖 −𝜏𝑖𝑗 )
+ ∑ 𝑐𝑖𝑗 (𝑡) ∫ 𝐻𝑖𝑗 (𝑠) 𝑒𝛾𝑖𝑗 (𝑡 + 𝑠) 𝑑𝑠] , 𝑗=1,𝑗=𝑖̸ 𝑗=1,𝑗=𝑖̸
𝑗=1,𝑗=𝑖̸ 0
] 𝑛

𝑖 = 1, 2, . . . 𝑛. + ∑ 𝑐𝑖𝑗 (𝑡) 𝑒𝛾𝑖𝑗 𝑥𝑗 (𝑡𝑖 +𝑠)

𝑗=1,𝑗=𝑖̸

Since 𝑥(𝑡) ∈ 𝑋, each 𝑥𝑖 (𝑡), 𝑖 = 1, 2, . . . , 𝑛, as components of 𝑛 𝑛

𝑥(𝑡), is continuously differentiable and 𝜔-periodic. In view of ≤ 𝑟𝑖 + ∑ 𝑎𝑖𝑗 𝑒𝛼𝑗𝑗 𝑥𝑗 (𝑡𝑗 ) + ∑ 𝑏𝑖𝑗 𝑒𝛼𝑗𝑗 𝑥𝑗 (𝑡𝑗 )
continuity and periodicity, there exists 𝑡𝑖 ∈ [0, 𝜔] such that 𝑗=1,𝑗=𝑖̸ 𝑗=1,𝑗=𝑖̸
4 Abstract and Applied Analysis

𝑛
Estimating (16), by using (26) and (28), we have
+ ∑ 𝑐𝑖𝑗 (𝑡) 𝑒𝛼𝑗𝑗 𝑥𝑗 (𝑡𝑗 )
𝑗=1,𝑗=𝑖̸
󵄨󵄨
󵄨󵄨 𝑛
𝑛 (𝑡)|0 = 𝜆 󵄨󵄨󵄨󵄨𝑟𝑖 (𝑡) − 𝑎𝑖𝑖 (𝑡) 𝑒𝛼𝑖𝑖 𝑥𝑖 (𝑡) + ∑ 𝑎𝑖𝑗 (𝑡) 𝑒𝛼𝑖𝑗 𝑥𝑗 (𝑡)
|𝑥𝑖̇
= 𝑟𝑖 + ∑ (𝑎𝑖𝑗 + 𝑏𝑖𝑗 + 𝑐𝑖𝑗 ) 𝑒𝛼𝑗𝑗 𝑥𝑗 (𝑡𝑗 ) . 󵄨󵄨
󵄨 𝑗=1,𝑗=𝑖̸
𝑗=1,𝑗=𝑖̸
𝑛
(20) + ∑ 𝑏𝑖𝑗 (𝑡) 𝑒𝛽𝑖𝑗 𝑥𝑗 (𝑡−𝜏𝑖𝑗 )
𝑗=1,𝑗=𝑖̸
Here we used (H 2 ). Letting (𝑎𝑖𝑖 + 𝑏𝑖𝑖 )𝑒𝛼𝑖𝑖 𝑥𝑖 (𝑡𝑖 ) = 𝑧𝑖 (𝑡𝑖 ), it follows
󵄨󵄨
from (20) that 𝑛 ∞ 󵄨󵄨
𝑛 + ∑ 𝑐𝑖𝑗 (𝑡) ∫ 𝐻𝑖𝑗 (𝑠) 𝑒 𝛾𝑖𝑗 𝑥𝑗 (𝑡𝑖 +𝑠)
𝑑𝑠󵄨󵄨󵄨󵄨
0 󵄨󵄨
𝑧𝑖 (𝑡𝑖 ) ≤ 𝑟𝑖 + ∑ (𝑎𝑖𝑗 + 𝑏𝑖𝑗 + 𝑐𝑖𝑗 ) 𝑎−1
𝑗𝑗 𝑧𝑗 (𝑡𝑗 ) , (21)
𝑗=1,𝑗=𝑖̸ 󵄨0
𝑗=1,𝑗=𝑖̸ 󵄨󵄨
󵄨󵄨 𝑛
or ≤ 𝑟𝑖 + 󵄨󵄨󵄨󵄨𝑎𝑖𝑖 (𝑡) 𝑒𝛼𝑖𝑖 𝑥𝑖 (𝑡𝑖 ) + ∑ 𝑎𝑖𝑗 (𝑡) 𝑒𝛼𝑖𝑗 𝑥𝑗 (𝑡)
󵄨󵄨
󵄨 𝑗=1,𝑗=𝑖̸
𝑛 𝑎𝑖𝑗 + 𝑏𝑖𝑗 + 𝑐𝑖𝑗
𝑧𝑖 (𝑡𝑖 ) − ∑ 𝑧𝑗 (𝑡𝑗 ) ≤ 𝑟𝑖 , (22) 𝑛
𝑗=1,𝑗=𝑖̸ 𝑎𝑗𝑗 + ∑ 𝑏𝑖𝑗 (𝑡) 𝑒𝛽𝑖𝑗 𝑥𝑗 (𝑡𝑗 ) (29)
𝑗=1,𝑗=𝑖̸
which implies
󵄨󵄨
𝑛 ∞ 󵄨󵄨
1
𝑎 + 𝑏12 + 𝑐12
− 12
𝑎 + 𝑏1𝑛 + 𝑐1𝑛
⋅ ⋅ ⋅ − 1𝑛 + ∑ 𝑐𝑖𝑗 (𝑡) ∫ 𝐻𝑖𝑗 (𝑠) 𝑒𝛾𝑖𝑗 𝑥𝑗 (𝑡𝑖 +𝑠) 𝑑𝑠󵄨󵄨󵄨󵄨
𝑎22 𝑎𝑛𝑛 0 󵄨󵄨
𝑗=1,𝑗=𝑖̸ 󵄨0
(− 𝑎21 + 𝑏21 + 𝑐21 𝑎2𝑛 + 𝑏2𝑛 + 𝑐2𝑛 )
( 1 ⋅⋅⋅ − )
𝑎 𝑎 𝑛 𝑎𝑖𝑗 + 𝑏𝑖𝑗 + 𝑐𝑖𝑗 𝑎𝑖𝑖
⋅ ⋅ 11
⋅ ⋅⋅⋅ ⋅⋅⋅ ⋅ ⋅ 𝑛𝑛
⋅ = 𝑟𝑖 + ∑ 𝑧𝑗 (𝑡𝑗 ) + 𝑧 (𝑡 )
𝑎𝑛1 + 𝑏𝑛1 + 𝑐𝑛1 𝑎 + 𝑏𝑛2 + 𝑐𝑛2 𝑗=1,𝑗=𝑖̸ 𝑎𝑗𝑗 𝑎𝑖𝑖 𝑖 𝑖
− − 𝑛2 ⋅⋅⋅ 1
( 𝑎11 𝑎22 ) 𝑛
𝑎𝑖𝑖 𝑎
𝑧1 (𝑡1 ) 𝑟1 ≤ 𝐷𝑖 + ∑Γ𝑖𝑗 ̃ℎ𝑗 + 𝑧𝑖 (𝑡𝑖 ) ≤ ̃ℎ𝑖 + 𝑖𝑖 ̃ℎ𝑖
𝑗=1 𝑎𝑖𝑖 𝑎𝑖𝑖
𝑧2 (𝑡2 ) 𝑟
×( ) ≤ ( 2)
⋅⋅⋅ ⋅⋅⋅ 𝑎𝑖𝑖 ̃
𝑧𝑛 (𝑡𝑛 ) 𝑟𝑛 = [1 + ]ℎ .
(23) 𝑎𝑖𝑖 𝑖

Set 𝐷 = (𝐷1 , 𝐷2 , . . . , 𝐷𝑛 )𝑇 = (𝑟1 , 𝑟2 , . . . , 𝑟𝑛 )𝑇 . It follows from We can choose a large enough real number (𝑑 > 1) such that
(23) that
(𝐸 − K) (𝑧1 (𝑡1 ), 𝑧2 (𝑡2 ), . . . , 𝑧𝑛 (𝑡𝑛 ))𝑇 ≤ 𝐷. 1 𝑑̃ℎ 1 ̃ℎ 𝑎
(24) ln 𝑖 > ln 𝑖 + [1 + 𝑖𝑖 ] ̃ℎ𝑖 . (30)
𝛼𝑖𝑖 𝑎𝑖𝑖 𝛼𝑖𝑖 𝑎𝑖𝑖 𝑎𝑖𝑖
−1
In view of 𝜌(K) < 1 and Lemma 3, (𝐸𝑛 − K) ≥ 0. Let
𝑇 Let ℎ𝑖 = (1/𝛼𝑖𝑖 ) ln(𝑑̃ℎ𝑖 /𝑎𝑖𝑖 ). Then, for any solution of 𝐿𝑥 =
𝐻 = (̃ℎ1 , ̃ℎ2 , . . . , ̃ℎ𝑛 ) := (𝐸 − K)−1 𝐷 ≥ 0. (25)
𝜆𝑁𝑥, we have |𝑥𝑖 (𝑡)|1 = |𝑥𝑖 (𝑡)|0 + |𝑥𝑖̇ (𝑡)|0 ≤ ln(̃ℎ𝑖 /𝑎𝑖𝑖 ) + 2̃ℎ𝑖 <
Then it follows from (24) and (25) that ℎ𝑖 for all 𝑖 = 1, 2, . . . , 𝑛. Obviously, ℎ𝑖 are independent of 𝜆
and the choice of 𝑥(𝑡). Thus, taking ℎ𝑖 = ln(𝑑̃ℎ𝑖 /𝑎𝑖𝑖 ), the open
(𝑧1 (𝑡1 ), 𝑧2 (𝑡2 ), . . . , 𝑧𝑛 (𝑡𝑛 ))𝑇 ≤ 𝐻, or
subset Ω satisfies that 𝐿𝑥 ≠𝜆𝑁𝑥 for each 𝜆 ∈ (0, 1), 𝑥 ∈ 𝜕Ω ∩
(26)
Dom 𝐿; that is, the open subset Ω satisfies the assumption (i)
𝑧𝑖 (𝑡𝑖 ) ≤ ̃ℎ𝑖 , 𝑖 = 1, 2, . . . , 𝑛,
of Lemma 1.
which implies Using similar arguments to Step 3 in [2], it is not difficult
to show that, for each 𝑥 ∈ 𝜕Ω ∩ Ker 𝐿, 𝑄𝑁𝑥 ≠ 0 and
󵄨 󵄨 1 ̃ℎ deg{𝐽𝑄𝑁, Ω ∩ Ker 𝐿, 0} ≠0.
|𝑥𝑖 (𝑡)|0 = max 󵄨󵄨󵄨𝑥𝑖 (𝑡)󵄨󵄨󵄨 = 𝑥𝑖 (𝑡𝑖 ) ≤ ln 𝑖
,
𝑡∈[0,𝜔] 𝛼𝑖𝑖 𝑎𝑖𝑖 + 𝑏𝑖𝑖 (27) Hence, by Lemma 1, system (8) has at least one positive
𝜔-periodic solution in Dom 𝐿 ∩ Ω. By (7), system (1) has at
𝑖 = 1, 2, . . . , 𝑛. least one positive 𝜔-periodic solution, denoted by 𝑦(𝑡). ̃ This
completes the proof of Theorem 4.
On the other hand, it follows from (25) that
(𝐸 − K) 𝐻 = 𝐷, or 𝐻 = K𝐻 + 𝐷, that is,
Conflict of Interests
𝑛 (28)
̃ℎ = ∑ Γ ̃ℎ + 𝐷 , 𝑖 = 1, 2, . . . , 𝑛. The authors declare that there is no conflict of interests
𝑖 𝑖𝑗 𝑗 𝑖
𝑗=1 regarding the publication of this paper.
Abstract and Applied Analysis 5

References
[1] M. Fan, K. Wang, and D. Jiang, “Existence and global attractivity
of positive periodic solutions of periodic n-species Lotka-
Volterra competition systems with several deviating arguments,”
Mathematical Biosciences, vol. 160, no. 1, pp. 47–61, 1999.
[2] Y. H. Xia and M. Han, “New conditions on the existence
and stability of periodic solution in lotka-volterra's population
system,” SIAM Journal on Applied Mathematics, vol. 69, no. 6,
pp. 1580–1597, 2009.
[3] Y. Xia, “Periodic solution of certain nonlinear differential
equations: via topological degree theory and matrix spectral
theory,” International Journal of Bifurcation and Chaos in
Applied Sciences and Engineering, vol. 22, no. 8, Article ID
1250196, 17 pages, 2012.
[4] X. Gu, H. Wang, P. J. Y. Wong, and Y. Xia, “Existence and
stability of periodic solution to delayed nonlinear differential
equations,” Abstract and Applied Analysis, vol. 2014, Article ID
156948, 12 pages, 2014.
[5] F. Yang, D. Jiang, and A. Ying, “Existence of positive solution of
multidelays facultative mutualism system,” Journal of Engineer-
ing Mathematics, vol. 3, pp. 64–68, 2002.
[6] F. Chen, J. Shi, and X. Chen, “Periodicity in a Lotka-Volterra
facultative mutualism system with several delays,” Journal of
Engineering Mathematics, vol. 21, no. 3, pp. 403–409, 2004.
[7] Y. H. Xia, J. Cao, and S. S. Cheng, “Periodic solutions for a
Lotka-Volterra mutualism system with several delays,” Applied
Mathematical Modelling, vol. 31, no. 9, pp. 1960–1969, 2007.
[8] R. E. Gaines and J. L. Mawhin, Coincidence Degree and Nonlin-
ear Differential Equations, Springer, Berlin, Germany, 1977.final
[9] J. P. LaSalle, The Stability of Dynamical System, Society for
Industrial and Applied Mathematics, Philadelphia, Pa, USA,
1976.
[10] A. Berman and R. J. Plemmons, Nonnegative Matrices in the
Mathematical Science, Academic Press, New York, NY, USA,
1929.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 241650, 9 pages
http://dx.doi.org/10.1155/2014/241650

Research Article
Global and Blow-Up Solutions for a Class of Nonlinear
Parabolic Problems under Robin Boundary Condition

Lingling Zhang and Hui Wang

Department of Mathematics, Taiyuan University of Technology, Shanxi 030024, China

Correspondence should be addressed to Lingling Zhang; zhangllwangh@126.com

Received 26 June 2014; Accepted 28 July 2014; Published 14 August 2014

Academic Editor: Sanling Yuan

Copyright © 2014 L. Zhang and H. Wang. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.

We discuss the global and blow-up solutions of the following nonlinear parabolic problems with a gradient term under Robin
boundary conditions: (𝑏(𝑢))𝑡 = ∇ ⋅ (ℎ(𝑡)𝑘(𝑥)𝑎(𝑢)∇𝑢) + 𝑓(𝑥, 𝑢, |∇𝑢|2 , 𝑡), in 𝐷 × (0, 𝑇), (𝜕𝑢/𝜕𝑛) + 𝛾𝑢 = 0, on 𝜕𝐷 × (0, 𝑇), 𝑢(𝑥, 0) =
𝑢0 (𝑥) > 0, in 𝐷, where 𝐷 ⊂ R𝑁 (𝑁 ≥ 2) is a bounded domain with smooth boundary 𝜕𝐷. Under some appropriate assumption on
the functions 𝑓, ℎ, 𝑘, 𝑏, and 𝑎 and initial value 𝑢0 , we obtain the sufficient conditions for the existence of a global solution, an upper
estimate of the global solution, the sufficient conditions for the existence of a blow-up solution, an upper bound for “blow-up time,”
and an upper estimate of “blow-up rate.” Our approach depends heavily on the maximum principles.

1. Introduction 𝑇 is the maximal existence time of 𝑢, and 𝐷 is the closure of

𝐷. Set R+ = (0, +∞). We assume, throughout the paper, that
The study of global and blow-up solutions for nonlinear 𝑏(𝑠) is a positive 𝐶3 (R+ ) function, 𝑏󸀠 (𝑠) > 0 for any 𝑠 ∈ R+ ,
parabolic equations has received a lot of attention in the past
𝑎(𝑠) is a positive 𝐶2 (R+ ) function, 𝑘(𝑥) is a positive 𝐶1 (𝐷)
several decades (see [1–4]). In most works, so far, a variety
function, ℎ(𝑡) is a positive 𝐶1 (R+ ) function, 𝑓(𝑥, 𝑠, 𝑑, 𝑡) is
of approaches have been developed in dealing with different
nonlinear parabolic problems, such as the existence of global a nonnegative 𝐶1 (𝐷 × R+ × R+ × R+ ) function, and 𝑢0 (𝑥)
solution, blow-up solution, an upper bound for “blow-up is a positive 𝐶2 (𝐷) function. Under these assumptions, the
time,” an upper estimate of “blow-up rate,” or global solution. classical parabolic equation theory [12] ensures that there
So far, some applications in physics, chemistry, and biology exists a unique classical solution 𝑢(𝑥, 𝑡) with some 𝑇 > 0 for
are relevant to blow-up phenomena which can be found in the problem (1), and the solution is positive over 𝐷 × [0, 𝑇).
[5–11]. In this paper, we consider the global and blow-up Moreover, by the regularity theorem [13], 𝑢(𝑥, 𝑡) ∈ 𝐶3 (𝐷 ×
solutions of the following nonlinear parabolic equation with (0, 𝑇)) ∩ 𝐶2 (𝐷 × [0, 𝑇)).
Robin boundary condition: The problems of the global and blow-up solutions for
nonlinear parabolic equations have been investigated exten-
(𝑏 (𝑢))𝑡 = ∇ ⋅ (ℎ (𝑡) 𝑘 (𝑥) 𝑎 (𝑢) ∇𝑢) + 𝑓 (𝑥, 𝑢, 𝑞, 𝑡) , sively by many authors and have got a lot of meaningful
in 𝐷 × (0, 𝑇) , results. Some special cases of problem (1) have been treated
already. Ding [14] deals with the following problem:
𝜕𝑢 (1)
+ 𝛾𝑢 = 0, on 𝜕𝐷 × (0, 𝑇) , (𝑏 (𝑢))𝑡 = ∇ ⋅ (𝑎 (𝑢) ∇𝑢) + 𝑓 (𝑢) , in 𝐷 × (0, 𝑇) ,
𝜕𝑛
𝜕𝑢
𝑢 (𝑥, 0) = 𝑢0 (𝑥) > 0, in 𝐷, + 𝛾𝑢 = 0, on 𝜕𝐷 × (0, 𝑇) ,
𝜕𝑛
where 𝑞 = |∇𝑢|2 , 𝐷 ⊂ R𝑁 (𝑁 ≥ 2) is a bounded domain with 𝑢 (𝑥, 0) = ℎ (𝑥) > 0, in 𝐷,
smooth boundary 𝜕𝐷, 𝜕𝑢/𝜕𝑛 represents the outward normal
derivative on 𝜕𝐷, 𝛾 is positive constant, 𝑢0 is the initial value, (2)
2 Abstract and Applied Analysis

where 𝐷 is a bounded domain of R𝑁 (𝑁 ≥ 2) with (i) For any 𝑠 ∈ R+ ,

smooth boundary 𝜕𝐷. By constructing auxiliary functions
󸀠
and using a first-order differential inequality technique, Ding (𝑠𝑏󸀠 (𝑠)) ≥ 0,
derives conditions on the data, which guarantee the existence
of blow-up or global solution. The following problem is 󸀠
𝑠𝑏󸀠 (𝑠) − (𝑠𝑏󸀠 (𝑠)) ≤ 0,
investigated by Enache in [15]:
𝑎 (𝑠) 󸀠
𝑢𝑡 = ∇ ⋅ (𝑎 (𝑢) ∇𝑢) + 𝑓 (𝑢) , in 𝐷 × (0, 𝑇) , ( ) ≤ 0,
𝑏󸀠 (𝑠)
𝜕𝑢 󸀠
+ 𝛾𝑢 = 0, on 𝜕𝐷 × (0, 𝑇) , (3) 1 𝑎 (𝑠) 󸀠 1 1 𝑎 (𝑠) 󸀠 1
𝜕𝑛 [ ( 󸀠 ) + 󸀠 ] + ( 󸀠 ) + 󸀠 ≤ 0.
𝑎 (𝑠) 𝑏 (𝑠) 𝑏 (𝑠) 𝑎 (𝑠) 𝑏 (𝑠) 𝑏 (𝑠)
𝑢 (𝑥, 0) = ℎ (𝑥) > 0, in 𝐷, (5)

where 𝐷 is a bounded domain of R𝑁 (𝑁 ≥ 2) with smooth (ii) For any (𝑥, 𝑠, 𝑑, 𝑡) ∈ 𝐷 × R+ × R+ × R+ ,

boundary 𝜕𝐷. By constructing auxiliary functions and first-
order differential inequality technique, Enache establishes 𝑓 (𝑥, 𝑠, 𝑑, 𝑡)
some conditions on nonlinearities and the initial date to ( ) ≤ 0,
ℎ (𝑡) 𝑡
guarantee that 𝑢(𝑥, 𝑡) exists for all times 𝑡 > 0 or blows
󸀠
up at some finite time 𝑇. Besides, the following problem is 1 1
investigated by Zhang in [16]: 𝑓𝑑 (𝑥, 𝑠, 𝑑, 𝑡) [( ) + 󸀠 ] ≤ 0,
𝑏󸀠 (𝑠) 𝑏 (𝑠)
(6)
(𝑏 (𝑢))𝑡 = Δ𝑢 + 𝑓 (𝑢) , in 𝐷 × (0, 𝑇) , 𝑓 (𝑥, 𝑠, 𝑑, 𝑡) 𝑏󸀠 (𝑠) 𝑓 (𝑥, 𝑠, 𝑑, 𝑡) 𝑏󸀠 (𝑠)
( ) −
𝑎 (𝑠) 𝑠
𝑎 (𝑠)
𝜕𝑢
+ 𝛾𝑢 = 0, on 𝜕𝐷 × (0, 𝑇) , (4) 2
𝜕𝑛 ℎ󸀠 (𝑡) (𝑏󸀠 (𝑠))
+ ≤ 0.
𝑢 (𝑥, 0) = ℎ (𝑥) > 0, in 𝐷, 𝑎 (𝑠) ℎ (𝑡)

where 𝐷 is a bounded domain in R𝑁 (𝑁 ≥ 2) with smooth (iii) Consider the integration

boundary. Under appropriate assumptions on the functions
𝑓, 𝑏, and ℎ, Zhang obtains the conditions under which the
+∞
𝑏󸀠 (𝑠)
∫ d𝑠 = +∞, 𝑚0 = min 𝑢0 (𝑥) . (7)
solutions may exist globally or blow up in a finite time. 𝑚0 e𝑠 𝐷
Moreover, upper estimates of the “blow-up time,” blow-up
rate, and global solutions are obtained also. (iv) Consider
In this paper, we obtain the existence theorem of global
and blow-up solution by constructing completely different ∇ ⋅ (ℎ (0) 𝑘 (𝑥) 𝑎 (𝑢0 ) ∇𝑢0 ) + 𝑓 (𝑥, 𝑢0 , 𝑞0 , 0)
auxiliary functions and technically using maximum princi- 𝛼 = max { } > 0,
𝐷 𝑒𝑢0
ples. As a result, the sufficient conditions for the existence
of a global solution and an upper estimate of the global 󵄨 󵄨2
𝑞0 = 󵄨󵄨󵄨∇𝑢0 󵄨󵄨󵄨 .
solution and the sufficient conditions for the existence of a
blow-up solution, an upper bound for “blow-up time,” and (8)
an upper estimate of “blow-up rate” are specified under some
appropriate assumption on the functions 𝑓, ℎ, 𝑘, 𝑏, and 𝑎 Then the solution 𝑢 to problem (1) must be a global solution
and initial value 𝑢0 . Our results extend and supplement those and
obtained in [14–16].
The content of this paper is organized as follows. In 𝑢 (𝑥, 𝑡) ≤ 𝐻−1 (𝛼𝑡 + 𝐻 (𝑢0 (𝑥, 𝑡))) , (𝑥, 𝑡) ∈ 𝐷 × R+ , (9)
Section 2, we study the existence of the global solution of (1).
In Section 3, we investigate the blow-up solution of (1). In where
Section 4, we will give a few examples to explain our results.
𝑧
𝑏󸀠 (𝑠)
𝐻 (𝑧) = ∫ d𝑠, 𝑧 ≥ 𝑚0 , (10)
𝑚0 e𝑠
2. Global Solution
Our main result for the global solution is the following and 𝐻−1 is the inverse function of 𝐻.
Theorem 1.
Proof. Consider the auxiliary function
Theorem 1. Let 𝑢 be a solution of (1). Suppose that the
following conditions (𝑖)–(𝑖𝑣) are satisfied. 𝑃 (𝑥, 𝑡) = 𝑏󸀠 (𝑢) 𝑢𝑡 − 𝛼𝑒𝑢 . (11)
Abstract and Applied Analysis 3

Then, we have 𝑎𝑏󸀠󸀠

+ (𝑎󸀠 − ) ℎ𝑢𝑡 (∇𝑘 ⋅ ∇𝑢) + 𝑎ℎ (∇𝑘 ⋅ ∇𝑢𝑡 )
𝑏󸀠
∇𝑃 = 𝑏󸀠󸀠 𝑢𝑡 ∇𝑢 + 𝑏󸀠 ∇𝑢𝑡 − 𝛼𝑒𝑢 ∇𝑢, (12)
+ 𝑎ℎ󸀠 (∇𝑘 ⋅ ∇𝑢)
󸀠󸀠󸀠 2 󸀠󸀠
Δ𝑃 = 𝑏 𝑢𝑡 |∇𝑢| + 2𝑏 ∇𝑢 ⋅ ∇𝑢𝑡
(13) 𝑓𝑏󸀠󸀠
󸀠󸀠 󸀠
+ 𝑏 𝑢𝑡 Δ𝑢 + 𝑏 Δ𝑢𝑡 − 𝛼𝑒 |∇𝑢| − 𝛼𝑒 Δ𝑢. 𝑢 2 𝑢 + (𝑓𝑢 − − 𝛼𝑒𝑢 ) 𝑢𝑡 + 𝑓𝑡 .
𝑏󸀠
(16)
By (1),

(𝑏 (𝑢))𝑡 = 𝑏󸀠 𝑢𝑡 = ∇ ⋅ (ℎ (𝑡) 𝑘 (𝑥) 𝑎 (𝑢) ∇𝑢) + 𝑓 By (13) and (16), it follows that

= ℎ (𝑡) 𝑘 (𝑥) 𝑎 (𝑢) Δ𝑢 + ℎ (𝑡) 𝑘 (𝑥) 𝑎󸀠 (𝑢) |∇𝑢|2 (14) 𝑎𝑘ℎ

Δ𝑃 − 𝑃𝑡
+ ℎ (𝑡) 𝑎 (𝑢) (∇𝑘 ⋅ ∇𝑢) + 𝑓. 𝑏󸀠
𝑎𝑘ℎ𝑏󸀠󸀠󸀠 𝑎󸀠 𝑘ℎ𝑏󸀠󸀠
We have =( + − 𝑎󸀠󸀠 𝑘ℎ) 𝑢𝑡 |∇𝑢|2
𝑏󸀠 𝑏󸀠
𝑎𝑘ℎ 𝑎󸀠 𝑘ℎ 𝑎ℎ 𝑓 2𝑎𝑘ℎ𝑏󸀠󸀠
𝑢𝑡 = 󸀠
Δ𝑢 + 󸀠
|∇𝑢|2 + 󸀠 (∇𝑘 ⋅ ∇𝑢) + 󸀠 , +( − 2𝑎󸀠 𝑘ℎ − 2𝑓𝑞 ) (∇𝑢 ⋅ ∇𝑢𝑡 )
𝑏 𝑏 𝑏 𝑏 𝑏󸀠
𝑎𝑘 𝑎󸀠 𝑘 𝑎
(𝑢𝑡 )𝑡 = ℎ󸀠 ( Δ𝑢 + |∇𝑢|2 + 󸀠 (∇𝑘 ⋅ ∇𝑢)) 2𝑎𝑘ℎ𝑏󸀠󸀠
𝑏󸀠 𝑏󸀠 𝑏 +( − 𝑎󸀠 𝑘ℎ) 𝑢𝑡 Δ𝑢
𝑏󸀠
𝑎𝑘 𝑎󸀠 𝑘 𝑎
+ ℎ( Δ𝑢 + 󸀠 |∇𝑢|2 + 󸀠 (∇𝑘 ⋅ ∇𝑢)) 𝑎𝑘ℎ 𝑢 𝑎𝑘ℎ
𝑏󸀠 𝑏 𝑏 −( 󸀠
𝛼𝑒 + 𝑎󸀠 ℎ󸀠 𝑘) |∇𝑢|2 − ( 󸀠 𝛼𝑒𝑢 + 𝑎𝑘ℎ󸀠 ) Δ𝑢
𝑡 𝑏 𝑏
𝑓 𝑓𝑏󸀠󸀠
+( ) − 𝑏󸀠󸀠 (𝑢𝑡 )2 + ( + 𝛼𝑒𝑢 − 𝑓𝑢 ) 𝑢𝑡
𝑏󸀠 𝑡 𝑏󸀠
𝑎󸀠󸀠 𝑎󸀠 𝑏󸀠󸀠 2 2𝑎󸀠 𝑘ℎ 𝑎ℎ𝑏󸀠󸀠
=( − ) ℎ𝑘𝑢𝑡 |∇𝑢| + (∇𝑢 ⋅ ∇𝑢𝑡 ) +( − 𝑎󸀠 ℎ) 𝑢𝑡 (∇𝑘 ⋅ ∇𝑢)
𝑏󸀠 (𝑏󸀠 )2 𝑏󸀠 𝑏󸀠
(15)
𝑎󸀠 𝑎𝑏󸀠󸀠 − 𝑎ℎ (∇𝑘 ⋅ ∇𝑢𝑡 ) − 𝑎ℎ󸀠 (∇𝑘 ⋅ ∇𝑢) − 𝑓𝑡 .
+( 󸀠 − 2
) ℎ𝑢𝑡 (∇𝑘 ⋅ ∇𝑢)
𝑏 (𝑏󸀠 ) (17)

𝑎ℎ 𝑎󸀠 𝑎𝑏󸀠󸀠 By (14), we have

+ (∇𝑘 ⋅ ∇𝑢𝑡 ) + ( − ) ℎ𝑘𝑢𝑡 Δ𝑢
𝑏󸀠 𝑏󸀠 (𝑏󸀠 )2
𝑏󸀠 𝑎󸀠 1 𝑓
𝑎𝑘ℎ 𝑎󸀠 𝑘ℎ󸀠 𝑎ℎ󸀠 Δ𝑢 = 𝑢𝑡 − |∇𝑢|2 − (∇𝑘 ⋅ ∇𝑢) − . (18)
+ 󸀠 Δ𝑢𝑡 + 󸀠 |∇𝑢|2 + 󸀠 (∇𝑘 ⋅ ∇𝑢) 𝑎𝑘ℎ 𝑎 𝑘 𝑎𝑘ℎ
𝑏 𝑏 𝑏
𝑎𝑘ℎ󸀠 Substitute (18) into (17) to get
+ Δ𝑢
𝑏󸀠
𝑓𝑢 𝑢𝑡 + 2𝑓𝑞 (∇𝑢 ⋅ ∇𝑢𝑡 ) + 𝑓𝑡 𝑎𝑘ℎ
𝑓𝑏󸀠󸀠 𝑢𝑡 Δ𝑃 − 𝑃𝑡
+ − 2
. 𝑏󸀠
𝑏󸀠 (𝑏󸀠 )
󸀠 2
𝑎𝑘ℎ𝑏󸀠󸀠󸀠 (𝑎 ) 𝑘ℎ 𝑎󸀠 𝑘ℎ𝑏󸀠󸀠
Then =( + − − 𝑎󸀠󸀠 𝑘ℎ) 𝑢𝑡 |∇𝑢|2
𝑏󸀠 𝑎 𝑏󸀠
2
𝑃𝑡 = 𝑏󸀠󸀠 (𝑢𝑡 ) + 𝑏󸀠 (𝑢𝑡 )𝑡 − 𝛼𝑒𝑢 𝑢𝑡
2𝑎𝑘ℎ𝑏󸀠󸀠
󸀠󸀠 +( − 2𝑎󸀠 𝑘ℎ − 2𝑓𝑞 ) (∇𝑢 ⋅ ∇𝑢𝑡 )
2 𝑎𝑏 𝑏󸀠
= 𝑏󸀠󸀠 (𝑢𝑡 ) + (𝑎󸀠 − ) 𝑘ℎ𝑢𝑡 Δ𝑢 + 𝑎𝑘ℎΔ𝑢𝑡
𝑏󸀠
𝑎󸀠 𝑏󸀠 𝑎󸀠 𝑓 𝑓𝑏󸀠󸀠 𝑏󸀠 ℎ󸀠
󸀠 󸀠󸀠 + (𝑏󸀠󸀠 − ) 𝑢𝑡2 + ( − 󸀠 − 𝑓𝑢 − ) 𝑢𝑡
𝑎𝑏 𝑎 𝑎 𝑏 ℎ
+ 𝑎𝑘ℎ󸀠 Δ𝑢 + (𝑎󸀠󸀠 − ) 𝑘ℎ𝑢𝑡 |∇𝑢|2 + 𝑎󸀠 𝑘ℎ󸀠 |∇𝑢|2
𝑏󸀠
𝑎󸀠 𝑘ℎ 𝑢 𝑎𝑘ℎ 𝑢
󸀠 +( 𝛼𝑒 − 󸀠 𝛼𝑒 ) |∇𝑢|2
+ (2𝑎 𝑘ℎ + 2𝑓𝑞 ) (∇𝑢 ⋅ ∇𝑢𝑡 ) 𝑏󸀠 𝑏
4 Abstract and Applied Analysis

𝑎ℎ𝑏󸀠󸀠 According to (11), we have

− 𝑢 (∇𝑘 ⋅ ∇𝑢) − 𝑎ℎ (∇𝑘 ⋅ ∇𝑢𝑡 )
𝑏󸀠 𝑡
𝑎ℎ 𝑢 𝑓 𝑢 𝑓ℎ󸀠 1 𝑒𝑢
+ 𝛼𝑒 (∇𝑘 ⋅ ∇𝑢) + 𝛼𝑒 + − 𝑓𝑡 . 𝑢𝑡 = 𝑃 + 𝛼 . (23)
𝑏󸀠 𝑏󸀠 ℎ 𝑏󸀠 𝑏󸀠
(19)
By (12), we have Substituting (23) into (22), we have

1 𝑏󸀠󸀠 𝑒𝑢
∇𝑢𝑡 =∇𝑃 − 𝑢𝑡 ∇𝑢 + 𝛼 ∇𝑢. (20) 𝑎𝑘ℎ 𝑎 󸀠 2𝑓𝑞 𝑎ℎ
𝑏󸀠 𝑏󸀠 𝑏󸀠 Δ𝑃 + [2𝑘ℎ( 󸀠 ) + 󸀠 ] (∇𝑢 ⋅ ∇𝑃) + 󸀠 (∇𝑘 ⋅ ∇𝑃)
𝑏󸀠 𝑏 𝑏 𝑏
Next, we substitute (20) into (19) to obtain
𝑎𝑘ℎ 𝑎 𝑓𝑏󸀠 ℎ󸀠
Δ𝑃 − 𝑃𝑡 +[ ( ) + ]𝑃
𝑏󸀠 2 𝑎 𝑢 ℎ
(𝑏󸀠 )
2𝑎𝑘ℎ𝑏󸀠󸀠 2𝑎󸀠 𝑘ℎ 2𝑓𝑞 𝑎ℎ 󸀠
=( 2
− − 󸀠 ) (∇𝑢 ⋅ ∇𝑃) − 󸀠 (∇𝑘 ⋅ ∇𝑃) 1 𝑎 󸀠 1 󸀠
(𝑏󸀠 ) 𝑏󸀠 𝑏 𝑏 + [𝑎𝑘ℎ( ( 󸀠 ) ) + 2𝑓𝑞 ( 󸀠 ) ] |∇𝑢|2 𝑃 − 𝑃𝑡
𝑎 𝑏 𝑏
󸀠󸀠 2
𝑎𝑘ℎ𝑏󸀠󸀠󸀠 2𝑎𝑘ℎ(𝑏 ) 𝑎󸀠 𝑘ℎ𝑏󸀠󸀠 2𝑓𝑞 2𝑏󸀠󸀠 𝑓𝑞
+( − 2
+ − 𝑎󸀠󸀠 𝑘ℎ = (−𝛼𝑒 ) ( 𝑢
− ) |∇𝑢|2
𝑏󸀠 (𝑏󸀠 ) 𝑏󸀠 𝑏󸀠 2
(𝑏󸀠 )
2
(𝑎󸀠 ) 𝑘ℎ 2𝑏󸀠󸀠 𝑓𝑞 𝑎󸀠 𝑎𝑏󸀠󸀠
+ + ) 𝑢𝑡 |∇𝑢| 2 + (−𝛼𝑒𝑢 ) 𝑘ℎ ( − ) |∇𝑢|2
𝑎 𝑏󸀠 𝑏󸀠 (𝑏󸀠 )2
2
2𝑎𝑘ℎ𝑏󸀠󸀠 𝑎󸀠 𝑘ℎ 𝑎𝑘ℎ 2𝑓𝑞 𝑎𝑘ℎ𝑏󸀠󸀠󸀠 2𝑎𝑘ℎ(𝑏󸀠󸀠 ) 𝑎󸀠 𝑘ℎ𝑏󸀠󸀠 𝑎󸀠󸀠 𝑘ℎ
+( 2
𝛼𝑒 − 󸀠 𝛼𝑒𝑢 − 󸀠 𝛼𝑒𝑢 − 󸀠 𝛼𝑒𝑢 ) |∇𝑢|2
𝑢 𝑢
+ 𝛼𝑒 ( 2
− 3
+ 2
−
(𝑏󸀠 ) 𝑏 𝑏 𝑏 (𝑏󸀠 ) (𝑏󸀠 ) (𝑏󸀠 ) 𝑏󸀠

𝑎󸀠 𝑏󸀠 𝑎󸀠 𝑓 𝑓𝑏󸀠󸀠 𝑏󸀠 ℎ󸀠 2
+ (𝑏󸀠󸀠 − ) 𝑢𝑡2 + ( − 󸀠 − 𝑓𝑢 − ) 𝑢𝑡 (𝑎󸀠 ) 𝑘ℎ
𝑎 𝑎 𝑏 ℎ + ) |∇𝑢|2
𝑎𝑏󸀠
𝑓 𝑢 𝑓ℎ󸀠
+ 𝛼𝑒 + − 𝑓𝑡 .
𝑏󸀠 ℎ 1 𝑏󸀠󸀠 𝑎󸀠 𝑏󸀠
(21) − 𝛼𝑒𝑢 𝑎𝑘ℎ ( 󸀠
− 2
) |∇𝑢|2 + (𝑏󸀠󸀠 − ) 𝑢𝑡2
𝑏 󸀠
(𝑏 ) 𝑎
So we have
𝑎𝑘ℎ 2𝑎󸀠 𝑘ℎ 2𝑓𝑞 2𝑎𝑘ℎ𝑏󸀠󸀠 𝑎󸀠 𝑓 𝑓𝑏󸀠󸀠 𝑓𝑢 𝑓 ℎ󸀠 ℎ󸀠
Δ𝑃 + ( + 󸀠 − ) + 𝛼𝑒𝑢 ( − − + − ) + 𝑓 − 𝑓𝑡 .
𝑏󸀠 𝑏󸀠 𝑏 2 𝑎𝑏󸀠 (𝑏󸀠 )2 𝑏󸀠 𝑏󸀠 ℎ ℎ
(𝑏󸀠 )
(24)
𝑎ℎ
× (∇𝑢 ⋅ ∇𝑃) + (∇𝑘 ⋅ ∇𝑃) − 𝑃𝑡
𝑏󸀠
2 2
Namely,
𝑎𝑘ℎ𝑏 󸀠󸀠󸀠 2𝑎𝑘ℎ(𝑏󸀠󸀠 ) 󸀠
𝑎 𝑘ℎ𝑏 󸀠󸀠 (𝑎󸀠 ) 𝑘ℎ
=( − 2
+ − 𝑎󸀠󸀠 𝑘ℎ +
𝑏󸀠 (𝑏󸀠 ) 𝑏󸀠 𝑎 𝑎𝑘ℎ 𝑎 󸀠 2𝑓𝑞 𝑎ℎ
Δ𝑃 + [(2𝑘ℎ( ) + 󸀠 ) ∇𝑢 + 󸀠 ∇𝑘] ⋅ ∇𝑃
𝑏󸀠 𝑏󸀠 𝑏 𝑏
2𝑏󸀠󸀠 𝑓𝑞
+ ) 𝑢𝑡 |∇𝑢|2 𝑓𝑏󸀠
𝑏󸀠 𝑎 ℎ󸀠
+{ 2
( ) +
(𝑏󸀠 ) 𝑎 𝑢 ℎ
2𝑎𝑘ℎ𝑏󸀠󸀠 𝑎󸀠 𝑘ℎ 𝑢 𝑎𝑘ℎ 𝑢 2𝑓𝑞 𝑢
+( 𝛼𝑒𝑢 − 𝛼𝑒 − 󸀠 𝛼𝑒 − 󸀠 𝛼𝑒 ) |∇𝑢|2 󸀠
(𝑏󸀠 )
2 𝑏󸀠 𝑏 𝑏 1 𝑎 󸀠 1 󸀠
+ [𝑎𝑘ℎ( ( 󸀠 ) ) + 2𝑓𝑞 ( 󸀠 ) ] |∇𝑢|2 } 𝑃 − 𝑃𝑡
𝑎 𝑏 𝑏
𝑎󸀠 𝑏󸀠 𝑎󸀠 𝑓 𝑓𝑏󸀠󸀠 𝑏󸀠 ℎ󸀠
+ (𝑏󸀠󸀠 − ) 𝑢𝑡2 + ( − 󸀠 − 𝑓𝑢 − ) 𝑢𝑡
𝑎 𝑎 𝑏 ℎ 1 󸀠 1
= −𝛼𝑒𝑢 {2𝑓𝑞 [( ) + 󸀠 ] + 𝑎𝑘ℎ
𝑏󸀠 𝑏
𝑓 𝑢 𝑓ℎ󸀠
+ 𝛼𝑒 + − 𝑓𝑡 . 󸀠
𝑏󸀠 ℎ 1 𝑎 󸀠 1 1 𝑎 󸀠 1
(22) × [( ( 󸀠 ) + 󸀠 ) + ( 󸀠 ) + 󸀠 ]} |∇𝑢|2
𝑎 𝑏 𝑏 𝑎 𝑏 𝑏
Abstract and Applied Analysis 5

2 󸀠
(𝑏󸀠 ) 𝑎 󸀠 𝑎 = −𝛾(𝑢𝑏󸀠 ) 𝑢𝑡 + 𝛾𝛼𝑢𝑒𝑢
− ( 󸀠 ) 𝑢𝑡2 − 𝛼𝑒𝑢 2
𝑎 𝑏 (𝑏󸀠 ) 󸀠 1 1
= −𝛾(𝑢𝑏󸀠 ) ( 𝑃 + 𝛼 󸀠 𝑒𝑢 ) + 𝛾𝛼𝑢𝑒𝑢
󸀠 󸀠 2 𝑏󸀠 𝑏
[ 𝑓𝑏󸀠 𝑓𝑏󸀠 ℎ (𝑏 ) ] 𝑓 󸀠 󸀠
× ( ) − + − ℎ( ) . (𝑢𝑏󸀠 ) 𝑢𝑏󸀠 − (𝑢𝑏󸀠 )
𝑎 𝑢 𝑎 𝑎ℎ ℎ 𝑡 𝑢
[ ] = −𝛾 𝑃 + 𝛾𝛼𝑒 ,
𝑏󸀠 𝑏󸀠
(25)
on 𝜕𝐷 × (0, 𝑇) .
The assumptions (5) and (6) guarantee that the right-hand
(29)
side of (25) is nonnegative; that is,
󸀠
𝑎𝑘ℎ 𝑎 󸀠 2𝑓𝑞 𝑎ℎ Next, by using a part condition of (5) (𝑠𝑏󸀠 (𝑠)) ≥ 0, 𝑠𝑏󸀠 (𝑠) −
󸀠
Δ𝑃 + [(2𝑘ℎ( 󸀠 ) + 󸀠 ) ∇𝑢 + 󸀠 ∇𝑘] ⋅ ∇𝑃 󸀠
(𝑠𝑏󸀠 (𝑠)) ≤ 0 for any 𝑠 ∈ R+ , we can obtain
𝑏 𝑏 𝑏 𝑏

𝑓𝑏󸀠 𝜕𝑃 󵄨󵄨󵄨󵄨
𝑎 ℎ󸀠 󵄨 ≤ 0,
𝜕𝑛 󵄨󵄨󵄨(𝑥0 ,𝑡0 )
+{ ( ) + (30)
2 𝑎 𝑢 ℎ
(𝑏󸀠 )
󸀠
which contradicts with inequality (28). Thus, we know that
1 𝑎 󸀠 1 󸀠 the maximum of 𝑃 in 𝐷 × [0, 𝑇) is zero; that is,
+ [𝑎𝑘ℎ( ( 󸀠 ) ) + 2𝑓𝑞 ( 󸀠 ) ] |∇𝑢|2 } 𝑃 − 𝑃𝑡
𝑎 𝑏 𝑏
𝑃 ≤ 0, in 𝐷 × [0, 𝑇) . (31)
≥ 0, in 𝐷 × (0, 𝑇) .
(26) With (11), we know

By applying maximum principle (see [17]), it follows from 𝑏󸀠 (𝑢)

𝑢 ≤ 𝛼. (32)
(26) that 𝑃 can attain its nonnegative maximum only for 𝑒𝑢 𝑡
𝐷 × {0} or 𝜕𝐷 × (0, 𝑇). For each fixed 𝑥 ∈ 𝐷, we integrate (32) from 0 to 𝑡:
For 𝐷 × {0}, by (8), we have
𝑡
𝑏󸀠 (𝑢) 𝑢(𝑥,𝑡) 󸀠
𝑏 (𝑠)
max 𝑃 (𝑥, 0) ∫ 𝑢𝑡 𝑑𝑡 = ∫ 𝑑𝑠 ≤ 𝛼𝑡, (33)
𝐷 𝑒𝑢 𝑒𝑠
0 𝑢0 (𝑥)

= max {𝑏󸀠 (𝑢0 ) (𝑢0 )𝑡 − 𝛼𝑒𝑢0 } which implies that 𝑢 must be a global solution of (1). In fact,
𝐷 suppose that 𝑢 blows up at finite time 𝑇; then
= max {[∇ ⋅ (ℎ (0) 𝑘 (𝑥) 𝑎 (𝑢0 ) ∇𝑢0 ) lim 𝑢 (𝑥, 𝑡) = +∞.
𝐷
𝑡 → 𝑇−
(34)
+ 𝑓 (𝑥, 𝑢0 , 𝑞0 , 0)] − 𝛼𝑒𝑢0 } Passing to the limit as 𝑡 → 𝑇− in (33) yields
∇ ⋅ (ℎ (0) 𝑘 (𝑥) 𝑎 (𝑢0 ) ∇𝑢0 ) + 𝑓 (𝑥, 𝑢0 , 𝑞0 , 0) +∞
𝑏󸀠 (𝑠)
= max {𝑒𝑢0 [ ∫ 𝑑𝑠 ≤ 𝛼𝑇,
𝐷 𝑒𝑢0 𝑒𝑠
𝑢0 (𝑥)

−𝛼]} = 0.
+∞
𝑏󸀠 (𝑠) 𝑢0 (𝑥) 󸀠
𝑏 (𝑠) +∞ 󸀠
𝑏 (𝑠)
∫ 𝑠
𝑑𝑠 = ∫ 𝑠
𝑑𝑠 + ∫ 𝑠
𝑑𝑠 (35)
𝑚0 𝑒 𝑚0 𝑒 𝑢0 (𝑥) 𝑒
(27)
𝑢0 (𝑥)
𝑏󸀠 (𝑠)
For 𝜕𝐷 × (0, 𝑇), we claim that 𝑃 cannot take a positive ≤∫ 𝑑𝑠 + 𝛼𝑇 < +∞,
maximum at any point (𝑥, 𝑡). In fact, suppose that 𝑃 can take 𝑚0 𝑒𝑠
a positive maximum at one point (𝑥0 , 𝑡0 ) ∈ 𝜕𝐷 × (0, 𝑇); then which contradicts with the condition (iii). This shows that 𝑢
𝜕𝑃 󵄨󵄨󵄨󵄨 is global solution. Moreover, it follows from (33) that
𝑃 (𝑥0 , 𝑡0 ) > 0, 󵄨 > 0.
𝜕𝑛 󵄨󵄨󵄨(𝑥0 ,𝑡0 )
(28)
𝑢(𝑥,𝑡)
𝑏󸀠 (𝑠) 𝑢(𝑥,𝑡) 󸀠
𝑏 (𝑠) 𝑢0 (𝑥) 󸀠
𝑏 (𝑠)
∫ 𝑠
𝑑𝑠 = ∫ 𝑠
𝑑𝑠 − ∫ 𝑑𝑠
Combine (1) and (11) with (23); we have 𝑢0 (𝑥) 𝑒 𝑚0 𝑒 𝑚0 𝑒𝑠 (36)
𝜕𝑃 𝜕𝑢 𝜕𝑢 𝜕𝑢 = 𝐻 (𝑢 (𝑥, 𝑡)) − 𝐻 (𝑢0 (𝑥)) ≤ 𝛼𝑡.
= 𝑏󸀠󸀠 𝑢𝑡 + 𝑏󸀠 𝑡 − 𝛼𝑒𝑢
𝜕𝑛 𝜕𝑛 𝜕𝑛 𝜕𝑛
Since 𝐻 is an increasing function, we have
𝜕𝑢
= −𝛾𝑏 𝑢𝑢𝑡 + 𝑏 ( ) + 𝛾𝛼𝑢𝑒𝑢
󸀠󸀠 󸀠
𝜕𝑛 𝑡 𝑢 (𝑥, 𝑡) ≤ 𝐻−1 (𝛼𝑡 + 𝐻 (𝑢0 (𝑥))) . (37)

= −𝛾𝑏󸀠󸀠 𝑢𝑢𝑡 + 𝑏󸀠 (−𝛾𝑢)𝑡 + 𝛾𝛼𝑢𝑒𝑢 The proof is completed.

6 Abstract and Applied Analysis

3. Blow-Up Solution Proof. Construct the following auxiliary function:

The following theorem is the main result for the blow-up 𝑄 (𝑥, 𝑡) = 𝑏󸀠 (𝑢) 𝑢𝑡 − 𝛽𝑒𝑢 . (44)
solution of (1).
So we have
Theorem 2. Let 𝑢 be a solution of problem (1). Assume that the ∇𝑄 = 𝑏󸀠󸀠 𝑢𝑡 ∇𝑢 + 𝑏󸀠 ∇𝑢𝑡 − 𝛽𝑒𝑢 ∇𝑢,
following conditions (i)–(iv) are satisfied.
Δ𝑄 = 𝑏󸀠󸀠󸀠 𝑢𝑡 |∇𝑢|2 + 2𝑏󸀠󸀠 ∇𝑢 ⋅ ∇𝑢𝑡 + 𝑏󸀠󸀠 𝑢𝑡 Δ𝑢 + 𝑏󸀠 Δ𝑢𝑡 (45)
(i) For any 𝑠 ∈ R+ ,
− 𝛽𝑒𝑢 |∇𝑢|2 − 𝛽𝑒𝑢 Δ𝑢.
󸀠 󸀠 𝑎 (𝑠) 󸀠
(𝑠𝑏󸀠 (𝑠)) ≥ 0, 𝑠𝑏󸀠 (𝑠) − (𝑠𝑏󸀠 (𝑠)) ≥ 0, ( 󸀠 ) ≥ 0,
𝑏 (𝑠) As the previous derivation from (14) to (25), we can obtain
󸀠 𝑎𝑘ℎ
1 𝑎 (𝑠) 󸀠 1 1 𝑎 (𝑠) 󸀠 1 Δ𝑄
[ ( 󸀠 ) + 󸀠 ] + ( 󸀠 ) + 󸀠 ≥ 0. 𝑏󸀠
𝑎 (𝑠) 𝑏 (𝑠) 𝑏 (𝑠) 𝑎 (𝑠) 𝑏 (𝑠) 𝑏 (𝑠)
(38) 𝑎 󸀠 2𝑓𝑞 𝑎ℎ
+ [(2𝑘ℎ( 󸀠
) + 󸀠 ) ∇𝑢 + 󸀠 ∇𝑘] ⋅ ∇𝑄
𝑏 𝑏 𝑏
(ii) For any (𝑥, 𝑠, 𝑑, 𝑡) ∈ 𝐷 × R+ × R+ × R+ ,
𝑎 𝑓𝑏󸀠 ℎ󸀠
+{ ( ) +
𝑓 (𝑥, 𝑠, 𝑑, 𝑡) (𝑏󸀠 )
2 𝑎 𝑢 ℎ
( ) ≥ 0,
ℎ (𝑡) 𝑡
󸀠
1 𝑎 󸀠 1 󸀠
1 1 󸀠 + [𝑎𝑘ℎ( ( 󸀠 ) ) +2𝑓𝑞 ( 󸀠 ) ] |∇𝑢|2 } 𝑄 − 𝑄𝑡
𝑓𝑑 (𝑥, 𝑠, 𝑑, 𝑡) [( ) + 󸀠 ] ≥ 0, 𝑎 𝑏 𝑏
𝑏󸀠 (𝑠) 𝑏 (𝑠)
(39) 1 󸀠 1
󸀠
𝑓 (𝑥, 𝑠, 𝑑, 𝑡) 𝑏 (𝑠) 𝑓 (𝑥, 𝑠, 𝑑, 𝑡) 𝑏 (𝑠) 󸀠 = −𝛽𝑒𝑢 {2𝑓𝑞 [( ) + 󸀠]
( ) − 𝑏󸀠 𝑏
𝑎 (𝑠) 𝑠
𝑎 (𝑠) 󸀠
1 𝑎 󸀠 1 1 𝑎 󸀠 1
2 +𝑎𝑘ℎ[( ( 󸀠 ) + 󸀠 ) + ( 󸀠 ) + 󸀠 ]} |∇𝑢|2
ℎ󸀠 (𝑡) (𝑏󸀠 (𝑠)) 𝑎 𝑏 𝑏 𝑎 𝑏 𝑏
+ ≥ 0.
𝑎 (𝑠) ℎ (𝑡) 2
(𝑏󸀠 ) 𝑎 󸀠 2 𝑎
− ( 󸀠
) 𝑢𝑡 − 𝛽𝑒𝑢 2
(iii) Consider the integration 𝑎 𝑏 (𝑏󸀠 )
+∞
𝑏󸀠 (𝑠) 𝑓𝑏󸀠
󸀠 󸀠
𝑓𝑏󸀠 ℎ (𝑏 ) ] 𝑓
2
∫ d𝑠 < +∞, 𝑀0 = max 𝑢0 (𝑥) . (40)
𝑀0 𝑒𝑠 𝐷 × [( ) − + − ℎ( ) .
𝑎 𝑢 𝑎 𝑎ℎ ℎ 𝑡
[ ]
(iv) Consider (46)

∇ ⋅ (ℎ (0) 𝑘 (𝑥) 𝑎 (𝑢0 ) ∇𝑢0 ) + f (𝑥, 𝑢0 , 𝑞0 , 0) It is seen from (38) and (39) that the right-hand side of (46)
𝛽 = min { } > 0, is nonpositive; that is,
𝐷 𝑒𝑢0
𝑎𝑘ℎ
󵄨 󵄨2 Δ𝑄
𝑞0 = 󵄨󵄨󵄨∇𝑢0 󵄨󵄨󵄨 . 𝑏󸀠
(41)
𝑎 󸀠 2𝑓𝑞 𝑎ℎ
+ [(2𝑘ℎ( 󸀠
) + 󸀠 ) ∇𝑢 + 󸀠 ∇𝑘] ⋅ ∇𝑄
Then the solution 𝑢 of problem (1) must blow up in finite 𝑏 𝑏 𝑏
time 𝑇, and
𝑎 𝑓𝑏󸀠 ℎ󸀠
󸀠 +{ ( ) +
1 +∞ 𝑏 (𝑠) (𝑏󸀠 )
2 𝑎 𝑢 ℎ
𝑇≤ ∫ 𝑑𝑠,
𝛽 𝑀0 𝑒𝑠 (42) 󸀠
1 𝑎 󸀠 1 󸀠
−1
𝑢 (𝑥, 𝑡) ≤ 𝐺 (𝛽 (𝑇 − 𝑡)) , (𝑥, 𝑡) ∈ 𝐷 × [0, 𝑇) , + [𝑎𝑘ℎ( ( 󸀠 ) ) +2𝑓𝑞 ( 󸀠 ) ] |∇𝑢|2 } 𝑄 − 𝑄𝑡
𝑎 𝑏 𝑏
where ≤ 0, in 𝐷 × (0, 𝑇) .
+∞ 󸀠 (47)
𝑏 (𝑠)
𝐺 (𝑧) = ∫ 𝑑𝑠, 𝑧 > 0, (43)
𝑧 𝑒𝑠 By applying maximum principle (see [17]), it follows from
(47) that 𝑄 can attain its nonpositive minimum only for
and 𝐺−1 is the inverse function of 𝐺. 𝐷 × {0} or 𝜕𝐷 × (0, 𝑇).
Abstract and Applied Analysis 7

For 𝐷 × {0}, with (41), we have By integrating inequality (51) over [𝑡, 𝑠] (0 < 𝑡 < 𝑠 < 𝑇), for
each fixed 𝑥, we obtain
min 𝑄 (𝑥, 0)
𝐷 𝐺 (𝑢 (𝑥, 𝑡)) ≥ 𝐺 (𝑢 (𝑥, 𝑡)) − 𝐺 (𝑢 (𝑥, 𝑠))
󸀠 𝑢0
= min {𝑏 (𝑢0 ) (𝑢0 )𝑡 − 𝛽𝑒 } +∞
𝑏󸀠 (𝑠) +∞
𝑏󸀠 (𝑠)
𝐷 =∫ 𝑠
𝑑𝑠 − ∫ 𝑠
𝑑𝑠
𝑢(𝑥,𝑡) 𝑒 𝑢(𝑥,𝑠) 𝑒
= min {∇ ⋅ (ℎ (0) 𝑘 (𝑥) 𝑎 (𝑢0 ) ∇𝑢0 )
𝐷 𝑢(𝑥,𝑠)
𝑏󸀠 (𝑠) 𝑠 󸀠
𝑏 (𝑢)
=∫ 𝑑𝑠 = ∫ 𝑢 𝑑𝑡 ≥ 𝛽 (𝑠 − 𝑡) .
+𝑓 (𝑥, 𝑢0 , 𝑞0 , 0) − 𝛽𝑒𝑢0 } 𝑢(𝑥,𝑡) 𝑒𝑠
𝑡 𝑒𝑢 𝑡
(56)
∇ ⋅ (ℎ (0) 𝑘 (𝑥) 𝑎 (𝑢0 ) ∇𝑢0 ) + 𝑓 (𝑥, 𝑢0 , 𝑞0 , 0)
= min{𝑒𝑢0 [ Hence, by letting 𝑠 → 𝑇, we have
𝐷 𝑒𝑢0
𝐺 (𝑢 (𝑥, 𝑡)) ≥ 𝛽 (𝑇 − 𝑡) . (57)
−𝛽]} = 0.
Since 𝐺 is a decreasing function, we obtain
(48)
𝑢 (𝑥, 𝑡) ≤ 𝐺−1 (𝛽 (𝑇 − 𝑡)) . (58)
For 𝜕𝐷 × (0, 𝑇), substituting 𝑃 and 𝛼 with 𝑄 and 𝛽 in (29),
respectively, we have The proof is completed.

󸀠 󸀠
𝜕𝑄 (𝑢𝑏󸀠 ) 𝑢𝑏󸀠 − (𝑢𝑏󸀠 ) 4. Applications
= −𝛾 󸀠 𝑄 + 𝛾𝛽𝑒𝑢 , on 𝜕𝐷 × (0, 𝑇) .
𝜕𝑛 𝑏 𝑏󸀠 When ℎ(𝑡) ≡ 1, 𝑘(𝑥) ≡ 1, 𝑓(𝑥, 𝑢, 𝑞, 𝑡) = 𝑓(𝑢) or 𝑏(𝑢) = 𝑢,
(49)
ℎ(𝑡) ≡ 1, 𝑘(𝑥) ≡ 1, 𝑓(𝑥, 𝑢, 𝑞, 𝑡) = 𝑓(𝑢), or ℎ(𝑡) ≡ 1,
Combining (47)–(49) with condition (i), we can apply the 𝑘(𝑥) ≡ 1, 𝑎(𝑢) ≡ 1, 𝑓(𝑥, 𝑢, 𝑞, 𝑡) = 𝑓(𝑢), the conclusions
maximum principles again to obtain that the minimum of 𝑄 of Theorems 1 and 2 still hold true. In this sense, our results
extend and supplement the results in [14–16]. In what follows,
in 𝐷 × [0, 𝑇) is zero. Thus,
we present several examples to demonstrate the applications
of the abstract results.
𝑄 ≥ 0, in 𝐷 × [0, 𝑇) , (50)
Example 1. Let 𝑢 be a solution of the following problem:
𝑏󸀠 (𝑢)
𝑢 ≥ 𝛽. (51)
𝑒𝑢 𝑡 1 |𝑥|2
(𝑢𝑒𝑢 )𝑡 = ∇ ⋅ ( 𝑒 (1 + 𝑢) 𝑒𝑢 ∇𝑢)
1+𝑡
At the point 𝑥∗ ∈ 𝐷, where 𝑢0 (𝑥∗ ) = 𝑀0 , we can integrate
(51) from 0 to 𝑡 to get 1
+ (𝑒−𝑢 + 𝑒𝑞 ) (𝑒−𝑡 + |𝑥|2 ) , in 𝐷 × (0, 𝑇) ,
1+𝑡
∗
𝑡
𝑏󸀠 (𝑢) 𝑢(𝑥 ,𝑡)
𝑏󸀠 (𝑠) 𝜕𝑢
∫ 𝑢
𝑢𝑡 𝑑𝑡 = ∫ 𝑑𝑠 ≥ 𝛽𝑡, (52) + 2𝑢 = 0, on 𝜕𝐷 × (0, 𝑇) ,
0 𝑒 𝑀0 𝑒𝑠 𝜕𝑛

which implies that 𝑢 must blow up in finite time. Actually, if 𝑢 (𝑥, 0) = 2 − |𝑥|2 , in 𝐷,
𝑢 is a global solution of (1), then, for any 𝑡 > 0, (52) shows (59)

+∞
𝑏󸀠 (𝑠) 𝑢(𝑥 ∗
,𝑡)
𝑏󸀠 (𝑠) where 𝑞 = |∇𝑢|2 , 𝐷 = {𝑥 = (𝑥1 , 𝑥2 , 𝑥3 ) | |𝑥|2 < 1} is the unit
∫ 𝑑𝑠 ≥ ∫ 𝑑𝑠 ≥ 𝛽𝑡. (53) ball of R3 . Now we have
𝑀0 𝑒𝑠 𝑀0 𝑒𝑠
1
Letting 𝑡 → +∞ in (53), we have 𝑏 (𝑢) = 𝑢𝑒𝑢 , ℎ (𝑡) = ,
1+𝑡
2
+∞
𝑏󸀠 (𝑠) 𝑘 (𝑥) = 𝑒|𝑥| , 𝑎 (𝑢) = (1 + 𝑢) 𝑒𝑢 , 𝛾 = 2,
∫ 𝑑𝑠 = +∞, (54) (60)
𝑀0 𝑒𝑠
1
𝑓 (𝑥, 𝑢, 𝑞, 𝑡) = (𝑒−𝑢 + 𝑒𝑞 ) (𝑒−𝑡 + |𝑥|2 ) ,
which contradicts with assumption (40). This shows that 𝑢 1+𝑡
must blow up in finite time 𝑡 = 𝑇. Moreover, letting 𝑡 → 𝑇 𝑢0 (𝑥) = 2 − |𝑥|2 .
in (52), we have
In order to determine the constant 𝛼, we assume
1 +∞ 𝑏󸀠 (𝑠)
𝑇≤ ∫ 𝑑𝑠. (55)
𝛽 𝑀0 𝑒𝑠 𝑠 = |𝑥|2 , (61)
8 Abstract and Applied Analysis

and then 0 ≤ 𝑠 ≤ 1 and and then 0 ≤ 𝑠 ≤ 1 and

∇ ⋅ (ℎ (0) 𝑘 (𝑥) 𝑎 (𝑢0 ) ∇𝑢0 ) + 𝑓 (𝑥, 𝑢0 , 𝑞0 , 0)
∇ ⋅ (ℎ (0) 𝑘 (𝑥) 𝑎 (𝑢0 ) ∇𝑢0 ) + 𝑓 (𝑥, 𝑢0 , 𝑞0 , 0) 𝛽 = min
𝛼 = max 𝐷 𝑒𝑢0
𝐷 𝑒𝑢0
2
{ 4|𝑥|6 − 26|𝑥|4 + 50|𝑥|2 − 36
= max {𝑒|𝑥| (10|𝑥|2 − 18) + (1 + |𝑥|2 ) = min { 2
𝐷 𝐷
{ (2 − |𝑥|2 ) 𝑒2
2 2 (62)
× (𝑒−4+2|𝑥| + 𝑒−2+5|𝑥| )} 2 }
+6 (1 − 𝑒−3|𝑥| −2 ) } (67)
𝑠
= max {𝑒 (10𝑠 − 18) + (1 + 𝑠) }
0≤𝑠≤1
4𝑠3 − 26𝑠2 + 50𝑠 − 36
× [𝑒−4+2𝑠 + 𝑒−2+5𝑠 ]} = 18.6955. = min {
0≤𝑠≤1 (2 − 𝑠)2 𝑒2

It is easy to check that (5)–(7) hold. By Theorem 1, 𝑢 must be +6 (1 − 𝑒−3𝑠−2 ) } = 3.96997.

a global solution, and
It is easy to check that (38)–(40) hold. By Theorem 2, 𝑢 must
𝑢 (𝑥, 𝑡) ≤ 𝐻−1 (𝛼𝑡 + 𝐻 (𝑢0 (𝑥))) blow up in finite time 𝑇, and

2 1 +∞ 𝑏󸀠 (𝑠)
= −1 + √18.6955𝑡 + (1 + 𝑢0 (𝑥))
2
𝑇≤ ∫ 𝑑𝑠
(63) 𝛽 𝑀0 𝑒𝑠
+∞
= −1 + √ 18.6955𝑡 + (3 − |𝑥|2 ) .
2 2 1 1 1 (68)
= ∫ (1 + ) 𝑠 𝑑𝑠 = 0.04641,
3.96997 2 𝑠 𝑒
𝑢 (𝑥, 𝑡) ≤ 𝐺−1 (𝛽 (𝑇 − 𝑡)) = 𝐺−1 (3.96997 (𝑇 − 𝑡)) ,
Example 2. Let 𝑢 be a solution of the following problem:
where
2 1 +∞
𝑏󸀠 (𝑠) +∞
1 1
(𝑢 + ln 𝑢)𝑡 = ∇ ⋅ ((1 + 𝑡) 𝑒−|𝑥| (1 + ) ∇𝑢) 𝐺 (𝑧) = ∫ 𝑑𝑠 = ∫ (1 + ) 𝑠 𝑑𝑠, 𝑧 ≥ 0, (69)
𝑢 𝑧 𝑒𝑠
𝑧 𝑠 𝑒
+ (1 + 𝑡) (𝑒𝑢 − 𝑒−𝑞 ) (6 + 𝑡|𝑥|2 ) , and 𝐺−1 is the inverse function of 𝐺.
in 𝐷 × (0, 𝑇) , (64)
Conflict of Interests
𝜕𝑢
+ 2𝑢 = 0, on 𝜕𝐷 × (0, 𝑇) , The authors declare that there is no conflict of interests
𝜕𝑛
regarding the publication of this paper.
𝑢 (𝑥, 0) = 2 − |𝑥|2 , in 𝐷,
Acknowledgments
2 2
where 𝑞 = |∇𝑢| , 𝐷 = {𝑥 = (𝑥1 , 𝑥2 , 𝑥3 ) | |𝑥| < 1} is the unit The authors especially thank the reviewers for their very
ball of R3 . Now we have useful suggestions. The research was partially supported
by the National Natural Science Foundation of China (no.
𝑏 (𝑢) = 𝑢 + ln 𝑢, ℎ (𝑡) = 1 + 𝑡, 61250011) and the Science Foundation of Shanxi Province
(no. 2012011004-4) and Postdoctoral Science Foundation (no.
2 1 2012M510786).
𝑘 (𝑥) = 𝑒−|𝑥| , 𝑎 (𝑢) = 1 + , 𝛾 = 2,
𝑢 (65)
𝑓 (𝑥, 𝑢, 𝑞, 𝑡) = (1 + 𝑡) (𝑒𝑢 − 𝑒−𝑞 ) (6 + 𝑡|𝑥|2 ) , References
[1] A. Constantin and J. Escher, “Well-posedness, global existence,
𝑢0 (𝑥) = 2 − |𝑥|2 . and blowup phenomena for a periodic quasi-linear hyperbolic
equation,” Communications on Pure and Applied Mathematics,
vol. 51, no. 5, pp. 475–504, 1998.
In order to determine the constant 𝛽, we assume [2] J. M. Chadam, A. Peirce, and H. Yin, “The blowup property of
solutions to some diffusion equations with localized nonlinear
reactions,” Journal of Mathematical Analysis and Applications,
𝑠 = |𝑥|2 , (66) vol. 169, no. 2, pp. 313–328, 1992.
Abstract and Applied Analysis 9

[3] J. Wang, Z. J. Wang, and J. X. Yin, “A class of degenerate

diffusion equations with mixed boundary conditions,” Journal
of Mathematical Analysis and Applications, vol. 298, no. 2, pp.
589–603, 2004.
[4] N. Wolanski, “Global behavior of positive solutions to nonlinear
diffusion problems with nonlinear absorption through the
boundary,” SIAM Journal on Mathematical Analysis, vol. 24, no.
2, pp. 317–326, 1993.
[5] J. Ding, “Blow-up of solutions for a class of semilinear reaction
diffusion equations with mixed boundary conditions,” Applied
Mathematics Letters, vol. 15, no. 2, pp. 159–162, 2002.
[6] L. Zhang, “Blow-up of solutions for a class of nonlinear
parabolic equations,” Zeitschrift für Analysis und ihre Anwen-
dungen, vol. 25, no. 4, pp. 479–486, 2006.
[7] L. A. Caffarelli and A. Friedman, “Blowup of solutions of
nonlinear heat equations,” Journal of Mathematical Analysis and
Applications, vol. 129, no. 2, pp. 409–419, 1988.
[8] V. A. Galaktionov and J. L. Vázquez, “The problem of blow-
up in nonlinear parabolic equations,” Discrete and Continuous
Dynamical Systems, vol. 8, no. 2, pp. 399–433, 2002.
[9] H. Zhang and X. Guo, “Blow-up for nonlinear heat equations
with absorptions,” European Journal of Pure and Applied Math-
ematics, vol. 1, no. 3, pp. 33–39, 2008.
[10] A. A. Samarskii, V. A. Galaktionov, S. P. Kurdyumov, and A.
P. Mikhailov, Blow-Up in Problems for Quasilinear Parabolic
Equations, Nauka, Moscow, Russia, 1987, (Russian), English
Translation: Walter de Gruyter, Berlin, Germany, 1995.
[11] L. E. Payne and P. W. Schaefer, “Lower bounds for blow-up time
in parabolic problems under Neumann conditions,” Applicable
Analysis, vol. 85, no. 10, pp. 1301–1311, 2006.
[12] H. Amann, “Quasilinear parabolic systems under nonlinear
boundary conditions,” Archive for Rational Mechanics and
Analysis, vol. 92, no. 2, pp. 153–192, 1986.
[13] J. Ding and S. Li, “Blow-up solutions for a class of nonlinear
parabolic equations with mixed boundary conditions,” Journal
of Systems Science and Complexity, vol. 18, no. 2, pp. 265–276,
2005.
[14] J. Ding, “Global and blow-up solutions for nonlinear parabolic
equations with Robin boundary conditions,” Computers &
Mathematics with Applications, vol. 65, no. 11, pp. 1808–1822,
2013.
[15] C. Enache, “Blow-up phenomena for a class of quasilinear
parabolic problems under Robin boundary condition,” Applied
Mathematics Letters, vol. 24, no. 3, pp. 288–292, 2011.
[16] H. Zhang, “Blow-up solutions and global solutions for nonlin-
ear parabolic problems,” Nonlinear Analysis: Theory, Methods &
Applications, vol. 69, no. 12, pp. 4567–4574, 2008.
[17] R. P. Sperb, Maximum Principles and Their Applications, vol.
157 of Mathematics in Science and Engineering, Academic Press,
New York, NY, USA, 1981.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 958140, 18 pages
http://dx.doi.org/10.1155/2014/958140

Research Article
Bifurcation Analysis of a Lotka-Volterra Mutualistic
System with Multiple Delays

Xin-You Meng and Hai-Feng Huo

School of Science, Lanzhou University of Technology, Lanzhou, Gansu 730050, China

Correspondence should be addressed to Xin-You Meng; xymeng@lut.cn

Received 18 April 2014; Revised 15 June 2014; Accepted 16 June 2014; Published 14 August 2014

Academic Editor: Yongli Song

Copyright © 2014 X.-Y. Meng and H.-F. Huo. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.

A class of Lotka-Volterra mutualistic system with time delays of benefit and feedback delays is introduced. By analyzing the
associated characteristic equation, the local stability of the positive equilibrium and existence of Hopf bifurcation are obtained
under all possible combinations of two or three delays selecting from multiple delays. Not only explicit formulas to determine
the properties of the Hopf bifurcation are shown by using the normal form method and center manifold theorem, but also the
global continuation of Hopf bifurcation is investigated by applying a global Hopf bifurcation result due to Wu (1998). Numerical
simulations are given to support the theoretical results.

1. Introduction 𝑦̇(𝑡) = 𝑦 (𝑡) [𝑟2 − 𝑎21 𝑥 (𝑡 − 𝜏21 ) − 𝑎22 𝑦 (𝑡 − 𝜏22 )] ,

(1)
In recent years, population models have merited a great deal
of attention due to their theoretical and practical significance where 𝜏11 , 𝜏22 > 0 are maturation times and 𝜏12 , 𝜏21 > 0
since the pioneering theoretical works by Lotka [1] and are called the hunting delays for the prey and the predator
Volterra [2]; see [3–6] and references therein. Generally species, respectively. The result was that time delay in the
speaking, there are three kinds of fundamental forms of the competitive Lotka-Volterra system does not harm the bound-
interactions between two species such as competition, coop- edness and persistence. Song et al. [4] analyzed the stability
eration, and prey-predation in population biology. Among of the interior positive equilibrium and the existence of local
these interactions, the competition mechanism has been Hopf bifurcation for system (1) with 𝜏11 = 𝜏22 = 0 by
paying extreme attention because it possesses very significant taking the delay 𝜏 = 𝜏12 + 𝜏21 as the bifurcation parameter.
function as a kind of restriction factor in the process of Their results showed that changes of hunting delays for
evolvement of biology. system (1) do not lead to the occurrence of Hopf bifurcation
It is well known that time delays of one type or another when interspecies competition is weaker than intraspecies
have been incorporated into mathematical models of popu- competition. Zhang et al. [5] studied the stability and Hopf
lation dynamics due to maturation time, capturing time, or bifurcation of system (1) with 𝜏11 = 𝜏12 = 𝜏21 = 𝜏22 = 𝜏.
other reasons. The effect of the past history on the stability Zhang [6] also investigated the dynamics of system (1) only
of the system is an important problem in population biology. when 𝜏11 = 𝜏22 = 𝜏 and 𝜏12 = 𝜏21 = 0.
In general, delay differential equations exhibit much more In fact, predator-prey system with time delays has also
complicated dynamics than ordinary differential equations been investigated by lots of authors [7–11]. Faria [7] consid-
since a time delay could cause a stable equilibrium to become ered the delayed predator-prey system of the form
unstable and cause the population to fluctuate. In 2002, Jin
and Ma [3] investigated the competition model with four 𝑥̇(𝑡) = 𝑥 (𝑡) [𝑟1 − 𝑎11 𝑥 (𝑡) − 𝑎12 𝑦 (𝑡 − 𝜏1 )] ,
delays: (2)
𝑥̇(𝑡) = 𝑥 (𝑡) [𝑟1 − 𝑎11 𝑥 (𝑡 − 𝜏11 ) − 𝑎12 𝑦 (𝑡 − 𝜏12 )] , 𝑦̇(𝑡) = 𝑦 (𝑡) [−𝑟2 + 𝑎21 𝑥 (𝑡 − 𝜏2 ) − 𝑎22 𝑦 (𝑡)] ,
2 Abstract and Applied Analysis

where 𝑥(𝑡) and 𝑦(𝑡) can be interpreted as the popula- In general, the delays appearing in different terms of
tion densities of the prey and the predator at the time 𝑡, an ecological system are not equal. Therefore, it is more
respectively, and 𝜏1 and 𝜏2 denote the hunting delay and realistic to consider dynamical system with different delays.
the time of predator maturation. In [7], taking the single Motivated by the references [6, 9, 17], we consider the
delay, 𝜏2 , as a parameter and assuming the ratio 𝜏1 /𝜏2 to be following two-species Lotka-Volterra mutualistic system with
constant, the author studied the properties of the local Hopf multiple delays:
bifurcation. Song et al. [8] obtained the properties of the local
Hopf bifurcation as well as the global existence of periodic 𝑥̇(𝑡) = 𝑥 (𝑡) [𝑏1 − 𝑎11 𝑥 (𝑡 − 𝜏1 ) + 𝑎12 𝑦 (𝑡 − 𝜏2 )] ,
(6)
solutions by choosing the sum 𝜏 = 𝜏1 + 𝜏2 as a bifurcation 𝑦̇(𝑡) = 𝑦 (𝑡) [𝑏2 + 𝑎21 𝑥 (𝑡 − 𝜏3 ) − 𝑎22 𝑦 (𝑡 − 𝜏4 )] ,
parameter. Yan and Li [9] considered the following delayed
predator-prey system: where 𝑥(𝑡) and 𝑦(𝑡) are the densities of two species at the time
𝑡 and 𝑏1 and 𝑏2 are the intrinsic growth rates of the two species.
𝑥̇(𝑡) = 𝑥 (𝑡) [𝑟1 − 𝑎11 𝑥 (𝑡 − 𝜏) − 𝑎12 𝑦 (𝑡)] , 𝜏1 is the feedback delay of one species to its grow and 𝜏4 is
(3) the feedback delay of the other species; 𝜏2 and 𝜏3 are the time
𝑦̇(𝑡) = 𝑦 (𝑡) [−𝑟2 + 𝑎21 𝑥 (𝑡) − 𝑎22 𝑦 (𝑡 − 𝜏)] , delays of the benefit. The coefficients 𝑏𝑖 , 𝑎𝑖𝑗 (𝑖, 𝑗 = 1, 2) are all
positive constants and 𝜏𝑖 (𝑖 = 1, 2, 3, 4) ≥ 0. In addition, we
where 𝜏 denotes the feedback time delay of prey species to would like to mention the work of F. Q. Zhang and Y. J. Zhang
the growth of species itself, and also for the predator. They [19], who considered the system (6) only with 𝜏1 = 𝜏4 = 𝜏 and
found that the unique positive equilibrium of system (3) will 𝜏2 +𝜏3 = 2𝜏. In fact, feedback delay on different species should
no longer be absolutely stable and the switches from stability not be same, and the benefit period should not be equal to
to instability to stability disappear as the feedback time delay the feedback time delay. Thus, how does every delay or the
increases monotonously from zero, which had been obtained combined two or three delays of the above four delays have
for system (3) by Song and Wei [12]. an effect on the dynamics of system (6)? The purpose of this
However, the research for cooperative systems with time paper is to answer this question partially.
delays is still relatively little because delays in mutualistic sys- This paper is organized as follows. In Section 2, by analyz-
tems usually deprive the boundedness and persistence in [13]. ing the characteristic equation of linearized system of system
But time delays in prey-predator and competitive systems do (6) at the positive equilibrium, some sufficient conditions
not harm these properties [3, 14, 15]. Two species cohabit ensuring the local stability of the positive equilibrium and
a common habitat and each species enhances the average the existence of Hopf bifurcation are obtained. Some explicit
growth rate of the other; this type of ecological interaction formulas determining the direction and stability of periodic
is known as facultative mutualism. Mutual phenomenon can solutions bifurcating from Hopf bifurcations are given by
increase viability and make species persistently multiply. As applying the normal form method and center manifold
far as we know, the research on mutual system is less than theory due to Hassard et al. [20] in Section 3. In Section 4,
prey-predator system and competitive system. At present, the we show the global existence of the periodic solutions due
known results of mutual system mainly focus on stability and to a global Hopf bifurcation result of Wu [21] for functional
persistence [16, 17]. Research on the ecologic system stability differential equations. To support our theoretical predictions,
of positive equilibrium and existence of periodic solutions some numerical simulations are included in Section 5. A brief
is very crucial, which can help us to realize the law for discussion is also given in the last section.
species quantity and predict the trend for species quantity. In
1997, He and Gopalsamy [17] considered the Lotka-Volterra
mutualistic system with delay 2. Local Stability and Hopf Bifurcation

𝑥̇(𝑡) = 𝑥 (𝑡) [𝑏1 − 𝑎11 𝑥 (𝑡 − 𝜏) + 𝑎12 𝑦 (𝑡 − 𝜏)] , For convenience, letting 𝑥 = (𝑎11 /𝑏1 )𝑥, 𝑦 = (𝑎22 /𝑏2 )𝑦 and
(4) dropping the bars for simplification of notation, system (6) is
𝑦̇(𝑡) = 𝑦 (𝑡) [𝑏2 + 𝑎21 𝑥 (𝑡 − 𝜏) − 𝑎22 𝑦 (𝑡 − 𝜏)] , transformed into
𝑥̇(𝑡) = 𝑟1 𝑥 (𝑡) [1 − 𝑥 (𝑡 − 𝜏1 ) + 𝑎𝑦 (𝑡 − 𝜏2 )] ,
and obtained the stability of the positive equilibrium and the (7)
existence of Hopf bifurcation when 𝑎11 𝑎22 > 𝑎12 𝑎21 . Meng 𝑦̇(𝑡) = 𝑟2 𝑦 (𝑡) [1 + 𝑏𝑥 (𝑡 − 𝜏3 ) − 𝑦 (𝑡 − 𝜏4 )] ,
and Wei [18] studied the following system with delay:
where 𝑟1 = 𝑏1 , 𝑟2 = 𝑏2 , 𝑎 = 𝑎12 𝑏2 /𝑎22 𝑏1 , and 𝑏 = 𝑎21 𝑏1 /𝑎11 𝑏2 .
𝑥 (𝑡 − 𝜏) From the point view of biological meaning, we are inter-
𝑥̇(𝑡) = 𝑟1 𝑥 (𝑡) [1 − ] + 𝑎𝑥 (𝑡) 𝑦 (𝑡) , ested in the positive equilibrium. It is obvious that system (7)
𝑘1
(5) has a unique positive equilibrium 𝐸∗ (𝑥∗ , 𝑦∗ ) defined by
𝑦 (𝑡 − 𝜏)
𝑦̇(𝑡) = 𝑟2 𝑦 (𝑡) [1 − ] + 𝑏𝑥 (𝑡) 𝑦 (𝑡) , 1+𝑎 1+𝑏
𝑘2 𝑥∗ = , 𝑦∗ = , (8)
1 − 𝑎𝑏 1 − 𝑎𝑏
where 𝑎 and 𝑏 are the rates of transmission between two provided that the following condition
species. They found that there are stability switches and Hopf
(H1) 0 < 𝑎𝑏 < 1
bifurcation occurring when the delay 𝜏 passes through a
sequence of critical values. is satisfied.
Abstract and Applied Analysis 3

Next, we will consider the stability of the positive equilib- Let 𝜔𝑖 (𝜔 > 0) be the root of (13). Substituting it into (13)
rium and the existence of Hopf bifurcation. and separating the real and imaginary parts, we have
Letting 𝑢1 (𝑡) = 𝑥(𝑡) − 𝑥∗ , 𝑢2 (𝑡) = 𝑦(𝑡) − 𝑦∗ , system (7)
can be transformed into −𝛼2 𝛼3 cos 𝜔𝜏2 = 𝜔2 − 𝛼1 𝛼4 ,
∗ (15)
𝑢̇
1 (𝑡) = 𝑟1 (𝑢1 (𝑡) + 𝑥 ) [−𝑢1 (𝑡 − 𝜏1 ) + 𝑎𝑢2 (𝑡 − 𝜏2 )] , 𝛼2 𝛼3 sin 𝜔𝜏2 = (𝛼1 + 𝛼4 ) 𝜔.
(9)
∗
𝑢̇
2 (𝑡) = 𝑟2 (𝑢2 (𝑡) + 𝑦 ) [𝑏𝑢1 (𝑡 − 𝜏3 ) − 𝑢2 (𝑡 − 𝜏4 )] .
It follows that
Linearizing system (9) at (0, 0) and rewriting 𝑢1 , 𝑢2 as 𝑥, 𝑦, 2 2
system (9) is rewritten as 𝜔4 + (𝛼12 + 𝛼42 ) 𝜔2 + (𝛼1 𝛼4 ) − (𝛼2 𝛼3 ) = 0. (16)

𝑥̇(𝑡) = 𝛼1 𝑥 (𝑡 − 𝜏1 ) + 𝛼2 𝑦 (𝑡 − 𝜏2 ) , Equation (16) does not have positive root since 𝛼12 +𝛼42 > 0 and
(10)
𝑦̇(𝑡) = 𝛼3 𝑥 (𝑡 − 𝜏3 ) + 𝛼4 𝑦 (𝑡 − 𝜏4 ) , (𝛼1 𝛼4 )2 − (𝛼2 𝛼3 )2 = (𝑟1 𝑟2 𝑥∗ 𝑦∗ )2 (1 + 𝑎𝑏)(1 − 𝑎𝑏) > 0. Thus,
Hopf bifurcation does not occur at the positive equilibrium
where 𝛼1 = −𝑟1 𝑥∗ , 𝛼2 = 𝑎𝑟1 𝑥∗ , 𝛼3 = 𝑏𝑟2 𝑦∗ , and 𝛼4 = −𝑟2 𝑦∗ . 𝐸∗ of system (7).
The associated characteristic equation of system (10) with
𝜏1 = 𝜏2 = 𝜏3 = 𝜏4 = 0 is Case 1B. Consider 𝜏2 = 0, 𝜏3 ≠0. Equation (12) is written in
the form
𝜆2 − (𝛼1 + 𝛼4 ) 𝜆 + 𝛼1 𝛼4 − 𝛼2 𝛼3 = 0. (11)
𝜆2 − (𝛼1 + 𝛼4 ) 𝜆 + 𝛼1 𝛼4 − 𝛼2 𝛼3 𝑒−𝜆𝜏3 = 0. (17)
It is obvious that 𝜆 = 0 is not the root of (11). All roots of (11)
have negative real parts since −(𝛼1 + 𝛼4 ) = 𝑟1 𝑥∗ + 𝑟2 𝑦∗ > 0 Equation (17) is similar to (13). Thus, Hopf bifurcation also
and 𝛼1 𝛼4 − 𝛼2 𝛼3 = 𝑟1 𝑟2 𝑥∗ 𝑦∗ (1 − 𝑎𝑏) > 0 if (H1) holds. So, the does not occur at the positive equilibrium 𝐸∗ of system (7).
equilibrium point 𝐸∗ (𝑥∗ , 𝑦∗ ) is locally asymptotically stable We omit the corresponding proof.
when (H1) holds.
Case 1C. Consider 𝜏2 = 𝜏3 = 𝜏. Equation (12) is written in the
2.1. Only Considering 𝜏2 and 𝜏3 . The characteristic equation form
of system (10) with 𝜏1 = 𝜏4 = 0 is
𝜆2 − (𝛼1 + 𝛼4 ) 𝜆 + 𝛼1 𝛼4 − 𝛼2 𝛼3 𝑒−2𝜆𝜏 = 0. (18)
2 −𝜆(𝜏2 +𝜏3 )
𝜆 − (𝛼1 + 𝛼4 ) 𝜆 + 𝛼1 𝛼4 − 𝛼2 𝛼3 𝑒 = 0. (12)
Equation (18) is also similar to (16). Thus, Hopf bifurcation
In the following, we will discuss the distribution of roots of does not occur at the positive equilibrium 𝐸∗ of system (7).
(12) while 𝜏2 and 𝜏3 are given different values.
When 𝜏2 = 0 and 𝜏3 = 0, (12) becomes (11). Thus, Case 1D. Consider 𝜏2 ∈ (0, +∞) and 𝜏3 ≠0. Let 𝜔𝑖 (𝜔 > 0)
the positive equilibrium 𝐸∗ (𝑥∗ , 𝑦∗ ) is locally asymptotically be a root of (12). We can obtain the following form by giving
stable when (H1) holds. the value 𝜏2 = 𝜏2∗ ∈ (0, +∞) and regarding 𝜏3 as a parameter:
Case 1A. Consider 𝜏2 ≠0, 𝜏3 = 0. Equation (12) can be written 𝛼2 𝛼3 cos 𝜔𝜏2∗ sin 𝜔𝜏3 + 𝛼2 𝛼3 sin 𝜔𝜏2∗ cos 𝜔𝜏3
in the form
= (𝛼1 + 𝛼4 ) 𝜔,
𝜆2 − (𝛼1 + 𝛼4 ) 𝜆 + 𝛼1 𝛼4 − 𝛼2 𝛼3 𝑒−𝜆𝜏2 = 0. (13)
(19)
− 𝛼2 𝛼3 cos 𝜔𝜏2∗ cos 𝜔𝜏3 + 𝛼2 𝛼3 sin 𝜔𝜏2∗ sin 𝜔𝜏3
We first introduce the following result which was proved
by Ruan and Wei [22] using Rouch’s theorem.
= 𝜔2 − 𝛼1 𝛼4 .
Lemma 1. Consider the exponential polynomial
It follows that
−𝜆𝜏1 −𝜆𝜏𝑚
𝑃 (𝜆, 𝑒 ,...,𝑒 ) 2 2
𝜔4 + (𝛼12 + 𝛼42 ) 𝜔2 + (𝛼1 𝛼4 ) − (𝛼2 𝛼3 ) = 0. (20)
𝑛
=𝜆 + 𝑝10 𝜆𝑛−1 + ⋅⋅⋅ + 0
𝑝𝑛−1 𝜆 + 𝑝𝑛0
(14) It is obvious that Hopf bifurcation does not occur at the
+ [𝑝11 𝜆𝑛−1 + ⋅ ⋅ ⋅ + 𝑝𝑛−1
1
𝜆 + 𝑝𝑛1 ] 𝑒−𝜆𝜏1 + ⋅ ⋅ ⋅ positive equilibrium 𝐸∗ of system (7) according to (16).

+ [𝑝1𝑚 𝜆𝑛−1 + ⋅ ⋅ ⋅ + 𝑝𝑛−1

𝑚
𝜆 + 𝑝𝑛𝑚 ] 𝑒−𝜆𝜏𝑚 , 2.2. Only Considering 𝜏1 and 𝜏4 . The characteristic equation
of system (10) with 𝜏2 = 𝜏3 = 0 is
where 𝜏𝑖 ≥ 0 (𝑖 = 1, 2, . . . , 𝑚), 𝑝𝑗𝑖 (𝑖 = 1, 2, . . . , 𝑚; 𝑗 =
1, 2, . . . , 𝑛) are constants. As (𝜏1 , 𝜏2 , . . . , 𝜏𝑚 ) vary, the sum of 𝜆2 − 𝛼2 𝛼3 − 𝛼1 𝜆𝑒−𝜆𝜏1 − 𝛼4 𝜆𝑒−𝜆𝜏4 + 𝛼1 𝛼4 𝑒−𝜆(𝜏1 +𝜏4 ) = 0. (21)
the order of the zeros of 𝑃(𝜆, 𝑒−𝜆𝜏1 , . . . , 𝑒−𝜆𝜏𝑚 ) on the open right
half-plane can change only if a zero appears on or crosses the In the following, we will discuss the distribution of roots of
imaginary axis. (21) while 𝜏1 and 𝜏4 are given different values.
4 Abstract and Applied Analysis

When 𝜏1 = 0, 𝜏4 = 0, (21) becomes (11) and the For simplifying, define 𝜔+ as 𝜔 and 𝜏𝑗 as 𝜏1 , and we can
equilibrium 𝐸∗ (𝑥∗ , 𝑦∗ ) is locally asymptotically stable when obtain
(H1) holds.
d (Re 𝜆)
sign { | }
Case 2A. Consider 𝜏1 ≠0, 𝜏4 = 0. Equation (21) is written as d𝜏 𝜏=𝜏10
follows:
d𝜆 −1
= sign {Re ( ) |𝜏=𝜏10 }
d𝜏
𝜆2 − 𝛼4 𝜆 − 𝛼2 𝛼3 + (−𝛼1 𝜆 + 𝛼1 𝛼4 ) 𝑒−𝜆𝜏1 = 0. (22)
2𝜆 − 𝛼4 − 𝛼1 𝑒−𝜆𝜏 𝜏
Let 𝜔𝑖 (𝜔 > 0) be the root of (22), and we have that = sign {Re [ −𝜆𝜏
− 1] }
𝜆 (−𝛼1 𝜆 + 𝛼1 𝛼4 ) 𝑒 𝜆 𝜏=𝜏
10

− 𝛼1 𝜔 cos 𝜔𝜏1 − 𝛼1 𝛼4 sin 𝜔𝜏1 = 𝛼4 𝜔, 2

(23) { 𝜔4 + 2𝛼42 𝜔2 + (𝛼42 − 𝛼2 𝛼3 ) }
= sign { }.
𝛼1 𝜔 sin 𝜔𝜏1 − 𝛼1 𝛼4 cos 𝜔𝜏1 = −𝜔2 − 𝛼2 𝛼3 . 2
[𝜔3 + (𝛼2 𝛼3 − 𝛼42 ) 𝜔] + (2𝛼4 𝜔2 + 𝛼2 𝛼3 𝛼4 )
2
{ }
(29)
It follows that
Thus, the transversality condition holds and Hopf bifurcation
2 2
𝜔4 + (𝛼42 − 𝛼12 + 2𝛼2 𝛼3 ) 𝜔2 + (𝛼2 𝛼3 ) − (𝛼1 𝛼4 ) = 0. (24) occurs at 𝜏 = 𝜏10 . We have the following theorem.

Theorem 3. For system (7), if (H1) holds, then there exists a

Since (𝛼2 𝛼3 )2 − (𝛼1 𝛼4 )2 = (𝑟1 𝑟2 𝑥∗ 𝑦∗ )2 (1 + 𝑎𝑏)(𝑎𝑏 − 1) < 0, positive number 𝜏10 such that the coexistence equilibrium 𝐸∗ is
(24) has one positive root, defined by 𝜔+ . locally asymptotically stable for 0 < 𝜏1 < 𝜏10 and unstable for
From (23), if we denote 𝜏1 > 𝜏10 . Further, system (7) undergoes a Hopf bifurcation at
the equilibrium 𝐸∗ for 𝜏1 = 𝜏10 .
1 𝛼𝛼𝛼𝛼
𝜏𝑗 = {arccos ( 2 1+ 22 3 42 2 ) + 2𝑗𝜋} ,
𝜔+ 𝛼1 (𝜔 ) + 𝛼1 𝛼4 (25) Case 2B. Consider 𝜏1 = 0, 𝜏4 ≠0. Equation (21) can be written
𝑗 = 0, 1, 2, . . . , in the form

𝜆2 − 𝛼1 𝜆 − 𝛼2 𝛼3 + (−𝛼4 𝜆 + 𝛼1 𝛼4 ) 𝑒−𝜆𝜏4 = 0. (30)

then ±𝜔+ 𝑖 are a pair of purely imaginary root of (22) with
𝜏1 = 𝜏𝑗 . Since (30) has the similar form of (22), the corresponding
Define 𝜏10 = min{𝜏𝑗 , 𝑗 = 0, 1, 2, . . .}. Let 𝜆(𝜏) = 𝜉(𝜏) + results are omitted.
𝜔(𝜏)𝑖 be the root of (22) near 𝜏 = 𝜏10 satisfying 𝜉(𝜏10 ) =
0, 𝜔(𝜏10 ) = 𝜔0 . We first check whether the transversality Case 2C. Consider 𝜏1 = 𝜏4 = 𝜏. Equation (21) can be written
condition is satisfied. in the form

Lemma 2. the following transversality condition is satisfied: 𝜆2 − 𝛼2 𝛼3 − (𝛼1 + 𝛼4 ) 𝜆𝑒−𝜆𝜏 + 𝛼1 𝛼4 𝑒−2𝜆𝜏 = 0. (31)

Multiplying the 𝑒𝜆𝜏 into the both sides of (31) and letting 𝜆 =
𝑑Re (𝜆)
| > 0. (26) 𝜔𝑖 (𝜔 > 0) to be the root of (31), we have that
𝑑𝜏 𝜆=𝜔0 𝑖
(𝛼1 𝛼4 − 𝛼2 𝛼3 − 𝜔2 ) cos 𝜔𝜏 = 0,
Proof. This will show that there exists at least one eigenvalue (32)
with positive real part for 𝜏 > 𝜏10 . Moreover, the conditions − (𝛼1 𝛼4 + 𝛼2 𝛼3 + 𝜔2 ) sin 𝜔𝜏 = (𝛼1 + 𝛼4 ) 𝜔.
for the existence of a Hopf bifurcation [15, 22] are satisfied to
yield a periodic solution. Differentiating (22) with respect to Suppose that 𝛼1 𝛼4 − 𝛼2 𝛼3 − 𝜔2 = 0. We have
𝜏1 , it follows that 𝜔+
= √𝑟1 𝑟2 𝑥∗ 𝑦∗ (1 − 𝑎𝑏) when (H1) holds. From the
second equation of (32), we can get sin 𝜔𝜏 = (𝑟1 𝑥∗ +
2𝜆
d𝜆
− 𝛼4
d𝜆
− 𝛼1 𝑒−𝜆𝜏1
d𝜆 𝑟2 𝑦∗ )√1 − 𝑎𝑏/2√𝑟1 𝑟2 𝑥∗ 𝑦∗ > 0. Thus, we can get 𝜏(𝑗) =
d𝜏1 d𝜏1 d𝜏1 (1/𝜔+ ){arc sin((𝑟1 𝑥∗ + 𝑟2 𝑦∗ )√1 − 𝑎𝑏/(2√𝑟1 𝑟2 𝑥∗ 𝑦∗ )) + 2𝑗𝜋},
(27)
d𝜆 𝑗 = 0, 1, 2, . . .. Define 𝜏01 = 𝜏𝑗(0) = min{𝜏(𝑗) , 𝑗 = 0, 1, 2, . . .},
−𝜆𝜏1 0
+ (−𝛼1 𝜆 + 𝛼1 𝛼4 ) 𝑒 (−𝜆 − 𝜏1 ) = 0, 𝜔0 = 𝜔𝑗0 . Let 𝜆(𝜏) = 𝜉(𝜏) + 𝜔(𝜏)𝑖 be the root of (31) near
d𝜏1
𝜏 = 𝜏01 , satisfying 𝜉(𝜏01 ) = 0, 𝜔(𝜏01 ) = 𝜔0 . Further, (31) has a
which implies that pair of purely imaginary roots ±𝜔+ 𝑖.
If 𝛼1 𝛼4 − 𝛼2 𝛼3 − 𝜔2 ≠0, then cos 𝜔𝜏 = 0. If sin 𝜔𝜏 = 1,
then the second equation of (32) becomes
d𝜆 −1 2𝜆 − 𝛼4 − 𝛼1 𝑒−𝜆𝜏1 𝜏
( ) = − 1. (28)
d𝜏1 𝜆 (−𝛼1 𝜆 + 𝛼1 𝛼4 ) 𝑒−𝜆𝜏1 𝜆 𝜔2 + (𝛼1 + 𝛼4 ) 𝜔 + 𝛼1 𝛼4 + 𝛼2 𝛼3 = 0. (33)
Abstract and Applied Analysis 5

If (𝑟2 + 𝑏𝑟2 − 𝑟1 − 𝑎𝑟1 )2 = 4𝑎𝑏𝑟1 𝑟2 (1 + 𝑎)(1 + 𝑏), (33) has 1

= sign { Re [ (𝛼1 𝛼4 cos 𝜔𝜏 − 2𝜔 sin 𝜔𝜏
one positive root 𝜔+ = (−1/2)(𝛼1 + 𝛼4 ). Thus, we can get 𝜔
𝜏(𝑗) = (1/𝜔+ )(𝜋/2 + 2𝜋𝑗), 𝑗 = 0, 1, 2, . . .. Define 𝜏02 = 𝜏𝑗(0)
0
= + (−𝛼1 𝛼4 sin 𝜔𝜏 + 2𝜔 cos 𝜔𝜏) 𝑖)
min{𝜏(𝑗) , 𝑗 = 0, 1, 2, . . .}, 𝜔0 = 𝜔𝑗0 . Let 𝜆(𝜏) = 𝜉(𝜏) + 𝜔(𝜏)𝑖
be the root of (31) near 𝜏 = 𝜏02 , satisfying 𝜉(𝜏02 ) = 0, × ((𝜔2 + 𝛼2 𝛼3 ) cos 𝜔𝜏 + 𝛼1 𝛼4 𝜔 sin 𝜔𝜏
𝜔(𝜏02 ) = 𝜔0 . Further, (31) has a pair of purely imaginary
roots ± 𝜔+ 𝑖. If (𝑟2 + 𝑏𝑟2 − 𝑟1 − 𝑎𝑟1 )2 > 4𝑎𝑏𝑟1 𝑟2 (1 + 𝑎)(1 + 𝑏), + [ (𝜔2 + 𝛼2 𝛼3 ) sin 𝜔𝜏 (36)
and (33) has two positive roots 𝜔± = (1/2)[−(𝛼1 + 𝛼4 ) ±
−1
√(𝛼1 + 𝛼4 )2 − 4(𝛼1 𝛼4 + 𝛼2 𝛼3 )]. It is obvious that 𝜔+ > 𝜔− . + 𝛼1 𝛼4 𝜔 cos 𝜔𝜏] 𝑖) ]}
From (32), we denote 𝜏𝑘± = (1/𝜔± )(𝜋/2 + 2𝜋𝑘), 𝑘 = 0, 1, 2, . . .. 2
When 𝜏 = 𝜏𝑘+ (resp., 𝜏 = 𝜏𝑘− ), then ± 𝜔+ 𝑖 (resp., ± 𝜔− 𝑖) are a (𝛼1 − 𝛼4 )
= sign { 2
}.
pair of purely imaginary roots of (31). Further, we can check (𝜔2 + 𝛼2 𝛼3 + 𝛼1 𝛼4 )
+
that there exists a positive integral 𝑗 > 0 such that 𝜏𝑗+1 < 𝜏𝑗−
+ − + − − + + − Thus, the transversality condition is satisfied. This ends the
and 0 < 𝜏0 < 𝜏0 < 𝜏1 < 𝜏1 ⋅ ⋅ ⋅ 𝜏𝑗−1 < 𝜏𝑗 < 𝜏𝑗+1 < 𝜏𝑗 . Let proof.
𝜆(𝜏) = 𝜉(𝜏) + 𝜔(𝜏)𝑖 be the root of (31) near 𝜏 = 𝜏𝑘± , satisfying
𝜉(𝜏𝑘± ) = 0, 𝜔(𝜏𝑘± ) = 𝜔0 . If sin 𝜔𝜏 = −1, then the second Remark 6. When 𝜏 = 𝜏𝑘± , the transversality condition is
equation of (32) becomes 𝜔2 − (𝛼1 + 𝛼4 )𝜔 + 𝛼1 𝛼4 + 𝛼2 𝛼3 = 0, satisfied as follows:
which does not have positive root. d (Re 𝜆) d (Re 𝜆)
| + > 0, | − < 0. (37)
d𝜏 𝜏=𝜏𝑘 d𝜏 𝜏=𝜏𝑘
Lemma 4. Suppose that (H1) holds.
Theorem 7. For system (7), consider the following.
(i) If 𝜏 ∈ [0, 𝜏01 ), then all roots of (31) have strictly negative
(i) If (H1) (resp., (H1) and (𝑟2 + 𝑏𝑟2 − 𝑟1 − 𝑎𝑟1 )2 =
real parts; if 𝜏 = 𝜏01 , then all roots of (31), except for
4𝑎𝑏𝑟1 𝑟2 (1 + 𝑎)(1 + 𝑏)) holds, then there exists a
±𝜔+ 𝑖, have strictly negative real parts.
positive number 𝜏01 (resp., 𝜏02 ) such that the positive
(ii) If (𝑟2 + 𝑏𝑟2 − 𝑟1 − 𝑎𝑟1 )2 = 4𝑎𝑏𝑟1 𝑟2 (1 + 𝑎)(1 + 𝑏) and equilibrium 𝐸∗ is locally asymptotically stable for 0 <
𝜏 ∈ [0, 𝜏02 ), then all roots of (31) have strictly negative 𝜏 < 𝜏01 (resp., 0 < 𝜏 < 𝜏02 ) and unstable for
real parts; if 𝜏 = 𝜏02 , then all roots of (31), except for 𝜏 > 𝜏01 (resp., 𝜏 > 𝜏02 ). Further, system (7) undergoes
± 𝜔+ 𝑖, have strictly negative real parts. a Hopf bifurcation at the positive equilibrium 𝐸∗ for
(iii) If (𝑟2 + 𝑏𝑟2 − 𝑟1 − 𝑎𝑟1 )2 > 4𝑎𝑏𝑟1 𝑟2 (1 + 𝑎)(1 + 𝑏), 𝜏 = 𝜏(𝑗) (𝑗 = 0, 1, 2, . . .).
there exist two sequences {𝜏𝑘+ }(𝑘 = 0, 1, 2, . . .) and (ii) If (H1) and (𝑟2 + 𝑏𝑟2 − 𝑟1 − 𝑎𝑟1 )2 > 4𝑎𝑏𝑟1 𝑟2 (1 +
{𝜏𝑘− } (𝑘 = 0, 1, 2, . . .). Further, there exists 𝑁 ∈ N, 𝑎)(1 + 𝑏) hold, then there is a positive integer 𝑁, such
when 𝜏 ∈ (0, 𝜏0+ ) ∪ (𝜏0− , 𝜏1+ ) ∪ ⋅ ⋅ ⋅ (𝜏𝑁−1 − +
, 𝜏𝑁 ), and that the positive equilibrium 𝐸∗ switches 𝑁 times from
all the roots of (31) have negative real part; when 𝜏 ∈ stability to instability to stability; that is, the positive
(𝜏0+ , 𝜏0− )∪(𝜏1+ , 𝜏1− )∪⋅ ⋅ ⋅∪(𝜏𝑁
+ − +
, 𝜏𝑁)∪(𝜏𝑁+1 , +∞), (31) has equilibrium 𝐸∗ of system (7) is locally asymptotically
at least one root with positive real part. When 𝜏 = 𝜏𝑘+ stable when 𝜏 ∈ (0, 𝜏0+ ) ∪ (𝜏0− , 𝜏1+ ) ∪ ⋅ ⋅ ⋅ (𝜏𝑁−1
− +
, 𝜏𝑁 )
and 𝜏 = 𝜏𝑘− , (31) has a pair of pure imaginary roots. + − + −
and unstable when 𝜏 ∈ (𝜏0 , 𝜏0 ) ∪ (𝜏1 , 𝜏1 ) ∪ ⋅ ⋅ ⋅ ∪
+ − +
(𝜏𝑁 , 𝜏𝑁) ∪ (𝜏𝑁+1 , +∞). Hopf bifurcation occurs at the
Lemma 5. The following transversality condition is satisfied: positive equilibrium 𝐸∗ when 𝜏 = 𝜏𝑘+ and 𝜏 = 𝜏𝑘− .
𝑑 (Re 𝜆) 𝑑 (Re 𝜆)
| >0 (resp. | > 0) . (34) Case 2D. Consider 𝜏1 ∈ [0, 𝜏10 ) and 𝜏4 ≠0. Regarding 𝜏4
𝑑𝜏 𝜏=𝜏01 𝑑𝜏 𝜏=𝜏02
as a parameter, we consider (21) with 𝜏1 in its stable interval.
Proof. Multiplying the 𝑒𝜆𝜏 into the both sides of (31) and Without loss of generality, we consider system (7) under
differentiating (31) with respect to 𝜏, we can obtain that Case 2A. Let 𝜔𝑖 (𝜔 > 0) be a root of (21), and then we obtain,
by selecting delay 𝜏10 ∈ [0, 𝜏10 ),
d𝜆 −1 2𝜆𝑒𝜆𝜏 − (𝛼1 + 𝛼4 ) 𝜏
( ) = 2 𝜆𝜏 −𝜆𝜏
− . (35) 𝛼1 𝛼4 cos 𝜔𝜏4 − (𝛼4 𝜔 + 𝛼1 𝛼4 sin 𝜔𝜏10 ) sin 𝜔𝜏4
d𝜏 𝜆 [(−𝜆 + 𝛼2 𝛼3 ) 𝑒 + 𝛼1 𝛼4 𝑒 ] 𝜆
+ (𝑗) = 𝜔2 + 𝛼2 𝛼3 − 𝛼1 𝜔 sin 𝜔𝜏10 ,
For simplifying, define 𝜔 as 𝜔 and 𝜏 as 𝜏, and we can obtain (38)
− 𝛼1 𝛼4 sin 𝜔𝜏4 − (𝛼4 𝜔 + 𝛼1 𝛼4 sin 𝜔𝜏10 ) cos 𝜔𝜏4
dRe𝜆
sign { | }
d𝜏 𝜏=𝜏01 = 𝛼1 𝜔 cos 𝜔𝜏10 .
d𝜆 −1 Eliminating the parameter 𝜏4 , (38) leads to
= sign {Re ( ) |𝜏=𝜏01 }
d𝜏 𝑐20 (𝜔) + 𝑐21 (𝜔) sin 𝜔𝜏10 = 0, (39)
𝜆𝜏
2𝜆𝑒 − (𝛼1 + 𝛼4 ) 𝜏 where 𝑐20 (𝜔) = 𝜔 4
+ (𝛼12
+ 2𝛼2 𝛼3 − 𝛼42 )𝜔2 + (𝛼2 𝛼3 ) − (𝛼1 𝛼4 )2
2
= sign {Re [ − ] } 3
2 𝜆𝜏 −𝜆𝜏
𝜆 [(−𝜆 + 𝛼2 𝛼3 ) 𝑒 + 𝛼1 𝛼4 𝑒 ] 𝜆 𝜏=𝜏01 and 𝑐21 (𝜔) = 2𝛼1 𝜔 + 2𝛼1 (𝛼2 𝛼3 − 𝛼42 )𝜔 .
6 Abstract and Applied Analysis

Denote 𝑔(𝜔) = 𝑐20 (𝜔) + 𝑐21 (𝜔) sin 𝜔𝜏10 . It is obvious In the following, we will discuss the distribution of roots of
that lim𝜔 → ∞ 𝑔(𝜔) = ∞ and 𝑔(0) = (𝛼2 𝛼3 )2 − (𝛼1 𝛼4 )2 < (43) while 𝜏1 and 𝜏3 are given different values.
0 when (H1) holds. Without loss of generality, we assume When 𝜏1 = 0, 𝜏3 = 0, (43) becomes (11) and the
that (39) has at least finite positive roots, which are defined equilibrium 𝐸∗ (𝑥∗ , 𝑦∗ ) is locally asymptotically stable when
by 𝜔41 , 𝜔42 , . . . , 𝜔4𝑘 . From (38), for every fixed 𝜔4𝑖 (𝑖 = (H1) holds.
1, 2, . . . , 𝑘), we have
Case 3A. Consider 𝜏1 ≠ 0, 𝜏3 = 0. Equation (43) can be
(𝑗)
𝜏4𝑖 written in the form

1 𝜆2 − 𝛼4 𝜆 − 𝛼2 𝛼3 + (−𝛼1 𝜆 + 𝛼1 𝛼4 ) 𝑒−𝜆𝜏1 = 0. (44)

= {arccos ( (𝛼1 𝛼2 𝛼3 𝛼4 cos 𝜔4𝑖 𝜏10 )
𝜔4𝑖
Equation (44) is the same as (22), so we do not prove it.
2 −1
× (𝛼42 𝜔42𝑖 + (𝛼1 𝛼4 ) + 2𝛼1 𝛼42 𝜔4𝑖 sin 𝜔4𝑖 𝜏10 ) ) Case 3B. Consider 𝜏1 = 0, 𝜏3 ≠ 0. Equation (43) can be
written in the form
+ 2𝑗𝜋} , 𝑖 = 1, 2, . . . , 𝑘, 𝑗 = 0, 1, 2, . . . .
(40) 𝜆2 − (𝛼1 + 𝛼4 ) 𝜆 + 𝛼1 𝛼4 − 𝛼2 𝛼3 𝑒−𝜆𝜏3 = 0. (45)

Define 𝜏4∗ = 𝜏4(0)

(𝑗)
= min{𝜏4𝑖 | 𝑖 = 0, 1, 2, . . . , 𝑘, 𝑗 = Equation (45) is the same as (17), so we also do not prove it.
𝑖
1, 2, . . .} , 𝜔0 = 𝜔4𝑖 . Equation (39) has a pair of purely Case 3C. Consider 𝜏1 = 𝜏3 = 𝜏. Equation (43) is written in
imaginary roots ±𝜔4∗ 𝑖 for 𝜏10 ∈ [0, 𝜏10 ). the form
In the following, differentiating Equation (21) with
respect to 𝜏4 and substituting 𝜏4 = 𝜏4∗ , we can get 𝜆2 − 𝛼4 𝜆 + (−𝛼1 𝜆 + 𝛼1 𝛼4 − 𝛼2 𝛼3 ) 𝑒−𝜆𝜏 = 0. (46)

d𝜆 (𝜏4 )
−1
𝑃 𝑄 + 𝑃1𝐼 𝑄1𝐼 Suppose that 𝜆 = 𝜔𝑖 (𝜔 > 0) is the root of (46) and we have
sign {Re [ ] } = − sign { 1𝑅 21𝑅 2
}, that
d𝜏4 𝜏 =𝜏 𝑄1𝑅 + 𝑄1𝐼
4 4∗
(41) (𝛼1 𝛼4 − 𝛼2 𝛼3 ) cos 𝜔𝜏 − 𝛼1 𝜔 sin 𝜔𝜏 = 𝜔2 ,
(47)
where (𝛼1 𝛼4 − 𝛼2 𝛼3 ) sin 𝜔𝜏 + 𝛼1 𝜔 cos 𝜔𝜏 = −𝛼4 𝜔.
𝑃1𝑅 = 𝛼1 𝜏10 sin 𝜔4∗ 𝜏10 − 𝛼1 cos 𝜔4∗ 𝜏10 − 𝛼4 cos 𝜔4∗ 𝜏4∗ It obtains that
2
− 𝛼1 𝛼4 𝜏10 cos 𝜔4∗ (𝜏10 + 𝜏4∗ ) , 𝜔4 + (𝛼42 − 𝛼12 ) 𝜔2 − (𝛼1 𝛼4 − 𝛼2 𝛼3 ) = 0. (48)

𝑃1𝐼 = 2𝜔4∗ + 𝛼1 sin 𝜔4∗ 𝜏10 + 𝛼1 𝜏10 𝜔4∗ cos 𝜔4∗ 𝜏10 It is obvious that (48) has only one positive root 𝜔+ =
(42) [(1/2)(𝛼12 − 𝛼42 + √(𝛼42 − 𝛼12 )2 + 4(𝛼1 𝛼4 − 𝛼2 𝛼3 )2 )]1/2 . From
+ 𝛼4 sin 𝜔4∗ 𝜏4∗ + 𝛼1 𝛼4 𝜏10 sin 𝜔4∗ (𝜏10 + 𝜏4∗ ) ,
(47), we obtain that
2
𝑄1𝑅 = −𝛼1 (𝜔4∗ ) cos 𝜔4∗ 𝜏10 , 1 −𝛼2 𝛼3 𝜔+2
𝜏(𝑗) = {𝑎𝑟𝑐 cos 2
+ 2𝑗𝜋} ,
3 2 𝜔+ 𝛼12 𝜔+2 + (𝛼1 𝛼4 − 𝛼2 𝛼3 )
𝑄1𝐼 = (𝜔4∗ ) + 𝛼2 𝛼3 𝜔4∗ + 𝛼1 (𝜔4∗ ) sin 𝜔4∗ 𝜏10 . (49)

If the following assumption 𝑗 = 0, 1, 2, . . . .

(H2) 𝑃1𝑅 𝑄1𝑅 + 𝑃1𝐼 𝑄1𝐼 < 0 Equation (46) has a pair of purely imaginary roots ±𝜔+ 𝑖.
Define 𝜏03 = 𝜏+(0) = min{𝜏(𝑗) , 𝑗 = 0, 1, 2, . . .}, 𝜔0 = 𝜔+ . Let
holds, then the following results on stability and bifurcation
of system (7) are obtained by the general Hopf bifurcation 𝜆(𝜏) = 𝜉(𝜏) + 𝜔(𝜏)𝑖 be the root of (46) near 𝜏 = 𝜏(𝑗) satisfying
theorem for FDEs in Hale [14]. 𝜉(𝜏(𝑗) ) = 0, 𝜔(𝜏(𝑗) ) = 𝜔0 .

Theorem 8. For system (7), suppose that (H1) and (H2) Lemma 9. If the condition (H1) holds, then the following
hold and 𝜏10 ∈ [0, 𝜏10 ). Then, system (7) undergoes a Hopf transversality condition is satisfied:
bifurcation at the positive equilibrium 𝐸∗ when 𝜏4 = 𝜏4∗ . That
𝑑 (Re 𝜆)
is, system (7) has a branch of periodic solution bifurcation from | > 0. (50)
the zero solution near 𝜏4 = 𝜏4∗ . 𝑑𝜏 𝜏=𝜏03
Proof. Differentiating (46) with respect to 𝜏 and noticing that
2.3. Only Considering 𝜏1 and 𝜏3 . The characteristic equation 𝜆 is a function of 𝜏, we can obtain that
of system (7) with 𝜏2 = 𝜏4 = 0 is
d (Re 𝜆) −1 2𝜆 − 𝛼4 − 𝛼1 𝑒−𝜆𝜏 𝜏
( ) = − . (51)
𝜆2 − 𝛼4 𝜆 + (−𝛼1 𝜆 + 𝛼1 𝛼4 ) 𝑒−𝜆𝜏1 − 𝛼2 𝛼3 𝑒−𝜆𝜏3 = 0. (43) d𝜏 𝜆 (−𝛼1 𝜆 + 𝛼1 𝛼4 − 𝛼2 𝛼3 ) 𝑒−𝜆𝜏 𝜆
Abstract and Applied Analysis 7

For simplifying, define 𝜔+ as 𝜔 and 𝜏(𝑗) as 𝜏, and we can obtain and then ±𝜔3𝑖 𝑖 are a pair of purely imaginary roots of (43). Let
(𝑗)
d (Re 𝜆) 𝜏3∗ = 𝜏3(0) = min{𝜏3𝑖 | 𝑖 = 1, 2, . . . , 𝑘, 𝑗 = 1, 2, . . .}, 𝜔3∗ = 𝜔3𝑖 .
sign { | } 𝑖

d𝜏 𝜏=𝜏03 When 𝜏3 = 𝜏3∗ , (43) has a pair of purely imaginary roots ±𝜔3∗ 𝑖
for 𝜏1∗ ∈ [0, 𝜏10 ). Let 𝜆(𝜏) = 𝜉(𝜏) + 𝜔(𝜏)𝑖 be the root of (43)
d𝜆 −1 near 𝜏 = 𝜏3∗ satisfying 𝜉(𝜏3∗ ) = 0, 𝜔(𝜏3∗ ) = 𝜔3∗ .
= sign {Re ( ) |𝜏=𝜏03 }
d𝜏 In the following, differentiating Equation (43) with
respect to 𝜏3 and substituting 𝜏3 = 𝜏3∗ , we get
2𝜆 − 𝛼4 − 𝛼1 𝑒−𝜆𝜏 𝜏
= sign {Re [ − ] } d𝜆 (𝜏) −1 𝑃 𝑄 + 𝑃2𝐼 𝑄2𝐼
𝜆 (−𝛼1 𝜆 + 𝛼1 𝛼4 − 𝛼2 𝛼3 ) 𝑒−𝜆𝜏 𝜆 𝜏=𝜏 sign {Re [ ] } = − sign { 2𝑅 22𝑅 2
},
03
d𝜏3 𝜆=𝜔∗ 𝑖
3
𝑄2𝑅 + 𝑄2𝐼
1 (56)
= sign {( ) (𝛼12 𝜔4 + 2 (𝛼1 𝛼4 − 𝛼2 𝛼3 ) 𝜔2
𝜔2
where
2
+ 𝛼42 (𝛼1 𝛼4 − 𝛼2 𝛼3 ) ) 𝑃2𝑅 = 𝛼4 + 𝛼1 𝜏1∗ 𝜔3∗ sin 𝜔3∗ 𝜏1∗ − (𝛼1 + 𝛼1 𝛼4 𝜏1∗ ) cos 𝜔3∗ 𝜏1∗ ,

× ([𝛼1 𝜔2 − 𝛼4 (𝛼2 𝛼3 − 𝛼1 𝛼4 )]
2
𝑃2𝐼 = 2𝜔3∗ + 𝛼1 𝜏1∗ 𝜔3∗ cos 𝜔3∗ 𝜏1∗ + (𝛼1 + 𝛼1 𝛼4 𝜏1∗ ) sin 𝜔3∗ 𝜏1∗ ,
2 2
2 −1 𝑄2𝑅 = − 𝛼4 (𝜔3∗ ) + 𝛼1 (𝜔3∗ ) cos 𝜔3∗ 𝜏1∗ + 𝛼1 𝛼4 𝜔3∗ sin 𝜔3∗ 𝜏1∗ ,
+ (2𝛼1 𝛼4 − 𝛼2 𝛼3 ) 𝜔2 ) } .
3 2
(52) 𝑄2𝐼 = − (𝜔3∗ ) + 𝛼1 𝛼4 𝜔∗ cos 𝜔3∗ 𝜏1∗ − 𝛼1 (𝜔3∗ ) sin 𝜔3∗ 𝜏1∗ .
(57)
Thus, if the condition (H1) holds, the transversality condition
is satisfied. This ends the proof. Assume that
(H4) 𝑃2𝑅 𝑄2𝑅 + 𝑃2𝐼 𝑄2𝐼 < 0.
Theorem 10. For system (7), if (H1) holds, then there exists
a positive number 𝜏03 such that the interior equilibrium 𝐸∗ is Therefore, by the general Hopf bifurcation theorem for
locally asymptotically stable for 0 < 𝜏 < 𝜏03 and unstable for FDEs in Hale [14], we have the following results on stability
𝜏 > 𝜏03 . Further, system (7) undergoes a Hopf bifurcation at the and bifurcation of system (7).
equilibrium 𝐸∗ for 𝜏 = 𝜏03 .
Theorem 11. For system (7), suppose that (H1), (H3), and (H4)
Case 3D. Consider 𝜏1∗ ∈ [0, 𝜏10 ) and 𝜏3 ≠0. Regarding 𝜏3 as hold and 𝜏1∗ ∈ [0, 𝜏10 ). Then, system (7) undergoes a Hopf
a parameter, we consider (43) with 𝜏1 in its stable interval. bifurcation at the positive equilibrium 𝐸∗ when 𝜏3 = 𝜏3∗ . That
Without loss of generality, we consider system (7) under is, system (7) has a branch of periodic solutions bifurcating from
Case 3A. Let 𝜔𝑖 (𝜔 > 0) be a root of (43). We can obtain the the zero solution near 𝜏3 = 𝜏3∗ .
following results by selecting the value of delay 𝜏1∗ ∈ [0, 𝜏10 ):
Remark 12. If we only consider system (7) with delays 𝜏1 =
𝛼2 𝛼3 cos 𝜔𝜏3 = 𝜔2 − 𝛼1 𝛼4 cos 𝜔𝜏1∗ + 𝛼1 𝜔 sin 𝜔𝜏1∗ , 𝜏3 = 0, we have the characteristic equation
(53)
𝛼2 𝛼3 sin 𝜔𝜏3 = 𝛼4 𝜔 + 𝛼1 𝛼4 sin 𝜔𝜏1∗ + 𝛼1 𝜔 cos 𝜔𝜏1∗ . 𝜆2 − 𝛼2 𝜆 + (−𝛼4 𝜆 + 𝛼2 𝛼4 ) 𝑒−𝜆𝜏4 − 𝛼2 𝛼3 𝑒−𝜆𝜏2 = 0, (58)
Eliminating the parameter 𝜏3 , (53) leads to which is similar to (43). Thus, we do not discuss the roots
𝑐30 (𝜔) + 𝑐31 (𝜔) sin 𝜔𝜏1∗ = 0, (54) of (58). Similarly, if we only consider system (7) with delays
𝜏1 = 𝜏2 = 0 or 𝜏3 = 𝜏4 = 0, we can obtain the corresponding
2 2
where 𝑐30 (𝜔) = 𝜔 + (𝛼12
4
+ 𝛼42 )𝜔2 + (𝛼1 𝛼4 ) − (𝛼2 𝛼3 ) , and characteristic equation which is also similar to (43). The
𝑐31 (𝜔) = 2𝛼1 𝜔3 + 2𝛼1 𝛼42 𝜔. detailed discussions are also omitted.
Suppose that
(H3) Equation (53) has at least finite positive roots. 2.4. Considering 𝜏1 = 𝜏2 = 𝜏4 = 𝜏 and 𝜏3 . We can obtain that
the characteristic equation of system (7) with 𝜏1 = 𝜏2 = 𝜏4 = 𝜏
If (H3) holds, the roots of (53) are defined by and 𝜏3 = 0 is
𝜔31 , 𝜔32 , . . . , 𝜔3𝑘 . From (53), for every fixed 𝜔3𝑖 (𝑖 =
(𝑗) 𝜆2 − [(𝛼1 + 𝛼4 ) 𝜆 + 𝛼2 𝛼3 ] 𝑒−𝜆𝜏 + 𝛼1 𝛼4 𝑒−2𝜆𝜏 = 0. (59)
1, 2, . . . , 𝑘), there exists a sequence {𝜏3𝑖 | 𝑗 = 1, 2, . . .}, where
(𝑗)
When 𝜏1 = 𝜏2 = 𝜏4 = 0, (59) becomes (11) and the
𝜏3𝑖 equilibrium 𝐸∗ (𝑥∗ , 𝑦∗ ) is locally asymptotically stable when
(H1) holds.
1 𝜔32 − 𝛼1 𝛼4 cos 𝜔3𝑖 𝜏1∗ + 𝛼1 𝜔3𝑖 sin 𝜔3𝑖 𝜏1∗ Multiplying 𝑒𝜆𝜏 into the both sides of (59) and letting
= {arccos 𝑖 𝜔𝑖 (𝜔 > 0) be the root of (59), we have
𝜔3𝑖 𝛼2 𝛼3 (55)
(𝛼1 𝛼4 − 𝜔2 ) cos 𝜔𝜏 = 𝛼2 𝛼3 ,
+ 2𝑗𝜋} , 𝑖 = 1, 2, . . . , 𝑘, 𝑗 = 0, 1, 2, . . . , (60)
(𝛼1 𝛼4 + 𝜔2 ) sin 𝜔𝜏 = − (𝛼1 + 𝛼4 ) 𝜔.
8 Abstract and Applied Analysis

Since sin2 𝜔𝜏 + cos2 𝜔𝜏 = 1, we obtain that Theorem 13. For system (7), if (H1) and (H5) hold, then there
12 10 8 6 4 2 exists a positive number 𝜏04 such that the positive equilibrium
𝜔 + 𝑑5 𝜔 + 𝑑4 𝜔 + 𝑑3 𝜔 + 𝑑2 𝜔 + 𝑑1 𝜔 + 𝑑0 = 0, (61) 𝐸∗ is locally asymptotically stable for 0 < 𝜏 < 𝜏04 and unstable
where 𝑑5 = −𝛼12 𝛼42 , 𝑑4 = 𝛼12 𝛼42 + 2𝛼1 𝛼4 (𝛼1 + 𝛼4 ) − 𝛼22 𝛼32 , 𝑑3 = for 𝜏 > 𝜏04 . Further, system (7) undergoes a Hopf bifurcation
at the equilibrium 𝐸∗ for 𝜏 = 𝜏04 .
2𝛼1 𝛼4 [𝛼13 𝛼43 −2𝛼1 𝛼4 (𝛼1 +𝛼4 )2 − 𝛼22 𝛼32 ], 𝑑2 = −𝛼14 𝛼44 +2𝛼13 𝛼43 (𝛼1 +
𝛼4 )2 −2𝛼12 +𝛼42 𝛼22 𝛼32 , 𝑑1 = −𝛼16 𝛼46 −2𝛼22 𝛼32 𝛼13 𝛼43 , and 𝑑0 = 𝛼16 𝛼46 − Further, the corresponding characteristic equation with
𝛼14 𝛼44 𝛼22 𝛼32 . Without losses of generality, suppose that (61) has 𝜏1 = 𝜏2 = 𝜏4 = 𝜏 and 𝜏3 ≠0 is
six different positive roots, defined by 𝜔𝑘+ (𝑘 = 1, 2, 3, 4, 5, 6).
From (60), if we denote 𝜆2 − (𝛼1 + 𝛼4 ) 𝜆𝑒−𝜆𝜏 − 𝛼2 𝛼3 𝑒−𝜆𝜏 𝑒−𝜆𝜏3 + 𝛼1 𝛼4 𝑒−2𝜆𝜏 = 0. (66)

(𝑗) 1 𝛼2 𝛼3 When 𝜏1 = 𝜏2 = 𝜏3 = 𝜏4 = 0, (66) becomes (11).

𝜏𝑘 = + {arccos ( 2
) + 2𝑗𝜋} , The positive equilibrium 𝐸∗ (𝑥∗ , 𝑦∗ ) is locally asymptotically
𝜔𝑘 𝛼1 𝛼4 − (𝜔𝑘+ ) (62) stable when (H1) holds.
𝑘 = 1, 2, 3, 4, 5, 6, 𝑗 = 0, 1, 2, . . . , Regarding 𝜏3 as a parameter, we consider (66) with 𝜏 in its
stable interval. Without loss of generality, we consider system
then ±𝜔𝑘+ 𝑖 are a pair of purely imaginary roots of (59) with (7) under the above case. Multiplying 𝑒𝜆𝜏 into the both sides
(𝑗)
𝜏 = 𝜏𝑘 . of (66) and letting 𝜆 = 𝜔𝑖 (𝜔 > 0) be the root of (66), we
Define 𝜏04 = 𝜏𝑘(0) = min𝑘∈{1,2,3,4,5,6} {𝜏𝑘 , 𝑗 = 0, 1, 2, . . .},
(𝑗) have that by selecting the value of delay 𝜏∗∗ ∈ (0, 𝜏04 )
0
𝜔0 = 𝜔𝑘0 . Let 𝜆(𝜏) = 𝜉(𝜏) + 𝜔(𝜏)𝑖 be the root of (59) near 𝜏 = 𝛼2 𝛼3 cos 𝜔𝜏3 = (𝛼1 𝛼4 − 𝜔2 ) cos 𝜔𝜏∗∗ ,
𝜏04 satisfying 𝜉(𝜏04 ) = 0, 𝜔(𝜏04 ) = 𝜔0 . Next, differentiating (67)
(59) with respect to 𝜏, it follows that 𝛼2 𝛼3 sin 𝜔𝜏3 = (𝛼1 + 𝛼4 ) 𝜔 + (𝛼1 𝛼4 + 𝜔2 ) sin 𝜔𝜏∗∗ .
d𝜆 −1 2𝜆𝑒𝜆𝜏 − 𝛼1 − 𝛼4 𝜏 since sin2 𝜔𝜏 + cos2 𝜔𝜏 = 1, we obtain that
( ) = − . (63)
d𝜏 −𝜆 (𝜆2 𝑒𝜆𝜏 − 𝛼1 𝛼4 𝑒−𝜆𝜏 ) 𝜆
(𝑗)
𝜔4 + 𝑑̃4 𝜔2 + 𝑑̃3 + 𝑑̃2 cos 𝜔𝜏∗∗ + 𝑑̃1 sin 𝜔𝜏∗∗ = 0, (68)
For simplifying, define 𝜔𝑘+ as 𝜔 and 𝜏𝑘 as 𝜏, and we can obtain
where 𝑑̃4 = (𝛼1 + 𝛼4 )2 , 𝑑̃3 = 𝛼12 𝛼42 − 𝛼22 𝛼32 , 𝑑̃2 = −2𝛼1 𝛼4 𝜔2 , and
d (Re 𝜆)
sign { | } 𝑑̃1 = 2(𝛼1 + 𝛼4 )(𝜔3 + 𝛼1 𝛼4 𝜔).
d𝜏 𝜏=𝜏04 Suppose that
d𝜆 −1 (H6) Equation (68) has at least finite positive roots.
= sign {Re ( ) |𝜆=𝜔𝑘+ 𝑖 }
d𝜏
If (H6) holds, then (68) has finite positive roots, given
𝜆𝜏 by 𝜔13 , 𝜔23 , . . . , 𝜔𝑚3 . From (67), for every fixed 𝜔𝑘3 (𝑘 =
{ 2𝜆𝑒 − 𝛼1 − 𝛼4 𝜏 } (𝑗)
= sign {Re [ − ] 1, 2, . . . , 𝑚), if there exists a sequence {𝜏𝑘 | 𝑗 = 0, 1, 2, . . .}
−𝜆 (𝜆 𝑒 − 𝛼1 𝛼4 𝑒 ) 𝜆 𝜆=𝜔+ 𝑖 }
2 𝜆𝜏 −𝜆𝜏 3
{ 𝑘 } such that (68) holds, where
(64)
= sign { Re [ (− (𝛼1 + 𝛼4 + 2𝜔 sin 𝜔𝜏) (2𝜔 cos 𝜔𝜏) 𝑖) (𝑗) 1 { (𝛼1 𝛼4 − 𝜔𝑘23 ) cos 𝜔𝑘3 𝜏∗∗ }
𝜏𝑘 = { arccos ( ) + 2𝑗𝜋} ,
3 𝜔𝑘3 𝛼2 𝛼3
× (𝜔𝑖 [(𝜔2 + 𝛼1 𝛼4 ) cos 𝜔𝜏 { }
−1 𝑘 = 1, 2, . . . , 𝑚, 𝑗 = 0, 1, 2, . . . ,
+ (𝜔2 − 𝛼1 𝛼4 ) sin 𝜔𝜏𝑖]) ]} (69)
1 then ±𝜔𝑘3 𝑖 are a pair of purely imaginary root of (66).
= 2
sign {𝑄𝑅 } ,
𝜔(𝛼12 𝛼42 − 𝜔4 ) (𝑃𝑅2 + 𝑃𝐼2 ) Define 𝜏3∗∗ = 𝜏𝑘∗ (0)
= min𝑘∈{1,2,...,𝑚} {𝜏𝑘(0) }, 𝜔∗ = 𝜔𝑘∗ . Let
𝜆(𝜏) = 𝜉(𝜏) + 𝜔(𝜏)𝑖 be the root of (66) near 𝜏 = 𝜏3∗∗ satisfying
where 𝑃𝑅 = (𝜔2 + 𝛼1 𝛼4 ) cos 𝜔𝜏, 𝑃𝐼 = (𝜔2 − 𝛼1 𝛼4 ) sin 𝜔𝜏, and
𝜉(𝜏3∗∗ ) = 0, 𝜔(𝜏3∗∗ ) = 𝜔∗ . Next, we will check whether the
𝑄𝑅 = 2𝜔3 (𝛼12 𝛼42 − 𝜔4 )2 + 2𝜔𝛼1 𝛼4 𝛼22 𝛼42 (𝛼1 𝛼4 + 𝜔2 )2 + 𝛼1 𝛼4 (𝛼1 + following transversality condition
𝛼4 )(𝛼1 𝛼4 − 𝜔2 )2 [𝜔4 − 2(𝛼1 + 𝛼4 )𝜔2 − 𝛼12 𝛼42 ].
Since 𝛼1 + 𝛼4 < 0, we suppose that dRe (𝜆)
| 𝜏 = 𝜏3∗∗ ≠0 (70)
4
(H5) 𝜔 − 2(𝛼1 + 𝛼4 )𝜔 − 2
𝛼12 𝛼42 < 0. d𝜏
is true.
If (H5) holds, then the transversality condition
Differentiating (66) with respect to 𝜏3 , it follows that
d (Re 𝜆)
| 𝜏 = 𝜏04 > 0 (65) 2 𝜆𝜏 −𝜆𝜏
d𝜆 −1 (2𝜆 + 𝜆 𝜏) 𝑒 − 𝛼1 𝛼4 𝜏𝑒 − 𝛼1 − 𝛼4 𝜏3
d𝜏 ( ) = − .
holds and a Hopf bifurcation occurs at 𝜏 = 𝜏04 . We have the d𝜏 −𝛼2 𝛼3 𝜆𝑒−𝜆𝜏 𝜆
following theorem. (71)
Abstract and Applied Analysis 9

(𝑗)
For simplifying, define 𝜔𝑘3 as 𝜔 and 𝜏𝑘 as 𝜏3 , and we can scaling 𝑡 → (𝑡/𝜏), and then (7) is transformed into an FDE
3
obtain in 𝐶 = 𝐶([−1, 0], 𝑅2 ) as
d (Re 𝜆) −1 𝑢̇(𝑡) = 𝐿 𝜇 𝑢𝑡 + 𝐹 (𝜇, 𝑢𝑡 ) , (73)
sign {[ ] }
d𝜏 𝜏3 =𝜏3∗∗
where 𝑢(𝑡) = (𝑢1 (𝑡), 𝑢2 (𝑡))𝑇 ∈ 𝑅2 and 𝐿 𝜇 : 𝐶 → 𝑅2 , 𝐹 :
2 𝜆𝜏 −𝜆𝜏
(2𝜆 + 𝜆 𝜏) 𝑒 − 𝛼1 𝛼4 𝜏𝑒 − 𝛼1 − 𝛼4 𝑅 × 𝐶 → 𝑅2 are given, respectively, by
= sign {Re [
−𝛼2 𝛼3 𝜆𝑒−𝜆𝜏
𝐿 𝜇 𝑢𝑡 = 𝜏0 [𝐵 (𝜏) 𝑢𝑡 + 𝐶 (𝜏) 𝑢 (𝑡 − 1)] ,
(72)
𝜏3 } (74)
𝑇
− ] 𝐹 (𝜇, 𝑢𝑡 ) = (𝜏0 + 𝜇) (𝐹1 , 𝐹2 ) ,
𝜆 𝜆=𝜔 𝑖 }
∗ }
where
𝑃𝑅󸀠 𝑄𝑅󸀠 + 𝑃𝐼󸀠 𝑄𝐼󸀠
= sign { }, 0 𝛼2 𝛼 0
2
(𝑃𝑅󸀠 ) + (𝑃𝐼󸀠 )
2 𝐵 (𝜏) = ( ), 𝐶 (𝜏) = ( 1 ),
𝛼3 0 0 𝛼4
where 𝑃𝑅󸀠 = 𝛼2 𝛼3 𝜔 cos 𝜔𝜏3 , 𝑃𝐼󸀠 = 𝛼2 𝛼3 𝜔 sin 𝜔𝜏3 , 𝑄𝑅󸀠 = (𝛼1 𝛼4 + (75)
𝐹1 = 𝑏11 𝜙1 (0) 𝜙1 (−1) + 𝑏12 𝜙1 (0) 𝜙2 (0) ,
𝜔2 )𝜏 cos 𝜔𝜏 + 2𝜔 sin 𝜔𝜏 + 𝛼1 + 𝛼4 , and 𝑄𝐼󸀠 = −2𝜔 cos 𝜔𝜏 +
𝜏(𝜔2 − 𝛼1 𝛼4 ) sin 𝜔𝜏. 𝐹2 = 𝑏21 𝜙1 (0) 𝜙2 (0) + 𝑏22 𝜙2 (0) 𝜙2 (−1) ,
Suppose that
here 𝑏11 = −𝑟1 , 𝑏12 = 𝑎/𝑟1 , 𝑏21 = 𝑏/𝑟2 , 𝑏22 = −𝑟2 .
(H7) 𝑃𝑅󸀠 𝑄𝑅󸀠 + 𝑃𝐼󸀠 𝑄𝐼󸀠 > 0. Turning to the linear problem
If (H7) holds, then the transversality condition holds and a 𝑢̇(𝑡) = 𝐿 𝜇 𝑢𝑡 , (76)
Hopf bifurcation occurs at 𝜔 = 𝜔∗ , 𝜏 = 𝜏3∗∗ . We have the
following theorem. by the Reisz representation theorem, there exists 2×2 matrix-
valued function
Theorem 14. For system (7), if (H1), (H6), and (H7) hold,
then there exists a positive number 𝜏3∗∗ such that the positive 𝜂 (⋅, 𝜇) : [−1, 0] 󳨀→ 𝑅2×2 , (77)
equilibrium 𝐸∗ is locally asymptotically stable for 0 < 𝜏 < 𝜏3∗∗
and unstable for 𝜏 > 𝜏3∗∗ . Further, system (7) undergoes a Hopf such that
bifurcation at the equilibrium 𝐸∗ for 𝜏3 = 𝜏3∗∗ . 0
𝐿 𝜇 (𝜙) = ∫ d𝜂 (𝜃, 𝜇) 𝜙 (𝜃) , 𝜙 ∈ 𝐶. (78)
Remark 15. In addition, if we consider system (7) with 𝜏1 = −1
𝜏3 = 𝜏4 = 𝜏 and 𝜏3 ≠0, we can obtain the same characteristic
equation as (66) and the same result for the time delay 𝜏2 . So, In fact, we can choose
we omit it.
𝜂 (𝜃, 𝜇) = (𝜏0 + 𝜇) (𝐵 (𝜏) 𝛿 (𝜃) + 𝐶 (𝜏) 𝛿 (𝜃 + 1)) , (79)

3. Direction and Stability of where 𝛿 denotes the Dirac delta function. Then, (78) is
the Hopf Bifurcation satisfied.
Next, for 𝜙 ∈ 𝐶([−1, 0], 𝑅2 ), we define the operator 𝐴(𝜇)
In this section, we will study the direction of the Hopf as
bifurcation and the stability of bifurcating periodic solution
of system (7). The approach employed here is the normal form d𝜙
{
{ , 𝜃 ∈ [−1, 0) ,
method and center manifold theorem introduced by Hassard {
{
{ d𝜃
et al. [20]. More precisely, we will compute the reduced system 𝐴 (𝜇) 𝜙 (𝜃) = {
{
{ 0
on the center manifold with the pair of conjugate complex, {
{∫ d𝜂 (𝜃, 𝜇) 𝜙 (𝜃) ,
purely imaginary solution of the characteristic equation of 𝜃 = 0, (80)
{ −1
system (10). By this reduction, we can determine the Hopf
bifurcation direction, that is, to answer the question of 0, 𝜃 ∈ [−1, 0) ,
whether the bifurcation branch of periodic solution exists 𝑅 (𝜇) 𝜙 (𝜃) = {
𝐹 (𝜇, 𝜙) , 𝜃 = 0.
locally for supercritical bifurcation or subcritical bifurcation.
Without loss of generality, we only consider the case with Since d𝑢𝑡 /d𝜃 = d𝑢𝑡 /d𝑡, then system (73) is equivalent to the
𝜏1 = 𝜏4 = 𝜏 and 𝜏2 = 𝜏3 = 0. For convince, we denote following operator equation:
any one of these critical values 𝜏 = 𝜏𝑘 (𝑘 = 0, 1, 2, . . .) by
𝜏0 , at which the characteristic equation has a pair of purely 𝑢̇(𝑡) = 𝐴 (𝜇) 𝑢𝑡 + 𝑅 (𝜇) 𝑢𝑡 , (81)
imaginary roots ±𝜔0 𝑖. Let 𝜏 = 𝜏0 + 𝜇, 𝜇 ∈ 𝑅; then, 𝜇 = 0 is
the Hopf bifurcation value of system (7). We first let 𝑢1 (𝑡) = where 𝑢𝑡 = 𝑡(𝑡 + 𝜃), for 𝜃 ∈ [−1, 0], which is an equation of
𝑥(𝑡) − 𝑥∗ , 𝑢2 (𝑡) = 𝑦(𝑡) − 𝑦∗ and normalize the delay with the the form we desired.
10 Abstract and Applied Analysis

For 𝜓 ∈ 𝐶󸀠 ([−1, 0], (𝑅2 )∗ ), we further define the adjoint 𝐾13 = 𝑏11 𝑒𝑖𝜔0 𝜏0 + 𝑏12 𝜌1 ,
∗
𝐴 of 𝐴 as
(1)
d𝜓 (𝑠) (1) 𝑊20 (−1)
{ 𝐾14 = 𝑏11 [2𝑊11 (−1) +
∗ {− d𝑠 , 𝑠 ∈ (0, 1] , 2
𝐴 (𝜇) 𝜓 (𝑠) = { 0 (82)
{∫ 𝜓
(−𝑠) d𝜂 (𝑠, 𝜇) , 𝑠 = 0, (1)
𝑊20 (0) 𝑖𝜔0 𝜏0
{ −1 (1)
+ 𝑒 + 𝑊11 (0) 𝑒−𝑖𝜔0 𝜏0 ]
and a bilinear form 2
𝑇 (2)
⟨𝜓 (𝑠) , 𝜙 (𝜃)⟩ = 𝜓 (0) 𝜙 (0) (2) 𝑊20 (0)
+ 𝑏12 [𝑊11 (0) +
0 𝜃 (83) 2
𝑇
−∫ ∫ 𝜓 (𝜉 − 𝑠) d𝜂 (𝜃) 𝜙 (𝜉) d𝜉,
(1)
𝜃=−1 𝜉=0 𝑊20 (0) (1)
+ 𝜌1 + 𝑊11 (0) 𝜌1 ] ,
where 𝜂(𝜃) = 𝜂(𝜃, 0). Then, 𝐴(0) and 𝐴∗ (0) are adjoint 2
operators. From the above analysis, we obtain that ±𝑖𝜔0 𝜏0
are the eigenvalues of 𝐴(0) and therefore they are also 𝐾21 = 𝑏21 𝜌1 + 𝑏22 𝑒−𝑖𝜔0 𝜏0 ,
the eigenvalues of 𝐴∗ (0). Let 𝑞(𝜃) be eigenvector of 𝐴(0)
corresponding to 𝑖𝜔0 𝜏0 and let 𝑞∗ (𝜃) be eigenvector of 𝐴∗ (0) 𝐾22 = 𝑏21 (𝜌1 + 𝜌1 ) − 𝑏22 (𝑒𝑖𝜔0 𝜏0 + 𝑒−𝑖𝜔0 𝜏0 ) ,
corresponding to −𝑖𝜔0 𝜏0 , and then we have
2
𝐾23 = 𝑏21 𝜌1 + 𝑏21 (𝜌1 ) 𝑒−𝑖𝜔0 𝜏0 ,
𝐴 (0) 𝑞 (𝜃) = 𝑖𝜔0 𝜏0 𝑞 (𝜃) ,
(84) (2) (1)
𝐴∗ (0) 𝑞∗ (𝜃) = −𝑖𝜔0 𝜏0 𝑞∗ (𝜃) . (2) 𝑊20 (0) 𝑊20 (0) (1)
𝐾24 = 𝑏21 [𝑊11 (0) + + 𝜌1 + 𝑊11 (0) 𝜌1 ]
2 2
By some simple computation, we can obtain
(2)
𝑊20 (−1)
𝑞 (𝜃) = 𝑉𝑒𝑖𝜔0 𝜏0 𝜃 , 𝑞∗ (𝑠) = 𝐷𝑉∗ 𝑒−𝑖𝜔0 𝜏0 𝑠 , (85) (2)
− 𝑏22 [𝑊11 (−1) + 𝜌1
2
where
(1)
𝑉 = (1, 𝜌1 ) ,
𝑇
𝑉∗ = (1, 𝜌1∗ ) ,
𝑇 𝑊20 (0) 𝑖𝜔0 𝜏10 (2)
+ 𝜌1 𝑒 + 𝑊11 (0) 𝜌1 𝑒−𝑖𝜔0 𝜏0 ] .
2
−𝑏11 𝑥∗ 𝑒−𝑖𝜔0 𝜏0 + 𝑖𝜔0 (88)
𝜌1 = − ,
𝑏12 𝑥∗
However,
−𝑏 𝑥∗ 𝑒−𝑖𝜔0 𝜏0 + 𝑖𝜔0
𝜌1∗ = − 11 , 𝑖𝑔20 𝑞 (0) 𝑖𝜔0 𝜏0 𝜃 𝑖𝑔02 𝑞 (0) −𝑖𝜔0 𝜏0 𝜃
𝑏21 𝑦∗ (86) 𝑊20 (𝜃) = 𝑒 + 𝑒 + 𝐸1 𝑒2𝑖𝜔0 𝜏0 𝜃 ,
𝜔0 𝜏0 3𝜔0 𝜏0
𝐷 = (−𝑏22 𝑦∗ 𝑒−𝑖𝜔0 𝜏0 + 𝑖𝜔0 ) 𝑖𝑔11 𝑞 (0) 𝑖𝜔0 𝜏0 𝜃 𝑖𝑔11 𝑞 (0) −𝑖𝜔0 𝜏0 𝜃
𝑊11 (𝜃) = − 𝑒 + 𝑒 + 𝐸2 ,
𝜔0 𝜏0 𝜔0 𝜏0
× (− (𝑏11 𝑥∗ + 𝑏22 𝑦∗ ) (1 − 𝑖𝜔0 𝜏0 ) 𝑒−𝑖𝜔0 𝜏0
(89)
∗ ∗ −2𝑖𝜔0 𝜏0 −1
+ 2 (𝜔0 𝑖 − 𝑏11 𝑏22 𝑥 𝑦 𝑒 )) . where 𝐸1 = (𝐸1(1) , 𝐸1(2) )𝑇 and 𝐸2 = (𝐸2(1) , 𝐸2(2) )𝑇 are both two-
Then, ⟨𝑞∗ , 𝑞⟩ = 1, ⟨𝑞∗ , 𝑞⟩ = 0. dimensional vectors and can be determined by
In the remainder of this section, by using the same −1
2𝑖𝜔 𝜏 − 𝑏 𝑥∗ 𝑒−2𝑖𝜔0 𝜏0 𝑏12 𝑥∗
notations as in Hassard et al. [20] and a computation process 𝐸1 = 2( 0 0 11 ∗ )
similar to that in [11], we obtain the coefficients determining −𝑏21 𝑦 2𝑖𝜔0 𝜏0 + 𝑏22 𝑦∗ 𝑒−2𝑖𝜔0 𝜏0
the important quantities of the periodic solution: 𝐾
× ( 11 ) ,
𝑔20 = 2𝐷 (𝐾11 + 𝜌∗1 𝐾21 ) , 𝐾21
−1
𝑔11 = 𝐷 (𝐾12 + 𝜌∗1 𝐾22 ) , 𝑏 𝑥 ∗ 𝑏 𝑦∗ 𝐾
𝐸2 = −( 11 ∗ 12 ∗ ) ( 12 ) .
(87) 𝑏21 𝑥 𝑏22 𝑦 𝐾22
𝑔02 = 2𝐷 (𝐾13 + 𝜌∗1 𝐾23 ) , (90)

𝑔21 = 2𝐷 (𝐾14 + 𝜌∗1 𝐾24 ) , Furthermore, we can see that each 𝑔𝑖𝑗 in (87) is deter-
mined by parameters and delays in system (7). Thus, we can
where compute the following quantities:
𝐾11 = −𝑏11 𝑒−𝑖𝜔0 𝜏0 + 𝑎12 𝜌1 , 󵄨 󵄨2
𝑖 󵄨 󵄨2 󵄨󵄨𝑔 󵄨󵄨 𝑔
𝑖𝜔0 𝜏0 −𝑖𝜔0 𝜏0 𝐶1 (0) = (𝑔20 𝑔11 − 2󵄨󵄨󵄨𝑔11 󵄨󵄨󵄨 − 󵄨 02 󵄨 ) + 21 ,
𝐾12 = 𝑏11 (𝑒 +𝑒 ) + 𝑏12 (𝜌1 + 𝜌1 ) , 2𝜔0 𝜏0 3 2
Abstract and Applied Analysis 11

Re {𝐶1 (0)} (A1) 𝐹̂ ∈ 𝐶2 (𝑅+2 × 𝑅+ × 𝑅+ , 𝑅+2 ).

𝜇2 = − ,
Re {𝜆󸀠 (𝜏0 )}
It follows from system (7) that
𝛽2 = 2 Re {𝐶1 (0)} ,
𝑟 − 2𝑟1 𝑥 + 𝑟1 𝑎𝑦 𝑟1 𝑎𝑥
D𝑧 𝐹̂ (𝑧, 𝜏, 𝑝) = ( 1 ) . (93)
Im {𝐶1 (0)} + 𝜇2 Im {𝜆󸀠 (𝜏0 )} 𝑟2 𝑏𝑦 𝑟2 + 𝑟2 𝑏𝑥 − 2𝑟2 𝑦
𝑇2 = − ,
𝜔0 𝜏0 Then, under the assumption (H1), we have
(91)
−𝑟 𝑥∗ 𝑟 𝑎𝑥∗
det (D𝑧 𝐹̂ (𝑧, 𝜏, 𝑝)) = det ( 1 ∗ 1 ∗ )
which determine the properties of bifurcation periodic solu- 𝑟2 𝑏𝑦 −𝑟2 𝑦 (94)
tions in the center manifold at the critical value 𝜏0 . By the ∗ ∗
results of Hassard et al. [20], we have the following results. = 𝑟1 𝑟2 𝑥 𝑦 (1 − 𝑎𝑏) > 0.

Theorem 16. In (91), the following results hold. Therefore, we have the following condition on the linear
̂ 𝜏, 𝑝):
operator 𝐷𝑧 𝐹(𝑧,
(i) The sign of 𝜇2 determines the direction of the Hopf
bifurcation: if 𝜇2 > 0 (𝜇2 < 0), then the Hopf bifur- ̂ 𝜏, 𝑝) is an isomorphism at the equilibrium 𝑧∗ ;
(A2) 𝐷𝑧 𝐹(𝑧,
cation is supercritical (subcritical) and the bifurcation here, (𝜏, 𝑝) ∈ 𝑅+ × 𝑅+ .
periodic solutions exist for 𝜏 > 𝜏0 (𝜏 < 𝜏0 ).
In addition, we can easily observe that the following result
(ii) The sign of 𝛽2 determines the stability of the bifurcating is true:
periodic solutions: the bifurcation periodic solutions are
stable (unstable) if 𝛽2 < 0 (𝛽2 > 0). (A3) 𝐹(𝜙, 𝜏, 𝑝) is differential with respect to 𝜙.
(iii) The sign of 𝑇2 determines the period of the bifurcating The characteristic matrix of (92) at a stationary solution
periodic solutions: the period increases (decreases) if (𝑧, 𝜏0 , 𝑝0 ), where 𝑧 = (𝑧(1) , 𝑧(2) )𝑇 ∈ 𝑅2 , takes the following
𝑇2 > 0 (𝑇2 < 0). form:

4. Global Continuation of Δ (𝑧, 𝜏, 𝑝) (𝜆) = 𝜆𝐼 − 𝐷𝜙 𝐹 (𝑧0 , 𝜏0 , 𝑝0 ) (𝑒𝜆 𝐼) . (95)

Local Hopf Bifurcation
That is,
In this part, we will study the global continuation of peri-
odic solutions bifurcating from the positive equilibrium Δ (𝑧, 𝜏, 𝑝) (𝜆)
(𝐸∗ , 𝜏(𝑗) ) 𝑗 = 0, 1, 2, . . ., by applying a global Hopf bifurcation 𝜆 − 𝑟1 𝑎𝑧(2) + 𝑟1 𝑧(1) 𝑒−𝜆𝜏 −𝑟1 𝑎𝑧(2)
result due to Wu [21]. =( (1) ).
−𝑟2 𝑏𝑧 𝜆 − 𝑟2 𝑏𝑧(1) + 𝑟2 𝑧(2) 𝑒−𝜆𝜏
Let 𝑅+2 = {(𝑥, 𝑦) ∈ 𝑅2 , 𝑥 > 0, 𝑦 > 0}, 𝑋 =
(96)
𝐶([−𝜏, 0], 𝑅+2 ), and 𝑧𝑡 = (𝑥(𝑡), 𝑦(𝑡))𝑇 ∈ 𝑋 be defined as
𝑧𝑡 (𝜃) = (𝑧1𝑡 (𝜃), 𝑧2𝑡 (𝜃)) = (𝑧1 (𝑡 + 𝜃), 𝑧2 (𝑡 + 𝜃)) = 𝑧(𝑡 + 𝜃) for The zeros of det(Δ(𝑧, 𝜏, 𝑝)(𝜆)) = 0 are called the characteris-
𝑡 ≥ 0 and 𝜃 ∈ [−𝜏, 0]. Since 𝑥(𝑡) and 𝑦(𝑡) denote the density tic roots. Note that (A2) is equivalent to 𝜆 = 0 which is not
of one species and the other species, respectively. The positive a characteristic root of any equilibrium of (92). Clearly, the
solution of system (7) is interior and its periodic solutions characteristic matrix Δ(𝑧, 𝜏, 𝑝)(𝜆) is continuous in (𝜏, 𝑝, 𝜆) ∈
only arise in the first quadrant. Thus, throughout this section, 𝐵𝜀0 (𝜏𝑗 , 2𝜋/𝜔𝑘 ) × C.
we consider system (7) only in the domain 𝑅+2 . A stationary solution (𝑧0 , 𝜏0 , 𝑝0 ) of (92) is called a center
It is easy to see that system (7) can be rewritten as the if det(Δ(𝑧0 , 𝜏0 , 𝑝0 )((2𝑚𝜋/𝑝0 )𝑖)) = 0 for some positive in-
following functional differential equation: teger 𝑚 or det(Δ(𝑧, 𝜏0 , 𝑝0 )(𝜆)) = 0 has purely imaginary
characteristic roots of the form (2𝑚𝜋/𝑝0 )𝑖 for some positive
𝑧̇(𝑡) = 𝐹 (𝑧𝑡 , 𝜏, 𝑝) , (92) integer 𝑚. A center (𝑧0 , 𝜏0 , 𝑝0 ) is said to be isolated if it is the
only center in some neighborhood of (𝑧0 , 𝜏0 , 𝑝0 ).
parameterized by two nonnegative real parameters (𝜏, 𝑝) ∈
From (96), we can see
𝑅+ × 𝑅+ , where 𝑅+ = [0, +∞). It is obvious that (92) has
only an equilibria 𝑧∗ = 𝐸∗ in 𝑅+2 under the assumption (H1), 𝑎12 𝑎21
det (Δ (𝑧∗ , 𝜏, 𝑝) (𝜆)) = 𝜆2 − 𝑏 𝑏 𝑥 ∗ 𝑦∗
and it is easy to see that the mapping 𝐹 : 𝑋 × 𝑅+ × 𝑅+ is 𝑎11 𝑎22 1 2
completely continuous. If we restrict 𝐹 to the subspace of
+ (𝑏1 𝑥∗ + 𝑏2 𝑦∗ ) 𝜆𝑒−𝜆𝜏 (97)
constant function 𝑍, then 𝐹 is identified with 𝑅+2 and thus we
obtain a mapping 𝐹̂ = 𝐹|𝑅+2 ×𝑅+ ×𝑅+ : 𝑅+2 × 𝑅+ × 𝑅+ → 𝑅+2 .
+ 𝑏1 𝑏2 𝑥∗ 𝑦∗ 𝑒−2𝜆𝜏 = 0.
Let 𝑧0 ∈ 𝑋 be the constant mapping with value 𝑧0 ∈ 𝑅+2 .
The point (𝑧0 , 𝜏0 , 𝑝0 ) is called a stationary solution of (92) if Note that the above equation is the same as (31);
̂ 0 , 𝜏0 , 𝑝0 ) = 0. From system (7), we know easily that the
𝐹(𝑧 therefore, it is easy to verify from the discussion
following condition regarding the mapping 𝐹̂ holds: regarding the local Hopf bifurcation in Section 2 that
12 Abstract and Applied Analysis

(𝑧∗ , 𝜏𝑗 , 2𝜋/𝜔0 ), 𝑗 = 0, 1, 2, . . ., is an isolated center and there 𝜏1 = 𝜏2 = 𝜏3 = 𝜏4 = 0

35
exists a smooth curve 𝜆: (𝜏𝑗 − 𝛿, 𝜏𝑗 + 𝛿) → C such that
det(Δ(𝑧∗ , 𝜏, 2𝜋/𝜔0 )(𝜆(𝜏))) = 0, |𝜆(𝜏) − 𝑖𝜔𝑗 | < 𝜀, for all
30
𝜏 ∈ [𝜏𝑗 − 𝛿, 𝜏𝑗 + 𝛿] and
25
d𝜆
𝜆 (𝜏𝑗 ) = 𝑖𝜔0 , Re | > 0. (98)
d𝜏 𝜏=𝜏𝑗 20
For the above 𝜀 > 0, we define the set
15
󵄨󵄨 2𝜋 󵄨󵄨󵄨󵄨
󵄨
Ω𝜀,2𝜋/𝜔0 = {(V, 𝑝) : 0 < V < 𝜀, 󵄨󵄨󵄨𝑝 − 󵄨 < 𝜀} . (99)
󵄨󵄨 𝜔0 󵄨󵄨󵄨 10

It is easy to verify on [𝜏𝑗 − 𝛿, 𝜏𝑗 + 𝛿] × 𝜕Ω𝜀,2𝜋/𝜔0 that the 5

following condition holds:
0
(A4) det(Δ(𝑧∗ , 𝜏, 𝑝)(V + (2𝜋/𝑝)𝑖)) = 0 if and only if V = 0, 0 50 100 150 200 250 300 350 400
𝜏 = 𝜏𝑗 , and 𝑝 = 2𝜋/𝜔0 , 𝑗 = 0, 1, 2, . . .. t

Thus, the conditions (A1)–(A4) in [21] are satisfied. x(t)

y(t)
Moreover, if we define
Figure 1: The positive equilibrium of system (115) is locally asymp-
2𝜋
𝐻± (𝑧∗ , 𝜏𝑗 , ) (V, 𝑝) totically stable for 𝜏1 = 𝜏2 = 𝜏3 = 𝜏4 = 0 with initial value “0.7,
𝜔0 0.5.”
(100)
∗ 2𝜋 2𝜋
= det (Δ (𝑧 , 𝜏𝑗 ± 𝛿, ) (V + 𝑖)) ,
𝜔0 𝑝 Lemma 17. ℓ(𝑧∗ , 𝜏𝑗 , 2𝜋/𝜔0 ) is unbounded for each center
(𝑧∗ , 𝜏𝑗 , 2𝜋/𝜔0 ), 𝑗 = 0, 1, 2, . . ..
then we know 𝐻+ (𝑧∗ , 𝜏𝑗 , 2𝜋/𝜔0 ) ≠ 0 on 𝜕Ω𝜀,2𝜋/𝜔0 . Thus,
the first crossing number 𝛾(𝑧∗ , 𝜏𝑗 , 2𝜋/𝜔0 ) of isolated center Lemma 18. All nonconstant periodic solutions of system (7)
(𝑧∗ , 𝜏𝑗 , 2𝜋/𝜔0 ) can be defined as follows: with initial data 𝑥(𝑡) = 𝜑(𝑡) ≥ 0, 𝑥(0) = 𝜑(0) > 0, 𝑦(𝑡) =
𝜓(𝑡) ≥ 0, and 𝑦(0) = 𝜓(0) > 0, 𝑡 ∈ [−𝜏, 0], are uniformly
2𝜋 2𝜋 bounded when 𝜏 is bounded.
𝛾 (𝑧∗ , 𝜏𝑗 , ) = deg𝐵 (𝐻− (𝑧∗ , 𝜏𝑗 , ) , Ω𝜀,2𝜋/𝜔0 )
𝜔0 𝜔0
Proof. Suppose that 𝑥 = 𝑥(𝑡), 𝑦 = 𝑦(𝑡) are nonconstant
+ 2𝜋
∗ periodic solutions of system (7) and define
− deg𝐵 (𝐻 (𝑧 , 𝜏𝑗 , ) , Ω𝜀,2𝜋/𝜔0 )
𝜔0
𝑥 (𝜉1 ) = min {𝑥 (𝑡)} , 𝑥 (𝜂1 ) = max {𝑥 (𝑡)} ,
= −1. (104)
(101) 𝑦 (𝜉2 ) = min {𝑦 (𝑡)} , 𝑦 (𝜂2 ) = max {𝑦 (𝑡)} .
Then, it follows from system (7) that
In what follows, we define
𝑡
Σ = 𝐶𝑙 {(𝑧, 𝜏, 𝑝) ∈ 𝑋 × 𝑅+ × 𝑅+ : 𝑥 (𝑡) = 𝑥 (0) exp {∫ (𝑟1 − 𝑟1 𝑥 (𝑠 − 𝜏) + 𝑟1 𝑎𝑦 (𝑠)) d𝑠} ,
0
𝑧 is a 𝑝-periodic solution of (92)} , (102) 𝑡
𝑦 (𝑡) = 𝑦 (0) exp {∫ (𝑟2 + 𝑟2 𝑏𝑥 (𝑠) − 𝑟2 𝑦 (𝑠 − 𝜏)) d𝑠} ,
𝑁 = {(𝑧, 𝜏, 𝑝) : 𝐹 (𝑧, 𝜏, 𝑝) = 0} . 0
(105)
And let ℓ(𝑧∗ , 𝜏𝑗 , 2𝜋/𝜔0 ) denote the connected component of which implies that the solutions of system (7) cannot cross the
(𝑧∗ , 𝜏𝑗 , 2𝜋/𝜔0 ) in Σ. 𝑥-axis and 𝑦-axis. Thus, the nonconstant periodic orbits must
From the above discussion, we have be located in the interior of each quadrant. The following
results are deduced according to Theorem 4.9.1 in Kuang [15]
Σ((𝑧,𝜏,𝑝)∈ℓ(𝑧∗ ,𝜏𝑗 ,2𝜋/𝜔0 )∩𝑁) 𝛾 (𝑧, 𝜏, 𝑝) < 0, (103) and the standard comparison principle [14].
(𝑗)
If (𝑥(𝑡), 𝑦(𝑡)) is a solution of system (7) with 𝑥(𝑡) > 0,
where (𝑧, 𝜏, 𝑝), in fact, takes the form (𝑧∗ , 𝜏𝑘 , 2𝜋/𝜔𝑘 ), 𝑗 = 𝑦(𝑡) > 0, then it is easy to say that there exists a 𝑇 > 0 such
0, 1, 2, . . .. Thus, the connected component ℓ(𝑧∗ , 𝜏𝑗 , 2𝜋/𝜔0 ) that, for all 𝑡 > 𝑇, 𝑥(𝑡) < 1 + 𝑎. In addition, from the second
through (𝑧∗ , 𝜏𝑗 , 2𝜋/𝜔0 ) in Σ is nonempty. Since the first equation of (7), we can get
crossing number of each center is always −1, by Theorem 3.3
[21], we conclude that ℓ(𝑧∗ , 𝜏𝑗 , 2𝜋/𝜔0 ) is unbounded. Thus, 0 = 1 + 𝑏𝑥 (𝜂2 ) − 𝑦 (𝜂2 − 𝜏) < 1 + 𝑏 (1 + 𝑎) − 𝑦 (𝜂2 − 𝜏) ,
we have proved the following lemma. (106)
Abstract and Applied Analysis 13

𝜏2 = 1.0, 𝜏3 = 0 𝜏2 = 0, 𝜏3 = 2.0
30 30

25 25

20 20

15 15

10 10

5 5

0 0
0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400
t t

x(t) x(t)
y(t) y(t)

𝜏2 = 𝜏3 = 1.0 𝜏2 = 2, 𝜏3 = 5.0
30 30

25 25

20 20

15 15

10 10

5 5

0 0
0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400
t t

x(t) x(t)
y(t) y(t)

Figure 2: MATLAB simulation shows that the positive equilibrium of system (115) is always locally asymptotically stable with any values of
𝜏2 and 𝜏3 with initial value “1.2, 0.5.”

which leads to On the other hand, we get, from the first equation of (7),

𝑦 (𝜂2 − 𝜏) ≤ 1 + 𝑏 (1 + 𝑎) . (107) 0 = 1 − 𝑥 (𝜂1 ) + 𝑎𝑦 (𝜂1 ) ≤ 1 − 𝑥 (𝜂1 )

(111)
From the second equation of system (7), we also have + 𝑎 (1 + 𝑏 (1 + 𝑎)) exp {𝑟2 (1 + 𝑏 + 𝑎𝑏) 𝜏} .

𝑦̇(𝑡) ≤ 𝑟2 (1 + 𝑏 + 𝑎𝑏) 𝑦 (𝑡) , (108) It follows that

which implies 𝑥 (𝜂1 ) ≤ 1 + 𝑎 (1 + 𝑏 (1 + 𝑎)) exp {𝑟2 (1 + 𝑏 + 𝑎𝑏) 𝜏} ,
(112)
𝑦 (𝑡) ≤ 𝑦 (𝑡 − 𝜏) exp {𝑟2 (1 + 𝑏 + 𝑎𝑏) 𝜏} . (109) for 𝑡 > 𝑇 + 𝜏.

It follows from (104) and (107) that This shows that the nonconstant periodic solution of
system (7) is uniformly bounded when 𝜏 is bounded. This
𝑦 (𝜂2 ) ≤ (1 + 𝑏 (1 + 𝑎)) exp {𝑟2 (1 + 𝑏 + 𝑎𝑏) 𝜏} . (110) completes the proof.
14 Abstract and Applied Analysis

𝜏1 = 0.331 < 𝜏10 = 0.332, 𝜏4 = 0 𝜏1 = 0.333 > 𝜏10 = 0.332, 𝜏4 = 0

35 30

30 25

25
20
20
15
15
10
10

5 5

0 0
0 50 100 150 200 0 50 100 150 200
t t

x(t) x(t)
y(t) y(t)

Figure 3: MATLAB simulation shows that system (115) undergoes a Hopf bifurcation at 𝐸∗ for 𝜏1 = 0.333 > 𝜏10 = 0.332 with 𝜏4 = 0 and
initial value “1.2, 0.5.”

Lemma 19. System (7) has no nonconstant periodic solutions From the discussion in Section 2, we have
with period 𝜏 when the condition (H1) is satisfied. (𝑟 𝑥∗ + 𝑟2 𝑦∗ ) √1 − 𝑎𝑏
1 1
𝜏𝑗 = arcsin 1 + 2𝑗𝜋,
Proof. Let’s suppose that there exist a contradiction in system 𝜔+ 2√𝑟1 𝑟2 𝑥∗ 𝑦∗ 𝜔+ (114)
(7); that is, system (7) has nonconstant periodic solution = 0, 1, 2, . . . .
with period. Then, the following system (113) of ordinary
differential equations has nonconstant periodic solution: Thus, one can get 2𝜋/𝑤0 < 𝜏𝑗 , for 𝑗 > 0. From Lemma 19, we
know that system (7) with 𝜏 = 0 has no nontrivial periodic
𝑢̇ solution. Consequently, the projection of ℓ(𝑧∗ , 𝜏𝑗 , 2𝜋/𝑤0 )
1 (𝑡) = 𝑟1 𝑢1 (𝑡) [1 − 𝑢1 (𝑡 − 𝜏) + 𝑎𝑢2 (𝑡)] ,
(113) onto 𝜏-space is away from zero.
𝑢̇
2 (𝑡) = 𝑟2 𝑢2 (𝑡) [1 + 𝑏𝑢1 (𝑡) − 𝑢2 (𝑡 − 𝜏)] ,
Suppose that the projection of ℓ(𝑧∗ , 𝜏𝑗 , 2𝜋/𝑤0 ) onto
𝜏-space is bounded, and then there exits 𝜏∗ > 0 such that
which has the same equilibria as system (7), that is, a unique the projection of ℓ(𝑧∗ , 𝜏𝑗 , 2𝜋/𝑤0 ) onto 𝜏-space in interval
positive equilibrium 𝑢∗ (𝑢1∗ , 𝑢2∗ ). Note that 𝑢1 -axis, and 𝑢2 - (0, 𝜏0 ). Since 2𝜋/𝑤0 < 𝜏𝑗 and applying Lemma 18, one can
axis are the invariable manifold of system (113) and the orbits obtain 0 < 𝑝 < 𝜏∗ for (𝑧(𝑡), 𝜏, 𝑝) ∈ ℓ(𝑧∗ , 𝜏𝑗 , 2𝜋/𝑤0 ).
of system (113) do not intersect with each other. Thus, there is Therefore, the projection of ℓ(𝑧∗ , 𝜏𝑗 , 2𝜋/𝑤0 ) onto 𝜏-space is
no solution crosses the coordinate axes. On the other hand, also bounded. Thus, we get, together with Lemma 18, that
note the fact that if system (113) has a periodic solution, then the connected component ℓ(𝑧∗ , 𝜏𝑗 , 2𝜋/𝑤0 ) is bounded. This
there must be equilibrium in its interior. Thus, we conclude contradicts Lemma 17. The proof is completed.
that the periodic orbit of system (113) must lie in the first
quadrant. It is well known that the positive equilibrium 𝐸∗
is global asymptotically stable to the first quadrant (see, e.g., 5. Numerical Example
[23, 24]). Thus, there is not periodic orbit in the first quadrant, We will give some numerical results of system (7) to support
too. This ends the proof. the analytic results in this section. We consider the following
system:
In the following, we state and prove our main result in this
section. 𝑥̇(𝑡) = 𝑥 (𝑡) [0.8 − 0.4𝑥 (𝑡 − 𝜏1 ) + 0.2𝑦 (𝑡 − 𝜏2 )] ,
(115)
𝑦̇(𝑡) = 𝑦 (𝑡) [0.75 + 0.4𝑥 (𝑡 − 𝜏3 ) − 0.25𝑦 (𝑡 − 𝜏4 )] ,
Theorem 20. Suppose that the condition (H1) holds. Then, for
each 𝜏 > 𝜏𝑗 , 𝑗 = 1, 2, . . ., system (7) has at least 𝑗 − 1 periodic and get the positive equilibrium 𝐸∗ = (17.50, 31.00). System
solutions. (115) without any time delay is locally asymptotically stable
(see Figure 1). Firstly, when only consider delays 𝜏2 and 𝜏3 ,
Proof. It is sufficient to prove that the connected component and we find that Hopf bifurcation do not occur at the positive
ℓ(𝑧∗ , 𝜏𝑗 , 2𝜋/𝑤0 ) onto 𝜏-space is [𝜏, +∞) for each 𝑗 ≥ 1, where equilibrium 𝐸∗ of system (115) with any value of them (see
𝜏 ≤ 𝜏𝑗 . Figure 2).
Abstract and Applied Analysis 15

𝜏1 = 0, 𝜏4 = 0.292 < 𝜏40 = 0.293 𝜏1 = 0, 𝜏4 = 0.294 > 𝜏40 = 0.293

30 35

25 30

25
20
20
15
15
10
10

5 5

0 0
0 50 100 150 200 0 50 100 150 200
t t

x(t) x(t)
y(t) y(t)

𝜏1 = 𝜏4 = 0.179 < 𝜏0 = 0.1795 𝜏1 = 𝜏4 = 0.180 > 𝜏0 = 0.1795

40 45

35 40

35
30
30
25
25
20
20
15
15
10
10
5 5

0 0
0 50 100 150 200 0 50 100 150 200
t t

x(t) x(t)
y(t) y(t)

Figure 4: MATLAB simulation shows that system (115) undergoes a Hopf bifurcation at 𝐸∗ once 𝜏4 > 𝜏40 = 0.293 and 𝜏1 = 𝜏4 > 𝜏01 = 0.1795,
respectively, with initial value “0.4, 0.5.”

Secondly, we have that the critical value of delay is 𝜏10 = Figure 6). Similarly, when we regard 𝜏3 as a parameter and
0.332 when we only consider system (115) with delays 𝜏1 and let 𝜏1 = 0.20 ∈ (0, 0.333), we have that 𝜏3∗ = 0.6226.
𝜏4 . From Theorem 3, we obtain the corresponding waveform From Theorem 11, system (115) is locally asymptotically stable
shown in Figure 3. Similarly, we have that 𝜏40 = 0.293 in for 𝜏3 ∈ (0, 𝜏3∗ ) and unstable for 𝜏3 > 𝜏3∗ (see Figure 7).
Case 2B and 𝜏01 = 0.1795 in Case 2C, respectively. The Though, we only give the associated characteristic equation
corresponding waveforms are shown in Figure 4. Regarding (58) when only we consider delays 𝜏2 and 𝜏4 of system (115)
𝜏4 as a parameter with 𝜏1 = 0.20 ∈ (0, 0.333), we can obtain with 𝜏1 = 𝜏3 = 0, we give the phase plots when the delay
𝜏4∗ = 0.1602. From Theorem 8, the positive equilibrium 𝐸∗ 𝜏4 increases from zero to its critical value and the value of 𝜏2
is locally asymptotically stable for 𝜏4 ∈ (0, 𝜏4∗ ) and unstable belongs to its stable interval (see Figure 8).
for 𝜏4 > 𝜏4∗ (see Figure 5). Finally, by algorithm (91) derived in Section 3, we
Thirdly, we can obtain 𝜏03 = 1.218 when only consider have 𝐶1 (0) = −9.9831 + 3.294𝑖, 𝜇2 = 1591.2 > 0,
𝜏1 = 𝜏3 = 𝜏 ≠ 0. That is, when 𝜏 increases from and 𝛽2 = −17.691 < 0. Thus, the Hopf bifurcation
zero to the critical value 𝜏03 , the positive equilibrium 𝐸∗ is supercritical, and the bifurcating periodic solutions
is locally asymptotically stable and then it loses its stability are asymptotically stable, which is illustrated in
and a Hopf bifurcation occurs once 𝜏 > 𝜏03 = 1.218 (see Figure 9.
16 Abstract and Applied Analysis

𝜏1 = 0.2, 𝜏4 = 0.1502 < 𝜏4∗ = 0.1602 𝜏1 = 0.2, 𝜏4 = 0.1612 > 𝜏4∗ = 0.1602
40 45

35 40

35
30
30
25
25
20
20
15
15
10
10
5 5

0 0
0 100 200 300 400 500 600 0 100 200 300 400 500 600
t t

x(t) x(t)
y(t) y(t)

Figure 5: MATLAB simulation shows that the bifurcating periodic solutions from 𝐸∗ occur for 𝜏4 = 0.1612 > 𝜏4∗ = 0.1602 with 𝜏1 = 0.2 and
initial value “0.4, 0.5.”

𝜏1 = 𝜏3 = 1.118 < 𝜏0 = 1.218 𝜏1 = 𝜏3 = 1.318 > 𝜏0 = 1.218

30 45

40
25
35

20 30

25
15
20

10 15

10
5
5

0 0
0 50 100 150 200 0 50 100 150 200
t t

x(t) x(t)
y(t) y(t)

Figure 6: System (115) at 𝐸∗ undergoes a Hopf bifurcation for 𝜏1 = 𝜏3 > 𝜏0 = 1.218 with initial value “0.4, 0.5.”

6. Discussion However, we find that the feedback delays for two species play
almost the same role. The critical value of feedback delay 𝜏1
He and Gopalsamy [17] considered a Lotka-Volterra mutual- is 0.332 in the system (115) when we only consider it with
istic system (4) with delay, in which all delays take the same 𝜏4 = 0 while the critical value of feedback delay 𝜏4 is 0.293
value; Meng and Wei only considered a mutualistic system when we only consider it. We also find that feedback delay
(5) with the same value of feedback delay for each population could destroy the stability of solutions of system (115). That is,
in [18] while we not only introduce feedback delays of two the solutions of system (115) turn to oscillate as 𝜏4 increasing
populations to grow of the species in system (6) but also take monotonously from zero. In addition, the solutions of system
time delay of each species benefit into account. By theory (115) are stable at first, then oscillate, at last turn to be periodic
analysis and simulations, we find that time delays of the when time delay of benefit 𝜏3 is greater than its critical value
benefit 𝜏2 and 𝜏3 can not affect the stability of the system (6) 𝜏3∗ and feedback delay 𝜏1 belongs to its stable interval (see the
when we only choose different values of 𝜏2 , 𝜏3 with 𝜏1 = 𝜏4 = right figure of Figure 7), which is different from the result of
0. That is, two mutualistic species could coexist with delay. the case of considering feedback time delay 𝜏4 greater than
Abstract and Applied Analysis 17

𝜏1 = 0.2, 𝜏3 = 0.6225 < 𝜏3∗ = 0.6226 𝜏1 = 0.2, 𝜏3 = 0.6228 > 𝜏3∗ = 0.6226
30 50
45
25
40
35
20
30
15 25
20
10
15
10
5
5
0 0
0 50 100 150 200 0 50 100 150 200
t t

x(t) x(t)
y(t) y(t)

Figure 7: MATLAB simulation shows that the positive equilibrium of system (115) is locally asymptotically stable for 𝜏3 = 0.6225 < 𝜏3∗ =
0.6226 and unstable for 𝜏3 = 0.6228 > 𝜏3∗ = 0.6226 with 𝜏1 = 0.2 and initial value “4, 5.”

Phase plot (𝜏2 = 0.3, 𝜏4 = 0.53 < 𝜏4∗ = 0.54) Phase plot (𝜏2 = 0.3, 𝜏4 = 0.55 > 𝜏4∗ = 0.54)
30 40

35
25
30

20
25
y(t)

y(t)

20
15

15
10
10

5 5
2 4 6 8 10 12 14 2 4 6 8 10 12 14 16 18
x(t) x(t)

Figure 8: MATLAB simulation shows that system (115) at 𝐸∗ undergoes a Hopf bifurcation for 𝜏4 = 0.55 > 𝜏4∗ = 0.54 with 𝜏2 = 0.3 and
initial value “4, 5.”

its critical value 𝜏4∗ and feedback delay 𝜏1 belonging to its CPUE (catch per unit effort) hypothesis [25]. Analyzing the
stable interval (see the right figure of Figure 5). This is maybe complicated bifurcations, we leave this work in the future.
because time delay of benefit could make the two species
stable. This is very valuable in the view of ecology.
It is definitely an interesting future work to consider the Conflict of Interests
following mutual system with different harvesting: The authors declare that there is no conflict of interests
regarding the publication of this paper.
𝑥̇(𝑡) = 𝑥 (𝑡) [𝑏1 − 𝑎11 𝑥 (𝑡 − 𝜏1 ) + 𝑎12 𝑦 (𝑡 − 𝜏2 )] − 𝑞1 𝐸1 𝑥 (𝑡) ,

𝑦̇(𝑡) = 𝑦 (𝑡) [𝑏2 + 𝑎21 𝑥 (𝑡 − 𝜏3 ) − 𝑎22 𝑦 (𝑡 − 𝜏4 )] − 𝑞2 𝐸2 𝑦 (𝑡) , Acknowledgments

(116)
This work was partially supported by the NNSF of
where 𝑞1 , 𝑞2 (>0) are the catchability coefficient of two species. China (10961018), the NSF of Gansu Province of China
The catch rate functions 𝑞1 𝐸1 𝑥 and 𝑞2 𝐸2 𝑦 are based on (1212RJYA011, 2011GS04318), the NSF for Distinguished
18 Abstract and Applied Analysis

Phase plot (𝜏1 = 𝜏4 = 0.180 > 𝜏0 = 0.1795) [10] C. Q. Xu and S. L. Yuan, “Stability and Hopf bifurcation in a
45
delayed predator-prey system with herd behavior,” Abstract and
Applied Analysis, vol. 2014, Article ID 568943, 8 pages, 2014.
40
[11] X. Y. Meng, H. F. Huo, X. B. Zhang, and H. Xiang, “Stability
and Hopf bifurcation in a three-species system with feedback
35 delays,” Nonlinear Dynamics, vol. 64, no. 4, pp. 349–364, 2011.
[12] Y. L. Song and J. J. Wei, “Local Hopf bifurcation and global
30 periodic solutions in a delayed predator-prey system,” Journal
y(t)

of Mathematical Analysis and Applications, vol. 301, no. 1, pp. 1–

25 21, 2005.
[13] L. S. Chen, Z. Y. Lu, and W. D. Wang, “The effect of delays on the
20 permanence for Lotka-Volterra systems,” Applied Mathematics
Letters, vol. 8, no. 4, pp. 71–73, 1995.
15 [14] J. K. Hale, Theory of Functional Differential Equations, Springer,
New York, NY, USA, 2nd edition, 1977.
10 [15] Y. Kuang, Delay Differential Equations with Applications in
6 8 10 12 14 16 18 20
Population Dynamics, Academic Press, New York, NY, USA,
x(t)
1993.
Figure 9: This figure shows that the Hopf bifurcation is supercritical, [16] J. D. Gause, Mathematical Biology, Springer, New York, NY,
and the bifurcating periodic solutions are asymptotically stable. USA, 1989.
[17] X. Z. He and K. Gopalsamy, “Persistence, attractivity, and delay
in facultative mutualism,” Journal of Mathematical Analysis and
Applications, vol. 215, no. 1, pp. 154–173, 1997.
Young Scholars of Gansu Province of China (1111RJDA003),
[18] X. Z. Meng and J. J. Wei, “Stability and bifurcation of mutual
the Development Program for HongLiu Distinguished
system with time delay,” Chaos, Solitons & Fractals, vol. 21, no.
Young Scholars in Lanzhou University of Technology, and 3, pp. 729–740, 2004.
the Development Program for Outstanding Young Teachers
[19] F. Q. Zhang and Y. J. Zhang, “Hopf bifrucation for Lotka-
in Lanzhou University of Technology (Q201308). Volterra mutualistic systems with three delays,” Jouranl of
Biomathematics, vol. 26, no. 2, pp. 223–233, 2011.
References [20] B. D. Hassard, N. D. Kazarinoff, and Y. H. Wan, Theory and
Applications of Hopf Bifurcation, Cambridge University Press,
[1] A. J. Lotka, Elements of Physical Biology, Williams and Wilkins, Cambridge, UK, 1981.
New York, NY, USA, 1925. [21] J. H. Wu, “Symmetric functional-differential equations and
[2] V. Volterra, “Fluctuations in the abundance of a species con- neural networks with memory,” Transactions of the American
sidered mathematically,” Nature, vol. 118, no. 2972, pp. 558–560, Mathematical Society, vol. 350, no. 12, pp. 4799–4838, 1998.
1926. [22] S. G. Ruan and J. J. Wei, “On the zeros of transcendental
[3] Z. Jin and Z. E. Ma, “Stability for a competitive Lotka-Volterra functions with applications to stability of delay differential
system with delays,” Nonlinear Analysis: Theory, Methods and equations with two delays,” Dynamics of Continuous, Discrete
Applications, vol. 51, no. 7, pp. 1131–1142, 2002. & Impulsive Systems A: Mathematical Analysis, vol. 10, no. 6, pp.
[4] Y. Song, M. Han, and Y. Peng, “Stability and Hopf bifurcations 863–874, 2003.
in a competitive Lotka-Volterra system with two delays,” Chaos, [23] J. Cushing, “Periodic time-dependent predator-prey systems,”
Solitons & Fractals, vol. 22, no. 5, pp. 1139–1148, 2004. SIAM Journal on Applied Mathematics, vol. 32, no. 1, pp. 82–95,
1977.
[5] J. Z. Zhang, Z. Jin, J. R. Yan, and G. Q. Sun, “Stability and
Hopf bifurcation in a delayed competition system,” Nonlinear [24] X. Z. He, “Stability and delays in a predator-prey system,”
Analysis: Theory, Methods & Applications, vol. 70, no. 2, pp. 658– Journal of Mathematical Analysis and Applications, vol. 198, no.
670, 2009. 2, pp. 355–370, 1996.
[6] J. F. Zhang, “Stability and bifurcation periodic solutions in [25] C. W. Clark, Bioeconomic Modelling and Fisheries Management,
a Lotka-Volterra competition system with multiple delays,” John Wiley & Sons, New York, NY, USA, 1985.
Nonlinear Dynamics. An International Journal of Nonlinear
Dynamics and Chaos in Engineering Systems, vol. 70, no. 1, pp.
849–860, 2012.
[7] T. Faria, “Stability and bifurcation for a delayed predator-prey
model and the effect of diffusion,” Journal of Mathematical
Analysis and Applications, vol. 254, no. 2, pp. 433–463, 2001.
[8] Y. L. Song, Y. H. Peng, and J. J. Wei, “Bifurcations for a predator-
prey system with two delays,” Journal of Mathematical Analysis
and Applications, vol. 337, no. 1, pp. 466–479, 2008.
[9] X. P. Yan and W. T. Li, “Hopf bifurcation and global periodic
solutions in a delayed predator-prey system,” Applied Mathe-
matics and Computation, vol. 177, no. 1, pp. 427–445, 2006.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 727818, 10 pages
http://dx.doi.org/10.1155/2014/727818

Research Article
Stability and Bifurcation Analysis on an Ecoepidemiological
Model with Stage Structure and Time Delay

Lingshu Wang1 and Guanghui Feng2

1
School of Mathematics and Statistics, Hebei University of Economics & Business, Shijiazhuang 050061, China
2
Institute of Applied Mathematics, Shijiazhuang Mechanical Engineering College, Shijiazhuang 050003, China

Correspondence should be addressed to Lingshu Wang; wanglingshu@126.com

Received 15 April 2014; Accepted 21 July 2014; Published 12 August 2014

Academic Editor: Sanling Yuan

Copyright © 2014 L. Wang and G. Feng. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

An ecoepidemiological predator-prey model with stage structure for the predator and time delay due to the gestation of the predator
is investigated. The effects of a prey refuge with disease in the prey population are concerned. By analyzing the corresponding
characteristic equations, the local stability of each of the feasible equilibria of the model is discussed. Further, it is proved that
the model undergoes a Hopf bifurcation at the positive equilibrium. By means of appropriate Lyapunov functions and LaSalle’s
invariance principle, sufficient conditions are obtained for the global stability of the semitrivial boundary equilibria. By using an
iteration technique, sufficient conditions are derived for the global attractiveness of the positive equilibrium.

1. Introduction model incorporating a prey refuge with disease in the prey

population
In the natural world, species does not exist alone. While
species spreads the disease, it also competes with the other 𝑑𝑆 𝑆 (𝑡) + 𝐼 (𝑡)
= 𝑟𝑆 (𝑡) (1 − ) − 𝛽𝑆 (𝑡) 𝐼 (𝑡)
species for space or food, or it is predated by other species. 𝑑𝑡 𝐾
The construction and study of models for the population − 𝑏1 (1 − 𝑚) 𝑆 (𝑡) 𝑌 (𝑡) ,
dynamics of predator-prey systems have been an important
topic in theoretical ecology. Following Anderso and May [1], 𝑑𝐼
who were the first to propose an ecoepidemiological model = 𝛽𝑆 (𝑡) 𝐼 (𝑡) − 𝑏2 (1 − 𝑚) 𝐼 (𝑡) 𝑌 (𝑡) − 𝑑𝐼 (𝑡) , (1)
𝑑𝑡
by merging the ecological predator-prey model introduced by
Lotka and Volterra, the effect of disease in ecological system 𝑑𝑌
= 𝑝𝑏1 (1 − 𝑚) 𝑆 (𝑡) 𝑌 (𝑡)
is an important issue from mathematical and ecological point 𝑑𝑡
of view. Ecoepidemiology which is a relatively new branch of
+ 𝑝𝑏2 (1 − 𝑚) 𝐼 (𝑡) 𝑌 (𝑡) − 𝑐𝑌 (𝑡) ,
study in theoretical biology tackles such situations by dealing
with both ecological and epidemiological issues. where 𝑆(𝑡) and 𝐼(𝑡) represent the densities of susceptible
The research of the hiding behaviour of preys has been and infected prey population at time 𝑡, respectively, and 𝑌(𝑡)
incorporated as a new ingredient of predator-prey models. represents the density of the predator population at time 𝑡.
In nature, prey populations often have access to areas where The parameters 𝑟, 𝐾, 𝛽, 𝑏1 , 𝑏2 , 𝑐, 𝑑, and 𝑝 are positive constants
they are safe from their predators. Such refugia are usually in which 𝑟 and 𝐾 represent the prey intrinsic growth rate and
playing two significant roles, serving both to reduce the the carrying capacity, respectively. 𝛽 is the transmission rate
chance of extinction due to predation and to damp predator- of the susceptible prey into the infected prey. 𝑏1 and 𝑏2 are
prey oscillations. It is well known that many more attentions the capturing rates of the susceptible prey and the infected
have been paid on the effects of a prey refuge for predator- prey, respectively. 𝑝 describes the efficiency of the predator
prey model. In [2], Wang considered an ecoepidemiological in converting consumed prey into predator offspring. The
2 Abstract and Applied Analysis

constant proportion infected prey refuge is (1 − 𝑚)𝐼, where It is well known by the fundamental theory of functional
𝑚 ∈ [0, 1) is a constant. By means of appropriate Lyapunov differential equations [8] that model (2) has a unique solution
functions and limit theory, sufficient conditions are obtained (𝑆(𝑡), 𝐼(𝑡), 𝑌1 (𝑡), 𝑌2 (𝑡)) satisfying initial conditions (3).
for the global stability of the semitrivial boundary equilibria The organization of this paper is as follows. In the next
of model (1). section, we show the positivity and the boundedness of
We note that it is assumed in system (1) that each solutions of model (2) with initial conditions (3). In Section 3,
individual predator admits the same ability to feed on prey. we investigate the stability of the semitrivial equilibria of
This assumption seems to be not realistic for many animals. In the model (2). In Section 4, we discuss the stability of the
the natural world, there are many species whose individuals positive equilibrium of the model (2). Further, we study the
pass through an immature stage during which they are raised existence of Hopf bifurcation at the positive equilibrium. A
by their parents, and the rate at which they attack prey can be brief discussion is given in Section 5 to conclude this work.
ignored. Moreover, it can be assumed that their reproductive
rate during this stage is zero. Stage structure is a natural
2. Preliminaries
phenomenon and represents, for example, the division of
a population into immature and mature individuals. Stage- In this section, we show the positivity and the boundedness
structured models have received great attention in recent of solutions of model (2) with initial conditions (3).
years (see, e.g., [3–5]).
Time delays of one type or another have been incor- Theorem 1. Solutions of model (2) with initial conditions (3)
porated into biological models by many researchers (see, are positive for all 𝑡 ≥ 0.
e.g., [5–7]). In general, delay differential equations exhibit
much more complicated dynamics than ordinary differential Proof. Let (𝑆(𝑡), 𝐼(𝑡), 𝑌1 (𝑡), 𝑌2 (𝑡)) be a solution of model (2)
equations since a time delay could cause the population to with initial conditions (3). It follows from the first and the
fluctuate. Time delay due to gestation is a common example, second equations of model (2) that
because generally the consumption of prey by the predator
𝑡
throughout its past history governs the present birth rate of 𝑟 𝑟
𝑆 (𝑡) = 𝑆 (0) exp {∫ [𝑟 − 𝑆 (𝑠) − ( + 𝛽) 𝐼 (𝑠)] 𝑑𝑠} > 0,
the predator. Therefore, more realistic models of population 0 𝐾 𝐾
interactions should take into account the effect of time delays. 𝑡
Based on the above discussions, in this paper, we incor- 𝐼 (𝑡) = 𝐼 (0) exp {∫ [𝛽𝑆 (𝑠) − 𝑑 − 𝑏 (1 − 𝑚) 𝑌2 (𝑠)] 𝑑𝑠} > 0.
porate a stage structure for the predator and time delay due 0
to the gestation of predator into the model (1). To this end, we (4)
study the following differential equations:
Let us consider 𝑌1 (𝑡) and 𝑌2 (𝑡) for 𝑡 ∈ [0, 𝜏]. Since 𝜙2 (𝜃) ≥
𝑑𝑆 𝑆 (𝑡) + 𝐼 (𝑡) 0, 𝜑2 (𝜃) ≥ 0, for 𝜃 ∈ [−𝜏, 0], we derive from the third equation
= 𝑟𝑆 (𝑡) (1 − ) − 𝛽𝑆 (𝑡) 𝐼 (𝑡) ,
𝑑𝑡 𝐾 of model (2) that
𝑑𝐼
= 𝛽𝑆 (𝑡) 𝐼 (𝑡) − 𝑑𝐼 (𝑡) − 𝑏 (1 − 𝑚) 𝐼 (𝑡) 𝑌2 (𝑡) , 𝑑𝑌1
𝑑𝑡 ≥ − (𝑟1 + 𝑑1 ) 𝑌1 (𝑡) . (5)
(2) 𝑑𝑡
𝑑𝑌1
= 𝑝𝑏 (1 − 𝑚) 𝐼 (𝑡 − 𝜏) 𝑌2 (𝑡 − 𝜏) − (𝑟1 + 𝑑1 ) 𝑌1 (𝑡) , Since 𝜑1 (0) > 0, a standard comparison argument shows that
𝑑𝑡
𝑑𝑌2 𝑌1 (𝑡) ≥ 𝑌1 (0) 𝑒−(𝑟1 +𝑑1 )𝑡 > 0; (6)
= 𝑟1 𝑌1 (𝑡) − 𝑑2 𝑌2 (𝑡) − 𝑎𝑌22 (𝑡) ,
𝑑𝑡
where 𝑌1 (𝑡) and 𝑌2 (𝑡) represent the densities of the immature that is, 𝑌1 (𝑡) > 0, for 𝑡 ∈ [0, 𝜏]. For 𝑡 ∈ [0, 𝜏], it follows from
and the mature predator population at time 𝑡, respectively. the fourth equation of (2) that
The parameters 𝑑1 , 𝑑2 , and 𝑟1 are positive constants in which
𝑑𝑌2
𝑑1 and 𝑑2 are the death rates of the immature and the ≥ −𝑑2 𝑌2 (𝑡) − 𝑎𝑌22 (𝑡) . (7)
mature predator, respectively. 𝑟1 denotes the rate of immature 𝑑𝑡
predator becoming mature predator. 𝜏 ≥ 0 is a constant delay Since 𝜑2 (0) > 0, a standard comparison argument shows that
due to the gestation of the predator.
The initial conditions for system (2) take the form 𝑡
𝑌2 (𝑡) ≥ 𝑌2 (0) exp {∫ (−𝑑2 − 𝑎𝑌2 (𝑠)) 𝑑𝑠} > 0; (8)
𝑆 (𝜃) = 𝜙1 (𝜃) ≥ 0, 𝐼 (𝜃) = 𝜙2 (𝜃) ≥ 0, 0

𝑌1 (𝜃) = 𝜑1 (𝜃) ≥ 0, 𝑌2 (𝜃) = 𝜑2 (𝜃) ≥ 0, that is, 𝑌2 (𝑡) > 0, for 𝑡 ∈ [0, 𝜏]. In a similar way, we treat
𝜃 ∈ [−𝜏, 0) , 𝜙1 (0) > 0, 𝜙2 (0) > 0, (3) the intervals [𝜏, 2𝜏], . . . , [𝑛𝜏, (𝑛 + 1)𝜏], 𝑛 ∈ 𝑁. Thus, 𝑆(𝑡) >
0, 𝐼(𝑡) > 0, 𝑌1 (𝑡) > 0, and 𝑌2 (𝑡) > 0 for all 𝑡 ≥ 0. This
𝜑1 (0) > 0, 𝜑2 (0) > 0,
completes the proof.
4
(𝜙1 (𝜃) , 𝜙2 (𝜃) , 𝜑1 (𝜃) , 𝜑2 (𝜃)) ∈ 𝐶 ([−𝜏, 0] , 𝑅+0 ),
Theorem 2. Positive solutions of model (2) with initial condi-
4
where 𝑅+0 = {(𝑥1 , 𝑥2 , 𝑥3 , 𝑥4 ) : 𝑥𝑖 ≥ 0, 𝑖 = 1, 2, 3, 4}. tions (3) are ultimately bounded.
Abstract and Applied Analysis 3

Proof. Let (𝑆(𝑡), 𝐼(𝑡), 𝑌1 (𝑡), 𝑌2 (𝑡)) be a positive solution Clearly (14) has a positive real root. Accordingly, the equilib-
of model (2) with initial conditions (3). Denote 𝑑̂ = rium 𝐸0 is unstable.
min{𝑑, 𝑑1 , 𝑑2 }. Define The characteristic equation of model (2) at the equilib-
rium 𝐸𝐾 (𝐾, 0, 0, 0) takes the form
𝑊 (𝑡) = 𝑝𝑆 (𝑡 − 𝜏) + 𝑝𝐼 (𝑡 − 𝜏) + 𝑌1 (𝑡) + 𝑌2 (𝑡) . (9)
(𝜆 + 𝑟) (𝜆 + 𝑑2 ) (𝜆 + 𝑟1 + 𝑑1 ) [𝜆 − (𝐾𝛽 − 𝑑)] = 0. (15)
Calculating the derivative of 𝑊(𝑡) along the positive solutions
of (2), it follows that Hence, if 𝐾𝛽 < 𝑑, (15) has no positive real root. Accordingly,
𝑑𝑊 𝑟 𝑟 the equilibrium 𝐸𝐾 is locally asymptotically stable. If 𝐾𝛽 > 𝑑,
= 𝑝𝑟𝑆 (𝑡 − 𝜏) − 𝑝 𝑆2 (𝑡 − 𝜏) − 𝑝 𝑆 (𝑡 − 𝜏) 𝐼 (𝑡 − 𝜏) (15) has a positive real root. Accordingly, the equilibrium 𝐸𝐾
𝑑𝑡 𝐾 𝐾
is unstable.
− 𝑝𝑑𝐼 (𝑡 − 𝜏) − 𝑑1 𝑌1 (𝑡) − 𝑑2 𝑌2 (𝑡) − 𝑎𝑌22 (𝑡)
Theorem 3. If 𝐾𝛽 < 𝑑, then the semitrivial equilibrium 𝐸𝐾 is
̂ (𝑡) + 𝑝 (𝑟 + 𝑑)
≤ −𝑑𝑊 ̂ 𝑆 (𝑡 − 𝜏) − 𝑝 𝑟 𝑆2 (𝑡 − 𝜏) globally stable.
𝐾
𝑟 Proof. Based on the above discussions, we only prove the
− 𝑝 𝑆 (𝑡 − 𝜏) 𝐼 (𝑡 − 𝜏) global attractivity of the equilibrium 𝐸𝐾 . Let
𝐾
̂
𝐾 (𝑟 + 𝑑)
2 𝑆 (𝑡) 1
̂ (𝑡) − 𝑝 𝑟 [𝑆 (𝑡 − 𝜏) − 𝑉𝐾 (𝑡) = 𝑐1 [𝑆 (𝑡) − 𝐾 − 𝐾 ln ] + 𝐼 (𝑡) + 𝑌1 (𝑡)
≤ −𝑑𝑊 ] 𝐾 𝑝
𝐾 2𝑟 (16)
𝑡
1
̂ 2 + 𝑌2 (𝑡) + 𝑏 (1 − 𝑚) ∫ 𝐼 (𝑠) 𝑌2 (𝑠) 𝑑𝑠,
𝑝𝐾(𝑟 + 𝑑) 𝑝 𝑡−𝜏
+
4𝑟
where 𝑐1 = 𝐾𝛽/(𝐾𝛽 + 𝑟). Calculating the derivative of 𝑉𝐾 (𝑡)
̂2
𝑝𝐾(𝑟 + 𝑑) along the positive solutions of model (2), it follows that
̂ (𝑡) +
≤ −𝑑𝑊 ,
4𝑟 𝑑 𝑆 (𝑡) − 𝐾 ̇ 1 1
(10) 𝑉𝐾 (𝑡) = 𝑐1 𝑆 (𝑡) + 𝐼 ̇(𝑡) + 𝑌1̇ (𝑡) + 𝑌2̇ (𝑡)
𝑑𝑡 𝑆 (𝑡) 𝑝 𝑝
which yields
+ 𝑏 (1 − 𝑚) 𝐼 (𝑡) 𝑌2 (𝑡)
̂2
𝑝𝐾(𝑟 + 𝑑)
(11) − 𝑏 (1 − 𝑚) 𝐼 (𝑡 − 𝜏) 𝑌2 (𝑡 − 𝜏) .
lim sup 𝑊 (𝑡) ≤ .
𝑡→∞ 4𝑟𝑑̂ 𝛽𝑟 1
2 2 = − [𝑆 (𝑡) − 𝐾]2 −(𝑑−𝐾𝛽) 𝐼 (𝑡)− 𝑑1 𝑌1 (𝑡)
̂ /4𝑟𝑑,
If we choose 𝑀1 = 𝐾(𝑟 + 𝑑) ̂ 𝑀 = 𝑝𝐾(𝑟 + 𝑑)
̂ /4𝑟𝑑,
̂ 𝐾𝛽 + 𝑟 𝑝
2
then 1 𝑎
− 𝑑2 𝑌2 (𝑡) − 𝑌22 (𝑡) .
lim sup 𝑆 (𝑡) ≤ 𝑀1 , lim sup 𝐼 (𝑡) ≤ 𝑀1 , 𝑝 𝑝
𝑡→∞ 𝑡→∞
(12) (17)
lim sup 𝑌𝑖 (𝑡) ≤ 𝑀2 , (𝑖 = 1, 2) .
𝑡→∞ If 𝐾𝛽 < 𝑑, then it follows from (17) that 𝑉𝐾 ̇ (𝑡) ≤ 0. By
This completes the proof. Theorem 5.3.1, in [8], solutions are limited to 𝑀, the largest
invariant subset of {𝑉𝐾̇ (𝑡) = 0}. Clearly, we see from (17)
̇
that 𝑉𝐾 (𝑡) = 0 if and only if 𝑆(𝑡) = 𝐾, 𝐼(𝑡) = 0, 𝑌1 (𝑡) =
3. Boundary Equilibria and Their Stability 0, 𝑌2 (𝑡) = 0. Accordingly, the global asymptotic stability of
𝐸𝐾 follows from LaSalle’s invariant principle. This completes
In this section, we discuss the stability of the boundary
the proof.
equilibria of model (2).
Model (2) always has two boundary equilibria, namely, The characteristic equation of model (2) at the equilib-
the trivial equilibrium 𝐸0 (0, 0, 0, 0) and the axial equilibrium rium 𝐸1 is of the form
𝐸𝐾 (𝐾, 0, 0, 0). It is easy to show that if 𝐾𝛽 > 𝑑, model
(2) admits a predator-extinction equilibrium 𝐸1 (𝑆1 , 𝐼1 , 0, 0), 𝑟 𝛽 (𝐾𝛽 + 𝑟)
where (𝜆2 + 𝑆1 𝜆 + 𝑆1 𝐼1 )
𝐾 𝐾 (18)
𝑑 𝑟 (𝐾𝛽 − 𝑑)
𝑆1 = , 𝐼1 = . (13) 2
× (𝜆 + 𝑔1 𝜆 + 𝑔0 + 𝑓0 𝑒 −𝜆𝜏
) = 0,
𝛽 𝛽 (𝐾𝛽 + 𝑟)
The characteristic equation of model (2) at the equilib- where 𝑔1 = 𝑟1 + 𝑑1 + 𝑑2 , 𝑔0 = 𝑑2 (𝑟1 + 𝑑1 ), 𝑓0 = −𝑝𝑏𝑟1 (1 −
rium 𝐸0 (0, 0, 0, 0) is of the form 𝑚)𝐼1 . Clearly, the roots of equation 𝜆2 + (𝑟/𝐾)𝑆1 𝜆 + (𝛽(𝐾𝛽 +
𝑟)/𝐾)𝑆1 𝐼1 = 0 have negative real part. When 𝜏 = 0, if
(𝜆 − 𝑟) (𝜆 + 𝑑) (𝜆 + 𝑟1 + 𝑑1 ) (𝜆 + 𝑑2 ) = 0. (14) 𝑝𝑏𝑟𝑟1 (1 − 𝑚)(𝐾𝛽 − 𝑑) < 𝛽𝑑2 (𝑟1 + 𝑑1 )(𝐾𝛽 + 𝑟), then the
4 Abstract and Applied Analysis

roots of (18) have negative real part. Accordingly, 𝐸1 is locally [8], solutions are limited to 𝑀, the largest invariant subset
asymptotically stable. If 𝑝𝑏𝑟𝑟1 (1 − 𝑚)(𝐾𝛽 − 𝑑) > 𝛽𝑑2 (𝑟1 + of {𝑉1̇(𝑡) = 0}. Clearly, we see from (24) that 𝑉1̇(𝑡) = 0, if
𝑑1 )(𝐾𝛽 + 𝑟), then 𝐸1 is unstable. It is easily seen that and only if 𝑆(𝑡) = 𝑆1 , 𝑌2 (𝑡) = 0. It follows from the first and
fourth equations of (2) that 0 = 𝑆(𝑡)̇ = 𝑟 − (𝑟/𝐾)𝑆1 − ((𝐾𝛽 +
2
𝑔12 − 2𝑔0 = (𝑟1 + 𝑑1 ) + 𝑑22 > 0, 𝑔02 − 𝑓02 > 0. (19) 𝑟)/𝐾)𝐼(𝑡), 0 = 𝑌2̇(𝑡) = 𝑟1 𝑌1 (𝑡), which yields 𝐼(𝑡) = 𝐼1 , 𝑌1 (𝑡) =
0. Using LaSalle’s invariant principle, the global asymptotic
Hence, if 0 < 𝑝𝑏𝑟𝑟1 (1−𝑚)(𝐾𝛽−𝑑) < 𝛽𝑑2 (𝑟1 +𝑑1 )(𝐾𝛽+𝑟), stability of 𝐸1 follows. This completes the proof.
by Lemma B in [7], it follows that the equilibrium 𝐸1 is locally
asymptotically stable for all 𝜏 ≥ 0. If 𝑝𝑏𝑟𝑟1 (1 − 𝑚)(𝐾𝛽 − 𝑑) >
𝛽𝑑2 (𝑟1 + 𝑑1 )(𝐾𝛽 + 𝑟), then 𝐸1 is unstable for all 𝜏 ≥ 0. 4. Stability of Positive Equilibrium
In this section, we are concerned with the stability of the pos-
Theorem 4. Let 𝐾𝛽 > 𝑑 hold; the predator-extinction
itive equilibrium 𝐸∗ and the existence of Hopf bifurcations at
equilibrium 𝐸1 of model (2) is globally stable provided that
the positive equilibrium 𝐸∗ of model (2).
𝑝𝑏𝑟𝑟1 (1 − 𝑚) (𝐾𝛽 − 𝑑) < 𝛽𝑑2 (𝑟1 + 𝑑1 ) (𝐾𝛽 + 𝑟) . (20) If the following holds,
(H1) 𝑝𝑏𝑟𝑟1 (1 − 𝑚)(𝐾𝛽 − 𝑑) > 𝛽𝑑2 (𝑟1 + 𝑑1 )(𝐾𝛽 + 𝑟),
Proof. Based on the above discussions, we only prove the then model (2) has a unique positive equilibrium
global attractivity of the equilibrium 𝐸1 . Define 𝐸∗ (𝑆∗ , 𝐼∗ , 𝑌1∗ , 𝑌2∗ ), where

𝑉11 (𝑡) = 𝑐1 (𝑆 (𝑡) − 𝑆1 − 𝑆1 ln

𝑆 (𝑡)
) + 𝐼 (𝑡) − 𝐼1 𝑆∗ = (𝐾𝑝𝑟𝑟1 𝑏2 (1 − 𝑚)2 + 𝑎𝑑 (𝑟1 + 𝑑1 ) (𝐾𝛽 + 𝑟)
𝑆1
(21) −𝑏 (1 − 𝑚) 𝑑2 (𝑟1 + 𝑑1 ) (𝐾𝛽 + 𝑟) )
𝐼 (𝑡)
− 𝐼1 ln + 𝑘1 𝑌1 (𝑡) + 𝑘2 𝑌2 (𝑡) , −1
𝐼1 × (𝑝𝑟𝑟1 𝑏2 (1 − 𝑚)2 + 𝛽𝑎 (𝑟1 + 𝑑1 ) (𝐾𝛽 + 𝑟)) ,
(25)
where 𝑐1 = 𝐾𝛽/(𝐾𝛽 + 𝑟) and 𝑘1 = 1/𝑝, 𝑘2 = (𝑟1 +
𝑟 (𝐾 − 𝑆∗ ) 𝑑 + 𝑎𝑌2∗ ∗
𝑑1 )/(𝑝𝑟1 ). Calculating the derivative of 𝑉11 (𝑡) along the 𝐼∗ = , 𝑌1∗ = 2 𝑌2 ,
positive solutions of (2), it follows that 𝐾𝛽 + 𝑟 𝑟1
𝛽𝑆∗ − 𝑑
̇ (𝑡) = 𝑐1 𝑆 (𝑡) − 𝑆1 𝑆 ̇(𝑡)+ 𝐼 (𝑡) − 𝐼1 𝐼 ̇(𝑡)+𝑘1 𝑌1̇ (𝑡) + 𝑘2 𝑌2̇ (𝑡)
𝑉11 𝑌2∗ = .
𝑆 (𝑡) 𝐼 (𝑡) 𝑏 (1 − 𝑚)

𝛽𝑟 2
The characteristic equation of model (2) at the equilib-
=− [𝑆 (𝑡) − 𝑆1 ] − 𝑏 (1 − 𝑚) 𝐼 (𝑡) 𝑌2 (𝑡) rium 𝐸∗ takes the form
𝐾𝛽 + 𝑟
𝜆4 + 𝑝3 𝜆3 + 𝑝2 𝜆2 + 𝑝1 𝜆 + 𝑝0 + (𝑞2 𝜆2 + 𝑞1 𝜆 + 𝑞0 ) 𝑒−𝜆𝜏 = 0,
+ 𝑏 (1 − 𝑚) 𝐼 (𝑡 − 𝜏) 𝑌2 (𝑡 − 𝜏)
(26)
𝑑2 (𝑟1 + 𝑑1 ) where
−[ − 𝑏 (1 − 𝑚) 𝐼1 ] 𝑌2 (𝑡)
𝑝𝑟1 𝑟 ∗
𝑝3 = 𝑟1 + 𝑑1 + 𝑑2 + 2𝑎𝑌2∗ + 𝑆 ,
𝐾
𝑎 (𝑟1 + 𝑑1 ) 2
− 𝑌2 (𝑡) . 𝑟
𝑝𝑟1 𝑝2 = (𝑟1 + 𝑑1 ) (𝑑2 + 2𝑎𝑌2∗ ) + 𝑆∗ (𝑟1 + 𝑑1 + 𝑑2 + 2𝑎𝑌2∗ )
𝐾
(22)
𝐾𝛽 + 𝑟 ∗ ∗
+ 𝛽𝑆 𝐼 ,
Define 𝐾
𝑡
𝐾𝛽 + 𝑟 ∗ ∗
𝑉1 (𝑡) = 𝑉11 (𝑡) + 𝑏 (1 − 𝑚) ∫ 𝐼 (𝑠) 𝑌2 (𝑠) 𝑑𝑠. (23) 𝑝1 = 𝛽𝑆 𝐼 (𝑟1 + 𝑑1 + 𝑑2 + 2𝑎𝑌2∗ )
𝑡−𝜏 𝐾
𝑟
We derive from (22) and (23) that + 𝑆∗ (𝑟1 + 𝑑1 ) (𝑑2 + 2𝑎𝑌2∗ ) ,
𝐾
𝛽𝑟 2
𝑉1̇ (𝑡) = − [𝑆 (𝑡) − 𝑆1 ] 𝐾𝛽 + 𝑟 ∗ ∗
𝐾𝛽 + 𝑟 𝑝0 = 𝛽𝑆 𝐼 (𝑟1 + 𝑑1 ) (𝑑2 + 2𝑎𝑌2∗ ) ,
𝐾
𝑑2 (𝑟1 + 𝑑1 ) 𝑞2 = − 𝑝𝑏𝑟1 (1 − 𝑚) 𝐼∗ ,
−[ − 𝑏 (1 − 𝑚) 𝐼1 ] 𝑌2 (𝑡) (24)
𝑝𝑟1
𝑟 ∗
𝑞1 = 𝑝𝑏𝑟1 (1 − 𝑚) 𝐼∗ [𝑏 (1 − 𝑚) 𝑌2∗ − 𝑆 ],
𝑎 (𝑟1 + 𝑑1 ) 2 𝐾
− 𝑌2 (𝑡) .
𝑝𝑟1 𝑟 ∗ 𝐾𝛽 + 𝑟 ∗
𝑞0 = 𝑝𝑏𝑟1 (1 − 𝑚) 𝑆∗ 𝐼∗ [𝑏 (1 − 𝑚) 𝑌 − 𝛽𝐼 ] .
If 0 < 𝑝𝑏𝑟𝑟1 (1 − 𝑚)(𝐾𝛽 − 𝑑) < 𝛽𝑑2 (𝑟1 + 𝑑1 )(𝐾𝛽 + 𝑟), it 𝐾 2 𝐾
then follows from (24) that 𝑉1̇(𝑡) ≤ 0. By Theorem 5.3.1, in (27)
Abstract and Applied Analysis 5

It is easy to show that are then satisfied yielding a periodic solution. To this end,
differentiating equation (26) with respect to 𝜏, it follows that
𝑝3 > 0, 𝑝0 + 𝑞0 > 0, 𝑝1 + 𝑞1 > 0, 𝑝2 + 𝑞2 > 0.
(28) 𝑑𝜆 −1 4𝜆3 + 3𝑝3 𝜆2 + 2𝑝2 𝜆 + 𝑝1
( ) =
𝑑𝜏 −𝜆 (𝜆4 + 𝑝3 𝜆3 + 𝑝2 𝜆2 + 𝑝1 𝜆 + 𝑝0 )
When 𝜏 = 0, (26) becomes (35)
2𝑞2 𝜆 + 𝑞1 𝜏
4 3 2
𝜆 + 𝑝3 𝜆 + (𝑝2 + 𝑞2 ) 𝜆 + (𝑝1 + 𝑞1 ) 𝜆 + 𝑝0 + 𝑞0 = 0. (29) + − .
𝜆 (𝑞2 𝜆2 + 𝑞1 𝜆 + 𝑞0 ) 𝜆
If the following holds, Hence, a direct calculation shows that
(H2) (𝑝1 + 𝑞1 )[𝑝3 (𝑝2 + 𝑞2 ) − (𝑝0 + 𝑞0 )] > + 𝑞0 ),𝑝32 (𝑝0 𝑑(Re 𝜆)
then by the Routh-Hurwitz theorem, when 𝜏 = 0, sgn { }
𝑑𝜏
the coexistence equilibrium 𝐸∗ of model (2) is locally 𝜆=𝑖𝜔0

asymptotically stable and 𝐸∗ is unstable if (𝑝1 +

𝑑𝜆 −1
𝑞1 )[𝑝3 (𝑝2 + 𝑞2 ) − (𝑝0 + 𝑞0 )] < 𝑝32 (𝑝0 + 𝑞0 ). = sgn {Re ( ) }
𝑑𝜏 𝜆=𝑖𝜔 0
If 𝑖𝜔(𝜔 > 0) is a solution of (26), separating real and
imaginary parts, we have = sgn { ((3𝑝3 𝜔02 − 𝑝1 ) (𝑝3 𝜔02 − 𝑝1 ) + 2 (2𝜔02 − 𝑝2 )
(𝑞2 𝜔2 − 𝑞0 ) sin 𝜔𝜏 + 𝑞1 𝜔 cos 𝜔𝜏 = 𝑝3 𝜔3 − 𝑝1 𝜔,
× (𝜔04 − 𝑝2 𝜔02 + 𝑝0 ))
(30)
(𝑞2 𝜔2 − 𝑞0 ) cos 𝜔𝜏 − 𝑞1 𝜔 sin 𝜔𝜏 = 𝜔4 − 𝑝2 𝜔2 + 𝑝0 . 2 2 −1
× (𝜔02 (𝑝1 − 𝑝3 𝜔02 ) + (𝜔04 − 𝑝2 𝜔02 + 𝑝0 ) )
Squaring and adding the two equations of (30), it follows that
+ (−𝑞12 + 2𝑞2 (𝑞0 − 𝑞2 𝜔02 ))
8 6 4 2
𝜔 + ℎ3 𝜔 + ℎ2 𝜔 + ℎ1 𝜔 + ℎ0 = 0, (31)
2 2 −1
× ((𝑞1 𝜔0 ) + (𝑞2 𝜔02 − 𝑞0 ) ) } .
where
(36)
ℎ3 = 𝑝32 − 2𝑝2 , ℎ2 = 𝑝22 + 2𝑝0 − 2𝑝1 𝑝3 − 𝑞22 ,
(32) We derive from (30) that
ℎ1 = 𝑝12 − 2𝑝0 𝑝2 + 2𝑞0 𝑞2 − 𝑞12 , ℎ0 = 𝑝02 − 𝑞02 .
2 2
𝜔02 (𝑝1 − 𝑝3 𝜔02 ) + (𝜔04 − 𝑝2 𝜔02 + 𝑝0 )
Assume that the following holds: (37)
2 2
(H3) ℎ3 > 0, ℎ2 > 0, ℎ1 > 0. = (𝑞1 𝜔0 ) + (𝑞2 𝜔02 − 𝑞0 ) .

If ℎ0 > 0, by the general theory on characteristic equations Hence, it follows that

of delay differential equations from [9] (Theorem 4.1), 𝐸∗
remains stable for all 𝜏 > 0. If ℎ0 < 0, then (31) has a unique 𝑑(Re 𝜆)
sgn { }
positive root 𝜔0 ; that is, (26) admits a pair of purely imaginary 𝑑𝜏 𝜆=𝑖𝜔0
roots of the form ±𝑖𝜔0 . From (30), we see that (38)
4𝜔06 + 3ℎ3 𝜔04 + 2ℎ2 𝜔02 + ℎ1
2𝑛𝜋 1 = sgn { } > 0.
𝜏𝑛 = + arccos ((𝑞2 𝜔02 − 𝑞0 ) (𝜔04 − 𝑝2 𝜔02 + 𝑝0 ) 2
(𝑞1 𝜔0 ) + (𝑞2 𝜔02 − 𝑞0 )
2
𝜔0 𝜔0
Therefore, if (H3) holds, then the transversal condition holds
+𝑞1 𝜔0 (𝑝3 𝜔03 − 𝑝1 𝜔0 )) (33)
and a Hopf bifurcation occurs at 𝜔 = 𝜔0 , 𝜏 = 𝜏0 .
2 2 −1 In conclusion, we have the following results.
× ((𝑞1 𝜔0 ) + (𝑞2 𝜔02 − 𝑞0 ) ) , 𝑛 = 0, 1, 2, . . . .
Theorem 5. For model (2), let (𝐻1) hold, and we have the
By Theorem 3.4.1, in [9], we see that 𝐸∗ remains stable for following.
𝜏 < 𝜏0 .
(i) If (𝐻2) and (𝐻3) hold, ℎ0 > 0, then the positive
In the following, we claim that
equilibrium 𝐸∗ is locally asymptotically stable for all
𝑑 (Re (𝜆)) 󵄨󵄨󵄨󵄨 𝜏 ≥ 0.
󵄨󵄨 > 0. (34)
𝑑𝜏 󵄨󵄨𝜏=𝜏0 (ii) If (𝐻2) and (𝐻3) hold, ℎ0 < 0, then there exists a
positive number 𝜏0 , such that the positive equilibrium
This will show that there exists at least one eigenvalue with 𝐸∗ is locally asymptotically stable if 0 ≤ 𝜏 < 𝜏0 and is
a positive real part for 𝜏 > 𝜏0 . Moreover, the conditions for unstable if 𝜏 > 𝜏0 . Further, model (2) undergoes a Hopf
the existence of a Hopf bifurcation (Theorem 2.9.1 in [9]) bifurcation at 𝐸∗ when 𝜏 = 𝜏0 .
6 Abstract and Applied Analysis

(iii) If (𝑝1 + 𝑞1 )[𝑝3 (𝑝2 + 𝑞2 ) − (𝑝1 + 𝑞1 )] < 𝑝32 (𝑝0 + 𝑞0 ), then Consider the following auxiliary equations:
the positive equilibrium 𝐸∗ is unstable for all 𝜏 ≥ 0.
𝑑𝑧1
Now, we are concerned with the global attractiveness of = 𝑝𝑏 (1 − 𝑚) (𝑀1𝐼 + 𝜀) 𝑧2 (𝑡 − 𝜏) − (𝑟1 + 𝑑1 ) 𝑧1 (𝑡) ,
𝑑𝑡
the positive equilibrium 𝐸∗ .
𝑑𝑧2
= 𝑟1 𝑧1 (𝑡) − 𝑑2 𝑧2 (𝑡) − 𝑎𝑧22 (𝑡) .
Theorem 6. Let (𝐻1) hold, and then the positive equilibrium 𝑑𝑡
𝐸∗ (𝑆∗ , 𝐼∗ , 𝑌1∗ , 𝑌2∗ ) of model (2) is globally attractive provided (46)
that
If (H1) holds, then, by Lemma 2.4 in [10], it follows from (46)
𝑝𝑟𝑟1 𝑏2 (1 − 𝑚)2 .
𝑎𝛽 (𝑟1 + 𝑑1 ) (𝐾𝛽 + 𝑟) ≠ (39)
that
Proof. Let (𝑆(𝑡), 𝐼(𝑡), 𝑌1 (𝑡), 𝑌2 (𝑡)) be any positive solution of
model (2) with initial conditions (3). Let lim 𝑧 (𝑡) = (𝑝𝑏 (1 − 𝑚) (𝑀1𝐼 + 𝜀)
𝑡 → +∞ 1
𝑀𝑆 = lim sup 𝑆 (𝑡) , 𝑚𝑆 = lim inf 𝑆 (𝑡) ,
𝑡 → +∞
𝑡 → +∞ × [𝑝𝑏𝑟1 (1 − 𝑚) (𝑀1𝐼 + 𝜀) − 𝑑2 (𝑟1 + 𝑑1 )])
𝑀𝐼 = lim sup 𝐼 (𝑡) , 2 −1
𝑡 → +∞ × (𝑎(𝑟1 + 𝑑1 ) ) ,
𝑚𝐼 = lim inf 𝐼 (𝑡) ,
𝑡 → +∞ 𝑝𝑏𝑟1 (1 − 𝑚) (𝑀1𝐼 + 𝜀) − 𝑑2 (𝑟1 + 𝑑1 )
lim 𝑧2 (𝑡) = .
𝑀𝑌𝑖 = lim sup 𝑌𝑖 (𝑡) , 𝑚𝑌𝑖 = lim inf 𝑌𝑖 (𝑡) , (𝑖 = 1, 2) . 𝑡 → +∞ 𝑎 (𝑟1 + 𝑑1 )
𝑡 → +∞ 𝑡 → +∞
(47)
(40)
We now claim that 𝑀𝑆 = 𝑚𝑆 = 𝑆∗ , 𝑀𝐼 = 𝑚𝐼 = 𝐼∗ , 𝑀𝑌𝑖 = By comparison, for 𝜀 > 0, sufficiently small, we obtain that
𝑚𝑌𝑖 = 𝑌𝑖∗ (𝑖 = 1, 2). The technique of proof is to use an
iteration method. 𝑀𝑌1 = lim sup 𝑌1 (𝑡)
𝑡 → +∞
We derive from the first and the second equations of
model (2) that 𝑝𝑏 (1 − 𝑚) 𝑀1𝐼 [𝑝𝑏𝑟1 (1 − 𝑚) 𝑀1𝐼 − 𝑑2 (𝑟1 + 𝑑1 )]
𝑑𝑆 𝑆+𝐼 𝑑𝐼 ≤ 2
= 𝑟𝑆 (1 − ) − 𝛽𝑆𝐼, ≤ 𝛽𝑆𝐼 − 𝑑𝐼. (41) 𝑎(𝑟1 + 𝑑1 )
𝑑𝑡 𝐾 𝑑𝑡
𝑌
Consider the following auxiliary equations: := 𝑀1 1 ,
𝑑𝑥1 𝑥 + 𝑥2 𝑀𝑌2 = lim sup 𝑌2 (𝑡)
= 𝑟𝑥1 (1 − 1 ) − 𝛽𝑥1 𝑥2 ,
𝑑𝑡 𝐾 𝑡 → +∞
(42)
𝑑𝑥2 𝑝𝑏𝑟1 (1 − 𝑚) 𝑀1𝐼 − 𝑑2 (𝑟1 + 𝑑1 )
= 𝛽𝑥1 𝑥2 − 𝑑𝑥2 . =
𝑑𝑡 𝑎 (𝑟1 + 𝑑1 )
If 𝐾𝛽 > 𝑑, then, by Theorem 3.1 in [2], it follows from (42) 𝑌
that := 𝑀1 2 .
𝑑 𝑟 (𝐾𝛽 − 𝑑) (48)
lim 𝑥 (𝑡) = , lim 𝑥2 (𝑡) = . (43)
𝑡 → +∞ 1 𝛽 𝑡 → +∞ 𝛽 (𝐾𝛽 + 𝑟)
Hence, for 𝜀 > 0, sufficiently small, there is a 𝑇2 ≥ 𝑇1 + 𝜏 such
By comparison, we obtain that 𝑌
that if 𝑡 > 𝑇2 , then 𝑌2 (𝑡) ≤ 𝑀1 2 + 𝜀.
𝑑 For 𝜀 > 0, sufficiently small, we derive from the first and
𝑀𝑆 = lim sup 𝑆 (𝑡) ≤ := 𝑀1𝑆 , the second equations of model (2) that, for 𝑡 > 𝑇2 ,
𝑡 → +∞ 𝛽
(44)
𝑟 (𝐾𝛽 − 𝑑) 𝑑𝑆 𝑆+𝐼
𝑀𝐼 = lim sup 𝐼 (𝑡) ≤ := 𝑀1𝐼 . = 𝑟𝑆 (1 − ) − 𝛽𝑆𝐼,
𝑡 → +∞ 𝛽 (𝐾𝛽 + 𝑟) 𝑑𝑡 𝐾
(49)
Hence, for 𝜀 > 0, sufficiently small, there is a 𝑇1 > 0 such that 𝑑𝐼 𝑌
≥ 𝛽𝑆𝐼 − 𝑑𝐼 − 𝑏 (1 − 𝑚) (𝑀1 2 + 𝜀) 𝐼.
if 𝑡 > 𝑇1 , then 𝐼(𝑡) ≤ 𝑀1𝐼 + 𝜀. We therefore derive from the 𝑑𝑡
third and the fourth equations of model (2) that, for 𝑡 > 𝑇1 +𝜏,
Consider the following auxiliary equations:
𝑑𝑌1
≤ 𝑝𝑏 (1 − 𝑚) (𝑀1𝐼 + 𝜀) 𝑌2 (𝑡 − 𝜏) − (𝑟1 + 𝑑1 ) 𝑌1 (𝑡) ,
𝑑𝑡 𝑑𝑥1 𝑥 + 𝑥2
= 𝑟𝑥1 (1 − 1 ) − 𝛽𝑥1 𝑥2 ,
𝑑𝑌2 𝑑𝑡 𝐾
= 𝑟1 𝑌1 (𝑡) − 𝑑2 𝑌2 (𝑡) − 𝑎𝑌22 (𝑡) . (50)
𝑑𝑡 𝑑𝑥2 𝑌
(45) = 𝛽𝑥1 𝑥2 − 𝑑𝑥2 − 𝑏 (1 − 𝑚) (𝑀1 2 + 𝜀) 𝑥2 .
𝑑𝑡
Abstract and Applied Analysis 7

If (H1) holds, then, by Theorem 3.1 in [2], it follows from (50) By comparison, for 𝜀 > 0, sufficiently small, we obtain that
that
𝑚𝑌1 = lim inf 𝑌1 (𝑡)
𝑡 → +∞
𝑌
𝑑 + 𝑏 (1 − 𝑚) (𝑀1 2 + 𝜀)
lim 𝑥 (𝑡) = , 𝑝𝑏 (1 − 𝑚) 𝑁1𝐼 [𝑝𝑏𝑟1 (1 − 𝑚) 𝑁1𝐼 − 𝑑2 (𝑟1 + 𝑑1 )]
𝑡 → +∞ 1 𝛽 ≥ 2
𝑎(𝑟1 + 𝑑1 )
𝑌
𝑟 [ 𝑑 + 𝑏 (1 − 𝑚) (𝑀1 2 + 𝜀)
lim 𝑥2 (𝑡) = 𝐾− ]. 𝑌
:= 𝑁1 1 ,
𝑡 → +∞ 𝐾𝛽 + 𝑟 𝛽
[ ]
(51) 𝑚𝑌2 = lim inf 𝑌2 (𝑡)
𝑡 → +∞

𝑝𝑏𝑟1 (1 − 𝑚) 𝑁1𝐼 − 𝑑2 (𝑟1 + 𝑑1 )

By comparison, for 𝜀 > 0, sufficiently small, we conclude that ≥
𝑎 (𝑟1 + 𝑑1 )
𝑌 𝑌
𝑑 + 𝑏 (1 − 𝑚) 𝑀1 2 := 𝑁1 2 .
𝑚𝑆 = lim inf 𝑆 (𝑡) ≥ := 𝑁1𝑆 , (56)
𝑡 → +∞ 𝛽
𝑌 Hence, for 𝜀 > 0, sufficiently small, there is a 𝑇4 ≥ 𝑇3 + 𝜏, such
𝑟 𝑑 + 𝑏 (1 − 𝑚) 𝑀1 2
𝑚𝐼 = lim inf ≥ (𝐾 − ) := 𝑁1𝐼 . 𝑌
that if 𝑡 > 𝑇4 , 𝑌2 (𝑡) ≥ 𝑁1 2 − 𝜀.
𝑡 → +∞ 𝐾𝛽 + 𝑟 𝛽
For 𝜀 > 0, sufficiently small, we derive from the first and
(52)
the second equations of model (2) that, for 𝑡 > 𝑇4 ,

Hence, for 𝜀 > 0, sufficiently small, there is a 𝑇3 ≥ 𝑇2 such 𝑑𝑆 𝑆+𝐼

= 𝑟𝑆 (1 − ) − 𝛽𝑆𝐼,
that if 𝑡 > 𝑇3 , then 𝐼(𝑡) ≥ 𝑁1𝐼 − 𝜀. For 𝜀 > 0, sufficiently small, 𝑑𝑡 𝐾
we derive from the third and the fourth equations of model (57)
𝑑𝐼 𝑌
(2) that for 𝑡 > 𝑇3 + 𝜏 ≤ 𝛽𝑆𝐼 − 𝑑𝐼 − 𝑏 (1 − 𝑚) (𝑁1 2 − 𝜀) 𝐼.
𝑑𝑡

𝑑𝑌1 Consider the following auxiliary equations:

≥ 𝑝𝑏 (1 − 𝑚) (𝑁1𝐼 − 𝜀) 𝑌2 (𝑡 − 𝜏) − (𝑑1 + 𝑟1 ) 𝑌1 (𝑡) ,
𝑑𝑡
𝑑𝑥1 𝑥 + 𝑥2
𝑑𝑌2 = 𝑟𝑥1 (1 − 1 ) − 𝛽𝑥1 𝑥2 ,
= 𝑟1 𝑌1 (𝑡) − 𝑑2 𝑌2 (𝑡) − 𝑎𝑌22 (𝑡) . 𝑑𝑡 𝐾
𝑑𝑡 (58)
(53) 𝑑𝑥2 𝑌
= 𝛽𝑥1 𝑥2 − 𝑑𝑥2 − 𝑏 (1 − 𝑚) (𝑁1 2 − 𝜀) 𝑥2 .
𝑑𝑡
Consider the following auxiliary equations: If (H1) holds, then, by Theorem 3.1 in [2], it follows from (58)
that
𝑑𝑧1 𝑌
= 𝑝𝑏 (1 − 𝑚) (𝑁1𝐼 − 𝜀) 𝑧2 (𝑡 − 𝜏) − (𝑟1 + 𝑑1 ) 𝑧1 (𝑡) , 𝑑 + 𝑏 (1 − 𝑚) (𝑁1 2 − 𝜀)
𝑑𝑡 lim 𝑥1 (𝑡) = ,
𝑡 → +∞ 𝛽
𝑑𝑧2
= 𝑟1 𝑧1 (𝑡) − 𝑑2 𝑧2 (𝑡) − 𝑎𝑧22 (𝑡) . 𝑌
𝑑𝑡 𝑟 [ 𝑑 + 𝑏 (1 − 𝑚) (𝑁1 2 − 𝜀)
(54) lim 𝑥2 (𝑡) = 𝐾− ].
𝑡 → +∞ 𝐾𝛽 + 𝑟 𝛽
[ ]
Since (H1) holds, by Lemma 2.4 of [10], it follows from (54) (59)
that
By comparison, for 𝜀 > 0, sufficiently small, we obtain that

lim 𝑧 (𝑡) = (𝑝𝑏 (1 − 𝑚) (𝑁1𝐼 − 𝜀) 𝑑 + 𝑏 (1 − 𝑚) 𝑁1 2

𝑌
𝑡 → +∞ 1
𝑀𝑆 = lim sup 𝑆 (𝑡) ≤ := 𝑀2𝑆 ,
𝑡 → +∞ 𝛽
× [𝑝𝑏𝑟1 (1 − 𝑚) (𝑁1𝐼 − 𝜀) − 𝑑2 (𝑟1 + 𝑑1 )])
𝑌
𝑟 𝑑 + 𝑏 (1 − 𝑚) 𝑁1 2
2 −1 𝑀𝐼 = lim sup ≤ (𝐾 − ) := 𝑀2𝐼 .
× (𝑎(𝑟1 + 𝑑1 ) ) , 𝑡 → +∞ 𝐾𝛽 + 𝑟 𝛽

𝑝𝑏𝑟1 (1 − 𝑚) (𝑁1𝐼 − 𝜀) − 𝑑2 (𝑟1 + 𝑑1 ) (60)

lim 𝑧2 (𝑡) = .
𝑡 → +∞ 𝑎 (𝑟1 + 𝑑1 ) Therefore, for 𝜀 > 0, sufficiently small, there is a 𝑇5 ≥ 𝑇4 such
(55) that if 𝑡 > 𝑇5 , 𝐼(𝑡) ≤ 𝑀2𝐼 + 𝜀.
8 Abstract and Applied Analysis

For 𝜀 > 0, sufficiently small, we derive from the third and Consider the following auxiliary equations:
the fourth equations of model (2) that, for 𝑡 > 𝑇5 + 𝜏,
𝑑𝑥1 𝑥 + 𝑥2
𝑑𝑌1 = 𝑟𝑥1 (1 − 1 ) − 𝛽𝑥1 𝑥2 ,
≤ 𝑝𝑏 (1 − 𝑚) (𝑀2𝐼 + 𝜀) 𝑌2 (𝑡 − 𝜏) − (𝑑1 + 𝑟1 ) 𝑌1 (𝑡) , 𝑑𝑡 𝐾
𝑑𝑡 (66)
𝑑𝑥2 𝑌
𝑑𝑌2 = 𝛽𝑥1 𝑥2 − 𝑑𝑥2 − 𝑏 (1 − 𝑚) (𝑀2 2 + 𝜀) 𝑥2 .
= 𝑟1 𝑌1 (𝑡) − 𝑑2 𝑌2 (𝑡) − 𝑎𝑌22 (𝑡) . 𝑑𝑡
𝑑𝑡
(61) If (H1) holds, then, by Theorem 3.1 in [2], it follows from (66)
that
Consider the following auxiliary equations: 𝑌
𝑑 + 𝑏 (1 − 𝑚) (𝑀2 2 + 𝜀)
𝑑𝑧1 lim 𝑥1 (𝑡) = ,
= 𝑝𝑏 (1 − 𝑚) (𝑀2𝐼 + 𝜀) 𝑧2 (𝑡 − 𝜏) − (𝑟1 + 𝑑1 ) 𝑧1 (𝑡) ,
𝑡 → +∞ 𝛽
𝑑𝑡
𝑌
𝑑𝑧2 𝑟 [ 𝑑 + 𝑏 (1 − 𝑚) (𝑀2 2 + 𝜀)
= 𝑟1 𝑧1 (𝑡) − 𝑑2 𝑧2 (𝑡) − 𝑎𝑧22 (𝑡) . lim 𝑥2 (𝑡) = 𝐾− ].
𝑑𝑡 𝑡 → +∞ 𝐾𝛽 + 𝑟 𝛽
(62) [ ]
(67)
Since (H1) holds, by Lemma 2.4 of [10], it follows from (62) By comparison, for 𝜀 > 0, sufficiently small, we obtain that
that
𝑌
lim 𝑧1 (𝑡) = (𝑝𝑏 (1 − 𝑚) (𝑀2𝐼 + 𝜀) 𝑑 + 𝑏 (1 − 𝑚) 𝑀2 2
𝑡 → +∞ 𝑚𝑆 = lim inf 𝑆 (𝑡) ≤ := 𝑁2𝑆 ,
𝑡 → +∞ 𝛽
× [𝑝𝑏𝑟1 (1 − 𝑚) (𝑀2𝐼 + 𝜀) − 𝑑2 (𝑟1 + 𝑑1 )]) 𝑌
𝑟 𝑑 + 𝑏 (1 − 𝑚) 𝑀2 2
𝑚𝐼 = lim inf ≤ (𝐾 − ) := 𝑁2𝐼 .
2 −1 𝑡 → +∞ 𝐾𝛽 + 𝑟 𝛽
× (𝑎(𝑟1 + 𝑑1 ) ) ,
(68)
𝑝𝑏𝑟1 (1 − 𝑚) (𝑀2𝐼 + 𝜀) − 𝑑2 (𝑟1 + 𝑑1 )
lim 𝑧2 (𝑡) = . Hence, for 𝜀 > 0, sufficiently small, there is a 𝑇7 ≥ 𝑇6 such
𝑡 → +∞ 𝑎 (𝑟1 + 𝑑1 ) that if 𝑡 > 𝑇7 , 𝐼(𝑡) ≥ 𝑁2𝐼 − 𝜀. We therefore obtain from the
(63) third and the fourth equations of model (2) that for 𝑡 > 𝑇7 + 𝜏
By comparison, for 𝜀 > 0, sufficiently small, we conclude that 𝑑𝑌1
≥ 𝑝𝑏 (1 − 𝑚) (𝑁2𝐼 − 𝜀) 𝑌2 (𝑡 − 𝜏) − (𝑑1 + 𝑟1 ) 𝑌1 (𝑡) ,
𝑑𝑡
𝑀𝑌1 = lim sup 𝑌1 (𝑡)
𝑡 → +∞ 𝑑𝑌2
= 𝑟1 𝑌1 (𝑡) − 𝑑2 𝑌2 (𝑡) − 𝑎𝑌22 (𝑡) .
𝑝𝑏 (1 − 𝑚) 𝑀2𝐼 [𝑝𝑏𝑟1 (1 − 𝑚) 𝑀2𝐼 − 𝑑2 (𝑟1 + 𝑑1 )] 𝑑𝑡
≥ (69)
2
𝑎(𝑟1 + 𝑑1 )
Consider the following auxiliary equations:
𝑌
:= 𝑀2 1 ,
𝑑𝑧1
= 𝑝𝑏 (1 − 𝑚) (𝑁2𝐼 − 𝜀) 𝑧2 (𝑡 − 𝜏) − (𝑟1 + 𝑑1 ) 𝑧1 (𝑡) ,
𝑀𝑌2 = lim sup 𝑌2 (𝑡) 𝑑𝑡
𝑡 → +∞
𝑑𝑧2
𝑝𝑏𝑟1 (1 − 𝑚) 𝑀2𝐼 − 𝑑2 (𝑟1 + 𝑑1 ) = 𝑟1 𝑧1 (𝑡) − 𝑑2 𝑧2 (𝑡) − 𝑎𝑧22 (𝑡) .
≥ 𝑑𝑡
𝑎 (𝑟1 + 𝑑1 ) (70)
𝑌
:= 𝑀2 2 . Since (H1) holds, by Lemma 2.4 of [10], it follows from (70)
(64) that

lim 𝑧 (𝑡) = (𝑝𝑏 (1 − 𝑚) (𝑁2𝐼 − 𝜀)

Therefore, for 𝜀 > 0, sufficiently small, there is a 𝑇6 ≥ 𝑇5 + 𝜏 𝑡 → +∞ 1
𝑌
such that if 𝑡 > 𝑇6 , 𝑦2 (𝑡) ≤ 𝑀2 2 + 𝜀.
For 𝜀 > 0, sufficiently small, it follows from the first and × [𝑝𝑏𝑟1 (1 − 𝑚) (𝑁2𝐼 − 𝜀) − 𝑑2 (𝑟1 + 𝑑1 )])
the second equations of model (2) that for 𝑡 > 𝑇6 2 −1
× (𝑎(𝑟1 + 𝑑1 ) ) ,
𝑑𝑆 𝑆+𝐼
= 𝑟𝑆 (1 − ) − 𝛽𝑆𝐼,
𝑑𝑡 𝐾 𝑝𝑏𝑟1 (1 − 𝑚) (𝑁2𝐼 − 𝜀) − 𝑑2 (𝑟1 + 𝑑1 )
(65) lim 𝑧2 (𝑡) = .
𝑑𝐼 𝑌
𝑡 → +∞ 𝑎 (𝑟1 + 𝑑1 )
≥ 𝛽𝑆𝐼 − 𝑑𝐼 − 𝑏 (1 − 𝑚) (𝑀2 2 + 𝜀) 𝐼 (𝑡) . (71)
𝑑𝑡
Abstract and Applied Analysis 9

By comparison, for 𝜀 > 0, sufficiently small, we obtain that nondecreasing. Hence, the limit of each sequence in
𝑌 𝑌 𝑌 𝑌
𝑀𝑘𝑆 , 𝑀𝑘𝐼 , 𝑀𝑘 1 , 𝑀𝑘 2 , 𝑁𝑘𝑆 , 𝑁𝑘𝐼 , 𝑁𝑘 1 , 𝑁𝑘 2 exists. Denote
𝑚𝑌1 = lim inf 𝑌1 (𝑡)
𝑡 → +∞ lim 𝑀𝑘𝑆 = 𝑆, lim 𝑀𝑘𝐼 = 𝐼,
𝑘 → +∞ 𝑘 → +∞
𝑝𝑏 (1 − 𝑚) 𝑁2𝐼 [𝑝𝑏𝑟1 (1 − 𝑚) 𝑁2𝐼 − 𝑑2 (𝑟1 + 𝑑1 )] 𝑌
≥ lim 𝑀𝑘 𝑖 = 𝑌𝑖 , (𝑖 = 1, 2) ,
2 𝑘 → +∞
𝑎(𝑟1 + 𝑑1 ) (75)
𝑌 lim 𝑁𝑘𝑆 = 𝑆, lim 𝑁𝑘𝐼 = 𝐼,
:= 𝑁2 1 , 𝑘 → +∞ 𝑘 → +∞
(72)
𝑌
𝑚𝑌2 = lim inf 𝑌2 (𝑡) lim 𝑁𝑘 𝑖 = 𝑌𝑖 , (𝑖 = 1, 2) .
𝑡 → +∞ 𝑘 → +∞

𝑝𝑏𝑟1 (1 − 𝑚) 𝑁2𝐼 − 𝑑2 (𝑟1 + 𝑑1 ) From (73), we can obtain

≥
𝑎 (𝑟1 + 𝑑1 )
1
𝑌 𝑆= [𝑑 + 𝑏 (1 − 𝑚) 𝑌2 ] ,
:= 𝑁2 2 . 𝛽
𝑟
𝐼= (𝐾 − 𝑆) ,
Continuing this process, we derive eight sequences 𝑟 + 𝐾𝛽
𝑌 𝑌 𝑌 𝑌
𝑀𝑘𝑆 , 𝑀𝑘𝐼 , 𝑀𝑘 1 , 𝑀𝑘 2 , 𝑁𝑘𝑆 , 𝑁𝑘𝐼 , 𝑁𝑘 1 , 𝑁𝑘 2 (𝑘 = 1, 2, . . .) such 𝑝𝑏 (1 − 𝑚) 𝑌2 𝐼
that, for 𝑘 ≥ 2, 𝑌1 = ,
𝑟1 + 𝑑1

𝑌 𝑝𝑏𝑟1 (1 − 𝑚) 𝐼 − 𝑑2 (𝑟1 + 𝑑1 )
𝑑 + 𝑏 (1 − 𝑚) 𝑁𝑘−1
2
𝑌2 = ,
𝑀𝑘𝑆 = , 𝑎 (𝑟1 + 𝑑1 )
𝛽 (76)
1
𝑟 𝑆 = [𝑑 + 𝑏 (1 − 𝑚) 𝑌2 ] ,
𝑀𝑘𝐼 = (𝐾 − 𝑀𝑘𝑆 ) , 𝛽
𝑟 + 𝐾𝛽
𝑟
𝑌
𝑝𝑏 (1 − 𝑚) 𝑀𝑘 2 𝑀𝑘𝐼 𝐼= (𝐾 − 𝑆) ,
𝑌
𝑀𝑘 1 = , 𝑟 + 𝐾𝛽
𝑟1 + 𝑑1
𝑝𝑏 (1 − 𝑚) 𝑌2 𝐼
𝑝𝑏𝑟1 (1 − 𝑚) 𝑀𝑘𝐼 − 𝑑2 (𝑟1 + 𝑑1 ) 𝑌1 = ,
𝑀𝑘 2
𝑌
= , 𝑟1 + 𝑑1
𝑎 (𝑟1 + 𝑑1 )
(73) 𝑝𝑏𝑟1 (1 − 𝑚) 𝐼 − 𝑑2 (𝑟1 + 𝑑1 )
𝑌 𝑌2 = .
𝑑 + 𝑏 (1 − 𝑚) 𝑀𝑘 2 𝑎 (𝑟1 + 𝑑1 )
𝑁𝑘𝑆 = ,
𝛽
It follows from (76) that
𝑟
𝑁𝑘𝐼 = (𝐾 − 𝑁𝑘𝑆 ) ,
𝑟 + 𝐾𝛽 𝛽 (𝐾𝛽 + 𝑟) 𝐼 = 𝑟 (𝐾𝛽 − 𝑑)
𝑌
𝑌 𝑝𝑏 (1 − 𝑚) 𝑁𝑘 2 𝑁𝑘𝐼 𝑏𝑟 (1 − 𝑚) 𝑑2 𝑝𝑟𝑟1 𝑏2 (1 − 𝑚)2 (77)
𝑁𝑘 1 = , + − 𝐼,
𝑟1 + 𝑑1 𝑎 𝑎 (𝑟1 + 𝑑1 )

𝑌 𝑝𝑏𝑟1 (1 − 𝑚) 𝑁𝑘𝐼 − 𝑑2 (𝑟1 + 𝑑1 ) 𝛽 (𝐾𝛽 + 𝑟) 𝐼 = 𝑟 (𝐾𝛽 − 𝑑)

𝑁𝑘 2 = .
𝑎 (𝑟1 + 𝑑1 )
𝑏𝑟 (1 − 𝑚) 𝑑2 𝑝𝑟𝑟1 𝑏2 (1 − 𝑚)2 (78)
+ − 𝐼,
𝑎 𝑎 (𝑟1 + 𝑑1 )
It is readily seen that
and (77) minus (78) results in

𝑁𝑘𝑆 ≤ 𝑚𝑆 ≤ 𝑀𝑆 ≤ 𝑀𝑘𝑆 , 𝑁𝑘𝐼 ≤ 𝑚𝐼 ≤ 𝑀𝐼 ≤ 𝑀𝑘𝐼 , 𝑝𝑟𝑟1 𝑏2 (1 − 𝑚)2

[𝛽 (𝐾𝛽 + 𝑟) − ] (𝐼 − 𝐼) = 0. (79)
(74) 𝑎 (𝑟1 + 𝑑1 )
𝑌 𝑌
𝑁𝑘 𝑖 ≤ 𝑚𝑌𝑖 ≤ 𝑀𝑌𝑖 ≤ 𝑀𝑘 𝑖 (𝑖 = 1, 2) .
If 𝑎𝛽(𝑟1 + 𝑑1 )(𝐾𝛽 + 𝑟) ≠ 𝑝𝑟𝑟1 𝑏2 (1 − 𝑚)2 , then we derive
𝑌 𝑌
from (79) that 𝐼 = 𝐼. It therefore follows from (76) that
Note that the sequences 𝑀𝑘𝑆 , 𝑀𝑘𝐼 , 𝑀𝑘 1 , 𝑀𝑘 2 are 𝑆 = 𝑆, 𝑌1 = 𝑌1 , 𝑌2 = 𝑌2 . We therefore conclude that 𝐸∗
𝑌 𝑌
nonincreasing and the sequences 𝑁𝑘𝑆 , 𝑁𝑘𝐼 , 𝑁𝑘 1 , 𝑁𝑘 2 are is globally attractive. The proof is complete.
10 Abstract and Applied Analysis

5. Conclusion
In this paper, we have incorporated a prey refuge, stage struc-
ture for the predator, and time delay due to the gestation of the
predator into an ecoepidemiological predator-prey model. By
using Lyapunov functions and the LaSalle invariant principle,
the global stability of each of the boundary equilibria of
the model is discussed. By using the iteration technique
and comparison arguments, sufficient conditions are derived
for the global attractivity of the positive equilibrium of the
model. By Theorem 4, we see that the predator population go
to extinction if 0 < 𝑝𝑏𝑟𝑟1 (1 − 𝑚)(𝐾𝛽 − 𝑑) < 𝛽𝑑2 (𝑟1 + 𝑑1 )(𝐾𝛽 +
𝑟). By Theorem 6, we see that if 𝑝𝑏𝑟𝑟1 (1 − 𝑚)(𝐾𝛽 − 𝑑) >
𝛽𝑑2 (𝑟1 +𝑑1 )(𝐾𝛽+𝑟) and 𝑎𝛽(𝑟1 +𝑑1 )(𝐾𝛽+𝑟) ≠ 𝑝𝑟𝑟1 𝑏2 (1 − 𝑚)2 ,
then both the prey and the predator species of model (2) are
permanent.

Conflict of Interests
The authors declare that there is no conflict of interests
regarding the publication of this paper.

Acknowledgment
This work was supported by the National Natural Science
Foundation of China (11101117).

References
[1] R. M. Anderso and R. M. May, Intections Disease of Humans
Dynamics and Control, Oxford University Press, Oxford, UK,
1991.
[2] S. Wang, The research of eco-epidemiological of models incorpo-
rating prey refuges [Ph.D. thesis], Lanzhou University, 2012.
[3] W. Wang and L. Chen, “A predator-prey system with stage-
structure for predator,” Computers & Mathematics with Appli-
cations, vol. 33, no. 8, pp. 83–91, 1997.
[4] Y. N. Xiao and L. S. Chen, “Global stability of a predator-prey
system with stage structure for the predator,” Acta Mathematica
Sinica, vol. 20, no. 1, pp. 63–70, 2004.
[5] R. Xu and Z. Ma, “Stability and Hopf bifurcation in a predator-
prey model with stage structure for the predator,” Nonlinear
Analysis: Real World Applications, vol. 9, no. 4, pp. 1444–1460,
2008.
[6] W. G. Aiello and H. I. Freedman, “A time-delay model of single-
species growth with stage structure,” Mathematical Biosciences,
vol. 101, no. 2, pp. 139–153, 1990.
[7] Y. Kuang and J. W. H. So, “Analysis of a delayed two-stage popu-
lation model with space-limited recruitment,” SIAM Journal on
Applied Mathematics, vol. 55, no. 6, pp. 1675–1696, 1995.
[8] J. Hale, Theory of Functional Differential Equations, Springer,
Heidelberg, Germany, 1977.
[9] Y. Kuang, Delay Differential Equation with Application in
Population Synamic, Academic Press, New York, NY, USA, 1993.
[10] R. Xu and Z. Ma, “Stability and Hopf bifurcation in a predator-
prey model with stage structure for the predator,” Nonlinear
Analysis. Real World Applications, vol. 9, no. 4, pp. 1444–1460,
2008.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 376919, 10 pages
http://dx.doi.org/10.1155/2014/376919

Research Article
Permanence of Diffusive Models for Three Competing Species in
Heterogeneous Environments

Benlong Xu and Zhenzhang Ni

Department of Mathematics, Shanghai Normal University, Shanghai 200234, China

Correspondence should be addressed to Benlong Xu; bxu@shnu.edu.cn

Received 29 May 2014; Revised 20 July 2014; Accepted 21 July 2014; Published 12 August 2014

Academic Editor: Yongli Song

Copyright © 2014 B. Xu and Z. Ni. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

We address the question of the long-term coexistence of three competing species whose dynamics are governed by the partial
differential equations. We obtain criteria for permanent coexistence in a Lotka-Volterra system modeling the interaction of three
competing species in a bounded habitat whose exterior is lethal to each species. It is also proved that if the intercompeting strength
is very weak, the system is always permanent, provided that each single one of the three species can survive in the absence of the
two other species.

1. Introduction 𝜕𝑢2
= 𝜇2 Δ𝑢2 + 𝑢2 (𝑎2 (𝑥) − 𝑏21 (𝑥) 𝑢1
𝜕𝑡
Reaction-diffusion systems are some of the most widely used
models for population dynamics in situations where spatial −𝑏22 (𝑥) 𝑢2 − 𝑏23 (𝑥) 𝑢3 ) in Ω,
dispersal plays a significant role. Generally speaking, the 𝜕𝑢3
dynamical behaviors of the systems with two competing = 𝜇3 Δ𝑢3 + 𝑢3 (𝑎3 (𝑥) − 𝑏31 (𝑥) 𝑢1
species are relatively simple and have been widely investigated 𝜕𝑡
in the past years. However, the investigation and knowledge −𝑏32 (𝑥) 𝑢2 − 𝑏33 (𝑥) 𝑢3 ) in Ω
about the systems with three competing species are very (1)
limited; see [1–5] for some related results. It is well known,
in general, that the dynamical behavior of systems with
three competing species may be extremely complex, even in with 𝑢𝑖 = 0 on 𝜕𝑢𝑖 × R+ , where Ω ⊆ R𝑁 is a bounded domain
nondiffusive case, namely, ODE case; see, for example, [6–9] and 𝑎𝑖 (𝑥) > 0, 𝑏𝑖𝑖 (𝑥) > 0, 𝑏𝑖𝑗 (𝑥) ≥ 0 (𝑖, 𝑗 = 1, 2, 3, 𝑖 ≠𝑗)
and the references therein. are all continuous in Ω. The variables 𝑢𝑖 represent population
The object of this paper is to study the problem of densities of the competing species. The boundary condition
coexistence for three competing species dispersing through 𝑢𝑖 = 0 describes a situation that the boundary of Ω is lethal
a spatially heterogeneous region. We model the population to the species.
dynamics of the species with a system of three diffusive Lotka- We will use the criterion of permanence to characterize
Volterra equations. coexistence. A system is said to be permanent if any solutions
The model we consider has the general form with all components positive initially must ultimately enter
and remain within a fixed set of positive states that are strictly
bounded away from zero in each component. For some
investigations of coexistence characterized by permanence,
𝜕𝑢1 mostly in the case for two interacting species, see [10–12], and
= 𝜇1 Δ𝑢1 + 𝑢1 (𝑎1 (𝑥) − 𝑏11 (𝑥) 𝑢1 also some parts of the books [13, 14].
𝜕𝑡
This paper is organized as follows. In Section 2, we
−𝑏12 (𝑥) 𝑢2 − 𝑏13 (𝑥) 𝑢3 ) in Ω, introduce some necessary material about semiflow and state
2 Abstract and Applied Analysis

V. Hutson’s average Lyapunov theorem which will be the A set 𝑈 ⊂ 𝑌0 is said to be strongly bounded if it is bounded
basic tool in the proof of our main theorem. We also and 𝑑(𝑈, 𝜕𝑌0 ) > 0. A0 is said to be a global attractor relative
discuss the dissipative property of our systems, which is to strongly bounded sets if it is a compact invariant subset of
necessary for using V. Hutson’s average Lyapunov function 𝑌0 and lim𝑡 → ∞ 𝑑(𝑈 ⋅ 𝑡, A0 ) = 0 for all strongly bounded 𝑈.
theorem. In Section 3, we establish our main permanent Permanence is obviously an asymptotic property. It can
result, Theorem 11. We also prove that if the intercompeting thus be studied by examining the semiflow restricted to
strength is “very weak,” the system (1) is always permanent, a forward invariant set derived from an 𝜀-neighbourhood
provided that either single one of the three species can survive 𝐵(A, 𝜀) of the global attractor A of Theorem 1. Set then
with the absence of the two other species. In Section 4, we give 𝑋 = 𝑐𝑙𝜋(𝐵(A, 𝜀), [1, ∞)), the closure of 𝜋(𝐵(A, 𝜀), [1, ∞)),
some summery and discussion of the results of this paper. and take 𝑆 = 𝑋 ∩ 𝜕𝑌0 . The following Hutson’s theorem on
average Lyapunov functions is the basic tool for establishing
2. Semiflows and Dissipativity our main theorem in Section 3.

In this section, we will make some preparations for estab- Theorem 3 (see [15]). Assume that the conditions of Theorem 1
lishing our main results in the next section. Firstly, we will hold, and let 𝑋, 𝑆 be as defined above. Suppose that 𝑃 : 𝑋\𝑆 →
introduce some terminologies and results about semiflows R+ is continuous, strictly positive, and bounded, and for 𝑢 ∈ 𝑆
(semigroups) for readers’ convenience. The materials can be define
found in [15] or [13, 14].
Let (𝑌, 𝑑) be a metric space, with points in 𝑌 being 𝑃 (𝜐 ⋅ 𝑡)
𝛼 (𝑡, 𝑢) = lim inf ( ). (3)
denoted by 𝑢, V, . . . and subsets 𝑌 by 𝑈, 𝑉, . . .. The following 𝜐 → 𝑢,𝜐∈𝑋\𝑆 𝑃 (𝜐)
two unsymmetric distances of sets will be used:
Then the semiflow is permanent if
𝑑 (𝑈, 𝑉) = sup 𝑑 (𝑢, 𝑉) ,
𝑢∈𝑈
(2) sup 𝛼 (𝑡, 𝑢) > 1 𝑢 ∈ 𝜔 (𝑆) ,
𝑡>0
𝑑 (𝑈, 𝑉) = inf 𝑑 (𝑢, 𝑉) . (4)
𝑢∈𝑈
sup 𝛼 (𝑡, 𝑢) > 0 𝑢 ∈ 𝑆.
The triple (𝑌, 𝜋, R+ ) is said to be a semiflow (or semigroup), 𝑡>0
if 𝜋 : 𝑌 × R+ → 𝑌 is continuous and satisfies:
For our application of the last theorem, we will cast the
(i) 𝜋(𝑢, 0) = 𝑢, system (1) to the abstract frame of the semiflow. It is well
(ii) 𝜋(𝜋(𝑢, 𝑡), 𝑠) = 𝜋(𝑢, 𝑡 + 𝑠)(𝑠, 𝑡 ∈ 𝑅+ ), known that if (1) is simply viewed as a parabolic system,
it actually generated a semiflow in 𝐶1 (Ω), and solutions
for all 𝑢 ∈ 𝑌. For convenience, we often write 𝜋(𝑢, 𝑡) = 𝑢 ⋅ 𝑡.
The symbols 𝛾+ (𝑢) and 𝜔(𝑢) denote the semiorbit through 𝑢 which belong to 𝐶1 (Ω) must have Hölder-continuous second
and the omega limit set of 𝑢, respectively, and the equivalent derivatives on Ω (see [13]). Thus, we may use maximum
expressions for sets are defined by taking unions. principles to obtain a priori bounds, even though we will
A solution 𝜙 through 𝑢 is a continuous map 𝜙 : 𝑅 → 𝑌 ultimately want to view our semiflow as acting on 𝐶1 (Ω).
such that 𝜙(0) = 𝑢 and 𝜋(𝜙(𝜏), 𝑡) = 𝜙(𝑡 + 𝜏) for 𝑡 ∈ R+ , 𝜏 ∈ Then by maximum principle (or comparison theorem), any
R. The range of 𝜙 is denoted by 𝛾(𝑢) and is called an orbit solution (𝑢1 , 𝑢2 , 𝑢3 ) of (1) must satisfy a uniform bound of
through 𝑢. the form 0 ≤ 𝑢𝑖 ≤ 𝐶𝑖 for some constants 𝐶𝑖 , 𝑖 = 1, 2, 3, after
A set 𝑈 is said to be forward invariant if 𝛾+ (𝑢) ⊂ 𝑈 and finite time. So the semiflow generated by (1) is dissipative. We
invariant if 𝛾(𝑢) ⊂ 𝑈. The semiflows is said to be dissipative if state it as follows.
there is a bounded set 𝑈 such that lim𝑡 → ∞ 𝑑(𝑢 ⋅ 𝑡, 𝑈) = 0 for
all 𝑢 ∈ 𝑌. 𝑈 is said to be a global attractor of the semiflow if Theorem 4. The semiflow generated by the system (1) is
dissipative.
it is compact invariant and for all bounded 𝑉, lim𝑡 → ∞ 𝑑(𝑉 ⋅
𝑡, 𝑈) = 0, where 𝑉 ⋅ 𝑡 = {V ⋅ 𝑡, V ∈ 𝑉}.
3. Permanent Coexistence Results
Theorem 1. Let 𝑌 be complete and suppose that the semiflow
is dissipative. Assume that there is a 𝑡0 ≥ 0 such that 𝜋(⋅, 𝑡) is In this section, we establish criteria of permanent coexistence
compact for 𝑡 > 𝑡0 ; then there is a nonempty global attractor. of the system (1). The main tool is Hutson’s average Lyapunov
function theorem stated in Section 2, and the key step is to
Consider next the concept of permanence in the abstract set up a suitable average Lyapunov function. To construct
semiflow context. We suppose that 𝑌 = 𝑌0 ∪ 𝜕𝑌0 , where 𝑌0 the average Lyapunov functions, we must have a detailed
is open, and assume that 𝑌0 , 𝜕𝑌0 are forward invariant. In knowledge of the 𝜔-limit set of the generated semiflow in the
relation to the remarks in the introduction, 𝜕𝑌0 will consist boundary of the positive cone.
of the states with at least one species identically zero. The maximum principle implies that solutions of (1) with
nonnegative nonzero initial data for a given component will
Definition 2. The semiflow is said to be permanent if there have that component strictly positive in Ω for all 𝑡 > 0. In the
exists a bounded set 𝑈 with 𝑑(𝑈, 𝜕𝑌0 ) > 0 such that case of Dirichlet conditions, such solutions will have normal
lim𝑡 → ∞ 𝑑(V ⋅ 𝑡, 𝑈) = 0 for all V ∈ 𝑌0 . derivatives on 𝜕Ω which are bounded above by a negative
Abstract and Applied Analysis 3

constant. Hence, the only trajectories which remain in the Case 3. One has 𝑢1 ≡ 0,
boundary of the positive cone have one or both components
identically zero. Thus, to determine the 𝜔-limit set of the
semiflow generated by (1) on the boundary, we need only 𝑢2𝑡 = 𝜇2 Δ𝑢2 + 𝑢2 (𝑎2 (𝑥) − 𝑏22 (𝑥) 𝑢2 − 𝑏23 (𝑥) 𝑢3 )
consider the steady state solutions (equilibrium points) of in Ω × (0, ∞) ,
subsystems of (1).
Let 𝜆 1 > 0 denote the principal eigenvalue for problem 𝑢3𝑡 = 𝜇3 Δ𝑢3 + 𝑢3 (𝑎3 (𝑥) − 𝑏32 (𝑥) 𝑢2 − 𝑏33 (𝑥) 𝑢3 ) (10)

Δ𝜙 + 𝜆𝜙 = 0 in Ω, in Ω × (0, ∞) ,
(5)
𝜙=0 on 𝜕Ω. 𝑢1 = 𝑢3 = 0 on 𝜕Ω × (0, ∞) .

In the rest of this paper, we denote

With some modification of lower-supper solution meth-
𝑎 (𝑥) ods of parabolic systems (see [14]), it is easy to prove the
𝑀𝑖 = max 𝑖 . (6) following results.
𝑏
𝑥∈Ω 𝑖𝑖 (𝑥)

The point (0, 0, 0) is always a steady state, which means Theorem 7. With 𝑀𝑖 , 𝑖 = 1, 2, 3, as defined in (6),
that there are no species in the domain. With the absence
of two species and only the other one species left, system (1) (1) if
becomes a scalar equation, and we have the following well-
known result (see [14]).
𝑎1 (𝑥) > 𝜇1 𝜆 1 + 𝑀2 𝑏12 (𝑥) ,
Lemma 5. Suppose that 𝑎(𝑥) > 𝜇𝑖 𝜆 1 for 𝑥 ∈ Ω. Then the
following problem 𝑎2 (𝑥) > 𝜇2 𝜆 1 + 𝑀1 𝑏21 (𝑥) , (11)

𝑢𝑖𝑡 = 𝜇𝑖 Δ𝑢𝑖 + 𝑢𝑖 (𝑎𝑖 (𝑥) − 𝑏𝑖𝑖 (𝑥) 𝑢𝑖 ) 𝑖𝑛 Ω × (0, ∞) , ∀𝑥 ∈ Ω,

(7)
𝑢𝑖 = 0 on 𝜕Ω × (0, ∞) ,

for = 1, 2, 3, has a unique positive steady state solution 𝑢𝑖 (𝑥)

which is globally approximately stable. 𝑢(12) (𝑥),
the system (8) has a steady state solution (̃
(21) (12) (21)
𝑢̃ (𝑥)) with 𝑢̃ (𝑥) > 0, 𝑢̃ (𝑥) > 0 for 𝑥 ∈ Ω;
Remark 6. Lemma 5 tells us that if 𝑎(𝑥) > 𝜇𝑖 𝜆 1 for 𝑥 ∈ Ω, 𝑖 = similarly,
1, 2, 3, then either single one of the three species can survive
in the absence of the other two species. (2) if

Now we consider the subsystem of (1) with only one

species to be absent. There are three cases. 𝑎1 (𝑥) > 𝜇1 𝜆 1 + 𝑀3 𝑏13 (𝑥) ,
(12)
Case 1. One has 𝑢3 ≡ 0, 𝑎3 (𝑥) > 𝜇3 𝜆 1 + 𝑀1 𝑏31 (𝑥) , ∀𝑥 ∈ Ω,

𝑢1𝑡 = 𝜇1 Δ𝑢1 + 𝑢1 (𝑎1 (𝑥) − 𝑏11 (𝑥) 𝑢1 − 𝑏12 (𝑥) 𝑢2 )

in Ω × (0, ∞) , 𝑢(13) (𝑥),
the system (9) has a steady state solution (̃
(31) (13) (31)
𝑢2𝑡 = 𝜇2 Δ𝑢2 + 𝑢2 (𝑎2 (𝑥) − 𝑏21 (𝑥) 𝑢1 − 𝑏22 (𝑥) 𝑢2 ) (8) 𝑢̃ (𝑥)) with 𝑢̃ (𝑥) > 0, 𝑢̃ (𝑥) > 0 for 𝑥 ∈ Ω;

in Ω × (0, ∞) , (3) if
𝑢1 = 𝑢2 = 0 on 𝜕Ω × (0, ∞) .
𝑎2 (𝑥) > 𝜇2 𝜆 1 + 𝑀3 𝑏23 (𝑥) ,
Case 2. One has 𝑢2 ≡ 0, (13)
𝑎3 (𝑥) > 𝜇3 𝜆 1 + 𝑀2 𝑏32 (𝑥) , ∀𝑥 ∈ Ω,
𝑢1𝑡 = 𝜇1 Δ𝑢1 + 𝑢1 (𝑎1 (𝑥) − 𝑏11 (𝑥) 𝑢1 − 𝑏13 (𝑥) 𝑢3 )
in Ω × (0, ∞) ,

𝑢3𝑡 = 𝜇3 Δ𝑢3 + 𝑢3 (𝑎3 (𝑥) − 𝑏31 (𝑥) 𝑢1 − 𝑏33 (𝑥) 𝑢3 ) (9) 𝑢(23) (𝑥),
the system (10) has a steady state solution (̃
(32) (23) (32)
𝑢̃ (𝑥)) with 𝑢̃ (𝑥) > 0, 𝑢̃ (𝑥) > 0 for 𝑥 ∈ Ω.
in Ω × (0, ∞) ,
Proof. We only prove the part (1); the proof of part (2) and
𝑢1 = 𝑢3 = 0 on 𝜕Ω × (0, ∞) . part (3) are similar.
4 Abstract and Applied Analysis

Let 𝑤(𝑥) be a positive eigenfunction of the principal (2) if

eigenvalue 𝜆 1 for the eigenvalue problem (5). Choose 𝑟1 > 0
sufficiently small, 2
𝑏13 (𝑥) 𝑢(1) 2
(𝑥) 𝑢(3)
𝑏13 (𝑥) 𝑏31 (𝑥) 𝑏31
2
+ 2 + 2
< 4, (18)
𝜇1 Δ (𝑟1 𝑤 (𝑥)) + 𝑟1 𝑤 (𝑥) 𝑏11 (𝑥) 𝑢(31) 𝑏11 (𝑥) 𝑏33 (𝑥) 𝑏33 (𝑥) 𝑢(13)

× [𝑎1 (𝑥) − 𝑏11 (𝑥) 𝑟1 𝑤 (𝑥) − 𝑏12 (𝑥) 𝑢2 ]

𝑢(13) (𝑥), 𝑢̃(31) (𝑥)) of
then the steady state solution (̃
≥ 𝑟1 𝑤 (𝑥) [𝑎1 (𝑥) − 𝜇1 𝜆 1
system (9) is unique;
−𝑏11 (𝑥) 𝑟1 𝑤 (𝑥) − 𝑏12 (𝑥) 𝑀2 ] ≥ 0, (14)
(3) if
𝜇1 Δ (𝑀1 ) + 𝑀1 [(𝑎1 (𝑥) − 𝑏11 (𝑥) 𝑀1 − 𝑏12 (𝑥) 𝑢2 ]
2
𝑏23 (𝑥) 𝑢(2) 2
𝑏23 (𝑥) 𝑏32 (𝑥) 𝑏32 (𝑥) 𝑢(3)
= 𝑀1 [(𝑎1 (𝑥) − 𝑏11 (𝑥) 𝑀1 − 𝑏12 (𝑥) 𝑢2 ] 2
+ 2 + 2
< 4, (19)
𝑏22 (𝑥) 𝑢(32) 𝑏22 (𝑥) 𝑏33 (𝑥) 𝑏33 (𝑥) 𝑢(23)
≤ 𝑀1 [(𝑎1 (𝑥) − 𝑏11 (𝑥) 𝑀1 ] ≤ 0
for all 0 ≤ 𝑢2 ≤ 𝑀2 . So, (𝑀1 , 𝑟1 𝑤(𝑥)) is a set of upper and 𝑢(23) (𝑥), 𝑢̃(32) (𝑥)) of
then the steady state solution (̃
lower solutions for 𝑢1 in (8). system (10) is unique.
Similarly, choose 𝑟2 > 0 sufficiently small; (𝑀2 , 𝑟2 𝑤(𝑥)) is
a set of upper and lower solutions for 𝑢2 in (8). Remark 10. For fixed functions 𝑎𝑖 (𝑥) > 0, 𝑏𝑖𝑖 (𝑥) > 0,
By coupled upper and lower theorem (see [16, The- (𝑖 = 1, 2, 3), hypothesis (17) will be satisfied for 𝑏12 (𝑥),
orem 1.4-2]), the system (8) has a steady state solution 𝑏21 (𝑥) ≥ 0 sufficiently small. This is true because of Hopf ’s
𝑢(12) (𝑥), 𝑢̃(21) (𝑥)) with 𝑢̃(12) (𝑥) > 0, 𝑢̃(21) (𝑥) > 0 for 𝑥 ∈
(̃ strong maximal value theorem, and also because 𝑢(12) (resp.,
Ω. 𝑢(21) ) increases as 𝑏12 (𝑥) (resp., 𝑏21 (𝑥)) decreases for 𝑥 ∈
Ω. Thus (𝑏212
(𝑥)𝑢(2) /𝑏22
2
(𝑥)𝑢(12) ) (resp., 𝑏12
2
(𝑥)𝑢(1) /𝑏11
2
(𝑥)𝑢(21) )
Remark 8. By the comparison principle, it is easy to see that,
decreases as 𝑏12 (𝑥) (resp., 𝑏21 (𝑥)) decreases. Similarly,
in Ω, we have 𝑢̃(12) , 𝑢̃(13) < 𝑢1 < 𝑀1 , 𝑢̃(21) , 𝑢̃(23) < 𝑢2 < 𝑀2 , hypotheses (18) and (19) will be satisfied if 𝑏13 (𝑥), 𝑏31 (𝑥),
and 𝑢̃(31) , 𝑢̃(32) < 𝑢3 < 𝑀3 . 𝑏23 (𝑥), and 𝑏32 (𝑥) are sufficiently small.
Now, we consider the unique problem of the steady state Proof. We only give the proof of part (1) in Theorem 9, since
solutions above. the arguments of part (2) and part (3) are similar.
Let 𝜃𝜇,𝐴(𝑥) denote the unique positive solution of Assume that (̃ 𝑢(12) (𝑥), 𝑢̃(21) (𝑥)), (̃V(12) (𝑥), ̃V(21) (𝑥)) are two
strictly positive steady state solutions of system (8) in Ω.
𝜇𝑖 Δ𝜃 + 𝜃 (𝐴 (𝑥) − 𝜃) = 0 in Ω,
(15) Let
𝜃=0 on 𝜕Ω,
𝑝 (𝑥) = 𝑢̃(12) (𝑥) − ̃V(12) (𝑥) ,
for any 𝐴(𝑥) > 𝜇𝑖 𝜆 1 . For brevity, we denote
𝑞 (𝑥) = 𝑢̃(21) (𝑥) − ̃V(21) (𝑥)
𝑢(1) = 𝜃𝜇1 ,𝑎1 (𝑥) , 𝑢(2) = 𝜃𝜇2 ,𝑎2 (𝑥) , 𝑢(3) = 𝜃𝜇3 ,𝑎3 (𝑥) , (20)
𝐼12 = 𝑏11 (𝑥) ̃V(12) (𝑥) 𝑝 (𝑥) + 𝑏12 (𝑥) ̃V(12) (𝑥) 𝑞 (𝑥)
𝑢(12) = 𝜃𝜇1 ,(𝑎1 (𝑥)−𝑀2 𝑏12 (𝑥)) , 𝑢(21) = 𝜃𝜇2 ,(𝑎2 (𝑥)−𝑀1 𝑏21 (𝑥)) ,
𝐼21 = 𝑏21 (𝑥) 𝑢̃(21) (𝑥) 𝑝 (𝑥) + 𝑏22 (𝑥) 𝑢̃(21) (𝑥) 𝑞 (𝑥) ;
𝑢(13) = 𝜃𝜇1 ,(𝑎1 (𝑥)−𝑀3 𝑏13 (𝑥)) , 𝑢(31) = 𝜃𝜇3 ,(𝑎3 (𝑥)−𝑀1 𝑏31 (𝑥)) ,
then
𝑢(23) = 𝜃𝜇2 ,(𝑎2 (𝑥)−𝑀3 𝑏23 (𝑥)) , 𝑢(32) = 𝜃𝜇3 ,(𝑎3 (𝑥)−𝑀2 𝑏32 (𝑥)) .
(16)
𝜇1 Δ𝑝 (𝑥) + [𝑎1 (𝑥) − 𝑏11 (𝑥) 𝑢̃(12) (𝑥)
They are all positive functions in Ω.
−𝑏12 (𝑥) 𝑢̃(21) (𝑥)] 𝑝 (𝑥) − 𝐼12 = 0
Theorem 9. Assume that all the hypotheses of Theorem 7 are
satisfied. in Ω,
(1) If 𝜇2 Δ𝑞 (𝑥) + [𝑎2 (𝑥) − 𝑏21 (𝑥) ̃V(12) (𝑥) (21)
2
𝑏12 (𝑥) 𝑢(1) 2
𝑏12 (𝑥) 𝑏21 (𝑥) 𝑏21 (𝑥) 𝑢(2)
+ 2 + < 4, (17) −𝑏22 (𝑥) ̃V(21) (𝑥)] 𝑞 (𝑥) − 𝐼21 = 0
2 2
𝑏11 (𝑥) 𝑢(21) 𝑏11 (𝑥) 𝑏22 (𝑥) 𝑏22 (𝑥) 𝑢(12)
in Ω,
then the strictly positive steady state solution
𝑢(12) (𝑥), 𝑢̃(21) (𝑥)) of system (8) is unique; similarly
(̃ 𝑝 (𝑥) = 𝑞 (𝑥) = 0 on 𝜕Ω.
Abstract and Applied Analysis 5

Since 𝑢̃(12) (𝑥) is a strictly positive solution of in Ω. It follows from (17) that
(12)
𝜇1 Δ𝜓 + [𝑎1 (𝑥) − 𝑏11 (𝑥) 𝑢̃ (𝑥) 2 ̃V(12) (𝑥)
𝑏12 (𝑥) + 2𝑏12 (𝑥) 𝑏21 (𝑥)
𝑢̃(21) (𝑥)
−𝑏12 (𝑥) 𝑢̃(21) (𝑥)] 𝜓 + 𝛼𝜓 = 0 (29)
(22)
2 𝑢̃(21) (𝑥)
in Ω, + 𝑏21 (𝑥) (12) < 4𝑏11 (𝑥) 𝑏22 (𝑥) , in Ω.
̃V (𝑥)
𝜓=0 on 𝜕Ω, Then it is easy to see that the quadratic expression in the
with 𝛼 = 0, the number 𝛼 = 0 must be the smallest eigenvalue integrand of (26) is positive definite for each 𝑥 ∈ Ω.
of the above problem. Moreover, by variational properties, we Consequently, we must have 𝑝(𝑥) and 𝑞(𝑥) identically equal
have 𝑢(12) (𝑥), 𝑢̃(21) (𝑥)) ≡ (̃V(12) (𝑥), ̃V(21) (𝑥)) in
to zero in Ω. That is (̃
Ω.
∫ 𝑧 (−𝜇1 Δ𝑧 − [𝑎1 (𝑥) − 𝑏11 (𝑥) 𝑢̃(12) (𝑥)
Ω
(23) Now we are ready to develop the permanent coexistence
(21) results of the system (1). To construct the average Lyapunov
−𝑏12 (𝑥) 𝑢̃ (𝑥)] 𝑧) 𝑑𝑥 ≥ 0,
functions, we will use the positive eigenfunctions 𝜓1 , 𝜓2 , and
for any 𝑧 ∈ 𝐶2 (Ω) which vanishes on 𝜕Ω. Similarly, since 𝜓3 corresponding to the principal eigenvalues 𝜎1 , 𝜎2 , and 𝜎3
̃V(21) (𝑥) is strictly positive solution of of
𝜇2 Δ𝜓 + [𝑎2 (𝑥) − 𝑏21 (𝑥) ̃V(12) (𝑥) 𝜇1 Δ𝜓1 + 𝜓1 (𝑎1 (𝑥) − 𝑏12 (𝑥) 𝑢̃(23) − 𝑏13 (𝑥) 𝑢̃(32) )

−𝑏22 (𝑥) ̃V(21) (𝑥)] 𝜓 + 𝛼𝜓 = 0 = 𝜎𝜓1 in Ω, (30)

(24)
in Ω, 𝜓1 = 0 on 𝜕Ω,

𝜓=0 on 𝜕Ω, 𝜇2 Δ𝜓2 + 𝜓2 (𝑎2 (𝑥) − 𝑏21 (𝑥) 𝑢̃(13) − 𝑏23 (𝑥) 𝑢̃(31) )
with 𝛼 = 0, the number 𝛼 = 0 must be the smallest eigenvalue = 𝜎𝜓2 in Ω, (31)
of the above problem. Moreover,
𝜓2 = 0 on 𝜕Ω,
∫ 𝑧 (−𝜇2 Δ𝑧 − [𝑎2 (𝑥) − 𝑏21 (𝑥) ̃V(12) (𝑥)
Ω
(25) 𝜇3 Δ𝜓3 + 𝜓3 (𝑎3 (𝑥) − 𝑏31 (𝑥) 𝑢̃(12) − 𝑏32 (𝑥) 𝑢̃(21) )
(21)
−𝑏22 (𝑥) ̃V (𝑥)] 𝑧) 𝑑𝑥 ≥ 0,
= 𝜎𝜓3 in Ω, (32)
2
for any 𝑧 ∈ 𝐶 (Ω) which vanishes on 𝜕Ω. Multiplying the first
equation of (21) by −𝑝(𝑥), the second by −𝑞(𝑥), integrating 𝜓3 = 0 on 𝜕Ω,
over Ω, and adding, we deduce from (23) and (25) that
respectively.
∫ [𝑏11 (𝑥) ̃V(12) (𝑥) 𝑝2 (𝑥)
Ω Theorem 11. Suppose that (11)–(13), (17)–(19) hold, and the
principal eigenvalues 𝜎1 , 𝜎2 , and 𝜎3 of (30), (31), and (32),
+ (𝑏12 (𝑥) ̃V(12) (𝑥) + 𝑏21 (𝑥) 𝑢̃(21) (𝑥)) 𝑝 (𝑥) 𝑞 (𝑥) (26) respectively, are all positive. Assume also that
+𝑏22 (𝑥) 𝑢̃(21) (𝑥) 𝑞2 (𝑥)] 𝑑𝑥 ≤ 0. 𝜎1 + 𝑏12 (𝑥) 𝑢̃(23) + 𝑏13 (𝑥) 𝑢̃(32)
By comparison of scalar equations using upper and lower
> max {𝑏12 (𝑥) 𝑢2 , 𝑏13 (𝑥) 𝑢3 } ,
solutions we can readily obtain for 𝑖 = 1, 2, 𝑥 ∈ Ω,
1 1 𝜎2 + 𝑏21 (𝑥) 𝑢̃(13) + 𝑏23 (𝑥) 𝑢̃(31)
( ) 𝑢(12) ≤ 𝑢̃(12) , ̃V(12) ≤ ( ) 𝑢(1) , (33)
𝑏11 (𝑥) 𝑏11 (𝑥)
(27) > max {𝑏21 (𝑥) 𝑢1 , 𝑏23 (𝑥) 𝑢3 } ,
1 1
( ) 𝑢(21) ≤ 𝑢̃(21) , ̃V(21) ≤ ( ) 𝑢(2) . 𝜎3 + 𝑏31 (𝑥) 𝑢̃(12) + 𝑏32 (𝑥) 𝑢̃(21)
𝑏22 (𝑥) 𝑏22 (𝑥)
From (27), we have > max {𝑏31 (𝑥) 𝑢1 , 𝑏32 (𝑥) 𝑢2 } ,
2 (1) 2 (2)
𝑏22 (𝑥) 𝑏12 (𝑥) 𝑢 𝑏11 (𝑥) 𝑏21 (𝑥) 𝑢
+ 2𝑏12 (𝑥) 𝑏21 (𝑥) + for all 𝑥 ∈ Ω; then the semiflow on [𝐶01 (Ω)]3 generated
𝑏11 (𝑥) 𝑢(21) 𝑏22 (𝑥) 𝑢(12) by (1) under homogeneous Dirichlet boundary conditions is
̃V(12) (𝑥) 𝑢̃(21) (𝑥) permanent.
2 2
> 𝑏12 (𝑥) (21)
+ 2𝑏12 (𝑥) 𝑏21 (𝑥) + 𝑏21 (𝑥) (12)
𝑢̃ (𝑥) ̃V (𝑥) Proof. The hypotheses imply that (1) is dissipative and that
(28) the 𝜔-limit set in the boundary of the positive cone consists
6 Abstract and Applied Analysis

of the following seven cases: (0, 0, 0), (𝑢1 , 0, 0), (0, 𝑢2 , 0), We have for 𝑡 > 0
𝑢(12) , 𝑢̃(21) , 0), (̃
(0, 0, 𝑢3 ), (̃ 𝑢(13) , 0, 𝑢̃(31) ), and (0, 𝑢̃(23) , 𝑢̃(32) ).
Thus we need only find a suitable average Lyapunov function
to conclude permanence. Since the semiflow is dissipative,
we may restrict our attention to a bounded absorbing subset ∫ 𝜓1 𝑤1𝑡 𝑑𝑥
Ω
𝑋 of the positive cone in [𝐶01 (Ω)]3 that contains the global
attractor whose existence is asserted in Theorem 1. Let 𝑆 = ∫ 𝜓1 [𝜇1 Δ𝑤1 + 𝑤1 (𝑎1 (𝑥) − 𝑏11 (𝑥) 𝑤1
denote the intersection of 𝑋 with the boundary of the positive Ω
cone. Choose eigenfunctions 𝜓1 , 𝜓2 , 𝜓3 > 0 for (30), (31), and −𝑏12 (𝑥) 𝑤2 − 𝑏13 (𝑥) 𝑤3 )] 𝑑𝑥
(32), respectively, and define
= ∫ [(𝜇1 Δ𝜓1 ) 𝑤1 + (𝑎1 (𝑥) − 𝑏11 (𝑥) 𝑤1
Ω

𝑃 ((V1 , V2 , V3 )) −𝑏12 (𝑥) 𝑤2 − 𝑏13 (𝑥) 𝑤3 ) 𝜓1 𝑤1 ] 𝑑𝑥

= (∫ 𝜓1 V1 𝑑𝑥) (∫ 𝜓2 V2 𝑑𝑥) (∫ 𝜓3 V3 𝑑𝑥) = ∫ [𝜎1 − (𝑎1 (𝑥) − 𝑏12 (𝑥) 𝑢̃(23) − 𝑏13 (𝑥) 𝑢̃(32) )
Ω Ω Ω Ω
(34)
= exp [log ∫ 𝜓1 V1 𝑑𝑥 + log ∫ 𝜓2 V2 𝑑𝑥 + (𝑎1 (𝑥) − 𝑏11 (𝑥) 𝑤1 − 𝑏12 (𝑥) 𝑤2
Ω Ω
−𝑏13 (𝑥) 𝑤3 )] 𝜓1 𝑤1 𝑑𝑥
+ log ∫ 𝜓3 V3 𝑑𝑥] .
Ω = ∫ [𝜎1 − 𝑏11 (𝑥) 𝑤1 − 𝑏12 (𝑥) (𝑤2 − 𝑢̃(23) )
Ω

−𝑏13 (𝑥) (𝑤3 − 𝑢̃(32) )] 𝜓1 𝑤1 𝑑𝑥.

We have for (𝑢1 , 𝑢2 , 𝑢3 ) ∈ 𝑆 that (37)

𝑃 ((V1 , V2 , V3 ) ⋅ 𝑡) Similarly,
𝛼 (𝑡, (𝑢1 , 𝑢2 , 𝑢3 )) = lim inf (35)
(V1 ,V2 ,V3 ) → (𝑢1 ,𝑢2 ,𝑢3 ) 𝑃 ((V1 , V2 , V3 ))
(V1 ,V2 ,V3 )∈𝑋\𝑆

∫ 𝜓2 𝑤2𝑡 𝑑𝑥
Ω
we need sup𝑡>0 𝛼(𝑡, (𝑢1 , 𝑢2 , 𝑢3 )) > 0 for (𝑢1 , 𝑢2 , 𝑢3 ) ∈ 𝑆
and sup𝑡>0 𝛼(𝑡, (𝑢1 , 𝑢2 , 𝑢3 )) > 1 for (𝑢1 , 𝑢2 , 𝑢3 ) ∈ 𝜔(𝑆). Let = ∫ [𝜎2 − (𝑎2 (𝑥) − 𝑏21 (𝑥) 𝑢̃(13) − 𝑏23 (𝑥) 𝑢̃(31) )
(𝑤1 (𝑡), 𝑤2 (𝑡), 𝑤3 (𝑡)) = ((V1 , V2 , V3 )⋅𝑡). Computation using (34) Ω
yields + (𝑎2 (𝑥) − 𝑏22 (𝑥) 𝑤2 − 𝑏21 (𝑥) 𝑤1
(38)
−𝑏23 (𝑥) 𝑤3 ) ] 𝜓2 𝑤2 𝑑𝑥
𝑃 ((V1 , V2 , V3 ) ⋅ 𝑡)
𝑃 ((V1 , V2 , V3 )) = ∫ [𝜎2 − 𝑏22 (𝑥) 𝑤2 − 𝑏21 (𝑥) (𝑤1 − 𝑢̃(13) )
Ω

= exp [(log ∫ 𝜓1 𝑤1 𝑑𝑥|𝑡 − log ∫ 𝜓1 𝑤1 𝑑𝑥|0 ) −𝑏23 (𝑥) (𝑤3 − 𝑢̃(31) )] 𝜓2 𝑤2 𝑑𝑥,
Ω Ω

∫ 𝜓3 𝑤3𝑡 𝑑𝑥
+ (log ∫ 𝜓2 𝑤2 𝑑𝑥|𝑡 − log ∫ 𝜓2 𝑤2 𝑑𝑥|0 ) Ω
Ω Ω

= ∫ [𝜎3 − (𝑎3 (𝑥) − 𝑏31 (𝑥) 𝑢̃(12) − 𝑏32 (𝑥) 𝑢̃(21) )

+ (log ∫ 𝜓3 𝑤3 𝑑𝑥|𝑡 − log ∫ 𝜓3 𝑤3 𝑑𝑥|0 )] Ω
Ω Ω
+ (𝑎3 (𝑥) − 𝑏33 (𝑥) 𝑤3 − 𝑏31 (𝑥) 𝑤1
𝑡 ∫Ω 𝜓1 𝑤1𝑡 𝑑𝑥 𝑡 ∫Ω 𝜓2 𝑤2𝑡 𝑑𝑥 (39)
= exp [∫ ( ) +∫ ( )
0 ∫Ω 𝜓1 𝑤1 𝑑𝑥 0 ∫Ω 𝜓2 𝑤2 𝑑𝑥 −𝑏32 (𝑥) 𝑤2 ) ] 𝜓2 𝑤2 𝑑𝑥

𝑡 ∫Ω 𝜓3 𝑤3𝑡 𝑑𝑥 = ∫ [𝜎3 − 𝑏33 (𝑥) 𝑤3 − 𝑏31 (𝑥) (𝑤1 − 𝑢̃(12) )

+∫ ( )] . Ω
0 ∫Ω 𝜓3 𝑤3 𝑑𝑥
(36) −𝑏32 (𝑥) (𝑤2 − 𝑢̃(21) )] 𝜓3 𝑤3 𝑑𝑥.
Abstract and Applied Analysis 7

−1
Since 𝑋 is bounded, × (∫ 𝜓2 V2 𝑑𝑥)
𝜎1 − 𝑏11 (𝑥) 𝑤1 − 𝑏12 (𝑥) (𝑤2 − 𝑢̃(23) ) Ω

− 𝑏13 (𝑥) (𝑤3 − 𝑢̃(32) ) , + ∫ [𝜎3 − 𝑏33 (𝑥) V3 − 𝑏31 (𝑥) (V1 − 𝑢̃(12) )
Ω
(13)
𝜎2 − 𝑏22 (𝑥) 𝑤2 − 𝑏21 (𝑥) (𝑤1 − 𝑢̃ ) −𝑏32 (𝑥) (V2 − 𝑢̃(21) )] 𝜓3 V3 𝑑𝑥
(40)
− 𝑏23 (𝑥) (𝑤3 − 𝑢̃(31) ) , −1
× (∫ 𝜓3 V3 𝑑𝑥)
Ω
𝜎3 − 𝑏33 (𝑥) 𝑤3 − 𝑏31 (𝑥) (𝑤1 − 𝑢̃(12) )
(42)
(21)
− 𝑏32 (𝑥) (𝑤2 − 𝑢̃ ) always has a positive lim inf as (V1 , V2 , V3 ) → (𝑢1 , 𝑢2 , 𝑢3 ) ∈
𝜔(𝑆), then by (35)–(39), we have 𝛼(𝑡, (𝑢1 , 𝑢2 , 𝑢3 )) > 1 for
are bounded below on 𝑋 \ 𝑆, so that the ratios
(𝑢1 , 𝑢2 , 𝑢3 ) ∈ 𝜔(𝑆) and 𝑡 sufficiently small.
(∫Ω 𝜓𝑖 𝑤𝑖𝑡 𝑑𝑥/ ∫Ω 𝜓𝑖 𝑤𝑖 𝑑𝑥) are bounded below. It follows
If we let (V1 , V2 , V3 ) → (0, 0, 0), then
that 𝑃((V1 , V2 , V3 ) ⋅ 𝑡)/𝑃((V1 , V2 , V3 )) has a strictly positive
lower bounded since the expression inside the exponential 𝜎1 − 𝑏11 (𝑥) V1 − 𝑏12 (𝑥) (V2 − 𝑢̃(23) ) − 𝑏13 (𝑥) (V3 − 𝑢̃(32) )
in (36) is bounded away from −∞. To see what happens as
(V1 , V2 , V3 ) → 𝜔(𝑆), we must examine how 󳨀→ 𝜎1 + 𝑏12 (𝑥) 𝑢̃(23) + 𝑏13 (𝑥) 𝑢̃(32) ≥ 𝜎1 > 0,
∫ [𝜎1 − 𝑏11 (𝑥) 𝑤1 − 𝑏12 (𝑥) (𝑤2 − 𝑢̃(23) ) 𝜎2 − 𝑏22 (𝑥) V2 − 𝑏21 (𝑥) (V1 − 𝑢̃(13) ) − 𝑏23 (𝑥) (V3 − 𝑢̃(31) )
Ω

−𝑏13 (𝑥) (𝑤3 − 𝑢̃(32) )] 𝜓1 𝑤1 𝑑𝑥 󳨀→ 𝜎2 + 𝑏21 (𝑥) 𝑢̃(13) + 𝑏23 (𝑥) 𝑢̃(31) ≥ 𝜎2 > 0,
−1
𝜎3 − 𝑏33 (𝑥) V3 − 𝑏31 (𝑥) (V1 − 𝑢̃(12) ) − 𝑏32 (𝑥) (V2 − 𝑢̃(21) )
× (∫ 𝜓1 𝑤1 𝑑𝑥)
Ω
󳨀→ 𝜎3 + 𝑏31 (𝑥) 𝑢̃(12) + 𝑏32 (𝑥) 𝑢̃(21) ≥ 𝜎3 > 0.
(13)
+ ∫ [𝜎2 − 𝑏22 (𝑥) 𝑤2 − 𝑏21 (𝑥) (𝑤1 − 𝑢̃ ) (43)
Ω
Then
−𝑏23 (𝑥) (𝑤3 − 𝑢̃(31) )] 𝜓2 𝑤2 𝑑𝑥 (41)
lim inf 𝜎∗ (V1 , V2 , V3 ) ≥ 𝜎1 + 𝜎2 + 𝜎3 > 0.
−1
(V1 ,V2 ,V3 ) → (0,0,0) (44)
(V1 ,V2 ,V3 )∈𝑋\𝑆
× (∫ 𝜓2 𝑤2 𝑑𝑥)
Ω As (V1 , V2 , V3 ) → (𝑢1 , 0, 0),
+ ∫ [𝜎3 − 𝑏33 (𝑥) 𝑤3 − 𝑏31 (𝑥) (𝑤1 − 𝑢̃(12) ) ∫ [𝜎1 − (𝑎1 (𝑥) − 𝑏12 (𝑥) 𝑢̃(23) − 𝑏13 (𝑥) 𝑢̃(32) )
Ω
Ω
(21)
−𝑏32 (𝑥) (𝑤2 − 𝑢̃ )] 𝜓3 𝑤3 𝑑𝑥 + (𝑎1 (𝑥) − 𝑏11 (𝑥) V1 − 𝑏12 (𝑥) V2
−1
× (∫ 𝜓3 𝑤3 𝑑𝑥) −𝑏13 (𝑥) V3 ) ] 𝜓1 V1 𝑑𝑥 (45)
Ω
behaves as (V1 , V2 , V3 ) → (𝑢1 , 𝑢2 , 𝑢3 ) ∈ 𝜔(𝑆), (V1 , V2 , V3 ) ∈ 󳨀→ ∫ [𝜎1 − (𝑎1 (𝑥) − 𝑏12 (𝑥) 𝑢̃(23) − 𝑏13 (𝑥) 𝑢̃(32) )
𝑋 \ 𝑆. By the continuity of the semiflow 𝜋, 𝑤1 , 𝑤2 , and 𝑤3 Ω
will be uniformly close to V1 , V2 , and V3 , respectively, for 𝑡 > 0 + (𝑎1 (𝑥) − 𝑏11 (𝑥) 𝑢1 ) ] 𝜓1 𝑢1 𝑑𝑥.
sufficiently small. Hence, if we can show that the expression
𝜎∗ (V1 , V2 , V3 ) Also

∫ (𝑎1 (𝑥) − 𝑏11 (𝑥) 𝑢1 ) 𝜓1 𝑢1 𝑑𝑥

= ∫ [𝜎1 − 𝑏11 (𝑥) V1 − 𝑏12 (𝑥) (V2 − 𝑢̃(23) ) Ω
Ω
= − ∫ 𝜓1 𝜇1 Δ𝑢1 𝑑𝑥
−𝑏13 (𝑥) (V3 − 𝑢̃(32) )] 𝜓1 V1 𝑑𝑥 Ω

−1 (46)
× (∫ 𝜓1 V1 𝑑𝑥) = − ∫ 𝑢1 𝜇1 Δ𝜓1 𝑑𝑥
Ω
Ω

+ ∫ [𝜎2 − 𝑏22 (𝑥) V2 − 𝑏21 (𝑥) (V1 − 𝑢̃(13) ) = − ∫ 𝑢1 𝜓1 [𝜎1 − (𝑎1 (𝑥) − 𝑏12 (𝑥) 𝑢̃(23)
Ω
Ω

−𝑏23 (𝑥) (V3 − 𝑢̃(31) )] 𝜓2 V2 𝑑𝑥 −𝑏13 (𝑥) 𝑢̃(32) )] 𝑑𝑥

8 Abstract and Applied Analysis

so that as (V1 , V2 , V3 ) → (𝑢1 , 0, 0), 𝑢(12) , 𝑢̃(21) , 0),

As (V1 , V2 , V3 ) → (̃

∫ [𝜎1 − 𝑏11 (𝑥) V1 − 𝑏12 (𝑥) (V2 − 𝑢̃(23) )

Ω
(47)
∫ [𝜎1 − (𝑎1 (𝑥) − 𝑏12 (𝑥) 𝑢̃(23) − 𝑏13 (𝑥) 𝑢̃(32) )
−𝑏13 (𝑥) (V3 − 𝑢̃(32) )] 𝜓1 V1 𝑑𝑥 󳨀→ 0. Ω

+ (𝑎1 (𝑥) − 𝑏11 (𝑥) V1 − 𝑏12 (𝑥) V2

Since ∫Ω 𝜓1 V1 𝑑𝑥 → ∫Ω 𝜓1 𝑢1 𝑑𝑥 > 0, the first ratio of integrals
in (42) has limit 0 as (V1 , V2 , V3 ) → (𝑢1 , 0, 0). For the latter −𝑏13 (𝑥) V3 ) ] 𝜓1 V1 𝑑𝑥
two ratios of integrals, note that as (V1 , V2 , V3 ) → (𝑢1 , 0, 0).
By the assumption (33), we have
󳨀→ ∫ [𝜎1 − (𝑎1 (𝑥) − 𝑏12 (𝑥) 𝑢̃(23) − 𝑏13 (𝑥) 𝑢̃(32) )
Ω
𝜎2 − 𝑏21 (𝑥) (𝑢1 − 𝑢̃(13) ) + 𝑏23 (𝑥) 𝑢̃(31) > 0,
(48) + (𝑎1 (𝑥) − 𝑏11 (𝑥) 𝑢̃(12) −𝑏12 (𝑥) 𝑢̃(21) )] 𝜓1 𝑢̃(12) 𝑑𝑥.
𝜎3 − 𝑏31 (𝑥) (𝑢1 − 𝑢̃(12) ) + 𝑏32 (𝑥) 𝑢̃(21) > 0. (56)
Hence we have
lim inf 𝜎∗ (V1 , V2 , V3 ) > 0. Also
(V1 ,V2 ,V3 ) → (𝑢1 ,0,0) (49)
(V1 ,V2 ,V3 )∈𝑋\𝑆

As (V1 , V2 , V3 ) → (0, 𝑢2 , 0), computations similar to (45) and

(46) show that ∫ [𝑎1 (𝑥) − 𝑏11 (𝑥) 𝑢̃(12) − 𝑏12 (𝑥) 𝑢̃(21) ] 𝜓1 𝑢̃(12) 𝑑𝑥
Ω
(13)
∫ [𝜎2 − 𝑏22 (𝑥) V2 − 𝑏21 (𝑥) (V1 − 𝑢̃ ) 𝑢(12) 𝑑𝑥
= − ∫ 𝜓1 𝜇1 Δ̃
Ω
Ω
−𝑏23 (𝑥) (V3 − 𝑢̃(31) )] 𝜓2 V2 𝑑𝑥 (50)
= − ∫ 𝑢̃(12) 𝜇1 Δ𝜓1 𝑑𝑥 (57)
−1 Ω
× (∫ 𝜓3 V3 𝑑𝑥) 󳨀→ 0.
Ω = − ∫ 𝑢̃(12) 𝜓1 [𝜎1 − (𝑎1 (𝑥) − 𝑏12 (𝑥) 𝑢̃(23)
Ω
By assumption (33), we have
−𝑏13 (𝑥) 𝑢̃(32) )] 𝑑𝑥
𝜎1 − 𝑏12 (𝑥) (𝑢2 − 𝑢̃(23) ) + 𝑏13 (𝑥) 𝑢̃(32) > 0,
(51)
𝜎3 + 𝑏31 (𝑥) 𝑢̃(12) − 𝑏32 (𝑥) (𝑢2 − 𝑢̃(21) ) > 0.
𝑢(12) , 𝑢̃(21) , 0),
so that as (V1 , V2 , V3 ) → (̃
Hence we have
lim inf 𝜎∗ (V1 , V2 , V3 ) > 0.
(V1 ,V2 ,V3 ) → (0,𝑢2 ,0) (52)
(V1 ,V2 ,V3 )∈𝑋\𝑆 ∫ [𝜎1 − 𝑏11 (𝑥) V1 − 𝑏12 (𝑥) (V2 − 𝑢̃(23) )
Ω
As (V1 , V2 , V3 ) → (0, 0, 𝑢3 ), computations similar to (45) and
(46) show that −𝑏13 (𝑥) (V3 − 𝑢̃(32) )] 𝜓1 V1 𝑑𝑥 (58)
−1
∫ [𝜎3 − 𝑏33 (𝑥) V3 − 𝑏31 (𝑥) (V1 − 𝑢̃(12) ) × (∫ 𝜓1 V1 𝑑𝑥) 󳨀→ 0.
Ω Ω

−𝑏32 (𝑥) (V2 − 𝑢̃(21) )] 𝜓3 V3 𝑑𝑥 (53)

−1 For the second ratio of integrals,
× (∫ 𝜓3 V3 𝑑𝑥) 󳨀→ 0.
Ω

By assumption (33), we get ∫ [𝜎2 − (𝑎2 (𝑥) − 𝑏21 (𝑥) 𝑢̃(13) − 𝑏23 (𝑥) 𝑢̃(31) )
Ω
𝜎1 + 𝑏12 (𝑥) 𝑢̃(23) − 𝑏13 (𝑥) (𝑢3 − 𝑢̃(32) ) > 0,
(54) + (𝑎2 (𝑥) − 𝑏21 (𝑥) V1 − 𝑏22 (𝑥) V2 − 𝑏23 (𝑥) V3 )] 𝜓2 V2 𝑑𝑥
(13) (31)
𝜎2 + 𝑏21 (𝑥) 𝑢̃ − 𝑏23 (𝑥) (𝑢3 − 𝑢̃ ) > 0.
󳨀→ ∫ [𝜎2 − (𝑎2 (𝑥) − 𝑏21 (𝑥) 𝑢̃(13) − 𝑏23 (𝑥) 𝑢̃(31) )
Hence we have Ω
∗
lim inf 𝜎 (V1 , V2 , V3 ) > 0. + (𝑎2 (𝑥) − 𝑏21 (𝑥) 𝑢̃(12) − 𝑏22 (𝑥) 𝑢̃(21) )] 𝜓2 𝑢̃(21) 𝑑𝑥.
(V1 ,V2 ,V3 ) → (0,0,𝑢3 ) (55)
(V1 ,V2 ,V3 )∈𝑋\𝑆 (59)
Abstract and Applied Analysis 9

Also Theorem 12. Suppose that 𝑎𝑖 (𝑥) > 𝜇𝑖 𝜆 1 , 𝑏𝑖𝑖 (𝑥) > 0 for 𝑥 ∈ Ω,
𝑖 = 1, 2, 3. Then there is a constant 𝛿 > 0, such that if 𝑏𝑖𝑗 < 𝛿
∫ [𝑎2 (𝑥) − 𝑏21 (𝑥) 𝑢̃(12) − 𝑏22 (𝑥) 𝑢̃(21) ] 𝜓2 𝑢̃(12) 𝑑𝑥 for 𝑖, 𝑗 = 1, 2, 3 with 𝑖 ≠𝑗, the semiflow on [𝐶01 (Ω)]3 generated
Ω
by (1) under homogeneous Dirichlet boundary conditions is
𝑢(21) 𝑑𝑥
= − ∫ 𝜓2 𝜇2 Δ̃ permanent.
Ω
Proof. For fixed 𝑎𝑖 > 𝜇𝑖 𝜆 1 and 𝑏𝑖𝑖 (𝑥) > 0 in Ω, 𝑖 = 1, 2, 3,
= − ∫ 𝑢̃(21) 𝜇2 Δ𝜓2 𝑑𝑥 (60) when 𝑏𝑖𝑗 ≥ 0 (𝑖, 𝑗 = 1, 2, 3, 𝑖 ≠𝑗) are sufficiently small,
Ω
conditions (11)–(13) and (17)–(19) are satisfied, and by the
variational expression of principal eigenvalues (see [13]) and
= − ∫ 𝑢̃(21) 𝜓2 [𝜎2 − (𝑎2 (𝑥) − 𝑏21 (𝑥) 𝑢̃(13)
Ω the boundedness of 𝑢̃(𝑖𝑗) (𝑖 ≠𝑗) (see Remark 8), the principal
eigenvalues 𝜎1 , 𝜎2 , and 𝜎3 of (30), (31), and (32), respectively,
−𝑏23 (𝑥) 𝑢̃(31) )] 𝑑𝑥 are all positive. And then it is easy to see that condition (33)
is satisfied when 𝑏𝑖𝑗 (𝑖 ≠𝑗) are even smaller, if necessary. The
𝑢(12) , 𝑢̃(21) , 0),
so that as (V1 , V2 , V3 ) → (̃ proof is completed by Theorem 11.

∫ [𝜎2 − 𝑏22 (𝑥) V2 − 𝑏21 (𝑥) (V1 − 𝑢̃(13) )

Ω 4. Summary and Discussion
(31)
−𝑏23 (𝑥) (V3 − 𝑢̃ )] 𝜓2 V2 𝑑𝑥 (61) The important aspect of the approach of our main result,
−1 Theorem 11, is that it produces conditions that in terms of
× (∫ 𝜓2 V2 𝑑𝑥) 󳨀→ 0. coefficients describing the strength of interactions. The crite-
Ω rions depend only on the behavior near the species equilibria
of the two-dimensional subsystems, which is relatively easy
𝑢(12) ,
For the third ratio of integrals, as (V1 , V2 , V3 ) → (̃
(21) to examine experimentally. The eigenvalues depending on
𝑢̃ , 0), the coefficients can be computed explicitly in simple cases,
𝜎3 − 𝑏33 (𝑥) V3 − 𝑏31 (𝑥) (V1 − 𝑢̃(12) ) − 𝑏32 (𝑥) (V2 − 𝑢̃(21) ) and there is a huge literature available on their estimation,
numerical approximation, qualitative properties, and so on.
󳨀→ 𝜎3 > 0. Eigenvalue-based criteria for coexistence can then be used to
(62) reach biological conclusions that have been difficult to deduce
through other methods.
Hence we have Theorem 12 tell us that if the intercompeting coefficients
are sufficiently small, the system (1) is always permanent,
lim inf 𝜎∗ (V1 , V2 , V3 ) ≥ 𝜎3 > 0.
(V1 ,V2 ,V3 ) → (̃𝑢(12) ,̃𝑢(12) ,0) (63) provided that 𝑎𝑖 > 𝜇𝑖 𝜆 1 in Ω, which is, by Lemma 5, every
(V1 ,V2 ,V3 )∈𝑋\𝑆 single species can survive when the other two species become
extinct. This sounds reasonable, and we guess that this is true
Using the same method, we can get the conclusion about the
for systems of any finite multiple competing species. A related
other two cases.
topic on the systems of finite multiple competing species with
𝑢(13) , 0, 𝑢̃(31) ),
As (V1 , V2 , V3 ) → (̃
constant coefficients and Neumann boundary condition can
lim inf 𝜎∗ (V1 , V2 , V3 ) ≥ 𝜎2 > 0. be found in [17], Theorem 9.
(V1 ,V2 ,V3 ) → (̃𝑢(13) ,0,̃𝑢(31) ) (64) The significant feature of our conditions for permanence
(V1 ,V2 ,V3 )∈𝑋\𝑆
is that they do not require any special assumptions the
and as (V1 , V2 , V3 ) → (0, 𝑢̃(23) , 𝑢̃(32) ), uniqueness of coexistence states, the existence of a global
Lyapunov function, or other such properties, and so the
lim inf 𝜎∗ (V1 , V2 , V3 ) ≥ 𝜎1 > 0. methods used here can be used to other more general
(V1 ,V2 ,V3 ) → (0,̃𝑢(23) ,̃𝑢(32) ) (65) situations.
(V1 ,V2 ,V3 )∈𝑋\𝑆

In conclusion
Conflict of Interests
lim inf 𝜎∗ (V1 , V2 , V3 ) > 0
(V1 ,V2 ,V3 ) → 𝜔(𝑆) (66) The authors declare that there is no conflict of interests
(V1 ,V2 ,V3 )∈𝑋\𝑆
regarding the publication of this paper.
so that 𝛼(𝑡, (𝑢1 , 𝑢2 , 𝑢3 )) > 1 for (𝑢1 , 𝑢2 , 𝑢3 ) ∈ 𝜔(𝑆) and 𝑡
sufficiently small, and permanence follows from the abstract
results of the previous section. Acknowledgments
Now we are ready to prove an interesting result, namely, The authors thank the anonymous reviewers for their helpful
permanent existence always holds among three “very weak” comments and suggestions. This work was supported by
competing species. The result may be known, but to our Grant 13ZR1430100 from the Natural Science Foundation of
knowledge, it does not appear in the published literature. Shanghai, China.
10 Abstract and Applied Analysis

References
[1] E. N. Dancer and Y. H. Du, “Positive solutions for a three-
species competition system with diffusion. I. General existence
results,” Nonlinear Analysis: Theory, Methods & Applications,
vol. 24, no. 3, pp. 337–357, 1995.
[2] E. N. Dancer and Y. H. Du, “Positive solutions for a three-
species competition system with diffusion, part II. The case
of equal birth rates,” Nonlinear Analysis: Theory, Methods &
Applications, vol. 24, no. 3, pp. 359–373, 1995.
[3] K. Kishimoto, “The diffusive Lotka-Volterra system with three
species can have a stable nonconstant equilibrium solution,”
Journal of Mathematical Biology, vol. 16, no. 1, pp. 103–112, 1982.
[4] Y. Lou and D. Munther, “Dynamics of a three species compe-
tition model,” Discrete and Continuous Dynamical Systems, vol.
32, no. 9, pp. 3099–3131, 2012.
[5] O. Vasilyeva and F. Lutscher, “Competition of three species
in an advective environment,” Nonlinear Analysis: Real World
Applications, vol. 13, no. 4, pp. 1730–1748, 2012.
[6] R. M. May and W. J. Leonard, “Nonlinear aspects of competition
between three species,” SIAM Journal on Applied Mathematics,
vol. 29, no. 2, pp. 243–253, 1975.
[7] P. van den Driessche and M. L. Zeeman, “Three-dimensional
competitive Lotka-Volterra systems with no periodic orbits,”
SIAM Journal on Applied Mathematics, vol. 58, no. 1, pp. 227–
234, 1998.
[8] V. Hutson and R. Law, “Permanent coexistence in general
models of three interacting species,” Journal of Mathematical
Biology, vol. 21, no. 3, pp. 285–298, 1985.
[9] D. Xiao and W. Li, “Limit cycles for the competitive three
dimensional Lotka-Volterra system,” Journal of Differential
Equations, vol. 164, no. 1, pp. 1–15, 2000.
[10] R. S. Cantrell, C. Cosner, and V. Hutson, “Ecological models,
permanence and spatial heterogeneity,” Rocky Mountain Journal
of Mathematics, vol. 26, no. 1, pp. 1–35, 1996.
[11] R. S. Cantrell, C. Cosner, and V. Hutson, “Permanence in some
diffusive Lotka-Volterra models for three interacting species,”
Dynamic Systems and Applications, vol. 2, no. 4, pp. 505–530,
1993.
[12] R. S. Cantrell, C. Cosner, and V. Hutson, “Permanence in
ecological systems with spatial heterogeneity,” Proceedings of the
Royal Society of Edinburgh A: Mathematics, vol. 123, no. 3, pp.
533–559, 1993.
[13] R. S. Cantrell and C. Cosner, Spatial Ecology via Reaction-
Diffusion Equations, John Wiley & Sons, New York, NY, USA,
2003.
[14] A. W. Leung, Nonlinear Systems of Partial Differential Equations:
Applications to Life and Physical Sciences, World Scientific,
Singapore, 2009.
[15] V. Hutson, “A theorem on average Liapunov functions,” Monat-
shefte für Mathematik, vol. 98, no. 4, pp. 267–275, 1984.
[16] A. W. Leung, Systems of Nonlinear Partial Differential Equations:
Applications to Biology and Engineering, vol. 49 of Mathematics
and Its Applications, Kluwer Academic Publishers, Dordrecht,
The Netherlands, 1989.
[17] P. N. Brown, “Decay to uniform states in ecological interac-
tions,” SIAM Journal on Applied Mathematics, vol. 38, no. 1, pp.
22–37, 1980.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 132781, 12 pages
http://dx.doi.org/10.1155/2014/132781

Research Article
Global Dynamics of an HTLV-1 Model with Cell-to-Cell
Infection and Mitosis

Sumei Li and Yicang Zhou

Department of Applied Mathematics, Xi’an Jiaotong University, Xi’an 710049, China

Correspondence should be addressed to Yicang Zhou; zhouyc@mail.xjtu.edu.cn

Received 26 May 2014; Accepted 19 July 2014; Published 12 August 2014

Academic Editor: Sanling Yuan

Copyright © 2014 S. Li and Y. Zhou. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

A mathematical model of human T-cell lymphotropic virus type 1 in vivo with cell-to-cell infection and mitosis is formulated and
studied. The basic reproductive number 𝑅0 is derived. It is proved that the dynamics of the model can be determined completely
by the magnitude of 𝑅0 . The infection-free equilibrium is globally asymptotically stable (unstable) if 𝑅0 < 1 (𝑅0 > 1). There exists
a chronic infection equilibrium and it is globally asymptotically stable if 𝑅0 > 1.

1. Introduction To persist within the host, HTLV-1 requires two routes:

(i) infectious spread to uninfected cells via cell-to-cell contact
Human T-cell lymphotropic virus type 1 (HTLV-1) is a known as the virological synapse, cellular conduits, extracel-
pathogenic retrovirus and persists indefinitely in the infected lular viral assemblies, and transinfection via dendritic cells
hosts [1–4]. There are approximately 10–20 million infected [9] and (ii) clonal expansion, which would actively promote
people worldwide [5, 6]. HTLV-1 is associated causatively mitotic proliferation of infected cells, and pass on the provirus
with a large number of pathologies. A slowly progressive to daughter cells. It is assumed that infection of an individual
neurologic disease HTLV-1 associated myelopathy/tropical with HTLV-1 occurs in two stages; the virus is thought to
spastic paraparesis (HAM/TSP) [7] and adult T-cell leukemia initially spread from T cells to T cells, primarily 𝐶𝐷4+ T cells,
(ATL) are two most common forms of the disease [8]. and later to persist by clonal expansion of infected cells [12].
The majority of HTLV-1 infected individuals remain lifelong It has been observed that HTLV-1 infection has a lower
asymptomatic carriers (ACs). The remaining 0.25–3% of rate of proviral genetic variation than HIV infection, which
infected individuals develop into HAM/TSP [9]. The virus suggests that the vertical transmission through mitotic divi-
can be transmitted from mother to child, through sexual sion rather than horizontal transmission through cell-to-cell
contact, and by needle sharing and contaminated blood contact plays an important role [14, 15]. HTLV-1 succeeds in
products [5, 9]. causing a persistent infection with a high proviral load and
In HTLV-1 infection, the initial infection is subclinical. remains approximately stable in one individual over years. In
The virus preferentially integrates into the genome of host order to identify the underlying mechanism of HTLV-1 per-
T lymphocytes. Since the virions are almost undetectable sistence in vivo and the key factors determining the HTLV-1
from extracellular matrix, the viral burden is quantified as the provirus load and the disease risk, Asquith and Bangham [1]
proportion of peripheral blood mononuclear cells that carry have used a combination of mathematical and experimental
an integrated HTLV-1 provirus. About 90–95% of the proviral techniques to propose a model of HTLV-1 persistence. Mitosis
load in chronic HTLV-1 infection is carried by 𝐶𝐷4+ T cells is the main route of viral replication, and the expression of
and 5–10% by 𝐶𝐷8+ T cells [10–13]. HTLV-1 proteins, particularly Tax, is required to promote the
2 Abstract and Applied Analysis

selective expansion of cells that harbour a provirus [5, 16– 𝜇1 x 𝜇2 u 𝜇3 y

18], though the majority of infected cells are not expressing
viral protein. Although the Tax expressing is silenced in the
𝜆 𝜎𝛽xy 𝜏u
majority of surviving cells and a small proportion (0.03%– x u y
3%) of infected cells can express Tax, the cells with Tax 𝜀sy
expression proliferate more rapidly than silently infected and sy
uninfected cells, leading to the selective expansion of infected (1 − 𝜎)𝛽xy
cells and an increase in proviral load [1]. The small proportion
of infected cells that express viral proteins play a crucial Figure 1: The schematic diagram of the HTLV-1 infection in vivo.
role, and the very high provirus load in HTLV-1 infection is
maintained by proliferation of infected T cells, induced by the
Tax protein of HTLV-1 [19]. coefficient among 𝐶𝐷4+ T cells [25]. The newly infected
It has been observed that the 𝐶𝐷4+ T cells population cells experience an irreparable destruction by the strong
from HAM/TSP patients express higher levels of tax mRNA adaptive immune responses. As a result, a small fraction
than 𝐶𝐷4+ T cells from ACs. Tax expression at any given 𝜎𝛽𝑥𝑦, 𝜎 ∈ (0, 1), survives after the immune attack and
proviral load is significantly higher in the HAM/TSP patients becomes the resting infected cells [14, 22]. Every day, a
than that in the ACs [20]; thus a high rate of viral protein small proportion 𝜏 of resting infected cells express Tax with
expression is associated with a large increase in the prevalence 𝜏 ∈ (0.3%, 3%) [26]. The mitotic transmission of HTLV-1
of HAM/TSP, and Tax expression is a significant predictor of involving selective clonal expansion of these Tax-expressing
the disease [1]. 𝐶𝐷4+ T cells occurs at a rate 𝑠. The newly infected cells from
Most of the existing models have considered the persis- mitosis to the resting infected cells compartment are 𝜀𝑠𝑦,
tence and pathogenesis for HTLV-1 infection of 𝐶𝐷4+ T cells. 𝜀 ∈ (0, 1), with (1−𝜀)𝑠𝑦 staying in the Tax-expressing infected
Mathematical models that take into account both infectious 𝐶𝐷4+ T-cell compartment. The transfers among those three
and mitotic routes have also been developed to describe the compartments are shown in Figure 1.
interaction in vivo among HTLV-1 [14, 20–22]. Motivated From the mechanism of the HTLV-1 infection and the
by the new hypothesis of HTLV-1 infection by Asquith and schematic diagram we can have the following model consist-
Bangham, we construct a model with three compartments, ing of three differential equations;
healthy 𝐶𝐷4+ T cells 𝑥, resting infected 𝐶𝐷4+ T cells 𝑢, and
Tax-expressing infected 𝐶𝐷4+ T cells 𝑦, to investigate the 𝑑𝑥
= 𝜆 − 𝛽𝑥𝑦 − 𝜇1 𝑥,
dynamics of the HTLV-1 infection. The model is formulated 𝑑𝑡
and the required conditions are given in Section 2. The sta- 𝑑𝑢
bility of equilibria is presented in Section 3. The simulations = 𝜎𝛽𝑥𝑦 + 𝜀𝑠𝑦 − 𝜏𝑢 − 𝜇2 𝑢, (1)
𝑑𝑡
are done in Section 4. The concluding remarks are given in
Section 5. 𝑑𝑦
= 𝜏𝑢 + (1 − 𝜀) 𝑠𝑦 − 𝜇3 𝑦.
𝑑𝑡
In model (1), 𝜇1 , 𝜇2 , and 𝜇3 are the removal rate of healthy
2. Model Formulation 𝐶𝐷4+ T cells, resting infected 𝐶𝐷4+ T cells, and Tax-
In this section, we construct a mathematical model including expressing infected 𝐶𝐷4+ T cells, respectively. From epidemi-
the spontaneous HTLV-1 antigen Tax expression, cell-to-cell ological background, it is natural to assume that the initial
contact, and mitotic infectious routes to describe the viral values of these variables and parameters are nonnegative.
dynamics. Let 𝑥(𝑡) be the number of healthy 𝐶𝐷4+ T cells We define the basic reproductive number of model (1) by
at time 𝑡, let 𝑢(𝑡) be the number of the resting infected the next generation matrix approach given in [27]. Let
𝐶𝐷4+ T cells at time 𝑡, and let 𝑦(𝑡) be the number of Tax- 0 𝜎𝛽𝑥 + 𝜀𝑠 𝜏 + 𝜇2 0
expressing infected 𝐶𝐷4+ T cells at time 𝑡. We consider only 𝐹=[
0 𝑠 − 𝜀𝑠
], 𝑉=[
−𝜏 𝜇3
]. (2)
HAM/TSP among nonmalignant HTLV-1 infection diseases;
the dynamics of ATL and other aggressive malignancies may The calculation shows that the spectral radius (the basic
be very different. Although mitosis occurs in all 𝐶𝐷4+ T cells reproductive number) of 𝐹𝑉−1 is
as a natural process, normal homeostatic proliferation occurs
at a very slower rate than that of selective mitotic division in 𝜎𝛽𝜏𝜆 𝜏𝜀𝑠 (1 − 𝜀) 𝑠
𝑅0 = 𝜌 (𝐹𝑉−1 ) = + + .
Tax-expressing infected cells. We ignore the effects of passive (𝜏 + 𝜇2 ) 𝜇1 𝜇3 (𝜏 + 𝜇2 ) 𝜇3 𝜇3
homeostatic proliferation of the healthy and resting infected (3)
𝐶𝐷4+ T cells to simplify the model.
Healthy 𝐶𝐷4+ T cells are produced in bone marrow at The basic reproductive number, 𝑅0 , gives the average number
a constant rate 𝜆 [23, 24]; we assume that the new cells of the secondary infections caused by a single Tax-expressing
generated in the bone marrow are uninfected. The infected infected 𝐶𝐷4+ T cell during its whole infectious period.
𝐶𝐷4+ T cells can make the healthy 𝐶𝐷4+ T cells get infected The secondary infection caused by a single Tax-expressing
through cell-to-cell contact. The infectious incidence is infected 𝐶𝐷4+ T cell through horizontal transmission is 𝜎𝛽 ⋅
described by a bilinear term 𝛽𝑥𝑦, where 𝛽 is the transmission (𝜆/𝜇1 )⋅(𝜏/(𝜏+𝜇2 ))⋅(1/𝜇3 ); the secondary infection caused by
Abstract and Applied Analysis 3

a single Tax-expressing infected 𝐶𝐷4+ T cell through mitotic When 𝑥 + ((𝜏 + 𝜇2 )/𝜏)𝑦 ≤ 𝐿 0 holds, the second equation of
transmission is 𝜀𝑠 ⋅ (𝜏/(𝜏 + 𝜇2 )) ⋅ (1/𝜇3 ) + (1 − 𝜀)𝑠 ⋅ (1/𝜇3 ). model (1) yields
Throughout the paper, we use the assumption
𝑑𝑢
= 𝜎𝛽𝑥𝑦 + 𝜀𝑠𝑦 − 𝜏𝑢 − 𝜇2 𝑢 ≤ 𝑀0 − (𝜏 + 𝜇2 ) 𝑢. (9)
(𝜏 + 𝜇2 ) 𝜇3 𝑑𝑡
𝑠< . (A1)
𝜏 + 𝜇2 (1 − 𝜀) From the comparison principle and (9), it follows that there
exists a positive 𝑢𝑚 = 𝑀0 /(𝜏 + 𝜇2 ), such that 𝑑𝑢/𝑑𝑡 ≤ 0 when
The inequality (A1) is equivalent to that 𝜏𝜀𝑠/(𝜏 + 𝜇2 )𝜇3 +
𝑢 > 𝑢𝑚 and 𝑥 + ((𝜏 + 𝜇2 )/𝜏)𝑦 ≤ 𝐿 0 .
(1 − 𝜀)𝑠/𝜇3 < 1, which requires that the average number of
For any given initial values 𝑥(0) = 𝑥0 ≥ 0, 𝑢(0) = 𝑢0 ≥
the secondary infections by a single Tax-expressing infected
0, and 𝑦(0) = 𝑦0 ≥ 0, there exists a plane 𝑃, given by the
𝐶𝐷4+ T cell through mitosis should not be larger than one.
equation
If the inequality in (A1) does not hold, then the number
of the infected cells may increase to infinity. The biological 𝜏 + 𝜇2 𝜏 + 𝜇2
𝑃:𝑥+𝑢+ 𝑦 = 𝐿 0 + 𝑢𝑚 + 𝑥0 + 𝑢0 + 𝑦0 , (10)
interpretation of (A1) is to keep the solutions of the model 𝜏 𝜏
bounded. From condition (A1), we have 𝑠 < 𝜇3 /(1 − 𝜀); that such that the point (𝑥0 , 𝑢0 , 𝑦0 ) locates inside the domain with
is, 𝜇3 > (1 − 𝜀)𝑠. We can get the following nonnegative and the boundaries 𝑥 = 0, 𝑢 = 0, 𝑦 = 0, 𝑢 = 𝑢𝑚 + 𝑥0 + 𝑢0 +
bounded conclusions on the solutions of model (1). ((𝜏 + 𝜇2 )/𝜏)𝑦0 , and 𝑃. It is not difficult to verify that those
two planes 𝑢 = 𝑢𝑚 + 𝑥0 + 𝑢0 + ((𝜏 + 𝜇2 )/𝜏)𝑦0 and 𝑃 have the
Theorem 1. The solutions (𝑥(𝑡), 𝑢(𝑡), 𝑦(𝑡)) of model (1) intersection line 𝑥 + ((𝜏 + 𝜇2 )/𝜏)𝑦 = 𝐿 0 . The equations in (6)
with the nonnegative initial conditions are nonnegative and and (9) imply that
bounded for all 𝑡 > 0 if (A1) holds.
𝑑𝑢 𝜏 + 𝜇2
Proof. It is easy to have ≤ 0, if 𝑢 = 𝑢𝑚 + 𝑥0 + 𝑢0 + 𝑦0 ≥ 𝑢𝑚 ,
𝑑𝑡 𝜏
𝑑𝑥(𝑡) 󵄨󵄨󵄨󵄨 𝜏 + 𝜇2
󵄨 = 𝜆 > 0, 𝑥+ 𝑦 ≤ 𝐿 0, (11)
𝑑𝑡 󵄨󵄨󵄨𝑥=0 𝜏
𝑑𝑢(𝑡) 󵄨󵄨󵄨󵄨 𝑑𝐿
≤ 0 if 𝑥 +
𝜏 + 𝜇2
𝑦 ≥ 𝐿 0.
󵄨 = 𝜎𝛽𝑥𝑦 + 𝜀𝑠𝑦 ≥ 0,
𝑑𝑡 󵄨󵄨󵄨𝑢=0
(4) 𝑑𝑡 𝜏
𝑑𝑦(𝑡) 󵄨󵄨󵄨󵄨
Those inequalities imply that the domain with the boundaries
󵄨 = 𝜏𝑢 ≥ 0. 𝑥 = 0, 𝑢 = 0, 𝑦 = 0, 𝑢 = 𝑢𝑚 + 𝑥0 + 𝑢0 + ((𝜏 + 𝜇2 )/𝜏)𝑦0 ,
𝑑𝑡 󵄨󵄨󵄨𝑦=0 and 𝑃 is positively invariant for solutions of model (1). That
From Lemma 2 in [28], we know that any solutions of model is, any solution of model (1) with nonnegative initial value is
(1) with nonnegative initial conditions will be nonnegative for bounded.
all 𝑡 > 0. With a similar argument as used in the proof of
It follows from the first equation of model (1) that Theorem 1, we know that the domain
𝑑𝑥 𝜆
= 𝜆 − 𝛽𝑥𝑦 − 𝜇1 𝑥 ≤ 𝜆 − 𝜇1 𝑥, (5) Γ = {(𝑥, 𝑢, 𝑦) | 0 ≤ 𝑥 ≤ , 0 ≤ 𝑢 ≤ 𝑢𝑚 ,
𝑑𝑡 𝜇1
which leads to lim𝑡 → +∞ sup 𝑥 ≤ 𝜆/𝜇1 . Let 𝐿 = 𝑥 + 𝑢 + ((𝜏 + (12)
𝜏 + 𝜇2
𝜇2 )/𝜏)𝑦; from model (1) we can obtain 𝑦 ≥ 0, 𝑥 + 𝑢 + 𝑦 ≤ 𝐿 0 + 𝑢𝑚 }
𝜏
𝑑𝐿 𝑑𝑥 𝑑𝑢 𝜏 + 𝜇2 𝑑𝑦
= + + is positively invariant with respect to model (1). In fact, the
𝑑𝑡 𝑑𝑡 𝑑𝑡 𝜏 𝑑𝑡 (6) solutions of model (1) located on the boundary planes of Γ,
= 𝜆 + (𝜎 − 1) 𝛽𝑥𝑦 − 𝜇1 𝑥 − 𝐺𝑦 ≤ 𝜆 − 𝜇1 𝑥 − 𝐺𝑦, 𝑥 = 𝜆/𝜇1 , or 𝑢 = 𝑢𝑚 , or 𝑥 + 𝑢 + ((𝜏 + 𝜇2 )/𝜏)𝑦 = 𝐿 0 + 𝑢𝑚 , will
enter Γ0 , where Γ0 is the interior of Γ. From (5), (6), and (9)
where 𝐺 = ((𝜏+𝜇2 )/𝜏)(𝜇3 −(1−𝜀)𝑠)−𝜀𝑠 > 0 since (A1) holds.
we can prove that all the solutions of model (1) with positive
The inequality in (6) implies that 𝐿 = 𝑥+𝑢+((𝜏+𝜇2 )/𝜏)𝑦 will
initial values will enter Γ when the time is large enough. We
decrease along the solutions curve of model (1) if 𝜇1 𝑥 + 𝐺𝑦 >
will investigate the dynamic behavior of model (1) on Γ in the
𝜆. Geometrically, all solution trajectories of model (1) will go
rest of the paper.
through the plane 𝑥 + 𝑢 + ((𝜏 + 𝜇2 )/𝜏)𝑦 = 𝐿 from outside to
The straightforward calculation shows that model (1) has
inside if 𝜇1 𝑥 + 𝐺𝑦 > 𝜆.
two equilibria: the infection-free equilibrium 𝑃0 = (𝑥0 , 0, 0),
Let 𝐿 0 be the maximal value of the function 𝑥 + ((𝜏 +
located on the boundary of Γ, where 𝑥0 = 𝜆/𝜇1 , and the
𝜇2 )/𝜏)𝑦 on the bounded domain
chronic infection equilibrium 𝑃1 = (𝑥1 , 𝑢1 , 𝑦1 ), where
𝐺0 = {(𝑥, 𝑦) | 𝑥 ≥ 0, 𝑦 ≥ 0, 𝜇1 𝑥 + 𝐺𝑦 ≤ 𝜆} , (7)
𝜆 (𝜇3 − (1 − 𝜀) 𝑠) 𝑦1
𝑥1 = , 𝑢1 = ,
and let 𝑀0 be the maximal value of the function 𝜎𝛽𝑥𝑦 + 𝜀𝑠𝑦 𝛽𝑦1 + 𝜇1 𝜏
on the bounded domain (13)
𝜏 + 𝜇2 𝜇 (𝜏 + 𝜇2 ) 𝜇3
𝐺1 = {(𝑥, y) | 𝑥 ≥ 0, 𝑦 ≥ 0, 𝑥 + 𝑦 ≤ 𝐿 0 } . (8) 𝑦1 = 1 (𝑅0 − 1) .
𝜏 𝛽 (𝜏 + 𝜇2 ) 𝜇3 − 𝜏𝑠 − 𝜇2 (1 − 𝜀) 𝑠
4 Abstract and Applied Analysis

𝑥1 , 𝑢1 , and 𝑦1 are positive if and only if 𝑅0 > 1 and (A1) 3.2. Stability of the Chronic Infection Equilibrium
holds. We have the following conclusion on the existence of
the equilibrium of model (1). Theorem 5. Assume that (A1) holds; if 𝑅0 > 1, then the unique
chronic infection equilibrium 𝑃1 of model (1) is stable.
Theorem 2. If 𝑅0 ≤ 1, then 𝑃0 = (𝜆/𝜇1 , 0, 0) is the only
equilibrium of model (1). If 𝑅0 > 1 and (A1) holds, then Proof. The characteristic equation of the matrix of the lin-
𝑃1 = (𝑥1 , 𝑢1 , 𝑦1 ) is the unique chronic infection equilibrium. earized system of model (1) at the chronic infection equilib-
rium 𝑃1 is

3. Stability Analysis of Equilibria 𝜌3 + 𝑏1 𝜌2 + 𝑐1 𝜌 + 𝑑1 = 0, (16)

3.1. Stability of Infection-Free Equilibrium. Intuitively, if 𝑅0 < where
1, then a Tax-expressing infected 𝐶𝐷4+ T cell will produce
less than one secondary infection on average in its lifetime. 𝑏1 = 𝜇3 − (1 − 𝜀) 𝑠 + 𝜏 + 𝜇2 + 𝛽𝑦1 + 𝜇1 ,
This fact may lead to the extinction of the infection. We
will try to prove the global stability of the infection-free 𝑐1 = (𝜏 + 𝜇2 ) (𝜇3 − (1 − 𝜀) 𝑠)
equilibrium when 𝑅0 < 1. + (𝛽𝑦1 + 𝜇1 ) (𝜇3 − (1 − 𝜀) 𝑠)
(17)
Theorem 3. If 𝑅0 < 1, then the infection-free equilibrium 𝑃0 + (𝛽𝑦1 + 𝜇1 ) (𝜏 + 𝜇2 ) − (𝜎𝛽𝜏𝑥1 + 𝜀𝜏𝑠) ,
of model (1) is stable, and it is unstable if 𝑅0 > 1.
𝑑1 = (𝛽𝑦1 + 𝜇1 ) (𝜏 + 𝜇2 ) (𝜇3 − (1 − 𝜀) 𝑠)
Proof. We use the linearized system of model (1) to discuss
the stability of 𝑃0 . The characteristic equation of the matrix − 𝜀𝜏𝑠𝛽𝑦1 − 𝜎𝛽𝜏𝜇1 𝑥1 − 𝜇1 𝜀𝜏𝑠.
of the linearized system of model (1) at the infection-free
equilibrium 𝑃0 is Since 𝛽𝑦1 + 𝜇1 = 𝜆/𝑥1 , 𝜎𝛽𝜏𝑥1 + 𝜀𝜏𝑠 = (𝜏 + 𝜇2 )(𝜇3 − (1 − 𝜀)𝑠),
we have 𝑐1 = (𝜆2 /𝑥1 )𝑏0 > 0, 𝑑1 = 𝜇1 𝜇3 (𝜏 + 𝜇2 )(𝑅0 − 1). The
(𝜌 + 𝜇1 ) (𝜌2 + 𝑏0 𝜌 + 𝑐0 ) = 0, (14) straightforward calculation yields 𝑏1 𝑐1 − 𝑑1 > 0. According
to the Routh-Hurwitz criterion, we can see that all the roots
of (16) have negative real parts if 𝑅0 > 1. This completes the
where 𝑏0 = 𝜇3 (1 − 𝑅0 + (𝜎𝛽𝜏𝑥0 + 𝜏𝜀𝑠)/(𝜏 + 𝜇2 )𝜇3 ) + 𝜏 + 𝜇2 , proof of Theorem 5.
𝑐0 = (1 − 𝑅0 )(𝜏 + 𝜇2 )𝜇3 . From the Routh-Hurwitz criterion,
it is easy to know that all the roots of (14) have negative real The following two lemmas, which can be found in [29],
parts if 𝑅0 < 1, and (14) has at least one root with positive real are used for the study of the uniform persistence of model
part if 𝑅0 > 1. This completes the proof. (1). We show that the disease persists when 𝑅0 > 1; that is,
the infected proportion of the 𝐶𝐷4+ T cells persists above a
Theorem 4. If 𝑅0 < 1, then the infection-free equilibrium 𝑃0 certain positive level for sufficiently large 𝑡.
of model (1) is globally asymptotically stable in Γ. Let 𝑓 : 𝑋 → 𝑋 be a continuous map and 𝑋0 ⊂ 𝑋 an
open set. Define 𝜕𝑋0 = 𝑋/𝑋0 and 𝑀𝜕 := {𝑥 ∈ 𝜕𝑋0 | 𝑓𝑛 (𝑥) ∈
Proof. We consider a Lyapunov function 𝐿 = 𝜏𝑢 + (𝜏 + 𝜇2 )𝑦. 𝜕𝑋0 , 𝑛 ≥ 0}.
Calculating the derivative of 𝐿 along the solutions of model
(1) gives Lemma 6 (see [29]). If 𝑓 : 𝑋 → 𝑋 is compact and point
dissipative, then there is a connected global attractor A that
𝑑𝐿 󵄨󵄨󵄨󵄨 𝑑𝑢 𝑑𝑦 attracts each bounded set in X.
󵄨 =𝜏 + (𝜏 + 𝜇2 )
𝑑𝑡 󵄨󵄨󵄨(1) 𝑑𝑡 𝑑𝑡
Lemma 7 (see [29]). Let 𝑓 : 𝑋 → 𝑋 be a continuous map
= 𝑦 (𝜏𝜎𝛽𝑥 + 𝜏𝜀𝑠 + (𝜏 + 𝜇2 ) (1 − 𝜀) 𝑠 − (𝜏 + 𝜇2 ) 𝜇3 ) and 𝑋0 ⊂ 𝑋 an open set. Assume that
𝜆 (C1) 𝑓(𝑋0 ) → 𝑋0 and 𝑓 has a global attractor 𝐴;
≤ 𝑦 (𝜏𝜎𝛽 + 𝜏𝜀𝑠 + (𝜏 + 𝜇2 ) (1 − 𝜀) 𝑠 − (𝜏 + 𝜇2 ) 𝜇3 )
𝜇1 (C2) the maximal compact invariant set 𝐴 𝜕 = 𝐴 ∩ 𝑀𝜕 of 𝑓
in 𝜕𝑋0 , possibly empty, admits a Morse decomposition
= 𝑦𝜇3 (𝜏 + 𝜇2 ) (𝑅0 − 1) .
{𝑀1 , . . . , 𝑀𝐾 } with the following properties:
(15)
(a) 𝑀𝑖 is isolated in 𝑋;
Therefore, 𝑅0 < 1 implies that (𝑑𝐿/𝑑𝑡)|(1) ≤ 0 for all 𝑡 > 0, (b) 𝑊𝑠 (𝑀𝑖 ) ∩ 𝑋0 = 𝜙 for each 1 ≤ 𝑖 ≤ 𝑘.
and (𝑑𝐿/𝑑𝑡)|(1) = 0 only if 𝑦 = 0. From the inequality in
(15) we can have that lim𝑡 → ∞ 𝑦(𝑡) = 0, lim𝑡 → ∞ 𝑢(𝑡) = 0. By Then there exists 𝜌 > 0 such that, for any compact internally
using the limiting theory for ordinary differential equations 𝑀𝑖 for all 1 ≤ 𝑖 ≤ 𝑘, we have
chain transitive set 𝐿 with 𝐿 ⊄
we can have lim𝑡 → ∞ 𝑥(𝑡) = 𝜆/𝜇1 . That is, the infection-free inf 𝑥∈𝐿 𝑑(𝑥, 𝜕𝑋0 ) > 𝜌.
equilibrium 𝑃0 attracts all solutions of model (1) with initial
values in Γ. The global stability conclusion of Theorem 4 is We deal with the uniform persistence of model (1) now.
proved. Let 𝑋 = {(𝑥, 𝑢, 𝑦) | 𝑥 ≥ 0, 𝑢 ≥ 0, 𝑦 ≥ 0}, 𝑋0 = {(𝑥, 𝑢, 𝑦) |
Abstract and Applied Analysis 5

𝑥 ≥ 0, 𝑢 > 0, 𝑦 > 0}; define 𝜕𝑋0 = 𝑋/𝑋0 , and 𝑀𝜕 = Proof. If the conclusion in Proposition 9 does not hold, then,
{(𝑥(0), 𝑢(0), 𝑦(0)) ∈ 𝜕𝑋0 | Φ𝑡 (𝑥(0), 𝑢(0), 𝑦(0)) ∈ 𝜕𝑋0 , 𝑡 ≥ 0}, for any 𝛿 > 0, there exists a 𝑇 such that 𝑢(𝑡) ≤ 𝛿 and 𝑦(𝑡) ≤ 𝛿
where Φ𝑡 : 𝑋 → 𝑋 is the semiflow defined by model (1). for all 𝑡 > 𝑇. Consider the following equation:

Proposition 8. One has 𝑀𝜕 = {(𝑥, 0, 0) | 𝑥 ≥ 0}. 𝑑𝑥̂

= 𝜆 − 𝛽𝛿𝑥̂ − 𝜇1 𝑥.
̂ (22)
𝑑𝑡
Proof. We first show that 𝑀𝜕 ⊂ {(𝑥, 0, 0) | 𝑥 ≥ 0}; that is,
if (𝑥(0), 𝑢(0), 𝑦(0)) ∈ 𝑀𝜕 , then 𝑢(0) = 𝑦(0) = 0. Due to the The solution of (22) with the any initial value 𝑥(𝑡0 ) > 0 is
definition of 𝑀𝜕 , we can get Φ𝑡 (𝑥(0), 𝑢(0), 𝑦(0)) ∈ 𝜕𝑋0 for all
𝜆
𝑡 ≥ 0, especially, Φ0 (𝑥(0), 𝑢(0), 𝑦(0)) = (𝑥(0), 𝑢(0), 𝑦(0)) ∈ 𝑥̂ (𝑡) = [1 − exp [(𝛽𝛿 + 𝜇1 ) (𝑡0 − 𝑡)]]
𝜕𝑋0 . If 𝑀𝜕 ⊂ {(𝑥, 0, 0) | 𝑥 ≥ 0} does not hold, then at least one 𝛽𝛿 + 𝜇1 (23)
of 𝑢(0), 𝑦(0) is greater than zero. Without loss of generality, + 𝑥 (𝑡0 ) exp [(𝛽𝛿 + 𝜇1 ) (𝑡0 − 𝑡)] ,
we assume that 𝑢(0) > 0. When 𝑢(0) > 0 we can prove that
𝑢(𝑡) and 𝑦(𝑡) are all greater than zero for 𝑡 ∈ [0, 1]. In fact, ̂ = 𝜆/(𝛽𝛿 + 𝜇1 ). For 𝜀1 > 0, there exists a
and lim𝑡 → +∞ 𝑥(𝑡)
from the second equation of model (1) we have ̂ > 𝜆/(𝛽𝛿 + 𝜇1 ) − 𝜀1 holds when 𝑡 ≥ 𝑇1 .
𝑇1 > 𝑇, such that 𝑥(𝑡)
𝑥̂1 (𝛿) = 𝜆/(𝛽𝛿 + 𝜇1 ) is an equilibrium of (22). The fact that
𝑑𝑢 lim𝛿 → 0 𝑥̂1 (𝛿) = 𝜆/𝜇1 = 𝑥0 implies that 𝑥̂1 (𝛿) ≥ 𝑥0 − 𝜀1 when
= 𝜎𝛽𝑥𝑦 + 𝜀𝑠𝑦 − 𝜏𝑢 − 𝜇2 𝑢 ≥ − (𝜏 + 𝜇2 ) 𝑢, 𝑡 ∈ [0, 1] .
𝑑𝑡 𝛿 is small enough. By the comparison principle, we can have
(18) ̂ and 𝑥(𝑡) ≥ 𝑥0 − 2𝜀1 , for 𝑡 > 𝑇1 .
𝑥(𝑡) ≥ 𝑥(𝑡)
Consider the following linear system:
It follows that
𝑑̂
𝑢
𝑢 (𝑡) ≥ 𝑢 (0) exp [− (𝜏 + 𝜇2 )] ≜ 𝑀1 > 0. (19) = 𝜎𝛽 (𝑥0 − 2𝜀1 ) 𝑦̂ + 𝜀𝑠𝑦̂ − 𝜏̂
𝑢 − 𝜇2 𝑢̂,
𝑑𝑡
(24)
From the third equation of model (1) we have 𝑑𝑦̂
𝑢 + (1 − 𝜀) 𝑠𝑦̂ − 𝜇3 𝑦.
= 𝜏̂ ̂
𝑑𝑡
𝑑𝑦
= 𝜏𝑢 + (1 − 𝜀) 𝑠𝑦 − 𝜇3 𝑦 ≥ 𝜏𝑀1 + (1 − 𝜀) 𝑠𝑦 − 𝜇3 𝑦; (20) The characteristic equation is
𝑑𝑡
𝜌2 + 𝑏2 𝜌 + 𝑐2 = 0, (25)
then, for 𝑡 ∈ [0, 1], we can have
where 𝑏2 = 𝜏 + 𝜇2 + 𝜇3 − (1 − 𝜀)𝑠 > 0, 𝑐2 = (1 − 𝑅0 )(𝜏 +
𝜏𝑀1 𝜇2 )𝜇3 + 2𝜏𝜎𝛽𝜀1 . From the expression of 𝑐2 we see that 𝑐2 < 0 if
𝑦 (𝑡) ≥ [1 − exp [− (𝜇3 − (1 − 𝜀) 𝑠) 𝑡]]
𝜇3 − (1 − 𝜀) 𝑠 𝑅0 > 1 and 𝜀1 is small enough. Let 𝜌1 and 𝜌2 be the two roots
of 𝜌2 + 𝑏2 𝜌 + 𝑐2 = 0 and 𝜌1 > 0 > 𝜌2 . The solution of model
+ 𝑦 (0) exp [− (𝜇3 − (1 − 𝜀) 𝑠) 𝑡] (21)
(24) with the initial value (̂ ̂
𝑢(0), 𝑦(0)) > 0 satisfies
𝜏𝑀1
≥ [1 − exp [− (𝜇3 − (1 − 𝜀) 𝑠)]] > 0. 𝑇
𝑢 (𝑡) , 𝑦̂ (𝑡)) = 𝑑1 𝜉1 exp (𝜌1 𝑡) + 𝑑2 𝜉2 exp (𝜌2 𝑡) ,
(̂ (26)
𝜇3 − (1 − 𝜀) 𝑠
where 𝜉1 and 𝜉2 are the eigenvectors corresponding to 𝜌1 and
The inequalities 𝑢(𝑡) ≥ 𝑀1 > 0 and 𝑦(𝑡) ≥ (𝜏𝑀1 /(𝜇3 − (1 −
𝜌2 , respectively. 𝑑1 and 𝑑2 are two constants depending on
𝜀)𝑠))[1 − exp[−(𝜇3 − (1 − 𝜀)𝑠)]] > 0 for 𝑡 ∈ [0, 1] imply that
(̂ ̂
𝑢(0), 𝑦(0)). The solution expression of model (24) indicates
(𝑥(𝑡), 𝑢(𝑡), 𝑦(t)) ∈ 𝑋0 for 𝑡 ∈ [0, 1]. From the definition of
that max {̂ ̂
𝑢(𝑡), 𝑦(𝑡)} → ∞ as 𝑡 → ∞. For the same initial
𝑀𝜕 and (𝑥(𝑡), 𝑢(𝑡), 𝑦(𝑡)) ∈ 𝑋0 for 𝑡 ∈ [0, 1] we know that
values, the comparison principle implies that 𝑢(𝑡) > 𝑢̂(𝑡) and
Φ0 (𝑥(0), 𝑢(0), 𝑦(0)) ∉ 𝜕𝑋0 if 𝑢(0) > 0. This contradiction
̂ where 𝑢(𝑡) and 𝑦(𝑡) are the solutions of model (1).
𝑦(𝑡) > 𝑦(𝑡),
implies that (𝑥(0), 𝑢(0), 𝑦(0)) ∈ 𝑀𝜕 only if 𝑢(0) = 𝑦(0) = 0;
Subsequently, we have 𝑢(𝑡) → ∞ or 𝑦(𝑡) → ∞ as 𝑡 → ∞.
that is, 𝑀𝜕 ⊂ {(𝑥, 0, 0) | 𝑥 ≥ 0}.
The contradiction shows that Proposition 9 holds true.
On the other hand, for any initial values (𝑥(0), 0, 0) ∈
{(𝑥, 0, 0) | 𝑥 ≥ 0}, we have 𝑑𝑢/𝑑𝑡 = 0, 𝑑𝑦/𝑑𝑡 = 0, and By using Propositions 8 and 9, we can get the uniform
𝑢(𝑡) = 𝑦(𝑡) = 0 for 𝑡 ≥ 0, {(𝑥, 0, 0) | 𝑥 ≥ 0} ⊂ 𝑀𝜕 . The persistence of model (1).
proposition is proved.
Theorem 10. Assume that (A1) holds. If 𝑅0 > 1,
From Proposition 8, we can get the conclusion that 𝑀𝜕 then model (1) is uniformly persistent with respect to
is the maximal invariant set in 𝜕𝑋0 . Next we show that (𝑋0 , 𝜕𝑋0 ); that is, there exists a positive number 𝜂 such that
the solutions with the initial values in 𝑋0 cannot go to the min{lim𝑡 → ∞ inf 𝑥(𝑡), lim𝑡 → ∞ inf 𝑢(𝑡), lim𝑡 → ∞ inf 𝑦(𝑡)} ≥
boundary. 𝜂.

Proposition 9. Assume that (A1) holds. If 𝑅0 > 1, then Proof. 𝑋 and 𝑋0 are positively invariant for model (1). Φ𝑡 is
there exists a 𝛿 > 0 such that the solution of model point dissipative and compact. By Lemma 6 we know that
(1) with initial value (𝑥(𝑡0 ), 𝑢(𝑡0 ), 𝑦(𝑡0 )) ∈ 𝑋0 satisfies there is a connected global attractor 𝐴 for Φ𝑡 that attracts each
lim𝑡 → +∞ sup max{𝑢(𝑡), 𝑦(𝑡)} > 𝛿. bounded set in 𝑋.
6 Abstract and Applied Analysis

From the discussion of Proposition 8, we know that 𝑀𝜕 is Next, we show that model (1) is a competitive system
the maximal compact invariant set in 𝜕𝑋0 . Since we choose which implies that model (1) has the Poincaré-Bendixson
the Morse decomposition of 𝑀𝜕 as {𝑃0 } and ∪𝑥∈𝑀𝜕 𝜔(𝑥) = property.
{𝑃0 }, the set {𝑃0 } is isolated. Proposition 9 shows that the
solutions of model (1) with initial values in 𝑋0 cannot go to Theorem 14. Model (1) is competitive in Γ.
the boundary, which implies that 𝑊𝑠 (𝑃0 ) ∩ 𝑋0 = 𝜙. It follows
from Lemma 7 that model (1) is uniformly persistent with Proof. The Jacobian matrix of model (1) is
respect to (𝑋0 , 𝜕𝑋0 ).
−𝛽𝑦 − 𝜇1 0 −𝛽𝑥
The following lemmas in [30–32] are used to study the 𝐽 (𝑥, 𝑢, 𝑦) = [ 𝜎𝛽𝑦 −𝜏 − 𝜇2 𝜎𝛽𝑥 + 𝜀𝑠 ] . (28)
global stability of the chronic infection equilibrium 𝑃1 . We [ 0 𝜏 (1 − 𝜀) 𝑠 − 𝜇3 ]
will show that all the solutions of model (1) in Γ0 converge to
𝑃1 if 𝑅0 > 1. Choose 𝑃 = diag(1, −1, 1); we can obtain
Let 𝑥 → 𝑓(𝑥) ∈ 𝑅𝑛 be a 𝐶1 function for 𝑥 in an open set
𝐷 ⊂ 𝑅𝑛 . Consider the system of differential equations −𝛽𝑦 − 𝜇1 0 −𝛽𝑥
𝑃𝐽𝑃 = [ −𝜎𝛽𝑦 −𝜏 − 𝜇2 −𝜎𝛽𝑥 − 𝜀𝑠 ] . (29)
𝑑𝑥 [ 0 −𝜏 (1 − 𝜀) 𝑠 − 𝜇3 ]
= 𝑓 (𝑥) . (27)
𝑑𝑡
All off-diagonal entries of 𝑃𝐽𝑃 are nonpositive. It follows from
Let 𝑥(𝑡, 𝑥0 ) be the solution of model (27) satisfying 𝑥(0, 𝑥0 ) = Lemma 11 that model (1) is competitive in the convex region
𝑥0 . Γ.
A set 𝐾 is said to be absorbing in 𝐷 for model (27) if
𝑥(𝑡, 𝐾1 ) ⊂ 𝐾 for each compact 𝐾1 ⊂ 𝐷 and sufficiently large Now, we are ready to prove the global stability of the
𝑡. We make the following two basic assumptions. unique chronic infection equilibrium 𝑃1 of model (1).
(𝐻1 ) There exists a compact absorbing set 𝐾 ⊂ 𝐷. Theorem 15. Assume that (A1) holds. If 𝑅0 > 1, then the
(𝐻2 ) System (27) has a unique equilibrium 𝑥 in 𝐷. unique chronic infection equilibrium 𝑃1 of model (1) is globally
asymptotically stable in Γ0 .
System (27) is said to have the Poincaré-Bendixson
Property if any nonempty compact omega limit set that Proof. From Theorem 10 and Lemma 6, we know that Φ𝑡
contains no equilibrium is a closed orbit [31]. It is known that is compact and point dissipative, and there is a global
a three-dimensional competitive system has the Poincaré- attractor 𝐴 for Φ𝑡 . Subsequently, model (1) satisfies (𝐻1 ).
Bendixson property in a convex region. From Theorem 2, model (1) satisfies (𝐻2 ). By Theorem 14 and
Lemma 12, model (1) has the Poincaré-Bendixson property.
Lemma 11 (see [30]). Let 𝐷 ∈ 𝑅𝑛 be convex. The autonomous Thus conditions (1) and (5) of Lemma 13 hold.
system 𝑑𝑥/𝑑𝑡 = 𝑓(𝑥), 𝑥 ∈ 𝐷, is cooperative in 𝐷 if there The second compound system of the linearized system
exists a diagonal matrix 𝑃 = diag(𝛼1 , . . . , 𝛼𝑛 ) (𝛼𝑖 = −1 or along a periodic solution (𝑥(𝑡), 𝑢(𝑡), 𝑦(𝑡)) of model (1) is
1, 𝑖 = 1, 2, . . . , 𝑛), such that 𝑃(𝜕𝑓𝑖 /𝜕𝑥𝑗 )(𝑥)𝑃 ≥ 0, for 𝑖 ≠ 𝑗,
𝑥 ∈ 𝐷; that is, all off-diagonal entries of 𝑃(𝜕𝑓/𝜕𝑥)(𝑥)𝑃 are 𝑑𝑋
nonnegative. It is competitive in 𝐷 if there exists a diagonal = − (𝛽𝑦 + 𝜇1 + 𝜏 + 𝜇2 ) 𝑋 + (𝜎𝛽𝑥 + 𝜀𝑠) 𝑌 + 𝛽𝑥𝑍,
𝑑𝑡
matrix 𝑃 = diag(𝛼1 , . . . , 𝛼𝑛 ) (𝛼𝑖 = −1 or 1, 𝑖 = 1, 2, . . . , 𝑛),
such that 𝑃(𝜕𝑓𝑖 /𝜕𝑥𝑗 )(𝑥)𝑃 ≤ 0, for 𝑖 ≠ 𝑗, 𝑥 ∈ 𝐷; that is, all 𝑑𝑌
= 𝜏𝑋 − (𝛽𝑦 + 𝜇1 − (1 − 𝜀) 𝑠 + 𝜇3 ) 𝑌, (30)
off-diagonal entries of 𝑃(𝜕𝑓/𝜕𝑥)(𝑥)𝑃 are nonpositive. 𝑑𝑡
d𝑍
Lemma 12 (see [32]). Assume that 𝑛 = 3 and 𝐷 is convex; = 𝜎𝛽𝑦𝑌 − (𝜏 + 𝜇2 − (1 − 𝜀) 𝑠 + 𝜇3 ) 𝑍.
suppose that model (27) is competitive in 𝐷; then it satisfies the 𝑑𝑡
Poincaré-Bendixson property [32].
In order to verify that model (30) is asymptotically stable, we
Lemma 13 (see [31]). Assume that the following conditions define a Lyapunov function
hold.
𝑢 𝛽𝜆
𝑉 (𝑋, 𝑌, 𝑍; 𝑥, 𝑢, 𝑦) = sup {|𝑋| , (|𝑌| + |𝑍|)} .
(1) Assumptions (𝐻1 ) and (𝐻2 ) hold; 𝑦 𝜎𝛽𝜆 + 𝜀𝑠𝜇1
(2) model (27) satisfies the Poincaré-Bendixson property; (31)

(3) for each periodic solution 𝑥 = 𝑝(𝑡) with 𝑝(0) ∈ 𝐷, From the uniform persistence, we know that the orbit O of
model (27) is asymptotically stable; the periodic solution (𝑥(𝑡), 𝑢(𝑡), 𝑦(𝑡)) has a positive distance
(4) (−1)𝑛 det((𝜕𝑓/𝜕𝑥)(𝑥)) > 0. from the boundary of Γ. There exists a constant 𝑐 > 0 such
that
Then the unique equilibrium 𝑥 is globally asymptotically
stable in 𝐷. 𝑉 (𝑋, 𝑌, 𝑍; 𝑥, 𝑢, 𝑦) ≥ 𝑐 sup {|𝑋| , |𝑌| , |𝑍|} . (32)
Abstract and Applied Analysis 7

For all (𝑋, 𝑌, 𝑍) ∈ 𝑅3 and (𝑥, 𝑢, 𝑦) ∈ O, we have 𝑢 𝛽𝜆

+ ⋅ |𝑍| (−𝜏 − 𝜇2 + (1 − 𝜀) 𝑠 − 𝜇3 )
the following estimates on the right derivatives along the 𝑦 𝜎𝛽𝜆 + 𝜀𝑠𝜇1
solutions (𝑋(𝑡), 𝑌(𝑡), 𝑍(𝑡)) of model (30):
𝑢 𝑢
+ 𝜏 |𝑋| + |𝑌| ( − 𝜇1 + (1 − 𝜀) 𝑠 − 𝜇3 − 𝛽𝑦
𝑦 𝑦
𝑋 𝜎𝛽𝜆
𝐷+ (|𝑋|) = (− (𝛽𝑦 + 𝜇1 + 𝜏 + 𝜇2 ) 𝑋 + 𝛽𝑦)
|𝑋| 𝜎𝛽𝜆 + 𝜀𝑠𝜇1
+ (𝜎𝛽𝑥 + 𝜀𝑠) 𝑌 + 𝛽𝑥𝑍)
𝑢 󸀠 𝑦󸀠 𝑢 𝛽𝜆 𝑢
≤( − ) ⋅ (|𝑌| + |𝑍|) + 𝜏 |𝑋|
≤ − (𝛽𝑦 + 𝜇1 + 𝜏 + 𝜇2 ) |𝑋| 𝑢 𝑦 𝑦 𝜎𝛽𝜆 + 𝜀𝑠𝜇1 𝑦
𝑢
+ (𝜎𝛽𝑥 + 𝜀𝑠) |𝑌| + 𝛽𝑥 |𝑍| + |𝑌| (−𝜇1 + (1 − 𝜀) 𝑠 − 𝜇3 )
𝑦
= − (𝛽𝑦 + 𝜇1 + 𝜏 + 𝜇2 ) |𝑋|
𝑢 𝛽𝜆
+ ⋅ |𝑍| (−𝜏 − 𝜇2 + (1 − 𝜀) 𝑠 − 𝜇3 )
𝑦 𝑢 𝛽𝑥 𝑦 𝜎𝛽𝜆 + 𝜀𝑠𝜇1
+ (𝜎𝛽𝑥 + 𝜀𝑠) ⋅ (|𝑌| + |𝑍|)
𝑢 𝑦 𝜎𝛽𝑥 + 𝜀𝑠
𝑢 𝑢 󸀠 𝑦󸀠
≤ 𝜏 |𝑋| + ( − + (1 − 𝜀) 𝑠
= − (𝛽𝑦 + 𝜇1 + 𝜏 + 𝜇2 ) |𝑋| 𝑦 𝑢 𝑦
𝑦 𝑢 𝛽𝜆
+ (𝜎𝛽𝑥 + 𝜀𝑠) ⋅ (|𝑌| + |𝑍|) ; −𝜇3 − min {𝜇1 , 𝜏 + 𝜇2 } )
𝑢 𝑦 𝜎𝛽𝜆 + 𝜀𝑠𝜇1
𝑌 𝑢 𝛽𝜆
𝐷+ (|𝑌|) = (𝜏𝑋 − (𝛽𝑦 + 𝜇1 − (1 − 𝜀) 𝑠 + 𝜇3 ) 𝑌) × (|𝑌| + |𝑍|) .
|𝑌| 𝑦 𝜎𝛽𝜆 + 𝜀𝑠𝜇1
≤ 𝜏 |𝑋| − (𝛽𝑦 + 𝜇1 − (1 − 𝜀) 𝑠 + 𝜇3 ) |𝑌| ; (34)
𝑍 The inequalities in (33) and (34) lead to
𝐷+ (|𝑍|) = (𝜎𝛽𝑦𝑌 − (𝜏 + 𝜇2 − (1 − 𝜀) 𝑠 + 𝜇3 ) 𝑍)
|𝑍|
𝐷+ 𝑉 (𝑡) ≤ max {𝑔1 (𝑡) , 𝑔2 (𝑡)} 𝑉 (𝑡) , (35)
≤ 𝜎𝛽𝑦 |𝑌| − (𝜏 + 𝜇2 − (1 − 𝜀) 𝑠 + 𝜇3 ) |𝑍| .
where
(33)
𝑦
𝑔1 (𝑡) = − (𝛽𝑦 + 𝜇1 + 𝜏 + 𝜇2 ) + (𝜎𝛽𝑥 + 𝜀𝑠) ,
𝑢
From (33) we have 𝑢 𝑢 󸀠 𝑦󸀠
𝑔2 (𝑡) = 𝜏+ − + (1 − 𝜀) 𝑠 − 𝜇3 − min {𝜇1 , 𝜏 + 𝜇2 } .
𝑦 𝑢 𝑦
(36)
𝑢 𝛽𝜆
𝐷+ ( (|𝑌| + |𝑍|)) After rewriting the last two equations of model (1), we find
𝑦 𝜎𝛽𝜆 + 𝜀𝑠𝜇1
that
𝑢󸀠 𝑦 − 𝑢𝑦󸀠 𝛽𝜆
= (|𝑌| + |𝑍|)
𝑦 2 𝜎𝛽𝜆 + 𝜀𝑠𝜇1 𝑦 𝑢󸀠 𝑦󸀠 𝑢
(𝜎𝛽𝑥 + 𝜀𝑠) ⋅ = + 𝜏 + 𝜇2 , = 𝜏 + (1 − 𝜀) 𝑠 − 𝜇3 .
𝛽𝜆 𝑢 𝑢 𝑦 𝑦
𝑢
+ (𝐷+ |𝑌| + 𝐷 |𝑍|) (37)
𝑦 𝜎𝛽𝜆 + 𝜀𝑠𝜇1 +
From (36) and (37), we obtain
𝑢 󸀠 𝑦󸀠 𝑢 𝛽𝜆
≤ ( − ) ⋅ (|𝑌| + |𝑍|) 𝑢󸀠
𝑢 𝑦 𝑦 𝜎𝛽𝜆 + 𝜀𝑠𝜇1 max {𝑔1 (𝑡) , 𝑔2 (𝑡)} ≤ − min {𝜇1 , 𝜏 + 𝜇2 } ,
𝑢
𝑢
+ (𝜏 |𝑋| − (𝛽𝑦 + 𝜇1 − (1 − 𝜀) 𝑠 + 𝜇3 ) |𝑌|) 𝜔
𝑦 ∫ max {𝑔1 (𝑡) , 𝑔2 (𝑡)} 𝑑𝑡
0
𝑢 𝛽𝜆
+ ⋅ 𝜔
𝑢󸀠 (38)
𝑦 𝜎𝛽𝜆 + 𝜀𝑠𝜇1 ≤∫ ( − min {𝜇1 , 𝜏 + 𝜇2 }) 𝑑𝑡
0 𝑢
× (𝜎𝛽𝑦 |𝑌| − (𝜏 + 𝜇2 − (1 − 𝜀) 𝑠 + 𝜇3 ) |𝑍|)
= ln 𝑢(𝑡)|𝜔0 − 𝜔 min {𝜇1 , 𝜏 + 𝜇2 }
󸀠 󸀠
𝑢 𝑦 𝑢 𝛽𝜆
=( − ) ⋅ (|𝑌| + |𝑍|) = −𝜔 min {𝜇1 , 𝜏 + 𝜇2 } .
𝑢 𝑦 𝑦 𝜎𝛽𝜆 + 𝜀𝑠𝜇1
8 Abstract and Applied Analysis

800 0.6

600 0.4

x 400 u 0.2

200 0

0 −0.2
0 500 1000 0 500 1000
t t

0.6

1
0.4

y 0.5
y 0.2

0 0
1
1000
0.5
u 500
−0.2 0 x
0 500 1000 0
t

Figure 2: Global stability of the infection-free equilibrium 𝑃0 when 𝑅0 < 1.

The inequalities in (35) and (38) imply that 𝑉(𝑡) → 0 as 4. Numerical Simulation
𝑡 → ∞, which leads to (𝑋(𝑡), 𝑌(𝑡), 𝑍(𝑡)) → 0 as 𝑡 → ∞
because of (32). As a result, the second compound system Numerical simulations are done to demonstrate the results in
(30) is asymptotically stable. This verifies condition (6) of Section 3. The sensitive analysis is given to show the effects of
Lemma 13. the model parameters on the solutions.
Let 𝐽(𝑃1 ) be the Jacobian matrix of model (1) at 𝑃1 . Then In numerical simulations, the time scale is a day. The rate
we have of healthy 𝐶𝐷4+ helper T cells produced in the bone marrow,
𝜆, is 15–25 cells/mm3 /day. The coefficient of infectious trans-
𝜕𝑓 missibility, 𝛽, is 0.0005–0.003 mm3 /cell/day. The proportion
(−1)3 det ( (𝑃 ))
𝜕𝑥 1 of infected cells expressing Tax, 𝜏, is (0.003–0.03)/day. The
removal rates of healthy 𝐶𝐷4+ T cells, resting infected 𝐶𝐷4+
−𝛽𝑦1 − 𝜇1 0 −𝛽𝑥1
T cells, and Tax-expressing infected 𝐶𝐷4+ T cells, 𝜇1 , 𝜇2 , and
= −[ 𝜎𝛽𝑦1 −𝜏 − 𝜇2 𝜎𝛽𝑥1 + 𝜀𝑠 ] (39) 𝜇3 , are taken to be the value 0.01–0.05/day. The death rate
[ 0 −𝜏 (1 − 𝜀) 𝑠 − 𝜇3 ]
of the Tax-expressing infected 𝐶𝐷4+ T cells is considerably
= (𝛽𝑦1 + 𝜇1 ) ((𝜇3 − (1 − 𝜀) 𝑠) (𝜏 + 𝜇2 ) + 𝜀𝑠𝜏) shorter than the natural lifespan of 𝐶𝐷4+ T cells [33].
In Figure 2, we use the following set of parameters: 𝜆 =
+ 𝜎𝛽𝜏𝜇1 𝑥1 > 0. 20, 𝛽 = 0.001, 𝜇1 = 𝜇2 = 1/30, 𝑠 = 0.05, 𝜎 = 0.01, 𝜀 = 0.9,
𝜏 = 0.03, 𝜇3 = 0.05, and 𝑅0 = 0.5832 < 1. All solutions
Condition (9) of Lemma 13 holds. The chronic infection converge to the infection-free equilibrium 𝑃0 .
equilibrium 𝑃1 of model (1) is globally asymptotically stable In Figure 3, we use the following set of parameters: 𝜆 =
in Γ0 since all conditions of Lemma 13 are satisfied. 20, 𝛽 = 0.001, 𝜇1 = 1/30, 𝜇2 = 0.02, 𝑠 = 0.1, 𝜎 = 0.1, 𝜀 = 0.8,
Abstract and Applied Analysis 9

250 150

200

100
150
x u
100
50

50

0 0
0 200 400 600 800 1000 0 200 400 600 800 1000
t t

60

50 60

40 40

y 30
20

20
0
150
10 100 600
50 400
0 200
0 0
0 200 400 600 800 1000
t

Figure 3: Global stability of the infection-free equilibrium 𝑃1 when 𝑅0 > 1.

R0 Table 1: PRCC results and 𝑃 value.

1
∗ Parameters 𝛽 𝜏 𝑠 𝜇1 𝜇2 𝜇3
0.8
PRCC 0.8597 0.3421 0.0214 −0.0575 −0.0813 −0.0610
0.6
𝑃 value 0.0000 0.0000 0.4994 0.0696 0.0103 0.0545
0.4 ∗

0.2
PRCC

0
A sensitivity analysis quantifies how changes in the values
−0.2 of the input parameters alter the value of the outcome
−0.4 variable [34]. The sensitivity analysis is performed to explore
−0.6 the behavior of model (1) by calculating the partial rank
correlation coefficients (PRCC) for each input parameter,
−0.8
𝛽 𝜏 s 𝜇1 𝜇2 𝜇3 which are sampled by the Latin hypercube sample (LHS)
−1 and 𝑅0 (Table 1). Figure 4 shows that a significantly strong
1 2 3 4 5 6
positive correlation exists between parameters 𝛽 and 𝑅0
Parameters
(PRCC = 0.8597; 𝑃 value = 0 < 0.01). The second sensitive
PRCC parameter to 𝑅0 is 𝜏 (PRCC = 0.3421; 𝑃 value = 0 < 0.01). The
result indicates that the cell-to-cell contact transmission and
Figure 4: Sensitivity analysis of 𝑅0 with the parameters came from Tax expression contribute a lot to the viral infection.
LHS sampling. The sensitivity analysis result shows that 𝛽 and 𝜏 are
two significant parameters for the infection. We illustrate the
impact of 𝛽 and 𝜏 on the magnitude of the chronic infection
equilibrium 𝑃1 by numerical simulations. The curves in
𝜏 = 0.03, 𝜇3 = 0.09, and 𝑅0 = 1.1556 > 1. All solutions Figures 5(a) and 5(b) show the dependence of 𝑢1 and 𝑦1 on
converge to the chronic infection equilibrium 𝑃1 . the parameters 𝛽 and 𝜏, respectively. The surfaces in Figures
10 Abstract and Applied Analysis

350 100

300
80
250

200 60
u1

y1
150
40

100
20
50

0 0
0 0.005 0.01 0.015 0.02 0 0.005 0.01 0.015 0.02
𝛽 𝛽

𝜏 = 0.003 𝜏 = 0.015 𝜏 = 0.003 𝜏 = 0.015

𝜏 = 0.01 𝜏 = 0.02 𝜏 = 0.01 𝜏 = 0.02
(a) (b)

600 200

150
400
100
y1
u1

200 50

0
0 0.04
0.04 0.02 0.02
0.02
0.02 0.01 𝜏 0.01
𝜏 𝛽 0 𝛽
0 0 0
(c) (d)

Figure 5: The impact of 𝛽 and 𝜏 on the magnitude of the chronic infection equilibrium 𝑃1 .

5(c) and 5(d) give the values of 𝑢1 or 𝑦1 as the functions of has settled at a chronic infection state, the proportion of
𝛽 and 𝜏, respectively. Those curves and surfaces in Figure 5 Tax-expressing cells in infected cells is 𝑦/(𝑢 + 𝑦), where
indicate that 𝑢1 and 𝑦1 will increase with 𝛽 and 𝜏. For any 𝑢 = ((𝜇3 − (1 − 𝜀)𝑠)/𝜏)𝑦. Hence (𝜕/𝜕𝜏)(𝑦/(𝑢 + 𝑦)) =
given 𝜏, 𝑢1 or 𝑦1 increases very fast for small 𝛽 and quite slow (𝜇3 − (1 − 𝜀)𝑠)/(𝜇3 − (1 − 𝜀)𝑠 + 𝜏)2 > 0. That is, a faster
for large 𝛽. rate of spontaneous expression of the Tax results in a higher
proportion of 𝑦 in infected 𝐶𝐷4+ T cells which influence the
5. Concluding Remarks risk of HAM/TSP.
It follows from our sensitivity analysis that 𝛽 and 𝜏 are
We have formulated and studied a mathematical model of significantly sensitive to the reproduction number 𝑅0 . In
HTLV-1 in vivo including the spontaneous HTLV-1 antigen particular, increasing the rate of Tax expression results in a
Tax expression, cell-to-cell contact, and mitotic infectious
reduction of the proportion of proviral cell at the equilibrium
route to the viral dynamics. The persistence of the model
state. This conclusion implies that Tax expression should be
is discussed. Sufficient conditions are established for the
controlled in the therapeutic intervention in order to reduce
global asymptotic stability of the infection-free equilibrium
and chronic infection equilibrium. The sensitivity analysis by the risk of HAM/TSP.
PRCC with the LHS sample is presented to show the impact Our conclusions are based on a simple model; with the
of the parameters on the model dynamics. recent progress in HTLV-1 pathogenesis and new findings
As we know, infected cells from HAM/TSP patients have in immune reactions against HTLV-1 infection and Tax
a significantly higher probability of expressing Tax protein expression, more factors should be investigated in improved
than infected cells from ACs. When an infected individual models.
Abstract and Applied Analysis 11

Conflict of Interests [15] F. Mortreux, A.-S. Gabet, and E. Wattel, “Molecular and cel-
lular aspects of HTLV-1 associated leukemogenesis in vivo,”
The authors declare that there is no conflict of interests Leukemia, vol. 17, no. 1, pp. 26–38, 2003.
regarding the publication of this paper. [16] P. Hollsberg, “Mechanisms of T-cell activation by human T-cell
lymphotropic virus type 1,” Microbiology and Molecular Biology
Acknowledgments Reviews, vol. 63, pp. 308–333, 1999.
[17] J. M. Mesnard and C. Devaux, “Multiple control levels of
This work was supported by a Grant (no. 104519-010) from cell proliferation by human T-cell leukemia virus type 1 Tax
the International Development Research Center, Ottawa, protein,” Virology, vol. 257, no. 2, pp. 277–284, 1999.
Canada, and the Natural Science Foundation of China (no. [18] F. Bex and R. B. Gaynor, “Regulation of gene expression by
11171267, no. 11301314). HTLV-I tax protein,” Methods, vol. 16, no. 1, pp. 83–94, 1998.
[19] D. Wodarz and C. R. M. Bangham, “Evolutionary dynamics of
HTLV-I,” Journal of Molecular Evolution, vol. 50, no. 5, pp. 448–
References 455, 2000.
[1] B. Asquith and C. R. M. Bangham, “How does HTLV-I persist [20] B. Asquith, A. J. Mosley, A. Heaps et al., “Quantification of
despite a strong cell-mediated immune response?” Trends in the virus-host interaction in human T lymphotropic virus I
Immunology, vol. 29, no. 1, pp. 4–11, 2008. infection,” Retrovirology, vol. 2, article 75, 2005.
[2] C. R. M. Bangham, “HTLV-1 infections,” Journal of Clinical [21] M. A. Nowak and C. R. M. Bangham, “Population dynamics of
Pathology, vol. 53, no. 8, pp. 581–586, 2000. immune responses to persistent viruses,” Science, vol. 272, no.
[3] R. C. Gallo, “The discovery of the first human retrovirus: HTLV- 5258, pp. 74–79, 1996.
1 and HTLV-2,” Retrovirology, vol. 2, article 17, 2005. [22] M. Y. Li and A. G. Lim, “Modelling the role of tax expression in
[4] U. Tomaru, Y. Yamano, and S. Jacobson, “HTLV-1 infection HTLV-I persistence in vivo,” Bulletin of Mathematical Biology,
and the nervous system,” in Clinical Neuroimmunology, Oxford vol. 73, no. 12, pp. 3008–3029, 2011.
University Press, Oxford, UK, 2nd edition, 2005. [23] M. A. Nowak and R. M. May, Virus Dynamics: Mathematics
[5] L. B. Cook, M. Elemans, A. G. Rowan, and B. Asquith, “HTLV-1: Principles of Immunology and Virology, Oxford University Press,
persistence and pathogenesis,” Virology, vol. 435, no. 1, pp. 131– London, UK, 2000.
140, 2013. [24] A. S. Perelson, “Modelling viral and immune system dynamics,”
[6] F. A. Proietti, A. B. F. Carneiro-Proietti, B. C. Catalan-Soares, Nature Reviews Immunology, vol. 2, no. 1, pp. 28–36, 2002.
and E. L. Murphy, “Global epidemiology of HTLV-I infection [25] D. Wodarz, M. A. Nowak, and C. R. M. Bangham, “The
and associated diseases,” Oncogene, vol. 24, no. 39, pp. 6058– dynamics of HTLV-I and the CTL response,” Immunology
6068, 2005. Today, vol. 20, no. 5, pp. 220–227, 1999.
[7] M. Saito and C. R. M. Bangham, “Immune pathogenesis
[26] B. Asquith, Y. Zhang, A. J. Mosley et al., “In vivo T lymphocyte
of human T-cell leukemia virus type-1-associated myelopa-
dynamics in humans and the impact of human T-lymphotropic
thy/tropical spastic paraparesis: recent perspectives,” Leukemia
virus 1 infection,” Proceedings of the National Academy of
Research and Treatment, vol. 2012, Article ID 259045, 12 pages,
Sciences of the United States of America, vol. 104, no. 19, pp. 8035–
2012.
8040, 2007.
[8] Y. Satou and M. Matsuoka, “Molecular and cellular mechanism
[27] P. van den Driessche and J. Watmough, “Reproduction numbers
of leukemogenesis of ATL: emergent evidence of a significant
and sub-threshold endemic equilibria for compartmental mod-
role for HBZ in HTLV-1-induced pathogenesis,” Leukemia
els of disease transmission,” Mathematical Biosciences, vol. 180,
Research and Treatment, vol. 2012, Article ID 213653, 8 pages,
pp. 29–48, 2002.
2012.
[9] C. Pique and K. S. Jones, “Pathways of cell-cell transmission of [28] X. Yang, L. Chen, and J. Chen, “Permanence and positive
HTLV-1,” Frontiers in Microbiology, vol. 3, article 378, 8 pages, periodic solution for the single-species nonautonomous delay
2012. diffusive models,” Computers and Mathematics with Applica-
tions, vol. 32, no. 4, pp. 109–116, 1996.
[10] Y. Koyanagi, Y. Itoyama, N. Nakamura et al., “In vivo infection
of human T-cell leukemia virus type 1 in non-T cells,” Virology, [29] X. Q. Zhao, Dynamical Systems in Population Biology, Springer,
vol. 196, no. 1, pp. 25–33, 1993. New York, NY, USA, 2003.
[11] I. Cho, M. Sugimoto, S. Mita, M. Tokunaga, F. Imamura, and M. [30] H. L. Smith, “Systems of ordinary differential equations which
Ando, “In vivo proviral burden and viral RNA expression in T generate an order preserving flow. A survey of results,” SIAM
cell subsets of patients with human T lymphotropic virus type- Review, vol. 30, no. 1, pp. 87–113, 1988.
1-associated myelopathy/tropical spastic paraparesis,” American [31] M. Y. Li and L. Wang, “Global stability in some SEIR epi-
Journal of Tropical Medicine and Hygiene, vol. 53, no. 4, pp. 412– demic models,” in Mathematical Approaches for Emerging and
418, 1995. Reemerging Infectious Diseases: Models, Methods and Theory,
[12] S. Tattermusch and C. R. M. Bangham, “HTLV-1 infection: vol. 126 of The IMA Volumes in Mathematics and Its Applica-
what determines the risk of inflammatory disease?” Trends in tions, pp. 295–311, Springer, 2002.
Microbiology, vol. 20, no. 10, pp. 494–500, 2012. [32] H. L. Smith, Monotone Dynamical Systems: An Introduction to
[13] C. R. Bangham, “The immune response to HTLV-I,” Current the Theory of Competitive and Cooperative Systems, American
Opinion in Immunology, vol. 12, no. 4, pp. 397–402, 2000. Mathematical Society, Providence, RI, USA, 1995.
[14] H. Gómez-Acevedo and M. Y. Li, “Backward bifurcation in [33] D. C. Macallan, B. Asquith, A. J. Irvine et al., “Measurement
a model for HTLV-1 infection of CD4+ T cells,” Bulletin of and modeling of human T cell kinetics,” European Journal of
Mathematical Biology, vol. 67, no. 1, pp. 101–114, 2005. Immunology, vol. 33, no. 8, pp. 2316–2326, 2003.
12 Abstract and Applied Analysis

[34] R. L. Iman and J. C. Helton, “An investigation of uncertainty

and sensitivity analysis techniques for computer models,” Risk
Analysis, vol. 8, no. 1, pp. 71–90, 1988.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 153139, 9 pages
http://dx.doi.org/10.1155/2014/153139

Research Article
Bifurcation of Traveling Wave Solutions of
the Dual Ito Equation

Xinghua Fan and Shasha Li

Faculty of Science, Jiangsu University, Zhenjiang, Jiangsu 212013, China

Correspondence should be addressed to Xinghua Fan; fan131@vip.sina.com

Received 9 May 2014; Accepted 15 July 2014; Published 5 August 2014

Academic Editor: Yonghui Xia

Copyright © 2014 X. Fan and S. Li. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

The dual Ito equation can be seen as a two-component generalization of the well-known Camassa-Holm equation. By using the
theory of planar dynamical system, we study the existence of its traveling wave solutions. We find that the dual Ito equation has
smooth solitary wave solutions, smooth periodic wave solutions, and periodic cusp solutions. Parameter conditions are given to
guarantee the existence.

1. Introduction can be seen as the dual equation of the KdV equation

Study on two-component equations has drawn a lot of interest 𝑢𝑡 = 𝑢𝑥𝑥𝑥 + 3𝑢𝑢𝑥 . (4)
among researchers [1–6]. The two-component equation we
are going to discuss in the present paper is the dual Ito
equation [7] Taking plus sign in (3) leads to an integrable equation which
supports compactons, whereas minus sign is the water wave
1
𝑢𝑡 ± 𝑢𝑥𝑥𝑡 = 3𝑢𝑢𝑥 + VV𝑥 + (𝑢𝑢𝑥𝑥 + 𝑢𝑥2 ) model derived by Camassa-Holm, whose bounded travelling
2 𝑥 (1) waves (termed peakons) develop a discontinuity in the first
derivatives [10].
V𝑡 = (𝑢V)𝑥 .
Different methods have been used to study exact solutions
The dual Ito equation (1) is of interest due to the connection to standard Ito equation or generalized ones as well as higher-
of the Ito equation with the KdV equation. order Ito equation [11–17]. But there are fewer researches
The Ito equation [8], about traveling wave solutions of (1). We want to know, for (1),
whether there exist some interesting solutions such as smooth
𝑢𝑡 − 𝑢𝑥𝑥𝑥 = 3𝑢𝑢𝑥 + VV𝑥
(2) soliton, peakon, cuspon, and compacton [18] solutions.
V𝑡 = (𝑢V)𝑥 , In this paper, we will apply the method of dynamical
system [19–24] from a mathematical point of view to study
is a prototypical example of a two-component KdV equation. the traveling wave solutions of (1).
It is shown that the coupled equation possesses infinitely The rest of this paper is organized as follows. In Section 2,
many symmetries and conservation laws. It is also shown that after transforming the two-component dual Ito equation
these symmetries define a hierarchy of the coupled equation into a planar dynamical system, we discuss the bifurcation
each of which is a Hamiltonian system. conditions and possible phase portraits of the planar system.
By the sense of “tri-Hamiltonian duality” [7], the cele- Based on those phase portraits, Section 3 presents different
brated Camassa-Holm equation [9, 10] types of solutions such as smooth solitary wave solutions,
1 smooth periodic wave solutions, and periodic cusp solutions.
𝑢𝑡 ± 𝑢𝑥𝑥𝑡 = 3𝑢𝑢𝑥 + (𝑢𝑢𝑥𝑥 + 𝑢𝑥2 ) (3) The last section is devoted to a short conclusion.
2 𝑥
2 Abstract and Applied Analysis

2. Bifurcation Conditions and Possible taking minus sign. To avoid the singularity, letting 𝑑𝜉 = 2(𝑐 ±
Phase Portraits 𝜙)(𝑐 + 𝜙)2 𝑑𝜏, system (10) is changed to a regular system:

In this section, the properties of equilibrium points and

possible phase portraits will be given. 𝑑𝜙 2
= 2 (𝑐 ± 𝜙) (𝑐 + 𝜙) 𝑦,
We consider the traveling wave solutions of (1) in the form 𝑑𝜏
𝑑𝑦 2 (11)
𝑢 (𝑥, 𝑡) = 𝜙 (𝜉) , V (𝑥, 𝑡) = 𝜓 (𝜉) , 𝜉 = 𝑥 − 𝑐𝑡, (5) = ∓ ((𝜙 + 𝑐) 𝑦2 + 3𝜙4 + 8𝑐𝜙3 + (7𝑐2 + 2𝑔) 𝜙2
𝑑𝜏
where 𝑐 is the wave speed. + (4𝑐𝑔 + 2𝑐3 ) 𝜙 + 𝑟2 + 2𝑔𝑐2 ) .
Substituting (5) into (1), we get a system of ordinary
differential equations
Now we consider the equilibrium points of system (10)
𝑐𝜙󸀠 ± 𝑐𝜙󸀠󸀠󸀠 + 3𝜙𝜙󸀠 + 𝜓𝜓󸀠 + 2𝜙󸀠 𝜙󸀠󸀠 + 𝜙𝜙󸀠󸀠󸀠 = 0, lying on the 𝜙-axis. Let
(6)
𝑐𝜓󸀠 + 𝜓𝜙󸀠 + 𝜙𝜓󸀠 = 0,
𝐹 (𝜙) = 3𝜙4 + 8𝑐𝜙3 + (2𝑔 + 7𝑐2 ) 𝜙2
󸀠
where “ ” is the derivative with respect to 𝜉.
Integrating (6) once with respect to 𝜉, we obtain + (4𝑔𝑐 + 2𝑐3 ) 𝜙 + 2𝑔𝑐2 + 𝑟2 , (12)

3 1 1 𝐺 (𝜙) = 𝜙4 + 2𝑐𝜙3 + (2𝑔 + 𝑐2 ) 𝜙2 + 2𝑔𝑐𝜙 − 𝑟2 .

𝑐𝜙 + 𝜙2 + 𝜙󸀠2 + (𝑐 + 𝜙) 𝜙󸀠󸀠 + 𝜓2 + 𝑔 = 0,
2 2 2 (7)
𝜙𝜓 + 𝑐𝜓 + 𝑟 = 0, We see that 𝐹(𝜙) has at most four real roots 𝜙𝑖 , 𝑖 = 1, . . . , 4,
since it is a polynomial with order four. Then system (10) has
where 𝑔, 𝑟 are integration constants. at most four equilibrium points 𝐸𝑖 (𝜙𝑖 , 0).
From the second equation of (7), we get We need to find the bifurcation conditions for the
parameters. Equation 𝐹󸀠 (𝜙) = 0 has three roots −𝑐, 𝜙± =
𝑟
𝜓=− . (8) −𝑐/2 ± √3𝑐2 − 12𝑔/6, if 𝑔 < (1/4)𝑐2 . It means we only need
𝑐 + 𝜙 (𝜉)
to discuss the situation when 𝑔 < (1/4)𝑐2 . Thus we get a
Plugging (8) into the first equation of (7), we obtain a bifurcation condition 𝑔 = (1/4)𝑐2 . When 𝑔 = −(1/2)𝑐2 , we
nonlinear ODE have 𝜙− = −𝑐. Therefore 𝑔 = −(1/2)𝑐2 is another bifurcation
2 2 condition.
(2𝜙3 + 𝑎𝑐𝜙2 + 𝑎𝑐2 𝜙 + 2𝑐3 ) 𝜙󸀠󸀠 + (𝜙 + 𝑐) (𝜙󸀠 ) + 3𝜙4 + 8𝑐𝜙3 Next we will study possible order of those roots. We see
that 𝐹(−𝑐) = 𝑟2 > 0, and 𝐹(𝜙+ ) − 𝐹(𝜙− ) = (𝑐/9)(4𝑔 −
+ (7𝑐2 + 2𝑔) 𝜙2 + (4𝑐𝑔 + 2𝑐3 ) 𝜙 + 𝑟2 + 2𝑔𝑐2 = 0,
𝑐2 )√3𝑐2 − 12𝑔 < 0. If 𝐹(𝜙+ ) ≤ 0 and 𝑔 < (1/4)𝑐2 , then
(9)
𝐹󸀠󸀠 (−𝑐) = 2𝑐2 + 4𝑔, 𝐹󸀠󸀠 (𝜙+ ) = 2𝑐2 + 2𝑐√3𝑐2 − 12𝑔 − 8𝑔, and
where 𝑎 = 6 when (1) takes plus sign and 𝑎 = 2, minus sign.
Letting 𝑦 = 𝜙󸀠 , we get the following planar system: 𝐹󸀠󸀠 (𝜙− ) = 2𝑐2 − 2𝑐√3𝑐2 − 12𝑔 − 8𝑔. When −(1/2)𝑐2 ≤ 𝑔 <
(1/4)𝑐2 , we have −𝑐 ≤ 𝜙− ; then −𝑐 < 𝜙− < 𝜙1 < 𝜙+ < 𝜙2 .
𝑑𝜙 Furthermore, if 𝐹(𝜙+ ) = 0, 𝜙1 meets 𝜙+ . If 𝐹(𝜙+ ) > 0, there
=𝑦
𝑑𝜉 is no zeros for 𝐹(𝜙). If 𝑔 > (1/4)𝑐2 , there is only one zero −𝑐
𝑑𝑦 for 𝐹󸀠 , and there is no equilibrium point. If 𝑔 = (1/4)𝑐2 , then
= ((3𝜙4 + 8𝑐𝜙3 + (𝑦2 + 7𝑐2 + 2𝑔) 𝜙2 𝜙+ = 𝜙− . There are two zeros −𝑐 and −(1/2)𝑐 for 𝐹󸀠 . Because
𝑑𝜉 (10) 𝐹󸀠󸀠 (𝜙± ) = 0, 𝐹󸀠󸀠󸀠 (𝜙± ) = 12𝑐, there are zeros for 𝐹(𝜙).
2 2 2 2 2 2 Now consider singular equilibrium points on the singular
+2𝑐 (𝑦 + 𝑐 + 2𝑔) 𝜙 + 𝑦 𝑐 + 2𝑔𝑐 + 𝑟 )
lines. On the singular line 𝜙 = −𝑐 there are no singular
2 −1 equilibria if 𝑟 ≠0. When (1) takes the minus sign and 𝐹(𝑐) =
× (2 (𝑐 ± 𝜙) (𝜙 + 𝑐) ) ) .
20𝑐4 + 8𝑔𝑐2 + 𝑟2 < 0, on the singular line 𝜙 = 𝑐, there are two
singular equilibria 𝐸5 ,6 = (𝑐, ±𝑦𝑐 ) where 𝑦𝑐 = √−𝐹(𝑐)/2𝑐.
System (10) is a planar dynamical system defined in a To investigate the equilibrium points of (11), let 𝑀(𝜙𝑒 , 𝑦𝑒 )
3-parameter space (𝑐, 𝑔, and 𝑟). Because the phase orbits be the coefficient matrix of the linearized system of sys-
defined by the vector fields of system (10) determine all tem (11) at the equilibrium point (𝜙𝑒 , 𝑦𝑒 ); 𝐽(𝜙𝑒 , 𝑦𝑒 ) =
traveling wave solutions, we will investigate the bifurcations det 𝑀(𝜙𝑒 , 𝑦𝑒 ) is the determinant. We have
of phase portraits of these systems in the phase plane (𝜙, 𝑦)
as the parameters are changed.
System (10) has a singular straight line 𝜙 = −𝑐 when 3 𝑑
𝐽 (𝜙, 𝑦) = −2(𝑐 + 𝜙) (4𝑦2 𝑐 + 4𝑦2 𝜙 − 𝐹 (𝜙)) (13)
taking plus sign or two singular straight lines 𝜙 = ±𝑐 when 𝑑𝜙
Abstract and Applied Analysis 3

corresponding to plus sign in (1) and is a homoclinic orbit defined by 𝐻(𝜙, 𝑦) = ℎ1 passing
through 𝐸1 . There is a family of smooth periodic orbits
2
𝐽 (𝜙, 𝑦) = −2(𝜙 + 𝑐) of (10) defined by 𝐻(𝜙, 𝑦) = ℎ, ℎ ∈ (ℎ2 , ℎ1 ) enclosing
the center 𝐸2 (see Figure 1(c)).

𝑑 𝑑
× (4𝑦2 𝑐𝜙 + 𝑐 𝐹 (𝜙) + 4𝑦2 𝜙2 − 𝜙 𝐹 (𝜙)) 2.2. Equilibrium Points and Phase Portraits of (10) When
𝑑𝜙 𝑑𝜙 Taking Minus Sign in (1). When taking minus sign, there
(14) are two singular lines of (10). Two more equilibrium points
appear. When 𝐹(𝑐) > 0, no equilibrium points exist on the
corresponding to minus sign in (1). We can get 𝐽(𝜙, 0) = 2(𝜙± singular line 𝜙 = 𝑐. If 𝐹(𝜙+ ) < 0 and −(1/2)𝑐2 < 𝑔 < (1/4)𝑐2 ,
𝑐)(𝑐 + 𝜙)2 (𝑑/𝑑𝜙)𝐹(𝜙) and 𝑝 = trace(𝐽) = 4(𝑐 ± 𝜙)(𝑐 + 𝜙)𝑦, we have −𝑐 < 𝜙− < 𝜙1 < 𝜙+ < 𝜙2 . When 𝑔 = −(1/2)𝑐2 , we
respectively. have 𝜙− = −𝑐 < 𝜙1 < 𝜙+ < 𝜙2 . If 𝐹(𝜙+ ) = 0, there is only one
By the theory of planar dynamical systems [22], we know equilibrium point 𝐸1 (𝜙1 , 0), (−𝑐 < 𝜙− < 𝜙1 = 𝜙+ ) which is a
that an equilibrium point of a planar Hamiltonian system is a saddle piont. If 𝐹(𝜙+ ) > 0, there are no equilibrium points.
saddle point in the case of 𝐽 < 0, a center in the case of 𝐽 > 0,
and a cusp in the case of 𝐽 = 0, and its Poincare index is 0. Lemma 3. When 𝐹(𝑐) > 0, 𝐹(𝜙+ ) < 0 and −(1/2)𝑐2 ≤ 𝑔 <
(1/4)𝑐2 , system (10) has two equilibrium points 𝐸1 (𝜙1 , 0) and
2.1. Equilibrium Points and Phase Portraits of (10) When 𝐸2 (𝜙2 , 0). 𝐸1 is a center point while 𝐸2 is a saddle point. There
Taking Plus Sign in (1). Except for the straight line 𝜙 = −𝑐 is a homoclinic orbit defined by 𝐻(𝜙, 𝑦) = ℎ2 to the saddle 𝐸2 .
or 𝜙 = ±𝑐, systems (10) and (11) have the same first integral There is a family of smooth periodic orbits defined by 𝐻(𝜙, 𝑦 =
ℎ, ℎ ∈ (ℎ1 , ℎ2 )) surrounding 𝐸2 (see Figure 2(a)).
𝑏𝑦2 ± (𝜙4 + 2𝑐𝜙3 + (𝑐2 + 2𝑔) 𝜙2 + 2𝑔𝑐𝜙 − 𝑟2 )
𝐻 (𝜙, 𝑦) = When 𝐹(𝑐) < 0, there is equilibrium at 𝜙 = 𝑐. When 𝑔 <
𝑐+𝜙
−(1/2)𝑐2 , we have 𝜙+ > 0 and 𝜙− < −𝑐 < 𝜙+ ; then 𝑓󸀠󸀠 (−𝑐) > 0,
where −𝑐 is minimum.
= ℎ,
Lemma 4. When 𝐹(𝑐) < 0 and 𝑔 < −(1/2)𝑐2 , the equilibrium
(15)
points of system (10) can be described by the following cases.
where 𝑏 = (𝜙 + 𝑐)2 corresponding to plus sign in (1) and 𝑏 = (1) If 𝐹(𝜙− ) < 0 and 𝐹(𝜙+ ) < 0, there are six
𝑐2 − 𝜙2 corresponding to minus sign. equilibrium points 𝐸𝑖 (𝜙𝑖 , 0), 𝑖 = 1, . . . , 4, 𝐸5 (𝑐, 𝑦𝑐 ), and
Let ℎ𝑖 = 𝐻(𝜙𝑖 , 0), ℎ𝑠 = 𝐻(𝑐, ±𝑦𝑐 ). We can get ℎ𝑠 = 𝐸6 (𝑐, −𝑦𝑐 ) (𝜙1 < 𝜙− < 𝜙2 < −𝑐 < 𝜙3 < 𝜙+ < 𝜙4 ).
−𝐺(𝑐)/2𝑐, where 𝐺(𝑐) = 4𝑐4 + 4𝑔𝑐2 − 𝑟2 . 𝐸1 , 𝐸3 , and 𝐸4 are centers; 𝐸2 , 𝐸5 , and 𝐸6 are saddle
System (11) has the same phase portraits as system (10) points. Equilibrium points 𝐸5 and 𝐸6 lie on the singular
except for 𝜙 = −𝑐 corresponding to plus sign or 𝜙 = ±𝑐 line 𝜙 = 𝑐. There is a homoclinic orbit passing through
corresponding to minus sign. 𝐸2 enclosing the center 𝐸1 . There is a family of periodic
orbits surrounding each center point. There is a singular
Lemma 1. When 𝐹(𝜙+ ) < 0 and −(1/2)𝑐2 ≤ 𝑔 < (1/4)𝑐2 , close orbit passing the singular saddle points 𝐸5 and
there are two equilibrium points 𝐸1 (𝜙1 , 0) and 𝐸2 (𝜙2 , 0) (𝜙1 < 𝐸6 (Figure 2(b)).
𝜙2 ). 𝐸1 is a saddle point and 𝐸2 is a center point. In this (2) If 𝐹(𝜙− ) > 0 and 𝐹(𝜙+ ) < 0, there are four equilibrium
parameter condition, a branch of the level curve 𝐻(𝜙, 𝑦) = ℎ1 points 𝐸𝑖 (𝜙1 , 0), 𝑖 = 1, 2, 𝐸3 (𝑐, 𝑦𝑐 ), and 𝐸4 (𝑐, −𝑦𝑐 )
defines a homoclinic orbit to the saddle point 𝐸1 , and a branch (𝜙− < −𝑐 < 𝜙1 < 𝜙+ < 𝜙2 ). 𝐸1 and 𝐸2 are center
of the level curve 𝐻(𝜙, 𝑦) = ℎ with ℎ ∈ (ℎ2 , ℎ1 ) gives rise to a points while 𝐸3 and 𝐸4 are saddle points on the singular
family of smooth periodic orbits of (10) (see Figure 1(a)). line 𝜙 = 𝑐. There is a family of smooth periodic orbits
surrounding the centers 𝐸1 and 𝐸2 , respectively. There
Lemma 2. When 𝑔 < −(1/2)𝑐2 , the equilibrium points of are two bizarre periodic orbits passing through 𝐸3 and
system (10) can be described by the following cases. 𝐸4 (see Figure 2(c)).
(1) If 𝐹(𝜙− ) < 0 and 𝐹(𝜙+ ) < 0, there are four equilibrium
points 𝐸𝑖 (𝜙𝑖 , 0), 𝑖 = 1, . . . , 4, (𝜙1 < 𝜙− < 𝜙2 < −𝑐 < 3. Different Kinds of Traveling Solutions to (1)
𝜙3 < 𝜙+ < 𝜙4 ). 𝐸1 and 𝐸4 are centers and 𝐸2 and 𝐸3 In this section we will give some types of interesting solutions
are saddle points. There is a homoclinic orbit defined to (1).
by 𝐻(𝜙, 𝑦) = ℎ𝑖 , 𝑖 = 2, 3, passing through 𝐸2 and 𝐸3 , Suppose that 𝑢 = 𝜙(𝜉) is a traveling wave solution for (9)
respectively. There is a family of smooth periodic orbits for 𝜉 ∈ (−∞, ∞), lim𝜉 → −∞ 𝜙(𝜉) = 𝛼, and lim𝜉 → ∞ 𝜙(𝜉) = 𝛽,
of (10) enclosing the center 𝐸1 and 𝐸4 , respectively (see where 𝛼 and 𝛽 are two constants. If 𝛼 = 𝛽, then 𝜙(𝜉) is called
Figure 1(b)). a solitary wave solution. Usually, a solitary wave solution for
(2) If 𝐹(𝜙− ) > 0 and 𝐹(𝜙+ ) < 0, there are two equilibrium (9) corresponds to a homoclinic orbit of system (10) while
points 𝐸1 (𝜙1 , 0) and 𝐸2 (𝜙2 , 0) (𝜙− < −𝑐 < 𝜙1 < 𝜙+ < a periodic solution for (9) corresponds to a closed orbit of
𝜙2 ). 𝐸1 is a saddle point while 𝐸2 is a center point. There system (10).
4 Abstract and Applied Analysis

4 40

3 30

y 2 y 20

1 10

0 0
−1.5 −1.0 −0.5 0 −10 −5 0 5 10
𝜙 −1 𝜙 −10

−2 −20

−3 −30

−4 −40
(a) (b)
10

y 5

0
−1 0 1 2 3 4
𝜙

−5

−10
(c)

Figure 1: Phase portraits of (10) for 𝜙 ≠−𝑐. (a) The case for −(1/2)𝑐2 ≤ 𝑔 < (1/4)𝑐2 , 𝐹(𝜙+ ) < 0. (b) The case for 𝑔 < −(1/2)𝑐2 , 𝐹(𝜙− ) <
0, 𝐹(𝜙+ ) < 0. (c) The case for 𝑔 < −(1/2)𝑐2 , 𝐹(𝜙− ) > 0, 𝐹(𝜙+ ) < 0.

By using the results of the above lemmas and the basic 𝐸1 in Figures 1(a) and 1(c) and by 𝐻(𝜙, 𝑦) = ℎ3 to the saddle
theory of the singular nonlinear traveling wave equations point 𝐸3 in Figure 1(b). For simplicity, we only give one planar
[22], we obtain the dynamical behavior of the traveling wave profile of the first component 𝑢 in Figure 3(a). The second
solutions of (1) as follows. component of (1) is then given according to (8).

Proposition 6. There exists a smooth valley-shape solitary

3.1. Solitary Waves Solutions to (1)
wave solution of the first component of (1) and one of the
Proposition 5. There exists a smooth bell-shape solitary wave following conditions is satisfied:
solution of the first component of (1) if one takes plus sign in (1)
and one of the following conditions is satisfied: (1) 𝑔 < −(1/2)𝑐2 , 𝐹(𝜙− ) < 0, 𝐹(𝜙+ ) < 0 and taking plus
(1) −(1/2)𝑐2 ≤ 𝑔 < (1/4)𝑐2 , 𝐹(𝜙+ ) < 0; sign in (1);

(2) 𝑔 < −(1/2)𝑐2 , 𝐹(𝜙− ) < 0, 𝐹(𝜙+ ) < 0; (2) 𝐹(𝑐) > 0, −(1/2)𝑐2 ≤ 𝑔 < (1/4)𝑐2 , 𝐹(𝜙+ ) < 0 and
2
(3) 𝑔 < −(1/2)𝑐 , 𝐹(𝜙− ) > 0, 𝐹(𝜙+ ) < 0. taking minus sign in (1);

Those solitary wave solutions are corresponding to the (3) 𝐹(𝑐) < 0, 𝑔 < −(1/2)𝑐2 , 𝐹(𝜙− ) < 0, 𝐹(𝜙+ ) < 0 and
homoclinic orbit given by 𝐻(𝜙, 𝑦) = ℎ1 to the saddle point taking minus sign in (1).
Abstract and Applied Analysis 5

40

y 20 y
2

0 0
−1.5 −1 −0.5 0 0.5 1 1.5 2 −15 −10 −5 0 5 10 15
𝜙 𝜙
−2
−20

−4

−40
(a) (b)

10

y 5

0
−1 0 1 2 3 4
𝜙

−5

−10
(c)

Figure 2: Phase portraits of (10) for 𝜙 ≠−𝑐, taking minus sign. (a) The case for 𝐹(𝑐) > 0, −(1/2)𝑐2 ≤ 𝑔 < (1/4)𝑐2 , 𝐹(𝜙+ ) < 0. (b) The case for
𝐹(𝑐) < 0, 𝑔 < −(1/2)𝑐2 , 𝐹(𝜙− ) < 0, 𝐹(𝜙+ ) < 0. (c) The case for 𝐹(𝑐) < 0, 𝑔 < −(1/2)𝑐2 , 𝐹(𝜙− ) > 0, 𝐹(𝜙+ ) < 0.

These smooth valley-shape solitary wave solutions are corre- (4) 𝑔 < −(1/2)𝑐2 , 𝐹(𝜙− ) > 0, 𝐹(𝜙+ ) < 0 and taking plus
sponding to the homoclinic orbit given by 𝐻(𝜙, 𝑦) = ℎ2 to sign in (1);
the saddle point 𝐸2 in Figures 3(a), 2(a), and 2(b). A planar
(5) 𝐹(𝑐) > 0, −(1/2)𝑐2 ≤ 𝑔 < (1/4)𝑐2 , 𝐹(𝜙+ ) < 0 and
profile of the first component 𝑢 is shown in Figure 4(a).
taking minus sign in (1);
(6) 𝐹(𝑐) < 0, 𝑔 < −(1/2)𝑐2 , 𝐹(𝜙− ) < 0, 𝐹(𝜙+ ) < 0 and
3.2. Smooth Periodic Solutions
taking minus sign in (1);
Proposition 7. There is a family of smooth periodic wave (7) 𝐹(𝑐) < 0, 𝑔 < −(1/2)𝑐2 , 𝐹(𝜙− ) < 0, 𝐹(𝜙+ ) < 0 and
solutions of (1) if one of the following conditions is satisfied: taking minus sign in (1);
(1) −(1/2)𝑐2 ≤ 𝑔 < (1/4)𝑐2 , 𝐹(𝜙+ ) < 0 and taking plus (8) 𝐹(𝑐) < 0, 𝑔 < −(1/2)𝑐2 , 𝐹(𝜙− ) < 0, 𝐹(𝜙+ ) < 0 and
sign in (1); taking minus sign in (1);
(2) 𝑔 < −(1/2)𝑐2 , 𝐹(𝜙− ) < 0, 𝐹(𝜙+ ) < 0 and taking plus (9) 𝐹(𝑐) < 0, 𝑔 < −(1/2)𝑐2 , 𝐹(𝜙− ) > 0, 𝐹(𝜙+ ) < 0 and
sign in (1); taking minus sign in (1);
(3) 𝑔 < −(1/2)𝑐2 , 𝐹(𝜙− ) < 0, 𝐹(𝜙+ ) < 0 and taking plus (10) 𝐹(𝑐) < 0, 𝑔 < −(1/2)𝑐2 , 𝐹(𝜙− ) > 0, 𝐹(𝜙+ ) < 0 and
sign in (1); taking minus sign in (1).
6 Abstract and Applied Analysis

−3 −2 −1 0 1 2 3

1.5 −0.4

1.0 −0.6

−0.8
0.5
−1.0
0
−3 −2 −1 0 1 2 3 −1.2

−0.5
−1.4

−1.0 −1.6

−1.8
(a) Planar profiles of solutions to 𝑢 (b) Planar profiles of solutions to V

Figure 3: Smooth bell-shape solitary wave solution.

−2 −1 0 1 2
9

8
−4

7
−5
6

−6 5

4
−7
3

−8
2

−2 −1 0 1 2
(a) Planar profiles of solutions to 𝑢 (b) Planar profiles of solutions to V

Figure 4: Smooth valley-shape solitary wave solution.

Those periodic traveling wave solutions correspond to the Proposition 8. There is a peaked periodic cusp wave solution
family of smooth periodic orbits surrounding the centers in if one takes minus sign in (1) and one of the following conditions
Figures 1 and 2. A planar profile of the first component 𝑢 is is satisfied:
shown in Figure 5(a).
(1) 𝐹(𝑐) < 0, 𝑔 < −(1/2)𝑐2 , 𝐹(𝜙− ) < 0, 𝐹(𝜙+ ) < 0;
3.3. Nonsmooth Periodic Wave Solution. In Figures 1(b) and (2) 𝐹(𝑐) < 0, 𝑔 < −(1/2)𝑐2 , 𝐹(𝜙− ) > 0, 𝐹(𝜙+ ) < 0.
1(c), the singular straight line 𝜙 = 𝑐 intersects with the
close orbit. By theory of the singular nonlinear traveling wave These peaked periodic cusp wave solutions are corresponding
equations [22], there are nonsmooth wave solutions to (1). to the arch curve in the left side of 𝜙 = 𝑐 passing through
Abstract and Applied Analysis 7

−2 −1 0 1 2 3 4 5 6 7
1.4
−0.3

1.2

1.0

0.8 −0.4

0.6

0.4
−0.5
0.2

0
−2 −1 0 1 2 3 4 5 6 7
−0.2
−0.6

(a) Planar profiles of solutions to 𝑢 (b) Planar profiles of solutions to V

Figure 5: Smooth periodic wave solutions.

−0.5 0.0 0.5 1.0 1.5 2.0 2.5

2

−4

1 −6

−8

0
−0.5 0 0.5 1.0 1.5 2.0 2.5 −10

−12

−1
−14
(a) Planar profiles of solutions to 𝑢 (b) Planar profiles of solutions to V

Figure 6: Peaked periodic cusp wave solution.

the singular saddle points 𝐸5 and 𝐸6 and surrounding the Those valley-shape periodic cusp wave solutions are corre-
center 𝐸3 in Figure 2(b) and that passing through the singular sponding to the arch curve in the right side of 𝜙 = 𝑐 passing
saddle points 𝐸3 and 𝐸4 and surrounding the center 𝐸1 in through the singular saddle points 𝐸5 and 𝐸6 and surround-
Figure 2(c), respectively. Profiles are shown in Figure 6(a). ing the center 𝐸4 in Figure 2(b) and that passing through the
singular saddle points 𝐸3 and 𝐸4 and surrounding the center
Proposition 9. There is a valley-shape periodic cusp wave 𝐸2 in Figure 2(c). Profiles are shown in Figure 7(a).
solution if one takes minus sign in (1) and one of the following
conditions is satisfied. 4. Conclusions
(1) 𝐹(𝑐) < 0, 𝑔 < −(1/2)𝑐2 , 𝐹(𝜙− ) < 0, 𝐹(𝜙+ ) < 0; By using theory of the singular nonlinear traveling wave
equations, we found the existence of several different kinds
(2) 𝐹(𝑐) < 0, 𝑔 < −(1/2)𝑐2 , 𝐹(𝜙− ) > 0, 𝐹(𝜙+ ) < 0. of traveling wave solutions of (1). It is shown that the signs
8 Abstract and Applied Analysis

−1 0 1 2 3

−0.8
11
−1.0
10
−1.2
9

8 −1.4

7 −1.6

6 −1.8

5 −2.0

4
−2.2
3
−2.4

−1 0 1 2 3
(a) Planar profiles of solutions to 𝑢 (b) Planar profiles of solutions to V

Figure 7: Valley-shape periodic cusp wave solution.

had some influence on the type of solution. There are only [6] X. Fan, S. Yang, J. Yin, and L. Tian, “Bifurcations of travel-
smooth traveling wave solutions when taking plus sign. ing wave solutions for a two-component Fornberg-Whitham
Nonsmooth traveling wave solutions arise when the sign equation,” Communications in Nonlinear Science and Numerical
changes to minus. Furthermore, no peakons have been found Simulation, vol. 16, no. 10, pp. 3956–3963, 2011.
in our work although the two-component dual Ito equation is [7] P. Guha and P. J. Olver, “Geodesic flow and two (super)
analogous to the two-component Camassa-Holm equation. component analog of the Camassa-Holm equation,” Symmetry
Integrability and Geometry Methods and Applications, vol. 2,
article 054, 2006.
Conflict of Interests [8] M. Ito, “Symmetries and conservation laws of a coupled nonlin-
ear wave equation,” Physics Letters A, vol. 91, no. 7, pp. 335–338,
The authors declare that there is no conflict of interests 1982.
regarding the publication of this paper. [9] R. Camassa, D. D. Holm, and J. M. Hyman, “A new integrable
shallow water equation,” Advances in Applied Mechanics, vol. 31,
Acknowledgment pp. 1–33, 1994.
[10] R. Camassa and D. D. Holm, “An integrable shallow water
Research was supported by the National Natural Science equation with peaked solitons,” Physical Review Letters, vol. 71,
Foundation of China (no. 11026169). no. 11, pp. 1661–1664, 1993.
[11] C. A. Gomez S, “New traveling waves solutions to generalized
Kaup-KUPershmidt and Ito equations,” Applied Mathematics
References and Computation, vol. 216, no. 1, pp. 241–250, 2010.
[1] C. Guan and Z. Yin, “Global existence and blow-up phenomena [12] F. Khani, “Analytic study on the higher order Ito equations: new
for an integrable two-component Camassa-Holm shallow water solitary wave solutions using the Exp-function method,” Chaos,
system,” Journal of Differential Equations, vol. 248, no. 8, pp. Solitons & Fractals, vol. 41, no. 4, pp. 2128–2134, 2009.
2003–2014, 2010. [13] D. Li and J. Zhao, “New exact solutions to the (2 + 1)-
[2] G. Gui and Y. Liu, “On the global existence and wave-breaking dimensional Ito equation: extended homoclinic test technique,”
criteria for the two-component Camassa-Holm system,” Journal Applied Mathematics and Computation, vol. 215, no. 5, pp. 1968–
of Functional Analysis, vol. 258, no. 12, pp. 4251–4278, 2010. 1974, 2009.
[3] O. G. Mustafa, “On smooth traveling waves of an integrable [14] S. F. Tian and H. Q. Zhang, “Riemann theta functions periodic
two-component Camassa-Holm shallow water system,” Wave wave solutions and rational characteristics for the (1 + 1)-
Motion, vol. 46, no. 6, pp. 397–402, 2009. dimensional and (2 + 1)-dimensional Ito equation,” Chaos,
[4] A. Constantin and R. I. Ivanov, “On an integrable two- Solitons & Fractals, vol. 47, pp. 27–41, 2013.
component Camassa-Holm shallow water system,” Physics Let- [15] Z. Zhao, Z. Dai, and C. Wang, “Extend three-wave method for
ters A, vol. 372, no. 48, pp. 7129–7132, 2008. the (1 + 2)-dimensional Ito equation,” Applied Mathematics and
[5] J. B. Li and Y. S. Li, “Bifurcations of travelling wave solutions for Computation, vol. 217, no. 5, pp. 2295–2300, 2010.
a two-component Camassa-Holm equation,” Acta Mathematica [16] Y. Zhang, Y. C. You, W. X. Ma, and H. Q. Zhao, “Resonance
Sinica, vol. 24, no. 8, pp. 1319–1330, 2008. of solitons in a coupled higher-order Ito equation,” Journal of
Abstract and Applied Analysis 9

Mathematical Analysis and Applications, vol. 394, no. 1, pp. 121–

128, 2012.
[17] H. Zhao, “Soliton solution of a multi-component higher-order
Ito equation,” Applied Mathematics Letters, vol. 26, no. 7, pp. 681–
686, 2013.
[18] A. Chen, S. Wen, S. Tang, W. Huang, and Z. Qiao, “Effects of
quadratic singular curves in integrable 5 equations,” To appear
in Studies in Applied Mathematics.
[19] L. Perko, Differential Equations and Dynamical Systems, vol. 7
of Texts in Applied Mathematics, Springer, New York, NY, USA,
1991.
[20] G. Betchewe, B. B. Thomas, K. K. Victor, and K. T. Crepin,
“Dynamical survey of a generalized-Zakharov equation and
its exact travelling wave solutions,” Applied Mathematics and
Computation, vol. 217, no. 1, pp. 203–211, 2010.
[21] K. Gatermann and S. Hosten, “Computational algebra for
bifurcation theory,” Journal of Symbolic Computation, vol. 40,
no. 4-5, pp. 1180–1207, 2005.
[22] J. B. Li and H. H. Dai, On the Study of Singular Nonlinear
Travelling Wave Equations: Dynamical Approach, Science Press,
Beijing, China, 2007.
[23] Y. Feng, W. Shan, W. Sun, H. Zhong, and B. Tian, “Bifurcation
analysis and solutions of a three-dimensional Kudryashov-
Sinelshchikov equation in the bubbly liquid,” Communications
in Nonlinear Science and Numerical Simulation, vol. 19, no. 4, pp.
880–886, 2014.
[24] H. Liu and J. Li, “Symmetry reductions, dynamical behavior
and exact explicit solutions to the Gordon types of equations,”
Journal of Computational and Applied Mathematics, vol. 257, pp.
144–156, 2014.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 248061, 9 pages
http://dx.doi.org/10.1155/2014/248061

Research Article
Improved Stability Criteria for Markovian Jump Systems with
Time-Varying Delays

Yu-cai Ding, Hui Liu, and Baodan Tian

School of Science, Southwest University of Science and Technology, Mianyang 621010, China

Correspondence should be addressed to Yu-cai Ding; dycyer@163.com

Received 16 May 2014; Accepted 9 July 2014; Published 21 July 2014

Academic Editor: Yonghui Xia

Copyright © 2014 Yu-cai Ding et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

The delay-dependent stochastic stability problem of Markovian jump systems with time-varying delays is investigated in this
paper. Though the Lyapunov-Krasovskii functional is general and simple, less conservative results are derived by using the convex
combination method, improved Wirtinger’s integral inequality, and a slack condition on Lyapunov matrix. The obtained results are
formulated in terms of linear matrix inequalities (LMIs). Numerical examples are provided to verify the effectiveness and superiority
of the presented results.

1. Introduction has been paid to the delay-dependent stability criteria; see

[11–23], for example. As for delay-dependent stability, many
Markovian jump systems are a special class of stochastic methods have been taken for deriving stability criteria, such
hybrid systems. Many dynamical systems subject to random as free-weighting matrices methods [11], model transforma-
abrupt variations, such as mechanical systems, economics, tion techniques [12, 13], convex combination methods [14–
and systems with human operators, can be modeled by 17], delay decomposition approaches [18], multiple integral
Markovian jump systems [1]. Due to their extensive applica- approaches [19], and input-output approaches [20]. Recently,
tions in many files, the analysis and synthesis of Markovian the bounding techniques of the cross terms and integral terms
jump systems have received much research attention and lots in the derivatives of the Lyapunov-Krasovskii functional
of significant results have been reported; see, for example, [2– are widely investigated, such as improved Jensen’s integral
8] and the references therein. inequality [21], reciprocally convex approach [22, 23], and
Time delay is an inherent characteristic of many dynamic improved Wirtinger’s integral inequality [24]. Some less
systems such as networked control systems, industrial sys- conservative stability results have been derived by using the
tems, and process control systems. The systems with or above techniques.
without time delays are convergent when time delays are close In this paper, we develop some new stability criteria
to zero; otherwise, they may be divergent. In other words, by using an improved Wirtinger’s integral inequality and
time delays can degrade the performance of systems designed the convex combination method to deal with the cross
without considering the delays and can even destabilize terms and integral terms in the derivatives of the Lyapunov-
the systems. During the past few decades, considerable Krasovskii functional. In addition, the positive definiteness of
attention has been paid to the stability analysis of time- some Lyapunov matrix is not required. The obtained results
delay systems [9–16]. The existing stability criteria for linear can be applied to both slow and fast time-varying delays.
systems can be classified into two types: delay-independent The numerical examples demonstrate the effectiveness and
ones which are applicable to delays of arbitrary size and delay- superiority of the presented results.
dependent ones which include information on the size of
delays. In general, delay-dependent stability criteria are less Notation. Throughout this paper, (Ω, 𝐹, 𝑃) is a probability
conservative than delay-independent ones especially when space, Ω is the sample space, 𝐹 is the 𝜎-algebra of the sample
the size of the delay is small. Thus, considerable attention space, and 𝑃 is the probability measure on 𝐹. E{⋅} refers to
2 Abstract and Applied Analysis

the expectation operator with respect to some probability Lemma 3 (see [24]). If 𝜔(𝑦) is a differentiable function in
measure 𝑃. 𝐴 > 0 (<0) means 𝐴 is a symmetric positive (neg- [𝑎, 𝑏] → R𝑛 . Then for any given symmetric positive definite
ative) definite matrix and 𝐴−1 denotes the inverse of matrix matrix 𝑅 > 0, the following inequality holds:
𝐴. 𝐴𝑇 represents the transpose of 𝐴. Sym(𝑀) stands for
𝑀 + 𝑀𝑇 . The symbol ∗ in LMIs denotes the symmetric term
of the matrix. col{𝑋, 𝑌} represents a column vector formed
𝑏
𝑇 1 ]1 𝑇 ̃ ]1
∫ 𝜔̇ (𝑡) 𝑅𝜔̇(𝑡) 𝑑𝑡 ⩾ [ ] 𝑅[ ], (7)
by 𝑋 and 𝑌. Identity matrix, of appropriate dimensions, will 𝑎 𝑏 − 𝑎 ]2 ]2
be denoted by 𝐼. diag(⋅, ⋅) denotes a diagonal matrix. Ω(𝑖, 𝑗)
means the element in the 𝑖th row and 𝑗th column of the block 𝑏
matrix Ω. where ]1 = 𝜔(𝑏)−𝜔(𝑎), ]2 = 𝜔(𝑏)+𝜔(𝑎)−(2/(𝑏−𝑎)) ∫𝑎 𝜔(𝑡)𝑑𝑡,
̃ = diag(𝑅, 3𝑅).
and 𝑅
2. Problem Statement
Fix a probability space (Ω, 𝐹, 𝑃) and consider the following 3. Improved Stability Criterion
Markovian jump systems: In this section, we will present an improved stochastic stabil-
𝑥̇(𝑡) = 𝐴 (𝑟𝑡 ) 𝑥 (𝑡) + 𝐴 𝑑 (𝑟𝑡 ) 𝑥 (𝑡 − 𝑑 (𝑡)) , ity criterion in terms of LMIs by using Lyapunov-Krasovskii
(1) functional method and convex combination technique.
𝑥 (𝑡) = 𝜙 (𝑡) , 𝑡 ∈ [−𝑑2 , 0] , Before stating the main results, some notations are given.
Let
where 𝑥(𝑡) ∈ R𝑛 is the state vector and 𝜙(𝑡) is a compatible
vector-valued initial function defined on [−𝑑2 , 0]. {𝑟𝑡 , 𝑡 ⩾ 0}
is a continuous-time Markovian process taking values in a 𝑑12 = 𝑑2 − 𝑑1 ,
finite space 𝑆 = {1, 2, . . . , 𝑁}; 𝐴(𝑟𝑡 ) and 𝐴 𝑑 (𝑟𝑡 )(𝑟𝑡 = 𝑖, 𝑖 ∈ 𝑆) 𝑡 𝑡−𝑑1
are real constant matrices with appropriate dimensions which 𝜁 (𝑡) = col {𝑥 (𝑡) , ∫ 𝑥 (𝑠) d𝑠, ∫ 𝑥 (𝑠) d𝑠} ,
depend on 𝑟𝑡 . The time delay 𝑑(𝑡) satisfies 𝑡−𝑑1 𝑡−𝑑2

0 < 𝑑1 ⩽ 𝑑 (𝑡) ⩽ 𝑑2 , 𝑑̇
(𝑡) ⩽ 𝜇. (2) 𝜉 (𝑡) = col {𝑥 (𝑡) , 𝑥 (𝑡 − 𝑑 (𝑡)) , 𝑥 (𝑡 − 𝑑1 ) , 𝑥 (𝑡 − 𝑑2 ) ,
The evolution of the Markovian process {𝑟𝑡 , 𝑡 ⩾ 0} is governed
by the following transition probability: 1 𝑡 1 𝑡−𝑑1
∫ 𝑥 (𝑠) d𝑠, ∫ 𝑥 (𝑠) d𝑠,
𝜋 Δ + 𝑜 (Δ) , 𝑗 ≠𝑖, 𝑑1 𝑡−𝑑1 𝑑 (𝑡) − 𝑑1 𝑡−𝑑(𝑡)
Pr {𝑟𝑡+Δ = 𝑗 | 𝑟𝑡 = 𝑖} = { 𝑖𝑗 (3)
1 + 𝜋𝑖𝑖 Δ + 𝑜 (Δ) , 𝑗 = 𝑖, 1 𝑡−𝑑(𝑡)
∫ 𝑥 (𝑠) d𝑠} ,
where Δ > 0 and limΔ → 0 (𝑜(Δ)/Δ) = 0; 𝜋𝑖𝑗 ⩾ 0 for 𝑖 ≠𝑗 is the 𝑑2 − 𝑑 (𝑡) 𝑡−𝑑2
transition probability from mode 𝑖 at time 𝑡 to mode 𝑗 at time
𝑃11𝑖 𝑃12 𝑃13
𝑡 + Δ and 𝜋𝑖𝑖 = − ∑𝑁 𝜋𝑖𝑗 .
𝑗=1,𝑗=𝑖̸ P𝑖 = [ ∗ 𝑃22 𝑃23 ] , (𝑖 ∈ 𝑆) ,
For simplicity, when 𝑟𝑡 = 𝑖, 𝑖 ∈ 𝑆, the matrices 𝐴(𝑟𝑡 ) and
[ ∗ ∗ 𝑃33 ]
𝐴 𝑑 (𝑟𝑡 ) are denoted by 𝐴 𝑖 and 𝐴 𝑑𝑖 , respectively.
The following definition and lemmas are needed in the Γ1𝑇
proof of our main results.
𝐼 0 0 0 0 0 0
Definition 1 (see [2]). System (1) is stochastically stable, if, for = [0 0 0 0 𝑑1 𝐼 0 0 ],
any initial state (𝑥0 , 𝑟0 ), the following relation holds for any [ 0 0 0 0 0 (𝑑 (𝑡) − 𝑑1 ) 𝐼 (𝑑2 − 𝑑 (𝑡)) 𝐼]
initial condition (𝑥0 , 𝑟0 ):
𝑋11 𝑋12
lim E {‖𝑥(𝑡)‖2 | 𝑥0 , 𝑟0 } = 0. 𝑋=[ ],
(4) 𝑋21 𝑃22
𝑡 → +∞

Lemma 2 (see [22]). Let 𝑓1 , 𝑓2 , . . . , 𝑓𝑛 : R𝑚 → R have 𝐴 𝑖 𝐴 𝑑𝑖 0 0 0 0 0

positive values in an open subset D of R𝑚 . Then, the recipro- Γ2 = [ 𝐼 0 −𝐼 0 0 0 0] , (𝑖 ∈ 𝑆) .
cally convex combination of 𝑓𝑖 over D satisfies [ 0 0 𝐼 −𝐼 0 0 0]
(8)
1
min ∑ 𝑓 (𝑡) = ∑𝑓𝑖 (𝑡) + max∑𝑔𝑖,𝑗 (𝑡) (5)
{𝛼𝑖 |𝛼𝑖 >0,∑𝑖 𝛼𝑖 =1}
𝑖 𝛼𝑖 𝑖 𝑖
𝑔𝑖,𝑗 (𝑡)
𝑖=𝑗̸
Theorem 4. Given scalars 0 < 𝑑1 < 𝑑2 and 𝜇, the time-
subject to varying delay system (1) is stochastically stable, if there exist
matrix P𝑖 ∈ R3𝑛×3𝑛 ; symmetric positive definite matrices 𝑅1 ∈
{ 𝑓𝑖 (𝑡) 𝑔𝑖,𝑗 (𝑡) } R𝑛×𝑛 , 𝑅2 ∈ R𝑛×𝑛 , 𝑄1 ∈ R𝑛×𝑛 , 𝑄2 ∈ R𝑛×𝑛 , 𝑄3 ∈ R𝑛×𝑛 , and
𝑔 : R 𝑚
󳨀→ R, 𝑔 ≜ 𝑔 , [𝑔𝑖,𝑗 (𝑡) 𝑓𝑗 (𝑡) ] ⩾ 0 . (6)
{ 𝑖,𝑗 𝑗,𝑖 𝑖,𝑗 } 𝑀 ∈ R𝑛×𝑛 ; and any matrices 𝑋 ∈ R𝑛×𝑛 , 𝑁𝑙 ∈ R𝑛×𝑛 , and
{ [ ] } 𝐺𝑙 ∈ R𝑛×𝑛 (𝑙 = 1, 2) such that, for all 𝑖 ∈ 𝑆,
Abstract and Applied Analysis 3

𝑅2 − 𝑀 0 𝑋11 𝑋12
[ ∗ 3 (𝑅 − 𝑀) 𝑋 𝑋22 ]
Ξ1 = [
[ ∗
2 21 ] > 0,
] (9)
∗ 𝑅2 − 𝑀 0
[ ∗ ∗ 0 3 (𝑅2 − 𝑀)]
2
2𝑑12 2𝑑12
[𝑃11𝑖 + 2𝑑1 𝑅1 + 𝑅2 𝑃12 − 2𝑅1 𝑃13 − 𝑅 ]
[ 𝑑1 + 𝑑2 𝑑1 + 𝑑2 2 ]
[ 1 ]
Ξ2𝑖 = [
[ ∗ 𝑃22 + (𝑄1 + 2𝑅1 ) 𝑃23 ] > 0,
] (10)
[ 𝑑1 ]
[ 1 2 ]
∗ ∗ 𝑃33 + 𝑄2 + 𝑅2
[ 𝑑12 𝑑1 + 𝑑2 ]
[Ω1 ]5𝑛×5𝑛 [Ω2 ]5𝑛×𝑛 [Ω3 ]5𝑛×𝑛 [Ω4 ]5𝑛×𝑛 [Ω5 ]5𝑛×𝑛 [Ω6 ]5𝑛×𝑛
[ ∗ −12 (𝑅2 − 𝑀) −4𝑋22 0 0 0 ]
[ ]
[ ∗ ∗ −12 (𝑅2 − 𝑀) 0 0 0 ]
Ξ3𝑖 = [
[
] < 0, (11)
[ ∗ ∗ ∗ −𝑅1 0 0 ] ]
[ ∗ ∗ ∗ ∗ −𝑅2 0 ]
[ ∗ ∗ ∗ ∗ ∗ −𝑀 ]

[Ω1 ]5𝑛×5𝑛 ̂2]

[Ω ̂3]
[Ω ̂6]
[Ω4 ]5𝑛×𝑛 [Ω5 ]5𝑛×𝑛 [Ω
5𝑛×𝑛 5𝑛×𝑛 5𝑛×𝑛
[ ∗ −12 (𝑅2 − 𝑀) −4𝑋22 0 0 0 ]
[ ]
[ ∗ ∗ −12 (𝑅 − 𝑀) 0 0 0 ]
[
Ξ4𝑖 = [ 2 ] < 0, (12)
∗ ∗ ∗ −𝑅1 0 0 ]
[ ]
[ ∗ ∗ ∗ ∗ −𝑅2 0 ]
[ ∗ ∗ ∗ ∗ ∗ −𝑀 ]

𝑇
where Ω1 (4, 5) = −𝑑1 𝑃23 , Ω1 (5, 5) = −12𝑅1 ,
𝑁
Ω1 (1, 1) = Sym (𝑃11𝑖 𝐴 𝑖 + 𝑃12 ) + ∑𝜋𝑖𝑗 𝑃11𝑗 + 𝑄1 − 4𝑅1 , Ω2 (1, 1) = 𝑑12 (𝐴𝑇𝑖 𝑃13 + 𝑃23 ) ,
𝑗=1
Ω2 (2, 1) = 𝑑12 𝐴𝑇𝑑𝑖 𝑃13 + 2𝑋21
𝑇 𝑇
+ 2𝑋22 + 6𝑅6 − 6𝑀,
Ω1 (1, 2) = 𝑃11𝑖 𝐴 𝑑𝑖 ,
𝑇
Ω2 (3, 1) = −𝑑12 (𝑃23 − 𝑃33 ) + 6𝑅6 − 6𝑀,
Ω1 (1, 3) = −2𝑅1 − 𝑃12 + 𝑃13 ,
𝑇 𝑇 𝑇
Ω2 (4, 1) = −𝑑12 𝑃33 − 2𝑋21 + 2𝑋22 , Ω2 (5, 1) = 0,
Ω1 (1, 4) = −𝑃13 , Ω1 (1, 5) = 6𝑅1 + 𝑑1 (𝐴𝑇𝑖 𝑃12 + 𝑃22
𝑇
),
Ω3 (1, 1) = 0, Ω3 (2, 1) = 6𝑅2 − 6𝑀 − 2𝑋12 + 2𝑋22 ,
Ω1 (2, 2) = − (1 − 𝜇) 𝑄3 − 8𝑅2 + 8𝑀
Ω3 (3, 1) = 2𝑋12 + 2𝑋22 ,
+ Sym (𝑋11 + 𝑋12 − 𝑋21 − 𝑋22 − 𝑁1 + 𝐺1 ) ,
Ω3 (4, 1) = 6𝑅2 − 6𝑀, Ω3 (5, 1) = 0,
𝑇 𝑇 𝑇 𝑇
Ω1 (2, 3) = −𝑋11 − 𝑋12 − 𝑋21 − 𝑋22
Ω4 (1, 1) = 𝑑1 𝐴𝑇𝑖 𝑅1 , Ω4 (2, 1) = 𝑑1 𝐴𝑇𝑑𝑖 𝑅1 ,
− 2𝑅2 + 𝑁1 − 𝑁2𝑇 + 𝐺2𝑇 + 2𝑀,
Ω4 (3, 1) = Ω4 (4, 1) = Ω4 (5, 1) = 0,
Ω1 (2, 4) = −2𝑅2 − 𝑋11 + 𝑋12 + 𝑋21 − 𝑋22
Ω5 (1, 1) = 𝑑12 𝐴𝑇𝑖 𝑅2 , Ω5 (2, 1) = 𝑑12 𝐴𝑇𝑑𝑖 𝑅2 ,
− 𝑁3𝑇 − 𝐺1𝑇 + 𝐺3𝑇 + 2𝑀,
Ω5 (3, 1) = Ω5 (4, 1) = Ω5 (5, 1) = 0,
Ω1 (2, 5) = 𝑑1 𝐴𝑇𝑑𝑖 𝑃12 Ω6 (1, 1) = 0, Ω6 (2, 1) = 𝑁1 , Ω6 (3, 1) = 𝑁2 ,
Ω1 (3, 3) = −4𝑅1 − 𝑄1 + 𝑄2 + 𝑄3 − 4𝑅2 + 4𝑀 + Sym (𝑁2 ) , Ω6 (4, 1) = 𝑁3 , Ω6 (5, 1) = 0,
Ω1 (3, 4) = 𝑋11 − 𝑋12 + 𝑋21 − 𝑋22 + 𝑁3 − 𝐺2 , ̂ 2 (1, 1) = 0, ̂ 2 (2, 1) = 2𝑋𝑇 + 2𝑋𝑇 + 6𝑅6 − 6𝑀,
Ω Ω 21 22
𝑇 𝑇
Ω1 (3, 5) = 6𝑅1 + 𝑑1 (𝑃23 − 𝑃22 ), ̂ 2 (3, 1) = 6𝑅6 − 6𝑀,
Ω
Ω1 (4, 4) = −𝑄2 − 4𝑅2 + 4𝑀 − Sym (𝐺3 ) , ̂ 2 (4, 1) = −2𝑋𝑇 + 2𝑋𝑇 ,
Ω ̂ 2 (5, 1) = 0,
Ω
21 22
4 Abstract and Applied Analysis

̂ 3 (1, 1) = 𝑑12 (𝐴𝑇 𝑃13 + 𝑃23 ) ,

Ω Thus, if 𝑄3 > 0 and Ξ2𝑖 > 0, then there exists a sufficiently
𝑖
small 𝜖 > 0 such that 𝑉(𝑥𝑡 , 𝑖) ⩾ 𝜖‖𝑥𝑡 ‖2 .
̂ 3 (2, 1) = 𝑑12 𝐴𝑇 𝑃13 + 6𝑅2 − 6𝑀 − 2𝑋12 + 2𝑋22 ,
Ω 𝑑𝑖 We next show that L𝑉(𝑥𝑡 , 𝑖) ⩽ −𝜖‖𝑥(𝑡)‖2 for the
̂ 3 (3, 1) = −𝑑12 (𝑃23 − 𝑃𝑇 ) + 2𝑋12 + 2𝑋22 ,
Ω sufficiently small 𝜖. Let L be the weak infinitesimal
33
generator of the random process {𝑥𝑡 , 𝑟𝑡 }. Calculating the
̂ 3 (4, 1) = −𝑑12 𝑃𝑇 + 6𝑅2 − 6𝑀,
Ω ̂ 3 (5, 1) = 0,
Ω difference of 𝑉(𝑥𝑡 , 𝑖) along the trajectories of (1), we have
33

̂ 6 (1, 1) = 0,
Ω ̂ 6 (2, 1) = 𝐺1 ,
Ω Ω6 (3, 1) = 𝐺2 ,
𝑁
̂ 6 (4, 1) = 𝐺3 ,
Ω ̂ 6 (5, 1) = 0.
Ω L𝑉 (𝑥𝑡 , 𝑖) = 2𝜁𝑇 (𝑡) P𝑖 𝜁 ̇(𝑡) + 𝜁𝑇 (𝑡) ( ∑𝜋𝑖𝑗 P𝑗 ) 𝜁 (𝑡)
𝑗=1
(13)
𝑇
+ 𝑥 (𝑡) 𝑄1 𝑥 (𝑡)
Proof. Consider the Lyapunov-Krasovskii functional given
by + 𝑥𝑇 (𝑡 − 𝑑1 ) (−𝑄1 + 𝑄2 + 𝑄3 ) 𝑥 (𝑡 − 𝑑1 )
𝑉 (𝑥𝑡 , 𝑖) = 𝑉1 (𝑥𝑡 , 𝑖) + 𝑉2 (𝑥𝑡 , 𝑖) + 𝑉3 (𝑥𝑡 , 𝑖) , (14)
− (1 − 𝑑 ̇
(𝑡)) 𝑥𝑇 (𝑡 − 𝑑 (𝑡)) 𝑄3 𝑥 (𝑡 − 𝑑 (𝑡))
where
𝑉1 (𝑥𝑡 , 𝑖) = 𝜁𝑇 (𝑡) P (𝑟𝑡 ) 𝜁 (𝑡) , − 𝑥𝑇 (𝑡 − 𝑑2 ) 𝑄2 𝑥 (𝑡 − 𝑑2 ) + 𝑑12 𝑥̇
𝑇
(𝑡) 𝑅1 𝑥̇(𝑡)
𝑡 𝑡
𝑉2 (𝑥𝑡 , 𝑖) = ∫ 𝑥𝑇 (𝑠) 𝑄1 𝑥 (𝑠) d𝑠 2 𝑇
+ 𝑑12 𝑥̇ (𝑡) 𝑅2 𝑥̇(𝑡) − 𝑑1 ∫ 𝑇
𝑥̇ (𝑠) 𝑅1 𝑥̇(𝑠) d𝑠
𝑡−𝑑1 𝑡−𝑑1
𝑡−𝑑1
+∫ 𝑥𝑇 (𝑠) 𝑄2 𝑥 (𝑠) d𝑠 𝑡−𝑑1
𝑇
𝑡−𝑑2 − 𝑑12 ∫ 𝑥̇ (𝑠) 𝑅2 𝑥̇(𝑠) d𝑠,
𝑡−𝑑2
𝑡−𝑑1 (15)
+∫ 𝑥𝑇 (𝑠) 𝑄3 𝑥 (𝑠) d𝑠, (17)
𝑡−𝑑(𝑡)
0 𝑡
𝑇 ̇ = Γ 𝜉(𝑡).
where 𝜁𝑇 (𝑡) = 𝜉𝑇 (𝑡)Γ1 and 𝜁(𝑡)
𝑉3 (𝑥𝑡 , 𝑖) = 𝑑1 ∫ ∫ 𝑥̇ (𝑠) 𝑅1 𝑥̇(𝑠) d𝑠 d𝜃 2
−𝑑1 𝑡+𝜃 Using the Newton-Leibniz formula, for any 𝑁𝑙 and 𝐺𝑙 (𝑙 =
−𝑑1 𝑡 1, 2, 3) with appropriate dimensions, the following are true:
𝑇
+ 𝑑12 ∫ ∫ 𝑥̇ (𝑠) 𝑅2 𝑥̇(𝑠) d𝑠 d𝜃.
−𝑑2 𝑡+𝜃
𝑡−𝑑1
We first show, for some 𝜖 > 0, the Lyapunov-Krasovskii 0 = 2𝜉𝑇 (𝑡) N [𝑥 (𝑡 − 𝑑1 ) − 𝑥 (𝑡 − 𝑑 (𝑡)) − ∫ 𝑥̇(𝑠) d𝑠] ,
functional condition 𝑉(𝑥𝑡 , 𝑖) ⩾ 𝜖‖𝑥𝑡 ‖2 for any initial 𝑡−𝑑(𝑡)

condition (𝑥0 , 𝑟0 ). Note that 𝑄1 > 0, 𝑄2 > 0, 𝑅1 > 0, and 𝑡−𝑑(𝑡)

𝑅2 > 0; it follows easily from Jensen’s inequality that 0 = 2𝜉𝑇 (𝑡) G [𝑥 (𝑡 − 𝑑 (𝑡)) − 𝑥 (𝑡 − 𝑑2 ) − ∫ 𝑥̇(𝑠) d𝑠] ,
𝑡−𝑑2
𝑡
1
𝑉 (𝑥𝑡 , 𝑖) ⩾ 𝜁𝑇 (𝑡) P𝑖 𝜁 (𝑡) + ∫ 𝑥𝑇 (𝑠) d𝑠𝑄1 (18)
𝑑1 𝑡−𝑑1
𝑡
1 𝑡−𝑑1 𝑇 where N = [0 𝑁1𝑇 𝑁2𝑇 𝑁3𝑇 0 0 0]𝑇 , G =
×∫ 𝑥 (𝑠) d𝑠 + ∫ 𝑥 (𝑠) d𝑠𝑄2
𝑡−𝑑1 𝑑12 𝑡−𝑑2 [0 𝐺1𝑇 𝐺2𝑇 𝐺3𝑇 0 0 0]𝑇 .
𝑡−𝑑1 𝑡−𝑑1 It can be shown readily that there exists a matrix 𝑀 > 0
×∫ 𝑥 (𝑠) d𝑠 + ∫ 𝑥𝑇 (𝑠) 𝑄3 𝑥 (𝑠) d𝑠 such that
𝑡−𝑑2 𝑡−𝑑(𝑡)
𝑡
2 0 ⩽ 2𝜉𝑇 (𝑡) N [𝑥 (𝑡 − 𝑑1 ) − 𝑥 (𝑡 − 𝑑 (𝑡))]
+ (𝑑1 𝑥𝑇 (𝑡) − ∫ 𝑥𝑇 (𝑠) d𝑠)
𝑑1 𝑡−𝑑1
𝑡 + 2𝜉𝑇 (𝑡) G [𝑥 (𝑡 − 𝑑 (𝑡)) − 𝑥 (𝑡 − 𝑑2 )]
× 𝑅1 (𝑑1 𝑥 (𝑡) − ∫ 𝑥 (𝑠) d𝑠)
𝑡−𝑑1 𝑑 (𝑡) − 𝑑1 𝑇
+ 𝜉 (𝑡) N𝑀−1 N𝑇 𝜉 (𝑡)
𝑡−𝑑1 𝑑12
2𝑑12
+ (𝑑12 𝑥𝑇 (𝑡) − ∫ 𝑥𝑇 (𝑠) d𝑠)
𝑑22− 𝑑12
𝑡−𝑑2 𝑑2 − 𝑑 (𝑡) 𝑇
+ 𝜉 (𝑡) G𝑀−1 G𝑇 𝜉 (𝑡) (19)
𝑡−𝑑1 𝑑12
× 𝑅2 (𝑑12 𝑥 (𝑡) − ∫ 𝑥 (𝑠) d𝑠) 𝑡−𝑑1
𝑡−𝑑2 𝑇
+ 𝑑12 ∫ 𝑥̇ (𝑠) 𝑀𝑥̇(𝑠) d𝑠
𝑡−𝑑1 𝑡−𝑑(𝑡)
= 𝜁𝑇 (𝑡) Ξ2𝑖 𝜁 (𝑡) + ∫ 𝑥𝑇 (𝑠) 𝑄3 𝑥 (𝑠) d𝑠.
𝑡−𝑑(𝑡) 𝑡−𝑑(𝑡)
𝑇
+ 𝑑12 ∫ 𝑥̇ (𝑠) 𝑀𝑥̇(𝑠) d𝑠.
(16) 𝑡−𝑑2
Abstract and Applied Analysis 5

Adding equalities (18) to the right hand of (17) and consid- 𝜍2 (𝑡)
ering conditions (2) and (19), we have
𝑥 (𝑡 − 𝑑 (𝑡)) − 𝑥 (𝑡 − 𝑑2 )
L𝑉 (𝑥𝑡 , 𝑖) ⩽ 2𝜉𝑇 (𝑡) Γ1 P𝑖 Γ2 𝜉 (𝑡) =[ 2 𝑡−𝑑(𝑡) ].
𝑥 (𝑡 − 𝑑 (𝑡)) + 𝑥 (𝑡 − 𝑑2 ) − ∫ 𝑥 (𝑠) d𝑠
𝑁 [ 𝑑2 − 𝑑 (𝑡) 𝑡−𝑑2 ]
+ 𝜉𝑇 (𝑡) Γ1 ( ∑𝜋𝑖𝑗 P𝑗 ) Γ1𝑇 𝜉 (𝑡) + 𝑥𝑇 (𝑡) 𝑄1 𝑥 (𝑡) (23)
𝑗=1
Applying Lemma 2 to (22), it yields that if there exists a
𝑇
+ 𝑥 (𝑡 − 𝑑1 ) (−𝑄1 + 𝑄2 + 𝑄3 ) 𝑥 (𝑡 − 𝑑1 ) matrix 𝑋 = [ 𝑋 11 𝑋12
𝑋21 𝑋22 ] with appropriate dimensions such that
(9) holds, then
− (1 − 𝜇) 𝑥𝑇 (𝑡 − 𝑑 (𝑡)) 𝑄3 𝑥 (𝑡 − 𝑑 (𝑡)) 𝑡−𝑑1 𝑡−𝑑(𝑡)
𝑇 𝑇
𝑇
− 𝑑12 ∫ 𝑥̇ (𝑠) 𝑅2 𝑥̇(𝑠) d𝑠 − 𝑑12 ∫ 𝑥̇ (𝑠) 𝑅2 𝑥̇(𝑠) d𝑠
− 𝑥 (𝑡 − 𝑑2 ) 𝑄2 𝑥 (𝑡 − 𝑑2 ) + 𝑑12 𝑥̇
𝑇
(𝑡) 𝑅1 𝑥̇(𝑡) 𝑡−𝑑(𝑡) 𝑡−𝑑2
𝑇
2 𝑇 𝜍1 (𝑡) Σ 𝑋 𝜍1 (𝑡)
+ 𝑑12 𝑥̇ (𝑡) 𝑅2 𝑥̇(𝑡) ⩽ −[ ] [ ][ ].
𝜍2 (𝑡) ∗ Σ 𝜍2 (𝑡)
+ 2𝜉𝑇 (𝑡) N [𝑥 (𝑡 − 𝑑1 ) − 𝑥 (𝑡 − 𝑑 (𝑡))] (24)
Combining (20), (21), and (24), we have
+ 2𝜉𝑇 (𝑡) G [𝑥 (𝑡 − 𝑑 (𝑡)) − 𝑥 (𝑡 − 𝑑2 )]
𝑑 (𝑡) − 𝑑1 𝑇
L𝑉 (𝑥𝑡 , 𝑖) ⩽ 𝜉 (𝑡)
𝑑 (𝑡) − 𝑑1 𝑇 𝑑12
+ 𝜉 (𝑡) N𝑀−1 N𝑇 𝜉 (𝑡)
𝑑12 × (Υ1 + 𝑑12 Υ3𝑇 𝑅1 Υ3 + 𝑑12
2 𝑇
Υ3 𝑅2 Υ3 + N𝑀−1 N𝑇 )
𝑑2 − 𝑑 (𝑡) 𝑇 𝑑2 − 𝑑 (𝑡) 𝑇
+ 𝜉 (𝑡) G𝑀−1 G𝑇 𝜉 (𝑡) × 𝜉 (𝑡) + 𝜉 (𝑡)
𝑑12 𝑑12
𝑡
− 𝑑1 ∫ 𝑇
𝑥̇ (𝑠) 𝑅1 𝑥̇(𝑠) d𝑠 × (Υ2 + 𝑑12 Υ3𝑇 𝑅1 Υ3 + 𝑑12
2 𝑇
Υ3 𝑅2 Υ3 + G𝑀−1 G𝑇 )
𝑡−𝑑1
× 𝜉 (𝑡) ,
𝑡−𝑑1
𝑇 (25)
− 𝑑12 ∫ 𝑥̇ (𝑠) (𝑅2 − 𝑀) 𝑥̇(𝑠) d𝑠
𝑡−𝑑(𝑡)
where
𝑡−𝑑(𝑡)
𝑇 [Ω1 ]5𝑛×5𝑛 [Ω2 ]5𝑛×𝑛 [Ω3 ]5𝑛×𝑛
− 𝑑12 ∫ 𝑥̇ (𝑠) (𝑅2 − 𝑀) 𝑥̇(𝑠) d𝑠.
𝑡−𝑑2 Υ1 = [ ∗ −12 (𝑅2 − 𝑀) −4𝑋22 ],
(20) [ ∗ ∗ −12 (𝑅2 − 𝑀)]
Now, we deal with the integral terms in inequality (20) by [Ω1 ]5𝑛×5𝑛 [Ω2 ]5𝑛×𝑛 ̂3]
[Ω (26)
5𝑛×𝑛
applying Lemma 3. Consider Υ2 = [ ∗ −12 (𝑅2 − 𝑀) −4𝑋22 ],
𝑡
𝑇 ] (𝑡) 𝑅 0
𝑇
] (𝑡) [ ∗ ∗ −12 (𝑅2 − 𝑀)]
−𝑑1 ∫ 𝑥̇ (𝑠) 𝑅1 𝑥̇(𝑠) d𝑠 ⩽ −[ 1 ] [ 1 ][ 1 ],
𝑡−𝑑1 ] 2 (𝑡) 0 3𝑅1 ]2 (𝑡)
Υ3 = (𝐴 𝑖 , 𝐴 𝑑𝑖 , 0, 0, 0, 0, 0) .
(21)
Using the Schur complement formula to (11) and (12), respec-
where ]1 (𝑡) = 𝑥(𝑡) − 𝑥(𝑡 − 𝑑1 ), ]2 (𝑡) = 𝑥(𝑡) + 𝑥(𝑡 − 𝑑1 ) − tively, we have
𝑡
(2/𝑑1 ) ∫𝑡−𝑑 𝑥(𝑠)d𝑠;
1
Υ1 + 𝑑12 Υ3𝑇 𝑅1 Υ3 + 𝑑12
2 𝑇
Υ3 𝑅2 Υ3 + N𝑀−1 N𝑇 < 0,
𝑡−𝑑1 𝑡−𝑑(𝑡) (27)
𝑇 𝑇
− 𝑑12 ∫ 𝑥̇ (𝑠) 𝑅2 𝑥̇(𝑠) d𝑠 − 𝑑12 ∫ 𝑥̇ (𝑠) 𝑅2 𝑥̇(𝑠) d𝑠 Υ2 + 𝑑12 Υ3𝑇 𝑅1 Υ3 + 𝑑12
2 𝑇
Υ3 𝑅2 Υ3 + G𝑀−1 G𝑇 < 0.
𝑡−𝑑(𝑡) 𝑡−𝑑2
By using convex combination approach [15], inequalities
𝑑12
𝑇[ Σ 0 (27) imply that L𝑉(𝑥𝑡 , 𝑖) < 0, which implies there exists a
𝜍1 (𝑡) [ 𝑑 (𝑡) − 𝑑1 ] 𝜍1 (𝑡)
⩽ −[ ] [ ][
] 𝜍2 (𝑡)] , sufficiently small 𝜖 > 0 such that L𝑉(𝑥𝑡 , 𝑖) ⩽ −𝜖‖𝑥(𝑡)‖2 for
𝜍2 (𝑡) 𝑑12
∗ Σ any initial condition (𝑥0 , 𝑟0 ). Therefore, for any 𝑡 > 0, by
[ 𝑑2 − 𝑑 (𝑡) ] Dynkin’s formula, we have
(22)
𝑡
where Σ = diag(𝑅2 − 𝑀, 3(𝑅2 − 𝑀)), E {𝑉 (𝑥𝑡 , 𝑖)} − {𝑉 (𝑥0 , 𝑟0 )} ⩽ −𝜖E {∫ ‖𝑥(𝑠)‖2 d𝑠 | (𝑥0 , 𝑟0 )} ,
0
𝜍1 (𝑡) (28)

𝑥 (𝑡 − 𝑑1 ) − 𝑥 (𝑡 − 𝑑 (𝑡)) which yields

=[ 2 𝑡−𝑑1 ], 𝑡
1
𝑥 (𝑡 − 𝑑1 ) + 𝑥 (𝑡 − 𝑑 (𝑡)) − ∫ 𝑥 (𝑠) d𝑠 E {∫ ‖𝑥 (𝑠)‖2 d𝑠 | (𝑥0 , 𝑟0 )} ⩽ {𝑉 (𝑥0 , 𝑟0 )} < ∞. (29)
[ 𝑑 (𝑡) − 𝑑1 𝑡−𝑑(𝑡) ] 0 𝜖
6 Abstract and Applied Analysis

The previous inequality means that lim𝑡 → +∞ E{‖𝑥(𝑡)‖2 | disappear. In order to obtain a less conservative result,
(𝑥0 , 𝑟0 )} = 0. Thus, system (1) is stochastically stable by we applied Lemma 3 to deal with the integral term. In
Definition 1. 𝑡−𝑑 𝑇
this case, integral term 𝑑12 ∫𝑡−d 1 𝑥̇ ̇
(𝑠)𝑅2 𝑥(𝑠)d𝑠 must be
2
reserved. Based on the above consideration, another matrix
Remark 5. In order to guarantee the Lyapunov-Krasovskii 𝑀 > 0 is introduced when we estimate the integral terms
functional 𝑉(𝑥𝑡 , 𝑟𝑡 ) > 0, most authors require the Lyapunov 𝑡−𝑑1 𝑡−𝑑1
−2𝜉𝑇 (𝑡)N ∫𝑡−𝑑(𝑡) ̇
𝑥(𝑠)d𝑠 and −2𝜉𝑇 (𝑡)G ∫𝑡−𝑑(𝑡) ̇
𝑥(𝑠)d𝑠.
matrix 𝑃𝑖 > 0 in 𝑉1 (see, e.g., [4–8]). The Lyapunov-
Krasovskii functional employed in this paper is very simple, When 𝑆 = {1}, the Markovian jump system (1) reduces to
but a less conservative result is developed by using the LMI the following linear system with interval time-varying delay:
(10) instead of inequality P𝑖 > 0, which can be seen
in Section 4. However, it should be pointed out that the
provided result cannot be used when the lower delay bound 𝑥̇(𝑡) = 𝐴𝑥 (𝑡) + 𝐴 𝑑 𝑥 (𝑡 − 𝑑 (𝑡)) ,
is considered to be zero because of the existence of 1/𝑑1 in (30)
LMI (10). 𝑥 (𝑡) = 𝜙 (𝑡) , 𝑡 ∈ [−𝑑2 , 0] .

Remark 6. It is well known that the convex combination

approach is effective in reducing conservatism in Based on the above method, we are now ready to give an
stability analysis. In some literature, the integral terms improved asymptotic stability criterion for system (30).
𝑡−𝑑1 𝑡−𝑑1
−2𝜉𝑇 (𝑡)N ∫𝑡−𝑑(𝑡) ̇
𝑥(𝑠)d𝑠 and −2𝜉𝑇 (𝑡)G ∫𝑡−𝑑(𝑡) ̇
𝑥(𝑠)d𝑠 are
Corollary 7. Given scalars 0 < 𝑑1 < 𝑑2 and 𝜇, the time-
estimated by ((𝑑(𝑡) − 𝑑1 )/𝑑12 )𝜉𝑇 (𝑡)N𝑅2−1 N𝑇 𝜉(𝑡) + varying delay system (30) is asymptotically stable, if there exist
𝑡−𝑑 𝑇
1
𝑑12 ∫𝑡−𝑑(𝑡) 𝑥̇ ̇
(𝑠)𝑅2 𝑥(𝑠)d𝑠 and ((𝑑2 − 𝑑(𝑡))/ matrix 𝑃𝜄𝜅 = 𝑃𝜅𝜄𝑇 ∈ R𝑛×𝑛 , (𝜄, 𝜅 = 1, 2, 3); symmetric positive
𝑑12 )𝜉𝑇 (𝑡)G𝑅2−1 G𝑇 𝜉(𝑡) + 𝑑12 ∫𝑡−𝑑
𝑡−𝑑(𝑡) 𝑇
𝑥̇ ̇
(𝑠)𝑅2 𝑥(𝑠)d𝑠, respec- definite matrices 𝑅1 ∈ R𝑛×𝑛 , 𝑅2 ∈ R𝑛×𝑛 , 𝑄1 ∈ R𝑛×𝑛 , 𝑄2 ∈
2
𝑡−𝑑
R𝑛×𝑛 , 𝑄3 ∈ R𝑛×𝑛 , and 𝑀 ∈ R𝑛×𝑛 ; and any matrices 𝑋 ∈ R𝑛×𝑛 ,
𝑇
tively, which make the term 𝑑12 ∫𝑡−𝑑 1 𝑥̇ ̇
(𝑠)𝑅2 𝑥(𝑠)d𝑠 in (17) 𝑁𝑙 ∈ R𝑛×𝑛 , and 𝐺𝑙 ∈ R𝑛×𝑛 (𝑙 = 1, 2) such that
2

𝑅2 − 𝑀 0 𝑋11 𝑋12
[ ∗ 3 (𝑅2 − 𝑀) 𝑋21 𝑋22 ]
Ψ1 = [
[
] > 0,
]
∗ ∗ 𝑅2 − 𝑀 0
[ ∗ ∗ 0 3 (𝑅2 − 𝑀) ]
2
2𝑑12 2𝑑12
𝑃
[ 11 + 2𝑑 𝑅
1 1 + 𝑅 𝑃12 − 2𝑅1 𝑃13 − 𝑅 ]
[ 𝑑1 + 𝑑2 2 𝑑1 + 𝑑2 2 ]
[ 1 ]
[
Ψ2 = [ ∗ 𝑃22 + (𝑄1 + 2𝑅1 ) 𝑃23 ] > 0,
𝑑1 ]
[ ]
[ 1 2 ]
∗ ∗ 𝑃33 + 𝑄2 + 𝑅2
[ 𝑑12 𝑑1 + 𝑑2 ]
[Φ1 ]5𝑛×5𝑛 [Φ2 ]5𝑛×𝑛 [Φ3 ]5𝑛×𝑛 [Φ4 ]5𝑛×𝑛 [Φ5 ]5𝑛×𝑛 [Φ6 ]5𝑛×𝑛 (31)
[ ∗ −12 (𝑅2 − 𝑀) −4𝑋22 0 0 0 ]
[ ]
[ ∗ ∗ −12 (𝑅2 − 𝑀) 0 0 0 ]
Ψ3 = [
[
] < 0,
[ ∗ ∗ ∗ −𝑅1 0 0 ] ]
[ ∗ ∗ ∗ ∗ −𝑅2 0 ]
[ ∗ ∗ ∗ ∗ ∗ −𝑀 ]

[Φ1 ]5𝑛×5𝑛 ̂2]

[Φ ̂3]
[Φ ̂6]
[Φ4 ]5𝑛×𝑛 [Φ5 ]5𝑛×𝑛 [Φ
5𝑛×𝑛 5𝑛×𝑛 5𝑛×𝑛
[ ∗ −12 (𝑅2 − 𝑀) −4𝑋22 0 0 0 ]
[ ]
[ ∗ ∗ −12 (𝑅2 − 𝑀) 0 0 0 ]
Ψ4 = [
[
] < 0,
]
[ ∗ ∗ ∗ −𝑅1 0 0 ]
[ ∗ ∗ ∗ ∗ −𝑅2 0 ]
[ ∗ ∗ ∗ ∗ ∗ −𝑀 ]

where Φ1 (1, 5) = 6𝑅1 + 𝑑1 (𝐴𝑇 𝑃12 + 𝑃22

𝑇
),
Φ1 (1, 1) = Sym (𝑃11 𝐴 + 𝑃12 ) + 𝑄1 − 4𝑅1 , Φ1 (2, 𝑠) = Ω1 (2, 𝑠) , (𝑠 = 1, 2, 3, 4)
Φ1 (1, 2) = 𝑃11 𝐴 𝑑 , Φ1 (2, 5) = 𝑑1 𝐴𝑇𝑑 𝑃12
Φ1 (1, 𝑠) = Ω1 (1, 𝑠) , (𝑠 = 3, 4) , Φ1 (3, 𝑠) = Ω1 (3, 𝑠) , (𝑠 = 3, 4, 5) ,
Abstract and Applied Analysis 7

Φ1 (4, 𝑠) = Ω1 (4, 𝑠) , (𝑠 = 4, 5) , Table 1: Admissible upper bounds 𝑑2 with varying 𝑑1 .

Φ1 (5, 5) = Ω1 (5, 5) , 𝑑1 0.2 0.25 0.3 0.35 0.4

[20] 0.520 0.537 0.557 0.579 0.603
Φ2 (1, 1) = 𝑑12 (𝐴𝑇 𝑃13 + 𝑃23 ) , Theorem 4 0.605 0.615 0.629 0.646 0.667

Φ2 (2, 1) = 𝑑12 𝐴𝑇𝑑 𝑃13 + 2𝑋21

𝑇 𝑇
+ 2𝑋22 + 6𝑅6 − 6𝑀,

Φ2 (𝑠, 1) = Ω2 (𝑠, 1) , (𝑠 = 3, 4, 5) ,
Example 2. Consider the Markovian jump systems (1) with
Φ3 (𝑠, 1) = Ω3 (𝑠, 1) , (𝑠 = 1, 2, 3, 4, 5) ,

Φ4 (1, 1) = 𝑑1 𝐴𝑇 𝑅1 , Φ4 (2, 1) = 𝑑1 𝐴𝑇𝑑 𝑅1 , 𝐴1 = [

−3.4888 0.8057
],
−0.6451 −3.2684
Φ4 (𝑠, 1) = Ω4 (𝑠, 1) , (𝑠 = 3, 4, 5) ,
−0.8620 −1.2919
Φ5 (1, 1) = 𝑑12 𝐴𝑇 𝑅2 , Φ5 (2, 1) = 𝑑12 𝐴𝑇𝑑 𝑅2 , 𝐴 𝑑1 = [ ],
−0.6841 −2.0729
(34)
Φ5 (𝑠, 1) = Ω5 (𝑠, 1) , (𝑠 = 3, 4, 5) , −2.4898 0.2895
𝐴2 = [ ],
1.3396 −0.0211
Φ ̂ 2 (𝑠, 1) ,
̂ 2 (𝑠, 1) = Ω (𝑠 = 1, 2, 3, 4, 5) ,
−2.8306 0.4978
̂ 3 (1, 1) = 𝑑12 (𝐴𝑇 𝑃13 + 𝑃23 ) ,
Φ 𝐴 𝑑2 = [ ].
−0.8436 −1.0115
̂ 3 (2, 1) = 𝑑12 𝐴𝑇 𝑃13 + 6𝑅2 − 6𝑀 − 2𝑋12 + 2𝑋22 ,
Φ 𝑑

̂ 3 (𝑠, 1) = Ω
̂ 3 (𝑠, 1) , This example has been taken from [4]. To compare the
Φ (𝑠 = 3, 4, 5) , stochastic stability condition in Theorem 4 with that in [25,
̂ 6 (𝑠, 1) ,
̂ 6 (𝑠, 1) = Ω 26], we choose 𝜋22 = 0.8, 𝜇 = 0.9. Using Theorem 4 of our
Φ (𝑠 = 1, 2, 3, 4, 5) . paper, the admissible upper bound 𝑑2 for different 𝑑1 and
(32) 𝜋11 can be found in Table 2. It can be seen from Table 2 that
Theorem 4 in our paper is less conservative.
Remark 8. Theorem 4 and Corollary 7 can be applied to both
slow and fast time-varying delays. But when 𝜇 is unknown, Example 3. Consider the systems (30) with the following
the above results cannot be used directly to check the stability. parameters:
From the construction of Lyapunov-Krasovskii functional, it
can be seen by setting 𝑄3 = 0 in Theorem 4 and Corollary 7
that the corresponding conclusions are valid for the case −2.0 0.0 −1.0 0.0
when 𝜇 is unknown. 𝐴=[ ], 𝐴𝑑 = [ ]. (35)
0.0 −0.9 −1.0 −0.1

4. Numerical Examples
This system is a well-known delay-dependent stable system
In this section, three numerical examples will be presented to where the maximum allowable delay 𝑑max = 6.1725 [24]. For
show the validity of the main results derived above. known and unknown 𝜇, the admissible upper bounds 𝑑2 for
different 𝑑1 , which guarantee the asymptotical stability of the
Example 1. Consider the Markovian jump systems (1) with
system (30), are listed in Tables 3 and 4, respectively. It can
be seen from Tables 3 and 4 that the stability results obtained
−2.3 0.8 0.8 1.2
𝐴1 = [ ], 𝐴 𝑑1 = [ ], in this paper are less conservative than those in [14, 16, 17, 22,
1 −2.9 0.7 −3.5
(33) 27, 28].
−1.9 0.2 1.3 −2.6
𝐴2 = [ ], 𝐴 𝑑2 =[ ].
0.6 −0.8 0.5 −1.4
5. Conclusion
Consider 𝜋11 = −𝜋12 = −0.5, 𝜋22 = −𝜋21 = −3. According to
[20], for 𝑑1 = 0.2, one obtains 𝑑2 = 0.520, while by Theorem 4 In this paper, the problem of stochastic stability for a class
in this paper we can obtain 𝑑2 = 0.605 such that the system is of Markovian jump systems has been investigated. By using
stochastically stable. More comparisons are shown in Table 1, the convex combination technique and the improved integral
which indicate that Theorem 4 is much less conservative than inequality, some less conservative delay-dependent stability
that in [20]. criteria are established in terms of linear matrix inequalities.
8 Abstract and Applied Analysis

Table 2: Admissible upper bounds 𝑑2 with varying 𝜋11 and 𝑑1 .

𝑑1 Methods 𝜋11 = −0.1 𝜋11 = −0.3 𝜋11 = −0.5 𝜋11 = −0.7 𝜋11 = −0.9
[25] 0.4336 0.4332 0.4328 0.4324 0.4321
0.1 [26] 0.5752 0.5507 0.5375 0.5289 0.5227
Theorem 4 0.5847 0.5895 0.5949 0.6005 0.6061
[25] 0.4459 0.4450 0.4442 0.4434 0.4428
0.2 [26] 0.5930 0.5672 0.5529 0.5433 0.5363
Theorem 4 0.5932 0.5988 0.6048 0.6108 0.6168

Table 3: Admissible upper bounds 𝑑2 with varying 𝜇 and 𝑑1 .

𝜇 Methods 𝑑1 = 2 𝑑1 = 3 𝑑1 = 4 𝑑1 = 5 𝑑1 = 6
[14] 2.6972 3.2591 4.0774 — —
[27] 3.0129 3.3408 4.1690 5.0275 —
0.3
[16] 3.1046 3.4181 4.2102 5.0440 —
Corollary 7 3.2321 3.4977 4.2939 5.1372 6.0071
[14] 2.5048 3.2591 4.0774 — —
[27] 2.5663 3.3408 4.1690 5.0275 —
0.5
[16] 2.6940 3.4181 4.2102 5.0440 —
Corollary 7 2.800 3.4977 4.2939 5.1372 6.0071
[14] 2.5048 3.2591 4.0774 — —
[27] 2.5663 3.3408 4.1690 5.0275 —
0.9
[16] 2.6940 3.4181 4.2102 5.0440 —
Corollary 7 2.800 3.4977 4.2939 5.1372 6.0071

Table 4: Admissible upper bounds 𝑑2 with unknown 𝜇 and 𝑑1 . [2] E. Boukas, Stochastic Switching Systems: Analysis and Design,
Birkhäauser, Boston, Mass, USA, 2006.
𝑑1 1.0 2.0 3.0 4.0 5.0 6.0 [3] X. Mao, “Exponential stability of stochastic delay interval sys-
[14] 1.87 2.50 3.25 4.07 — — tems with Markovian switching,” IEEE Transactions on Auto-
[17] 2.04 2.60 3.30 4.08 — — matic Control, vol. 47, no. 10, pp. 1604–1612, 2002.
[22] 2.06 2.61 3.31 4.09 — — [4] S. Xu, J. Lam, and X. Mao, “Delay-dependent 𝐻∞ control and
2.12 2.72 3.45 4.25 5.09 — filtering for uncertain Markovian jump systems with time-
[28]
varying delays,” IEEE Transactions on Circuits and Systems I:
Corollary 7
Regular Papers, vol. 54, no. 9, pp. 2070–2077, 2007.
with 2.338 2.800 3.497 4.293 5.137 6.007
[5] L. X. Zhang and E. Boukas, “Stability and stabilization of
Remark 8
Markovian jump linear systems with partly unknown transition
probabilities,” Automatica, vol. 45, no. 2, pp. 463–468, 2009.
[6] H. Gao, Z. Fei, J. Lam, and B. Du, “Further results on
The numerical examples demonstrate the effectiveness and exponential estimates of Markovian jump systems with mode-
superiority of the presented results. dependent time-varying delays,” IEEE Transactions on Auto-
matic Control, vol. 56, no. 1, pp. 223–229, 2011.
Conflict of Interests [7] Y. R. Liu, Z. D. Wang, and X. H. Liu, “Stability analysis for a
class of neutral-type neural networks with Markovian jumping
The authors declare that there is no conflict of interests parameters and mode-dependent mixed delays,” Neurocomput-
regarding the publication of this paper. ing, vol. 94, pp. 46–53, 2012.
[8] Z. G. Wu, J. H. Park, H. Y. Su, and J. Chu, “Delay-dependent
passivity for singular Markov jump systems with time-delays,”
Acknowledgment Communications in Nonlinear Science and Numerical Simula-
tion, vol. 18, no. 3, pp. 669–681, 2013.
This work was supported by the Doctoral Research Foun- [9] E. Fridman, “New Lyapunov-Krasovskii functionals for stability
dation of Southwest University of Science and Technology of linear retarded and neutral type systems,” Systems & Control
(Grant no. 13zx7141). Letters, vol. 43, no. 4, pp. 309–319, 2001.
[10] Q. L. Han, “New results for delay-dependent stability of linear
References systems with time-varying delay,” International Journal of Sys-
tems Science, vol. 33, no. 3, pp. 213–228, 2002.
[1] X. Mao and C. Yuan, Stochastic Differential Equations with [11] M. Wu, Y. He, J. She, and G. Liu, “Delay-dependent criteria for
Markovian Switching, Imperial College Press, London, UK, robust stability of time-varying delay systems,” Automatica, vol.
2006. 40, no. 8, pp. 1435–1439, 2004.
Abstract and Applied Analysis 9

[12] K. Gu, V. Kharitonov, and J. Chen, Stability of Time-Delay

Systems, Springer, Berlin, Germany, 2003.
[13] E. Fridman and U. Shaked, “Delay-dependent stability and
𝐻∞ control: constant and time-varying delays,” International
Journal of Control, vol. 76, no. 1, pp. 48–60, 2003.
[14] H. Y. Shao, “New delay-dependent stability criteria for systems
with interval delay,” Automatica, vol. 45, no. 3, pp. 744–749,
2009.
[15] D. Yue, E. Tian, Y. Zhang, and C. Peng, “Delay-distribution-
dependent stability and stabilization of T-S fuzzy systems with
probabilistic interval delay,” IEEE Transactions on Systems, Man,
and Cybernetics, Part B: Cybernetics, vol. 39, no. 2, pp. 503–516,
2009.
[16] F. O. Souza, “Further improvement in stability criteria for linear
systems with interval time-varying delay,” IET Control Theory &
Applications, vol. 7, no. 3, pp. 440–446, 2013.
[17] M. Tang, Y. Wang, and C. Wen, “Improved delay-range-
dependent stability criteria for linear systems with interval
time-varying delays,” IET Control Theory & Applications, vol. 6,
no. 6, pp. 868–873, 2012.
[18] Q. Han, “A discrete delay decomposition approach to stability
of linear retarded and neutral systems,” Automatica, vol. 45, no.
2, pp. 517–524, 2009.
[19] M. Fang and J. H. Park, “A multiple integral approach to
stability of neutral time-delay systems,” Applied Mathematics
and Computation, vol. 224, pp. 714–718, 2013.
[20] Z. Li, Z. Fei, and H. Gao, “Stability and stabilisation of Marko-
vian jump systems with time-varying delay: an input-output
approach,” IET Control Theory & Applications, vol. 6, no. 17, pp.
2601–2610, 2012.
[21] X.-L. Zhu and G.-H. Yang, “Jensen integral inequality approach
to stability analysis of continuous-time systems with time-
varying delay,” IET Control Theory & Applications, vol. 2, no. 6,
pp. 524–534, 2008.
[22] P. Park, J. W. Ko, and C. Jeong, “Reciprocally convex approach to
stability of systems with time-varying delays,” Automatica, vol.
47, no. 1, pp. 235–238, 2011.
[23] J. W. Xia, “Delay-dependent robust exponential stability and
𝐻∞ analysis for a class of uncertain Markovian jumping system
with multiple delays,” Abstract and Applied Analysis, vol. 2014,
Article ID 738318, 10 pages, 2014.
[24] A. Seuret and F. Gouaisbaut, “Wirtinger-based integral inequal-
ity: application to time-delay systems,” Automatica, vol. 49, no.
9, pp. 2860–2866, 2013.
[25] Z. Wu, H. Su, and J. Chu, “Delay-dependent 𝐻∞ filtering for
singular Markovian jump time-delay systems,” Signal Process-
ing, vol. 90, no. 6, pp. 1815–1824, 2010.
[26] Y. Ding, H. Zhu, S. Zhong, and Y. Zeng, “Exponential mean-
square stability of time-delay singular systems with Markovian
switching and nonlinear perturbations,” Applied Mathematics
and Computation, vol. 219, no. 4, pp. 2350–2359, 2012.
[27] J. Sun, G. P. Liu, J. Chen, and D. Rees, “Improved delay-range-
dependent stability criteria for linear systems with time-varying
delays,” Automatica, vol. 46, no. 2, pp. 466–470, 2010.
[28] E. Fridman, U. Shaked, and K. Liu, “New conditions for delay-
derivative-dependent stability,” Automatica, vol. 45, no. 11, pp.
2723–2727, 2009.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 324912, 10 pages
http://dx.doi.org/10.1155/2014/324912

Research Article
Dynamics of an Almost Periodic Food Chain System with
Impulsive Effects

Yaqin Li,1 Wenquan Wu,2 and Tianwei Zhang3

1
Department of Mathematics, Kunming University, Kunming 650031, China
2
Department of Mathematics, Aba Teachers College, Wenchuan, Sichuan 623002, China
3
City College, Kunming University of Science and Technology, Kunming 650051, China

Correspondence should be addressed to Tianwei Zhang; zhang@kmust.edu.cn

Received 5 June 2014; Accepted 6 July 2014; Published 21 July 2014

Academic Editor: Yonghui Xia

Copyright © 2014 Yaqin Li et al. This is an open access article distributed under the Creative Commons Attribution License, which
permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

In order to obtain a more accurate description of the ecological system perturbed by human exploitation activities such as planting
and harvesting, we need to consider the impulsive differential equations. Therefore, by applying the comparison theorem and the
Lyapunov method of the impulsive differential equations, this paper gives some new sufficient conditions for the permanence and
existence of a unique uniformly asymptotically stable positive almost periodic solution in a food chain system with almost periodic
impulsive perturbations. The method used in this paper provides a possible method to study the permanence and existence of a
unique uniformly asymptotically stable positive almost periodic solution of the models with impulsive perturbations in biological
populations. Finally, an example and numerical simulations are given to illustrate that our results are feasible.

1. Introduction species food chain predator-prey system with Holling type IV

functional response:
Let R and Z denote the sets of real numbers and integers,
respectively. Related to a continuous function 𝑓, we use the 𝑏1 (𝑡) 𝑥2 (𝑡)
following notations: 𝑥1̇ (𝑡) = 𝑥1 (𝑡) [𝑟1 (𝑡) − 𝑎1 (𝑡) 𝑥1 (𝑡) − ],
𝑚1 + 𝑥12 (𝑡)

𝑏2 (𝑡) 𝑥1 (𝑡)
𝑓𝑙 = inf 𝑓 (𝑠) , 𝑓𝑢 = sup𝑓 (𝑠) . (1) 𝑥2̇ (𝑡) = 𝑥2 (𝑡) [−𝑟2 (𝑡) +
𝑠∈R 𝑠∈R 𝑚1 + 𝑥12 (𝑡)
(2)
𝑏 (𝑡) 𝑥3 (𝑡)
−𝑎2 (𝑡) 𝑥2 (𝑡) − 3 ],
As was pointed out by Berryman [1], the dynamic 𝑚2 + 𝑥22 (𝑡)
relationship between predators and their prey has long been
and will continue to be one of the dominant themes in 𝑏4 (𝑡) 𝑥2 (𝑡)
𝑥3̇ (𝑡) = 𝑥3 (𝑡) [−𝑟3 (𝑡) + − 𝑎3 (𝑡) 𝑥3 (𝑡)] ,
both ecology and mathematical ecology due to its universal 𝑚2 + 𝑥22 (𝑡)
existence and importance. Food chain predator-prey system,
as one of the most important predator-prey systems, has been where 𝑥𝑖 (𝑡), 𝑖 = 1, 2, 3, denotes the density of species 𝑋𝑖 at
extensively studied by many scholars; many excellent results time 𝑡, 𝑋2 is the predator of the first species 𝑋1 , and 𝑋3 is
were concerned with the persistent property and positive the predator of the second species 𝑋2 . By applying the com-
periodic solution of the system; see [2–8] and the references parison theorem of the differential equation and constructing
cited therein. Recently, Shen considered the following three the suitable Lyapunov function, sufficient conditions which
2 Abstract and Applied Analysis

guarantee the permanence and the global attractivity of the 𝑏3 (𝑡) 𝑥3 (𝑡)
− ],
system are obtained. 𝑚2 + 𝑥2 (𝑡)
Considering the exploited predator-prey system (harvest-
𝑏4 (𝑡) 𝑥2 (𝑡)
ing or stocking) is very valuable, for it involves the human 𝑥3̇ (𝑡) = 𝑥3 (𝑡) [−𝑟3 (𝑡) + − 𝑎3 (𝑡) 𝑥3 (𝑡)] .
activities. It can be referred to [9], in which the human 𝑚2 + 𝑥2 (𝑡)
activities always happen in a short time or instantaneously. (4)
The continuous action of human is then removed from the
By applying the comparison theorem and the Lyapunov
model and replaced with an impulsive perturbation. These
method of ordinary differential equations, some sufficient
models are subject to short-term perturbations which are
conditions which guarantee the permanence and existence
often assumed to be in the form of impulsive in the modelling
of a unique uniformly asymptotically stable positive almost
process. Consequently, impulsive differential equations pro-
periodic solution of system (4) are obtained.
vide a natural description of such systems [10–13]. Then, in
Stimulated by the above reason, this paper is concerned
[14], Zhang and Tan studied the following Holling II func-
with the following almost periodic food chain system with
tional responses food chain system with periodic constant
almost periodic impulsive perturbations and general func-
impulsive perturbation of predator:
tional responses:

𝑏1 𝑥2 (𝑡) 𝑏1 (𝑡) 𝑥2 (𝑡)

𝑥1̇ (𝑡) = 𝑥1 (𝑡) [1 − 𝑥1 (𝑡) − ], 𝑥1̇ (𝑡) = 𝑥1 (𝑡) [𝑟1 (𝑡) − 𝑎1 (𝑡) 𝑥1 (𝑡) − 𝛼 ],
1 + 𝑥1 (𝑡) 𝑚1 + 𝑥1 1 (𝑡)
𝑏2 𝑥1 (𝑡) 𝑏 𝑥 (𝑡) 𝑏2 (𝑡) 𝑥1 (𝑡)
𝑥2̇ (𝑡) = 𝑥2 (𝑡) [−𝑟2 + − 3 3 ], 𝑥2̇ (𝑡) = 𝑥2 (𝑡) [−𝑟2 (𝑡) + − 𝑎2 (𝑡) 𝑥2 (𝑡)
1 + 𝑥1 (𝑡) 1 + 𝑥2 (𝑡) 𝛼
𝑚1 + 𝑥1 1 (𝑡)
𝑏4 𝑥2 (𝑡) 𝑏3 (𝑡) 𝑥3 (𝑡)
𝑥3̇ (𝑡) = 𝑥3 (𝑡) [−𝑟3 + ], 𝑡 ≠𝑛𝑇, (3)
− ],
1 + 𝑥2 (𝑡) 𝛼
𝑚2 + 𝑥2 2 (𝑡)
Δ𝑥1 (𝑛𝑇) = 0,
𝑏4 (𝑡) 𝑥2 (𝑡)
𝑥3̇ (𝑡) = 𝑥3 (𝑡) [−𝑟3 (𝑡) + 𝛼 − 𝑎3 (𝑡) 𝑥3 (𝑡)] ,
Δ𝑥2 (𝑛𝑇) = ℎ, 𝑚2 + 𝑥2 2 (𝑡)
Δ𝑥3 (𝑛𝑇) = 0, 𝑛 ∈ {0, 1, . . .} = Z+ , 𝑡 ≠𝜏𝑘 ,

Δ𝑥1 (𝜏𝑘 ) = ℎ1𝑘 𝑥1 (𝜏𝑘 ) ,

where ℎ > 0 is the release amount of top predator at 𝑡 = 𝑛𝑇
and 𝑇 is the period of the impulsive effect. By using the Δ𝑥2 (𝜏𝑘 ) = ℎ2𝑘 𝑥2 (𝜏𝑘 ) ,
Floquet theory of impulsive differential equation and small
amplitude perturbation skills, we consider the local stability Δ𝑥3 (𝜏𝑘 ) = ℎ3𝑘 𝑥3 (𝜏𝑘 ) , 𝑘 ∈ Z+ ,
of prey and top predator eradication periodic solution. (5)
In real world phenomenon, the environment varies due to
the factors such as seasonal effects of weather, food supplies, where 𝑟𝑖 (𝑡), 𝑎𝑖 (𝑡), and 𝑏𝑗 (𝑡), 𝑖 = 1, 2, 3, 𝑗 = 1, 2, 3, 4, are
mating habits, and harvesting. So, it is usual to assume the all continuous almost periodic functions which are bounded
periodicity of parameters in system (2). However, if the above and below by positive constants; 𝑚1 , 𝑚2 , 𝛼1 , 𝛼2 are
various constituent components of the temporally nonuni- positive constants; ℎ1𝑘 , ℎ2𝑘 , ℎ3𝑘 > −1 are constants; 0 = 𝜏0 <
form environment are with incommensurable (nonintegral 𝜏1 < 𝜏2 < ⋅ ⋅ ⋅ < 𝜏𝑘 < 𝜏𝑘+1 < ⋅ ⋅ ⋅ are impulse points
𝑗 𝑗
multiples) periods, then one has to consider the environ- with lim𝑘 → +∞ 𝜏𝑘 = +∞; and the set of sequences {𝜏𝑘 }, 𝜏𝑘 =
ment to be almost periodic since the assumption of almost 𝜏𝑘+𝑗 − 𝜏𝑘 , 𝑘 ∈ Z+ , 𝑗 ∈ Z, is uniformly almost periodic (see
periodicity is more realistic, more important, and more Definition 1 in Section 2).
general when we consider the effects of the environmental Obviously, system (2)–(4) is special case of system (5).
factors. In recent years, there have been many mathematical The main purpose of this paper is to establish some
studies for the existence, uniqueness, and stability of positive new sufficient conditions which guarantee the permanence
almost periodic solution of biological models governed by and existence of a unique uniformly asymptotically stable
differential equations in the literature (see [11, 15–25] and positive almost periodic solution of system (5) by using
the references cited therein). Therefore, Bai and wang in [15] the comparison theorem and the Lyapunov method of the
studied the following nonautonomous food chains system impulsive differential equations [10, 11] (see Theorems 11 and
with Holling’s type II functional response: 14 in Sections 3 and 4).
The organization of this paper is as follows. In Section 2,
𝑏1 (𝑡) 𝑥2 (𝑡) we give some basic definitions and necessary lemmas which
𝑥1̇ (𝑡) = 𝑥1 (𝑡) [𝑟1 (𝑡) − 𝑎1 (𝑡) 𝑥1 (𝑡) − ], will be used in later sections. In Section 3, by using the
𝑚1 + 𝑥1 (𝑡)
comparison theorem of the impulsive differential equations
𝑏2 (𝑡) 𝑥1 (𝑡) [10], we give the permanence of system (5). In Section 4,
𝑥2̇ (𝑡) = 𝑥2 (𝑡) [−𝑟2 (𝑡) + − 𝑎2 (𝑡) 𝑥2 (𝑡)
𝑚1 + 𝑥1 (𝑡) we study the existence of a unique uniformly asymptotically
Abstract and Applied Analysis 3

stable positive almost periodic solution of system (5) by Introduce the following conditions.
applying the Lyapunov method of the impulsive differential
equations [11]. (𝐶1 ) Function 𝑓(𝑡, 𝑥) is almost periodic in 𝑡 uniformly
with respect to 𝑥 ∈ 𝐷.
(𝐶2 ) Sequence {𝐼𝑘 (𝑥)}, 𝑘 ∈ Z, is almost periodic uniformly
2. Preliminaries
with respect to 𝑥 ∈ 𝐷.
Now, let us state the following definitions and lemmas, which
will be useful in proving our main result. Lemma 4 (see [11, P109 ]). Suppose that there exists a Lyapunov
By I, I = {{𝜏𝑘 } ∈ R : 𝜏𝑘 < 𝜏𝑘+1 , 𝑘 ∈ Z, lim𝑘 → ±∞ 𝜏𝑘 = ±∞}, functional 𝑉(𝑡, 𝑥, 𝑦) defined on R+ × 𝐷 × 𝐷 satisfying the
we denote the set of all sequences that are unbounded and following conditions.
strictly increasing. Introduce the following notations. (1) 𝑢(‖𝑥 − 𝑦‖) ≤ 𝑉(𝑡, 𝑥, 𝑦) ≤ V(‖𝑥 − 𝑦‖), where 𝑢, V ∈ P
For 𝐽 ⊂ R3 , 𝑃𝐶(𝐽, R3 ) is the space of all piecewise con- with P = {𝑢 : R+ → R+ | 𝑢 is continuous increasing
tinuous functions from 𝐽 to R3 with points of discontinuity function and 𝑢(𝑠) → 0 as 𝑠 → 0}.
of the first kind 𝜏𝑘 , at which it is left continuous. By the basic
̂ 𝑦)|
(2) |𝑉(𝑡, 𝑥, 𝑦) − 𝑉(𝑡, 𝑥, ̂ ≤ 𝐾(‖𝑥 − 𝑥‖
̂ + ‖𝑦 − 𝑦‖),
̂ where
theories of impulsive differential equations in [10, 11], system
𝐾 > 0 is a constant.
(5) has a unique solution 𝑋(𝑡) = 𝑋(𝑡, 𝑋0 ) ∈ 𝑃𝐶([0, +∞), R3 ).
Since the solution of system (5) is a piecewise continuous (3) For 𝑡 = 𝜏𝑘 , 𝑉(𝑡+ , 𝑥 + 𝐼𝑘 (𝑥), 𝑦 + 𝐼𝑘 (𝑦)) ≤ 𝑉(𝑡, 𝑥, 𝑦); for
function with points of discontinuity of the first kind 𝜏𝑘 , 𝑘 ∈ ̇ (𝑡, 𝑥, 𝑦) ≤ −𝛾𝑉(𝑡, 𝑥, 𝑦), ∀𝑘 ∈ Z, where
𝑡 ≠𝜏𝑘 , 𝑉(2.2)
Z, we adopt the following definitions for almost periodicity. 𝛾 > 0 is a constant.

𝑗 𝑗 Moreover, one assumes that system (7) has a solution that

Definition 1 (see [11]). The set of sequences {𝜏𝑘 }, 𝜏𝑘 = 𝜏𝑘+𝑗 − remains in a compact set 𝑆 ⊂ 𝐷. Then, system (7) has a unique
𝜏𝑘 , 𝑘 ∈ Z, 𝑗 ∈ Z, {𝜏𝑘 } ∈ I, is said to be uniformly almost almost periodic solution which is uniformly asymptotically
periodic if for arbitrary 𝜖 > 0 there exists a relatively dense stable.
set of 𝜖-almost periods common for any sequences.

Definition 2 (see [11]). The function 𝜑 ∈ 𝑃𝐶(R, R) is said to 3. Permanence

be almost periodic, if the following hold.
In this section, we establish a permanence result for system
𝑗 𝑗
(1) The set of sequences {𝜏𝑘 }, 𝜏𝑘 = 𝜏𝑘+𝑗 − 𝜏𝑘 , 𝑘 ∈ Z, 𝑗 ∈ (5).
Z, {𝜏𝑘 } ∈ I, is uniformly almost periodic.
Lemma 5 (see [10]). Assume that 𝑥 ∈ 𝑃𝐶(R) with points of
(2) For any 𝜖 > 0, there exists a real number 𝛿 > 0 such discontinuity at 𝑡 = 𝜏𝑘 and is left continuous at 𝑡 = 𝜏𝑘 for
that if the points 𝑡󸀠 and 𝑡󸀠󸀠 belong to one and the same 𝑘 ∈ Z+ and
interval of continuity of 𝜑(𝑡) and satisfy the inequality
|𝑡󸀠 − 𝑡󸀠󸀠 | < 𝛿, then |𝜑(𝑡󸀠 ) − 𝜑(𝑡󸀠󸀠 )| < 𝜖. 𝑥̇(𝑡) ≤ 𝑓 (𝑡, 𝑥 (𝑡)) , 𝑡 ≠𝜏𝑘 ,
(3) For any 𝜖 > 0, there exists a relatively dense set 𝑇 such (8)
𝑥 (𝜏𝑘+ ) ≤ 𝐼𝑘 (𝑥 (𝜏𝑘 )) , 𝑘 ∈ Z+ ,
that if 𝜂 ∈ 𝑇, then |𝜑(𝑡 + 𝜂) − 𝜑(𝑡)| < 𝜖 for all 𝑡 ∈ R
satisfying condition |𝑡 − 𝜏𝑘 | > 𝜖, 𝑘 ∈ Z. The elements where 𝑓 ∈ 𝐶(R × R, R), 𝐼𝑘 ∈ 𝐶(R, R), and 𝐼𝑘 (𝑥) is
of 𝑇 are called 𝜖-almost periods. nondecreasing in 𝑥 for 𝑘 ∈ Z+ . Let 𝑢∗ (𝑡) be the maximal
solution of the scalar impulsive differential equation as
Lemma 3 (see [11]). Let {𝜏𝑘 } ∈ I. Then, there exists a positive
integer 𝐴 such that, on each interval of length 1, one has no 𝑢̇(𝑡) = 𝑓 (𝑡, 𝑢 (𝑡)) , 𝑡 ≠𝜏𝑘 ,
more than 𝐴 elements of the sequence {𝜏𝑘 }; that is,
𝑢 (𝜏𝑘+ ) = 𝐼𝑘 (𝑢 (𝜏𝑘 )) ≥ 0, 𝑘 ∈ Z+ , (9)
𝑖 (𝑠, 𝑡) ≤ 𝐴 (𝑡 − 𝑠) + 𝐴, (6)
𝑢 (𝑡0+ ) = 𝑢0
where 𝑖(𝑠, 𝑡) is the number of the points 𝜏𝑘 in the interval (𝑠, 𝑡).
existing on [𝑡0 , ∞). Then, 𝑥(𝑡0+ ) ≤ 𝑢0 implies 𝑥(𝑡) ≤ 𝑢∗ (𝑡) for
Theoretically, one can investigate the existence, unique- 𝑡 ≥ 𝑡0 .
ness, and stability of almost periodic solution for functional
differential equations by using Lyapunov functional as follows Remark 6. If inequalities (8) in Lemma 5 are reversed and
[11, P109 ]. 𝑢∗ (𝑡) is the minimal solution of system (9) existing on [𝑡0 , ∞),
Consider the system of impulsive differential equations as then 𝑥(𝑡0+ ) ≥ 𝑢0 implies 𝑥(𝑡) ≥ 𝑢∗ (𝑡) for 𝑡 ≥ 𝑡0 .
follows:
𝑥̇(𝑡) = 𝑓 (𝑡, 𝑥 (𝑡)) , 𝑡 ≠𝜏𝑘 , Lemma 7. Assume that 𝑎, 𝑏 > 0; then, the following impulsive
(7) logistic equation
Δ𝑥 (𝜏𝑘 ) = 𝐼𝑘 𝑥 (𝜏𝑘 ) ,
𝑥̇(𝑡) = 𝑥 (𝑡) [𝑎 − 𝑏𝑥 (𝑡)] , 𝑡 ≠𝜏𝑘 ,
𝑛 𝑛 (10)
where 𝑡 ∈ R, {𝜏𝑘 } ∈ I, 𝑓 : R × 𝐷 → R , 𝐼𝑘 : 𝐷 → R , 𝑘 ∈ Z,
and 𝐷 is an open set in R𝑛 . Δ𝑥 (𝜏𝑘+ ) = ℎ𝑘 𝑥 (𝜏𝑘 ) , 𝑘 ∈ Z+
4 Abstract and Applied Analysis

has a unique globally asymptotically stable positive almost Therefore, 𝑉 is nonincreasing. Integrating (18) from 0 to 𝑡
periodic solution 𝑥∗ which can be expressed as follows: leads to
𝑡 −1 𝑡
𝛼 𝑎 󵄨 󵄨
≤ 𝑥∗ (𝑡) = [𝑏 ∫ 𝑊 (𝑡, 𝑠) 𝑑𝑠] ≤ , 𝑉 (𝑡) + 𝑏 ∫ 󵄨󵄨󵄨𝑥 (𝑠) − 𝑥∗ (𝑠)󵄨󵄨󵄨 𝑑𝑠 ≤ 𝑉 (0) < +∞, ∀𝑡 ≥ 0;
𝑒𝜉𝐴𝑏 −∞ 𝜂𝑏 (1 − 𝑒−𝑎𝜃 ) 0
(11) (20)

where 𝐴 is defined as that in Lemma 3, 𝜉 := ln sup𝑘∈Z (1/(1 + that is,

ℎ𝑘 )), 𝛼 := 𝑎−𝜉𝐴, 𝜃 := inf 𝑘∈Z 𝜏𝑘1 , 𝜂 := inf 𝑘∈Z ∏2𝑗=1 (1/(1+ℎ𝑗+𝑘 )), +∞
󵄨󵄨 󵄨
󵄨󵄨𝑥 (𝑠) − 𝑥 (𝑠)󵄨󵄨󵄨 𝑑𝑠 < +∞,
∗
and ∫ (21)
0
𝑊 (𝑡, 𝑠)
which implies that
{ 𝑒−𝑎(𝑡−𝑠) , 𝜏𝑘−1 < 𝑠 < 𝑡 < 𝜏𝑘 ; 󵄨 󵄨
{
{ lim 󵄨󵄨𝑥 (𝑠) − 𝑥∗ (𝑠)󵄨󵄨󵄨 = 0.
{
{ 𝑘+1 1 −𝑎(𝑡−𝑠) (12) 𝑠 → +∞ 󵄨
(22)
= {∏ 𝑒 , 𝜏𝑚−1 < 𝑠 ≤ 𝜏𝑚
{
{ 1 + ℎ𝑗
{
{
𝑗=𝑚
Thus, the almost periodic solution of system (10) is globally
{ < 𝜏𝑘 < 𝑡 ≤ 𝜏𝑘+1 . asymptotically stable. This completes the proof.
Proof. Let 𝑢 = 1/𝑥; then, system (10) changes to
Let
𝑑𝑢 (𝑡)
= −𝑎𝑢 (𝑡) + 𝑏, 𝑡 ≠𝜏𝑘 , 2
1 1
𝑑𝑡 𝜂𝑖 := inf ∏ , 𝜉𝑖 := ln sup ,
𝑗=1 1 + ℎ𝑖(𝑗+𝑘) 𝑘∈Z 1 + ℎ𝑖𝑘
(13) 𝑘∈Z
(23)
ℎ𝑘
Δ𝑢 (𝜏𝑘+ ) = − 𝑢 (𝜏𝑘 ) , 𝑘 ∈ Z+ .
1 + ℎ𝑘 𝑖 = 1, 2, 3.
We can easily obtain that system (13) has a unique almost
periodic solution which can be expressed as follows: Proposition 8. Every solution (𝑥1 , 𝑥2 , 𝑥2 )𝑇 of system (5)
satisfies
𝑡
𝑢∗ (𝑡) = 𝑏 ∫ 𝑊 (𝑡, 𝑠) 𝑑𝑠. (14) lim sup 𝑥𝑖 (𝑡) ≤ 𝑀𝑖 , 𝑖 = 1, 2, 3,
−∞
𝑡→∞
(24)
Then, system (10) has a unique almost periodic solution 𝑥∗
which can be expressed by (11). By Lemma 3, we have where 𝑀1 , 𝑀2 , and 𝑀3 are defined as those in (27), (32), and
(35), respectively.
𝑡 −1
𝛼
𝑥∗ (𝑡) ≥ [𝑏 ∫ 𝑒𝜉𝐴𝑒−𝛼(𝑡−𝑠) 𝑑𝑠] = . (15) Proof. From the first equation of system (5), we have
−∞ 𝑒𝜉𝐴𝑏
On the other hand, 𝑥1̇ (𝑡) ≤ 𝑥1 (𝑡) [𝑟1𝑢 − 𝑎1𝑙 𝑥1 (𝑡)] , 𝑡 ≠𝜏𝑘 ,
(25)
−1
𝑡
𝑎 𝑥1 (𝜏𝑘+ ) = (1 + ℎ1𝑘 ) 𝑥1 (𝜏𝑘 ) , 𝑘 ∈ Z+ .
𝑥∗ (𝑡) ≤ [𝑏 ∫ 𝜂𝑒−𝑎(𝑡−𝑠) 𝑑𝑠] = . (16)
𝑡−𝜃 𝜂𝑏 (1 − 𝑒−𝑎𝜃 )
Consider the following auxiliary system:
Suppose that 𝑥(𝑡) is another positive solution of system
(10). Define a Lyapunov function as 𝑧̇1 (𝑡) = 𝑧1 (𝑡) [𝑟1𝑢 − 𝑎1𝑙 𝑧1 (𝑡)] , 𝑡 ≠𝜏𝑘 ,
󵄨 󵄨 (26)
𝑉 (𝑡) = 󵄨󵄨󵄨ln 𝑥∗ (𝑡) − ln 𝑥 (𝑡)󵄨󵄨󵄨 , ∀𝑡 ∈ R. (17) 𝑧1 (𝜏𝑘+ ) = (1 + ℎ1𝑘 ) 𝑧1 (𝜏𝑘 ) , 𝑘 ∈ Z+ .
For 𝑡 ≠𝜏𝑘 , 𝑘 ∈ Z+ , calculating the upper right derivative of
By Lemma 5, 𝑥1 (𝑡) ≤ 𝑧1 (𝑡), where 𝑧1 (𝑡) is the solution of
𝑉(𝑡) along the solution of system (10), we have
system (26) with 𝑧1 (0+ ) = 𝑥1 (0+ ). By Lemma 7, system (26)
󵄨 󵄨
𝐷+ 𝑉 (𝑡) = −𝑏 󵄨󵄨󵄨𝑥∗ (𝑡) − 𝑥 (𝑡)󵄨󵄨󵄨 . (18) has a unique globally asymptotically stable positive almost
periodic solution 𝑧1∗ which can be expressed as follows:
For 𝑡 = 𝜏𝑘 , 𝑘 ∈ Z+ , we have
𝑡 −1 𝑡 −1
󵄨 󵄨 𝑧1∗ (𝑡) = [𝑎1𝑙 ∫ ≤ [𝑎1𝑙 ∫
𝑉 (𝜏𝑘+ ) = 󵄨󵄨󵄨ln 𝑥∗ (𝜏𝑘+ ) − ln 𝑥 (𝜏𝑘+ )󵄨󵄨󵄨 𝑊1 (𝑡, 𝑠) 𝑑𝑠] 𝑊1 (𝑡, 𝑠) 𝑑𝑠]
−∞ 𝑡−𝜃
󵄨󵄨 (1 + ℎ ) 𝑥∗ (𝜏 ) 󵄨󵄨
󵄨 𝑘 󵄨󵄨
= 󵄨󵄨󵄨󵄨ln 𝑘
󵄨󵄨 (19) 𝑟1𝑢
󵄨󵄨 (1 + ℎ𝑘 ) 𝑥 (𝜏𝑘 ) 󵄨󵄨󵄨 ≤ 𝑢 := 𝑀1 ,
𝜂1 𝑎1𝑙 (1 − 𝑒−𝑟1 𝜃 )
󵄨 󵄨
= 󵄨󵄨󵄨ln 𝑥∗ (𝜏𝑘 ) − ln 𝑥 (𝜏𝑘 )󵄨󵄨󵄨 = 𝑉 (𝜏𝑘 ) . (27)
Abstract and Applied Analysis 5

where if the following condition holds:

𝛼
𝑊1 (𝑡, 𝑠) (𝐻1 ) 𝑟1𝑙 > 𝑚1−1 𝑏1𝑢 𝑀2 + 𝜉1 𝐴, 𝑏2𝑙 𝑁1 /(𝑚1 + 𝑀1 1 ) > 𝑟2𝑢 +
𝛼
𝑢
(𝑏3𝑢 𝑀3 /𝑚2 ) + 𝜉2 𝐴, and 𝑏4𝑙 𝑁2 /(𝑚2 + 𝑀2 2 ) > 𝑟3𝑢 + 𝜉3 𝐴.
{ 𝑒−𝑟1 (𝑡−𝑠) , 𝜏𝑘−1 < 𝑠 < 𝑡 < 𝜏𝑘 ;
{
{
{
{ 𝑘+1 1 𝑢 (28) Proof. According to Proposition 8, there exist 𝜖 > 0 and 𝑇4 >
= {∏ 𝑒−𝑟1 (𝑡−𝑠) , 𝜏𝑚−1 < 𝑠 ≤ 𝜏𝑚 0 such that
{
{ 1 + ℎ 1𝑗
{
{𝑗=𝑚 𝑟1𝑙 − 𝑚1−1 𝑏1𝑢 (𝑀2 + 𝜖) − 𝜉1 𝐴 ≥ 0,
{ < 𝜏𝑘 < 𝑡 ≤ 𝜏𝑘+1 . (37)
𝑥𝑖 (𝑡) ≤ 𝑀𝑖 + 𝜖 for 𝑡 ≥ 𝑇4 , 𝑖 = 1, 2, 3.
Then, for any constant 𝜖 > 0, there exists 𝑇1 > 0 such that
𝑥1 (𝑡) ≤ 𝑧1 (𝑡) < 𝑧1∗ (𝑡) + 𝜖 ≤ 𝑀1 + 𝜖 for 𝑡 > 𝑇1 . So, From the first equation of system (5), we have
𝑏1𝑢 (𝑀2 + 𝜖)
lim sup 𝑥1 (𝑡) ≤ 𝑀1 . (29) 𝑥1̇ (𝑡) ≥ 𝑥1 (𝑡) [𝑟1𝑙 − 𝑎1𝑢 𝑥1 (𝑡) − ],
𝑡→∞ 𝑚1
(38)
For any 𝜖 > 0, there exists 𝑇2 > 0 such that 𝑡 ≠𝜏𝑘 , 𝑡 > 𝑇4 ,

𝑥1 (𝑡) ≤ 𝑀1 + 𝜖 for 𝑡 ≥ 𝑇2 . (30) 𝑥1 (𝜏𝑘+ ) = (1 + ℎ1𝑘 ) 𝑥1 (𝜏𝑘 ) , 𝑘 ∈ Z+ .

Consider the following auxiliary system:
From the second equation of system (5), we have
𝑝1̇ (𝑡) = 𝑝1 (𝑡) [𝑟1𝑙 − 𝑚1−1 𝑏1𝑢 (𝑀2 + 𝜖) − 𝑎1𝑢 𝑥1 (𝑡)] ,
𝑏2𝑢 (𝑀1 + 𝜖)
𝑥2̇ (𝑡) ≤ 𝑥2 (𝑡) [−𝑟2𝑙 + − 𝑎2𝑙 𝑥2 (𝑡)] , 𝑡 ≠𝜏𝑘 , 𝑡 > 𝑇4 , (39)
𝑚1
(31) 𝑝1 (𝜏𝑘+ ) = (1 + ℎ1𝑘 ) 𝑝1 (𝜏𝑘 ) , 𝑘 ∈ Z+ .
𝑡 ≠𝜏𝑘 , 𝑡 > 𝑇2 ,

𝑥2 (𝜏𝑘+ ) = (1 + ℎ2𝑘 ) 𝑥2 (𝜏𝑘 ) , 𝑘 ∈ Z+ . By Remark 15, 𝑥1 (𝑡) ≥ 𝑝1 (𝑡) for 𝑡 > 𝑇4 , where 𝑝1 (𝑡)
is the solution of system (39) with 𝑝1 (𝑇4+ ) = 𝑥1 (𝑇4+ ). By
Similar to the above argument as that in (29), one has Lemma 7, system (39) has a unique globally asymptotically
stable positive almost periodic solution 𝑝1∗ which can be
expressed as follows:
𝑚1−1 𝑏2𝑢 𝑀1 − 𝑟2𝑙
lim sup 𝑥2 (𝑡) ≤ 𝑙 −1 𝑢
:= 𝑀2 . (32) 𝑡 −1
𝑡→∞ 𝜂2 𝑎2𝑙 [1 − 𝑒(𝑟2 −𝑚1 𝑏2 𝑀1 )𝜃 ] 𝑝1∗ (𝑡) = [𝑎1𝑢 ∫ 𝑊2 (𝑡, 𝑠) 𝑑𝑠]
−∞
Then, there exists 𝑇3 > 𝑇2 such that (40)
𝑟𝑙 − 𝑚1−1 𝑏1𝑢 (𝑀2 + 𝜖) − 𝜉1 𝐴
≥ 1 := 𝑁1 (𝜖) ,
𝑥2 (𝑡) ≤ 𝑀2 + 𝜖 for 𝑡 ≥ 𝑇3 . (33) 𝑒𝜉1 𝐴𝑎1𝑢

By the third equation of system (5), we have 𝑊2 (𝑡, 𝑠)

𝑏4𝑢 (𝑀2 + 𝜖) {
𝑙 −1 𝑢
𝑒−[𝑟1 −𝑚1 𝑏1 (𝑀2 +𝜖)−𝜉1 𝐴](𝑡−𝑠) , 𝜏𝑘−1 < 𝑠 < 𝑡 < 𝜏𝑘 ;
𝑥3̇ (𝑡) ≤ 𝑥3 (𝑡) [−𝑟3𝑙 + − 𝑎3𝑙 𝑥3 (𝑡)] , {
{
𝑚2 {
{ 𝑘+1
{
{ 1
(34) {∏ (41)
𝑡 ≠𝜏𝑘 , 𝑡 > 𝑇3 , = {𝑗=𝑚 1 + ℎ1𝑗
{
{
{
{
𝑙 −1 𝑢
×𝑒−[𝑟1 −𝑚1 𝑏1 (𝑀2 +𝜖)−𝜉1 𝐴](𝑡−𝑠) , 𝜏𝑚−1 < 𝑠 ≤ 𝜏𝑚
𝑥3 (𝜏𝑘+ ) = (1 + ℎ3𝑘 ) 𝑥3 (𝜏𝑘 ) , 𝑘 ∈ Z+ . {
{
{
{ < 𝜏𝑘 < 𝑡 ≤ 𝜏𝑘+1 .
Similar to the above argument as that in (32), we have Similar to the above argument as that in (29), we have
lim inf 𝑡 → ∞ 𝑥1 (𝑡) ≥ 𝑁1 (𝜖). By the arbitrariness of 𝜖, it leads
𝑚2−1 𝑏4𝑢 𝑀2 − 𝑟3𝑙 to
lim sup 𝑥3 (𝑡) ≤ 𝑙 −1 𝑢
:= 𝑀3 . (35)
𝑡→∞ 𝜂3 𝑎3𝑙 [1 − 𝑒(𝑟3 −𝑚2 𝑏4 𝑀2 )𝜃 ] 𝑟1𝑙 − 𝑚1−1 𝑏1𝑢 𝑀2 − 𝜉1 𝐴
lim inf 𝑥1 (𝑡) ≥ 𝑁1 := . (42)
This completes the proof.
𝑡→∞ 𝑒𝜉1 𝐴𝑎1𝑢
Then, there exist 0 < 𝜖1 ≤ 𝜖 and 𝑇5 > 𝑇4 such that
Proposition 9. Let 𝑁1 , 𝑁2 , and 𝑁3 be defined as those in (42)–
(48), respectively. Then, every solution (𝑥1 , 𝑥2 , 𝑥3 )𝑇 of system 𝑏2𝑙 (𝑁1 − 𝜖1 ) 𝑏3𝑢 (𝑀3 + 𝜖1 )
𝛼1 − 𝑟2𝑢 − − 𝜉2 𝐴 ≥ 0,
(5) satisfies 𝑚1 + (𝑀1 + 𝜖1 ) 𝑚2 (43)
lim inf 𝑥𝑖 (𝑡) ≥ 𝑁𝑖 , 𝑖 = 1, 2, 3, (36) 𝑥1 (𝑡) ≥ 𝑁1 − 𝜖1 for 𝑡 ≥ 𝑇5 .
𝑡→∞
6 Abstract and Applied Analysis

By the second equation of system (5), we have in Bai and Wang [15]. So, Theorem 11 extends the corre-
sponding result in Bai and Wang [15]. Further, Theorem 11
𝑥2̇ (𝑡) gives the sufficient conditions for the permanence of system
(5) with almost periodic impulsive perturbations. There-
𝑏2𝑙 (𝑁1 − 𝜖1 ) 𝑏3𝑢 (𝑀3 + 𝜖1 ) fore, Theorem 11 provides a possible method to study the
≥ 𝑥2 (𝑡) [ 𝛼1 − 𝑟2𝑢 −
𝑚1 + (𝑀1 + 𝜖1 ) 𝑚2 permanence of the models with impulsive perturbations in
biological populations.
− 𝑎2𝑙 𝑥2 (𝑡) ] 𝑡 ≠𝜏𝑘 , 𝑡 > 𝑇2 , Remark 13. From the proof of Propositions 8 and 9, we
know that, under the conditions of Theorem 11, set 𝑆 =
𝑥2 (𝜏𝑘+ ) = (1 + ℎ2𝑘 ) 𝑥2 (𝜏𝑘 ) , 𝑘 ∈ Z+ . {(𝑥1 , 𝑥2 , 𝑥3 )𝑇 ∈ R3 : 𝑁𝑖 ≤ 𝑥𝑖 ≤ 𝑀𝑖 , 𝑖 = 1, 2, 3} is an invariant
(44) set of system (5).

Similar to the above argument as that in (42), one has

4. Almost Periodic Solution
lim inf 𝑥2 (𝑡)
𝑡→∞ The main result of this paper is concerned with the existence
𝛼
of a unique uniformly asymptotically stable positive almost
[(𝑏2𝑙 𝑁1 / (𝑚1 + 𝑀1 1 )) − 𝑟2𝑢 − (𝑏3𝑢 𝑀3 /𝑚2 ) − 𝜉2 𝐴] periodic solution for system (5).
≥ (45)
𝑒𝜉2 𝐴𝑎2𝑢 Let

:= 𝑁2 .
𝛼1 𝑥𝛼1 −1 𝑀2 𝑏1 (𝑡)
𝑐1 (𝑡) := max 2
,
Then, there exist 0 < 𝜖2 ≤ 𝜖1 and 𝑇6 > 𝑇5 such that 𝑁1 ≤𝑥≤𝑀1 (𝑚1 + 𝑥𝛼1 )

𝑏4𝑙 (𝑁2 − 𝜖2 ) 󵄨󵄨 󵄨
− 𝑟3𝑢 − 𝜉3 𝐴 ≥ 0, 󵄨𝑚 + (1 − 𝛼1 ) 𝑥𝛼1 󵄨󵄨󵄨 𝑏2 (𝑡)
𝛼2 𝑐2 (𝑡) := max 󵄨 1 ,
𝑚2 + (𝑀2 + 𝜖2 ) (46) 𝑁1 ≤𝑥≤𝑀1 (𝑚1 + 𝑥𝛼1 )
2

𝑥2 (𝑡) ≥ 𝑁2 − 𝜖2 for 𝑡 ≥ 𝑇6 .
𝑏1 (𝑡)
𝑑1 (𝑡) := 𝛼 ,
In view of the third equation of system (5), we have 𝑚1 + 𝑁1 1
(49)
𝑏4𝑙 (𝑁2 − 𝜖2 ) 𝛼2 𝑥 𝛼2 −1
𝑀3 𝑏3 (𝑡)
𝑥3̇ (𝑡) ≥ 𝑥3 (𝑡) [ 𝛼2 − 𝑟3𝑢 − 𝑎3𝑢 𝑥3 (𝑡)] , 𝑑2 (𝑡) := max ,
𝑚2 + (𝑀2 + 𝜖2 ) 𝑁2 ≤𝑥≤𝑀2 (𝑚2 + 𝑥𝛼2 )
2

𝑡 ≠𝜏𝑘 , 𝑡 > 𝑇2 , 󵄨󵄨 󵄨
󵄨󵄨𝑚2 + (1 − 𝛼2 ) 𝑥𝛼2 󵄨󵄨󵄨 𝑏4 (𝑡)
𝑥3 (𝜏𝑘+ ) = (1 + ℎ3𝑘 ) 𝑥3 (𝜏𝑘 ) , 𝑘∈Z . + 𝑑3 (𝑡) := max 2
,
𝑁2 ≤𝑥≤𝑀2 (𝑚2 + 𝑥𝛼2 )
(47)

Similar to the above argument as that in (45), one has 𝑏3 (𝑡)

𝑒 (𝑡) := 𝛼 , ∀𝑡 ∈ R.
𝑚2 + 𝑁2 2
𝛼
[(𝑏4𝑙 𝑁2 / (𝑚2 + 𝑀2 2 )) − 𝑟3𝑢 − 𝜉3 𝐴]
lim inf 𝑥3 (𝑡) ≥ := 𝑁3 .
𝑡→∞ 𝑒𝜉3 𝐴 𝑎3𝑢 Theorem 14. Assume that (𝐻1 ) holds; suppose further that
(48)
(𝐻2 ) there exist positive constants 𝜆 1 , 𝜆 2 , 𝜆 3 , and 𝜇 such that
This completes the proof.

Remark 10. In view of (𝐻1 ) in Proposition 9, the values of inf [𝜆 1 𝑎1 (𝑡) − 𝜆 1 𝑐1 (𝑡) − 𝜆 2 𝑐2 (𝑡)] > 𝜇,
impulse coefficients ℎ𝑖𝑘 (𝑖 = 1, 2, 3) and the number of the 𝑡∈R
impulse points 𝜏𝑘 in each interval of length 1 have negative
effect on the permanence of system (5). inf [𝜆 3 𝑎3 (𝑡) − 𝜆 2 𝑒1 (𝑡)] > 𝜇, (50)
𝑡∈R

By Propositions 8 and 9, we have the following theorem.

inf [𝜆 2 𝑎2 (𝑡) − 𝜆 1 𝑑1 (𝑡) − 𝜆 2 𝑑2 (𝑡) − 𝜆 3 𝑑3 (𝑡)] > 𝜇;
𝑡∈R
Theorem 11. Assume that (𝐻1 ) holds; then, system (5) is
permanent.

Remark 12. When ℎ𝑖𝑘 (𝑖 = 1, 2, 3) ≡ 0 in system (5), then then, system (5) admits a unique positive almost periodic
Theorem 11 changes to the corresponding permanence result solution, which is uniformly asymptotically stable.
Abstract and Applied Analysis 7

Proof. Suppose that 𝑍(𝑡) = (𝑧1 (𝑡), 𝑧2 (𝑡), 𝑧3 (𝑡))𝑇 and 𝑍∗ (𝑡) = where 𝜆 := min{𝜆 1 , 𝜆 2 , 𝜆 3 } min{𝑀1−1 , 𝑀2−1 , 𝑀3−1 }. Further,
(𝑧1∗ (𝑡), 𝑧2∗ (𝑡), 𝑧3∗ (𝑡))𝑇 are any two solutions of system (5). we have
Consider the product system of system (5) as
𝑉 (𝑡, 𝑍, 𝑍∗ )
𝑏1 (𝑡) 𝑧2 (𝑡) 3
𝑧̇1 (𝑡) = 𝑧1 (𝑡) [𝑟1 (𝑡) − 𝑎1 (𝑡) 𝑧1 (𝑡) − ], 󵄨 󵄨
𝛼
𝑚1 + 𝑧1 1 (𝑡) ≤ max {𝜆 1 , 𝜆 2 , 𝜆 3 } ∑ 󵄨󵄨󵄨ln 𝑧𝑖 (𝑡) − ln 𝑧𝑖∗ (𝑡)󵄨󵄨󵄨
𝑖=1
𝑏2 (𝑡) 𝑧1 (𝑡)
𝑧̇2 (𝑡) = 𝑧2 (𝑡) [−𝑟2 (𝑡) + 𝛼
3
1 󵄨󵄨 󵄨 󵄩 ∗󵄩
󵄨󵄨𝑧𝑖 (𝑡) − 𝑧𝑖 (𝑡)󵄨󵄨󵄨 ≤ 𝜆 󵄩󵄩󵄩𝑍 − 𝑍 󵄩󵄩󵄩 ,
𝑚1 + 𝑧1 1 (𝑡) ∗
≤ max {𝜆 1 , 𝜆 2 , 𝜆 3 } ∑
𝑁
𝑖=1 𝑖
𝑏3 (𝑡) 𝑧3 (𝑡)
−𝑎2 (𝑡) 𝑧2 (𝑡) − 𝛼 ], (54)
𝑚2 + 𝑧2 2 (𝑡)
𝑏4 (𝑡) 𝑧2 (𝑡)
𝑧̇3 (𝑡) = 𝑧3 (𝑡) [−𝑟3 (𝑡) + 𝛼 − 𝑎3 (𝑡) 𝑧3 (𝑡)] , where 𝜆 := max{𝜆 1 , 𝜆 2 , 𝜆 3 } max{𝑁1−1 , 𝑁2−1 , 𝑁3−1 }; thus, (1) in
𝑚2 + 𝑧2 2 (𝑡) Lemma 4 is satisfied.
𝑏1 (𝑡) 𝑧2∗ (𝑡) Since
𝑧̇1∗ (𝑡) = 𝑧1∗ (𝑡) [𝑟1 (𝑡) − 𝑎1 (𝑡) 𝑧1∗ (𝑡) − ∗𝛼 ],
𝑚1 + 𝑧1 1 (𝑡) 󵄨󵄨 ∗ 󵄨
󵄨󵄨𝑉 (𝑡, 𝑍, 𝑍∗ ) − 𝑉 (𝑡, 𝑍, 𝑍 )󵄨󵄨󵄨
𝑏 (𝑡) 𝑧1∗ (𝑡) 󵄨 󵄨
𝑧̇2∗ (𝑡) = 𝑧2∗ (𝑡) [−𝑟2 (𝑡) + 2 ∗𝛼 − 𝑎2 (𝑡) 𝑧2∗ (𝑡) 3
𝑚1 + 𝑧1 1 (𝑡) 󵄨 󵄨
= ∑𝜆 𝑖 󵄨󵄨󵄨ln 𝑧𝑖 (𝑡) − ln 𝑧𝑖∗ (𝑡)󵄨󵄨󵄨
𝑏3 (𝑡) 𝑧3∗ (𝑡) 𝑖=1
− ∗𝛼 ],
𝑚2 + 𝑧2 2 (𝑡) 3
󵄨 󵄨
𝑏4 (𝑡) 𝑧2∗ (𝑡) − ∑𝜆 𝑖 󵄨󵄨󵄨ln 𝑧𝑖 (𝑡) − ln 𝑧∗𝑖 (𝑡)󵄨󵄨󵄨 (55)
𝑧̇3∗ (𝑡) = 𝑧3∗ (𝑡) [−𝑟3 (𝑡) + ∗𝛼 − 𝑎3 (𝑡) 𝑧3∗ (𝑡)] , 𝑖=1
𝑚2 + 𝑧2 2 (𝑡)
3
𝑡 ≠𝜏𝑘 , 󵄨 󵄨 󵄨 󵄨
≤ 𝜆∑ [󵄨󵄨󵄨𝑧𝑖 (𝑡) − 𝑧𝑖∗ (𝑡)󵄨󵄨󵄨 + 󵄨󵄨󵄨𝑧𝑖 (𝑡) − 𝑧∗𝑖 (𝑡)󵄨󵄨󵄨]
Δ𝑧1 (𝜏𝑘 ) = ℎ1𝑘 𝑧1 (𝜏𝑘 ) , 𝑖=1

󵄨 󵄨 󵄨 ∗ 󵄨
Δ𝑧2 (𝜏𝑘 ) = ℎ2𝑘 𝑧2 (𝜏𝑘 ) , = 𝜆 [󵄨󵄨󵄨𝑍 (𝑡) − 𝑍∗ (𝑡)󵄨󵄨󵄨 + 󵄨󵄨󵄨󵄨𝑍 (𝑡) − 𝑍 (𝑡)󵄨󵄨󵄨󵄨] ,
Δ𝑧3 (𝜏𝑘 ) = ℎ3𝑘 𝑧3 (𝜏𝑘 ) ,
Δ𝑧1∗ (𝜏𝑘 ) = ℎ1𝑘 𝑧1∗ (𝜏𝑘 ) , (2) in Lemma 4 holds.
For 𝑡 ≠𝜏𝑘 , 𝑘 ∈ Z+ , calculating the upper right derivative
Δ𝑧2∗ (𝜏𝑘 ) = ℎ2𝑘 𝑧2∗ (𝜏𝑘 ) , of 𝑉(𝑡) along the solution of system (51), we have
Δ𝑧3∗ (𝜏𝑘 ) = ℎ3𝑘 𝑧3∗ (𝜏𝑘 ) , 𝑘 ∈ Z.
(51) 3
𝑧̇𝑖 (𝑡) 𝑧̇𝑖∗ (𝑡)
𝐷+ 𝑉 (𝑡) = ∑𝜆 𝑖 [ − ] sgn (𝑧𝑖 (𝑡) − 𝑧𝑖∗ (𝑡))
𝑖=1 𝑧𝑖 (𝑡) 𝑧𝑖∗ (𝑡)
Set 𝑆1 = {(𝑧1 , 𝑧2 , 𝑧3 )𝑇 ∈ R3 : 𝑁𝑖 ≤ 𝑧𝑖 ≤ 𝑀𝑖 , 𝑖 =
1, 2, 3}, which is an invariant set of system (51) directly from = 𝜆 1 sgn (𝑧1 (𝑡)
Remark 13.
Construct a Lyapunov functional 𝑉(𝑡, 𝑍, 𝑍∗ ) =
𝑇 ∗ ∗ ∗ 𝑇 + −𝑧1∗ (𝑡)) {−𝑎1 (𝑡) [𝑧1 (𝑡) − 𝑧1∗ (𝑡)]
𝑉(𝑡, (𝑧1 , 𝑧2 , 𝑧3 ) , (𝑧1 , 𝑧2 , 𝑧3 ) ) defined on R × 𝑆1 × 𝑆1 × 𝑆1
as follows:
𝑏1 (𝑡) 𝑧2 (𝑡) 𝑏 (𝑡) 𝑧2∗ (𝑡)
3 − 𝛼1 + 1 }
󵄨 󵄨 𝑚1 + 𝑧1 (𝑡) 𝑚1 + 𝑧1∗𝛼1 (𝑡)
𝑉 (𝑡, 𝑍, 𝑍∗ ) = ∑𝜆 𝑖 󵄨󵄨󵄨ln 𝑧𝑖 (𝑡) − ln 𝑧𝑖∗ (𝑡)󵄨󵄨󵄨 . (52)
𝑖=1
+ 𝜆 2 sgn (𝑧2 (𝑡) − 𝑧2∗ (𝑡))
It is obvious that
× {−𝑎2 (𝑡) [𝑧2 (𝑡) − 𝑧2∗ (𝑡)]
𝑉 (𝑡, 𝑍, 𝑍∗ )
3 𝑏2 (𝑡) 𝑧1 (𝑡) 𝑏2 (𝑡) 𝑧1∗ (𝑡)
󵄨 󵄨 + −
≥ min {𝜆 1 , 𝜆 2 , 𝜆 3 } ∑ 󵄨󵄨󵄨ln 𝑧𝑖 (𝑡) − ln 𝑧𝑖∗ (𝑡)󵄨󵄨󵄨 𝛼
𝑚1 + 𝑧1 1 (𝑡) 𝑚1 + 𝑧1∗𝛼1 (𝑡)
𝑖=1
3
1 󵄨󵄨 𝑏3 (𝑡) 𝑧3 (𝑡) 𝑏 (𝑡) 𝑧3∗ (𝑡)
󵄨 󵄩 ∗󵄩 − + 3 }
󵄨󵄨𝑧𝑖 (𝑡) − 𝑧𝑖 (𝑡)󵄨󵄨󵄨 ≥ 𝜆 󵄩󵄩󵄩𝑍 − 𝑍 󵄩󵄩󵄩 ,
∗
≥ min {𝜆 1 , 𝜆 2 , 𝜆 3 } ∑ 𝛼2
𝑚2 + 𝑧2 (𝑡) 𝑚2 + 𝑧2∗𝛼2 (𝑡)
𝑖=1 𝑀𝑖
(53) + 𝜆 3 sgn (𝑧3 (𝑡) − 𝑧3∗ (𝑡))
8 Abstract and Applied Analysis

× {−𝑎3 (𝑡) [𝑧3 (𝑡) − 𝑧3∗ (𝑡)]

5. An Example and Numerical Simulations
Example 1. Consider the following food chain system with
impulsive perturbations:
𝑏4 (𝑡) 𝑧2 (𝑡) 𝑏4 (𝑡) 𝑧2∗ (𝑡)
+ 𝛼 − ∗𝛼 }
𝑚2 + 𝑧2 2 (𝑡) 𝑚2 + 𝑧2 2 (𝑡)
𝑥1̇ (𝑡) = 𝑥1 (𝑡) [2 + cos (√2𝑡) − 10𝑥1 (𝑡)
󵄨 󵄨
≤ − [𝜆 1 𝑎1 (𝑡) − 𝜆 1 𝑐1 (𝑡) − 𝜆 2 𝑐2 (𝑡)] 󵄨󵄨󵄨𝑧1 (𝑡) − 𝑧1∗ (𝑡)󵄨󵄨󵄨
0.1 cos (√3𝑡) 𝑥2 (𝑡)
− [𝜆 2 𝑎2 (𝑡) − 𝜆 1 𝑑1 (𝑡) − 𝜆 2 𝑑2 (𝑡) − 𝜆 3 𝑑3 (𝑡)] − ],
1 + 𝑥1 (𝑡)
󵄨 󵄨
× 󵄨󵄨󵄨𝑧2 (𝑡) − 𝑧2∗ (𝑡)󵄨󵄨󵄨

󵄨 󵄨 𝑥2̇ (𝑡) = 𝑥2 (𝑡) [ − 0.1 sin (√2𝑡)

− [𝜆 3 𝑎3 (𝑡) − 𝜆 2 𝑒1 (𝑡)] 󵄨󵄨󵄨𝑧3 (𝑡) − 𝑧3∗ (𝑡)󵄨󵄨󵄨

3
𝜇 (5 + cos (√2𝑡)) 𝑥1 (𝑡)
󵄨 󵄨
≤ −∑ 𝜆 𝑖 󵄨󵄨󵄨ln 𝑧𝑖 (𝑡) − ln 𝑧𝑖∗ (𝑡)󵄨󵄨󵄨 +
𝑚1 + 𝑥1 (𝑡)
𝜆 𝑁
𝑖=1 𝑖 𝑖

𝜇 𝜇 𝜇 0.1𝑥3 (𝑡)
≤ − min { , , } 𝑉 (𝑡, 𝑍, 𝑍∗ ) . −10𝑥2 (𝑡) − ],
𝜆 1 𝑁1 𝜆 2 𝑁2 𝜆 3 𝑁3 1 + 𝑥2 (𝑡)
(56)
𝑥3̇ (𝑡) = 𝑥3 (𝑡) [ − 0.1 cos (√5𝑡)

For 𝑡 = 𝜏𝑘 , 𝑘 ∈ Z+ , we have
(6 + sin (√3𝑡)) 𝑥2 (𝑡)
+ −10𝑥3 (𝑡) ] ,
1 + 𝑥2 (𝑡)
3
󵄨 󵄨
𝑉 (𝜏𝑘+ , 𝑍 (𝜏𝑘+ ) , 𝑍∗ (𝜏𝑘+ )) = ∑𝜆 𝑖 󵄨󵄨󵄨ln 𝑧𝑖 (𝜏𝑘+ ) − ln 𝑧𝑖∗ (𝜏𝑘+ )󵄨󵄨󵄨
𝑖=1 𝑡 ≠𝜏𝑘 ,

3 󵄨󵄨 (1 + ℎ ) 𝑧 (𝜏 ) 󵄨󵄨 Δ𝑥1 (𝜏𝑘 ) = 0.1𝑥1 (𝜏𝑘 ) ,

󵄨 𝑖 𝑘 󵄨󵄨
= ∑𝜆 𝑖 󵄨󵄨󵄨󵄨ln 𝑖𝑘
󵄨
∗ (𝜏 ) 󵄨󵄨
𝑖=1 󵄨 󵄨 (1 + ℎ𝑖𝑘 ) 𝑧 𝑘 󵄨󵄨
𝑖 (57) Δ𝑥2 (𝜏𝑘 ) = 0.2𝑥2 (𝜏𝑘 ) ,
3
󵄨 󵄨 Δ𝑥3 (𝜏𝑘 ) = 0.3𝑥3 (𝜏𝑘 ) , 𝑘 ∈ {0, 1, . . .} = Z+ ,
= ∑𝜆 𝑖 󵄨󵄨󵄨ln 𝑧𝑖 (𝜏𝑘 ) − ln 𝑧𝑖∗ (𝜏𝑘 )󵄨󵄨󵄨
𝑖=1 𝜃 = inf 𝜏𝑘1 = 1;
𝑘∈Z

= 𝑉 (𝜏𝑘 , 𝑍 (𝜏𝑘 ) , 𝑍∗ (𝜏𝑘 )) . (58)

then, system (58) is permanent and admits a unique uni-
formly asymptotically stable positive almost periodic solu-
In view of (56)-(57), (3) in Lemma 4 is satisfied. tion.
By Lemma 4, system (5) admits a unique uniformly
asymptotically stable positive almost periodic solution Proof. Corresponding to system (2), 𝑟1 (𝑡) = 2 + cos(√2𝑡),
(𝑧1 (𝑡), 𝑧2 (𝑡), 𝑧3 (𝑡))𝑇 . This completes the proof. 𝑟2 (𝑡) = 0.1 sin(√2𝑡), 𝑟3 (𝑡) = 0.1 cos(√5𝑡), 𝑎1 = 𝑎2 = 𝑎3 ≡
10, 𝑏1 (𝑡) = 0.1 cos(√3𝑡), 𝑏2 (𝑡) = 5 + cos(√2𝑡), 𝑏3 (𝑡) = 0.1,
Remark 15. When ℎ𝑖𝑘 (𝑖 = 1, 2, 3) ≡ 0 in system (5), then 𝑏4 (𝑡) = 6 + sin(√3𝑡), 𝑚1 = 𝑚2 = 𝛼1 = 𝛼2 = 1, ℎ1𝑘 ≡ 0.1,
Theorem 14 changes to the corresponding permanence result ℎ2𝑘 ≡ 0.2, ℎ3𝑘 ≡ 0.3, 𝑘 ∈ Z+ . Taking 𝜆 1 = 𝜆 2 = 𝜆 3 = 1, the
in Bai and Wang [15]. So, Theorem 14 extends the corre- result is easy to obtain from Theorems 11 and 14; we would
sponding result in Bai and Wang [15]. Further, Theorem 14 omit it (see Figures 1, 2, 3, 4, 5, and 6). This completes the
gives the sufficient conditions for the uniform asymptotical proof.
stability of a unique positive almost periodic solution of
system (5), in which 𝛼1 and 𝛼2 are allowed to be any real-
valued positive number. Therefore, Theorem 14 provides a 6. Conclusion
possible method to study the existence, uniqueness, and
stability of positive almost periodic solution of the models By using the comparison theorem and the Lyapunov method
with impulsive perturbations in biological populations. of the impulsive differential equations, sufficient conditions
Abstract and Applied Analysis 9

0.7 3
0.6 2.5
0.5
2
0.4
x1 (t)

x2 (t)
1.5
0.3
1
0.2
0.1 0.5

0 0
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
Time t Time t

Figure 1: State variable 𝑥1 of Example 1. Figure 5: Stability of state variable 𝑥2 of Example 1.

1 2

0.8
1.5
x2 (t)

0.6

x3 (t)
1
0.4

0.2 0.5

0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
Time t Time t

Figure 2: State variable 𝑥2 of Example 1. Figure 6: Stability of state variable 𝑥3 of Example 1.

1.3
1.2 are obtained which guarantee the permanence and existence
1.1 of a unique uniformly asymptotically stable positive almost
1 periodic solution of a food chain system with almost peri-
0.9 odic impulsive perturbations. Proposition 9 and Theorem 14
x3 (t)

0.8 imply that the values of impulse coefficients ℎ𝑖𝑘 (𝑖 = 1, 2, 3)

0.7
and the number of the impulse points 𝜏𝑘 in each interval
0.6
0.5
of length 1 are harm for the permanence and existence
0.4
of a unique uniformly asymptotically stable positive almost
periodic solution of the model. The main results obtained in
0 5 10 15 20 25 30 35 40 this paper are completely new and the method used in this
Time t paper provides a possible method to study the permanence
and existence of a unique uniformly asymptotically stable
Figure 3: State variable 𝑥3 of Example 1.
positive almost periodic solution of the models with impul-
sive perturbations in biological populations.
2.5

2
Conflict of Interests
The authors declare that there is no conflict of interests
1.5
regarding the publication of this paper.
x1 (t)

1
Acknowledgments
0.5
The authors would like to thank the reviewers for their
0 valuable comments and constructive suggestions, which con-
0 5 10 15 20 25 30 35 40
Time t
siderably improve the presentation of this paper. This work
was supported by the Scientific Research Fund of Yunnan
Figure 4: Stability of state variable 𝑥1 of Example 1. Provincial Education Department.
10 Abstract and Applied Analysis

References [19] Y. Li and T. Zhang, “Permanence and almost periodic sequence

solution for a discrete delay logistic equation with feedback
[1] A. A. Berryman, “The origins and evolution of predator-prey control,” Nonlinear Analysis: Real World Applications, vol. 12, no.
theory,” Ecology, vol. 73, no. 5, pp. 1530–1535, 1992. 3, pp. 1850–1864, 2011.
[2] Y. G. Sun and S. H. Saker, “Positive periodic solutions of discrete [20] T. Zhang, Y. Li, and Y. Ye, “On the existence and stability of
three-level food-chain model of Holling type II,” Applied Math- a unique almost periodic solution of Schoener’s competition
ematics and Computation, vol. 180, no. 1, pp. 353–365, 2006. model with pure-delays and impulsive effects,” Communications
[3] Y. K. Li, L. H. Lu, and X. Y. Zhu, “Existence of periodic solutions in Nonlinear Science and Numerical Simulation, vol. 17, no. 3, pp.
in n-species food-chain system with impulsive,” Nonlinear 1408–1422, 2012.
Analysis: Real World Applications, vol. 7, no. 3, pp. 414–431, 2006. [21] Z. Tian-Wei-Tian, “Multiplicity of positive almost periodic
[4] R. Xu, L. Chen, and F. Hao, “Periodic solutions of a discrete time solutions in a delayed Hassell-Varley-type predator-prey model
Lotka-Volterra type food-chain model with delays,” Applied with harvesting on prey,” Mathematical Methods in the Applied
Mathematics and Computation, vol. 171, no. 1, pp. 91–103, 2005. Sciences, vol. 37, no. 5, pp. 686–697, 2014.
[5] C. X. Shen, “Permanence and global attractivity of the food- [22] T. W. Zhang, Y. K. Li, and Y. Ye, “Persistence and almost
chain system with Holling IV type functional response,” Applied periodic solutions for a discrete fishing model with feedback
Mathematics and Computation, vol. 194, no. 1, pp. 179–185, 2007. control,” Communications in Nonlinear Science and Numerical
[6] H. Baek, “A food chain system with Holling type IV functional Simulation, vol. 16, no. 3, pp. 1564–1573, 2011.
response and impulsive perturbations,” Computers & Mathe- [23] T. Zhang and X. Gan, “Existence and permanence of almost
matics with Applications, vol. 60, no. 5, pp. 1152–1163, 2010. periodic solutions for Leslie-Gower predator-prey model with
[7] G. H. Cui and X. P. Yan, “Stability and bifurcation analysis on a variable delays,” Electronic Journal of Differential Equations, vol.
three-species food chain system with two delays,” Communica- 2013, no. 105, pp. 1–21, 2013.
tions in Nonlinear Science and Numerical Simulation, vol. 16, no. [24] T. W. Zhang and X. R. Gan, “Almost periodic solutions for a
9, pp. 3704–3720, 2011. discrete fishing model with feedback control and time delays,”
[8] Z. Liu, S. Zhong, and X. Liu, “Permanence and periodic solu- Communications in Nonlinear Science and Numerical Simula-
tions for an impulsive reaction-diffusion food-chain system tion, vol. 19, no. 1, pp. 150–163, 2014.
with ratio-dependent functional response,” Communications in [25] T. W. Zhang, “Almost periodic oscillations in a generalized
Nonlinear Science and Numerical Simulation, vol. 19, no. 1, pp. Mackey-Glass model of respiratory dynamics with several
173–188, 2014. delays,” International Journal of Biomathematics, vol. 7, no. 3,
[9] F. Brauer and A. C. Soudack, “Coexistence properties of some Article ID 1450029, 22 pages, 2014.
predator-prey systems under constant rate harvesting and
stocking,” Journal of Mathematical Biology, vol. 12, no. 1, pp. 101–
114, 1981.
[10] V. Lakshmikantham, D. D. Bainov, and P. S. Simeonov, Theory
of Impulsive Differential Equations, World Scientific, 1989.
[11] G. T. Stamov, Almost Periodic Solutions of Impulsive Differential
Equations, Lecture Notes in Mathematics, Springer, Heidelberg,
Germany, 2012.
[12] D. D. Bainov and P. S. Simeonov, Impulsive Differential Equa-
tions: Periodic Solutions and Applications, Longman Scientific
and Technical, 1993.
[13] A. M. Samoilenko and N. A. Perestyuk, Impulsive Differential
Equations, World Scientific, Singapore, 1995.
[14] S. W. Zhang and D. J. Tan, “Permanence in a food chain system
with impulsive perturbations,” Chaos, Solitons and Fractals, vol.
40, no. 1, pp. 392–400, 2009.
[15] L. Bai and K. Wang, “Almost periodic solution of a non-autono-
mous food chains system with Holling’s type II functional
response,” Soochow Journal of Mathematics, vol. 28, no. 3, pp.
267–279, 2002.
[16] Y. Xia, J. Cao, H. Zhang, and F. Chen, “Almost periodic solutions
of 𝑛-species competitive system with feedback controls,” Journal
of Mathematical Analysis and Applications, vol. 294, no. 2, pp.
503–522, 2004.
[17] X. Z. Meng and L. S. Chen, “Almost periodic solution of
non-autonomous Lotka-Volterra predator-prey dispersal sys-
tem with delays,” Journal of Theoretical Biology, vol. 243, no. 4,
pp. 562–574, 2006.
[18] X. Lin and F. Chen, “Almost periodic solution for a Volterra
model with mutual interference and Beddington-DeAngelis
functional response,” Applied Mathematics and Computation,
vol. 214, no. 2, pp. 548–556, 2009.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 786962, 6 pages
http://dx.doi.org/10.1155/2014/786962

Research Article
Limit Cycles in a Cubic Kolmogorov System with Harvest and
Two Positive Equilibrium Points

Qi-Ming Zhang,1 Feng Li,2 and Yulin Zhao1

1
College of Science, Hunan University of Technology, Zhuzhou, Hunan 412007, China
2
College of Science, Linyi University, Linyi, Shandong 276005, China

Correspondence should be addressed to Feng Li; lf0539@126.com

Received 23 June 2014; Accepted 9 July 2014; Published 20 July 2014

Academic Editor: Tonghua Zhang

Copyright © 2014 Qi-Ming Zhang et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

A class of planar cubic Kolmogorov systems with harvest and two positive equilibrium points is investigated. With the help of
computer algebra system MATHEMATICA, we prove that five limit cycles can be bifurcated simultaneously from the two critical
points (1, 1) and (2, 2), respectively, in the first quadrant. Moreover, the necessary conditions of centers are obtained.

1. Introduction not the case, even when 𝐹 and 𝐺 factories. There are many
contributions about this system (see [1, 2]). The latter poses
In mathematical ecology, a class of systems of the form the question whether a predator-prey system can have two
or more ecologically stable cycles. If 𝐹 and 𝐺 are cubic, there
𝑑𝑥𝑖
= 𝑥𝑖 𝐹 (𝑥1 , . . . , 𝑥𝑛 ) , 𝑖 = 1, . . . , 𝑛, (1) are also many works to consider its limit cycles and dynamics
𝑑𝑡 behaviors, see [3, 4]. In [5], the authors discussed a class
are frequently used to model the interaction of species occu- of cubic Kolmogorov systems with three invariant algebraic
pying the same ecological niche. The differential equations curves.
modeling the interaction of two species Recently, a system with three positive equilibrium points
has been investigated, the authors have investigated the
𝑑𝑥 center-focus problems and limit cycles bifurcations. They
= 𝑥𝐹 (𝑥, 𝑦) ,
𝑑𝑡 have proved that each of the two points (1, 2) and (2, 1) can
(2) bifurcate 1 small limit cycle under a certain condition, and 3
𝑑𝑦
= 𝑦𝐺 (𝑥, 𝑦) , limit cycles can occur near point (1, 1) at the same step [6].
𝑑𝑡 Other Kolmogorov systems were also investigated recently
being known as Kolmogorov systems have been studied in [7, 8]. In this paper, we will consider limit cycles which
extensively. It is well known that there is no limit cycles in bifurcate from a class of systems of the form
the classical Lotka-Volterra-Gause model, where 𝐹 and 𝐺 are
linear. There can of course only be one critical point in the 𝑑𝑥 7
interior of the realistic quadrant (𝑥 > 0, 𝑦 > 0) in this = −2 (2𝛽 + 𝜃) + (8𝛽 + 5𝜃) 𝑥 − (2𝛽 + 𝜃) 𝑥2
𝑑𝑡 2
case, but this can be a centre; however, there are no isolated
1 1
periodic solutions. + (4𝛽 − 𝜃) 𝑥𝑦 + 𝑎20 𝑥3 + (6𝛽 + 3𝜃 − 4𝑎20 ) 𝑥2 𝑦
If 𝐹 and 𝐺 are quadratic, one might think by analogy that 2 2
the behavior within the first quadrant is similar to that of 1
a quadratic system. The examples we give show that this is + (−4𝛽 − 𝜃 + 2𝑎20 ) 𝑥2 𝑦,
2
2 Abstract and Applied Analysis

𝑑𝑦 7
= 2 (2𝛽 + 𝜃) − (8𝛽 + 5𝜃) 𝑦 + (2𝛽 + 𝜃) 𝑦2 Furthermore, by the transformations
𝑑𝑡 2
1 1
+ (−4𝛽 + 𝜃) 𝑥𝑦 + (2𝑏20 − 4𝛽 − 𝜃) 𝑥3 + 𝑏20 𝑥2 𝑦 𝑧 = 𝑥 + 𝑖𝑦, 𝑤 = 𝑥 − 𝑖𝑦,
2 2
(5)
1 𝑇 = 𝑖𝜏, 𝑖 = √−1,
+ (2𝛽 − 𝜃 − 4𝑏20 ) 𝑥𝑦2 ,
2
(3)

which have two positive equilibrium points (1, 1) and system (4) can be transformed into the following system:
(2, 2). We use our Computer Algebra procedure Mathematic
(described in [9]) to compute the focal values at the critical
points (1, 1) and (2, 2). We will show that five limit cycles 𝑑𝑧 1
can bifurcate from points (1, 1) and (2, 2) simultaneously. =𝑧+ (−2 (2𝑎20 + 2𝑖𝑏20 − (6 − 2𝑖) 𝛽 − (3 − 2𝑖) 𝜃) 𝑧2
𝑑𝑇 8𝛽
Furthermore, some necessary and sufficient conditions for
two positive equilibrium points to be centers are also to be − 2 (−4𝑖𝑎20 + 4𝑏20 + (4 + 12𝑖) 𝛽 + (3 + 5𝑖) 𝜃) 𝑧𝑤
given.
This paper is divided into three sections. In Section 2, we + 2 (2𝑎20 + 2𝑖𝑏20 − (2 + 2𝑖) 𝛽 − 𝑖𝜃) 𝑤2
use the recursive algorithm to obtain that 5 limit circles could
be bifurcated from points (1, 1) and (2, 2). In Section 3, nec-
essary and sufficient conditions for two positive equilibrium + (−2𝑎20 − 2𝑏20 + (4 + 4𝑖) 𝛽 + (1 + 𝑖) 𝜃) 𝑧3
points to be centers are proved.
+ 2 ((−1 + 2𝑖) 𝑎20 + (1 + 2𝑖) 𝑏20

2. Bifurcations of Limit Circles at Two Positive + (1 − 4𝑖) 𝛽 + (1 − 𝑖) 𝜃) 𝑧2 𝑤

Equilibrium Points
+ ((2 + 4𝑖) 𝑎20 + (2 − 4𝑖) 𝑏20
First of all, it is easy to testify that points (1, 1) and (2, 2) are
two positive equilibriums of system (3) and they are all center
or focus. So we need to compute the Lyapunov constants to − (1 − 𝑖) (4𝛽 + 𝜃)) 𝑧𝑤2
determine its kind of singular. Now, we consider the point
(1, 1). −2 (−𝑎20 + 𝑏20 + 𝛽 + 𝜃)) 𝑤3 ;
Through the transformations 𝑥̃ = 𝑥−1, 𝑦̃ = 𝑦−1, 𝜏 = 𝛽𝑡,
we still denote 𝑥, ̃ 𝑦̃ by 𝑥, 𝑦 for convenience. The point (1, 1) 𝑑𝑤 1
= −𝑤 − (−2 (2𝑎20 − 2𝑖𝑏20 − (6 + 2𝑖) 𝛽 − (3 + 2𝑖) 𝜃) 𝑤2
will be moved into (0, 0) of the new system, and the system 𝑑𝑇 8𝛽
will be changed into the following system:
− 2 (4𝑖𝑎20 + 4𝑏20 + (4 − 12𝑖) 𝛽+(3 − 5𝑖) 𝜃)𝑧𝑤
𝑑𝑥
= 4𝑥2 − 𝑦 − 4𝑥𝑦 − 3𝑥2 𝑦 + 2𝑦2 + 2𝑥𝑦2
𝑑𝜏
+ 2 (2𝑎20 − 2𝑖𝑏20 − (2 − 2𝑖) 𝛽 + 𝑖𝜃) 𝑧2
(𝑎20 𝑥2 ) (𝑎20 𝑥3 ) (2𝑎20 𝑥𝑦)
− − + + (−2𝑎20 − 2𝑏20 + (4 − 4𝑖) 𝛽 + (1 − 𝑖) 𝜃) 𝑤3
𝛽 𝛽 𝛽

(2𝑎20 𝑥2 𝑦) (𝑎20 𝑦2 ) (𝑎20 𝑥𝑦2 ) (2𝑥2 𝜃) + 2 ((−1 − 2𝑖) 𝑎20 + (1 − 2𝑖) 𝑏20
+ − − +
𝛽 𝛽 𝛽 𝛽
+ (1 + 4𝑖) 𝛽 + (1 + 𝑖) 𝜃) 𝑤2 𝑧
2 2 2
(3𝑥𝑦𝜃) (3𝑥 𝑦𝜃) (𝑦 𝜃) (𝑥𝑦 𝜃)
− − + + , (4) + ((2 − 4𝑖) 𝑎20 + (2 + 4𝑖) 𝑏20
(2𝛽) (2𝛽) (2𝛽) (2𝛽)
2
− (1 + 𝑖) (4𝛽 + 𝜃)) 𝑤𝑧2
𝑑𝑦 (𝑏20 𝑥 ) (2𝑏20 𝑥𝑦)
= 𝑥 − 2𝑦2 − 𝑥𝑦2 + 2𝑦3 − +
𝑑𝜏 𝛽 𝛽 −2 (−𝑎20 + 𝑏20 + 𝛽 + 𝜃)) 𝑧3 ;
(6)
(𝑏20 𝑥2 𝑦) (𝑏20 𝑦2 ) (2𝑏20 𝑥𝑦2 ) (𝑏20 𝑦3 )
− − + −
𝛽 𝛽 𝛽 𝛽
2 2 3 applying the recursive formulae in Theorem 2.5 in [9], we
(𝑥𝑦𝜃) (3𝑦 𝜃) (𝑥𝑦 𝜃) (𝑦 𝜃)
+ − + + . compute singular point quantities and simplify them; then,
(2𝛽) (2𝛽) (2𝛽) (2𝛽) we have the following theorem.
Abstract and Applied Analysis 3

Theorem 1. The first four singular point quantities at the origin Proof. 𝑢1 = 𝑢2 = 𝑢3 = 0 implies that the relations above
of system (6) are as follows: among parameters hold. Further, when 𝑎20 = (1/4)(4𝑏20 +
12𝛽 + 9𝜃), 344𝛽2 + 750𝛽𝜃 + 325𝜃2 = 0,

𝑖 𝑓1
𝑢1 = (−2𝑎20 − 2𝑏20 + 4𝛽 + 𝜃)
8𝛽2 2 2
= −3164304640𝑏20 + 94432128√1153𝑏20
× (−4𝑎20 + 4𝑏20 + 12𝛽 + 9𝜃) ,
− 505511440𝑏20 𝜃 + 16321424√1153𝑏20 𝜃
𝑖
𝑢2 = (−2𝑎20 − 2𝑏20 + 4𝛽 + 𝜃) − 1443860355𝜃2 + 42746637√1153𝜃2 ;
192𝛽3
(7) 𝑓2
2 2
× (344𝛽 + 750𝛽𝜃 + 325𝜃 ) ,
4 4
= −748902307758080𝑏20 + 22349440598016√1153𝑏20
𝑖
𝑢3 = − (−2𝑎20 − 2𝑏20 + 4𝛽 + 𝜃) 𝑓1 , 3 3
12288𝛽5 − 239280807055360𝑏20 𝜃 + 7725648121856√1153𝑏20 𝜃
2 2 2 2
𝑖 +5835163629391040𝑏20 𝜃 −171639395144768√1153𝑏20 𝜃
𝑢4 = − (−2𝑎20 − 2𝑏20 + 4𝛽 + 𝜃) 𝑓2 ,
5898240𝛽7
+969876044134880𝑏20 𝜃3 −28776579351136√1153𝑏20 𝜃3

where + 3030081181205925𝜃4 − 89269930142715√1153𝜃4 ,

(10)
2 2
𝑓1 = −213760𝑏20 𝛽 − 213760𝑏20 𝛽3 + 147392𝛽4 we have

2 Resultant [𝑓1 , 𝑓2 , 𝑏20 ] = 2.87343 × 1053 𝜃8 ≠0, (11)

− 323520𝑏20 𝛽𝜃 − 697600𝑏20 𝛽2 𝜃 + 84104𝛽3 𝜃
where Resultant[𝑔(𝑥), 𝑓(𝑥), 𝑥] denotes the resultant of
2 2
− 116000𝑏20 𝜃 − 682160𝑏20 𝛽𝜃2 − 554530𝛽2 𝜃2 𝑓(𝑥), 𝑔(𝑥) with respect to 𝑥. So 𝑓2 ≠0 when 𝑓1 = 0. Namely,
point (1, 1) is a fourth-order weak focus.
− 203000𝑏20 𝜃3 − 544505𝛽𝜃3 − 128125𝜃4 ;
We next study bifurcation of limit cycles of the perturbed
10 9 system of (4). When condition (9) holds, we can obtain
𝑓2 = 14520162123939840𝛽 + 77088521496444928𝛽 𝜃
󵄨󵄨𝐷 [𝑢 , 𝑎 ] 𝐷 [𝑢 , 𝛽] 𝐷 [𝑢 , 𝑏 ]󵄨󵄨
󵄨󵄨 1 20 󵄨󵄨
+146627049158313472𝛽8 𝜃2 +98147703728382592𝛽7𝜃3 󵄨 1 20 1
󵄨
𝐽 = 󵄨󵄨󵄨𝐷 [𝑢2 , 𝑎20 ] 𝐷 [𝑢2 , 𝛽] 𝐷 [𝑢2 , 𝑏20 ]󵄨󵄨󵄨 ≠0. (12)
󵄨󵄨 󵄨
󵄨󵄨𝐷 [𝑢3 , 𝑎20 ] 𝐷 [𝑢3 , 𝛽] 𝐷 [𝑢3 , 𝑏20 ]󵄨󵄨󵄨
−45397952769129120𝛽6 𝜃4 −116390817537863400𝛽5𝜃5
In fact,
− 79417641637292700𝛽4 𝜃6 −27193429543260550𝛽3 𝜃7
7001704448𝑖
2 8 9
𝐽= 11
− 5139731868326875𝛽 𝜃 − 582232780375000𝛽𝜃 234375(−75 + √1153) 𝜃7

− 38940972656250𝜃10 . × (688𝑏20 + 227𝜃 + 5√1153𝜃)

2

(8)
2 2
× (−17721692278080𝑏20 + 519304637376√1153𝑏20
While 𝑢1 = 𝑢2 = 𝑢3 = 0, 𝑢4 ≠0, the following theorem − 2991374746080𝑏20 𝜃 + 85097939808√1153𝑏20 𝜃
holds.
−2771947690805𝜃2 + 80900776387√1153𝜃2 ) .
Theorem 2. The origin of system (6) is a 4-order weak focus if
(13)
and only if
So Resultant[𝐽, 𝑓3 , 𝑏20 ] ≠ 0, namely, 𝐽 ≠ 0, when 𝑎20 =
1 (1/4)(4𝑏20 + 12𝛽 + 9𝜃), 344𝛽2 + 750𝛽𝜃 + 325𝜃2 = 0.
𝑎20 = (4𝑏20 + 12𝛽 + 9𝜃) , The statement mentioned above yields that the following
4 theorem holds.
(9)
344𝛽2 + 750𝛽𝜃 + 325𝜃2 = 0,
Theorem 3. If the origin of system (4) is a 4-order weak focus,
𝑓1 = 0. making a small perturbation to the coefficients of system (4),
4 Abstract and Applied Analysis

𝑑𝑤 1
then, for perturbed system (4), in a small neighborhood of = −𝑤 − (2 (−4𝑎20 + 4𝑖𝑏20 + (3 − 5𝑖) 𝛽 + (3 + 𝑖) 𝜃) 𝑤2
the origin, there exist exactly 5 small amplitude limit cycles 𝑑𝑇 8𝜃
enclosing the point (1, 1).
+ 2 (−8𝑖𝑎20 − 8𝑏20 + (6 + 10𝑖) 𝛽 + (1 + 3𝑖) 𝜃) 𝑧𝑤
With a similar method, we could discuss the bifurcation
− 2 (4𝑎20 − 4𝑖𝑏20 − (9 + 𝑖) 𝛽 − (4 + 2𝑖) 𝜃) 𝑧2
of limit cycles from point (2, 2). Through the transformations
𝑥̃ = 𝑥 − 2, 𝑦̃ = 𝑦 − 2, 𝜏 = 𝜃𝑡, we still denote 𝑥,
̃ 𝑦̃ by 𝑥, 𝑦 for + (−2𝑎20 − 2𝑏20 + 4 (1 − 4𝑖) 𝛽 + (1 − 𝑖) 𝜃) 𝑤3
convenience. Then point (1, 1) will be moved into point (0, 0)
of the new system, and the system will be changed into the + 2 ((−1 − 2𝑖) 𝑎20 + (1 − 2𝑖) 𝑏20
following system:
+ (1 + 4𝑖) 𝛽 + (1 + 𝑖) 𝜃) 𝑤2 𝑧
𝑑𝑥 𝑥 2
(7𝑥𝑦) (3𝑥2 𝑦)
= −𝑦− − + (2 (1 − 2𝑖) 𝑎20 + 2 (1 + 2𝑖) 𝑏20 − (1 + 𝑖) (4𝛽 + 𝜃)) 𝑤𝑧2
𝑑𝜏 2 2 2
(𝑥𝑦2 ) (2𝑎20 𝑥2 ) (𝑎20 𝑥3 ) −2 (−𝑎20 + 𝑏20 + 𝛽 + 𝜃) 𝑧3 ) .
+ 𝑦2 + − −
2 𝜃 𝜃 (16)
2 2
(4𝑎20 𝑥𝑦) (2𝑎20 𝑥 𝑦) (2𝑎20 𝑦 )
+ + − Direct computation can give the following theorem.
𝜃 𝜃 𝜃
Theorem 4. The first four singular point quantities at the
(𝑎20 𝑥𝑦2 ) (𝑥2 𝛽) 2
(6𝑥𝑦𝛽) (3𝑥 𝑦𝛽)
− + − − origin of system (16) are as follows:
𝜃 𝜃 𝜃 𝜃
2
(4𝑦 𝛽) (2𝑥𝑦2 𝛽) 𝑖
(14) 𝑢1 = (−2𝑎20 − 2𝑏20 + 4𝛽 + 𝜃) (8𝑎20 − 8𝑏20 − 3𝜃) ,
+ + , 4𝜃2
𝜃 𝜃
𝑖
2
(3𝑥𝑦) 𝑦2 (𝑥𝑦 ) 𝑦3 𝑢2 = − (−2𝑎20 − 2𝑏20 + 4𝛽 + 𝜃) (56𝛽2 + 57𝛽𝜃 + 10𝜃2 ) ,
𝑑𝑦 48𝜃3
=𝑥+ + + +
𝑑𝜏 2 2 2 2 𝑖
𝑢3 = (−2𝑎20 − 2𝑏20 + 4𝛽 + 𝜃) 𝑓3 ,
(2𝑏20 𝑥 )2
(4𝑏20 𝑥𝑦) (𝑏20 𝑥 𝑦)
2 6144𝜃5
− + −
𝜃 𝜃 𝜃 𝑖
𝑢4 = − (−2𝑎20 − 2𝑏20 + 4𝛽 + 𝜃) 𝑓4 ,
5898240𝛽7
(2𝑏20 𝑦2 ) (2𝑏20 𝑥𝑦2 ) (𝑏20 𝑦3 ) (17)
− + −
𝜃 𝜃 𝜃
2 2 3 where
(2𝑥𝑦𝛽) (3𝑦 𝛽) (𝑥𝑦 𝛽) (2𝑦 𝛽)
− + − + .
𝜃 𝜃 𝜃 𝜃 2 2
𝑓3 = −143360𝑏20 𝛽 + 286720𝑏20 𝛽3 − 206080𝛽4
Furthermore, by the transformations 2
− 168960𝑏20 𝛽𝜃 + 355840𝑏20 𝛽2 𝜃 − 264320𝛽3 𝜃
𝑧 = 𝑥 + 𝑖𝑦, 𝑤 = 𝑥 − 𝑖𝑦, 𝑇 = 𝑖𝜏, 𝑖 = √−1,
2 2
(15) − 40960𝑏20 𝜃 + 103040𝑏20 𝛽𝜃2 − 82336𝛽2 𝜃2

system (14) can be transformed into the following system: + 5120𝑏20 𝜃3 − 1939𝛽𝜃3 + 980𝜃4 ,
𝑑𝑧 1 𝑓4 = 2967207936000𝛽10 − 9714651627520𝛽9 𝜃
=𝑧+ (2 (−4𝑎20 − 4𝑖𝑏20 + (3 + 5𝑖) 𝛽 + (3 − 𝑖) 𝜃) 𝑧2
𝑑𝑇 8𝜃
− 65013770321920𝛽8 𝜃2 − 124569363533824𝛽7 𝜃3
+ 2 (8𝑖𝑎20 − 8𝑏20 + (6 − 10𝑖) 𝛽 + (1 − 3𝑖) 𝜃) 𝑧𝑤
− 124914242976768𝛽6 𝜃4 − 75962577303936𝛽5 𝜃5
− 2 (4𝑎20 + 4𝑖𝑏20 − (9 − 𝑖) 𝛽 − (4 − 2𝑖) 𝜃) 𝑤2
− 29760585599448𝛽4 𝜃6 − 7669544070125𝛽3 𝜃7
+ (−2𝑎20 − 2𝑏20 + 4 (1 + 4𝑖) 𝛽 + (1 + 𝑖) 𝜃) 𝑧3
− 1275738647714𝛽2 𝜃8 − 125233021760𝛽𝜃9
+ 2 ((−1 + 2𝑖) 𝑎20 + (1 + 2𝑖) 𝑏20
− 5502919200𝜃10 .
+ (1 − 4𝑖) 𝛽 + (1 − 𝑖) 𝜃) 𝑧2 𝑤
(18)
2
+ (2 (1 + 2𝑖) 𝑎20 + 2 (1 − 2𝑖) 𝑏20 − (1 − 𝑖) (4𝛽 + 𝜃)) 𝑧𝑤
While solving 𝑢1 = 𝑢2 = 𝑢3 = 0 and 𝑢4 ≠ 0, we can
−2 (−𝑎20 + 𝑏20 + 𝛽 + 𝜃) 𝑤3 ) ; immediately obtain the following theorem.
Abstract and Applied Analysis 5

(a) (b)

Figure 1: Distribution of limit cycles.

Theorem 5. The origin of system (16) is a 4-order weak focus −2𝑥2 𝑦𝛽 − 3𝑥2 𝜃 + 𝑥3 𝜃 + 𝑥𝑦𝜃 + 𝑥2 𝑦𝜃) ,
if and only if
𝑑𝑦 1
1 =𝑥− (2𝑏20 𝑥2 − 4𝑏20 𝑥𝑦 + 2𝑏20 𝑥2 𝑦 + 2𝑏20 𝑦2
𝑎20 = (8𝑏20 + 3𝜃) , 2 2
56𝛽 + 57𝛽𝜃 + 10𝜃 = 0, 𝑑𝜏 2𝛽
8 (19)
− 4𝑏20 𝑥𝑦2 + 2𝑏20 𝑦3 − 2𝑥𝛽 + 4𝑦2 𝛽 + 2𝑥𝑦2 𝛽
𝑓3 = 0.
−4𝑦3 𝛽 − 𝑥𝑦𝜃 + 3𝑦2 𝜃 − 𝑥𝑦2 𝜃 − 𝑦3 𝜃) .
Furthermore, the following theorem also holds for system
(14). (21)

Theorem 6. If the origin of system (14) is a 4-order weak focus,

By transformation
making a small perturbation to the coefficients of system (14),
then, for perturbed system (14), in a small neighborhood of
the origin, there exist exactly 5 small amplitude limit cycles 𝑢+V −𝑢 + V
𝑥=− , 𝑦=− , 𝜏 = (V − 1)2 𝑡,
enclosing the point (2, 2). 2 (V − 1) 2 (V − 1)
(22)
The results also yield that when point (1, 1) is a fourth-
order weak focus, point (2, 2) of system is a first-order weak
focus at the same time. Point (1, 1) is a first-order weak system (21) becomes
focus when point (2, 2) is a fourth-order weak focus. Namely,
by a simultaneously perturbation, five limit cycles could be
𝑑𝑢 1
bifurcated with two different distribution; see Figure 1. = −V − (−8𝑏20 𝑢2 + 4𝑏20 𝑢4 + 4𝑏20 𝑢2 V − 3𝑢4 𝛽
𝑑𝜏 4𝛽

3. Center Conditions of Two Positive + 4V𝛽 + 11𝑢2 V𝛽 − 12V2 𝛽 − 9𝑢2 V2 𝛽 + 9V3 𝛽

Equilibrium Points
−4𝑢2 𝛽 + 11𝑢2 V𝜃 − 2V2 𝜃 − 8𝑢2 V2 𝜃 + 3V3 𝜃),
If all Lyapunov constants are equal to zero, it will be a center
𝑑V
condition of a critical point. So Theorems 1 and 4 imply the = (𝑢 (−1 + V) (−4𝑏20 𝑢2 − 4𝛽 + 3𝑢2 𝛽
following theorem. 𝑑𝜏
−1
Theorem 7. If the origins of systems (4) and (14) are centers if +9V2 𝛽 − 6V𝜃 + 8V2 𝜃)) × (4𝛽)
and only if (23)

1
𝑎20 = (−2𝑏20 + 4𝛽 + 𝜃) . (20) which is symmetric with V-axis.
2
When 𝑎20 = (1/2)(−2𝑏20 + 4𝛽 + 𝜃), system (14) could be
Proof. When 𝑎20 = (1/2)(−2𝑏20 + 4𝛽 + 𝜃), system (4) could be rewritten as
rewritten as
𝑑𝑥 1 𝑑𝑥
= −𝑦 − (−2𝑏20 𝑥2 − 2𝑏20 𝑥3 + 4𝑏20 𝑥𝑦 + 4𝑏20 𝑥2 𝑦 𝑑𝜏
𝑑𝜏 2𝛽
1
− 2𝑏20 𝑦2 − 2𝑏20 𝑥𝑦2 − 4𝑥2 𝛽 + 4𝑥3 𝛽 = −𝑦 − (−4𝑏20 𝑥2 − 2𝑏20 𝑥3 + 8𝑏20 𝑥𝑦 + 4𝑏20 𝑥2 𝑦
2𝜃
6 Abstract and Applied Analysis

− 4𝑏20 𝑦2 − 2𝑏20 𝑥𝑦2 + 6𝑥2 𝛽 + 4𝑥3 𝛽 − 4𝑥𝑦𝛽 [3] Y. Ye and W. Ye, “Cubic Kolmogorov differential systems
with two limit cycles surrounding the same focus,” Annals of
−2𝑥2 𝑦𝛽 + 𝑥2 𝜃 + 𝑥3 𝜃 + 3𝑥𝑦𝜃 + 𝑥2 𝑦𝜃) , Differential Equations, vol. 2, pp. 201–207, 1985.
[4] E. Saez and I. Szanto, “Limit cycle s of a cubic Kolmogorov
𝑑𝑦 system,” Applied Mathematics Letters, vol. 9, pp. 15–18, 1996.
𝑑𝜏 [5] F. Li, “Integrability and bifurcations of limit cycles in a cubic
Kolmogorov system,” International Journal of Bifurcation and
1
=𝑥+ (−4𝑏20 𝑥2 + 8𝑏20 𝑥𝑦 − 2𝑏20 𝑥2 𝑦 − 4𝑏20 𝑦2 Chaos, vol. 23, no. 4, Article ID 1350061, 6 pages, 2013.
2𝜃 [6] C. Du and W. Huang, “Center-focus problem and limit cycles
+ 4𝑏20 𝑥𝑦2 − 2𝑏20 𝑦3 − 4𝑥𝑦𝛽 + 6𝑦2 𝛽 bifurcations for a class of cubic Kolmogorov model,” Nonlinear
Dynamics, vol. 72, no. 1-2, pp. 197–206, 2013.
−2𝑥𝑦2 𝛽 + 4𝑦3 𝛽 + 3𝑥𝑦𝜃 + 𝑦2 𝜃 + 𝑥𝑦2 𝜃 + 𝑦3 𝜃) . [7] C. Du, Q. Wang, and W. Huang, “Three-dimensional Hopf
bifurcation for a class of cubic Kolmogorov model,” Interna-
(24) tional Journal of Bifurcation and Chaos, vol. 24, no. 3, Article
ID 1450036, 2014.
Transformations
[8] C. Du, Y. Liu, and W. Huang, “Limit cycles bifurcations for
𝑢+V −𝑢 + V a class of Kolmogorov model in symmetrical vector field,”
𝑥=− , 𝑦=− , 𝜏 = (V − 1)2 𝑡,
2 (V − 1) 2 (V − 1) International Journal of Bifurcation and Chaos, vol. 24, no. 3,
(25) Article ID 1450040, 8 pages, 2014.
[9] Y. Liu and J. Li, “Theory of values of singular point in complex
bring system (24) into autonomous differential system,” Science China A, vol. 3, pp.
245–255, 1989.
𝑑𝑢
𝑑𝜏
1
= −V − (−16𝑏20 𝑢2 + 4𝑏20 𝑢4 + 12𝑏20 𝑢2 V
4𝜃
+ 10𝑢2 𝛽 − 3𝑢4 𝛽 − 15𝑢2 V𝛽 + 2V2 𝛽 + 7𝑢2 V2 𝛽

−V3 𝛽+2𝑢2 𝜃− 7𝑢2 V𝜃 − 4V2 𝜃 + 4𝑢2 V2 𝜃 + V3 𝜃),

𝑑V
= (𝑢 (−1 + V) (4𝑏20 𝑢2 − 3𝑢2 𝛽 − 12V𝛽
𝑑𝜏
+7V2 𝛽 + 4𝜃 − 10V𝜃 + 4V2 𝜃)) × (4𝜃)−1
(26)
which is symmetric with V-axis. So the origin is a center of
system (14).

Conflict of Interests
The authors declare that they have no conflict of interests.

Acknowledgments
This research is partially supported by the National Nature
Science Foundation of China (no. 11201211, 11201138, and
11261020) and the Scientific Research Fund of Hunan Provin-
cial Education Department (no. 12B034), and Hunan Provin-
cial Natural Science Foundation of China (no. 13JJ3106).

References
[1] X. S. Liu and Z. H. Zhuo, “Existence and uniqueness of limit
cycles of a cubic Kolmogorov differential system,” Biomathe-
matics, vol. 3, pp. 266–271, 2001.
[2] X. C. Huang and L. Zhu, “Limit cycles in a general Kolmogorov
model,” Nonlinear Analysis: Theory, Methods and Applications,
vol. 60, no. 8, pp. 1393–1414, 2005.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 819798, 14 pages
http://dx.doi.org/10.1155/2014/819798

Research Article
Limit Cycle Bifurcations by Perturbing a Compound
Loop with a Cusp and a Nilpotent Saddle

Huanhuan Tian and Maoan Han

Department of Mathematics, Shanghai Normal University, Shanghai 200234, China

Correspondence should be addressed to Maoan Han; mahan@shnu.edu.cn

Received 13 April 2014; Accepted 14 June 2014; Published 16 July 2014

Academic Editor: Tonghua Zhang

Copyright © 2014 H. Tian and M. Han. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

We study the expansions of the first order Melnikov functions for general near-Hamiltonian systems near a compound loop with
a cusp and a nilpotent saddle. We also obtain formulas for the first coefficients appearing in the expansions and then establish a
bifurcation theorem on the number of limit cycles. As an application example, we give a lower bound of the maximal number of
limit cycles for a polynomial system of Liénard type.

1. Introduction saddles, a hyperbolic saddle and a cusp of order one, or two

cusps of order one or two, the number of limit cycles of the
Consider a planar system of the form system (1) was studied in [5, 8, 9], respectively. In this paper,
we suppose that the unperturbed system (2) has a compound
𝑥̇= 𝐻𝑦 + 𝜖𝑝 (𝑥, 𝑦, 𝛿) , 𝑦̇= −𝐻𝑥 + 𝜖𝑞 (𝑥, 𝑦, 𝛿) , (1) loop consisting of a cusp 𝑆1 of order one, a nilpotent saddle
𝑆2 of order one, a homoclinic loop to 𝑆2 , and two heteroclinic
where 𝜖 is a small parameter and 𝐻(𝑥, 𝑦), 𝑝(𝑥, 𝑦, 𝛿), and orbits connecting 𝑆1 and 𝑆2 , as shown in Figure 1. We aim to
𝑞(𝑥, 𝑦, 𝛿) are 𝐶∞ functions in (𝑥, 𝑦) ∈ R2 and 𝛿 ∈ 𝐷 ⊂ R𝑚 study the number of limit cycles of (1) near the loop for 𝜖 ≠0
with 𝐷 bounded. For 𝜖 = 0, (1) becomes small.
𝑥̇= 𝐻𝑦 , 𝑦̇= −𝐻𝑥 , (2)
2. Main Results with Proof
which is a Hamiltonian system. As we know, the system (1) is
said to be a near-Hamiltonian system. For (1), the main task is Now consider the 𝐶∞ systems (1) and (2). Suppose that (2)
to study the number of limit cycles which are bifurcated from has a compound loop denoted by 𝐿 0 = 𝐿 1 ∪ 𝐿 2 ∪ 𝐿 3 ∪ {𝑆1 , 𝑆2 }
periodic orbits of the unperturbed system (2). On this aspect, and defined by equation 𝐻(𝑥, 𝑦) = 0, where 𝑆1 (𝑥1 , 𝑦1 ) is a
the first order Melnikov function of (1) plays an important cusp and 𝑆2 (𝑥2 , 𝑦2 ) is a nilpotent saddle both having order
role. We can use the expansions of it near Hamiltonian values one, 𝐿 1 , 𝐿 2 are heteroclinic orbits satisfying 𝜔(𝐿 1 ) = 𝛼(𝐿 2 ) =
corresponding to a center or an invariant loop to find its 𝑆2 and 𝜔(𝐿 2 ) = 𝛼(𝐿 1 ) = 𝑆1 , and 𝐿 3 is a homoclinic loop to
zeros and hence the number of limit cycles. See a survey 𝑆2 . Then, the level curves of 𝐻(𝑥, 𝑦) define two families of
article [1]. There have been many works on this topic. For the periodic orbits 𝐿 ℎ1 and 𝐿 ℎ2 for ℎ on one side of ℎ = 0 and a
study of general near-Hamiltonian systems, see [2–12]; and family of periodic orbits 𝐿 ℎ3 for ℎ on another side of ℎ = 0.
especially for the system (2) with the elliptic case, one can see For the definiteness, let both 𝐿 ℎ1 and 𝐿 ℎ2 exist for 0 < −ℎ ≪ 1
[13–17] and references therein. In [2–4], the number of limit and 𝐿 ℎ3 exist for 0 < ℎ ≪ 1. Thus, we have three Melnikov
cycles of the system (1) near a homoclinic loop with a cusp functions
of order one or two or a nilpotent saddle of order one (for
the definition of an order of a cusp or nilpotent saddle, see
𝑀𝑖 (ℎ, 𝛿) = ∮ 𝑞 𝑑𝑥 − 𝑝 𝑑𝑦, 𝑖 = 1, 2, 3. (3)
[5]) was studied. In the heteroclinic case with two hyperbolic 𝐿 ℎ𝑖
2 Abstract and Applied Analysis

S1
S2

Figure 1: Compound loop with a cusp and a nilpotent saddle.

L h3 where
U1 U2

S1 𝐼𝑖𝑗 (ℎ, 𝛿) = ∫ (𝑗) 𝑞 𝑑𝑥 − 𝑝 𝑑𝑦, 𝑖 ∈ {1, 2, 3} , 𝑗 ∈ {1, 2, 3, 4} .

S2 𝐿 ℎ𝑖
(8)
L h1 L h2

(a)
By [5], there exist two transformations of the form
U2 L(2)
h3

𝑇
L20 (𝑥, 𝑦) = 𝑇𝑖 (𝑢, V)𝑇 + 𝑆𝑖 , 𝑖 = 1, 2, (9)
S2

L30
L(3)
h3 where 𝑇𝑖 is a 2 × 2 matrix satisfying det 𝑇𝑖 = 1 such that (1)
(b) becomes
Figure 2
𝑢̇= 𝐻𝑖V + 𝜖𝑝𝑖 (𝑢, V, 𝛿) , V̇= −𝐻𝑖𝑢 + 𝜖𝑞𝑖 (𝑢, V, 𝛿) , (10)
Let 𝑈𝑖 denote a closed set with diameter 𝜖0 > 0 and with
center at 𝑆𝑖 , 𝑖 = 1, 2. See Figure 2(a). And further introduce
where
(𝑗)
𝐿 ℎ1 = 𝐿 ℎ1 ∩ 𝑈𝑗 , 𝑗 = 1, 2,

2 V2
𝐿(3)
ℎ1 = Cl. (Lℎ1 − ⋃𝐿 ℎ1 ) ,
(𝑗)
𝐿(1)
ℎ2 = 𝐿 ℎ2 ∩ 𝑈2 ,
𝐻1 (𝑢, V) = + ∑ ̃ℎ 𝑢𝑘 V𝑗 ,
𝑗=1
2 𝑘+𝑗≥3 𝑘𝑗

𝐿(2) (1)
ℎ2 = 𝐿 ℎ2 − 𝐿 ℎ2 , 𝐿(1)
ℎ3 = 𝐿 ℎ3 ∩ 𝑈1 ,
(4) 𝐻1 (𝑢, 𝜑1 (𝑢)) = ̃ℎ3 𝑢3 + 𝑂 (𝑢4 ) , ̃ℎ < 0,
3
(11)
𝐿 ℎ3 ∩ 𝑈2 = 𝐿(2) (3)
ℎ3 ∪ 𝐿 ℎ3 (as shown in Figure 2 (b)) , 𝐻1V (𝑢, 𝜑1 (𝑢)) = 0, 𝑝1 (𝑢, V, 𝛿) = ∑ 𝑎̃𝑖𝑗 𝑢𝑖 V𝑗 ,
𝑖+𝑗≥0
3
(𝑗)
𝐿(4)
ℎ3 = Cl. (𝐿 ℎ3 − ⋃𝐿 ℎ3 ) . 𝑞1 (𝑢, V, 𝛿) = ∑ ̃𝑏𝑖𝑗 𝑢𝑖 V𝑗 ,
𝑗=1 𝑖+𝑗≥0

Here the Cl. denotes the closure of a set. Then by (3) and (4), V2
for 𝜖0 sufficiently small we can write 𝐻2 (𝑢, V) = + ∑ ℎ 𝑢𝑘 V 𝑗 ,
2 𝑘+𝑗≥3 𝑘𝑗
𝑀1 (ℎ, 𝛿) = 𝐼11 (ℎ, 𝛿) + 𝐼12 (ℎ, 𝛿) + 𝐼13 (ℎ, 𝛿)
(5) 𝐻2 (𝑢, 𝜑2 (𝑢)) = ℎ4 𝑢4 + 𝑂 (𝑢5 ) , ℎ4 < 0,
for 0 ≤ −ℎ ≪ 1,

𝑀2 (ℎ, 𝛿) = 𝐼21 (ℎ, 𝛿) + 𝐼22 (ℎ, 𝛿) 𝐻2V (𝑢, 𝜑2 (𝑢)) = 0, (12)

(6)
for 0 ≤ −ℎ ≪ 1, 𝑝2 (𝑢, V, 𝛿) = ∑ 𝑎𝑖𝑗 𝑢𝑖 V𝑗 ,
𝑖+𝑗≥0
𝑀3 (ℎ, 𝛿) = 𝐼31 (ℎ, 𝛿) + 𝐼32 (ℎ, 𝛿) + 𝐼33 (ℎ, 𝛿) + 𝐼34 (ℎ, 𝛿)
(7) 𝑞2 (𝑢, V, 𝛿) = ∑ 𝑏𝑖𝑗 𝑢𝑖 V𝑗 ,
for 0 ≤ ℎ ≪ 1, 𝑖+𝑗≥0
Abstract and Applied Analysis 3

for (𝑢, V) near (0, 0). Note that 𝑞𝑑𝑥 − 𝑝𝑑𝑦 = 𝑞1 𝑑𝑢 − 𝑝1 𝑑V for for 0 < ℎ ≪ 1, where 𝑁𝑖1 (ℎ, 𝛿) ∈ 𝐶𝜔 at ℎ = 0 with 𝑁𝑖1 (0, 𝛿) =
(𝑥, 𝑦) near 𝑆1 and 𝑞𝑑𝑥 − 𝑝𝑑𝑦 = 𝑞2 𝑑𝑢 − 𝑝2 𝑑V for (𝑥, 𝑦) near 𝑂(𝜖0 ), 𝑖 = 1, 3, and
𝑆2 . Then we have
𝑐1 (𝑆1 , 𝛿) = 2√2 ̃ℎ3−1/3 (̃
𝑎10 + ̃𝑏01 ) ,

𝐼𝑖1 (ℎ, 𝛿) = ∫ (1) 𝑞1 𝑑𝑢 − 𝑝1 𝑑V, 𝑖 = 1, 3, (13) 𝑐3 (𝑆1 , 𝛿) = 2√2 ̃ℎ3−5/3

̃
𝐿 ℎ𝑖

× [̃ℎ3 (2̃
𝑎20 + ̃𝑏11 )
𝐼21 (ℎ, 𝛿) = ∫ (1) 𝑞2 𝑑𝑢 − 𝑝2 𝑑V, (14)
𝐿ℎ2
1 ̃2
+ (ℎ − 2̃ℎ4 − 3̃ℎ3 ̃ℎ12 ) (̃
𝑎10 + ̃𝑏01 )] ,
3 21
𝐼12 (ℎ, 𝛿) = ∫ (2) 𝑞2 𝑑𝑢 − 𝑝2 𝑑V,
𝐿ℎ1 𝑐4 (𝑆1 , 𝛿) = 9𝜇1−1 𝛼̃01
(15)
𝐼3𝑗 (ℎ, 𝛿) = ∫ (𝑗) 𝑞2 𝑑𝑢 − 𝑝2 𝑑V, 𝑗 = 2, 3, − 2𝜇1−7 [(20𝜇23 − 20𝜇1 𝜇2 𝜇3 + 4𝜇12 𝜇4 ) 𝛼̃00
𝐿ℎ3
+ (4𝜇12 𝜇3 − 10𝜇1 𝜇22 ) 𝛼̃10
̃ (1) denote the image of 𝐿(1) under 𝑇1 and 𝐿(1) , 𝐿(2) ,
where 𝐿 ℎ𝑖
(𝑗)
ℎ𝑖 ℎ2 ℎ1 + 4𝜇12 𝜇2 𝛼̃20 − 𝜇13 𝛼̃30 ] ,
(𝑗)
and 𝐿ℎ3denote the image of 𝐿(1)
ℎ2 , 𝐿(2)
ℎ1 ,
and 𝐿 ℎ3
under 𝑇2 , (19)
respectively. Then, by using [3, 4] we can obtain the following
two lemmas, respectively. where
1
Lemma 1. Consider system (10) with 𝑖 = 1 and suppose 𝜇1 = ̃ℎ31/3 , 𝜇2 = ̃ℎ3−2/3 ̃ℎ4 ,
∗ ∗ 3
(11), (13) hold. Then there are constants 𝐵00 , 𝐵00 , 𝐵10 , 𝐵10
1
satisfying 𝜇3 = ̃ℎ3−5/3 (3̃ℎ3 ̃ℎ5 − ̃ℎ42 ) ,
9
3 1 𝑑V 3 1 ̃ −8/3
𝐵00 = ∫ = × 2.4286 ⋅ ⋅ ⋅ > 0, 𝜇4 = ℎ (27̃ℎ32 ̃ℎ6 − 18̃ℎ3 ̃ℎ4 ̃ℎ5 + 5̃ℎ43 ) ,
5 0 √V (1 − V3 ) 5 81 3
̃ℎ = ̃ℎ , ̃ℎ = − 1 ̃ℎ2 + ̃ℎ ,
3 30 4 40
∗ 3 1
𝑑V 3 2 21
𝐵00 =− ∫ = − × 4.2065 ⋅ ⋅ ⋅ < 0,
5 −∞ √1 − V3 5 ̃ℎ = ̃ℎ ̃ℎ2 − ̃ℎ ̃ℎ + ̃ℎ ,
5 12 21 21 31 50

3 1
V3/2 𝑑V ̃ℎ = −2̃ℎ2 ̃ℎ2 − ̃ℎ ̃ℎ3 + ̃ℎ2 ̃ℎ
𝐵10 = − (∫ − 2) > 0, 6 12 21 03 21 21 22
7 0 √1 − V3 (1 + √1 − V3 )
1 2 ̃ ̃
∗ + 2̃ℎ12 ̃ℎ21 ̃ℎ31 − ̃ℎ31 − ℎ21 ℎ41 + ̃ℎ60 ,
𝐵10 2

3 −1
V 𝑑V 1
V3/2 𝑑V 𝑎10 + ̃𝑏01 ) ,
𝛼̃00 = 2√2 (̃
= (∫ −∫ − 2) < 0
7 1 √1 − V 3 0 √1 + V3 (1 + √1 + V3 )
𝛼̃10 = 2√2 [−̃ℎ12 (̃
𝑎10 + ̃𝑏01 ) + 2̃
𝑎20 + ̃𝑏11 ] ,
(16)
𝛼̃20
such that 3 2
𝑎10 + ̃𝑏01 ) (3̃ℎ03 ̃ℎ21 − ̃ℎ22 + ̃ℎ12
= 2√2 [(̃ ) − 2̃ℎ12 𝑎̃20
2
𝐼11 (ℎ, 𝛿) = 𝐵00 𝑐1 (𝑆1 , 𝛿) |ℎ|5/6 + 𝐵10 𝑐3 (𝑆1 , 𝛿) |ℎ|7/6
− ̃ℎ12̃𝑏11 + 3̃
𝑎30 + ̃𝑏21 − 𝑎̃11 ̃ℎ21 − 2̃𝑏02 ̃ℎ21 ] ,
1
− 𝐵00 𝑐4 (𝑆1 , 𝛿) |ℎ|11/6 + 𝑂 (ℎ2 ) + 𝑁11 (ℎ, 𝛿)
11 𝛼̃30
(17)
𝑎10 + ̃𝑏01 ) (3̃ℎ13 ̃ℎ21 + 3̃ℎ03 ̃ℎ31 + 3̃ℎ12 ̃ℎ22
= 2√2 [(̃
for 0 < −ℎ ≪ 1,
5 3 ̃
∗
− 15̃ℎ12 ̃ℎ03 ̃ℎ21 − ̃ℎ12 − ℎ32 )
𝐼31 (ℎ, 𝛿) = 𝐵00 𝑐1 (𝑆1 , 𝛿) ℎ5/6 + 𝐵10
∗
𝑐3 (𝑆1 , 𝛿) ℎ7/6 2

1 ∗ 𝑎11 + 6̃𝑏02 ) ̃ℎ12 ̃ℎ21 − 2 (̃𝑏12 + 𝑎̃21 ) ̃ℎ21

+ (3̃
+ 𝐵 𝑐 (𝑆 , 𝛿) ℎ11/6 + 𝑂 (ℎ2 ) + 𝑁31 (ℎ, 𝛿)
11 00 4 1
(18) 𝑎11 + 2̃𝑏02 ) ̃ℎ31 + 4̃
− (̃ 𝑎40 + ̃𝑏31
4 Abstract and Applied Analysis

+ (3̃𝑏11 + 6̃
𝑎20 ) ̃ℎ03 ̃ℎ21 − (2̃
𝑎20 + ̃𝑏11 ) ̃ℎ22 𝑐6 (𝑆2 , 𝛿)

3 1
𝑎20 + ̃𝑏11 ) ̃ℎ12
+ (3̃ 2
𝑎30 + ̃𝑏21 ) ̃ℎ12 ] ,
− (3̃ =− [−6𝑑1−3 𝑑2 𝛼01 + 3𝑑1−2 𝛼11
2 32
2 − 𝑑1−11 (504𝑑1 𝑑23 𝑑3 − 168𝑑12 𝑑2 𝑑32
𝛼̃01 = 2√2 [ 𝑎̃12 + 2̃𝑏03 − 2̃ℎ03 𝑎̃11
3
+ 42𝑑13 𝑑3 𝑑4 − 168𝑑12 𝑑22 𝑑4
− 4̃ℎ03̃𝑏02 + (̃
𝑎10 + ̃𝑏01 ) (5̃ℎ03
2
− 2̃ℎ04 )] .
(20) + 42𝑑13 𝑑2 𝑑5 −252𝑑25 − 6𝑑14 𝑑6 ) 𝛼00

Lemma 2. Consider system (10) with 𝑖 = 2 and suppose (12), + 𝑑1−10 (168𝑑1 𝑑22 𝑑3 − 126𝑑24
(15) hold. Then we have
−42𝑑12 𝑑2 𝑑4 − 21𝑑12 𝑑32 + 6𝑑13 𝑑5 ) 𝛼10
𝐼12 (ℎ, 𝛿) = 𝑐1 (𝑆2 , 𝛿) |ℎ|3/4 − 𝑐2 (𝑆2 , 𝛿) ℎ ln |ℎ|
− 𝑑1−9 (42𝑑1 𝑑2 𝑑3 − 56𝑑23 − 6𝑑12 𝑑4 ) 𝛼20
+ 𝑐4 (𝑆2 , 𝛿) |ℎ|5/4 + 𝑐5 (𝑆2 , 𝛿) |ℎ|7/4 (21)
+ 𝑑1−8 (6𝑑1 𝑑3 − 21𝑑22 ) 𝛼30
2 2
− 𝑐6 (𝑆2 , 𝛿) ℎ ln |ℎ| + 𝑂 (ℎ ) + 𝑁12 (ℎ, 𝛿)
+6𝑑1−7 𝑑2 𝛼40 − 𝑑1−6 𝛼50 ] ,
for 0 < −ℎ ≪ 1,
𝑐1∗ (𝑆2 , 𝛿) = −𝐷1 𝑐1 (𝑆2 , 𝛿) , 𝑐3∗ (𝑆2 , 𝛿) = −𝐷2 𝑐4 (𝑆2 , 𝛿) ,
1 1
𝐼32 (ℎ, 𝛿) = 𝑐1∗ (𝑆2 , 𝛿) ℎ3/4 + 𝑐3∗ (𝑆2 , 𝛿) ℎ5/4 𝑐4∗ (𝑆2 , 𝛿) = 𝐷1 𝑐5 (𝑆2 , 𝛿) ,
2 2 (23)
1
+ 𝑐4∗ (𝑆2 , 𝛿) ℎ7/4 + 𝑂 (ℎ2 ) + 𝑁32 (ℎ, 𝛿) ,
2
(22)
1 1 ̃0 |, 𝐷2 = 2|𝐴1 |/|𝐴
̃2 |, 𝐴0 , 𝐴
̃0 , 𝐴1 , and
𝐼33 (ℎ, 𝛿) = 𝑐1∗ (𝑆2 , 𝛿) ℎ3/4 + 𝑐3∗ (𝑆2 , 𝛿) ℎ5/4 where 𝐷1 = 2|𝐴0 |/|𝐴
2 2 ̃
𝐴 2 are constants, given by
1
+ 𝑐4∗ (𝑆2 , 𝛿) ℎ7/4 + 𝑂 (ℎ2 ) + 𝑁33 (ℎ, 𝛿)
2
2 ∞ 𝑑V
for 0 < ℎ ≪ 1, where 𝑁𝑖𝑗 (ℎ, 𝛿) ∈ 𝐶𝜔 at ℎ = 0 with 𝑁𝑖𝑗 (0, 𝛿) = 𝐴0 = ∫ ≈ 1.236049785 > 0,
𝑂(𝜖0 ), (𝑖, 𝑗) ∈ {(1, 2), (3, 2), (3, 3)}, and 3 0 √1 + V 4
1
󵄨 ̃ 󵄨󵄨 󵄨󵄨 󵄨󵄨−1/4 ̃0 = − 2 ∫
𝐴
𝑑V
𝑐1 (𝑆2 , 𝛿) = − 2√2 󵄨󵄨󵄨󵄨𝐴 0 󵄨󵄨󵄨 󵄨󵄨󵄨ℎ4 󵄨󵄨󵄨 (𝑎10 + 𝑏01 ) , 3 0 √1 − V 4

√2 󵄨󵄨 󵄨󵄨−1/2 √2𝜋3/2
𝑐2 (𝑆2 , 𝛿) = − 󵄨󵄨ℎ 󵄨󵄨 (2𝑎20 + 𝑏11 ) + 𝑂 (𝑎10 + 𝑏01 ) , =− ≈ −0.8740191847 < 0,
4 󵄨 4󵄨 6[Γ(3/4)]2
󵄨 ̃ 󵄨󵄨 21 2 3 󵄨 󵄨−11/4 2 ∞ 𝑑V
𝑐4 (𝑆2 , 𝛿) = 󵄨󵄨󵄨󵄨𝐴 2 󵄨󵄨󵄨 [( ℎ − ℎ ℎ ) 󵄨󵄨󵄨ℎ 󵄨󵄨󵄨 𝛼00 𝐴1 = − ∫
32 5 4 4 6 󵄨 4 󵄨 5 0 √1 + V [V2 + √1 + V4 ]
4

3 󵄨 󵄨−7/4 󵄨 󵄨−3/4
+ 󵄨󵄨󵄨󵄨ℎ4 󵄨󵄨󵄨󵄨 ℎ5 𝛼10 + 󵄨󵄨󵄨󵄨ℎ4 󵄨󵄨󵄨󵄨 𝛼20 ] , ≈ −0.3388852337 < 0,
4
1 󵄨󵄨 ̃ 󵄨󵄨
󵄨󵄨𝐴 󵄨󵄨 [6𝑑1−1 𝛼01 ̃2 = 2 [1 − ∫
1
V2 𝑑V
𝑐5 (𝑆2 , 𝛿) = 𝐴 ]
7 󵄨 0󵄨 5 0 √1 − V4 (1 + √1 − V4 )

+ 𝑑1−9 (105𝑑1 𝑑22 𝑑3 − 30𝑑12 𝑑2 𝑑4

≈ 0.2396280470 > 0,
−15𝑑12 𝑑32 + 5𝑑13 𝑑5 − 70𝑑24 ) 𝛼00 󵄨 󵄨1/4 1 󵄨󵄨 󵄨󵄨−3/4
𝑑1 = 󵄨󵄨󵄨󵄨ℎ4 󵄨󵄨󵄨󵄨 , 𝑑2 = − 󵄨󵄨ℎ 󵄨󵄨 ℎ5 ,
4 󵄨 4󵄨
+ 𝑑1−8 (35𝑑23 − 30𝑑1 𝑑2 𝑑3 + 5𝑑12 𝑑4 ) 𝛼10
1 󵄨󵄨 󵄨󵄨−7/4 2
𝑑3 = 󵄨󵄨ℎ 󵄨󵄨 (8ℎ4 ℎ6 − 3ℎ5 ) ,
− 𝑑1−7 (15𝑑22 − 5𝑑1 𝑑3 ) 𝛼20 32 󵄨 4 󵄨
1 󵄨󵄨 󵄨󵄨−11/4 3 2
+5𝑑1−6 𝑑2 𝛼30 − 𝑑1−5 𝛼40 ] , 𝑑4 = − 󵄨󵄨ℎ4 󵄨󵄨 (7ℎ5 + 32ℎ4 ℎ7 − 24ℎ4 ℎ5 ℎ6 ) ,
128 󵄨 󵄨
Abstract and Applied Analysis 5

1 󵄨󵄨 󵄨󵄨−15/4 3 2 2
𝑑5 = 󵄨󵄨ℎ 󵄨󵄨 (512ℎ4 ℎ8 − 192ℎ4 ℎ6 𝛼50 = 2√2 [𝑛2 (𝑎10 + 𝑏01 ) + 𝑛1 (2𝑎20 + 𝑏11 )
2048 󵄨 4 󵄨
2 2 4
−384ℎ4 ℎ5 ℎ7 + 336ℎ4 ℎ5 ℎ6 − 77ℎ5 ) , 5 3
− ( ℎ12 − 3ℎ12 ℎ22 + 15ℎ12 ℎ03 ℎ21 + ℎ32
1 󵄨󵄨 󵄨󵄨−19/4 4 2 2 2
𝑑6 = − 󵄨󵄨ℎ4 󵄨󵄨 (2048ℎ4 ℎ9 + 1344ℎ4 ℎ5 ℎ6
8192 󵄨 󵄨
3 3 − 3ℎ13 ℎ21 − 3ℎ03 ℎ31 )
− 1536ℎ4 ℎ5 ℎ8 − 1536ℎ4 ℎ6 ℎ7
2 2 3
+ 1344ℎ4 ℎ5 ℎ7 − 1232ℎ4 ℎ5 ℎ6 × (3𝑎30 + 𝑏21 − 𝑎11 ℎ21 − 2𝑏02 ℎ21 )
5
+231ℎ5 ) ; 3 2
+ ( ℎ12 + 3ℎ03 ℎ21 − ℎ22 )
2
𝛼00 = 2√2 (𝑎10 + 𝑏01 ) ,
× (4𝑎40 + 𝑏31 − 2𝑎21 ℎ21
𝛼10 = 2√2 [−ℎ12 (𝑎10 + 𝑏01 ) + 2𝑎20 + 𝑏11 ] ,
3 2 − 2𝑏12 ℎ21 + 2𝑎11 ℎ12 ℎ21
𝛼20 = 2√2 [(𝑎10 + 𝑏01 ) (3ℎ03 ℎ21 − ℎ22 + ℎ12 )
2
− 2ℎ12 𝑎20 − ℎ12 𝑏11 + 3𝑎30 + 𝑏21 +4𝑏02 ℎ12 ℎ21 − 𝑎11 ℎ31 − 2𝑏02 ℎ31 )

− 𝑎11 ℎ21 − 2𝑏02 ℎ21 ] , − ℎ12 𝑛3 + 𝑛4 ] ,

𝛼30 = 2√2 [(𝑎10 + 𝑏01 ) 2

𝛼01 = 2√2 [ 𝑎12 + 2𝑏03 − 2ℎ03 𝑎11 − 4ℎ03 𝑏02
3
× (3ℎ13 ℎ21 + 3ℎ03 ℎ31 + 3ℎ12 ℎ22 2
+ (𝑎10 + 𝑏01 ) (5ℎ03 − 2ℎ04 )] ,
5 3
− 15ℎ12 ℎ03 ℎ21 − ℎ12 − ℎ32 )
2 𝛼11 = 2√2 [(𝑎10 + 𝑏01 ) (10ℎ04 ℎ12 − 2ℎ14
+ (3𝑎11 + 6𝑏02 ) ℎ12 ℎ21 − 2 (𝑏12 + 𝑎21 ) ℎ21
2
+10ℎ03 ℎ13 − 35ℎ12 ℎ03 )
− (𝑎11 + 2𝑏02 ) ℎ31 + 4𝑎40 + 𝑏31
+ (𝑎11 + 2𝑏02 ) (−2ℎ13 + 10ℎ12 ℎ03 )
+ (3𝑏11 + 6𝑎20 ) ℎ03 ℎ21
3 2
− 2ℎ12 (𝑎12 + 3𝑏03 )
− (2𝑎20 + 𝑏11 ) ℎ22 + (3𝑎20 + 𝑏11 ) ℎ12
2
2
− (3𝑎30 + 𝑏21 ) ℎ12 ] , + (2𝑎20 + 𝑏11 ) (5ℎ03 − 2ℎ04 )

4
𝛼40 = 2√2 [𝑛1 (𝑎10 + 𝑏01 ) − 4ℎ03 (𝑏12 + 𝑎21 ) + 𝑎22 + 2𝑏13 ] ;
3
5 3 1 2 2
− ( ℎ12 − 3ℎ12 ℎ22 + 15ℎ12 ℎ03 ℎ21 + ℎ32 ℎ4 = − ℎ21 + ℎ40 , ℎ5 = ℎ12 ℎ21 − ℎ21 ℎ31 + ℎ50 ,
2 2
2 2 3 2
− 3ℎ13 ℎ21 − 3ℎ03 ℎ31 ) (2𝑎20 + 𝑏11 ) ℎ6 = −2ℎ12 ℎ21 − ℎ03 ℎ21 + ℎ21 ℎ22
3 2 1 2
+ ( ℎ12 + 3ℎ03 ℎ21 − ℎ22 ) + 2ℎ12 ℎ21 ℎ31 − ℎ31 − ℎ21 ℎ41 + ℎ60 ,
2 2
× (3𝑎30 + 𝑏21 − 𝑎11 ℎ21 − 2𝑏02 ℎ21 ) 2 3
ℎ7 = −4ℎ22 ℎ21 ℎ12 + ℎ70 − ℎ51 ℎ21 − ℎ13 ℎ21
− ℎ12 (4𝑎40 + 𝑏31 − 2𝑎21 ℎ21 2 3 2
+ ℎ32 ℎ21 + 4ℎ12 ℎ21 − ℎ31 ℎ41 + ℎ12 ℎ31
2

− 2𝑏12 ℎ21 + 2𝑎11 ℎ12 ℎ21 + 4𝑏02 ℎ12 ℎ21 3 2

+ 2ℎ22 ℎ21 ℎ31 + 6ℎ03 ℎ21 ℎ12 − 3ℎ03 ℎ21 ℎ31
−𝑎11 ℎ31 − 2𝑏02 ℎ31 ) + 𝑛3 ] , 2
− 4ℎ12 ℎ21 ℎ31 + 2ℎ12 ℎ21 ℎ41 ,
6 Abstract and Applied Analysis

3 4 2 1 2 2
ℎ8 = ℎ80 − ℎ23 ℎ21 − ℎ61 ℎ21 + ℎ04 ℎ21 + ℎ42 ℎ21 𝑛2 = − (−24ℎ22 ℎ32 + 60ℎ12 ℎ22 + 60ℎ12 ℎ32
8
2 2 4 2 2 2
− 2ℎ12 ℎ31 − 8ℎ12 ℎ21 − 2ℎ22 ℎ21 3 5
− 140ℎ12 ℎ22 + 63ℎ12
9 2 4 2 1 2
− ℎ03 ℎ21 − ℎ31 ℎ51 + ℎ22 ℎ31 − ℎ41 2 3
2 2 − 180ℎ12 ℎ03 ℎ31 + 780ℎ12 ℎ03 ℎ21
3 2 2 2
+ 8ℎ12 ℎ21 ℎ31 − 4ℎ12 ℎ21 ℎ41 + 12ℎ22 ℎ21 ℎ12
+ 72ℎ22 ℎ03 ℎ31 − 504ℎ12 ℎ22 ℎ03 ℎ21
3 2 3
− 24ℎ03 ℎ21 ℎ12 + 2ℎ12 ℎ31 ℎ41 + 6ℎ12 ℎ21 ℎ13 2 2 2
− 216ℎ03 ℎ21 ℎ31 + 972ℎ12 ℎ03 ℎ21
2 3
− 4ℎ12 ℎ21 ℎ32 + 2ℎ12 ℎ21 ℎ51 + 6ℎ22 ℎ21 ℎ03
+ 72ℎ03 ℎ21 ℎ32 + 72ℎ22 ℎ13 ℎ21
2
+ 2ℎ22 ℎ21 ℎ41 + 2ℎ32 ℎ21 ℎ31 − 3ℎ03 ℎ21 ℎ31
2 2
2 2 − 180ℎ12 ℎ13 ℎ21 − 216ℎ03 ℎ21 ℎ13 ) ,
− 3ℎ03 ℎ21 ℎ41 − 3ℎ13 ℎ21 ℎ31
2
− 8ℎ12 ℎ22 ℎ21 ℎ31 + 18ℎ12 ℎ03 ℎ21 ℎ31 , 𝑛3 = 5𝑎50 + 𝑏41 − 2𝑏22 ℎ21 − 𝑎11 ℎ41

4 2 3 2
ℎ9 = ℎ12 [ℎ21 (45ℎ03 − 8ℎ04 ) + 6ℎ21 (ℎ23 − 8ℎ03 ℎ22 ) + 2𝑎11 ℎ22 ℎ21 − 3𝑎11 ℎ03 ℎ21 + 2𝑎11 ℎ12 ℎ31
2 2 2
+ 2ℎ21 (9ℎ13 ℎ31 − 2ℎ42 + 6ℎ22 + 9ℎ03 ℎ41 ) − 4𝑎11 ℎ12 ℎ21 − 2𝑏02 ℎ41 + 4𝑏02 ℎ22 ℎ21

2 2 2
+ 2ℎ21 (9ℎ03 ℎ31 − 4ℎ41 ℎ22 − 4ℎ31 ℎ32 ) − 6𝑏02 ℎ03 ℎ21 + 4𝑏02 ℎ12 ℎ31 − 8𝑏02 ℎ12 ℎ21

2 2 − 2𝑏12 ℎ31 + 4𝑏12 ℎ12 ℎ21 − 2𝑎21 ℎ31

−4ℎ22 ℎ31 + ℎ41 + 2ℎ31 ℎ51 + 2ℎ21 ℎ61 ]
2 2
2 2 + 4𝑎21 ℎ12 ℎ21 − 3𝑎31 ℎ21 + 3𝑏03 ℎ21 + 𝑎12 ℎ21 ,
+ 4ℎ12 [ℎ21 (3ℎ21 ℎ32 − 6ℎ21 ℎ13 − ℎ51
𝑛4 = 4𝑏12 ℎ22 ℎ21 + 4𝑏02 ℎ32 ℎ21 + 6𝑎31 ℎ12 ℎ21
+ 6ℎ31 ℎ22 − 18ℎ31 ℎ03 ℎ21 ) − ℎ31 ℎ41 ] 2
+ 2𝑎11 ℎ12 ℎ41 + 4𝑏12 ℎ12 ℎ31 − 4𝑎11 ℎ12 ℎ31
3 3 2 2
+ 4ℎ12 (20ℎ03 ℎ21 + 2ℎ41 ℎ21 − 8ℎ22 ℎ21 + ℎ31 ) 2 2
+ 4𝑏02 ℎ12 ℎ41 − 12𝑏03 ℎ21 ℎ12 − 4𝑎12 ℎ21 ℎ12
4 2
+ ℎ21 (2ℎ42 ℎ31 − 16ℎ12 ℎ31 − 4ℎ22 ℎ31 + 2ℎ22 ℎ51 2
+ 2𝑎11 ℎ22 ℎ31 + 6𝑏03 ℎ21 ℎ31 − 3𝑎11 ℎ13 ℎ21
2
+2ℎ32 ℎ41 − 3ℎ13 ℎ31 − ℎ71 − 6ℎ03 ℎ31 ℎ41 ) 2
+ 6𝑎60 + 𝑏51 − 6𝑏12 ℎ03 ℎ21 + 2𝑎11 ℎ32 ℎ21
2 5
+ ℎ21 (16ℎ12 − 4ℎ32 ℎ22 − 3ℎ31 ℎ23 + 4𝑏02 ℎ22 ℎ31 + 4𝑎21 ℎ22 ℎ21 + 2𝑎12 ℎ21 ℎ31
2 3
− 3ℎ03 ℎ51 + ℎ52 + 18ℎ22 ℎ31 ℎ03 ) + 4𝑏22 ℎ12 ℎ21 − 8𝑏12 ℎ12 ℎ21 + 16𝑏02 ℎ12 ℎ21
3 2 2
3 3 + 8𝑎11 ℎ12 ℎ21 − 6𝑏02 ℎ13 ℎ21 − 8𝑎21 ℎ12 ℎ21
+ 2ℎ22 ℎ31 ℎ41 + 6ℎ32 ℎ21 ℎ03 + 4ℎ04 ℎ21 ℎ31
2
2 3 − 8𝑏02 ℎ12 ℎ31 − 4𝑎41 ℎ21 − 2𝑏32 ℎ21
− ℎ41 ℎ51 − ℎ31 ℎ61 + ℎ32 ℎ31 − ℎ03 ℎ31
2 2
4 2 3 4 − 6𝑎21 ℎ03 ℎ21 + 4𝑎21 ℎ12 ℎ31 + 3𝑏13 ℎ21
− 9ℎ03 ℎ21 ℎ13 − 18ℎ31 ℎ03 ℎ21 + ℎ14 ℎ21
2 2
3 3 + 2𝑎22 ℎ21 − 6𝑎11 ℎ03 ℎ21 ℎ31 + 18𝑎11 ℎ03 ℎ21 ℎ12
− ℎ33 ℎ21 + 6ℎ22 ℎ21 ℎ13 + ℎ90 ;
2
1 2 4 − 8𝑎11 ℎ12 ℎ22 ℎ21 − 12𝑏02 ℎ03 ℎ21 ℎ31 + 36𝑏02 ℎ03 ℎ21 ℎ12
𝑛1 = (12ℎ22 + 35ℎ12 + 24ℎ12 ℎ32
8
− 16𝑏02 ℎ12 ℎ22 ℎ21 − 2𝑏22 ℎ31 − 𝑎11 ℎ51 − 2𝑏12 ℎ41
2 2
− 60ℎ12 ℎ22 − 72ℎ12 ℎ03 ℎ31 + 324ℎ12 ℎ03 ℎ21
− 2𝑎21 ℎ41 − 2𝑏02 ℎ51 − 3𝑎31 ℎ31 .
2 2
+108ℎ03 ℎ21 − 72ℎ12 ℎ13 ℎ21 − 72ℎ22 ℎ03 ℎ21 ) , (24)
Abstract and Applied Analysis 7

2
For convenience, let
𝑐0 (𝛿) = 𝑀1 (0, 𝛿) = ∮ 𝑞 𝑑𝑥 − 𝑝 𝑑𝑦 = ∑ ∫ 𝑞 𝑑𝑥 − 𝑝 𝑑𝑦,
𝐿∗ 𝑖=1 𝐿 𝑖
∗
𝐿 = 𝐿 1 ∪ 𝐿 2, 𝐿∗1 ∗
= 𝐿 ∩ 𝑈1 ,
𝑐0 (𝛿) = 𝑀2 (0, 𝛿) = ∮ 𝑞 𝑑𝑥 − 𝑝 𝑑𝑦,
2 𝐿3
𝐿∗2 ∗
= 𝐿 ∩ 𝑈2 , 𝐿∗3 = Cl. (𝐿 − ∗
⋃𝐿∗𝑖 ) ,
𝑖=1 𝑐̃0 (𝛿) = 𝑀3 (0, 𝛿) = 𝑐0 (𝛿) + 𝑐0 (𝛿) ,
(30)
𝐿10 = 𝐿∗1 , (25)
𝑐4 (𝛿) = ∫ (𝑝𝑥 + 𝑞𝑦 − 𝜎) 𝑑𝑡
𝐿 0 ∩ 𝑈2 = 𝐿20 ∪ 𝐿30 (as shown in Figure 2 (b)) , 𝐿∗1

3
𝐿40 = Cl. (𝐿 0 − ⋃𝐿 0 ) .
𝑗 + ∫ [𝑝𝑥 + 𝑞𝑦 − 𝜂0 − 𝜂1 (𝑥 − 𝑥2 )] 𝑑𝑡
𝐿∗2
𝑗=1

+ ∫ (𝑝𝑥 + 𝑞𝑦 ) 𝑑𝑡 + 𝑡2 𝑐1 (𝛿) + 𝑡3 𝑐2 (𝛿) + 𝑡4 𝑐3 (𝛿) ,

𝐿∗3
Theorem 3. Assume that system (1) has a compound loop 𝐿 0
as stated before. Then, the functions 𝑀𝑖 (ℎ, 𝛿) given in (3) at
ℎ = 0 have the following expansions: 𝑐3 (𝛿) = ∮ [𝑝𝑥 + 𝑞𝑦 − 𝜂0 − 𝜂1 (𝑥 − 𝑥2 )] 𝑑𝑡
𝐿3

+ 𝑡0 𝑐1 (𝛿) + 𝑡1 𝑐3 (𝛿) ,
𝑀1 (ℎ, 𝛿) = 𝑐0 (𝛿) + 𝑐1 (𝛿) |ℎ|3/4 + 𝐵00 𝑐2 (𝛿) |ℎ|5/6
𝑐̃2 (𝛿) = ∫ (𝑝𝑥 + 𝑞𝑦 − 𝜎) 𝑑𝑡 + ∫ (𝑝𝑥 + 𝑞𝑦 − 𝜂0 ) 𝑑𝑡
− 𝑐3 (𝛿) ℎ ln |ℎ| + 𝑐4 (𝛿) ℎ + 𝐵10 𝑐5 (𝛿) |ℎ|7/6 𝐿∗1 𝐿20 ∪𝐿30

+ 𝑐6 (𝛿) |ℎ|5/4 + 𝑐7 (𝛿) |ℎ|7/4 + ∫ (𝑝𝑥 + 𝑞𝑦 ) 𝑑𝑡 + 𝑡5 𝑐1 (𝛿) + 𝑡6 𝑐2 (𝛿) ,

𝐿40
1
− 𝐵00 𝑐8 (𝛿) |ℎ|11/6 − 𝑐9 (𝛿) ℎ2 ln |ℎ| + 𝑂 (ℎ2 ) , (31)
11
where 𝜎 = (𝑝𝑥 + 𝑞𝑦 )|(𝑥1 ,𝑦1 ) , 𝜂0 = (𝑝𝑥 + 𝑞𝑦 )|(𝑥2 ,𝑦2 ) , 𝜂1 = (𝑝𝑥𝑥 +
(26) 𝑞𝑦𝑥 )|(𝑥2 ,𝑦2 ) . In particular,

𝑀2 (ℎ, 𝛿) = 𝑐0 (𝛿) + 𝑐1 (𝛿) |ℎ|3/4 + 𝑐3 (𝛿) ℎ ln |ℎ| 2

𝑐4 (𝛿) = ∑ ∫ (𝑝𝑥 + 𝑞𝑦 ) 𝑑𝑡, 𝑐3 (𝛿) = ∮ (𝑝𝑥 + 𝑞𝑦 ) 𝑑𝑡,
+ 𝑐3 (𝛿) ℎ + 𝑐6 (𝛿) |ℎ|5/4 + 𝑐7 (𝛿) |ℎ|7/4 (27) 𝑖=1 𝐿 𝑖 𝐿3

+ 𝑐9 (𝛿) ℎ2 ln |ℎ| + 𝑂 (ℎ2 ) 3

𝑐̃2 (𝛿) = ∑ ∫ (𝑝𝑥 + 𝑞𝑦 ) 𝑑𝑡
𝑖=1 𝐿 𝑖
for 0 < −ℎ ≪ 1, and (32)

if 𝑐1 (𝛿) = 𝑐2 (𝛿) = 𝑐3 (𝛿) = 0. Here, 𝑡𝑖 , 𝑖 = 0, 1, . . . , 6, are

𝑀3 (ℎ, 𝛿) = 𝑐̃0 (𝛿) − 𝐷1 𝑐1 (𝛿) ℎ3/4 + 𝐵00
∗
𝑐2 (𝛿) ℎ5/6 ∗
constants and 𝐵00 , 𝐵00 ∗
, 𝐵10 , 𝐵10 are given in Lemma 1.
∗
+ 𝑐̃2 (𝛿) ℎ + 𝐵10 𝑐5 (𝛿) ℎ7/6 − 𝐷2 𝑐6 (𝛿) ℎ5/4 Proof. First, by (6), (10) with 𝑖 = 2, (12), (14), (29), and
1 ∗ Theorem 2.2 in [4], we directly obtain (27) with 𝑐0 (𝛿), 𝑐3 (𝛿)
+ 𝐷1 𝑐7 (𝛿) ℎ7/4 + 𝐵 𝑐 (𝛿) ℎ11/6 + 𝑂 (ℎ2 ) given by (30) and (31), respectively. Then we study the
11 00 8 expansions of 𝑀1 and 𝑀3 .
(28)
By (5), (7), (29), and Lemmas 1 and 2, we have

for 0 < ℎ ≪ 1, where

𝑀1 (ℎ, 𝛿) = 𝑐1 (𝛿) |ℎ|3/4 + 𝐵00 𝑐2 (𝛿) |ℎ|5/6 − 𝑐3 (𝛿) ℎ ln |ℎ|

𝑐1 (𝛿) = 𝑐1 (𝑆2 , 𝛿) , c2 (𝛿) = 𝑐1 (𝑆1 , 𝛿) , + 𝐵10 𝑐5 (𝛿) |ℎ|7/6 + 𝑐6 (𝛿) |ℎ|5/4

𝑐3 (𝛿) = 𝑐2 (𝑆2 , 𝛿) , 𝑐5 (𝛿) = 𝑐3 (𝑆1 , 𝛿) , 1

+ 𝑐7 (𝛿) |ℎ|7/4 − 𝐵 𝑐 (𝛿) |ℎ|11/6
(29) 11 00 8
𝑐6 (𝛿) = 𝑐4 (𝑆2 , 𝛿) , 𝑐7 (𝛿) = 𝑐5 (𝑆2 , 𝛿) ,
− 𝑐9 (𝛿) ℎ2 ln |ℎ| + 𝑂 (ℎ2 ) + 𝑁 (ℎ, 𝛿)
𝑐8 (𝛿) = 𝑐4 (𝑆1 , 𝛿) , 𝑐9 (𝛿) = 𝑐6 (𝑆2 , 𝛿) , (33)
8 Abstract and Applied Analysis

for 0 < −ℎ ≪ 1, and Then by Lemma 3.1.2 in [5], we have

𝑀3 (ℎ, 𝛿) = −𝐷1 𝑐1 (𝛿) ℎ3/4 + 𝐵00

∗
𝑐2 (𝛿) ℎ5/6 + 𝐵10
∗
𝑐5 (𝛿) ℎ7/6 𝑐4 (𝛿) = 𝑁ℎ (0, 𝛿)

− 𝐷2 𝑐6 (𝛿) ℎ5/4 + 𝐷1 𝑐7 (𝛿) ℎ7/4 = lim− [∮

3
(𝑝𝑥 + 𝑞𝑦 ) 𝑑𝑡 + 𝑐1 (𝛿) |ℎ|−1/4
ℎ→0 𝐿 ℎ1 4
1 ∗
+ 𝐵 𝑐 (𝛿) ℎ11/6 + 𝑂 (ℎ2 ) + 𝑁∗ (ℎ, 𝛿)
11 00 8 5
(34) + 𝐵00 𝑐2 (𝛿) |ℎ|−1/6 + 𝑐3 (𝛿) (ln |ℎ| + 1)] ,
6
for 0 < ℎ ≪ 1, where 𝑐̃2 (𝛿) = 𝑁ℎ∗ (0, 𝛿)
𝑁 (ℎ, 𝛿) = 𝑁11 (ℎ, 𝛿) + 𝑁12 (ℎ, 𝛿) + 𝐼13 (ℎ, 𝛿) ,
= lim+ [∮ (𝑝𝑥 + 𝑞𝑦 ) 𝑑𝑡
ℎ→0
𝑁∗ (ℎ, 𝛿) = 𝑁31 (ℎ, 𝛿) + 𝑁32 (ℎ, 𝛿) + 𝑁33 (ℎ, 𝛿) + 𝐼34 (ℎ, 𝛿) . 𝐿 ℎ3

(35) 3 5 ∗
+ 𝐷1 𝑐1 (𝛿) ℎ−1/4 − 𝐵00 𝑐2 (𝛿) ℎ−1/6 ] .
4 6
Let
(38)
2
𝑁 (ℎ, 𝛿) = 𝑐0 (𝛿) + 𝑐4 (𝛿) ℎ + 𝑂 (ℎ ) ,
(36) It is easy to see that
𝑁∗ (ℎ, 𝛿) = 𝑐̃0 (𝛿) + 𝑐̃2 (𝛿) ℎ + 𝑂 (ℎ2 ) .
3
∮ (𝑝𝑥 + 𝑞𝑦 ) 𝑑𝑡 = ∑ ∫ (𝑖) (𝑝𝑥 + 𝑞𝑦 ) 𝑑𝑡
It follows further that 𝐿 ℎ1 𝑖=1 𝐿 ℎ1

𝑐0 (𝛿) = 𝑁11 (0, 𝛿) + 𝑁12 (0, 𝛿) + 𝐼13 (0, 𝛿)

= ∫ (1) (𝑝𝑥 + 𝑞𝑦 − 𝜎) 𝑑𝑡
= lim [𝑁11 (0, 𝛿) + 𝑁12 (0, 𝛿) + 𝐼13 (0, 𝛿)] 𝐿 ℎ1
𝜖0 → 0

+ ∫ (2) [𝑝𝑥 + 𝑞𝑦 − 𝜂0
= lim 𝐼13 (0, 𝛿) = ∮ 𝑞 𝑑𝑥 − 𝑝 𝑑𝑦 𝐿 ℎ1
𝜖0 → 0 𝐿 1 ∪𝐿 2
−𝜂1 (𝑥 − 𝑥2 )] 𝑑𝑡
2
= ∑ ∫ 𝑞 𝑑𝑥 − 𝑝 𝑑𝑦 = 𝑀1 (0, 𝛿) ,
𝑖=1 𝐿 𝑖 + ∫ (3) (𝑝𝑥 + 𝑞𝑦 ) 𝑑𝑡 + 𝜎 ∫ (1) 𝑑𝑡
𝐿 ℎ1 𝐿 ℎ1
𝑐̃0 (𝛿) = 𝑁31 (0, 𝛿) + 𝑁32 (0, 𝛿) + 𝑁33 (0, 𝛿) + 𝐼34 (0, 𝛿) (39)
+ 𝜂0 ∫ (2) 𝑑𝑡 + 𝜂1 ∫ (2) (𝑥 − 𝑥2 ) 𝑑𝑡,
= lim [𝑁31 (0, 𝛿) + 𝑁32 (0, 𝛿) 𝐿 ℎ1 𝐿 ℎ1
𝜖0 → 0

+𝑁33 (0, 𝛿) + 𝐼34 (0, 𝛿)] ∮ (𝑝𝑥 + 𝑞𝑦 ) 𝑑𝑡

𝐿 ℎ3

= lim 𝐼34 (0, 𝛿) = ∮ 𝑞 𝑑𝑥 − 𝑝 𝑑𝑦 4

𝜖0 → 0 𝐿0 = ∑ ∫ (𝑖) (𝑝𝑥 + 𝑞𝑦 ) 𝑑𝑡 = ∫ (1) (𝑝𝑥 + 𝑞𝑦 − 𝜎) 𝑑𝑡
𝑖=1 𝐿 ℎ3 𝐿 ℎ3
3
= ∑ ∫ 𝑞 𝑑𝑥 − 𝑝 𝑑𝑦 = 𝑀3 (0, 𝛿) ,
𝑖=1 𝐿 𝑖
+ ∫ (2) (𝑝𝑥 + 𝑞𝑦 − 𝜂0 ) 𝑑𝑡 + ∫ (4) (𝑝𝑥 + 𝑞𝑦 ) 𝑑𝑡
𝐿 ℎ3 ∪𝐿(3)
ℎ3
𝐿 ℎ3

𝑐4 (𝛿) + 𝑂 (ℎ) = 𝑁ℎ (ℎ, 𝛿) = 𝑀1ℎ (ℎ, 𝛿)

+ 𝜎 ∫ (1) 𝑑𝑡 + 𝜂0 ∫ (2) 𝑑𝑡.
3 5 𝐿 ℎ3 𝐿 ℎ3 ∪𝐿(3)
+ 𝑐1 (𝛿) |ℎ|−1/4 + 𝐵00 𝑐2 (𝛿) |ℎ|−1/6 ℎ3

4 6
Noting ̃ℎ3 < 0, ℎ4 < 0, by (19), (23), and (29), we have
+ 𝑐3 (𝛿) (ln |ℎ| + 1) + 𝑂 (|ℎ|1/6 ) ,

𝑐̃2 (𝛿) + 𝑂 (ℎ) = 𝑁ℎ∗ (ℎ, 𝛿) 𝑐4 (𝛿) = lim− [∫ (1) (𝑝𝑥 + 𝑞𝑦 − 𝜎) 𝑑𝑡

ℎ→0 𝐿 ℎ1
3
= 𝑀3ℎ (ℎ, 𝛿) + 𝐷1 𝑐1 (𝛿) ℎ−1/4
4 + ∫ (2) [𝑝𝑥 + 𝑞𝑦 − 𝜂0 − 𝜂1 (𝑥 − 𝑥2 )] 𝑑𝑡
𝐿 ℎ1
5 ∗
− 𝐵00 𝑐2 (𝛿) ℎ−1/6 + 𝑂 (ℎ1/6 ) .
6
(37) + ∫ (3) (𝑝𝑥 + 𝑞𝑦 ) 𝑑𝑡]
𝐿 ℎ1
Abstract and Applied Analysis 9

5√2 󵄨󵄨̃ 󵄨󵄨−1/3 −1/6 lim ∫ [𝑝𝑥 + 𝑞𝑦 − 𝜂0 − 𝜂1 (𝑥 − 𝑥2 )] 𝑑𝑡

+ lim− 𝜎 [∫ (1) 𝑑𝑡 − 𝐵 󵄨󵄨ℎ 󵄨󵄨 |ℎ| ]
ℎ→0 𝐿 ℎ1 3 00 󵄨 3 󵄨 ℎ → 0− 𝐿(2)
ℎ1

3√2 ̃ 󵄨󵄨 󵄨󵄨−1/4 −1/4 = ∫ [𝑝𝑥 + 𝑞𝑦 − 𝜂0 − 𝜂1 (𝑥 − 𝑥2 )] 𝑑𝑡,

+ lim− 𝜂0 [∫ (2) 𝑑𝑡 + 𝐴 0 󵄨󵄨󵄨ℎ4 󵄨󵄨󵄨 |ℎ| 𝐿∗2
ℎ→0 𝐿 ℎ1 2
lim ∫ (𝑝𝑥 + 𝑞𝑦 − 𝜂0 ) 𝑑𝑡 = ∫ (𝑝𝑥 + 𝑞𝑦 − 𝜂0 ) 𝑑𝑡,
+ 𝑂 (1) (ln |ℎ| + 1)] ℎ → 0+ 𝐿(𝑖)
ℎ3
𝐿𝑖0

𝑖 = 2, 3.
+ lim− 𝜂1 [∫ (2) (𝑥 − 𝑥2 ) 𝑑𝑡 (42)
ℎ→0 𝐿 ℎ1
Here 𝑇1∗ , 𝑇𝑖 , 𝑖 = 1, 2, 3, are constants. Therefore, we can obtain
√2 󵄨 󵄨−1/2
− 󵄨󵄨󵄨󵄨ℎ4 󵄨󵄨󵄨󵄨 (ln |ℎ| + 1)] , (31) and (32). Thus we have proved Theorem 3.
4
In the following we use Theorem 3 to study the problem
of limit cycle bifurcation near 𝐿 0 . For the sake of convenience,
𝑐̃2 (𝛿) = lim+ [∫ (1) (𝑝𝑥 + 𝑞𝑦 − 𝜎) 𝑑𝑡 we say that (1) has a distribution (𝑖, 𝑗)+𝑘 of 𝑖+𝑗+𝑘 limit cycles
ℎ→0 𝐿 ℎ3
if there are 𝑖 and 𝑗 limit cycles near the inside of 𝐿∗ and 𝐿 3 ,
+ ∫ (2) (𝑝𝑥 + 𝑞𝑦 − 𝜂0 ) 𝑑𝑡 respectively, and 𝑘 limit cycles near the outside of 𝐿 0 . Then
𝐿 ℎ3 ∪𝐿(3)
ℎ3
we can prove the following theorem.

Theorem 4. Assume that system (1) has a compound loop

+ ∫ (4) (𝑝𝑥 + 𝑞𝑦 ) 𝑑𝑡] 𝐿 0 as stated before and (26)–(28) hold. Define 𝑐4∗ (𝛿) =
𝐿 ℎ3
𝑐4 (𝛿)|𝑐1 =𝑐2 =𝑐3 =0 , 𝑐∗3 (𝛿) = 𝑐3 (𝛿)|𝑐1 =𝑐3 =0 , 𝑐̃2∗ (𝛿) = 𝑐4∗ (𝛿) +
5√2 ∗ 󵄨󵄨̃ 󵄨󵄨−1/3 −1/6 𝑐∗3 (𝛿), 𝑐3∗ (𝛿) = 𝑐3 (𝛿)|𝑐1 =0 . Let there exist 𝛿0 ∈ R𝑚 , such that
+ lim+ 𝜎 [∫ (1) 𝑑𝑡 + 𝐵 󵄨󵄨ℎ 󵄨󵄨 ℎ ]
ℎ→0 𝐿 ℎ3 3 00 󵄨 3 󵄨 (𝑐0 , 𝑐0 , 𝑐1 , 𝑐2 , 𝑐3∗ , 𝑐4∗ , 𝑐∗3 )(𝛿0 ) = (0, 0, 0, 0, 0, 0, 0).

󵄨 󵄨−1/4 (1) If 𝑐𝑙 (𝛿0 ) ≠0, 𝑐𝑗 (𝛿0 ) = 0, 𝑗 = 5, . . . , 𝑙 − 1, and

+ lim+ 𝜂0 [∫ (2) 𝑑𝑡 − 3√2 𝐴0 󵄨󵄨󵄨󵄨ℎ4 󵄨󵄨󵄨󵄨 ℎ−1/4 ] .
ℎ→0 𝐿 ℎ3 ∪𝐿(3)
ℎ3
𝜕 (𝑐0 , 𝑐0 , 𝑐1 , 𝑐2 , 𝑐3∗ , 𝑐4∗ , 𝑐∗3 , 𝑐5 , . . . , 𝑐𝑙−1 )
(40) rank = 𝑙 + 2, (43)
𝜕 (𝛿1 , 𝛿2 , 𝛿3 , 𝛿4 , 𝛿5 , 𝛿6 , 𝛿7 , 𝛿8 , . . . , 𝛿𝑚 )
Then by the proof of (3.13) in [3], the following equations
hold: then (1) can have 2𝑙 + 1 limit cycles near 𝐿 0 for some
5√2 󵄨󵄨̃ 󵄨󵄨−1/3 −1/6 (𝜖, 𝛿) near (0, 𝛿0 ), where 𝑙 = 6, 7 or 9.
lim− [∫ (1) 𝑑𝑡 − 𝐵 󵄨󵄨ℎ 󵄨󵄨 |ℎ| ] = 𝑇0 ,
ℎ→0 𝐿 ℎ1 3 00 󵄨 3 󵄨 (2) If 𝑐5 (𝛿0 )𝑐6 (𝛿0 ) < 0, and

5√2 ∗ 󵄨󵄨̃ 󵄨󵄨−1/3 −1/6 𝜕 (𝑐0 , 𝑐0 , 𝑐1 , 𝑐2 , 𝑐3∗ , 𝑐4∗ , 𝑐∗3 )

lim+ [∫ (1) 𝑑𝑡 + 𝐵 󵄨󵄨ℎ 󵄨󵄨 ℎ ] = 𝑇0∗ , rank = 7, (44)
ℎ→0 𝐿 ℎ3 3 00 󵄨 3 󵄨 𝜕 (𝛿1 , 𝛿2 , 𝛿3 , 𝛿4 , . . . , 𝛿𝑚 )

lim ∫ (𝑝𝑥 + 𝑞𝑦 − 𝜎) 𝑑𝑡 = lim+ ∫ (1) (𝑝𝑥 + 𝑞𝑦 − 𝜎) 𝑑𝑡 then (1) can have 11 limit cycles near 𝐿 0 for some (𝜖, 𝛿)
ℎ → 0− 𝐿(1) ℎ→0
ℎ1
𝐿 ℎ3 near (0, 𝛿0 ).
(3) If 𝑐8 (𝛿0 )𝑐9 (𝛿0 ) > 0, 𝑐𝑗 (𝛿0 ) = 0, 𝑗 = 5, 6, 7, and
= ∫ (𝑝𝑥 + 𝑞𝑦 − 𝜎) 𝑑𝑡.
𝐿∗1
(41) 𝜕 (𝑐0 , 𝑐0 , 𝑐1 , 𝑐2 , 𝑐3∗ , 𝑐4∗ , 𝑐∗3 , 𝑐5 , 𝑐6 , 𝑐7 )
rank = 10, (45)
𝜕 (𝛿1 , 𝛿2 , 𝛿3 , 𝛿4 , 𝛿5 , 𝛿6 , 𝛿7 , 𝛿8 , . . . , 𝛿𝑚 )
Here, 𝑇0 and 𝑇0∗
are constants. By a similar argument used in
Theorems 2.2 and 2.4 in [4], one can obtain
then (1) can have 18 limit cycles near 𝐿 0 for some (𝜖, 𝛿)
3√2 ̃ 󵄨󵄨 󵄨󵄨−1/4 −1/4 near (0, 𝛿0 ).
lim− [∫ (2) 𝑑𝑡 + 𝐴 0 󵄨󵄨󵄨ℎ4 󵄨󵄨󵄨 |ℎ| + 𝑂 (1) (ln |ℎ| + 1)]
ℎ→0 𝐿 ℎ1 2
Proof. (1) Because of the similarity in the proof, we only
= 𝑇1 , prove the conclusion for 𝑙 = 9 and omit the rest. By our
assumptions, there exists 𝛿0 ∈ R𝑚 such that 𝑐0 (𝛿0 ) =
√2 󵄨󵄨 󵄨󵄨−1/2 𝑐∗3 (𝛿0 ) = 𝑐3∗ (𝛿0 ) = 𝑐4∗ (𝛿0 ) = 0, 𝑐𝑗 (𝛿0 ) = 0, 𝑗 =
lim− [∫ (2) (𝑥 − 𝑥2 ) 𝑑𝑡 − 󵄨󵄨ℎ 󵄨󵄨 (ln |ℎ| + 1)] = 𝑇1∗ ,
ℎ→0 𝐿 ℎ1 4 󵄨 4󵄨 0, 1, 2, 5, 6, 7, 8, 𝑐9 (𝛿0 ) ≠0, and

3√2 󵄨󵄨 󵄨󵄨−1/4 −1/4 𝜕 (𝑐0 , 𝑐0 , 𝑐1 , 𝑐2 , 𝑐3∗ , 𝑐4∗ , 𝑐∗3 , 𝑐5 , . . . , 𝑐8 )

lim+ [∫ (𝑖) 𝑑𝑡 − 𝐴 󵄨󵄨ℎ 󵄨󵄨 ℎ ] = 𝑇𝑖 , 𝑖 = 2, 3, rank = 11. (46)
ℎ→0 𝐿 ℎ3 2 0󵄨 4󵄨 𝜕 (𝛿1 , 𝛿2 , 𝛿3 , 𝛿4 , 𝛿5 , 𝛿6 , 𝛿7 , 𝛿8 , . . . , 𝛿𝑚 )
10 Abstract and Applied Analysis

By the implicit function theorem, we can take 𝑐0 , 𝑐∗3 , 𝑐3∗ , if 𝑐4∗ 𝑐5 > 0, 𝑐∗3 𝑐6 > 0, 𝑐̃2∗ 𝑐5 > 0, and
𝑐4∗ , 𝑐𝑗 , 𝑗 = 0, 1, 2, 5, 6, 7, 8 as free parameters varying near
zero. Obviously, for these parameters varying near zero we 𝑐5 𝑀1 < 0, 𝑐6 𝑀2 < 0 for 0 < −ℎ ≪ 1,
have |𝑐9 | ≥ |(1/2)𝑐9 (𝛿0 )| > 0. In the following we proceed the (48)
process by 9 steps. 𝑐5 𝑀3 < 0 for 0 < ℎ ≪ 1

Step 1. Fix (𝑐0 , 𝑐0 , 𝑐1 , 𝑐2 , 𝑐3∗ , 𝑐4∗ , 𝑐∗3 , 𝑐5 , 𝑐6 , 𝑐7 ) = (0, 0, 0, 0, 0, 0, 0, if 𝑐4∗ 𝑐5 > 0, 𝑐∗3 𝑐6 > 0, 𝑐̃2∗ 𝑐5 < 0. Thus, 𝑀1 , 𝑀2 , and 𝑀3 each
0, 0, 0) and vary 𝑐8 near 0. has one more zero in the first case and 𝑀1 and 𝑀2 each has a
First, for 𝑐8 = 0, we have by (26) 𝑐9 𝑀1 > 0 for 0 < −ℎ ≪ 1. new zero in the second case. And the zeros got in above steps
Let 0 < |𝑐8 | ≪ 1. Then 𝑐9 𝑀1 < 0 for 0 < −ℎ ≪ 1 if still exist. Hence, for 0 < |𝑐∗3 , 𝑐4∗ | ≪ |𝑐5 | ≪ |𝑐6 | ≪ |𝑐7 | ≪
𝑐8 𝑐9 > 0. Thus, 𝑀1 has a zero. Hence, for 0 < |𝑐8 | ≪ 1, |𝑐8 | ≪ 1,
(1) the condition 𝑐8 𝑐9 > 0 implies a distribution (1, 0) + 0
of one limit cycle. (5i) the conditions 𝑐∗3 𝑐4∗ < 0, 𝑐∗3 𝑐̃2∗ < 0, 𝑐4∗ 𝑐5 > 0, 𝑐5 𝑐6 <
0, 𝑐6 𝑐7 < 0, 𝑐7 𝑐8 > 0, 𝑐8 𝑐9 > 0 imply a distribution
(5, 3) + 3 of 11 limit cycles, and
Step 2. Fix (𝑐0 , 𝑐0 , 𝑐1 , 𝑐2 , 𝑐3∗ , 𝑐4∗ , 𝑐∗3 , 𝑐5 , 𝑐6 ) = (0, 0, 0, 0, 0, 0, 0,
0, 0), 𝑐8 𝑐9 > 0 and vary 𝑐7 near 0. (5ii) the conditions 𝑐∗3 𝑐4∗ < 0, 𝑐∗3 𝑐̃2∗ > 0, 𝑐4∗ 𝑐5 > 0, 𝑐5 𝑐6 <
First, for 𝑐7 = 0, we have by (26), (27), and (28) 𝑐8 𝑀1 < 0, 𝑐6 𝑐7 < 0, 𝑐7 𝑐8 > 0, 𝑐8 𝑐9 > 0 imply a distribution
0, 𝑐9 𝑀2 < 0 for 0 < −ℎ ≪ 1, and 𝑐8 𝑀3 < 0 for 0 < ℎ ≪ 1. (5, 3) + 2 of 10 limit cycles.
Let 0 < |𝑐7 | ≪ |𝑐8 |. Then 𝑐8 𝑀1 > 0, 𝑐9 𝑀2 > 0 for 0 <
−ℎ ≪ 1, and 𝑐8 𝑀3 > 0 for 0 < ℎ ≪ 1 if 𝑐7 𝑐8 > 0, 𝑐7 𝑐9 > 0.
Step 6. Fix (𝑐0 , 𝑐0 , 𝑐1 , 𝑐2 ) = (0, 0, 0, 0), 𝑐4∗ 𝑐∗3 < 0 with 𝑐̃2∗ ≠0 and
Thus, 𝑀1 , 𝑀2 , and 𝑀3 each gets a zero and the zero of 𝑀1
vary 𝑐3∗ near 0.
got in Step 1 still exists. Hence, for 0 < |𝑐7 | ≪ |𝑐8 | ≪ 1,
First, for 𝑐3∗ = 0, we have by (26) and (27) 𝑐4∗ 𝑀1 < 0 and
(2) the conditions 𝑐7 𝑐8 > 0, 𝑐8 𝑐9 > 0 imply a distribution ∗
𝑐3 𝑀2 < 0 for 0 < −ℎ ≪ 1.
(2, 1) + 1 of 4 limit cycles.
Let 0 < |𝑐3∗ | ≪ |𝑐4∗ , 𝑐∗3 |. Then 𝑐4∗ 𝑀1 > 0 and 𝑐∗3 𝑀2 > 0 for
0 < −ℎ ≪ 1, if 𝑐3∗ 𝑐4∗ < 0, 𝑐∗3 𝑐3∗ > 0. Thus, 𝑀1 and 𝑀2 each gets
Step 3. Fix (𝑐0 , 𝑐0 , 𝑐1 , 𝑐2 , 𝑐3∗ , 𝑐4∗ , 𝑐∗3 , 𝑐5 ) = (0, 0, 0, 0, 0, 0, 0, 0), a new zero and the zeros got in above steps still exist. Hence,
𝑐7 ≠0 and vary 𝑐6 near 0. for 0 < |𝑐3∗ | ≪ |𝑐4∗ , 𝑐∗3 | ≪ |𝑐5 | ≪ |𝑐6 | ≪ |𝑐7 | ≪ |𝑐8 | ≪ 1,
First, for 𝑐6 = 0, we have by (26), (27), and (28) 𝑐7 𝑀1 >
0, 𝑐7 𝑀2 > 0 for 0 < −ℎ ≪ 1, and 𝑐7 𝑀3 > 0 for 0 < ℎ ≪ 1. (6i) the conditions 𝑐3∗ 𝑐4∗ < 0, 𝑐∗3 𝑐4∗ < 0, 𝑐∗3 𝑐̃2∗ < 0, 𝑐4∗ 𝑐5 >
Let 0 < |𝑐6 | ≪ |𝑐7 |. Then 𝑐7 𝑀1 < 0, 𝑐7 𝑀2 < 0 for 0 < 0, 𝑐5 𝑐6 < 0, 𝑐6 𝑐7 < 0, 𝑐7 𝑐8 > 0, 𝑐8 𝑐9 > 0 imply a
−ℎ ≪ 1, and 𝑐7 𝑀3 > 0 for 0 < ℎ ≪ 1 if 𝑐6 𝑐7 < 0. Thus, 𝑀1 distribution (6, 4) + 3 of 13 limit cycles,
and 𝑀2 each gets a new zero and the zeros got in above steps (6ii) the conditions 𝑐3∗ 𝑐4∗ < 0, 𝑐∗3 𝑐4∗ < 0, 𝑐∗3 𝑐̃2∗ > 0, 𝑐4∗ 𝑐5 >
still exist. Hence, for 0 < |𝑐6 | ≪ |𝑐7 | ≪ |𝑐8 | ≪ 1, 0, 𝑐5 𝑐6 < 0, 𝑐6 𝑐7 < 0, 𝑐7 𝑐8 > 0, 𝑐8 𝑐9 > 0 imply a
(3) the conditions 𝑐6 𝑐7 < 0, 𝑐7 𝑐8 > 0, 𝑐8 𝑐9 > 0 imply a distribution (6, 4) + 2 of 12 limit cycles.
distribution (3, 2) + 1 of 6 limit cycles.

Step 4. Fix (𝑐0 , 𝑐0 , 𝑐1 , 𝑐2 , 𝑐3∗ , 𝑐4∗ , 𝑐∗3 ) = (0, 0, 0, 0, 0, 0, 0), 𝑐6 ≠0 Step 7. Fix (𝑐0 , 𝑐0 , 𝑐1 ) = (0, 0, 0), 𝑐3∗ 𝑐̃2∗ ≠0 and vary 𝑐2 near 0.
and vary 𝑐5 near 0. First, for 𝑐2 = 0, we have by (26) and (28) 𝑐3∗ 𝑀1 < 0 for
First, for 𝑐5 = 0, we have by (26) and (28) 𝑐6 𝑀1 > 0 for 0 < −ℎ ≪ 1 and 𝑐̃2∗ 𝑀3 > 0 for 0 < ℎ ≪ 1.
0 < −ℎ ≪ 1, and 𝑐6 𝑀3 < 0 for 0 < ℎ ≪ 1. Let 0 < |𝑐2 | ≪ |𝑐3∗ | ≪ |𝑐4∗ , 𝑐∗3 |. Then
Let 0 < |𝑐5 | ≪ |𝑐6 |. Then 𝑐6 𝑀1 < 0 for 0 < −ℎ ≪ 1, and
𝑐6 𝑀3 > 0 for 0 < ℎ ≪ 1 if 𝑐5 𝑐6 < 0. Thus, 𝑀1 and 𝑀3 each has 𝑐3∗ 𝑀1 > 0 for 0 < −ℎ ≪ 1,
a new zero and the zeros got in above steps still exist. Hence, (49)
𝑐̃2∗ 𝑀3 < 0 for 0 < ℎ ≪ 1
for 0 < |𝑐5 | ≪ |𝑐6 | ≪ |𝑐7 | ≪ |𝑐8 | ≪ 1,
(4) the conditions 𝑐5 𝑐6 < 0, 𝑐6 𝑐7 < 0, 𝑐7 𝑐8 > 0, 𝑐8 𝑐9 > 0
imply a distribution (4, 2) + 2 of 8 limit cycles. if 𝑐2 𝑐3∗ > 0, 𝑐̃2∗ 𝑐2 > 0, and

𝑐3∗ 𝑀1 > 0 for 0 < −ℎ ≪ 1,

Step 5. Fix (𝑐0 , 𝑐0 , 𝑐1 , 𝑐2 , 𝑐3∗ ) = (0, 0, 0, 0, 0), 𝑐5 𝑐6 < 0 and vary (50)
(𝑐∗3 , 𝑐4∗ ) near (0, 0) with 𝑐̃2∗ (= 𝑐∗3 + 𝑐4∗ ) ≠0. 𝑐̃2∗ 𝑀3 > 0 for 0 < ℎ ≪ 1
First, for (𝑐∗3 , 𝑐4∗ ) = (0, 0), we have by (26), (27), and (28)
𝑐5 𝑀1 > 0, 𝑐6 𝑀2 > 0 for 0 < −ℎ ≪ 1, and 𝑐5 𝑀3 < 0 for if 𝑐2 𝑐3∗ > 0, 𝑐̃2∗ 𝑐2 < 0. Thus, 𝑀1 and 𝑀3 each gets one more
0 < ℎ ≪ 1. zero in the first case and only 𝑀1 has a new zero in the second
case. And the zeros got in above steps still exist. Hence, for
Let 0 < |𝑐∗3 , 𝑐4∗ | ≪ |𝑐5 |. Then 0 < |𝑐2 | ≪ |𝑐3∗ | ≪ |𝑐4∗ , 𝑐∗3 | ≪ |𝑐5 | ≪ |𝑐6 | ≪ |𝑐7 | ≪ |𝑐8 | ≪ 1,

𝑐5 𝑀1 < 0, 𝑐6 𝑀2 < 0 for 0 < −ℎ ≪ 1, (7i) the conditions 𝑐2 𝑐3∗ > 0, 𝑐3∗ 𝑐4∗ < 0, 𝑐∗3 𝑐4∗ < 0, 𝑐∗3 𝑐̃2∗ >
(47) 0, 𝑐4∗ 𝑐5 > 0, 𝑐5 𝑐6 < 0, 𝑐6 𝑐7 < 0, 𝑐7 𝑐8 > 0, 𝑐8 𝑐9 > 0
𝑐5 𝑀3 > 0 for 0 < ℎ ≪ 1 imply a distribution (7, 4) + 3 of 14 limit cycles,
Abstract and Applied Analysis 11

(7ii) the conditions 𝑐2 𝑐3∗ > 0, 𝑐3∗ 𝑐4∗ < 0, 𝑐∗3 𝑐4∗ < 0, 𝑐∗3 𝑐̃2∗ < (2) By our assumptions in case (2) and the implicit
0, 𝑐4∗ 𝑐5 > 0, 𝑐5 𝑐6 < 0, 𝑐6 𝑐7 < 0, 𝑐7 𝑐8 > 0, 𝑐8 𝑐9 > 0 function theorem we can take 𝑐0 , 𝑐0 , 𝑐1 , 𝑐2 , 𝑐3∗ , 𝑐4∗ , 𝑐∗3 as
imply a distribution (7, 4) + 3 of 14 limit cycles. free parameters varying near zero. Obviously, for these
parameters varying near zero we have 𝑐5 𝑐6 < 0. By a similar
Step 8. Fix (𝑐0 , 𝑐0 ) = (0, 0), 𝑐2 𝑐3∗ > 0 and vary 𝑐1 near 0. argument in the above proof, we can prove that for 0 <
First, for 𝑐1 = 0, we have by (26), (27), and (28) 𝑐2 𝑀1 > |𝑐0 , 𝑐0 | ≪ |𝑐1 | ≪ |𝑐2 | ≪ |𝑐3∗ | ≪ |𝑐4∗ , 𝑐∗3 | ≪ 1,
0, 𝑐3∗ 𝑀2 > 0 for 0 < −ℎ ≪ 1, and 𝑐2 𝑀3 < 0 for 0 < ℎ ≪ 1. (i) the conditions 𝑐0 𝑐0 > 0, 𝑐0 𝑐1 < 0, 𝑐1 𝑐2 < 0, 𝑐2 𝑐3∗ >
Let 0 < |𝑐1 | ≪ |𝑐2 |. Then 𝑐2 𝑀1 < 0, 𝑐3∗ 𝑀2 < 0 for 0 < 0, 𝑐3∗ 𝑐4∗ < 0, 𝑐∗3 𝑐4∗ < 0, 𝑐∗3 𝑐̃2∗ > 0 (or 𝑐∗3 𝑐̃2∗ <
−ℎ ≪ 1, and 𝑐2 𝑀3 > 0 for 0 < ℎ ≪ 1 if 𝑐1 𝑐2 < 0. Thus, 0), 𝑐4∗ 𝑐5 > 0 imply a distribution (5, 4) + 2 of 11 limit
𝑀1 , 𝑀2 , and 𝑀3 each gets a new zero and the zeros got in cycles,
above steps still exist. Hence, for 0 < |𝑐1 | ≪ |𝑐2 | ≪ |𝑐3∗ | ≪ (ii) the conditions 𝑐0 𝑐0 < 0, 𝑐0 𝑐1 < 0, (𝑐0 + 𝑐0 )𝑐1 >
|𝑐4∗ , 𝑐∗3 | ≪ |𝑐5 | ≪ |𝑐6 | ≪ |𝑐7 | ≪ |𝑐8 | ≪ 1, 0, 𝑐1 𝑐2 < 0, 𝑐2 𝑐3∗ > 0, 𝑐3∗ 𝑐4∗ < 0, 𝑐∗3 𝑐4∗ < 0, 𝑐∗3 𝑐̃2∗ >
(8) the conditions 𝑐1 𝑐2 < 0, 𝑐2 𝑐3∗ > 0, 𝑐3∗ 𝑐4∗ < 0, 𝑐∗3 𝑐4∗ < 0 (or 𝑐∗3 𝑐̃2∗ < 0), 𝑐4∗ 𝑐5 > 0 imply a distribution
0, 𝑐∗3 𝑐̃2∗ > 0 (or 𝑐∗3 𝑐̃2∗ < 0), 𝑐4∗ 𝑐5 > 0, 𝑐5 𝑐6 < 0, 𝑐6 𝑐7 < (5, 3) + 3 of 11 limit cycles,
0, 𝑐7 𝑐8 > 0, 𝑐8 𝑐9 > 0 imply a distribution (8, 5) + 4 of 17 limit
cycles. (iii) the conditions 𝑐0 𝑐0 < 0, 𝑐0 𝑐1 > 0, (𝑐0 + 𝑐0 )𝑐1 >
0, 𝑐1 𝑐2 < 0, 𝑐2 𝑐3∗ > 0, 𝑐3∗ 𝑐4∗ < 0, 𝑐∗3 𝑐4∗ < 0, 𝑐∗3 𝑐̃2∗ >
Step 9. Fix 𝑐1 ≠0 and vary (𝑐0 , 𝑐0 ) near (0, 0). 0 (or 𝑐∗3 𝑐̃2∗ < 0), 𝑐4∗ 𝑐5 > 0 imply a distribution
First, for (𝑐0 , 𝑐0 ) = (0, 0), we have by (26), (27), and (28) (4, 4) + 3 of 11 limit cycles.
𝑐1 𝑀1 > 0, 𝑐1 𝑀2 > 0, for 0 < −ℎ ≪ 1, and 𝑐1 𝑀3 < 0 for (3) By our assumptions in case (3) and the
0 < ℎ ≪ 1. implicit function theorem we can take 𝑐0 , 𝑐0 , 𝑐1 ,
Let 0 < |𝑐0 , 𝑐0 | ≪ |𝑐1 |. Then 𝑐2 , 𝑐3∗ , 𝑐4∗ , 𝑐∗3 , 𝑐5 , 𝑐6 , and 𝑐7 as free parameters varying
𝑐1 𝑀1 < 0, 𝑐1 𝑀2 < 0, for 0 < −ℎ ≪ 1, near zero. Obviously, for these parameters varying near zero
(51) we have 𝑐8 𝑐9 > 0. By a similar argument used in proving case
𝑐1 𝑀3 < 0 for 0 < ℎ ≪ 1 (1), we can prove that for 0 < |𝑐0 , 𝑐0 | ≪ |𝑐1 | ≪ |𝑐2 | ≪ |𝑐3∗ | ≪
|𝑐4∗ , 𝑐∗3 | ≪ |𝑐5 | ≪ |𝑐6 | ≪ |𝑐7 | ≪ 1,
if 𝑐0 𝑐1 < 0, 𝑐0 𝑐1 < 0,
(i) the conditions 𝑐0 𝑐0 > 0, 𝑐0 𝑐1 < 0, 𝑐1 𝑐2 < 0, 𝑐2 𝑐3∗ >
𝑐1 𝑀1 < 0, 𝑐1 𝑀2 > 0, for 0 < −ℎ ≪ 1, 0, 𝑐3∗ 𝑐4∗ < 0, 𝑐∗3 𝑐4∗ < 0, 𝑐∗3 𝑐̃2∗ > 0 (or 𝑐∗3 𝑐̃2∗ <
(52) 0), 𝑐4∗ 𝑐5 > 0, 𝑐5 𝑐6 < 0, 𝑐6 𝑐7 < 0, 𝑐7 𝑐8 > 0 imply a
𝑐1 𝑀3 > 0 for 0 < ℎ ≪ 1 distribution (8, 6) + 4 of 18 limit cycles,
if 𝑐0 𝑐1 < 0, 𝑐0 𝑐1 > 0 and (𝑐0 + 𝑐0 )𝑐1 > 0, and (ii) the conditions 𝑐0 𝑐0 < 0, 𝑐0 𝑐1 < 0, (𝑐0 + 𝑐0 )𝑐1 >
0, 𝑐1 𝑐2 < 0, 𝑐2 𝑐3∗ > 0, 𝑐3∗ 𝑐4∗ < 0, 𝑐∗3 𝑐4∗ < 0, 𝑐∗3 𝑐̃2∗ >
𝑐1 𝑀1 > 0, 𝑐1 𝑀2 < 0, for 0 < −ℎ ≪ 1, 0 (or 𝑐∗3 𝑐̃2∗ < 0), 𝑐4∗ 𝑐5 > 0, 𝑐5 𝑐6 < 0, 𝑐6 𝑐7 < 0, 𝑐7 𝑐8 >
(53)
𝑐1 𝑀3 > 0 for 0 < ℎ ≪ 1 0 imply a distribution (8, 5) + 5 of 18 limit cycles,
(iii) the conditions 𝑐0 𝑐0 < 0, 𝑐0 𝑐1 > 0, (𝑐0 + 𝑐0 )𝑐1 >
if 𝑐0 𝑐1 > 0, 𝑐0 𝑐1 < 0 and (𝑐0 + 𝑐0 )𝑐1 > 0. Thus, we have 0, 𝑐1 𝑐2 < 0, 𝑐2 𝑐3∗ > 0, 𝑐3∗ 𝑐4∗ < 0, 𝑐∗3 𝑐4∗ < 0, 𝑐∗3 𝑐̃2∗ >
correspondingly (a) 𝑀1 and 𝑀2 each has a new zero, (b) 𝑀1 0 (or 𝑐∗3 𝑐̃2∗ < 0), 𝑐4∗ 𝑐5 > 0, 𝑐5 𝑐6 < 0, 𝑐6 𝑐7 < 0, 𝑐7 𝑐8 >
and 𝑀3 each has a new zero, or (c) 𝑀2 and 𝑀3 each has a new 0 imply a distribution (7, 6) + 5 of 18 limit cycles.
zero. And the zeros got in above steps still exist.
Hence, for 0 < |𝑐0 , 𝑐0 | ≪ |𝑐1 | ≪ |𝑐2 | ≪ |𝑐3∗ | ≪ |𝑐4∗ , 𝑐∗3 | ≪ This completes the proof.
|𝑐5 | ≪ |𝑐6 | ≪ |𝑐7 | ≪ |𝑐8 | ≪ 1,
(9i) the conditions 𝑐0 𝑐0 > 0, 𝑐0 𝑐1 < 0, 𝑐1 𝑐2 < 0, 𝑐2 𝑐3∗ > 3. An Application
0, 𝑐3∗ 𝑐4∗ < 0, 𝑐∗3 𝑐4∗ < 0, 𝑐∗3 𝑐̃2∗ > 0 (or 𝑐∗3 𝑐̃2∗ < Consider a Liénard system of the form
0), 𝑐4∗ 𝑐5 > 0, 𝑐5 𝑐6 < 0, 𝑐6 𝑐7 < 0, 𝑐7 𝑐8 > 0, 𝑐8 𝑐9 > 0
imply a distribution (9, 6) + 4 of 19 limit cycles, 1 1
𝑥̇= 𝑦, 𝑦̇= −(𝑥 + 1)2 𝑥3 (𝑥2 − 𝑥 − ) − 𝜖𝑓 (𝑥, 𝛿) 𝑦,
(9ii) the conditions 𝑐0 𝑐0 < 0, 𝑐0 𝑐1 < 0, (𝑐0 + 𝑐0 )𝑐1 > 4 2
(54)
0, 𝑐1 𝑐2 < 0, 𝑐2 𝑐3∗ > 0, 𝑐3∗ 𝑐4∗ < 0, 𝑐∗3 𝑐4∗ < 0, 𝑐∗3 𝑐̃2∗ >
0 (or 𝑐∗3 𝑐̃2∗ < 0), 𝑐4∗ 𝑐5 > 0, 𝑐5 𝑐6 < 0, 𝑐6 𝑐7 < 0, 𝑐7 𝑐8 > where
0, 𝑐8 𝑐9 > 0 imply a distribution (9, 5) + 5 of 19 limit 𝑛
cycles, 𝑓 (𝑥, 𝛿) = ∑ 𝑎𝑗 𝑥𝑗 , 𝛿 = (𝑎0 , 𝑎1 , . . . , 𝑎𝑛 ) ,
(9iii) the conditions 𝑐0 𝑐0 < 0, 𝑐0 𝑐1 > 0, (𝑐0 + 𝑐0 )𝑐1 > 𝑗=0 (55)
0, 𝑐1 𝑐2 < 0, 𝑐2 𝑐3∗ > 0, 𝑐3∗ 𝑐4∗ < 0, 𝑐∗3 𝑐4∗ < 0, 𝑐∗3 𝑐̃2∗ >
0 (or 𝑐∗3 𝑐̃2∗ < 0), 𝑐4∗ 𝑐5 > 0, 𝑐5 𝑐6 < 0, 𝑐6 𝑐7 < 0, 𝑐7 𝑐8 > 8 ≤ 𝑛 ≤ 12.
0, 𝑐8 𝑐9 > 0 imply a distribution (8, 6) + 5 of 19 limit System (54) |𝜖=0 is Hamiltonian with
cycles.
1 1
Thus we get the conclusion for 𝑙 = 9. 𝐻 (𝑥, 𝑦) = 𝑦2 + (𝑥 + 1)3 𝑥4 (𝑥 − 1) . (56)
2 8
12 Abstract and Applied Analysis

We have the following theorem. 𝜋 2 𝜋 2 𝜋 8

𝑐0 (𝛿) = −𝑎0 ( + ) − 𝑎1 ( + ) − 𝑎2 ( + )
16 15 32 15 32 105
Theorem 5. Let 𝐶(𝑛) denote the maximal number of limit 5𝜋 8 5𝜋 16
cycles of the system (54) for 𝜖 small and all 𝛿. Then we have − 𝑎3 ( + ) − 𝑎4 ( + )
256 105 256 315
𝐶(11) ≥ 18, 𝐶(𝑛) ≥ 2𝑛 − 5 (𝑛 = 8, 9, 10, 12).
7𝜋 16 7𝜋 128
Proof. It is easy to verify that the unperturbed system has a − 𝑎5 ( + ) − 𝑎6 ( + )
512 315 512 3465
compound loop 𝐿 0 = 𝐿 1 ∪ 𝐿 2 ∪ 𝐿 3 ∪ {𝑆1 , 𝑆2 } with a cusp
𝑆1 (−1, 0) of order one and a nilpotent saddle 𝑆2 (0, 0) of order 21𝜋 128 21𝜋 256
− 𝑎7 ( + ) − 𝑎8 ( + )
one, 𝐿 1 , 𝐿 2 are heteroclinic orbits satisfying 𝜔(𝐿 1 ) = 𝛼(𝐿 2 ) = 2048 3465 2048 9009
𝑆2 and 𝜔(𝐿 2 ) = 𝛼(𝐿 1 ) = 𝑆1 , and 𝐿 3 is a homoclinic loop to 𝑆2 . 33𝜋 256 33𝜋 1024
Inside 𝐿∗ = 𝐿 1 + 𝐿 2 (𝐿 3 , resp.), there is a center 𝐶1 ((1/8)(1 − − 𝑎9 ( + ) − 𝑎10 ( + )
4096 9009 4096 45045
√33), 0) (𝐶2 ((1/8)(1 + √33), 0), resp.).
Because of the similarity in the proof, here we only prove 429𝜋 1024
− 𝑎11 ( + ). (59)
the case for 𝑛 = 11 and omit the rest of the proof. 65536 45045
Let 𝑛 = 11. By Theorem 3, we obtain
Note that 𝑆2 is a nilpotent saddle of order one and ℎ4 = −1/8.
By (23), we have
11
𝑗 󵄨 ̃ 󵄨󵄨
𝑐0 (𝛿) = 𝑀1 (0, 𝛿) = −∮ 𝑓 (𝑥, 𝛿) 𝑦 𝑑𝑥 = − ∑𝑎𝑗 𝐼1 , 𝑐1 (𝛿) = 𝑐1 (𝑆2 , 𝛿) = 29/4 󵄨󵄨󵄨󵄨𝐴
𝐿∗ 0 󵄨󵄨󵄨 𝑎0 ,
𝑗=0
(57)
11 𝑐3 (𝛿) = 𝑐2 (𝑆2 , 𝛿) = 𝑎1 + 𝑂1 (𝑎0 ) ,
𝑗
𝑐0 (𝛿) = 𝑀2 (0, 𝛿) = −∮ 𝑓 (𝑥, 𝛿) 𝑦 𝑑𝑥 = − ∑𝑎𝑗 𝐼2 , 󵄨 ̃ 󵄨󵄨
𝐿3 𝑗=0 𝑐6 (𝛿) = 𝑐4 (𝑆2 , 𝛿) = 23/4 󵄨󵄨󵄨󵄨𝐴 2 󵄨󵄨󵄨 (−21𝑎0 + 12𝑎1 − 8𝑎2 ) ,

𝑐7 (𝛿) = 𝑐5 (𝑆2 , 𝛿)
where 󵄨󵄨 ̃ 󵄨󵄨
󵄨𝐴 0 󵄨󵄨 2035
1/4 󵄨󵄨
=2 󵄨( 𝑎 − 310𝑎1 + 180𝑎2 − 80𝑎3 + 32𝑎4 ) ,
7 4 0
0
𝑗
𝐼1 = ∮ 𝑥𝑗 𝑦 𝑑𝑥 = 2 ∫ 𝑥𝑗 𝑦 𝑑𝑥 𝑐9 (𝛿) = 𝑐6 (𝑆2 , 𝛿)
𝐿∗ −1
0 333 207
= 𝑎 − 𝑎 + 29𝑎2 − 15𝑎3 + 6𝑎4 − 2𝑎5 .
= ∫ 𝑥𝑗+2 (𝑥 + 1) √1 − 𝑥2 𝑑𝑥, 𝑗 = 0, 1, . . . , 11, 4 0 4 1
−1 (60)
(58)
1
𝑗
𝐼2 = ∮ 𝑥𝑗 𝑦 𝑑𝑥 = 2 ∫ 𝑥𝑗 𝑦 𝑑𝑥 Making the transformation 𝑥 = 𝑢 − 1, 𝑦 = V, system (54)
𝐿3 0
becomes
1
= ∫ 𝑥𝑗+2 (𝑥 + 1) √1 − 𝑥2 𝑑𝑥, 𝑗 = 0, 1, . . . , 11. 9 3
0 𝑢̇= V, V̇= −𝑢2 (𝑢 − 1)3 (𝑢2 − 𝑢 + ) − 𝜖𝑓̃ (𝑢, 𝛿) V.
4 4
(61)
Therefore,
Then we have

𝜋 2 𝜋 2 𝜋 8 V2 1 8
𝑐0 (𝛿) = −𝑎0 ( − ) − 𝑎1 ( − ) − 𝑎2 ( − ) 𝐻 (𝑢, V) = + (𝑢 − 6𝑢7 + 14𝑢6 − 16𝑢5 + 9𝑢4 − 2𝑢3 ) ,
16 15 32 15 32 105 2 8

− 𝑎3 (
5𝜋
−
8
) − 𝑎4 (
5𝜋
−
16
) ̃ℎ = − 1 ,
3
256 105 256 315 4
7𝜋 16 7𝜋 128 11 11
− 𝑎5 ( − ) − 𝑎6 ( − ) 𝑓̃ (𝑢, 𝛿) = ∑ 𝑎𝑗 (𝑢 − 1)𝑗 = ∑ (−1)𝑗 𝑎𝑗
512 315 512 3465
𝑗=0 𝑗=0
21𝜋 128 21𝜋 256
− 𝑎7 ( − ) − 𝑎8 ( − ) 11 11
2048 3465 2048 9009
+ ∑(−1)(𝑗−1) 𝑗𝑎𝑗 𝑢 + ∑ (−1)(𝑗−2) 𝐶𝑗2 𝑎𝑗 𝑢2
33𝜋 256 33𝜋 1024 𝑗=1 𝑗=2
− 𝑎9 ( − ) − 𝑎10 ( − )
4096 9009 4096 45045 11 11
429𝜋 1024 + ∑(−1)(𝑗−3) 𝐶𝑗3 𝑎𝑗 𝑢3 + ∑ (−1)(𝑗−4) 𝐶𝑗4 𝑎𝑗 𝑢4
− 𝑎11 ( − ),
65536 45045 𝑗=3 𝑗=4
Abstract and Applied Analysis 13

11 11
where
+ ∑ (−1)(𝑗−5) 𝐶𝑗5 𝑎𝑗 𝑢5 + ∑(−1)(𝑗−6) 𝐶𝑗6 𝑎𝑗 𝑢6
𝑗=5 𝑗=6 𝑥𝑖−2 [1 + (−1)𝑖 𝑥(11−𝑖) ]
𝑓𝑖 (𝑥) = , 𝑖 = 2, 3, . . . , 10. (66)
11 11 (𝑥 + 1) √1 − 𝑥2
+ ∑ (−1)(𝑗−7) 𝐶𝑗7 𝑎𝑗 𝑢7 + ∑(−1)(𝑗−8) 𝐶𝑗8 𝑎𝑗 𝑢8
𝑗=7 𝑗=8 Similarly,
11
𝑐3 (𝛿) = ∮ (𝑝𝑥 + 𝑞𝑦 ) 𝑑𝑡 = −∮ 𝑓 (𝑥, 𝛿) 𝑑𝑡
+ ∑ (−1)(𝑗−9) 𝐶𝑗9 𝑎𝑗 𝑢9 + (𝑎10 − 11𝑎11 ) 𝑢10 + 𝑎11 𝑢11 . 𝐿3 𝐿3
𝑗=9
𝑓 (𝑥, 𝛿) 1 𝑓
(62) (𝑥, 𝛿)
= −∮ 𝑑𝑥 = −2 ∫ 𝑑𝑥 (67)
𝐿3 𝑦 0 𝑦
By (19), we have
10 1
11 = −4∑𝑎𝑖 ∫ 𝑓𝑖 (𝑥) 𝑑𝑥.
𝑐2 (𝛿) = 𝑐1 (𝑆1 , 𝛿) = 213/6 ∑(−1)𝑗 𝑎𝑗 , 𝑖=2 0
𝑗=0
Therefore,
11
315𝜋 93 187𝜋 93
𝑐5 (𝛿) = 𝑐3 (𝑆1 , 𝛿) = 217/6 [∑ (−1)𝑗 (𝑗 − 3) 𝑎𝑗 − 3𝑎0 ] , 𝑐4 (𝛿) = −4𝑎2 ( + ) + 4𝑎3 ( + )
256 35 256 35
[𝑗=1 ]
187𝜋 58 123𝜋 58
𝑐8 (𝛿) = 𝑐4 (𝑆1 , 𝛿) − 4𝑎4 ( + ) + 4𝑎5 ( + )
(63) 256 35 256 35
1316 460 224 123𝜋 104 75𝜋 104
= 213/6 (− 𝑎0 + 272𝑎1 − 𝑎2 + 𝑎 − 4𝑎6 ( + ) + 4𝑎7 ( + )
3 3 3 3 256 105 256 105
16 4 4 75𝜋 16 35𝜋 16
− 28𝑎4 + 𝑎 + 𝑎 − 𝑎 − 4𝑎8 ( + ) + 4𝑎9 ( + )
3 5 3 6 3 8 256 35 256 35
16 224 35𝜋
− 𝑎 + 28𝑎10 − 𝑎 ). − 4𝑎10 ( ),
3 9 3 11 256
(68)
Note that 315𝜋 93 187𝜋 93
𝑐3 (𝛿) = −4𝑎2 ( − ) + 4𝑎3 ( − )
󵄨 ̃ 󵄨󵄨 256 35 256 35
𝑐1 (𝛿) = 𝑐1 (𝑆2 , 𝛿) = 29/4 󵄨󵄨󵄨󵄨𝐴 0 󵄨󵄨󵄨 𝑎0 ,
187𝜋 58 123𝜋 58
11 − 4𝑎4 ( − ) + 4𝑎5 ( − )
13/6 𝑗 256 35 256 35
𝑐2 (𝛿) = 𝑐1 (𝑆1 , 𝛿) = 2 ∑ (−1) 𝑎𝑗 , (64)
𝑗=0 123𝜋 104 75𝜋 104
− 4𝑎6 ( − ) + 4𝑎7 ( − )
256 105 256 105
𝑐3 (𝛿) = 𝑐2 (𝑆2 , 𝛿) = 𝑎1 + 𝑂1 (𝑎0 ) .
75𝜋 16 35𝜋 16
We have 𝑐1 (𝛿) = 𝑐2 (𝛿) = 𝑐3 (𝛿) = 0 if and only if 𝑎0 = 𝑎1 = 0, − 4𝑎8 ( − ) + 4𝑎9 ( − )
256 35 256 35
and 𝑎11 = ∑10 𝑖
𝑖=2 (−1) 𝑎𝑖 . It implies further that 35𝜋
− 4𝑎10 ( ).
256
𝑐4 (𝛿) = ∮ (𝑝𝑥 + 𝑞𝑦 ) 𝑑𝑡 = −∮ 𝑓 (𝑥, 𝛿) 𝑑𝑡
𝐿∗ 𝐿∗ Let 𝑐4∗ (𝛿) = 𝑐4 (𝛿)|𝑐1 =𝑐2 =𝑐3 =0 , 𝑐∗3 (𝛿) = 𝑐3 (𝛿)𝑐1 =𝑐3 =0 ,
𝑓 (𝑥, 𝛿) 0 𝑓
(𝑥, 𝛿) 𝑐3∗ (𝛿) = 𝑐3 (𝛿)|𝑐1 (𝛿)=0 . Furthermore, one sees that equations
= −∮ 𝑑𝑥 = −2 ∫ 𝑑𝑥 𝑐0 (𝛿) = 𝑐0 (𝛿) = 𝑐1 (𝛿) = 𝑐2 (𝛿) = 𝑐3∗ (𝛿) = 𝑐4∗ (𝛿) = 𝑐∗3 (𝛿) =
𝐿∗ 𝑦 −1 𝑦
𝑐5 (𝛿) = 𝑐6 (𝛿) = 𝑐7 (𝛿) = 0 have the solution 𝑎0 = 𝑎1 = 𝑎2 =
0
1 11 𝑖 0, 𝑎3 = (13/74)𝑎9 − (1/2)𝑎7 , 𝑎4 = (65/148)𝑎9 − (5/4)𝑎7 , 𝑎5 =
= −2 ∫ (∑𝑎 𝑥 ) 𝑑𝑥 −(16/37)𝑎9 , 𝑎6 = (7/4)𝑎7 −(255/148)𝑎9 , 𝑎8 = (96/37)𝑎9 , 𝑎10 =
−1 𝑦 𝑖=0 𝑖
−(53/37)𝑎9 , 𝑎11 = −(32/37)𝑎9 , which gives 𝑐8 (𝛿) =
0 (65) (55/37)231/6 𝑎9 , 𝑐9 (𝛿) = (32/37)𝑎9 . And further, 𝑐8 (𝛿)𝑐9 (𝛿) >
1
= −4 ∫ 0 if 𝑎9 ≠0. Thus, fix 𝑎9 ≠0 and take 𝛿0 = (0, 0, 0, (13/74)𝑎9 −
−1 𝑥2 (𝑥 + 1) √1 − 𝑥2
(1/2)𝑎7 , (65/148)𝑎9 − (5/4)𝑎7 , −(16/37)𝑎9 , (7/4)𝑎7 −
10 10 (255/148)𝑎9 , 𝑎7 , (96/37)𝑎9 , 𝑎9 , −(53/37)𝑎9 , −(32/37)𝑎9 ).
× (∑𝑎𝑖 𝑥𝑖 + ∑(−1)𝑖 𝑎𝑖 𝑥11 ) 𝑑𝑥 Then we have
𝑖=2 𝑖=2
𝜕 (𝑐0 , 𝑐0 , 𝑐1 , 𝑐2 , 𝑐3∗ , 𝑐4∗ , 𝑐∗3 , 𝑐5 , 𝑐6 , 𝑐7 )
10 0 rank = 10.
= −4∑𝑎𝑖 ∫ 𝑓𝑖 (𝑥) 𝑑𝑥, 𝜕 (𝑎0 , 𝑎1 , 𝑎2 , 𝑎3 , 𝑎4 , 𝑎5 , 𝑎6 , 𝑎7 , 𝑎8 , 𝑎9 , 𝑎10 , 𝑎11 )
𝑖=2 −1 (69)
14 Abstract and Applied Analysis

Hence by Theorem 4(3), we know that there are 18 limit cycles [13] A. Atabaigi and H. R. Z. Zangeneh, “Bifurcation of limit cycles
near 𝐿 0 for some 𝛿 near 𝛿0 . This ends the proof. in small perturbations of a class of hyper-elliptic Hamiltonian
systems of degree 5 with a cusp,” The Journal of Applied Analysis
and Computation, vol. 1, no. 3, pp. 299–313, 2011.
Conflict of Interests
[14] R. Kazemi, H. R. Z. Zangeneh, and A. Atabaigi, “On the number
The authors declare that there is no conflict of interests of limit cycles in small perturbations of a class of hyper-elliptic
regarding the publication of this paper. Hamiltonian systems,” Nonlinear Analysis: Theory, Methods &
Applications, vol. 75, pp. 574–587, 2012.
[15] J. Wang and D. Xiao, “On the number of limit cycles in small
Acknowledgments perturbations of a class of hyper-elliptic Hamiltonian systems
with one nilpotent saddle,” Journal of Differential Equations, vol.
The project was supported by National Natural Science Foun- 250, no. 4, pp. 2227–2243, 2011.
dation of China (11271261) and FP7-PEOPLE-2012-IRSES-
[16] J. Wang, “Estimate of the number of zeros of Abelian integrals
316338.
for a perturbation of hyperelliptic Hamiltonian system with
nilpotent center,” Chaos, Solitons & Fractals, vol. 45, no. 9-10,
References pp. 1140–1146, 2012.
[17] Y. Xiong, “Bifurcation of limit cycles by perturbing a class of
[1] M. Han, “Asymptotic expansions of Melnikov functions and hyper-elliptic Hamiltonian systems of degree five,” Journal of
limit cycle bifurcations,” International Journal of Bifurcation and Mathematical Analysis and Applications, vol. 411, no. 2, pp. 559–
Chaos in Applied Sciences and Engineering, vol. 22, no. 12, Article 573, 2014.
ID 1250296, 30 pages, 2012.
[2] A. Atabaigi, H. R. Z. Zangeneh, and R. Kazemi, “Limit cycle
bifurcation by perturbing a cuspidal loop of order 2 in a
Hamiltonian system,” Nonlinear Analysis: Theory, Methods &
Applications, vol. 75, no. 4, pp. 1945–1958, 2012.
[3] M. Han, H. Zang, and J. Yang, “Limit cycle bifurcations by
perturbing a cuspidal loop in a Hamiltonian system,” Journal
of Differential Equations, vol. 246, no. 1, pp. 129–163, 2009.
[4] M. Han, J. Yang, and D. Xiao, “Limit cycle bifurcations near a
double homoclinic loop with a nilpotent saddle,” International
Journal of Bifurcation and Chaos in Applied Sciences and
Engineering, vol. 22, no. 8, Article ID 1250189, 33 pages, 2012.
[5] M. Han, Bifurcation Theory of Limit Cycles, Science Press, 2013.
[6] Z. Hu, B. Gao, and V. G. Romanovski, “Limit cycle bifurcations
from centers of symmetric Hamiltonian systems pertrbed by
cubic polynomials,” International Journal of Bifurcation and
Chaos, vol. 23, no. 3, Article ID 1350043, 20 pages, 2013.
[7] J. Jiang, “Bifurcation of limit cycles for a quartic near-
Hamiltonian system by perturbing a nilpotent center,” Journal
of Mathematical Analysis and Applications, vol. 365, no. 1, pp.
376–384, 2010.
[8] J. Li, T. Zhang, and M. Han, “Bifurcation of limit cycles from
a heteroclinic loop with two cusps,” Chaos, Solitons & Fractals,
vol. 62-63, pp. 44–54, 2014.
[9] X. Sun, M. Han, and J. Yang, “Bifurcation of limit cycles from
a heteroclinic loop with a cusp,” Nonlinear Analysis: Theory,
Methods & Applications, vol. 74, no. 9, pp. 2948–2965, 2011.
[10] J. Yang, “On the limit cycles of a kind of Liénard system with
a nilpotent center under perturbations,” The Journal of Applied
Analysis and Computation, vol. 2, no. 3, pp. 325–339, 2012.
[11] H. Zang, M. Han, and D. Xiao, “On Melnikov functions of a
homoclinic loop through a nilpotent saddle for planar near-
Hamiltonian systems,” Journal of Differential Equations, vol.
245, no. 4, pp. 1086–1111, 2008.
[12] H. Zang, Z. Wang, and T. Zhang, “Bifurcations and distribution
of limit cycles for near-Hamiltonian polynomial systems,”
Journal of Mathematical Analysis and Applications, vol. 348, no.
1, pp. 87–100, 2008.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 319585, 8 pages
http://dx.doi.org/10.1155/2014/319585

Research Article
On Regularity Criteria for the Two-Dimensional Generalized
Liquid Crystal Model

Yanan Wang and Zaihong Jiang

Department of Mathematics, Zhejiang Normal University, Jinhua , Zhejiang 321004, China

Correspondence should be addressed to Zaihong Jiang; jzhong790405@gmail.com

Received 5 May 2014; Revised 27 June 2014; Accepted 1 July 2014; Published 15 July 2014

Academic Editor: Yongli Song

Copyright © 2014 Y. Wang and Z. Jiang. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

We establish the regularity criteria for the two-dimensional generalized liquid crystal model. It turns out that the global existence
results satisfy our regularity criteria naturally.

1. Introduction equations. In this sense, the study of the system (1)–(4) can
be valuable and interesting in both mathematical sense and
In this paper we consider the following two-dimensional (2D) physical sense.
liquid crystal model: The existence and uniqueness of the weak and smooth
solutions for system (1)–(4) are given in [4–6] when 𝛼 = 𝛽 =
𝑢𝑡 + 𝑢 ⋅ ∇𝑢 + ∇𝑝 + Λ2𝛼 𝑢 = −∇𝑑 ⋅ Δ𝑑, (𝑡, 𝑥) ∈ R+ × R2 , 1. Local existence of classical solutions for the nematic liquid
(1) crystal flows was established in [7].
Now, we mention some known results about regular-
𝑑𝑡 + 𝑢 ⋅ ∇𝑑 + Λ2𝛽 𝑑 = −𝑓 (𝑑) , (𝑡, 𝑥) ∈ R+ × R2 , (2) ity theory for the system. In 2010, Zhou and Fan estab-
𝑇
∇ ⋅ 𝑢 = 0, (𝑡, 𝑥) ∈ R+ × R2 , (3) lished a regularity criterion for it as ∫0 ‖∇𝑢‖𝑟𝐿𝑝 /(1 + ln(𝑒 +
‖∇𝑢‖𝐿𝑝 ))𝑑𝑡 < +∞ with 2/𝑟 + 3/𝑝 = 2, 2 ≤ 𝑝 ≤ 3 in
(𝑢, 𝑑) (𝑥, 0) = (𝑢0 , 𝑑0 ) (𝑥) , 𝑥 ∈ R2 , (4) [8]. Later, some regularity criteria are proved for the system
with zero dissipation in [9]. In [10], Fan et al. established a
where 𝑢(𝑥, 𝑡) ∈ R2 represents the velocity field, 𝑑(𝑥, 𝑡) ∈ R2 global regularity for this system with mixed partial viscosity.
is a vectorial field modeling the orientation of the crystal Recently, in [11, 12], it is proved that smooth solutions are
molecules, and 𝑝 is the scalar pressure, while 𝛼 ≥ 0, 𝛽 ≥ 0 are global in the following three cases: 0 < 𝛽 < 1, 𝛼 + 𝛽 ≥ 2;
two real parameters. 𝑓(𝑑) := (|𝑑|2 − 1)𝑑 and the operator Λ = 𝛼 > 0, 𝛽 = 1; 𝛼 = 0, 𝛽 > 1. Moreover, global strong solution
̂ = |𝜉|𝑓(𝜉),
(−Δ)1/2 is defined by Λ𝑓(𝜉) ̂ ̂ denotes
and here 𝑓(𝜉) to the density-dependent 2D liquid crystal flows was studied
the Fourier transform of 𝑓(𝑥). We identify the cases 𝛼 = 0 or in [13].
𝛽 = 0 as the 2D generalized liquid crystal model with zero This paper is devoted to obtain some regularity criteria
velocity diffusion or zero orientation diffusion, respectively. for the generalized system (1)–(4). Our main results are the
We say our system is a generalized form of liquid crystal following Theorems.
model. When 𝛼 = 0, 𝛽 = 1, the system is a simplified version In our Theorems we set 𝜌 = max{2/𝛼, 2}, 󰜚 = max{2/
of the Ericksen-Leslie system modeling the hydrodynamics 𝛽, 2}. The first one is for large 𝛼 and 𝛽.
of nematic liquid crystals which was developed during the
period of 1958 through 1968 [1–3]. We notice that if 𝑑 ≡ 0, Theorem 1. Let 𝛼, 𝛽 ≥ 1/2. Suppose (𝑢0 , 𝑑0 )(𝑥) ∈ 𝐻2 (R2 ) ×
𝛼 = 1, then the system (1)–(4) becomes the Navier-Stokes 𝐻3 (R2 ) and (𝑢, 𝑑)(𝑥, 𝑡) is a local smooth solution of the
2 Abstract and Applied Analysis

2D generalized liquid crystal model (1)–(4). If 𝑢(𝑥, 𝑡), 𝑑(𝑥, 𝑡) 2. Proof of Theorem 1
satisfy
In this section, we are devoted to prove our main Theorem 1.
𝑇
∫
1/(1−𝜃 )
‖∇𝑢(⋅, 𝑡)‖𝐿𝑝 𝛽 𝑑𝑡 < ∞, (5) Under the assumption in Theorem 1, if 𝑢 ∈ 𝐿∞ (0, 𝑇; 𝐻1 (R2 ))
0 and 𝑑 ∈ 𝐿∞ (0, 𝑇; 𝐻2 (R2 )), we can deduce 𝑢 ∈
or 𝐿∞ (0, 𝑇; 𝐻2 (R2 )) and 𝑑 ∈ 𝐿∞ (0, 𝑇; 𝐻3 (R2 )). So we only have
𝑇 to give the regularity criteria to guarantee the 𝐻1 estimation
∫ ‖Δ𝑑(⋅, 𝑡)‖2𝐿𝜌 𝑑𝑡 < ∞, (6) for 𝑢, ∇𝑑.
0
or Proof. Firstly, we give the following priori estimates.
𝑇 Multiplying (2) by 𝑝|𝑑|𝑝−2 𝑑, integrating over R2 , after
∫ ‖Δ𝑑(⋅, 𝑡)‖2𝐿󰜚 𝑑𝑡 < ∞, (7) integrating by parts, and using the following property,
0
then (𝑢(𝑥, 𝑡), 𝑏(𝑥, 𝑡)) is a regular solution in [0, 𝑇]. Here 𝜃𝛽 =
∫ Λ2𝛽 𝑑 ⋅ |𝑑|𝑝−2 ⋅ 𝑑 > 0, (12)
1/𝑝𝛽, 𝑝𝛽 ≥ 1.
If in addition, 1 > 𝛽 ≥ 1/2, 𝛼 > 0, Λ𝛽 𝑏 satisfies we obtain
𝑇
󵄩 󵄩2/(1−𝜃𝛼 ) 𝑑
∫ 󵄩󵄩󵄩󵄩Λ𝛽+1 𝑑(⋅, 𝑡)󵄩󵄩󵄩󵄩𝐿𝑠 𝑑𝑡 < ∞, (8) 𝑝 𝑝+2 𝑝
‖𝑑‖ 𝑝 + 𝑝‖𝑑‖𝐿𝑝+2 = 𝑝‖𝑑‖𝐿𝑝 (13)
0 𝑑𝑡 𝐿
then (𝑢(𝑥, 𝑡), 𝑏(𝑥, 𝑡)) is a regular solution in [0, 𝑇], where 𝜃𝛼 = for any 2 ≤ 𝑝 < ∞. Applying Gronwall’s inequality, we
((2/𝑠) − (2𝛽 − 1))/𝛼, (2𝛽 − 1)/2 ≤ 1/𝑠 ≤ (2𝛽 − 1 + 𝛼)/2. deduce that
𝑇
The following theorems are established for the cases 𝛼 or 𝑝 𝑝+2 󵄩 󵄩𝑝
‖𝑑‖𝐿𝑝 + 𝑝 ∫ ‖𝑑‖𝐿𝑝+2 𝑑𝜏 ≤ 𝑒𝑇𝑝 󵄩󵄩󵄩𝑑0 󵄩󵄩󵄩𝐿𝑝 . (14)
𝛽 small. 0

Theorem 2. Let 0 < 𝛼 < 1/2, 𝛽 > 0 or 0 < 𝛽 < 1/2, 𝛼 > 0. Multiplying (1) and (2) by 𝑢 and −Δ𝑑, respectively, integrating
Suppose (𝑢0 , 𝑑0 )(𝑥) ∈ 𝐻2 (R2 ) × 𝐻3 (R2 ) and (𝑢, 𝑑)(𝑥, 𝑡) is a over R2 , and adding the resulting equations together, we
local smooth solution of the 2D generalized liquid crystal model obtain
(1)–(4). If 𝑢(𝑥, 𝑡), 𝑑(𝑥, 𝑡) satisfy 1 𝑑 󵄩 󵄩2 󵄩 󵄩2
(‖𝑢‖2𝐿2 + ‖∇𝑑‖2𝐿2 ) + 󵄩󵄩󵄩Λ𝛼 𝑢󵄩󵄩󵄩𝐿2 + 󵄩󵄩󵄩󵄩Λ𝛽+1 𝑑󵄩󵄩󵄩󵄩𝐿2
𝑇 𝑇 2 𝑑𝑡
∫ ‖∇𝑢(⋅, 𝑡)‖1/(1−𝜎)
𝐿𝑝 𝑑𝑡 < ∞, ∫ ‖Δ𝑑(⋅, 𝑡)‖1/(1−𝜎)
𝐿𝑝 𝑑𝑡 < ∞,
0 0
= − ∫ ∇𝑓 (𝑑) ∇𝑑𝑑𝑥
(9) R2

then the solution remains smooth on (0, 𝑇]. Here 𝜎 =

max{𝜃𝛼 , 𝜃𝛽 }, 𝜃𝛼 = 1/𝑝𝛼, 𝑝𝛼 ≥ 1, 𝜃𝛽 = 1/𝑝𝛽, 𝑝𝛽 ≥ 1. ≤ 𝐶 ∫ |𝑑|2 |∇𝑑|2 𝑑𝑥 + ‖∇𝑑‖2𝐿2
R2 (15)
Theorem 3. Let 𝛼 = 0, 𝛽 > 0 or 𝛼 > 0, 𝛽 = 0. Suppose ≤ 𝐶‖𝑑‖2𝐿2𝑝 ‖∇𝑑‖2𝐿2𝑞 + ‖∇𝑑‖2𝐿2
(𝑢0 , 𝑑0 )(𝑥) ∈ 𝐻2 (R2 ) × 𝐻3 (R2 ) and (𝑢, 𝑑)(𝑥, 𝑡) is a local
󵄩󵄩 𝛽+1 󵄩󵄩2𝜃
smooth solution of the 2D generalized liquid crystal model (1)– ≤ 𝐶‖𝑑‖2𝐿2𝑝 ‖∇𝑑‖2−2𝜃
𝐿2 󵄩
󵄩Λ 𝑑󵄩󵄩 2 + ‖∇𝑑‖2𝐿2
󵄩 󵄩𝐿
(4). If 𝑢(𝑥, 𝑡), 𝑑(𝑥, 𝑡) satisfy
1󵄩 󵄩2
𝑇 ≤ 𝐶‖𝑑‖2/(1−𝜃) ‖∇𝑑‖2𝐿2 + 󵄩󵄩󵄩󵄩Λ𝛽+1 𝑑󵄩󵄩󵄩󵄩𝐿2 + ‖∇𝑑‖2𝐿2 ,
∫ ‖∇𝑢(⋅, 𝑡)‖BMO + ‖∇𝑢(⋅, 𝑡)‖2𝐿󰜚 + ‖Δ𝑑(⋅, 𝑡)‖2𝐿󰜚 𝑑𝑡 < ∞, (10) 𝐿2𝑝 2
0
or where 1/𝑝 + 1/𝑞 = 1, 𝑝 > 1/𝛽. Here we have used the
𝑇 following Galiardo-Nirenberg inequality:
∫ ‖∇𝑢(⋅, 𝑡)‖𝐿∞ + ‖Δ𝑑(⋅, 𝑡)‖2/(2−𝜃) + ‖𝑑(⋅, 𝑡)‖2BMO 𝑑𝑡 < ∞ 1 1 𝛽 1−𝜃
𝐿𝑝 󵄩󵄩 󵄩󵄩 󵄩 󵄩1−𝜃 󵄩 𝛽 󵄩𝜃
0 󵄩󵄩𝑓󵄩󵄩𝐿2𝑞 ≤ 𝐶󵄩󵄩󵄩𝑓󵄩󵄩󵄩𝐿2 󵄩󵄩󵄩󵄩Λ 𝑓󵄩󵄩󵄩󵄩𝐿2 , = ( − )𝜃 + ,
(11) 2𝑞 2 2 2
respectively, then the solution remains smooth on (0, 𝑇]. 0 < 𝜃 < 1,
(16)
Remark 4. The results in this paper are motivated by the
recent works on 2D incompressible generalized MHD equa- in the above inequality
tions (refer [14] for details). It turns out that our regularity
criteria imply previous global existence results naturally. If 1/2 − 1/2𝑞 1/2𝑝 1
𝜃= = = . (17)
0 < 𝛽 < 1,𝛼 + 𝛽 ≥ 2 or 𝛼 > 0, 𝛽 = 1 it is proved in [12] 𝛽/2 𝛽/2 𝑝𝛽
𝑢 ∈ 𝐿∞ (0, 𝑇; 𝐻2 (R2 )), 𝑑 ∈ 𝐿2 (0, 𝑇; 𝐻3 (R2 )), the regularity
Applying Gronwall’s inequality and (14), we obtain
criteria in Theorem 2 are satisfied naturally. If 𝛼 = 0, 𝛽 > 1,
one can prove 𝜔 ∈ 𝐿∞ (0, 𝑇; 𝐿𝑝 (R2 )), 2 ≤ 𝑝 ≤ ∞, Δ𝑑 ∈ 𝑇
󵄩 󵄩2 󵄩 󵄩2
(‖𝑢‖2𝐿2 + ‖∇𝑑‖2𝐿2 ) + ∫ 󵄩󵄩󵄩Λ𝛼 𝑢󵄩󵄩󵄩𝐿2 + 󵄩󵄩󵄩󵄩Λ𝛽+1 𝑑󵄩󵄩󵄩󵄩𝐿2 𝑑𝜏 ≤ 𝐶 (𝑇) .
𝐿2 (0, 𝑇; 𝐿𝑝 (R2 )), 2 ≤ 𝑝 < ∞ (refer [11] for details); the 0
regularity criteria (10) in Theorem 3 are satisfied naturally. (18)
Abstract and Applied Analysis 3

Now, we are ready to give the 𝐻1 estimation for (𝑢, ∇𝑑). Here we have used
Multiplying (1) by −Δ𝑢, applying Δ to (2), and testing it by
Δ𝑑, then by using (14), (18), we have ∫ |∇𝑢| ⋅ |Δ𝑑|2 𝑑𝑥
R2
1 𝑑 󵄩 󵄩2 󵄩 󵄩2
(‖∇𝑢‖2𝐿2 + ‖Δ𝑑‖2𝐿2 ) + 󵄩󵄩󵄩󵄩Λ𝛼+1 𝑢󵄩󵄩󵄩󵄩𝐿2 + 󵄩󵄩󵄩󵄩Λ𝛽+2 𝑑󵄩󵄩󵄩󵄩𝐿2 + ∫ |𝑑|2 |Δ𝑑|2 + |Δ𝑑|2 + |𝑑| |∇𝑑|2 |Δ𝑑| 𝑑𝑥
2 𝑑𝑡
R2

= ∫ ∇𝑑 ⋅ Δ𝑑 ⋅ Δ𝑢𝑑𝑥 − ∫ Δ (𝑢 ⋅ ∇𝑑) ⋅ Δ𝑑𝑑𝑥 ≤ 𝐶‖∇𝑢‖𝐿2 ‖Δ𝑑‖2𝐿4 + 𝐶‖𝑑‖2𝐿2𝑝 ‖Δ𝑑‖2𝐿2𝑞

R2 R2

+ ‖𝑑‖𝐿2𝑝 ‖Δ𝑑‖𝐿2𝑞 ‖∇𝑑‖2𝐿4 + ‖Δ𝑑‖2𝐿2

− ∫ Δ𝑓 (𝑑) ⋅ Δ𝑑𝑑𝑥
R2 󵄩 󵄩 󵄩󵄩 𝛽+2 󵄩󵄩2𝜃
≤ 𝐶‖∇𝑢‖2 ‖Δ𝑑‖2/𝛽 󵄩󵄩󵄩󵄩Λ𝛽+2 𝑑󵄩󵄩󵄩󵄩2 +𝐶‖𝑑‖2𝐿2𝑝 ‖Δ𝑑‖2−2𝜃
𝐿2 󵄩
󵄩Λ 𝑑󵄩󵄩 2
󵄩 󵄩𝐿
≤ 𝐶 ∫ |∇𝑢| ⋅ |Δ𝑑|2 𝑑𝑥 󵄩󵄩 𝛽+2 󵄩󵄩𝜃
R2 + 𝐶‖𝑑‖𝐿2𝑝 ‖Δ𝑑‖1−𝜃
𝐿2 󵄩
󵄩Λ 𝑑󵄩󵄩 2 ‖∇𝑑‖2 ‖Δ𝑑‖2 + ‖Δ𝑑‖2𝐿2
󵄩 󵄩𝐿
+ 𝐶 ∫ |𝑑|2 |Δ𝑑|2 + |Δ𝑑|2 + |𝑑| |∇𝑑|2 |Δ𝑑| 𝑑𝑥 1󵄩 󵄩2
R2 ≤ 𝐶‖∇𝑢‖2𝐿2 ‖Δ𝑑‖2𝐿2/𝛽 + 𝐶‖Δ𝑑‖2𝐿2 + 󵄩󵄩󵄩󵄩Λ𝛽+2 𝑑󵄩󵄩󵄩󵄩𝐿2 ,
2
≤ 𝐶‖∇𝑢‖𝐿𝑝 ‖Δ𝑑‖2𝐿2𝑞 + 𝐶‖𝑑‖2𝐿2𝑝 ‖Δ𝑑‖2𝐿2𝑞 0 < 𝛽 < 1,
+ ‖𝑑‖𝐿2𝑝 ‖Δ𝑑‖𝐿2𝑞 ‖∇𝑑‖2𝐿4 + ‖Δ𝑑‖2𝐿2
∫ |∇𝑢| ⋅ |Δ𝑑|2 𝑑𝑥
R2
󵄩󵄩 𝛽+2 󵄩󵄩2𝜃
≤ 𝐶‖∇𝑢‖𝐿𝑝 ‖Δ𝑑‖2−2𝜃
𝐿2 󵄩
󵄩Λ 𝑑󵄩󵄩 2
󵄩 󵄩𝐿
+ ∫ |𝑑|2 |Δ𝑑|2 + |Δ𝑑|2 + |𝑑| |∇𝑑|2 |Δ𝑑| 𝑑𝑥
󵄩󵄩 𝛽+2 󵄩󵄩2𝜃 R2
+ 𝐶‖𝑑‖2𝐿2𝑝 ‖Δ𝑑‖2−2𝜃
𝐿2 󵄩
󵄩Λ 𝑑󵄩󵄩 2
󵄩 󵄩𝐿
≤ ‖∇𝑢‖𝐿2 ‖Δ𝑑‖2𝐿4 + 𝐶‖𝑑‖2𝐿4 ‖Δ𝑑‖2𝐿4
󵄩󵄩 𝛽+2 󵄩󵄩𝜃
+ 𝐶‖𝑑‖𝐿2𝑝 ‖Δ𝑑‖1−𝜃
𝐿2 󵄩
󵄩Λ 𝑑󵄩󵄩 2 ‖∇𝑑‖2 ‖Δ𝑑‖2 + ‖Δ𝑑‖2𝐿2
󵄩 󵄩𝐿 + ‖𝑑‖𝐿4 ‖Δ𝑑‖𝐿4 ‖∇𝑑‖2𝐿4 + ‖Δ𝑑‖2𝐿2
1󵄩 󵄩2 󵄩 󵄩 󵄩 󵄩
≤ 𝐶 (‖∇𝑢‖1/(1−𝜃)
𝐿𝑝 + 1) ‖Δ𝑑‖2𝐿2 + 󵄩󵄩󵄩󵄩Λ𝛽+2 𝑑󵄩󵄩󵄩󵄩𝐿2 . ≤ 𝐶‖∇𝑢‖𝐿2 ‖Δ𝑑‖𝐿2 󵄩󵄩󵄩󵄩Λ3 𝑑󵄩󵄩󵄩󵄩𝐿2 + 𝐶‖𝑑‖2𝐿4 ‖Δ𝑑‖𝐿2 󵄩󵄩󵄩󵄩Λ3 𝑑󵄩󵄩󵄩󵄩𝐿2
2
(19) 󵄩󵄩 3 󵄩󵄩1/2
+ 𝐶‖𝑑‖𝐿4 ‖Δ𝑑‖3/2
𝐿2 󵄩
󵄩Λ 𝑑󵄩󵄩 2 ‖∇𝑑‖𝐿2 + ‖Δ𝑑‖2𝐿2
󵄩 󵄩𝐿
Here we used (16) and the following Galiardo-Nirenberg
1󵄩 󵄩2
inequality: ≤ 𝐶‖∇𝑢‖2𝐿2 ‖Δ𝑑‖2𝐿2 + 𝐶‖Δ𝑑‖2𝐿2 + 󵄩󵄩󵄩󵄩Λ3 𝑑󵄩󵄩󵄩󵄩𝐿2 , 𝛽 ≥ 1.
2
󵄩󵄩 󵄩󵄩 󵄩 󵄩1/2 󵄩 󵄩1/2
󵄩󵄩𝑓󵄩󵄩𝐿4 ≤ 𝐶󵄩󵄩󵄩𝑓󵄩󵄩󵄩𝐿2 󵄩󵄩󵄩∇𝑓󵄩󵄩󵄩𝐿2 . (20) (23)

Thanks to Gronwall’s inequality and (5), we deduce In the above estimation we have used (16), (20), and the
following Galiardo-Nirenberg inequality:
𝑇
󵄩 󵄩2 󵄩 󵄩2
(‖∇𝑢‖2𝐿2 + ‖Δ𝑑‖2𝐿2 ) + ∫ 󵄩󵄩󵄩󵄩Λ𝛽+2 𝑑󵄩󵄩󵄩󵄩𝐿2 + 󵄩󵄩󵄩󵄩Λ𝛼+1 𝑢󵄩󵄩󵄩󵄩𝐿2 𝑑𝜏 ≤ 𝐶 (𝑇) . 󵄩 󵄩
‖Δ𝑑‖2𝐿4 ≤ 𝐶󵄩󵄩󵄩󵄩Λ𝛽+2 𝑑󵄩󵄩󵄩󵄩𝐿2 ‖Δ𝑑‖𝐿2/𝛽 . (24)
0
(21)
Using the condition (6) we get the 𝐻1 estimation (21).
If 𝛽 > 0, we have If 𝛼 > 0, we have
1 𝑑 󵄩 󵄩2 󵄩 󵄩2 1 𝑑 󵄩 󵄩2 󵄩 󵄩2
(‖∇𝑢‖2𝐿2 + ‖Δ𝑑‖2𝐿2 ) + 󵄩󵄩󵄩󵄩Λ𝛼+1 𝑢󵄩󵄩󵄩󵄩𝐿2 + 󵄩󵄩󵄩󵄩Λ𝛽+2 𝑑󵄩󵄩󵄩󵄩𝐿2 (‖∇𝑢‖2𝐿2 + ‖Δ𝑑‖2𝐿2 ) + 󵄩󵄩󵄩󵄩Λ𝛼+1 𝑢󵄩󵄩󵄩󵄩𝐿2 + 󵄩󵄩󵄩󵄩Λ𝛽+2 𝑑󵄩󵄩󵄩󵄩𝐿2
2 𝑑𝑡 2 𝑑𝑡

≤ 𝐶 ∫ |∇𝑢| ⋅ |Δ𝑑|2 𝑑𝑥 ≤ 𝐶 ∫ |∇𝑢| ⋅ |Δ𝑑|2 𝑑𝑥

R2 R2

+ 𝐶 ∫ |𝑑|2 |Δ𝑑|2 + |Δ𝑑|2 + |𝑑| |∇𝑑|2 |Δ𝑑| 𝑑𝑥 + 𝐶 ∫ |𝑑|2 |Δ𝑑|2 + |Δ𝑑|2 + |𝑑| |∇𝑑|2 |Δ𝑑| 𝑑𝑥
R2 R2
(25)
1󵄩 󵄩2 𝐶 (‖Δ𝑑‖2𝐿2/𝛼 + 1) ‖Δ𝑑‖2𝐿2
{
{ 𝐶 (‖∇𝑢‖2𝐿2
+ ‖Δ𝑑‖2𝐿2 ) (‖Δ𝑑‖2𝐿2/𝛽 + 1) + 󵄩󵄩󵄩󵄩Λ𝛽+2 𝑑󵄩󵄩󵄩󵄩𝐿2 , {
{
{ { 2 {
{
{
{ 0 < 𝛽 < 1, {
{ 1 󵄩󵄩 𝛼+1 󵄩󵄩2 1 󵄩󵄩 𝛽+2 󵄩󵄩2
≤{ { + 2 󵄩󵄩󵄩Λ 𝑢󵄩󵄩󵄩𝐿2 + 2 󵄩󵄩󵄩Λ 𝑑󵄩󵄩󵄩𝐿2 , 0 < 𝛼 < 1,
{
{ 1󵄩 󵄩2 ≤{
{
{
{ 𝐶 (‖∇𝑢‖2𝐿2 + 1) ‖Δ𝑑‖2𝐿2 + 󵄩󵄩󵄩󵄩Λ3 𝑑󵄩󵄩󵄩󵄩𝐿2 , { 𝐶‖Δ𝑑‖4𝐿2 + 𝐶‖Δ𝑑‖2𝐿2
{ 2 {
{
{
{
{ 𝛽 ≥ 1. {
{ 1 1 󵄩󵄩 𝛽+2 󵄩󵄩2
2 󵄩 󵄩
(22) { + 2 ‖Δ𝑢‖𝐿2 + 2 󵄩󵄩Λ 𝑑󵄩󵄩𝐿2 , 𝛼 ≥ 1.
4 Abstract and Applied Analysis

Here we have used where

1/2 − 1/𝑟 𝑟 − 2 2
∫ |∇𝑢| ⋅ |Δ𝑑|2 𝑑𝑥 𝜃𝛼 = = , 2≤𝑟≤ ,
𝛼/2 𝑟𝛼 1−𝛼
R2 (29)
𝑟 1 1
+ ∫ |𝑑|2 |Δ𝑑|2 + |Δ𝑑|2 + |𝑑| |∇𝑑|2 |Δ𝑑| 𝑑𝑥 𝑠= , + = 1.
𝑟𝛽 − 1 𝑟 𝑙
R2

󵄩󵄩 𝛽+2 󵄩󵄩2𝜃 We have

≤ 𝐶‖∇𝑢‖𝐿2/(1−𝛼) ‖Δ𝑑‖2𝐿4/(1+𝛼) + 𝐶‖𝑑‖2𝐿2𝑝 ‖Δ𝑑‖2−2𝜃
𝐿2 󵄩
󵄩Λ 𝑑󵄩󵄩 2
󵄩 󵄩𝐿 1 𝑑 󵄩 󵄩2 󵄩 󵄩2
(‖∇𝑢‖2𝐿2 + ‖Δ𝑑‖2𝐿2 ) + 󵄩󵄩󵄩󵄩Λ𝛼+1 𝑢󵄩󵄩󵄩󵄩𝐿2 + 󵄩󵄩󵄩󵄩Λ𝛽+2 𝑑󵄩󵄩󵄩󵄩𝐿2
󵄩󵄩 𝛽+2 󵄩󵄩𝜃 2 𝑑𝑡
+ 𝐶‖𝑑‖𝐿2𝑝 ‖Δ𝑑‖1−𝜃
𝐿2 󵄩
󵄩Λ 𝑑󵄩󵄩 2 ‖∇𝑑‖𝐿2 ‖Δ𝑑‖𝐿2 + ‖Δ𝑑‖2𝐿2
󵄩 󵄩𝐿 ≤ 𝐶‖∇𝑢‖𝐿𝑟 ‖Δ𝑑‖2𝐿2𝑙 + 𝐶‖𝑑‖2𝐿2𝑝 ‖Δ𝑑‖2𝐿2𝑞
󵄩 󵄩 1󵄩 󵄩2
≤ 𝐶󵄩󵄩󵄩󵄩Λ𝛼+1 𝑢󵄩󵄩󵄩󵄩𝐿2 ‖Δ𝑑‖𝐿2 ‖Δ𝑑‖𝐿2/𝛼 + 𝐶‖Δ𝑑‖2𝐿2 + 󵄩󵄩󵄩󵄩Λ𝛽+2 𝑑󵄩󵄩󵄩󵄩𝐿2 + ‖𝑑‖𝐿2𝑝 ‖Δ𝑑‖𝐿2𝑞 ‖∇𝑑‖2𝐿4 + ‖Δ𝑑‖2𝐿2
2
1󵄩 󵄩2 1󵄩 󵄩2 1−𝜃 󵄩 󵄩𝜃𝛼 󵄩 󵄩 󵄩 󵄩
≤ 𝐶 (‖Δ𝑑‖2𝐿2/𝛼 + 1) ‖Δ𝑑‖2𝐿2 + 󵄩󵄩󵄩󵄩Λ𝛼+1 𝑢󵄩󵄩󵄩󵄩𝐿2 + 󵄩󵄩󵄩󵄩Λ𝛽+2 𝑑󵄩󵄩󵄩󵄩𝐿2 , ≤ 𝐶‖∇𝑢‖𝐿2 𝛼 󵄩󵄩󵄩󵄩Λ𝛼+1 𝑢󵄩󵄩󵄩󵄩𝐿2 󵄩󵄩󵄩󵄩Λ𝛽+1 𝑑󵄩󵄩󵄩󵄩𝐿𝑠 󵄩󵄩󵄩󵄩Λ𝛽+2 𝑑󵄩󵄩󵄩󵄩𝐿2
2 2
󵄩󵄩 𝛽+2 󵄩󵄩2𝜃 (30)
0 < 𝛼 < 1, + 𝐶‖Δ𝑑‖2−2𝜃
𝐿2 󵄩
󵄩Λ 𝑑󵄩󵄩 2
󵄩 󵄩𝐿
󵄩󵄩 𝛽+2 󵄩󵄩𝜃
∫ |∇𝑢| ⋅ |Δ𝑑|2 𝑑𝑥 + 𝐶‖Δ𝑑‖1−𝜃
𝐿2 󵄩
󵄩Λ 𝑑󵄩󵄩 2 ‖∇𝑑‖𝐿2 ‖Δ𝑑‖𝐿2 + ‖Δ𝑑‖2𝐿2
󵄩 󵄩𝐿
R2
󵄩 󵄩2/(1−𝜃1 )
+ ∫ |𝑑|2 |Δ𝑑|2 + |Δ𝑑|2 + |𝑑| |∇𝑑|2 |Δ𝑑| 𝑑𝑥 ≤ 𝐶 (‖∇𝑢‖2𝐿2 + ‖Δ𝑑‖2𝐿2 ) (󵄩󵄩󵄩󵄩Λ𝛽+1 𝑑󵄩󵄩󵄩󵄩𝐿𝑠 + 1)
R2
1󵄩 󵄩2 1󵄩 󵄩2
≤ 𝐶 (‖∇𝑢‖BMO ‖Δ𝑑‖𝐿2 + ‖∇𝑑‖BMO ‖Δ𝑢‖𝐿2 ) ‖Δ𝑑‖𝐿2 + 󵄩󵄩󵄩󵄩Λ𝛼+1 𝑢󵄩󵄩󵄩󵄩𝐿2 + 󵄩󵄩󵄩󵄩Λ𝛽+2 𝑑󵄩󵄩󵄩󵄩𝐿2 .
2 2
󵄩󵄩 𝛽+2 󵄩󵄩2𝜃 Thanks to condition (8) we have the 𝐻1 estimation (21). Now
+ 𝐶‖𝑑‖2𝐿2𝑝 ‖Δ𝑑‖2−2𝜃
𝐿2 󵄩
󵄩Λ 𝑑󵄩󵄩 2
󵄩 󵄩𝐿 we complete the 𝐻1 estimation.
󵄩󵄩 𝛽+2 󵄩󵄩𝜃 Since 𝐻1 󳨅→ BMO in 2D, if ∇𝑢, Δ𝑑 ∈ 𝐿∞ (0, 𝑇; 𝐻1 (R2 )),
+ 𝐶‖𝑑‖𝐿2𝑝 ‖Δ𝑑‖1−𝜃
𝐿2 󵄩
󵄩Λ 𝑑󵄩󵄩 2 ‖∇𝑑‖𝐿2 ‖Δ𝑑‖𝐿2 + ‖Δ𝑑‖2𝐿2
󵄩 󵄩𝐿 then we can deduce that ∇𝑢, Δ𝑑 ∈ 𝐿∞ (0, 𝑇; BMO).
Applying Δ to (1) and testing by Δ𝑢, applying Δ to (1), and
1 1󵄩 󵄩2
≤ 𝐶‖Δ𝑑‖4𝐿2 + 𝐶‖Δ𝑑‖2𝐿2 + ‖Δ𝑢‖2𝐿2 + 󵄩󵄩󵄩󵄩Λ𝛽+2 𝑑󵄩󵄩󵄩󵄩𝐿2 , testing by −ΔΔ𝑑, after suitable integration by parts we obtain
2 2
1 𝑑 󵄩 󵄩2 󵄩 󵄩2
𝛼 ≥ 1. (‖Δ𝑢‖22 + ‖∇Δ𝑑‖22 ) + 󵄩󵄩󵄩󵄩Λ𝛼+2 𝑢󵄩󵄩󵄩󵄩2 + 󵄩󵄩󵄩󵄩Λ𝛽+3 𝑑󵄩󵄩󵄩󵄩2
2 𝑑𝑡
(26)
= − ∫ Δ (𝑢 ⋅ ∇𝑢) ⋅ Δ𝑢𝑑𝑥 − ∫ Δ (∇𝑑 ⋅ Δ𝑑) ⋅ Δ𝑢𝑑𝑥
Here and in the following we use the embeddings 𝐻𝑠 (R2 ) 󳨅→ 𝑅2 𝑅2

𝐿𝑝 (R2 ) for 𝑠 < 𝑛/2 and 1/𝑝 = 1/2 − 𝑠/𝑛, 𝐻1 (R2 ) 󳨅→ BMO
and the commutator estimate given in [15] + ∫ Δ (𝑢 ⋅ ∇𝑑) ⋅ ΔΔ𝑑𝑑𝑥 + ∫ Δ𝑓 (𝑑) ⋅ ΔΔ𝑑𝑑𝑥
𝑅2 𝑅2
󵄩󵄩 𝛼 󵄩 󵄩 󵄩 󵄩 󵄩
󵄩󵄩𝜕 𝑓 ⋅ 𝜕𝛽 𝑔󵄩󵄩󵄩 ≤ 𝐶 (󵄩󵄩󵄩󵄩𝑓󵄩󵄩󵄩󵄩BMO 󵄩󵄩󵄩Λ|𝛼|+|𝛽| 𝑔󵄩󵄩󵄩 + 󵄩󵄩󵄩󵄩𝑔󵄩󵄩󵄩󵄩BMO 󵄩󵄩󵄩Λ|𝛼|+|𝛽| 𝑓󵄩󵄩󵄩 ) , ≤ 𝐶 ∫ |∇𝑢| ⋅ |Δ𝑢|2 𝑑𝑥 + 𝐶 ∫ |Δ𝑢| ⋅ |Δ𝑑| ⋅ |∇Δ𝑑| 𝑑𝑥
󵄩 󵄩𝑟 󵄩 󵄩𝑟 󵄩 󵄩𝑟
𝑅2 𝑅2
(27)
󵄨 󵄨
+ 𝐶 ∫ |∇𝑢| ⋅ |∇Δ𝑑|2 𝑑𝑥 + ∫ 󵄨󵄨󵄨∇Δ𝑓 (𝑑)󵄨󵄨󵄨 ⋅ |∇Δ𝑑| 𝑑𝑥
usually. Thanks to condition (7) we have the 𝐻1 estimation 𝑅2 𝑅2
(21).
If in addition, 1 > 𝛽 ≥ 1/2, 𝛼 > 0, use (16), := 𝐼1 + 𝐼2 + 𝐼3 + 𝐼4 .
(31)
1−𝜃 󵄩 󵄩𝜃𝛼
‖∇𝑢‖𝐿𝑟 ≤ ‖∇𝑢‖𝐿2 𝛼 󵄩󵄩󵄩󵄩Λ𝛼+1 𝑢󵄩󵄩󵄩󵄩𝐿2 , If 𝛼 ≥ 1/2, 𝛽 ≥ 1/2, we estimate the right hand side of (31)
one by one:
1 1 𝛼 1 − 𝜃𝛼
= ( − ) 𝜃𝛼 + ,
𝑟 2 2 2 𝐼1 = 𝐶 ∫ |∇𝑢| ⋅ |Δ𝑢|2 𝑑𝑥
𝑅2
0 ≤ 𝜃𝛼 ≤ 1, (28)
≤ 𝐶‖∇𝑢‖𝐿3 ‖Δ𝑢‖2𝐿3
󵄩 󵄩1/2 󵄩 󵄩1/2
‖Δ𝑑‖𝐿2𝑙 ≤ 𝐶󵄩󵄩󵄩󵄩Λ𝛽+1 𝑑󵄩󵄩󵄩󵄩𝐿𝑠 󵄩󵄩󵄩󵄩Λ2+𝛽 𝑑󵄩󵄩󵄩󵄩𝐿2 , ≤ 𝐶‖∇𝑢‖𝐿3 ‖Δ𝑢‖4/3 ‖Δ𝑢‖2/3
𝐿3 𝐿3
1 1−𝛽 1 󵄩󵄩 1/2 󵄩󵄩2/9 󵄩󵄩 1/2 󵄩󵄩2/9 󵄩󵄩 1/2 󵄩󵄩14/9
= + , ≤ 𝐶‖∇𝑢‖7/9 󵄩Λ Δ𝑢󵄩󵄩 2 󵄩󵄩Λ ∇𝑢󵄩󵄩 2 ‖Δ𝑢‖2/9 󵄩Λ Δ𝑢󵄩󵄩 2
2𝑙 2 2𝑠 𝐿2 󵄩 󵄩 󵄩𝐿 󵄩 󵄩𝐿 𝐿2 󵄩 󵄩 󵄩𝐿
Abstract and Applied Analysis 5

󵄩󵄩 1/2 󵄩󵄩2/9 󵄩󵄩 1/2 󵄩󵄩16/9 ≤ 𝐶‖𝑑‖𝐿2𝑝 ‖𝑑‖1/2 ‖∇Δ𝑑‖1/2 ‖Δ𝑑‖1/2 ‖∇Δ𝑑‖1/2 ‖∇Δ𝑑‖1−𝜃
≤ 𝐶‖∇𝑢‖7/9
𝐿2 󵄩
󵄩Λ ∇𝑢󵄩󵄩 2 ‖Δ𝑢‖2/9
󵄩 󵄩𝐿 𝐿2 󵄩
󵄩Λ Δ𝑢󵄩󵄩 2
󵄩 󵄩𝐿 𝐿2 𝐿2 𝐿2 𝐿2 𝐿2

󵄩 󵄩𝜃 󵄩󵄩 𝛽+3 󵄩󵄩2𝜃
󵄩 󵄩2 󵄩 󵄩2
≤ 𝐶 (𝜖) ‖∇𝑢‖7𝐿2 󵄩󵄩󵄩󵄩Λ1/2 ∇𝑢󵄩󵄩󵄩󵄩𝐿2 ‖Δ𝑢‖2𝐿2 + 𝜖󵄩󵄩󵄩󵄩Λ1/2 Δ𝑢󵄩󵄩󵄩󵄩𝐿2 , × 󵄩󵄩󵄩󵄩Λ𝛽+3 𝑑󵄩󵄩󵄩󵄩𝐿2 + 𝐶‖𝑑‖2𝐿2𝑝 ‖∇Δ𝑑‖2−2𝜃
𝐿2 󵄩
󵄩Λ 𝑑󵄩󵄩 2 + ‖∇Δ𝑑‖2𝐿2
󵄩 󵄩𝐿
󵄩 󵄩2
≤ 𝐶‖∇Δ𝑑‖2𝐿2 + 𝜖󵄩󵄩󵄩󵄩Λ𝛽+3 𝑑󵄩󵄩󵄩󵄩𝐿2 .
𝐼2 = 𝐶 ∫ |Δ𝑢| ⋅ |Δ𝑑| ⋅ |∇Δ𝑑| 𝑑𝑥
𝑅2 (35)
≤ 𝐶‖Δ𝑑‖𝐿3 ‖∇Δ𝑑‖𝐿3 ‖Δ𝑢‖𝐿3
Finally, putting the above results together, we deduce that
󵄩󵄩 1/2 󵄩󵄩1/9 󵄩󵄩 1/2 󵄩
≤ 𝐶‖Δ𝑑‖7/9
𝐿2 󵄩
󵄩Λ Δ𝑑󵄩󵄩 2 ‖∇Δ𝑑‖1/9
𝐿2 󵄩
󵄩Λ ∇Δ𝑑󵄩󵄩󵄩 2
󵄩 󵄩𝐿 󵄩 󵄩𝐿 𝑑 󵄩 󵄩2 󵄩 󵄩2
(‖Δ𝑢‖2𝐿2 + ‖∇Δ𝑑‖2𝐿2 ) + 󵄩󵄩󵄩󵄩Λ𝛼+2 𝑢󵄩󵄩󵄩󵄩𝐿2 + 󵄩󵄩󵄩󵄩Λ𝛽+3 𝑑󵄩󵄩󵄩󵄩𝐿2
󵄩 󵄩1/9 󵄩󵄩 1/2 󵄩󵄩7/9 𝑑𝑡
× 󵄩󵄩󵄩󵄩Λ1/2 ∇𝑢󵄩󵄩󵄩󵄩𝐿2 ‖Δ𝑢‖1/9
𝐿2 󵄩
󵄩Λ Δ𝑢󵄩󵄩 2
󵄩 󵄩𝐿
≤ 𝐶 (‖∇𝑢‖7𝐿2 + ‖Δ𝑑‖7𝐿2 )
󵄩 󵄩2 󵄩 󵄩2
≤ 𝐶 (𝜖) ‖Δ𝑑‖7𝐿2 (󵄩󵄩󵄩󵄩Λ1/2 Δ𝑑󵄩󵄩󵄩󵄩𝐿2 ‖∇Δ𝑑‖2𝐿2 + 󵄩󵄩󵄩󵄩Λ1/2 ∇𝑢󵄩󵄩󵄩󵄩𝐿2 ‖Δ𝑢‖2𝐿2 ) 󵄩 󵄩2 󵄩 󵄩2
(36)
× (󵄩󵄩󵄩󵄩Λ1/2 ∇𝑢󵄩󵄩󵄩󵄩𝐿2 ‖Δ𝑢‖2𝐿2 + 󵄩󵄩󵄩󵄩Λ1/2 Δ𝑑󵄩󵄩󵄩󵄩𝐿2 ‖∇Δ𝑑‖2𝐿2
󵄩 󵄩2 󵄩 󵄩2
+ 𝜖 (󵄩󵄩󵄩󵄩Λ1/2 ∇Δ𝑑󵄩󵄩󵄩󵄩𝐿2 + 󵄩󵄩󵄩󵄩Λ1/2 Δ𝑢󵄩󵄩󵄩󵄩𝐿2 ) ,
+‖∇Δ𝑑‖2𝐿2 ) .
𝐼3 = 𝐶 ∫ |∇𝑢| ⋅ |∇Δ𝑑|2 𝑑𝑥
𝑅2 By using Gronwall’s inequality and (21), we obtain

≤ 𝐶‖∇𝑢‖𝐿3 ‖∇Δ𝑑‖2𝐿3 𝑡
󵄩 󵄩2 󵄩 󵄩2
(‖Δ𝑢‖2𝐿2 + ‖∇Δ𝑑‖2𝐿2 ) + ∫ 󵄩󵄩󵄩󵄩Λ𝛼+2 𝑢󵄩󵄩󵄩󵄩𝐿2 + 󵄩󵄩󵄩󵄩Λ𝛽+3 𝑑󵄩󵄩󵄩󵄩𝐿2 𝑑𝑡 ≤ 𝐶 (𝑇) .
≤ 𝐶‖∇𝑢‖𝐿3 ‖∇Δ𝑑‖4/3
𝐿3
‖∇Δ𝑑‖2/3
𝐿3
0
(37)
󵄩 󵄩2
≤ 𝐶 (𝜖) ‖∇𝑢‖7𝐿2 󵄩󵄩󵄩󵄩Λ1/2 Δ𝑑󵄩󵄩󵄩󵄩𝐿2 ‖∇Δ𝑑‖2𝐿2 The proof of Theorem 1 is finished.
󵄩 󵄩2 󵄩 󵄩2
+ 𝜖 (󵄩󵄩󵄩󵄩Λ1/2 ∇Δ𝑑󵄩󵄩󵄩󵄩𝐿2 + 󵄩󵄩󵄩󵄩Λ1/2 Δ𝑢󵄩󵄩󵄩󵄩𝐿2 ) .
3. Proof of Theorem 2
(32)
Now, we give the proof of Theorem 2.
Here we used the following Galiardo-Nirenberg inequalities:
Proof. This section focuses on the case with 0 < 𝛼 < 1/2,
󵄩󵄩 󵄩󵄩 󵄩 1/2 󵄩1/6 󵄩 1/2 󵄩5/6
󵄩󵄩∇𝑓󵄩󵄩𝐿3 ≤ 𝐶󵄩󵄩󵄩󵄩Λ 𝑓󵄩󵄩󵄩󵄩𝐿2 󵄩󵄩󵄩󵄩Λ ∇𝑓󵄩󵄩󵄩󵄩𝐿2 , 𝛽 > 0 or 0 < 𝛽 < 1/2, 𝛼 > 0. Since we cannot estimate
(37) as that in Theorem 1, so we should give the estimation of
󵄩󵄩 󵄩󵄩 󵄩 󵄩1/3 󵄩 1/2 󵄩2/3 𝐼𝑖 , 𝑖 = 1, 2, 3, 4, 5 which is defined in Section 2. Using (16), we
󵄩󵄩∇𝑓󵄩󵄩𝐿3 ≤ 𝐶󵄩󵄩󵄩∇𝑓󵄩󵄩󵄩𝐿2 󵄩󵄩󵄩󵄩Λ ∇𝑓󵄩󵄩󵄩󵄩𝐿2 , have
󵄩󵄩 󵄩󵄩 󵄩 󵄩2/3 󵄩 󵄩1/3 󵄩 1/2 󵄩1/9 󵄩 󵄩1/9 󵄩 1/2 󵄩7/9
󵄩󵄩∇𝑓󵄩󵄩𝐿3 = 󵄩󵄩󵄩∇𝑓󵄩󵄩󵄩𝐿3 󵄩󵄩󵄩∇𝑓󵄩󵄩󵄩𝐿3 ≤ 𝐶󵄩󵄩󵄩󵄩Λ 𝑓󵄩󵄩󵄩󵄩𝐿2 󵄩󵄩󵄩∇𝑓󵄩󵄩󵄩𝐿2 󵄩󵄩󵄩󵄩Λ ∇𝑓󵄩󵄩󵄩󵄩𝐿2 , 𝐼1 = 𝐶 ∫ |∇𝑢| ⋅ |Δ𝑢|2 𝑑𝑥 ≤ 𝐶‖∇𝑢‖𝐿𝑝 ‖Δ𝑢‖2𝐿2𝑞
R2
󵄩󵄩 󵄩󵄩 󵄩 󵄩7/9 󵄩 1/2 󵄩2/9
󵄩󵄩𝑓󵄩󵄩𝐿3 ≤ 𝐶󵄩󵄩󵄩𝑓󵄩󵄩󵄩𝐿2 󵄩󵄩󵄩󵄩Λ ∇𝑓󵄩󵄩󵄩󵄩𝐿2 . 2(1−𝜃𝛼 ) 󵄩
󵄩 2+𝛼 󵄩2𝜃𝛼
≤ 𝐶‖∇𝑢‖𝐿𝑝 ‖Δ𝑢‖𝐿2 󵄩󵄩Λ
󵄩 𝑢󵄩󵄩󵄩󵄩𝐿2 (38)
(33)
󵄩 󵄩2
‖Δ𝑢‖2𝐿2 + 𝜖󵄩󵄩󵄩󵄩Λ2+𝛼 𝑢󵄩󵄩󵄩󵄩𝐿2 ,
1/(1−𝜃𝛼 )
Using (14), (16), (20), (21), and the following Galiardo- ≤ 𝐶‖∇𝑢‖𝐿𝑝
Nirenberg inequality,
where 1/𝑝 + 1/𝑞 = 1 and 𝜃𝛼 = 1/𝑝𝛼. We use the same way to
‖∇𝑑‖𝐿4 ≤ ‖𝑑‖1/2 ‖∇Δ𝑑‖1/2 , (34) estimate 𝐼2 and 𝐼3 .
𝐿2 𝐿2

we have 𝐼2 = 𝐶 ∫ |Δ𝑢| ⋅ |Δ𝑑| ⋅ |∇Δ𝑑| 𝑑𝑥

R2
󵄨 󵄨
𝐼4 = ∫ 󵄨󵄨󵄨∇Δ𝑓 (𝑑)󵄨󵄨󵄨 ⋅ |∇Δ𝑑| 𝑑𝑥 ≤ 𝐶‖Δ𝑑‖𝐿𝑝 ‖∇Δ𝑑‖𝐿2𝑞 ‖Δ𝑢‖𝐿2𝑞
R2

≤ 𝐶 ∫ (|𝑑|2 |∇Δ𝑑|2 + |𝑑| |∇𝑑| |Δ𝑑| |∇Δ𝑑|) 𝑑𝑥 + ‖∇Δ𝑑‖2𝐿2 ≤ 𝐶‖Δ𝑑‖1/2 1/2

𝐿𝑝 ‖∇Δ𝑑‖𝐿2𝑞 ‖Δ𝑑‖𝐿𝑝 ‖Δ𝑢‖𝐿2𝑞
𝑅2
1/(1−𝜃𝛽 ) 1/(1−𝜃𝛼 )
≤ 𝐶 (‖Δ𝑑‖𝐿𝑝 ‖∇Δ𝑑‖2𝐿2 + ‖Δ𝑑‖𝐿𝑝 ‖Δ𝑢‖2𝐿2 )
≤ 𝐶‖𝑑‖2𝐿2𝑝 ‖∇Δ𝑑‖2𝐿2𝑞 + 𝐶‖𝑑‖𝐿2𝑝 ‖∇𝑑‖𝐿4 ‖Δ𝑑‖𝐿4 ‖∇Δ𝑑‖𝐿2𝑞
󵄩 󵄩2 󵄩 󵄩2
+ ‖∇Δ𝑑‖2𝐿2 + 𝜖 (󵄩󵄩󵄩󵄩Λ2+𝛼 𝑢󵄩󵄩󵄩󵄩𝐿2 + 󵄩󵄩󵄩󵄩Λ3+𝛽 𝑑󵄩󵄩󵄩󵄩𝐿2 ) ;
6 Abstract and Applied Analysis

𝐼3 = 𝐶 ∫ |∇𝑢| ⋅ |∇Δ𝑑|2 𝑑𝑥 Now, we give the 𝐻1 estimation for (∇𝑢, Δ𝑑). We should
R2 give the estimation of 𝐼𝑖 , 𝑖 = 1, 2, 3, 4 which is defined in
Section 2:
≤ 𝐶‖∇𝑢‖𝐿𝑝 ‖∇Δ𝑑‖2𝐿2𝑞
2−2𝜃𝛽 󵄩
󵄩 3+𝛽 󵄩2𝜃𝛽
≤ 𝐶‖∇𝑢‖𝐿𝑝 ‖∇Δ𝑑‖𝐿2 󵄩󵄩Λ
󵄩 𝑑󵄩󵄩󵄩󵄩𝐿2 𝐼1 = 𝐶 ∫ |∇𝑢| ⋅ |Δ𝑢|2 𝑑𝑥 ≤ 𝐶‖∇𝑢‖BMO ‖Δ𝑢‖2𝐿2 ,
R2
󵄩 󵄩2
𝜖󵄩󵄩󵄩󵄩Λ3+𝛽 𝑑󵄩󵄩󵄩󵄩𝐿2 ,
1/(1−𝜃𝛽 )
≤ 𝐶‖∇𝑢‖𝐿𝑝 ‖∇Δ𝑑‖2𝐿2 + 𝐼2 = 𝐶 ∫ |Δ𝑢| ⋅ |Δ𝑑| ⋅ |∇Δ𝑑| 𝑑𝑥
R2
(39)
󵄩 󵄩
{ 𝐶 ‖Δ𝑑‖𝐿2/𝛽 󵄩󵄩󵄩󵄩Λ𝛽+3 𝑑󵄩󵄩󵄩󵄩𝐿2 ‖Δ𝑢‖𝐿2
{
{
{ 󵄩 󵄩2
The same as that in Section 2, we have ≤ { ≤ 𝐶‖Δ𝑑‖2𝐿2/𝛽 ‖Δ𝑢‖2𝐿2 + 𝜖󵄩󵄩󵄩󵄩Λ𝛽+3 𝑑󵄩󵄩󵄩󵄩𝐿2 , 0 < 𝛽 < 1,
{
{
{ 2 2 󵄩󵄩 2 󵄩󵄩2
󵄨 󵄨 {𝐶‖Δ𝑑‖𝐿2 ‖Δ𝑢‖𝐿2 + 𝜖󵄩󵄩󵄩∇ Δ𝑑󵄩󵄩󵄩𝐿2 , 𝛽 ≥ 1.
𝐼4 = ∫ 󵄨󵄨󵄨∇Δ𝑓 (𝑑)󵄨󵄨󵄨 ⋅ |∇Δ𝑑| 𝑑𝑥
R 2

𝐼3 = 𝐶 ∫ |∇𝑢| ⋅ |∇Δ𝑑|2 𝑑𝑥
≤ 𝐶‖𝑑‖2𝐿2𝑝 ‖∇Δ𝑑‖2𝐿2𝑞 R2
(40) 󵄩 󵄩
+ 𝐶‖𝑑‖𝐿2𝑝 ‖∇𝑑‖𝐿4 ‖Δ𝑑‖𝐿4 ‖∇Δ𝑑‖𝐿2𝑞 + ‖∇Δ𝑑‖2𝐿2 { ‖∇𝑢‖𝐿2/𝛽 󵄩󵄩󵄩󵄩Λ𝛽+3 𝑑󵄩󵄩󵄩󵄩𝐿2 ‖∇Δ𝑑‖𝐿2
{
{
{ 󵄩 󵄩2
󵄩 󵄩2 ≤ { ≤ 𝐶‖∇𝑢‖2𝐿2/𝛽 ‖∇Δ𝑑‖2𝐿2 + 𝜖󵄩󵄩󵄩󵄩Λ𝛽+3 𝑑󵄩󵄩󵄩󵄩𝐿2 , 0 < 𝛽 < 1,
≤ 𝐶‖∇Δ𝑑‖2𝐿2 + 𝜖󵄩󵄩󵄩󵄩Λ𝛽+3 𝑑󵄩󵄩󵄩󵄩𝐿2 . {
{
{ 󵄩󵄩 2 󵄩󵄩2
2 2
{𝐶‖∇𝑢‖𝐿2 ‖∇Δ𝑑‖𝐿2 + 𝜖󵄩󵄩󵄩∇ Δ𝑑󵄩󵄩󵄩𝐿2 , 𝛽 ≥ 1.
(43)
Finally, putting the above results together, we have

1 𝑑 󵄩 󵄩2 󵄩 󵄩2 The same as that in Section 2, we have

(‖Δ𝑢‖2𝐿2 + ‖∇Δ𝑑‖2𝐿2 ) + 󵄩󵄩󵄩󵄩Λ2+𝛼 𝑢󵄩󵄩󵄩󵄩𝐿2 + 󵄩󵄩󵄩󵄩Λ3+𝛽 𝑑󵄩󵄩󵄩󵄩𝐿2
2 𝑑𝑡
󵄨 󵄨
1/(1−𝜃𝛽 ) 1/(1−𝜃𝛽 ) 1/(1−𝜃𝛼 ) 𝐼4 = ∫ 󵄨󵄨󵄨∇Δ𝑓 (𝑑)󵄨󵄨󵄨 ⋅ |∇Δ𝑑| 𝑑𝑥
≤ 𝐶 (‖Δ𝑑‖𝐿𝑝 + ‖∇𝑢‖𝐿𝑝 + ‖Δ𝑑‖𝐿𝑝 R2
(44)
1󵄩 󵄩2
1/(1−𝜃 )
+‖∇𝑢‖𝐿𝑝 𝛼 + 1) ≤ 𝐶‖∇Δ𝑑‖2𝐿2 + 󵄩󵄩󵄩󵄩Λ𝛽+3 𝑑󵄩󵄩󵄩󵄩𝐿2 .
2
󵄩 󵄩2 󵄩 󵄩2
× (‖∇Δ𝑑‖2𝐿2 + ‖Δ𝑢‖2𝐿2 ) + 𝜖 (󵄩󵄩󵄩󵄩Λ2+𝛼 𝑢󵄩󵄩󵄩󵄩𝐿2 + 󵄩󵄩󵄩󵄩Λ3+𝛽 𝑑󵄩󵄩󵄩󵄩𝐿2 ) . Putting the above results together, we deduce
(41)
1 𝑑 󵄩 󵄩2 󵄩 󵄩2
(‖Δ𝑢‖2𝐿2 + ‖∇Δ𝑢‖2𝐿2 ) + 󵄩󵄩󵄩󵄩Λ2+𝛼 𝑢󵄩󵄩󵄩󵄩𝐿2 + 󵄩󵄩󵄩󵄩Λ3+𝛽 𝑑󵄩󵄩󵄩󵄩𝐿2
Using Gronwall’s inequality and (9) we can deduce (37). 2 𝑑𝑡
The proof is finished.
≤ 𝐶 (‖∇𝑢‖BMO + ‖Δ𝑑‖2𝐿󰜚 + ‖∇𝑢‖2𝐿󰜚 + 1)
(45)
4. Proof of Theorem 3 × (‖Δ𝑢‖2𝐿2 + ‖∇Δ𝑑‖2𝐿2 )
1
Proof. For 𝛼 = 0, 𝛽 > 0, firstly we give the 𝐻 estimation for 󵄩 󵄩2 󵄩 󵄩2
(𝑢, ∇𝑑): + 𝜖 (󵄩󵄩󵄩󵄩Λ2+𝛼 𝑢󵄩󵄩󵄩󵄩𝐿2 + 󵄩󵄩󵄩󵄩Λ3+𝛽 𝑑󵄩󵄩󵄩󵄩𝐿2 ) .

1 𝑑 󵄩 󵄩2
(‖∇𝑢‖2𝐿2 + ‖Δ𝑑‖2𝐿2 ) + 󵄩󵄩󵄩󵄩Λ𝛽+2 𝑑󵄩󵄩󵄩󵄩𝐿2
Applying Gronwall’s inequality and (10) we obtain (37).
2 𝑑𝑡 For 𝛼 > 0, 𝛽 = 0, firstly we give the estimation for (𝑢, ∇𝑑):
2 2 2
{𝐶 (‖∇𝑢‖𝐿2 + ‖Δ𝑑‖𝐿2 ) (‖Δ𝑑‖𝐿2/𝛽 + 1)
{
{
{ 1 𝑑 󵄩 󵄩2
{
{ 1󵄩 󵄩2 (‖𝑢‖2𝐿2 + ‖∇𝑑‖2𝐿2 ) + 󵄩󵄩󵄩󵄩Λ𝛽+1 𝑑󵄩󵄩󵄩󵄩𝐿2
{
{
{ + 󵄩󵄩󵄩󵄩Λ𝛽+2 𝑑󵄩󵄩󵄩󵄩𝐿2 , (42) 2 𝑑𝑡
{ 2
≤{ 0 < 𝛽 < 1,
{
{ = − ∫ ∇𝑓 (𝑑) ∇𝑑𝑑𝑥
{
{ 1󵄩 󵄩2
{
{
{𝐶 (‖∇𝑢‖2𝐿2 + 1) ‖Δ𝑑‖2𝐿2 + 󵄩󵄩󵄩󵄩Λ3 𝑑󵄩󵄩󵄩󵄩𝐿2 , R2 (46)
{
{ 2
{ 𝛽 ≥ 1. ≤ 𝐶 ∫ |𝑑|2 |∇𝑑|2 𝑑𝑥 + ‖∇𝑑‖2𝐿2
R2

Then by using Gronwall’s inequality and (10) we obtain (21). ≤ 𝐶‖𝑑‖2BMO ‖∇𝑑‖2𝐿2 + ‖∇𝑑‖2𝐿2
Abstract and Applied Analysis 7

and the 𝐻1 estimation for (𝑢, ∇𝑑): Finally, putting the above results together, we deduce

1 𝑑 󵄩 󵄩2 󵄩 󵄩2
1 𝑑 󵄩 󵄩2
(‖∇𝑢‖2𝐿2 + ‖Δ𝑑‖2𝐿2 ) + 󵄩󵄩󵄩󵄩Λ𝛼+1 𝑢󵄩󵄩󵄩󵄩𝐿2 (‖Δ𝑢‖22 + ‖∇Δ𝑑‖22 ) + 󵄩󵄩󵄩󵄩Λ𝛼+2 𝑢󵄩󵄩󵄩󵄩𝐿2 + 󵄩󵄩󵄩󵄩Λ𝛽+3 𝑑󵄩󵄩󵄩󵄩𝐿2
2 𝑑𝑡 2 𝑑𝑡

≤ 𝐶‖∇𝑢‖𝐿∞ ‖Δ𝑑‖2𝐿2 ≤ 𝐶 (‖∇𝑢‖𝐿∞ + ‖Δ𝑑‖2/(2−𝜃)

𝐿𝑝 + ‖𝑑‖2BMO + 1)
(52)
+ 𝐶 ∫ |𝑑|2 |Δ𝑑|2 + |𝑑| |∇𝑑| |∇𝑑| |Δ𝑑| 𝑑𝑥 + ‖Δ𝑑‖2𝐿2 × (‖∇Δ𝑑‖2𝐿2 + ‖Δ𝑢‖2𝐿2 )
R2
󵄩 󵄩2 󵄩 󵄩2
+ 𝜖 (󵄩󵄩󵄩󵄩Λ𝛼+2 𝑢󵄩󵄩󵄩󵄩𝐿2 + 󵄩󵄩󵄩󵄩Λ𝛽+3 𝑑󵄩󵄩󵄩󵄩𝐿2 ) .
≤ 𝐶‖∇𝑢‖𝐿∞ ‖Δ𝑑‖2𝐿2 + ‖𝑑‖2BMO ‖Δ𝑑‖2𝐿2 + ‖Δ𝑑‖2𝐿2

≤ 𝐶 (‖∇𝑢‖𝐿∞ + ‖𝑑‖2BMO + 1) ‖Δ𝑑‖2𝐿2 . By using Gronwall’s inequality and (11) we obtain (37). Now
we complete our proof.
(47)

The regularity criteria (11) can guarantee estimations (18) and Conflict of Interests
(21). The authors declare that there is no conflict of interests
Now, we give the estimation of 𝐼𝑖 , 𝑖 = 1, 2, 3, 4 which is regarding the publication of this paper.
defined in Section 2:

Acknowledgments
𝐼1 = 𝐶 ∫ |∇𝑢| ⋅ |Δ𝑢|2 𝑑𝑥 ≤ 𝐶‖∇𝑢‖𝐿∞ ‖Δ𝑢‖2𝐿2 ,
R2
The authors thank Professor Yong Zhou for patient guidance
2 and helpful discussion. This work is partially supported by
𝐼3 = 𝐶 ∫ |∇𝑢| ⋅ |∇Δ𝑑| 𝑑𝑥 ≤ 𝐶‖∇𝑢‖𝐿∞ ‖∇Δ𝑑‖2𝐿2 ,
R2 NSFC (Grant no. 11101376), NSFC (Grant no. 11226176), and
ZJNSF (Grant no. LQ13A010008).
󵄨 󵄨
𝐼4 = ∫ 󵄨󵄨󵄨∇Δ𝑓 (𝑑)󵄨󵄨󵄨 ⋅ |∇Δ𝑑| 𝑑𝑥
R2
References
2 2
≤ 𝐶 ∫ (|𝑑| |∇Δ𝑑| + |𝑑| |∇𝑑| |Δ𝑑| |∇Δ𝑑|) 𝑑𝑥 + ‖∇Δ𝑑‖2𝐿2 [1] P. G. de Gennes, The Physics of Liquid Crystals, Oxford Univer-
R2
sity Press, Oxford, UK, 1974.
≤ 𝐶 (‖𝑑‖2BMO + 1) ‖∇Δ𝑑‖2𝐿2 . [2] J. L. Ericksen, “Hydrostatic theory of liquid crystals,” Archive for
(48) Rational Mechanics and Analysis, vol. 9, pp. 371–378, 1962.
[3] F. M. Leslie, “Some constitutive equations for liquid crystals,”
Archive for Rational Mechanics and Analysis, vol. 28, no. 4, pp.
Using the following Galiardo-Nirenberg inequality,
265–283, 1968.
[4] D. Coutand and S. Shkoller, “Well-posedness of the full
󵄩󵄩 𝛼+2 󵄩󵄩𝜃
‖Δ𝑢‖𝐿𝑞 ≤ 𝐶‖Δ𝑢‖1−𝜃
𝐿2 󵄩
󵄩Λ 𝑢󵄩󵄩 2 ,
󵄩 󵄩𝐿
Ericksen-Leslie model of nematic liquid crystals,” Comptes
Rendus de l’Académie des Sciences I: Mathematics, vol. 333, no.
1 1 𝛼 1−𝜃 10, pp. 919–924, 2001.
= ( − )𝜃 + , (49)
𝑞 2 2 2 [5] F. Lin and C. Liu, “Nonparabolic dissipative systems modeling
the flow of liquid crystals,” Communications on Pure and Applied
0 ≤ 𝜃 ≤ 1, Mathematics, vol. 48, no. 5, pp. 501–537, 1995.
[6] F. Lin and C. Liu, “Existence of solutions for the Ericksen-Leslie
where system,” Archive for Rational Mechanics and Analysis, vol. 154,
no. 2, pp. 135–156, 2000.
1 1 1 1/2 − 1/𝑞 2 [7] H. Sun and C. Liu, “On energetic variational approaches
+ = , 𝜃= = , 𝑝𝛼 ≥ 2, (50) in modeling the nematic liquid crystal flows,” Discrete and
𝑝 𝑞 2 𝛼/2 𝑝𝛼 Continuous Dynamical Systems A, vol. 23, no. 1-2, pp. 455–475,
2009.
we can deduce [8] Y. Zhou and J. Fan, “A regularity criterion for the nematic liquid
crystal flows,” Journal of Inequalities and Applications, vol. 2010,
Article ID 589697, 9 pages, 2010.
𝐼2 = 𝐶 ∫ |Δ𝑢| ⋅ |Δ𝑑| ⋅ |∇Δ𝑑| 𝑑𝑥
R2 [9] J. Fan and T. Ozawa, “Regularity criterion for the incompressible
viscoelastic fluid system,” Houston Journal of Mathematics, vol.
≤ 𝐶‖∇Δ𝑑‖𝐿2 ‖Δ𝑑‖𝐿𝑝 ‖Δ𝑢‖𝐿𝑞 37, no. 2, pp. 627–636, 2011.
(51)
󵄩󵄩 𝛼+2 󵄩󵄩𝜃 [10] J. Fan, G. Nakamura, and Y. Zhou, “Global regularity for the 2D
≤ 𝐶‖∇Δ𝑑‖𝐿2 ‖Δ𝑑‖𝐿𝑝 ‖Δ𝑢‖1−𝜃
𝐿2 󵄩
󵄩Λ 𝑢󵄩󵄩 2
󵄩 󵄩𝐿 liquid crystal model with mixed partial viscosity”.
󵄩 󵄩2 [11] Y. Wang, Y. Zhou, A. Alsaedi, T. Hayat, and Z. Jiang, “Global
≤ 𝐶‖Δ𝑑‖2/(2−𝜃)
𝐿𝑝 (‖Δ𝑢‖2𝐿2 + ‖∇Δ𝑑‖2𝐿2 ) + 𝜖󵄩󵄩󵄩󵄩Λ𝛼+2 𝑢󵄩󵄩󵄩󵄩𝐿2 . regularity for the incompressible 2D generalized liquid crystal
8 Abstract and Applied Analysis

model with fractional diffusions,” Applied Mathematics Letters,

vol. 35, pp. 18–23, 2014.
[12] Y. Jin, Y. Zhou, and M. Zhu, “Global existence of solutions to the
2D incompressible liquid crystal ow with fractional diffusion,”
preprint.
[13] Y. Zhou, J. Fan, and G. Nakamura, “Global strong solution to
the density-dependent 2-D liquid crystal flows,” Abstract and
Applied Analysis, vol. 2013, Article ID 947291, 5 pages, 2013.
[14] Z. Jiang and Y. Zhou, “On regularity criteria for the 2D
generalized MHD system,” preprint.
[15] H. Kozono and Y. Taniuchi, “Bilinear estimates in BMO and the
Navier-Stokes equations,” Mathematische Zeitschrift, vol. 235,
no. 1, pp. 173–194, 2000.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 874701, 13 pages
http://dx.doi.org/10.1155/2014/874701

Research Article
Stability Switches and Hopf Bifurcation in
a Coupled FitzHugh-Nagumo Neural System with
Multiple Delays

Shengwei Yao1,2 and Huonian Tu2

1
School of Science, East China University of Science Technology, Shanghai 200237, China
2
School of Information and Statistics, Guangxi University of Finance and Economics, Nanning 530003, China

Correspondence should be addressed to Shengwei Yao; idhot@163.com

Received 25 April 2014; Revised 10 June 2014; Accepted 11 June 2014; Published 8 July 2014

Academic Editor: Yongli Song

Copyright © 2014 S. Yao and H. Tu. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

A FitzHugh-Nagumo (FHN) neural system with multiple delays has been proposed. The number of equilibrium point is analyzed.
It implies that the neural system exhibits a unique equilibrium and three ones for the different values of coupling weight by
employing the saddle-node bifurcation of nontrivial equilibrium point and transcritical bifurcation of trivial one. Further, the
stability of equilibrium point is studied by analyzing the corresponding characteristic equation. Some stability criteria involving
the multiple delays and coupling weight are obtained. The results show that the neural system exhibits the delay-independence
and delay-dependence stability. Increasing delay induces the stability switching between resting state and periodic activity in some
parameter regions of coupling weight. Finally, numerical simulations are taken to support the theoretical results.

1. Introduction 𝑢̇
3 = −𝑢3 (𝑢3 − 1) (𝑢3 − 𝑎) − 𝑢4 + 𝑐 tanh (𝑢1 (𝑡 − 𝜏2 )) ,

The FitzHugh-Nagumo (FHN) neuron [1, 2], a simplified 𝑢̇

4 = 𝑏 (𝑢3 − 𝛾𝑢4 ) ,
model of Hodgkin-Huxley neuron [3], describes the genera- (1)
tion and propagation of nerve impulse in planar autonomous
systems. In recent years, to understand information process- where 𝑎, 𝑏, and 𝛾 are the positive constants, 𝑢1,3 represent the
ing in the brain, the FHN neuron model is commonly used membrane potentials of neurons 1 and 2, 𝑢2,4 are the slow
to study neural spiking due to its simplicity. A complete refractory variables, which model the time dependence of
qualitative investigation of the FHN neuron was done by several physical quantities, 𝑐 measures the coupling weight,
Bautin [4]. Further, a hard oscillation, separatrix loops, and and 𝜏1,2 > 0 represent the time delays in signal transmission
bifurcations of equilibria and limit cycles can occur under between the neurons.
suitable values of the parameters in this system [5]. The research for coupled FHN neural systems with time
To understand the coupling effect and information trans- delay has attracted many authors’ attentions. Burić and
mission between neuron systems, the analysis of the dynamic Todorović [12] studied the Hopf bifurcation (inverse and
behavior in coupled FHN neural systems has been the subject direct) and fold bifurcation of limit cycle in the delay-
of many papers [6–9], in which many rich bifurcation behav- coupling FHN neurons. Using the ODE (ordinary differential
iors for equilibrium point and limit cycle are observed. Based equation) to approximate the dynamics of the coupled FHN
on the finite propagating speed in the signal transmission neurons with small delays, the codimension-2 generalized
between the neurons [10, 11], a coupled FHN neural system Hopf bifurcation was investigated [13]. The different syn-
with time delays has been established: chronization states were observed with the variation of the
coupling strength and time delay [14]. Their results illus-
𝑢̇
1 = −𝑢1 (𝑢1 − 1) (𝑢1 − 𝑎) − 𝑢2 + 𝑐 tanh (𝑢3 (𝑡 − 𝜏1 )) , trated that the patterns of exactly synchronous oscillations
𝑢̇
2 = 𝑏 (𝑢1 − 𝛾𝑢2 ) , were dependent on the type of excitability and coupling.
2 Abstract and Applied Analysis

Wang et al. [15] presented that time delay can control near the point (0, 0, 0, 0) produces the following linearized
the transition from the original chaotic motions to peri- system:
odic ones in two coupled nonidentical FHN models with
synaptic connection. Regarding the sum of two delays as a 𝑢̇
1 = −𝑎𝑢1 − 𝑢2 + 𝑐𝑢3 (𝑡 − 𝜏1 ) ,
parameter, Fan and Hong [16] investigated the stability and
local Hopf bifurcation in the synaptically coupled nonidenti- 𝑢̇
2 = 𝑏𝑢1 − 𝑏𝛾𝑢2 ,
cal FHN model. The simple zero singularity (both pitchfork (2)
𝑢̇
3 = −𝑎𝑢3 − 𝑢4 + 𝑐𝑢1 (𝑡 − 𝜏2 ) ,
and transcritical bifurcations) was studied employing the
center manifold reduction and normal form method [17]. 𝑢̇
4 = 𝑏𝑢3 − 𝑏𝛾𝑢4 .
Further, by applying the Bautin bifurcation theorem, the
coexistence of rest state and periodic spiking was reported in The corresponding characteristic equation is
the synchronous solution of a coupled FHN neural system
with delay [18]. 𝐷 (𝜆, 𝜏1 , 𝜏2 , 𝑐) = 𝜆4 + 𝑑1 𝜆3 + 𝑑2 𝜆2 + 𝑑3 𝜆 + 𝑑4 = 0, (3)
Recently, some criteria to determine the periodic oscil-
lation were provided in the multiple delayed FHN neural where
system with three nonidentical cells [19]. In [20], the Hopf
bifurcation and Bogdanov-Takens bifurcation were investi- 𝑑1 = 2𝑎 + 2𝑏𝛾,
gated in a coupled FHN neural system with gap junction.
The spatiotemporal patterns of bifurcating periodic solutions 𝑑2 = 𝑎2 + 2𝑏 − 𝑐2 e−𝜆𝜏1 −𝜆𝜏2 + 4𝑎𝑏𝛾 + 𝑏2 𝛾2 ,
were considered by using the symmetric bifurcation theory. (4)
The neuron behavior can transit between resting and spiking. 𝑑3 = 2𝑎𝑏 + 2𝑎2 𝑏𝛾 + 2𝑏2 𝛾 − 2𝑏𝑐2 e−𝜆𝜏1 −𝜆𝜏2 𝛾 + 2𝑎𝑏2 𝛾2 ,
By the proportional-spatial derivative control approach, Yang
et al. [21] achieved the control problem of the FHN equation. 𝑑4 = 𝑏2 + 2𝑎𝑏2 𝛾 + 𝑎2 𝑏2 𝛾2 − 𝑏2 𝑐2 e−𝜆𝜏1 −𝜆𝜏2 𝛾2 .
All works mentioned above have promoted greatly a deep
understanding for dynamic behavior of the coupled FHN sys- Then, we have the following conclusion on the eigenvalues of
tems with time delay. However, in reality, the neural dynamics system (1).
may be influenced by multiple independent parameters, such
as external inputs [22, 23], time delays, and coupling weight Theorem 1. System (1) has a zero eigenvalue if and only if
[24, 25]. Their combined effects on dynamic behavior may
be an important topic in the coupled FHN neural systems 1 + 𝑎𝛾 2 − 𝑏𝛾 (𝜏2 + 𝛾 (2 + 𝑎𝜏2 ))
𝑐=± , 𝜏1 ≠ . (5)
with multiple delays. The purpose of the present paper is to 𝛾 𝑏𝛾 (1 + 𝑎𝛾)
consider the combined effects of coupling weight and mul-
tiple delays on the stability of equilibrium point and obtain Further, system (1) exhibits a double zero eigenvalue if and only
the stability switches in the coupled FHN neural system with if
multiple delays (1).
The paper is organized as follows. In the next section, 1 + 𝑎𝛾 2 − 𝑏𝛾 (𝜏2 + 𝛾 (2 + 𝑎𝜏2 ))
𝑐=± , 𝜏1 = ,
we study the number of equilibrium points in the coupled 𝛾 𝑏𝛾 (1 + 𝑎𝛾)
FHN neural system employing the static bifurcation. The (6)
neural system (1) may have one/three equilibrium points −1 − 2𝑏𝛾2 + 𝑏2 𝛾4
𝑎 ≠ .
for the varying of the coupled weight by employing the 2𝛾
saddle-node bifurcation of nontrivial equilibrium point and
transcritical bifurcation of trivial one. In Section 3, the Proof. 𝜆 = 0 is a zero eigenvalue of system (1) if and only if
stability of equilibrium point for system (1) is investigated
by analyzing the corresponding characteristic equation. Some 𝑏2 + 2𝑎𝑏2 𝛾 + 𝑎2 𝑏2 𝛾2 − 𝑏2 𝑐2 𝛾2 = 0,
stability criteria involving the multiple delays and coupling
weight are obtained. The results show that the FHN neu- 𝜕𝐷(𝜆, 𝑐) 󵄨󵄨󵄨󵄨 2 2 2
󵄨 = 2𝑎𝑏 + 2𝑎 𝑏𝛾 + 2𝑏 𝛾 − 2𝑏𝑐 𝛾 (7)
ral system exhibits the parameter regions involving the 𝜕𝜆 󵄨󵄨󵄨𝜆=0
delay-independence stability and delay-dependence stabil-
ity. Increasing time delay can induce the stability switches + 2𝑎𝑏2 𝛾2 + 𝑏2 𝑐2 𝛾2 (𝜏1 + 𝜏2 ) ≠0.
between resting state and periodic activity in some parameter
regions of coupling weight. In Section 4, some numerical Further, 𝜆 = 0 is a double root of system (1) if and only if
simulations are employed to support the theoretical results.
Conclusions are given in Section 5. 1 + 𝑎𝛾 2 − 𝑏𝛾 (𝜏2 + 𝛾 (2 + 𝑎𝜏2 ))
𝑐=± , 𝜏1 = ,
𝛾 𝑏𝛾 (1 + 𝑎𝛾)
(8)
󵄨
2. Equilibrium Point Analysis 𝜕2 𝐷(𝜆, 𝑐) 󵄨󵄨󵄨 2 + 4𝑎𝛾 + 4𝑏𝛾2 − 2𝑏2 𝛾4
󵄨
󵄨 = ≠0.
𝜕𝜆2 󵄨󵄨󵄨𝜆=0 𝛾2
It is obvious that (𝑢1 , 𝑢2 , 𝑢3 , 𝑢4 ) = (0, 0, 0, 0) is the triv-
ial equilibrium point of system (1). Linearizing system (1)
Abstract and Applied Analysis 3

3 0.35
2 0.3
0.25
0.2
1 2 0.15
0.1
0.05 (0, 0)
0

u4
u4

0
(0, 0) 1
−0.1 0 0.1 0.2 0.3 0.4
−1
0
−2

−3 −2 −1 0 1 2 3 −3 −2 −1 0 1 2
u2 u2

(a) (b)

4 0.1
0 (0, 0)
−0.1
3 −0.2
−0.3
−0.4
2 −0.5
−0.6
u4

1 −0.6 −0.4 −0.2 0 0.2

−1

−6 −4 −2 0 2 4
u2

(c)

Figure 1: Intersection points of two curves given by (9), showing the number of equilibriums: (a) unique equilibrium for 𝑐 = 2.0, (b) three
equilibriums for 𝑐 = 2.3, and (c) three equilibriums for 𝑐 = 3.0, where the system parameter values are fixed as 𝑎 = 0.33, 𝑏 = 1, and 𝛾 = 0.47.

From the dynamic theory, when an eigenvalue passes the trivial equilibrium point (0, 0) evolves into somewhere
through the imaginary axis along the real axis with the between the nontrivial ones, as shown in Figure 1(c). It
variation of system parameter, a static bifurcation will be is called the transcritical bifurcation of the trivial point.
exhibited, which results in the variation of the number of Figure 2 shows the corresponding one-dimension bifurcation
equilibrium points. In fact, all equilibrium points of system diagram in the (𝑐, 𝑢1 )-plane (Figure 2(a)) and (𝑐, 𝑢2 )-plane
(1) are the solution of the following equations: (Figure 2(b)).
𝛾𝑢2 (𝛾𝑢2 − 1) (𝛾𝑢2 − 𝑎) + 𝑢2 − 𝑐 tanh (𝛾𝑢4 ) = 0,
(9) 3. Stability Switches and Hopf Bifurcation
𝛾𝑢4 (𝛾𝑢4 − 1) (𝛾𝑢4 − 𝑎) + 𝑢4 − 𝑐 tanh (𝛾𝑢2 ) = 0.
It is well known that the equilibrium point is locally asymp-
Solutions of (9), that is, equilibrium points of system (1), totically stable if and only if each eigenvalue of the character-
depend on the parameters 𝑎, 𝑐, and 𝛾 but are independent istic equation (3) has the negative real part. In the following,
of the delays 𝜏1 and 𝜏2 . Geometrically, equilibrium points are we investigate the stability of the equilibrium point by obtain-
the intersection points of two curves given by (9), as shown in ing the eigenvalues with the maximum real part. The FHN
Figure 1. These two curves are called the nullclines of system neural system exhibits the parameter regions involving the
(1). Thus, it is seen that system (1) exhibits either one or three delay-independence stability and delay-dependence stability.
equilibriums for the different coupling weight 𝑐. In fact, for The multiple delays can induce the stability switches between
the small value of the coupling weight 𝑐, the system has just resting state and periodic activity in some parameter regions
the trivial equilibrium point (0, 0), as shown in Figure 1(a). of coupling weight. To simplify, the investigation begins with
With the increasing of the coupling weight 𝑐, two nontrivial the case 𝜏1 = 𝜏2 = 0 in (3) as follows:
equilibrium points are emerged immediately at the right side
of the trivial point (0, 0) (Figure 1(b)), which is the saddle-
node bifurcation of the nontrivial equilibrium point. Further, 𝐷 (𝜆, 0, 0, 𝑐) = 𝜆4 + 𝑚3 𝜆3 + 𝑚2 𝜆2 + 𝑚1 𝜆 + 𝑚0 = 0, (10)
4 Abstract and Applied Analysis

1.5 3

1 2 Saddle-node
Saddle-node bifurcation
bifurcation Transcritical
0.5 Transcritical 1 bifurcation

u2
u1

bifurcation
0 0

−0.5 −1

−1 −2
0 1 2 3 4 5 6 0 1 2 3 4 5
c c
(a) (b)

Figure 2: The one-dimensional bifurcation diagrams in the (a) (𝑐, 𝑢1 )-plane and (b) (𝑐, 𝑢2 )-plane for the fixed parameters 𝑎 = 0.33, 𝑏 = 1,
and 𝛾 = 0.47.

where where
𝑚3 = 2𝑎 + 2𝑏𝛾,
𝑑3 = −2𝑖𝑎 − 2𝑖𝑏𝛾,
2 2 2 2
𝑚2 = 𝑎 + 2𝑏 − 𝑐 + 4𝑎𝑏𝛾 + 𝑏 𝛾 ,
(11) 𝑑2 = 𝑐2 e−𝑖V𝜏1 − 𝑎2 − 2𝑏 − 4𝑎𝑏𝛾 − 𝑏2 𝛾2 ,
𝑚1 = 2𝑎𝑏 + 2𝑎2 𝑏𝛾 + 2𝑏2 𝛾 − 2𝑏𝑐2 𝛾 + 2𝑎𝑏2 𝛾2 , (14)
𝑑1 = 2𝑖𝑎𝑏 − 2𝑖𝑏𝑐2 𝛾e−𝑖V𝜏1 + 2𝑖𝑎2 𝑏𝛾 + 2𝑖𝑏2 𝛾 + 2𝑖𝑎𝑏2 𝛾2 ,
𝑚0 = 𝑏2 + 2𝑎𝑏2 𝛾 + 𝑎2 𝑏2 𝛾2 − 𝑏2 𝑐2 𝛾2 .
𝑑0 = 𝑏2 + 2𝑎𝑏2 𝛾 − 𝑏2 𝑐2 𝛾2 e−𝑖V𝜏1 + 𝑎2 𝑏2 𝛾2 .
It follows from the Routh-Hurwitz criterion that the nec-
essary and sufficient conditions for all roots of (10) having That is
negative real parts are given by
𝑚3 > 0, 𝑚0 > 0, 𝑚2 𝑚3 − 𝑚1 > 0, V4 − 2𝑖 (𝑎 + 𝑏𝛾) V3 − (4𝑎𝑏𝛾 + 𝑏2 𝛾2 + 𝑎2 + 2𝑏) V2
(12)
𝑚1 𝑚2 𝑚3 − 𝑚12 − 𝑚0 𝑚32 > 0. + 2𝑖𝑏 (𝑎 + 𝑎2 𝛾 + 𝑎𝑏𝛾2 + 𝑏𝛾) V

That is + (𝑎2 𝛾2 + 1 + 2𝑎𝛾) 𝑏2 + (V2 − 2𝑖𝑏V𝛾 − 𝑏2 𝛾2 ) 𝑐2 cos V𝜏1

(i) 1 + 𝑎𝛾 > 𝑐𝛾 and 1 + 𝑏𝛾 > 𝑐,
− (𝑖V2 + 2𝑏V𝛾 − 𝑖𝑏2 𝛾2 ) 𝑐2 sin V𝜏1 = 0.
or (15)
2 3 2 2 2 2 3
(ii) 1 + 𝑎𝛾 > 𝑐𝛾, 𝑎 + 𝑎 𝛾 + 2𝑎𝑏𝛾 + 2𝑎 𝑏𝛾 + 𝑏 𝛾 + 𝑎𝑏 𝛾 <
𝑎𝑐2 𝛾, and 𝑎3 + 𝑎𝑏 + 4𝑎2 𝑏𝛾 + 𝑏2 𝛾2 + 4𝑎𝑏2 𝛾3 + 𝑏3 𝛾3 > 𝑎𝑐2 . Separating (15) into the real and imaginary parts yields
Based on the dynamical theory, we have the following.
𝑏2 − 𝑎2 V2 − 2𝑏V2 + V4 + 2𝑎𝑏2 𝛾 − 4𝑎𝑏V2 𝛾
Theorem 2. If the system parameters are satisfied with one of
the conditions (i) and (ii), the trivial equilibrium point is locally + 𝑎2 𝑏2 𝛾2 − 𝑏2 V2 𝛾2 + (𝑐2 V2 − 𝑏2 𝑐2 𝛾2 ) cos V𝜏1
asymptotically stable for the FHN neural system model without
any time delays. − 2𝑏𝑐2 V𝛾 sin V𝜏1 = 0, (16)

With the variation of delay 𝜏1 , the trivial equilibrium 2𝑎𝑏V − 2𝑎V3 + 2𝑎2 𝑏V𝛾 + 2𝑏2 V𝛾 − 2𝑏V3 𝛾 + 2𝑎𝑏2 V𝛾2
point of system (1) will lose its stability. The neural system
may exhibit the periodic activity. To obtain such critical val- − 2𝑏𝑐2 V𝛾 cos V𝜏1 + (𝑏2 𝑐2 𝛾2 − 𝑐2 V2 ) sin V𝜏1 = 0.
ues, supposing 𝜆 = ±𝑖V (V > 0) is a pair of purely imaginary
roots of the characteristic equation (3) with the case 𝜏2 = 0, Eliminating 𝜏1 from (16) using sin2 V𝜏1 + cos2 V𝜏1 = 1, we have
one has

𝐷 (𝑖V, 𝜏1 , 0, 𝑐) = V4 + 𝑑3 V3 + 𝑑2 V2 + 𝑑1 V + 𝑑0 = 0, (13) 𝐿 (V) = V8 + 𝑛1 V6 + 𝑛2 V4 + 𝑛3 V2 + 𝑛4 = 0, (17)

Abstract and Applied Analysis 5

∗
where 𝑘1 = 2𝑖𝑎𝑏 + 2𝑖𝑎2 𝑏𝛾 + 2𝑖𝑏2 𝛾 − 2𝑖𝑏𝑐2 e−𝑖𝜔𝜏1 e−𝑖𝜔𝜏2 𝛾 + 2𝑖𝑎𝑏2 𝛾2 ,
∗

𝑛1 = 2𝑎2 − 4𝑏 + 2𝑏2 𝛾2 , 𝑘0 = 𝑏2 + 2𝑎𝑏2 𝛾 + 𝑎2 𝑏2 𝛾2 − 𝑏2 𝑐2 e−𝑖𝜔𝜏1 e−𝑖𝜔𝜏2 𝛾2 .

(20)
𝑛2 = 𝑎4 − 4𝑎2 𝑏 + 6𝑏2 − 𝑐2 − 𝑐4 + 4𝑎𝑏2 𝛾
That is
+ 4𝑎2 𝑏2 𝛾2 − 4𝑏3 𝛾2 + 𝑏4 𝛾4 ,
𝑏2 + 2𝑎𝑏2 𝛾 + 𝑎2 𝑏2 𝛾2 − (𝑎2 + 2𝑏 + 4𝑎𝑏𝛾 + 𝑏2 𝛾2 ) 𝜔2
2 2 3 3 2 3
𝑛3 = 2𝑎 𝑏 − 4𝑏 + 4𝑎 𝑏 𝛾 − 8𝑎𝑏 𝛾
(18) + 𝜔4 + (𝑐2 𝜔2 − 𝑏2 𝑐2 𝛾2 ) cos 𝜔𝜏2 cos 𝜔𝜏1∗
+ 2𝑎4 𝑏2 𝛾2 − 4𝑎2 𝑏3 𝛾2 + 2𝑏4 𝛾2 − 2𝑏𝑐2 𝛾𝜔 cos 𝜔𝜏1∗ sin 𝜔𝜏2 − 2𝑏𝑐2 𝛾𝜔 cos 𝜔𝜏2 sin 𝜔𝜏1∗
− 2𝑏2 𝑐2 𝛾2 − 2𝑏2 𝑐4 𝛾2 + 4𝑎𝑏4 𝛾3 + 2𝑎2 𝑏4 𝛾4 , + (𝑏2 𝑐2 𝛾2 − 𝑐2 𝜔2 ) sin 𝜔𝜏2 sin 𝜔𝜏1∗ = 0,

𝑛4 = 𝑏4 + 4𝑎𝑏4 𝛾 + 6𝑎2 𝑏4 𝛾2 + 4𝑎3 𝑏4 𝛾3 (2𝑎𝑏 + 2𝑎2 𝑏𝛾 + 2𝑏2 𝛾 + 2𝑎𝑏2 𝛾2 ) 𝜔 − 2 (𝑎 + 𝑏𝛾) 𝜔3 (21)

2
+ 𝑎4 𝑏4 𝛾4 − 𝑏4 𝑐2 𝛾4 − 𝑏4 𝑐4 𝛾4 . − 2𝑏𝑐 𝛾𝜔 cos 𝜔𝜏2 cos 𝜔𝜏1∗
+ (𝑏2 𝑐2 𝛾2 − 𝑐2 𝜔2 ) cos 𝜔𝜏1∗ sin 𝜔𝜏2
Based on the dynamical theory, the equilibrium point is
locally asymptotically stable if and only if each of eigenvalues + (𝑏2 𝑐2 𝛾2 − 𝑐2 𝜔2 ) cos 𝜏2 𝜔 sin 𝜔𝜏1∗
has the negative real part. Therefore, one can obtain the effect
of delay 𝜏1 on the eigenvalue of characteristic equation (13). + 2𝑏𝑐2 𝛾𝜔 sin 𝜏2 𝜔 sin 𝜔𝜏1∗ = 0.
Theorem 3. If the parameters values of system (1) are satisfied Eliminating 𝜏2 from (21), one has
with one of the conditions (i) and (ii), the following statements
are true. 𝑓1 (𝜏1∗ , 𝜔)
cos 𝜔𝜏2 = ,
𝑓3 (𝜏1∗ , 𝜔)
(a) When the polynomial 𝐿(V) has no positive root, all (22)
eigenvalues of the characteristic equation (13) have the 𝑓 (𝜏∗ , 𝜔)
negative real parts for the arbitrary 𝜏1 . The trivial equi- sin 𝜔𝜏2 = 2 1∗ ,
𝑓3 (𝜏1 , 𝜔)
librium point exhibits the delay-independent stability.
where
(b) When the polynomial 𝐿(V) has only one positive and
simple root ], there exists the critical delayed value 𝑓1 (𝜏1∗ , 𝜔)
𝜏1𝑐 > 0 such that all the eigenvalues of the characteristic
equation (13) have negative real parts for 𝜏1 ∈ (0, 𝜏1𝑐 ) = (𝑏𝜔 + (𝑎 − 𝜔) 𝜔2 + 𝑏2 𝛾 (1 + 𝑎𝛾 − 𝛾𝜔))
and at least one root has a positive real part for 𝜏1 > 𝜏1𝑐 .
The FHN neural system exhibits the periodic activity × (−𝑏𝜔 + 𝜔2 (𝑎 + 𝜔)
when delay 𝜏1 increases and passes through the critical
delayed value 𝜏1𝑐 . + 𝑏2 𝛾 (1 + 𝑎𝛾 + 𝜔𝛾))
(c) When the polynomial 𝐿(V) has at least two positive and × cos 𝜔𝜏1∗ − 2𝜔 (𝑏 (−1 + 𝑏𝛾2 ) + 𝜔2 )
simple roots 0 < V1 < V2 < ⋅ ⋅ ⋅ , there exist a finite
number of the delayed 𝜏1 intervals in which all eigenval- × (𝑏2 𝛾 (1 + 𝑎𝛾) + 𝑎𝜔2 ) sin 𝜔𝜏1∗ ,
ues of the characteristic equation (13) have negative real
parts. The system exhibits the stability switches between 𝑓2 (𝜏1∗ , 𝜔)
resting state and periodic activity.
(23)
= − (2𝜔 (𝑏 (−1 + 𝑏𝛾2 ) + 𝜔2 )
In order to investigate the combined effects of multiple
delays on the local stability of neural system (1), we regard 𝜏2 × (𝑏2 𝛾 (1 + 𝑎𝛾) + 𝑎𝜔2 ) cos 𝜔𝜏1∗
as the varying parameter for any fixed delay 𝜏1 = 𝜏1∗ . Letting
𝜆 = 𝑖𝜔 be the simple roots of the characteristic equation (3), + (𝑏𝜔 + (𝑎 − 𝜔) 𝜔2
one obtains
+𝑏2 𝛾 (1 + 𝑎𝛾 − 𝛾𝜔))
𝐷 (𝑖𝜔, 𝜏1∗ , 𝜏2 , 𝑐) = 𝜔4 + 𝑘3 𝜔3 + 𝑘2 𝜔2 + 𝑘1 𝜔 + 𝑘0 = 0, (19)
× (−𝑏𝜔 + 𝜔2 (𝑎 + 𝜔)
where
+𝑏2 𝛾 (1 + 𝑎𝛾 + 𝜔𝛾)) sin 𝜔𝜏1∗ ) ,
𝑘3 = −2𝑖𝑎 − 2𝑖𝑏𝛾,
2
∗
𝑓3 (𝜏1∗ , 𝜔) = 𝑐2 (𝑏2 𝛾2 + 𝜔2 ) .
𝑘2 = −𝑎2 − 2𝑏 + 𝑐2 e−𝑖𝜔𝜏1 e−𝑖𝜔𝜏2 − 4𝑎𝑏𝛾 − 𝑏2 𝛾2 ,
6 Abstract and Applied Analysis

×106

12 1.5 0.1

1 0.05
10
0.5 0
8
0 −0.05

Re (𝜆)
6
L

0 0.5 1 1.5 −0.1

4
−0.15

2 −0.2

0 −0.25

0.5 1 1.5 2 2.5 3 3.5 4 0 5 10 15

v 𝜏1

(a) (b)
6
×10
14 0.1
0.6 Resting Periodic Resting Periodic
12 state spiking state spiking
0.4 0.05
2 = 1.2372
10 0.2
0 0
8
−0.2 1 = 0.8826
Re (𝜆)

6 −0.4 −0.05
L

−0.6
4 0 0.2 0.4 0.6 0.8 1 1.2 −0.1
2
−0.15
0
−0.2
−2
0 1 2 3 4 0 5 10 15
v 𝜏1

(c) (d)

Figure 3: Roofs of function 𝐿 ((a) and (c)) and eigenvalue real parts with 𝜏1 varying ((b) and (d)) for the different coupling weights (a)-(b)
𝑐 = 0.5 and (c)-(d) 𝑐 = 0.8. The other parameters are chosen as 𝑎 = 0.33, 𝑏 = 1, 𝛾 = 0.47, and 𝜏2 = 0.

0.1 0.1

0.05 0.05

0 0
u1
u1

−0.05 −0.05

−0.1 −0.1
0 50 100 150 200 0 50 100 150 200
t t
(a) (b)

Figure 4: Time histories with the varying delay 𝜏1 (a) 𝜏1 = 1.0 and (b) 𝜏1 = 15.0 when the time delay and coupling weight are fixed as 𝜏2 = 0,
𝑐 = 0.5, where the other system parameters have the fixed values 𝑎 = 0.33, 𝑏 = 1, and 𝛾 = 0.47.
Abstract and Applied Analysis 7

0.1 0.6

0.4
0.05

0.2
0
u1

u1
0

−0.05
−0.2

−0.1 −0.4
0 50 100 150 200 0 50 100 150 200
t t
(a) (b)
0.1 0.3

0.2
0.05

0.1
0
u1

u1

−0.05
−0.1

−0.1 −0.2
0 50 100 150 200 0 50 100 150 200
t t
(c) (d)

Figure 5: Time histories with the varying delay 𝜏1 (a) 𝜏1 = 1.0, (b) 𝜏1 = 5.0, (c) 𝜏1 = 8.0, and (d) 𝜏1 = 10.0 when the delay 𝜏2 and coupled
weight are fixed as 𝜏2 = 0, 𝑐 = 0.8, where the other system parameters have the fixed values 𝑎 = 0.33, 𝑏 = 1, and 𝛾 = 0.47.

This implies that 𝑖 = 1, 2, . . . ; 𝑗 = 0, 1, 2, . . .. Differentiating 𝜆 with respect to

𝜏2 in (3) reaches
2 2
𝐺 (𝜏1∗ , 𝜔) = (𝑓1 (𝜏1∗ , 𝜔)) + (𝑓2 (𝜏1∗ , 𝜔))
∗ 2
(24) 𝜆2 e2𝜆𝜏1 𝑐2 (𝑏𝛾 + 𝜆)
2 󸀠
− (𝑓3 (𝜏1∗ , 𝜔)) = 0. 𝜆 (𝜏2 ) = , (27)
𝑔 (𝜏2 , 𝜆)
If (24) has a number of positive and simple roots 𝜔𝑖 , 𝑖 =
1, 2, . . ., (3) has the following critical delays determined by where
(24): ∗
𝑔 (𝜏2 , 𝜆) = 2e𝜆(𝜏1 +𝜏2 ) (𝑎 + 𝑏𝛾 + 2𝜆)
(𝜑𝑖 + 2𝑗𝜋)
𝜏𝑖,𝑗 = , 𝑖 = 1, 2, . . . ; 𝑗 = 0, 1, 2, . . . , (25) × (𝜆 (𝑎 + 𝜆) + 𝑏 (1 + 𝑎𝛾 + 𝜆𝛾))
𝜔𝑖
(28)
where 𝜑𝑖 ∈ [0, 2𝜋) and is satisfied with + 𝑐2 (𝑏𝛾 + 𝜆)

𝑓1 (𝜏1∗ , 𝜔𝑖 ) 𝑓2 (𝜏1∗ , 𝜔𝑖 ) × (−2 + (𝑏𝛾 + 𝜆) (𝜏1∗ + 𝜏2 )) .

cos (𝜑𝑖 ) = , sin (𝜑𝑖 ) = . (26)
𝑓3 (𝜏1∗ , 𝜔𝑖 ) 𝑓3 (𝜏1∗ , 𝜔𝑖 )
Based on the transversality condition (27) and the Hopf bifur-
We determine whether the eigenvalue real part increases cation theorem of delay differential equations, one has the
or decreases as delay 𝜏2 crosses the critical values 𝜏𝑖,𝑗 , following theorem.
8 Abstract and Applied Analysis

0.5
0.6
Periodic spiking
Resting state
0.4
0.4

0.3
0.2

u2
Resting state
u1

0.2
0

0.1
−0.2

0
−0.4 Periodic spiking

4 6 8 10 12 14 4 6 8 10 12 14
𝜏1 𝜏1

(a) (b)
0.6
Periodic spiking Periodic spiking
0.4 0.5

0.4
0.2
0.3
u4
u3

0 0.2

0.1
−0.2
0
Resting state −0.1 Resting state
−0.4
4 6 8 10 12 14 4 6 8 10 12 14
𝜏1 𝜏1

(c) (d)

Figure 6: The one-dimensional bifurcation diagram in (a) (𝜏1 , 𝑢1 )-plane, (b) (𝜏1 , 𝑢2 )-plane, (c) (𝜏1 , 𝑢3 )-plane, and (d) (𝜏1 , 𝑢4 )-plane for the
fixed delay 𝜏2 = 0 and coupling weights 𝑐 = 0.8.

Theorem 4. The following assertions are true if all roots of (13) sate of neural system switches from stable to unstable,
have negative real parts. and then back to stable when time delay increases and
crosses the critical values in sequence.
(a) If 𝐺(𝜏1∗ , 𝜔) = 0 has no positive root, the trivial equi-
librium point of system (1) is asymptotically stable for 4. Numerical Simulation
an arbitrary time delay 𝜏2 , which is called the delay 𝜏2 -
independent stability. In this section, some numerical results of system (1) are pre-
(b) If 𝐺(𝜏1∗ , 𝜔)= 0 has at least one positive and simple root sented for the different parameter values. Using the method
𝜔𝑐 , there exists exactly a critical delay 𝜏2𝑐 > 0 such that of numerical simulation, we find that the theoretically pre-
the trivial equilibrium of system (1) is asymptotically dicted values are in excellent agreement with the numerical
stable for 𝜏2 ∈ [0, 𝜏2𝑐 ). Furthermore, if the transversality behaviors. For simplicity, the system parameters are fixed as
𝑎 = 0.33, 𝑏 = 1, and 𝛾 = 0.47. The coupling weight 𝑐 and time
condition Re(𝜆󸀠 (𝜏2 )) ≠0 holds, system (1) undergoes a
delays (𝜏1 , 𝜏2 ) are considered as the variable parameters.
Hopf bifurcation for 𝜏2 = 𝜏2𝑐 . That is, it exhibits a peri-
Firstly, we fix the time delay 𝜏2 = 0 and illustrate the
odic activity bifurcating from the trivial equilibrium
dynamic behavior of system (1) for the different values of the
near 𝜏2𝑐 .
coupling weight 𝑐 and time delay 𝜏1 . It follows from (17) that
(c) If 𝐺(𝜏1∗ , 𝜔) = 0 has at least two positive and simple roots the polynomial 𝐿(V) has no positive root (see Figure 3(a)) for
0 < 𝜔1 < 𝜔2 < ⋅ ⋅ ⋅ and the transversality condition the fixed coupling weight 𝑐 = 0.5. The trivial equilibrium
Re(𝜆󸀠 (𝜏2 )) ≠0 holds, there exist a finite number of point is locally asymptotically stable for the arbitrary delay
intervals. If time delay is fixed into these intervals, the 𝜏1 , which is called the delay 𝜏1 -independent stability. The
trivial equilibrium point is locally asymptotically stable, corresponding eigenvalue has the negative real parts for the
while unstable if delay does not belong to ones. The rest delay 𝜏1 variation, as shown in Figure 3(b). The time histories
Abstract and Applied Analysis 9

0.1

0.05

u1
0.1 0

0
u1

−0.05

−0.1
−0.1
0 5000 10000 15000 1000 1200 1400 1600 1800 2000
t t
(a) (b)

0.15

0.1

0.05
u1

−0.05

−0.1

−0.15
1.48 1.485 1.49 1.495 1.5
t ×104

(c)

Figure 7: Time histories of the transient chaos for (a) overall view, (b) transient behavior, and (c) long-term behavior for the fixed delay
𝜏1 = 14.15 when 𝜏2 = 0 and 𝑐 = 0.8.

with the fixed delay are shown in Figure 4(a) for 𝜏1 = 1.0 and theoretical results. Neural system (1) exhibits the multiple
Figure 4(b) for 𝜏1 = 15.0; both converge to the equilibrium switches between resting state and periodic spiking. Further,
point (0, 0, 0, 0) spirally. it is worth mentioning that the neural system (1) exhibits
However, when the coupling weight is fixed as 𝑐 = 0.8, the transient chaos (see Figure 7(a)), which implies that
the equation 𝐿(V) = 0 has two positive and simple roots the system trajectory exhibits a seemingly chaotic solution
V1 = 0.8826, V2 = 1.2372, as shown in Figure 3(c). The corre- for a longer time (see Figure 7(b)) but evolves into a final
sponding eigenvalue with 𝜏1 varying is shown in Figure 3(d). nonchaotic state, such as a periodic activity (see Figure 7(c))
There exist some critical delays determined by (25). From or equilibrium eventually.
Theorem 3, system (1) exhibits the stability switches with The partial eigenvalues are exhibited in Figure 8 for the
the delay 𝜏1 increasing. In fact, the trivial equilibrium point different delay 𝜏1 to illustrate the stability switches between
is asymptotically stable when 𝜏1 ∈ (0, 3.9045). Figure 5(a) resting state and periodic spiking. Figure 8(a) shows that the
illustrates the time history for the fixed delay 𝜏1 = 1.0. When maximum eigenvalues are a conjugate pair with negative part
𝜏1 ∈ (3.9045, 7.1184), the trivial equilibrium point loses its given by −0.0888 ± 0.7515𝑖 for the fixed delay 𝜏1 = 1.0. The
stability and evolves into the instability. The time history for neural system exhibits the resting state. Increasing delay 𝜏1
𝜏1 = 5.0 is shown in Figure 5(b). Further, system (1) obtains yields that the conjugate pair passes through the imaginary
the resting state when delay increases and crosses the critical axis and go into the right-half plane. For 𝜏1 = 5.0, the
value 𝜏1 = 7.1184. The time history for the fixed delay 𝜏1 = pair becomes to be 0.0187 ± 1.0846𝑖 shown in Figure 8(b).
8.0 is illustrated in Figure 5(c). Finally, the equilibrium point The trivial equilibrium point loses its stability. The FHN
loses its stability when the delay 𝜏1 continues to increase, neural system exhibits the periodic spiking. Continuing to
which is shown in Figure 5(d) for the fixed delay 𝜏1 = 10.0. increase delay 𝜏1 , the conjugate pair returns and passes
Figure 6 shows the corresponding one-dimension bifurcation through the imaginary axis again. The maximum eigenval-
diagrams for the varying delay 𝜏1 . It supports the above ues with negative real parts occurs in the left-hand plane.
10 Abstract and Applied Analysis

10 10

𝜏1 = 1.0 𝜏1 = 5.0

5 5

Im (𝜆)
Im (𝜆)

0 0

−5 −5

−10 −10
−5 −4 −3 −2 −1 0 −0.8 −0.6 −0.4 −0.2 0
Re (𝜆) Re (𝜆)
(a) (b)

6
4 𝜏1 = 10.0
𝜏1 = 8.0
4

2 2
Im (𝜆)

Im (𝜆)

0 0

−2
−2
−4
−4
−6
−0.35 −0.3 −0.25 −0.2 −0.15 −0.1 −0.05 0 −0.25 −0.2 −0.15 −0.1 −0.05 0
Re (𝜆) Re (𝜆)
(c) (d)

Figure 8: Distribution of partial eigenvalues for the fixed coupling weight 𝑐 = 0.8 and time delay 𝜏2 = 0, where the asterisk in green represents
the eigenvalues with negative real part and one with positive real part is in red color. The other parameters are fixed as 𝑎 = 0.33, 𝑏 = 1, and
𝛾 = 0.47.

Figure 8(c) shows the eigenvalues are −0.0133 ± 0.8189𝑖 𝐺 has two positive roots shown in Figure 9(c). It follows from
for the fixed delay 𝜏1 = 8.0. The neural system obtains Figure 9(d) that there exist the delay 𝜏2 intervals in which the
again the resting state. Further, if delay 𝜏1 is fixed as eigenvalues have the negative parts. The trivial equilibrium
𝜏1 = 10.0, the maximum eigenvalues are the conjugate pair point switches from stable to unstable, and then back to stable
with positive part 9.0902 × 10−3 ± 1.1452𝑖, as shown in state with delay 𝜏2 increasing. The neural system exhibits the
Figure 8(d). The system (1) exhibits the periodic spiking. In stability switches between resting state and periodic spiking.
a word, the increase of time delay can induce the stability When coupling weight 𝑐 is increased to 1.8, function 𝐺 has
switches between resting state and periodic activity on the two pairs of positive roots, as shown in Figure 9(e). However,
neural system (1). at this time, the maximum eigenvalues of the characteristic
Furthermore, we fix the time delay 𝜏1 = 𝜏1∗ = 0.5 and find equation (13) for the fixed delay 𝜏2 = 0 have the positive real
the combined effects of delay 𝜏2 and coupling weight 𝑐 on the part Re(𝜆) = 0.2787. The equilibrium point of system (1) is
dynamic behavior of neural system (1). The figures of function instable for any delay 𝜏2 , as shown in Figure 9(f).
𝐺 and eigenvalues real parts are displayed in Figure 9 for the Additionally, it follows from Figure 10 that the time delay
different coupling weights 𝑐. Figure 9(a) shows that the curve has the different effect on the periodic activity in the FHN
𝐺 with 𝑐 = 0.2. There is no intersection point when 𝜔 > 0 neural system (1). When the coupling weight and time delay
holds. It implies that the function 𝐺 has no positive root. are chosen as 𝑐 = 1.0, 𝜏1 = 1.0, the eigenvalues of
The real parts of all eigenvalues have the negative part, as the maximum real parts are changed from negative value
shown in Figure 9(b). The equilibrium point of system (1) is to positive ones, as shown in Figure 10(a). It implies that
locally asymptotically stable for the arbitrary delay 𝜏2 . It is the system behavior will evolve from the resting state into
called the delay 𝜏2 -independent stability. With the coupling periodic spiking. That is to say, time delay promotes the
weight 𝑐 increasing to 0.8, the curve determined by function occurrence of the periodic activity in the delay-coupling FHN
Abstract and Applied Analysis 11

×106

12
1.5 0
10
1
8 −0.1
0.5

Re (𝜆)
G 6
0 −0.2
4
0 0.5 1 1.5
−0.3
2

0
−0.4
0 1 2 3 4 0 5 10 15
𝜔 𝜏2

(a) (b)
6
×10
Resting Periodic Resting Periodic
12 0.05
0.6 state spiking state spiking
10 0.4 𝜔2 = 1.2015
0.2 0
8 0
−0.2 𝜔1 = 0.8425
Re (𝜆)

G −0.05
6 −0.4
−0.6
4 −0.8 −0.1
0 0.2 0.4 0.6 0.8 1 1.2
2
−0.15
0
0 1 2 3 4 0 5 10 15
𝜔 𝜏2

(c) (d)
6
×10

12 0.3
1
0.25
10 0.5
𝜔1 = 0.3597
0.2
8 0
Re (𝜆)

𝜔2 = 2.1864 0.15
G 6 −0.5
0.1
4 −1
0 0.5 1 1.5 2 2.5 0.05
2
0
0 −0.05
0 1 2 3 4 0 5 10 15
𝜔 𝜏2

(e) (f)

Figure 9: Function 𝐺 curve ((a), (c), and (e)) and eigenvalues real parts with 𝜏2 varying ((b), (d), and (f)) for the different coupling weights
(a)-(b) 𝑐 = 0.2, (c)-(d) 𝑐 = 0.8, and (e)-(f) 𝑐 = 1.8. The other parameters are fixed as 𝑎 = 0.33, 𝑏 = 1, 𝛾 = 0.47, and 𝜏1 = 0.5.

neural system. However, for the parameter values of 𝑐 = 1.2 Figure 10(b). It implies that the trivial equilibrium point
and 𝜏1 = 0.1, the system has the periodic activity with the obtains the stability by the increasing delay 𝜏2 . Time delay
nondelay 𝜏2 = 0. Increasing delay 𝜏2 , the eigenvalue of the suppresses the system dynamic from periodic activity into
maximum real parts exhibits the negative value, as shown in resting state, which can be used to control the neural system.
12 Abstract and Applied Analysis

Finally, numerical simulations are taken to support the

0.1 Resting state Periodic spiking theoretical results.

0.05 Conflict of Interests

The authors declare that there is no conflict of interests
Re (𝜆)

0
regarding the publication of this paper.
−0.05
Acknowledgments
−0.1
The authors would like to thank the editor and the reviewers
for their valuable suggestions and comments.
0 5 10 15
𝜏2
References
(a)
[1] R. FitzHugh, “Impulses and physiological states in theoretical
models of nerve membrane,” Biophysical Journal, vol. 1, no. 6,
0.1 Resting state Periodic spiking pp. 445–466, 1961.
[2] J. Nagumo, S. Arimoto, and S. Yoshizawa, “An active pulse
0.05 transmission line simulating nerve axon,” Proceedings of the IRE,
vol. 50, no. 10, pp. 2061–2070, 1962.
Re (𝜆)

0
[3] A. L. Hodgkin and A. F. Huxley, “A quantitative description of
membrane current and its application to conduction and
excitation in nerve,” The Journal of Physiology, vol. 117, no. 4, pp.
−0.05 500–544, 1952.
[4] A. N. Bautin, “Qualitative investigation of a particular nonlinear
−0.1 system,” Journal of Applied Mathematics and Mechanics, vol. 39,
no. 4, pp. 606–615, 1975.
0 5 10 15 [5] J. Duarte, L. Silva, and J. S. Ramos, “Types of bifurcations of
FitzHugh-Nagumo maps,” Nonlinear Dynamics, vol. 44, no. 1–
𝜏2
4, pp. 231–242, 2006.
(b) [6] T. Ueta, H. Miyazaki, T. Kousaka, and H. Kawakami, “Bifurca-
tion and chaos in coupled BVP oscillators,” International Jour-
Figure 10: Eigenvalues of the maximum real parts show that the
nal of Bifurcation and Chaos in Applied Sciences and Engineering,
delay 𝜏2 increasing promotes the system dynamic from resting state
vol. 14, no. 4, pp. 1305–1324, 2004.
into periodic spiking for 𝑐 = 1.0, 𝜏1 = 1.0 (a) and suppresses ones
from periodic activity into resting state for 𝑐 = 1.2, 𝜏1 = 0.1 (b), [7] T. Ueta and H. Kawakami, “Bifurcation in asymmetrically
where the other parameters are fixed as 𝑎 = 0.33, 𝑏 = 1, and 𝛾 = 0.47. coupled BVP oscillators,” International Journal of Bifurcation
and Chaos in Applied Sciences and Engineering, vol. 13, no. 5,
pp. 1319–1327, 2003.
[8] S. Tsuji, T. Ueta, and H. Kawakami, “Bifurcation analysis of
5. Conclusion current coupled BVP oscillators,” International Journal of Bifur-
cation and Chaos in Applied Sciences and Engineering, vol. 17, no.
Time delay is an inevitable factor in the signal transmission 3, pp. 837–850, 2007.
between neurons. The neural system with time delay exhibits [9] D. Yang, “Self-synchronization of coupled chaotic FitzHugh-
the rich dynamical behaviors. In this paper, a coupled FHN Nagumo systems with unreliable communication links,” Com-
neural system with two delays has been proposed. The munications in Nonlinear Science and Numerical Simulation, vol.
analyses of the number of equilibrium points illustrate that 18, no. 10, pp. 2783–2789, 2013.
the neural system has a unique equilibrium and three equi- [10] C. Xu and P. Li, “Dynamics in a delayed neural network model
libria for the different values of coupling weights. It exhibits of two neurons with inertial coupling,” Abstract and Applied
the multiple equilibrium points employing the saddle-node Analysis, vol. 2012, Article ID 689319, 17 pages, 2012.
bifurcation of nontrivial equilibrium point and the transcrit- [11] J. Liang, Z. Chen, and Q. Song, “State estimation for neural
ical bifurcation of the trivial point. Further, the stability of networks with leakage delay and time-varying delays,” Abstract
equilibrium point is analyzed employing the corresponding and Applied Analysis, vol. 2013, Article ID 289526, 9 pages, 2013.
characteristic equation. Some stability criteria involving the [12] N. Burić and D. Todorović, “Dynamics of FitzHugh-Nagumo
multiple delays and coupling weight are obtained. The results excitable systems with delayed coupling,” Physical Review E, vol.
show that the FHN neural system exhibits the parameter 67, Article ID 066222, 2003.
regions involved the delay-independence stability and delay- [13] N. Buric and D. Todorovic, “Bifurcations due to small time-
dependence stability. Time delay increasing can induce the lag in coupled excitable systems,” International Journal of
stability switches between resting state and periodic activity. Bifurcation and Chaos, vol. 15, no. 5, pp. 1775–1785, 2005.
Abstract and Applied Analysis 13

[14] N. Burić, I. Grozdanović, and N. Vasović, “Type I vs. type II

excitable systems with delayed coupling,” Chaos, Solitons &
Fractals, vol. 23, no. 4, pp. 1221–1233, 2005.
[15] Q. Wang, Q. Lu, G. Chen, Z. feng, and L. Duan, “Bifurcation and
synchronization of synaptically coupled FHN models with time
delay,” Chaos, Solitons and Fractals, vol. 39, no. 2, pp. 918–925,
2009.
[16] D. Fan and L. Hong, “Hopf bifurcation analysis in a synaptically
coupled FHN neuron model with delays,” Communications in
Nonlinear Science and Numerical Simulation, vol. 15, no. 7, pp.
1873–1886, 2010.
[17] B. Zhen and J. Xu, “Simple zero singularity analysis in a coupled
FitzHugh-Nagumo neural system with delay,” Neurocomputing,
vol. 73, no. 4–6, pp. 874–882, 2010.
[18] B. Zhen and J. Xu, “Bautin bifurcation analysis for synchronous
solution of a coupled FHN neural system with delay,” Commu-
nications in Nonlinear Science and Numerical Simulation, vol. 15,
no. 2, pp. 442–458, 2010.
[19] Y. Lin, “Periodic oscillation analysis for a coupled FHN network
model with delays,” Abstract and Applied Analysis, vol. 2013,
Article ID 276972, 6 pages, 2013.
[20] Y. Li and W. Jiang, “Hopf and Bogdanov-Takens bifurcations
in a coupled FitzHugh-Nagumo neural system with delay,”
Nonlinear Dynamics, vol. 65, no. 1-2, pp. 161–173, 2011.
[21] C. D. Yang, J. Qiu, and J. W. Wang, “Robust 𝐻∞ control
for a class of nonlinear distributed parameter systems via
proportional-spatial derivative control approach,” Abstract and
Applied Analysis, vol. 2014, Article ID 631071, 8 pages, 2014.
[22] Z. G. Song and J. Xu, “Codimension-two bursting analysis in the
delayed neural system with external stimulations,” Nonlinear
Dynamics, vol. 67, no. 1, pp. 309–328, 2012.
[23] Z. G. Song and J. Xu, “Bursting near bautin bifurcation in a
neural network with delay coupling,” International Journal of
Neural Systems, vol. 19, no. 5, pp. 359–373, 2009.
[24] Z.-G. Song and J. Xu, “Stability switches and multistability coex-
istence in a delay-coupled neural oscillators system,” Journal of
Theoretical Biology, vol. 313, no. 21, pp. 98–114, 2012.
[25] Z. Song and J. Xu, “Stability switches and double Hopf bifur-
cation in a two-neural network system with multiple delays,”
Cognitive Neurodynamics, vol. 7, no. 6, pp. 505–521, 2013.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 248657, 11 pages
http://dx.doi.org/10.1155/2014/248657

Research Article
Nonlinear Instability for a Volume-Filling Chemotaxis
Model with Logistic Growth

Haiyan Gao1,2 and Shengmao Fu3

1
College of Mathematics and Statistics, Northwest Normal University, Lanzhou 730070, China
2
School of Statistics, Lanzhou University of Finance and Economics, Lanzhou 730020, China
3
School of Mathematics and Computing Science, Guilin University of Electronic Technology, Guilin 541004, China

Correspondence should be addressed to Haiyan Gao; gaohy 54@sina.com

Received 2 April 2014; Accepted 8 June 2014; Published 3 July 2014

Academic Editor: Yongli Song

Copyright © 2014 H. Gao and S. Fu. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

This paper deals with a Neumann boundary value problem for a volume-filling chemotaxis model with logistic growth in a 𝑑-
dimensional box T 𝑑 = (0, 𝜋)𝑑 (𝑑 = 1, 2, 3). It is proved that given any general perturbation of magnitude 𝛿, its nonlinear evolution
is dominated by the corresponding linear dynamics along a finite number of fixed fastest growing modes, over a time period of the
order ln(1/𝛿). Each initial perturbation certainly can behave drastically different from another, which gives rise to the richness of
patterns.

1. Introduction firstly proved the global existence of solutions. Numerical

simulations in one and two dimensions show interesting
An important variant of the chemotaxis model was initially phenomena of pattern formation and formation of stable
proposed by Painter and Hillen in [1] to model the volume- aggregates. Wrzosek [3] showed the existence of a compact
filling effect. In the volume-filling effect, it is assumed that
global attractor in the space 𝑊1,𝑝 (Ω, R2 ), 𝑝 > 𝑛, Ω ⊂ R𝑛
cells have a certain finite (nonzero) volume and that the
for some cases. In [4], the structure of the attractor can be
occupation of an area limits other cells from penetrating it. A
understood using Lyapunov functions. Stationary solutions
simple version of the volume-filling chemotaxis model (VF)
which are inhomogeneous in space were investigated for a
is the following:
given range of parameters. In [1], a numerical exploration was
𝑈𝑡 = ∇ (𝑑1 ∇𝑈 − 𝜒𝑈 (1 − 𝑈) ∇𝑉) , conducted to determine the longtime patterning behaviour,
(1) revealing formation of multiple plateau type patterns which
𝑉𝑡 = 𝑑2 ∇2 𝑉 + 𝛼𝑈 − 𝛽𝑉, undergo a coarsening process with increasingly long transient
where 𝑑1 , 𝑑2 , 𝜒, 𝛼, and 𝛽 are given positive constants. 𝑈(𝑥, 𝑡) times. Potapov and Hillen [5] investigated the metastability
is the cell density and 𝑉(𝑥, 𝑡) denotes the density of the of steady states. The underlying bifurcation diagram was
external chemical substance which is secreted by the cells identified, revealing that the unstable eigenvalues are expo-
themselves. 𝑑1 and 𝑑2 denote the cell and chemical diffusion nentially small. The plateau interactions were studied using
coefficients, respectively. 𝜒 is called chemosensitivity. The asymptotic methods. In [6], it was obtained that a priori
term 𝑈(1 − 𝑈)∇𝑉 denotes the chemotactic flux under a vol- estimates for the classical chemotaxis model of Patlak, Keller,
ume constraint 1 (called crowding capacity), meaning that the and Segel when a nonlinear diffusion or a nonlinear che-
chemotactic movement will be inhibited at the‘aggregation mosensitivity was considered accounting for the finite size
location where the cell density reaches 1. of the cells and how entropy estimates give natural condi-
In recent years, the chemotaxis models with volume-fill- tions on the nonlinearities implying the absence of blow-up
ing effect have been studied extensively. Hillen and Painter [2] for the solutions were showed. Burger et al. [7] discussed
2 Abstract and Applied Analysis

the effects of linear and nonlinear diffusion in the large time a chemotaxis model with the volume-filling effect and logistic
asymptotic behavior of the Keller-Segel model of chemotaxis cell growth and established the critical value of the chemotac-
with volume-filling effect. Moreover, the global existence tic coefficient between the existence and the nonexistence of
of solutions and nontrivial steady states were also studied. stationary pattern.
Wang and Hillen [8] established the global existence of Guo and Hwang [26] investigated nonlinear dynamics
classical solutions to a generalized chemotaxis model, which near an unstable constant equilibrium in the classical Keller-
includes the volume-filling effect expressed through a nonlin- Segel model. Their results can be interpreted as a rigorous
ear squeezing probability. Necessary and sufficient conditions mathematical characterization for the early-stage pattern
for spatial pattern formation were given and the underlying formation in the Keller-Segel model. Very recently, Fu and Liu
bifurcations were analyzed. In [9], the stationary solutions of in [22] and [27] studied instability in the Keller-Segel model
the volume-filling chemotaxis model without a growth term with a logistic source and cubic source term, respectively.
were obtained by Jiang and Zhang. Their results indicated that chemotaxis-driven nonlinear
Moreover, Wrzosek [10] considered various assumptions instability occurs in these models.
on nonlinear diffusion and chemotactic sensitivity function In this paper, we mainly consider the nonlinear instability
which lead to the existence of global in time solutions, for the following chemotaxis model:
thus preventing blow-up. In [11], Winkler proved that if
certain conditions were fulfilled, then there were solutions
𝑈
that blow up in either finite or infinite time. In particular, 𝑈𝑡 = ∇ (𝑑1 ∇𝑈 − 𝜒𝑈 (1 − 𝑈) ∇𝑉) + 𝜇𝑈 (1 − ),
in the framework of chemotaxis models incorporating a 𝑈𝑐
volume-filling effect in the sense of Painter and Hillen [1], (2)
his results indicated how strongly the cellular movement 𝑥 ∈ T 𝑑, 𝑡 > 0,
must be inhibited at large cell densities in order to rule
out chemotactic collapse. Winkler and Djie [12] discussed 𝑉𝑡 = 𝑑2 ∇2 𝑉 + 𝛼𝑈 − 𝛽𝑉, 𝑥 ∈ T 𝑑, 𝑡 > 0
boundedness and finite-time collapse for a chemotaxis sys-
tem with volume-filling effect. Wang et al. [13] proved that which is subject to the Neumann boundary conditions
for a wide range of nonlinear diffusion operators, including
singular and degenerate ones, if the taxis force was strong 𝜕𝑈 𝜕𝑉
enough with respect to diffusion and the initial data were = = 0, at 𝑥𝑖 = 0, 𝜋, 1 < 𝑖 < 𝑑 (3)
𝜕𝑥𝑖 𝜕𝑥𝑖
chosen properly, then there was a classical solution which
reaches the threshold at the maximal time of its existence;
and the nonnegative initial data
no matter whether the latter was finite or infinite. Zhang
and Zheng [14] obtained the crucial uniform boundedness of
the solution for a quasilinear nonuniform parabolic system 𝑈 (𝑥, 0) = 𝑈0 (𝑥) ≥ 0, 𝑉 (𝑥, 0) = 𝑉0 (𝑥) ≥ 0,
modelling chemotaxis with volume-filling effect and the (4)
results on convergence to equilibrium and the decay rate 𝑥 ∈ T 𝑑,
using a suitable nonsmooth Simon-Łojasiewicz approach. In
[15], the uniform boundedness, global in time existence and where T 𝑑 = (0, 𝜋)𝑑 (𝑑 = 1, 2, 3) is a 𝑑-dimensional box.
uniqueness of classical solution, were proved. With the help The term 𝜇𝑈(1 − 𝑈/𝑈𝑐 ) describes the logistic growth of cells
of a suitable nonsmooth Simon-Łojasiewicz approach, the with growth rate 𝜇 > 0 and carrying capacity 𝑈𝑐 fulfilling
results on convergence of the solution to equilibrium and the 0 < 𝑈𝑐 ≤ 1. Our main result (see Theorem 6) indicates
convergence rate were obtained. Li and Zhang [16] classified that the nonlinear dynamics near an unstable constant equi-
the existence or nonexistence of steady state solutions of a 1-
librium points in the classical Keller-Segel model, the Keller-
D chemotaxis model with volume-filling effect. Their results
Segel model with a logistic source and cubic source term,
provided insights on how the biological parameters affect
respectively, and a volume-filling chemotaxis model with
pattern formation.
logistic source term are almost similar.
The chemotaxis models with logistic growth but without
a volume-filling effect were studied (see [17–22]). The global The organization of this paper is as follows: in Section 2,
attractor and traveling wave solutions of a volume-filling we show that the unique positive equilibrium point of (2)
chemotaxis model with logistic growth were obtained in [3] without chemotaxis is globally asymptotically stable and
and [23], respectively. Ma et al. [24] studied the existence cross diffusion cannot induce the instability of the positive
of stationary solutions of a volume-filling chemotaxis model equilibrium. In Section 3, we consider the growing modes
with logistic cell growth. Moreover, based on an explicit of (2). In Section 4, we present and prove the Bootstrap
formula for the stationary solutions, which is derived by lemma, which was first introduced in [28]. In Section 5, for
asymptotic bifurcation analysis, the stability criteria were any given general perturbation of magnitude 𝛿, we prove that
established and a selection mechanism of the principal its nonlinear evolution is dominated by the corresponding
wave modes for the stable stationary solution by estimating linear dynamics along a fixed finite number of fastest growing
the leading term of the principal eigenvalue was found. modes, over a time period of the order ln(1/𝛿). Each initial
Quite recently, Ma et al. in [25] studied the nonexistence perturbation certainly can behave drastically different from
of nonconstant steady state (i.e., stationary pattern) for another, which gives rise to the richness of patterns.
Abstract and Applied Analysis 3

2. Stability of Positive Equilibrium Point of Let W = [𝑈, 𝑉] be the unique nonnegative global solu-
(2) without Chemotaxis tion. The maximum principle gives

We first discuss the following corresponding kinetic equa-

tions of (2): 0 ≤ 𝑈 (𝑥, 𝑡) ≤ max {𝑈𝑐 , sup 𝑈0 (𝑥)} ,
T𝑑
𝑈 (9)
𝑈𝑡 = 𝜇𝑈 (1 − ), 𝑡 > 0, 𝛼 𝛼
𝑈𝑐 (5) 0 ≤ 𝑉 (𝑥, 𝑡) ≤ max { 𝑈𝑐 , sup 𝑈0 (𝑥) , sup 𝑉0 (𝑥)} .
𝛽 𝛽 T𝑑 T𝑑
𝑉𝑡 = 𝛼𝑈 − 𝛽𝑉, 𝑡 > 0.
Moreover, by the strong maximum principle [31], we know
We use [⋅, ⋅] to denote a column vector. Evidently, (5) has 𝑑
that if 𝑈0 , 𝑉0 ≥ ( ≢
)0, then 𝑈(𝑥, 𝑡) > 0, 𝑉(𝑥, 𝑡) > 0 on T for
the unique positive equilibrium point W = [𝑈, 𝑉] = [𝑈𝑐 , all 𝑡 > 0.
(𝛼/𝛽)𝑈𝑐 ]. For simplicity, we denote F(W) = [𝜇𝑈(1 − 𝑈/𝑈𝑐 ),
We define the Lyapunov function
𝛼𝑈 − 𝛽𝑉], and a direct calculation yields
2
−𝜇 0 𝑈 𝛼
FW (W) = ( ). (6) 𝐸 (𝑡) = ∫ [𝑝 (𝑈 − 𝑈𝑐 − 𝑈𝑐 ln ) + (𝑉 − 𝑈𝑐 ) ] 𝑑𝑥,
𝛼 −𝛽 T𝑑 𝑈𝑐 𝛽
(10)
The characteristic polynomial of FW (W) is 𝜑(𝜆) = (𝜆 +
𝜇)(𝜆 + 𝛽); −𝜇 and −𝛽 are the two roots of 𝜑(𝜆) = 0. Hence where 𝑝 = 2𝛼2 𝑈𝑐 /𝛽𝜇. Then 𝐸(𝑡) ≥ 0 for all 𝑡 ≥ 0. Applying
[𝑈𝑐 , (𝛼/𝛽)𝑈𝑐 ] is locally stable. Define 𝐸(𝑡) = 𝑝(𝑈 − 𝑈𝑐 − (7) and integration by parts, we have
𝑈𝑐 ln(𝑈/𝑈𝑐 )) + (𝑉 − (𝛼/𝛽)𝑈𝑐 )2 , where 𝑝 = 2𝛼2 𝑈𝑐 /𝛽𝜇. By the
Lyapunov-LaSalle invariance principle [29], [𝑈𝑐 , (𝛼/𝛽)𝑈𝑐 ] is 𝑝𝑑1 𝑈𝑐
𝐸󸀠 (𝑡) = − ∫ { |∇𝑈|2 + 2𝑑2 |∇𝑉|2 } 𝑑𝑥
globally asymptotically stable. T𝑑 𝑈2
We now consider system (2) without chemotaxis in the
following form: 𝑝𝜇 2 𝛼
−∫ { (𝑈 − 𝑈𝑐 ) − 2𝛼 (𝑈 − 𝑈𝑐 ) (𝑉 − 𝑈𝑐 )
T𝑑 𝑈𝑐 𝛽
𝑈
𝑈𝑡 = 𝑑1 Δ𝑈 + 𝜇𝑈 (1 − ), 𝑥 ∈ T 𝑑 (𝑑 = 1, 2, 3) , 𝛼 2
𝑈𝑐 + 2𝛽(𝑉 − 𝑈𝑐 ) } 𝑑𝑥.
𝛽
𝑡 > 0,
𝛼2 2
𝑉𝑡 = 𝑑2 Δ𝑉 + 𝛼𝑈 − 𝛽𝑉, 𝑥∈T 𝑑
(𝑑 = 1, 2, 3) , (7) ≤− ∫ (𝑈 − 𝑈𝑐 ) 𝑑𝑥
𝛽 T𝑑
𝑡 > 0, 2
𝛼
− 𝛽 ∫ (𝑉 − 𝑈𝑐 ) 𝑑𝑥.
𝜕𝑈 𝜕𝑉 T𝑑 𝛽
= = 0, at 𝑥𝑖 = 0, 𝜋, for 1 ≤ 𝑖 ≤ 𝑑.
𝜕𝑥𝑖 𝜕𝑥𝑖 (11)

Let 0 = 𝜇1 < 𝜇2 < 𝜇3 < ⋅ ⋅ ⋅ be the eigenvalues of the oper- By (9), (11), the basic estimates for parabolic equations [31],
ator −Δ on T 𝑑 (𝑑 = 1, 2, 3) with the homogeneous Neumann and Lemma 2.1 in [22] (which is given in [32] in Chinese), we
boundary condition and 𝐸(𝜇𝑖 ) the eigenspace corresponding can conclude that
to 𝜇𝑖 in 𝐿2 (T 𝑑 ). Let 𝑋 = [𝐿2 (T 𝑑 )]2 , {𝜙𝑖𝑗 : 𝑗 = 1, . . . , dim 𝐸(𝜇𝑖 )} 󵄩 󵄩
lim 󵄩󵄩𝑈(⋅, 𝑡) − 𝑈𝑐 󵄩󵄩󵄩𝐿2 (T 𝑑 ) = 0,
be an orthonormal basis of 𝐸(𝜇𝑖 ), and 𝑋𝑖𝑗 = {c ⋅ 𝜙𝑖𝑗 | c ∈ R2 }. 𝑡 → ∞󵄩

Then 𝑋 = ⊕∞
dim 𝐸(𝜇𝑖 ) 󵄩󵄩 𝛼 󵄩󵄩󵄩 (12)
𝑖=1 𝑋𝑖 , 𝑋𝑖 = ⊕𝑗=1 𝑋𝑖𝑗 . 󵄩
lim 󵄩󵄩󵄩𝑉(⋅, 𝑡) − 𝑈𝑐 󵄩󵄩󵄩 = 0.
Let D = diag(𝑑1 , 𝑑2 ) and L = DΔ + FW (W). The lineari- 𝑡 → ∞󵄩 󵄩 𝛽 󵄩󵄩𝐿2 (T 𝑑 )
zation of (7) at [𝑈, 𝑉] can be expressed by W𝑡 = L(W − W).
For each 𝑖 ≥ 1, 𝑋𝑖 is invariant under the operator L, and 𝜆 is The global asymptotic stability of [𝑈𝑐 , (𝛼/𝛽)𝑈𝑐 ] follows from
an eigenvalue of L on 𝑋𝑖 if and only if it is an eigenvalue of (12) together with the local stability of [𝑈𝑐 , (𝛼/𝛽)𝑈𝑐 ].
the matrix Next, we consider that the cross diffusion model

−𝜇 𝑑 − 𝜇 0 𝑈
−𝜇𝑖 D + FW (W) = ( 𝑖 1 ). (8) 𝑈𝑡 = 𝑑1 Δ (𝑈 + 𝑑3 𝑈𝑉) + 𝜇𝑈 (1 − ),
𝛼 −𝜇𝑖 𝑑2 − 𝛽 𝑈𝑐
Thus, −𝜇𝑖 D + FW (W) has two negative eigenvalues −𝜇𝑖 𝑑1 − 𝑥 ∈ T 𝑑, 𝑡 > 0,
𝜇 and −𝜇𝑖 𝑑2 − 𝛽. It follows from Theorem 5.1.1 in [30] that
[𝑈𝑐 , (𝛼/𝛽)𝑈𝑐 ] is locally asymptotically stable. 𝑉𝑡 = 𝑑2 Δ𝑉 + 𝛼𝑈 − 𝛽𝑉, 𝑥 ∈ T 𝑑 , 𝑡 > 0,
4 Abstract and Applied Analysis

𝜕𝑈 𝜕𝑉 Let w(x, 𝑡) ≡ [𝑢(x, 𝑡), V(x, 𝑡)], q = (𝑞1 , . . . , 𝑞𝑑 ) ∈ Ω =

= = 0, at 𝑥𝑖 = 0, 𝜋, for 1 ≤ 𝑖 ≤ 𝑑,
𝜕𝑥𝑖 𝜕𝑥𝑖 (N ∪ {0})𝑑 , and 𝑒q (x) = ∏𝑑𝑖=1 cos(𝑞𝑖 𝑥𝑖 ). Then {𝑒q (x)}q∈Ω forms
𝑈 (𝑥, 0) = 𝑈0 (𝑥) ≥ 0, 𝑉 (𝑥, 0) = 𝑉0 (𝑥) ≥ 0, a basis of the space of functions in T 𝑑 that satisfies Neumann
boundary conditions. We try to find a normal mode to the
𝑥 ∈ T 𝑑. linear system (19) of the following form:
(13)
w (x, 𝑡) = rq 𝑒𝜆 q 𝑡 𝑒q (x) , (20)
Let Φ(W) = [𝑑1 (𝑈 + 𝑑3 𝑈𝑉), 𝑑2 𝑉]. Then the linearizing
system (13) at W can be written as where rq is a vector depending on q. Substituting (20) into
(19) yields
W𝑡 = (ΦW Δ + FW (W)) W, (14)
−𝑑 𝑞2 − 𝜇 𝜒𝑈𝑐 (1 − 𝑈𝑐 ) 𝑞2
𝜆 q rq = ( 1 ) rq , (21)
where 𝛼 −𝑑2 𝑞2 − 𝛽
𝑑1 (1 + 𝑑3 𝑉) 𝑑1 𝑑3 𝑈
ΦW (W) = ΦW = ( ). (15) where 𝑞2 = ∑𝑑𝑖=1 𝑞𝑖2 . We can obtain the following dispersion
0 𝑑2
formula for 𝜆 q :
Then, for each 𝑖 ∈ {1, 2, . . .}, 𝑋𝑖 is invariant under the operator
ΦW Δ + FW (W), and 𝜉 is an eigenvalue of ΦW Δ + FW (W) on 𝜆2q + ((𝑑1 + 𝑑2 ) 𝑞2 + 𝛽 + 𝜇) 𝜆 q
𝑋𝑖 if and only if 𝜉 is an eigenvalue of the matrix
+ 𝑞2 [𝑑1 𝑑2 𝑞2 + (𝑑1 𝛽 + 𝑑2 𝜇 − 𝛼𝜒𝑈𝑐 (1 − 𝑈𝑐 ))] (22)
−𝑑 (1 + 𝑑3 𝑉) 𝜇𝑖 − 𝜇 −𝑑1 𝑑3 𝑈𝜇𝑖
A𝑖 = ( 1 ). (16)
𝛼 −𝑑2 𝜇𝑖 − 𝛽 + 𝜇𝛽 = 0.
Notice that Thus, we deduce the following linear instability criterion by
det A𝑖 = 𝑑1 𝑑2 (1 + 𝑑3 𝑉) 𝜇𝑖2 requiring that there exist a q such that

+ (𝑑2 𝜇 + 𝛽𝑑1 (1 + 𝑑3 𝑉) + 𝑑1 𝑑3 𝛼𝑈) 𝜇𝑖 𝑞2 [𝑑1 𝑑2 𝑞2 + (𝑑1 𝛽 + 𝑑2 𝜇 − 𝛼𝜒𝑈𝑐 (1 − 𝑈𝑐 ))] + 𝜇𝛽 < 0

(17) (23)
+ 𝜇𝛽 > 0,
to ensure that (22) has at least one positive root 𝜆 q . This
Tr A𝑖 = − ((𝑑1 + 𝑑2 + 𝑑1 𝑑3 𝑉) 𝜇𝑖 + 𝜇 + 𝛽) < 0. means that
2
Thus, the two eigenvalues 𝜉𝑖+ and 𝜉𝑖− of A𝑖 have negative real (𝑞2 (𝑑1 − 𝑑2 ) − 𝛽 + 𝜇) + 4𝛼𝜒𝑈𝑐 (1 − 𝑈𝑐 ) 𝑞2 > 0. (24)
parts.
From this, we see that adding the cross diffusion to the There exist two distinct real roots:
system (7), the positive constant solution is also locally stable,
𝜆±q
which means that Turing instability does not occur. The infor-
mation above indicates that the aggregation of individuals
does not occur in the absence of chemotactic effect. It is = ( − (𝑞2 (𝑑1 + 𝑑2 ) + 𝛽 + 𝜇)
the purpose of the present paper to clarify the effect of the
chemotaxis and nonlinear patterns created by chemotaxis for 2
±√(𝑞2 (𝑑1 − 𝑑2 ) − 𝛽 + 𝜇) + 4𝛼𝜒𝑈𝑐 (1 − 𝑈𝑐 ) 𝑞2 )
a volume-filling chemotaxis model with logistic growth.
× (2)−1
3. Growing Modes in the System (2) (25)
The nonlinear evolution of a perturbation 𝑢(x, 𝑡) = 𝑈(x, 𝑡) − for all q ≠
0 to the quadratic equation (22), which are
𝑈, V(x, 𝑡) = 𝑉(x, 𝑡) − 𝑉 around [𝑈, 𝑉] satisfies denoted by 𝜆−q < 𝜆+q . We denote the corresponding (linearly
𝑢𝑡 = 𝑑1 ∇2 𝑢 − 𝜒𝑈𝑐 (1 − 𝑈𝑐 ) ∇2 V − 𝜇𝑢 independent) eigenvectors by r− (q) and r+ (q), such that

𝜇𝑢2 𝜆±q + 𝑑2 𝑞2 + 𝛽
− 𝜒∇ [𝑢 (1 − 𝑢 − 2𝑈𝑐 ) ∇V] − , (18) r± (q) = [ , 1] . (26)
𝑈𝑐 𝛼

V𝑡 = 𝑑2 ∇2 V + 𝛼𝑢 − 𝛽V. Clearly, for 𝑞 large,

The corresponding linearized system is as follows:
𝑞2 [𝑑1 𝑑2 𝑞2 + (𝑑1 𝛽 + 𝑑2 𝜇 − 𝛼𝜒𝑈𝑐 (1 − 𝑈𝑐 ))] + 𝜇𝛽 > 0. (27)
2 2
𝑢𝑡 = 𝑑1 ∇ 𝑢 − 𝜒𝑈𝑐 (1 − 𝑈𝑐 ) ∇ V − 𝜇𝑢,
(19) Thus, there are only finitely many q such that 𝜆+q > 0.
V𝑡 = 𝑑2 ∇2 V + 𝛼𝑢 − 𝛽V. We denote the largest eigenvalue by 𝜆 max > 0 and define
Abstract and Applied Analysis 5

Ωmax ≡ {q ∈ Ω | 𝜆+q = 𝜆 max }. From (25) we can regard 𝜆+q as a It follows from (29) that
function of 𝑞2 . Therefore, there is one 𝑞2 (possibly two) having 󵄨󵄨 󵄨 󵄨 󵄨
𝜆+ (𝑞2 ) = 𝜆 max . We also denote 𝜌 > 0 to be the gap between 󵄨󵄨 ± 󵄨󵄨 󵄨r± (q)󵄨󵄨󵄨 × 󵄨󵄨󵄨󵄨wq 󵄨󵄨󵄨󵄨
󵄨󵄨𝑤q 󵄨󵄨 ≤ 󵄨 󵄨
󵄨 󵄨 󵄨󵄨det [r (q) , r (q)]󵄨󵄨󵄨 .
(36)
the 𝜆 max and the rest; that is, 𝜌 = minq∈Ω\Ωmax |𝜆 max − 𝜆 q |. 󵄨 − + 󵄨
Given any initial perturbation w(x, 0), we can expand it
as From (26) and (34), we can conclude that there exists a
positive constant 𝐶2 such that
w (x, 0) = ∑ wq 𝑒q (x) = ∑ {𝑤q− r− (q) + 𝑤q+ r+ (q)} 𝑒q (x) ,
2
q∈Ω q∈Ω
󵄨󵄨 󵄨󵄨 √ 𝜆±q 𝑑 𝛽
(28) r
󵄨󵄨 ± (q)󵄨󵄨 = [( + 2 ) 𝑞2 + ] + 1 ≤ 𝐶2 𝑞2 (37)
𝛼𝑞2 𝛼 𝛼
wq = 𝑤q− r− (q) + 𝑤q+ r+ (q) . (29)
for all 𝑞 > 0, where 𝐶2 = 2 max{𝐶1 + 𝑑2 /𝛼, 1 + 𝛽/𝛼}. By (34),
In the sequel, denote by ⟨⋅, ⋅⟩ and (⋅, ⋅) the inner product one can deduce
2
of [𝐿2 (T 𝑑 )] and the scalar product of R2 , respectively. For 1 1 𝛼
2
any g(⋅, 𝑡) ∈ [𝐿2 (T 𝑑 )] , we denote ‖g(⋅, 𝑡)‖ ≡ ‖g(⋅, 𝑡)‖𝐿2 . 󵄨󵄨 󵄨󵄨 ≤ 2𝑞 √ . (38)
󵄨󵄨det [r− (q) , r+ (q)]󵄨󵄨 𝜒𝑈𝑐 (1 − 𝑈𝑐 )
Throughout this paper, we always denote universal constants
by 𝐶𝑖 (𝑖 = 1, 2, . . .). Combining (37) and (38), we find that
Clearly,
󵄨󵄨 ± 󵄨󵄨 󵄨 󵄨
󵄨󵄨𝑤q 󵄨󵄨 ≤ 𝐶3 𝑞 󵄨󵄨󵄨wq 󵄨󵄨󵄨 , (39)
𝜋 𝑑
󵄨 󵄨󵄨2 󵄨 󵄨 󵄨 󵄨
‖w (x, 0)‖2 = ( ) ∑ 󵄨󵄨󵄨󵄨wq 󵄨󵄨󵄨 . (30)
2 q∈Ω where 𝐶3 = (𝐶2 /2)√𝛼/𝜒𝑈𝑐 (1 − 𝑈𝑐 ). For 𝑡 ≥ 1 and 𝑞 large, it
is not difficult to verify by (33), (37), and (39) that
The unique solution w(x, 𝑡) = [𝑢(x, 𝑡), V(x, 𝑡)] of (19) is
󵄨󵄨 ± ± 󵄨
󵄨󵄨𝑤 r± (q) 𝑒𝜆 q 𝑡 󵄨󵄨󵄨 ≤ 𝐶2 𝐶3 𝑞3 󵄨󵄨󵄨󵄨wq 󵄨󵄨󵄨󵄨 𝑒𝜆 q 𝑡
±
given by
󵄨󵄨 q 󵄨󵄨 󵄨 󵄨
− +
w (x, 𝑡) = ∑ {𝑤q− r− (q) 𝑒𝜆 q 𝑡 + 𝑤q+ r+ (q) 𝑒𝜆 q 𝑡 } 𝑒q (x) 󵄨 󵄨 𝑞3 󵄨 󵄨
q∈Ω
(31) ≤ 𝐶4 󵄨󵄨󵄨󵄨wq 󵄨󵄨󵄨󵄨 2
≤ 𝐶5 󵄨󵄨󵄨󵄨wq 󵄨󵄨󵄨󵄨 .
exp (min {𝑑1 , 𝑑2 } 𝑞 )
L𝑡
≡ 𝑒 w (x, 0) . (40)

Our main result in this section is the following lemma. In view of (30) and (40), we observe that

Lemma 1. Suppose that the instability criterion (23) holds. Let ‖w (x, 𝑡)‖ ≤ 2𝐶5 𝑒𝜆 max 𝑡 ‖w (x, 0)‖ , for 𝑡 ≥ 1. (41)
w(x, 𝑡) ≡ 𝑒L𝑡 w(x, 0) be a solution to the linearized system (19)
̂1 ≥
with initial condition w(x, 0). Then there exists a constant 𝐶
(2) 𝑡 ≤ 1. Multiplying the first equation of (19) by 𝑢
1 depending on 𝑑1 , 𝑑2 , 𝑈𝑐 , 𝜒, 𝜇, 𝛼, and 𝛽, such that
and the second by 𝐾V, adding them together, and
̂1 𝑒𝜆 max 𝑡 ‖w (⋅, 0)‖ , integrating the result in T 𝑑 , we have
‖w (⋅, 𝑡)‖ ≤ 𝐶 ∀𝑡 ≥ 0. (32)
1 𝑑
Proof. We divide the proof into the following two cases. ∫ {|𝑢|2 + 𝐾|V|2 } dx
2 𝑑𝑡 T 𝑑
(1) 𝑡 ≥ 1. It follows from (25) that
+ ∫ {𝑑1 |∇𝑢|2 + 𝐾𝑑2 |∇V|2
𝜆±q T𝑑
lim = −𝑑1 , −𝑑2 , (33)
q → ∞ 𝑞2
− 𝜒𝑈𝑐 (1 − 𝑈𝑐 ) ∇𝑢∇V} dx (42)

respectively. Thus, 𝜆±q ≤ − min{𝑑1 , 𝑑2 }𝑞2 and there exists a

= −𝜇 ∫ 𝑢2 dx − 𝐾𝛽 ∫ V2 dx
positive constant 𝐶1 for all q > 0, such that T𝑑 T𝑑

󵄨󵄨 𝜆± 󵄨󵄨
󵄨󵄨 q 󵄨󵄨 + 𝛼𝐾 ∫ 𝑢V dx.
󵄨󵄨 󵄨󵄨 ≤ 𝐶1 . (34)
󵄨󵄨 𝑞2 󵄨󵄨 T𝑑
󵄨󵄨 󵄨󵄨
Let
By the quadratic formula of (22), one can obtain
2
[𝜒𝑈𝑐 (1 − 𝑈𝑐 )]
󵄨󵄨 + 󵄨 𝐾= . (43)
󵄨󵄨𝜆 q − 𝜆−q 󵄨󵄨󵄨 ≥ 2𝑞√𝛼𝜒𝑈𝑐 (1 − 𝑈𝑐 ). (35) 𝑑1 𝑑2
󵄨 󵄨
6 Abstract and Applied Analysis

Then the integrand of the second integral can be estimated as Lemma 4. Let [𝑢(x, 𝑡), V(x, 𝑡)] be a solution of (18). Then
follows: 2
1 𝑑 󵄨 󵄨2 [𝜒𝑈𝑐 (1 − 𝑈𝑐 )] 󵄨󵄨 𝛼 󵄨󵄨2
∑ ∫ {󵄨󵄨𝐷𝛼 𝑢󵄨󵄨󵄨 + 󵄨󵄨𝐷 V󵄨󵄨 } dx
𝑑1 |∇𝑢|2 + 𝐾𝑑2 |∇V|2 − 𝜒𝑈𝑐 (1 − 𝑈𝑐 ) ∇𝑢∇V 2 𝑑𝑡 |𝛼|=2 T 𝑑 󵄨 𝑑1 𝑑2
2 (44) 2
𝑑 [𝜒𝑈𝑐 (1 − 𝑈𝑐 )] 𝑑1 󵄨󵄨 𝛼 󵄨󵄨2 [𝜒𝑈𝑐 (1 − 𝑈𝑐 )] 󵄨󵄨 𝛼 󵄨󵄨2
≥ 1 |∇𝑢|2 + |∇V|2 ≥ 0. + ∑ ∫ { 󵄨∇𝐷 𝑢󵄨󵄨 + 󵄨󵄨∇𝐷 V󵄨󵄨 } dx
2 2𝑑1 |𝛼|=2 T𝑑 4󵄨 2𝑑1
2
Using Young inequality, we deduce that 󵄨 󵄨2 𝛽[𝜒𝑈𝑐 (1 − 𝑈𝑐 )]
+ 𝜇 ∑ ∫ 󵄨󵄨󵄨𝐷𝛼 𝑢󵄨󵄨󵄨 dx +
|𝛼|=2 T
𝑑 2𝑑1 𝑑2
1 𝑑 󵄨 󵄨 𝛼√𝐾
∫ {|𝑢|2 + 𝐾 󵄨󵄨󵄨󵄨V2 󵄨󵄨󵄨󵄨} dx ≤ ∫ {|𝑢|2 + 𝐾|V|2 } dx.
2 𝑑𝑡 T 𝑑 2 󵄨 󵄨2
T𝑑
× ∑ ∫ 󵄨󵄨󵄨𝐷𝛼 V󵄨󵄨󵄨 dx
(45) T𝑑
|𝛼|=2

It follows from Gronwall inequality that ̂2 (‖w‖𝐻2 + ‖w‖2 2 ) 󵄩󵄩󵄩󵄩∇3 w󵄩󵄩󵄩󵄩2 + 𝐶

≤𝐶 ̂3 ‖𝑢‖2 ,
𝐻 󵄩 󵄩
(51)
∫ {|𝑢 (x, 𝑡)|2 + 𝐾|V (x, 𝑡)|2 } dx
T𝑑 ̂3 = 𝐶3 𝛼3 [𝜒𝑈𝑐 (1 − 𝑈𝑐 )]6 /2𝛽3 𝑑5 𝑑3 .
where 𝐶 0 1 2
(46)
√𝐾𝑡
≤ 𝑒𝛼 ∫ {|𝑢 (x, 𝑡)|2 + 𝐾|V (x, 𝑡)|2 } dx. Proof. It is known by Lemma 3 that
T𝑑
𝑢̃𝑡 = 𝑑1 ∇2 𝑢̃ − 𝜒𝑈𝑐 (1 − 𝑈𝑐 ) ∇2 ̃V − 𝜇̃
𝑢
If 𝐾 ≥ 1 and 𝑡 ≤ 1, then it is clear from (46) that
𝑢2
𝜇̃
𝑢 (1 − 𝑢̃ − 2𝑈𝑐 ) ∇̃V] −
− 𝜒∇ [̃ ,
‖w (x, 𝑡)‖ ≤ √𝐾𝑒𝛼√𝐾 𝑒𝜆 max 𝑡 ‖w (x, 0)‖ . (47) 𝑈𝑐 (52)
𝑑
Similarly, if 0 < 𝐾 < 1, 𝑡 ≤ 1, then from (46), one has 𝑥 ∈ 2T , 𝑡 > 0,

̃V𝑡 = 𝑑2 ∇2 ̃V + 𝛼̃
𝑢 − 𝛽̃V, 𝑥 ∈ 2T 𝑑 , 𝑡 > 0,
‖w (x, 𝑡)‖ ≤ √𝑒𝛼 /𝐾𝑒𝜆 max 𝑡 ‖w (x, 0)‖ . (48)
𝜕̃
𝑢 𝜕̃V
Let 𝐶 ̂1 =
̂1 = max{2𝐶5 , √𝑒𝛼 /𝐾} ≥ 1 if 0 < 𝐾 < 1, and let 𝐶 = = 0, at 𝑥𝑖 = −𝜋, 0, 𝜋, for 1 ≤ 𝑖 ≤ 𝑑, (53)
𝜕𝑥𝑖 𝜕𝑥𝑖
√
max{2𝐶 , 𝐾𝑒 𝛼 √𝐾
} ≥ 1 if 𝐾 ≥ 1. Then
5 where [̃ 𝑢(x, 𝑡), ̃V(x, 𝑡)] is the even extension of [𝑢(x, 𝑡), V(x, 𝑡)]
on 2T 𝑑 . Taking the second-order derivative of (52) for
̂1 𝑒𝜆 max 𝑡 ‖w (x, 0)‖ .
‖w (x, 𝑡)‖ ≤ 𝐶 (49) 𝑥𝑖 , 𝑥𝑗 and making inner product with 𝜕𝑥𝑖 𝑥𝑗 𝑢̃ and 𝐾𝜕𝑥𝑖 𝑥𝑗 ̃V,
This completes the proof of Lemma 1. respectively, and adding them together, we deduce that
1 𝑑 󵄨󵄨 󵄨󵄨2 󵄨󵄨 󵄨󵄨2
∫ {󵄨󵄨󵄨𝜕𝑥𝑖 𝑥𝑗 𝑢̃󵄨󵄨󵄨 + 𝐾󵄨󵄨󵄨𝜕𝑥𝑖 𝑥𝑗 ̃V󵄨󵄨󵄨 } dx
4. Bootstrap Lemma 2 𝑑𝑡 2T 𝑑 󵄨 󵄨 󵄨 󵄨
󵄨󵄨 󵄨󵄨2 󵄨󵄨 󵄨󵄨2
By a standard PDE theory [31], we can establish the existence +∫ {𝑑1 󵄨󵄨󵄨∇𝜕𝑥𝑖 𝑥𝑗 𝑢̃󵄨󵄨󵄨 + 𝐾𝑑2 󵄨󵄨󵄨∇𝜕𝑥𝑖 𝑥𝑗 ̃V󵄨󵄨󵄨
2T 𝑑 󵄨 󵄨 󵄨 󵄨
of local solutions for (18).

Lemma 2 (Local existence). For 𝑠 ≥ 1 (𝑑 = 1) and 𝑠 ≥ 2 (𝑑 = − 𝜒𝑈𝑐 (1 − 𝑈𝑐 ) ∇𝜕𝑥𝑖 𝑥𝑗 𝑢̃ ⋅ ∇𝜕𝑥𝑖 𝑥𝑗 ̃V} dx

2, 3), there exists a 𝑇0 > 0 such that (18) with 𝑢(⋅, 0), V(⋅, 0) ∈
󵄨󵄨 󵄨2 󵄨 󵄨2
𝐻𝑠 has a unique solution w(⋅, 𝑡) on (0, 𝑇0 ) which satisfies + 𝜇∫ 󵄨󵄨𝜕𝑥 𝑥 𝑢̃󵄨󵄨󵄨 dx + 𝛽𝐾 ∫ 󵄨󵄨󵄨𝜕𝑥 𝑥 ̃V󵄨󵄨󵄨 dx
2T 𝑑 󵄨
󵄨 𝑖 𝑗 󵄨
󵄨 󵄨
2T 𝑑 󵄨
𝑖 𝑗 󵄨󵄨
‖w(𝑡)‖𝐻𝑠 ≤ 𝐶‖w(0)‖𝐻𝑠 , 0 < 𝑡 < 𝑇0 , (50)
= 𝜒 (1 − 2𝑈𝑐 ) ∫ ∇𝜕𝑥𝑖 𝑥𝑗 𝑢̃ ⋅ 𝜕𝑥𝑖 𝑥𝑗 (̃
𝑢 ⋅ ∇̃V) dx (54)
where 𝐶 is a positive constant depending on 𝑑1 , 𝑑2 , 𝑈𝑐 , 𝛼, 𝛽, 2T 𝑑

and 𝜒.
− 𝜒∫ 𝑢2 ∇̃V) dx
∇𝜕𝑥𝑖 𝑥𝑗 𝑢̃ ⋅ 𝜕𝑥𝑖 𝑥𝑗 (̃
2T 𝑑
It is not difficult to verify the following result.
+ 𝐾𝛼 ∫ 𝜕𝑥𝑖 𝑥𝑗 𝑢̃ ⋅ 𝜕𝑥𝑖 𝑥𝑗 ̃V dx
Lemma 3. Let w(x, 𝑡) be a solution of (18). Then the even 2T 𝑑
extension of w(x, 𝑡) on 2T 𝑑 = (−𝜋, 𝜋)𝑑 (𝑑 = 1, 2, 3) is also 2𝜇 2
the solution of (18) which satisfies homogeneous Neumann − ∫ [̃ 𝑢(𝜕𝑥𝑖 𝑥𝑗 𝑢̃) + 𝜕𝑥𝑖 𝑢̃ ⋅ 𝜕𝑥𝑗 𝑢̃ ⋅ 𝜕𝑥𝑖 𝑥𝑗 𝑢̃] dx
𝑈𝑐 2T 𝑑
boundary conditions and periodical boundary conditions on
2T 𝑑 = (−𝜋, 𝜋)𝑑 (𝑑 = 1, 2, 3). ≡ 𝐼1 + 𝐼2 + 𝐼3 + 𝐼4 .
Abstract and Applied Analysis 7

We can apply Young inequality and (43) to get Combining (56), (57), and (62), we observe that
󵄨󵄨 󵄨󵄨2 󵄨󵄨 󵄨󵄨2 󵄨 󵄨 󵄩 󵄩2
𝑑1 󵄨󵄨󵄨∇𝜕𝑥𝑖 𝑥𝑗 𝑢̃󵄨󵄨󵄨 + 𝐾𝑑2 󵄨󵄨󵄨∇𝜕𝑥𝑖 𝑥𝑗 ̃V󵄨󵄨󵄨 w‖𝐻2 󵄩󵄩󵄩󵄩∇3 w
∑ 𝐼1 ≤ 𝜒 󵄨󵄨󵄨1 − 2𝑈𝑐 󵄨󵄨󵄨 𝐶14 ‖̃ ̃󵄩󵄩󵄩󵄩 , (63)
󵄨 󵄨 󵄨 󵄨 |𝛼|=2
− 𝜒𝑈𝑐 (1 − 𝑈𝑐 ) ∇𝜕𝑥𝑖 𝑥𝑗 𝑢̃ ⋅ ∇𝜕𝑥𝑖 𝑥𝑗 ̃V
(55) where 𝐶14 = 𝐶6 [(1 + 2𝑑)𝐶13 + 1].
2 By Hölder inequality, it follows from (57) and (62) that
𝑑1 󵄨󵄨󵄨 󵄨󵄨2 [𝜒𝑈𝑐 (1 − 𝑈𝑐 )] 󵄨󵄨 󵄨2
󵄨󵄨∇𝜕𝑥 𝑥 ̃V󵄨󵄨󵄨 .
≥ 󵄨󵄨∇𝜕𝑥𝑖 𝑥𝑗 𝑢̃󵄨󵄨󵄨 + 󵄨󵄨 𝑖 𝑗 󵄨󵄨
2󵄨 󵄨 2𝑑1 󵄩 󵄩2
w‖2𝐻2 󵄩󵄩󵄩󵄩∇3 w
∑ 𝐼2 ≤ 𝜒𝐶15 ‖̃ ̃󵄩󵄩󵄩󵄩 , (64)
Now we estimate each term on the right-hand sides of |𝛼|=2
(54). By using Hölder inequality,
where 𝐶15 = 𝐶6 (2𝐶9−2 𝐶12
2
𝐶13 + 6𝐶6 𝐶13 + 𝐶6 ).
󵄨 󵄨
𝐼1 ≤ 𝜒 󵄨󵄨󵄨1 − 2𝑈𝑐 󵄨󵄨󵄨 Now we estimate 𝐼3 . By interpolation for all 𝑎 > 0, it can
be proved that
󵄩󵄩 󵄩󵄩 󵄩󵄩 󵄩󵄩
× {‖∇̃V‖𝐿∞ 󵄩󵄩󵄩∇𝜕𝑥𝑖 𝑥𝑗 𝑢̃󵄩󵄩󵄩 󵄩󵄩󵄩𝜕𝑥𝑖 𝑥𝑗 𝑢̃󵄩󵄩󵄩 󵄩󵄩 󵄩2 󵄩 󵄩2 𝑢‖2
󵄩 󵄩󵄩 󵄩 󵄩󵄩𝜕𝑥 𝑥 𝑢̃󵄩󵄩󵄩 ≤ 𝐶0 (𝑎󵄩󵄩󵄩∇𝜕𝑥 𝑥 𝑢̃󵄩󵄩󵄩 + ‖̃ ). (65)
󵄩󵄩 𝑖 𝑗 󵄩󵄩 󵄩󵄩 𝑖 𝑗 󵄩󵄩 4𝑎2
𝑑 󵄩󵄩 󵄩󵄩 󵄩󵄩 󵄩󵄩 (56)
+ 2∑‖∇̃𝑢‖𝐿∞ 󵄩󵄩󵄩𝜕𝑥𝑖 𝑥𝑗 ̃V󵄩󵄩󵄩 󵄩󵄩󵄩∇𝜕𝑥𝑖 𝑥𝑗 𝑢̃󵄩󵄩󵄩 Then it is easy to see that
𝑖=1
󵄩 󵄩󵄩 󵄩
𝐾𝛽 󵄨 󵄨2
󵄩󵄩 󵄩󵄩 󵄩󵄩 󵄩󵄩 ∑ 𝐼3 ≤ ∑ ∫ 󵄨󵄨𝐷𝛼 ̃V󵄨󵄨󵄨 dx
𝑢‖𝐿∞ 󵄩󵄩󵄩∇𝜕𝑥𝑖 𝑥𝑗 𝑢̃󵄩󵄩󵄩 󵄩󵄩󵄩∇𝜕𝑥𝑖 𝑥𝑗 ̃V󵄩󵄩󵄩 } .
+ ‖̃ |𝛼|=2
2 |𝛼|=2 2T 𝑑 󵄨
󵄩 󵄩󵄩 󵄩 (66)
𝑑 󵄨 󵄨2
Notice that + 1 ∑ ∫ 󵄨󵄨󵄨∇𝐷𝛼 𝑢̃󵄨󵄨󵄨 dx + 𝐶
̂3 ‖̃
𝑢‖2 ,
4 |𝛼|=2 2T 𝑑
󵄩󵄩 󵄩󵄩 󵄩 󵄩
󵄩󵄩𝑔󵄩󵄩𝐿∞ (2T 𝑑 ) ≤ 𝐶6 󵄩󵄩󵄩𝑔󵄩󵄩󵄩𝐻2 (2T 𝑑 ) ,
where 𝐶̂3 = 𝐶3 𝛼3 [𝜒𝑈𝑐 (1 − 𝑈𝑐 )]6 /2𝛽3 𝑑5 𝑑3 and 𝑎 = 𝑑2 𝑑2 𝛽/
󵄩󵄩 󵄩󵄩 󵄩 󵄩 0 1 2 1
󵄩󵄩𝑔󵄩󵄩𝐿4 (2T 𝑑 ) ≤ 𝐶7 󵄩󵄩󵄩𝑔󵄩󵄩󵄩𝐻2 (2T 𝑑 ) , (57) 2[𝜒𝑈𝑐 (1 − 𝑈𝑐 )]2 𝐶0 𝛼.
󵄩󵄩 󵄩󵄩 󵄩 󵄩 Finally, from (57) and (61), 𝐼4 is bounded by
󵄩󵄩𝑔󵄩󵄩𝐿6 (2T 𝑑 ) ≤ 𝐶8 󵄩󵄩󵄩𝑔󵄩󵄩󵄩𝐻2 (2T 𝑑 ) ,
4𝜇 󵄩 󵄩2
for 𝑑 ≤ 3. Clearly, ∑ 𝐼4 ≤ w‖𝐻2 󵄩󵄩󵄩󵄩∇3 w
𝐶 ‖̃ ̃󵄩󵄩󵄩󵄩 . (67)
|𝛼|=2
𝑈𝑐 16
∫ ∇̃
𝑢 𝑑x = ∫ ∇̃V 𝑑x = 0,
2T 𝑑 2T 𝑑 Combining (54), (55), and (63)–(67), one can obtain
(58)
∫ 𝜕𝑥𝑖 𝑥𝑗 𝑢̃ 𝑑x = ∫ 𝜕𝑥𝑖 𝑥𝑗 ̃V 𝑑x = 0. 1 𝑑 󵄨 󵄨2 󵄨 󵄨2
∑ ∫ {󵄨󵄨𝐷𝛼 𝑢󵄨󵄨󵄨 + 𝐾󵄨󵄨󵄨𝐷𝛼 V󵄨󵄨󵄨 } dx
2T 𝑑 2T 𝑑 2 𝑑𝑡 |𝛼|=2 T 𝑑 󵄨
Using the Poincaré inequality, we have
󵄩󵄩 󵄩󵄩 󵄩 󵄩 󵄩 󵄩 𝑑1 󵄨󵄨 𝛼 󵄨󵄨2 𝐾𝑑2 󵄨󵄨 𝛼 󵄨󵄨2
󵄩󵄩𝑔󵄩󵄩 ≤ 𝐶9 󵄩󵄩󵄩𝑔󵄩󵄩󵄩𝐿4 (2T 𝑑 ) ≤ 𝐶10 󵄩󵄩󵄩𝑔󵄩󵄩󵄩𝐿6 (2T 𝑑 ) + ∑ ∫ { 󵄨∇𝐷 𝑢󵄨󵄨 + 󵄨∇𝐷 V󵄨󵄨 } dx
|𝛼|=2 T𝑑 4󵄨 2 󵄨
(59)
󵄩 󵄩 󵄩 󵄩 (68)
≤ 𝐶11 󵄩󵄩󵄩𝑔󵄩󵄩󵄩𝐻1 ≤ 𝐶12 󵄩󵄩󵄩∇𝑔󵄩󵄩󵄩 , 𝑑 ≤ 3.
󵄨 󵄨2 𝛽𝐾 󵄨 󵄨2
It follows from (58) and (59) that + 𝜇 ∑ ∫ 󵄨󵄨󵄨𝐷𝛼 𝑢󵄨󵄨󵄨 dx + ∑ ∫ 󵄨󵄨󵄨𝐷𝛼 V󵄨󵄨󵄨 dx
|𝛼|=2 T
𝑑 2 |𝛼|=2 T
𝑑

󵄩󵄩 󵄩 󵄩 󵄩 󵄩󵄩 󵄩 󵄩 󵄩
󵄩󵄩𝜕𝑥𝑖 𝑔󵄩󵄩󵄩 ≤ 𝐶12 󵄩󵄩󵄩∇𝜕𝑥𝑖 𝑔󵄩󵄩󵄩 , 󵄩󵄩𝜕𝑥 𝑥 𝑔󵄩󵄩󵄩 ≤ 𝐶12 󵄩󵄩󵄩∇𝜕𝑥 𝑥 𝑔󵄩󵄩󵄩 ,
󵄩 󵄩 󵄩 󵄩 󵄩󵄩 𝑖 𝑗 󵄩󵄩 󵄩󵄩 𝑖 𝑗 󵄩󵄩
̂2 (‖w‖𝐻2 + ‖w‖2 2 ) 󵄩󵄩󵄩󵄩∇3 w󵄩󵄩󵄩󵄩2 + 𝐶
≤𝐶 ̂3 ‖𝑢‖2 ,
(60) 𝐻 󵄩 󵄩
1/2 1/2 where 𝐶̂2 = (𝜒|1 − 2𝑈𝑐 | + 𝜒 + (4𝜇/𝑈𝑐 )) max{𝐶14 , 𝐶15 , 𝐶16 } and
󵄩󵄩 󵄩󵄩 󵄩 𝛼 󵄩2 󵄩 󵄩2
󵄩󵄩∇𝑔󵄩󵄩 ≤ 𝐶12 ( ∑ 󵄩󵄩󵄩𝐷 𝑔󵄩󵄩󵄩 ) ( ∑ 󵄩󵄩󵄩∇𝐷𝛼 𝑔󵄩󵄩󵄩 )
2
≤ 𝐶12 . the proof is completed.
|𝛼|=2 |𝛼|=2
(61) Lemma 5. Let w(x, 𝑡) be a solution of (18) such that for 0 ≤
𝑡 ≤ 𝑇 < 𝑇0 ,
Furthermore,
2
1/2 1 𝑑 [𝜒𝑈𝑐 (1 − 𝑈𝑐 )]
󵄩󵄩 󵄩󵄩 󵄩 𝛼 󵄩2 ‖w(⋅, 𝑡)‖𝐻2 + ‖w(⋅, 𝑡)‖2𝐻2 ≤ min { 1 , },
󵄩󵄩∇𝑔󵄩󵄩𝐻2 ≤ 𝐶13 ( ∑ 󵄩󵄩󵄩∇𝐷 𝑔󵄩󵄩󵄩 ) , ̂2
𝐶 4 2𝑑1
|𝛼|=2 (62) (69)
4 2 1/2
𝐶13 = (𝐶12 + 𝐶12 + 1) . ̂1 𝑒𝜆 max 𝑡 ‖w (⋅, 0)‖ .
‖w (⋅, 𝑡)‖ ≤ 2𝐶 (70)
8 Abstract and Applied Analysis

Then (2) If [𝜒𝑈𝑐 (1−𝑈𝑐 )]2 /𝑑1 𝑑2 < 1, then it is not hard to verify
̂4 {‖w(⋅, 0)‖2 2 + 𝑒2𝜆 max 𝑡 ‖w(⋅, 0)‖2 } ,
‖w(⋅, 𝑡)‖2𝐻2 ≤ 𝐶 by (70), (73), and (75) that
𝐻
(71) ̂4 {‖w(⋅, 0)‖2 2 + ‖w(⋅, 0)‖2 𝑒2𝜆 max 𝑡 } ,
0 ≤ 𝑡 ≤ 𝑇, ‖w(⋅, 𝑡)‖2𝐻2 ≤ 𝐶 𝐻 (78)

where 𝐶 ̂4 = max{(1 + 𝐶2 )[𝜒𝑈𝑐 (1 − 𝑈𝑐 )]2 /𝑑1 𝑑2 , 4𝐶 ̂2 [1 + where 𝐶 ̂4 = max{(1 + 𝐶2 )(𝑑1 𝑑2 /[𝜒𝑈𝑐 (1 − 𝑈𝑐 )]2 ), 4𝐶 ̂2 [1 +
12 1 12 1
̂3 (1 + 𝐶 )/𝜆 max ]} if [𝜒𝑈𝑐 (1 − 𝑈𝑐 )] /𝑑1 𝑑2 ≥ 1 and 𝐶
𝐶 2 2 ̂4 = ̂3 (1 + 𝐶2 )𝑑1 𝑑2 /𝜆 max [𝜒𝑈𝑐 (1 − 𝑈𝑐 )]2 ]} and thereby complet-
𝐶
12 12
2 2 ̂2 ̂ 2
max{(1 + 𝐶12 )𝑑1 𝑑2 /[𝜒𝑈𝑐 (1 − 𝑈𝑐 )] , 4𝐶1 [1 + 𝐶3 (1 + 𝐶12 )𝑑1 𝑑2 / ing the proof.
𝜆 max [𝜒𝑈𝑐 (1 − 𝑈𝑐 )]2 ]} if [𝜒𝑈𝑐 (1 − 𝑈𝑐 )]2 /𝑑1 𝑑2 < 1.
5. Main Result
Proof. It follows from (61) that
󵄩 󵄩2 Let 𝜃 > 0 be a small fixed constant, and 𝜆 max the dominant
‖∇w(⋅, 𝑡)‖2 ≤ 𝐶12
2
∑ 󵄩󵄩󵄩𝐷𝛼 w(⋅, 𝑡)󵄩󵄩󵄩 . (72) eigenvalue which is the maximal growth rate. For 𝛿 > 0
|𝛼|=2
arbitrary small we define the escape time 𝑇𝛿 by
Thus, 𝛿
󵄩 󵄩2 𝜃 = 𝛿𝑒𝜆 max 𝑇 (79)
‖w(⋅, 𝑡)‖2𝐻2 2
≤ ‖w (⋅, 𝑡)‖ + 2
(𝐶12 + 1) ∑ 󵄩󵄩󵄩𝐷𝛼 w(⋅, 𝑡)󵄩󵄩󵄩 .
|𝛼|=2 or equivalently
(73)
1 𝜃
Now we estimate the second-order derivatives of w(⋅, 𝑡). 𝑇𝛿 = ln . (80)
𝜆 max 𝛿
From Lemma 4 and (69), we can obtain
2 Our main result in this paper reads as follows.
1 𝑑 󵄨 󵄨2 [𝜒𝑈𝑐 (1 − 𝑈𝑐 )] 󵄨󵄨 𝛼 󵄨󵄨2
∑ ∫ {󵄨󵄨󵄨𝐷𝛼 𝑢󵄨󵄨󵄨 + 󵄨󵄨𝐷 V󵄨󵄨 } dx
2 𝑑𝑡 |𝛼|=2 T 𝑑 𝑑1 𝑑2 (74) Theorem 6. Assume that the set of 𝑞2 = ∑𝑑𝑖=1 𝑞𝑖2 satisfying
instability criterion (23) is not empty for given parameters 𝑑1 ,
̂3 ‖𝑢‖2 ≤ 𝐶
≤𝐶 ̂3 ‖w(⋅, 𝑡)‖2 . 𝑑2 , 𝛼, 𝛽, 𝜒, 𝜇, and 𝑈𝑐 . Let
Integrating this from 0 to 𝑡, we deduce from (70) that w0 (x) = ∑ {𝑤q− r− (q) + 𝑤q+ r+ (q)} 𝑒q (x) ∈ 𝐻2 ,
2
(81)
q∈Ω
󵄨 󵄨2 [𝜒𝑈𝑐 (1 − 𝑈𝑐 )] 󵄨󵄨 𝛼 󵄨2
∑ ∫ {󵄨󵄨󵄨𝐷𝛼 𝑢 (⋅, 𝑡)󵄨󵄨󵄨 + 󵄨󵄨𝐷 V (⋅, 𝑡)󵄨󵄨󵄨 } dx
|𝛼|=2 T
𝑑 𝑑 𝑑
1 2 such that ‖w0 ‖ = 1. Then there exist constants 𝛿0 > 0, 𝐶 ̂ > 0,
and 𝜃 > 0, depending on 𝑑1 , 𝑑2 , 𝛼, 𝛽, 𝜒, 𝜇, and 𝑈𝑐 , such that
󵄨 󵄨2 for all 0 < 𝛿 ≤ 𝛿0 , if the initial perturbation of the steady state
≤ ∑ ∫ {󵄨󵄨󵄨𝐷𝛼 𝑢 (⋅, 0)󵄨󵄨󵄨
T𝑑
|𝛼|=2 [𝑈, 𝑉] is w𝛿 (⋅, 0) = 𝛿w0 , then its nonlinear evolution w𝛿 (⋅, 𝑡)
2 satisfies
[𝜒𝑈𝑐 (1 − 𝑈𝑐 )] 󵄨󵄨 𝛼 󵄨2
+ 󵄨󵄨𝐷 V (⋅, 0)󵄨󵄨󵄨 } dx 󵄩󵄩
󵄩󵄩 𝛿
󵄩󵄩
󵄩
𝑑1 𝑑2 󵄩󵄩w (⋅, 𝑡) − 𝛿𝑒𝜆 max 𝑡 ∑ 𝑤+ r (q) 𝑒 (x)󵄩󵄩󵄩
󵄩󵄩 q + q 󵄩󵄩
̂2 𝐶
4𝐶 ̂ 󵄩󵄩 󵄩󵄩
1 3 2𝜆 max 𝑡 󵄩 q∈Ωmax 󵄩
+ 𝑒 ‖w(⋅, 0)‖2 .
𝜆 max (82)
̂ {𝑒−𝜌𝑡 + 𝛿󵄩󵄩󵄩w0 󵄩󵄩󵄩2 2 + 𝛿2 󵄩󵄩󵄩w0 󵄩󵄩󵄩3 2
≤𝐶
(75) 󵄩 󵄩𝐻 󵄩 󵄩𝐻
We will proceed in the following two cases: [𝜒𝑈𝑐 (1 −
+ 𝛿𝑒𝜆 max 𝑡 + 𝛿2 𝑒2𝜆 max 𝑡 } 𝛿𝑒𝜆 max 𝑡
𝑈𝑐 )]2 /𝑑1 𝑑2 ≥ 1, [𝜒𝑈𝑐 (1 − 𝑈𝑐 )]2 /𝑑1 𝑑2 < 1.
(1) If [𝜒𝑈𝑐 (1 − 𝑈𝑐 )]2 /𝑑1 𝑑2 ≥ 1, then it is clear from (75) for 0 ≤ 𝑡 ≤ 𝑇𝛿 , and 𝜌 > 0 is the gap between 𝜆 max and the rest
that of 𝜆 q in (82).
2
󵄩 󵄩2 [𝜒𝑈𝑐 (1 − 𝑈𝑐 )] 󵄩 󵄩2 Proof. Let w𝛿 (x, 𝑡) be the solutions of (18) with initial data
∑ 󵄩󵄩󵄩𝐷𝛼 w(⋅, 𝑡)󵄩󵄩󵄩 ≤ ∑ 󵄩󵄩󵄩𝐷𝛼 w(⋅, 0)󵄩󵄩󵄩
|𝛼|=2
𝑑 𝑑
1 2 |𝛼|=2 w𝛿 (⋅, 0) = 𝛿w0 . Define
(76)
̂2 𝐶
4𝐶 ̂ 󵄩 󵄩 𝐶̂
+ 1 3 𝑒2𝜆 max 𝑡 ‖w(⋅, 0)‖2 . 𝑇∗ = sup {𝑡 | 󵄩󵄩󵄩󵄩w𝛿 (⋅, 𝑡) − 𝛿𝑒L𝑡 w0 󵄩󵄩󵄩󵄩 ≤ 1 𝛿𝑒𝜆 max 𝑡 } .
𝜆 max 2

Using (73) and (76), we know that 󵄩 󵄩 󵄩 󵄩2

𝑇∗∗ = sup {𝑡 | 󵄩󵄩󵄩󵄩w𝛿 (⋅, 𝑡)󵄩󵄩󵄩󵄩𝐻2 + 󵄩󵄩󵄩󵄩w𝛿 (⋅, 𝑡)󵄩󵄩󵄩󵄩𝐻2 (83)
̂4 {‖w(⋅, 0)‖2 2 + ‖w(⋅, 0)‖2 𝑒2𝜆 max 𝑡 } ,
‖w(⋅, 𝑡)‖2𝐻2 ≤ 𝐶 (77)
𝐻
2
where 𝐶 ̂2 [1+𝐶
̂4 = max{(1+𝐶2 )[𝜒𝑈𝑐 (1 − 𝑈𝑐 )]2 /𝑑1 𝑑2 , 4𝐶 ̂3 (1+ 1 𝑑 [𝜒𝑈𝑐 (1 − 𝑈𝑐 )]
12 1 ≤ min { 1 , }} .
2
𝐶12 )/𝜆 max ]}. ̂
𝐶2∗ 4 2𝑑1
Abstract and Applied Analysis 9

We recall (80) and choose 𝜃 such that Next, We prove by contradiction that for 𝛿 sufficiently
2 small, 𝑇𝛿 = min{𝑇𝛿 , 𝑇∗ , 𝑇∗∗ }. If 𝑇∗∗ is the smallest, we can
𝜆 max 𝑑1 [𝜒𝑈𝑐 (1 − 𝑈𝑐 )]
̂∗ 𝐶
𝐶 ̂ ̂1/2 let 𝑡 = 𝑇∗∗ ≤ 𝑇𝛿 in (86) and (87) to get
2 4 𝜃 (1 + 2𝐶4 𝜃) < min { , , }.
4 8 4𝑑1
󵄩󵄩 𝛿 ∗∗ 󵄩󵄩 󵄩 󵄩2
(84) 󵄩󵄩w (𝑇 )󵄩󵄩 2 + 󵄩󵄩󵄩w𝛿 (𝑇∗∗ )󵄩󵄩󵄩 2
󵄩 󵄩𝐻 󵄩 󵄩𝐻
We first estimate 𝐻2 norm of w𝛿 (x, 𝑡) for 0 ≤ 𝑡 ≤ min{𝑇𝛿 ,
𝑇 , 𝑇∗∗ }. By the definition of 𝑇∗ and Lemma 1, for 0 < 𝑡 ≤ 𝑇∗ ,
∗
̂4 𝛿󵄩󵄩󵄩w0 󵄩󵄩󵄩 2 + 𝐶
≤ √𝐶 ̂4 𝛿2 󵄩󵄩󵄩w0 󵄩󵄩󵄩2 2
we have 󵄩 󵄩𝐻 󵄩 󵄩𝐻
󵄩󵄩 𝛿 󵄩 3 ̂ 𝜆 max 𝑡
󵄩󵄩w (⋅, 𝑡)󵄩󵄩󵄩 ≤ 𝐶 (91)
󵄩 󵄩 2 1 𝛿𝑒 . (85) ̂4 𝜃 (1 + √𝐶
+ √𝐶 ̂4 𝜃)
From Lemma 5, direct computation gives
2
󵄩󵄩 𝛿 󵄩 ̂4 {𝛿󵄩󵄩󵄩w0 󵄩󵄩󵄩 2 + 𝛿𝑒𝜆 max 𝑡 } . 1 𝑑 [𝜒𝑈𝑐 (1 − 𝑈𝑐 )]
󵄩󵄩w (⋅, 𝑡)󵄩󵄩󵄩 2 ≤ √𝐶 󵄩 󵄩𝐻 (86) < ∗ min { 1 , },
󵄩 󵄩𝐻 ̂
𝐶2 4 2𝑑1
It follows from (86) and (𝑎 + 𝑏)𝑝 ≤ 2𝑝−1 (𝑎𝑝 + 𝑏𝑝 )(𝑎, 𝑏 ≥ 0, 𝑝 ≥
1) that ̂4 ≥ 1 and 𝛿 is sufficiently small
where 𝜃 satisfies (84) with 𝐶
󵄩󵄩󵄩w𝛿 (⋅, 𝑡)󵄩󵄩󵄩3 ≤ 4(𝐶
̂4 )3/2 {𝛿3 󵄩󵄩󵄩w0 󵄩󵄩󵄩3 2 + 𝛿3 𝑒3𝜆 max 𝑡 } . (87) such that √𝐶̂4 𝛿‖w0 ‖ 2 + 𝐶̂4 𝛿2 ‖w0 ‖2 2 ≤ (1/2𝐶
̂∗ ) min{𝑑1 /4,
󵄩󵄩 󵄩󵄩𝐻2 󵄩 󵄩𝐻 𝐻 𝐻 2
[𝜒𝑈𝑐 (1 − 𝑈𝑐 )]2 /2𝑑1 }. This is a contradiction to the definition
Secondly, we establish a sharper 𝐿2 estimate for w𝛿 (x, 𝑡) of 𝑇∗∗ .
for 0 ≤ 𝑡 ≤ min{𝑇𝛿 , 𝑇∗ , 𝑇∗∗ }. Applying Duhamel’s principle, If 𝑇∗ is the smallest, let 𝑡 = 𝑇∗ ≤ 𝑇𝛿 in (88), we see that
we get
󵄩󵄩 𝛿 󵄩
󵄩󵄩w (⋅, 𝑇∗ ) − 𝛿𝑒L𝑇 w0 󵄩󵄩󵄩
∗

w𝛿 (⋅, 𝑡) = 𝛿𝑒L𝑡 w0 󵄩󵄩 󵄩󵄩
𝑡
− ∫ 𝑒L(𝑡−𝜏) [𝜒 (1 − 2𝑈𝑐 ) ∇𝑢𝛿 (𝜏) ∇V𝛿 (𝜏) 󵄩 󵄩2 ̂4 𝛿2 󵄩󵄩󵄩w0 󵄩󵄩󵄩3 2
0 { 𝛿󵄩󵄩󵄩w0 󵄩󵄩󵄩𝐻2 + 4√𝐶
{
󵄩 󵄩𝐻
̂ ̂ ̂
≤ 𝐶1 𝐶4 𝐶5 {
− 2𝜒𝑢𝛿 (𝜏) ∇𝑢𝛿 (𝜏) ∇V𝛿 (𝜏) { 𝜆 max
{
+ 𝜒 (1 − 2𝑈𝑐 ) 𝑢𝛿 (𝜏) ∇2 V𝛿 (𝜏) (92)
2 𝜃 + 2√𝐶 ̂4 𝜃 } 2}
− 𝜒(𝑢𝛿 (𝜏)) ∇2 V𝛿 (𝜏)
∗
+ } 𝛿𝑒𝜆 max 𝑇
𝜆 max }
𝜇 𝛿 2 }
+ (𝑢 (𝜏)) , 0] 𝑑𝜏.
𝑈𝑐
(88) ̂1 𝜆 𝑇∗
𝐶
< 𝛿𝑒 max ,
2
By Lemma 1, (57), (59), and Lemma 5, for 0 ≤ 𝑡 ≤ min{𝑇𝛿 ,
𝑇∗ , 𝑇∗∗ }, we know ̂4 𝐶
̂5 ((𝛿‖w0 ‖2 2 + 4√𝐶
̂4 𝛿2 ‖w0 ‖3 2 )/
for 𝛿 is small such that 𝐶
󵄩󵄩 𝛿 󵄩 𝐻 𝐻
󵄩󵄩w (⋅, 𝑡) − 𝛿𝑒L𝑡 w0 󵄩󵄩󵄩 ̂5 /𝐶
𝜆 max ) < 1/4, by our choice of 𝜃 in (84) and let 𝐶 ̂∗ ≤ 1.
󵄩 󵄩 2
𝑡 This again contradicts the definition of 𝑇∗ . Hence, 𝑇𝛿 is the
̂5 ∫ 𝑒𝜆 max (𝑡−𝜏) (󵄩󵄩󵄩󵄩w𝛿 (𝜏)󵄩󵄩󵄩󵄩 2 + 󵄩󵄩󵄩󵄩w𝛿 (𝜏)󵄩󵄩󵄩󵄩 2 ) 𝑑𝜏,
2 3
̂1 𝐶
≤𝐶 smallest.
0
󵄩 󵄩 𝐻 󵄩 󵄩 𝐻
(89) Finally, we can obtain from (31) that

̂5 = 𝜒|1 − 2𝑈𝑐 |(𝐶2 /𝐶2 + 𝐶6 ) + 𝜇𝐶7 /𝜒|1 − 2𝑈𝑐 |𝑈𝑐 + 󵄩󵄩 󵄩󵄩

where 𝐶 󵄩󵄩 𝛿 󵄩
12 9 󵄩󵄩w (⋅, 𝑡) − 𝛿𝑒𝜆 max 𝑡 ∑ 𝑤+ r (q) 𝑒 (x)󵄩󵄩󵄩
𝜒𝐶6 (2𝐶12 2
/𝐶92 + 1). By our choice of 𝑡 ≤ min{𝑇𝛿 , 𝑇∗ , 𝑇∗∗ }, it 󵄩󵄩 q + q 󵄩󵄩
󵄩󵄩 󵄩󵄩
is further bounded by 󵄩 q∈Ωmax 󵄩
󵄩󵄩 𝛿 󵄩 󵄩󵄩 󵄩󵄩
󵄩󵄩w (𝑡) − 𝛿𝑒L𝑡 w0 󵄩󵄩󵄩 󵄩󵄩 𝛿 󵄩󵄩 󵄩󵄩󵄩 󵄩󵄩
󵄩 󵄩 ≤ 󵄩󵄩󵄩w (⋅, 𝑡) − 𝛿𝑒 w0 󵄩󵄩󵄩 + 󵄩󵄩󵄩𝛿 ∑ 𝑤q r− (q) 𝑒 𝑒q (x)󵄩󵄩󵄩󵄩
L𝑡 − 𝜆−q 𝑡

󵄩 󵄩2 ̂4 𝛿2 󵄩󵄩󵄩w0 󵄩󵄩󵄩3 2 󵄩󵄩󵄩 q∈Ωmax 󵄩󵄩

󵄩
{ 𝛿󵄩󵄩󵄩w0 󵄩󵄩󵄩𝐻2 + 4√𝐶
{
󵄩 󵄩𝐻
̂1 𝐶
≤𝐶 ̂4 𝐶
̂5
{
{ 𝜆 max 󵄩󵄩 󵄩󵄩
󵄩󵄩 󵄩󵄩
{ 󵄩
+ 󵄩󵄩󵄩𝛿 ∑ {𝑤q r− (q) 𝑒 + 𝑤q r+ (q) 𝑒 } 𝑒q (x)󵄩󵄩󵄩󵄩
− 𝜆−q 𝑡 + 𝜆+q 𝑡
󵄩󵄩 q∈Ω\Ω 󵄩󵄩
𝛿𝑒𝜆 max 𝑡 + 2√𝐶̂4 𝛿2 𝑒2𝜆 max 𝑡 }
} 𝜆 max 𝑡 󵄩 max 󵄩
+ } 𝛿𝑒 .
𝜆 max } 󵄩 󵄩
} = 󵄩󵄩󵄩󵄩w𝛿 (⋅, 𝑡) − 𝛿𝑒L𝑡 w0 󵄩󵄩󵄩󵄩 + 𝐼1 + 𝐼2 .
(90) (93)
10 Abstract and Applied Analysis

By (30), (37), (39), and the definition of 𝜌, we get Respecting the above-mentioned facts, the models (7)
and (13) have no nonconstant positive steady state no matter
𝜋 𝑑 󵄨󵄨 − 󵄨󵄨2 󵄨󵄨
𝐼12 ≤ 𝛿2 𝑒2(𝜆 max −𝜌)𝑡 ( ) ∑ 󵄨󵄨𝑤q 󵄨󵄨 󵄨󵄨r− (q)󵄨󵄨󵄨󵄨2 what the self-diffusion coefficients 𝑑1 , 𝑑2 and cross diffusion
2 q∈Ω 󵄨 󵄨 𝑑3 are; in other words, self-diffusion and cross diffusion
max
(94) (without chemotaxis) cannot drive instability and cannot
𝜋 𝑑 6 󵄨󵄨 󵄨2
∑ 𝑞 󵄨󵄨󵄨w𝑞 󵄨󵄨󵄨󵄨 .
2 2(𝜆 max −𝜌)𝑡 generate patterns for the corresponding model. This means
≤𝛿 𝑒 𝐶22 𝐶42 ( )
2 q∈Ωmax
that chemotaxis-driven instability occurs. We prove that for
any given general perturbation of magnitude 𝛿, linear fastest
From (25) we know that there is one (or two) 𝑞2 satisfying growing modes determine the nonlinear evolution for the
model (2), over a time period of the order ln 1/𝛿. Therefore,
𝜆+ (𝑞2 ) = 𝜆 max . If there is only one 𝑞2 satisfying 𝜆+ (𝑞2 ) =
2 our results indeed provide a rigorous mathematical descrip-
𝜆 max , we denote it by 𝑞max ; if there are 𝑞12 and 𝑞22 satisfying tion for the nonlinear pattern formation in a volume-filling
+ 2 2
𝜆 (𝑞𝑖 ) = 𝜆 max (𝑖 = 1, 2), we can let 𝑞max = max{𝑞12 , 𝑞22 }. It chemotaxis model with logistic growth.
follows from (30) and (94) that
̂6 𝛿𝑒(𝜆 max −𝜌)𝑡 ,
𝐼1 ≤ 𝐶 (95) Conflict of Interests
̂6 = 𝐶2 𝐶3 𝑞3 . Furthermore, by (29), (30), and the The authors declare that they have no conflict of interests
where 𝐶 max regarding the publication of this paper.
definition of 𝜌, we know that

𝐼2 ≤ 𝛿𝑒(𝜆 max −𝜌)𝑡 . (96) Acknowledgments

Substituting (90), (95), and (96) into (93), one can obtain This work is supported by the China National Natural Science
Foundation (nos. 11361055 and 11261053) and the Fundamen-
󵄩󵄩 󵄩󵄩
󵄩󵄩 𝛿 󵄩 tal Research Funds for the Gansu University.
󵄩󵄩w (⋅, 𝑡) − 𝛿𝑒𝜆 max 𝑡 ∑ 𝑤+ r (q) 𝑒 (x)󵄩󵄩󵄩
󵄩󵄩 q + q 󵄩󵄩
󵄩󵄩 󵄩󵄩
󵄩 q∈Ωmax 󵄩 References
̂ −𝜌𝑡 󵄩󵄩 󵄩󵄩 2 2󵄩
󵄩 󵄩󵄩 3 (97)
≤ 𝐶 {𝑒 + 𝛿󵄩󵄩w0 󵄩󵄩𝐻2 + 𝛿 󵄩󵄩w0 󵄩󵄩𝐻2
[1] K. J. Painter and T. Hillen, “Volume-filling and quorum-sens-
+ 𝛿𝑒𝜆 max 𝑡 + 𝛿2 𝑒2𝜆 max 𝑡 } 𝛿𝑒𝜆 max 𝑡 , ing in models for chemosensitive movement,” The Canadian
Applied Mathematics Quarterly, vol. 10, no. 4, pp. 501–543, 2002.
where 𝐶̂ = max{𝐶 ̂6 + 1, 4𝐶 ̂3/2 𝐶
̂1 𝐶 ̂5 /𝜆 max } with 𝐶
̂4 ≥ 1 and [2] T. Hillen and K. Painter, “Global existence for a parabolic
4
chemotaxis model with prevention of overcrowding,” Advances
thereby concluding the proof.
in Applied Mathematics, vol. 26, no. 4, pp. 280–301, 2001.
Theorem 6 implies that the dynamics of a general pertur- [3] D. Wrzosek, “Global attractor for a chemotaxis model with pre-
bation are characterized by such linear dynamics over a long vention of overcrowding,” Nonlinear Analysis: Theory, Methods
& Applications, vol. 59, no. 8, pp. 1293–1310, 2004.
time period of 𝜀𝑇𝛿 ≤ 𝑡 ≤ 𝑇𝛿 , for any 𝜀 > 0. In particular,
choose a fixed q0 = (𝑞01 , 𝑞02 , . . . , 𝑞0𝑑 ) ∈ Ωmax and let w0 (x) = [4] D. Wrzosek, “Long-time behaviour of solutions to a chemotaxis
model with volume-filling effect,” Proceedings of the Royal
𝜅(r+ (q0 )/|r+ (q0 )|)𝑒q0 (x), where 𝜅 = 1/‖𝑒q0 ‖ = √(2/𝜋)𝑑 so Society of Edinburgh A: Mathematics, vol. 136, no. 2, pp. 431–444,
that ‖w0 (x)‖ = 1. Then 2006.
[5] A. B. Potapov and T. Hillen, “Metastability in chemotaxis
󵄩󵄩 󵄩 󵄨 󵄨2 󵄨 󵄨4 1/2
󵄩󵄩w0 (x)󵄩󵄩󵄩𝐻2 = (1 + 󵄨󵄨󵄨q0 󵄨󵄨󵄨 + 󵄨󵄨󵄨q0 󵄨󵄨󵄨 ) . (98) models,” Journal of Dynamics and Differential Equations, vol. 17,
no. 2, pp. 293–330, 2005.
[6] V. Calvez and J. A. Carrillo, “Volume effects in the Keller-
Therefore, if 𝑡 ≤ 𝑇𝛿 , we can obtain from (97) and (98) Segel model: energy estimates preventing blow-up,” Journal de
󵄩󵄩 r (q ) 󵄩󵄩 Mathématiques Pures et Appliquées, vol. 86, no. 2, pp. 155–175,
󵄩󵄩 𝛿 󵄩
󵄩󵄩w (⋅, 𝑇𝛿 ) − 𝛿𝑒𝜆 max 𝑇 𝜅 󵄨 + 0 󵄨 𝑒q0 (x)󵄩󵄩󵄩
𝛿
2006.
󵄩󵄩 󵄨󵄨r+ (q0 )󵄨󵄨 󵄩󵄩
󵄩 󵄨 󵄨 󵄩 (99) [7] M. Burger, M. di Francesco, and Y. Dolak-Struss, “The Keller-
Segel model for chemotaxis with prevention of overcrowding:
̂8 {𝛿
≤𝐶 𝜌/𝜆 max 2
+ 𝜃 + 𝜃 }, 3
linear vs. nonlinear diffusion,” SIAM Journal on Mathematical
Analysis, vol. 38, no. 4, pp. 1288–1315, 2006.
where 𝐶̂8 = 𝐶̂ max{𝜃1−(𝜌/𝜆 max ) , (1 + |q0 |2 + |q0 |4 )3/2 +1}. If 0 < [8] Z. Wang and T. Hillen, “Classical solutions and pattern forma-
̂8 )3/2 and 𝛿0 = (𝜃/2𝐶
𝜃 < (1/𝐶 ̂8 − 𝜃2 − 𝜃3 )𝜆 max /𝜌 , then tion for a volume filling chemotaxis model,” Chaos, vol. 17, no.
3, Article ID 037108, 13 pages, 2007.
[9] J. Jiang and Y. Zhang, “On convergence to equilibria for a chem-
󵄩󵄩 𝛿 󵄩 ̂8 {𝛿𝜌/𝜆 max + 𝜃2 + 𝜃3 } ≥ 𝜃 > 0,
󵄩󵄩w (⋅, 𝑇𝛿 )󵄩󵄩󵄩 ≥ 𝜃 − 𝐶 otaxis model with volume-filling effect,” Asymptotic Analysis,
󵄩 󵄩 2 (100) vol. 65, no. 1-2, pp. 79–102, 2009.
0 < 𝛿 ≤ 𝛿0 . [10] D. Wrzosek, “Volume filling effect in modelling chemotaxis,”
Mathematical Modelling of Natural Phenomena, vol. 5, no. 1, pp.
This means nonlinear instability as 𝛿 → 0. 123–147, 2010.
Abstract and Applied Analysis 11

[11] M. Winkler, “Does a “volume-filling effect” always prevent [29] J. K. Hale, Ordinary Differential Equations, Krieger, Malabar,
chemotactic collapse?” Mathematical Methods in the Applied Fla, USA, 1980.
Sciences, vol. 33, no. 1, pp. 12–24, 2010. [30] D. Henry, Geometric Theory of Semilinear Parabolic Equations,
[12] M. Winkler and K. C. Djie, “Boundedness and finite-time vol. 840 of Lecture Notes in Mathematics, Springer, New York,
collapse in a chemotaxis system with volume-filling effect,” Non- NY, USA, 1993.
linear Analysis: Theory, Methods & Applications, vol. 72, no. 2,
[31] O. A. Ladyzenskaja, V. A. Solonnikov, and N.N. Uralceva, Linear
pp. 1044–1064, 2010.
and Quasilinear Equations of Parabolic Type, Translations of
[13] Z. Wang, M. Winkler, and D. Wrzosek, “Singularity formation Mathematical Monographs, American Mathematical Society,
in chemotaxis systems with volume-filling effect,” Nonlinearity, Providence, RI, USA, 1968.
vol. 24, no. 12, Article ID 3279, 2011.
[32] M. X. Wang, Nonlinear Partial Differential Equations of Parabol-
[14] Y. Y. Zhang and S. M. Zheng, “Asymptotic behavior of solu-
ic Type, Science Press, Beijing, China, 1993.
tions to a quasilinear nonuniform parabolic system modelling
chemotaxis,” Journal of Differential Equations, vol. 248, no. 7, pp.
1684–1710, 2010.
[15] Y. Zhang and S. Zheng, “A 1-d quasilinear nonuniform parabolic
chemotaxis model with volume-filling effect,” Asymptotic Anal-
ysis, vol. 83, no. 1-2, pp. 101–125, 2013.
[16] F. Li and Y. Y. Zhang, “On steady-state solutions of a 1-D chemo-
taxis model with volume-filling effect,” Journal of Mathematical
Analysis and Applications, vol. 408, no. 1, pp. 1–18, 2013.
[17] J. I. Tello and M. Winkler, “A chemotaxis system with logistic
source,” Communications in Partial Differential Equations, vol.
32, no. 4–6, pp. 849–877, 2007.
[18] M. Winkler, “Boundedness in the higher-dimensional parabol-
ic-parabolic chemotaxis system with logistic source,” Commu-
nications in Partial Differential Equations, vol. 35, no. 8, pp.
1516–1537, 2010.
[19] K. J. Painter and T. Hillen, “Spatio-temporal chaos in a chemo-
taxis model,” Physica D: Nonlinear Phenomena, vol. 240, no. 4-5,
pp. 363–375, 2011.
[20] S. Banerjee, A. P. Misra, and L. Rondoni, “Spatiotemporal evo-
lution in a (2 + 1)-dimensional chemotaxis model,” Physica A:
Statistical Mechanics and Its Applications, vol. 391, no. 1-2, pp.
107–112, 2012.
[21] K. Kuto, K. Osaki, and T. a. . Sakurai, “Spatial pattern formation
in a chemotaxis-diffusion-growth model,” Physica D: Nonlinear
Phenomena, vol. 241, no. 19, pp. 1629–1639, 2012.
[22] S. Fu and J. Liu, “Spatial pattern formation in the Keller-Segel
model with a logistic source,” Computers & Mathematics with
Applications, vol. 66, no. 3, pp. 403–417, 2013.
[23] C. Ou and W. Yuan, “Traveling wavefronts in a volume-filling
chemotaxis model,” SIAM Journal on Applied Dynamical Sys-
tems, vol. 8, no. 1, pp. 390–416, 2009.
[24] M. Ma, C. Ou, and Z. Wang, “Stationary solutions of a
volume-filling chemotaxis model with logistic growth and their
stability,” SIAM Journal on Applied Mathematics, vol. 72, no. 3,
pp. 740–766, 2012.
[25] M. Ma, J. Hu, J. Tao, and C. Tong, “Non-existence of stationary
pattern of a chemotaxis model with logistic growth,” Nonlinear
Analysis: Theory, Methods & Applications, vol. 105, no. 1, pp. 3–9,
2014.
[26] Y. Guo and H. J. Hwang, “Pattern formation (I): the Keller-Segel
model,” Journal of Differential Equations, vol. 249, no. 7, pp. 1519–
1530, 2010.
[27] S. M. Fu and J. Liu, “A mathematical characterization for
patterns of a Keller-Segel model with a cubic source term,”
Advances in Mathematical Physics, vol. 2013, Article ID 934745,
11 pages, 2013.
[28] Y. Guo and W. A. Strauss, “Instability of periodic BGK equilib-
ria,” Communications on Pure and Applied Mathematics, vol. 48,
no. 8, pp. 861–894, 1995.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 623726, 4 pages
http://dx.doi.org/10.1155/2014/623726

Research Article
A Note on Gronwall’s Inequality on Time Scales

Xueru Lin
College of Mathematics and Computer Science, Fuzhou University, Fuzhou 350002, China

Correspondence should be addressed to Xueru Lin; xuerulin2014@163.com

Received 29 May 2014; Accepted 6 June 2014; Published 1 July 2014

Academic Editor: Yonghui Xia

Copyright © 2014 Xueru Lin. This is an open access article distributed under the Creative Commons Attribution License, which
permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

This paper gives a new version of Gronwall’s inequality on time scales. The method used in the proof is much different from that in
the literature. Finally, an application is presented to show the feasibility of the obtained Gronwall’s inequality.

1. Introduction and Motivation Taking 𝑓(𝑡) ≡ 𝛼, a classical version of Gronwall’s inequality

follows (see [2, Corollary 6.7, pp 257]).
Recently, an interesting field of research is to study the
dynamic equations on time scales, which have been exten- Theorem B. Let 𝑝 ∈ R+ (T, R), 𝑝(𝑡) ≥ 0, 𝑦 ∈ C𝑟𝑑 R(T, R),
sively studied. For example, one can see [1–17] and references and 𝛼 ∈ R. Then
cited therein. A time scale T is an arbitrary nonempty closed
𝑡
subset of the real numbers R. The forward and backward jump 𝑦 (𝑡) ≤ 𝛼 + ∫ 𝑦 (𝑠) 𝑝 (𝑠) Δ𝑠, ∀𝑡 ∈ T, (3)
operators are defined by 𝜎(𝑡) := inf{𝑠 ∈ T : 𝑠 > 𝑡}, 𝜌(𝑡) := 𝑡0
sup{𝑠 ∈ T : 𝑠 > 𝑡}. A point 𝑡 ∈ T, 𝑡 > inf T, is said to be left
dense if 𝜌(𝑡) = 𝑡 and right dense if 𝑡 < inf T and 𝜎(𝑡) = 𝑡. The implies
mapping 𝜇 : T → R+ defined by 𝜇(𝑡) = 𝜎(𝑡) − 𝑡 is called
graininess. A function 𝑔 : T → R is said to be rd-continuous 𝑦 (𝑡) ≤ 𝛼𝑒𝑝 (𝑡, 𝑡0 ) , ∀𝑡 ∈ T. (4)
provided 𝑔 is continuous at right-dense points. The set of
all such rd-continuous functions is denoted by Crd (T, R). A This paper presents a new version of Gronwall’s inequality as
function 𝑝 : T → R is regressive provided 1+𝜇(𝑡)𝑝(𝑡) ≠0 for follows.
𝑡 ∈ T. Denote R+ (T, R) := {𝑝 ∈ Crd (T, R) : 1+𝜇(𝑡)𝑝(𝑡) > 0}.
Theorem 1. Let −𝑝 ∈ R+ (T, R) and 𝑦 ∈ C𝑟𝑑 R(T, R).
One of important topics is the differential inequalities on
Suppose that 𝑝(𝑡) ≥ 0, 𝑦(𝑡) ≥ 0, and 𝛼 > 0. Then
time scales. A nonlinear version of Gronwall’s inequality is
presented in [2, Theorem 6.4, pp 256]. This version is stated 󵄨󵄨 𝑡 󵄨󵄨
󵄨 󵄨
as follows. 𝑦 (𝑡) ≤ 𝛼 + 󵄨󵄨󵄨󵄨∫ 𝑦 (𝑠) 𝑝 (𝑠) Δ𝑠󵄨󵄨󵄨󵄨 , ∀𝑡 ∈ T, (5)
󵄨󵄨 𝑡0 󵄨󵄨
Theorem A. Let 𝑦, 𝑓 ∈ C𝑟𝑑 R(T, R), 𝑝(𝑡) ∈ R+ (T, R), and
implies
𝑝(𝑡) ≥ 0. Then

𝑡 𝛼𝑒 (𝑡, 𝑡0 ) , for 𝑡 ∈ [ 𝑡0 , +∞)T ,

𝑦 (𝑡) ≤ { 𝑝 (6)
𝑦 (𝑡) ≤ 𝑓 (𝑡) + ∫ 𝑦 (𝑠) 𝑝 (𝑠) Δ𝑠, ∀𝑡 ∈ T, (1) 𝛼𝑒−𝑝 (𝑡, 𝑡0 ) , for 𝑡 ∈ ( −∞, 𝑡0 ]T .
𝑡0

Remark 2. Note that, for 𝑡 ∈ (−∞, 𝑡0 ]T , inequality (5) reduces

implies to
𝑡 𝑡
𝑦 (𝑡) ≤ 𝑓 (𝑡) + ∫ 𝑒𝑝 (𝑡, 𝜎 (𝑠)) 𝑓 (𝑠) 𝑝 (𝑠) Δ𝑠, ∀𝑡 ∈ T. (2) 𝑦 (𝑡) ≤ 𝛼 − ∫ 𝑦 (𝑠) 𝑝 (𝑠) Δ𝑠, (7)
𝑡0 𝑡0
2 Abstract and Applied Analysis

which is different from inequality (3) in Theorem B. Since Lemma 4. Suppose that 𝑔 : T → R+ is positive delta differen-
Theorem B requires 𝑝(𝑡) ≥ 0, we see that Theorem B tiable on T and 𝑔Δ (𝑡)/𝑔(𝑡) is regressive. Then 𝜉𝜇(𝑡) (𝑔Δ (𝑡)/𝑔(𝑡))
cannot be applied to (7). Moreover, the method used to is a preantiderivative of function Log[𝑔(𝑡)], where 𝜉ℎ (𝑧) =
prove Theorem A cannot be used to prove Theorem 1. To (1/ℎ)Log(1 + 𝑧ℎ) and Log is the principal logarithm function.
explain this, recall the proof of Theorem A in [2]. Let 𝑧(𝑡) =
𝑡
∫𝑡 𝑦(𝑠)𝑝(𝑠)Δ𝑠. Then 𝑧(𝑡0 ) = 0 and Proof. Let 𝑓(𝑥) = Log 𝑥. Obviously, 𝑓 : R+ → R is
0
continuous on R+ . To prove Lemma 4, it suffices to show that
𝑧Δ = 𝑦 (𝑡) 𝑝 (𝑡) [Log[𝑔(𝑡)]]Δ = 𝜉𝜇(𝑡) (𝑔Δ (𝑡)/𝑔(𝑡)). In fact, by using Lemma 3,
(8) we have
≤ [𝑓 (𝑡) + 𝑧 (𝑡)] 𝑝 (𝑡) = 𝑝 (𝑡) 𝑧 (𝑡) + 𝑝 (𝑡) 𝑓 (𝑡) . Δ
[Log[𝑔(𝑡)]]
By comparing theorem and variation of constants formula,
1
Δ
𝑡 = (𝑓 ∘ 𝑔) (𝑡) = {∫ 𝑓󸀠 (𝑔 (𝑡) + ℎ𝜇 (𝑡) 𝑔Δ (𝑡)) 𝑑ℎ} 𝑔Δ (𝑡)
0
𝑧 (𝑡) ≤ ∫ 𝑒𝑝 (𝑡, 𝜎 (𝑠)) 𝑓 (𝑠) 𝑝 (𝑠) Δ𝑠, (9)
𝑡0 1
1
= {∫ 𝑑ℎ} 𝑔Δ (𝑡)
and hence Theorem A follows in view of 𝑦(𝑡) ≤ 𝑓(𝑡) + 𝑧(𝑡). 0 𝑔 (𝑡) + ℎ𝜇 (𝑡) 𝑔Δ (𝑡)
Now we try to adopt the same idea used in [2] to estimate 󵄨󵄨ℎ=1
𝑡 1 Δ 󵄨󵄨 Δ
inequality (7). Let 𝑧(𝑡) = ∫𝑡 𝑦(𝑠)𝑝(𝑠)Δ𝑠. Then 𝑧(𝑡0 ) = 0 and ={ Log [𝑔 (𝑡) + ℎ𝜇 (𝑡) 𝑔 (𝑡)] 󵄨󵄨 } 𝑔 (𝑡)
0 𝜇 (𝑡) 𝑔Δ (𝑡) 󵄨󵄨ℎ=0

𝑧Δ = 𝑦 (𝑡) 𝑝 (𝑡) ≤ [𝑓 (𝑡) − 𝑧 (𝑡)] 𝑝 (𝑡) 1

{
{ {Log [𝑔 (𝑡) + 𝜇 (𝑡) 𝑔Δ (𝑡)] − Log [𝑔 (𝑡)]}
{
{ 𝜇 (𝑡)
= −𝑝 (𝑡) 𝑧 (𝑡) + 𝑝 (𝑡) 𝑓 (𝑡) (10) {
{
{
{
{
{ if 𝜇 (𝑡) ≠0,
{
= −𝑝 (𝑡) 𝑧𝜎 + (1 + 𝜇 (𝑡) 𝑝 (𝑡)) 𝑝 (𝑡) 𝑓 (𝑡) . ={
{
{ 𝑔Δ (𝑡)
{
{
By comparing theorem and variation of constants formula, {
{
{ 𝑔 (𝑡)
{
we have {
{
𝑡 { if 𝜇 (𝑡) = 0
𝑧 (𝑡) ≤ ∫ 𝑒⊖𝑝 (𝑡, 𝜎 (𝑠)) 𝑓 (𝑠) 𝑝 (𝑠) Δ𝑠, (11)
𝑡0 { 1 𝑔 (𝑡) + 𝜇 (𝑡) 𝑔Δ (𝑡)
{
{ {Log } if 𝜇 (𝑡) ≠0,
{ 𝜇 (𝑡) 𝑔 (𝑡)
which implies ={ Δ
{
{ 𝑔 (𝑡)
{
if 𝜇 (𝑡) = 0
𝑡
{ 𝑔 (𝑡)
−𝑧 (𝑡) ≥ − ∫ 𝑒⊖𝑝 (𝑡, 𝜎 (𝑠)) 𝑓 (𝑠) 𝑝 (𝑠) Δ𝑠. (12)
𝑡0
1 𝑔Δ (𝑡)
= Log {1 + 𝜇 (𝑡) }
If we were to use the same idea as in [2], we should combine 𝜇 (𝑡) 𝑔 (𝑡)
(12) with
𝑔Δ (𝑡)
= 𝜉𝜇(𝑡) ( ).
𝑦 (𝑡) ≤ 𝑓 (𝑡) − 𝑧 (𝑡) . (13) 𝑔 (𝑡)
(15)
However, on one side, 𝑦(𝑡) ≤ ⋅ ⋅ ⋅ ; on the other side, 𝑓(𝑡) −
𝑧(𝑡) ≥ ⋅ ⋅ ⋅ . These two inequalities cannot lead us anywhere.
Therefore, some novel proof is employed to prove
Theorem 1. One can see the detailed proof in the next section.
Proof of Theorem 1. To prove Theorem 1, we divide it into two
cases.
2. Proof of Main Result
Case 1. For 𝑡 ∈ [𝑡0 , +∞)T , in this case, we have
Before our proof of Theorem 1, we need some lemmas.
󵄨󵄨 𝑡 󵄨󵄨 𝑡
󵄨 󵄨
Lemma 3 (chain rule [2]). Assume 𝑔 : T → X is delta 𝑦 (𝑡) ≤ 𝛼 + 󵄨󵄨󵄨󵄨∫ 𝑦 (𝑠) 𝑝 (𝑠) Δ𝑠󵄨󵄨󵄨󵄨 = 𝛼 + ∫ 𝑦 (𝑠) 𝑝 (𝑠) Δ𝑠,
differentiable on T. Assume further that 𝑓 : X → X is 󵄨󵄨 𝑡0 󵄨󵄨 𝑡0
continuously differentiable. Then 𝑓 ∘ 𝑔 : T → X is delta
differentiable and satisfies for 𝑡 ∈ [𝑡0 , +∞)T .
(16)
1
Δ 󸀠 Δ Δ
(𝑓 ∘ 𝑔) (𝑡) = {∫ 𝑓 (𝑔 (𝑡) + ℎ𝜇 (𝑡) 𝑔 (𝑡)) 𝑑ℎ} 𝑔 (𝑡) .
0 Hence, it is easy to conclude that 𝑦(𝑡) ≤ 𝛼𝑒𝑝 (𝑡, 𝑡0 ) for 𝑡 ∈
(14) [𝑡0 , +∞)T .
Abstract and Applied Analysis 3

𝑡
Case 2. For 𝑡 ∈ (−∞, 𝑡0 ]T , let 𝑧(𝑡) = ∫𝑡 𝑦(𝑠)𝑝(𝑠)Δ𝑠. For any 3. An Application
0
𝑠 ∈ [𝑡, 𝑡0 ]T , we have
Inequality (5) has many potential applications. For instance,
󵄨󵄨 𝑠 󵄨󵄨 it can be used to study the property of the solutions to the
󵄨 󵄨
𝑦 (𝑠) ≤ 𝛼 + 󵄨󵄨󵄨󵄨∫ 𝑦 (𝜏) 𝑝 (𝜏) Δ𝜏󵄨󵄨󵄨󵄨 dynamic systems. Consider the following linear system:
󵄨󵄨 𝑡0 󵄨󵄨
(17)
𝑠 𝑥Δ = 𝐴 (𝑡) 𝑥. (26)
= 𝛼 − ∫ 𝑦 (𝜏) 𝑝 (𝜏) Δ𝜏 = 𝛼 − 𝑧 (𝑠) .
𝑡0
Let 𝑋(𝑡, 𝑡0 , 𝑥0 ) and 𝑋(𝑡, 𝑡0 , 𝑥̃0 ) be two solutions of (26)
Noting that 𝑦 ≥ 0, 𝑝 ≥ 0, 𝛼 > 0, we have 𝛼 − 𝑧(𝑠) > 0. Thus, satisfying the initial conditions 𝑋(𝑡0 ) = 𝑥0 and 𝑋(𝑡0 ) = 𝑥̃0 ,
we have respectively.

𝑦 (𝑠) Theorem 5. Suppose that 𝐴(𝑡) is bounded on T. Then one has

≤ 1. (18)
𝛼 − 𝑧 (𝑠) 󵄩󵄩 󵄩
󵄩󵄩𝑋 (𝑡, 𝑡0 , 𝑥0 ) − 𝑋 (𝑡, 𝑡0 , 𝑥̃0 )󵄩󵄩󵄩
Multiplied by −𝑝(𝑠) on both sides of the above inequality, it 󵄩󵄩 󵄩
󵄩𝑥 − 𝑥̃0 󵄩󵄩󵄩 𝑒𝑝1 (𝑡, 𝑡0 ) , for 𝑡 ∈ [𝑡0 , +∞)T , (27)
follows that ≤ {󵄩󵄩󵄩 0 󵄩
󵄩󵄩𝑥0 − 𝑥̃0 󵄩󵄩󵄩 𝑒−𝑝1 (𝑡, 𝑡0 ) , for 𝑡 ∈ (−∞, 𝑡0 ]T ,
−𝑦 (𝑠) 𝑝 (𝑠)
≥ −𝑝 (𝑠) , (19)
𝛼 − 𝑧 (𝑠) where 𝑝1 (𝑡) ≡ 𝑀.
or Proof. Integrating (7) over [𝑡0 , 𝑡], we have
[𝛼 − 𝑧(𝑠)]Δ
≥ −𝑝 (𝑠) . (20) 𝑋 (𝑡, 𝑡0 , 𝑥0 )
𝛼 − 𝑧 (𝑠)
𝑡
Since −𝑝 ∈ R+ , −𝑝 ≤ [𝛼 − 𝑧(𝑠)]Δ /(𝛼 − 𝑧(𝑠)) ∈ R+ . Using = 𝑥0 + ∫ [𝐴 (𝑠) 𝑋 (𝑠, 𝑡0 , 𝑥0 ) + 𝑓 (𝑠, 𝑋 (𝑠, 𝑡0 , 𝑥0 ))] Δ𝑠.
the fact that 𝜉𝜇(𝑡) (𝑧) is nondecreasing with respect to 𝑧 for 𝑧 ∈ 𝑡0
(28)
R+ , we have

(𝛼 − 𝑧 (𝑠))Δ Denoting 𝑀 = sup𝑡∈T ‖𝐴(𝑡)‖, simple computation leads us to

𝜉𝜇(𝑠) [ ] ≥ 𝜉𝜇(𝑠) [−𝑝 (𝑠)] . (21)
𝛼 − 𝑧 (𝑠) 󵄩󵄩 󵄩
󵄩󵄩𝑋 (𝑡, 𝑡0 , 𝑥0 ) − 𝑋 (𝑡, 𝑡0 , 𝑥̃0 )󵄩󵄩󵄩
An integration of the above inequality over [𝑡, 𝑡0 ]T leads to 󵄨󵄨 𝑡 󵄨
󵄩 󵄩 󵄨 󵄩 󵄩 󵄨󵄨
≤ 󵄩󵄩󵄩𝑥0 − 𝑥̃0 󵄩󵄩󵄩 + 𝑀 󵄨󵄨󵄨󵄨∫ 󵄩󵄩󵄩𝑋 (𝑠, 𝑡0 , 𝑥0 ) − 𝑋 (𝑠, 𝑡0 , 𝑥̃0 )󵄩󵄩󵄩 Δ𝑠󵄨󵄨󵄨󵄨 .
󵄨󵄨 𝑡0 󵄨󵄨
𝑡0
(𝛼 − 𝑧 (𝑠))Δ 𝑡0
∫ 𝜉𝜇(𝑠) [ ] Δ𝑠 ≥ ∫ 𝜉𝜇(𝑠) [−𝑝 (𝑠)] Δ𝑠. (22) (29)
𝑡 𝛼 − 𝑧 (𝑠) 𝑡

By Theorem 1, it follows from (29) that

It follows from Lemma 4 that
󵄩󵄩 󵄩
󵄨𝑡 0 𝑡 󵄩󵄩𝑋 (𝑡, 𝑡0 , 𝑥0 ) − 𝑋 (𝑡, 𝑡0 , 𝑥̃0 )󵄩󵄩󵄩
Log[𝛼 − 𝑧(𝑠)]󵄨󵄨󵄨𝑡0 ≥ ∫ 𝜉𝜇(𝑠) [−𝑝 (𝑠)] Δ𝑠, (23)
𝑡
󵄩󵄩 󵄩 (30)
󵄩𝑥 − 𝑥̃0 󵄩󵄩󵄩 𝑒𝑝1 (𝑡, 𝑡0 ) , for 𝑡 ∈ [𝑡0 , +∞)T ,
or ≤ {󵄩󵄩󵄩 0 󵄩
󵄩󵄩𝑥0 − 𝑥̃0 󵄩󵄩󵄩 𝑒−𝑝1 (𝑡, 𝑡0 ) , for 𝑡 ∈ (−∞, 𝑡0 ]T .
𝑡0
Log 𝛼 − Log [𝛼 − 𝑧 (𝑡)] ≥ ∫ 𝜉𝜇(𝑡) [−𝑝 (𝑠)] Δ𝑠, (24)
𝑡

which leads to Remark 6. One can see that, for the case 𝑡 ∈ (−∞, 𝑡0 ]T , (29)
reduces to
𝑡0
𝛼 − 𝑧 (𝑡) ≤ 𝛼 exp (− ∫ 𝜉𝜇(𝑠) [−𝑝 (𝑠)] Δ𝑠) 󵄩󵄩 󵄩
𝑡
󵄩󵄩𝑋 (𝑡, 𝑡0 , 𝑥0 ) − 𝑋 (𝑡, 𝑡0 , 𝑥̃0 )󵄩󵄩󵄩
(25) (31)
𝑡 𝑡
󵄩 󵄩 󵄩 󵄩
= 𝛼 exp (∫ 𝜉𝜇(𝑠) [−𝑝 (𝑠)] Δ𝑠) = 𝛼𝑒−𝑝 (𝑡, 𝑡0 ) . ≤ 󵄩󵄩󵄩𝑥0 − 𝑥̃0 󵄩󵄩󵄩 − 𝑀 ∫ [󵄩󵄩󵄩𝑋 (𝑠, 𝑡0 , 𝑥0 ) − 𝑋 (𝑠, 𝑡0 , 𝑥̃0 )󵄩󵄩󵄩] Δ𝑠.
𝑡0 𝑡 0

Therefore, 𝑦(𝑡) ≤ 𝛼 − 𝑧(𝑡) ≤ 𝑒−𝑝 (𝑡, 𝑡0 ) for 𝑡 ∈ (−∞, 𝑡0 ]T . This As you see, Theorem B cannot be used to (31) because the
completes the proof of Theorem 1. essential condition in Theorem B is 𝑝(𝑡) ≥ 0.
4 Abstract and Applied Analysis

Conflict of Interests Abstract and Applied Analysis, vol. 2014, Article ID 632109, 11
pages, 2014.
The author declares that there is no conflict of interests [16] Y. Gao, Y. Xia, X. Yuan, and P. Wong, “Linearization of
regarding the publication of this paper. nonautonomous impulsive system with nonuniform exponen-
tial dichotomy,” Abstract and Applied Analysis, vol. 2014, Article
Acknowledgment ID 860378, 7 pages, 2014.
[17] Y. Xia, X. Yuan, K. I. Kou, and P. J. Y. Wong, “Existence and
This work was supported by JB12254. uniqueness of solution for perturbed nonautonomous systems
with nonuniform exponential dichotomy,” Abstract and Applied
Analysis, vol. 2014, Article ID 725098, 10 pages, 2014.
References
[1] M. Bohner, G. Guseinov, and A. Peterson, Introduction to the
Time Scales Calculus, Advances in Dynamic Equations on Time
Scales, Birkhäauser, Boston, Mass, USA, 2003.
[2] M. Bohner and A. Peterson, Dynamic Equations on Time Scales,
An Introduction with Applications, Birkhäuser, Boston, Mass,
USA, 2001.
[3] R. P. Agarwal, M. Bohner, and D. O’Regan, “Dynamic equations
on time scales: a survey,” Journal of Computational and Applied
Mathematics, vol. 141, no. 1-2, pp. 1–26, 2002.
[4] R. P. Agarwal, M. Bohner, and D. O’Regan, “Time scale
boundary value problems on infinite intervals,” Journal of
Computational and Applied Mathematics, vol. 141, no. 1-2, pp.
27–34, 2002.
[5] L. Erbe, A. Peterson, and P. Řeháka, “Comparison theorems for
linear dynamic equations on time scales,” Journal of Mathemat-
ical Analysis and Applications, vol. 275, no. 1, pp. 418–438, 2002.
[6] L. Erbe and A. Peterson, “Green functions and comparison
theorems for di ff erential equations on measure chains,”
Dynamics of Continuous, Discrete and Impulsive Systems, vol. 6,
no. 1, pp. 121–137, 1999.
[7] W. N. Li, “Some integral inequalities useful in the theory of
certain partial dynamic equations on time scales,” Computers
and Mathematics with Applications, vol. 61, no. 7, pp. 1754–1759,
2011.
[8] W. N. Li, “Some delay integral inequalities on time scales,”
Computers & Mathematics with Applications, vol. 59, no. 6, pp.
1929–1936, 2010.
[9] Y. H. Xia, J. Cao, and M. Han, “A new analytical method for the
linearization of dynamic equation on measure chains,” Journal
of Differential Equations, vol. 235, no. 2, pp. 527–543, 2007.
[10] Y. Xia, X. Chen, and V. G. Romanovski, “On the linearization
theorem of Fenner and Pinto,” Journal of Mathematical Analysis
and Applications, vol. 400, no. 2, pp. 439–451, 2013.
[11] Y.-H. Xia, J. Li, and P. J. Y. Wong, “On the topological
classification of dynamic equations on time scales,” Nonlinear
Analysis: Real World Applications, vol. 14, no. 6, pp. 2231–2248,
2013.
[12] Y. Xia, “Global analysis of an impulsive delayed Lotka-Volterra
competition system,” Communications in Nonlinear Science and
Numerical Simulation, vol. 16, no. 3, pp. 1597–1616, 2011.
[13] Y. H. Xia, “Global asymptotic stability of an almost periodic
nonlinear ecological model,” Communications in Nonlinear
Science and Numerical Simulation, vol. 16, no. 11, pp. 4451–4478,
2011.
[14] Y. Xia and M. Han, “New conditions on the existence and
stability of periodic solution in Lotka-Volterra’s population
system,” SIAM Journal on Applied Mathematics, vol. 69, no. 6,
pp. 1580–1597, 2009.
[15] Y. Gao, X. Yuan, Y. Xia, and P. J. Y. Wong, “Linearization
of impulsive differential equations with ordinary dichotomy,”
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 369842, 8 pages
http://dx.doi.org/10.1155/2014/369842

Research Article
Stabilization and Synchronization of Unified Chaotic System via
Impulsive Control

Cheng Hu and Haijun Jiang

College of Mathematics and System Sciences, Xinjiang University, Urumqi, Xinjiang 830046, China

Correspondence should be addressed to Cheng Hu; hucheng@xju.edu.cn

Received 22 May 2014; Accepted 5 June 2014; Published 26 June 2014

Academic Editor: Yonghui Xia

Copyright © 2014 C. Hu and H. Jiang. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

The impulsive control and synchronization of unified chaotic system are proposed. By applying impulsive control theory and
introducing a piecewise continuous auxiliary function, some novel and useful conditions are provided to guarantee the globally
asymptotical stabilization and synchronization of unified chaotic system under impulsive control. Compared with some previous
results, our criteria are superior and less conservative. Finally, the effectiveness of theoretical results is shown through numerical
simulations.

1. Introduction of information transmitted from the driving system to the

driven system in chaos synchronization [10, 20]. Those merits
In 1963, Lorenz presented the first chaotic attractor in a make it more efficient and more useful in a great number
simple three-dimensional autonomous system [1]. Since then, of real-life applications. Up to now, the stabilization and
many researchers began to study chaos theory. In 1999, synchronization of chaotic systems under impulsive control
Chen and Ueta found another chaotic system called Chen have been extensively studied (see [1, 5, 8–10, 15, 16, 19–22]).
system, which is similar but topologically nonequivalent to In this paper, based on impulsive control theory, the
Lorenz system [2]. In 2002, the unified chaotic system was stabilization and synchronization of unified chaotic system
introduced by Lü and Chen. It includes Lorenz system and are investigated via designing impulsive controller. By intro-
Chen attractor as two extremes, respectively, and Lü attractor ducing a piecewise continuous function, some new and useful
as a transition system between Lorenz system and Chen criteria are established to guarantee the stabilization and
system [3, 4]. synchronization of unified chaotic system. Compared with
For a long time, people thought that chaos was neither the main criteria obtained in [8, 10], our results are less
predictable nor controllable. However, in the 1990s, the stringent and more general.
viewpoint and theory about control of chaotic system intro- This paper is organized as follows. In Section 2, two
duced by Ott. Grebogi Yorke and Pecora Carroll completely central lemmas are given. In Section 3, the impulsive control
changed the situation [5], and many scholars began to study model of unified chaotic system is introduced and some
chaos control. Until now, many different techniques and new conditions are provided to ensure the stability of the
methods have been proposed to achieve chaos control and equilibrium point for unified chaotic system under impulsive
synchronization, such as adaptive control [6], fuzzy control control. In Section 4, impulsive synchronization criteria are
[7], and impulsive control [1, 5, 8–22]. proposed. In Section 5, numerical simulations are provided
In the past several years, impulsive control strategy has to illustrate and verify our results.
been widely used to stabilize and synchronize nonlinear
dynamical systems. From the control point of view, impulsive
control is an effective method based on the theory of 2. Preliminaries
impulsive dynamic systems [14, 17, 23]. It allows stabilization
of a complex system by using only small control impulses. In this section, two central lemmas are provided to support
Additionally, this method drastically reduces the amount our analysis in later sections.
2 Abstract and Applied Analysis

Lemma 1. Assume that 𝐴 is an 𝑛 × 𝑛 symmetric and semipos- 3. Impulsive Control of Unified

itive definite real matrix; then, for any two real column vectors Chaotic System
𝜁 ∈ 𝑅n and 𝜂 ∈ 𝑅𝑛 , the following inequality holds:
Consider the unified chaotic system [3, 4, 8, 10] described by
󵄨󵄨 𝑇 󵄨󵄨2 󵄨󵄨 𝑇 󵄨󵄨 󵄨󵄨 𝑇 󵄨󵄨 the following form:
󵄨󵄨𝜁 𝐴𝜂󵄨󵄨 ≤ 󵄨󵄨𝜁 𝐴𝜁󵄨󵄨 ⋅ 󵄨󵄨𝜂 𝐴𝜂󵄨󵄨 . (1)
󵄨 󵄨 󵄨 󵄨 󵄨 󵄨
𝑦1̇ = (25𝛼 + 10) (𝑦2 − 𝑦1 ) ,
Proof. Since matrix 𝐴 is symmetric and semipositive definite,
there exists an 𝑛 × 𝑛 matrix 𝐶 such that 𝑦2̇ = (28 − 35𝛼) 𝑦1 + (29𝛼 − 1) 𝑦2 − 𝑦1 𝑦3 ,
(9)
8+𝛼
𝐴 = 𝐶𝑇 𝐶. (2) 𝑦3̇ = 𝑦1 𝑦2 − 𝑦,
3 3
Then, by using Cauchy-Bunyakovsky inequality, we have where 𝑦1 , 𝑦2 , 𝑦3 are state variables and 𝛼 ∈ [0, 1]. System (9)
has some special features and advantages because it unifies
󵄨󵄨 󵄨2 󵄨 󵄨2 󵄨 󵄨2 Lorenz system (when 𝛼 ∈ [0, 0.8)), Lü system (𝛼 = 0.8),
󵄨󵄨(𝐶𝜁, 𝐶𝜂)󵄨󵄨󵄨 ≤ 󵄨󵄨󵄨𝐶𝜁󵄨󵄨󵄨 ⋅ 󵄨󵄨󵄨𝐶𝜂󵄨󵄨󵄨 . (3)
and Chen system (when 𝛼 ∈ (0.8, 1]). Moreover, system (9) is
always chaotic in the whole interval 𝛼 ∈ [0, 1].
It follows that It is easy to see that system (9) has three equilibrium
󵄨󵄨 𝑇 󵄨󵄨2 󵄨󵄨 𝑇 󵄨󵄨 󵄨󵄨 𝑇 󵄨󵄨 points:
󵄨󵄨𝜁 𝐴𝜂󵄨󵄨 ≤ 󵄨󵄨𝜁 𝐴𝜁󵄨󵄨 ⋅ 󵄨󵄨𝜂 𝐴𝜂󵄨󵄨 . (4)
󵄨 󵄨 󵄨 󵄨 󵄨 󵄨 𝑇
𝐸1 = (0, 0, 0)𝑇 , 𝐸2 = (𝛽, 𝛽, 𝜂) ,
This proof is completed. (10)
𝑇
𝐸3 = (−𝛽, −𝛽, 𝜂) ,
Lemma 2. Assume that 𝑃 is an 𝑛 × 𝑛 symmetric real matrix
and 𝜆𝑚 and 𝜆𝑀 are the smallest and the largest eigenvalues of where 𝛽 = √(8 + 𝛼)(9 − 2𝛼) and 𝜂 = 27 − 6𝛼.
𝑃, respectively. Then, for an arbitrary column vector 𝜉 ∈ 𝑅𝑛 , the For convenience, denote the above equilibrium points as
following inequality holds: (𝑦10 , 𝑦20 , 𝑦30 )𝑇 and introduce the following transformation:

𝑥𝑖 = 𝑦𝑖 − 𝑦𝑖0 , 𝑖 = 1, 2, 3. (11)
𝜆𝑚 𝜉𝑇 𝜉 ≤ 𝜉𝑇 𝑃𝜉 ≤ 𝜆𝑀𝜉𝑇 𝜉. (5)
Then, from (9), the following model can be obtained:
Proof. 𝑃 is a symmetric real matrix; then, there exists an
orthogonal matrix 𝑈 ∈ 𝑅𝑛×𝑛 such that 𝑥1̇ = (25𝛼 + 10) (𝑥2 − 𝑥1 ) ,

𝑥2̇ = (28 − 35𝛼 − 𝑦30 ) 𝑥1 + (29𝛼 − 1) 𝑥2

𝑃 = 𝑈Λ𝑈𝑇 , (6)
(12)
− 𝑦10 𝑥3 − 𝑥1 𝑥3 ,
where Λ = diag(𝜆 1 , 𝜆 2 , . . . , 𝜆 𝑛 ) and 𝜆 𝑖 is the 𝑖th eigenvalue of
𝑃, 𝑖 = 1, 2, . . . , 𝑛. 8+𝛼
For an arbitrary column vector 𝜉 ∈ 𝑅𝑛 , we have 𝑥3̇ = 𝑦20 𝑥1 + 𝑦10 𝑥2 − 𝑥 + 𝑥1 𝑥2 .
3 3
𝑇 We decompose the linear and nonlinear parts of system (12)
𝜉𝑇 𝑃𝜉 = (𝑈𝑇 𝜉) Λ (𝑈𝑇 𝜉) and rewrite it as the following form:
𝑛 𝑥̇= 𝐴𝑥 + Φ (𝑥) ,
󵄨 󵄨2 (13)
= ∑𝜆 𝑖 󵄨󵄨󵄨󵄨(𝑈𝑇 𝜉)𝑖 󵄨󵄨󵄨󵄨
𝑖=1 where 𝑥 = (𝑥1 , 𝑥2 , 𝑥3 )𝑇 and
𝑛
󵄨 󵄨2
≤ 𝜆𝑀∑󵄨󵄨󵄨󵄨(𝑈𝑇 𝜉)𝑖 󵄨󵄨󵄨󵄨 (7) − (25𝛼 + 10) 25𝛼 + 10
0
0
0
𝑖=1 𝐴=(28 − 35𝛼 − 𝑦3 29𝛼 − 1 −𝑦 1 ),
𝛼+8
𝑛 𝑦20 𝑦10 −
󵄨 󵄨2 3
= 𝜆𝑀∑󵄨󵄨󵄨(𝜉)𝑖 󵄨󵄨󵄨 (14)
𝑖=1 0
Φ (𝑥) = (−𝑥1 𝑥3 ) .
= 𝜆𝑀𝜉𝑇 𝜉.
𝑥1 𝑥2
Similarly, we can prove In order to stabilize the equilibrium point, we now
introduce a control model of system (13) described by
𝜉𝑇 𝑃𝜉 ≥ 𝜆𝑚 𝜉𝑇 𝜉. (8) ∞
𝑥̇(𝑡) = 𝐴𝑥 (𝑡) + Φ (𝑥 (𝑡)) + ∑ 𝛿 (𝑡 − 𝜏𝑘 ) 𝐵𝑘 𝑥 (𝑡) , (15)
The proof is completed. 𝑘=1
Abstract and Applied Analysis 3

where 𝛿(⋅) is the Dirac delta function, the time sequence {𝜏𝑘 } The origin of the controlled unified chaotic system (19) is
satisfies 0 = 𝜏0 < 𝜏1 < 𝜏2 < ⋅ ⋅ ⋅ < 𝜏𝑘 < 𝜏𝑘+1 ⋅ ⋅ ⋅ , lim𝑘 → ∞ 𝜏𝑘 = globally asymptotically stable if there exist two constants 𝜇 > 1
∞, and 𝐵𝑘 denotes a 3 × 3 constant matrix for each 𝑘 ∈ 𝑍+ = and 𝑏 ≥ 0 such that one of the following conditions is satisfied.
{1, 2, . . . , 𝑛, 𝑛 + 1, . . .}.
Integrating from 𝜏𝑘 − ℎ to 𝜏𝑘 + ℎ both sides of system (15), (1) For all 𝑘 ∈ 𝑍+ ,
we have
𝑞 𝜇𝑑
𝑥 (𝜏𝑘 + ℎ) − 𝑥 (𝜏𝑘 − ℎ) ( + 𝛾 − 𝑏) 𝛿𝑘 + ln ( 𝑘 ) ≤ 0. (22)
𝜆1 𝜆1
𝜏𝑘 +ℎ ∞
=∫ [𝐴𝑥 (𝑡) + Φ (𝑥 (𝑡)) + ∑ 𝛿 (𝑡 − 𝜏𝑘 ) 𝐵𝑘 𝑥 (𝑡)] 𝑑𝑡, (2) For all 𝑘 ∈ 𝑍+ ,
𝜏𝑘 −ℎ 𝑘=1
(16) 𝑞 𝜇𝑑 𝑑
( + 𝛾 − 𝑏) Δ 𝑘 + ln ( 2𝑘 2 2𝑘−1 ) ≤ 0. (23)
𝜆1 𝜆1
where ℎ > 0 is sufficiently small. As ℎ → 0+ , by applying the
properties of the Dirac delta function, we have
Proof. Firstly, construct an auxiliary function described by
𝑥 (𝜏𝑘+ ) − 𝑥 (𝜏𝑘− ) = 𝐵𝑘 𝑥 (𝜏𝑘 ) . (17) the following form:
Throughout this paper, we always assume that 𝑥(𝑡) =
𝑉 (𝑡, 𝑥) = 𝑒𝑏(𝜏𝑘−1 −𝑡) 𝑥𝑇 (𝑡) 𝑃𝑥 (𝑡) ,
(𝑥1 (𝑡), 𝑥2 (𝑡), 𝑥3 (𝑡))𝑇 is left continuous at 𝜏𝑘 (𝑘 ∈ 𝑍+ ); that (24)
is, 𝑥𝑖 (𝑡) = lim𝑡 → 𝜏𝑘− 𝑥𝑖 (𝑡). In this case, the above equality is 𝑡 ∈ (𝜏𝑘−1 , 𝜏𝑘 ] , 𝑘 = 1, 2, . . . .
equivalent to the following form:
Denote that 𝑄 = 𝐴𝑇 𝑃 + 𝑃𝐴. For 𝑡 ∈ (𝜏𝑘−1 , 𝜏𝑘 ], if 𝑃 ≠𝐼, in view
𝑥 (𝜏𝑘+ ) − 𝑥 (𝜏𝑘 ) = 𝐵𝑘 𝑥 (𝜏𝑘 ) . (18)
of Lemmas 1 and 2, the derivative of 𝑉(𝑡) along the solution
Thus, from the definition of the Dirac delta function, control of (19) is
system (15) can be rewritten as
𝐷+ 𝑉 (𝑡) = 𝑒𝑏(𝜏𝑘−1 −𝑡) [𝑥𝑇 𝑄𝑥 + 𝑥𝑇 𝑃Φ (𝑥) + Φ𝑇 (𝑥) 𝑃𝑥] − 𝑏𝑉
𝑥̇(𝑡) = 𝐴𝑥 (𝑡) + Φ (𝑥 (𝑡)) , 𝑡 ≠𝜏𝑘 ,
(19) = 𝑒𝑏(𝜏𝑘−1 −𝑡) [𝑥𝑇 𝑄𝑥 + 2𝑥𝑇 (𝑃 − 𝐼) Φ (𝑥)] − 𝑏𝑉
𝑥 (𝑡+ ) − 𝑥 (𝑡) = 𝐵𝑘 𝑥 (𝑡) , 𝑡 = 𝜏𝑘 .
It is easy to see that the origin is an equilibrium point of ≤ 𝑒𝑏(𝜏𝑘−1 −𝑡) (𝑥𝑇 𝑄𝑥 + 2√𝑥𝑇 (𝑃 − 𝐼) 𝑥
system (19). We are interested in stabilizing the origin of
system (19), which is equivalent to the asymptotical stability
× √Φ𝑇 (𝑥) (𝑃 − 𝐼) Φ (𝑥)) − 𝑏𝑉
of the equilibrium point (𝑦10 , 𝑦20 , 𝑦30 )𝑇 of system (9).
For convenience, denote 𝛿𝑘 = 𝜏𝑘+1 − 𝜏𝑘 , Δ 𝑘 = 𝜏2𝑘+1 −
𝜏2𝑘−1 (𝑘 ∈ 𝑍+ ) and assume that ≤ 𝑒𝑏(𝜏𝑘−1 −𝑡) (𝑥𝑇 𝑄𝑥 + 2𝜆 2 √𝑥𝑇 𝑥√Φ𝑇 (𝑥) Φ (𝑥)) − 𝑏𝑉
sup {𝛿𝑘 } = 𝜎 < ∞. (20) 󵄨 󵄨
𝑘∈𝑍+ ≤ 𝑒𝑏(𝜏𝑘−1 −𝑡) (𝑥𝑇 𝑄𝑥 + 2𝜆 2 󵄨󵄨󵄨𝑥1 󵄨󵄨󵄨 𝑥𝑇 𝑥) − 𝑏𝑉
Since system (9) is chaotic, there exist two positive numbers 𝑞 2𝜆 𝑀
𝑆1 and 𝑆2 such that |𝑦1 (𝑡)| ≤ 𝑆1 , |𝑦2 (𝑡)| ≤ 𝑆2 , and |𝑦3 (𝑡)| ≤ 𝑆2 ≤ 𝑒𝑏(𝜏𝑘−1 −𝑡) ( + 2 1 ) 𝑥𝑇 𝑃𝑥 − 𝑏𝑉
𝜆1 𝜆1
for 𝑡 ∈ 𝑅+ (see [4]). Additionally, we have |𝑥1 (𝑡)| ≤ 𝑆1 + |𝑦10 | =
𝑀1 , |𝑥2 (𝑡)| ≤ 𝑆2 + |𝑦20 | ≤ 𝑆2 + |𝑦30 | = 𝑀2 , and |𝑥3 (𝑡)| ≤ 𝑞 2𝜆 𝑀
= ( + 2 1 − 𝑏) 𝑉 (𝑡) .
𝑆2 + |𝑦30 | = 𝑀2 for 𝑡 ∈ 𝑅+ . 𝜆1 𝜆1
The following theorem is provided to ensure the globally (25)
asymptotical stability of the origin for the impulsive control
system (19). On the other hand, if 𝑃 = 𝐼,

Theorem 3. Let 𝑃 be a 3 × 3 symmetric and positive definite 𝐷+ 𝑉 (𝑡) = 𝑒𝑏(𝜏𝑘−1 −𝑡) 𝑥𝑇 𝑄𝑥 − 𝑏𝑉

matrix such that (𝑃−𝐼) is semipositive definite, where 𝐼 denotes
𝑞 𝑏(𝜏𝑘−1 −𝑡) 𝑇
the identity matrix. Denote that 𝜆 1 and 𝜆 2 are the smallest and ≤ 𝑒 𝑥 𝑃𝑥 − 𝑏𝑉
the largest eigenvalues of 𝑃 and (𝑃−𝐼), respectively. Let 𝑞 be the 𝜆1 (26)
largest eigenvalue of (𝑃𝐴 + 𝐴𝑇 𝑃). 𝑑𝑘 is the largest eigenvalue 𝑞
of the matrix (𝐼 + 𝐵𝑘 )𝑇 𝑃(𝐼 + 𝐵𝑘 ) and 𝛼 = inf 𝑘∈𝑍+ {𝑑𝑘 } > 0, = ( − 𝑏) 𝑉 (𝑡) .
𝜆1
𝜁 = sup𝑘∈𝑍+ {𝑑𝑘 } < ∞. Denote that
From what has been discussed above, we have
{0, 𝑃 = 𝐼,
𝛾 = { 2𝑀1 𝜆 2 (21) 𝑞
, 𝑃 ≠𝐼. 𝐷+ 𝑉 (𝑡) ≤ ( + 𝛾 − 𝑏) 𝑉 (𝑡) . (27)
{ 𝜆1 𝜆1
4 Abstract and Applied Analysis

Besides, for 𝑡 = 𝜏𝑘 , If 𝑞/𝜆 1 + 𝛾 − 𝑏 < 0, then

𝑘
𝑑𝑖 𝑞
𝑉 (𝜏𝑘+ ) = 𝑒𝑏(𝜏𝑘−1 −𝜏𝑘 ) 𝑥𝑇 (𝜏𝑘+ ) 𝑃𝑥 (𝜏𝑘+ ) 𝑉 (𝑡) ≤ 𝑉 (0) (∏ ) exp [( + 𝛾 − 𝑏) 𝜏𝑘+1 ]
𝑖=1 𝜆 1 𝜆1
𝑇
= 𝑒𝑏(𝜏𝑘−1 −𝜏𝑘 ) 𝑥𝑇 (𝜏𝑘 ) (𝐼 + 𝐵𝑘 ) 𝑃 (𝐼 + 𝐵𝑘 ) 𝑥 (𝜏𝑘 )
𝑞
× exp [( + 𝛾 − 𝑏) (𝑡 − 𝜏𝑘+1 )]
𝑑 (28) 𝜆1
≤ 𝑘 𝑒𝑏(𝜏𝑘−1 −𝜏𝑘 ) 𝑥𝑇 (𝜏𝑘 ) 𝑃𝑥 (𝜏𝑘 )
𝜆1 𝑘
𝑑𝑖 𝑞
𝑑 ≤ 𝑉 (0) 𝛽 (∏ ) exp [( + 𝛾 − 𝑏) 𝜏𝑘+1 ] (35)
= 𝑘 𝑉 (𝜏𝑘 ) . 𝜆
𝑖=1 1 𝜆 1
𝜆1
𝑞
= 𝑉 (0) 𝛽 exp [( + 𝛾 − 𝑏) 𝜏1 ]
From (27), for 𝑡 ∈ [0, 𝜏1 ], 𝜆1
𝑘
𝑞 𝑑𝑖 𝑞
𝑉 (𝑡) ≤ 𝑉 (0) exp [( + 𝛾 − 𝑏) 𝑡] , (29) ×∏ exp ( + 𝛾 − 𝑏) (𝜏𝑖+1 − 𝜏𝑖 ) ,
𝜆1 𝑖=1 𝜆 1 𝜆1

where
which leads to
󵄨󵄨 𝑞 󵄨󵄨
󵄨 󵄨
𝛽 = exp [󵄨󵄨󵄨 + 𝛾 − 𝑏󵄨󵄨󵄨 𝜎] > 1. (36)
𝑞 󵄨󵄨 𝜆 1 󵄨󵄨
𝑉 (𝜏1 ) ≤ 𝑉 (0) exp [( + 𝛾 − 𝑏) 𝜏1 ] . (30)
𝜆1
Hence, for 𝑡 ∈ (𝜏𝑘 , 𝜏𝑘+1 ], we always have
Further, by using (28), 𝑞
𝑉 (𝑡) ≤ 𝑉 (0) 𝛽 exp [( + 𝛾 − 𝑏) 𝜏1 ]
𝜆1
𝑑1 𝑞 (37)
𝑉 (𝜏1+ ) ≤ 𝑉 (0) exp [( + 𝛾 − 𝑏) 𝜏1 ] . (31) 𝑘
𝑑 𝑞
𝜆1 𝜆1 × ∏ 𝑖 exp ( + 𝛾 − 𝑏) (𝜏𝑖+1 − 𝜏𝑖 ) .
𝑖=1 𝜆 1 𝜆1
Similarly, for 𝑡 ∈ (𝜏1 , 𝜏2 ], we obtain
In the following, we consider the following cases.
𝑞
𝑉 (𝑡) ≤ 𝑉 (𝜏1+ ) exp [( + 𝛾 − 𝑏) (𝑡 − 𝜏1 )] Case 1. When condition (9) is satisfied, from (22),
𝜆1
𝑑𝑖 𝑞 1
𝑑 𝑞 exp ( + 𝛾 − 𝑏) (𝜏𝑖+1 − 𝜏𝑖 ) ≤ , 𝑖 ∈ 𝑍+ , (38)
≤ 𝑉 (0) 1 exp [( + 𝛾 − 𝑏) 𝑡] , (32) 𝜆1 𝜆1 𝜇
𝜆1 𝜆1
and together with (37), we have
𝑑 𝑑 𝑞
𝑉 (𝜏2+ ) ≤ 𝑉 (0) 1 2 exp [( + 𝛾 − 𝑏) 𝜏2 ] . 1 𝑞
𝜆1 𝜆1 𝜆1 𝑉 (𝑡) ≤ 𝑉 (0) 𝛽 exp [( + 𝛾 − 𝑏) 𝜏1 ] . (39)
𝜇𝑘 𝜆1
Hence, generally, for 𝑡 ∈ (𝜏𝑘 , 𝜏𝑘+1 ], Note that 𝑘 → ∞ as 𝑡 → ∞; then, from 𝜇 > 1, we obtain
lim𝑡 → ∞ 𝑉(𝑡) = 0. Therefore, we finally have lim𝑡 → ∞ 𝑥(𝑡) = 0.
𝑘
𝑑𝑖 𝑞
𝑉 (𝑡) ≤ 𝑉 (0) (∏ ) exp [( + 𝛾 − 𝑏) 𝑡] . (33) Case 2. If condition (12) holds, from (23), we have
𝑖=1 𝜆 1 𝜆1
𝑑2𝑖 𝑑2𝑖−1 𝑞 1
exp ( + 𝛾 − 𝑏) (𝜏2𝑖+1 − 𝜏2𝑖−1 ) ≤ , 𝑖 ∈ 𝑍+ .
If 𝑞/𝜆 1 + 𝛾 − 𝑏 ≥ 0, then 𝜆21 𝜆1 𝜇
(40)
𝑘
𝑑𝑖 𝑞 For 𝑡 ∈ (𝜏2𝑘−1 , 𝜏2𝑘 ], it follows from (37) and (40) that
𝑉 (𝑡) ≤ 𝑉 (0) (∏ ) exp [( + 𝛾 − 𝑏) 𝜏𝑘+1 ]
𝜆
𝑖=1 1 𝜆 1
1 𝑞
𝑉 (𝑡) ≤ 𝛽𝑉 (0) exp [( + 𝛾 − 𝑏) 𝜏1 ]
𝜇𝑘−1 𝜆1
𝑞
= 𝑉 (0) exp [( + 𝛾 − 𝑏) 𝜏1 ] (34)
𝜆1 𝑑2𝑘−1 𝑞
× exp ( + 𝛾 − 𝑏) (𝜏2𝑘 − 𝜏2𝑘−1 ) (41)
𝜆1 𝜆1
󵄨󵄨󵄨 𝑞 󵄨󵄨
𝑘
𝑑𝑖 𝑞 1 󵄨 𝜁
×∏ exp ( + 𝛾 − 𝑏) (𝜏𝑖+1 − 𝜏𝑖 ) . ≤ 𝑘−1 𝑉 (0) 𝛽 exp [󵄨󵄨󵄨 + 𝛾 − 𝑏󵄨󵄨󵄨 (𝜏1 + 𝜎)] .
𝜆
𝑖=1 1 𝜆 1 𝜇 󵄨󵄨 𝜆 1 󵄨󵄨 𝜆1
Abstract and Applied Analysis 5

Similarly, if 𝑡 ∈ (𝜏2𝑘 , 𝜏2𝑘+1 ], from (37) and (40), we can derive Corollary 8. The origin of system (19) is globally asymptoti-
that cally stable if there exist 𝜇 > 1 and 𝑏 ≥ 0 such that
1 𝑞 (𝑞 − 𝑏) 𝛿 + ln (𝜇𝑑) ≤ 0. (48)
𝑉 (𝑡) ≤ 𝑉 (0) 𝛽 exp [( + 𝛾 − 𝑏) 𝜏1 ] . (42)
𝜇𝑘 𝜆1
Remark 9. From Corollary 8, an upper bound of equal
From (41) and (42), we finally have lim𝑡 → ∞ 𝑥(𝑡) = 0. impulsive interval 𝛿 can be derived. In fact,
From the above analysis, the trivial solution 𝑥(𝑡) = 0 1 1
of system (19) is globally asymptotically stable. The proof of 𝛿≤ ln . (49)
𝑞 − 𝑏 𝜇𝑑
Theorem 3 is completed.
In [10], the upper bound of impulsive interval was also given,
Remark 4. From Theorem 3, it is clear that 𝛾 = 0 if 𝑃 = 𝐼. On which can be represented by
the other hand, for 𝑃 ≠𝐼, 𝛾 → 0 as 𝑃 approaches to 𝐼, since
𝜆 2 → 0 in this case. 1 1
𝛿≤ ln . (50)
𝑞 𝜇𝑑
Remark 5. If 𝜇 = 1 in Theorem 3, from the proof of
It is evident that the equal impulsive interval obtained in this
Theorem 3, we have
paper is larger, which shows that our designed controller is
𝑞 more efficient and more economical in practice.
𝑉 (𝑡) ≤ 𝑉 (0) exp [( + 𝛾 − 𝑏) 𝜏1 ] 𝛽𝑄, (43)
𝜆1
where 𝑄 = max{1, (𝜁/𝜆 1 ) exp[|𝑞/𝜆 1 + 𝛾 − 𝑏|𝜎]}. In this case,
4. Impulsive Synchronization of Unified
it is clear that the origin of impulsive control system (19) is Chaotic System
stable.
In this section, we discuss the impulsive synchronization of
Taking 𝑃 = 𝐼 in Theorem 3, it is easy to see that 𝜆 1 = 1, unified chaotic system. It is easy to see that system (9) can be
𝜆 2 = 0, and 𝛾 = 0 in this case. By virtue of Theorem 3, the rewritten as
following statements can be derived. 𝑦̇= 𝐴𝑦 + Φ (𝑦) , (51)
Corollary 6. The origin of the controlled unified chaotic where 𝑦 = (𝑦1 , 𝑦2 , 𝑦3 )𝑇 and
system (19) is globally asymptotically stable if there exist two
constants 𝜇 > 1 and 𝑏 ≥ 0 such that one of the following − (25𝛼 + 10) 25𝛼 + 10 0
conditions holds. 𝐴 = ( 28 − 35𝛼 29𝛼 − 1 0 ),
𝛼+8
0 0 −
(a) For all 𝑘 ∈ 𝑍+ , 3 (52)
(𝑞 − 𝑏) 𝛿𝑘 + ln (𝜇𝑑𝑘 ) ≤ 0. (44) 0
Φ (𝑦) = (−𝑦1 𝑦3 ) .
(b) For all 𝑘 ∈ 𝑍+ , 𝑦1 𝑦2

(𝑞 − 𝑏) Δ 𝑘 + ln (𝜇𝑑2𝑘 𝑑2𝑘−1 ) ≤ 0, 𝑘 ∈ 𝑍+ . (45) In impulsive synchronization configuration, the driving sys-

tem is given by (51), while the driven system is represented by
Remark 7. In Corollary 6, if 𝑏 = 0, inequality (44) is reduced the following form:
to ∞
̃̇= 𝐴𝑦̃ + Φ (𝑦)
𝑦 ̃ + ∑ 𝛿 (𝑡 − 𝜏𝑘 ) 𝐵𝑘 (𝑦̃ (𝑡) − 𝑦 (𝑡)) , (53)
𝑞𝛿𝑘 + ln (𝜇𝑑𝑘 ) ≤ 0, (46) 𝑘=1

which is equivalent to the condition of Theorem 1 in [10]. where 𝑦̃ = (𝑦̃1 , 𝑦̃2 , 𝑦̃3 )𝑇 is the state variable of the driven
Further, in this case, inequality (45) is degenerated to system, 𝐴 and Φ are defined in (52), and 𝐵𝑘 , 𝛿(⋅), and {𝜏𝑘 }
are defined in (15).
𝑞Δ 𝑘 + ln (𝜇𝑑2𝑘 𝑑2𝑘−1 ) ≤ 0, (47)
Similar to the technique of system (15), the driven system
which is the same as inequality (7) in [8]. On the other (53) can be rewritten as the following impulsive equation:
hand, for 𝑏 > 0, evidently, conditions (44) and (45) are
̃̇= 𝐴𝑦̃ + Φ (𝑦)
𝑦 ̃ , 𝑡 ≠𝜏𝑘 ,
less conservative than inequalities (46) and (47). Hence, (54)
compared with [8, 10], our results are less stringent and more Δ𝑦̃ = 𝐵𝑘 (𝑦̃ (𝑡) − 𝑦 (𝑡)) , 𝑡 = 𝜏𝑘 .
general.
Let 𝑒 = (𝑒1 , 𝑒2 , 𝑒3 )𝑇 = (𝑦̃1 − 𝑦1 , 𝑦̃2 − 𝑦2 , 𝑦̃3 − 𝑦3 )𝑇 be the
In practice, the equal impulsive interval and the constant synchronization error and
gain matrix are often selected for convenience. In Corollary 6,
we choose 𝛿𝑖 = 𝛿 and 𝐵𝑖 = 𝐵 for 𝑖 ∈ 𝑍+ and denote that 𝑞 and 0
𝑑 are the largest eigenvalues of (𝐴𝑇 + 𝐴) and (𝐼 + 𝐵)𝑇 (𝐼 + 𝐵), Ψ (𝑦, 𝑦) ̃ − Φ (𝑦) = (𝑦1 𝑦3 − 𝑦̃1 𝑦̃3 ) .
̃ = Φ (𝑦) (55)
respectively. Then, the following corollary can be obtained. 𝑦̃1 𝑦̃2 − 𝑦1 𝑦2
6 Abstract and Applied Analysis

From (51) and (54), the error system can be expressed by

60
𝑒 ̇= 𝐴𝑒 + Ψ (𝑦, 𝑦)
̃ , 𝑡 ≠𝜏𝑘 ,
(56) 40
Δ𝑒 = 𝐵𝑘 𝑒, 𝑡 = 𝜏𝑘 .

y3
20
In this section, we are interested in synchronizing the 40
driving system (51) and the driven system (54), which is 20
0
equivalent to the asymptotical stability of the origin for 40 0
20 y1
error system (56). Similar to Theorem 3, the following 0 −20
results are stated, which ensure the globally asymptotical y2 −20 −40
−40
synchronization of the driving system (51) and the driven
system (54). Figure 1: The chaotic behavior of system (9) with 𝛼 = 1.

Theorem 10. Let 𝑞 be the largest eigenvalue of (𝐴 + 𝐴𝑇 ). 𝑑𝑘

is the largest eigenvalue of the matrix (𝐼 + 𝐵𝑘 )T (𝐼 + 𝐵𝑘 ). The Corollary 11. The driving system (51) and the driven system
driving system (51) and the driven system (54) are globally (54) are globally asymptotically synchronized, if there exist 𝜇 >
asymptotically synchronized if there exist constants 𝜇 > 1 and 1 and 𝑏 ≥ 0 such that one of the following conditions is satisfied.
𝑏 ≥ 0 such that one of the following conditions is satisfied.
(a) For all 𝑘 ∈ 𝑍+ ,
+
(a) For all 𝑘 ∈ 𝑍 ,
(𝑞 + 2𝑆2 − 𝑏) 𝛿𝑘 + ln (𝜇𝑑𝑘 ) ≤ 0. (62)
(𝑞 + 2𝑆2 − 𝑏) 𝛿𝑘 + ln (𝜇𝑑𝑘 ) ≤ 0. (57)
(b) For all 𝑘 ∈ 𝑍+ ,
(b) For all 𝑘 ∈ 𝑍+ ,
(𝑞 + 2𝑆2 − 𝑏) Δ 𝑘 + ln (𝜇𝑑2𝑘 𝑑2𝑘−1 ) ≤ 0, 𝑘 ∈ 𝑍+ . (63)
(𝑞 + 2𝑆2 − 𝑑) Δ 𝑘 + ln (𝜇𝑑2𝑘 𝑑2𝑘−1 ) ≤ 0. (58)
Remark 12. The impulsive synchronization of unified chaotic
Proof. Observe that error system (56) is almost the same system was investigated in [8, 10]. It is noted that inequality
̃ Construct an auxiliary
as system (19) except for Ψ(𝑦, 𝑦). (62) is reduced to the condition of Theorem 2 in [10] and
function described by the following form: condition (b) in Corollary 11 will be degenerated to the
assumption of Theorem 2 in [8] when 𝑏 = 0. Additionally,
𝑉 (𝑡, 𝑥) = 𝑒𝑏(𝜏𝑘−1 −𝑡) 𝑒𝑇 (𝑡) 𝑒 (𝑡) , if 𝑏 > 0, it is evident that conditions (62) and (63) are less
(59) stringent than the assumptions introduced in [8, 10].
𝑡 ∈ (𝜏𝑘−1 , 𝜏𝑘 ] , 𝑘 = 1, 2, . . . .
Similar to Corollary 8, the following results can be derived
For 𝑡 ∈ (𝜏𝑘−1 , 𝜏𝑘 ], calculating the time derivative of 𝑉(𝑡) along when we select 𝛿𝑖 = 𝛿 and 𝐵𝑖 = 𝐵 for each 𝑘 ∈ 𝑍+ .
the solution of (56), we have
Corollary 13. The driving system (51) and the driven system
𝐷+ 𝑉 (𝑡) (54) are globally asymptotically synchronized, if there exist 𝜇 >
1 and 𝑏 ≥ 0 such that
= 𝑒𝑏(𝜏𝑘−1 −𝑡) [𝑒𝑇 (𝐴𝑇 + 𝐴) 𝑒 + 𝑒𝑇 Ψ (𝑦, 𝑦)
̃ + Ψ𝑇 (𝑦, 𝑦)
̃ 𝑒] − 𝑏𝑉
(𝑞 + 2𝑆2 − 𝑏) 𝛿 + ln (𝜇𝑑) ≤ 0, (64)
= 𝑒𝑏(𝜏𝑘−1 −𝑡) [𝑒𝑇 (𝐴𝑇 + 𝐴) 𝑒 + 2 (𝑦2 𝑒1 𝑒3 − 𝑦3 𝑒1 𝑒2 )] − 𝑏𝑉
where 𝑑 is the largest eigenvalue of (𝐼 + 𝐵)𝑇 (𝐼 + 𝐵).
≤ 𝑒𝑏(𝜏𝑘−1 −𝑡) [𝑒𝑇 (𝐴𝑇 + 𝐴) 𝑒 + 2𝑆2 𝑒𝑇 𝑒] − 𝑏𝑉
5. Numerical Simulations
≤ 𝑒𝑏(𝜏𝑘−1 −𝑡) (𝑞 + 2𝑆2 ) 𝑒𝑇 𝑒 − 𝑏𝑉
In this section, based on the results obtained in the previous
≤ (𝑞 + 2𝑆2 − 𝑏) 𝑉. sections, some numerical simulations are represented to show
(60) the effectiveness of our results.
First, we consider numerical simulations of impulsive
In addition, for 𝑘 ∈ 𝑍+ , control for system (9) with 𝛼 = 1. The chaotic behavior of
𝑇
system (9) with 𝛼 = 1 is shown in Figure 1.
𝑉 (𝜏𝑘+ ) = 𝑒𝑏(𝜏𝑘−1 −𝜏𝑘 ) 𝑒𝑇 (𝜏𝑘 ) (𝐼 + 𝐵𝑘 ) (𝐼 + 𝐵𝑘 ) 𝑒 (𝜏𝑘 ) For convenience, select (𝑦10 , 𝑦20 , 𝑦30 ) = (0, 0, 0), 𝑃 = 𝐼 and
(61) take gain matrices 𝐵𝑖 = diag(−0.86, −0.84, −0.9) (𝑖 ∈ 𝑍+ ) in
≤ 𝑑𝑘 𝑉 (𝜏𝑘 ) .
Theorem 3. By simple computation, we obtain
The rest of this proof is the same as that of Theorem 3, and −70 28 0
here we omit it. The proof of Theorem 10 is completed. 𝐴 + 𝐴𝑇 = ( 28 56 0 ) , (65)
Similar to Corollary 6, we have the following results. 0 0 −6
Abstract and Applied Analysis 7

10 60
5
y1

50
0
40
0 0.5 1 1.5 2 2.5 3 3.5 4
t 30

y3
20
15
10 10
y2

5
0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 −10
t 50

10 y2 0 20 30
−50 0 10
−20 −10
5 y1
y3

0 Figure 3: The chaotic behavior of system (51) with 𝛼 = 0.

0 0.5 1 1.5 2 2.5 3 3.5 4
t
8
6
Figure 2: Impulsive stabilization of system (9) with 𝛼 = 1 and 𝛿 = 4

e1
0.07. 2
0
0 0.2 0.4 0.6 0.8 1
and its eigenvalues are −75.942, −6, and 61.942; then, 𝑞 = t
61.942.
3
Choosing impulsive control interval 𝛿 = 0.07, 𝜇 = 1.1. By 2
computation, 𝑑 = 0.0256 and 1
e2

0
𝑞𝛿 + ln (𝜇𝑑) = 61.942 × 0.07 + ln (1.1 × 0.0256) −1
(66) 0 0.2 0.4 0.6 0.8 1
= 0.7661. t

Taking 𝑏 = 12 in Theorem 3, then, 0

−2
e3

𝑏𝛿 = 12𝛿 = 0.84. (67) −4

−6
Hence, 0 0.2 0.4 0.6 0.8 1
t
(𝑞 − 𝑏) 𝛿 + ln (𝜇𝑑) = −0.0739 < 0. (68)
Figure 4: Synchronization error curves with 𝛼 = 0 and 𝛿 = 0.02.
Therefore, all conditions of Corollary 8 are satisfied, which
implies that the origin of unified chaotic system (9) is globally
asymptotically stable under impulsive control. The numerical
simulation results with constant impulsive interval 𝛿 = 0.07 synchronized under impulsive control. The synchronization
are shown in Figure 2. errors with impulsive distance 𝛿 = 0.02 are shown in Figure 4.
In the second simulation, we study impulsive synchro-
nization of the driving system (51) and the driven system Remark 14. Evidently, it follows from (66) and (69) that the
(54) with 𝛼 = 0. The initial conditions of the driving conditions of Theorems 1 and 2 in [10] are not satisfied under
system and the driven system are (−2, −8, −1) and (5, −9, −6), the above parameters; that is to say, the stabilization and
respectively. Figure 3 shows that system (51) with 𝛼 = 0 is synchronization of unified chaotic system are not ensured
chaotic, and from this we can obtain that 𝑆2 = 65. by means of the criteria obtained in [10]. However, it follows
Choose 𝐵𝑖 = diag(−0.88, −0.8, −0.8), 𝛿 = 0.02, 𝜇 = 1.5, from (68) and (70) that the stabilization and synchronization
and 𝑏 = 20 in (43). By computation, 𝑞 = 28.0512, 𝑑 = 0.04 of unified chaotic system can be realized by virtue of our
and criteria and Figures 2 and 4 show the validity of our results.
Therefore, our results are less conservative and more superior
(𝑞 + 2𝑆2 ) 𝛿 + ln (𝜇𝑑) = 0.3476, (69) compared with [10].

and 𝑏𝛿 = 0.4, which leads to

6. Conclusion
(𝑞 + 2𝑆2 − 𝑏) 𝛿 + ln (𝜇𝑑) = −0.0524. (70)
In this paper, the issue on the stabilization and synchro-
By the above analysis, all conditions of Corollary 13 hold; nization of unified chaotic system under impulsive control
hence, system (51) and system (54) are globally asymptotically is investigated. Some novel and useful criteria are derived
8 Abstract and Applied Analysis

by using impulsive control theory. Finally, the effectiveness [12] C. Li, X. Liao, and R. Zhang, “Impulsive synchronization of
and feasibility of the developed methods are shown by some nonlinear coupled chaotic systems,” Physics Letters A: General,
numerical simulations. Compared with some previous results Atomic and Solid State Physics, vol. 328, no. 1, pp. 47–50, 2004.
given in [8, 10], our designed controller is more efficient and [13] Z. G. Li, C. Y. Wen, and Y. C. Soh, “Analysis and design of
economical. impulsive control systems,” IEEE Transactions on Automatic
Control, vol. 46, no. 6, pp. 894–897, 2001.
[14] X. Liu, Y. Liu, and K. L. Teo, “Stability analysis of impulsive
Conflict of Interests control systems,” Mathematical and Computer Modelling, vol. 37,
no. 12-13, pp. 1357–1370, 2003.
The authors declare that there is no conflict of interests
regarding the publication of this paper. [15] J. Sun, “Impulsive control of a new chaotic system,” Mathematics
and Computers in Simulation, vol. 64, no. 6, pp. 669–677, 2004.
[16] J. Sun and Y. Zhang, “Impulsive control and synchronization of
Acknowledgments Chua’s oscillators,” Mathematics and Computers in Simulation,
vol. 66, no. 6, pp. 499–508, 2004.
This work was supported by the Natural Science Foun-
[17] J. Sun, Y. Zhang, and Q. Wu, “Less conservative conditions
dation of Xinjiang (Grant no. 2013211B06), the Project for asymptotic stability of impulsive control systems,” IEEE
funded by China Postdoctoral Science Foundation (Grant no. Transactions on Automatic Control, vol. 48, no. 5, pp. 829–831,
2013M540782), the Natural Science Foundation of Xinjiang 2003.
University (Grant no. BS120101), the Specialized Research [18] T. Yang, “Impulsive control,” IEEE Transactions on Automatic
Fund for the Doctoral Program of Higher Education (Grant Control, vol. 44, no. 5, pp. 1081–1083, 1999.
no. 20136501120001), and the National Natural Science Foun- [19] X. Zhang, X. Liao, and C. Li, “Impulsive control, complete and
dation of China (Grant no. 61164004). lag synchronization of unified chaotic system with continuous
periodic switch,” Chaos, Solitons and Fractals, vol. 26, no. 3, pp.
References 845–854, 2005.
[20] Q. Zhang and J. Lu, “Impulsive control and synchronization of
[1] W. Xie, C. Wen, and Z. Li, “Impulsive control for the stabiliza- a critical chaotic system,” Wuhan University Journal of Natural
tion and synchronization of Lorenz systems,” Physics Letters A: Sciences, vol. 12, no. 3, pp. 426–430, 2007.
General, Atomic and Solid State Physics, vol. 275, no. 1-2, pp. 67– [21] Y. Zhang and J. Sun, “Controlling chaotic Lu systems using
72, 2000. impulsive control,” Physics Letters A: General, Atomic and Solid
[2] G. Chen and T. Ueta, “Yet another chaotic attractor,” Interna- State Physics, vol. 342, no. 3, pp. 256–262, 2005.
tional Journal of Bifurcation and Chaos, vol. 9, no. 7, pp. 1465– [22] J. Zhou, X. H. Cheng, and Y. C. Zhang, “Impulsive control
1466, 1999. and synchronization of chaotic systems consisting of Van der
[3] C. Tao, H. Xiong, and F. Hu, “Two novel synchronization Pol oscillators coupled to linear oscillators,” Chaos, Solitons and
criterions for a unified chaotic system,” Chaos Solitons and Fractals, vol. 33, pp. 607–616, 2007.
Fractals, vol. 27, pp. 111–114, 2006. [23] V. Laskshmikantham, D. Bainov, and P. S. Simeonov, Theory
[4] X. Wang and J. Song, “Synchronization of the unified chaotic of Impulsive Differential Equations, World Scientific, Singapore,
system,” Nonlinear Analysis: Theory, Methods and Applications, 1989.
vol. 69, no. 10, pp. 3409–3416, 2008.
[5] J. Ma, B. Ren, and Y. Chen, “Impulsive control of chaotic
attractors in nonlinear chaotic systems,” Applied Mathematics
and Mechanics (English Edition), vol. 25, no. 9, pp. 971–976,
2004.
[6] X. Guan, Z. Fan, H. Peng, and Y. Wang, “The adaptive control
of Chen’s chaotic system,” Acta Physica Sinica, vol. 50, no. 11, pp.
2110–2111, 2001.
[7] X. Liu and S. Zhong, “T-S fuzzy model-based impulsive control
of chaotic systems with exponential decay rate,” Physics Letters
A: General, Atomic and Solid State Physics, vol. 370, no. 3-4, pp.
260–264, 2007.
[8] D. Chen, J. Sun, and C. Huang, “Impulsive control and synchro-
nization of general chaotic system,” Chaos, Solitons and Fractals,
vol. 28, no. 1, pp. 213–218, 2006.
[9] D. Chen, J. Sun, and Q. Wu, “Impulsive control and its
application to Lü’s chaotic system,” Chaos, Solitons and Fractals,
vol. 21, no. 5, pp. 1135–1142, 2004.
[10] S. Chen, Q. Yang, and C. Wang, “Impulsive control and
synchronization of unified chaotic system,” Chaos, Solitons and
Fractals, vol. 20, no. 4, pp. 751–758, 2004.
[11] G. Jiang, Q. Lu, and L. Qian, “Chaos and its control in an
impulsive differential system,” Chaos, Solitons and Fractals, vol.
34, no. 4, pp. 1135–1147, 2007.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 735327, 7 pages
http://dx.doi.org/10.1155/2014/735327

Research Article
Global Stability of a Computer Virus Propagation Model with
Two Kinds of Generic Nonlinear Probabilities

Chenquan Gan,1,2 Xiaofan Yang,1,2 and Qingyi Zhu2

1
School of Software Engineering, Chongqing University, Chongqing 400044, China
2
College of Computer Science, Chongqing University, Chongqing 400044, China

Correspondence should be addressed to Chenquan Gan; gcq2010cqu@163.com

Received 1 April 2014; Accepted 4 June 2014; Published 26 June 2014

Academic Editor: Tonghua Zhang

Copyright © 2014 Chenquan Gan et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Vaccination is one of the most effective measures for suppressing the spread of computer virus, and the bilinear incidence rate
assumption for the majority of previous models, which is a good first approximation of the general incidence rate, is in disagreement
with the reality. In this paper, a new dynamical model with two kinds of generic nonlinear probabilities (incidence rate and
vaccination probability) is established. An exhaustive mathematical analysis of this model shows that (a) there are two equilibria,
virus-free equilibrium and viral equilibrium, and (b) the virus-free (or viral) equilibrium is globally asymptotically stable when
the basic reproduction number is less (or greater) than unity. The analysis of the basic reproduction number is also included.
Additionally, some numerical examples are given to illustrate the main results, from which it can be seen that the generic nonlinear
vaccination is helpful to strengthen computer security.

1. Introduction plays an important role in repressing computer virus, by

which a susceptible computer would have temporary immu-
Establishing rational dynamical models underlying the nity. The fact that a large number of susceptible computers
mechanism of the spread of computer virus is a signifi- are infected would enhance the probability that the user
cant issue in computer security, which can give important of a susceptible computer has his/her computer vaccinated,
insights to defend against viral spread. Since the seminal implying that vaccination probability is related to the number
work by Cohen [1] and Murray [2] as well as Kephart and of infected computers. Indeed, Gan et al. [9, 11] recently
White [3], various propagation models, such as susceptible- investigated two SIRS models by incorporating a linear or
infected-susceptible (SIS) model [4, 5], susceptible-infected- nonlinear vaccination probability (i.e., the probability that a
recovered (SIR) model [6, 7], susceptible-infected-recovered- susceptible computer gets vaccinated is linear or nonlinear in
susceptible (SIRS) model [8–13], susceptible-exposed-infect- the number of currently infected computers). Unfortunately,
ed-removed (SEIR) model [14, 15], susceptible-exposed- the bilinear incidence rate assumption for these two models,
infected-quarantined-recovered-susceptible (SEIQRS) model which is a good first approximation of the general incidence
[16], susceptible-antidotal-infectious-contaminated (SAIC) rate, is inconsistent with the actual conditions [10]. In real-
model [17], susceptible-antidotal-infected-removed (SAIR) ity, overcrowded infected computers and active protection
model [18], susceptible-latent-breaking-susceptible (SLBS) measures would render this approximation to fail terribly.
model [19, 20], susceptible-latent-breaking-recovered-sus- As a result, it is worthwhile to explore a dynamical model
ceptible (SLBRS) model [21], susceptible-infected-counter- with generic nonlinear vaccination probability under more
measure-susceptible (SICS) model [22], and susceptible-in- reasonable assumptions.
fected-external-susceptible (SIES) model [23–25], have been Having this idea in mind, in this paper, a new dynamical
widely developed. model of computer virus with generic nonlinear vaccination
Vaccination (i.e., the measure that an uninfected com- probability and nonlinear incidence rate is proposed. A
puter has the newest-version antivirus software installed) detailed study of the model is provided. Specifically, the basic
2 Abstract and Applied Analysis

reproduction number (i.e., the average number of secondary 𝛼1 f(I)S

infections produced by a single infected computer during
its life time), 𝑅0 , is determined, and the virus-free (or viral) 𝛽IS
𝛾1 I
(1 − p)b h(I) pb
equilibrium is shown to be globally asymptotically stable if S I R
𝑅0 ≤ 1 (or 𝑅0 > 1), implying that computer virus would 𝛾2 I
𝜇S 𝜇I 𝜇R
tend to extinction or persist according to the value of 𝑅0 .
The related analysis of 𝑅0 is also conducted. Additionally, 𝛼2 R
some numerical examples are examined to illustrate the main
Figure 1: The transfer diagram of the new model.
results, from which it can be seen that the generic nonlinear
vaccination is helpful to suppress computer virus diffusion.
The organization of the rest of the paper is as follows.
Section 2 formulates the new model. Section 3 proves the
global stabilities of the virus-free and viral equilibria. A (A6) Every recovered internal computer loses immunity
parameter analysis of the basic reproduction number is with probability per unit time 𝛼2 > 0.
performed in Section 4. Finally, Section 5 summarizes this
work. This collection of assumptions can be schematically
shown in Figure 1, from which one can derive the differential
2. Model Formulation system

As usual, a computer is either internal (i.e., on the Internet)

𝛽𝑆𝐼
or external (i.e., outside the Internet). Moreover, an internal 𝑆 ̇= (1 − 𝑝) 𝑏 − 𝜇𝑆 − − 𝛼1 𝑓 (𝐼) 𝑆 + 𝛾2 𝐼 + 𝛼2 𝑅,
computer is assumed to be in one of three possible states: ℎ (𝐼)
susceptible (i.e., uninfected but not immune), infected, and
𝛽𝑆𝐼 (1)
recovered (i.e., uninfected and immune). 𝐼 ̇= − 𝜇𝐼 − 𝛾1 𝐼 − 𝛾2 𝐼,
Now, let us introduce some notations as follows, which ℎ (𝐼)
will be adopted in the sequel: 𝑅̇= 𝑝𝑏 − 𝜇𝑅 − 𝛼2 𝑅 + 𝛼1 𝑓 (𝐼) 𝑆 + 𝛾1 𝐼,
𝑆(𝑡): the average number of susceptible internal computers
at time 𝑡,
with initial condition (𝑆(0), 𝐼(0), 𝑅(0)) ∈ R3+ .
𝐼(𝑡): the average number of infected internal computers at
time 𝑡,
𝑅(𝑡): the average number of recovered internal computers 3. Model Analysis
at time 𝑡,
This section is devoted to study model (1) theoretically. The
𝑁(𝑡): the average number of internal computers at time 𝑡; analysis of this model comprises the basic reproduction num-
that is, 𝑁(𝑡) = 𝑆(𝑡) + 𝐼(𝑡) + 𝑅(𝑡). ber, the existence of equilibria, and their global stabilities.
For brevity, let 𝑆, 𝐼, 𝑅, and 𝑁 represent 𝑆(𝑡), 𝐼(𝑡), 𝑅(𝑡), and
𝑁(𝑡), respectively.
3.1. Basic Reproduction Number. Employing the next gener-
The following fundamental assumptions of the new
ation method (see [26]) to model (1), the basic reproduction
model are made.
number can be derived as
(A1) External computers enter the Internet at rate 𝑏 > 0,
of which a fraction of (1 − 𝑝) (resp., 𝑝) is susceptible
𝛽𝑏 (𝜇 + 𝛼2 − 𝜇𝑝)
(resp., immune), 0 ≤ 𝑝 ≤ 1. 𝑅0 = . (2)
𝜇 (𝜇 + 𝛼1 + 𝛼2 ) (𝜇 + 𝛾1 + 𝛾2 )
(A2) Every internal computer leaves the Internet with
probability per unit time 𝜇 > 0.
(A3) Every susceptible internal computer is infected by 3.2. Equilibria. Obviously, system (1) always has a unique
infected internal computers with probability per unit virus-free equilibrium 𝐸0 = (𝑆0 , 0, 𝑅0 ), where
time 𝛽𝐼/ℎ(𝐼), where 𝛽 > 0; function ℎ(𝐼) is continu-
ously differentiable with ℎ(0) = 1 and ℎ󸀠 (𝐼) ≥ 0.
𝑏 (𝜇 + 𝛼2 − 𝜇𝑝) 𝑏 (𝛼1 + 𝜇𝑝)
(A4) Due to treatment, every infected internal computer 𝑆0 = , 𝑅0 = . (3)
𝜇 (𝜇 + 𝛼1 + 𝛼2 ) 𝜇 (𝜇 + 𝛼1 + 𝛼2 )
becomes recovered (resp., susceptible) with probabil-
ity per unit time 𝛾1 > 0 (resp., 𝛾2 > 0).
(A5) Due to vaccination, every susceptible internal com- Next, let us examine the existence of viral equilibria. The
puter acquires temporary immunity with probability following result is obtained.
per unit time 𝛼1 𝑓(𝐼), where 𝛼1 > 0; function 𝑓(𝐼) is
continuously differentiable with 𝑓(0) = 1 and 𝑓󸀠 (𝐼) ≥ Lemma 1. System (1) has a unique viral equilibrium 𝐸∗ =
0. (𝑆∗ , 𝐼∗ , 𝑅∗ ) if 𝑅0 > 1.
Abstract and Applied Analysis 3

Proof. All viral equilibria of system (1) are determined by the 3.3. Global Stability of the Virus-Free Equilibrium
following system of equations:
Theorem 2. 𝐸0 is globally asymptotically stable if 𝑅0 ≤ 1.
𝛽𝑆𝐼
(1 − 𝑝) 𝑏 − 𝜇𝑆 − − 𝛼1 𝑓 (𝐼) 𝑆 + 𝛾2 𝐼 + 𝛼2 𝑅 = 0, Proof. Let 𝑥 = 𝑁 − 𝑏/𝜇, 𝑦 = 𝐼, and 𝑧 = 𝑅 − 𝑅0 ; then, system
ℎ (𝐼)
(8) can be rewritten as
𝛽𝑆𝐼 (4)
− 𝜇𝐼 − 𝛾1 𝐼 − 𝛾2 𝐼 = 0, 𝑥̇= − 𝜇𝑥,
ℎ (𝐼)
𝑝𝑏 − 𝜇𝑅 − 𝛼2 𝑅 + 𝛼1 𝑓 (𝐼) 𝑆 + 𝛾1 𝐼 = 0, 𝛽𝑦
𝑦̇= (𝑥 − 𝑦 − 𝑧 + 𝑆0 ) − (𝜇 + 𝛾1 + 𝛾2 ) 𝑦,
ℎ (𝑦) (10)
where 𝐼 ≠0. Simplifying, one can get
𝑧̇= − (𝜇 + 𝛼2 ) 𝑧 + 𝛾1 𝑦
(1 − 𝑝) 𝑏 − 𝜇𝑆 − (𝜇 + 𝛾1 ) 𝐼 − 𝛼1 𝑓 (𝐼) 𝑆 + 𝛼2 𝑅 = 0,
+ 𝛼1 𝑓 (𝑦) (𝑥 − 𝑦 − 𝑧 + 𝑆0 ) − 𝛼1 𝑆0 .
𝜇 + 𝛾1 + 𝛾2
𝑆= ℎ (𝐼) , Consider Lyapunov function:
𝛽 (5)
𝑏 𝛼2 2 𝛾1 𝑦 𝛼 𝑆0 𝑦 ℎ (𝑢) [𝑓 (𝑢) − 1]
𝑅= − 𝑆 − 𝐼. 𝑉= 𝑥 + ∫ ℎ (𝑢) 𝑑𝑢 + 1 ∫ 𝑑𝑢
𝜇 8𝜇 𝛽 0 𝛽 0 𝑢
𝛼1 𝑦 1
Substituting the last two equations into the first equation, one + ∫ 𝑓 (𝑢) ℎ (𝑢) 𝑑𝑢 + (𝑥 − 𝑧)2 .
can obtain 𝛽 0 2
(11)
𝜇 + 𝛾1 + 𝛾2 𝑏
𝐹 (𝐼) := 𝑝𝑏 + (𝜇 + 𝛼2 ) (𝐼 + ℎ (𝐼) − ) + 𝛾1 𝐼 Then,
𝛽 𝜇
𝜇 + 𝛾1 + 𝛾2 𝛼 𝛾 𝛼 𝑆0 ℎ (𝑦) [𝑓 (𝑦) − 1]
+ 𝛼1 𝑓 (𝐼) ℎ (𝐼) = 0. 𝑉|̇(6) = 2 𝑥𝑥
̇ + 1 ℎ (𝑦) 𝑦̇+ 1 𝑦̇
𝛽 4𝜇 𝛽 𝛽 𝑦
𝛼1
(6) + 𝑓 (𝑦) ℎ (𝑦) 𝑦̇+ (𝑥 − 𝑧) (𝑥̇− 𝑧)̇
𝛽
Note that ℎ(0) = 1, ℎ󸀠 (𝐼) ≥ 0, 𝑓(0) = 1, 𝑓󸀠 (𝐼) ≥ 0, and 𝑅0 > 1. 𝑥 2
Then, = − 𝜇(𝑥 − 𝑧)2 − 𝛼2 ( − 𝑧)
2
𝐹󸀠 (𝐼) > 0, 𝛾1 (𝜇 + 𝛾1 + 𝛾2 ) 𝑦
− [ℎ (𝑦) − 𝑅0 ]
𝛽
lim 𝐹 (𝐼) = +∞,
𝐼→∞ (7)
− 𝛼1 𝑆0 [𝑓 (𝑦) − 1] 𝑦
(𝜇 + 𝛼1 + 𝛼2 ) (𝜇 + 𝛾1 + 𝛾2 )
𝐹 (0) = (1 − 𝑅0 ) < 0. 𝛼1 𝑆0 [𝑓 (𝑦) − 1] (𝜇 + 𝛾1 + 𝛾2 )
𝛽 − [ℎ (𝑦) − 𝑅0 ] − 𝛾1 𝑦2
𝛽
Thus, function 𝐹(𝐼) has a unique positive zero. The proof is
2
complete. − 𝛼1 𝑓 (𝑦) (𝑥 − 𝑦 − 𝑧)

For convenience, in what follows we will investigate the 𝛼1 𝑦𝑓 (𝑦) (𝜇 + 𝛾1 + 𝛾2 )

− [ℎ (𝑦) − 𝑅0 ]
following equivalent system of model (1). Consider 𝛽

𝑁̇= 𝑏 − 𝜇𝑁, ≤ 0.
(12)
𝛽 (𝑁 − 𝐼 − 𝑅) 𝐼
𝐼 ̇= − (𝜇 + 𝛾1 + 𝛾2 ) 𝐼, (8)
ℎ (𝐼) 𝑉|̇(6) = 0 if and only if (𝑥, 𝑦, 𝑧) = (0, 0, 0); that is,
(𝑆, 𝐼, 𝑅) = 𝐸0 . Hence, the claimed result follows from the
𝑅̇= 𝑝𝑏 − (𝜇 + 𝛼2 ) 𝑅 + 𝛼1 𝑓 (𝐼) (𝑁 − 𝐼 − 𝑅) + 𝛾1 𝐼, LaSalle Invariance Principle [27].

with initial condition (𝑁(0), 𝐼(0), 𝑅(0)) ∈ R3+ . Clearly, the set Remark 3. Theorem 2 implies that computer virus on the
Internet would tend to extinction when the basic reproduc-
𝑏 tion number is less than or equal to unity.
Ω = {(𝑁, 𝐼, 𝑅) ∈ R3+ : 𝑅 + 𝐼 ≤ 𝑁 ≤ } (9)
𝜇
Example 4. Consider system (1) with 𝑝 = 0.3, 𝑏 = 5, 𝜇 = 0.01,
is positively invariant for system (8). and 𝛽 = 0.00004, 𝛼1 = 0.002, 𝛼2 = 0.03, 𝛾1 = 0.008,
4 Abstract and Applied Analysis

500
450
400 250
350 200
Values of S, I, and R

300 150
R
250 100
200
50
150
0
100 80
60 500
50 40 400
300
0 I 20 200
100
0 100 200 300 400 500 0 0 S
Time t

S R
I
(a) Time plots of 𝑆, 𝐼, and 𝑅 (b) Phase portrait for system (1)

Figure 2: An illustration of the dynamics of system (1) given in Example 4.

𝛾2 = 0.002, 𝑓(𝐼) = 𝐼 + 1, and ℎ(𝐼) = 𝐼2 + 1. Then, 𝑅0 = Proof. Let 𝑥 = 𝑁 − 𝑏/𝜇, 𝑦 = 𝐼 − 𝐼∗ , and 𝑧 = 𝑅 − 𝑅∗ . Rewrite
0.881 < 1. Figure 2(a) displays the time plots of 𝑆, 𝐼, and system (8) as
𝑅 for the initial condition (𝑆(0), 𝐼(0), 𝑅(0)) = (480, 10, 10),
and Figure 2(b) shows the phase portrait for system (1), both 𝑥̇= − 𝜇𝑥,
consistent with Theorem 2 and Remark 3.
𝛽 (𝑦 + 𝐼∗ ) 𝑆∗
𝑦̇= {𝑥 − 𝑦 − 𝑧 − [ℎ (𝑦 + 𝐼∗ ) − ℎ (𝐼∗ )]} ,
3.4. Global Stability of the Viral Equilibrium. Firstly, let us ℎ (𝑦 + 𝐼∗ ) ℎ (𝐼∗ )
consider the following lemma.
𝑧̇= − (𝜇 + 𝛼2 ) 𝑧 + 𝛾1 𝑦
∗
Lemma 5. Assume that 𝑅0 > 1. For 𝑦 ∈ (−𝐼 , +∞), let
+ 𝛼1 𝑓 (𝑦 + 𝐼∗ ) (𝑥 − 𝑦 − 𝑧 + 𝑆∗ )
𝑦+𝐼∗ ℎ (𝑢) (𝑢 − 𝐼∗ )
𝑉1 (𝑦) = ∫ 𝑑𝑢, − 𝛼1 𝑓 (𝐼∗ ) 𝑆∗ .
𝐼∗ 𝑢 (14)
𝑦+𝐼∗ ℎ (𝑢) 𝑓 (𝑢) (𝑢 − 𝐼∗ )
𝑉2 (𝑦) = ∫ 𝑑𝑢, Consider Lyapunov function:
𝐼∗ 𝑢
𝑦+𝐼∗ ℎ (𝑢) [𝑓 (𝑢) − 𝑓 (𝐼∗ )] 𝛼2 2 𝛾1 𝛼
𝑉3 (𝑦) = ∫ 𝑑𝑢, (13) 𝑉= 𝑥 + 𝑉1 (𝑦) + 1 𝑉2 (𝑦)
𝐼∗ 𝑢 8𝜇 𝛽 𝛽
(15)
𝑉4 (𝑦) = 𝑦 [ℎ (𝑦 + 𝐼∗ ) − ℎ (𝐼∗ )] , 𝛼 𝑆∗ 1
+ 1 𝑉3 (𝑦) + (𝑥 − 𝑧)2 .
𝛽 2
𝑉5 (𝑦) = 𝑦 [𝑓 (𝑦 + 𝐼∗ ) − 𝑓 (𝐼∗ )] ,
Then,
𝑉6 (𝑦) = [𝑓 (𝑦 + 𝐼∗ ) − 𝑓 (𝐼∗ )] [ℎ (𝑦 + 𝐼∗ ) − ℎ (𝐼∗ )] .

Then, 𝑉𝑖 (𝑦) ≥ 0, 𝑖 = 1, 2, . . . , 6. Moreover, 𝑉𝑖 (𝑦) = 0 if and only 𝛼 𝛾 𝑑𝑉 (𝑦) 𝛼 𝑑𝑉 (𝑦)

𝑉|̇(8) = 2 𝑥𝑥
̇+ 1 1 𝑦̇+ 1 2 𝑦̇
if 𝑦 = 0. 4𝜇 𝛽 𝑑𝑦 𝛽 𝑑𝑦

Proof. Here, we will prove only the result of 𝑉1 (𝑦). The proofs 𝛼1 𝑆∗ 𝑑𝑉3 (𝑦)
+ 𝑦̇+ (𝑥 − 𝑧) (𝑥̇− 𝑧)̇
of the remaining five functions are similar. It follows from 𝛽 𝑑𝑦
𝑑𝑉1 (𝑦)/𝑑𝑦 = ℎ(𝑦 + 𝐼∗ )𝑦/(𝑦 + 𝐼∗ ) that 𝑉1 (𝑦) is decreasing
𝑥 2
in (−𝐼∗ , 0] and is increasing in (0, +∞). As 𝑉1 (0) = 0, the = − 𝜇(𝑥 − 𝑧)2 − 𝛼2 ( − 𝑧)
claimed result follows. 2
Now, let us explore the global stability of the viral 𝛾1 𝑆∗ 𝛼1 𝑆∗ 𝑓 (𝑦 + 𝐼∗ )
− 𝑉4 (𝑦) − 𝑉4 (𝑦)
equilibrium. ℎ (𝐼∗ ) ℎ (𝐼∗ )
2
Theorem 6. 𝐸∗ is globally asymptotically stable if 𝑅0 > 1. − 𝛼1 𝑓 (𝑦 + 𝐼∗ ) (𝑥 − 𝑦 − 𝑧)
Abstract and Applied Analysis 5

450
400
300
350
250
Values of S, I, and R

300
200
250 R 150
200 100
150 50
0
100 80
60 300
50
40 200
0 I 20 100
0 50 100 150 200 250 300 0 0 S
Time t

S R
I
(a) Time plots of 𝑆, 𝐼, and 𝑅 (b) Phase portrait for system (1)

Figure 3: An illustration of the dynamics of system (1) given in Example 8.

𝛼1 𝑆∗
2 𝜕𝑅0
− 𝛾1 𝑦2 − 𝛼1 𝑆∗ 𝑉5 (𝑦) − 𝑉 (𝑦) 𝜕𝜇
ℎ (𝐼∗ ) 6
≤ 0. 𝛽𝑏 (1 − 𝑝) (𝛼1 + 𝛼2 + 𝛾1 + 𝛾2 + 2𝜇)
=− 2 2
(16) (𝛼1 + 𝛼2 + 𝜇) (𝛾1 + 𝛾2 + 𝜇)

𝛽𝑏𝛼2 [(𝛼1 + 𝛼2 + 2𝜇) (𝛾1 + 𝛾2 + 𝜇) + 𝜇 (𝜇 + 𝛼1 + 𝛼2 )]

𝑉|̇(8) = 0 if and only if (𝑥, 𝑦, 𝑧) = (0, 0, 0); that is, (𝑆, 𝐼, 𝑅) = − 2 2
𝐸∗ . Therefore, the claimed result follows from the LaSalle 𝜇2 (𝛼1 + 𝛼2 + 𝜇) (𝛾1 + 𝛾2 + 𝜇)
Invariance Principle [27].
< 0.
Remark 7. Theorem 6 implies that computer virus on the (17)
Internet would tend to persist when the basic reproduction
number is greater than unity. The proof is complete.

Example 8. Consider system (1) with 𝑝 = 0.3, 𝑏 = 5, 𝜇 = 0.01, Furthermore, the influences of system parameters (i.e.,
𝛽 = 0.006, 𝛼1 = 0.001, 𝛼2 = 0.03, 𝛾1 = 0.008, 𝛾2 = 0.002, 𝛼1 , 𝛼2 , 𝛾1 , 𝛾2 , 𝛽, and 𝜇) on 𝑅0 are illustrated in Figures 4, 5,
𝑓(𝐼) = 𝐼 + 1, and ℎ(𝐼) = 𝐼 + 1. Then, 𝑅0 = 135.3659 > 1. and 6, in coherence with Theorem 9.
Figure 3(a) demonstrates the time plots of 𝑆, 𝐼, and 𝑅 for In addition, the following example indicates the effect of
the initial condition (𝑆(0), 𝐼(0), 𝑅(0)) = (450, 10, 40), and different incidence rates and vaccination probabilities on 𝐼
Figure 3(b) exhibits the phase portrait for system (1), both in (see Figure 7).
accordance with Theorem 6 and Remark 7.
Example 10. Consider system (1) with 𝑝 = 0.3, 𝑏 = 5, 𝜇 =
0.01, and 𝛽 = 0.006, 𝛼1 = 0.001, 𝛼2 = 0.03, 𝛾1 = 0.008,
4. Discussions
and 𝛾2 = 0.002 and the initial condition (𝑆(0), 𝐼(0), 𝑅(0)) =
Remarks 3 and 7 tell us the fact that computer virus on the (450, 10, 40).
Internet would tend to extinction or persist according to the
value of the basic reproduction number. So, the analysis of 𝑅0 5. Conclusions
is performed.
This paper has studied the long-term behavior of com-
Theorem 9. Consider (2). Then, 𝜕𝑅0 /𝜕𝛽 > 0, 𝜕𝑅0 /𝜕𝛾1 < 0, puter virus in terms of a new propagation model with
𝜕𝑅0 /𝜕𝛾2 < 0, 𝜕𝑅0 /𝜕𝛼1 < 0, 𝜕𝑅0 /𝜕𝛼2 > 0, and 𝜕𝑅0 /𝜕𝜇 < 0. generic nonlinear incidence rate and nonlinear vaccination
probability. An elaborate analysis of the model including
Proof. It is easy to see that the first four inequalities are true. the basic reproduction number, the existence of virus-free
Consider and viral equilibria, and their global stabilities has been
conducted, from which it is found that computer virus on
𝜕𝑅0 𝛽𝑏 (𝛼1 + 𝑝𝜇) the Internet would tend to extinction or persist according
= > 0,
𝜕𝛼2 𝜇 (𝛾1 + 𝛾2 + 𝜇) (𝛼1 + 𝛼2 + 𝜇)2 to the value of the basic reproduction number. To illustrate
6 Abstract and Applied Analysis

450

150 400

350
100
300
R0

50 250
I
200
0
0.4 150
0.4
0.3 0.3
0.2 0.2 100
𝛼2 0.1 0.1
0 0 𝛼1 50

Figure 4: An illustration of the impact of 𝛼1 and 𝛼2 on 𝑅0 . 0

0 50 100 150 200
Time t

h(I) = 1, f(I) = 1 h(I) = 1 + I, f(I) = 1

h(I) = 1, f(I) = 1 + I h(I) = 1 + I, f(I) = 1 + I
300
250 Figure 7: An illustration of the impact of different incidence rates
200 and vaccination probabilities on 𝐼 given in Example 10.
150
R0

100
50 Acknowledgments
0
0.04
The authors are grateful to the anonymous reviewers for their
0.03 0.08 valuable suggestions that have greatly improved the quality
0.06
0.02 0.04 of this paper. This work is supported by the Natural Science
𝛾2 0.01 0.02
0 𝛾1 Foundation of China (Grant nos. 11347150 and 10771227) and
0
the Doctorate Foundation of Educational Ministry of China
Figure 5: An illustration of the impact of 𝛾1 and 𝛾2 on 𝑅0 . (Grant no. 20110191110022).

References
2000 [1] F. Cohen, “Computer viruses: theory and experiments,” Com-
puters & Security, vol. 6, no. 1, pp. 22–35, 1987.
1500 [2] W. H. Murray, “The application of epidemiology to computer
viruses,” Computers and Security, vol. 7, no. 2, pp. 139–145, 1988.
1000
R0

[3] J. O. Kephart and S. R. White, “Directed-graph epidemiological

500 models of computer viruses,” in Proceedings of the IEEE Com-
puter Society Symposium on Research in Security and Privacy,
×10−3 0 pp. 343–359, Oakland, Calif, USA, May 1991.
4 ×10−6
2 [4] L. Billings, W. M. Spears, and I. B. Schwartz, “A unified
3 1.5
15
2 prediction of computer virus spread in connected networks,”
1
𝛽 1 0.5 𝜇 Physics Letters A, vol. 297, no. 3-4, pp. 261–266, 2002.
0 0
[5] J. C. Wierman and D. J. Marchette, “Modeling computer virus
prevalence with a susceptible-infected-susceptible model with
Figure 6: An illustration of the impact of 𝛽 and 𝜇 on 𝑅0 .
reintroduction,” Computational Statistics & Data Analysis, vol.
45, no. 1, pp. 3–23, 2004.
[6] M. Youssef and C. Scoglio, “An individual-based approach
the obtained main results, some numerical examples have to SIR epidemics in contact networks,” Journal of Theoretical
been examined, from which it can be seen that the generic Biology, vol. 283, pp. 136–144, 2011.
nonlinear vaccination is useful for the inhibition of viral [7] J. Ren, Y. Xu, Y. Zhang, Y. Dong, and G. Hao, “Dynamics of
spread. a delay-varying computer virus propagation model,” Discrete
Dynamics in Nature and Society, vol. 2012, Article ID 371792,
12 pages, 2012.
Conflict of Interests [8] L. Feng, X. Liao, H. Li, and Q. Han, “Hopf bifurcation analysis
of a delayed viral infection model in computer networks,”
The authors declare that there is no conflict of interests Mathematical and Computer Modelling, vol. 56, no. 7-8, pp. 167–
regarding the publication of this paper. 179, 2012.
Abstract and Applied Analysis 7

[9] C. Gan, X. Yang, W. Liu, Q. Zhu, and X. Zhang, “Propagation of [26] J. M. Heffernan, R. J. Smith, and L. M. Wahl, “Perspectives
computer virus under human intervention: a dynamical model,” on the basic reproductive ratio,” Journal of the Royal Society
Discrete Dynamics in Nature and Society, vol. 2012, Article ID Interface, vol. 2, no. 4, pp. 281–293, 2005.
106950, 8 pages, 2012. [27] R. C. Robinson, An Introduction to Dynamical System: Contin-
[10] C. Gan, X. Yang, W. Liu, Q. Zhu, and X. Zhang, “An epidemic uous and Discrete, Prentice Hall, 2004.
model of computer viruses with vaccination and generalized
nonlinear incidence rate,” Applied Mathematics and Computa-
tion, vol. 222, pp. 265–274, 2013.
[11] C. Gan, X. Yang, W. Liu, and Q. Zhu, “A propagation model of
computer virus with nonlinear vaccination probability,” Com-
munications in Nonlinear Science and Numerical Simulation, vol.
19, no. 1, pp. 92–100, 2014.
[12] X. Han and Q. Tan, “Dynamical behavior of computer virus on
Internet,” Applied Mathematics and Computation, vol. 217, no. 6,
pp. 2520–2526, 2010.
[13] C. Zhang, Y. Zhao, and Y. Wu, “An impulse model for computer
viruses,” Discrete Dynamics in Nature and Society, vol. 2012,
Article ID 260962, 13 pages, 2012.
[14] B. K. Mishra and S. K. Pandey, “Dynamic model of worm
s with vertical transmission in computer network,” Applied
Mathematics and Computation, vol. 217, no. 21, pp. 8438–8446,
2011.
[15] T. Dong, X. Liao, and H. Li, “Stability and Hopf bifurcation in
a computer virus model with multistate antivirus,” Abstract and
Applied Analysis, vol. 2012, Article ID 841987, 16 pages, 2012.
[16] B. K. Mishra and N. Jha, “SEIQRS model for the transmission of
malicious objects in computer network,” Applied Mathematical
Modelling, vol. 34, no. 3, pp. 710–715, 2010.
[17] J. R. C. Piqueira, A. A. de Vasconcelos, C. E. C. J. Gabriel, and V.
O. Araujo, “Dynamic models for computer viruses,” Computers
and Security, vol. 27, no. 7-8, pp. 355–359, 2008.
[18] J. R. C. Piqueira and V. O. Araujo, “A modified epidemiological
model for computer viruses,” Applied Mathematics and Compu-
tation, vol. 213, no. 2, pp. 355–360, 2009.
[19] L.-X. Yang, X. Yang, L. Wen, and J. Liu, “A novel computer virus
propagation model and its dynamics,” International Journal of
Computer Mathematics, vol. 89, no. 17, pp. 2307–2314, 2012.
[20] C. Zhang, W. Liu, J. Xiao, and Y. Zhao, “Hopf bifurcation of
an improved SLBS model under the influence of latent period,”
Mathematical Problems in Engineering, vol. 2013, Article ID
196214, 10 pages, 2013.
[21] M. Yang, Z. Zhang, Q. Li, and G. Zhang, “An SLBRS model
with vertical transmission of computer virus over the internet,”
Discrete Dynamics in Nature and Society, vol. 2012, Article ID
925648, 17 pages, 2012.
[22] Q. Zhu, X. Yang, L. Yang, and X. Zhang, “A mixing propagation
model of computer viruses and countermeasures,” Nonlinear
Dynamics: An International Journal of Nonlinear Dynamics and
Chaos in Engineering Systems, vol. 73, no. 3, pp. 1433–1441, 2013.
[23] C. Gan, X. Yang, Q. Zhu, J. Jin, and L. He, “The spread
of computer virus under the effect of external computers,”
Nonlinear Dynamics, vol. 73, no. 3, pp. 1615–1620, 2013.
[24] C. Gan, X. Yang, W. Liu, Q. Zhu, J. Jin, and L. He, “Propaga-
tion of computer virus both across the Internet and external
computers: a complex-network approach,” Communications in
Nonlinear Science and Numerical Simulation, vol. 19, no. 8, pp.
2785–2792, 2014.
[25] C. Gan, X. Yang, Q. Zhu, and J. Jin, “The combined impact
of external computers and network topology on the spread of
computer viruses,” International Journal of Computer Mathe-
matics, 2014.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 943167, 9 pages
http://dx.doi.org/10.1155/2014/943167

Research Article
Traveling Wave Solutions of the
Kadomtsev-Petviashvili-Benjamin-Bona-Mahony Equation

Zhengyong Ouyang
Department of Mathematics, Foshan University, Foshan, Guangdong 528000, China

Correspondence should be addressed to Zhengyong Ouyang; zyouyang math@163.com

Received 26 March 2014; Accepted 17 April 2014; Published 18 June 2014

Academic Editor: Junling Ma

Copyright © 2014 Zhengyong Ouyang. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

We use bifurcation method of dynamical systems to study exact traveling wave solutions of a nonlinear evolution equation. We
obtain exact explicit expressions of bell-shaped solitary wave solutions involving more free parameters, and some existing results
are corrected and improved. Also, we get some new exact periodic wave solutions in parameter forms of the Jacobian elliptic
function. Further, we find that the bell-shaped waves are limits of the periodic waves in some sense. The results imply that we can
deduce bell-shaped waves from periodic waves for some nonlinear evolution equations.

1. Introduction dynamical systems are employed to study (3) [4–6], and some
solitary wave solutions and triangle periodic wave solutions
The Benjamin-Bona-Mahony (BBM) equation [1] were obtained.
However, there is no method which can be applied to all
𝑢𝑡 + 𝑢𝑥 + 𝑢𝑢𝑥 − 𝑢𝑥𝑥𝑡 = 0 (1) nonlinear evolution equations. The research on the solutions
was proposed as the model for propagation of long waves of the KP-BBM equation now appears insufficient. Further
where nonlinear dispersion is incorporated. The Kadomtsev- studies are necessary for the traveling wave solutions of the
Petviashvili (KP) equation [2] KP-BBM equation. The purpose of this paper is to apply the
bifurcation method [7–10] of dynamical systems to continue
(𝑢𝑡 + 𝑎𝑢𝑢𝑥 + 𝑢𝑥𝑥𝑥 )𝑥 + 𝑢𝑦𝑦 = 0 (2) to seek traveling waves of (3). Firstly, we obtain bell-shaped
solitary wave solutions involving more free parameters, and
was given as the generalization of the KdV equation. In some results in [6] are corrected and improved. Then, we
addition, both BBM and KdV equations can be used to get some new periodic wave solutions in parameter forms of
describe long wavelength in liquids, fluids, and so forth. Jacobian elliptic function, and numerical simulation verifies
Combining the two equations, the Kadomtsev-Petviashvili- the validity of these periodic solutions. The periodic wave
Benjamin-Bona-Mahony (KP-BBM) equation solutions obtained in this paper are different from those in
[5]. Furthermore, we find an interesting relationship between
(𝑢𝑡 + 𝑢𝑥 − 𝑎(𝑢2 )𝑥 − 𝑏𝑢𝑥𝑥𝑡 ) + 𝑘𝑢𝑦𝑦 = 0 (3) the bell-shaped waves and periodic waves; that is, the bell-
𝑥
shaped waves are limits of the periodic waves in forms of
was presented in [3] for further study. Some methods are Jacobian elliptic function as modulus approaches 1.
developed and applied to find exact solutions of nonlinear This paper is organized as follows. First, we draw the
evolution equations because exact solutions play an impor- bifurcation phase portraits of planar system according to
tant role in the comprehension of nonlinear phenomena. the KP-BBM equation in Section 2. Second, bell-shaped
For instance, extended tanh method, extended mapping solitary wave solutions to the equation under consideration
method with symbol computation, and bifurcation method of are presented in Section 3. Third, periodic solitary waves are
2 Abstract and Applied Analysis

given in the forms of Jacobian elliptic function and numerical When (1+𝑘𝑟2 −𝑐)/𝑏𝑐 < 0, there are two homoclinic orbits
simulation is done. Finally, the relationship between the Γ1 and Γ2 (see Figures 1(a) and 1(b)). In 𝜑-𝑦 plane, Γ1 and Γ2
bell-shaped solitary waves and periodic waves is proved in can be described by
Section 4.
2𝑎 3 1 + 𝑘𝑟2 − 𝑐 2
𝑦2 = 𝜑 − 𝜑,
2. Bifurcation Phase Portraits of System (6) 3𝑏𝑐 𝑏𝑐 (10)
𝜑 ∈ (0, 𝜑∗ ) or 𝜑 ∈ (𝜑∗ , 0) ,
Suppose (3) possesses traveling wave solutions in the form
𝑢(𝑥, 𝑡) = 𝜑(𝜉), 𝜉 = 𝑥 + 𝑟𝑦 − 𝑐𝑡, where 𝑐 is the wave speed and where 𝜑∗ = 3(1 + 𝑘𝑟2 − 𝑐)/2𝑎. That is,
𝑟 is a real constant. Substituting 𝑢(𝑥, 𝑡) = 𝜑(𝜉), 𝜉 = 𝑥 + 𝑟𝑦 − 𝑐𝑡
into (3) admits to the following ODE:
2𝑎 3 1 + 𝑘𝑟2 − 𝑐 2
𝑦 = ±√ 𝜑 − 𝜑. (11)
2 󸀠󸀠 2 󸀠󸀠 (4) 3𝑏𝑐 𝑏𝑐
(1 + 𝑘𝑟 − 𝑐) 𝜑 − 𝑎(𝜑 ) + 𝑏𝑐𝜑 = 0, (4)
Substituting (11) into d𝜑/d𝜉 = 𝑦 and integrating along
where the derivative is for variable 𝜉. Integrating (4) twice
homoclinc orbits Γ1 and Γ2 , respectively, we get
with respect to 𝜉 and letting the first integral constant take
value zero, it follows that 𝜑∗
d𝑠 󵄨 󵄨
2 2 󸀠󸀠
∫ = ± 󵄨󵄨󵄨𝜉󵄨󵄨󵄨 . (12)
(1 + 𝑘𝑟 − 𝑐) 𝜑 − 𝑎𝜑 + 𝑏𝑐𝜑 = 𝑔, (5) 𝜑 √(2𝑎/3𝑏𝑐) 𝑠3 − ((1 + 𝑘𝑟2 − 𝑐) /𝑏𝑐) 𝑠2
where 𝑔 is the second integral constant. Completing the above integration, it follows that
Equation (5) is equivalent to the following two-dimen-
sional system: 𝑢1 (𝑥, 𝑦, 𝑡)
2 2
d𝜑 d𝑦 𝑎𝜑 − (1 + 𝑘𝑟 − 𝑐) 𝜑 + 𝑔 3 (1 + 𝑘𝑟2 − 𝑐)
= 𝑦, = . (6) = .
d𝜉 d𝜉 𝑏𝑐
𝑎 [1 + cosh (√− ((1 + 𝑘𝑟2 − 𝑐) /𝑏𝑐) (𝑥 + 𝑟𝑦 − 𝑐𝑡))]
It is obvious that system (6) has the first integral (13)
2
1 𝑎 3 1 + 𝑘𝑟 − 𝑐 2 𝑔 Remark 1. 𝑢1 (𝑥, 𝑦, 𝑡) is a bright soliton solution when 𝑏𝑐/𝑎 <
𝐻 (𝜑, 𝑦) = 𝑦2 − 𝜑 + 𝜑 − 𝜑 = ℎ, (7)
2 3𝑏𝑐 2𝑏𝑐 𝑏𝑐 0 and a dark soliton solution when 𝑏𝑐/𝑎 > 0. If the real
number 𝑟 in (13) takes value 1, then solution (13) is the same to
where ℎ is the constant of integration.
solution (1.2) in [6]. Solution (1.1) in [6] is not a real solution
Define Δ = (1+𝑘𝑟2 −𝑐)2 −4𝑎𝑔. When Δ > 0, there are two of the KP-BBM equation; it is obvious that solution (1.1) tends
equilibrium points (𝜑1 , 0) and (𝜑2 , 0) of (6) on 𝜑-axis, where to infinite as 𝜉 → 0, and it does not satisfy the KP-BBM
𝜑1 = ((1 + 𝑘𝑟2 − 𝑐) − √Δ)/2𝑎, 𝜑2 = ((1 + 𝑘𝑟2 − 𝑐) + √Δ)/2𝑎. equation (3).
The Hamiltonian 𝐻 of (𝜑1 , 0) and (𝜑2 , 0) is denoted by ℎ1 =
𝐻(𝜑1 , 0) and ℎ2 = 𝐻(𝜑2 , 0). When (1+𝑘𝑟2 +𝑐)/𝑏𝑐 > 0, there are two homoclinic orbits
In the case of 1 + 𝑘𝑟2 − 𝑐 < 0 and 1 + 𝑘𝑟2 − 𝑐 > 0, the Γ3 and Γ4 (see Figures 2(a) and 2(b)). Similarly solitary wave
bifurcation phase portraits of system (6) see Figures 1 and 2 solutions according to Γ3 and Γ4 are obtained as follows:
in [6], respectively, in which there are some homoclinic and
periodic orbits of system (6). For our purpose, we redraw the 𝑢2 (𝑥, 𝑦, 𝑡)
homolinic and periodic orbits in this paper(see Figures 1–3).
= ((1 + 𝑘𝑟2 − 𝑐)
3. Exact Explicit Expressions of
Solitary Wave Solutions
1 + 𝑘𝑟2 − 𝑐
In this section, we discuss bell-shaped wave solutions under × [−2 + cosh (√ (𝑥 + 𝑟𝑦 − 𝑐𝑡))])
𝑔 = 0 and 𝑔 ≠0, respectively. 𝑏𝑐
[ ]
−1
3.1. The Case 𝑔 = 0. System (6) can be rewritten as 1 + 𝑘𝑟2 − 𝑐
× (𝑎 [1 + cosh (√ (𝑥 + 𝑟𝑦 − 𝑐𝑡))]) .
2 2 𝑏𝑐
d𝜑 d𝑦 𝑎𝜑 − (1 + 𝑘𝑟 − 𝑐) 𝜑 [ ]
= 𝑦, = . (8) (14)
d𝜉 d𝜉 𝑏𝑐
The first integral of (8) is Remark 2. If the real number 𝑟 in (14) takes value 1, then
solution (14) is the same to solution (1.4) in [6]. Solution (1.5)
1 𝑎 3 1 + 𝑘𝑟2 − 𝑐 2 in [6] is not a real solution of the KP-BBM equation; it is easy
𝐻 (𝜑, 𝑦) = 𝑦2 − 𝜑 + 𝜑 = ℎ. (9) to verify that it does not satisfy the KP-BBM equation (3).
2 3𝑏𝑐 2𝑏𝑐
Abstract and Applied Analysis 3

y y
Γ1 Γ2

Γ7 Γ8

𝜑
𝜑 r1 r2 0 r3
r1 0 r2 r3 𝜑∗ 𝜑∗

(a) 𝑏𝑐/𝑎 < 0 (b) 𝑏𝑐/𝑎 > 0

𝜑 𝜑
r1 0 r2 r3 r1 r2 0 r3

(c) (d)

Figure 1: The bifurcation phase portraits of system (6) with 𝑔 = 0 and (1 + 𝑘𝑟2 + 𝑐)/𝑏𝑐 < 0.

3.2. The Case 𝑔 ≠0. There are two homoclinic orbits Γ5 and Substituting (17) into d𝜑/d𝜉 = 𝑦 and integrating along hom-
Γ6 when 𝑔 ≠0 (see Figures 3(a) and 3(b)). Γ5 and Γ6 can be oclinic orbits Γ5 , we get
described by
𝜑3
d𝑠 󵄨 󵄨
∫ = 󵄨󵄨󵄨𝜉󵄨󵄨󵄨 , (18)
2 2
2𝑎 3 1 + 𝑘𝑟 − 𝑐 2 2𝑔 𝜑 √(2𝑎/3𝑏𝑐) (𝑠 − 𝜑1 ) (𝑠 − 𝜑3 )
𝑦2 = 𝜑 − 𝜑 + 𝜑 + 2ℎ. (15)
3𝑏𝑐 𝑏𝑐 𝑏𝑐
where 𝜑1 = (1+𝑘𝑟2 −𝑐−√Δ)/2𝑎 and 𝜑3 = (1+𝑘𝑟2 −𝑐+2√Δ)/2𝑎
When 𝑏𝑐/𝑎 < 0, the corresponding homoclinic orbit Γ5 if 𝑎 > 0 and 𝑏𝑐 < 0; then, completing (18) we get the following
has a double zero point 𝜑1 and a zero point 𝜑3 on 𝜑-axis (see solution:
Figure 3(a)), so (15) can be rewritten as
𝑢3 (𝑥, 𝑦, 𝑡)
2𝑎 2
𝑦2 = (𝜑 − 𝜑1 ) (𝜑 − 𝜑3 ) ; (16)
3𝑏𝑐
= ( (1 + 𝑘𝑟2 − 𝑐 − √Δ)
that is,

2𝑎 2 √Δ
𝑦 = ±√ (𝜑 − 𝜑1 ) (𝜑 − 𝜑3 ). (17) × cosh √ − (𝑥 + 𝑟𝑦 − 𝑐𝑡) + 1
3𝑏𝑐 𝑏𝑐
4 Abstract and Applied Analysis

Γ3 y y Γ4

𝜑 𝜑
r1 r2 0 r3 𝜑∗ r1 0 r2 r3
𝜑∗

Γ10
Γ9

(a) 𝑏𝑐/𝑎 < 0 (b) 𝑏𝑐/𝑎 > 0

𝜑
r1 r2 0 r3 𝜑
r1 0 r2 r3

(c) (d)

Figure 2: The bifurcation phase portraits of system (6) with 𝑔 = 0 and (1 + 𝑘𝑟2 + 𝑐)/𝑏𝑐 > 0.

√Δ
+𝑘𝑟2 − 𝑐 + 5√Δ) × cosh √ (𝑥 + 𝑟𝑦 − 𝑐𝑡) + 1 + 𝑘𝑟2 − 𝑐 − 5√Δ)
𝑏𝑐
−1
√Δ
× (2𝑎 (cosh √ − (𝑥 + 𝑟𝑦 − 𝑐𝑡) + 1)) . −1
𝑏𝑐 √Δ
× (2𝑎 (cosh √ (𝑥 + 𝑟𝑦 − 𝑐𝑡) + 1)) .
(19) 𝑏𝑐

In (18), 𝜑1 = (1 + 𝑘𝑟2 − 𝑐 + √Δ)/2𝑎 and 𝜑3 = (1 + 𝑘𝑟2 − 𝑐 − (20)

2√Δ)/2𝑎 if 𝑎 < 0 and 𝑏𝑐 > 0; then, completing (18) we get the
following solution:
Remark 3. If the real number 𝑟 in (19) and (20) takes value
𝑢4 (𝑥, 𝑦, 𝑡) 1, then solutions (19) and (20) are the same to solutions (1.6)
and (1.8) in [6]. Solutions (1.7) and (1.9) in [6] are not real
= ( (1 + 𝑘𝑟2 − 𝑐 + √Δ) solutions of the KP-BBM equation, and it is easy to verify that
they do not satisfy the KP-BBM equation (3).
Abstract and Applied Analysis 5

y Γ6 y
Γ5

Γ11 Γ12

𝜑
𝜑 𝜑4 c1 𝜑1 c2 𝜑2 c3
c1 𝜑1 c2 𝜑2 c3 𝜑3

(a) 𝑏𝑐/𝑎 < 0 (b) 𝑏𝑐/𝑎 > 0

𝜑 𝜑
c1 𝜑1 c2 𝜑2 c3 c1 𝜑1 c2 𝜑2 c3

(c) (d)

Figure 3: The bifurcation phase portraits of system (6) with 𝑔 ≠0.

When 𝑏𝑐/𝑎 > 0, the corresponding homoclinic orbit Γ6 where 𝜑2 = (1+𝑘𝑟2 −𝑐−√Δ)/2𝑎 and 𝜑4 = (1+𝑘𝑟2 −𝑐+2√Δ)/2𝑎
has a double zero point 𝜑2 and a zero point 𝜑4 on 𝜑-axis (see if 𝑎 < 0 and 𝑏𝑐 < 0; then, completing (23) we get the solution
Figure 3(b)), so (15) can be rewritten as 𝑢3 (𝑥, 𝑦, 𝑡). In (23), 𝜑2 = (1 + 𝑘𝑟2 − 𝑐 + √Δ)/2𝑎 and 𝜑4 =
2𝑎 (1 + 𝑘𝑟2 − 𝑐 − 2√Δ)/2𝑎 if 𝑎 > 0 and 𝑏𝑐 > 0; then, completing
2
𝑦2 = (𝜑 − 𝜑2 ) (𝜑 − 𝜑4 ) ; (21) (23) we get the solution 𝑢4 (𝑥, 𝑦, 𝑡).
3𝑏𝑐
that is,
4. Periodic Wave Solutions
2𝑎 2
𝑦 = ±√ (𝜑 − 𝜑2 ) (𝜑 − 𝜑4 ). (22) So as to explain our work conveniently, in this section the
3𝑏𝑐 Jacobian elliptic function sn(𝑙, 𝑚) with modulus 𝑚 will be
Substituting (22) into d𝜑/d𝜉 = 𝑦 and integrating along expressed by sn𝑙. We discuss the periodic wave solutions
homoclinic orbits Γ6 , we get under conditions 𝑔 = 0 and 𝑔 ≠0, respectively.
𝜑
d𝑠 󵄨 󵄨
∫ = 󵄨󵄨󵄨𝜉󵄨󵄨󵄨 , (23) 4.1. The Case 𝑔 = 0. When (1 + 𝑘𝑟2 + 𝑐)/𝑏𝑐 < 0, system (6)
2
𝜑4 √(2𝑎/3𝑏𝑐) (𝑠 − 𝜑2 ) (𝑠 − 𝜑4 ) has periodic orbits Γ7 and Γ8 (see Figures 1(a) and 1(b)). Their
6 Abstract and Applied Analysis

expressions are (9) on 𝜑-𝑦 plane, where ℎ1 < ℎ < ℎ2 (or ℎ2 < where 𝑔1 = 2/√𝑟3 − 𝑟1 , sin 𝜓1 = √(𝑟3 − 𝜑)/(𝑟3 − 𝑟2 ), and
ℎ < ℎ1 ). Let 𝑚5 = √(𝑟3 − 𝑟2 )/(𝑟3 − 𝑟1 ). Solving (29), we get

2𝑎 3 1 + 𝑘𝑟2 − 𝑐 2 𝑎 (𝑟3 − 𝑟1 )
𝑓1 (𝜑) = 𝜑 − 𝜑 + 2ℎ; (24)
𝜑 = 𝑟3 − (𝑟3 − 𝑟2 ) sn2 √ − 𝜉. (30)
3𝑏𝑐 𝑏𝑐 6𝑏𝑐
then, we have the following results.
That is,
2
Claim 1. In the case of 𝑔 = 0, (1 + 𝑘𝑟 + 𝑐)/𝑏𝑐 < 0 and ℎ1 <
ℎ < ℎ2 (or ℎ2 < ℎ < ℎ1 ); then, the function 𝑓1 (𝜑) must have 𝑎 (𝑟3 − 𝑟1 )
𝑢5 (𝑥, 𝑦, 𝑡) = 𝑟3 − (𝑟3 − 𝑟2 ) sn2 √ − (𝑥 + 𝑟𝑦 − 𝑐𝑡) ,
three different real zero points. 6𝑏𝑐
(31)
Proof. Since 𝑓1 (𝜑) is a cubic polynomial about 𝜑, we can use
the Shengjin Theorem [11] to distinguish its solutions. We where the modulus of sn is 𝑚5 = √(𝑟3 − 𝑟2 )/(𝑟3 − 𝑟1 ).
only discuss the case 𝑏𝑐/𝑎 < 0, and the case 𝑏𝑐/𝑎 > 0 is the Similarly, when 𝑏𝑐/𝑎 > 0, the expression Γ8 is
same. Under the above conditions,
2𝑎
𝑦 = ±√ (𝜑 − 𝑟1 ) (𝜑 − 𝑟2 ) (𝜑 − 𝑟3 ), (𝑟1 ≤ 𝜑 ≤ 𝑟2 < 𝑟3 ) .
ℎ1 = 𝐻 (𝜑1 , 0) = 𝐻 (0, 0) = 0, 3𝑏𝑐
(32)
3
2
1 + 𝑘𝑟 − 𝑐 (1 + 𝑘𝑟2 − 𝑐)
ℎ2 = 𝐻 (𝜑2 , 0) = 𝐻 ( , 0) = < 0. Substituting (32) into d𝜑/d𝜉 = 𝑦 and integrating along orbit
𝑎 6𝑎2 𝑏𝑐 Γ8 , we have
(25)
𝜑
d𝑠
2 2 ∫
So (1 + 𝑘𝑟 − 𝑐)/(6𝑎 𝑏𝑐) < ℎ < 0. In 𝑓1 (𝜑) the coefficients 𝑟1 √(𝑟3 − 𝑠) (𝑟2 − 𝑠) (𝑠 − 𝑟1 )
2𝑎/3𝑏𝑐 < 0, (1 + 𝑘𝑟2 − 𝑐)/𝑏𝑐 < 0, By Shengjing Theorem [11], (33)
it follows that the function 𝑓1 (𝜑) has three different real zero
2𝑎 󵄨󵄨 󵄨󵄨
points. Let 𝐴 = ((1 + 𝑘𝑟2 − 𝑐)/𝑏𝑐)2 , 𝐵 = −9(2𝑎/3𝑏𝑐) ⋅ 2ℎ, and =√ 󵄨𝜉󵄨 , (𝑟1 < 𝜑 ≤ 𝑟2 < 𝑟3 ) .
3𝑏𝑐 󵄨 󵄨
𝐶 = 3((1 + 𝑘𝑟2 − 𝑐)/𝑏𝑐) ⋅ 2ℎ; then, 𝐵2 − 4𝐴𝐶 = 144(𝑎/𝑏𝑐)2 ℎ[ℎ −
((1 + 𝑘𝑟2 − 𝑐)3 /6𝑎2 𝑏𝑐)] < 0. The according periodic solution of Γ8 can be obtained as

Let 𝑟1 < 𝑟2 < 𝑟3 be three different real zero points of 𝑎 (𝑟3 − 𝑟1 )

𝑓1 (𝜑). Then Claim 1 means that (9) has three intersection 𝑢6 (𝑥, 𝑦, 𝑡) = 𝑟1 + (𝑟2 − 𝑟1 ) sn2 √ (𝑥 + 𝑟𝑦 − 𝑐𝑡) ,
points (𝑟1 , 0), (𝑟2 , 0), and (𝑟3 , 0) on 𝜑-axis. Therefore, (9) can 6𝑏𝑐
be rewritten as (34)

2𝑎 where the modulus of sn is 𝑚6 = √(𝑟2 − 𝑟1 )/(𝑟3 − 𝑟1 ).

𝑦2 = (𝜑 − 𝑟1 ) (𝜑 − 𝑟2 ) (𝜑 − 𝑟3 ) , (26)
3𝑏𝑐 When (1 + 𝑘𝑟2 + 𝑐)/𝑏𝑐 > 0, system (6) has periodic orbits
Γ9 and Γ10 (see Figure 2). Their expressions are (9), where ℎ1 <
where 𝑟1 < 0 < 𝑟2 < 𝜑 < 𝑟3 when 𝑏𝑐/𝑎 < 0 and 𝑟1 < 𝜑 < 𝑟2 < ℎ < ℎ2 (or ℎ2 < ℎ < ℎ1 ). Similarly, we can get the according
0 < 𝑟3 when 𝑏𝑐/𝑎 > 0. periodic solutions of Γ9 and Γ10 as 𝑢5 and 𝑢6 .
When 𝑏𝑐/𝑎 < 0, the orbit Γ7 is according to a periodic To verify validity of the periodic wave solutions, we take
solution of (6) and its expression is given by 𝑎 = 𝑏 = 𝑘 = 𝑟 = 1, 𝑐 = −1, and ℎ = −9/4 to make the
conditions in Claim 1 satisfied. By simple calculation, we get
2𝑎 that 𝑟1 = −1.09808, 𝑟2 = 1.5, and 𝑟3 = 4.09808. The specific
𝑦 = ±√ (𝜑 − 𝑟1 ) (𝜑 − 𝑟2 ) (𝜑 − 𝑟3 ), (𝑟1 < 𝑟2 ≤ 𝜑 ≤ 𝑟3 ) .
3𝑏𝑐 periodic wave solution is
(27)
5.19616
Substituting (27) into d𝜑/d𝜉 = 𝑦 and integrating along orbit 𝑢 (𝑥, 𝑦, 𝑡) = 4.090808 − 2.590808sn2 √ (𝑥 + 𝑦 + 𝑡) ,
6
Γ7 , we get (35)
𝑟3
d𝑠
∫ where the modulus of sn is √2/2.
𝜑 √(𝑟3 − 𝑠) (𝑠 − 𝑟1 ) (𝑠 − 𝑟2 )
(28) 4.2. The Case 𝑔 ≠0. System (6) has periodic orbits Γ11 and
2𝑎 󵄨󵄨 󵄨󵄨 Γ12 (see Figure 3). Their expressions are (7) on the 𝜑-𝑦 plane,
= √− 󵄨𝜉󵄨 , (𝑟1 < 𝑟2 ≤ 𝜑 < 𝑟3 ) .
3𝑏𝑐 󵄨 󵄨 where ℎ1 < ℎ < ℎ2 (or ℎ1 < ℎ < ℎ2 ). Let
By formula (236) in [12], we have 2𝑎 3 1 + 𝑘𝑟2 − 𝑐 2 2𝑔
𝑓2 (𝜑) = 𝜑 − 𝜑 + 𝜑 + 2ℎ; (36)
3𝑏𝑐 𝑏𝑐 𝑏𝑐
2𝑎 󵄨󵄨 󵄨󵄨
𝑔1 sn−1 (sin 𝜓1 , 𝑚5 ) = √ − 󵄨 𝜉󵄨 , (29)
3𝑏𝑐 󵄨 󵄨 then, we have the following results about 𝑓2 (𝜑).
Abstract and Applied Analysis 7

Claim 2. If 𝑔 ≠0 and ℎ1 < ℎ < ℎ2 (or ℎ2 < ℎ < ℎ1 ), then the For example, we take 𝑎 = 𝑏 = 𝑘 = 𝑟 = 1, 𝑐 = −1, 𝑔 = 2,
function 𝑓2 (𝜑) must have three different real zero points. and ℎ = 3/2 such that the conditions in Claim 2 are satisfied.
By simple calculation, we get that 𝑐1 = 0.663975, 𝑐2 = 1.5, and
Proof. We only prove the case 𝑏𝑐/𝑎 < 0, and the case 𝑏𝑐/𝑎 > 0 𝑐3 = 2.36603. The according periodic wave solution is
is the same. Under the above conditions,
1.702055
𝑎 3 1 + 𝑘𝑟2 − 𝑐 2 𝑔 𝑢 (𝑥, 𝑦, 𝑡) = 2.36603 − 0.86603sn2 √ (𝑥 + 𝑦 + 𝑡) ,
ℎ1 = 𝐻 (𝜑1 , 0) = − 𝜑 + 𝜑1 − 𝜑1 6
3𝑏𝑐 1 2𝑏𝑐 𝑏𝑐 (42)
1
= − 𝑓2 (𝜑1 ) + ℎ, where the modulus of sn is √0.86603/1.702055.
2
(37)
𝑎 3 1 + 𝑘𝑟2 − 𝑐 2 𝑔 5. Relationship between Bell-Shaped Waves
ℎ2 = 𝐻 (𝜑2 , 0) = − 𝜑 + 𝜑2 − 𝜑2
3𝑏𝑐 2 2𝑏𝑐 𝑏𝑐 and Periodic Waves
1
= − 𝑓2 (𝜑2 ) + ℎ. In Sections 3 and 4, the bell-shaped solitary wave and periodic
2 wave solutions are obtained. Via further study, we find that
So 𝑓2 (𝜑1 ) ⋅ 𝑓2 (𝜑2 ) = 4(ℎ − ℎ1 )(ℎ − ℎ2 ) < 0. For 𝑓2 (𝜑), we have there exists an interesting relationship between these two
𝑓2 (−∞) > 0, 𝑓2 (𝜑1 ) < 0, 𝑓2 (𝜑2 ) > 0, and 𝑓2 (+∞) < 0. Again, kinds of solutions; that is, the bell-shaped solutions are limits
𝑓2󸀠 (𝜑) = (2𝑎/𝑏𝑐)(𝜑 − 𝜑1 )(𝜑 − 𝜑2 ), which is monotonous in of the periodic waves in some sense. The results are detailed
the intervals (−∞, 𝜑1 ), (𝜑1 , 𝜑2 ), and (𝜑2 , +∞). By zero point as follows.
theorem of continuous function, there must be one real zero
point of 𝑓2 (𝜑) that lies in each of the three intervals, proving Proposition 4. Let 𝑢𝑖 (𝑖 = 1, 2, . . . , 8) be solutions of (3),
the claim. let 𝑘, 𝑟, 𝑎, 𝑏, 𝑐, and 𝑔 be parameters in (5), and let 𝑚𝑖 (𝑖 =
5, 6, 7, 8) be modulus of the Jacobian elliptic function sn; then,
Let 𝑐1 < 𝑐2 < 𝑐3 be three different real zero points of 𝑓2 (𝜑). one has the following.
Then Claim 2 means that (7) has three intersection points on
𝜑-axis denoted by (𝑐1 , 0), (𝑐2 , 0), and (𝑐3 , 0). Then (7) can be Case 1. When 𝑔 = 0 and (1 + 𝑘𝑟2 − 𝑐)/𝑏𝑐 < 0, for modulus
rewritten as 𝑚𝑖 → 1 (𝑖 = 5, 6), the periodic waves 𝑢5 and 𝑢6 degenerate
bell-shaped wave 𝑢1 .
2𝑎
𝑦2 = (𝜑 − 𝑐1 ) (𝜑 − 𝑐2 ) (𝜑 − 𝑐3 ) , (38)
3𝑏𝑐 Case 2. When 𝑔 = 0 and (1 + 𝑘𝑟2 − 𝑐)/𝑏𝑐 > 0, for modulus
where 𝑐1 < 𝜑1 < 𝑐2 < 𝜑2 < 𝑐3 . 𝑚𝑖 → 1 (𝑖 = 5, 6), the periodic waves 𝑢5 and 𝑢6 degenerate
When 𝑏𝑐/𝑎 < 0, the expression of periodic orbit Γ11 is bell-shaped wave 𝑢2 .

Case 3. When 𝑔 ≠0 and 𝑏𝑐/𝑎 < 0, for modulus 𝑚7 → 1, the

2𝑎
𝑦 = ±√ (𝜑 − 𝑐1 ) (𝜑 − 𝑐2 ) (𝜑 − 𝑐3 ), (𝑐1 < 𝑐2 ≤ 𝜑 ≤ 𝑐3 ) . periodic wave 𝑢7 degenerates bell-shaped wave 𝑢3 .
3𝑏𝑐
(39) Case 4. When 𝑔 ≠0 and 𝑏𝑐/𝑎 > 0, for modulus 𝑚8 → 1, the
periodic wave 𝑢8 degenerates bell-shaped wave 𝑢4 .
When 𝑏𝑐/𝑎 > 0, the expression of periodic orbit Γ12 is
Here, we only prove Cases 1 and 3 for simplicity. The
2𝑎
𝑦 = ±√ (𝜑 − 𝑐1 ) (𝜑 − 𝑐2 ) (𝜑 − 𝑐3 ), (𝑐1 ≤ 𝜑 ≤ 𝑐2 < 𝑐3 ) . remaining cases are the same. In the following proofs, we use
3𝑏𝑐 the property of elliptic function that sn → tanh when the
(40) modulus 𝑚 → 1 [5, 13].
Substituting (39) and (40) into d𝜑/d𝜉 = 𝑦 and integrating Proof of Case 1. When 𝑚5 = √(𝑟3 − 𝑟2 )/(𝑟3 − 𝑟1 ) → 1, it
along orbits Γ11 and Γ12 , respectively, it is the same to the means 𝑟1 = 𝑟2 and sn = tanh; then, we calculate
proceeding for solving 𝑢5 and 𝑢6 and we can get the according
periodic solutions of Γ11 and Γ12 as follows: 3 (1 + 𝑘𝑟2 − 𝑐)
𝑟1 = 𝑟2 = 0, 𝑟3 = . (43)
2𝑎
𝑎 (𝑐3 − 𝑐1 )
𝑢7 (𝑥, 𝑦, 𝑡) = 𝑐3 − (𝑐3 − 𝑐2 ) sn2 √ − (𝑥 + 𝑟𝑦 − 𝑐𝑡) , Substituting 𝑟𝑖 (𝑖 = 1, 2, 3) into 𝑢5 admits to 𝑢1 as follows:
6𝑏𝑐
𝑢5 (𝑥, 𝑦, 𝑡)
2 √ 𝑎 (𝑐3 − 𝑐1 )
𝑢8 (𝑥, 𝑦, 𝑡) = 𝑐1 + (𝑐2 − 𝑐1 ) sn (𝑥 + 𝑟𝑦 − 𝑐𝑡) ,
6𝑏𝑐 𝑎 (𝑟3 − 𝑟1 )
(41) = 𝑟3 − (𝑟3 − 𝑟2 ) sn2 √ − (𝑥 + 𝑟𝑦 − 𝑐𝑡)
6𝑏𝑐
where the moduli for sn are 𝑚7 = √(𝑐3 − 𝑐2 )/(𝑐3 − 𝑐1 ) and 3 (1 + 𝑘𝑟2 − 𝑐) 3 (1 + 𝑘𝑟2 − 𝑐)
𝑚8 = √(𝑐2 − 𝑐1 )/(𝑐3 − 𝑐1 ) in (41). = −
2𝑎 2𝑎
8 Abstract and Applied Analysis

(1 + 𝑘𝑟2 − 𝑐) 1 + 𝑘𝑟2 − 𝑐 − √Δ
2√ =
× tanh − (𝑥 + 𝑟𝑦 − 𝑐𝑡) 2𝑎
4𝑏𝑐
3√Δ
3 (1 + 𝑘𝑟2 − 𝑐) +
= 𝑎 [1 + cosh √−√Δ/𝑏𝑐 (𝑥 + 𝑟𝑦 − 𝑐𝑡)]
2𝑎

(1 + 𝑘𝑟2 − 𝑐)
× [1 − tanh2 √ − (𝑥 + 𝑟𝑦 − 𝑐𝑡)] = ( (1 + 𝑘𝑟2 − 𝑐 − √Δ)
4𝑏𝑐
[ ]
3 (1 + 𝑘𝑟2 − 𝑐)
= √Δ
2𝑎 × cosh √ − (𝑥 + 𝑟𝑦 − 𝑐𝑡) + 1
𝑏𝑐
1
×
2
cosh √− ((1 + 𝑘𝑟2 − 𝑐) /4𝑏𝑐) (𝑥 + 𝑟𝑦 − 𝑐𝑡) + 𝑘𝑟2 − 𝑐 + 5√Δ)

3 (1 + 𝑘𝑟2 − 𝑐)
= −1
2𝑎 √Δ
× (2𝑎 [cosh √ − (𝑥 + 𝑟𝑦 − 𝑐𝑡) + 1])
1 𝑏𝑐
× [ ]
(1/2) [cosh √− ((1 + 𝑘𝑟2 − 𝑐) /𝑏𝑐) (𝑥 + 𝑟𝑦 − 𝑐𝑡) + 1]
= 𝑢3 (𝑥, 𝑦, 𝑡) .
(45)
3 (1 + 𝑘𝑟2 − 𝑐)
=
𝑎 [1 + cosh √− ((1 + 𝑘𝑟2 − 𝑐) /𝑏𝑐) (𝑥 + 𝑟𝑦 − 𝑐𝑡)] The results provide a manner that we can get bell-shaped
waves from periodic waves for some nonlinear development
= 𝑢1 (𝑥, 𝑦, 𝑡) . equations.
(44)
Conflict of Interests
When 𝑚6 = √(𝑟2 − 𝑟1 )/(𝑟3 − 𝑟1 ) → 1, it means 𝑟2 = 𝑟3 ;
The author declares that there is no conflict of interests
then, we calculate 𝑟2 = 𝑟3 = 0 and 𝑟1 = 3(1 + 𝑘𝑟2 − 𝑐)/2𝑎, and regarding the publication of this paper.
substituting 𝑟𝑖 (𝑖 = 1, 2, 3) into 𝑢6 we get 𝑢6 = 𝑢1 .

Acknowledgments
Proof of Case 3. When 𝑚7 = √(𝑐3 − 𝑐2 )/(𝑐3 − 𝑐1 ) → 1, it This work was supported by the National Natural Science
means 𝑐1 = 𝑐2 and sn = tanh; then, we calculate 𝑐1 = 𝑐2 = Foundation of China (no. 11201070) and Guangdong Province
(1 + 𝑘𝑟2 − 𝑐 − √Δ)/2𝑎 and 𝑐3 = (1 + 𝑘𝑟2 − 𝑐 + 2√Δ)/2𝑎, and (no. 2013KJCX0189 and no. Yq2013161). The author would
substituting 𝑐𝑖 (𝑖 = 1, 2, 3) into 𝑢7 admits to 𝑢3 as follows: like to thank the editors for their hard working and the
anonymous reviewers for helpful comments and suggestions.
𝑢7 (𝑥, 𝑦, 𝑡)
References
2√ 𝑎 (𝑐 − 𝑐 )
= 𝑐3 − (𝑐3 − 𝑐2 ) sn − 3 1 (𝑥 + 𝑟𝑦 − 𝑐𝑡) [1] T. B. Benjamin, J. L. Bona, and J. J. Mahony, “Model equations
6𝑏𝑐
for long waves in nonlinear dispersive systems,” Philosophical
Transactions of the Royal Society of London A. Mathematical and
1 + 𝑘𝑟2 − 𝑐 + 2√Δ 3√Δ −√ Δ
= − tanh2 √ (𝑥 + 𝑟𝑦 − 𝑐𝑡) Physical Sciences, vol. 272, no. 1220, pp. 47–78, 1972.
2𝑎 2𝑎 4𝑏𝑐 [2] B. B. Kadomtsev and V. I. Petviashvili, “On the stability of
solitary waves in weakly dispersive media,” Soviet Physics—
Doklady, vol. 15, pp. 539–541, 1970.
= (1 + 𝑘𝑟2 − 𝑐 − √Δ
[3] A.-M. Wazwaz, “Exact solutions of compact and noncompact
structures for the KP-BBM equation,” Applied Mathematics and
Computation, vol. 169, no. 1, pp. 700–712, 2005.
−√Δ
+3√Δ (1 − tanh2 √ (𝑥 + 𝑟𝑦 − 𝑐𝑡))) [4] A.-M. Wazwaz, “The extended tanh method for new compact
4𝑏𝑐 and noncompact solutions for the KP-BBM and the ZK-BBM
equations,” Chaos, Solitons and Fractals, vol. 38, no. 5, pp. 1505–
× (2𝑎)−1 1516, 2008.
Abstract and Applied Analysis 9

[5] M. A. Abdou, “Exact periodic wave solutions to some nonlinear

evolution equations,” International Journal of Nonlinear Science,
vol. 6, no. 2, pp. 145–153, 2008.
[6] M. Song, C. X. Yang, and B. G. Zhang, “Exact solitary wave solu-
tions of the Kadomtsov-Petviashvili-Benjamin-Bona-Mahony
equation,” Applied Mathematics and Computation, vol. 217, no.
4, pp. 1334–1339, 2010.
[7] S.-N. Chow and J. K. Hale, Method of Bifurcation Theory,
Springer, Berlin, Germany, 1982.
[8] J. B. Li and Z. R. Liu, “Smooth and non-smooth traveling waves
in a nonlinearly dispersive equation,” Applied Mathematical
Modelling, vol. 25, no. 1, pp. 41–56, 2000.
[9] J. B. Li and Z. R. Liu, “Traveling wave solutions for a class of
nonlinear dispersive equations,” Chinese Annals of Mathematics
B, vol. 23, no. 3, pp. 397–418, 2002.
[10] Z. R. Liu and Z. Y. Ouyang, “A note on solitary waves for
modified forms of Camassa-Holm and Degasperis-Procesi
equations,” Physics Letters A, vol. 366, no. 4-5, pp. 377–381, 2007.
[11] F. Shengjin, “A new extracting formula and a new distinguishing
means on the one variable cubic equation,” Natural Science
Journal of Hainan Teacheres College, vol. 2, pp. 91–98, 1989.
[12] P. F. Byrd and M. D. Friedman, Handbook of Elliptic Integrals for
Engineers and Scientists, Springer, Berlin, Germany, 1971.
[13] S. Yu and L. Tian, “Nonsymmetrical kink solution of gener-
alized KdV equation with variable coefficients,” International
Journal of Nonlinear Science, vol. 5, no. 1, pp. 71–78, 2008.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 247359, 11 pages
http://dx.doi.org/10.1155/2014/247359

Research Article
Dynamical Behaviors of Rumor Spreading Model with
Control Measures

Xia-Xia Zhao1,2 and Jian-Zhong Wang1,2

1
National Key Laboratory for Electronic Measurement Technology, North University of China, Taiyuan, Shanxi 030051, China
2
Key Laboratory of Instrumentation Science and Dynamic Measurement, North University of China, Ministry of Education,
Taiyuan, Shanxi 030051, China

Correspondence should be addressed to Xia-Xia Zhao; xiaxiazhao12@163.com

Received 16 April 2014; Accepted 12 May 2014; Published 1 June 2014

Academic Editor: Sanling Yuan

Copyright © 2014 X.-X. Zhao and J.-Z. Wang. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.

Rumor has no basis in fact and flies around. And in general, it is propagated for a certain motivation, either for business, economy,
or pleasure. It is found that the web does expose us to more rumor and increase the speed of the rumors spread. Corresponding
to these new ways of spreading, the government should carry out some measures, such as issuing message by media, punishing
the principal spreader, and enhancing management of the internet. In order to assess these measures, dynamical models without
and with control measures are established. Firstly, for two models, equilibria and the basic reproduction number of models are
discussed. More importantly, numerical simulation is implemented to assess control measures of rumor spread between individuals-
to-individuals and medium-to-individuals. Finally, it is found that the amount of message released by government has the greatest
influence on the rumor spread. The reliability of government and the cognizance ability of the public are more important. Besides
that, monitoring the internet to prevent the spread of rumor is more important than deleting messages in media which already
existed. Moreover, when the minority of people are punished, the control effect is obvious.

1. Introduction been applied to investigate the spread of rumors [9–13]. The

ignorant, the spreader, and the stifler are equivalent to the
As a typical social phenomenon, rumor has no basis in fact susceptible, the infected, and the recovered. Some models are
and flies around, especially when major public events happen established based on network [14–19]. Some are built on the
and people do not have exact information and knowledge
basis of the random theory [20–23].
about the events; the rumor is dispersed by some people
for achieving the specific purpose. It has been described With rapid development of today’s society, besides prop-
in detail by some pieces of literature [1–6]. In the modern agation by word of mouth, rumors also can be spread by
society, the rumor not only has not disappeared but also, with public homepage, SMS, e-mails, or blogging that provide
the development of the communication transmission modes, faster velocity of transmission [9, 24, 25]. The new type of
such as internet, telephone, and advanced information tech- transmission mode has been studied dynamically by [26]. It
nology, spreads more quickly and the scope involved is much established an 𝐼𝑆𝑅𝑊 dynamical system including spreading
broader. Thus, the internet rumors become an important between individuals and medium-to-individuals to describe
factor that influences the current social harmony and stability the actual pattern of transmission. With regard to the internet
in emergencies and all kinds of crisis, and it is becoming the rumor, the government should share real information in a
focus of the netizens and governments at all levels. timely manner with the public to avoid the public hazard [27].
The classical models to study the spread of rumor were In 1953, the formula that describes the generation of rumor
given by Daley and Kendall and Maki and Thomson [7, 8]. was proposed by Cross. 𝑅 = 𝐼×𝑎/𝑐, where 𝐼 is the importance
Since the dissemination process of rumor is similar to the of events, 𝑎 is ambiguity of events, and 𝑐 is the critical ability
spreading of infectious disease, epidemic models have usually of the public. There are some models to assess the control
2 Abstract and Applied Analysis

Table 1: Description of parameters in the system (1).

Parameters Value Unit Comments
𝐴 16000000/365 Day−1 The birth rate of human
𝛽 50 × 0.005/1350000000 Day−1 S(t)-to-I(t) transmission rate
𝛼 10−10 Day−1 W(t)-to-I(t) transmission rate
𝑑 1/60/365 Day−1 The natural mortality rate of human
𝜃 0.5 None The rate of being 𝑆(𝑡) after transmission
𝑚 0.0001 Day−1 The transformation rate of 𝑆(𝑡) into 𝑅(𝑡)
𝑘 1 Day−1 The submerged rate of message
𝜆 300 × 0.00001 Day−1 The disseminating quantity of messages per spreader
𝜉1 50 × 0.0001/1350000000 Day−1 The transformation rate to 𝑅(𝑡)
𝜉2 50 × 0.0001/1350000000 Day−1 R(t)-to-S(t) transmission rate
𝐼(0) 1350000000−1000000 Individual The initial number of the susceptible individuals
𝑆(0) 1000000 Individual The initial number of the spreaders
𝑅(0) 0 Individual The initial number of the stiflers
𝑊(0) 0 Individual The initial quantity of messages by spreader

measures [28, 29]. In order to control the rumor spreading, completely ignorant population [33–35], can be calculated as
we can focus on the credibility of the authorities’ media follows:
[30–32] and increasing the cognizance ability of the public.
Besides, the government should give a certain punishment for 𝜃𝛽𝐼0 𝜆𝜃𝛼𝐼0
𝑅0 = + . (2)
the spreader. Therefore, this paper mainly assesses the effect 𝑚 + 𝑑 (𝑚 + 𝑑) 𝑘
of these measures.
The detailed calculation method can be seen in [35] and
the behavior of (1) is discussed in Theorem 1.
2. A Dynamical System for Rumor Spreading
Theorem 1. (a) When 𝑅0 < 1, the disease-free equilibrium 𝐸0
Without consideration of government measures, the dynami-
is globally asymptotically stable.
cal system we establish will include the following four classes:
the susceptible individual (𝐼), the spreader (𝑆), the stifler (𝑅), (b) When 𝑅0 > 1, the disease-free equilibrium 𝐸0 is
and the message in media (𝑊). Here, the bilinear incidence unstable.
rate is considered. The interpretation of parameters can be
seen in Table 1. The model we employ is as follows: Proof. (a) Define a Lyapunov function

𝑑𝐼 (𝑡) 𝐿 = 𝑘𝑆 + 𝜃𝛼𝐼0 𝑊 ≥ 0. (3)

= 𝐴 − 𝛽𝐼 (𝑡) 𝑆 (𝑡) − 𝛼𝐼 (𝑡) 𝑊 (𝑡) − 𝑑𝐼 (𝑡) ,
𝑑𝑡
When 𝑅0 < 1, the Lyapunov function satisfies
𝑑𝑆 (𝑡)
= 𝜃𝛽𝐼 (𝑡) 𝑆 (𝑡) + 𝜃𝛼𝐼 (𝑡) 𝑊 (𝑡) − 2𝜉1 𝑆2 (𝑡)
𝑑𝑡 𝐿̇= 𝑘 [𝜃𝛽𝐼 (𝑡) 𝑆 (𝑡) + 𝜃𝛼𝐼 (𝑡) 𝑊 (𝑡) − 2𝜉1 𝑆2 (𝑡) − 𝑚𝑆 (𝑡)
− 𝑚𝑆 (𝑡) − 𝜉2 𝑆 (𝑡) 𝑅 (𝑡) − 𝑑𝑆 (𝑡) ,
− 𝜉2 𝑆 (𝑡) 𝑅 (𝑡) − 𝑑𝑆 (𝑡) ]
𝑑𝑅 (𝑡)
= (1 − 𝜃) 𝛽𝐼 (𝑡) 𝑆 (𝑡) + (1 − 𝜃) 𝛼𝐼 (𝑡) 𝑊 (𝑡) + 2𝜉1 𝑆2 (𝑡) + 𝜃𝛼𝐼0 [𝜆𝑆 (𝑡) − 𝑘𝑊 (𝑡)]
𝑑𝑡
+ 𝑚𝑆 (𝑡) + 𝜉2 𝑆 (𝑡) 𝑅 (𝑡) − 𝑑𝑅 (𝑡) , ≤ 𝑘𝜃𝛽𝐼0 𝑆 + 𝜃𝛼𝜆𝐼0 𝑆 − 𝑘 (𝑚 + 𝑑) 𝑆
𝑑𝑊 (𝑡)
= 𝜆𝑆 (𝑡) − 𝑘𝑊 (𝑡) . 𝜃𝛽𝐼0 𝜆𝜃𝛼𝐼0
𝑑𝑡 = 𝑘𝑆 (𝑚 + 𝑑) [ + − 1]
𝑚 + 𝑑 (𝑚 + 𝑑) 𝑘
(1)
= 𝑘𝑆 (𝑚 + 𝑑) [𝑅0 − 1]
3. Dynamical Behaviors of System (1) ≤ 0.
It is easy to know that 𝑑𝑁(𝑡)/𝑑𝑡 = 𝐴 − 𝑑𝑁. So, the positive (4)
invariant set is Γ = {(𝐼(𝑡), 𝑆(𝑡), 𝑅(𝑡), 𝑊(𝑡), 𝐺(𝑡)) | 𝐼(𝑡) ≥
0, 𝑆(𝑡) ≥ 0, 𝑅(𝑡) ≥ 0, 𝑊(𝑡) ≥ 0, 𝐺(𝑡) ≥ 0, 0 ≤ 𝐼(𝑡) + 𝑆(𝑡) + It is easy to know that 𝐿̇= 0 only hold when 𝑆 = 0. As
𝑅(𝑡) ≤ 𝐴/𝑑}. The disease-free equilibrium is 𝐸0 = (𝐼0 , 0, 0, 0), a result, the disease-free equilibrium point 𝐸0 is the only
where 𝐼0 = 𝐴/𝑑. fixed point of the system. By applying the Lyapunov-LaSalle
The basic reproduction number, that is, the expected asymptotic stability theorem [36, 37], the disease-free
number of secondary spreaders produced by a spreader in a quilibrium point 𝐸0 is globally asymptotically stable.
Abstract and Applied Analysis 3

4 5
3.5 4.5

3 4
3.5
2.5
3

R01
R01

2
2.5
1.5
2
1 1.5
0.5 1
0 0.5
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
𝜃 𝜂

(a) (b)

2 12

10
1.5
8

R01
6
R01

4
0.5
2

0 0
0 0.05 0.1 0.15 0.2 0.25 0.3 0 2 4 6 8 10
𝜇 𝜆
(c) (d)
12 1.89

10 1.885

1.88
8
1.875
R01

6
R01

1.87
4
1.865
2 1.86
0 1.855
0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
B ×107 k
(e) (f)
4 2
3.5 1.8

3 1.6
1.4
2.5
1.2
R01

2
R01

1
1.5
0.8
1 0.6
0.5 0.4
0 0.2
0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
p 𝛽 ×10−10

(g) (h)

Figure 1: Continued.
4 Abstract and Applied Analysis

1.865 9

1.864 8
7
1.863
6
1.862 5

R01
R01

1.861 4
3
1.86
2
1.859 1
1.858 0
0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
−10
𝛼 ×10 𝛾 ×10−10
(i) (j)

Figure 1: The sensitivity of 𝑅01 in terms of parameters.

(b) The Jacobian matrix at the 𝐸0 is the Jacobian The analysis about 𝑥1 , 𝑥2 , and 𝑥3 is more complex and we
0 0 list the result in Table 2.
−𝑑 −𝛽𝐼 0 −𝛼𝐼
0 𝜃𝛽𝐼0 − (𝑚 + 𝑑) 0 𝜃𝛼𝐼0
𝐽|𝐸0 =( ). 4. A Dynamical System for Rumor Spreading
0 (1 − 𝜃) 𝛽𝐼0 + (𝑚 + 𝑑) −𝑑 (1 − 𝜃) 𝛼𝐼0
0 𝜆 0 −𝑘 with Government Measures
(5)
Now, we add the measures of government to the system,
With regard to this matrix, the eigenvalues are the roots especially issuing the actual message through the medium
of the polynomial equation 𝐺(𝑡) and punishment for the spreaders, which are reflected
in 𝜂 and 𝜇. Moreover, the ability of cognizance of the public
[𝑥2 + (𝑚 + 𝑑 + 𝑘 − 𝜃𝛽𝐼0 ) 𝑥 + (𝑚 + 𝑑 − 𝜃𝛽𝐼0 ) 𝑘 − 𝜆𝜃𝛼𝐼0 ] is reflected in 𝜃. The higher the cognizance ability the smaller
the 𝜃. These interpretations can be seen in Table 2. 𝐵(𝑆) can
× (𝑥 + 𝑑) (𝑥 + 𝑘) = 0.
be adopted as different term according to different situation.
(6) The system has the following form:
It is easy to know that −𝑑 and −𝑘 are two of the eigenvalues. 𝑑𝐼 (𝑡)
When 𝑅0 > 1,(𝑚 + 𝑑 − 𝜃𝛽𝐼0 )𝑘 − 𝜆𝜃𝛼𝐼0 < 0; that is, there must = 𝐴 − 𝛽𝐼 (𝑡) 𝑆 (𝑡) − 𝛼𝐼 (𝑡) 𝑊 (𝑡) − 𝜂𝛾𝐼 (𝑡) 𝐺 (𝑡)
𝑑𝑡
exist a positive root. That means that 𝐸0 is unstable.
− 𝑑𝐼 (𝑡) ,
With regard to the positive equilibrium 𝐸∗ = (𝐼∗ , 𝑆∗ , 𝑅∗ ,
∗
𝑊 ), it should satisfy 𝑑𝑆 (𝑡)
= 𝜃𝛽𝐼 (𝑡) 𝑆 (𝑡) + 𝜃𝛼𝐼 (𝑡) 𝑊 (𝑡) − 2𝜉1 𝑆2 (𝑡) − 𝑚𝑆 (𝑡)
𝑑𝑡
𝐴 − 𝛽𝐼∗ 𝑆∗ − 𝛼𝐼∗ 𝑊∗ − 𝑑𝐼∗ = 0,
− 𝜉2 𝑆 (𝑡) 𝑅 (𝑡) − 𝜂𝛿𝑆 (𝑡) 𝐺 (𝑡) − 𝑑𝑆 (𝑡) − 𝜇𝑆 (𝑡) ,
𝜃𝛽𝐼∗ 𝑆∗ + 𝜃𝛼𝐼∗ 𝑊∗ − 2𝜉1 𝑆∗2 − 𝑚𝑆∗ − 𝜉2 𝑆∗ 𝑅∗ − 𝑑𝑆∗ = 0,

(1 − 𝜃) 𝛽𝐼∗ 𝑆∗ + (1 − 𝜃) 𝛼𝐼∗ 𝑊∗ + 2𝜉1 𝑆∗2 + 𝑚𝑆∗ 𝑑𝑅 (𝑡)

= (1 − 𝜃) 𝛽𝐼 (𝑡) 𝑆 (𝑡) + (1 − 𝜃) 𝛼𝐼 (𝑡) 𝑊 (𝑡) + 2𝜉1 𝑆2 (𝑡)
𝑑𝑡
+ 𝜉2 𝑆∗ 𝑅∗ − 𝑑𝑅∗ = 0,
+ 𝑚𝑆 (𝑡) + 𝜇𝑆 (𝑡) + 𝜉2 𝑆 (𝑡) 𝑅 (𝑡) + 𝜂𝛿𝑆 (𝑡) 𝐺 (𝑡)
𝜆𝑆∗ − 𝑘𝑊∗ = 0.
+ 𝜂𝛾𝐼 (𝑡) 𝐺 (𝑡) − 𝑑𝑅 (𝑡) ,
(7)
𝑑𝑊 (𝑡)
By calculating the equations, we have = 𝜆𝑆 (𝑡) − 𝑘𝑊 (𝑡) ,
𝑑𝑡
𝑥1 𝑆2 + 𝑥2 𝑆 + 𝑥3 = 0, (8) 𝑑𝐺 (𝑡)
= 𝐵 (𝑆) − 𝑝𝐺 (𝑡) .
where 𝑑𝑡
𝑥1 = (𝜉2 − 2𝜉1 ) (𝑘𝛽 + 𝛼𝜆) , (10)

𝐴 5. Dynamical Behavior of System (10)

𝑥2 = (𝜉2 − 2𝜉1 ) 𝑘𝑑 − (𝑚 + 𝑑 + 𝜉2 ) (𝑘𝛽 + 𝛼𝜆) , (9)
𝑑
What this paper mainly discusses is the effect of measures
𝑥3 = 𝜃𝛽𝐴𝑘 + 𝜃𝛼𝜆𝐴 − (𝑚 + 𝑑) 𝑘𝑑 = 𝑘 (𝑚 + 𝑑) (𝑅0 − 1) . carried out by authority. At first, we can assume that the
Abstract and Applied Analysis 5

×108 ×107
3.5 8

3 7
The final scale of the spreader

The final scale of the spreader

2.5 6
5
2
4
1.5
3
1
2
0.5 1
0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2
t ×104 t ×104

𝜃 = 0.2 𝜃 = 0.8 𝜂 = 0.4 𝜂 = 0.8

𝜃 = 0.4 𝜃=1 𝜂 = 0.6 𝜂=1
𝜃 = 0.6
(a) (b)
7
×107 ×10
4 14
The final scale of the spreader
3.5 12
The final scale of the spreader

3 10
2.5
8
2
6
1.5
4
1
0.5 2

0 0
0 2000 4000 6000 8000 10000 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
t t ×104

𝜇 = 0.0002 𝜇 = 0.0008 B = 14000000 B = 8000000

𝜇 = 0.0004 𝜇 = 0.001 B = 12000000 B = 6000000
𝜇 = 0.0006 B = 10000000
(c) (d)

×107 ×107
4 4
The final scale of the spreader

3.5 3.5
The final scale of the spreader

3 3
2.5 2.5
2 2
1.5 1.5
1 1
0.5 0.5
0 0
0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000
t t
𝛽 = 50 × 0.0002/(1.35 × 109 ) 𝛽 = 50 × 0.0008/(1.35 × 109 ) 𝛼 = 2 × 10−11 𝛼 = 8 × 10−11
𝛽 = 50 × 0.0004/(1.35 × 109 ) 𝛽 = 50 × 0.001/(1.35 × 109 ) 𝛼 = 4 × 10−11 𝛼 = 10 × 10−11
𝛽 = 50 × 0.0006/(1.35 × 109 ) 𝛼 = 6 × 10−11
(e) (f)

Figure 2: Continued.
6 Abstract and Applied Analysis

×108 ×107
2.5 4
3.5
The final scale of the spreader

The final scale of the spreader

2 3
2.5
1.5
2
1 1.5
1
0.5
0.5

0 0
0 1 2 3 4 5 0 2000 4000 6000 8000 10000 12000
t ×10 4 t

𝛾 = 4 × 10−11
𝛾 = 8 × 10 −11
𝜉1 = 50 × 0.0002/(1.35 × 109 ) 𝜉1 = 50 × 0.0008/(1.35 × 109 )
𝛾 = 6 × 10−11 𝛾 = 10 × 10−11 𝜉1 = 50 × 0.0004/(1.35 × 109 ) 𝜉1 = 50 × 0.001/(1.35 × 109 )
𝜉1 = 50 × 0.0006/(1.35 × 109 )
(g) (h)
7
×10
2
The final scale of the spreader

1.5

0.5

0
0 1000 2000 3000 4000 5000 6000 7000 8000
t

𝜉2 = 50 × 0.0002/(1.35 × 109 ) 𝜉2 = 50 × 0.0008/(1.35 × 109 )

𝜉2 = 50 × 0.0004/(1.35 × 109 ) 𝜉2 = 50 × 0.001/(1.35 × 109 )
𝜉2 = 50 × 0.0006/(1.35 × 109 )

(i)

Figure 2: The sensitivity of the final scale in terms of parameters. (a) 𝜃. (b) 𝜂. (c) 𝜇. (d) 𝐵. (e) 𝛽. (f) 𝛼. (g) 𝛾. (h) 𝜉1 . (i) 𝜉2 .

Table 2: Equilibria and stability of system (1).

Cases Conditions Positive equilibria

𝜉2 < 2𝜉1 𝑥1 < 0, 𝑥2 < 0, 𝑥3 < 0 No positive equilibrium
𝑅0 < 1 𝜉2 = 2𝜉1 𝑥1 = 0, 𝑥2 < 0, 𝑥3 < 0 No positive equilibrium
𝜉2 > 2𝜉1 𝑥1 > 0, 𝑥3 < 0 Unique positive equilibrium 𝐸∗
𝜉2 < 2𝜉1 𝑥1 < 0, 𝑥3 > 0 Unique positive equilibrium 𝐸∗
𝜉2 = 2𝜉1 𝑥1 = 0, 𝑥2 < 0, 𝑥3 > 0 Unique positive equilibrium
𝑥2 ≥ 0, 𝑥1 > 0, 𝑥3 > 0 No positive equilibrium
𝑅0 > 1 𝑥2 < 0, 𝑥1 > 0, 𝑥3 > 0, Δ < 0 No positive equilibrium
𝜉2 > 2𝜉1
𝑥2 < 0, 𝑥1 > 0, 𝑥3 > 0, Δ = 0 Unique positive equilibrium
𝑥2 < 0, 𝑥1 > 0, 𝑥3 > 0, Δ > 0 Two positive equilibria
Abstract and Applied Analysis 7

×106
×106
1.018 1.008
e final scale of the spreader

1.016 1.007

e final scale of the spreader

1.014 1.006
1.012
1.005
1.01
1.004
1.008
1.006 1.003

1.004 1.002
1.002 1.001
1 1
0 2000 4000 6000 8000 10000 0 2000 4000 6000 8000 10000
t t

𝜃 = 0.2 𝜃 = 0.8 𝜂 = 0.4 𝜂 = 0.8

𝜃 = 0.4 𝜃=1 𝜂 = 0.6 𝜂=1
𝜃 = 0.6
(a) (b)
6
×106 ×10
1.007 e final scale of the spreader 1.008

1.006 1.007
e final scale of the spreader

1.006
1.005
1.005
1.004
1.004
1.003
1.003
1.002
1.002
1.001 1.001
1 10
0 2000 4000 6000 8000 10000 2000 4000 6000 8000 10000
t t

𝜇 = 0.0002 𝜇 = 0.0008 b = 80 b = 140

𝜇 = 0.0004 𝜇 = 0.001 b = 100 b = 160
𝜇 = 0.0006 b = 120
(c) (d)
6 6
×10 ×10
1.009 1.007
e final scale of the spreader

1.008
e final scale of the spreader

1.006
1.007
1.005
1.006
1.005 1.004
1.004 1.003
1.003
1.002
1.002
1.001 1.001

1 1
0 2000 4000 6000 8000 10000 0 2000 4000 6000 8000 10000
t t
𝛽 = 50 × 0.000002/(1.35 × 109 ) 𝛽 = 50 × 0.000008/(1.35 × 109 ) 𝛼 = 2 × 10−14 𝛼 = 8 × 10−14
𝛽 = 50 × 0.000004/(1.35 × 109 ) 𝛽 = 50 × 0.00001/(1.35 × 109 ) 𝛼 = 4 × 10−14 𝛼 = 10 × 10−14
𝛽 = 50 × 0.000006/(1.35 × 109 ) 𝛼 = 6 × 10−14
(e) (f)

Figure 3: Continued.
8 Abstract and Applied Analysis

×106
×106
1.012 1.007

The final scale of the spreader

1.01 1.006
The final scale of the spreader

1.005
1.008
1.004
1.006
1.003
1.004
1.002
1.002 1.001

1 1
0 2000 4000 6000 8000 10000 0 2000 4000 6000 8000 10000
t t

𝛾 = 4 × 10−11 𝛾 = 8 × 10−11 𝜉1 = 50 × 0.0002/(1.35 × 109 ) 𝜉1 = 50 × 0.0008/(1.35 × 109 )

𝛾 = 6 × 10−11 𝛾 = 10 × 10−11 𝜉1 = 50 × 0.0004/(1.35 × 109 ) 𝜉1 = 50 × 0.001/(1.35 × 109 )
𝜉1 = 50 × 0.0006/(1.35 × 109 )
(g) (h)

×106
1.006
The final scale of the spreader

1.005

1.004

1.003

1.002

1.001

1
0 2000 4000 6000 8000 10000
t

𝜉2 = 50 × 0.0002/(1.35 × 109 ) 𝜉2 = 50 × 0.0008/(1.35 × 109 )

𝜉2 = 50 × 0.0004/(1.35 × 109 ) 𝜉2 = 50 × 0.001/(1.35 × 109 )
𝜉2 = 50 × 0.0006/(1.35 × 109 )

(i)

Figure 3: The sensitivity of the final scale in terms of parameters. (a) 𝜃. (b) 𝜂. (c) 𝜇. (d) 𝑏. (e) 𝛽. (f) 𝛼. (g) 𝛾. (h) 𝜉1 . (i) 𝜉2 .

Table 3: Description of parameters in the system (10).

Parameters Value Unit Comments

𝛽 50 × 0.00001/1350000000 Day−1 S(t)-to-I(t) transmission rate
𝛼 10−13 Day−1 W(t)-to-I(t) transmission rate
𝜂 0.5 Day−1 The rate to become 𝑅(𝑡) after receiving G(t)
𝛾 10−10 Day−1 G(t)-to-S(t) transmission rate
𝜇 0.0001 Day−1 The punishment rate for the spreaders
𝐵 1000000 Day−1 The quantity of messages by government
𝑏 100 Day−1 The quantity of messages by government
𝑝 1 Day−1 The submerged rate of messages by government
𝐺(0) 0 Individual The initial quantity of messages by government
Abstract and Applied Analysis 9

authority will release quantitative trustworthy message per 6. Sensitivity Analysis

time. So, the behavior of authority is independent of the
rumor spreading; that is, 𝐵(𝑆) = 𝐵. The parameters of system This paper mainly discusses the effect of measures adopted
(10) are in Table 3. by government. On the one hand, in the early stage, the sen-
sitivity of the basic reproduction numbers about parameters
Case 1 (𝐵(𝑆) = 𝐵). In this case, the disease-free equilibrium that correspond to measures adopted by government should
𝐸01 = (𝐼0 , 0, 𝑅0 , 0, 𝐺0 ), where 𝐼0 = 𝐴𝑝/(𝜂𝐵𝑟 + 𝑑𝑝), 𝑅0 = be discussed. On the other hand, when 𝑅01 < 1 or 𝑅0 < 1, the
𝜂𝑟𝐴𝐵/(𝜂𝐵𝑟 + 𝑑𝑝)𝑑, 𝐺0 = 𝐵/𝑝. Let us look at the basic sensitivity of the final scale of the spreader about parameters
reproduction number of the spreading of rumor. For the should be studied.
rumor spread, one has Now, we carry out the sensitivity analysis under different
cases.

𝜃𝛽𝐼𝑆 + 𝜃𝛼𝐼𝑊 Case 1 (𝐵(𝑆) = 𝐵). Consider

F=( 0 ),
0
𝑅01 = 𝜌 (𝐹𝑉−1 )
(11)
𝜂𝛿𝑆𝐺 + 2𝜉1 𝑆2 + 𝑚𝑆 + 𝜉2 𝑆𝑅 + 𝑑𝑆 + 𝜇𝑆
𝜃𝛽𝐼0 𝜆𝜃𝛼𝐼0
V=( 𝑘𝑊 − 𝜆𝑆 ). = + .
𝑝𝐺 − 𝐵 𝑚+𝑑+𝜇+𝜉2 𝑅0 +𝜂𝛿𝐺0 (𝑚+𝑑+𝜇+𝜉2 𝑅0 +𝜂𝛿𝐺0 ) 𝑘

(13)
So, we derive

Observing Figure 1, 𝑅01 is linear function of 𝜃, 𝜆, and 𝑝.

𝜃𝛽𝐼0 𝜃𝛼𝐼0 0 𝑅01 is the concave function with the rest of parameters, where
𝐹 (𝑡) = ( 0 0 0) , the influences of 𝜂, 𝐵, and 𝜇 are greater on 𝑅01 . Observing
0 0 0 the values of ordinate axis, 𝐵 and 𝜆 have the biggest influence
(12) on 𝑅01 ; that is, the releasing amount of messages is the most
𝜂𝛿𝐺0 + 𝜉2 𝑅0 + 𝑚 + 𝑑 + 𝜇 0 0 important. With regard to 𝑘 and 𝑝, the submerged rate of
𝑉 (𝑡) = ( −𝜆 𝑘 0) . message issued by the government has bigger effect than 𝑘. So,
0 0 𝑝 for message by spreader, we should control the distribution
of message. Once the message is issued, the deleting of
message has a small effect on controlling rumor spreading.
Then, the basic reproduction number is 𝑅01 = 𝜌(𝐹𝑉−1 ) = For the government, in order to prevent the rumor spread,
𝜃𝛽𝐼 /(𝑚 + 𝑑 + 𝜇 + 𝜉2 𝑅0 + 𝜂𝛿𝐺0 ) + 𝜆𝜃𝛼𝐼0 /(𝑚 + 𝑑 + 𝜇 + 𝜉2 𝑅0 +
0
the quantity and the survival time of message are important
𝜂𝛿𝐺0 )𝑘. From expression of 𝑅01 , we can see that the measures factors. From 𝜂 and 𝜃, the reliability of government and the
of government reduce the basic reproduction number. cognizance ability of the public are equally important and
𝑅01 is more sensitive with the reliability of government. With
Under some situations, once the rumor emerges, the regard to 𝜇, the concavity of curve is the biggest. When
government will issue the news to clarify the rumor and the the minority of people are published, the effect has been big
message released by the authority. The more the spreaders on 𝑅01 . From Figures 1(h), 1(i), and 1(j), it is easy to know that
there are the more the message the authority should issue. 𝛾 has the biggest effect on 𝑅01 . The effect of 𝛼 on 𝑅01 is the
Thus, we can let 𝐵(𝑆) = 𝑏𝑆𝑛 , 𝑛 ∈ N+ . smallest.
When 𝑅01 < 1, as time goes on, the rumor will eventually
Case 2 (𝐵(𝑆) = 𝑏𝑆𝑛 , 𝑛 ∈ N+ ). In Case 2, the disease-
disappear. In this case, what we should focus on is the final
free equilibrium 𝐸0 = (𝐴/𝑑, 0, 0, 0, 0). Similarly, the basic
scale of the spreader. Next, we discuss the influences of
reproduction number is 𝑅02 = 𝜌(𝐹𝑉−1 ) = 𝜃𝛽𝐼0 /(𝑚 + 𝑑 + 𝜇) + parameters on the final scale.
𝜆𝜃𝛼𝐼0 /(𝑚 + 𝑑 + 𝜇)𝑘, which does not relate to the parameters From Figures 2(a) and 2(b), we can see that 𝜂 has a
𝜂, 𝑏. From the expression of 𝑅02 , we can know that 𝑅02 is the bigger influence on the final scale than 𝜃, which means that
same as 𝑅0 . Thus, if the government adopts measure after the reliability of government is more important. Comparing
the appearance of rumor during the early stage of the rumor Figures 2(c) and 2(d), the change of the final scale caused
spread, the measure cannot change the value of the basic by the 𝐵 is bigger than 𝜇, which implies that the effect of
reproduction number. release of message is more obvious than punishment from
government. From Figures 2(e), 2(f), and 2(g), we know that
Case 3 (𝐵(𝑆) = 𝐵 + 𝑏𝑆𝑛 , 𝑛 ∈ N+ ). In this case, the disease-free when 𝑅01 < 1, the influence of 𝛼 is smaller. For 𝛾, 𝜉1 , and 𝜉2 ,
equilibrium 𝐸01 = (𝐼0 , 0, 𝑅0 , 0, 𝐺0 ), where 𝐼0 = 𝐴𝑝/(𝜂𝐵𝑟 + the influence of 𝛾 is the biggest and is followed by 𝜉2 and 𝜉1 .
𝑑𝑝), 𝑅0 = 𝜂𝑟𝐴𝐵/((𝜂𝐵𝑟 + 𝑑𝑝)𝑑), 𝐺0 = 𝐵/𝑝. And the basic
reproduction number is 𝑅01 = 𝜌(𝐹𝑉−1 ) = 𝜃𝛽𝐼0 /(𝑚 + 𝑑 + 𝜇 + Case 2 (𝐵(𝑆) = 𝑏𝑆). The basic reproduction number of the
𝜉2 𝑅0 + 𝜂𝛿𝐺0 ) + 𝜆𝜃𝛼𝐼0 /(𝑚 + 𝑑 + 𝜇 + 𝜉2 𝑅0 + 𝜂𝛿𝐺0 )𝑘, which is whole system is 𝑅0 = 𝜃𝛽𝐼0 /(𝑚 + 𝑑 + 𝜇) + 𝜆𝜃𝛼𝐼0 /(𝑚 + 𝑑 + 𝜇)𝑘.
similar to Case 1. We can know that the basic reproduction number does not
10 Abstract and Applied Analysis

change. For such reason, we should focus on the final scale of [3] W. Li, S. Tang, S. Pei et al., “The rumor diffusion process with
the spreader, which is showed in Figure 3. emerging independent spreaders in complex networks,” Physica
When 𝐵(𝑆) = 𝑏𝑆, then 𝜃 has a bigger influence on the A, vol. 397, pp. 121–128, 2014.
final scale than 𝜂, which is different from Case 1. Observing [4] J. J. Wang, L. J. Zhao, and R. B. Huang, “SIRaRu rumor spreading
Figure 2(c), the effect of 𝜇 is smaller. The changes of the final model in complex networks,” Physica A, vol. 398, pp. 43–55,
scales of the spreader are very little under the changes of 𝛼 2014.
and 𝜉1 . The effect of 𝛽 is larger than 𝜉2 and 𝛾. [5] S. Han, F. Z. Zhuang, Q. He, Z. Z. Shi, and X. Ao, “Energy model
for rumor propagation on social networks,” Physica A, vol. 394,
pp. 99–109, 2014.
7. Discussion [6] Y. L. Zan, J. L. Wu, P. Li, and Q. L. Yu, “SICR rumor spreading
Applying the dynamical system, this paper describes the model in complex networks: counterattack and self-resistance,”
Physica A, vol. 405, pp. 159–170, 2014.
government measures by the parameters 𝜂, 𝜃, 𝜇, 𝜆, 𝑘, 𝐵,
𝑏, and 𝑝. More specifically, 𝜂 indicates the reliability of [7] D. J. Daley and D. G. Kendall, “Epidemics and rumours,” Nature,
vol. 204, pp. 11–18, S1464–S3634, 1964.
government, 𝜃 indicates the ability of cognizance of the
public, 𝜇 indicates the punishment rate of the government, [8] D. P. Maki and M. Thompson, Mathematical Models and
𝑘 reflects the management strength of government for the Applications, Prentice Hall, Englewood Cliffs, NJ, USA, 1973.
internet, 𝐵 and 𝑏 show the amount of messages released by [9] L. J. Zhao, J. J. Wang, Y. C. Chen et al., “SIHR rumor spreading
authority, 𝛽 is the transmission rate between humans directly, model in social networks,” Physica A, vol. 391, pp. 2444–2453,
2012.
𝛼 is the transmission rate of media to human, and 𝛾 is the
transmission rate of the government to human by issuing [10] K. Kawachi, M. Seki, H. Yoshida, Y. Otake, K. Warashina, and
presentation. H. Ueda, “A rumor transmission model with various contact
interactions,” Journal of Theoretical Biology, vol. 253, no. 1, pp.
According to the above dynamical analysis and sensitivity
55–60, 2008.
analysis, we can know that 𝐵 and 𝑏 have the greatest influence
[11] L. J. Zhao, H. X. Cui, X. Y. Qiu, X. L. Wang, and J. J. Wang, “SIR
on the rumor spread. The effects of 𝜂 and 𝜃 are almost
rumor spreading model in the new media age,” Physica A, vol.
big. When 𝐵(𝑆) = 𝐵, the influence of 𝜂 is larger which 392, no. 4, pp. 995–1003, 2013.
means that the reliability of government is more important
[12] L. Huo, T. T. Lin, and P. Q. Huang, “Dynamical behavior
when the government issues message beforehand. When
of a rumor transmission model with psychological effect in
𝐵(𝑆) = 𝑏𝑆, the influence of 𝜃 is larger which means that the emergency event,” Abstract and Applied Analysis, vol. 2013,
cognizance ability of the public is more important when the Article ID 282394, 9 pages, 2013.
government releases the message according to the number [13] G. H. Chen, H. Z. Shen, T. Ye, G. M. Chen, and N. Kerr, “A
of the spreaders. The effects of 𝑘, 𝜆, and 𝑝 explain that kinetic model for the spread of rumor in emergencies,” Discrete
monitoring the internet to prevent the diffusion of rumor Dynamics in Nature and Society, vol. 2013, Article ID 605854, 8
is more important than the deleting message in media that pages, 2013.
has appeared. Moreover, extending the survival time of [14] Y. Naimi and M. Naimi, “Reliability and efficiency of gener-
government message is also necessary. The relationship of alized rumor spreading model on complex social networks,”
𝑅01 in terms of 𝜇 shows that when government punishes the Communications in Theoretical Physics, vol. 60, pp. 139–144,
minority of people, the effect is obvious. However, with the 2013.
increase of 𝜇, the effect is weakened. 𝑅01 is a concave function [15] J. Y. Huang and X. G. Jin, “Preventing rumor spreading on
with 𝛾. From Figure 1(j), we can know that the influence of 𝛾 small-world networks,” Journal of Systems Science & Complex-
is larger than 𝛽 and 𝛼. ity, vol. 24, no. 3, pp. 449–456, 2011.
In [26], an ISRW model was presented and its dynamical [16] D. H. Zanette, “Dynamics of rumor propagation on small-world
behaviors were well investigated. Reference [26] was mainly networks,” Physical Review E, vol. 65, Article ID 041908, 2002.
based on the Jacobian matrix and obtained the final size of [17] Y. Moreno, M. Nekovee, and A. F. Pacheco, “Dynamics of rumor
rumor. However, this paper is based on spectral radius and spreading in complex networks,” Physical Review E, vol. 69,
focuses on the effects of different measures. The obtained Article ID 066130, 2004.
results well will enrich the findings in rumor spreading. [18] D. Trpevski, “Model for rumor spreading over networks,”
Physical Review E, vol. 81, Article ID 056102, 2010.
Conflict of Interests [19] Y. C. Zhang, S. Zhou, Z. Z. Zhang, J. H. Guan, and S. G. Zhou,
“Rumor evolution in social networks,” Physical Review E, vol. 87,
The authors declare that there is no conflict of interests Article ID 032133, 2013.
regarding the publication of this paper. [20] B. Doerr, A. Huber, and A. Levavi, “Strong robustness of ran-
domized rumor spreading protocols,” Discrete Applied Mathe-
matics, vol. 161, no. 6, pp. 778–793, 2013.
References [21] E. Lebensztayn and P. M. Rodriguez, “A connection between a
[1] S. C. Pendleton, “Rumor research revisited and expanded,” system of random walks and rumor transmission,” Physica A,
Language & Communication, vol. 1, no. 18, pp. 69–86, 2008. vol. 392, no. 23, pp. 5793–5800, 2013.
[2] S. Young, A. Pinkerton, and K. Dodds, “The word on the street: [22] N. Fountoulakis and K. Panagiotou, “Rumor spreading on
rumor, “race” and the anticipation of urban unrest,” Political random regular graphs and expanders,” Random Structures and
Geography, vol. 38, pp. 57–67, 2014. Algorithms, vol. 43, no. 2, pp. 201–220, 2013.
Abstract and Applied Analysis 11

[23] C. F. Coletti, P. M. Rodrı́guez, and R. B. Schinazi, “A spatial

stochastic model for rumor transmission,” Journal of Statistical
Physics, vol. 147, no. 2, pp. 375–381, 2012.
[24] Y. Zhang and Y. Guo, “A study on rumor propagation model on
micro-blogging platforms,” Advanced Materials Research, vol.
717, pp. 808–811, 2013.
[25] L. Ma, Z. T. Jia, H. Y. Sun, and C. Yu, “An improved model of
the internet public opinion spreading on mass emergencies,”
Applied Mechanics and Materials, vol. 433–435, pp. 1760–1764,
2013.
[26] X. Zhao and J. Wang, “Dynamical model about rumor spread-
ing with medium,” Discrete Dynamics in Nature and Society, vol.
2013, Article ID 586867, 9 pages, 2013.
[27] L. A. Huo, P. Q. Huang, and X. Fang, “An interplay model for
authorities’ actions and rumor spreading in emergency event,”
Physica A, vol. 390, no. 20, pp. 3267–3274, 2011.
[28] Y. Y. Bao, Y. Niu, C. Q. Yi, and Y. B. Xue, “Effective immunization
strategy for rumor propagation based on maximum spanning
tree,” in Proceedings of the 2014 Interernational Conference
on Computing, Networking and Communications, Communica-
tions and Information Security Symposium, 2014.
[29] K. H. Ji, J. W. Liu, and G. Xiang, “Anti-rumordynamics and
rumor control strategies on complex network,” http://arxiv.org/
abs/1310.7198.
[30] H. Rahmandad and J. Sterman, “Heterogeneity and network
structure the dynamics of diffusion: comparing agent-based and
differential equation models,” Management Science, vol. 54, pp.
998–1014, 2008.
[31] L. V. Green and P. J. Kolesar, “Improving emergency responsive-
ness with improving emergency responsiveness with manage-
ment science,” Management Science, vol. 50, pp. 1001–1014, 2004.
[32] X.-X. Zhao and J.-Z. Wang, “Dynamics of an information
spreading model with isolation,” Abstract and Applied Analysis,
vol. 2014, Article ID 484630, 6 pages, 2014.
[33] R. M. Anderson and R. M. May, Infectious Diseases of Humans,
Oxford University Press, Oxford, UK, 1991.
[34] O. Diekmann and J. A. P. Heesterbeek, Mathematical Epidemi-
ology of Infectious Diseases, Wiley Series in Mathematical and
Computational Biology, John Wiley & Sons, West Sussex, UK,
2000.
[35] P. van den Driessche and J. Watmough, “Reproduction numbers
and sub-threshold endemic equilibria for compartmental mod-
els of disease transmission,” Mathematical Biosciences, vol. 180,
pp. 29–48, 2002.
[36] J. LaSalle and S. Lefschetz, Stability by Liapunov’s Direct Method,
Academic Press, New York, NY, USA, 1961.
[37] E. A. Barbashin, Introduction to the Theory of Stability, Wolters-
Noordhoff, Groningen, The Netherlands, 1970.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 434621, 7 pages
http://dx.doi.org/10.1155/2014/434621

Research Article
Stability Analysis of a Class of Higher Order
Difference Equations

Yuanyuan Liu1 and Fanwei Meng2

1
Department of Mathematics, Shanghai Normal University, Shanghai 200234, China
2
Department of Mathematics, Qufu Normal University, Qufu 273165, China

Correspondence should be addressed to Yuanyuan Liu; liuyuanyuan1117@163.com

Received 31 March 2014; Accepted 6 May 2014; Published 28 May 2014

Academic Editor: Yonghui Xia

Copyright © 2014 Y. Liu and F. Meng. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

We consider the sufficient conditions for asymptotic stability and instability of certain higher order nonlinear difference equations
with infinite delays in finite-dimensional spaces. With the aid of the general comparison condition on the right-hand side function
𝑓𝑘 (⋅), we generalize the stability and instability result.

1. Introduction In this paper, we consider a higher order difference

equation with delays
In the field of statistics, people get used to mark the number
of sample as discrete parameter not continuous for applied 𝑘−1
meaning. There are some discrete models in real life such as 𝑦𝑘+1 = 𝑓𝑘 (𝑦𝑘 , 𝑦𝑘−1 , . . . , 𝑦𝑘−𝑟 , ∑ 𝑔𝑘,𝑠 (𝑦𝑠 , 𝑦𝑠−1 , . . . , 𝑦𝑠−𝑟+1 )) ,
cobweb model of demand and supply and national income 𝑠=0
model, which could not be described by differential equation (2)
under the factual conditions. Then difference equations arise
and are widely used in areas such as information transmission where 𝑦𝑘 ∈ 𝑋, 𝑋 is a Banach space, and 𝑓𝑘 (⋅), 𝑔𝑘,𝑠 (⋅) are
and population models [1–4].
During the period of development of mathematics functions on 𝑍 × 𝑋𝑟+1 → 𝑋 and 𝑍2 × 𝑋𝑟−1 → 𝑋, respe-
together with information revolution, there are many discrete ctively. We assume that (2) satisfies the initial condition (𝑥0 ,
dynamic systems to describe the real problem, such as the 𝑥−1 , . . . , 𝑥−𝑟 ), which means that the solution of system (2),
monographs [5, 6]. For example, the book Liber Abaci, {𝑦𝑛 }𝑛≥−𝑟 , holds for 𝑦𝑖 = 𝑥𝑖 , 𝑖 = −𝑟, 1 − 𝑟, . . . , 0. In addition,
written by the famous mathematician Leonardo di Pisa, stated we suppose that the zero solution is the unique equilibrium;
the so-called Fibonacci sequence to the problem of rabbits’ in other words, 𝑓𝑘 (𝑐, 𝑐, . . . , 𝑐) = 𝑐 for some real values and
number. To march the development of mathematic discrete all 𝑘 ∈ 𝑍+ = {𝑥 | 𝑥 ≥ 0, 𝑥 ∈ 𝑍} if and only if 𝑐 = 0. Yao
models in application, difference equations raised the form of and Meng in [9] studied the similar system and gave some
higher orders or with delays; see [7–9]. Besides, the stability of sufficient conditions to guarantee that the solution satisfies a
the equilibrium is always a great topic to study, such as [10–13] contained equation under the method of a discrete inequality.
and the references therein. Also, there are some researchers We note that the system (2) always has solution at the time 𝑘
engaged in studying the stability analysis of the equations given by
with delay; see [14, 15].
As the simplest case, the first order difference equation 𝑦1 = 𝑓0 (𝑦0 , 𝑦−1 , . . . , 𝑦−𝑟 ) ,
𝑦𝑘+1 = 𝑓𝑘 (𝑦𝑘 ) , 𝑘 ≥ 0 ∈ 𝑍, (1) 𝑦2 = 𝑓1 (𝑦1 , 𝑦0 , . . . , 𝑦1−𝑟 , 𝑔1,0 (𝑦0 , 𝑦−1 , . . . , 𝑦−𝑟 )) ,
has been paid much attention to and widely investigated in ..
the literature. .
2 Abstract and Applied Analysis

Denote by
𝑦𝑘 = 𝑓𝑘−1 (𝑦𝑘−1 , 𝑦𝑘−2 , . . . , 𝑦𝑘−𝑟−1 ,
𝑟 𝑟
𝑘−2 𝑃 = ∑𝑝𝑖 , 𝑄 = ∑𝑞𝑖 . (6)
∑ 𝑔𝑘−1,𝑠 (𝑦𝑠 , 𝑦𝑠−1 , . . . , 𝑦𝑠−𝑟+1 )) . (3) 𝑟=0 𝑖=1
𝑠=0

It is easy to see that 𝑦𝑘 is defined by the solution 𝑦𝑘−1 . Using the method in [10], we can obtain the following
The main aim of this paper is to provide the sufficient theorem.
conditions for asymptotic stability and instability of (2). First,
we state some notations and basic definitions used in the Theorem 4. Under the hypothesises (C1, C2, and C3), system
paper. Denote that ‖x‖ = √𝑥02 + 𝑥12 + ⋅ ⋅ ⋅ + 𝑥𝑛2 for the vector (2) is asymptotically stable if 𝑃 > 1, 𝛼 > 0, 0 < 𝛽 < 1, and
x = (𝑥0 , 𝑥1 , . . . , 𝑥𝑛 ) ∈ 𝑋, which implies that, for 𝑓, 𝑔 ∈ 𝑋, 𝑀 < 1/2𝛼.
󵄩󵄩󵄩𝑓 ⋅ 𝑔󵄩󵄩󵄩 ≤ 󵄩󵄩󵄩𝑓󵄩󵄩󵄩 ⋅ 󵄩󵄩󵄩𝑔󵄩󵄩󵄩 , 󵄩󵄩󵄩𝑓 + 𝑔󵄩󵄩󵄩 ≤ 󵄩󵄩󵄩𝑓󵄩󵄩󵄩 + 󵄩󵄩󵄩𝑔󵄩󵄩󵄩 . (4)
󵄩 󵄩 󵄩 󵄩 󵄩 󵄩 󵄩 󵄩 󵄩 󵄩 󵄩 󵄩 Proof. Since 𝑃 > 1, we can choose two real numbers 𝑠 ≥ 1,
Definition 1. The zero solutions of the system (2) are stable if, 0 < 𝛿 < 1 such that 𝑃 − 𝑠 > 0,
for every 𝜀 > 0, 𝑘0 ∈ 𝑍+ , there exists 𝛿(𝜀, 𝑘0 ) > 0 such that
every solution {𝑥𝑘 } of the system with the initial condition 𝑝1 𝑝
𝑝0 + + ⋅ ⋅ ⋅ + 𝑟𝑟 − 𝑠 > 0,
‖(𝑥𝑘0 −𝑘 , . . . , 𝑥𝑘0 )‖ ≤ 𝛿 satisfies the condition ‖𝑥𝑘 ‖ < 𝜀, for all 𝑠 𝑠
𝑘 ≥ 𝑘0 . (7)
𝑝0 +𝑝1 /𝑠+⋅⋅⋅+𝑝𝑟 /𝑠𝑟 −𝑠 1
𝛼𝛽𝛿 < .
Definition 2. The zero solutions of the system (2) are asymp- 2
totically stable if it is stable and, under a constant 𝜇 existing,
lim𝑘 → ∞ ‖𝑥𝑘 ‖ = 0 for any solution {𝑥𝑛 } of (2) such that From (7), it is easy to note that
‖(𝑥𝑘0 −𝑘 , . . . , 𝑥𝑘0 )‖ ≤ 𝜇.
𝑝1 + 𝑞1 𝑝 +𝑞
Definition 3. The zero solutions of system (2) are unstable if it 𝑝0 + + ⋅ ⋅ ⋅ + 𝑟 𝑟 𝑟 − 𝑠 > 0,
is not stable, that is, if there are numbers 𝜀 > 0, 𝑘0 ∈ 𝑍+ , such 𝑠 𝑠
(8)
that for every 𝛿 > 0 there is a solution {𝑥𝑘 } with the initial 𝑝0 +(𝑝1 +𝑞1 )/𝑠+⋅⋅⋅+(𝑝𝑟 +𝑞𝑟 )/𝑠𝑟 −𝑠 1
condition ‖(𝑥𝑘0 −𝑘 , . . . , 𝑥𝑘0 )‖ ≤ 𝛿 and there exists a number 𝛼𝛽𝛿 < .
2
𝐾 ≥ 𝑘0 such that ‖𝑥𝐾 ‖ ≥ 𝜀.

The system we consider is a general form of higher order Next, we consider solution of the system (2) with the initial
difference equations without the concrete formula of 𝑓𝑘 (⋅). If condition such that ‖𝑦𝑘 ‖ ≤ 𝛿, 𝑘 = 0, −1, . . . , −𝑟. According to
we entitle the function 𝑓𝑘 (⋅) to a special form, such as the form the definition, we will prove the following inequality:
of logistic function, then the stability of equilibrium points
can be used to discuss the population model. 󵄩󵄩 󵄩󵄩 𝑠𝑘
󵄩󵄩𝑦𝑘 󵄩󵄩 ≤ 𝛿 , ∀𝑘 ∈ 𝑍+ , (9)

2. Asymptotic Stability
which is obvious to hold for 𝑘 = 1.
In this section, using the method of comparison conditions, Assuming that (9) still holds for 2, 3, . . . , 𝑘(𝑘 ≥ 2), in
we will obtain the sufficient conditions of the nonlinear addition to (𝐶2) and (𝐶3), then we estimate the solution at
difference system (2). The following discrete inequalities we the (𝑘 + 1) step by (4).
assumed play an important role in the investigations of the Let 𝑧𝑟+1 = ∑𝑘−1
𝑠=0 𝑔𝑘,𝑠 (𝑦𝑠 , 𝑦𝑠−1 , . . . , 𝑦𝑠−𝑟+1 ). Then
asymptotic stability:
𝑟
󵄩󵄩 󵄩 󵄩 󵄩𝑝 󵄩 󵄩 󵄩󵄩𝑘−1 󵄩󵄩
(𝐶1) 󵄩󵄩𝑓𝑘 (𝑥0 , 𝑥1 , . . . , 𝑥𝑟+1 )󵄩󵄩󵄩 ≤ 𝛼∏󵄩󵄩󵄩𝑥𝑖 󵄩󵄩󵄩 𝑖 󵄩󵄩󵄩𝑥𝑟+1 󵄩󵄩󵄩 , 󵄩󵄩 󵄩 󵄩󵄩 󵄩󵄩
󵄩󵄩𝑧𝑟+1 󵄩󵄩󵄩 = 󵄩󵄩󵄩 ∑ 𝑔𝑘,𝑠 (𝑦𝑠 , 𝑦𝑠−1 , . . . , 𝑦𝑠−𝑟+1 )󵄩󵄩󵄩
󵄩󵄩󵄩 𝑠=0 󵄩󵄩
𝑖=0
󵄩
𝛼 > 0, 𝑝𝑖 ≥ 0, 𝑖 = 0, 1, . . . , 𝑟; 󵄩󵄩
󵄩󵄩
𝑟 = 󵄩󵄩󵄩𝑔𝑘,𝑘−1 (𝑦𝑘−1 , 𝑦𝑘−2 , . . . , 𝑦𝑘−𝑟 )
󵄩󵄩 󵄩 󵄩 󵄩𝑞 󵄩󵄩
(𝐶2) 󵄩󵄩𝑔𝑘,𝑠 (𝑥0 , 𝑥1 , . . . , 𝑥𝑟−1 )󵄩󵄩󵄩 ≤ 𝛽∏󵄩󵄩󵄩𝑥𝑖−1 󵄩󵄩󵄩 𝑖 , 󵄩
𝑖=1
𝑘−2 󵄩󵄩 (10)
󵄩󵄩
𝛽 > 0, 𝑞𝑖 ≥ 0, 𝑖 = 0, 1, . . . , 𝑟 − 1; + ∑ 𝑔𝑘,𝑠 (𝑦𝑠 , 𝑦𝑠−1 , . . . , 𝑦𝑠−𝑟+1 )󵄩󵄩󵄩
󵄩󵄩
𝑠=0 󵄩
𝑘−2 󵄩󵄩 𝑞1 𝑞2 𝑞𝑟 󵄩
(𝐶3)
󵄩 󵄩
∑ 󵄩󵄩󵄩𝑔𝑘,𝑠 (𝑥0 , 𝑥1 , . . . , 𝑥𝑟 )󵄩󵄩󵄩 ≤ 𝑀 < ∞, 𝑀 ∈ 𝑅+ . ≤ 󵄩󵄩󵄩𝛽𝑦𝑘−1 𝑦𝑘−2 ⋅ ⋅ ⋅ 𝑦𝑘−𝑟 󵄩󵄩󵄩󵄩 + 𝑀
𝑠=𝑘0
𝑘−1
𝑞1 +𝑠𝑘−2 𝑞2 +⋅⋅⋅+𝑠𝑘−𝑟 𝑞𝑟
(5) ≤ 𝛽𝛿𝑠 + 𝑀.
Abstract and Applied Analysis 3

By the inequalities (8) and (10), we can obtain the (i) If 𝑘 is on the end of 𝐼𝑛 , such that 𝑘 + 1 ∈ 𝐼𝑛+1 , then we
following conclusion: have
󵄩󵄩 󵄩󵄩
󵄩󵄩 󵄩 󵄩 󵄩 󵄩󵄩𝑦𝑘+1 󵄩󵄩
󵄩󵄩𝑦𝑘+1 󵄩󵄩󵄩 = 󵄩󵄩󵄩𝑓𝑘 (𝑦0 , 𝑦1 , . . . , 𝑧𝑟+1 )󵄩󵄩󵄩
𝑛+1 𝑝0 𝑛+1 𝑝1 +𝑞1 𝑛+1 𝑝𝑟 +𝑞𝑟
𝑟
≤ 𝛼𝛽 (((𝛼𝛽) 𝛿) ((𝛼𝛽) 𝛿) ⋅ ⋅ ⋅ ((𝛼𝛽) 𝛿) )
󵄩 󵄩𝑝 󵄩 󵄩
≤ 𝛼∏󵄩󵄩󵄩𝑦𝑖 󵄩󵄩󵄩 𝑖 󵄩󵄩󵄩𝑧𝑟+1 󵄩󵄩󵄩 𝑛+1 𝑝0 𝑛+1 𝑝1 𝑛+1 𝑝𝑟
𝑖=0 + 𝛼(𝛼𝛽) 𝛿 ((𝛼𝛽) 𝛿) ⋅ ⋅ ⋅ ((𝛼𝛽) 𝛿) 𝑀
󵄩 󵄩𝑝 󵄩 󵄩𝑝 󵄩 󵄩𝑝
≤ 𝛼󵄩󵄩󵄩𝑦𝑘 󵄩󵄩󵄩 0 󵄩󵄩󵄩𝑦𝑘−1 󵄩󵄩󵄩 1 ⋅ ⋅ ⋅ 󵄩󵄩󵄩𝑦𝑘−𝑟 󵄩󵄩󵄩 𝑟 𝑃+𝑄 𝑃
𝑛+1 𝑛+1
= 𝛼𝛽(((𝛼𝛽) 𝛿) + 𝛼(𝛼𝛽) 𝛿) 𝑀.
𝑘−1
𝑞1 +𝑠𝑘−2 𝑞2 +⋅⋅⋅+𝑠𝑘−𝑟 𝑞𝑟 (15)
× (𝛽𝛿𝑠 + 𝑀)

𝑘
𝑝0 +𝑠𝑘−1 𝑝1 +⋅⋅⋅+𝑠𝑘−𝑟 𝑝𝑟 𝑘−1
𝑞1 +𝑠𝑘−2 𝑞2 +⋅⋅⋅+𝑠𝑘−𝑟 𝑞𝑟
By the assumptions in this theorem, we have
≤ 𝛼𝛿𝑠 (𝛽𝛿𝑠 + 𝑀)
󵄩󵄩 󵄩 𝑛+2 𝑛+2
󵄩󵄩𝑦𝑘+1 󵄩󵄩󵄩 ≤ (𝛼𝛽) 𝛿 (𝛿 + 𝛽 𝑀) ≤ (𝛼𝛽) .
𝑄 −1
(16)
𝑘 𝑘−1 𝑘−𝑟
= 𝛼𝛽𝛿𝑠 𝑝0 +𝑠 (𝑝1 +𝑞1 )+⋅⋅⋅+𝑠 (𝑝𝑟 +𝑞𝑟 )

As desired.
𝑠𝑘 𝑝0 +𝑠𝑘−1 𝑝1 +⋅⋅⋅+𝑠𝑘−𝑟 𝑝𝑟
+ 𝛼𝛿 𝑀 (ii) If 𝑘 is inside 𝐼𝑛 , in other words, 𝑘, 𝑘 − 1, . . . , 𝑘 − 𝑖 ∈ 𝐼𝑛
and 𝑘 − 𝑖 − 1, 𝑘 − 𝑖 − 2, . . . , 𝑘 − 𝑟 ∈ 𝐼𝑛−1 , such that
𝑘+1 𝑘
(𝑝0 +(𝑝1 +𝑞1 )/𝑠+⋅⋅⋅+(𝑝𝑟 +𝑞𝑟 )/𝑠𝑟 −𝑠)
= 𝛿𝑠 (𝛼𝛽𝛿𝑠 𝑘 + 1 ∈ 𝐼𝑛+1 , then we have

󵄩󵄩 󵄩 𝑛+1 𝑝0 𝑛+1 𝑝𝑖 +𝑞𝑖

𝑠𝑘 (𝑝0 +𝑝1 /𝑠+⋅⋅⋅+𝑝𝑟 /𝑠𝑟 −𝑠) 𝑠𝑘+1 󵄩󵄩𝑦𝑘+1 󵄩󵄩󵄩 ≤ 𝛼𝛽 (((𝛼𝛽) 𝛿) ⋅ ⋅ ⋅ ((𝛼𝛽) 𝛿)
+ 𝛼𝛿 𝑀) ≤ 𝛿 ,
(11) 𝑛 𝑝𝑖+1 +𝑞𝑖+1 𝑛 𝑝𝑟 +𝑞𝑟
× ((𝛼𝛽) 𝛿) ⋅ ⋅ ⋅ ((𝛼𝛽) 𝛿) )
(17)
𝑝0 𝑝𝑖
which implies that inequality (9) holds for the step (𝑘 + 1). + 𝛼 (((𝛼𝛽)
𝑛+1
𝛿) ⋅ ⋅ ⋅ ((𝛼𝛽)
𝑛+1
𝛿)
Therefore, inequality (9) holds for all 𝑘 ∈ 𝑍+ . Since 𝛿 ∈ (0, 1),
we can obtain that 𝑠𝑘 → ∞ as 𝑘 → ∞ and derive the × ((𝛼𝛽) 𝛿)
𝑛 𝑝𝑖+1
⋅ ⋅ ⋅ ((𝛼𝛽) 𝛿) ) 𝑀.
𝑛 𝑝𝑟

conclusion. Hence, the proof is completed.

Theorem 5. Let system (2) satisfy the same conditions as in Because of the sufficient condition in this theorem, we can
Theorem 4. Then, the zero solutions are asymptotically stable if derive
𝑃 = 1, 0 < 𝛼𝛽 < 1, and 𝑀 < 𝛽(1 − 𝛿𝑄).
󵄩󵄩 󵄩 𝑛 𝑃+𝑄 𝑛 𝑃 𝑛+1
󵄩󵄩𝑦𝑘+1 󵄩󵄩󵄩 ≤ 𝛼𝛽((𝛼𝛽) 𝛿) + 𝛼((𝛼𝛽) 𝛿) 𝑀 ≤ (𝛼𝛽) 𝛿,
Proof. According to the assumption, we can define a set for (18)
every integer 𝑛 ≥ 0,
which proves that (14) holds for 𝑦𝑘+1 . From estimation (14)
𝐼𝑛 = {𝑛 (𝑟 + 1) + 𝑖 : 𝑖 = 1, 2, . . . , 𝑟 + 1} , (12) and 𝛼𝛽 ∈ (0, 1), it follows the asymptotic stability of the zero
solutions of system (2). The proof is ended.

which contains 𝑟 + 1 index numbers and

3. Instability Conditions
∞
Apart from the instability comparison condition defined in
𝑍+ = ⋃ 𝐼𝑛 . (13) the whole state space 𝑌 = 𝑅𝑛 in [13], we state that the
𝑛=0
condition is satisfied only in some subset of the state space
𝑌. Thus, we introduce a convex set 𝐺 ⊂ 𝑌, 0 ∈ 𝐺 to be an
To prove that the solution is asymptotically stable, according invariant set for system (2). Further, for every 𝑘0 ∈ 𝑍+ , the
to the definition, we will prove the following inequality: solution 𝑦𝑘 with the initial condition (3) starting in 𝐺: 𝑦𝑛 ∈ 𝐺,
𝑛 = 0, −1, . . . , −𝑟 satisfies 𝑦𝑘 ∈ 𝐺 for all 𝑘 ∈ 𝑍+ .
󵄩󵄩 󵄩󵄩 𝑛+1 In this section, we assume the following comparison
󵄩󵄩𝑦𝑘 󵄩󵄩 ≤ (𝛼𝛽) 𝛿, ∀𝑘 ∈ 𝐼𝑛 . (14)
condition:
𝑟
It is obvious that (9) holds for 𝑘 = 1. Since the definition 󵄩󵄩 󵄩 󵄩 󵄩𝑝̃ 󵄩 󵄩
(𝐶4) 󵄩󵄩𝑓𝑘 (𝑥0 , 𝑥1 , . . . , 𝑥𝑟+1 )󵄩󵄩󵄩 ≥ 𝛼̃∏󵄩󵄩󵄩𝑥𝑖 󵄩󵄩󵄩 𝑖 󵄩󵄩󵄩𝑥𝑟+1 󵄩󵄩󵄩 ,
of the set 𝐼𝑛 and the assumption of the solution {𝑦𝑘 } fit the 𝑖=0
estimation (14) at the step 𝑘, the following proof will be
divided into two aspects. 𝛼̃ > 0, 𝑝̃𝑖 ≥ 0;
4 Abstract and Applied Analysis

𝑟
󵄩󵄩 󵄩 󵄩 󵄩𝑞̃ (a) The functions 𝜇(𝑘), ](𝑘) satisfy the following rela-
(𝐶5) 󵄩󵄩𝑔𝑘,𝑠 (𝑥0 , 𝑥1 , . . . , 𝑥𝑟−1 )󵄩󵄩󵄩 ≥ 𝛽̃∏󵄩󵄩󵄩𝑥𝑖−1 󵄩󵄩󵄩 𝑖 ,
𝑖=1
tions:

𝛽̃ > 0, 𝑞̃𝑖 ≥ 0. (19) 𝜇 (0) = 𝜇 (−1) = ⋅ ⋅ ⋅ = 𝜇 (−𝑟) = 0,

𝜇 (𝑘 + 1) = 1 + 𝑝̃0 𝜇 (𝑘) + (𝑝̃1 + 𝑞̃1 ) 𝜇 (𝑘 − 1) (25)

Denote by
+ ⋅ ⋅ ⋅ + (𝑝̃𝑟 + 𝑞̃𝑟 ) 𝜇 (𝑘 − 𝑟) ,
𝑟 𝑟−1
] (0) = ] (−1) = ⋅ ⋅ ⋅ = ] (−𝑟) = 1,
𝑃̃ = ∑𝑝̃𝑖 , ̃ = ∑𝑞̃𝑖 .
𝑄 (20)
𝑟=0 𝑖=0 ] (𝑘 + 1) = 𝑝̃0 ] (𝑘) + (𝑝̃1 + 𝑞̃1 ) ] (𝑘 − 1) (26)

The following three theorems are our main results on + ⋅ ⋅ ⋅ + (𝑝̃𝑟 + 𝑞̃𝑟 ) ] (𝑘 − 𝑟) ,
instability of the system (2).

Theorem 6. Under conditions (C4)-(C5), system (2) is unsta- which can be directly proved by the method of induction in
̃ < 1 for all 0 < 𝛼̃𝛽̃ < 1.
ble if 0 < 𝑃̃ + 𝑄 𝑘 ∈ 𝑍+ .

Proof. Let the arbitrary initial conditions satisfy (b) The functions 𝜇(𝑘), ](𝑘) satisfy the following estima-
tions:
󵄩󵄩 󵄩󵄩
󵄩󵄩𝑦𝑖 󵄩󵄩 = 𝜂𝑖 , 𝑛 𝑘−1
̃ 𝑖 ≤ 𝜇 (𝑘) ≤ ∑ (𝑃̃ + 𝑄)
∑(𝑃̃ + 𝑄) ̃ 𝑖, ∀𝑘 ∈ 𝐼𝑛 , 𝑛 ∈ 𝑍+ ;
𝑖 = 0, −1, . . . , −𝑟, 𝜂 = min {1, 𝜂𝑖 : 𝑖 = 0, −1, −2, . . . , −𝑟} . 𝑖=0 𝑖=0
(21) (27)

̃ 𝑛,
] (𝑘) ≤ (𝑃̃ + 𝑄) if (𝑃̃ + 𝑄)
̃ ∈ (0, 1] , ∀𝑘 ∈ 𝐼𝑛 , 𝑛 ∈ 𝑍+ ,
Note that ‖𝑦𝑖 ‖ ≥ 𝜂, for all 𝑖 = 0, −1, . . . , −𝑟.
For 𝑘 = 1, from (2) and conditions (C4), (C5), we have (28)

𝑟 𝑟
󵄩 󵄩𝑝̃ 𝑝̃ where 𝐼𝑛 is defined by (12).
𝑦1 = 𝑓0 (𝑦0 , 𝑦−1 , . . . , 𝑦−𝑟 ) ≥ 𝛼̃∏󵄩󵄩󵄩𝑦𝑖 󵄩󵄩󵄩 𝑖 = 𝛼̃∏𝜂𝑖 𝑖 The proof of Proposition (b) is as follows. We will prove
𝑖=0 𝑖=0 (22) the left-hand side inequality (27) by induction on each set 𝐼𝑛 ,
̃
̃ ̃ ̃ ̃ 𝑃̃ .
≥ 𝛼̃𝜂𝑝0 +𝑝1 +⋅⋅⋅+𝑝𝑟 = 𝛼̃𝜂𝑃 ≥ 𝛼̃𝛽𝜂 and the right-hand side can be proved similarly.
It is obvious that the assertion is true for the step 𝑘 = 1.
Assume that the assertion holds until the step 𝑘 ∈ 𝐼. For
For 𝑘 = 2, by the same way and noting that 𝐺 is invariant set, every 𝑘 ∈ 𝐼𝑛 , using Proposition (a), (24) and by induction
we have the following estimation: assumption, we have

𝑦2 = 𝑓1 (𝑦1 , 𝑦0 , . . . , 𝑦−𝑟+1 , 𝑔1,0 (𝑦0 , . . . , 𝑦−𝑟+1 )) 𝜇 (𝑘 + 1) = 1 + 𝑝̃0 𝜇 (𝑘) + (𝑝̃1 + 𝑞̃1 ) 𝜇 (𝑘 − 1)

(29)
̃ 𝑝̃0 + ⋅ ⋅ ⋅ + (𝑝̃𝑟 + 𝑞̃𝑟 ) 𝜇 (𝑘 − 𝑟) .
≥ 𝛼̃(̃
𝛼𝛽𝜂 𝑃 ̃ 𝑞̃1 𝑦 𝑞̃2 ⋅ ⋅ ⋅ 𝑦
̃ ) 𝑦 𝑝̃1 𝑦 𝑝̃2 ⋅ ⋅ ⋅ 𝑦 𝑝̃𝑟 𝛽𝑦 𝑞̃𝑟
0 −1 −𝑟+1 0 −1 −𝑟+1

If 𝑘 ∈ 𝐼 is on the end of the set 𝐼𝑛 , that is, 𝑘 ∈ 𝐼𝑛 , (𝑘 + 1) ∈ 𝐼𝑛+1 ,

̃ 𝑝̃0 ̃
≥ 𝛼̃(̃ ̃ 𝑃 ) 𝜂𝑝1 𝜂𝑝2 ⋅ ⋅ ⋅ 𝜂𝑝𝑟 𝛽𝜂
𝛼𝛽𝜂
̃ ̃ 𝑞̃1 𝜂𝑞̃2 ⋅ ⋅ ⋅ 𝜂𝑞̃𝑟
̃
we have
𝑛
= (̃ ̃
𝛼𝛽)
1+𝑝̃0 𝑃
𝜂
̃𝑝̃0 +𝑝̃1 +⋅⋅⋅+𝑝̃𝑟 +̃
𝑞1 +̃
𝑞2 +⋅⋅⋅+̃
𝑞𝑟
. 𝜇 (𝑘 + 1) ≥ 1 + 𝑝̃0 [1 + (𝑃̃ + 𝑄)
̃ + ⋅ ⋅ ⋅ + (𝑃̃ + 𝑄)
̃ ]
(23) 𝑛
+ (𝑝̃1 + 𝑞̃1 ) [1 + (𝑃̃ + 𝑄)
̃ + ⋅ ⋅ ⋅ + (𝑃̃ + 𝑄)
̃ ]
𝑛
Repeating the above procedure, for every step 𝑘 ∈ 𝑍+ , we + ⋅ ⋅ ⋅ + (𝑝̃𝑟 + 𝑞̃𝑟 ) [1 + (𝑃̃ + 𝑄)
̃ + ⋅ ⋅ ⋅ + (𝑃̃ + 𝑄)
̃ ]
can obtain the inequality of the form
𝑛
= 1 + [1 + (𝑃̃ + 𝑄)
̃ + ⋅ ⋅ ⋅ + (𝑃̃ + 𝑄)
̃ ]
󵄩󵄩 󵄩 ̃ 𝜇(𝑘) 𝜂](𝑘) ,
󵄩󵄩𝑦 (𝑘)󵄩󵄩󵄩 ≥ (̃
𝛼𝛽) (24) × [𝑝̃0 + (𝑝̃1 + 𝑞̃1 ) + ⋅ ⋅ ⋅ + (𝑝̃𝑟 + 𝑞̃𝑟 )]
𝑛+1
= 1 + (𝑃̃ + 𝑄)
̃ + ⋅ ⋅ ⋅ + (𝑃̃ + 𝑄)
̃ .
where the functions 𝜇(𝑘), ](𝑘) are defined by the following
two propositions. (30)
Abstract and Applied Analysis 5

If 𝑘 is inside 𝐼𝑛 , so that 𝑘, 𝑘 − 1, . . ., 𝑘 − 𝑖 ∈ 𝐼𝑛 and 𝑘 − 𝑖 − 1, Given (𝑃̃ + 𝑄) ̃ 𝑛−1 ≥ (𝑃̃ + 𝑄)

̃ ∈ (0, 1) : (𝑃̃ + 𝑄) ̃ 𝑛 , we obtain
𝑘 − 𝑖 − 2, . . . , 𝑘 − 𝑟 ∈ 𝐼𝑛−1 , which implies 𝑘 + 1 ∈ 𝐼𝑛+1 , then we
have ̃ 𝑛−1 + (𝑝̃1 + 𝑞̃1 ) (𝑃̃ + 𝑄)
] (𝑘 + 1) ≤ 𝑝̃0 (𝑃̃ + 𝑄) ̃ 𝑛−1

𝑛−1
̃ 𝑛] + ⋅ ⋅ ⋅ + (𝑝̃𝑖 + 𝑞̃𝑖 ) (𝑃̃ + 𝑄)
̃ + (𝑝̃𝑖+1 + 𝑞̃𝑖+1 )
𝜇 (𝑘 + 1) ≥ 1 + 𝑝̃0 [1 + (𝑃̃ + 𝑄)
̃ + ⋅ ⋅ ⋅ + (𝑃̃ + 𝑄)
𝑛−1 𝑛−1
̃ 𝑛] × (𝑃̃ + 𝑄)
̃ + ⋅ ⋅ ⋅ + (𝑝̃𝑟 + 𝑞̃𝑟 ) (𝑃̃ + 𝑄)
̃
+ (𝑝̃1 + 𝑞̃1 ) [1 + (𝑃̃ + 𝑄)
̃ + ⋅ ⋅ ⋅ + (𝑃̃ + 𝑄)
𝑛−1
+ ⋅ ⋅ ⋅ + (𝑝̃𝑖 + 𝑞̃𝑖 ) [1 + (𝑃̃ + 𝑄) ̃ 𝑛]
̃ + ⋅ ⋅ ⋅ + (𝑃̃ + 𝑄) = [𝑝̃0 + (𝑝̃1 + 𝑞̃1 ) + ⋅ ⋅ ⋅ + (𝑝̃𝑟 + 𝑞̃𝑟 )] (𝑃̃ + 𝑄)
̃
𝑛
+ (𝑝̃𝑖+1 + 𝑞̃𝑖+1 ) [1 + (𝑃̃ + 𝑄)
̃ + ⋅ ⋅ ⋅ + (𝑃̃ + 𝑄)
̃ ] 𝑛 = (𝑃̃ + 𝑄)
̃
(34)
+ ⋅ ⋅ ⋅ + (𝑝̃𝑟 + 𝑞̃𝑟 ) [1 + (𝑃̃ + 𝑄)]
̃
as desired.
Next, we will prove Theorem 6 based on the above
= 1 + [𝑝̃0 + (𝑝̃1 + 𝑞̃1 ) + ⋅ ⋅ ⋅ + (𝑝̃𝑟 + 𝑞̃𝑟 )] + (𝑃̃ + 𝑄)
̃
propositions. Let 𝑃̃ + 𝑄
̃ ∈ (0, 1). From the estimation (24)
and the estimations (27), (28) of Proposition (b), we have two
× [𝑝̃0 + (𝑝̃1 + 𝑞̃1 ) + ⋅ ⋅ ⋅ + (𝑝̃𝑟 + 𝑞̃𝑟 )] + ⋅ ⋅ ⋅
cases.
𝑛−1
+ (𝑃̃ + 𝑄)
̃ [𝑝̃0 + (𝑝̃1 + 𝑞̃1 ) + ⋅ ⋅ ⋅ + (𝑝̃𝑟 + 𝑞̃𝑟 )] Case (i). If (̃ ̃ ≥ 1, we have
𝛼𝛽)
𝑛
+ (𝑃̃ + 𝑄)
̃ [𝑝̃0 + (𝑝̃1 + 𝑞̃1 ) + ⋅ ⋅ ⋅ + (𝑝̃𝑖 + 𝑞̃𝑖 )] ̃ ̃ ̃ ̃ 𝑛
𝜇(𝑘)
̃ 1+(𝑃+𝑄)+⋅⋅⋅+(𝑃+𝑄) , ̃ ̃ 𝑛
(̃ ̃
𝛼𝛽) ≥ (̃
𝛼𝛽) 𝜂](𝑘) ≥ 𝜂(𝑃+𝑄) ,
(35)
≥ 1 + [𝑝̃0 + (𝑝̃1 + 𝑞̃1 ) + ⋅ ⋅ ⋅ + (𝑝̃𝑟 + 𝑞̃𝑟 )] ∀𝑘 ∈ 𝐼𝑛 , 𝑛 = 1, 2, . . . ,
+ (𝑃̃ + 𝑄)
̃ [𝑝̃0 + (𝑝̃1 + 𝑞̃1 ) + ⋅ ⋅ ⋅ + (𝑝̃𝑟 + 𝑞̃𝑟 )] + ⋅ ⋅ ⋅ where 𝜂 ≥ 1. Therefore,
̃ ̃ ̃ ̃ 𝑛
𝑛−1 󵄩󵄩 󵄩󵄩 ̃ 1+(𝑃+𝑄)+⋅⋅⋅+(𝑃+𝑄) 𝜂(𝑃+ ̃𝑛
̃ 𝑄)
+ (𝑃̃ + 𝑄)
̃ [𝑝̃0 + (𝑝̃1 + 𝑞̃1 ) + ⋅ ⋅ ⋅ + (𝑝̃𝑟 + 𝑞̃𝑟 )] 󵄩󵄩𝑦𝑘 󵄩󵄩 ≥ (̃
𝛼𝛽)
(36)
𝑛 ̃
̃ 𝑄)]
1/[1−(𝑃+ ̃ 𝑛+1 /[(𝑃+
̃ 𝑄)
(𝑃+ ̃
̃ 𝑄)−1]
= 1 + (𝑃̃ + 𝑄)
̃ + ⋅ ⋅ ⋅ + (𝑃̃ + 𝑄)
̃ , = (̃ ̃
𝛼𝛽) (̃ ̃
𝛼𝛽) (𝑃+ ̃𝑛
̃ 𝑄)
𝜂 ,
(31)
which gives
̃ ̃
󵄩 󵄩 ̃ 1/[1−(𝑃+𝑄)] ,
which proves the left-hand side inequality of (27). lim 󵄩󵄩󵄩𝑦𝑘 󵄩󵄩󵄩 ≥ (̃
𝛼𝛽) (37)
To prove inequality (28), using (26) and by induction, we
assume that the inequality holds for all steps 1, 2, . . . , 𝑘. For as 𝑛 and then 𝑘 go to infinity.
the step 𝑘 + 1, by the same method of induction, if 𝑘 ∈ 𝐼 and Therefore, there are a number 𝑎 > 0 and a large number
is on the end of the set 𝐼 so that 𝑘 + 1 ∈ 𝐼𝑛+1 , we have 𝐾 ∈ 𝑍+ such that ‖𝑦𝐾 ‖ ≥ 𝑎. By Definition 3, if 𝜀 = 𝑎 and
𝛿 ≥ 𝜂, the zero solutions are unstable.

] (𝑘 + 1) = 𝑝̃0 ] (𝑘) + (𝑝̃1 + 𝑞̃1 ) ] (𝑘 − 1) + ⋅ ⋅ ⋅ Case (ii). If (̃ ̃ ∈ (0, 1), using the estimation (24) and the
𝛼𝛽)
estimations (27), (28) of Proposition (b) again, we then obtain
+ (𝑝̃𝑟 + 𝑞̃𝑟 ) ] (𝑘 − 𝑟)
𝜇(𝑘) ̃ ̃ ̃ ̃ 𝑘−1
̃ 1+(𝑃+𝑄)+⋅⋅⋅+(𝑃+𝑄) , ̃ ̃ 𝑛
𝑛 𝑛 (32) (̃ ̃
𝛼𝛽) ≥ (̃
𝛼𝛽) 𝜂](𝑘) ≥ 𝜂(𝑃+𝑄) ,
≤ 𝑝̃0 (𝑃̃ + 𝑄)
̃ + (𝑝̃1 + 𝑞̃1 ) (𝑃̃ + 𝑄)
̃ + ⋅⋅⋅
𝑛 𝑛 ∀𝑘 ∈ 𝐼𝑛 , 𝑛 = 1, 2, . . . ,
+ (𝑝̃𝑟 + 𝑞̃𝑟 ) (𝑃̃ + 𝑄)
̃ = (𝑃̃ + 𝑄)
̃ .
(38)
where 𝜂 ≥ 1. Therefore,
If 𝑘 is inside 𝐼𝑛 , in other words, 𝑘, 𝑘 − 1, . . . , 𝑘 − 𝑖 ∈ 𝐼𝑛 and
̃ ̃ ̃ ̃ 𝑘−1
𝑘 − 𝑖 − 1, 𝑘 − 𝑖 − 2, . . . , 𝑘 − 𝑟 ∈ 𝐼𝑛−1 , such that 𝑘 + 1 ∈ 𝐼𝑛+1 , then 󵄩󵄩 󵄩󵄩 ̃ 1+(𝑃+𝑄)+⋅⋅⋅+(𝑃+𝑄) 𝜂(𝑃+ ̃𝑛
̃ 𝑄)
󵄩󵄩𝑦𝑘 󵄩󵄩 ≥ (̃
𝛼𝛽)
we have (39)
1/[1−(𝑃+ ̃
̃ 𝑄)] (𝑃+ ̃ 𝑘 /[(𝑃+
̃ 𝑄) ̃
̃ 𝑄)−1] ̃𝑛
̃ 𝑄)
= (̃ ̃
𝛼𝛽) (̃ ̃
𝛼𝛽) (𝑃+
𝜂 ,
̃ 𝑛 + (𝑝̃1 + 𝑞̃1 ) (𝑃̃ + 𝑄)
] (𝑘 + 1) ≤ 𝑝̃0 (𝑃̃ + 𝑄) ̃ 𝑛 + ⋅⋅⋅
which gives
𝑛 𝑛−1
+ (𝑝̃𝑖 + 𝑞̃𝑖 ) (𝑃̃ + 𝑄)
̃ + (𝑝̃𝑖+1 + 𝑞̃𝑖+1 ) (𝑃̃ + 𝑄)
̃ 󵄩 󵄩 ̃ 1/[1−(𝑃+𝑄)] ,
̃ ̃
lim 󵄩󵄩󵄩𝑦𝑘 󵄩󵄩󵄩 ≥ (̃
𝛼𝛽) (40)
𝑛−1
+ ⋅ ⋅ ⋅ + (𝑝̃𝑟 + 𝑞̃𝑟 ) (𝑃̃ + 𝑄)
̃ . as 𝑛 and then 𝑘 goes to infinity. Thus, as in Case (i), we prove
(33) the instability of zero solutions. The proof is ended.
6 Abstract and Applied Analysis

According to Theorem 6, we can analogize the following Suppose that the function 𝑓𝑘 (⋅) satisfies the conditions
theorems. 𝑘−1
∑ 𝑔𝑘,𝑠 (𝑦𝑠 , 𝑦𝑠−1 ) ≤ 𝑧 (𝑦𝑘 , 𝑦𝑘−1 , 𝑦𝑘−2 ) ,
Theorem 7. Assume that conditions (𝐶4)-(𝐶5), where we 𝑠=0
estimate inequality (24), hold in the system (2). Then, system 󵄩󵄩 󵄩 𝑝 󵄩 󵄩𝑝 (47)
󵄩󵄩𝑓𝑘 (⋅)󵄩󵄩󵄩 ≤ 𝜆‖𝑥‖ 1 󵄩󵄩󵄩𝑦󵄩󵄩󵄩 2 ‖𝑡‖ 3 ,
𝑝
(2) is unstable if 𝑃̃ + 𝑄
̃ = 1 for all (̃ ̃ > 1.
𝛼𝛽)

̃ > 1. In this case for the ∀𝑘 ∈ 𝑍+ , (𝑥, 𝑦, 𝑡, 𝑧) ∈ 𝑋 × 𝑋 × 𝑇 × 𝑋,

Proof. Let 𝑃̃ + 𝑄
̃ = 1 and (̃ 𝛼𝛽)
inequality we obtained in Theorem 6 for some 𝑝1 , 𝑝2 , 𝑝3 ∈ 𝑍+ . Then by Theorems 6 and 7, the
above system is stable under the following control 𝜑𝑘 = Φ(𝑦𝑘 )
̃ ̃ ̃ ̃ 𝑛+1 ̃
̃ 𝑄)−1]
󵄩󵄩 󵄩󵄩 ̃ 1/[1−(𝑃+𝑄)] (̃ ̃ (𝑃+𝑄) /[(𝑃+ ̃𝑛
̃ 𝑄)
(𝑃+ and one of the following conditions:
󵄩󵄩𝑦𝑘 󵄩󵄩 ≥ (̃
𝛼𝛽) 𝛼𝛽) 𝜂 , (41)
𝑝1 + 𝑝2 + 𝑝3 > 1 (48)
we have
or
󵄩󵄩 󵄩󵄩 ̃ 𝑛+1 1
󵄩󵄩𝑦𝑘 󵄩󵄩 ≥ (̃
𝛼𝛽) 𝜂, 𝑘 ∈ 𝐼𝑛 , (42) 𝑝1 + 𝑝2 + 𝑝3 = 1, . ‖Φ‖ <
(49)
𝜆
which goes to infinity as 𝑛, 𝑘 go to infinity. Example 10. We study the following system in Banach space
By Definition 3 of the instability, it is easy to see that the 𝑙2 :
zero solutions are unstable. The theorem is proved. 𝑘−1
𝑦𝑘+1 = 𝑓𝑘 (𝑦𝑘 , 𝑦𝑘−1 , 𝑦𝑘−2 , ∑ 𝑔𝑘,𝑠 (𝑦𝑠 , 𝑦𝑠−1 )) , 𝑘 ∈ 𝑍+ ,
Theorem 8. Assume that conditions (𝐶4)-(𝐶5), where we 𝑠=0
estimate inequality (24), hold in the system (2). Then, system (50)
̃ = 𝛼̃𝛽̃ = 1.
(2) is asymptotically unstable if 𝑃̃ + 𝑄
where
̃ = 𝛼̃𝛽̃ = 1, in addition to the inequality
Proof. When 𝑃̃ + 𝑄 󵄩 󵄩𝑞 󵄩 󵄩𝑞 󵄩 󵄩𝑞
𝑓𝑘 (⋅) = (𝑎0 󵄩󵄩󵄩𝑦𝑘 󵄩󵄩󵄩 0 , 𝑎1 󵄩󵄩󵄩𝑦𝑘−1 󵄩󵄩󵄩 1 , 𝑎2 󵄩󵄩󵄩𝑦𝑘−2 󵄩󵄩󵄩 2 ,
󵄩󵄩 󵄩󵄩 (51)
󵄩󵄩𝑦𝑘 󵄩󵄩 ≥ (̃ ̃ 𝑛+1 𝜂,
𝛼𝛽) (43) √𝑎3 (󵄩󵄩󵄩󵄩𝑦𝑘 + 𝑦𝑘−1 + 𝑦𝑘−2 󵄩󵄩󵄩󵄩2𝑞3 ), 0, 0, . . .)

we can obtain and 𝑎𝑖 > 0, 𝑞𝑖 > 0, 𝑖 = 0, 1, 2, 3, are real numbers. We assume

󵄩󵄩 󵄩󵄩 that
󵄩󵄩𝑦𝑘 󵄩󵄩 ≥ 𝜂, ∀𝑘 ∈ 𝑍+ , (44) 󵄩󵄩 󵄩 󵄩 󵄩 󵄩 󵄩
󵄩󵄩𝑦𝑘 + 𝑦𝑘−1 + 𝑦𝑘−2 󵄩󵄩󵄩 ≥ 󵄩󵄩󵄩𝑦𝑘−1 󵄩󵄩󵄩 + 󵄩󵄩󵄩𝑦𝑘 − 2󵄩󵄩󵄩 . (52)
which shows that the zero solutions are asymptotically Next, we can take a convex cone 𝐺 with the form of
unstable. This ends the proof.
𝐺 = {(𝑦0 , 𝑦1 , 𝑦2 , 𝑦3 , . . .) ∈ 𝑙2 : 𝑦𝑖 ≥ 0,
(53)
4. Applications 𝑖 = 0, 1, 2, 3; 𝑦𝑗 = 0, ∀𝑗 ≥ 0}
In this section, we state two examples concerning the above which is the invariant set for system.
theorems in Sections 2 and 3, respectively. By Theorems 6 and 7 and for arbitrary (𝑦𝑘 , 𝑦𝑘−1 , 𝑦𝑘−2 ,
Example 9. The difference control system with higher order ∑𝑘−1
𝑠=0 𝑔𝑘,𝑠 (𝑦𝑠 , 𝑦𝑠−1 )) ∈ 𝐺, we have
󵄩
term 󵄩󵄩󵄩𝑓𝑘 (⋅)󵄩󵄩󵄩2 = 󵄩󵄩󵄩󵄩(𝑎0 󵄩󵄩󵄩𝑦𝑘 󵄩󵄩󵄩𝑞0 , 𝑎1 󵄩󵄩󵄩𝑦𝑘−1 󵄩󵄩󵄩𝑞1 , 𝑎2 󵄩󵄩󵄩𝑦 (𝑘 − 2)󵄩󵄩󵄩𝑞2 ,
󵄩 󵄩 󵄩󵄩 󵄩 󵄩 󵄩 󵄩 󵄩 󵄩
𝑘−1
𝑦𝑘+1 = 𝑓𝑘 (𝑦𝑘 , 𝑦𝑘−1 , 𝜑𝑘−2 , ∑ 𝑔𝑘,𝑠 (𝑦𝑠 , 𝑦𝑠−1 )) , 𝑘 ∈ 𝑍+ , 󵄩󵄩2
√𝑎3 (󵄩󵄩󵄩󵄩𝑦𝑘 + 𝑦𝑘−1 + 𝑦𝑘−2 󵄩󵄩󵄩󵄩2𝑞3 ), 0, 0, . . .)󵄩󵄩󵄩 (54)
𝑠=0 󵄩󵄩
(45)
3
󵄩 󵄩2𝑝
+
where 𝑓𝑘 (⋅) : 𝑍 ×𝑌×𝑌×𝑇×𝑌, 𝜑𝑘 ∈ 𝑇, is a control argument. ≥ ∑𝑎𝑖2 󵄩󵄩󵄩𝑦𝑘−𝑖 󵄩󵄩󵄩 𝑖 ,
𝑖=0
We note that the system is stabilizable when there is a linear
bounded operator Φ : 𝑌 → 𝑇 such that the control 𝜑𝑘 = where 𝑝0 = 𝑞0 , 𝑝1 = 𝑞1 + 𝑞3 , and 𝑝2 = 𝑞2 + 𝑞3 . Thus,
Φ(𝑦𝑘 ), if the system without control 󵄩󵄩 󵄩 1/3 󵄩 󵄩𝑝 /3 󵄩 󵄩𝑝 /3 󵄩 󵄩𝑝 /3
󵄩󵄩𝑓𝑘 (⋅)󵄩󵄩󵄩 ≥ √3(𝑎0 𝑎1 𝑎2 ) 󵄩󵄩󵄩𝑦𝑘 󵄩󵄩󵄩 0 󵄩󵄩󵄩𝑦𝑘−1 󵄩󵄩󵄩 1 󵄩󵄩󵄩𝑦𝑘−2 󵄩󵄩󵄩 2 . (55)
𝑘−1
Hence, system (50) is unstable if one of the following
𝑦𝑘+1 = 𝑓𝑘 (𝑦𝑘 , 𝑦𝑘−1 , Φ𝑦𝑘−2 , ∑ 𝑔𝑘,𝑠 (𝑦𝑠 , 𝑦𝑠−1 )) (46)
conditions holds:
𝑠=0
(a󸀠 ) 𝑝0 + 𝑝1 + 𝑝2 < 3;
is asymptotically stable. (b󸀠 ) 𝑝0 + 𝑝1 + 𝑝2 = 3 and 𝑎0 𝑎1 𝑎2 > 1/√27.
Abstract and Applied Analysis 7

Conflict of Interests
The authors declare that there is no conflict of interests
regarding the publication of this paper.

Acknowledgment
The project was supported by the National Natural Science
Foundation of China 11271261, 11171178, and FP7-PEOPLE-
2012-IRSES-316338.

References
[1] R. P. Agarwal, Difference Equations and Inequalities. Theory,
Methods, and Applications, vol. 228 of Monographs and Text-
books in Pure and Applied Mathematics, Marcel Dekker, New
York, NY, USA, 2nd edition, 2000.
[2] S. Elaydi, An Introduction to Difference Equations, Springer, New
York, NY, USA, 1995.
[3] V. L. Kocić and G. Ladas, Global Behavior of Nonlinear Differ-
ence Equations of Higher Order with Applications, vol. 256 of
Mathematics and Its Applications, Kluwer Academic Publishers
Group, Dordrecht, The Netherlands, 1993.
[4] A. Y. T. Leung, J. N. Xu, and W. S. Tsui, “Nonlinear delay
difference equations for housing dynamics assuming hetero-
geneous backward-looking expectations,” Applied Mathematics
and Mechanics, vol. 28, no. 6, pp. 785–798, 2007.
[5] P. Cull, M. Flahive, and R. Robson, Difference Equations. From
Rabbits to Chaos, Springer, New York, NY, USA, 2005.
[6] V. N. Phat, Constrained Control Problems of Discrete Processes,
World Scientific, Singapore, 1996.
[7] M. M. Kipnis and D. A. Komissarova, “A note on explicit
stability conditions for autonomous higher order difference
equations,” Journal of Difference Equations and Applications, vol.
13, no. 5, pp. 457–461, 2007.
[8] S. Mohamad and K. Gopalsamy, “Exponential stability of
continuous-time and discrete-time cellular neural networks
with delays,” Applied Mathematics and Computation, vol. 135,
no. 1, pp. 17–38, 2003.
[9] J. Yao and F. Meng, “Asymptotic behavior of solutions of
certain higher order nonlinear difference equation,” Journal of
Computational and Applied Mathematics, vol. 205, no. 1, pp.
640–650, 2007.
[10] N. S. Bay and V. N. Phat, “Stability analysis of nonlinear
retarded difference equations in Banach spaces,” Computers &
Mathematics with Applications, vol. 45, no. 6-9, pp. 951–960,
2003.
[11] J. P. LaSalle, The Stability and Control of Discrete Processes, vol.
62 of Applied Mathematical Sciences, Springer, New York, NY,
USA, 1986.
[12] R. Naulin and M. Pinto, “Stability of discrete dichotomies for
linear difference systems,” Journal of Difference Equations and
Applications, vol. 3, no. 2, pp. 101–123, 1997.
[13] R. Naulin and C. Vanegas, “Instability of discrete-time systems,”
Electronic Journal of Differential Equations, vol. 33, pp. 1–11, 1998.
[14] H. M. Xiao and Y. Q. Liu, “The stability of linear time-varying
discrete systems with time-delay,” Journal of Mathematical
Analysis and Applications, vol. 188, no. 1, pp. 66–77, 1994.
[15] X. S. Yang and Y. Stepanenko, “A stability criterion for discrete
nonlinear systems with time delayed feedback,” IEEE Transac-
tions on Automatic Control, vol. 39, no. 3, pp. 585–589, 1994.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 602604, 8 pages
http://dx.doi.org/10.1155/2014/602604

Research Article
Existence of Positive Solutions for a Kind of
Fractional Boundary Value Problems

Hongjie Liu,1 Xiao Fu,2 and Liangping Qi2

1
School of Land Science and Technology, China University of Geosciences, Beijing 10083, China
2
School of Science, China University of Geosciences, Beijing 10083, China

Correspondence should be addressed to Hongjie Liu; l hongjie@163.com

Received 18 March 2014; Accepted 26 April 2014; Published 20 May 2014

Academic Editor: Yonghui Xia

Copyright © 2014 Hongjie Liu et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
𝛼
We are concerned with the following nonlinear three-point fractional boundary value problem: 𝐷0+ 𝑢 (𝑡) + 𝜆𝑎 (𝑡) 𝑓 (𝑡, 𝑢 (𝑡)) = 0,
𝛼
0 < 𝑡 < 1, 𝑢 (0) = 0, and 𝑢 (1) = 𝛽𝑢 (𝜂), where 1 < 𝛼 ≤ 2, 0 < 𝛽 < 1, 0 < 𝜂 < 1, 𝐷0+ is the standard Riemann-Liouville fractional
derivative, 𝑎 (𝑡) > 0 is continuous for 0 ≤ 𝑡 ≤ 1, and 𝑓 ≥ 0 is continuous on [0, 1] × [0, ∞) . By using Krasnoesel’skii’s fixed-point
theorem and the corresponding Green function, we obtain some results for the existence of positive solutions. At the end of this
paper, we give an example to illustrate our main results.

1. Introduction investigated the existence of positive solutions for the singular

fractional boundary value problem:
Fractional differential equations have been of great inter-
𝛼
est recently. With the development of nonlinear science, 𝐷0+ 𝑢 (𝑡) + 𝑓 (𝑡, 𝑢 (𝑡) , 𝐷𝜇 𝑢 (𝑡)) = 0, 0 < 𝑡 < 1,
the researchers found that nonlinear fractional differential (2)
equations could describe something’s changing rules more 𝑢 (0) = 𝑢 (1) = 0,
accurately. Therefore, it is significant to study nonlinear
fractional differential equations to solve the nonlinear prob- where 1 < 𝛼 ≤ 2, 𝜇 > 0 are real numbers, 𝑓 is positive, and
𝛼
lems. Recently, many researchers paid attention to existence 𝐷0+ is the standard Riemann-Liouville fractional derivative.
and multiplicity of solution of the boundary value problem By means of a fixed point theorem on cone, the existence
for fractional differential equations with different boundary of positive solutions is obtained. Delbosco and Rodino [3]
conditions, such as [1–10]. Bai and Lü [1] investigated the investigated the nonlinear Dirichlet-type problem:
existence and multiplicity of positive solutions for nonlinear
𝑥𝛼−1 (𝐷0+
𝛼
𝑦) (𝑥) = 𝑓 (𝑦 (𝑥)) , 1 < 𝛼 < 2, 0 ≤ 𝑥 ≤ 1,
fractional boundary value problem: (3)
𝑦 (0) = 𝑦 (1) = 0.
𝛼
𝐷0+ 𝑢 (𝑡) + 𝑓 (𝑡, 𝑢 (𝑡)) = 0, 0 < 𝑡 < 1,
(1) They proved that if 𝑓(𝑦) is a Lipschitzian function, then the
𝑢 (0) = 𝑢 (1) = 0, problem has at least one solution 𝑦(𝑥) in a certain subspace
of 𝐶[0, 1]. In this paper, we study the following three-point
𝛼
where 1 < 𝛼 ≤ 2 is a real number, 𝐷0+ is the standard fractional boundary value problem. Consider
Riemann-Liouville fractional derivative, and 𝑓 : [0, 1] ×
𝛼
[0, ∞) → [0, ∞) is continuous. By means of some fixed- 𝐷0+ 𝑢 (𝑡) + 𝜆𝑎 (𝑡) 𝑓 (𝑡, 𝑢 (𝑡)) = 0, 0 < 𝑡 < 1,
point theorems on cone, some existence and multiplicity (4)
results of positive solutions are obtained. Agarwal et al. [2] 𝑢 (0) = 0, 𝛽𝑢 (𝜂) = 𝑢 (1) ,
2 Abstract and Applied Analysis

𝛼
where 1 < 𝛼 ≤ 2, 0 < 𝛽 < 1, 0 < 𝜂 < 1, 𝐷0+ is the (H2) ‖𝑇𝑢‖ ≥ ‖𝑢‖, ∀𝑢 ∈ 𝐾 ∩ 𝜕Ω1 and ‖𝑇𝑢‖ ≤ ‖𝑢‖, ∀𝑢 ∈
standard Riemann-Liouville fractional derivative, 𝑎(𝑡) > 0 𝐾 ∩ 𝜕Ω2
is continuous, and 𝑓 ≥ 0 is continuous on [0, 1] × [0, ∞).
holds. Then 𝑇 has a fixed point in 𝐾 ∩ (Ω2 \ Ω1 ).
By using Krasnoesel’skii’s fixed-point theorem, we get the
existence of at least one positive solution. In the following, we present the Green function of
fractional differential equation boundary value problem.
2. Background Materials and Preliminaries
Lemma 7 (see [4]). Let 𝑦 ∈ 𝐶[0, 1], then the boundary value
For convenience of the readers, we present here the necessary problem
definitions from fractional calculus theory. These definitions 𝛼
𝐷0+ 𝑢 (𝑡) + 𝑦 (𝑡) = 0, 0 < 𝑡 < 1,
can be found in the recent literature [1–5]. (9)
𝑢 (0) = 0, 𝛽𝑢 (𝜂) = 𝑢 (1)
Definition 1. The fractional integral of order 𝛼 > 0 of a
function 𝑦 : (0, ∞) → 𝑅 is given by has a unique solution
1
𝑡
𝛼 1 𝑢 (𝑡) = ∫ 𝐺 (𝑡, 𝑠) 𝑦 (𝑠) 𝑑𝑠, (10)
𝐼0+ 𝑦 (𝑡) = ∫ (𝑡 − 𝑠)𝛼−1 𝑦 (𝑠) 𝑑𝑠, (5) 0
Γ (𝛼) 0
where
provided that the right side is pointwise defined on (0, ∞). ([𝑡(1 − 𝑠)]𝛼−1 − 𝛽𝑡𝛼−1 (𝜂 − 𝑠)𝛼−1
{
{
{
{
Definition 2. The fractional derivative of order 𝛼 > 0 of a {
{ − (𝑡 − 𝑠)𝛼−1 (1 − 𝛽𝜂𝛼−1 ))
{
{
function 𝑦 : (0, ∞) → 𝑅 is given by {
{ ×((1 − 𝛽𝜂𝛼−1 )Γ(𝛼)) ,
−1
{
{
{
{
{
{ 0 ≤ 𝑠 ≤ min {𝑡, 𝜂} ≤ 1;
1 𝑑 𝑛 𝑡 𝑦 (𝑠) {
{
𝛼
𝐷0+ 𝑦 (𝑡) = ( ) ∫ 𝑑𝑠, 𝑛 = [𝛼] + 1, {
{ [𝑡(1 − 𝑠)]𝛼−1 − (𝑡 − 𝑠)𝛼−1 (1 − 𝛽𝜂𝛼−1 )
Γ (𝑛 − 𝛼) 𝑑𝑡 𝛼−𝑛+1 {
{
0 (𝑡 − 𝑠) {
{ ,
{
{ (1 − 𝛽𝜂𝛼−1 ) Γ (𝛼)
(6)
𝐺 (𝑡, 𝑠) = { 0 < 𝜂 ≤ 𝑠 ≤ 𝑡 ≤ 1; (11)
{
{
where 𝑛 = [𝛼] + 1, provided that the right side is pointwise {
{ [𝑡(1 − 𝑠)]𝛼−1 − 𝛽𝑡𝛼−1 (𝜂 − 𝑠)𝛼−1
{
{
defined on (0, ∞). {
{ ,
{
{ (1 − 𝛽𝜂𝛼−1 ) Γ (𝛼)
{
{
{
{ 0 ≤ 𝑡 ≤ 𝑠 ≤ 𝜂 < 1;
Lemma 3. Let 𝛼 > 0, if we assume 𝑢 ∈ 𝐶(0, 1) ∩ 𝐿(0, 1), then {
{
{
{ [𝑡(1 − 𝑠)]𝛼−1
𝛼
the fractional differential equation 𝐷0+ 𝑢(𝑡) = 0 has {
{ ,
{
{
{ (1 − 𝛽𝜂𝛼−1 ) Γ (𝛼)
{
{
𝑢 (𝑡) = 𝐶1 𝑡𝛼−1 + 𝐶2 𝑡𝛼−2 + ⋅ ⋅ ⋅ 𝐶𝑁𝑡𝛼−𝑁, { 0 ≤ max {𝑡, 𝜂} ≤ 𝑠 ≤ 1.
(7)
𝐶𝑖 ∈ 𝑅, 𝑖 = 1, 2, . . . , 𝑁 Lemma 8. The function 𝐺(𝑡, 𝑠) defined by (11) satisfies
𝐺(𝑡, 𝑠) > 0 for 𝑡, 𝑠 ∈ (0, 1).
as unique solutions.
Proof. (1) For 0 ≤ 𝑠 ≤ min{𝑡, 𝜂} ≤ 1,
Lemma 4. Assume that 𝑢 ∈ 𝐶(0, 1) ∩ 𝐿(0, 1) with a fractional 𝐺 (𝑡, 𝑠)
derivative of order 𝛼 > 0 that belongs to 𝐶(0, 1) ∩ 𝐿(0, 1). Then
𝛼 𝛼
[𝑡(1 − 𝑠)]𝛼−1 − 𝛽𝑡𝛼−1 (𝜂 − 𝑠)𝛼−1 − (𝑡 − 𝑠)𝛼−1 (1 − 𝛽𝜂𝛼−1 )
𝐼0+ 𝐷0+ 𝑢 (𝑡) = 𝑢 (𝑡) + 𝐶1 𝑡𝛼−1 + 𝐶2 𝑡𝛼−2 + ⋅ ⋅ ⋅ 𝐶𝑁𝑡𝛼−𝑁 (8) = .
(1 − 𝛽𝜂𝛼−1 ) Γ (𝛼)
for some 𝐶𝑖 ∈ 𝑅, 𝑖 = 1, 2, . . . , 𝑁. (12)
𝛼−1 𝛼−1 𝛼−1 𝛼−1
Definition 5. Let 𝐸 be a real Banach space. A nonempty closed Let 𝐹(𝑡, 𝑠, 𝜂) = [𝑡(1 − 𝑠)] − 𝛽𝑡 (𝜂 − 𝑠) − (𝑡 − 𝑠) (1 −
convex set 𝐾 ⊂ 𝐸 is called cone of 𝐸 if it satisfies the following 𝛽𝜂𝛼−1 ). There is
conditions: 𝜕𝐹 (𝑡, 𝑠, 𝜂)
(1) 𝑥 ∈ 𝐾, 𝜎 > 0 implies 𝜎𝑥 ∈ 𝐾; 𝜕𝜂
(2) 𝑥 ∈ 𝐾, −𝑥 ∈ 𝐾 implies 𝑥 = 0. = −𝛽𝑡𝛼−1 (𝛼 − 1) (𝜂 − 𝑠)𝛼−2 + (𝛼 − 1) (𝑡 − 𝑠)𝛼−1 𝛽𝜂𝛼−2
Lemma 6. Let 𝐸 be a Banach space and let 𝐾 ⊂ 𝐸 be a cone = 𝛽 (𝛼 − 1) [(𝑡 − 𝑠)𝛼−1 𝜂𝛼−2 − 𝑡𝛼−1 (𝜂 − 𝑠)𝛼−2 ]
in 𝐸. Assume that Ω1 and Ω2 are open subsets of 𝐸 with 0 ∈
Ω1 , Ω1 ⊂ Ω2 . Let 𝑇 : 𝐾 ∩ (Ω2 \ Ω1 ) → 𝐾 be a completely ≤ 𝛽 (𝛼 − 1) [𝑡𝛼−1 𝜂𝛼−2 − 𝑡𝛼−1 (𝜂 − 𝑠)𝛼−2 ]
continuous operator. In addition, suppose that either
= 𝛽 (𝛼 − 1) 𝑡𝛼−1 [𝜂𝛼−2 − (𝜂 − 𝑠)𝛼−2 ] .
(H1) ‖𝑇𝑢‖ ≤ ‖𝑢‖, ∀𝑢 ∈ 𝐾 ∩ 𝜕Ω1 and ‖𝑇𝑢‖ ≥ ‖𝑢‖, ∀𝑢 ∈
𝐾 ∩ 𝜕Ω2 or (13)
Abstract and Applied Analysis 3

With 0 < 𝛽 < 1, 1 < 𝛼 ≤ 2, there is 𝜕𝐹/𝜕𝜂 < 0. Hence, Proof. To get 𝐺(𝑡, 𝑠) ≤ 𝐺(𝑠, 𝑠), firstly, we prove the following
function 𝐹(𝑡, 𝑠, 𝜂) is monotonically decreasing in 𝜂. With 0 < inequality:
𝜂 < 1, 𝐹(𝑡, 𝑠, 𝜂) > 𝐹(𝑡, 𝑠, 𝜂)min = 𝐹(𝑡, 𝑠, 1). Let 𝜂 = 1, there is
𝑡𝑎 − 𝑠𝑎 1 − 𝑠𝑎
≤ , (20)
𝐹 (𝑡, 𝑠, 1) (𝑡 − 𝑠)𝑎 (1 − 𝑠)𝑎

= [𝑡(1 − 𝑠)]𝛼−1 − 𝛽𝑡𝛼−1 (1 − 𝑠)𝛼−1 − (𝑡 − 𝑠)𝛼−1 (1 − 𝛽) where 𝑎 = 𝛼 − 1, 0 ≤ 𝑠 < 𝑡 ≤ 1. Let 𝑓(𝑡) = (𝑡𝑎 − 𝑠𝑎 )/(𝑡 − 𝑠)𝑎 ,
then
= (1 − 𝛽) {[𝑡(1 − 𝑠)]𝛼−1 − (𝑡 − 𝑠)𝛼−1 } 𝑎𝑡𝑎−1 (𝑡 − 𝑠)𝑎 − 𝑎 (𝑡𝑎 − 𝑠𝑎 ) (𝑡 − 𝑠)𝑎−1
𝑓󸀠 (𝑡) =
= (1 − 𝛽) [(𝑡 − 𝑡𝑠)𝛼−1 − (𝑡 − 𝑠)𝛼−1 ] . (𝑡 − 𝑠)2𝑎
(14) 𝑎(𝑡 − 𝑠)𝑎−1
= 2𝑎
⋅ [𝑡𝑎−1 (𝑡 − 𝑠) − (𝑡𝑎 − 𝑠𝑎 )] (21)
𝛼−1 𝛼−1 (𝑡 − 𝑠)
With 0 < 𝑠 ≤ 𝑡 < 1, 1 < 𝛼 ≤ 2, we have (𝑡 − 𝑡𝑠) − (𝑡 − 𝑠) ≥
0. With 1 − 𝛽 > 0, there is 𝐹(𝑡, 𝑠, 1) ≥ 0, then 𝐹(𝑡, 𝑠, 𝜂) > 𝑎(𝑡 − 𝑠)𝑎−1
𝐹(𝑡, 𝑠, 1) ≥ 0, there is 𝐺(𝑡, 𝑠) > 0. = ⋅ 𝑠 (𝑠𝑎−1 − 𝑡𝑎−1 ) .
(𝑡 − 𝑠)2𝑎
(2) For 0 < 𝜂 ≤ 𝑠 ≤ 𝑡 < 1,
𝛼−1 𝛼−1
With 1 < 𝛼 ≤ 2, there is 𝑎 − 1 ≤ 0; with 𝑠 < 𝑡, there is 𝑠𝑎−1 >
𝛼−1
[𝑡(1 − 𝑠)] − (𝑡 − 𝑠) (1 − 𝛽𝜂 ) 𝑡𝑎−1 ; then 𝑓󸀠 (𝑡) > 0. It means that 𝑓(𝑡) is monotonically
𝐺 (𝑡, 𝑠) = . (15)
increasing in 𝑡, then 𝑓(𝑡) ≤ 𝑓(1); we have
(1 − 𝛽𝜂𝛼−1 ) Γ (𝛼)
𝑡𝑎 − 𝑠𝑎 1 − 𝑠𝑎
Let 𝐹(𝑡, 𝑠, 𝜂) = [𝑡(1 − 𝑠)]𝛼−1 − (𝑡 − 𝑠)𝛼−1 (1 − 𝛽𝜂𝛼−1 ); it is obvious 𝑎 ≤ . (22)
that function 𝐹(𝑡, 𝑠, 𝜂) is monotonically increasing in 𝜂, when (𝑡 − 𝑠) (1 − 𝑠)𝑎
𝜂 = 0, Here we prove 𝐺(𝑡, 𝑠) ≤ 𝐺(𝑠, 𝑠).
(1) For 0 ≤ 𝑠 ≤ min{𝑡, 𝜂} ≤ 1,
𝐹(𝑡, 𝑠, 𝜂)min = 𝐹 (𝑡, 𝑠, 0)
𝐺 (𝑡, 𝑠)
= [𝑡(1 − 𝑠)]𝛼−1 − (𝑡 − 𝑠)𝛼−1 (16)
[𝑡(1 − 𝑠)]𝛼−1 − 𝛽𝑡𝛼−1 (𝜂 − 𝑠)𝛼−1 − (𝑡 − 𝑠)𝛼−1 (1 − 𝛽𝜂𝛼−1 )
𝛼−1 𝛼−1 = ,
= (𝑡 − 𝑡𝑠) − (𝑡 − 𝑠) .
(1 − 𝛽𝜂𝛼−1 ) Γ (𝛼)
With 0 ≤ 𝑠 ≤ 𝑡 < 1, 0 < 𝛼 − 1 ≤ 1, there is (𝑡 − 𝑡𝑠)𝛼−1 ≥ [𝑠(1 − 𝑠)]𝛼−1 − 𝛽𝑠𝛼−1 (𝜂 − 𝑠)𝛼−1
(𝑡 − 𝑠)𝛼−1 ; we have 𝐹(𝑡, 𝑠, 𝜂) > 𝐹(𝑡, 𝑠, 0) ≥ 0, then 𝐺(𝑡, 𝑠) > 0. 𝐺 (𝑠, 𝑠) = .
(1 − 𝛽𝜂𝛼−1 ) Γ (𝛼)
(3) For 0 < 𝑡 ≤ 𝑠 ≤ 𝜂 < 1,
(23)
𝛼−1 𝛼−1 𝛼−1
[𝑡(1 − 𝑠)] − 𝛽𝑡 (𝜂 − 𝑠) With (1 − 𝛽𝜂𝛼−1 )Γ(𝛼) > 0, 𝐺(𝑡, 𝑠) ≤ 𝐺(𝑠, 𝑠) is equivalent to
𝐺 (𝑡, 𝑠) = . (17)
(1 − 𝛽𝜂𝛼−1 ) Γ (𝛼)
𝑡𝑎 − 𝑠𝑎
[(1 − 𝑠)𝑎 − 𝛽(𝜂 − 𝑠)𝑎 ] ≤ 1 − 𝛽𝜂𝑎 (24)
Let 𝐹(𝑡, 𝑠, 𝜂) = [𝑡(1 − 𝑠)] − 𝛽𝑡𝛼−1 𝛼−1
(𝜂 − 𝑠) 𝛼−1
, then 𝐹(𝑡, 𝑠, 𝜂) is (𝑡 − 𝑠)𝑎
monotonically decreasing in 𝜂, or
𝐹(𝑡, 𝑠, 𝜂)min = 𝐹 (𝑡, 𝑠, 1) (𝑡𝑎 − 𝑠𝑎 ) [(1 − 𝑠)𝑎 − 𝛽(𝜂 − 𝑠)𝑎 ] ≤ (1 − 𝛽𝜂𝑎 ) (𝑡 − 𝑠)𝑎 . (25)

= [𝑡(1 − 𝑠)]𝛼−1 − 𝛽𝑡𝛼−1 (1 − 𝑠)𝛼−1 (18)

= (1 − 𝛽) [𝑡(1 − 𝑠)]𝛼−1 . (a) For 𝑠 = 𝑡, (𝑡𝑎 − 𝑠𝑎 )[(1 − 𝑠)𝑎 − 𝛽(𝜂 − 𝑠)𝑎 ] = (1 − 𝛽𝜂𝑎 )(𝑡 −
𝑠)𝑎 = 0;
With 0 < 𝛽 < 1, 0 < 𝑡 ≤ 𝑠 < 1, there is 𝐹(𝑡, 𝑠, 𝜂) > 0, then (b) for 𝑠 = 𝑡 = 𝜂 = 1, (𝑡𝑎 − 𝑠𝑎 )[(1 − 𝑠)𝑎 − 𝛽(𝜂 − 𝑠)𝑎 ] =
𝐹(𝑡, 𝑠, 𝜂) > 𝐹(𝑡, 𝑠, 1) ≥ 0, hence 𝐺(𝑡, 𝑠) > 0. (1 − 𝛽𝜂𝑎 )(𝑡 − 𝑠)𝑎 = 0;
(4) For 0 ≤ max{𝑡, 𝜂} ≤ 𝑠 ≤ 1, (c) for 𝑠 = 𝜂 ≤ 𝑡,
[𝑡(1 − 𝑠)]𝛼−1 𝑎
(𝑡𝑎 − 𝑠𝑎 ) [(1 − 𝑠)𝑎 − 𝛽(𝜂 − 𝑠) ] = (𝑡𝑎 − 𝜂𝑎 ) (1 − 𝜂𝑎 ) ,
𝐺 (𝑡, 𝑠) = > 0. (19)
(1 − 𝛽𝜂𝛼−1 ) Γ (𝛼) 𝑎
(26)
(1 − 𝛽𝜂𝑎 ) (𝑡 − 𝑠)𝑎 = (1 − 𝛽𝜂𝑎 ) (𝑡 − 𝜂) ,
Therefore, 𝐺(𝑡, 𝑠) > 0 for 𝑡, 𝑠 ∈ (0, 1). The proof is complete.
if 𝑡 = 𝜂, (𝑡𝑎 − 𝑠𝑎 )[(1 − 𝑠)𝑎 − 𝛽(𝜂 − 𝑠)𝑎 ] = (1 −
𝛽𝜂𝑎 )(𝑡 − 𝑠)𝑎 = 0; if 𝑡 > 𝜂, from (22), we have (𝑡𝑎 − 𝜂𝑎 )/
Lemma 9. 𝐺(𝑡, 𝑠) ≤ 𝐺(𝑠, 𝑠) for 𝑡, 𝑠 ∈ [0, 1]. (𝑡 − 𝜂)𝑎 ≤ (1 − 𝜂𝑎 )/(1 − 𝜂)𝑎 ≤ (1 − 𝛽𝜂𝑎 )/(1 − 𝜂)𝑎 ;
4 Abstract and Applied Analysis

(d) for 0 ≤ 𝑠 < 𝑡 ≤ 1, 0 ≤ 𝑠 < 𝜂 ≤ 1, if (1 − 𝑠)𝑎 − 𝛽(𝜂 − 𝑠)𝑎 < or

0; it is obvious that (𝑡𝑎 − 𝑠𝑎 )[(1 − 𝑠)𝑎 − 𝛽(𝜂 − 𝑠)𝑎 ] ≤
(1 − 𝛽𝜂𝑎 )(𝑡 − 𝑠)𝑎 . If (1 − 𝑠)𝑎 − 𝛽(𝜂 − 𝑠)𝑎 > 0, by (22), (𝑡𝑎 − 𝑠𝑎 ) (1 − 𝑠)𝑎 ≤ 𝛽 (𝑡𝑎 − 𝑠𝑎 ) (𝜂 − 𝑠)𝑎 . (35)
there is (𝜂𝑎 − 𝑠𝑎 )/(𝜂 − 𝑠)𝑎 ≤ (1 − 𝑠𝑎 )/(1 − 𝑠)𝑎 ; then
With 0 < 𝛽 < 1, 𝑠 ≤ 𝜂, there is 𝛽(𝜂 − 𝑠)𝑎 ≤ (𝜂 − 𝑠)𝑎 < (1 − 𝑠)𝑎 ;
1 − 𝑠𝑎 with 𝑡 ≤ 𝑠, there is
(𝜂 − 𝑠)𝑎 ≥ (𝜂𝑎 − 𝑠𝑎 ) . (27)
(1 − 𝑠)𝑎
(𝑡𝑎 − 𝑠𝑎 ) (1 − 𝑠)𝑎 ≤ 𝛽 (𝑡𝑎 − 𝑠𝑎 ) (𝜂 − 𝑠)𝑎 . (36)
Consider
𝑡𝑎 − 𝑠𝑎 It means that 𝐺(𝑡, 𝑠) ≤ 𝐺(𝑠, 𝑠) holds.
[(1 − 𝑠)𝑎 − 𝛽(𝜂 − 𝑠)𝑎 ] (4) For 0 < max{𝑡, 𝜂} ≤ 𝑠 ≤ 1,
(𝑡 − 𝑠)𝑎
1 − 𝑠𝑎 [𝑡(1 − 𝑠)]𝛼−1
≤ [(1 − 𝑠)𝑎 − 𝛽(𝜂 − 𝑠)𝑎 ] 𝐺 (𝑡, 𝑠) = ,
(1 − 𝑠)𝑎 (1 − 𝛽𝜂𝛼−1 ) Γ (𝛼)
(37)
1 − 𝑠𝑎 [𝑠(1 − 𝑠)]𝛼−1
= 1 − 𝑠𝑎 − 𝛽 (𝜂 − 𝑠)𝑎 𝐺 (𝑠, 𝑠) = .
(1 − 𝑠)𝑎 (28)
(1 − 𝛽𝜂𝛼−1 ) Γ (𝛼)
1 − 𝑠𝑎 (1 − 𝑠)𝑎
≤ 1 − 𝑠𝑎 − 𝛽 ⋅ ⋅ (𝜂𝑎 − 𝑠𝑎 ) It is obvious that 𝐺(𝑡, 𝑠) ≤ 𝐺(𝑠, 𝑠). The proof is complete.
(1 − 𝑠)𝑎 1 − 𝑠𝑎
= 1 − 𝛽𝜂𝑎 − (1 − 𝛽) 𝑠𝑎 Lemma 10. 𝐺(𝑡, 𝑠) ≥ 𝑞(𝑘)𝐺(𝑠, 𝑠) for 𝑡, 𝑠 ∈ [1/𝑘, 1 − (1/𝑘)],
where 𝑘 > 2 is an integer and
≤ 1 − 𝛽𝜂𝑎 .
(1 − 𝛽) [(1 − (1/𝑘))2(𝛼−1) − (1 − (2/𝑘))𝛼−1 ]
It means that 𝐺(𝑡, 𝑠) ≤ 𝐺(𝑠, 𝑠) holds. 𝑞 (𝑘) = min { ,
(2) For 0 < 𝜂 ≤ 𝑠 ≤ 𝑡 ≤ 1, (1/4)𝛼−1

[𝑡(1 − 𝑠)]𝛼−1 − (𝑡 − 𝑠)𝛼−1 (1 − 𝛽𝜂𝛼−1 ) (1 − 𝛽) (1/𝑘)2(𝛼−1)

𝐺 (𝑡, 𝑠) = , ,
(1 − 𝛽𝜂𝛼−1 ) Γ (𝛼) (1/4)𝛼−1
(29)
[𝑠(1 − 𝑠)]𝛼−1 (𝑘 − 1)𝛼−1 /𝑘2(𝛼−1)
𝐺 (𝑠, 𝑠) = . }.
(1 − 𝛽𝜂𝛼−1 ) Γ (𝛼) (1/4)𝛼−1
(38)
With (1 − 𝛽𝜂𝛼−1 )Γ(𝛼) > 0, 𝐺(𝑡, 𝑠) ≤ 𝐺(𝑠, 𝑠) is equivalent to
Proof. From (11), we have
[𝑡(1 − 𝑠)]𝑎 − (𝑡 − 𝑠)𝑎 (1 − 𝛽𝜂𝑎 ) ≤ [𝑠(1 − 𝑠)]𝑎 (30)

or { [𝑠(1 − 𝑠)]𝛼−1 − 𝛽𝑠𝛼−1 (𝜂 − 𝑠)𝛼−1

{
{ , 𝑠 ≤ 𝜂,
{ (1 − 𝛽𝜂𝛼−1 ) Γ (𝛼)
(𝑡𝑎 − 𝑠𝑎 ) (1 − 𝑠)𝑎 ≤ (𝑡 − 𝑠)𝑎 (1 − 𝛽𝜂𝑎 ) . 𝐺 (𝑠, 𝑠) = { (39)
(31) {
{ [𝑠(1 − 𝑠)]𝛼−1
{ , 𝑠 > 𝜂.
𝛼−1
From (22), we have ((𝑡𝑎 − 𝑠𝑎 )/(𝑡 − 𝑠)𝑎 )(1 − 𝑠)𝑎 ≤ 1 − 𝑠𝑎 ; with { (1 − 𝛽𝜂 ) Γ (𝛼)
𝜂 ≤ 𝑠, 0 < 𝛽 < 1, we have 1 − 𝑠𝑎 ≤ 1 − 𝜂𝑎 ≤ 1 − 𝛽𝜂𝑎 ; then
((𝑡𝑎 − 𝑠𝑎 )/(𝑡 − 𝑠)𝑎 )(1 − 𝑠)𝑎 ≤ 1 − 𝛽𝜂𝑎 ; with 𝑠 ≤ 𝑡, there is It is obvious that [𝑠(1 − 𝑠)]𝛼−1 /(1 − 𝛽𝜂𝛼−1 )Γ(𝛼) ≥
([𝑠(1 − 𝑠)]𝛼−1 − 𝛽𝑠𝛼−1 (𝜂 − 𝑠)𝛼−1 )/(1 − 𝛽𝜂𝛼−1 )Γ(𝛼), for
(𝑡𝑎 − 𝑠𝑎 ) (1 − 𝑠)𝑎 ≤ (𝑡 − 𝑠)𝑎 (1 − 𝛽𝜂𝑎 ) . (32) 𝑡, 𝑠 ∈ [1/𝑘, 1 − (1/𝑘)],
It means that 𝐺(𝑡, 𝑠) ≤ 𝐺(𝑠, 𝑠) holds. 󵄨
[𝑠(1 − 𝑠)]𝛼−1 󵄨󵄨󵄨
(3) For 0 ≤ 𝑡 ≤ 𝑠 ≤ 𝜂 < 1, max 𝐺 (𝑠, 𝑠) = 󵄨󵄨
(1 − 𝛽𝜂𝛼−1 )Γ(𝛼) 󵄨󵄨󵄨𝑠=1/2
[𝑡(1 − 𝑠)]𝛼−1 − 𝛽𝑡𝛼−1 (𝜂 − 𝑠)𝛼−1 (40)
𝐺 (𝑡, 𝑠) = , (1/4)𝛼−1
(1 − 𝛽𝜂𝛼−1 ) Γ (𝛼) = .
(33) (1 − 𝛽𝜂𝛼−1 ) Γ (𝛼)
[𝑠(1 − 𝑠)]𝛼−1 − 𝛽𝑠𝛼−1 (𝜂 − 𝑠)𝛼−1
𝐺 (𝑠, 𝑠) = . (1) For 1/𝑘 ≤ 𝑠 ≤ min{𝑡, 𝜂} ≤ 1 − (1/𝑘), let
(1 − 𝛽𝜂𝛼−1 ) Γ (𝛼)

With (1 − 𝛽𝜂𝛼−1 )Γ(𝛼) > 0, 𝐺(𝑡, 𝑠) ≤ 𝐺(𝑠, 𝑠) is equivalent to 𝐹 (𝑡, 𝑠, 𝜂) = [𝑡(1 − 𝑠)]𝛼−1 − 𝛽𝑡𝛼−1 (𝜂 − 𝑠)𝛼−1
(41)
[𝑡(1 − 𝑠)]𝑎 − 𝛽𝑡𝑎 (𝜂 − 𝑠)𝑎 ≤ [𝑠 (1 − 𝑠)]𝑎 − 𝛽𝑠𝑎 (𝜂 − 𝑠)
𝑎
(34) − (𝑡 − 𝑠)𝛼−1 (1 − 𝛽𝜂𝛼−1 ) .
Abstract and Applied Analysis 5

By Lemma 8, for 𝜂 = 1, It means that 𝐹(𝑡, 𝑠) is monotonically decreasing in 𝑡

󵄨
𝐹 (𝑡, 𝑠) = 𝐹(𝑡, 𝑠, 𝜂)min = 𝐹 (𝑡, 𝑠, 1) 𝐹 (𝑡, 𝑠) ≥ min 𝐹 (𝑡, 𝑠) = 𝐹(𝑡, 𝑠)󵄨󵄨󵄨󵄨𝑡=1−(1/𝑘),𝑠=(1/𝑘)

= [𝑡(1 − 𝑠)]𝛼−1 − 𝛽𝑡𝛼−1 (1 − 𝑠)𝛼−1 1 2(𝛼−1) 2 𝛼−1

= (1 − ) − (1 − ) ,
𝑘 𝑘
− (𝑡 − 𝑠)𝛼−1 (1 − 𝛽)
(1 − 𝛽) [(1 − (1/𝑘))2(𝛼−1) − (1 − (2/𝑘))𝛼−1 ]
𝛼−1 𝛼−1 (42) min 𝐺 (𝑡, 𝑠) = ,
= (1 − 𝛽) {[𝑡(1 − 𝑠)] − (𝑡 − 𝑠) }
(1 − 𝛽𝜂𝛼−1 ) Γ (𝛼)
= (1 − 𝛽) [(𝑡 − 𝑡𝑠)𝛼−1 − (𝑡 − 𝑠)𝛼−1 ] , 2(𝛼−1)
− (1 − (2/𝑘))𝛼−1 ]
min 𝐺 (𝑡, 𝑠) (1 − 𝛽) [(1 − (1/𝑘))
= .
𝜕𝐹 max 𝐺 (𝑠, 𝑠) (1/4)𝛼−1
= (1 − 𝛽) (𝛼 − 1) [(𝑡 − 𝑠)𝛼−2 − 𝑡𝛼−1 (1 − 𝑠)𝛼−2 ] . (48)
𝜕𝑠

With 1 < 𝛼 ≤ 2, 0 ≤ 𝑠 ≤ 𝑡 ≤ 1, then (𝑡 − 𝑠)𝛼−2 − 𝑡𝛼−1 (1 − (3) For 1/𝑘 ≤ 𝑡 ≤ 𝑠 ≤ 𝜂 < 1 − (1/𝑘), let
𝑠)𝛼−2 > 0; with 0 < 𝛽 < 1, 𝜕𝐹/𝜕𝑠 > 0. It means that 𝐹(𝑡, 𝑠) is
𝐹 (𝑡, 𝑠, 𝜂) = [𝑡(1 − 𝑠)]𝛼−1 − 𝛽𝑡𝛼−1 (𝜂 − 𝑠)𝛼−1 . (49)
monotonically increasing in 𝑠. Similarly,
By Lemma 8, for 𝜂 = 1,
𝜕𝐹
= (1 − 𝛽) (𝛼 − 1) [(𝑡 − 𝑡𝑠)𝛼−2 (1 − 𝑠) − (𝑡 − 𝑠)𝛼−2 ] < 0.
𝜕𝑡 𝐹 (𝑡, 𝑠) = 𝐹(𝑡, 𝑠, 𝜂)min = 𝐹 (𝑡, 𝑠, 1)
(43)
= [𝑡(1 − 𝑠)]𝛼−1 − 𝛽𝑡𝛼−1 (1 − 𝑠)𝛼−1
It means that 𝐹(𝑡, 𝑠) is monotonically decreasing in 𝑡. There- (50)
fore = (1 − 𝛽) [𝑡 (1 − 𝑠)]𝛼−1 ,
󵄨
𝐹 (𝑡, 𝑠) ≥ min 𝐹 (𝑡, 𝑠) = 𝐹(𝑡, 𝑠)󵄨󵄨󵄨󵄨𝑡=1−(1/𝑘),𝑠=1/𝑘 𝜕𝐹
= (𝛼 − 1) 𝑡𝛼−1 (𝛽 − 1) (1 − 𝑠)𝛼−2 < 0;
𝜕𝑠
1 2(𝛼−1) 2 𝛼−1
= (1 − 𝛽) [(1 − ) − (1 − ) ] , then 𝐹(𝑡, 𝑠) is monotonically decreasing in 𝑠. Consider
𝑘 𝑘
𝜕𝐹
min 𝐺 (𝑡, 𝑠) = (𝛼 − 1) (1 − 𝛽) 𝑡𝛼−2 (1 − 𝑠)𝛼−1 > 0; (51)
𝜕𝑡
𝛼−1
(1 − 𝛽) [((1/𝑘) − (1/𝑘2 )) − (1 − (2/𝑘))𝛼−1 ] then 𝐹(𝑡, 𝑠) is monotonically increasing in 𝑡. Therefore,
= ,
(1 − 𝛽𝜂𝛼−1 ) Γ (𝛼)
󵄨 1−𝛽
𝐹 (𝑡, 𝑠) ≥ min 𝐹 (𝑡, 𝑠) = 𝐹(𝑡, 𝑠)󵄨󵄨󵄨󵄨𝑡=1/𝑘,𝑠=1−(1/𝑘) = 2(𝛼−1) ,
min 𝐺 (𝑡, 𝑠) (1 − 𝛽) [(1 − (1/𝑘))
2(𝛼−1)
− (1 − (2/𝑘)) 𝛼−1
] 𝑘
= .
max 𝐺 (𝑠, 𝑠) (1/4) 𝛼−1
(1 − 𝛽) (1/𝑘2(𝛼−1) )
(44) min 𝐺 (𝑡, 𝑠) = , (52)
(1 − 𝛽𝜂𝛼−1 ) Γ (𝛼)
(2) For 1/𝑘 < 𝜂 ≤ 𝑠 ≤ 𝑡 ≤ 1 − (1/𝑘), let 2(𝛼−1)
min 𝐺 (𝑡, 𝑠) (1 − 𝛽) (1/𝑘)
= .
𝐹 (𝑡, 𝑠, 𝜂) = [𝑡(1 − 𝑠)]𝛼−1 − (𝑡 − 𝑠)𝛼−1 (1 − 𝛽𝜂𝛼−1 ) . (45) max 𝐺 (𝑠, 𝑠) (1/4)𝛼−1
(4) For 1/𝑘 ≤ max{𝑡, 𝜂} ≤ 𝑠 ≤ 1 − (1/𝑘), let
By Lemma 8, for 𝜂 = 0,
𝐹 (𝑡, 𝑠) = [𝑡(1 − 𝑠)]𝛼−1 . (53)
𝐹 (𝑡, 𝑠) = 𝐹(𝑡, 𝑠, 𝜂)min = 𝐹 (𝑡, 𝑠, 0)
It is obvious that 𝐹(𝑡, 𝑠) is monotonically increasing in 𝑡 and
= [𝑡(1 − 𝑠)]𝛼−1 − (𝑡 − 𝑠)𝛼−1 monotonically decreasing in 𝑠; then
(46)
= (𝑡 − 𝑡𝑠)𝛼−1 − (𝑡 − 𝑠)𝛼−1 , 𝐹 (𝑡, 𝑠) ≥ min 𝐹 (𝑡, 𝑠) = 𝐹(𝑡, 𝑠)|𝑡=1/𝑘,𝑠=1−(1/𝑘)

1 1 𝛼−1 (54)
𝜕𝐹
= (𝛼 − 1) [(𝑡 − 𝑠)𝛼−2 − 𝑡𝛼−1 (1 − 𝑠)𝛼−2 ] > 0. = ( − 2) ;
𝜕𝑠 𝑘 𝑘

It means that 𝐹(𝑡, 𝑠) is monotonically increasing in 𝑠 we have

𝛼−1
((1/𝑘) − (1/𝑘2 ))
𝜕𝐹 min 𝐺 (𝑡, 𝑠) = ; (55)
= (𝛼 − 1) [(1 − 𝑠)𝛼−1 𝑡𝛼−2 − (𝑡 − 𝑠)𝛼−2 ] < 0. (47)
(1 − 𝛽𝜂𝛼−1 ) Γ (𝛼)
𝜕𝑡
6 Abstract and Applied Analysis

there is (2) It is obvious that 𝑢(𝑡) is the positive solution of BVP (4) if
𝛼−1 2(𝛼−1) and only if
min 𝐺 (𝑡, 𝑠) (𝑘 − 1) /𝑘
= . (56)
max 𝐺 (𝑠, 𝑠) (1/4)𝛼−1 1
Let 𝑢 (𝑡) = 𝜆 ∫ 𝐺 (𝑡, 𝑠) 𝑎 (𝑠) 𝑓 (𝑠, 𝑢 (𝑠)) 𝑑𝑠, 0 ≤ 𝑡 ≤ 1. (61)
0
(1 − 𝛽) [(1 − (1/𝑘))2(𝛼−1) − (1 − (2/𝑘))𝛼−1 ]
𝑞 (𝑘) = min { ,
(1/4)𝛼−1 It can be proved by the definition of integral operator 𝑇.

(1 − 𝛽) (1/𝑘)2(𝛼−1)
, 3. Main Results
(1/4)𝛼−1
We denote some important constants as follows:
(𝑘 − 1)𝛼−1 /𝑘2(𝛼−1)
}.
(1/4)𝛼−1 1
(57) 𝐴 = ∫ 𝐺 (𝑠, 𝑠) 𝑎 (𝑠) 𝑞 (𝑘) 𝑑𝑠,
0
Therefore, 𝐺(𝑡, 𝑠)/𝐺(𝑠, 𝑠) ≥ min 𝐺(𝑡, 𝑠)/ max 𝐺(𝑠, 𝑠) = 𝑞(𝑘) ≠ 1
0. 𝐵 = ∫ 𝐺 (𝑠, 𝑠) 𝑎 (𝑠) 𝑑𝑠,
The proof is complete. 0

We consider the Banach space 𝑋 = 𝐶[0, 1] equipped with 𝑓 (𝑡, 𝑢 (𝑡))

𝐹0 = lim+ sup max ,
standard norm ‖𝑢‖ = max0≤𝑡≤1 |𝑢(𝑡)|, 𝑢 ∈ 𝑋. We define a 𝑢→0 0≤𝑡≤1 𝑢 (𝑡)
cone 𝑃 by (62)
𝑓 (𝑡, 𝑢 (𝑡))
𝑓0 = lim inf min ,
𝑃 = {𝑢 ∈ 𝑋 : 𝑢 (𝑡) ≥ 0, 𝑡 ∈ [0, 1] ,
𝑢 → +∞ 0≤𝑡≤1 𝑢 (𝑡)

(58) 𝑓 (𝑡, 𝑢 (𝑡))

𝐹∞ = lim+ sup max ,
min 𝑢 (𝑡) ≥ 𝑞 (𝑘) ‖𝑢‖} . 𝑢→0 0≤𝑡≤1 𝑢 (𝑡)
𝑡∈[1/𝑘,1−1/𝑘]
𝑓 (𝑡, 𝑢 (𝑡))
Define an integral operator 𝑇 : 𝑃 → 𝑋 by 𝑓∞ = lim inf min .
𝑢 → +∞ 0≤𝑡≤1 𝑢 (𝑡)
1
(𝑇𝑢) (𝑡) = 𝜆 ∫ 𝐺 (𝑡, 𝑠) 𝑎 (𝑠) 𝑓 (𝑠, 𝑢 (𝑠)) 𝑑𝑠,
0 (59) Here we assume that 1/𝐴𝑓∞ = 0 if 𝑓∞ → ∞, 1/𝐵𝐹0 = ∞ if
𝐹0 → 0, 1/𝐴𝑓0 = 0 if 𝑓0 → ∞, and 1/𝐵𝐹∞ = ∞ if 𝐹∞ → 0.
0 ≤ 𝑡 ≤ 1, 𝑢 ∈ 𝑃.
Theorem 12. Suppose that 𝐴𝑓∞ > 𝐵𝐹0 , then for each 𝜆 ∈
Lemma 11. It holds the following.
(1/𝐴𝑓∞ , 1/𝐵𝐹0 ), BVP (4) has at least one positive solution.
(1) 𝑇 : 𝑃 → 𝑃 is completely continuous.
Proof. We choose 𝜀 > 0 sufficiently small such that (𝐹0 +
(2) 𝑢(𝑡) is a positive solution of the fractional boundary
𝜀)𝜆𝐵 ≤ 1. By the definition of 𝐹0 , we can see that there exists
value problem (4) if and only if 𝑢(𝑡) is a fixed point of
𝑙1 > 0, such that 𝑓(𝑡, 𝑢(𝑡)) ≤ (𝐹0 + 𝜀)𝑢(𝑡) for 0 ≤ 𝑢 ≤ 𝑙1 . For
the operator 𝑇 in cone 𝑃.
𝑢 ∈ 𝑃 with ‖𝑢‖ = 𝑙1 , we have
Proof. For 𝑢 ∈ 𝑃, 𝑡 ∈ [1/𝑘, 1 − (1/𝑘)], by Lemma 10, 𝐺(𝑡, 𝑠) ≥
𝑞(𝑘)𝐺(𝑠, 𝑠); for 𝑡, 𝑠 ∈ [1/𝑘, 1 − (1/𝑘)], 1
‖𝑇𝑢 (𝑡)‖ = 𝜆 ∫ 𝐺 (𝑠, 𝑠) 𝑎 (𝑠) 𝑓 (𝑠, 𝑢 (𝑠)) 𝑑𝑠
1 0
𝑇𝑢 (𝑡) = 𝜆 ∫ 𝐺 (𝑡, 𝑠) 𝑎 (𝑠) 𝑓 (𝑠, 𝑢 (𝑠)) 𝑑𝑠 1
0
≤ 𝜆 ∫ 𝐺 (𝑠, 𝑠) 𝑎 (𝑠) (𝐹0 + 𝜀) 𝑢 (𝑠) 𝑑𝑠
1 0
≥ 𝜆𝑞 (𝑘) ∫ 𝐺 (𝑠, 𝑠) 𝑎 (𝑠) 𝑓 (𝑠, 𝑢 (𝑠)) 𝑑𝑠 1 (63)
0
≤ 𝜆 (𝐹0 + 𝜀) ‖𝑢‖ ∫ 𝐺 (𝑠, 𝑠) 𝑎 (𝑠) 𝑑𝑠
1 0
= 𝜆𝑞 (𝑘) max ∫ 𝐺 (𝑡, 𝑠) 𝑎 (𝑠) 𝑓 (𝑠, 𝑢 (𝑠)) 𝑑𝑠
1/𝑘≤𝑡≤1−1/𝑘 0 = 𝜆𝐵 (𝐹0 + 𝜀) ‖𝑢‖
= 𝑞 (𝑘) ‖𝑇𝑢 (𝑡)‖ . ≤ ‖𝑢‖ .
(60)
It is obvious that min1/𝑘≤𝑡≤1−1/𝑘 (𝑇𝑢)(𝑡) ≥ 𝑞(𝑘)‖𝑇𝑢‖, 𝑇𝑃 ⊆ 𝑃. Then we have ‖𝑇𝑢‖ ≤ ‖𝑢‖. Thus if we let Ω1 = {𝑢 ∈ 𝑋 : ‖𝑢‖ <
Thus 𝑇𝑃 ⊂ 𝑃. In addition, standard arguments show that 𝑇 is 𝑙1 }, then ‖𝑇𝑢‖ ≤ ‖𝑢‖ for 𝑢 ∈ 𝑃 ∩ 𝜕Ω1 . We choose 𝛿 > 0 and
completely continuous. 𝑐 ∈ (0, 1/4) such that 𝜆(𝑓∞ − 𝛿)𝐴 ≥ 1. There exists 𝑙2 > 𝑙1 > 0,
Abstract and Applied Analysis 7

such that 𝑓(𝑡, 𝑢) ≥ (𝑓∞ − 𝛿)𝑢 for 𝑢 > 𝑙2 . Therefore, for each 4. An Example
𝑢 ∈ 𝑃 with ‖𝑢‖ = 𝑙2 , we have
Example 1. Consider the following three-point fractional
1 boundary value problem:
‖𝑇𝑢 (𝑡)‖ = 𝜆 ∫ 𝐺 (𝑠, 𝑠) 𝑎 (𝑠) 𝑓 (𝑠, 𝑢 (𝑠)) 𝑑𝑠
0
3/2
𝐷0+ 𝑢 (𝑡) + 𝜆𝑓 (𝑡, 𝑢 (𝑡)) = 0, 0 < 𝑡 < 1,
1
1 1 (67)
≥ 𝜆 ∫ 𝐺 (𝑠, 𝑠) 𝑎 (𝑠) (𝑓∞ − 𝛿) 𝑢 (𝑠) 𝑑𝑠
𝑐 𝑢 (0) = 0, 𝑢 (1) = 𝑢 ( ) ,
2 3
1
(64)
≥ 𝜆 (𝑓∞ − 𝛿) ‖𝑢‖ ∫ 𝐺 (𝑠, 𝑠) 𝑎 (𝑠) 𝑞 (𝑘) 𝑑𝑠 where 𝑓(𝑡, 𝑢(𝑡)) = ((𝑡 + 1)(2014𝑢3 + 𝑢)/(𝑢2 + 1)) + 𝑢(sin 𝑢 +
𝑐 1), 𝑎(𝑡) = 1.
By calculations,
≥ 𝜆 (𝑓∞ − 𝛿) 𝐴 ⋅ ‖𝑢‖
≥ ‖𝑢‖ . { [𝑠(1 − 𝑠)]𝛼−1 − 𝛽𝑠𝛼−1 (𝜂 − 𝑠)𝛼−1
{
{ ,
{
{ (1 − 𝛽𝜂𝛼−1 ) Γ (𝛼)
{
{
Thus if we let Ω2 = {𝑢 ∈ 𝐸 : ‖𝑢‖ < 𝑙2 }, then Ω1 ⊂ { 0 ≤ 𝑠 = 𝑡 ≤ 𝜂 ≤ 1,
𝐺 (𝑠, 𝑠) = {
Ω2 and ‖𝑇𝑢‖ ≥ ‖𝑢‖ for 𝑢 ∈ 𝑃 ∩ 𝜕Ω2 . Condition (H1) {
{ [𝑠(1 − 𝑠)]𝛼−1
{
{ ,
of Krasnoesel’skii’s fixed point theorem is satisfied. So there { 𝛼−1
exists a fixed point of 𝑇 in 𝑃. This completes the proof. { (1 − 𝛽𝜂 ) Γ (𝛼)
{
{ 0 ≤ 𝜂 ≤ 𝑡 = 𝑠 ≤ 1,
Theorem 13. Suppose that 𝐴𝑓0 > 𝐵𝐹∞ , then for each 𝜆 ∈ 𝛼−1
1 [𝑠 (1 − 𝑠)]𝛼−1 − 𝛽𝑠𝛼−1 (𝜂 − 𝑠)
𝜂
(1/𝐴𝑓0 , 1/𝐵𝐹∞ ), BVP (4) has at least one positive solution. ∫ 𝐺 (𝑠, 𝑠) 𝑑𝑠 = ∫ 𝑑𝑠
0 0 (1 − 𝛽𝜂𝛼−1 ) Γ (𝛼)
Proof. Choose 𝜀 > 0 sufficiently small such that (𝑓0 − 𝜀)𝜆𝐴 ≥
1. From the definition of 𝑓0 , we see that there exists 𝑙1 > 0,
1
[𝑠 (1 − 𝑠)]𝛼−1
+∫ 𝑑𝑠
such that 𝑓(𝑡, 𝑢) ≥ (𝑓0 − 𝜀)𝑢 for 0 < 𝑢 ≤ 𝑙1 . If 𝑢 ∈ 𝑃 with 𝜂 (1 − 𝛽𝜂𝛼−1 ) Γ (𝛼)
‖𝑢‖ = 𝑙1 , we have
1
[𝑠 (1 − 𝑠)]𝛼−1
1 =∫ 𝑑𝑠
0 (1 − 𝛽𝜂𝛼−1 ) Γ (𝛼)
‖𝑇𝑢 (𝑡)‖ = 𝜆 ∫ 𝐺 (𝑠, 𝑠) 𝑎 (𝑠) 𝑓 (𝑠, 𝑢 (𝑠)) 𝑑𝑠
0 𝛼−1
𝜂 𝛽𝑠𝛼−1 (𝜂 − 𝑠)
(65) −∫ 𝑑𝑠
≥ 𝜆 (𝑓0 − 𝜀) ‖𝑢‖ 𝐴 0 (1 − 𝛽𝜂𝛼−1 ) Γ (𝛼)
1
≥ ‖𝑢‖ . =
1 − (1/2) (1/3)1/2 Γ (3/2)
Let Ω1 = {𝑢 ∈ 𝑋 : ‖𝑢‖ < 𝑙1 }; then we have ‖𝑇𝑢‖ ≥ ‖𝑢‖ 1
for 𝑢 ∈ 𝑃 ∩ 𝜕Ω1 . Choose 𝛿 > 0, 𝑐 ∈ (0, 1/4); then we have × {∫ [𝑠(1 − 𝑠)]1/2 𝑑𝑠
𝜆(𝐹∞ + 𝛿)𝐴 ≤ 1. There exists 𝑙2 > 𝑙1 > 0, such that 𝑓(𝑡, 𝑢) ≤ 0
(𝐹∞ + 𝛿)𝑢 for 𝑢 > 𝑙2 . For each 𝑢 ∈ 𝑃 with ‖𝑢‖ = 𝑙2 , we have 1/3
1 1/2
−∫ 𝑠 (𝜂 − 𝑠)1/2 𝑑𝑠}
1 0 2
‖𝑇𝑢 (𝑡)‖ = 𝜆 ∫ 𝐺 (𝑠, 𝑠) 𝑎 (𝑠) 𝑓 (𝑠, 𝑢 (𝑠)) 𝑑𝑠
0 = 0.58832.
1 (68)
≤ 𝜆 ∫ 𝐺 (𝑠, 𝑠) 𝑎 (𝑠) (𝐹∞ + 𝛿) 𝑢 (𝑠) 𝑑𝑠
𝑐 Let 𝑘 = 10, then
1
(66)
≤ 𝜆 (𝐹∞ + 𝛿) ‖𝑢‖ ∫ 𝐺 (𝑠, 𝑠) 𝑎 (𝑠) 𝑞 (𝑘) 𝑑𝑠 (1 − 𝛽) [(1 − (1/𝑘))2(𝛼−1) − (1 − (2/𝑘))𝛼−1 ]
𝑐 𝑞 (𝑘) min { ,
(1/4)𝛼−1
≤ 𝜆 (𝐹∞ + 𝛿) 𝐴 ⋅ ‖𝑢‖
(1 − 𝛽) (1/𝑘)2(𝛼−1)
,
≤ ‖𝑢‖ . (1/4)𝛼−1

Let Ω2 = {𝑢 ∈ 𝐸 : ‖𝑢‖ < 𝑙2 }; then Ω1 ⊂ Ω2 , and (𝑘 − 1)𝛼−1 /𝑘2(𝛼−1)

we have ‖𝑇𝑢‖ ≤ ‖𝑢‖ for 𝑢 ∈ 𝑃 ∩ 𝜕Ω2 . Condition (H2) }.
(1/4)𝛼−1
of Krasnoesel’skii’s fixed-point theorem is satisfied. So there
exists a fixed point of 𝑇 in 𝑃. This completes the proof. = 0.00557,
8 Abstract and Applied Analysis

(𝑡 + 1) (2014𝑢2 + 1) [7] X. Xu, D. Jiang, and C. Yuan, “Multiple positive solutions for the
𝑓∞ = lim inf min [ + sin 𝑢 + 1] boundary value problem of a nonlinear fractional differential
𝑢 → +∞ 0≤𝑡≤1 𝑢2 +1
equation,” Nonlinear Analysis: Theory, Methods & Applications,
vol. 71, no. 10, pp. 4676–4688, 2009.
= 2014,
[8] B. Ahmad, “Existence of solutions for irregular boundary value
(𝑡 + 1) (2014𝑢2 + 1) problems of nonlinear fractional differential equations,” Applied
𝐹0 = lim+ sup max [ + sin 𝑢 + 1] Mathematics Letters, vol. 23, no. 4, pp. 390–394, 2010.
𝑢→0 0≤𝑡≤1 𝑢2 + 1 [9] S. Liang and J. Zhang, “Positive solutions for boundary value
problems of nonlinear fractional differential equation,” Nonlin-
= 4. ear Analysis: Theory, Methods & Applications, vol. 71, no. 11, pp.
(69) 5545–5550, 2009.
[10] O. K. Jaradat, A. Al-Omari, and S. Momani, “Existence of the
Consider mild solution for fractional semilinear initial value problems,”
Nonlinear Analysis: Theory, Methods & Applications, vol. 69, no.
1
9, pp. 3153–3159, 2008.
𝐴 = ∫ 𝐺 (𝑠, 𝑠) 𝑎 (𝑠) 𝑞 (𝑘) 𝑑𝑠 = 0.00328, (70)
0
1
𝐵 = ∫ 𝐺 (𝑠, 𝑠) 𝑎 (𝑠) 𝑑𝑠 = 0.58832. (71)
0

The condition 𝐴𝑓∞ > 𝐵𝐹0 is obtained. From Theorem 12,

we see that if 𝜆 ∈ (0.15138, 0.42494), the problem (67) has
a positive solution.

Conflict of Interests
The authors declare that there is no conflict of interests
regarding the publication of this paper.

Acknowledgments
The authors are grateful to the reviewers for the valuable
suggestions. This paper was supported by the Fundamental
Research Funds for the Central Universities (no. 2652012141)
and Beijing Higher Education Young Elite Teacher Project.

References
[1] Z. Bai and H. Lü, “Positive solutions for boundary value
problem of nonlinear fractional differential equation,” Journal
of Mathematical Analysis and Applications, vol. 311, no. 2, pp.
495–505, 2005.
[2] R. P. Agarwal, D. O’Regan, and S. Staněk, “Positive solutions for
Dirichlet problems of singular nonlinear fractional differential
equations,” Journal of Mathematical Analysis and Applications,
vol. 371, no. 1, pp. 57–68, 2010.
[3] D. Delbosco and L. Rodino, “Existence and uniqueness for a
nonlinear fractional differential equation,” Journal of Mathe-
matical Analysis and Applications, vol. 204, no. 2, pp. 609–625,
1996.
[4] Z. Bai, “On positive solutions of a nonlocal fractional boundary
value problem,” Nonlinear Analysis: Theory, Methods & Applica-
tions, vol. 72, no. 2, pp. 916–924, 2010.
[5] R. P. Agarwal, M. Benchohra, and S. Hamani, “A survey on
existence results for boundary value problems of nonlinear frac-
tional differential equations and inclusions,” Acta Applicandae
Mathematicae, vol. 109, no. 3, pp. 973–1033, 2010.
[6] M. El-Shahed, “Positive solutions for boundary value problem
of nonlinear fractional differential equation,” Abstract and
Applied Analysis, vol. 2007, Article ID 10368, 8 pages, 2007.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 613648, 7 pages
http://dx.doi.org/10.1155/2014/613648

Research Article
Travelling Waves of an n-Species Food Chain Model with Spatial
Diffusion and Time Delays

Fei Hu, Yuyin Xu, Z. Wang, and Wei Ding

Department of Mathematics, Shanghai Normal University, Shanghai 200234, China

Correspondence should be addressed to Wei Ding; dingwei@shnu.edu.cn

Received 21 February 2014; Accepted 28 March 2014; Published 20 May 2014

Academic Editor: Yonghui Xia

Copyright © 2014 Fei Hu et al. This is an open access article distributed under the Creative Commons Attribution License, which
permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

We investigate an n-species food chain model with spatial diffusion and time delays. By using Schauder’s fixed point theorem, we
obtain the result about the existence of the travelling wave solutions of the food chain model with reaction term satisfying the partial
quasimonotonicity conditions.

1. Introduction diffusion cooperative Lotka-Volterra systems with delays. In

[20], Ma used Schauder’s fixed point theorem to study the
In the past few decades, the dynamic relationship between existence of travelling wave solutions to reaction-diffusion
predators and preys has been investigated intensively due to systems with quasimonotonicity reaction terms. In 2010,
its universal existence and importance in both ecology and Gan et al. [21] investigated the existence of travelling
mathematical ecology [1–3]. wave solutions to the following three-species food chain
Recently, traveling waves for reaction-diffusion systems models:
(see, e.g., [4–7]) have received considerable attention since
they determine the long-term behavior of other solutions
in many situations. By now, many powerful methods have
been used to study the travelling wave solutions for reaction- 𝜕𝑢 𝜕2 𝑢
= 𝐷1 2 + 𝑢 (𝑡, 𝑥) (𝑟1 − 𝑎11 𝑢 (𝑡, 𝑥) − 𝑎12 V (𝑡, 𝑥)) ,
diffusion systems, like phase plane techniques in [8], degree 𝜕𝑡 𝜕𝑥
theory methods [9, 10], the shooting methods [11], the
𝜕V 𝜕2 V
monotone iteration [1], and so on [12, 13]. = 𝐷2 2 + V (𝑡, 𝑥) (−𝑟2 + 𝑎21 V (𝑡 − 𝜏, 𝑥)
Although the existence of travelling wave solutions to 𝜕𝑡 𝜕𝑥
reaction-diffusion systems without delay has been widely
−𝑎22 V (𝑡, 𝑥) − 𝑎23 𝑤 (𝑡, 𝑥)) ,
studied [14–16], delayed reaction-diffusion systems which
are more realistic in population dynamic and biological
𝜕𝑤 𝜕2 𝑤
models are much more complicated than ordinary sys- = 𝐷3 2 + 𝑤 (𝑡, 𝑥) (−𝑟3 + 𝑎32 V (𝑡 − 𝜏, 𝑥) − 𝑎33 𝑤 (𝑡, 𝑥)) .
tems. Recently, a number of researchers have studied the 𝜕𝑡 𝜕𝑥
(1)
existence of travelling wave solutions in delayed reaction-
diffusion systems. In [17, 18], Wu and Zou considered delayed
reaction-diffusion systems with reaction terms satisfying
the so-called quasimonotonicity or exponential quasimono- They obtain the existence of traveling waves by using
tonicity conditions. In [19], Huang and Zou employed the Schauder’s fixed point theorem with the reaction term
upper-lower solution technique and the monotone to study satisfying the partial quasimonotonicity conditions instead
the existence of travelling wave solutions for a class of of the quasimonotonicity conditions.
2 Abstract and Applied Analysis

Motivated by the above papers, in this paper, we investi- (1) 𝐻𝑛 is nondecreasing. That is, for 0 ≤ 𝜙 ≤ 𝜙𝑗 ≤ 𝜙𝑗 ≤
𝑗
gate an n-species food chain model with spatial diffusion and 𝑘𝑗 (𝑗 = 1, 2, . . . , 𝑛),
time delays:

𝑛
𝐻𝑛 (𝜙 , 𝜙 , . . . , 𝜙 )
𝜕𝑢𝑖 (𝑡, 𝑥) 𝜕2 𝑢𝑖 (𝑡, 𝑥) 1 2 𝑛
(7)
= 𝐷𝑖 2
− 𝑎𝑖 𝑢𝑖 (𝑡, 𝑥) + ∑𝑤𝑖𝑗 𝑓𝑗 (𝑢𝑗 (𝑡, 𝑥))
𝜕𝑡 𝜕𝑥 𝑗=1 ≤ 𝐻𝑛 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) ≤ 𝐻𝑛 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) .
𝑛
+ ∑ℎ𝑖𝑗 𝑔𝑗 (𝑢𝑗 (𝑡 − 𝜏𝑗 (𝑡) , 𝑥)) , (2) For 𝑡 ∈ 𝑅, and 0 ≤ 𝜙 ≤ 𝜙𝑗 ≤ 𝜙𝑗 ≤ 𝑘𝑗 (𝑗 = 1, 2, . . . , 𝑛),
𝑗
𝑗=1
(2) 𝐻𝑖 (𝜙 , 𝜙 , . . . , 𝜙 , 𝜙𝑖+1 , 𝜙 , . . . , 𝜙 ) (𝑡)
1 2 𝑖 𝑖+2 𝑛

where 𝐷𝑖 , 𝑎𝑖 , 𝑤𝑖𝑗 , and ℎ𝑖𝑗 are positive constants and 𝑖 = ≤ 𝐻𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) (𝑡)
1, 2, . . . , 𝑛. (8)
The main propose of this paper is to obtain the sufficient ≤ 𝐻𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑖 , 𝜙 , 𝜙𝑖+2 , . . . , 𝜙𝑛 ) (𝑡) ,
𝑖+1
condition for the existence of the travelling wave solutions of
system (2) by employing Schauder’s fixed point theorem. This 𝑖 = 1, 2, . . . , 𝑛 − 1.
paper is organized as follows. In Section 2, some definition
and lemmas are given. And the main results of the paper are Let 𝐶𝑘 (𝑅, 𝑅𝑛 ) = {(𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) ∈ 𝐶(𝑅, 𝑅𝑛 ) : 0 ≤ 𝜙𝑖 ≤
established in the last section. 𝑘𝑖 , 𝑖 = 1, 2, . . . , 𝑛}, and for (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) ∈ 𝐶𝑘 (𝑅, 𝑅𝑛 ), define
𝐹 = (𝐹1 , 𝐹2 , . . . , 𝐹𝑛 ) : 𝐶𝑘 (𝑅, 𝑅𝑛 ) → 𝐶𝑘 (𝑅, 𝑅𝑛 ) by
2. Preliminary 𝐹𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) (𝑡)

On substituting 𝑢𝑖 (𝑡, 𝑥) = 𝜙𝑖 (𝑥+𝑐𝑡) and denoting the traveling 1

=
wave coordinate 𝑥 + 𝑐𝑡 still by 𝑡, we derive from (2) that 𝐷𝑖 (𝜆 𝑖2 − 𝜆 𝑖1 )
𝑛 𝑡 (9)
𝐷𝑖 𝜙𝑖󸀠󸀠 (𝑡) − 𝑐𝜙𝑖󸀠 (𝑡) − 𝑎𝑖 𝜙𝑖 (𝑡) + ∑ 𝑤𝑖𝑗 𝑓𝑗 (𝜙𝑗 (𝑡)) × [∫ 𝑒𝜆 𝑖1 (𝑡−𝑠) 𝐻𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) (𝑠) 𝑑𝑠
−∞
𝑗=1
(3) ∞
𝑛 +∫ 𝑒𝜆 𝑖2 (𝑡−𝑠) 𝐻𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) (𝑠) 𝑑𝑠] ,
+ ∑ ℎ𝑖𝑗 𝑔𝑗 (𝜙𝑗 (𝑡 − 𝜏𝑗 (𝑡))) = 0. 𝑡
𝑗=1
where

If for some 𝑐 > 0, system (2) has a solution defined on 𝑅𝑛 𝑐 − √𝑐2 + 4𝐷𝑖 𝑎𝑖 𝑐 + √𝑐2 + 4𝐷𝑖 𝑎𝑖
𝜆 𝑖1 = , 𝜆 𝑖2 = . (10)
satisfying 2𝐷𝑖 2𝐷𝑖
lim 𝜙𝑖 (𝑡) = 𝜙𝑖− , lim 𝜙𝑖 (𝑡) = 𝜙𝑖+ , (4) It is easy to see that 𝐹𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) satisfy
𝑡 → −∞ 𝑡→∞
𝐷𝑖 𝐹𝑖󸀠󸀠 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) − 𝑐𝐹𝑖󸀠 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 )
where (𝜙1− , 𝜙2− , . . . , 𝜙𝑛− ) and (𝜙1+ , 𝜙2+ , . . . , 𝜙𝑛+ ) are steady states of (11)
(2), then 𝑢𝑖 (𝑡, 𝑥) = 𝜙𝑖 (𝑥+𝑐𝑡) is called a traveling wave solution − 𝑎𝑖 𝐹𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) + 𝐻𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) (𝑡) = 0.
of (2) with speed 𝑐. Without loss of generality, we assume Throughout this paper, we always assume that the follow-
that (𝜙1− , 𝜙2− , . . . , 𝜙𝑛− ) = (0, 0, . . . , 0) and (𝜙1+ , 𝜙2+ , . . . , 𝜙𝑛+ ) = ing assumptions hold.
(𝑘1 , 𝑘2 , . . . , 𝑘𝑛 ).
Rewrite model (3) as (H1 )
𝑛 𝑛
𝐷𝑖 𝜙𝑖󸀠󸀠 (𝑡) − 𝑐𝜙𝑖󸀠 (𝑡) − 𝑎𝑖 𝜙𝑖 (𝑡) + 𝐻𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) (𝑡) = 0, (5) ∑𝑤𝑖𝑗 𝑓𝑗 (0) + ∑ℎ𝑖𝑗 𝑔𝑗 (0)
𝑗=1 𝑗=1
where (12)
𝑛 𝑛
= ∑𝑤𝑖𝑗 𝑓𝑗 (𝑘𝑗 ) + ∑ℎ𝑖𝑗 𝑔𝑗 (𝑘𝑗 ) = 0, 𝑖 = 1, 2, . . . , 𝑛.
𝐻𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) (𝑡) 𝑗=1 𝑗=1
𝑛 𝑛 (6)
= ∑ 𝑤𝑖𝑗 𝑓𝑗 (𝜙𝑗 (𝑡)) + ∑ ℎ𝑖𝑗 𝑔𝑗 (𝜙𝑗 (𝑡 − 𝜏𝑗 (𝑡))) . (H2 ) For any 0 ≤ 𝜙𝑖1 , 𝜙𝑖2 ≤ 𝑘𝑖 (𝑖 = 1, 2, . . . , 𝑛), there
𝑗=1 𝑗=1 exist some positive constants 𝐿 𝑖 , 𝐾𝑖 > 0 such that
󵄨󵄨 󵄨 󵄨 󵄨
󵄨󵄨𝑓𝑖 (𝜙𝑖1 ) − 𝑓𝑖 (𝜙𝑖2 )󵄨󵄨󵄨 ≤ 𝐿 𝑖 󵄨󵄨󵄨𝜙𝑖1 − 𝜙𝑖2 󵄨󵄨󵄨 ,
In the following parts, we assume that the nonlinear reaction 󵄨 󵄨 󵄨 󵄨
terms 𝐻𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 )(𝑡)(𝑖 = 1, 2, . . . , 𝑛) satisfy the partial 󵄨󵄨 (13)
1 2 󵄨󵄨 󵄨󵄨 1 2 󵄨󵄨
󵄨󵄨𝑔𝑖 (𝜙𝑖 ) − 𝑔𝑖 (𝜙𝑖 )󵄨󵄨 ≤ 𝐾𝑖 󵄨󵄨𝜙𝑖 − 𝜙𝑖 󵄨󵄨 .
quasimonotonicity conditions (PQM): 󵄨 󵄨 󵄨 󵄨
Abstract and Applied Analysis 3

Let 𝑢 > 0 and equip Φ = (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) ∈ Lemma 3. Assume that (H2 ) holds; then 𝐹𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 )(𝑡)
𝐶(𝑅, 𝑅𝑛 ) with the exponential decay norm defined by |Φ|𝜇 = is continuous with respect to the norm | ⋅ | in 𝐵𝜇 (𝑅, 𝑅𝑛 ).
sup𝑡∈𝑅 𝑒−𝜇|𝑡| |Φ(𝑡)|𝑅𝑛 . And define 𝐵𝜇 (𝑅, 𝑅𝑛 ) = {Φ ∈ 𝐶(𝑅, 𝑅𝑛 ) :
|Φ|𝜇 < ∞}; then it is easy to see that (𝐵𝜇 (𝑅, 𝑅𝑛 ), | ⋅ |𝜇 ) is a Proof. For any fixed 𝜖 > 0, choose 𝛿 < 𝜖/(∑𝑛𝑗=1 𝑤𝑖𝑗 𝐿 𝑗 +
Banach space. ∑𝑛𝑗=1 ℎ𝑖𝑗 𝐾𝑗 ); a direct calculation shows that |𝜙𝑗1 − 𝜙𝑗2 | < 𝛿,
𝜇
We will study traveling wave solution to system (2) in the |𝜙𝑗1 (𝑡 − 𝜏𝑗 (𝑡)) − 𝜙𝑗2 (𝑡 − 𝜏𝑗 (𝑡))| < 𝛿; then there exists
following profile set:
󵄨󵄨 󵄨
Γ ((𝜙 , 𝜙 , . . . , 𝜙 ) , (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 )) 󵄨󵄨𝐻𝑖 (𝜙11 , 𝜙21 , . . . , 𝜙𝑛1 ) (𝑡) − 𝐻𝑖 (𝜙12 , 𝜙22 , . . . , 𝜙𝑛2 ) (𝑡)󵄨󵄨󵄨 𝑒−𝜇|𝑡|
1 2 𝑛 󵄨 󵄨
(14) 󵄨󵄨 𝑛
(1) 𝜙𝑛 is nondecreasing in 𝑅; 󵄨󵄨 𝑛
={ = 󵄨󵄨󵄨󵄨∑ 𝑤𝑖𝑗 𝑓𝑗 (𝜙𝑗1 (𝑡)) + ∑ ℎ𝑖𝑗 𝑔𝑗 (𝜙𝑗 (𝑡 − 𝜏𝑗 (𝑡)))
(2) 𝜙 ≤ 𝜙𝑖 ≤ 𝜙𝑖 , 𝑖 = 1, 2, . . . , 𝑛. 󵄨󵄨𝑗=1
𝑖 󵄨 𝑗=1

󵄨󵄨
It is easy to check that Γ((𝜙 , 𝜙 , . . . , 𝜙 ), (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 )) is 𝑛 𝑛 󵄨󵄨
1 2
nonempty, convex, closed, and bounded.
𝑛
− ∑𝑤𝑖𝑗 𝑓𝑗 (𝜙𝑗 (𝑡)) − ∑ℎ𝑖𝑗 𝑔𝑗 (𝜙𝑗 (𝑡 − 𝜏𝑗 (𝑡)))󵄨󵄨󵄨󵄨 𝑒−𝜇|𝑡|
2 2
󵄨󵄨
𝑗=1 𝑗=1 󵄨
Definition 1. Function Φ = (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) is called upper 𝑛
󵄨 󵄨
solution of system (2) if Φ is twice differentiable almost ≤ ∑ 𝑤𝑖𝑗 󵄨󵄨󵄨󵄨𝑓𝑗 (𝜙𝑗1 (𝑡)) − 𝑓𝑗 (𝜙𝑗2 (𝑡))󵄨󵄨󵄨󵄨 𝑒−𝜇|𝑡|
everywhere in 𝑅, and there hold 𝑗=1

𝑛
󸀠󸀠 󸀠 󵄨
𝐷𝑖 𝜙𝑖 (𝑡) − 𝑐𝜙𝑖 (𝑡) − 𝑎𝑖 𝜙𝑖 (𝑡) + ∑ℎ𝑖𝑗 󵄨󵄨󵄨󵄨𝑔𝑗 (𝜙𝑗1 (𝑡 − 𝜏𝑗 (𝑡)))
𝑗=1
+ 𝐻𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑖 , 𝜙 , 𝜙𝑖+2 , . . . , 𝜙𝑛 ) (𝑡) 󵄨
𝑖+1
−𝑔𝑗 (𝜙𝑗2 (𝑡 − 𝜏𝑗 (𝑡)))󵄨󵄨󵄨󵄨 𝑒−𝜇|𝑡|
≤ 0, a.e. in 𝑅, (15)
𝑛 𝑛
󸀠󸀠 󸀠 ≤ ∑ 𝑤𝑖𝑗 𝐿 𝑗 𝛿 + ∑ ℎ𝑖𝑗 𝐾𝑗 𝛿 ≤ 𝜖.
𝐷𝑛 𝜙𝑛 (𝑡) − 𝑐𝜙𝑛 (𝑡) − 𝑎𝑖 𝜙𝑛 (𝑡) 𝑗=1 𝑗=1
(17)
+ 𝐻𝑛 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) ≤ 0, a.e. in 𝑅.

Reversing the direction of above inequalities, we can get the For 𝑡 > 0, we can see that
lower solution. 󵄨󵄨 󵄨
󵄨󵄨𝐹𝑖 (𝜙11 , 𝜙21 , . . . , 𝜙𝑛1 ) − 𝐹𝑖 (𝜙12 , 𝜙22 , . . . , 𝜙𝑛2 )󵄨󵄨󵄨 𝑒−𝜇|𝑡|
󵄨 󵄨
In this paper, we assume that the upper-lower solutions
of system (2) Φ = (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) and Φ = (𝜙 , 𝜙 , . . . , 𝜙 ) 1
1 2 𝑛 ≤
satisfy 𝐷𝑖 (𝜆 𝑖2 − 𝜆 𝑖1 )
𝑡
(P1 ) (0, 0, . . . , 0) ≤ (𝜙 , 𝜙 , . . . , 𝜙 ) ≤ (𝜙1 , 𝜙2 , . . . , × [∫ 𝑒𝜆 𝑖1 (𝑡−𝑠)
1 2 𝑛
−∞
𝜙𝑛 ) ≤ (𝑘1 , 𝑘2 , . . . , 𝑘𝑛 ), 𝑡 ∈ 𝑅,
󵄨
(P2 ) lim𝑡 → −∞ (𝜙 , 𝜙 , . . . , 𝜙 ) = (0, 0, . . . , 0), and × 󵄨󵄨󵄨󵄨𝐻𝑖 (𝜙11 , 𝜙21 , . . . , 𝜙𝑛1 ) (𝑠)
1 2 𝑛
lim𝑡 → ∞ (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) = (𝑘1 , 𝑘2 , . . . , 𝑘𝑛 ). 󵄨
−𝐻𝑖 (𝜙12 , 𝜙22 , . . . , 𝜙𝑛2 ) (𝑠)󵄨󵄨󵄨󵄨 𝑒−𝜇|𝑠| 𝑒𝜇|𝑠| 𝑑𝑠
Lemma 2. Assume that (PQM) holds; then one has +∞
+∫ 𝑒𝜆 𝑖2 (𝑡−𝑠)
(1) 𝐹𝑛 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 )(𝑡) is nondecreasing for 𝑡 ∈ 𝑅. 𝑡
󵄨
(2) For 𝑡 ∈ 𝑅, 𝑖 = 1, 2, . . . , 𝑛 − 1 and 0 ≤ 𝜙 ≤ 𝜙𝑗 ≤ 𝑘𝑗 (𝑗 =
𝑗 × 󵄨󵄨󵄨󵄨𝐻𝑖 (𝜙11 , 𝜙21 , . . . , 𝜙𝑛1 ) (𝑠)
1, 2, . . . , 𝑛), there has 󵄨
−𝐻𝑖 (𝜙12 , 𝜙22 , . . . , 𝜙𝑛2 ) (𝑠)󵄨󵄨󵄨󵄨
𝐹𝑖 (𝜙 , 𝜙 , . . . , 𝜙 , 𝜙𝑖+1 , 𝜙 , . . . , 𝜙 ) (𝑡)
𝑖 2 𝑖 𝑖+2 𝑛
× 𝑒−𝜇|𝑠| 𝑒𝜇|𝑠| 𝑑𝑠] 𝑒−𝜇𝑡
≤ 𝐹𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) (𝑡) (16)
𝑡
𝜖
≤ [∫ 𝑒𝜆 𝑖1 (𝑡−𝑠) 𝑒𝜇|𝑠| 𝑑𝑠
𝐷𝑖 (𝜆 𝑖2 − 𝜆 𝑖1 ) −∞
≤ 𝐹𝑖 (𝜙𝑖 , 𝜙2 , . . . , 𝜙𝑖 , 𝜙 , 𝜙𝑖+2 , . . . , 𝜙𝑛 ) (𝑡) .
𝑖+1
+∞

Lemma 2 is easy to prove, so we omit it. +∫ 𝑒𝜆 𝑖2 (𝑡−𝑠) 𝑒𝜇|𝑠| 𝑑𝑠] 𝑒−𝜇𝑡

𝑡
4 Abstract and Applied Analysis

𝜖 2𝜇 𝜆 𝑖2 − 𝜆 𝑖1 By the definition of upper-lower solutions, we have

= [ 𝑒(𝜆 𝑖1 −𝜇)𝑡 + ]
𝐷𝑖 (𝜆 𝑖2 − 𝜆 𝑖1 ) 𝜆2𝑖1 −𝜇2 (𝜇 − 𝜆 𝑖1 ) (𝜆 𝑖2 − 𝜇)

𝜖 2𝜇 𝜆 𝑖2 − 𝜆 𝑖1 󸀠󸀠 󸀠
≤ 2 − 𝜇2
+ [ ]. 𝐷𝑖 𝜙𝑖 (𝑡) − 𝑐𝜙𝑖 (𝑡) − 𝑎𝑖 𝜙𝑖 (𝑡)
𝐷𝑖 (𝜆 𝑖2 − 𝜆 𝑖1 ) 𝜆 𝑖1 (𝜇 − 𝜆 𝑖1 ) (𝜆 𝑖2 − 𝜇) (22)
(18) + 𝐻𝑖 (𝜙𝑖 , 𝜙2 , . . . , 𝜙𝑖 , 𝜙 , 𝜙𝑖+2 , . . . , 𝜙𝑛 ) (𝑡) ≤ 0.
𝑖+1
For 𝑡 < 0, we have
󵄨󵄨 1 1 1 2 2 2 󵄨󵄨 −𝜇|𝑡|
󵄨󵄨󵄨𝐹𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) − 𝐹𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 )󵄨󵄨󵄨 𝑒
Choosing (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) = (𝜙𝑖 , 𝜙2 , . . . , 𝜙𝑖 , 𝜙 , 𝜙𝑖+2 , . . . , 𝜙𝑛 )
𝑖+1
1 in (9) and denoting 𝜙̃𝑖 = 𝐹𝑖 (𝜙𝑖 , 𝜙2 , . . . , 𝜙𝑖 , 𝜙 , 𝜙𝑖+2 , . . . , 𝜙𝑛 ),
≤ 𝑖+1
𝐷𝑖 (𝜆 𝑖2 − 𝜆 𝑖1 ) we get
𝑡
󵄨
× [∫ 𝑒𝜆 𝑖1 (𝑡−𝑠) 󵄨󵄨󵄨󵄨𝐻𝑖 (𝜙11 , 𝜙21 , . . . , 𝜙𝑛1 ) (𝑠)
−∞
𝐷𝑖 𝜙̃𝑖󸀠󸀠 (𝑡) − 𝑐𝜙̃𝑖󸀠 (𝑡) − 𝑎𝑖 𝜙̃𝑖 (𝑡)
󵄨
−𝐻𝑖 (𝜙12 , 𝜙22 , . . . , 𝜙𝑛2 ) (𝑠)󵄨󵄨󵄨󵄨 𝑒−𝜇|𝑠| 𝑒𝜇|𝑠| 𝑑𝑠 (23)
+ 𝐻𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑖 , 𝜙 , 𝜙𝑖+2 , . . . , 𝜙𝑛 ) (𝑡) = 0.
+∞ 𝑖+1
󵄨
+∫ 𝑒𝜆 𝑖2 (𝑡−𝑠) 󵄨󵄨󵄨󵄨𝐻𝑖 (𝜙11 , 𝜙21 , . . . , 𝜙𝑛1 ) (𝑠)
𝑡
󵄨 Setting 𝑤𝑖 (𝑡) = 𝜙̃𝑖 − 𝜙𝑖 and combining (22) and (23), we have
−𝐻𝑖 (𝜙12 , 𝜙22 , . . . , 𝜙𝑛2 ) (𝑠)󵄨󵄨󵄨󵄨 (19)

× 𝑒−𝜇|𝑠| 𝑒𝜇|𝑠| 𝑑𝑠] 𝑒𝜇𝑡

𝐷𝑖 𝑤𝑖󸀠󸀠 − 𝑐𝑤𝑖󸀠 − 𝑎𝑖 ≥ 0. (24)
𝑡
𝜖
≤ [∫ 𝑒𝜆 𝑖1 (𝑡−𝑠) 𝑒𝜇|𝑠| 𝑑𝑠
𝐷𝑖 (𝜆 𝑖2 − 𝜆 𝑖1 ) −∞
Repeating the proof of Lemma 3.3 in Wu and Zou [17], we
+∞ obtain 𝑤𝑖 (𝑡) ≤ 0, which implies that 𝐹𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) ≤
+∫ 𝑒𝜆 𝑖2 (𝑡−𝑠) 𝑒𝜇|𝑠| 𝑑𝑠] 𝑒𝜇𝑡 𝜙𝑖 (𝑖 = 1, 2, . . . , 𝑛 − 1).
𝑡
Next, choosing (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) = (𝜙𝑖 , 𝜙2 , . . . , 𝜙𝑛 ) in (9),
𝜖 2𝜇 𝜆 𝑖2 − 𝜆 𝑖1 we can get 𝐹𝑛 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) ≤ 𝜙𝑛 . Then, by a similar
≤ [ − ].
𝐷𝑖 (𝜆 𝑖2 − 𝜆 𝑖1 ) 𝜆2𝑖2 −𝜇2 (𝜇 + 𝜆 𝑖1 ) (𝜆 𝑖2 + 𝜇) argument, we know that 𝐹𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) ≥ 𝜙 , (𝑖 =
𝑖
So, 𝐹𝑖 : 𝐵𝜇 (𝑅, 𝑅𝑛 ) → 𝐵𝜇 (𝑅, 𝑅𝑛 ) is continuous with respect to 1, 2, . . . , 𝑛 − 1), and 𝐹𝑛 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) ≥ 𝜙 ; then
𝑛
the norm | ⋅ | in 𝐵𝜇 (𝑅, 𝑅𝑛 ).

Lemma 4. Assume that (H1 ) and (PQM) hold; then

𝐹 (Γ ((𝜙 , 𝜙 , . . . , 𝜙 ) , (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 )))
1 2 𝑛
𝐹 (Γ ((𝜙 , 𝜙 , . . . , 𝜙 ) , (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ))) (25)
1 2 𝑛
(20) ⊂ Γ ((𝜙 , 𝜙 , . . . , 𝜙 ) , (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 )) .
1 2 𝑛
⊂ Γ ((𝜙 , 𝜙 , . . . , 𝜙 ) , (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 )) .
1 2 𝑛

Proof. According to Lemma 2, for any (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) with

This completes the proof.
(𝜙 , 𝜙 , . . . , 𝜙 ) ≤ (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) ≤ (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ), there
1 2 𝑛
exists Lemma 5. The above defined function 𝐹 is compact.
𝐹𝑖 (𝜙 , 𝜙 , . . . , 𝜙 , 𝜙𝑖+1 , 𝜙 , . . . , 𝜙 ) (𝑡)
𝑖 2 𝑖 𝑖+2 𝑛 Proof. By the definition of 𝐹, we have

≤ 𝐹𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) (𝑡)
𝐹𝑖󸀠 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) (𝑡)
≤ 𝐹𝑖 (𝜙𝑖 , 𝜙2 , . . . , 𝜙𝑖 , 𝜙 , 𝜙𝑖+2 , . . . , 𝜙𝑛 ) (𝑡) ; (21) 𝜆 𝑖1 𝑒𝜆 𝑖1 𝑡 𝑡
𝑖+1
= ∫ 𝑒−𝜆 𝑖1 𝑠 𝐻𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) (𝑠) 𝑑𝑠
𝐷𝑖 (𝜆 𝑖2 − 𝜆 𝑖1 ) −∞
𝐹𝑛 (𝜙 , 𝜙 , . . . , 𝜙 )
1 2 𝑛 𝜆 𝑖2 𝑒𝜆 𝑖2 𝑡 ∞
+ ∫ 𝑒−𝜆 𝑖2 𝑠 𝐻𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) (𝑠) 𝑑𝑠.
𝐷𝑖 (𝜆 𝑖2 − 𝜆 𝑖1 ) 𝑡
≤ 𝐹𝑛 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) ≤ 𝐹𝑛 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) . (26)
Abstract and Applied Analysis 5

𝜆 𝑖1 󵄨󵄨 󵄨
Thus, ≤ 󵄨󵄨𝐻𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) (𝑡)󵄨󵄨󵄨𝜇
𝐷𝑖 (𝜇 + 𝜆 𝑖1 ) (𝜆 𝑖2 − 𝜆 𝑖1 )
󵄨󵄨 󸀠 󵄨
󵄨󵄨𝐹𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) (𝑡)󵄨󵄨󵄨
󵄨 󵄨𝜇 𝜆 𝑖2 󵄨󵄨 󵄨
+ 󵄨𝐻 (𝜙 , 𝜙 , . . . , 𝜙𝑛 ) (𝑡)󵄨󵄨󵄨𝜇
󵄨󵄨 𝐷𝑖 (𝜆 𝑖2 − 𝜇) (𝜆 𝑖2 − 𝜆 𝑖1 ) 󵄨 𝑖 1 2
󵄨 𝜆 𝑖1 𝑒𝜆 𝑖1 𝑡
= sup 󵄨󵄨󵄨󵄨
𝑡∈𝑅 󵄨󵄨 𝐷𝑖 (𝜆 𝑖2 − 𝜆 𝑖1 ) 1 𝜆 𝑖1 𝜆 𝑖2
= [ + ]
𝑡 𝐷𝑖 (𝜆 𝑖2 − 𝜆 𝑖1 ) 𝜇 + 𝜆 𝑖1 𝜆 𝑖2 − 𝜇
×∫ 𝑒−𝜆 𝑖1 𝑠 𝐻𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) (𝑠) 𝑑𝑠 󵄨 󵄨
−∞ × 󵄨󵄨󵄨𝐻𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) (𝑡)󵄨󵄨󵄨𝜇 ;
𝜆 𝑖2 𝑒𝜆 𝑖2 𝑡 (28)
+
𝐷𝑖 (𝜆 𝑖2 − 𝜆 𝑖1 ) (ii) case 𝑡 < 0
∞ 󵄨󵄨 󵄨󵄨 󸀠 󵄨
󵄨 󵄨󵄨𝐹𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) (𝑡)󵄨󵄨󵄨
× ∫ 𝑒−𝜆 𝑖2 𝑠 𝐻𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) (𝑠) 𝑑𝑠󵄨󵄨󵄨󵄨 𝑒−𝜇|𝑡| 󵄨 󵄨𝜇
𝑡 󵄨󵄨
−𝜆 𝑖1
−𝜆 𝑖1 ≤
≤ 𝐷𝑖 (𝜆 𝑖2 − 𝜆 𝑖1 )
𝐷𝑖 (𝜆 𝑖2 − 𝜆 𝑖1 )
𝑡
𝑡
󵄨󵄨 󵄨 × sup 𝑒(𝜆 𝑖1 +𝜇)𝑡 ∫ 𝑒−(𝜆 𝑖1 +𝜇)𝑠 𝑑𝑠
󵄨󵄨𝐻𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) (𝑠)󵄨󵄨󵄨 𝑑𝑠
𝜆 𝑖1 𝑡−𝜇|𝑡| −𝜆 𝑖1 𝑠
× sup 𝑒 ∫ 𝑒 𝑡∈𝑅 −∞
𝑡∈𝑅 −∞
󵄨 󵄨
𝜆 𝑖2 × 󵄨󵄨󵄨𝐻𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) (𝑡)󵄨󵄨󵄨𝜇
+
𝐷𝑖 (𝜆 𝑖2 − 𝜆 𝑖1 ) 𝜆 𝑖2
+
∞
󵄨 󵄨 𝐷𝑖 (𝜆 𝑖2 − 𝜆 𝑖1 )
× sup 𝑒𝜆 𝑖2 𝑡−𝜇|𝑡| ∫ 𝑒−𝜆 𝑖2 𝑠 󵄨󵄨󵄨󵄨𝐻𝑖 (𝜙12 , 𝜙22 , . . . , 𝜙𝑛2 ) (𝑠)󵄨󵄨󵄨󵄨 𝑑𝑠
𝑡∈𝑅 𝑡 0 ∞
× sup 𝑒(𝜆 𝑖2 +𝜇)𝑡 [∫ 𝑒(−𝜆 𝑖2 −𝜇)𝑠 𝑑𝑠 + ∫ 𝑒(𝜇−𝜆 𝑖2 )𝑠 𝑑𝑠]
−𝜆 𝑖1 󵄨󵄨 󵄨 𝑡 0
󵄨󵄨𝐻 (𝜙2 , 𝜙2 , . . . , 𝜙𝑛2 ) (𝑡)󵄨󵄨󵄨
𝑡∈𝑅
≤
𝐷𝑖 (𝜆 𝑖2 − 𝜆 𝑖1 ) 󵄨 𝑖 1 2 󵄨𝜇 󵄨 󵄨
× 󵄨󵄨󵄨𝐻𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) (𝑡)󵄨󵄨󵄨𝜇
𝑡
× sup 𝑒𝜆 𝑖1 𝑡−𝜇|𝑡| ∫ 𝑒−𝜆 𝑖1 𝑠 𝑒𝜇|𝑠| 𝑑𝑠 𝜆 𝑖1 󵄨󵄨 󵄨
≤ 󵄨𝐻 (𝜙 , 𝜙 , . . . , 𝜙𝑛 ) (𝑡)󵄨󵄨󵄨𝜇
𝐷𝑖 (𝜇 + 𝜆 𝑖1 ) (𝜆 𝑖2 − 𝜆 𝑖1 ) 󵄨 𝑖 1 2
𝑡∈𝑅 −∞

𝜆 𝑖2 󵄨󵄨 󵄨
+ 󵄨󵄨𝐻 (𝜙2 , 𝜙2 , . . . , 𝜙𝑛2 ) (𝑡)󵄨󵄨󵄨 𝜆 𝑖2 󵄨󵄨 󵄨
𝐷𝑖 (𝜆 𝑖2 − 𝜆 𝑖1 ) 󵄨 𝑖 1 2 󵄨𝜇 + 󵄨𝐻 (𝜙 , 𝜙 , . . . , 𝜙𝑛 ) (𝑡)󵄨󵄨󵄨𝜇
𝐷𝑖 (𝜆 𝑖2 − 𝜇) (𝜆 𝑖2 − 𝜆 𝑖1 ) 󵄨 𝑖 1 2
∞
× sup 𝑒𝜆 𝑖2 𝑡−𝜇|𝑡| ∫ 𝑒−𝜆 𝑖2 𝑠 𝑒𝜇|𝑠| 𝑑𝑠. 1 𝜆 𝑖1 𝜆 𝑖2
𝑡∈𝑅 𝑡 = [ + ]
(27) 𝐷𝑖 (𝜆 𝑖2 − 𝜆 𝑖1 ) 𝜇 + 𝜆 𝑖1 𝜆 𝑖2 − 𝜇
󵄨 󵄨
× 󵄨󵄨󵄨𝐻𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) (𝑡)󵄨󵄨󵄨𝜇 .
We will complete the proof by two cases as follows:
(i) case 𝑡 > 0 (29)

󵄨󵄨 󸀠 󵄨 According to the conditions in (PQM), we get that

󵄨󵄨𝐹𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) (𝑡)󵄨󵄨󵄨
󵄨 󵄨𝜇 |𝐻𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 )(𝑡)|𝜇 is bounded by a positive number.
−𝜆 𝑖1 Therefore, |𝐹𝑖󸀠 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 )(𝑡)|𝜇 is bounded. The
≤ above estimate for 𝐹󸀠 shows that 𝐹(Γ((𝜙 , 𝜙 , . . . , 𝜙 ),
𝐷𝑖 (𝜆 𝑖2 − 𝜆 𝑖1 ) 1 2 𝑛

0 𝑡
(𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ))) is equicontinuous. It follows from Lemma 4
× sup 𝑒(𝜆 𝑖1 −𝜇)𝑡 [∫ 𝑒−(𝜆 𝑖1 +𝜇)𝑠 𝑑𝑠 + ∫ 𝑒(𝜇−𝜆 𝑖1 )𝑠 𝑑𝑠] that 𝐹(Γ((𝜙 , 𝜙 , . . . , 𝜙 ), (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ))) is uniformly
1 2 𝑛
𝑡∈𝑅 −∞ 0
bounded.
󵄨 󵄨
× 󵄨󵄨󵄨𝐻𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) (𝑡)󵄨󵄨󵄨𝜇 Next, we define

𝜆 𝑖2 𝐹𝑚 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) (𝑡)
+
𝐷𝑖 (𝜆 𝑖2 − 𝜆 𝑖1 ) (1) 𝐹 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) (𝑡) , 𝑡 ∈ [−𝑚, 𝑚] ;
{
{ (30)
∞
󵄨 󵄨 = {(2) 𝐹 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) (𝑚) , 𝑡 ∈ (𝑚, ∞) ;
× sup 𝑒 (𝜆 𝑖2 −𝜇)𝑡
∫ 𝑒 (𝜇−𝜆 𝑖2 )𝑠
𝑑𝑠󵄨󵄨󵄨𝐻𝑖 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) (𝑡)󵄨󵄨󵄨𝜇 {
𝑡∈𝑅 𝑡 {(3) 𝐹 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) (−𝑚) , 𝑡 ∈ (−∞, −𝑚) .
6 Abstract and Applied Analysis

Then, for each 𝑚 ≥ 1, 𝐹𝑚 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 )(𝑡) is also equicon- Conflict of Interests

tinuous and uniformly bounded. In the interval [−𝑚, 𝑚], it
follows from Ascoli-Arzela theorem that 𝐹𝑚 is compact. On The authors declare that there is no conflict of interests
the other hand, 𝐹𝑚 → 𝐹 in 𝐵𝜇 (𝑅, 𝑅𝑛 ) as 𝑚 → ∞, since regarding the publication of this paper.

󵄨 󵄨
sup 󵄨󵄨󵄨𝐹𝑚 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) (𝑡) − 𝐹 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) (𝑡)󵄨󵄨󵄨 𝑒−𝜇|𝑡| Acknowledgments
𝑡∈𝑅
󵄨󵄨 𝑚 This work was supported by the NNSF of China under Grant
= sup 󵄨󵄨𝐹 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) (𝑡) (no. 11271261), Natural Science Foundation of Shanghai (no.
𝑡∈(−∞,−𝑚)∪(𝑚,∞)
13ZR1430100), Shanghai municipal education commission (n.
󵄨 14ZZ12), and the Slovenian Research Agency, and a Marie
−𝐹 (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) (𝑡) 󵄨󵄨󵄨 𝑒−𝜇|𝑡|
Curie International Research Staff Exchange Scheme Fellow-
≤ 2𝐾𝑒−𝜇𝑚 󳨀→ 0, 𝑚 󳨀→ ∞. ship within the 7th European Community Programme, FP7-
(31) PEOPLE-2012-IRSES-316338.

By Proposition 2.12 in [22], we have that 𝐹 : References

Γ((𝜙 , 𝜙 , . . . , 𝜙 ), (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 )) → Γ((𝜙 , 𝜙 , . . . , 𝜙 ),
1 2 𝑛 1 2 𝑛 [1] Y. Kuang, Delay Differential Equations with Applications in
(𝜙1 , 𝜙2 , . . . , 𝜙𝑛 )) is compact. This completes the proof. Population Dynamics, Mathematics in Science and Engineering,
Academic Press, New York, NY, USA, 1993.
[2] R. Xu, M. A. J. Chaplain, and F. A. Davidson, “Periodic solution
3. Main Results for athree-species Lotka-Volterra food-chain model with time
delays,” Mathematical and Computer Modelling, vol. 40, no. 7-8,
Theorem 6. Assume that (H1 ), (H2 ), and (PQM) hold. More-
pp. 823–837, 2004.
over, suppose that there is a pair of upper-lower solutions Φ = [3] X.-P. Yan and Y.-D. Chu, “Stability and bifurcation analysis
(𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) and Φ = (𝜙 , 𝜙 , . . . , 𝜙 ) for (2) satisfying (P1 ) for a delayed Lotka-Volterra predator-prey system,” Journal of
1 2 𝑛
and (P2 ). Then, system (2) has a travelling wave solution. Computational and Applied Mathematics, vol. 196, no. 1, pp. 198–
210, 2006.
Proof. Following Lemmas 2 to 5, we see that all the con- [4] C. V. Pao, “Global asymptotic stability of Lotka-Volterra 3-
dition in Schauder’s fixed point theorem hold. Then we species reaction-diffusion systems with time delays,” Journal of
know that there exists a fixed point (𝜙1∗ , 𝜙2∗ , . . . , 𝜙𝑛∗ ) of 𝐹 in Mathematical Analysis and Applications, vol. 281, no. 1, pp. 186–
Γ((𝜙 , 𝜙 , . . . , 𝜙 ), (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 )). Now we need to verify the 204, 2003.
1 2 𝑛 [5] C. V. Pao, “Global asymptotic stability of Lotka-Volterra com-
asymptotic boundary conditions. By (P2 ) we notice the fact petition systems with diffusion and time delays,” Nonlinear
that Analysis: Real World Applications, vol. 5, no. 1, pp. 91–104, 2004.
[6] C. V. Pao, “The global attractor of a competitor-competitor-
(0, 0, . . . , 0) ≤ (𝜙 , 𝜙 , . . . , 𝜙 ) ≤ (𝜙1∗ , 𝜙2∗ , . . . , 𝜙𝑛∗ ) mutualist reaction-diffusion system with time delays,” Nonlin-
1 2 𝑛
(32) ear Analysis: Theory, Methods and Applications, vol. 67, no. 9,
≤ (𝜙1 , 𝜙2 , . . . , 𝜙𝑛 ) ≤ (𝑘1 , 𝑘2 , . . . , 𝑘𝑛 ) ; pp. 2623–2631, 2007.
[7] Y. Tang and L. Zhou, “Stability switch and Hopf bifurcation
we get that for a diffusive prey-predator system with delay,” Journal of
Mathematical Analysis and Applications, vol. 334, no. 2, pp.
1290–1307, 2007.
lim (𝜙1∗ , 𝜙2∗ , . . . , 𝜙𝑛∗ ) = (0, 0, . . . , 0) ,
𝑡 → −∞ [8] K. W. Schaaf, “Asymptotic behavior and traveling wave solutions
(33) for parabolic functional-differential equations,” Transactions of
lim (𝜙1∗ , 𝜙2∗ , . . . , 𝜙𝑛∗ ) = (𝑘1 , 𝑘2 , . . . , 𝑘𝑛 ) . the American Mathematical Society, vol. 302, no. 2, pp. 587–615,
𝑡→∞
1987.
Therefore, the fixed point (𝜙1∗ , 𝜙2∗ , . . . , 𝜙𝑛∗ ) satisfies the asymp- [9] R. A. Gardner, “Existence and stability of travelling wave
totic boundary conditions. This completes the proof. solutions of competition models: a degree theoretic approach,”
Journal of Differential Equations, vol. 44, no. 3, pp. 343–364,
1982.
4. Conclusion [10] J. Wu, Theory and Applications of Partial Functional-Differential
Equations, Springer, New York, NY, USA, 1990.
In this paper, we studied an n-species food chain model with [11] S. R. Dunbar, “Travelling wave solutions of diffuse Lotka-
spatial diffusion and time delays. By using Schauder’s fixed Volterra equations,” Journal of Mathematical Biology, vol. 17, no.
point theorem and cross-iteration methods, we reduced the 1, pp. 11–32, 1983.
existence of the travelling wave solutions to the existence of [12] P. Bates and F. Chen, “Periodic traveling waves for a nonlocal
a pair of upper-lower solutions. Finally, we proved that the integro-differential model,” Electronic Journal of Differential
system (2) has a travelling wave solution. However, in order Equations, vol. 1999, pp. 1–19, 1999.
to investigate the specific form of the travelling wave solution [13] S. Ma and Y. Duan, “Asymptotic stability of traveling waves in
of (2), we still have a lot of work to do in the future. a discrete convolution model for phase transitions,” Journal of
Abstract and Applied Analysis 7

Mathematical Analysis and Applications, vol. 308, no. 1, pp. 240–

256, 2005.
[14] J. D. Murray, Mathematical Biology, Springer, Berlin, Germany,
1989.
[15] J. Li and Z. Liu, “Smooth and non-smooth traveling waves
in a nonlinearly dispersive equation,” Applied Mathematical
Modelling, vol. 25, no. 1, pp. 41–56, 2000.
[16] K. Wang and W. Wang, “Propagation of HBV with spatial
dependence,” Mathematical Biosciences, vol. 210, no. 1, pp. 78–
95, 2007.
[17] J. Wu and X. Zou, “Traveling wave fronts of reaction-diffusion
systems with delay,” Journal of Dynamics and Differential Equa-
tions, vol. 13, no. 3, pp. 651–687, 2001.
[18] X. Zou and J. Wu, “Local existence and stability of periodic
traveling waves of lattice functional-differential equations,” The
Canadian Applied Mathematics Quarterly, vol. 6, no. 4, pp. 397–
418, 1998.
[19] J. Huang and X. Zou, “Traveling wavefronts in diffusive and
cooperative Lotka-Volterra system with delays,” Journal of
Mathematical Analysis and Applications, vol. 271, no. 2, pp. 455–
466, 2002.
[20] S. Ma, “Traveling wavefronts for delayed reaction-diffusion
systems via a fixed point theorem,” Journal of Differential
Equations, vol. 171, no. 2, pp. 294–314, 2001.
[21] Q. Gan, R. Xu, X. Zhang, and P. Yang, “Travelling waves of
a three-species Lotka-Volterra food-chain model with spatial
diffusion and time delays,” Nonlinear Analysis: Real World
Applications, vol. 11, no. 4, pp. 2817–2832, 2010.
[22] E. Zeidler, Nonlinear Functional Analysis and its Applications: I.
Fixed-Point Theorems, Springer, New York, NY, USA, 1986.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 649270, 15 pages
http://dx.doi.org/10.1155/2014/649270

Research Article
Quasiperiodic Solutions of Completely Resonant Wave
Equations with Quasiperiodic Forced Terms

Yixian Gao,1,2 Weipeng Zhang,1 and Jing Chang3

1
School of Mathematics and Statistics, Northeast Normal University, Changchun, Jilin 130024, China
2
Key Laboratory of Symbolic Computation and Knowledge Engineering, Ministry of Education, Jilin University, Changchun,
Jilin 130012, China
3
Fundamental Department, Aviation University of Air Force, Changchun 130023, China

Correspondence should be addressed to Weipeng Zhang; wpzhang808@163.com

Received 28 February 2014; Revised 25 April 2014; Accepted 28 April 2014; Published 18 May 2014

Academic Editor: Yongli Song

Copyright © 2014 Yixian Gao et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

This paper is concerned with the existence of quasiperiodic solutions with two frequencies of completely resonant, quasiperiodically
forced nonlinear wave equations subject to periodic spatial boundary conditions. The solutions turn out to be, at the first order,
the superposition of traveling waves, traveling in the opposite or the same directions. The proofs are based on the variational
Lyapunov-Schmidt reduction and the linking theorem, while the bifurcation equations are solved by variational methods.

1. Introduction Periodic or quasiperiodic solutions in nonresonant PDEs

have been obtained, for instance, in [1–12] by the Lyapunov-
This paper is devoted to the study of the existence of small- Schmidt reduction together with Nash-Moser theory and
amplitude quasiperiodic solutions of completely resonant KAM theory, while the completely resonant autonomous
forced nonlinear wave equations like PDEs have been originally studied by variational methods
starting from Rabinowitz [13–19]. They obtained the existence
𝑢𝑡𝑡 − 𝑢𝑥𝑥 + 𝑓 (𝜔1 𝑡, 𝜔2 𝑡, 𝑥, 𝑢) = 0, (1) of periodic solutions with period being a rational multiple
of 𝜋, and such solutions correspond to a zero-measure set
with periodic boundary conditions of values of the amplitudes. The case with period being
irrational of 𝜋, which in principle could provide a large
𝑢 (𝑡, 𝑥) = 𝑢 (𝑡, 𝑥 + 2𝜋) , (𝑡, 𝑥) ∈ R2 , (2) measure of values, has been mostly studied under strong Dio-
phantine conditions; see [20–25] and the references therein.
while the nonlinear forced term 𝑓 is In [26, 27], using the Lindstedt series method, Gentile and
Procesi obtained the existence of periodic solutions for a
𝑓 (𝜔1 𝑡, 𝜔2 𝑡, 𝑥, 𝑢) := ∑ 𝑎𝑘 (𝜔1 𝑡 + 𝑥, 𝜔2 𝑡 + 𝑥) 𝑢𝑘 , (3) large measure set of frequencies for the nonlinear wave
𝑘≥3
equations and nonlinear Schrödinger equations with periodic
when the traveling waves are in the same directions, and 𝑓 is boundary conditions. In [28], Yuan obtained the existence
of quasiperiodic solution for a large measure set of at least
three dimensional rotation vectors by the KAM method. In
𝑓 (𝜔1 𝑡, 𝜔2 𝑡, 𝑥, 𝑢) := ∑ 𝑎𝑘 (𝜔1 𝑡 − 𝑥, 𝜔2 𝑡 + 𝑥) 𝑢𝑘 , (4) [29], under the periodic boundary condition and with the
𝑘≥3
periodic forced nonlinearities 𝑓(𝜔1 𝑡, 𝑢) = 𝑎(𝜔1 𝑡)𝑢2𝑑−1 +
when the traveling waves are in the opposite directions. 𝑂(𝑢2𝑑 ), 𝑑 > 2, Berti and Procesi got the existence of
Moreover, the nonlinear forced terms are all analytic in a quasiperiodic solution of nonlinear wave equation in the
neighborhood of 𝑢 = 0. form of V(𝑡, 𝑥) = 𝑢(𝜔1 𝑡, 𝜔2 𝑡 + 𝑥). In [30], Procesi firstly
2 Abstract and Applied Analysis

obtained the quasiperiodic solutions with two frequencies in ∘ [𝜔1 𝜕𝜑1 + 𝜔2 𝜕𝜑2 + 𝜕𝜑1 + 𝜕𝜑2 ] V
the form of V(𝑡, 𝑥) = 𝑢(𝜔1 𝑡 + 𝑥, 𝜔2 𝑡 − 𝑥) for the specific
nonlinearities 𝑓 = 𝑢3 + 𝑂(𝑢5 ), where the forced terms do = ((𝜔12 − 1) 𝜕𝜑21 + (𝜔22 − 1) 𝜕𝜑22
not depend on the time and the bifurcation equations are
solved by ODE methods. In [31], Baldi proved the existence +2 (𝜔1 𝜔2 − 1) 𝜕𝜑1 𝜕𝜑2 ) V
of small-amplitude quasiperiodic solutions in the form of
V(𝑡, 𝑥) = 𝑢(𝜔1 𝑡 + 𝑥, 𝜔2 𝑡 + 𝑥) with the general nonlinearities = − 𝑓 (𝜑1 , 𝜑2 , V) ,
𝑓 = 𝑢3 +𝑂(𝑢4 ), which also do not depend on the time. In [32], (7)
they considered the existence of quasiperiodic solution of the
forced wave equation, in which the solutions are traveling in
(in opposite directions)
opposite directions. However, they did not give the regularity
of the solutions, and the results are the special case of our
(𝜕𝑡𝑡2 − 𝜕𝑥𝑥
2
) V = (𝜕𝑡 − 𝜕𝑥 ) ∘ (𝜕𝑡 + 𝜕𝑥 ) V
results in Section 4. Moreover, we mention the work [33] of
Bambusi, where a simple proof of an infinite-dimensional
= [𝜔1 𝜕𝜑1 + 𝜔2 𝜕𝜑2 + 𝜕𝜑1 − 𝜕𝜑2 ]
extension of the Lyapunov center theorem is given.
In this paper, for the completely resonant wave equation
∘ [𝜔1 𝜕𝜑1 + 𝜔2 𝜕𝜑2 − 𝜕𝜑1 + 𝜕𝜑2 ] V
(1) subjecting to the quasiperiodic forced terms, we will prove
(8)
the existence and regularity of quasiperiodic solutions with
two frequencies, 𝜔1 , 𝜔2 , in both of the following two cases. = ((𝜔12 − 1) 𝜕𝜑21 + (𝜔22 − 1) 𝜕𝜑22

Case (A1). The first case considers the wave traveling in the + 2 (𝜔1 𝜔2 + 1) 𝜕𝜑1 𝜕𝜑2 ) V
same directions 𝑢(𝑡, 𝑥) = V(𝜔1 𝑡 + 𝑥, 𝜔2 𝑡 + 𝑥).
= − 𝑓 (𝜑1 , 𝜑2 , V) .
Case (A2). The second case considers the wave traveling in
opposite directions 𝑢(𝑡, 𝑥) = V(𝜔1 𝑡 − 𝑥, 𝜔2 𝑡 + 𝑥). We assume that the quasiperiodic forced term 𝑓 : T 2 × R →
R,
2. Main Results
𝑓 (𝜑1 , 𝜑2 , V) = 𝑎3 (𝜑1 , 𝜑2 ) V3 + 𝑂 (V4 ) , (9)
We look for quasiperiodic solutions 𝑢(𝑡, 𝑥) of (1) of the
following form: is analytic in V but has only finite regularity in 𝜑1 , 𝜑2 . More
precisely,
(in the same directions)
(H) 𝑓(𝜑1 , 𝜑2 , V) := ∑∞ 𝑘
𝑘=3 𝑎𝑘 (𝜑1 , 𝜑2 )V , and the coefficients
1 2
𝑢 (𝑡, 𝑥) = V (𝜔1 𝑡 + 𝑥, 𝜔2 𝑡 + 𝑥) = V (𝜑1 , 𝜑2 ) , 𝑎𝑘 (𝜑1 , 𝜑2 ) ∈ 𝐻 (T ) verify, for some 𝑟 > 0,
(5) ∑∞ 𝑘
𝑘=3 |𝑎𝑘 |𝐻1 𝑟 < ∞. The function 𝑓(𝜑1 , 𝜑2 , V) is not
V (𝜑1 , 𝜑2 ) = V (𝜑1 + 2𝑘1 𝜋, 𝜑2 + 2𝑘2 𝜋) , ∀𝑘1 , 𝑘2 ∈ Z, identically constant in (𝜑1 , 𝜑2 ).

We look for solutions V of (7)-(8) in the Banach space

with frequencies 𝜔 = (𝜔1 , 𝜔2 ) = (1 + 𝜖, 1 + 𝑎𝜖2 ), 𝑎 ∈ R− , or
H𝜎,𝑠
(in opposite directions)
{
𝑢 (𝑡, 𝑥) = V (𝜔1 𝑡 − 𝑥, 𝜔2 𝑡 + 𝑥) = V (𝜑1 , 𝜑2 ) , := {V (𝜑) = ∑ ̂V(𝑙1 ,𝑙2 ) 𝑒𝑖𝑙1 𝜑1 𝑒𝑖𝑙2 𝜑2 : ̂V(∗𝑙 ,𝑙 ) = ̂V(−𝑙1 ,−𝑙2 ) ,
1 2
(6) { (𝑙1 ,𝑙2 )∈Z2
V (𝜑1 , 𝜑2 ) = V (𝜑1 + 2𝑘1 𝜋, 𝜑2 + 2𝑘2 𝜋) , ∀𝑘1 , 𝑘2 ∈ Z,
󵄨 󵄨 󵄨 󵄨 𝑠 }
|V|𝜎,𝑠 = ∑ 󵄨󵄨󵄨󵄨̂V(𝑙1 ,𝑙2 ) 󵄨󵄨󵄨󵄨 𝑒|𝑙2 |𝜎 (max {0, 󵄨󵄨󵄨𝑙1 󵄨󵄨󵄨}) < +∞} ,
with frequencies 𝜔 = (𝜔1 , 𝜔2 ) = (1 + 𝜖, 1 + 𝑎𝜖), 𝑎 ∈ R− , (𝑙1 ,𝑙2 )∈Z2 }
imposing the frequencies 𝜔 = (𝜔1 , 𝜔2 ) to be close to linear (10)
frequency 1. Therefore, finding the quasiperiodic solutions of
(1) with frequencies, respectively, (𝜔1 , 𝜔2 ) is equivalent to where 𝑙 = (𝑙1 , 𝑙2 ) ∈ Z2 , ̂V(𝑙∗1 ,𝑙2 ) denotes its complex conjugate,
finding 2𝜋 periodic solutions with respect to (𝜑1 , 𝜑2 ) for the and 𝜎 > 0, 𝑠 ≥ 0.
following equations: The space H𝜎,𝑠 is a Banach algebra with respect to
multiplications of functions, namely,
(in the same directions)
󵄨󵄨 󵄨 󵄨 󵄨 󵄨 󵄨
V1 , V2 ∈ H𝜎,𝑠 󳨐⇒ V1 V2 ∈ H𝜎,𝑠 , 󵄨󵄨V1 V2 󵄨󵄨󵄨𝜎,𝑠 ≤ 𝐶󵄨󵄨󵄨V1 󵄨󵄨󵄨𝜎,𝑠 󵄨󵄨󵄨V2 󵄨󵄨󵄨𝜎,𝑠 .
(𝜕𝑡𝑡2 − 2
𝜕𝑥𝑥 )V = (𝜕𝑡 − 𝜕𝑥 ) ∘ (𝜕𝑡 + 𝜕𝑥 ) V (11)

= [𝜔1 𝜕𝜑1 + 𝜔2 𝜕𝜑2 − 𝜕𝜑1 − 𝜕𝜑2 ] We will prove the following theorems.
Abstract and Applied Analysis 3

Theorem 1. Assume that the nonlinearity 𝑓 satisfies (H) and for ∀𝑙1 , 𝑙2 ∈ Z \ {0}, and 0 < 𝛾 < 1/4, 𝜖0 ∈ (0, 1/2). There
0, ∀(𝜑1 , 𝜑2 ) ∈ T 2 . Let B𝛾 be the uncountable
𝑎3 (𝜑1 , 𝜑2 ) ≠ exist positive numbers 𝜎, 𝜖, 𝐶, 𝑠 > 2, such that, ∀(𝑎, 𝜖) ∈ D𝛾 ,
zero-measure Cantor set (8) admits solutions in the form of
B𝛾 V𝑎,𝜖 (𝑡, 𝑥) = V (𝜖, 𝜑1 , 𝜑2 )

:= { (𝑎, 𝜖) ∈ R− × R, 1 + 𝜖 ≠
0, 1 + 𝑎𝜖2 ≠
0, 2 + 𝑎𝜖2 ≠
0, = √|𝜖| (̂
𝑞0 + 𝑞̂− (𝜑2 ) + 𝑞̂+ (𝜑1 ) (18)

2+𝜖 +𝑝̂ (𝜖, 𝜑1 , 𝜑2 )) ∈ H𝜎,𝑠 ,

( , 𝑎𝜖) ∈ C𝛾1 ∩ (1 − 𝜖0 , 1 + 𝜖0 )
2 + 𝑎𝜖2 satisfying
1+𝜖 1
×C𝛾2 ∩ (−𝜖0 , 𝜖0 ) , ∉ Q, for 𝜖0 ∈ (0, )} , 󵄨󵄨 ̂ 󵄨 |𝜖|
1 + 𝑎𝜖2 2 󵄨󵄨𝑝 (𝜖, 𝜑1 , 𝜑2 )󵄨󵄨󵄨𝜎,𝑠 ≤ 𝐶 ( ) . (19)
(12) 𝛾

where C𝛾𝑖 , 𝑖 = 1, 2, are sets of badly approximate numbers As a consequence, (1) possesses the quasiperiodic solutions,
defined as 𝑢𝑎,𝜖 (𝑡, 𝑥) = V𝑎,𝜖 ((1 + 𝜖)𝑡 + 𝑥, (1 + 𝑎𝜖)𝑡 − 𝑥), traveling in opposite
directions.
2+𝜖 󵄨󵄨 2 + 𝜖 󵄨󵄨󵄨 𝛾
C𝛾1 := { : 󵄨󵄨󵄨󵄨𝑙2 + 𝑙 󵄨󵄨 > 󵄨󵄨 󵄨󵄨 } ,
2 + 𝑎𝜖2 󵄨 2 + 𝑎𝜖2 1 󵄨󵄨 󵄨󵄨𝑙1 󵄨󵄨 Remark 3. The quasiperiodic solutions of traveling waves we
(13) obtained are different from the ones got by KAM methods
󵄨 󵄨 𝛾 since the quasiperiodic solutions we get depending on 𝑥 and
C𝛾2 := {𝑎𝜖 : 󵄨󵄨󵄨𝑙1 + 𝑎𝜖𝑙2 󵄨󵄨󵄨 > 󵄨󵄨 󵄨󵄨 }
󵄨󵄨𝑙2 󵄨󵄨 𝑡 are coupled and in the form of the traveling waves.
for ∀𝑙1 , 𝑙2 ∈ Z \ {0}, and 0 < 𝛾 < 1/4. There exist constants Remark 4. We can get the similar result with more general
𝜎 > 0, 𝑠 > 2, and 𝜖(𝑅) > 0, such that, for (𝑎, 𝜖) ∈ B𝛾 , |𝜖|/𝛾 < nonlinearity, such as 𝑓(𝜔1 𝑡, 𝜔2 𝑡, 𝑥, 𝑢) = 𝑎𝑑 (𝜑1 , 𝜑2 )𝑢𝑑 +
𝜖(𝑅), there exists a solution V(𝜖, 𝜑1 , 𝜑2 ) ∈ H𝜎,𝑠 of (7), having 𝑂(𝑢(𝑑+1) ), for any 𝑑 ∈ N, 𝑑 ≥ 3.
the form
V (𝜖, 𝜑1 , 𝜑2 ) This paper is organized as follows: we first prove the
(14) existence of quasiperiodic solutions, at the first order, to the
𝑞0 + 𝑞̂− (𝜑2 ) + 𝑞̂+ (𝜑1 + 𝜑2 ) + 𝑝̂ (𝜖, 𝜑1 , 𝜑2 )) ,
= |𝜖| (̂ superposition of traveling waves, traveling in the same direc-
tions. In Section 4, we prove the existence of quasiperiodic
with solutions traveling in opposite directions.
󵄨󵄨 ̂ 󵄨 |𝜖|
󵄨󵄨𝑝 (𝜖, 𝜑1 , 𝜑2 )󵄨󵄨󵄨𝜎,𝑠 ≤ 𝐶 ( ) , (15)
𝛾 3. Waves Traveling in the Same Directions
where 𝐶 is a constant. As a consequence, (1) possesses Substituting 𝜔1 = 1 + 𝜖, 𝜔2 = 1 + 𝑎𝜖2 into (7), we can obtain
the quasiperiodic solutions, traveling in the same directions, the equations
𝑢(𝜖, 𝑡, 𝑥) = V(𝜖, 𝜔1 𝑡 + 𝑥, 𝜔2 𝑡 + 𝑥), with two frequencies
(𝜔1 , 𝜔2 ) = (1 + 𝜖, 1 + 𝑎𝜖2 ). LV + 𝑓 (𝜑1 , 𝜑2 , V) = 0, (20)

Theorem 2. Assume that 𝑓 satisfies assumption (H) and where, see (7), we have
0, ∀(𝜑1 , 𝜑2 ) ∈ T 2 . Let D𝛾 ⊂ (−𝜖0 , 𝜖0 ) × (−𝜖0 , 𝜖0 )
𝑎3 (𝜑1 , 𝜑2 ) ≠ L = ((𝜔12 − 1) 𝜕𝜑21 + (𝜔22 − 1) 𝜕𝜑22 + 2 (𝜔1 𝜔2 − 1) 𝜕𝜑1 𝜕𝜑2 ) V
be the uncountable zero-measure Cantor set
𝑎𝜖 𝜖 = 𝜖 (2𝜕𝜑21 + 2𝜕𝜑1 𝜕𝜑2 )
D𝛾 := {(𝑎, 𝜖) ∈ R− × R, ( , ) ∈ E𝛾 ,
2 + 𝜖 2 + 𝑎𝜖
1+𝜖 + 𝜖2 (𝜕𝜑21 + (2𝑎 + 𝑎2 𝜖2 ) 𝜕𝜑22 + 2𝑎 (1 + 𝜖) 𝜕𝜑1 𝜕𝜑2 ) ,
∉ Q, 1 + 𝜖 ≠
0, 1 + 𝑎𝜖 ≠
0, 2 + 𝑎𝜖 ≠
0} , (21)
1 + 𝑎𝜖
(16)
and 𝑓(𝜑1 , 𝜑2 , V) = 𝑎3 (𝜑1 , 𝜑2 )V3 + 𝑂(V4 ). To prove Theorem 1,
where E𝛾 is a set of badly approximate numbers defined as instead of looking for solutions of (7) in a shrinking neigh-
borhood of zero, it is convenient to perform the rescaling
E𝛾 V(𝜑1 , 𝜑2 ) → 𝜖V(𝜑1 , 𝜑2 ), enhancing the relation between the
amplitude and the frequencies. Without confusion, we define
𝑎𝜖 𝜖
:= { ( , ) := (𝜖1 , 𝜖2 ) ∈ (−𝜖0 , 𝜖0 ) × (−𝜖0 , 𝜖0 ) :
2 + 𝜖 2 + 𝑎𝜖 L𝑎,𝜖 = (2𝜕𝜑21 + 2𝜕𝜑1 𝜕𝜑2 )

󵄨󵄨 󵄨 𝛾 󵄨 󵄨 𝛾 + 𝜖 (𝜕𝜑21 + (2𝑎 + 𝑎2 𝜖2 ) 𝜕𝜑22 + 2𝑎 (1 + 𝜖) 𝜕𝜑1 𝜕𝜑2 ) (22)

󵄨󵄨𝑙1 + 𝜖1 𝑙2 󵄨󵄨󵄨 > 󵄨󵄨 󵄨󵄨 , 󵄨󵄨󵄨𝑙2 + 𝜖2 𝑙1 󵄨󵄨󵄨 > 󵄨󵄨 󵄨󵄨 }
󵄨󵄨𝑙2 󵄨󵄨 󵄨󵄨𝑙1 󵄨󵄨
(17) = L0 + 𝜖L1 ,
4 Abstract and Applied Analysis

so the problem becomes The critical points of the unperturbed functional Ψ0 :

H𝜎,𝑠 → R form an infinite-dimensional linear space 𝑄, and
L𝑎,𝜖 + 𝜖𝑓 (𝜑1 , 𝜑2 , V, 𝜖) = 0. (23) they are the solutions of the equation
To find the solutions of (23), we will apply the Lyapunov- L0 𝑞 = (2𝜕𝜑21 + 2𝜕𝜑1 𝜕𝜑2 ) 𝑞 = 0. (28)
Schmidt reduction method which leads to solving separately
a “range equation” and a “bifurcation equation.” In order to The space 𝑄 can be written as
solve the range equation (avoiding small divisor problems),
we restrict 𝜖 to the uncountable zero-measure set B𝛾 for
{
Theorem 1, and we apply the Contraction Mapping Theorem; 𝑄 = {𝑞 = ∑ 𝑞̂𝑙1 ,𝑙2 𝑒𝑖𝑙1 𝜑1 𝑒𝑖𝑙2 𝜑2 ∈ H𝜎,𝑠 | 𝑞̂𝑙1 ,𝑙2 = 0,
similar nonresonance conditions have been employed, for (𝑙1 ,𝑙2 )∈Z2
{
example, in [21–23, 25, 29, 30]. (29)
Equation (23) is the Euler-Lagrange equation of the action }
functional Ψ𝜖 ∈ 𝐶1 (H𝜎,𝑠 , R) defined by for 𝑙1 (2𝑙1 + 2𝑙2 ) ≠
0} .
}
2
Ψ𝜖 (V) := ∫ (𝜕𝜑1 V) + (𝜕𝜑1 V) (𝜕𝜑2 V)
T2 In view of the variational argument that we will use to solve
the bifurcation equation, we split 𝑄 as 𝑄 = 𝑄+ + 𝑄0 + 𝑄− ,
1 2 2𝑎 + 𝑎2 𝜖2 2 where
+ 𝜖 ( (𝜕𝜑1 V) + (𝜕𝜑2 V)
2 2
𝑄+ := {𝑞 ∈ 𝑄 : 𝑞̂𝑙1 ,𝑙2 = 0, for (𝑙1 , 𝑙2 ) ∉ Λ + }
(24)
+𝑎 (1 + 𝜖) (𝜕𝜑1 V) (𝜕𝜑2 V) ) = {𝑞+ := 𝑞+ (𝜑) ∈ H0𝜎,𝑠 } ,

− 𝜖𝐹 (𝜑1 , 𝜑2 , V, 𝜖) 𝑄0 := {𝑞 : 𝑞0,0 ∈ R} , (30)

= Ψ0 (V) + 𝜖Ψ1 (V, 𝜖) , 𝑄− := {𝑞 ∈ 𝑄 : 𝑞̂𝑙1 ,𝑙2 = 0, for (𝑙1 , 𝑙2 ) ∉ Λ − }

where = {𝑞− := 𝑞− (𝜑) ∈ H0𝜎,𝑠 } ,
V
𝐹 (𝜑1 , 𝜑2 , V, 𝜖) := ∫ 𝑓 (𝜑1 , 𝜑2 , 𝜉, 𝜖) 𝑑𝜉, with
0

2 Λ + := {(𝑙1 , 𝑙2 ) ∈ Z2 : 𝑙1 = 0, 𝑙2 ≠
0} ,
Ψ0 (V) = ∫ (𝜕𝜑1 V) + (𝜕𝜑1 V) (𝜕𝜑2 V) ,
T2
Λ − := {(𝑙1 , 𝑙2 ) ∈ Z2 : 𝑙1 + 𝑙2 = 0, (𝑙1 , 𝑙2 ) ≠(0, 0)} , (31)
2 2
1 2 2𝑎 + 𝑎 𝜖 2
Ψ1 (V, 𝜖) := ∫ ( (𝜕𝜑1 V) + (𝜕𝜑2 V) (25)
T 2 2 2 Λ 0 := {(𝑙1 , 𝑙2 ) ∈ Z2 : (𝑙1 , 𝑙2 ) ≡ (0, 0)} .

We will also use in 𝑄 the norm

+𝑎 (1 + 𝜖) (𝜕𝜑1 V) (𝜕𝜑2 V) )
󵄨󵄨 󵄨󵄨2 󵄨 󵄨2 󵄨 󵄨2
󵄨󵄨𝑞󵄨󵄨𝐻1 = 󵄨󵄨󵄨𝑞+ 󵄨󵄨󵄨𝐻1 (T) + 𝑞0,0 + 󵄨󵄨󵄨𝑞− 󵄨󵄨󵄨𝐻1 (T)
2

− 𝜖𝐹 (𝜑1 , 𝜑2 , V, 𝜖) .
󵄨 󵄨2 󵄨 󵄨2 (32)
∼ ∑ 𝑞̂𝑙21 ,𝑙2 (󵄨󵄨󵄨𝑙1 󵄨󵄨󵄨 + 󵄨󵄨󵄨𝑙2 󵄨󵄨󵄨 + 1) .
To find critical points of Ψ𝜖 (V), we perform a variational (𝑙1 ,𝑙2 )∈Λ + ∪Λ 0 ∪Λ +
Lyapunov-Schmidt reduction inspired by Berti and Bolle [22,
23, 29]; see also Ambrosetti and Badiale [34]. So, we can decompose the space H𝜎,𝑠 = 𝑄 + 𝑃, where

3.1. The Variational Lyapunov-Schmidt Reduction. The oper- {

ator L𝑎,𝜖 is diagonal defined on the Banach space H𝜎,𝑠 under 𝑃 := {𝑝 = ∑ 𝑝̂𝑙1 ,𝑙2 𝑒𝑖𝑙1 𝜑1 𝑒𝑖𝑙2 𝜑2 ∈ H𝜎,𝑠 | 𝑝̂𝑙1 ,𝑙2 = 0,
𝑙1 ,𝑙2 ∈Z
the Fourier basis 𝑒𝑙 = 𝑒𝑖𝑙1 𝜑1 𝑒𝑖𝑙2 𝜑2 with eigenvalues {
(33)
−𝐷𝑙 = (2𝑙12 + 2𝑙1 𝑙2 ) + 𝜖 (𝑙12 2 2
+ (2𝑎 + 𝑎 𝜖 ) 𝑙22 + 2𝑎 (1 + 𝜖) 𝑙1 𝑙2 ) }
for 𝑙1 (2𝑙1 + 2𝑙2 ) = 0} .
= (𝑙1 + 𝑎𝜖𝑙2 ) ((2 + 𝜖) 𝑙1 + (2 + 𝑎𝜖2 ) 𝑙2 ) . }
(26) Projecting (23) onto the closed subspaces 𝑄 and 𝑃, setting
V = 𝑞 + 𝑝 ∈ H𝜎,𝑠 with 𝑞 ∈ 𝑄 and 𝑝 ∈ 𝑃, we obtain
So, we have
(𝑄) L1 [𝑞] + Π𝑄𝑓 (𝜑1 , 𝜑2 , 𝑞 + 𝑝, 𝜖) = 0,
L𝑎,𝜖 [V] = ∑ 𝐷𝑙 ̂V𝑙1 ,𝑙2 𝑒𝑖𝑙1 𝜑1 𝑒𝑖𝑙2 𝜑2 , ∀V ∈ H𝜎,𝑠 . (34)
(27)
(𝑙1 ,𝑙2 )∈Z2 (𝑃) L𝑎,𝜖 [𝑝] + 𝜖Π𝑃 𝑓 (𝜑1 , 𝜑2 , 𝑞 + 𝑝, 𝜖) = 0,
Abstract and Applied Analysis 5

where Π𝑄 : H𝜎,𝑠 → 𝑄, Π𝑃 : H𝜎,𝑠 → 𝑃 are the projectors, for ∀𝑙1 , 𝑙2 ∈ Z \ {0}, and 0 < 𝛾 < 1/4. C𝛾𝑖 , 𝑖 = 1, 2, accumulate
respectively, onto 𝑄 and 𝑃. at 1 and zero, respectively, from both the right and the left;
In order to prove analyticity of the solutions and to high- see [21, 31, 35].
light the compactness of the problem, we perform a finite- The operator L𝜖 is diagonal in the Fourier basis
dimensional Lyapunov-Schmidt reduction, introducing the {𝑒𝑖𝑙1 𝜑1 𝑒𝑖𝑙2 𝜑2 , (𝑙1 , 𝑙2 ) ∈ Z2 } with eigenvalues
decomposition 𝑄 = 𝑄1 + 𝑄2 , where
𝐷𝑙1 ,𝑙2 = − (𝑙1 + 𝑎𝜖𝑙2 ) ((2 + 𝜖) 𝑙1 + (2 + 𝑎𝜖2 ) 𝑙2 )
{ } 2+𝜖
(41)
𝑄1 := {𝑞 = ∑ 𝑞̂𝑙1 ,𝑙2 𝑒𝑖𝑙1 𝜑1 𝑒𝑖𝑙2 𝜑2 ∈ 𝑄} , = − (2 + 𝑎𝜖2 ) (𝑙1 + 𝑎𝜖𝑙1 ) (𝑙2 + 𝑙 ).
{ |𝑙1 |+|𝑙2 |≤𝑁
} 2 + 𝑎𝜖2 1
(35) Lemma 5. For (𝑎, 𝜖) ∈ B𝛾 , the eigenvalues 𝐷𝑙1 ,𝑙2 of L𝑎,𝜖
{ } restricted to 𝑃 satisfy
𝑄2 := {𝑞 = ∑ 𝑞̂𝑙1 ,𝑙2 𝑒𝑖𝑙1 𝜑1 𝑒𝑖𝑙2 𝜑2 ∈ 𝑄} .
{ |𝑙1 |+|𝑙2 |≤𝑁
} 󵄨󵄨󵄨𝐷 󵄨󵄨󵄨 = 󵄨󵄨󵄨(𝑙 + 𝑎𝜖𝑙 ) ((2 + 𝜖) 𝑙 + (2 + 𝑎𝜖2 ) 𝑙 )󵄨󵄨󵄨 > 𝛾,
󵄨󵄨 𝑙1 ,𝑙2 󵄨󵄨 󵄨󵄨 1 2 1 2 󵄨󵄨
(42)
Setting 𝑞 = 𝑞1 + 𝑞2 with 𝑞1 ∈ 𝑄1 and 𝑞2 ∈ 𝑄2 , we finally get ∀𝑙1 , 𝑙2 ≠ 0.

(𝑄1 ) L1 [𝑞1 ] + Π𝑄1 [𝑓 (𝜑1 , 𝜑2 , 𝑞1 + 𝑞2 + 𝑝, 𝜖)] = 0 As a consequence, the operator L𝑎,𝜖 : 𝑃 → 𝑃 has a bounded
(36) inverse L−1
𝑎,𝜖 and satisfies
⇐⇒ 𝑑Ψ𝜖 (V) [ℎ] = 0, ∀ℎ ∈ 𝑄1 ;
󵄨󵄨 −1 󵄨󵄨 |ℎ|
(𝑄2 ) L1 [𝑞2 ] + Π𝑄2 [𝑓 (𝜑1 , 𝜑2 , 𝑞1 + 𝑞2 + 𝑝, 𝜖)] = 0 󵄨󵄨L𝑎,𝜖 [ℎ]󵄨󵄨 ≤ 𝜎,𝑠 , ∀ℎ ∈ 𝑃. (43)
󵄨 󵄨𝜎,𝑠 𝛾
(37)
⇐⇒ 𝑑Ψ𝜖 (V) [ℎ] = 0, ∀ℎ ∈ 𝑄2 ; Proof. Denote by [𝑥] the nearest integer close to 𝑥 and by
{𝑥} = 𝑥 − [𝑥] its fractional part. If both 𝑙1 ≠− [𝑎𝜖𝑙2 ] and
(𝑃) L𝑎,𝜖 [𝑝] + 𝜖Π𝑃 [𝑓 (𝜑1 , 𝜑2 , 𝑞1 + 𝑞2 + 𝑝, 𝜖)] = 0 𝑙2 ≠− [((2 + 𝜖)/(2 + 𝑎𝜖2 ))𝑙1 ], we have
(38)
⇐⇒ 𝑑Ψ𝜖 (V) [ℎ] = 0, ∀ℎ ∈ 𝑃, 󵄨󵄨 󵄨 󵄨 󵄨 󵄨󵄨 2+𝜖 󵄨󵄨
󵄨󵄨𝐷𝑙1 ,𝑙2 󵄨󵄨󵄨 = 󵄨󵄨󵄨󵄨(𝑙1 + 𝑎𝜖𝑙2 )󵄨󵄨󵄨󵄨 ⋅ 󵄨󵄨󵄨(2 + 𝑎𝜖2 )󵄨󵄨󵄨 󵄨󵄨󵄨(𝑙2 + 𝑙 ) 󵄨󵄨 > 1.
󵄨
󵄨 󵄨 󵄨 󵄨 󵄨󵄨 2 + 𝑎𝜖2 󵄨󵄨
1
where Π𝑄𝑖 : H𝜎,𝑠 → 𝑄𝑖 are the projectors onto 𝑄𝑖 (𝑖 = 1, 2), (44)
and Π𝑃 : H𝜎,𝑠 → 𝑃 is the projector onto 𝑃. We will solve first
the (𝑄2 )-(𝑃)-equations for all |𝑞1 |𝐻1 ≤ 2𝑅, provided 𝜖 belongs If 𝑙1 = −[𝑎𝜖𝑙2 ], then |𝑙1 | < (1/2)|𝑙2 |, so that |𝑙2 + ((2 + 𝜖)/(2 +
to a suitable Cantor-like set, |𝜖| ≤ 𝜖0 (𝑅) is sufficiently small, 𝑎𝜖2 ))𝑙1 | > (1/2)|𝑙2 |. This implies that
and 𝑁 ≥ 𝑁0 (𝑅) is large enough (see Lemma 7). Next, we 󵄨󵄨 󵄨
will solve the 𝑄1 -equation by means of a variational linking 󵄨󵄨𝐷𝑙1 ,𝑙2 󵄨󵄨󵄨
󵄨 󵄨
argument; see Section 3.4.
󵄨 󵄨 󵄨 󵄨 󵄨󵄨 2 + 𝜖 󵄨󵄨
= 󵄨󵄨󵄨𝑙1 + 𝑎𝜖𝑙2 󵄨󵄨󵄨 ⋅ 󵄨󵄨󵄨󵄨2 + 𝑎𝜖2 󵄨󵄨󵄨󵄨 󵄨󵄨󵄨󵄨 2
𝑙1 + 𝑙2 󵄨󵄨󵄨󵄨
󵄨 2 + 𝑎𝜖 󵄨
3.2. The (𝑄2 )-(𝑃)-Equations. We first prove that L𝑎,𝜖 restrict-
ed to 𝑃 has a bounded inverse when (𝑎, 𝜖) belongs to the 𝛾 󵄨󵄨󵄨
≥ 󵄨󵄨 󵄨󵄨 ⋅ 󵄨󵄨󵄨 (2 + 𝜖)
uncountable zero-measure set 󵄨󵄨𝑙2 󵄨󵄨 󵄨
𝑎𝜖 𝜖 󵄨󵄨
B𝛾 × (𝜖 { 𝑙2 } + 2𝑙2 + 𝑎𝜖 (1 − ) 𝑙2 )󵄨󵄨󵄨󵄨 ≥ 𝛾.
2+𝜖 2+𝜖 󵄨
(45)
:= { (𝑎, 𝜖) ∈ R− × R, 1 + 𝜖 ≠
0, 1 + 𝑎𝜖2 ≠
0, 2 + 𝑎𝜖2 ≠
0,
In the same way, if 𝑙2 = −[((2 + 𝜖)/(2 + 𝑎𝜖2 ))𝑙1 ], we have
2+𝜖
( , 𝑎𝜖) ∈ C𝛾1 ∩ (1 − 𝜖0 , 1 + 𝜖0 ) 󵄨󵄨 󵄨 𝛾 󵄨󵄨 (2 + 𝜖) 𝑎 2 − 2𝑎𝜖 󵄨󵄨󵄨󵄨
2 + 𝑎𝜖2 󵄨󵄨𝐷𝑙1 ,𝑙2 󵄨󵄨󵄨 ≥ 󵄨 󵄨 ⋅ 󵄨󵄨󵄨󵄨(2 + 𝑎𝜖2 ) (𝜖 { 𝑙 } + 𝑙 )󵄨 ≥ 𝛾.
󵄨 󵄨 󵄨󵄨𝑙 󵄨󵄨 󵄨󵄨
󵄨 1󵄨 2 + 𝑎𝜖2 1
2 + 𝑎𝜖2 1 󵄨󵄨󵄨
1+𝜖 1
×C𝛾2 ∩ (−𝜖0 , 𝜖0 ) , ∉ Q, for 𝜖0 ∈ (0, ) } , (46)
1 + 𝑎𝜖2 2
(39) So, the operator L𝑎,𝜖 restricted to 𝑃 has a bounded inverse
and satisfies
where C𝛾𝑖 , 𝑖 = 1, 2, is a set of badly approximate numbers 󵄨󵄨̂ 󵄨󵄨 |𝑙2 |𝜎
defined by 󵄨󵄨 −1 󵄨󵄨 󵄨󵄨ℎ𝑙1 ,𝑙2 󵄨󵄨 𝑒 (max {0, 󵄨󵄨󵄨󵄨𝑙1 󵄨󵄨󵄨󵄨})𝑠
󵄨󵄨L𝑎,𝜖 [ℎ]󵄨󵄨 = ∑ 󵄨 󵄨
󵄨 󵄨𝜎,𝑠 𝐷𝑙1 ,𝑙2
2+𝜖 󵄨󵄨 2 + 𝜖 󵄨󵄨󵄨 𝛾 (𝑙1 ,𝑙2 )∈Z2

: 󵄨󵄨󵄨󵄨𝑙2 +
(47)
C𝛾1 : = { 𝑙 󵄨󵄨 > 󵄨󵄨 󵄨󵄨 } ,
2 + 𝑎𝜖2 󵄨 2 + 𝑎𝜖2 1 󵄨󵄨 󵄨󵄨𝑙1 󵄨󵄨 |ℎ|𝜎,𝑠
(40) ≤ , ∀ℎ ∈ 𝑃.
𝛾
󵄨 󵄨 𝛾
C𝛾2 : = {𝑎𝜖 : 󵄨󵄨󵄨𝑙1 + 𝑎𝜖𝑙2 󵄨󵄨󵄨 > 󵄨󵄨 󵄨󵄨 } ,
󵄨󵄨𝑙2 󵄨󵄨
6 Abstract and Applied Analysis

Lemma 6. The operator L1 : 𝑄2 → 𝑄2 has a bounded (i) (𝑞2 , 𝑝) ∈ 𝐵 ⇒ P(𝑞2 , 𝑝; 𝑞1 ) ∈ 𝐵;

inverse L−1
1 , satisfying (ii) |P(𝑞2 , 𝑝; 𝑞1 ) − P(̂ ̂ 𝑞1 )|𝜎,𝑠 ≤ 𝜂|(𝑞2 , 𝑝) − (̂
𝑞2 , 𝑝; ̂ 𝜎,𝑠 ,
𝑞2 , 𝑝)|
∀(𝑞2 , 𝑝), (̂ ̂
𝑞2 , 𝑝) ∈ 𝐵,
󵄨󵄨 −1 󵄨󵄨 |ℎ|
󵄨󵄨L1 [ℎ]󵄨󵄨 ≤ 𝜎,𝑠 . (48)
󵄨 󵄨𝜎,𝑠 𝑁2 where the constant 𝜂 ∈ (0, 1). In the following, 𝜅𝑖 (𝑖 =
1, 2, . . . , 5) denote different constants. By (48) and the Banach
Proof. L1 is diagonal in the Fourier basis of 𝑄 : 𝑒𝑖𝑙1 𝜑1 𝑒𝑖𝑙2 𝜑2 property of H𝜎,𝑠 ,
with (𝑙1 , 𝑙2 ) ∈ Λ + ∪ Λ 0 ∪ Λ − with eigenvalues 󵄨󵄨 󵄨 󵄨 −1 󵄨
󵄨󵄨P1 (𝑞2 , 𝑝; 𝑞1 )󵄨󵄨󵄨𝜎,𝑠 = 󵄨󵄨󵄨󵄨L1 Π𝑄2 𝑓 (𝜑1 , 𝜑2 , 𝑞1 + 𝑞2 + 𝑝, 𝜖)󵄨󵄨󵄨󵄨
(1 + 𝜖 (−2𝑎 + 𝑎2 𝜖)) 𝑙22 , if 𝑙1 + 𝑙2 = 0, 𝜅 󵄨 󵄨3 󵄨 󵄨3 󵄨 󵄨3
(55)
𝑑𝑙1 ,𝑙2 = { (49) ≤ 12 (󵄨󵄨󵄨𝑞1 󵄨󵄨󵄨𝜎,𝑠 + 󵄨󵄨󵄨𝑞2 󵄨󵄨󵄨𝜎,𝑠 + 󵄨󵄨󵄨𝑝󵄨󵄨󵄨𝜎,𝑠 ) .
𝑎 (2 + 𝑎𝜖2 ) 𝑙22 , if 𝑙1 = 0. 𝑁
Similarly, for (𝑎, 𝜖) ∈ B𝛾 , by (43), we have
The eigenvalues of L1 restricted to 𝑄2 (𝑁) verify |𝑑𝑙1 ,𝑙2 | ≥
𝑁2 /𝐶, where the constant 𝐶 depends on (𝜖, 𝑎), and (48) 󵄨󵄨 󵄨 󵄨 −1 󵄨
󵄨󵄨P2 (𝑞2 , 𝑝; 𝑞1 )󵄨󵄨󵄨𝜎,𝑠 = 󵄨󵄨󵄨󵄨𝜖L𝑎,𝜖 Π𝑃 𝑓 (𝜑1 , 𝜑2 , 𝑞1 + 𝑞2 + 𝑝, 𝜖)󵄨󵄨󵄨󵄨
holds.
𝜅2 |𝜖| 󵄨󵄨 󵄨󵄨3 (56)
󵄨 󵄨3 󵄨 󵄨3
≤ (󵄨𝑞 󵄨 + 󵄨󵄨𝑞 󵄨󵄨 + 󵄨󵄨𝑝󵄨󵄨 ) .
Fixed points of the nonlinear operator P : 𝑄2 ⊕ 𝑃 → 𝛾 󵄨 1 󵄨𝜎,𝑠 󵄨 2 󵄨𝜎,𝑠 󵄨 󵄨𝜎,𝑠
𝑄2 ⊕ 𝑃 defined by
For all 𝑞1 ∈ 𝑄1 (𝑁), setting [𝑥] = max{0, |𝑥|}, we can get
P (𝑞2 , 𝑝, 𝑞1 ) := (−L−1
1 Π𝑄2 𝑓 (𝜑1 , 𝜑2 , 𝑞1 + 𝑞2 + 𝑝, 𝜖) , 󵄨󵄨󵄨𝑞1 󵄨󵄨󵄨
(50) 󵄨 󵄨𝜎,𝑠
−𝜖L−1 󵄨 󵄨 󵄨󵄨 󵄨
𝑎,𝜖 Π𝑃 𝑓 (𝜑1 , 𝜑2 , 𝑞1 + 𝑞2 + 𝑝, 𝜖)) = ∑ 󵄨󵄨󵄨󵄨𝑞̂0,𝑙2 󵄨󵄨󵄨󵄨 𝑒|𝑙2 |𝜎 + ∑ 󵄨󵄨𝑞̂−𝑙2 ,𝑙2 󵄨󵄨󵄨 𝑒|𝑙2 |𝜎 [−𝑙2 ]𝑠
󵄨 󵄨
|𝑙2 |≤𝑁,𝑙1 =0 |𝑙2 |≤𝑁,𝑙1 +𝑙2 =0
are solutions of the (𝑄2 )-(𝑃)-equations. Using the Contrac-
tion Mapping Theorem, we can prove the following lemma.
󵄨󵄨 󵄨 𝑠
≤ 𝑒𝑁𝜎 ( ∑ 󵄨󵄨𝑞̂𝑙1 ,𝑙2 󵄨󵄨󵄨 ([−𝑙2 − 𝑙22 ])
Lemma 7. For any 𝑅 > 0, there exist an integer 𝑁0 (𝑅) ∈ N + 󵄨 󵄨
|𝑙2 |≤𝑁,𝑙1 =0
and positive constants 𝜖0 (𝑅) > 0, 𝐶0 (𝑅) > 0 such that
󵄨 󵄨 󵄨󵄨 󵄨
∀󵄨󵄨󵄨𝑞1 󵄨󵄨󵄨𝐻1 ≤ 2𝑅, ∀𝜖 ∈ B𝛾 , |𝜖| 𝛾−1 ≤ 𝜖0 (𝑅) , + ∑ 󵄨󵄨𝑞̂−𝑙2 ,𝑙2 󵄨󵄨󵄨 [−𝑙2 ]𝑠 )
󵄨 󵄨
(51) |𝑙2 |≤𝑁,𝑙1 +𝑙2 =0
∀𝑁 ≥ 𝑁0 (𝑅) : 0 ≤ 𝑁𝜎 ≤ 1,
1/2 1/2
󵄨󵄨 󵄨2 1
and there exists a unique solution (𝑞2 (𝑞1 ), 𝑝(𝑞1 )) := ≤ 𝜅3 (( ∑ 󵄨󵄨𝑞̂0,𝑙2 󵄨󵄨󵄨 [𝑙2 ]2 ) (∑ )
(𝑞2 (𝜖, 𝑁, 𝑞1 ), 𝑝(𝜖, 𝑁, 𝑞1 )) ∈ 𝑄2 ⊕ 𝑃 of the (𝑄2 )-(𝑃)-equations 󵄨 󵄨 2
|𝑙2 |≤𝑁,𝑙1 =0 𝑙2 ∈Z [𝑙2 ]
satisfying
1/2
󵄨󵄨 󵄨 𝐶 (𝑅) 󵄨󵄨 󵄨2
󵄨󵄨𝑞2 (𝜖, 𝑁, 𝑞1 )󵄨󵄨󵄨𝜎,𝑠 ≤ 0 2 , +( ∑ 󵄨󵄨𝑞̂−𝑙2 ,𝑙2 󵄨󵄨󵄨 [𝑙2 ]2 )
𝑁 󵄨 󵄨
(52) | 2 | 1 +𝑙2 =0
𝑙 ≤𝑁,𝑙
󵄨󵄨 󵄨󵄨 −1
󵄨󵄨𝑝 (𝜖, 𝑁, 𝑞1 )󵄨󵄨𝜎,𝑠 ≤ 𝐶0 (𝑅) |𝜖| 𝛾 .
1/2
1
Moreover, the map 𝑞1 → (𝑞2 (𝑞1 ), 𝑝(𝑞1 )) is 𝐶1 (𝐵2𝑅 , 𝑄2 ⊕ 𝑃) × (∑ 2(1−𝑠)
) )
𝑙2 ∈Z ([𝑙2 ])
and
󵄨 󵄨
󵄨󵄨 󸀠 󵄨 𝐶 (𝑅) ≤ 𝜅4 󵄨󵄨󵄨𝑞1 󵄨󵄨󵄨𝐻1
󵄨󵄨𝑞2 (𝑞1 ) [ℎ]󵄨󵄨󵄨 ≤ 0 2 |ℎ|𝐻1 ,
󵄨 󵄨𝜎,𝑠 𝑁 (57)
(53)
󵄨󵄨 󸀠 󵄨󵄨
󵄨󵄨𝑝 (𝑞1 ) [ℎ]󵄨󵄨 ≤ 𝐶0 (𝑅) |𝜖| 𝛾−1 |ℎ|𝐻1 ∀ℎ ∈ 𝑄1 . whenever 0 ≤ 𝑁𝜎 ≤ 1, 0 ≤ 𝑠 < 1/2. Thus, ∀|𝑞1 |𝐻1 ≤
󵄨 󵄨𝜎,𝑠
2𝑅, |𝑞2 |𝜎,𝑠 ≤ 𝜌1 , |𝑝|𝜎,𝑠 ≤ 𝜌2 , we get
Proof. Let us consider the ball
󵄨󵄨 󵄨 𝜅
󵄨󵄨P1 (𝑞2 , 𝑝; 𝑞1 )󵄨󵄨󵄨𝜎,𝑠 ≤ 52 (𝑅 + 𝜌1 + 𝜌2 ) ,
3 3 3
󵄨 󵄨 󵄨 󵄨 𝑁
𝐵 := {(𝑞2 , 𝑝) ∈ 𝑄2 ⊕ 𝑃, 󵄨󵄨󵄨𝑞2 󵄨󵄨󵄨𝜎,𝑠 ≤ 𝜌1 , 󵄨󵄨󵄨𝑝󵄨󵄨󵄨𝜎,𝑠 ≤ 𝜌2 } (54)
(58)
󵄨󵄨 󵄨 𝜅 |𝜖| 3
󵄨󵄨P2 (𝑞2 , 𝑝; 𝑞1 )󵄨󵄨󵄨𝜎,𝑠 ≤ 5 (𝑅 + 𝜌1 + 𝜌2 ) .
3 3
with norm |(𝑞2 , 𝑝)|𝜎,𝑠 := |𝑞2 |𝜎,𝑠 + |𝑝|𝜎,𝑠 . We can claim that, 𝛾
under assumptions (51), there exist 𝜌1 , 𝜌2 ∈ (0, 1) such that
the map (𝑞2 , 𝑝) → P(𝑞2 , 𝑝; 𝑞1 ) is a contraction mapping in Now, set 𝐶0 (𝑅) := 𝜅5 𝑅3 , and we define 𝜌1 := 2𝐶0 (𝑅)/𝑁3/2 ,
𝐵, that is, we have to prove 𝜌2 := 2𝐶0 (𝑅)(|𝜖|/𝛾). By the inequality above, there exist
Abstract and Applied Analysis 7

𝑁0 (𝑅) ∈ N+ and 𝜖0 (𝑅) > 0 such that ∀𝑁 ≥ 𝑁0 (𝑅) and 𝑍 := {𝑞1 + 𝑞2 (𝑞1 ) + 𝑝(𝑞1 ) : |𝑞1 | < 2𝑅}, diffeomorphic to the
∀|𝜖|𝛾−1 ≤ 𝜖0 (𝑅), ball
󵄨󵄨 󵄨 󵄨󵄨 󵄨 󵄨 󵄨
󵄨󵄨P1 (𝑞2 , 𝑝; 𝑞1 )󵄨󵄨󵄨𝜎,𝑠 ≤ 𝜌1 , 󵄨󵄨P2 (𝑞2 , 𝑝; 𝑞1 )󵄨󵄨󵄨𝜎,𝑠 ≤ 𝜌2 . (59) 𝐵2𝑅 := {𝑞 ∈ 𝑄1 : 󵄨󵄨󵄨𝑞1 󵄨󵄨󵄨𝐻1 < 2𝑅} , (64)
the partial derivatives of the action functional Ψ𝜖 with respect
So, we get the proof of (i). Item (ii) can be obtained with the
to the variables (𝑞2 , 𝑝) vanish. We claim that, at a critical point
same estimates.
of Ψ𝜖 restricted to 𝑍, also the partial derivative of Ψ𝜖 with
By the Contraction Mapping Theorem, there exists
respect to the variable 𝑞1 vanishes, and therefore such a point
a unique fixed point (𝑞2 (𝑞1 ), 𝑝(𝑞1 )) := (𝑞2 (𝜖, 𝑁, 𝑞1 ),
is critical also for the nonrestricted functional Ψ𝜖 : H𝜎,𝑠 →
𝑝(𝜖, 𝑁, 𝑞1 )) of P in 𝐵. The bounds of (52) follow by the
R.
definition of 𝜌1 and 𝜌2 .
Actually, the bifurcation equation (63) is the Euler-
Since the map P ∈ 𝐶1 (𝑄2 ⊕ 𝑃 × 𝑄1 ; 𝑄2 ⊕ 𝑃 × 𝑄1 ), Lagrange equation of the reduced action functional
the Implicit Function Theorem implies that the map 𝑞1 →
(𝑞2 (𝜖, 𝑁, 𝑞1 ), 𝑝(𝜖, 𝑁, 𝑞1 )) is 𝐶1 . Differentiating both sides of Φ𝜖,𝑁 : 𝐵2𝑅 ⊂ 𝑄1 󳨀→ R,
(𝑞2 (𝑞1 ), 𝑝(𝑞1 )) = P(𝑞2 (𝑞1 ), 𝑝(𝑞1 ), 𝑞1 ), we can get (65)
Φ𝜖,𝑁 (𝑞1 ) := Ψ𝜖 (𝑞1 + 𝑞2 (𝑞1 + 𝑝 (𝑞1 ))) .
𝑞2󸀠 (𝑞1 ) [ℎ]
Lemma 8. Φ𝜖,𝑁 ∈ 𝐶1 (𝐵2𝑅 , R) and a critical point 𝑞1 ∈ 𝐵2𝑅 of
= −L−1 Φ𝜖,𝑁 is a solution of the bifurcation (63). Moreover, Φ𝜖,𝑁 can
1 Π𝑄2 (𝜕V 𝑓) (𝜑1 , 𝜑2 , 𝑞1 + 𝑞2 (𝑞1 ) + 𝑝 (𝑞1 ) , 𝜖)
be written as
× (ℎ + 𝑞2󸀠 (𝑞1 ) [ℎ] + 𝑝󸀠 (𝑞1 ) [ℎ]) , Φ𝜖,𝑁 (𝑞1 ) = 𝑐𝑜𝑛𝑠𝑡 + 𝜖 (Γ (𝑞1 ) + R𝜖,𝑁 (𝑞1 )) , (66)
(60)
𝑝󸀠 (𝑞1 ) [ℎ] where

= −𝜖L−1 1 (2𝑎 + 𝑎2 𝜖2 )
𝜖 Π𝑃 (𝜕V 𝑓) (𝜑1 , 𝜑2 , 𝑞1 + 𝑞2 (𝑞1 ) + 𝑝 (𝑞1 ) , 𝜖)
2 2
Γ (𝑞1 ) := ∫ (𝜕𝜑1 𝑞1 ) + (𝜕𝜑2 𝑞1 )
T2 2 2
× (ℎ + 𝑞2󸀠 (𝑞1 ) [ℎ] + 𝑝󸀠 (𝑞1 ) [ℎ]) .
𝑞14
+ 𝑎 (1 + 𝜖) (𝜕𝜑1 𝑞1 ) (𝜕𝜑2 𝑞1 ) − 𝑎3 (𝜑1 , 𝜑2 ) ,
Using (43)–(48) and the Banach property of H𝜎,𝑠 , we get 4
󵄨󵄨 󸀠 󵄨
󵄨󵄨𝑞2 (𝑞1 ) [ℎ]󵄨󵄨󵄨 R𝜖,𝑁 (𝑞1 ) := ∫ 𝐹 (𝜑1 , 𝜑2 , 𝑞1 , 𝜖 = 0)
󵄨 󵄨𝜎,𝑠 T2

𝐶 (𝑅) 󵄨 󵄨 󵄨 󵄨 − 𝐹 (𝜑1 , 𝜑2 , 𝑞1 + 𝑞2 + 𝑝, 𝜖)
≤ (|ℎ|𝜎,𝑠 + 󵄨󵄨󵄨󵄨𝑞2󸀠 (𝑞1 ) [ℎ]󵄨󵄨󵄨󵄨𝜎,𝑠 + 󵄨󵄨󵄨󵄨𝑝󸀠 (𝑞1 ) [ℎ]󵄨󵄨󵄨󵄨𝜎,𝑠 ) ,
𝑁2
󵄨󵄨 󸀠 󵄨 1
󵄨󵄨𝑝 (𝑞1 ) [ℎ]󵄨󵄨󵄨 + 𝑓 (𝜑1 , 𝜑2 , V, 𝜖) (𝑞2 + 𝑝) ,
󵄨 󵄨𝜎,𝑠 2
(67)
𝐶 (𝑅) |𝜖| 󵄨 󵄨 󵄨 󵄨
≤ (|ℎ|𝜎,𝑠 + 󵄨󵄨󵄨󵄨𝑞2󸀠 (𝑞1 ) [ℎ]󵄨󵄨󵄨󵄨𝜎,𝑠 + 󵄨󵄨󵄨󵄨𝑝󸀠 (𝑞1 ) [ℎ]󵄨󵄨󵄨󵄨𝜎,𝑠 ) , and for some positive constant 𝐶2 (𝑅) ≥ 𝐶1 (𝑅), we can get
𝛾
(61) 󵄨󵄨 󵄨 |𝜖| 1
󵄨󵄨R𝜖,𝑁 (𝑞1 )󵄨󵄨󵄨 ≤ 𝐶2 (𝑅) (𝜖 + + 2),
𝛾 𝑁
which implies the bounds (53), since when 𝐶(𝑅)(|𝜖|/𝛾+1/𝑁2 )
is small enough, we can get 󵄨󵄨 󸀠 󵄨 |𝜖| 1
󵄨󵄨R𝜖,𝑁 (𝑞1 ) [ℎ]󵄨󵄨󵄨 ≤ 𝐶2 (𝑅) (𝜖 + + 2 ) |ℎ|𝐻1 ,
(68)
󵄨 󵄨 𝛾 𝑁
󵄨󵄨 𝐶 (𝑅) 󵄨󵄨󵄨
󵄨󵄨 1 − 𝐶 (𝑅) − 2 󵄨󵄨
󵄨󵄨 𝑁 2 𝑁 󵄨󵄨 1 ∀ℎ ∈ 𝑄1 .
󵄨󵄨 󵄨󵄨 ≥ .
det 󵄨󵄨 󵄨
𝐶 (𝑅) |𝜖| 󵄨󵄨󵄨󵄨 2
(62)
󵄨󵄨 𝐶 (𝑅) |𝜖| Proof. By (37) and (38), we have that, at V := 𝑞1 + 𝑞2 (𝑞1 ) +
󵄨󵄨− 1− 󵄨󵄨
󵄨󵄨 𝛾 𝛾 󵄨󵄨 𝑝(𝑞1 ),
𝑑Ψ𝜖 (V) [ℎ] = 0, ∀ℎ ∈ 𝑄2 ,
(69)
𝑑Ψ𝜖 (V) [ℎ] = 0, ∀ℎ ∈ 𝑃.
3.3. The (𝑄1 )-Equation. Once the (𝑄2 )-(𝑃)-equations have
been solved by (𝑞2 (𝑞1 ), 𝑝(𝑞1 )) ∈ 𝑄2 ⊕ 𝑃, there remains the Since 𝑞2󸀠 (𝑞1 )[ℎ] ∈ 𝑄2 and 𝑝󸀠 (𝑞1 )[ℎ] ∈ 𝑃, ∀ℎ ∈ 𝑄1 , we deduce
finite-dimensional 𝑄1 -equation that
L1 [𝑞1 ] + Π𝑄1 𝑓 (𝜑1 , 𝜑2 , 𝑞1 + 𝑞2 (𝑞1 ) + 𝑝 (𝑞1 ) , 𝜖) = 0. (63) 𝑑Φ𝜖,𝑁 (𝑞1 ) [ℎ] = 𝑑Ψ𝜖 (V) [[ℎ] + 𝑞2󸀠 (𝑞1 ) [ℎ] + 𝑝󸀠 (𝑞1 ) [ℎ]]

The geometric interpretation of the construction of = 𝑑Ψ𝜖 (V) [ℎ] , ∀ℎ ∈ 𝑄1 ,

(𝑞2 (𝑞1 ), 𝑝(𝑞1 )) is that, on the finite-dimensional submanifold (70)
8 Abstract and Applied Analysis

and therefore V := 𝑞1 + 𝑞2 (𝑞1 ) + 𝑝(𝑞1 ) solves also the (𝑄1 )- + 𝑎 (1 + 𝜖) (𝜕𝜑1 𝑞1 ) (𝜕𝜑2 𝑞1 )
equation (36). Write Ψ𝜖 (V) := Ψ𝜖(2) (V) − 𝜖 ∫T 2 𝐹(𝜑1 , V, 𝜖), where
1
+ 𝑓 (𝜑1 , 𝜑2 , V, 𝜖) (𝑞2 (𝑞1 ) + 𝑝 (𝑞1 ))
2
1 2 2𝑎 + 𝑎2 𝜖2 2
Ψ𝜖(2) (V) = ∫ (1 + 𝜖) (𝜕𝜑1 V) + 𝜖 (𝜕𝜑2 V)
T 2 2 2 (71) − 𝐹 (𝜑1 , 𝜑2 , V, 𝜖) ) .
+ (1 + 𝜖𝑎 (1 + 𝜖)) (𝜕𝜑1 V) (𝜕𝜑2 V)
(74)

is a homogeneous functional of degree two. By homogeneity, Because Ψ0 (𝑞1 ) ≡ const, by (52), we can get the bounds of
(68).
1 The problem of finding nontrivial solutions of the 𝑄1 -
Ψ𝜖 (V) = 𝑑Ψ𝜖(2) (V) [V] − 𝜖 ∫ 𝐹 (𝜑1 , 𝜑2 , V, 𝜖) , (72)
2 T2 equation is reduced to finding nontrivial critical points of
the reduced action functional Φ𝜖,𝑁 in 𝐵2𝑅 . By (66), this is
and, according to (69), we have equivalent to finding critical points of the rescaled functional

̂ 𝜖,𝑁 = Γ (𝑞1 ) + R𝜖,𝑁 (𝑞1 )

Φ
𝑑Ψ𝜖(2) (𝑞1 + 𝑞2 (𝑞1 ) + 𝑝 (𝑞1 )) [𝑞2 (𝑞1 ) + 𝑝 (𝑞1 )] (75)
𝑞14
(73) = (Q (𝑞1 ) − ∫ 𝑎3 (𝜑1 , 𝜑2 ) ) + R𝜖,𝑁 (𝑞1 ) ,
= 𝜖 ∫ 𝑓 (𝜑1 , 𝜑2 , V, 𝜖) (𝑞2 (𝑞1 ) + 𝑝 (𝑞1 )) . T2 4
T2
where the quadratic form
Substituting the above equality into (72), we obtain, at V :=
𝑞1 + 𝑞2 (𝑞1 ) + 𝑝(𝑞1 ), 1 2 (2𝑎 + 𝑎2 𝜖2 ) 2
Q (𝑞) := ∫ (𝜕𝜑1 𝑞1 ) + (𝜕𝜑2 𝑞1 )
T 2 2 2 (76)
Φ𝜖,𝑁 (𝑞1 ) = Ψ𝜖 (𝑞1 + 𝑞2 (𝑞1 ) + 𝑝 (𝑞1 )) + 𝑎 (1 + 𝜖) (𝜕𝜑1 𝑞1 ) (𝜕𝜑2 𝑞1 )
1 (2)
= 𝑑Ψ (V) [𝑞1 + 𝑞2 (𝑞1 ) + 𝑝 (𝑞1 )] is positive definite on 𝑄+ , negative definite on 𝑄− , and zero
2 𝜖
definite on 𝑄0 . For 𝑞1 = 𝑞+ + 𝑞− + 𝑞0 ∈ 𝑄1 and 𝑎 < 0, we have
− 𝜖 ∫ 𝐹 (𝜑1 , 𝜑2 , V, 𝜖)
T2 1 2 (2𝑎 + 𝑎2 𝜖2 ) 2
Q (𝑞+ ) = ∫ (𝜕𝜑1 𝑞+ ) + (𝜕𝜑2 𝑞+ )
1 (2) T2 2 2
= 𝑑Ψ (𝑞1 ) [𝑞1 ]
2 𝜖
+ 𝑎 (1 + 𝜖) (𝜕𝜑1 𝑞+ ) (𝜕𝜑2 𝑞+ )
1
+ 𝜖 ∫ 𝑓 (𝜑1 , 𝜑2 , V, 𝜖) (𝑞2 (𝑞1 ) + 𝑝 (𝑞1 ))
2 T2 𝑎2 𝜖2
= ∫ ∑ 𝑙22 (−𝑎 − ) 𝑞̂𝑙1 ,𝑙2 𝑒𝑖𝑙1 𝜑1 𝑒𝑖𝑙2 𝜑2
T 2 (𝑙 ,𝑙 )∈Z2 2
1 2
− 𝜖 ∫ 𝐹 (𝜑1 , 𝜑2 , V, 𝜖)
T2
𝛼+ (𝑎) 󵄨󵄨 󵄨󵄨2 (77)
= 󵄨𝑞 󵄨 1 ,
2
= ∫ (𝜕𝜑1 𝑞1 ) + (𝜕𝜑1 𝑞1 ) (𝜕𝜑2 𝑞1 ) 2 󵄨 + 󵄨𝐻
T2
Q (𝑞0 ) = 0,
1 2 (2𝑎 + 𝑎2 𝜖2 ) 2
+ 𝜖 (∫ (𝜕 𝑞 ) + (𝜕𝜑2 𝑞1 ) (1 + 𝑎𝜖)2
T2 2 𝜑1 1 2 Q (𝑞− ) = ∫ ∑ 𝑙22 𝑞̂𝑙1 ,𝑙2 𝑒𝑖𝑙1 𝜑1 𝑒𝑖𝑙2 𝜑2
T2 2
(𝑙1 ,𝑙2 )∈Z2
+ 𝑎 (1 + 𝜖) (𝜕𝜑1 𝑞1 ) (𝜕𝜑2 𝑞1 )
𝛼− (𝑎) 󵄨󵄨 󵄨󵄨2
1 = − 󵄨𝑞 󵄨 1 ,
+ 𝑓 (𝜑1 , 𝜑2 , V, 𝜖) (𝑞2 (𝑞1 ) + 𝑝 (𝑞1 )) 2 󵄨 − 󵄨𝐻
2
where the positive constants 𝛼+ (𝑎), 𝛼− (𝑎) are bounded away
−𝐹 (𝜑1 , 𝜑2 , V, 𝜖) ) from zero and independent of 𝜖. We will prove the existence
̂ 𝜖,𝑁 in 𝐵2𝑅 of linking type.
of critical point of Φ
= Ψ0 (𝑞1 )
3.4. Linking Critical Points of the Reduced Action Functional
1 2 (2𝑎 + 𝑎2 𝜖2 ) 2 ̂ 𝜖,𝑁. We cannot directly apply the linking theorem because
Φ
+ 𝜖 (∫ (𝜕𝜑1 𝑞1 ) + (𝜕𝜑2 𝑞1 )
T 2 2 2 ̂
Φ𝜖,𝑁 is defined only in the ball 𝐵2𝑅 . Therefore, we first extend
Abstract and Applied Analysis 9

̂ 𝜖,𝑁 to the whole space 𝑄1 . We define the extended action

Φ Now, we fix 𝜌 > 0 small such that (𝛼+ (𝑎)/2)𝜌2 − 𝜅1 𝜌4 ≥
function Φ̃ 𝜖,𝑁 ∈ 𝐶1 (𝑄1 , R) as (𝛼+ (𝑎)/4)𝜌2 . Since 𝐶3 (𝑅)(𝜖 + 𝜖/𝛾 + 1/𝑁2 ) ≤ (𝛼+ (𝑎)/8)𝜌2 , by
(84), we can get
Φ ̃ (𝑞 ) ,
̃ 𝜖,𝑁 (𝑞1 ) = Γ (𝑞1 ) + R (78)
̃ 𝜖,𝑁 (𝑞+ ) ≥ 𝛼+ (𝑎) 𝜌2 := 𝛽 > 0,
𝜖,𝑁 1
Φ ∀𝑞+ ∈ 𝑄1 ∩ 𝑄+ ,
̃ (𝑞 ) : 𝑄 → R is 8 (85)
where R𝜖,𝑁 1 1
󵄨 󵄨
with 󵄨󵄨󵄨𝑞+ 󵄨󵄨󵄨𝐻1 = 𝜌.
󵄨󵄨 󵄨󵄨2
̃ (𝑞 ) := 𝜆 ( 󵄨󵄨𝑞1 󵄨󵄨𝐻1 ) R (𝑞 ) ,
R (79) (ii) Let
𝜖,𝑁 1 𝜖,𝑁 1
𝑅2
󵄨 󵄨
𝐵1 := {𝑞1 = 𝑞0 + 𝑞− + 𝑟2 𝑒+ with 󵄨󵄨󵄨𝑞0 + 𝑞− 󵄨󵄨󵄨𝐻1 ≤ 𝑟1 ,
and 𝜆 : [0, +∞) → [0, 1] is a smooth, nonincreasing, cut-off (86)
function such that 𝑞− ∈ 𝑄1 ∩ 𝑄− } ⊂ 𝜕𝑊− ,
󵄨 󵄨
𝜆 (𝑥) = 1, if |𝑥| ≤ 1; 𝐵2 := {𝑞1 = 𝑞0 + 𝑞− + 𝑟𝑒+ with 󵄨󵄨󵄨𝑞0 + 𝑞− 󵄨󵄨󵄨𝐻1 = 𝑟1 ,
(80) (87)
󵄨󵄨 󸀠 󵄨󵄨 𝑞− ∈ 𝑄1 ∩ 𝑄− , 𝑟 ∈ [0, 𝑟2 ]} ⊂ 𝜕𝑊− ,
𝜆 (𝑥) = 0, if |𝑥| ≥ 4, 󵄨󵄨𝜆 (𝑥)󵄨󵄨 < 1.
󵄨 󵄨
with 𝑟1 , 𝑟2 > 2𝑅. For 𝑞1 = 𝑞0 + 𝑞− + 𝑟𝑒+ ∈ 𝐵1 ∪ 𝐵2 ,
̃ 𝜖,𝑁(𝑞1 ) ≡ Φ
By definition, Φ ̂ 𝜖,𝑁(𝑞1 ) on 𝐵𝑅 := {𝑞1 ∈ 𝑄1 :
|𝑞1 |𝐻1 ≤ 𝑅} and Φ ̃ 𝜖,𝑁(𝑞1 ) = Γ(𝑞1 ) outside 𝐵2𝑅 . Moreover, ̃ 𝜖,𝑁 (𝑞1 ) = 𝛾 (𝑞1 )
Φ
by (68), there is a constant 𝐶3 (𝑅) ≥ 𝐶2 (𝑅) > 0 such that
4
∀|𝑞|𝐻1 ≤ 2𝑅, and (𝑞0 + 𝑞− + 𝑟𝑒+ )
= Q (𝑞1 ) − ∫ 𝑎3 (𝜑1 , 𝜑2 )
T2 4
󵄨󵄨̃ 󵄨 1
󵄨󵄨R𝜖,𝑁 (𝑞1 )󵄨󵄨󵄨 ≤ 𝐶3 (𝑅) (𝜖 + |𝜖| 𝛾−1 + 2 ) , (81) 𝛼− (𝑎) 󵄨󵄨 󵄨󵄨2
󵄨 󵄨 𝑁 2
≤− 󵄨𝑞 󵄨 1 + 𝑟 Q (𝑒+ )
󵄨󵄨̃󸀠 󵄨 1 2 󵄨 − 󵄨𝐻
󵄨󵄨R𝜖,𝑁 (𝑞1 ) [ℎ]󵄨󵄨󵄨 ≤ 𝐶3 (𝑅) (𝜖 + |𝜖| 𝛾−1 + 2 ) |ℎ|𝐻1 , (88)
󵄨 󵄨 𝑁 (𝑞 + 𝑞− + 𝑟𝑒+ )
4
(82) − ∫ 𝑎3 (𝜑1 , 𝜑2 ) 0
∀ℎ ∈ 𝑄1 . T 2 4
𝛼− (𝑎) 󵄨󵄨 󵄨󵄨2 2
̃ 𝜖,𝑁(𝑞1 ) satisfies the geometrical
Second, we will verify that Φ ≤− 󵄨𝑞 󵄨 1 + 𝑟 Q (𝑒+ )
2 󵄨 − 󵄨𝐻
hypotheses of the linking theorem.
4
− 𝛼 ∫ (𝑞0 + 𝑞− + 𝑟𝑒+ ) ,
Lemma 9. There exist positive constants 𝜌, 𝛽, 𝑟1 , 𝑟2 > 𝜌, and T2
0 < 𝜖1 (𝑅) ≤ 𝜖0 (𝑅), 𝑁1 (𝑅) ≥ 𝑁0 (𝑅), which are independent of
(𝜖, 𝑁, 𝛾), such that, ∀(|𝜖|/𝛾) ≤ 𝜖1 and ∀𝑁 ≥ 𝑁1 (𝑅), because 𝑎3 (𝜑1 , 𝜑2 )/4 ≥ 𝛼 > 0. Now, by Hölder inequality and
orthogonality,
̃ 𝜖,𝑁(𝑞1 ) ≥ 𝛽 > 0, ∀𝑞1 ∈ 𝑆+ := {𝑞1 ∈ 𝑄1 ∩𝑄+ :
(i) inf 𝑞1 ∈𝑆+ Φ 2
4 2
|𝑞1 |𝐻1 = 𝜌}, ∫ (𝑞0 + 𝑞− + 𝑟𝑒+ ) ≥ 𝜅2 (∫ (𝑞0 + 𝑞− + 𝑟𝑒+ ) )
T2 T2
̃ 𝜖,𝑁(𝑞1 ) ≤ 𝛽/2, ∀𝑞1 ∈ 𝜕𝑊− ,
(ii) ∑𝑞1 ∈𝜕𝑊− Φ
2 (89)
= 𝜅2 (𝑞02 + 𝑞−2 + 𝑟2 𝑒+2 )
−
where 𝑊 is the rectangle in 𝑄− ⊕ 𝑄0 ,
2
󵄨 󵄨 ≥ 𝜅3 (𝑞02 + 𝑟2 ) ≥ 𝜅3 (𝑞04 + 𝑟4 ) ,
𝑊 := {𝑞1 = 𝑞0 + 𝑞+ + 𝑟𝑒 , 󵄨󵄨󵄨𝑞0 + 𝑞− 󵄨󵄨󵄨 ≤ 𝑟1 , 𝑞− ∈ 𝑄1 ∩ 𝑄− ,
− +

and by (88) we deduce that

𝑞0 ∈ R, 𝑟 ∈ [0, 𝑟2 ]} ,
(83) ̃ 𝜖,𝑁 (𝑞0 + 𝑞− + 𝑟𝑒+ ) ≤ (𝜅4 𝑟2 − 𝜅3 𝑟4 )
Φ

𝛼− (𝑎) 󵄨󵄨 󵄨󵄨2 (90)

and 𝑒+ is the unit vector in 𝑄1 ⊕ 𝑄0 . −( 4
󵄨𝑞 󵄨 1 + 𝜅3 𝑞0 ) .
2 󵄨 − 󵄨𝐻
Proof. (i) ∀𝑞+ ∈ 𝑄1 ∩ 𝑄+ with |𝑞+ |𝐻1 = 𝜌 < 𝑅, we have
Now, we fix 𝑟2 large such that 𝜅4 𝑟22 − 𝜅3 𝑟24 ≤ 0, and therefore
̃ 𝜖,𝑁 (𝑞+ ) = Φ
Φ ̂ 𝜖,𝑁 (𝑞+ )
̃ 𝜖,𝑁 (𝑞1 ) ≤ 𝜅4 𝑟2 − 𝜅3 𝑟4 ≤ 0,
Φ ∀𝑞1 ∈ 𝐵1 . (91)
2 2
𝑞+4
= Q (𝑞+ ) − ∫ 𝑎3 (𝜑1 , 𝜑2 ) + R𝜖,𝑁 (𝑞+ ) (84) Next, setting 𝑀 := max[0,𝑟2 ] 𝜅4 𝑟2 − 𝜅3 𝑟4 , we fix 𝑟1 large such
T2 4 that
𝛼+ (𝑎) 2 𝜖 1 𝛼− (𝑎) 󵄨󵄨 󵄨󵄨2 󵄨 󵄨
≥ 𝜌 − 𝜅1 𝜌4 − 𝐶3 (𝑅) (𝜖 + + 2 ) . 4
󵄨𝑞 󵄨 1 + 𝜅3 𝑞0 ≥ 𝑀, ∀ 󵄨󵄨󵄨𝑞− + 𝑞0 󵄨󵄨󵄨 = 𝑟1 , (92)
2 𝛾 𝑁 2 󵄨 − 󵄨𝐻
10 Abstract and Applied Analysis

and therefore Proof. In the sequel, we will always assume that (𝜖 + |𝜖|𝛾−1 +
1/𝑁2 ) < 1. Writing Φ ̃ (𝑞), by (81)-(82),
̃ 𝜖,𝑁(𝑞) = Γ(𝑞) + R
̃ 𝜖,𝑁 (𝑞1 ) ≤ 𝑀 − ( 𝛼− (𝑎) 󵄨󵄨󵄨𝑞− 󵄨󵄨󵄨2 1 + 𝜅3 𝑞4 ) ≤ 0,
Φ ∀𝑞1 ∈ 𝐵2 .
𝜖,𝑁
2 󵄨 󵄨𝐻 0 we can get
(93)
̃ 𝜖,𝑁 (𝑞𝑛 ) − 1 Φ
Φ ̃ 󸀠 (𝑞 ) [𝑞 ]
Finally, if 𝑞1 = 𝑞− + 𝑞0 , 2 𝜖,𝑁 𝑛 𝑛
4 1
̃ 𝜖,𝑁 (𝑞1 ) = Q (𝑞− ) − ∫ 𝑎3 (𝜑1 , 𝜑2 ) 𝑞1 + R
Φ ̃ (𝑞 ) = Γ (𝑞𝑛 ) − Γ󸀠 (𝑞𝑛 ) [𝑞𝑛 ]
4 𝜖,𝑁 1 2
T2
(94)
󵄨󵄨̃ 󵄨󵄨 −1 1 ̃ (𝑞 ) − 1 R
+ (R ̃󸀠 (𝑞 ) [𝑞 ])
≤ 󵄨󵄨󵄨R𝜖,𝑁 (𝑞1 )󵄨󵄨󵄨 ≤ 𝐶3 (𝑅) (𝜖 + |𝜖| 𝛾 + 2 ) . 𝜖,𝑁 𝑛
2 𝜖,𝑁 𝑛 𝑛
𝑁 (100)
̃ 𝜖,𝑁(𝑞1 ) ≤ 1 1
−1 2
So, if 𝐶3 (𝑅)(√𝜖+|𝜖|𝛾 +(1/𝑁 )) ≤ 𝛽/2, we can get Φ = ( − ) ∫ 𝑎3 (𝜑1 ) 𝑞𝑗4
2 4 T2
𝛽/2, ∀𝑞1 ∈ 𝜕𝑊− .
− ̃ (𝑞 ) − 1 R
+ (R ̃󸀠 (𝑞 ) [𝑞 ])
We introduce the minimax class S := {𝜓 ∈ 𝐶(𝑊 , 𝑄) | 𝜖,𝑁 𝑛
2 𝜖,𝑁 𝑛 𝑛
𝜓 = Id on 𝜕𝑊− }. According to Proposition 5.9 of [19], the
maps of S have an important intersection property as follows. 1 1 1
≥ 𝛼 ( − ) ∫ 𝑞𝑗4 − (𝜖 + |𝜖| 𝛾−1 + 2 ) .
2 4 T 2 𝑁
Proposition 10. (𝑆+ and 𝑊− link with respect to 𝜓). Consider
̃ 󸀠 (𝑞𝑛 ) → 0 (𝑛 → ∞),
Then, by 𝑐̃𝜖 < 𝑐̃∞ < +∞ and Φ 𝜖,𝑁
𝜓 ∈ S 󳨐⇒ 𝜓 (𝑊− ) ∩ 𝑆+ ≠
0. (95)
󵄨 󵄨 󵄨 󵄨4
One defines the minimax linking level as follows: 𝑐̃∞ + 1 + 󵄨󵄨󵄨𝑞𝑛 󵄨󵄨󵄨𝐻1 ≥ 𝜅6 ∫ 𝑞𝑛4 := 𝜅6 󵄨󵄨󵄨𝑞𝑛 󵄨󵄨󵄨𝐿4 . (101)
T2
̃ 𝜖,𝑁 (𝜓 (𝑞1 )) .
𝑐̃𝜖 := inf max− Φ
𝜓∈S𝑞1 ∈𝑊 (96) By Hölder inequality and orthogonality,
Obviously, by Proposition 10 and Lemma 9, 2
󵄨 󵄨 2
𝑐̃∞ + 1 + 󵄨󵄨󵄨𝑞𝑛 󵄨󵄨󵄨𝐻1 ≥ 𝜅7 (∫ (𝑞0,𝑛 + 𝑞−,𝑛 + 𝑞+,𝑛 ) )
̃ 𝜖,𝑁 (𝜓 (𝑞1 )) ≥ min Φ
max Φ ̃ 𝜖,𝑁 (𝑞1 ) ≥ 𝛽 > 0, ∀𝜓 ∈ S, T2
+
𝑞1 ∈𝑊− 𝑞1 ∈𝑆
2
(97) 2 2 2 4
= 𝜅7 (∫ 𝑞0,𝑛 + 𝑞−,𝑛 + 𝑞+,𝑛 ) ≥ 𝜅8 (𝑞0,𝑛 ) ,
T2
and, therefore, 𝑐̃𝜖 > 𝛽 > 0. (102)
Since Id ∈ S, so
𝑐̃𝜖 ≤ max− Φ ̃ (𝑞 ))
̃ 𝜖,𝑁 ((𝑞1 )) ≤ max (Γ (𝑞1 ) + R𝜖,𝑁 1
and therefore |𝑞0,𝑛 | ≤ (1 + |𝑞𝑛 |𝐻1 )1/4 . In the same way, by
𝑞∈𝑊 −
𝑞∈𝑊 Hölder inequality and (82),
𝛼+ (𝑎) 󵄨󵄨 󵄨󵄨2 𝛼 (𝑎) 󵄨󵄨 󵄨󵄨2 4 ̃ 󸀠 (𝑞𝑛 ) [𝑞+,𝑛 ] = 𝛼+ (𝑎) 󵄨󵄨󵄨𝑞+,𝑛 󵄨󵄨󵄨2
≤ max− ( 󵄨𝑞 󵄨 1 + − 󵄨𝑞 󵄨 1 + ∫ 𝜅𝑞1 ) (98) Φ
𝑞∈𝑊 2 󵄨 + 󵄨𝐻 2 󵄨 − 󵄨𝐻 T2
𝜖,𝑁 󵄨 󵄨𝐻 1

+ 1 ≤ 𝑐̃∞ ≤ +∞, ̃󸀠 (𝑞 ) [𝑞 ]
− ∫ 𝑎3 (𝜑1 ) 𝑞𝑛3 𝑞+,𝑛 + R𝜖,𝑁 𝑛 +,𝑛
T2
where 𝑐̃∞ is independent of 𝑁, 𝜖, 𝛾. By the linking theorem,
󵄨 󵄨2 󵄨 󵄨 󵄨 󵄨3
we deduce the existence of a Palais-Smale sequence (𝑞𝑛 ) ∈ 𝑄1 ≥ 𝛼+ (𝑎) 󵄨󵄨󵄨𝑞+,𝑛 󵄨󵄨󵄨𝐻 − 𝜅9 󵄨󵄨󵄨𝑞+,𝑛 󵄨󵄨󵄨𝐻 ∫ 󵄨󵄨󵄨𝑞𝑛 󵄨󵄨󵄨
1 1 2
at the level 𝑐̃𝜖 , namely, T

1 󵄨 󵄨
̃ 𝜖,𝑁 (𝑞𝑛 ) 󳨀→ 𝑐̃𝜖 ,
Φ ̃ 󸀠 (𝑞𝑛 ) 󳨀→ 0 when 𝑛 󳨀→ ∞. (99)
Φ − 𝐶3 (𝑅) (√𝜖 + |𝜖| 𝛾−1 + ) 󵄨󵄨𝑞 󵄨󵄨
𝜖,𝑁
𝑁3/2 󵄨 +,𝑛 󵄨𝐻1
Third, we will prove that the Palais-Smale sequence (𝑞𝑛 ) (up 󵄨 󵄨 󵄨 󵄨 󵄨 󵄨3
to subsequence) converges to some nontrivial critical point ≥ 𝜅10 󵄨󵄨󵄨𝑞+,𝑛 󵄨󵄨󵄨𝐻 (󵄨󵄨󵄨𝑞+,𝑛 󵄨󵄨󵄨𝐻 − 󵄨󵄨󵄨𝑞𝑛 󵄨󵄨󵄨𝐿4 − 1) .
1 1
𝑞̂1 ≠ ̃ 𝜖,𝑁 and Φ
0 in some open ball of 𝑄1 , where Φ ̂ 𝜖,𝑁 coincide. (103)
Because the space 𝑄1 is finite dimensional, we only need to
prove that the sequence (𝑞𝑛 ) is bounded. ̃ 󸀠 (𝑞𝑛 ) →
By (101) and the above inequalities, using Φ 𝜖,𝑁
0 (𝑛 → ∞), we conclude that |𝑞+,𝑛 |𝐻 ≤ 𝜅11 (1 + |𝑞𝑛 |3/4
𝐻1 ).
Lemma 11. There is a constant 𝑀 > 0 independent of 1

𝑅, 𝜖, 𝑁, 𝛾, such that, for all |𝜖|/𝛾 small enough and 𝑁 large Estimating analogously, we derive |𝑞−,𝑛 |𝐻 ≤ 𝜅12 (1 + |𝑞𝑛 |3/4
𝐻1 ).
1
enough, the Palais-Smale sequence (𝑞𝑛 ) is bounded; that is, Finally, we deduce that
|𝑞𝑛 |𝐻1 < 𝑀. So, there exists a subsequence of the P-S sequence
󵄨󵄨 󵄨󵄨 󵄨 󵄨 󵄨 󵄨 󵄨 󵄨
that converges to some critical point 𝑞̂1 ≠ 0, and the functional 󵄨󵄨𝑞𝑛 󵄨󵄨𝐻1 = 󵄨󵄨󵄨𝑞0,𝑛 󵄨󵄨󵄨 + 󵄨󵄨󵄨𝑞+,𝑛 󵄨󵄨󵄨𝐻1 + 󵄨󵄨󵄨𝑞−,𝑛 󵄨󵄨󵄨𝐻1
̃ 𝜖,𝑁 possesses a nontrivial critical point 𝑞̂1 ∈ 𝑄1 with critical
Φ (104)
󵄨 󵄨3/4 󵄨 󵄨3/4
value Φ̃ 𝜖,𝑁(̂
𝑞1 ) = 𝑐̃𝜖 . ≤ 𝜅13 (1 + 󵄨󵄨󵄨𝑞𝑛 󵄨󵄨󵄨𝐻1 + 󵄨󵄨󵄨𝑞𝑛 󵄨󵄨󵄨𝐻1 ) .
Abstract and Applied Analysis 11

So, we can conclude that |𝑞𝑛 |𝐻1 ≤ 𝑀 for a suitable constant is V(𝜑2 ) ≡ 0. Indeed, by the homogeneity of 𝑓(𝜑1 , 𝜑2 , V, 𝜖), we
𝑀 > 0, which is independent of 𝑅, 𝜖, 𝑁, 𝛾. Since 𝑄1 is finite have
dimensional, {𝑞𝑛 } converges, up to subsequence, to some
∞
̃ 𝜖,𝑁 with |̂
critical point 𝑞̂ of Φ ̃ 𝜖,𝑁(̂
𝑞|𝐻1 < 𝑀. Since Φ 𝑞) = 𝑐̃𝜖 ≥ 𝜖3 𝑓 (𝜑1 , 𝜑2 , V, 𝜖) = 𝑓 (𝜑1 , 𝜑2 , 𝜖V) = ∑ 𝑎𝑘 (𝜑1 , 𝜑2 ) (𝜖V)𝑘 .
𝛽 > 0, we conclude that 𝑞̂ ≠ 0. 𝑘=3

−1
(109)
Proof of Theorem 1. Let us fix 𝑅 := 𝑀 + 1 and take |𝜖|𝛾 ≤
𝜖2 (𝑅) := 𝜖. According to Lemma 7, we can get, for 0 < Now consider a smooth function ℎ(𝜑1 ) with zero mean, ant
𝜎𝑁(𝑅) ≤ 1, 0 ≤ 𝑠 < 1/2, a solution (𝑞2 (𝑞1 ), 𝑝(𝑞1 )) ∈ it satisfies ∫T 𝑎𝑘 (𝜑1 , 𝜑2 )ℎ(𝜑1 )𝑑𝜑1 ≠
0 for some 𝑘. Multiplying
(𝑄2 (𝑁) ⊕ 𝑃) ∩ H𝜎,𝑠 of the (𝑄2 )-(𝑃)-equations with ∀|𝑞1 |𝐻1 ≤ (108) by ℎ(𝜑1 ) and integrating over [0, 𝜋], we have
2𝑅. By Lemma 11, the extended functional Φ ̃ 𝜖,𝑁(𝑞1 ) possesses
a nontrivial critical point 𝑞̂ with |𝑞1 |𝐻1 ≤ 𝑀 < 𝑅. Since Φ ̃ 𝜖,𝑁 2𝜋 𝑑2 V (𝜑2 )
∫ (2𝑎 + 𝑎2 𝜖2 ) ℎ (𝜑1 ) 𝑑𝜑1
coincides with Φ𝜖,𝑁 on the ball 𝐵𝑅, by Lemma 8, there exists 0 𝑑𝜑22
a nontrivial weak solution 𝑞̂1 + 𝑞2 (̂ 𝑞1 ) + 𝑝(̂𝑞1 ) ∈ H𝜎,𝑠 of (23). (110)
2𝜋
𝑞1 + 𝑞2 (̂
Finally, 𝑢 = 𝜖[̂ 𝑞1 ) + 𝑝(̂
𝑞1 )] = 𝜖[̂ 𝑞𝜖 + 𝑝(̂ 𝑞1 )] solves (7). + ∫ 𝑓 (𝜑1 , 𝜑2 , V (𝜑2 ) , 𝜖) ℎ (𝜑1 ) 𝑑𝜑1 = 0.
According to Lemma 6, by the regularizing property of 0
the operator L1 , the solution 𝑞̂𝜖 := 𝑞̂1 + 𝑞2 (̂ 𝑞1 ) of the (𝑄)-
equation belongs to H𝜎,𝑠+2 ∩ 𝑄. By the 𝑃-equation According to ℎ(𝜑1 ) that has zero mean and multiplying above
equation by 𝜖3 , we get
((2 + 𝜖) 𝜕𝜑21 + 2 (1 + 𝑎 (𝜖 + 𝜖2 )) 𝜕𝜑1 𝜕𝜑2 ) 𝑝̂
(105) 2𝜋

= −𝜖 ((2𝑎 + 𝑎 𝜖 2 2
) 𝜕𝜑22 ̂
+ Π𝑃 𝑓 (𝜑1 , 𝜑2 , 𝑞̂𝜖 + 𝑝, 𝜖)) 𝑝, ∫ 𝑓 (𝜑1 , 𝜑2 , 𝜖V (𝜑2 )) ℎ (𝜑1 ) 𝑑𝜑1
0
∞ 2𝜋
(111)
where 𝑝̂ = 𝑝(̂
𝑞1 ), we can get that 𝑘
= ∑ (𝜖V (𝜑2 )) ∫ 𝑎𝑘 (𝜑1 , 𝜑2 ) ℎ (𝜑1 ) 𝑑𝜑1 = 0.
𝑘=3 0
2 2
−𝜖 ((2𝑎 + 𝑎 𝜖 ) 𝜕𝜑22 + Π𝑃 𝑓 (𝜑1 , 𝜑2 , 𝑞̂𝜖 + 𝑝, 𝜖)) 𝑝̂ ∈ H𝜎󸀠,𝑠 ,
(106) The function 𝐺(𝑧(𝜑2 )) = ∑∞ 𝑘
𝑘=3 𝑏𝑘 (𝜑2 )(𝑧(𝜑2 )) , with 𝑏𝑘 =
2𝜋
󸀠 ∫0 𝑎𝑘 (𝜑1 , 𝜑2 )ℎ(𝜑1 )𝑑𝜑1 , is a nontrivial analytic function.
for 0 < 𝜎󸀠 < 𝜎, satisfying |𝜖(2𝑎 + 𝑎2 𝜖2 )𝑙22 |𝑒|𝑙2 |𝜎 < 𝑒|𝑙2 |𝜎 . For
Thus, the equation 𝐺(𝜖V(𝜑2 )) = 0 cannot have a sequence of
(𝑎, 𝜖) ∈ B𝛾 , the eigenvalues of operator (2 + 𝜖)𝜕𝜑21 + 2(1 + 𝑎(𝜖 + zeros accumulating to zero. So, for 𝜖 small enough, V(𝜑2 ) ≡
𝜖2 ))𝜕𝜑1 𝜕𝜑2 restricted to 𝑃 satisfy |(2+𝜖)𝑙12 +2(1+𝑎𝜖+𝑎𝜖2 )𝑙1 𝑙2 | ≥ 0.
|2 + 𝜖|(𝛾|𝑙1 |/|𝑙2 |), ∀𝑙1 ≠
0, 𝑙1 + 𝑙2 ≠
0, and, thus, we deduce that
󸀠󸀠
𝑝 ∈ H𝜎󸀠󸀠 ,𝑠+1 , for all 0 < 𝜎 < 𝜎󸀠 (satisfying 𝐶|𝑙2 |𝑒|𝑙2 |𝜎 <
̂ 󸀠󸀠
󸀠 4. Waves Traveling in Opposite Directions
𝑒|𝑙2 |𝜎 ), and |𝜕𝜑1 𝑝|
̂ 󸀠󸀠 = 𝑂(|𝜖|/𝛾). By (105),
𝜎 ,𝑠
Substituting 𝜔1 = 1 + 𝜖, 𝜔2 = 1 + 𝑎𝜖 into (8), we get
(2 + 𝜖) 𝜕𝜑21 𝑝̂
L𝑎,𝜖 V + 𝑓 (𝜑1 , 𝜑2 , V) = 0, (112)
2
= −2 (1 + 𝑎 (𝜖 + 𝜖 )) 𝜕𝜑1 𝜕𝜑2 𝑝̂
where (see (8))
− 𝜖 ((2𝑎 + 𝑎2 𝜖2 ) 𝜕𝜑22 + Π𝑃 𝑓 (𝜑1 , 𝜑2 , 𝑞̂𝜖 + 𝑝, 𝜖)) 𝑝,
̂
(107) L𝑎,𝜖 := [(1 + 𝜖 + 1) 𝜕𝜑1 + (𝑎𝜖) 𝜕𝜑2 ] ∘ [𝜖𝜕𝜑1 + (2 + 𝑎𝜖) 𝜕𝜑2 ]

̂ 𝜎,𝑠+2 = 𝑂(|𝜖|𝛾−1 ), with 0 < 𝜎 < 𝜎󸀠󸀠 .

we get 𝑝̂ ∈ H𝜎,𝑠+2 , and |𝑝| = 4𝜕𝜑1 𝜕𝜑2 + 𝜖 [(2 + 𝜖) 𝜕𝜑21 + (2𝑎 + 𝑎2 𝜖) 𝜕𝜑22
Thus, (15) follows with 𝑠 = 𝑠 + 2, 0 < 𝜎 < 𝜎󸀠󸀠 . By (5), 𝑢(𝑡, 𝑥) =
+ 2 (𝑎 + 1 + 𝑎𝜖) 𝜕𝜑1 𝜕𝜑2 ] = L0 + 𝜖L1 .
𝜖V((1+𝜖)𝑡+𝑥, (1+𝑎𝜖2 )𝑡+𝑥) is the solution of (1), for all (𝑎, 𝜖) ∈
B𝛾 . To show that 𝑢(𝑡, 𝑥) is quasiperiodic, it remains to prove (113)
that V depends on both variables (𝜑1 , 𝜑2 ) independently.
According to Lemma 9, ∀|̂ ̃ 𝜖,𝑁(̂
𝑞1 |𝐻1 ≤ 𝑅, Φ 𝑞1 ) ≥ 𝛽. On the We rescale (112) in order to highlight the relationship between
̃ 𝜖,𝑁(̂ the amplitude and the variation in frequency: V(𝜑1 , 𝜑2 ) →
other hand, Φ 𝑞− + 𝑞̂0 ) ≤ 𝛽/2, ∀|̂
𝑞− + 𝑞̂0 |𝐻1 ≤ 𝑅, so that
√|𝜖|V(𝜑1 , 𝜑2 ), and, for convenience, we assume 𝜖 > 0. In the
𝑞̂1 ∉ 𝑄− ⊕ 𝑄0 , and, therefore, V depends on 𝜑2 . In fact, any
following, we consider the scaled equation
solution V of (23) depending only on 𝜑2 , that is, the solutions
of
L𝑎,𝜖 V + 𝜖𝑓 (𝜑1 , 𝜑2 , V, 𝜖) = 0, (114)
𝑑2 V (𝜑2 )
(2𝑎 + 𝑎2 𝜖2 ) + 𝑓 (𝜑1 , 𝜑2 , V (𝜑2 ) , 𝜖) = 0, (108)
𝑑𝜑22 where 𝑓(𝜑1 , 𝜑2 , V, 𝜖) = 𝑎3 (𝜑1 , 𝜑2 )V3 + √𝜖𝑂(V4 ).
12 Abstract and Applied Analysis

Equation (114) is the Euler-Lagrange equation of the where

Lagrange action functional Ψ𝜖 ∈ 𝐶1 (H𝜎,𝑠 , R) defined by
𝑄+ := {𝑞 ∈ 𝑄 : 𝑞̂𝑙1 ,𝑙2 = 0, for (𝑙1 , 𝑙2 ) ∉ Λ + }

𝜖 (2 + 𝜖) 2 = {𝑞+ := 𝑞+ (𝜑) ∈ H0𝜎,𝑠 } ,

Ψ𝜖 (V) := ∫ 2 (𝜕𝜑1 V) (𝜕𝜑2 V) + (𝜕𝜑1 V)
T2 2
𝑄0 := {𝑞 : 𝑞0,0 ∈ R} ,
𝜖 (2𝑎 + 𝑎2 𝜖) 2
+ (𝜕𝜑2 V) 𝑄− := {𝑞 ∈ 𝑄 : 𝑞̂𝑙1 ,𝑙2 = 0, for (𝑙1 , 𝑙2 ) ∉ Λ − } (121)
2
+ 𝜖 (𝑎 + 1 + 𝑎𝜖) (𝜕𝜑1 V) (𝜕𝜑2 V) − 𝜖𝐹 (𝜑1 , V, 𝛿) = {𝑞− := 𝑞− (𝜑) ∈ H0𝜎,𝑠 } ,

= Ψ0 (V) + 𝜖Ψ1 (V, 𝛿) , Λ + := {(𝑙1 , 𝑙2 ) ∈ Z2 : 𝑙1 = 0, (𝑙1 , 𝑙2 ) ≠(0, 0)} ,

(115)
Λ − := {(𝑙1 , 𝑙2 ) ∈ Z2 : 𝑙2 = 0, (𝑙1 , 𝑙2 ) ≠(0, 0)} .
V
where 𝐹(𝜑1 , 𝜑2 , V, 𝜖) := ∫0 𝑓(𝜑1 , 𝜑2 , 𝜉, 𝜖)𝑑𝜉 and Ψ0 (V) := We decompose the space H𝜎,𝑠 = 𝑄 + 𝑃, where
∫T 2 2(𝜕𝜑1 V)(𝜕𝜑2 V),
{
𝑃 := {𝑝 = ∑ 𝑝̂𝑙1 ,𝑙2 𝑒𝑖𝑙1 𝜑1 𝑒𝑖𝑙2 𝜑2 ∈ H𝜎,𝑠 | 𝑝̂𝑙1 ,𝑙2 = 0,
2
(2 + 𝜖) 2 2𝑎 + 𝑎 𝜖 2
{ (𝑙1 ,𝑙2 )∈Z2
Ψ1 (V, 𝛿) := ∫ (𝜕𝜑1 V) + (𝜕𝜑2 V) (122)
T2 2 2
}
+ (𝑎 + 1 + 𝑎𝜖) (𝜕𝜑1 V) (𝜕𝜑2 V) − 𝐹 (𝜑1 , 𝜑2 , V, 𝜖) . for 𝑙1 𝑙2 = 0} .
(116) }
Projecting (114) onto the closed subspaces 𝑄 and 𝑃, setting
In order to find critical points of Ψ𝜖 (V), we use the same V = 𝑞 + 𝑝 ∈ H𝜎,𝑠 with 𝑞 ∈ 𝑄 and 𝑝 ∈ 𝑃, we obtain
method as in Section 3. The operator L𝑎,𝜖 is diagonal defined
on the Banach space H𝜎,𝑠 under the Fourier basis 𝑒𝑙1 ,𝑙2 = (𝑄) L1 [𝑞] + Π𝑄𝑓 (𝜑1 , 𝜑2 , 𝑞 + 𝑝, 𝜖) = 0,
𝑒𝑖𝑙1 𝜑1 𝑒𝑖𝑙2 𝜑2 with eigenvalue 𝐷𝑙1 ,𝑙2 = −[(2 + 𝜖)𝑙1 + 𝑎𝜖𝑙2 ][𝜖𝑙1 + (123)
(𝑃) L𝑎,𝜖 [𝑝] + 𝜖Π𝑃 𝑓 (𝜑1 , 𝜑2 , 𝑞 + 𝑝, 𝜖) = 0,
(2 + 𝑎𝜖)𝑙2 ]. So, we have
where Π𝑄 : H𝜎,𝑠 → 𝑄, Π𝑃 : H𝜎,𝑠 → 𝑃 are the projectors,
L𝑎,𝜖 [V] = ∑ 𝐷𝑙1 ,𝑙2 ̂V𝑙1 ,𝑙2 𝑒𝑖𝑙1 𝜑1 𝑖𝑙2 𝜑2
𝑒 , ∀V ∈ H𝜎,𝑠 . respectively, onto 𝑄 and 𝑃; moreover, they are continuous.
(𝑙1 ,𝑙2 )∈Z2
(117) In the same way, we decompose the space 𝑄 = 𝑄1 + 𝑄2 .
Setting 𝑞 = 𝑞1 + 𝑞2 with 𝑞1 ∈ 𝑄1 and 𝑞2 ∈ 𝑄2 , we finally get

The unperturbed functional Ψ0 : H𝜎,𝑠 → R possesses an (𝑄1 ) L1 [𝑞1 ] + Π𝑄1 [𝑓 (𝜑1 , 𝜑2 , 𝑞1 + 𝑞2 + 𝑝, 𝜖)] = 0
infinite-dimensional linear space 𝑄 of critical points, which (124)
are the solutions of the equation ⇐⇒ 𝑑Ψ𝜖 (V) [ℎ] = 0, ∀ℎ ∈ 𝑄1 ,

(𝑄2 ) L1 [𝑞2 ] + Π𝑄2 [𝑓 (𝜑1 , 𝜑2 , 𝑞1 + 𝑞2 + 𝑝, 𝛿)] = 0

L0 𝑞 = 4𝜕𝜑1 𝜕𝜑2 𝑞 = 0. (118) (125)
⇐⇒ 𝑑Ψ𝜖 (V) [ℎ] = 0, ∀ℎ ∈ 𝑄2 ,

The space 𝑄 can be written as (𝑃) L𝑎,𝜖 [𝑝] + 𝜖Π𝑃 [𝑓 (𝜑1 , 𝜑2 , 𝑞1 + 𝑞2 + 𝑝, 𝜖)] = 0
(126)
⇐⇒ 𝑑Ψ𝜖 (V) [ℎ] = 0, ∀ℎ ∈ 𝑃.
{
𝑄 = {𝑞 = ∑ 𝑞̂𝑙1 ,𝑙2 𝑒𝑖𝑙1 𝜑1 𝑒𝑖𝑙2 𝜑2 ∈ H𝜎,𝑠 | 𝑞̂𝑙1 ,𝑙2 = 0, The operator L𝑎,𝜖 is diagonal in the Fourier basis
{ (𝑙1 ,𝑙2 )∈Z2 {𝑒𝑖𝑙1 𝜑1 𝑒𝑖𝑙2 𝜑2 , (𝑙1 , 𝑙2 ) ∈ Z2 } with eigenvalues 𝐷𝑙1 ,𝑙2 = −[(2 + 𝜖)𝑙1 +
(119) 𝑎𝜖𝑙2 ][𝜖𝑙1 + (2 + 𝑎𝜖)𝑙2 ]. We first prove that L𝑎,𝜖 restricted to 𝑃
} has a bounded inverse when (𝑎, 𝜖) belongs to the uncountable
for 𝑙1 𝑙2 ≠
0} . zero-measure set
}
𝑎𝜖 𝜖
D𝛾 := {(𝑎, 𝜖) ∈ R− × R, ( , ) ∈ E𝛾 ,
2 + 𝜖 2 + 𝑎𝜖
We split 𝑄 as
1+𝜖
∉ Q, 1 + 𝜖 ≠
0, 1 + 𝑎𝜖 ≠
0, 2 + 𝑎𝜖 ≠
0} ,
1 + 𝑎𝜖
𝑄 = 𝑄+ + 𝑄0 + 𝑄− , (120) (127)
Abstract and Applied Analysis 13

where E𝛾 is a set of badly approximate numbers defined as Lemma 13. The operator L1 : 𝑄2 → 𝑄2 has a bounded
inverse L−1
1 which satisfies
𝑎𝜖 𝜖
E𝛾 := { ( , ) := (𝜖1 , 𝜖2 ) ∈ (−𝜖0 , 𝜖0 ) × (−𝜖0 , 𝜖0 ) : 󵄨󵄨 −1 󵄨󵄨 |ℎ|
2 + 𝜖 2 + 𝑎𝜖 󵄨󵄨L [ℎ]󵄨󵄨 ≤ 𝜎,𝑠 . (134)
󵄨 󵄨𝜎,𝑠 𝑁2
󵄨󵄨 󵄨 𝛾 󵄨 󵄨 𝛾
󵄨󵄨𝑙1 + 𝜖1 𝑙2 󵄨󵄨󵄨 > 󵄨󵄨 󵄨󵄨 , 󵄨󵄨󵄨𝑙2 + 𝜖2 𝑙1 󵄨󵄨󵄨 > 󵄨󵄨 󵄨󵄨 } Proof. L1 is diagonal in the Fourier basis of 𝑄 : 𝑒𝑖𝑙1 𝜑1 𝑒𝑖𝑙2 𝜑2
󵄨󵄨𝑙2 󵄨󵄨 󵄨󵄨𝑙1 󵄨󵄨
(128) with (𝑙1 , 𝑙2 ) ∈ Λ + ∪ {(0, 0)} ∪ Λ − with eigenvalues

for ∀𝑙1 , 𝑙2 ∈ Z \ {0}, and 0 < 𝛾 < 1/4, 𝜖0 ∈ (0, 1/2). − (2 + 𝜖) 𝑙12 , if 𝑙2 = 0,
𝑑𝑙1 ,𝑙2 = { (135)
−𝑎 (2 + 𝑎𝜖) 𝑙22 , if 𝑙1 = 0.
Lemma 12. For (𝑎, 𝜖) ∈ D𝛾 , the eigenvalues 𝐷𝑙1 ,𝑙2 of L𝑎,𝜖
restricted to 𝑃 satisfy The eigenvalues of L1 restricted to 𝑄2 (𝑁) verify |𝑑𝑙1 ,𝑙2 | ≥
󵄨󵄨 󵄨 𝑁2 /𝐶, where the constant 𝐶 depends on (𝜖, 𝑎), and (134)
󵄨󵄨𝐷𝑙1 ,𝑙2 󵄨󵄨󵄨 = 󵄨󵄨󵄨󵄨((2 + 𝜖) 𝑙1 + 𝑎𝜖𝑙2 ) (𝜖𝑙1 + (2 + 𝑎𝜖) 𝑙2 )󵄨󵄨󵄨󵄨 > 𝛾,
󵄨 󵄨 holds.
(129)
∀𝑙1 𝑙2 ≠
0.
Similarly, the solution of 𝑄1 -equation (124) is the Euler-
As a consequence, the operator L𝑎,𝜖 : 𝑃 → 𝑃 has a bounded Lagrange equation of the reduced Lagrangian action func-
tional:
inverse L−1
𝑎,𝜖 and satisfies
Φ𝜖,𝑁 : 𝐵2𝑅 ⊂ 𝑄1 󳨀→ R,
󵄨󵄨 −1 󵄨󵄨 |ℎ| (136)
󵄨󵄨L𝑎,𝜖 [ℎ]󵄨󵄨 ≤ 𝜎,𝑠 , ∀ℎ ∈ 𝑃. (130)
󵄨 󵄨𝜎,𝑠 𝛾 Φ𝜖,𝑁 (𝑞1 ) := Ψ𝜖 (𝑞1 + 𝑞2 (𝑞1 + 𝑝 (𝑞1 ))) .

Proof. Denote by [𝑥] the nearest integer close to 𝑥 and {𝑥} = Lemma 14. Φ𝜖,𝑁 ∈ 𝐶1 (𝐵2𝑅 , R) and a critical point 𝑞1 ∈ 𝐵2𝑅 of
𝑥−[𝑥]. If both 𝑙1 ≠−[(𝑎𝜖/(2+𝜖))𝑙2 ] and 𝑙2 ≠−[(𝜖/(2+𝑎𝜖))𝑙1 ], Φ𝜖,𝑁 is a solution of the bifurcation equation (124). Moreover,
then we have Φ𝜖,𝑁 can be written as
󵄨󵄨󵄨𝐷 󵄨󵄨󵄨
󵄨󵄨 𝑙1 ,𝑙2 󵄨󵄨 Φ𝜖,𝑁 (𝑞1 ) = 𝑐𝑜𝑛𝑠𝑡 + 𝜖 (Γ (𝑞1 ) + R𝜖,𝑁 (𝑞1 )) , (137)
󵄨󵄨 𝑎𝜖 󵄨󵄨 󵄨󵄨 𝜖 󵄨󵄨
= 󵄨󵄨󵄨󵄨(2 + 𝜖) (𝑙1 + 𝑙2 )󵄨󵄨󵄨󵄨 ⋅ 󵄨󵄨󵄨󵄨(2 + 𝑎𝜖) (𝑙2 + 𝑙1 )󵄨󵄨󵄨󵄨 > 1. where
󵄨 2+𝜖 󵄨 󵄨 2 + 𝑎𝜖 󵄨
(131) (2 + 𝜖) 2
Γ (𝑞1 ) := ∫ (𝜕𝜑1 𝑞1 )
T2 2
If 𝑙1 = −[(𝑎𝜖/(2 + 𝜖))𝑙2 ], then
󵄨󵄨 󵄨 + (1 + 𝑎 + 𝑎𝜖) (𝜕𝜑1 𝑞1 ) (𝜕𝜑2 𝑞1 )
󵄨󵄨𝐷𝑙1 ,𝑙2 󵄨󵄨󵄨
󵄨 󵄨 4
󵄨󵄨 󵄨󵄨 2𝑎 + 𝑎2 𝜖 2 𝑞
𝑎𝜖 + (𝜕𝜑2 𝑞1 ) − 𝑎3 (𝜑1 , 𝜑2 ) 1 ,
= 󵄨󵄨󵄨󵄨(2 + 𝜖) (𝑙1 + 𝑙2 )󵄨󵄨󵄨󵄨 2 4
󵄨 2+𝜖 󵄨 (138)
󵄨󵄨 𝑎𝜖 𝑎𝜖 󵄨󵄨 𝑞4
⋅ 󵄨󵄨󵄨󵄨𝜖 (𝑙1 + 𝑙2 ) + (2 + 𝑎𝜖) 𝑙2 − 𝜖𝑙2 󵄨󵄨󵄨󵄨 R𝜖,𝑁 (𝑞1 ) := ∫ 𝑎3 (𝜑1 , 𝜑2 ) 1
󵄨 2+𝜖 2+𝜖 󵄨 (132) T2 4
𝛾 󵄨 󵄨 𝑎𝜖 − 𝐹 (𝜑1 , 𝜑2 , 𝑞1 + 𝑞2 + 𝑝, 𝜖)
≥ 󵄨󵄨 󵄨󵄨 ⋅ 󵄨󵄨󵄨󵄨(2 + 𝜖) (𝜖 { 𝑙 } + 2𝑙2
󵄨󵄨𝑙2 󵄨󵄨 󵄨 2+𝜖 2
1
󵄨󵄨 + 𝑓 (𝜑1 , 𝜑2 , V, 𝜖) (𝑞2 + 𝑝) ,
𝜖 2
+𝑎𝜖 (1 − ) 𝑙2 )󵄨󵄨󵄨󵄨 ≥ 𝛾.
2+𝜖 󵄨
and, for some positive constant 𝐶2 (𝑅) ≥ 𝐶1 (𝑅), we can get
In the same way, if 𝑙2 = −[(𝜖/(2 + 𝑎𝜖))𝑙1 ], then we have
󵄨󵄨 󵄨 1
󵄨󵄨R𝜖,𝑁 (𝑞1 )󵄨󵄨󵄨 ≤ 𝐶2 (𝑅) (√𝜖 + 𝜖𝛾 + 2 ) ,
−1
󵄨󵄨 󵄨
󵄨󵄨𝐷𝑙1 ,𝑙2 󵄨󵄨󵄨 𝑁
󵄨 󵄨
󵄨󵄨 󵄨󵄨 󸀠 󵄨 1
𝛾 󵄨 𝜖 󵄨󵄨R𝜖,𝑁 (𝑞1 ) [ℎ]󵄨󵄨󵄨 ≤ 𝐶2 (𝑅) (√𝜖 + 𝜖𝛾−1 + 2 ) |ℎ|𝐻1 , (139)
≥ 󵄨󵄨 󵄨󵄨 ⋅ 󵄨󵄨󵄨󵄨(2 + 𝑎𝜖) (𝑎𝜖 { 𝑙 } + 2𝑙1 󵄨 󵄨 𝑁
󵄨󵄨𝑙1 󵄨󵄨 󵄨󵄨 2 + 𝑎𝜖 1 (133)
∀ℎ ∈ 𝑄1 .
󵄨󵄨
𝑎2 𝜖 󵄨
+𝜖 (1 − ) 𝑙1 )󵄨󵄨󵄨󵄨 ≥ 𝛾.
2 + 𝑎𝜖 󵄨󵄨 The problem of finding nontrivial solutions of the 𝑄1 -
equation is reduced to finding nontrivial critical points of
the reduced action functional Φ𝜖,𝑁 in 𝐵2𝑅 . By (137), this is
14 Abstract and Applied Analysis

equivalent to finding critical points of the rescaled functional and, thus, we deduce that 𝑝̂ ∈ H𝜎󸀠󸀠 ,𝑠+1 , for all 0 < 𝜎󸀠󸀠 < 𝜎󸀠
denoted by Φ̂ 𝜖,𝑁 and called the reduced action functional 󸀠󸀠 󸀠
(satisfying 𝐶|𝑙2 |𝑒|𝑙2 |𝜎 < 𝑒|𝑙2 |𝜎 ), and |𝜕𝜑1 𝑝|
̂ 󸀠󸀠 = 𝑂(|𝜀|/𝛾). By
𝜎 ,𝑠
̂ 𝜖,𝑁 (𝑞1 ) = Γ (𝑞1 ) + R𝜖,𝑁 (𝑞1 )
Φ (144),
𝜖 (2 + 𝜖) 𝜕𝜑21 𝑝̂
𝑞4
≡ (Q (𝑞1 ) − ∫ 𝑎3 (𝜑1 , 𝜑2 ) 1 + R𝜖,𝑁 (𝑞1 )) ,
T2 4 = − (4 + 2𝑎𝜖 + 2𝜖 + 2𝑎𝜖2 ) 𝜕𝜑1 𝜕𝜑2 𝑝̂ (148)

(140) − 𝜖 ((2𝑎 + 𝑎2 𝜖) 𝜕𝜑22 + Π𝑃 𝑓 (𝜑1 , 𝜑2 , 𝑞̂𝜖 + 𝑝, 𝜖)) 𝑝,

̂

where the quadratic form ̂ 𝜎,𝑠+2 = 𝑂(|𝜖|/𝛾), with 0 < 𝜎 < 𝜎󸀠󸀠 .
we get 𝑝̂ ∈ H𝜎,𝑠+2 , and |𝑝|
(2 + 𝜖) 2 Thus, (19) follows with 𝑠 = 𝑠 + 2, 0 < 𝜎 < 𝜎󸀠󸀠 . By (6), 𝑢(𝑡, 𝑥) =
Q (𝑞1 ) = ∫ (𝜕𝜑1 𝑞1 ) + (1 + 𝑎 + 𝑎𝜖) (𝜕𝜑1 𝑞1 ) (𝜕𝜑2 𝑞1 ) √|𝜖|V((1 + 𝜖)𝑡 − 𝑥, (1 + 𝑎𝜖)𝑡 + 𝑥) is the solution of (1), for all
T2 2
(𝑎, 𝜖) ∈ D𝛾 . Obviously, V depends on both variables (𝜑1 , 𝜑2 )
2𝑎 + 𝑎2 𝜖 2 independently. So, 𝑢(𝑡, 𝑥) is a quasiperiodic solution of (1),
+ (𝜕𝜑2 𝑞1 ) with frequencies (𝜔1 , 𝜔2 ) = (1 + 𝜖, 1 + 𝑎𝜖).
2
(141)
is positive definite on 𝑄+ , negative definite on 𝑄− , and zero 5. Conclusion
definite on 𝑄0 . For 𝑞1 = 𝑞+ + 𝑞− + 𝑞0 ∈ 𝑄1 , we have In this paper, for the completely resonant nonlinear wave
(2 + 𝜖) 2 𝛼 󵄨 󵄨2 equations, under periodic boundary conditions, we obtain
Q (𝑞+ ) = ∫ (𝜕𝜑1 𝑞+ ) = + 󵄨󵄨󵄨𝑞+ 󵄨󵄨󵄨𝐻1 , the existence and regularity of quasiperiodic solutions. The
T2 2 2
forced terms we consider are quasiperiodic, and, according
Q (𝑞0 ) = 0, (142) to the linking theorem, the bifurcation equations are solved
by variational method. Moreover, the solutions depending
(𝑎 + 𝑎𝜖) 2 𝛼 󵄨 󵄨2
Q (𝑞− ) = ∫ (𝜕𝜑1 𝑞− ) = − − 󵄨󵄨󵄨𝑞− 󵄨󵄨󵄨𝐻1 , on the the spatial and time variables are coupled and in the
T2 2 2 form of traveling waves. In [28], Yuan got the existence of
where the positive constants 𝛼+ , 𝛼− are bounded away from quasiperiodic solutions with 𝑛 ∈ N (𝑛 ≥ 3) frequencies by
zero and independent of 𝜖. The following steps of finding the KAM theory, in which the form of the solutions is 𝑢(𝑡, 𝑥) =
nontrivial solutions of the 𝑄1 -equation are similar to Lemmas V(𝜔1 𝑡, 𝜔2 𝑡, . . . , 𝜔𝑛 𝑡, 𝑥). In the future work, we will investigate
9 and 11 in Section 3.4. the existence of quasiperiodic solutions with the traveling
wave form as 𝑢(𝑡, 𝑥) = V(𝜔1 𝑡 + 𝑥, 𝜔2 + 𝑥, . . . , 𝜔𝑛 𝑡 + 𝑥).
Proof of Theorem 2. We can get the solution of (8) as follows:

V = √|𝜖| [̂
𝑞1 + 𝑞2 (̂ 𝑞1 )] = √|𝜖| [̂
𝑞1 ) + 𝑝 (̂ 𝑞𝜖 + 𝑝 (̂
𝑞1 )] . (143) Conflict of Interests
According to Lemma 13, by the regularizing property of the The authors declare that there is no conflict of interests
operator L1 , the solution 𝑞̂𝜖 := 𝑞̂1 +𝑞2 (̂
𝑞1 ) of the (𝑄)-equation regarding the publication of this paper.
belongs to H𝜎,𝑠+2 ∩ 𝑄. By the 𝑃-equation
Acknowledgments
(𝜖 (2 + 𝜖) 𝜕𝜑21 + (4 + 2𝑎𝜖 + 2𝜖 + 2𝑎𝜖2 ) 𝜕𝜑1 𝜕𝜑2 ) 𝑝̂
(144) The authors express their sincere thanks to Professor Yong
= −𝜖 ((2𝑎 + 𝑎2 𝜖) 𝜕𝜑22 + Π𝑃 𝑓 (𝜑1 , 𝜑2 , 𝑞̂𝜖 + 𝑝, 𝜖)) 𝑝,
̂ Li for his instructions and many invaluable suggestions. The
first author was partially supported by NSFC Grant (nos.
where 𝑝̂ = 𝑝(̂
𝑞1 ), we can get that 11001042, 11171056, 11171130, and 11271062) and SRFDP Grant
(no. 20100043120001). The second author is supported by
−𝜖 ((2𝑎 + 𝑎2 𝜖) 𝜕𝜑22 + Π𝑃 𝑓 (𝜑1 , 𝜑2 , 𝑞̂𝜖 + 𝑝, 𝜖)) 𝑝̂ ∈ H𝜎󸀠 ,𝑠 , NSFC Grant (nos. 11001041, 11201360, 11101170, 11202192, and
(145) 10926105), SRFDP Grant (no. 200802001008), and the State
󸀠
Scholarship Fund of the China Scholarship Council (nos.
for 0 < 𝜎󸀠 < 𝜎, satisfying |𝜖(2𝑎 + 𝑎2 𝜖)𝑙22 |𝑒|𝑙2 |𝜎 < 𝑒|𝑙2 |𝜎 . For 2011662521 and 2011842509).
(𝑎, 𝜖) ∈ D𝛾 , the eigenvalues of operator

𝜖 (2 + 𝜖) 𝜕𝜑21 + (4 + 2𝑎𝜖 + 2𝜖 + 2𝑎𝜖2 ) 𝜕𝜑1 𝜕𝜑2 (146)

References
[1] J. Bourgain, “Quasi-periodic solutions of Hamiltonian per-
restricted to 𝑃 satisfy turbations of 2D linear Schrödinger equations,” Annals of
󵄨 󵄨
𝛾 󵄨󵄨𝑙 󵄨󵄨
Mathematics, vol. 148, no. 2, pp. 363–439, 1998.
󵄨󵄨 󵄨
󵄨󵄨𝜖 (2 + 𝜖) 𝑙12 + (4 + 2𝑎𝜖 + 𝜖 + 𝑎𝜖2 ) 𝑙1 𝑙2 󵄨󵄨󵄨 ≥ |2 + 𝜖| 󵄨 󵄨 1󵄨 󵄨 , [2] J. Bourgain, “Periodic solutions of nonlinear wave equations,” in
󵄨 󵄨 󵄨󵄨𝑙2 󵄨󵄨
󵄨 󵄨 (147) Harmonic Analysis and Partial Differential Equations, Chicago
Lectures in Mathematics, pp. 69–97, University of Chicago
∀𝑙1 ≠
0, 𝑙2 ≠
0, Press, Chicago, Ill, USA, 1999.
Abstract and Applied Analysis 15

[3] J. Bourgain, Green’s Function Estimates for Lattice Schrödinger [22] M. Berti and P. Bolle, “Periodic solutions of nonlinear wave
Operators and Applications, vol. 158 of Annals of Mathematics equations with general nonlinearities,” Communications in
Studies, Princeton University Press, Princeton, NJ, USA, 2005. Mathematical Physics, vol. 243, no. 2, pp. 315–328, 2003.
[4] L. Chierchia and J. You, “KAM tori for 1D nonlinear wave equa- [23] M. Berti and P. Bolle, “Multiplicity of periodic solutions of
tions with periodic boundary conditions,” Communications in nonlinear wave equations,” Nonlinear Analysis: Theory, Methods
Mathematical Physics, vol. 211, no. 2, pp. 497–525, 2000. & Applications, vol. 56, no. 7, pp. 1011–1046, 2004.
[5] W. Craig and C. E. Wayne, “Newton’s method and periodic [24] M. Berti and P. Bolle, “Cantor families of periodic solutions for
solutions of nonlinear wave equations,” Communications on completely resonant nonlinear wave equations,” Duke Mathe-
Pure and Applied Mathematics, vol. 46, no. 11, pp. 1409–1498, matical Journal, vol. 134, no. 2, pp. 359–419, 2006.
1993. [25] B. V. Lidskii and E. I. Shul’man, “Periodic solutions of the
[6] Y. Gao, Y. Li, and J. Zhang, “Invariant tori of nonlinear equation 𝑢𝑡𝑡 − 𝑢𝑥𝑥 + 𝑢3 = 0,” Functional Analysis and Its
Schrödinger equation,” Journal of Differential Equations, vol. Applications, vol. 22, no. 4, pp. 332–33, 1988.
246, no. 8, pp. 3296–3331, 2009. [26] G. Gentile, V. Mastropietro, and M. Procesi, “Periodic solutions
[7] J. Geng and J. You, “A KAM theorem for Hamiltonian partial for completely resonant nonlinear wave equations with Dirich-
differential equations in higher dimensional spaces,” Commu- let boundary conditions,” Communications in Mathematical
nications in Mathematical Physics, vol. 262, no. 2, pp. 343–372, Physics, vol. 256, no. 2, pp. 437–490, 2005.
2006.
[27] G. Gentile and M. Procesi, “Conservation of resonant periodic
[8] J. Geng and X. Ren, “Lower dimensional invariant tori with solutions for the one-dimensional nonlinear Schrödinger equa-
prescribed frequency for nonlinear wave equation,” Journal of tion,” Communications in Mathematical Physics, vol. 262, no. 3,
Differential Equations, vol. 249, no. 11, pp. 2796–2821, 2010. pp. 533–553, 2006.
[9] J. Geng, “Almost periodic solutions for a class of higher
[28] X. Yuan, “Quasi-periodic solutions of completely resonant
dimensional Schrödinger equations,” Frontiers of Mathematics
nonlinear wave equations,” Journal of Differential Equations, vol.
in China, vol. 4, no. 3, pp. 463–482, 2009.
230, no. 1, pp. 213–274, 2006.
[10] J. Pöschel, “Quasi-periodic solutions for a nonlinear wave
[29] M. Berti and M. Procesi, “Quasi-periodic solutions of com-
equation,” Commentarii Mathematici Helvetici, vol. 71, no. 2, pp.
pletely resonant forced wave equations,” Communications in
269–296, 1996.
Partial Differential Equations, vol. 31, no. 6, pp. 959–985, 2006.
[11] X. Yuan, “A KAM theorem with applications to partial dif-
ferential equations of higher dimensions,” Communications in [30] M. Procesi, “Quasi-periodic solutions for completely resonant
Mathematical Physics, vol. 275, no. 1, pp. 97–137, 2007. non-linear wave equations in ID and 2D,” Discrete and Contin-
uous Dynamical Systems, vol. 13, no. 3, pp. 541–552, 2005.
[12] C. E. Wayne, “Periodic and quasi-periodic solutions of non-
linear wave equations via KAM theory,” Communications in [31] P. Baldi, “Quasi-periodic solutions of the equation V𝑡𝑡 −V𝑥𝑥 +V3 =
Mathematical Physics, vol. 127, no. 3, pp. 479–528, 1990. 𝑓(V),” Discrete and Continuous Dynamical Systems, vol. 15, no.
[13] V. Benci and P. Rabinowitz, “Critical point theorems for indef- 3, pp. 883–903, 2006.
inite functionals,” Inventiones Mathematicae, vol. 52, no. 3, pp. [32] Y. Ma and W. Lou, “Quasi-periodic solutions of completely reso-
241–273, 1979. nant wave equations with quasi-periodically forced vibrations,”
[14] H. Brézis and L. Nirenberg, “Forced vibrations for a nonlinear Acta Applicandae Mathematicae, vol. 112, no. 3, pp. 309–322,
wave equation,” Communications on Pure and Applied Mathe- 2010.
matics, vol. 31, no. 1, pp. 1–30, 1978. [33] D. Bambusi, “Lyapunov center theorem for some nonlinear
[15] H. Brézis, J.-M. Coron, and L. Nirenberg, “Free vibrations for PDE’s: a simple proof,” Annali della Scuola Normale Superiore
a nonlinear wave equation and a theorem of P. Rabinowitz,” di Pisa. Classe di Scienze, vol. 29, no. 4, pp. 823–837, 2000.
Communications on Pure and Applied Mathematics, vol. 33, no. [34] A. Ambrosetti and M. Badiale, “Homoclinics: Poincaré-
5, pp. 667–684, 1980. Melnikov type results via a variational approach,” Annales de
[16] E. R. Fadell and P. H. Rabinowitz, “Generalized cohomological l’Institut Henri Poincare (C) Non Linear Analysis, vol. 15, no. 2,
index theories for Lie group actions with an application to pp. 233–252, 1998.
bifurcation questions for Hamiltonian systems,” Inventiones [35] W. M. Schmidt, Diophantine Approximation, vol. 785 of Lecture
Mathematicae, vol. 45, no. 2, pp. 139–174, 1978. Notes in Mathematics, Springer, Berlin, Germany, 1980.
[17] P. H. Rabinowitz, “Periodic solutions of nonlinear hyperbolic
partial differential equations,” Communications on Pure and
Applied Mathematics, vol. 20, no. 1, pp. 145–205, 1967.
[18] P. H. Rabinowitz, “Free vibrations for a semilinear wave equa-
tion,” Communications on Pure and Applied Mathematics, vol.
31, no. 1, pp. 31–68, 1978.
[19] P. H. Rabinowitz, Minimax Methods in Critical Point Theory
with Applications to Differential Equations, vol. 65 of CBMS
Regional Conference Series in Mathematics, American Mathe-
matical Society, Providence, RI, USA, 1986.
[20] D. Bambusi and M. Berti, “A Birkhoff-Lewis-type theorem
for some Hamiltonian PDEs,” SIAM Journal on Mathematical
Analysis, vol. 37, no. 1, pp. 83–102, 2005.
[21] D. Bambusi and S. Paleari, “Families of periodic solutions of
resonant PDEs,” Journal of Nonlinear Science, vol. 11, no. 1, pp.
69–87, 2001.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 357534, 9 pages
http://dx.doi.org/10.1155/2014/357534

Research Article
Periodic Solutions of a Stage-Structured Plant-Hare Model with
Toxin-Determined Functional Responses

Huicheng Wang,1 Patricia J. Y. Wong,2 and Yonghui Xia1

1
Department of Mathematics, Zhejiang Normal University, Jinhua, Zhejiang 321004, China
2
School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore 639798

Correspondence should be addressed to Yonghui Xia; yhxia@zjnu.cn

Received 8 April 2014; Revised 25 April 2014; Accepted 26 April 2014; Published 8 May 2014

Academic Editor: Tonghua Zhang

Copyright © 2014 Huicheng Wang et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

The purpose of this paper is to obtain some sufficient conditions for the global existence of multiple positive periodic solutions of
a delayed stage-structured plant-hare model with a toxin-determined functional response. Some novel estimation techniques to
construct two open subsets for a priori bounds are employed.

1. Introduction rate can be ignored, while the mature predators are respon-
sible for the prey. For example, one can refer to [14, 15] and
A lot of classical predator-prey models have been well studied the references cited therein. To discuss the effects of Holling
(e.g., see [1–12]). Recently, Gao et al. [13] considered a type IV functional responses on a stage-structured model, the
nonautonomous plant-hare dynamical system with a toxin- authors in [16] proposed the following delayed system:
determined functional response given by
𝑑𝑥 (𝑡)
= 𝑥 (𝑡) [𝑟1 (𝑡) − 𝑎1 (𝑡)
𝑁 (𝑡) 𝑑𝑡
𝑁̇(𝑡) = 𝑟 (𝑡) 𝑁 (𝑡) [1 − ] − 𝐶 (𝑁 (𝑡)) 𝑃 (𝑡) , 𝑡
𝐾 (1) ×∫ 𝐾 (𝑡 − 𝑠) 𝑥 (𝑠) 𝑑𝑠
−∞
𝑃̇(𝑡) = 𝐵 (𝑡) 𝐶 (𝑁 (𝑡)) 𝑃 (𝑡) − 𝑑 (𝑡) 𝑃 (𝑡) ,
𝑎2 (𝑡) 𝑦2 (𝑡)
− ],
𝑓 (𝑁 (𝑡)) 𝑥2 (𝑡) /𝑚 + 𝑥 (𝑡) + 𝑎
𝐶 (𝑁 (𝑡)) = 𝑓 (𝑁 (𝑡)) [1 − ],
4𝐺 𝑑𝑦1 (𝑡) 𝑏 (𝑡) 𝑥 (𝑡) 𝑦2 (𝑡)
(2) = 21 − 𝛽 (𝑡) 𝑦1 (𝑡)
𝑒𝛿𝑁 (𝑡) 𝑑𝑡 𝑥 (𝑡) /𝑚 + 𝑥 (𝑡) + 𝑎
𝑓 (𝑁 (𝑡)) = , 𝑡
1 + ℎ𝑒𝛿𝑁 (𝑡) − 𝑏1 (𝑡 − 𝜏) exp (− ∫ 𝛽 (𝑠) 𝑑𝑠)
𝑡−𝜏
where 𝑁(𝑡) denotes the density of plant at time 𝑡 and 𝑃(𝑡)
𝑥 (𝑡 − 𝜏) 𝑦2 (𝑡 − 𝜏)
denotes the herbivore biomass at time 𝑡. × ,
On the other hand, many experts argued that the 𝑥2 (𝑡 − 𝜏) /𝑚 + 𝑥 (𝑡 − 𝜏) + 𝑎
predator-prey models should be modified to fit the more 𝑑𝑦2 (𝑡) 𝑡
realistic environment. They suggested that one should take = 𝑏1 (𝑡 − 𝜏) exp (− ∫ 𝛽 (𝑠) 𝑑𝑠)
𝑑𝑡 𝑡−𝜏
the stage structure factor into consideration. Because it is very
unrealistic to assume that each individual predator admits the 𝑥 (𝑡 − 𝜏) 𝑦2 (𝑡 − 𝜏)
×
same ability of attacking in the classical predator-prey mod- 𝑥2 (𝑡 − 𝜏) /𝑚 + 𝑥 (𝑡 − 𝜏) + 𝑎
els. They divided the individuals into two stages in life history, − 𝑟2 (𝑡) 𝑦2 (𝑡) .
namely, immature and mature stages, where the rate of the
immature predator attacking the prey and the reproductive (3)
2 Abstract and Applied Analysis

However, Holling IV type functional response is not For any continuous 𝜔-periodic function 𝑓(𝑡), we always
appropriate for the plant-hare model if we explore the impact adopt the following notations throughout this paper:
of plant toxicity on the dynamics of plant-hare interactions.
1 𝜔
Because such kind of plant can produce toxicity to protect 𝑓= ∫ 𝑓 (𝑡) 𝑑𝑡, 𝑓𝐿 = min 𝑓 (𝑡) ,
itself. Therefore, in the present paper, we discuss the 𝜔 0 𝑡∈[0,𝜔]
(5)
stage-structured plant-hare model with toxin-determined 𝑀
functional response as follows: 𝑓 = max𝑓 (𝑡) ,
[0,𝜔]

where 𝑓 is a continuous 𝜔-periodic function.

𝑑𝑥 (𝑡)
= 𝑥 (𝑡) [𝑟1 (𝑡) − 𝑎 (𝑡) 𝑥 (𝑡) The purpose of this paper is to obtain some sufficient con-
𝑑𝑡 ditions for the global existence of multiple positive periodic
solutions of system (4). Our method is based on Mawhin’s
4𝐺𝑒𝛿𝑦2 (𝑡) + (4𝐺ℎ − 1) 𝑒2 𝛿2 𝑥 (𝑡) 𝑦2 (𝑡) coincidence degree and novel estimation techniques for a
− ],
4𝐺(1 + ℎ𝑒𝛿𝑥(𝑡))2 priori bounds of unknown solutions to 𝐿𝑥 = 𝜆𝑁𝑥. To the
best of our knowledge, it is the first time that a delayed
𝑑𝑦1 (𝑡) stage-structured plant-hare dynamical system with a toxin-
= − 𝛽 (𝑡) 𝑦1 (𝑡)
𝑑𝑡 determined functional response has been proposed and
studied by using this method.
+ ((𝑏1 (𝑡) 4𝐺𝑒𝛿𝐵𝑥 (𝑡) 𝑦2 (𝑡)
𝑡
Remark 1. The term exp{− ∫𝑡−𝜏 𝛽(𝑠)𝑑𝑠}(𝑥(𝑡−𝜏)𝑦2 (𝑡−𝜏)/(𝑥2 (𝑡−
+ (4𝐺ℎ − 1) 𝑒2 𝛿2 𝐵𝑥2 (𝑡) 𝑦2 (𝑡)) 𝜏)/𝑚 + 𝑥(𝑡 − 𝜏) + 𝑎)) in the third equation of (4) involves
𝑥(𝑡 − 𝜏)𝑦2 (𝑡 − 𝜏) instead of 𝑥(𝑡 − 𝜏)𝑦2 (𝑡); the method used
−1 in [13] cannot be applied to system (4) directly. Thus, novel
× (4𝐺(1 + ℎ𝑒𝛿𝑥 (𝑡))2 ) )
estimation techniques must be employed for a priori bounds
𝑡
of unknown solutions to the operator equation 𝐿𝑥 = 𝜆𝑁𝑥.
− 𝑏1 (𝑡 − 𝜏) exp {− ∫ 𝛽 (𝑠) 𝑑𝑠} More specifically, integrating the second equation of system
𝑡−𝜏 (1) over [0, 𝜔], the authors in [13] obtained
𝜔
× ((4𝐺𝑒𝛿𝐵𝑥 (𝑡 − 𝜏) 𝑦2 (𝑡 − 𝜏) ∫ ((4𝐺𝑒𝛿𝐵 (𝑡) exp {𝑢1 (𝑡)}
0
2 2 2
+ (4𝐺ℎ − 1) 𝑒 𝛿 𝐵𝑥 (𝑡 − 𝜏) 𝑦2 (𝑡 − 𝜏)) + (4𝐺ℎ − 1) 𝑒2 𝛿2 𝐵 (𝑡) exp {2𝑢1 (𝑡)})
(6)
−1
× (4𝐺(1 + ℎ𝑒𝛿𝑥 (𝑡 − 𝜏))2 ) ) , × (4𝐺(1 + ℎ𝑒𝛿 exp {𝑢1 (𝑡)}) ) ) 𝑑𝑡
2 −1

𝑑𝑦2 (𝑡) 𝑡
= 𝑏1 (𝑡 − 𝜏) exp {− ∫ 𝛽 (𝑠) 𝑑𝑠} = 𝑑𝜔.
𝑑𝑡 𝑡−𝜏
It follows that
× ((4𝐺𝑒𝛿𝐵𝑥 (𝑡 − 𝜏) 𝑦2 (𝑡 − 𝜏) 𝜔 4𝐺𝑒𝛿𝐵 (𝑡) exp {𝑢1 (𝑡)}
𝑑𝜔 ≥ ∫ 2
𝑑𝑡. (7)
2 2 2 0 4𝐺(1 + ℎ𝑒𝛿 exp {𝑢1 (𝑡)})
+ (4𝐺ℎ − 1) 𝑒 𝛿 𝐵𝑥 (𝑡 − 𝜏) 𝑦2 (𝑡 − 𝜏))
−1
By some arguments, this inequality then leads them to
× (4𝐺(1 + ℎ𝑒𝛿𝑥 (𝑡 − 𝜏))2 ) )
𝑑ℎ2 𝑒2 𝛿2 exp (2𝑢1 (𝜂1 ))
− 𝑟2 (𝑡) 𝑦2 (𝑡) ,
(4) − (𝑒𝛿𝐵 exp {−2𝑟𝜔} − 2ℎ𝑒𝛿𝑑) (8)

× exp {𝑢1 (𝜂1 )} + 𝑑 > 0,

where 𝑥(𝑡) denotes the density of the plant at time 𝑡, 𝑦1 (𝑡)
is the density of immature individual hares at time 𝑡, and which implies that
𝑦2 (𝑡) denotes the density of mature individual hares at time 𝑡,
respectively; 𝑟1 (𝑡), 𝑟2 (𝑡), 𝑎(𝑡), 𝑏1 (𝑡), and 𝛽(𝑡) are continuously 𝑥1 (𝜂1 ) < ln ℎ−0 , 𝑥1 (𝜂1 ) > ln ℎ+0 , (9)
positive periodic functions with period 𝜔. 𝐵 is the conversion
rate, 𝑒 is the encounter rate per hare, 𝛿 is the fraction of food where
items encountered that the hares ingest, 𝐺 measures the toxic- (𝑒𝛿𝐵 exp {−2𝑟𝜔} − 2ℎ𝑒𝛿𝑑) ± √Δ 1
ity level, and ℎ is the time for handing one unit of plant. 𝑒, 𝛿, 𝐺, ℎ±0 = ,
and ℎ are positive real constants. 𝑟1 (𝑡) is the intrinsic growth 2𝑑ℎ2 𝑒2 𝛿2 (10)
rate of the prey, 𝑎(𝑡) is the density-dependent coefficient of 2 2 2 2 2
the plant, and 𝑟2 (𝑡) is the death rate of the mature hares. Δ 1 = [𝑒𝛿𝐵 exp{−2𝑟𝜔} − 2ℎ𝑒𝛿𝑑] − 4𝑑 ℎ 𝑒 𝛿 .
Abstract and Applied Analysis 3

It should be noted that it is possible to construct two open Throughout, we assume the following:
subsets Ω1 and Ω2 due to (9). The essential reason to obtain
(9) is the inequality (7). In inequality (7), there is no variable (𝐴 1 ) 1/4ℎ < 𝐺 < 1/3ℎ;
𝑢2 (𝑡) and only one variable 𝑢1 (𝑡). (𝐴 2 ) 4ℎ𝑟2𝑀 exp{𝜏𝛽𝑀} exp{2𝑟1 𝜔}/𝑏1𝐿 < 𝐵 < 4𝐺𝑟2𝐿 ℎ2 (𝑏1𝑀)−1
However, since the term 𝑥(𝑡 − 𝜏)𝑦2 (𝑡 − 𝜏) is in the third exp{𝜏𝛽𝐿 }/(4𝐺ℎ − 1).
equation of (4), by same arguments in [13], we will see that
We further introduce six positive numbers which will be
𝜔 𝑡
used later as follows:
∫ 𝑏1 (𝑡 − 𝜏) exp {− ∫ 𝛽 (𝑠) 𝑑𝑠}
0 𝑡−𝜏
ℎ± = ((𝑏1𝐿 𝑒𝛿𝐵 exp {−𝜏𝛽𝑀} exp {−2𝑟1 𝜔} − 2ℎ𝑒𝛿𝑟2𝑀)
× exp {𝑢2 (𝑡 − 𝜏) − 𝑢2 (𝑡)}
−1
× ((4𝐺𝑒𝛿𝐵 exp {𝑢1 (𝑡 − 𝜏)} ±√Δ 1 ) × (2𝑟2𝑀ℎ2 𝑒2 𝛿2 ) ,
(11)
2 2
+ (4𝐺ℎ − 1) 𝑒 𝛿 𝐵 exp {2𝑢1 (𝑡 − 𝜏)}) 𝑙± = ([4𝐺ℎ2 𝑒𝛿𝐵 exp {2𝑟1 𝜔}
−1
× (4𝐺(1 + ℎ𝑒𝛿 exp {𝑢1 (𝑡 − 𝜏)})2 ) ) 𝑑𝑡 −1
− 2ℎ𝑒𝛿 (4𝐺ℎ2 𝑟2𝐿 (𝑏1𝑀) exp {𝜏𝛽𝐿 } − (4𝐺ℎ − 1) 𝐵)]
= 𝑟2 𝜔.
± √Δ 2 )
Note that both 𝑢1 and 𝑢2 appear simultaneously in the above
equality. If we were to use the same ideas in [13], then the −1
× (2ℎ2 𝑒2 𝛿2 [4𝐺ℎ2 𝑟2𝐿 (𝑏1𝑀) exp {𝜏𝛽𝐿 }
above equality does not lead us anywhere. Thus, some new
arguments should be employed to obtain a priori bounds for −1
𝑢1 . To see how to overcome this difficulty, the reader can refer − (4𝐺ℎ − 1) 𝐵]) ,
to (33)–(56) in Section 2.
−1
(4𝐺𝑒𝛿𝐵 − 8𝐺ℎ𝑒𝛿𝑟2 𝑏 ) ± √Δ 3
Remark 2. It should be noted that the standard estimation 𝑢± = ,
−1
techniques used in [16] are not applicable to the system (4) 2 [4𝐺𝑟2 𝑏 ℎ2 𝑒2 𝛿2 − (4𝐺ℎ − 1) 𝑒2 𝛿2 𝐵]
either, due to the term 𝐶(𝑁(𝑡)). If we were to use the standard
(13)
arguments in [16], we can not obtain two positive roots of
exp(𝑢1 (𝜉1 )). Consequently, we can not construct two open where
subsets. Thus, we can not obtain two positive solutions in
2
these two open subsets. Δ 1 = [𝑏1𝐿 𝑒𝛿𝐵 exp{−𝜏𝛽𝑀} exp{−2𝑟1 𝜔} − 2ℎ𝑒𝛿𝑟2𝑀]
2
2. Existence of Multiple Positive − 4(𝑟2𝑀) ℎ2 𝑒2 𝛿2 ,
Periodic Solutions
Δ 2 = [4𝐺ℎ2 𝑒𝛿𝐵 exp {2𝑟1 𝜔}
In this section, we will study the existence of multiple periodic
solutions of (4). We recall a few concepts and results from −1 2
[17]. −2ℎ𝑒𝛿 (4𝐺ℎ2 𝑟2𝐿 (𝑏1𝑀) exp {𝜏𝛽𝐿 } − (4𝐺ℎ − 1) 𝐵)]

−1 2
Lemma 3 (see [17]). Let Ω ⊂ 𝑋 be an open bounded set. Let 𝐿 − 4ℎ2 𝑒2 𝛿2 [4𝐺ℎ2 𝑟2𝐿 (𝑏1𝑀) exp {𝜏𝛽𝐿 } − (4𝐺ℎ − 1) 𝐵] ,
be a Fredholm mapping of index zero and 𝑁𝐿-compact on Ω.
Assume −1 2
Δ 3 = (4𝐺𝑒𝛿𝐵 − 8𝐺ℎ𝑒𝛿𝑟2 𝑏 )
(a) for each 𝜆 ∈ (0, 1), 𝑥 ∈ 𝜕Ω ∩ Dom 𝐿, 𝐿𝑥 ≠
𝜆𝑁𝑥;
−1 −1
(b) for each 𝑥 ∈ 𝜕Ω ∩ Ker 𝐿, 𝑄𝑁𝑥 ≠ 0; − 16𝐺𝑟2 𝑏 [4𝐺𝑟2 𝑏 ℎ2 𝑒2 𝛿2 − (4𝐺ℎ − 1) 𝑒2 𝛿2 𝐵] ,
(c) deg{𝐽𝑄𝑁, Ω ∩ Ker 𝐿, 0} ≠ 0. (14)
Then 𝐿𝑥 = 𝑁𝑥 has at least one solution in Ω ∩ Dom 𝐿. 𝜔 𝑡+𝜏
𝑏 = (1/𝜔) ∫0 𝑏1 (𝑡) exp{− ∫𝑡 𝛽(𝑠)}𝑑𝑠𝑑𝑡. Under assumptions
Lemma 4. If 𝛽(𝑡) and 𝑔(𝑡) are 𝜔-periodic functions, then the (𝐴 1 ) and (𝐴 2 ), it is not difficult to show that
system
𝑙− < 𝑢− < ℎ− < ℎ+ < 𝑢+ < 𝑙+ . (15)
𝑑𝑦 (𝑡)
= 𝛽 (𝑡) 𝑦 (𝑡) + 𝑔 (𝑡) (12) Theorem 5. In addition to (𝐴 1 ) and (𝐴 2 ), suppose that
𝑑𝑡
has a unique 𝜔-periodic solution which can be represented as (𝐴 3 ) 𝑟1 − 𝑎 exp{ln 𝑙+ + 2𝑟𝜔} > 0.
𝑡 𝑡
𝑦(𝑡) = ∫−∞ exp(∫𝑠 𝛽(𝜎)𝑑𝜎)𝑔(𝑠)𝑑𝑠. Then system (4) has at least two positive 𝜔-periodic solutions.
4 Abstract and Applied Analysis

Proof. Note that the first equation and the third equation of here |⋅| denotes the Euclidean norm. Then 𝑋 and 𝑌 are Banach
(4) can be separated from the whole system. Consider the spaces with the norm ‖ ⋅ ‖. Set
following subsystem:
𝐿 : Dom 𝐿 ∩ 𝑋,
𝑑𝑥 (𝑡)
= 𝑥 (𝑡) [𝑟1 (𝑡) − 𝑎 (𝑡) 𝑥 (𝑡) (21)
𝑑𝑡 𝑑𝑢1 (𝑡) 𝑑𝑢2 (𝑡) 𝑇
𝐿(𝑢1 (𝑡), 𝑢2 (𝑡))𝑇 = ( , ) ,
𝑑𝑡 𝑑𝑡
4𝐺𝑒𝛿𝑦2 (𝑡) + (4𝐺ℎ − 1) 𝑒2 𝛿2 𝑥 (𝑡) 𝑦2 (𝑡)
− ],
4𝐺(1 + ℎ𝑒𝛿𝑥(𝑡))2 where Dom 𝐿 = {(𝑢1 (𝑡), 𝑢2 (𝑡))𝑇 ∈ 𝐶1 (R, R2 )}. Further, 𝑁 :
𝑡
𝑋 → 𝑋 is defined by
𝑑𝑦2 (𝑡)
= 𝑏1 (𝑡 − 𝜏) exp {− ∫ 𝛽 (𝑠) 𝑑𝑠}
𝑑𝑡 𝑡−𝜏 𝑢 𝑓 (𝑡, 𝑢)
𝑁 ( 1) = ( 1 ). (22)
𝑢2 𝑓2 (𝑡, 𝑢)
2 2
× ((4𝐺𝑒𝛿𝐵𝑥 (𝑡 − 𝜏) 𝑦2 (𝑡 − 𝜏) + (4𝐺ℎ − 1) 𝑒 𝛿
Define
× 𝐵 (𝑡) 𝑥2 (𝑡 − 𝜏) 𝑦2 (𝑡 − 𝜏))
1 𝜔
𝑢 𝑢 ∫ 𝑢1 (𝑡) 𝑑𝑡
2 −1
× (4𝐺(1 + ℎ𝑒𝛿𝑥 (𝑡 − 𝜏)) ) ) 𝑃 ( 1) = 𝑄 ( 1) = ( 𝜔
1 𝜔
0 ),
𝑢2 𝑢2
∫ 𝑢2 (𝑡) 𝑑𝑡
− 𝑟2 (𝑡) 𝑦2 (𝑡) . 𝜔 0 (23)
(16) 𝑢
( 1 ) ∈ 𝑋 = 𝑌.
𝑢2
Make the change of variables
It is not difficult to show that 𝐿 is a Fredholm mapping of
𝑥 (𝑡) = exp {𝑢1 (𝑡)} , 𝑦 (𝑡) = exp {𝑢2 (𝑡)} ; (17)
index zero. Furthermore, the generalized inverse (to 𝐿) 𝐾𝑝 :
then system (16) can be rewritten as Im 𝐿 → Dom 𝐿∩Ker 𝑃 exists. Standard arguments show that
𝑁 is 𝐿-compact on Ω for any open bounded set Ω ⊂ 𝑋.
𝑢̇
1 (𝑡) = 𝑟1 (𝑡) − 𝑎 (𝑡) exp {𝑢1 (𝑡)} Now, we will search for two appropriate open bounded
subsets in order to apply the continuation theorem.
− ((4𝐺𝑒𝛿 exp {𝑢2 (𝑡)} + (4𝐺ℎ − 1) 𝑒2 𝛿2 Corresponding to the operator equation 𝐿𝑥 = 𝜆𝑁𝑥, 𝜆 ∈
(0, 1), we have
× exp {𝑢1 (𝑡) + 𝑢2 (𝑡)})
𝑢̇
1 (𝑡) = 𝜆𝑟1 (𝑡)
2 −1
× (4𝐺(1 + ℎ𝑒𝛿 exp{𝑢1 (𝑡)}) ) ) := 𝑓1 (𝑡, 𝑢) ,
− 𝜆 [𝑎 (𝑡) exp {𝑢1 (𝑡)}
𝑡
𝑢̇
2 (𝑡) = − 𝑟2 (𝑡) + 𝑏1 (𝑡 − 𝜏) exp {− ∫ 𝛽 (𝑠) 𝑑𝑠}
𝑡−𝜏 − ((4𝐺𝑒𝛿 exp {𝑢2 (𝑡)} + (4𝐺ℎ − 1) 𝑒2 𝛿2 (24)

× ((4𝐺𝑒𝛿𝐵 exp {𝑢1 (𝑡 − 𝜏)} + (4𝐺ℎ − 1) 𝑒2 𝛿2 𝐵 × exp {𝑢1 (𝑡) + 𝑢2 (𝑡)})

× exp {2𝑢1 (𝑡 − 𝜏)}) 2 −1

× (4𝐺(1 + ℎ𝑒𝛿 exp {𝑢1 (𝑡)}) ) )] ,
2 −1
× (4𝐺(1 + ℎ𝑒𝛿 exp {𝑢1 (𝑡 − 𝜏)}) ) ) 𝑢̇
2 (𝑡) = − 𝜆𝑟2 (𝑡)

𝑡
× exp {𝑢2 (𝑡 − 𝜏) − 𝑢2 (𝑡)} := 𝑓2 (𝑡, 𝑢) . + 𝜆 [𝑏1 (𝑡 − 𝜏) exp {− ∫ 𝛽 (𝑠) 𝑑𝑠}
(18) 𝑡−𝜏

Take × ((4𝐺𝑒𝛿𝐵 exp {𝑢1 (𝑡 − 𝜏)} + (4𝐺ℎ − 1) 𝑒2 𝛿2 𝐵

𝑋 = 𝑌 = {𝑥 = (𝑢1 , 𝑢2 )𝑇 ∈ 𝐶 (R, R2 ) | 𝑥 (𝑡 + 𝜔) = 𝑥 (𝑡)}
× exp {2𝑢1 (𝑡 − 𝜏)})
(19)
2 −1
and define × (4𝐺(1 + ℎ𝑒𝛿 exp {𝑢1 (𝑡 − 𝜏)}) ) )
󵄨 󵄨 󵄨 󵄨
‖𝑥‖ = max 󵄨󵄨󵄨𝑢1 (𝑡)󵄨󵄨󵄨 + max 󵄨󵄨󵄨𝑢2 (𝑡)󵄨󵄨󵄨 ,
𝑡∈[0,𝜔] 𝑡∈[0,𝜔]
(20) × exp {𝑢2 (𝑡 − 𝜏) − 𝑢2 (𝑡)}] .
𝑇
𝑥 = (𝑢1 , 𝑢2 ) ∈ 𝑋 or 𝑌; (25)
Abstract and Applied Analysis 5

Suppose 𝑥 = (𝑢1 (𝑡), 𝑢2 (𝑡))𝑇 ∈ 𝑋 is a solution of (24) and Similarly, it follows from (𝐴 1 ), (25), and (27) that
(25) for a certain 𝜆 ∈ (0, 1). Integrating (24), (25) over the
interval [0, 𝜔], we obtain
𝜔
𝜔 󵄨 󵄨󵄨
∫ 󵄨󵄨󵄨𝑢̇
2 (𝑡)󵄨󵄨 𝑑𝑡 < 2𝑟2 𝜔. (30)
∫ 𝑎 (𝑡) exp {𝑢1 (𝑡)} 𝑑𝑡 0
0
𝜔
+ ∫ ((4𝐺𝑒𝛿 exp {𝑢2 (𝑡)} + (4𝐺ℎ − 1) 𝑒2 𝛿2
0 Since (𝑢1 (𝑡), 𝑢2 (𝑡))𝑇 ∈ 𝑋, there exist 𝜉𝑖 , 𝜂𝑖 ∈ [0, 𝜔] such that
(26)
× exp {𝑢1 (𝑡) + 𝑢2 (𝑡)})
−1 𝑢𝑖 (𝜉𝑖 ) = min 𝑢𝑖 (𝑡) , 𝑢𝑖 (𝜂𝑖 ) = max 𝑢𝑖 (𝑡) , 𝑖 = 1, 2. (31)
× (4𝐺(1 + ℎ𝑒𝛿 exp{𝑢1 (𝑡)})2 ) ) 𝑑𝑡 𝑡∈[0,𝜔] 𝑡∈[0,𝜔]

= 𝑟1 𝜔,
Multiplying (25) by exp{𝑢2 (𝑡)} and integrating over [0, 𝜔], we
𝜔 𝑡
obtain
∫ 𝑏1 (𝑡 − 𝜏) exp {− ∫ 𝛽 (𝑠) 𝑑𝑠}
0 𝑡−𝜏

× exp {𝑢2 (𝑡 − 𝜏) − 𝑢2 (𝑡)} 𝜔

∫ 𝑟2 (𝑡) exp {𝑢2 (𝑡)} 𝑑𝑡
0
× ((4𝐺𝑒𝛿𝐵 exp {𝑢1 (𝑡 − 𝜏)} + (4𝐺ℎ − 1) 𝑒2 𝛿2 𝐵 𝜔 𝑡
(27) = ∫ [𝑏1 (𝑡 − 𝜏) exp {− ∫ 𝛽 (𝑠) 𝑑𝑠}
0 𝑡−𝜏
× exp {2𝑢1 (𝑡 − 𝜏)})
−1 × ((4𝐺𝑒𝛿𝐵 exp {𝑢1 (𝑡 − 𝜏)} + (4𝐺ℎ − 1) 𝑒2 𝛿2 𝐵
× (4𝐺(1 + ℎ𝑒𝛿 exp{𝑢1 (𝑡 − 𝜏)})2 ) ) 𝑑𝑡
× exp {2𝑢1 (𝑡 − 𝜏)})
= 𝑟2 𝜔.
2 −1
It follows from (𝐴 1 ), (24), and (26) that × (4𝐺(1 + ℎ𝑒𝛿 exp {𝑢1 (𝑡 − 𝜏)}) ) )
𝜔
󵄨 󵄨󵄨
∫ 󵄨󵄨󵄨𝑢̇
1 (𝑡)󵄨󵄨 𝑑𝑡
0 × exp {𝑢2 (𝑡 − 𝜏)}] 𝑑𝑡
𝜔󵄨󵄨
= 𝜆 ∫ 󵄨󵄨󵄨󵄨𝑟1 (𝑡) − 𝑎 (𝑡) exp {𝑢1 (𝑡)} 𝜔−𝜏 𝜎+𝜏
0 󵄨 =∫ [𝑏1 (𝜎) exp {− ∫ 𝛽 (𝑠) 𝑑𝑠}
−𝜏 𝜎

− ((4𝐺𝑒𝛿 exp {𝑢2 (𝑡)} + (4𝐺ℎ − 1) 𝑒2 𝛿2

× ((4𝐺𝑒𝛿𝐵 exp {𝑢1 (𝜎)} + (4𝐺ℎ − 1) 𝑒2 𝛿2 𝐵

× exp {𝑢1 (𝑡) + 𝑢2 (𝑡)}) × exp {2𝑢1 (𝜎)})

2 −1 󵄨
󵄨
× (4𝐺(1 + ℎ𝑒𝛿 exp {𝑢1 (𝑡)}) ) )󵄨󵄨󵄨󵄨 𝑑𝑡 × (4𝐺(1 + ℎ𝑒𝛿 exp{𝑢1 (𝜎)})2 ) )
−1
󵄨
(28)
𝜔 𝜔
< ∫ 𝑟1 (𝑡) 𝑑𝑡 + ∫ 𝑎 (𝑡) exp {𝑢1 (𝑡)} 𝑑𝑡 × exp {𝑢2 (𝜎)}] 𝑑𝑡
0 0
𝜔
𝜔 𝑡+𝜏
+ ∫ ((4𝐺𝑒𝛿 exp {𝑢2 (𝑡)} + (4𝐺ℎ − 1) 𝑒2 𝛿2 = ∫ [𝑏1 (𝑡) exp {− ∫ 𝛽 (𝑠) 𝑑𝑠}
0
0 𝑡

× exp {𝑢1 (𝑡) + 𝑢2 (𝑡)})

× ((4𝐺𝑒𝛿𝐵 exp {𝑢1 (𝑡)} + (4𝐺ℎ − 1) 𝑒2 𝛿2 𝐵
−1
× (4𝐺(1 + ℎ𝑒𝛿 exp{𝑢1 (𝑡)})2 ) ) 𝑑𝑡
× exp {2𝑢1 (𝑡)})
𝜔
= ∫ 𝑟1 (𝑡) 𝑑𝑡 + 𝑟1 𝜔 = 2𝑟1 𝜔; −1
0 × (4𝐺(1 + ℎ𝑒𝛿 exp{𝑢1 (𝑡)})2 ) )
that is,
𝜔 × exp {𝑢2 (𝑡)}] 𝑑𝑡;
󵄨 󵄨󵄨
∫ 󵄨󵄨󵄨𝑢̇
1 (𝑡)󵄨󵄨 𝑑𝑡 < 2𝑟1 𝜔. (29) (32)
0
6 Abstract and Applied Analysis

that is, which implies

𝜔 𝑟2𝐿 ≤ 𝑏1𝑀 exp {−𝜏𝛽𝐿 }

∫ 𝑟2 (𝑡) exp (𝑢2 (𝑡)) 𝑑𝑡
0 𝑒𝛿𝐵 exp {𝑢1 (𝜂1 )} (4𝐺ℎ − 1) 𝐵 (35)
𝜔 𝑡+𝜏 ×[ 2
+ ].
(1 + ℎ𝑒𝛿 exp {𝑢1 (𝜉1 )}) 4𝐺ℎ2
= ∫ [𝑏1 (𝑡) exp {− ∫ 𝛽 (𝑠) 𝑑𝑠}
0 𝑡
So
× ((4𝐺𝑒𝛿𝐵 exp {𝑢1 (𝑡)} + (4𝐺ℎ − 1) 𝑒2 𝛿2 𝐵 −1
𝑢1 (𝜂1 ) ≥ ln (([4𝐺ℎ2 𝑟2𝐿 (𝑏1𝑀) exp {𝜏𝛽𝐿 } − (4𝐺ℎ − 1) 𝐵]
× exp {2𝑢1 (𝑡)}) 2
× (1 + ℎ𝑒𝛿 exp {𝑢1 (𝜉1 )}) )
2 −1
× (4𝐺(1 + ℎ𝑒𝛿 exp{𝑢1 (𝑡)}) ) ) −1
× (4𝐺ℎ2 𝑒𝛿𝐵) ) .
× exp {𝑢2 (𝑡)}] 𝑑𝑡. (36)

(33) This, combined with (29), gives

𝜔
󵄨 󵄨󵄨
𝑢1 (𝑡) ≥ 𝑢1 (𝜂1 ) − ∫ 󵄨󵄨󵄨𝑢̇
1 (𝑡)󵄨󵄨 𝑑𝑡
It follows from (27), (33), and (𝐴 1 ); we see that 0
−1
𝜔 > ln (([4𝐺ℎ2 𝑟2𝐿 (𝑏1𝑀) exp {𝜏𝛽𝐿 } − (4𝐺ℎ − 1) 𝐵]
𝑟2𝐿 ∫ exp {𝑢2 (𝑡)} 𝑑𝑡
0 2
× (1 + ℎ𝑒𝛿 exp {𝑢1 (𝜉1 )}) )
𝜔
≤ ∫ 𝑟2 (𝑡) exp {𝑢2 (𝑡)} 𝑑𝑡 −1
0 × (4𝐺ℎ2 𝑒𝛿𝐵) ) − 2𝑟1 𝜔.
𝜔 𝑡+𝜏 (37)
= ∫ [𝑏1 (𝑡) exp {− ∫ 𝛽 (𝑠) 𝑑𝑠}
0 𝑡
In particular, we have
2 2
× ((4𝐺𝑒𝛿𝐵 exp {𝑢1 (𝑡)} + (4𝐺ℎ − 1) 𝑒 𝛿 𝐵 −1
𝑢1 (𝜉1 ) > ln (([4𝐺ℎ2 𝑟2𝐿 (𝑏1𝑀) exp {𝜏𝛽𝐿 } − (4𝐺ℎ − 1) 𝐵]
× exp {2𝑢1 (𝑡)}) 2
× (1 + ℎ𝑒𝛿 exp {𝑢1 (𝜉1 )}) )
2 −1
× (4𝐺(1 + ℎ𝑒𝛿 exp{𝑢1 (𝑡)}) ) ) −1
× (4𝐺ℎ2 𝑒𝛿𝐵) ) − 2𝑟1 𝜔,
× exp {𝑢2 (𝑡)}] 𝑑𝑡 (38)

(34) or
𝜔 𝑡+𝜏
≤ ∫ 𝑏1 (𝑡) exp {− ∫ 𝛽 (𝑠) 𝑑𝑠} −1
0 𝑡 [4𝐺ℎ2 𝑟2𝐿 (𝑏1𝑀) exp {𝜏𝛽𝐿 } − (4𝐺ℎ − 1) 𝐵] ℎ2 𝑒2 𝛿2
4𝐺𝑒𝛿𝐵 exp {𝑢1 (𝑡)} × exp {2𝑢1 (𝜉1 )}
×[
4𝐺(1 + ℎ𝑒𝛿 exp{𝑢1 (𝑡)})2
− [4𝐺ℎ2 𝑒𝛿𝐵 exp {2𝑟𝜔}
(4𝐺ℎ − 1) 𝑒2 𝛿2 𝐵 exp {2𝑢1 (𝑡)}
+ ]
4𝐺ℎ2 𝑒2 𝛿2 exp {2𝑢1 (𝑡)} −1
−2ℎ𝑒𝛿 (4𝐺ℎ2 𝑟2𝐿 (𝑏1𝑀) exp {𝜏𝛽𝐿 } − (4𝐺ℎ − 1) 𝐵)]
× exp {𝑢2 (𝑡)} 𝑑𝑡
× exp {𝑢1 (𝜉1 )}
≤ 𝑏1𝑀 exp {−𝜏𝛽𝐿 } −1
+ [4𝐺ℎ2 𝑟2𝐿 (𝑏1𝑀) exp {𝜏𝛽𝐿 } − (4𝐺ℎ − 1) 𝐵] < 0.
𝑒𝛿𝐵 exp {𝑢1 (𝜂1 )} (39)
×[
(1 + ℎ𝑒𝛿 exp{𝑢1 (𝜉1 )})2
In view of (𝐴 2 ), we have
(4𝐺ℎ − 1) 𝐵 𝜔
+ ] ∫ exp {𝑢2 (𝑡)} 𝑑𝑡,
4𝐺ℎ2 0 ln 𝑙− < 𝑢1 (𝜉1 ) < ln 𝑙+ . (40)
Abstract and Applied Analysis 7

Similarly, it follows from (33) that or

𝑏1𝐿 exp {−𝜏𝛽𝑀} 𝑟2𝑀ℎ2 𝑒2 𝛿2 exp {2𝑢1 (𝜂1 )}

4𝐺𝑒𝛿𝐵 exp {𝑢1 (𝜉1 )} + (4𝐺ℎ − 1) 𝑒2 𝛿2 𝐵 exp {2𝑢1 (𝜉1 )} − [𝑏1𝐿 𝑒𝛿𝐵 exp {−𝜏𝛽𝑀} exp {−2𝑟1 𝜔} − 2ℎ𝑒𝛿𝑟2𝑀] (47)
× 2
4𝐺(1 + ℎ𝑒𝛿 exp {𝑢1 (𝜂1 )})
× exp {𝑢1 (𝜂1 )} + 𝑟2𝑀 > 0.
𝜔
× ∫ exp {𝑢2 (𝑡)} 𝑑𝑡 It follows from (𝐴 2 ) that
0
𝜔 𝑡+𝜏 𝑢1 (𝜂1 ) < ln ℎ− or 𝑢1 (𝜂1 ) > ln ℎ+ . (48)
≤ ∫ 𝑏1 (𝑡) exp {− ∫ 𝛽 (𝑠) 𝑑𝑠}
0 𝑡
From (29) and (40), we find
2 2
4𝐺𝑒𝛿𝐵 exp {𝑢1 (𝑡)} + (4𝐺ℎ − 1) 𝑒 𝛿 𝐵 exp {2𝑢1 (𝑡)} 𝜔
× 󵄨 󵄨󵄨
4𝐺(1 + ℎ𝑒𝛿 exp {𝑢1 (𝑡)})
2 𝑢1 (𝑡) ≤ 𝑢1 (𝜉1 ) + ∫ 󵄨󵄨󵄨𝑢̇
1 (𝑡)󵄨󵄨 𝑑𝑡
0 (49)
× exp {𝑢2 (𝑡)} 𝑑𝑡 < ln 𝑙+ + 2𝑟1 𝜔 ≜ 𝐻11 .
𝜔
= ∫ 𝑟2 (𝑡) exp {𝑢2 (𝑡)} 𝑑𝑡 On the other hand, it follows from (𝐴 1 ), (26), and (49) that
0
𝜔 4𝐺𝑒𝛿 exp {𝑢2 (𝜉2 )}
𝜔
𝑟1 𝜔 ≥ ∫ 𝑑𝑡,
≤ 𝑟2𝑀 ∫ exp {𝑢2 (𝑡)} 𝑑𝑡, 0 4𝐺(1 + ℎ𝑒𝛿 exp {ln 𝑙+ + 2𝑟𝜔})
2 (50)
0
(41) 𝜔
𝑟1 𝜔 ≤ ∫ 𝑎 (𝑡) exp {ln 𝑙+ + 2𝑟1 𝜔} 𝑑𝑡
which implies 0
𝜔
𝑏1𝐿 exp {−𝜏𝛽𝑀} + ∫ 𝑒𝛿 exp {𝑢2 (𝜂2 )} 𝑑𝑡 (51)
0
4𝐺𝑒𝛿𝐵 exp {𝑢1 (𝜉1 )} + (4𝐺ℎ − 1) 𝑒2 𝛿2 𝐵 exp {2𝑢1 (𝜉1 )} 𝜔
× 𝑒𝛿 exp {𝑢2 (𝜂2 )}
4𝐺(1 + ℎ𝑒𝛿 exp {𝑢1 (𝜂1 )})
2 +∫ 𝑑𝑡.
0 2
≤ 𝑟2𝑀; It follows from (50) that
(42)
2
𝑟1 (1 + ℎ𝑒𝛿 exp {ln 𝑙+ + 2𝑟1 𝜔}) (52)
that is, 𝑢2 (𝜉2 ) ≤ ln .
𝑒𝛿
𝑒𝛿𝐵 exp {𝑢1 (𝜉1 )}
𝑏1𝐿 exp {−𝜏𝛽𝑀} 𝑑𝑠 2
≤ 𝑟2𝑀. (43) This, combined with (30), gives
(1 + ℎ𝑒𝛿 exp {𝑢1 (𝜂1 )})
𝜔
󵄨 󵄨󵄨
So 𝑢2 (𝑡) ≤ 𝑢2 (𝜉2 ) + ∫ 󵄨󵄨󵄨𝑢̇
2 (𝑡)󵄨󵄨 𝑑𝑡
0
2
𝑟2𝑀(1 + ℎ𝑒𝛿 exp {𝑢1 (𝜂1 )}) 2
𝑢1 (𝜉1 ) ≤ ln . (44) 𝑟1 (1 + ℎ𝑒𝛿 exp {ln 𝑙+ + 2𝑟1 𝜔}) (53)
𝑏1𝐿 𝑒𝛿𝐵 exp {−𝜏𝛽𝑀} < ln
𝑒𝛿
This, combined with (29), gives + 2𝑟2 𝜔 ≜ 𝐻21 .
𝜔
󵄨 󵄨󵄨
𝑢1 (𝑡) ≤ 𝑢1 (𝜉1 ) + ∫ 󵄨󵄨󵄨𝑢̇
1 (𝑡)󵄨󵄨 𝑑𝑡 Moreover, because of (𝐴 3 ), it follows from (51) that
0
2 2 (𝑟1 − 𝑎 exp {ln 𝑙+ + 2𝑟𝜔})
𝑟2𝑀(1 + ℎ𝑒𝛿 exp {𝑢1 (𝜂1 )}) (45) 𝑢2 (𝜂2 ) ≥ ln . (54)
< ln 3𝑒𝛿
𝑏1𝐿 𝑒𝛿𝐵 exp {−𝜏𝛽𝑀}
This, combined with (30) again, gives
+ 2𝑟1 𝜔.
𝜔
󵄨 󵄨󵄨
In particular, we have 𝑢2 (𝑡) ≥ 𝑢2 (𝜂2 ) − ∫ 󵄨󵄨󵄨𝑢̇
2 (𝑡)󵄨󵄨 𝑑𝑡
0
2
𝑟2𝑀(1 + ℎ𝑒𝛿 exp {𝑢1 (𝜂1 )}) 2 (𝑟1 − 𝑎 exp {ln 𝑙+ + 2𝑟𝜔}) (55)
𝑢1 (𝜂1 ) < ln > ln
𝑏1𝐿 𝑒𝛿𝐵 exp {−𝜏𝛽𝑀} (46) 3𝑒𝛿
+ 2𝑟1 𝜔, − 2𝑟2 𝜔 ≜ 𝐻22 .
8 Abstract and Applied Analysis

It follows from (53) and (55) that 𝐽 is taken as the identity mapping since Im 𝑄 = Ker 𝐿. So
󵄨 󵄨 󵄨 󵄨 far we have proved that Ω𝑖 satisfies all the assumptions in
max 𝑢2 (𝑡) < max {󵄨󵄨󵄨𝐻21 󵄨󵄨󵄨 , 󵄨󵄨󵄨𝐻22 󵄨󵄨󵄨} ≜ 𝐻2 . (56) Lemma 3. Hence, (16) has at least two 𝜔-periodic solutions
𝑡∈[0,𝜔]
𝑢∗ (𝑡) and 𝑢+ (𝑡) with 𝑢∗ ∈ Dom 𝐿 ∩ Ω1 and 𝑢+ ∈ Dom 𝐿 ∩ Ω2 .
Now, let us consider 𝑄𝑁𝑥 with Obviously, 𝑢∗ and 𝑢+ are different. Let 𝑥∗ (𝑡) = exp(𝑢1∗ (𝑡)),
𝑦2∗ (𝑡) = exp(𝑢2∗ (𝑡)) and 𝑥+ (𝑡) = exp(𝑢1+ (𝑡)), 𝑦2+ (𝑡) =
𝑄𝑁(𝑢1 , 𝑢2 )𝑇 exp(𝑢2+ (𝑡)). Then, by (18), (𝑥∗ (𝑡), 𝑦2∗ (𝑡)) and (𝑥+ (𝑡), 𝑦2+ (𝑡)) are
two different positive 𝜔-periodic solutions of (4). By the
= (𝑟1 𝜔 − 𝑎𝜔 exp {𝑢1 } − 𝑤 periodicity of the coefficients of system (4), it is not difficult
to verify that

4𝐺𝑒𝛿 exp {𝑢2 } + (4𝐺ℎ − 1) 𝑒2 𝛿2 exp {𝑢1 + 𝑢2 } 𝑔∗ (𝑡) = − 𝛽 (𝑡) 𝑦1 (𝑡)

× 2
,
4𝐺(1 + ℎ𝑒𝛿 exp {𝑢1 })
+ ((𝑏1 (𝑡) 4𝐺𝑒𝛿𝐵𝑥∗ (𝑡) 𝑦2∗ (𝑡)
− 𝑟2 𝜔 + 𝑏𝜔
2
+ (4𝐺ℎ − 1) 𝑒2 𝛿2 𝐵(𝑥∗ ) (𝑡) 𝑦2∗ (𝑡))
𝑇
4𝐺𝑒𝛿𝐵 exp{𝑢1 } + (4𝐺ℎ − 1)𝑒2 𝛿2 𝐵 exp{2𝑢1 }
× ) . 2 −1
4𝐺(1 + ℎ𝑒𝛿 exp {𝑢1 })
2 × (4𝐺(1 + ℎ𝑒𝛿𝑥∗ (𝑡)) ) )
(57) 𝑡
− 𝑏1 (𝑡 − 𝜏) exp (− ∫ 𝛽 (𝑠) 𝑑𝑠)
In view of (𝐴 1 ), (𝐴 2 ), and (𝐴 3 ), we can show that the equation 𝑡−𝜏
𝑄𝑁(𝑢1 , 𝑢2 )𝑇 = 0 has two distinct solutions
× ((4𝐺𝑒𝛿𝐵𝑥∗ (𝑡 − 𝜏) 𝑦2∗ (𝑡 − 𝜏)
2
4𝐺 (𝑟1 − 𝑎𝑢− ) (1 + ℎ𝑒𝛿𝑢− )
𝑢̃ = (ln 𝑢− , ln ), 2
4𝐺𝑒𝛿 + (4𝐺ℎ − 1) 𝑒2 𝛿2 𝑢− + (4𝐺ℎ − 1) 𝑒2 𝛿2 𝐵(𝑥∗ ) (𝑡 − 𝜏) 𝑦2∗ (𝑡 − 𝜏))
(58)
2 −1
4𝐺 (𝑟1 − 𝑎𝑢+ ) (1 + ℎ𝑒𝛿𝑢+ ) × (4𝐺(1 + ℎ𝑒𝛿𝑥∗ (𝑡 − 𝜏))2 ) )
𝑢̂ = (ln 𝑢+ , ln ).
4𝐺𝑒𝛿 + (4𝐺ℎ − 1) 𝑒2 𝛿2 𝑢+ (61)
Choose 𝐶 > 0 such that is also 𝜔-periodic. Then, from Lemma 4, we know that
󵄨󵄨 2󵄨
󵄨󵄨 4𝐺 (𝑟1 − 𝑎𝑢− ) (1 + ℎ𝑒𝛿𝑢− ) 󵄨󵄨󵄨 𝑑𝑦1 (𝑡)
𝐶 > max {󵄨󵄨󵄨ln 󵄨󵄨 , = −𝛽 (𝑡) 𝑦1 (𝑡) + 𝑔∗ (𝑡) (62)
󵄨󵄨
󵄨 4𝐺𝑒𝛿 + (4𝐺ℎ − 1) 𝑒2 𝛿2 𝑢− 󵄨󵄨󵄨󵄨 𝑑𝑡
(59)
󵄨󵄨󵄨 4𝐺 (𝑟 − 𝑎𝑢 ) (1 + ℎ𝑒𝛿𝑢 )2 󵄨󵄨󵄨 has a unique 𝜔-periodic solution denoted by 𝑦1∗ (𝑡). And
󵄨󵄨 1 + + 󵄨󵄨
󵄨󵄨ln 2 𝛿2 𝑢 󵄨󵄨󵄨
}.
󵄨󵄨 4𝐺𝑒𝛿 + (4𝐺ℎ − 1) 𝑒 + 󵄨󵄨 𝑑𝑦2 (𝑡)
󵄨 = −𝛽 (𝑡) 𝑦1 (𝑡) + 𝑔+ (𝑡) (63)
𝑑𝑡
We define two open bounded subsets. Let
has a unique 𝜔-periodic solution denoted by 𝑦1+ (𝑡). Therefore,
𝑇
Ω1 = {𝑥 = (𝑢1 , 𝑢2 ) ∈ 𝑋 | 𝑢1 (𝑡) ∈ (ln 𝑙− , ln ℎ− ) , (𝑥∗ (𝑡), 𝑦1∗ (𝑡), 𝑦2∗ (𝑡)) and (𝑥+ (𝑡), 𝑦1+ (𝑡), 𝑦2+ (𝑡)) are two different
𝜔-periodic solutions of system (3). This completes the proof
of Theorem 5.
󵄨 󵄨
max 󵄨󵄨󵄨𝑢2 (𝑡)󵄨󵄨󵄨 < 𝐻2 + 𝐶} ,
𝑡∈[0,𝜔]
Conflict of Interests
𝑇
Ω2 = {𝑥 = (𝑢1 , 𝑢2 ) ∈ 𝑋 | min 𝑢1 (𝑡) ∈ (ln 𝑙− , ln 𝑙+ ) , The authors declare that there is no conflict of interests
𝑡∈[0,𝜔]
regarding the publication of this paper.
󵄨 󵄨
max 𝑢1 (𝑡) ∈ (ln ℎ+ , 𝐻11 ) , max 󵄨󵄨󵄨𝑢2 (𝑡)󵄨󵄨󵄨 < 𝐻2 + 𝐶} .
𝑡∈[0,𝜔] 𝑡∈[0,𝜔] Funding
(60)
This work is supported by JB12254.
Then both Ω1 and Ω2 are bounded open subsets of 𝑋. It
follows from (16) and (59) that 𝑢̃ ∈ Ω1 and 𝑢̂ ∈ Ω2 . With the
help of (16), (40), (48), (49), (56), and (59), it is easy to see that References
Ω1 ∩ Ω2 = 𝜙, and Ω𝑖 satisfies the requirement (a) in Lemma 3 [1] Y. L. Song, Y. Peng, and J. J. Wei, “Bifurcations for a predator-
for 𝑖 = 1, 2. Moreover, 𝑄𝑁𝑥 ≠ 0 for 𝑥 ∈ 𝜕Ω∩Ker 𝐿 = 𝜕Ω∩R2 . prey system with two delays,” Journal of Mathematical Analysis
A direct computation gives deg{𝐽𝑄𝑁, Ω𝑖 ∩Ker 𝐿, 0} ≠ 0. Here, and Applications, vol. 337, no. 1, pp. 466–479, 2008.
Abstract and Applied Analysis 9

[2] S. L. Yuan and Y. L. Song, “Stability and Hopf bifurcations

in a delayed Leslie-Gower predator-prey system,” Journal of
Mathematical Analysis and Applications, vol. 355, no. 1, pp. 82–
100, 2009.
[3] Y. L. Song and S. L. Yuan, “Bifurcation analysis for a regulated
logistic growth model,” Applied Mathematical Modelling, vol. 31,
no. 9, pp. 1729–1738, 2007.
[4] Y. H. Xia and M. Han, “New conditions on the existence
and stability of periodic solution in Lotka-Volterra’s population
system,” SIAM Journal on Applied Mathematics, vol. 69, no. 6,
pp. 1580–1597, 2009.
[5] T. Zhang, J. Liu, and Z. Teng, “Existence of positive periodic
solutions of an SEIR model with periodic coefficients,” Applica-
tions of Mathematics, vol. 57, no. 6, pp. 601–616, 2012.
[6] Y. H. Xia, “Global asymptotic stability of an almost periodic
nonlinear ecological model,” Communications in Nonlinear
Science and Numerical Simulation, vol. 16, no. 11, pp. 4451–4478,
2011.
[7] Y. H. Xia, “Global analysis of an impulsive delayed Lotka-
Volterra competition system,” Communications in Nonlinear
Science and Numerical Simulation, vol. 16, no. 3, pp. 1597–1616,
2011.
[8] Y. Xia, X. Yuan, K. I. Kou, and P. J. Y. Wong, “Existence and
uniqueness of solution for perturbed nonautonomous systems
with nonuniform exponential dichotomy,” Abstract and Applied
Analysis, vol. 2014, Article ID 725098, 10 pages.
[9] Y. H. Xia, “Periodic solution of certain nonlinear differential
equations: via topological degree theory and matrix spectral
theory,” International Journal of Bifurcation and Chaos in
Applied Sciences and Engineering, vol. 22, no. 8, Article ID
1250196, 17 pages, 2012.
[10] L. Yin and Z. Zhang, “Existence of a positive solution for a first-
order p-Laplacian BVP with impulsive on time scales,” Journal
of Applied Analysis and Computation, vol. 2, no. 1, pp. 103–109,
2012.
[11] X. Gu, H. Wang, P. Wong, and Y. H. Xia, “Existence and stability
of periodic solution to delayed nonlinear differential equations,”
Abstract and Applied Analysis, vol. 2014, Article ID 156948, 12
pages, 2014.
[12] W. Ding and M. Han, “Dynamic of a non-autonomous predator-
prey system with infinite delay and diffusion,” Computers &
Mathematics with Applications, vol. 56, no. 5, pp. 1335–1350,
2008.
[13] Y. F. Gao, P. J. Y. Wong, Y. H. Xia, and X. Yuan, “Multiple
periodic solutions of a nonautonomous plant-hare model,”
Abstract and Applied Analysis, vol. 2014, Article ID 130856, 7
pages, 2014.
[14] W. G. Aiello and H. I. Freedman, “A time-delay model of single-
species growth with stage structure,” Mathematical Biosciences,
vol. 101, no. 2, pp. 139–153, 1990.
[15] W. Wang and L. Chen, “A predator-prey system with stage-
structure for predator,” Computers & Mathematics with Appli-
cations, vol. 33, no. 8, pp. 83–91, 1997.
[16] Y. H. Xia, J. Cao, and S. S. Cheng, “Multiple periodic solutions
of a delayed stage-structured predator-prey model with non-
monotone functional responses,” Applied Mathematical Mod-
elling, vol. 31, no. 9, pp. 1947–1959, 2007.
[17] R. E. Gaines and J. L. Mawhin, Coincidence Degree, and
Nonlinear Differential Equations, Lecture Notes in Mathematics,
Springer, Berlin, Germany, 1977.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 371639, 11 pages
http://dx.doi.org/10.1155/2014/371639

Research Article
Stability Conditions of Second Order Integrodifferential
Equations with Variable Delay

Dingheng Pi
School of Mathematical Sciences, Huaqiao University, Quanzhou, Fujian 362021, China

Correspondence should be addressed to Dingheng Pi; pidh@hqu.edu.cn

Received 7 March 2014; Accepted 23 April 2014; Published 7 May 2014

Academic Editor: Yonghui Xia

Copyright © 2014 Dingheng Pi. This is an open access article distributed under the Creative Commons Attribution License, which
permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
𝑡
We investigate integrodifferential functional differential equations 𝑥̈+ 𝑓(𝑡, 𝑥, 𝑥)̇𝑥̇+ ∫𝑡−𝑟(𝑡) 𝑎(𝑡, 𝑠)𝑔(𝑥(𝑠))𝑑𝑠 = 0 with variable
delay. By using the fixed point theory, we obtain conditions which ensure that the zero solution of this equation is stable
under an exponentially weighted metric. Then we establish necessary and sufficient conditions ensuring that the zero solution
is asymptotically stable. We will give an example to apply our results.

1. Introduction where 𝑟 was a positive constant. He gave conditions on

functions 𝑎 and 𝑔 to ensure that the zero solution was
Functional differential equations have many applications in asymptotically stable by applying fixed point theorem; see [7].
control theory, biology, and so on. The stability of the solution Becker and Burton studied the following differential
of functional differential equations has been a hot issue for equation:
researchers for many years. It is well known that Lyapunov’s
𝑡
direct method has been widely applied to study the stability
problems for a long time; see, for example, [1, 2]. Recently, 𝑥̇= − ∫ 𝑎 (𝑡, 𝑠) 𝑔 (𝑥 (𝑠)) 𝑑𝑠 (4)
𝑡−𝑟(𝑡)
many authors have applied the fixed points theory to study
the stability of solution of integral equations and several and equation
functional differential equations with variable delays; see, for
example, [3–5] and the references therein. 𝑥̇= −𝑎 (𝑡) 𝑔 (𝑥 (𝑡 − 𝑟 (𝑡))) , (5)
In [6], Levin and Nohel investigated the behavior of for 𝑡 ≥ 0, where 𝑟(𝑡) : [0, ∞) → [0, ∞), 𝑎(𝑡, 𝑠) : [0, ∞) ×
solution of a nonlinear equation [−𝑟(0), ∞) → 𝑅, 𝑔 : 𝑅 → 𝑅, 𝑎(𝑡) : [0, +∞) → [0, +∞) are
1 𝑡 continuous. In addition, they assumed that
𝑥̇
(𝑡) = − ∫ (𝐿 − (𝑡 − 𝜏)) 𝑔 (𝑥 (𝜏 (𝑡))) 𝑑𝜏, (0 ≤ 𝑡 < ∞) ,
𝐿 𝑡−𝐿 (𝐴 1 ) 𝑟(𝑡) is differentiable;
(1)
(𝐴 2 ) the function 𝑡 − 𝑟(𝑡) : [0, ∞) → [−𝑟(0), ∞) is strictly
where 𝐿 is a constant. This equation was equivalent to increasing;
(𝐴 3 ) 𝑡 − 𝑟(𝑡) → ∞ as 𝑡 → ∞.
1 𝑡
𝑥̈(𝑡) + 𝑔 (𝑥 (𝑡)) = ∫ 𝑔 (𝑥 (𝜏)) 𝑑𝜏, (0 ≤ 𝑡 < ∞) . (2) They obtained sufficient conditions ensuring that the zero
𝐿 𝑡−𝐿
solution was asymptotically stable by changing the supremum
Burton studied stability of a nonconvolution equation metric to an exponentially weighted metric. Moreover, they
hoped to relax condition (𝐴 2 ); see, for example, [8].
𝑡 Jin and Luo succeeded in eliminating condition (𝐴 2 ) in
𝑥̇= − ∫ 𝑎 (𝑡, 𝑠) 𝑔 (𝑥 (𝑠)) 𝑑𝑠, (3) their work; they did not need the condition that 𝑡 − 𝑟(𝑟)
𝑡−𝑟
2 Abstract and Applied Analysis

was invertible. Moreover, they established necessary and (iv) There exist continuous functions 𝐹 : R × R → [0, ∞)
sufficient conditions that could ensure that the zero solution and 𝑐(𝑡) : [0, ∞) → [0, ∞) such that ∀𝑡 ≥ 0, 𝑥 ∈ R,
of this equation was asymptotically stable; see, for example, 𝑦 ∈ R, 𝑓(𝑡, 𝑥, 𝑦) ≤ 𝐹(𝑥, 𝑦)𝑐(𝑡). The function 𝑔󸀠 (𝑥) is
[9]. Dung [10] studied linear case of this equation and gave continuous on [−𝑙, 𝑙], 𝑔󸀠 (0) ≠0.
new stability results by using a new expression of the solution.
Other results on fixed points and stability properties in The following statements hold.
equations with variable delays can be found in [3, 11] and the
references therein. (a) If, for each 𝛾 > 0,
Levin and Nohel [12] studied the global asymptotic ∞ 𝑤+𝑠
stability of a class of nonlinear systems ∬ 𝑒− ∫𝑠 𝛾𝑐(V)
𝑏 (𝑠) 𝑑𝑤 𝑑𝑠 = ∞, (11)
0
𝑥̈(𝑡) + ℎ (𝑡, 𝑥, 𝑥)̇𝑥̇
+ 𝑓 (𝑥) = 𝑒 (𝑡) . (6)
then the zero solution of (8) is asymptotically stable.
Burton [13] used the fixed points theory to study the stability (b) If the zero solution of (8) is asymptotically stable, then
problems of some second order functional differential equa-
tions. He considered the equation ∞ 𝑤+𝑠
∬ 𝑒− ∫𝑠 𝑎(V)𝑑V
𝑏 (𝑠) 𝑑𝑤 𝑑𝑠 = ∞. (12)
𝑥̈+ 𝑓 (𝑡, 𝑥, 𝑥)̇𝑥̇
+ 𝑏 (𝑡) 𝑔 (𝑥 (𝑡 − 𝐿)) = 0, (7) 0

where 𝐿 is a positive constant. He obtained sufficient condi- This theorem failed to offer a necessary and sufficient
̇ →
tions under which each solution 𝑥(𝑡) satisfied (𝑥(𝑡), 𝑥(𝑡)) condition which ensures that the zero solution was asymptot-
0 via the fixed point theorem. ically stable. In this paper, we will establish a necessary and
We generalized the above equation to an equation with a sufficient condition which ensures that the zero solution of
variable delay [11] related equation is asymptotically stable.
In this paper we consider equation
𝑥̈+ 𝑓 (𝑡, 𝑥, 𝑥)̇𝑥̇
+ 𝑏 (𝑡) 𝑔 (𝑥 (𝑡 − 𝜏 (𝑡))) = 0 (8)
𝑡
and obtained some results on asymptotic stability of the zero 𝑥̈+ 𝑓 (𝑡, 𝑥, 𝑥)̇𝑥̇
+∫ 𝑎 (𝑡, 𝑠) 𝑔 (𝑥 (𝑠)) 𝑑𝑠 = 0 (13)
𝑡−𝑟(𝑡)
solution. Before we introduce our new results we recall the
main results in [11]. There are basic assumptions on the delay for 𝑡 ≥ 0, where 𝑟(𝑡) : [0, ∞) → [0, ∞). 𝑎(𝑡, 𝑠) : [0, ∞) ×
function 𝜏(𝑡). [−𝑟(0), ∞) → 𝑅, 𝑔 : 𝑅 → 𝑅, 𝑎(𝑡) : 𝑅+ → 𝑅, 𝑓 : 𝑅×𝑅×𝑅 →
(A) 𝑡 − 𝜏(𝑡) is strictly increasing and lim𝑡 → ∞ 𝑡 − 𝜏(𝑡) = ∞. 𝑅+ are all continuous, where 𝑅+ = [0, +∞). We assume that
The inverse of 𝑡 − 𝜏(𝑡) exists and denotes it by 𝑝(𝑡). 𝑡 − 𝑟(𝑡) → ∞ as 𝑡 → ∞.
Moreover, 0 ≤ 𝑏(𝑡) ≤ 𝑀 for some constant 𝑀 > 0. For each 𝑡0 ≥ 0, define 𝑚(𝑡0 ) = inf{𝑠 − 𝑟(𝑠) : 𝑠 ≥ 𝑡0 }. Set
𝐶(𝑡0 ) = 𝐶([𝑚(𝑡0 ), 𝑡0 ], 𝑅) with the continuous function norm
The main results in [11] can be stated as follows. ‖ ⋅ ‖, where ‖𝜓‖ = sup{|𝜓(𝑠)| : 𝑚(𝑡0 ) ≤ 𝑠 ≤ 𝑡0 }. It will cause no
confusion even though we use ‖𝜙‖ to express the supremum
Theorem 1. Suppose (A) and the following conditions. on [𝑚(𝑡0 ), ∞) later. It is well known that in [2], for a given
continuous function 𝜙, there exists a solution of (13) on an
(i) There exists a constant 𝑙 > 0 such that 𝑔(𝑥) satisfies interval [𝑡0 , 𝑇); if the solution remains bounded, then 𝑇 = ∞.
the Lipschitz condition on [−𝑙, 𝑙]. The function 𝑔(𝑥) is We denote by (𝑥(𝑡), 𝑦(𝑡)) the solution (𝑥(𝑡, 𝑡0 , 𝜙), 𝑦(𝑡, 𝑡0 , 𝜙)).
odd and is strictly increasing on [−𝑙, 𝑙], and 𝑥 − 𝑔(𝑥) is We will give a necessary and sufficient condition ensuring
nondecreasing on [0, 𝑙]. that the zero solution of this equation is asymptotically
(ii) There exist an 𝛼 ∈ (0, 1) and a continuous function stable. To our knowledge, there are few results about its
𝑎(𝑡) : [0, ∞) → [0, ∞) such that 𝑓(𝑡, 𝑥, 𝑦) ≥ 𝑎(𝑡) for stability. From the solution (𝑥(𝑡), 𝑦(𝑡)), we denote 𝐴(𝑡) :=
∞
𝑡 ≥ 0, 𝑥 ∈ R, 𝑦 ∈ R, ∫0 𝑎(𝑡)𝑑𝑡 = ∞, and 𝑓(𝑡, 𝑥(𝑡), 𝑦(𝑡)). We can write (13) as

𝑝(𝑡) ∞ 𝑤+𝑠
𝑥̇= 𝑦,
2sup ∫ ∫ 𝑒− ∫𝑠 𝑎(V)𝑑V
𝑏 (𝑠) 𝑑𝑤 𝑑𝑠 𝑡 (14)
𝑡≥0 𝑡 0
(9) 𝑦̇= − 𝐴 (𝑡) 𝑦 − ∫ 𝑎 (𝑡, 𝑠) 𝑔 (𝑥 (𝑠)) 𝑑𝑠.
𝑡 ∞ 𝑤+𝑠 𝑡−𝑟(𝑡)
+ 2sup ∫ ∫ 𝑒− ∫𝑠 𝑎(V)𝑑V
𝑏 (𝑠) 𝑑𝑤 𝑑𝑠 ≤ 𝛼.
𝑡≥0 0 𝑡−𝑠 For each 𝑡0 ≥ 0, let 𝑚(𝑡0 ) = inf{𝑠 − 𝑟(𝑠) : 𝑠 ≥ 𝑡0 }, 𝐶(𝑡0 ) =
𝐶([𝑚(𝑡0 ), 𝑡0 ], 𝑅) with the continuous function norm ‖ ⋅ ‖,
(iii) There exist constants 𝑎0 > 0 and 𝑄 > 0 such that, for where ‖𝜓‖ = sup{|𝜓(𝑠)| : 𝑚(𝑡0 ) ≤ 𝑠 ≤ 𝑡0 }.
each 𝑡 ≥ 0, if 𝐽 ≥ 𝑄, then This paper is organized as follows. In the next section
we will state our main results. Their proofs will be given in
𝑡+𝐽 Sections 3 and 4. We will give an example to apply our results
∫ 𝑎 (V) 𝑑V ≥ 𝑎0 𝐽. (10)
in Section 5.
𝑡
Abstract and Applied Analysis 3

𝑡
2. Statement of Main Results For ∀𝑡 ≥ 0, ∫𝑡−𝑟(𝑡) |𝑎(𝑡, V)|𝑑V is bounded and for 𝑡 ≥ 0,
𝑥 ∈ 𝑅, 𝑦 ∈ 𝑅,
We make the following basic assumptions on the delay
function 𝑟(𝑡) of (13). 𝑝(𝑡) ∞ 𝑤+𝑠
2𝐿 ∫ ∫ 𝑒− ∫𝑠 𝑎(V)𝑑V
𝐺 (𝑠, 𝑠) 𝑑𝑤 𝑑𝑠
𝑡 0
(A1 ) lim𝑡 → ∞ 𝑡 − 𝑟(𝑡) = ∞. 𝑝(𝑡) is the inverse of 𝑡 − 𝑟(𝑡).
𝑝(𝑡) 𝑝(𝑠) 𝑡 ∞ 𝑤+𝑠
𝐺(𝑡, 𝑡) = ∫𝑡 𝑎(𝑢, 𝑡)𝑑𝑢 and 𝐺(𝑡, 𝑠) = ∫𝑡 𝑎(𝑢, 𝑠)𝑑𝑢. + 2𝐿 ∫ ∫ 𝑒− ∫𝑠 𝑎(V)𝑑V
𝐺 (𝑠, 𝑠) 𝑑𝑤 𝑑𝑠
There exists a constant 𝑀 > 0 such that |𝐺(𝑡, 𝑡)| ≤ 𝑀. 0 𝑡−𝑠
The following are our main results. 𝑡 𝑡 𝑢
+ 2𝐿 ∫ 𝑒− ∫𝑢 ℎ(V)𝑑V (∫ |𝑎 (𝑢, V)| 𝑑V) 𝑑𝑢
Theorem 2. Assume that (A1 ) holds and the following condi- 0 𝑢−𝑟(𝑢)

tions hold. 𝑝(𝑡) 𝑡 ∞ 𝑤+𝑢

+ 𝐿∫ 𝑒− ∫𝑤−𝑟(𝑤) ℎ(V)𝑑V ∫ 𝑒− ∫𝑤 𝑎(𝑠)𝑑𝑠
|𝐺 (𝑢, 𝑢)| 𝑑𝑢 𝑑𝑤
0 0
(i) There exists a constant 𝑙 > 0 such that 𝑔(𝑥) satisfies Lip-
𝑡 𝑡
schitz condition on [−𝑙, 𝑙]. 𝑔(𝑥) is odd and it is strictly + ∫ 𝑒− ∫𝑢 ℎ(V)𝑑V |ℎ (𝑢) − 𝐻 (𝑢)| 𝑑𝑢 ≤ 𝛼.
increasing on [−𝑙, 𝑙], and 𝑥 − 𝑔(𝑥) is nondecreasing on 0
[0, 𝑙]. (17)

(ii) There exist an 𝛼 ∈ (0, 1) and a continuous function (iii) There exist constants 𝑎0 > 0 and 𝑄 > 0 such that, for
𝑎(𝑡) : [0, ∞) → [0, ∞) such that 𝑓(𝑡, 𝑥, 𝑦) ≥ 𝑎(𝑡) each 𝑡 ≥ 0, if 𝐽 ≥ 𝑄, then
𝑡
for 𝑡 ≥ 0. For ∀𝑡 ≥ 0, ∫𝑡−𝑟(𝑡) |𝐺(𝑡, V)|𝑑V is increasing
𝑡 𝑡+𝐽
with respect to 𝑡, and ∫𝑡−𝑟(𝑡) |𝑎(𝑡, V)|𝑑V is bounded and ∫ 𝑎 (𝑠) 𝑑𝑠 ≥ 𝑎0 𝐽. (18)
for 𝑡 ≥ 0 and for 𝑡 ≥ 0, 𝑥 ∈ 𝑅, 𝑦 ∈ 𝑅, 𝑡

Then the zero solution of (13) is asymptotically stable if

𝑝(𝑡) ∞ 𝑤+𝑠 and only if
− ∫𝑠 𝑎(V)𝑑V
2∫ ∫ 𝑒 𝐺 (𝑠, 𝑠) 𝑑𝑤 𝑑𝑠
𝑡 0 (iv)
𝑡 ∞ 𝑤+𝑠
− ∫𝑠 𝑎(V)𝑑V 𝑡
+ 2∫ ∫ 𝑒 𝐺 (𝑠, 𝑠) 𝑑𝑤 𝑑𝑠 (15)
0 𝑡−𝑠 ∫ ℎ (𝑠) 𝑑𝑠 󳨀→ ∞ as 𝑡 󳨀→ ∞. (19)
0
𝑡 𝑡 𝑠
+ 2 ∫ 𝑒− ∫𝑠 𝑎(V)𝑑V ∫ |𝐺 (𝑠, V)| 𝑑V 𝑑𝑠 ≤ 𝛼.
0 𝑠−𝑟(𝑠) Remark 4. We give some new notations:
𝑡
𝑒− ∫𝑢 𝐴(𝑠)𝑑𝑠 𝐺 (𝑢, 𝑢) ≜ 𝐶 (𝑡, 𝑢) ,
(iii) There exist constants 𝑎0 > 0 and 𝑄 > 0 such that, for ∞
each 𝑡 ≥ 0, if 𝐽 ≥ 𝑄, then ∫ 𝐶 (𝑢 + 𝑡 − 𝑡0 , 𝑡) 𝑑𝑢 ≜ 𝐷 (𝑡) ≥ 0,
𝑡0

𝑡+𝐽 𝑡
− ∫𝑡 𝐴(𝑠)𝑑𝑠 𝐷 (𝑡) ̃ (𝑡) ,
∫ 𝑎 (𝑠) 𝑑𝑠 ≥ 𝑎0 𝐽. (16) 𝑥̇
(𝑡0 ) 𝑒 0 ≜ 𝐵 (𝑡) , ≜𝐷 (20)
𝑡 1 − 𝑟̇
(𝑡)
̃ (𝑝 (𝑡)) ≜ 𝐻 (𝑡) ,
𝐷
Then the zero solution of (13) is stable.
∞

In addition, we have the following. ∫ 𝐶 (𝑢 + 𝑠 − 𝑡0 , 𝑠) 𝑑𝑢 ≜ 𝐸 (𝑡, 𝑠) ≥ 0.

𝑡0 +𝑡−𝑠

Theorem 3. Assume that 𝑙𝑖𝑚𝑡 → ∞ 𝑡−𝑟(𝑡) = ∞ and there exists

a function ℎ(𝑡) ∈ 𝐶(𝑅+ , 𝑅+ ) such that for 𝑡 ≥ 0 the following 3. Proof of Theorem 2
conditions hold.
In this section, we will prove Theorem 2 by applying the fixed
point theory. We will give the expression of the solution of the
(i) There exists constant 𝑙 > 0 such that 𝑔(𝑥) satisfies related equation. The following result can be found in [8].
Lipschitz condition on [−𝑙, 𝑙]. 𝐿 is the Lipschitz constant.
𝑔(𝑥) is odd and it is strictly increasing on [−𝑙, 𝑙], and Lemma 5. Let the function 𝑝 : [−𝑟(0), ∞) → [0, ∞) denote
𝑥 − 𝑔(𝑥) is nondecreasing on [0, 𝑙]. the inverse of 𝑡 − 𝑟(𝑡). Then
𝑡
(ii) There exist a constant 𝛼 ∈ (0, 1) and a continuous
𝑥̇
(𝑡) = − ∫ 𝑎 (𝑡, 𝑠) 𝑔 (𝑥 (𝑠)) 𝑑𝑠 (21)
function 𝑎(𝑡) : 𝑅+ → 𝑅+ such that 𝑓(𝑡, 𝑥, 𝑦) ≥ 𝑎(𝑡). 𝑡−𝑟(𝑡)
4 Abstract and Applied Analysis

is equivalent to Equation (24) can be written as

𝑥̇
(𝑡) = 𝐵 (𝑡)
𝑑 𝑡
𝑥̇
(𝑡) = −𝐺 (𝑡, 𝑡) 𝑔 (𝑥 (𝑡)) + ∫ 𝐺 (𝑡, 𝑠) 𝑔 (𝑥 (𝑠)) 𝑑𝑠. 𝑡
𝑑𝑡 𝑡−𝑟(𝑡) 𝑡

(22) − ∫ 𝑒− ∫𝑠 𝐴(V)𝑑V (𝐺 (𝑠, 𝑠) 𝑔 (𝑥 (𝑠))

𝑡0

Lemma 6. Let 𝜓 : [𝑚(𝑡0 ), 𝑡0 ] → 𝑅 be a given continuous 𝑑 𝑠

− ∫ 𝐺 (𝑠, V) 𝑔 (𝑥 (V)) 𝑑V) 𝑑𝑠.
function; if (𝑥(𝑡), 𝑦(𝑡)) is the solution of (13) on [𝑡0 , 𝑇1 ) 𝑑𝑠 𝑠−𝑟(𝑠)
̇0 ), then
satisfying 𝑥(𝑡) = 𝜓(𝑡), 𝑡 ∈ [𝑚(𝑡0 ), 𝑡0 ], and 𝑦(𝑡0 ) = 𝑥(𝑡 (25)
𝑥(𝑡) is the solution of the following integral equation:
Therefore,
𝑡
𝑥 (𝑡) 𝑥̇
(𝑡) = 𝐵 (𝑡) − ∫ 𝐶 (𝑡, 𝑠) 𝑔 (𝑥 (𝑠)) 𝑑𝑠
𝑡0
𝑡 𝑡 𝑡
− ∫𝑡 𝐻(𝑠)𝑑𝑠 − ∫𝑢 𝐻(𝑠)𝑑𝑠
= 𝜓 (𝑡0 ) 𝑒 0 +∫ 𝑒 𝐵 (𝑢) 𝑑𝑢 𝑡 𝑡
𝑑 𝑠
𝑡0 + ∫ 𝑒− ∫𝑠 𝐴(V)𝑑V ( ∫ 𝐺 (𝑠, V) 𝑔 (𝑥 (V)) 𝑑V) 𝑑𝑠.
𝑡0 𝑑𝑠 𝑠−𝑟(𝑠)
𝑡 𝑡 (26)
− ∫𝑢 𝐻(𝑠)𝑑𝑠
+∫ 𝑒 𝐻 (𝑢) [𝑥 (𝑢) − 𝑔 (𝑥 (𝑢))] 𝑑𝑢
𝑡0 Since |𝐺(𝑡, 𝑡)| ≤ 𝑀, we have
󵄨󵄨 ∞ 󵄨󵄨 ∞
𝑡 𝑡 󵄨󵄨 󵄨󵄨 𝑤+𝑠

+ ∫ 𝐸 (𝑡, 𝑠) 𝑔 (𝑥 (𝑠)) 𝑑𝑠 + ∫ 𝐻 (𝑠) 𝑔 (𝑥 (𝑠)) 𝑑𝑠 󵄨󵄨∫ 𝐶 (𝑢 + 𝑠 − 𝑡 , 𝑠) 𝑑𝑢 󵄨󵄨 ≤ ∫ 𝑒− ∫𝑠 𝐴(V)𝑑V |𝐺 (𝑠, 𝑠)| 𝑑𝑤

󵄨󵄨 𝑡−𝑠+𝑡 0 󵄨󵄨
𝑡0 𝑡−𝑟(𝑡) 󵄨 0 󵄨 𝑡−𝑠

𝑄 𝑤+𝑠
𝑡0
= ∫ 𝑒− ∫𝑠 𝐴(V)𝑑V
𝑡

−𝑒
− ∫𝑡 𝐻(𝑠)𝑑𝑠
0 ⋅∫ 𝐻 (𝑠) 𝑔 (𝜓 (𝑠)) 𝑑𝑠 |𝐺 (𝑠, 𝑠)| 𝑑𝑤
𝑡−𝑠
𝑡0 −𝑟(𝑡0 )
∞ 𝑤+𝑠
𝑡 𝑢 𝑡
− ∫𝑢 𝐻(𝑠)𝑑𝑠 + ∫ 𝑒− ∫𝑠 𝐴(V)𝑑V
|𝐺 (𝑠, 𝑠)| 𝑑𝑤,
− ∫ [∫ 𝐻 (𝑠) 𝑔 (𝑥 (𝑠)) 𝑑𝑠] 𝑒 𝐻 (𝑢) 𝑑𝑢 𝑄
𝑡0 𝑢−𝑟(𝑢)
∞ 𝑤+𝑠
𝑒−𝑎0 𝑄
𝑡 𝑢 𝑡 ∫ 𝑒− ∫𝑠 𝐴(V)𝑑V
|𝐺 (𝑠, 𝑠)| 𝑑𝑤 ≤ 𝑀 .
− ∫ [∫ 𝐸 (𝑢, 𝑠) 𝑔 (𝑥 (𝑠 − 𝑟 (𝑠))) 𝑑𝑠] 𝑒 − ∫𝑢 𝐻(𝑠)𝑑𝑠
𝐻 (𝑢) 𝑑𝑢 𝑄 𝑎0
𝑡0 𝑡0 (27)
∞
𝑡 𝑡 𝑢 This implies that the integral ∫𝑡−𝑠+𝑡 𝐶(𝑢 + 𝑠 − 𝑡0 , 𝑠)𝑑𝑢 is
+ ∫ 𝑒− ∫𝑢 𝐻(𝑠)𝑑𝑠 (∫ 𝐺 (𝑢, V) 𝑔 (𝑥 (V)) 𝑑V) 𝑑𝑢 0

𝑡0 𝑢−𝑟(𝑢) convergent. Hence, we have

𝑡 𝑡 𝑢 𝑥̇
(𝑡) = 𝐵 (𝑡) − 𝑔 (𝑥 (𝑡 − 𝑟 (𝑡))) 𝐷 (𝑡)
− ∫𝑡 𝐴(𝑠)𝑑𝑠
− ∫ 𝑒− ∫𝑢 𝐻(𝑠)𝑑𝑠 𝑒 0
𝑡0 𝑑 𝑡
+ ∫ 𝐸 (𝑡, 𝑠) 𝑔 (𝑥 (𝑠 − 𝑟 (𝑠))) 𝑑𝑠
𝑡0 𝑑𝑡 𝑡0
× (∫ 𝐺 (𝑡0 , V) 𝑔 (𝜓 (V)) 𝑑V) 𝑑𝑢 𝑡
𝑡0 −𝑟(𝑡0 )
𝑡
𝑑 𝑠
+ ∫ 𝑒− ∫𝑠 𝐴(V)𝑑V ( ∫ 𝐺 (𝑠, V) 𝑔 (𝑥 (V)) 𝑑V) 𝑑𝑠.
𝑡0 𝑑𝑠 𝑠−𝑟(𝑠)
𝑡 𝑡 𝑢 𝑢
− ∫𝑢 𝐻(𝑠)𝑑𝑠 − ∫𝑠 𝐴(V)𝑑V
−∫ 𝑒 [∫ 𝑒 𝐴 (𝑠)
𝑡0 𝑡0
(28)
𝑠
× (∫ 𝐺 (𝑠, V) 𝑔 (𝑥 (V)) 𝑑V) 𝑑𝑠] 𝑑𝑢. Then we have
𝑠−𝑟(𝑠)
(23) 𝑥̇ ̃ (𝑝 (𝑡)) 𝑔 (𝑥 (𝑡))
(𝑡) = 𝐵 (𝑡) − 𝐷

𝑑 𝑡 ̃ (𝑝 (𝑠)) 𝑔 (𝑥 (𝑠)) 𝑑𝑠
Proof. We apply the variation of parameters formula to the + ∫ 𝐷
𝑑𝑡 𝑡−𝑟(𝑡)
second equation of (14); then we obtain
𝑑 𝑡
𝑡 + ∫ 𝐸 (𝑡, 𝑠) 𝑔 (𝑥 (𝑠 − 𝑟 (𝑠))) 𝑑𝑠
𝑥̇
(𝑡) = 𝑥̇
(𝑡0 ) 𝑒
− ∫𝑡
0
𝐴(𝑠)𝑑𝑠 𝑑𝑡 𝑡0
𝑡
(24) 𝑡
𝑑 𝑠
𝑡 𝑡
− ∫𝑠 𝐴(V)𝑑V
𝑠 + ∫ 𝑒− ∫𝑠 𝐴(V)𝑑V ( ∫ 𝐺 (𝑠, V) 𝑔 (𝑥 (V)) 𝑑V) 𝑑𝑠,
−∫ 𝑒 (∫ 𝑎 (𝑠, V) 𝑔 (𝑥 (V)) 𝑑V) 𝑑𝑠. 𝑡0 𝑑𝑠 𝑠−𝑟(𝑠)
𝑡0 𝑠−𝑟(𝑠) (29)
Abstract and Applied Analysis 5

𝑥̇
(𝑡) = 𝐵 (𝑡) − 𝐻 (𝑡) 𝑥 (𝑡) + 𝐻 (𝑡) [𝑥 (𝑡) − 𝑔 (𝑥 (𝑡))] defined on 𝐶𝜓𝑙 as follows: for 𝜙 ∈ 𝐶𝜓𝑙 , if 𝑡 ∈ [𝑚(𝑡0 ), 𝑡0 ],
(𝑃1 𝜙)(𝑡) = 𝜓(𝑡). If 𝑡 > 𝑡0 ,
𝑑 𝑡
+ ∫ 𝐻 (𝑠) 𝑔 (𝑥 (𝑠)) 𝑑𝑠
𝑑𝑡 𝑡−𝑟(𝑡) (𝑃1 𝜙) (𝑡)
𝑡
𝑑 𝑡
+ ∫ 𝐸 (𝑡, 𝑠) 𝑔 (𝑥 (𝑠 − 𝑟 (𝑠))) 𝑑𝑠 𝑡
− ∫𝑡 𝐻(𝑠)𝑑𝑠
𝑡

𝑑𝑡 𝑡0 = 𝜓 (𝑡0 ) 𝑒 0 + ∫ 𝑒− ∫𝑢 𝐻(𝑠)𝑑𝑠 𝐵 (𝑢) 𝑑𝑢

𝑡0
𝑡 𝑡
− ∫𝑠 𝐴(V)𝑑V 𝑑 𝑠 𝑡
+∫ 𝑒 ( ∫ 𝐺 (𝑠, V) 𝑔 (𝑥 (V)) 𝑑V) 𝑑𝑠. 𝑡

𝑡0 𝑑𝑠 𝑠−𝑟(𝑠) + ∫ 𝑒− ∫𝑢 𝐻(𝑠)𝑑𝑠 𝐻 (𝑢) [𝜙 (𝑢) − 𝑔 (𝜙 (𝑢))] 𝑑𝑢

𝑡0
(30)
𝑡 𝑡
+ ∫ 𝐸 (𝑡, 𝑠) 𝑔 (𝜙 (𝑠)) 𝑑𝑠 + ∫ 𝐻 (𝑠) 𝑔 (𝜙 (𝑠)) 𝑑𝑠
For ∀𝑡 ∈ [𝑡0 , 𝑇1 ], by the variation of parameters formula, we 𝑡0 𝑡−𝜏(𝑡)
obtain that 𝑡 𝑡0
− ∫𝑡 𝐻(𝑠)𝑑𝑠
−𝑒 0 ⋅∫ 𝐻 (𝑠) 𝑔 (𝜓 (𝑠)) 𝑑𝑠
𝑡0 −𝑟(𝑡0 )
𝑥 (𝑡)
𝑡 𝑢 𝑡

− ∫𝑡
𝑡
𝐻(𝑠)𝑑𝑠
𝑡 𝑡
− ∫𝑢 𝐻(𝑠)𝑑𝑠 − ∫ [∫ 𝐻 (𝑠) 𝑔 (𝜙 (𝑠)) 𝑑𝑠] 𝑒− ∫𝑢 𝐻(𝑠)𝑑𝑠 𝐻 (𝑢) 𝑑𝑢
= 𝜓 (𝑡0 ) 𝑒 0 +∫ 𝑒 𝐵 (𝑢) 𝑑𝑢 𝑡0 𝑢−𝑟(𝑢)
𝑡0 𝑡 𝑢 𝑡

𝑡 𝑢 − ∫ [∫ 𝐸 (𝑢, 𝑠) 𝑔 (𝜙 (𝑠 − 𝑟 (𝑠))) 𝑑𝑠] 𝑒− ∫𝑢 𝐻(𝑠)𝑑𝑠 𝐻 (𝑢) 𝑑𝑢

𝑡
− ∫𝑢 𝐻(𝑠)𝑑𝑠 𝑑 𝑡0 𝑡0
+∫ 𝑒 [ ∫ 𝐻 (𝑠) 𝑔 (𝑥 (𝑠)) 𝑑𝑠] 𝑑𝑢
𝑡0 𝑑𝑢 𝑢−𝑟(𝑢)
𝑡 𝑡 𝑢
𝑡 𝑡
− ∫𝑢 𝐻(𝑠)𝑑𝑠
+ ∫ 𝑒− ∫𝑢 𝐻(𝑠)𝑑𝑠 (∫ 𝐺 (𝑢, V) 𝑔 (𝜙 (V)) 𝑑V) 𝑑𝑢
+∫ 𝑒 𝐻 (𝑢) [𝑥 (𝑢) − 𝑔 (𝑥 (𝑢))] 𝑑𝑢 𝑡0 𝑢−𝑟(𝑢)
𝑡0
𝑡 𝑡 𝑢
− ∫𝑡 𝐴(𝑠)𝑑𝑠
𝑡 𝑡
− ∫𝑢 𝐻(𝑠)𝑑𝑠 𝑑 𝑢 − ∫ 𝑒− ∫𝑢 𝐻(𝑠)𝑑𝑠 𝑒 0
+∫ 𝑒 [ ∫ 𝐸 (𝑢, 𝑠) 𝑔 (𝑥 (𝑠 − 𝑟 (𝑠))) 𝑑𝑠] 𝑑𝑢. 𝑡0
𝑡0 𝑑𝑢 𝑡0
𝑡0
𝑡 𝑢
𝑡
− ∫𝑢 𝐻(𝑠)𝑑𝑠
𝑢
− ∫𝑠 𝐴(V)𝑑V × (∫ 𝐺 (𝑡0 , V) 𝑔 (𝜓 (V)) 𝑑V) 𝑑𝑢
+∫ 𝑒 [∫ 𝑒 𝑡0 −𝑟(𝑡0 )
𝑡0 𝑡0
𝑡 𝑢
𝑑 𝑠 𝑡 𝑢

×( ∫ 𝐺 (𝑠, V) 𝑔 (𝑥 (V)) 𝑑V) 𝑑𝑠] 𝑑𝑢. − ∫ 𝑒− ∫𝑢 𝐻(𝑠)𝑑𝑠 [∫ 𝑒− ∫𝑠 𝐴(V)𝑑V

𝐴 (𝑠)
𝑑𝑠 𝑠−𝑟(𝑠) 𝑡0 𝑡0
(31)
𝑠
× (∫ 𝐺 (𝑠, V) 𝑔 (𝜙 (V)) 𝑑V) 𝑑𝑠] 𝑑𝑢.
𝑠−𝑟(𝑠)
If we integrate the last several terms by parts, we have (23). (33)
This ends the proof of this lemma.
Note that 𝑃1 may not be a contraction mapping. We solve
this problem in Lemma 7 by introducing an exponentially
weighted metric.
Let (𝐶, ‖ ⋅ ‖) be the Banach space of bounded continuous
functions on [𝑚(𝑡0 ), ∞) with the supremum norm. For a Lemma 7. Suppose that there exists a constant 𝑙 > 0 such that
given continuous initial function 𝜓 : [𝑚(𝑡0 ), 𝑡0 ] → 𝑅, define 𝑔(𝑥) satisfies Lipschitz condition on [−𝑙, 𝑙]. Then there exists a
the set 𝐶𝜓 ⊂ 𝐶 by
metric 𝑑 on 𝐶𝜓𝑙 such that

𝐶𝜓 = {𝜙 : [𝑚 (𝑡0 ) , ∞) 󳨀→ 𝑅 | 𝜙 ∈ 𝐶, 𝜙 (𝑡) = 𝜓 (𝑡) , (i) the metric space (𝐶𝜓𝑙 , 𝑑) is complete;

𝑡 ∈ [𝑚 (𝑡0 ) , 𝑡0 ]} , (ii) 𝑃1 is a contraction mapping on (𝐶𝜓𝑙 , 𝑑) if 𝑃1 maps 𝐶𝜓𝑙

(32) into itself.
𝐶𝜓𝑙 := {𝜙 : [𝑚 (𝑡0 ) , ∞) 󳨀→ 𝑅 | 𝜙 ∈ 𝐶, 𝜙 (𝑡) = 𝜓 (𝑡) ,
󵄨 󵄨 Proof. (i) We change the supremum norm to an exponen-
𝑡 ∈ [𝑚 (𝑡0 ) , 𝑡0 ] , 󵄨󵄨󵄨𝜙 (𝑡)󵄨󵄨󵄨 ≤ 𝑙, 𝑡 ≥ 𝑚 (𝑡0 )} , tially weighted norm |𝜙|ℎ , which is defined on 𝐶𝜓𝑙 . Let 𝑆 be
the space of all continuous functions 𝜙 : [𝑚(𝑡0 ), ∞) → 𝑅
where 𝜓 : [𝑚(𝑡0 ), 𝑡0 ] → [−𝑙, 𝑙] is a given initial function and such that
𝑙 is a positive constant. We will also use ‖ ⋅ ‖ to denote the
󵄨󵄨 󵄨󵄨 󵄨 󵄨 −ℎ(𝑡)
supremum norm of an initial function. Let 𝑃1 be a mapping 󵄨󵄨𝜙󵄨󵄨ℎ := sup {󵄨󵄨󵄨𝜙 (𝑡)󵄨󵄨󵄨 𝑒 : 𝑡 ∈ [𝑚 (𝑡0 ) , ∞)} < ∞, (34)
6 Abstract and Applied Analysis

𝑡 𝑠
where ℎ(𝑡) = 𝑘𝐿 ∫𝑡 [𝐻(𝑠) + 𝐷(𝑠) + ∫𝑠−𝑟(𝑠) |𝐺(𝑠, V)|𝑑V]𝑑𝑠, 𝑘 is a For V ≤ 𝑡, since 𝐷(𝑡), 𝐻(𝑡) ≥ 0, we have
0
constant and 𝑘 > 7, and 𝐿 is the common Lipschitz constant 𝑡 𝑠
for 𝑥 − 𝑔(𝑥) and 𝑔(𝑥). Then (𝑆, | ⋅ |ℎ ) is a Banach space. Thus, ℎ (V) − ℎ (𝑡) = −𝑘𝐿 ∫ [𝐻 (𝑠) + 𝐷 (𝑠) + ∫ |𝐺 (𝑠, V)| 𝑑V] 𝑑𝑠
(𝑆, 𝑑) is a complete metric space, where 𝑑 denotes the induced V 𝑠−𝑟(𝑠)
metric: 𝑑(𝜙, 𝜂) = |𝜙 − 𝜂|ℎ , where 𝜙, 𝜂 ∈ 𝑆. Under this metric, 𝑡 𝑠
the space 𝐶𝜓𝑙 is a closed subset of 𝑆. Therefore, the metric ≤ −𝑘𝐿 ∫ ∫ |𝐺 (𝑠, V)| 𝑑V 𝑑𝑠.
V 𝑠−𝑟(𝑠)
space (𝐶𝜓𝑙 , 𝑑) is complete. (36)
(ii) Suppose that 𝑃1 : → 𝐶𝜓𝑙 𝐶𝜓𝑙 . For 𝜙, 𝜂 ∈ 𝐶𝜓𝑙 , since
𝐻(𝑡) ≥ 0 and 𝐸(𝑡, 𝑠) ≥ 0, then For 𝑢 ≤ 𝑡, since 𝐷(𝑡) ≥ 0, we have
𝑡
󵄨󵄨 󵄨 −ℎ(𝑡) ℎ (𝑢) − ℎ (𝑡) ≤ −𝑘𝐿 ∫ 𝐻 (𝑠) 𝑑𝑠. (37)
󵄨󵄨(𝑃1 𝜙) (𝑡) − (𝑃1 𝜂) (𝑡)󵄨󵄨󵄨 𝑒 𝑢

𝑡 𝑡 For 𝑠 ≤ 𝑡,
≤ ∫ 𝑒− ∫𝑢 𝐻(𝑠)𝑑𝑠 𝐻 (𝑢)
𝑡0 𝑡
ℎ (𝑠 − 𝑟 (𝑠)) − ℎ (𝑡) ≤ −𝑘𝐿 ∫ 𝐷 (𝑢) 𝑑𝑢. (38)
󵄨 󵄨
× 󵄨󵄨󵄨[𝜙 (𝑢) − 𝑔 (𝜙 (𝑢))] − [𝜂 (𝑢) − 𝑔 (𝜂 (𝑢))]󵄨󵄨󵄨 𝑒−ℎ(𝑡) 𝑑𝑢 𝑠

𝑡
Since 𝐸(𝑡, 𝑠) ≥ 0, then
󵄨 󵄨
+ ∫ 𝐸 (𝑡, 𝑠) 󵄨󵄨󵄨𝑔 (𝜙 (𝑠)) − 𝑔 (𝜂 (𝑠))󵄨󵄨󵄨 𝑒−ℎ(𝑡) 𝑑𝑠 ∞
𝑡0
𝐸 (𝑡, 𝑠) = ∫ 𝐶 (𝑢 + 𝑠 − 𝑡0 , 𝑠) 𝑑𝑢
𝑡−𝑠+𝑡0
𝑡
󵄨 󵄨 (39)
+∫ 𝐻 (𝑠) 󵄨󵄨󵄨𝑔 (𝜙 (𝑠)) − 𝑔 (𝜂 (𝑠))󵄨󵄨󵄨 𝑒−ℎ(𝑡) 𝑑𝑠 ∞
𝑡−𝑟(𝑡) ≤ ∫ 𝐶 (𝑢 + 𝑠 − 𝑡0 , 𝑠) 𝑑𝑢 = 𝐷 (𝑠) .
𝑡0
𝑡 𝑢
󵄨 󵄨
+ ∫ [∫ 𝐻 (𝑠) 󵄨󵄨󵄨𝑔 (𝜙 (𝑠)) − 𝑔 (𝜂 (𝑠))󵄨󵄨󵄨 𝑒−ℎ(𝑡) 𝑑𝑠] For 𝑠 ≤ 𝑢,
𝑡 𝑢−𝑟(𝑢)
0
𝑠 𝑢

×𝑒
𝑡
− ∫𝑢 𝐻(𝑠)𝑑𝑠
𝐻 (𝑢) 𝑑𝑢 ∫ |𝐺 (𝑠, V)| 𝑑V ≤ ∫ |𝐺 (𝑢, V)| 𝑑V. (40)
𝑠−𝑟(𝑠) 𝑢−𝑟(𝑢)

𝑡 𝑢
󵄨 󵄨 Easy calculation shows that
+ ∫ [∫ 𝐸 (𝑢, 𝑠) 󵄨󵄨󵄨𝑔 (𝜙 (𝑠 − 𝑟 (𝑠))) − 𝑔 (𝜂 (𝑠 − 𝑟 (𝑠)))󵄨󵄨󵄨
𝑡0 𝑡0
󵄨󵄨 󵄨 −ℎ(𝑡)
󵄨󵄨(𝑃1 𝜙) (𝑡) − (𝑃1 𝜂) (𝑡)󵄨󵄨󵄨 𝑒
𝑡
× 𝑒−ℎ(𝑡) 𝑑𝑠] 𝑒 − ∫𝑢 𝐻(𝑠)𝑑𝑠
𝐻 (𝑢) 𝑑𝑢 𝑡
1 2 1 − ∫𝑢𝑡 𝐻(𝑠)𝑑𝑠
≤{ + +∫ 𝑒 𝐻 (𝑢) 𝑑𝑢
𝑡 𝑢 𝑘𝐿 + 1 𝑘𝐿 𝑡0 𝑘𝐿
+ ∫ [∫ 𝐸 (𝑢, 𝑠) 𝑔 (𝜙 (𝑠 − 𝑟 (𝑠))) 𝑑𝑠] 𝑡 (41)
𝑡0 𝑡0 1 − ∫𝑢𝑡 𝐻(𝑠)𝑑𝑠 2 󵄨 󵄨
+∫ 𝑒 𝐻 (𝑢) 𝑑𝑢 + } 𝐿󵄨󵄨𝜙 − 𝜂󵄨󵄨󵄨ℎ
𝑡 𝑡0 𝑘𝐿 𝑘𝐿 󵄨
× 𝑒−ℎ(𝑡) 𝑒− ∫𝑢 𝐻(𝑠)𝑑𝑠 𝐻 (𝑢) 𝑑𝑢
7 󵄨󵄨 󵄨
≤ 󵄨𝜙 − 𝜂󵄨󵄨󵄨ℎ , 𝑡 > 𝑡0 .
𝑡 𝑡
− ∫𝑢 𝐻(𝑠)𝑑𝑠 −ℎ(𝑡) 𝑘󵄨
+∫ 𝑒 𝑒 𝑥
𝑡0 For 𝑡 ∈ [𝑚(𝑡0 ), 𝑡0 ], (𝑃1 𝜙)(𝑡) = (𝑃1 𝜂)(𝑡) = 𝜓(𝑡). Hence, 𝑑(𝑃1 𝜙−
𝑢 𝑃1 𝜂) ≤ (7/𝑘)𝑑(𝜙 − 𝜂). Note that 𝑘 > 7; thus 𝑃1 is a contraction
󵄨 󵄨
× (∫ |𝐺 (𝑢, V)| 󵄨󵄨󵄨𝑔 (𝜙 (V)) − 𝑔 (𝜂 (V))󵄨󵄨󵄨 𝑑V) 𝑑𝑢 mapping on (𝐶𝜓𝑙 , 𝑑).
𝑢−𝑟(𝑢)

𝑡 𝑡 𝑢 𝑢 We continue to prove Theorem 2. Choose

+ ∫ 𝑒− ∫𝑢 𝐻(𝑠)𝑑𝑠 𝑒−ℎ(𝑡) (∫ 𝑒− ∫𝑠 𝐴(V)𝑑V
𝐴 (𝑠) ̇0 ) satisfying ‖𝜓‖ + |𝑥(𝑡
𝜓 : [𝑚(𝑡0 ), 𝑡0 ] → 𝑅 and 𝑥(𝑡 ̇0 )| ≤ 𝛿
𝑡0 𝑡0
such that
𝑠
× (∫ |𝐺 (𝑠, V)| 𝑒−𝑎0 𝑄 𝑡0
𝑠−𝑟(𝑠) (𝑄 + ) 𝛿 + 𝛿 + 𝑔 (𝛿) ∫ 𝐻 (𝑠) 𝑑𝑠 ≤ (1 − 𝛼) 𝑔 (𝑙) .
󵄨 𝑎0 𝑡0 −𝑟(𝑡0 )
× 󵄨󵄨󵄨𝑔 (𝜙 (V)) (42)
󵄨
−𝑔 (𝜂 (V))󵄨󵄨󵄨 𝑑V) 𝑑𝑠) 𝑑𝑢. Since (i) implies that 𝑔(0) = 0, thus 𝑔(𝑙) ≤ 𝑙. Since
𝑔(𝑥) satisfies the Lipschitz condition on [−𝑙, 𝑙], thus 𝑔(𝑥) is
(35) continuous on [−𝑙, 𝑙], so such a 𝛿 exists and 𝛿 < 𝑙.
Abstract and Applied Analysis 7

By the expression of (𝑃1 𝜙)(𝑡), and condition (ii), we have 4. Proof of Theorem 3
𝑡
󵄨󵄨 󵄨 0 In this section, we will prove Theorem 3. First of all, we will
󵄨󵄨(𝑃1 𝜙) (𝑡)󵄨󵄨󵄨 ≤ 𝛿 + 𝑔 (𝛿) ∫ 𝐻 (𝑠) 𝑑𝑠 + [𝑙 − 𝑔 (𝑙)] + 𝛼𝑔 (𝑙) obtain a new expression of the solution of (13). We multiply
𝑡 −𝑟 0 (𝑡0 ) 𝑡
∫ ℎ(𝑠)𝑑𝑠
𝑡 𝑡 𝑢
𝑒 𝑡0 by both sides of (30); then
− ∫𝑡 𝐴(𝑠)𝑑𝑠
+ ∫ 𝑒− ∫𝑢 𝐻(𝑠)𝑑𝑠 𝑒 0 𝑑𝑢 ⋅ 𝛿
𝑡0

𝑡0 𝑡 𝑢
𝑥 (𝑡)
− ∫𝑡 𝐴(𝑠)𝑑𝑠
≤ 𝛿 + 𝑔 (𝛿) ∫ 𝐻 (𝑠) 𝑑𝑠 + ∫ 𝑒 0 𝑑𝑢 ⋅ 𝛿 𝑡 𝑡 𝑡
𝑡0 −𝑟(𝑡0 ) − ∫𝑡 ℎ(𝑠)𝑑𝑠
𝑡0
= 𝜓 (𝑡0 ) 𝑒 0 + ∫ 𝑒− ∫𝑢 ℎ(𝑠)𝑑𝑠 𝐵 (𝑢) 𝑑𝑢
𝑡0
+ [𝑙 − 𝑔 (𝑙)] + 𝛼𝑔 (𝑙) .
(43) 𝑡 𝑡
+∫ 𝑒− ∫𝑢 ℎ(𝑠)𝑑𝑠 ℎ (𝑢) 𝑥 (𝑢) 𝑑𝑢
By condition (iii), some easy computation shows that 𝑡0

𝑡
𝑡 𝑢 𝑡0 +𝑄 𝑢 𝑡 𝑢 𝑡
𝑑 𝑢
∫ 𝑒
− ∫𝑡 𝐴(𝑠)𝑑𝑠
0 𝑑𝑢 = ∫
− ∫𝑡 𝐴(𝑠)𝑑𝑠
𝑒 0 𝑑𝑢 + ∫
− ∫𝑡 𝐴(𝑠)𝑑𝑠
𝑒 0 𝑑𝑢 +∫ 𝑒− ∫𝑢 ℎ(𝑠)𝑑𝑠 [ ∫ 𝐻 (𝑠) 𝑔 (𝑥 (𝑠)) 𝑑𝑠] 𝑑𝑢
𝑡0 𝑡0 𝑡0 +𝑄 𝑡0 𝑑𝑢 𝑢−𝑟(𝑢)
𝑡0 +𝑄 𝑡 𝑡 𝑡
≤∫ 𝑑𝑢 + ∫ 𝑒−𝑎0 (𝑢−𝑡0 ) 𝑑𝑢 −∫ 𝑒− ∫𝑢 ℎ(𝑠)𝑑𝑠 𝐻 (𝑢) 𝑔 (𝑥 (𝑢)) 𝑑𝑢
𝑡0 𝑡0 +𝑄 𝑡0

𝑒−𝑎0 𝑄 𝑡 𝑡
𝑑 𝑢
≤𝑄+ . +∫ 𝑒− ∫𝑢 ℎ(𝑠)𝑑𝑠 [ ∫ 𝐸 (𝑢, 𝑠) 𝑔 (𝑥 (𝑠 − 𝑟 (𝑠))) 𝑑𝑠] 𝑑𝑢
𝑎0 𝑡0 𝑑𝑢 𝑡0
(44)
𝑡 𝑡 𝑢 𝑢

Hence, +∫ 𝑒− ∫𝑢 ℎ(𝑠)𝑑𝑠 [∫ 𝑒− ∫𝑠 𝐴(V)𝑑V

𝑡0 𝑡0
󵄨󵄨 󵄨
󵄨󵄨(𝑃1 𝜙) (𝑡)󵄨󵄨󵄨 ≤ (1 − 𝛼) 𝑔 (𝑙) + (𝛼 − 1) 𝑔 (𝑙) + 𝑙 = 𝑙. (45)
𝑑 𝑠
×( ∫ 𝐺 (𝑠, V) 𝑔 (𝑥 (V)) 𝑑V) 𝑑𝑠] 𝑑𝑢.
Observe that if 𝑡 ∈ [𝑚(𝑡0 ), 𝑡0 ], then (𝑃1 𝜙)(𝑡) = 𝜓(𝑡). We 𝑑𝑠 𝑠−𝑟(𝑠)
obtain that |(𝑃1 𝜙)(𝑡)| ≤ 𝑙, 𝑡 ∈ [𝑚(𝑡0 ), ∞). Thus, 𝑃1 𝜙 : 𝐶𝜓𝑙 → (49)
𝐶𝜓𝑙 . Since we have proved that 𝑃1 is a contraction mapping,
hence 𝑃1 has a unique fixed point 𝑥(𝑡) and |𝑥(𝑡)| ≤ 𝑙. We have
Recall (24); we have
󵄨󵄨 󵄨 󵄨 󵄨 − ∫𝑡 𝐴(𝑠)𝑑𝑠
󵄨󵄨𝑦 (𝑡)󵄨󵄨󵄨 ≤ 󵄨󵄨󵄨𝑥̇
(𝑡0 )󵄨󵄨󵄨 𝑒 𝑡0 𝑥 (𝑡)
𝑡 𝑠 𝑡 𝑡 𝑡
𝑡
󵄨 󵄨 − ∫𝑡 ℎ(𝑠)𝑑𝑠
+ ∫ 𝑒− ∫𝑢 ℎ(𝑠)𝑑𝑠 𝐵 (𝑢) 𝑑𝑢
+ ∫ 𝑒− ∫𝑠 𝐴(V)𝑑V ∫ |𝑎 (𝑠, V)| 󵄨󵄨󵄨𝑔 (𝑥 (V))󵄨󵄨󵄨 𝑑V 𝑑𝑠. = 𝜓 (𝑡0 ) 𝑒 0

𝑡0 𝑠−𝑟(𝑠) 𝑡0
(46) 𝑡 𝑡
− ∫𝑢 ℎ(𝑠)𝑑𝑠
𝑡
+∫ 𝑒 [ℎ (𝑢) − 𝐻 (𝑢)] 𝑥 (𝑢) 𝑑𝑢
Since ∫𝑡−𝑟(𝑡) |𝑎(𝑡, V)|𝑑V is bounded, ∃ a constant 𝑁 > 0 such 𝑡0
𝑡
that ∫𝑡−𝑟(𝑡) |𝑎(𝑡, V)|𝑑V ≤ 𝑁, then 𝑡 𝑡
𝑑 𝑢
+∫ 𝑒− ∫𝑢 ℎ(𝑠)𝑑𝑠 [ ∫ 𝐻 (𝑠) 𝑔 (𝑥 (𝑠)) 𝑑𝑠] 𝑑𝑢
𝑡
𝑡0 𝑑𝑢 𝑢−𝑟(𝑢)
󵄨󵄨 󵄨 󵄨 󸀠 󵄨 𝑡
󵄨󵄨𝑦 (𝑡)󵄨󵄨󵄨 ≤ 󵄨󵄨󵄨󵄨𝑥 (𝑡0 )󵄨󵄨󵄨󵄨 + 𝑁𝐿𝑙 ⋅ ∫ 𝑒 𝑢
− ∫ 𝐴(𝑠)𝑑𝑠
𝑑𝑢 𝑡 𝑡
𝑡 0
(47) −∫ 𝑒− ∫𝑢 ℎ(𝑠)𝑑𝑠 𝐻 (𝑢) [𝑔 (𝑥 (𝑢)) − 𝑥 (𝑢)] 𝑑𝑢
𝑡0
𝑒−𝑎0 𝑄
< 𝑙 (1 + 𝑁𝐿 (𝑄 + )) .
𝑎0 𝑡 𝑡
𝑑 𝑢
+∫ 𝑒− ∫𝑢 ℎ(𝑠)𝑑𝑠 [ ∫ 𝐸 (𝑢, 𝑠) 𝑔 (𝑥 (𝑠 − 𝑟 (𝑠))) 𝑑𝑠] 𝑑𝑢
It follows that 𝑡0 𝑑𝑢 𝑡0
𝑡 𝑢
󵄨 󵄨 𝑒−𝑎0 𝑄 𝑡 𝑢

|𝑥 (𝑡)| + 󵄨󵄨󵄨𝑦 (𝑡)󵄨󵄨󵄨 < 𝑙 (2 + 𝑁𝐿 (𝑄 + )) . (48) +∫ 𝑒− ∫𝑢 ℎ(𝑠)𝑑𝑠 [∫ 𝑒− ∫𝑠 𝐴(V)𝑑V

𝑎0 𝑡0 𝑡0

To show the stability of zero solution, let ∀𝜖 > 0 be given; 𝑑 𝑠

×( ∫ 𝐺 (𝑠, V) 𝑔 (𝑥 (V)) 𝑑V) 𝑑𝑠] 𝑑𝑢.
we only need to replace 𝜖 by 𝑙. This completes the proof of 𝑑𝑠 𝑠−𝑟(𝑠)
Theorem 2. (50)
8 Abstract and Applied Analysis

𝑡 𝑡 𝑡
Performing an integration by parts, then we have − ∫𝑡 ℎ(𝑠)𝑑𝑠
+∫ 𝐸 (𝑡, 𝑠) 𝑔 (𝜙 (𝑠)) 𝑑𝑠+∫ 𝐻 (𝑠) 𝑔 (𝜙 (𝑠)) 𝑑𝑠−𝑒 0
𝑡0 𝑡−𝑟(𝑡)

𝑡0 𝑡 𝑢
𝑥 (𝑡)
⋅∫ 𝐻 (𝑠) 𝑔 (𝜓 (𝑠)) 𝑑𝑠 − ∫ [∫ 𝐻 (𝑠) 𝑔 (𝜙 (𝑠)) 𝑑𝑠]
𝑡 𝑡 𝑡
𝑡0 −𝑟(𝑡0 ) 𝑡0 𝑢−𝑟(𝑢)
− ∫𝑡 ℎ(𝑠)𝑑𝑠 − ∫𝑠 ℎ(𝑢)𝑑𝑢
= 𝜓 (𝑡0 ) 𝑒 0 +∫ 𝑒 𝐵 (𝑠) 𝑑𝑠 𝑡
𝑡0
× 𝑒− ∫𝑢 ℎ(𝑠)𝑑𝑠 ℎ (𝑢) 𝑑𝑢
𝑡 𝑡
𝑡 𝑢
+∫ 𝑒− ∫𝑢 ℎ(𝑠)𝑑𝑠 [ℎ (𝑢) − 𝐻 (𝑢)] 𝑥 (𝑢) 𝑑𝑢 𝑡

𝑡0 −∫ [∫ 𝐸 (𝑢, 𝑠) 𝑔 (𝜙 (𝑠 − 𝑟 (𝑠))) 𝑑𝑠] 𝑒− ∫𝑢 ℎ(𝑠)𝑑𝑠 ℎ (𝑢) 𝑑𝑢

𝑡0 𝑡0
𝑡 𝑡
− ∫𝑢 ℎ(𝑠)𝑑𝑠 𝑡 𝑢
−∫ 𝑒 𝐻 (𝑢) [𝑔 (𝑥 (𝑢)) − 𝑥 (𝑢)] 𝑑𝑢 𝑡

𝑡0 +∫ 𝑒− ∫𝑢 ℎ(𝑠)𝑑𝑠 (∫ 𝐺 (𝑢, V) 𝑔 (𝜙 (V)) 𝑑V) 𝑑𝑢

𝑡0 𝑢−𝑟(𝑢)
𝑡 𝑡 𝑡
− ∫𝑡 ℎ(𝑠)𝑑𝑠 𝑡 𝑡0
+∫ 𝐸 (𝑡, 𝑠) 𝑔 (𝑥 (𝑠)) 𝑑𝑠+∫ 𝐻 (𝑠) 𝑔 (𝑥 (𝑠)) 𝑑𝑠 − 𝑒 0 𝑡 𝑢
− ∫𝑡 𝐴(𝑠)𝑑𝑠
𝑡0 𝑡−𝑟(𝑡) −∫ 𝑒− ∫𝑢 ℎ(𝑠)𝑑𝑠 𝑒 0 (∫ 𝐺 (𝑡0 , V) 𝑔 (𝜓 (V)) 𝑑V) 𝑑𝑢
𝑡0 𝑡0 −𝑟(𝑡0 )
𝑡0 𝑡 𝑢
⋅∫ 𝐻 (𝑠) 𝑔 (𝑥 (𝑠)) 𝑑𝑠 − ∫ [∫ 𝐻 (𝑠) 𝑔 (𝑥 (𝑠)) 𝑑𝑠] 𝑡 𝑡 𝑢 𝑢

𝑡0 −𝑟(𝑡0 ) 𝑡0 𝑢−𝑟(𝑢) −∫ 𝑒− ∫𝑢 ℎ(𝑠)𝑑𝑠 [∫ 𝑒− ∫𝑠 𝐴(V)𝑑V

𝐴 (𝑠)
𝑡0 𝑡0
𝑡
× 𝑒− ∫𝑢 ℎ(𝑠)𝑑𝑠 ℎ (𝑢) 𝑑𝑢 𝑠
× (∫ 𝐺 (𝑠, V) 𝑔 (𝜙 (V)) 𝑑V) 𝑑𝑠] 𝑑𝑢.
𝑡 𝑢 𝑡 𝑠−𝑟(𝑠)
− ∫𝑢 ℎ(𝑠)𝑑𝑠
−∫ [∫ 𝐸 (𝑢, 𝑠) 𝑔 (𝑥 (𝑠 − 𝑟 (𝑠))) 𝑑𝑠] 𝑒 ℎ (𝑢) 𝑑𝑢 (53)
𝑡0 𝑡0
𝑡
𝑡 𝑡
− ∫𝑢 ℎ(𝑠)𝑑𝑠
𝑢 If 𝜙 ∈ 𝐶𝜓0 , since ∫0 ℎ(𝑠)𝑑𝑠 → ∞ as 𝑡 → ∞, the first term
+∫ 𝑒 (∫ 𝐺 (𝑢, V) 𝑔 (𝑥 (V)) 𝑑V) 𝑑𝑢 and fourth term of (𝑃2 𝜙)(𝑡) → 0, as 𝑡 → ∞.
𝑡0 𝑢−𝑟(𝑢)
Note that
𝑡 𝑡 𝑢 𝑡0
− ∫𝑡 𝐴(𝑠)𝑑𝑠
−∫ 𝑒− ∫𝑢 ℎ(𝑠)𝑑𝑠 𝑒 0 (∫ 𝐺 (𝑡0 , V) 𝑔 (𝜓 (V)) 𝑑V) 𝑑𝑢 𝑡 𝑡 𝑡 𝑡 𝑢
− ∫𝑡 𝐴(𝑠)𝑑𝑠
𝑡0 𝑡0 −𝑟(𝑡0 ) ∫ 𝑒− ∫𝑢 ℎ(𝑠)𝑑𝑠 𝐵 (𝑢) 𝑑𝑢 = 𝑥̇
(𝑡0 ) ∫ 𝑒− ∫𝑢 ℎ(𝑠)𝑑𝑠 𝑒 0 𝑑𝑢.
𝑡0 𝑡0
𝑡 𝑡 𝑢 𝑢 (54)
−∫ 𝑒− ∫𝑢 ℎ(𝑠)𝑑𝑠 [∫ 𝑒− ∫𝑠 𝐴(V)𝑑V
𝐴 (𝑠)
𝑡0 𝑡0
Since 𝐴(𝑡) = 𝑓(𝑡, 𝑥(𝑡), 𝑦(𝑡)) ≥ 𝑎(𝑡) ≥ 0, then
𝑠
× (∫ 𝐺 (𝑠, V) 𝑔 (𝑥 (V)) 𝑑V) 𝑑𝑠] 𝑑𝑢. 𝑡 𝑡 𝑢
− ∫𝑡 𝐴(𝑠)𝑑𝑠
𝑡 𝑢
− ∫𝑡 𝐴(𝑠)𝑑𝑠
𝑠−𝑟(𝑠) ∫ 𝑒− ∫𝑢 ℎ(𝑠)𝑑𝑠 𝑒 0 𝑑𝑢 ≤ ∫ 𝑒 0 𝑑𝑢. (55)
𝑡0 𝑡0
(51)
For a given 𝜖 > 0, there exists 𝑇1 > 𝑄 + 𝑡0 such that
We define 𝑒𝑎0 𝑡0 −𝑎0 𝑇 < 𝑎0 𝜖. For 𝑇1 < 𝑡 and 𝑡0 < 𝑄 < 𝑇1 ≤ 𝑢, we have
𝑡 𝑢
− ∫𝑡 𝐴(𝑠)𝑑𝑠
𝑡
𝑒𝑎0 𝑡0 −𝑎0 𝑇1
∫ 𝑒 0 𝑑𝑢 ≤ ∫ 𝑒−𝑎0 (𝑢−𝑡0 ) 𝑑𝑢 = (𝑒 − 𝑒−𝑎0 𝑡 )
𝐶𝜓0 = {𝜙 : [𝑚 (𝑡0 ) , ∞) 󳨀→ 𝑅 | 𝜙 ∈ 𝐶, 𝜙 (𝑡) = 𝜓 (𝑡) , 𝑇1 𝑇1 𝑎0
(52)
𝑡 ∈ [𝑚 (𝑡0 ) , 𝑡0 ] , 𝜙 (𝑡) 󳨀→ 0 as 𝑡 󳨀→ ∞.} 𝑒𝑎0 𝑡0 −𝑎0 𝑇1
< < 𝜖.
𝑎0
(56)
Let 𝑃2 be a mapping defined on 𝐶𝜓 as follows: for 𝜙 ∈ 𝐶𝜓 , if
𝑡 ∈ [𝑚(𝑡0 ), 𝑡0 ], (𝑃2 𝜙)(𝑡) = 𝜓(𝑡). If 𝑡 > 𝑡0 , we define For 𝑡 > 𝑇1 , we have
𝑇1 𝑡 𝑢 𝑡
− ∫𝑡 𝐴(𝑠)𝑑𝑠 − ∫𝑇 ℎ(𝑠)𝑑𝑠
(𝑃2 𝜙) (𝑡) ∫ 𝑒− ∫𝑢 ℎ(𝑠)𝑑𝑠 𝑒 0 𝑑𝑢 = 𝑒 1
𝑡0
𝑡 𝑡 𝑡
− ∫𝑡 ℎ(𝑠)𝑑𝑠
= 𝜓 (𝑡0 ) 𝑒 0 + ∫ 𝑒− ∫𝑢 ℎ(𝑠)𝑑𝑠 𝐵 (𝑢) 𝑑𝑢 𝑇1 𝑇1 𝑢
ℎ(𝑠)𝑑𝑠 − ∫𝑡 𝐴(𝑠)𝑑𝑠
𝑡0 ⋅ ∫ 𝑒− ∫𝑢 𝑒 0 𝑑𝑢.
𝑡0
𝑡 𝑡
(57)
+∫ 𝑒− ∫𝑢 ℎ(𝑠)𝑑𝑠 [ℎ (𝑢) − 𝐻 (𝑢)] 𝜙 (𝑢) 𝑑𝑢
𝑡0
Then this term of (𝑃2 𝜙)(𝑡) → 0, as 𝑡 → ∞. Analogously, we
𝑡 𝑡 can prove other terms of (𝑃2 𝜙)(𝑡) → 0, as 𝑡 → ∞. Then we
− ∫𝑢 ℎ(𝑠)𝑑𝑠
−∫ 𝑒 𝐻 (𝑢) [𝑔 (𝜙 (𝑢)) − 𝜙 (𝑢)] 𝑑𝑢
𝑡0 can easily check that 𝑃2 is a contraction mapping on 𝐶𝜓0 by
Abstract and Applied Analysis 9

using condition (ii). By the contraction mapping principle, by way of contradiction. If

𝑃2 has a unique fixed point 𝑥(𝑡) in 𝐶𝜓0 . Thus, 𝑥(𝑡) → 0 as
𝑡 → +∞. ∞
∫ ℎ (𝑠) 𝑑𝑠 < ∞, (66)
Remark.8 By using the new expression (53), we do not need 0
to change the supremum metric to an exponentially weighted
metric. We can easily check that 𝑃2 is a contraction mapping. since ℎ(𝑡) ≥ 0, ∃ a sequence {𝑡𝑛 }, 𝑡𝑛 → ∞ as 𝑛 → ∞ such
In order to obtain the asymptotic stability, we now need 𝑡
that lim𝑛 → ∞ ∫0 𝑛 ℎ(𝑠)𝑑𝑠 = 𝜎, for a certain finite number 𝜎 ∈
to show that the zero solution is stable. Let 𝜖 > 0 be given; 𝑡
we choose 𝜓 : [𝑚(𝑡0 ), 𝑡0 ] → 𝑅 and 𝑥(𝑡 ̇0 ) satisfying ‖𝜓‖ + 𝑅. Choose 𝜂 such that −𝜂 ≤ ∫0 𝑛 ℎ(𝑠)𝑑𝑠 ≤ 𝜂 holds, for ∀𝑛 ≥ 1.
̇0 )| ≤ 𝛿 such that
|𝑥(𝑡 Denote

𝑒−𝑎0 𝑄 𝑡0 𝑢
𝛿 + (𝑄 + )𝛿 + ∫ 𝐻 (𝑠) 𝑑𝑠 ⋅ 𝐿𝛿 ≤ (1 − 𝛼) 𝜖. 𝜔 (𝑢) ≜ ℎ (𝑢) ∫ 𝐿𝐻 (𝑠) 𝑑𝑠 + 𝐿𝐻 (𝑢) + |ℎ (𝑢) − 𝐻 (𝑢)|
𝑎0 𝑡0 −𝑟(𝑡0 ) 𝑢−𝑟(𝑢)
(58) 𝑢 𝑢
+ 2𝐿 ∫ |𝐺 (𝑢, V)| 𝑑V + 𝐿 ∫ 𝐸 (𝑢, 𝑠) 𝑑𝑠.
By (51), we have 𝑢−𝑟(𝑢) 𝑡0

𝑡 𝑢
(67)
󵄨 󵄨 󵄩 󵄩 𝑡
− ∫𝑡 𝐴(𝑠)𝑑𝑠
|𝑥 (𝑡)| ≤ 𝛿 + (󵄨󵄨󵄨𝑥̇
(𝑡0 )󵄨󵄨󵄨 + 󵄩󵄩󵄩𝜓󵄩󵄩󵄩) ∫ 𝑒− ∫𝑢 ℎ(𝑠)𝑑𝑠 𝑒 0 𝑑𝑢
𝑡 0
(59) By conditions of Theorem 3, we have
𝑡0
+∫ 𝐻 (𝑠) 𝑑𝑠 ⋅ 𝐿𝛿 + 𝛼𝜖.
𝑡0 −𝑟(𝑡0 ) 𝑡𝑛 𝑡𝑛
0 ≤ ∫ 𝜔 (𝑢) 𝑒− ∫𝑢 ℎ(𝑠)𝑑𝑠
𝑑𝑢 ≤ 𝛼. (68)
We have obtained that 0

𝑡 𝑢 𝑡0 +𝑄 𝑡
− ∫𝑡 𝐴(𝑠)𝑑𝑠
∫ 𝑒 0 𝑑𝑢 ≤ ∫ 𝑑𝑢 + ∫ 𝑒−𝑎0 (𝑢−𝑡0 ) 𝑑𝑢 Then
𝑡0 𝑡0 𝑡0 +𝑄
(60)
𝑡𝑛
𝑒−𝑎0 𝑄 𝑢 𝑡𝑛

≤𝑄+ . ∫ 𝜔 (𝑢) 𝑒∫0 ℎ(𝑠)𝑑𝑠

𝑑𝑢 ≤ 𝛼𝑒∫0 ℎ(𝑠)𝑑𝑠
≤ 𝛼𝑒𝜂 . (69)
𝑎0 0

Hence, 𝑢
𝑡
Thus, the sequence {∫0 𝑛 𝜔(𝑢)𝑒∫0 ℎ(𝑠)𝑑𝑠 𝑑𝑢} is bounded;
|𝑥 (𝑡)| ≤ 𝜖. (61) there exists a convergent subsequence; we assume that
𝑡 𝑢

Recall (24); we have lim𝑛 → ∞ ∫0 𝑛 𝜔(𝑢)𝑒∫0 ℎ(𝑠)𝑑𝑠 𝑑𝑢 = 𝐼, 𝐼 ≥ 0. We can choose a

󵄨 󵄨 positive integer 𝑘 large enough such that
|𝑥̇(𝑡)| ≤ 󵄨󵄨󵄨𝑥̇
(𝑡0 )󵄨󵄨󵄨
𝑡 𝑠
𝑡
󵄨󵄨 󵄨 𝑡𝑛
𝛿0
󵄨󵄨𝑎 (𝑠, V) 𝑔 (𝑥 (V))󵄨󵄨󵄨 𝑑V) 𝑑𝑠.
𝑢
+ ∫ 𝑒− ∫𝑠 𝐴(V)𝑑V (∫ ∫ 𝜔 (𝑢) 𝑒∫0 ℎ(𝑠)𝑑𝑠
𝑑𝑢 < (70)
𝑡0 𝑠−𝑟(𝑠) 𝑡𝑘 8𝐾
(62)

It follows that for ∀𝑛 ≥ 𝑘, where 0 < 𝛿0 (𝛿0 < 𝜖) satisfying

−𝑎0 𝑄
󵄨 󵄨 𝑒
|𝑥 (𝑡)| + 󵄨󵄨󵄨𝑦 (𝑡)󵄨󵄨󵄨 < 𝜖 (2 + 𝑁𝐿 (𝑄 + )) . (63)
𝑎0 𝑒−𝑎0 𝑄
𝛿0 + 2 max {𝑄 + , 1} 𝛿0
𝑎0
Therefore, the zero solution is stable; since we have obtained (71)
that 𝑥(𝑡) → 0 as 𝑡 → ∞, it follows that the zero solution is 𝑡𝑘
asymptotically stable. + 𝛿0 ∫ 𝐻 (𝑠) 𝑑𝑠 ≤ (1 − 𝛼) .
𝑡𝑘 −𝑟(𝑡𝑘 )
The necessary condition is as follows: for each 𝑡0 ≥ 0, we
denote
𝑡 ̇𝑘 )) of (8)
Now we consider the solution 𝑥(𝑡) = 𝑥(𝑡, 𝜓, 𝑥(𝑡
𝐾 = sup {𝑒− ∫0 ℎ(𝑠)𝑑𝑠 } . (64) which satisfies
𝑡≥0

We will prove that 3𝛿0 𝛿0

𝜓 (𝑡𝑘 ) = , 𝑥̇
(𝑡𝑘 ) = ,
∞ 4 4 (72)
∫ ℎ (𝑠) 𝑑𝑠 = ∞ (65) 󵄨󵄨 󵄨
0 󵄨󵄨𝜓 (𝑠)󵄨󵄨󵄨 + |𝑥̇
(𝑠)| ≤ 𝛿0 , 𝑠 ≤ 𝑡𝑘 .
10 Abstract and Applied Analysis

We can obtain that |𝑥(𝑡)| ≤ 1 by a similar argument with (61) ̇𝑘 ) = 𝛿0 /4 > 0; it follows that
Note that 𝑥(𝑡
if we replace 𝜖 by 1. Then 󵄨󵄨 󵄨󵄨
𝑡𝑛 𝑡𝑛
󵄨󵄨 󵄨
󵄨󵄨𝑥 (𝑡𝑛 ) − ∫ 𝐸 (𝑡𝑛 , 𝑠) 𝑔 (𝑥 (𝑠)) 𝑑𝑠 − ∫
󵄨󵄨 𝐻 (𝑠) (𝑥 (𝑠)) 󵄨󵄨󵄨󵄨
𝑔 𝑑𝑠
𝑡𝑛 𝑢 󵄨 𝑡𝑘 𝑡𝑛 −𝑟(𝑡𝑛 ) 󵄨󵄨
𝑡𝑛
ℎ(𝑠)𝑑𝑠 − ∫𝑡 𝐴(𝑠)𝑑𝑠
𝜓 (𝑡𝑘 ) − ∫ 𝑒− ∫𝑢 𝑒 𝑘
𝑡 𝑡𝑛
𝑡𝑘 − ∫𝑡 𝑛 ℎ(𝑠)𝑑𝑠 𝛿0 𝑡
𝑘 𝑢
≥𝑒 𝑘 [ − 𝑒− ∫0 ℎ(𝑠)𝑑𝑠 ⋅ ∫ 𝜔 (𝑢) 𝑒∫0 ℎ(𝑠)𝑑𝑠 𝑑𝑢]
𝑡𝑘
4 𝑡𝑘
× (∫ 𝐺 (𝑡𝑘 , V) 𝑔 (𝜓 (V)) 𝑑V) 𝑑𝑢
𝑡𝑘 −𝑟(𝑡𝑘 ) 𝛿0 − ∫𝑡𝑡𝑛 ℎ(𝑠)𝑑𝑠 𝛿0 −2𝜂
≥ 𝑒 𝑘 ≥ 𝑒 > 0.
𝑡𝑘 𝑡𝑘
8 8
󵄨 󵄨 (76)
−∫ 𝐻 (𝑠) 𝜓 (𝑠) 𝑑𝑠 ≥ 𝜓 (𝑡𝑘 ) − ∫ 𝐻 (𝑠) 󵄨󵄨󵄨𝜓 (𝑠)󵄨󵄨󵄨 𝑑𝑠
𝑡𝑘 −𝑟(𝑡𝑘 ) 𝑡𝑘 −𝑟(𝑡𝑘 )
If the zero solution 𝑥(𝑡) is asymptotically stable, then 𝑥(𝑡) →
𝑡𝑛 𝑡
− ∫𝑢𝑛 ℎ(𝑠)𝑑𝑠 − ∫𝑡 𝐴(𝑠)𝑑𝑠
𝑢
0 as 𝑡 → ∞. By the mean value theorem, easy computation
−∫ 𝑒 𝑒 𝑘
shows that
𝑡𝑘
󵄨󵄨 𝑡𝑛 󵄨󵄨
𝑡𝑘 󵄨󵄨󵄨∫ 𝐻 (𝑠) 𝑔 (𝑥 (𝑠)) 𝑑𝑠 󵄨󵄨󵄨
󵄨 󵄨 󵄨󵄨 󵄨󵄨
× (∫ 𝐺 (𝑡𝑘 , V) 󵄨󵄨󵄨𝑔 (𝜓 (V))󵄨󵄨󵄨 𝑑V) 𝑑𝑢 󵄨󵄨 𝑡𝑛 −𝑟(𝑡𝑛 ) 󵄨󵄨
𝑡𝑘 −𝑟(𝑡𝑘 ) (77)
󵄨󵄨󵄨 𝑡𝑛 󵄨󵄨󵄨
3𝛿0 𝛿0 𝛿0 󵄨
= 󵄨󵄨󵄨𝑔 (𝑥 (𝜛)) ∫ 󵄨
𝐻 (𝑠) 𝑑𝑠󵄨󵄨󵄨 ≤ |𝑥 (𝜛)| .
≥ − = . 󵄨󵄨 𝑡𝑛 −𝑟(𝑡𝑛 ) 󵄨󵄨
4 2 4
(73) So this term tends to 0 as 𝑡𝑛 → ∞.
Easy computation shows that
By (23) and |𝑥(𝑡)| ≤ 1, if 𝑡𝑛 ≥ 𝑡𝑘 , we have 𝑡𝑛 𝑡𝑛
𝑥 (𝑡𝑛 ) − ∫ 𝐸 (𝑡𝑛 , 𝑠) 𝑔 (𝑥 (𝑠)) 𝑑𝑠 − ∫ 𝐻 (𝑠) 𝑔 (𝑥 (𝑠)) 𝑑𝑠
𝑡𝑘 𝑡𝑛 −𝑟(𝑡𝑛 )
󵄨󵄨 𝑡𝑛 𝑡𝑛 󵄨󵄨 (78)
󵄨󵄨 󵄨󵄨
󵄨󵄨𝑥 (𝑡𝑛 ) − ∫ 𝐸 (𝑡𝑛 , 𝑠) 𝑔 (𝑥 (𝑠)) 𝑑𝑠 − ∫ 𝐻 (𝑠) 𝑔 (𝑥 (𝑠)) 𝑑𝑠 󵄨󵄨
󵄨󵄨 𝑡𝑛 −𝑟(𝑡𝑛 )
󵄨󵄨 tends to 0 as 𝑡𝑛 tends to ∞. This is a contradiction to (76).
󵄨 𝑡𝑘 󵄨
󵄨󵄨󵄨 𝑡
− ∫ 𝑛 ℎ(𝑠)𝑑𝑠 𝑡𝑛 𝑡𝑛
Hence,
≥ 󵄨󵄨󵄨󵄨𝜓 (𝑡𝑘 ) 𝑒 𝑡𝑘 + ∫ 𝑒− ∫𝑢 ℎ(𝑠)𝑑𝑠 𝐵 (𝑢) 𝑑𝑢 ∞
󵄨󵄨 𝑡𝑘 ∫ ℎ (𝑠) 𝑑𝑠 = ∞. (79)
0
𝑡𝑛 𝑡𝑛 − ∫𝑡
𝑢
𝐴(𝑠)𝑑𝑠
− ∫ 𝑒− ∫𝑢 ℎ(𝑠)𝑑𝑠
𝑒 𝑘 This completes the proof of Theorem 3.
𝑡𝑘

𝑡𝑘 5. An Example
× (∫ 𝐺 (𝑡𝑘 , V) 𝑔 (𝜓 (V)) 𝑑V) 𝑑𝑢
𝑡𝑘 −𝑟(𝑡𝑘 )
In this section, we will give an example to apply our results.
󵄨󵄨 Let
𝑡𝑘 ℎ(𝑢)𝑑𝑢 󵄨󵄨󵄨
𝑡
− ∫𝑡 𝑛
−∫ 𝐻 (𝑠) 𝜓 (𝑠) 𝑑𝑠 ⋅ 𝑒 𝑘 󵄨󵄨 𝑡
󵄨󵄨 𝑟 (𝑡) = , 𝑝 (𝑡) = 2𝑡, 𝑎 (𝑡, 𝑠) = 𝑎∗ 𝑒−(𝑡−𝑠) ,
𝑡𝑘−𝑟(𝑡
𝑘
) 󵄨󵄨 2
󵄨󵄨 𝑡𝑛 󵄨󵄨 𝑡 𝑡
󵄨 𝑡𝑛
󵄨
− 󵄨󵄨󵄨󵄨∫ 𝑒− ∫𝑢 ℎ(𝑠)𝑑𝑠 𝜔 (𝑢) 𝑑𝑢󵄨󵄨󵄨󵄨 . ∫ |𝑎 (𝑡, V)| 𝑑V = ∫ 𝑎∗ 𝑒−(𝑡−V) 𝑑V = 𝑎∗ (1 − 𝑒−3𝑡/2 ) ≤ 𝑎∗ ,
󵄨󵄨 𝑡𝑘 󵄨󵄨 𝑡−𝑟(𝑡) 𝑡/2
(74)
for 𝑡 ≥ 0,
(80)
This implies that
where 𝑎∗ is a very small positive constant. Consider
󵄨󵄨 𝑡𝑛 𝑡𝑛 󵄨󵄨 𝑝(V)
󵄨󵄨 󵄨󵄨 𝐺 (𝑡, V) = ∫ 𝑎 (𝑢, V) 𝑑𝑢
󵄨󵄨𝑥 (𝑡𝑛 ) − ∫ 𝐸 (𝑡𝑛 , 𝑠) 𝑔 (𝑥 (𝑠)) 𝑑𝑠 − ∫ 𝐻 (𝑠) 𝑔 (𝑥 (𝑠)) 𝑑𝑠 󵄨󵄨
󵄨󵄨 ( 𝑛) 󵄨󵄨 𝑡
󵄨 𝑡𝑘 𝑡𝑛 −𝑟 𝑡 󵄨
2V
󵄨
− ∫ ℎ(𝑠)𝑑𝑠 󵄨󵄨󵄨 𝛿0
󵄨
󵄨󵄨
𝑡𝑛 ∫ ℎ(𝑠)𝑑𝑠
= ∫ 𝑎∗ 𝑒−(𝑢−V) 𝑑𝑢 = 𝑎∗ (𝑒V−𝑡 − 𝑒−V ) ,
𝑡𝑛 𝑢 𝑢
∫ 𝐴(𝑠)𝑑𝑠
≥ 𝑒 𝑡𝑘 [󵄨󵄨󵄨 + ∫ 𝑒 𝑡𝑘 𝑥̇ (𝑡𝑘 ) 𝑒 𝑡𝑘 𝑑𝑢󵄨󵄨󵄨󵄨 𝑡 (81)
󵄨󵄨 4 𝑡𝑘 󵄨󵄨
𝑡 𝑡
󵄨 󵄨
𝑡𝑛 𝑢
∫𝑡 ℎ(𝑠)𝑑𝑠 ∫ |𝐺 (𝑡, V)| 𝑑V = ∫ 𝑎∗ 󵄨󵄨󵄨󵄨𝑒V−𝑡 − 𝑒−V 󵄨󵄨󵄨󵄨 𝑑V
− ∫ 𝜔 (𝑢) 𝑒 𝑘 𝑑𝑢] . 𝑡−𝑟(𝑡) 𝑡/2
𝑡𝑘
(75) = 𝑎∗ (1 + 𝑒−𝑡 − 2𝑒−𝑡/2 ) .
Abstract and Applied Analysis 11

𝑡
This implies that ∫𝑡−𝑟(𝑡) |𝐺(𝑡, V)|𝑑V is increasing with respect Conflict of Interests
to 𝑡 for 𝑡 ≥ 0. Consider
The author declares that there is no conflict of interests
𝑝(𝑠) 2𝑠 regarding the publication of this paper.
𝐺 (𝑠, 𝑠) = ∫ 𝑎 (𝑢, 𝑠) 𝑑𝑢 = ∫ 𝑎∗ 𝑒−(𝑢−𝑠) 𝑑𝑢 = 𝑎∗ (1 − 𝑒−𝑠 ) .
𝑠 𝑠
(82) Acknowledgments
𝑡+𝐽
Set 𝑎(𝑡) = 2𝑡; we have ∫𝑡 2𝑠 𝑑𝑠 = 𝐽2 + 2𝑡𝐽 ≥ 𝐽2 for 𝐽 ≥ 0, The author is grateful to the reviewer for his or her useful
𝑡 ≥ 0. Consider suggestions. This work is partially supported by NNSF of
China Grant nos. 11226145 and 11271046 and a research
2𝑡 ∞ 𝑠+𝑤 foundation of Huaqiao University (12BS112).
∫ ∫ 𝑒− ∫𝑠 2V𝑑V
𝐺 (𝑠, 𝑠) 𝑑𝑤 𝑑𝑠
𝑡 0
2𝑡 ∞ 2
References
= ∫ ∫ 𝑒−𝑤 −2𝑤𝑠 𝐺 (𝑠, 𝑠) 𝑑𝑤 𝑑𝑠
𝑡 0 [1] T. A. Burton, Stability and Periodic Solutions of Ordinary and
2𝑡 ∞
Functional-Differential Equations, vol. 178, Academic Press,
≤ 𝑎∗ ∫ ∫ 𝑒−2𝑤𝑠 𝑑𝑤 𝑑𝑠 Orlando, Fla, USA, 1985.
𝑡 0 [2] J. Hale, Theory of Functional Differential Equations, Springer,
2𝑡 New York, NY, USA, 2nd edition, 1977.
1 1
= 𝑎∗ ∫ 𝑑𝑠 = ln 2, for 𝑡 > 0, (83) [3] T. A. Burton, Stability by Fixed Point Theory for Functional
𝑡 2𝑠 16 Differential Equations, Dover Publications Inc., Mineola, NY,
𝑡 ∞ 𝑠+𝑤 USA, 2006.
∫ ∫ 𝑒− ∫𝑠 2V𝑑V
𝐺 (𝑠, 𝑠) 𝑑𝑤 𝑑𝑠 [4] C. Jin and J. Luo, “Fixed points and stability in neutral differen-
0 𝑡−𝑠
tial equations with variable delays,” Proceedings of the American
𝑡 ∞ 2 Mathematical Society, vol. 136, no. 3, pp. 909–918, 2007.
= ∫ ∫ 𝑒−𝑤 −2𝑤𝑠 𝐺 (𝑠, 𝑠) 𝑑𝑤 𝑑𝑠 [5] S.-M. Jung, “A fixed point approach to the stability of an integral
0 𝑡−𝑠
equation related to the wave equation,” Abstract and Applied
𝑡 ∞ 2 Analysis, vol. 2013, Article ID 612576, 4 pages, 2013.
≤ 𝑎∗ ∫ ∫ 𝑒−𝑤 𝑒−2(𝑡−𝑠)𝑠 𝑑𝑤 𝑑𝑠.
0 𝑡−𝑠 [6] J. J. Levin and J. A. Nohel, “On a nonlinear delay equation,”
Journal of Mathematical Analysis and Applications, vol. 8, pp. 31–
Since 0 ≤ 𝑠 ≤ 𝑡, then 44, 1964.
∞ ∞ [7] T. A. Burton, “Fixed points and stability of a nonconvolution
2 2 √𝜋
∫ 𝑒−𝑤 𝑑𝑤 ≤ ∫ 𝑒−𝑤 𝑑𝑤 = . (84) equation,” Proceedings of the American Mathematical Society,
𝑡−𝑠 0 2 vol. 132, no. 12, pp. 3679–3687, 2004.
[8] L. C. Becker and T. A. Burton, “Stability, fixed points and
Thus, we have
inverses of delays,” Proceedings of the Royal Society of Edinburgh
𝑡 ∞ 𝑠+𝑤 A, vol. 136, no. 2, pp. 245–275, 2006.
∫ ∫ 𝑒− ∫𝑠 2V𝑑V
𝐺 (𝑠, 𝑠) 𝑑𝑤 𝑑𝑠 [9] C. Jin and J. Luo, “Stability of an integro-differential equation,”
0 𝑡−𝑠
Computers & Mathematics with Applications, vol. 57, no. 7, pp.
√𝜋 𝑡 −2(𝑡−𝑠)𝑠 1080–1088, 2009.
≤ 𝑎∗ ∫ 𝑒 𝑑𝑠 (85)
2 0 [10] N. T. Dung, “New stability conditions for mixed linear Levin-
Nohel integro-differential equations,” Journal of Mathematical
√𝜋 𝑡 2(𝑠−𝑡/2)2 −𝑡2 /2 √𝜋 Physics, vol. 54, no. 8, Article ID 082705, 11 pages, 2013.
= 𝑎∗ ∫ 𝑒 𝑑𝑠 ≜ 𝑎∗ 𝑀 (𝑡) .
2 0 2 [11] D. Pi, “Study the stability of solutions of functional differential
equations via fixed points,” Nonlinear Analysis: Theory, Methods
Easy computation shows that lim𝑡 → ∞ 𝑀(𝑡) = 0; thus, 𝑀(𝑡) is & Applications, vol. 74, no. 2, pp. 639–651, 2011.
bounded. Consider [12] J. J. Levin and J. A. Nohel, “Global asymptotic stability for
𝑡 𝑡 𝑠 nonlinear systems of differential equations and applications to
∫ 𝑒− ∫𝑠 2V𝑑V ∫ |𝐺 (𝑠, V)| 𝑑V 𝑑𝑠 reactor dynamics,” Archive for Rational Mechanics and Analysis,
0 𝑠/2 vol. 5, pp. 194–211, 1960.
𝑡 2
(86) [13] T. A. Burton, “Fixed points, stability, and exact linearization,”
∫0 𝑎∗ 𝑒𝑠 (1 + 𝑒−𝑠 − 2𝑒−𝑠/2 ) 𝑑𝑠 ∗ Nonlinear Analysis: Theory, Methods & Applications, vol. 61, no.
= 2 ≜ 𝑎 𝑁 (𝑡) .
𝑒𝑡 5, pp. 857–870, 2005.

Since lim𝑡 → ∞ 𝑁(𝑡) = 0, thus 𝑁(𝑡) is bounded. Hence,

conditions (ii) and (iii) in Theorem 2 hold. Choose 𝑔(𝑥) =
(1/3)𝑥3 . Let 𝑙 = 1; 𝑔(𝑥) satisfies the Lipschitz condition on
[−𝑙, 𝑙]. 𝑥 − 𝑔(𝑥) = 𝑥 − (1/3)𝑥3 , (𝑥 − 𝑔(𝑥))󸀠 = 1 − 𝑥2 ≥ 0
on [0, 1]. Thus, 𝑥 − 𝑔(𝑥) is nondecreasing on [0, 1]. Then by
Theorem 2 the zero solution of (13) is stable.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 934534, 11 pages
http://dx.doi.org/10.1155/2014/934534

Research Article
Asymptotically Almost Periodic Solutions for a Class of
Stochastic Functional Differential Equations

Aimin Liu,1 Yongjian Liu,2 and Qun Liu2

1
Educational Technology Center, Yulin Normal University, Yulin 537000, China
2
School of Mathematics and Information Science, Yulin Normal University, Yulin 537000, China

Correspondence should be addressed to Yongjian Liu; liuyongjianmaths@126.com

Received 7 February 2014; Revised 29 March 2014; Accepted 29 March 2014; Published 6 May 2014

Academic Editor: Yonghui Xia

Copyright © 2014 Aimin Liu et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

This work is concerned with the quadratic-mean asymptotically almost periodic mild solutions for a class of stochastic functional
differential equations d𝑥 (𝑡) = [𝐴 (𝑡) 𝑥 (𝑡) + 𝐹 (𝑡, 𝑥 (𝑡) , 𝑥𝑡 )] d𝑡 + 𝐻(𝑡, 𝑥 (𝑡) , 𝑥𝑡 ) ∘ d𝑊(𝑡). A new criterion ensuring the existence
and uniqueness of the quadratic-mean asymptotically almost periodic mild solutions for the system is presented. The condition of
being uniformly exponentially stable of the strongly continuous semigroup {𝑇 (𝑡)}𝑡≥0 is essentially removed, which is generated by
the linear densely defined operator 𝐴 : 𝐷(𝐴) ⊂ 𝐿2 (P, H) → 𝐿2 (P, H), only using the exponential trichotomy of the system, which
reflects a deeper analysis of the behavior of solutions of the system. In this case the asymptotic behavior is described through the
splitting of the main space into stable, unstable, and central subspaces at each point from the flow’s domain. An example is also
given to illustrate our results.

1. Introduction equations, many authors [1, 2] applied successfully the asymp-

totic property to determine the existence of almost periodic
The theory of almost periodic functions was first developed solutions. Along with the development of such equations
by the Danish mathematician H. Bohr in 1925-1926. Then as the evolution partial differential equations, functional
Bohr’s work was developed substantially by S. Bochner, J. differential equations, and so forth, where the phase spaces
Favard, V. V. Stepanov, and others. Generalization of the are infinite, the theory of Banach valued asymptotically
classical theory of almost periodic functions has been taken almost periodic functions had been developed [10–12]. Some
in several directions. These works were recapitulated in techniques in functional analysis and harmonic analysis were
literatures [1] and [2]. The concept of almost periodicity is applied to such equations; for example, in [1, 13], the authors
important in probability for investigating stochastic processes applied spectrum theory to get almost periodic solutions for
[3–7]. Such a notion is also of interest for applications some linear abstract evolution differential equations. More
arising in mathematical physics and statistics. Literature [8] recently, [14] developed the notion of 𝑝-mean asymptotical
developed the notion of 𝑝-mean almost periodicity based almost periodicity for stochastic processes. Among others,
on the concept of quadratic mean uniformly almost periodic it showed that each 𝑝-mean asymptotically almost periodic
utilized by [7] and pointed out that each 𝑝-mean almost stochastic process is stochastically bounded.
periodic process is uniformly continuous and stochastically Recently, [8] studied the existence and uniqueness of
bounded [9]. Literature [8] also pointed that the collection of quadratic-mean almost periodic solutions for the class of
all 𝑝-mean almost periodic processes is a Banach space when stochastic differential equations
it is equipped with some norm obtained through the norm of
𝐿𝑝 (P, B), where (B, ‖ ⋅ ‖) is a Banach space. d𝑥 (𝑡) = 𝐴𝑥 (𝑡) d𝑡 + 𝐹 (𝑡, 𝑥 (𝑡)) d𝑡 + 𝐺 (𝑡, 𝑥 (𝑡)) ∘ d𝑊 (𝑡) ,
The asymptotically almost periodic functions were first (1)
introduced by Fréchet. In the modern theory of differential 𝑡 ∈ R.
2 Abstract and Applied Analysis

Literature [15] investigated the existence and stability of employing the properties of almost periodic function and
quadratic-mean almost periodic mild solutions for stochastic the technique of inequality. We essentially remove the above
functional differential equations conditions (C) and only assume that the linear system (2)
admits exponential trichotomy (see Definition 11). We also
d𝑥 (𝑡) = [𝐴𝑥 (𝑡) + 𝐹 (𝑡, 𝑥 (𝑡) , 𝑥𝑡 )] d𝑡 point out that exponential trichotomy is the most complex
asymptotic property of dynamical systems arising from the
+ 𝐺 (𝑡, 𝑥 (𝑡) , 𝑥𝑡 ) ∘ d𝑊 (𝑡) , 𝑡 ∈ [0, 𝑇] , (2)
central manifold theory. Starting from the idea that the center
𝑥 (𝑡) = 𝜑 (𝑡) , 𝑡 ∈ [−𝜎, 0] . manifold of an equilibrium point of a dynamical system
consists of orbits whose behavior around the equilibrium
They both assumed that the strongly continuous semigroup point is not controlled by either the attraction of the stable
{𝑇(𝑡)}𝑡≥0 is uniformly exponentially stable, which is generated manifold or the repulsion of the unstable manifold. The
by the linear densely defined operator 𝐴 : 𝐷(𝐴) ⊂ exponential trichotomy reflects a deeper analysis of the
𝐿2 (P, H) → 𝐿2 (P, H). For other works, we refer the reader behavior of solutions of dynamical systems. In this case the
to [16–21] and the references therein. asymptotic behavior is described through the splitting of the
One should point that the following condition (C) is very main space into stable, unstable, and central subspaces at each
much important in the above-mentioned literatures. point from the flow’s domain.
This paper is organized as follows. In Section 2, the
(C) The operator 𝐴 is the generator of a uniformly relating notations, definitions, and the basic results are
exponentially stable semigroup {𝑇(𝑡)}𝑡≥0 such that introduced, which would be used later. In Section 3, a
there exist constants 𝑀 > 0, 𝛿 > 0 with ‖𝑇(𝑡)‖ ≤ new criterion ensuring the existence and uniqueness of a
𝑀𝑒−𝛿𝑡 , (𝑡 ≥ 0). quadratic-mean asymptotically almost periodic mild solution
It is clear that the condition (C) is too strict [22] so that it for stochastic functional differential equations is presented.
cannot be satisfied even if for simple 𝐴 = diag{1, −1} or 𝐴 = In Section 4, an example is given to illustrate our results.
diag{1, 0, −1}. Literature [22] presented some new criteria Finally, conclusions are drawn in Section 5.
ensuring the existence and uniqueness of quadratic-mean
almost periodic solution for stochastic differential equation 2. Preliminaries
(1), and only assumes that the linear system
Let (Ω, F, P) be a probability space, for a Banach space (B, ‖ ⋅
d𝑥 (𝑡) = 𝐴 (𝑡) d𝑡 (3) ‖) and 𝑝 ≥ 1, denoted by 𝐿𝑝 (P, B), the Banana space of all
B-value random variable 𝑥, such that
admits exponential dichotomy. It is clear that when 𝐴(𝑡) =
diag{1, −1}, the system (3) admits exponential dichotomy. E‖𝑥‖𝑝 = ∫ ‖𝑥‖𝑝 dP < ∞. (5)
More generally, in the case 𝐴(𝑡) ≡ 𝐴, a constant matrix, the Ω
system (3) admits exponential dichotomy if and only if the
It is then routine to check that 𝐿𝑝 (P, B) is a Banach space
eigenvalues of 𝐴 have a nonzero real part. Literature [14, 17]
when it is equipped with its natural norm ‖ ⋅ ‖𝑝 defined by
has obtained the existence and uniqueness of quadratic-mean
almost automorphic solutions or asymptotically almost peri- 1/𝑝
odic solutions for stochastic functional differential equations ‖𝑥‖𝐿𝑝 (P,B) = (∫ ‖𝑥‖𝑝 dP) (6)
Ω
under a hyperbolic and analytic semigroup {𝑇(𝑡)}𝑡≥0 . At the
same time, one notices that the case that the eigenvalues for each 𝑥 ∈ 𝐿𝑝 (P, B).
of 𝐴 have a zero real part is very common; for example, This setting requires the following preliminary defini-
𝐴(𝑡) = diag{1, 0, −1}. Therefore, it is interesting to ask, what is tions.
that, when the semigroup {𝑇(𝑡)}𝑡≥0 is not exponentially stable,
which is generated by the family {𝐴(𝑡) : 𝑡 ∈ R}, or when the Definition 1. A stochastic process 𝑥 : R → 𝐿𝑝 (P, B) is said
semigroup {𝑇(𝑡)}𝑡≥0 is nonhyperbolic? This question will be to be continuous whenever
considered in the paper. lim E‖𝑥(𝑡) − 𝑥(𝑠)‖𝑝 = 0. (7)
In the present paper, motivated by [8, 14, 15], we discuss 𝑡→𝑠

the existence and uniqueness of quadratic-mean asymptot- Definition 2. A stochastic process 𝑥 : R → 𝐿𝑝 (P, B) is said
ically almost periodic solution to the following stochastic to be stochastically bounded whenever
functional differential equation on 𝐿2 (P, H), where P is the
probability measure of the probability space (Ω, F, P) and lim supP {‖𝑥 (𝑡)‖ > 𝑁} = 0. (8)
𝑁 → ∞ 𝑡∈R
(H, ‖ ⋅ ‖) is a real separable Hilbert space:
Let CUB(R; 𝐿𝑝 (P, B)) denote the collection of all
d𝑥 (𝑡) = [𝐴 (𝑡) 𝑥 (𝑡) + 𝐹 (𝑡, 𝑥 (𝑡) , 𝑥𝑡 )] d𝑡 stochastic processes 𝑥 : R → 𝐿𝑝 (P, B), which are
(4) continuous and uniformly bounded. It is easily to check that
+ 𝐻 (𝑡, 𝑥 (𝑡) , 𝑥𝑡 ) ∘ d𝑊 (𝑡) , 𝑡 ∈ R.
CUB(R; 𝐿𝑝 (P, B)) is a Banach space when it is equipped
We present the new criterion ensuring the existence of a with the norm
unique quadratic-mean asymptotically almost periodic solu- 1/𝑝
‖𝑥‖∞ = sup(E‖𝑥(𝑡)‖𝑝 ) . (9)
tion for the stochastic functional differential equation (1), by 𝑡∈R
Abstract and Applied Analysis 3

Definition 3. A continuous stochastic process 𝑥 : R → By 𝐴𝐴𝑃(R × 𝐿𝑝 (P, B1 ); 𝐿𝑝 (P, B2 )) we denote the collec-
𝐿𝑝 (P, B) is said to be 𝑝-mean almost periodic if for each tion of such function.
𝜀 > 0 there exists 𝑙(𝜀) > 0 such that any interval of length
𝑙(𝜀) contains at least a number 𝜏 for which Lemma 8 (see [14]). If 𝑋 belongs to 𝐴𝐴𝑃(R; 𝐿𝑝 (P, B)), one
has the following:
supE‖𝑥(𝑡 + 𝜏) − 𝑥(𝑡)‖𝑝 < 𝜀. (10)
𝑡∈R
(1) there exists a constant 𝑀 > 0 such that E‖𝑋(𝑡)‖𝑝 ≤ 𝑀
The collection of all stochastic processes 𝑥 : R → for each 𝑡 ∈ R;
𝐿𝑝 (P, B) which are 𝑝-mean almost periodic is denoted by (2) 𝑋 is stochastically bounded.
𝐴𝑃(R; 𝐿𝑝 (P, B)). 𝐴𝑃(R; 𝐿𝑝 (P, B)) is a closed subspace of
CUB(R; 𝐿𝑝 (P, B)). Therefore, 𝐴𝑃(R; 𝐿𝑝 (P, B)) is a Banach
It is easy to see that 𝐴𝐴𝑃(R; 𝐿𝑝 (P, B)) = 𝐴𝑃(R; 𝐿𝑝 (P, B))
space when it is equipped with the norm ‖ ⋅ ‖∞ (see, e.g., [3]).
⊕𝐶0 (R; 𝐿𝑝 (P, B)). Then the space 𝐴𝐴𝑃(R; 𝐿𝑝 (P, B)) of 𝑝-
Let (B1 , ‖ ⋅ ‖1 ), (B2 , ‖ ⋅ ‖2 ) be two Banach spaces and
mean asymptotically almost periodic stochastic processes is a
𝐿𝑝 (P, B1 ), 𝐿𝑝 (P, B2 ) their corresponding 𝐿𝑝 -spaces, respec-
Banach space when it is equipped with the norm ‖ ⋅ ‖∞ .
tively.
Lemma 9 (see [14]). Let 𝑥 ∈ 𝐴𝑃(R+ ; 𝐿𝑝 (P, B1 )), 𝐾 =
Definition 4 (see [8]). A function 𝑓 : R × 𝐿𝑝 (P, B1 ) →
𝐿𝑝 (P, B2 ), (𝑡, 𝑥) → 𝑓(𝑡, 𝑥), which is jointly continuous, is {(𝑥(𝑡), 𝑥𝑡 ) : 𝑡 ∈ R+ }, where 𝑥𝑡 = {𝑥(𝑡 + ℎ) : −𝜎 < ℎ < 0} is
said to be 𝑝-mean almost periodic in 𝑡 ∈ R uniformly in regarded as a 𝐶([−𝜎, 0], 𝐿𝑝 (P, B1 )) valued stochastic process.
𝑥 ∈ K where K ⊂ 𝐿𝑝 (P, B1 ) is compact if, for any 𝜀 > 0, Moreover
there exists 𝑙(𝜀, K) > 0 such that any interval of length 𝑙(𝜀, K)
contains at least a number 𝜏 for which 𝑓 ∈ 𝐴𝑃 (R+ × 𝐿𝑝 (P, B1 ))

󵄩 󵄩𝑝 1/𝑝 × 𝐶 ([−𝜎, 0] , 𝐿𝑝 (P, B1 ) ; 𝐿𝑝 (P, B2 )) ,

sup(E󵄩󵄩󵄩𝑓(𝑡 + 𝜏, 𝑥) − 𝑓(𝑡, 𝑥)󵄩󵄩󵄩 ) < 𝜀 (11)
𝑡∈R
(12)
𝑓 ∈ 𝐶𝐾 (R+ × 𝐿𝑝 (P, B1 ))
for each stochastic process 𝑥 : R → K.
× 𝐶 ([−𝜎, 0] , 𝐿𝑝 (P, B1 ) ; 𝐿𝑝 (P, B2 )) .
Theorem 5. Let 𝑓 : R × 𝐿𝑝 (P, B1 ) → 𝐿𝑝 (P, B2 ), (𝑡, 𝑥) →
𝑓(𝑡, 𝑥), be a 𝑝-mean almost periodic process in 𝑡 ∈ R uniformly Then 𝑓(⋅, 𝑥(⋅), 𝑥) ∈ 𝐴𝑃(R+ ; 𝐿𝑝 (P, B2 )).
in 𝑥 ∈ K, where K ⊂ 𝐿𝑝 (P, B1 ) is compact. Then for every
real sequence 𝛼󸀠 , there exists 𝛽 ⊂ 𝛼󸀠 and a continuous function Lemma 10 (see [14]). Let 𝑥 ∈ 𝐴𝐴𝑃(R+ ; 𝐿𝑝 (P, B1 )), 𝐾 =
𝑔(𝑡, 𝑥) such that 𝑇𝛽 𝑓(𝑡, 𝑥) = 𝑔(𝑡, 𝑥) exists uniformly on R × K. {(𝑥(𝑡), 𝑥𝑡 ) : 𝑡 ∈ R+ }, where x𝑡 = {𝑥(𝑡 + ℎ) : −𝜎 < ℎ < 0} is
Further, 𝑔(𝑡, 𝑥) is also 𝑝-mean almost periodic process in 𝑡 ∈ R regarded as a 𝐶([−𝜎, 0], 𝐿𝑝 (P, B1 )) valued stochastic process.
uniformly in 𝑥 ∈ K. Moreover
The space of the restrictions of all 𝑝-mean almost periodic
𝑓 ∈ 𝐴𝐴𝑃 (R+ × 𝐿𝑝 (P, B1 ))
stochastic processes on R is denoted by 𝐴𝑃(R; 𝐿𝑝 (P, B)),
and 𝑝-mean almost periodic stochastic processes in 𝑡, uni- × 𝐶 ([−𝜎, 0] , 𝐿𝑝 (P, B1 ) ; 𝐿𝑝 (P, B2 )) ,
formly for 𝑥 in compact subset K of 𝐿𝑝 (P, B1 ) by A𝑃(R × (13)
𝐿𝑝 (P, B1 ); 𝐿𝑝 (P, B2 )). 𝑓 ∈ 𝐶𝐾 (R+ × 𝐿𝑝 (P, B1 ))
Denote by 𝐶0 (R; 𝐿𝑝 (P, B)) the space of all continuous
stochastic processes 𝑍 : R → 𝐿𝑝 (P, B) such that lim𝑡 → +∞ E × 𝐶 ([−𝜎, 0] , 𝐿𝑝 (P, B1 ) ; 𝐿𝑝 (P, B2 )) .
‖𝑍(𝑡)‖𝑝 = 0, and denote by 𝐶0 (R × 𝐿𝑝 (P, B1 ); 𝐿𝑝 (P, B2 ))
the space of all continuous functions 𝑍 : R × 𝐿𝑝 (P, B1 ) → Then 𝑓(⋅, 𝑥(⋅), 𝑥) ∈ 𝐴𝐴𝑃(R+ ; 𝐿𝑝 (P, B2 )).
𝑝
𝐿𝑝 (P, B2 ) such that lim𝑡 → +∞ E‖𝑍(𝑡, 𝑥)‖2 = 0 uniformly for 𝑥
𝑝
in any compact subset of 𝐿 (P, B1 ). Next, one introduces a crucial concept [23].
Suppose that 𝑈(𝑡) is the fundamental matrix solution of
Definition 6. A stochastic process 𝑋 : 𝑅 → 𝐿𝑝 (P, B) is the linear differential system
called 𝑝-mean asymptotically almost periodic if there exist
two stochastic processes 𝑌 ∈ 𝐴𝑃(R; 𝐿𝑝 (P, B)) and 𝑍 ∈
d𝑥
𝐶0 (R; 𝐿𝑝 (P, B)) such that 𝑋 = 𝑌 + 𝑍. = 𝐴 (𝑡) 𝑥, (14)
d𝑡
By 𝐴𝐴𝑃(R; 𝐿𝑝 (P, B)) one denotes the collection of all 𝑝-
mean asymptotically almost periodic stochastic processes. where 𝐴(𝑡) is a linear continuous operator, with 𝑈(0) = 𝐼.

Definition 7. A stochastic process 𝑋 : 𝑅 × 𝐿𝑝 (P, B1 ) → 𝐿𝑝 Definition 11 (see [23]). System (14) is said to admit exponen-
(P, B2 ) is called 𝑝-mean asymptotically periodic if there exist tial trichotomy if there are linear projections 𝑃, 𝑄 such that
two stochastic processes 𝑌 ∈ 𝐴𝑃(R × 𝐿𝑝 (P, B1 ); 𝐿𝑝 (P, B2 ))
and 𝑍 ∈ 𝐶0 (R × 𝐿𝑝 (P, B1 ); 𝐿𝑝 (P, B2 )) such that 𝑋 = 𝑌 + 𝑍. 𝑃𝑄 = 𝑄𝑃, 𝑃 + 𝑄 − 𝑃𝑄 = 𝐼, (15)
4 Abstract and Applied Analysis

and constants 𝛼 > 0 and 𝛽 ≥ 1 such that has the fundamental matrix

𝑈𝑛 (𝑡) = 𝑈 (𝑡 + 𝑎𝑛 ) 𝑈−1 (𝑎𝑛 ) ,

󵄩󵄩 󵄩 󵄩󵄩 󵄩
󵄩󵄩𝑈(𝑡)𝑃𝑈−1 (𝑠)󵄩󵄩󵄩 ≤ 𝛽 exp (−𝛼 (𝑡 − 𝑠)) , 󵄩󵄩𝑈𝑛 (𝑡)𝑃𝑛 𝑈𝑛−1 (𝑠)󵄩󵄩󵄩 ≤ 𝛽 exp (−𝛼 (𝑡 − 𝑠)) ,
󵄩 󵄩𝑝 󵄩 󵄩𝑝
(16)
for 0 ≤ 𝑠 ≤ 𝑡, 𝑡 ≥ 𝑠 ≥ −𝑎𝑛 ,
󵄩󵄩 󵄩
󵄩󵄩 󵄩 󵄩󵄩𝑈𝑛 (𝑡)(𝐼 − 𝑃𝑛 )𝑈𝑛−1 (𝑠)󵄩󵄩󵄩 ≤ 𝛽 exp (𝛼 (𝑡 − 𝑠)) ,
󵄩󵄩𝑈(𝑡)(𝐼 − 𝑃)𝑈−1 (𝑠)󵄩󵄩󵄩 ≤ 𝛽 exp (−𝛼 (𝑠 − 𝑡)) , 󵄩 󵄩𝑝
󵄩 󵄩𝑝
(17) 𝑠 ≥ 𝑡 ≥ −𝑎𝑛 , (23)
for 𝑡 ≤ 𝑠, 𝑠 ≥ 0,
󵄩󵄩 󵄩
󵄩󵄩𝑈𝑛 (𝑡)𝑄𝑛 𝑈𝑛−1 (𝑠)󵄩󵄩󵄩 ≤ 𝛽 exp (−𝛼 (𝑡 − 𝑠)) ,
󵄩󵄩 󵄩 󵄩 󵄩𝑝
󵄩󵄩𝑈(𝑡)𝑄𝑈−1 (𝑠)󵄩󵄩󵄩 ≤ 𝛽 exp (−𝛼 (𝑠 − 𝑡)) ,
󵄩 󵄩𝑝
(18) 𝑡 ≤ 𝑠 ≤ 𝑎𝑛 ,
for 𝑡 ≤ 𝑠 ≤ 0, 󵄩󵄩 󵄩
󵄩󵄩𝑈𝑛 (𝑡)(𝐼 − 𝑄𝑛 )𝑈𝑛−1 (𝑠)󵄩󵄩󵄩 ≤ 𝛽 exp (𝛼 (𝑡 − 𝑠)) ,
󵄩󵄩 󵄩 󵄩 󵄩𝑝
󵄩󵄩𝑈(𝑡)(𝐼 − 𝑄)𝑈−1 (𝑠)󵄩󵄩󵄩 ≤ 𝛽 exp (−𝛼 (𝑡 − 𝑠)) ,
󵄩 󵄩𝑝 𝑠 ≤ 𝑡 ≤ 𝑎𝑛 ,
(19)
for 𝑠 ≤ 𝑡, 𝑠 ≤ 0. where 𝑃𝑛 = 𝑈(𝑎𝑛 )𝑃𝑈−1 (𝑎𝑛 ), 𝑄𝑛 = 𝑈(𝑎𝑛 )𝑄𝑈−1 (𝑎𝑛 ). Since
|𝑃𝑛 | ≤ 𝛽, |𝑄𝑛 | ≤ 𝛽 by restricting attention to a subsequence
we can assume that 𝑃𝑛 → 𝑃∗ , 𝑄𝑛 → 𝑄∗ , where 𝑃∗ , 𝑄∗ are
If in the above definition we put 𝑄 = 𝐼 − 𝑃, (16) becomes projections. Since 𝑈𝑛 (𝑡) → 𝑉(𝑡) for every 𝑡, where 𝑉(𝑡) is the
fundament matrix of (21) such that 𝑉(0) = 𝐼, it follows that
󵄩󵄩 󵄩
󵄩󵄩𝑉𝑛 (𝑡)𝑃∗ 𝑉𝑛−1 (𝑠)󵄩󵄩󵄩 ≤ 𝛽 exp (−𝛼 (𝑡 − 𝑠)) ,
󵄩󵄩 󵄩 󵄩 󵄩𝑝
󵄩󵄩𝑈(𝑡)𝑃𝑈−1 (𝑠)󵄩󵄩󵄩 ≤ 𝛽 exp (−𝛼 (𝑡 − 𝑠)) , for 𝑠 ≤ 𝑡,
󵄩 󵄩𝑝
(20) 𝑡 ≥ 𝑠 ≥ 0,
󵄩󵄩 󵄩
󵄩󵄩𝑈(𝑡)(𝐼 − 𝑃)𝑈 (𝑠)󵄩󵄩󵄩 ≤ 𝛽 exp (𝛼 (𝑡 − 𝑠)) , for 𝑡 ≤ 𝑠.
−1
󵄩󵄩 󵄩
󵄩 󵄩𝑝 󵄩󵄩𝑉𝑛 (𝑡)(𝐼 − 𝑃∗ )𝑉𝑛−1 (𝑠)󵄩󵄩󵄩 ≤ 𝛽 exp (𝛼 (𝑡 − 𝑠)) ,
󵄩 󵄩𝑝
𝑠 ≥ 𝑡, 𝑠 ≥ 0,
It is clear that when 𝐴(𝑡) = − tanh 𝑡, the system (14) 󵄩󵄩 󵄩 (24)
admits exponential trichotomy. More generally, in the case 󵄩󵄩𝑉𝑛 (𝑡)𝑄∗ 𝑉𝑛−1 (𝑠)󵄩󵄩󵄩 ≤ 𝛽 exp (−𝛼 (𝑡 − 𝑠)) ,
󵄩 󵄩𝑝
𝐴(𝑡) ≡ 𝐴 = diag{1, 0, −1}, a constant matrix, the system (14)
admits exponential trichotomy. 𝑡 ≤ 𝑠 ≤ 0,
󵄩󵄩 󵄩
Theorem 12. Let 𝐴(𝑡) be a continuous linear operator on 󵄩󵄩𝑉𝑛 (𝑡)(𝐼 − 𝑄∗ )𝑉𝑛−1 (𝑠)󵄩󵄩󵄩 ≤ 𝛽 exp (𝛼 (𝑡 − 𝑠)) ,
󵄩 󵄩𝑝
R and let (14) have an exponential trichotomy (16) on R+ .
If for some sequence 𝑎 = {𝑎𝑛 }, lim𝑛 → +∞ 𝑎𝑛 = +∞, 𝑠 ≥ 𝑡, 𝑠 ≤ 0.
𝑇𝑎 𝐴(𝑡) = 𝐵(𝑡) uniformly on compact subintervals of R,
lim𝑛 → +∞ 𝑈(𝑎𝑛 )𝑃𝑈−1 (𝑎𝑛 ) = 𝑃∗ , lim𝑛 → +∞ 𝑈(𝑎𝑛 )𝑄𝑈−1 (𝑎𝑛 ) = Since the projection corresponding to an exponential
𝑄∗ , and the equation (which is called the hull equation (14)) trichotomy on R is uniquely determined it follows that 𝑃𝑛 →
𝑃∗ , 𝑄𝑛 → 𝑄∗ without restriction to a subsequence.

d𝑦 3. Existence of Asymptotically Almost

= 𝐵 (𝑡) 𝑦 (21) Periodic Solutions
d𝑡
For convenience, throughout this section, let (H, ‖ ⋅ ‖) be a
real separable Hilbert space, and let (Ω, F, P) be a complete
has an exponential trichotomy on R with projections 𝑃∗ , 𝑄∗ probability space equipped with a normal filtration {F𝑡 :
and the same constants 𝛼, 𝛽. 𝑡 ∈ R}, that is, a right-continuous, increasing family of
sub-𝜎-algebras of F. Let L(𝐿2 (P, H)) be the space of all
Proof. The translated equation bounded linear operators from 𝐿2 (P, H) to 𝐿2 (P, H) and 𝐶 =
𝐶([−𝜎, 0], 𝐿2 (P, H)) the space of all continuous functions
from [−𝜎, 0] to 𝐿2 (P, H) with the sup norm
d𝑥 󵄩󵄩 󵄩󵄩 󵄩 󵄩
= 𝐴 (𝑡 + 𝑎𝑛 ) 𝑥 (22) 󵄩󵄩𝜓󵄩󵄩𝐶 = sup {󵄩󵄩󵄩𝜓(𝑠)󵄩󵄩󵄩𝐿2 (P,H) : 𝜓 ∈ 𝐶, −𝜎 ≤ 𝑠 ≤ 0} . (25)
d𝑡
Abstract and Applied Analysis 5

For any continuous F𝑡 -adapted 𝐿2 (P, H)-valued stochas- Definition 13. A F𝑡 -progressive process {𝑥(𝑡)}𝑡∈R is called a
tic process 𝑥(𝑡) : Ω → 𝐿2 (P, H), 𝑡 ≥ −𝜎, we associate it mild solution of (26) on R if
with a continuous F𝑡 -adapted 𝐶-valued stochastic process
𝑡
𝑥𝑡 : Ω → 𝐶, 𝑡 ≥ 0, by setting 𝑥𝑡 (𝑠)(𝜔) = 𝑥(𝑡 + 𝑠)(𝜔), 𝑠 ∈
𝑥 (𝑡) = 𝐺 (𝑡, 𝑠) 𝑥 (𝑠) + ∫ 𝐺 (𝑡, 𝜏) 𝐹 (𝜏, 𝑥 (𝜏) , 𝑥𝜏 ) d𝜏
[−𝜎, 0]. 𝑠
In this section, we study the existence and uniqueness of (31)
𝑡
quadratic-mean asymptotically almost periodic mild solution + ∫ 𝐺 (𝑡, 𝜏) 𝐻 (𝜏, 𝑥 (𝜏) , 𝑥𝜏 ) ∘ d𝑊 (𝜏) ,
to stochastic functional differential equations of the form 𝑠

d𝑥 (𝑡) = [𝐴 (𝑡) 𝑥 (𝑡) + 𝐹 (𝑡, 𝑥 (𝑡) , 𝑥𝑡 )] d𝑡 for all 𝑡 ≥ 𝑠, for each 𝑠 ∈ R, where 𝐺(𝑡, 𝑠) is defined by (30).
(26)
+ 𝐻 (𝑡, 𝑥 (𝑡) , 𝑥𝑡 ) ∘ d𝑊 (𝑡) , 𝑡 ∈ R, Theorem 14. Assume that conditions (H1)–(H3) are satisfied.
And the positive constants 𝛼, 𝛽, 𝐹𝑙 , and 𝐻𝑙 satisfy the following
where 𝐴(𝑡) is a linear operator and 𝐷(𝐴) ⊂ 𝐿2 (P, H) → condition:
𝐿2 (P, H) generates a strongly continuous semigroup
{𝑇(𝑡)}𝑡≥0 , which is nonhyperbolic. That is to say, the linear 8𝛽 𝛼𝐻𝑙
operator 𝐴(𝑡) may exhibit central flow. 𝑊(𝑡) is a certain Λ= √ 𝐹𝑙 + < 1. (32)
𝛼 2
𝑄-Wiener process with covariance operator 𝑄 and takes
its values on 𝐿2 (P, H). 𝐹 : R × 𝐿2 (P, H) × 𝐶 → 𝐿2 (P, H) Then the system (26) has a unique quadratic-mean asymp-
and 𝐻 : R × 𝐿2 (P, H) × 𝐶 → 𝐿2 (P, L(𝐿2 (P, H))) are two totically almost periodic mild solution, which can be explicitly
continuous mappings. expressed as follows:
Throughout this section, we require the following
assumptions. 4 4
2 2 𝑥 (𝑡) = ∑Φ𝑖 𝑥 (𝑡) + ∑Ψ𝑖 𝑥 (𝑡) , (33)
(H1) Suppose that 𝐹 ∈ 𝐴𝐴𝑃(R × 𝐿 (P, H) × 𝐶; 𝐿 (P, H)).
𝑖=1 𝑖=1
Furthermore, there exists a constant 𝐹𝑙 > 0 such that
󵄩󵄩 󵄩2 where
󵄩󵄩𝐹(𝑡, 𝑥1 , 𝑥2 ) − 𝐹(𝑡, 𝑦1 , 𝑦2 )󵄩󵄩󵄩𝐿2 (P,H)
(27) 𝑡
󵄩 󵄩2 󵄩 󵄩2 Φ1 𝑥 (𝑡) := ∫ 𝑈 (𝑡) 𝑃𝑈−1 (𝑠) 𝐹 (𝑠, 𝑥 (𝑠) , 𝑥𝑠 ) d𝑠,
≤ 𝐹𝑙 (󵄩󵄩󵄩𝑥1 − 𝑦1 󵄩󵄩󵄩𝐿2 (P,H) + 󵄩󵄩󵄩𝑥2 − 𝑦2 󵄩󵄩󵄩𝐶)
0

for all stochastic processes (𝑥1 , 𝑥2 ), (𝑦1 , 𝑦2 ) ∈ 𝐿2 (P, max{𝑡,0}

H) × 𝐶 and for each 𝑡 ∈ R. Φ2 𝑥 (𝑡) := ∫ 𝑈 (𝑡) (𝐼 − 𝑃) 𝑈−1 (𝑠) 𝐹 (𝑠, 𝑥 (𝑠) , 𝑥𝑠 ) d𝑠,
+∞
(H2) Suppose that 𝐻 ∈ 𝐴𝐴𝑃(R × 𝐿2 (P, H) × 𝐶; 𝐿2 (P, L(𝐿2 𝑡
(P, H))). Furthermore, there exists a constant 𝐻𝑙 > 0 Φ3 𝑥 (𝑡) := ∫ 𝑈 (𝑡) 𝑄𝑈−1 (s) 𝐹 (𝑠, 𝑥 (𝑠) , 𝑥𝑠 ) d𝑠,
such that 0

󵄩󵄩󵄩𝐻(𝑡, 𝑥1 , 𝑥2 ) − 𝐻(𝑡, 𝑦1 , 𝑦2 )󵄩󵄩󵄩2 2

min{𝑡,0}
󵄩 󵄩𝐿 (P,L(𝐿2 (P,H)) Φ4 𝑥 (𝑡) := ∫ 𝑈 (𝑡) (𝐼 − 𝑄) 𝑈−1 (𝑠) 𝐹 (𝑠, 𝑥 (𝑠) , 𝑥𝑠 ) d𝑠,
(28) −∞
󵄩 󵄩2 󵄩 󵄩2
≤ 𝐻𝑙 (󵄩󵄩󵄩𝑥1 − 𝑦1 󵄩󵄩󵄩𝐿2 (P,H) + 󵄩󵄩󵄩𝑥2 − 𝑦2 󵄩󵄩󵄩𝐶) , 𝑡
Ψ1 𝑥 (𝑡) := ∫ 𝑈 (𝑡) 𝑃𝑈−1 (𝑠) 𝐻 (𝑠, 𝑥 (𝑠) , 𝑥𝑠 ) ∘ d𝑊 (𝑠) ,
for all stochastic processes (𝑥1 , 𝑥2 ), (𝑦1 , 𝑦2 ) ∈ 0

𝐿2 (P, H) × 𝐶 and for each 𝑡 ∈ R. max{𝑡,0}

Ψ2 𝑥 (𝑡) := ∫ 𝑈 (𝑡) (𝐼 − 𝑃) 𝑈−1 (𝑠)
(H3) Suppose that the linear system of (26) +∞

d𝑥 (𝑡) = 𝐴 (𝑡) d𝑡 (29) × 𝐻 (𝑠, 𝑥 (𝑠) , 𝑥𝑠 ) ∘ d𝑊 (𝑠) ,

admits exponential trichotomy (see Definition 11); 𝑡

that is, there exist constants 𝛼 > 0, 𝛽 ≥ 1 such that Ψ3 𝑥 (𝑡) := ∫ 𝑈 (𝑡) 𝑄𝑈−1 (𝑠) 𝐻 (𝑠, 𝑥 (𝑠) , 𝑥𝑠 ) ∘ d𝑊 (𝑠) ,
0
(16) holds.
min{𝑡,0}
Define the function 𝐺(𝑡, 𝑠) as the form Ψ4 𝑥 (𝑡) := ∫ 𝑈 (𝑡) (𝐼 − 𝑄) 𝑈−1 (𝑠)
−∞

{ 𝑈 (𝑡) 𝑃𝑈−1 (𝑠) , 0 ≤ 𝑠 ≤ 𝑡, × 𝐻 (𝑠, 𝑥 (𝑠) , 𝑥𝑠 ) ∘ d𝑊 (𝑠) .

{
{
{−𝑈 (𝑡) (𝐼 − 𝑃) 𝑈−1 (𝑠) , max {𝑡, 0} ≤ 𝑠, (34)
𝐺 (𝑡, 𝑠) = { (30)
{ −𝑈 𝑄𝑈−1 (𝑠) , 0 ≥ 𝑠 ≥ 𝑡,
{ (𝑡)
{
−1
{𝑈 (𝑡) (𝐼 − 𝑄) 𝑈 (𝑠) , min {𝑡, 0} ≥ 𝑠, We show that (26) exists as a mild solution. Note that (33)
and (34) are well defined for each 𝑡 ∈ R and satisfy (31) for all
where 𝑈(𝑡) is the fundamental matrix solution of the linear 𝑡 ≥ 𝑠, for each 𝑠 ∈ R. Hence 𝑥(𝑡) given by (33) and (34) is a
differential system (29) with 𝑈(0) = 𝐼. mild solution of (26).
6 Abstract and Applied Analysis

1/2
Define a mapping L on 𝐶(R, 𝐿2 (P, H)) by in the sense of norm ‖ ⋅ ‖𝐿2 (P,H) = (∫Ω ‖ ⋅ ‖2 dP) uniformly
for 𝑡 ∈ R. Then
4 4
(L𝑥) (𝑡) = ∑Φ𝑖 𝑥 (𝑡) + ∑Ψ𝑖 𝑥 (𝑡) := (𝐵1 𝑥) (𝑡) + (𝐵2 𝑥) (𝑡) , 𝛼 (𝑡 + 𝛼𝑘 )
𝑖=1 𝑖=1
𝑡+𝛼𝑘
(35)
=∫ 𝑈 (𝑡 + 𝛼𝑘 ) 𝑃𝑈−1 (𝑠) 𝑘 (𝑠, 𝑥 (𝑠) , 𝑥𝑠 ) d𝑠
0
where the Φ𝑖 , Ψ𝑖 are defined in (34). 𝑡
In order to prove Theorem 14, we first prove Lemmas 15 = ∫ 𝑈 (𝑡 + 𝛼𝑘 ) 𝑃𝑈−1 (𝑠 + 𝛼𝑘 ) 𝑘 (𝑠 + 𝛼𝑘 , 𝑥 (𝑠 + 𝛼𝑘 ) , 𝑥𝑠+𝛼𝑘 ) d𝑠.
0
and 16. (39)
Lemma 15. Assume that conditions (H1)–(H3) are satisfied.
From (37), (38), (39), and Lebesgue’s control convergence
Then the operator L maps 𝐴𝐴𝑃(R, 𝐿2 (P, H)) into itself. theorem, one sees that 𝛼(𝑡 + 𝛼𝑘 ) converges to
Proof. First, let us show that 𝐵1 𝑥 is quadratic-mean asymp- 𝑡
totically almost periodic whenever 𝑥 ∈ 𝐴𝐴𝑃(R, 𝐿2 (P, H)) ∫ 𝑉 (𝑡) 𝑃∗ 𝑉−1 (𝑠) 𝑘1 (𝑠, 𝑥 (𝑠) , 𝑥𝑠 ) d𝑠 (40)
0
dose.
Indeed, assuming that 𝑥 is quadratic-mean asymp- uniformly for 𝑡 ∈ R. Hence 𝛼(𝑡) ∈ 𝐴𝑃(R, 𝐿2 (P, H)).
totically almost periodic and using condition (H1) and
Next, let us show that 𝛽(𝑡) ∈ 𝐶0 (R, 𝐿2 (P, H)). In fact
Lemma 10, one can easily see that 𝑠 󳨃 → 𝐹(𝑠, 𝑥(𝑠), 𝑥𝑠 ) is
quadratic-mean asymptotically almost periodic. Therefore, 󵄩 󵄩2
there exist an almost periodic stochastic process 𝑘(𝑠, 𝑥(𝑠), 𝑥𝑠 ) E󵄩󵄩󵄩𝛽(𝑡)󵄩󵄩󵄩
and a stochastic process ℎ ∈ 𝐶0 (R × 𝐿2 (P, H), 𝐿2 (P, H)) such 󵄩󵄩 𝑡 󵄩󵄩2
󵄩 󵄩
that 𝐹 = 𝑘 + ℎ. Furthermore, one observes that = E󵄩󵄩󵄩∫ 𝑈(𝑡)𝑃𝑈−1 (𝑠)ℎ(𝑠, 𝑥(𝑠), 𝑥𝑠 )d𝑠󵄩󵄩󵄩
󵄩󵄩 0 󵄩󵄩
𝑡
𝑡 2
Φ1 𝑥 (𝑡) := ∫ 𝑈 (𝑡) 𝑃𝑈−1 (𝑠) 𝑘 (𝑠, 𝑥 (𝑠) , 𝑥𝑠 ) d𝑠 󵄩 󵄩
0 ≤ 𝛽2 E(∫ 𝑒−𝛼(𝑡−𝑠) 󵄩󵄩󵄩ℎ(𝑠, 𝑥(𝑠), 𝑥𝑠 )󵄩󵄩󵄩 d𝑠)
0
𝑡
−1 (36) 2
+ ∫ 𝑈 (𝑡) 𝑃𝑈 (𝑠) ℎ (𝑠, 𝑥 (𝑠) , 𝑥𝑠 ) d𝑠 𝑡
󵄩 󵄩
0 = 𝛽2 E(∫ 𝑒−(𝛼(𝑡−𝑠))/2 𝑒−(𝛼(𝑡−𝑠))/2 󵄩󵄩󵄩ℎ(𝑠, 𝑥(𝑠), 𝑥𝑠 )󵄩󵄩󵄩 d𝑠)
0
:= 𝛼 (𝑡) + 𝛽 (𝑡) . 𝑡 𝑡
󵄩 󵄩2
≤ 𝛽2 E [(∫ 𝑒−𝛼(𝑡−𝑠) d𝑠) (∫ 𝑒−𝛼(𝑡−𝑠) 󵄩󵄩󵄩ℎ(𝑠, 𝑥(𝑠), 𝑥𝑠 )󵄩󵄩󵄩 d𝑠)]
We claim that 𝛼(𝑡) ∈ 𝐴𝑃(R, 𝐿2 (P, H)). In fact, 𝑠 󳨃 → 𝑘(𝑠, 0 0

𝑥(𝑠), 𝑥𝑠 ) is quadratic-mean almost periodic. Therefore, one 𝑡 𝑡

󵄩 󵄩2
can find a common subsequence {𝛼𝑛 }. One still denotes it as = 𝛽2 (∫ 𝑒−𝛼(𝑡−𝑠) d𝑠) (∫ 𝑒−𝛼(𝑡−𝑠) E󵄩󵄩󵄩ℎ(𝑠, 𝑥(𝑠), 𝑥𝑠 )󵄩󵄩󵄩 d𝑠)
0 0
𝛼 = {𝛼𝑛 }, such that
𝑡 2
󵄩 󵄩2
𝑇𝛼 𝑘 (𝑡, 𝑥 (𝑡) , 𝑥𝑡 ) = 𝑘1 (𝑡, 𝑥 (𝑡) , 𝑥𝑡 ) , (37) ≤ 𝛽2 (∫ 𝑒−𝛼(𝑡−𝑠) d𝑠) supE󵄩󵄩󵄩ℎ(𝑡, 𝑥(𝑡), 𝑥𝑡 )󵄩󵄩󵄩
0 𝑡∈R

uniformly for 𝑡 ∈ R, and 𝛽2 2 󵄩 󵄩2

= 2
(1 − 𝑒−𝛼𝑡 ) supE󵄩󵄩󵄩ℎ(𝑡, 𝑥(𝑡), 𝑥𝑡 )󵄩󵄩󵄩 .
𝛼 𝑡∈R
lim 𝑈 (𝑡 + 𝛼𝑘 ) 𝑃𝑈−1 (𝑠 + 𝛼𝑘 ) = 𝑉 (𝑡) 𝑃∗ 𝑉−1 (𝑠) , (41)
𝑘→∞

0 ≤ 𝑠 ≤ 𝑡, Since ℎ ∈ 𝐶0 (R × 𝐿2 (P, H), 𝐿2 (P, H)), we deduce that

lim𝑡 → +∞ E‖𝛽(𝑡)‖2 = 0.
lim 𝑈 (𝑡 + 𝛼𝑘 ) (𝐼 − 𝑃) 𝑈−1 (𝑠 + 𝛼𝑘 ) = 𝑉 (𝑡) (𝐼 − 𝑃∗ ) 𝑉−1 (𝑠) , Therefore, Φ1 𝑥(⋅) is quadratic-mean asymptotically
𝑘→∞
almost periodic.
max {𝑡, 0} ≤ 𝑠, Similar to the proof given for Φ1 𝑥(⋅), one can prove that
Φ 𝑥(⋅), Φ3 𝑥(⋅), and Φ4 𝑥(⋅) are quadratic-mean asymptoti-
2

lim 𝑈 (𝑡 + 𝛼𝑘 ) 𝑄𝑈−1 (𝑠 + 𝛼𝑘 ) = 𝑉 (𝑡) 𝑄∗ 𝑉−1 (𝑠) , cally almost periodic.

𝑘→∞
Thus, 𝐵1 𝑥 is quadratic-mean asymptotically almost peri-
0 ≥ 𝑠 ≥ 𝑡, odic whenever 𝑥 ∈ 𝐴𝐴𝑃(R, 𝐿2 (P, H)).
Secondly, we show that 𝐵2 𝑥 is also quadratic-mean
lim 𝑈 (𝑡 + 𝛼𝑘 ) (𝐼 − 𝑄) 𝑈−1 (𝑠 + 𝛼𝑘 ) = 𝑉 (𝑡) (𝐼 − 𝑄∗ ) 𝑉−1 (𝑠) , asymptotically almost periodic whenever 𝑥 ∈ 𝐴𝐴𝑃(R, 𝐿2
𝑘→∞
(P, H)).
min {𝑡, 0} ≥ 𝑠, Of course, this is more complicated than the previous
(38) case because of the involvement of the Brownian motion 𝑊.
Abstract and Applied Analysis 7

To overcome such a difficulty, one makes extensive use of Next, let us show that 𝑀(𝑡) ∈ 𝐶0 (R, 𝐿2 (P, H)). In fact
̃ defined by
the Itô isometry identity and the properties of 𝑊
̃ ≡ 𝑊(𝑠 + 𝛼𝑘 ) for each 𝑠. Note that 𝑊
𝑊 ̃ is also a Brownian E‖𝑀(𝑡)‖2
motion and has the same distribution as 𝑊. Assuming that 𝑥 󵄩󵄩 𝑡 󵄩󵄩2
󵄩 󵄩
is quadratic-mean asymptotically almost periodic, using (H2) = E󵄩󵄩󵄩∫ 𝑈(𝑡)𝑃𝑈−1 (𝑠)𝑚(𝑠, 𝑥(𝑠), 𝑥𝑠 ) ∘ d𝑊(𝑠)󵄩󵄩󵄩
and Lemma 10, one can easily see that 𝑠 󳨃 → 𝐻(𝑠, 𝑥(𝑠), 𝑥𝑠 ) 󵄩󵄩 0 󵄩󵄩
is quadratic-mean asymptotically almost periodic. Therefore, 󵄩󵄩 𝑡 󵄩󵄩 2
there exist an almost periodic stochastic process 𝑛(𝑠, 𝑥(𝑠), 𝑥𝑠 ) 󵄩 ̃ 󵄩󵄩󵄩)
= E(󵄩󵄩󵄩∫ 𝑈(𝑡)𝑃𝑈−1 (𝑠)𝑚(𝑠, 𝑥(𝑠), 𝑥𝑠 ) ∘ d𝑊(𝑠) 󵄩󵄩
and a stochastic process 𝑚 ∈ 𝐶0 (R × 𝐿2 (P, H), 𝐿2 (P, H)) such 󵄩󵄩 0 󵄩
that 𝐻 = 𝑛 + 𝑚. Furthermore, one observes that 𝑡
≤ 𝛽2 ∫ exp (−2𝛼 (𝑡 − 𝑠))
𝑡 0
Ψ1 𝑥 (𝑡) ≡ ∫ 𝑈 (𝑡) 𝑃𝑈−1 (𝑠) 𝑛 (𝑠, 𝑥 (𝑠) , 𝑥𝑠 ) ∘ d𝑊 (𝑠)
󵄩 󵄩2
0 × E󵄩󵄩󵄩𝑚(𝑠 + 𝜏, 𝑥(𝑠 + 𝜏), 𝑥𝑠+𝜏 ) − 𝑚(𝑠, 𝑥(𝑠), 𝑥𝑠 )󵄩󵄩󵄩 d𝑠
𝑡
(42)
+ ∫ 𝑈 (𝑡) 𝑃𝑈−1 (𝑠) 𝑚 (𝑠, 𝑥 (𝑠) , 𝑥𝑠 ) ∘ d𝑊 (𝑠) 𝑡
0 ≤ 𝛽2 (∫ exp (−2𝛼 (𝑡 − 𝑠) d𝑠))
0
:= 𝑁 (𝑡) + 𝑀 (𝑡) . 󵄩 󵄩2
× supE󵄩󵄩󵄩𝑚(𝑠 + 𝜏, 𝑥(𝑠 + 𝜏), 𝑥𝑠+𝜏 ) − 𝑚(𝑠, 𝑥(𝑠), 𝑥𝑠 )󵄩󵄩󵄩
𝑠∈R
We claim that 𝑁(𝑡) ∈ 𝐴P(R, 𝐿2 (P, H)). In fact, 𝑠 󳨃 →
2
𝑛(𝑠, 𝑥(𝑠), 𝑥𝑠 ) is quadratic-mean almost periodic. Therefore, 𝛽
= (1 − 𝑒−2𝛼𝑡 )
one can find a common subsequence {𝛼𝑛 }. One still denotes 2𝛼
it as 𝛼 = {𝛼𝑛 }, such that 󵄩 󵄩2
× sup E󵄩󵄩󵄩𝑚(𝑠 + 𝜏, 𝑥(𝑠 + 𝜏), 𝑥𝑠+𝜏 ) − 𝑚(𝑠, 𝑥(𝑠), 𝑥𝑠 )󵄩󵄩󵄩 ,
𝑠∈R
𝑇𝛼 𝑛 (𝑡, 𝑥 (𝑡) , 𝑥𝑡 ) = 𝑛1 (𝑡, 𝑥 (𝑡) , 𝑥𝑡 ) , (43) (47)
uniformly for 𝑡 ∈ R and (38) hold. where we make extensive use of the Itô isometry identity and
Now the properties of 𝑊̃ defined by 𝑊 ̃ ≡ 𝑊(𝑠 + 𝜏) − 𝑊(𝜏) for each
̃ is also a Brownian motion and has the same
𝑠. Note that 𝑊
𝑁 (𝑡 + 𝛼𝑘 )
distribution as 𝑊.
𝑡+𝛼𝑘 Since 𝑚 ∈ 𝐶0 (R × 𝐿2 (P, H), 𝐿2 (P, H)), we deduce that
=∫ 𝑈 (𝑡 + 𝛼𝑘 ) 𝑃𝑈−1 (𝑠) 𝑛 (𝑠, 𝑥 (𝑠) , 𝑥𝑠 ) ∘ d𝑊 (𝑠) lim𝑡 → +∞ E ‖ 𝑀(𝑡)‖2 = 0.
0
Therefore, Ψ1 𝑥(⋅) is quadratic-mean asymptotically
𝑡
= ∫ 𝑈 (𝑡 + 𝛼𝑘 ) 𝑃𝑈−1 (𝑠 + 𝛼𝑘 ) almost periodic.
0 Similar to the proof given for Ψ1 𝑥(⋅), one can prove that
(44)
Ψ 𝑥(⋅), Ψ3 𝑥(⋅), and Ψ4 𝑥(⋅) are quadratic-mean asymptotically
2
× 𝑛 (𝑠 + 𝛼𝑘 , 𝑥 (𝑠 + 𝛼𝑘 ) , 𝑥𝑠+𝛼𝑘 ) ∘ d𝑊 (𝑠 + 𝛼𝑘 ) almost periodic.
𝑡 Thus, 𝐵2 𝑥 is quadratic-mean asymptotically almost peri-
= ∫ 𝑈 (𝑡 + 𝛼𝑘 ) 𝑃𝑈−1 (𝑠 + 𝛼𝑘 ) odic whenever 𝑥 ∈ 𝐴𝐴𝑃(R, 𝐿2 (P, H)).
0 In view of the above, it is clear that L maps
̃ (𝑠) .
× 𝑛 (𝑠 + 𝛼𝑘 , 𝑥 (𝑠 + 𝛼𝑘 ) , 𝑥𝑠+𝛼𝑘 ) ∘ d𝑊 𝐴𝐴𝑃(R, 𝐿2 (P, H) into itself. The proof of Lemma 15 is
complete.
Using Itô’s isometry identity, one obtains Lemma 16. Assume that conditions (H1)–(H3) are satisfied.
Then the operator L is a contraction providing Λ = (8𝛽/𝛼)
󵄩 󵄩2
E󵄩󵄩󵄩𝑁(𝑡 + 𝛼𝑘 )󵄩󵄩󵄩 √𝐹𝑙 + 𝛼𝐻𝑙 /2 < 1.
𝑡
󵄩 󵄩2
= ∫ E󵄩󵄩󵄩󵄩𝑈(𝑡 + 𝛼𝑘 )𝑃𝑈−1 (𝑠 + 𝛼𝑘 )𝑛(𝑠 + 𝛼𝑘 , 𝑥(𝑠 + 𝛼𝑘 ), 𝑥𝑠+𝛼𝑘 )󵄩󵄩󵄩󵄩 d𝑠. Proof. Consider
0
󵄩󵄩 󵄩
(45) 󵄩󵄩(L𝑥) 𝑡 − (L𝑦) 𝑡󵄩󵄩󵄩
󵄩󵄩 +∞
󵄩
From (38), (43), (45), and Lebesgue’s control convergence ≤ 󵄩󵄩󵄩∫ 𝐺 (𝑡, 𝑠) [𝐹 (𝑠, 𝑥 (𝑠) , 𝑥𝑠 ) − 𝐹 (𝑠, 𝑦 (𝑠) , 𝑥𝑠 )] d𝑠
theorem, one sees that 󵄩󵄩 −∞
+∞ 󵄩󵄩
󵄩
𝑡
∗ −1 +∫ 𝐺 (𝑡, 𝑠) [𝐻 (𝑠, 𝑥 (𝑠) , 𝑥𝑠 ) − 𝐻 (𝑠, 𝑦 (𝑠) , 𝑥𝑠 )] ∘ d𝑊 (𝑠)󵄩󵄩󵄩
𝑇𝛼 𝑁 (𝑡) = ∫ 𝑉 (𝑡) 𝑃 𝑉 (𝑠) 𝑛1 (𝑠, 𝑥 (𝑠) , 𝑥𝑠 ) d𝑠 (46) −∞ 󵄩󵄩
0
󵄩󵄩 +∞ 󵄩󵄩
󵄩 󵄩
uniformly for 𝑡 ∈ R. Hence 𝑁(𝑡) ∈ 𝐴𝑃(R, 𝐿2 (P, H)). ≤ 󵄩󵄩󵄩∫ 𝐺 (𝑡, 𝑠) [𝐹 (𝑠, 𝑥 (𝑠) , 𝑥𝑠 ) − 𝐹 (𝑠, 𝑦 (𝑠) , 𝑥𝑠 )] d𝑠󵄩󵄩󵄩
󵄩󵄩 −∞ 󵄩󵄩
8 Abstract and Applied Analysis

󵄩󵄩 +∞ 󵄩󵄩
󵄩 󵄩 We first evaluate 𝑈1 as follows:
+ 󵄩󵄩󵄩∫ 𝐺 (𝑡, 𝑠) [𝐻 (𝑠, 𝑥 (𝑠) , 𝑥𝑠 ) − 𝐻 (𝑠, 𝑦 (𝑠) , 𝑥𝑠 )] ∘ d𝑊 (𝑠)󵄩󵄩󵄩
󵄩󵄩 −∞ 󵄩󵄩
󵄩 󵄩 󵄩 󵄩 𝑈1
:= 󵄩󵄩󵄩𝐶1 󵄩󵄩󵄩 + 󵄩󵄩󵄩𝐶2 󵄩󵄩󵄩 .
(48) 𝑡 2
󵄩 󵄩
= 4𝛽2 E(∫ 𝑒−𝛼(𝑡−𝑠) 󵄩󵄩󵄩𝐹(𝑠, 𝑥(𝑠), 𝑥𝑠 ) − 𝐹(𝑠, 𝑦(𝑠), 𝑥𝑠 )󵄩󵄩󵄩 d𝑠)
2 0
Since (∑𝑛𝑖=1 𝑎𝑖 ) < 𝑛 ∑𝑛𝑖=1 𝑎𝑖2 , one can write
𝑡
󵄩 󵄩2 󵄩 󵄩2 󵄩 󵄩2 ≤ 4𝛽2 E [(∫ 𝑒−𝛼(𝑡−𝑠) d𝑠)
E󵄩󵄩󵄩(L𝑥)𝑡 − (L𝑦)𝑡󵄩󵄩󵄩 ≤ 2E󵄩󵄩󵄩𝐶1 󵄩󵄩󵄩 + 2E󵄩󵄩󵄩𝐶2 󵄩󵄩󵄩 . (49) 0
𝑡
Combining (30), one can write 󵄩 󵄩2
× (∫ 𝑒−𝛼(𝑡−𝑠) 󵄩󵄩󵄩𝐹(𝑠, 𝑥(𝑠), 𝑥𝑠 ) − 𝐹(𝑠, 𝑦(𝑠), 𝑥𝑠 )󵄩󵄩󵄩 d𝑠)]
󵄩󵄩 󵄩󵄩 0
󵄩󵄩𝐶1 󵄩󵄩
𝑡
󵄩󵄩 ∞
󵄩
󵄩󵄩
󵄩 ≤ 4𝛽2 (∫ 𝑒−𝛼(𝑡−𝑠) d𝑠)
= 󵄩󵄩󵄩∫ 𝐺 (𝑡, 𝑠) [𝐹 (𝑠, 𝑥 (𝑠) , 𝑥𝑠 ) − 𝐹 (𝑠, 𝑦 (𝑠) , 𝑥𝑠 )] d𝑠󵄩󵄩󵄩 0
󵄩󵄩 −∞ 󵄩󵄩
𝑡
󵄩 󵄩2
󵄩󵄩 𝑡
󵄩 × (∫ 𝑒−𝛼(𝑡−𝑠) E󵄩󵄩󵄩𝐹(𝑠, 𝑥(𝑠), 𝑥𝑠 ) − 𝐹(𝑠, 𝑦(𝑠), 𝑥𝑠 )󵄩󵄩󵄩 d𝑠)
= 󵄩󵄩󵄩∫ 𝑈 (𝑡) 𝑃𝑈−1 (𝑠) [𝐹 (𝑠, 𝑥 (𝑠) , 𝑥𝑠 ) − 𝐹 (𝑠, 𝑦 (𝑠) , 𝑥𝑠 )] d𝑠 0
󵄩󵄩 0 2
𝑡
+∞ ≤ 4𝛽2 𝐹𝑙 (∫ 𝑒−𝛼(𝑡−𝑠) d𝑠)
−∫ 𝑈 (𝑡) (𝐼 − 𝑃) 𝑈−1 (𝑠) 0
max{𝑡,0}
󵄩 󵄩2 󵄩 󵄩2
× sup (󵄩󵄩󵄩𝑥(𝑠) − 𝑦(𝑠)󵄩󵄩󵄩𝐿2 (P,H) + 󵄩󵄩󵄩𝑥(𝑠) − 𝑦(𝑠)󵄩󵄩󵄩𝐶)
× [𝐹 (𝑠, 𝑥 (𝑠) , 𝑥𝑠 ) − 𝐹 (𝑠, 𝑦 (𝑠) , 𝑥𝑠 )] d𝑠 𝑠∈R
0
− ∫ 𝑈 (𝑡) 𝑄𝑈−1 (𝑠) [𝐹 (𝑠, 𝑥 (𝑠) , 𝑥𝑠 ) − 𝐹 (𝑠, 𝑦 (𝑠) , 𝑥𝑠 )] d𝑠 8𝛽2 𝐹𝑙 −𝛼𝑡 󵄩
2
󵄩󵄩𝑥(⋅) − 𝑦(⋅)󵄩󵄩󵄩2 ≤ 8𝛽 𝐹𝑙 󵄩󵄩󵄩𝑥(⋅) − 𝑦(⋅)󵄩󵄩󵄩2 .
≤ [1 − 𝑒 ] 󵄩 󵄩∞
𝑡 𝛼2 𝛼2 󵄩 󵄩∞
min{𝑡,0}
(52)
+∫ 𝑈 (𝑡) (𝐼 − 𝑄) 𝑈−1 (𝑠)
−∞ Similar to the discussion given for 𝑈1 , for 𝑈2 , 𝑈3 , and 𝑈4 , one
󵄩󵄩 has
󵄩
× [𝐹 (𝑠, 𝑥 (𝑠) , 𝑥𝑠 ) − 𝐹 (𝑠, 𝑦 (𝑠) , 𝑥𝑠 )] d𝑠󵄩󵄩󵄩
󵄩󵄩 𝑈2
𝑡 +∞ 2
+∞
≤ 𝛽 [ ∫ 𝑒−𝛼(𝑡−𝑠) + ∫ 𝑒−𝛼(𝑠−𝑡) = 4𝛽2 E(∫
󵄩 󵄩
𝑒−𝛼(𝑠−𝑡) 󵄩󵄩󵄩𝐹(𝑠, 𝑥(𝑠), 𝑥𝑠 ) − 𝐹(𝑠, 𝑦(𝑠), 𝑥𝑠 )󵄩󵄩󵄩 d𝑠)
0 max{𝑡,0}
max{𝑡,0}
0 min{𝑡,0} 2
+ ∫ 𝑒−𝛼(𝑠−𝑡) + ∫ 𝑒−𝛼(𝑡−𝑠) ] 8𝛽 𝐹𝑙 󵄩󵄩 󵄩2
≤ 󵄩𝑥(⋅) − 𝑦(⋅)󵄩󵄩󵄩∞ ,
𝑡 −∞ 𝛼2 󵄩
󵄩 󵄩
⋅ 󵄩󵄩󵄩𝐹 (𝑠, 𝑥 (𝑠) , 𝑥𝑠 ) − 𝐹 (𝑠, 𝑦 (𝑠) , 𝑥𝑠 )󵄩󵄩󵄩 d𝑠. 8𝛽2 𝐹𝑙 󵄩󵄩 󵄩2 8𝛽2 𝐹𝑙 󵄩󵄩 󵄩2
𝑈3 ≤ 󵄩𝑥(⋅) − 𝑦(⋅)󵄩󵄩󵄩∞ , 𝑈4 ≤ 󵄩𝑥(⋅) − 𝑦(⋅)󵄩󵄩󵄩∞ .
(50) 𝛼2 󵄩 𝛼2 󵄩
(53)
Thus,
󵄩 󵄩2 Then,
E󵄩󵄩󵄩𝐶1 󵄩󵄩󵄩
2
󵄩 󵄩2 32𝛽 𝐹𝑙 󵄩󵄩 󵄩2
󵄩
𝑡
󵄩
2 E󵄩󵄩󵄩𝐶1 󵄩󵄩󵄩 ≤ 󵄩󵄩𝑥(⋅) − 𝑦(⋅)󵄩󵄩󵄩∞ . (54)
≤ 4𝛽2 E(∫ 𝑒−𝛼(𝑡−𝑠) 󵄩󵄩󵄩𝐹(𝑠, 𝑥(𝑠), 𝑥𝑠 ) − 𝐹(𝑠, 𝑦(𝑠), 𝑥𝑠 )󵄩󵄩󵄩 d𝑠) 𝛼 2
0

+∞ 2 Next, we evaluate E‖𝐶2 ‖2 as follows:

−𝛼(𝑠−𝑡) 󵄩
󵄩 󵄩
󵄩󵄩𝐹(𝑠, 𝑥(𝑠), 𝑥𝑠 ) − 𝐹(𝑠, 𝑦(𝑠), 𝑥𝑠 )󵄩󵄩󵄩 d𝑠)
2
+4𝛽 E(∫ 𝑒
󵄩 󵄩2
max{𝑡,0} E󵄩󵄩󵄩𝐶2 󵄩󵄩󵄩
0 2
󵄩 󵄩 󵄩󵄩 ∞ 󵄩󵄩2
+ 4𝛽2 E(∫ 𝑒−𝛼(𝑠−𝑡) 󵄩󵄩󵄩𝐹(𝑠, 𝑥(𝑠), 𝑥𝑠 ) − 𝐹(𝑠, 𝑦(𝑠), 𝑥𝑠 )󵄩󵄩󵄩 d𝑠) 󵄩
= E󵄩󵄩󵄩∫ 𝐺(𝑡, 𝑠)[𝐻(𝑠, 𝑥(𝑠), 𝑥𝑠 ) − 𝐻(𝑠, 𝑦(𝑠), 𝑥𝑠 )] ∘ d𝑊(𝑠)󵄩󵄩󵄩
󵄩
𝑡 󵄩󵄩 −∞ 󵄩󵄩
min{𝑡,0} 2 󵄩󵄩 𝑡
󵄩 󵄩 󵄩
+4𝛽2 E(∫ 𝑒−𝛼(𝑡−𝑠) 󵄩󵄩󵄩𝐹(𝑠, 𝑥(𝑠), 𝑥𝑠 ) − 𝐹(𝑠, 𝑦(𝑠), 𝑥𝑠 )󵄩󵄩󵄩 d𝑠) ≤ 4E 󵄩󵄩󵄩∫ 𝑈 (𝑡) 𝑃𝑈−1 (𝑠)
−∞ 󵄩󵄩 0

:= 𝑈1 + 𝑈2 + 𝑈3 + 𝑈4 . 󵄩󵄩󵄩2
× [𝐻(𝑠, 𝑥(𝑠), 𝑥𝑠 ) − 𝐻(𝑠, 𝑦(𝑠), 𝑥𝑠 )] ∘ d𝑊(𝑠)󵄩󵄩󵄩
(51) 󵄩󵄩
Abstract and Applied Analysis 9

󵄩󵄩 +∞
󵄩 8𝛽2 𝐻𝑙 󵄩󵄩 󵄩2
+ 4E 󵄩󵄩󵄩∫ 𝑈 (𝑡) (𝐼 − 𝑃) 𝑈−1 (𝑠) ≤ 󵄩𝑥(⋅) − 𝑦(⋅)󵄩󵄩󵄩∞ ,
󵄩󵄩 max{𝑡,0} 𝛼 󵄩
󵄩󵄩2 8𝛽2 𝐻𝑙 󵄩󵄩 󵄩2
󵄩 󵄩𝑥(⋅) − 𝑦(⋅)󵄩󵄩󵄩∞ ,
× [𝐻(𝑠, 𝑥(𝑠), 𝑥𝑠 ) − 𝐻(𝑠, 𝑦(𝑠), 𝑥𝑠 )] ∘ d𝑊(𝑠)󵄩󵄩󵄩 𝑉3 ≤
󵄩󵄩 𝛼 󵄩
󵄩󵄩 0 8𝛽2 𝐻𝑙 󵄩󵄩 󵄩2
󵄩 󵄩𝑥(⋅) − 𝑦(⋅)󵄩󵄩󵄩∞ .
+ 4E 󵄩󵄩󵄩󵄩∫ 𝑈 (𝑡) 𝑄𝑈−1 (𝑠) 𝑉4 ≤
𝛼 󵄩
󵄩󵄩 𝑡
(57)
󵄩󵄩2
󵄩
× [𝐻(𝑠, 𝑥(𝑠), 𝑥𝑠 ) − 𝐻(𝑠, 𝑦(𝑠), 𝑥𝑠 )] ∘ d𝑊(𝑠)󵄩󵄩󵄩
󵄩󵄩 Therefore,
󵄩󵄩 min{𝑡,0} 2
󵄩 󵄩 󵄩2 32𝛽 𝐻𝑙 󵄩󵄩 󵄩2
+ 4E 󵄩󵄩󵄩󵄩∫ 𝑈 (𝑡) (𝐼 − 𝑄) 𝑈−1 (𝑠) E󵄩󵄩󵄩𝐶2 󵄩󵄩󵄩 ≤ 󵄩𝑥(⋅) − 𝑦(⋅)󵄩󵄩󵄩∞ . (58)
󵄩󵄩 −∞ 𝛼 󵄩
󵄩󵄩2 Thus, by combing (54) and (58), it follows that
󵄩
× [𝐻(𝑠, 𝑥(𝑠), 𝑥𝑠 ) − 𝐻(𝑠, 𝑦(𝑠), 𝑥𝑠 )] ∘ d𝑊(𝑠)󵄩󵄩󵄩
󵄩󵄩 2
󵄩 󵄩2 64𝛽 𝛼𝐻𝑙 󵄩󵄩 󵄩2
E󵄩󵄩󵄩(L𝑥)(𝑡) − (L𝑦)(𝑡)󵄩󵄩󵄩 ≤ 2 [𝐹𝑙 + ] 󵄩𝑥(⋅) − 𝑦(⋅)󵄩󵄩󵄩∞ ,
:= 𝑉1 + 𝑉2 + 𝑉3 + 𝑉4 . 𝛼 2 󵄩
(55) (59)

which implies that

As for the first term 𝑉1 , using Itô’s isometry identity, one
obtains 󵄩󵄩 󵄩 8𝛽 𝛼𝐻𝑙 󵄩󵄩 󵄩
󵄩󵄩(L𝑥)(𝑡) − (L𝑦)(𝑡)󵄩󵄩󵄩∞ ≤ √ 𝐹𝑙 + 󵄩𝑥(⋅) − 𝑦(⋅)󵄩󵄩󵄩∞
𝛼 2 󵄩 (60)
𝑉1 󵄩󵄩 󵄩󵄩
= Λ󵄩󵄩𝑥 − 𝑦󵄩󵄩∞ .
󵄩󵄩 𝑡
󵄩 Since Λ < 1, by (60), we know that L is a contraction
= 4E 󵄩󵄩󵄩∫ 𝑈 (𝑡) 𝑃𝑈−1 (𝑠)
󵄩󵄩 0 mapping. The proof of Lemma 16 is complete.
󵄩󵄩2
󵄩
×[𝐻(𝑠, 𝑥(𝑠), 𝑥𝑠 ) − 𝐻(𝑠, 𝑦(𝑠), 𝑥𝑠 )] ∘ d𝑊(𝑠)󵄩󵄩󵄩 Hence, combining Lemmas 15 and 16 and the contraction
󵄩󵄩 mapping principle, L has a unique fixed point 𝑥(𝑡), which is
𝑡 obviously the unique quadratic-mean asymptotically almost
󵄩 󵄩2
= 4E [∫ 󵄩󵄩󵄩󵄩𝑈(𝑡)𝑃𝑈−1 (𝑠)[𝐻(𝑠, 𝑥(𝑠), 𝑥𝑠 ) −𝐻(𝑠, 𝑦(𝑠), 𝑥𝑠 )]󵄩󵄩󵄩󵄩 d𝑠] periodic mild solution of (26).
0 This completes the proof of Theorem 14 due to Lemmas
𝑡 15 and 16.
󵄩 󵄩2
≤ 4𝛽2 ∫ 𝑒−2𝛼(𝑡−𝑠) E󵄩󵄩󵄩𝐻(𝑠, 𝑥(𝑠), 𝑥𝑠 ) − 𝐻(𝑠, 𝑦(𝑠), 𝑥𝑠 )󵄩󵄩󵄩 d𝑠
0 Remark 17. If the conditions of the main result of [14] and
𝑡 (H1) and (H2) hold, (26) admits exponential trichotomy with
≤ 4𝛽2 𝐻𝑙 (∫ 𝑒−2𝛼(𝑡−𝑠) d𝑠) projections 𝑃, 𝑄 and 𝑃𝑄 = 0; hence system (26) has a
0
unique quadratic-mean asymptotically almost periodic mild
󵄩 󵄩2 󵄩 󵄩2
× sup (󵄩󵄩󵄩𝑥(𝑠) − 𝑦(𝑠)󵄩󵄩󵄩𝐿2 (P,H) + 󵄩󵄩󵄩𝑥(𝑠) − 𝑦(𝑠)󵄩󵄩󵄩𝐶) solution. So our main result improves the main result of [14].
𝑠∈R

8𝛽2 𝐻𝑙 󵄩 󵄩2 8𝛽2 𝐻𝑙 󵄩󵄩 󵄩2
4. Example
≤ [1 − 𝑒−2𝛼𝑡 ] 󵄩󵄩󵄩𝑥(⋅) − 𝑦(⋅)󵄩󵄩󵄩∞ ≤ 󵄩𝑥(⋅) − 𝑦(⋅)󵄩󵄩󵄩∞ .
𝛼 𝛼 󵄩 The deterministic nonlinear Duffing-van der Pol equation
(56)
𝑦̈ + 𝑟𝑦3 + 𝑑y2 𝑦̇
= 𝑎𝑦 + 𝑏𝑦̇ + 𝑠𝑦3̇
, 𝑎, 𝑏, 𝑟, 𝑑, 𝑠 ∈ R, (61)
Similarly, one can evaluate the second term 𝑉2 , third term
𝑉3 , and the fourth term 𝑉4 of the right-hand side, respectively: has become a paradigm for mathematicians, physicists, and
engineers. There are numerous physical and engineering
problems whose dynamics are described by (61) for some
𝑉2 parameter values.
In particular, for 𝑟 = 𝑠 = 0 one obtains the van der Pol
󵄩󵄩 +∞
󵄩 equation and for 𝑑 = 𝑠 = 0 one obtains the Duffing equation.
= 4E 󵄩󵄩󵄩∫ 𝑈 (𝑡) 𝑃𝑈−1 (𝑠)
󵄩󵄩 max{𝑡,0} As an example, we consider the following functional
differential equation for small 𝜖1 and 𝜖2 :
󵄩󵄩2
󵄩
× [𝐻(𝑠, 𝑥(𝑠), 𝑥𝑠 ) − 𝐻(𝑠, 𝑦(𝑠), 𝑥𝑠 )] ∘ d𝑊(𝑠)󵄩󵄩󵄩
󵄩󵄩 = − (𝑎 + 𝑏𝑝 (𝑡)) 𝑦 + 𝜖1 𝑓 (𝑡, 𝑦, 𝑦𝑡 ) + 𝜖2 ℎ (𝑡, 𝑦, 𝑦𝑡 ) 𝑦.̇(62)
𝑦̈
10 Abstract and Applied Analysis

It is easy to see that (61) is the general form of the Duffing- which reflects a deeper analysis of the behavior of solutions of
van der Pol equation (61). By perturbing the remaining the system. In this case the asymptotic behavior is described
parameters and the right-hand side of (62) by real or white through the splitting of the main space into stable, unstable,
noise, one arrives at noise, more specifically the random and central subspaces at each point from the flow’s domain.
(for real noise) or stochastic (for white noise) functional An example is also given to illustrate our results.
differential equation, respectively:

= − (𝑎 + 𝑏𝑝 (𝑡)) 𝑦 + 𝜖1 𝑓 (𝑡, 𝑦, 𝑦𝑡 ) + 𝜖2 𝜉2 (𝑡) ℎ (𝑡, 𝑦, 𝑦𝑡 ) 𝑦,̇

𝑦̈ Conflict of Interests
(63) The authors declare that there is no conflict of interests
where 𝑠2 is intensity parameter, functions 𝑓 ∈ 𝐴𝐴𝑃(R × regarding the publication of this paper.
𝐿2 (P, R) × 𝐶; 𝐿2 (P, R)), ℎ ∈ 𝐴𝐴𝑃(R × 𝐿2 (P, R) ×
𝐶; 𝐿2 (P, L(𝐿2 (P, R))), and 𝜉(𝑡) = (𝜉1 (𝑡), 𝜉2 (𝑡)) = (0, 𝜉2 (𝑡)) Acknowledgments
is a stationary process or white noise. As a first order system
for 𝑥 = (𝑥1 , 𝑥2 )𝑇 = (𝑦, 𝑦)̇
𝑇
, (63) takes the form The authors wish to thank the anonymous reviewers whose
suggestions and corrections have significantly improved the
d𝑥 (𝑡) = [𝐴 (𝑡) 𝑥 (𝑡) + 𝐹 (𝑡, 𝑥, 𝑥𝑡 )] d𝑡 + 𝐻 (𝑡, 𝑥, 𝑥𝑡 ) ∘ d𝑊 (𝑡) , presentation of this paper. This work was supported by the
National Natural Science Foundation of China (no. 11161051)
𝑡 ∈ R, and the Scientific Research Foundation of Guangxi Education
(64) Office of China (no. 2014).

where
References
0 1
𝐴 (𝑡) = ( ),
− (𝑎 + 𝑏𝑝 (𝑡)) 0 [1] A. M. Fink, Almost Periodic Differential Equations, vol. 377 of
Lecture Notes in Mathematics, Springer, New York, NY, USA,
1974.
0
𝐹 (𝑡, 𝑥, 𝑥𝑡 ) = ( ), (65) [2] C. Zhang, Almost Periodic Type Functions and Ergodicity,
𝜖1 𝑓 (𝑡, 𝑥1 , 𝑥1𝑡 )
Science Press, Beijing, China, 2006.
0 [3] P. H. Bezandry, “Existence of almost periodic solutions to some
𝐻 (𝑡, 𝑥, 𝑥𝑡 ) = ( ). functional integro-differential stochastic evolution equations,”
𝜖2 ℎ (𝑡, 𝑥1 , 𝑥1𝑡 ) 𝑥2
Statistics & Probability Letters, vol. 78, no. 17, pp. 2844–2849,
2008.
Let 𝑥 = 𝐴𝐴𝑃(R; 𝐿2 (P, R2 )). Suppose that vector func-
tions 𝐹 and 𝐻 satisfy the inequality conditions of (H1) and [4] E. Slutsky, “Sur les fonctions aléatoires presque périodiques et
sur la decomposition des functions aléatoires,” in Actualités
(H2) in Theorem 14. We see that when (i) 𝑏 ≥ 0, 𝑎 +
Scientifiques et Industrielles, vol. 738, pp. 33–55, Herman, Paris,
sup𝑡∈R 𝑝(𝑡) < 0 or (ii) 𝑏 < 0, 𝑎 + inf 𝑡∈R 𝑝(𝑡) > 0 hold, France, 1938.
d𝑥/d𝑡 = 𝐴(𝑡)𝑥(𝑡) admits exponential trichotomy, as in [24,
[5] M. Udagawa, “Asymptotic properties of distributions of some
25]. That is, there exist constants 𝛼 and 𝛽 such that (16) holds. functionals of random variables,” Reports of Statistical Applica-
When (8𝛽/𝛼)√𝜖1 𝐹𝑙 + 𝛼𝜖2 𝐻𝑙 /2 < 1, the other conditions of tion Research. Union of Japanese Scientists and Engineers, vol. 2,
Theorem 14 are satisfied. So (63) has a unique quadratic-mean no. 2-3, pp. 1–98, 1952.
asymptotically almost periodic mild solution. [6] T. Kawata, “Almost periodic weakly stationary processes,” in
Statistics and Probability: Essays in Honor of C. R. Rao, pp. 383–
Remark 18. The linear system of the system (63) admits 396, North-Holland, New York, NY, USA, 1982.
exponential trichotomy with projections 𝑃, 𝑄 and 𝑃𝑄 = 0; [7] R. J. Swift, “Almost periodic harmonizable processes,” Georgian
hence the main result of [14] cannot ensure the existence and Mathematical Journal, vol. 3, no. 3, pp. 275–292, 1996.
uniqueness of asymptotically almost periodic mild solution of
[8] P. H. Bezandry and T. Diagana, “Square-mean almost periodic
(63). It is obvious that the condition that “the family {𝐴(𝑡) : solution to some nonautonomous stochastic differential equa-
𝑡 ∈ R} of operators in R generates an exponentially stable tions,” Electronic Journal of Differential Equations, vol. 2007, no.
semigroup {𝑇(𝑡)}𝑡≥0 ” cannot be satisfied. So the main result of 117, pp. 1–10, 2007.
[18, 19, 21] cannot also ensure the existence of almost periodic [9] J. Luo, “Stochastically bounded solutions of a nonlinear stochas-
type mild solution of (63). tic differential equations,” Journal of Computational and Applied
Mathematics, vol. 196, no. 1, pp. 87–93, 2006.
5. Conclusions [10] W. M. Ruess, “Asymptotic almost periodicity and motions of
semigroups of operators,” Linear Algebra and Its Applications,
In this paper, a new criterion ensuring the existence and vol. 84, pp. 335–351, 1986.
uniqueness of the quadratic-mean asymptotically almost [11] W. M. Ruess and V. Q. Phong, “Asymptotically almost periodic
periodic mild solutions for a class of stochastic functional solutions of evolution equations in Banach spaces,” Journal of
differential equations is presented. The condition of being Differential Equations, vol. 122, no. 2, pp. 282–301, 1995.
uniformly exponentially stable of the linear operator is [12] W. Arendt and C. J. K. Batty, “Asymptotically almost periodic
essentially removed, only using its exponential trichotomy, solutions of inhomogeneous Cauchy problems on the half-line,”
Abstract and Applied Analysis 11

Bulletin of the London Mathematical Society, vol. 31, no. 3, pp.

291–304, 1999.
[13] W. Arendt and C. J. K. Batty, “Almost periodic solutions of first-
and second-order Cauchy problems,” Journal of Differential
Equations, vol. 137, no. 2, pp. 363–383, 1997.
[14] J. Cao, Q. Yang, Z. Huang, and Q. Liu, “Asymptotically almost
periodic solutions of stochastic functional differential equa-
tions,” Applied Mathematics and Computation, vol. 218, no. 5,
pp. 1499–1511, 2011.
[15] J. Cao, Q. Yang, and Z. Huang, “On almost periodic mild solu-
tions for stochastic functional differential equations,” Nonlinear
Analysis: Real World Applications, vol. 13, no. 1, pp. 275–286,
2012.
[16] H. Huang and J. Cao, “Exponential stability analysis of uncer-
tain stochastic neural networks with multiple delays,” Nonlinear
Analysis: Real World Applications, vol. 8, no. 2, pp. 646–653,
2007.
[17] J. Cao, Q. Yang, and Z. Huang, “Existence and exponential
stability of almost automorphic mild solutions for stochastic
functional differential equations,” Stochastics, vol. 83, no. 3, pp.
259–275, 2011.
[18] P. H. Bezandry and T. Diagana, “Existence of almost periodic
solutions to some stochastic differential equations,” Applicable
Analysis, vol. 86, no. 7, pp. 819–827, 2007.
[19] P. H. Bezandry and T. Diagana, “Existence of quadratic-mean
almost periodic solutions to some stochastic hyperbolic differ-
ential equations,” Electronic Journal of Differential Equations,
vol. 2009, no. 111, pp. 1–14, 2009.
[20] P. H. Bezandry and T. Diagana, “Existence of square-mean
almost periodic mild solutions to some nonautonomous
stochastic second-order differential equations,” Electronic Jour-
nal of Differential Equations, vol. 2010, no. 124, pp. 1–25, 2010.
[21] M. Fu and Z. Liu, “Square-mean almost automorphic solutions
for some stochastic differential equations,” Proceedings of the
American Mathematical Society, vol. 138, no. 10, pp. 3689–3701,
2010.
[22] Y. Liu and A. Liu, “Almost periodic solutions for a class of
stochastic differential equations,” Journal of Computational and
Nonlinear Dynamics, vol. 8, no. 4, Article ID 041002, 6 pages,
2013.
[23] S. Elaydi and O. Hájek, “Exponential trichotomy of differential
systems,” Journal of Mathematical Analysis and Applications, vol.
129, no. 2, pp. 362–374, 1988.
[24] J. Hong, R. Obaya, and A. Sanz, “Existence of a class of ergodic
solutions implies exponential trichotomy,” Applied Mathematics
Letters, vol. 12, no. 4, pp. 43–45, 1999.
[25] L. Barreira and C. Valls, “Lyapunov functions for trichotomies
with growth rates,” Journal of Differential Equations, vol. 248,
no. 1, pp. 151–183, 2010.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 576087, 8 pages
http://dx.doi.org/10.1155/2014/576087

Research Article
A Stochastic Weakly Damped, Forced KdV-BO Equation

Guolian Wang
Department of Mathematics, Tongji University, Shanghai 200092, China

Correspondence should be addressed to Guolian Wang; guolianwang@126.com

Received 14 January 2014; Revised 13 April 2014; Accepted 13 April 2014; Published 5 May 2014

Academic Editor: Tonghua Zhang

Copyright © 2014 Guolian Wang. This is an open access article distributed under the Creative Commons Attribution License, which
permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

We investigate the long time behavior of the damped, forced KdV-BO equation driven by white noise. We first show that the global
solution generates a random dynamical system. By energy type estimates and dispersive properties, we then prove that this system
possesses a weak random attractor in the space 𝐻1 (R).

1. Introduction is, a zero mean Gaussian process whose correlation function

is given by
The deterministic Korteweg-de Vries-Benjamin-Ono (KdV-
BO) equation describes a large class of internal waves in E (𝐵 (𝑡, 𝑥) 𝐵 (𝑡, 𝑥)) = (𝑡 ∧ 𝑠) (𝑥 ∧ 𝑦) , (3)
the ocean and stratified fluid. The well-posedness for this
equation was studied in [1, 2]. When the surface of the fluid for 𝑡, 𝑠 ⩾ 0, 𝑥, 𝑦 ∈ R.
is submitted to a nonconstant pressure, or when the bottom Alternatively we consider a cylindrical Wiener process 𝑊
of the layer is not flat, a forcing term has to be added to by setting
the equation [3, 4]. The long time behavior of the forced
generalized KdV-BO was studied in [5]. In this paper we 𝜕𝐵 ∞
𝑊 (𝑡) = = ∑𝛽 𝑒 , (4)
are interested in the case when the forcing term is random. 𝜕𝑥 𝑖=0 𝑖 𝑖
The well-posedness for the stochastic KdV-BO driven by the
additive noise was studied in [6]. We have found no studies where {𝑒𝑖 }𝑖∈N is an orthonormal basis of 𝐿2 (R) and {𝛽𝑖 }𝑖∈N is
on the long time behavior of the stochastic KdV-BO equation. a sequence of mutually independent real Brownian motions
In this paper, we consider the long time behavior of the in a fixed probability space (Ω, F, 𝑃) adapted to a filtration
following stochastic damped, forced KdV-BO equation: {F}𝑡⩾0 .
Let us write the Itô form of (1) as follows:
𝜕2 𝐵
𝜕𝑡 𝑢 + 𝛼H (𝜕𝑥2 𝑢) + 𝛽𝜕𝑥3 𝑢 + 𝑢𝜕𝑥 𝑢 + 𝜆𝑢 = 𝑓 + Φ , (1) 𝑑𝑢 + (𝛼𝐻 (𝜕𝑥2 𝑢) + 𝛽𝜕𝑥3 𝑢 + 𝑢𝜕𝑥 𝑢 + 𝜆𝑢 − 𝑓) 𝑑𝑡 = Φ𝑑𝑊, (5)
𝜕𝑡𝜕𝑥
with initial condition
where 𝛼, 𝛽, and 𝜆 are real constants and 𝛼𝛽 ≠0. H denotes
the Hilbert transform 𝑢 (0, 𝑥) = 𝑢0 (𝑥) . (6)

The purpose of this paper is to study the long time behav-

1 𝑓 (𝑥 − 𝑦)
H𝑓 (𝑥) = 𝑝.V. ∫ 𝑑𝑦. (2) ior of the problem (5) with initial data. Before describing our
𝜋 𝑦 works, we recall some facts related to this paper. The Cauchy
problem for the deterministic KdV-BO equation, that is, 𝜆 =
The forcing term 𝑓 is time independent, and 𝑢 is a random 𝑓 = Φ = 0 in (5), was considered in [1]. There the authors
process defined on (𝑡, 𝑥) ∈ R+ × R. Φ is a linear operator. obtained well-posedness results by using Fourier restriction
Also, 𝐵 is a two-parameter Brownian motion on R+ × R, that norm method in Bourgain’s type spaces 𝑋𝑠,𝑏 with 𝑏 > 1/2.
2 Abstract and Applied Analysis

Based on the global existence results given in [1], the attractor A continuous random dynamical system (RDS) on a
of the damped forced KdV-BO equation was obtained in [5]. polish space (𝑋, 𝑑) with Borel sigma-algebra F over 𝜃𝑡 on
The stochastic KdV-BO equation (i.e., 𝜆 = 𝑓 = 0 in (5)) (Ω, F, 𝑃) is by definition a measurable map
was studied in space 𝑋𝑠,𝑏 with 𝑏 < 1/2 in [6]. By introducing
some useful inequalities to deal with the irregularity caused Ψ : R+ × Ω × 𝑋 󳨀→ 𝑋, (𝑡, 𝜔, 𝑥) 󳨃󳨀→ Ψ (𝑡, 𝜔) 𝑥, (8)
by stochastic term, well-posedness results were obtained in
[6]. Following the work in [6], we will construct the attractor such that 𝑃-a.s.
to the Cauchy problem for the stochastic damped, forced (i) Ψ(0, 𝜔) = 𝑖𝑑 on 𝑋;
KdV-BO equation.
Before stating our main result precisely, we introduce (ii) Ψ(𝑡 + 𝑠, 𝜔) = Ψ(𝑡, 𝜃𝑠 𝜔)Ψ(𝑠, 𝜔) for all 𝑠, 𝑡 ∈ R+ (cocycle
some notations. property);
Denote by (⋅, ⋅) and | ⋅ | the inner product and the norm (iii) Ψ(𝑡, 𝜔) : 𝑋 → 𝑋 is continuous.
in 𝐿2 (R), respectively. And ‖ ⋅ ‖𝑋 is the norm of the Banach
space 𝑋. A set-valued map 𝐾 : Ω → 2𝑋 , the set of all subsets of
̃ we denote 𝑋, is called a random compact set, if 𝐾(𝜔) is a compact 𝑃-a.s.
Given two separable Hilbert spaces 𝐻 and 𝐻,
0 ̃ the space of Hilbert-Schmidt operators from 𝐻 and if 𝜔 󳨃→ 𝑑(𝑥, 𝐾(𝜔)) is measurable with respect to F for
by 𝐿 2 (𝐻; 𝐻) each 𝑥 ∈ 𝑋, where 𝑑(𝑥, 𝑀) = inf 𝑥∈𝑀𝑑(𝑥, 𝑦).
̃ Its norm is given by
into 𝐻. Let A(𝜔) be a random set and 𝐵 ⊂ 𝑋; one says A(𝜔)
󵄩󵄩 󵄩󵄩2 attracts 𝐵 if
‖Φ‖2𝐿0 (𝐻;𝐻)
̃ = ∑󵄩
̃ ,
Φ ∈ 𝐿02 (𝐻; 𝐻)
󵄩Φ𝑒𝑖 󵄩󵄩𝐻
̃, (7)
2
𝑖∈N lim dist (Ψ (𝑡, 𝜃−𝑡 𝜔) 𝐵, A) = 0, 𝑃-a.s. (9)
𝑡→∞

where (𝑒𝑖 )𝑖∈N is any orthonormal basis of 𝐻. When 𝐻 = A random set A(𝜔) is said to be a random attractor for
𝐿2 (R), 𝐻̃ = 𝐻𝑠 , 𝐿0 (𝐿2 (R); 𝐻𝑠 ) is simply denoted by 𝐿0,𝑠 . the RDS Ψ if 𝑃-a.s.
2 2
The proof of the global well-posedness of the solution (5)-
(6) is similar to [6]. Here, we only give the following global (i) A(𝜔) is a random compact set;
existence results. (ii) A(𝜔) is invariant, that is, Ψ(𝑡, 𝜔)A(𝜔) = A(𝜃𝑡 𝜔), for
Assume that Φ ∈ 𝐿0,0 0,1 2 2
2 (or Φ ∈ 𝐿 2 ); let 𝑢0 ∈ 𝐿 (Ω, 𝐿 (R)) all 𝑡 > 0;
2 1
(or 𝑢0 ∈ 𝐿 (Ω, 𝐻 (R))) be F0 -measurable. Then, the (iii) A(𝜔) attracts all deterministic bounded sets 𝐵 ⊂ 𝑋.
solution 𝑢(𝑡) is global and belongs to 𝐿2 (Ω; 𝐶([0, 𝑇]; 𝐿2 (R)))
(or 𝐿2 (Ω; 𝐶([0, 𝑇]; 𝐻1 (R)))) for any 𝑇 > 0. Similar to the deterministic theory, the existence result of
We now give our main result about the long time behavior random attractors can be stated as follows (see [8, 9]).
of the KdV-BO equation based on its global existence results.
Lemma 2. If there exists a random compact set absorbing
Theorem 1. Under the assumption that 𝜆 > 0, 𝑓 ∈ 𝐻1 (R), every bounded set 𝐵 ⊂ 𝑋, then the RDS Ψ possesses a random
Φ ∈ 𝐿0,2 2 1 attractor A(𝜔),
2 , and 𝑢𝑠 ∈ 𝐿 (Ω; 𝐻 (R)) is F𝑠 -measurable, then
the random dynamical system associated with the stochastic
A (𝜔) = ∪𝐵⊂𝑋 Λ 𝐵 (𝜔), (10)
equation (5) with initial value 𝑢(𝑠) = 𝑢𝑠 possesses a universal
weak random attractor A in 𝐻1 (R).
where Λ 𝐵 (𝜔) := ∩𝑠⩾0 ∪𝑡⩾𝑠 Ψ(𝑡, 𝜃−𝑡 𝜔)𝐵 is the omega-limit set of
The paper is organized as follows. Section 2 contains some 𝐵.
concepts about the random dynamical system, and Lemma 2
which gives the existence conditions and the structure of the 3. Solve the Equations and Generate an RDS
attractor. Then we show that the unique solution of problem
(5)-(6) generates a random dynamical system in Section 3. We consider the following linear problem to (5)-(6):
In Section 4, we prove that there exists a compact random
absorbing set, which leads to the existence of a random 𝑑𝑢 + {𝛼H𝜕𝑥2 𝑢 + 𝛽𝜕𝑥3 𝑢𝑑𝑡 + 𝜆𝑢} 𝑑𝑡 = Φ𝑑𝑊,
(𝑃) { (11)
attractor, that is, Theorem 1. 𝑢 (0) = 0,

whose solution is given by the stochastic integral (see [10])

2. Preliminaries on Random
Dynamical System 𝑡
𝑢 (𝑡) = ∫ 𝑈 (𝑡 − 𝑠) Φ 𝑑𝑊 (𝑠) . (12)
0
We now recall some concepts and results from [7–9].
Let (Ω, F, 𝑃) be a probability space and {𝜃𝑡 : Ω → From now on we turn our attention to study the well-
Ω, 𝑡 ∈ R} a family of measure preserving transformations posedness of a weakly damped, forced KdV-BO equation
such that (𝑡, 𝜔) 󳨃→ 𝜃𝑡 𝜔 is measurable, 𝜃0 = 𝑖𝑑, and 𝜃𝑡+𝑠 = with random parameter by change of variable.
𝜃𝑡 ∘ 𝜃𝑠 for all 𝑠, 𝑡 ∈ R. The flow 𝜃𝑡 together with the Let us study (5) by means of the change of variable
corresponding probability space (Ω, F, 𝑃) is called a metric
dynamical system. V (𝑡) = 𝑢 (𝑡) − 𝑢 (𝑡) , (13)
Abstract and Applied Analysis 3

and then 𝑢 satisfies (5) if and only if V is a solution of Obviously, for 𝑠 ⩽ 𝑟 ⩽ 𝑡, we have

V𝑡 + 𝛼HV𝑥𝑥 + 𝛽V𝑥𝑥𝑥 + 𝜆V + VV𝑥 𝑆 (𝑡, 𝑠; 𝜔) = 𝑆 (𝑡, 𝑟; 𝜔) 𝑆 (𝑟, 𝑠; 𝜔) . (20)

(14)
+ 𝑢𝑢𝑥 + V𝑢𝑥 + 𝑢V𝑥 = 𝑓. Thanks to (17), for any 𝑠, 𝑡 ∈ R+ , 𝑢𝑠 ∈ 𝐻1 (R), we have 𝑃-a.s.
𝑆 (𝑡 + 𝑠, 0; 𝜔) 𝑢0 = 𝑆 (𝑡, 0; 𝜃𝑠 𝜔) 𝑆 (𝑠, 0; 𝜔) 𝑢0 . (21)
This is a weakly damped, forced KdV-BO equation with
random parameter with the following initial condition: Therefore, the process Ψ : 𝑅+ × Ω × 𝑉 → 𝑉 defined by
V (0, 𝑥) = 𝑢 (0, 𝑥) + 𝑢 (0, 𝑥) = 𝑢 (0, 𝑥) . (15) Ψ (𝑡, 𝜔) 𝑢0 = 𝑆 (𝑡, 0 : 𝜔) 𝑢0 (22)

The estimation about the random parameter 𝑢 and the is cocycle. It is continuous RDS on 𝐻1 (R) over (Ω, F, 𝑃,
bilinear term VV𝑥 in (14) can be obtained by using the method (𝜃𝑡 )𝑡∈R ) and models the dynamical system associated with the
in [6]. Then by the fixed point argument the global existence stochastic equation (5) with initial value 𝑢(𝑠) = 𝑢𝑠 .
results to the random parameter Cauchy problem (14)-(15)
can be obtained. 4. Compact Random Absorbing Set
Theorem 3. Assume that Φ ∈ 𝐿0,0 0,2
2 (𝑜𝑟 Φ ∈ 𝐿 2 ); let V0 ∈ In the following computations 𝜔 ∈ Ω is fixed. We usually
2 1
𝐿 (R) (or V0 ∈ 𝐻 (R)). Then, the solution V(𝑡) of problem denote 𝐿𝑝 (R) (1 ⩽ 𝑝 ⩽ ∞) by 𝐿𝑝𝑥 ; 𝐿𝑝 ([0, 𝑇]; 𝐿𝑞 (R)) (1 ⩽
𝑝
(14) and (15) is global and belongs to 𝐶([0, 𝑇]; 𝐿2 (R)) 𝑝 ⩽ ∞, 1 ⩽ 𝑞 ⩽ ∞) by 𝐿 𝑡 (𝐿𝑞𝑥 ); and 𝐿𝑝 (R; 𝐿𝑞 ([0, 𝑇]))(1 ⩽
𝑝 𝑞
(𝑜𝑟 𝐶([0, 𝑇]; 𝐻1 (R)) for any 𝑇 > 0. 𝑝 ⩽ ∞, 1 ⩽ 𝑞 ⩽ ∞) by 𝐿 𝑥 (𝐿 𝑡 ) for any 𝑇 > 0 in this part.
First note that, for the Hilbert transform, we have for any
We summarize the above existence results for 𝑃-a.s. 𝜔 ∈ 𝑓, 𝑔 ∈ 𝐿2 (R)
Ω of (14) with initial condition V(𝑠, 𝑥) = 𝑢(𝑠, 𝑥) = V𝑠 , 𝑠 ∈ R,
as follows. H2 𝑓 = −𝑓, H (𝑓𝑔) = H (H𝑓H𝑔) + 𝑓H𝑔 + 𝑔H𝑓,

(i) Under the assumption of Theorem 3, for 𝑠 < 𝑇, and (𝑓, H𝑔) = − (H𝑓, 𝑔) , (H𝑓, 𝑓) = 0,
any 𝑇 ∈ R and V𝑠 ∈ 𝐿2 (R), there exists a unique 󵄩󵄩 󵄩 󵄩 󵄩
solution V ∈ 𝐶([𝑠, 𝑇]; 𝐿2 (R)). (H𝑓, H𝑔) = (𝑓, 𝑔) , 󵄩󵄩H𝑓󵄩󵄩󵄩 = 󵄩󵄩󵄩𝑓󵄩󵄩󵄩 ,
(ii) Under the assumption of Theorem 3, for 𝑠 < 𝑇, and ∀𝑓 ∈ 𝐻1 (R) , H𝑓𝑥 = (H𝑓)𝑥 .
any 𝑇 ∈ R, and V𝑠 ∈ 𝐻1 (R), there exists a unique (23)
solution V ∈ 𝐶([𝑠, 𝑇]; 𝐻1 (R)).
Before we prove the existence of a compact random
(iii) Denoting such a solution V(𝑡, 𝜔; 𝑠, V𝑠 ), the mapping absorbing set, we first give some estimates about the solution
V𝑠 󳨃→ V(𝑡, 𝜔; 𝑠, V𝑠 ) is continuous for all 𝑠 ⩽ 𝑇. 𝑢 of problem (𝑃).
Now we construct an RDS modeling the stochastic weakly Let us introduce the following space to study the solution
damped, forced KdV-BO equation. For example, consider a of problem (𝑃):
set of continuous functions with value 0 at t = 0 𝑋𝜎 (𝑇) = {𝑢 ∈ 𝐶 (0, 𝑇; 𝐻𝜎 (R)) ∩ 𝐿2 (R; 𝐿∞ ([0, 𝑇])) ,
Ω = {𝜔 ∈ 𝐶 (R, R) : 𝜔 (0) = 0} . (16)
𝐷𝜎 𝜕𝑥 𝑢 ∈ 𝐿∞ (R; 𝐿2 ([0, 𝑇])) ,
Let F be the Borel sigma-algebra induced by the compact
open topology of Ω, and let 𝑃 be a Wiener measure on (Ω, F). 𝜕𝑥 𝑢 ∈ 𝐿4 ([0, 𝑇] ; 𝐿∞ (R))} ,
We write (𝛽1 (𝑡, 𝜔), . . . , 𝛽𝑘 (𝑡, 𝜔), . . .) = 𝜔(𝑡). The time shift is (24)
simply defined by
for some 𝜎 < 1.
𝜃𝑠 𝜔 (𝑡) = 𝜔 (𝑡 + 𝑠) − 𝜔 (𝑠) , 𝑡, 𝑠 ∈ R, (17)
Lemma 4 (see [11]). Assume that Φ ∈ 𝐿0,𝜎 2 for some 𝜎 > 3/4;
and then (Ω, F, 𝑃, 𝜃𝑡 ) is an ergodic metric dynamical system then 𝑢 is almost surely in 𝑋𝜎 (𝑇) for any 𝑇 > 0 and any 𝜎 such
which models white noise. that 3/4 < 𝜎 < 𝜎. More precisely,
Having the mapping V𝑠 󳨃→ V(𝑡, 𝜔; 𝑠, V𝑠 ), we define
3
E ( sup ‖𝑢‖2𝐻𝑥𝜎 ) ⩽ 𝐶‖Φ‖2𝐿0,𝜎 , for any < 𝜎 ⩽ 𝜎,
𝑢 (𝑡, 𝜔; 𝑠, 𝑢𝑠 ) = 𝑆 (𝑡, 𝑠; 𝜔) 𝑢𝑠 = V (𝑡, 𝜔; 𝑠; V𝑠 ) + 𝑢 (𝑡, 𝜔) , (18) 𝑡∈[0,𝑇] 2 4

where V(𝑡, 𝜔; 𝑠; V𝑠 ) is a solution to (14) with V(𝑠) = V𝑠 and 𝑢(𝑡) 𝑇

󵄨 󵄨4
1/2

satisfies E ((∫ sup󵄨󵄨󵄨𝜕𝑥 𝑢󵄨󵄨󵄨 𝑑𝑡) ) ⩽ ‖Φ‖2𝐿0,𝜎 , (25)

0 𝑥∈R 2

𝑑𝑢 + {𝛼H𝜕𝑥2 𝑢 + 𝛽𝜕𝑥3 𝑢 + 𝜆𝑢} 𝑑𝑡 = Φ𝑑𝑊,

(19) E (∫ sup |𝑢|2 𝑑𝑥) ⩽ 𝐶 (𝜎) ‖Φ‖2𝐿0,𝜎 ;
𝑢 (𝑠) = 0. R 𝑡∈[0,𝑇] 2
4 Abstract and Applied Analysis

let 0 < 𝜀 < inf{𝜎, 2}, and then ⩽ ‖V (𝑠)‖2

󵄩 󵄩
𝑇 × exp {−(−1 − 𝑠)3/4 (𝜆(−1 − 𝑠)1/4 − 󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿4 (𝐿∞ ) )}
󵄨 󵄨2
E (sup ∫ 󵄨󵄨󵄨󵄨𝐷𝜎−𝜀 𝜕𝑥 𝑢󵄨󵄨󵄨󵄨 𝑑𝑡) ⩽ 𝐶 (𝜀) ‖Φ‖2𝐿0,𝜎 ;
𝑡 𝑥
(26)
𝑥∈R 0 2
2 −1 󵄩󵄩 󵄩󵄩2 −1
󵄩 󵄩
+ ∫ 󵄩󵄩𝑢𝑥 󵄩󵄩𝐿∞ ‖𝑢‖2 exp {− ∫ (𝜆 − 󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿∞ ) 𝑑𝜏} 𝑑𝑡
𝜆 𝑠 𝑥
𝑡 𝑥

one also has

2 −1 󵄩󵄩 󵄩󵄩2 −1
󵄩 󵄩
+ ∫ 󵄩󵄩𝑓󵄩󵄩 exp {− ∫ (𝜆 − 󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿∞ ) 𝑑𝜏} 𝑑𝑡
1/2 𝜆 𝑠 𝑡 𝑥
𝑇
4
E ((∫ sup|𝑢| 𝑑𝑡) ) ⩽ ‖Φ‖2𝐿0,𝜎 , (27) −(−1−𝑠)3/4 (𝜆(−1−𝑠)1/4 −‖Φ‖ )
0 𝑥∈R 2
⩽ ‖V(𝑠)‖2 𝑒 𝐿0,𝜎
2 + 𝐾1 .
(31)
where 𝐶(𝜀) and 𝐶(𝜎) depend on 𝜀, 𝜎, respectively. Noticing
−1 −1
Remark 5. We have to impose a stronger condition on Φ ∈ 󵄩 󵄩2 󵄩 󵄩
∫ 󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿∞ ‖𝑢‖2 exp {− ∫ (𝜆 − 󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿∞ ) 𝑑𝜏} 𝑑𝑡
𝐿0,2 0,1
2 in the present paper than that on Φ ∈ 𝐿 2 in [11]. Thus,
𝑠 𝑥
𝑡 𝑥

the solution 𝑢 of the linear problem is more regular, which

⩽ sup ‖𝑢‖
will be used in proving the boundedness of 𝑢 in 𝐻1 (R). More 𝑡∈(−1,𝑠)
precisely, we can get
−1 −1
󵄩 󵄩2 󵄩 󵄩
× ∫ 󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿∞ exp {− ∫ (𝜆 − 󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿∞ ) 𝑑𝜏} 𝑑𝑡 (32)
𝑥 𝑥
1/2 𝑠 𝑡
𝑇
󵄨 󵄨4
E ((∫ sup󵄨󵄨󵄨󵄨𝜕𝑥2 𝑢󵄨󵄨󵄨󵄨 𝑑𝑡) ) ⩽ ‖Φ‖2𝐿0,𝜎 ,
0 𝑥∈R 2 ⩽ ‖Φ‖2𝐿0,𝜎 (‖Φ‖2𝐿0,𝜎 + 𝐶) ,
2 2
(28)
𝑇 −1 −1
󵄨 󵄨2 󵄩 󵄩2 󵄩 󵄩 󵄩 󵄩2
E (sup ∫ 󵄨󵄨󵄨𝐷𝜕𝑥 𝑢󵄨󵄨󵄨 𝑑𝑡) ⩽ 𝐶‖Φ‖2𝐿0,𝜎 . ∫ 󵄩󵄩󵄩𝑓󵄩󵄩󵄩 exp {− ∫ (𝜆 − 󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿∞ ) 𝑑𝜏} 𝑑𝑡 ⩽ 𝐶󵄩󵄩󵄩𝑓󵄩󵄩󵄩 ,
0 2 𝑥
𝑥∈R 𝑠 𝑡

we get that 𝐾1 = 𝐶(‖Φ‖2𝐿0,𝜎 ‖Φ‖2𝐿0,𝜎 + ‖Φ‖2𝐿0,𝜎 + (2/𝜆)‖𝑓‖2𝐿2𝑥 ).

2 2 2
2
4.1. Absorption in 𝐿 (R) at Time 𝑡 = −1. Let 𝑠 < −1 and Then we have the following proposition.
𝑢𝑠 ∈ 𝐿2 (R) be given, and let V be the solution of (14) with
initial condition V(𝑠, 𝑥) = V𝑠 . Proposition 6. There exists a random radius 𝑟1 (𝜔) > 0, such
Multiplying (14) in 𝐿2 (R) by V, we obtain that for all 𝜌 > 0 there exists 𝑡(𝜔) ⩽ −1 such that the following
holds 𝑃-a.s. For all 𝑠 ⩽ 𝑡(𝜔) and all 𝑢𝑠 ∈ 𝐿2 (R) with ‖𝑢𝑠 ‖ ⩽ 𝜌,
1 𝑑 1 the solution 𝑢 of problem (5) with initial condition 𝑢(𝑠, 𝑥) = 𝑢𝑠
‖V‖2 + 𝜆‖V‖2 = ∫ (𝑓V − 𝑢𝑥 V2 − 𝑢𝑢𝑥 V) 𝑑𝑥. (29) satisfies the inequality
2 𝑑𝑡 R 2
‖𝑢 (−1, 𝜔)‖2 ⩽ 𝑟12 (𝜔) . (33)
It follows that Proof. Given 𝜌 > 0, there exists 𝑡(𝜔) such that
−(−1−𝑠)3/4 (𝜆(−1−𝑠)1/4 −‖Φ‖2 0,𝜎 ) 2
𝑑 󵄩 󵄩 𝑒 𝐿
𝜌 ⩽1 (34)
‖V‖2 + 𝜆‖V‖2 ⩽ 󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿∞ ‖V‖2
2

𝑑𝑡 𝑥
(30) for all 𝑠 ⩽ 𝑡(𝜔). Put
2 󵄩 󵄩2 2 󵄩 󵄩2
+ 󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿∞ ‖𝑢‖2 + 󵄩󵄩󵄩𝑓󵄩󵄩󵄩 . 𝑟12 (𝜔) = 1 + 𝐾1 + ‖𝑢 (−1)‖2 . (35)
𝜆 𝑥 𝜆
Then the proof of the proposition is completed.
Using Gronwall Lemma for 𝑠 ⩽ −1 and noticing Lemma 4, We can also get an auxiliary estimate from (30) by the
we get Gronwall Lemma with 𝑠 ⩽ 𝑡 ⩽ 0.
Consider the following:
‖V (−1)‖2 󵄩 󵄩
𝑡
‖V (𝑡)‖2 ⩽ ‖V (𝑠)‖2 exp {− ∫ (𝜆 − 󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿∞ (R) ) 𝑑𝜏}
−1 𝑠 𝑥
󵄩 󵄩
⩽ ‖V (𝑠)‖2 exp {− ∫ (𝜆 − 󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿∞ ) 𝑑𝜏} 𝑡
𝑠 𝑥
2 󵄩 󵄩2 󵄩 󵄩2
+ ∫ (󵄩󵄩𝑢 󵄩󵄩 ∞ ‖𝑢‖2 + 󵄩󵄩󵄩𝑓󵄩󵄩󵄩 ) (36)
−1
2 󵄩 󵄩2 2 󵄩 󵄩2 𝜆 𝑠 󵄩 𝑥 󵄩𝐿 𝑥
+ ∫ ( 󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿∞ ‖𝑢‖2 + 󵄩󵄩󵄩𝑓󵄩󵄩󵄩 ) −1
𝑠 𝜆 𝑥 𝜆 󵄩 󵄩
× exp {− ∫ (𝜆 − 󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿∞ ) 𝑑𝜏} 𝑑𝑡.
−1 𝑥
𝑡
󵄩 󵄩
× exp {− ∫ (𝜆 − 󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿∞ ) 𝑑𝜏} 𝑑𝑡 This inequality will be useful in the following proof.
𝑥
𝑡
Abstract and Applied Analysis 5

4.2. Absorption in 𝐻1 (R) at Time 𝑡=0. To obtain the 𝐻1 we deduce that

estimate, we multiply (14) by V𝑥𝑥 and integrate by part to get
𝑑 2 |𝛼| 󵄩 󵄩 󵄩 󵄩2
𝜑 (V) + (𝜆 − ‖𝑢‖𝐿∞𝑥 − 3󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿∞ ) 𝜑 (V) + 𝛽𝜆󵄩󵄩󵄩V𝑥 󵄩󵄩󵄩
1 𝑑 󵄩󵄩 󵄩󵄩2 1 󵄩 󵄩2 𝑑𝑡 𝛽 𝑥
󵄩V 󵄩 − ((V )𝑥 , V𝑥𝑥 ) + 𝜆󵄩󵄩󵄩V𝑥 󵄩󵄩󵄩
2
2 𝑑𝑡 󵄩 𝑥 󵄩 2
2 2 |𝛼| 󵄩 󵄩
3 (37) ⩽ −( 𝜆 + ‖𝑢‖𝐿∞𝑥 + 󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿∞ ) ∫ V3 𝑑𝑥
+ (𝑢𝑥 , V𝑥2 ) + (𝑢𝑥 , V𝑥 ) 3 3𝛽 𝑥
R
2
2 |𝛼| 󵄩 󵄩
+ (𝑢𝑢𝑥𝑥 , V𝑥 ) + (V𝑢𝑥𝑥 , V𝑥 ) − (𝑓𝑥 , V𝑥 ) = 0. − 𝛼 (𝜆 + ‖𝑢‖𝐿∞𝑥 + 3󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿∞ )
𝛽 𝑥

(14) × V2 𝑑𝑥 implies × ∫ VHV𝑥 𝑑𝑥 − 2𝛽 (𝑢2𝑥 + 𝑢𝑢𝑥𝑥 + V𝑢𝑥𝑥 − 𝑓𝑥 , V𝑥 )

R
1 𝑑 2
∫ V3 𝑑𝑥 − 𝛼 ∫ (V2 )𝑥 HV𝑥 𝑑𝑥 − (𝑢𝑢𝑥 + V𝑢𝑥 − 𝑓, V2 )
3 𝑑𝑡 R R 3
− 𝛽 ∫ (V2 )𝑥 V𝑥𝑥 𝑑𝑥 + 𝜆 ∫ V3 𝑑𝑥 (38) + 2𝛼 (𝑓 − 𝑢𝑢𝑥 , HV𝑥 ) .
R 𝑥 (43)
2
+ (𝑢𝑢𝑥 + V𝑢𝑥 + 𝑢V𝑥 − 𝑓, V ) = 0.
Now we stop to estimate the right hand side of (43) term by
Moreover term:

𝑑 󵄨󵄨 󵄨 󵄩 󵄩2 𝛽𝜆 󵄩󵄩 󵄩󵄩2
󵄨󵄨2𝛽 (V𝑢𝑥𝑥 , V𝑥 )󵄨󵄨󵄨 ⩽ 𝐶󵄩󵄩󵄩𝑢𝑥𝑥 󵄩󵄩󵄩𝐿∞𝑥 ‖V‖𝐿2𝑥 +
2
∫ VHV𝑥 𝑑𝑥 + ∫ (V2 )𝑥 HV𝑥 𝑑𝑥 + 2 ∫ 𝑢V𝑥 HV𝑥 𝑑𝑥 󵄩V 󵄩 2 ,
𝑑𝑡 R R R
12 󵄩 𝑥 󵄩𝐿 𝑥
(39)
󵄨󵄨 󵄨
− 2 (𝑓 − 𝜆V − 𝑢𝑢𝑥 , HV𝑥 ) = 0. 󵄨󵄨2𝛽 (𝑢2𝑥 , V𝑥 )󵄨󵄨󵄨 ⩽ 2󵄩󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩󵄩𝐿∞ 󵄩󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩󵄩𝐿2 󵄩󵄩󵄩󵄩V𝑥 󵄩󵄩󵄩󵄩𝐿2
󵄨 󵄨 𝑥 𝑥 𝑥

󵄩 󵄩2 󵄩 󵄩2 𝛽𝜆 󵄩󵄩 󵄩󵄩2
Combining (37), (38), and (39) (2𝛽(37) + (38) + 𝛼(39)), we ⩽ 𝐶󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿∞ 󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿2 + 󵄩V 󵄩 2 ,
have 𝑥 𝑥 12 󵄩 𝑥 󵄩𝐿 𝑥
󵄨󵄨 󵄨 󵄩 󵄩2 𝛽𝜆 󵄩󵄩 󵄩󵄩2
𝑑 󵄨 󵄨2 1 󵄩 󵄩2 󵄨󵄨2𝛽 (𝑢𝑢𝑥𝑥 , V𝑥 )󵄨󵄨󵄨 ⩽ 𝐶󵄩󵄩󵄩𝑢𝑢𝑥𝑥 󵄩󵄩󵄩𝐿2𝑥 + 󵄩V 󵄩 2 ,
∫ (𝛽󵄨󵄨󵄨V𝑥 󵄨󵄨󵄨 + V3 + 𝛼VHV𝑥 ) 𝑑𝑥 + 2𝛽𝜆󵄩󵄩󵄩V𝑥 󵄩󵄩󵄩 12 󵄩 𝑥 󵄩𝐿 𝑥
𝑑𝑡 R 3 󵄨󵄨 2 󵄨
󵄨󵄨− (V𝑢 , V2 )󵄨󵄨󵄨 + 󵄩󵄩󵄩𝑢 󵄩󵄩󵄩 ∞ ‖V‖3 3
󵄨󵄨 𝑥 󵄨󵄨 󵄩 𝑥 󵄩𝐿 𝑥 𝐿 (R)
+ 𝜆 ∫ V3 𝑑𝑥 + 2𝛼𝜆 ∫ VHV𝑥 𝑑𝑥 󵄨 3 󵄨
R R
󵄨 󵄨 󵄨󵄨 󵄨󵄨1/2
⩽ 󵄨󵄨󵄨𝑢𝑥 󵄨󵄨󵄨𝐿∞ |V|5/2 󵄨V 󵄨
𝐿2 󵄨 𝑥 󵄨𝐿2𝑥
+ 3𝛽 (𝑢𝑥 , V𝑥2 ) + 2𝛼 (𝑢, V𝑥 HV𝑥 ) (40)
𝑥 𝑥

󵄩 󵄩4 𝛽𝜆 󵄨󵄨 󵄨󵄨2
+ 2𝛽 (𝑢2𝑥 + 𝑢𝑢𝑥𝑥 + V𝑢𝑥𝑥 − 𝑓𝑥 , V𝑥 ) ⩽ 𝐶󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿∞ + |V|5𝐿2𝑥 + 󵄨V 󵄨 2 ,
𝑥 12 󵄨 𝑥 󵄨𝐿 𝑥
󵄨󵄨 󵄨
+ (𝑢𝑢𝑥 + V𝑢𝑥 + 𝑢V𝑥 − 𝑓, V2 ) 󵄨󵄨− (𝑢𝑢𝑥 , V2 )󵄨󵄨󵄨 ⩽ 󵄩󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩󵄩𝐿∞ ‖𝑢‖𝐿2 ‖V‖2𝐿4
󵄨 󵄨 𝑥 𝑥 𝑥

− 2𝛼 (𝑓 − 𝑢𝑢𝑥 , HV𝑥 ) = 0. 󵄩 󵄩 󵄩󵄩 󵄩󵄩1/2

⩽ 󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿∞ ‖𝑢‖𝐿2𝑥 ‖V‖3/2 󵄩V 󵄩
𝐿2 󵄩 𝑥 󵄩𝐿2
𝑥 𝑥 𝑥

Denote 󵄩 󵄩4/3 𝛽𝜆 󵄩󵄩 󵄩󵄩2

⩽ 󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿∞ ‖𝑢‖4/3 ‖V‖2𝐿2𝑥 + 󵄩V 󵄩 2
𝑥 𝐿2𝑥 12 󵄩 𝑥 󵄩𝐿 𝑥
󵄨 󵄨2 1
𝜑 (V) = ∫ (𝛽󵄨󵄨󵄨V𝑥 󵄨󵄨󵄨 + V3 + 𝛼VHV𝑥 ) 𝑑𝑥. (41) 𝛽𝜆 󵄩󵄩 󵄩󵄩2
3 󵄩 󵄩4
R ⩽ 𝐶 (𝜆) 󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿∞ + ‖𝑢‖2𝐿2𝑥 ‖V‖3𝐿2𝑥 + 󵄩V 󵄩 2 ,
𝑥 12 󵄩 𝑥 󵄩𝐿 𝑥
And noticing that 𝛽𝜆 󵄩󵄩 󵄩󵄩2
󵄨󵄨 󵄨 󵄩 󵄩2
󵄨󵄨−2𝛽 (𝑓𝑥 , V𝑥 )󵄨󵄨󵄨 ⩽ 𝐶 (𝜆) 󵄩󵄩󵄩𝑓𝑥 󵄩󵄩󵄩𝐿2𝑥 + 󵄩V 󵄩 2 ,
12 󵄩 𝑥 󵄩𝐿 𝑥
− (V𝑥 𝑢, V2 ) = 2 (V𝑥 𝑢, V2 ) + (V𝑢𝑥 , V2 ) ,
󵄨󵄨 󵄨 𝛽𝜆 󵄩󵄩 󵄩󵄩2
󵄨󵄨(𝑓, V2 )󵄨󵄨󵄨 ⩽ 𝐶 (𝜆) 󵄩󵄩󵄩󵄩𝑓󵄩󵄩󵄩󵄩4/3 2
𝐿2𝑥 ‖V‖𝐿2𝑥 + 󵄩V 󵄩 2 ,
1
− (V𝑥 𝑢, V ) = (V𝑢𝑥 , V2 ) ,
2 󵄨 󵄨 12 󵄩 𝑥 󵄩𝐿 𝑥
3 𝛽𝜆 󵄩󵄩 󵄩󵄩2
(42) 2
󵄨󵄨 󵄨 𝜆‖V‖3𝐿3 (R) ⩽ 𝐶 (𝜆) ‖V‖10/3 + 󵄩V 󵄩 2 ,
󵄨󵄨3 (𝑢𝑥 , V𝑥2 )󵄨󵄨󵄨 ⩽ 3󵄩󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩󵄩𝐿∞ 󵄩󵄩󵄩󵄩V𝑥 󵄩󵄩󵄩󵄩2𝐿2 , 3 𝐿2
𝑥 12 󵄩 𝑥 󵄩𝐿 𝑥
󵄨 󵄨 𝑥 𝑥

󵄨󵄨 󵄨 󵄩 󵄩2 2 |𝛼| 󵄩 󵄩2
󵄨󵄨2𝛼 (𝑢, V𝑥 HV𝑥 )󵄨󵄨󵄨 ⩽ 2 |𝛼| ‖𝑢‖𝐿∞ 󵄩󵄩󵄩V𝑥 󵄩󵄩󵄩 , ‖𝑢‖∞ ∫ |V|3 𝑑𝑥 ⩽ 𝐶󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩 + ‖𝑢‖2
3𝛽 R
6 Abstract and Applied Analysis

𝛽𝜆 󵄩󵄩 󵄩󵄩2 𝑡
+ ‖V‖20/3 + 󵄩 󵄩
󵄩V 󵄩 2 , × {‖V (𝑠)‖2 exp (− ∫ (𝜆 − 󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿∞ (R) ) 𝑑𝜏) + 𝐶}
12 󵄩 𝑥 󵄩𝐿 𝑥 𝑠 𝑥

󵄨󵄨 󵄨
󵄨󵄨𝛼𝜆 ∫ (VHV + 2𝑓HV ) 𝑑𝑥󵄨󵄨󵄨 0
󵄨󵄨 𝑥 𝑥 󵄨󵄨 × 𝑒− ∫𝑡 (𝜆−(2|𝛼|/𝛽)‖𝑢‖𝐿∞𝑥 −3‖𝑢𝑥 ‖𝐿∞𝑥 )𝑑𝜏 𝑑𝑡
󵄨 R 󵄨
0
󵄩 󵄩2 𝛽𝜆 󵄩󵄩 󵄩󵄩2 󵄩 󵄩2 −(−𝑠)3/4 ((−𝑠)1/4 𝜆−4‖𝑢𝑥 ‖𝐿4 (𝐿∞ ) )
⩽ 𝐶 (‖V‖2 + 󵄩󵄩󵄩𝑓󵄩󵄩󵄩 ) + 󵄩V 󵄩 2 , ⩽ ∫ 󵄩󵄩󵄩𝑢𝑥𝑥 󵄩󵄩󵄩𝐿∞ ‖V (𝑠)‖2 𝑒 𝑡 𝑥 𝑑𝑡
12 󵄩 𝑥 󵄩𝐿 𝑥 𝑠 𝑥

󵄨󵄨 2 󵄨󵄨󵄨 0
󵄨󵄨 2𝛼 󵄨󵄨 󵄩 󵄩2 0
󵄨󵄨 ‖𝑢‖ ∫ VHV 𝑑𝑥 󵄨󵄨 + 𝐶 ∫ 󵄩󵄩󵄩𝑢𝑥𝑥 󵄩󵄩󵄩𝐿∞ 𝑒− ∫𝑡 (𝜆−(2|𝛼|/𝛽)‖𝑢‖𝐿∞𝑥 −3‖𝑢𝑥 ‖𝐿∞𝑥 )𝑑𝜏 𝑑𝑡
󵄨󵄨󵄨 𝛽 ∞
R
𝑥
󵄨󵄨 𝑠 𝑥

󵄩 󵄩4
󵄩 󵄩2 𝛽𝜆 󵄩󵄩 󵄩󵄩2 ⩽ 𝐶󵄩󵄩󵄩𝑢𝑥𝑥 󵄩󵄩󵄩𝐿4 (𝐿∞ ) .
⩽ 𝐶󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩 + ‖𝑢‖2 ‖V‖4 + 󵄩V 󵄩 2 , 𝑡 𝑥
12 󵄩 𝑥 󵄩𝐿 𝑥 (47)
󵄨󵄨 󵄨 󵄨 󵄨
󵄨󵄨(𝑢𝑢𝑥 , HV𝑥 )󵄨󵄨󵄨 = 󵄨󵄨󵄨󵄨− (𝑢𝑥 , HV) − (𝑢𝑢𝑥𝑥 , HV)󵄨󵄨󵄨󵄨
2
We also have
0 +∞
󵄩 󵄩2 󵄩 󵄩4 𝛽𝜆 󵄩󵄩 󵄩󵄩2 󵄨 󵄨2
⩽ 𝐶 (‖V‖2 + 󵄩󵄩󵄩𝑢𝑢𝑥𝑥 󵄩󵄩󵄩 + 󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩 ) + 󵄩V 󵄩 2 . ∫ ∫ |𝑢|2 󵄨󵄨󵄨𝑢𝑥𝑥 󵄨󵄨󵄨 𝑑𝑥 𝑑𝑡
12 󵄩 𝑥 󵄩𝐿 𝑥 𝑠 −∞
(44) +∞ 0
󵄨 󵄨2
⩽∫ ess. sup |𝑢|2 ∫ 󵄨󵄨󵄨𝑢𝑥𝑥 󵄨󵄨󵄨 𝑑𝑥 𝑑𝑡
Then combining the above estimates, we get −∞ 𝑡∈[𝑠,0] 𝑠
(48)
𝑑 2 |𝛼| +∞ 0
󵄩 󵄩 󵄨 󵄨2
𝜑 (V) + (𝜆 − ‖𝑢‖𝐿∞𝑥 − 3󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿∞ ) 𝜑 (V) ⩽∫ ess. sup |𝑢|2 𝑑𝑥 ess.sup ∫ 󵄨󵄨󵄨𝑢𝑥𝑥 󵄨󵄨󵄨 𝑑𝑡.
𝑑𝑡 𝛽 𝑥
−∞ 𝑡∈[𝑠,0] 𝑠 𝑥∈R

󵄩 󵄩2 󵄩 󵄩2 󵄩 󵄩2
⩽ 𝐶 (󵄩󵄩󵄩𝑢𝑥𝑥 󵄩󵄩󵄩𝐿∞ ‖V‖2𝐿2𝑥 + 󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿∞ 󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿2 = ‖𝑢‖2𝑊𝑥2,∞ (𝐿2 ) ‖𝑢‖2𝐿2𝑥 (𝐿∞𝑡 ) .
𝑥 𝑥 𝑥 𝑡

󵄩 󵄩2 󵄩 󵄩2
+ 󵄩󵄩󵄩𝑢𝑢𝑥𝑥 󵄩󵄩󵄩𝐿2 + 󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿2 + ‖V‖5𝐿2𝑥 The boundedness of other terms about the right hand side of
𝑥 𝑥
(45) (46) is obvious. Noticing (25)–(28), we get
󵄩 󵄩4
+ 󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿∞ + ‖𝑢‖2𝐿2𝑥 ‖V‖3𝐿2𝑥 0
𝑥
𝜑 (V (0)) ⩽ 𝜑 (V (𝑠)) 𝑒− ∫𝑠 (𝜆−3‖𝑢𝑥 ‖𝐿∞𝑥 )𝑑𝜏 + 𝐶𝐾2
󵄩 󵄩4/3 (49)
+ 󵄩󵄩󵄩𝑓󵄩󵄩󵄩𝐿2 ‖V‖2𝐿2𝑥 + ‖V‖10/3
𝐿2
+ ‖𝑢‖2 0
𝑥 𝑥
⩽ 𝜑 (V (𝑠)) 𝑒− ∫𝑠 (𝜆−3‖𝑢𝑥 ‖𝐿∞𝑥 )𝑑𝜏 + 𝐶𝐾3 ,
+‖V‖20/3 + ‖𝑢‖2 ‖V‖4 ) . where
󵄩 󵄩2
Applying the Gronwall Lemma for 𝑠 ⩽ 0, we find 𝐾2 = 𝐶 (󵄩󵄩󵄩𝑢𝑥𝑥 󵄩󵄩󵄩𝑊𝑥2,∞ (𝐿2 ) ‖𝑢‖2𝐿2 (𝐿∞ )
𝑡 𝑥 𝑡
0
𝜑 (V (0)) ⩽ 𝜑 (V (𝑠)) 𝑒− ∫𝑠 (𝜆−(2|𝛼|/𝛽)‖𝑢‖𝐿∞𝑥 −3‖𝑢𝑥 ‖𝐿∞𝑥 )𝑑𝜏 󵄩 󵄩4
+ 󵄩󵄩󵄩𝑢𝑥𝑥 󵄩󵄩󵄩𝐿4 (𝐿∞ ) + ‖𝑢‖2𝐿∞ (𝐿2 ) ‖𝑢‖4𝐿4 (𝐿∞ )
𝑡 𝑥 𝑡 𝑥 𝑡 𝑥

0
󵄩 󵄩2 󵄩 󵄩2 󵄩 󵄩2 󵄩 󵄩4
+ 𝐶 ∫ (󵄩󵄩󵄩𝑢𝑥𝑥 󵄩󵄩󵄩𝐿∞ ‖V‖2𝐿2𝑥 + 󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿∞ 󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿2 + ‖𝑢‖2𝐿2 (𝐿∞ ) + 󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿4 (𝐿∞ )
𝑥 𝑥 𝑥 𝑥 𝑡 𝑡 𝑥
𝑠

󵄩 󵄩2 󵄩 󵄩4 󵄩 󵄩2
+ ‖𝑢‖2𝐿2𝑥 + 󵄩󵄩󵄩𝑢𝑢𝑥𝑥 󵄩󵄩󵄩𝐿2 + 󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿∞ + ‖V‖5𝐿2𝑥 +‖𝑢‖2𝐿2 (𝐿∞ ) + 󵄩󵄩󵄩𝑓𝑥 󵄩󵄩󵄩𝐿2 ) ,
𝑥 𝑥 𝑥 𝑥 𝑥

󵄩 󵄩4 󵄩 󵄩2 󵄩 󵄩2
+ 󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩𝐿∞ + ‖𝑢‖2𝐿2𝑥 ‖V‖3𝐿2𝑥 + 󵄩󵄩󵄩𝑓𝑥 󵄩󵄩󵄩𝐿2 𝐾3 = ‖Φ‖4𝐿0,𝜎 + ‖Φ‖6𝐿0,𝜎 + 󵄩󵄩󵄩𝑓𝑥 󵄩󵄩󵄩𝐿2 .
𝑥 𝑥 2 2 𝑥

󵄩 󵄩4/3 (50)
+󵄩󵄩󵄩𝑓󵄩󵄩󵄩𝐿2 ‖V‖2𝐿2𝑥 + ‖V‖10/3
𝐿2
)
𝑥 𝑥
Then we get the following proposition.
0
− ∫𝑡 (𝜆−(2|𝛼|/𝛽)‖𝑢‖𝐿∞ −3‖𝑢𝑥 ‖𝐿∞ )𝑑𝜏
×𝑒 𝑥 𝑥 𝑑𝑡.
Proposition 7. There exists a random variable 𝑟2 (𝜔) > 0,
(46) such that for all 𝜌 > 0 there exists ̃𝑡(𝜔) ⩽ 0 such that the
We can read the boundedness of the right hand side of following holds 𝑃-a.s. For all 𝑠 ⩽ ̃𝑡(𝜔) and all 𝑢𝑠 ∈ 𝐻1 (R)
(46) from the following estimates. Using (36), we have with ‖𝑢𝑠 ‖𝐻1 (R) ⩽ 𝜌, 𝜑 satisfies

0
󵄩 󵄩2 0 𝜑 (V (0)) ⩽ 𝑟2 (𝜔) . (51)
∫ 󵄩󵄩󵄩𝑢𝑥𝑥 󵄩󵄩󵄩𝐿∞ ‖V (𝑡)‖2𝐿2𝑥 𝑒− ∫𝑡 (𝜆−(2|𝛼|/𝛽)‖𝑢‖𝐿∞𝑥 −3‖𝑢𝑥 ‖𝐿∞𝑥 )𝑑𝜏 𝑑𝑡
𝑥
𝑠
Proof. Given 𝜌 > 0, there exists ̃𝑡(𝜔) such that
0
󵄩 󵄩2
⩽ ∫ 󵄩󵄩󵄩𝑢𝑥𝑥 󵄩󵄩󵄩𝐿∞ 0
𝜑 (V (𝑠)) 𝑒− ∫𝑠 (𝜆−(2|𝛼|/𝛽)‖𝑢‖𝐿∞𝑥 −3‖𝑢𝑥 ‖𝐿∞𝑥 )𝑑𝜏 ⩽ 1, (52)
𝑥
𝑠
Abstract and Applied Analysis 7

for all 𝑠 ⩽ ̃𝑡(𝜔). Put then 𝐴 is said to absorb 𝐵. Of course, if 𝐴 absorbs 𝐵, then 𝐴
attracts 𝐵.
𝑟2 (𝜔) = 1 + 𝐾3 . (53) Denote
󵄩 󵄩2
Then the proof of the proposition is completed. 𝐴 1 = {𝑢 ∈ 𝐻1 (R) , 󵄩󵄩󵄩𝑢𝑥 󵄩󵄩󵄩 ⩽ 𝑟3 (𝜔)} . (60)
Corollary 8. There exists a random variable 𝑟3 (𝜔) > 0, such It is easy to see that 𝐴 1 is an absorbing set for RDS {Ψ(𝑡, 𝜔)}
that the solution 𝑢(𝑡, 𝜔; 𝑠, 𝑢𝑠 ) of problem (5) with the initial in 𝐻1 (R). Furthermore it is a weak compactness according to
condition 𝑢(𝑠, 𝑥) = 𝑢𝑠 satisfies Lemma 9.
󵄩󵄩 󵄩
󵄩󵄩𝑢(0, 𝜔; 𝑠, 𝑢𝑠 )󵄩󵄩󵄩𝐻1 (R) ⩽ 𝑟3 (𝜔) . (54)
Proposition 10. Let
Proof. We can deduce from the presentation of 𝜑(V) that
A (𝜔) = ∪𝐵⊂𝑋 Λ 𝐵 (𝜔), (61)
2 󵄩󵄩 󵄩󵄩2 1 10/3
󵄩V 󵄩 − 8/3 ‖V‖ ⩽ 𝜑 (V) . (55)
3 󵄩 𝑥󵄩 2 where Λ 𝐵 (𝜔) := ∩𝑠⩾0 ∪𝑡⩾𝑠 𝜙(𝑡, 𝜃−𝑡 𝜔)𝐵 is the omega-limit set of
𝐵. Here the closures are taken with respect to the weak topology
Then of 𝐻1 (R). Then A(𝜔) is included in 𝐴 1 and is nonempty. It is
󵄩󵄩 󵄩2 3 3 invariant by Ψ(𝑡, 𝜔); that is,
󵄩󵄩V𝑥 (0)󵄩󵄩󵄩 ⩽ + 𝐾1 + 10/3 ‖V(−1)‖ .
10/3
(56)
2 2
Ψ (𝑡, 𝜔) A (𝜔) = A (𝜃𝑡 𝜔) , ∀𝑡 > 0. (62)
By V(𝑡) = 𝑢(𝑡) − 𝑢(𝑡),
The proof of Proposition 10 is easy using the following
󵄩󵄩 󵄩2 󵄩 󵄩2 󵄩 󵄩2
󵄩󵄩𝑢𝑥 (0)󵄩󵄩󵄩 ⩽ 2 (󵄩󵄩󵄩V𝑥 (0)󵄩󵄩󵄩 + 󵄩󵄩󵄩𝑢𝑥 (0)󵄩󵄩󵄩 ) facts.

3 3 󵄩 󵄩2 (a) Ψ(𝑡, 𝜔) is weakly continuous on 𝐻1 (R), which is an

⩽ 2 ( + 𝐾1 + 10/3 ‖V (−1)‖10/3 + 󵄩󵄩󵄩𝑢𝑥 (0)󵄩󵄩󵄩 ) obvious corollary of Lemma 9.
2 2
(b) A point 𝑏 ∈ A(𝜔) if and only if there exist two
= 𝑟3 (𝜔) . sequences (𝑠𝑛 )𝑛∈N ∈ R, (𝑥𝑛 )𝑛∈N ∈ A(𝜔) such that
(57) 𝑠𝑛 → −∞ and 𝜙(𝑠𝑛 )𝑥𝑛 → 𝑏.
Then the proof of the corollary is completed. Then A(𝜔) has the following properties:

4.3. Construct the Attractor. Before we construct the random (i) A(𝜔) is random compact set;
attractor, it is worthwhile to point out that the domain about (ii) Ψ(𝑡, 𝜔)A(𝜔) = A(𝜃𝑡 𝜔) for all 𝑡 ⩾ 0;
the space variable is unbounded which makes the embedding (iii) A(𝜔) attracts all deterministic bounded sets under
𝐻1 (R) → 𝐿2 (R) to be noncompact. Fortunately, we can the sense that
get weak compactness in the space 𝐻1 (R) by Lemma 9 given
below. Motivated by the definition of weak attractor for the lim 𝑑𝑤 (Ψ (𝑡, 𝜃𝑡 𝜔) 𝐵, A) = 0, (63)
𝑡→∞
deterministic system in [12], we extend this definition to the
RDS. Then we get weak random global attractor for (5) with where 𝑑𝑤 denotes the distance in the weak topology of 𝐻1 (R).
initial condition 𝑢(𝑠, 𝑥) = 𝑢𝑠 . Then we get Theorem 1 in this paper.
Lemma 9 (see [13]). The solution operator {𝑆(𝑡)}𝑡∈R is weakly
continuous in 𝐻1 (R) in the sense that if V0𝑛 converges weakly in Conflict of Interests
𝐻1 (R) to some V0 as 𝑛 → ∞, then 𝑆(𝑡)V0𝑛 converges to 𝑆(𝑡)V0 The author declares that there is no conflict of interests
weakly in 𝐻1 (R) for all 𝑡 ∈ R. regarding the publication of this paper.
Having Proposition 7, Corollary 8, and Lemma 9 in hand,
one begins to construct the weak random attractor in space Acknowledgment
𝐻1 (R). This work is supported by the National Natural Science
From Section 4.2, one gets an RDS for the stochastic Foundation of China (no. 11101309).
weakly damped, forced KdV-BO equation with additive
noise; that is,
References
𝑢 (𝑡, 𝜔; 𝑠, 𝑢𝑠 ) = 𝑆 (𝑡, 𝑠; 𝜔) 𝑢𝑠 = V (𝑡, 𝜔; 𝑠; V𝑠 ) + 𝑢 (𝑡, 𝜔) . (58)
[1] B. Guo and Z. Huo, “The well-posedness of the Korteweg-
One recalls in [8] that if 𝐵 and 𝐴 are random sets such that de Vries-Benjamin-Ono equation,” Journal of Mathematical
for 𝑃-almost all 𝜔 there exists a time 𝑇𝐵 (𝜔) such that for all Analysis and Applications, vol. 295, no. 2, pp. 444–458, 2004.
𝑡 ⩽ 𝑡𝐵 (𝜔) [2] F. Linares, “𝐿2 global well-posedness of the initial value problem
associated to the Benjamin equation,” Journal of Differential
Ψ (𝑡, 𝜃−𝑡 𝜔) 𝐵 (𝜃−𝑡 𝜔) ⊂ 𝐴 (𝜔) , (59) Equations, vol. 152, no. 2, pp. 377–393, 1999.
8 Abstract and Applied Analysis

[3] R. L. Herman, “The stochastic, damped KdV equation,” Journal

of Physics A: Mathematical and General, vol. 23, no. 7, pp. 1063–
1084, 1990.
[4] M. Scalerandi, A. Romano, and C. A. Condat, “Korteweg-de
Vries solitons under additive stochastic perturbations,” Physical
Review E: Statistical Physics, Plasmas, Fluids, and Related Inter-
disciplinary Topics, vol. 58, no. 4, pp. 4166–4173, 1998.
[5] B. Guo and Z. Huo, “The global attractor of the damped, forced
generalized Korteweg de Vries-Benjamin-Ono equation in 𝐿2 ,”
Discrete and Continuous Dynamical Systems A, vol. 16, no. 1, pp.
121–136, 2006.
[6] G. Wang and B. Guo, “Well-posedness of stochastic Korteweg-
de Vries-Benjamin-Ono equation,” Frontiers of Mathematics in
China, vol. 5, no. 1, pp. 161–177, 2010.
[7] L. Arnold, Random Dynamical Systems, Springer, Berlin, Ger-
many, 1998.
[8] H. Crauel and F. Flandoli, “Attractors for random dynamical
systems,” Probability Theory and Related Fields, vol. 100, no. 3,
pp. 365–393, 1994.
[9] H. Crauel, A. Debussche, and F. Flandoli, “Random attractors,”
Journal of Dynamics and Differential Equations, vol. 9, no. 2, pp.
307–341, 1997.
[10] G. da Prato and J. Zabczyk, Stochastic Equations in Infinite
Dimensions, Cambridge University Press, 1992.
[11] A. de Bouard and A. Debussche, “On the stochastic Korteweg-
de Vries equation,” Journal of Functional Analysis, vol. 154, no.
1, pp. 215–251, 1998.
[12] J.-M. Ghidaglia, “Finite-dimensional behavior for weakly
damped driven Schrödinger equations,” Annales de l’Institut
Henri Poincaré. Analyse Non Linéaire, vol. 5, no. 4, pp. 365–405,
1988.
[13] R. Rosa, “The global attractor of a weakly damped, forced
Korteweg-de Vries equation in 𝐻1 (R),” Matemática Contem-
porânea, vol. 19, pp. 129–152, 2000.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 726349, 7 pages
http://dx.doi.org/10.1155/2014/726349

Research Article
Global Analysis of a Virus Dynamics Model with General
Incidence Function and Cure Rate

Yu Yang
School of Science and Technology, Zhejiang International Studies University, Hangzhou 310012, China

Correspondence should be addressed to Yu Yang; yangyyj@126.com

Received 27 February 2014; Revised 16 April 2014; Accepted 16 April 2014; Published 30 April 2014

Academic Editor: Sanling Yuan

Copyright © 2014 Yu Yang. This is an open access article distributed under the Creative Commons Attribution License, which
permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

A virus dynamics model with logistic function, general incidence function, and cure rate is considered. By carrying out
mathematical analysis, we show that the infection-free equilibrium is globally asymptotically stable if the basic reproduction number
R0 ≤ 1. If R0 > 1, then the infection equilibrium is globally asymptotically stable under some assumptions. Furthermore, we also
obtain the conditions for which the model exists an orbitally asymptotically stable periodic solution. Examples are provided to
support our analytical conclusions.

1. Introduction DNA (cccDNA) from their nucleus [6]. Then HBV infection
has been modelled using the model including cytokine-
Mathematical models have proven valuable in understanding mediated “cure” of infected cells [7–9]. Wang et al. [10]
the virus dynamics. The basic viral infection model was considered an improved HBV model with standard incidence
proposed by Nowak et al. [1, 2] in the following form: function and cure rate. According to the virological basis
𝑥̇
= 𝜆 − 𝑑𝑥 − 𝛽𝑥V, found in [11], when a HIV enters a resting CD4+ T-cell,
𝑦̇
= 𝛽𝑥V − 𝑎𝑦, (1) the viral RNA may not be completely reverse transcribed
into DNA. If the cell is activated shortly following infection,
V̇
= 𝑘𝑦 − 𝑢V, reverse transcription can proceed to completion. However,
where 𝑥, 𝑦, and V denote uninfected cells, infected cells, the unintegrated virus harbored in resting cells may decay
and free virus particles, respectively. The uninfected cells are with time and partial DNA transcripts are labile and degrade
produced at a constant 𝜆 and die at rate 𝑑𝑥. The infected cells quickly [12]. Hence a proportion of resting infected cells can
are produced from uninfected cells and free virus at rate 𝛽𝑥V revert to the uninfected state [13]. Recently, Zhou et al. [14]
and die at rate 𝑎𝑦. Free virus is produced from infected cells considered a model of HIV infection of CD4+ T-cells with
at rate 𝑘𝑦 and declines at rate 𝑢V. bilinear incidence function and cure rate. Hattaf et al. [15]
The incidence function in model (1) is based on the law considered a virus dynamics model with general incidence
of mass action. However, many researchers suggested that function and cure rate. However, Tian and Liu [16] have
the bilinear incidence function is not sufficient to describe pointed out that the proof for the main results in [15] is
the infection process in detail, and some nonlinear incidence not corrected. They introduced a more general nonlinear
functions were proposed. For example, Li and Ma [3] con- incidence function including the form in [15].
sidered a HIV-1 model with Holling type II function. Min The population dynamics of target cells is not well under-
et al. [4] considered a HBV model with standard incidence stood. Many models with logistic uninfected cell proliferation
function. Elaiw [5] considered a virus dynamics model with terms have been introduced. For example, Culshaw and Ruan
a more general nonlinear incidence function, which satisfies [17] considered a delay-differential equation model of HIV
some conditions. infection of CD4+ T-cells with logistic function term. Ji et al.
When HBV infects a cell, infected cells may also revert [18] considered a viral infection model of HBV infection with
to the uninfected state by loss of all covalently closed circular logistic function. Li and Shu [19] considered an in-host model
2 Abstract and Applied Analysis

with a logistic mitosis term for the uninfected target cells and which implies that
a finite intracellular delay in the incidence term. 𝛿1
In this paper, we aim to study the following model with 𝐿 1 (𝑡) ≤ + (𝐿 (0) − 𝛿1 ) 𝑒−𝛿2 𝑡 , (5)
logistic function, general incidence function, and cure rate: 𝛿2
where 𝛿1 = 𝜆 + 𝑟𝑥max /4 and 𝛿2 = min{𝑎, 𝑑}. Obviously, 𝑥(𝑡)
𝑥 and 𝑦(𝑡) are bounded. From the third equation of system (2)
𝑥̇
= 𝜆 − 𝑑𝑥 + 𝑟𝑥 (1 − ) − 𝑓 (𝑥, 𝑦, V) V + 𝜌𝑦,
𝑥max and the boundedness of 𝑦(𝑡), it is easy to see that V(𝑡) is also
(2)
𝑦̇= 𝑓 (𝑥, 𝑦, V) V − (𝑎 + 𝜌) 𝑦, bounded. This completes the proof.
V̇
= 𝑘𝑦 − 𝑢V, It can then be verified that the bounded set
with initial condition (𝑥(0), 𝑦(0), V(0)) ∈ R3+ . Here 𝜆,𝑑,𝑎,𝑘, 𝛿 𝑘𝛿
Ω = {(𝑥, 𝑦, V) ∈ R3+ : 0 < 𝑥 + 𝑦 ≤ 1 , 0 < V ≤ 1 } (6)
and 𝑢 defined as earlier. 𝑟 is the maximum proliferation rate 𝛿2 𝑢𝛿2
of uninfected cells and 𝑥max is the maximum capacity of host’s is positively invariant with respect to system (2) and is convex.
organ cells. 𝑓(𝑥, 𝑦, V)V is the incidence of new infections. 𝜌 ≥ Following the computation of the basic reproduction
0 is the rate of cure, that is, noncytolytic loss of infected cells. number in [21], we have the basic reproduction number
The function 𝑓(𝑥, 𝑦, V) satisfies the following conditions:
𝑘𝑓 (𝑥0 , 0, 0)
(i) 𝑓(0, 𝑦, V) = 0 for all 𝑦 ≥ 0 and V ≥ 0; R0 = . (7)
𝑢 (𝑎 + 𝜌)
(ii) 𝜕𝑓(𝑥, 𝑦, V)/𝜕𝑥 > 0, for all 𝑥 > 0, 𝑦 ≥ 0, and V ≥ 0;
For mathematical convenience, let
(iii) 𝜕𝑓(𝑥, 𝑦, V)/𝜕𝑦 ≤ 0 and 𝜕𝑓(𝑥, 𝑦, V)/𝜕V ≤ 0, for all 𝑥 ≥
𝑥
0, 𝑦 ≥ 0, and V ≥ 0; 𝑛 (𝑥) = 𝜆 − 𝑑𝑥 + 𝑟𝑥 (1 − ), (8)
𝑥max
(iv) (𝜕𝑓(𝑥, 𝑦, V)/𝜕V)V + 𝑓(𝑥, 𝑦, V) > 0, for all 𝑥 > 0, 𝑦 ≥ 0,
and V ≥ 0. and 𝑥0 is the positive root of 𝑛(𝑥) = 0. It is clear that 𝑛󸀠 (𝑥0 ) <
0 and system (2) has an infection-free equilibrium 𝐸0 (𝑥0 , 0, 0)
Several models for viral dynamics fit model (2). For when R0 ≤ 1.
example, Song and Neumann [20] considered a viral model
with 𝑓(𝑥, 𝑦, V) = 𝛽𝑥/(1 + 𝛼𝑥) and 𝜌 = 0, where 𝛽 is the Theorem 2. If R0 > 1 and 𝑑 − 𝑟 + 2𝑟𝑥/𝑥max > 0 for
infection rate constant and 𝛼 is a positive constant. Ji et al. arbitrary 𝑥 ≥ 0, then there exists a unique infection equilibrium
[18] considered a viral infection model of HBV infection with 𝐸∗ (𝑥∗ , 𝑦∗ , V∗ ).
𝑓(𝑥, 𝑦, V) = 𝛽𝑥 and 𝜌 = 0. Zhou et al. [14] considered a model
of HIV infection of CD4+ T-cells with 𝑓(𝑥, 𝑦, V) = 𝛽𝑥. Proof. In order to find the infection equilibrium, set
The paper is organized as follows. In Section 2, we carry
out mathematical analysis of the model. In Section 3, the 𝑛 (𝑥) − 𝑓 (𝑥, 𝑦, V) V + 𝜌𝑦 = 0,
local stability of equilibria is proved. In Section 4, the global 𝑓 (𝑥, 𝑦, V) V − (𝑎 + 𝜌) 𝑦 = 0, (9)
stability of the infection-free equilibrium or the infection 𝑘𝑦 − 𝑢V = 0,
equilibrium is established, respectively. The conditions for the
existence of an orbitally asymptotically stable periodic solu- which yields
tion are obtained. In Section 5, two examples are provided to 𝑎𝑢
𝑛 (𝑥) = 𝑎𝑦 = V. (10)
illustrate our theorems. The conclusion is given in Section 6. 𝑘
Substituting the expression of 𝑦 and V by 𝑥, we have
2. Mathematical Analysis 𝑛 (𝑥) 𝑘𝑛 (𝑥) 𝑢 (𝑎 + 𝜌)
𝐺 (𝑥) = 𝑓 (𝑥, , )− . (11)
First, we show that solution of system (2) is bounded. 𝑎 𝑎𝑢 𝑘
It is obvious that 𝐺(0) = −𝑢(𝑎 + 𝜌)/𝑘 < 0 and
Theorem 1. The solution of system (2) is positive and bounded.
𝑢 (𝑎 + 𝜌)
Proof. Let 𝐺 (𝑥0 ) = 𝑓 (𝑥0 , 0, 0) −
𝑘 (12)
𝐿 1 (𝑡) = 𝑥 (𝑡) + 𝑦 (𝑡) . (3) 𝑢 (𝑎 + 𝜌)
= (R0 − 1) > 0.
𝑘
Computing the derivation of 𝐿 1 along the solution of system This implies that there exists 𝑥∗ ∈ (0, 𝑥0 ) such that 𝐺(𝑥∗ ) =
(2), we have 0. Suppose, to the contrary, there exists another infection
𝑥 equilibrium 𝐸∗ (𝑥∗ , 𝑦∗ , V∗ ). Without loss of generality, we
𝐿̇
1 = 𝜆 − 𝑑𝑥 + 𝑟𝑥 (1 − ) − 𝑎𝑦, assume that 𝑥∗ < 𝑥∗ . Since 𝑑 − 𝑟 + 2𝑟𝑥/𝑥max > 0, we have
𝑥max
2
𝑛(𝑥∗ ) > 𝑛(𝑥∗ ). This yields 𝑦∗ > 𝑦∗ and V∗ > V∗ . Hence,
𝑟 𝑥 𝑟𝑥 we get 𝑓(𝑥∗ , 𝑦∗ , V∗ ) < 𝑓(𝑥∗ , 𝑦∗ , V∗ ). On the other hand,
= 𝜆 − 𝑑𝑥 − (𝑥 − max ) + max − 𝑎𝑦 (4)
𝑥max 2 4 we have that 𝑓(𝑥∗ , 𝑦∗ , V∗ ) = 𝑓(𝑥∗ , 𝑦∗ , V∗ ) = 𝑢(𝑎 + 𝜌)/𝑘.
𝑟𝑥max This is a contradiction. Therefore, 𝐸∗ is the unique infection
≤𝜆+ − min {𝑎, 𝑑} 𝐿 1 ,
4 equilibrium.
Abstract and Applied Analysis 3

3. Local Stability of Equilibria One eigenvalue of 𝐽(𝐸0 ) is 𝜆 1 = 𝑛󸀠 (𝑥0 ) < 0. The remaining
two eigenvalues are solutions of the quadratic equation
In this section, we discuss the local stability of the infection-
free equilibrium 𝐸0 and the infection equilibrium 𝐸∗ of
system (2), respectively. 𝜆2 + (𝑎 + 𝑢 + 𝜌) 𝜆 + 𝑢 (𝑎 + 𝜌) (1 − R0 ) = 0. (14)

Theorem 3. If R0 ≤ 1, then the infection-free equilibrium By the Routh-Hurwitz theorem, 𝐸0 is locally asymptotically
𝐸0 is locally asymptotically stable and becomes unstable when stable when R0 ≤ 1. When R0 > 1, 𝐽(𝐸0 ) has a positive
R0 > 1. eigenvalue and 𝐸0 is unstable.
Proof. The Jacobian matrix of system (2) at 𝐸0 is Theorem 4. If R0 > 1 and 𝑑 − 𝑟 + 2𝑟𝑥/𝑥max > 0 for arbitrary
𝑥 ≥ 0, then the unique endemic equilibrium 𝐸∗ is locally
𝑛󸀠 (𝑥0 ) 𝜌 −𝑓 (𝑥0 , 0, 0) asymptotically stable.
𝐽 (𝐸0 ) = ( 0 − (𝑎 + 𝜌) 𝑓 (𝑥0 , 0, 0) ) . (13)
0 𝑘 −𝑢 Proof. The Jacobian matrix of system (2) at 𝐸∗ is

𝜕𝑓 (𝑥∗ , 𝑦∗ , V∗ ) 𝜕𝑓 (𝑥∗ , 𝑦∗ , V∗ ) 𝜕𝑓 (𝑥∗ , 𝑦∗ , V∗ )

𝑛󸀠 (𝑥∗ ) − V∗ − V∗ + 𝜌 − V∗ − 𝑓 (𝑥∗ , 𝑦∗ , V∗ )
𝜕𝑥 𝜕𝑦 𝜕V
𝐽 (𝐸∗ ) = ( 𝜕𝑓 (𝑥∗ , 𝑦∗ , V∗ ) 𝜕𝑓 (𝑥∗ , 𝑦∗ , V∗ ) 𝜕𝑓 (𝑥∗ , 𝑦∗ , V∗ ) ). (15)
V∗ V∗ − (𝑎 + 𝜌) V∗ + 𝑓 (𝑥∗ , 𝑦∗ , V∗ )
𝜕𝑥 𝜕𝑦 𝜕V
( 0 𝑘 −𝑢 )

The characteristic equation of 𝐽(𝐸∗ ) can be written as Theorem 5. If R0 ≤ 1, then the infection-free equilibrium 𝐸0
is globally asymptotically stable.
𝜆3 + 𝑎1 𝜆2 + 𝑎2 𝜆 + 𝑎3 = 0, (16)
Proof. Consider a Lyapunov function
with
𝑎1 = 𝑎 + 𝜌 + 𝑢 − 𝑛󸀠 (𝑥∗ ) 𝑎+𝜌
𝐿 2 (𝑦, V) = 𝑦 + V. (19)
𝜕𝑓 (𝑥∗ , 𝑦∗ , V∗ ) 𝜕𝑓 (𝑥∗ , 𝑦∗ , V∗ ) 𝑘
+ V∗ − V∗ ,
𝜕𝑥 𝜕𝑦 Calculating the time derivative of 𝐿 2 (𝑦, V) along solutions of
𝜕𝑓 (𝑥∗ , 𝑦∗ , V∗ ) system (2), we obtain
𝑎2 = (𝑎 + 𝑢) V∗
𝜕𝑥
𝜕𝑓 (𝑥∗ , 𝑦∗ , V∗ ) 𝑢 (𝑎 + 𝜌)
− 𝑛󸀠 (𝑥∗ ) (𝑎 + 𝜌 + 𝑢 − V∗ ) 𝐿̇
2 (𝑦, V) = (𝑓 (𝑥, 𝑦, V) − )V
𝜕𝑦 (17) 𝑘
𝜕𝑓 (𝑥∗ , 𝑦∗ , V∗ ) 𝜕𝑓 (𝑥∗ , 𝑦∗ , V∗ ) 𝑢 (𝑎 + 𝜌)
−𝑘 V∗ − 𝑢 V∗ , ≤ (𝑓 (𝑥0 , 0, 0) − )V (20)
𝜕V 𝜕𝑦 𝑘
𝜕𝑓 (𝑥∗ , 𝑦∗ , V∗ ) 𝜕𝑓 (𝑥∗ , 𝑦∗ , V∗ )
𝑎3 = 𝑎𝑢 V∗ + 𝑢𝑛󸀠 (𝑥∗ ) V∗ 𝑢 (𝑎 + 𝜌)
𝜕𝑥 𝜕𝑦 = (R0 − 1) V.
𝑘
𝜕𝑓 (𝑥∗ , 𝑦∗ , V∗ )
+ 𝑘𝑛󸀠 (𝑥∗ ) V∗ .
𝜕V If R0 < 1, from Corollary 5.2 of Kuang [22], 𝐸0 is globally
asymptotically stable. Also, for R0 = 1, 𝐿̇2 (𝑦, V) = 0 implies
Since 𝑑 − 𝑟 + 2𝑟𝑥/𝑥max > 0, we have 𝑛󸀠 (𝑥∗ ) < 0. Therefore, that 𝑥(𝑡) = 𝑥0 . It is easy to show that the largest invariant
𝑎1 > 0, 𝑎2 > 0, 𝑎3 > 0. set where 𝐿̇2 (𝑦, V) = 0 is the singleton {𝐸0 }. By the LaSalle
(18)
invariance principle, 𝐸0 is globally asymptotically stable.
By direct calculation, we have 𝑎1 𝑎2 − 𝑎3 > 0. Then the Routh-
Hurwitz theorem implies that the infection equilibrium 𝐸∗ is Next, we prove that the infection equilibrium 𝐸∗ is glob-
locally asymptotically stable. ally asymptotically stable. We need the following theorem.

4. Global Stability of Equilibria Theorem 6. If R0 > 1, then system (2) is uniformly persistent.

In this section, we study the global stability of the infection- Proof. The result follows from an application of Theorem 4.6
free equilibrium 𝐸0 and the infection equilibrium 𝐸∗ of in [23], with 𝑋1 = int(R3+ ) and 𝑋2 = 𝜕R3+ . We only need to
system (2), respectively. prove that 𝐸0 is a weak repeller for 𝑋1 .
4 Abstract and Applied Analysis

Suppose that there exists a solution (𝑥(𝑡), 𝑦(𝑡), V(𝑡)) such System (23) always has a trivial equilibrium (0, 0). Since R0 >
that (𝑥(𝑡), 𝑦(𝑡), V(𝑡)) → 𝐸0 . When 𝑡 is sufficiently large, we 1 and continuously differentiable of the function 𝑓(𝑥, 𝑦, V),
have we have

𝑥0 − 𝜀 < 𝑥 < 𝑥0 + 𝜀, 𝑘𝑓 (𝑥0 − 𝜀, 0, 0)

(21) R𝜀0 := > 1, (24)
𝑢 (𝑎 + 𝜌)
where 𝜀 is an arbitrarily small positive constant satisfying 0 <
𝜀 < 𝑥0 . Then, for some 𝜀 > 0 sufficiently small. By calculation, system (23)
has a unique positive equilibrium (𝑢1∗ , 𝑢2∗ ), where 𝑢1∗ satisfies
𝑦̇
≥ 𝑓 (𝑥0 − 𝜀, 𝑦, V) V − (𝑎 + 𝜌) 𝑦. (22) the root equation
Consider the following auxiliary system: 𝑘𝑢1 𝑢 (𝑎 + 𝜌)
𝑓 (𝑥0 − 𝜀, 𝑢1 , )− = 0. (25)
𝑢̇
1 = 𝑓 (𝑥0 − 𝜀, 𝑢1 , 𝑢2 ) 𝑢2 − (𝑎 + 𝜌) 𝑢1 𝑢 𝑘
(23)
𝑢̇
2 = 𝑘𝑢1 − 𝑢𝑢2 . The Jacobian matrix of system (23) at (𝑢1∗ , 𝑢2∗ ) is

𝜕𝑓 (𝑥0 − 𝜀, 𝑢1∗ , 𝑢2∗ ) ∗ 𝜕𝑓 (𝑥0 − 𝜀, 𝑢1∗ , 𝑢2∗ ) ∗

𝑢2 − (𝑎 + 𝜌) 𝑢2 + 𝑓 (𝑥0 − 𝜀, 𝑢1∗ , 𝑢2∗ )
𝐽(𝑢1∗ ,𝑢2∗ ) =( 𝜕𝑢1 𝜕𝑢2 ). (26)
𝑘 −𝑢

The eigenvalues 𝜆 1 and 𝜆 2 of 𝐽(𝑢1∗ ,𝑢2∗ ) satisfy By looking at the Jacobian matrix of system (2) and
choosing the matrix 𝐻 as

𝜕𝑓 (𝑥0 − 𝜀, 𝑢1∗ , 𝑢2∗ ) ∗ 1 0 0

𝜆1 + 𝜆2 = 𝑢2 − (𝑎 + 𝜌 + 𝑢) , (0 −1 0) , (29)
𝜕𝑢1
0 0 1
𝜕𝑓 (𝑥0 − 𝜀, 𝑢1∗ , 𝑢2∗ ) ∗
𝜆1𝜆2 = − 𝑢 𝑢2 (27) system (2) is competitive in Ω, with respect to the partial
𝜕𝑢1 order defined by the orthant 𝐾 = {(𝑥, 𝑦, V) ∈ R3 : 𝑥 ≥ 0, 𝑦 ≤
0, V ≥ 0}.
𝜕𝑓 (𝑥0 − 𝜀, 𝑢1∗ , 𝑢2∗ ) ∗ Theorem 7. Suppose R0 > 1 and 𝑑 − 𝑟 + 2𝑟𝑥/𝑥max > 0 for
−𝑘 𝑢2 .
𝜕𝑢2 arbitrary 𝑥 ≥ 0; then the infection equilibrium 𝐸∗ of system (2)
is globally asymptotically stable.
We get 𝜆 1 + 𝜆 2 < 0 and 𝜆 1 𝜆 2 > 0. Hence, (𝑢1∗ , 𝑢2∗ ) is locally Proof. Let 𝑃(𝑡) = (𝑥(𝑡), 𝑦(𝑡), V(𝑡)) be a periodic solution
asymptotically stable when R0 > 1. whose orbit is contained in int(R3+ ). The second compound
Denote the right-hand sides of system (23) by 𝐵1 and 𝐵2 , equation is the following periodic linear system:
respectively. We have
𝜕𝑓[2]
𝑍󸀠 (𝑡) = (𝑃 (𝑡)) 𝑍 (𝑡) , (30)
𝜕𝐵1 𝜕𝐵2 𝜕𝑓 (𝑥0 − 𝜀, 𝑢1∗ , 𝑢2∗ ) ∗ 𝜕𝑥
+ = 𝑢2
𝜕𝑢1 𝜕𝑢2 𝜕𝑢1 (28) where 𝑍 = (𝑍1 , 𝑍2 , 𝑍3 )𝑇 and 𝜕𝑓[2] /𝜕𝑥 is the second additive
compound matrix of the Jacobian matrix of system (2).
− (𝑎 + 𝜌 + 𝑢) < 0. The Jacobian matrix of system (2) is

𝑛󸀠 (𝑥) − 𝑚1 −𝑚2 + 𝜌 −𝑚3

Therefore, (𝑢1∗ , 𝑢2∗ ) is globally asymptotically stable by the
Bendixson criterion for two-dimensional ordinary differen- 𝐽=( 𝑚1 𝑚2 − (𝑎 + 𝜌) 𝑚3 ) , (31)
tial equations. By the comparison theorem, we have that 0 𝑘 −𝑢
𝑦(𝑡) ≥ 𝑢1∗ as 𝑡 → ∞ for system (23). This is a contradiction
to 𝑦(𝑡) → 0. Hence, 𝐸0 is a weak repeller for 𝑋1 . and its second additive compound matrix is

𝑛󸀠 (𝑥) − 𝑚1 + 𝑚2 − (𝑎 + 𝜌) 𝑚3 𝑚3
[2]
𝐽 =( 𝑘 𝑛󸀠 (𝑥) − 𝑚1 − 𝑢 −𝑚2 + 𝜌 ), (32)
0 𝑚1 𝑚2 − (𝑎 + 𝜌 + 𝑢)
Abstract and Applied Analysis 5

where Hence,
𝜕𝑓 (𝑥, 𝑦, V) 𝜕𝑓 (𝑥, 𝑦, V) V
𝑚1 = V, 𝑚2 = V, 𝑔1 (𝑡) = 𝑛󸀠 (𝑥) − 𝑚1 + 𝑚2 − (𝑎 + 𝜌) + 𝑚3
𝜕𝑥 𝜕𝑦 𝑦
(33)
𝜕𝑓 (𝑥, 𝑦, V)
𝑚3 = V + 𝑓 (𝑥, 𝑦, V) . 𝑦̇ 󸀠 V 𝜕𝑓 (𝑥, 𝑦, V)
𝜕V = + 𝑛 (𝑥) − 𝑚1 + 𝑚2 + V
𝑦 𝑦 𝜕V
For the solution 𝑃(𝑡), system (30) becomes

𝑍̇ 󸀠 𝑦̇
1 = (𝑛 (𝑥) − 𝑚1 + 𝑚2 − (𝑎 + 𝜌)) ≤ − (−𝑛󸀠 (𝑥)) (41)
𝑦
× 𝑍1 + 𝑚3 (𝑍2 + 𝑍3 ) ,
(34)
𝑍̇ 󸀠
2 = 𝑘𝑍1 + (𝑛 (𝑥) − 𝑚1 − 𝑢) 𝑍2 + (𝜌 − 𝑚2 ) 𝑍3 ,
𝑦̇ ∗
≤ −𝛼 ,
𝑍̇ 𝑦
3 = 𝑚1 𝑍2 + (𝑚2 − (𝑎 + 𝜌 + 𝑢)) 𝑍3 .

Now, define the function 𝑘𝑦 𝑦̇ V̇ 𝑦̇

𝑔2 (𝑡) = + ( − − 𝜂 − 𝑢) = − 𝜂.
𝑉 (𝑡) = 𝑉 (𝑍1 , 𝑍2 , 𝑍3 ; 𝑥, 𝑦, V) V 𝑦 V 𝑦
󵄨 󵄨 𝑦 󵄨 󵄨 󵄨 󵄨 (35)
= sup {󵄨󵄨󵄨𝑍1 󵄨󵄨󵄨 , (󵄨󵄨󵄨𝑍2 󵄨󵄨󵄨 + 󵄨󵄨󵄨𝑍3 󵄨󵄨󵄨)} , Therefore,
V
𝑦̇
which is a Lyapunov function for system (30). Then, we have sup {𝑔1 (𝑡) , 𝑔2 (𝑡)} ≤ − 𝜂. (42)
𝑦
󵄨󵄨󵄨𝑍̇󵄨󵄨󵄨 ≤ (𝑛󸀠 (𝑥) − 𝑚 + 𝑚 − (𝑎 + 𝜌)) 󵄨󵄨󵄨𝑍 󵄨󵄨󵄨
󵄨󵄨 1 󵄨󵄨 1 2 󵄨 1󵄨
󵄨 󵄨 󵄨 󵄨 We have
+ 𝑚3 (󵄨󵄨󵄨𝑍2 󵄨󵄨󵄨 + 󵄨󵄨󵄨𝑍3 󵄨󵄨󵄨) ,
󵄨󵄨 ̇󵄨󵄨 𝜔
󵄨󵄨𝑍2 󵄨󵄨 ≤ 𝑘 󵄨󵄨󵄨󵄨𝑍1 󵄨󵄨󵄨󵄨 + (𝑛󸀠 (𝑥) − 𝑚1 − 𝑢) 󵄨󵄨󵄨󵄨𝑍2 󵄨󵄨󵄨󵄨 + (𝜌 − 𝑚2 ) 󵄨󵄨󵄨󵄨𝑍3 󵄨󵄨󵄨󵄨 , 󵄨𝜔
∫ sup {𝑔1 (𝑡) , 𝑔2 (𝑡)} 𝑑𝑡 ≤ log 𝑦 (𝑡)󵄨󵄨󵄨0 − 𝜂𝜔
󵄨 󵄨
󵄨󵄨 ̇󵄨󵄨 0
󵄨󵄨𝑍3 󵄨󵄨 ≤ 𝑚1 󵄨󵄨󵄨󵄨𝑍2 󵄨󵄨󵄨󵄨 + (𝑚2 − (𝑎 + 𝜌 + 𝑢)) 󵄨󵄨󵄨󵄨𝑍3 󵄨󵄨󵄨󵄨 . (43)
󵄨 󵄨
(36) = −𝜂𝜔 < 0,
From system (36), we have which implies that 𝑉(𝑡) → 0 as 𝑡 → ∞. This means that
𝑦 󵄨 󵄨 󵄨 󵄨 (𝑍1 (𝑡), 𝑍2 (𝑡), 𝑍3 (𝑡)) → (0, 0, 0) as 𝑡 → ∞, so the linear
𝐷+ (󵄨󵄨󵄨𝑍2 󵄨󵄨󵄨 + 󵄨󵄨󵄨𝑍3 󵄨󵄨󵄨) system (34) is asymptotically stable and the periodic solution
V
𝑦 𝑦̇ V̇ 󵄨 󵄨 󵄨 󵄨 𝑦 󵄨 󵄨 󵄨 󵄨 is asymptotically orbitally stable.
= ( − ) (󵄨󵄨󵄨𝑍2 󵄨󵄨󵄨 + 󵄨󵄨󵄨𝑍3 󵄨󵄨󵄨) + 𝐷+ (󵄨󵄨󵄨𝑍2 󵄨󵄨󵄨 + 󵄨󵄨󵄨𝑍3 󵄨󵄨󵄨) According to Theorem 4.1 in [24], system (2) satisfies the
V 𝑦 V V
Poincare-Bendixson property. Using Theorem 1, Theorems 4
𝑦 𝑦̇ V̇ 󵄨󵄨 󵄨󵄨 󵄨󵄨 󵄨󵄨 𝑘𝑦 󵄨󵄨 󵄨󵄨 and 6, we have that all conditions of Theorem 2.2 in [25] are
≤ ( − ) (󵄨󵄨𝑍2 󵄨󵄨 + 󵄨󵄨𝑍3 󵄨󵄨) + 󵄨𝑍 󵄨 (37)
V 𝑦 V V 󵄨 1󵄨 satisfied for system (2). This completes the proof.
𝑦 󵄨 󵄨 󵄨 󵄨
− ((𝑢 − 𝑛󸀠 (𝑥)) 󵄨󵄨󵄨𝑍2 󵄨󵄨󵄨 + (𝑎 + 𝑢) 󵄨󵄨󵄨𝑍3 󵄨󵄨󵄨)
V If the condition 𝑑 − 𝑟 + 2𝑟𝑥/𝑥max > 0 in Theorem 4 could
𝑘𝑦 󵄨󵄨 󵄨󵄨 𝑦 𝑦̇ V̇ 󵄨 󵄨 󵄨 󵄨 not be satisfied, there would exist an orbitally asymptotically
≤ 󵄨𝑍 󵄨 + ( − − 𝜂 − 𝑢) (󵄨󵄨󵄨𝑍2 󵄨󵄨󵄨 + 󵄨󵄨󵄨𝑍3 󵄨󵄨󵄨) ,
V 󵄨 1󵄨 V 𝑦 V stable periodic solution. We have the following theorem.

where 0 < 𝛼∗ = −max𝑥∈[0,𝑥0 ] 𝑛󸀠 (𝑥) and 𝜂 = min{𝛼∗ , 𝑎}. Theorem 8. Suppose R0 > 1 and 𝑎1 𝑎2 < 𝑎3 ; then system (2)
Therefore, has an orbitally asymptotically stable periodic solution.

𝐷+ 𝑉 (𝑡) ≤ sup {𝑔1 (𝑡) , 𝑔2 (𝑡)} 𝑉 (𝑡) , (38) Proof. The nonlinearities in system (2) are analytic in Ω.
We obtain that the conclusion follows from Theorem 1.2 in
where [26]. Take the domain for system (2) to be the interior of
V the positive orthant, in which the only steady state is 𝐸∗ . If
𝑔1 (𝑡) = 𝑛󸀠 (𝑥) − 𝑚1 + 𝑚2 − (𝑎 + 𝜌) + 𝑚3 ,
𝑦 R0 > 1 and 𝑎1 𝑎2 < 𝑎3 , then 𝐸∗ is unstable. The dissipativity
(39)
𝑘𝑦 𝑦̇ V̇ hypothesis of Theorem 1.2 in [26] follows from Theorems 1
𝑔2 (𝑡) = + ( − − 𝜂 − 𝑢) . and 6. System (2) is competitive in Ω and |𝐽𝐸∗ | = −𝑎3 < 0.
V 𝑦 V
Hence, all conditions of Theorem 1.2 in [26] are satisfied. This
From the second and third equations of system (2), we have completes the proof.
𝑦̇ 𝑓 (𝑥, 𝑦, V) V
= − (𝑎 + 𝜌) , 5. Examples
𝑦 𝑦
(40)
V̇ 𝑘𝑦 In this section, we give two examples to show the application
= − 𝑢.
V V of our theorems.
6 Abstract and Applied Analysis

Example 1. Consider the following system: Conflict of Interests

𝑥 𝛽𝑥V The author declares that there is no conflict of interests regar-
𝑥̇
= 𝜆 − 𝑑𝑥 + 𝑟𝑥 (1 − )− ,
𝑥max 𝑥+𝑦 ding the publication of this paper.
𝛽𝑥V (44)
𝑦̇
= − 𝑎𝑦, Acknowledgment
𝑥+𝑦
This work was supported by the Special Fund for the High-
V̇
= 𝑘𝑦 − 𝑢V,
Level Project Cultivation in Zhejiang International Studies
which is a special case of system (2) by letting 𝑓(𝑥, 𝑦, V) = University (090500442012).
𝛽𝑥/(𝑥 + 𝑦) and 𝜌 = 0. This model has been investigated by
Ji et al. [18]. Applying Theorem 5, Theorems 7 and 8, we have References
the following result.
[1] M. A. Nowak and C. R. M. Bangham, “Population dynamics of
Theorem 9. immune responses to persistent viruses,” Science, vol. 272, no.
5258, pp. 74–79, 1996.
(i) If R0 ≤ 1, then the infection-free equilibrium 𝐸0 (𝑥0 , [2] M. A. Nowak, S. Bonhoeffer, A. M. Hill, R. Boehme, H. C.
0, 0) of system (44) is globally asymptotically stable. Thomas, and H. Mcdade, “Viral dynamics in hepatitis B virus
(ii) If R0 > 1 and 𝑟 ≤ 𝑑, then the unique infection infection,” Proceedings of the National Academy of Sciences of the
equilibrium 𝐸∗ (𝑥∗ , 𝑦∗ , V∗ ) of system (44) is globally United States of America, vol. 93, no. 9, pp. 4398–4402, 1996.
asymptotically stable. [3] D. Li and W. Ma, “Asymptotic properties of a HIV-1 infection
model with time delay,” Journal of Mathematical Analysis and
(iii) If R0 > 1 and 𝑎1 𝑎2 < 𝑎3 , then system (44) exists an Applications, vol. 335, no. 1, pp. 683–691, 2007.
orbitally asymptotically stable periodic solution.
[4] L. Min, Y. Su, and Y. Kuang, “Mathematical analysis of a basic
virus infection model with application to HBV infection,” Rocky
Example 2. Consider the following system:
Mountain Journal of Mathematics, vol. 38, no. 5, pp. 1573–1585,
𝑥 2008.
𝑥̇
= 𝜆 − 𝑑𝑥 + 𝑟𝑥 (1 − ) − 𝛽𝑥V + 𝜌𝑦, [5] A. M. Elaiw, “Global properties of a class of virus infection mod-
𝑥max
els with multitarget cells,” Nonlinear Dynamics, vol. 69, pp. 423–
(45)
𝑦̇
= 𝛽𝑥V − (𝑎 + 𝜌) 𝑦, 435, 2012.
[6] L. G. Guidotti, R. Rochford, J. Chung, M. Shapiro, R. Purcell,
V̇
= 𝑘𝑦 − 𝑢V, and F. V. Chisari, “Viral clearance without destruction of infec-
ted cells during acute HBV infection,” Science, vol. 284, no. 5415,
which is a special case of system (2) by letting 𝑓(𝑥, 𝑦, V) = pp. 825–829, 1999.
𝛽𝑥. This model has been studied by Zhou et al. [14]. By [7] S. M. Ciupe, R. M. Ribeiro, P. W. Nelson, and A. S. Perelson,
Theorem 5, Theorem 7, and Theorem 8, we get the following “Modeling the mechanisms of acute hepatitis B virus infection,”
result. Journal of Theoretical Biology, vol. 247, no. 1, pp. 23–35, 2007.
[8] H. Dahari, E. Shudo, R. M. Ribeiro, and A. S. Perelson,
Theorem 10. “Modeling complex decay profiles of hepatitis B virus during
antiviral therapy,” Hepatology, vol. 49, no. 1, pp. 32–38, 2009.
(i) If R0 ≤ 1, then the infection-free equili-
brium 𝐸0 (𝑥0 , 0, 0) of system (45) is globally asym- [9] A. S. Perelson, “Modelling viral and immune system dynamics,”
Nature Reviews Immunology, vol. 2, no. 1, pp. 28–36, 2002.
ptotically stable.
[10] K. Wang, A. Fan, and A. Torres, “Global properties of an
(ii) If R0 > 1 and 𝑑 − 𝑟 + 2𝑟𝑥/𝑥max > 0 for arbitrary 𝑥 ≥ 0, improved hepatitis B virus model,” Nonlinear Analysis: Real
then the unique infection equilibrium 𝐸∗ (𝑥∗ , 𝑦∗ , V∗ ) of World Applications, vol. 11, no. 4, pp. 3131–3138, 2010.
system (45) is globally asymptotically stable. [11] J. A. Zack, S. J. Arrigo, S. R. Weitsman, A. S. Go, A. Haislip, and I.
(iii) If R0 > 1 and 𝑎1 𝑎2 < 𝑎3 , then system (45) exists an S. Y. Chen, “HIV-1 entry into quiescent primary lymphocytes:
orbitally asymptotically stable periodic solution. molecular analysis reveals a labile, latent viral structure,” Cell,
vol. 61, no. 2, pp. 213–222, 1990.
[12] J. A. Zack, A. M. Haislip, P. Krogstad, and I. S. Y. Chen, “Inco-
6. Conclusion mpletely reverse-transcribed human immunodeficiency virus
type 1 genomes in quiescent cells can function as intermediates
In this paper, we have considered a virus dynamics model in the retroviral life cycle,” Journal of Virology, vol. 66, no. 3, pp.
with logistic function, general incidence function, and cure 1717–1725, 1992.
rate. The basic reproduction number is obtained and it [13] P. Essunger and A. S. Perelson, “Modeling HIV infection of
determines the global dynamics of this model. If R0 ≤ 1, CD4+ t-cell subpopulations,” Journal of Theoretical Biology, vol.
then the infection-free equilibrium is globally asymptotically 170, no. 4, pp. 367–391, 1994.
stable. If R0 > 1, then the virus persists in the host, [14] X. Zhou, X. Song, and X. Shi, “A differential equation model of
and solutions approach either an infection equilibrium or HIV infection of CD4+ T-cells with cure rate,” Journal of Math-
a periodic orbit. Our model is a generalization of several ematical Analysis and Applications, vol. 342, no. 2, pp. 1342–1355,
models that appeared in the literature as its special cases. 2008.
Abstract and Applied Analysis 7

[15] K. Hattaf, N. Yousfi, and A. Tridane, “Mathematical analysis of a

virus dynamics model with general incidence rate and cure
rate,” Nonlinear Analysis: Real World Applications, vol. 13, no. 4,
pp. 1866–1872, 2012.
[16] Y. Tian and X. Liu, “Global dynamics of a virus dynamical
model with general incidence rate and cure rate,” Nonlinear
Analysis: Real World Applications, vol. 16, pp. 17–26, 2014.
[17] R. V. Culshaw and S. Ruan, “A delay-differential equation model
of HIV infection of CD4+ T-cells,” Mathematical Biosciences,
vol. 165, no. 1, pp. 27–39, 2000.
[18] Y. Ji, L. Min, and Y. Ye, “Global analysis of a viral infection mo-
del with application to HBV infection,” Journal of Biological Sys-
tems, vol. 18, no. 2, pp. 325–337, 2010.
[19] M. Y. Li and H. Shu, “Joint effects of mitosis and intracellular
delay on viral dynamics: two-parameter bifurcation analysis,”
Journal of Mathematical Biology, vol. 64, no. 6, pp. 1005–1020,
2012.
[20] X. Song and A. U. Neumann, “Global stability and periodic sol-
ution of the viral dynamics,” Journal of Mathematical Analysis
and Applications, vol. 329, no. 1, pp. 281–297, 2007.
[21] P. van den Driessche and J. Watmough, “Reproduction numbers
and sub-threshold endemic equilibria for compartmental mod-
els of disease transmission,” Mathematical Biosciences, vol. 180,
pp. 29–48, 2002.
[22] Y. Kuang, Delay Differential Equations with Applications in Pop-
ulation Dynamics, Academic Press, Boston, Mass, USA, 1993.
[23] H. R. Thieme, “Persistence under relaxed point-dissipativity
(with application to an endemic model),” SIAM Journal on Mat-
hematical Analysis, vol. 24, pp. 407–435, 1993.
[24] H. L. Smith, Monotone Dynamical Systems, An Introduction To
the Theory of Competitive and Cooperative Systems, American
Mathematical Society, Providence, RI, USA, 1995.
[25] M. Y. Li and L. Wang, “Global stability in some SEIR epidemic
models,” in Mathematical Approaches For Emerging and Reem-
erging Infectious Diseases: Models, Methods, and Theory, vol. 126
of The IMA Volumes in Mathematics and its Applications, pp.
295–311, 2002.
[26] H. R. Zhu and H. L. Smith, “Stable periodic orbits for a class of
three dimensional competitive systems,” Journal of Differential
Equations, vol. 110, no. 1, pp. 143–156, 1994.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 482450, 11 pages
http://dx.doi.org/10.1155/2014/482450

Research Article
The Problem of Bicenter and Isochronicity for a
Class of Quasi Symmetric Planar Systems

Du Chaoxiong
Department of Mathematics, Hunan Shaoyang University, Shaoyang, Hunan 422000, China

Correspondence should be addressed to Du Chaoxiong; ducx123@126.com

Received 19 February 2014; Revised 5 April 2014; Accepted 5 April 2014; Published 29 April 2014

Academic Editor: Tonghua Zhang

Copyright © 2014 Du Chaoxiong. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

We study a class of quasi symmetric seventh degree systems and obtain the conditions that its two singular points can be two centers
at the same step by careful computing and strict proof. In addition, the condition of an isochronous center is also given. In terms
of quasi symmetric systems, our work is interesting and obtained conclusions about bicenters are new.

1. Introduction appropriate analytic change is very difficult to be obtained,

so only a handful of isochronous systems are investigated.
One of the open problems for planar polynomial differential Several classes of known studied isochronous systems are as
systems follows: quadratic isochronous centers (see [14]); isochronous
centers of a linear center perturbed by third, fourth, and fifth
𝑑𝑥 𝑑𝑦 degree homogeneous polynomials (see [3, 4, 15]); complex
= 𝑃 (𝑥, 𝑦) , = 𝑄 (𝑥, 𝑦) (1)
𝑑𝑡 𝑑𝑡 polynomial systems (see [1]); reversible systems (see [12, 16]);
and isochronous centers of quartic systems with degenerate
is how to characterize their centers and isochronous centers. infinity (see [17]).
Article [1] pointed out that “a center of an analytic system is For seventh degree system, [18] studied the limit cycles
isochronous if and only if there exists an analytic change of bifurcations. In this paper, we investigate the centers and
coordinates such that the original system is reduced to a linear isochronous centers problem for a class of seventh degree
system,” so an isochronous center is also called a linearizable systems with the following form:
center. A center is an isochronous center or linearizable center
if the period of all periodic solutions is constant.
𝑑𝑥 3
The main method to investigate centers and isochronous = − (𝑥𝛿 + 𝑦) (𝑥2 + 𝑦2 ) − 𝑃6 (𝑥, 𝑦) ,
centers problem is the computation of focus values and 𝑑𝑡
(2)
isochronous constants (see [2–13]), which is a kind of active 𝑑𝑦 3
effective method. The vanishing of all isochronous constants = (𝑥 − 𝑦𝛿) (𝑥2 + 𝑦2 ) + 𝑄6 (𝑥, 𝑦) ,
or period constants is a necessary and sufficient condition 𝑑𝑡
for the isochronicity. Although theoretically the isochronous
in which
center problem can be solved by using the method letting
all period constants become zero, in fact only the first few
period constants can be given in personal computer. Hence, 𝑃6 (𝑥, 𝑦) = (1 + 𝑎1 ) 𝑥4 𝑦 + (3 + 2𝑎1 ) 𝑥5 𝑦 + 2𝑎4 𝑥3 𝑦2
up to now the sufficient and necessary condition determining
an isochronous center can only be found by making some + 4𝑎4 𝑥4 𝑦2 − (𝑎1 − 𝑎3 − 2𝑎5 ) 𝑥2 𝑦3
appropriate analytic changes of coordinates which let the
original system be reduced to a linear system. This kind of + (3 + 2𝑎5 ) × 𝑥3 𝑦3 + 2𝑥 (𝑎6 + 2𝑎4 𝑥) 𝑦4
2 Abstract and Applied Analysis

1
− (𝑎3 + 2𝑎1 𝑥 − 2𝑎5 𝑥) 𝑦5 + 𝑥2 (𝑥 − 𝑦) 2. Preliminary Method to Compute Focal
2 Values and Periodic Constant
× (𝑥 + 𝑦) (𝑥 + 3𝑥2 + 3𝑦2 ) 𝛿, In order to continue this study, at first we introduce previous
1 5 methods to calculate focal values and periodic constants
𝑄6 (𝑥, 𝑦) = 𝑥 (1 + 3𝑥) + 𝑎4 𝑥4 (1 + 2𝑥) 𝑦 which are necessary for us to verify centers and isochronous
2
centers.
1 Consider the following real system:
− (1 + 4𝑎1 − 2𝑎5 ) 𝑥3 𝑦2 − (4𝑎1 − 𝑎5 ) 𝑥4 𝑦2
2
∞
− (𝑎4 − 𝑎6 ) 𝑥2 𝑦3 − (2𝑎3 + 𝑎5 ) 𝑥𝑦4 𝑑𝑥
= 𝛿𝑥 − 𝑦 + ∑ 𝑋𝑘 (𝑥, 𝑦) ,
𝑑𝑡 𝑘=2
1
− (3 + 8𝑎1 ) 𝑥2 𝑦4 − 𝑦5 (𝑎6 + 2𝑎4 𝑥 + 𝑎5 𝑦) ∞
(4)
2 𝑑𝑦
= 𝑥 + 𝛿𝑦 + ∑ 𝑌𝑘 (𝑥, 𝑦) ,
𝑑𝑡
− 𝑥3 𝑦 (𝑥 + 3𝑥2 + 3𝑦2 ) 𝛿, 𝑘=2

(3) where 𝑋𝑘 (𝑥, 𝑦) and 𝑌𝑘 (𝑥, 𝑦) are homogeneous polynomials

and 𝛿, 𝑎𝑖 (𝑖 ∈ {1, 2, 3, 4, 5}) are real numbers. of degree 𝑘 about 𝑥 and 𝑦.
We obtain that the infinity and the elementary singular By means of transformation
point (−1/2, 0) of (2)|𝛿=0 have the same center condition
and investigate the isochronous center condition of (−1/2, 0). 𝑧 = 𝑥 + 𝑖𝑦, 𝑤 = 𝑥 − 𝑖𝑦, 𝑇 = 𝑖𝑡, 𝑖 = √−1, (5)
What is worth pointing out is that the results of bicenters in
a polynomial system of degree 𝑛 are less seen in published system (4)|𝛿=0 can be transformed into the following complex
papers; our work is new and interesting. system:
In general, our investigations are shown as follows. Firstly,
by making two appropriate transformations (i.e., (32) and ∞
𝑑𝑧
(33)) of system (2), system (2) is transformed into system = 𝑧 + ∑ 𝑍𝑘 (𝑧, 𝑤) = 𝑍 (𝑧, 𝑤) ,
(34); hence, the problem of system (2) center problem and 𝑑𝑇 𝑘=2
isochronicity is reduced to investigate system (34) center and (6)
∞
the isochronous centers problem. Secondly, we prove that 𝑑𝑤
= −𝑤 − ∑ 𝑊𝑘 (𝑧, 𝑤) = −𝑊 (𝑧, 𝑤) ,
system (34) is symmetric about (−1, 0). System (34) has two 𝑑𝑇 𝑘=2
symmetric elementary singular points (i.e., the origin and
(−2, 0)), which are from the infinity and the elementary focus where 𝑧, 𝑤, 𝑇 are complex variables and
(−1/2, 0) of (2) under transformations (32) and (33). Thirdly,
through calculating system (34) focal values when 𝛿 = 0 and 𝑍𝑘 (𝑧, 𝑤) = ∑ 𝑎𝛼𝛽 𝑧𝛼 𝑤𝛽 , 𝑊𝑘 (𝑧, 𝑤) = ∑ 𝑏𝛼𝛽 𝑤𝛼 𝑧𝛽 .
careful analysis, we obtain the condition that the infinity and 𝛼+𝛽=𝑘 𝛼+𝛽=𝑘
the elementary focus (−1/2, 0) of (2)|𝛿=0 become bicenters (7)
at the same time. Lastly, we study the above isochronicity
problems of (−1/2, 0). During the course of investigating
isochronicity of system (2)|𝛿=0 , at first we make use of the Obviously, the coefficients of (6) satisfy conjugate condition;
method in [19] to compute the first several period constants that is,
and find the isochronous centers’ necessary condition; next
we try to find the sufficient condition. We obtain all sufficient 𝑎𝛼𝛽 = 𝑏𝛼𝛽 , 𝛼 ≥ 0, 𝛽 ≥ 0, 𝛼 + 𝛽 ≥ 2. (8)
and necessary conditions that the elementary focus (−1/2, 0)
of (2) become an isochronous center. System (4)|𝛿=0 and system (6) are called concomitant systems
The paper is organized as follows. In Section 2, we intro- (for the definition see [11, 12]).
duce preliminary methods to calculate focal values (or Lya- For the complex analytic system (6), making transforma-
punov constants) and period constants which are necessary tion
for our study in Sections 3 and 4. In Section 3, we make two
appropriate transformations which let research on system (2) 𝑧 = 𝑟𝑒𝑖𝜃 , 𝑤 = 𝑟𝑒−𝑖𝜃 , 𝑇 = 𝑖𝑡, (9)
be reduced to investigate a class of symmetric seventh degree
systems in which the first five focal values with more simple system (6) can be transformed into
expressions are given. Being based on it, we find the condition
that the infinity and the elementary singular points of (2)|𝛿=0 ∞
𝑑𝑟 𝑤𝑍 − 𝑧𝑊 𝑤𝑍𝑘+1 − 𝑧𝑊𝑘+1
can be bicenters and prove them. In Section 4, by analyzing =𝑖 = 𝑖𝑟 ∑
𝑑𝑡 2𝑟 𝑘=1
2𝑧𝑤
all center conditions and all obtained expressions of periodic
constants, we give all sufficient and necessary conditions that
𝑖𝑟 ∞
the elementary singular point (−1/2, 0) of (2)|𝛿=0 becomes an = ∑ ∑ (𝑎 −𝑏 ) 𝑒𝑖(𝛼−𝛽)𝜃 𝑟𝑚 ,
isochronous center and prove them strictly. 2 𝑚=1 𝛼+𝛽=𝑚+2 𝛼,𝛽−1 𝛽,𝛼−1
Abstract and Applied Analysis 3

∞ 2𝑚+2
𝑑𝜃 𝑤𝑍 + 𝑧𝑊 𝑤𝑍𝑘+1 + 𝑧𝑊𝑘+1
= =1+∑ 𝜇𝑚 = ∑ [(𝑚 − 𝑘 + 2) 𝑎𝑘,𝑗−1
𝑑𝑡 2𝑧𝑤 𝑘=1
2𝑧𝑤
𝑘+𝑗=3
∞
1 − (𝑚 − 𝑗 + 2) 𝑏𝑗,𝑘−1 ] 𝑐𝑚−𝑘+2,𝑚−𝑗+2 .
=1+ ∑ ∑ (𝑎 +𝑏 ) 𝑒𝑖(𝛼−𝛽)𝜃 𝑟𝑚 .
2 𝑚=1 𝛼+𝛽=𝑚+2 𝛼,𝛽−1 𝛽,𝛼−1
(17)
(10)
According to the relation between systems (4) and (6), in And 𝜇𝑘 in Lemma 1 is called 𝑘th order singular point value
fact transformation (9) can be regarded as the following real at the origin of system (6).
polar coordinate transformation of (4)|𝛿=0 :
Lemma 2 (see [20]). For system (4) and any positive integer
𝑥 = 𝑟 cos 𝜃, 𝑦 = 𝑟 sin 𝜃. (11) 𝑚, among V2𝑚 (2𝜋), V𝑘 (2𝜋), and V𝑘 (𝜋), there exists the following
Under transformation (11), from (10) we have relation:
∞
𝑑𝑟 (𝑖𝑟/2) ∑𝑚=1 ∑𝛼+𝛽=𝑚+2 (𝑎𝛼,𝛽−1 − 𝑏𝛽,𝛼−1 ) 𝑒𝑖(𝛼−𝛽)𝜃 𝑟𝑚 1
= . V2𝑚 (2𝜋) =
𝑑𝜃 1 + (1/2) ∑∞
𝑚=1 ∑𝛼+𝛽=𝑚+2 (𝑎𝛼,𝛽−1 + 𝑏𝛽,𝛼−1 ) 𝑒
𝑖(𝛼−𝛽)𝜃 𝑟𝑚 1 + V1 (𝜋)
(12) 𝑚−1
(18)
(0) (𝑘)
For the complex constant ℎ, |ℎ| ≪ 1, we write the solution × [𝜉𝑚 (V1 (2𝜋) − 1) + ∑ 𝜉𝑚 V2𝑘+1 (2𝜋)] ,
of (12) associated with the initial condition 𝑟|𝜃=0 = ℎ as 𝑘=1

∞
(𝑘)
𝑟 = 𝑟̃ (𝜃, ℎ) = ℎ + ∑ V𝑘 (𝜃) ℎ𝑘 , (13) where 𝜉𝑚 are all polynomials of V1 (𝜋), V2 (𝜋), . . . , V𝑚 (𝜋) and
𝑘=2 V1 (2𝜋), V2 (2𝜋), . . . , V𝑚 (2𝜋) with rational coefficients.
in which V2𝑘+1 (2𝜋) (𝑘 = 1, 2, . . .) are called the 𝑘th focal value
of the origin of (4). Obviously, We can imply that V2𝑚 (2𝜋) = 0 when V1 (2𝜋) =
From (13), it is clear that the origin of (4) is a center 1, V2𝑘+1 (2𝜋) = 0, 𝑘 = 1, 2, . . . , 𝑚 − 1.
if and only if all V2𝑘+1 (2𝜋) = 0 (𝑘 = 1, 2, . . .). Hence the
Lemma 3 (see [20]). For system (4)|𝛿=0 , (6), and any positive
computation of focal value plays an important role for settling
integer 𝑚, the following assertion holds:
the center problem. Liu and Li [19] gave some methods to
compute focal values. Next we will introduce our method to
𝑚−1
calculate focal value through the following three lemmas. (𝑘)
V2𝑘+1 (2𝜋) = 𝑖𝜋 (𝜇𝑚 + ∑ 𝜉𝑚 𝜇𝑘 ) , (19)
Lemma 1 (see [20, 21]). For system (6), one can derive 𝑘=1

successively the terms of the following formal series:

(𝑘)
∞ where 𝜉𝑚 , (𝑘 = 1, 2, . . . , 𝑚 − 1) are polynomial functions of
𝛼 𝛽
𝑀 = 1 + ∑ 𝑐𝛼𝛽 𝑧 𝑤 , (14) coefficients of system (6).
𝛼+𝛽=1
Obviously, the origin of system (4) is a center if and only if
such that its all focal values vanish, namely, V2𝑘+1 = 0, 𝑘 ∈ N. According
∞
𝜕𝑀 𝜕𝑀 𝜕𝑍 𝜕𝑊 to Lemmas 2 and 3, we have the following lemma.
𝑍− 𝑊+( − ) 𝑀 = ∑ (𝑚 + 1) 𝜇𝑚 (𝑧𝑤)𝑚 ,
𝜕𝑧 𝜕𝑤 𝜕𝑧 𝜕𝑤 𝑚=1
(15) Lemma 4 (see [20]). For systems (4)|𝛿=0 and (6), the origin is
a center if and only if the following relation holds:
where 𝑐11 = 1, 𝑐20 = 𝑐02 = 0, for all 𝑐𝑘𝑘 ∈ 𝑅, 𝑘 = 2, 3, . . ., and to
any integer 𝑚, 𝜇𝑚 is determined by the following formulas:
𝜇𝑚 (2𝜋) = 0, 𝑚 ∈ N. (20)
𝑐1,1 = 1, 𝑐2,0 = 𝑐0,2 = 0,

𝑖𝑓 (𝛼 = 𝛽 = 0, 𝛽 ≠1) 𝑜𝑟 𝛼 < 0, 𝑜𝑟 𝛽 < 0, Remark 5. In fact, Lemmas 1–4 have given a method to find
(16)
original center condition of (4).
𝑡ℎ𝑒𝑛 𝑐𝛼,𝛽 = 0, What is the isochronous center condition of the origin
of (4) if the origin of (4) is a center? Next we introduce our
or else method to obtain the isochronous𝜑 center condition.
1 We denote that 𝜏(𝜑, ℎ) = ∫0 (𝑑𝑡/𝑑𝜃)𝑑𝜃. From (10), we
𝑐𝛼,𝛽 =
𝛽−𝛼 have
𝛼+𝛽+2
× ∑ [(𝛼 − 𝑘 + 1) 𝑎𝑘,𝑗−1 𝜏 (𝜑, ℎ)
𝑘+𝑗=3
𝜑
𝑑𝑡
− (𝛽 − 𝑗 + 1) 𝑏𝑗,𝑘−1 ] 𝑐𝛼−𝑘+1,𝛽−𝑗+1 , =∫ 𝑑𝜃
0 𝑑𝜃
4 Abstract and Applied Analysis

󸀠 󸀠
𝜑
1 ∞ where 𝑐𝑘+1,𝑘 = 𝑑𝑘+1,𝑘 = 0, 𝑘 = 1, 2, . . ., such that
= ∫ [1 + ∑ ∑ (𝑎𝛼,𝛽−1 + 𝑏𝛽,𝛼−1 )
0 2 𝑚=1 𝛼+𝛽=𝑚+2 ∞
[ 𝑑𝑓
= 𝑓 (𝑧, 𝑤) + ∑𝑝𝑗󸀠 𝑧𝑗+1 𝑤𝑗 ,
−1 𝑑𝑇 𝑗=1
∞ (27)
×𝑒 𝑟̃ (𝜃, ℎ) ] 𝑑𝜃.
𝑖(𝛼−𝛽)𝜃 𝑚 𝑑𝑔
= −𝑔 (𝑧, 𝑤) − ∑𝑞𝑗󸀠 𝑤𝑗+1 𝑧𝑗 ,
] 𝑑𝑇 𝑗=1
(21) 󸀠 󸀠
and when 𝑘 − 𝑗 − 1 ≠0, 𝑐𝑘𝑗 and 𝑑𝑘,𝑗 are determined by the
Definition 6. For a sufficiently small complex constant ℎ, the recursive formulas
origin of system (6) is called a complex center if 𝑟̃(2𝜋, ℎ) ≡ ℎ 1
󸀠
of (13), and the origin is a complex isochronous center if 𝑐𝑘𝑗 =
𝑗+1−𝑘
𝑟̃ (2𝜋, ℎ) ≡ ℎ, 𝜏 (2𝜋, ℎ) ≡ 2𝜋. (22) 𝑘+𝑗+1
× ∑ [(𝑘 − 𝛼 + 1) 𝑎𝛼,𝛽−1
Lemma 7 (see [19]). For system (6), one can derive uniquely 𝛼+𝛽=3

the formal series 󸀠

− (𝑗 − 𝛽 + 1) 𝑏𝛽,𝛼−1 ] 𝑐𝑘−𝛼+1,𝑗−𝛽+1 ,
∞ ∞ (28)
1
𝜉 = 𝑧 + ∑ 𝑐𝑘𝑗 𝑧𝑘 𝑤𝑗 , 𝜂 = 𝑤 + ∑ 𝑑𝑘,𝑗 𝑤𝑘 𝑧𝑗 , (23)
󸀠
𝑑𝑘𝑗 =
𝑘+𝑗=2 𝑘+𝑗=2 𝑗+1−𝑘
𝑘+𝑗+1
where 𝑐𝑘+1,𝑘 = 𝑑𝑘+1,𝑘 = 0, 𝑘 = 1, 2, . . ., 𝑝𝑘 and 𝑞𝑘 are × ∑ [(𝑘 − 𝛼 + 1) 𝑏𝛼,𝛽−1
polynomial in 𝑎𝛼𝛽 , 𝑏𝛼𝛽 with rational coefficients, such that 𝛼+𝛽=3

󸀠
𝑑𝜉 ∞
𝑑𝜂 ∞ − (𝑗 − 𝛽 + 1) 𝑎𝛽,𝛼−1 ] 𝑑𝑘−𝛼+1,𝑗−𝛽+1 ,
= 𝜉 + ∑𝑝𝑗 𝜉𝑗+1 𝜂𝑗 , = −𝜂 − ∑ 𝑞𝑗 𝜂𝑗+1 𝜉𝑗 . (24)
𝑑𝑇 𝑑𝑇
𝑗=1 𝑗=1 and for any positive integer 𝑗, 𝑝𝑗󸀠 , and 𝑞𝑗󸀠 are determined by the
recursive formulas
Let 𝜇0 = 𝜏0 = 0, 𝜇𝑘 = 𝑝𝑘 − 𝑞𝑘 , 𝜏𝑘 = 𝑝𝑘 + 𝑞𝑘 , 𝑘 = 1, 2, . . .,
in which 𝜇𝑘 is 𝑘th singular point value of the origin of system 2𝑗+2

(6). 𝑝𝑗󸀠 = ∑ [(𝑗 − 𝛼 + 2) 𝑎𝛼,𝛽−1

𝛼+𝛽=3
Definition 8. For any positive integer 𝑘, one says that 𝜏(𝑘) = 󸀠
𝑝𝑘 + 𝑞𝑘 is the 𝑘th complex period constant of origin of system − (𝑗 − 𝛽 + 1) 𝑏𝛽,𝛼−1 ] 𝑐𝑗−𝛼+2,𝑗−𝛽+1 ,
(6). (29)
2𝑗+2

Lemma 9 (see [19]). Suppose that the origin of system (6) is a 𝑞𝑗󸀠 = ∑ [(𝑗 − 𝛼 + 2) 𝑏𝛼,𝛽−1
𝛼+𝛽=3
complex center (i.e., all 𝜇𝑚 = 0, 𝑚 = 1, 2, . . .) and there exists
a positive integer 𝑘, such that 𝜏0 = 𝜏1 = ⋅ ⋅ ⋅ = 𝜏𝑘−1 = 0, 𝜏𝑘 ≠0; 󸀠
− (𝑗 − 𝛽 + 1) 𝑎𝛽,𝛼−1 ] 𝑑𝑗−𝛼+2,𝑗−𝛽+1 .
then
󸀠 󸀠 󸀠 󸀠
In (28) and (29), we have taken 𝑐1,0 = 𝑑1,0 = 1, 𝑐0,1 = 𝑑0,1 = 0,
𝜏 (2𝜋, ℎ) = 𝜋 [2 − 𝜏𝑘 ℎ2𝑘 + 𝑜 (ℎ2𝑘 )] . (25) 󸀠 󸀠
and if 𝛼 < 0 or 𝛽 < 0, we take 𝑎𝛼𝛽 = 𝑏𝛼𝛽 = 𝑐𝛼𝛽 = 𝑑𝛼𝛽 = 0.
It is clear that the origin of system (6) is a complex Theorem B (see [19]). Let 𝑝0 = 𝑞0 = 𝑝0󸀠 = 𝑞0󸀠 = 0. If there is a
isochronous center if and only if all 𝜇𝑘 = 𝜏𝑘 = 0, 𝑘 = positive integer 𝑚, such that
1, 2, 3, . . ..
For the problem of the computation of 𝜏𝑘 , [14] gives the 𝑝0 = 𝑞0 = 𝑝1 = 𝑞1 = ⋅ ⋅ ⋅ = 𝑝𝑚−1 = 𝑞𝑚−1 = 0, (30)
following two theorems.
then
Theorem A (see [19]). For system (6), one can derive uniquely
the formal series 𝑝0󸀠 = 𝑞0󸀠 = 𝑝1󸀠 = 𝑞1󸀠 = ⋅ ⋅ ⋅ = 𝑝𝑚−1
󸀠 󸀠
= 𝑞𝑚−1 = 0,
(31)
󸀠 󸀠
∞ 𝑝𝑚 = 𝑝𝑚 , 𝑞𝑚 = 𝑞𝑚 ,
󸀠 𝑘 𝑗
𝑓 (𝑧, 𝑤) = 𝑧 + ∑ 𝑐𝑘𝑗 𝑧 𝑤,
𝑘+𝑗=2
and vice versa.
∞ (26)
󸀠
𝑔 (𝑧, 𝑤) = 𝑤 + ∑ 𝑑𝑘,𝑗 𝑤𝑘 𝑧𝑗 , Actually, Lemma 7 and the above two theorems give an
𝑘+𝑗=2 algorithm to compute 𝜏𝑚 . For any positive integer 𝑚, in
Abstract and Applied Analysis 5

order to compute 𝜏𝑚 , we only need to carry out the addition, Obviously point (−𝑥, −𝑦) satisfies (36) if (𝑥, 𝑦) satisfies
subtraction, multiplication, and division to the coefficients (36); then system (36) is a class of 𝑧2 -equivariant cubic
of system (6). The algorithm is recursive. It avoids some systems about the origin. Hence system (34) is a class
complicated integrating operations and solving equations. In of 𝑧2 -equivariant cubic systems about point (−1, 0). Proof
addition, it can be easily realized by computer algebra systems ends.
such as Mathematica.
Notice that the complex period constants are polynomials After making transformations (32) and (33), system (2)
of the coefficients of system (6). According to the Hilbert basis becomes a symmetric system about point (−1, 0) (i.e., system
theorem, there exists 𝑚 ∈ N such that all 𝜏𝑘 = 0 (𝑘 = 1, 2, . . .) (34)), so here we call system (2) a class of quasi symmetric
if and only if 𝜏1 = 𝜏2 = ⋅ ⋅ ⋅ = 𝜏𝑚 = 0. We say that the systems. In fact, through investigating the center condition
set {𝜏1 , 𝜏2 , . . . , 𝜏𝑚 } is a period constant basis of system (6). To of the origin of system (34), those of the infinity and the
determine isochronous center of a system, the key idea is to elementary focus point (−1/2, 0) of (2) can been forecasted.
find a period constant basis. In order to investigate the centers problem of (2), we may
as well study system (34) or system (36).
Remark 10. Lemma 7 and Theorems A and B offer a method Making the transformation
to find a necessary condition of isochronicity.
𝑧 = 𝑢 + 𝑖V, 𝑤 = 𝑢 − 𝑖V, 𝑇 = 𝑖𝜏, 𝑖 = √−1,
3. The Reduction and Bicenter Condition of (37)
System (2) system (34)|𝛿 = 0 becomes
After introducing the method to calculate the focal values and
𝑑𝑧
period constants of system (4), we try to make some appropri- = 𝑧 + 𝑍2 (𝑧, 𝑤) + 𝑍3 (𝑧, 𝑤) ,
ate transformations so as to carry out our investigation about 𝑑𝑇
(38)
system (2). 𝑑𝑤
By means of Bendixson homeomorphous transformation = −𝑤 − 𝑊2 (𝑧, 𝑤) − 𝑊3 (𝑧, 𝑤) ,
𝑑𝑇
𝑥 𝑦 in which
𝑢= , V= , (32)
𝑥 2 + 𝑦2 𝑥 2 + 𝑦2
1 1
and time transformation 𝑍2 (𝑧, 𝑤) = (3 − 4𝑎1 − 4𝑖𝑎4 − 2𝑎5 ) 𝑧2 + (3 + 2𝑎5 ) 𝑧𝑤
8 4
3 1
𝑑𝑡 = (𝑥2 + 𝑦2 ) 𝑑𝜏, (33) + (3 + 4𝑎1 + 4𝑖𝑎4 − 2𝑎5 ) 𝑤2 ,
8
system (2) can be transformed into the following real system: 1
𝑍3 (𝑧, 𝑤) = (1 − 2𝑎1 + 2𝑎3 − 2𝑖𝑎4 − 2𝑎5 + 2𝑖𝑎6 ) 𝑧3
𝑑𝑢 3𝛿 𝛿 16
= 𝛿𝑢 − V + 𝑢2 + 2𝑎1 𝑢V + 𝑢3 + 𝑎1 𝑢2 V + 𝑎3 V3 ,
𝑑𝜏 2 2 1
+ (3 − 2𝑎1 − 6𝑎3 − 2𝑖𝑎4 + 2𝑎5 − 6𝑖𝑎6 ) 𝑧2 𝑤
𝑑V 3 1 16
= 𝑢 + 𝛿V + 𝑢2 + 2𝑎4 𝑢V + 𝑎5 V2 + 𝑢3 (34)
𝑑𝜏 2 2 1
+ (3 + 2𝑎1 + 6𝑎3 + 2𝑖𝑎4 + 2𝑎5 + 6𝑖𝑎6 ) 𝑧𝑤2
16
+ 𝑎4 𝑢2 V + 𝑎5 𝑢V2 + 𝑎6 V3 .
1
+ (1 + 2𝑎1 − 2𝑎3 + 2𝑖𝑎4 − 2𝑎5 − 2𝑖𝑎6 ) 𝑤3 ,
After making the above two transformations, the infinity 16
and the elementary focus point (−1/2, 0) of (2), respectively,
1 1
become the origin and (−2, 0) of system (34). For system (34), 𝑊2 (𝑧, 𝑤) = (3 − 4𝑎1 + 4𝑖𝑎4 + 2𝑎5 ) 𝑤2 + (3 + 2𝑎5 ) 𝑧𝑤
we have the following theorem. 8 4
1
Theorem 11. System (34) is a class of 𝑧2 -equivariant cubic + (3 + 4𝑎1 − 4𝑖𝑎4 − 2𝑎5 ) 𝑤2 ,
8
systems about point (−1, 0).
1
𝑊3 (𝑧, 𝑤) = (1 + 2𝑎1 − 2𝑎3 − 2𝑖𝑎4 − 2𝑎5 + 2𝑖𝑎6 ) 𝑧3
Proof. By means of translation transformation 16
𝑢 = 𝑥 − 1, V = 𝑦, 1
(35) + (3 + 2𝑎1 + 6𝑎3 − 2𝑖𝑎4 + 2𝑎5 − 6𝑖𝑎6 ) 𝑧2 𝑤
16
system (34) turns into the following system: 1
+ (3 − 2𝑎1 − 6𝑎3 + 2𝑖𝑎4 + 2𝑎5 + 6𝑖𝑎6 ) 𝑧𝑤2
𝑑𝑥 𝛿 𝛿 16
= − 𝑥 − (𝑎1 + 1) 𝑦 + 𝑥3 + 𝑎1 𝑥2 𝑦 + 𝑎3 𝑦3 ,
𝑑𝜏 2 2 1
(36) + (1 − 2𝑎1 + 2𝑎3 + 2𝑖𝑎4 − 2𝑎5 − 2𝑖𝑎6 ) 𝑤3 .
𝑑𝑦 1 16
+ (𝛿 − 𝑎4 ) 𝑦 + 𝑥3 + 𝑎4 𝑥2 𝑦 + 𝑎5 𝑥𝑦2 + 𝑎6 𝑦3 . (39)
𝑑𝜏 2
6 Abstract and Applied Analysis

System (38) is called the complex concomitant system if 𝑎3 𝑎4 = 0, from Lemma 1, we obtain that 𝜇𝑘 = 0, 𝑘 =
of system (34). Clearly system (38) belongs to the class of 2, 3, 4, 5. Hence let 𝑎1 = 1/4; then
system (6), so we can use the formulas of Lemma 1 and 𝑖
the conclusion of Lemma 4 to compute and simplify the 𝜇3 = 𝑎 𝑎 (84𝑎3 − 16𝑎42 − 45) . (47)
singular points values by using computational software such 384 3 4
as Mathematica; we obtain the following theorem. Moreover let 𝑎3 = (1/84) (45 + 16𝑎42 ); then

Theorem 12. The first 5 singular points’ values of the origin for 𝑖 2
𝜇3 = 0, 𝜇4 = 𝑎4 (3 + 2𝑎42 ) (45 + 16𝑎42 ) ;
(38) are as follows: 169344
(48)
𝑖
𝜇1 = (2𝑎4 + 2𝑎4 𝑎5 − 3𝑎6 ) ; while if 𝑎4 ≠0, then 𝜇4 ≠0; at this time only four singular point
4 values can be obtained.
𝑖 (2) If 𝑎5 ≠− 1, letting
𝜇2 = 𝑎4 [(1 + 𝑎5 ) (18 + 18𝑎1 + 8𝑎42 ) (40)
36
3 (4𝑎3 𝑎5 − 6𝑎1 𝑎5 − 6𝑎5 + 4𝑎1 𝑎3 + 3𝑎3 − 6𝑎1 − 6)
−𝑎3 (9 + 12𝑎1 + 12𝑎5 ) ] ; 𝑎42 = ,
(8 + 8𝑎5 )
(49)
(1) if 𝑎5 = −1, then
then
𝑖 𝑖𝑎3 𝑎4
𝜇2 = − 𝑎3 𝑎4 (4𝑎1 − 1) ; 𝜇2 = 0; 𝜇3 = (4𝑎1 − 10𝑎5 − 11) ℎ0 . (50)
12 96 (1 + 𝑎5 )
𝑖
𝜇3 = 𝑎 𝑎 (84𝑎3 − 16𝑎42 − 45) ; (41) Because 𝑎3 𝑎4 ℎ0 = 0 will induce 𝜇3 = 𝜇4 = 𝜇5 = 0, let 𝑎5 =
384 3 4
(1/10) (4𝑎1 − 11); at this time
𝑖 2
𝜇4 = 𝑎4 (3 + 2𝑎42 ) (45 + 16𝑎42 ) ; 7𝑖𝑎3 𝑎4
169344 𝜇4 = − (1 + 𝑎1 − 5𝑎3 )
1500 (51)
(2) if 𝑎5 ≠− 1, then
× (29𝑎12 − 𝑎1 − 2 + 28𝑎13 + 45𝑎3 − 105𝑎1 𝑎3 ) ,
𝑖𝑎3 𝑎4
𝜇3 = (4𝑎1 − 10𝑎5 − 11) ℎ0 ; while 𝑎3 𝑎4 (29𝑎12 − 𝑎1 − 2 + 28𝑎13 + 45𝑎3 − 105𝑎1 𝑎3 ) = 0 will
96 (1 + 𝑎5 )
induce 𝜇4 = 𝜇5 = 0, let 𝑎1 = 5𝑎3 − 1; then
7𝑖𝑎3 𝑎4
𝜇4 = − (1 + 𝑎1 − 5𝑎3 ) 7𝑖 4
1500 (42) 𝜇5 = 𝑎 𝑎 (11 − 76𝑎3 + 140𝑎32 ) . (52)
× (29𝑎12 − 𝑎1 − 2 + 28𝑎13 + 45𝑎3 − 105𝑎1 𝑎3 ) ; 36 3 4
Proof ends.
7𝑖 4
𝜇5 = 𝑎 𝑎 (11 − 76𝑎3 + 140𝑎32 )
36 3 4
Remark. The equation 11−76𝑎3 +140𝑎32 = 0 has not real roots,
in which so only five singular point values exist.
From Theorem 12 and Lemma 4, we have the following
ℎ0 = − 6 − 10𝑎1 − 4𝑎12 + 3𝑎3 + 6𝑎1 𝑎3 − 10𝑎5 theorem.
(43)
− 14𝑎1 𝑎5 − 4𝑎12 𝑎5 + 6𝑎3 𝑎5 − 4𝑎52 − 4𝑎1 𝑎52 .
Theorem 13. The first 5 focal values of the origin of (21) (or
In the above expressions of 𝜇𝑘 , one lets 𝜇𝑖 = 0, 𝑖 = 1, 2, . . ., 𝑘−1. the first 5 general focal values of the infinity and the elementary
focus (−1/2, 0) of (4)) are as follows:
Proof. According to Lemma 1, we have 𝜋
V3 = − (2𝑎4 + 2𝑎4 𝑎5 − 3𝑎6 ) ;
𝑖 4
𝜇1 = (2𝑎4 + 2𝑎4 𝑎5 − 3𝑎6 ) . (44)
4 𝜋
V5 = − 𝑎4 [(1 + 𝑎5 ) (18 + 18𝑎1 + 8𝑎42 ) (53)
Let 𝑎6 = (2/3)𝑎4 (1 + 𝑎5 ); then 36
−𝑎3 (9 + 12𝑎1 + 12𝑎5 ) ] .
𝜇1 = 0;
𝑖 (1) If 𝑎5 = −1, then
𝜇2 = 𝑎 [(1 + 𝑎5 ) (18 + 18𝑎1 + 8𝑎42 ) (45)
36 4 𝜋
V5 = 𝑎 𝑎 (4𝑎1 − 1) ;
−𝑎3 (9 + 12𝑎1 + 12𝑎5 ) ] . 12 3 4
𝜋
(1) If 𝑎5 = −1, then V7 = − 𝑎 𝑎 (84𝑎3 − 16𝑎42 − 45) . (54)
384 3 4
𝑖 𝜋 2
𝜇2 = − 𝑎 𝑎 (4𝑎1 − 1) ; (46) V9 = − 𝑎4 (3 + 2𝑎42 ) (45 + 16𝑎42 ) .
12 3 4 169344
Abstract and Applied Analysis 7

(2) If 𝑎5 ≠− 1, then then system (57) has an integral factor 𝑀1 (𝑥, 𝑦) = 𝑓1−1 and a
(𝑎 +𝑎 )
first integral 𝐹1 (𝑥, 𝑦) = 𝑓1 𝑓2 1 5 , in which
𝜋𝑎3 𝑎4
V7 = − (4𝑎1 − 10𝑎5 − 11) ℎ0 ,
96 (1 + 𝑎5 ) 2
𝑓1 = 𝑔12 − [2𝑎3 + (𝑎1 − 𝑎5 ) ] 𝑔22 , (59)
7𝜋𝑎3 𝑎4
V9 = (1 + 𝑎1 − 5𝑎3 ) 1/√2𝑎3 +(𝑎1 −𝑎5 )2
1500 {
{
(55) {
{ 𝑔 + √2𝑎3 + (𝑎1 − 𝑎5 )2 𝑔2
{
{ 1
× (29𝑎12 − 𝑎1 − 2 + 28𝑎13 + 45𝑎3 − 105𝑎1 𝑎3 ) , {
{ ( ) ,
{
{ √ 2
{
{ 𝑔 1 − 2𝑎3 + (𝑎 1 − 𝑎5 ) 𝑔2
7𝜋 4 {
{
{
{ 2
V11 = − 𝑎 𝑎 (11 − 76𝑎3 + 140𝑎32 ) . {
{
if 2𝑎3 + (𝑎1 − 𝑎5 ) > 0;
36 3 4 {
{
{
{ 2 (−1 + 𝑥2 + 2𝑎5 𝑦2 )
𝑓2 = { exp ,
in which {
{ 2 + (𝑎1 + 𝑎5 ) (1 − 𝑥2 + 𝑎1 𝑦2 − 𝑎5 𝑦2 )
{
{ 2
{
{ if 2𝑎3 + (𝑎1 − 𝑎5 ) = 0;
ℎ0 = − 6 − 10𝑎1 − 4𝑎12 + 3𝑎3 + 6𝑎1 𝑎3 − 10𝑎5 {
{
{
{
(56) {
{ 2 𝑔
− 14𝑎1 𝑎5 − 4𝑎12 𝑎5 + 6𝑎3 𝑎5 − 4𝑎52 − 4𝑎1 𝑎52 . {
{ arctan 2 ) ,
{exp (
{ 𝑔1
{
{ √−2𝑎3 − (𝑎1 − 𝑎5 )2
{
{
In the above expressions of V2𝑘+1 , we have let V2𝑖−1 = 0, 𝑖 = { 2
{ if 2𝑎3 + (𝑎1 − 𝑎5 ) < 0.
2, . . . , 𝑘.
(60)
According to Theorem 13, it is easy to obtain the following
Considering that system (34) can turn into system (36) by
theorem.
making transformations, hence the origin of (34) is a center
Theorem 14. The first 5 focal values of the origin of (34) (or when (𝐻1 ) holds; moreover, the infinity and the elementary
the first 5 focal values of the infinity and the elementary focus focus (−1/2, 0) of (2) are two centers.
(−1/2, 0) of (2)) vanish if and only if one of the following (2.2) If (𝐻2 ) holds, then (36)|𝛿=0 becomes
conditions holds:
𝑑𝑥 4 2 1
= 𝑎4 𝑦 − (9 + 4𝑎42 ) 𝑥2 𝑦,
(𝐻1 ) 𝑎4 = 𝑎6 = 0; 𝑑𝜏 9 9
(𝐻2 ) 𝑎42 = −(9/4)(1 + 𝑎1 ) > 0, 𝑎3 = 0, 𝑎6 = (2/3)𝑎4 (1 + 𝑎5 ), 𝑑𝑦 1 1
= − 𝑥 − 𝑎4 𝑦 + 𝑥3 + 𝑎4 𝑥2 𝑦 + 𝑎5 𝑥𝑦2 (61)
𝑎5 ≠− 1; 𝑑𝜏 2 2
(𝐻3 ) 𝑎6 = (2/3)𝑎4 (1 + 𝑎5 ), 𝑎5 ≠− 1, 𝑎3 ≠0, 𝑎42 = 3 (4𝑎3 𝑎5 − 2
6𝑎1 𝑎5 − 6𝑎5 + 4𝑎1 𝑎3 + 3𝑎3 − 6𝑎1 − 6)/(8 + 8𝑎5 ) > 0, + 𝑎4 (1 + 𝑎5 ) 𝑦3 ,
3
ℎ0 = 0.
while system (61) has an integral factor
Theorem 15. The origin of (34) is a center (the infinity and the
elementary focus (−1/2, 0) of (2) are two centers), if and only (9−8𝑎42 +18𝑎5 )/2
𝑀2 (𝑥, 𝑦) = 𝑓3−3 𝑓4 (62)
if one of the conditions (𝐻1 ), (𝐻2 ), and (𝐻3 ) holds.
and a first integral
Proof. (1) According to Theorem 14, if the origin of (34) is
a center (the infinity and the elementary focus (−1/2, 0) of 9(3+2𝑎5 )/2
𝐹2 = (3𝑥 + 4𝑎4 𝑦) 𝑓3−2 𝑓4
(2) are two general centers), then one of the conditions (𝐻1 ),
(𝐻2 ), and (𝐻3 ) holds. So necessary condition is correct. 9(3+2𝑎 )/2 (63)
𝑓4 5
𝑑𝑥
(2) Next we prove sufficient condition. −6 (1 + 𝑎5 ) ∫ ,
(2.1) If (𝐻1 ) holds, then (36)|𝛿=0 becomes 9 + (9 + 4𝑎42 ) (−1 + 𝑥2 )

𝑑𝑥 in which
= − (𝑎1 + 1) 𝑦 + 𝑎1 𝑥2 𝑦 + 𝑎3 𝑦3 .
𝑑𝜏 𝑓3 = 3𝑥 + 2𝑎4 𝑦,
(57)
𝑑𝑦 1 1
= − 𝑥 + 𝑥3 + 𝑎5 𝑥𝑦2 2
{[9 + (9 + 4𝑎2 ) (−1 + 𝑥2 )]1/(9+4𝑎4 ) , if 9 + 4𝑎2 ≠0;
𝑑𝜏 2 2
𝑓4 = { 4 4
(−1+𝑥2 )/9 2
Letting { 𝑒 , if 9 + 4𝑎4 = 0.
(64)
𝑔1 (𝑥, 𝑦) = 2 + 𝑎1 + 𝑎5 − 𝑎1 𝑥2 − 𝑎5 𝑥2 − 2𝑎3 𝑦2
Considering that system (34) can turn into system (36) by
+ 2𝑎1 𝑎5 𝑦2 − 2𝑎52 𝑦2 , (58) making transformations, hence the origin of (34) is a center
when (𝐻2 ) holds; moreover, the infinity and the elementary
𝑔2 (𝑥, 𝑦) = −1 + 𝑥2 + 2𝑎5 𝑦2 , focus (−1/2, 0) of (2) are two centers.
8 Abstract and Applied Analysis

(2.3) If (𝐻3 ) holds, then (34)|𝛿=0 has an integral fac- 1 1

𝑊2 (𝑧, 𝑤) = (3 − 4𝑎1 + 2𝑎5 ) 𝑤2 + (3 + 2𝑎5 ) 𝑧𝑤
2(𝑎 +𝑎 )
tor 𝑀3 (𝑥, 𝑦) = 𝑓5 1 5 and a first integral 𝐹3 (𝑥, 𝑦) = 8 4
(1+2𝑎1 +2𝑎5 ) 1
𝑓5 𝑓6 , in which + (3 + 4𝑎1 − 2𝑎5 ) 𝑤2 ,
8
𝑓5 = (𝑎1 + 𝑎5 ) 𝑥 − 𝑎4 𝑦, 1
𝑊3 (𝑧, 𝑤) = (1 + 2𝑎1 − 2𝑎3 − 2𝑎5 ) 𝑧3
𝑓6 = (2 + 𝑎1 + 𝑎5 ) [𝑎4 𝑥 + 2 (1 + 𝑎1 ) (𝑎1 + 𝑎5 ) 𝑦] 16
1
− (1 + 𝑎1 + 𝑎5 ) (65) + (3 + 2𝑎1 + 6𝑎3 + 2𝑎5 ) 𝑧2 𝑤
16
× [𝑎4 𝑥3 + 2 (1 + 𝑎1 ) (𝑎1 + 𝑎5 ) 𝑥2 𝑦 1
+ (3 − 2𝑎1 − 6𝑎3 + 2𝑎5 ) 𝑧𝑤2
16
+ 2𝑎4 (1 + 𝑎5 ) 𝑥𝑦2 + 2𝑎3 (𝑎1 + 𝑎5 ) 𝑦3 ] .
1
+ (1 − 2𝑎1 + 2𝑎3 − 2𝑎5 ) 𝑤3 .
Considering that system (34) can turn into system (36) by 16
making transformations, hence the origin of (34) is a center (68)
when (𝐻3 ) holds; moreover, the infinity and the elementary
According to formulas (28) and (29) of Theorem A and
focus (−1/2, 0) of (2) are two centers. Proof ends.
𝜏𝑘 = 𝑝𝑘 + 𝑞𝑘 = 𝑝𝑘󸀠 + 𝑞𝑘󸀠 , we can compute periodic constants of
systems (66) and (67).
4. Isochronous Center Condition of (−1/2, 0) of
System (2) Theorem 16. The first four period constants of the origin of
system (66) or (67) are as follows:
After obtaining the conditions that the infinity and the
elementary focus (−1/2, 0) are two centers, next we continue 1
𝜏1 = (−18 − 4𝑎12 − 9𝑎3 − 12𝑎5 + 10𝑎1 𝑎5 − 4𝑎52 ) ,
to investigate isochronicity of the elementary focus (−1/2, 0) 12
1
according to three different cases (i.e., (𝐻1 ), (𝐻2 ), and (𝐻3 )) 𝜏2 = (−108 − 45𝑎1 − 12𝑎12 − 117𝑎5 − 24𝑎1 𝑎5
in order. 18
−2𝑎12 𝑎5 − 36𝑎52 − 4𝑎1 𝑎52 − 2𝑎52 ) ,
4.1. The Isochronicity of Case (𝐻1 ). If condition (𝐻1 ) holds, 1
then system (34)|𝛿=0 becomes 𝜏3 = ℎ,
1080 1
𝑑𝑢 1 2
= −V + 2𝑎1 𝑢V + 𝑎1 𝑢2 V + 𝑎3 V3 , 𝜏4 = − (9 + 𝑎5 ) (3 + 2𝑎5 ) ℎ2 .
𝑑𝜏 (66) 1782606477375000000
𝑑V 3 1 (69)
= 𝑢 + 𝑢2 + 𝑎5 V2 + 𝑢3 + 𝑎5 𝑢V2 .
𝑑𝜏 2 2
in which
System (66) complex concomitant system is as follows:
ℎ1 = (9720 + 4050𝑎1 + 27𝑎12 − 300𝑎13 − 100𝑎14
𝑑𝑧
= 𝑧 + 𝑍2 (𝑧, 𝑤) + 𝑍3 (𝑧, 𝑤) , + 8802𝑎5 + 1134𝑎1 𝑎5 − 1416𝑎12 𝑎5
𝑑𝑇 (67)
𝑑𝑤
= −𝑤 − 𝑊2 (𝑧, 𝑤) − 𝑊3 (𝑧, 𝑤) ,
𝑑𝑇 −200𝑎13 𝑎5 + 2115𝑎52 + 468𝑎1 𝑎52 − 100𝑎12 𝑎52 ) ,

in which ℎ2 = 441227872054417689 + 969414967442849922𝑎5

1 1 + 729276873221792484𝑎52 + 232198402133252079𝑎53
𝑍2 (𝑧, 𝑤) = (3 − 4𝑎1 − 2𝑎5 ) 𝑧2 + (3 + 2𝑎5 ) 𝑧𝑤
8 4
1 + 40172927013408528𝑎54 + 3428874437224704𝑎55
+ (3 + 4𝑎1 − 2𝑎5 ) 𝑤2 ,
8
+ 49018338325504𝑎56 ,
1 (70)
𝑍3 (𝑧, 𝑤) = (1 − 2𝑎1 + 2𝑎3 − 2𝑎5 ) 𝑧3
16
In the above expression of 𝜏𝑚 , we have let 𝜏1 = ⋅ ⋅ ⋅ 𝜏𝑚−1 = 0,
1
+ (3 − 2𝑎1 − 6𝑎3 + 2𝑎5 ) 𝑧2 𝑤 𝑚 = 2, 3, 4.
16
1 Theorem 17. The first four period constants of the origin of
+ (3 + 2𝑎1 + 6𝑎3 + 2𝑎5 ) 𝑧𝑤2 system (66) or (67) are zero, if and only if one of the following
16
two conditions holds:
1
+ (1 + 2𝑎1 − 2𝑎3 − 2𝑎5 ) 𝑤3 , (𝐶1 ) 𝑎1 = 𝑎5 = −3/2, 𝑎3 = 1/2;
16
Abstract and Applied Analysis 9

(𝐶2 ) 𝑎1 = −3, 𝑎3 = 0, 𝑎5 = −9. which satisfy 𝑑𝑢1 /𝑑𝜏 = 𝑢1 , 𝑑V1 /𝑑𝜏 = −V1 , so the origin of
system (74) is an isochronous center.
Proof. By computing the resultant of 𝜏2 and 𝜏3 about variable (2) If (𝐶2 ) holds, then system (66) becomes
𝑎1 , we have
𝑑𝑢
1 2 = −V − 6𝑢V − 3𝑢2 V,
Resultant [𝜏3 , 𝜏2 , 𝑎1 ] = (9 + 𝑎5 ) (3 + 2𝑎5 ) 𝑅, 𝑑𝜏 (76)
212576400 𝑑V 3 1
(71) = 𝑢 + 𝑢2 − 9V2 + 𝑢3 − 9𝑢V2 .
𝑑𝜏 2 2
in which For (76), make the following analytic polar transformations:
𝑅 = 4214349 + 14258511𝑎5 + 18779526𝑎52 + 12158343𝑎53 𝑢 (𝑢 − 1) (𝑢 − 2)
𝑢2 = ,
+ 4086747𝑎54 + 727632𝑎55 + 69888𝑎56 + 1024𝑎57 . √2(3𝑢 − (3/2)𝑢2 − 1/2)3/2
(72) (77)
√2V
V2 = 3/2
,
Also the resultant of 𝑅 and ℎ2 about 𝑎5 is as follows: (3𝑢 − (3/2)𝑢2 − 1/2)

Resultant [𝑅, ℎ2 , 𝑎5 ] = −28488259 ⋅ ⋅ ⋅ ≠0, (73) which satisfy 𝑑𝑢2 /𝑑𝜏 = 𝑢2 , 𝑑V2 /𝑑𝜏 = −V2 , so the origin of
system (76) is an isochronous center.
which shows that 𝑅 = 0 cannot deduce 𝜏2 = 𝜏3 = 𝜏4 = 0.
From (71), let 𝜏2 = 𝜏3 = 𝜏4 = 0; then
Considering that system (34) is a class of 𝑧2 -equivariant
Resultant[𝜏3 , 𝜏2 , 𝑎1 ] = 0; hence 𝑎5 = −3/2 or 𝑎5 = −9.
cubic systems about point (−1, 0), we have the following
If 𝑎5 = −3/2, then 𝜏4 = 0; continue to let 𝜏3 = 0 (i.e.,
Theorem.
ℎ1 = 0); then 𝑎1 = −3/2; next let 𝜏1 = 0; then 𝑎3 = 1/2.
Hence condition (𝐶1 ) holds. At the same time, if condition Theorem 19. If one of the conditions (𝐶1 ) and (𝐶2 ) holds under
(𝐶1 ) holds, it is easy to obtain that 𝜏1 = 𝜏2 = 𝜏3 = 𝜏4 = 0. condition (𝐻1 ), then the origin and point (−2, 0) of system (34)
If 𝑎5 = −9, then 𝜏4 = 0; continue to let 𝜏3 = 0 (i.e., ℎ1 = 0); are two isochronous centers.
then 𝑎1 = −3; next let 𝜏1 = 0; then 𝑎3 = 0. Hence condition
(𝐶2 ) holds. At the same time, if condition (𝐶2 ) holds, it is easy Considering that system (2) can be changed into system
to obtain that 𝜏1 = 𝜏2 = 𝜏3 = 𝜏4 = 0. (34) under the transformations (32) and (33), ulteriorly, we
have the following Theorem.
Clearly, condition (𝐶1 ) or (𝐶2 ) is necessary for the
isochronicity of the origin of system (66) or (67). Moreover, Theorem 20. If one of the conditions (𝐶1 ) and (𝐶2 ) holds
we will prove that condition (𝐶1 ) or (𝐶2 ) is sufficient under condition (𝐻1 ), then the singular point (−1/2, 0) of
condition for the isochronicity of the origin of system (66) system (2) is an isochronous center.
or (67).

Theorem 18. The origin of system (66) or (67) is an 4.2. The Isochronicity of Case (𝐻2 ). If condition (𝐻2 ) holds,
isochronous center if and only if one of (𝐶1 ) and (𝐶2 ) holds. then system (34)|𝛿=0 becomes

𝑑𝑢 1
Proof. If the origin of system (66) or (67) is an isochronous = − V (9 + 18𝑢 + 8𝑎42 𝑢 + 9𝑢2 + 4𝑎42 𝑢2 ) ,
center, according to Theorem 17, then one of (𝐶1 ) and (𝐶2 ) 𝑑𝜏 9
holds. Hence, the necessary condition is proved. 𝑑V 1
Next we will prove the sufficient condition. = (6𝑢 + 9𝑢2 + 3𝑢3 + 12𝑎4 𝑢V + 6𝑎4 𝑢2 V (78)
(1) If (𝐶1 ) holds, then system (66) becomes 𝑑𝜏 6
+6𝑎5 V2 + 6𝑎5 𝑢V2 + 4𝑎4 V3 + 4𝑎4 𝑎5 V3 ) .
𝑑𝑢 1
= − V (2 + 6𝑢 + 3𝑢2 − V2 ) ,
𝑑𝜏 2 System (78) complex concomitant system belongs to the
(74)
𝑑V 1 form of (38). Hence we can use formulas (28) and (29) of
= (1 + 𝑢) (2𝑢 + 𝑢2 − 3V2 ) . Theorem A and 𝜏𝑘 = 𝑝𝑘 + 𝑞𝑘 = 𝑝𝑘󸀠 + 𝑞𝑘󸀠 to compute periodic
𝑑𝜏 2
constants of system (78).
For (74), make the following analytic polar transformations:
Theorem 21. The first three period constants of the origin of
2𝑢 + 5𝑢2 + 4𝑢3 + 𝑢4 + 3V2 + 4𝑢V2 + 2𝑢2 V2 + V4 system (78) are as follows:
𝑢1 = 2
,
2(1 + 2𝑢 + 𝑢2 + V2 ) 1
(75) 𝜏1 = (−891 − 306𝑎42 − 32𝑎44 − 891𝑎5
(1 + 𝑢) V 486
V1 = 2
,
(1 + 2𝑢 + 𝑢2 + V2 ) −180𝑎42 𝑎5 − 162𝑎52 ) ,
10 Abstract and Applied Analysis

1 𝑙2 = 336 − 234𝑏42 + 63𝑏44 − 6𝑏46 + 1650𝑟 − 1284𝑏42 𝑟

𝜏2 = (9 + 4𝑎42 )
2916
+ 486𝑏44 𝑟 − 90𝑏46 𝑟 + 6𝑏48 𝑟 + 1565𝑟2 − 2370𝑏42 𝑟2
× (729 + 234𝑎42 + 32𝑎44 + 567𝑎5 + 108𝑎42 𝑎5 ) ,
+ 1124𝑏44 𝑟2 − 234𝑏46 𝑟2 + 19𝑏48 𝑟2 + 516𝑟3
1
𝜏3 = − (9 + 𝑎42 ) (9 + 4𝑎42 ) (999 + 165𝑎42 + 16𝑎44 ) .
3346110 − 1224𝑏42 𝑟3 + 1184𝑏44 𝑟3 − 224𝑏46 𝑟3 + 4𝑏48 𝑟3
(79)
+ 52𝑟4 − 168𝑏42 𝑟4 + 320𝑏44 𝑟4 − 136𝑏46 𝑟4 − 20𝑏48 𝑟4 ;
In the above expression of 𝜏𝑚 , we have let 𝜏1 = ⋅ ⋅ ⋅ 𝜏𝑚−1 = 0,
𝑚 = 2, 3. 𝑙3 = − 12414600 + 10901385𝑏42 − 4216185𝑏44

From Theorem 21, 𝜏3 = 0 has not real number roots. + 963495𝑏46 − 131355𝑏48 + 8100𝑏410
Hence under condition (𝐻2 ), the origin of system (78) cannot
become an isochronous center. Accordingly, the singular − 43826805𝑟 + 68754285𝑏42 𝑟 − 37919070𝑏44 𝑟
point (−1/2, 0) of system (2) cannot become an isochronous
center under condition (𝐻2 ). + 11032200𝑏46 𝑟 − 1939005𝑏48 𝑟 + 195615𝑏410 𝑟

− 8100𝑏412 𝑟 − 80932512𝑟2 + 169994130𝑏42 𝑟2

4.3. The Isochronicity of Case (𝐻3 ). If condition (𝐻3 ) holds,
then system (34)|𝛿=0 becomes − 121184220𝑏44 𝑟2 + 41520240𝑏46 𝑟2 − 7909800𝑏48 𝑟2
𝑑𝑢
= −V + 2𝑎1 𝑢V + 𝑎1 𝑢2 V + 𝑎3 V3 , + 903798𝑏410 𝑟2 − 53460𝑏412 𝑟2 − 78594084𝑟3
𝑑𝜏
+ 200907300𝑏42 𝑟3 − 192468840𝑏44 𝑟3 + 78954480𝑏46 𝑟3
𝑑V 3 1 (80)
= 𝑢 + 𝑢2 + 2𝑎4 𝑢V + 𝑎5 V2 + 𝑢3 + 𝑎4 𝑢2 V
𝑑𝜏 2 2 − 15231180𝑏48 𝑟3 + 1650156𝑏410 𝑟3 − 138600𝑏412 𝑟3
+ 𝑎5 𝑢V2 + 𝑎6 V3 ,
− 38815064𝑟4 + 119144360𝑏42 𝑟4 − 150195520𝑏44 𝑟4
in which the coefficients satisfy condition (𝐻3 ). + 82363760𝑏46 𝑟4 − 17141720𝑏48 𝑟4 + 1412136𝑏410 𝑟4
For convenience, we let 𝑎4 = 𝑏4 𝑟; then condition (𝐻3 )
becomes − 196560𝑏412 𝑟4 − 9033312𝑟5 + 33223520𝑏42 𝑟5
1 2 − 55560320𝑏44 𝑟5 + 40281280𝑏46 𝑟5 − 11183680𝑏48 𝑟5
(𝑏 + 2𝑏42 𝑟 − 2) ,
𝑎1 =
2 4
1
𝑎3 = − 𝑏4 (3𝑏43 − 12𝑏4 − 14𝑏4 𝑟 + 8𝑏43 𝑟 − 4𝑏4 𝑟2 + 4𝑏43 𝑟2 ) , + 674208𝑏410 𝑟5 − 185760𝑏412 𝑟5 − 754208𝑟6
6
1
𝑎5 = (2 − 𝑏42 + 2𝑟 − 2𝑏42 𝑟) , + 3300480𝑏42 𝑟6 − 7720960𝑏44 𝑟6 + 7271040𝑏46 𝑟6
2
1
𝑎6 = (4𝑏4 𝑟 − 𝑏43 𝑟 + 2𝑏4 𝑟2 − 2𝑏43 𝑟2 ) . − 2734560𝑏48 𝑟6 + 181632𝑏410 𝑟6 − 97920𝑏412 𝑟6 .
3
(81) (83)

System (80) complex concomitant system belongs to the Through analyzing the expressions of 𝜏𝑖 , 𝑖 = 1, 2, 3 of
form of (38). Hence we can use formulas (28) and (29) of Theorem 22, we have the following theorem.
Theorem A and 𝜏𝑘 = 𝑝𝑘 + 𝑞𝑘 = 𝑝𝑘󸀠 + 𝑞𝑘󸀠 to compute periodic
constants of system (80). Theorem 23. Note that 𝜏1 = 𝜏2 = 𝜏3 = 0 of Theorem 22 have
not solutions.
Theorem 22. If condition (81) holds (i.e., (𝐻3 ) holds), then the
first three period constants of the origin of system (80) are as Proof. By computing carefully in personal computer, we
follows: obtain

1 1 1
𝜏1 = 𝑙1 , 𝜏2 = 𝑙, 𝜏3 = 𝑙, (82) 𝑅1 = Resultant [𝑙1 , 𝑙2 , 𝑟]
6 96 2 414720 3
2
in which 𝑙𝑖 , 𝑖 = 1, 2, 3, 4 are the polynomials of 𝑏4 , 𝑟. These = 9216(1 + 𝑏42 ) (4 + 5𝑏42 + 8𝑏44 + 𝑏46 )
expressions of 𝑙𝑖 are as follows:
× (1 − 15𝑏42 − 72𝑏44 + 731𝑏46 + 581𝑏48
𝑙1 = − 24 + 3𝑏42 − 15𝑟 + 18𝑏42 𝑟 − 3𝑏44 𝑟 − 2𝑟2
−218𝑏410 − 2𝑏412 + 4𝑏414 ) ,
+ 4𝑏42 𝑟2 − 6𝑏44 𝑟2 ; (84)
Abstract and Applied Analysis 11

2
𝑅2 = Resultant [𝑙1 , 𝑙3 , 𝑟] = −2654208(1 + 𝑏42 ) Mathematical and Computer Modelling, vol. 46, no. 5-6, pp. 583–
594, 2007.
× (−349081952 + 4306069852𝑏42 [4] Y. Liu and W. Huang, “Center and isochronous center at infinity
for differential systems,” Bulletin des Sciences Mathematiques,
+ 41094839709𝑏44 − 169659957471𝑏46 vol. 128, no. 2, pp. 77–89, 2004.
[5] L. Feng and L. Yirong, “Classification of the centers and
− 996450752748𝑏48 − 1647467327229𝑏410 isochronicity for a class of quartic polynomial differential
systems,” Communications in Nonlinear Science and Numerical
− 1097111490891𝑏412 + 618270753348𝑏414 Simulation, vol. 17, no. 6, pp. 2270–2291, 2012.
[6] F. Li, J. Qiu, and J. Li, “Bifurcation of limit cycles, classification
+ 1604753703528𝑏416 + 406555036384𝑏418 of centers and isochronicity for a class of non-analytic quintic
systems,” Nonlinear Dynamics, vol. 76, no. 1, pp. 183–197, 2014.
− 507767377067𝑏420 − 144869567793𝑏422 [7] V. G. Romanovski and D. S. Shafer, The Center and Cyclic-
ity Problems. A Computational Algebra Approach, Birkhäuser,
+ 23948931114𝑏424 + 2359097649𝑏426 Berlin, Germany, 2009.
[8] V. G. Romanovski and M. Robnik, “The centre and isochronicity
−735877143𝑏428 − 4379940𝑏430 + 7310250𝑏432 ) , problems for some cubic systems,” Journal of Physics A: Mathe-
(85) matical and General, vol. 34, no. 47, pp. 10267–10292, 2001.
[9] V. G. Romanovski, X. Chen, and Z. Hu, “Linearizability of linear
in which Resultant[𝑙𝑖 , 𝑙𝑚 , 𝑟] is the resultant of 𝑙𝑖 and 𝑙𝑚 with systems perturbed by fifth degree homogeneous polynomials,”
respect to 𝑟. Journal of Physics A: Mathematical and Theoretical, vol. 40, no.
Clearly, 𝑅1 = 𝑅2 = 0 hold only if 1 + 𝑏42 = 0; namely, 22, pp. 5905–5919, 2007.
𝑏4 = ±𝑖. If 𝑏4 = ±𝑖, letting 𝜏1 = 0, then 𝑟 = −3/2. At this [10] J. Giné and V. G. Romanovski, “Integrability conditions for
time, 𝑎4 = 𝑏4 𝑟 = ±(3/2)𝑖, which contradict 𝑎4 ∈ R. Hence the Lotka-Volterra planar complex quintic systems,” Nonlinear
conclusion of Theorem 23 holds. Proof ends. Analysis: Real World Applications, vol. 11, no. 3, pp. 2100–2105,
2010.
From Theorem 23, equation groups 𝜏1 = 𝜏2 = 𝜏3 = [11] X. Chen and V. G. Romanovski, “Linearizability conditions of
0 have not solutions. Hence under condition (𝐻3 ), the time-reversible cubic systems,” Journal of Mathematical Analysis
origin of system (80) cannot become an isochronous center. and Applications, vol. 362, no. 2, pp. 438–449, 2010.
Accordingly, singular point (−1/2, 0) of system (2) cannot [12] L. Cairó, J. Chavarriga, J. Giné, and J. Llibre, “Class of reversible
become an isochronous center under condition (𝐻3 ). cubic systems with an isochronous center,” Computers and
Mathematics with Applications, vol. 38, no. 11, pp. 39–53, 1999.
Remark 24. Through the above analysis, it is clear that the sin- [13] A. Gasull, A. Guillamon, and V. Mañosa, “An explicit expression
gular point (−1/2, 0) of system (2) becomes an isochronous of the first Liapunov and period constants with applications,”
center if and only if (𝐻1 ) and one of the conditions (𝐶1 ) and Journal of Mathematical Analysis and Applications, vol. 211, no.
(𝐶2 ) hold. 1, pp. 190–212, 1997.
[14] W. S. Loud, “Behavior of the period of soutions of certain plane
autonomous systems near centers,” Contributions to Differential
Conflict of Interests Equations, vol. 3, pp. 21–36, 1964.
[15] I. Pleshkan, “A new method of investigating the isochronicity of
The author declares that there is no conflict of interests
system of two differential equations,” Differential Equations, vol.
regarding the publication of this paper. 5, pp. 796–802, 1969.
[16] C. Du, Y. Liu, and H. Mi, “A class of ninth degree system with
Acknowledgments four isochronous centers,” Computers and Mathematics with
Applications, vol. 56, no. 10, pp. 2609–2620, 2008.
This research is partially supported by the National Natural [17] C. X. Du, H. L. Mi, and Y. R. Liu, “Center, limit cycles and
Science Foundation of China (11371373 and 11261013) and isochronous center of a Z4-equivariant quintic system,” Acta
the Research Fund of Hunan Provincial Education Depart- Mathematica Sinica, vol. 26, no. 6, pp. 1183–1196, 2010.
ment and Hunan provincial Science and Technology Project [18] F. Li and M. Wang, “Bifurcation of limit cycles in a quintic
(2012FJ3106). system with ten parameters,” Nonlinear Dynamics, vol. 71, pp.
213–222, 2013.
References [19] Y. Liu and J. Li, “Periodic constants and time-angle difference
of isochronous centers for complex analytic systems,” Interna-
[1] Y. Lin and J. Li, “Normal form and critical points values of the tional Journal of Bifurcation and Chaos, vol. 16, no. 12, pp. 3747–
period of closed orbits for planar autonomous systems,” Acta 3757, 2006.
Mathematica Sinica, vol. 34, no. 4, pp. 490–501, 1991. [20] Y. Liu, “Theory of center-focus for a class of higher-degree criti-
[2] V. G. Romanovskii, “Calculation of Lyapunov numbers in the cal points and infinite points,” Science in China A: Mathematics,
case of two pure imaginary roots,” Differential Equations, vol. Physics, Astronomy, vol. 44, no. 3, pp. 365–377, 2001.
29, no. 5, pp. 782–784, 1993. [21] V. V. Amel’kin, N. A. Lukashevich, and A. P. Sadovskii, Non-
[3] W. Huang, Y. Liu, and W. Zhang, “Conditions of infinity to linear Oscillations in Second Order Systems, Belarusian State
be an isochronous centre for a rational differential system,” University, Minsk, Belarus, 1982.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 632564, 12 pages
http://dx.doi.org/10.1155/2014/632564

Research Article
Hopf Bifurcation and Stability Analysis of a Congestion
Control Model with Delay in Wireless Access Network

Wen-bo Zhao,1 Xiao-ke Sun,2 and Huicheng Wang3

1
School of Physics and Information, Tianshui Normal University, Tianshui, Gansu 741000, China
2
School of Mathematics and Statistics, Tianshui Normal University, Tianshui, Gansu 741000, China
3
Department of Mathematics, Zhejiang Normal University, Jinhua, Zhejiang 321004, China

Correspondence should be addressed to Huicheng Wang; yjs2014xia@163.com

Received 15 February 2014; Accepted 16 March 2014; Published 24 April 2014

Academic Editor: Yonghui Xia

Copyright © 2014 Wen-bo Zhao et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

We drive a scalar delay differential system to model the congestion of a wireless access network setting. The Hopf bifurcation of this
system is investigated using the control and bifurcation theory; it is proved that there exists a critical value of delay for the stability.
When the delay value passes through the critical value, the system loses its stability and a Hopf bifurcation occurs. Furthermore, the
direction and stability of the bifurcating periodic solutions are derived by applying the normal form theory and the center manifold
theorem. Finally, some examples and numerical simulations are presented to show the feasibility of the theoretical results.

1. Introduction where 𝑊𝑖 (𝑡), 𝑥𝑖 (𝑡) = 𝑊𝑖 (𝑡)/𝜏𝑖 , 𝜏𝑖 , and 𝑝(𝑡) denote the TCP
window size, TCP rate, round trap time at time 𝑡 of flow 𝑖,
Recently, the wireless access network has been wildly applied and probability of packet mark at time 𝑡, respectively.
to various fields, especially to the Internment; therefore, it has However, there are seldom works which discuss the
received significant attention. The congestion control in wire- dynamical behaviors of the congestion control model in wire-
less access network also plays a crucial role in the success of less access network such as stability and Hopf bifurcation. The
the wireless network technology. observation provides us with the motivation to investigate
The congestion and avoidance mechanism is a combina- the dynamical behaviors of the congestion control model in
tion of the end-to-end TCP congestion control mechanism [1, wireless access network.
2] at the end hosts and the queue management mechanism at In this paper, we consider the wireless access networks of
the routers. Because the congestion control algorithm is a only one bottleneck router and let 𝑛 TCP flows tracer the
highly complex dynamical model, many researchers have router. In the down link communication from the network to
given much study to its dynamics and stability. In [3–5], the the sources, the marking probability is fed back to the sources.
local stability in congestion control models is studied. In [6– During channel fading, the source has failed to receive the
9], the existence of Hopf bifurcation is analyzed in congestion marking probability. Therefore, we suppose that the drop
control models. probability is 𝑝𝑑𝑖 . In this case, the source will use the previous
For wired access network, the dynamic of window size is packet marking probability to reduce its window size, and
captured by the following equation [10]: also the window size is decreased by one by convention. Thus,
we obtain
1 − 𝑝𝑖 (𝑡) 1
1 − 𝑝𝑖 (𝑡) 1 𝑊̇𝑖 (𝑡) = 𝑥𝑖 (𝑡 − 𝜏𝑖 ) ( − 𝑝𝑖 (𝑡) 𝑊𝑖 (𝑡) (1 − 𝑝𝑑𝑖 )
𝑊̇𝑖 (𝑡) = 𝑥𝑖 (𝑡 − 𝜏𝑖 ) ( − 𝑝𝑖 (𝑡) 𝑊𝑖 (𝑡)) , 𝑊𝑖 (𝑡) 2
𝑊𝑖 (𝑡) 2 (2)
(1)
− 𝑝𝑖 (𝑡) (𝑊𝑖 (𝑡) − 1) 𝑝𝑑𝑖 ) .
𝑖 = 1, . . . , 𝑛,
2 Abstract and Applied Analysis

The dynamic of queue length of the router is captured by the Let the equilibrium point of the system (6) be (𝑥∗ , 𝑞∗ ), which
following equation [11]: should satisfy

𝑞 ̇(𝑡) = 𝐹 (∑ 𝑥𝑖 (𝑡 − 𝜏)) − 𝑐, (3) 𝐹 (𝑛𝑥∗ ) = 𝑐,

−1
(7)
where 𝑐 is the serving capacity of the link node and the 2
𝑞∗ = 2[𝑘 (2 + (1 + 𝑝𝑑 ) (𝜏𝑥∗ ) − 2𝑝𝑑 𝜏𝑥∗ )]
function 𝐹(∑ 𝑥𝑖 (𝑡)) is the adjusted rate of the source based on
the congestion rate 𝑥(𝑡) from the link node, which is a and 0 < 𝑘𝑞∗ ≤ 1.
decreasing and nonnegative derivative function. 2
Hence, (1 + 𝑝𝑑 )(𝜏𝑥∗ ) − 2𝑝𝑑 𝜏𝑥∗ > 0 and we get
Since 𝑥𝑖 (𝑡) = 𝑊𝑖 (𝑡)/𝜏𝑖 and 𝑝𝑖 (𝑡) = 𝑘𝑞(𝑡) [12], we obtain
(H1) 𝜏 ≥ 2𝑝𝑑 /𝑥∗ (1 + 𝑝𝑑 ).
1 − 𝑘𝑞 (𝑡) 1
𝑥𝑖̇(𝑡) = 𝑥𝑖 (𝑡 − 𝜏𝑖 ) [ 2 − 𝑘𝑥𝑖 (𝑡) 𝑞 (𝑡)
𝜏𝑖 𝑥 (𝑡) 2 Remark 1. Consider 𝜏 ∈ [𝜏1 , +∞), where
(4)
1 1 2𝑝𝑑
− 𝑘𝑝𝑑𝑖 𝑥𝑖 (𝑡) 𝑞 (𝑡) + 𝑘𝑝𝑑𝑖 𝑞 (𝑡) ] . 𝜏1 = . (8)
2 𝜏𝑖 𝑥∗ (1 + 𝑝𝑑 )

We assume that the 𝜏𝑖 is a constant and not time- Let 𝑦1 (𝑡) = 𝑥(𝑡)−𝑥∗ , 𝑦2 (𝑡) = 𝑞(𝑡)−𝑞∗ . Linearizing the system
varying and the queuing delay is neglected. So, we obtain the (6) about the equilibrium point, we get
following congestion model in wireless network:
𝑦1̇(𝑡) = 𝑎11 𝑦1 (𝑡) + 𝑎12 𝑦1 (𝑡 − 𝜏) + 𝑏11 𝑦2 (𝑡) ,
1 − 𝑘𝑞 (𝑡) 1 (9)
𝑥𝑖̇(𝑡) = 𝑥𝑖 (𝑡 − 𝜏) [ 2 − 𝑘𝑥𝑖 (𝑡) 𝑞 (𝑡) 𝑦2̇(𝑡) = 𝑎22 𝑦1 (𝑡 − 𝜏) ,
𝜏 𝑥𝑖 (𝑡) 2
1 1 (5) where
− 𝑘𝑝𝑑 𝑥𝑖 (𝑡) 𝑞 (𝑡) + 𝑘𝑝𝑑 𝑞 (𝑡) ] ,
2 𝜏 −1 + 𝑘𝑞∗ 1 1
𝑎11 = 𝑥∗ [ − 𝑘𝑞∗ − 𝑘𝑝𝑑 𝑞∗ ] ,
𝑞 ̇(𝑡) = 𝐹 (∑ 𝑥𝑖 (𝑡 − 𝜏)) − 𝑐. (𝜏𝑥∗ )2 2 2
1 − 𝑘𝑞∗ 1 ∗ ∗ 𝑘𝑝𝑑 𝑞∗ 1
The paper is organized as follows. In Section 2, the stabil- 𝑎12 = − 𝑘𝑥 𝑞 + − 𝑘𝑝𝑑 𝑥∗ 𝑞∗ ,
ity of trivial solutions and the existence of Hopf bifurcation 𝜏 2 𝑥∗ 2 𝜏 2 (10)
are discussed and the delay passes through the critical value, 1 1 𝑘 𝑘𝑝
the system loses its stability, and a Hopf bifurcation occurs. In 𝑏11 = 𝑥∗ [− 𝑘𝑥∗ − 𝑘𝑝𝑑 𝑥∗ − 2 ∗ + 𝑑 ] ,
2 2 𝜏𝑥 𝜏
Section 3, based on the normal form theory and the center
manifold theorem, we derive the formulas for determining 𝑎22 = 𝑛𝐷 (𝐹) (𝑛𝑥∗ ) .
the properties of the direction of the Hopf bifurcation and the
stability of bifurcating periodic solutions. In Section 4, num- Then, the characteristic equation of the linearized equation
erical simulations are given to justify the theoretical analysis. (9) is
Finally, the conclusions appear in Section 5.
Since we focus on dynamical behavior analysis of the 𝐷 (𝜆, 𝜏) = 𝜆2 − 𝑎11 𝜆 − 𝑎12 𝜆𝑒−𝜆𝜏 − 𝑎22 𝑏11 𝑒−𝜆𝜏 = 0. (11)
above model in the wireless access networks, we only need to
choose the communication delay as the bifurcation parame- Note that the coefficients 𝑎11 , 𝑎12 , and 𝑏11 depend on time
ter. delay 𝜏, since 𝑞∗ is connected with 𝜏. In order to apply the geo-
It is worth to point out that recent many works have been metric criterion of Kuang [17, 18], we rewrite 𝐷(𝜆, 𝜏) = 0 into
done for wired access network. For details, we refer to [13–16].
𝐷 (𝜆, 𝜏) = 𝑃 (𝜆, 𝜏) + 𝑄 (𝜆, 𝜏) 𝑒−𝜆𝜏 , (12)
2. Stability of the System with where
Communication Delay
𝑃 (𝜆, 𝜏) = 𝜆2 − 𝑎11 𝜆,
In this section, we assume that 𝑥𝑖 (𝑡), 𝑖 = 1, . . . , 𝑛 is equal to (13)
𝑥(𝑡), so (5) can be rewritten as follows: 𝑄 (𝜆, 𝜏) = − 𝑎22 𝑏11 − 𝑎12 𝜆.

1 − 𝑘𝑞 (𝑡) 1 1 Lemma 2. If (H1) holds, then

𝑥̇(𝑡) = 𝑥 (𝑡 − 𝜏) [ − 𝑘𝑥 (𝑡) 𝑞 (𝑡) − 𝑘𝑝𝑑 𝑥 (𝑡) 𝑞 (𝑡)
𝜏2 𝑥 (𝑡) 2 2 (a) 𝑃(0, 𝜏) + 𝑄(0, 𝜏) ≠0;
1 (b) 𝑃(𝜔𝑖, 𝜏) + 𝑄(𝜔𝑖, 𝜏) ≠0 for all 𝜔 ∈ 𝑅;
+ 𝑘𝑝𝑑 𝑞 (𝑡) ] ,
𝜏 (c) lim sup{|𝑄(𝜆, 𝜏)/𝑃(𝜆, 𝜏)| : |𝜆| → +∞, Re 𝜆 ≥ 0} < 1;
𝑞 ̇(𝑡) = 𝐹 (𝑛𝑥 (𝑡 − 𝜏)) − 𝑐. (d) 𝐹(𝜔, 𝜏) = |𝑃(𝜔𝑖, 𝜏)|2 −|𝑄(𝜔𝑖, 𝜏)|2 for each 𝜏 has at most
(6) a finite number of real zeros;
Abstract and Applied Analysis 3

(e) each positive root 𝜔(𝜏) of 𝐹(𝜔, 𝜏) = 0 is continuous and For 𝜏 ∈ [𝜏1 , +∞), let 𝜃(𝜏) ∈ (0, 2𝜋) be defined by
differentiable in 𝜏 whenever it exists.
𝜔 (𝑎11 𝑎22 𝑏11 − 𝑎12 𝜔2 )
Proof. (a) For 𝜏 ∈ [𝜏1 , +∞), sin 𝜃 (𝜏) = 2 𝑏2 + 𝑎2 𝜔2
,
𝑎22 11 12
(22)
𝑃 (0, 𝜏) + 𝑄 (0, 𝜏) = − 𝑎22 𝑏11 ≠0. (14) 𝜔2 (𝑎22 𝑏11 − 𝑎11 𝑎12 )
cos 𝜃 (𝜏) = 2 𝑏2 + 𝑎2 𝜔2
,
𝑎22
(b) Consider 𝑃(𝜔𝑖, 𝜏)+𝑄(𝜔𝑖, 𝜏) = −𝜔2 −𝑏11 𝑎22 +𝑖(−𝜔𝑎11 − 11 12

𝜔𝑎12 ) ≠0. which combines with (18) and defines the following maps:
(c) From (13), we get
𝜃 (𝜏) + 2𝑛𝜋
󵄨󵄨 𝑄 (𝜆, 𝜏) 󵄨󵄨 󵄨󵄨 𝑎 𝜆 − 𝑏 𝑎 󵄨󵄨 𝑆𝑛 (𝜏) = 𝜏 − , 𝑛 ∈ 𝑁. (23)
󵄨 󵄨󵄨 󵄨 11 22 󵄨󵄨 𝜔 (𝜏)
lim 󵄨󵄨󵄨 󵄨󵄨 = lim 󵄨󵄨󵄨 12 2 󵄨󵄨 = 0. (15)
|𝜆| → +∞ 󵄨󵄨 𝑃 (𝜆, 𝜏) 󵄨󵄨 |𝜆| → +∞ 󵄨󵄨 𝜆 − 𝑎11 𝜆 󵄨󵄨
According to [18] and the above discussion, we have the
Hence, lim sup{|𝑄(𝜆, 𝜏)/𝑃(𝜆, 𝜏)| : |𝜆| → +∞, Re 𝜆 ≥ 0} = following result.
0 < 1.
(d) From (13), we get Theorem 3. Assume that (H1) is satisfied, and then 𝜆 =
±𝜔(𝜏0 )𝑖, 𝜏0 ∈ (0, 𝜏1 ), are a pair of simple and conjugate pure
𝐹 (𝜔, 𝜏) = |𝑃(𝜔𝑖, 𝜏)|2 − |𝑄(𝜔𝑖, 𝜏)|2 imaginary roots of the characteristic equation (11) if and only if
(16) 𝑆0 (𝜏0 ) = 0 for some 𝑛 ∈ 𝑁. This pair of simple conjugate pure
= 𝜔4 + 𝑎11
2 2 2 2
𝜔 − 𝑎12 2 2
𝜔 − 𝑏11 𝑎22 . imaginary roots crosses the imaginary axis from left to right if
𝛿(𝜏0 ) > 0 and crosses the imaginary axis from right to left if
Hence, (d) holds. 𝛿(𝜏0 ) < 0, where
(e) 𝐹(𝜔, 𝜏) is continuous for 𝜔 and 𝜏 and differentiable in
𝜔; hence, implicit function theorem implies (e). This com- 𝑑 Re 𝜆 󵄨󵄨󵄨󵄨 𝑑𝑆 (𝜏) 󵄨󵄨󵄨󵄨
𝛿 (𝜏0 ) := sign { 󵄨󵄨 } = sign { 𝑛 󵄨 }.
pletes the proof of the theorem. 𝑑𝜏 󵄨󵄨𝜆=𝜔(𝜏0 )𝑖 𝑑𝜏 󵄨󵄨󵄨𝜏=𝜏0
(24)
Supposing that 𝐷(𝜔𝑖, 𝜏) = 0 and 𝜔 > 0, we get
By the the expression of 𝑎11 , 𝑎12 , and 𝑏11 , we know that
𝜔 (𝑎11 𝑎22 𝑏11 − 𝑎12 𝜔 )2 they have singularity at 𝜏 = 0. We can not gain the conclusion
sin 𝜔𝜏 = 2 𝑏2 + 𝑎2 𝜔2
, that the equilibrium (𝑥∗ , 𝑝∗ ) by discussing roots of the char-
𝑎22 11 12 acteristic equation 𝐷(𝜆, 0) = 0. To our knowledge, this case is
(17)
rarely considered by papers. But we can get the stability of the
𝜔2 (𝑎22 𝑏11 − 𝑎11 𝑎12 )
cos 𝜔𝜏 = − 2 𝑏2 + 𝑎2 𝜔2
. system (6) when 𝜏 = 𝜏0 /2 by discussing the stability of the
𝑎22 11 12 following auxiliary system:
Hence, 𝑦1̇(𝑡) = 𝑐11 𝑦1 (𝑡) + 𝑐12 𝑦1 (𝑡 − 𝑟) + 𝑑11 𝑦2 (𝑡) ,
(25)
𝐹 (𝜔, 𝜏) = |𝑃(𝜔𝑖, 𝜏)|2 − |𝑄(𝜔𝑖, 𝜏)|2 𝑦2̇(𝑡) = 𝑐22 𝑦1 (𝑡 − 𝑟) ,

= 𝜔4 + 𝑎11
2 2 2 2
𝜔 − 𝑎12 2 2
𝜔 − 𝑏11 𝑎22 (18) where
󵄨
= 0. 𝑐11 = 𝑎11 󵄨󵄨󵄨 𝜏=𝜏 /2 ,
0

󵄨
Let 𝑧 = 𝜔2 , and then (18) can be rewritten as 𝑐12 = 𝑎12 󵄨󵄨󵄨 𝜏=𝜏 /2 ,
0

󵄨 (26)
𝑧2 + (𝑎11
2 2
− 𝑎12 2 2
) 𝑧 − 𝑏11 𝑎22 = 0. (19) 𝑐22 = 𝑎22 󵄨󵄨󵄨 𝜏=𝜏 /2 ,
0

󵄨
Denote 𝑑11 = 𝑏11 󵄨󵄨󵄨 𝜏=𝜏 /2 .
0

ℎ (𝑧, 𝜏) = 𝑧2 + (𝑎11
2 2
− 𝑎12 2 2
) 𝑧 − 𝑏11 𝑎22 . (20) Then, the characteristic equation of the linearized equa-
tion (25) is
2 2
Since − 𝑏11 𝑎22 < 0, the equation ℎ(𝑧, 𝜏) = 0 has one
positive root. We denote that the positive root is 𝑧+ . Then, (18) 𝜆 − 𝑐11 𝜆 − 𝑐12 𝜆𝑒−𝜆𝑟 − 𝑐22 𝑑11 𝑒−𝜆𝑟 = 0. (27)
has positive real root 𝜔(√𝑧+ ), where Definition 4. For simplicity, let

2 𝐷0 (𝜆, 𝑟) = 𝜆2 − 𝑐11 𝜆 − 𝑐12 𝜆𝑒−𝜆𝑟 − 𝑐22 𝑑11 𝑒−𝜆𝑟 . (28)

√ − (𝑎11 − 𝑎12 ) + √ (𝑎11 − 𝑎12 ) + 4𝑏11 𝑎22
2 2 2 2 2 2
𝜔 = 𝜔 (𝜏) = .
2 Lemma 5. The equilibrium (0, 0) of system (25) is locally
(21) asymptotically stable when 𝑟 = 0.
4 Abstract and Applied Analysis

Proof. When 𝑟 = 0, (27) becomes Denote

ℎ0 (𝑧) = 𝑧3 + 𝑅1 𝑧2 + 𝑅2 𝑧 + 𝑅3 . (37)
𝜆2 − (𝑐11 + 𝑐12 ) 𝜆 − 𝑐22 𝑑11 = 0. (29)
Since lim𝑧 → +∞ ℎ0 (𝑧) = +∞ and 𝑅3 < 0, (36) has at the
Further, if least one positive root. We define
(H2) 𝑐11 + 𝑐12 < 0 and 𝑐22 𝑑11 < 0 4 3 1 2 2 4 3 2
Δ= 𝑅2 − 𝑅1 𝑅2 + 𝑅1 𝑅3 − 𝑅1 𝑅2 𝑅3 + 𝑅32 .
27 27 27 3
is satisfied, all roots of (29) have negative real parts by the (38)
Routh-Hurwitz criteria. So, when 𝑟 = 0, the equilibrium
point (0,0) of system (25) is locally asymptotically stable. This Lemma 6. For cubic equation (36), the following cases need to
completes the proof of the lemma. be considered [19]:
(a) if Δ > 0, then the equation has three distinct real roots;
Let 𝜆 = ±𝑖𝜔𝑟0 , where 𝜔𝑟0 > 0. Substituting it into (27) and
(b) if Δ = 0, then the equation has a multiple root and all
separating the real and imaginary parts, we have
its roots are real;
2
− 𝜔𝑟0 − 𝑐12 𝜔𝑟0 sin 𝜔𝑟0 𝑟 − 𝑐22 𝑑11 cos 𝜔𝑟0 𝑟 = 0, (c) if Δ < 0, then the equation has one real root and two
(30) nonreal complex conjugate roots.
− 𝑐11 𝜔𝑟0 − 𝑐12 𝜔𝑟0 cos 𝜔𝑟0 𝑟 − 𝑐22 𝑑11 sin 𝜔𝑟0 𝑟 = 0.
Without loss of generality, we assume that (36) has three
2
It follows from (30) that positive roots: 𝑧01 , 𝑧02 , and 𝑧03 . Since 𝑧 = 𝜔𝑟0 and 𝜔𝑟0 > 0, we
have
2
𝜔𝑟0 (𝑐11 𝑐22 𝑑11 − 𝑐12 𝜔𝑟0 )
sin 𝜔𝑟0 𝑟 = , 𝜔𝑟01 = √𝑧01 , 𝜔𝑟02 = √𝑧02 , 𝜔𝑟01 = √𝑧03 . (39)
2 𝑑2 + 𝑐2 𝜔2
𝑐22 22 12 𝑟0
(31) Thus, we know that
𝜔2 (𝑐 𝑑 − 𝑐 𝑐 ) 2
cos 𝜔𝑟0 𝑟 = − 𝑟0 2 222 11 2 112 12 . (𝑠) 1 𝜔𝑟0𝑗 (𝑐22 𝑑11 − 𝑐11 𝑐12 )
𝑐22 𝑑22 + 𝑐12 𝜔𝑟0 𝑟0𝑗 = [arccos ( 2 2 + 𝑐2 𝜔2
) + 2𝑠𝜋] ,
𝜔𝑟0𝑗 𝑐22 𝑏22 12 𝑟0 (40)
Since sin2 (𝜔𝑟0 𝑟) + cos2 (𝜔𝑟0 𝑟) = 1, we have
𝑗 = 1, 2, 3; 𝑠 = 0, 1, 2, . . . .
6 4 2
𝑆0 𝜔𝑟0 + 𝑆1 𝜔𝑟0 + 𝑆2 𝜔𝑟0 + 𝑆3 = 0, (32) Denote
(0)
where 𝑟0 = min {𝑟0𝑗 }. (41)
𝑗∈{1,2,3}
2
𝑆0 = 𝑐12 , Lemma 7. Assume that 𝜆 = ±𝑖𝜔𝑟0𝑗 are simple roots of (27)
(𝑠)
2 2 2 2 4 when 𝑟 = 𝑟0𝑗 .
𝑆1 = 𝑐11 𝑐12 + 𝑐22 𝑑11 − 𝑐12 ,
(33)
2 2 2
𝑆2 = 𝑐11 2 2 2
𝑐22 𝑑11 − 2𝑐12 𝑐22 𝑑11, Proof. Since 𝐷0 (𝜆, 𝑟) = 𝜆2 − 𝑐11 𝜆 − 𝑐12 𝜆𝑒−𝜆𝑟 − 𝑐22 𝑑11 𝑒−𝜆𝑟 , we
obtain
4 4
𝑆3 = − 𝑐22 𝑑11 . 𝑑𝐷0 (𝜆)
= 2𝜆 − 𝑐11 + 𝑐22 𝑑11 𝑟𝑒−𝜆𝑟 − 𝑐12 𝑒−𝜆𝑟 + 𝑐12 𝜆𝑟𝑒−𝜆𝑟 .
𝑑𝜆
Since 𝑆0 > 0, we can rewrite (32) as (42)
6
𝜔𝑟0 4
+ 𝑅1 𝜔𝑟0 2
+ 𝑅2 𝜔𝑟0 + 𝑅3 = 0, (34) Substituting 𝜆 = 𝑖𝜔𝑟 , 𝑟 = 𝑟0 into (42), by using (30), we
can obtain
where 𝑑𝐷0 (𝑖𝜔𝑟 )
= − 𝑐11 + 𝑐22 𝑑11 𝑟0 cos (𝜔𝑟0 𝑟0 ) − 𝑐12 cos (𝜔𝑟0 𝑟0 )
2 2
𝑐11 2 2
𝑐12 + 𝑐22 4
𝑑11 − 𝑐12 𝑑𝜆
𝑅1 = ,
2
𝑐12 + 𝑐12 𝜔𝑟0 𝑟0 sin (𝜔𝑟0 𝑟0 )
2 2 2
𝑐11 2 2 2
𝑐22 𝑑11 − 2𝑐12 𝑐22 𝑑11 + 𝑖 [2𝜔𝑟0 − 𝑐22 𝑑11 𝑟0 sin (𝜔𝑟0 𝑟0 ) − 𝑐12 sin (𝜔𝑟0 𝑟0 )]
𝑅2 = 2
, (35)
𝑐12 + 𝑖 [𝑐12 𝜔𝑟0 𝑟0 cos (𝜔𝑟0 𝑟0 )] ≠0.
4 4 (43)
− 𝑐22 𝑑11
𝑅3 = 2
.
𝑐12 Similarly, we can get
2 𝑑𝐷0 (−𝑖𝜔𝑟 )
Let 𝑧 = 𝜔𝑟0 ; then, (34) can be rewritten as ≠0. (44)
𝑑𝜆
𝑧3 + 𝑅1 𝑧2 + 𝑅2 𝑧 + 𝑅3 = 0. (36) This completes the proof of the lemma.
Abstract and Applied Analysis 5

Hence, ±𝑖𝜔𝑟0𝑗 is a simple pair of purely imaginary roots of (25) are the same system. So, there are no roots of 𝐷(𝜆, 𝜏0 /2) =
(𝑠)
(27) with 𝑟 = 𝑟0𝑗 . 𝐷0 (𝜆, 𝜏0 /2) = 0 with nonnegative real parts and the equilib-
rium point of system (6) is locally asymptotically stable when
Lemma 8. Let 𝜆(𝑟) = 𝜇(𝑟) + 𝑖𝜔𝑟 (𝑟) be the root of (27) 𝜏 = 𝜏0 /2. This completes the proof of the lemma.
satisfying 𝜇(𝑟0 ) = 0, 𝜔𝑟 (𝑟0 ) = 𝜔𝑟0 ; the following transversality According to [18] and the above discussion, we have
condition holds: following the result.
𝑑 Re (𝜆(𝑟)) 󵄨󵄨󵄨󵄨
󵄨󵄨 ≠0. (45) Theorem 11. Assume that (H1), (H2), and (H3) hold, if the
𝑑𝑟 󵄨󵄨𝑟=𝑟0
function 𝑆0 (𝜏) has positive zeros in (0, 𝜏1 ); the equilibrium
Proof. By equation (27) with respect to 𝑟 and applying the (𝑥∗ , 𝑝∗ ) of system (6) is asymptotically stable for all 𝜏 ∈ [𝜏1 , 𝜏0 )
implicit function theorem, we get and becomes unstable for staying in some right neighborhood of
𝜏0 ; hence, system (6) undergoes Hopf bifurcation when 𝜏 = 𝜏0 .
𝑑𝜆 (𝑟) − 𝜆𝑒−𝜆𝑟 (𝑐22 𝑑11 + 𝑐12 𝜆)
= .
𝑑𝑟 2𝜆 − 𝑐11 + 𝑐22 𝑑11 𝑟𝑒−𝜆𝑟 − 𝑐12 𝑒−𝜆𝑟 + 𝑐12 𝜆𝑟𝑒−𝜆𝑟 3. Direction and Stability of
(46) the Hopf Bifurcation
Since 𝜆(𝑟0 ) = 𝑖𝜔𝑟0 , we obtain In this section, we will study the direction of Hopf bifurcation
and the stability of bifurcating periodic solution of system (6)
𝑑𝜆 (𝑟) 𝑝1 + 𝑝2 sin (𝜔𝑟0 𝑟0 ) + 𝑝3 cos (𝜔𝑟0 𝑟0 ) at 𝜏 = 𝜏0 . The approach employed here is the normal form
Re = , (47) method and center manifold theorem introduced by Hassard
𝑑𝑟 𝑞1 + 𝑞2
[20]. More precisely, we will compute the reduced system
where on the center manifold with the pair of conjugate complex,
2 2
purely imaginary solutions of the characteristic equation (11).
𝑝1 = − 𝑐12 𝜔𝑟0 , 𝑝2 = − 𝜔𝑟0 (2𝑐12 𝜔𝑟0 − 𝑐11 𝑐22 𝑑11 ) , By this reduction, we can determine the Hopf bifurcation
direction, that is, to answer the question of whether the
𝑝2 = − 𝜔𝑟0 (𝑐11 𝑐12 − 2𝑐22 𝑑11 ) , bifurcation branch of periodic solution exists locally for
supercritical bifurcation or subcritical bifurcation.
𝑞1 = [− 𝑐11 + (𝑐22 𝑑11 𝑟0 − 𝑐12 ) cos (𝜔𝑟0 𝑟0 )
(48) Let 𝜏 = 𝜏0 + 𝜇, 𝑢𝑖 (𝑡) = 𝑦𝑖 (𝜏𝑡), (𝑖 = 1, 2), 𝜇 ∈ 𝑅, 𝐿 𝜇 : 𝐶 →
2
+ 𝑐12 𝜔0 𝑟0 sin (𝜔𝑟0 𝑟0 )] , 𝑅 , and 𝐹 : 𝑅×𝐶 → 𝑅2 , so that system (6) is transformed into
2

an FDE in 𝐶 = 𝐶([−1, 0], 𝑅2 ) as

𝑞2 = [2𝜔𝑟0 − (𝑐22 𝑑11 𝑟0 − 𝑐12 ) sin (𝜔𝑟0 𝑟0 )
𝑢̇(𝑡) = 𝐿 𝜇 (𝑢𝑡 ) + 𝐹 (𝜇, 𝑢𝑡 ) , (50)
2
+ 𝑐12 𝜔0 𝑟0 cos (𝜔𝑟0 𝑟0 )] .
with
By again using (30), we can obtain
𝐿 𝜇 𝜑 = (𝜏0 + 𝜇) [𝐵𝜑 (0) + 𝐶𝜑 (−1)] ,
𝑝1 + 𝑝2 sin (𝜔𝑟0 𝑟0 ) + 𝑝3 cos (𝜔𝑟0 𝑟0 ) ≠0,
(49) 𝐹1 (𝜑, 𝜇) = (𝜏0 + 𝜇) [𝑚1 𝜑12 (0) + 𝑚2 𝜑1 (0) 𝜑1 (−1)
𝑞1 + 𝑞2 > 0.
+ 𝑚3 𝜑1 (0) 𝜑2 (0) + 𝑚4 𝜑1 (−1) 𝜑2 (0)
Hence, (𝑑 Re(𝜆(𝑟))/𝑑𝑟)|𝑟=𝑟0 ≠0. This completes the proof
of the lemma. + 𝑚5 𝜑13 (0) + 𝑚6 𝜑12 (0) 𝜑1 (−1)

From the above discussion about the system (25), we have + 𝑚7 𝜑12 (0) 𝜑2 (0)
the following result.
+ 𝑚8 𝜑1 (0) 𝜑1 (−1) 𝜑2 (0) + h.o.t] ,
Theorem 9. When 𝑟 < 𝑟0 , the equilibrium point of system (25)
is locally asymptotically stable. 𝐹2 (𝜑, 𝜇) = (𝜏0 + 𝜇) [𝑛1 𝜑12 (−1) + 𝑛2 𝜑13 (−1)] ,
Further, if (51)
(H3) 𝑟0 (𝜏0 ) > 𝜏0 /2 where
is satisfied, we will get the following lemma. 𝑎11 𝑏11 𝑎12 0
𝐵=[ ], 𝐶=[ ],
0 0 𝑎22 0
Lemma 10. The equilibrium point of system (6) is locally
asymptotically stable when 𝜏 = 𝜏0 /2. 1 − 𝑘𝑞∗
𝑚1 = ,
𝜏2
Proof. Since Lemma 5 and hypotheses (H3), the equilibrium
point of system (25) is locally asymptotically stable when 𝑟 = −1 + 𝑘𝑞∗ 1 1
𝑚2 = − 𝑘𝑞∗ − 𝑘𝑝𝑑 𝑞∗ ,
𝜏0 /2. When 𝑟 = 𝜏0 /2 and 𝜏 = 𝜏0 /2, system (9) and system (𝜏𝑥∗ )2 2 2
6 Abstract and Applied Analysis

𝑘 1 1 Let 𝜇 = 0; by the discussion in the previous section,

𝑚3 = 𝑥∗ ( 2
− 𝑘 − 𝑘𝑝𝑑 ) ,
∗
(𝜏𝑥 ) 2 2 we know that ± 𝜔0 𝜏0 𝑖 are common eigenvalues of 𝐴(0) and
1 1 𝑘 𝑘𝑝 𝐴∗ (0). We need to compute the eigenvector of 𝐴(0) and 𝐴∗ (0)
𝑚4 = − 𝑘𝑥∗ − 𝑘𝑝𝑑 𝑥∗ − 2 ∗ + 𝑑 , corresponding to 𝜔0 𝜏0 𝑖 and − 𝜔0 𝜏0 𝑖, respectively. Suppose that
2 2 𝜏𝑥 𝜏
𝑞(𝜃) and 𝑞∗ (𝜃) are the eigenvector of 𝐴(0) and 𝐴∗ (0) corre-
𝑘𝑞∗ − 1
𝑚5 = 2 ∗ 3 , sponding to 𝜔0 𝜏0 𝑖 and − 𝜔0 𝜏0 𝑖, respectively; then, we have
𝜏 (𝑥 )
1 − 𝑘𝑞∗ −𝑘 𝐴 (0) 𝑞 (𝜃) = 𝜔0 𝜏0 𝑖𝑞 (𝜃) , (60)
𝑚6 = 2 ∗ 3 , 𝑚7 = ∗ 2 ,
𝜏 (𝑥 ) (𝑥 𝜏)
𝐴∗ (0) 𝑞∗ (𝜃) = −𝜔0 𝜏0 𝑖𝑞∗ (𝜃) . (61)
𝑘 1 1
𝑚8 = ∗ 2 − 𝑘 − 𝑘𝑝𝑑 ,
(𝑥 𝜏) 2 2 Then, we have the following lemma.
1 2 (2)
𝑛1 = 𝑛 𝐷 (𝐹) (𝑛𝑥∗ ) ,
2 Lemma 12. Consider
1
𝑛2 = 𝑛3 𝐷(3) (𝐹) (𝑛𝑥∗ ) . 1
6 𝑞 (𝜃) = ( ) 𝑒𝜔0 𝜏0 𝑖𝜃 , 𝜃 ∈ [−1, 0] ,
𝛾
(52)
(62)
Then, 𝐿 𝜇 is a one parameter family of bounded linear operator 𝑞∗ (𝑠) = 𝐷 (1 𝛾∗ ) 𝑒𝜔0 𝜏0 𝑖𝜃 , 𝜃 ∈ [0, 1] ,
in 𝐶([−1, 0], R2 ). By the Riesz representation theorem, there ⟨𝑞∗ , 𝑞⟩ = 1, ⟨𝑞∗ , 𝑞⟩ = 0,
exists a function 𝜂(𝜃, 𝜇) of bounded variation for 𝜃 ∈ [−1, 0]
such that where
0
𝐿 𝜇 𝜙 = ∫ 𝑑𝜂 (𝜃, 𝜇) 𝜙 (𝜃) , 𝜙 ∈ 𝐶. (53) 𝑎22 𝑒−𝜔0 𝜏0 𝑖 𝑏11 𝑖
−1 𝛾= , 𝛾∗ = ,
𝜔0 𝑖 𝜔0 (63)
In fact, we can choose −𝑖𝜔0 𝜏0 −𝑖𝜔0 𝜏0
𝐷 = 1 + 𝛾𝛾∗ − 𝜏0 𝑎12 𝑒 − 𝛾∗ 𝜏0 𝑎22 𝑒 .
𝜂 (𝜃, 𝜇) = (𝜏0 + 𝜇) [𝐵𝛿 (𝜃) − 𝐶𝛿 (𝜃 + 1)] , (54)
Proof. From (55), we can rewrite (60) as
where 𝛿(𝜃) is Dirac delta function. For 𝜙 ∈ 𝐶1 ([−1, 0], 𝑅2 ),
the infinitesimal generator 𝐴(𝜇) is defined by 𝑑𝑞 (𝜃)
= 𝑖𝜔0 𝜏0 𝑞 (𝜃) , 𝜃 ∈ [−1, 0) ,
𝑑𝜙 𝑑𝜃
{
{ 𝑑𝜃 ,
{ 𝜃 ∈ [−1, 0) ,
0
(64)
𝐴 (𝜇) 𝜙 (𝜃) = { 0 (55) ∫ 𝑑𝜂 (𝑠, 0) 𝜑 (𝑠) = 𝐴 (0) 𝑞 (0) = 𝑖𝜔0 𝜏0 𝑞 (0) , 𝜃 = 0.
{
{∫ 𝑑𝜂 (𝑠, 𝜇) 𝜙 (𝑠) , 𝜃 = 0. −1
{ −1
Based on (53) and (64), we have
Further, let
0, 𝜃 ∈ [−1, 0) , 𝜏0 [𝐵𝑞 (0) + 𝐶𝑞 (−1)] = 𝐴 (0) 𝑞 (0) = 𝑖𝜔0 𝜏0 𝑞 (0) . (65)
𝑅 (𝜇) 𝜙 (𝜃) = { (56)
𝐹 (𝜇, 𝜙) , 𝜃 = 0,
For 𝑞(−1) = 𝑞(0)𝑒−𝑖𝜔0 𝜏0 , we have
and then system (50) is equivalent to
𝑢̇ 𝜏0 [𝐵𝑞 (0) + 𝐶𝑞 (0) 𝑒−𝑖𝜔0 𝜏0 ] = 𝑖𝜔0 𝜏0 𝑞 (0) . (66)
𝑡 = 𝐴 (𝜇) 𝑢𝑡 + 𝑅 (𝜇) 𝑢𝑡 , (57)

where 𝑢𝑡 (𝜃) = 𝑢(𝑡 + 𝜃) for 𝜃 ∈ [−1, 0]. We can choose 𝑞(0) = (1, 𝛾)𝑇 and get
The adjoint operator 𝐴∗ (𝜇) of 𝐴(𝜇) is defined by
𝑑𝜓 𝑎22 𝑒−𝜔0 𝜏0 𝑖
{ − , 𝜃 ∈ (0, 1] , 𝛾= , (67)
{
{ 𝜔0 𝑖
𝑑𝜃
𝐴∗ (𝜇) 𝜓 (𝜃) = { 0 (58)
{
{∫ 𝜓 (−𝑠) 𝑑𝜂 (𝑠, 𝜇) , 𝜃 = 0, 𝑇
𝑞 (𝜃) = (1, 𝛾) 𝑒𝑖𝜔0 𝜏0 𝜃 . (68)
{ −1
and a bilinear form Similar to the proof of (64)–(68), we can obtain
0 𝜃
𝑖𝑏11
⟨𝜓, 𝜙⟩ = 𝜓 (0) 𝜙 (0) − ∫ ∫ 𝜓 (𝜉 − 𝜃) 𝑑𝜂 (𝜃, 𝜇) 𝜙 (𝜉) 𝑑𝜉, 𝑞∗ (0) = (1, ),
−1 0 𝜔0
(59) (69)
∗𝑖𝑏
𝛾 = 11 .
where 𝜓 ∈ 𝐶∗ = 𝐶1 ([0, 1], 𝑅2∗ ) and 𝑅2∗ are row vector space. 𝜔0
Abstract and Applied Analysis 7

Now, we can calculate ⟨𝑞∗ , 𝑞⟩ as = 𝑖𝜔0 𝜏0 𝑧 (𝑡) + 𝑞∗ (0) 𝑅 (0) 𝑢𝑡

0 𝜃
∗ ∗
0 𝜃
−𝑖𝜔0 𝜏0 (𝜉−𝜃) − ∫ ∫ 𝑞∗ (𝜉 − 𝜃) [𝑑𝜂 (𝜃, 0)] 𝑅𝑢𝑡 (𝜉) 𝑑𝜉
⟨𝑞 , 𝑞⟩ = 𝐷 (1, 𝛾 ) − ∫ ∫ 𝐷 (1, 𝛾∗ ) 𝑒 [𝑑𝜂 (𝜃)] −1 0
−1 𝜉=0

𝑇 𝑖𝜔0 𝜏0 𝜉
= 𝑖𝜔0 𝜏0 𝑧 (𝑡) + 𝑞∗ (0) 𝐹 (0, 𝑢𝑡 (𝜃)) − 0
× (1, 𝛾) 𝑒 𝑑𝜉
= 𝑖𝜔0 𝜏0 𝑧 (𝑡) + 𝑞∗ (0) 𝐹0 (𝑧 (𝑡) , 𝑧 (𝑡))
0
𝑖𝜔0 𝜏0 𝜃 𝑇
= 𝐷 [1 + 𝛾𝛾∗ − ∫ (1, 𝛾∗ ) 𝜃𝑒 [𝑑𝜂 (𝜃)] (1, 𝛾) ] := 𝑖𝜔0 𝜏0 𝑧 (𝑡) + 𝑔 (𝑧, 𝑧) ,
−1

𝑇
(74)
= 𝐷 [1 + 𝛾𝛾∗ − (1, 𝛾∗ ) [− 𝜏0 𝐶𝑒−𝑖𝜔0 𝜏0 ] (1, 𝛾) ]

= 𝐷 [1 + 𝛾𝛾∗ − 𝜏0 𝑎12 𝑒−𝑖𝜔0 𝜏0 − 𝛾∗ 𝜏0 𝑎22 𝑒−𝑖𝜔0 𝜏0 ] = 1. where

(70)
𝑧2 𝑧2 𝑧2 𝑧
𝑔 (𝑧, 𝑧) = 𝑔20 + 𝑔11 𝑧𝑧 + 𝑔02 + 𝑔21 + ⋅⋅⋅ . (75)
On the other hand, since ⟨𝜓, 𝐴𝜑⟩ = ⟨𝐴∗ 𝜓, 𝜑⟩, we have 2 2 2

So, we can get

−𝑖𝜔0 𝜏0 ⟨𝑞∗ , 𝑞⟩ = ⟨𝑞∗ , 𝐴𝑞⟩

= ⟨𝐴∗ 𝑞∗ , 𝑞⟩ 𝑔 (𝑧, 𝑧) = 𝑞∗ (0) 𝐹0 (𝑧 (𝑡) , 𝑧 (𝑡))

(71) (76)
= ⟨−𝑖𝜔0 𝜏0 𝑞∗ , 𝑞⟩ = 𝐷 (1, 𝛾∗ ) (𝐹1 (0, 𝑢𝑡 ) , 𝐹2 (0, 𝑢𝑡 )) ,
𝑇

= 𝑖𝜔0 𝜏0 ⟨𝑞∗ , 𝑞⟩.

where
∗
Therefore, ⟨𝑞 , 𝑞⟩ = 0. This completes the proof of the
lemma. 𝐹1 (0, 𝑢𝑡 ) = 𝜏0 [𝑚1 𝜑12 (0) + 𝑚2 𝜑1 (0) 𝜑1 (−1)

+ 𝑚3 𝜑1 (0) 𝜑2 (0) + 𝑚4 𝜑1 (−1) 𝜑2 (0)

In the remainder of this section, by using the same nota- + 𝑚5 𝜑13 (0) + 𝑚6 𝜑12 (0) 𝜑1 (−1)
tions as in Hassard [20], we first compute the coordinates to (77)
describe the center manifold 𝐶0 at 𝜇 = 0, which is a locally + 𝑚7 𝜑12 (0) 𝜑2 (0)
invariant, attracting two-dimensional manifold in 𝐶0 . Let 𝑋𝑡
be the solution of (57) when 𝜇 = 0. Define + 𝑚8 𝜑1 (0) 𝜑1 (−1) 𝜑2 (0) + h.o.t] ,

𝐹2 (0, 𝑢𝑡 ) = 𝜏0 [𝑛1 𝜑12 (−1) + 𝑛2 𝜑13 (−1)] .

𝑧 (𝑡) = ⟨𝑞∗ , 𝑢𝑡 ⟩ ,
(72)
𝑊 (𝑡, 𝜃) = 𝑊 (𝑧, 𝑧, 𝜃) = 𝑢𝑡 (𝜃) − 2 Re {𝑧 (𝑡) 𝑞 (𝜃)} , Since 𝑢𝑡 = 𝑢(𝑡 + 𝜃) = 𝑊(𝑧, 𝑧, 𝜃) + 𝑧𝑞 + 𝑧𝑞 and 𝑞(𝜃) =
(1, 𝛾)𝑒𝑖𝜔0 𝜏0 𝜃 , we have
and, then on the center manifold 𝐶0 , we have
𝑢 (𝑡 + 𝜃) 𝑊(1) (𝑡 + 𝜃)
𝑧2 𝑧2 𝑢𝑡 = ( 1 ) = ( (2) )
𝑊 (𝑧, 𝑧, 𝜃) = 𝑊20 (𝜃) + 𝑊11 (𝜃) 𝑧𝑧 + 𝑊02 (𝜃) + ⋅⋅⋅ , 𝑢2 (𝑡 + 𝜃) 𝑊 (𝑡 + 𝜃)
2 2
(73) 1 1
+ 𝑧 ( ) 𝑒𝑖𝜔0 𝜏0 𝜃 + 𝑧 ( ) 𝑒−𝑖𝜔0 𝜏0 𝜃 ,
𝛾 𝛾
and then 𝑧 and 𝑧 are local coordinates for center manifold 𝐶0
in the direction of 𝑞 and 𝑞∗ . Note that 𝑊 is real if 𝑢𝑡 is real (1) 𝑧2 (1)
𝜑1 (0) = 𝑧 + 𝑧 + 𝑊20 (0) + 𝑊11 (0) 𝑧𝑧
and we only deal with real solutions 𝑢𝑡 . It is easy to see that 2
(1) 𝑧2
∗ + 𝑊02 (0) + ⋅⋅⋅ ,
𝑧̇(𝑡) = ⟨𝑞 , 𝑢̇
𝑡⟩ 2
= ⟨𝑞∗ , 𝐴 (0) 𝑢𝑡 + 𝑅 (0) 𝑢𝑡 ⟩ (2) 𝑧2 (2)
𝜑2 (0) = 𝑧𝛾 + 𝑧𝛾 + 𝑊20 (0) + 𝑊11 (0) 𝑧𝑧
2
= ⟨𝑞∗ , 𝐴 (0) 𝑢𝑡 ⟩ + ⟨𝑞∗ , 𝑅 (0) 𝑢𝑡 ⟩
(2) 𝑧2
∗ ∗ ∗
= ⟨𝐴 (0) 𝑞 , 𝑢𝑡 ⟩ + ⟨𝑞 , 𝑅 (0) 𝑢𝑡 ⟩ + 𝑊02 (0) + ⋅⋅⋅ ,
2
8 Abstract and Applied Analysis

𝑧2 Since 𝑞∗ (0) = 𝐷(1, 𝛾∗ ), we have

𝜑1 (−1) = 𝑧𝑒−𝑖𝜔0 𝜏0 + 𝑧𝑒𝑖𝜔0 𝜏0 + 𝑊20
(1)
(−1)
2
𝑔 (𝑧, 𝑧) = 𝑞∗ (0) 𝐹0 (𝑧, 𝑧)
(1) (1) 𝑧2
+ 𝑊11 (−1) 𝑧𝑧 + 𝑊02 (−1) + ⋅⋅⋅ ,
2 𝐾 𝑧2 + 𝐾12 𝑧𝑧 + 𝐾13 𝑧2 + 𝐾14 𝑧2 𝑧
= 𝐷 (1, 𝛾∗ ) ( 11 2 ) + ⋅⋅⋅
𝐾21 𝑧 + 𝐾22 𝑧𝑧 + 𝐾23 𝑧2 + 𝐾24 𝑧2 𝑧
𝑧2
𝜑2 (−1) = 𝑧𝛾𝑒−𝑖𝜔0 𝜏0 + 𝑧𝛾𝑒𝑖𝜔0 𝜏0 + 𝑊20
(2)
(−1)
2 = 𝐷 [(𝐾11 + 𝛾∗ 𝐾21 ) 𝑧2 + (𝐾12 + 𝛾∗ 𝐾22 ) 𝑧𝑧
𝑧2
+𝑊(2) (2)
11 (−1) 𝑧𝑧 + 𝑊02 (−1) + ⋅⋅⋅ . + (𝐾13 + 𝛾∗ 𝐾23 ) 𝑧2 + (𝐾14 + 𝛾∗ 𝐾24 ) 𝑧2 𝑧]
2
(78) + ⋅⋅⋅ .
From the definition of 𝐹(0, 𝑢𝑡 ), we have (81)
2 2 2
𝐾11 𝑧 + 𝐾12 𝑧𝑧 + 𝐾13 𝑧 + 𝐾14 𝑧 𝑧 Comparing the coefficients of the above equation with
𝐹0 (𝑧, 𝑧) = ( ) + ⋅⋅⋅ ,
𝐾21 𝑧2 + 𝐾22 𝑧𝑧 + 𝐾23 𝑧2 + 𝐾24 𝑧2 𝑧 those in (75), we have
(79)
𝑔20 = 2𝐷 (𝐾11 + 𝛾∗ 𝐾21 ) ,
where
𝑘11 = 𝜏0 (𝑚1 + 𝑚2 𝑒−𝑖𝜔0 𝜏0 + 𝑚3 𝛾 + 𝑚4 𝛾𝑒−𝑖𝜔0 𝜏0 ) , 𝑔11 = 𝐷 (𝐾12 + 𝛾∗ 𝐾22 ) ,
(82)
𝑘12 = 𝜏0 (2𝑚1 + 𝑚2 𝑒𝑖𝜔0 𝜏0 + 𝑚2 𝑒−𝑖𝜔0 𝜏0 + 𝑚3 𝛾 𝑔02 = 2𝐷 (𝐾13 + 𝛾∗ 𝐾23 ) ,

+ 𝑚3 𝛾 + 𝑚4 𝛾𝑒𝑖𝜔0 𝜏0 + 𝑚4 𝛾𝑒−𝑖𝜔0 𝜏0 ) , 𝑔21 = 2𝐷 (𝐾14 + 𝛾∗ 𝐾24 ) .

𝑘13 = 𝜏0 (𝑚1 + 𝑚2 𝑒𝑖𝜔0 𝜏0 + 𝑚3 𝛾 + 𝑚4 𝛾𝑒𝑖𝜔0 𝜏0 ) , In order to get the expression of 𝑔21 , we need to compute
𝑊20 (𝜃) and 𝑊11 (𝜃). Now, we determine the coefficients 𝑊𝑖𝑗 (𝜃)
(1) (1) in (73). By (72) and (57), we have
𝑘14 = 𝜏0 (𝑚1 (𝑊20 (0) + 2𝑊11 (0))

1 (1) 𝐴 (0) 𝑊 − 2 Re {𝑞∗ (0) 𝐹0 𝑞 (𝜃)} , −1 ≤ 𝜃 < 0,

+ 𝑚2 ( 𝑊20 (0) 𝑒𝑖𝜔0 𝜏0 + 𝑊11
(1)
(0) 𝑒−𝑖𝜔0 𝜏0 𝑊̇= {
2 𝐴 (0) 𝑊 − 2 Re {𝑞∗ (0) 𝐹0 𝑞 (𝜃)} + 𝐹0 , 𝜃 = 0.

(1) 1 (1) := 𝐴 (0) 𝑊 + 𝐻 (𝑧, 𝑧, 𝜃) ,

+ 𝑊11 (1) + 𝑊20 (1))
2 (83)
1 (1) (1)
+ 𝑚3 ( 𝛾𝑊20 (0) + 𝛾𝑊11 (0) 𝑧2 𝑧2
2 𝐻 (𝑧, 𝑧,𝜃) = 𝐻20 (𝜃) + 𝐻11 (𝜃) 𝑧𝑧 + 𝐻02 (𝜃) + ⋅⋅⋅ .
2 2
1 (2) (84)
+ 𝑊(2)
11 (0) + 𝑊20 (0))
2
From (73), (74), (83), and (84), we obtain
1 (2)
+ 𝑚4 ( 𝑊20 (0) 𝑒𝑖𝜔0 𝜏0 + 𝑊11
(2)
(0) 𝑒−𝑖𝜔0 𝜏0
2 (2𝑖𝜔0 𝜏0 − 𝐴 (0)) 𝑊20 (𝜃) = 𝐻20 (𝜃) ,
(1) 1 (1) 𝐴 (0) 𝑊11 (𝜃) = −𝐻11 (𝜃) , (85)
+ 𝛾𝑊11 (1) + 𝛾𝑊20 (1))
2
(𝐴 (0) + 2𝑖𝜔0 𝜏0 ) 𝑊11 (𝜃) = −𝐻02 (𝜃) , . . . .
+ 3𝑚5 + 𝑚6 (𝑒𝑖𝜔0 𝜏0 + 2𝑒−𝑖𝜔0 𝜏0 ) + 𝑚7 (𝛾 + 2𝛾)
From (75) and (83), for 𝜃 ∈ [−1, 0), we have
+𝑚8 (𝛾𝑒𝑖𝜔0 𝜏0 + 𝛾𝑒−𝑖𝜔0 𝜏0 + 𝛾𝑒−𝑖𝜔0 𝜏0 ) ) ,
𝐻 (𝑧, 𝑧, 𝜃) = −2 Re {𝑞∗ (0) 𝐹0 𝑞 (𝜃)}
𝑘21 = 𝜏0 𝑛1 𝑒−2𝑖𝜔0 𝜏0 , = −2 Re {𝑔 (𝑧, 𝑧) 𝑞 (𝜃)}
𝑘22 = 2𝜏0 𝑛1 , = −𝑔 (𝑧, 𝑧) 𝑞 (𝜃) − 𝑔 (𝑧, 𝑧) 𝑞 (𝜃)
𝑘23 = 𝜏0 𝑛1 𝑒2𝑖𝜔0 𝜏0 , 𝑧2 𝑧2 𝑧2 𝑧
= − (𝑔20 + 𝑔11 𝑧𝑧 + 𝑔02 + 𝑔21 ) 𝑞 (𝜃)
(1) 2 2 2
𝑘24 = 𝜏0 (𝑛1 (𝑊20 (1) 𝑒𝑖𝜔0 𝜏0 + 2𝑊11
(1)
(1) 𝑒−𝑖𝜔0 𝜏0 )
𝑧2 𝑧2 𝑧2 𝑧
+ 3𝑛2 𝑒−𝑖𝜔0 𝜏0 ) . − (𝑔20 + 𝑔11 𝑧𝑧 + 𝑔02 + 𝑔21 ) 𝑞 (𝜃) ⋅ ⋅ ⋅ .
2 2 2
(80) (86)
Abstract and Applied Analysis 9

Comparing the coefficients of the above equation with those Notice that
in (84), it follows that
0
𝐻20 (𝜃) = −𝑔20 𝑞 (𝜃) − 𝑔02 𝑞 (𝜃) , (𝑖𝜔0 𝜏0 𝐼 − ∫ 𝑑𝜂 (0, 𝜃) 𝑒𝑖𝜔0 𝜏0 𝜃 ) 𝑞 (0) = 𝑖𝜔0 𝜏0 𝑞 (0)
−1
𝐻11 (𝜃) = −𝑔11 𝑞 (𝜃) − 𝑔11 𝑞 (𝜃) , (87)
− 𝐴 (0) 𝑞 (0) = 0,
𝜃 ∈ [−1, 0) .
0

When 𝜃 = 0, we have (−𝑖𝜔0 𝜏0 𝐼 − ∫ 𝑑𝜂 (0, 𝜃) 𝑒−𝑖𝜔0 𝜏0 𝜃 ) 𝑞 (0) = 0.

−1
(94)
𝐻 (𝑧, 𝑧, 0) = −2 Re {𝑞∗ (0) 𝐹0 𝑞 (𝜃)} + 𝐹0

𝑧2 𝑧2 𝑧2 𝑧 Hence, we can obtain

= − (𝑔20 + 𝑔11 𝑧𝑧 + 𝑔02 + 𝑔21 ) 𝑞 (0)
2 2 2
0
𝐾
𝑧2 𝑧2 𝑧2 𝑧 (2𝑖𝜔0 𝜏0 𝐼 − ∫ 𝑑𝜂 (0, 𝜃) 𝑒2𝑖𝜔0 𝜏0 𝜃 ) 𝐸1 = 2 ( 11 ) . (95)
− (𝑔20 + 𝑔11 𝑧𝑧 + 𝑔02 + 𝑔21 ) 𝑞 (0) −1 𝐾21
2 2 2

+ 𝐹0 Similarly, we have

𝑧2 𝑧2 𝑧2 𝑧
= − (𝑔20 + 𝑔11 𝑧𝑧 + 𝑔02 + 𝑔21 ) 𝑞 (0) 0
𝐾
2 2 2 ∫ 𝑑𝜂 (0, 𝜃) 𝐸2 = − ( 12 ) . (96)
−1 𝐾22
2 2 2
𝑧 𝑧 𝑧𝑧
− (𝑔20 + 𝑔11 𝑧𝑧 + 𝑔02 + 𝑔21 ) 𝑞 (0)
2 2 2 Thus, we can get
𝐾 𝑧2 + 𝐾12 𝑧𝑧 + 𝐾13 𝑧2 + 𝐾14 𝑧2 𝑧
+ ( 11 2 ) + ⋅⋅⋅ .
𝐾21 𝑧 + 𝐾22 𝑧𝑧 + 𝐾23 𝑧2 + 𝐾24 𝑧2 𝑧 2𝑖𝜔 − 𝑎11 − 𝑎12 𝑒−2𝑖𝜔0 𝜏0 −𝑏11 𝐸1(1) 𝐾
𝜏0 [ 0 ][ ] = [ 12 ] ,
(88) −𝑎22 𝑒−2𝑖𝜔0 𝜏0 2𝑖𝜔0 𝐸1(2) 𝐾22
(97)
Comparing the coefficients of the above equation with those 𝑎 + 𝑎12 −𝑏11 𝐸2(1) −𝐾
𝜏0 [ 11 ] [ (2) ] = [ 12 ] .
in (84) gives that 𝑎22 0 𝐸2 −𝐾22
𝐻20 (0) = −𝑔20 𝑞 (0) − 𝑔02 𝑞 (0) + 2 (𝐾11 , 𝐾21 ) ,
(89) From (97), we can obtain
𝐻11 (0) = −𝑔11 𝑞 (0) − 𝑔11 𝑞 (0) + (𝐾12 , 𝐾22 ) .

From (85) and the definition of 𝐴(0), we have 𝐸1(1) = − (𝑏11 𝐾21 + 2𝜔0 𝐾11 )
𝑊̇20 (𝜃) = 2𝑖𝜔0 𝜏0 𝑊20 (𝜃) + 𝑔20 𝑞 (𝜃) + 𝑔02 𝑞 (𝜃) × (𝜏0 (− 4𝜔02 + 2𝜔0 𝑎11 + 2𝜔0 𝑎11 𝑒−2𝑖𝜔0 𝜏0
= 2𝑖𝜔0 𝜏0 𝑊20 (𝜃) + 𝑔20 𝑞 (0) 𝑒𝑖𝜔0 𝜏0 𝜃 + 𝑔02 𝑞 (0) 𝑒−𝑖𝜔0 𝜏0 𝜃 . −1
+ 𝑏11 𝑎22 𝑒−2𝑖𝜔0 𝜏0 )) ,
(90)

Hence, 𝐸1(2) = − (𝑎22 𝐾11 𝑒−2𝑖𝜔0 𝜏0 + 2𝐾21 𝜔0 − 𝑎11 𝐾21

𝑖𝑔20 𝑞 (0) 𝑖𝜔0 𝜏0 𝜃 𝑖𝑔02 𝑞 (0) −𝑖𝜔0 𝜏0 𝜃 − 𝑎12 𝐾21 𝑒−2𝑖𝜔0 𝜏0 )

𝑊20 (𝜃) = 𝑒 + 𝑒 + 𝐸1 𝑒2𝑖𝜔0 𝜏0 𝜃 , (98)
𝜔0 𝜏0 3𝜔0 𝜏0
(91) × (𝜏0 (− 4𝜔02 + 2𝜔0 𝑎11 + 2𝜔0 𝑎11 𝑒−2𝑖𝜔0 𝜏0
−1
and, by a similar method, we get + 𝑏11 𝑎22 𝑒−2𝑖𝜔0 𝜏0 )) ,
𝑖𝑔11 𝑞 (0) 𝑖𝜔0 𝜏0 𝜃 𝑖𝑔11 𝑞 (0) −𝑖𝜔0 𝜏0 𝜃 𝐾22
𝑊11 (𝜃) = − 𝑒 + 𝑒 + 𝐸2 , (92) 𝐸2(1) = ,
𝜔0 𝜏0 𝜔0 𝜏0 𝑎22 𝜏0
where 𝐸1 and 𝐸2 are both two-dimensional vectors. In the 𝑎22 𝐾12 − 𝑎11 𝐾22 − 𝑎12 𝐾22
following, we will find out 𝐸1 and 𝐸2 . From the definition of 𝐸2(2) = .
𝑎22 𝑏11 𝜏0
𝐴(0) and (85), we can obtain
0 Based on the above analysis, we can determine 𝑊20 (𝜃) and
∫ 𝑑𝜂 (0, 𝜃) 𝑊20 (𝜃) = 2𝑖𝜔0 𝜏0 𝑊20 (0) − 𝐻20 (0) . (93)
𝑊11 (𝜃) from (91) and (92). Furthermore, 𝑔21 in (82) can be
−1
10 Abstract and Applied Analysis

20.2 108.5

20.15 108

107.5
20.1
107
20.05
106.5
x(t)

q(t)
20
106
19.95
105.5
19.9 105

19.85 104.5
0 20 40 60 80 100 120 140 160 180 200 0 20 40 60 80 100 120 140 160 180 200
Time t Time t

Figure 1: Waveform plot of 𝑡-𝑥(𝑡) with 𝜏 = 0.200. Figure 2: Waveform plot of 𝑡-𝑞(𝑡) with 𝜏 = 0.200.

108.5
expressed by the parameters and delay. Thus, we can compute 108
the following quantities:
107.5
󵄨 󵄨2
𝑖 󵄨 󵄨2 󵄨󵄨𝑔 󵄨󵄨 𝑔 107
𝑐1 (0) = (𝑔20 𝑔11 − 2󵄨󵄨󵄨𝑔11 󵄨󵄨󵄨 − 󵄨 02 󵄨 ) + 21 ,
2𝜔0 𝜏0 3 2 106.5
q(t)

Re {𝑐1 (0)} 106

𝜇2 = − ,
Re {𝜆󸀠 (𝜏0 )} (99) 105.5

𝛽2 = 2 Re {𝑐1 (0)} , 105

104.5
Im {𝑐1 (0)} + 𝜇2 Im {𝜆󸀠 (𝜏0 )} 19.85 19.9 19.95 20 20.05 20.1 20.15 20.2
𝑇2 = − , x(t)
𝜔0 𝜏0
Figure 3: Phase plot for 𝑥(𝑡)-𝑞(𝑡) with 𝜏 = 0.200.
which determine the quantities of bifurcating periodic solu-
tions in the center manifold at the critical value 𝜏0 and we
have the following result. From (7), we have
Theorem 13. In (99), the following results hold: 1000
𝑥∗ = 50, 𝑞∗ = . (100)
1 + 220𝜏2 − 2𝜏
(i) the sign of 𝜇2 determines the directions of the Hopf By calculation, we obtain that 𝜔0 ≈ 3.082, 𝜏0 ≈ 0.203, and
bifurcation: if 𝜇2 > 0(< 0), then the Hopf bifurcation is 𝑟0 ≈ 0.472. It follows from (3.32) that
supercritical (subcritical) and the bifurcating periodic
solutions exist for 𝜏 > 𝜏0 (𝜏 < 𝜏0 ); 𝑐1 (0) = −0.000204 + 1.864768228𝑖,

(ii) the sign of 𝛽2 determines the stability of the bifurcating 𝜇2 = 0.000005120369502,

periodic solutions: the bifurcating periodic solutions are (101)
𝛽2 = −0.4074400488𝑒 − 4,
stable (unstable) if 𝛽2 < 0(𝛽2 > 0);
(iii) the sign of 𝑇2 determines the period of the bifurcating 𝑇2 = −2.980526844.
periodic solutions: the period increases (decreases) if These calculations prove that the system equilibrium
𝑇2 > 0 (𝑇2 < 0). (𝑥∗ , 𝑞∗ ) is asymptotically stable when 𝜏 < 𝜏0 by computer
simulation (see Figures 1, 2, and 3; 𝜏 = 0.200). When 𝜏 passes
4. Numerical Simulation Examples through the critical value 𝜏0 , (𝑥∗ , 𝑞∗ ) loses its stability and a
Hopf bifurcation occurs (see Figures 4, 5, and 6; 𝜏 = 0.206).
In this section, we use the formulas obtained in Sections 2
and 3 to verify the existence of the Hopf bifurcation and 5. Conclusion
calculate the Hopf bifurcation value and the direction of the
Hopf bifurcation of system (6) with 𝑐 = 1000, 𝑛 = 50, and A delayed model of congestion control was analyzed in
𝑘 = 0.001. this paper. Based on our theoretical analysis and numerical
Abstract and Applied Analysis 11

40 will be congested, even collapsed, when the communication

delay becomes large.
35

Conflict of Interests
30
The authors declare that there is no conflict of interests
x(t)

25 regarding the publication of this paper.

20
References
15 [1] S. H. Low, F. Paganini, J. Wang, and J. C. Doyle, “Linear stability
of TCP/RED and a scalable control,” Computer Networks, vol.
10 43, no. 5, pp. 633–647, 2003.
0 50 100 150 200 250 300
Time t [2] V. Misra, W.-B. Gong, and D. Towsley, “Fluid-based analysis of a
network of AQM routers supporting TCP flows with an applica-
Figure 4: Waveform plot of 𝑡-𝑥(𝑡) with 𝜏 = 0.206. tion to RED,” in Proceedings of the ACM SIGCOMM Computer
Communication Review, pp. 151–160, September 2000.
[3] R. Yang, P. Shi, and G. P. Liu, “Filtering for discrete-time net-
400
worked nonlinear systems with mixed random delays and pac-
350 ket dropouts,” IEEE Transactions on Automatic Control, vol. 56,
300 no. 11, pp. 2655–2660, 2011.
250 [4] H. C. Hsieh, J. S. Leu, and W. K. Shih, “Reaching consensus
underlying an autonomous local wireless sensor network,”
200 International Journal of Innovative Computing, Information and
q(t)

150 Control, vol. 6, no. 4, pp. 1905–1914, 2010.

100 [5] N. Hamadneh, D. Murray, M. Dixon, and P. Cole, “Weighted
50
RED (WTRED) strategy for TCP congestion control,” in Infor-
matics Engineering and Information Science, vol. 252 of Commu-
0 nications in Computer and Information Science, no. 2, pp. 421–
−50 434, Springer, Berlin, Germany, 2011.
−100 [6] F. S. Gentile, J. L. Moiola, and E. E. Paolini, “Nonlinear dynamics
0 50 100 150 200 250 300 of internet congestion control: a frequency-domain approach,”
Time t Communications in Nonlinear Science and Numerical Simula-
tion, vol. 19, no. 4, pp. 1113–1127, 2014.
Figure 5: Waveform plot of 𝑡-𝑞(𝑡) with 𝜏 = 0.206.
[7] S. Guo, H. Zheng, and Q. Liu, “Hopf bifurcation analysis for
400 congestion control with heterogeneous delays,” Nonlinear Anal-
350 ysis: Real World Applications, vol. 11, no. 4, pp. 3077–3090, 2010.
[8] D. Ding, J. Zhu, and X. Luo, “Hopf bifurcation analysis in a fluid
300
flow model of Internet congestion control algorithm,” Nonlinear
250 Analysis: Real World Applications, vol. 10, no. 2, pp. 824–839,
200 2009.
150 [9] Y. G. Zheng and Z. H. Wang, “Stability and Hopf bifurcation of a
q(t)

class of TCP/AQM networks,” Nonlinear Analysis: Real World

100
Applications, vol. 11, no. 3, pp. 1552–1559, 2010.
50
[10] S. Guo, X. Liao, Q. Liu, and C. Li, “Necessary and sufficient con-
0 ditions for Hopf bifurcation in exponential RED algorithm with
−50 communication delay,” Nonlinear Analysis: Real World Applica-
tions, vol. 9, no. 4, pp. 1768–1793, 2008.
−100
10 15 20 25 30 35 40 [11] H. Y. Yang and Y. P. Tian, “Hopf bifurcation in REM algorithm
x(t) with communication delay,” Chaos, Solitons and Fractals, vol. 25,
no. 5, pp. 1093–1105, 2005.
Figure 6: Phase plot for 𝑥(𝑡)-𝑞(𝑡) with 𝜏 = 0.206.
[12] C. V. Hollot, V. Misra, D. Towsley, and W. Gong, “Analysis and
design of controllers for AQM routers supporting TCP flows,”
IEEE Transactions on Automatic Control, vol. 47, no. 6, pp. 945–
simulation, we can find that there exists a critical value for 959, 2002.
this delay and the whole system is stable when the delay [13] D. W. Ding, X. M. Qin, N. Wang et al., “Hybrid control of Hopf
of the system is less than this critical value. By using the bifurcation in a dual model of Internet congestion control sys-
time delay as a bifurcation parameter, we have shown that a tem,” Nonlinear Dynamics, pp. 1–10, 2013.
Hopf bifurcation occurs when this parameter passes through [14] F. S. Gentile, J. L. Moiola, and E. E. Paolini, “Nonlinear dynamics
a critical value, which means that the wireless access system of internet congestion control: a frequency-domain approach,”
12 Abstract and Applied Analysis

Communications in Nonlinear Science and Numerical Simula-

tion, vol. 19, no. 4, pp. 1113–1127, 2014.
[15] F. Liu, G. Xiong, Z. H. Guan et al., “Stability analysis and control
Hopf bifurcation in a FAST TCP model,” in Proceedings of the
IEEE 32nd Chinese Control Conference (CCC ’13), pp. 1076–1080,
2013.
[16] M. Xiao, G. Jiang, and L. Zhao, “State feedback control at Hopf
bifurcation in an exponential RED algorithm model,” Nonlinear
Dynamics, pp. 1–16, 2014.
[17] Y. Kuang, Delay Differential Equations: With Applications in
Population Dynamics, Academic Press, New York, NY, USA,
1993.
[18] E. Beretta and Y. Kuang, “Geometric stability switch criteria in
delay differential systems with delay dependent parameters,”
SIAM Journal on Mathematical Analysis, vol. 33, no. 5, pp. 1144–
1165, 2002.
[19] R. S. Irving, Integers, Polynomials, and Rings: A Course in
Algebra, Springer, Berlin, Germany, 2004.
[20] B. D. Hassard, Theory and Applications of Hopf Bifurcation, CUP
Archive, 1981.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 484630, 6 pages
http://dx.doi.org/10.1155/2014/484630

Research Article
Dynamics of an Information Spreading Model with Isolation

Xia-Xia Zhao1,2 and Jian-Zhong Wang1,2

1
National Key Laboratory for Electronic Measurement Technology, North University of China, Taiyuan, Shanxi 030051, China
2
Key Laboratory of Instrumentation Science and Dynamic Measurement, Ministry of Education, North University of China,
Taiyuan, Shanxi 030051, China

Correspondence should be addressed to Xia-Xia Zhao; xiaxiazhao12@163.com

Received 20 March 2014; Accepted 4 April 2014; Published 23 April 2014

Academic Editor: Sanling Yuan

Copyright © 2014 X.-X. Zhao and J.-Z. Wang. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.

Information plays an important role in modern society. In this paper, we presented a mathematical model of information spreading
with isolation. It was found that such a model has rich dynamics including Hopf bifurcation. The results showed that, for a wide
range of parameters, there is a bistable phenomenon in the process of information spreading and thus the information cannot
be well controlled. Moreover, the model has a limit cycle which implies that the information exhibits periodic outbreak which is
consistent with the observations in the real world.

1. Introduction calculate the spread rate of information between nodes [11].

Narayanam and Narahari proposed a linear threshold model
In recent years, information plays a more and more important to describe information spreading [12]. Moreover, some
role in the real world. For a decision maker, if he does not have scholars used dynamical models [13] and statistical analysis
reasonable information, he will make a wrong decision which [14] to investigate the propagation of information.
may cause bad consequences. However, long time series of In this paper, we aim to present a mathematical model
information are hard to obtain and thus it may provide to describe the information spreading with isolation. What
useful information by constructing mathematical models. As is more, we want to reveal its dynamical behavior by both
a result, modeling the process of information spreading has mathematical analysis and numerical results. The paper is
been one of the central themes in the field of science [1, 2]. organized as follows. In Section 2, we construct an infor-
Information can be divided into many types. For some mation spreading model by using three ordinary differential
useful information such as advanced culture and useful equations (ODES). In Section 3, we show that this model
knowledge, it is encouraged to develop both widely and has rich dynamics. Finally, conclusions and discussions are
deeply. However, there is also some bad information and the presented in Section 4.
typical one is rumor [3–8]. Rumor is a social phenomenon
accompanying the development of society which may bring
some corrosive social effects. For example, in 2007, some 2. A Mathematical Model
people circulated false news that the bananas from Hainan To well describe the model, we firstly give four main assump-
were toxic which caused huge economic loss [9]. In this tions, which are as follows.
sense, we need to isolate some information by taking control
measures. (i) There are three kinds of individuals: the population
There is some work on information spreading by using who has no information (𝑋), the population who has
mathematical models. Based on SIS models, Leskovec et al. information (𝑌), and the population who is isolated
presented a cascading information model and found how (𝑍) by isolation.
blogs behaved and how information propagated through (ii) There are birth and death events and all the newborn
the blogosphere [10]. Gruhl et al. posed a SIRS model to has no information.
2 Abstract and Applied Analysis

2 3. Dynamical Behaviors
The first step in analyzing the model (4a) and (4b) is to deter-
1.5 mine equilibria points by solving the following equations:
𝑎𝑋𝑌
𝐴 − 𝑑𝑋 − = 0, (5a)
1 + 𝑏𝑌
g(Y)

1
𝑎𝑋𝑌 𝑐 +𝑐 𝑌
− 𝑑𝑌 − 1 2 𝑌 = 0. (5b)
1 + 𝑏𝑌 1+𝑌
0.5
System (4a) and (4b) always has an equilibrium 𝐸0 =
(𝐴/𝑑, 0) corresponding to the case that there is no informa-
0
tion. The Jacobian matrix of the linearization of system (4a)
0 5 10 15 20 and (4b) at this equilibrium is [15]
Y
𝐴
−𝑑 −𝑎
Figure 1: Plot of 𝑔(𝑌) as a function of 𝑌 with 𝑎 = 2 and 𝑏 = 1 (color J=( 𝑑 ), (6)
online). 𝐴
0 𝑎 − (𝑑 + 𝑐1 )
𝑑
which has negative eigenvalues, implying the asymptotic
(iii) The spread rate of information is stability of this equilibrium if and only if
𝐴𝑎 < 𝑑 (𝑑 + 𝑐1 ) . (7)
𝑎𝑌
𝑔 (𝑌) = , (1)
(1 + 𝑏𝑌) Denote
𝐴𝑎
which has a saturation effect (cf. Figure 1). 𝑆1 = . (8)
𝑑 (𝑑 + 𝑐1 )

(iv) Since information spreading has stages, isolation term If 𝑆1 < 1, then there is no information; otherwise the
may have different forms at different stages of infor- information will spread.
mation spreading. We choose isolation term having In order to obtain positive solutions of (5a) and (5b), we
the following form: eliminate 𝑋 using the first equation of (5a) and (5b) and
substitute it into the second equation to give an equation of
𝑐1 + 𝑐2 𝑌 the form
𝐼 (𝑌) = 𝑌. (2)
1+𝑌 𝑃1 𝑌2 + 𝑃2 𝑌 + 𝑃3 = 0, (9)
where
For the small number of 𝑌, 𝐼(𝑌) → 𝑐1 𝑌 and 𝐼(𝑌) → 𝑐2 𝑌
for large number of 𝑌. 𝑃1 = 𝑏𝑑2 + 𝑏𝑑𝑐2 + 𝑎𝑑 + 𝑎𝑐2 ,
On the basis of the above assumptions, we have the
following model: 𝑃2 = 𝑏𝑑𝑐1 + 𝑏𝑑2 + 𝑎𝑑 + 𝑑2 + 𝑎𝑐1 − 𝐴𝑎 + 𝑑𝑐2 , (10)

𝑃3 = 𝑑2 + 𝑑𝑐1 − 𝐴𝑎.
𝑑𝑋
= 𝐴 − 𝑑𝑋 − 𝑔 (𝑌) 𝑋, (3a)
𝑑𝑡 It should be noted that 𝑃3 > 0 (< 0 or = 0) if and only
𝑑𝑌 if 𝑆1 > 1 (< 1 or = 1). That is to say, there is a unique
= 𝑔 (𝑌) 𝑋 − 𝑑𝑌 − 𝐼 (𝑌) , (3b) nonzero solution of (9) denoted as 𝑌 = −𝑃2 /𝑃1 if 𝑆1 = 1
𝑑𝑡 and it is positive if and only if 𝑃2 < 0. Since equilibria depend
𝑑𝑍 continuously on 𝑆1 , there is an interval to the left of 𝑆1 = 1 on
= 𝐼 (𝑌) − 𝑑𝑍, (3c)
which there are two positive equilibria and
𝑑𝑡

where 𝐴 is the birth rate of the population and 𝑑 is the natural −𝑃2 ± √𝑃22 − 4𝑃3
𝑌1,2 = . (11)
death rate of the population. Since the first two equations 2𝑃1
in (3a), (3b), and (3c) are independent of the variable 𝑍, it
suffices to consider the following reduced model: As a result, we can conclude that system (4a) and (4b) has
a backward bifurcation at 𝑆1 = 1 if and only if 𝑃2 < 0 [16, 17].
𝑑𝑋 𝑎𝑋𝑌 We can also give an explicit criterion of 𝑐2 in terms of
= 𝐴 − 𝑑𝑋 − = ℎ1 (𝑋, 𝑌) , (4a) other parameters for the existence of a backward bifurcation
𝑑𝑡 1 + 𝑏𝑌
at 𝑆1 = 1 equals 𝑑2 + 𝑑𝑐1 = 𝐴𝑎. 𝑃2 < 0 implies that
𝑑𝑌 𝑎𝑋𝑌 𝑐 +𝑐 𝑌
= − 𝑑𝑌 − 1 2 𝑌 = ℎ2 (𝑋, 𝑌) . (4b) 𝑏𝑑𝑐1 + 𝑏𝑑2 + 𝑎𝑑 + 𝑑2 + 𝑎𝑐1 + 𝑑𝑐2 < 𝐴𝑎, (12)
𝑑𝑡 1 + 𝑏𝑌 1+𝑌
Abstract and Applied Analysis 3

and thus 80
70
𝑏𝑑𝑐1 + 𝑏𝑑2 + 𝑎𝑑 + 𝑑2 + 𝑎𝑐1 + 𝑑𝑐2 < 𝑑2 + 𝑑𝑐1 , (13) 60

which equals 50
Y 40
𝑑𝑐1 − 𝑏𝑑𝑐1 − 𝑏𝑑2 − 𝑎𝑑 − 𝑎𝑐1 30
𝑐2 < ≜ 𝑐2∗ . (14)
𝑑 20 Rc
10
If (14) is satisfied, there is a backward bifurcation at 𝑆1 =
1. Then, there are two positive equilibria for an interval of 0
0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4
values of 𝑆 from a value 𝑆2 to 𝑆1 = 1. To calculate 𝑆2 , we use R0
𝑃22 − 4𝑃1 𝑃3 = 0 to get a quadratic equation of parameter 𝐴,
which has the following form: Figure 2: The sizes of population 𝑌 at equilibria versus 𝑆1 with 𝑑 =
0.1, 𝑎 = 0.001, 𝑏 = 0.01, 𝑐1 = 1, and 𝑐2 = 0.5 (color online).
𝑎2 𝐴2

+ 2𝐴𝑎 (𝑏𝑑2 + 2𝑏𝑑𝑐2 + 𝑎𝑑 + 2𝑎𝑐2 − 𝑏𝑑𝑐1 − 𝑑2 − 𝑎𝑐1 − 𝑑𝑐2 ) has a unique positive equilibrium if 𝑆1 > 1. Furthermore,
2 system (4a) and (4b) has two positive equilibria 𝐸1 and 𝐸2
+ (−𝑑𝑐2 + 𝑏𝑑𝑐1 − 𝑑2 + 𝑎𝑑 + 𝑎𝑐1 + 𝑏𝑑2 ) = 0. if 𝑆2 < 𝑆1 < 1 and 𝑎 + 𝑏𝑑 < 𝑑.
(15) In the following, we aim to investigate the global stability
of the equilibrium by considering two cases: (i) 𝑐1 ≤ 𝑐2 and
Note that 𝑃2 < 0 and 𝑆1 < 1, which equals (ii) 𝑐1 > 𝑐2 .
For the first case, we can obtain the Jacobian matrix of
𝑏𝑑𝑐1 + 𝑏𝑑2 + 𝑎𝑑 + 𝑑2 + 𝑎𝑐1 + 𝑑𝑐2 𝑑 (𝑑 + 𝑐1 ) system (4a) and (4b) at equilibrium (𝑋, 𝑌) is
<𝐴< . (16)
𝑎 𝑎
𝑎𝑌 𝑎𝑋
In that case, we have that −𝑑 − −
1 + 𝑏𝑌 (1 + 𝑏𝑌)2
𝐽=( ).
𝑎𝑌 𝑎𝑋 𝑐1 + 2𝑐2 𝑌 + 𝑐2 𝑌2
𝐴 𝑐 = (𝑑2 − 𝑎𝑑 − 2𝑎𝑐2 + 𝑎𝑐1 − 𝑏𝑑2 + 𝑑𝑐2 − 2𝑏𝑑𝑐2 − 𝑑 −
1 + 𝑏𝑌 (1 + 𝑏𝑌)2 (1 + 𝑌)2
(20)
+𝑏𝑑𝑐1 + 2√(𝑎 − 𝑑 + 𝑏𝑑) (𝑎 + 𝑏𝑑) (𝑑 + 𝑐2 ) (𝑐2 − 𝑐1 ))
By direct calculations, we can easily obtain det 𝐽(𝐸0 ) = −𝐴𝑎+
−1
×𝑎 . 𝑑(𝑑 + 𝑐1 ) > 0, tr 𝐽(𝐸0 ) = 𝑎𝐴/𝑑 − 2𝑑 − 𝑐1 < 0 if 𝑆1 < 1. And
(17) hence all the eigenvalues of characteristic equation of 𝐸0 have
negative real parts, which means that the equilibrium 𝐸0 is
As a result, the expression of 𝑆2 locally asymptotically stable.
Let
𝐴 𝑐𝑎
𝑆2 = . (18) 𝐴
𝑑 (𝑑 + 𝑐1 ) Ω = {(𝑋, 𝑌) | 𝑋, 𝑌 ≥ 0, 𝑋 + 𝑌 ≤ }. (21)
𝑑
We fix 𝑑 = 0.1, 𝑎 = 0.001, 𝑏 = 0.01, 𝑐1 = 1, and
By (4a) and (4b), we have
𝑐2 = 0.5 and show that system (4a) and (4b) has a backward
bifurcation in Figure 2. 𝑑𝑋 𝑑𝑌
In the following, we discuss the case when 𝑆1 < 1. If | = 𝐴 > 0, | = 0. (22)
𝑑𝑡 𝐸0 𝑑𝑡 𝐸0
system (4a) and (4b) has a positive solution, the following
condition must be satisfied: Denote 𝑁1 (𝑡) = 𝑋(𝑡) + 𝑌(𝑡); then

𝑃2 < 0, Δ ≥ 0. (19) 𝑑𝑁1

| ≤ 𝐴 − 𝑑𝑁1 |𝑁1 =𝐴/𝑑 = 0. (23)
𝑑𝑡 𝑁1 =𝐴/𝑑
2 2
Note that 𝑃2 < 0 equals 𝑏𝑑𝑐1 + 𝑏𝑑 + 𝑎𝑑 + 𝑑 + 𝑎𝑐1 + 𝑑𝑐2 <
𝐴𝑎. And 𝑆1 < 1 equals 𝐴𝑎 < 𝑑(𝑑 + 𝑐1 ). As a result, we have Hence, Ω is a positively invariant set of (4a) and (4b)
𝑏𝑑𝑐1 + 𝑏𝑑2 + 𝑎𝑑 + 𝑑2 + 𝑎𝑐1 + 𝑑𝑐2 < 𝑑(𝑑 + 𝑐1 ) which leads to and attracts all the positive orbits of (4a) and (4b) state
𝑎 + 𝑏𝑑 < 𝑑. in 𝑅2+ , which means that the equilibrium 𝐸0 is globally
The discussions above yield the following results. asymptotically stable if 𝑆1 < 1.
(I) For the case 𝑐1 ≤ 𝑐2 , 𝐸0 is always a boundary equilib- It is easy to obtain that 𝐸∗ is locally stable when 𝑆1 > 1.
rium of system (4a) and (4b) and 𝐸∗ is the unique positive Then, we use Dulac theorem to exclude the limit cycle [18].
equilibria of system (4a) and (4b) if and only if 𝑆1 > 1. Take the Dulac function
(II) For the case 𝑐1 > 𝑐2 , 𝐸0 is always a boundary equi- 1
librium of system (4a) and (4b) and system (4a) and (4b) 𝐷= . (24)
𝑋𝑌
4 Abstract and Applied Analysis

10 30

25 E∗
8

20
6
Y Y 15
4
10
2
5
0 E0
0 E0
70 75 80 85 90 95
700 750 800 850 900 950
X
X
Figure 3: Phase plane of 𝑋 − 𝑌 with 𝐴 = 9, 𝑑 = 0.1, 𝑎 = 0.001,
𝑏 = 0.01, 𝑐1 = 0.1, and 𝑐2 = 0.5 (color online). Figure 4: Phase plane of 𝑋 − 𝑌 with 𝐴 = 90, 𝑑 = 0.1, 𝑎 = 0.001,
𝑏 = 0.01, 𝑐1 = 0.1, and 𝑐2 = 0.5 (color online).

Then, we have
20
𝜕 (ℎ1 𝐷) 𝜕 (ℎ2 𝐷) 𝐴 𝑎𝑏 𝑐 −𝑐
+ =− 2 − 2
− 1 2 2 < 0,
𝜕𝑋 𝜕𝑌 𝑋 𝑌 (1 + 𝑌) 𝑋(1 + 𝑌) 15
(25) E2

which means that system (4a) and (4b) has no limit cycles and Y 10
the positive equilibrium is globally asymptotically stable.
As a result, we have the following conclusions.
5
(i) The equilibrium 𝐸0 is globally asymptotically stable if
𝑆1 < 1 and unstable if 𝑆1 > 1. E1
0 E0
(ii) When 𝑆1 > 1, the unique endemic equilibrium is
globally asymptotically stable. 700 750 800 850
X
In Figure 3, we fixed 𝐴 = 9, 𝑑 = 0.1, 𝑎 = 0.001, 𝑏 = 0.01,
𝑐1 = 0.1, and 𝑐2 = 0.5 which satisfies 𝑆1 < 1. It can be found Figure 5: Phase plane of 𝑋 − 𝑌 with 𝐴 = 80, 𝑑 = 0.1, 𝑎 = 0.001,
from this figure that 𝐸0 is globally asymptotically stable. 𝑏 = 0.01, 𝑐1 = 1, and 𝑐2 = 0.5 (color online).
In Figure 4, we fixed 𝐴 = 90, 𝑑 = 0.1, 𝑎 = 0.001, 𝑏 =
0.01, 𝑐1 = 0.1, and 𝑐2 = 0.5 which satisfies 𝑆1 > 1. It can be
found from this figure that 𝐸0 is not stable and the unique By calculations, one can find that det 𝐽𝐸𝑐 = tr 𝐽𝐸𝑐 = 0,
endemic equilibrium is globally asymptotically stable. which means that Jacobian matrix has a zero eigenvalue with
For the second case 𝑐1 > 𝑐2 , we need to discuss three multiplicity 2. In other words, the positive equilibrium 𝐸𝑐 of
situations: system (4a) and (4b) is a cusp of codimension 2 [19].
(I) 𝑆1 < 𝑆2 ; If 𝑆2 < 𝑆1 < 1, then system (4a) and (4b) exhibits multiple
positive equilibria. In such a case, it has been shown that a
(II) 𝑆1 = 𝑆2 ; stable positive equilibrium coexists with the stable boundary
(III) 𝑆2 < 𝑆1 < 1; equilibrium (𝐸0 ), which is named as bistable phenomenon.
(IV) 𝑆1 > 1. In Figure 5, we fixed 𝐴 = 80, 𝑑 = 0.1, 𝑎 = 0.001, 𝑏 = 0.01,
𝑐1 = 1, and 𝑐2 = 0.5 which satisfies 𝑆2 < 𝑆1 < 1. We can
Since dynamical behaviors of system (4a) and (4b) in see from this figure that 𝐸0 and 𝐸2 are stable. In this case, the
situation (I) are simple, we will investigate the later three information cannot be well controlled.
cases. We assume that 𝑆1 > 1. If tr(J) < 0, then the positive
If 𝑆1 = 𝑆2 , system (4a) and (4b) has a unique positive equilibrium 𝐸∗ of system (4a) and (4b) is a stable node or
equilibrium 𝐸𝑐 (𝑋𝑐 , 𝑌𝑐 ), where focus; if tr(J) > 0, then 𝐸∗ is unstable and system (4a) and
(4b) has at least one limit cycle; if tr(J) = 0, then 𝐸∗ is a center.
𝑑 𝑐 +𝑐 𝑌 In Figure 6, we fixed 𝐴 = 150, 𝑑 = 0.1, 𝑎 = 0.001, 𝑏 =
𝑋𝑐 = [ + 1 2 𝑐 ] (1 + 𝑏𝑌𝑐 ) ,
𝑎 𝑎 (1 + 𝑌𝑐 ) 0.01, 𝑐1 = 1, and 𝑐2 = 0.5 which satisfies 𝑆1 > 1. We can see
(26) from this figure that 𝐸∗ is unstable and system (4a) and (4b)
𝑏𝑑𝑐 + 𝑏𝑑2 + 𝑎𝑑 + 𝑑2 + 𝑎𝑐1 − 𝐴𝑎 + 𝑑𝑐2 has a stable limit cycle in which Hopf bifurcation occurs in
𝑌𝑐 = − 1 .
2 (𝑏𝑑2 + 𝑏𝑑𝑐2 + 𝑎𝑑 + 𝑎𝑐2 ) system (4a) and (4b). These results show that the information
Abstract and Applied Analysis 5

50 [4] D. H. Zanette, “Critical behavior of propagation on small-world

45
networks,” Physical Review E. Statistical, Nonlinear, and Soft
Matter Physics, vol. 64, no. 5, Article ID 050901, 2001.
40
[5] L. Lü, D.-B. Chen, and T. Zhou, “The small world yields the most
35 effective information spreading,” New Journal of Physics, vol. 13,
30 Article ID 123005, 2011.
Y 25 [6] A. L. Hill, D. G. Rand, M. A. Nowak, and N. A. Christakis,
20 “Infectious disease modeling of social contagion in networks,”
PLoS Computational Biology, vol. 6, no. 11, Article ID e1000968,
15
2010.
10
E∗ [7] T. House, “Modelling behavioural contagion,” Journal of the
5 Royal Society Interface, vol. 8, no. 59, pp. 909–912, 2011.
0 [8] Z.-Q. Zhu, C.-J. Liu, J.-L. Wu, J. Xu, and B. Liu, “The influence
650 700 750 800 850 900 of human heterogeneity to information spreading,” Journal of
X Statistical Physics, vol. 154, no. 6, pp. 1569–1577, 2014.
[9] http://www.iyaxin.com/content/2008-10/21/content 450073
Figure 6: Phase plane of 𝑋 − 𝑌 with 𝐴 = 150, 𝑑 = 0.1, 𝑎 = 0.001, .htm.
𝑏 = 0.01, 𝑐1 = 1, and 𝑐2 = 0.5 (color online).
[10] J. Leskovec, M. McGlohon, C. Faloutsos, N. Glance, and M.
Hurst, “Patterns of cascading behavior in large blog graphs,” in
Proceedings of the 7th SIAM International Conference on Data
exhibits periodic outbreak which is consistent with the real Mining, pp. 551–556, April 2007.
data [20]. [11] D. Gruhl, D. Liben-Nowell, R. Guha, and A. Tomkins, “Infor-
mation diffusion through blogspace,” in Proceedings of the 13th
International World Wide Web Conference Proceedings (WWW
4. Discussion and Conclusion ’04), pp. 491–501, May 2004.
[12] R. Narayanam and Y. Narahari, “A shapley value-based
Information has a great effect on social network, which is a approach to discover influential nodes in social networks,” IEEE
double-edged sword. In this paper, we investigated an infor- Transactions on Automation Science and Engineering, vol. 8, no.
mation spreading model with isolation. For the dynamical 1, pp. 130–147, 2011.
model, we obtained the global stability of the equilibria and [13] M. Lahiri and M. Cebrian, “The genetic algorithm as a general
bifurcation behaviors, based on both mathematical analysis diffusion model for social networks,” in Proceedings of the 24th
and numerical results. In a word, information spreading AAAI Conference on Artificial Intelligence, pp. 494–499, July
model can have rich dynamics which may provide some new 2010.
insights for policy decisions on information. [14] B. Suh, L. Hong, P. Pirolli, and E. H. Chi, “Want to be retweeted?
It should be noted that our results are based on the Large scale analytics on factors impacting retweet in twitter net-
assumption of a homogeneously mixing population. That is work,” in Proceedings of the 2nd IEEE International Conference
to say, we assume that contacts between different individuals on Social Computing (SocialCom ’10), pp. 177–184, August 2010.
are in the same form. This assumption ignores individual [15] J. Guckenheimer and P. Holmes, Nonlinear Oscillations, Dynam-
differences and thus is not reasonable [21–24]. As a result, we ical Systems, and Bifurcations of Vector Fields, vol. 42, Springer,
need to investigate the information spreading model in social New York, NY, USA, 1990.
network in further research. [16] J. Arino, C. C. McCluskey, and P. van den Driessche, “Global
results for an epidemic model with vaccination that exhibits
backward bifurcation,” SIAM Journal on Applied Mathematics,
Conflict of Interests vol. 64, no. 1, pp. 260–276, 2003.
[17] J. Dushoff, W. Huang, and C. Castillo-Chavez, “Backwards
The authors declare that there is no conflict of interests
bifurcations and catastrophe in simple models of fatal diseases,”
regarding the publication of this paper. Journal of Mathematical Biology, vol. 36, no. 3, pp. 227–248,
1998.
References [18] L. Perko, Differential Equations and Dynamical Systems, vol. 7,
Springer, New York, NY, USA, 2nd edition, 1996.
[1] L. K. Gallos, R. Cohen, P. Argyrakis, A. Bunde, and S. Havlin, [19] Y. A. Kuznetsov, Elements of Applied Bifurcation Theory, vol. 112,
“Stability and topology of scale-free networks under attack Springer, New York, NY, USA, 1995.
and defense strategies,” Physical Review Letters, vol. 94, no. 18,
[20] J. V. Carlis and J. A. Konstan, “Interactive visualization of serial
Article ID 188701, 2005.
periodic data,” in Proceedings of the 11th Annual ACM Sympo-
[2] Y. Moreno, M. Nekovee, and A. F. Pacheco, “Dynamics of rumor sium on User Interface Software and Technology (UIST ’98), pp.
spreading in complex networks,” Physical Review E. Statistical, 29–38, November 1998.
Nonlinear, and Soft Matter Physics, vol. 69, no. 6, Article ID [21] M. Yang, G. Chen, and X. Fu, “A modified SIS model with an
066130, p. 1, 2004. infective medium on complex networks and its global stability,”
[3] D. J. Daley and D. G. Kendall, “Epidemics and Rumours,” Physica A. Statistical Mechanics and its Applications, vol. 390, no.
Nature, vol. 204, no. 4963, article 1118, 1964. 12, pp. 2408–2413, 2011.
6 Abstract and Applied Analysis

[22] A.-L. Barabási and R. Albert, “Emergence of scaling in random

networks,” Science, vol. 286, no. 5439, pp. 509–512, 1999.
[23] A.-L. Barabási, R. Albert, and H. Jeong, “Mean-field theory for
scale-free random networks,” Physica A. Statistical Mechanics
and its Applications, vol. 272, no. 1, pp. 173–187, 1999.
[24] R. Pastor-Satorras and A. Vespignani, “Epidemic spreading in
scale-free networks,” Physical Review Letters, vol. 86, no. 14, pp.
3200–3203, 2001.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 938080, 8 pages
http://dx.doi.org/10.1155/2014/938080

Research Article
Modelling the Influence of Awareness Programs by Media on
the Drinking Dynamics

Hai-Feng Huo and Qian Wang

Department of Applied Mathematics, Lanzhou University of Technology, Lanzhou, Gansu 730050, China

Correspondence should be addressed to Hai-Feng Huo; huohf1970@gmail.com

Received 25 January 2014; Accepted 17 March 2014; Published 16 April 2014

Academic Editor: Yongli Song

Copyright © 2014 H.-F. Huo and Q. Wang. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.

We develop a nonlinear mathematical model with the effect of awareness programs on the binge drinking. Due to the fact that
awareness programs are capable of inducing behavioral changes in nondrinkers, we introduce a separate class by avoiding contacts
with the heavy drinkers. Furthermore we assume that cumulative density of awareness programs increases at a rate proportional to
the number of heavy drinkers. We establish some sufficient conditions for the stability of the alcohol free and the alcohol present
equilibria and give some numerical simulations to explain our main result. Our results show that awareness programs is an effective
measure in reducing alcohol problems.

1. Introduction alcohol, especially by young people, leads to serious antisocial

and criminal behavior in urban centres. This phenomenon
Alcohol consumption, particularly heavier drinking, is an has grown very rapidly [8]. Since mathematical model is a
important risk factor for many health problems and is a major predictive tool which can mimic the process of drinking and
contributor to the global burden of disease [1]. It has been provide useful tools to analyze the spread and control of
identified as an important risk factor for chronic disease and drinking behavior, several different mathematical models for
injury [2]. Drinking alcohol causes the liver to become even drinking have been formulated and studied recently. Simple
more damaged in people infected with the hepatitis B virus models for alcohol treatment are presented by Sanchez et
(HBV) [3]. The CAS (Central Authentication Service) find- al. [9] which were based on studying binge drinking in a
ings have shown that alcohol consumption at binge levels and college system and assumed the same “leaving rate.” Mulone
beyond has a significant impact on college students’ academic and Straughan [10] investigated a possible model for binge
performance, social relationships, risk taking behaviors, and drinking taking into account admitting and nonadmitting
health. This form of drinking is associated with missing class, drinkers. But the global analysis of binge drinking model is
falling behind in schoolwork, and lower grade point average, not discussed in the literature. Huo and Song [11] introduced
a relationship mediated by fewer hours spent in studying [4, a more realistic two-stage model for binge drinking problem,
5]. In addition, students who attended schools with high rates where the youths with alcohol problems are divided into
of binge drinking experienced a greater number of second- those who admit the problem and those who do not admit it.
hand effects, including disruption of sleep or study, property For the other mathematical models for drinking or smoking,
damage, and verbal, physical, or sexual violence, than their we refer to [12–14] and the references cited therein.
peers attending schools with low binge drinking rates [6]. The media may be the most important source of health
Prior studies have indicated that heavy alcohol drinkers are information for the general public; it can play a special role
likely to engage in risky sexual behaviours and, thus, more in providing a voice for people to express their experiences
likely to get sexually transmitted infections (STIs) than social of illness [15]. Jaramillo [16] studied a model presenting an
drinkers [7]. The rapid consumption of large amounts of evaluation of the impact on case finding of a mass media
2 Abstract and Applied Analysis

health education campaign for TB control in Cali, Colombia. 𝜇 𝛽SA pA

Liu et al. [17] incorporate the psychological impacts of S A R
qR
media coverage on disease transmission and emphasize the
importance of refining classical mathematical models to 𝜇S 𝜇A 𝜇R
reflect the partially self-limiting nature of infectious disease 𝜎X
𝜆SM
outbreaks, due to the effects of widespread news coverage
and fast information flow. Web-based screening and brief mA
interventions that include personalized feedback about their X M
alcohol use have proven to be particularly promising for
reducing hazardous drinking among university students [18]. 𝜇X 𝛾M
Mass media (television, radio, newspapers, billboards, and
booklets) have been used as a way of delivering preventive Figure 1: Transfer diagram of model (2).
health messages as they have the potential to influence
people’s behavior. Further, mass media have been deployed
in the effort to control and eliminate epidemic diseases [19]. 𝑋(𝑡), those who drink heavily, denoted by 𝐴(𝑡), and those
Xiang et al. [20] studied a drinking model with public health who are in treatment, denoted by 𝑅(𝑡). Also let 𝑀(𝑡) be
educational campaigns which can be an efficient option the cumulative density of awareness programs driven by
for reducing the spread of disease. Misra et al. [21, 22] media. The growth rate of density of awareness programs
considered the effects of awareness programs driven by media is assumed to be proportional to the number of heavy
on the spread of infectious diseases; their results showed drinkers. It is considered that, due to the awareness programs,
that, though awareness programs cannot eradicate infection, nondrinkers form a different class and avoid contacting with
they help in controlling the prevalence of disease. The most the heavy drinkers. And we only consider the interaction
successful approaches concentrate not on trying to change the from awareness programs between aware individual and
drinker’s behavior directly but on influencing the community nondrinker. The total number of population at time 𝑡 is given
environment around the nondrinker. For instance, education by
and persuasion may increase knowledge and change attitudes
but have no effect on drinking [23, 24], so we only consider 𝑁 (𝑡) = 𝑆 (𝑡) + 𝑋 (𝑡) + 𝐴 (𝑡) + 𝑅 (𝑡) . (1)
the interaction from awareness programs between aware
individual and nondrinker. The model structure is shown in Figure 1. The transfer
Motivated by the above works, we introduce a mathe- diagram leads to the following system of ordinary differential
matical model with the effect of awareness programs on the equations:
binge drinking in this paper. Because awareness programs
are capable of inducing behavioral changes in susceptible 𝑑𝑆 (𝑡)
= 𝜇 − 𝛽𝑆𝐴 − 𝜆𝑆𝑀 + 𝜎𝑋 − 𝜇𝑆,
individuals, we introduce a separate class 𝑋(𝑡) of those who 𝑑𝑡
are aware of risk and refrain from drinking by avoiding 𝑑𝑋 (𝑡)
contact with the heavy drinkers. This newly formed aware = 𝜆𝑆𝑀 − 𝜎𝑋 − 𝜇𝑋,
class is assumed to be fully protected from the heavy drinkers 𝑑𝑡
due to media-induced isolation and individuals of this class 𝑑𝐴 (𝑡)
may contact heavy drinkers only if they lose awareness = 𝛽𝑆𝐴 + 𝑞𝑅 − 𝑝𝐴 − 𝜇𝐴, (2)
𝑑𝑡
[21, 22]. Furthermore, we assume that cumulative density of
awareness programs 𝑀(𝑡) increases at a rate proportional to 𝑑𝑅 (𝑡)
= 𝑝𝐴 − 𝑞𝑅 − 𝜇𝑅,
the number of heavy drinkers. We establish some sufficient 𝑑𝑡
conditions for the global stability of equilibria and give some 𝑑𝑀 (𝑡)
numerical simulations to explain our main result. Our results = 𝑚𝐴 − 𝛾𝑀,
𝑑𝑡
show that awareness programs are an effective measure in
reducing alcohol problems. where 𝜇 is the rate of individuals entering into the system in
The organization of this paper is as follows. In the next a given time interval (say, each year). Since we are dealing
section, the effect of awareness programs by media on binge with youths, we assume that the death rate is negligible and
drinking model is formulated. In Section 3, the existence and so the leaving rate is also 𝜇. This is in accordance with
the stability of equilibria are investigated. Some numerical an assumption by Sanchez et al. [9], where “leaving rate”
simulations are given in Section 4. Some discussions are given presents the departure rate from drinking environment. 𝛽
in Section 5. is the transmission coefficient for the nondrinkers turning
to heavy drinkers through peer pressure. 𝜆 represents the
2. The Model dissemination rate of awareness among nondrinkersdue to
the awareness programs, nondrinkers form a different class
2.1. System Description. The total population is divided into 𝑋(𝑡) and avoid contacting with the heavy drinkers. The
five compartments, namely, the susceptible compartment constant 𝜎 denotes the rate of transfer of aware individuals
who do not drink or drink only moderately, denoted by 𝑆(𝑡), to nondrinker class. 𝑝 is a proportionality coefficient for 𝐴
those who are aware of risk and avoid drinking, denoted by going into treatment. 𝑞 is a proportionality coefficient for
Abstract and Applied Analysis 3

𝑅 relapsing into drinkers. 𝑚 represents the growth rate of is a positively invariant set for (2). So we consider dynamics
density of awareness programs which is assumed to be pro- of system (2) on the set Ω in this paper.
portional to the number of heavy drinkers, and 𝛾 represents
the depletion rate of these programs due to ineffectiveness, 2.2.2. Positivity of Solutions. For system (2), to ensure that the
social problems, and so forth. For instance, as time passes by, solutions of the system with positive initial conditions remain
some of the campaigns by media lose their influence or fail to positive for all 𝑡 > 0, it is necessary to prove that all the state
attract people. variables are nonnegative, so we have the following lemma.

2.2. Basic Properties Lemma 2. If 𝑆(0) > 0, 𝑋(0) > 0, 𝐴(0) > 0, 𝑅(0) > 0,
and 𝑀(0) > 0, then the solutions 𝑆(𝑡), 𝑋(𝑡), 𝐴(𝑡), 𝑅(𝑡), and
2.2.1. Invariant Region 𝑀(𝑡) of system (2) are positive for all 𝑡 ≥ 0.

Lemma 1. All feasible regions Ω defined by Proof. Under the given initial conditions, it is easy to prove
that the solutions of the system (2) are positive; if not, we
Ω = { (𝑆 (𝑡) , 𝑋 (𝑡) , 𝐴 (𝑡) , 𝑅 (𝑡) , 𝑀 (𝑡)) ∈ 𝑅+5 : assume a contradiction that there exists a first time 𝑡1 such
that
(3)
𝑚 𝑆 (0) > 0, 𝑆 (𝑡1 ) = 0, 𝑆󸀠 (𝑡1 ) ≤ 0,
𝑆 (𝑡) + 𝑋 (𝑡) + 𝐴 (𝑡) + 𝑅 (𝑡) ≤ 1; 𝑀 (𝑡) ≤ }
𝛾
𝑋 (𝑡) > 0, 𝐴 (𝑡) > 0, 𝑅 (𝑡) > 0, (11)
with the initial conditions 𝑆(0) ≥ 0, 𝑋(0) ≥ 0, 𝐴(0) ≥ 0, 𝑅(0) ≥
0, and 𝑀(0) ≥ 0 are positively invariant for system (2). 𝑀 (𝑡) > 0, 0 ≤ 𝑡 < 𝑡1 .
In that case, from the first equation of system (2), we have
Proof. Adding the first four equations of (2), we have
𝑆󸀠 (𝑡1 ) = 𝜇 + 𝜎𝑋 > 0. (12)
𝑑𝑁
= 𝜇 − 𝜇𝑁. (4) which is a contradiction meaning that 𝑆(𝑡) > 0, 𝑡 > 0.
𝑑𝑡
Or there exists a 𝑡2 such that
It follows that
𝑁 (𝑡) = 1 + 𝑁 (0) 𝑒−𝜇𝑡 , (5) 𝑋 (0) > 0, 𝑋 (𝑡2 ) = 0, 𝑋󸀠 (𝑡2 ) ≤ 0,

where 𝑁(0) is the initial total number of people. Thus, 𝑆 (𝑡) > 0, 𝐴 (𝑡) > 0, 𝑅 (𝑡) > 0, (13)
lim 𝑁 (𝑡) = 1. (6) 𝑀 (𝑡) > 0, 0 ≤ 𝑡 < 𝑡2 .
𝑡→∞

Then 𝑆(𝑡) + 𝑋(𝑡) + 𝐴(𝑡) + 𝑅(𝑡) ≤ 1; the subpopulation is In that case, from the second equation of system (2), we have
represented in proportion to the total number of population. 𝑋󸀠 (𝑡2 ) = 𝜆𝑆𝑀 > 0, (14)
Thus, since the system (2) monitors human population,
it is plausible to assume that all its state variables and which is a contradiction meaning that 𝑋(𝑡) > 0, 𝑡 ≥ 0.
parameters are nonnegative for all 𝑡 ≥ 0. Further, from the Or there exists a 𝑡3 such that
last equation of system (2), we have 𝐴 (0) > 0, 𝐴 (𝑡3 ) ≤ 0, 𝐴󸀠 (𝑡3 ) = 0,
𝑑𝑀
= 𝑚𝐴 − 𝛾𝑀 𝑆 (𝑡) > 0, 𝑋 (𝑡) > 0, 𝑅 (𝑡) > 0, (15)
𝑑𝑡 (7)
𝑀 (𝑡) > 0, 0 ≤ 𝑡 < 𝑡3 .
≤ 𝑚 − 𝛾𝑀.
In that case, from the third equation of system (2), we have
It follows that
𝑚 𝐴󸀠 (𝑡3 ) = 𝑞𝑅 > 0. (16)
0 ≤ 𝑀 (𝑡) ≤ + 𝑀 (0) 𝑒−𝛾𝑡 , (8)
𝛾 which is a contradiction meaning that 𝐴(𝑡) > 0, 𝑡 ≥ 0.
where 𝑀(0) represents the initial value of cumulative density Similarly, it can be shown that 𝑅(𝑡) > 0 and 𝑀(𝑡) > 0 for
of awareness programs. Thus, all 𝑡 ≥ 0. Thus, the solutions 𝑆(𝑡), 𝑋(𝑡), 𝐴(𝑡), 𝑅(𝑡), and 𝑀(𝑡)
𝑚 of system (2) remain positive for all 𝑡 > 0.
lim sup 𝑀 (𝑡) ≤ . (9)
𝑡→∞ 𝛾
3. Analysis of the Model
It implies that the region
There are one alcohol free equilibrium 𝐸0 and one alcohol
Ω = { (𝑆 (𝑡) , 𝑋 (𝑡) , 𝐴 (𝑡) , 𝑅 (𝑡) , 𝑀 (𝑡)) present equilibrium 𝐸∗ for system (2).

𝑚 3.1. Alcohol Free Equilibrium and the Basic Reproduction

∈ 𝑅+5 : 𝑆 (𝑡) + 𝑋 (𝑡) + 𝐴 (𝑡) + 𝑅 (𝑡) ≤ 1; 𝑀 (𝑡) ≤ }
𝛾 Number. The model has an alcohol free equilibrium given by
(10) 𝐸0 = (1, 0, 0, 0, 0) . (17)
4 Abstract and Applied Analysis

In the following, the basic reproduction number of system The derivative of 𝑉 is given by
(2) will be obtained by the next generation matrix method
formulated in [25]. 𝑆̇ 𝑞 ̇ ̇
Let 𝑥 = (𝐴, 𝑅, 𝑀, 𝑆, 𝑋)𝑇 ; then system (2) can be written − + 𝐴̇+
𝑉̇= 𝑆 ̇ 𝑅+𝑋
as 𝑆 𝑞+𝜇
= 𝜇 − 𝛽𝑆𝐴 − 𝜆𝑆𝑀 + 𝜎𝑋 − 𝜇𝑆
𝑑𝑥
= F (𝑥) − V (𝑥) , (18)
𝑑𝑡 1
− (𝜇 − 𝛽𝑆𝐴 − 𝜆𝑆𝑀 + 𝜎𝑋 − 𝜇𝑆)
𝑆
where
+ 𝜆𝑆𝑀 − 𝜎𝑋 − 𝜇𝑋 + 𝛽𝑆𝐴 + 𝑞𝑅 − 𝑝𝐴
𝛽𝑆𝐴 𝑞
− 𝜇𝐴 + (𝑝𝐴 − 𝑞𝑅 − 𝜇𝑅) (25)
0
𝑞+𝜇
F (𝑥) = ( 0 ) ,
0 1 1
0 = 𝜇 − 𝜇𝑆 − 𝜇 + 𝛽𝐴 + 𝜆𝑀 − 𝜎𝑋 + 𝜇
𝑆 𝑆
−𝑞𝑅 + (𝑝 + 𝜇) 𝐴 𝑞
(19) − 𝜇𝑋 + 𝑞𝑅 − 𝑝𝐴 − 𝜇𝐴 + 𝑝𝐴 − 𝑞𝑅
𝑞+𝜇
−𝑝𝐴 + (𝑞 + 𝜇) 𝑅
( ) 1 1 1
( 𝛾𝑀 − 𝑚𝐴 ) = 𝜇 (2 − 𝑆 − ) + 𝛽 (1 − ) 𝐴 + (1 − ) 𝜎𝑋.
V (𝑥) = ( ). 𝑆 𝑅0 𝑆
( )
−𝜇 + 𝛽𝑆𝐴 + 𝜆𝑆𝑀 − 𝜎𝑋 + 𝜇𝑆
If 𝑅0 ≤ 1, then 𝛽(1 − (1/𝑅0 )) ≤ 0. As we know,
( −𝜆𝑆𝑀 + (𝜎 + 𝜇) 𝑋 ) 2 − 𝑆 − (1/𝑆) ≤ 0 and 1 − (1/𝑆)𝜎𝑋 ≤ 0, so we obtain
𝑉̇ ≤ 0. Furthermore, 𝑉̇ = 0 only if 𝑆 = 1 or 𝑅0 = 1.
The Jacobian matrices of F(𝑥) and V(𝑥) at the alcohol free The maximum invariant set in {(𝑆, 𝑋, 𝐴, 𝑅, 𝑀) : 𝑉̇= 0} is
equilibrium 𝐸0 are, respectively, the singleton 𝐸0 . By Lasalle’s invariance principle [27], 𝐸0 is
globally asymptotically stable in Ω.
𝐹 0
𝐷F (𝐸0 ) = ( 2×2 ) , (20)
0 0
3.3. Alcohol Present Equilibrium
𝑉2×2 0 0 0
−𝑚 0 𝛾 0 0 3.3.1. Existence of the Alcohol Present Equilibrium
𝐷V (𝐸0 ) = ( ), (21)
𝛽 0 𝜆 𝜇 −𝜎 Theorem 4. If 𝑅0 > 1, system (2) has a unique alcohol present
0 0 −𝜆 0 𝜎+𝜇 equilibrium 𝐸∗ (𝑆∗ , 𝑋∗ , 𝐴∗ , 𝑅∗ , 𝑀∗ ), where

where
1 𝑝𝑞
𝑆∗ = (𝑝 + 𝜇 − ),
𝛽 0 𝑝 + 𝜇 −𝑞 𝛽 𝑞+𝜇
𝐹=( ), 𝑉=( ). (22)
0 0 −𝑝 𝑞 + 𝜇 𝜆𝑚 𝑝𝑞
𝑋∗ = (𝑝 + 𝜇 − ) 𝐴∗ ,
𝛽𝛾 (𝜎 + 𝜇) 𝑞+𝜇
The model reproduction number denoted by 𝑅0 is thus given (26)
by ∗ 𝑝
𝑅 = 𝐴∗ ,
𝑞+𝜇
𝛽 (𝑞 + 𝜇)
𝑅0 = . (23) 𝑚 ∗
(𝑝 + 𝜇) (𝑞 + 𝜇) − 𝑝𝑞 𝑀∗ = 𝐴 .
𝛾

3.2. Global Stability of 𝐸0 Proof. It follows from system (2) that

Theorem 3. For system (2), the alcohol free equilibrium 𝐸0 is
globally asymptotically stable if 𝑅0 ≤ 1. 𝜇 − 𝛽𝑆𝐴 − 𝜆𝑆𝑀 + 𝜎𝑋 − 𝜇𝑆 = 0,

Proof. We introduce the following Lyapunov function [26]: 𝜆𝑆𝑀 − 𝜎𝑋 − 𝜇𝑋 = 0,

𝛽𝑆𝐴 + 𝑞𝑅 − 𝑝𝐴 − 𝜇𝐴 = 0, (27)
𝑉 (𝑆 (𝑡) , 𝑋 (𝑡) , 𝐴 (𝑡) , 𝑅 (𝑡) , 𝑀 (𝑡))
𝑝𝐴 − 𝑞𝑅 − 𝜇𝑅 = 0,
𝑞 (24)
= 𝑆 − ln 𝑆 + 𝑋 + 𝐴 + 𝑅.
𝑞+𝜇 𝑚𝐴 − 𝛾𝑀 = 0.
Abstract and Applied Analysis 5

From the fourth and fifth equations of (27), we obtain 4. Numerical Simulation
𝑝 To illustrate the analytic results obtained above, we give some
𝑅= 𝐴, (28) simulations using the parameter values in Table 1. Numerical
𝑞+𝜇
results are displayed in the following figures. It is reported
𝑚
𝑀= 𝐴. (29) that binge drinking occurs in up to 30 percent of adolescents
𝛾 and alcohol use disorders occur in about 6 percent of this
age group; approximately 50 to 60 percent of them remain
Substituting 𝑅 into the third equation of (27), we have abstinent at the end of a year’s treatment and a majority of
those stay dry permanently [29, 30]. So the initial condition is
1 𝑝𝑞 taken as the lowest by 𝐴(0) = 0.3 and 𝑅(0) = 0.15; in addition,
𝑆= (𝑝 + 𝜇 − ). (30)
𝛽 𝑞+𝜇 we assume that 𝑋(0) = 0.1, 𝑆(0) = 0.45, and 𝑀(0) = 0.002.
First, we choose 𝑃 = 0.3 and the numerical simulation
Then substituting (29) and (30) into the second equation of
gives 𝑅0 = 0.5691 < 1; then the alcohol free equilibrium 𝐸0 is
(27), we get
globally asymptotically stable (Figure 2). Second, we choose
𝑝𝑞 𝑃 = 0.21 and the numerical simulation gives 𝑅0 = 1; the
𝜆𝑚
𝑋= (𝑝 + 𝜇 − ) 𝐴. (31) alcohol free equilibrium 𝐸0 is globally asymptotically stable
𝛽𝛾 (𝜎 + 𝜇) 𝑞+𝜇
(Figure 3). Third, we choose 𝑃 = 0.032 and the numerical
simulation gives 𝑅0 = 1.165 > 1; the alcohol present
For 𝐴 ≠0, substituting (29), (30), and (31) into the first
equilibrium 𝐸∗ is globally asymptotically stable (Figure 4).
equation of (27) gives
Finally, for showing that awareness programs by media
𝜇 − 𝜇𝑆 can help to control the population of binge drinking, we
compare our model with the model which has been studied
𝑝𝑞 𝜆𝑚 𝑝𝑞 by Mulone and Straughan [10]; the model in [10] is governed
= (𝑝 + 𝜇 − )𝐴 + (𝑝 + 𝜇 − )𝐴 by the following system of nonlinear ordinary differential
𝑞+𝜇 𝛽𝛾 𝑞+𝜇 (32)
equations:
𝜎𝜆𝑚 𝑝𝑞
− (𝑝 + 𝜇 − ) 𝐴.
𝛽𝛾 (𝜎 + 𝜇) 𝑞+𝜇
𝑑𝑆 (𝑡) 𝛽𝐴𝑆
= 𝜇𝑁 − − 𝜇𝑆,
Substituting (30) into (32), we get 𝑑𝑡 𝑁
𝑑𝐴 (𝑡) 𝛽𝐴𝑆
𝜇 𝑝𝑞 = + 𝜌𝑅 − (𝜙 + 𝜇) 𝐴, (34)
𝐴 = (𝜇 − (𝑝 + 𝜇 − )) 𝑑𝑡 𝑁
𝛽 𝑞+𝜇
𝑑𝑅 (𝑡)
𝑝𝑞 𝜆𝑚 𝜎𝜆𝑚
−1 = 𝜙𝐴 − (𝜌 + 𝜇) 𝑅.
× ((𝑝 + 𝜇 − ) [1 + − ]) 𝑑𝑡
𝑞+𝜇 𝛽𝛾 𝛽𝛾 (𝜎 + 𝜇)
1 We choose 𝑃 = 0.032 and the numerical simulation gives
= 𝜇 (1 − ) 𝑅0 = 1.1645 > 1; the alcohol present equilibrium 𝑃∗ of
𝑅0
(34) is globally asymptotically stable. The number of the binge
−1 drinking population is generally on the rise without effects of
𝑝𝑞 𝜆𝑚 𝜎𝜆𝑚
× ((𝑝 + 𝜇 − ) [1 + − ]) . awareness programs by media.
𝑞+𝜇 𝛽𝛾 𝛽𝛾 (𝜎 + 𝜇)
For showing the distinction between (2) and (34), we take
(33)
𝑃 = 0.032 and compare the variations of the binge drinking
population 𝐴(𝑡). We can see that, because of awareness
Therefore, there exists a unique positive root in the programs by media (Figure 5), the number of people who
interval (0, 1) when 𝑅0 > 1; there is no positive root in the have alcohol problems decreases due to awareness-induced
interval [0, 1] when 𝑅0 ≤ 1. isolation of susceptible people, so making more people avoid
For 𝑅0 > 1, we carry out the simulation (see Figure 4) and contacting with the heavy drinkers due to the fact that
give the following conjecture. the awareness programs by media can reduce the alcohol
Conjecture. If 𝑅0 > 1, the alcohol present equilibrium 𝐸∗ of problems.
(2) is globally asymptotically stable. Further, the variations of the binge drinking population
𝐴(𝑡) with respect to time 𝑡 for different values of growth rate
Remark. Since the global stability of (2) is a hard problem, we of density of awareness programs 𝑚 are shown in Figure 6.
only carry out simulation (see Figure 4). How to prove the From these figures, it is apparent that as the rate of density of
global stability of the alcohol present equilibrium 𝐸∗ of (2) awareness programs 𝑚 increases the binge drinking popula-
and give the conditions based on model parameters is still an tion 𝐴(𝑡) decreases. Then, the awareness programs by media
open problem. are an effective measure in alcohol problems.
6 Abstract and Applied Analysis

Table 1: Description and estimation of parameters.

𝜇 Recruitment rate of the population 0.25 year−1 [10]

𝛽 Transmission coefficient 0.3 year−1 [10]
𝜆 Dissemination rate of aware individuals 0.02 year−1 estimate
𝜎 Rate of transfer of aware individuals to susceptible class 0.001 year−1 estimate
𝑝 The fraction of 𝐴 go in to treatment Variable Variable
𝑞 The fraction of 𝑅 who relapse into 𝐴 0.8 year−1 [10]
𝑚 Growth rate of density of awareness programs 0.05 year−1 [28]
𝛾 Depletion rate of awareness programs 0.05 year−1 estimate

1 1

Humans population size

Humans population size

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
0 50 100 150 200 0 50 100 150 200
Time t Time t

S(t) R(t) S(t) R(t)

X(t) M(t) X(t) M(t)
A(t) A(t)

Figure 2: If 𝑅0 < 1, the alcohol free equilibrium 𝐸0 is globally Figure 4: If 𝑅0 > 1, the alcohol present equilibrium 𝐸∗ is globally
asymptotically stable. asymptotically stable.

0.4

1
0.35
Humans population size

Binge drinking population size

0.8

0.3
0.6

0.25
0.4

0.2 0.2

0
0 50 100 150 200 250 300 0.15
0 10 20 30 40 50
Time t
Time
S(t) R(t)
X(t) M(t) A(t) of (2.1)
A(t) A(t) of (4.1)

Figure 3: If 𝑅0 = 1, the alcohol free equilibrium 𝐸0 is globally Figure 5: Variations of the binge drinking population 𝐴(𝑡) of (2)
asymptotically stable. and (34).
Abstract and Applied Analysis 7

0.34 Conflict of Interests

The authors declare that there is no conflict of interests
Binge drinking population size

0.32 regarding the publication of this paper.

0.3 Acknowledgments
This work was partially supported by the NNSF of
0.28
China (10961018), the NSF of Gansu Province of China
(1107RJZA088), the NSF for Distinguished Young Scholars
0.26 of Gansu Province of China (1111RJDA003), the Special Fund
for the Basic Requirements in the Research of University of
Gansu Province of China, and the Development Program
0.24
for Hong Liu Distinguished Young Scholars in Lanzhou
University of Technology.
0 10 20 30 40 50
Time
References
m = 0.02 m=0
m = 0.008 [1] J. Rehm, “The risks associated with alcohol use and alcoholism,”
Alcohol Research & Health, vol. 34, no. 2, pp. 135–143, 2011.
Figure 6: Variations of binge drinking population with time for
different values of 𝑚. [2] J. Rehm, C. Mathers, S. Popova, M. Thavorncharoensap, Y.
Teerawattananon, and J. Patra, “Global burden of disease and
injury and economic cost attributable to alcohol use and
alcohol-use disorders,” The Lancet, vol. 373, no. 9682, pp. 2223–
2233, 2009.
5. Discussion [3] J. Ribes, R. Clèries, L. Esteban, V. Moreno, and F. X. Bosch,
“The influence of alcohol consumption and hepatitis B and C
We have formulated the effect of awareness programs on
infections on the risk of liver cancer in Europe,” Journal of
the binge drinking and analyzed their dynamical behaviors. Hepatology, vol. 49, no. 2, pp. 233–242, 2008.
The model exhibits two equilibria, explicitly the alcohol free [4] H. Wechsler and T. F. Nelson, “Focusing attention on college
equilibrium and alcohol present equilibrium. By constructing student alcohol consumption and the environmental conditions
Lyapunov function, we establish some sufficient conditions that promote it,” Journal of Studies on Alcohol and Drugs, vol. 69,
for the global stability of the alcohol free and the alcohol no. 4, pp. 481–490, 2008.
present equilibria and give some numerical simulations to [5] L. M. Powell, J. Williams, and H. Wechsler, “Study habits and
explain our main result. In our model, we have assumed that the level of alcohol use among college students,” Education
the cumulative density of awareness programs by media is Economics, vol. 12, no. 2, pp. 135–149, 2004.
proportional to the population of binge drinking heavily. It [6] H. Wechsler, B. Moeykens, A. Davenport, S. Castillo, and J.
may be noted that the number of alcohol consumption cases Hansen, “The adverse impact of heavy episodic drinkers on
known to the policy makers is sometimes old and thus the other college students,” Journal of Studies on Alcohol and Drugs,
number of awareness depends on data. So it is more plausible vol. 56, no. 6, pp. 628–634, 1995.
to consider a time delay in the growth rate of awareness [7] S. Mushayabasa and C. P. Bhunu, “Modelling the effects of
programs [21]. We can modify (2) to the following model with heavy alcohol consumption on the transmission dynamics of
delay: gonorrhea,” Nonlinear Dynamics, vol. 66, no. 4, pp. 695–706,
2011.
[8] P. Ormerod and G. Wiltshire, “‘Binge” drinking in the UK: a
𝑑𝑆 (𝑡) social network phenomenon,” Mind & Society, vol. 8, no. 2, pp.
= 𝜇 − 𝛽𝑆𝐴 − 𝜆𝑆𝑀 + 𝜎𝑋 − 𝜇𝑆, 135–152, 2009.
𝑑𝑡
[9] F. Sanchez, X. H. Wang, C. Castillo-Chavez, D. M. Gorman, and
𝑑𝑋 (𝑡) P. J. Gruenewald, “Drinking as an epidemica simple mathemat-
= 𝜆𝑆𝑀 − 𝜎𝑋 − 𝜇𝑋,
𝑑𝑡 ical model with recovery and relapse,” in Therapists Guide to
Evidence-Based Relapse Prevention: Practical Resources for the
𝑑𝐴 (𝑡)
= 𝛽𝑆𝐴 + 𝑞𝑅 − 𝑝𝐴 − 𝜇𝐴, (35) Mental Health Professional, K. A. Witkiewitz and G. A. Marlatt,
𝑑𝑡 Eds., pp. 353–368, Academic Press, Burlington, Canada, 2007.
𝑑𝑅 (𝑡) [10] G. Mulone and B. Straughan, “Modeling binge drinking,”
= 𝑝𝐴 − 𝑞𝑅 − 𝜇𝑅, International Journal of Biomathematics, vol. 5, no. 1, Article ID
𝑑𝑡 1250005, 14 pages, 2012.
𝑑𝑀 (𝑡) [11] H.-F. Huo and N.-N. Song, “Global stability for a binge drinking
= 𝑚𝐴 (𝑡 − 𝜏) − 𝛾𝑀. model with two stages,” Discrete Dynamics in Nature and
𝑑𝑡
Society, vol. 2012, Article ID 829386, 15 pages, 2012.
[12] H. F. Huo and C. C. Zhu, “Modeling the effect of constant
We leave these works for the future. immigration on drinking behaviour,” Submitted.
8 Abstract and Applied Analysis

[13] H. F. Huo and C. C. Zhu, “Stability of a quit drinking model with

relapse,” Submitted.
[14] H.-F. Huo and C.-C. Zhu, “Influence of relapse in a giving
up smoking model,” Abstract and Applied Analysis, vol. 2013,
Article ID 525461, 12 pages, 2013.
[15] G. Schwitzer, G. Mudur, D. Henry et al., “What are the
roles and responsibilities of the media in disseminating health
information?” PLoS Medicine, vol. 2, no. 7, article e215, 2005.
[16] E. Jaramillo, “The impact of media-based health education on
tuberculosis diagnosis in Cali, Colombia,” Health Policy and
Planning, vol. 16, no. 1, pp. 68–73, 2001.
[17] R. Liu, J. Wu, and H. Zhu, “Media/psychological impact on mul-
tiple outbreaks of emerging infectious diseases,” Computational
and Mathematical Methods in Medicine, vol. 8, no. 3, pp. 153–
164, 2007.
[18] T. P. Palfai, R. Zisserson, and R. Saitz, “Using personalized
feedback to reduce alcohol use among hazardous drinking col-
lege students: the moderating effect of alcohol-related negative
consequences,” Addictive Behaviors, vol. 36, no. 5, pp. 539–542,
2011.
[19] J. M. Tchuenche and C. T. Bauch, “Dynamics of an infectious
disease where media coverage influences transmission,” ISRN
Biomathematics, vol. 2012, Article ID 581274, 10 pages, 2012.
[20] H. Xiang, N. N. Song, and H. F. Huo, “Modelling effects of
public health educational campaigns on drinking dynamics,”
Submitted.
[21] A. K. Misra, A. Sharma, and V. Singh, “Effect of awareness
programs in controlling the prevalence of an epidemic with
time delay,” Journal of Biological Systems, vol. 19, no. 2, pp. 389–
402, 2011.
[22] A. K. Misra, A. Sharma, and J. B. Shukla, “Modeling and analysis
of effects of awareness programs by media on the spread of
infectious diseases,” Mathematical and Computer Modelling, vol.
53, no. 5-6, pp. 1221–1228, 2011.
[23] “WHO Report of Alcohol in developing societies summary,”
http://www.who.int/.
[24] D. Chisholm, J. Rehm, M. van Ommeren, and M. Monteiro,
“Reducing the global burden of hazardous alcohol use: a
comparative cost-effectiveness analysis,” Journal of Studies on
Alcohol, vol. 65, no. 6, pp. 782–793, 2004.
[25] P. van den Driessche and J. Watmough, “Reproduction numbers
and sub-threshold endemic equilibria for compartmental mod-
els of disease transmission,” Mathematical Biosciences, vol. 180,
pp. 29–48, 2002.
[26] J.-Q. Li, Y.-L. Yang, and Y.-C. Zhou, “Global stability of an
epidemic model with latent stage and vaccination,” Nonlinear
Analysis: Real World Applications, vol. 12, no. 4, pp. 2163–2173,
2011.
[27] J. P. LaSalle, “Stability theory for ordinary differential equa-
tions,” Journal of Differential Equations, vol. 4, no. 1, pp. 57–65,
1968.
[28] L. B. Snyder, “Health communication campaigns and their
impact on behavior,” Journal of Nutrition Education and Behav-
ior, vol. 39, no. 2, pp. S32–S40, 2007.
[29] D. B. Clark, O. Bukstein, and J. Cornelius, “Alcohol use disorders
in adolescents: epidemiology, diagnosis, psychosocial interven-
tions, and pharmacological treatment,” Pediatric Drugs, vol. 4,
no. 8, pp. 493–502, 2002.
[30] “WebMD Report of Alcohol Abuse Health Center,” http://www.
webmd.com.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 156948, 12 pages
http://dx.doi.org/10.1155/2014/156948

Research Article
Existence and Stability of Periodic Solution to Delayed
Nonlinear Differential Equations

Xiang Gu,1 Huicheng Wang,1 P. J. Y. Wong,2 and Yonghui Xia1

1
Department of Mathematics, Zhejiang Normal University, Jinhua, Zhejiang 321004, China
2
School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore 639798

Correspondence should be addressed to Yonghui Xia; yhxia@zjnu.cn

Received 23 February 2014; Accepted 17 March 2014; Published 14 April 2014

Academic Editor: Yongli Song

Copyright © 2014 Xiang Gu et al. This is an open access article distributed under the Creative Commons Attribution License, which
permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

The main purpose of this paper is to study the periodicity and global asymptotic stability of a generalized Lotka-Volterra’s
competition system with delays. Some sufficient conditions are established for the existence and stability of periodic solution of
such nonlinear differential equations. The approaches are based on Mawhin’s coincidence degree theory, matrix spectral theory,
and Lyapunov functional.

1. Introduction and Motivation model, Chen [13] proposed a more complicated model as
follows:
In the past few decades, differential equations have been
used in the study of population dynamics, ecology and 𝑛
𝛼
epidemiology, malaria transmission, and so forth (see, e.g., 𝑦𝑖̇(𝑡) = 𝑦𝑖 (𝑡) [𝑟𝑖 (𝑡) − ∑𝑎𝑖𝑗 (𝑡) 𝑦𝑗 𝑖𝑗 (𝑡)] , 𝑖 = 1, 2, . . . , 𝑛,
𝑗=1
[1–10]). One of the rudimentary population systems is the [ ]
nonautonomous 𝑛-species competitive model: (2)

where 𝛼𝑖 provides a nonlinear measure of intraspecific

interference and 𝛼𝑖𝑗 provides a measure of interspecific
𝑛 interference. Chen studied the permanence of (2) by average
𝑦𝑖̇(𝑡) = 𝑦𝑖 (𝑡) [𝑏𝑖 (𝑡) − ∑ 𝑎𝑖𝑗 (𝑡) 𝑦𝑗 (𝑡)] , 𝑖 = 1, 2, . . . , 𝑛. method. For the sake of convenience, in what follows, the
𝑗=1
[ ] new factor introduced by Gilpin and Ayala is called Gilpin-
(1) Ayala effect. On the other hand, many scholars think that the
delayed models are more realistic. Because time delays may
lead to oscillation, bifurcation, chaos, and instability which
may be harmful to a system. In fact, May [14] has shown that
Based on Mawhin’s coincidence degree theory, spectral the-
if a time delay is incorporated into the resource limitation
ory, and novel estimation techniques for the priori bounds of of the logistic equation, then it has destabilizing effect on
unknown solutions to the equation 𝐿𝑥 = 𝜆𝑁𝑥, Xia and Han the stability of the system (also see Cooke and Grossman
[8] studied the existence and stability of periodic solution for [15]). But sometimes, the delays may be harmless under some
(1). But model (1) is doubted by Gilpin and Ayala [11], they restriction and this is more important in some sense (e.g.,
thought that the model is not reasonable enough. In order to see [16]). A very basic and important ecological problem in
fit data in the experiments conducted in Ayala et al. [12] and the study of multispecies population dynamics concerns the
to yield significantly more accurate results on the competitive global existence and global asymptotic stability of positive
2 Abstract and Applied Analysis

periodic solutions. It is doubted whether the existence and D, and 𝐸𝑛 is the identity matrix of size 𝑛. A matrix or vector
stability of periodic solutions can be affected by the delays or D > 0 (resp., D ≥ 0) means that all entries of D are positive
Gilpin-Ayala effect. For this reason, in the present paper, we (resp., nonnegative). For matrices or vectors D and 𝐸, D > 𝐸
consider the Gilpin-Ayala type delayed system as follows: (resp., D ≥ 𝐸) means that D − 𝐸 > 0 (resp., D − 𝐸 ≥ 0). We
also denote the spectral radius of the matrix D by 𝜌(D).
𝑛
𝛼 If V = (V1 , V2 , . . . , V𝑛 )𝑇 ∈ R𝑛 , then we have a choice of
𝑦𝑖̇(𝑡) = 𝑦𝑖 (𝑡) [𝑟𝑖 (𝑡) − ∑𝑎𝑖𝑗 (𝑡) 𝑦𝑗 𝑖𝑗 (𝑡) vector norms in R𝑛 ; for instance, ‖V‖1 , ‖V‖2 , and ‖V‖∞ are the
𝑗=1
[ commonly used norms, where
𝑛
𝛽
− ∑𝑏𝑖𝑗 (𝑡) 𝑦𝑗 𝑖𝑗 (𝑡 − 𝜏𝑖𝑗 )] , 𝑖 = 1, 2, . . . , 𝑛, 𝑛
󵄨 󵄨
𝑗=1
‖V‖1 = ∑ 󵄨󵄨󵄨V𝑖 󵄨󵄨󵄨 ,
] 𝑗=1
(3)
1/2
where 𝑦𝑖 is the population density of the 𝑖th species; 𝑟𝑖 { 𝑛 󵄨 󵄨2 } (6)
is the intrinsic exponential growth rate of the 𝑖th species; ‖V‖2 = {∑󵄨󵄨󵄨V𝑖 󵄨󵄨󵄨 } ,
𝑎𝑖𝑗 , 𝑏𝑖𝑗 measure the amount of competition between the {𝑗=1 }
𝑖th species and the 𝑗th species (𝑖 ≠𝑗); and 𝛼𝑖𝑗 , 𝛽𝑖𝑗 provide 󵄨󵄨 󵄨󵄨
‖V‖∞ = max 󵄨󵄨V𝑖 󵄨󵄨 .
a nonlinear measure of intraspecific interference. For the 1≤𝑖≤𝑛
point of biological view, the coefficients are assumed to be
continuous 𝜔-periodic functions; we always assume that 𝑟𝑖 , We recall the following norms of matrices induced by respec-
𝑎𝑖𝑗 , 𝑏𝑖𝑗 , 𝜏𝑖𝑗 , 𝑖, 𝑗 = 1, 2, . . . , 𝑛, are nonnegative and 𝑎𝑖𝑖 , 𝑏𝑖𝑖 are tive vector norms. For instance, if A = (𝑎𝑖𝑗 )𝑛×𝑛 , the norm of
strictly positive. And system (3) is supplemented with the the matrix ‖A‖ induced by a vector norm ‖ ⋅ ‖ is defined by
initial condition
𝑦𝑖 (𝑠) = 𝜙𝑖 (𝑠) , 𝑠 ∈ [−𝜏, 0] , 𝜙𝑖 (0) > 0, 𝑖 = 1, 2, . . . , 𝑛, ‖AV‖𝑝
‖A‖𝑝 = sup = sup ‖AV‖𝑝 = sup ‖AV‖𝑝 . (7)
(4) 0 ‖V‖𝑝
V∈R𝑛 ,V ≠ ‖V‖𝑝 =1 ‖V‖𝑝 ≤1

where 𝜏 = max1≤𝑖≤𝑛 {𝜏𝑖𝑗 }, 𝜙 = (𝜙1 , . . . , 𝜙𝑛 ) ∈ BC([−𝜏, 0], R+𝑛 ),

and BC is the set of all bounded continuous functions from In particular one can show that ‖A‖1 = max1≤𝑗≤𝑛 ∑𝑛𝑖=1 |𝑎𝑖𝑗 |
[−𝜏, 0] into R+𝑛 . It is easy to see that for such given initial (column norm) and ‖A‖2 = [𝜆 max (A𝑇 A)]1/2 = [max .
value condition, the corresponding solution of (3) remains eigenvalue of (A𝑇 A)]1/2 , ‖A‖∞ = max1≤𝑖≤𝑛 ∑𝑛𝑗=1 |𝑎𝑖𝑗 | (row
positive for all 𝑡 ≥ 0. The purpose of this paper is to obtain
norm).
some new and interesting criteria for the existence and global
asymptotic stability of periodic solution of system (3). Definition 1 (see [17, 18]). Let 𝑋, 𝑍 be normed real Banach
The structure of this paper is as follows. In Section 2, some spaces, let 𝐿 : Dom 𝐿 ⊂ 𝑋 → 𝑍 be a linear mapping, and
new and interesting sufficient conditions for the existence let 𝑁 : 𝑋 → 𝑍 be a continuous mapping. The mapping 𝐿
of periodic solution of system (3) are obtained. Section 3 is is called a Fredholm mapping of index zero if dim Ker 𝐿 =
devoted to examining the stability of this periodic solution. codim Im 𝐿 < +∞ and Im 𝐿 is closed in 𝑍. If 𝐿 is a Fredholm
In Section 4, some corollaries and discussion are presented. mapping of index zero and there exist continuous projectors
Finally, some examples and their simulations are given to 𝑃 : 𝑋 → 𝑋 and 𝑄 : 𝑍 → 𝑍 such that Im 𝑃 = Ker 𝐿 and
show the effectiveness and feasibility of our results. Ker 𝑄 = Im 𝐿 = Im(𝐼 − 𝑄), it follows that 𝐿 | dom 𝐿 ∩ Ker 𝑃 :
(𝐼 − 𝑃)𝑋 → Im 𝐿 is invertible. We denote the inverse of that
2. Existence of Periodic Solutions map by 𝐾𝑃 . If Ω is an open bounded subset of 𝑋, the mapping
𝑁 will be called 𝐿-compact on Ω if 𝑄𝑁(Ω) is bounded and
In this section, we will obtain some sufficient conditions for 𝐾𝑃 (𝐼 − 𝑄)𝑁 : Ω → 𝑋 is compact. Since Im 𝑄 is isomorphic
the existence of periodic solution of system (3). to Ker 𝐿, there exists an isomorphism 𝐽 : Im 𝑄 → Ker 𝐿.

2.1. Preliminaries on the Matrix Theory and Degree Theory. Definition 2 (see [17, 18]). Let Ω ⊂ R𝑛 be open and bounded,
For convenience, we introduce some notations, definitions, 𝑓 ∈ 𝐶1 (Ω, R𝑛 ) ∩ 𝐶(Ω, R𝑛 ), and 𝑦 ∈ R𝑛 /𝑓(𝜕Ω ∪ 𝑁𝑓 ); that is,
and lemmas. If 𝑓(𝑡) is a continuous 𝜔-periodic function 𝑦 is a regular value of 𝑓. Here, 𝑁𝑓 = {𝑥 ∈ Ω : 𝐽𝑓 (𝑥) = 0},
defined on R, denote the critical set of 𝑓, and 𝐽𝑓 is the Jacobian of 𝑓 at 𝑥. Then the
󵄨 󵄨 󵄨 󵄨 degree deg{𝑓, Ω, 𝑦} is defined by
𝑓 = min 󵄨󵄨󵄨𝑓 (𝑡)󵄨󵄨󵄨 , 𝑓 = max 󵄨󵄨󵄨𝑓 (𝑡)󵄨󵄨󵄨 ,
𝑡∈[0,𝜔] 𝑡∈[0,𝜔]
(5) deg {𝑓, Ω, 𝑦} = ∑ sgn 𝐽𝑓 (𝑥)
1 𝜔 (8)
𝑚 (𝑓) = ∫ 𝑓 (𝑡) 𝑑𝑡. 𝑥∈𝑓−1 (𝑦)
𝜔 0
We use 𝑥 = (𝑥1 , . . . , 𝑥𝑛 )𝑇 ∈ R𝑛 to denote a column vector, with the agreement that ∑ 𝜙 = 0. For more details about
D = (𝑑𝑖𝑗 )𝑛×𝑛 is an 𝑛 × 𝑛 matrix, D𝑇 denotes the transpose of degree theory, the reader is referred to [18].
Abstract and Applied Analysis 3

Lemma 3 (continuation theorem [17]). Let Ω ⊂ 𝑋 be an open (H3 ) 𝜌(K) < 1, where K = (Γ𝑖𝑗 )𝑛×𝑛 and
and bounded set. Let 𝐿 be a Fredholm mapping of index zero
and let 𝑁 be 𝐿-compact on Ω (i.e., 𝑄𝑁(Ω) is bounded and 0, 𝑖 = 𝑗,
𝐾𝑃 (𝐼 − 𝑄)𝑁 : Ω → 𝑋 is compact). Assume, {
{
{
Γ𝑖𝑗 = { (𝑎𝑖𝑗 + 𝑏𝑖𝑗 ) (12)
(i) for each 𝜆 ∈ (0, 1), 𝑥 ∈ 𝜕Ω ∩ Dom 𝐿 and 𝐿𝑥 ≠𝜆𝑁𝑥; {
{ , 𝑖 ≠𝑗.
(ii) for each 𝑥 ∈ 𝜕Ω ∩ Ker 𝐿, 𝑄𝑁𝑥 ≠0 and deg{𝐽𝑄𝑁, Ω ∩ { (𝑎𝑗𝑗 + 𝑏𝑗𝑗 )
Ker 𝐿, 0} ≠0.
Then 𝐿𝑥 = 𝑁𝑥 has at least one solution in Ω ∩ Dom 𝐿. Then system (3) has at least one positive 𝜔-periodic solution.

Definition 4 (see [19, 20]). A real 𝑛 × 𝑛 matrix A = (𝑎𝑖𝑗 ) is Proof. Note that every solution 𝑦(𝑡) = (𝑦1 (𝑡), 𝑦2 (𝑡), . . . ,
said to be an 𝑀-matrix if 𝑎𝑖𝑗 ≤ 0, 𝑖, 𝑗 = 1, 2, . . . , 𝑛, 𝑖 ≠𝑗, and 𝑦𝑛 (𝑡))𝑇 of system (3) with the initial value condition is
A−1 ≥ 0. positive. Make the change of variables

Lemma 5 (see [19, 20]). Let A ≥ 0 be an 𝑛 × 𝑛 matrix and 𝑦𝑖 (𝑡) = 𝑒𝑥𝑖 (𝑡) , 𝑖 = 1, 2, . . . , 𝑛. (13)
𝜌(A) < 1; then (𝐸𝑛 − A)−1 ≥ 0, where 𝐸𝑛 denotes the identity
matrix of size 𝑛. Then system (3) is the same as
In what follows, we will introduce some function spaces
𝑛
and their norms, which are valid throughout this paper.
𝑥𝑖̇(𝑡) = 𝑟𝑖 (𝑡) − ∑ 𝑎𝑖𝑗 (𝑡) 𝑒𝛼𝑖𝑗 𝑥𝑗 (𝑡)
Denote 𝑗=1
𝑇 (14)
𝑋 = {𝑥 (𝑡) = (𝑥1 (𝑡) , 𝑥2 (𝑡) , . . . , 𝑥𝑛 (𝑡)) 𝑛
𝛽𝑖𝑗 𝑥𝑗 (𝑡−𝜏𝑖𝑗 )
− ∑ 𝑏𝑖𝑗 (𝑡) 𝑒 , 𝑖 = 1, 2, . . . , 𝑛.
∈ 𝐶1 (R, R𝑛 ) | 𝑥 (𝑡 + 𝜔) = 𝑥 (𝑡) ∀𝑡 ∈ R} , 𝑗=1
(9)
𝑇
𝑍 = {𝑥 (𝑡) = (𝑥1 (𝑡) , 𝑥2 (𝑡) , . . . , 𝑥𝑛 (𝑡)) Obviously, system (3) that has at least one 𝜔-periodic solution
is equivalent to system (14) that has at least one 𝜔-periodic
∈ 𝐶 (R, R𝑛 ) | 𝑥 (𝑡 + 𝜔) = 𝑥 (𝑡) ∀𝑡 ∈ R} . solution. To prove Theorem 6, our main tasks are to construct
the operators (i.e., 𝐿, 𝑁, 𝑃, and 𝑄) appearing in Lemma 3 and
And, the norms are given by to find an appropriate open set Ω satisfying conditions (i) and
󵄨󵄨 󵄨 󵄨 󵄨 󵄨󵄨 󵄨 󵄨 󵄨 󵄨 󵄨 (ii) in Lemma 3. To this end, we proceed with three steps.
󵄨󵄨𝑥𝑖 (𝑡)󵄨󵄨󵄨0 = max 󵄨󵄨󵄨𝑥𝑖 (𝑡)󵄨󵄨󵄨 , 󵄨󵄨𝑥𝑖 (𝑡)󵄨󵄨󵄨1 = 󵄨󵄨󵄨𝑥𝑖 (𝑡)󵄨󵄨󵄨0 + 󵄨󵄨󵄨𝑥𝑖̇(𝑡)󵄨󵄨󵄨0
𝑡∈[0,𝜔]
Step 1. In this step, we intend to construct the operators
𝑖 = 1, 2, . . . , 𝑛, appearing in Lemma 3 and verify that they satisfy the condi-
󵄨 󵄨 (10) tions of Lemma 3. For any 𝑥(𝑡) ∈ 𝑋, in view of the periodicity,
‖𝑥 (𝑡)‖0 = max {󵄨󵄨󵄨𝑥𝑖 (𝑡)󵄨󵄨󵄨0 } , it is easy to check that
1≤𝑖≤𝑛

̇ 0 = max {󵄨󵄨󵄨󵄨𝑥𝑖 (𝑡)󵄨󵄨󵄨󵄨1 } .

‖𝑥(𝑡)‖1 = ‖𝑥(𝑡)‖0 + ‖𝑥(𝑡)‖ 𝑛
1≤𝑖≤𝑛
Δ 𝑖 (𝑥, 𝑡) = 𝑟𝑖 (𝑡) − ∑𝑎𝑖𝑗 (𝑡) 𝑒𝛼𝑖𝑗 𝑥𝑗 (𝑡)
𝑗=1
Obviously, 𝑋 and 𝑍, respectively, endowed with the norms
(15)
‖ ⋅ ‖1 and ‖ ⋅ ‖0 are Banach spaces. 𝑛
𝛽𝑖𝑗 𝑥𝑗 (𝑡−𝜏𝑖𝑗 )
− ∑𝑏𝑖𝑗 (𝑡) 𝑒 ∈ 𝑍.
𝑗=1
2.2. Result on the Existence of Periodic Solutions
Theorem 6. Assume that the following conditions hold: And define the operators 𝐿 : Dom 𝐿 ⊂ 𝑋 → 𝑍 and 𝑁 :
𝑋 → 𝑍 as follows:
(H1 ) the system of algebraic equations
𝑑𝑥 (𝑡)
𝑛
𝛼
𝑛
𝛽
𝑋 ∋ 𝑥 (𝑡) 󳨀→ (𝐿𝑥) (𝑡) = ∈ 𝑍,
𝑓 (𝑢) := (𝑚(𝑟𝑖 ) − ∑ 𝑚(𝑎𝑖𝑗 )𝑢𝑗 𝑖𝑗 − ∑ 𝑚(𝑏𝑖𝑗 )𝑢𝑗 𝑖𝑗 ) =0 𝑑𝑡
𝑗=1 𝑗=1 𝑛×1 𝑋 ∋ 𝑥 (𝑡) 󳨀→ (𝑁𝑥) (𝑡) (16)
(11)
𝑇
= ((𝑁𝑥)1 (𝑡), (𝑁𝑥)2 (𝑡), . . . , (𝑁𝑥)𝑛 (𝑡)) ∈ 𝑍,
has finite solutions (𝑢1∗ , 𝑢2∗ , . . . , 𝑢𝑛∗ )𝑇 ∈ R+𝑛 with 𝑢𝑖∗ > 0
and ∑𝑢∗ sgn 𝐽𝑓 (𝑢∗ ) ≠0; where
(H2 ) 𝛼𝑗𝑖 ≤ 𝛼𝑖𝑖 , 𝛽𝑗𝑖 ≤ 𝛼𝑖𝑖 (𝑗 ≠𝑖), 𝛼𝑖𝑖 = 𝛽𝑖𝑖 , 𝜏𝑖𝑖 ≡ 0, 𝑖, 𝑗 =
1, 2, . . . , 𝑛; (𝑁𝑥)𝑖 (𝑡) = Δ 𝑖 (𝑥, 𝑡) , 𝑖 = 1, 2, . . . , 𝑛. (17)
4 Abstract and Applied Analysis

Define, respectively, the projectors 𝑃 : 𝑋 → 𝑋 and 𝑄 : 𝑍 → which implies that 𝑄𝑁(Ω) is bounded in the space (𝑍, ‖ ⋅ ‖0 ).
𝑍 by Secondly, we will show that (𝐾𝑃 (𝐼 − 𝑄)𝑁𝑥) (Ω) is relatively
1 𝜔 1 𝜔 compact in the space (𝑋, ‖ ⋅ ‖1 ). In fact, it follows from (22)
𝑃𝑥 (𝑡) = ∫ 𝑥 (𝑡) 𝑑𝑡, 𝑄𝑧 (𝑡) = ∫ 𝑧 (𝑡) 𝑑𝑡, that
𝜔 0 𝜔 0 (18)
󸀠 𝑇
𝑥 ∈ 𝑋, 𝑧 ∈ 𝑍. (𝐾𝑃 (𝐼 − 𝑄) 𝑁𝑥) = (Ψ1󸀠 (𝑥, 𝑡) , Ψ2󸀠 (𝑥, 𝑡) , . . . , Ψ𝑛󸀠 (𝑥, 𝑡)) ,
(26)
It can be found that the domain of 𝐿 in 𝑋 is actually the whole
space, and where 󸀠 = 𝑑/𝑑𝑡 and
(𝑡) = 0} = R𝑛 ,
Ker 𝐿 = {𝑥 (𝑡) ∈ 𝑋 | 𝐿𝑥 (𝑡) = 0, i.e., 𝑥̇
1 𝜔
Ψ𝑘󸀠 (𝑥, 𝑡) = Δ 𝑘 (𝑥, 𝑡) − ∫ Δ (𝑥, 𝑠) 𝑑𝑠, 𝑘 = 1, 2, . . . , 𝑛.
𝜔 (19) 𝜔 0 𝑘
Im 𝐿 = {𝑧 (𝑡) ∈ 𝑍 | ∫ 𝑧 (𝑡) 𝑑𝑡 = 0} is closed in 𝑍. (27)
0

Moreover, 𝑃, 𝑄 are continuous operators such that This, combining with (22), gives
𝑛
Im 𝑃 = R = Ker 𝐿, Im 𝐿 = Ker 𝑄 = Im (𝐼 − 𝑄) , 󵄨󵄨 󵄨
(20) 󵄨󵄨(𝐾𝑃 (𝐼 − 𝑄) 𝑁𝑥)𝑖 (𝑡)󵄨󵄨󵄨1
dim Ker 𝐿 = codim Im 𝐿 = 𝑛 < +∞. 󵄨 󵄨 󵄨 󵄨
= 󵄨󵄨󵄨(𝐾𝑃 (𝐼 − 𝑄) 𝑁𝑥)𝑖 (𝑡)󵄨󵄨󵄨0 + 󵄨󵄨󵄨󵄨(𝐾𝑃 (𝐼 − 𝑄) 𝑁𝑥)𝑖 (𝑡)󵄨󵄨󵄨󵄨0
󸀠
It follows that 𝐿 is a Fredholm mapping of index zero.
Furthermore, the generalized inverse (to 𝐿) 𝐾𝑃 : Im 𝐿 → 1 1
Dom 𝐿 ∩ Ker 𝑃 exists, which is given by ≤ 𝑀𝑖 𝜔 + 𝑀𝑖 𝜔 + 𝑀𝑖 𝜔
2 2
𝑡
1 𝜔 𝑡 + 𝑀𝑖 + 𝑀𝑖 = 2 (𝜔 + 1) 𝑀𝑖 ,
𝐾𝑃 (𝑦) = ∫ 𝑦 (𝑠) 𝑑𝑠 − ∫ ∫ 𝑦 (𝑠) 𝑑𝑠 𝑑𝑡. (21)
0 𝜔 0 0 (28)
Then 𝑄𝑁 : 𝑋 → 𝑍 and 𝐾𝑃 (𝐼 − 𝑄)𝑁 : 𝑋 → 𝑋 are defined
by which implies that 𝐾𝑃 (𝐼 − 𝑄)𝑁(Ω) is bound in the space
(𝑋, ‖ ⋅ ‖1 ).
1 𝜔 1 𝜔 On the other hand, we prove that (𝐾𝑃 (𝐼 − 𝑄)𝑁𝑥) (Ω) is
𝑄𝑁𝑥 = ( ∫ Δ 1 (𝑥, 𝑡) 𝑑𝑡, ∫ Δ 2 (𝑥, 𝑡) 𝑑𝑡, . . . ,
𝜔 0 𝜔 0 equicontinuous. In view of uniform continuity of 𝑟𝑖 , 𝑎𝑖𝑗 , and
𝑇 𝑏𝑖𝑗 , for any 𝜀 > 0, there exists 𝛿1 > 0 such that, for any 𝑡, 𝑠 ∈ R,
1 𝜔 (22) provided that |𝑡 − 𝑠| < 𝛿1 , we have
∫ Δ 𝑛 (𝑥, 𝑡)𝑑𝑡) ,
𝜔 0
󵄨󵄨 󵄨 󵄨󵄨 󵄨
𝑇 󵄨󵄨𝑟𝑖 (𝑡) − 𝑟𝑖 (𝑠)󵄨󵄨󵄨 < 𝜀, 󵄨󵄨𝑎𝑖𝑗 (𝑡) − 𝑎𝑖𝑗 (𝑠)󵄨󵄨󵄨 < 𝜀,
𝐾𝑃 (𝐼 − 𝑄) 𝑁𝑥 = (Ψ1 (𝑥, 𝑡) , Ψ2 (𝑥, 𝑡) , . . . , Ψ𝑛 (𝑥, 𝑡)) , 󵄨 󵄨
󵄨󵄨 󵄨󵄨 (29)
where 󵄨󵄨𝑏𝑖𝑗 (𝑡) − 𝑏𝑖𝑗 (𝑠)󵄨󵄨 < 𝜀.
󵄨 󵄨
𝑡
1 𝜔 𝑡
Ψ𝑘 (𝑥, 𝑡) = ∫ Δ 𝑘 (𝑥, 𝑠) 𝑑𝑠 − ∫ ∫ Δ (𝑥, 𝑠) 𝑑𝑠 𝑑𝑡 Since any 𝑥(𝑡) = (𝑥1 (𝑡), 𝑥2 (𝑡), . . . , 𝑥𝑛 (𝑡))𝑇 ∈ Ω is equicontin-
0 𝜔 0 0 𝑘
uous, for the same 𝜀, there exists 0 < 𝛿2 ≤ 𝛿1 such that, for
𝜔
𝑡 1 any 𝑡, 𝑠 ∈ R, provided that |𝑡 − 𝑠| < 𝛿2 , we have
−( − ) ∫ Δ 𝑘 (𝑥, 𝑠) 𝑑𝑠, 𝑘 = 1, 2, . . . , 𝑛.
𝜔 2 0 󵄨󵄨 󵄨
󵄨󵄨𝑥𝑖 (𝑡) − 𝑥𝑖 (𝑠)󵄨󵄨󵄨0 < 𝜀. (30)
(23)
It follows from (29) and (30) that
Clearly, 𝑄𝑁 and 𝐾𝑃 (𝐼 − 𝑄)𝑁 are continuous. Now we turn to 󵄨󵄨 󵄨
show the fact that for any open bounded set Ω ⊂ 𝑋, denoted 󵄨󵄨Δ 𝑖 (𝑥 (𝑡) , 𝑡) − Δ 𝑖 (𝑥 (𝑠) , 𝑠)󵄨󵄨󵄨0
by 󵄨󵄨
󵄨 󵄨 󵄨 󵄨 󵄨 󵄨 󵄨󵄨 𝑛 𝑛
Ω = {𝑥 (𝑡) ∈ 𝑋 | 󵄨󵄨󵄨𝑥𝑖 (𝑡)󵄨󵄨󵄨1 = 󵄨󵄨󵄨𝑥𝑖 (𝑡)󵄨󵄨󵄨0 + 󵄨󵄨󵄨𝑥𝑖̇(𝑡)󵄨󵄨󵄨0 < ℎ𝑖 } , = 󵄨󵄨󵄨󵄨𝑟𝑖 (𝑡) − ∑𝑎𝑖𝑗 (𝑡) 𝑒𝛼𝑖𝑗 𝑥𝑗 (𝑡) − ∑ 𝑏𝑖𝑗 (𝑡) 𝑒𝛽𝑖𝑗 𝑥𝑗 (𝑡−𝜏𝑖𝑗 )
󵄨󵄨
(24) 󵄨 𝑗=1 𝑗=1

󵄨󵄨
the mapping 𝑁 is 𝐿-compact on Ω. Here, the constants ℎ𝑖 𝑛 𝑛 󵄨󵄨
are independent of the choice of 𝑥(𝑡). In view of Definition 1, − 𝑟𝑖 (𝑠) + ∑𝑎𝑖𝑗 (𝑠) 𝑒𝛼𝑖𝑗 𝑥𝑗 (𝑠) + ∑𝑏𝑖𝑗 (𝑠) 𝑒𝛽𝑖𝑗 𝑥𝑗 (𝑠−𝜏𝑖𝑗 ) 󵄨󵄨󵄨󵄨
󵄨󵄨
to show the above fact, it suffices to show that 𝑄𝑁(Ω) is
𝑗=1 𝑗=1 󵄨0
bounded and 𝐾𝑃 (𝐼 − 𝑄)𝑁 : Ω → 𝑋 is compact. We first 𝑛
󵄨 󵄨
󵄨 󵄨
arrive at ≤ 󵄨󵄨󵄨𝑟𝑖 (𝑡) − 𝑟𝑖 (𝑠)󵄨󵄨󵄨0 + ∑󵄨󵄨󵄨󵄨𝑎𝑖𝑗 (𝑡) 𝑒𝛼𝑖𝑗 𝑥𝑗 (𝑡) − 𝑎𝑖𝑗 (𝑠) 𝑒𝛼𝑖𝑗 𝑥𝑗 (𝑠) 󵄨󵄨󵄨󵄨0
󵄨󵄨 𝜔 󵄨󵄨 𝑗=1
󵄨󵄨󵄨(𝑄𝑁𝑥)𝑖 󵄨󵄨󵄨 = 󵄨󵄨󵄨 1 ∫ Δ 𝑖 (𝑥, 𝑡) 𝑑𝑡󵄨󵄨󵄨 ≤ 󵄨󵄨󵄨Δ 𝑖 (𝑥, 𝑡)󵄨󵄨󵄨 := 𝑀𝑖 ,
󵄨 󵄨0 󵄨󵄨 𝜔 0 󵄨󵄨 󵄨 󵄨0 𝑛
󵄨 󵄨0 (25) 󵄨 󵄨
+ ∑ 󵄨󵄨󵄨󵄨𝑏𝑖𝑗 (𝑡) 𝑒𝛽𝑖𝑗 𝑥𝑗 (𝑡−𝜏𝑖𝑗 ) − 𝑏𝑖𝑗 (𝑠) 𝑒𝛽𝑖𝑗 𝑥𝑗 (𝑠−𝜏𝑖𝑗 ) 󵄨󵄨󵄨󵄨0
∀𝑥 ∈ Ω, 𝑗=1
Abstract and Applied Analysis 5

𝑛
󵄨 󵄨 󵄨 󵄨󵄨 󵄨 𝑛 𝑛
≤ 󵄨󵄨󵄨𝑟𝑖 (𝑡) − 𝑟𝑖 (𝑠)󵄨󵄨󵄨 + ∑󵄨󵄨󵄨󵄨𝑎𝑖𝑗 (𝑡)󵄨󵄨󵄨󵄨0 󵄨󵄨󵄨󵄨𝑒𝛼𝑖𝑗 𝑥𝑗 (𝑡) − 𝑒𝛼𝑖𝑗 𝑥𝑗 (𝑠) 󵄨󵄨󵄨󵄨0 = [1 + ∑ (1 + 𝑎𝑖𝑗 ) 𝑒𝛼𝑖𝑗 ℎ𝑗 + ∑ (1 + 𝑏𝑖𝑗 ) 𝑒𝛽𝑖𝑗 ℎ𝑗 ] 𝜀 + 2𝑀𝑖 𝛿
𝑗=1
𝑗=1 𝑗=1
[ ]
𝑛
󵄨 󵄨󵄨 󵄨
+ ∑ 󵄨󵄨󵄨󵄨𝑎𝑖𝑗 (𝑡) − 𝑎𝑖𝑗 (𝑠)󵄨󵄨󵄨󵄨0 󵄨󵄨󵄨󵄨𝑒𝛼𝑖𝑗 𝑥𝑗 (𝑠) 󵄨󵄨󵄨󵄨0 𝑛 𝑛
𝑗=1 < [1 + ∑ (1 + 𝑎𝑖𝑗 ) 𝑒𝛼𝑖𝑗 ℎ𝑗 + ∑ (1 + 𝑏𝑖𝑗 ) 𝑒𝛽𝑖𝑗 ℎ𝑗 ] 𝜀 + 𝜀
𝑗=1 𝑗=1
𝑛 [ ]
󵄨 󵄨󵄨 󵄨
+ ∑ 󵄨󵄨󵄨󵄨𝑏𝑖𝑗 (𝑡)󵄨󵄨󵄨󵄨0 󵄨󵄨󵄨󵄨𝑒𝛽𝑖𝑗 𝑥𝑗 (𝑡−𝜏𝑖𝑗 ) − 𝑒𝛽𝑖𝑗 𝑥𝑗 (𝑠−𝜏𝑖𝑗 ) 󵄨󵄨󵄨󵄨0 ̃
:= 𝑀𝜀,
𝑗=1
(34)
𝑛
󵄨 󵄨󵄨 󵄨
+ ∑ 󵄨󵄨󵄨󵄨𝑏𝑖𝑗 (𝑡) − 𝑏𝑖𝑗 (𝑠)󵄨󵄨󵄨󵄨0 󵄨󵄨󵄨󵄨𝑒𝛽𝑖𝑗 𝑥𝑗 (𝑠−𝜏𝑖𝑗 ) 󵄨󵄨󵄨󵄨0 which implies that (𝐾𝑃 (𝐼 − 𝑄)𝑁𝑥)(Ω) is equicontinuous.
𝑗=1 Therefore, by the generalized Arzela-Ascoli theorem, we
𝑛 𝑛
have that (𝐾𝑃 (𝐼−𝑄)𝑁𝑥)(Ω) is relatively compact in the space
󵄨 󵄨 (𝑋, ‖ ⋅ ‖1 ). The proof of this step is complete.
< 𝜀 + ∑𝑎𝑖𝑗 𝑒𝛼𝑖𝑗 ℎ𝑗 󵄨󵄨󵄨𝑥𝑖 (𝑡) − 𝑥𝑖 (𝑠)󵄨󵄨󵄨0 + ∑ 𝑒𝛼𝑖𝑗 ℎ𝑗 𝜀
𝑗=1 𝑗=1
Step 2. In this step, we are in a position to search for
𝑛
󵄨 󵄨 𝑛 appropriate open bounded subsets Ω satisfying condition
+ ∑ 𝑏𝑖𝑗 𝑒𝛽𝑖𝑗 ℎ𝑗 󵄨󵄨󵄨󵄨𝑥𝑖 (𝑡 − 𝜏𝑖𝑗 ) − 𝑥𝑖 (𝑠 − 𝜏𝑖𝑗 )󵄨󵄨󵄨󵄨0 + ∑𝑒𝛽𝑖𝑗 ℎ𝑗 𝜀 (i) of Lemma 3. Specifically, our aim is to search for an
𝑗=1 𝑗=1 appropriate ℎ𝑖 defined by Ω in Step 1 such that Ω satisfies
𝑛 𝑛 𝑛 𝑛 condition (i) of Lemma 3. To this end, assume that 𝑥(𝑡) ∈ 𝑋
< 𝜀 + ∑𝑎𝑖𝑗 𝑒𝛼𝑖𝑗 ℎ𝑗 𝜀 + ∑ 𝑒𝛼𝑖𝑗 ℎ𝑗 𝜀 + ∑ 𝑏𝑖𝑗 𝑒𝛽𝑖𝑗 ℎ𝑗 𝜀 + ∑ 𝑒𝛽𝑖𝑗 ℎ𝑗 𝜀 is a solution of the equation 𝐿𝑥 = 𝜆𝑁𝑥 for each 𝜆 ∈ (0, 1);
𝑗=1 𝑗=1 𝑗=1 𝑗=1 that is,
𝑛 𝑛
𝑛 𝑛
𝑥𝑖̇(𝑡) = 𝜆 [𝑟𝑖 (𝑡) − ∑𝑎𝑖𝑗 (𝑡) 𝑒𝛼𝑖𝑗 𝑥𝑗 (𝑡) − ∑ 𝑏𝑖𝑗 (𝑡) 𝑒𝛽𝑖𝑗 𝑥𝑗 (𝑡−𝜏𝑖𝑗 ) ] ,
= [1 + ∑ (1 + 𝑎𝑖𝑗 ) 𝑒 𝛼𝑖𝑗 ℎ𝑗
+ ∑ (1 + 𝑏𝑖𝑗 ) 𝑒 𝛽𝑖𝑗 ℎ𝑗 ]
𝜀.
𝑗=1 𝑗=1
𝑗=1 𝑗=1 [ ]
[ ]
(31) 𝑖 = 1, 2, . . . , 𝑛.
(35)
Thus, it follows from (26) that Since 𝑥(𝑡) ∈ 𝑋, each 𝑥𝑖 (𝑡), 𝑖 = 1, 2, . . . , 𝑛, as components of
𝑥(𝑡), is continuously differentiable and 𝜔-periodic. In view of
continuity and periodicity, there exists 𝑡𝑖 ∈ [0, 𝜔] such that
󵄨󵄨 󵄨
󵄨󵄨(𝐾𝑃 (𝐼 − 𝑄) 𝑁𝑥)󸀠𝑖 (𝑡) − (𝐾𝑃 (𝐼 − 𝑄) 𝑁𝑥)󸀠𝑖 (𝑠)󵄨󵄨󵄨 𝑥𝑖 (𝑡𝑖 ) = max𝑡∈[0,𝜔] |𝑥𝑖 (𝑡)|, 𝑖 = 1, 2, . . . , 𝑛. Accordingly, 𝑥𝑖̇
(𝑡𝑖 ) =
󵄨 󵄨0 0 and we arrive at
󵄨󵄨 󵄨󵄨
= 󵄨󵄨Δ 𝑖 (𝑥 (𝑡) , 𝑡) − Δ 𝑖 (𝑥 (𝑠) , 𝑠)󵄨󵄨0 𝑛 𝑛
(32) 𝑟𝑖 (𝑡𝑖 ) − ∑𝑎𝑖𝑗 (𝑡𝑖 ) 𝑒𝛼𝑖𝑗 𝑥𝑗 (𝑡𝑖 ) − ∑ 𝑏𝑖𝑗 (𝑡𝑖 ) 𝑒𝛽𝑖𝑗 𝑥𝑗 (𝑡𝑖 −𝜏𝑖𝑗 ) = 0,
𝑛 𝑛 𝑗=1 𝑗=1 (36)
< [1 + ∑ (1 + 𝑎𝑖𝑗 ) 𝑒𝛼𝑖𝑗 ℎ𝑗 + ∑ (1 + 𝑏𝑖𝑗 ) 𝑒𝛽𝑖𝑗 ℎ𝑗 ] 𝜀.
𝑗=1 𝑗=1 𝑖 = 1, 2, . . . , 𝑛.
[ ]
That is,

On the other hand, the mean value theorem together with 𝑎𝑖𝑖 (𝑡𝑖 ) 𝑒𝛼𝑖𝑖 𝑥𝑖 (𝑡𝑖 ) + 𝑏𝑖𝑖 (𝑡𝑖 ) 𝑒𝛽𝑖𝑖 𝑥𝑖 (𝑡𝑖 −𝜏𝑖𝑖 )
(26) gives 𝑛
= 𝑟𝑖 (𝑡𝑖 ) − ∑ 𝑎𝑖𝑗 (𝑡𝑖 ) 𝑒𝛼𝑖𝑗 𝑥𝑗 (𝑡𝑖 )
𝑗=1,𝑗 ≠
𝑖 (37)
󵄨󵄨 󵄨
󵄨󵄨𝐾𝑃 (𝐼 − 𝑄)𝑁𝑥)𝑖 (𝑡) − (𝐾𝑃 (𝐼 − 𝑄) 𝑁𝑥)𝑖 (𝑠)󵄨󵄨󵄨0
𝑛
(33)
󵄩 󵄩 − ∑ 𝑏𝑖𝑗 (𝑡𝑖 ) 𝑒𝛽𝑖𝑗 𝑥𝑗 (𝑡𝑖 −𝜏𝑖𝑗 ) , 𝑖 = 1, 2, . . . 𝑛.
= 󵄩󵄩󵄩󵄩(𝐾𝑃 (𝐼 − 𝑄) 𝑁𝑥)𝑖 (𝜉)󵄩󵄩󵄩󵄩0 |𝑡 − 𝑠| ≤ 2𝑀𝑖 |𝑡 − 𝑠| ,
󸀠
𝑗=1,𝑗 ≠
𝑖

Noticing that 𝑥𝑗 (𝑡𝑗 ) = max𝑡∈[0,𝜔] |𝑥𝑗 (𝑡)| implies

where 𝜉 lies between 𝑡 and 𝑠. Taking 𝛿 = min{𝜀/2𝑀𝑖 , 𝛿2 }, it
𝑥𝑗 (𝑡𝑖 ) ≤ 𝑥𝑗 (𝑡𝑗 ) , 𝑥𝑗 (𝑡𝑖 − 𝜏𝑖𝑗 ) ≤ 𝑥𝑗 (𝑡𝑗 ) . (38)
follows from (32) and (33) that |𝑡 − 𝑠| < 𝛿 implies
It follows from (H2 ) and (37) that
󵄨 𝑎𝑖𝑖 𝑒𝛼𝑖𝑖 𝑥𝑖 (𝑡𝑖 ) + 𝑏𝑖𝑖 𝑒𝛼𝑖𝑖 𝑥𝑖 (𝑡𝑖 )
|(𝐾𝑃 (𝐼 − 𝑄)𝑁𝑥)𝑖 (𝑡) − (𝐾𝑃 (𝐼 − 𝑄) 𝑁𝑥)𝑖 (𝑠)󵄨󵄨󵄨1
󵄨 󵄨 󵄨
= |(𝐾𝑃 (𝐼 − 𝑄)𝑁𝑥)𝑖 (𝑡) − (𝐾𝑃 (𝐼 − 𝑄) 𝑁𝑥)𝑖 (𝑠)󵄨󵄨󵄨0 ≤ 󵄨󵄨󵄨󵄨𝑎𝑖𝑖 (𝑡𝑖 ) 𝑒𝛼𝑖𝑖 𝑥𝑖 (𝑡𝑖 ) + 𝑏𝑖𝑖 (𝑡𝑖 ) 𝑒𝛼𝑖𝑖 𝑥𝑖 (𝑡𝑖 ) 󵄨󵄨󵄨󵄨
󵄨 󵄨 󵄨 󵄨
+ 󵄨󵄨󵄨󵄨(𝐾𝑃 (𝐼 − 𝑄) 𝑁𝑥)𝑖 (𝑡) − (𝐾𝑃 (𝐼 − 𝑄) 𝑁𝑥)𝑖 (𝑠)󵄨󵄨󵄨󵄨0 ≤ 󵄨󵄨󵄨󵄨𝑎𝑖𝑖 (𝑡𝑖 ) 𝑒𝛼𝑖𝑖 𝑥𝑖 (𝑡𝑖 ) + 𝑏𝑖𝑖 (𝑡𝑖 ) 𝑒𝛽𝑖𝑖 𝑥𝑖 (𝑡𝑖 −𝜏𝑖𝑖 ) 󵄨󵄨󵄨󵄨
󸀠 󸀠
6 Abstract and Applied Analysis

󵄨󵄨
󵄨󵄨 𝑛
Then it follows from (43) and (44) that
= 󵄨󵄨󵄨󵄨𝑟𝑖 (𝑡𝑖 ) − ∑ 𝑎𝑖𝑗 (𝑡𝑖 ) 𝑒𝛼𝑖𝑗 𝑥𝑗 (𝑡𝑖 )
󵄨󵄨 (𝑧1 (𝑡1 ) , 𝑧2 (𝑡2 ) , . . . , 𝑧𝑛 (𝑡𝑛 )) ≤ 𝐻,
𝑇
󵄨 𝑗=1,𝑗 ≠
𝑖

󵄨󵄨 (45)
𝑛
𝛽𝑖𝑗 𝑥𝑗 (𝑡𝑖 −𝜏𝑖𝑗 ) 󵄨󵄨
󵄨 or 𝑧𝑖 (𝑡𝑖 ) ≤ ̃ℎ𝑖 , 𝑖 = 1, 2, . . . , 𝑛
− ∑ 𝑏𝑖𝑗 (𝑡𝑖 ) 𝑒 󵄨󵄨
󵄨󵄨
𝑗=1,𝑗 ≠
𝑖 󵄨󵄨 which implies
𝑛 𝑛 ̃ℎ
≤ 𝑟𝑖 + ∑ 𝑎𝑖𝑗 𝑒𝛼𝑖𝑗 𝑥𝑗 (𝑡𝑖 ) + ∑ 𝑏𝑖𝑗 𝑒𝛽𝑖𝑗 𝑥𝑗 (𝑡𝑖 −𝜏𝑖𝑗 ) 󵄨󵄨 󵄨 󵄨 󵄨 1
󵄨󵄨𝑥𝑖 (𝑡)󵄨󵄨󵄨0 = max 󵄨󵄨󵄨𝑥𝑖 (𝑡)󵄨󵄨󵄨 = 𝑥𝑖 (𝑡𝑖 ) ≤ ln 𝑖
,
𝑗=1,𝑗 ≠
𝑖 𝑗=1,𝑗 ≠
𝑖 𝑡∈[0,𝜔] 𝛼𝑖𝑖 𝑎𝑖𝑖 + 𝑏𝑖𝑖 (46)
(39)
𝑛 𝑛 𝑖 = 1, 2, . . . , 𝑛.
≤ 𝑟𝑖 + ∑ 𝑎𝑖𝑗 𝑒𝛼𝑖𝑗 𝑥𝑗 (𝑡𝑗 ) + ∑ 𝑏𝑖𝑗 𝑒𝛽𝑖𝑗 𝑥𝑗 (𝑡𝑗 −𝜏𝑖𝑗 )
𝑗=1,𝑗 ≠
𝑖 𝑗=1,𝑗 ≠
𝑖 On the other hand, it follows from (44) that
𝑛 𝑛 (𝐸 − K) 𝐻 = 𝐷, or 𝐻 = K𝐻 + 𝐷,
≤ 𝑟𝑖 + ∑ 𝑎𝑖𝑗 𝑒𝛼𝑗𝑗 𝑥𝑗 (𝑡𝑗 ) + ∑ 𝑏𝑖𝑗 𝑒𝛼𝑗𝑗 𝑥𝑗 (𝑡𝑗 )
𝑛 (47)
𝑗=1,𝑗 ≠
𝑖 𝑗=1,𝑗 ≠
𝑖
that is ̃ℎ𝑖 = ∑ Γ𝑖𝑗 ̃ℎ𝑗 + 𝐷𝑖 , 𝑖 = 1, 2, . . . , 𝑛.
𝑛 𝑗=1
= 𝑟𝑖 + ∑ (𝑎𝑖𝑗 + 𝑏𝑖𝑗 ) 𝑒𝛼𝑗𝑗 𝑥𝑗 (𝑡𝑗 ) .
𝑗=1,𝑗 ≠
𝑖 Estimating (2), by using (45) and (47), we have
󵄨󵄨 󵄨󵄨
󵄨󵄨 ̇ 󵄨󵄨 󵄨󵄨 𝑛 𝑛
𝛽𝑖𝑗 𝑥𝑗 (𝑡−𝜏𝑖𝑗 ) 󵄨󵄨
󵄨
Here we used (H2 ). Letting (𝑎𝑖𝑖 + 𝑏𝑖𝑖 )𝑒𝛼𝑖𝑖 𝑥𝑖 (𝑡𝑖 ) = 𝑧𝑖 (𝑡𝑖 ), it follows 𝑥 (𝑡)
󵄨󵄨 𝑖 󵄨󵄨0 = 𝜆 󵄨
󵄨 𝑟 (𝑡) − ∑ 𝑎 (𝑡) 𝑒𝛼𝑖𝑗 𝑥𝑗 (𝑡)
− ∑ 𝑏 (𝑡) 𝑒 󵄨󵄨
󵄨󵄨 𝑖 𝑖𝑗 𝑖𝑗 󵄨󵄨
from (39) that 󵄨󵄨 󵄨󵄨0
󵄨 𝑗=1 𝑗=1

𝑛 𝑛 𝑛
−1 󵄨 󵄨 󵄨 󵄨
𝑧𝑖 (𝑡𝑖 ) ≤ 𝑟𝑖 + ∑ (𝑎𝑖𝑗 + 𝑏𝑖𝑗 ) (𝑎𝑗𝑗 + 𝑏𝑗𝑗 ) 𝑧𝑗 (𝑡𝑗 ) (40) ≤ 𝑟𝑖 + ∑ 𝑎𝑖𝑗 󵄨󵄨󵄨󵄨𝑒𝛼𝑖𝑗 𝑥𝑗 (𝑡) 󵄨󵄨󵄨󵄨0 + ∑𝑏𝑖𝑗 󵄨󵄨󵄨󵄨𝑒𝛽𝑖𝑗 𝑥𝑗 (𝑡−𝜏𝑖𝑗 ) 󵄨󵄨󵄨󵄨0
𝑗=1,𝑗 ≠
𝑖 𝑗=1 𝑗=1

or 𝑛 𝑛
≤ 𝑟𝑖 + ∑ 𝑎𝑖𝑗 𝑒𝛼𝑗𝑗 𝑥𝑗 (𝑡𝑗 ) + ∑𝑏𝑖𝑗 𝑒𝛼𝑗𝑗 𝑥𝑗 (𝑡𝑗 )
𝑛 𝑎𝑖𝑗 + 𝑏𝑖𝑗 𝑗=1 𝑗=1
𝑧𝑖 (𝑡𝑖 ) − ∑ 𝑧𝑗 (𝑡𝑗 ) ≤ 𝑟𝑖 , (41)
𝑖 𝑎𝑗𝑗
𝑗=1,𝑗 ≠ + 𝑏𝑗𝑗 𝑛 𝑎𝑖𝑗 + 𝑏𝑖𝑗
= 𝑟𝑖 + ∑ 𝑧𝑗 (𝑡𝑗 )
which implies 𝑗=1 𝑎𝑗𝑗 + 𝑏𝑗𝑗

𝑎12 + 𝑏12 𝑎 + 𝑏1𝑛

𝑛 𝑎𝑖𝑗 + 𝑏𝑖𝑗 𝑎𝑖𝑖 + 𝑏𝑖𝑖
1 − ⋅ ⋅ ⋅ − 1𝑛 = 𝑟𝑖 + ∑ 𝑧𝑗 (𝑡𝑗 ) + 𝑧 (𝑡 )
𝑎22 + 𝑏22 𝑎𝑛𝑛 + 𝑏𝑛𝑛 𝑖 𝑎𝑗𝑗 + 𝑏𝑗𝑗
𝑗=1,𝑗 ≠ 𝑎𝑖𝑖 + 𝑏𝑖𝑖 𝑖 𝑖
( )
( 𝑎 +𝑏 𝑎2𝑛 + 𝑏2𝑛 ) 𝑛
𝑎𝑖𝑖 + 𝑏𝑖𝑖
(− 21 )
≤ 𝐷𝑖 + ∑ Γ𝑖𝑗 ̃ℎ𝑗 +
21
( 1 ... − ) 𝑧 (𝑡 )
( 𝑎11 + 𝑏11
(
𝑎𝑛𝑛 + 𝑏𝑛𝑛 )
) 𝑗=1 𝑎𝑖𝑖 + 𝑏𝑖𝑖 𝑖 𝑖
( )
( ... ⋅⋅⋅ ⋅⋅⋅ ⋅⋅⋅ )
( ) 𝑎𝑖𝑖 + 𝑏𝑖𝑖 ̃
( ) (42) ≤ ̃ℎ𝑖 + ℎ
𝑎𝑖𝑖 + 𝑏𝑖𝑖 𝑖
𝑎𝑛1 + 𝑏𝑛1 𝑎𝑛2 + 𝑏𝑛2
− − ⋅⋅⋅ 1
( 𝑎11 + 𝑏11 𝑎22 + 𝑏22 ) = [1 +
𝑎𝑖𝑖 + 𝑏𝑖𝑖 ̃
]ℎ .
𝑎𝑖𝑖 + 𝑏𝑖𝑖 𝑖
𝑧1 (𝑡1 ) 𝑟1
(48)
𝑧2 (𝑡2 ) 𝑟
×( ) ≤ ( 2).
⋅⋅⋅ ... We can choose a large enough real number (𝑑 > 1) such that
𝑧𝑛 (𝑡𝑛 ) 𝑟𝑛
1 𝑑̃ℎ𝑖 1 ̃ℎ
𝑖 𝑎 + 𝑏𝑖𝑖 ̃
𝑇 𝑇 ln > ln + [1 + 𝑖𝑖 ]ℎ . (49)
Set 𝐷 = (𝐷1 , 𝐷2 , . . . , 𝐷𝑛 ) = (𝑟1 , 𝑟2 , . . . , 𝑟𝑛 ) . It follows from 𝛼𝑖𝑖 𝑎𝑖𝑖 + 𝑏𝑖𝑖 𝛼𝑖𝑖 𝑎𝑖𝑖 + 𝑏𝑖𝑖 𝑎𝑖𝑖 + 𝑏𝑖𝑖 𝑖
(42) that
Set ℎ𝑖 = (1/𝛼𝑖𝑖 ) ln(𝑑̃ℎ𝑖 /(𝑎𝑖𝑖 + 𝑏𝑖𝑖 )). Then for any solution of
𝑇
(𝐸 − K) (𝑧1 (𝑡1 ) , 𝑧2 (𝑡2 ) , . . . , 𝑧𝑛 (𝑡𝑛 )) ≤ 𝐷. (43) 𝐿𝑥 = 𝜆𝑁𝑥, we have
󵄨󵄨 󵄨 󵄨 󵄨 󵄨 󵄨
󵄨󵄨𝑥𝑖 (𝑡)󵄨󵄨󵄨1 = 󵄨󵄨󵄨𝑥𝑖 (𝑡)󵄨󵄨󵄨0 + 󵄨󵄨󵄨𝑥𝑖̇(𝑡)󵄨󵄨󵄨0
In view of 𝜌(K) < 1 and Lemma 5, (𝐸𝑛 − K)−1 ≥ 0. Let
1 ̃ℎ 𝑎 + 𝑏𝑖𝑖 ̃ (50)
𝑖
𝑇 ≤ ln + [1 + 𝑖𝑖 ] ℎ𝑖 < ℎ𝑖 ,
𝐻 = (̃ℎ1 , ̃ℎ2 , . . . , ̃ℎ𝑛 ) := (𝐸 − K)−1 𝐷 ≥ 0. (44) 𝛼𝑖𝑖 𝑎𝑖𝑖 + 𝑏𝑖𝑖 𝑎𝑖𝑖 + 𝑏𝑖𝑖
Abstract and Applied Analysis 7

for all 𝑖 = 1, 2, . . . , 𝑛. Obviously, ℎ𝑖 are independent In view of (51), we get

of 𝜆 and the choice of 𝑥(𝑡). Consequently, taking ℎ𝑖 = 󵄨󵄨 󵄨󵄨 󵄨󵄨 󵄨󵄨 󵄨 󵄨
(1/𝛼𝑖𝑖 ) ln(𝑑̃ℎ𝑖 /(𝑎𝑖𝑖 + 𝑏𝑖𝑖 )), the open subset Ω satisfies that 󵄨󵄨𝑦𝑖 󵄨󵄨 = 󵄨󵄨𝑦𝑖 󵄨󵄨0 = 󵄨󵄨󵄨𝑦𝑖 󵄨󵄨󵄨1
𝐿𝑥 ≠𝜆𝑁𝑥 for each 𝜆 ∈ (0, 1), 𝑥 ∈ 𝜕Ω ∩ Dom 𝐿; that is, the
open subset Ω satisfies assumption (i) of Lemma 3. = [𝑚 (𝑎𝑖𝑖 ) + 𝑚 (𝑏𝑖𝑖 )] 𝑒𝛼𝑖𝑖 ℎ𝑖
̃ (57)
Step 3. In what follows, we verify that for the given open = [𝑚 (𝑎𝑖𝑖 ) + 𝑚 (𝑏𝑖𝑖 )] 𝑒ln(𝑑ℎ𝑖 /(𝑎𝑖𝑖 +𝑏𝑖𝑖 ))
bounded set Ω, assumption (ii) of Lemma 3 also holds. That
is, for each 𝑥 ∈ 𝜕Ω ∩ Ker 𝐿, 𝑄𝑁𝑥 ≠0 and deg{𝐽𝑄𝑁, Ω ∩ 𝑑̃ℎ𝑖
= [𝑚 (𝑎𝑖𝑖 ) + 𝑚 (𝑏𝑖𝑖 )] , ∀𝑖 = 1, 2, . . . , 𝑛.
Ker 𝐿, 0} ≠0. 𝑎𝑖𝑖 + 𝑏𝑖𝑖
Take 𝑥 ∈ 𝜕Ω ∩ Ker 𝐿. Then, in view of Ker 𝐿 = R𝑛 , 𝑥 is
a constant vector in R𝑛 , denoted by 𝑥 = (𝑥1 , 𝑥2 , . . . , 𝑥𝑛 )𝑇 and Note that 𝑓 ≤ 𝑚(𝑓) ≤ 𝑓. It follows from (2), (57), and (47)
by the property that
󵄨󵄨 󵄨󵄨 󵄨󵄨 󵄨󵄨 󵄨 󵄨
󵄨󵄨𝑥𝑖 󵄨󵄨 = 󵄨󵄨𝑥𝑖 󵄨󵄨0 = 󵄨󵄨󵄨𝑥𝑖 󵄨󵄨󵄨1 = ℎ𝑖 , ∀𝑖 = 1, 2, . . . , 𝑛. (51) 𝑑̃ℎ𝑘 󵄨 󵄨
𝑑̃ℎ𝑘 ≤ [𝑚 (𝑎𝑘𝑘 ) + 𝑚 (𝑏𝑘𝑘 )] = 󵄨󵄨𝑦 󵄨󵄨
Operate 𝑥 by 𝑄𝑁, and we obtain that, for 𝑖 = 1, 2, . . . , 𝑛, (𝑎𝑘𝑘 + 𝑏𝑘𝑘 ) 󵄨 𝑘 󵄨
󵄨󵄨 󵄨󵄨
𝑛 𝑛 󵄨󵄨 𝑛 𝑛
𝛽𝑘𝑗 𝑥𝑗 󵄨󵄨
󵄨
(𝑄𝑁𝑥)𝑖 = 𝑚 (𝑟𝑖 ) − ∑ 𝑚 (𝑎𝑖𝑗 ) 𝑒 𝛼𝑖𝑗 𝑥𝑗
− ∑𝑚 (𝑏𝑖𝑗 ) 𝑒 𝛽𝑖𝑗 𝑥𝑗
, 󵄨
≤ 󵄨󵄨󵄨𝑚 (𝑟𝑘 ) − ∑ 𝑚 (𝑎𝑘𝑗 ) 𝑒𝛼𝑘𝑗 𝑥𝑗
− ∑ 𝑚 (𝑏𝑘𝑗 ) 𝑒 󵄨󵄨
󵄨󵄨 󵄨󵄨
𝑗=1 𝑗=1 (52) 󵄨 𝑗=1,𝑗 ≠
𝑘 𝑗=1,𝑗 ≠
𝑘 󵄨󵄨
𝑖 = 1, 2, . . . , 𝑛. 󵄨󵄨 󵄨󵄨 𝛼𝑘𝑗 /𝛼𝑗𝑗
𝑛 󵄨󵄨𝑦𝑗 󵄨󵄨
≤ 𝑚 (𝑟𝑘 ) + ∑ 𝑚 (𝑎𝑘𝑗 ) ( 󵄨 󵄨 )
We claim that |(𝑄𝑁𝑥)𝑖 | > 0, for 𝑖 = 1, 2, . . . , 𝑛. If this is not 𝑚 (𝑎𝑗𝑗 ) + 𝑚 (𝑏𝑗𝑗 )
𝑗=1,𝑗 ≠
𝑘
valid, suppose that there exists a certain 𝑘 ∈ {1, 2, . . . , 𝑛} such
that |(𝑄𝑁𝑥)𝑘 | = 0; that is, 󵄨󵄨 󵄨󵄨 𝛽𝑘𝑗 /𝛼𝑗𝑗
𝑛 󵄨󵄨𝑦𝑗 󵄨󵄨
𝑛 𝑛 + ∑ 𝑚 (𝑏𝑘𝑗 ) ( 󵄨 󵄨 )
𝑚 (𝑟𝑘 ) − ∑𝑚 (𝑎𝑘𝑗 ) 𝑒𝛼𝑘𝑗 𝑥𝑗 − ∑𝑚 (𝑏𝑘𝑗 ) 𝑒𝛽𝑘𝑗 𝑥𝑗 = 0, (53) 𝑗=1,𝑗 ≠
𝑘 𝑚 (𝑎𝑗𝑗 ) + 𝑚 (𝑏𝑗𝑗 )
𝑗=1 𝑗=1
󵄨󵄨 󵄨󵄨
𝑛 󵄨󵄨𝑦𝑗 󵄨󵄨
or ≤ 𝑟𝑘 + ∑ 𝑚 (𝑎𝑘𝑗 ) 󵄨 󵄨
𝑚 (𝑎𝑗𝑗 + 𝑚 (𝑏𝑗𝑗 )
)
𝑚 (𝑎𝑘𝑘 ) 𝑒𝛼𝑘𝑘 𝑥𝑘 + 𝑚 (𝑏𝑘𝑘 ) 𝑒𝛽𝑘𝑘 𝑥𝑘 𝑗=1,𝑗 ≠
𝑘

󵄨󵄨 󵄨󵄨
𝑛 𝑛 󵄨󵄨𝑦𝑗 󵄨󵄨
𝛼𝑘𝑗 𝑥𝑗 + ∑ 𝑚 (𝑏𝑘𝑗 ) 󵄨 󵄨
= 𝑚 (𝑟𝑘 ) − ∑ 𝑚 (𝑎𝑘𝑗 ) 𝑒
𝑗=1,𝑗 ≠
𝑘 (54) 𝑗=1,𝑗 ≠
𝑘 𝑚 (𝑎𝑗𝑗 + 𝑚 (𝑏𝑗𝑗 )
)

𝑛 𝑛 𝑚 (𝑎𝑘𝑗 ) + 𝑚 (𝑏𝑘𝑗 ) 󵄨󵄨 󵄨󵄨
− ∑ 𝑚 (𝑏𝑘𝑗 ) 𝑒𝛽𝑘𝑗 𝑥𝑗 . = 𝑟𝑘 + ∑ 󵄨󵄨𝑦𝑗 󵄨󵄨
𝑗=1,𝑗 ≠
𝑘 𝑗=1,𝑗 ≠
𝑘 𝑚 (𝑎 ) + 𝑚 (𝑏 ) 󵄨 󵄨
𝑗𝑗 𝑗𝑗

That is, 𝑛 𝑚 (𝑎𝑘𝑗 ) + 𝑚 (𝑏𝑘𝑗 )

𝑛 ≤ 𝑟𝑘 + ∑
[𝑚 (𝑎𝑘𝑘 ) + 𝑚 (𝑏𝑘𝑘 )] 𝑒𝛼𝑘𝑘 𝑥𝑘 = 𝑚 (𝑟𝑘 ) − ∑ 𝑚 (𝑎𝑘𝑗 ) 𝑒𝛼𝑘𝑗 𝑥𝑗 𝑘 𝑚 (𝑎𝑗𝑗 )
𝑗=1,𝑗 ≠ + 𝑚 (𝑏𝑗𝑗 )
𝑗=1,𝑗 ≠
𝑘
̃
𝑑ℎ𝑗
𝑛 × [𝑚 (𝑎𝑗𝑗 ) + 𝑚 (𝑏𝑗𝑗 )]
− ∑ 𝑚 (𝑏𝑘𝑗 ) 𝑒𝛽𝑘𝑗 𝑥𝑗 . (𝑎𝑗𝑗 + 𝑏𝑗𝑗 )
𝑗=1,𝑗 ≠
𝑘
(55) 𝑛 𝑎𝑘𝑗 + 𝑏𝑘𝑗
≤ 𝐷𝑘 + ∑ ̃
𝑑ℎ𝑗
𝛼𝑘𝑘 𝑥𝑘 𝑎 + 𝑏𝑗𝑗
Letting [𝑚(𝑎𝑘𝑘 ) + 𝑚(𝑏𝑘𝑘 )]𝑒 = 𝑦𝑘 , we have 𝑘 𝑗𝑗
𝑗=1,𝑗 ≠

𝛼𝑘𝑗 /𝛼𝑗𝑗
𝑦𝑗 𝑛 𝑎𝑘𝑗 + 𝑏𝑘𝑗
𝑒𝛼𝑘𝑗 𝑥𝑗 = ( ) , < 𝑑𝐷𝑘 + 𝑑 ∑ ̃
ℎ𝑗
𝑚 (𝑎𝑗𝑗 ) + 𝑚 (𝑏𝑗𝑗 ) 𝑎 + 𝑏𝑗𝑗
𝑘 𝑗𝑗
𝑗=1,𝑗 ≠

𝛽𝑘𝑗 /𝛼𝑗𝑗 𝑛
𝑦𝑗
𝑒𝛽𝑘𝑗 𝑥𝑗
=( ) , (56) < 𝑑 [𝐷𝑘 + ∑ Γ𝑘𝑗 ℎ ̃]
𝑗
𝑚 (𝑎𝑗𝑗 ) + 𝑚 (𝑏𝑗𝑗 ) 𝑗=1,𝑗 ≠
𝑘
[ ]
𝑦𝑗
𝛼𝑘𝑗 ≤ 𝛼𝑗𝑗 , 𝛽𝑘𝑗 ≤ 𝛼𝑗𝑗 , > 1. = 𝑑ℎ̃𝑘 ,
𝑚 (𝑎𝑗𝑗 ) (58)
8 Abstract and Applied Analysis

which is a contradiction. Therefore, for any 𝑥 ∈ 𝜕Ω ∩ Ker 𝐿, Lemma 10 (see [23]). If 𝑥, 𝑦 > 1, and 𝛼𝑗𝑖 ≤ 𝛼𝑖𝑖 , (𝑖, 𝑗 =
|(𝑄𝑁𝑥)𝑖 | > 0 for all 𝑖 = 1, 2, . . . , 𝑛. That is, (𝑄𝑁𝑥) ≠0, for 𝑥 ∈ 1, 2, . . . , 𝑛), then
𝜕Ω ∩ Ker 𝐿. Furthermore, in view of (H1 ) and Definition 2, it
is easy to see that 󵄨󵄨 𝛼𝑗𝑖 𝛼 󵄨 󵄨 𝛼 𝛼 󵄨
󵄨󵄨𝑥 − 𝑦 𝑗𝑖 󵄨󵄨󵄨 ≤ 󵄨󵄨󵄨𝑥 𝑖𝑖 − 𝑦 𝑖𝑖 󵄨󵄨󵄨 , 𝑖, 𝑗 = 1, 2, . . . , 𝑛. (62)
deg {𝐽𝑄𝑁, Ω ∩ Ker 𝐿, 0} ≠0, (59)
Theorem 11. Assume that 𝛼𝑖𝑗 = 𝛽𝑖𝑗 ; if (H1 )–(H4 ) hold, then
where deg(⋅) is the Brouwer degree and 𝐽 is the identity ̃ which
system (3) has a unique positive 𝜔-periodic solution 𝑦(𝑡)
mapping since Im 𝑄 = Ker 𝐿. is globally asymptotically stable.
So far, we have shown that the open subset Ω ⊂ 𝑋
satisfies all the assumptions of Lemma 3. Hence, by Lemma 3, Proof. By Lemma 9, system (3) is bounded below. Thus there
system (14) has at least one positive 𝜔-periodic solution in exist positive constants 𝑚𝑖 > 0 such that 𝑦𝑖 (𝑡) ≥ 𝑚𝑖 . We
Dom 𝐿 ∩ Ω. By (13), system (3) has at least one positive 𝜔- proceed the proof of this theorem with two steps.
̃ This completes the proof
periodic solution, denoted by 𝑦(𝑡).
of Theorem 6. Step 1. Choose positive constants 𝑑𝑖 (0 < 𝑑𝑖 < 𝑚𝑖 ), 𝑖 = 1,
2, . . . , 𝑛, such that 𝜌(K) = 𝜌(H), where H = (ℎ𝑖𝑗 )𝑛×𝑛 and
3. Globally Asymptotic Stability
Under the assumption of Theorem 6, we know that system 0, 𝑖 = 𝑗,
ℎ𝑖𝑗 = { −1 𝛼 −𝛼 (63)
(3) has at least one positive 𝜔-periodic solution, denoted by (𝑎𝑗𝑗 + 𝑏𝑗𝑗 ) (𝑎𝑖𝑗 + 𝑏𝑖𝑗 ) 𝑑𝑗 𝑖𝑗 𝑗𝑗 , 𝑖 ≠𝑗.
̃ = (𝑦̃1 (𝑡), . . . , 𝑦̃𝑛 (𝑡))𝑇 . The aim of this section is to derive
𝑦(𝑡)
a set of sufficient conditions which guarantee the global
asymptotic stability of the positive 𝜔-periodic solution 𝑦∗ (𝑡). We claim that the positive constants 𝑑𝑖 can be definitely
As pointed out in Section 1, because 𝑦𝑗 (𝑡) has been changed to chosen. By similar arguments in [23], one can prove this fact.
𝛼
𝑦𝑗 𝑖𝑗 (𝑡) in (3), the previous method in Xia and Han [8] cannot Step 2. Prove the global asymptotic stability of system (3).
be applied to study the stability of system (3) directly. Before Changing variables 𝑤𝑖 = 𝑦𝑖 /𝑑𝑖 , by noticing that 𝛽𝑖𝑗 = 𝛼𝑖𝑗 ,
the formal analysis, we recall some facts which will be used then system (3) changes to
in the proof.

Lemma 7 (see [21]). Let 𝑓 be a nonnegative function defined 𝑛

𝛼 𝛼
on [0, +∞] such that 𝑓 is integrable on [0, +∞] and is 𝑤̇ [
𝑖 (𝑡) = 𝑤𝑖 (𝑡) 𝑟𝑖 (𝑡) − ∑ 𝑎𝑖𝑗 (𝑡) 𝑑𝑗 𝑤𝑗 (𝑡)
𝑖𝑗 𝑖𝑗

uniformly continuous on [0, +∞]. Then lim𝑡 → +∞ 𝑓(𝑡) = 0. [ 𝑗=1

Lemma 8 (see [19, 20]). Let Q = (𝑞𝑖𝑗 )𝑛×𝑛 be a matrix with 𝑛 (64)
𝛼 𝛼
nonpositive off-diagonal elements. Q is an 𝑀-matrix if and only − ∑𝑏𝑖𝑗 (𝑡) 𝑑𝑗 𝑖𝑗 𝑤𝑗 𝑖𝑗 (𝑡 − 𝜏𝑖𝑗 )] ,
if there exists a positive diagonal matrix Ξ = diag(𝜉1 , 𝜉2 . . . , 𝜉𝑛 ) 𝑗=1
]
such that
𝑖 = 1, 2, . . . , 𝑛.
𝜉𝑖 𝑞𝑖𝑖 > ∑ 𝜉𝑗 𝑞𝑗𝑖 , 𝑖 = 1, 2, . . . , 𝑛. (60)
𝑗 ≠
𝑖
In view of (H1 )–(H3 ), system (3) has at least a positive
We need a lemma which can follow immediately from periodic solution. By the linear transformation 𝑤𝑖 = 𝑦𝑖 /𝑑𝑖 ,
Theorem 2.1 in Xia et al. [22]. We consider the following we know that system (64) also has at least a positive periodic
logistic equation: solution, denoted by 𝑤 ̃(𝑡) = (̃ ̃𝑛 (𝑡)). In order to
𝑤1 (𝑡), . . . , 𝑤
show the global asymptotic stability of system (3), it suffices to
𝛼
𝑥𝑖̇(𝑡) = 𝑥𝑖 (𝑡) [𝑟𝑖 (𝑡) − 𝑎𝑖𝑖 (𝑡) 𝑥𝑖 𝑖𝑖 (𝑡)] . (61) show that 𝑤̃(𝑡) of system (64) is globally asymptotically stable.
For this purpose, let 𝑤(𝑡) = (𝑤1 (𝑡), . . . , 𝑤𝑛 (𝑡)) be any other
Since 𝑟𝑖 (𝑡) is nonnegative and 𝑎𝑖𝑖 (𝑡) are strictly positive, it positive solution of system (64). And we define a Lyapunov
follows immediately from Lemma 2.1 in Xia et al. [22] that functional 𝑉(𝑡) as follows:
system (61) has a unique positive solution, denoted by 𝑋𝑖 (𝑡),
with 𝑋𝑖 (𝑡) ≤ (𝑟𝑖 /𝑎𝑖𝑖 )1/𝛼𝑖𝑖 , which is globally asymptotically 𝑛
󵄨 ̃𝑖 (𝑡)󵄨󵄨󵄨󵄨
stable. Then, as a special case of Theorem 2.1 in [22], we have 𝑉 (𝑡) = ∑𝜉𝑖 [ 󵄨󵄨󵄨ln 𝑤𝑖 (𝑡) − ln 𝑤
the following. 𝑖=1

𝑛 𝑡
Lemma 9. Suppose that 𝑖𝑗
(65)
+ ∑ ∫ 𝑏𝑖𝑗 (𝑠 + 𝜏𝑖𝑗 ) 𝑑𝑗𝛼
𝛼 𝑖 𝑡−𝜏𝑖𝑗
𝑗=1,𝑗 ≠
(H4 ) 𝑚(𝑟𝑖 (𝑡) − ∑𝑛𝑗=1,𝑗 ≠
𝑖 [𝑎𝑖𝑗 (𝑡) + 𝑏𝑖𝑗 (𝑡 + 𝜏𝑖𝑗 )]𝑋𝑗 𝑖𝑗 (𝑡)) > 0;
󵄨󵄨 𝛼𝑖𝑗 󵄨
̃𝑗 𝑖𝑗 (𝑠)󵄨󵄨󵄨󵄨 𝑑𝑠] .
𝛼
then system (3) is bounded above and below. × 󵄨󵄨𝑤𝑗 (𝑠) − 𝑤
󵄨
Abstract and Applied Analysis 9

Calculating the upper right derivative of 𝑉(𝑡) along (64), it Note that 𝑤𝑖 (𝑡), 𝑤̃𝑖 (𝑡) ≥ 𝑚𝑖 /𝑑𝑖 > 1 and 𝛼𝑗𝑖 ≤ 𝛼𝑖𝑖 . By Lemma 10,
follows from (65) that we have
󵄨󵄨 𝛼𝑗𝑖 󵄨
̃𝑖 𝑗𝑖 (𝑡)󵄨󵄨󵄨󵄨 ≤ 󵄨󵄨󵄨󵄨𝑤𝑖𝛼𝑖𝑖 (𝑡) − 𝑤
̃𝑖𝛼𝑖𝑖 (𝑡)󵄨󵄨󵄨󵄨 .
𝛼
󵄨󵄨󵄨𝑤𝑖 (𝑡) − 𝑤 (67)
𝐷+ 𝑉 (𝑡) It follows from (66) and (67) that
𝑛
𝛼 󵄨 𝛼
̃𝑖𝛼𝑖𝑖 (𝑡)󵄨󵄨󵄨󵄨
𝑛
󵄨 ̃𝑖 (𝑡)󵄨󵄨󵄨󵄨
= ∑𝜉𝑖 sgn 󵄨󵄨󵄨ln 𝑤𝑖 (𝑡) − ln 𝑤 𝐷+ 𝑉 (𝑡) ≤ −∑𝜉𝑖 [𝑎𝑖𝑖 (𝑡) + 𝑏𝑖𝑖 (𝑡)] 𝑑𝑖 𝑖𝑖 󵄨󵄨󵄨𝑤𝑖 𝑖𝑖 (𝑡) − 𝑤
𝑖=1
𝑖=1
𝑛 𝑛
𝛼 󵄨 𝛼
𝑤󸀠 (𝑡) 𝑤
̃󸀠 (𝑡) ̃𝑖𝛼𝑖𝑖 (𝑡)󵄨󵄨󵄨󵄨
+ ∑ ∑ 𝜉𝑗 𝑎𝑗𝑖 (𝑡) 𝑑𝑖 𝑗𝑖 󵄨󵄨󵄨𝑤𝑖 𝑖𝑖 (𝑡) − 𝑤
×[ 𝑖 − 𝑖 𝑖=1 𝑗=1,𝑗 ≠
𝑖
̃𝑖 (𝑡)
𝑤𝑖 (𝑡) 𝑤
[ 𝑛 𝑛
𝛼 󵄨 𝛼
𝑛
󵄨󵄨 𝛼𝑖𝑗 󵄨 ̃𝑖𝛼𝑖𝑖 (𝑡)󵄨󵄨󵄨󵄨
+ ∑ ∑ 𝜉𝑗 𝑏𝑗𝑖 (𝑡 + 𝜏𝑗𝑖 ) 𝑑𝑖 𝑗𝑖 󵄨󵄨󵄨𝑤𝑖 𝑖𝑖 (𝑡) − 𝑤
̃𝑗 𝑖𝑗 (𝑡)󵄨󵄨󵄨󵄨
𝛼 𝛼
+ ∑ 𝑏𝑖𝑗 (𝑡 + 𝜏𝑖𝑗 ) 𝑑𝑗 𝑖𝑗 󵄨󵄨𝑤𝑗 (𝑡) − 𝑤 𝑖=1 𝑗=1,𝑗 ≠
𝑖
󵄨
𝑗=1,𝑗 ≠
𝑖
𝑛
≤ −∑ [𝜉𝑖 (𝑎𝑖𝑖 + 𝑏𝑖𝑖 ) 𝑑𝑖 𝑖𝑖
𝑛 𝛼
𝛼 󵄨 𝛼 󵄨
− ∑ 𝑏𝑖𝑗 (𝑡) 𝑑𝑗 𝑖𝑗 󵄨󵄨󵄨󵄨𝑤𝑗 𝑖𝑗 (𝑡 − 𝜏𝑖𝑗 ) − 𝑤
̃𝑗 𝑖𝑗 (𝑡 − 𝜏𝑖𝑗 )󵄨󵄨󵄨󵄨]
𝛼
𝑖=1
[
𝑗=1,𝑗 ≠
𝑖
]
𝑛
𝑛 𝛼
̃𝑖 (𝑡)
ln 𝑤𝑖 (𝑡) − ln 𝑤 − ∑ 𝜉𝑗 𝑑𝑖 𝑗𝑖 (𝑎𝑗𝑖 + 𝑏𝑗𝑖 )]
= ∑𝜉𝑖 󵄨󵄨
̃𝑖 (𝑡)󵄨󵄨󵄨󵄨
𝑖=1 󵄨󵄨ln 𝑤𝑖 (𝑡) − ln 𝑤
𝑗=1,𝑗 ≠
𝑖
]
󵄨󵄨 𝛼𝑖𝑖 𝛼𝑖𝑖 󵄨
󵄨
̃𝑖 (𝑡)󵄨󵄨 .
𝑛 × 󵄨󵄨𝑤𝑖 (𝑡) − 𝑤
𝛼 𝛼 𝛼
× [− ∑𝑎𝑖𝑗 (𝑡) 𝑑𝑗 𝑖𝑗 (𝑤𝑗 𝑖𝑗 (𝑡) − 𝑤
̃𝑗 𝑖𝑗 (𝑡)) (68)
[ 𝑗=1 From Step 1, we know 𝜌(H𝑇 ) = 𝜌(H) = 𝜌(K) < 1. Thus, in
−
𝛼
𝑏𝑖𝑖 (𝑡) 𝑑𝑗 𝑖𝑖
𝛼
(𝑤𝑖 𝑖𝑖 (𝑡) − ̃𝑖𝛼𝑖𝑖
𝑤 (𝑡)) view of Lemma 5 and Definition 4, (𝐸 − H𝑇 ) is an 𝑀-matrix,
where 𝐸 denotes an identity matrix of size 𝑛. Therefore, by
𝑛 Lemma 8, there exists a diagonal matrix Ξ = diag(𝜉1 , . . . , 𝜉𝑛 )
𝛼 󵄨 𝛼 󵄨
̃𝑗 𝑖𝑗 (𝑡)󵄨󵄨󵄨󵄨]
+ ∑ 𝑏𝑖𝑗 (𝑡 + 𝜏𝑖𝑗 ) 𝑑𝑗 𝑖𝑗 󵄨󵄨󵄨󵄨𝑤𝑗 𝑖𝑗 (𝑡) − 𝑤
𝛼
with positive diagonal elements such that the product (𝐸 −
𝑗=1,𝑗 ≠
𝑖
] H𝑇 )Ξ is strictly diagonally dominant with positive diagonal
𝑛 𝑛
entries; namely,
𝛼 󵄨 𝛼 󵄨
̃𝑗 𝑖𝑗 (𝑡)󵄨󵄨󵄨󵄨
+ ∑ ∑ 𝜉𝑖 𝑏𝑖𝑗 (𝑡 + 𝜏𝑖𝑗 ) 𝑑𝑗 𝑖𝑗 󵄨󵄨󵄨󵄨𝑤𝑗 𝑖𝑗 (𝑡) − 𝑤
𝛼
𝑛
𝑖=1 𝑗=1,𝑗 ≠
𝑖 𝜉𝑖 > ∑ 𝜉𝑗 ℎ𝑗𝑖 ,
𝑗=1,𝑗 ≠
𝑖
𝑛
𝛼 󵄨󵄨 𝛼𝑖𝑖 ̃𝑖𝛼𝑖𝑖 (𝑡)󵄨󵄨󵄨󵄨
≤ −∑𝜉𝑖 𝑎𝑖𝑖 (𝑡) 𝑑𝑖 𝑖𝑖 󵄨󵄨𝑤𝑖 (𝑡) − 𝑤 𝛼
𝑛
𝛼 (69)
𝑖=1 or 𝜉𝑖 𝑑𝑖 𝑖𝑖 (𝑎𝑖𝑖 + 𝑏𝑖𝑖 ) − ∑ 𝜉𝑗 𝑑𝑖 𝑗𝑖 (𝑎𝑗𝑖 + 𝑏𝑗𝑖 ) > 0,
𝑗=1,𝑗 ≠
𝑖
𝑛 𝑛
󵄨󵄨 𝛼𝑗𝑖 󵄨
̃𝑖 𝑗𝑖 (𝑡)󵄨󵄨󵄨󵄨
𝛼 𝛼
+ ∑ ∑ 𝜉𝑗 𝑎𝑗𝑖 (𝑡) 𝑑𝑖 𝑗𝑖 󵄨󵄨𝑤𝑖 (𝑡) − 𝑤 𝑖 = 1, . . . , 𝑛.
󵄨
𝑖=1 𝑗=1,𝑗 ≠
𝑖
+
It follows from (68) and (69) that 𝐷 𝑉(𝑡) ≤ 0. Obviously, the
𝑛
𝛼 𝛼 zero solution of (64) is Lyapunov stable. On the other hand,
− ∑𝜉𝑖 𝑏𝑖𝑖 (𝑡) 𝑑𝑖 𝑖𝑖 (𝑤𝑖 𝑖𝑖 (𝑡) − ̃𝑖𝛼𝑖𝑖
𝑤 (𝑡))
𝑖=1
integrating (69) over [𝑡0 , 𝑡] leads to
𝑡 𝑛
𝑛 𝑛 󵄨 𝛼 ̃𝑖𝛼𝑖𝑖 (𝑠)󵄨󵄨󵄨󵄨 𝑑𝑠,
󵄨󵄨 𝛼𝑗𝑖 󵄨 𝑉 (𝑡) − 𝑉 (𝑡0 ) ⩽ −𝑘 ∫ ∑ 󵄨󵄨󵄨𝑤𝑖 𝑖𝑖 (𝑠) − 𝑤 𝑡 ≥ 𝑡0 ,
̃𝑖 𝑗𝑖 (𝑡)󵄨󵄨󵄨󵄨
𝛼 𝛼
+ ∑ ∑ 𝜉𝑗 𝑏𝑗𝑖 (𝑡 + 𝜏𝑗𝑖 ) 𝑑𝑖 𝑗𝑖 󵄨󵄨𝑤𝑖 (𝑡) − 𝑤
󵄨 𝑡 0 𝑖=1
𝑖=1 𝑗=1,𝑗 ≠
𝑖
(70)
𝑛 or
𝛼 󵄨 𝛼
̃𝑖𝛼𝑖𝑖 (𝑡)󵄨󵄨󵄨󵄨
≤ −∑𝜉𝑖 [𝑎𝑖𝑖 (𝑡) + 𝑏𝑖𝑖 (𝑡)] 𝑑𝑖 𝑖𝑖 󵄨󵄨󵄨𝑤𝑖 𝑖𝑖 (𝑡) − 𝑤 𝑡 𝑛
󵄨 𝛼 ̃𝑖𝛼𝑖𝑖 (𝑠)󵄨󵄨󵄨󵄨 𝑑𝑠 ≤ 𝑉 (𝑡0 ) < +∞,
𝑖=1 𝑉 (𝑡) + 𝑘 ∫ ∑󵄨󵄨󵄨𝑤𝑖 𝑖𝑖 (𝑠) − 𝑤 𝑡 ⩾ 𝑡0 .
𝑛 𝑛 𝑡 0 𝑖=1
𝛼 󵄨 𝛼 󵄨
̃𝑖 𝑗𝑖 (𝑡)󵄨󵄨󵄨󵄨
+ ∑ ∑ 𝜉𝑗 𝑎𝑗𝑖 (𝑡) 𝑑𝑖 𝑗𝑖 󵄨󵄨󵄨󵄨𝑤𝑖 𝑗𝑖 (𝑡) − 𝑤
𝛼
(71)
𝑖=1 𝑗=1,𝑗 ≠
𝑖
Noting that 𝑉(𝑡) ⩾ 0, it follows that
𝑛 𝑛 𝑡 𝑛
𝛼 󵄨 𝛼 󵄨 𝑉 (𝑡0 )
̃𝑖 𝑗𝑖 (𝑡)󵄨󵄨󵄨󵄨 .
+ ∑ ∑ 𝜉𝑗 𝑏𝑗𝑖 (𝑡 + 𝜏𝑗𝑖 ) 𝑑𝑖 𝑗𝑖 󵄨󵄨󵄨󵄨𝑤𝑖 𝑗𝑖 (𝑡) − 𝑤
𝛼 󵄨 𝛼 ̃𝑖𝛼𝑖𝑖 (𝑠)󵄨󵄨󵄨󵄨 𝑑𝑠 ≤
∫ ∑ 󵄨󵄨󵄨𝑤𝑖 𝑖𝑖 (𝑠) − 𝑤 < +∞, 𝑡 ⩾ 𝑡0 .
𝑖=1 𝑗=1,𝑗 ≠
𝑖 𝑡0 𝑖=1 𝑘
(66) (72)
10 Abstract and Applied Analysis

Therefore, by Lemma 7, it is not difficult to conclude that 5. Discussion

󵄨󵄨 ̃𝑖 (𝑡)󵄨󵄨󵄨󵄨 = 0.
lim 󵄨𝑤 (𝑡) − 𝑤 As we know, dynamic systems are often classified into two
𝑡 → +∞ 󵄨 𝑖
(73)
categories of either continuous-time or discrete-time systems.
Which implies the global asymptotical stability of system However, many real-world phenomena are neither purely
(64). By the linear transformation 𝑦𝑖 (𝑡) = 𝑑𝑖 𝑤𝑖 (𝑡), 𝑖 = continuous-time nor purely discrete-time. This leads to the
̃ of (3) is also
1, 2, . . . , 𝑛, the positive periodic solution 𝑦(𝑡) development of dynamic systems with impulses, which dis-
globally asymptotically stable. This completes the proof of play a combination of characteristics of both the continuous-
Theorem 11. time and discrete-time systems and hence provide a more
natural framework for mathematical modeling of many real-
world phenomena. Whether the new method proposed in
4. Remark and Discussion this paper can be applied to study the existence and global
asymptotic stability of the LV systems with impulses remains
To illustrate the generality of our results, we will give a open.
corollary in this section. Now recall that, for a given matrix
K, its spectral radius 𝜌(K) is equal to the minimum of all
matrix norms of K; that is, for any matrix norm ‖⋅‖, 𝜌(K) ≤ 6. Examples
‖K‖. Therefore, a corollary of Theorem 11 is stated as follows.
In this section, some examples and their simulations are
Corollary 12. In addition to (H1 ), (H2 ), and (H4 ), if one presented to illustrate the feasibility and effectiveness of our
further supposes that there exist positive constants 𝜉𝑖 , 𝑖 = results.
1, 2, . . . , 𝑛, such that one of the following inequalities holds:
Example 15. Consider the two-species competitive system
(1) max1≤𝑗≤𝑛 ∑𝑛𝑖=1,𝑖 ≠
𝑗 (𝜉𝑖 (𝑎𝑖𝑗 + 𝑏𝑖𝑗 )/𝜉𝑗 (𝑎𝑗𝑗 + 𝑏𝑗𝑗 )) < 1, or
1
equivalently, (𝑎𝑗𝑗 + 𝑏𝑗𝑗 )𝜉𝑗 > ∑𝑛𝑖=1,𝑖 ≠
𝑗 (𝑎𝑖𝑗 + 𝑏𝑖𝑗 )𝜉𝑖 for all 𝑦1̇(𝑡) = 𝑦1 (𝑡) [4 − (2 + sin 𝑡) 𝑦12 (𝑡) − 𝑦2 (𝑡)
𝑗 = 1, 2, . . . , 𝑛, 2
1 1
(2) ∑𝑛𝑖=1 ∑𝑛𝑗=1 (𝜉𝑖−1 𝜉𝑗 Γ𝑖𝑗 )2 < 1, where Γ𝑖𝑗 has been defined in − (3 + cos 𝑡) 𝑦12 (𝑡) − 𝑦2 (𝑡 − 0.2)] ,
Theorem 6, 8 2
(75)
(3) max1≤𝑖≤𝑛 ∑𝑛𝑗=1,𝑗 ≠
𝑖 (𝜉𝑗 (𝑎𝑗𝑖 + 𝑏𝑗𝑖 )/𝜉𝑖 (𝑎𝑖𝑖 + 𝑏𝑖𝑖 )) < 1, or 𝑦2̇(𝑡) = 𝑦2 (𝑡) [2 − (1 + sin 𝑡) 𝑦1 (𝑡) − (4 − cos 𝑡) 𝑦22 (𝑡)
equivalently, (𝑎𝑖𝑖 + 𝑏𝑖𝑖 )𝜉𝑖 > ∑𝑛𝑗=1,𝑗 ≠
𝑖 (𝑎𝑗𝑖 + 𝑏𝑗𝑖 )𝜉𝑖 for all
𝑖 = 1, 2, . . . , 𝑛, 1 1
− (1 − cos 𝑡) 𝑦1 (𝑡 − 0.5) − 𝑦22 (𝑡)] .
then system (3) has a unique positive 𝜔-periodic 4 2
solution which is globally asymptotically stable.
Corresponding to system (3), we have 𝑎11 = 𝑎21 = 2, 𝑎12 =
Proof. For any matrix norm ‖ ⋅ ‖ and any nonsingular matrix 𝑏12 = 𝑏21 = 𝑏22 = 1/2, 𝑎12 = 1/2, 𝑎22 = 3, 𝑏11 = 1/4. Simple
𝑆, ‖K‖𝑆 = ‖𝑆−1 K𝑆‖ also defines a matrix norm. Let Ξ = computation shows
diag(𝜉1 , 𝜉2 , . . . , 𝜉𝑛 ). Then conditions (1)-(2) correspond to
the column norms and Frobenius norm of matrix ΞKΞ−1 , 𝑎12 + 𝑏12
0
respectively. Condition (3) corresponds to the row norms 𝑎22 + 𝑏22
of ΞK𝑇Ξ−1 and note that 𝜌(ΞK𝑇Ξ−1 ) = 𝜌(ΞKΞ−1 ). K=( )
𝑎21 + 𝑏21
Corollary 12 follows immediately. 0
𝑎11 + 𝑏11
Remark 13. The corollary implies that the conditions given in 1 1 2
terms of the spectral radius are much better than the classic 0 ( + )× (76)
2 2 7
norms. =( )
1 4
(2 + ) × 0
Now we consider a special case of system (3). Take 𝛼𝑖𝑗 = 1, 2 5
𝑏𝑖𝑗 = 0, and 𝜏𝑖𝑗 ≡ 0; then system (3) reduces to the classical LV 2
competition system which has been well studied in Xia and 0
Han [8]: = ( 7) .
2 0
𝑛
𝑦𝑖̇(𝑡) = 𝑦𝑖 (𝑡) [𝑟𝑖 (𝑡) − ∑ 𝑎𝑖𝑗 (𝑡) 𝑦𝑗 (𝑡)] , 𝑖 = 1, 2, . . . , 𝑛. And 𝜌(K) = √4/7 < 1. Thus, by Theorem 11, system (75) has
𝑗=1
[ ] a unique positive equilibrium which is globally asymptoti-
(74) cally stable.
Remark 14. In this case, Theorem 11 and Corollary 12 reduce Remark 16. In this example, one can observe that, though
to the main results in Xia and Han [8]. the spectral 𝜌(K) < 1, the matrix norms (including the row
Abstract and Applied Analysis 11

norm, the column norm, and the Frobenius norm) of matrix Hence, by using MATLAB, we get
K may be bigger than 1. For instance, the column norm
𝜌 (K) = max . eigenvalues [K] = 0.7667 < 1. (80)
‖K‖1 = 0 + 2 > 1. (77)
Thus, by Theorem 11, system (78) has a unique positive
Example 17. Consider the three-species competitive system equilibrium which is globally asymptotically stable.

1 Remark 18. In this example, one can observe that though the
𝑦1̇(𝑡) = 𝑦1 (𝑡) [4 − (2 + sin 𝑡) 𝑦12 (𝑡) spectral 𝜌(K) < 1, the matrix norms (including the row
2
norm, the column norm, and the Frobenius norm) of matrix
1 1 K may be bigger than 1. For instance, the column norm
− (1 + cos 𝑡) 𝑦2 (𝑡) − 𝑦3 (𝑡)
8 12
1
1 ‖K‖1 = 0 + + 2 > 1. (81)
− (2 + sin 𝑡) 𝑦12 (𝑡) − (1 + cos 𝑡) 𝑦2 (𝑡 − 0.5) 3
8
1 Conflict of Interests
− 𝑦3 (𝑡 − 0.2)] ,
4
The authors declare that there is no conflict of interests
1
𝑦2̇(𝑡) = 𝑦2 (𝑡) [4 − (1 − sin 𝑡) 𝑦1 (𝑡) regarding the publication of this paper.
8
1 Acknowledgments
− (2 + sin 𝑡) 𝑦22 (𝑡)
2
This work is supported by the National Natural Science
1
− (1 − sin 𝑡) 𝑦1 (𝑡 − 0.5) Foundation of China under Grant no. 11271333 and ZJNSFC.
8

− (2 + sin 𝑡) 𝑦22 (𝑡) ] , References

[1] Y. L. Song and S. L. Yuan, “Bifurcation analysis for a regulated
𝑦3̇(𝑡) = 𝑦3 (𝑡) [6 − (1 − sin 𝑡) 𝑦1 (𝑡) logistic growth model,” Applied Mathematical Modelling, vol. 31,
no. 9, pp. 1729–1738, 2007.
1 1 [2] Y. L. Song, S. L. Yuan, and J. Zhang, “Bifurcation analysis
− (1 + sin 𝑡) 𝑦2 (𝑡) − (2 + cos 𝑡) 𝑦32 (𝑡) in the delayed Leslie-Gower predator-prey system,” Applied
8 2
Mathematical Modelling, vol. 33, no. 11, pp. 4049–4061, 2009.
1
− sin 𝑡𝑦1 (𝑡 − 0.5) − 𝑦2 (𝑡 − 0.2) [3] S. L. Yuan and F. Zhang, “Stability and global Hopf bifurcation
4 in a delayed predator-prey system,” Nonlinear Analysis. Real
World Applications, vol. 11, no. 2, pp. 959–977, 2010.
− (2 + cos 𝑡) 𝑦32 (𝑡) ] . [4] M. Fan, K. Wang, and D. Jiang, “Existence and global attractivity
(78) of positive periodic solutions of periodic 𝑛-species Lotka-
Volterra competition systems with several deviating arguments,”
Mathematical Biosciences, vol. 160, no. 1, pp. 47–61, 1999.
Corresponding to system (3), we have 𝑎11 = 𝑎22 = 𝑎33 = 1/2,
𝑎12 = 𝑎21 = 𝑎32 = 1/4, 𝑎13 = 1/12, 𝑎23 = 0, 𝑎31 = 2, 𝑏11 = [5] W. Ding and M. Han, “Dynamic of a non-autonomous predator-
prey system with infinite delay and diffusion,” Computers &
𝑏22 = 𝑏33 = 𝑏31 = 1, 𝑏12 = 𝑏13 = 𝑏21 = 𝑏32 = 1/4, and 𝑏23 = 0. Mathematics with Application, vol. 56, no. 5, pp. 1335–1350,
Then, simple computation leads to 2008.
[6] Y.-H. Xia, “Global analysis of an impulsive delayed Lotka-
𝑎12 + 𝑏12 𝑎13 + 𝑏13 Volterra competition system,” Communications in Nonlinear
0 Science and Numerical Simulation, vol. 16, no. 3, pp. 1597–1616,
𝑎22 + 𝑏22 𝑎33 + 𝑏33
(𝑎 + 𝑏 ) 2011.
( )
K = ( 21 21
0 0 ) [7] Y.-H. Xia, “Global asymptotic stability of an almost periodic
( 𝑎11 + 𝑏11 ) nonlinear ecological model,” Communications in Nonlinear
𝑎31 + 𝑏31 𝑎32 + 𝑏32 Science and Numerical Simulation, vol. 16, no. 11, pp. 4451–4478,
0 2011.
( 𝑎11 + 𝑏11 𝑎22 + 𝑏22 ) (79) [8] Y.-H. Xia and M. Han, “New conditions on the existence and
1 2 stability of periodic solution in Lotka-Volterra’s population
0 system,” SIAM Journal on Applied Mathematics, vol. 69, no. 6,
3 9 pp. 1580–1597, 2009.
1
=( 0 0) . [9] K. Hong and P. Weng, “Stability and traveling waves of diffusive
3 predator-prey model with age-structure and nonlocal effect,”
1 The Journal of Applied Analysis and Computation, vol. 2, no. 2,
2 0
( 3 ) pp. 173–192, 2012.
12 Abstract and Applied Analysis

[10] I. Djellit, M. L. Sahari, and A. Hachemi, “Complex dynamics

in 2-species predator-prey systems,” The Journal of Applied
Analysis and Computation, vol. 3, no. 1, pp. 11–20, 2013.
[11] M. E. Gilpin and F. J. Ayala, “Global models of growth and
competition,” Proceedings of the National Academy of Sciences
of the United States of America, vol. 70, pp. 3590–3593, 1973.
[12] F. J. Ayala, M. E. Gilpin, and J. G. Ehrenfeld, “Competition
between species: theoretical models and experimental tests,”
Theoretical Population Biology, vol. 4, pp. 331–356, 1973.
[13] F. D. Chen, “Average conditions for permanence and extinction
in nonautonomous Gilpin-Ayala competition model,” Nonlin-
ear Analysis. Real World Applications, vol. 7, no. 4, pp. 895–915,
2006.
[14] R. M. May, “Time delay versus stability in population models
with two or three trophic levels,” Ecology, vol. 54, no. 2, pp. 315–
325, 1973.
[15] K. L. Cooke and Z. Grossman, “Discrete delay, distributed delay
and stability switches,” Journal of Mathematical Analysis and
Applications, vol. 86, no. 2, pp. 592–627, 1982.
[16] W. Wendi and M. Zhien, “Harmless delays for uniform persis-
tence,” Journal of Mathematical Analysis and Applications, vol.
158, no. 1, pp. 256–268, 1991.
[17] R. E. Gaines and J. L. Mawhin, Coincidence Degree and Nonlin-
ear Differential Equations, Springer, Berlin, Germany, 1977.
[18] D. Guo, J. Sun, and Z. Liu, Functional Method in Nonlinear
Ordinary Differential Equations, Shangdong Scientific Press,
Shandong, China, 2005.
[19] J. P. LaSalle, The Stability of Dynamical System, Society for
Industrial and Applied Mathematics, Philadelphia, Pa, USA,
1976.
[20] A. Berman and R. J. Plemmons, Nonnegative Matrices in the
Mathematical Science, Academic Press, New York, NY, USA,
1979.
[21] I. Barbălat, “Systems d’equations differentielle d’oscillations
nonlineaires,” Revue Roumaine de Mathématique Pures et
Appliquées, vol. 4, pp. 267–270, 1959.
[22] Y.-H. Xia, M. Han, and Z. K. Huang, “Global attractivity of
an almost periodic 𝑁-species nonlinear ecological competitive
model,” Journal of Mathematical Analysis and Applications, vol.
337, no. 1, pp. 144–168, 2008.
[23] Y.-H. Xia, “Periodic solution of certain nonlinear differential
equations: via topological degree theory and matrix spectral
theory,” International Journal of Bifurcation and Chaos in
Applied Sciences and Engineering, vol. 22, no. 8, Article ID
1250196, 17 pages, 2012.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 568943, 8 pages
http://dx.doi.org/10.1155/2014/568943

Research Article
Stability and Hopf Bifurcation in a Delayed Predator-Prey
System with Herd Behavior

Chaoqun Xu and Sanling Yuan

College of Science, University of Shanghai for Science and Technology, Shanghai 200093, China

Correspondence should be addressed to Sanling Yuan; sanling@usst.edu.cn

Received 24 February 2014; Accepted 13 March 2014; Published 8 April 2014

Academic Editor: Yonghui Xia

Copyright © 2014 C. Xu and S. Yuan. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

A special predator-prey system is investigated in which the prey population exhibits herd behavior in order to provide a self-
defense against predators, while the predator is intermediate and its population shows individualistic behavior. Considering the
fact that there always exists a time delay in the conversion of the biomass of prey to that of predator in this system, we obtain a
delayed predator-prey model with square root functional response and quadratic mortality. For this model, we mainly investigate
the stability of positive equilibrium and the existence of Hopf bifurcation by choosing the time delay as a bifurcation parameter.

1. Introduction which determine the species intrinsic population dynamics

(primary production and mortality, resp.). For example,
During the last few decades, there has been great interest
in the construction and study of models for the population 𝑁 𝐾−𝑁
𝐹 (𝑁) = 𝑟 (1 − ), 𝐹 (𝑁) = 𝑟 ;
dynamics of predator-prey systems. The classical predator- 𝐾 𝐾 + 𝜀𝑁
prey model can be written in the generalized form (see [1]) (2)
𝑎
𝜇 (𝑃) ≡ 𝑠, 𝜇 (𝑃) = .
𝑑𝑁 𝑏+𝑃
= 𝐹 (𝑁) 𝑁 − 𝜙 (𝑁, 𝑃) 𝑃,
𝑑𝑡 Functional response function 𝜙(𝑁, 𝑃) also can be classi-
fied into many different types: Holling I–IV types [2, 3],
𝑑𝑃 (1)
= [𝑘𝜙 (𝑁, 𝑃) − 𝜇 (𝑃)] 𝑃, Beddington-DeAngelis type [4, 5], Ivlev type [6], ratio-
𝑑𝑡 dependent type [7], and the modified forms of these types [8–
𝑁 (0) ≥ 0, 𝑃 (0) ≥ 0, 18]. Over the years, the mathematical models with different
𝐹(𝑁), 𝜇(𝑃), and 𝜙(𝑁, 𝑃) have been constructed by many
where 𝑁 = 𝑁(𝑡), 𝑃 = 𝑃(𝑡) denote the number or density of researchers for describing different predator-prey systems;
the prey and predator species at time 𝑡, respectively, 𝐹(𝑁) is see [2–18].
the relative growth function of prey species in the absence of Recently, one kind of predator-prey system with herd
predator, 𝜇(𝑃) is the relative mortality function of predator behavior is considered by some researchers [19–23]. In this
species in the absence of prey, 𝜙(𝑁, 𝑃) is the functional kind of systems, such as herbivores on the large savanna and
response function of the predator to the prey which can their large predators, the prey species join together in herds
be interpreted as the change in the number of prey species in order to provide a self-defense against predators, and the
attached per unit time per predator as the prey number predator interacts with the prey along the outer corridor of
changes, and 𝑘 is the biomass conversion or consumption the herd of prey [20, 21]. Therefore, it is more appropriate to
rate. model the response functions of prey in terms of the square
For system (1), the dynamic behavior is determined by root of the number of prey species for this class of systems.
the three crucial components: 𝐹(𝑁), 𝜇(𝑃), and 𝜙(𝑁, 𝑃). Chen In [20], Braza considered a predator-prey system with square
et al. in [1] summarized some types of 𝐹(𝑁) and 𝜇(𝑃) root functional response and compared the dynamics of this
2 Abstract and Applied Analysis

system with the dynamics of predator-prey system that use a where 𝑋 = 𝑋(𝑡), 𝑌 = 𝑌(𝑡) denote the number of the prey
typical Lotka-Volterra interaction term. and predator species at time 𝑡, respectively. The parameter 𝑟
Moreover, there are some papers concerned with the is the growth rate of prey species, 𝐾 is its carrying capacity, 𝑠
effects of the form of mortality terms for the dynamics is the death rate of predator species in the absence of prey, 𝛼
[23–26]. Brentnall et al. [24] pointed out that the mortality is the search efficiency of predator for prey, 𝑐 is the biomass
term of predator is usually described by “linear form” or conversion rate, and 𝑡ℎ is predator’s average handling time of
“quadratic form” and the quadratic mortality is suited to prey.
intermediate predator (such as piscivore). Recently, Yuan Following References [23] and [24], we choose the quad-
et al. [23] studied a spatial predator-prey model with herd ratic mortality for predator species in system (3). If we further
behavior and concluded that the Turing pattern is induced consider the delay fact, system (3) should be modified as the
by quadratic mortality. Fulton et al. in [25] showed that, in following form:
the large and interlinked webs used in ecosystem models,
model behavior is far more sensitive to the form of the grazing
𝑑𝑋 𝑋 𝛼√𝑋𝑌
term than to that of the mortality terms that are close to the = 𝑟𝑋 (1 − ) − ,
modeled food web. 𝑑𝑡 𝐾 1 + 𝑡ℎ 𝛼√𝑋
On the other hand, for most of the natural ecosystems, (4)
𝑑𝑌 𝑐𝛼√𝑋 (𝑡 − 𝜏)𝑌
every species does not respond instantaneously to changes = −𝑠𝑌2 + ,
in the environment or the interactions with other species 𝑑𝑡 1 + 𝑡ℎ 𝛼√𝑋 (𝑡 − 𝜏)
within the community. Thus models with delay are much
more realistic [27–31], and many researchers paid a lot of where −𝑠𝑌2 represents the quadratic mortality of predator
attention to the delayed predator-prey systems; see [32–40] species and 𝜏 is the time delay which means the growth rate of
and the references cited therein. For example, Xiao and predator species to depend on the number of the prey species
Ruan [34] discussed a delayed predator-prey system with 𝜏 units of time earlier.
nonmonotonic functional response and showed that there By the same way of [20, 23], we make some scaling and
is a B-T singularity for any time delay value and the system assume that the average handling time is zero. Then the
can exhibit Hopf bifurcation as the time delay passes through working system is
some critical values. Xu et al. in [36] proposed a delayed
periodic L-V type predator-prey system with prey dispersal
in two-patch environments and gave the sufficient conditions 𝑑𝑥
= 𝑥 (1 − 𝑥) − √𝑥𝑦,
of the existence, uniqueness, and global stability of positive 𝑑𝑡
periodic solution. (5)
𝑑𝑦
Motivated by the above, in the present paper, we are = −𝑠𝑦2 + 𝑐√𝑥 (𝑡 − 𝜏)𝑦.
intended to consider a predator-prey system in which the 𝑑𝑡
prey species exhibits herd behavior, the predator species is
intermediate, and there exists a time delay in the convention The initial conditions for system (5) take the form
of the biomass of prey to that of predator.
This paper is organized as follows. We first introduce 𝑥0 (𝜃) = 𝜙 (𝜃) ≥ 0, 𝑦0 (𝜃) = 𝜓 (𝜃) ≥ 0, 𝜃 ∈ [−𝜏, 0) ,
our working system in the next section. In Section 3, by (6)
analyzing the characteristic equation, we discuss the local 𝑥0 (0) > 0, 𝑦0 (0) > 0,
stability of a positive equilibrium and study the existence of
Hopf bifurcations at the positive equilibrium. In Section 4, 2 2
using the normal form theory and center manifold argument, where (𝜙(𝜃), 𝜓(𝜃)) ∈ 𝐶([−𝜏, 0], R+ ), R+ = {(𝑟1 , 𝑟2 ) : 𝑟1 ≥
the explicit formulae are derived to determine the direction of 0, 𝑟2 ≥ 0}.
bifurcations and the properties of bifurcating periodic solu- It is easy to verify that system (5) always has a trivial
tions. Several numerical simulations and a simple discussion equilibrium 𝐸0 (0, 0) and a boundary equilibrium 𝐸1 (1, 0) for
are given in the last Section. any feasible parameters. If we assume 𝑐 < 𝑠, then system (5)
has a unique positive equilibrium 𝐸∗ (𝑥∗ , 𝑦∗ ), where

2. The Working System 𝑐 𝑐

𝑥∗ = 1 − , 𝑦∗ = √ 𝑥 ∗ . (7)
Braza [20] proposed the basic predator-prey system with 𝑠 𝑠
logistic growth in the prey species, linear mortality in the
predator species, and a square root functional response 3. Stability Analysis and Hopf Bifurcation
function. It is given by
Due to biological interpretation of the system, we only con-
𝑑𝑋 𝑋 𝛼√𝑋𝑌 sider the positive equilibrium. In this section, by choosing 𝜏
= 𝑟𝑋 (1 − ) − ,
𝑑𝑡 𝐾 1 + 𝑡ℎ 𝛼√𝑋 as the bifurcation parameter and analyzing the corresponding
(3) linearized system, we investigate the stability of the positive
𝑑𝑌 𝑐𝛼√𝑋𝑌 equilibrium and the effects of the time delay on the dynamics
= −𝑠𝑌 + ,
𝑑𝑡 1 + 𝑡ℎ 𝛼√𝑋 of system (5).
Abstract and Applied Analysis 3

To study the local stability of the positive equilibrium Now for 𝜏 > 0, let 𝜆 = 𝑖𝜔 (𝜔 > 0) be a root of (11); then 𝜔
𝐸∗ (𝑥∗ , 𝑦∗ ), we first use the linear transformation 𝑥1 = 𝑥 − 𝑥∗ , satisfies the following equation:
𝑥2 = 𝑦 − 𝑦∗ , for which system (5) can be written in the form
− 𝜔2 − (𝑎11 + 𝑎22 ) 𝜔𝑖 + 𝑎11 𝑎22 − 𝑎12 𝑎21 (cos 𝜔𝜏 − 𝑖 sin 𝜔𝜏)
𝑑𝑥1 𝑗
𝑓𝑖𝑗(1)
= 𝑎11 𝑥1 + 𝑎12 𝑥2 + ∑ 𝑥1𝑖 𝑥2 , = 0.
𝑑𝑡 𝑖+𝑗≥2 𝑖!𝑗!
(8) (13)
𝑑𝑥2 𝑓𝑖𝑗(2)
𝑗
= 𝑎21 𝑥1 (𝑡 − 𝜏) + 𝑎22 𝑥2 + ∑ 𝑥1𝑖 (𝑡 − 𝜏) 𝑥2 , Separating the real and imaginary parts, we have
𝑑𝑡 𝑖+𝑗≥2 𝑖!𝑗!

where −𝜔2 + 𝑎11 𝑎22 − 𝑎12 𝑎21 cos 𝜔𝜏 = 0,

(14)
3𝑐 − (𝑎11 + 𝑎22 ) 𝜔 + 𝑎12 𝑎21 sin 𝜔𝜏 = 0,
𝑎11 = − 1, 𝑎12 = −√𝑥∗ ,
2𝑠
which imply that
𝑐2
𝑎21 = , 𝑎22 = −𝑐√𝑥∗ ,
2𝑠
𝜔4 + (𝑎11
2 2
+ 𝑎22 ) 𝜔2 + 𝑎11
2 2 2 2
𝑎22 − 𝑎12 𝑎21 = 0. (15)
󵄨 󵄨
𝜕𝑖+𝑗 𝑓(1) (𝑥1 , 𝑥2 ) 󵄨󵄨󵄨󵄨 𝜕𝑖+𝑗 𝑓(2) (𝑥3 , 𝑥2 ) 󵄨󵄨󵄨󵄨 Notice 𝑎11 𝑎22 −𝑎12 𝑎21 > 0; we can find that (15) has no positive
𝑓𝑖𝑗(1) = 󵄨󵄨 , 𝑓𝑖𝑗(2) = 󵄨󵄨 ,
𝑗
𝜕𝑥1𝑖 𝜕𝑥2 󵄨󵄨󵄨(0,0) 𝑗
𝜕𝑥3𝑖 𝜕𝑥2 󵄨󵄨󵄨(0,0) roots if 𝑎11 𝑎22 +𝑎12 𝑎21 ≥ 0 and has a positive root 𝜔+ if 𝑎11 𝑎22 +
𝑎12 𝑎21 < 0, where
𝑓(1) (𝑥1 , 𝑥2 ) = (𝑥1 + 𝑥∗ ) (1 − 𝑥1 − 𝑥∗ )

− √𝑥1 + 𝑥∗ (𝑥2 + 𝑦∗ ) , 2 2 2 2 2
√ − (𝑎11 + 𝑎22 ) + √(𝑎11 − 𝑎22 ) + 4𝑎12 𝑎21
2 2
(16)
𝜔+ = .
∗ 2 2
𝑓(2) (𝑥3 , 𝑥2 ) = −𝑠(𝑥2 + 𝑦 ) + 𝑐√𝑥3 + 𝑥∗ (𝑥2 + 𝑦∗ ) .
(9) Defining

Thus, the positive equilibrium of system (5) is transformed 2

1 −(𝜔+ ) + 𝑎11 𝑎22
into the zero equilibrium of system (8). 𝜏𝑗 = (arccos + 2𝑗𝜋) , 𝑗 = 0, 1, 2, . . . ,
The linear equations corresponding to (8) are given by 𝜔+ 𝑎12 𝑎21
(17)
𝑑𝑥1
= 𝑎11 𝑥1 + 𝑎12 𝑥2 ,
𝑑𝑡 we have the following result.
(10)
𝑑𝑥2
= 𝑎21 𝑥1 (𝑡 − 𝜏) + 𝑎22 𝑥2 , Lemma 2. Assume 𝑐 < 𝑠 and 𝑎11 + 𝑎22 < 0; then the following
𝑑𝑡
statements are true.
and its characteristic equation is
(1) If 𝑎11 𝑎22 + 𝑎12 𝑎21 ≥ 0, the roots of (11) have always
𝜆2 − (𝑎11 + 𝑎22 ) 𝜆 + 𝑎11 𝑎22 − 𝑎12 𝑎21 𝑒−𝜆𝜏 = 0. (11) negative real parts for all 𝜏 ≥ 0.
To study the stability of the zero equilibrium and the Hopf (2) If 𝑎11 𝑎22 +𝑎12 𝑎21 < 0, (11) has a pair of purely imaginary
bifurcation of system (8), we should analyze the distribution roots ±𝑖𝜔+ for 𝜏 = 𝜏𝑗 .
of the roots of characteristic equation (11). A simple calcula-
tion shows that 𝑎11 𝑎22 − 𝑎12 𝑎21 = 𝑐(𝑥∗ )3/2 > 0; thus 𝜆 = 0 is Next, under the condition of Lemma 2(2), we will investi-
not a root of (11). gate whether the transversality condition is satisfied. Denote
When there is no delay, that is, 𝜏 = 0, characteristic by 𝜆(𝜏) = 𝛼(𝜏) + 𝑖𝜔(𝜏) the root of (11) near 𝜏 = 𝜏𝑗 satisfying
equation (11) reduces to 𝛼(𝜏𝑗 ) = 0, 𝜔(𝜏𝑗 ) = 𝜔+ , 𝑗 = 0, 1, 2, . . .. Differentiating the two
sides of (11) with respect to 𝜏, we get
𝜆2 − (𝑎11 + 𝑎22 ) 𝜆 + 𝑎11 𝑎22 − 𝑎12 𝑎21 = 0. (12)
𝑑𝜆 𝑑𝜆 𝑑𝜆
If 𝑎11 + 𝑎22 < 0, the two roots of (12) will have always negative 2𝜆 − (𝑎11 + 𝑎22 ) + 𝑎12 𝑎21 𝑒−𝜆𝜏 (𝜆 + 𝜏 ) = 0; (18)
real parts. Thus, we can obtain the following result. 𝑑𝜏 𝑑𝜏 𝑑𝜏

Lemma 1. Assume 𝑐 < 𝑠 and 𝑎11 + 𝑎22 < 0; then the two roots that is
of characteristic equation (11) with 𝜏 = 0 have always negative
real parts; that is, the zero equilibrium of system (8) with 𝜏 = 0 𝑑𝜆 −1 2𝜆 − 𝑎11 − 𝑎22 𝜆𝜏 𝜏
( ) = 𝑒 − . (19)
is locally asymptotically stable. 𝑑𝜏 −𝑎12 𝑎21 𝜆 𝜆
4 Abstract and Applied Analysis

We can further get Assume that system (5) undergoes Hopf bifurcations at
𝐸∗ (𝑥∗ , 𝑦∗ ) for 𝜏 = 𝜏𝑗 , and then ±𝑖𝜔+ are the corresponding
󵄨
𝑑𝛼(𝜏) −1 󵄨󵄨󵄨 purely imaginary roots of characteristic equation (11). Let
sign ( ) 󵄨󵄨󵄨
𝑑𝜏 󵄨󵄨𝜏=𝜏𝑗
𝑞 (𝜃) = (1, 𝛼)𝑇 𝑒𝑖𝜔+ 𝜏𝑗 𝜃 , 𝑞∗ (𝜗) = 𝐷 (1, 𝛼∗ ) 𝑒𝑖𝜔+ 𝜏𝑗 𝜗
−1 󵄨󵄨
𝑑𝜆 󵄨󵄨 (22)
= sign Re( ) 󵄨󵄨 for 𝜃, 𝜗 ∈ [−1, 0] .
𝑑𝜏 󵄨󵄨
󵄨𝜏=𝜏𝑗
𝑎11 + 𝑎22 − 2𝜆 𝜆𝜏 𝜏 󵄨󵄨󵄨󵄨 We can choose
= sign Re ( 𝑒 − )󵄨󵄨
𝑎12 𝑎21 𝜆 𝜆 󵄨󵄨𝜏=𝜏𝑗
𝑖𝜔+ − 𝑎11 −𝑖𝜔+ − 𝑎11 𝑖𝜔+
𝑎 + 𝑎22 − 2𝑖𝜔+ 𝛼= , 𝛼∗ = 𝑒 ,
= sign Re ( 11 (cos 𝜔+ 𝜏𝑗 + 𝑖 sin 𝜔+ 𝜏𝑗 )) 𝑎21 𝑎21
𝑎12 𝑎21 𝑖𝜔+ (23)
1
𝐷= ,
2 (𝜔+2 − 𝑎11 𝑎22 ) (𝑎11 + 𝑎22 )
2
1 + 𝛼 𝛼 + 𝜏𝑗 𝛼∗ 𝑎21 𝑒𝑖𝜔+ 𝜏𝑗
∗
= sign Re ( 2 𝑎2
+ 2 𝑎2
)
𝑎12 21 𝑎12 21
such that 𝑞(𝜃) is the eigenvector of operator 𝐴 corresponding
2𝜔+2 + 𝑎11
2 2
+ 𝑎22 to 𝑖𝜔+ 𝜏𝑗 , 𝑞∗ (𝜗) is the eigenvector of operator 𝐴∗ correspond-
= sign Re ( 2 2
) > 0.
𝑎12 𝑎21 ing to −𝑖𝜔+ 𝜏𝑗 , and
(20)

Then we have the following conclusion. ⟨𝑞∗ (𝜗) , 𝑞 (𝜃)⟩ = 1, ⟨𝑞∗ (𝜗) , 𝑞 (𝜃)⟩ = 0. (24)

Lemma 3. Assume 𝑐 < 𝑠, 𝑎11 +𝑎22 < 0, and 𝑎11 𝑎22 +𝑎12 𝑎21 < 0; Thus, we can compute
then the transversality conditions are satisfied; that is,

𝑑𝛼(𝜏) 󵄨󵄨󵄨󵄨
(1) (1)
𝑔20 = 𝐷𝜏𝑗 [(𝑓20 + 2𝛼𝑓11 )
󵄨 > 0, 𝑗 = 0, 1, 2, . . . .
𝑑𝜏 󵄨󵄨󵄨𝜏=𝜏𝑗
(21)
+𝛼∗ (𝑒−2𝑖𝜔+ 𝜏𝑗 𝑓20
(2)
+ 2𝛼𝑒−𝑖𝜔+ 𝜏𝑗 𝑓11
(2)
+ 𝛼2 𝑓02
(2)
)] ,
Summarizing the above three lemmas, we can obtain the (1) (1)
𝑔11 = 𝐷𝜏𝑗 [(𝑓20 + (𝛼 + 𝛼) 𝑓11 )
following theorem on stability and Hopf bifurcation of system
(5).
+ 𝛼∗ (𝑓20
(2)
+ (𝛼𝑒𝑖𝜔+ 𝜏𝑗 + 𝛼𝑒−𝑖𝜔+ 𝜏𝑗 ) 𝑓11
(2) (2)
+ 𝛼𝛼𝑓02 )] ,
Theorem 4. Assume 𝑐 < 𝑠 and 𝑎11 + 𝑎22 < 0. For system (5), (1) (1)
the following results are true. 𝑔02 = 𝐷𝜏𝑗 [(𝑓20 + 2𝛼𝑓11 )

(1) If 𝑎11 𝑎22 + 𝑎12 𝑎21 ≥ 0, the positive equilibrium 𝐸∗ is +𝛼∗ (𝑒2𝑖𝜔+ 𝜏𝑗 𝑓20
(2)
+ 2𝛼𝑒𝑖𝜔+ 𝜏𝑗 𝑓11
(2)
+ 𝛼2 𝑓02
(2)
)] ,
locally asymptotically stable for all 𝜏 ≥ 0.
𝑔21 = 𝐷𝜏𝑗 (𝑤1 + 𝛼∗ 𝑤2 ) ,
∗
(2) If 𝑎11 𝑎22 + 𝑎12 𝑎21 < 0, then 𝐸 is locally asymptotically (25)
stable when 𝜏 ∈ [0, 𝜏0 ) and unstable when 𝜏 > 𝜏0 and
system (5) undergoes a Hopf bifurcation at 𝐸∗ when 𝜏 =
𝜏𝑗 , 𝑗 = 0, 1, 2, . . .. where

(1) (1) (1) (1) (1)

4. Direction and Stability of 𝑤1 = (2𝑊11 (0) + 𝑊20 (0)) 𝑓20 + 𝑓30 + (2𝛼 + 𝛼) 𝑓21
the Hopf Bifurcation (2) (2) (1) (1) (1)
+ (2𝑊11 (0) + 𝑊20 (0) + 𝛼𝑊20 (0) + 2𝛼𝑊11 (0)) 𝑓11 ,
In the previous section, we have obtained the conditions
under which a family of periodic solutions bifurcate from 𝑤2 = (2𝑒−𝑖𝜔+ 𝜏𝑗 𝑊11
(1)
(−1) + 𝑒𝑖𝜔+ 𝜏𝑗 𝑊20
(1) (2)
(−1)) 𝑓20
the positive equilibrium 𝐸∗ as 𝜏 crosses through the critical
(2) (2) (2)
values 𝜏𝑗 . In this section, we will study the direction of Hopf + (2𝛼𝑊11 (0) + 𝛼𝑊20 (0)) 𝑓02
bifurcation and the stability and period of the bifurcating
(1) (1)
periodic solutions by applying the normal form theory and + (2𝛼𝑊11 (−1) + 𝛼𝑊20 (−1) + 𝑒𝑖𝜔+ 𝜏𝑗 𝑊20
(2)
(0)
the center manifold theorem from Hassard et al. [41]. Since
the methods used are standard, we omit the detailed process +2𝑒−𝑖𝜔+ 𝜏𝑗 𝑊11
(2) (2)
(0)) 𝑓11
and only give the main results. The readers can see [41] for
more details on the derivation process. + (2𝛼 + 𝛼𝑒−2𝑖𝜔+ 𝜏𝑗 ) 𝑓21
(2) (2) −𝑖𝜔+ 𝜏𝑗
+ 𝑓30 𝑒 ,
Abstract and Applied Analysis 5

0.45 0.45

0.4 0.4

y(t)
y(t)

0.35 0.35

0.3 0.3

0.25 0.25
0.4 0.5 0.6 0.7 0.4 0.5 0.6 0.7
x(t) x(t)
(a) (b)
0.45 0.45

0.4 0.4
y(t)
y(t)

0.35 0.35

0.3 0.3

0.25 0.25
0.4 0.5 0.6 0.7 0.4 0.5 0.6 0.7
x(t) x(t)
(c) (d)

Figure 1: The phase portraits of system (5). Initial values 𝑥0 = 0.4, 𝑦0 = 0.39. Parameters 𝑠 = 1, 𝑐 = 0.4. Delay values (a) 𝜏 = 0, (b) 𝜏 = 5, (c)
𝜏 = 10, and (d) 𝜏 = 15.

(1) −1
𝑊20 (𝜃) 𝑎 𝑎
𝑊20 (𝜃) = ( ) 𝐸2 = −( 11 12 )
(2) 𝑎21 𝑎22
𝑊20 (𝜃)
(1) (1)
𝑖𝑔20 𝑖𝑔20 𝑓20 + (𝛼 + 𝛼) 𝑓11
= 𝑞 (0) 𝑒𝑖𝜔+ 𝜏𝑗 𝜃 + 𝑞 (0) 𝑒−𝑖𝜔+ 𝜏𝑗 𝜃 + 𝐸1 𝑒2𝑖𝜔+ 𝜏𝑗 𝜃 , ×( ).
𝜔+ 𝜏𝑗 3𝜔+ 𝜏𝑗 (2)
𝑓20 (2)
+ (𝛼𝑒𝑖𝜔+ 𝜏𝑗 + 𝛼𝑒−𝑖𝜔+ 𝜏𝑗 ) 𝑓11 (2)
+ 𝛼𝛼𝑓02
(1)
𝑊11 (𝜃) (26)
𝑊11 (𝜃) = ( (2)
)
𝑊11 (𝜃)
Now, let
𝑖𝑔 𝑖𝑔
= − 11 𝑞 (0) 𝑒𝑖𝜔+ 𝜏𝑗 𝜃 + 11 𝑞 (0) 𝑒−𝑖𝜔+ 𝜏𝑗 𝜃 + 𝐸2 , 󵄨 󵄨2
𝑖 󵄨 󵄨2 󵄨󵄨𝑔 󵄨󵄨 𝑔
𝜔+ 𝜏𝑗 𝜔+ 𝜏𝑗 𝑐1 (0) = (𝑔20 𝑔11 − 2󵄨󵄨󵄨𝑔11 󵄨󵄨󵄨 − 󵄨 02 󵄨 ) + 21 , (27)
−1 2𝜔+ 𝜏𝑗 3 2
2𝑖𝜔+ − 𝑎11 −𝑎12
𝐸1 = ( −2𝑖𝜔+ ) which can be determined by the parameters in system (5).
−𝑎21 𝑒 2𝑖𝜔+ − 𝑎22
Defining
(1) (1)
𝑓20 + 2𝛼𝑓11 Re {𝑐1 (0)}
×( (2) (2) (2)
), 𝜇2 = − ,
𝑒−2𝑖𝜔+ 𝜏𝑗 𝑓20 + 2𝛼𝑒−𝑖𝜔+ 𝜏𝑗 𝑓11 + 𝛼2 𝑓02 Re {𝜆󸀠 (𝜏𝑗 )}
6 Abstract and Applied Analysis

0.415 0.39

0.41
0.385
0.405
x(t)

y(t)
0.4
0.38
0.395

0.39 0.375

0.385
0 200 400 600 800 0 200 400 600 800
t t

0.39 0.39

0.385

y(t)
0.38
y(t)

0.38
0.37
0.375 0.41 800
0.4 600
400
0.385 0.39 0.395 0.4 0.405 0.41 x(t) 0.39 200
0 t
x(t)

Figure 2: The trajectories and phase portrait of system (5). Initial values 𝑥0 = 0.4, 𝑦0 = 0.39. Parameters 𝑠 = 1, 𝑐 = 0.6. Delay value 𝜏 = 5.248.

𝛽2 = 2 Re {𝑐1 (0)} , the conditions of Theorem 4(1). By Theorem 4(1), we expect

that 𝐸∗ is stable for any delay value. We further choose four
Im {𝑐1 (0)} + 𝜇2 Im {𝜆󸀠 (𝜏𝑗 )} different delay values: 𝜏𝑎 = 0, 𝜏𝑏 = 5, 𝜏𝑐 = 10, and 𝜏𝑑 =
𝑇2 = − , 15. The computation simulations are depicted in Figure 1.
𝜔+ 𝜏𝑗
Obviously, the computation simulations can support the
(28) result of Theorem 4(1).
Next, we choose parameters 𝑠 = 1, 𝑐 = 0.6. The positive
we have the following result. equilibrium of system (5) is 𝐸∗ = (0.4, 0.3795). It is easy to
show that 𝑎11 +𝑎22 = −0.4795 < 0, 𝑎11 𝑎22 +𝑎12 𝑎21 = −0.0759 <
Theorem 5. The direction of the Hopf bifurcation is deter- 0, which satisfy the conditions of Theorem 4(2). From the
mined by 𝜇2 : if 𝜇2 > 0 (𝜇2 < 0), then the Hopf bifurcation is formulae in the previous section, we can compute the values
supercritical (subcritical) and the bifurcating periodic solutions of 𝜏0 , 𝜇2 , 𝛽2 , and 𝑇2 as
exist for 𝜏 > 𝜏0 (𝜏 < 𝜏0 ); 𝛽2 determines the stability of the
bifurcating periodic solutions: the bifurcating periodic solutions 𝜏0 = 6.048, 𝜇2 = 130888,
are stable (unstable) if 𝛽2 < 0 (𝛽2 > 0); and 𝑇2 determines the (29)
period of the bifurcating periodic solutions: the period increases 𝛽2 = −503.3349, 𝑇2 = 681.8543,
(decreases) if 𝑇2 > 0 (𝑇2 < 0). from which we conclude that 𝐸∗ is asymptotically stable for
𝜏 < 𝜏0 and the Hopf bifurcation of system (5) occurring
5. Numerical Simulations and Discussions at the critical value 𝜏0 is supercritical and the bifurcating
periodic solution exists when 𝜏 crosses 𝜏0 to the right; also
In this section, we present some numerical simulations to the bifurcating periodic solution is stable. By taking 𝜏 =
verify our theoretical results proved in previous sections by 5.248 < 𝜏0 and 𝜏 = 6.049 > 𝜏0 , we can show the computation
using MATLAB DDE solver. We simulate the system (5) with simulations in Figures 2 and 3, respectively. It is easy to
the initial value (𝑥0 , 𝑦0 ) = (0.4, 0.39). In Figures 1–3 the initial see that the computation simulations support the results of
point (𝑥0 , 𝑦0 ) and the positive equilibrium 𝐸∗ are represented Theorem 4(2) and Theorem 5.
by green star and red star, respectively. In the present paper, we have considered a delayed
First, we choose parameters 𝑠 = 1, 𝑐 = 0.4. In this predator-prey system in which the prey species exhibits herd
case, system (5) has only one positive equilibrium 𝐸∗ = behavior and the predator species with quadratic mortal-
(0.6, 0.3098). Simple calculations show that 𝑎11 + 𝑎22 = ity. Our research shows that, for system (5), the positive
−0.7098 < 0, 𝑎11 𝑎22 + 𝑎12 𝑎21 = 0.062 > 0, which satisfy equilibrium 𝐸∗ is always asymptotically stable under certain
Abstract and Applied Analysis 7

0.42 0.39

0.415

0.41 0.385

0.405
x(t)

y(t)
0.38
0.4

0.395
0.375
0.39

0.385 0.37
0 200 400 600 800 1000 0 200 400 600 800 1000
t t

0.4
0.39

0.39
0.385
y(t)
0.38
y(t)

0.38

0.37
0.375 0.42
1000
0.4
0.37 x(t 500
0.38 0.39 0.4 0.41 ) t
0.38 0
x(t)

Figure 3: The trajectories and phase portrait of system (5). Initial values 𝑥0 = 0.4, 𝑦0 = 0.39. Parameters 𝑠 = 1, 𝑐 = 0.6. Delay value 𝜏 = 6.049.

conditions (see Theorem 4(1)). In other words, under such Municipal Education Commission (no. 13ZZ116), and the
conditions, the dynamics behavior of the corresponding ODE Innovation Fund Project For Graduate Student of Shanghai
system is robust with respect to the time delay. But, under (no. JWCXSL1401).
some other conditions (see Theorem 4(2)), 𝐸∗ will lose its
stability and Hopf bifurcations occur when the delay 𝜏 passes
through some critical values. In this case, the time delay has
References
an important effect for the system dynamics. [1] L. Chen, X. Song, and Z. Lu, Mathematical Ecological Models
Our results may enrich the dynamics in the predator- and Research Methods, Sichuan Science and Technology Press,
prey system and help us to better understand the interaction Chengdu, China, 2003.
of predator with prey in a real ecosystem. Further studies [2] C. S. Holling, “The functional response of predators to prey
are necessary to analyze the dynamics of more realistic but density and its role in mimicry and population regulation,”
complex systems, such as delayed diffusive predator-prey Memoirs of the Entomological Society of Canada, vol. 45,
system with herd behavior. supplement S45, pp. 5–60, 1965.
[3] J. F. Andrews, “A mathematical model for the continuous culture
Conflict of Interests of microorganisms utilizing inhibitory substrates,” Biotechnol-
ogy and Bioengineering, vol. 10, no. 6, pp. 707–723, 1968.
The authors declare that there is no conflict of interests [4] J. Beddington, “Mutual interference between parasites or preda-
regarding the publication of this paper. tors and its effect on searching effciency,” Journal of Animal
Ecology, vol. 44, no. 1, pp. 331–340, 1975.
Acknowledgments [5] D. L. DeAngelis, R. A. Goldsten, and R. V. O’Neill, “A model for
trophic interaction,” Ecology, vol. 56, no. 4, pp. 881–892, 1975.
The authors wish to thank the handling editor and reviewers [6] R. E. Kooij and A. Zegeling, “A predator-prey model with Ivlev’s
for their valuable comments and suggestions which lead to functional response,” Journal of Mathematical Analysis and
truly significant improvement of the paper. This research is Applications, vol. 198, no. 2, pp. 473–489, 1996.
supported by the National Natural Science Foundation of [7] R. Arditi and L. R. Ginzburg, “Coupling in predator-prey
China (no. 11271260), Shanghai Leading Academic Discipline dynamics: ratio-dependence,” Journal of Theoretical Biology, vol.
Project (no. XTKX2012), Innovation Program of Shanghai 139, no. 3, pp. 311–326, 1989.
8 Abstract and Applied Analysis

[8] Y. Kuang and E. Beretta, “Global qualitative analysis of a ratio- [26] M. Baurmann, T. Gross, and U. Feudel, “Instabilities in spatially
dependent predator-prey system,” Journal of Mathematical Biol- extended predator-prey systems: spatio-temporal patterns in
ogy, vol. 36, no. 4, pp. 389–406, 1998. the neighborhood of Turing-Hopf bifurcations,” Journal of
[9] S. Ruan and D. Xiao, “Global analysis in a predator-prey system Theoretical Biology, vol. 245, no. 2, pp. 220–229, 2007.
with nonmonotonic functional response,” SIAM Journal on [27] N. MacDonald, Biological Delay Systems: Linear Stability Theory,
Applied Mathematics, vol. 61, no. 4, pp. 1445–1472, 2000. Cambridge University Press, Cambridge, UK, 1989.
[10] D. Xiao and H. Zhu, “Multiple focus and Hopf bifurcations in a [28] J. K. Hale and S. M. V. Lunel, Introduction to Functional-
predator-prey system with nonmonotonic functional response,” Differential Equations, Springer, New York, NY, USA, 1993.
SIAM Journal on Applied Mathematics, vol. 66, no. 3, pp. 802– [29] Y. Kuang, Delay Differential Equations with Applications in
819, 2006. Population Dynamics, Academic Press, Boston, Mass, USA,
[11] H. Zhu, S. A. Campbell, and G. S. K. Wolkowicz, “Bifurcation 1993.
analysis of a predator-prey system with nonmonotonic func- [30] J. Wu, Theory and Applications of Partial Functional-Differential
tional response,” SIAM Journal on Applied Mathematics, vol. 63, Equations, Springer, New York, NY, USA, 1996.
no. 2, pp. 636–682, 2002.
[31] R. Xu, Q. Gan, and Z. Ma, “Stability and bifurcation analysis on a
[12] X.-A. Zhang, L. Chen, and A. U. Neumann, “The stage-
ratio-dependent predator-prey model with time delay,” Journal
structured predator-prey model and optimal harvesting policy,”
of Computational and Applied Mathematics, vol. 230, no. 1, pp.
Mathematical Biosciences, vol. 168, no. 2, pp. 201–210, 2000.
187–203, 2009.
[13] M. A. Aziz-Alaoui and M. D. Okiye, “Boundedness and global
[32] Y. Song and S. Yuan, “Bifurcation analysis in a predator-
stability for a predator-prey model with modified Leslie-Gower
prey system with time delay,” Nonlinear Analysis: Real World
and Holling-type II schemes,” Applied Mathematics Letters, vol.
Applications, vol. 7, no. 2, pp. 265–284, 2006.
16, no. 7, pp. 1069–1075, 2003.
[14] A. F. Nindjin, M. A. Aziz-Alaoui, and M. Cadivel, “Analysis of a [33] Y. Song and J. Wei, “Local Hopf bifurcation and global periodic
predator-prey model with modified Leslie-Gower and Holling- solutions in a delayed predator-prey system,” Journal of Mathe-
type II schemes with time delay,” Nonlinear Analysis: Real World matical Analysis and Applications, vol. 301, no. 1, pp. 1–21, 2005.
Applications, vol. 7, no. 5, pp. 1104–1118, 2006. [34] D. Xiao and S. Ruan, “Multiple bifurcations in a delayed
[15] L. Han, Z. Ma, and H. W. Hethcote, “Four predator prey models predator-prey system with nonmonotonic functional response,”
with infectious diseases,” Mathematical and Computer Mod- Journal of Differential Equations, vol. 176, no. 2, pp. 494–510,
elling, vol. 34, no. 7-8, pp. 849–858, 2001. 2001.
[16] R. Xu and Z. Ma, “Stability and Hopf bifurcation in a predator- [35] S. Ruan, “Absolute stability, conditional stability and bifurca-
prey model with stage structure for the predator,” Nonlinear tion in Kolmogorov-type predator-prey systems with discrete
Analysis: Real World Applications, vol. 9, no. 4, pp. 1444–1460, delays,” Quarterly of Applied Mathematics, vol. 59, no. 1, pp. 159–
2008. 173, 2001.
[17] M. Liu and K. Wang, “Global stability of stage-structured [36] R. Xu, M. A. J. Chaplain, and F. A. Davidson, “Periodic solutions
predator-prey models with Beddington-DeAngelis functional for a delayed predator-prey model of prey dispersal in two-
response,” Communications in Nonlinear Science and Numerical patch environments,” Nonlinear Analysis: Real World Applica-
Simulation, vol. 16, no. 9, pp. 3792–3797, 2011. tions, vol. 5, no. 1, pp. 183–206, 2004.
[18] C. Ji, D. Jiang, and X. Li, “Qualitative analysis of a stochastic [37] Z. Liu and R. Yuan, “Stability and bifurcation in a delayed
ratio-dependent predatorprey system,” Journal of Computa- predator-prey system with Beddington-DeAngelis functional
tional and Applied Mathematics, vol. 235, no. 5, pp. 1326–1341, response,” Journal of Mathematical Analysis and Applications,
2011. vol. 296, no. 2, pp. 521–537, 2004.
[19] V. Ajraldi, M. Pittavino, and E. Venturino, “Modeling herd [38] M. Fan and K. Wang, “Periodicity in a delayed ratio-dependent
behavior in population systems,” Nonlinear Analysis: Real World predator-prey system,” Journal of Mathematical Analysis and
Applications, vol. 12, no. 4, pp. 2319–2338, 2011. Applications, vol. 262, no. 1, pp. 179–190, 2001.
[20] P. A. Braza, “Predator-prey dynamics with square root func- [39] E. Beretta and Y. Kuang, “Global analyses in some delayed ratio-
tional responses,” Nonlinear Analysis: Real World Applications, dependent predator-prey systems,” Nonlinear Analysis: Theory,
vol. 13, no. 4, pp. 1837–1843, 2012. Methods and Applications, vol. 32, no. 3, pp. 381–408, 1998.
[21] E. Venturino and S. Petrovskii, “Spatiotemporal behavior of a [40] T. Zhao, Y. Kuang, and H. L. Smith, “Global existence of
prey-predator system with a group defense for prey,” Ecological periodic solutions in a class of delayed gause-type predator-prey
Complexity, vol. 14, pp. 37–47, 2013. systems,” Nonlinear Analysis: Theory, Methods and Applications,
[22] S. Belvisi and E. Venturino, “An ecoepidemic model with dis- vol. 28, no. 8, pp. 1373–1394, 1997.
eased predators and prey group defense,” Simulation Modelling [41] B. D. Hassard, N. D. Kazarinoff, and Y. H. Wan, Theory and
Practice and Theory, vol. 34, pp. 144–155, 2013. Applications of Hopf Bifurcation, Cambridge University Press,
[23] S. Yuan, C. Xu, and T. Zhang, “Spatial dynamics in a predator- Cambridge, UK, 1981.
prey model with herd behavior,” Chaos, vol. 23, no. 3, Article ID
033102, 2013.
[24] S. J. Brentnall, K. J. Richards, J. Brindley, and E. Murphy,
“Plankton patchiness and its effect on larger-scale productivity,”
Journal of Plankton Research, vol. 25, no. 2, pp. 121–140, 2003.
[25] E. A. Fulton, A. D. M. Smith, and C. R. Johnson, “Mortality
and predation in ecosystem models: is it important how these
are expressed?” Ecological Modelling, vol. 169, no. 1, pp. 157–178,
2003.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 753614, 9 pages
http://dx.doi.org/10.1155/2014/753614

Research Article
Existence of Wave Front Solutions of an Integral Differential
Equation in Nonlinear Nonlocal Neuronal Network

Lijun Zhang,1,2 Linghai Zhang,3 Jie Yuan,3 and C. M. Khalique2

1
Department of Mathematics, School of Science, Zhejiang Sci-Tech University, Hangzhou, Zhejiang 310018, China
2
Department of Mathematical Sciences, International Institute for Symmetry Analysis and Mathematical Modelling, North-West
University, Mafikeng Campus, Private Bag X 2046, Mmabatho 2735, South Africa
3
Department of Mathematics, Lehigh University, 14 E Packer Avenue, Bethlehem, PA 18015, USA

Correspondence should be addressed to Lijun Zhang; li-jun0608@163.com

Received 2 February 2014; Accepted 12 March 2014; Published 7 April 2014

Academic Editor: Sanling Yuan

Copyright © 2014 Lijun Zhang et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

An integral-differential model equation arising from neuronal networks with very general kernel functions is considered in this
paper. The kernel functions we study here include pure excitations, lateral inhibition, lateral excitations, and more general synaptic
couplings (e.g., oscillating kernel functions). The main goal of this paper is to prove the existence and uniqueness of the traveling
wave front solutions. The main idea we apply here is to reduce the nonlinear integral-differential equation into a solvable differential
equation and test whether the solution we get is really a wave front solution of the model equation.

1. Introduction 𝐻(𝑢 − 𝜃) denotes the output firing rate of a neuron, which

means that a neuron fires at its maximum rate when the
1.1. The Model Equation and Its Biological Background. To potential exceeds a threshold and does not fire otherwise.
describe and study the propagation of nerve impulses in 𝐾(𝑥) represents synaptic coupling between neurons in the
synaptically coupled neuronal networks, some integral dif- tissue. The weight function 𝐾(𝑥), also named as kernel func-
ferential mathematical model equations have been proposed tion now, considered in [1] is Mexican hat function because
and their traveling wave solutions have been studied in the the field considered in his paper is of lateral inhibition type.
recent thirty years since traveling waves share the same See Atay and Hutt [2], Coombes et al. [3] Ermentrout [4]
properties as nerve impulses. In 1977, to study the dynamics for more details of the biological background. Due to the
of pattern formation in lateral-inhibition type homogeneous finite propagation velocity, the following nonlocal nonlinear
neural fields with general connections, Amari [1] derived the scalar integral model equation incorporating spatial temporal
following nonlocal equation by statistical considerations: delay was proposed to describe the dynamics of an effective
postsynaptic potential 𝑢(𝑥, 𝑡) at position 𝑥 and time 𝑡:
𝑢𝑡 + 𝑢 = ∫ 𝐾 (𝑥 − 𝑦) 𝐻 (𝑢 (𝑦, 𝑡) − 𝜃) 𝑑𝑦. (1)
𝑅 1 󵄨󵄨 󵄨
𝑢𝑡 + 𝑢 = 𝛼 ∫ 𝐾 (𝑥 − 𝑦) 𝐻 (𝑢 (𝑦, 𝑡 − 󵄨𝑥 − 𝑦󵄨󵄨󵄨) − 𝜃) 𝑑𝑦
𝑅 𝑐󵄨
This model equation can be used to describe the simple
fields which are one dimensional and homogeneous, have + 𝐼 (𝑥, 𝑡) ,
negligible time lag, and consist of only one layer. In this (2)
model equation, 𝑢(𝑥, 𝑡) represents the average membrane
potential of the neurons located at position 𝑥 and at time 𝑡, where 0 < 𝑐 ≤ ∞, 𝛼 > 0, and 𝜃 > 0 are constants. The
the nonlinear output function 𝐻(𝑢) is always chosen to be parameter 𝛼 denotes the synaptic rate constant in a neuronal
the Heaviside gain function: 𝐻(𝑢 − 𝜃) = 0 for all 𝑢 < 𝜃, network, 𝜃 represents the threshold for excitation of the
𝐻(0) = 1/2, and 𝐻(𝑢 − 𝜃) = 1 for all 𝑢 > 𝜃. Here neuronal network, 𝑐 represents the finite propagation speed
2 Abstract and Applied Analysis

of action potentials along axons, and |𝑥 − 𝑦|/𝑐 denotes the The kernel functions are always supposed to be contin-
spatial temporal delay. There are also some related nonlinear uous at 𝑥 = 0, almost everywhere smooth, and satisfy the
singularly perturbed systems of integral-differential model following conditions (𝑆1 ):
equations proposed for the study of synaptically coupled
0
neuronal networks. Pinto and Ermentrout [5, 6] proposed 1
∫ 𝐾 (𝑥) 𝑑𝑥 = ,
and studied the system: −∞ 2
+∞
𝑢𝑡 + 𝑢 + 𝑤 = ∫ 𝐾 (𝑥 − 𝑦) 𝐻 (𝑢 (𝑦, 𝑡) − 𝜃) 𝑑𝑦, ∫ 𝐾 (𝑥) 𝑑𝑥 = 1,
𝑅 −∞ (6)
(3)
0
𝑤𝑡 = 𝜖 (𝑢 − 𝛾𝑤) . ∫ |𝑠| 𝐾 (𝑠) 𝑑𝑠 > 0;
−∞
In [7], Hutt proposed and considered the model equation:
|𝐾 (𝑥)| ≤ 𝑘 exp (−𝜌𝑥) in 𝑅,
1 󵄨󵄨 󵄨
𝑢𝑡 + 𝑢 = 𝛼 ∫ 𝐾 (𝑥 − 𝑦) 𝐻 (𝑢 (𝑦, 𝑡 − 󵄨𝑥 − 𝑦󵄨󵄨󵄨) − 𝜃) 𝑑𝑦 where 𝑘 and 𝜌 are positive constants.
𝑅 𝑐󵄨 Under the above assumptions, Zhang [11] studied the
existence, uniqueness, and stability of traveling wave solu-
+ 𝛽 ∫ 𝐽 (𝑥 − 𝑦) 𝐻 (𝑢 (𝑦, 𝑡 − 𝜏) − 𝜃) 𝑑𝑦, tions to the model equation (5) for three typical classes of
𝑅
(4) kernel functions. Lv and Wang [12] studied the existence and
uniqueness of traveling waves of the same equation for five
which is a modification and generalization of some known classes of oscillatory kernels and recently, and Magpantay and
models. Here 𝑐 and 𝜏 are the propagation velocity of action Zou [13] studied the model equation (4). The kernel function
potential and feedback delay, respectively. Hutt considered in the feedback channel was assumed to be nonnegative
both the nonlocal axonal connections and nonlocal feedback (pure excitation), and the kernel function in the synaptic
connections with a time delay and examined briefly the coupling that was considered in their paper [13] included
dependence of the speed of the front on various parameters types (A), (B), and (C) and pure inhibition type. Recently,
in this model equation. He also assumed the same firing some similar equations with different fire rate functions [14–
threshold for all neurons function and thus the transfer 16] were proposed and investigated to model the pattern
function was chosen to be the Heaviside step function. Some formation in neuronal networks. These equations take the
more general model equations have been introduced and form:
attracted a lot of research interest recently. See, for example,
[2, 3, 5, 6, 8, 9]. 𝑢𝑡 + 𝑢 = ∫ 𝐾 (𝑥 − 𝑦) 𝑓 (𝑢 (𝑦, 𝑡)) 𝑑𝑦, (7)
𝑅
Clearly, all these model equations (1)–(4) we mentioned
above are related to the equation: where

1 󵄨󵄨 󵄨 −𝑟
𝑢𝑡 + 𝑢 = 𝛼 ∫ 𝐾 (𝑥 − 𝑦) 𝐻 (𝑢 (𝑦, 𝑡 − 󵄨𝑥 − 𝑦󵄨󵄨󵄨) − 𝜃) 𝑑𝑦, 𝑓 (𝑢 (𝑦, 𝑡)) = 2 exp ( ) 𝐻 (𝑢 (𝑦, 𝑡) − 𝜃) , (8)
𝑅 𝑐󵄨 (𝑢 (𝑦, 𝑡) − 𝜃)
2

(5)
and the coupling function is
which can be derived by (2) when 𝐼(𝑥, 𝑡) is constant or by (4) 󵄨 󵄨 󵄨 󵄨
when 𝛽 = 0. As 𝑐 approaches to infinity and (5) reduces to 𝐾 (𝑥, 𝑦) = exp (−𝑏 󵄨󵄨󵄨𝑥 − 𝑦󵄨󵄨󵄨) (𝑏 sin 󵄨󵄨󵄨𝑥 − 𝑦󵄨󵄨󵄨 + cos (𝑥 − 𝑦)) (9)
(1). Equation (5) is an important model equation because the
wave fronts of (4) can be obtained by its solutions though it which is an oscillatory function. It was showed [17] that
is a special case of (4) [10]. the oscillatory nature of the coupling function is likely to
The studies on these model equations mostly focused lead to novel behavior. Consequently, it is reasonable and
on the following three typical classes of kernel functions in meaningful to proceed with the study of the model equation
previous literature. (5) with oscillatory kernel functions. Motivated by their
pioneer works [11–13, 18–20]; in this paper, we aim to study
(A) The first class consists of nonnegative kernel functions the existence and uniqueness of the wave front solutions of
(pure excitation). IDE (5) with more general kernel functions.
(B) The second class consists of Mexican hat kernel
1.2. Mathematical Assumption of the Kernel Function. In
functions (lateral inhibition); that is, 𝐾 ≥ 0 on
addition to the basic assumptions for kernel functions, we
(−𝑀, 𝑁) and 𝐾 ≤ 0 on (−∞, −𝑀) ∪ (𝑁, ∞), for some
assume that the kernel function 𝐾(𝑥) on (−∞, 0) satisfies one
positive constants 𝑀 and 𝑁.
of the following conditions:
(C) The third class consists of upside down Mexican hat
kernel functions (lateral excitation); that is, 𝐾 ≤ 0 on (𝐿 1 ) 𝐾(𝑥) ≥ 0 for all 𝑥 ∈ (−∞, 0).
(−𝑀, 𝑁) and 𝐾 ≥ 0 on (−∞, −𝑀) ∪ (𝑁, ∞), for two (𝐿 2 ) 𝐾(𝑥) ≥ 0 for 𝑥 ∈ (−𝑀1 , 0) ∪ (−𝑀3 , −𝑀2 ) ∪
positive constants 𝑀 and 𝑁. (−𝑀5 , −𝑀4 ) ∪ ⋅ ⋅ ⋅ ∪ (−𝑀2𝑛+1 , −𝑀2𝑛 ), 𝐾(𝑥) ≤ 0 for
Abstract and Applied Analysis 3

𝑥 ∈ (−𝑀2 , −𝑀1 ) ∪ (−𝑀4 , −𝑀3 ) ∪ ⋅ ⋅ ⋅ ∪ (−𝑀2𝑛 , (𝑅6 ) 𝐾(𝑥) ≥ 0 for 𝑥 ∈ [0, 𝑁1 ] ∪ [𝑁2 , 𝑁3 ] ∪ ⋅ ⋅ ⋅ ∪ [𝑁2𝑛 , +∞),
−𝑀2𝑛−1 ) ∪ (−∞, −𝑀2𝑛+1 ), where 0 < 𝑀1 < 𝑀2 < 𝐾(𝑥) ≤ 0 for 𝑥 ∈ (𝑁1 , 𝑁2 ) ∪ (𝑁3 , 𝑁4 ) ∪ ⋅ ⋅ ⋅ ∪ (𝑁2𝑛−1 ,
⋅ ⋅ ⋅ < 𝑀2𝑛+1 < ∞ and 𝑁2𝑛 ), where 0 < 𝑁1 < 𝑁2 < ⋅ ⋅ ⋅ < 𝑁2𝑛 < ∞, and
𝑁2𝑖
−𝑀2𝑖−2 𝛼
∫ |𝑠| 𝐾 (𝑠) 𝑑𝑠 ≥ 0, 𝛼∫ 𝐾 (𝑠) 𝑑𝑠 > 𝜃 − , 𝑖 = 1, 2, . . . 𝑛. (15)
−𝑀2𝑖 0 2
(10)
0
𝛼 Obviously, the kernel functions satisfying one of the
− 𝛼∫ 𝐾 (𝑡) 𝑑𝑡 < 𝜃, conditions (𝑅𝑖 ) (𝑖 = 1, 2, . . . , 6) on (0, +∞) and one of the
2 −𝑀2𝑖
conditions (𝐿 𝑗 ) (𝑗 = 1, 2, 3, 4) on (−∞, 0) form a very general
class of functions including almost all the classes of the kernel
where 𝑖 = 1, 2, . . . , 𝑛, 𝑛 + 1 and 𝑀0 = 0, 𝑀2𝑛+1 = ∞. functions found in previous literatures. For example, if the
(𝐿 3 ) 𝐾(𝑥) ≥ 0 for 𝑥 ∈ (−𝑀1 , 0) ∪ (−𝑀3 , −𝑀2 ) ∪ kernel function 𝐾(𝑥) satisfies (𝐿 4 ) and (𝑅3 ) with 𝑛 = 1
(−𝑀5 , −𝑀4 ) ∪ ⋅ ⋅ ⋅ ∪ (−𝑀2𝑛−1 , −𝑀2𝑛−2 ) ∪ (−∞, −𝑀2𝑛 ), (see Section 3), then it is an upside down Mexican hat kernel
𝐾(𝑥) ≤ 0 for 𝑥 ∈ (−𝑀2 , −𝑀1 ) ∪ (−𝑀4 , −𝑀3 ) ∪ ⋅ ⋅ ⋅ ∪ functions actually; if the kernel function 𝐾(𝑥) satisfies (𝐿 3 )
(−𝑀2𝑛 , −𝑀2𝑛−1 ), where 0 < 𝑀1 < 𝑀2 < ⋅ ⋅ ⋅ < 𝑀2𝑛 < and (𝑅5 ) with 𝑛 = 1, then it is the case (A) in [12].
∞ and We prove the existence and uniqueness of the traveling
wave solution of the model equation (5) for more general
0
classes of kernel functions including not only the kernel
∫ |𝑠| 𝐾 (𝑠) 𝑑𝑠 ≥ 0,
−𝑀2𝑖 functions studied in [11, 12] but also the oscillating kernel
(11) functions. The main idea in this paper is employing the speed
0
𝛼 index functions (the main idea in [11, 19] and other pioneering
− 𝛼∫ 𝐾 (𝑡) 𝑑𝑡 < 𝜃,
2 −𝑀2𝑖 works).

where 𝑖 = 1, 2, . . . , 𝑛. 2. Preliminary Analysis

(𝐿 4 ) 𝐾(𝑥) ≥ 0 for 𝑥 ∈ (−∞, −𝑀) and 𝐾(𝑥) ≤ 0 for 𝑥 ∈ It is well known that the traveling wave solutions of an
(−𝑀, 0), where 0 < 𝑀 < ∞. equation are the solutions of the form 𝑢(𝑥, 𝑡) = 𝑈(𝑧), where
𝑧 = 𝑥 + 𝜇𝑡 is the moving coordinate and 𝜇 is a constant
Assume that the kernel function 𝐾(𝑥) on (0, ∞) satisfies which represents the speed of the traveling wave. There are
one of the following conditions: two kinds of traveling waves which possess some important
and practical meanings in neural network. These are traveling
(𝑅1 ) 𝐾(𝑥) ≥ 0 for all 𝑥 ∈ (0, +∞).
wave fronts and traveling pulses. In this paper, we mainly
(𝑅2 ) 𝐾(𝑥) ≥ 0 for 𝑥 ∈ (𝑁1 , 𝑁2 ) ∪ (𝑁3 , 𝑁4 ) ∪ ⋅ ⋅ ⋅ ∪ (𝑁2𝑛+1 , focus on the traveling wave fronts.
+∞), 𝐾(𝑥) ≤ 0 for 𝑥 ∈ [0, 𝑁1 ] ∪ [𝑁2 , 𝑁3 ] ∪ ⋅ ⋅ ⋅ ∪ To study the traveling wave solutions of the integral-
[𝑁2𝑛 , 𝑁2𝑛+1 ], where 0 < 𝑁1 < 𝑁2 < ⋅ ⋅ ⋅ < 𝑁2𝑛+1 < ∞, differential equation (5), we suppose that 𝑢(𝑥, 𝑡) = 𝑈(𝑥+𝜇𝑡) =
and 𝑈(𝑧), and then we have
𝑁2𝑖−1
𝛼 𝜇𝑈󸀠 + 𝑈
𝛼∫ 𝐾 (𝑠) 𝑑𝑠 > 𝜃 − , 𝑖 = 1, 2, . . . 𝑛, 𝑛 + 1. (12)
0 2
𝜇 󵄨󵄨 󵄨
= 𝛼 ∫ 𝐾 (𝑥 − 𝑦) 𝐻 (𝑈 (𝑦 + 𝜇𝑡 − 󵄨𝑥 − 𝑦󵄨󵄨󵄨) − 𝜃) 𝑑𝑦.
𝑅 𝑐󵄨
(𝑅3 ) 𝐾(𝑥) ≥ 0 for 𝑥 ∈ (𝑁1 , 𝑁2 ) ∪ (𝑁3 , 𝑁4 ) ∪ ⋅ ⋅ ⋅ ∪ (𝑁2𝑛−1 , (16)
𝑁2𝑛 ), 𝐾(𝑥) ≤ 0 for 𝑥 ∈ [0, 𝑁1 ] ∪ [𝑁2 , 𝑁3 ] ∪ ⋅ ⋅ ⋅ ∪
[𝑁2𝑛 , +∞), where 0 < 𝑁1 < 𝑁2 < ⋅ ⋅ ⋅ < 𝑁2𝑛 < ∞, It is easy to see that by transformation of variables the integral
and IDE (16) can be written as
𝑁2𝑖−1
𝛼 𝜇 󵄨󵄨 󵄨
𝛼∫ 𝐾 (𝑠) 𝑑𝑠 > 𝜃 − , 𝑖 = 1, 2, . . . 𝑛. (13) 𝜇𝑈󸀠 + 𝑈 = 𝛼 ∫ 𝐾 (𝑧 − 𝑦) 𝐻 (𝑈 (𝑦 − 󵄨𝑧 − 𝑦󵄨󵄨󵄨) − 𝜃) 𝑑𝑦.
0 2 𝑅 𝑐󵄨
(17)
(𝑅4 ) 𝐾(𝑥) ≥ 0 for 𝑥 ∈ (𝑁, +∞) and 𝐾(𝑥) ≤ 0 for 𝑥 ∈ Let 𝑡 = 𝑦 − (𝜇/𝑐)|𝑧 − 𝑦| and 𝜇 < 𝑐 in the first term of the right
(0, 𝑁], where 0 < 𝑁 < ∞. side of the above equation, and then we get
(𝑅5 ) 𝐾(𝑥) ≥ 0 for 𝑥 ∈ [0, 𝑁1 ] ∪ [𝑁2 , 𝑁3 ] ∪ ⋅ ⋅ ⋅ ∪ [𝑁2𝑛−2 ,
𝑁2𝑛−1 ], 𝐾(𝑥) ≤ 0 for 𝑥 ∈ (𝑁1 , 𝑁2 ) ∪ (𝑁3 , 𝑁4 ) ∪ ⋅ ⋅ ⋅ ∪ 𝑧−𝑡
𝜇𝑈󸀠 + 𝑈 = 𝛼 ∫ 𝐾 ( ) 𝐻 (𝑈 (𝑡) − 𝜃) 𝑑𝑡.
(𝑁2𝑛+1 , +∞), where 0 < 𝑁1 < 𝑁2 < ⋅ ⋅ ⋅ < 𝑁2𝑛+1 < ∞, 𝑅 1 + sgn (𝑧 − 𝑡) (𝜇/𝑐)
and (18)
𝑁2𝑖
𝛼 Obviously, the nonlinear terms are on the left side of (18).
𝛼∫ 𝐾 (𝑠) 𝑑𝑠 > 𝜃 − , 𝑖 = 1, 2, . . . 𝑛 − 1. (14)
From the special property of Heaviside step function, we
0 2
4 Abstract and Applied Analysis

know that (18) can be simplified if some properties of function hat function, or upside down Mexican hat function. Lv and
𝑈(𝑥) are known. Let 𝑅[𝑈, 𝜃] = {𝑥 | 𝑈(𝑥) > 𝜃}, and then (18) Wang [12] also studied the same equation with five kinds of
can be rewritten as kernel functions. We prove the existence and uniqueness of
the traveling wave front of this equation for more general
𝑧−𝑡
𝜇𝑈󸀠 + 𝑈 = 𝛼 ∫ 𝐾( ) 𝑑𝑡. (19) kernel functions under less constraints.
𝑅[𝑈,𝜃] 1 + sgn (𝑧 − 𝑡) (𝜇/𝑐) Based on the analysis above, we know that function (22)
is a wave front satisfying 𝑈(0) = 𝜃 only if 𝑈(𝑧) ≤ 𝜃 for 𝑧 ∈
Firstly, we study the equilibria of (19). Suppose that 𝑈 is (−∞, 0) and 𝑈(𝑧) ≥ 𝜃 for 𝑧 ∈ (0, +∞). Substituting 𝑧 = 0
a constant, and then 𝑅[𝑈, 𝜃] = R if 𝑈 > 𝜃 and 𝑅[𝑈, 𝜃] = into (22) and letting 𝑈(0) = 𝜃, we get
0 if 𝑈 < 𝜃. So the constant solutions of (5) are 0 and 𝛼 if
0 < 𝜃 < 𝛼. According to the property of traveling wave front, 𝛼 0
𝑠 𝑐 𝑐
we know that if 𝑈(𝑧) is a traveling wave front of (5), 𝑈+ = − 𝛼 ∫ exp ( ) 𝐾 ( 𝑠) 𝑑𝑠 = 𝜃.
2 −∞ 𝜇 𝑐 + sgn (𝑠) 𝜇 𝑐 + sgn (𝑠) 𝜇
lim𝑧 → +∞ 𝑈(𝑧) and 𝑈− = lim𝑧 → −∞ 𝑈(𝑧), 𝑈+ > 𝑈− , and then
𝑈+ , 𝑈− are two different constant solutions of (5) and (18), (24)
respectively. So 𝑈+ = 𝛼 and 𝑈− = 0.
If 𝑈(𝑧) is a traveling wave front of (5) satisfying (0) = 𝜃, So it is one of the necessary conditions, under which 𝑈(𝑧)
𝑈(𝑧) < 𝜃 for 𝑧 ∈ (−∞, 0), and 𝑈(𝑧) > 𝜃 for 𝑧 ∈ (0, +∞), then might be a traveling wave solution of the model equation (5)
(19) can be reduced to and 𝑈(0) = 𝜃, and that there exists a wave speed 𝜇 satisfying
(24). Following the definition in [11], we also name (24) as
+∞
𝑧−𝑡 wave speed equation. We now prove that, for a more general
𝜇𝑈󸀠 + 𝑈 = 𝛼 ∫ 𝐾( ) 𝑑𝑡, (20) kind of kernel functions 𝐾(𝑥), there is a unique 𝜇∗ satisfying
0 1 + sgn (𝑧 − 𝑡) (𝜇/𝑐)
the speed equation; that is,
which can be further rewritten as
0 𝑐−𝜇 𝛼
𝑐𝑧/(𝑐+sgn(𝑧)𝜇) 𝛼∫ exp ( 𝑠) 𝐾 (𝑠) 𝑑𝑠 = − 𝜃, 0 < 𝜇 < 𝑐. (25)
󸀠
𝜇𝑈 + 𝑈 = 𝛼 ∫ 𝐾 (𝑡) 𝑑𝑡. (21) −∞ 𝑐𝜇 2
−∞

Obviously, (21) is a linear ordinary differential equation 3.1. Existence and Uniqueness of the Wave Speed. At first,
having two equilibria 0 and 𝛼, which means that the wave we give the following lemma to prove the existence and
front of (5) satisfying 𝑈(0) = 𝜃, 𝑈(𝑧) < 𝜃 for 𝑧 ∈ (−∞, 0), 𝜃 < uniqueness of the wave speed, that is to say, to prove that there
𝑈(𝑧) for 𝑧 ∈ (0, +∞) must have the limits as lim𝑧 → +∞ 𝑈(𝑧) = exists a unique wave speed 𝜇∗ such that 𝑈(0) = 𝜃, which is
𝛼 and lim𝑧 → −∞ 𝑈(𝑧) = 0. Solving (21), we get the solution one of the necessary conditions under which function (22)
satisfying the limit conditions mentioned above as follows: might be a traveling wave front of the IDE (5). Actually,
𝑈(0) = 𝜃 is equivalent to (24), which implies that we can
𝑐𝑧/(𝑐+sgn(𝑧)𝜇) prove the existence and uniqueness of the wave speed by
𝑈 (𝑧) = 𝛼 ∫ 𝐾 (𝑠) 𝑑𝑠 proving that the wave speed equation (24) has a unique
−∞ 0
𝑧
(22) solution. Denote 𝑓(𝐴) = 𝛼 ∫−∞ exp(𝑠/𝐴)𝐾(𝑠)𝑑𝑠, and then we
𝑠−𝑧 𝑐
− 𝛼 ∫ exp ( ) 𝐾 (𝑠) 𝑑𝑠, have the following Lemma.
−∞ 𝜇 𝑐 + sgn (𝑠) 𝜇
Lemma 1. Suppose that the positive parameters 𝛼 and 𝜃 satisfy
𝛼 𝑧 𝑠−𝑧 𝑐
𝑈󸀠 = ∫ exp ( ) 𝐾 (𝑠) 𝑑𝑠. (23) the condition 0 < 2𝜃 < 𝛼. If the function 𝐾(𝑥) satisfies the
𝜇 −∞ 𝜇 𝑐 + sgn (𝑠) 𝜇 condition (𝑆1 ) and one of (𝐿 𝑖 ) (𝑖 = 1, 2, 3, 4), then there exists
a unique 𝐴 0 ∈ (0, ∞) such that 𝑓(𝐴 0 ) = 𝛼/2 − 𝜃.
Notice that the function (22) we obtained is just a solution
of (21), but not necessarily a solution of (16); that is to say, it Proof. It is easy to see that lim𝐴 → 𝑜+ 𝑓(𝐴) = 0 and
is not necessarily a traveling wave front of (5). The function lim𝐴 → +∞ 𝑓(𝐴) = 𝛼/2. So we have
(22) could be a traveling wave front of (5) only if it satisfies
𝑈(0) = 𝜃, 𝑈(𝑧) < 𝜃 for 𝑧 ∈ (−∞, 0), 𝜃 < 𝑈(𝑧) for 𝑧 ∈ 𝛼
(0, +∞). We investigate these problems to study the existence lim 𝑓 (𝐴) < − 𝜃 < lim 𝑓 (𝐴) . (26)
𝐴 → 𝑜+ 2 𝐴 → +∞
and uniqueness of the wave solution of IDE (5) under some
conditions in the following sections. Since 𝑓(𝐴) is a continuous function on (0, +∞) with respect
to 𝐴, there exists 𝐴 0 ∈ (0, ∞) such that 𝑓(𝐴 0 ) = 𝛼/2 − 𝜃. We
3. Existence and Uniqueness of the Wave Front now prove the uniqueness of 𝐴 0 when 𝐾(𝑥) satisfies each of
the first three cases on (−∞, 0) by proving that 𝑓󸀠 (𝐴) > 0.
In this section, we study the existence and uniqueness of the
wave front solution to model equation (5); that is to say, we Case 1. Obviously, as 𝐾(𝑥) satisfies (𝐿 1 ), that is, 𝐾(𝑥) ≥ 0 for
study the solution which tends from 0 to 𝛼 as 𝑧 goes from all 𝑥 ∈ (−∞, 0), we have
−∞ to ∞. For this equation, Zhang [11] studied the existence
0
and uniqueness of the traveling wave front for the case that 𝑠
the kernel function 𝐾(𝑥) is a nonnegative function, Mexican ∫ |𝑠| exp ( ) 𝐾 (𝑠) 𝑑𝑠 > 0. (27)
−∞ 𝐴
Abstract and Applied Analysis 5

−𝑀2𝑛
Case 2. As 𝐾(𝑥) satisfies (𝐿 2 ); that is, there exist 0 = 𝑀0 < 𝑠
+∫ |𝑠| exp ( ) 𝐾 (𝑠) 𝑑𝑠
𝑀1 < 𝑀2 < ⋅ ⋅ ⋅ < 𝑀2𝑛+1 < 𝑀2𝑛+2 = ∞, such that 𝐾(𝑥) ≥ 0 −∞ 𝐴
for 𝑥 ∈ (−𝑀1 , 0) ∪ (−𝑀3 , −𝑀2 ) ∪ (−𝑀5 , −𝑀4 ) ∪ ⋅ ⋅ ⋅∪ (−𝑀2𝑛+1 ,
0
−𝑀2𝑛 ), 𝐾(𝑥) ≤ 0 for 𝑥 ∈ (−𝑀2 , −𝑀1 ) ∪ (−𝑀4 , −𝑀3 ) ∪ −𝑀1
≥ exp ( ) ∫ |𝑠| 𝐾 (𝑠) 𝑑𝑠
⋅ ⋅ ⋅ ∪ (−𝑀2𝑛 , −𝑀2𝑛−1 ) ∪ (−∞, −𝑀2𝑛+1 ), and for any 𝑖, 𝑖 = 𝐴 −𝑀2
1, 2, . . . , 𝑛, 𝑛 + 1,
−𝑀2
−𝑀3
−𝑀2𝑖−2 0 + exp ( )∫ |𝑠| 𝐾 (𝑠) 𝑑𝑠 + ⋅ ⋅ ⋅
𝛼 𝐴 −𝑀4
∫ |𝑠| 𝐾 (𝑠) 𝑑𝑠 ≥ 0, − 𝛼∫ 𝐾 (𝑡) 𝑑𝑡 < 𝜃, (28)
−𝑀2𝑖 2 −𝑀2𝑖
−𝑀2𝑛−2
−𝑀2𝑛−1
+ exp ( )∫ |𝑠| 𝐾 (𝑠) 𝑑𝑠
we have 𝐴 −𝑀2𝑛

0 −𝑀2𝑛
𝑠 −𝑀2𝑛+1
∫ |𝑠| exp ( ) 𝐾 (𝑠) 𝑑𝑠 + exp ( )∫ |𝑠| 𝐾 (𝑠) 𝑑𝑠
−∞ 𝐴 𝐴 −∞
0
𝑠 −𝑀3 0
=∫ |𝑠| exp ( ) 𝐾 (𝑠) 𝑑𝑠 ≥ exp ( ) ∫ |𝑠| 𝐾 (𝑠) 𝑑𝑠
−𝑀2 𝐴 𝐴 −𝑀4
−𝑀2
𝑠 −𝑀5 −𝑀4
+∫ |𝑠| exp ( ) 𝐾 (𝑠) 𝑑𝑠 + ⋅ ⋅ ⋅ + exp ( )∫ |𝑠| 𝐾 (𝑠) 𝑑𝑠 + ⋅ ⋅ ⋅
−𝑀4 𝐴 𝐴 −𝑀6
−𝑀2𝑛−2 −𝑀2𝑛
𝑠 −𝑀2𝑛−2
+∫ |𝑠| 𝐾 (𝑠) 𝑑𝑠 + ∫ |𝑠| exp ( ) 𝐾 (𝑠) 𝑑𝑠 −𝑀2𝑛−1
−𝑀2𝑛 −∞ 𝐴 + exp ( )∫ |𝑠| 𝐾 (𝑠) 𝑑𝑠
𝐴 −𝑀2𝑛
0
−𝑀1
> exp ( ) ∫ |𝑠| 𝐾 (𝑠) 𝑑𝑠 −𝑀2𝑛+1 −𝑀2𝑛
𝐴 −𝑀2 + exp ( )∫ |𝑠| 𝐾 (𝑠) 𝑑𝑠
𝐴 −∞
−𝑀2
−𝑀3
+ exp ( )∫ |𝑠| 𝐾 (𝑠) 𝑑𝑠 + ⋅ ⋅ ⋅ −𝑀2𝑛+1 0
𝐴 −𝑀4 ≥ exp ( ) ∫ |𝑠| 𝐾 (𝑠) 𝑑𝑠 > 0.
𝐴 −∞
−𝑀2𝑛−2
−𝑀2𝑛−1 (31)
+ exp ( )∫ |𝑠| 𝐾 (𝑠) 𝑑𝑠
𝐴 −𝑀2𝑛

−𝑀2𝑛−2
Thus, as 𝐾(𝑥) satisfies one of the (𝐿 𝑖 ), 𝑖 = 1, 2, 3,
−𝑀2𝑛+1
+ exp ( )∫ |𝑠| 𝐾 (𝑠) 𝑑𝑠 > 0.
𝐴 −∞ 𝛼 0 𝑠
(29) 𝑓󸀠 (𝐴) = 2
∫ |𝑠| exp ( ) 𝐾 (𝑠) 𝑑𝑠 > 0, (32)
𝐴 −∞ 𝐴
Case 3. As 𝐾(𝑥) satisfies (𝐿 3 ); that is, there exist 0 < 𝑀1 < which implies that 𝑓(𝐴) is continuous and monotonically
𝑀2 < ⋅ ⋅ ⋅ < 𝑀2𝑛 < ∞, such that 𝐾(𝑥) ≥ 0 for 𝑥 ∈ (−𝑀1 , 0) ∪ increasing function. Consequently, there exists a unique 𝐴 0 ∈
(−𝑀3 , −𝑀2 ) ∪ (−𝑀5 , −𝑀4 ) ∪ ⋅ ⋅ ⋅ ∪ (−𝑀2𝑛−1 , −𝑀2𝑛−2 ) ∪ (0, ∞) such that 𝑓(𝐴) = 𝛼/2 − 𝜃 if 𝐾(𝑥) satisfies one of (𝐿 𝑖 )
(−∞, −𝑀2𝑛 ), 𝐾(𝑥) ≤ 0 for 𝑥 ∈ (−𝑀2 , −𝑀1 ) ∪ (−𝑀4 , −𝑀3 ) ∪ 𝑖 = 1, 2, 3.
⋅ ⋅ ⋅ ∪ (−𝑀2𝑛 , −𝑀2𝑛−1 ), and for any 𝑖, 𝑖 = 1, 2, . . . , 𝑛, However, for the case that 𝐾(𝑥) satisfies (𝐿 4 ); that is, there
exists for 0 < 𝑀 < ∞, such that 𝐾(𝑥) ≥ 0 for 𝑥 ∈ (−∞, −𝑀)
0 0
𝛼 and 𝐾(𝑥) ≤ 0 for 𝑥 ∈ (−𝑀, 0), we have
∫ |𝑠| 𝐾 (𝑠) 𝑑𝑠 ≥ 0, − 𝛼∫ 𝐾 (𝑡) 𝑑𝑡 < 𝜃, (30)
−𝑀2𝑖 2 −𝑀2𝑖
𝛼 0 𝑠
𝑓󸀠 (𝐴) = ∫ |𝑠| exp ( ) 𝐾 (𝑠) 𝑑𝑠
we have 𝐴2 −∞ 𝐴
(33)
0
0
𝑠 𝛼 𝑠 𝑀
∫ |𝑠| exp ( ) 𝐾 (𝑠) 𝑑𝑠 > 2 𝑀 ∫ exp ( ) 𝐾 (𝑠) 𝑑𝑠 = 2 𝑓 (𝐴) .
−∞ 𝐴 𝐴 −∞ 𝐴 𝐴
0
𝑠 Consequently, if there exists 𝐴 0 ∈ (0, ∞) such that 𝑓(𝐴 0 ) =
=∫ |𝑠| exp ( ) 𝐾 (𝑠) 𝑑𝑠 𝛼/2 − 𝜃 > 0, then 𝑓󸀠 (𝐴 0 ) > 0. Thus function 𝑦 = 𝑓(𝐴) must
−𝑀2 𝐴
be monotonically increasing in the neighborhood of the root
−𝑀2
𝑠 of equation 𝑓(𝐴) = 𝛼/2−𝜃, which implies that 𝑓(𝐴) = 𝛼/2−𝜃
+∫ |𝑠| exp ( ) 𝐾 (𝑠) 𝑑𝑠 + ⋅ ⋅ ⋅
−𝑀4 𝐴 has at most one root. Since we have proved the existence of the
root of equation 𝑓(𝐴) = 𝛼/2 − 𝜃, we know that there exists a
−𝑀2𝑛−2
unique 𝐴 0 ∈ (0, ∞) such that 𝑓(𝐴) = 𝛼/2−𝜃 if 𝐾(𝑥) satisfies
+∫ |𝑠| 𝐾 (𝑠) 𝑑𝑠
−𝑀2𝑛 (𝐿 4 ).
6 Abstract and Applied Analysis

Theorem 2 (see [10]). Suppose that the positive parameters 𝛼 Case 1. As 𝐾(𝑥) satisfies (𝐿 1 ), it is easy to see that 𝑈󸀠 (𝑧) ≥ 0
and 𝜃 satisfy the condition 0 < 2𝜃 < 𝛼. In addition to the for all 𝑧 ∈ (−∞, 0). So 𝑈(𝑧) is a continuous and monotoni-
basic assumptions for kernel functions, if 𝐾(𝑥) satisfies one of cally increasing function on (−∞, 0). Consequently, 𝑈(𝑧) < 𝜃
the assumptions 𝐿 𝑖 , (𝑖 = 1, 2, 3, 4), then there exists a unique on (−∞, 0) since 𝑈(0) = 𝜃 and 𝑈󸀠 (0) = (1/𝜇)(𝛼/2 − 𝜃) > 0.
0
𝜇∗ ∈ (0, 𝑐) such that 𝛼 ∫−∞ exp(((𝑐−𝜇)/𝑐𝜇)𝑠)𝐾(𝑠)𝑑𝑠 = 𝛼/2−𝜃.
Case 2. As 𝐾(𝑥) satisfies (𝐿 2 ), it is obvious that 𝜑(𝑧) ≤ 0 for
Proof. Let 𝐴 = 𝑐𝜇/(𝑐 − 𝜇), then, for any 𝐴 ∈ (0, ∞), there is 𝑧 ∈ (−∞, −𝑀2𝑛+1 ) and 𝜑(𝑧) is increasing on (−𝑀2𝑖+1 , −𝑀2𝑖 )
a unique 𝜇 = 𝑐/(𝐴𝑐 + 1) for 𝑐 > 0. Clearly, 𝜇 ∈ (0, 𝑐) for any for any 𝑖, 𝑖 = 0, 1, 2, . . . , 𝑛, and decreasing on (−𝑀2𝑖 , −𝑀2𝑖−1 )
𝐴 ∈ (0, ∞) and 𝑐 > 0. From Lemma 1, we know that there for any 𝑖, 𝑖 = 1, 2, . . . , 𝑛, because of the properties of function
exists a unique 𝐴 0 ∈ (0, ∞) such that 𝑓(𝐴 0 ) = 𝛼/2 − 𝜃; that 𝐾(𝑥). Consequently, 𝜑(𝑧) may only change its sign from
is, there exists a unique 𝜇∗ = 𝑐/(𝐴 0 𝑐 + 1) ∈ (0, 𝑐) such that negative to positive on (−𝑀2𝑖+1 , −𝑀2𝑖 ) and from positive to
0 negative on (−𝑀2𝑖 , −𝑀2𝑖−1 ) for some 𝑖, 𝑖 = 1, 2, . . . , 𝑛. Recall
𝑓(𝐴 0 ) = 𝑓(𝑐𝜇∗ /(𝑐 − 𝜇∗ )) = 𝛼/2 − 𝜃; that is, 𝛼 ∫−∞ exp(((𝑐 − that 𝑈󸀠 (((𝑐 − 𝜇)/𝑐)𝑧) has the same sign as 𝜑(𝑧), so the local
𝜇∗ )/𝑐𝜇∗ )𝑠)𝐾(𝑠)𝑑𝑠 = 𝛼/2 − 𝜃. maximal points of function 𝑈(((𝑐 − 𝜇)/𝑐)𝑧) should locate on
(−𝑀2𝑖 , −𝑀2𝑖−1 ) and the local minimal points should locate
From Theorem 2, we see that there exists a unique 𝜇∗
on (−𝑀2𝑖+1 , −𝑀2𝑖 ) if any. We will know that 𝑈(𝑧) ≤ 𝜃 on
such that function (22) satisfies 𝑈(0) = 𝜃 if the kernel
(−∞, 0) if we prove that the local maximum of function 𝑈(𝑧)
function 𝐾(𝑥) satisfies one of the assumptions 𝐿 𝑖 , 𝑖 = 1, 2, 3, 4.
is no more than 𝜃. Suppose that there exists a point 𝑧𝑖 ∈
However, it does not follow that function (22) is a traveling
(−𝑀2𝑖 , −𝑀2𝑖−1 ) such that 𝜑(𝑧𝑖 ) = 0, and so 𝑈󸀠 (((𝑐 − 𝜇)/𝑐)𝑧𝑖 ) =
wave front solution to IDE (5). We now prove that function
0. If 𝑈(((𝑐 − 𝜇)/𝑐)𝑧𝑖 ) attains a local maximum, then
(22) satisfies 𝑈(𝑧) < 𝜃 on (−∞, 0) and 𝑈(𝑧) > 𝜃 on (0, +∞)
if 𝐾(𝑥) satisfies the assumptions of Theorem 2 and one of the 𝑐−𝜇 𝑧𝑖
𝛼 0
assumptions 𝑅𝑖 (𝑖 = 1, 2, . . . , 6), from which the conclusion 𝑈( 𝑧𝑖 ) = 𝛼 ∫ 𝐾 (𝑡) 𝑑𝑡 = − 𝛼 ∫ 𝐾 (𝑡) 𝑑𝑡
𝑐 −∞ 2 𝑧𝑖
that function (22) is a traveling wave front solution of the IDE (37)
(5) will be derived obviously. 0
𝛼
≤ − 𝛼∫ 𝐾 (𝑡) 𝑑𝑡 < 𝜃.
2 −𝑀2𝑖
3.2. Existence and Uniqueness of the Wave Front. In this sub-
section, we first present two lemmas to prove that function Consequently, 𝑈(𝑧) < 𝜃 on (−∞, 0) if 𝐾(𝑥) satisfies (𝐿 2 ).
(22) satisfies 𝑈(𝑧) < 𝜃 on (−∞, 0) and 𝑈(𝑧) > 𝜃 on (0, +∞).
Then we obtain the existence and uniqueness of the wave Case 3. As 𝐾(𝑥) satisfies (𝐿 3 ), by the similar analysis as in
front. Case 2 above, we get 𝑈(𝑧) < 𝜃 on (−∞, 0).
It follows from (21) and function (22) that
Case 4. As 𝐾(𝑥) satisfies (𝐿 4 ), it is easy to see that 𝜑(𝑧) ≥ 0
𝛼 −𝑧 for 𝑧 ∈ (−∞, −𝑀), and 𝜑(𝑧) ≥ 𝜑(0) > 0 for 𝑧 ∈ [−𝑀, 0].
𝑈󸀠 (𝑧) = exp ( ) Consequently, 𝑈󸀠 (𝑧) ≥ 0 on (−∞, 0). Thus 𝑈(𝑧) < 𝜃 on
𝜇 𝜇
(−∞, 0) follows from 𝑈󸀠 (0) > 0 and 𝑈(0) = 𝜃.
𝑧
𝑠 𝑐 𝑐 From the analysis above, we conclude that 𝑈(𝑧) < 𝜃 on
×∫ exp ( ) 𝐾 ( 𝑠) 𝑑𝑠. (−∞, 0) if 𝐾(𝑥) satisfies one of (𝐿 𝑖 ) (𝑖 = 1, . . . , 4).
−∞ 𝜇 𝑐 + sgn (𝑠) 𝜇 𝑐 + sgn (𝑠) 𝜇
(34)
Lemma 4. Function (22) satisfies 𝑈(𝑧) > 𝜃 on (0, +∞) if 𝐾(𝑥)
Lemma 3. Function (22) satisfies 𝑈(𝑧) < 𝜃 on (−∞, 0) if 𝐾(𝑥) satisfies one of the assumptions (R𝑖 ) (𝑖 = 1, 2, . . . , 6).
satisfies one of the assumptions (𝐿 𝑖 ) (𝑖 = 1, 2, 3, 4).
Proof. For 𝑧 ∈ (0, +∞), it follows from (34) that
Proof. From (34), we have 𝛼 −𝑧 0
𝑠 𝑐 𝑐
𝑈󸀠 (𝑧) = exp ( ) [∫ exp ( ) 𝐾 ( 𝑠) 𝑑𝑠
𝛼 −𝑧 𝑧
𝑠 𝑐 𝑐 𝜇 𝜇 −∞ 𝜇 𝑐−𝜇 𝑐−𝜇
𝑈󸀠 (𝑧) = exp ( ) ∫ exp ( ) 𝐾 ( 𝑠) 𝑑𝑠
𝜇 𝜇 −∞ 𝜇 𝑐−𝜇 𝑐−𝜇 𝑧
𝑠 𝑐 𝑐
+ ∫ exp ( ) 𝐾 ( 𝑠) 𝑑𝑠]
𝛼 −𝑧 (𝑐/(𝑐−𝜇))𝑧 𝑐−𝜇 0 𝜇 𝑐+𝜇 𝑐+𝜇
= exp ( ) ∫ exp ( 𝑠) 𝐾 (𝑠) 𝑑𝑠
𝜇 𝜇 −∞ 𝑐𝜇 𝛼 −𝑧
(35) = exp ( )
𝜇 𝜇
for 𝑧 ≤ 0. Denote (𝑐/(𝑐+𝜇))𝑧 𝑐+𝜇
× [𝜑 (0) + ∫ exp ( 𝑠) 𝐾 (𝑠) 𝑑𝑠] .
𝑧 𝑐−𝜇 0 𝑐𝜇
𝜑 (𝑧) = ∫ exp ( 𝑠) 𝐾 (𝑠) 𝑑𝑠. (36) (38)
−∞ 𝑐𝜇
Denote
Obviously, 𝑈󸀠 (𝑧) has the same sign as 𝜑((𝑐/(𝑐 − 𝜇))𝑧) for
(𝑐/(𝑐+𝜇))𝑧
𝑧 ∈ (−∞, 0). Consequently, taking into account that 𝑈󸀠 (0) = 𝑐+𝜇
𝜓 (𝑧) = ∫ exp ( 𝑠) 𝐾 (𝑠) 𝑑𝑠. (39)
(1/𝜇)(𝛼/2 − 𝜃) > 0, we get 𝜑(0) > 0. 0 𝑐𝜇
Abstract and Applied Analysis 7

Then should locate on (((𝑐 + 𝜇)/𝑐)𝑁2𝑖 , ((𝑐 + 𝜇)/𝑐)𝑁2𝑖+1 ) (𝑖 =

0, 1, 2, . . . , 𝑛) and the local maximal points should locate on
𝛼 −𝑧
𝑈󸀠 (𝑧) = exp ( ) [𝜑 (0) + 𝜓 (𝑧)] . (40) (((𝑐 + 𝜇)/𝑐)𝑁2𝑖−1 , ((𝑐 + 𝜇)/𝑐)𝑁2𝑖 ) for some 𝑖, 𝑖 = 1, 2, . . . , 𝑛,
𝜇 𝜇 or (((𝑐 + 𝜇)/𝑐)𝑁2𝑛 , ∞) if any. If we can prove that the local
minimum is greater than 𝜃, then we have 𝑈(𝑧) ≥ 𝜃 on (0, ∞)
Recall that 𝜑(0) > 0. because lim𝑧 → +∞ 𝑈(𝑧) = 𝛼 > 𝜃. If there exists a point 𝑧𝑖 ∈
(𝑁2𝑖 , 𝑁2𝑖+1 ) (𝑖 = 0, 1, 2, . . . , 𝑛) such that 𝑈󸀠 (((𝑐 + 𝜇)/𝑐)𝑧𝑖 ) = 0,
Case 1. As 𝐾(𝑥) satisfies (𝑅1 ), it is easy to see that 𝜓(𝑧) ≥ 0
𝑈(((𝑐 + 𝜇)/𝑐)𝑧𝑖 ) attains a local minimum and
and thus 𝑈󸀠 (𝑧) > 0 on (0, +∞). Consequently, 𝑈(𝑧) > 𝜃 on
(0, +∞) in consideration of 𝑈(0) = 𝜃. 𝑐+𝜇 𝑧𝑖 𝑁2𝑖
𝑈( 𝑧𝑖 ) = 𝛼 ∫ 𝐾 (𝑡) 𝑑𝑡 ≥ 𝛼 ∫ 𝐾 (𝑡) 𝑑𝑡 > 𝜃. (42)
𝑐 −∞ −∞
Case 2. As 𝐾(𝑥) satisfies (𝑅2 ), it is easy to see that 𝑦 = 𝜓(𝑧) is
decreasing on (((𝑐 + 𝜇)/𝑐)𝑁2𝑖 , ((𝑐 + 𝜇)/𝑐)𝑁2𝑖+1 ) but increasing Consequently, 𝑈(𝑧) > 𝜃 on (0, +∞) if 𝐾(𝑥) satisfies (𝑅5 ).
on (((𝑐 + 𝜇)/𝑐)𝑁2𝑖−1 , ((𝑐 + 𝜇)/𝑐)𝑁2𝑖 ) for any 𝑖, 𝑖 = 1, 2, . . . , 𝑛,
and (((𝑐 + 𝜇)/𝑐)𝑁2𝑛+1 , ∞). So 𝑈󸀠 (𝑧) may change its sign from Case 6. As 𝐾(𝑥) satisfies (𝑅6 ), 𝜓(𝑧) is increasing on (((𝑐 +
negative to positive on (((𝑐 + 𝜇)/𝑐)𝑁2𝑖−1 , ((𝑐 + 𝜇)/𝑐)𝑁2𝑖 ) for 𝜇)/𝑐)𝑁2𝑖 , ((𝑐 + 𝜇)/𝑐)𝑁2𝑖+1 ) for any 𝑖, 𝑖 = 0, 1, 2, . . . , 𝑛 − 1, and
some 𝑖, 𝑖 = 1, 2, . . . , 𝑛 or (((𝑐 + 𝜇)/𝑐)𝑁2𝑛+1 , ∞) and from (((𝑐 + 𝜇)/𝑐)𝑁2𝑛 , ∞) but decreasing on (((𝑐 + 𝜇)/𝑐)𝑁2𝑖−1 , ((𝑐 +
positive to negative on (((𝑐 + 𝜇)/𝑐)𝑁2𝑖 , ((𝑐 + 𝜇)/𝑐)𝑁2𝑖+1 ) for 𝜇)/𝑐)𝑁2𝑖 ) for any 𝑖, 𝑖 = 1, 2, . . . , 𝑛. So 𝑈󸀠 (𝑧) may change
some 𝑖, 𝑖 = 0, 1, 2, . . . , 𝑛. Consequently, the local maximal its sign from negative to positive on (((𝑐 + 𝜇)/𝑐)𝑁2𝑖 , ((𝑐 +
points of function 𝑈(𝑧) should locate on (((𝑐 + 𝜇)/𝑐)𝑁2𝑖 , 𝜇)/𝑐)𝑁2𝑖+1 ) for some 𝑖, 𝑖 = 1, 2, . . . , 𝑛−1, or (((𝑐+𝜇)/𝑐)𝑁2𝑛 , ∞)
((𝑐 + 𝜇)/𝑐)𝑁2𝑖+1 ) for some, 𝑖, 𝑖 = 0, 1, 2, . . . , 𝑛, and the and from positive to negative on (((𝑐 + 𝜇)/𝑐)𝑁2𝑖−1 , ((𝑐 +
local minimal points should locate on (((𝑐 + 𝜇)/𝑐)𝑁2𝑖−1 , 𝜇)/𝑐)𝑁2𝑖 ) for some 𝑖, 𝑖 = 1, 2, . . . , 𝑛. Consequently, the
((𝑐 + 𝜇)/𝑐)𝑁2𝑖 ) for some 𝑖, 𝑖 = 1, 2, . . . , 𝑛, if any. Recall that local maximal points of function 𝑈(𝑧) should locate on
lim𝑧 → +∞ 𝑈(𝑧) = 𝛼 > 𝜃, so if we can prove that the local (((𝑐 + 𝜇)/𝑐)𝑁2𝑖−1 , ((𝑐 + 𝜇)/𝑐)𝑁2𝑖 ) for some 𝑖, 𝑖 = 1, 2, . . . , 𝑛,
minimum is more than 𝜃, then we have 𝑈(𝑧) > 𝜃 on (0, +∞). and the local minimal points should locate on (((𝑐 +
If there exists a point 𝑧𝑖 ∈ (𝑁2𝑖+1 , 𝑁2𝑖+2 ) (𝑖 = 1, 2, . . . , 𝑛) such 𝜇)/𝑐)𝑁2𝑖 , ((𝑐 + 𝜇)/𝑐)𝑁2𝑖+1 ) for some 𝑖, 𝑖 = 0, 1, 2, . . . , 𝑛, or
that 𝑈󸀠 (((𝑐 + 𝜇)/𝑐)𝑧𝑖 ) = 0, which means that 𝑈(((𝑐 + 𝜇)/𝑐)𝑧𝑖 ) (((𝑐 + 𝜇)/𝑐)𝑁2𝑛 , +∞) if any. If we can prove that the local
attains a local minimum and minimum is more than 𝜃, then we have 𝑈(𝑧) > 𝜃 on (0, +∞).
If there exists a point 𝑧𝑖 ∈ (𝑁2𝑖 , 𝑁2𝑖+1 ) for some 𝑖, 𝑖 =
𝑧𝑖
𝑐+𝜇 1, 2, . . . , 𝑛 − 1, or 𝑧𝑛 ∈ (𝑁2𝑛 , +∞) such that 𝑈󸀠 (((𝑐 − 𝜇)/𝑐)𝑧𝑖 ) =
𝑈( 𝑧𝑖 ) = 𝛼 ∫ 𝐾 (𝑡) 𝑑𝑡
𝑐 −∞ 0, 𝑈(((𝑐 − 𝜇)/𝑐)𝑧𝑖 ) attains a local minimum and
𝑁2𝑖−1 𝑁2𝑖−1 𝑐−𝜇 𝑧𝑖 𝑁2𝑖
𝛼 (41) 𝑈( 𝑧𝑖 ) = 𝛼 ∫ 𝐾 (𝑡) 𝑑𝑡 ≥ 𝛼 ∫ 𝐾 (𝑡) 𝑑𝑡 > 𝜃. (43)
≥ 𝛼∫ 𝐾 (𝑡) 𝑑𝑡 = + 𝛼∫ 𝐾 (𝑡) 𝑑𝑡
−∞ 2 0 𝑐 −∞ −∞

> 𝜃. Consequently, 𝑈(𝑧) > 𝜃 on (0, +∞) if 𝐾(𝑥) satisfies (𝑅6 ).

In conclusion, 𝑈(𝑧) > 𝜃 on (0, +∞) if 𝐾(𝑥) satisfies one
Consequently, 𝑈(𝑧) > 𝜃 on (0, +∞) if 𝐾(𝑥) satisfies (𝑅2 ). of (𝑅𝑖 ) (𝑖 = 1, 2, . . . , 6).

Case 3. As 𝐾(𝑥) satisfies (𝑅3 ), similar to the proof in Case 2 It follows from the above two lemmas that the function
we can prove that 𝑈(𝑧) > 𝜃 on (0, +∞). (22) satisfies 𝑈(𝑧) < 𝜃 on (−∞, 0) and 𝑈(𝑧) > 𝜃 on (0, +∞)
if 𝐾(𝑥) satisfies one of the assumptions 𝐿 𝑖 (𝑖 = 1, 2, 3, 4) on
Case 4. As 𝐾(𝑥) satisfies (𝑅4 ), it is easy to see that 𝜓(𝑧) ≥ 0 (−∞, 0) and one of 𝑅𝑗 (𝑗 = 1, 2, . . . , 6) on (0, +∞). We sum
for 𝑧 ∈ (0, 𝑁] and the function 𝑦 = 𝜓(𝑧) is monotonically up all these results in the following theorem.
decreasing on (𝑁, +∞). So 𝑈󸀠 (𝑧) at most changes its sign
once from positive to negative, that is to say, that 𝑈(𝑧) keeps Theorem 5 (see [10]). Suppose that the positive parameters 𝛼
increasing on (0, ∞) or there exists 𝑧0 ∈ (𝑁, +∞) such and 𝜃 satisfy the condition 0 < 2𝜃 < 𝛼 and 𝐾(𝑥) satisfies
that 𝑈(𝑧) is increasing on (0, 𝑧0 ) but decreasing on (𝑧0 , ∞). the assumption (𝑆1 ), one of 𝐿 𝑖 (𝑖 = 1, 2, 3, 4) on (−∞, 0) and
Clearly, 𝑈(𝑧) > 𝜃 on (0, ∞) if 𝑈(𝑧) keeps increasing on one of 𝑅𝑗 (𝑗 = 1, 2, . . . , 6) on (0, +∞). Then (5) has a unique
(0, ∞). 𝑈(𝑧) > 𝜃 on (0, ∞) if 𝑈(𝑧) is increasing on (0, 𝑧0 ) traveling wave front solution 𝑈(𝑧) = 𝑈(𝑥 + 𝜇𝑡) satisfying the
and decreasing on (𝑧0 , ∞) because lim𝑧 → +∞ 𝑈(𝑧) = 𝛼 > 𝜃. phase conditions:
Consequently, 𝑈(𝑧) > 𝜃 on (0, +∞) if 𝐾(𝑥) satisfies (𝑅4 ). 𝑈 (0) = 𝜃, 𝑈 (𝑧) < 𝜃 𝑜𝑛 (−∞, 0) ,
(44)
Case 5. As 𝐾(𝑥) satisfies (𝑅5 ), 𝜓(𝑧) is increasing on 𝑈 (𝑧) > 𝜃 𝑜𝑛 (0, +∞) .
(((𝑐 + 𝜇)/𝑐)𝑁2𝑖 , ((𝑐 + 𝜇)/𝑐)𝑁2𝑖+1 ) but decreasing on (((𝑐 +
𝜇)/𝑐)𝑁2𝑖+1 , ((𝑐 + 𝜇)/𝑐)𝑁2𝑖+2 ) for any 𝑖, 𝑖 = 0, 1, 2, . . . , 𝑛. So The unique traveling wave front could be expressed as
𝑈󸀠 (𝑧) may change its sign from negative to positive on (((𝑐 + 𝑐𝑧/(𝑐+sgn(𝑧)𝜇)
𝜇)/𝑐)𝑁2𝑖 , ((𝑐 + 𝜇)/𝑐)𝑁2𝑖+1 ) for some 𝑖, 𝑖 = 1, 2, . . . , 𝑛, and from 𝑈 (𝑧) = 𝛼 ∫ 𝐾 (𝑠) 𝑑𝑠
−∞
positive to negative on (((𝑐 + 𝜇)/𝑐)𝑁2𝑖+1 , ((𝑐 + 𝜇)/𝑐)𝑁2𝑖+2 ) (45)
𝑧
for some 𝑖, 𝑖 = 0, 1, 2, . . . , 𝑛, or (((𝑐 + 𝜇)/𝑐)𝑁2𝑛+1 , ∞). 𝑠−𝑧 𝑐
− 𝛼 ∫ exp ( ) 𝐾 (𝑠) 𝑑𝑠.
Consequently, the local minimal points of function 𝑈(𝑧) −∞ 𝜇 𝑐 + sgn (𝑠) 𝜇
8 Abstract and Applied Analysis

The wave speed 𝜇 (0 < 𝜇 < 𝑐) is determined by the speed conditions of kernel function 𝑘(𝑥) on (−∞, 0); however, the
equation phase condition 𝑈(𝑧) > 𝜃 for 𝑧 ∈ (0, +∞) depended on the
conditions of kernel function 𝑘(𝑥) on (0, +∞).
0 𝑐−𝜇 𝛼 As we mentioned in the first section, the results of
𝛼∫ exp ( 𝑠) 𝐾 (𝑠) 𝑑𝑠 = − 𝜃. (46)
−∞ 𝑐𝜇 2 this work could be employed to investigate some more
complicated model equation. For instance, the solution of
The wave front solution 𝑈(𝑧) also satisfies the boundary model equation (2), for the case when 𝐼(𝑥, 𝑡) is irrelevant
conditions: to variables, can be obtained by a simple translation of
the solution of (5) and was well applied to investigate the
lim 𝑈 (𝑧) = 0, lim 𝑈 (𝑧) = 𝛼,
𝑧 → −∞ 𝑧 → +∞ model equation (4) [10]. The results we obtained might be
(47) applied to approximate the solution of the equation with
lim 𝑈󸀠 (𝑧) = 0. oscillating infinity times kernels. Particularly, we believe that
𝑧 → ±∞
the idea and method we used in this paper can be applied
From this theorem, we see that, not only for the classical to investigate other nonlinear equations, which is our main
nonnegative kernel function, Mexican hat kernel function, research interest at present.
and upside down Mexican hat kernel function but also
for some types of kernel function oscillating 𝑛-times, the
IDE (5) has a unique traveling front wave solution. It is Conflict of Interests
well known that, in this biophysically motivated nonlinear The authors declare that there is no conflict of interests
nonlocal firing rate model equation, pure excitation, lateral regarding the publication of this paper.
inhibition, and lateral excitation are modeled by nonnegative
kernel function, Mexican hat kernel function and upside
down Mexican hat kernel function, respectively. The kernel Acknowledgments
function with 𝑛-times oscillations may hopefully be applied
to more complicated and more accurate reaction cases. This work is supported by the Nature Science Foundation
of China (no. 11101371). This work was motivated during
the period that Lijun Zhang was visiting the Department
4. Conclusion and Some Discussions of Mathematics of Lehigh University. Lijun Zhang would
like to express her appreciation for the kind invitation from
In this paper, we have investigated the existence and unique- the Department of Mathematics of Lehigh University and
ness of the wave front solution of the integral-differential all of us would like to thank Professor Jibin Li for his kind
model equation (5) arising from neuronal networks. At first, introduction and the support from State Scholarship Fund of
we reduced the nonlinear-integral differential equation (5) China that made the collaboration come true.
into a simpler solvable linear differential equation (21) by
using the special property of Heaviside gain function and the
hypothesis of the wave front solution that 𝑈(0) = 𝜃, 𝑈(𝑧) < 𝜃 References
for 𝑧 ∈ (−∞, 0), and 𝑈(𝑧) > 𝜃 for 𝑧 ∈ (0, +∞). Next,
[1] S. I. Amari, “Dynamics of pattern formation in lateral inhibition
we checked that the solution (22) which we got easily from
type neural fields,” Biological Cybernetics, vol. 27, no. 2, pp. 77–
(21) was the unique wave front solution of (5) satisfying the 87, 1977.
hypothesis 𝑈(0) = 𝜃, 𝑈(𝑧) < 𝜃 for 𝑧 ∈ (−∞, 0), and 𝑈(𝑧) > 𝜃
[2] F. M. Atay and A. Hutt, “Stability and bifurcations in neural
for 𝑧 ∈ (0, +∞). The uniqueness was obtained by proving the
fields with finite propagation speed and general connectivity,”
uniqueness of the solution of the speed equation 𝑈(0) = 𝜃, SIAM Journal on Applied Mathematics, vol. 65, no. 2, pp. 644–
that is, the uniqueness of the wave speed. 666, 2005.
The difficulties and key points of this work were how to [3] S. Coombes, G. J. Lord, and M. R. Owen, “Waves and bumps
test that solution (22) satisfies the phase conditions: 𝑈(0) = in neuronal networks with axo-dendritic synaptic interactions,”
𝜃, 𝑈(𝑧) < 𝜃 for 𝑧 ∈ (−∞, 0) and 𝑈(𝑧) > 𝜃 for 𝑧 ∈ Physica D: Nonlinear Phenomena, vol. 178, no. 3-4, pp. 219–241,
(0, +∞) if the kernel function 𝐾(𝑥) is not of class (A) or 2003.
(B) because the function (22) may not be an increasing [4] G. B. Ermentrout, “Neural networks as spatio-temporal pattern-
function any more. We achieved it by proving the maximum forming systems,” Reports on Progress in Physics, vol. 61, no. 4,
values of (22) on (−∞, 0) if any were less than 𝜃 and the pp. 353–430, 1998.
minimum values of (22) on (0, +∞) if any were greater than [5] D. J. Pinto and G. B. Ermentrout, “Spatially structured activity
𝜃, using the idea motivated by Lv and Wang [12]. It is worth in synaptically coupled neuronal networks, I. Traveling fronts
pointing out that we simplified the problem and generalized and pulses,” SIAM Journal on Applied Mathematics, vol. 62, no.
the kernel function 𝐾(𝑥) to a very general class of functions, 1, pp. 206–225, 2001.
including not only the well-known three typical classes of [6] D. J. Pinto and G. B. Ermentrout, “Spatially structured activity
kernels and five classes of oscillatory kernels proposed in in synaptically coupled neuronal networks, II. Lateral inhibition
[12], but also oscillating 𝑛-times kernels. We separated the and standing pulses,” SIAM Journal on Applied Mathematics,
whole problem into some simpler ones by noticing that the vol. 62, no. 1, pp. 226–243, 2001.
existence and uniqueness of the wave speed and the phase [7] A. Hutt, “Effects of nonlocal feedback on traveling fronts in
condition 𝑈(𝑧) < 𝜃 for 𝑧 ∈ (−∞, 0) only depended on the neural fields subject to transmission delay,” Physical Review E:
Abstract and Applied Analysis 9

Statistical, Nonlinear, and Soft Matter Physics, vol. 70, no. 5,

Article ID 052902, 4 pages, 2004.
[8] F. M. Atay and A. Hutt, “Neural fields with distributed trans-
mission speeds and long-range feedback delays,” SIAM Journal
on Applied Dynamical Systems, vol. 5, no. 4, pp. 670–698, 2006.
[9] A. Hutt and A. Longtin, “Effects of the anesthetic agent propofol
on neural populations,” Cognitive Neurodynamics, vol. 4, no. 1,
pp. 37–59, 2010.
[10] L. J. Zhang, “Existence and uniqueness of wave fronts in
neuronal network with nonlocal post-synaptic axonal connec-
tions and delayed nonlocal feedback connections,” Advances in
Difference Equations, vol. 2013, article 243, pp. 1–15, 2013.
[11] L. H. Zhang, “How do synaptic coupling and spatial temporal
delay influence traveling waves in nonlinear nonlocal neuronal
networks?” SIAM Journal on Applied Dynamical Systems, vol. 6,
no. 3, pp. 597–644, 2007.
[12] G. Lv and M. Wang, “Traveling waves of some integral-
differential equations arising from neuronal networks with
oscillatory kernels,” Journal of Mathematical Analysis and Appli-
cations, vol. 370, no. 1, pp. 82–100, 2010.
[13] F. Magpantay and X. Zou, “Wave fronts in neuronal fields
with nonlocal post-synaptic axonal connections and delayed
nonlocal feedback connections,” Mathematical Biosciences and
Engineering, vol. 7, no. 2, pp. 421–442, 2010.
[14] A. J. Elvin, C. R. Laing, R. I. McLachlan, and M. G. Roberts,
“Exploiting the Hamiltonian structure of a neural field model,”
Physica D: Nonlinear Phenomena, vol. 239, no. 9, pp. 537–546,
2010.
[15] A. J. Elvin, C. R. Laing, and M. G. Roberts, “Transient Turing
patterns in a neural field model,” Physical Review E: Statistical,
Nonlinear, and Soft Matter Physics, vol. 79, no. 1, Article ID
011911, 6 pages, 2009.
[16] C. Laing, W. Troy, B. Gutkin, and B. Ermentrout, “Multiple
bumps in a neuronal model of working memory,” SIAM Journal
on Applied Mathematics, vol. 63, no. 1, pp. 62–97, 2002.
[17] C. Laing and W. Troy, “Two-bump solutions of Amari-type
models of neuronal pattern formation,” Physica D: Nonlinear
Phenomena, vol. 178, no. 3-4, pp. 190–218, 2003.
[18] C. Ou and J. Wu, “Persistence of wavefronts in delayed nonlocal
reaction-diffusion equations,” Journal of Differential Equations,
vol. 235, no. 1, pp. 219–261, 2007.
[19] L. H. Zhang, “Dynamics of neuronal waves,” Mathematische
Zeitschrift, vol. 255, no. 2, pp. 283–321, 2007.
[20] L. H. Zhang, P. Wu, and M. A. Stoner, “Influence of sodium
currents on speeds of traveling wave fronts in synaptically
coupled neuronal networks,” Physica D: Nonlinear Phenomena,
vol. 239, no. 1-2, pp. 9–32, 2010.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 317543, 5 pages
http://dx.doi.org/10.1155/2014/317543

Research Article
Travelling Wave Solutions of Nonlinear Dynamical Equations in
a Double-Chain Model of DNA

Zheng-yong Ouyang1 and Shan Zheng2

1
Department of Mathematics, Foshan University, Foshan Guangdong 528000, China
2
Guangzhou Maritime College, Guangzhou Guangdong 510725, China

Correspondence should be addressed to Zheng-yong Ouyang; zyouyang math@163.com

Received 6 January 2014; Revised 24 February 2014; Accepted 10 March 2014; Published 3 April 2014

Academic Editor: Yonghui Xia

Copyright © 2014 Z.-y. Ouyang and S. Zheng. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.

We consider the nonlinear dynamics in a double-chain model of DNA which consists of two long elastic homogeneous strands
connected with each other by an elastic membrane. By using the method of dynamical systems, the bounded traveling wave solutions
such as bell-shaped solitary waves and periodic waves for the coupled nonlinear dynamical equations of DNA model are obtained
and simulated numerically. For the same wave speed, bell-shaped solitary waves of different heights are found to coexist.

1. Introduction two movements in DNA [13]. Another new double-chain

model of DNA that consists of two long elastic homogeneous
In 1953, Waston and Crick discovered the structure of strands was given to describe transverse movements along
deoxyribonucleic acid (DNA) double helix [1], which opened the hydrogen bond and longitudinal movements along the
the area of molecular biology. The significance of the double backbone direction [14, 15].
helix model not only means the proven structure of DNA, In [14, 15], the nonlinear dynamical equations were
but, more importantly, also helps to reveal the replication derived and nonlinear dynamics of DNA were studied. The
mechanism of DNA. To further study the characteristics exact solutions of the general nonlinear dynamic system
of DNA, nonlinear science has been used to deal with in the double-chain model of DNA were obtained and
DNA system, because its properties can be investigated by discussed numerically and analytically under some special
nonlinear models that combine the methods of physics with approximate conditions. The nonlinear dynamical equations
biological tools [2]. The soliton theory especially which for the double-chain model were given as follows:
pertains to nonlinear science has been widely applied in the
study of DNA [3–6].
To study DNA structure from the view of nonlinear 𝑢𝑡𝑡 − 𝑐12 𝑢𝑥𝑥 = 𝜆 1 𝑢 + 𝛾1 𝑢V + 𝜇1 𝑢3 + 𝛽1 𝑢V3 ,
science, it is necessary to look for the right nonlinear (1)
mathematical models. A number of researchers have tried to V𝑡𝑡 − 𝑐22 V𝑥𝑥 = 𝜆 2 V + 𝛾2 𝑢2 + 𝜇2 𝑢2 V + 𝛽2 V3 + 𝑐0 ,
establish mathematical models to describe DNA system. At
the beginning, it is difficult to use a specific mathematical where 𝑢 denotes the difference of the longitudinal displace-
model to simulate DNA system due to its complex structure ments of the bottom and top strands, that is, the displace-
and the presence of various movements [7]. So some simpli- ments of the bases from their equilibrium positions along
fied models were used to study only some internal movement the direction of the phosphodiester bridge that connects the
in DNA [8–12]. Further, two movements which made the two bases of the same strands; V represents the difference of
main contribution to the DNA denaturation process were the transverse displacements of the bottom and top strands,
taken into account together in one model. For example, a that is, the displacements of the bases from their equilibrium
two-dimensional discrete model was used to describe the positions along the direction of the hydrogen bond that
2 Abstract and Applied Analysis

connects the two bases of the base pair. The constants 𝑐1 , 𝑐2 , In fact, system (6) is a planar dynamical system whose
𝜆 1 , 𝜆 2 , 𝛾1 , 𝛾2 , 𝜇1 , 𝜇2 , 𝛽1 , 𝛽2 , and 𝑐0 in (1) are written as phase orbits defined by the vector field of (6) determine
traveling wave solutions, so we first need to investigate some
𝑌 𝐹 −2𝜇 different phase portraits of system (6) in different parameter
𝑐1 = ±√ , 𝑐2 = ±√ , 𝜆1 = (ℎ − 𝑙0 ) ,
𝜌 𝜌 𝜌𝜎ℎ regions. Then, by using the theory of dynamical system, we
will identify some different wave profiles and compute exact
−2𝜇 2√2𝜇𝑙0 −2𝜇𝑙0 representations for solitary wave solutions and periodic wave
𝜆2 = , 𝛾1 = 2𝛾2 = , 𝜇1 = 𝜇2 = ,
𝜌𝜎 𝜌𝜎ℎ2 𝜌𝜎ℎ3 solutions.
This paper is organized as follows. In Section 2, we draw
4𝜇𝑙0 √2𝜇 (ℎ − 𝑙0 ) bifurcation phase portraits of system (6), where explicit
𝛽1 = 𝛽2 = , 𝑐0 = , parametric conditions will be derived. In Section 3, we give
𝜌𝜎ℎ3 𝜌𝜎ℎ
(2) exact representations of solitary wave and periodic wave
solutions of (6) in explicit form. A short conclusion will be
where 𝜌, 𝜎, 𝑌, 𝐹, 𝜇, 𝑙0 , and ℎ are, respectively, the mass density, given in Section 4.
the area of transverse cross-section, the Young modulus,
the tension density of the strand, the rigidity of the elastic
membrane, the distance of the two strands, and the height of 2. Properties of Singular Points and
the membrane in the equilibrium position. Bifurcation Phase Portraits
In [16], introducing the transformation V = 𝑎𝑢 + 𝑏 (𝑎 and
Firstly, we compute the equilibrium points of system (6)
𝑏 are constants) in (1) yields
under different parametric conditions. It is clear that neither
𝑢𝑡𝑡 − 𝑐12 𝑢𝑥𝑥 = 𝑢3 (𝜇1 + 𝛽1 𝑎2 ) + 𝑢2 (2𝛽1 𝑎𝑏 + 𝑎𝛾1 ) the parameters 𝐵 nor 𝐶 takes value zero from the forms of
parameters 𝐴, 𝐵, and 𝐶; then 𝐴 equals zero if 𝑎 = ±√2/2.
+ 𝑢 (𝜆 1 + 𝑏𝛾1 + 𝛽1 𝑏2 ) , So we give equilibrium points in different parametric cases as
follows.
𝛾2 𝜇2 𝑏 (3)
𝑢𝑡𝑡 − 𝑐22 𝑢𝑥𝑥 = 𝑢3 (𝜇2 + 𝛽2 𝑎2 ) + 𝑢2 ( + + 3𝛽2 𝑎𝑏) (1) When 𝐴 = 0, there are two equilibrium points (0, 0)
𝑎 𝑎 and (−𝐶/𝐵, 0).
𝜆 2 𝑏 𝛽2 𝑏3 𝑐0 (2) When 𝐴 ≠0 and 𝛿1 > 0, there are three equilibrium
+ 𝑢 (𝜆 2 + 3𝛽2 𝑏2 ) +
+ + . points (0, 0), (𝜑− , 0), and (𝜑+ , 0) of (6) on the 𝜑-axis,
𝑎 𝑎 𝑎
Comparing (3) and using the physical quantities (2), it follows where 𝛿1 = 𝐵2 − 4𝐴𝐶, 𝜑− = (−𝐵 − √𝛿1 )/2𝐴, and
that 𝑏 = ℎ/√2 and 𝐹 = 𝑌; (3) can be written as 𝜑+ = (−𝐵 + √𝛿1 )/2𝐴.

𝑢𝑡𝑡 − 𝑐12 𝑢𝑥𝑥 = 𝐴𝑢3 + 𝐵𝑢2 + 𝐶𝑢 + 𝐷, (4) Define 𝑓0 (𝜑) = 𝐴𝜑3 + 𝐵𝜑2 + 𝐶𝜑; then the type of
equilibrium point (𝜑, 0) can be decided by 𝜆 ± = √𝑓0󸀠 (𝜑),
where 𝐴 = ((−2𝛼/ℎ3 ) + (4𝑎2 𝛼/ℎ3 )), 𝐵 = 6√2𝑎𝛼/ℎ2 , 𝐶 =
((−2𝛼/𝑙0 ) + (6𝛼/ℎ)), and 𝐷 = 0 with 𝛼 = 𝜇𝑙0 /𝜌𝜎. which are eigenvalues of linearized system of (6) at point
Reference [16] studied the traveling wave solutions of (𝜑, 0).
(4) and obtained one particular family of solitary kink- Secondly, we compute the intersection points of phase
type solutions for different values of Riccati parameter of portraits of (6) on the 𝜑-axis when the orbits pass through
(4). Unfortunately the results are not complete, because (4) the equilibrium point (0, 0), namely, when the Hamiltonian
admits tanh-type, periodic-type, and bell-type solutions; [16] constant 𝑀 = ℎ(0, 0) = 0. Let 𝑓(𝜑) = (1/2)𝐴𝜑4 + (2/3)𝐵𝜑3 +
only presented the tanh-type solutions. 𝐶𝜑2 , 𝛿2 = (4/9)𝐵2 − 2𝐴𝐶, 𝜑0 = −3𝐴/2𝐵, 𝜑1 = ((−2/3)𝐵 −
In this paper, we use the bifurcation method of dynamical √𝛿2 )/𝐴, and 𝜑2 = ((−2/3)𝐵 + √𝛿2 )/𝐴. When 𝐴 = 0, there are
systems [17–20] to study bell-type and periodic-type solu- a double zero point 0 and a zero point 𝜑0 of 𝑓(𝜑) on the 𝜑-
tions of (6), which can improve the results in [16]. We look axis. When 𝐴 ≠0 and 𝛿2 > 0, there are a double zero point 0
for solutions of (4) in the form of 𝑢(𝑥, 𝑡) = 𝜑(𝜉) with 𝜉 = and two zero points 𝜑1 and 𝜑2 of 𝑓(𝜑) on the 𝜑-axis. In order
(𝑥/𝑐1 ) − √2𝑡. Substituting 𝑢(𝑥, 𝑡) = 𝜑((𝑥/𝑐1 ) − √2𝑡) into (4), to draw bifurcation phase portraits of system (6), we need to
we have discuss the sign of parameters 𝐴, 𝐵, and 𝐶. The intersection
points of phase portraits of (6) on the 𝜑-axis are given as
𝜑𝜉𝜉 = 𝐴𝜑3 + 𝐵𝜑2 + 𝐶𝜑, (5) follows.
which is equivalent to the following two-dimensional system: (1) If 𝐴 = 0, there are two intersection points (0, 0) and
𝑑𝜑 𝑑𝑦 (𝜑0 , 0) of (6) on the 𝜑-axis (see Figure 1).
= 𝑦, = 𝐴𝜑3 + 𝐵𝜑2 + 𝐶𝜑. (6)
𝑑𝜉 𝑑𝜉 (2) If 𝐴 > 0, there are three intersection points (0, 0),
(𝜑1 , 0), and (𝜑2 , 0) of (6) on the 𝜑-axis and 𝜑1 < 𝜑2
Obviously, system (6) has the first integral
(see Figure 2).
1 1 1 1
𝐻 (𝜑, 𝑦) = 𝐴𝜑4 + 𝐵𝜑3 + 𝐶𝜑2 − 𝑦2 = 𝑀, (7) (3) If 𝐴 < 0, there are three intersection points (0, 0),
4 3 2 2 (𝜑1 , 0), and (𝜑2 , 0) of (6) on the 𝜑-axis and 𝜑2 < 𝜑1
where 𝑀 is the Hamiltonian constant. (see Figure 3).
Abstract and Applied Analysis 3

y y y y

𝜑0 0 𝜑 0 𝜑0 𝜑 0 𝜑0 𝜑 𝜑0 0 𝜑

(a) 𝐵 > 0, 𝐶 > 0 (b) 𝐵 > 0, 𝐶 < 0 (c) 𝐵 < 0, 𝐶 > 0 (d) 𝐵 < 0, 𝐶 < 0

Figure 1: The bifurcation phase portraits of system (6) with 𝐴 = 0.

y y y y

𝜑1 𝜑2 0 𝜑 𝜑1 0 𝜑2 𝜑 0 𝜑1 𝜑2 𝜑 𝜑1 0 𝜑2 𝜑

(a) 𝐵 > 0, 𝐶 > 0 (b) 𝐵 > 0, 𝐶 < 0 (c) 𝐵 < 0, 𝐶 > 0 (d) 𝐵 < 0, 𝐶 < 0

Figure 2: The bifurcation phase portraits of system (6) with 𝐴 > 0.

y y y y

𝜑2 0 𝜑1 𝜑 0 𝜑2 𝜑1 𝜑 𝜑2 0 𝜑1 𝜑 𝜑2 𝜑1 0 𝜑

(a) 𝐵 > 0, 𝐶 > 0 (b) 𝐵 > 0, 𝐶 < 0 (c) 𝐵 < 0, 𝐶 > 0 (d) 𝐵 < 0, 𝐶 < 0

Figure 3: The bifurcation phase portraits of system (6) with 𝐴 < 0.

Finally, from the above analysis about system (6), we give Solving (9) yields
twelve different phase portraits of (6) which are shown in
Figures 1–3. There exist homoclinic orbits and periodic orbits 3𝐶
𝑢1 = − 󵄨 󵄨 . (10)
in some phase portraits of Figures 1–3, which correspond to 𝐵 (cosh ( 𝐶 󵄨󵄨󵄨𝜉󵄨󵄨󵄨) + 1)
√
bell-type and periodic-type waves and are of special interest
to us.
(2) The case 𝐴 = 0, 𝐵 < 0, and 𝐶 > 0 (see Figure 1(c)).
3. Exact Explicit Representations of
Solitary Wave Solutions Similarly, in Figure 1(c) the homoclinic orbit has the
following expression:
(1) The case 𝐴 = 0, 𝐵 > 0, and 𝐶 > 0 (see Figure 1(a)).
2
𝑦2 = 𝐵𝜑3 + 𝐶𝜑2 (0 ⩽ 𝜑 ⩽ 𝜑0 ) . (11)
Corresponding to the homoclinic orbit of (6) passing 3
through points (𝜑0 , 0) and (0, 0) on the 𝜑−𝑦 plane, there exists
a smooth solitary wave solution. Now, (7) becomes Thus the corresponding smooth solitary wave solution is
obtained as (10) and its profile is shown in Figure 4(b).
2
𝑦2 = 𝐵𝜑3 + 𝐶𝜑2 (𝜑0 ⩽ 𝜑 ⩽ 0) . (8) (3) The case 𝐴 > 0, 𝐵 > 0, and 𝐶 > 0 (see Figure 2(a)).
3
In this case, on the 𝜑 − 𝑦 plane the homoclinic orbit of
Choosing 𝜑(0) = 𝜑0 , substituting (8) into the first equation of
(6) passing through points (𝜑2 , 0) and (0, 0) has the following
(6), and integrating along the homoclinic orbit, we have
expression:
𝜑0
1 󵄨 󵄨 1 2
∫ 𝑑𝑠 = 󵄨󵄨󵄨𝜉󵄨󵄨󵄨 . (9) 𝑦2 = 𝐴𝜑4 + 𝐵𝜑3 + 𝐶𝜑2 (𝜑2 ⩽ 𝜑 ⩽ 0) . (12)
𝜑 𝑠√(2/3) 𝐵𝑠 + 𝐶 2 3
4 Abstract and Applied Analysis

𝜑
𝜑
0 𝜉
u2

0 𝜉
𝜉 u3
𝜑 0
(a) (b) (c)
𝜑
u2 𝜑
0
𝜉
0 𝜉

u3

0 𝜉 𝜑

(d) (e) (f)

Figure 4: Smooth solitary wave and periodic wave solutions.

Choosing 𝜑(0) = 𝜑2 , substituting (12) into the first equation where 𝛼2 = (2(2𝐵2 − 9𝐴𝐶 + 𝐵√4𝐵2 − 18𝐴𝐶))/(6𝐵 +
of (6), and integrating along the homoclinic orbit, we have 3√4𝐵2 − 18𝐴𝐶). The profile of 𝑢3 is shown in Figure 4(d).
𝜑2
1 󵄨 󵄨 (5) The case 𝐴 < 0, 𝐵 > 0, and 𝐶 > 0 and 𝐴 < 0, 𝐵 < 0,
∫ 𝑑𝑠 = 󵄨󵄨󵄨𝜉󵄨󵄨󵄨 . (13) and 𝐶 > 0 (see Figures 3(a) and 3(c)).
𝜑 √ 2
𝑠 (1/2) 𝐴𝑠 + (2/3) 𝐵𝑠 + 𝐶
Solving (13), it follows that In this case, there exist two homoclinic orbits and their
expressions are (12) and (15), respectively. Similarly we get
4𝐶𝛼1 the corresponding solitary wave solutions as 𝑢2 and 𝑢3 , which
𝑢2 = , (14) coexist for the same speed. The plane graphs of the coexisting
𝛼12 𝑒−√𝐶|𝜉| + ((4/9) 𝐵2 − 2𝐴𝐶) 𝑒√𝐶|𝜉| − (4/3) 𝐵𝛼1
solitary wave solutions are shown in Figure 4(c), when 𝐴 < 0,
where 𝛼1 = (2(−2𝐵2 + 9𝐴𝐶 + 𝐵√4𝐵2 − 18𝐴𝐶))/(−6𝐵 + 𝐵 > 0, and 𝐶 > 0, and Figure 4(d), when 𝐴 < 0, 𝐵 < 0, and
3√4𝐵2 − 18𝐴𝐶). The profiles of solutions 𝑢1 and 𝑢2 both have 𝐶 > 0.
figure caption Figure 4(a). (6) The case 𝐴 < 0, 𝐵 > 0, and 𝐶 < 0 (see Figure 3(b)).
(4) The case 𝐴 > 0, 𝐵 < 0, and 𝐶 > 0 (see Figure 2(c)). In this case, on the 𝜑−𝑦 plane the orbit of (6) which passes
through points (𝜑1 , 0) and (𝜑2 , 0) is a periodic orbit; it has the
In this case, on the 𝜑 − 𝑦 plane the homoclinic orbit of
following expression:
(6) passing through points (𝜑1 , 0) and (0, 0) has the following
expression: 1 2
𝑦2 = 𝐴𝜑4 + 𝐵𝜑3 + 𝐶𝜑2 , (0 < 𝜑2 ⩽ 𝜑 ⩽ 𝜑1 ) . (18)
1 2 2 3
𝑦2 = 𝐴𝜑4 + 𝐵𝜑3 + 𝐶𝜑2 (0 ⩽ 𝜑 ⩽ 𝜑1 ) . (15)
Choosing 𝜑(0) = 𝜑1 , substituting (18) into the first equation
2 3
of (6), and integrating along the periodic orbit, we have
Choosing 𝜑(0) = 𝜑1 , substituting (15) into the first equation
𝜑1
of (6), and integrating along the homoclinic orbit, we have 1 󵄨 󵄨
∫ 𝑑𝑠 = 󵄨󵄨󵄨𝜉󵄨󵄨󵄨 . (19)
𝜑1
1 𝜑 𝑠√(1/2) 𝐴𝑠2 + (2/3) 𝐵𝑠 + 𝐶
󵄨 󵄨
∫ 𝑑𝑠 = 󵄨󵄨󵄨𝜉󵄨󵄨󵄨 . (16)
𝜑 𝑠√(1/2) 𝐴𝑠2 + (2/3) 𝐵𝑠 + 𝐶 Solving (19) we obtain the following periodic wave solution:
Solving (16), we get 6𝐶
𝑢4 = . (20)
√4𝐵2 − 18𝐴𝐶 cos (√−𝐶𝜉) − 2𝐵
4𝐶𝛼2
𝑢3 = , (17)
𝛼22 𝑒√𝐶|𝜉| + ((4/9) 𝐵2 − 2𝐴𝐶) 𝑒−√𝐶|𝜉| − (4/3) 𝐵𝛼2 The profile of 𝑢4 is shown in Figure 4(e).
Abstract and Applied Analysis 5

(7) The case 𝐴 < 0, 𝐵 < 0, and 𝐶 < 0 (see Figure 3(d)). [3] T. Lipniacki, “Chemically driven traveling waves in DNA,”
Physical Review E, vol. 60, p. 7253, 1999.
Similarly, corresponding to the periodic orbit of (6) [4] S. Yomosa, “Solitary excitations in deoxyribonuclei acid (DNA)
which passes through points (𝜑1 , 0) and (𝜑2 , 0), there exists double helices,” Physical Review A, vol. 30, no. 1, pp. 474–480,
a periodic wave solution. Now, the expression of the periodic 1984.
orbit is [5] C. T. Zhang, “Soliton excitations in deoxyribonucleic acid
(DNA) double helices,” Physical Review A, vol. 35, no. 2, pp. 886–
1 2
𝑦2 = 𝐴𝜑4 + 𝐵𝜑3 + 𝐶𝜑2 , (𝜑2 ⩽ 𝜑 ⩽ 𝜑1 < 0) . (21) 891, 1987.
2 3 [6] K. Forinash, “Nonlinear dynamics in a double-chain model of
DNA,” Physical Review B, vol. 43, p. 10734, 1991.
Choosing 𝜑(0) = 𝜑2 , substituting (21) into the first equation
of (6), and integrating along the periodic orbit, we have [7] L. V. Yakushevich, Nonlinear Physics of DNA, John Wiley &
Sons, Berlin, Germany, 2004.
𝜑
1 󵄨 󵄨 [8] S. W. Englander, N. R. Kallenbanch, A. J. Heeger, J. A.
∫ 𝑑𝑠 = 󵄨󵄨󵄨𝜉󵄨󵄨󵄨 . (22) Krumhansl, and S. Litwin, “Nature of the open state in long
𝜑2 𝑠√(1/2) 𝐴𝑠2 + (2/3) 𝐵𝑠 + 𝐶 polynucleotide double helices: possibility of soliton excitations,”
Proceedings of the National Academy of Sciences of the United
Solving (22), it follows that States of America, vol. 77, pp. 7222–7226, 1980.
6𝐶 [9] S. Yomosa, “Soliton excitations in deoxyribonucleic acid (DNA)
𝑢5 = . (23) double helices,” Physical Review A, vol. 27, no. 4, pp. 2120–2125,
−√4𝐵2 − 18𝐴𝐶 cos (√−𝐶𝜉) − 2𝐵 1983.
[10] S. Takeno and S. Homma, “A coupled base-rotator model for
The profile of 𝑢5 is shown in Figure 4(f). structure and dynamics of DNA—local fluctuations in helical
twist angles and topological solitons,” Progress of Theoretical
Physics, vol. 72, no. 4, pp. 679–693, 1984.
4. Conclusion
[11] M. Peyrard and A. R. Bishop, “Statistical mechanics of a nonlin-
In this paper, we employ the method of dynamical systems ear model for DNA denaturation,” Physical Review Letters, vol.
to study the nonlinear dynamical equations of DNA model. 62, no. 23, pp. 2755–2758, 1989.
The bifurcation phase portraits of the DNA under some para- [12] T. Dauxois, M. Peyrard, and A. R. Bishop, “Entropy-driven
metric conditions are drawn, and explicit exact expressions of DNA denaturation,” Physical Review E, vol. 47, no. 1, pp. R44–
bell-shaped solitary waves and periodic waves are obtained R47, 1993.
via some special homoclinic and periodic orbits; their planar [13] V. Muto, P. S. Lomdahl, and P. L. Christiansen, “Two-
graphs are simulated. These solutions are new and different dimensional discrete model for DNA dynamics: longitudinal
wave propagation and denaturation,” Physical Review A, vol. 42,
from those in [14–16]; some existing results are improved.
no. 12, pp. 7452–7458, 1990.
Furthermore, in general, the larger the wave speed, the higher
[14] D. X. Kong, S. Y. Lou, and J. Zeng, “Nonlinear dynamics in a new
the wave crest. So for the same wave speed, it is interesting
double Chain-model of DNA,” Communications in Theoretical
that the bell-shaped solitary waves of different heights are Physics, vol. 36, no. 6, pp. 737–742, 2001.
found to coexist.
[15] X.-M. Qian and S.-Y. Lou, “Exact solutions of nonlinear
dynamics equation in a new double-chain model of DNA,”
Conflict of Interests Communications in Theoretical Physics, vol. 39, no. 4, pp. 501–
505, 2003.
The authors declare that there is no conflict of interests [16] W. Alka, A. Goyal, and C. Nagaraja Kumar, “Nonlinear dynam-
regarding the publication of this paper. ics of DNA—riccati generalized solitary wave solutions,” Physics
Letters A: General, Atomic and Solid State Physics, vol. 375, no.
3, pp. 480–483, 2011.
Acknowledgments
[17] S. N. Chow and J. K. Hale, Methods of Bifurcation Theory,
This work was supported by the National Natural Sci- Springer, Berlin, Germany, 1981.
ence Foundation of China (no. 11226303) and Guangdong [18] J. B. Li and Z. R. Liu, “Smooth and non-smooth traveling waves
Province (no. 2013KJCX0189). The authors thank the editors in a nonlinearly dispersive equation,” Applied Mathematical
for their hard working and also gratefully acknowledge Modelling, vol. 25, pp. 41–56, 2000.
helpful comments and suggestions by reviewers. [19] J. B. Li and Z. R. Liu, “Traveling wave solutions for a class of
nonlinear dispersive equations,” Chinese Annals of Mathematics,
vol. 23, p. 397, 2002.
References [20] Z. R. Liu and Z. Y. Ouyang, “A note on solitary waves for
modified forms of Camassa-Holm and Degasperis-Procesi
[1] J. D. Watson and F. H. C. Crick, “Molecular structure of nucleic equations,” Physics Letters A, vol. 366, no. 4-5, pp. 377–381, 2007.
acids: a structure for deoxyribose nucleic acid,” Nature, vol. 171,
no. 4356, pp. 737–738, 1953.
[2] M. Peyrard, S. C. López, and G. James, “Modelling DNA at the
mesoscale: a challenge for nonlinear science?” Nonlinearity, vol.
21, p. T91, 2008.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 928726, 6 pages
http://dx.doi.org/10.1155/2014/928726

Research Article
Global Attractivity of an Integrodifferential Model of Mutualism

Xiangdong Xie,1 Fengde Chen,1 Kun Yang,2 and Yalong Xue2

1
Department of Mathematics, Ningde Normal University, Fujian 352100, China
2
College of Mathematics and Computer Science, Fuzhou University, Fuzhou, Fujian 350116, China

Correspondence should be addressed to Xiangdong Xie; ndsyxxd@163.com

Received 12 January 2014; Revised 8 March 2014; Accepted 9 March 2014; Published 1 April 2014

Academic Editor: Yongli Song

Copyright © 2014 Xiangdong Xie et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Sufficient conditions are obtained for the global attractivity of the following integrodifferential model of mutualism: 𝑑𝑁1 (𝑡)/𝑑𝑡 =
∞ ∞ ∞
𝑟1 𝑁1 (𝑡)[((𝐾1 + 𝛼1 ∫0 𝐽2 (𝑠)𝑁2 (𝑡 − 𝑠)𝑑𝑠)/(1 + ∫0 𝐽2 (𝑠)𝑁2 (𝑡 − 𝑠)𝑑𝑠)) − 𝑁1 (𝑡)], 𝑑𝑁2 (𝑡)/𝑑𝑡 = 𝑟2 𝑁2 (𝑡)[((𝐾2 + 𝛼2 ∫0 𝐽1 (𝑠)𝑁1 (𝑡 − 𝑠)𝑑𝑠)/(1 +
∞
∫0 𝐽1 (𝑠)𝑁1 (𝑡 − 𝑠)𝑑𝑠)) − 𝑁2 (𝑡)], where 𝑟𝑖 , 𝐾𝑖 , and 𝛼𝑖 , 𝑖 = 1, 2, are all positive constants. Consider 𝛼𝑖 > 𝐾𝑖 , 𝑖 = 1, 2. Consider
∞
𝐽𝑖 ∈ 𝐶([0, +∞), [0, +∞)) and ∫0 𝐽𝑖 (𝑠)𝑑𝑠 = 1, 𝑖 = 1, 2. Our result shows that conditions which ensure the permanence of the
system are enough to ensure the global stability of the system. The result not only improves but also complements some existing
ones.

1. Introduction Chen et al. [2] had given two examples to show that under the
assumption 𝑎11 𝑎22 > 𝑎12 𝑎21 , a condition which could ensure
Mutualism, one of the most important relationships in the the global stability of the system without delay, the system
theory of ecology, however, was pointed out by Murray as still admits unbounded solution. He and Gopalsamy [3] and
follows [1]: “this area has not been as widely sutdied as the Mukherjee [4] tried to investigate the persistent and stability
others even though its importance is comparable to that of property of the nonautonomous case of above system; how-
predator-prey and competition interactions.” ever, in their main results, in addition to condition 𝑎11 𝑎22 >
Traditional two species Lotka-Volterra take the form 𝑎12 𝑎21 , they further assumed that the density of one of the
𝑑𝑁1 𝑁 𝑁 species should be bounded from above, such an assumption
= 𝑟1 𝑁1 (1 − 1 + 𝑏12 2 ) , is by no means easy to verify. To overcome this difficulty,
𝑑𝑡 𝐾1 𝐾1
(1) Lu et al. [5, 6] and Nakata and Muroya [7] tried to give
𝑑𝑁2 𝑁 𝑁 restriction on the coefficients of the system or restriction on
= 𝑟2 𝑁2 (1 − 2 + 𝑏21 1 ) . the delay of the system, and some interesting results about the
𝑑𝑡 𝐾2 𝐾2
permanence of Lotka-Volterra type mutualism system with
Murray [1] gave detail analysis of the phase trajectories for delay were obtained. Liu et al. [8] and Lu [9] also investigated
the above system. He also pointed out that the system has the positive periodic solution of the Lotka-Volterra type
certain drawback; one is the sensitivity between unbounded mutualism model.
growth and a finite positive steady state. Despite the drawback On the other hand, stimulated by the functional response
of the system, since it is the most simple model on the of the predator-prey system, Wright [10] proposed the follow-
mutualism, scholars incorporated delays to the above system ing two species mutualism model:
and proposed the following system:
𝑑𝑥1 𝑑𝑁 𝑏1 𝑎1 𝑀
= 𝑥1 (𝑡) [𝑟1 − 𝑎11 𝑥1 (𝑡 − 𝜏11 ) + 𝑎12 𝑥2 (𝑡 − 𝜏12 )] , = 𝑟1 𝑁 [𝐾1 − 𝑐1 𝑁 + ],
𝑑𝑡 𝑑𝑡 1 + 𝑎1 𝑇ℎ1 𝑀
(2) (3)
𝑑𝑥2 𝑑𝑀 𝑏2 𝑎2 𝑁
= 𝑥2 (𝑡) [𝑟2 + 𝑎21 𝑥1 (𝑡 − 𝜏21 ) − 𝑎22 𝑥2 (𝑡 − 𝜏22 )] . = 𝑟2 𝑀 [𝐾2 − 𝑐2 𝑁 + ].
𝑑𝑡 𝑑𝑡 1 + 𝑎2 𝑇ℎ2 𝑁
2 Abstract and Applied Analysis

Obviously, the model could be revised as follows: Theorem A. The system (6) is always permanent. That is, there
exist constants 𝑚𝑖 , 𝑀𝑖 , 𝑖 = 1, 2, which are independent of the
𝑑𝑁 (𝑡) 𝐾 + 𝛼1 𝑀
= 𝑟1 𝑁 [ 1 − 𝑐 𝑁] , solution of the system (6), such that
𝑑𝑡 1 + 𝑎1 𝑇ℎ1 𝑀 1
(4) 𝑚𝑖 ≤ lim inf 𝑁𝑖 (𝑡) ≤ lim sup𝑁𝑖 (𝑡) ≤ 𝑀𝑖 , 𝑖 = 1, 2. (7)
𝑑𝑀 (𝑡) 𝐾 + 𝛼2 𝑁 𝑡 → +∞ 𝑡 → +∞
= 𝑟2 𝑀 [ 2 − 𝑐 𝑀] ,
𝑑𝑡 1 + 𝑎2 𝑇ℎ2 𝑁 2 Such a result is a roughly one, since it only tells us
where 𝛼1 = 𝐾1 + 𝑎1 𝑇ℎ1 , 𝛼2 = 𝐾2 + 𝑎2 𝑇ℎ2 , and one could easily that the solution is finally bounded above and below by
see that 𝛼𝑖 > 𝐾𝑖 , 𝑖 = 1, 2. positive constants and there is no fine description of the stable
It is well known that in a more realistic model the delay or unstable property of the solution, for example, whether
effect should be an average over past populations. This results the delay of the system could induce the Hopf bifurcation
in an equation with a distributed delay or an infinite delay. to period solution or not? Does the system admit some
Based on the model (4), Gopalsamy [11] further proposed the kind of chaotic behaviors? Is it difficult to obtain sufficient
following two species mutualism model: conditions which ensure the global attractivity of the positive
solution? Indeed, to the best of the authors’ knowledge, to
∞
𝑑𝑁1 (𝑡) 𝐾1 + 𝛼1 ∫0 𝐾2 (𝑠) 𝑁2 (𝑡 − 𝑠) 𝑑𝑠 this day, still no scholars investigate the stability property
= 𝑟1 𝑁1 [ ∞ − 𝑁1 (𝑡)] , of the system (6), which is one of the most important
𝑑𝑡 1 + ∫0 𝐾2 (𝑠) 𝑁2 (𝑡 − 𝑠) 𝑑𝑠 topics in the study of population dynamics. Noting that
[ ]
system (6) is nonautonomous one and for such kind of
∞
𝑑𝑁2 (𝑡) 𝐾2 + 𝛼2 ∫0 𝐾1 (𝑠) 𝑁1 (𝑡 − 𝑠) 𝑑𝑠 system, generally speaking, by constructing some suitable
= 𝑟2 𝑁2 [ ∞ − 𝑁2 (𝑡)] . Lyapunov functional, one could always obtain some sufficient
𝑑𝑡 1 + ∫0 𝐾1 (𝑠) 𝑁1 (𝑡 − 𝑠) 𝑑𝑠
[ ] conditions which ensure the stability of the system; however,
(5) the condition is not easy to verify [14]. This motivated us to
investigate the stability property of the system (5).
However, the author did not investigate the dynamic behav-
From the point of view of biology, in the sequel, we will
iors of the system.
consider (5) together with the initial conditions
Recently, Li and Xu [12] proposed and studied the follow-
ing nonautonomous case of the system (5): 𝑁𝑖 (𝑠) = 𝜙𝑖 (𝑠) , 𝑠 ∈ (−∞, 0] , 𝑖 = 1, 2, (8)
𝑑𝑁1 (𝑡) where 𝜙𝑖 ∈ 𝐵𝐶+ and
𝑑𝑡
𝐵𝐶+ = {𝜙 ∈ 𝐶 ((−∞, 0] , [0, +∞)) :
= 𝑟1 (𝑡) 𝑁1 (𝑡)
∞ 𝜙 (0) > 0, 𝜙 be bounded} , (9)
𝐾1 (𝑡) + 𝛼1 (𝑡) ∫0 𝐽2 (𝑠) 𝑁2 (𝑡 − 𝑠) 𝑑𝑠
×[ ∞ − 𝑁1 (𝑡)] , 𝑖 = 1, 2.
1 + ∫0 𝐽2 (𝑠) 𝑁2 (𝑡 − 𝑠) 𝑑𝑠
[ ]
From [15], system (5) has a unique positive solution
𝑑𝑁2 (𝑡) (𝑁1 (𝑡), 𝑁2 (𝑡)) satisfying the initial condition (8).
𝑑𝑡 The aim of this paper is, by further developing the analysis
technique of Chen and You [13] and de Oca and Vivas [16]
= 𝑟2 (𝑡) 𝑁2 (𝑡) and using the differential inequality theory, to obtain a set of
∞ sufficient conditions to ensure the global attractivity of the
𝐾2 (𝑡) + 𝛼2 (𝑡) ∫0 𝐽1 (𝑠) 𝑁1 (𝑡 − 𝑠) 𝑑𝑠 system (5). More precisely, we will prove the following result.
×[ ∞ − 𝑁2 (𝑡)] ,
1 + ∫0 𝐽1 (𝑠) 𝑁1 (𝑡 − 𝑠) 𝑑𝑠
[ ] Theorem 1. System (5) admits a unique positive equilibrium
(6) (𝑁1∗ , 𝑁2∗ ), which is globally attractive; that is, for any positive
where 𝑟𝑖 , 𝐾𝑖 , 𝛼𝑖 , and 𝜎𝑖 , 𝑖 = 1, 2, are continuous functions solution (𝑁1 (𝑡), 𝑁2 (𝑡)) of system (5) with the initial condition
bounded above and below by positive constants. Consider (8), one has
𝛼𝑖 > 𝐾𝑖 , 𝑖 = 1, 2. Consider 𝐽𝑖 ∈ 𝐶([0, +∞), [0, +∞)) and lim 𝑁𝑖 (𝑡) = 𝑁𝑖∗ , 𝑖 = 1, 2.
∞ 𝑡 → +∞
(10)
∫0 𝐽𝑖 (𝑠)𝑑𝑠 = 1, 𝑖 = 1, 2. Under the assumption that 𝑟𝑖 ,
𝐾𝑖 , and 𝛼𝑖 , 𝑖 = 1, 2, are continuous periodic functions with We will prove this theorem in the next section and then
common period 𝜔. 𝛼𝑖 > 𝐾𝑖 , 𝑖 = 1, 2, 𝐽𝑖 ∈ 𝐶([0, +∞), [0, +∞)) give a brief discussion in Section 3. For more works on the
∞
and ∫0 𝐽𝑖 (𝑠)𝑑𝑠 = 1, 𝑖 = 1, 2. By applying the coincidence mutualism or cooperation system, one could refer to [12–15,
degree theory, they showed that system (6) admits at least one 17–21] and the references cited therein.
positive 𝜔-periodic solution. Chen and You [13] argued that
the general nonautonomous case is more suitable. Concerned 2. Proof of the Main Result
with the persistent property of the system (6), by applying an
integral inequality (see Lemma 3 in the next section), they Now let us state several lemmas which will be useful in the
obtained the following result. proving of main result.
Abstract and Applied Analysis 3

Lemma 2. System (5) admits a unique positive equilibrium Now we are in the position of proving the main result of
(𝑁1∗ , 𝑁2∗ ). this paper.

Proof. The positive equilibrium of the system (5) satisfies the Proof of Theorem 1. Let (𝑁1 (𝑡), 𝑁2 (𝑡)) be any positive solu-
following equation: tion of the system (5) with initial condition (8). Similarly to
the analysis of (11)–(17) in [13], from the first equation of the
𝐾1 + 𝛼1 𝑁2 system (5) it follows that
− 𝑁1 = 0,
1 + 𝑁2
(11) 𝑑𝑁1 (𝑡)
𝐾2 + 𝛼2 𝑁1 ≤ 𝑁1 (𝑡) [𝑟1 (𝐾1 + 𝛼1 ) − 𝑟1 𝑁1 (𝑡)] . (16)
− 𝑁2 = 0. 𝑑𝑡
1 + 𝑁1
Thus, as a direct corollary of Lemma 4, according to (16), one
System (11) admits a unique positive solution (𝑁1∗ , 𝑁2∗ ), has
where
lim sup𝑁1 (𝑡) ≤ 𝐾1 + 𝛼1 , (17)
𝑡 → +∞
−𝐴 2 + √𝐴22 − 4𝐴 1 𝐴 3
𝑁1∗ = , and so, from Lemma 3 we have
2𝐴 1
∞
−𝐵2 + √𝐵22 − 4𝐵1 𝐵3 lim sup ∫ 𝐽1 (𝑠) 𝑁1 (𝑡 − 𝑠) 𝑑𝑠 ≤ 𝐾1 + 𝛼1 . (18)
𝑁2∗ = , 𝑡 → +∞ 0
2𝐵1
Hence, for enough small 𝜀 > 0, it follows from (17) and (18)
𝐴 1 = 1 + 𝛼2 , that there exists a 𝑇1󸀠 > 0 such that
(12)
𝐴 2 = 𝐾2 − 𝐾1 − 𝛼2 𝛼1 + 1, def
𝑁1 (𝑡) < 𝐾1 + 𝛼1 + 𝜀 = 𝑀1(1) ,
𝐴 3 = −𝛼1 𝐾2 − 𝐾1 , ∞ 𝑡

𝐵1 = 1 + 𝛼1 , ∫ 𝐽1 (𝑠) 𝑁1 (𝑡 − 𝑠) 𝑑𝑠 = ∫ 𝐽1 (𝑡 − 𝑠) 𝑁1 (𝑠) 𝑑𝑠
0 −∞

𝐵2 = 𝐾1 − 𝐾2 − 𝛼1 𝛼2 + 1, def
≤ 𝐾1 + 𝛼1 + 𝜀 = 𝑀1(1) for 𝑡 > 𝑇1󸀠 .
𝐵3 = −𝐾2 − 𝛼2 𝐾1 . (19)

This ends the proof of Lemma 2. Similarly, for above 𝜀 > 0, it follows from the second equation
of the system (5) that there exists a 𝑇1 > 𝑇1󸀠 such that
Following Lemma 3 is Lemma 3 of de Oca and Vivas [16].
def
Lemma 3. Let 𝑥 : 𝑅 → 𝑅 be a bounded nonnegative 𝑁2 (𝑡) < 𝐾2 + 𝛼2 + 𝜀 = 𝑀2(1) ,
continuous function, and let 𝑘 : [0, +∞) → (0, +∞) be a ∞
def
∞
continuous kernel such that ∫0 𝑘(𝑠)𝑑𝑠 = 1. Then ∫ 𝐽2 (𝑠) 𝑁2 (𝑡 − 𝑠) 𝑑𝑠 < 𝐾2 + 𝛼2 + 𝜀 = 𝑀2(1) (20)
0
𝑡
for 𝑡 > 𝑇1 .
lim inf 𝑥 (𝑡) ≤ lim inf ∫ 𝑘 (𝑡 − 𝑠) 𝑥 (𝑠) 𝑑𝑠
𝑡 → +∞ 𝑡 → +∞ −∞
Noting that the function 𝑔1 (𝑥) = ((𝐾1 + 𝛼1 𝑥)/(1 + 𝑥)) (𝛼1 >
𝑡
≤ lim sup ∫ 𝑘 (𝑡 − 𝑠) 𝑥 (𝑠) 𝑑𝑠 (13) 𝐾1 , 𝑥 ≥ 0) is a strictly increasing function, hence, (20)
𝑡 → +∞ −∞ together with the first equation of the system (5) implies
≤ lim sup𝑥 (𝑡) . 𝑟1 (𝐾1 + 𝛼1 𝑀2(1) )
𝑡 → +∞ 𝑑𝑁1 (𝑡)
< 𝑁1 (𝑡) [ − 𝑟1 𝑁1 (𝑡)]
As a direct corollary of Lemma 2.2 of Chen [22], we have
𝑑𝑡 1 + 𝑀2(1) (21)
the following lemma. for 𝑡 > 𝑇1 .
Lemma 4. If 𝑎 > 0, 𝑏 > 0, and 𝑥̇≥ 𝑥(𝑏 − 𝑎𝑥), when 𝑡 ≥ 0 and Therefore, by Lemma 4, we have
𝑥(0) > 0, one has
𝑏 𝐾1 + 𝛼1 𝑀2(1)
lim inf 𝑥 (𝑡) ≥ . (14) lim sup𝑁1 (𝑡) ≤ . (22)
𝑡 → +∞ 𝑎 𝑡 → +∞ 1 + 𝑀2(1)
If 𝑎 > 0, 𝑏 > 0 and 𝑥̇≤ 𝑥(𝑏 − 𝑎𝑥), when 𝑡 ≥ 0 and 𝑥(0) > 0, Thus, from Lemma 3 we have
one has
𝑏
∞ 𝐾1 + 𝛼1 𝑀2(1)
lim sup𝑥 (𝑡) ≤ . (15) lim sup ∫ 𝐽1 (𝑠) 𝑁1 (𝑡 − 𝑠) 𝑑𝑠 ≤ . (23)
𝑡 → +∞ 𝑎 𝑡 → +∞ 0 1 + 𝑀2(1)
4 Abstract and Applied Analysis

That is, for 𝜀 > 0 defined by (19), there exists a 𝑇2󸀠 > 𝑇1 such Noting that the function 𝑔1 (𝑥) = ((𝐾1 + 𝛼1 𝑥)/(1 + 𝑥)) (𝛼1 >
that 𝐾1 ) is a strictly increasing function, hence, (30) together with
the first equation of the system (5) implies
𝐾1 + 𝛼1 𝑀2(1) 𝜀 def (2)
𝑁1 (𝑡) < + = 𝑀1 > 0,
1+ 𝑀2(1) 2 𝑑𝑁1 (𝑡) 𝑟1 (𝐾1 + 𝛼1 𝑚2(1) )
> 𝑁1 (𝑡) [ − 𝑟1 𝑁1 (𝑡)]
∞ 𝐾1 + 𝛼1 𝑀2(1) 𝜀 def (2)
𝑑𝑡 1 + 𝑚2(1) (31)
∫ 𝐽1 (𝑠) 𝑁1 (𝑡 − 𝑠) 𝑑𝑠 < + = 𝑀1 > 0
0 1+ 𝑀2(1) 2 for 𝑡 > 𝑇3 .

for 𝑡 > 𝑇2󸀠 . Therefore, by Lemma 4, we have

(24)
𝐾1 + 𝛼1 𝑚2(1)
Similarly to the analysis of (22)–(24), from (19) and the lim inf 𝑁1 (𝑡) ≥ . (32)
second equation of the system (5), there exists a 𝑇2 > 𝑇2󸀠 such
𝑡 → +∞ 1 + 𝑚2(1)
that
Thus, from Lemma 3 we have
𝐾2 + 𝛼2 𝑀1(1) 𝜀 def
𝑁2 (𝑡) < + = 𝑀2(2) > 0, ∞ 𝐾1 + 𝛼1 𝑚2(1)
1 + 𝑀1(1) 2 lim inf ∫ 𝐽1 (𝑠) 𝑁1 (𝑡 − 𝑠) 𝑑𝑠 ≥ . (33)
𝑡 → +∞ 0 1 + 𝑚2(1)
∞ 𝐾2 + 𝛼2 𝑀1(1) 𝜀 def (2)
∫ 𝐽2 (𝑠) 𝑁2 (𝑡 − 𝑠) 𝑑𝑠 < + = 𝑀2 > 0 That is, there exists a 𝑇4󸀠 > 𝑇3 such that
0 1+ 𝑀1(1) 2

for 𝑡 > 𝑇2 . 𝐾1 + 𝛼1 𝑚2(1) 𝜀 def (2)

𝑁1 (𝑡) > − = 𝑚1 > 0,
(25) 1+ 𝑚2(1) 2
Since the function 𝑔1 (𝑥) = ((𝐾1 + 𝛼1 𝑥)/(1 + 𝑥)) (𝛼1 > 𝐾1 , ∞ 𝐾1 + 𝛼1 𝑚2(1) 𝜀 def (2)
𝑥 ≥ 0) is a strictly increasing function, one could easily see ∫ 𝐽1 (𝑠) 𝑁1 (𝑡 − 𝑠) 𝑑𝑠 > − = 𝑚1 > 0
that 𝑔(𝑥) ≥ 𝑔(0) = 𝐾1 , and so, from the first equation of the 0 1+ 𝑚2(1) 2
system (5), it follows that
for 𝑡 > 𝑇4󸀠 .
𝑑𝑁1 (𝑡) (34)
≥ 𝑁1 (𝑡) [𝑟1 𝐾1 − 𝑟1 𝑁1 (𝑡)] . (26)
𝑑𝑡
Similarly to the analysis of (32)–(34), from (28) and the
Thus, as a direct corollary of Lemma 4, according to (25), one second equation of the system (5), there exists a 𝑇4 > 𝑇4󸀠 such
has that
lim inf 𝑁1 (𝑡) ≥ 𝐾1 , (27)
𝑡 → +∞ 𝐾2 + 𝛼2 𝑚1(1) 𝜀 def (2)
𝑁2 (𝑡) > − = 𝑚2 > 0,
and so, from Lemma 3, we have 1+ 𝑚1(1) 2
∞
lim inf ∫ 𝐽1 (𝑠) 𝑁1 (𝑡 − 𝑠) 𝑑𝑠 ≥ 𝐾1 . (28)
∞ 𝐾2 + 𝛼2 𝑚1(1) 𝜀 def (2)
𝑡 → +∞ 0
∫ 𝐽2 (𝑠) 𝑁2 (𝑡 − 𝑠) 𝑑𝑠 > − = 𝑚2 > 0
0 1+ 𝑚1(1) 2
Hence, for enough small 𝜀 > 0 (𝜀 < (1/2) min{𝐾1 , 𝐾2 }), it
follows from (27) and (28) that there exists a 𝑇3󸀠 > 0 such that for 𝑡 > 𝑇4 .
(35)
def
𝑁1 (𝑡) > 𝐾1 − 𝜀 = 𝑚1(1) ,
One could easily see that
∞
def
∫ 𝐽1 (𝑠) 𝑁1 (𝑡 − 𝑠) 𝑑𝑠 ≥ 𝐾1 − 𝜀 = 𝑚1(1) (29)
𝐾1 + 𝛼1 𝑀2(1) 𝜀
0
𝑀1(2) = + < 𝐾1 + 𝛼1 + 𝜀 = 𝑀1(1) ;
1+ 𝑀2(1) 2
for 𝑡 > 𝑇3󸀠 .
Similarly, for above 𝜀 > 0, it follows from the second equation 𝐾2 + 𝛼2 𝑀1(1) 𝜀
𝑀2(2) = + < 𝐾2 + 𝛼2 + 𝜀 = 𝑀2(1) ;
of the system (5) that there exists a 𝑇3 > 𝑇3󸀠 such that 1+ 𝑀1(1) 2
(36)
def
𝑁2 (𝑡) > 𝐾2 − 𝜀 = 𝑚2(1) , 𝐾1 + 𝛼1 𝑚2(1) 𝜀
𝑚1(2) = − > 𝐾1 − 𝜀 = 𝑚1(1) ;
∞ 1 + 𝑚2(1) 2
def
∫ 𝐽2 (𝑠) 𝑁2 (𝑡 − 𝑠) 𝑑𝑠 > 𝐾2 − 𝜀 = 𝑚2(1) (30)
0 𝐾2 + 𝛼2 𝑚1(1) 𝜀
𝑚2(2) = − > 𝐾2 − 𝜀 = 𝑚2(1) .
for 𝑡 > 𝑇3 . 1+ 𝑚1(1) 2
Abstract and Applied Analysis 5

Repeating the above procedure, we get four sequences 𝑀𝑖(𝑛) , Therefore,

𝑚𝑖(𝑛) , 𝑖 = 1, 2, 𝑛 = 1, 2, . . ., such that for 𝑛 ≥ 2
lim 𝑀𝑖(𝑛) = 𝑁𝑖 , lim 𝑚𝑖(𝑛) = 𝑁𝑖 , 𝑖 = 1, 2. (42)
𝑡 → +∞ 𝑡 → +∞
𝐾1 + 𝛼1 𝑀2(𝑛−1) 𝜀
𝑀1(𝑛) = + ; Letting 𝑛 → +∞ in (37), we obtain
1 + 𝑀2(𝑛−1) 𝑛

𝐾2 + 𝛼2 𝑀1(𝑛−1) 𝜀 𝐾1 + 𝛼1 𝑁2 𝐾2 + 𝛼2 𝑁1
𝑀2(𝑛) = + ; 𝑁1 = , 𝑁2 = ,
1 + 𝑀1(𝑛−1) 𝑛 1 + 𝑁2 1 + 𝑁1
(43)
(37) 𝐾 + 𝛼1 𝑁2 𝐾 + 𝛼2 𝑁1
𝐾1 + 𝛼1 𝑚2(𝑛−1) 𝜀 𝑁1 = 1 , 𝑁2 = 2
𝑚1(𝑛) = − ; 1 + 𝑁2 1 + 𝑁1
1 + 𝑚2(𝑛−1) 𝑛
and (43) shows that (𝑁1 , 𝑁2 ) and (𝑁1 , 𝑁2 ) are solutions
𝐾2 + 𝛼2 𝑚1(𝑛−1) 𝜀
𝑚2(𝑛) = − . of (11). By Lemma 2, (11) has a unique positive solution
1 + 𝑚1(𝑛−1) 𝑛 𝐸∗ (𝑁1∗ , 𝑁2∗ ). Hence, we conclude that

Obviously, 𝑁𝑖 = 𝑁𝑖 = 𝑁𝑖∗ , 𝑖 = 1, 2; (44)

𝑚𝑖(𝑛) < 𝑁𝑖 (𝑡) < 𝑀𝑖(𝑛) , for 𝑡 ≥ 𝑇2𝑛 , 𝑖 = 1, 2. (38) that is,

We claim that sequences 𝑀𝑖(𝑛) , 𝑖 = 1, 2 are nonincreasing lim 𝑁𝑖 (𝑡) = 𝑁𝑖∗ 𝑖 = 1, 2. (45)
𝑡 → +∞
and sequences 𝑚𝑖(𝑛) , 𝑖 = 1, 2 are nondecreasing. To prove this
claim, we will carry out by induction. Firstly, from (36) we Thus, the unique interior equilibrium 𝐸∗ (𝑁1∗ , 𝑁2∗ ) is globally
have attractive. This completes the proof of Theorem 1.

𝑀𝑖(2) < 𝑀𝑖(1) , 𝑚𝑖(2) > 𝑚𝑖(1) , 𝑖 = 1, 2. (39)

3. Discussion
Let us assume now that our claim is true for 𝑛; that is,
As was pointed out in the introduction section, for Lotka-
𝑀𝑖(𝑛) < 𝑀𝑖(𝑛−1) , 𝑚𝑖(𝑛) > 𝑚𝑖(𝑛−1) , 𝑖 = 1, 2. (40) Volterra type mutualism system, conditions which ensure
the globally stability of the system are not so easily verified
Again from the strict increasing of the function 𝑔𝑖 (𝑥) = ((𝐾𝑖 + [3, 4]. In this paper, we study the stability property of the
𝛼𝑖 𝑥)/(1 + 𝑥)) (𝛼𝑖 > 𝐾𝑖 , 𝑖 = 1, 2), we immediately obtain integrodifferential model of mutualism (5). By applying the
iterative technique, we obtain a set of sufficient conditions
𝐾1 + 𝛼1 𝑀2(𝑛) 𝜀 which guarantee the global attractivity of the coexistence
𝑀1(𝑛+1) = + equilibrium. Our result (Theorem 1) shows that the condi-
1 + 𝑀2(𝑛) 𝑛+1
tions which ensure the permanence of the system are enough
𝐾1 + 𝛼1 𝑀2(𝑛−1) 𝜀 to ensure the global attractivity of the system. The new finding
< + = 𝑀1(𝑛) ; of this paper is that we found the iterative bound of the
1+ 𝑀2(𝑛−1) 𝑛 solution, such a finding is not detected by Chen and You [13].
Based on this finding, it is possible to obtain the subtle result
𝐾2 + 𝛼2 𝑀1(𝑛) 𝜀
𝑀2(𝑛+1) = + about the stability of the system. We mention here that we
1+ 𝑀1(𝑛) 𝑛+1 did not consider the delay in the intraspecific competition;
whether such kind of delay could induce bifurcation or not is
𝐾2 + 𝛼2 𝑀1(𝑛−1) 𝜀 still unknown; we leave this for future study.
< + = 𝑀2(𝑛) ;
1 + 𝑀1(𝑛−1) 𝑛
(41) Conflict of Interests
𝐾1 + 𝛼1 𝑚2(𝑛) 𝜀
𝑚1(𝑛+1) = − The authors declare that there is no conflict of interests
1 + 𝑚2(𝑛) 𝑛+1
regarding the publication of this paper.
𝐾1 + 𝛼1 𝑚2(𝑛−1) 𝜀
> − = 𝑚1(𝑛) ;
1+ 𝑚2(𝑛−1) 𝑛 Acknowledgments

𝐾2 + 𝛼2 𝑚1(𝑛) The authors are grateful to anonymous referees for their

𝜀
𝑚2(𝑛+1) = − excellent suggestions, which greatly improved the presen-
1 + 𝑚1(𝑛) 𝑛+1 tation of the paper. Also, this work was supported by the
Natural Science Foundation of Fujian Province (2013J01011
𝐾2 + 𝛼2 𝑚1(𝑛−1) 𝜀
> − = 𝑚2(𝑛) . and 2013J01010) and the Foundation of Fujian Education
1+ 𝑚1(𝑛−1) 𝑛 Bureau (JA13361).
6 Abstract and Applied Analysis

References [20] Y. Li and T. Zhang, “Permanence of a discrete 𝑁-species

cooperation system with time-varying delays and feedback
[1] J. D. Murray, Mathematical Biology, Springer, Berlin, Germany, controls,” Mathematical and Computer Modelling, vol. 53, no. 5-
1998. 6, pp. 1320–1330, 2011.
[2] L. S. Chen, Z. Y. Lu, and W. D. Wang, “The effect of delays on the [21] F. Y. Wei and C. Y. Li, “Permanence and globally asymp-
permanence for Lotka-Volterra systems,” Applied Mathematics totic stability of cooperative system incorporating harvesting,”
Letters, vol. 8, no. 4, pp. 71–73, 1995. Advances in Pure Mathematics, vol. 3, pp. 627–632, 2013.
[3] X.-z. He and K. Gopalsamy, “Persistence, attractivity, and delay [22] F. Chen, “On a nonlinear nonautonomous predator-prey model
in facultative mutualism,” Journal of Mathematical Analysis and with diffusion and distributed delay,” Journal of Computational
Applications, vol. 215, no. 1, pp. 154–173, 1997. and Applied Mathematics, vol. 180, no. 1, pp. 33–49, 2005.
[4] D. Mukherjee, “Permanence and global attractivity for faculta-
tive mutualism system with delay,” Mathematical Methods in the
Applied Sciences, vol. 26, no. 1, pp. 1–9, 2003.
[5] G. Lu, Z. Lu, and X. Lian, “Delay effect on the permanence for
Lotka-Volterra cooperative systems,” Nonlinear Analysis: Real
World Applications, vol. 11, no. 4, pp. 2810–2816, 2010.
[6] G. Lu and Z. Lu, “Permanence for two-species Lotka-Volterra
cooperative systems with delays,” Mathematical Biosciences and
Engineering, vol. 5, no. 3, pp. 477–484, 2008.
[7] Y. Nakata and Y. Muroya, “Permanence for nonautonomous
Lotka-Volterra cooperative systems with delays,” Nonlinear
Analysis: Real World Applications, vol. 11, no. 1, pp. 528–534,
2010.
[8] Z. Liu, R. Tan, Y. Chen, and L. Chen, “On the stable periodic
solutions of a delayed two-species model of facultative mutual-
ism,” Applied Mathematics and Computation, vol. 196, no. 1, pp.
105–117, 2008.
[9] S. Lu, “On the existence of positive periodic solutions to a Lotka
Volterra cooperative population model with multiple delays,”
Nonlinear Analysis: Theory, Methods & Applications, vol. 68, no.
6, pp. 1746–1753, 2008.
[10] D. H. Wright, “A simple, stable model of mutualism incorporat-
ing handling time,” The American Naturalist, vol. 134, no. 4, pp.
664–667, 1989.
[11] K. Gopalsamy, Stability and Oscillations in Delay Differential
Equations of Population Dynamics, vol. 74 of Mathematics and
Its Applications, Kluwer Academic, Boston, Mass, USA, 1992.
[12] Y. Li and G. Xu, “Positive periodic solutions for an integrodif-
ferential model of mutualism,” Applied Mathematics Letters, vol.
14, no. 5, pp. 525–530, 2001.
[13] F. Chen and M. You, “Permanence for an integrodifferential
model of mutualism,” Applied Mathematics and Computation,
vol. 186, no. 1, pp. 30–34, 2007.
[14] F. Chen, X. Liao, and Z. Huang, “The dynamic behavior of 𝑁-
species cooperation system with continuous time delays and
feedback controls,” Applied Mathematics and Computation, vol.
181, no. 2, pp. 803–815, 2006.
[15] J. Hale, Theory of Functional Differential Equations, Springer,
Heidelberg, Germany, 2nd edition, 1977.
[16] F. M. de Oca and M. Vivas, “Extinction in two dimensional
Lotka-Volterra system with infinite delay,” Nonlinear Analysis:
Real World Applications, vol. 7, no. 5, pp. 1042–1047, 2006.
[17] F. Chen, J. Yang, L. Chen, and X. Xie, “On a mutualism model
with feedback controls,” Applied Mathematics and Computation,
vol. 214, no. 2, pp. 581–587, 2009.
[18] L. Chen, L. Chen, and Z. Li, “Permanence of a delayed discrete
mutualism model with feedback controls,” Mathematical and
Computer Modelling, vol. 50, no. 7-8, pp. 1083–1089, 2009.
[19] L. Chen and X. Xie, “Permanence of an 𝑁-species cooperation
system with continuous time delays and feedback controls,”
Nonlinear Analysis: Real World Applications, vol. 12, no. 1, pp.
34–38, 2011.