PDF The Mathematics of Finite Networks An Introduction To Operator Graph Theory 1st Edition Michael Rudolph Download
PDF The Mathematics of Finite Networks An Introduction To Operator Graph Theory 1st Edition Michael Rudolph Download
PDF The Mathematics of Finite Networks An Introduction To Operator Graph Theory 1st Edition Michael Rudolph Download
com
https://ebookmeta.com/product/the-mathematics-of-
finite-networks-an-introduction-to-operator-graph-
theory-1st-edition-michael-rudolph/
OR CLICK BUTTON
DOWLOAD EBOOK
https://ebookmeta.com/product/spatial-networks-a-complete-
introduction-from-graph-theory-and-statistical-physics-to-real-
world-applications-barthelemy/
https://ebookmeta.com/product/discrete-mathematics-with-graph-
theory-1st-edition-santosh-kumar-yadav/
https://ebookmeta.com/product/graph-and-network-theory-an-
applied-approach-using-mathematica-1st-edition-michael-a-henning/
https://ebookmeta.com/product/introduction-to-graph-theory-with-
solutions-to-selected-problems-2nd-edition-khee-meng-koh/
Guide to Discrete Mathematics: An Accessible
Introduction to the History, Theory, Logic and
Applications 2nd Edition Gerard O'Regan
https://ebookmeta.com/product/guide-to-discrete-mathematics-an-
accessible-introduction-to-the-history-theory-logic-and-
applications-2nd-edition-gerard-oregan/
https://ebookmeta.com/product/an-introduction-to-chemistry-
michael-mosher/
https://ebookmeta.com/product/a-walk-through-combinatorics-an-
introduction-to-enumeration-and-graph-theory-and-selected-other-
topics-5th-edition-miklos-bona/
https://ebookmeta.com/product/qualitative-theory-of-odes-an-
introduction-to-dynamical-systems-theory-1st-edition-henryk-
zoladek/
https://ebookmeta.com/product/an-undergraduate-introduction-to-
financial-mathematics-4th-edition-buchanan/
The Mathematics of Finite Networks
Since the early eighteenth century, the theory of networks and graphs has matured into
an indispensable tool for describing countless real-world phenomena. However, the
study of large-scale features of a network often requires unrealistic limits, such as
taking the network size to infinity or assuming a continuum. These asymptotic and
analytic approaches can significantly diverge from real or simulated networks when
applied at the finite scales of real-world applications. This book offers an approach to
overcoming these limitations by introducing operator graph theory, an exact,
non-asymptotic set of tools combining graph theory with operator calculus. The book
is intended for mathematicians, physicists, and other scientists interested in discrete
finite systems and their graph-theoretical description, and in delineating the abstract
algebraic structures that characterise such systems. All the necessary background on
graph theory and operator calculus is included for readers to understand the potential
applications of operator graph theory.
M I C H A E L RU D O L P H
CNRS / Institute Denis Poisson, University of Tours
University Printing House, Cambridge CB2 8BS, United Kingdom
One Liberty Plaza, 20th Floor, New York, NY 10006, USA
477 Williamstown Road, Port Melbourne, VIC 3207, Australia
314–321, 3rd Floor, Plot 3, Splendor Forum, Jasola District Centre,
New Delhi – 110025, India
103 Penang Road, #05–06/07, Visioncrest Commercial, Singapore 238467
www.cambridge.org
Information on this title: www.cambridge.org/9781107134430
DOI: 10.1017/9781316466919
© Michael Rudolph 2022
This publication is in copyright. Subject to statutory exception
and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
First published 2022
A catalogue record for this publication is available from the British Library.
Library of Congress Cataloging-in-Publication Data
Names: Rudolph, Michael, 1969– author.
Title: The mathematics of finite networks : an introduction to operator
graph theory / Michael Rudolph.
Description: Cambridge ; New York, NY : Cambridge University Press, 2022. |
Includes bibliographical references and index.
Identifiers: LCCN 2021058077 (print) | LCCN 2021058078 (ebook) |
ISBN 9781107134430 (hardback) | ISBN 9781316466919 (epub)
Subjects: LCSH: Graph theory. | Operator theory. | System analysis. |
BISAC: COMPUTERS / General
Classification: LCC QA166 .R83 2022 (print) | LCC QA166 (ebook) |
DDC 511/.5–dc23/eng/20220126
LC record available at https://lccn.loc.gov/2021058077
LC ebook record available at https://lccn.loc.gov/2021058078
ISBN 978-1-107-13443-0 Hardback
Cambridge University Press has no responsibility for the persistence or accuracy of
URLs for external or third-party internet websites referred to in this publication
and does not guarantee that any content on such websites is, or will remain,
accurate or appropriate.
For Uesat
— in memoriam —
Contents
Preface page ix
1 Introduction 1
PART II APPLICATIONS
5 Generating Graphs 181
5.1 Erdős–Rényi Random Graphs 181
5.2 Gilbert Random Graphs 200
vii
viii Contents
Afterthought 325
Bibliography 327
Index of Notations 333
Subject Index 336
Preface
Imagine for a moment that you stroll through a beautiful city like London
with its dizzying array of inspiring detail. An overwhelming number of small
streets, unpredictably intersecting with one another, wind around busy alley-
ways which themselves appear to stretch endlessly into all directions. Finding
your way in this arguably chaotic mesh of interconnected paths and road-
ways by foot or by car without resorting to a memorised representation or
detailed map of its layout would require the often time-consuming search
through a vast number of possibilities until, eventually, the desired destination
is reached. Taking a bird’s-eye perspective, however, this confusing maze of
small streets and mews seems to magically disappear and reveal, with concen-
tric rings of expressways framing a star of boulevards which emanate from the
metropole’s centre, a different, much simpler pattern. Although such a coarser
structural view certainly helps to confine your search, it does so at the expense
of abstracting from finer details, – the small, narrow streets which are not only
an integral part of the cultural and historical lifeblood of a city like London but
reside at the very heart of its flair. Naturally, you might then wonder whether it
is possible to hold on to these pertinent details alongside the more crude struc-
tural peculiarities which emerge at larger scales and hold, not unlike a skeleton,
those details together.
When thinking of city maps, we instinctively envision a network of links
along which an ever-changing flow of traffic is carried. Such an idealised
description, however, is not limited to the maps we are all familiar with. From
the interactions between atoms and subatomic particles to the gravitational
forces which act between the billions of galaxies stretching across the known
universe, from the transmission of electrical signals in our brains to the com-
plexity of social interactions between people, most if not all of the phenomena
we encounter, consciously or not, all around us find a natural representation
in the form of networks. Indeed, it can be indubitably argued that the abstract
ix
x Preface
notion of interacting objects resides at the very heart of our conceptual under-
standing of nature as it touches upon the very fabric of physical reality with
its finite and discrete makeup. It is for that reason that we now find the math-
ematical study of interconnected objects, the theory of networks and graphs,
branching deeply through all fields of science and its innumerable applications.
Since its formal inception in the early eighteenth century, graph theory has
constructed a solid foundation on which its conceptual and mathematical roots
now comfortably rest. In doing so, it has amassed a plethora of useful tools
which allow for a thorough qualification and quantification of the sheer
uncountable number of networks we find all around us. Unfortunately, such a
realisation of graph-theoretical concepts does not come without its limitations.
All too often, it is the very nature of a studied system which also can, and does,
throw stones in the way of its graph-theoretical formalisation or viable qual-
ifying generalisations. Worse even, many real-world systems defiantly refuse
an analytical treatment with the toolkit of graph theory and can only be studied
numerically, thus, they are inadvertently yet inescapably tied to the efficiency
of our computational hardware, or the lack thereof.
To bypass these limitations, and to capture the essential structural properties
of real-world networks, the construction of simplified mathematical models
is required. However, as in the example of London’s map, many of the cur-
rent mathematical approaches provide only a bird’s-eye view, at the expense
of the finer details, in order to reveal the large-scale features of the network
that may exist, or to allow certain simplifications to become effective. Indeed,
these approaches regularly consider unrealistic limits, such as taking the net-
work size to infinity or abstracting from a network’s defining discrete makeup
by assuming a continuum. While such ‘asymptotic’ and ‘analytic’ approaches
can certainly provide simple representations of the dominant structural features
and, thus, often succeed at describing network structure in the aggregate, how-
ever, the equations at their core can diverge from real or simulated networks
quite significantly, especially when applied to the finite scales occurring in the
real world. What if the algorithms behind network models – and, for good
models, behind real-world data as well – could be cast in mathematical terms
that would allow for direct manipulations and, in some cases, exact solutions
without the necessity of imposing limits? What if such algorithms could be
cast into a form which would allow one to consider specific graph ensembles
all at once, without the need to computationally construct and assess concrete
individual realisation?
In this book, we aim to introduce, or at least propose, exactly such an
approach to graph theory. The mathematical language used for formalising
this discrete and finite algebraic framework lends itself to operator calculus,
Preface xi
I humbly admit that this book has been some years in the making, during a
time of great professional and personal hardship. For that reason, I must first
and foremost thank my ever-patient editor Lauren Cowles and her editorial
assistant Amy He for their unwavering support and understanding nature. The
same applies to my fellow colleagues at the Institute Denis Poisson in Tours
and Orleans, who gave me in the ‘Year of Crazy’ a new comfortable home and
awoke a memory of still-existing places in which only the sky is the limit, a
cherished memory I thought had long been lost. When thinking about the tran-
quility of a new home, it is impossible not to see Madame Gigi and Daniel
‘Rambo72’ who not only helped me to build, from nothing, a secluded Asgar-
dian paradise deep in the intoxicating beauty of Sarthe, France, but gave me
xii Preface
When on sunny days in the early years of the eighteenth century people has-
tened through the narrow streets of Königsberg, now Kaliningrad in Russia, it
was not for catching a breeze of fresh air, but for being the lucky first to unlock
the mystery of its now famous bridges. Not a valuable golden treasure was to
be unearthed, but the solution to the simple problem of finding a path over the
seven bridges that connect both banks of the river Pregel with the two islands
around which that beautiful Prussian town was built. The goal, however, was
not to find just any path, but one along which each bridge was crossed only
once. A futile endeavour indeed, and it is unknown how many of Königsberg’s
inhabitants and intrigued travellers from afar succumbed to this formidable
challenge before in 1736 the brilliant mind of a young genius named Leonard
Euler provided the definite answer. Such a path does not exist! [42].
What was Euler’s deep insight which allowed him to solve the problem of
Königsberg’s bridges without setting foot on any of them, the insight which
had eluded all mathematical treasure hunters, professional and recreational
alike, before him? In short, seven bridges connect the four main geographi-
cal areas of Königsberg, but only one or two of these areas can serve as start
and finish of the sought-after path through the city. This means that at least two
areas must be traversed and, in order to do so, must be connected by an even
number of bridges across the river Pregel. However, as a quick look at the map
of Königsberg reveals (Fig. 1.1a), each of its four areas was accessible only
through an odd number of bridges, thus rendering the existence of a valid path
an impossibility.
Although it still took many decades after Euler’s simple yet brilliant reso-
lution of the now famous Königsberg bridge problem before a coherent and
rigorous mathematical framework was established, it undoubtedly marks both
historically and conceptually the beginning of what is now known as network
or graph theory. Indeed, his contribution in this regard cannot be overstated.
1
2 Introduction
a b c
A B C D
C g
A
c d B
e
A D C
D
a b
B f weight
1 2 3
By stripping colourful yet less relevant details off the given original problem,
Euler single-handedly introduced an abstract notion which not just allowed
him to uncover its defining essence, but in fact touches upon the very fabric
of physical reality. Each real-world system is finite and discrete, comprised of
discernible parts which share links to, interact with or assert an action on other
parts of the system. Be it the plethora of elementary particles interacting with
one another through the exchange of virtual bosons, the complex dendritic and
axonal trees which carry electrical signals through a network of synaptically
interlinked neurons in our brains, or the arrangement of galaxies into large
filaments through the action of the gravitational force, all natural phenomena
we discovered so far can be abstractly described by and modelled as networks
of interconnected objects. These objects are called nodes or vertices, their con-
nections edges, and graph theory, in its most general sense, is the mathematical
framework for the study of networks formed by nodes and edges.
Truth be told, it would be an almost futile endeavour to present just a glimpse
of the large body of graph-theoretical literature which emerged in the past three
centuries since Euler’s time. For that reason, we must, and only humbly can,
refer the interested reader to the historically first textbook written by Kőnig
[65], the still-unmatched definite textbook on the subject by Harary [59] and
perhaps the most comprehensive presentation of modern applications of graph
Introduction 3
Figure 1.2 Solving the problem of Königsberg’s seven bridges through an exhaus-
tive search. Each branch in this tree constructs a possible Eulerian walk, with
bridges being crossed marked in grey and bridges already traversed in black.
Shown are only Eulerian walks starting from one bank of the river Pregel, omit-
ting all walks which begin by crossing the second bridge to the south. Each branch
of this tree of possibilities leaves at least one bridge uncrossed; thus, none of the
constructed walks constitutes a sought-after Eulerian path.
thus, an Eulerian path. Without having ever set foot in Königsberg, we just
solved the problem of its bridges by brute force on a piece of paper. But today,
one can find some 100 bridges and overpasses in Kaliningrad. How long would
it take now to draw a tree of all possible walks through the city?
In short, about 10150 times as long, give or take some orders of magnitude!
A formidable exercise which undoubtedly lies beyond the capacity of even the
most modern computers, let alone a wholeheartedly motivated researcher with
paper and pencil. The longer and more nuanced answer to this question neces-
sitates the introduction of the notion of computational complexity, a measure
Introduction 5
which, in the most general sense, quantifies the scaling behaviour of computa-
tional algorithms with respect to the magnitude of their input variables, in our
case the number of bridges and overpasses. To distinguish further the time it
takes to run a given algorithm from the storage space it requires, both defining
aspects quantifying the efficiency of a computational algorithm in relation to
the stringent constraints imposed by the utilised computational hardware, one
typically differentiates time complexity and space complexity, respectively.
Let us first take a look at the time complexity of our little exercise. Draw-
ing a single walk through the city which traverses each bridge – that is, each
edge in the associated graph – at most once certainly scales linearly with the
total number of bridges E. In mathematical terms, one expresses such a scal-
ing behaviour with the notion O(E). However, our exercise requires us to draw
all Eulerian walks and identify potential Eulerian paths, a rather brute-force
approach which is called exhaustive search. In order to do so, we must start at
every possible point and traverse all the possible paths from that point onwards
(Fig. 1.2), a task with a staggering time complexity of not less than O(E!).
Luckily for us, the original problem absolves us from having to cope with
such an unmanageable complexity, as it requires us only to decide wether an
Eulerian path exists or not. Indeed, following Euler’s inspiring solution, we
only need to calculate the number of edges connected to each node and test
wether any of these numbers is even, a task which can certainly be performed
in O(E) and leaves us off the hook, this time. Unfortunately, however, many
of the quantitative graph-theoretical methods and measures do not allow for
such an elegant approach, and they typically scale polynomially with the size
of the graph, that is, with a time complexity of O(E k ) for some k 1. Undeni-
ably, such a scaling is certainly better than the factorial scaling we encountered
above in a brute-force construction. But even with the development of compu-
tational hardware in the foreseeable future or the conception of novel, highly
efficient computational algorithms, such a polynomial time complexity is still
hard to cope with and likely renders the exact quantification of larger networks
difficult or even impossible for many decades to come.
The situation regarding space complexity does not look much less gloomy
when dealing with graph-theoretical formulations of real-world phenomena.
Although efficient representations for a small subset of special graphs exist,
generally one is left with a matrix representation of a graph’s adjacency rela-
tions (see Fig. 1.1c), an approach with a space complexity of O(N 2 ) in the
number of nodes N. Certainly, neither the seven bridges of Königsberg during
Euler’s time nor the more numerous bridges in current-day Kaliningrad pose a
representational problem here. But what about the graph describing the World
Wide Web with its hundreds of millions of nodes [1, 34]? Even if we assign
6 Introduction
only a Boolean number in the form of a computational bit to indicate the pres-
ence, or absence, of an edge between any given pair of nodes, it would require
a staggering 1000 terabytes to store the complete adjacency matrix of a graph
describing only 100 million linked websites. Luckily for us, in this specific
case, a linear list of source and target nodes would suffice, as each node of
this graph is, at average, connected to only seven or eight other nodes (e.g.,
see [26, 54] but also [77]). But even such a sparse representation would still
require about one gigabyte of memory alone for storing the graph’s defining
structure, without any computational analysis having yet been performed on
it. When considering still larger or less sparsely connected graphs, such as the
network of neurons in the human brain with thousands of connections embed-
ding each of its 100 billion nodes [11, 23], we quickly run out of options as
the memory requirements for their representation alone will breach the storage
capacity made available in modern computers, perhaps even the computational
devices of the next few generations.
The question, then, is, where does this leave us? Of course, a computa-
tional analysis is just one of several available options when facing a graph-
theoretical formulation of a natural phenomenon. Indeed, the very limitation of
a purely numerical approach has led to the development of a whole host of new
and exciting methods and continues to stimulate novel ideas and techniques
beyond the mere optimisation of computational algorithms. Arguably, one of
the conceptually most important and fruitful of these ideas is the distinction of
real-world networks into classes of simplified, heavily idealised graph models.
Examples of such idealised models include perhaps the most-widely known
Erdős–Rényi model [22, 38] which generally serves as the prototype when
referring to random graphs, the small-world graph of Watts and Strogatz [109]
inspired by Stanley Milgram’s infamous ‘six degrees of separation’ [78] and
the scale-free graph of Barabási and Albert [12] which captures the property of
self-similarity found in complex systems, to mention but a few. Most of these
models are conceived by abstracting from many of the finer structural pecu-
liarities we observe in real-world networks, details which are, as we argued
above, increasingly harder to delineate through computational approaches in
cases where larger or densely connected networks are concerned. With these
graph models typically requiring less than a handful of parameters for their
construction and characterisation, the simplified nature of the models then not
only allows for the application of numerical optimisation techniques which
significantly reduce their computational complexity, but eventually opens the
door for a more rigorous mathematical exploration by allowing for an algebraic
formalisation of their algorithmic construction.
Introduction 7
In this book, we will propose and motivate exactly such a rigorous math-
ematical approach. By embarking on this adventurous journey, however, we
will seek to avoid at least one of the major caveats accompanying many of the
idealised graph-theoretical models proposed as descriptions of real-world phe-
nomena since the time of Euler. To briefly illustrate this caveat, let us consider
perhaps the simplest of these abstract models, that of the Erdős–Rényi random
graph. For each point in its two-dimensional parameter space, which is spanned
by the number of nodes and the number of edges, or, alternatively but with sub-
tle conceptual differences, the connection probability between pairs of nodes,
a whole ensemble of possible realisations exists, each of which describes a
distinct connectivity pattern between all possible pairs of nodes. Although the
algebraic formalisation of this model allows, at least in principle, access to this
ensemble, most of the defining characteristics of the Erdős–Rényi model can
still only be assessed numerically, ideally so over a large number of realisations
for each parameter set in order to ensure statistical significance of the obtained
results.
Naturally, just as in the case of graph-theoretical descriptions of real-world
phenomena, such a computational approach is subject to the stringent limita-
tions mentioned earlier, especially when models of larger graphs are consid-
ered. Due to its mathematically well-rooted conception, however, the Erdős–
Rényi random graph model also invites a more rigorous analytical examina-
tion. Some of its characteristics, such as the number and distribution of edges
connected to each node, are indeed governed by exact formulae which, due to
the random makeup of the Erdős–Rényi model, are probabilistic in nature and
deliver the expectation value for those characteristics across the entire param-
eter space and, thus, the whole ensemble of possible realisations. With other
properties – such as the shortest distance between any pair of nodes, measured
in number of edges which must at least be traversed to reach a target node, or
the clustering, quantified by the abundance of triangular cycles – we are unfor-
tunately not so lucky. Indeed, many if not most of the properties allow for an
analytical take only in some parts of the model’s parameter space, that is, under
certain and often stringent conditions imposed on the graph’s makeup.
It was the meticulous work of Paul Erdős and Alfréd Rényi in the early sec-
ond half of the last century which eventually established what is now called
random graph theory, the mathematical study of properties of the probability
space of random graphs, a field which stretches today with applications far
beyond the random graph model named after its original architects (for a com-
prehensive introduction, see [22]). The conceptual core of this theory defines a
set of conditions which allow for a mathematically rigorous quantification of a
8 Introduction
random graph’s properties. But – and herein lies the problem or caveat hinted at
above – these very conditions are rooted in asymptotic considerations, specif-
ically and most importantly the heavy-weighting assumption which demands
that the network size should become infinitely large. Although for the math-
savvy reader such a demand might certainly appear as nothing unusual, as
something being widely employed in mathematics and its many applications –
a trick if you will, to establish some form of analytical framework for describ-
ing a physical phenomenon or, as in the case of random graphs, for formalising
a theoretical model – we must stop here and evaluate its potentially pernicious
repercussions and pitfalls.
The important question to ask before issuing any demands on our approach
is, do we find in nature any system which we know for certain is infinite in
one or more of its properties or, alternatively, is genuinely continuous in an
analytical or, if you like, Cantorian sense? Or is it that our models, describing
these systems, assume such an idealising makeup for reasons of simplicity, or
the sheer lack of a more suitable approach in tune with our study of nature
and, thus, epistemology? We could go as far as posing the heretical question
of whether the fabric of physical reality itself is algebraic, that is finite and
discrete, or analytic, that is infinite and continuous, in nature? Although on
first sight such questions might appear far removed from an graph-theoretical
inquiry of our world – indeed, as being of little or no relevance today as our
analytical formulation of real-world phenomena has undeniably demonstrated
tremendous success – their gravity continues to penetrate philosophical discus-
sions and, arguably, lies at the very bedrock of modern science and its mathe-
matical foundation (a comprehensive review of arguments from both sides can
be found in Hagar [56]).
To illustrate this point, arguably naively so, in the context of this book, let
us take a look at a physical phenomenon which many of us encounter on a
daily basis, namely preparing, and eventually enjoying, a nice cup of coffee.
Hot water is poured on finely grained coffee and finds its way – ‘percolates’ –
through the porous brown powder to finally yield the tasty, longed-for bever-
age. This mundane process is just one example of a whole class of phenomena
in condensed matter physics which finds a natural representation in terms of
random graph theory. However, due to its analytical foundation, the obtained
results are, at best, valid only in the case that the number of nodes in the asso-
ciated graph-theoretical models approaches infinity – that is, we deal with an
impossible-to-imagine amount of coffee in front of us – or if the connection
probability between nodes is small enough that the connectivity pattern of the
model graph resembles that of a tree, a structure that is devoid of cycles or
loops and resembles, to stick with our illustrative example, anything but the
Introduction 9
coffee we are all familiar with. Indeed, the reality of making coffee certainly
looks much different from the model we use to describe it, as we do not have
at our disposal an (almost) infinite quantity of coffee powder for quenching
our desire, nor do we find a resemblance between a sparsely connected tree
and the gaps separating the coffee grains through which the water finds its way
into our cups!
Despite the fact that theoretical models, by their very conceptual defini-
tion, can provide only a simplified, more abstract view of natural phenomena,
the above highlighted differences are certainly, or at least arguably, not even
close to being a viable reflection of the real-world system in question. Thus,
reiterating the above questions, we must ask whether random graph theory,
indeed graph theory in general, with its analytical core is the right mathemati-
cal approach to describe the discrete and finite phenomena, such as percolation,
which we find all around us. After all, every system we can discern in nature,
in lack of a proof to the opposite, is both finite and discrete. An answer to this
question becomes even more pressing when considering systems which, by
their very definition, are bound in size, such as the game of chess, or display
boundaries which crucially shape the system’s defining behavioural charac-
teristics, such as the aromatic powder in a coffee machine. In such cases we
have no choice but to strictly adhere to a finite description, as each simplifying
asymptotic assumption will necessarily push us outside the defining premise
of the studied phenomenon and eventually deliver a model which no longer
observes the phenomenon being considered.
Is it possible to conceive of a mathematical framework which is capable
of more precisely capturing finite and discrete graph-theoretical models, thus
allowing for a more viable and, arguably, more accurate and thus valid descrip-
tion of real-world phenomena? In this book, we will answer this question affir-
matively and motivate a rigorous algebraic approach for the construction, anal-
ysis and characterisation of finite graphs. This approach, which we will term
operator graph theory, comes at the price of abandoning the original definition
of a network, or graph, which resides at the very core of classical graph theory
in favour of a more dynamic, operational viewpoint. A graph will no longer
be a mere static collection of nodes and edges, but instead a construct which
can evolve and change due to the actions of operators. We will argue that by
studying the algebraic properties typically associated with such operators, this
approach opens up a new dimension of qualitative and quantitative insights
into the very properties which define a given finite graph, insights which find
a viable mathematical representation without resorting to the dangerous tool
bag of asymptotic approximations and throttling limitations. In order to further
stress, on epistemological grounds and in full awareness of its heretical nature,
10 Introduction
13
14 Classical Graph Theory
Although this now most commonly cited definition constitutes the very foun-
dation of classical graph theory, it must be cautioned that the defined notion
also remains rather limited in its scope and suffices only as a portrayal of the
principal algebraic building blocks which describe the structural makeup of a
given finite graph. Indeed, as each student of real-world graphs will quickly
realise, an algebraically more precise description and subsequent exploration
of a network’s characteristics necessitates a more suitable mathematical repre-
sentation of the set of nodes and their existing relations. This often includes, or
even requires, a quantification of the strength, or weight, of the relations estab-
lished between the nodes beyond the mere existence, or absence, of adjacency
relations themselves. Such a representation can be achieved by assigning to
each pair of nodes (i, j) with i, j ∈ N a number wi j ∈ Q, such that
wi j , 0 iff (i, j) ∈ E,
(
wi j : N × N → Q with (2.1)
wi j = 0 otherwise.
Referring to a node i by its order in the set N, this mapping is typically rep-
resented as |N| × |N| matrix and called the weight matrix w of the graph G.
Throughout this book – and, arguably, somewhat in defiance of the vast major-
ity of introductory textbooks of graph theory – we will accommodate this more
2.1 Graphs and Networks 15
general stance and consider each of a finite graph’s edges as being weighted.
As we will see below, the various types of networks utilised in the applied
graph-theoretical literature then appear as special cases defined by imposing
constraints on the weight matrix.
Before utilising Definition 2.1 and the weight matrix (2.1) to furnish a classi-
cal graph-theoretical framework, however, let us briefly illustrate the algebraic
representation of a graph in the case of Königsberg’s bridges. The city’s four
areas (Kneiphof, Haberberg or Vorstadt, Löbenicht or Altstadt, and Lomse)
comprise all nodes of this graph and find a suitable representation in form of
the set N = {1, 2, 3, 4} by assigning a number to each area consecutively. With
this enumeration, the seven bridges connecting between the city’s areas can
then be listed as pairs (1, 2) and (2, 1) for the Grüne Brücke and Köttelbrücke,
(1, 3) and (3, 1) for the Krämerbrücke and Schmiedebrücke, (1, 4) and (4, 1)
for the Honigbrücke, (2, 4) and (4, 2) for the Hohebrücke, and (3, 4) and (4, 3)
for the Holzbrücke, and comprise the elements of E. In order to account for the
two bridges connecting Kneiphof and Haberberg, the adjacency relations (1, 2)
and (2, 1) are both assigned a weight of two. The same applies to the adjacency
relation between nodes 1 (Kneiphof) and 3 (Löbenicht), thus leaving us with
the weight matrix
0 2 2 1
2 0 0 1
w = (2.2)
2 0 0 1
1 1 1 0
as not only an algebraic description of the graph associated with Königsberg’s
bridges (Fig. 2.1a), but also a suitable mathematical representation which will
ease the further analysis of this historically important graph.
1 4
d e
Figure 2.1 Examples of various types of graphs. a: The seven bridges of Königs-
berg form an undirected multigraph, with the weight of the graph’s edges repre-
senting the number of bridges connecting each of the city’s four areas (thin solid:
1 bridge, thick solid: 2 bridges). b: A relational digraph in which pairs of nodes
are connected by directed edges. c: The same graph with undirected edges, form-
ing a simple graph. d: A self-looped undirected multigraph. Although all edges
between pairs of distinct nodes are unweighted, a self-loop is considered as either
two directed links each carrying a weight of 1 or, as shown here, as one undirected
edge of weight 2. e: A self-looped directed multigraph in which each self-loop is
represented by two directed edges (indicated by double-arrows) each carrying a
weight of 1. Graphs a, c and d are examples of undirected graphs, while b and e
are digraphs. Due to the presence of weighted edges and self-loops, graphs dis-
played in a, d and e require a weight matrix (2.3) for their correct representation,
while in the cases displayed in b and c, an adjacency matrix (2.4) suffices.
In terms of the weight matrix, we can thus define a graph as being directed
if ∃ (i, j) ∈ E with i, j ∈ N : wi j , 0 ∧ w ji = 0. On the other hand, the
Königsberg’s bridge graph is a classical example of an undirected graph, as
here for each (i, j) ∈ E also ( j, i) ∈ E such that wi j = w ji for all i, j ∈ N
and each unordered pair (i, j) : wi j , 0 defines an undirected edge (Fig. 2.1a).
On closer inspection, we furthermore notice that in this particular graph none
of its nodes serves both as source and target of an edge. A graph for which
@ i ∈ N : (i, i) ∈ E is called a not self-looped graph, and a self-looped graph if
and only if there exists at least one self-loop, that is, an edge which connects a
node with itself such that wii , 0 for at least one i ∈ N.
Following historically established terminology, the assignment of a weight
wi j ∈ Q to each adjacency relation (i, j) ∈ E of a directed graph with wi j , 1
2.1 Graphs and Networks 17
Table 2.1 Major graph types and their weight matrix representation.
directed ∃i, j ∈ N : wi j , 0 ∧ w ji = 0 undirected ∀i, j ∈ N : wi j = w ji
self-looped ∃i ∈ N : wii , 0 not self-looped ∀i ∈ N : wii = 0
weighted ∃(i, j) ∈ E : wi j , 1 unweighted ∀i, j ∈ N : wi j ∈ B
multidigraph ∀i, j ∈ N: wi j ∈ N0 ∧ ∃i, j ∈ N : wi j , w ji ∧ ∃(i, j) ∈ E : wi j > 1
multigraph ∀i, j ∈ N: wi j ∈ N0 ∧ wi j = w ji ∧ ∃(i, j) ∈ E : wi j > 1
relational digraph ∀i, j ∈ N: wi j ∈ B ∧ ∀i ∈ N : wii = 0
simple graph ∀i, j ∈ N: wi j ∈ B ∧ wi j = w ji ∧ ∀i ∈ N : wii = 0
for at least one (i, j) ∈ E defines a weighted digraph. A first special case of
a weighted digraph is obtained if we set wi j = w ji for each (i, j) ∈ E. In this
case, the graph is undirected and simply called a weighted graph. By further
restricting each non-zero element of the weight matrix to the set of natural
numbers N, we have
wi j ∈ N iff (i, j) ∈ E,
(
wi j : N × N → N0 with (2.3)
wi j = 0 otherwise,
and the corresponding graph is called a directed multigraph, or multidigraph,
if ∃(i, j) ∈ E : wi j > 1, and an undirected multigraph or simply multigraph if,
in addition, its weight matrix is symmetric. In a multigraph or multidigraph,
each adjacency relation (i, j) ∈ E with weight wi j > 1 can be viewed as repre-
senting either wi j edges of weight 1 linking nodes i and j, or a single multiedge
of weight wi j . Recalling the weight matrix (2.2), the graph of Königsberg’s
bridges provides a classical example of such a multigraph (Fig. 2.1a). A final
special type of weighted digraph we will entertain here is realised if each ele-
ment of the weight matrix of a not self-looped digraph maps a pair of nodes
into the Boolean set B := {0, 1}. The weight matrix is then commonly referred
to as adjacency matrix a with elements
1 iff (i, j) ∈ E and i , j,
(
ai j : N × N → B with ai j = (2.4)
0 otherwise,
and defines a relational digraph (Fig. 2.1b) or, in the undirected case with
symmetric adjacency matrix ai j = a ji , a simple graph (Fig. 2.1c).
It is this type of simple – that is, undirected, not self-looped and single-
edged – graph which most prominently features in the contemporary applied
graph-theoretical literature. Indeed, and somewhat unfortunately, the over-
whelming majority of notions and measures introduced into graph theory since
Euler’s time is attuned to utilising the adjacency matrix (2.4), a simplification
which, arguably, only rarely is justified when dealing with graph-theoretical
descriptions of real-world phenomena. For that reason, the presentation in the
18 Classical Graph Theory
remainder of this book will focus, whenever possible, on the most general, i.e.
weighted, directed and self-looped, type of graph formalised in Definition 2.1
together with equation (2.1), and highlight, whenever necessary, results and
subtleties for the specific types of graphs introduced above as special cases.
One of these subtleties concerns the incorporation of nodes with self-loops
in a graph. In order to explore the complication associated with the presence
of self-loops, let us first consider the case of an unweighted undirected graph.
Here, for each ordered pair (i, j) ∈ E of distinct nodes, ( j, i) must also be an
element of E. If the source and target node of an edge are identical and each
edge carries a weight of 1, we obtain for the associated adjacency relations
which suggests that each pair (i, i) must be counted twice in E. Although it is
possible to account for this doubling in the mathematical definition of graph-
theoretical quantities by treating self-loops separately from undirected edges
between distinct nodes, it is often more convenient to represent a self-loop as
one edge and assign to it a weight of two, as illustrated above and exemplified
in Fig. 2.1d. However, such an approach carries the subtle disadvantage that
unweighted undirected graphs with self-loops can no longer be represented by
employing an adjacency matrix (2.4), as in this case aii = 2 < B if (i, i) ∈ E, and
thus limits the commonly used notion of an adjacency matrix to an unweighted
undirected graph without self-loops; that is, simple graphs. In the presence of
self-loops, such graphs are multigraphs and therefore require a weight matrix
with wii = 2 ∀ (i, i) ∈ E for their representation.
Naturally, this convention can be generalised to incorporate undirected multi-
graphs with self-loops and arbitrary edge weights, in which case wii = 2N0 for
all i ∈ N, as well as undirected graphs with edges taking weights in Q. More-
over, and most importantly, the above approach with its associated conditions
imposed on the diagonal elements of the weight matrices (2.1) and (2.3) also
applies to directed graphs if we consider that, by definition, each self-looped
node serves as both source and target of an adjacency relation and, thus, each
self-loop necessarily corresponds to an undirected edge (Fig. 2.1e). Indeed, in
a directed graph, an undirected adjacency relation of weight w between two
distinct nodes i and j can be generated by either adding a reciprocal directed
edge to E (symmetrisation) or discarding the direction of the given edge which,
again, necessitates the addition of a reciprocal directed edge to E (doubling).
In both cases, a self-loop is conceptually consistent with the presence of two
2.1 Graphs and Networks 19
w w 2w
i=j
i j ⇒ i ⇒ i
w
⇒ w w
i j ⇒
w w 2w.
i=j
i j ⇒ i ⇒ i
w w
e f g h
however, that such a generalisation runs into conflict somewhat with the type
of directed graphs introduced in the previous section, a situation which will be
salvaged in Section 2.2.3 with the introduction of the type of weight-directed
graph. An illustrative example of a complete digraph is given by a small collab-
oration network of scientists in which each individual member will contribute,
depending on expertise and relevance, to further the other members’ research.
As a representative example, for a complete graph we find the network of grav-
itational interactions between galaxies, as each galaxy exerts an attractive force
on all others while the strength of this force and, thus, the weight of each edge
in the associated graph will vary between different pairs as a function of the
galaxies’ spatial distance.
Inherently related to the notion of a complete graph is that of a graph com-
plement, also called a graph edge-complement or graph inverse [53]. Given a
relational digraph G = (N, E), the complement of G is typically defined as a
relational digraph Ḡ = (N̄, Ē) with the same set of nodes N̄ = N and a set
of edges Ē such that for all i, j ∈ N : i , j we have (i, j) ∈ Ē if and only if
(i, j) < E. If ai j denotes the adjacency matrix of G, the adjacency matrix of its
complement is thus given by āi j = 1 − ai j for all i, j ∈ N : i , j and āii = 0.
With this, the sum of G and its complement Ḡ then yields a complete digraph
formed by the set of nodes N. Here, the graph sum of two relational digraphs
2.1 Graphs and Networks 21
G1 = (N, E1 ) and G2 = (N, E2 ) sharing the same set of nodes N and edge sets
E1 , E2 : E1 ∩E2 = ∅ is defined as the relational digraph G1 5 G2 := (N, E1 ∪E2 ).
In case two given graphs share at least one common edge or at least one of the
graphs is weighted, their graph sum will necessarily yield a weighted graph.
This generalisation allows naturally extending the notion of a graph comple-
ment to include not self-looped weighted graphs, such that in the remainder of
this book we will refer to a graph complement Ḡ of a not self-looped graph or
digraph G if their sum G 5 Ḡ respectively forms a complete graph or digraph.
Another important graph class is comprised of simple graphs in which each
node establishes the same number of edges to other nodes in the graph, and
is called a regular graph or k-regular graph (Fig. 2.2b). Mathematically, the
requirement defining a k-regular graph can be expressed as
X
∀i ∈ N: ai j = k = const,
j∈N
where the sum runs over all nodes in the graph. As in the case of a complete
graph, this notion also can be generalised to cover relational digraphs, in which
case each node is required to have the same number of outgoing and incoming
directed edges, that is,
X X
∀i ∈ N: ai j = a ji = k = const .
j∈N j∈N
Such a graph is then called a k-regular digraph. Finally, if the edges of a not
self-looped directed graph are weighted, we can formulate
X X
∀i ∈ N: wi j = w ji = w = const
j∈N j∈N
Table 2.2 Important graph classes and their weight matrix representation.
Class Typea Representation
d ud w uw
complete • • • • ∀i, j ∈ N, i , j : wi j , 0
k-regular • • • ∀i, j ∈ N P
: wi j ∈ B ∧P ∀i ∈ N : j∈N wi j = j∈N w ji = k
P P
w-regular • • • ∀i ∈ N : j∈N wi j = j∈N w ji = w
connected • • • ∀i, j ∈ N, i , j ∃ jl ∈ N, l ∈ [1, k], j1 = i, jk+1 = j : w jl jl+1 , 0
bipartite • • • • ∃N 0 ⊂ N, N 0 , ∅ such that
∀i, j ∈ N 0 : wi j = 0 ∧ ∀i, j ∈ N \ N 0 : wi j = 0
and ∃i ∈ N 0 , j ∈ N \ N 0 : wi j , 0 ∨ w ji , 0
biclique • • • • ∃N 0 ⊂ N, N 0 , ∅ such that
∀i, j ∈ N 0 : wi j = 0 ∧ ∀i, j ∈ N \ N 0 : wi j = 0
and ∀i ∈ N 0 , j ∈ N \ N 0 : wi j , 0 ∧ w ji , 0
a Graph types (see Table 2.1) to which the given representation applies. Here d, ud, w and uw
denote directed, undirected, weighted and unweighted graphs respectively.
N1 and N2 of nodes can be identified such that adjacency relations exist only
between nodes which reside in different subsets, that is, if N = N1 ∪ N1 and
for all (i, j) ∈ E : (i ∈ N1 ∧ j ∈ N2 ) ∨ (i ∈ N2 ∧ j ∈ N1 ), the graph is
called a bipartite graph, and a complete bipartite graph, or simply biclique,
in the case all nodes in N1 are connected to all nodes in set N2 (Fig. 2.2e).
This notion applies to both weighted and unweighted as well as directed and
undirected not self-looped graphs. Due to their conceptional simplicity, bipar-
tite graphs find widespread use not only as models of interactions between
groups of objects whose members share common characteristics [10], but also
in the modelling of feedforward networks such as artificial neural networks
[46, 75] and a variety of biological neural networks along with their functional
properties [93].
The final and, historically, arguably most intriguing class of graphs to be
introduced in this abridged presentation of classical graph theory is comprised
of networks which allow for a planar embedding. Such an embedding is defined
as a visual representation of a given graph on a two-dimensional surface such
that edges intersect only at the nodes to which they connect. A simple graph
which allows for such a representation is then called a planar graph, and it
is called a non-planar graph if no such planar embedding exists. The graph of
Königsberg’s bridges (Fig 2.2f) and the complete simple graph with four nodes
depicted in Fig 2.2g are planar, while the bipartite graph shown in Fig 2.2e and
complete simple graph with five nodes in Fig 2.2h are examples of non-planar
graphs. Indeed, the latter two graphs are instrumental in deciding whether a
given simple graph allows for a planar embedding or not, a remarkable insight
formalised in Kuratowski’s theorem [72]. Unfortunately, a just presentation
of this historically significant result along with its many proofs [64, 104] is
beyond the scope of this brief introduction, and we must refer the interested
reader to the excellent textbook by Nishizeki and Chiba [89].
Instead, let us take a closer look at another important contribution which
not only resides at the very heart of the theory of planar graphs but unearthed
one of the most beautiful equations in the theory of finite graphs altogether,
Euler’s formula. Owing to its definition, a planar graph G = (N, E) necessarily
divides the plane into distinct regions or faces, each encircled by a subset of
edges forming a cycle. Let us denote with F the set of these faces, including
the unbounded region outside the graph, such that the cardinality |F | yields the
number of faces. Euler’s formula then states that the number of nodes, given by
|N|, the number of edges, given by |E|, and the number of faces |F | are subject
to the strict identity
for any finite connected planar graph G. In the graph of Königsberg’s bridges
(Fig. 2.2f), for example, we find |N| = 4, |E| = 5 and |F | = 3. For the complete
graph with |N| = 4 depicted in Fig. 2.2g, we have |E| = 6 and |F | = 4. Thus
both examples of planar graphs satisfy (2.5).
Due to the historical significance of Euler’s formula, and its importance for
many real-world applications, we will present here a simple proof demonstrat-
ing its validity. To that end, let us first introduce the notion of a tree graph as
a simple planar graph which has only one face, and show by induction on the
number of nodes that (2.5) holds for this type of graph. If a tree graph contains
only one node and no edge, Euler’s formula clearly applies. We now assume
that (2.5) is true for all tree graphs with |N| ≥ 1 nodes. Following its defini-
tion, each such graph must have |N| − 1 edges. Let T 0 = (N 0 , E0 ) be a tree
graph with |N 0 | = |N| + 1 nodes and |E0 | edges. Starting at any of its nodes, we
then follow an arbitrary path comprised of edges between connected nodes but
without revisiting nodes. As a tree graph has only one face, and thus no cycle,
such a path will not return to the initially chosen node but instead must neces-
sarily terminate at a node which is connected by only one edge. By removing
the path’s final node along with its sole edge, we then construct a new tree
T 00 = (N 00 , E00 ) with |N 00 | = |N 0 | − 1 = |N| nodes and |E00 | = |E0 | − 1 edges. By
the induction hypothesis, as T 00 is a tree graph with |N| nodes, we must have
|E00 | = |N| − 1, from which |E0 | = |N| for T 0 and the validity of Euler’s formula
for trees with |N| + 1 nodes, thus for all tree graphs, follows.
For arbitrary planar graphs, the proof of Euler’s formula (2.5) follows a simi-
lar approach using induction on the number of edges. A connected planar graph
with |N| ≥ 1 nodes must have at least |N| − 1 edges. If the graph has exactly
|N| − 1 edges, it is a tree graph and, as shown above, Euler’s formula cer-
tainly applies. Assuming that (2.5) is true for graphs with |N| ≥ 1 nodes and
|E| ≥ |N| − 1 edges, we have F ≥ 1. Let G0 = (N 0 , E0 ) be a connected simple
graph with |N 0 | = |N| nodes and |E0 | = |E| + 1 edges. Then F > 1, and G must
have at least one cycle. By choosing any of these cycles and removing one of
the chosen cycle’s edges, we then construct a new graph G00 = (N 00 , E00 ) with
|N 00 | = |N 0 | = |N| nodes, |E00 | = |E0 | − 1 = |E| edges and two merged faces,
i.e. |F 00 | = |F 0 | − 1. By the induction hypothesis, as G00 is a connected planar
graph with |N| nodes and |E| edges, we have |N 00 |−|E00 |+|F 00 | = 2, from which
|N 0 | − (|E0 | − 1) + (|F 0 | − 1) = |N 0 | − |E0 | + |F 0 | = 2 for G0 and eventually the
validity of (2.5) for all finite connected planar graphs follows.
The theory of planar graphs undoubtedly forms the root of one of the his-
torically most influential branches of graph theory, from which a whole host
of important real-world applications was spawned. Certainly high among these
applications ranks the four-colour theorem, which states that four colours
2.1 Graphs and Networks 25
Figure 2.3 The Platonic solids (tetrahedron, cube, octahedron, dodecahedron and
icosahedron) and their associated planar graphs (tetrahedral graph, cubical graph,
octahedral graph, dodecahedral graph and icosahedral graph). Using Euler’s for-
mula, it can easily be shown that only five Platonic solids are possible.
suffice to draw the faces of a planar graph in such a way that no adjacent faces
have the same colour [7, 8, 9], and which provides the mathematical founda-
tion for the accessible visual representation of contiguous regions on the many
maps we encounter in our everyday lives. Unfortunately, our necessarily brief
introduction to classical graph theory does not permit a worthy presentation of
this or other important results of planar graph theory. However, we will enter-
tain the intrigued reader with a beautiful yet simple graph-theoretical proof of
another historically significant mathematical theorem which states that there
can be only five Platonic solids: the tetrahedron, cube, octahedron, dodeca-
hedron and icosahedron (Fig. 2.3, top). Platonic solids are defined as convex
polyhedra whose faces are comprised of congruent convex regular polygons.
These iconic geometrical objects not only were the subject of extensive studies
in antiquity, but they played an important role in a variety of fields throughout
history. It was Euclid in his Elements who first showed rigorously that only
five such solids are possible [41].
Here we will present a simple proof utilising Euler’s formula (2.5). To that
end, we first represent each solid as a planar graph G = (N, E), called a
Platonic graph, by associating a node with each corner of the solid, and a
graph edge with each of the solid’s edges (Fig. 2.3). By construction, the
faces of the Platonic graphs correspond to the faces of the Platonic solid,
with each face of a given solid being encircled by the same number of edges
NeF , and each node establishing the same number of adjacency relations NeN .
As furthermore, each edge of our Platonic graphs belongs to exactly two dis-
tinct faces, we have NeF |F | = 2|E|, and as each edge links two distinct nodes,
we find NeN |N| = 2|E|. Employing Euler’s formula and some simple algebraic
26 Classical Graph Theory
Table 2.3 Platonic graphs and their associated five Platonic solids.
Platonic solid Platonic graph NeN NeF |N| |E| |F |
tetrahedron tetrahedral graph 3 3 4 6 4
cube cubical graph 3 4 8 12 6
octahedron octahedral graph 4 3 6 12 8
dodecahedron dodecahedral graph 3 5 20 30 12
icosahedron icosahedral graph 5 3 12 30 20
manipulations, which we leave as an exercise for the reader, we are then left
with the diophantine equation
1 1 1 1
N
+ F = +
Ne Ne |E| 2
as a necessary condition governing the existence of Platonic graphs. By noting
that NeN ≥ 3 and NeF ≥ 3, and that Platonic graphs must have at least one
edge such that |E| ≥ 1, we find that this equation can only have the five integer
solutions for NeN , NeF and |E| listed in Table 2.3, thus demonstrating that only
five Platonic graphs, hence only five Platonic solids, are possible.
With this brief excursion into the realms of our antique forefathers, we will
conclude our brief introduction to the basic notions and conceptual foundation
at the heart of the theory of graphs and networks. These preliminaries should
suffice as a framework for the material presented in the remainder of this book.
In the next section, we will return to general weighted digraphs and introduce,
on more rigorous grounds, some of the concepts and concrete quantities which
are commonly employed throughout the applied graph-theoretical literature to
characterise and measure graphs.
and we must refer the interested reader to the recent textbook of Newman
[86] or the excellent review by Boccaletti and colleagues [20] for a more
comprehensive and detailed overview. Instead, we will focus here on only
the most basic and prominent measures of relevance for presentation in later
chapters, and we will categorise these measures into general measures, degree-
based, weight-based and distance-based measures, as well as measures asso-
ciated with network motifs and graph spectra. With this necessarily abridged
presentation, we nevertheless hope to motivate the serious student to further
explore this vast and fluent branch of graph theory and its countless real-world
applications.
1 iff ∃ i, j ∈ N, (i, j) ∈ E : wi j , 0 ∧ w ji = 0,
(
λd ∈ B : λd := (2.6)
0 otherwise,
is weighted,
iff ∃ i, j ∈ N, (i, j) ∈ E : wi j , 1,
(
1
λw ∈ B : λw := (2.7)
0 otherwise,
or is self-looped
Nn ∈ N : Nn := |N| (2.9)
Another random document with
no related content on Scribd:
*** END OF THE PROJECT GUTENBERG EBOOK THE MINISTER
HAD TO WAIT ***
Updated editions will replace the previous one—the old editions will
be renamed.
1.D. The copyright laws of the place where you are located also
govern what you can do with this work. Copyright laws in most
countries are in a constant state of change. If you are outside the
United States, check the laws of your country in addition to the terms
of this agreement before downloading, copying, displaying,
performing, distributing or creating derivative works based on this
work or any other Project Gutenberg™ work. The Foundation makes
no representations concerning the copyright status of any work in
any country other than the United States.
• You pay a royalty fee of 20% of the gross profits you derive from
the use of Project Gutenberg™ works calculated using the
method you already use to calculate your applicable taxes. The
fee is owed to the owner of the Project Gutenberg™ trademark,
but he has agreed to donate royalties under this paragraph to
the Project Gutenberg Literary Archive Foundation. Royalty
payments must be paid within 60 days following each date on
which you prepare (or are legally required to prepare) your
periodic tax returns. Royalty payments should be clearly marked
as such and sent to the Project Gutenberg Literary Archive
Foundation at the address specified in Section 4, “Information
about donations to the Project Gutenberg Literary Archive
Foundation.”
• You comply with all other terms of this agreement for free
distribution of Project Gutenberg™ works.
1.F.
1.F.4. Except for the limited right of replacement or refund set forth in
paragraph 1.F.3, this work is provided to you ‘AS-IS’, WITH NO
OTHER WARRANTIES OF ANY KIND, EXPRESS OR IMPLIED,
INCLUDING BUT NOT LIMITED TO WARRANTIES OF
MERCHANTABILITY OR FITNESS FOR ANY PURPOSE.
Please check the Project Gutenberg web pages for current donation
methods and addresses. Donations are accepted in a number of
other ways including checks, online payments and credit card
donations. To donate, please visit: www.gutenberg.org/donate.
Most people start at our website which has the main PG search
facility: www.gutenberg.org.