See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/385056381

An analytic approach for nonlinear collisional fragmentation model arising in

bubble column

Article · October 2024

DOI: 10.1063/5.0231347

CITATIONS READS

0 25

3 authors:

Saddam Hussain Gourav Arora

Birla Institute of Technology and Science Pilani University of Petroleum & Energy Studies
12 PUBLICATIONS 52 CITATIONS 12 PUBLICATIONS 41 CITATIONS

SEE PROFILE SEE PROFILE

Rajesh Kumar
Birla Institute of Technology and Science Pilani
53 PUBLICATIONS 295 CITATIONS

SEE PROFILE

All content following this page was uploaded by Saddam Hussain on 31 October 2024.

The user has requested enhancement of the downloaded file.

 An Analytic Approach for Nonlinear
Collisional Fragmentation Model Arising in
Bubble Column
ै )∗1 , Gourav Arora (गौरव अरोड़ा)†2 and
Saddam Hussain (सद्दाम हुसन
Rajesh Kumar (राजेश कुमार)‡1

1
Department of Mathematics, Birla Institute of Technology and Science Pilani,
Rajasthan- 333031, India
2
Department of Mathematics, University of Petroleum and Energy Studies, Bidholi
Campus, Dehradun- 248007, India.

Abstract
The phenomenon of coagulation and breakage of particles plays a pivotal role
in diverse fields. It aids in tracking the development of aerosols and granules in the
pharmaceutical sector, coagulation or breakage of droplets in chemical engineer-
ing, understanding blood clotting mechanisms in biology, and facilitating cheese
production through the action of enzymes within the dairy industry. A significant
portion of research in this direction concentrates on coagulation or linear breakage
processes. In the case of linear case, bubble particles break down due to inherent
stresses or specific conditions of the breakage event. However, in many practical sit-
uations, particle division is primarily due to forces exerted during collisions between
particles, necessitating an approach that accounts for nonlinear collisional breakage.
Despite its critical role in a wide array of engineering and physical operations, the
study of this nonlinear fragmentation phenomenon has not been extensively pur-
sued. This article introduces an innovative semi-analytical method that leverages
the BLUES (beyond linear use of equation superposition) function to address the
nonlinear integro-partial differential model of collisional breakage population bal-
ance. This approach is versatile, allowing for the resolution of both linear/nonlinear
equations while sidestepping the complexities associated with discretization of do-
main. To assess the precision of this method, we conduct a thorough convergence
analysis. This process utilizes the principle of contractive mapping in the Banach
space, a globally recognized strategy for verifying convergence. We explore a vari-
ety of kernel parameters associated with collisional kernels, alongside breakage and
initial distribution functions, to derive novel iterative solutions. Comparing our
∗
Corresponding Author: p20200438@pilani.bits-pilani.ac.in
†
Email: gourav.arora@ddn.upes.ac.in
‡
Email: rajesh.kumar@pilani.bits-pilani.ac.in

1
2 Hussain, Arora and Kumar

findings with those obtained through the finite volume method (FVM) regarding
number density functions and their integral moments, we demonstrate the reliabil-
ity and accuracy of our approach. The consistency and correctness of our method
are further validated by depicting the errors between the exact and approximated
solutions in graphical and tabular formats.

Mathematics subject classifications (AMC). 34A12, 47J05, 45K05, 35R09, 65R20,

35Q70.
Keywords: Non-linear differential equation; Particulate process, Semi-analytical tech-
nique; Green function; Finite volume method; Convergence analysis.

1. Introduction
The particulate dynamics process significantly influences particle formation, wherein par-
ticles undergo transformations in their physical attributes. These transformations en-
compass properties such as size, volume, composition, enthalpy and porosity, manifesting
in forms like crystals, droplets, or bubbles. Several prominent particulate processes are
recognized within engineering and scientific domains, including nanotechnology, polymer-
ization, precipitation, crystallization, and emulsion. Among these processes, breakage, or
fragmentation, stands out as a pivotal mechanism wherein particles fracture into smaller
fragments. This breakage phenomenon finds application in specific events such as milling,
grinding, and protein filament division [1, 2].

Linear Breakage

External force ζ ϵ ζ −ϵ

Particle of volume ζ Fragments of volumes

in space ζ − ϵ and ϵ

Collision Induced Breakage

ϵ
ρ ζ −ϵ
ζ ρ ζ

ρ
Particles of volumes ζ and ρ Colliding particles
in space Fragments of volumes
ϵ, ρ and ζ − ϵ

Figure 1: Pictorial representation of linear and nonlinear breakage mechanisms

Analytical Approach for Nonlinear Collisional Fragmentation Model 3

Particle breakage is broadly categorized into linear and nonlinear (collisional-induced)

breakage. In linear breakage, matters split because of internal stresses, while collision-
induced breakage necessitates non-linear interactions between particles during collisions
[3, 4]. Figure 1 presents a graphical representation of both linear and nonlinear breakage
processes. In the linear case, a particle of volume ζ is subjected to an external force,
causing it to split into two particles with volumes ϵ and ζ − ϵ. In contrast, the nonlinear
process involves a collision between two particles of volumes ζ and ρ, resulting in the
formation of three particles with volumes ϵ, ρ, and ζ − ϵ. A notable distinction between
these two types lies in the resulting particle sizes. The linear breakage results in smaller
daughter particles, while breakage caused by collisions can transfer matter between the
colliding particles, potentially creating daughter particles that are larger than the original
ones [5]. For example, the chaotic movement in turbulent flow is critical to moving distinct
particles closer together [6]. Nonetheless, these particles will merge hinges on a variety
of intricate factors, including the energy lost during collisions, interactions influenced by
fluid dynamics, and the characteristics of the particle surfaces. Researchers have proposed
the collision kernels to forecast how often separate particles collide. By delving into these
advanced mechanisms, we enhance our comprehension of how particles behave and interact
within energetic settings. As such, collisional breakage encompasses a broader context
and significantly influences various engineering and natural science processes, including
astronomy, protein filament division, chemical engineering, pharmaceutical research, and
dispersed systems like raindrop development and fluidized beds [7, 8, 9, 10, 11, 12, 13]. It
can enhance the process by boosting the dissolution rate and permeability of the system,
or by limiting the maximum size of the granules [8, 9].

Despite its pivotal role and extensive contributions across a variety of real-world ap-
plications, the mathematical literature on collisional breakage equations (CBE) remains
relatively limited within the scientific community’s discourse. The focus of research in
this direction largely centers on coagulation and linear breakage processes [14, 15, 16, 17].
This research delves into the dynamics of nonlinear breakage, focusing specifically on the
phenomenon of pure binary collisional fragmentation. This particular subset of nonlinear
dynamics is defined by the inelastic and immediate collisions between particles, which
result in their breakage. The insights gained from these processes are pivotal for compre-
hending the collective behavior of raindrop distributions and the operational principles
of communication networks. Moreover, they play a critical role in various engineering
scenarios, including the operation of fluidized beds and the milling industry. Recent re-
search on bubble columns has mostly relied on straightforward linear models to describe
their behavior [17, 18, 19]. Nonetheless, this approach frequently misses the significant
chance of bubbles forming from collisions [20, 21], necessitating the use of more intri-
cate mathematical models. As a result, investigating nonlinear models that account for
collision-induced fragmentation is crucial for a more accurate depiction of bubble column
dynamics. The research aims to shed light on the complex nature of collisional fragmenta-
tion, contrasting its features with linear breakage and examining its broader implications
in the fields of natural sciences and engineering. It involves an in-depth examination of
the collisional breakage equation, underscoring its limited representation in mathematical
texts despite its significant impact on the enhancement of process efficiencies.
4 Hussain, Arora and Kumar

1.1 Collisional breakage model

Cheng and Redner [22] devised a mathematical model relying on the population balance
equation (PBE)’s framework to describe the complex dynamics of particle interactions
during collisions. The distribution of daughter particles is obtained by integrating the
cumulative number of breakage events for a parent particle with a given volume. This
leads to the formulation of the model as an integro-partial differential equation. Let f (t, ϵ)
represent the particle density distribution function, where ϵ is the particle volume, and
t is the time. Then temporal change in particle number density f (t, ϵ) ≥ 0 in a system
undergoing collisional breakage process is accurately marked out by the following equation
Z ∞Z ∞ Z ∞
∂f (t, ϵ)
= K(ρ, ζ)B(ϵ, ρ; ζ)f (t, ρ)f (t, ζ)dρdζ − K(ϵ, ρ)f (t, ϵ)f (t, ρ)dρ, (1)
∂t 0 ϵ 0

together with the initial condition

f (0, ϵ) = f in (ϵ) ≥ 0. (2)

Here, the kernel function K(ρ, ζ) quantifies the collision rate between two particles with
volumes ρ and ζ, leading to the fragmentation of at least one of the colliding particles.
Conversely, the breakage distribution function B(ϵ, ρ; ζ) specifies how particles with vol-
ume ϵ are generated from collisions involving particles with volumes ρ and ζ. Typically,
the collision kernel K(ρ, ζ) is assumed to be symmetric and non-negative with respect to
its parameters. Meanwhile, the breakage kernel B(ϵ, ρ; ζ) must fulfill certain conditions
outlined below.

• B(ϵ, ρ; ζ) = B(ϵ, ζ; ρ); Symmetric property.

• B(ϵ, ρ; ζ) ̸= 0, for ϵ ∈ (0, ρ) and B(ϵ, ζ; ρ) = 0, for ϵ > ρ.

Rρ
• 0 ϵB(ϵ, ζ; ρ)dϵ = ρ, ∀(ϵ, ρ) ∈ (0, ∞) × (0, ∞); Volume conservation property.
Rρ
• 0 B(ϵ, ζ; ρ)dϵ ≤ R < ∞ for all ρ, ζ > 0; Number of daughter particles after break-
age.

The left-hand side of equation (1) shows how the density distribution function changes
over time. The first term on the right-hand side, called the birth term, indicates the rate
at which particles of size ϵ form when a larger particle ρ breaks apart after colliding with
ζ. The next term, known as the death term, represents the rate at which particles of
volume ϵ disappear due to interactions with other particles.
Beyond the density distribution function, integral moments are also pivotal. The k th
moment of a number density function f (t, ϵ) is defined as
Z ∞
Mk (t) = ϵk f (t, ϵ)dϵ. (3)
0

Some initial moments are particularly noteworthy. The zeroth moment, when k = 0,
signifies the overall particle count within the system. The first moment, marked by
k = 1, quantifies the total mass of the system. Meanwhile, the second moment, with
k = 2, reflects the entire amount of energy the system expends throughout the process.
Analytical Approach for Nonlinear Collisional Fragmentation Model 5

1.2 State-of-the-art and its advancement

The model being discussed is an integro-partial differential equation that involves several
free parameters, such as the collisional kernel, number density function and breakage dis-
tribution function. Solving this model analytically is challenging because of its non-linear
characteristics and the complexity of the physical parameters involved. The mathematical
literature has extensively studied linear fragmentation, as evidenced by [23, 24, 25, 26]
and other references. The authors in [23] provided an analytical solution to the linear
breakage equation for certain kernel functions. The findings in [24, 25] discuss simi-
larity solutions, while [26] focuses on the model’s numerical solution using fixed pivot
and discrete element techniques. Extensive studies have focused on the well-posedness
of coagulation equation coupled with linear breakage. In the realm of numerical anal-
ysis related to aggregation and linear fragmentation equations, the literature reveals a
diverse range of methodologies, including the finite volume method [27, 28, 29], method
of moments [30], finite element scheme [31], and Monte Carlo method [32], among others.
However, literature on the collisional breakage mostly delves into the theoretical existence
and uniqueness of the solutions [33, 5, 34, 35, 36]. In [5, 34], the authors examined the
uniqueness, existence, and density conservation of the solutions. The existence of weak
solutions to the CBE is shown in [33] for a large class of unbounded collision kernels given
by K(ϵ, ρ) = ϵα ρβ + ϵβ ρα with α ≤ β ≤ 1. In their research on the existence of a solution
for the discrete collisional breakage with coagulation equation, Laurençot and Wrzosek
[5] imposed specific constraints on the kernels
K(ϵ, ρ) ≤ (ϵρ)γ , γ ∈ [0, 1), and B(ϵ, ρ; ζ) ≤ A < ∞, for 1 ≤ ϵ < ρ.
The investigation of gelation and its behavior over extended periods is another key area
of focus. In [35], the authors examined the global classical solution for coagulation and
collisional breakage equations with a collision kernel that supports infinite growth at large
volumes. Furthermore, [36] explores the analysis for mass-conserving solutions where the
collision kernel grows at most linearly as it approaches infinity. Additionally, numerical
literature on collisional breakage involves only finite volume methods [37, 38].
Numerical techniques are used to estimate the solutions at specific points within the
discrete domain. However, finding a suitable discretization can occasionally be more
challenging than solving the problem itself. For example, linear B-splines necessitate
numerous splines to obtain an adequate approximation, which complicates the general-
ization of these methods to spaces with more than three dimensions. Rounding error,
discretization error etc. frequently render numerical techniques unreliable. For certain
problems, a specialized numerical algorithm is often necessary to yield meaningful out-
comes. Therefore, developing more robust and versatile methods for solving differential
equations remains a vital and ongoing research area. Researchers have introduced semi-
analytical techniques such as the Adomian decomposition method (ADM) [39], Laplace
optimized decomposition method (LODM) [40], homotopy perturbation method (HPM)
[41, 42], variational iteration method (VIM) [43], HAM [44], and others [45, 46, 47]. These
methods address both linear and non-linear initial and boundary value problems with-
out relying on nonphysical assumptions such as linearization, discretization, initial guess
estimation, or a set of basis functions.
Recently, Indekeu et. al. [48] introduced a computational method ‘beyond linear use of
equation superposition’ (BLUES) for solving any kind of differential equations. A BLUES
6 Hussain, Arora and Kumar

function is identified as a solution to a nonlinear differential equation (DE) featuring a

delta source, which concurrently serves as a Green’s function for a corresponding linear
DE. For any given source, the BLUES function enables the construction of an exact solu-
tion for the nonlinear DE with a varied, yet associated source. Alternatively, the BLUES
function can facilitate the development of an approximate, piecewise analytical solution
for the nonlinear DE with any source. In this latter application, there is no requirement
to know the associated linear DE. The authors demonstrated the effectiveness of their
approach through various instances, employing either step sources or exponential corner
sources, within the framework of nonlinear reaction-diffusion-convection equations. Fur-
ther, [49] extends the BLUES function method for the partial differential equations. A
comparison is made with ADM, VIM and the VIM with Green function (GVIM). The
BLUES proved to be a viable alternative among various methods. The advantages it pro-
vides arise from the flexibility in selectively determining the linear portion, equipped with
an associated Green function, and the remainder that encompasses the nonlinear segment
of the given differential operator. Moreover, the study [50] delves into the interconnected
nonlinear ODEs found within the SIRS epidemiological model considering vaccination,
utilizing the BLUES function approach. This method facilitated the attainment of accu-
rate analytical approximations for the dynamics of the susceptible and infected population
percentages over time.
This motivates us to use the BLUES function method for solving CBEs considering
various kernel parameters and initial conditions. By leveraging this algorithm, we aim
to provide a rigorous yet computationally efficient framework for simulating collisional
breakage processes. Our research seeks to contribute to the ongoing efforts in developing
reliable solution techniques for population balance modeling, ultimately facilitating the
design and optimization of industrial processes involving particle breakage. The conver-
gence analysis and error estimations are also included to support the reliability of the
method.
The remaining of the paper is organised as follows. Section 2 describes the preliminar-
ies for the BLUES function method, including the development of the method for CBE
in subsection 2.1. The theoretical convergence analysis and error bounds are provided in
subsection 2.2. The numerical results and analysis in Section 3 demonstrate the method’s
effectiveness and precision. Section 4 summarizes the study’s findings with observations
and remarks. Detailed computations for accurate moments are included in the appendix.

2. BLUES function method

To grasp the concept of the BLUES method, let’s start by examining a linear equation in
the following form

Lf (t, ϵ) = 0 for t > 0, (4)

with the initial condition f (0, ϵ) = u(ϵ) and L as a linear operator acting on f (t, ϵ). As
the equation (4) is linear in nature, the solution f (t, ϵ) can be written as the convolution
G ∗ u, where G is the Green function which satisfies

LG(t, ϵ) = 0, for t > 0, (5)

Analytical Approach for Nonlinear Collisional Fragmentation Model 7

with the initial condition lim G(t, ϵ) = δ(ϵ), having δ(ϵ) as the dirac delta function. Thus,
t→0
the solution to the linear problem (4) is the convolution of the form
Z
f (t, ϵ) = G(t, ϵ − ϵ′ )u(ϵ′ )dϵ′ . (6)
R

Since our model (1) has first order derivative with respect to t, we restrict the nonlinear
equation to be the first order in time derivative, i.e., Lf = ft + L̃f with ft := ∂f /∂t
and L̃ represents the rest of the linear operator. To simplify the earlier partial differential
equation (PDE), we can introduce the initial condition, u(ϵ), by implementing a Dirac-
delta function in the time domain. By doing so, we obtain a dual-variable convolution
f (t, ϵ) = G ∗ uδ, which solves the following rearranged inhomogeneous PDE
Z tZ
Lf (t, ϵ) = L G(ϵ − ϵ′ , t − t′ )u(ϵ′ )δ(t′ )dϵ′ dt′ = u(ϵ)δ(t). (7)
0 R

It is important to mention that the above is equivalent to the original linear PDE (4), see
[51], Section 2 for more details.
In this approach to the PDE, an intrinsic source term effectively incorporates the initial
condition, facilitating the expansion of the BLUES iteration method to nonlinear PDEs.
This allows for the introduction of nonlinearity into the PDE with relative ease while
preserving the straightforward structure of the time-dependent component, denoted as

N f (t, ϵ) = ft + Ñ f = 0 for t > 0, (8)

with initial condition f (0, ϵ) = u(ϵ). According to the BLUES function method, solution
of the non-linear problem (8) is obtained by employing inhomogeneous PDE in the form
of two-variable convolution, i.e., f (t, ϵ) = G ∗ ϕ, so that
Z tZ
N f (t, ϵ) = N G(ϵ − ϵ′ , t − t′ )ϕ(t′ , ϵ′ )dϵ′ dt′ = u(ϵ)δ(t). (9)
0 R

The function G(t, ϵ) is known as the BLUES function which is taken to be Green function
for a specific, conveniently selected linear operator L associated with N . The challenge
is to calculate the new associated source ϕ(t, ϵ) based on the premise that G ∗ uδ solves
the linear PDE (7) with initial condition u(ϵ) and source term uδ. Now, for calculation,
following [51], a residual function R := L − N is defined and uses the implicit identity

N (G ∗ ϕ)(t, ϵ) = ϕ(t, ϵ) − R(G ∗ ϕ)(t, ϵ) = u(ϵ)δ(t), (10)

that follows from the property of Green function with respect to the chosen linear PDE.
In order to obtain the approximate solution of the nonlinear PDE (8) as a power series
in terms of the residual R, we can reformulate the above equation as

ϕ(t, ϵ) = u(ϵ)δ(t) + R(G ∗ ϕ)(t, ϵ). (11)

Therefore, iterations for the source function ϕ(t, ϵ) are obtained as

ϕn (t, ϵ) =u(ϵ)δ(t) + χn R(G ∗ ϕn−1 )(t, ϵ) for n ≥ 0,

8 Hussain, Arora and Kumar

where χn is the characteristic function such that

(
1 n≥1
χn =
0 n = 0.

Hence, the analytical approximation to the solution of (8) is

fn (t, ϵ) = (G ∗ ϕn )(t, ϵ) = f0 (t, ϵ) + (G ∗ Rfn−1 )(t, ϵ), (12)

Rt
with the initial guess f0 (t, ϵ) = 0
G(t − s)ψ(s, ϵ)ds where ψ(t, ϵ) = u(ϵ)δ(t).

2.1 Development of BLUES function method for collisional break-

age model
On comparing the model nonlinear problem (8) with the collisional breakage model (1),
we come across the following operators
Z ∞Z ∞ Z ∞
∂f (t, ϵ)
N f (t, ϵ) = − K(ρ, ζ)B(ϵ, ρ; ζ)f (t, ρ)f (t, ζ)dρdζ + K(ϵ, ρ)f (t, ϵ)f (t, ρ)dρ
∂t 0 ϵ 0
(13)

and the linear operator L = ft = ∂f∂t(t,ϵ)

. Therefore, the associated Green function G will
be Heaviside theta function given as G(t) = Θ(t). and the residual R = L − N is defined
as
Z ∞Z ∞ Z ∞
R := K(ρ, ζ)B(ϵ, ρ; ζ)f (t, ρ)f (t, ζ)dρdζ − K(ϵ, ρ)f (t, ϵ)f (t, ρ)dρ. (14)
0 ϵ 0

Now, based on the above defined methodology in Section 2, the zeroth order approximant
is given by the following convolution
Z t
f0 (t, ϵ) = Θ(t − s)f in (ϵ)δ(s)ds, (15)
0

and the higher order approximations are given as

Z t Z ∞Z ∞
fn (t, ϵ) = f0 (t, ϵ)+ Θ(t − s) K(ρ, ζ)B(ϵ, ρ; ζ)fn−1 (s, ρ)fn−1 (s, ζ)dρdζ
0
Z ∞ 0 ϵ

− K(ϵ, ρ)fn−1 (s, ϵ)fn−1 (s, ρ)dρ ds. (16)
0

2.2 Convergence analysis

This section gives a detailed description of the convergence analysis and error bounds for
the BLUES method. To analyze the convergence criterion of the proposed scheme (12),
fixed point theory is applied within a suitably constructed Banach space. Let us assume
the Banach space

Z :={f : [0, T ] × [0, ∞) → [0, ∞) | f, (∂f /∂t) are continuous and

Analytical Approach for Nonlinear Collisional Fragmentation Model 9

Z ∞
∥f ∥α = sup (λv)α |f (ζ, s)|dv < ∞, with λ, α > 0} .
s∈[0,t0 ] 0

Equation (1) can be expressed in an operator form as

f (t, ϵ) = A(f ) (17)

where, A(f ) is defined by

Z tZ ∞ Z ∞
A(f ) =f (ϵ) +
in
K(ρ, ζ)B(ϵ, ρ; ζ)f (s, ρ)f (s, ζ)dρdζ
Z ∞ 0 0 ϵ

− K(ϵ, ρ)f (s, ϵ)f (s, ρ)dρ ds. (18)
0

To illustrate the contractive property of operator A, the equivalent form of (18) is con-
sidered as
Z ∞Z ∞
∂
[f (t, ϵ) exp[T (t, ϵ, f )]] = exp[T (t, ϵ, f )] B(ϵ, ρ; ζ)K(ρ, ζ)f (t, ρ)f (t, ϵ)dρdϵ.
∂t 0 ϵ
(19)

where Z tZ ∞
T (t, ϵ, f ) = K(ϵ, ρ)f (s, ρ)dρds.
0 0

Consequently, the equivalent operator à is derived as

Ã(f ) =f in (ϵ) exp[−T (t, ϵ, f )]

Z t Z ∞ Z ∞
+ exp[T (s, ϵ, f ) − T (t, ϵ, f )] B(ϵ, ρ; ζ)K(ρ, ζ)f (s, ρ)f (s, ζ)dρdζds.
0 0 ϵ
(20)

Since A is equivalent to Ã, it suffices to demonstrate the existence of a solution for (20),
which in turn implies the existence of a solution for (18).

Theorem 2.1. The operator Ã, defined in (20) is contractive on Z, that is, Ãf − Ãf ∗ ≤
α
γ∥f − f ∗ ∥α , for all f, f ∗ ∈ Z if the following assumptions hold

Rρ
1. B(ϵ, ρ; ζ) = qϵk−1 /ρk , where k = 1, 2, . . . and q > 0 such that 0 ϵB(ϵ, ρ; ζ)dϵ = ρ,

2. K(ϵ, ρ) ≤ (ϵρ)α , α > 0 for all ϵ, ρ ∈ (0, ∞),

3. λ is so chosen that ρ < λ and,

4. γ := t0 exp (2t0 L) [∥f0 ∥α + (4qL/α) (1 + t0 L)] < 1 (for suitably chosen t0 ), where
L = ∥f0 ∥α + (q/α)T ∥f0 ∥21 .

Proof. For any f, f ∗ ∈ Z, we have

Ãf − Ãf ∗ =f in (ϵ) [exp[−T (t, ϵ, f )] − exp [−T (t, ϵ, f ∗ )]]

10 Hussain, Arora and Kumar

Z t Z ∞ Z ∞
+ exp[T (s, ϵ, f ) − T (t, ϵ, f )] K(ρ, ζ)B(ϵ, ρ; ζ)f (s, ρ)f (s, ζ)dρdζds
Z 0
t
0
Z ϵ
∞ Z ∞
∗ ∗
− exp [T (s, ϵ, f ) − T (t, ϵ, f )] K(ρ, ζ)B(ϵ, ρ; ζ)f ∗ (s, ρ)f ∗ (s, ζ)dρdζds
0
Z t Z ∞ Z 0∞ ϵ
=f in (ϵ)R̃(t, 0, ϵ) + R̃(t, s, ϵ) K(ρ, ζ)B(ϵ, ρ; ζ)f (s, ρ)f (s, ζ)dρdζds
Z t 0 0
Z ∞Z ∞
ϵ

∗ ∗ ∗
+ exp [T (s, ϵ, f ) − T (t, ϵ, f )] K(ρ, ζ)B(ϵ, ρ; ζ)f (s, ρ) f (s, ζ)
0
 Z ∞Z ∞ 0 ϵ

∗ ∗
− f (s, ζ) dρdϵ + K(ρ, ζ)B(ϵ, ρ; ζ)f (ζ, s) [f (s, ρ) − f (s, ρ)] dρdζ ds,
0 ϵ
(21)

where R̃(t, s, ϵ) = exp[T (s, ϵf )−T (t, ϵ, f )]−exp [T (s, ϵ, f ∗ ) − T (t, ϵ, f ∗ )]. It can be easily
shown that
 Z tZ ∞ 
|R̃(t, s, ϵ)| ≤ exp − K(ϵ, ρ)f (ρ, τ )dρdτ (t − s) ∥f − f ∗ ∥α
∗
s 0
≤ (t − s) exp{(t − s)A} ∥f − f ∗ ∥α ≤ J ∥f − f ∗ ∥α , (22)

where J = t exp{tA} and A = max {∥f ∥α , ∥f ∗ ∥α }. To demonstrate that the operator

à is contractive, we define the set S = {f ∈ Z : ∥f ∥α ≤ 2L}. It is easy to show that Ã
maps S into itself. For f, f ∗ ∈ S, we have A ≤ 2L. Applying the norm on both sides of
(21) gives
Z t
∗ ∗ q ∗
Ãf − Ãf ≤J ∥f0 ∥α ∥f − f ∥α + J ∥f − f ∥α ∥f ∥2α ds
α α
Z 0
q t
+ exp{tA} [(∥f ∥α + ∥f ∗ ∥α ) ∥f − f ∗ ∥α ] ds
α
h 0 q q i
=J ∥f0 ∥α + t∥f ∥α + (∥f ∥α + ∥f ∥α ) ∥f − f ∗ ∥α
2 ∗
α α
:=γ ∥f − f ∗ ∥α

if γ = t0 exp (2t0 L) [∥f0 ∥α + (4qL/α) (1 + t0 L)] < 1, then the operator à is contractive.

Theorem 2.2. Assuming that the operator à meets all the conditions specified in Theorem
2.1. If ∥f1 ∥α < ∞ and γ = t0 exp (2t0 L) [∥η0 ∥α + (4qL/α) (1 + t0 L)] < 1, then the
sequence fj defined by (12) converges to the exact solution f (t, ϵ) with the error bound

γk
∥f − fk ∥α ≤ ∥f1 ∥α .
(1 − γ)
Proof. Let us define C(fk−1 , s, ϵ) as
Z ∞Z ∞
C (fk−1 , s, ϵ) := K(ρ, ζ)B(ϵ, ρ; ζ)fk−1 (s, ρ)fk−1 (s, ζ)dρdζ
0
Z ∞ϵ
− K(ϵ, ζ)fk−1 (s, ϵ)fk−1 (s, ρ)dρ. (23)
0
Analytical Approach for Nonlinear Collisional Fragmentation Model 11

Therefore, we have
Z t
fk (t, ϵ) = f0 (t, ϵ) + Θ(t − s)C(fk−1 , s, ϵ)ds.
0

Now, as t > s, the above equation is reduced to

Z t Z ∞ Z ∞
fk (t, ϵ) = f0 (t, ϵ) + K(ρ, ζ)B(ϵ, ρ; ζ)fk−1 (s, ρ)fk−1 (s, ζ)dρdζ
0
Z ∞
0 ϵ

− K(ϵ, ζ)fk−1 (s, ϵ)fk−1 (s, ρ)dρ ds = A(fk−1 ).
0

Using the contraction property of the nonlinear operator A,

∥fj+1 − fj ∥α ≤ γ ∥fj − fj−1 ∥α ≤ γ 2 ∥fj−1 − fj−2 ∥α ≤ · · · ≤ γ j ∥f1 − f0 ∥α .

Now, for all j, k ∈ N with j > k, we have

∥fj − fk ∥α ≤ ∥fj − fj−1 ∥α + ∥fj−1 − fj−2 ∥α + · · · + ∥fk+1 − fk ∥α

 
≤ γ j−1 + γ j−2 + · · · + γ k ∥f1 − f0 ∥α
 
  1 − γ j−k
= γ 1 + γ + γ + ··· + γ
k 2 j−k−1
∥f1 − f0 ∥α = γ k
∥f1 ∥α .
1−γ


Since 0 < γ < 1, 1 − γ j−k < 1 and ∥f1 ∥ < ∞, it follows that

γk
∥fj − fk ∥α ≤ ∥f1 ∥α
1−γ
which converges to 0 as k → ∞. Therefore, there exists a function f (t, ϵ) such that

lim fk (t, ϵ) = f (t, ϵ).

k→∞

3. Numerical Results
In this section, we apply the BLUES method to solve the nonlinear collisional fragmen-
tation model (1) with varying combinations of collision kernels, breakage distribution,
and initial conditions. Given that analytical solutions for constant collision kernels have
been thoroughly investigated in the literature alongside various breakage distribution
functions, our goal is to address the PBE (1) for more intricate collision kernels such
as polymerization and product kernels. To highlight the originality of our approach, we
draw comparisons between our solutions and existing analytical ones, whenever they are
available, using tables and error graphs. For cases where precise answers are elusive, we
compared the outcomes derived from the BLUES method with those obtained through
the FVM discussed in [38]. The use of FVM is considered appropriate for this comparison
due to its capability in maintaining both the zeroth and first-order moments concurrently.
12 Hussain, Arora and Kumar

The fVM is the sole computational approach that has been validated and tested for the
collisional fragmentation equation [38]. It is noteworthy that, for one example, we manage
to derive the nth-order general term using the BLUES method, enabling us to articulate
the solution in a closed form. In certain instances, while we could ascertain the structure
of the solution in a complete form, the integration of more intricate functions precluded
the attainment of a purely complete solution. To navigate this challenge, we determine the
initial terms of the iterative solution, thereby facilitating the creation of semi-analytical
solutions.
Example 3.1. Product coagulation kernel, binary breakage function and ex-
ponential initial distribution
In the first example, we explore the collisional kernel as the product kernel, expressed
as K(ϵ, ρ) = ϵρ, alongside binary fragmentation, defined by B(ϵ|ρ, ζ) = ρ2 , and an exponen-
tial initial condition f (0, ϵ) = e−ϵ . This particular kernel is categorized under gelling ker-
nels and is applied in understanding the rennet-induced coagulation phenomenon within
the dairy industry [52]. It also elucidates the tendency for larger aerosol particles to en-
gage in collisions and subsequently disintegrate into smaller particles in the context of
aerosol science [53]. The analytical solution for the number density distribution is studied
in [54] and provided as

f (t, ϵ) = (1 + t)2 e−ϵ(1+t) .

Using the iterative scheme (12), the solution obtained via BLUES function method are

f0 (t, ϵ) =e−ϵ , f1 (t, ϵ) = e−ϵ (1 − tϵ + 2t),

 
−ϵ t 2 ϵ2
f2 (t, ϵ) =e 1 − tϵ + + 2t − 2t ϵ + t ,
2 2
2
   
−ϵ t 2 ϵ2 t 3 ϵ3 t 2 ϵ2
f3 (t, ϵ) =e 1 − tϵ + − + 2t 1 − tϵ + + t (1 − tϵ) .
2
2 3! 2
Following the similar pattern, we have the general term as
  
−ϵ t 2 ϵ2 (−1)n+1 tn ϵn t 2 ϵ2 (−1)n tn−1 ϵn−1
fn (t, ϵ) =e 1 − tϵ + + ··· + + 2t 1 − tϵ + + ··· +
2 n! 2 (n − 1)!
 2 2 n−1 n−2 n−2

tϵ (−1) t ϵ
+ t2 1 − tϵ + + ··· + ,
2 (n − 2)!
which, in the case of limn→∞ gives


lim fn (t, ϵ) = e−ϵ e−tϵ + 2te−tϵ + t2 e−tϵ
n→∞

= (1 + t)2 e−ϵ(1+t) .

Hence, the BLUES function method provides exactly the analytical solution for this ex-
ample. To evaluate its effectiveness, the 10th iterative approximation for the number
density function denoted as f10 (t, ϵ), is compared with the exact solution f (t, ϵ) and FVM
solution in Figure 2(a) at various time intervals. The figure demonstrates that the BLUES
solution predicts the density distribution function with remarkable precision, coinciding
Analytical Approach for Nonlinear Collisional Fragmentation Model 13

precisely with the analytical solution after just ten iterations. Initially, at time t = 0.37,
the system contains a smaller number of bubbles, which are also small in size. However,
as time progresses to t = 0.63, there is an observable increase in the density of these
small-sized bubbles. This increase is attributed to collisions among larger particles, lead-
ing to their fragmentation into smaller ones. Further, by the time the system reaches
t = 1, it is predominantly composed of these smaller-sized bubbles. Consequently, this
demonstrates that the BLUES method can accurately predict the behavior of bubbles
within the system. Table 1 provides the error between the exact and BLUES approxi-
mated solutions, highlighting how discrepancies progress over time. This progression of
error is methodically documented for 2, 3, 4, and 5 numbers of approximation across time
intervals from 0 to 1 in steps of 0.2. The findings illustrate that the error margin notably
escalates as time advances. This phenomenon underscores the growing deviation between
the precise solutions and their approximations over time. However, increasing the number
of terms in the approximation can significantly reduce this error. For instance, the error
for approximations with five terms is consistently lower at each time step compared to
approximations with two terms. This evidences the efficacy of increasing approximation
terms in mitigating error, highlighting its potential as a strategic approach to enhance
approximation accuracy over time.
Moreover, Figures 2(b), 2(c) and 2(d) illustrate the zeroth, first and second moments
of the density functions, respectively. The increasing trend of the zeroth moment in Figure
2(b) indicates that the total number of bubble particles in the system rises over time, which
is attributed to particle splitting following successful collisions. Despite this increase, the
total mass of the bubble particles remains conserved, as depicted in Figure 2(c). The
downward trend of the second moment in Figure 2(d) suggests that the system initially
absorbs a greater amount of energy, which decreases as time progresses. This happens
because the system starts with larger particles that require more energy to disintegrate.
As smaller particles are produced through fragmentation, less energy is needed to break
them. Notably, the BLUES method accurately predicts the zeroth and first moments,
even better than FVM. However, after a certain time, the second moment of BLUES
begins to show inconsistency, which can be compensated for by adding more iterations to
the BLUES solution.
Example 3.2. Product coagulation kernel, binary breakage function and Gaus-
sian initial distribution
This example considers product coagulation kernel K(ϵ, ρ) = ϵρ along with  the  binary
ϵ2
√
fragmentation β(ϵ|ρ, ζ) = ρ and Gaussian initial distribution f (0, ϵ) = exp − 2 / 2π.
2

Implementing the BLUES method as defined in Section 2, the iterative solutions are as
follows
 
2 2   t erfc √ϵ
− ϵ2 − ϵ2
e e tϵ 2
f0 (t, ϵ) = √ , f1 (t, ϵ) = √ 1− √ + √ ,
2π 2π 2π 2π
ϵ2     
e− 2 tϵ t 2 ϵ2 t2 ϵ 1 3tϵ
f2 (t, ϵ) = √ 1− √ + √ + + t erfc √ √ − ,
2π 2π 2!( 2π)2 2π 2 2π 4π
ϵ2   
e− 2 tϵ t 2 ϵ2 t 3 ϵ3 t2 tϵ
f3 (t, ϵ) = √ 1− √ + √ − √ + 1− √
2π 2π 2!( 2π)2 3!( 2π)3 2π 2π
14 Hussain, Arora and Kumar

4.5 1

2
, FVM
4 0.95
2
, BLUES

Normalized first moments

number density function

3.5 0.9 2
, Exact

3
0.85
2.5
0.8
2
0.75
1.5
0.7
1
0.65
0.5

0 0.6

0.55
10-2 10-1 100 101 102 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
volume time

(a) Number density (b) Zeroth moment

2 1

1
, FVM 2
, FVM
1.8 0.95
, BLUES , BLUES

Normalized second moments

1 2
Normalized first moments

1.6 , Exact , Exact

1 0.9 2

1.4
0.85
1.2
0.8
1
0.75
0.8
0.7
0.6
0.65
0.4

0.2 0.6

0 0.55
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
time time

(c) First moment (d) Second moment

Figure 2: Particle density distribution and moments functions for Example 3.1

Error between exact and BLUES method approximated solutions

t
n=2 n=3 n=4 n=5

0.2 2.69561 ×10−3 3.71262 ×10−4 5.46632 ×10−5 8.34915 ×10−6

0.4 1.97518 × 10−2 5.37117 × 10−3 1.57093 × 10−3 4.77082×10−4

0.6 6.13831 × 10−2 2.47539 × 10−2 1.07983 × 10−2 4.89821 × 10−3

0.8 1.34663 × 10−1 7.16763×10−2 4.14893×10−2 2.50124 × 10−2

1.0 2.44562 ×10−1 1.61257 ×10−1 1.162 ×10−1 8.73452 ×10−2

Table 1: Error distribution for Example 3.1 at different values of n.

  
ϵ 1 3tϵ t 2 ϵ2
+ t erfc √ √ − + √ 3 ,
2 2π 4π 2 2π 2
Analytical Approach for Nonlinear Collisional Fragmentation Model 15

ϵ2   
e− 2 tϵ t 2 ϵ2 t 3 ϵ3 t 4 ϵ4 t2 tϵ t 2 ϵ2
f4 (t, ϵ) = √ 1− √ + √ − √ + √ + 1− √ + √
2π 2π 2!( 2π)2 3!( 2π)3 4!( 2π)4 2π 2π 2!( 2π)2
  
ϵ 1 3tϵ t 2 ϵ2 5t3 ϵ3
+ t erfc √ √ − + √ 3 − ,
2 2π 4π 2 2π 2 48π 2

where, erfc(ϵ) is the complementary error function given by

Z ∞
1
erfc(ϵ) = 2 √ exp(−z 2 )dz.
ϵ π

By observing the above solution approximations, it can be seen that solutions will converge
to the function of the form
2
− ϵ2 − √tϵ     
e 2π t2 ϵ 1 3tϵ
f (t, ϵ) = √ 1+ + t erfc √ √ − + ··· .
2π 2π 2 2π 4π

Since the analytical solution is not available in the literature for this case, the 5-term

4.5 2.8

0
, FVM
4 2.6
, BLUES
Normalized zeroth moments

0
number density function

3.5
2.4
3
2.2
2.5
2
2
1.8
1.5
1.6
1

0.5 1.4

0 1.2

-0.5 1
10-3 10-2 10-1 100 101 102 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
volume time

(a) Number density (b) Zeroth moment

2 1

1
, FVM 2
, FVM
1.8 0.95
, BLUES , BLUES
Normalized second moments

1 2
Normalized first moments

1.6
0.9
1.4
0.85
1.2
0.8
1
0.75
0.8
0.7
0.6
0.65
0.4

0.2 0.6

0 0.55
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 0.5 1 1.5 2 2.5 3
time time

(c) First moment (d) Second moment

Figure 3: Particles density distribution and moments functions for Example 3.2

BLUES method solution, i.e., f5 (t, ϵ) is plotted against the FVM approximated number
16 Hussain, Arora and Kumar

density function in Figure 3(a) to demonstrate the novelty of the method. Results ob-
tained via both the schemes are overlapping at various times, indicating the accuracy
of the algorithms. Further, the zeroth moment is given in Figure 3(b) which shows the
linear growth for the chosen kernel parameters and results are almost overlapping to each
other. The first moment in Figure 3(c) proves the mass conservation of the system. Also,
the second moment obtained using BLUES matches well with the FVM’s outcomes, see
Figure 3(d).

Example 3.3. Polymerization coagulation kernel, binary breakage function

and exponential initial distribution
√ √
Let us take into account the polymerization coagulation kernel K(ϵ, ρ) = 3 ϵ + a 3 ρ + a,
binary brakage B(ϵ|ρ, ζ) = ρ2 and exponential initial condition f (0, ϵ) = exp(−ϵ). This
aggregation kernel is crucial in systems like colloidal suspensions and ferrofluids where the
particle has polarization properties, such as electrostatic or magnetic charges [53]. For
the sake of simplicity, the value of a is assumed to be 0, although it should be noted that
any value of a may be considered. These kernel parameters lead to the following BLUES
iterative solution components
f0 (t, ϵ) =e−ϵ ,
   
−ϵ 1 −ϵ 3
√
f1 (t, ϵ) =e + t 1.78Γ , ϵ − 0.89e ϵ ,
3
  
−ϵ −ϵ 1 √
√

f2 (t, ϵ) =e + te ϵ
eΓ , ϵ t 0.40 − 5.58 3 ϵ − 1.67t2 3 ϵ + 1.78
3
 
√ √ √
2
+ ϵ 0.11t ϵ + 0.39t ϵ − 0.20t − 0.89 + t(1.19t + 3.98)e Γ
3 2 3 3 ϵ
,ϵ ,
3
  
−ϵ 2 √
√
√

f3 (t, ϵ) =te ϵ
te Γ ,ϵ t4 0.003 − 0.23 3 ϵ + t3 0.07 − 1.40 3 ϵ + t2 0.49 − 4.68 3 ϵ
3
  
√
√ 1 √

+ t 1.79 − 8.30 ϵ − 0.014t ϵ + 3.986 + e Γ
3 5 3 ϵ
,ϵ t 0.402 − 5.580 3 ϵ
3

√
√

+ 0.011t ϵ + t 0.190ϵ − 0.005 ϵ + t 1.122ϵ − 0.103 ϵ + t 3.745ϵ2/3
6 2/3 5 2/3 3 4 2/3 3 3

 
√ √

− 0.691 ϵ + 0.005 + t 6.644ϵ − 2.517 ϵ + 0.0869 + 1.785 − 0.0002t6 ϵ
3 2 2/3 3

+ 0.004t6 + 0.0003t5 ϵ2/3 − 0.003t5 ϵ + 0.067t5 + 0.007t4 ϵ2/3 − 0.020t4 ϵ + 0.400t4

√ √
+ 0.049t3 ϵ2/3 − 0.002t3 3 ϵ − 0.066t3 ϵ + 1.33t3 + 0.179t2 ϵ2/3 − 0.043t2 3 ϵ

√ √
− 0.118t ϵ + 2.372t + 0.398tϵ − 0.201t ϵ − 0.892 ϵ + e−ϵ ,
2 2 2/3 3 3

and so on, where Γ(a, ϵ) is the incomplete Gamma function given by,
Z ∞
Γ(a, ϵ) = ta−1 e−t dt.
ϵ

Further, iterative solution components can be easily computed using any computational
software such as MATHEMATICA. The higher terms are ommited here due to long
Analytical Approach for Nonlinear Collisional Fragmentation Model 17

expressions. Similar to the previous case, here also, the density function and integral
moments obtained using BLUES method are presented with the FVM obtained density
function and moments in Figure 4. It can be observed from the figure that the BLUES
method predicts the number density (Figure 4(a)) and moments (Figures 4(b), 4(c) and
4(d)) with very good agreement with the FVM’s moments.

3.5 1.3
0
, FVM

0
, BLUES
3

Normalized zeroth moments

1.25
number density function

2.5
1.2

2
1.15
1.5

1.1
1

0.5 1.05

0 1
10-8 10-6 10-4 10-2 100 102 0 0.05 0.1 0.15 0.2 0.25 0.3
volume time

(a) Number density (b) Zeroth moment

2 1

1
, FVM 2
, FVM
1.8
, BLUES , BLUES
Normalized second moments

1 0.98 2
Normalized first moments

1.6

1.4 0.96

1.2
0.94
1
0.92
0.8

0.6 0.9

0.4
0.88
0.2

0 0.86
0 0.05 0.1 0.15 0.2 0.25 0.3 0 0.05 0.1 0.15 0.2 0.25 0.3
time time

(c) First moment (d) Second moment

Figure 4: Particle density distribution and moments functions for Example 3.3

Example 3.4. Product coagulation kernel K(ϵ, ρ) = ϵρ and breakage kernel

3
B(ϵ|ρ, ζ) = 1/2 1/2 with exponential initial condition f (0, ϵ) = e−ϵ
2ϵ ρ

The methodology defined in Section 2 provides the following sequence of approxima-

tions

f0 (t, ϵ) =e−ϵ ,
  √ √
−ϵ 3t 3 π erfc ( ϵ) t
f1 (t, ϵ) =e 1 − tϵ + + √ ,
2 4 ϵ
  √ √  
−ϵ t2 ϵ2 3t 9t2 3 πt erfc ( ϵ) 3t
f2 (t, ϵ) =e 1 − tϵ + + (1 − tϵ) + + √ 1+
2 2 16 4 ϵ 8
18 Hussain, Arora and Kumar

15 √ √
√
− π erfc ϵ t2 ϵ,
16
   
−ϵ t2 ϵ2 t3 ϵ3 3t t 2 ϵ2 9t2 35t3 ϵ 15t3
f3 (t, ϵ) =e 1 − tϵ + − + 1 − tϵ + + − −
2 3! 2 2 16 64 128
√ √    
3 πt erfc ( ϵ) 3t 5t 2 √ √
2 √ 15 15 35
+ √ 1+ − − π erfc ϵ t ϵ − t + tϵ .
4 ϵ 8 64 16 64 64

The above iterative expressions follows a pattern with some extra terms. These solutions
components may converge to a function of the form
  √ √ 
−ϵ −tϵ 3t −tϵ 9t2 35t3 ϵ 15t3 3 πt erfc ( ϵ)
lim fn (t, ϵ) = e e + e + − − ··· + √ 1+
n→∞ 2 16 64 128 4 ϵ
  
3t 5t2 √ √
2 √ 15 15 35
− + · · · − π erfc ϵ t ϵ − t + tϵ · · · .
8 64 16 64 64

In Figure 5(a), the FVM solution and BLUES solution with 5th order approximation
f5 (t, ϵ) are compared at various time levels. The FVM and BLUES solutions align well
with each other. The zeroth and first moments computed using BLUES also match closely
with the moments obtained from the FVM solution, as provided in Figures 5(b) and 5(c).
This simplification results in a linear increase in the number of bubble particles, as shown
in Figure 5(b), and the proposed technique accurately captures this behavior. Further-
more, Figure 5(c) compares the first moments calculated using BLUES with the first
moment from the FVM solution, confirming mass conservation in the system. However,
Figure 5(d) shows that the BLUES method captures the second moment accurately for
the shorter time period. As time increases, the BLUES second moment starts to exhibit
discrepancies from the FVM second moment.

Example 3.5. Coagulation kernel K(ϵ, ρ) = ϵρ/20, with binary breakage function
2
B(ϵ|ρ, ζ) = and initial distribution f (0, ϵ) = ϵ2 e−ϵ
ρ
To test the novelty of the proposed scheme, here in this case, we combine different
kernel parametes. The rate of coagulation kernel is reduces to K(ϵ, ρ) = ϵρ/20 and the
initial distribution is chosen as f (0, ϵ) = ϵ2 e−ϵ . The breakage function is selected to ran-
dom binary breakage. Fortunately, in this case we were able to find the analytical zeroth
and first order moments. The calculation for these moments are provided in Appendix
and the expressions are given as

9t
M0 (t) = + 2, and M1 (t) = 6. (24)
5
The second-order moment could not be calculated analytically due to the involvement of
higher-order moments in the calculation.
The BLUES iterative solutions for this case are given as

f0 (t, ϵ) =e−ϵ ϵ2 ,
   
−ϵ 1 3t 6tϵ 6t
f1 (t, ϵ) =e − 3tϵ +
3 2
+1 ϵ + + ,
10 5 5 5
Analytical Approach for Nonlinear Collisional Fragmentation Model 19

160

0
, FVM
140 5
, BLUES

Normalized zeroth moments

0
number density function

120 4.5

100 4

80 3.5

60 3

40 2.5

20 2

0 1.5

-20 1
10-3 10-2 10-1 100 101 102 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
volume time

(a) Number density (b) Zeroth moment

2 1

1
, FVM 2
, FVM
1.8
, BLUES 0.95 , BLUES

Normalized second moments

1 2
Normalized first moments

1.6
0.9
1.4
0.85
1.2

1 0.8

0.8
0.75
0.6
0.7
0.4
0.65
0.2

0 0.6
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
time time

(c) First moment (d) Second moment

Figure 5: Particle density distribution and moments functions for Example 3.4

     
9t2 ϵ4 9t2 3t 3 −ϵ 27t2 3t 27t2 6tϵ 6t
f2 (t, ϵ) =e + − − ϵ + − + +1 ϵ + 2
+ + ,
200 50 10 100 5 50 5 5
      
−ϵ 9t3 ϵ5 6t 81t3 27t2 27t3 9t2 27t3 9t2 3t 3
f3 (t, ϵ) =e − + − ϵ+ + + 4
ϵ + − − ϵ
2000 5 500 50 1000 200 1000 50 10
  
27t3 27t2 3t 6t
+ − − + +1 ϵ + 2
,
500 100 5 5

and so on. As the terms are getting complex for higher iterative solutions, we do not
include the further iterations here, though these iterations are easily calculable with the
help of equation (12). Since, the analytical solution for the number density function is
not availaible in this instance, therefore the findings of approximated density distribution
using BLUES are compared with findings of FVM solution. The results indicate that the
density functions obtained using the BLUES method align closely with the FVM solutions,
as illustrated in Figure 6(a). This figure demonstrates the high degree of accuracy achieved
by BLUES in approximating the number density functions compared to the FVM. The
exact moments are calculated in this example and compared with the FVM and BLUES
method moments. The expected growth in the number of particles is observed in Figure
6(b). Also, as observed in previous examples, both approaches consistently uphold the
20 Hussain, Arora and Kumar

mass conservation law. This is demonstrated in Figure 6(c), where the first-order moment
accurately reflects the conservation of mass. This consistency highlights the reliability and
robustness of the BLUES approach in ensuring that fundamental physical principles are
maintained. Furthermore, similar energy dissipation was captured by both FVM and
BLUES methods as in Figure 6(d).

5 3
0
, FVM
2.8
, BLUES

Normalized zeroth moments

0
4
number density function

2.6 0
, Exact

2.4
3
2.2

2 2

1.8
1
1.6

1.4
0
1.2

-1 1
10-3 10-2 10-1 100 101 102 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
volume time

(a) Number density (b) Zeroth moment

2 1

1
, FVM 2
, FVM
1.8
, BLUES , BLUES
Normalized second moments

1 0.95 2
Normalized first moments

1.6

1.4 0.9

1.2
0.85
1
0.8
0.8

0.6 0.75

0.4
0.7
0.2

0 0.65
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
time time

(c) First moment (d) Second moment

Figure 6: Particle density distribution and moments functions for Example 3.5

4. Concluding Remarks
This article presented a significant advancement in the field of semi-analytical methods
by demonstrating the BLUES function method’s effectiveness in solving the CBE, even in
cases where exact solutions were previously unavailable. By meticulously analyzing five
different test cases, which incorporated variations in initial conditions, collisional kernels,
and breakage functions, the study showcased the precision of the BLUES method and
its superiority in estimating number density and moments after a minimal number of
iterations. Remarkably, in the first example, the series solution exactly matched precise
solution documented in existing literature, reinforcing the credibility and accuracy of the
BLUES method. Furthermore, in other examples without precise solutions, the algorithm
Analytical Approach for Nonlinear Collisional Fragmentation Model 21

consistently mirrored the model’s predicted behavior with remarkable fidelity and pro-
vided the solution in the form of functions. The comprehensive numerical comparison
with the FVM and graphical analysis further affirmed the reliability and robustness of
the proposed scheme, especially in determining the number density function and first two
integral moments. A pivotal contribution of this work was its in-depth analysis of the
series solutions’ convergence towards analytical solution and the establishment of error
bounds. Therefore, this article highlighted the use of the BLUES function method for
solving the CBE with very good accuracy.
In future, due to high efficiency and accuracy, the current approach can be extended
to solve similar models such as multi-dimensional fragmentation and aggregation equa-
tions or complex partial differential equations. To strengthen the theoretical results, a
comparison with experimental data can be explored.

Acknowledgement
Mr. Saddam Hussain is thankful to Council of Scientific and Industrial Research (CSIR),
New Delhi, Government of India, for supporting this research work under the research
Grant No. 09/719(0125)/2021-EMR-I. Dr. Rajesh Kumar wishes to thank Science and
Engineering Research Board (SERB), DST India, for the funding through the project
MTR/2021/000866.

References
[1] M. Tournus, M. Escobedo, W.-F. Xue, and M. Doumic, “Insights into the dynamic
trajectories of protein filament division revealed by numerical investigation into the
mathematical model of pure fragmentation,” PLoS Computational Biology, vol. 17,
no. 9, p. e1008964, 2021.

[2] L. G. Wang, R. Ge, X. Chen, R. Zhou, and H.-M. Chen, “Multiscale digital twin for
particle breakage in milling: From nanoindentation to population balance model,”
Powder Technology, vol. 386, pp. 247–261, 2021.

[3] J. Gillespie and R. List, “Effects of collision-induced breakup on drop size distribu-
tions in steady state rainshafts,” Pure and Applied Geophysics, vol. 117, pp. 599–626,
1978.

[4] R. List and J. Gillespie, “Evolution of raindrop spectra with collision-induced

breakup,” Journal of Atmospheric Sciences, vol. 33, no. 10, pp. 2007–2013, 1976.

[5] P. Laurençot and D. Wrzosek, “The discrete coagulation equations with collisional
breakage,” Journal of Statistical Physics, vol. 104, pp. 193–220, 2001.

[6] S. Chen and S. Li, “Collision-induced breakage of agglomerates in homogenous

isotropic turbulence laden with adhesive particles,” Journal of Fluid Mechanics, vol.
902, p. A28, 2020.
22 Hussain, Arora and Kumar

[7] J. Kiefer and H. Straaten, “A model of ion track structure based on classical collision
dynamics (radiobiology application),” Physics in Medicine & Biology, vol. 31, no. 11,
p. 1201, 1986.

[8] Y. Zeng, F. Jia, X. Meng, Y. Han, and Y. Xiao, “The effects of friction characteristic
of particle on milling process in a horizontal rice mill,” Advanced Powder Technology,
vol. 29, no. 5, pp. 1280–1291, 2018.

[9] S. Ma, C. Wang, L. Li, and X. Wang, “Effects of particle size on the quality attributes
of wheat flour made by the milling process,” Cereal Chemistry, vol. 97, no. 2, pp.
172–182, 2020.

[10] J. Blum, “Dust evolution in protoplanetary discs and the formation of planetesimals:
What have we learned from laboratory experiments?” Space Science Reviews, vol.
214, no. 2, p. 52, 2018.

[11] L. Fries, S. Antonyuk, S. Heinrich, D. Dopfer, and S. Palzer, “Collision dynamics in

fluidised bed granulators: A dem-cfd study,” Chemical Engineering Science, vol. 86,
pp. 108–123, 2013.

[12] M. Lombart, M. Hutchison, and Y.-N. Lee, “Fragmentation with discontinuous

galerkin schemes: non-linear fragmentation,” Monthly Notices of the Royal Astro-
nomical Society, vol. 517, no. 2, pp. 2012–2027, 2022.

[13] G. M. McFarquhar, “A new representation of collision-induced breakup of raindrops

and its implications for the shapes of raindrop size distributions,” Journal of the
Atmospheric Sciences, vol. 61, no. 7, pp. 777–794, 2004.

[14] W. Wang, Q. Pan, Z. Shen, Y. Dong, L. Zhou, Z. Yu, G. Chen, and J. Duan,
“Investigation of hydrodynamic performance in a staggered multistage internal airlift
loop reactor,” Physics of Fluids, vol. 36, no. 2, 2024.

[15] Z. Ansari, M. Rae, and M. Singh, “Two moments preserving sectional approach for
an enzymatic coagulation equation,” Physics of Fluids, vol. 36, no. 6, 2024.

[16] J. Mao, Z.-T. Wang, Y.-C. Yang, and D. Li, “Numerical treatments for large eddy
simulations of liquid–liquid dispersions via population balance equation,” Physics of
Fluids, vol. 35, no. 6, 2023.

[17] C. Martinez-Bazan, J. Rodriguez-Rodriguez, G. Deane, J. Montanes, and J. Lasheras,

“Considerations on bubble fragmentation models,” Journal of fluid mechanics, vol.
661, pp. 159–177, 2010.

[18] D. B. Gaylo, K. Hendrickson, and D. K. Yue, “Fundamental time scales of bubble

fragmentation in homogeneous isotropic turbulence,” Journal of Fluid Mechanics,
vol. 962, p. A25, 2023.

[19] A. Aiyer, D. Yang, M. Chamecki, and C. Meneveau, “A population balance model for
large eddy simulation of polydisperse droplet evolution,” Journal of Fluid Mechanics,
vol. 878, pp. 700–739, 2019.
Analytical Approach for Nonlinear Collisional Fragmentation Model 23

[20] G. Yang, K. Guo, and T. Wang, “Numerical simulation of the bubble column at
elevated pressure with a cfd-pbm coupled model,” Chemical Engineering Science,
vol. 170, pp. 251–262, 2017.

[21] K. Guo, T. Wang, Y. Liu, and J. Wang, “Cfd-pbm simulations of a bubble column
with different liquid properties,” Chemical Engineering Journal, vol. 329, pp. 116–
127, 2017.

[22] Z. Cheng and S. Redner, “Scaling theory of fragmentation,” Physical review letters,
vol. 60, no. 24, p. 2450, 1988.

[23] R. M. Ziff, “New solutions to the fragmentation equation,” Journal of Physics A:

Mathematical and General, vol. 24, no. 12, p. 2821, 1991.

[24] T. W. Peterson, “Similarity solutions for the population balance equation describing
particle fragmentation,” Aerosol Science and Technology, vol. 5, no. 1, pp. 93–101,
1986.

[25] G. Breschi and M. A. Fontelos, “A note on the self-similar solutions to the sponta-
neous fragmentation equation,” Proceedings of the Royal Society A: Mathematical,
Physical and Engineering Sciences, vol. 473, no. 2201, p. 20160740, 2017.

[26] E. S. Hosseininia and A. Mirghasemi, “Numerical simulation of breakage of two-

dimensional polygon-shaped particles using discrete element method,” Powder Tech-
nology, vol. 166, no. 2, pp. 100–112, 2006.

[27] R. Kumar, J. Kumar, and G. Warnecke, “Moment preserving finite volume schemes
for solving population balance equations incorporating aggregation, breakage, growth
and source terms,” Mathematical Models and Methods in Applied Sciences, vol. 23,
no. 07, pp. 1235–1273, 2013.

[28] J.-P. Bourgade and F. Filbet, “Convergence of a finite volume scheme for coagulation-
fragmentation equations,” Mathematics of Computation, vol. 77, no. 262, pp. 851–
882, 2008.

[29] L. Forestier-Coste and S. Mancini, “A finite volume preserving scheme on nonuniform

meshes and for multidimensional coalescence,” SIAM Journal on Scientific Comput-
ing, vol. 34, no. 6, pp. B840–B860, 2012.

[30] M. M. Attarakih, C. Drumm, and H.-J. Bart, “Solution of the population balance
equation using the sectional quadrature method of moments (sqmom),” Chemical
Engineering Science, vol. 64, no. 4, pp. 742–752, 2009.

[31] N. Ahmed, G. Matthies, and L. Tobiska, “Stabilized finite element discretization

applied to an operator-splitting method of population balance equations,” Applied
Numerical Mathematics, vol. 70, pp. 58–79, 2013.

[32] Y. Lin, K. Lee, and T. Matsoukas, “Solution of the population balance equation
using constant-number monte carlo,” Chemical Engineering Science, vol. 57, no. 12,
pp. 2241–2252, 2002.
24 Hussain, Arora and Kumar

[33] A. K. Giri and P. Laurençot, “Existence and nonexistence for the collision-induced
breakage equation,” SIAM Journal on Mathematical Analysis, vol. 53, no. 4, pp.
4605–4636, 2021.

[34] C. Walker, “Coalescence and breakage processes,” Mathematical Methods in the Ap-
plied Sciences, vol. 25, no. 9, pp. 729–748, 2002.

[35] P. K. Barik and A. K. Giri, “Global classical solutions to the continuous coagula-
tion equation with collisional breakage,” Zeitschrift für Angewandte Mathematik und
Physik, vol. 71, pp. 1–23, 2020.

[36] A. K. Giri and P. Laurençot, “Weak solutions to the collision-induced breakage equa-
tion with dominating coagulation,” Journal of Differential Equations, vol. 280, pp.
690–729, 2021.

[37] A. Das, P. Kushwah, J. Saha, and M. Singh, “Improved higher-order finite volume
scheme and its convergence analysis for collisional breakage equation,” Applied Nu-
merical Mathematics, vol. 196, pp. 118–132, 2024.

[38] A. Das, J. Kumar, M. Dosta, and S. Heinrich, “On the approximate solution and
modeling of the kernel of nonlinear breakage population balance equation,” SIAM
journal on scientific computing, vol. 42, no. 6, pp. B1570–B1598, 2020.

[39] M. Kumar and Umesh, “Recent development of adomian decomposition method for
ordinary and partial differential equations,” International Journal of Applied and
Computational Mathematics, vol. 8, no. 2, p. 81, 2022.

[40] S. Kaushik, S. Hussain, and R. Kumar, “Laplace transform-based approximation

methods for solving pure aggregation and breakage equations,” Mathematical Meth-
ods in the Applied Sciences, vol. 46, pp. 1–20, 2023.

[41] G. Arora, R. Kumar, and Y. Mammeri, “Homotopy perturbation and adomian de-
composition methods for condensing coagulation and lifshitz-slyzov models,” GEM-
International Journal on Geomathematics, vol. 14, no. 1, p. 4, 2023.

[42] N. Yadav, A. Das, M. Singh, S. Singh, and J. Kumar, “Homotopy perturbation

method and its convergence analysis for nonlinear collisional fragmentation equa-
tions,” Proceedings of the Royal Society A, vol. 479, no. 2279, p. 20230567, 2023.

[43] G. Arora, S. Hussain, and R. Kumar, “Comparison of variational iteration and ado-
mian decomposition methods to solve growth, aggregation and aggregation-breakage
equations,” Journal of Computational Science, vol. 67, p. 101973, 2023.

[44] S. Liao, Beyond perturbation: introduction to the homotopy analysis method. Chap-
man and Hall/CRC, 2003.

[45] Z. Eshkuvatov, “New development of homotopy analysis method for non-linear

integro-differential equations with initial value problems,” Mathematical Modeling
and Computing, no. 9, Num. 4, pp. 842–859, 2022.
Analytical Approach for Nonlinear Collisional Fragmentation Model 25

[46] S. Hussain and R. Kumar, “Elzaki projected differential transform method for multi-
dimensional aggregation and combined aggregation-breakage equations,” Journal of
Computational Science, vol. 75, p. 102211, 2024.

[47] S. Hussain, G. Arora, and R. Kumar, “Semi-analytical methods for solving non-linear
differential equations: A review,” Journal of Mathematical Analysis and Applications,
vol. 531, p. 127821, 2023.

[48] J. O. Indekeu and K. K. Müller-Nedebock, “Blues function method in computational

physics,” Journal of Physics A: Mathematical and Theoretical, vol. 51, no. 16, p.
165201, 2018.

[49] J. Berx and J. O. Indekeu, “Blues function method applied to partial differential
equations and analytic approximants for interface growth under shear,” Physical
Review Research, vol. 3, no. 3, p. 033113, 2021.

[50] ——, “Epidemic processes with vaccination and immunity loss studied with the blues
function method,” Physica A: Statistical Mechanics and its Applications, vol. 590, p.
126724, 2022.

[51] J. Berx, “Deposition, diffusion and convection: Blues approximants and some exact
results,” Ph.D. dissertation, Ph. D. thesis, KU Leuven, 2021.

[52] D. J. McMahon and R. J. Brown, “Enzymic coagulation of casein micelles: a review,”

Journal of Dairy science, vol. 67, no. 5, pp. 919–929, 1984.

[53] H. R. Kobraei and G. C. Duncan, “Evolution of discrete coagulation equation,”

Journal of colloid and interface science, vol. 113, no. 2, pp. 557–565, 1986.

[54] R. M. Ziff and E. McGrady, “The kinetics of cluster fragmentation and depolymeri-
sation,” Journal of Physics A: Mathematical and General, vol. 18, no. 15, p. 3027,
1985.
 26 Hussain, Arora and Kumar

A. Appendix
As previously mentioned, the process of computing the analytical moments is presented
here.
2
Considering K(ϵ, ρ) = ϵρ/20, B(ϵ|ρ, ζ) = , and f in (ϵ) = ϵ2 e−ϵ in equation (1) give
ρ
Z ∞Z ∞ Z ∞
∂f (t, ϵ) ζ ϵρ
= f (t, ρ)f (t, ζ)dρdζ − f (t, ϵ)f (t, ρ)dρ. (25)
∂t 0 ϵ 10 0 20

Integrating equation (25) from 0 to ∞ with respect to ϵ yields

Z ∞Z ∞Z ∞
dM0 (t) ζ M 2 (t)
= f (t, ρ)f (t, ζ)dρdζdϵ − 1 . (26)
dt 0 0 x 10 20

Changing the order of integration lead to the following

Z ∞Z ∞
dM0 (t) 1 1
= ρζf (t, ρ)f (t, ζ)dρdζ − M12 (t)
dt 10 0 0 20
1 1
= M12 (t) − M12 (t),
10 20
dM0 (t) 1 2
= M1 (t). (27)
dt 20
To Solve the above DE, we need the first moment M1 . For that, multiplying the equation
(25) by ϵ and integrating from 0 to ∞ with respect to ϵ provide
Z ∞Z ∞Z ∞ Z ∞Z ∞ 2
dM1 (t) ϵζ ϵρ
= f (t, ρ)f (t, ζ)dρdζdϵ − f (t, ϵ)f (t, ρ)dρ.
dt 0 0 ϵ 10 0 0 20

Again, changing the order of integration and some further calculations yield

dM1 (t) 1 1
= M2 (t)M1 (t) − M2 (t)M1 (t) = 0.
dt 20 20

Therefore, dMdt1 (t) = 0 or M1 (t) = Constant. Since, the first moment is constant through-
out the process, it can also be computed using the initial condition. Multiplying the
initial data by ϵ and integrating from 0 to ∞ with respect to ϵ, we get the first moment
M1 (τ ) = 6. Using the value of M1 (t) in (27) provides the zeroth moment as M0 (t) = 9t5 +2.

View publication stats