Random Walks and Heat Kernels on

Graphs 1st Edition Martin T. Barlow

Visit to download the full and correct content document:
https://textbookfull.com/product/random-walks-and-heat-kernels-on-graphs-1st-edition
-martin-t-barlow/
More products digital (pdf, epub, mobi) instant
download maybe you interests ...

Probability on Graphs Random Processes on Graphs and

Lattices Geoffrey Grimmett

https://textbookfull.com/product/probability-on-graphs-random-
processes-on-graphs-and-lattices-geoffrey-grimmett/

Probability on graphs random processes on graphs and

lattices Second Edition Grimmett

https://textbookfull.com/product/probability-on-graphs-random-
processes-on-graphs-and-lattices-second-edition-grimmett/

Generating Random Networks and Graphs 1st Edition

Annibale

https://textbookfull.com/product/generating-random-networks-and-
graphs-1st-edition-annibale/

Random Graphs Geometry and Asymptotic Structure 1st

Edition Michael Krivelevich

https://textbookfull.com/product/random-graphs-geometry-and-
asymptotic-structure-1st-edition-michael-krivelevich/
From Random Walks to Random Matrices - Selected Topics
in Modern Theoretical Physics 1st Edition Jean Zinn-
Justin

https://textbookfull.com/product/from-random-walks-to-random-
matrices-selected-topics-in-modern-theoretical-physics-1st-
edition-jean-zinn-justin/

Limit Theorems for Some Long Range Random Walks on

Torsion Free Nilpotent Groups (SpringerBriefs in
Mathematics) Chen

https://textbookfull.com/product/limit-theorems-for-some-long-
range-random-walks-on-torsion-free-nilpotent-groups-
springerbriefs-in-mathematics-chen/

Heat Exchangers First Edition Holger Martin

https://textbookfull.com/product/heat-exchangers-first-edition-
holger-martin/

Random Walks in the Quarter Plane Algebraic Methods

Boundary Value Problems Applications to Queueing
Systems and Analytic Combinatorics 2nd Edition Guy
Fayolle
https://textbookfull.com/product/random-walks-in-the-quarter-
plane-algebraic-methods-boundary-value-problems-applications-to-
queueing-systems-and-analytic-combinatorics-2nd-edition-guy-
fayolle/

ASL Grammar: The Workbook 1st Edition Rochelle Barlow

https://textbookfull.com/product/asl-grammar-the-workbook-1st-
edition-rochelle-barlow/
LONDON MATHEMATICAL SOCIETY LECTURE NOTE SERIES

Managing Editor: Professor M. Reid, Mathematics Institute,

University of Warwick, Coventry CV4 7AL, United Kingdom

The titles below are available from booksellers, or from Cambridge University Press at
http://www.cambridge.org/mathematics

315 Structured ring spectra, A. BAKER & B. RICHTER (eds)

316 Linear logic in computer science, T. EHRHARD, P. RUET, J.-Y. GIRARD & P. SCOTT (eds)
317 Advances in elliptic curve cryptography, I.F. BLAKE, G. SEROUSSI & N.P. SMART (eds)
318 Perturbation of the boundary in boundary-value problems of partial differential equations, D. HENRY
319 Double affine Hecke algebras, I. CHEREDNIK
320 L-functions and Galois representations, D. BURNS, K. BUZZARD & J. NEKOVÁŘ (eds)
321 Surveys in modern mathematics, V. PRASOLOV & Y. ILYASHENKO (eds)
322 Recent perspectives in random matrix theory and number theory, F. MEZZADRI & N.C. SNAITH (eds)
323 Poisson geometry, deformation quantisation and group representations, S. GUTT et al (eds)
324 Singularities and computer algebra, C. LOSSEN & G. PFISTER (eds)
325 Lectures on the Ricci flow, P. TOPPING
326 Modular representations of finite groups of Lie type, J.E. HUMPHREYS
327 Surveys in combinatorics 2005, B.S. WEBB (ed)
328 Fundamentals of hyperbolic manifolds, R. CANARY, D. EPSTEIN & A. MARDEN (eds)
329 Spaces of Kleinian groups, Y. MINSKY, M. SAKUMA & C. SERIES (eds)
330 Noncommutative localization in algebra and topology, A. RANICKI (ed)
331 Foundations of computational mathematics, Santander 2005, L.M PARDO, A. PINKUS, E. SÜLI & M.J. TODD (eds)
332 Handbook of tilting theory, L. ANGELERI HÜGEL, D. HAPPEL & H. KRAUSE (eds)
333 Synthetic differential geometry (2nd Edition), A. KOCK
334 The Navier–Stokes equations, N. RILEY & P. DRAZIN
335 Lectures on the combinatorics of free probability, A. NICA & R. SPEICHER
336 Integral closure of ideals, rings, and modules, I. SWANSON & C. HUNEKE
337 Methods in Banach space theory, J.M.F. CASTILLO & W.B. JOHNSON (eds)
338 Surveys in geometry and number theory, N. YOUNG (ed)
339 Groups St Andrews 2005 I, C.M. CAMPBELL, M.R. QUICK, E.F. ROBERTSON & G.C. SMITH (eds)
340 Groups St Andrews 2005 II, C.M. CAMPBELL, M.R. QUICK, E.F. ROBERTSON & G.C. SMITH (eds)
341 Ranks of elliptic curves and random matrix theory, J.B. CONREY, D.W. FARMER, F. MEZZADRI & N.C. SNAITH (eds)
342 Elliptic cohomology, H.R. MILLER & D.C. RAVENEL (eds)
343 Algebraic cycles and motives I, J. NAGEL & C. PETERS (eds)
344 Algebraic cycles and motives II, J. NAGEL & C. PETERS (eds)
345 Algebraic and analytic geometry, A. NEEMAN
346 Surveys in combinatorics 2007, A. HILTON & J. TALBOT (eds)
347 Surveys in contemporary mathematics, N. YOUNG & Y. CHOI (eds)
348 Transcendental dynamics and complex analysis, P.J. RIPPON & G.M. STALLARD (eds)
349 Model theory with applications to algebra and analysis I, Z. CHATZIDAKIS, D. MACPHERSON, A. PILLAY & A.
WILKIE (eds)
350 Model theory with applications to algebra and analysis II, Z. CHATZIDAKIS, D. MACPHERSON, A. PILLAY & A.
WILKIE (eds)
351 Finite von Neumann algebras and masas, A.M. SINCLAIR & R.R. SMITH
352 Number theory and polynomials, J. MCKEE & C. SMYTH (eds)
353 Trends in stochastic analysis, J. BLATH, P. MÖRTERS & M. SCHEUTZOW (eds)
354 Groups and analysis, K. TENT (ed)
355 Non-equilibrium statistical mechanics and turbulence, J. CARDY, G. FALKOVICH & K. GAWEDZKI
356 Elliptic curves and big Galois representations, D. DELBOURGO
357 Algebraic theory of differential equations, M.A.H. MACCALLUM & A.V. MIKHAILOV (eds)
358 Geometric and cohomological methods in group theory, M.R. BRIDSON, P.H. KROPHOLLER & I.J. LEARY (eds)
359 Moduli spaces and vector bundles, L. BRAMBILA-PAZ, S.B. BRADLOW, O. GARCÍA-PRADA & S. RAMANAN (eds)
360 Zariski geometries, B. ZILBER
361 Words: Notes on verbal width in groups, D. SEGAL
362 Differential tensor algebras and their module categories, R. BAUTISTA, L. SALMERÓN & R. ZUAZUA
363 Foundations of computational mathematics, Hong Kong 2008, F. CUCKER, A. PINKUS & M.J. TODD (eds)
364 Partial differential equations and fluid mechanics, J.C. ROBINSON & J.L. RODRIGO (eds)
365 Surveys in combinatorics 2009, S. HUCZYNSKA, J.D. MITCHELL & C.M. RONEY-DOUGAL (eds)
366 Highly oscillatory problems, B. ENGQUIST, A. FOKAS, E. HAIRER & A. ISERLES (eds)
367 Random matrices: High dimensional phenomena, G. BLOWER
368 Geometry of Riemann surfaces, F.P. GARDINER, G. GONZÁLEZ-DIEZ & C. KOUROUNIOTIS (eds)
369 Epidemics and rumours in complex networks, M. DRAIEF & L. MASSOULIÉ
370 Theory of p-adic distributions, S. ALBEVERIO, A.YU. KHRENNIKOV & V.M. SHELKOVICH
371 Conformal fractals, F. PRZYTYCKI & M. URBAŃSKI
372 Moonshine: The first quarter century and beyond, J. LEPOWSKY, J. MCKAY & M.P. TUITE (eds)
373 Smoothness, regularity and complete intersection, J. MAJADAS & A. G. RODICIO
374 Geometric analysis of hyperbolic differential equations: An introduction, S. ALINHAC
375 Triangulated categories, T. HOLM, P. JØRGENSEN & R. ROUQUIER (eds)
376 Permutation patterns, S. LINTON, N. RUŠKUC & V. VATTER (eds)
377 An introduction to Galois cohomology and its applications, G. BERHUY
378 Probability and mathematical genetics, N. H. BINGHAM & C. M. GOLDIE (eds)
379 Finite and algorithmic model theory, J. ESPARZA, C. MICHAUX & C. STEINHORN (eds)
380 Real and complex singularities, M. MANOEL, M.C. ROMERO FUSTER & C.T.C WALL (eds)
381 Symmetries and integrability of difference equations, D. LEVI, P. OLVER, Z. THOMOVA & P. WINTERNITZ (eds)
382 Forcing with random variables and proof complexity, J. KRAJÍČEK
383 Motivic integration and its interactions with model theory and non-Archimedean geometry I, R. CLUCKERS, J. NICAISE
& J. SEBAG (eds)
384 Motivic integration and its interactions with model theory and non-Archimedean geometry II, R. CLUCKERS, J. NICAISE
& J. SEBAG (eds)
385 Entropy of hidden Markov processes and connections to dynamical systems, B. MARCUS, K. PETERSEN &
T. WEISSMAN (eds)
386 Independence-friendly logic, A.L. MANN, G. SANDU & M. SEVENSTER
387 Groups St Andrews 2009 in Bath I, C.M. CAMPBELL et al (eds)
388 Groups St Andrews 2009 in Bath II, C.M. CAMPBELL et al (eds)
389 Random fields on the sphere, D. MARINUCCI & G. PECCATI
390 Localization in periodic potentials, D.E. PELINOVSKY
391 Fusion systems in algebra and topology, M. ASCHBACHER, R. KESSAR & B. OLIVER
392 Surveys in combinatorics 2011, R. CHAPMAN (ed)
393 Non-abelian fundamental groups and Iwasawa theory, J. COATES et al (eds)
394 Variational problems in differential geometry, R. BIELAWSKI, K. HOUSTON & M. SPEIGHT (eds)
395 How groups grow, A. MANN
396 Arithmetic differential operators over the p-adic integers, C.C. RALPH & S.R. SIMANCA
397 Hyperbolic geometry and applications in quantum chaos and cosmology, J. BOLTE & F. STEINER (eds)
398 Mathematical models in contact mechanics, M. SOFONEA & A. MATEI
399 Circuit double cover of graphs, C.-Q. ZHANG
400 Dense sphere packings: a blueprint for formal proofs, T. HALES
401 A double Hall algebra approach to affine quantum Schur–Weyl theory, B. DENG, J. DU & Q. FU
402 Mathematical aspects of fluid mechanics, J.C. ROBINSON, J.L. RODRIGO & W. SADOWSKI (eds)
403 Foundations of computational mathematics, Budapest 2011, F. CUCKER, T. KRICK, A. PINKUS & A. SZANTO (eds)
404 Operator methods for boundary value problems, S. HASSI, H.S.V. DE SNOO & F.H. SZAFRANIEC (eds)
405 Torsors, étale homotopy and applications to rational points, A.N. SKOROBOGATOV (ed)
406 Appalachian set theory, J. CUMMINGS & E. SCHIMMERLING (eds)
407 The maximal subgroups of the low-dimensional finite classical groups, J.N. BRAY, D.F. HOLT & C.M. RONEY-DOUGAL
408 Complexity science: the Warwick master’s course, R. BALL, V. KOLOKOLTSOV & R.S. MACKAY (eds)
409 Surveys in combinatorics 2013, S.R. BLACKBURN, S. GERKE & M. WILDON (eds)
410 Representation theory and harmonic analysis of wreath products of finite groups, T. CECCHERINI-SILBERSTEIN,
F. SCARABOTTI & F. TOLLI
411 Moduli spaces, L. BRAMBILA-PAZ, O. GARCÍA-PRADA, P. NEWSTEAD & R.P. THOMAS (eds)
412 Automorphisms and equivalence relations in topological dynamics, D.B. ELLIS & R. ELLIS
413 Optimal transportation, Y. OLLIVIER, H. PAJOT & C. VILLANI (eds)
414 Automorphic forms and Galois representations I, F. DIAMOND, P.L. KASSAEI & M. KIM (eds)
415 Automorphic forms and Galois representations II, F. DIAMOND, P.L. KASSAEI & M. KIM (eds)
416 Reversibility in dynamics and group theory, A.G. O’FARRELL & I. SHORT
417 Recent advances in algebraic geometry, C.D. HACON, M. MUSTAŢĂ & M. POPA (eds)
418 The Bloch–Kato conjecture for the Riemann zeta function, J. COATES, A. RAGHURAM, A. SAIKIA & R. SUJATHA
(eds)
419 The Cauchy problem for non-Lipschitz semi-linear parabolic partial differential equations, J.C. MEYER & D.J. NEEDHAM
420 Arithmetic and geometry, L. DIEULEFAIT et al (eds)
421 O-minimality and Diophantine geometry, G.O. JONES & A.J. WILKIE (eds)
422 Groups St Andrews 2013, C.M. CAMPBELL et al (eds)
423 Inequalities for graph eigenvalues, Z. STANIĆ
424 Surveys in combinatorics 2015, A. CZUMAJ et al (eds)
425 Geometry, topology and dynamics in negative curvature, C.S. ARAVINDA, F.T. FARRELL & J.-F. LAFONT (eds)
426 Lectures on the theory of water waves, T. BRIDGES, M. GROVES & D. NICHOLLS (eds)
427 Recent advances in Hodge theory, M. KERR & G. PEARLSTEIN (eds)
428 Geometry in a Fréchet context, C. T. J. DODSON, G. GALANIS & E. VASSILIOU
429 Sheaves and functions modulo p, L. TAELMAN
430 Recent progress in the theory of the Euler and Navier-Stokes equations, J.C. ROBINSON, J.L. RODRIGO, W. SADOWSKI
& A. VIDAL-LÓPEZ (eds)
431 Harmonic and subharmonic function theory on the real hyperbolic ball, M. STOLL
432 Topics in graph automorphisms and reconstruction (2nd Edition), J. LAURI & R. SCAPELLATO
433 Regular and irregular holonomic D-modules, M. KASHIWARA & P. SCHAPIRA
434 Analytic semigroups and semilinear initial boundary value problems (2nd Edition), K. TAIRA
435 Graded rings and graded Grothendieck groups, R. HAZRAT
436 Groups, graphs and random walks, T. CECCHERINI-SILBERSTEIN, M. SALVATORI & E. SAVA-HUSS (eds)
437 Dynamics and analytic number theory, D. BADZIAHIN, A. GORODNIK & N. PEYERIMHOFF (eds)
438 Random Walks and Heat Kernels on Graphs, MARTIN T. BARLOW
London Mathematical Society Lecture Note Series: 438

Random Walks and Heat

Kernels on Graphs

M A RT I N T. BA R L OW
University of British Columbia, Canada
 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
4843/24, 2nd Floor, Ansari Road, Daryaganj, Delhi – 110002, India
79 Anson Road, #06–04/06, Singapore 079906

Cambridge University Press is part of the University of Cambridge.

It furthers the University’s mission by disseminating knowledge in the pursuit of
education, learning, and research at the highest international levels of excellence.

www.cambridge.org
Information on this title: www.cambridge.org/9781107674424
DOI: 10.1017/9781107415690

c Martin T. Barlow 2017
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 2017
Printed in the United Kingdom by Clays, St Ives plc
A catalogue record for this publication is available from the British Library.
Library of Congress Cataloging-in-Publication Data
Names: Barlow, M. T.
Title: Random walks and heat kernels on graphs / Martin T. Barlow,
University of British Columbia, Canada.
Description: Cambridge : Cambridge University Press, [2017] |
Series: London Mathematical Society lecture note series ; 438 |
Includes bibliographical references and index.
Identifiers: LCCN 2016051295 | ISBN 9781107674424
Subjects: LCSH: Random walks (Mathematics) | Graph theory. |
Markov processes. | Heat equation.
Classification: LCC QA274.73 .B3735 2017 | DDC 511/.5–dc23
LC record available at https://lccn.loc.gov/2016051295
ISBN 978-1-107-67442-4 Paperback
Cambridge University Press has no responsibility for the persistence or accuracy of
URLs for external or third-party Internet Web sites referred to in this publication
and does not guarantee that any content on such Web sites is, or will remain,
accurate or appropriate.
To my mother, Yvonne Barlow, and in memory of my father, Andrew Barlow.
 Contents

Preface page ix

1 Introduction 1
1.1 Graphs and Weighted Graphs 1
1.2 Random Walks on a Weighted Graph 6
1.3 Transition Densities and the Laplacian 11
1.4 Dirichlet or Energy Form 15
1.5 Killed Process 21
1.6 Green’s Functions 22
1.7 Harmonic Functions, Harnack Inequalities, and the
Liouville Property 26
1.8 Strong Liouville Property for Rd 31
1.9 Interpretation of the Liouville Property 32

2 Random Walks and Electrical Resistance 38

2.1 Basic Concepts 38
2.2 Transience and Recurrence 42
2.3 Energy and Variational Methods 44
2.4 Resistance to Infinity 55
2.5 Traces and Electrical Equivalence 61
2.6 Stability under Rough Isometries 67
2.7 Hitting Times and Resistance 70
2.8 Examples 73
2.9 The Sierpinski Gasket Graph 75

3 Isoperimetric Inequalities and Applications 80

3.1 Isoperimetric Inequalities 80
3.2 Nash Inequality 85
3.3 Poincaré Inequality 91

vii
viii Contents

3.4 Spectral Decomposition for a Finite Graph 97

3.5 Strong Isoperimetric Inequality and Spectral Radius 101
4 Discrete Time Heat Kernel 106
4.1 Basic Properties and Bounds on the Diagonal 106
4.2 Carne–Varopoulos Bound 111
4.3 Gaussian and Sub-Gaussian Heat Kernel Bounds 116
4.4 Off-diagonal Upper Bounds 124
4.5 Lower Bounds 128
5 Continuous Time Random Walks 132
5.1 Introduction to Continuous Time 132
5.2 Heat Kernel Bounds 140
6 Heat Kernel Bounds 149
6.1 Strongly Recurrent Graphs 149
6.2 Gaussian Upper Bounds 155
6.3 Poincaré Inequality and Gaussian Lower Bounds 160
6.4 Remarks on Gaussian Bounds 168
7 Potential Theory and Harnack Inequalities 172
7.1 Introduction to Potential Theory 172
7.2 Applications 179
Appendix 183
A.1 Martingales and Tail Estimates 183
A.2 Discrete Time Markov Chains and the Strong Markov
Property 186
A.3 Continuous Time Random Walk 190
A.4 Invariant and Tail σ -fields 197
A.5 Hilbert Space Results 202
A.6 Miscellaneous Estimates 205
A.7 Whitney Type Coverings of a Ball 206
A.8 A Maximal Inequality 211
A.9 Poincaré Inequalities 213
References 219
Index 224
 Preface

The topic of random walks on graphs is a vast one, and has close connections
with many other areas of probability, as well as analysis, geometry, and alge-
bra. In the probabilistic direction, a random walk on a graph is just a reversible
or symmetric Markov chain, and many results on random walks on graphs also
hold for more general Markov chains. However, pursuing this generalisation
too far leads to a loss of concrete interest, and in this text the context will be
restricted to random walks on graphs where each vertex has a finite number of
neighbours.
Even with these restrictions, there are many topics which this book does not
cover – in particular the very active field of relaxation times for finite graphs.
This book is mainly concerned with infinite graphs, and in particular those
which have polynomial volume growth. The main topic is the relation between
geometric properties of the graph and asymptotic properties of the random
walk. A particular emphasis is on properties which are stable under minor per-
turbations of the graph – for example the addition of a number of diagonal
edges to the Euclidean lattice Z2 . The precise definition of ‘minor perturbation’
is given by the concept of a rough isometry, or quasi-isometry. One example
of a property which is stable under these perturbations is transience; this sta-
bility is proved in Chapter 2 using electrical networks. A considerably harder
theorem, which is one of the main results of this book, is that the property of
satisfying Gaussian heat kernel bounds is also stable under rough isometries.
The second main theme of this book is deriving bounds on the transition
density of the random walk, or the heat kernel, from geometric information on
the graph. Once one has these bounds, many properties of the random walk
can then be obtained in a straightforward fashion.
Chapter 1 gives the basic definition of graphs and random walks, as well
as that of rough isometry. Chapter 2 explores the close connections between

ix
x Preface

random walks and electrical networks. The key contribution of network the-
ory is to connect the hitting properties of the random walk with the effective
resistance between sets. The effective resistance can be bounded using varia-
tional principles, and this introduces tools which have no purely probabilistic
counterpart.
Chapter 3 introduces some geometric and analytic inequalities which will
play a key role in the remainder of the book – two kinds of isoperimetric
inequality, Nash inequalities, and Poincaré inequalities. It is shown how the
two analytic inequalities (i.e. Nash and Poincaré) can be derived from the
geometric information given by the isoperimetric inequalities.
Chapter 4 studies the transition density of the discrete time random walk,
or discrete time heat kernel. Two initial upper bounds on this are proved – an
‘on-diagonal’ upper bound from a Nash inequality, and a long range bound,
the Carne–Varopoulos bound, which holds in very great generality. Both these
proofs are quite short. To obtain full Gaussian or sub-Gaussian bounds much
more work is needed. The second half of Chapter 4 makes the initial steps, and
introduces conditions (bounds on exit times from balls, and a ‘near-diagonal
lower bound’) which if satisfied will lead to full Gaussian bounds.
Chapter 5 introduces the continuous time random walk, which is in many
respects an easier and more regular object to study than the discrete time ran-
dom walk. In this chapter it is defined from the discrete time walk and an
independent Poisson process – another construction, from the transition densi-
ties, is given in the Appendix. It is proved that Gaussian or sub-Gaussian heat
kernel bounds hold for the discrete time random walk if and only if they hold
for the continuous time walk.
Chapter 6 proves sub-Gaussian or Gaussian bounds for two classes of
graphs. The first are ‘strongly recurrent’ graphs; given the work done in
Chapter 4 the results for these come quite easily from volume and resistance
estimates. The second class are graphs which satisfy a Poincaré inequality –
these include Zd and graphs which are roughly isometric to Zd . The proof of
Gaussian bounds for these graphs is based on the tools developed by Nash to
study divergence form PDE; these methods also work well in the graph context.
Chapter 7 gives a brief introduction to potential theory. Using the method
of balayage, the heat kernel bounds of Chapter 6 lead to a Harnack inequality,
and this in turn is used to prove the strong Liouville property (i.e. all positive
harmonic functions are constant) for graphs which are roughly isometric to Zd .
This book is based on a course given at the first Pacific Institute for the
Mathematical Sciences (PIMS) summer school in Probability, which was held
at UBC in 2004. It is written for first year graduate students, though it is prob-
ably too long for a one semester course at most institutions. A much shorter
 Preface xi

version of this course was also given in Kyoto in 2005, and I wish to thank
both PIMS and the Research Institute for the Mathematical Sciences (RIMS)
for their invitations to give these courses.
I circulated a version of these notes several years ago, and I would like to
thank those who have read them and given feedback, and in particular, Alain
Sznitman and Gerard Ben-Arous, for encouraging me to complete them. I also
wish to thank Alain for making available to me some lecture notes of his own
on this material.
I have learnt much from a number of previous works in this area – the
classical text of Spitzer [Spi], as well as the more recent works [AF, LL, LP,
Wo].
I am grateful to those who have sent me corrections, and in particular to
Yoshihiro Abe, Alma Sarai Hernandez Torres, Guillermo Martinez Dibene,
Jun Misumi, and Zichun Ye.
 1

Introduction

1.1 Graphs and Weighted Graphs

We start with some basic definitions. To be quite formal initially, a graph is a
pair  = (V, E). Here V is a finite or countably infinite set, and E is a subset
of P2 (V) = {subsets A of V with two elements}. Given a set A we write | A|
for the number of elements of A. The elements of V are called vertices, and
the elements of E edges. In practice, we think of the elements of V as points
(often embedded in some space such as Rd ), and the edges as lines between
the vertices.
Now for some general definitions.

(1) We write x ∼ y to mean {x, y} ∈ E, and say that y is a neighbour

of x. Since all edges have two distinct elements, we have not allowed
edges between a point x and itself. Also, we do not allow multiple edges
between points, though we could; anything which can be done with mul-
tiple edges can be done better with weights, which will be introduced
shortly.
(2) A path γ in  is a sequence x0 , x1 , . . . , x n with xi−1 ∼ xi for 1 ≤
i ≤ n. The length of a path γ is the number of edges in γ . A path
γ = (x 0 , . . . , xn ) is a cycle or loop if x0 = xn . A path is self-avoiding
if the vertices in the path are distinct.
(3) We define d(x, y) to be the length n of the shortest path x =
x0 , x1 , . . . , x n = y. If there is no such path then we set d(x, y) = ∞.
(This is the graph or chemical metric on .) We also write for x ∈ V,
A ⊂ V,

d(x, A) = min{d(x, y) : y ∈ A}.

(4)  is connected if d(x, y) < ∞ for all x, y.

1
2 Introduction

(5)  is locally finite if N (x) = {y : y ∼ x} is finite for each x ∈ V, that is,

every vertex has a finite number of neighbours.  has bounded geometry
if supx |N (x)| < ∞.
(6) Define balls in  by
B(x, r ) = {y : d(x, y) ≤ r }, x ∈ V, r ∈ [0, ∞).

Note that while the metric d(x, ·) is integer valued, we allow r here to be
real. Of course B(x, r ) = B(x, r ).
(7) We define the exterior boundary of A by
∂ A = {y ∈ Ac : there exists x ∈ A with x ∼ y}.
Set also

∂i A = ∂(Ac ) = {y ∈ A : there exists x ∈ Ac with x ∼ y },

A = A ∪ ∂ A,
Ao = A − ∂i A.

We use the notation An ↑↑ V to mean that An is an increasing sequence of

finite sets with ∪n An = V.

Notation We write ci , Ci for (strictly) positive constants, which will remain

fixed in each argument. In some proofs we will write c, c etc. for positive
constants, which may change from line to line. Given functions f, g : A → R
we write f  g to mean that there exists c ≥ 1 such that

c−1 f (x) ≤ g(x) ≤ c f (x) for all x ∈ A.

We write 1 for the function f which is identically 1, and 1 A for the indicator
function (or characteristic function) of the set A. We write 1x for 1{x} .

From now on we will always assume:

 is locally finite, (H1)

 is connected; (H2)

and in addition from time to time we will assume:

 has bounded geometry. (H3)

We will treat weighted graphs.

Definition 1.1 We assume there exist weights (also called conductances)

μx y , x, y ∈ V satisfying:
 1.1 Graphs and Weighted Graphs 3

(i) μx y = μ yx ,
(ii) μx y ≥ 0 for all x, y ∈ V,
(iii) if x  = y then μx y > 0 if and only if x ∼ y.

We call (, μ) a weighted

 graph. Note that condition
 (iii) allows us to have
μx x > 0. Since E = {x, y} : x  = y, μx y > 0 , we can recover the set of
edges from μ.

The natural weights on  are given by


1 if x ∼ y,
μx y =
0 otherwise.
Whenever we discuss a graph without explicit mention of weights, we will
assume we are using the natural weights.

Let μx = μ(x) = y μx y , and extend μ to a measure on V by setting

μ(A) = μx . (1.1)
x∈A

Since  is locally finite we have:

(i) |B(x, r )| < ∞ for any x and r ,

(ii) μ(A) < ∞ for any finite set A.
We will sometimes need additional conditions on the weights.

Definition 1.2 We say (, μ) has bounded weights if there exists C1 < ∞
such that C 1−1 ≤ μe ≤ C 1 for all e ∈ E, and μx x ≤ C1 for all x ∈ V. We say
(, μ) has controlled weights if there exists C 2 < ∞ such that
μx y 1
≥ whenever x ∼ y.
μx C2

We introduce the conditions:

(, μ) has bounded weights, (H4)

(, μ) has controlled weights. (H5)

Controlled weights is called ‘the p0 condition’ in [GT1, GT2]; note that it

holds if (H3) and (H4) hold. While a little less obvious than (H4), it turns out
that controlled weights is the ‘right’ condition in many circumstances. As an
example of its use, we show how it gives bounds on both the cardinality and
the measure of a ball.
4 Introduction

Lemma 1.3 Suppose (, μ) satisfies (H5) with constant C2 . Then (H3) holds
and, for n ≥ 0, |B(x, n)| ≤ 2C2n , μ(B(x, n)) ≤ 2C2n μx .

Proof Since μx y ≥ μx /C2 if x ∼ y, we have |N (x)| ≤ C2 for all x and thus

(H3) holds. Further, we have C 2 ≥ 2 unless |V| ≤ 2. Then writing S(x, n)
= {y : d(x, y) = n}, we have |S(x, n)| ≤ C2 |S(x, n − 1)|, and so by induction
|B(x, n)| ≤ (C 2n+1 − 1)/(C 2 − 1) ≤ 2C2n . Also
 
μ(N (x)) = μ y ≤ C2 μx y = C 2 μ x .
y∼x y∼x

Hence again by induction μ(B(x, n)) ≤ 2C2n μx .

Examples 1.4 (1) The Euclidean lattice Zd . Here V = Zd , and x ∼ y if

and only if |x − y| = 1. Strictly speaking we should denote this graph by
(Zd , E d ), where E d is the set of edges defined above.
(2) The d-ary tree. (‘Binary’ if d = 2.) This is the unique infinite graph with
|N (x)| ≡ d + 1, and with no cycles.
(3) We will also consider the ‘rooted binary tree’ B. Let B0 = {o}, and for
n ≥ 1 let Bn = {0, 1}n . Then the vertex set is given by B = ∪∞ n=0 Bn . For
x = (x 1 , . . . , xn ) ∈ Bn with n ≥ 2 set a(x) = (x 1 , . . . , xn−1 ) – we call
a(x) the ancestor of x. (We set a(z) = o for z ∈ B1 .) Then the edge set of
the rooted binary tree is given by
 
E(B) = {x, a(x)} : x ∈ B − B0 .

(4) The canopy tree is a subgraph of B defined as follows. Let

U = {o} ∪ {x = (x1 , . . . , x n ) ∈ B : x1 = · · · = xn = 0}.

Define f : B → B by taking f (x) to be the point in U closest to x, and set

BCan = {x ∈ B : d(x, f (x)) ≤ d(o, f (x))}. Then the canopy tree is the
subgraph of (B, E(B)) generated by BCan . The canopy tree has exponential
volume growth, but only one self-avoiding path from o to infinity.
(5) Let G be a finitely generated group, and let  = {g1 , . . . , gn }
be a set of generators, not necessarily minimal. Write ∗ =
{g1 , . . . , gn , g1−1 , . . . , gn−1 }. Let V = G and let {g, h} ∈ E if and only
if g −1 h ∈ ∗ . Then  = (V, E) is the Cayley graph of the group G with
generators .
Zd is the Cayley graph of the group Z ⊕ · · · ⊕ Z, with generators gk ,
1 ≤ k ≤ d; here gk has 1 in the kth place and is zero elsewhere. Since
different sets of generators will give rise to different Cayley graphs, in
general a group G will have many different Cayley graphs. The d-ary tree
 1.1 Graphs and Weighted Graphs 5

is the Cayley graph of the group generated by  = {x1 , x2 , . . . , xd+1 } with

the relations x 12 = · · · = x d+1
2 = 1. Since the ternary tree is also the Cayley
graph of the free group with two generators, two distinct (non-isomorphic)
groups can have the same Cayley graph.
(6) Let α > 0. Consider the graph Z+ = {0, 1, . . . }, but with weights
(α)
μn,n+1 = α n . This satisfies (H1)–(H3) and (H5), but (H4) fails unless
α = 1. If α < 1 then (Z+ , μ(α) ) is an infinite graph, but μ(α) (Z+ ) < ∞.
(7) Given a weighted graph (V, E, μ), the lazy weighted graph (V, E, μ(L) )
(L)
is defined by setting μx y = μx y if y  = x, and

μ(L)
xx = μx y .
y=x

Definition 1.5 We introduce the following operations on graphs or weighted

graphs. Let (, μ) be a weighted graph.

(1) Subgraphs. Let H ⊂ V. Then the subgraph induced by H is the (weighted)

graph with vertex set H , edge set
 
E H = {x, y} ∈ E : x, y ∈ H ,

and weights (if appropriate) given by

μxHy = μx y , x, y ∈ H.

We define μxH = y∈H μxHy ; note that μxH ≤ μx . Clearly (H, E H ) need
not be connected, even if  is.
(2) Collapse of a subset to a point. Let A ⊂ V. The graph obtained by col-
lapsing A to a point a (where a  ∈ V) is the graph  = (V , E ) given
by

V = (V − A) ∪ {a},
   
E = {x, y} ∈ E : x, y ∈ V − A ∪ {x, a} : x ∈ ∂ A .

We define weights on (V , E ) by setting μx y = μx y if x, y ∈ V − A, and


μxa = μxb , x ∈ V − A.
b∈A

Note that μxa ≤ μx < ∞. In general this graph need not be locally finite,
but will be if |∂ A| < ∞. It is easy to check that if  is connected then so
is  .
6 Introduction

(3) Join of two graphs. Let i = (Vi , E i ), i = 1, 2 be two graphs. (We regard
V1 , V2 as being distinct sets.) If xi ∈ Vi , the join of 1 and 2 at x1 , x2 is
the graph (V, E) given by
 
V = V1 ∪ V2 , E = E 1 ∪ E 2 ∪ {x 1 , x2 } .

We define weights on (V, E) by keeping the weights on 1 and 2 , and

giving the new edge {x1 , x2 } weight 1.
(4) Graph products. Let i be graphs. There are two natural products, called
the Cartesian product (denoted 1  2 ) and the direct or tensor product
1 ×2 . In both cases we take the vertex set of the product to be V1 ×V2 =
{(x1 , x 2 ) : x 1 ∈ V1 , x2 ∈ V2 }. For the Cartesian product we define
 
E  = {(x, y1 ), (x, y2 )} : x ∈ V1 , {y1 , y2 } ∈ E 2
 
∪ {(x1 , y), (x 2 , y)} : {x1 , x2 } ∈ E 1 , y ∈ V2 ,

and for the tensor product

 
E × = {(x1 , y1 ), (x 2 , y2 )} : {x 1 , x 2 } ∈ E 1 , {y1 , y2 } ∈ E 2 .

Note that Z2 = Z  Z. If μ(i) are weights on i then we define weights

on the product by

μ (2)
(x,y1 ),(x,y2 ) = μ y1 ,y2 , μ (1)
(x 1 ,y),(x 2 ,y) = μx 1 ,x 2 ,
μ× (1) (2)
(x1 ,y1 ),(x2 ,y2 ) = μx 1 ,x 2 μ y1 ,y2 .

1.2 Random Walks on a Weighted Graph

Set
μx y
P(x, y) = . (1.2)
μx
Let X be the discrete time Markov chain X = (X n , n ≥ 0, Px , x ∈ V) with
transition matrix (P(x, y)), defined on a space (, F ) – see Appendix A.2.
Here Px is the law of the chain with X 0 = x, and the transition probabilities
are given by
Pz (X n+1 = y|X n = x) = P(x, y).

We call X the simple random walk (SRW) on (, μ) and (P(x, y)) the tran-
sition matrix of X . At each time step X moves from a vertex x to a neighbour
y with probability proportional to μx y . Since we can have μx x > 0 we do
in general allow X to remain at a vertex x with positive probability. If ν
 1.2 Random Walks on a Weighted Graph 7

is a probability measure on V we write Pν for the law of X started with

distribution ν.
Note that (P(x, y)) is μ-symmetric, that is:

μx P(x, y) = μ y P(y, x) = μx y . (1.3)

Set
Pn (x, y) = Px (X n = y).

The condition (1.3) implies that X is reversible.

Lemma 1.6 Let x0 , x1 , . . . , x n ∈ V. Then

μx0 Px0 (X 0 = x0 , X 1 = x1 , . . . , X n = xn )
= μxn Pxn (X 0 = xn , X 1 = x n−1 , . . . , X n = x0 ).

Proof Using the μ-symmetry of P it is easy to check that both sides equal
n
i=1 μxi−1 ,xi .

Definition 1.7 Let A ⊂ V. Define the hitting times

T A = min{n ≥ 0 : X n ∈ A},
T A+ = min{n ≥ 1 : X n ∈ A}.

Here we adopt the convention that min ∅ = +∞, so that T A = ∞ if and only
if X never hits A. Note also that if X 0  ∈ A then T A = T A+ . Write Tx = T{x}
and
τ A = T Ac = min{n ≥ 0 : X n ∈ A} (1.4)

for the time of the first exit from A. If A, B ⊂ V we will write {T A < TB }
for the event that could more precisely be denoted {T A < TB , T A < ∞}.
If A, B are disjoint then the sample space  can be written as the disjoint
union

 = {T A < TB , T A < ∞} ∪ {TB < T A , TB < ∞}

∪ {T A = TB = ∞} ∪ {T A = TB < ∞}.

The following theorem is proved in most first courses on Markov chains –

see for example [Dur, KSK, Nor], or Appendix A.2.

Theorem 1.8 Let  be connected, locally finite, infinite.

(T) The following five conditions are equivalent:
8 Introduction

(a) There exists x ∈ V such that Px (Tx+ < ∞) < 1.

(b) For all x ∈ V, Px (Tx+ < ∞) < 1.
(c) For all x ∈ V,
∞
Pn (x, x) < ∞.
n=0

(d) For all x, y ∈ V with x  = y either Px (Ty < ∞) < 1 or P y (Tx < ∞) < 1.
(e) For all x, y ∈ V,
∞
Px (X hits y only finitely often) = Px 1(X n =y) < ∞ = 1.
n=0

(R) The following five conditions are equivalent:

(a) There exists x ∈ V such that Px (Tx+ < ∞) = 1.
(b) For all x ∈ V, Px (Tx+ < ∞) = 1.
(c) For all x ∈ V,
∞
Pn (x, x) = ∞.
n=0

(d) For all x, y ∈ V, Px (Ty < ∞) = 1.

(e) For all x, y ∈ V,


∞
Px (X hits y infinitely often) = Px 1(X n =y) = ∞ = 1.
n=0

Definition 1.9 If (T)(a)–(e) of Theorem 1.8 hold we say  (or X ) is transient;

otherwise (R)(a)–(e) of Theorem 1.8 hold and we say  or X is recurrent. The
type problem for a graph  is to determine whether it is transient or recur-
rent. A special case is given by the Cayley graphs of groups, and is sometimes
called Kesten’s problem: which groups have recurrent Cayley graphs? In this
case, recall that the Cayley graph depends on both the group G and the set of
generators .

Let us start with the Euclidean lattices. Polya [Pol] proved the following in
1921.

Theorem 1.10 Zd is recurrent if d ≤ 2 and transient if d ≥ 3.

Polya used the fact that X n is a sum of independent random variables, and
that therefore the Fourier transform of Pn (0, x) is a product. Inverting the
Fourier transform, he deduced that P2n (0, 0) ∼ cd n −d/2 , and hence proved
his theorem.
 1.2 Random Walks on a Weighted Graph 9

Most modern textbooks give a combinatorial proof. For d = 1 we have

2n 1
P2n (0, 0) = 2−2n ∼ ;
n (πn)1/2
this is easy using Stirling’s formula. For d = 2 we have

n
2n 2
1
P2n ((0, 0), (0, 0)) = 4−2n ∼ ,
k πn
k=0

after a bit more work. (Or we can use the trick of writing Z n = ( 12 (X n +
Yn ), 12 (X n − Yn )), where X and Y are independent SRW on Z.) In general we
obtain cd n −d/2 , so the series converges when d ≥ 3. This expression is to be
expected: by the local limit theorem (a variant of the central limit theorem),
we have that the distribution of X n is approximately the multivariate Gaussian
N (0, n 1/2 d −1/2 Id ). (Of course the fact that, after rescaling, the SRW X on Zd
with d ≥ 3 converges weakly to a transient process does not prove that X is
transient, since transience/recurrence is not preserved by weak convergence.)
The advantage of the combinatorial argument is that it is elementary. On
the other hand, the details are a bit messy for d ≥ 3. More significantly, the
technique is not robust. Consider the following three examples:
(1) The SRW on the hexagonal lattice in R2 .
(2) The SRW on a graph derived from a Penrose tiling.
(3) The graph (, μ) where  = Zd , and the weights μ satisfy μx y ∈ [c−1 , c].

Example (1) could be handled by the combinatorial method, but the details
would be awkward, since one now has to count six kinds of steps. Also it
is plainly a nuisance to have to give a new argument for each new lattice.
Polya’s method does work for this example, but both methods look hopeless
for (2) or (3), since they rely on having an exact expression for Pn (x, x) or
its transform. (For the convergence of SRW on a Penrose tiling to Brownian
motion see [T3, BT].)

We will be interested in how the geometry of  is related to the long run

behaviour of X . As far as possible we want techniques which are stable under
various perturbations of the graph.

Definition 1.11 Let P be some property of a weighted graph (, μ), or the
SRW X on it. P is stable under bounded perturbation of weights (weight
stable) if whenever (, μ) satisfies P, and μ are weights on  such that
c−1 μx y ≤ μx y ≤ cμx y , x, y ∈ V,
then (, μ ) satisfies P. (We say the weights μ and μ are equivalent.)
10 Introduction

Definition 1.12 Let (X i , di ), i = 1, 2 be metric spaces. A map ϕ : X 1 → X 2

is a rough isometry if there exist constants C1 , C 2 such that
C1−1 (d1 (x, y) − C2 ) ≤ d2 (ϕ(x), ϕ(y)) ≤ C1 (d1 (x, y) + C2 ), (1.5)

Bd2 (ϕ(x), C2 ) = X 2 . (1.6)

x∈X 1

If there exists a rough isometry between two spaces they are said to be roughly
isometric. (One can check that this is an equivalence relation.)

This concept was introduced for groups by Gromov [Grom1, Grom2]

under the name quasi-isometry, and in the context of manifolds by Kanai
[Kan1, Kan2]. A rough isometry between two spaces implies that the two
spaces have the same large scale structure. However, to get any useful con-
sequences of two spaces being roughly isometric one also needs some kind of
local regularity. This is usually done by considering rough isometries within a
family of spaces satisfying some fixed local regularity condition. (For exam-
ple, Kanai assumed the manifolds had bounded geometry: that is, that the Ricci
curvature was bounded below by a constant.) Sometimes one has to be careful
not to forget such ‘hidden’ side conditions.

Example Zd , Rd , and [0, 1] × Rd are all roughly isometric.

The following will be proved in Chapter 2 – see Proposition 2.59.

Proposition 1.13 Let G be a finitely generated infinite group, and ,  be

two sets of generators. Let ,  be the associated Cayley graphs. Then  and
 are roughly isometric.

For rough isometries of weighted graphs, the natural additional regularity

condition is to require that both graphs have controlled weights. Using Lemma
1.3 and the condition (1.7) below this allows one to relate the measures of balls
in the two graphs.

Definition 1.14 Let (i , μi ), i = 1, 2 be weighted graphs satisfying (H5). A

map ϕ : V1 → V2 is a rough isometry (between (1 , μ1 ) and (2 , μ2 )) if:
(1) ϕ is a rough isometry between the metric spaces (V1 , d1 ) and (V2 , d2 )
(with constants C1 and C2 );
(2) there exists C 3 < ∞ such that for all x ∈ V1
C3−1 μ1 (x) ≤ μ2 (ϕ(x)) ≤ C 3 μ1 (x). (1.7)
 1.3 Transition Densities and the Laplacian 11

Two weighted graphs are roughly isometric if there is a rough isometry

between them; one can check that this is also an equivalence relation. We define
stability under rough isometries of a property P of  or X in the obvious way.

Example Consider the graphs (Z+ , μ(α) ) defined in Examples 1.4, and let
ϕ : Z+ → Z+ be the identity map. Then ϕ is (of course) a rough isometry
between the graphs Z+ and Z+ , but if α  = β it is not a rough isometry between
the weighted graphs (Z+ , μ(α) ) and (Z+ , μ(β) ). We will see in Chapter 2 that
(Z+ , μ(α) ) is recurrent if and only if α ≤ 1, and that the type of a graph is
stable under rough isometries (of weighted graphs).

Question Is there any ‘interesting’ (global) property of the random walk on

a weighted graph (, μ) which is weight stable but not stable under rough
isometries? (Properties such as  being bipartite or a tree are not global.)

1.3 Transition Densities and the Laplacian

Rather than working with the transition probabilities P(x, y), it is often more
convenient to consider the transition density of X with respect to the measure
μ; we will also call this the (discrete time) heat kernel of the graph . We set
Pn (x, y) Px (X n = y)
pn (x, y) = = .
μy μy
We write p(x, y) = p1 (x, y), and note that
1x (y) μx y
p0 (x, y) = , p(x, y) = p1 (x, y) = .
μx μx μ y
It is easy to verify:

Lemma 1.15 The transition densities of X satisfy


pn+m (x, y) = pn (x, z) pm (z, y)μz , (1.8)
z
pn (x, y) = pn (y, x),
 
pn (x, y)μ y = μx pn (x, y) = 1.
y x

Define the function spaces

C(V) = RV = { f : V → R},
C0 (V) = { f : V → R such that f has finite support}.
12 Introduction

Let C+ (V) = { f : V → R+ }, and write C0,+ = C0 ∩ C+ etc. For f ∈ C(V)

we write
supp( f ) = {x : f (x)  = 0}.
For p ∈ [1, ∞) and f ∈ C(V) let
p

|| f || p = | f (x)| p μx ,
x∈V
and
L p (V, μ) = { f ∈ C(V) : || f || p < ∞}.
We set || f ||∞ = supx | f (x)| and L ∞ (V, μ) = { f : || f ||∞ < ∞}. If p ∈
[1, ∞) and An ↑↑ V then f 1 An → f in L p for any f ∈ L p , so that C0 is
dense in L p .
We write  f, g for the inner product on L 2 (V, μ):

 f, g = f (x)g(x)μx .
x∈V
We define the operators
 
Pn f (x) = Pn (x, y) f (y) = pn (x, y) f (y)μ y = Ex f (X n ), (1.9)
y∈V y∈V

and write P = P1 . Since we are assuming (H1), we have that Pn f is defined

for any f ∈ C(V). Using Lemma 1.15 we obtain
Pn = P n .
The (probabilistic) Laplacian is defined on C(V) by
1 
f (x) = μx y ( f (y) − f (x))
μx
y∈V

= p(x, y)( f (y) − f (x))μ y = Ex f (X 1 ) − f (x).
y

In operator terms we have

= P − I.

Notation We write A p→ p to denote the norm of A as an operator from

L p to L p , that is,
A p→ p = sup{A f  p :  f  p ≤ 1}.
Let q be the conjugate index of p . Since, by the duality of L p and L q ,
|| f || p = sup{|| f g||1 : ||g||q ≤ 1},
 1.3 Transition Densities and the Laplacian 13

we also have
A p→ p = sup{g A f 1 :  f  p ≤ 1, gq ≤ 1}. (1.10)

Proposition 1.16 (a) P1 = 1.

(b) For f ∈ C(V), |P f | ≤ P| f |.
(c) For each p ∈ [1, ∞], ||P|| p→ p ≤ 1, and || || p→ p ≤ 2.

Proof (a) and (b) are easy. For (c), let first p ∈ [1, ∞) and q be the conjugate
index. Let f ∈ L p (V, μ). Then, using Hölder’s inequality,
p
   p
||P f || p = |P f (x)| p μx = p(x, y) f (y)μ y μx
x x y
   1/ p   1/q p
≤ p(x, y)| f (y)| μ y p
p(x, y)1q μ y μx
x y y
  p
= p(x, y)| f (y)| p μx μ y = | f (y)| p μ y = || f || p .
x y y

If p = ∞ the result is easy:

 
|P f (x)| = | p(x, y) f (y)μ y | ≤ || f ||∞ p(x, y)μ y = || f ||∞ .
y y

As = P − I the second assertion is immediate.

Remark Note that this proof did not actually use the symmetry of p(x, y),

but just the fact that x μx p(x, y) = 1, that is, that μ is an invariant measure
for P.

Remark 1.17 The combinatorial Laplacian is defined by Com by


Com f (x) = μx y ( f (y) − f (x)). (1.11)
y∈V

In matrix terms Com = A − D, where A x y = μx y , and D is the diagonal

matrix with Dx x = μx . If  has its natural weights, then A is the adjacency
matrix of . It is easy to check that A and hence Com is self-adjoint with
respect to counting measure on V.
If M = supx μx < ∞ then one has
p
  p
||A f || p = μx A x y f (y)
x y
   1/ p   1/q  p
≤ μx A x y | f (y)| p A x y 1q
x y y
14 Introduction

  
≤ μx A x y | f (y)| p M p/q ≤ M 1+ p/q | f (y)| p μ y
x y y
p
=M p
|| f || p ,

which implies that ||A|| p→ p ≤ M. Since ||D|| p→ p ≤ M this gives

|| Com || p→ p ≤ 2M.

One can then define the ‘delayed RW’ on (, μ) by considering the random
walk associated with the operator
1
P=I− Com .
M
This is the SRW on the weighted graph (V, E, μ ), where μx y = M −1 μx y if

x  = y and μx x = 1 − y=x μx y .
If M = ∞ then the operator Com is not naturally associated with any
discrete time random walk. However it can be associated in a natural way with
a continuous time random walk – see Remark 5.7.

Proposition 1.18 P and are self-adjoint on L 2 .

Proof This is clear for P since


P f, g = p(x, y) f (y)μ y g(x)μx
x y

= p(x, y) f (y)g(x)μ y μx
x y

= p(y, x)g(x)μx f (y)μ y =  f, Pg.
y x

Hence = P − I is also self-adjoint.

Remark 1.19 See Example 1.35 in Section 1.6 for a graph  and functions
u, v such that both u and v have finite support, but
 
v(x) u(x)μx  = u(x) v(x)μx .
x x

(Of course the functions u, v are not in L 2 .)

On probability spaces one has L p2 ⊂ L p1 if p2 > p1 . On discrete spaces

such as  the inclusions are in the opposite direction, provided that μx is
bounded below.
 1.4 Dirichlet or Energy Form 15

Lemma 1.20 Suppose that μx ≥ a > 0 for each x ∈ V. Then if p2 > p1

and f ∈ C(V),
|| f || p2 ≤ a 1/ p2 −1/ p1 || f || p1 .

Proof By replacing f by f /|| f || p1 we can assume that || f || p1 = 1. If x ∈ V

then | f (x)| p1 ≤ 1/μx ≤ 1/a. Hence
 
| f (x)| p1 | f (x)| p2 − p1 μx ≤ | f (x)| p1 a −( p2 − p1 )/ p1 μx
p
|| f || p22 =
x x
≤ a 1− p2 / p1 .
Write pnx (·) = pn (x, ·). We can rewrite the equations in Lemma 1.15 as:
y
pn+m (x, y) =  pnx , pm , (1.12)
P pnx (y) = x
pn+1 (y), (1.13)
pnx (y) = (P − I ) pnx (y) = x
pn+1 (y) − pnx (y). (1.14)
Equation (1.14) is the discrete time heat equation. Note that (1.12) implies that
|| pnx ||22 ≤ μ−1
x for all x ∈ V, n ≥ 0.

1.4 Dirichlet or Energy Form

For f, g ∈ C(V) define the quadratic form

E ( f, g) = 12 μx y ( f (x) − f (y))(g(x) − g(y)), (1.15)
x∈V y∈V

whenever
 this sum converges absolutely. This is the discrete analogue of
∇ f · ∇g. Let
H 2 (V, μ) = H 2 (V) = H 2 = { f ∈ C(V) : E ( f, f ) < ∞}.
Choose a ‘base point’ o ∈ V and define
|| f ||2H 2 = E ( f, f ) + f (o)2 .
The sum in (1.15) is well defined for f, g ∈ H 2 (V). If A ⊂ V we will
sometimes use the notation

E A ( f, g) = 12 ( f (x) − f (y))(g(x) − g(y))μx y . (1.16)
x∈A y∈A

Proposition 1.21 (a) If x ∼ y then

−1/2
| f (x) − f (y)| ≤ μx y E ( f, f )1/2 .
Another random document with
no related content on Scribd:
of the clothed human form; in fact, an American scientist, John
Campbell, claims to have deciphered an ancient Japanese
inscription upon one of Grey’s figures referring to the “hopeless
number,” presumably of the castaways.

Fig. 44. Pictographic representation of emu-hunt.

In regard to the quaint head-gear which distinguishes these

designs, one need not go far from Australia to find something quite
analogous, for the Papuans wear an article which is quite similar to
what might be suggested by the drawings.

Fig. 45. Flying fox pattern.

We now turn to a more psychological aspect of primitive Australian

art, which includes such factors as convention and imagination.
These processes lie behind the symbolization of thought which has
evolved a means of pictographically conveying messages from one
individual to another, or, collectively, from one tribe to another.
Through long usage, the artist has learned to reduce the complexity
of a familiar, naturalistic design in such a way that, while still
retaining its intrinsic interpretation, he is able to demonstrate by a
few lines what ordinarily would require an intricate drawing. The
ultimate aim of such a system is, of course, to reduce the execution
of a design to a minimum of energy and time, without imperilling the
correctness of its interpretation. But whilst the artist is designing to
simplify the complexity of his thought symbol, the reciprocative factor
of assimilation on the part of his fellows is being stimulated. And by
means of this joint education, a design, with a simple motive from
Nature behind it, might gradually become so conventionalized that
the uninitiated fails entirely to grasp its significance.

PLATE XLIV

1. Hand marks in cave, Port George IV, Worora tribe.

2. Foot marks in cave, Port George IV, Worora tribe.

 During this evolution, or, should one say, metamorphosis, in the
perception of art, the tendency must necessarily be to smooth out
the irregular and unsymmetrical contours of Nature and bring the
design down to as nearly geometrical as possible. For the same
reason, a single design is often repeated indefinitely, so that a single
form derived from a simple motive may expand into a continuous
chain or ornate pattern covering a considerable area. A large series
of devices has been established, the majority of which are known all
over the Australian continent; but, as the same time, there are many
signs which are entirely of a “totemic” nature and can only be
understood by a person belonging to that particular “totem.”

Fig. 46. Conventional representation of hopping kangaroo.

Let us take a few simple cases illustrating the different points

enumerated above.
A simple drawing of a lizard would include a substantial body, with
rounded head and tapering tail; the legs are usually extended, and at
times have the claws marked. Vide Fig. 30.
The conventionalized form consists of a long straight line crossed
at right angles by two shorter bars, at points about one-quarter and
three-quarters distance respectively from one end.
A turtle design consists of an oval, representing the shield, from
which extend the head, tail, and paddles. In its modified form this
becomes a circle with six short lines radiating from the circumference
at equal distances apart (Fig. 31).
The picture of a frog in its simplified form becomes, like that of a
turtle, a circle, but has only four radiating lines (Fig. 32). When
designs like these are expanded symmetrically into patterns, the
result is after the style of that shown in the accompanying sketches.
These patterns are extensively used in totemic devices upon
tjuringas and implements.
The drawing of an echidna, or native hedgehog, ordinarily is like
the sketch shown above (Fig. 33), but as a result of its conventional
transformation it becomes a simple hexagon, from the corners of
which the limbs may or may not be shown as simply projecting lines.
The pattern obtained by linking up a number of hexagons is not
uncommonly found engraved upon weapons and implements.

Fig. 47. Crossed boomerangs, the

symbolic representation of a fight.

Many of the conventional patterns are not so apparent as the few

just mentioned. Let us take, for instance, the “Ladjia” or Yam
Tjuringa pattern of the Arunndta. Only the initiated would be able to
recognize in a number of groups of small, concentric circles,
regularly placed at the corners of rectangular figures, as the stems of
the yam plant, and a system of parallel lines connecting the circles
both straight and diagonally as the roots. Vide Fig. 34.
Tracks of animal, bird, and man are conspicuous among the
designs, generally, whether they be true copies from Nature upon
the walls of a cave, more or less modified gravures upon weapons,
or conventional patterns incised upon a message stick. A dog track
is represented by a larger dot, suggesting the imprint of the ball of
the paw, to which are attached four smaller dots, which lie in a row in
close proximity to the former; the smaller dots stand for the
impressions of the claws (Fig. 35).
A kangaroo track shows the two long parallel median impressions,
with a lateral at an acute angle to the former at either side. The same
figure is repeated a short distance away from the one described, and
in a straight line with it. The same design on a smaller scale denotes
a wallaby track. Occasionally the lateral lines are dispensed with
(Fig. 36).
Small oblong dots drawn in pair at equal distances from each
other indicate the hopping of a rabbit (Fig. 37).
The characteristic broad arrow-like footprint of an emu has already
been referred to; the smaller variety of the same design implies that
a wild turkey is meant. When a number of birds are to be
represented collectively, the archetype is developed into a
continuous pattern by linking one track up with another into a chain
(Fig. 38).

Fig. 48. Witchedy grub Tjuringa, Arunndta tribe (× 3/10). Tracing.

When attention is to be drawn to the fact that the birds are laying,
or sitting on eggs, a number of small circles are drawn about the
track (Fig. 39).
A lizard track is indicated by drawing a number of dots, equally
spaced along a straight line, and on alternate sides of it. The dots
are the claw impressions, the line the trail left by the tail (Fig. 40).
A single wavy line, or a number of parallel wavy lines, represents a
snake or a snake track (Fig. 41).
The human footprint is either correctly shown in detail, or is
reduced to a short, straight line at one end of which five
(occasionally only three or four) dots are drawn in a sloping line to
indicate the toes. When walking is to be implied, these footprints are
either shown one behind the other in a straight line, at equal
distances apart, or they stand alternately right and left by an
imaginary central line. The common way of showing the last-named
system conventionally is to connect the alternate footprints, whether
they be actually shown or not, by a zig-zag line (Fig. 42).
In all cases mentioned above, the track may stand equally well for
the object itself.

Fig. 49. Symbolic pictograph of kangaroo Tjuringa, Arunndta tribe.

It is a common thing to see two or more of the above systems

combined, after the style of the figures shown on page 344.
The interpretation of the first of these messages would be: “A man
is tracking a rabbit.”
The second (Fig. 44) would read: “A hunter is pursuing an emu
and is accompanied by his dog.”
In the same way as the natives of the Northern Territory have
applied their artistic talents to deciphering images of earthly objects
amongst the celestial bodies, and point with pride upon the great
emu, “Dangorra,” which nightly watches over them, so the
Wongapitcha and Aluridja of central Australia recognize in the
constellation we know as the Southern Cross the shape of an eagle-
hawk’s claw, and call it in consequence “Warridajinna”; the Milky
Way they consider to be a creek bed and assign to it the name
“Karru”; the northern Kimberleys tribes believe the Milky Way to be
the track of a great carpet snake they refer to as “Womma.”
 In the representation of a fish, the scales often take the form of a
cross-hatched pattern; but there are many cases in which the form of
the fish is not shown at all, yet the cross-hatched pattern remains
and is nevertheless indicative of the fish.

Fig. 50. Symbolic pictograph of caterpillar, Tjuringa, Arunndta tribe.

A design, fairly common in the north of Western Australia, consists

of two wavy lines which are parallel, in the inverted or reflected
sense, and joined at one end. The true significance of this pattern
does not become evident until one hides from view all but three of
the “waves,” say that portion lying to the right of the dotted line in the
accompanying sketch. When this is done, the form of a “flying fox” is
immediately recognized (Fig. 45).
In central, northern, and eastern Australia, a pattern frequently met
with on boomerangs, fighting sticks, and message sticks, consists of
strings of lenticles longitudinally striped and generally associated
with kangaroo tracks. This device is analogous to the one standing
for the walking of a man, viz. the zig-zag, in so far as it stands for the
hopping of a kangaroo. Here and there one finds the pattern
“finished off” at one end with the head of a kangaroo (Fig. 46).
A duel or tribal fight of any description is graphically recorded by
two crossed boomerangs, but the conventional derivative of this
design is simply supplied by two crossed lines (Fig. 47).
 Time is chronicled by two phases of the moon; a crescent standing
for new, and a circle for full moon.

Fig. 51. Symbolic drawing of “native-pear totem,” Arunndta tribe.

The substance, origin, home, or habitat of any creature figuring in

a drawing or gravure is particularized by the addition of a circle-
within-circle design. For instance, in the tjuringa of the witchedy grub
shown in Fig. 48, the parallel, straight lines, enclosed within a “U,” at
the left-hand side, represent the grub, the three concentric circle
groups in the centre are the gum trees in which it lives, the “U within
U” pattern at the right is an ancestor whose “totem” was the witchedy
grub, while the parallel lines at the extreme right of the tjuringa are
markings on the grub which have been adopted by the man who
owns the tjuringa, in the form of cicatrices, he cuts on his chest.
The “U within U” pattern is frequently met with engraved upon
tjuringas, and in most cases it conventionally conveys the idea of a
“sitting” person or animal. We have already noted something similar
in the peculiar concentric iron stains of Heavitree Gap (page 342),
but in the following three tjuringa drawings of the Arunndta additional
illustrations are given.

Fig. 52. Ochre drawing and tree-carving of man with shield,

Humbert River (× 1/10).

In the first instance, an “Arrera Knaninja” or Kangaroo Tjuringa,

the meat and the fat of the animal are represented by the two series
of concentric circles in the centre, whereas its sinew is indicated by
the horizontal lines connecting the circles. The numerous U groups
on both sides of the central figures stand for a great number of
kangaroo, each of which is sitting or lying on the ground (Fig. 49).
 PLATE XLV

1. Cave drawings, Forrest River, north-western Australia.

2. Decorating the body with pipe-clay, Humbert River, Northern Territory.

 The second example is from the “Yeapatja” or Caterpillar Tjuringa.
The large “circle-within-circle” groups in the centre are generally
recognized to be the bushes upon which the caterpillars were born,
while the U groups with smaller concentric circles in their centres are
intended for caterpillars attached to smaller plants upon which they
are feeding (Fig. 50).

Fig. 53. Human chain-pattern.

Lastly, a sacred drawing of the “Alangua Knaninja” or Native Pear

(Marsdenia) “totem,” which belongs to the Altjerringa women, is
composed of a central “circle-within-circle” group representing the
“totem,” whilst surrounding it a number of U groups are supposed to
be the mythic women seated on the ground (Fig. 51).

Fig. 54. Camps consisting of a man and his wife (left) and of eight men.

We shall now turn our attention to the consideration of the

representation of the human figure and its derivative forms. Several
more or less obvious designs have already been discussed. The first
step towards conventionalism is seen in the two figures from the
Humbert River district, the first an ochre cave drawing, the second a
carving on a boabab tree (Fig. 52). We notice in the ochre drawing,
which was one foot six inches high, a fairly shapely and solid figure
of a man holding a shield in his left hand; in the carving, which
measured one foot nine inches in height, the solid and shapely
outline has been reduced to a matter of just a few straight lines; that
is, if we neglect for the present the consideration of shield and
boomerang which the figure is holding. The result is, therefore, a
design resembling a Latin cross, at the lower end of which is
attached an inverted V-shaped symbol representing the legs; a small
circle may or may not be added to the top end to stand for the head.
 Fig. 55. Anthropomorphous designs, carved on
spear-throwers. Tracing.

This design is often repeated indefinitely in a lateral sense, so that

a rather ornate pattern results in which the individual figures “join
hands” and their “toes touch” below. A chained pattern, such as is
shown in the accompanying sketch (Fig. 53), may be noticed in Plate
XLV, 1, near the centre of the picture, below the ledge on which the
bold drawing of a snake appears, and on the same level as the semi-
human design on the extreme left.
The ultimate stage of this conventionalism so far as the human
figure is concerned is a simple, straight line, the upright arm of the
cross; this is extensively employed in carved representations of
people on message sticks. We might now hark back to the question
at the beginning of our disquisition on aboriginal art “what lies in a
mere line” and supply one answer at any rate.

Fig. 56. Anthropomorphous design, carved on pearl-shell, Sunday Island. Tracing

(× 1/4).

Any number of people might be represented by the same number

of short, straight, vertical lines. Followed by a zig-zag line, such
people are represented as being on the march. When a small dot
stands on each side of the straight line, or perhaps a number of dots
in intermediate positions between the lines, the design conveys the
idea that the people are camped, the dots standing for camp fires.
Moreover, when a horizontal line lies over the upright lines, the last-
named indicates that a hut, shelter, or breakwind was used during
the encampment (Fig. 54).
When there is an obvious, and intended, difference in the lengths
of the upright lines, the longer represent men, the shorter women.
Reverting now to the cross, and looking more closely into its
development, expansion, and embodiment in anthropomorphous
designs, we meet with one or two points of considerable interest.
We have had occasion to note that throughout central and
northern Australia, the tribes during the final acts of initiation
ceremonies make use of a sacred cross, called “Wanningi” or
“Wanninga” in the former region and “Parli” in the western portions of
the latter. These wanningi are constructed only immediately before
they are required and are destroyed again the moment the ceremony
is over. Women are not allowed to see them under any
consideration. A wanningi is made by fixing two pieces of wood
together in the shape of a cross, then, by starting at the intersection
of the arms, a long string, made of human hair, is wound spirally
round and round, from arm to arm, until the whole space between
the arms is filled in. The size of these crosses varies from three or
four inches up to two feet or more. This object is produced by the
Aluridja just before the mutilation of the neophyte is to take place. At
this critical moment of the youth’s life, when he is stepping from
adolescence across the great gap which will lead him to manhood,
the spirit of the Great Tukura presides invisibly concealed within the
wanningi, but returns to his high abode again when the function is
over. Vide Plate XLIII, 2.
 PLATE XLVI

Wordaman native with his body and head decorated in imitation of

skeleton and skull, Victoria River, Northern Territory.

The spiral winding of hair-string around the arms of the wanningi

associates the idea of the rhombic outline of the string with the arms
of the cross. In the representation of the human form, one often finds
the two patterns combined, or, it may be, the rhomb replaces the
cross.
In the example before us (Fig. 55, a), we have an engraved
pattern appearing on a spear-thrower, the motive of which, were it
not that the artist had added a human face, would have been very
difficult for the untrained eye to recognize. As it is, we have the
unmistakable evidence of an anthropomorphous design. Not only
does a modified rhomb represent the body of a man, but the figure
itself is flourishing the crossed arms of a “Wanningi” in its right hand.
The principal design thus identified passes, at the bottom, into a
pattern composed of several polygonal figures which may, no doubt,
be looked upon as derivatives of an original rhomb.
In the other illustration (Fig. 55, b), which is also a carving upon a
spear-thrower, the intricate association of the rhomb with the human
form is again apparent. The figure of a man, with face in profile, is
represented in a plain and more or less conventional way; the
straight trunk with the two arms resting upon the hips, symmetrically
on each side, in itself suggests the rhomb, but, in addition, most of
the intervening spaces have been filled in with parallel lines and a
cross-hatching pattern which embodies the rhomb.
Conventionalism in the representation of the human figure has
thus gone further than the mere inclusion of the cross or the rhomb
motive, by working in with the original design a derivative or new
pattern which fills up all the surrounding spaces.
By means of this system, a new element is introduced in the
shape of symmetry. If a vertical line, drawn through the centre of the
trunk, be taken as a median line in a simple design of the human
figure, all subsequent patterns which are drawn will be grouped
symmetrically about it. The most popular pattern used to fill up the
available spaces with, is one of a “concentric” type. By this method a
distinctive, bi-laterally symmetrical pattern is evolved, which after
prolonged usage may actually take the place of the original, and
have a true anthropomorphous significance.
Take the illustration of conventionalism of this kind shown on the
carved pearl shell covering from King Sound reproduced in Fig. 56.
The original motive was a simple line drawing of a human being after
the style of the one on the left-hand side. The next stage in its
evolution was brought about by blocking the areas between the
limbs, and between the head and arm, on either side, respectively, in
a manner which made the resulting pattern appear equally balanced
in respect of a median longitudinal line running through the back and
head of the original figure.
Very numerous designs of this nature are constantly met with in all
tribal areas of Australia, but in most cases the stranger who is not
aware of the intermediate or transitional stages may fail entirely to
grasp their meaning or origin.
 CHAPTER XXIX
STONE IMPLEMENTS

Survival of Stone Age in Australia—Stones used in their natural shape for

throwing, pounding, cooking, and grinding purposes—Hand-mills—Rasps—
Stone tomahawks—Scrapers—Operating knives—“Cores” or “nuclei”—Stone
knives—Spokeshaves—Awls—Concave scrapers—Slate scrapers of
Adelaide tribe—Scrapers embedded in resin—Adzes—Bladed spears and
knives—Stone spear-heads—Method of manufacture described.

There are not many places left in the world where the man of the
Stone Age can still be seen roaming the wilds of his inherited
possessions. Even in Australia, although there remain one or two
areas where comparatively little havoc has been wrought among the
primitive institutions of the indigenous man, yet the influence of
civilization is slowly, but very surely, encroaching indirectly upon his
ancient cults by such means as inter-tribal barter, if not actually by
the hand of the white intruder. Especially do these remarks apply to
the manufacture and utilization of stone implements; it is, of course,
only to be expected that the superiority of the metal blades of the
white man’s implements would appeal to the native who formerly had
to spend hours making a crude cutting edge which only too often
broke when applied to the test for the first time. We shall, however,
treat the subject regardless of the alterations which have been
brought about by our appearance upon the scene, and without
attempting to draw up a hard and fast scheme of classification.
At the present time, whilst there are not only some of the primitive
men alive still, but also a limited number of observers who have had
the good fortune of seeing them at work, it is of vastly greater
importance to record the living facts than to write exhaustively, nay,
even speculatively, upon the comparative shapes and embodied
techniques of artefacts whose stony composition will ensure their
keeping, even fossilized, long after the men who made them, and the
scientists who lived among them, have passed into oblivion.
The Australian aboriginal makes adequate use of any suitably
shaped pieces of stone he happens to find whilst in pursuit of game;
both in the Musgrave Ranges and the northern Kimberleys stones
are used in their natural shape for hurling into a flying flock of birds,
for shying at a bounding wallaby, for bringing down nuts of the
boabab, and for precipitating fledgelings out of a nest.
Other stones, usually oblong and rounded pebbles gathered in a
river bed, are used for pounding and cracking purposes. At any
camping ground these pebbles can be picked up in great numbers,
showing one or two places, usually the points, at which the
percussions have worn the stone away; pounding stones and
hammers of this description are equally plentiful in the sandhills on
the plains of Adelaide, all over central Australia, and along the north
coast. They are used for pounding seeds and foliage (the latter of
which is to serve for corrobboree decorations), for pulverizing ochre,
for cracking nuts and hollow bones containing marrow. Vide Plate
LIV, 1.
The underlying surface consists either of a level portion of an
outcrop or another, but larger, stone, which takes the place of an
anvil. Some of the coastal tribes of eastern Australia used to shape
their heavier pounding stones by chipping away material at one side
until a stout, cylindrical handle was formed, the whole resembling a
pestle; dumb-bell shaped pounders were also made, but were rare.
We have already learnt that natural pebbles or rock fragments are
also used, together with a wooden rod, for knocking out teeth during
initiation ceremonies.
 PLATE XLVII

1. Cave drawings (kangaroo, etc.), Forrest River, north-western

Australia.

2. Cave-drawing of kangaroo, Forrest River, north-western

Australia.