Visit https://ebookultra.

com to download the full version and

explore more ebooks

Towards an Information Theory of Complex

Networks Statistical Methods and Applications
2011th Edition Matthias Dehmer

_____ Click the link below to download _____

https://ebookultra.com/download/towards-an-information-
theory-of-complex-networks-statistical-methods-and-
applications-2011th-edition-matthias-dehmer/

Explore and download more ebooks at ebookultra.com

Here are some suggested products you might be interested in.
Click the link to download

The structure of complex networks theory and applications

First Published In Paperback Edition Estrada

https://ebookultra.com/download/the-structure-of-complex-networks-
theory-and-applications-first-published-in-paperback-edition-estrada/

Complex Heterogeneous Systems Thermodynamics Information

Theory Composites Networks and Electrochemistry 1st
Edition Markus Schmuck
https://ebookultra.com/download/complex-heterogeneous-systems-
thermodynamics-information-theory-composites-networks-and-
electrochemistry-1st-edition-markus-schmuck/

Information and Exponential Families In Statistical Theory

Barndorff-Nielsen

https://ebookultra.com/download/information-and-exponential-families-
in-statistical-theory-barndorff-nielsen/

Statistical Theory and Methods for Evolutionary Genomics

Xun Gu

https://ebookultra.com/download/statistical-theory-and-methods-for-
evolutionary-genomics-xun-gu/
A Multidisciplinary Introduction to Information Security
DISCRETE MATHEMATICS AND ITS APPLICATIONS 2011th Edition
Stig F. Mjolsnes
https://ebookultra.com/download/a-multidisciplinary-introduction-to-
information-security-discrete-mathematics-and-its-applications-2011th-
edition-stig-f-mjolsnes/

Mixed Methods Social Networks Research Design and

Applications Silvia Domínguez

https://ebookultra.com/download/mixed-methods-social-networks-
research-design-and-applications-silvia-dominguez/

Protein Microarrays Methods and Protocols 2011th Edition

Ulrike Korf

https://ebookultra.com/download/protein-microarrays-methods-and-
protocols-2011th-edition-ulrike-korf/

Game theory in wireless and communication networks theory

models and applications 1st Edition Zhu Han

https://ebookultra.com/download/game-theory-in-wireless-and-
communication-networks-theory-models-and-applications-1st-edition-zhu-
han/

Numerical and Statistical Methods for Bioengineering

Applications in MATLAB 1st Edition Michael R. King

https://ebookultra.com/download/numerical-and-statistical-methods-for-
bioengineering-applications-in-matlab-1st-edition-michael-r-king/
Towards an Information Theory of Complex Networks
Statistical Methods and Applications 2011th Edition
Matthias Dehmer Digital Instant Download
Author(s): Matthias Dehmer, Frank Emmert-Streib, Alexander Mehler
ISBN(s): 9780817649036, 0817649034
Edition: 2011
File Details: PDF, 5.56 MB
Year: 2011
Language: english
Matthias Dehmer
Frank Emmert-Streib
Alexander Mehler
Editors

Towards an Information
Theory of Complex Networks
Statistical Methods and Applications
Editors
Matthias Dehmer Frank Emmert-Streib
UMIT School of Medicine, Dentistry
Institute of Bioinformatics and Biomedical Sciences
and Translational Research Center for Cancer Research and Cell Biology
Eduard-Wallnöfer-Zentrum I Queen’s University Belfast
A-6060 Hall in Tirol 97 Lisburn Road
Austria Belfast BT9 7BL
mathias.dehmer@umit.at United Kingdom
v@bio-complexity.com

Alexander Mehler
Faculty of Computer Science
and Mathematics
Goethe-University Frankfurt
am Main Robert-Mayer-Straße 10
P.O. Box: 154
D-60325 Frankfurt am Main
Germany
mehler@em.uni-frankfurt.de

ISBN 978-0-8176-4903-6 e-ISBN 978-0-8176-4904-3

DOI 10.1007/978-0-8176-4904-3
Springer New York Dordrecht Heidelberg London

Library of Congress Control Number: 2011932673

Mathematics Subject Classification (2010): 68R10, 68P30, 94C15

c Springer ScienceCBusiness Media, LLC 2011

All rights reserved. This work may not be translated or copied in whole or in part without the written
permission of the publisher (Springer Science+Business Media, LLC, 233 Spring Street, New York,
NY 10013, USA), except for brief excerpts in connection with reviews or scholarly analysis. Use in
connection with any form of information storage and retrieval, electronic adaptation, computer software,
or by similar or dissimilar methodology now known or hereafter developed is forbidden.
The use in this publication of trade names, trademarks, service marks, and similar terms, even if they are
not identified as such, is not to be taken as an expression of opinion as to whether or not they are subject
to proprietary rights.

Printed on acid-free paper

www.birkhauser-science.com
Preface

For more than a decade, complex network analysis has evolved as a methodological
paradigm for a multitude of disciplines, including physics, chemistry, biology,
geography, sociology, computer science, statistics, media science, and linguistics.
Researchers in these fields share an interest in information processing subject to the
networking of their corresponding research object, for instance, genes, molecules,
individuals, semes, memes, etc. They start with the insight that any of these
research objects is extrinsically characterized, if not constituted, by its networking
with objects of the same provenance. In this way, networks, for example, gene
networks, food networks, city networks, networks of words, sentences, texts, or web
documents become important research objects in more and more disciplines.
This book, in line with these research developments, presents theoretical and
practical results of statistical models of complex networks in the formal sciences,
the natural sciences, and the humanities. One of its goals is to advocate and promote
combinations of graph-theoretic, information-theoretic, and statistical methods as a
way to better understand and characterize real-world networks.
On the one hand, networks appear as paradigmatic objects of approaches
throughout the natural and social sciences and the humanities. On the other hand,
networks are—irrespective of their disciplinary provenance—known for character-
istic distributions of graph-theoretic invariants which affect their robustness and
efficiency in information processing. The main goal of this book is to further develop
information-theoretic notions and to elaborate statistical models of information
processing in such complex networks. In this way, the book includes first steps
toward establishing a statistical information theory as a unified basis for complex
network analysis across a multitude of scientific disciplines.
The book presents work on the statistics of complex networks together with
applications of information theory in a range of disciplines such as quantitative
biology, quantitative chemistry, quantitative sociology, and quantitative linguistics.
It aims to integrate models of invariants of network topologies and dynamic
aspects of information processing in these networks or by means of these networks.

v
vi Preface

Thus, the book is in support of sharing and elaborating models and methods that may
help researchers get insights into complex problems emerging from interdisciplinary
reasoning.
The book is divided into two parts: Chaps. 1–4 deal with formal-theoretical issues
of network modeling, while Chaps. 5–13 further develop and apply these methods
to empirical networks from a wide range of areas. The book starts with a theoretical
contribution by Abbe Mowshowitz on the entropy of digraphs and infinite graphs.
The aim is to provide insights into more complex graph models that go beyond
the majority of network models based on finite undirected graphs. The chapter by
Nicolas Bonichon, Cyril Gavoille, and Nicolas Hanusse presents an information-
theoretic upper bound of planar graphs by means of the newly introduced notion of
well-orderly maps. Such a technique might be useful when studying properties of
the very important notion of planar graphs. Terence Chan and Raymond W. Yeung
study a statistical inference problem using network models. Richard Berkovits,
Lukas Jahnke, and Jan W. Kantelhardt examine phase transitions within complex
networks that help to examine their structural properties.
The remainder of the book combines the theoretical stance of the first section
with an empirical analysis of real networks. Elena Konstantinova provides a survey
on information-theoretic measures used in chemical graph theory. Prabhat K. Sahu
and Shyi-Long Lee develop a model of chemical graphs by example of molecular
networks. Exploring the spectral characteristics of these graphs, they provide a
successful classification of chemical graphs.
Biological or, more specifically, ecological networks are dealt with by Robert E.
Ulanowicz who describes a framework of quantifying patterns of the interaction
of networked trophic processes from the point of view of information theory.
Ecological networks are also the focus of the chapter of Linda J. Moniz, James D.
Nichols, Jonathan M. Nichols, Evan G. Cooch, and Louis M. Pecora, who provide
an approach to modeling the interaction dynamics of ecosystems and their change.
A comprehensive view of ontologically disparate networks is given by Cristian
R. Munteanu, J. Dorado, A. Pazos Sierra, F. Prado-Prado, L.G. Pérez-Montoto,
S. Vilar, F.M. Ubeira, A. Sanchez-Gonzaléz, M. Cruz-Monteagudo, S. Arrasate,
N. Sotomayor, E. Lete, A. Duardo-Sánchez, A. Dı́az-López, G. Patlewicz, and
H. González-Dı́az who use the notion of entropy centrality to compare various
systems such as chemical, biological, crime, and legislative networks, thereby
showing the interdisciplinary expressiveness of complex network theory.
The book continues with two contributions to linguistic networks: Alexander
Mehler develops a framework for analyzing the topology of social ontologies as
they evolve within Wikipedia and contrasts them with nonsocial, formal ontologies.
Olga Abramov and Tatjana Lokot present a comparative, classificatory study of
morphological networks by means of several measures of graph entropy.
Edward B. Allen discusses the measurement of the complexity and error prob-
ability of software systems represented as hypergraphs. Finally, in the chapter by
Philippe Blanchard and Dimitri Volchenkov, random walks are studied as a kind
of Markov process on graphs that allow insights into the dynamics of networks as
diverse as city and trade and exchange networks.
Preface vii

With such a broad field, it is clear that the present book addresses an interdisci-
plinary readership. It does not simply promote transdisciplinary research. Rather, it
is about interdisciplinary research that may be the starting point of developing an
overarching network science.

Matthias Dehmer
Frank Emmert-Streib
Alexander Mehler
Acknowledgments

Many colleagues have provided us with input, help, and support (consciously or
unconsciously) before and during the preparation of this book. In particular, we
would like to thank Andreas Albrecht, Gökmen Altay, Gabriel Altmann, Alain
Barrat, Igor Bass, David Bialy, Philippe Blanchard, Danail Bonchev, Stefan Borgert,
Mieczysław Borowiecki, Andrey A. Dobrynin, Michael Drmota, Ramon Ferrer
i Cancho, Maria and Gheorghe Duca, Maria Fonoberova, Armin Graber, Martin
Grabner, Peter Gritzmann, Ivan Gutman, Peter Hamilton, Wilfried Imrich, Patrick
Johnston, Elena Konstantinova, D. D. Lozovanu, Dennis McCance, Abbe Mow-
showitz, Arcady Mushegian, Andrei Perjan, Armindo Salvador, Maximilian Schich,
Heinz Georg Schuster, Helmut Schwegler, Andre Ribeiro, Burghard Rieger, Brigitte
Senn-Kircher, Fred Sobik, Doru Stefanescu, John Storey, Shailesh Tripathi, Kurt
Varmuza, Bohdan Zelinka, and Shu-Dong Zhang. Additionally, Matthias Dehmer
thanks Armin Graber for strong support and providing a fruitful atmosphere at
UMIT. Finally, we would like to thank our editor Tom Grasso who has been always
available and helpful.
The work on the chapters of Philippe Blanchard and Dimitri Volchenkov, Olga
Abramov, and Alexander Mehler have been supported by the German Federal
Ministry of Education and Research (BMBF) through the project Linguistic Net-
works.1 We gratefully acknowledge this financial support.

1
www.linguistic-networks.net.

ix
Contents

1 Entropy of Digraphs and Infinite Networks . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 1

A. Mowshowitz
2 An Information-Theoretic Upper Bound on Planar
Graphs Using Well-Orderly Maps . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 17
Nicolas Bonichon, Cyril Gavoille, and Nicolas Hanusse
3 Probabilistic Inference Using Function Factorization
and Divergence Minimization . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 47
Terence H. Chan and Raymond W. Yeung
4 Wave Localization on Complex Networks . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 75
Richard Berkovits, Lukas Jahnke, and Jan W. Kantelhardt
5 Information-Theoretic Methods in Chemical Graph Theory.. . . . . . . . . 97
Elena Konstantinova
6 On the Development and Application of Net-Sign Graph Theory . . . . 127
Prabhat K. Sahu and Shyi-Long Lee
7 The Central Role of Information Theory in Ecology . . . . . . . . . . . . . . . . . . . 153
Robert E. Ulanowicz
8 Inferences About Coupling from Ecological Surveillance
Monitoring: Approaches Based on Nonlinear Dynamics
and Information Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 169
L.J. Moniz, J.D. Nichols, J.M. Nichols, E.G. Cooch,
and L.M. Pecora

xi
xii Contents

9 Markov Entropy Centrality: Chemical, Biological, Crime,

and Legislative Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 199
C.R. Munteanu, J. Dorado, Alejandro Pazos-Sierra,
F. Prado-Prado, L.G. Pérez-Montoto, S. Vilar, F.M. Ubeira,
A. Sanchez-Gonzaléz, M. Cruz-Monteagudo, S. Arrasate,
N. Sotomayor, E. Lete, A. Duardo-Sánchez, A. Dı́az-López,
G. Patlewicz, and H. González-Dı́az
10 Social Ontologies as Generalized Nearly Acyclic Directed
Graphs: A Quantitative Graph Model of Social Tagging . . . . . . . . . . . . . . 259
Alexander Mehler
11 Typology by Means of Language Networks: Applying
Information Theoretic Measures to Morphological
Derivation Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 321
Olga Abramov and Tatiana Lokot
12 Information Theory-Based Measurement of Software . . . . . . . . . . . . . . . . . 347
Edward B. Allen
13 Fair and Biased Random Walks on Undirected Graphs
and Related Entropies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 365
Philippe Blanchard and Dimitri Volchenkov
Contributors

Olga Abramov University of Bielefeld, Universitätsstraße 25, 33615 Bielefeld,

Germany, olga.abramov@uni-bielefeld.de
Edward B. Allen Department of Computer Science and Engineering, Mississippi
State University, Box 9637, Mississippi State, MS 39762, USA,
edward.allen@computer.org
S. Arrasate Department of Organic Chemistry II, Faculty of Science and Technol-
ogy, University of the Basque Country/Euskal Herriko Unibertsitatea, Apto. 644,
48080 Bilbao, Spain, sonia.arrasate@ehu.es
Richard Berkovits Minerva Center and Department of Physics, Bar-Ilan Univer-
sity, Ramat-Gan 52900, Israel, berkov@mail.biu.ac.il
Philippe Blanchard Bielefeld – Bonn Stochastic Research Center (BiBoS),
University of Bielefeld, Postfach 100131, 33501 Bielefeld, Germany
Nicolas Bonichon LaBRI, University of Bordeaux, 351 Cours de la libération,
33405 Bordeaux, France, bonichon@labri.fr
Terence H. Chan Institute for Telecommunications Research, University
of South Australia, Adelaide, SA 5095, Australia, hlchan6@gmail.com;
terence.chan@unisa.edu.au
E.G. Cooch Department of Natural Resources, Cornell University, Ithaca,
NY 14853, USA, evan.cooch@cornell.edu
M. Cruz-Monteagudo CEQA, Faculty of Chemistry and Pharmacy, UCLV, Santa
Clara 54830, Cuba, gmailkelcm@yahoo.es
A. Dı́az-López Department of Special Public Law, Faculty of Law, University of
Santiago de Compostela, 15782 Santiago de Compostela, Spain,
antonio.lopez.diaz@usc.es

xiii
xiv Contributors

J. Dorado Department of Information and Communication Technologies,

Computer Science Faculty, University of A Coruña, 15071 A Coruña, Spain,
julian@udc.es
A. Duardo-Sánchez Department of Special Public Law, Faculty of Law,
University of Santiago de Compostela, 15782 Santiago de Compostela, Spain,
aliuskaduardo@yahoo.es
Cyril Gavoille LaBRI, University of Bordeaux, 351 Cours de la libération, 33405
Bordeaux, France, gavoille@labri.fr
H. González-Dı́az Department of Microbiology and Parasitology, Faculty of
Pharmacy, University of Santiago de Compostela, 15782 Santiago de Compostela,
Spain, humberto.gonzalez@usc.es
Nicolas Hanusse LaBRI, CNRS – University of Bordeaux, 351 Cours de la
libération, 33405 Bordeaux, France, hanusse@labri.fr
Lukas Jahnke Martin-Luther-Universität Halle-Wittenberg, 06099 Halle,
Germany
Jan W. Kantelhardt Martin-Luther-Universität Halle-Wittenberg, 06099 Halle,
Germany
Elena Konstantinova Sabolev Institute of Mathematics, Siberian Branch of
Russian Academy of Sciences, 630090 Novosibirsk, Russia,
e konsta@math.nsc.ru
Shyi-Long Lee Department of Chemistry and Biochemistry, National Chung
Cheng University, Chia-Yi, 621 Taiwan, chesll@ccu.edu.tw
E. Lete Department of Organic Chemistry II, Faculty of Science and Technology,
University of the Basque Country/Euskal Herriko Unibertsitatea, Apto. 644,
48080 Bilbao, Spain, esther.lete@ehu.es
Tatiana Lokot Faculty of Technology, University of Bielefeld, Universitaetsstr. 25,
33615 Bielefeld, Germany, tlokot@math.uni-bielefeld.de
Alexander Mehler Faculty of Computer Science and Mathematics, Goethe Uni-
versity Frankfurt am Main, D-60325 Frankfurt am Main, Germany,
mehler@em.uni-frankfurt.de
L.J. Moniz Johns Hopkins University, Applied Physics Laboratory, 11100 Johns
Hopkins Road, Laurel, MD 20723, USA, lindano@comcast.net
Abbe Mowshowitz Department of Computer Science, The City College of New
York (CUNY), 138th Street at Convent Avenue, New York, NY 10031, USA,
abbe@cs.ccny.cuny.edu
C.R. Munteanu Department of Information and Communication Technologies,
Computer Science Faculty, University of A Coruña, 15071 A Coruña, Spain,
cmunteaunu@udc.es
Contributors xv

J.D. Nichols U.S. Geological Survey, Patuxent Wildlife Research Center, Laurel,
MD 20708, USA, jnichols@usgs.gov
J.M. Nichols Naval Research Laboratory, Optical Sciences Division, Code 5673,
Washington, DC 20375, USA, jonathan.nichols@nrl.navy.mil
G. Patlewicz Institute for Health and Consumer Protection (IHPC), Joint Research
Centre (JRC), European Commission, via E. Fermi 2749–21027 Ispra (Varese), Italy
DuPont Haskell Global Centers for Health and Environmental Sciences, Newark,
DE 19711, USA, Grace.Y.Tier@usa.dupont.com
Alejandro Pazos-Sierra Department of Information and Communication Tech-
nologies, Computer Science Faculty, University of A Coruña,
15071 A Coruña, Spain, apazos@udc.es
L.M. Pecora Naval Research Laboratory, Code 6362, Washington, DC 20375,
USA, pecora@anvil.nrl.navy.mil
L.G. Pérez-Montoto Faculty of Pharmacy, University of Santiago de Compostela,
15782 Santiago de Compostela, Spain, lgmp2002@yahoo.es
F. Prado-Prado Faculty of Pharmacy, University of Santiago de Compostela,
15782 Santiago de Compostela, Spain, fenol1@hotmail.com
Prabhat K. Sahu Institüt für Physikalische und Theoretische Chemie, Universität
Würzburg, Am Hubland, 97074 Würzburg, Germany
Department of Chemistry and Biochemistry, National Chung Cheng University,
Chia-Yi, 621 Taiwan, sahu@chemie.uni-wuerzburg.de
A. Sanchez-Gonzaléz Department of Inorganic Chemistry, Faculty of Pharmacy,
University of Santiago de Compostela, 15782 Santiago de Compostela, Spain,
angeles.sanchez@usc.es
N. Sotomayor Department of Organic Chemistry II, Faculty of Science and
Technology, University of the Basque Country/Euskal Herriko Unibertsitatea, Apto.
644, 48080 Bilbao, Spain, nuria.sotomayor@ehu.es
F.M. Ubeira Department of Microbiology and Parasitology, Faculty of Pharmacy,
University of Santiago de Compostela, 15782 Santiago de Compostela, Spain,
fm.ubeira@usc.es
Robert E. Ulanowicz Department of Biology, University of Florida, Gainesville,
FL 32611-8525, USA
University of Maryland Center for Environmental Science, Solomons, MD 20688-
0038, USA, ulan@umces.edu
S. Vilar Faculty of Pharmacy, University of Santiago de Compostela, 15782
Santiago de Compostela, Spain, qosanti@yahoo.es
xvi Contributors

Dimitri Volchenkov The Center of Excellence Cognitive Interaction Technology

(CITEC), University of Bielefeld, Postfach 100131, 33501 Bielefeld, Germany,
volchenk@physik.uni-bielefeld.de
Raymond W. Yeung Department of Information Engineering, The Chinese
University of Hong Kong, whyeung@ie.cuhk.edu.hk
Chapter 1
Entropy of Digraphs and Infinite Networks

A. Mowshowitz

Abstract The information content of a graph G is defined in Mowshowitz

(Bull Math Biophys 30:175–204, 1968) as the entropy of a finite probability scheme
associated with the vertex partition determined by the automorphism group of G.
This provides a quantitative measure of the symmetry structure of a graph that has
been applied to problems in such diverse fields as chemistry, biology, sociology, and
computer science (Mowshowitz and Mitsou, Entropy, orbits and spectra of graphs,
Wiley-VCH, 2009). The measure extends naturally to directed graphs (digraphs)
and can be defined for infinite graphs as well (Mowshowitz, Bull Math Biophys
30:225–240, 1968).This chapter focuses on the information content of digraphs
and infinite graphs. In particular, the information content of digraph products and
recursively defined infinite graphs is examined.

Keywords Digraphs • Entropy • Infinite graphs • Information content

• Networks

MSC2000 Primary 68R10; Secondary 05C20, 05C25, 05C75, 94C15, 90B10.

1 Introduction

1.1 Overview

This chapter investigates the information content of directed and infinite graphs.
The information content of a finite graph (directed or undirected) is a quantitative
measure based on the symmetry structure of the graph. As explained in detail

A. Mowshowitz ()
Department of Computer Science, The City College of New York (CUNY),
138th Street at Convent Avenue, New York, NY 10031, USA
e-mail: abbe@cs.ccny.cuny.edu

M. Dehmer et al. (eds.), Towards an Information Theory of Complex Networks: 1

Statistical Methods and Applications, DOI 10.1007/978-0-8176-4904-3 1,
© Springer Science+Business Media, LLC 2011
2 A. Mowshowitz

below, the group of symmetries of a finite graph partitions the vertex set and thus
induces a unique finite probability scheme. The entropy of this scheme is taken to
be the information content of the graph. This “classical” notion differs from “graph
entropy” introduced in [16].
Development of the concept of entropy applied to finite graphs is discussed
in [17] and [20]. The application of entropy to graphs was introduced in the
1950s soon after the appearance of Shannon’s famous paper on information theory.
Entropy measurement has been used as a tool for characterizing molecules and
chemical structures. For example, measures characterizing the structural complexity
of chemical graphs have been developed and applied in [1, 3, 6]. Most of these
measures are based on graph invariants that generate an equivalence relation on
the vertices or edges of a graph. The resulting equivalence classes form a partition
to which a finite probability scheme [14] can be associated in a natural way. The
entropy of such a scheme provides a quantitative measure of structural complexity.
Various structural features of a graph have provided the basis for entropy
measures. The earliest centered on the symmetries of a graph [21]. Other features,
such as branching structure in molecular graphs, have been used to define entropy
measures [8]. Measures associated with graphs representing atoms and molecules
have been defined and applied to problems of discriminating chemical isomers and
to classifying atomic and chemical structures [7, 9, 15]. Such measures have also
been used for the analysis of biological networks [13]. Degree characteristics of a
graph have been used as basis for an entropy-based measure of disorder in complex
networks [23]. Interest in measuring the information content of graphs has also been
kindled in recent years by the growing importance of computer and social networks
in modern society [10, 24]. Relationships between graph entropy-based measures,
expressed as inequalities, have been demonstrated in [11].
The notion of information content can be extended to infinite graphs. The
approach adopted here is to consider an infinite graph as a sequence of finite graphs.
Each of the finite graphs in the sequence has a well-defined information content, and
if the corresponding sequence of information content values has an unambiguous
limit, that limit is defined to be the information content of the given infinite graph.
In Sect. 2, we will look into the existence of directed graphs with prescribed
information content and determine the information content of certain products of
directed graphs. Section 3 will focus on infinite graphs, investigating the information
content of some special classes of infinite graphs, and applying results from Sect. 2
to determine the information content of infinite graphs in general. Section 4 will
examine some applications of the information measure to problems in network
theory.

1.2 General Definitions

 
Definition 1. G D .V; E/; jV j < 1; E  V2 is called a finite undirected graph.
If G D .V; E/; jV j < 1, and E  V  V , then G is called a finite directed graph.
1 Entropy of Digraphs and Infinite Networks 3

Definition 2. A digraph Ln D .V; E/ is called a (directed) path of length n ( 1),

if V D fv0 ; v1 ;    ; vn g and E D f.v0 ; v1 /; .v1 ; v2 /;    ; .vn1 ; vn /g. The number of
vertices in Ln is n C 1, one more than the number of edges.
Definition 3. A digraph Cn D .V; E/ is called a (directed) cycle of length n ( 2),
if V D fv1 ; v2 ;    ; vn g and E D f.v1 ; v2 /; .v2 ; v3 /;    ; .vn ; v1 /g. Cn has the same
number (n) of vertices and edges.

Definition 4. The complete graph Kn has n vertices and n2 (undirected) edges.
See [12] for additional definitions of basic concepts in graph theory.

2 Entropy of Digraphs

2.1 Definition and Examples

The automorphism group of a digraph and the measure of information content based
on the group are defined below.
Definition 5. Let G D .V; E/ be a (directed or undirected) graph with vertex set
V (with jV j D n), and edge set E. The automorphism group of G, denoted by
Aut.G/, is the set of all adjacency preserving bijections of V .
Definition 6. Let fVi j1  i  kg be the collection of orbits of Aut.G/ and suppose
jVi j D ni f or 1  i  k. The entropy or information content of G is given by the
following formula [17]:
k
X ni n 
i
Ia .G/ D  log :
i D1
n n

Figure 1.1 illustrates the computation of the information content of a digraph.

2.2 Entropy of Digraph Products

Many different binary operations on graphs and digraphs appear in the literature
[19]. We will examine four such operations in some detail, namely, the sum, join,
Cartesian product, and the composition. Our aim is to determine the information
content of a digraph operation in relation to the information contents of the
respective digraphs in the operation. Such products are useful in defining classes
of digraphs with properties of interest in different applications, especially those
pertaining to the analysis of networks.
Definition 7. The sum of G1 and G2 is the digraph G1 [ G2 defined by V .G1 [
G2 / D V .G1 / [ V .G2 / and E.G1 [ G2 / D E.G1 / [ E.G2 /.
4 A. Mowshowitz

7 2
1
6 2

6 1 3
3 2
5 3
5 4

4
X Y Z

Orbit: {1,2,3,4,5,6} Orbits: {1}, {2,5}, {3,6}, {4,7} Orbits: {1}, {2}, {3}
Ia (X) = 0 Ia (Y) = −(1/7) log (1/7) Ia (Z) = log 3
−3(2/7) log (2/7)

Fig. 1.1 Computation of information content

G H G U H G+H GXH GoH

Fig. 1.2 Binary operations on digraphs

Definition 8. The join of G1 and G2 is the digraph G1 C G2 defined by V .G1 C

G2 / D V .G1 /[V .G2 / and E.G1 CG2 / D E.G1 /[E.G2 /[fŒu; vju 2 V .G1 /; v 2
V .G2 /g where Œu; v denotes the undirected edge joining u and v.
Definition 9. The Cartesian product of G1 and G2 is the digraph G1  G2
given by V .G1  G2 / D V .G1 /  V .G2 / and E.G1  G2 / D f.u; v/ D
..u1 ; u2 /; .v1 ; v2 //ju1 ; v1 2 V .G1 /; u2 ; v2 2 V .G2 /, and either u1 D v1 and .u2 ; v2 / 2
E.G2 / or u2 D v2 and .u1 ; v1 / 2 E.G1 /g
Definition 10. Two digraphs G and H are relatively prime with respect to the
Cartesian product if whenever G is isomorphic to G 0  D and H is isomorphic
to H 0  D, then D is the identity digraph K1 .
Definition 11. The composition of G1 and G2 is the digraph G1 ı G2
given by V .G1 ı G2 / D V .G1 /  V .G2 / and E.G1 ı G2 / D f.u; v/ D
..u1 ; u2 /; .v1 ; v2 //ju1 ; v1 2 V .G1 /; u2 ; v2 2 V .G2 /, and either .u1 ; v1 / 2 E.G1 /
or u1 D v1 and .u2 ; v2 / 2 E.G2 /g
The foregoing operations are illustrated in Fig. 1.2.
1 Entropy of Digraphs and Infinite Networks 5

These binary operations will be discussed with a view to characterizing the

information content of digraphs resulting from their application.

2.3 Sum and Join

Theorem 1. Let G and H be digraphs.

(a) Suppose Aut.G/ has orbits ViG with jViG j D mi for 1  i  m, and Aut.H /
has orbits ViH with jViH j D ni for 1  i  n. If no component of G is
isomorphic to a component of H , then

Ia .G [ H / D Ia .G C H /

1
D log.n C m/ C ŒnIa .G/ C mIa .H /
nCm
 n log.n/  m log.m/:

(b) If G and H are isomorphic, then Ia .G [ H / D Ia .G C H / D Ia .G/. More

generally, if each Gi .1  i  n/ is isomorphic to G, then

Ia .G1 [ G2    [ Gn / D Ia .G1 C G2    C Gn / D Ia .G/:

Proof. Ia .G [ H / D Ia .G C H / since the orbits of Aut.G [ H / are the same

as those of Aut.G C H /. This is a consequence of the fact that every vertex of
G is adjacent to every vertex of H in G C H . (a) Ia .G [ H / D Ia .G C H / D
Pk
ni ni P
k
mi mi 1
P
k Pk
 nCm
log. nCm / nCm
log. nCm / D nCm Πni log.nCm/C mi log.nC
i D1 i D1 i D1 i D1
1 P
k P
k
1
m/ C nCm
Πni log.ni /  mi log.mi / D log.n C m/ C nCm
ŒnIa .G/ C
i D1 i D1
mIa .H /  n log.n/  m log.m/, as required. (b) See [18]. t
u
When the two digraphs are of equal size, the information content of their join is
just one more than their average information content.
Corollary 1. Let G and H be as in the Theorem. If n D m, then Ia .G [ H / D
Ia .G C H / D 12 ŒIa .G/ C Ia .H / C 1.
Proof. The result follows immediately from the Theorem by setting m D n in the
expression for Ia .G [ H / D Ia .G C H /. t
u
6 A. Mowshowitz

Y
x

Fig. 1.3 Partial joins

Corollary 2. Let G and H be as in the Theorem and suppose Ia .G/ D Ia .H /.

1
Then (i) Ia .G [ H / D Ia .G C H / D Ia .G/ C log.n C m/  nCm Œn log.n/ C
m log.m/, and (ii) if in addition n D m, Ia .G [ H / D Ia .G C H / D Ia .G/ D
Ia .G/ C 1.
Of particular importance to the representation of real network growth is the
partial join operation.
Definition 12. A partial join of G1 and G2 for the set F is the digraph G1 ˚ G2
defined by V .G1 ˚G2 / D V .G1 /[V .G2 / and E.G1 ˚G2 / D E.G1 /[E.G2 /[F ,
where F  fŒu; vju 2 V .G1 /; v 2 V .G2 /g.
Figure 1.3 illustrates partial join operations for different sets F .
The information content of a partial join depends on the set F . For example, if
both graphs G and H are isomorphic to the directed cycle of length n and G ˚ H
is defined for set F consisting of a single undirected edge, Ia .G ˚ H / D log.n/
since each orbit of Aut.G ˚ H / consists of two of the 2n vertices. If there are two
edges joining G and H , one of which does not join corresponding vertices of the
directed n-cycles, the information content is log.2n/ since Aut.G ˚H / is the trivial
group in this case. Note that Ia .G/ D Ia .H / D 0 since the automorphism group of
a directed cycle with n vertices is the cyclic group of order n. Thus, it appears that
Ia .G ˚ H / can be expressed in terms of Ia .G/ and Ia .H / in special cases only.
Theorem 2. Let G1 and G2 be complete graphs with m and n vertices, respectively,
and suppose G D G1 ˚ G2 is a partial join with jF j D 1.
m1
 m1  n1
 m1 
(a) If m is different from n, Ia .G/ D mCn log mCn C mCn log mCn C
2
mCn
log.m C n/
(b) If m D n, then

1h  n  i
Ia .G/ D .n  1/ log C log.n/ :
n n1

Proof. Let Œx; y be the edge in F where x is in G1 and y is in G2 . If m is different

from n, the partial join G has four orbits A, B, C , and D, where A consists of the
m  1 vertices of G1 excluding x, B consists of the n  1 vertices of G2 excluding
y, and C and D are singletons containing x and y, respectively. If m D n there are
two orbits with 2 and 2.n  1/ vertices, respectively. t
u
1 Entropy of Digraphs and Infinite Networks 7

2.4 Cartesian Product and Composition

Theorem 3 ([18]). (a) Ia .G H /  Ia.G/CIa.H / for any digraphs G and H .

(b) Equality holds when G and H are weakly connected and relatively prime with
respect to the Cartesian product.
Proof. Part (a) follows from the fact that Aut.G  H / is a subgroup Aut.G/ 
Aut.H /. Part (b) is a consequence of the fact that Aut.G  H / is isomorphic to
Aut.G/  Aut.H / if and only if digraphs G and H are relatively prime with respect
to the Cartesian product. Note that being relatively prime is a sufficient but not a
necessary condition for equality in the theorem. t
u
The information content measure is also sub-additive for the composition
operation.
Theorem 4 ([18]). Ia .G  H /  Ia.G/ C Ia.H / for any digraphs G and H .
Figure 1.4 provides examples of the information content of the Cartesian product
and composition.

2.5 Existence Theorem

The join and Cartesian product can be used to construct digraphs with given
information content. More precisely, for any finite probability scheme there exists a
digraph with information content equal to the entropy of the scheme. This result is
stated in the following theorem originally presented in [18].
Theorem 5. Let n be any positive integer, and suppose P D fnij g is a partition of
n where nij D ni (1  j  ri /, ni1 ¤ ni2 (i1 ¤ i2 ), and i D 1; 2;    ; k. Then there
exists a weakly connected digraph G with n vertices such that Aut.G/ has exactly
Pk
rD ri orbits, and for each nij there is an orbit A with jAj D nij ; and, hence,
i D1

k
X ni n 
i
Ia .G/ D H.P / D  ri log :
i D1
n n

Proof. The proof is based on a simple construction. Let Gi D Lri 1  Cni where
Lri 1 is a directed path of length ri  1 and Cni is a directed cycle of length ni .
Since the path and cycle are relatively prime with respect to the Cartesian product,
the orbits of Aut.Gi / are the respective products of the orbits of Aut.Lri 1 / and
Aut.Cni /. Hence, Aut.Gi / has exactly ri orbits, each consisting of ni elements.
The digraph G formed by taking the join of the k non-isomorphic Gi has an
automorphism group with orbits corresponding to the partition specified in the
hypothesis of the theorem, and thus has the required information content. t
u
Figure 1.5 illustrates the Theorem for n D 25; P D f13 ; 24 ; 32 ; 42 g.
8 A. Mowshowitz

G H G’ H’ H G
Ia (G) = 0; Ia (H) = log 3 Ia (G’) = Ia (H’) Ia (H) = log 3; Ia (G) = 0;
= log 3 − 2/3

GXH G’ X H’ HoG
Ia (G X H) Ia (G’ X H’) Ia (G o H)
= Ia (G) + Ia(H) = 2 log 3 − 16/9 = Ia (H) + Ia (G)
= log 3 < Ia (G’) + Ia (H’) = log 3
= 2 log 3 − 12/9

Fig. 1.4 Information content of Cartesian product and composition

+ + +

L2 X C1 L3 X C2 L1 X C3 L1 X C4

Fig. 1.5 Construction of digraph with prescribed information content

1 Entropy of Digraphs and Infinite Networks 9

3 Entropy of Infinite Graphs

3.1 Preliminaries

Definition 13. A graph G D .V; E/ is countable if jV [ Ej is countable. G is

locally finite if the degree of every vertex of G is finite.
In what follows, we will restrict attention to countable graphs that may or may
not be locally finite.
Definition 14 ([18]). Let G D .V; E/ be a countable graph. A sequence fGn g1 nD1
of finite graphs Gn with Vn D V .Gn / and En D E.Gn / is said to converge to G as a
limit (written limn!1 Gn D G) if limn!1 Vn D V .G/ and limn!1 En D E.G/.
Note that both V and E are simply the limits of sequences of sets.
Definition 15 ([18]). A sequence fGn g1 nD1 of finite graphs Gn is a defining
sequence for a countable graph G if Gn  GnC1 for every n, and limn!1 Gn D G.
Since the limit of any monotonically increasing sequence fAn g1 nD1 of sets An exists
S1
and is equal to An , every countable graph G has a defining sequence.
nD1

A defining sequence for a countable graph G with V .G/ D fv1 ; v2 ; v3 ;    g can

be constructed as follows:
V .G1 D fv1 g and E.G1 / D ;,
V .GnC1 / D V .Gn / [ fvnC1g and E.GnC1 / D E.Gn / [ fŒvnC1; u 2 E.G/ju 2
V .Gn /g.
Definition 16 ([18]). Let fGn g1
nD1 be a defining sequence for a countable graph
G. The information content IO.GI Gn / of G with respect to the sequence fGn g1
nD1 is
given by IO.GI Gn // D limn!1 Ia .Gn / if the limit exists.
Figure 1.6 shows a countable graph with defining sequences that give rise to
different information content values.

, , ,

G1 G2 G3

...
,

G4

Fig. 1.6 A countable graph with more than one defining sequence
10 A. Mowshowitz

n
0 if n is odd
ia .Gn / D log.5/ 35 log.3/ 25 if n is even
Thus, for the subsequence Sn consisting
of the odd terms, IO.GI Sn / D 0; and for the subsequence Tn consisting of the even
terms, IO.GI Tn / D log.5/  35 log.3/  25 . The difference in this case is finite,
but it could be infinite as shown in [18]. Using a measure that depends on the
graph’s defining sequence is not necessarily a disadvantage. An infinite graph can
be viewed as an idealization of a growth process. Including the defining sequence
in the definition allows for capturing different principles of growth in practice.

3.2 Classes of Infinite Graphs

Infinite graphs can be built up recursively with the aid of graph products. The
following result makes use of the Cartesian product.
Lemma 1. Let G be a graph with n vertices. Ia .G  K2 / D Ia .G/.
Proof. Corresponding vertices of the two copies of G are in the same orbit of G 
K2 , so G and G  K2 have the same number of orbits, and each orbit of G  K2
has exactly double the number of vertices as the corresponding orbit of G. Thus, if
Aut.G/ has orbits A1 ; A2 ;    ; Ar with jA.i /j D ki ; 1  i  r, then Ia .G  K2 / D
P r  
2ki
 2n
log 2k
2n
i
D Ia G. t
u
i D1

Suppose G is a graph with n vertices. If Aut.G/ is the identity group, then

Ia .G/ D log.n/, and Ia .G  K2 / D log.n/. The sequence

H1 D G;
HnC1 D Hn  K2 ; for n  1

serves as a defining sequence of an infinite graph. Since Ia .Hn / D log.n/,

limn!1 Ia .Hn / D 1.
At the other extreme is the hypercube Hn , which can be defined recursively as
follows:
H1 D K2 ;
HnC1 D Hn  K2 ; for n  1:

Since the limit of the (defining) sequence fHn g1 nD1 exists, we can set H1 D
limn!1 Hn . Now, Ia .Hn / D 0 for all n  1 which implies by the lemma
that IO.H I Hn / = 0, i.e., the sequence of finite hypercubes yields a limit whose
information content is zero. The hypercube serves as a useful model in parallel
computation. A key feature in this context is the favorable maximum distance
between any two vertices in the graph. This allows for placing computational
units so as to minimize communication costs. The zero information content of the
1 Entropy of Digraphs and Infinite Networks 11

S3 X K2 W3 X K2

Fig. 1.7 Star and wheel products

hypercube reflects the high degree of symmetry of this graph, which allows for
simultaneous placement of elements at optimal distance from each other.
Other graphs of interest, with information content between the two extremes, can
be substituted for G in G  K2 .
Let S k denote the star of order k, a connected graph with one vertex of degree
k  1 and k  1 vertices of degree 1; and let W k denote the wheel of order k, a
connected graph obtained from the star by joining the degree 1 vertices in a cycle of
length k1. Once again using the Cartesian product, we can build infinite sequences
based on these simple graphs.

S1k D S k ;
k
SnC1 D Snk  K2 ; for n  1:

A sequence of graphs Wnk can be defined

 k1  similarly.
 
Ia .S k / D Ia .W k / D  k1
k
log k
 k1 log k1 D log.k/  k1
k
log.k  1/.
Denoting by S1 and W1 , respectively, the infinite graphs with defining sequences
fSnk g1 k 1 O k O k k1
nD1 and fWn gnD1 , we have I .S1 I Sn / D I .W1 I Wn / D log.k/  k
log.k  1/.
As k increases, almost all the vertices fall into one orbit and the information
content tends to zero.
Figure 1.7 shows the Cartesian products, respectively, of the star and the wheel
with K2 .
The information content of the line graph of order k is given by:
( k
k
log 2 if n is even
Ia .L / D k 
k1 1
k
log 2
C k
log k if n is odd:

The information content of the line graph increases without bound, so the informa-
tion content of the limit graph is infinite.
The cycle graph of order k has information content Ia .C k / D 0, so the limit
graph in this case has information content zero.
12 A. Mowshowitz

More complex graphs could be constructed by substituting for K2 in the

Cartesian products defining the terms in the infinite sequences considered above.

4 Applications

Preferential attachment has been studied extensively as a protocol for the growth of
large-scale networks like the Internet [5]. According to this protocol, a vertex added
to a network will be more likely to become attached to existing vertices of higher
rather than of lower degree. The “preference” of a vertex v as a target of attachment
might be expressed as the probability given by the degree of v divided by the sum of
the degrees in the graph. This introduces a random element in the growth process.
Perhaps the simplest way to realize a (relatively deterministic) version of growth by
preferential attachment is to add a single new vertex at each iteration, connecting the
new vertex to an existing one whose degree is maximal in the current graph. Call
this a type-0 preferential attachment protocol. If the starting graph is K1 , the result
is clearly a star. After the nth new vertex has been added, a star of order n C 1 has
n
been formed. This graph S nC1 has information content log.n C 1/  nC1 log.n/, and
as noted above, this value tends to zero as n increases without bound.
A variation on this simple protocol is to add k new vertices at each iteration and
attach each one of them to a different existing vertex, choosing the existing vertices
in nonincreasing order of degree, beginning with one of maximal degree.
Figure 1.8 illustrates the construction process according to this protocol, and the
following theorem gives the information content in the case where k equals the
number of vertices in the initial graph of the sequence.
Theorem 6. Let fGnk g1
1 be a sequence of graphs defined as follows:

G1k D S k

k
GnC1 is obtained from Gnk by adding k new vertices and joining each one to a
different vertex of maximal degree in Gnk .

G41 = S5 G42 G43

Fig. 1.8 A graph constructed with preferential attachment protocol type-0

1 Entropy of Digraphs and Infinite Networks 13

H21 = K2 H22 H23

Fig. 1.9 A graph constructed with preferential attachment protocol type-1

The information content of Gnk is given by:

   
.k  1/.n  1/ nk C 1 k nk C 1
Ia .Gnk / D log C log
nk C 1 .k  1/.n  1/ nk C 1 k
 
n1 nk C 1 1
C log C log.nk C 1/:
nk C 1 n1 nk C 1

Proof. Since k vertices are added for each iteration, Gnk , the nth graph in the
sequence has nk C 1 vertices. Let v be the vertex of highest degree in Gnk . The
orbits of Aut.Gnk / consist of the vertex v alone, the vertices of degree 1 adjacent to
v, the vertices of degree > 1 adjacent to v, and the vertices of degree 1 at distance 2
from v. Thus, the orbits of Aut.Gnk / have 1, n  1, k, and .k  1/.n  1/ vertices
from which the result follows. t
u
Corollary 3. Let Gnk be defined as in the theorem. Then IO.GI Gnk / D log k.
Proof. Simplifying the expression in the theorem gives Ia .Gnk / D log.nk C 1/ 
1
nkC1
Œ.k  1/.n  1/ log.k  1/.n  1/ C .n  1/ log.n  1/ C k log k. Taking the
limit as n ! 1 yields IO.GI Gnk / D log k as required. t
u
Greater connectivity in a network that grows by preferential attachment can
be achieved by allowing the newly added vertices to be joined to more than one
existing vertex [2]. Call this a type-1 preferential attachment protocol. This protocol
is illustrated in Fig. 1.9. The following theorem gives the information content of an
infinite graph that grows according to a type-1 protocol with k D 2.
Theorem 7. Let fHn2 g1
1 be a sequence of graphs defined as follows:

H12 D K2

2
HnC1 is obtained from Hn2 by adding 2 new vertices and joining each one to exactly
two different vertex of maximal degree in Hn2 .
14 A. Mowshowitz

(a) The information content of Hn2 is given by:

   
n nC1 1 nC1
Ia .Hn2 / D log C log :
nC1 n nC1 1

(b) IO.H I Hn2 / D 0

Proof. Every new vertex is joined to the two vertices of K2 ; so Aut.Hn2 / has exactly
two orbits consisting of 2 and 2n vertices, respectively. Part (b) follows immediately
n
from the fact that nC1 and nC1
n
1
tend to 1; and nC1 log.n C 1/ tends to 0. t
u
More realistic protocols call for making estimates of information content. If, for
example, each of the k new vertices added at a given step was joined to an existing
vertex v with probability equal to degree(v)/sum of degrees in the current graph,
one would try to obtain upper and lower bounds on the information content of the
infinite graph.

5 Summary and Conclusion

In this chapter, we have investigated the information content of directed graphs

and infinite graphs. We have determined the information content of various graph
products, computed the information content of some special classes of infinite
graphs, and investigated the information content of infinite graphs whose growth
is governed by preferential attachment protocols.
Digraphs and infinite graphs offer useful tools for modeling networks. In
particular, the information content of digraphs and infinite graphs may be useful
for classifying networks according to critical features such as diameter and vulner-
ability. The relationship between information content and these features is an open
question. For example, the graphs in Fig. 1.10 differ by the maximum possible on the

1 1

6 2 6 2

5 3 5 3

4 4
G H

Fig. 1.10 Information Ia (G) = log 6 Ia (H) = 0

content and diameter diameter G = 3 diameter H = 3
1 Entropy of Digraphs and Infinite Networks 15

measure of information content, but both have diameter 3 and for both the minimum
number of vertices whose removal disconnects the respective graphs is 2.
This example suggests that there is no simple correlation between information
content, on one hand, and diameter and vulnerability, on the other hand. Two graphs
with the same diameter can register as far apart as possible on the information
content scale; similarly, two graphs with the same (vertex) vulnerability rating can
be far apart on the information content measure. However, it is possible that graphs
that grow in a certain way (like the hypercube) may exhibit a high correlation
between diameter and information content.
Information content could possibly be correlated with diameter if only graphs
with similar distance properties were grouped together. In the example, the max-
imum distances from each vertex (from 1 to 6) to any other in G is given by
.2; 2; 3; 2; 2; 3/, whereas for H these maximum distances are .3; 3; 3; 3; 3; 3/. Al-
though the two graphs have the same diameter, the sequences of maximum distance
do not coincide. Another potentially useful discriminant is the metrical property
distance degree sequence [4]. This sequence gives the number of vertices at each
succeeding distance from a given vertex. Graph H in Fig. 1.10 is distance degree
regular but G is not. The hypercube is distance degree regular and has 0 information
content; however, a distance degree regular graph does not necessarily have a
transitive automorphism group. The more general metrical properties investigated in
[22] might prove useful in determining the relationship between information content
and diameter.
Another open question is how to define the information content of an infinite
graph independently of a defining sequence of finite graphs. Investigation of this
question could shed light on the properties of very large-scale networks.

Acknowledgments Research was sponsored by the US Army Research Laboratory and the UK
Ministry of Defence and was accomplished under Agreement Number W911NF-06-3-0001. The
views and conclusions contained in this document are those of the authors and should not be
interpreted as representing the official policies, either expressed or implied, of the US Army
Research Laboratory, the US Government, the UK Ministry of Defence, or the UK Government.
The US and UK Governments are authorized to reproduce and distribute reprints for Government
purposes notwithstanding any copyright notation hereon.

References

1. Balaban, A.T., Balaban, T.S.: New vertex invariants and topological indices of chemical graphs
based on information on distances. J. Math. Chem. 8, 383–397 (1991)
2. Bent, G., Dantressangle, P., Vyvyan, D., Mowshowitz, A., Mitsou, V.: A dynamic distributed
federated database. Proceedings of the Second Annual Conference of the International
Technology Alliance, Imperial College, London, September 2008
3. Bertz, S.H.: The first general index of molecular complexity. J. Am. Chem. Soc. 103,
3241–3243 (1981)
4. Bloom, G., Kennedy, J.W., Quintas, L.V.: Distance degree regular graphs. In: Chartrand, G.
(ed.) The Theory of Applications of Graphs, pp. 95–108. Wiley, New York (1981)
16 A. Mowshowitz

5. Bollobás, B., Riordan, O.: The diameter of a scalefree random graph. Combinatorika 24, 5–34
(2004)
6. Bonchev, D.: Information Theoretic Indices for Characterization of Chemical Structures.
Research Studies Press, Chichester, UK (1983)
7. Bonchev, D.: Complexity in Chemistry, Biology, and Ecology. Mathematical and Computa-
tional Chemistry series. Springer, New York (2005)
8. Bonchev, D., Trinajstic, N.: Information theory, distance matrix and molecular branching. J.
Chem. Phys. 67, 4517–4533 (1977)
9. Dehmer, M., Emmert-Streib, F.: Structural information content of chemical networks.
Zeitschrift für Naturforschung A 63a, 155–159 (2008)
10. Dehmer, M., Emmert-Streib, F., Mehler, A. (eds.): Towards an Information Theory of Complex
Networks: Statistical Methods and Applications. Springer/Birkhäuser, Berlin (2011)
11. Dehmer, M., Mowshowitz, A.: Inequalities for entropy-based measures of network information
content. Appl. Math. Comput. 215, 4263–4271 (2010)
12. Harary, F.: Graph Theory. Addison Wesley, Reading, MA (1969)
13. Hirata, H., Ulanowicz, R.E.: Information theoretical analysis of ecological networks. Int. J.
Syst. Sci. 15, 261–270 (1984)
14. Khinchin, A.I.: Mathematical Foundations of Information Theory. Dover Publications, New
York (1957)
15. Konstantinova, E.V., Skorobogatov, V.A., Vidyuk, M.V.: Applications of information theory in
chemical graph theory. Indian J. Chem. 42, 1227–1240 (2002)
16. Körner, J.: Coding of an information source having ambiguous alphabet and the entropy of
graphs. In: Transactions of the 6th Prague Conference on Information Theory, 411–425 (1973)
17. Mowshowitz, A.: Entropy and the complexity of graphs: I. An index of the relative complexity
of a graph. Bull. Math. Biophys. 30, 175–204 (1968)
18. Mowshowitz, A.: Entropy and the complexity of graphs: II. The information content of
digraphs and infinite graphs. Bull. Math. Biophys. 30, 225–240 (1968)
19. Mowshowitz, A., Mitsou, V., Bent, G.: Models of network growth by combination. Proceedings
of the Second Annual Conference of the International Technology Alliance, Imperial College,
London, September 2008
20. Mowshowitz, A., Mitsou, V.: Entropy, orbits and spectra of graphs. In: Dehmer, M. (ed.)
Analysis of Complex Networks: From Biology to Linguistics. Wiley-VCH, Weinheim (2009,
in press)
21. Rashevsky, N.: Life, information theory, and topology. Bull. Math. Biophys. 17, 229–235
(1955)
22. Skorobogatov, V.A., Dobrynin, A.A.: Metric analysis of graphs. MATCH 23, 105–151 (1988)
23. Sole, R.V., Valverde, S.: Information theory of complex networks: On evolution and architec-
tural constraints. Lect. Notes Phys. 650, 189–207 (2004)
24. Wasserman, S., Faust, K.: Social Network Analysis: Methods and Applications, Structural
Analysis in the Social Sciences. Cambridge University Press, Cambridge (1994)
Chapter 2
An Information-Theoretic Upper Bound
on Planar Graphs Using Well-Orderly Maps

Nicolas Bonichon, Cyril Gavoille, and Nicolas Hanusse

Abstract This chapter deals with compressed coding of graphs. We focus on planar
graphs, a widely studied class of graphs. A planar graph is a graph that admits an
embedding in the plane without edge crossings. Planar maps (class of embeddings
of a planar graph) are easier to study than planar graphs, but as a planar graph may
admit an exponential number of maps, they give little information on graphs. In
order to give an information-theoretic upper bound on planar graphs, we introduce
a definition of a quasi-canonical embedding for planar graphs: well-orderly maps.
This appears to be an useful tool to study and encode planar graphs. We present
upper bounds on the number of unlabeled1 planar graphs and on the number of
edges in a random planar graph. We also present an algorithm to compute well-
orderly maps and implying an efficient coding of planar graphs.

Keywords Compact coding • Enumerative combinatorics • Planar embedding •

Planar graph

MSC2000 Primary 05C10; Secondary 05C10, 05C30, 05C85.

1 Introduction

In graph theory, a planar graph is a graph which can be embedded in the plane,
i.e., it can be drawn on the plane in such a way that its edges intersect only at their
endpoints. A planar graph drawn in the plane without edge intersections is called a

1
Nodes and edges are not assumed to be labeled.
N. Bonichon ()
LaBRI, University of Bordeaux, 351 Cours de la libération, 33405 Bordeaux, France
e-mail: bonichon@labri.fr

M. Dehmer et al. (eds.), Towards an Information Theory of Complex Networks: 17

Statistical Methods and Applications, DOI 10.1007/978-0-8176-4904-3 2,
© Springer Science+Business Media, LLC 2011
18 N. Bonichon et al.

planar map or a planar embedding of the graph. The class of planar graphs is one of
the most studied graphs.
How much information can contain a simple planar graph of n nodes? The
question is highly related to the number of planar graphs. Counting the number
of (non-isomorphic) planar graphs with n nodes is a well-known and long-standing
unsolved graph-enumeration problem (cf. [24]). There is no known close formula,
neither asymptotic nor even an asymptotic on the logarithm of this number. Any
asymptotic on the logarithm would give a bound on the number of independent
random bits needed to generate a planar graph uniformly at random (but not
necessary in polynomial time).
Random combinatorial object generation is an important activity regarding
average case complexity analysis of algorithms and testing algorithms on typical in-
stances. Unlike random graphs (the Erdös–Rényi graph Model), still little is known
about random planar graphs. Indeed adding an edge in a planar graph highly depends
on the location of all previous edges. Random planar maps, i.e., plane embeddings
of planar graphs, have been investigated more successfully. Schaeffer [35] and then
Banderier et al. [2] have showed how to generate in polynomial time several planar
map families, e.g., 3-connected planar maps. Unfortunately, this generating does
not give much information about random planar graphs because there are many
ways to embed a planar graph into the plane. On the positive side, some families of
planar graphs support efficient random generation: trees [1], maximal outerplanar
graphs [3, 14], and more recently labeled and unlabeled outerplanar graphs [4].
Besides the combinatorial aspect and random generation, an important attention
is given in Computer Science to efficiently represent discrete objects. Efficiently
means that the representation is succinct, i.e., the storage of these objects uses few
bits, and that the time to compute such representation is polynomial in their size.
Fast manipulation of the so-encoded objects and easy access to a part of the code
are also desirable properties. At least two scopes of applications of high interests are
concerned with planar graph representation: Computer Graphics and Networking.
Surface discretization of a 3D object outputs a list of 3D coordinates and a set
of adjacency relations. In the case of convex objects, the set of adjacency relations
is an unlabeled planar graph. In general, small degree faces are used for surface
discretization, with triangle or quad meshes. Then, a compressor is applied on the
planar graph. Performances are expressed averaging the number of bits per edge or
per node. They are evaluated among a benchmark of standard examples [21], due to
the lack of “good” random planar graph generator, or typical instance generator.
For example, King and Rossignac [22, 34] gave a triangulation compressor that
guarantees 3:67 bits per node, the best possible rate being log2 .256=27/  3:24
bits per node from Tutte’s enumerative formula [39].
Routing table design for a network has been investigated in the case of planar
networks [15, 16, 26, 37]. The underlying graph of the network is preprocessed to
optimize routing tables, a data structure dedicated to each node in charge of finding
the next output port given the destination address of an incoming message. The
main objective is to minimize the size of the routing tables while maintaining routes
as short as possible. The strategy used by Gavoille and Hanusse [16] based on a
2 An Information-Theoretic Upper Bound on Planar Graphs 19

k-page embedding, and then improved by Lu [26] with orderly spanning trees,
demonstrates that a compact planar graph representation helps for the design of
compact routing tables, especially when shortest paths are required.

1.1 Related Works

Succinct representation of n-node m-edge planar graphs has a long history.

Turán [38] pioneered a 4m bit encoding, which has been improved later by Keeler
and Westbrook [20] to 3:58m. Munro and Raman [29] then proposed a 2m C 8n bit
encoding based on the 4-page embedding of planar graphs (see [40]). In a series of
articles, Lu et al. [8,11] refined the coding to 4m=3 C5n, thanks to orderly spanning
trees, a generalization of Schnyder’s trees [36]. Independently, codings have been
proposed for triangulations, where m D 3n6. A 4n bit encoding has been obtained
by several authors [5,11,34], interestingly with rather different techniques, and then
improved by the Rossignac’s Edgebreaker [22], who guaranteed 3:67n bits for
triangulations and computable in O.n/ time. Actually, He et al. [19] showed that,
in O.n log n/ time, a space optimal encoding for triangulations and for unlabeled
planar graphs can be achieved. Hence, a O.n log n/ time and a 3:24n bit encoder
for triangulations exist. For that, they use a recursive separator decomposition of
the graph, and an exponential coding algorithm for the very end components of sub-
logarithmic size. However, the time complexity hidden in the big-O notation could
be of limited use in practice. To implement the encoder, one needs, for instance,
to implement planar isomorphism and Lipton–Tarjan planar separator [23]. The
time complexity has been recently improved to O.n/ for planar graphs by Lu [27].
Although the length of the coding is optimal, the approach of [19, 27] does not give
any explicit bound of the number of bits used in the representation.
If we are interested only in the information-theoretic bound of planar graphs or
in statistical properties of planar graphs (what a random planar graph looks like:
number of edges, connectivity, etc.), other tools can be used. Denise et al. [12]
specified a Markov chain on the space of all labeled planar graphs whose limit
distribution is the uniform distribution. Their experiments show that random planar
graphs have approximately 2n edges, and are connected but not 2-connected.
Although the Markov chain converges to the uniform distribution, it is not proved
whether this Markov chain becomes close enough to the uniform distribution after a
polynomial number of steps. It is, however, proved that almost all labeled planar
graphs have at least 1:5n edges, and that the number p.n/ of unlabeled planar
graphs satisfies that n1 log2 p.n/ tends to a constant  such that log2 .256=27/ 6
 6 log2 .256=27/ C 3. The bounds on  easily derive from Tutte’s formula [39]:
Triangulations are planar graphs, and every planar graph is a subgraph of a
triangulation, thus having 23n6 possible subsets of edges. There are also no more
than nŠ2 nCo.n/ labeled planar graphs as there are at most nŠ ways to label the nodes
of a graph.
20 N. Bonichon et al.

Osthus et al. [31] investigated triangulations containing any planar graph, and
they showed that there is no more than nŠ25:22nCo.n/ labeled planar graphs. Osthus
et al. [31] also showed that almost all labeled planar graphs have at most 2:56n
edges, and that almost all unlabeled planar graphs have at most 2:69n edges.
A lower bound of 13n=7  1:85n has been obtained by Gerke and McDiarmid [17],
improving the 1:5n lower bound of the expected number of edges of [12]. Properties
of random planar graphs have also been investigated in [28].
Gimenez and Noy [18] show that the number of edges of a random labeled planar
graph is asymptotically normal and the mean is 2:213n and variance is 0:4303n.
Unlike general graphs, labeled and unlabeled planar graphs do not have the same
growing rate (up to the nŠ term) as proved in [28]. So upper bounds on labeled
planar graphs do not transfer to upper bounds on unlabeled planar graphs, but the
reverse is true.
Using generating function techniques, Noy and Gimenez [18] proved that the
number of labeled connected planar graphs tends to nŠ24:767nCO.log n/ . The number of
simple planar maps is asymptotic to 25:098nCO.log n/ (cf. algebraic generating function
presented in [25]) providing an upper bound for unlabeled planar graphs.

1.2 Presented Results

In this chapter, we present a new representation of planar graphs called well-orderly

maps. Starting from a planar graph, we show how to build and encode a well-orderly
map in linear time. Our construction leads to counting results about planar graphs.
More precisely, we show an upper bound of 24:91nCo.n/ on p.n/, the number of
unlabeled planar graphs with n nodes.
Since our upper bound can be parameterized with the number of edges, and
using the lower bound of [18], we are able to show that almost all unlabeled graphs
have at least 1:85n edges and at most 2:44n edges, setting a new lower bound and
improving the 2:69n upper bound of [31]. The presented results are a synthesis of
results presented in [6, 7].

1.3 Outline of the Chapter

Let us sketch our technique. Since the number of useful combinatorial objects are
numerous, we first briefly describe in Fig. 2.1 the different steps toward the compact
coding of planar graphs. Our real starting point is a planar map, sometimes called
planar embedding. To get a planar map from a planar graph, well-known linear time
algorithm can be used (see for instance [9]). Roughly speaking, we first present a
very particular embedding of a planar graph called well-orderly map and show how
to encode it using combinatorial tools like bijective combinatorics and a specific
compression technic of binary strings.
2 An Information-Theoretic Upper Bound on Planar Graphs 21

Planar graph
6 compressed binary strings

Planar map

5+1=6 binary strings

Well−orderly map

Super−triangulation
= Minimal Realizer Balanced trees with n−2 inner vertices
= 3 well−orderly trees

GRAPH THEORY AND GRAPH ALGORITHMIC COMBINATORICS AND BINARY TEXT COMPRESSION

Fig. 2.1 Roadmap toward the compact coding of a planar graph

A First Upper Bound Based on Triangulation

A natural approach to represent an n-node planar graph G is to consider a
triangulation of G, i.e., a supergraph S of G such that S is planar, has n nodes
and 3n  6 edges. Then, G can be obtained by coding S and a set MS of edges such
that E.S / n MS D E.G/. This way of representing a planar graph is suggested by
the .log2 .256=27/ C 3/n D 6:24n bit upper bound of [12] mentioned above.

Introduction of Well-Orderly Maps and Super-Triangulation

To obtain a representation more compact than 6:24n bits, we need to carefully
construct S . In particular, crucial steps are the way we embed G into the plane,
and the way we triangulate its faces. In Sect. 2.2, we introduce a specific embedding
of G called a well-orderly map, and we show that it can be computed in linear
time. Given a well-orderly map, we present how to build the supergraph S of G,
called hereafter super-triangulation and defined in Sect. 2. More precisely, a super-
triangulation S has the property that for a given node v 2 S only, one can perform
in a unique manner a traversal of S by following a specific spanning tree T rooted in
v, called a well-orderly tree, such that T is contained in G. Hence, given the super-
triangulation S of G, MS is of cardinality at most .3n  6/  .n  1/ D 2n  5, and
the edges of MS can be described among the possible edges of S n T only, i.e., with
at most 2n bits. This already provides a .log2 .256=27/ C 2/n D 5:24n bit upper
bound. Observe that the case G not connected can be easily transformed (in linear
time) into a new connected graph G, Q e.g., by linking all the connected components
of G into a single node (see Sect. 4 for more details).

Using Minimal Realizer Properties

The next step consists in encoding in a very compact way the super-triangulation.
In Sect. 3.1, we represent the super-triangulation S by a realizer, that is a partition
22 N. Bonichon et al.

of the edges into three trees .T0 ; T1 ; T2 / (see Schnyder’s trees [36]). In our case, the
partition has specific properties and corresponds to minimal realizer. We also show
how to uniquely recover the three trees from such a super-triangulation. Different
properties of minimal realizer are useful since the knowledge of two well-orderly
trees implies a canonical description of the third one and can exploited to save bits.
At this point, the following properties are only given as an illustration:
• Every edge .u; v/ of S such that (1) u is the parent of v in T1 and (2) u is an inner
node in T2 , must belong to G. This significantly saves bits in the coding of MS
since many edges of G can be guessed from S .
• An extra property is that two nodes belonging to the same branch of T2 have
the same parent in T1 (a branch is a maximal set of related nodes obtained
in a clockwise depth-first search of the tree, and such that a node belongs to
only one branch at the time, see Sect. 3). This latter property simplifies a lot the
representation of S . Knowing T2 , T1 does not need to be fully represented. Only
one relevant edge per branch of T2 is enough. As any tree of a realizer can be
deduced from the two others, the representation of S can be compacted in a very
efficient way, storing for instance T2 and the relevant edges of T1 .
Combining such properties and the optimal coding of realizer using the bijection of
Poulalhon and Schaeffer [33] (see also Theorem 2), we get an encoding of super-
triangulation presented in [7].

Compact Coding in Binary Strings

Finally, we show in Sect. 3 that the explicit representation of G is done with six
binary strings of different density (namely the ratio between the number of “ones”
it contains and its length): five for S and one for MS . We compact each string with
a variant of the Pagh’s compressor [32]. This allows to reach an optimal entropy
coding, i.e., with log2 kn Co.n/ bits for an n-bit string of k ones.2 If we parametrize
the number of branches of T2 (or equivalently its number of leaves), an entropy
analysis shows that 4:91n bits are enough to represent G.

2 Embedding and Triangulating Algorithms

A plane embedding of a graph, or shortly a plane graph, is a mapping of each node

to a point of the plane and of each edge to the continuous curve joining the two ends
of this edge such that edges do not cross except, possibly, on a common extremity.
A graph that has a plane embedding is a planar graph.

2
The original compressor runs in expected linear time. We give in this chapter a simpler guaranteed
linear time construction with asymptotically the same performances.
2 An Information-Theoretic Upper Bound on Planar Graphs 23

Fig. 2.2 Relationship a b BC

between realizer and orderly
tree: (a) edge-orientation rule T2 T1
around a node for a realizer,
and (b) blocks ordering
around an orderly node (T is B< B>
represented by directed edges
because the edge .v; w/ of T1 )

T0 BP

In this chapter, we deal with simple (no loops and no multi-edges) and undirected
graphs. If we cut the plane along the edges, the remainder falls into connected
regions of the plane, called faces. Each plane graph has a unique unbounded face,
called the outerface. The boundary of a face is the set of incident edges. The interior
edges are the edges non-incident to the boundary of the outerface, similarly for
interior nodes. Precise definitions can be founded for instance in [13, 30].
A triangulation is a plane embedding of a maximal planar graph, that is a planar
graph with n nodes and 3n  6 edges. There is only one way to embed in the plane
(up to a continuous transformation), a maximal planar graph whose three nodes are
chosen to lie on the outerface.

2.1 Well-Orderly Tree, Realizer and Super-Triangulation

Let T be a rooted spanning tree of a plane graph H . Two nodes are unrelated if
neither of them is an ancestor of the other in T . An edge of H is unrelated if its
endpoints are unrelated.
We introduce well-orderly trees, a special case of orderly spanning trees of
Chiang, Lin, and Lu in [8], referred as simply orderly trees later. Let v1 ; : : : ; vn be
the clockwise preordering of the nodes in T (nodes ordered by their first visit in a
clockwise traversal of the tree T ). Recall that a node vi is orderly in H with respect
to T if the incident edges of vi in H form the following four blocks (possibly empty
set of vertices) in clockwise order around vi (see Fig. 2.2b):
• BP .vi /: the edge incident to the parent of vi
• B< .vi /: unrelated edges incident to nodes vj with j < i
• BC .vi /: edges incident to the children of vi
• B> .vi /: unrelated edges incident to nodes vj with j > i
A node vi is well orderly in H with respect to T if it is orderly, and if:
• The clockwise first edge .vi ; vj / 2 B> .vi /, if it exists, verifies that the parent of
vj is an ancestor of vi (in T ).
In other words, if .vi ; vj / the first edge of B> .vi /, then the parent of vj is an
ancestor of vi in T .
24 N. Bonichon et al.

a r1 b r1

w w
r2 u r2 u
t t

v v

r0 r0

Fig. 2.3 Two realizers for a triangulation. The tree T 0 rooted in r0 (the tree with bold edges
augmented with the edges .r0 ; r1 / and .r0 ; r2 /) is well orderly in (b), and simply orderly in (a) (the
node v is not well orderly: .v; w/ is the clockwise first edge of B> .v/ and the parent of t is not an
ancestor of v). The clockwise preordering of T 0 in (a) is r0 ; r2 ; v; u; t; w; r1

Definition 1 (well-orderly tree). T is a well-orderly tree of H if all the nodes of T

are well orderly in H , and if the root of T belongs to the boundary of the outerface
of H (similarly for simply orderly tree).
Note that an orderly tree (simply or well orderly) is necessarily a spanning tree.
Observe also that the incident edges in H of a node of T are either in T or unrelated.
In particular, if an edge of H is related (i.e., one endpoint is a descendant of the other
one in T ), then it has to belong to T . It follows that all the neighbors in H of the
root of T are in T .
Definition 2 (well-orderly map). A plane graph H is a well-orderly map rooted
in v if H has a well-orderly tree of root v.
A convenient way to manipulate triangulations is to deal with realizers.
Definition 3 (realizer). A realizer of a triangulation is a partition of its interior
edges in three sets T0 , T1 , T2 of directed edges such that for each interior node v it
holds (see Fig. 2.2a):
• The clockwise order of the edges incident with v is: leaving in T0 , entering in T1 ,
leaving in T2 , entering in T0 , leaving in T1 , and entering in T2 .
• There is exactly one leaving edge incident with v in T0 , T1 , and T2 .
Observe that if .T0 ; T1 ; T2 / is a realizer, then .T1 ; T2 ; T0 / and .T2 ; T0 ; T1 / are
also realizers. Cyclic permutations of a realizer are not in general the only distinct
realizers of a given triangulation. Figure 2.3 depicts two realizers for a same
2 An Information-Theoretic Upper Bound on Planar Graphs 25

triangulation. Actually, the number of n-node realizers is asymptotically 24nCO.log n/

(cf. [5]), whereas the number of triangulations is only .256=27/nCO.log n/ (cf. [39]).
Schnyder showed in [36] that each set Ti of a realizer induces a tree rooted
in one node of the outerface. Moreover, he described a linear time algorithm to
compute such trees. Hereafter, if R D .T0 ; T1 ; T2 / is a realizer, then for notational
convenience R also denotes the underlying triangulation.
There are strong relationships between realizers and orderly trees (see Fig. 2.2).
In every realizer R D .T0 ; T1 ; T2 /, T0 (and by cyclic permutation each Ti ) is an
orderly tree of R n fr1 ; r2 g, where ri denotes the root of Ti . Indeed, the incident
edges with any node v that are not in T0 (thus that are unrelated with T0 ) are either
clockwise before the entering edges of T0 or clockwise after. Conversely, let T be
an orderly tree of a triangulation. Observe that the root of T has at least two children
(because its root is of degree at least two and all its neighbors must be in T ), and
thus T has at least two leaves. A realizer .T0 ; T1 ; T2 / can be obtained from T setting
T0 D T n fr1 ; r2 g, where r1 ; r2 are, respectively, the clockwise last and first leaf of
T – actually it is not difficult to see that ri is the root of Ti – and setting, for all inner
nodes v, that the clockwise first edge of B> .v/ and the clockwise last edge of B< .v/
belong to T1 and T2 , respectively (as illustrated in Fig. 2.2b). Observe that this latter
assignment for T1 and T2 is the only possible realizer with T0 D T n fr1 ; r2 g.
For each tree Ti of a realizer, we denote by T i the tree composed of Ti augmented
with the two edges of the outerface incident to the root of Ti . A node of a rooted
tree is inner if it is neither the root nor a leaf. For every non-root node u 2 Ti , we
denote by pi .u/ the parent of u in Ti .
Definition 4 (super-triangulation). A realizer S D .T0 ; T1 ; T2 / is a super-
triangulation of a graph G if:
1. V .S / D V .G/ and E.G/  E.S /
2. E.T0 /  E.G/
3. T 0 is a well-orderly tree of S
4. For every inner node v of T2 , .v; p1 .v// 2 E.G/
Intuitively, a super-triangulation of a graph G is a specific triangulation of
the faces of a specific plane embedding of G. Before exploring more deeply the
properties of super-triangulations, observe that, from Definition 4, the tree T0 does
not span in general the graph G (cf. example in Fig. 2.4). Moreover, a non-connected
graph may have a super-triangulation. For example, if G has an edge and two
isolated nodes, in that case, E.T0 / D E.G/ is possible.
Theorem 1. Every connected planar graph with at least three nodes has a super-
triangulation, computable in linear time.
In particular, Theorem 1 implies that every connected planar graph has an
embedding which is a well-orderly map.
26 N. Bonichon et al.

v6
b

a v5 v5

v2
v2 v8
v8
v4 v4
v7 v7
v6
v3 v3

c v6
v1 v1

v5

v2 v8
v4
v7

v3

v1

Fig. 2.4 A planar graph G (a), a well-orderly map of G rooted at v1 with its well-orderly tree
(bold edges) (b), and a super-triangulation of G (c) (dotted edges are non-edges of G)

2.2 Computing a Super-Triangulation from a Well-Orderly Map

In order to prove Theorem 1, we need the next three lemmas. The proofs of these
lemmas are given after the proof of Theorem 1.
Lemma 1. Every well-orderly map rooted in some node v has a unique well-orderly
tree of root v.
Lemma 2. Let G be a connected planar graph, and let v be any node of G. Then G
has a well-orderly map of root v. Moreover, well-orderly trees and the well-orderly
map can be computed in linear time.
In [8], a result similar to Lemma 2 about simply orderly trees and embeddings
is proved. However, the extra condition reduces much more the choice of the
embedding for the input planar graph and leads to the uniqueness of the tree
2 An Information-Theoretic Upper Bound on Planar Graphs 27

(Lemma 1). In the case of simply orderly embeddings, several orderly trees may
exist (cf. Fig. 2.3 where both orderly trees T 0 span the same triangulation). Actually,
the uniqueness concerns also the way to triangulate the faces of well-orderly maps,
thanks to the next lemma.
Lemma 3. Let T be the well-orderly tree of H rooted in some node r0 , and assume
that T has at least two leaves. Let r2 and r1 be the clockwise first and last leaves
of T , respectively. Then, there is a unique super-triangulation .T0 ; T1 ; T2 / of the
underlying graph of H , preserving the embedding H , and such that each Ti has
root ri . Moreover, T0 D T n fr1 ; r2 g and the super-triangulation are computable in
linear time.
First of all, let us show that Lemmas 1, 2, and 3 imply Theorem 1.
Proof of Theorem 1. Consider a connected planar graph G with at least three nodes,
and let v be any node of G with the only constraint that if G is a path, then v is
chosen to be of degree two (this is feasible since G has at least three nodes). Thanks
to Lemma 2, one can compute in linear time a well-orderly map H of G and a well-
orderly tree T rooted in v. Let us show that T has at least two leaves r1 ; r2 lying on
the outerface of H , r2 traversed before r1 in a clockwise preordering of T .
We show that T cannot be a chain, and thus has a node with at least two children
(and thus has two leaves). If G is a path, then T rooted in a node of degree two is
not a chain. Assume that G is not a path, but T is a chain. Then there exists an edge
of G that is not in T . However, all pairs of nodes of a chain are related, thus must
belong to T . Therefore, T is not a chain.
Lemma 3 can be therefore applied, and one can compute for G a super-
triangulation in linear time. t
u
Proof of Lemma 1. Assume that H has two well-orderly trees T; T 0 rooted in v. Let
v1 ; : : : ; vn (resp. v01 ; : : : ; v0n ) be the clockwise preordering of the nodes of T (resp.
T 0 ). Let vi be the node such that the neighbors of vi in T and in T 0 differ, and such
that i is minimum. We have vt D v0t for all t 6 i , and BC .vi / ¤ BC0 .vi /, where
BC0 .vi / denotes the children edge block around vi in T 0 .
W.l.o.g. assume jBC .vi /j 6 jBC0 .vi /j (the symmetric case is proved by exchang-
ing the role of T and T 0 ). Note that B< .vi / D B> .vi / D ¿ is impossible, otherwise
BC .vi / would consist of all the neighbors of vi (maybe the vi ’s parent excepted)
and jBC .vi /j 6 jBC0 .vi /j and BC .vi / ¤ BC .vi / would be incompatible. Let e1
(resp. e2 ) be the clockwise first (resp. last) edge of BC .vi /. Let e be an arbitrary
edge of BC0 .vi /. In the following, e1 6 e means either e1 D e, or e1 is clockwise
before e around vi .
Let us show that e1 6 e. This is clearly true if B< .vi / D ¿. If B< .vi / ¤ ¿, then
consider any edge .vi ; vh / 2 B< .vi /. Then, .vi ; vh / … BC0 .vi /. Indeed, as h < i , the
path from vh to vi in T exists also in T 0 , and the edge .vi ; vh / of T 0 would create a
cycle in T 0 . Thus, e1 6 e.
28 N. Bonichon et al.

If B> .vi / D ¿, then e 6 e2 . Hence, e1 6 e 6 e2 which is incompatible with the

fact that BC0 .vi / and BC .vi / are blocks of consecutive edges such that jBC .vi /j 6
jBC0 .vi /j. Thus, we must have B> .vi / ¤ ¿.
Let .vi ; vj / be the clockwise first edge of B> .vi /. Then, .vi ; vj / … BC0 .vi /.
Indeed, as T is well orderly, the vj ’s parent in T , say vk , is an ancestor of vi , so
k < i . As BC .vk / D BC0 .k/ for k < i , the edge .vk ; vj / exists in T 0 . Hence,
the path from vj to vi in T exists also in T 0 , and the edge .vi ; vj / of T 0 would
create a cycle in T 0 . It follows that every edge e 2 BC0 .vi / is such that .vi ; vj / 6 e
and e ¤ .vi ; vj /. As the path from vi to vj in T exists also in T 0 , the node vj is
after vi in a clockwise preorder of T 0 . It follows that vi is not well orderly in T 0 : a
contradiction. t
u
Proof of Lemma 2. We first give a simple algorithm to construct a well-orderly map
of G. Then we give some hints for an O.n/ time implementation.
We start by computing an arbitrary plane embedding H of G such that v belongs
to the outerface of H . This can be done in O.n/ time [10]. Then we traverse H from
v to build a well-orderly tree T . However, not every plane embedding allows the
construction of a well-orderly tree. If during the construction, T does not span all the
nodes, the embedding of H is modified, and a new traversal is run again. We show
that, after a finite number of steps, the number of nodes covered by the traversal
increase, and so the construction converges to a well-orderly tree. To describe more
precisely the traversal and the modification of H , we need some definitions.
Let T be any tree of H rooted in v, not necessarily a spanning tree. A node
is free if it does not belong to T . An edge is free if one of its endpoints if free.
We extended the notion of well-orderly node as follows: a node is partially well
orderly (with respect to H and T ) if it is well orderly except that its edge-blocks
B< and B> (relative to a clockwise preordering of the nodes of T ) may contain
zero or more free edges. Moreover, the clockwise last edge of B< and the clockwise
first edge of B> are never free. T is partially well orderly if all the nodes of T are
partially orderly. Every well-orderly tree is a partially well-orderly tree that spans
H . The four edge-blocks around a partially well-orderly node u in T are denoted by
BP .v; T /, B< .v; T /, BC .v; T /, and B> .v; T /. For convenience, the clockwise last
edge of B< .v; T / is named the back-edge of u in T , and the clockwise first edge of
B> .v; T / is named the front-edge in T .
We run a procedure traversal .H; v/ applied on the current embedding H of
G, which returns a partially well-orderly tree T of H rooted in v (see Fig. 2.5 for
example). It initializes T WD fvg and treats v as follows:
(1) List in a clockwise manner around v the edges .v; u1 /; : : : ; .v; uk / that are after
the back-edge and before the front-edge of v in T (if the back-edge or the front-
edge does not exist, the list consists of all the edges incident with v not already
in T )
(2) Update T with the edges .v; u1 /; : : : ; .v; uk /
(3) Recursively treat the nodes u1 ; : : : ; uk in that order
Another Random Scribd Document
with Unrelated Content
has elapsed without your mentioning it to me. If the plan was faulty,
if the letter was ill-expressed, why not have told me so? I should
have assembled you all to discuss the matter with me.”
We all stood near the arbour at the extremity of the path, while
the Emperor walked to and fro before us. At a moment, when he
had gone to a little distance from us, and was out of hearing, the
Grand Marshal, addressing himself to me, said:—“I fear I have
expressed myself improperly, and I am sorry for it.”—“We will leave
you alone with him,” said I; “you will soon make him forget the
offence.” I accordingly beckoned the other individuals who were
present to leave the garden.
In the evening, the Emperor, conversing with me about the events
of the morning, said:—“It was after we had made it up with the
Grand Marshal—It was before the misunderstanding with the Grand
Marshal,”—and other things of the same sort, which proved that the
affair had left no impression on his heart.

THE GENERALS OF THE ARMY OF ITALY.—ANCIENT ARMIES.—GENGISKAN.—

MODERN INVASIONS.—CHARACTER OF CONQUERORS.

6th.—The Emperor was somewhat unwell, and employed himself

in writing in his chamber. He dictated to me the portraits of the
Generals of the army of Italy—Massena, Augereau, Serrurier, &c.
Massena was endowed with extraordinary courage and firmness,
which seemed to increase in excess of danger. When conquered, he
was always as ready to fight the battle again as though he had been
the conqueror. Augereau, on the contrary, seemed to be tired and
disheartened by victory, of which he always had enough. His person,
his manners, and his language, gave him the air of a bravo. This,
however, says the narration, he was far from being, when he once
found himself sated with honours and riches, which he had
bestowed upon himself on every occasion that offered, and by every
means in his power. Serrurier, who retained the manners and
severity of an old major of infantry, was an honest and trustworthy
man; but an unfortunate general, &c.
 Among the various subjects of the day’s conversation, I note down
what the Emperor said respecting the armies of the Ancients. He
asked whether the accounts of the great armies mentioned in history
were to be credited. He was of opinion that those statements were
false and absurd. He placed no faith in the descriptions of the
innumerable armies of the Carthaginians in Sicily. “Such a multitude
of troops,” he observed, “would have been useless in so
inconsiderable an enterprise; and if Carthage could have assembled
such a force, a still greater one would have been raised in Hannibal’s
expedition, which was of much greater importance, but in which not
more than forty or fifty thousand men were employed.” He did not
believe the accounts of the millions of men composing the forces of
Darius and Xerxes, which might have covered all Greece, and which
would doubtless have been subdivided into a multitude of partial
armies. He even doubted the whole of that brilliant period of Greek
history; and he regarded the famous Persian war only as a series of
those undecided actions, in which each party claims the victory.
Xerxes returned triumphant, after taking, burning, and destroying
Athens; and the Greeks exulted in their victory, because they had
not surrendered at Salamis. “With regard to the pompous accounts
of the conquests of the Greeks, and the defeat of their numberless
enemies, it must be recollected,” observed the Emperor, “that the
Greeks, who wrote them, were a vain and hyperbolical people; and
that no Persian chronicle has ever been produced to set our
judgment right by contrary statements.”
But the Emperor attached credit to Roman history, if not in its
details, at least in its results; because these were facts as clear as
day-light. He also believed the descriptions of the armies of
Gengiskan and Tamerlane, however numerous they are said to have
been; because they were followed by gregarious nations, who, on
their part, were joined by other wandering tribes as they advanced;
“and it is not impossible,” observed the Emperor, “that Europe may
one day end thus. The revolution produced by the Huns, the cause
of which is unknown, because the tract is lost in the desert, may at a
future period be renewed.”
 The situation of Russia is admirably calculated to assist her in
bringing about such a catastrophe. She may collect at will
numberless auxiliaries, and scatter them over Europe. The
wandering tribes of the north will be the better disposed and the
more impatient to engage in such enterprises, in proportion as their
imaginations have been fired, and their avarice excited, by the
successes of those of their countrymen who lately visited us.
The conversation next turned on conquests and conquerors; and
the Emperor observed that, to be a successful conqueror, it was
necessary to be ferocious, and that, if he had been such, he might
have conquered the world. I presumed to dissent from this opinion,
which was doubtless expressed in a moment of vexation. I
represented that he, Napoleon, was precisely a proof of the
contrary; that he had not been ferocious, and yet had conquered the
world; and that, with the manners of modern times, ferocity would
certainly never have raised him to so high a point. I added that, at
the present day, terror could never subject us to the control of an
individual man; and that dominion was to be secured only by good
laws, joined to greatness of character, and that degree of energy,
which is proof against every trial, in him who is charged with the
execution of the laws. These, I affirmed, were precisely the causes
of Napoleon’s success, and of the submission and obedience of the
people over whom he ruled.
The Convention was ferocious, and inspired terror: it was
submitted to, but could not be endured. Had the power been vested
in an individual, his overthrow would soon have been accomplished.
But the Convention was a hydra, yet how many attempts were
hazarded for its destruction!—how many dangers did it escape as if
by miracle! It was reduced to the necessity of burying itself amidst
its triumphs.
For a conqueror to be ferocious, with success, he must of
necessity command troops who are themselves ferocious, and he
must wield dominion over unenlightened people. In this respect
Russia possesses an immense superiority over the rest of Europe.
She has the rare advantage of possessing a civilized government,
and barbarous subjects. There, information directs and commands,
while ignorance executes and destroys. A Turkish Sultan could not
long govern any enlightened European nation; the empire of
knowledge would be too strong for his power.
Speaking on another subject, the Emperor observed that, if the
French people had less energy than the Romans, they at least
evinced greater decorum. We should not have killed ourselves, as
the Romans did under the first Emperors; but at the same time we
should have afforded no examples of the turpitude and servility that
marked the later periods of the Roman empire. “Even in our most
corrupt days,” said he, “our baseness was not without certain
restrictions: courtiers, whom the sovereign could have prevailed on
to do any thing in his own palace, would have refused to bend the
knee to him at his levee.”
I have already mentioned that we had with us scarcely any
document relative to French affairs during the Emperor’s time. The
books that had been brought among his effects were merely a few
classics which he carried about with him in all his campaigns. I
received from Major Hudson, resident in the island, a political
compendium from 1793 to 1807, entitled “The Annual Register,”
which contains the succession of political events during each year,
together with some of the most important official documents. In our
destitute circumstances this was a valuable acquisition.

IDEAS.—PLANS.—POLITICAL SUGGESTIONS, &C.

7th.—The Emperor breakfasted alone, and was engaged during
the day in dictating to the Grand Marshal and M. de Montholon.
In the evening, as the Emperor and I were walking together in the
lower path, which had now become the favourite resort, I informed
him that a person of consequence, whose ideas and statements
might become the channel of mediation between ourselves and the
ruling world, and influence our future destiny, had, with sufficiently
significant forms and preambles, invited us to tell him
conscientiously what we believed to be the Emperor’s notions on
certain political subjects:—whether he had granted his last
constitution with the intention of maintaining it;—whether he had
sincerely abandoned his former plans as to the great empire;—
whether he would consent to leave England in the enjoyment of her
maritime supremacy, without envying her the tranquil possession of
India;—whether he would be willing to renounce the Colonies, and
to purchase colonial goods from the English alone, at the regular
market price;—whether he would not form an alliance with the
Americans, in case of their rupture with England;—whether he would
consent to the existence of a great kingdom in Germany, for that
branch of the English royal family which must immediately forfeit the
throne of Great Britain on the accession of the Princess Charlotte of
Wales, or, in default of Germany, whether he would consent to the
establishment of that dominion in Portugal, in case England should
conclude arrangements to that effect with the Court of Brazil.
These questions did not rest on vague ideas or idle opinions: the
individual supported them on positive facts. “We want,” said he, “a
long and lasting peace on the Continent—we want the tranquil
enjoyment of our present advantages, to help us out of the critical
circumstances in which we are now involved, and to relieve us from
the enormous debt with which we are burthened. The present state
of France and of Europe,” added he, “cannot bring about these
results. The victory of Waterloo has ruined you; but it was far from
saving us; every sensible man in England, every one who escapes
the momentary influence of passion, either does think, or will think,
as I do.”
The Emperor doubted a part of this statement, and treated the
rest as a reverie; then, changing his tone, he said to me, “Well, what
is your opinion? Come, sir, you are now in the Council of
State.”—“Sire,” I replied, “people often indulge in reveries on the
most serious subjects; and our being imprisoned at St. Helena does
not hinder us from composing romances. Here then is one. Why not
form a political marriage between the two nations, in which the one
would bring the army as a dowry: and the other the navy? This will
doubtless appear an absurd idea in the eyes of the vulgar, and will
perhaps be thought too bold by well-informed people, because it is
entirely novel and out of the usual routine. But it is one of those
unforeseen, luminous and useful creations of plans which
characterize your Majesty. You alone can cause it to be listened to,
and carry it into effect.”
Going even beyond the ideas of our English interlocutor, I said,
“Would not your Majesty give to-morrow, if you could, the whole
French navy to purchase Belgium and the bank of the Rhine? Would
you not give one hundred and fifty millions to purchase tens of
thousands of millions? Besides, such a bargain would procure to
both nations at once the object for which they have been wrangling
and fighting for so many years; it would reduce both countries to the
necessity of mutually assisting each other, instead of maintaining
perpetual enmity. Would it be nothing for France that her merchants
in the English colonies should henceforward be on the footing of
Englishmen; and thus secure, without striking a blow, the enjoyment
of the trade of the whole world? Would it not be every thing for
England, on the other hand, for the sake of insuring to herself the
sovereignty of the ocean, the universality of trade (for obtaining and
preserving which she has incurred so many risks), to attach France
to a system by which the latter would become the regulator and
arbitrator of the Continent?
“Henceforth secure from danger, and strengthened by all the
power of her ally, England might disband her army, in return for the
sacrifice which France would make of her navy. She might even
reduce considerably the number of her ships. She would thus pay
her debt, relieve her people, and prosper; and, far from envying
France, she would (the system being once fairly understood, and
passion having given place to real interests) herself labour for the
Continental aggrandizement of her neighbour; for France would then
be merely the advance-guard, while England would be the resource
and the reserve. Unity of legislation between the two nations, their
common interests, results so visibly advantageous, would make
amends for all the obstacles and difficulties which the passions of
rulers might oppose to the fulfilment of this plan.”
The Emperor heard me, but made no reply: it is seldom that one
can ascertain his private opinions, and he rarely enters into political
conversations. Lest I should not have expressed myself with
sufficient clearness, I requested that he would permit me to unfold
my ideas in writing. He consented, and said no more. It was now
very late, and he retired to rest.
8th.—The Emperor dictated in the garden to Messrs. Montholon
and Gourgaud, and then walked on his favourite path. He was
fatigued and indisposed. He observed some females about to
advance on the path, and to throw themselves awkwardly in his way,
for the sake of being introduced to him; this annoyed him, and he
turned away to avoid meeting them.
I suggested that riding on horseback might be beneficial to him;
we had three horses at our disposal. The Emperor replied that he
never could reconcile himself to the idea of having an English officer
constantly at his side; that he decidedly renounced riding on such
conditions; adding, that every thing in life must be reduced to
calculation, and that, if the vexation arising from the sight of his
jailor were greater than the advantage he might derive from riding,
it was of course advisable to renounce the recreation altogether.
The Emperor ate but little dinner. During the dessert he amused
himself in examining the paintings on some plates of very beautiful
Sevres porcelain. They were masterpieces in their kind, and were
worth thirty Napoleons each. The paintings represented views or
objects of antiquity in Egypt.
The Emperor closed the day with a walk on his favourite path. He
remarked that he had been very dull all day. After several broken
conversations, he looked at his watch, and was very glad to find it
was half-past ten.
The weather was delicious, and the Emperor insensibly recovered
his usual spirits. He complained of his constitution, which, though
vigorous, occasionally exposed him to fits of indisposition. He,
however, consoled himself with the thought that if, in imitation of the
ancients, he should ever feel inclined to escape from the disgusts
and vexations of life, his moral opinions were not of a nature to
prevent him. He said that sometimes he could not reflect without
horror on the many years he might still have to live, and on the
inutility of a protracted old age; and that, if he were convinced
France was happy and tranquil, and not needing his aid, he should
have lived long enough.
We ascended to the pavilion, it was past midnight, and we
thought we had gained a signal victory over time.
9th.—I called on Mr. Balcombe very early, to deliver to him my
letters for Europe, as a vessel was on the point of sailing. At Mr.
Balcombe’s house I met the officer who had been appointed as our
guard. Struck with the dejection which I had observed in the
Emperor the day before, and convinced of the necessity of his taking
exercise, I told the officer that I suspected the reason which
prevented the Emperor from riding on horseback; I added, that I
would speak to him the more candidly and openly, since I had
noticed the very delicate way in which he discharged his duty. I
inquired what were his instructions, and whether it would be
necessary to observe them literally, in case the Emperor merely took
a ride round the house; adverting to the repugnance which he must
naturally feel for arrangements that were calculated every moment
to revive the recollection of the situation in which he was placed. I
assured the officer that no reflections were intended to be cast on
him personally, and that I was convinced, when the Emperor wished
to take long rides, he would prefer having him to accompany him.
The officer replied that his instructions were to follow the Emperor;
but that, as he made it a rule not to do any thing that might be
offensive to him, he would take upon himself not to accompany him.
At breakfast I communicated to the Emperor the conversation I
had had with the Captain. He replied that it was all well meant on
the Captain’s part; but that he should not avail himself of the
indulgence, as it was not conformable with his sentiments to enjoy
an advantage which might be the means of compromising an officer.
This determination was very fortunate. When I went to Mr.
Balcombe’s in the evening, the Captain took me aside to inform me
that he had been to the town, in the course of the day, to speak
with the Admiral respecting our morning’s conversation, and that he
had been enjoined to abide by his instructions. I could not refrain
from replying, somewhat sharply, that I was certain the Emperor
would immediately send back the three horses that had been
assigned for his use. The officer, to whom I had also communicated
the reply which the Emperor had given me in the morning relative to
him, observed that it would be very right to send back the horses,
and that he thought nothing better could be done. This remark
appeared to me to be prompted by the mortification he himself
experienced at the part that was imposed on him.
When we left Mr. Balcombe’s, the Emperor walked up and down
the path in the garden. I mentioned to him what I had heard from
the English officer. He seemed to expect it. I was not deceived in my
conjecture; he ordered me to send away the horses. This vexed me
exceedingly; and I said, perhaps rather sharply, that with his leave I
would go and fulfil his orders immediately. On which he replied, with
great gravity and in a very peculiar tone of voice:—“No, Sir, you are
now out of temper. It rarely happens that any thing is done well
under such circumstances; it is always best to let the night pass over
after the offence of the day.”
We continued our walk till nearly midnight: the weather was
delightful.
10th.—To-day, when our usual task was ended, the Emperor
strolled out in a new direction. He proceeded towards the town, until
he came within sight of the road and the shipping. As he was
returning, he met Mrs. Balcombe, and a Mrs. Stuart, a very pretty
woman about twenty years of age, who was returning from Bombay
to England. The Emperor conversed with her respecting the manners
and customs of India, and the inconveniencies of a sea-voyage,
particularly for females. He also spoke of Scotland, which was Mrs.
Stuart’s native country; said a great deal about Ossian, and
complimented the lady on the climate of India not having spoiled her
clear Scottish complexion.
At this moment some slaves, carrying heavy boxes, passed by us
on the road; Mrs. Balcombe desired them, in rather an angry tone,
to keep back; but the Emperor interfered, saying, “Respect the
burden, Madam!” At these words, Mrs. Stuart, who had been
attentively observing the Emperor’s features, said, in a low tone of
voice, to her friend: “Heavens! what a countenance, and what a
character! How different from what I had been led to expect!”
MIDNIGHT CONVERSATIONS BY MOONLIGHT.—THE TWO EMPRESSES.—MARIA
LOUISA’S MARRIAGE.—HER HOUSEHOLD.—THE DUCHESS DE MONTEBELLO.
—MADAME DE MONTESQUIOU.—THE INSTITUTE OF MEUDON.—
SENTIMENTS OF THE HOUSE OF AUSTRIA, WITH REGARD TO NAPOLEON.—
ANECDOTES COLLECTED IN GERMANY, SINCE MY RETURN TO EUROPE.

11th—13th. We led a most regular life at Briars. Every day, after

dictating to me, the Emperor walked out between three and four
o’clock. He descended to the garden, where he walked up and
down, and dictated to one of the gentlemen who came from the
town for that purpose, and who wrote in the little arbour. About half-
past five o’clock, he passed Mr. Balcombe’s house, and went into the
lower walk, to which he became every day more and more attached.
At this time the family were at dinner, and he could enjoy his
promenade without interruption. I joined the Emperor here, and he
continued his walk until dinner was announced.
After dinner the Emperor again returned to the garden, where he
sometimes had his coffee brought to him. My son then visited Mr.
Balcombe’s family, and the Emperor and I walked up and down. We
frequently remained in the garden until the night was far advanced
and the moon rose to light us. In the mildness and serenity of the
night we forgot the burning heat of the day. The Emperor never was
more talkative, nor seemed more perfectly to forget his cares, than
during these moonlight walks. In the familiarity of the conversations
which I thus enjoyed with him, he took pleasure in relating
anecdotes of his boyhood, in describing the sentiments and illusions
which diffused a charm over the early years of his youth, and in
detailing the circumstances of his private life, since he had played so
distinguished a part on the great theatre of the world. I have
elsewhere noted down what I conceived myself at liberty to repeat.
Sometimes he seemed to think he had spoken too much at length,
and had detailed things too minutely. He would then say to me:
“Come, it is your turn now: let me have a little of your history; but
you are not a tale-teller.” Indeed I took especial care to be silent; I
was too much afraid of losing a syllable of what so deeply interested
me.
 In one of our nightly walks, the Emperor told me that he had, in
the course of his life, been attached to two women of very different
characters. The one was the votary of art and the graces; the other
was all innocence and simple nature; and each, he observed, had a
very high degree of merit.
The first, in no moment of her life ever assumed a position or
attitude that was not pleasing or captivating; it would have been
impossible ever to discover in her, or to experience from her, any
thing unpleasant. She employed every resource of art to heighten
natural attractions; but with such ingenuity as to render every trace
of it imperceptible. The other, on the contrary, never even suspected
that any thing was to be gained by innocent artifice. The one was
always beside the truth, her first answer was always in the negative;
the other was altogether frank and open, and was a stranger to
subterfuge. The first never asked her husband for any thing, but she
was in debt to every one: the second freely asked whenever she
wanted, which, however, very seldom happened; and she never
thought of receiving any thing without immediately paying for it.
Both were amiable and gentle in disposition, and strongly attached
to their husband. But it must already have been guessed who they
are; and those who have ever seen them will not fail to recognise
the two Empresses.
The Emperor declared that he had uniformly experienced from
both the greatest equality of temper and most implicit obedience.
The marriage of Maria-Louisa was consummated at Fontainebleau,
immediately after her arrival. The Emperor, setting aside all the
etiquette that had previously been arranged, went to meet her, and,
in disguise, got into her carriage. She was agreeably surprised when
she discovered him. She had always been given to understand that
Berthier, who had married her by proxy at Vienna, in person and age
exactly resembled the Emperor: she, however, signified that she
observed a very pleasing difference between them.
The Emperor wished to spare her all the details of domestic
etiquette, customary on such occasions: she had, however, received
careful instructions on the subject at Vienna. The Emperor inquired
what directions she had received from her illustrious relatives with
regard to him personally. To be entirely devoted to him, and to obey
him in all things, was the reply. This declaration, and not the
decisions of certain cardinals and bishops, as was reported, proved
the solution of all the Emperor’s scruples of conscience. Besides,
Henry IV. acted in the same way on a similar occasion.
Maria-Louisa’s marriage, said the Emperor, was proposed and
concluded under the same forms and conditions as that of Marie-
Antoinette, whose contract was adopted as a model. After the
separation from Josephine, negotiations were entered into with the
Emperor of Russia, for the purpose of soliciting the hand of one of
his sisters: the difficulties rested merely on the settling of certain
points relative to religion. Prince Eugene, conversing with M. de
Schwartzenberg, learned that the Emperor of Austria would not
object to an union between Napoleon and his daughter; and this
information was communicated to the Emperor. A council was
convoked to decide whether an alliance with Russia or Austria would
be most advantageous. Eugene and Talleyrand were for the Austrian
alliance, and Cambaceres against it. The majority were in favour of
an Archduchess. Eugene was appointed to make the official
overture, and the Minister for Foreign Affairs was empowered to sign
it that very day if an opportunity should present itself; which proved
to be the case.
Russia took umbrage at this; she thought herself trifled with,
though without just ground. Nothing of an obligatory nature had yet
transpired; both parties remained perfectly free. Political interests
predominated over every thing.
The Emperor appointed the Duchess de Montebello to be lady of
honour to Maria-Louisa; the Count de Beauharnais to be her
gentleman of honour, and the Prince Aldobrandini to be her equerry.
In the misfortunes of 1814, these individuals, said the Emperor, did
not evince the devotedness which the Empress was entitled to
expect from them. Her equerry deserted her without taking his
leave; her gentleman of honour refused to follow her; and her lady
of honour, notwithstanding the attachment which the Empress
entertained for her, thought she had completely fulfilled her duty in
attending her mistress as far as Vienna.
 The appointment of the Duchess de Montebello to the post of lady
of honour was one of those happy selections which, at the time it
was made, excited universal approbation. The Duchess was a young
and beautiful woman, of irreproachable character, and the widow of
a marshal, called the Orlando of the army, who had recently fallen
on the field of battle. This choice was very agreeable to the army,
and encouraged the national party, who were alarmed at the
marriage and the number and rank of the chamberlains who were
appointed. This retinue was, by many, looked upon as a step
towards the counter-revolution; and endeavours were made to
represent it as such. As for the Emperor, he had acted in ignorance
of the character of Maria-Louisa, and had been principally influenced
by the fear that she would be filled with prejudices respecting birth,
that might be offensive at the Court. When he came to know her
better, and found that she was wholly imbued with the prevailing
notions of the day, he regretted not having made another choice. He
conceived that he should have done better to select the Countess de
Beauveau, a woman of amiable, mild, and inoffensive manners, who
would have been influenced only by the family advice of her
numerous relatives, and who might thus have introduced a kind of
useful custom, and have occasioned the appointment of well-
recommended inferiors. She might also have rallied about the Court
many persons who were at a distance; and that without any
inconvenience, because these arrangements could only have been
brought about by the sanction of the Emperor, who was not the sort
of man to allow himself to be abused.
The Empress conceived the tenderest affection for the Duchess de
Montebello. This lady had at one time a chance of being Queen of
Spain. Ferdinand VII. when at Valency, requested the Emperor’s
permission to marry Mademoiselle de Tascher, cousin-german of
Josephine, and bearing the same name, after the example of the
Prince of Baden, who married Mademoiselle de Beauharnais. The
Emperor, who already contemplated a separation from Josephine,
refused his consent to the match, not wishing by this connexion to
add to the difficulties that already stood in the way of his divorce.
Ferdinand then solicited the hand of the Duchess de Montebello, or
of any other French lady whom the Emperor might think proper to
adopt. The Emperor subsequently gave Mademoiselle de Tascher in
marriage to the Duke d’Aremberg, whom he intended to create
Governor of the Netherlands; with the view of ultimately
compensating Brussels for the loss of the old Court. The Emperor
moreover wished to appoint the Count de Narbonne, who had taken
part in the Empress’s marriage, Gentleman of honour, in room of the
Count de Beauharnais; but the extreme aversion which Maria-Louisa
evinced for this change deterred the Emperor from carrying it into
effect. The Empress’s dislike to the Count de Narbonne was,
however, only occasioned by the intrigues of the individuals
composing her household, who had nothing to fear from M. de
Beauharnais, but who very much dreaded the influence and talent of
M. de Narbonne.
The Emperor informed us that, when he had to make
appointments to difficult posts, he usually asked the persons about
him to furnish him with a list of candidates; and from these lists, and
the information he obtained, he privately deliberated on his choice.
He mentioned several individuals who were proposed as lady of
honour to the Empress: they were the Princess de Vaudemont,
Madame de Rochefoucalt, afterwards Madame de Castellanes, and
some others. He then asked us to tell him whom we should have
proposed; which occasioned us to take a review of a good part of
the Court. One of us mentioned Madame de Montesquiou; upon
which the Emperor replied, “She would have done well, but she had
a post which suited her still better. She was a woman of singular
merit; her piety was sincere, and her principles excellent; she had
the highest claims on my esteem and regard. I wanted half a dozen
like her; I would have given them all appointments equal to their
deserts, and wished for more. She discharged her duties admirably
when with my son at Vienna.”
The following anecdote will afford a correct idea of the manner in
which Madame de Montesquiou managed the King of Rome. The
apartments of the young Prince were on the ground floor, and
looked out on the court, of the Tuileries. At almost every hour in the
day, numbers of people were looking in at the window, in the hope
of seeing him. One day when he was in a violent fit of passion, and
rebelling furiously against the authority of Madame de Montesquiou,
she immediately ordered all the shutters to be closed. The child,
surprised at the sudden darkness, asked Maman Quiou, as he used
to call her, what it all meant. “I love you too well,” she replied, “not
to hide your anger from the crowd in the court-yard. You, perhaps,
will one day be called to govern all those people, and what would
they say if they saw you in such a fit of rage? Do you think they
would ever obey you, if they knew you to be so naughty?” Upon
which, the child begged her pardon, and promised never again to
give way to such fits of anger.
“This,” observed the Emperor, “was language very different from
that addressed by M. de Villeroi to Louis XV. ‘Behold all those people,
my Prince,’ said he, ‘they belong to you; all the men you see yonder
are yours.’”
Madame de Montesquiou was adored by the young King of Rome.
At the time of her removal from Vienna it was found necessary to
employ stratagems to deceive the child: it was even feared that his
health would suffer from the separation.
The Emperor had conceived many novel ideas relative to the
education of the King of Rome. For this important object he decided
on the Institut de Meudon, of which he had already laid down the
principle, with the view of farther developing it at his leisure. There
he proposed to assemble the Princes of the Imperial house,
particularly the sons of those branches of the family who had been
raised to foreign thrones. This plan, he contended, would have
combined the attentions of private tuition with the advantages of
public education. “These children,” said the Emperor, “who were
destined to occupy different thrones, and to govern different
nations, would thus have acquired conformity of principles, manners,
and ideas. The better to facilitate the amalgamation and uniformity
of the federative parts of the Empire, each Prince was to bring with
him from his own country ten or twelve youths of about his own
age, the sons of the first families in the state. What an influence
would they not have exercised on their return home! I doubted not,”
continued the Emperor, “but that Princes of other dynasties,
unconnected with my family, would soon have solicited, as a great
favour, permission to place their sons in the Institute of Meudon.
What advantages would thence have arisen to the nations
composing the European association! All these young Princes,” said
he, “would have been brought together early enough to be united in
the tender and powerful bonds of youthful friendship: and they
would, at the same time, have been separated early enough to
obviate the fatal effects of rising passions—the ardour of partiality—
the ambition of success—the jealousy of love.”
The Emperor wished that the education of the Princes should be
founded on general information, extended views, summaries, and
results. He wished them to possess knowledge rather than learning,
judgment rather than attainments; he preferred the application of
details to the study of theories. Above all, he objected to the
pursuing of any particular study too deeply, for he regarded
perfection, or too great success in certain things, whether in the arts
or sciences, as a disadvantage to a prince. A nation, he said, will
never gain much by being governed by a poet, a virtuoso, a
naturalist, a chymist, a turner, a locksmith, &c. &c.
Maria-Louisa confessed to the Emperor that, when her marriage
with him was first proposed, she could not help feeling a kind of
terror, owing to the accounts she had heard of Napoleon from the
individuals of her family. When she mentioned these reports to her
uncles, the Archdukes, who were very urgent for the marriage, they
replied,—“That was all very true, while he was our enemy: but the
case is altered now.“
“To afford an idea of the sympathy and good will with which the
different members of the Austrian family were taught to regard me,”
said the Emperor, “it is sufficient to mention that one of the young
Archdukes frequently burned his dolls, which he called roasting
Napoleon. He afterwards declared he would not roast me any more,
for he loved me very much, because I had given his sister Louisa
plenty of money to buy him play-things.”
Since my return to Europe, I have had an opportunity of
ascertaining the sentiments entertained by the House of Austria
towards Napoleon. In Germany, a person of distinction informed me
that having had a private audience of the Emperor Francis, during
his tour in Italy in 1816, the conversation turned on Napoleon.
Francis spoke of him in the most respectful terms. One might almost
have supposed, said my informant, that he still regarded him as the
ruler of France, and that he was ignorant of his captivity at St.
Helena. He never alluded to him by any other title than the Emperor
Napoleon.
I learned from the same individual that the Archduke John, when
in Italy, visited a rotunda, on the ceiling of which was painted a
celebrated action of which Napoleon was the hero. As he raised his
head to look at the painting, his hat fell off, and one of his
attendants stooped to pick it up. “Let it be,” said he; “it is thus that
the man who is there portrayed should be contemplated.”
Now that I am on this subject, I will note down a few particulars
which I collected in Germany since my return to Europe; and to
mark the degree of credit to which they are entitled, I may mention
that I obtained them from individuals holding high diplomatic posts.
Every one knows that these members of diplomatic corps form
among themselves a sort of family, a kind of free-masonry, and that
their sources of information are of the most authentic kind.
The Empress Maria-Louisa complained that, when she quitted
France, M. de Talleyrand reserved to himself the honour of
demanding from her the restitution of the Crown jewels, and
ascertaining whether they had been restored with the most
scrupulous exactness.
In 1814, during the disasters of France, many tempting and
brilliant proposals were made to Prince Eugene. An Austrian General
offered him the crown of Italy, in the name of the Allies, on condition
of his joining them. This offer afterwards came from a still higher
quarter, and was several times repeated. During the reign of the
Emperor there had been some idea of raising the Prince to a throne;
and those of Portugal, Naples, and Poland, were thought of.
In 1815, men of high influence in European diplomacy
endeavoured to sound his opinions, with the view of ascertaining
whether, in case Napoleon should again be constrained to abdicate,
and the choice should fall on him, he would accept the Crown. On
this occasion, as on every other, the Prince steadily pursued a line of
duty and honour which will immortalize him. Honour and fidelity was
his constant reply; and posterity will make it his device.
On the distribution of States in 1814, the Emperor Alexander, who
frequently visited the Empress Josephine at Malmaison, signified a
wish to procure for her son the sovereignty of Genoa. She, however,
declined this proposition, at the instigation of one of the ruling
diplomatists, who falsely flattered her with the hope of something
better.
At the Congress of Vienna, the Emperor Alexander, who honoured
Prince Eugene with particular marks of favour, insisted that he
should be made the Sovereign of at least three hundred thousand
subjects. He testified the sincerest friendship for him, and they were
every day seen walking about together arm-in-arm. The landing at
Cannes put an end, if not to the sentiment, at least to the
manifestation of it; and changed the political interests of the
Emperor of Russia.—The Austrian government even entertained the
idea of seizing the person of Prince Eugene, and sending him a
prisoner to a fortress in Hungary; but the King of Bavaria, his father-
in-law, indignantly represented to the Emperor of Austria that
Eugene had gone to Vienna under his protection and guarantee, and
that they should not be violated. Thus Eugene remained free on his
own private parole and that of the King his father-in-law.
So lately as 1814, gold twenty and forty-franc pieces were struck
at Milan with the head of Napoleon and the date of 1814. Either
from motives of economy or some other cause, no new die had yet
been engraved.
After the fall of Napoleon, Alexander on several occasions
manifested a marked and decided dislike to him. In 1815 he was the
promoter of the second crusade against Napoleon; he directed every
hostile measure with the utmost degree of animosity, and seemed to
make it almost a personal affair; alleging, as the cause of his
aversion, that he had been deceived and trifled with. If this tardy
resentment was not a mere pretence, there is every reason to
believe that it was stirred up by an old confidant of Napoleon’s, who,
in private conversations, had artfully wounded the vanity of
Alexander, by statements, true or false, of the private opinion of
Napoleon with regard to his illustrious friend.
In 1814 there appeared reason to believe that Alexander would
not be averse to see young Napoleon placed on the throne of
France. After the Emperor’s second abdication, he seemed far less
favourably disposed to the continuance of Napoleon’s dynasty.
In the second crusade, the Emperor Alexander marched at the
head of innumerable forces. He was heard to declare, at that period,
that the war might last for three years; but that Napoleon would
nevertheless be subdued in the end.
On the first intelligence of the battle of Fleurus, the chiefs of all
the Russian columns immediately received orders to halt; while all
the Austrian and Bavarian corps instantly turned off, with the view of
detaching themselves and forming a separate force. Had the
Congress of Vienna been broken up on the 20th of March, it is
almost certain that the crusade would not have been renewed; and
had Napoleon been victorious at Waterloo, it is also tolerably certain
that the crusade would have been dissolved.
The news of Napoleon’s landing at Cannes was a thunderbolt to
the French plenipotentiary at Vienna. He indeed drew up the famous
declaration of the 13th of March; and, virulent as it is, the first draft
was still more so: it was amended by other ministers. The
countenance of this plenipotentiary, as he gradually learned the
advance of Napoleon, was a sort of thermometer, which excited the
laughter of all the members of the Congress.
Austria soon knew the real state of affairs: her couriers informed
her admirably well of all that was passing. The members of the
French Legation alone were involved in doubt: they were still
circulating a magnanimous letter from the King to the other
Sovereigns, informing them that he was resolved to die at the
Tuileries, when it was already known that Louis had left the capital,
and was on his way to the frontier.
A member of the Congress and Lord Wellington, in a confidential
conversation with the members of the French Legation, with the
map in their hands, assigned the 20th or the 21st for Napoleon’s
entry into Paris.
 As the Emperor Francis received the official publications from
Grenoble and Lyons, he regularly forwarded them to Schöenbrunn,
to Maria Louisa, to whom they afforded extreme joy. It is very true
that, at a somewhat later period, an idea was entertained of seizing
young Napoleon, in order to convey him to France.
The French Plenipotentiary at length quitted Vienna, and
proceeded to Frankfort and Wisbaden, whence he could more
conveniently negotiate either with Ghent or Paris. Never was a time-
serving courier thrown into greater embarrassment and anxiety. The
ardour with which he had been inspired, on receiving the intelligence
of Napoleon’s landing at Cannes, was very much abated when he
heard of the Emperor’s arrival at Paris; and he entered into an
understanding with Fouché that the latter should be his guarantee
with Napoleon, pledging himself, on the other hand, to be Fouché’s
guarantee with the Bourbons. There is good ground for believing
that the offers made by this Plenipotentiary to the new Sovereign
went very great lengths indeed; but Napoleon indignantly rejected
them, lest, as he said, he should degrade his policy too far.
In 1814, before M. de Talleyrand declared himself for the
Bourbons, he was for the Regency; in which, however, he himself
wished to play the principal part. Events fatal to the Napoleon
dynasty prevented this moment of uncertainty from being turned to
good account. Every thing tends to prove that the result which was
at that period adopted was far from being agreeable to the
intentions of Austria; that power was duped, betrayed, or at least
carried by assault.
The fatality attending the military movements was such that the
Allies entered Paris without the concurrence of the Austrian Cabinet.
Alexander’s famous declaration against Napoleon Buonaparte and his
family was also made without the Austrian Power being consulted;
and the Count d’Artois only entered France by contriving to slip in
secretly in spite of the orders at the Austrian head-quarters, where
he had been refused passports.
It appears that Austria, on the retreat from Moscow, exerted
sincere efforts in London for negotiating a peace with Napoleon; but
the influence of the Russian Cabinet was all-powerful in London, and
no proposals for peace were listened to. The armistice of Dresden
then arrived, and Austria declared for war.
During this interval, the Austrian minister in London could never
obtain a hearing. He however remained for a considerable time in
the English capital, and did not leave it until the Allies had reached
the heart of France, and Lord Castlereagh hinted at the possibility,
for a moment, that the heroic success of Napoleon might render
negotiations indispensable.
If this minister had not previously been sent to London, he would
have been destined for Paris; and there probably his influence might
have brought about a turn of negotiations different from those which
arose during his absence between the Tuileries and Vienna.
In the height of the crisis he found himself detained in England as
if by force. In his impatience to reach the grand centre of
negotiations, he quitted his post, and proceeded to Holland, braving
a violent tempest. No sooner had he arrived on the theatre of events
than he fell into the hands of Napoleon at Saint-Dizier; but the fate
of France was then decided, though the fact was not yet known at
the French head-quarters. Alexander was entering Paris.
The Austrian minister in London exerted every endeavour to
procure a passport to enable him to join his Sovereign by passing
through Calais and Paris; but in vain. This circumstance; whether
accidental, or premeditated, was another fatality. But for this
disappointment, the Austrian minister would have reached Paris
before the Allies, would have joined Maria-Louisa, would have
defeated the last projects of M. de Talleyrand, and would have
altogether produced new combinations.
Opinion was divided in the Austrian Cabinet. One party was for the
union with France; the other was for the alliance with Russia.
Intrigue or chance decided in favour of Russia, and Austria, from
that moment, was merely led on.
14th.—The coffee that was served at our breakfast this morning
was better than usual; it was even good. The Emperor expressed
himself pleased with it. Some moments after he observed, placing
his hand on his stomach, that he felt the benefit of it. It would be
difficult to express what were my feelings on hearing this simple
remark. The Emperor, by thus appreciating so trivial an enjoyment,
contrary to his custom, unconsciously proved to me the effect of all
the privations he had suffered, but of which he never complained.
When we returned from our evening walk, the Emperor read to
me a chapter on the Provisional Consuls, which he had dictated to
M. de Montholon. Having finished reading, the Emperor took a piece
of ribbon, and began to tie together the loose sheets of paper. It
was late; the silence of night prevailed around us. My reflections
were on that day of a melancholy cast. I gazed on the Emperor. I
looked on those hands which had wielded so many sceptres, and
which were now tranquilly, and perhaps, not without some degree of
pleasure, occupied in the humble task of tying together a few sheets
of paper. On these sheets, indeed, were traced events that will never
be forgotten; portraits that will decide the judgment of posterity. It is
the book of life or death to many whose names are recorded in it.
These were the reflections that passed in my mind. “And the
Emperor,” thought I, “reads to me what he writes; he familiarly
converses with me, asks my opinion, and I freely give it. After all, I
am not to be pitied in my exile at St. Helena.”
15th.—Immediately after dinner the Emperor walked in his
favourite path. He had his coffee carried down to him in the garden,
and he drank it as he walked about. The conversation turned on
love. I must have made some very fine and sentimental remarks on
this important subject; for the Emperor laughed at what he styled
my prattle, and said that he understood none of my romantic
verbiage. Then speaking with an air of levity, he wished to make me
believe that he was better acquainted with sensations than
sentiments. I made free to remark that he was trying to be thought
worse than he was described to be in the authentic, but very secret,
accounts that were circulated about the palace. “And what was said
of me?” resumed he, with an air of gaiety. “Sire,” I replied, “it is
understood that, when in the summit of your power, you suffered
yourself to be bound in the chains of love; that you became a hero
of romance; that, fired by an unexpected resistance, you conceived
an attachment for a lady in private life; that you wrote her above a
dozen love-letters; and that her power over you prevailed so far as
to compel you to disguise yourself, and to visit her secretly and
alone, at her own residence, in the heart of Paris.”—”And how came
this to be known?” said he, smiling; which of course amounted to an
admission of the fact. “And it was doubtless added,” continued he,
“that that was the most imprudent act of my whole life; for, had my
mistress proved treacherous, what might not have been my fate—
alone and disguised, in the circumstances in which I was placed,
amidst the snares with which I was surrounded? But what more is
said of me?”—“Sire, it is affirmed that your Majesty’s posterity is not
confined to the King of Rome. The secret chronicle states that he
has two elder brothers: one the offspring of a fair foreigner, whom
you loved in a distant country; the other, the fruit of a connection
nearer at hand, in the bosom of your own capital. It was asserted
that both had been conveyed to Malmaison, before our departure;
the one brought by his mother, and the other introduced by his
tutor; and they were described to be the living portraits of their
father.”[23]
The Emperor laughed much at the extent of my information, as he
termed it; and being now in a merry vein, he began to take a frank
retrospect of his early years, relating many of the love-affairs and
numerous adventures in which he had been engaged. I omit the
first; amongst the second he mentioned a supper that took place in
the neighbourhood of the Saone at the commencement of the
Revolution, and at which he had been present in company with the
faithful Desmazzis. He described the whole with the utmost
pleasantry.—He had got himself, he observed, into a wasp’s nest,
where his patriotic eloquence had to contend strenuously against the
contrary doctrines of the other guests, and had nearly brought him
into a serious scrape. “You and I,” he continued, “were at that time
very far from each other.”—”Not so very far, in point of distance,
Sire,” replied, “though certainly very remote with respect to
doctrines. At that time I was also in the neighbourhood of the
Saone, on one of the quays of Lyons, where crowds of patriots were
declaiming against the cannon which they had just discovered in
some boats, and which they termed a counter-revolution. I very
inopportunely proposed that they should make sure of the cannon,
by administering to them the civic oath. However, I narrowly
escaped being hanged for my folly. You see, Sire, that I might
precisely at that moment have balanced your account, had any
disaster befallen you among your aristocrat companions.” This was
not the only curious coincidence that was mentioned in the course of
the evening. The Emperor, having related to me an interesting
circumstance that took place in 1788, said, “Where were you at that
time.”—“Sire,” replied I, after a few moments recollection, “I was
then at Martinique, supping every evening with the future Empress
Josephine.”
A shower of rain came on and we were obliged to retire from our
favourite path, which, the Emperor observed, we might probably at
a future period look back to with pleasure. “Perhaps so,” I replied;
“but certainly that will not be until we have forsaken it for ever.
Meanwhile we must content ourselves with naming it the Path of
Philosophy, since it cannot be called the Path of Lethe.”

THE FAUXBOURG SAINT-GERMAIN, &C.—THE EMPEROR’S FREEDOM FROM

PREJUDICE AND ILL-WILL.—CHARACTERISTIC LANGUAGE.

16th.—To-day the Emperor put some questions to me relative to

the Fauxbourg Saint-Germain; that last bulwark of the old
aristocracy, that refuge of old-fashioned prejudices; the Germanic
League, as he called it. I told him that, before his last misfortune, his
power had extended into every part of it: it had been invaded, and
its name alone remained; it had been shaken and vanquished by
glory; and the victories of Austerlitz and Jena, and the triumph of
Tilsit, had achieved its conquest. The younger portion of the
inhabitants, and all who had generous hearts, could not be
insensible to the glory of their country. The Emperor’s marriage with
Maria-Louisa gave it the last blow. The few malcontents who
remained were either those whose ambition had not been gratified,
and who are to be found in all classes, or some obstinate old men,
and silly old women, bewailing their past influence. All reasonable
and sensible persons had yielded to the superior talents of the Head
of the State, and endeavoured to console themselves for their losses
in the hope of a better prospect for their children. This became the
point towards which all their ideas were directed. They gave the
Emperor credit for his partiality to old family names; they agreed
that any one else in his place would have annihilated them. They
prized very highly the confidence with which the Emperor had
collected individuals of ancient family about his person; and they
valued him no less for the language he had made use of in making
choice of their children to serve in the army:—“These names belong
to France and to History; I am the guardian of their glory, I will not
allow them to perish.” These and other such expressions had gained
him numbers of proselytes. The Emperor here expressed his
apprehension that sufficient favour had not been shewn to this party.
“My system of amalgamation,” said he, “required it: I wished and
even directed favours to be conferred on them: but the ministers,
who were the great mediators, never properly fulfilled my real
intentions in that respect; either because they had not sufficient
foresight, or because they feared that they might thus create rivals
for favour, and diminish their own chances. M. Talleyrand, in
particular, always shewed great opposition to such a measure, and
always resisted my favourable intuitions towards the old nobility.” I
observed, however, that the greater part of those whom he had
placed near him had soon shewn themselves attached to his person;
that they had served him conscientiously, and had, generally
speaking, remained faithful to him at the critical moment. The
Emperor did not deny it, and even went so far as to say that the
twofold event of the King’s return and his own abdication must
naturally have had great influence on certain doctrines; and that, for
his own part, he could see a great difference between the same
conduct pursued in 1814 and in 1815.
And here I must observe that, since I have become acquainted
with the Emperor’s character, I have never known him to evince, for
a single moment, the least feeling of anger or animosity against
those individuals who had been most to blame in their conduct
towards him. He gives no great credit to those who distinguished
themselves by their good conduct: they had only done their duty. He
is not very indignant against those who acted basely; he partly saw
through their characters: they yielded to the impulses of their
nature. He speaks of them coolly, and without animosity; attributing
their conduct in some measure to existing circumstances, which he
acknowledged were of a very perplexing nature, and placing the rest
to the account of human weakness. Vanity was the ruin of Marmont:
“Posterity will justly cast a shade upon his character,” said he; “yet
his heart will be more valued than the memory of his career. The
conduct of Augereau was the result of his want of information, and
the baseness of those who surrounded him; that of Berthier, of his
want of spirit, and his absolute nullity of character.”
I remarked that the latter had let slip the best and easiest
opportunity of rendering himself for ever illustrious, by frankly
making his submission to the King, and intreating his Majesty’s
permission to withdraw from the world, and mourn in solitude the
fate of him who had honoured him with the title of his companion in
arms, and had called him his friend. “Yes,” said the Emperor; “even
this step, simple as it was, was beyond his power.”—“His talents, his
understanding,” said I, “had always been a subject of doubt with us.
Your Majesty’s choice, your confidence, your great attachment,
surprised us exceedingly.”—“To say the truth,” replied the Emperor,
“Berthier was not without talent, and I am far from wishing to
disavow his merit, or my partiality for him; but his talent and merit
were special and technical; beyond a limited point he had no mind
whatever: and then he was so undecided.”—I observed that “he
was, notwithstanding, full of pretensions and pride in his conduct
towards us.”—“Do you think, then, that the title of Favourite goes for
nothing?” said the Emperor. I added, that “he was very harsh and
overbearing.” “And what,” said he, “my dear Las Cases, is more
overbearing than weakness which feels itself protected by strength?
Look at women, for example.”
Berthier accompanied the Emperor in his carriage during his
campaigns. As they drove along, the Emperor would examine the
order-book and the report of the positions, whence he formed his
resolutions, adopted his plans, and arranged the necessary
movements. Berthier noted down his directions, and at the first
station they came to, or during the first moments allotted to rest,
whether by night or by day, he made out, in his turn, all the orders
and individual details with admirable regularity, precision, and
despatch. This was a kind of duty at which he shewed himself
always ready and indefatigable. “This was the special merit of
Berthier,” said the Emperor: “it was most valuable to me; no other
talent could have made up for the want of it.”
I now return to notice some characteristic traits of the Emperor.
He invariably speaks with perfect coolness, without passion, without
prejudice, and without resentment, of the events and the persons
connected with his life. It is evident that he would be capable of
becoming the ally of his most cruel enemy, and of living with the
man who had done him the greatest wrong. He speaks of his past
history as if it had occurred three centuries ago: in his recitals and
his observations he speaks the language of past ages: he is like a
spirit discoursing in the Elysian fields; his conversations are true
Dialogues of the Dead. He speaks of himself as of a third person;
noticing the Emperor’s actions, pointing out the faults with which
history may reproach him, and analysing the reasons and the
motives which might be alleged in his justification.
He never can excuse himself, he says, by throwing blame on
others, since he never followed any but his own decision. He may
complain, at the worst, of false information, but never of bad
counsel. He had surrounded himself with the best possible advisers,
but he had always adhered to his own opinion, and he was far from
repenting of having done so. “It is,” said he, “the indecision and
anarchy of agents which produce anarchy and feebleness in results.
In order to form a just opinion respecting the faults produced by the
sole personal decision of the Emperor, it will be necessary to throw
into the scale the great actions which he would have been prevented
from performing, and the other faults which he would have been
induced to commit, by those very counsels which he is blamed for
not having followed.”
In viewing the complicated circumstances of his fall, looks upon
things so much in a mass, and from so high a point, that individuals
escape his notice. He never evinces the least symptom of virulence
towards those of whom it might be supposed he has the greatest
reason to complain. His greatest mark of reprobation, and I have
had frequent occasion to notice it, is to preserve silence with respect
to them, whenever they are mentioned in his presence. But how
often has he not been heard to restrain the violent and less reserved
expressions of those about him? “You are not acquainted with men,”
he has said to us; “they are difficult to comprehend, if one wishes to
be strictly just. Can they understand or explain even their own
characters? Almost all those who abandoned me would, had I
continued to be prosperous, never, perhaps have dreamed of their
own defection. There are vices and virtues which depend on
circumstances. Our last trials were beyond all human strength!
Besides I was forsaken rather than betrayed; there was more of
weakness than of perfidy around me. It was the denial of St. Peter:
tears and repentance are probably at hand. And where will you find,
in the page of history, any one possessing a greater number of
friends and partisans? Who was ever more popular and more
beloved? Who was ever more ardently and deeply regretted? Here,
from this very rock, on viewing the present disorders in France, who
would not be tempted to say that I still reign there? The Kings and
princes, my allies, have remained faithful to me to the last, they
were carried away by the people in a mass: and those who were
around me, found themselves enveloped and overwhelmed by an
irresistible whirlwind.... No! human nature might have appeared in a
light still more odious, and I might have had still greater cause of
complaint!”

ON THE OFFICERS OF THE EMPEROR’S HOUSEHOLD IN

1814.—PLAN OF ADDRESS TO THE KING.

17th.—The Emperor asked me some questions to-day relative to

the officers of his household. With the exception of two or three, at
the most, who had drawn upon themselves the contempt of the very
party to which they had gone over, nothing could be said against
them: the majority had even evinced an ardent devotion to the
Emperor’s interests. The Emperor then made enquiries respecting
some of these individuals in particular, calling them by their names;
Welcome to our website – the ideal destination for book lovers and
knowledge seekers. With a mission to inspire endlessly, we offer a
vast collection of books, ranging from classic literary works to
specialized publications, self-development books, and children's
literature. Each book is a new journey of discovery, expanding
knowledge and enriching the soul of the reade

Our website is not just a platform for buying books, but a bridge
connecting readers to the timeless values of culture and wisdom. With
an elegant, user-friendly interface and an intelligent search system,
we are committed to providing a quick and convenient shopping
experience. Additionally, our special promotions and home delivery
services ensure that you save time and fully enjoy the joy of reading.

Let us accompany you on the journey of exploring knowledge and

personal growth!

ebookultra.com