2018 Book ContactMechanics PDF

Solid Mechanics and Its Applications
J. R. Barber
Contact
Mechanics
Volume 250
Series editors
J. R. Barber, Ann Arbor, USA
Anders Klarbring, Linköping, Sweden
Founding editor
G. M. L. Gladwell, Waterloo, ON, Canada
Aims and Scope of the Series
The fundamental questions arising in mechanics are: Why?, How?, and How much?
The aim of this series is to provide lucid accounts written by authoritative
researchers giving vision and insight in answering these questions on the subject of
mechanics as it relates to solids.
The scope of the series covers the entire spectrum of solid mechanics. Thus it
includes the foundation of mechanics; variational formulations; computational
mechanics; statics, kinematics and dynamics of rigid and elastic bodies: vibrations
of solids and structures; dynamical systems and chaos; the theories of elasticity,
plasticity and viscoelasticity; composite materials; rods, beams, shells and
membranes; structural control and stability; soils, rocks and geomechanics;
fracture; tribology; experimental mechanics; biomechanics and machine design.
The median level of presentation is to the first year graduate student. Some texts
are monographs defining the current state of the field; others are accessible to final
year undergraduates; but essentially the emphasis is on readability and clarity.
More information about this series at http://www.springer.com/series/6557

J. R. Barber
Contact Mechanics
123
J. R. Barber
Department of Mechanical Engineering
University of Michigan
Ann Arbor, MI
USA
ISSN 0925-0042 ISSN 2214-7764 (electronic)

ISBN 978-3-319-70938-3 ISBN 978-3-319-70939-0 (eBook)
https://doi.org/10.1007/978-3-319-70939-0
Library of Congress Control Number: 2017958025
© Springer International Publishing AG 2018

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, express or implied, with respect to the material contained herein or
for any errors or omissions that may have been made. The publisher remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.
Printed on acid-free paper
This Springer imprint is published by Springer Nature

The registered company is Springer International Publishing AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
For many years, contact mechanics was more or less synonymous with Hertzian
contact where the contacting bodies are elastic with quadratic profiles, and most
of the applications were to traditional engineering components such as rolling
bearings, cams and gears. However, more recent applications cover an extraordi-
narily diverse range, including natural and artificial hip joints, the slip of tectonic
plates during earthquakes, the adhesion of gecko feet to a wall when climbing, the
interpretation of atomic force microscope (AFM) results and many others. Indeed in
almost all systems comprising more than a single object, loads are transmitted
between the components by contact, and the nature of this interaction is often
critical in determining the overall system behaviour.
For ‘macroscopic’ systems, the contact interaction can usually be simplified by
defining a dichotomy between the states of contact and separation. Bodies in
contact can transmit loads and conduct heat and electricity, whereas these processes
are either impossible or much reduced if the bodies are separated by even a small
gap. In idealized models, this distinction typically translates into a problem gov-
erned by inequalities and the resulting strong nonlinearity is a rich source of
interesting and complex mathematical phenomena. Additional inequalities are
introduced through the transition from stick to slip in problems involving friction.
However, recent applications increasingly involve very small length scales,
where the distinction between contact and separation is blurred, and we must also
recognize the possibility of tensile [adhesive] tractions between the bodies, for
example, due to van der Waals forces. In such cases, the inevitable roughness of the
contacting surfaces plays an important rôle. In biological applications, the materials
are likely to experience large strains and the characterization of the material con-
stitutive law is a challenge. Also, at extremely small scales, we must recognize that
continuum formulations will become inappropriate.
These considerations imply that contact problems are likely to be encountered by
scientists who are not primarily interested in contact mechanics per se, and that the
resulting problems can be quite complex, usually necessitating numerical solution
methods. Fortunately, modern finite element packages are very user-friendly and
increasingly contain modules describing a range of multiphysics interactions
v
vi Preface
between surfaces. However, the user will often encounter unexpected [and some-
times unexpectedly simple] predictions, and this always raises the question as to
whether an indealized analytical treatment may be sufficient to capture and indeed
‘explain’ the qualitative behaviour of the system, whilst providing a greater level of
generality and hence predictive power. Analytical models also have the benefit of
identifying the principal determinants of the behaviour of the system [particularly
when supplemented with finite element predictions for the same system], and hence
provide guidance as to which features of the underlying physics or geometry require
especially careful characterization or measurement.
These considerations have influenced the presentation of this book. Although
some relatively complicated mathematical treatments are addressed, there is an
emphasis on qualitative physical behaviour, and on situations where a simpler
approach gives useful results. Also, I have included problems at the end of each
chapter, principally as an indication to the reader of how the methods and concepts
discussed can be applied in different systems, though these problems are also
suitable as assignments in a course on contact mechanics.
The field of contact mechanics has expanded considerably since K. L. Johnson
published his classical monograph in 1985, and arguably it would be impossible to
achieve the same level of completeness in a single book today. I therefore make no
apology for focussing on topics with which I have had first-hand experience, and
this [rather than any misplaced sense of self-importance] explains the high pro-
portion of citations to my own work and to that of my graduate students and
collaborators. In particular, most of the book relates to linear elastic materials and
there is relatively little discussion of interior stress fields, even though these can be
of importance for failure analysis, particularly with reference to surface durability.
Other significant omissions include numerical methods, lubrication, plasticity and
viscoelasticity.
Ann Arbor, USA J. R. Barber

2017
Contents
1 Kinematics of Contact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Reference Frame and the Initial Gap Function . . . . . . . . . . . . . 2
1.2 Establishment of a Contact Region . . . . . . . . . . . . . . . . . . . . . 3
1.2.1 Definition of Contact . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.2 The Boundary Value Problem . . . . . . . . . . . . . . . . . . . 4
1.2.3 Signorini Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.4 Asymptotic Arguments . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.5 The Discrete Problem . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3 Nonlinear Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4 Almost Conformal Contact . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2 Three-Dimensional Frictionless Elastic Problems . . . . . . . . . . . . . . 13
2.1 The Half-Space Approximation . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 Normal Loading of the Half-Space . . . . . . . . . . . . . . . . . . . . . 14
2.2.1 The Point Force Solution . . . . . . . . . . . . . . . . . . . . . . 15
2.2.2 Similarity, Equilibrium and Anisotropy . . . . . . . . . . . . 16
2.2.3 The Composite Elastic Modulus . . . . . . . . . . . . . . . . . 17
2.3 Integral Equation Formulation . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.3.1 Field-Point Integration . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3.2 Indentation by a Flat Elliptical Punch . . . . . . . . . . . . . 20
2.4 Galin’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.4.1 A Special Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.5 Interior Stress Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.5.1 In-Plane Stress Components Near the Surface . . . . . . . 25
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3 Hertzian Contact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.1 Transformation of Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.1.1 Cylinders and Spheres . . . . . . . . . . . . . . . . . . . . . . . . 31
vii
viii Contents
3.1.2 More General Cases . . . . . . . . . . . . . . . . . . . . . . . . . . 32

3.2 Hertzian Pressure Distribution . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.3 Strategy for Hertzian Contact Calculations . . . . . . . . . . . . . . . . 34
3.3.1 Eccentricity of the Contact Area . . . . . . . . . . . . . . . . . 34
3.3.2 Dimensions of the Contact Area . . . . . . . . . . . . . . . . . 35
3.3.3 Highly Elliptical Contacts . . . . . . . . . . . . . . . . . . . . . . 38
3.4 First Yield . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4 More General Problems for the Half-Space . . . . . . . . . . . . . . . . . . 43
4.1 The Electrical–Mechanical Analogy . . . . . . . . . . . . . . . . . . . . . 44
4.1.1 Other Mathematical Analogies . . . . . . . . . . . . . . . . . . 46
4.1.2 Boyer’s Approximation . . . . . . . . . . . . . . . . . . . . . . . 48
4.1.3 Fabrikant’s Approximation . . . . . . . . . . . . . . . . . . . . . 49
4.2 General Theorems for Frictionless Contact . . . . . . . . . . . . . . . . 51
4.3 Superposition by Differentiation . . . . . . . . . . . . . . . . . . . . . . . . 55
4.4 The Force–Displacement Relation . . . . . . . . . . . . . . . . . . . . . . 57
4.4.1 Non-conformal Contact Problems . . . . . . . . . . . . . . . . 58
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5 Axisymmetric Contact Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.1 Green and Collins Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.1.1 The Flat Punch Solution . . . . . . . . . . . . . . . . . . . . . . . 65
5.2 Non-conformal Contact Problems . . . . . . . . . . . . . . . . . . . . . . 66
5.3 Annular Contact Regions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.4 The Non-axisymmetric Cylindrical Punch . . . . . . . . . . . . . . . . . 69
5.5 The Method of Dimensionality Reduction (MDR) . . . . . . . . . . 70
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
6 Two-Dimensional Frictionless Contact Problems . . . . . . . . . . . . . . . 77
6.1 The Line Force Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
6.2 Integral Equation Formulation . . . . . . . . . . . . . . . . . . . . . . . . . 80
6.2.1 Edge Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
6.3 Incremental Solution of Non-conformal Contact Problems . . . . . 85
6.3.1 Symmetric Problems . . . . . . . . . . . . . . . . . . . . . . . . . . 85
6.3.2 Bounded-Singular Problems . . . . . . . . . . . . . . . . . . . . 86
6.4 Solution by Fourier Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
6.4.1 Rigid-Body Rotation . . . . . . . . . . . . . . . . . . . . . . . . . . 87
6.4.2 Galin’s Theorem, Chebyshev Polynomials and
Recurrence Relations . . . . . . . . . . . . . . . . . . . . . . . . . 89
6.5 Periodic Contact Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
6.5.1 Sinusoidal Contact Pressure . . . . . . . . . . . . . . . . . . . . 91
6.5.2 Fourier Series Methods . . . . . . . . . . . . . . . . . . . . . . . . 92
6.5.3 The Periodic Green’s Function . . . . . . . . . . . . . . . . . . 93
6.5.4 The Cotangent Transform . . . . . . . . . . . . . . . . . . . . . . 93
Contents ix
6.5.5 Manners’ Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

6.5.6 Westergaard’s Problem . . . . . . . . . . . . . . . . . . . . . . . . 96
6.6 The Smirnov–Sobolev Transform . . . . . . . . . . . . . . . . . . . . . . . 97
6.6.1 Inversion of the Transform . . . . . . . . . . . . . . . . . . . . . 98
6.6.2 Example: Uniform Loading Over the Circle . . . . . . . . . 98
6.6.3 Anisotropic Problems . . . . . . . . . . . . . . . . . . . . . . . . . 99
6.7 Displacements in Two-Dimensional Problems . . . . . . . . . . . . . . 100
6.7.1 Kalker’s Line Contact Theory . . . . . . . . . . . . . . . . . . . 102
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
7 Tangential Loading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
7.1 Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
7.1.1 Gross Slip and Microslip . . . . . . . . . . . . . . . . . . . . . . 110
7.2 Green’s Functions for Tangential Forces and Displacements . . . 111
7.2.1 Three-Dimensional [point] Loading . . . . . . . . . . . . . . . 111
7.2.2 Two-Dimensional [line] Loading . . . . . . . . . . . . . . . . . 113
7.2.3 Normal-Tangential Coupling . . . . . . . . . . . . . . . . . . . . 114
7.3 Two-Dimensional Flat Rigid Punch with No Slip . . . . . . . . . . . 115
7.3.1 Uncoupled Problem . . . . . . . . . . . . . . . . . . . . . . . . . . 117
7.3.2 Oscillatory Singularities . . . . . . . . . . . . . . . . . . . . . . . 117
7.4 Axisymmetric Flat Rigid Punch with No Slip . . . . . . . . . . . . . . 119
7.5 The ‘Goodman’ Approximation . . . . . . . . . . . . . . . . . . . . . . . . 121
7.6 Uniform Tangential Displacement in a Prescribed Area . . . . . . . 123
7.6.1 Tangential Loading over a Circular Area . . . . . . . . . . . 123
7.6.2 Tangential Loading over an Elliptical Area . . . . . . . . . 124
7.6.3 Two Conjectures . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
7.7 Non-conformal Contact Problems with No Slip . . . . . . . . . . . . 127
7.7.1 Uncoupled Hertzian Contact with Tangential
Loading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
7.7.2 The Coupled Axisymmetric Problem under Purely
Normal Loading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
7.7.3 The Coupled Two-Dimensional Problem . . . . . . . . . . . 130
7.7.4 Relaxation Damping . . . . . . . . . . . . . . . . . . . . . . . . . . 132
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
8 Friction Laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
8.1 Amontons’ Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
8.1.1 Continuum Problems . . . . . . . . . . . . . . . . . . . . . . . . . 138
8.1.2 Two-Dimensional Problems . . . . . . . . . . . . . . . . . . . . 139
8.1.3 Existence and Uniqueness Theorems . . . . . . . . . . . . . . 139
8.2 The Klarbring Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
8.2.1 General Loading Scenarios . . . . . . . . . . . . . . . . . . . . . 142
8.2.2 The Critical Coefficient of Friction . . . . . . . . . . . . . . . 142
8.2.3 Wedging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
x Contents
8.3 Multinode Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144

8.3.1 The Evolution and Rate Problems . . . . . . . . . . . . . . . . 145
8.3.2 Algorithms for Two-Dimensional Problems with
Time-Varying Forces . . . . . . . . . . . . . . . . . . . . . . . . . 145
8.3.3 History-Dependence and Memory . . . . . . . . . . . . . . . . 146
8.3.4 Klarbring’s P-Matrix Criterion . . . . . . . . . . . . . . . . . . 147
8.4 Periodic Loading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
8.4.1 A Uniqueness Proof for Uncoupled Systems . . . . . . . . 149
8.4.2 Shakedown . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
8.4.3 Coupled Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
8.4.4 Asymptotic Approach to a Steady State . . . . . . . . . . . . 151
8.5 A Simple Continuum Frictional System . . . . . . . . . . . . . . . . . . 152
8.5.1 Unloading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
8.5.2 Periodic Loading . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
8.5.3 Discrete Model of the Strip Problem . . . . . . . . . . . . . . 157
8.5.4 The Inverse Problem . . . . . . . . . . . . . . . . . . . . . . . . . 157
8.6 More Complex Friction Laws . . . . . . . . . . . . . . . . . . . . . . . . . 158
8.6.1 Instabilities During Steady Sliding . . . . . . . . . . . . . . . . 159
8.6.2 Velocity-Dependent Friction Coefficient . . . . . . . . . . . . 159
8.6.3 Stick-Slip Vibrations . . . . . . . . . . . . . . . . . . . . . . . . . 161
8.6.4 Slip-Weakening Laws . . . . . . . . . . . . . . . . . . . . . . . . . 162
8.6.5 Rate-State Laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
9 Frictional Problems Involving Half-Spaces . . . . . . . . . . . . . . . . . . . 169
9.1 Cattaneo’s Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
9.2 The Ciavarella–Jäger Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 172
9.2.1 Three-Dimensional Problems . . . . . . . . . . . . . . . . . . . . 174
9.3 More General Loading Scenarios . . . . . . . . . . . . . . . . . . . . . . . 175
9.3.1 Constant Normal Force . . . . . . . . . . . . . . . . . . . . . . . . 175
9.3.2 Variable Normal Force . . . . . . . . . . . . . . . . . . . . . . . . 176
9.3.3 Memory and ‘Advancing Stick’ . . . . . . . . . . . . . . . . . 178
9.4 The Effect of Bulk Stress . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
9.4.1 Hertz Problem with Superposed Bulk Stress . . . . . . . . 179
9.4.2 Combined Bulk Stress and Tangential Force . . . . . . . . 181
9.5 Coupled Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
9.5.1 Indentation by a Two-Dimensional Flat
Rigid Punch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
9.5.2 Normal Loading for More General Geometries . . . . . . . 187
9.5.3 Combined Normal and Tangential Loading . . . . . . . . . 189
9.5.4 Unloading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
9.5.5 Periodic Loading . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
Contents xi
10 Asymptotic Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195

10.1 Indentation by a Frictionless Rigid Punch . . . . . . . . . . . . . . . . 195
10.1.1 Eigenfunction Series . . . . . . . . . . . . . . . . . . . . . . . . . . 197
10.1.2 More General Frictionless Indentation Problems . . . . . . 198
10.1.3 Non-conformal Problems . . . . . . . . . . . . . . . . . . . . . . 199
10.1.4 Both Materials Deformable . . . . . . . . . . . . . . . . . . . . . 200
10.2 No-Slip Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
10.3 Frictional Slip . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
10.3.1 Slip-Separation Transition . . . . . . . . . . . . . . . . . . . . . . 203
10.3.2 Slip–Stick Transition . . . . . . . . . . . . . . . . . . . . . . . . . 204
10.4 Indentation by an Elastic Wedge . . . . . . . . . . . . . . . . . . . . . . . 205
10.4.1 Right-Angle Wedge of the Same Material . . . . . . . . . . 206
10.4.2 A Slipping Interface . . . . . . . . . . . . . . . . . . . . . . . . . . 207
10.5 Local Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
10.5.1 The Flat and Rounded Indenter . . . . . . . . . . . . . . . . . . 209
10.5.2 Fretting in Non-conformal Contact . . . . . . . . . . . . . . . 211
10.5.3 Edge Slip Zones with a Rigid Punch . . . . . . . . . . . . . . 212
10.5.4 Slip Zones in Conformal Contact . . . . . . . . . . . . . . . . 214
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
11 Receding Contact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
11.1 Characteristics of Receding Contact . . . . . . . . . . . . . . . . . . . . . 222
11.1.1 Examples of Receding Contact . . . . . . . . . . . . . . . . . . 223
11.2 Frictional Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
11.2.1 Frictional Unloading . . . . . . . . . . . . . . . . . . . . . . . . . . 226
11.3 Thermoelastic Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
11.4 Almost Conformal Contact Problems . . . . . . . . . . . . . . . . . . . . 229
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
12 Adhesive Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
12.1 Adhesion Between Rigid Bodies . . . . . . . . . . . . . . . . . . . . . . . 236
12.2 The JKR Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
12.2.1 Axisymmetric Problems . . . . . . . . . . . . . . . . . . . . . . . 238
12.2.2 Indentation by a Sphere . . . . . . . . . . . . . . . . . . . . . . . 239
12.2.3 Energetic Considerations and Stability . . . . . . . . . . . . . 241
12.2.4 Hysteretic Energy Dissipation . . . . . . . . . . . . . . . . . . . 243
12.2.5 JKR Solution for More General Axisymmetric
Bodies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
12.2.6 Guduru’s Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
12.3 The Tabor Parameter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
12.3.1 An Adhesive Length Scale . . . . . . . . . . . . . . . . . . . . . 248
12.3.2 Limitations on the JKR Solution . . . . . . . . . . . . . . . . . 249
12.4 Solutions for Finite Tabor Parameter . . . . . . . . . . . . . . . . . . . . 250
12.4.1 Jump-In at Large Tabor Parameter . . . . . . . . . . . . . . . 251
xii Contents
12.4.2 Simplified Force Laws . . . . . . . . . . . . . . . . . . . . . . . . 252

12.4.3 Maugis’ Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
12.4.4 The ‘double-Hertz’ Approximation . . . . . . . . . . . . . . . 256
12.4.5 More General Axisymmetric Geometries . . . . . . . . . . . 258
12.5 Other Geometries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
12.5.2 Elliptical Contact Area . . . . . . . . . . . . . . . . . . . . . . . . 259
12.5.3 General Three-Dimensional Geometries . . . . . . . . . . . . 260
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
13 Beams, Plates, Membranes and Shells . . . . . . . . . . . . . . . . . . . . . . . 263
13.1 Contact of Beams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
13.1.1 A Heavy Beam Lifted from the Ground . . . . . . . . . . . . 265
13.1.2 Adhesive Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266
13.1.3 Piston Ring in a Cylinder . . . . . . . . . . . . . . . . . . . . . . 267
13.1.4 Two and Three-Dimensional Effects . . . . . . . . . . . . . . 270
13.1.5 Matched Asymptotic Expansions . . . . . . . . . . . . . . . . . 271
13.2 Contact of Plates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274
13.2.1 Displacement Due to a Concentrated Point Force . . . . . 275
13.2.2 Indentation by a Rigid Sphere . . . . . . . . . . . . . . . . . . . 275
13.3 Membrane Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
13.3.1 ‘Membrane Only’ Solutions . . . . . . . . . . . . . . . . . . . . 278
13.4 Contact of Shells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
13.5 Implications for Finite Element Solutions . . . . . . . . . . . . . . . . . 285
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
14 Layered Bodies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
14.1 Es El : Plate on an Elastic Foundation . . . . . . . . . . . . . . . . . . 290
14.1.1 Choice of Foundation Modulus . . . . . . . . . . . . . . . . . . 291
14.2 Es El : Layer on a Rigid Foundation . . . . . . . . . . . . . . . . . . . 295
14.2.1 Frictionless Unbonded Layer . . . . . . . . . . . . . . . . . . . . 296
14.2.2 Bonded Compressible Layer . . . . . . . . . . . . . . . . . . . . 298
14.2.3 Bonded Incompressible Layer . . . . . . . . . . . . . . . . . . . 298
14.2.4 Flat Punch Problems . . . . . . . . . . . . . . . . . . . . . . . . . . 303
14.2.5 Frictional Problems . . . . . . . . . . . . . . . . . . . . . . . . . . 304
14.2.6 Effect of Adhesive Forces . . . . . . . . . . . . . . . . . . . . . . 304
14.3 Winkler Layer on an Elastic Foundation . . . . . . . . . . . . . . . . . 307
14.3.1 Nonlinear Layers . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308
14.4 Fourier Transform Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
14.4.1 Elastic Layer Bonded to a Rigid Foundation . . . . . . . . 309
14.4.2 Multilayered Bodies . . . . . . . . . . . . . . . . . . . . . . . . . . 313
14.5 Functionally Graded Materials . . . . . . . . . . . . . . . . . . . . . . . . . 313
Contents xiii
14.5.1 Exponential Variation of Modulus . . . . . . . . . . . . . . . . 314

14.5.2 Power-Law Grading . . . . . . . . . . . . . . . . . . . . . . . . . . 315
14.5.3 Linear Variation of Modulus . . . . . . . . . . . . . . . . . . . . 318
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
15 Indentation Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
15.1 The Hardness Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
15.2 Power-Law Material . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
15.2.1 Graded Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . 326
15.3 Other Constitutive Laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 328
16 Contact of Rough Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
16.1 Bowden and Tabor’s Theory of Friction . . . . . . . . . . . . . . . . . . 329
16.1.1 The Ploughing Force . . . . . . . . . . . . . . . . . . . . . . . . . 330
16.1.2 Plastic Deformation at an Actual Contact . . . . . . . . . . . 331
16.1.3 The Effect of Surface Films . . . . . . . . . . . . . . . . . . . . 332
16.2 Profilometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
16.2.1 The Bearing Area Curve . . . . . . . . . . . . . . . . . . . . . . . 334
16.2.2 The Contact Problem . . . . . . . . . . . . . . . . . . . . . . . . . 336
16.3 Asperity Model Theories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337
16.3.1 The Exponential Distribution . . . . . . . . . . . . . . . . . . . 339
16.3.2 The Gaussian Distribution . . . . . . . . . . . . . . . . . . . . . . 340
16.3.3 The Plasticity Index . . . . . . . . . . . . . . . . . . . . . . . . . . 342
16.4 Statistical Models of Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . 343
16.4.1 Discrete Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
16.4.2 Random Process Models . . . . . . . . . . . . . . . . . . . . . . . 345
16.4.3 Determining Asperity Parameters . . . . . . . . . . . . . . . . 351
16.5 Fractal Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352
16.5.1 Archard’s Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352
16.5.2 Self-affine Fractals and the Fractal Dimension . . . . . . . 352
16.5.3 The Weierstrass Function . . . . . . . . . . . . . . . . . . . . . . 354
16.5.4 Generating Realizations of Fractal Profiles and
Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 356
16.6 Contact of Fractal Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
16.6.1 Majumdar and Bhushan’s Theory . . . . . . . . . . . . . . . . 359
16.6.2 Elastic Contact for a Fractal Surface . . . . . . . . . . . . . . 360
16.6.3 The Weierstrass Profile . . . . . . . . . . . . . . . . . . . . . . . . 362
16.6.4 Persson’s Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . 364
16.6.5 Implications for Coulomb’s Law of Friction . . . . . . . . 368
16.7 Adhesive Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369
16.7.1 Asperity Model Predictions . . . . . . . . . . . . . . . . . . . . . 370
16.7.2 The Sinusoidal Profile . . . . . . . . . . . . . . . . . . . . . . . . 371
16.7.3 Adhesion of Random Rough Surfaces . . . . . . . . . . . . . 374
xiv Contents
16.8 Incremental Stiffness and Contact Resistance . . . . . . . . . . . . . . 375

16.8.1 Asperity Model Predictions . . . . . . . . . . . . . . . . . . . . . 376
16.8.2 Clustering of Actual Contacts . . . . . . . . . . . . . . . . . . . 377
16.8.3 Bounds on Incremental Stiffness . . . . . . . . . . . . . . . . . 378
16.8.4 Persson’s Theory of Incremental Stiffness . . . . . . . . . . 380
16.8.5 Gaps and Fluid Leakage . . . . . . . . . . . . . . . . . . . . . . . 381
16.9 Finite-Size Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382
16.9.1 Integral Equation Formulation . . . . . . . . . . . . . . . . . . . 383
16.9.2 Unit Cells and the Constriction Alleviation Factor . . . . 386
16.9.3 Contact of Rough Spheres . . . . . . . . . . . . . . . . . . . . . 387
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 390
17 Thermoelastic Contact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395
17.1 Thermoelastic Deformation . . . . . . . . . . . . . . . . . . . . . . . . . . . 396
17.1.1 Fourier Transform Solutions . . . . . . . . . . . . . . . . . . . . 396
17.1.2 Steady-State Temperature . . . . . . . . . . . . . . . . . . . . . . 397
17.1.3 Thermoelastic Distortion Due to a Point Heat
Source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 398
17.1.4 Dundurs’ Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 399
17.1.5 Moving Heat Sources . . . . . . . . . . . . . . . . . . . . . . . . . 400
17.2 The Axisymmetric Thermoelastic Hertz Problem . . . . . . . . . . . 401
17.2.1 The Heat Conduction Problem . . . . . . . . . . . . . . . . . . 402
17.2.2 Thermoelastic Distortion . . . . . . . . . . . . . . . . . . . . . . . 403
17.2.3 Solution of the Contact Problem . . . . . . . . . . . . . . . . . 403
17.3 Existence and Uniqueness . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405
17.3.1 A One-Dimensional Model . . . . . . . . . . . . . . . . . . . . . 406
17.3.2 Effect of a Thermal Interface Resistance . . . . . . . . . . . 407
17.3.3 Imperfect Thermal Contact . . . . . . . . . . . . . . . . . . . . . 409
17.3.4 The Hertz Problem Revisited . . . . . . . . . . . . . . . . . . . 410
17.3.5 Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 410
17.3.6 Contact of Dissimilar Materials . . . . . . . . . . . . . . . . . . 413
17.3.7 Two-Dimensional Stability Problems . . . . . . . . . . . . . . 413
17.4 Solidification Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415
17.5 Frictional Heating . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 417
17.5.1 The Rod Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419
17.5.2 Burton’s Stability Analysis . . . . . . . . . . . . . . . . . . . . . 420
17.5.3 Out-of-Plane Sliding . . . . . . . . . . . . . . . . . . . . . . . . . . 421
17.5.4 In-Plane Sliding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423
17.5.5 Limiting Configurations . . . . . . . . . . . . . . . . . . . . . . . 425
17.5.6 Effect of Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . 427
17.5.7 Numerical Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . 429
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 431
Contents xv
18 Rolling and Sliding Contact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433

18.1 Rigid-Body Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433
18.1.1 Three-Dimensional Motions . . . . . . . . . . . . . . . . . . . . 435
18.2 Johnson’s Belt Drive Problem . . . . . . . . . . . . . . . . . . . . . . . . . 438
18.3 Tractive Rolling of Elastic Cylinders . . . . . . . . . . . . . . . . . . . . 441
18.3.1 Dissimilar Materials . . . . . . . . . . . . . . . . . . . . . . . . . . 445
18.3.2 Antiplane Loading . . . . . . . . . . . . . . . . . . . . . . . . . . . 446
18.3.3 Rolling of Misaligned Cylinders . . . . . . . . . . . . . . . . . 446
18.3.4 Three-Dimensional Rolling Contact Problems . . . . . . . 447
18.3.5 Kalker’s Strip Theory . . . . . . . . . . . . . . . . . . . . . . . . . 448
18.3.6 The Incipient Sliding Solution . . . . . . . . . . . . . . . . . . . 450
18.3.7 Transient Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . 450
18.3.8 Rail Corrugations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 451
18.4 Steady Sliding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 452
18.5 Wear . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 455
18.5.1 Archard’s Wear Law . . . . . . . . . . . . . . . . . . . . . . . . . 455
18.5.2 Long-Time Solution . . . . . . . . . . . . . . . . . . . . . . . . . . 456
18.5.3 Transient Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . 457
18.5.4 Galin’s Eigenfunction Method . . . . . . . . . . . . . . . . . . . 459
18.5.5 Non-conformal Contact Problems . . . . . . . . . . . . . . . . 461
18.6 Sliding of Rough Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . 462
18.6.1 Flash Temperatures . . . . . . . . . . . . . . . . . . . . . . . . . . 463
18.6.2 Bulk Temperatures . . . . . . . . . . . . . . . . . . . . . . . . . . . 468
18.6.3 Transient Asperity Interactions . . . . . . . . . . . . . . . . . . 469
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 470
19 Elastodynamic Contact Problems . . . . . . . . . . . . . . . . . . . . . . . . . . 475
19.1 Wave Speeds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 476
19.1.1 Rayleigh Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 477
19.2 Moving Contact Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 478
19.2.1 The Moving Line Force . . . . . . . . . . . . . . . . . . . . . . . 478
19.2.2 Integral Equation Formulation . . . . . . . . . . . . . . . . . . . 479
19.2.3 The Subsonic Problem . . . . . . . . . . . . . . . . . . . . . . . . 480
19.2.4 The Speed Range cR \V\c2 . . . . . . . . . . . . . . . . . . . 481
19.2.5 The Solution of Slepyan and Brun . . . . . . . . . . . . . . . 482
19.2.6 The Transonic Solution c2 \V\c1 . . . . . . . . . . . . . . . 484
19.2.7 The Superseismic Solution V [ c1 . . . . . . . . . . . . . . . 485
19.3 Interaction of a Bulk Wave with an Interface . . . . . . . . . . . . . . 490
19.3.1 SH-Waves Transmitted Across a Frictional
Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 490
19.3.2 In-Plane Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 496
xvi Contents
19.4 Interface Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 498

19.4.1 Slip Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 499
19.4.2 Slip Waves at a Sliding Interface . . . . . . . . . . . . . . . . 500
19.4.3 Slip–Stick Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . 501
19.5 Stability of Frictional Sliding . . . . . . . . . . . . . . . . . . . . . . . . . . 503
19.6 Transient Elastodynamic Contact Problems . . . . . . . . . . . . . . . . 504
19.6.1 Impulsive Line Force . . . . . . . . . . . . . . . . . . . . . . . . . 504
19.6.2 A Uniform Pressure Suddenly Applied . . . . . . . . . . . . 504
19.6.3 Integral Equation Formulation of the Transient
Contact Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 505
19.6.4 Normal Indentation by a Rigid Body . . . . . . . . . . . . . . 506
19.6.5 Superseismic Indentation . . . . . . . . . . . . . . . . . . . . . . 507
19.6.6 Self-Similar Indentation Problems . . . . . . . . . . . . . . . . 508
19.6.7 Three-Dimensional Transient Problems . . . . . . . . . . . . 509
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 510
20 Impact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . 513
20.1 Hertz’ Theory of Impact . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . 514
20.1.1 Duration of the Impact . . . . . . . . . . .
. . . . . . . . . . . . . 515
20.1.2 Homogeneous Sphere . . . . . . . . . . . .
. . . . . . . . . . . . . 517
20.1.3 Range of Validity of the Theory . . . .
. . . . . . . . . . . . . 517
20.1.4 The Superseismic Phase . . . . . . . . . .
. . . . . . . . . . . . . 518
20.2 Impact of a Cylinder . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . 519
20.3 Oblique Impact . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . 521
20.3.1 The Equation of Motion . . . . . . . . . .
. . . . . . . . . . . . . 522
20.3.2 The Tangential Contact Problem . . . .
. . . . . . . . . . . . . 523
20.3.3 Complete Stick . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . 523
20.3.4 Gross Slip . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . 526
20.3.5 Partial Slip . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . 526
20.3.6 The Complete Trajectory . . . . . . . . .
. . . . . . . . . . . . . 527
20.3.7 Rebound Conditions . . . . . . . . . . . . .
. . . . . . . . . . . . . 528
20.4 One-Dimensional Bar Problems . . . . . . . . . . .
. . . . . . . . . . . . . 529
20.4.1 The Semi-infinite Bar . . . . . . . . . . . .
. . . . . . . . . . . . . 530
20.4.2 The Infinite Bar . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . 531
20.4.3 Reflections . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . 532
20.4.4 The Impact Problem . . . . . . . . . . . . .
. . . . . . . . . . . . . 533
20.4.5 A Rigid Mass Impacting an Elastic Bar . . . . . . . . . . . . 533
20.4.6 Frictional Problems . . . . . . . . . . . .... . . . . . . . . . . . 536
20.4.7 Continuous Frictional Supports . . . .... . . . . . . . . . . . 538
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... . . . . . . . . . . . 540
Contents xvii
Appendix A: Potential Function Solutions for Elasticity Problems . . . . . 543

Appendix B: Integrals over Elliptical Domains . . . . . . . . . . . . . . . . . . . . . 547
Appendix C: Cauchy Singular Integral Equations . . . . . . . . . . . . . . . . . . 555
Appendix D: Dundurs’ Bimaterial Constants . . . . . . . . . . . . . . . . . . . . . . 559
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 561
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 581
Chapter 1
Kinematics of Contact
Most of the forces applied to engineering components arise from contact with one
or more other components. The contact may be nominally static as in the case of
components that are bolted together or assembled by a shrink fit, or it may involve
relative motion, as in the case of meshing gear teeth or the contact between a brake
pad and disc. In both cases, the contact can give rise to significantly enhanced local
stresses, implying the possibility of local material failure. Furthermore, the trans-
mission of force across a contact is affected by the local microtopography and may
involve local sliding, with consequent energy dissipation and the generation of heat.
All of these topics lie in the general field of tribology, but the mathematical analysis
of the transmission of force between contacting surfaces is the domain of the more
specific subject of contact mechanics.
It is important to distinguish between conformal or complete contact, where the
undeformed bodies share a common shape and can therefore be placed in contact over
an extended region without requiring any material deformation, and non-conformal
or incomplete contact, where the initial contact is necessarily restricted to one or more
isolated points. For example, Fig. 1.1a shows the conformal contact of a sphere in a
spherical depression of the same radius, whereas Fig. 1.1b shows the non-conformal
contact of the same sphere resting on a plane surface.
Notice that in conformal contact, the contact area is generally determined by the
shape of the contacting bodies and is therefore known a priori. By contrast, in non-
conformal contact, the contact area depends on the deformation of the bodies and
must be found as part of the solution. It generally increases as the load is increased.
© Springer International Publishing AG 2018 1

J.R. Barber, Contact Mechanics, Solid Mechanics and Its Applications 250,
https://doi.org/10.1007/978-3-319-70939-0_1
2 1 Kinematics of Contact
(a) (b)
Fig. 1.1 (a) Conformal and (b) non-conformal contact
1.1 Reference Frame and the Initial Gap Function
Consider the non-conformal contact problem shown in Fig. 1.2, where two bodies
identified by subscripts 1 and 2, respectively, make initial contact at the point O. So
far, no forces have been applied and the contact is restricted to a single point.
If the profiles of the bodies are smooth—i.e. if there are no sharp corners—we
can identify a common tangent plane at the point of contact and a common normal.
Points in the tangent plane will be defined by Cartesian coordinates (x, y) and we
choose the z-axis to point into body 1 as shown.
The undeformed profiles of the bodies can then be defined by the functions
g1 (x, y), g2 (x, y) which represent the gap that would exist if the body in ques-
tion were placed in contact with a plane surface. In many cases, we shall find that
the contact problem depends on the geometry of the two bodies only through the
composite initial gap function
g0 (x, y) = g1 (x, y) + g2 (x, y), (1.1)
which is the gap between the bodies in the undeformed configuration illustrated,
measured perpendicular to the tangent plane. We can then define a simpler equivalent
contact problem in which a body with a profile defined by g0 (x, y) makes contact
with a plane surface.
Fig. 1.2 Coordinate system 2.

for non-conformal contact
g2
O common
g1 tangent plane
common
z normal 1.
1.2 Establishment of a Contact Region 3
1.2 Establishment of a Contact Region
Suppose we now push the two bodies in Fig. 1.2 together by a normal force P. If
they really were rigid, this would imply a concentrated reaction force between the
bodies at O and hence an infinite contact pressure [force per unit area]. The bodies
will therefore inevitably deform slightly, establishing a finite contact region A, as
shown in Fig. 1.3.
In order to formulate the contact problem, we shall consider approaching this
condition by two steps. In the first step, we move the upper body downwards—i.e.
along the common normal—by some rigid-body displacement Δ. This will sim-
ply reduce the gap g0 (x, y) everywhere by Δ. We then superpose the deformation
of the bodies, which we characterize through the surface displacements u(1) (x, y),
u(2) (x, y). A vertically downward displacement u (1)
z (x, y) of the lower body 1 will
tend to increase the gap, whereas a downward displacement u (2) z (x, y) of the upper
body 2 will decrease it.
This superposition is illustrated schematically in Fig. 1.4 for a representative pair
of points P1 , P2 at (x, y) on the surfaces of the undeformed bodies. We conclude
that the final gap g(x, y) is given by the expression
g(x, y) = g0 (x, y) − Δ + u (1) (2)

z (x, y) − u z (x, y). (1.2)
Fig. 1.3 Establishment of a P

contact region A by local
deformation 2.
1.
P
Fig. 1.4 Effect of

P2
rigid-body displacement Δ
and surface displacements Δ
u (1) (2)
z (x, y), u z (x, y) on the
g 0 (x,y) u(2)
z (x,y)
final gap function
P1 (1) g(x,y)
u z (x,y)
1.2.1 Definition of Contact
We assume that the bodies are not allowed to interpenetrate each other, so the gap
cannot be negative. In fact, we can define the contact area A as the region in which
the final gap g(x, y) = 0. The rest of the interface then constitutes the separation
region Ā, where g(x, y) > 0.
For simplicity, suppose that the contact is frictionless so that there is a purely
normal traction [the contact pressure p(x, y)] transmitted between the bodies in A.
Generally, we assume that this traction cannot be tensile. In other words, we can
push against a frictionless surface, but we cannot ‘pull’ against it, so p(x, y) ≥ 0. We
also assume that a non-zero contact pressure can occur only in regions of contact. In
other words, if the gap g(x, y) > 0, the corresponding surfaces are locally traction-
free. Both of these assumptions must be relaxed at very small [e.g. nanometre] length
scales, where van der Waals attractive forces can be significant, but we shall postpone
discussion of this effect until Chap. 12.
With these definitions, we can establish a formal statement of the frictionless
contact problem through the conditions
g(x, y) = 0 (x, y) ∈ A
> 0 (x, y) ∈ Ā
(1.3)
p(x, y) = 0 (x, y) ∈ Ā
> 0 (x, y) ∈ A.
Notice that in each region we have one equation and one inequality. Also, the prod-
uct p(x, y)g(x, y) = 0 for all x, y. Problems of this kind for two non-negative com-
plementary variables whose product is everywhere zero are referred to as linear
complementarity problems (LCP).
1.2.2 The Boundary Value Problem
Problems in solid mechanics are usually well posed if either the traction or the
displacement is prescribed at each point on the boundary. This is the theoretical
counterpart of the fact that in an experiment, we are free to apply a prescribed
traction at any point on the surface of a deformable body, or to impose a prescribed

displacement, but not both.
For the special case where the materials are linear elastic, it can be shown that the
problem then has one and only one solution. The contact condition (1.3)1 together
with (1.2) requires that
u (1) (2)
z (x, y) − u z (x, y) = Δ − g0 (x, y) (x, y) ∈ A, (1.4)
and it follows that the equality conditions in (1.3) define a well-posed problem,
provided that A is known and the rigid-body approach Δ is prescribed.
1.2.3 Signorini Problems
The inequalities in (1.3) serve to determine the extent of the contact area A. In fact,
we might envisage a numerical procedure whereby we guess the value of A, solve
the resulting equations to determine p(x, y), g(x, y) and then check to see whether
the inequalities are satisfied. For any point where the appropriate inequality is not
satisfied, we could change the assumption—for example, take that point outside
the contact area if the contact pressure is negative—and re-solve, continuing the
iteration until the process converges on a value of A at which the inequalities are
satisfied everywhere. Some numerical solutions of the contact problem are based on
an iterative strategy of this kind.
The inequalities make the problem nonlinear, and there is no guarantee that the
iterative strategy suggested above [for example] would converge onto a solution or
that the solution once obtained would be unique. Frictionless linear elastic problems
of this form are generally known as Signorini problems, since Antonio Signorini was
the first to identify them as an extension of the class of conventional boundary value
problems. A formal proof of existence and uniqueness of solution for the Signorini
problem for an arbitrary geometry was given by Fichera (1964).
If the contact problem is conformal, the contact area A is generally known a
priori, but the inequalities should still be checked since there exist systems in which
separation occurs in regions that would have been in contact before the forces were
applied. This situation is known as receding contact [see Chap. 11].
1.2.4 Asymptotic Arguments
For linear elastic problems, the contact pressure p must tend to zero at the boundary of
the contact area as long as the surfaces of the bodies are continuous up to and including
the first derivative. The proof of this result depends on an asymptotic analysis of the
stress and displacement fields in the immediate vicinity of this boundary. Imagine that
we train a very powerful microscope on a point on the boundary, the magnification
(a)
(b)
1/2
p~r deformed surface p~r
1/2
3/2
1/2 δ∼r
uθ ~ r
deformed surface
Fig. 1.5 Deformation due to (a) a square-root singular pressure distribution, and (b) a square-root
bounded distribution
being so large that the rest of the features of the geometry are far beyond the field of
view and the local radius of curvature of the boundary appears so large as to make it
effectively straight. We shall show in Chap. 10 that the local elastic field can then be
expanded in an eigenfunction series, the terms of which involve increasing powers
of the distance r from the boundary of the contact area, and for points sufficiently
near to the boundary, only the first term in this series is significant.
If we apply this method to the equality boundary conditions in (1.3), we find that
the contact pressure is generally square-root singular at the boundary of the contact
area, implying a local deformation of the form shown in Fig. 1.5a. If the indenting
body has a smooth profile, as shown by the dotted curve, this would clearly imply
interpenetration of material, so we must conclude that the multiplier on the leading
singular term must be zero. The second term in the eigenfunction series will then be
dominant near the contact boundary, and this involves pressure going to zero with
a positive square-root form, as shown in Fig. 1.5b. We shall see numerous examples
of this behaviour in subsequent chapters.
This result suggests an alternative formulation of the non-conformal contact prob-
lem (1.3)—we enforce the two equality conditions as before, but we replace the
inequalities by the requirement that the contact pressure goes to zero at the boundary
of A. This method is very convenient in cases where the prospective contact area can
be characterized by a limited number of parameters. For example, in axisymmetric
problems, the contact pressure will be a function of radius r only and we can often
p(x,y)
B
deformed
A g(x,y) surface
Fig. 1.6 The contact pressure distribution p(x, y) satisfies the contact conditions at A, including
the boundedness condition, but the separation inequality is violated in a detached region near B
assume that the contact area will be a circle of radius a. The above condition then
reduces to p(a) = 0, which provides an equation for the unknown a.
Equations are much easier to manipulate than inequalities, so this approach is
widely used in the analytical solution of non-conformal contact problems. How-
ever, it is strictly equivalent to the physical requirements defined by the inequalities
only in cases where the resulting contact area is connected. If the physical problem
involves two or more unconnected contact regions, solutions satisfying the bounded-
ness condition may be possible in which only one of these regions is in contact and
the separation inequality (1.3)2 is violated near the other(s), as illustrated in Fig. 1.6.
This difficulty is easily overcome in particular problems, since once a solution
has been obtained using the equality condition [ p(x, y) → 0], we can then use the
results to check the inequalities. If tensile tractions are found at any interior point
of the contact area, it suggests an alternative solution where the contact area is
multiply connected, whereas if interpenetration (negative gap) is detected in the
separation region, it suggests the presence of an additional unconnected contact
area. The solution can in principle then be repeated using the equality condition with
this new topological partition of the potential contact region. More generally, the use
of the equality condition will define at most a finite number of distinct solutions to
the contact problem, only one of which will satisfy the strict inequality conditions
throughout the interface.
1.2.5 The Discrete Problem
Most practical problems involve the contact of deformable bodies that can be regarded
as continua for the purpose of analysis. Thus, the conditions in (1.3) are imposed at
a real infinity of points (x, y). However, the number of cases that can be treated by
analytical methods is limited and often we shall have recourse to numerical methods,
notably the finite element or boundary element methods. A finite element formulation
of a contact problem essentially reduces the contact interface A to a finite set of
discrete nodes, so that (1.3) constitutes a finite set of algebraic equations.
Mathematically, the discrete problem is very different from the continuum prob-
lem, though the value of the finite element method depends on the assumption that,
with appropriate mesh refinement, the predicted physical relations will approach
those of the (probably intractable) continuum problem in some asymptotic sense. In
particular, different techniques are needed for the proof of general theorems for the
two classes of problem.
The discrete problem is formally equivalent to a contact problem involving a set
of point masses connected by generalized springs that can make contact with one or
more rigid obstacles, and models of this kind can be very useful as aids to thinking
about the problem. For example, in the discrete version of (1.3), it is clear that the
number of unknowns is twice the number of nodes since there is an unknown contact
force and displacement at each node. However, the displacement at each node is
determined by the deformation of the body and hence must be a function of the
applied nodal forces. This furnishes another equal set of equations1 and ensures that
for a given set A, the number of equations is equal to the number of unknowns,
showing that the problem is well-posed.
1.3 Nonlinear Kinematics
Equations (1.2), (1.4) and the associated discussion are based on the assumption
of linear kinematics. In other words, strains, including rotations, are assumed to be
small compared with unity, so that a first-order approximation decouples orthogonal
components of displacement. This permits the deformation to be referred to the
undeformed configuration of the bodies and forms the basis of linear elasticity.
A simple example that illustrates this approximation is the indentation of an elastic
half-space by a frictionless rigid wedge, as shown in Fig. 1.7. Equation (1.4) states
that the elastic displacement at point A [originally at (x, 0)] is given by
u z (x, 0) = Δ − x tan α. (1.5)
However, the point A will generally have both a horizontal and vertical displacement,
u x , u z , respectively, and for it to remain on the surface of the wedge we actually
require
u z (x, 0) = Δ − [x + u x (x, 0)] tan α or u z (x, 0) + u x (x, 0) tan α = Δ − x tan α.

(1.6)
Now, the originally horizontal surface of the half-plane is rotated through the angle
α in the contact region, so if strains and rotations are to be mathematically small,
rigid uz
A
x ux
Δ
α
elastic
Fig. 1.7 Indentation of a half-plane by a rigid wedge. The horizontal dashed line indicates the
undeformed location of the surface
1 If
one of the bodies has one or more rigid-body degrees of freedom, there will be a deficit in this
second set of equations, which however is made up by the requirement that the applied forces satisfy
corresponding equilibrium conditions.
1.3 Nonlinear Kinematics 9
the term u x (x, 0) tan α in this equation is a second-order small quantity and should
therefore strictly be neglected in a linear analysis.
If tan α is relatively small, but not negligible, it is tempting to include the second-
order term, and indeed Hay et al. (1999) have shown that this leads to a solution
that is significantly closer to the results of a fully nonlinear finite element analysis.
However, once such second-order terms are included, we logically ought also to
consider the fact that the loaded surface is now inclined to the horizontal, so that
[for example] the frictionless condition seems to require that the traction component
parallel to this surface be zero. By contrast, if we refer the solution to the undeformed
configuration, we would instead require the horizontal traction σzx (x, 0) to be zero.
In fact, as in many branches of mechanics, once we pass beyond first-order terms in
small quantities, the development of a consistent second-order formulation is very
challenging, and generally it is preferable to have recourse immediately to a fully
nonlinear statement of the problem. However, this will almost always imply the use
of numerical methods which are beyond the scope of this book.
1.4 Almost Conformal Contact
The kinematic description of Sect. 1.1 and Fig. 1.2 is appropriate only as long as the
slope of the undeformed bodies is small in the contact region—i.e.
2 2
∂gi ∂gi
+ 1 (x, y) ∈ A; i = 1, 2. (1.7)
∂x ∂y
In many cases, this is also a necessary condition for the strains to be small. For
example, in the contact of a sphere and a plane, small strain theory is applicable only
if the radius of the contact area is small compared with the radius of the sphere, in
which case (1.7) is satisfied as well. However, even the small strain problem requires
(a) (b)
2.
n(θ)
g0 (θ)
P θ
O
θ O
1.
Fig. 1.8 (a) An almost conformal contact problem, (b) definition of the initial gap g0
a more general kinematic formulation if the contact is ‘almost conformal’ and the
slope of the surfaces varies significantly in the prospective contact area.
Figure 1.8a shows the unloaded configuration of a cylindrical body contacting
the surface of a slightly larger cylindrical hole in the second body. Initial contact
occurs at O, but it is clearly inappropriate to measure the initial gap in the direction
perpendicular to the common tangent at O, particularly for points near θ = π/2.
Instead, we could define the gap as the distance between the surfaces in the
direction of the normal n(θ) to the undeformed surface of body 1, as shown in
Fig. 1.8b. Notice that n will then not generally be exactly normal to the surface at Q
and hence the choice of body 1 as a reference is essentially arbitrary.2
More generally, we identify a representative point P1 (r) on the surface of body 1
by a position vector r and define the outward normal to the surface at that point by
the unit vector n(r). This allows us to determine the point P2 (r) at which the normal
at P1 intersects the surface of body 2. The initial gap g0 (r) is then defined as P1 P2 .
Suppose now that forces are applied to the bodies such as to cause displacements
u1 (r), u2 (r) on the surfaces of bodies 1 and 2, respectively. We also allow the possi-
bility of a rigid-body translation3 U of body 2 relative to body 1. The gap g projected
along the original normal is then given by

g(r) = g0 (r) + U · n + u2 (r) − u1 (r) · n. (1.8)
If the contact pressure p(r) is defined such that the traction t(r) on body 1 is
t(r) = − p(r)n(r), (1.9)
the Signorini problem can be defined by Eq. (1.3) with r replacing (x, y).
This example also illustrates the fact that (1.3) provides a fairly general state-
ment of the frictionless contact problem, provided appropriate interpretations can be
provided for the gap g and the contact pressure p.
Problems
1.Determine the gap function g0 (x, y) for the problem in which body 1 is a cylinder
of radius R with its axis aligned with the y-axis, and body 2 is a similar cylinder
whose axis bisects the angle between the x- and y-axes. Assume x, y R.
2. In a discrete formulation of a frictionless contact problem, the normal forces Pi
at a set of nodes are related to the corresponding normal displacements u j in body 1
by the linear matrix equation
2 Ifthe strains are small, it can be shown that the effect of changing the reference to the normal to
the surface of body 2 is negligible.
3 In some cases, a relative rigid-body rotation must also be included.
Problems 11

N
P = Ku or Pi = Ki j u j ,
j=1
where K is the contact stiffness matrix and can be assumed to be symmetric and
positive definite.
Consider the case where body 2 is rigid and there are only two nodes (N = 2). It
is proposed to solve the discrete version of the contact problem (1.3) by (i) assuming
both nodes are in contact, (ii) solving for P1 , P2 and (iii) changing the assumption to
separation at any node for which Pi < 0. Find the condition(s) that must be satisfied
by K if the gap at any node so released is to be positive. Can you extend this argument
to the case where N > 2?
3. Two bodies with sinusoidal surfaces are defined by the gap functions
g1 (x) = h 0 [1 + cos(mx)] ; g2 (x) = −h 0 (1 − )[1 + cos(mx)],
where 1 but mh 0 is O(1), so the mean slope is not small compared with unity.
Determine the direction of the local normal to the surface of body 1 as a function of
x and hence use the protocol of Sect. 1.4 to determine the initial gap g0 again as a
function of x.
Chapter 2
Three-Dimensional Frictionless Elastic
Problems
To complete the formulation of the problem of Eqs. (1.2)–(1.4), we need to relate the
contact pressure p(x, y) to the normal surface displacements u (1) (2)
z (x, y), u z (x, y) by
analyzing the deformation of the contacting bodies. In this chapter we shall assume
that the materials are linear elastic, which implies that
(i) the strains are everywhere small in comparison with unity;

(ii) superposition holds;
(iii) An Eulerian kinematics holds — i.e. the deformations can be referred to the
original undeformed configuration.
2.1 The Half-Space Approximation
We have already remarked that the contact area is generally small and hence that the
strains due to contact forces are concentrated in a small region. It follows that the
exact geometry of the bodies a long way away from the contact region is relatively
unimportant, since these regions experience at most a rigid-body motion. We can
therefore simplify the problem considerably by assuming that the body extends to
infinity.
In most cases, we can also simplify the elasticity problem by assuming that the
deformations due to a given traction distribution are the same as those that would be
produced in an equivalent body with a plane surface. For example, we assume that
the surface displacement u z due to the pressure distribution p(x, y) in Fig. 2.1a is
the same as that produced by the same distribution in Fig. 2.1b, where the loading is
applied to a half-space — i.e. a semi-infinite body bounded by a plane.

https://doi.org/10.1007/978-3-319-70939-0_2
14 2 Three-Dimensional Frictionless Elastic Problems
(a) (b)
p(x,y) p(x,y)
uz uz
Fig. 2.1 (a) Contact pressure acting on the body and (b) the equivalent half-space problem
Notice that this approximation will fail if there are any sharp corners or rapidly
varying slopes [small radii of curvature] on the surface near the contact area.1
2.2 Normal Loading of the Half-Space
If an isotropic half-space z > 0 is loaded by purely normal tractions, so that the shear
tractions
σzx (x, y, 0) = σzy (x, y, 0) = 0 (2.1)
for all x, y, the elastic stress field can conveniently be expressed in terms of a single
potential function ϕ, where
∂2ϕ ∂2ϕ ∂2ϕ

∇2ϕ ≡ + + 2 = 0, (2.2)
∂x 2 ∂ y2 ∂z
(Green and Zerna 1954, Sect. 5.7; Barber 2010, pp. 339–341). The stress and dis-
placement components for this solution in Cartesian coordinates (x, y, z) and cylin-
drical polar coordinates (r, θ, z) are given in Appendix A, Sect. A.1. In particular,
the normal traction and the normal displacement at the surface z = 0 are given by the
simpler expressions
∂2ϕ (1 − ν) ∂ϕ
σzz (x, y, 0) = − ; u z (x, y, 0) = − , (2.3)
∂z 2 G ∂z
where G, ν are, respectively, the shear modulus [modulus of rigidity] and Poisson’s
ratio for the material.
Equation (2.2) is known as Laplace’s equation and its solutions are described as
harmonic functions. We shall show in Chap. 4 that this representation of the elastic
fields permits us to establish mathematical analogies between frictionless elastic
contact problems and other physical processes leading to harmonic boundary-value
problems, such as electrostatics and the conduction of heat or electricity.
1 Except when the body in question is rigid, so that no elastic solution is required.
2.2 Normal Loading of the Half-Space 15
2.2.1 The Point Force Solution
The fundamental problem for the frictionless half-space is that in which the only
loading comprises a concentrated normal force acting at a point which we can take
as the origin [see Fig. 2.2]. If the material is isotropic, the stress and displacement
fields will be axisymmetric and hence functions of r and z only. In particular, we wish
to determine the normal surface displacement u z (r, 0) as a function of the distance
r from the point of application of the force.
The complete stress and displacement field for this problem can be obtained by
substituting the harmonic function
P
ln (R + z)
ϕ=− (2.4)
2π
√
into Eqs. (A.2), (A.3) of Appendix A, where R = r 2 +z 2 is the distance from the
origin. In particular, we obtain
∂3ϕ ∂2ϕ 3Pz 3

σzz = z − 2 =− , (2.5)
∂z 3 ∂z 2π R 5
which is zero at all points on the surface z = 0 except at the origin, where there
is a singularity. To verify that this corresponds to a concentrated normal force, we
consider the equilibrium of the layer 0 < z < h shown in Fig. 2.3.
The radial tractions σzr (r, θ, h) on the bottom surface z = h are axisymmetric
and hence self-equilibrating, but the normal tractions sum to a resultant force in the
z-direction equal to
∞ ∞
3Ph 3r dr
2π σzz (r, θ, h)r dr = − = −P, (2.6)
0 0 (r 2 + h 2 )5/2
which is independent of h. Thus to maintain equilibrium of the layer, there must be

a concentrated force P in the z-direction at the origin, as shown.
Fig. 2.2 Concentrated

normal force P applied to
P
the surface of the half-space r
z uz
R
Fig. 2.3 Equilibrium of the P

layer 0 < z < h
h r
σ zr
σ zz
Using the potential (2.4) and Eq. (2.3)2 , we can then obtain the normal surface
displacement as
P(1 − ν) P(1 − ν 2 )
u z (r, θ, 0) = = , (2.7)
2πGr π Er
where E = 2G(1+ν) is Young’s modulus.
2.2.2 Similarity, Equilibrium and Anisotropy
Notice that the surface displacement u z (r, θ, 0) decays inversely with distance r
from the point of application of the force. We could have reached this conclusion
by noting that the problem has no intrinsic length scale in the sense that an enlarged
view of the body including the applied loading is essentially identical to the original
figure. This implies that the solution of this ‘enlarged’ problem must be capable of
a linear mapping into the original problem, and hence that the elastic fields must
be self-similar, meaning that the stress and displacement contours must all have the
same shape. This, in turn, implies that the corresponding expressions for stress σi j
and displacement u i must take the separated variable form
σi j = f (R)Si j (θ, β); u i = g(R)Ui (θ, β), (2.8)
where σi j , Ui are functions only of the polar angles θ, β in spherical polar coordinates
(R, θ, β) centred on the point of application of the force. Equilibrium considerations
then dictate that f (R) be proportional to R −2 , since the force P must be transmitted
across a sequence of self-similar surfaces whose surface areas are proportional to
R 2 . Since the deformation is linear elastic, the strains must also decay with R −2 and
since strains are displacement derivatives, the displacements must decay with R −1 .
These arguments apply equally to the case where the material is anisotropic,
though in this case we would expect the surface displacement due to a point force to
depend on the angle θ in the form
Ph(θ)
u z (r, θ, 0) = . (2.9)
r
2.2 Normal Loading of the Half-Space 17
Fig. 2.4 If a normal force is C(a, θ)

applied at O, the normal
surface displacement at C
must be equal to that at D, a
even if the material is
generally anisotropic
O θ
D(a, θ + π)
The function h(θ) in Eq. (2.9) depends upon the elasticity tensor ci jkl for the mate-
rial (Willis 1967), which in the most general case involves 21 independent elastic
constants.
Figure 2.4 defines three equally spaced points D, O, C in a plan view of the
surface. Maxwell’s reciprocal theorem (Barber 2010, Sect. 34.1) requires that the
normal displacement at C(a, θ) due to a normal force P applied at O be equal to
the normal displacement at O due to an equal normal force P applied at C. But
since the material is homogeneous, this is also equal to the normal displacement at
D(a, θ+π) due to a normal force P applied at O. It follows that
h(θ + π) = h(θ), (2.10)
in Eq. (2.9) and hence that for the most general anisotropic material, h(θ) is capable
of the Fourier expansion
∞
∞

h(θ) = Am cos(2mθ) + Bm sin(2mθ), (2.11)
m=0 m=1
involving even multiples of θ only, where Am , Bm are constants.
2.2.3 The Composite Elastic Modulus
The kinematic contact conditions (1.2), (1.4) involve only the relative displacement
u z ≡ u (1) (2)
z − uz , (2.12)
and incremental forces associated with the contact pressure p(x, y) will tend to cause
positive displacements in body 1 and negative displacements in body 2, as shown in
Fig. 2.5.
Fig. 2.5 Surface

(1) (2)
displacements u z , u z due -uz(2)
to equal and opposite forces
P on the two bodies P
u (1)
z
The elasticity problems for the two bodies are clearly similar, so we obtain
Ph (1) (θ) Ph (2) (θ)

uz = + . (2.13)
r r
Except where otherwise stated, we shall assume in this book that the materials are
isotropic, in which case (2.13) reduces to
P
uz = , (2.14)
π E ∗r
∗
where the composite modulus E is defined by
1 (1 − ν12 ) (1 − ν22 )
∗ = + . (2.15)
E E1 E2
This is the only point at which the modulus of the materials enters into the calculation,
and hence the separate properties only influence the contact problem through their
∗
contribution to E . In particular, for any given frictionless contact problem, we can
always define an equivalent problem in which one of the two bodies is taken to be
∗
rigid (e.g. E 2 → ∞), in which case E is the ‘plane strain’ modulus of the remaining
fictitious deformable material, and u z is its normal surface displacement. We shall
generally make use of this simplification wherever possible.
2.3 Integral Equation Formulation
Since the elastic problem is linear, additional solutions can be obtained by superposi-
tion. For example, if a set of forces Pi , i = (1, N ) act at the points (xi , yi ), respectively,
the surface displacement is given by
1
N
Pi
u z (x, y) = ∗ . (2.16)
π E i=1 (x − xi )2 + (y − yi )2
2.3 Integral Equation Formulation 19
This procedure can be generalized to a continuous distribution of contact pressure

p(x, y) in a contact area (x, y) ∈ A. The force acting over the infinitesimal rectangle
ξ < x < ξ +δξ, η < y < η+δη is approximately equal to p(ξ, η) δξ δη. If we regard
this in the limit as a point force and use Eq. (2.14) to define the resulting surface dis-
placements, we can sum the effect of all such rectangles by integration [convolution],
obtaining
1 p(ξ, η)dξdη
u z (x, y) = , (2.17)
πE∗ A r
where the distance r from (ξ, η) to (x, y) is given by

r= (x − ξ)2 + (y − η)2 . (2.18)
In effect, Eq. (2.14) serves as the Green’s function for the more general problem
involving a distributed contact pressure. Using (1.2), (2.17), we can then express the
linear complementarity problem (1.3) in the integral equation form

1 p(ξ, η)dξdη
= Δ − g0 (x, y) (x, y) ∈ A
πE∗ A (x − ξ)2 + (y − η)2
> Δ − g0 (x, y) (x, y) ∈
/ A (2.19)
p(x, y) = 0 (x, y) ∈ A
>0 (x, y) ∈
/ A.
Notice that we have one equality condition for each point of the domain A+ Ā, but an
alternative way of looking at the problem is to consider the function p(ξ, η) in (2.19)
to be defined [as non-zero] only in A. In this case, the integral equation provides a
condition for determining this function at every point in A and this is reflected in the
fact that A is both the domain of integration and the domain in which the integral
equation is to be satisfied. This equality of domains is an essential requirement for
the solution of integral equations of this kind. In a crude, numerical sense, one might
envisage discretizing the integral, using a set of N nodes in A, in which case (2.19)
would reduce to a set of N algebraic equations for the N nodal values of p(ξ, η). In
fact, this is a simple and widely used method for obtaining numerical solutions to
contact problems.
In many contact problems, the total contact force

P= p(x, y)d xd y (2.20)
A
is prescribed and the rigid-body displacement Δ is unknown. In effect, we then have

one extra unknown and one extra equation.
Fig. 2.6 Field-point

integration S 2 dθ
r dθ dr
r
P θ
S1
2.3.1 Field-Point Integration
Evaluation of the integral (2.19) is complicated by the presence of the square root
term representing the distance r . A more convenient integral form can be obtained
by choosing the field point P(x, y) as the origin of a system of polar coordinates
(r, θ) as shown in Fig. 2.6.
The element of area is then dA =r dθdr and the factor of r cancels in Eq. (2.17),
leaving
π S2
1
u z (x, y) = p(r, θ) dr dθ, (2.21)
π E ∗ 0 S1
where the points S1 , S2 define the intersections between a line through P of inclina-
tion θ and the boundary of the contact area.
2.3.2 Indentation by a Flat Elliptical Punch
The integration in Eq. (2.21) is particularly straightforward if the contact area is

defined by the ellipse of Fig. 2.7 and the pressure distribution is
−1/2
x2 y2
p(x, y) = p0 1 − 2 − 2 , (2.22)
a b
corresponding to a total force
a b√1−x 2 /a 2
P= √ p(x, y)d yd x = 2π p0 ab. (2.23)
−a −b 1−x 2 /a 2
If the points P, Q in Fig. 2.7 have coordinates (x, y) and (ξ, η) respectively, we
have
ξ = x + r cos θ; η = y + r sin θ, (2.24)
Fig. 2.7 Elliptical contact

y,η
area
S2
Q
b r
P
θ
O a x,ξ
S1
so
ξ2 η2
1− 2
− 2 = C0 − C1 (θ)r − C2 (θ)r 2 , (2.25)
a b
where
x2 y2 x cos θ y sin θ
C0 = 1 − 2 − 2 ; C1 (θ) = 2 + (2.26)
a b a2 b2
cos2 θ sin2 θ (1 − e2 cos2 θ)

C2 (θ) = 2
+ 2
= , (2.27)
a b b2
and the eccentricity
b2
e= 1− . (2.28)
a2
Using this notation, the inner integral in (2.21) reduces to the form
S2 S2
p0 dr
p(r, θ)dr = .
S1 S1 C0 − C1 (θ)r − C2 (θ)r 2
If we then make the change of variable

2
C1 C0 C2 1 x sin θ − y cos θ
r =t− ; D2 = + 12 = − , (2.29)
2C2 C2 4C2 C2 abC2
we obtain
S2 D
p0 dt π p0
p(r, θ)dr = =√ (2.30)
S1 −D C2 (D − t )
2 2 C2
and hence, using this result and (2.27), (2.23) in (2.21),

π
p0 b dθ 2 p0 bK (e) P K (e)
u z (x, y) = ∗ √ = ∗ = , (2.31)
E 0 1− e2 cos2 θ E π E ∗a
where π/2
dθ
K (e) = √ (2.32)
0 1 − e2 cos2 θ
is the complete elliptic integral of the first kind.2

We note that this expression is independent of x, y and hence of the position of
the field point P, provided only that it lies within the ellipse of Fig. 2.7. It follows
that the contact pressure distribution under a flat rigid punch of elliptical planform is
−1/2
P x2 y2
p(x, y) = 1− 2 − 2 , (2.33)
2πab a b
and the indentation depth is

P K (e)
Δ= , (2.34)
π E ∗a
where P is the total force applied to the punch.
An interesting feature of this argument is that it can be extended to the case where
the indented body is anisotropic. The Green’s function now depends on the angle θ
as in Eq. (2.9), but functions of θ appear only in the outer integral in (2.21) at which
point dependence on the coordinates x, y has already been lost. We conclude that
Eq. (2.33) also defines the contact pressure distribution under a flat elliptical punch
indenting a generally anisotropic elastic half-space. The corresponding indentation
depth is given in Appendix B, Eq. (B.24).
Displacements outside the ellipse
The same method can be used to determine the displacement at a point P outside the
ellipse. We obtain
β
p0 b dθ p0 b
u z (x, y) = ∗ √ = ∗ [F(β, e) − F(α, e)] , (2.35)
E α 1 − e cos θ
2 2 E
where ψ
dθ
F(ψ, e) = √ (2.36)
0 1 − e2 cos2 θ
is the incomplete elliptic integral of the first kind, and the angles α, β are defined in
Fig. 2.8a.
To find these angles, we perform a linear transformation on the x-axis only, so as to
reduce the contact area to a circle of radius b, as shown in Fig. 2.8b. The transformed
angles are obtained as
α=
=
θ − φ; β θ + φ, (2.37)
2 Some authors define elliptic integrals in terms of the parameter m = e2

rather than e, and this nota-
tion is also used in Mathematica, so care must be taken when using these equations in conjunction
with results from other sources.
(a) (b)
α α
P(x, y) bx
φ P ( a , y(
φ
b
β θ β
O a O
b
Fig. 2.8 Computation of the normal surface displacements outside the contact area. A linear trans-
formation maps the original geometry (a) into the circle (b)
where
y b x2 y2
sin
θ = ; sin φ = and ρ=b + 2 (2.38)
ρ ρ a 2 b
to the field point in Fig. 2.8b. The transformation then

represents the distance O P
implies that
a cot
α
a cot β
cot α = ; cot β = . (2.39)
b b
2.4 Galin’s Theorem
A more general result, related to that obtained in Sect. 2.3.2, concerns pressure dis-
tributions of the form
−1/2
x2 y2
p(x, y) = Pn (x, y) 1 − 2 − 2 , (2.40)
a b
where Pn (x, y) is a polynomial of degree n in x and y.

Galin (1961), (2008) has shown that if the pressure distribution in the elliptical
region is defined by (2.40), the corresponding surface displacement inside the ellipse
will be of the following form
u z = Q n (x, y), (2.41)
where Q n (x, y) is another polynomial of degree n in x and y. The results for the
flat punch in Sect. 2.3.2 are simply the special case of Galin’s theorem for n = 0, in
which case both P0 and Q 0 are constants.
It is interesting to note that the dichotomy between contact and noncontact can
be blurred in this formulation by defining p(x, y) as
−1/2
x2 y2
p(x, y) = Pn (x, y) 1 − 2 − 2 (2.42)
a b
throughout the surface of the half-space. The polynomial will be real for all x, y, but
outside the ellipse the argument of the square root is negative, so the real part of this
expression is zero as required. A related idea was exploited by Manners (1998) to
give some simple solutions to problems involving periodic surfaces.3
Galin’s original proof is rather long and applies only to isotropic materials. Willis
(1967) has shown that the theorem also applies to generally anisotropic materials
and he gives a more efficient proof using the Radon transform (Willis 1970). Here,
we shall prove a restricted form of the theorem using the method of integration
introduced in Sect. 2.3 and developed more fully in Appendix B.
2.4.1 A Special Case
Consider the case where the pressure distribution is of the special form
n−1/2
x2 y2
p(x, y) = p0 1 − 2 − 2 , (2.43)
a b
where n is an integer. The normal displacement due to this distribution is

π S2 n−1/2
p0 x2 y2
uz = 1− 2 − 2 dr dθ, (2.44)
πE∗ 0 S1 a b
and using the change of variables of Eqs. (2.24), (2.29) and (2.26), this simplifies to
π D
p0 n−1/2
u z (x, y) = C2 (D 2 − t 2 )n−1/2 dtdθ. (2.45)
πE∗ 0 −D
The inner integral can then be performed using the substitution t = D cos φ with the
result
(2n − 1)!! p0 π 2n n−1/2
u z (x, y) = D C2 dθ, (2.46)
(2n)!! E ∗ 0
where (2n)!! ≡ 2.4.6 . . . (2n) and (2n−1)!! ≡ 1.3.5 . . . (2n−1).

Since C2 is independent of x, y and D 2 is a quadratic function of x, y [see
Eq. (2.29)], we conclude that u z (x, y) will be a polynomial in x, y of degree 2n.
3 See Sect. 6.5.5 below.

2.4 Galin’s Theorem 25
This is a special case of Galin’s theorem, but the same changes of a variable can be
used to show that more general distributions of the form (2.40) also lead to polynomial
displacement distributions. Alternatively, more general solutions can be generated
by appropriate differentiations of (2.43) with respect to x, y, as in Sect. 4.3, below.
2.5 Interior Stress Fields
So far we have considered only the surface tractions and displacements, and in par-
ticular the solution of the mixed boundary-value problem (1.3). However, in practical
cases we are often concerned with the interior stress field, since this will determine the
conditions under which plastic deformation or other failure modes may be initiated.
Once the contact pressure is known throughout the surface, the interior field can
be written down as appropriate convolution integrals of the point force result of
Sect. 2.2.1. This can be done in terms of specific stress components. For example,
we know from Eq. (2.5) that, for the point force P,
3Pz 3
σzz = − where R= x 2 + y2 + z2. (2.47)
2π R 5
It follows by convolution (superposition) that the same stress component for a contact
pressure distribution p(x, y) is
∞ ∞
3 z 3 p(ξ, η)dξdη
σzz = − . (2.48)
2π −∞ −∞ [(x − ξ)2 + (y − η)2 + z 2 ]5/2
Similar expressions can be written for other stress or displacement components,

though the resulting integrals usually require numerical evaluation, except in a few
simple cases. Johnson (1985), Sect. 3.5 gives some analytical results for stress com-
ponents along the axes for the Hertzian distribution

x2 y2
p(x, y) = p0 1 − 2
− 2. (2.49)
a b
2.5.1 In-Plane Stress Components Near the Surface
For general frictionless contact problems, the internal stress components σx x , σ yy

can be expressed in terms of the potential function ϕ as
∂3ϕ ∂2ϕ ∂2ϕ ∂3ϕ ∂2ϕ ∂2ϕ

σx x = z + + 2ν 2 ; σ yy = z 2 + + 2ν 2 (2.50)
∂x ∂z
2 ∂x 2 ∂y ∂ y ∂z ∂y 2 ∂x
from Eq. (A.2), so

2
∂ ∂ ϕ ∂2ϕ ∂2ϕ ∂3ϕ
σx x + σ yy = 1 + 2ν + z + = −(1 + 2ν) − z ,
∂z ∂x 2 ∂ y2 ∂z 2 ∂z 3
(2.51)
since ϕ and hence its Cartesian derivatives are harmonic. It follows that at the surface
z = 0,
∂2ϕ
σx x + σ yy = −(1 + 2ν) = (1 + 2ν)σzz = −(1 + 2ν) p, (2.52)
∂z 2
where we have used (2.3)1 for σzz (x, y, 0). In other words, the sum of the principal
in-plane stresses at the surface is proportional to the local contact pressure.
In the two-dimensional case where the stress field is independent of y, we have
∂2ϕ ∂2ϕ ∂2ϕ

= 0; = − , (2.53)
∂ y2 ∂x 2 ∂z 2
and hence
σx x (x, 0) = − p(x); σ yy (x, 0) = −2ν p(x), (2.54)
so the in-plane stress σx x is everywhere equal to the applied (frictionless) traction,

and in particular, it is zero in regions of separation. Notice, however, that these results
can be modified by the application of a far-field ‘bulk’ stress tangential to the contact
interface. This effect is discussed in Sect. 9.4, below.
Problems
1. As a more rigorous alternative to Fig. 2.3, it is proposed to consider the equilibrium

of the finite disc 0 ≤r < a, 0 < z < h. The integral in Eq. (2.6) must now have a finite
upper limit r = a and the shear tractions acting on the curved surface r = a will make
an additional contribution to the equilibrium equation. Show that these effects are
self-cancelling, and also that they both tend to zero as a → ∞.
2. An elastic half-space is loaded by a uniform pressure p0 inside the square −a <
x < a, −a < y < a. Use Eq. (2.17) to determine the normal surface displacement at
the centre of the square [i.e. at the origin].
3. An elastic half-space is loaded by a uniform pressure p0 inside the circle 0 ≤r < a.
Use the field-point integration method of Sect. 2.3.1 to determine the normal surface
displacement as a function of r for points inside the circle.
4. The flat elliptical rigid punch of Sect. 2.3.2 is loaded by a force P whose line of
action passes through the point (c, 0). This causes the punch to tilt about the y-axis.
Use Galin’s theorem to determine the corresponding contact pressure distribution,
Problems 27
assuming that the entire planform of the punch remains in contact with the half-space.
Hence, determine the maximum value of c if this assumption is to be correct.
5. Use the potential (2.4) and Appendix A, Eq. (A.3) to find the radial (tangential)
surface displacement u r (r, θ, 0) due to the concentrated normal force P. Hence, show
that the area of any circle r = a inscribed on the surface will be reduced as a result
of the deformation, and find the extent of this reduction.
6. Use the potential function solution of Appendix A, Eq. (A.2) to find a relation
between the surface dilatation
∂u x ∂u y
e S (x, y) ≡ + at z = 0,
∂x ∂y
and the contact pressure p(x, y), assuming frictionless conditions. Hence show that
the radial surface displacement u r (r, 0) and the contact pressure p(r ) in an axisym-
metric problem are related by the equation

(1 − 2ν) r
u r (r ) = − p(s)sds.
2Gr 0
7. Use a superposition similar to that in Eq. (2.5) to write expressions for the stress
components σrr , σzz on the axis r = 0 in cylindrical polar coordinates r, θ, z, when
the circle 0 ≤r < a is loaded by an axisymmetric contact pressure distribution p(r ).
Evaluate the resulting integrals for the case where p(r ) is uniform and equal to
p0 , and plot a graph showing how the maximum shear stress varies along the axis.
8. Use Eqs. (2.4), (2.50) to find the stress component σx x at a general point (x, y, 0) at
the surface due to the point force of Fig. 2.2. Hence determine this stress component
at a point on the x-axis when the half-space is loaded by a total normal force P
uniformly distributed along the line x = 0, −a < y < a.
Chapter 3
Hertzian Contact
If the contacting bodies are smooth, the gap function of Eq. (1.1) can be expanded as
a power series in x and y, for points sufficiently near the origin. Furthermore, since
the coordinate system in Fig. 1.2 satisfies the conditions
∂g0 ∂g0
g0 (0, 0) = 0; (0, 0) = 0; (0, 0) = 0, (3.1)
∂x ∂y
the first non-zero terms in this series are the quadratic terms
g0 (x, y) = Ax 2 + By 2 + C x y. (3.2)
The problem of Eq. (1.3) is influenced by the gap function only in the contact region
A which is generally small in non-conformal contact, so the higher order terms in g0
can often be neglected even when the surfaces are not strictly quadratic. The elastic
contact problem for a gap function defined by Eq. (3.2) was first solved by Hertz
(1882) and the resulting stress and displacement fields are generally referred to as
Hertzian contact.
3.1 Transformation of Coordinates
The second derivatives
∂ 2 g0 ∂ 2 g0 ∂ 2 g0
= 2 A, = 2B, = C. (3.3)
∂x 2 ∂ y2 ∂x∂ y
of the function (3.2) define the components of the two-dimensional curvature tensor,
and they obey the transformation rules for all Cartesian tensors—i.e. they can be

https://doi.org/10.1007/978-3-319-70939-0_3
30 3 Hertzian Contact
Fig. 3.1 Rotation of

coordinates in the tangent y
plane
y1
x1
α
α
O x
transformed into different Cartesian coordinate systems using Mohr’s circle. We can
prove this by defining a new Cartesian axis system x1 , y1 inclined to x, y at some
angle α, as shown in Fig. 3.1.
The coordinate pairs {x, y} and {x1 , y1 } are the components of the same position
vector r in the two systems, and hence they are related by the vector transformation
rules
x1 = x cos α + y sin α; y1 = y cos α − x sin α. (3.4)
Also, the vector operator [the gradient]
∂ ∂
∇=i + j (3.5)
∂x ∂y
obeys the same transformation rules, so
∂ ∂ ∂ ∂ ∂ ∂
= cos α + sin α ; = cos α − sin α . (3.6)
∂x1 ∂x ∂y ∂ y1 ∂y ∂x
Using these results, we conclude that

∂ 2 g0 ∂ ∂ ∂g0 ∂g0
= cos α + sin α cos α − sin α
∂x1 ∂ y1 ∂x ∂y ∂y ∂x
2
2 ∂ 2
g0 ∂ g0 ∂ 2 g0
= cos α − sin α
2
+ sin α cos α − . (3.7)
∂x∂ y ∂ y2 ∂x 2
Similarly, we find
∂ 2 g0 2 ∂ g0
2
2 ∂ g0
2
∂ 2 g0
= cos α + sin α + 2 sin α cos α (3.8)
∂x1 2 ∂x 2 ∂ y2 ∂x∂ y
∂ g0
2
∂ 2
g0 ∂ 2
g0 ∂ 2 g0
= cos 2
α + sin 2
α − 2 sin α cos α . (3.9)
∂ y1 2 ∂x 2 ∂ y2 ∂x∂ y
These are immediately recognizable as similar in form to the Mohr’s circle equations
for the transformation of stress in two dimensions.
3.1 Transformation of Coordinates 31
In particular, using (3.3), (3.7), we obtain
∂ 2 g0
= C cos2 α − sin2 α + 2 (B − A) sin α cos α, (3.10)
∂x1 ∂ y1
and this will be zero if α is chosen to satisfy the equation
C
tan(2α) = . (3.11)
(A − B)
The resulting axes x1 , y1 are then aligned with the principal directions of the quadratic
function g0 and it is convenient to derive the Hertzian contact solution in this principal
axis system, in which the constant C is zero.
3.1.1 Cylinders and Spheres
In many practical applications, the contacting bodies will be cylinders or spheres

with known principal directions and radii of curvature. For example, if a cylinder of
radius R is in contact with a plane surface, we can choose the y-axis to be parallel
with the axis of the cylinder in which case the gap function can be determined from
Fig. 3.2.
We obtain
g0 = R(1 − cos θ); x = R sin θ, (3.12)
but since we restrict attention to points close to the initial contact point O where
θ 1, we can use the approximations
θ2
sin θ ≈ θ; cos θ ≈ 1 − , (3.13)
2
Fig. 3.2 Gap function for a

cylindrical body of radius R x
O
g0
θ
from which
x2 1
g0 (x, y) = or A= ; B = 0. (3.14)
2R 2R
For a sphere of radius R, similar considerations yield
x2 y2 1
g0 (x, y) = + or A=B= . (3.15)
2R 2R 2R
3.1.2 More General Cases
For more general ellipsoidal bodies, we can write
x2 y2
g0 (x, y) = + , (3.16)
2Rx 2R y
where Rx , R y are the principal radii of curvature of the surface.

If both bodies are curved, the gap function is the sum of expressions derived from
each body separately, as in Eq. (1.1). If the principal curvatures of the two bodies are
aligned, we simply obtain
x2 y2 x2 y2
g0 (x, y) = Ax 2 + By 2 = + + + , (3.17)
2Rx(1) 2R (1)
y 2Rx(2) 2R (2)
y
where the superscripts (1) ,(2) refer to the radii of bodies 1 and 2, respectively. This
expression is identical to (3.16) if we define composite radii
1 1 1 1 1 1
= 2 A = (1) + (2) ; = 2B = (1) + (2) . (3.18)
Rx Rx Rx Ry Ry Ry
When both bodies are curved, it is possible for one of them to be concave, provided
that its radius of curvature is greater than that of the contacting convex body. This
case can be included in the above equations by denoting concave radii by equal
negative values.
If the principal curvatures of the two bodies are not aligned, we can use the trans-
formation equations (3.7)–(3.8) to determine the second derivatives of the separate
functions g1 (x, y), g2 (x, y) in any convenient but common coordinate system. These
functions can then be added, leading to an expression of the form (3.2), after which
the transformation equations are applied again to obtain the principal curvatures
A, B, or equivalently, the composite radii Rx , R y .
3.1 Transformation of Coordinates 33
We obtain1

1 1 1 1 1
A+B = + (1) + (2) + (2)
2 Rx(1) Ry Rx Ry
⎡ 2 2
1⎣ 1 1 1 1
|B − A| = − (1) + − (2)
2 Rx(1) Ry Rx(2) Ry

1/2
1 1 1 1
+2 − (1) − (2) cos(2α) , (3.19)
Rx(1) Ry Rx(2) Ry
where α is the angle between the x-axis in body 1 and that in body 2. In labelling
the final principal axes (x, y), we shall adopt the convention that B > A, so that the
absolute value sign can be dropped in Eq. (3.19). In particular, this requires that the
x- and y-axes be interchanged in the cylindrical contact problem of Fig. 3.2.
3.2 Hertzian Pressure Distribution
By setting n = 1 in Eqs. (2.43), (2.46), we find that the pressure distribution

x2 y2
p(x, y) = p0 1 − − (3.20)
a2 b2
applied over the ellipse
x2 y2
+ <1 (3.21)
a2 b2
generates the surface displacement
π
p0 1/2
u z (x, y) = D 2 C2 dθ. (3.22)
2E ∗ 0
Substituting for C2 , D from Eqs. (2.27), (2.29), respectively, and performing the
remaining integral2 for the case where b ≤ a, we obtain

p0 b E(e)
u z (x, y) = a 2 2
e K (e) − {K (e) − E(e)}x 2
− − K (e) y2 ,
E ∗ a 2 e2 (1 − e2 )
(3.23)
1A proof of these results is given by Johnson (1985), Appendix 2.

2 More details of the mathematical procedure for evaluating this integral are given in Appendix B,
Sect. B.2.1.
where K (e) is defined in Eq. (2.32),

π/2
E(e) = 1 − e2 cos2 θ dθ (3.24)
0
is the
complete elliptic integral of the second kind, and we recall that the eccentricity
e = 1−b2 /a 2 from (2.28). Equation (3.23) has the form
u z (x, y) = Δ − Ax 2 − By 2 , (3.25)
with

p0 b p0 b E(e)
A = ∗ 2 2 [K (e) − E(e)]; B= ∗ 2 2 − K (e) (3.26)
E a e E a e (1 − e2 )
p0 bK (e)
Δ= . (3.27)
E∗
This solution can therefore be used for the contact problem in which the gap function
is given by Eq. (3.17)—i.e. when the axis system has been chosen as in Sect. 3.1 to
align with the principal curvatures of g0 . We also record the value of the total normal
contact force
a b√1−x 2 /a 2 1/2
x2 y2 2π p0 ab
P=4 p0 1 − 2 − 2 d yd x = . (3.28)
0 0 a b 3
Notice that the pressure distribution (3.20) tends to zero at the boundary of the
ellipse, as required by the arguments of Sect. 1.2.4 and Chap. 10.
3.3 Strategy for Hertzian Contact Calculations
3.3.1 Eccentricity of the Contact Area
The first step in a Hertzian contact calculation is to determine the eccentricity e of

the contact area. The profiles of the contacting surfaces are known, so Eqs. (3.19)
can be used to determine A, B, after which we divide the two expressions (3.26) to
obtain

1 E(e) B
− K (e) = , (3.29)
[K (e) − E(e)] (1 − e ) 2 A
from which e can be determined.

3.3 Strategy for Hertzian Contact Calculations 35
Fig. 3.3 Eccentricity e of

the contact area as a function
of eg
Notice that contours of g0 [lines along which g0 (x, y) is constant] are also ellip-
tical, with eccentricity
A
eg = 1 − . (3.30)
B
However, the eccentricity of the contact area is generally larger than that of the
composite profile, the relation being plotted as the solid line in Fig. 3.3. The dashed
straight line represents the expression
2eg
e= √ , (3.31)
3
which provides a good approximation to the eccentricity in the range 0 < eg < 0.4.
3.3.2 Dimensions of the Contact Area
Equation (3.29) cannot be solved in closed form, but Maple or Mathematica will
solve it for a specified numerical value of B/A. Once e is known, the major and
minor axes a, b of the contact area can then be determined. Usually, either the total
compressive force P or the rigid-body approach Δ will be prescribed. Alternatively,
we may wish to determine the maximum allowable value of P or Δ if some critical
value of the maximum contact pressure p0 is not to be exceeded.
Total Force Prescribed

If the total force P is specified, we can eliminate p0 between Eqs. (3.26)1 , (3.28) to
obtain

3P[K (e) − E(e)] 1/3
a= , (3.32)
2πe2 E ∗ A
after which

b = a 1 − e2 . (3.33)
Once a, b are known, the maximum contact pressure p0 is obtained as
3P
p0 = (3.34)
2πab
from (3.28), and the relative rigid-body approach of the bodies is then given by
Eq. (3.27).
Rigid-Body Approach Prescribed
In this case, we can eliminate p0 between Eqs. (3.26)1 , (3.27) to obtain

Δ E(e)
a= 1− , (3.35)
Ae2 K (e)
after which
∗
E Δ
b = a 1 − e2 and p0 = . (3.36)
bK (e)
Finally, the compressive force P is obtained by substituting these results into

Eq. (3.28).
Notice that the dimensions a, b of the contact ellipse both increase with Δ1/2 . It
follows that the maximum contact pressure p0 ∼ Δ1/2 and the total force P ∼ Δ 3/2
from (3.34). In other words, the Hertzian contact acts as a stiffening spring.
Maximum Contact Pressure Prescribed
In this case, we use (3.33) to eliminate a from (3.26)1 , giving
p0 (1 − e2 ) b
b= [K (e) − E(e)] after which a=√ . (3.37)
E ∗ e2 A 1 − e2
The total force P and the rigid-body approach Δ are then determined from
Eqs. (3.28), (3.27), respectively.
These calculations are most easily performed in Maple or Mathematica, but
numerical values of the elliptic integrals K (e), E(e) can also be estimated to
3.3 Strategy for Hertzian Contact Calculations 37
Fig. 3.4 The complete

elliptic integrals K (e), E(e)
as functions of eccentricity e.
The definitions of these
functions are given in
Eqs. (2.32), (3.24),
respectively
acceptable accuracy from Fig. 3.4. For e < 0.6, the difference K (e)− E(e) is best
calculated from the approximation K (e)− E(e) ≈ π(e2 +15e4 /32)/4.
Example
A sphere of diameter 20 mm is pressed into the curved surface of a cylinder of
diameter 50 mm by a force of 20 N. Both bodies are of steel, for which E = 210 GPa,
ν = 0.3. Find the dimensions of the contact ellipse and the maximum contact pres-
sure p0 .
Solution
In the notation of Sect. 3.1.2,
Rx(1) = R (1)
y = 10mm; Rx(2) = ∞; R (2)
y = 25 mm,
so
1 1 1 1
A= + = 50 m−1 ; B= + = 70 m−1 ,
0.02 ∞ 0.02 0.05
from Eq. (3.18), and B > A as required.3 We then have

A 20
1− = = 0.535,
B 70
and Fig. 3.3 shows that e = 0.60 [Solution of Eq. (3.29) using Maple gives
e = 0.6011].
The force P = 20 N, and
1 2(1 − ν 2 ) ∗ 210
∗ = so E = = 115.4 GPa.
E E 2 × 0.91
3 Ifat this stage we had obtained B < A, we would simply interchange the x- and y-axes so as to
satisfy this convention.
The semi-major axis of the contact area is then obtained from Eq. (3.32) as
1/3
3 × 20[K (0.60) − E(0.60)]
a= = 0.115 × 10−3 m (0.115 mm)
2π × 0.602 × 115.4 × 109 × 50
and

b = a 1 − e2 = 0.092 mm.
Finally, the maximum contact pressure is obtained from (3.34) as
3 × 20
p0 = = 900 MPa.
2π × 0.115 × 0.092
3.3.3 Highly Elliptical Contacts
If the eccentricity of the contact area is close to unity, numerical errors may result from
the elliptic integral calculations. This will arise for example if two nominally parallel
cylindrical bodies are slightly misaligned due to manufacturing or support errors. For
e ≈ 1, the elliptic integrals have convenient asymptotic expansions. Writing
b
z= = 1 − e2 1, (3.38)
a
we have

4 z2 4
K (e) → ln + ln − 1 + ... (3.39)
z 4 z

z2 4 1
E(e) → 1 + ln − + ... (3.40)
2 z 2
For values of z < 0.02, the first term in each of (3.39), (3.40) will give an accuracy
of at least 1 in 103 , which is ample for practical problems. Using this approximation
in (3.29), we then obtain

A 4
≈ z 2 ln −1 , (3.41)
B z
which is compared with the exact relation in Fig. 3.5.

In the limit where two cylinders are perfectly aligned, e = 1 and the problem
becomes two-dimensional. We shall consider this case in the context of more general
two-dimensional problems in Chap. 6.
3.4 First Yield 39
Fig. 3.5 Ratio of the

semi-axes b/a as a function
of A/B for highly elliptical
contact areas. The dashed
line represents the
approximation (3.41)
3.4 First Yield
The Hertzian contact solution depends on the material remaining linear and elastic,
but we can use the resulting stress components, calculated as in Sect. 2.5, to determine
the value of the applied force at which yielding is predicted to initiate. Johnson
(1985) gives estimates based on the Tresca [maximum shear stress] and von Mises
[maximum distortion strain energy] criteria and shows that first yield is predicted in
a fairly narrow range
1.6SY < p0 < 1.8SY , (3.42)
where SY is the yield stress in uniaxial tension and p0 is the maximum contact
pressure defined in Eqs. (3.20), (3.34), (3.36).
The first point to yield lies on the z-axis at a depth d that depends on the eccentricity
e, the yield criterion used and Poisson’s ratio. However, in all cases, the depth lies in
the range 0.45b<d<0.8b, where b is the semi-minor axis of the ellipse. It is worth
noting that Hertzian contact is one of very few physical situations where first yield
occurs at an interior point, rather than at a point on the boundary of the loaded body.
For this reason, small amounts of plastic deformation in Hertzian contact are difficult
to detect, since sub-surface yielding has only a minor effect on the surface profile
after unloading. It is significant that in the hardness test, the average contact pressure
over the now visible permanent indentation is about 2.8S y , which is almost twice the
value of p0 at first yield.
Problems
1. Use Eqs. (3.7)–(3.8) to prove the relations (3.19) for the equivalent curvature in
the contact of misaligned quadratic profiles.
2. Figure 3.6 shows a ball bearing with an inner race of radius 36 mm, an outer race
of radius 48 mm and 16 balls of 12 mm diameter. Both inner and outer race surfaces
have a [concave] groove of radius 12 mm. All the components are of steel, for which
E = 210 GPa and ν = 0.3.
Find the maximum radial force which the bearing can transmit if the maximum
Hertzian contact pressure is not to exceed 1000 MPa. Make the pessimistic assump-
tion that at a given time, all the force is carried by a single ball.
3. Ball bearings are sometimes preloaded in order to increase the incremental stiffness
[the response to small changes in lateral force], since this [for example] influences
the dynamic behaviour of the bearing under vibratory loading and reduces errors
in precision machinery. Preload can be achieved by making the space between the
inner and outer race slightly smaller than the diameter of the balls. If the bearing in
Problem 2 is preloaded such that when assembled each ball transmits a normal force
of 20 N, determine the incremental stiffness of the bearing. Assume that all the balls
remain in contact and, hence, share in transmitting the incremental force.
4. Two elastic cylinders each of radius 300 mm and length 1200 mm are designed
to roll against each other whilst transmitting a normal force of 500 kN. However,
assembly errors cause the axes of the cylinders to be inclined to each other 0.2◦ ,
resulting in an elliptical Hertzian contact. Find the dimensions of the contact area and,
hence, determine whether contact will extend to the ends of the rollers. [E = 210 GPa,
ν = 0.3].
5. Use Eqs. (3.34), (3.36) and (3.35) to express the force P as a function of indentation
∗
Δ and A, e, E . Hence, find the incremental stiffness d P/dΔ and show that it is equal
to the [constant] stiffness for an elliptical flat rigid punch whose planform is identical
with the contact area in the Hertz problem at force P.
Fig. 3.6 Ball bearing. All

dimensions are in mm
12 mm
radius
48
A A
36
inner race
φ12
view on A-A
outer race
Problems 41
Fig. 3.7 Wheel/rail contact
0.5 m
6. If the eccentricity of the contact area is very close to unity, the stress state near
x = 0 will be approximately two-dimensional. Use Eqs. (3.37), (3.29) and (3.18) to
determine a relation between the maximum contact pressure p0 , the contact semi-
width b and the radius of curvature R y . Then, use the limiting expressions (3.39),
(3.40) to show that this relationship becomes independent of e and, hence, of Rx
when Rx R y .
For this limiting case, find the indentation depth Δ as a function of p0 , b, e
and comment on the results. In particular, what are the implications for the two-
dimensional contact problem of a cylinder pressed against a plane?
7. Figure 3.7 shows a cross-sectional view of a railway wheel in contact with a rail.
The radius of the wheel at the contact point is 0.5 m and the contact surface is a cone
of angle 5◦ as shown. The rail contact surface can be approximated as a cylinder of
radius 100 mm. Both bodies are made of steel for which E = 210 GPa, ν = 0.3.
If the wheel transmits a vertical force of 10 kN, find the maximum contact pressure
p0 and the dimensions of the contact area.
8. One component of an optical lens support system comprises a sphere of diameter
30 mm contacting a concave cylindrical groove of radius 20 mm. Both components
are made of steel for which E = 210 GPa, ν = 0.3. If the contact transmits a normal
force of 2 kN, find the maximum contact pressure p0 , the indentation depth Δ and
the incremental stiffness d P/dΔ to small changes in loading.
Chapter 4
More General Problems for the Half-Space
We have seen in Chaps. 1 and 2 that the general frictionless elastic contact problem for
two bodies approximated by half-spaces reduces to the indentation of a single elastic
∗
half-space with composite modulus E by a rigid punch defined by the combined
gap function g0 (x, y). If a rigid-body displacement Δ is imposed on this punch, the
problem is defined by the boundary conditions
u z (x, y) = Δ − g0 (x, y) (x, y) ∈ A (4.1)

p(x, y) = 0 (x, y) ∈
/A (4.2)

p(x, y)d xd y = P, (4.3)
A
where the contact area A is determined by the inequalities
u z (x, y) ≥ Δ − g0 (x, y) (x, y) ∈

/A (4.4)
p(x, y) > 0 (x, y) ∈ A. (4.5)
The stress and displacement fields in the deformable body are then conveniently
represented in terms of the elasticity solution of Appendix A, Sect. A.1.
In order to preserve generality, we express the results in terms of the composite
∗
modulus E of Eq. (2.15) and hence replace
(1 − ν) 2
by
G E∗
in Eq. (A.4). In particular, the contact pressure is given by

https://doi.org/10.1007/978-3-319-70939-0_4
44 4 More General Problems for the Half-Space
∂2ϕ
p(x, y) = −σzz (x, y, 0) = (x, y, 0), (4.6)
∂z 2
and the combined normal surface displacement is
2 ∂ϕ
u z (x, y) = − (x, y, 0). (4.7)
E ∗ ∂z
With this representation, the contact problem is reduced to the determination of

a function ϕ(x, y, z), harmonic [i.e. ∇ 2 ϕ = 0] in z > 0, satisfying the boundary
conditions
∗
∂ϕ E
(x, y, 0) = (g0 (x, y) − Δ) (x, y) ∈ A (4.8)
∂z 2
∂2ϕ
(x, y, 0) = 0 (x, y) ∈
/ A, (4.9)
∂z 2
where the contact area A is determined by the inequalities

∗
∂ϕ E
(x, y, 0) ≤ (g0 (x, y) − Δ) (x, y) ∈
/A (4.10)
∂z 2
∂2ϕ
(x, y, 0) > 0 (x, y) ∈ A. (4.11)
∂z 2
Here, we are assuming that the indentation depth Δ is given, in which case the
total applied force is a dependent variable given by

∂2ϕ
P= (x, y, 0)d xd y. (4.12)
A ∂z 2
Alternatively, if P is prescribed, Δ must be regarded as an additional dependent

variable.
If the gap function is not a quadratic function of x and y, the contact area A
will not generally be elliptical and an analytical solution of the above equations is
impracticable, except in the special case of axisymmetry which we shall discuss in
the next chapter. Here, we shall consider some general features of three-dimensional
problems.
4.1 The Electrical–Mechanical Analogy
If the punch is flat [g0 (x, y) = 0], the mathematical form of Eqs. (4.8)–(4.12) is
identical to that of the electrical conduction problem in which the region A is raised
to some constant electrical potential V = V0 , whilst the rest of the surface is insulated.
The relation between current density i and potential is
4.1 The Electrical–Mechanical Analogy 45
1
i = − ∇V, (4.13)
ρ
where ρ is the resistivity of the material, and continuity of current flux demands that

1
div i = − div ∇V = 0, (4.14)
ρ
so V must be harmonic if the resistivity is uniform [i.e. if the body is homogeneous].

The boundary conditions for the above electrical problem are therefore
V (x, y, 0) = V0 (x, y) ∈ A (4.15)

∂V
(x, y, 0) = 0 (x, y) ∈
/ A, (4.16)
∂z
where

1 ∂V
I = i z (x, y, 0)d xd y = − (x, y, 0)d xd y (4.17)
A ρ A ∂z
is the total current flowing into the half-space, and ∇ 2 V = 0.

The electrical and mechanical contact problems can be made mathematically
identical by writing
2V0 ∂ϕ
V (x, y, z) = − (x, y, z), (4.18)
E ∗Δ ∂z
and with this identity we deduce that the electrical conductance
I 2P
= (4.19)
V0 ρE ∗Δ
is proportional to the mechanical stiffness of the contact P/Δ. This relationship is

independent of the shape or dimensions of the contact area A.
Equation (4.19) can be generalized to the contact of two half-spaces of dissimilar
materials [resistivity ρ1 , ρ2 ] at electrical potentials V1 , V2 respectively, and to non-
conformal contact, since the incremental contact problem is then equivalent to a flat
punch indentation problem for the instantaneous contact area.1 We obtain the general
result (Barber 2003).
I 2 dP
= ∗ , (4.20)
V1 − V2 (ρ1 + ρ2 )E dΔ
This relation allows us to apply results from the study of electrical contact resistance
to deduce conclusions about contact stiffness. For example, the force–displacement
1 We shall explore this relation more fully in Sect. 6.3 below.

relation for an axisymmetric flat punch can be deduced from Holm’s solution for the
contact resistance associated with a circular contact area (Holm 1958).
Notice incidentally that the ratio V0 /ρI has dimensions of length−1 , so a more
illuminating way of writing Eq. (4.19) is
∗
V0 E Δ S
= = , (4.21)
ρI 2P a
where a is a representative length dimension of the contact area and S is a dimen-

sionless shape factor. For example, if the contact area is circular and a is taken as
its radius, S = 1/4. It follows that if the linear dimensions of the contact area are
all increased in the same ratio λ, thus preserving the same shape, the stiffness will
increase by the factor λ and the contact resistance will decrease by 1/λ.
4.1.1 Other Mathematical Analogies
Several other physical mechanisms lead to harmonic boundary value problems of

the form (4.15), (4.16) and analogies of this kind can often provide extra insight into
the behaviour to be expected in the elastic contact problem. An obvious parallel is
with the steady-state conduction of heat, for which the governing equations are
q = −K ∇T ; div q = 0, (4.22)
where q is the heat flux, T is the temperature and K is the thermal conductivity.
It follows in the same way that if two homogeneous half-spaces with conductivity
K 1 , K 2 make perfect thermal contact over some area A, the remaining part of the
interface being insulated, the total heat flux between the half-spaces

Q= qz (x, y, 0)d xd y (4.23)
A
will satisfy the equation

∗
Q 2K d P
= ∗ , (4.24)
T1 − T2 E dΔ
where the extremities of two half-spaces are maintained at steady temperatures T1 , T2 ,

respectively, and
1 1 1
∗ = + . (4.25)
K K1 K2
A related boundary value problem is that for the distribution of electrostatic charge
density σ(x, y) on a conductor in the form of a thin disk of planform A located on
the plane z = 0 in an infinite space. The potential V= V0 must be uniform throughout
the conductor, and using symmetry about the plane, we obtain the boundary value
problem
V (x, y, 0) = V0 (x, y) ∈ A (4.26)

∂V
(x, y, 0) = 0 (x, y) ∈
/ A, (4.27)
∂z
for the half-space z > 0, where ∇ 2 V = 0. The charge density in A is then given by
1 ∂V
σ(x, y) = − (x, y, 0), (4.28)
2π ∂z
(Maxwell 1892). If we then define the total charge

Σ= σ(x, y)d xd y, (4.29)
A
the mathematical analogy shows that
Σ 1 dP
= . (4.30)
V0 π E ∗ dΔ
The electrostatic analogy can be extended to the case where N thin planar con-
ductors Ai , i = (1, N ) are located at the plane z = 0, where the potential and the
total charge on conductor i are Vi , Σi , respectively. Since the problem is linear, these
quantities must be related by a matrix equation

N
Σj = Mi j Vi (4.31)
i=1
(Maxwell 1892, Sect. 86), where the symmetric matrix Mi j depends only on the
geometry and relative locations of the areas Ai . The analogous contact problem
involves N flat rigid punches of planform Ai , loaded by forces Pi , i = (1, N ) and
experiencing indentations Δi . It then follows that
∗
N
1
N
Pj = π E Mi j Δi or Δj = ∗ Ci j Pi , (4.32)
i=1
π E i=1
where C = M −1 . Several theorems in Sect. 4.2 (below) depend on mathematical

results that were originally developed in the context of electrostatics or Newtonian
gravitation (Kellogg 1929).
4.1.2 Boyer’s Approximation
Boyer (2001) describes an ingenious way of estimating the shape factor S for
‘tileable’ areas such as squares, rectangles and triangles. The method depends on the
superposition (4.32) and an approximation of the off-diagonal elements Mi j , i = j
by the point force result (2.14). If two contact areas Ai , A j are sufficiently far apart,
the displacement Δ j due to Pi alone will be approximately uniform within Ai and
given by
Pi Ci j Pi 1
Δj = ≈ so Ci j ≈ , (4.33)
πE∗ π E ∗ ri j ri j
where ri j is a measure of the distance between the two areas, typically the distance
between their centroids.
Figure 4.1 shows a square contact area of side 2a which we conceive as made
up of four identical squares of side a. The total force P must be shared equally
between the four squares by symmetry, and we suppose that the force P/4 acting
∗
over one of these squares acting alone produces a displacement of P S/2E a, from
Eq. (4.21). However, the same force will also produce a displacement at each of the
other squares, which we suppose to be given √ on average by (4.33) with P → P/4,
and ri j = a for the adjacent squares and ri j = a 2 for the diagonally opposite square.
The total displacement at each square is therefore due to four such forces, one on
each corner square, and is estimated as
PS P P PS
Δ= ∗ + ∗ + √ ∗
= ∗ , (4.34)
2E a 2π E a 4 2π E a E a
where the expression on the right is obtained simply by treating the area as a single
square of side 2a loaded by a force P, using (4.21). Equating the two expressions
∗
for Δ and cancelling the common factor P/E a, we can solve for S obtaining
√
4+ 2
S= = 0.431. (4.35)
4π
Fig. 4.1 A square contact

area of side 2a
a a 2
a
a
a a
Fig. 4.2 Characterization of

the contact area A for
Fabrikant’s approximation
a(θ)
θ
C
Nakamura (1995) gave a numerical solution of the problem and his result cor-
responds2 to a shape factor S = 0.434 in Eq. (4.21). Thus, Boyer’s approximation
underestimates the true shape factor by about 1%.
4.1.3 Fabrikant’s Approximation
We know from asymptotic considerations that the contact pressure under a flat rigid
punch must exhibit a square-root singularity at the edge of the contact area. Fabrikant
(1986) proposed an approximation satisfying this condition as
−1/2
r2
p(r, θ) = p0 1 − , (4.36)
a(θ)2
where the contact area A is defined by the boundary r = a(θ) in appropriate polar
coordinates, as shown in Fig. 4.2.
The origin of coordinates defines the point where the pressure is a minimum
and this must clearly lie on any axis of symmetry for the contact area A. For more
general shapes, it is reasonable to associate the origin with the centroid C of A. It is
easily verified that the distribution (4.36) is exact when the contact area is circular or
elliptical, so we might expect the approximation to be good for shapes that are not
too different from an ellipse.
The total indenting force is obtained as
2π a(θ) 2π
P= p(r, θ) r dr dθ = p0 [a(θ)]2 dθ, (4.37)
0 0 0
2 Nakamura considered the case where two electrodes of the same material make contact over a
square, so his dimensionless resistance is actually twice that reported here.
and the punch indentation Δ can be approximated by the displacement at the origin
which in turn can be written as
2π a(θ)
1
u z (0, 0) = p(r, θ) dr dθ, (4.38)
πE∗ 0 0
by treating C as the field point in Fig. 2.7 and Eq. (2.21), but sweeping out the contact
area as a set of sectors radiating from C. Substituting for p(r, θ) from (4.36) and
evaluating the inner integral, we then obtain

p0 L 2π
Δ ≈ u z (0, 0) = where L= a(θ)dθ. (4.39)
2E ∗ 0
Example
As an example, we consider the case where A is a square of side b. This can be
considered as the superposition of four triangles as shown in Fig. 4.3, in each of
which, with a suitable reference for θ,
b π π
a(θ) = ; − <θ< .
2 cos θ 4 4
It follows that
π/4 √ 2π π/4
dθ dθ
L = 2b = 4b ln 1 + 2 ; [a(θ)]2 dθ = b2 = 2b2
−π/4 cos θ −π/4 cos θ
2
0
and hence [from (4.39), (4.37)]

√
2 p0 b ln 1 + 2 E Δb
∗
∗
Δ= ; P = 2 p0 b = 2
√ ≈ 1.135E Δb.
E∗ ln 1 + 2
Fig. 4.3 Estimation of the

shape factor S for a square b
using Fabrikant’s
approximation
a(θ)
b θ
C
This corresponds to a shape factor S ≈ 0.441 from (4.21), which is about 1.6% higher
than Nakamura’s numerical result.
4.2 General Theorems for Frictionless Contact
The potential function formulation (4.8)–(4.11) can be used to prove general theorems
for frictionless contact problems for the half-space. Some of these results may appear
intuitively obvious, but it should be emphasised that exceptions can be found for
geometries that do not involve half-spaces, notably in the case of contact involving
beam-like structures [see Chap. 13].
Theorem 1 If a distribution of normal traction is applied to the surface of the

half-space, corresponding to a positive total compressive force P, the maximum
inward surface displacement u z (x, y) occurs at a point where the traction is non-
zero and compressive [σzz (x, y, 0) < 0]. Conversely, if the traction corresponds to
a tensile force, the maximum outward surface displacement occurs at a point where
σzz (x, y, 0) > 0.
Proof If the maximum value of a function ψ occurs at an interior point, then at this
point we must have
∂2ψ ∂2ψ ∂2ψ

<0; <0; <0 and hence ∇ 2 ψ < 0. (4.40)
∂x 2 ∂ y2 ∂z 2
No such point can exist if the function is harmonic and hence in such cases, the max-
imum [and minimum] values must occur on the boundary of the domain (Maxwell
1892, Sect. 112).
If the domain comprises the half-space z > 0, and if the maximum of a harmonic
function ψ occurs at some point B(x, y, 0) on the surface, then ex hypothesi, the
value of ψ at all other points is less than ψ B and hence at B
∂ψ
< 0. (4.41)
∂z
Now since the function ϕ in Sect. A.1 is harmonic, it follows that
∂ 2 ∂ϕ
∇ ϕ = ∇2 = 0. (4.42)
∂z ∂z
In other words, the function ∂ϕ/∂z is also harmonic. Setting ψ = −∂ϕ/∂z and using
the above result, we conclude that the maximum value of −∂ϕ/∂z occurs on the
surface of the half-space, and at this maximum point
∂ψ ∂2ϕ
= − 2 < 0. (4.43)
∂z ∂z
Applying this result to (4.6), we conclude that the maximum surface displacement
u z (x, y) occurs at a point where p(x, y) > 0—i.e. in a region of compressive normal
traction.
Notice that if the total force acting on the surface of the half-space is compressive,
the maximum outward displacement may occur at infinity [and hence be zero in the
absence of rigid-body displacement], and it will certainly so occur if the tractions
are everywhere compressive.
Theorem 2 The contact area A which satisfies the inequalities (4.10), (4.11) with
a prescribed value of Δ is that which maximizes the value of P in the solution of the
well-posed boundary value problem (4.8), (4.9).
Proof Consider the effect of increasing A by an infinitesimal increment δA. If the

result of this change is to increase P(A), we deduce from Theorem 1 that the max-
imum incremental inward surface displacement must occur in the loaded region
(A + δ A), and this incremental displacement is zero in A, so the maximum must
occur in δA and must be positive. It follows that δA must be a region in which there
was interpenetration before the incremental process was applied.
Now consider the effect of decreasing A by an infinitesimal increment δA. If
the result of this change is to increase P(A), it follows as before that the maximum
incremental inward surface displacement occurs in δA and hence the incremental
normal traction in δA must be compressive. But after the incremental process, this
pressure must be zero, since δA has been removed from the contact area, so before
the process, the traction in δA must have been tensile.
We conclude that P(A) can be increased only by removing from A regions that
where the tractions were originally tensile, or adding regions where originally there
was interpenetration, and hence the maximum value of P(A) will occur when there
are no more such regions available—i.e. when the inequalities (4.10), (4.11) are
satisfied.
Theorem 3 If the pressure distribution under a flat rigid punch of planform A

indented by a unit distance Δ = 1 is denoted by p ∗ (x, y), then the total indenting
force P(A) for a punch of general profile g0 (x, y) pressed in to a depth Δ sufficient
to ensure full contact is given by

P(A) = Δ − g0 (x, y) p ∗ (x, y)d xd y. (4.44)
A
Proof From Betti’s reciprocal theorem (Barber 2010, Chap. 34), if the tractions and
surface displacements for two different loadings of the same linear elastic body are
denoted by the vectors t 1 , u1 and t 2 , u2 respectively, then
4.2 General Theorems for Frictionless Contact 53

t 1 ·u2 dΓ = t 2 ·u1 dΓ, (4.45)
Γ Γ
where Γ denotes the surface of the body.

Theorem 3 follows immediately on taking the flat punch solution as loading 1 and
the more general shape solution as loading 2 (Shield 1967; Barber 1974).
Corollary
Equation (4.45) can also be used to determine the moment of the resultant force
about the x- and y-axes, or equivalently, the coordinates {x̄, ȳ} of the point through
which this resultant acts. We obtain

x̄ P(A) = Δ − g0 (x, y) px∗ (x, y)d xd y (4.46)
A

ȳ P(A) = Δ − g0 (x, y) p ∗y (x, y)d xd y, (4.47)
A
where px∗ (x, y), p ∗y (x, y) are the pressure distributions in A that would produce the
displacement distributions
u z (x, y) = x and u z (x, y) = y (4.48)
respectively.
Theorem 4 The contact area A and the contact pressure p(x, y) are non-decreasing
functions of the total compressive force P, and the gap g(x, y) in the separation
region (x, y) ∈
/ A is a non-increasing function of P.
Proof Suppose a force P is applied, establishing a contact area A. We now increase
P by some infinitesimal increment δ P. During this incremental process, we allow
new contact to be established in some region δA so as to satisfy the interpenetration
inequality (4.10), but we prevent any loss of contact in A by applying tensile tractions
if necessary.3
The incremental process involves a compressive force δ P and it follows from
Theorem 1 that the maximum incremental inward displacement δu(x, y) must occur
in a region where the incremental contact traction is compressive, and hence in
A + δA. Now in A the incremental displacement is uniform, comprising merely
the change δΔ in the rigid-body displacement, and δA was originally a separation
region, so the incremental displacement there must be less than δΔ. Thus, all points
(x, y) ∈ A are points of maximum incremental displacement and it follows from
Theorem 1 that the incremental tractions must be compressive throughout A.
We conclude that the local contact pressure p(x, y) is a monotonically increasing
function of P. Also, since loss of contact at any point in A would imply locally
tensile incremental contact tractions, A must be a non-decreasing function of P.
3 We shall show in the next paragraph that there cannot actually be any such tensile regions.
The incremental change in the gap g(x, y) is
δg(x, y) = δu(x, y) − δΔ,
and since the maximum incremental displacement is equal to δΔ, we conclude that
δg(x, y) ≤ 0. In other words, the gap is a non-increasing function of P.
Corollary
If the punch is flat [g0 (x, y) = 0], the contact pressure is everywhere compressive
and no point under the punch loses contact. If any point were to lose contact, the max-
imum normal displacement would occur in this region which is unloaded, contrary
to Theorem 1. The maximum normal displacement therefore comprises the whole
planform of the punch A and hence by the same argument as in Theorem 1, condition
(4.43) must apply throughout A.
Theorem 5 The incremental stiffness d P/dΔ is a non-decreasing function of P.
Proof This theorem is most easily proved by appealing to the electrical–mechanical

analogy of Sect. 4.1. We know from Theorem 4 that the contact area can only increase
when P is increased. In the electrical problem, if a conducting body makes contact
with a perfect conductor at potential V0 , and if the potential V = 0 at infinity, then
by the electrical analogue of Theorem 1, at all interior points, including points on the
interface that are not in contact (‘separation points’), the potential must satisfy the
inequality 0 < V < V0 .
If such a separation point is now allowed to make electrical contact with the surface
at potential V0 , the total current flow must increase, and hence the total conductance
of the contact must increase. By the analogy, this also implies that the incremental
stiffness of the contact must increase. Thus
d2 P
≥ 0, (4.49)
dΔ2
implying that the force–displacement curve is concave upwards.
Theorem 6 If two different contact problems for the half-space are defined by the ini-
tial gap functions g0 (x, y) = g1 (x, y), g0 (x, y) = g2 (x, y) and if g2 (x, y) > g1 (x, y)
for all (x, y), then the corresponding compressive forces P1 (Δ), P2 (Δ) for a given
rigid-body approach Δ satisfy the inequality P1 (Δ) > P2 (Δ).
Proof Suppose that the contact area A2 , the total force P2 (Δ) and the corresponding
contact pressure distribution p2 (x, y) are known for the initial gap function g2 (x, y).
To obtain the solution for g1 (x, y), we need to superpose the solution of an appropriate
incremental problem. We construct this incremental solution in two steps. In the first
step, we maintain the total contact area fixed as A2 . From Theorem 3 and Eq. (4.44),
the change in P in this step is
4.2 General Theorems for Frictionless Contact 55

δP = [g2 (x, y) − g1 (x, y)] p2∗ (x, y)d xd y, (4.50)
A2
where p2∗ (x, y) is the pressure under a flat punch of planform A2 with unit indentation.
Now for all (x, y), g2 > g1 ex hypothesi, and p2∗ > 0 from the Corollary to Theorem 4,
so δ P must be positive.
At the end of the first step, we anticipate some regions of A2 may involve tensile
contact tractions and some regions outside A2 may involve interpenetration with the
profile g1 (x, y). However, we know from Theorem 2 that relaxing these inequality
violations can only cause P to increase. It follows that P1 > P2 .
4.3 Superposition by Differentiation
In Sect. 2.3, we used integration as a form of linear superposition to obtain an inte-

gral equation formulation of the contact problem. Differentiation is also a form of
superposition and can be used in the same way. For example, suppose we place a
force P at the point (−δx, 0) and an equal negative force −P at the origin. This
force pair constitutes a moment M = Pδx about the y-axis and taking the limit as
δx → 0, we obtain the solution for a concentrated moment acting at the origin as

M 1 1
u z (x, y) = lim ∗ −
δx→0 π E δx (x + δx)2 + y 2 x 2 + y2

M ∂ 1 Mx
= ∗ =− ∗ 2 . (4.51)
π E ∂x x +y
2 2 π E (x + y 2 )3/2
More generally, if a given contact pressure distribution p(x, y) = f (x, y) produces

a surface displacement u z (x, y) = g(x, y), it follows that the contact pressure
∂ ∂
p(x, y) = f (x, y) will produce u z (x, y) = g(x, y). (4.52)
∂x ∂x
Notice that the differentiation must be one that can be achieved by a superposition
and limiting procedure, as in the moment example above. Thus, we can differentiate
with respect to a Cartesian coordinate (equivalent to a shift of origin) or with respect
to a parameter in the solution (such as the semi-axes a, b in the Hertz solution), but
not with respect to the coordinate r in polar coordinates (r, θ).
Example: Derivatives of the Hertz Solution
To illustrate the procedure, we note from Sect. 3.2 that the pressure distribution

x2 y2
p(x, y) = p0 1 − − (4.53)
a2 b2
produces the normal surface displacement

2
p0 b a
u z (x, y) = ∗ 2 2 a e K (e) − {K (e) − E(e)}x − 2 E(e) − K (e) y 2
2 2 2
E a e b
(4.54)
inside the ellipse x 2 /a 2 + y 2 /b2 < 1. Applying the result (4.52), we deduce immedi-
ately that the distribution
−1/2
p0 x x2 y2
p1 (x, y) = − 2 1− 2 − 2 (4.55)
a a b
will produce the displacement

2 p0 b K (e) − E(e) x
u z (x, y) = − (4.56)
E ∗ a 2 e2
inside the ellipse. Notice that as a result of the differentiation, the symbol p0 now
represents a quantity with the dimensions of force per unit length, not a traction. We
also note that the same differential operation on the resultant force shows that the
distribution (4.55) is equivalent to a moment
2π p0 ab
M= (4.57)
3
about the negative y-axis.
Now Eq. (4.56) represents an inclined plane surface inside the ellipse, so these
distributions define the solution of the problem where a flat rigid elliptical punch
is forced to make frictionless contact over the end face, which is rotated through a
small angle about the negative y-axis. By superposing an appropriate multiplier of
this solution and the solution for a flat rigid elliptical punch subjected to a centric
force P from Sect. 2.3.2 and Eqs. (2.31), (2.33), we can obtain the contact pressure
distribution for the problem of Fig. 4.4, in which the flat rigid elliptical punch is
loaded by a force P whose line of action passes through the point (c, 0).
We obtain
−1/2
P 3cx x2 y2
p(x, y) = 1+ 2 1− 2 − 2 . (4.58)
2πab a a b
It also follows that the punch will experience a rigid-body indentation Δ and rotation
α about the y-axis given by

P K (e) 3Pc K (e) − E(e)
Δ= ; α= . (4.59)
π E ∗a π E ∗ a 3 e2
4.4 The Force–Displacement Relation 57
Fig. 4.4 Flat rigid elliptical P

punch loaded by an c
off-centre force
α a x
4.4 The Force–Displacement Relation
Theorem 3 of Sect. 4.2 permits us to find the indenting force P for a given indentation
Δ, provided the contact area A and the corresponding flat punch pressure distribution
p ∗ (x, y) are known.
For example, if the indenting body is a rigid cylinder of radius a, the ‘flat punch’
pressure distribution for a unit indentation is
∗
E
p ∗ (r, θ) = √ , (4.60)
π a2 − r 2
from Eqs. (2.33), (2.34) with Δ = 1 and b = a. Alternatively, Eq. (4.60) can be
obtained using the axisymmetric analysis of Chap. 5, and particularly Sect. 5.1.1.
Using this result and Theorem 3, we conclude that if the punch has a fairly general
[i.e. not necessarily axisymmetric] shape defined by the gap function g0 (r, θ), then
the indenting force P for full contact is given by
∗ a
E 2π Δ − g0 (r, θ) r dθdr
P= √ . (4.61)
π 0 0 a2 − r 2
This force will generally not act through the origin, but its line of action {x̄, ȳ}
can be found from the Corollary to Theorem 3 and particularly Eqs. (4.46), (4.47).
We first take the limit of Eqs. (4.55), (4.56) as b → a, from which we determine that
for the circle of radius a,
∗ ∗
2E r cos θ 2E r sin θ
px∗ (r, θ) = √ ; p ∗y (r, θ) = √ , (4.62)
π a2 − r 2 π a2 − r 2
since x =r cos θ, y =r sin θ. Substituting into Eqs. (4.46), (4.47), we then obtain
∗ a
2E 2π
g0 (r, θ)r 2 cos θ dθdr
x̄ P = − √ (4.63)
π 0 0 a2 − r 2
∗ a
2E 2π
g0 (r, θ)r 2 sin θ dθdr
ȳ P = − √ , (4.64)
π 0 0 a2 − r 2
since the terms involving the rigid-body indentation Δ integrate to zero.
4.4.1 Non-conformal Contact Problems
If the problem is non-conformal, the contact area A is not known, but if it can be suit-
ably parametrized, it can be determined by the sequential application of Theorems 2
and 3.
For example, if the punch is axisymmetric, we anticipate a circular contact area
of as yet unknown radius a. We first determine the normal force P as a function of a
using Theorem 3. The gap function is now a function of r only, so Eq. (4.61) reduces
to a
∗ Δ − g0 (r ) r dr
P(a) = 2E √ . (4.65)
0 a2 − r 2
Theorem 2 then requires that the contact radius a be defined by the equation

∂P ∂ a Δ − g0 (r ) r dr
=0 or √ = 0. (4.66)
∂a ∂a 0 a2 − r 2
Once a is determined, it can be substituted back into Eq. (4.65) to give the force–
displacement relation P(Δ).
This method is limited to problems in which the shape of the contact area is one
for which the flat punch problem can be solved, and in practice this means it must be
circular or elliptical. However, the method can also be used in a Rayleigh–Ritz sense
to determine the contact area and the force–displacement relation in cases where the
contact area can reasonably be approximated by a circle or an ellipse.
An alternative approach is to use Fabrikant’s approximate solution for the flat
punch problem, so that p ∗ (r, θ) is approximated by Eq. (4.36) with p0 obtained from
(4.39)1 by setting Δ = 1. We find
∗ 2π a(θ)
2E a(θ)[Δ − g0 (r, θ)]r dr dθ
P= , (4.67)
L 0 0 a(θ)2 − r 2
where L is defined by (4.39)2 .

4.4 The Force–Displacement Relation 59
Barber and Billings (1990) describe an optimization procedure for choosing the
function a(θ) in Eq. (4.67) so as to minimize P. They give results for a punch whose
shape is defined by one vertex of a tetrahedron. Comparison with a numerical solu-
tion shows that the force–displacement relation in this case is within 5% of the
approximation defined using (4.67).
Problems
1. Use Boyer’s method to estimate the shape function S for an equilateral triangle of
side a. You will need to represent the triangle as the sum of four smaller triangles, as
shown in Fig. 4.5. Notice that the force on the central triangle will differ from those
on the other three.
2. A rigid flat punch with a rectangular cross section of dimensions a×5a is pressed
into the surface of an elastic half-space by a force P. Find an approximate expression
for the indentation depth Δ by considering five separate punches [see Fig. 4.6], each
of a ×a square cross section and each indented to the same depth. Use Nakamura’s
result for the diagonal elements in the resulting stiffness matrix and Eq. (4.33) for
the off-diagonal elements. Compare your result with that for an elliptical punch of
semi-axes 5a/2, a/2 using Eq. (2.34).
3. Use Eq. (2.17) to determine the surface displacement at the point (c, 0) due to
a uniform contact pressure p0 acting over the square −a/2 < x < a/2, −a/2 <
y < a/2, where c > a/2. Hence, determine whether Eq. (4.33) overestimates or
underestimates the off-diagonal elements of the stiffness matrix in Problem 2.
4. Show that the boundary of the ellipse in Fig. 2.7 can be expressed in polar coor-
dinates as
1
a(θ) = √ ,
C2 (θ)
Fig. 4.5 An equilateral

triangle of side 2a
a
Fig. 4.6 Planform of a 5a

rectangular punch considered
as five adjacent squares a
where C2 (θ) is defined by Eq. (2.27). Then use Eqs. (4.36), (4.39) to verify that
Fabrikant’s solution gives the exact result for the pressure distribution under an
elliptical rigid flat punch with indentation Δ.
5. Figure 4.7 shows a rigid punch with two parallel plane faces A1 , A2 , pressed into
an elastic half-space by a force P sufficient to cause all points in both of these areas
to be in contact. Show that the work done by P during loading is least when the
two areas have the same height [i.e. when they lie in the same plane]. Hence, or
otherwise, show that of all punches of given planform A [convex or concave], the
work done loading to a given force P [sufficient to establish full contact] is least
when the punch is flat.
6. Show that if the gap function g0 (x, y) is convex, meaning
∂ 2 g0 ∂ 2 g0
+ >0 all (x, y),
∂x 2 ∂ y2
the contact area A in a frictionless contact problem for the half-space must be simply
connected for all applied forces.
7. An axisymmetric rigid punch is defined by the piecewise-linear gap function g0 (r ),

where the slope g0 (r ) is a non-decreasing function of r . Show that the resulting force–
displacement relation P(Δ) is continuous up to the first derivative.
8. Use Eq. (2.14) to find an expression for the surface curvature
∂2uz ∂2uz
∇2 (u z ) ≡ +
∂x 2 ∂ y2
Fig. 4.7 A rigid punch with

P
two plane faces
A1
A2
Problems 61
when the surface is loaded by a concentrated compressive force P.

Use an integral formulation [as in Sect. 2.3] to generalize this expression to a
distribution of compressive normal tractions p(x, y), and use your result to prove
that a local maximum value of u z (x, y) can occur only in a loaded region.
9. Use Eqs. (2.43), (2.46) with a = b to obtain the normal displacement in the con-
tact area 0 ≤ r < a for the pressure distribution p(r ) = C(a 2 −r 2 )3/2 . Show that an
appropriate derivative of this distribution, in combination with lower order axisym-
metric fields, can be used to solve the problem of a cylindrical flat-ended rigid punch
of radius a indenting the curved surface of an elastic cylinder of radius R a if
the indenting force is sufficiently large to ensure full contact. Comment on possible
methods for solving this problem at lower values of the indenting force.
10. Use the method of Sect. 4.4 to determine the force–displacement relation for
Problem 9 in the range where the entire punch face makes contact with the cylindrical
surface.
Chapter 5
Axisymmetric Contact Problems
If the gap function g0 (r ) is axisymmetric, and if contact is assumed to occur only

within a circle of some radius a, the problem of Eqs. (4.8)–(4.11) is reduced to the
search for an axisymmetric harmonic function ϕ(r, z) in cylindrical polar coordinates
(r, θ, z) satisfying the equations
∗
∂ϕ E [g0 (r ) − Δ]
= 0≤r ≤a (5.1)
∂z 2
∂ ϕ
2
=0 r >a (5.2)
∂z 2
and the inequalities

∗
∂ϕ E [g0 (r ) − Δ] ∂2ϕ
< r >a; >0 0 ≤ r ≤ a. (5.3)
∂z 2 ∂z 2
The inequalities serve to determine the unknown contact radius a, but we can usually
replace them by the condition that the contact pressure p(r ) → 0 as r → a [see
Sects. 1.2.4 and 10.1.3]. Furthermore, it is often convenient to treat the radius a as
an independent variable and solve the resulting problem for the compressive force P
as an unknown, since the resulting mathematical problem is then completely linear.
5.1 Green and Collins Solution
The first general solution of the problem defined by Eqs. (5.1), (5.2) was given by
Sneddon (1947), using a Hankel transform method that was later formalized by
Gladwell (1980). Here, we shall use a method that was developed by Green and
Zerna (1954) and Collins (1959, 1963).

https://doi.org/10.1007/978-3-319-70939-0_5
64 5 Axisymmetric Contact Problems
We start by representing the axisymmetric function ϕ(r, z) in the form

a
ϕ(r, z) = F(r, z, t)h(t)dt, (5.4)
0
where h(t) is a real function of t,

F(r, z, t) = ln r 2 + (z + ıt)2 + z + ıt , (5.5)
√
and ı = −1.
The square root in Eq. (5.5) is interpreted as

r 2 + (z + ıt)2 = ρeıv/2 , (5.6)
where

2zt
ρ= 4
(r 2 + z 2 − t 2 )2 + 4z 2 t 2 ; v = arctan (5.7)
r 2 + z2 − t 2
and ρ ≥ 0, 0 ≤ v < π. It can be verified that the function F(r, z, t) is harmonic in

the domain z ≥ 0 for all values of the real parameter t, and hence (5.4) defines a
harmonic function for all h(t).
Differentiating (5.5) with respect to z, we have
∂F 1 ∂2F 1 ∂ z + ıt
= ; = (5.8)
∂z r 2 + (z + ıt)2 ∂z 2 r ∂r r 2 + (z + ıt)2
and hence
a
∂ϕ a
h(t)dt ∂2ϕ 1 ∂ (z + ıt)h(t)dt
= ; = . (5.9)
∂z 0 r 2 + (z + ıt)2 ∂z 2 r ∂r 0 r 2 + (z + ıt)2
On the surface z = 0, these expressions reduce to

a min(r,a)
∂ϕ h(t)dt h(t)dt
(r, 0) = √ = √ (5.10)
∂z 0 r 2 − t 2 0 r2 − t2

a
∂ ϕ
2
1 d ıth(t)dt
(r, 0) = √ . (5.11)
∂z 2 r dr 0 r2 − t2
Now, if r > a, the integrand in (5.11) is imaginary for all t ∈ (0, a) and hence the
boundary condition (5.2) is satisfied identically for all values of the function h(t).
The remaining boundary condition (5.1), with (5.10) then gives
∗
r
h(t)dt E [g0 (r ) − Δ]
√ = 0 ≤ r ≤ a, (5.12)
0 r −t
2 2 2
5.1 Green and Collins Solution 65
which is an Abel integral equation for the unknown function h(t) with solution
∗
E d t
[g0 (r ) − Δ] r dr
h(t) = √ , (5.13)
π dt 0 t2 − r2
(Barber 2010, Sect. 30.2.4). Once the function h(t) has been determined, the potential
function ϕ can be found from (5.4) and the complete stress and displacement field is
then given by Eq. (A.3) of Appendix A. In particular, the contact pressure is obtained
from Eqs. (4.6), (5.11) as
a
1 d th(t)dt
p(r ) = √ 0≤r <a (5.14)
r dr r t2 − r2
=0 r >a (5.15)
and the surface displacement is

r
2 h(t)dt
u z (r ) = − ∗ √ 0 ≤ r < a, (5.16)
E 0 r2 − t2
a
2 h(t)dt
=− ∗ √ r > a, (5.17)
E 0 r2 − t2
from (4.7), (5.10).

The total contact force is
a a r =a a
th(t)dt
P = 2π r p(r )dr = 2π √ = −2π h(t)dt. (5.18)
0 r t2 − r2 r =0 0
We can also obtain a direct relationship between the contact force P and the gap
function g0 (r ) by substituting (5.13) into (5.18) giving

∗
a
g0 (r )r dr
P = 2E aΔ − √ . (5.19)
0 a2 − r 2
5.1.1 The Flat Punch Solution
A simple example is the indentation of an elastic half-space by a flat-ended cylindrical

rigid punch of radius a, for which g0 (r ) = 0 and
∗ ∗
E Δ d t
r dr E Δ
h(t) = − √ =− . (5.20)
π dt 0 t −r
2 2 π
Substitution into Eqs. (5.14), (5.18) then yields

∗
E Δ P ∗
p(r ) = √ = √ ; P = 2E aΔ. (5.21)
π a −r
2 2 2πa a 2 − r 2
Of course, these results can also be obtained by setting b = a, e = 0 in Eqs. (2.31),

(2.33), since K (0) = π/2. We also obtain the displacement outside the contact area as
2Δ a
u z (r ) = arcsin r > a, (5.22)
π r
from Eq. (5.17).
5.2 Non-conformal Contact Problems
If the contact problem is non-conformal, the contact radius a must be chosen to

satisfy the contact inequalities, and we argued in Sect. 1.2.4 that this is equivalent to
the condition that p(r ) → 0 as r → a.
Integrating by parts, we obtain
a a
th(t)dt
√ = h(a) a 2 − r 2 − t 2 − r 2 h (t) dt
r t2 − r2 r
and hence, using this result in (5.14),

h(a) a
h (t)dt
p(r ) = − √ + √ . (5.23)
a2 − r 2 r t2 − r2
The second term is bounded as r → a and hence the tractions will be bounded
[and will, in fact, tend to zero] if and only if h(a) = 0. Alternatively, we can use
Theorem 2 of Sect. 4.2 to argue that a must satisfy the condition ∂ P/∂a = 0, which
with Eq. (5.18) leads directly to the same condition, h(a) = 0.
Using this result, we can then write a simpler expression for the contact pressure
in non-conformal contact problems as
a
h (t)dt
p(r ) = √ . (5.24)
r t2 − r2
Also, integrating by parts and using the result g0 (0) = 0, we have

t
t
[g0 (r ) − d ] r dr
√ = t 2 − r 2 g0 (r )dr − tΔ (5.25)
0 t2 − r2 0
5.2 Non-conformal Contact Problems 67
and hence, from (5.13),

∗
E t
g0 (r )dr
h(t) = t √ −Δ . (5.26)
π 0 t2 − r2
We can determine the central indentation Δ by setting h(a) = 0 in (5.26), giving

a
g (r )dr
Δ=a √0 . (5.27)
0 a2 − r 2
Finally, we substitute (5.26), (5.27) into (5.18) to obtain, after some algebraic sim-
plifications, a 2
∗ r g0 (r ) dr
P = 2E √ . (5.28)
0 a2 − r 2
The relations between P, Δ and a can often be found quite easily, even in problems
where the determination of the contact pressure p(r ) is algebraically tedious.
The Axisymmetric Hertzian Problem
The axisymmetric Hertzian problem could be solved by taking the limit as e → 0 in
Eqs. (3.32)–(3.34), but the analysis of the present section gives a more direct method
of solution. If an elastic half-plane is indented by a spherical body of radius R, the
initial gap function is
r2 r
g0 (r ) = ; g0 (r ) = , (5.29)
2R R
and substitution in Eqs. (5.27), (5.28), gives immediately
∗ ∗
a2 4E a 3 4E R 1/2 Δ3/2
Δ= ; P= = . (5.30)
R 3R 3
For the contact pressure distribution, we first use Eq. (5.26) to obtain
∗ t ∗ 2 ∗
E t r dr E t 2E t
h(t) = √ −Δ = −Δ so h (t) = .
π R 0 t2 − r2 π R πR
(5.31)
Substitution in (5.24) then gives
∗√ 2
2E a − r2
p(r ) = . (5.32)
πR
Example: A Truncated Conical Indenter
Figure 5.1 shows an elastic half-space indented by a rigid punch in the form of a
truncated cone. The cone is of angle π/2 − α where α
1 and the radius of the
Fig. 5.1 The truncated

conical indenter P
α
b
a
truncated end is b. Find the relation between the applied force P, the indentation
depth d and the contact radius a.
The derivative of the gap function g0 (r ) for this case is
g0 (r ) = 0 ; 0 ≤ r b,
so, substitution into Eqs. (5.27), (5.28) immediately yields the results
a
dr b
Δ = αa √ = αa arccos
b a2 − r 2 a
a 2
∗ r dr ∗ b
P = 2E α √ = E α a 2 arccos + b a 2 − b2 .
b a2 − r 2 a
Notice that both integrals have lower limits of r = b, since g0 (r ) = 0 for 0 ≤r < b.
5.3 Annular Contact Regions
The preceding method is restricted to problems in which the contact area comprises
a single circular contact area. However, it can be extended to problems with one or
more annular contact areas using linear superposition (Collins 1963; Jain and Kanwal
1972). To illustrate this procedure, we consider the case of a flat-ended annular punch
for which the contact area comprises the annulus b <r < a, as shown in Fig. 5.2.
Fig. 5.2 Indentation by an P

annular flat punch
a b
5.3 Annular Contact Regions 69
The boundary conditions in terms of the potential function ϕ(r, z) are
∂2ϕ
(r, 0) = 0 0≤r a (5.33)
∂z 2
∂ϕ
(r, 0) = Δ b < r < a, (5.34)
∂z
where Δ is the indentation depth.

We now write
ϕ = ϕ0 + ϕ1 + ϕ2 , (5.35)
where ϕ0 (r, z) is the potential corresponding to the cylindrical flat punch solution of
Sect. 5.1.1, defined by Eqs. (5.4), (5.20). This function already satisfies the condition
(5.34) and the remaining boundary conditions can then be satisfied by imposing the
conditions
∂ϕ1
(r, 0) = 0 r >b (5.36)
∂z
∂ϕ2
(r, 0) = 0 0≤r <a (5.37)
∂z
∂ 2 ϕ1 ∂ 2 ϕ0 ∂ 2 ϕ2
(r, 0) = − (r, 0) − (r, 0) 0≤r a. (5.39)
∂z 2 ∂z
Notice that the ranges of (5.36), (5.37) overlap in the region b < r < a so condition
(5.34) is still satisfied by (5.35). Using the notation of Sect. 5.1 and Eq. (5.5), it is
easily verified that these two conditions are satisfied identically by the choices
b ∞
ϕ1 = F(r, z, t)h 1 (t)dt ; ϕ2 = F(r, z, t)h 2 (t)dt, (5.40)
0 a
where h 1 (t), h 2 (t) are any two unknown functions. Substituting these expressions
into (5.38), (5.39), we then obtain two simultaneous Abel integral equations for
h 1 (t), h 2 (t), and these can be reduced to a single Fredholm integral equation in
either of the two functions by Abel inversion and back substitution.
5.4 The Non-axisymmetric Cylindrical Punch
The methods described in this chapter can be generalized to non-axisymmetric prob-

lems, subject to the rather restrictive condition that the contact area remains circular.
This arises [for example] if an elastic half-space is indented by a cylindrical rigid
punch with a fairly general non-plane surface, and the force is sufficient to ensure
contact throughout the punch end face.
Using a result due to Copson (1947) in combination with the potential function
solution of Sect. A.1, it can be shown that the contact pressure corresponding to the
mixed boundary value problem
u z (r, θ) = f (r ) cos(mθ) 0 ≤ r < a

p(r, θ) = 0 r >a (5.41)
is a
d th(t)dt
p(r, θ) = r m−1
cos(mθ) √ 0 ≤ r < a, (5.42)
dr r t2 − r2
where ∗
E d t
r m+1 f (r )dr
h(t) = − √ . (5.43)
πt 2m dt 0 t2 − r2
Thus, if the [now non-axisymmetric] gap function g0 (r, θ) in (5.1) is expanded as

a Fourier series in θ, each term of this series will define a separate problem, for
which Eqs. (5.42), (5.43) will define the corresponding term in a Fourier series for
the contact pressure p(r, θ). It is easily verified that these equations reduce to (5.14),
(5.13) for the axisymmetric term m = 0. Methods of determining the complete stress
and displacement fields are discussed by Barber (2010), Sect. 30.3.1.
5.5 The Method of Dimensionality Reduction (MDR)
Figure 5.3 shows, a two-dimensional rigid flat punch of width 2a pressed into a linear
elastic ‘Winkler’foundation by a force P per unit length [perpendicular to the figure].
The modulus of the foundation k is defined such that the contact pressure
p(x) = ku(x), (5.44)
where u(x) is the local indentation displacement. It follows that the indentation depth
Δ in Fig. 5.3 is defined by the equation
P = 2kaΔ. (5.45)
We found in Sect. 5.1.1, Eq. (5.21), that if a cylindrical flat rigid punch of radius
a is pressed into an elastic half-space to a depth Δ, the resulting force is given by
∗
P = 2E aΔ, and this can be made equal to (5.45) if the modulus of the founda-
∗
tion is set equal to the composite modulus E . This observation prompted Popov
and others to explore whether other simple relations can be established between
5.5 The Method of Dimensionality Reduction (MDR) 71
Fig. 5.3 Indentation of a

P
Winkler foundation by a
rigid flat punch
Δ modulus k = E *
a a
three-dimensional elastic contact problems and problems involving a Winkler foun-

dation (Geike and Popov 2007a; Popov and Heß 2015).
Suppose that a more general two-dimensional rigid punch is defined by a gap
function g1 (x), where g1 (0) = 0. We assume that the punch is symmetrical, so
g1 (−x) = g1 (x), and that the function g1 (x) is monotonically increasing in x > 0. If
this punch is pressed into the foundation to a depth Δ, contact will occur at all points
where g1 (x) < Δ and this defines the region −a < x < a, where
g1 (a) = Δ. (5.46)
In the contact region, the local normal displacement u(x) and the contact pressure
p(x) will be given by
∗
u(x) = Δ − g1 (x) ; p(x) = E [Δ − g1 (x)] , (5.47)
and the total force is therefore

a a
∗
P=2 p(x)d x = 2E aΔ − g1 (x)d x . (5.48)
0 0
This expression will be identical to the axisymmetric force–displacement relation

(5.19) if a a
g0 (r )r dr
g1 (x)d x = √ , (5.49)
0 0 a2 − r 2
and this condition will be satisfied for all a (and hence all P) if we choose1

d x
g0 (r )r dr x
g (r )dr
g1 (x) = √ =x √0 , (5.50)
dx 0 x2 − r2 0 x2 − r2
since g0 (0) = 0.
1 The perspicacious reader will note the close similarity between the definition of the transformed
two-dimensional profile g1 (x) and the function h(t) of Eq. (5.13). In effect, the method of dimen-
sionality reduction can be viewed as the creation of a fictitious problem to aid in the solution of the
mathematical problem of Sects. 5.1 and 5.2, in much the same spirit as the moment area method for
the solution of beam deflection problems.
The condition (5.46) then implies that

a
g (r )dr
Δ = g1 (a) = a √0 , (5.51)
0 a2 − r 2
which is identical to the axisymmetric result (5.27).

We conclude that the relations between the normal force P, the indentation depth
Δ and the contact radius a for a non-conformal axisymmetric contact problem can
be obtained by
(i) using the transformation (5.50) to define the profile g1 (x) of an equivalent sym-
metrical two-dimensional punch, and then
(ii) solving the simpler problem in which this punch is pressed into a Winkler foun-
∗
dation of modulus E .
We should note, however, that the predicted two-dimensional contact pressure
p(x) is not directly related to the axisymmetric value p(r ), and the surface displace-
ment outside the contact area is not predicted by the two-dimensional theory. Also,
the exact equivalency applies only if the contact area comprises a single circle, rather
than (for example) one or more annuli. It can, however, be extended to the case of
a set of axisymmetric asperities (Geike and Popov 2007b), provided these are suffi-
ciently sparse to be regarded as independent (Greenwood and Williamson 1966). In
other words, if the off-diagonal elements in the matrices M, C in Eq. (4.32) are small
enough to be neglected.
Pohrt et al. (2012) applied the same technique to a quasi-fractal surface2 defined
only through its power spectral density (PSD). Clearly, such surfaces will generally
not satisfy the above conditions for the transformation to be exact, but the authors
argue that comparisons with numerical solutions for specific realizations of the profile
statistics justify the use of the method.
Problems
1. A rigid conical punch of cone angle π/2−α, (α

1) is pressed into the surface
of an elastic half-space, as shown in Fig. 5.4. Find the contact pressure p(r ), the
indentation depth Δ and the applied force P, all as functions of the radius a of the
contact area.
2. An elastic half-space is indented by an axisymmetric rigid punch with the power-
law profile g0 (r ) = Cr λ , so the displacement in the contact area is
u z (r ) = Δ − Cr λ ,
2 see Sect. 16.5.

Problems 73
Fig. 5.4 The conical

indenter P
α
a x
where C, λ are constants. Show that the indentation force P, the contact radius a
and the indentation depth Δ are related by the equation

2EΔa λ
P= .
(1 − ν 2 ) λ+1
3. A rigid flat punch has rounded edges, as shown in Fig. 5.5. The punch is pressed
into an elastic half-space by a force P. Assuming that the contact is frictionless, find
the relation between P, the indentation depth Δ and the radius a of the contact area.
4. The elastic half-space is loaded by a uniform pressure p0 inside the circle 0 ≤r < a,
the rest of the surface being unloaded. Use Eq. (5.14) to determine the appropriate
function h(t) and hence find the surface displacement u z (r ) both inside and outside
the loaded region.
5. The profile of a smooth axisymmetric frictionless rigid punch is described by the
power law
g0 (r ) = An r 2n ,
where n is an integer. The punch is pressed into an elastic half-space by a force P. Find
the indentation Δ, the radius of the contact area a and the contact pressure distribution
p(r ). Check your results by comparison with the Hertz problem of Eq. (5.30) and
give simplified expressions for the case of the fourth order punch
g0 (r ) = A2 r 4 .
Fig. 5.5 Flat punch with

P
rounded corners
rigid
radius R
b
a elastic
6. An elastic half-space is indented by a rigid cylindrical punch of radius a with a

concave spherical end of radius R a. Find the contact pressure distribution p(r )
and hence determine the minimum force P0 required to maintain contact over the
entire punch surface. Do not attempt to solve the problem for P < P0 .
7.
(i) By representing the function ϕ in the form
b
ϕ= F(r, z, t)h(t)dt,
0
determine the surface tractions [tensile and compressive] needed to establish the
displacement field
2
r2
u z (r ) = Δ 1 − 2 0≤r b.
(ii) The otherwise flat surface of a rigid punch of radius a contains a number of
small widely spaced concave dimples of radius b
a and depth Δ. Use your
result from part (i) to find the minimum force that must be applied to the punch
to ensure that contact is established throughout the surface.
8. A rigid axisymmetric punch has a Hertzian profile perturbed by a set of concentric

sinusoidal waves, such that the initial gap function is defined as
r2
g0 (r ) = + A [1 − cos(mr )] ,
2R
where (m A)
1.
Assuming that the contact area comprises a single circle of radius a, determine
the indentation depth Δ and the normal force P as functions of a. Use these results
to make a parametric plot of P as a function of Δ and comment on the nature of this
plot as A is increased. [This problem requires the use of Maple or Mathematica].
9. Find expressions analogous to (5.10), (5.11) for the surface values of the derivatives
of the function ϕ2 in Eq. (5.40). In particular, verify that condition (5.37) is satisfied
for all functions h 2 (t).
10. Use the method of Sect. 5.4 to determine the contact pressure distribution under a
flat-ended cylindrical punch of radius a loaded by a force P applied through the point
(c, 0). Also, find the angle of tilt of the punch. Assume that the entire flat surface of
the punch makes contact. What is the maximum value of c for which this assumption
is correct?
Problems 75
11. A flat-ended rigid cylindrical punch of radius a is pressed into the curved surface
of an elastic cylinder of radius R a by a force P. Use the method of Sect. 5.4 to
find the contact pressure distribution for the case where P is sufficient to ensure that
the entire flat surface of the punch makes contact.
12. Use the method of dimensionality reduction to find the relations between the
indentation Δ, the radius of the contact area a and the applied force P for the power-
law punch of Problem 5.
Chapter 6
Two-Dimensional Frictionless Contact
Problems
If the gap function g0 (x, y) defined in Eq. (1.1) and Fig. 1.3 is independent of y,
contact will occur in one or more strips parallel to the y-axis and the resulting stress
and displacement fields will also be independent of y and hence two dimensional. In
fact, these fields will be of the form known as plane strain in the theory of elasticity,
meaning that the three strains e yx , e yy , e yz are everywhere zero. If the bodies are
actually of finite extent in the y-direction, we should expect some deviation from
plane strain conditions near the edges, but these edge effects will be localized in a
region comparable in dimension to the width of the contact area. Thus, the plane
strain assumption is a reasonable approximation when the bodies are long in the
y-direction, relative to the expected width of the contact area.
Approximately two-dimensional conditions can also arise at the opposite extreme,
when the contacting bodies are extremely short in the y-direction, in which case the
plane stress conditions are appropriate, obtained by assuming that the three stresses
σ yx , σ yy , σ yz are everywhere zero. However, the standard of comparison here is the
width of the contact area, and this is generally small compared with other dimensions
at least in non-conformal contact problems, so the conditions for plane stress contact
are actually rather seldom met. Here we shall restrict attention to the plane strain
case, noting that solutions for plane stress can always be obtained by making the
substitutions
ν E(1 + 2ν)
ν→ ; E→ (6.1)
(1 + ν) (1 + ν)2
in the solution for plane strain.1
1 These relations are inverse to those given in Eq. (3.18) of Barber (2010).
https://doi.org/10.1007/978-3-319-70939-0_6
78 6 Two-Dimensional Frictionless Contact Problems
Fig. 6.1 Concentrated P

normal force P per unit
length applied to the surface x
of the half-plane
x
uz
θ r
6.1 The Line Force Solution
As in Sect. 2.2.1, the starting point for the solution of the contact problem is to
determine the stress and displacement fields due to a concentrated normal force
on the boundary, as shown in Fig. 6.1. Notice however that since the problem and
hence the boundary conditions are assumed to be independent of the y-coordinate
perpendicular to the figure, the force P is here to be interpreted as a force per unit
length [along the y-axis]. Also, since all calculations will be confined to the zx-plane,
the body z > 0 shown in the figure is generally referred to as the half-plane.
The solution of the elasticity problem illustrated in Fig. 6.1 is readily obtained
using the Airy stress function2 and the resulting stress field takes a particularly
simple form3 in polar coordinates (r, θ) given by
P cos θ
σrr = − ; σr θ = σθθ = 0. (6.2)
πr
An expression for the corresponding normal surface displacements can be obtained
by distributing point forces uniformly along the line x = z = 0, using Eq. (2.7). If
we restrict these forces to the line segment −b < y < b, we obtain
√
P(1 − ν 2 ) b
dy 2P(1 − ν 2 ) b2 + x 2 + b
u z (x) = = ln . (6.3)
πE −b x 2 + y2 πE |x|
Unfortunately, this expression is unbounded as b → ∞. In other words, if the infinite

extremities of a half-space are prevented from moving and a uniform force per unit
length is applied along the complete y-axis, the displacement throughout the finite
domain will theoretically be infinite. This should not surprise us too much. After all,
if a semi-infinite bar as fixed at x = ∞ and a tensile force is applied at x = 0, the
resulting strain will be bounded, but the extension and hence the displacement at all
points other than at infinity will be unbounded.
2 Barber (2010), Chap. 12.

3 We shall show in Sect. 14.5.2 that the results σrr = P f (θ)/r, σr θ = σθθ = 0 can be deduced solely
from self-similarity and equilibrium arguments, without reference to the constitutive law, provided
only that it be linear. Thus, these results apply to generally anisotropic materials and to power-law
graded materials, though the function f (θ) will not then generally be a cosine.
6.1 The Line Force Solution 79
We did not encounter this problem in three dimensions, because the stress field
then varies with R −2 and, since displacements are obtained by integrating strains,
the displacements vary with R −1 which tends to zero at infinity. By contrast, in
two dimensions the stresses and strains vary with r −1 [see, for example, Eq. (6.2)]
and hence the displacements are logarithmically unbounded relative to the point at
infinity.
Of course, no real bodies are infinite, but one consequence of this result is that the
normal contact compliance in a two-dimensional problem depends logarithmically
on the finite size of the body, even when the contact area is small compared with the
other linear dimensions. For example, if we push a rigid punch a given distance d
into an elastic body, the resulting contact force and the associated contact pressure
distribution will depend on the finite size of the body. By contrast, if instead, we
impose a prescribed force on the punch, the resulting rigid-body indentation will
depend on the size of the body, but the contact pressure will not be, provided the
finite dimensions of the body are large compared with those of the contact area.
We shall discuss methods for estimating the influence of the size and shape of
the contacting bodies on contact compliance in Sect. 6.7 below. Until then we shall
restrict attention to problems where the contact force is prescribed, in which case,
the half-space (half-plane) geometry can be retained, resulting in considerable sim-
plification in the mathematics. We shall demonstrate two ways to do this.
Limiting form of Eq. (6.3)
We know from previous chapters that the solution (6.3) is needed only for points
within the contact area, for which we can assume that x b. We can, therefore
expand the equation in this limit, obtaining

2P(1 − ν 2 ) 2b 2P(1 − ν 2 )
u z (x) ≈ ln = ln(2b) − ln |x| . (6.4)
πE |x| πE
When this expression is used in the formulation of a contact problem, the term ln(2b)
will generate a displacement that is independent of x. This will affect the resulting
value of the indentation depth Δ, but will have no other effect on the solution. Thus,
if Δ is not prescribed, and we are not seeking to calculate it, we can drop the ln(2b)
term and use the simpler expression
2P ln |x|
u z (x) = − , (6.5)
πE∗
where we have generalized the result to two deformable materials by using the
∗
composite modulus E , as in Sect. 2.2.3. Notice that ln |x| → −∞ as |x| → 0, so a
positive displacement is produced under and near the force.
Formulation in Terms of Surface Slopes
The argument of a logarithm must be dimensionless, so Eq. (6.5) strictly requires
that the problem be stated in suitably normalized coordinates. However, the choice
of length scale affects only the rigid-body displacement Δ and hence if we temporar-
ily suspend mathematical rigour, it will be found that any logarithms appearing in
the solution for other physical quantities [such as relative displacements, strains or
contact tractions] have dimensionless arguments.
An alternative approach avoiding this contrivance is to argue that the gap function
g(x) is defined by (i) the rigid-body displacement and (ii) the derivative of the gap
function g0 (x), since g0 (0) = 0. We shall show that it is possible to formulate the
contact problem in terms of the corresponding derivative
du z 2P
=− ∗ , (6.6)
dx πE x
which is bounded and independent of b in the limit b → ∞.
6.2 Integral Equation Formulation
Suppose now that we have a contact pressure distribution p(x) applied in some region
x ∈ A. Applying a convolution on (6.6) analogous to that in Sect. 2.3, we can obtain
the surface slope as
du z 2 p(ξ)dξ
=− ∗ . (6.7)
dx π E A (x − ξ)
Differentiating (1.4) with respect to x we obtain
du z
= −g0 (x) x ∈A (6.8)
dx
and hence, using (6.7),
∗
p(ξ)dξ π E g0 (x)
= x ∈ A. (6.9)
A (x − ξ) 2
If the contact region A comprises a single strip b < x < a, we have

∗
a
= b < x < a, (6.10)
b (x − ξ) 2
which is a Cauchy singular integral equation for the unknown pressure distribution
p(x) (Muskhelishvili 1953). Solutions of this and related singular integral equations
are given in Appendix C. In particular, the solution of (6.10) is

∗ a √
1 E (ξ −b)(a −ξ) g0 (ξ)dξ
p(x) = √ P− b < x < a,
π (x −b)(a −x) 2 b (x − ξ)
(6.11)
where a
P= p(x)d x (6.12)
b
is the total normal contact force. Since x lies in the range of integration in Eqs. (6.10)
and (6.11), the factor (x − ξ) passes through zero and hence defines a non-integrable
singularity in the integrand. We must therefore interpret all integrals of this form in
the sense of principal values, so that, for example,
a
x− a
p(ξ)dξ p(ξ)dξ p(ξ)dξ
is replaced by lim + . (6.13)
b (x − ξ) →0 b (x − ξ) x+ (x − ξ)
6.2.1 Edge Conditions
Singular Solution
If an elastic half-plane is indented by a rigid indenter with a given force P sufficient to
cause the entire planform of the indenter to make contact as in Fig. 6.2a, the resulting
contact pressure is given by Eq. (6.11) and it will generally exhibit a square-root
singularity at x = a and x = b because of the premultiplier in this equation.
(a) (b) P
P
a b a
b
(c)
P
b a
Fig. 6.2 The contact area is determined by the planform of the indenter in case (a), but the point
x = b in case (b) and both points x = a, b in case (c) are determined by the requirement that the
contact pressure there be bounded
Indentation by a Flat Rigid Punch

In the special case where the indenter is flat, g0 (x) = 0 and we obtain
P
p(x) = √ b < x < a. (6.14)
π (x − b)(a − x)
One Edge Singular

However, if the force is insufficient to cause full contact as in Fig. 6.2b, c, the contact
pressure must be bounded at the appropriate edge as discussed in Sect. 1.2.4.
In Fig. 6.2b, the point x = b is still determined by the planform of the indenter,
but the point x = a must be determined from the condition that p(x) → 0 as x → a.
Imposing this condition on Eq. (6.11), we obtain
∗ √
E a
(ξ − b)(a − ξ) g0 (ξ)dξ
P− = 0, (6.15)
2 b (a − ξ)
or equivalently
∗
E a
ξ−b
P= g (ξ)dξ. (6.16)
2 b a−ξ 0
This equation defines the value of force P needed to establish contact in the range
b < x < a, but once this has been determined, the relation can be inverted to give b
as a function of P. Notice that this integral is not singular, so we do not need to have
recourse to principal values.
Example
Figure 6.3 shows a punch with a plane face inclined at an angle α( 1), so that
g0 (x) = α.
If the force P is insufficient to ensure contact over the whole punch face, we
conclude from Eq. (6.16) that

∗ ∗
E α a
ξ−b π E α(a − b) 4P
P= dξ = or a =b+ . (6.17)
2 b a−ξ 4 π E ∗α
Fig. 6.3 A rigid punch with P

an inclined flat face in partial
contact
α
b a
Thus, the extent of the contact region increases linearly with P and the mean pressure
P/(a −b) is independent of P and proportional to α.
Both Edges Bounded
In the case of Fig. 6.2c, the contact pressure must tend to zero at both edges of the
contact area, so in addition to Eq. (6.16) we require a similar condition at x = b
which is
∗ a
E a−ξ
P=− g (ξ) dξ. (6.18)
2 b ξ−b 0
Eliminating P between (6.16) and (6.18) and cancelling a non-zero multiplying

factor, we obtain a
g (ξ) dξ
√ 0 = 0, (6.19)
b (ξ − b)(a − ξ)
which is known as the consistency condition. Also, a more ‘symmetric’ expression

for the applied force P can be obtained by combining Eqs. (6.16) and (6.18) as
∗
E a
ξ g0 (ξ) dξ
P= √ . (6.20)
2 b (ξ − b)(a − ξ)
Equations (6.19) and (6.20) can be used to simplify the contact pressure (6.11) for
the bounded-bounded case, giving4
∗√
E (x −b)(a −x) a
g0 (ξ) dξ
p(x) = − √ b < x < a, (6.21)
2π b (x − ξ) (ξ −b)(a −ξ)
Notice that if the gap function g0 (x) is symmetric [i.e. an even function of x],
the consistency condition is identically satisfied if the contact area is also chosen to
be symmetric, and hence b = −a. The semi-width of the contact area a for a given
normal force can then be determined from the equation
∗
E a
ξ g0 (ξ) dξ
P= . (6.22)
2 −a a2 − ξ2
Two-Dimensional Hertzian Contact

In two-dimensional Hertzian contact, the initial gap function has the form
x2 x
g0 = so g0 (x) = , (6.23)
2R R
4 This result can also be derived from Eq. (C.6) and (C.9) of Appendix C.
Fig. 6.4 Flat Punch with

P
Rounded Corners
rigid
radius R
b
a elastic
where R is the composite radius defined as in Eq. (3.18). Substitution in (6.22) gives
∗ ∗
E a
ξ 2 dξ π E a2
P= = (6.24)
2R −a a2 − ξ2 4R
and the pressure distribution can be obtained from (6.21) as

∗√ 2 √
E a − x2 2P a 2 − x 2
p(x) = = . (6.25)
2R πa 2
Example: Rounded Flat Punch
Figure 6.4 shows a rigid flat punch with rounded corners, so that the derivative of the
initial gap function is defined by
g0 (x) = 0 −b < x b (6.26)
R
(x + b)
= x < −b.
R
When no force is applied (P = 0), contact will occur only in the flat portion −b < x <
b, but for P > 0 there will be contact in a region −a < x < a, where a > b. Substituting
(6.26) into (6.22) and noting that by symmetry the contributions from the segments
−a < x < −b and b < x < a will be equal, we obtain
∗ ∗

E a
ξ(ξ − b) dξ E b
P= = a 2 arccos − b a 2 − b2 , (6.27)
R b a2 − ξ2 2R a
which defines implicitly the relation between the unknown contact dimension a and
the applied force P.
6.3 Incremental Solution of Non-conformal Contact Problems 85
6.3 Incremental Solution of Non-conformal Contact

Problems
If at some stage in the loading process the normal force is P and the contact area is
defined by b < x < a, then increasing P by an infinitesimal increment δ P will cause
equal (infinitesimal) displacements throughout b < x < a. The incremental problem
is therefore similar in form to the flat punch problem and we conclude from Eq.
(6.14) that the increment in the contact pressure distribution will be
δP
δ p(x) = √ b < x < a. (6.28)
π (x − b)(a − x)
If the contact is non-conformal, we also anticipate that a and b will change during
this incremental process, but these changes will themselves be infinitesimal, so any
tractions in the newly established regions of contact will be second order small
quantities and can be neglected. This concept can be used to formulate non-conformal
contact problems as the superposition of a sequence of infinitesimal increments.5 One
advantage of the method is that it leads to Abel integrals, which are generally easier
to evaluate than Cauchy singular integrals.
6.3.1 Symmetric Problems
If the problem is symmetric so that b = −a, the force P can be determined as a

function of a from Eq. (6.22), and we can then determine the function
∗
dP E d a
ξg (ξ) dξ ∗ d
a
ξg (ξ) dξ
F(a) ≡ = 0 =E 0 , (6.29)
da 2 da −a a −ξ
2 2 da 0 a2 − ξ2
where the last expression follows from the fact that the integrand is an even func-
tion of ξ. The contact pressure distribution is then obtained by superposition of the
distribution (6.28) as
P(a) a
dP 1 F(s) ds
p(x) = = √ , (6.30)
P(x) π s(P)2 − x 2 π x s2 − x 2
where s(P) is the semi-width of the contact area when the applied force is P, and
P(s) is its inverse relation — i.e. the normal force at which the contact semi-width is
equal to s. Notice that the lower limit of the second integral is x, since contributions
to p(x) are made only when s > x.
5 This method was used by Segedin (1957) to solve a non-conformal axisymmetric contact problem.
The relative simplicity of this procedure is exemplified by the Hertzian problem,

∗
for which we found P(a) = π E a 2 /4R in Eq. (6.24). It follows immediately that
∗ ∗ ∗√ 2
d P πE a E a
sds E a − x2
F(a) = = and hence p(x) = √ = ,
da 2R 2R x s2 − x 2 2R
(6.31)
which agrees with (6.25) and does not require the evaluation of a Cauchy singular
integral.
6.3.2 Bounded-Singular Problems
Suppose that one edge of the contact region is determined by a sharp corner in the
punch, but the other varies with the applied force, as in Fig. 6.2b. The force is then
given by Eq. (6.16) and hence

∗
dP E d a
ξ−b
F(a) ≡ = g (ξ) dξ. (6.32)
da 2 da b a−ξ 0
A superposition similar to that in (6.30) then yields the pressure distribution

a
1 F(s) ds
p(x) = √ . (6.33)
π x (x − b)(s − x)
For example, for the inclined flat punch of Fig. 6.3, we obtain
∗
dP πE α
F(a) = = (6.34)
da 4
from (6.17) and hence
∗ ∗
E α a
ds E α a−x
p(x) = √ = , (6.35)
4 x (x − b)(s − x) 2 x −b
from (6.33).
6.4 Solution by Fourier Series
In this section, it is convenient to move the origin of coordinates to the mid-point of

the contact area, so that Eq. (6.10) takes the form
∗
a
= − a < x < a. (6.36)
−a (x − ξ) 2
6.4 Solution by Fourier Series 87
with 2a being the width of the contact area. Notice however that we shall not require
that the problem be symmetric in this coordinate system.
The change of variable
ξ = a cos θ; x = a cos φ (6.37)
then permits us to write

π ∗
p(θ) sin θdθ π E dg0 (φ)
=− 0 < φ < π. (6.38)
0 (cos φ − cos θ) 2a sin φ dφ
If we expand the two sides of this equation in the form of Fourier series, such that
∞
∞
dg0 (φ)
p(θ) sin θ = pn cos(nθ); = gn sin(nφ), (6.39)
n=0
dφ n=1
and equate coefficients using the result

π
cos(nθ)dθ π sin(nφ)
=− 0 < φ < π, (6.40)
0 (cos φ − cos θ) sin φ
we obtain ∗
E gn
pn = n = 0. (6.41)
2a
Notice that the integral (6.40) evaluates to zero for n = 0 and hence the coefficient
p0 cannot be determined from (6.41). Instead, we write the total applied force as
a π ∞
π
P= p(x)d x = a p(θ) sin θdθ = a pn cos(nθ)dθ = πap0 (6.42)
−a 0 n=0 0
and hence
P
p0 = . (6.43)
πa
These equations provide a general solution to the contact problem provided the gap
function g0 can be expanded as a Fourier series, as in Eq. (6.39).
6.4.1 Rigid-Body Rotation
In the preceding analysis, we have assumed that the bodies are free to approach each
other in the normal direction under the influence of the normal force P, but relative
rigid-body rotation is prevented. If instead rigid-body rotation is not constrained, and
if [for example] a flat rigid punch is loaded away from the mid-point, it will tend to
(a) P (b) P
α α
Fig. 6.5 (a) An unconstrained rigid flat punch loaded by an off-centre force; (b) a rotationally
constrained punch with an inclined flat face
rotate through some angle α, as shown in Fig. 6.5a. However, from a mathematical
point of view, the resulting contact problem is identical to that shown in Fig. 6.5b,
where a punch with a flat face inclined at α to the horizontal is pressed into a half-
plane with rotation constrained.
We first note that if the punch is unconstrained, an additional equilibrium condition
must be satisfied. For example, if the contact area is defined by −a < x < a and the
force P in Fig. 6.5a passes through the point x = −c, the contact pressure distribution
must satisfy the condition
a π
− Pc = p(x)xd x = a 2
p(θ) sin θ cos θdθ. (6.44)
−a 0
Using the representation (6.39)1 , we then have

∞
π
πa 2 p1
− Pc = a 2 pn cos θ cos(nθ)dθ = , (6.45)
n=0 0 2
and hence
2Pc
p1 = − . (6.46)
πa 2
To find the rigid-body rotation α( 1) in the problem of Fig. 6.5a, we note that it
is equivalent to an additional term αx in g0 (x) [as in the problem of Fig. 6.5b] and
hence to
dg0 (φ)
g0 (φ) = αa cos φ; = −αa sin φ. (6.47)
dφ
Using (6.39)2 , we then have
g1 2 p1 4Pc
g1 = −aα and hence α=− =− ∗ = , (6.48)
a E π E ∗a2
from (6.41) and (6.46).
For more general punch shapes, we can replace g0 (x) by g0 (x)+αx, solve the
resulting constrained contact problem, and finally determine α from Eq. (6.46).
6.4.2 Galin’s Theorem, Chebyshev Polynomials and

Recurrence Relations
A two-dimensional strip of contact can be regarded as the limit e → 1 of the condi-

tions across the minor axis in a three-dimensional elliptic contact and hence we can
conclude from Galin’s theorem [Sect. 2.4] that a traction distribution of the form
Pn (x)
p(x) = √ , (6.49)
a2 − x 2
will produce a normal surface displacement u z (x) = Q n (x), where Pn (x), Q n (x) are
polynomials of degree n.
This result is conveniently expressed in terms of the Chebyshev polynomials of
the first and second kind, defined as

sin (n + 1)θ
Tn (cos θ) = cos(nθ); Un (cos θ) = , (6.50)
sin θ
respectively. These are easily shown by trigonometric expansion to be polynomials
of degree n. Alternatively, the Chebyshev polynomials can be defined using the
recurrence relations
Tn+1 (x) = 2x Tn (x) − Tn−1 (x); Un+1 (x) = 2xUn (x) − Un−1 (x), (6.51)
with initial values
T0 (x) = 1; T1 (x) = x; U0 (x) = 1; U1 (x) = 2x. (6.52)
Using the change of variable inverse to (6.37) on Eq. (6.40), we then obtain
x
a
Tn (ξ/a)adξ
= −πUn−1 −a < x <a (6.53)
−a (x − ξ) a 2 − ξ 2 a
and hence, more generally, if the function Pn (x) in (6.49) is expanded as a finite
series of Chebyshev polynomials so that
a n x
p(x) = √ pk Tk − a < x < a, (6.54)
a 2 − x 2 k=0 a
the surface slope from Eq. (6.7) will be given by
du z 2
n x
= ∗ pk Uk−1 − a < x < a. (6.55)
dx E k=1 a
Notice incidentally that the right-hand side of this equation is a polynomial of degree
(n−1) rather than n because it is the x-derivative of the nth degree polynomial u z (x).
In two-dimensional problems for more complex domains, we shall encounter
integral equations with more complex kernels, but retaining the Cauchy singu-
larity (x −ξ) in the denominator. Generally, these require numerical solution, but
Chebyshev polynomials play an important rôle in this procedure (Erdogan and Gupta
1972; Krenk 1975a, b). For example, if we wish to approximate the contact pressure
distribution by a finite series of order n, as in Eq. (6.54), it is generally sufficient to
use the collocation method with the zeros of Tn+1 (x/a) serving as collocation points.
These zeros tend to be crowded near the end points x = ±a. Equivalently, if (6.37)
is used to cast the problem in the θ-domain, equal-spaced points should be used in
the range 0 < θ < π.
An alternative but related approach is to express the polynomial Pn (x) in (6.49)
in the power series form

n
n
pm x m
Pn (x) = pm x m or p(x) = √ − a < x < a, (6.56)
m=0 m=0 a2 − x 2
in which case Eq. (6.10) gives

∗
π E g0 (x)
n
= pm Im (x) − a < x < a, (6.57)
2 m=0
where we define a
ξ m dξ
Im (x) = . (6.58)
−a (x − ξ) a 2 − ξ 2
Using the identity
ξ = x − (x − ξ) so ξ m = xξ m−1 − (x − ξ)ξ m−1 (6.59)
we obtain the recurrence relation

π
ξ m−1 dξ a
Im (x) − x Im−1 (x) = − = −a m−1
cosm−1 θ dθ
−a a2 − ξ2 0
(m − 2)!!πa m−1
=− m odd (6.60)
(m − 1)!!
=0 m even, (6.61)
where
(2k)!! = 2.4.6....(2k); (2k − 1) = 1.3.5...(2k − 1). (6.62)
Expressions for any value of m can be obtained by applying this relation sequentially,
starting from the result
I0 (x) = 0 −a < x <a (6.63)

π sgn(x)
= √ |x| > a. (6.64)
x 2 − a2
It is readily verified that in the range −a < x < a, Im (x) is a polynomial of degree
(m −1) in x, and hence if g0 (x) is a known polynomial, the solution can be obtained
from (6.57) by equating coefficients. This method has the advantage of providing
a closed form solution for the displacements outside the contact area, using the
expressions developed from (6.64).
6.5 Periodic Contact Problems
So far we have assumed that the contact area is connected, meaning that it comprises
a single segment of the surface of the plane, but in many applications, notably in the
contact of rough surfaces, contact may occur in an array of unconnected strips. A
simple example is the periodic contact problem illustrated in Fig. 6.6, where a rigid
body with a sinusoidal profile is pressed into an elastic half-plane. This problem was
solved by Westergaard (1939) and we give the solution in Sect. 6.5.6 below, but first
we discuss some more general features of problems where the initial gap function
g0 (x) is periodic.
6.5.1 Sinusoidal Contact Pressure
We first consider the case where the half-plane is subjected to the sinusoidal contact
pressure
p(x) = p0 cos(ωx). (6.65)
The corresponding elasticity problem can be solved by assuming that the two
harmonic potential functions in Appendix A, Sect. A.3 have the form
φ = C exp(−ωz) cos(ωx); ψ = D exp(−ωz) cos(ωx), (6.66)
where C, D are two constants. It is easily verified that these functions are harmonic,
and all the stress and displacement components will vary sinusoidally with x. The
constants can, therefore be determined from the boundary conditions
σzz (x, 0) = − p(x) = − p0 cos(ωx); σzx (x, 0) = 0. (6.67)

Fig. 6.6 Indentation of an p

elastic half-plane by a rigid
body with a sinusoidal
surface profile rigid L
h0 a
x
elastic
In particular, we find that the normal surface displacment and surface slope are given
by
2 p0 cos(ωx) ∂u z 2 p0 sin(ωx)
u z (x, 0) = ∗ ; (x, 0) = − (6.68)
E ω ∂x E∗
respectively. Notice that the amplitude of the surface deformation is inverse with
ω, so the surface appears stiffer to high wavenumber [short wavelength] pressure
profiles. This result has important implications for the contact of rough surfaces, as
we shall see in Chap. 16.
6.5.2 Fourier Series Methods
More generally, if we write the contact pressure distribution in the form of a Fourier
series
∗ ∞
E
p(x) = p̄ + Cn cos(nωx) + Dn sin(nωx) , (6.69)
2 n=1
the corresponding surface slope will be
∞
du z
= − Cn sin(nωx) + Dn cos(nωx) . (6.70)
dx n=1
If the surface is periodic, with wavelength L as in Fig. 6.6, the fundamental

wavenumber ω is
2π
ω= . (6.71)
L
The contact condition (6.8) then gives
6.5 Periodic Contact Problems 93
∞

− Cn sin(nωx) + Dn cos(nωx) = −g0 (x) x ∈ A, (6.72)
n=1
and in the separation region, the contact pressure p(x) = 0, so

∗ ∞
E
p̄ + Cn cos(nωx) + Dn sin(nωx) = 0 x∈
/ A. (6.73)
2 n=1
Equations (6.72) and (6.73) define a pair of dual series equations from which the
constants Cn , Dn can be determined. This method was used by Dundurs et al. (1973)
for the problem of Fig. 6.6.
6.5.3 The Periodic Green’s Function
An alternative approach, leading to a Cauchy singular integral equation, is to start

from a Green’s function comprising a periodic set of equal concentrated normal
forces P separated by distances L. If these forces are located at x = ±n L where n is
any integer, we can write down the displacement derivative by superposition of (6.6)
as
du z 2P
∞
1 2P πx
=− ∗ = − ∗ cot , (6.74)
dx π E n=−∞ (x + n L) E L L
Block and Keer (2008). The same procedure as in Sect. 6.2 then shows that the
pressure distribution is defined by the equation

∗
1 π(x − ξ) E g0 (x)
cot p(ξ)dξ = x ∈ A, (6.75)
L A L 2
where A here denotes the contact area in a single segment of length L of the periodic
profile.
Equation (6.75) has a Cauchy singularity at ξ = x and clearly reduces to (6.9)
in the limit L → ∞. It can be reduced to a form similar to (6.10) by the change of
variable
L L
ξ = arctan(u); x = arctan(v), (6.76)
π π
Schmueser and Comninou (1979); Block and Keer (2008) used this formulation to
obtain closed-form solutions to a range of non-conformal periodic contact problems.
6.5.4 The Cotangent Transform
Manners (1998) defines the cotangent transform C { f (x)} such that


1 L/2
π(x − s)
g(s) = C { f (x)} = cot f (x)d x, (6.77)
L −L/2 L
where f (x) is a periodic function of x with period L and mean value zero. The
function f (x) can be expanded as a Fourier series
∞

f (x) = Cn cos(nωx) + Dn sin(nωx) , (6.78)
n=1
and substitution in (6.77) and evaluation of the intgegral then gives

∞

C { f (x)} = − Cn sin(nωs) + Dn cos(nωs) . (6.79)
n=1
In other words, C { f (x)} can be obtained from the Fourier series form (6.78) simply
by making the substitutions
cos → − sin; sin → cos (6.80)
in each term of the series. Applying this operation twice, we conclude that
C {C { f (x)}} = − f (x), (6.81)
or equivalently, the inverse transform is
C −1 {g(x)} = − C {g(x)} . (6.82)
With the notation (6.77), the contact condition (6.75) can be written
C {F(x)} = −g0 (s) s ∈ A, (6.83)
where the function F(x) is defined by

∗ L/2
E F(x) 1
p(x) = p̄ + and p̄ = p(x)d x (6.84)
2 L −L/2
is the mean contact pressure.
6.5.5 Manners’ Solution
For general periodic functions g0 (x), there may be more than one contact segment in
each period L. A very elegant solution of problems of this class is given by Manners
(1998), based on the technique used by Westergaard (1939).
6.5 Periodic Contact Problems 95
We first extend the idea introduced in Eq. (2.42), by defining a dimensionless real
function M(x) for all x such that
2 p(x)
M(x) = ∗ ; M(x) = g (x), (6.85)
E
where g(x) = g0 (x)−d +u z (x) is the gap in the loaded state and hence
du z
g (x) = g0 (x) + . (6.86)
dx
Notice that with this definition, we automatically satisfy the contact conditions
g (x) = 0 x ∈A p(x) = 0; x∈
/ A, (6.87)
where the contact region A is identified with those points at which M(x) > 0. Fur-
thermore, this representation automatically satisfies the asymptotic conditions that
the pressure and the gap derivative be square-root bounded at the edge of the contact
region, as long as M(x) is a continuous and differentiable function of x.
In the special case where no force is applied, p(x) = 0, g(x) = g0 (x) and hence
2
M(x) = − g0 (x) . (6.88)
Manners (1998) shows that for more general periodic problems, if the undeformed
profile
N
g0 (x) = C0 + Ck cos(kωx) + Dk sin(kωx), (6.89)
k=1
is truncated at N terms, M(x) can be expressed as a finite Fourier series as
N −1

2 4 p̄
M(x) = − g0 (x) + ∗ C g0 (x) + A0 + Ak cos(kωx) + Bk sin(kωx),
E k=1
(6.90)
where the cotangent transform C{·} is defined in Sect. 6.5.4. The coefficients
A0 , Ak , Bk must be determined from the conditions that (i) there exists at least one
interior point in each separation region at which g (x) = 0 and hence M(x) = M (x) =
0, and (ii) that in each separation region Sk

g (x)d x = 0, (6.91)
Sk
since otherwise the condition g (x) = 0 in A would not be sufficient to ensure that
g(x) = 0 in all contact regions.
Manners gives solutions to two examples involving simple two-term series, but
describes a numerical procedure for handling the calculations in more complex cases
in a later paper Manners (2003).
6.5.6 Westergaard’s Problem
For the simple sinusoidal profile of Fig. 6.6, the initial gap function is

2πx 2πh 0 2πx
g0 (x) = h 0 − h 0 cos and hence g0 (x) = sin . (6.92)
L L L
Thus N = 1 in Eq. (6.89), and the only unknown constant in (6.90) is A0 . Substituting
(6.92) into (6.90) and using (6.80) in the second term, we obtain

4π 2 h 20 2πx 8π p̄h 0 2πx
M(x) = − sin2 + cos + A0 , (6.93)
L2 L E∗L L
and A0 is obtained from the condition that, by symmetry, g (L/2) = 0, and hence
M(L/2) = 0. Solving for A0 and substituting back into (6.93), we obtain

4π 2 h 20 2πx 2 p̄L 2πx
M(x) = 1 + cos − 1 + cos
L2 L π E ∗h0 L
16π 2 h 20 πx
p̄L πx
= cos2 − sin2 . (6.94)
L2 L π E ∗h0 L
The contact semi-width a in Fig. 6.6 is determined from the condition that M(x) > 0
in the contact region, so
πa p̄L p̄
sin2 = = ∗, (6.95)
L π E ∗h0 p
where ∗
π E h0
p∗ = (6.96)
L
is the value of p̄ that is just sufficient to ensure full contact. The contact pressure in
−a < x < a can then be written
E
∗ ∗
2π E h 0 πx πa πx
p(x) = M(x) = cos sin2 − sin2
2 L L L L

πx p̄ πx
= 2 p ∗ cos − sin2 . (6.97)
L p∗ L
6.6 The Smirnov–Sobolev Transform 97
6.6 The Smirnov–Sobolev Transform
The relative simplicity of two-dimensional problems suggests the possibility of gen-

erating three-dimensional solutions by superposing two-dimensional fields with dif-
fering orientations. We define a fixed Cartesian coordinate system (x, y, z) such that
the half-space is defined by z > 0. We then define a second system (x1 , y1 , z), rotated
through an angle φ relative to the first, as shown in Fig. 6.7, so that
x1 = x cos φ + y sin φ; y1 = y cos φ − x sin φ. (6.98)
Now suppose that a two-dimensional field exists, such that a specific scalar field
quantity f (x1 , z) is independent of y1 . A more general three-dimensional field can
then be constructed by superposing such fields with different values of φ. The most
general such function is
π/2
g(x, y, z) = f (x1 , z, φ) dφ, (6.99)
−π/2
Sveklo (1964). Notice that we have introduced φ as a parameter into this equation,
since the superposed two-dimensional fields of different orientations may be differ-
ent. Also, it is not necessary to include the full range (0, 2π), since a rotation through
π transforms the coordinate system into itself except for a sign change.
If we define cylindrical polar coordinates (r, θ, z) through x =r cos θ, y =r sin θ,
Eq. (6.99) takes the form
π/2
g(r, θ, z) = f (r cos(φ − θ), z, φ) dφ. (6.100)
−π/2
In the special case where f is independent of φ, all the rotated two-dimensional fields
are the same and equal to f (x1 , z), and we obtain
π/2−θ
g(r, θ, z) = f (r cos ψ, z) dψ, (6.101)
−π/2−θ
Fig. 6.7 The rotated

coordinate system y
y1
x1
φ
φ
O x
where ψ = (φ−θ). If f is also even in x1 , so f (r cos(ψ), z) = f (r cos(π−ψ), z), we

then obtain π/2
g(r, z) = f (r cos ψ, z) dψ, (6.102)
−π/2
which is clearly axisymmetric.
6.6.1 Inversion of the Transform
The axisymmetric function g(r, z) can be regarded as a transform of the two-

dimensional function f (x1 , z). To invert the transform, we write r cos ψ = x1 ,
obtaining r
f (x1 , z) d x1
g(r, z) = 2 , (6.103)
0 r 2 − x12
which is an Abel integral equation for f (x1 , z) with solution

1 d x1
r g(r, z)dr
f (x1 , z) = , (6.104)
π d x1 0 x12 − r 2
(Barber (2010) and Sect. 30.2.4).
6.6.2 Example: Uniform Loading Over the Circle
As an illustration of this procedure, we consider the case where the half-space is

subjected to uniform loading p0 over the circle 0 ≤r < a. Writing g(r, 0) = p0 H (a −
r ), and evaluating the expression (6.104), we obtain
p0
f (x1 , 0) = |x1 | < a
π ⎛ ⎞
p0 ⎝ |x1 | ⎠
= 1− |x1 | > a. (6.105)
π x12 − a 2
Readers familiar with the fracture mechanics literature will immediately recognize
this equation as defining the compressive traction distribution on the crack plane for
a Griffith crack −a < x1 < a opened by a uniform pressure p0 /π Barber (2010) and
Sect. 13.3.2. In particular, the normal displacement u z on the surface z = 0 is6
6 Inthe Griffith crack problem, this is one half of the crack opening displacement, since the two
sides of the crack displace in opposite directions by the same amount.
6.6 The Smirnov–Sobolev Transform 99

2 p0 (1 − ν 2 ) a 2 − x12
u z (x1 , 0) = |x1 | < a
πE
=0 |x1 | > a. (6.106)
The corresponding displacement in the axisymmetric problem can then be obtained

by the transformation (6.103) as

4 p0 (1 − ν 2 ) min(r,a)
a 2 − x12
u z (r, 0) = d x1 . (6.107)
πE 0 r 2 − x12
6.6.3 Anisotropic Problems
This method is especially useful for problems involving generally anisotropic mate-
rials Sveklo (1964), since the corresponding two-dimensional problem can then be
solved using the classical methods of Stroh (1958, 1962) or Lekhnitskii (1963).
Detailed discussion of these methods and numerous applications are given in the
monograph by Ting (1996).
In Sect. 2.2.2 we used similarity and equilibrium arguments to show that the
normal surface displacement due to a normal point force P acting on the surface of
a generally anisotropic half-space must take the form
Ph(θ)
u z (r, θ, 0) = . (6.108)
r
To determine the function h(θ), we first note that the loading considered in Sect. 6.6.2
reduces to a normal point force if we set p0 = P/πa 2 and then proceed to the limit
as a → 0. The reader can verify that the corresponding displacement (6.107) tends
to Eq. (2.7) in the isotropic case.
If the material is anisotropic, we perform the same superposition over φ, but the
displacements due to the two-dimensional pressure distribution (6.105) will now
depend on φ. In fact, they correspond to the problem in which a Griffith crack of
width 2a and orientation π/2+φ is opened by a pressure
p0 P
= 2 2. (6.109)
π π a
This problem was solved by Stroh (1958) and the displacements at the crack plane
have the form

PC(φ) a 2 − x12
u z (x1 , 0) = |x1 | < a
π2 a 2
=0 |x1 | > a, (6.110)
where C(φ) is a function of φ only, which reflects the fact that the elasticity tensor
ci jkl must be trasformed into the rotated coordinate system (x1 , z). In the limit a → 0,
we obtain the delta function
PC(φ)
u z (x1 , 0) = δ(x1 ), (6.111)
2π
and substitution in the transform (6.100) yields
P π
u z (r, θ, 0) = C −θ . (6.112)
2πr 2
Thus, the normal displacement at (r, θ) depends only on the compliance in a two-
dimensional problem in which the y1 -axis is aligned with the direction θ. This method
uses the solution of the plane anisotropic crack problem to find the function C(φ),
but it can equally be obtained from Fourier transformation Barber and Sturla (1992)
or from the two-dimensional Green’s function Ting (1996).
6.7 Displacements in Two-Dimensional Problems
So far, we have avoided questions of rigid-body displacements in two-dimensional

problems by formulating the contact conditions in terms of displacement gradients.
However, there are two-dimensional problems in which rigid-body displacements are
of interest, notably those where the incremental stiffness is required. For example,
we may choose to preload a cylindrical roller bearing by making the rollers an
interference fit between the races [a two-dimensional version of Problem 3.3]. As
remarked in Sect. 6.1, such problems can only be solved by considering the finite
dimensions of the contacting body, even if these are large compared with the contact
area, and we shall generally find that the compliance increases logarithmically with
these dimensions.
A simple engineering approximation is to imagine ‘cutting out’ the real finite body
from a half-plane and then to set the rigid-body displacement to zero at an appropriate
point on the supported boundary. For example, Fig. 6.8a shows a rectangular elastic
block of height h and width w which is bonded to a rigid plane at the lower edge and
indented by a frictionless rigid cylinder of radius R at the upper edge. We assume
that the contact semi-width a (h, w), so that the stress field in the contact region
is well described by the Hertzian theory. Figure 6.8b shows the equivalent half-plane
solution, where we have indicated the extent of the actual finite body by dashed lines.
To approximate the incremental stiffness of the contact, we compute the displacement
at the point A and superpose a rigid-body displacement sufficient to bring this point
to rest. The vertical displacement of the indenter is therefore approximated by the
6.7 Displacements in Two-Dimensional Problems 101
(a) P (b) P
aa O
h h
w w
A
Fig. 6.8 (a) Indentation of an elastic block bonded to a rigid plane, (b) The block ‘cut out’ from a
half-plane indentation problem
reduction in length of the line O A, which is7

2
P 2h ν a
u≈ ∗ 2 ln − + O . (6.113)
πE a (1 − ν) h2
Greenwood and Barber (2012) give a solution to the case where the width w → ∞
and report results for the indentation depth that differ from Eq. (6.113) only in the
ν-dependent term, suggesting that the approximation is good provided that a/ h is
sufficiently small.
An alternative test of the accuracy of this approximation comprises an infinite strip
of depth h, loaded by a sinusoidal pressure p(x) = p0 cos(ωx), since this problem can
be solved in closed form [See Sect. 14.4]. Comparing the normal surface displacement
from Eq. (14.96) with the difference between the surface displacement and that at
depth h in the half-plane, we find that the percentage error of the approximation
is as shown in Fig. 6.9, where λ = 2π/ω is the wavelength of the loading. Clearly,
the approximation is excellent as long as the surface loading has no components of
wavelength comparable with or longer than the strip thickness.
Fig. 6.9 Percentage error in

the surface displacement of a
bonded strip subjected to a
sinusoidal pressure
distribution, based on change
of thickness of a strip cut
from the half-plane solution
7 Johnson (1985), Eq. (5.58).

Fig. 6.10 An elastic

cylinder compressed P
between rigid planes
A
a a
B
P
Another finite geometry problem for which a closed form solution is available
is that of an elastic cylinder of radius R compressed between two rigid planes, as
shown in Fig. 6.10.
The problem is clearly symmetrical, so if we choose a frame of reference in which
the centre O remains stationary, we find that the two rigid planes each approach O
by a distance
2
P 4R a
uA = ∗ 2 ln −1+O , (6.114)
πE a R2
Johnson (1985). By setting h = R and removing a factor of 2 from the logarithmic

term, we can express the approximation (6.113) as

2
P 4R ν a
uA = ∗ 2 ln − 2 ln(2) − +O . (6.115)
πE a (1 − ν) h2
This expression differs from (6.114) in the second [O(1)] term and defines a smaller
contact deformation, as we should expect because the additional surrounding material
in Fig. 6.8b constrains the elastic deformation of the contained disc. However, the
approximation is still good provided a R.
6.7.1 Kalker’s Line Contact Theory
In some situations, the contact problem is three dimensional, but the contact area is
‘slender’, so that conditions local to the contact area are approximately two dimen-
sional. For example, if a cylinder of finite length is pressed into the surface of a
half-space, we expect the width of the contact area to be much smaller than the
length of the cylinder, and hence, near the middle of the cylinder at least, the stress
gradients perpendicular to the cylinder axis will be much steeper than those parallel
to the axis.
Kalker (1972) exploited this idea by expanding the three-dimensional solution in
powers of a small parameter representing the ratio between the two length scales.
When the problem is viewed on the scale of the length of the contact area [the outer
problem], the contact pressure appears as a line force, except close to the point of
reference. But when viewed on the scale of the contact width [the inner problem],
deviations from the locally two-dimensional solution appear solely as an added rigid-
body displacement. The two problems can be related using the technique of matched
asymptotic expansions [Sivashinsky (1975) and Kalker (1977)].
It is convenient to make the small parameter explicit by defining dimensionless
coordinates ξ = x/b, η = y/b, where b is the half-length of the longer dimension of
the contact area, taken to be in the direction y. The normal surface displacement
u z (ξ, η) in the contact area [−(η) < ξ < (η), −1 < η < 1] is then approximated by

(η)
b
u z (ξ, η) = ∗ −2 p(s, η) ln |s − ξ|ds + F(η) ln 4(1 − η 2 )
πE −(η)
1
{F(t) − F(η)}dt
+ = d − g0 (ξ, η) (6.116)
−1 |t − η|
Panek and Kalker (1977), where

(η)
F(η) = p(ξ, η)dξ, (6.117)
−(η)
and the aspect ratio of the contact area implies that (η) 1 for all η.
Differentiating (6.116) with respect to ξ, we then obtain
(η) ∗
p(s, η)ds π E ∂g0 (ξ, η)
= − (η) < ξ < (η), (6.118)
−(η) (ξ − s) 2b ∂ξ
since the remaining terms are independent of ξ. This is a Cauchy singular integral
equation exactly analogous to (6.10). If the contact is conformal, so that (η) is a
known function of η, we require the singular solution which is
∗ (η)

1 E (η)2 − s 2 ∂g0 (s, η)
p(ξ, η) = F(η) − ds . (6.119)
π (η)2 − ξ 2 2b −(η) (ξ − s) ∂s
If this is now substituted into (6.116), the integral with respect to s can be performed,
and the solution (6.118) of the Cauchy equation guarantees that the terms involving
ξ then cancel, leaving a Fredholm integral equation on the domain −1 < η < 1 for
the unknown function F(η). Alternatively, we can simply set ξ = 0 in (6.116) before
substituting for p(s, η), leading to the same Fredholm equation. Once F(η) is known,
the total force is given by
1
P=b 2
F(η)dη. (6.120)
−1
For non-conformal contact problems, the semi-width of the contact area (η) appears
as an additional unknown, but it is related to F(η) by Eq. (C.9)2 .
Panek and Kalker (1977) and Sivashinsky (1975) used this method to approximate
the solution for the problem where a frictionless rigid punch of thin rectangular
planform [−a < x < a, −b < y < b, b a] is pressed into the surface of an elastic
half-space. It should be emphasized that the approximation involved is then reliable
away from the ends of the punch, but the predicted contact pressure distribution is
less accurate within distances of the order b from the ends x = ± a and is particularly
unreliable in the corners x = ± a, y = ± b where a more complex singular field is to
be expected. Similar considerations apply when a finite elastic cylinder is pressed
into the surface of an elastic half-space. The end effects in several cases of this kind
are discussed by Johnson (1985), Sect. 5.6.
Kalker (1977) extended the method to allow the centreline of the contact area to
be curved, and also included the effect of tangential forces, resulting from friction at
the interface [see Chaps. 9 and 18 below].
The method of matched asymptotic expansions is not restricted to half-space
problems. Castillo and Barber (1997) used it to analyze the problem of two elastic
cylinders whose axes are slightly misaligned. In this case, the classical Hertzian solu-
tion predicts a long narrow elliptical contact, but for sufficiently small misalignment
angles, the major axis can exceed the radius of the cylinders, which can then certainly
not be treated as half-spaces. Instead, the outer problem in this case comprises the
bending of two cylindrical beams, and is similar to that treated in Sect. 13.1 below.
However, the inner problem is defined by a two-dimensional Hertz contact in which
the contact force is resisted by a body force distributed in the form corresponding to
the transmission of a shear force along the cylinder, as shown in Fig. 6.11.
Fig. 6.11 The inner problem

for contact of a circular P
cylindrical beam
A
a a
R
Since the problem is not symmetric, the elastic compliance in the contact area
must be determined relative to an appropriate weighted average of the cylinder dis-
placements. Renton (1991) argues that the appropriate reference should be chosen
such that reaction forces do no net work. Using this criterion, Castillo and Barber
obtain the result

P 2R 85 + 291ν + 698ν 2 − 232ν 3 − 832ν 4
u= 2 ln − , (6.121)
πE∗ a 1152(1 − ν 2 )2
for the effective contact compliance. Notice however that this expression differs from
(6.115) only in the second term, which makes only a relatively small contribution to
the result.
Problems
1. The profiles of two half-planes are defined by the symmetric gap function g0 (x) =
C x 4 , where C is a constant. Use any of the various methods in this chapter to
determine the contact pressure distribution p(x) and the applied force P as functions
of the semi-width of the contact area a.
2. A rigid punch has the wedge-shaped profile shown in Fig. 6.12, where the angle
α 1. Find the contact semi-width a and the contact pressure p(x) as functions of
the applied force P. Discuss the nature of the pressure distribution near x = 0 and
x = a.
3. A rigid punch has the form of a truncated wedge as shown in Fig. 6.13. Find the
relation between the applied normal force P and the half-width of the contact area a.
4. A rigid punch in the form of a half-cylinder of radius R is pressed into an elas-
tic half-plane such that the plane side of the punch remains vertical, as shown in
Fig. 6.14. Find the relationship between the indenting force P and the width a of the
contact area, and hence determine the contact pressure distribution p(x) using the
incremental method of Sect. 6.3.
5. Use Eq. (6.27) and the incremental method of Sect. 6.3 to determine the contact
pressure distribution p(x) for the flat and rounded punch of Fig. 6.4. Plot the resulting
expression for representative values of b/a.
Fig. 6.12 The

wedge-shaped indenter P
α
a x

α
b
a
We know that the contact pressure for the flat punch is singular at the edges, so we
must anticipate a local maximum of contact pressure as a → b. Find the value of this
maximum and make a log-log dimensionless plot as a function of k = 1 − b2 /a 2 .
Comment on the shape of this plot as k → 0.
6. Use the Fourier series method of Sect. 6.4 to find the contact pressure distribution
p(x) for Problem 4.
7. Find the contact pressure distribution for the problem of Fig. 6.5a if the line of
action of the force P passes through the point x = −c and the contact area is defined
by −a < x < a. Hence determine the value c0 such that the entire punch face remains
in contact if and only if |c| < c0 .
Now suppose that c0 < c < a. Find the new contact pressure distribution p(x), the
angle of tilt α and the extent of the contact area.
8. Use the potentials (6.66) to find the complete stress and displacement field in the
half-plane due to the contact pressure distribution (6.65), and hence verify Eq. (6.68).
9. Solve Westergaard’s problem [Fig. 6.6] by substituting (6.76) in (6.75) and solving
the resulting Cauchy singular integral equation as in Sect. 6.2.
10. By writing p0 = P/πa 2 and proceeding to the limit where a → 0, show that Eq.
(6.107) reduces to the point force solution (2.7).
11. Use the method of Sect. 5.1 to determine the surface displacements due to the
axisymmetric pressure distribution p(r ) = p0 H (a −r ) and hence verify that they are
given by Eq. (6.107).
12. A rectangular block of height h and width w rests on two rigid cylinders, each
of radius R as shown in Fig. 6.15. The block is loaded only by its own weight. Use
arguments similar to those in Sect. 6.7 to estimate the vertical displacement of the
centre of gravity of the block due to elastic deformation. The material has density ρ,
Young’s modulus E and Poisson’s ratio ν.
Fig. 6.14 Indentation by a P

half-cylinder
a
Problems 107
Fig. 6.15 A rectangular

block resting on two w
cylinders h
radius R
Do you expect the spacing between the two supports to influence the result, and
if so, in what range of the parameters w, h, R?
13. A rigid flat punch with a rectangular cross section 20a ×2a is pressed into an
elastic half-space by a force P. Use Kalker’s line contact theory to estimate the
distribution of contact stress. How does your estimate for the indentation depth d
compare with that for an elliptical flat punch of semi-axes 10a and a?
Hint: Notice that g0 = 0 throughout the contact area, so the integral term in Eq.
(6.119) is also zero. You will need to write a simple numerical code to solve the
resulting Fredholm equation [e.g. by assuming that F(η) is piecewise constant and
using collocation at the mid-point of each element].
14. If the annular punch in Fig. 5.2 is ‘thin’, meaning 2c ≡ (a −b) a, the contact
pressure will be locally approximately two dimensional and given by
F
p(x) ≈ √ where x = r − R,
π c − x2
2
and F = P/2π R is the applied force per unit circumference around the mean line
R = (a +b)/2.
Estimate the rigid-body indentation Δ of the punch by (i) using Eq. (6.64) to
calculate the indentation of a two-dimensional flat punch relative to the points x =
±L, and then (ii) adding the average of the displacements at the points r = R − L and
r = R + L, equidistant on the two sides of a ring force of F per unit circumference,
for which
4F r
u z (r ) = K 0≤r < R
πE∗ R
4F R R
= ∗ K r > R.
πE r r
The resulting relation between P and Δ will depend on your choice of L which must
lie in the range c < L < R. How sensitive is the result to this choice and what do you
think is the most appropriate value?
Chapter 7
Tangential Loading
So far, we have considered only problems in which no tangential tractions are trans-
mitted across the contact interface. This will be the case if the contact is frictionless
[i.e. well lubricated] or if the contact interface is a plane of symmetry with regard to
geometry, material properties and loading. In most other cases, we should anticipate
that the normal contact tractions will tend to cause relative tangential displacement
or slip at the interface, and in many cases, this will be opposed by frictional tractions.
7.1 Kinematics
Suppose that the two bodies can be approximated by half-spaces, as in Sect. 2.1,
and that the contact area is denoted by A. We now impose a rigid-body tangential
displacement U(t) = {Ux (t), U y (t)} on body 2 [the upper body] in Fig. 7.1, where
t is time. In addition, we anticipate that there will be some tangential elastic surface
displacements in the two bodies that we identify as {u (k) (k)
x (x, y, t), u y (x, y, t)}, k =
1, 2. It follows that the relative tangential motion between two contacting points at
(x, y) [known as the shift h(x, y, t)] will be given by
h = U − u where u x = u (1) (2) (1) (2)

x − ux ; u y = u y − u y . (7.1)
With this sign convention, a positive shift is one where material points in the upper
body move in the direction of the corresponding positive coordinate axis relative to
points on the lower body.
We can also define the slip velocity
Vs ≡ ḣ = U̇ − u̇, (7.2)

https://doi.org/10.1007/978-3-319-70939-0_7
110 7 Tangential Loading
Fig. 7.1 Relative tangential

displacement Ux
2.
x
1.
z
which represents the relative velocity of a pair of material points in A, one in each
of the contacting bodies. The dot here represents differentiation with respect to time.
The spatial derivatives of h are related to tangential surface strains. For example
∂h x ∂u (2) ∂u (1)
= x
− x
= ex(2)x − ex(1)x , (7.3)
∂x ∂x ∂x
from (7.1), since the rigid-body displacement U(t) is a function of time only.
7.1.1 Gross Slip and Microslip
Suppose that through some sufficiently strong frictional mechanism, slip at a given
point (x, y) is completely prevented. We refer to this as a state of stick. It does not
necessarily imply that the shift h(x, y) = 0, since we may have imposed some rigid-
body displacement or elastic deformation before the bodies were placed in contact.
However, it does imply that no further shift occurs once contact has been established
and hence during a period of stick, ḣ(x, y) = 0.
Since the bodies are deformable, it is possible for part of the contact area to be in a
state of stick, whilst the rest is slipping. For example, suppose in a two-dimensional
problem, the contact area b < x < a is established with h x (x) = 0 and that during
subsequent tangential loading, the region b < x < c remains stuck, whilst c < x < a
is slipping. In the stick region, h(x) must remain equal to zero, including at the
stick-slip boundary x = c. In the slip region, we can, therefore, write

x
dh x x (2)
h x (x) = dx = ex x − ex(1)x d x c < x < a, (7.4)
c dx c
and since strains are generally small at least in elastic problems, it follows that the
amount of shift that can accumulate in the slip zone is itself small. This process is
therefore generally known as microslip.
By contrast, if the whole contact area slips, there is no restriction on the amount of
rigid-body displacement U that can accumulate, so shifts may be large. We refer to
7.1 Kinematics 111
this situation as gross slip or sliding. Typically, if two contacting bodies are subjected
to a tangential force, there will exist regions of stick and microslip at low forces,
but if the tangential force is increased, the stick region will shrink. When it has
shrunk to zero, sliding commences and the relative tangential motion then increases
dramatically.
There are significant kinematic differences between the formulation of problems
of microslip and gross slip. In the former case, we can usually refer the conditions
in the contact area to a coordinate system based on the undeformed and undisplaced
bodies, whereas during gross slip points on the two bodies move over each other
sufficiently far to require a material based [Lagrangian] kinematics.
7.2 Green’s Functions for Tangential Forces and

Displacements
In preparation for the formulation of contact problems involving tangential tractions,

we first need to generalize the results of Sect. 2.2.1 and 6.1 to include tangential forces
and displacements.
7.2.1 Three-Dimensional [point] Loading
If a normal compressive force P and tangential force Q = {Q x , Q y } act at the origin

on the surface of the elastic half-space z > 0 [body 1], as shown in Fig. 7.2, the
resulting surface displacements are1
(1 − 2ν)(1 + ν)P x (1 − ν 2 )Q x ν(1 + ν)x(x Q x + y Q y )

ux = − + +
2π Er 2 π Er π Er 3
(1 − 2ν)(1 + ν)P y (1 − ν )Q y
2
ν(1 + ν)y(x Q x + y Q y )
uy = − + +
2π Er 2 π Er π Er 3
(1 − ν 2 )P (1 − 2ν)(1 + ν)(x Q x + y Q y )
uz = + . (7.5)
π Er 2π Er 2
These results can also be expressed in polar coordinates as
(1 − 2ν)(1 + ν)P cos θ (1 + ν)(2 − ν)Q x ν(1 + ν)Q x cos(2θ)

ux = − + +
2π Er 2π Er 2π Er
ν(1 + ν)Q y sin(2θ)
+
2π Er
1 For the solution of this elasticity problem, see for example Westergaard (1964), Sect. 79.
(1 − 2ν)(1 + ν)P sin θ (1 + ν)(2 − ν)Q y ν(1 + ν)Q y cos(2θ)

uy = − + −
2π Er 2π Er 2π Er
ν(1 + ν)Q x sin(2θ)
+
2π Er
(1 − ν 2 )P (1 − 2ν)(1 + ν)(Q x cos θ + Q y sin θ)
uz = + . (7.6)
π Er 2π Er
Applying an equal and opposite force to the upper body 2, and using the same
equations with appropriate sign changes, we obtain the relative displacements as
−β P x Qx x(x Q x + y Q y )
u x ≡ u (1) (2)
x − ux = − + +
πE r ∗ 2 πE r∗ 3
π Er
−β P y Qy y(x Q x + y Q y )
u y ≡ u (1) (2)
y − uy =− + +
π E ∗r 2 π E ∗r 3
π Er
P β(x Q x + y Q y ))
u z ≡ u (1) (2)
z − uz = + , (7.7)
π E ∗r π E ∗r 2
where
(1 − 2ν1 )(1 + ν1 ) (1 − 2ν2 )(1 + ν2 )
β = E∗ − (7.8)
2E 1 2E 2
is another composite modulus defined

is one of Dundurs’ bimaterial constants2 and E
such that
1 ν1 (1 + ν1 ) ν2 (1 + ν2 )
= + . (7.9)

E E1 E2
Fig. 7.2 Concentrated

forces P, Q x , Q y applied to P
the surface of the half-space Qx
y r θ ux
Qy
x uz
uy
2 See Appendix D.
7.2 Green’s Functions for Tangential Forces and Displacements 113
7.2.2 Two-Dimensional [line] Loading
In two-dimensional problems, the Green’s function corresponds to line forces P,

Q x , Q y per unit length along the y-axis, as shown in Fig. 7.3
The corresponding surface displacements are
P(1 + ν)(1 − 2ν)sgn(x) 2Q x (1 − ν 2 ) ln |x|

u (1)
x = − −
2E πE
2Q y (1 + ν) ln |x|
u (1) =− (7.10)
y
πE
2P(1 − ν 2 ) ln |x| Q x (1 + ν)(1 − 2ν)sgn(x)
u (1) =− + ,
z
πE 2E
where sgn(x) is the signum function, defined by sgn(x) = 1 for x > 0 and sgn(x) = −1
for x < 0. Alternatively, we can write
sgn(x) = 2H (x) − 1, (7.11)
where H (x) is the Heaviside step function.

With two deformable bodies, the relative displacements are
β Psgn(x) 2Q x ln |x|
u x ≡ u (1) (2)
x − ux = − −
E∗ πE∗
2Q y ln |x|
u y ≡ u (1) (2)
y − uy =− (7.12)
πE
2P ln |x| β Q x sgn(x)
u z ≡ u (1) (2)
z − uz =− + ,
πE∗ E∗
where
1 1 + ν1 1 + ν2

= + , (7.13)
E E1 E2
and the differentiated forms are
Fig. 7.3 Concentrated line P

forces P, Q x , Q y per unit Qx
length applied to the surface
of the half-space ux
uy uz
Qy
y x
z
du x 2β Pδ(x) 2Q x du y 2Q y
=− ∗
− ; =−
dx E π E∗x dx πE x
du z 2P 2β Qδ(x)
=− ∗ + , (7.14)
dx πE x E∗
where
d H (x)
δ(x) = (7.15)
dx
is the Dirac delta function.
Since u y depends only on Q y and u x , u z are independent of Q y , the general two-
dimensional problem can be decomposed into an in-plane problem (u x , u z , Q x , P)
and an antiplane problem (u y , Q y ).
If the normal and tangential contact tractions are p(x), qx (x), q y (x) respectively,
the surface displacement derivatives can be written as convolution integrals on
Eq. (7.14), giving
a
du x 2β p(x) 2 qx (ξ)dξ
=− ∗
− ∗
(7.16)
dx E π E b (x − ξ)
a
du z 2 p(ξ)dξ 2βqx (x)
=− ∗ + (7.17)
dx π E b (x − ξ) E∗
for the in-plane problem, and

du y 2 a
q y (ξ)dξ
=− (7.18)
dx πE b (x − ξ)
for the antiplane problem. Equation (7.18) has the same mathematical form as (6.7),
and hence the techniques developed in Chap. 6 can be applied directly to the antiplane
problem.
7.2.3 Normal-Tangential Coupling
Notice that in three-dimensional problems and two-dimensional in-plane problems,

the normal force P produces tangential displacements and the tangential forces
Q, Q x produce normal displacements, except in the special case where β = 0. This
introduces coupling between normal and tangential effects which has a significant
influence on the behaviour of contact problems, as we shall see in this and subsequent
chapters.
Uncoupled problems are so much simpler than coupled problems that it is tempting
to set β = 0 even when it is simply small compared with unity. Three important
categories of problem in which β is strictly zero are:-
7.2 Green’s Functions for Tangential Forces and Displacements 115
• The materials of the two bodies are the same, so E 1 = E 2 and ν1 = ν2 .

• Both materials are incompressible, so ν1 = ν2 = 0.5. This is a reasonable approx-
imation for many polymeric materials, including rubber.
• One body is incompressible and the other is rigid. Of course, no real materials are
rigid, but β will be very small compared with unity as long as [for example] ν1 = 0.5
and E 2 E 1 , as in the case of contact between steel and rubber components, or
indentation tests on a polymeric material.
We also note that coupling is not present in antiplane problems, even when β = 0.
7.3 Two-Dimensional Flat Rigid Punch with No Slip
Figure 7.4 shows a two-dimensional flat rigid punch that is constrained to remain
vertical whilst being loaded by a normal force P and a tangential force Q x . We
consider the case where there is sufficient friction at the interface to prevent any
tangential relative motion. We then have g0 (x) = 0 in the contact area b < x < a for
all values of the normal and tangential forces P, Q x and, since there is no slip,
d ḣ
=0 b<x <a (7.19)
dx
for all t. It follows that
dh
=0 b < x < a, (7.20)
dx
since we can remove the time derivative on h by integrating in time from an unloaded
and undeformed initial condition. The kinematic boundary conditions can therefore
be written
du x du z
= 0; = 0 b < x < a, (7.21)
dx dx
from (7.1, 6.8) respectively. It then follows from (7.16, 7.17) that p(x), qx (x) must
Fig. 7.4 Flat rigid punch P

loaded by normal and
tangential forces
Qx
b a
satisfy the two integral equations

a
qx (ξ)dξ
+ πβ p(x) = 0 (7.22)
(x − ξ)
b
a
p(ξ)dξ
− πβqx (x) = 0, (7.23)
b (x − ξ)
in b < x < a, supplemented by the equilibrium equations

a a
qx (x)d x = Q x ; p(x)d x = P. (7.24)
b b
√
Multiplying (7.22) by ı = −1 and adding the result to (7.23), we obtain the single
complex equation
a
f (ξ)dξ
+ ıπβ f (x) = 0 b < x < a, (7.25)
b (x − ξ)
where
f (x) = p(x) + ıqx (x). (7.26)
The normalization (C.1) in Appendix C converts (7.25) to the Cauchy singular inte-
gral equation of the second kind (C.11), with

ı 1 1−β
λ=− and hence γ = ln , (7.27)
β 2πı 1+β
from (C.13). We define the alternative material mismatch parameter

1 1+β
= ln , (7.28)
2π 1−β
so that γ = ı, and we also find
1
cos(πγ) =
. (7.29)
1 − β2
Using these results, the solution of (7.25) is given by (C.14) as

ı
(P + ı Q x ) x −b
f (x) ≡ p(x) + ıqx (x) =
√ . (7.30)
π 1 − β 2 (a − x)(x − b) a−x
7.3 Two-Dimensional Flat Rigid Punch with No Slip 117
7.3.1 Uncoupled Problem
If β = 0, we also have = 0, and the traction distributions simplify to
P Qx
p(x) = √ ; qx (x) = √ . (7.31)
π (a − x)(x − b) π (a − x)(x − b)
Thus the tangential force has no effect on the normal traction distribution, which
is therefore identical to that in the frictionless problem (6.14). Also, the tangential
tractions have the same square-root singular form and depend only on the tangential
force Q x .
7.3.2 Oscillatory Singularities
If the problem is coupled, = 0 and Eq. (7.30) contains the additional factor
ı
x −b x −b x −b
= cos ln + ı sin ln . (7.32)
a−x a−x a−x
In particular, the normal traction distribution is then given by
1
p(x) = ( f (x)) =
√
π 1 − β (a − x)(x − b)
2

x −b x −b
× P cos ln − Q x sin ln
a−x a−x

P + Q x cos(θ + φ)
2 2
=
√ , (7.33)
π 1 − β 2 (a − x)(x − b)
where
x −b Qx
θ = ln ; tan φ = . (7.34)
a−x P
The pressure distribution p(x) near the left edge x = b is illustrated in Fig. 7.5. It
is bounded between the square-root singular limits

P 2 + Q 2x P 2 + Q 2x
−
√ ≤ p(x) ≤
√ , (7.35)
π 1 − β 2 (a − x)(x − b) π 1 − β 2 (a − x)(x − b)
shown as dashed lines in the figure. However, θ → ∞ as x → a and θ → − ∞ as

x → b, so the expression (7.33) oscillates an infinite number of times between these
limits in the vicinity of the each end point.
Fig. 7.5 Form of the contact pressure distribution near x = b. The distance x −b is plotted on a
logarithmic scale, since the oscillations are generally confined to a very small region near the ends.
Also, an unphysically high value of was used in this plot to clarify the qualitative behaviour
These oscillations are physically unacceptable, since they imply localized vio-
lations of the unilateral condition (1.3)4 . However, if |Q x |/P is not too large, the
region of violation is a small proportion of the total contact area, in which case the
above solution might be expected to be approximately correct, except very close to
the end points.
To explore this issue further, we can determine the first point at which p(x) goes
negative by setting θ+φ = ±π/2 and hence

x −b 1 ±π
= exp −φ . (7.36)
a−x 2
The parameter β cannot exceed 0.5, so the maximum possible value of [implying
the maximum influence of normal-tangential coupling] is
ln(3)
max = ≈ 0.175, (7.37)
2π
from Eq. (7.28). Using this value and assuming Q x = 0 and hence φ = 0, we find that
the first violation of the contact inequality occurs at a point distant 0.000125(a −b)
from the two end points. Thus, the violation is confined to a very small region near
the end points, where we might argue that the elasticity solution breaks down anyway
because the stresses and hence the strains are extremely large.
However, if |Q x | = 0, φ = 0 and the region of violation will be increased near one
end and decreased near the other. For example, if Q x = P, φ = π/4 and the larger
violation zone is now 0.0112(a − b) — i.e. 11% of the total contact area, which is
clearly unacceptable. Remember however that these calculations are based on the
maximum possible value of . For a more realistic case such as the indentation of a
7.3 Two-Dimensional Flat Rigid Punch with No Slip 119
half-plane with ν1 = 0.3 by a rigid punch [E 2 → ∞], we obtain ≈ 0.0935 and the
largest violation zone is extremely small even for |Q x | ∼ P.
These considerations can often be used to justify our ignoring the inequality
violations associated with the oscillatory tractions near the end points, but their
existence still represents a blemish in the theory and several modified formulations
of the problem have been advanced to generate a physically and mathematically
consistent contact model (e.g. Adams 1979). Of these, the most convincing involves
the assumption that the contact is frictional, with a finite friction coefficient (Spence
1973), in which case a solution satisfying all the inequalities can be obtained with a
frictional slip zone at each edge of the contact area. We shall discuss this solution in
Chap. 9 (Sect. 9.5.2).
7.4 Axisymmetric Flat Rigid Punch with No Slip
A related problem is that in which an axisymmetric flat punch of radius a is pressed

into an elastic half-space with no slip by a normal force P. In this case, first solved
by Mossakovskii (1954, 1963), the tangential force Q must be zero to preserve
axisymmetry, but if the problem is coupled [β = 0], we anticipate the development
of an axisymmetric tangential traction distribution qr (r ) = −σzr (r, 0). These tractions
constrain the deformation by preventing radial slip and hence should be expected
to increase the stiffness of the contact, defined as the ratio P/Δ, where Δ is the
indentation depth.
The surface outside the contact area is traction free, so the boundary conditions
can be stated as
σzr = σzz = 0 z = 0, r > a
u r = 0; u z = Δ z = 0, 0 ≤ r < a. (7.38)
The problem can be formulated by supplementing the solution of Sect. 2.2 [Appen-
dix A, Sect. A.1] by that of [Appendix A, Sect. A.2], where χ(r, z) is an axisymmetric
harmonic potential and ψ = 0. The non-zero surface tractions and displacements are
then given by
(1 − ν) ∂χ (1 − 2ν) ∂ϕ (1 − 2ν) ∂χ (1 − ν) ∂ϕ
u r (r, 0) = + ; u z (r, 0) = − −
G ∂r 2G ∂r 2G ∂z G ∂z
(7.39)
∂2χ ∂2ϕ
σzr (r, 0) = ; σzz (r, 0) = − 2 , (7.40)
∂r ∂z ∂z
and Eq. (7.39) can be generalized to the case of two deformable materials by writing

2 ∂ϕ ∂χ 2 ∂ϕ ∂χ
ur = β + ; uz = − −β . (7.41)
E∗ ∂r ∂r E∗ ∂z ∂z
This representation reduces the boundary value problem (7.38) to the search for
two harmonic potentials ϕ, χ satisfying the boundary conditions
∂2χ ∂2ϕ
= =0 z = 0, r >a (7.42)
∂r ∂z ∂z 2
∂ϕ ∂χ E ∗Δ ∂ϕ ∂χ
+β =− ; β + =0 z = 0, 0 ≤ r < a. (7.43)
∂z ∂z 2 ∂r ∂r
As in Sect. 5.1, we can satisfy the homogeneous conditions (7.42) by defining

a a
ϕ= F(r, z, t)h 1 (t)dt; χ = F(r, z, t)h 2 (t)dt, (7.44)
0 0
after which conditions (7.43) define two simultaneous Abel integral equations for
the real functions h 1 (t), h 2 (t). These can then be reduced to a single Cauchy integral
equation3
ıβ a f (t)dt E ∗Δ
f (x) = − − a < x < a, (7.45)
π −a (x − t) π
where f (x) = h 1 (x)+ı h 2 (x). This equation is of the same class as Eq. (7.25), and
can be solved by similar methods, though important differences are introduced by
the fact that the multiplier β now appears on the integral term.4
The contact tractions are obtained as
a
E ∗Δ 1 d t cos(2θ)dt
p(r ) = −σzz (r, 0) = −
√ (7.46)
π 1 − β 2 r dr r t2 − r2
a
E ∗ aΔ d sin(2θ)dt
q(r ) = −σzr (r, 0) = −
√ (7.47)
π 1−β 2 dr r t2 − r2
(Spence 1968), where

−1 t 1 a+t
θ(t) = tanh = ln . (7.48)
a 2 a−t
We also record the normal surface displacement outside the contact area [r > a],
which is
r 1
2 cos(2θ)ds
u z (r ) = Δφ where φ(ρ) =

, (7.49)
a π 1−β 0 2 ρ2 − s 2
with θ = tanh−1 s.
3 For an illustration of this method, see Barber (2010), Sect. 32.3.

4 See Gladwell (1980), p. 488.
7.4 Axisymmetric Flat Rigid Punch with No Slip 121
Fig. 7.6 Force–

displacement relation for the
axisymmetric flat punch with
no slip
P
2E aΔ
The relation between the normal force P and the indentation depth Δ is

E ∗ aΔ 1+β
P= ln . (7.50)
β 1−β
This expression is plotted in Fig. 7.6 and confirms that as the coupling parameter β
increases, the constraint on tangential displacement increases the required indenting
force. However, the maximum increase [at β = 0.5] is only about 10%.
It is easily verified that as β → 0 [weak coupling], → 0 [from (7.28)] and the
pressure distribution (7.46) tends to that given in Eq. (5.21), whilst the tangential
traction tends to zero, since the integrand in (7.47) becomes zero.
7.5 The ‘Goodman’ Approximation
The flat punch solutions of Sects 7.3 and 7.4 show that when the problem is coupled
[β = 0], tangential tractions are developed when slip is prevented, even when the
loading is purely normal, and that these tractions also modify the normal traction
distribution. However, the latter effect is generally small except in the immediate
vicinity of the edges of the contact region. For example, the contact pressure at the
centre of the contact area defined by Eq. (7.30) differs from the frictionless value by
only 10%, even for the maximum possible value β = 0.5 corresponding to ν = 0. For
a more realistic value ν = 0.3, the error in the normal tractions incurred by the use
of the frictionless solution is less than 3%, which is acceptable in many engineering
applications.
Goodman (1962) suggested a method of trading on this weak coupling to approx-
imate the solution of coupled contact problems. In effect, the non-zero tangential
tractions are neglected in the solution for the normal tractions, which therefore take
the same form as in the frictionless problem. However, the reciprocal effect of normal
tractions on tangential displacmements is included. This is particularly appropriate
in situations where the tangential tractions arise from friction, since (i) frictional
tractions are generally significantly less than the corresponding contact pressures
[coefficients of friction are usually less than unity, though this is not a physical
requirement] and (ii) associated frictional slip is often associated with wear and
other damage mechanisms, so it is desirable to obtain relatively accurate estimates.
In the two-dimensional problem, Goodman’s approximation is formally equiva-
lent to the replacement of Eqs. (7.16, 7.17) by
a
du x 2β p(x) 2 qx (ξ)dξ
=− − (7.51)
dx E∗ π E ∗ b (x − ξ)
a
du z 2 p(ξ)dξ
=− ∗ , (7.52)
dx π E b (x − ξ)
where the last term in (7.17) has been deleted. For the flat punch problem of Sect. 7.3,
we then obtain
P
p(x) = √ , (7.53)
π (a − x)(x − b)
as in (7.31), after which (7.51) and the no-slip condition gives

a
qx (ξ)dξ βP
= −πβ p(x) = − √ b < x < a. (7.54)
b (x − ξ) (a − x)(x − b)
(a) (b)
Fig. 7.7 Contact pressure (a) and tangential traction (b) for the flat punch of width 2a loaded by a
normal force P, with β = 0.3 and no slip. The solid line is the exact solution (7.30) and the dashed
line is the Goodman approximation (7.53, 7.55). For this case, the oscillations in the exact solution
described in Sect. 7.3.2 are confined to a region of order 10−7 from x/a = 1
7.5 The ‘Goodman’ Approximation 123
This is a Cauchy singular integral equation of the same form as (6.10), with solution

1 β P a dξ
qx (x) = √ Qx +
π (x − b)(a − x) π b (x − ξ)

1 βP a−x
= √ Qx − ln . (7.55)
π (x − b)(a − x) π x −b
Figure 7.7 compares the exact solution (7.30) for the contact tractions p(x), qx (x)
with the Goodman approximation [dashed line] for the case where b = −a, the loading
is purely normal [Q x = 0] and β = 0.3. The problem is symmetrical, so p(−x) = p(x)
and qx (−x) = − qx (x). The approximation is clearly acceptable for most engineering
purposes and also a fortiori for β < 0.3, which corresponds to ν > 0.28.
7.6 Uniform Tangential Displacement in a Prescribed Area
In this section, we shall consider problems in which all points inside a prescribed area
A are caused to displace tangentially by the same distance U = {Ux , U y }, whilst the
normal tractions are everywhere zero. These conditions define the tangential loading
phase of the problem in which a rigid punch of planform A is first pressed into a half-
space by a normal force P, and then loaded tangentially by a force Q = {Q x , Q y },
provided that β = 0 and there is no slip.
7.6.1 Tangential Loading over a Circular Area
If A is a circle of radius a, the tractions are
Q
q(r, θ) = √ , (7.56)
2πa a 2 − r 2
whilst the rigid-body tangential displacement is
Q(1 + ν)(2 − ν)
U= . (7.57)
4Ea
Westmann (1965) obtained these results using the method of Hankel transforms and
dual integral equations, but they can also be derived using the field-point integration
method of Sect. 2.3.1, as we shall show in the next section.
Of course, the rigid punch problem satisfies the condition β = 0 only for ν = 0.5,
but the relation (7.57) can also be used for the case where two half-spaces of the same
material [with any value of ν] make contact over a circle of radius a and undergo
a relative tangential displacement. In this case, the displacement of the contact area
Fig. 7.8 Elliptical contact

y,η
area
S2
Q
b r
P
θ
O a x,ξ
S1
relative to the point at infinity will be U in each half-space and the total relative
tangential motion of points at infinity in the two bodies will be 2U.
Notice that the normal compliance for the same geometry is given by Eq. (5.21)3
as Δ = P(1−ν 2 )/2Ea, so the contact is more flexible in tangential motion through
the ratio
(2 − ν)
RT = . (7.58)
2(1 − ν)
7.6.2 Tangential Loading over an Elliptical Area
If tangential tractions q(x, y) are applied inside the ellipse of Fig. 7.8, Eq. (7.6) can
be used with the field-point integration method of Sect. 2.3.1 to obtain the surface
tangential displacements at point P as
S2
(1 + ν)(2 − ν) π S2 ν(1 + ν) π
ux = qx (r, θ)dr dθ + cos(2θ) qx (r, θ)dr dθ
2π E 0 S1 2π E 0 S1
S2
ν(1 + ν) π
+ sin(2θ) q y (r, θ)dr dθ (7.59)
2π E 0 S1
S2
(1 + ν)(2 − ν) π S2 ν(1 + ν) π
uy = q y (r, θ)dr dθ − cos(2θ) q y (r, θ)dr dθ
2π E 0 S1 2π E 0 S1
S2
ν(1 + ν) π
+ sin(2θ) qx (r, θ)dr dθ. (7.60)
2π E 0 S1
The inner integrals have the same form as Eq. (2.21), so we conclude as in Sect. 2.4
that if the traction distribution has the form
−1/2
x2 y2
q(x, y) = P n (x, y) 1 − 2 − 2 , (7.61)
a b
7.6 Uniform Tangential Displacement in a Prescribed Area 125
and if the components of the vector P n (x, y) are polynomials of degree n, then u x
and u y will be polynomials of degree n inside the ellipse. In other words, Galin’s
theorem applies to tangential as well as normal loading.
In particular, for n = 0, if
−1/2
{Q x , Q y } x2 y2
{qx , q y }(x, y) = 1− 2 − 2 . (7.62)
2πab a b
corresponding to a tangential force Q = {Q x , Q y }, the integrals (7.59, 7.60) can be

evaluated as in Sect. 2.3.2 and Appendix B to give
(1 + ν)Q x
ux = [(2 − ν)I0 (0, e) + ν I0 (1, e)] (7.63)
4π Ea
(1 + ν)Q y
uy = [(2 − ν)I0 (0, e) − ν I0 (1, e)] , (7.64)
4π Ea
or
(1 − ν 2 )Q x K (e) ν(1 + ν)Q x [K (e) − E(e)]

ux = + (7.65)
π Ea πe2 Ea
(1 + ν)Q y K (e) ν(1 + ν)Q y [K (e) − E(e)]
uy = − , (7.66)
π Ea πe2 Ea
after substituting for the integrals I0 (0, e), I0 (1, e) from Appendix B, Eq. (B.33).
These results were first obtained by Mindlin (1949).
When the ellipse has a high eccentricity e → 1, Eqs. (7.65, 7.66) are dominated
by the terms involving K (e) which is logarithmically unbounded in this limit5 so
Q x (1 + ν)K (e) Q y (1 − ν 2 )K (e)

ux → ; uy → ; e → 1. (7.67)
π Ea π Ea
The corresponding normal compliance is
Δ (1 − ν 2 )K (e)
CN ≡ = , (7.68)
P π Ea
from (2.34), so the tangential compliance in the direction of the minor axis approaches
the normal compliance, whilst that in the direction of the major axis is greater in the
ratio 1/(1−ν). In this limit, the problems approach a two-dimensional state of in-
plane and antiplane deformation respectively.6
5 see Eq. (3.39).

6 See Sect. 7.2.2.
7.6.3 Two Conjectures
Mindlin’s results for the ellipse show that the ratio between tangential and normal
compliance is bounded between the in-plane and antiplane two-dimensional limits
and it seems likely that this will be the case for more general contact geometries.
However, the author is unaware of any proof of this result.
We also note that Eqs. (7.65, 7.66) can be written in the form
U = C Q, (7.69)
where U = {u x , u y } and C is a Cartesian tensor whose principal values are

(1 + ν) ν [K (e) − E(e) ]
C1 = (1 − ν)K (e) +
π Ea e2

(1 + ν) ν [K (e) − E(e) ]
C2 = K (e) − . (7.70)
π Ea e2
Figure 7.9 shows the Mohr’s circle representation for the coordinate transforma-
tion of the components of C. In particular, the centre of the circle is defined by the
coordinate c = (C1 +C2 )/2 and the radius is R = (C1 −C2 )/2.
If the force Q is directed along the x axis, the corresponding components of U
will be
Ux = C x x | Q|; U y = C x y | Q|, (7.71)
and hence the angle between the applied force Q and the resulting tangential dis-
placement U will be
Cx y
α = arctan . (7.72)
C x x
The maximum value of α occurs when the line joining {C x x , C x y } to the origin is
tangent to the Mohr’s circle in Fig. 7.9 and hence
R C1 − C2
tan (αmax ) = = . (7.73)
c C1 + C2
Fig. 7.9 Mohr’s circle for

transformation of the Cxy
α max
compliance tensor C
C2 R
C1
0 c Cxx
7.6 Uniform Tangential Displacement in a Prescribed Area 127
If the tangential compliance for any area A is indeed bounded between the in-plane
and antiplane limits, the ratio C2 /C1 must lie in the range
C2 ν
1−ν ≤ ≤ 1 so 0 ≤ tan (αmax ) ≤ , (7.74)
C1 2−ν
and the maximum possible value occurs for ν = 0.5 and is 18.43 o .
For the special case of the ellipse,
C1 + C2 (2 − ν)(1 + ν)K (e)

c= = = RT C N , (7.75)
2 2π Ea
from Eq. (7.58). In other words, the average tangential compliance differs from the
corresponding normal compliance by the factor RT for all ellipticities e. It is tempting
to assume that this result applies to the tangential compliance tensor for an arbitrary
region A on the surface of an elastic half-space, but again this result has not been
proved, and indeed Problem 7.7 offers a possible counterexample.
7.7 Non-conformal Contact Problems with No Slip
Consider a non-conformal contact in which the bodies initially make contact at a

single point and the normal force increases, causing a monotonic increase in the
extent of the contact area A. As soon as contact is established at a given point, we
assume no-slip conditions, so ḣ(x, y) = 0, (x, y) ∈ A, but we cannot now integrate this
relation to conclude that h(x, y) = 0, since some shift will generally have occurred
during the period before the point (x, y) came into contact.
Figure 7.10 illustrates this process for the case where an elastic half-space is
indented by an axisymmetric rigid non-conformal punch under the influence of a
purely normal force P. In this figure, the dashed line shows the locus of points
that lie at the edge of the contact region at different values of the normal force P.
All points on the surface are initially outside this locus and hence are free to move
inwards as well as downwards until they establish contact at the dashed line, after
which they must move vertically downwards to satisfy the no-slip condition.
Problems of this kind are generally history-dependent and must be solved in an
incremental sense. If during a small time increment the forces P, Q increase by
sufficiently small increments δ P, δ Q, the contact area will increase only infinitesi-
mally, and all points in the instantaneous contact area will experience uniform elastic
displacements equal to δΔ, δU respectively in the normal and tangential directions.
Thus, the incremental problem has the same form as the no-slip flat punch problem
and the complete non-conformal contact solution can be constructed by an appropri-
ate convolution on the flat punch solution.
The principal challenge with this method lies in determining the relation between
the parameter(s) defining the contact area and the applied forces, both of which vary
with time. It is usually straightforward in uncoupled problems, since the normal
Fig. 7.10 Trajectory of

surface points during P
indentation by a
non-conformal rigid punch rigid
with no-slip conditions undeformed surface punch
deformed surface
contact radius r = a
contact problem can then be solved first, independently of the tangential tractions
as in Chaps. 3–6 and is not history-dependent. In coupled problems, both normal
and tangential loading must be considered together, but this is still relatively simple
provided the contact area can be characterized by a single parameter a(t), which may
represent the radius of a circular contact area or the semi-width of the contact strip
in a symmetric two-dimensional problem. We here illustrate the method by several
examples.
7.7.1 Uncoupled Hertzian Contact with Tangential Loading
Consider the case of an elastic sphere of radius R indenting an elastic half-space

of the same material. Contact starts at time t = 0 and the normal force P(t) is a
monotonically increasing function of time t, with P(0) = 0. The sphere is also loaded
by tangential forces Q x (t), Q y (t) with Q x (0) = Q y (0) = 0, but these forces are not
necessarily monotonic.
From Eq. (5.30), we conclude that the relation between the contact radius a and
the normal force P at time t is given by
1/3
4E ∗ a(t)3 3P(t)R
P(t) = ; a(t) = , (7.76)
3R 4E ∗
and it follows that the changes in the tractions p(r ), qx (r ), q y (r ) during the time
increment δt are non-zero only in 0 ≤r < a(t) and given by

H [a(t) − r ]δt d P d Qx d Q y
{δ p, δqx , δq y } =
, , , (7.77)
2πa(t) a(t)2 − r 2 dt dt dt
7.7 Non-conformal Contact Problems with No Slip 129
using (5.21, 7.56). Integrating in time, we therefore obtain

1 t0
d P d Qx d Q y H [a(t) − r ]dt
{ p, qx , q y }(t0 ) = , ,
, (7.78)
2π 0 dt dt dt a(t) a(t)2 − r 2
for the traction distributions at time t0 , or equivalently

2E ∗ a0
d Qx d Q y ada
{ p, qx , q y }(a0 ) = 1, , √ , (7.79)
πR r dP dP a2 − r 2
where (7.76) is used to express d P/dt in terms of a and da/dt, and a0 is the contact
radius at time t0 . These integrals can be evaluated for a wide range of loading scenarios
[see for example, Problem 7.8].
7.7.2 The Coupled Axisymmetric Problem under Purely

Normal Loading
Consider a fairly general axisymmetric contact problem with an initial gap function
g0 (r ) under no-slip conditions. The rigid body indentation Δ = f (a) and the indenting
force P(a) are as yet unknown functions of the contact radius a. However, during
the incremental problem in which the radius increases from s to s + δs, all points
within the instantaneous contact area 0 <r < s are displaced downwards without slip
by a distance f (s)δs, so the incremental force is given by Eq. (7.50) as

E ∗ s f (s)δs 1+β
δP = ln , (7.80)
β 1−β
and the incremental displacements outside the contact area are

r
δu z (r ) = f (s)δs φ , (7.81)
s
where the function φ(ρ) is defined in (7.49). Now the complete solution when s = a
can be constructed by superposition of incremental distributions, and in particular,
the normal displacement outside the contact area must be given by the integral
a r
u z (r ) = f (s)φ ds r ≥ a. (7.82)
0 s
This equation applies also to the limiting point r = a which is just coming into contact,
so the contact condition u z (r ) = Δ−g0 (r ) = f (a)−g0 (r ) enables us to write
a a
f (s)φ ds = f (a) − g0 (a), (7.83)
0 s
which is a Volterra integral equation for the unknown function f (a). Once this
equation is solved, a similar superposition yields the normal indenting force
a
E∗ 1+β
P(a) = ln s f (s)ds, (7.84)
β 1−β 0
and the contact tractions can be obtained by a similar integration using Eqs. (7.46,
7.47) with Δ = f (s)ds.
If the function f (t) is capable of the power series expansion
∞

f (s) = Cn s n , (7.85)
n=1
substitution in Eq. (7.83) gives

∞
1
1
g0 (a) = Bn a n where Bn = Cn 1 − n ρn−1 φ dρ . (7.86)
n=1 0 ρ
Thus, if the initial gap function g0 (r ) can be expressed as a power series, the corre-
sponding power series for f (s) can be written down by equating coefficients. If the
punch has a power law profile, f (s) will contain a single term, as in the Hertzian
case where
1 −1
r2 s2 1
g0 (r ) = and f (s) = 1−2 ρφ dρ . (7.87)
2R 2R 0 ρ
Spence (1968) gives numerical results for these coefficients for conical and quadratic
[Hertzian] indenters and shows that the tangential tractions reduce the radius of
the contact circle for a given normal force P relative to the frictionless case. The
maximum reduction in the Hertzian case is about 6% for β = 0.5.
If g0 (r ) cannot be expanded as a convergent power series, Eq. (7.83) can be solved
by expressing g0 (r ) and f (s) as Mellin transforms.7
7.7.3 The Coupled Two-Dimensional Problem
For two-dimensional problems, the above procedure must be modified slightly

because the rigid-body displacement Δ is then ill defined, as explained in Sect. 6.1.
7 See for example, Spence (1968), Appendix E, or Sneddon (1951) Sect. 6.

However, we can define the incremental problem in terms of the x-derivative of the
normal surface displacement.
For normal loading by a symmetric indenter, the contact area must be symmetric
and we denote by P(s) the normal force at which the contact area is defined by
−s < x < s. The incremental problem in which the contact area extends to the range
−(s + δs) < x < s + δs constitutes a ‘flat punch’ problem for the incremental force
P (s)δs, for which the incremental tractions are given by Eqs. (7.30, 7.32) as

P (s)δs s+x
δ p(x) =
√ cos ln (7.88)
π 1 − β2 s2 − x 2 s−x

P (s)δs s+x
δqx (x) =
√ sin ln . (7.89)
π 1 − β2 s2 − x 2 s−x
The corresponding increment in the x-derivative of the normal surface displacement

outside the contact area is given by (7.17) as

du z 2 s
δ p(ξ)dξ
δ =− |x| > s, (7.90)
dx πE∗ −s (x − ξ)
since δqx (x) = 0 outside the contact region. Substituting for δ p(x) from (7.88), we
obtain x
du z 2P (s)δs
δ =−
φ , (7.91)
dx π2 1 − β 2 E ∗ s s
where
1
1+ξ dξ
φ(ρ) = cos ln
, (7.92)
−1 1−ξ (ρ − ξ) 1 − ξ 2
and integration with respect to s then yields

a x P (s)ds
du z 2
=−
φ |x| ≥ a. (7.93)
dx π2 1 − β 2 E ∗ 0 s s
At x = a this must be equal to −g0 (a) and hence we have

a P (s)ds
a
π 2 1 − β 2 E ∗ g0 (a)
φ = , (7.94)
0 s s 2
which is a Volterra integral equation for P (t). As in the axisymmetric case, we can
solve this equation if g0 (r ) can be expressed as a power series by writing
∞

P (s) = Cn s n . (7.95)
n=1
Substitution into (7.94) then yields

∞
1
2Cn 1 n−1
g0 (a) = Bn a n where Bn =
φ t dt. (7.96)
n=1 π2 1 − β 2 E ∗ 0 t
7.7.4 Relaxation Damping
If a non-conformal contact passes through a closed cycle of normal and tangential

loading, there is generally some energy loss, even if there is no slip, which seems to
preclude work being done against friction. Popov et al. (2015) refer to this process
as ‘relaxation damping’.
To illustrate the process, consider the case of two axisymmetric elastic bodies
pressed together by a normal force P and loaded by a tangential force Q = {Q x , Q y }
in the interfacial plane.8 We consider a loading cycle describing a closed loop in
{P, Q x , Q y }-space, such that the minimum normal force is Pmin and the maximum is
Pmax . It is convenient to restrict attention to cases where P increases monotonically
from Pmin to Pmax during the loading phase and decreases monotonically during
unloading. We also assume that β = 0, so the normal contact problem can be solved
independently of the tangential loading. It then follows from Theorems 4 and 5
of Sect. 4.2, that the normal indentation Δ(P) and the contact radius a(P) are
unique non-decreasing functions of P only, and the incremental normal compliance
C N (P) = Δ (P) is a non-increasing function of P — i.e.
C N (P) = Δ (P) ≤ 0. (7.97)
Since the bodies are axisymmetric, the contact area will be circular for all P and
the matrix C of Eq. (7.69) will be given by C = IRT C N (P). It follows that the
incremental tangential displacement δU due to an incremental tangential force δ Q
under no-slip conditions is
δU = RT C N (P)δ Q. (7.98)
The work done by the normal force is exactly recovered on unloading, since Δ is
a unique function of P, so we concentrate here on the work done by the tangential
force Q. If the tangential force and displacement during the loading phase Ṗ > 0 are
denoted by Q A (P), U A (P) respectively, the work done during loading is
P=Pmax Pmax
WA = Q A · dU A = RT Q A · Q A C N (P) d P, (7.99)
P=Pmin Pmin
using (7.98). We can integrate this equation by parts to obtain
8 The following derivation is adapted from that of Ahn (2017).

P
RT | Q|2 C N (P) max RT Pmax
WA = − | Q A |2 C N (P) d P. (7.100)
2 Pmin 2 Pmin
Unloading
If the tangential force during unloading is denoted by Q B (P), the corresponding
tangential displacement will be
U B (P) = U A (P) + RT ( Q B − Q A ) C N (P). (7.101)
Differentiating this expression with respect to P and using (7.98) to substitute for
U A (P), we obtain
U B = RT Q B C N (P) + RT ( Q B − Q A ) C N (P). (7.102)
The work recovered during unloading is then defined by

P=Pmax
WB = Q B · dU B
P=Pmin
Pmax Pmax
= RT Q B · Q B C N (P)d P + RT Q B · ( Q B − Q A )C N (P)d P.
Pmin Pmin
(7.103)
Integrating by parts and using (7.100), we obtain the energy dissipation per cycle as

RT Pmax
W = WA − WB = − | Q A − Q B |2 C N (P) d P. (7.104)
2 Pmin
The incremental normal compliance C N (P) is non-positive from (7.97) and hence
energy is always dissipated in a closed cycle, except in the special case where
Q A (P) = Q B (P) for all P — i.e. the loading and unloading paths are identical.
The physical mechanism of relaxation damping
During loading, the tangential traction will always be square-root bounded at the
edge of the contact area, but during unloading, there will be a square-root singularity
whenever Q A (P) = Q B (P). Thus, the energy dissipation is analogous to that during
crack propagation, but the process here is not unstable, so with sufficiently slow
loading rates, it seems unlikely that significant energy would be radiated as elastody-
namic waves. In most cases, the assumption of complete stick will be a mathematical
approximation to the state where the coefficient of friction is in some sense large.
However, if the problem is re-solved using a finite coefficient of friction f , (i) energy
will be dissipated in microslip whenever Q A (P) = Q B (P) [see Chap. 9], and (ii)
the frictional energy dissipation approaches (7.104) asymptotically as f → ∞. Thus,
one might regard relaxation damping as a limiting case of frictional damping, rather
than as a new physical process. However, the singularity in tangential tractions dur-
ing unloading also implies the likelihood of local inelastic [and hence dissipative]
processes such as plastic deformation.
More general contact problems
Popov et al. (2015) established Eq. (7.104) using the method of dimensionality reduc-
tion,9 which is rigorously correct only for axisymmetric problems with a single circu-
lar contact area. However, they showed by numerical examples that the results remain
true for a wide range of contact geometries. The proof given here can be extended
to more general geometries if the ratio between tangential and normal compliance
remains constant as P varies. We conjectured in Sect. 7.6.3 that this ratio is bounded
in the range (1, 1/(1−ν)), and we conclude that (7.104) should provide a good
approximation to the dissipation for all problems characterized by the contact of
elastic half-spaces with arbitrarily large friction coefficient.
Problems
1. Use Eq. (7.5) to find the surface dilatation
∂u x ∂u y
e S (x, y) ≡ + at z = 0,
∂x ∂y
due to a concentrated tangential force Q.

Hence or otherwise show that for any uncoupled problem [β = 0] with no slip, the
tangential contact tractions q = {qx , q y } must satisfy the equation

(r − r 0 ) ·q dA
= 0,
A |(r − r 0 )|3
where r 0 is any point in A. Comment on the interpretation of the apparent singularity

in this integral as r → r 0 .
2. Table 7.1 shows representative values of the elastic constants for a selection of
materials. Estimate Dundurs’ bimaterial constant β for the following material com-
binations: carbon steel on rubber, stainless steel on glass, nylon on glass, rubber on
ice, compact bone on stainless steel. Within this set of materials, which combination
gives the greatest value of β and what is that value?
3. A two-dimensional flat rigid punch is pressed into an elastic half-space by a normal
force P, such that the contact area is defined by −a < x < a. A tangential force
Q y is now applied in the y-direction [antiplane]. Assuming no-slip conditions, use
Eq. (7.18) to determine the additional contact tractions due to the application of Q y .
9 See Sect. 5.5.

Problems 135
Fig. 7.11 Welded area

between two half spaces
radius a b
4. A two-dimensional flat rigid punch is pressed into an elastic half-space by a normal

force P, such that the contact area is defined by −a < x < a. The line of action of the
force passes through the point x = c, so some rigid-body rotation is to be anticipated,
as in Fig. 6.5 a. Use the method of Sect. 7.3 to determine the resulting contact tractions
under the assumption of full contact and no slip.
5. If the Goodman approximation is applied to the problem of a normally loaded
axisymmetric flat punch, the first equation in (7.43) is simplified to
∂ϕ E ∗Δ
=−
∂z 2
and the solution for h 1 (t) is given in Sect. 5.1.1. Use this result and other results from
Sect. 5.1 to define an integral equation for the function h 2 (t), and hence determine
the tangential tractions σzr (r ) in the contact area.
6. Show that if the contact area A has three-fold symmetry [for example, like an
equilateral triangle], then the tensor C in Eq. (7.69) must be isotropic — i.e. C = C I,
where I is the identity matrix.
Two identical half-spaces are spot-welded together such that the welded area
comprises three circles of radius a whose centres are located at the vertices of an
equilateral triangle of side b, as shown in Fig. 7.11. The circles do not overlap, so
b > 2a.
Estimate the tangential compliance C by using Eq. (7.57) for the displacement at a
circle due to a force distributed over the same circle, and approximating the effect of
forces at the other circles by replacing them by equal point forces at their respective
centres. Comment on the possible relevance of your results to the second conjecture
in Sect. 7.6.3.
7. Due to surface roughness, two half-spaces of the same material make contact at a
large number N of small elliptical areas each of semi-major axis a and eccentricity
e. These ellipses are of random orientations and are sufficiently widely separated for
Table 7.1 Elastic properties of selected materials

Material Carbon Stainless Brass Compact Nylon Glass Granite Ice Aluminium Rubber
steel steel steel steel
E (GPa) 210 190 110 13.8 3.0 70 49 9.0 72 0.10
ν 0.3 0.3 0.33 0.42 0.4 0.24 0.28 0.33 0.31 0.49
their elastic fields to be independent of each other. Use Eqs. (7.68, 7.70) to estimate
the normal and tangential compliances of the system and comment on the effect of
ν and e on the ratio between them.
8. The uncoupled Hertzian contact of Sect. 7.7.1 is subject to the monotonically
increasing forces
0.075Ct 2
P = Ct; Q x = 0.2Ct; Qy = , where 0 < t < t0 .
t0
Find the contact tractions p, qx , q y under no-slip conditions as functions of x, y and

time t, and hence determine the minimum coefficient of friction f if this is to be a
reasonable assumption.
9. Use the Goodman approximation to estimate the tangential tractions developed
when a rigid cylinder of radius R is pressed into an elastic half-plane by a normal
force P. The normal tractions and the extent of the contact area will then be given
by the two-dimensional Hertzian solution, but the incremental tangential tractions
include the effect of normal-tangential coupling and are given by Eq. (7.55) [with
Q x = 0].
10. An elastic sphere of radius R is pressed into an elastic half-space and then
subjected to the periodic loading
P = P0 + P1 cos(ωt); Q x = Q 1 sin(ωt); Q y = 0,
where P0 > P1 > 0. Find the energy dissipation per cycle if both materials have
Young’s modulus E and Poisson’s ratio ν, and the friction coefficient is sufficiently
large to prevent all slip.
11. For a general uncoupled three-dimensional contact problem, the incremental
tangential compliance C(P) under no-slip conditions is a Cartesian tensor function
of the normal force P which we write in the form
∂U
= C(P) = Λ(P)C N (P),
∂Q
where C N (P) is the corresponding normal compliance. Show that Eq. (7.104) for the
energy dissipation per cycle is then generalized to

1 Pmax
W =− ( Q A − Q B ) · Λ(P)( Q A − Q B ) C N (P)d P.
2 Pmin
If we assume that Λ(P) is bounded between the in-plane and antiplane limits, as
conjectured in Sect. 7.6.3, can we place corresponding bounds on the error associated
with the approximation Λ(P) = RT I, where I is the identity matrix?
Chapter 8
Friction Laws
The tangential tractions discussed in the previous chapter must be associated with
some mechanism such as friction, unless the bodies are bonded [welded] together, in
which case we are not strictly in the domain of contact mechanics. The investigation
of the nature of such frictional tractions is more properly a subject for the related
discipline of Tribology, though we shall touch on the subject in Sect. 8.6, and also in
Chap. 16 where we investigate the contact of rough surfaces. In most of this book,
we shall make use of the simple theory of friction associated with the names of
Amontons and Coulomb, according to which the maximum tangential force | Q| that
can be transmitted between two bodies at a plane interface is linearly proportional
to the compressive normal force P and independent of the area of contact.
8.1 Amontons’ Law
We distinguish between states of stick and slip, such that during slip the magnitude
of the tangential force is
| Q| = f P, (8.1)
where f is the coefficient of friction, and the direction of the force must be such as
to oppose the slip velocity Vs as shown in Fig. 8.1.
More generally, we have
Q Vs
= , (8.2)
| Q| |Vs |
where the vector Q defines the tangential reaction on the lower body. During periods
of stick,
Vs = 0 and | Q| ≤ f P. (8.3)

https://doi.org/10.1007/978-3-319-70939-0_8
138 8 Friction Laws
Fig. 8.1 Frictional reactions F

induced during slip
Vs
Q
P
Q
We shall generally assume that the same coefficient of friction f applies in both
Eqs. (8.3, 8.1), in other words that the so-called static and dynamic coefficients of
friction are equal. Some of the implications of differing coefficients are discussed in
Sect. 8.6.3.
Numerous experimental investigations have documented significant deviations
from Amontons’ law, particularly at very small length scales, but it remains a remark-
ably good approximation to frictional behaviour at the macroscale and its mathemat-
ical simplicity recommends it for use in many engineering applications, if only as
an approximation.
8.1.1 Continuum Problems
The statement of Amontons’ law in Eqs. (8.1)–(8.3) is appropriate for the contact of
rigid bodies, and it can clearly be extended to systems of rigid bodies and springs, and
hence by extension to finite element models of contact, where the friction conditions
will be imposed at a discrete set of contact nodes, some of which may stick and some
slip. However, in problems for the continuum, we must restate the law in terms of
contact tractions [forces per unit area] obtaining
q(x, y) ḣ(x, y)
|q(x, y)| = f p(x, y) and = (8.4)
|q(x, y)| | ḣ(x, y)|
in slip regions, and
|q(x, y)| ≤ f p(x, y) and ḣ(x, y) = 0 (8.5)
in stick regions, where ḣ is defined in Eq. (7.2) and implies the sign convention of
Fig. 7.1, and q(x, y) is the tangential traction acting on the lower body 1.
8.1 Amontons’ Law 139
8.1.2 Two-Dimensional Problems
In three-dimensional problems, the slip velocity ḣ(x, y) and the frictional traction
q(x, y) are vectors within the plane, but in two-dimensional problems, we can only
distinguish between forward slip [ḣ(x) > 0] and backward slip [ḣ(x) < 0]. The
friction law for the continuum can then be stated as p(x) ≥ 0 for all x and
ḣ(x) = 0 g(x) = 0 |qx (x)| ≤ f p(x) stick

qx (x) = f p(x) g(x) = 0 ḣ(x) >0 forward slip
(8.6)
qx (x) = − f p(x) g(x) = 0 ḣ(x) <0 backward slip
qx (x) = 0 p(x) = 0 g(x) >0 separation,
where we have supplemented the frictional conditions by the normal contact condi-
tions (1.3). Notice that every point on the interfacial plane must be in one of four
states, separation, stick, forward slip or backward slip, and if the state is known for
all such points, we then have two linear equations for each x for the two unknowns
p(x), qx (x). We also note that the equations are supplemented by inequalities which,
as in frictionless contact, serve to determine the partition of the plane into regions of
each state. By contrast, the three-dimensional condition (8.4) is pointwise nonlinear,
even if the partition of the interfacial plane into zones in different states is known.
8.1.3 Existence and Uniqueness Theorems
Equations (8.6) show that frictional contact problems have some similarities with the
simpler frictionless problem of Eq. (1.3) and numerous investigators have endeav-
oured to establish theorems of existence and uniqueness. In this context, it is impor-
tant to distinguish between the discrete problem, where there exists a denumerable
set of discrete contact points, and the continuum problem discussed above, where
contact may occur at a non-denumerable set of real points. In the engineering liter-
ature, finite element methods are often used to reduce the continuum problem to a
discrete problem, where the contact and frictional conditions are imposed at a finite
set of contact nodes. Underlying such investigations lies the assumption that with
a sufficiently fine mesh, the solution will in some sense approach the continuum
solution asymptotically. However, from a mathematical perspective, the two prob-
lems are quite different and indeed considerably more progress has been made in the
investigation of the discrete problem.
The problem defined by (8.4) or (8.6) is history-dependent because it contains
the time derivative ḣ. In the mathematical literature, this is described as the rate
problem and is contrasted with the static problem in which conditions are imposed
on the instantaneous values of h rather than ḣ. From an engineering perspective, the
static problem formulation is meaningful if and only if the loading is proportional
and monotonic and all contact points were already in contact at zero force, in which
case ḣ can legitimately be integrated in time.
140 8 Friction Laws
Difficulties are encountered with both existence and uniqueness proofs for the
general quasi-static elastic contact problem with Coulomb friction.1 Some of these
difficulties can be avoided by ‘regularizing’ the friction law [i.e. smoothing the
discontinuities associated with the inequalities] or using a non-local friction law,
but uniqueness can still generally only be established subject to the proviso that the
coefficient of friction f should be ‘sufficiently small’ (Oden and Pires 1983). In
other words, there may exist a system-specific critical coefficient f 0 such that for
f > f 0 multiple solutions or non-existence of solution may occur. This was a serious
embarrassment to the developers of finite element codes, since these are required to
be ‘robust’ and capable of use by practitioners lacking a detailed knowledge of the
underlying algorithms.
8.2 The Klarbring Model
Insight into the behaviour of frictional systems was considerably advanced by

Klarbring’s investigation of the simple two-dimensional single node system (Klar-
bring 1990) shown in Fig. 8.2a.
A point mass M makes frictional contact with a rigid plane and is also supported
by a generalized linear spring K . The stiffness matrix K must be positive definite, so
k11 , k22 > 0, and we can choose the direction of x1 so as to ensure that k12 = k21 > 0
without loss of generality. A time-varying force F(t) = {F1 , F2 } is applied to the
mass and the resulting motion is described by a displacement vector u(t) = {v, w}.
Figure 8.2b shows a free-body diagram of the mass and in particular identifies the
normal and tangential reactions p, q respectively at the contact interface. Notice that
the positive direction of the tangential reaction q is opposite to that shown in Fig. 8.1,
since here the focus is on the equilibrium of the upper body, and it is convenient to
use the same coordinate system for {q, p} and {v, w}.
We assume that the force is applied sufficiently slowly for the response to be
quasi-static. In other words, the system passes through a sequence of equilibrium
states, so at all times t,
F1 + q − k11 v − k12 w = 0; F2 + p − k21 v − k22 w = 0. (8.7)
(a) (b)
K F2
F x2
M k 21 v + k 22 w
u k11 v + k12 w
x1 F1
p q
Fig. 8.2 (a) The Klarbring model, (b) Free-body diagram
1 See, for example, Kikuchi and Oden (1988).

8.2 The Klarbring Model 141
For this single node system, the discrete equivalents of conditions (8.6) are p ≥ 0
and
v̇ = 0 w=0 |q| ≤ f p stick
q = −fp w=0 v̇ > 0 forward slip
(8.8)
q = fp w=0 v̇ < 0 backward slip
q=0 p=0 w >0 separation.
Consider the case where the system is initially unloaded with v = w = 0 and the
external force F(t) = Ct, where C = {C1 , C2 } is a constant vector. In other words, the
two force components F1 , F2 are increased monotonically and in proportion. Using
(8.7, 8.8) to determine p, q, v, w, we then find that the inequalities governing the
four states are
f C2 < C1 < − f C2 stick

C1 + f C2 k11 C2 − k21 C1
> 0; <0 forward slip
k11 + f k21 k11 + f k21
C1 − f C2 k11 C2 − k21 C1
< 0; <0 backward slip
k11 − f k21 k11 − f k21
k21 C1 < k11 C2 separation. (8.9)
The assumed loading scenario corresponds to a straight line starting from the origin
in {F1 , F2 }-space and Fig. 8.3 shows the states predicted by (8.9) for a case where
k21 > 0 and f < k11 /k21 . In particular, we see that one and only one state is possible
for each set of values {C1 , C2 } and that the same state will persist as long as the
loading trajectory is defined by the straight line F(t) = Ct.
Fig. 8.3 Domains of stick, F2 k11

slip and separation for F1 = F
k 21 2
f < k11 /k21 and F(t) = Ct
separation
F1
backward forward
slip slip
F1 = fF2 stick F1= - f F2

142 8 Friction Laws
(a) F2 (b) F2
P
separation separation
F1 Q F1
P
Q
Fig. 8.4 Movement of the stick cone
8.2.1 General Loading Scenarios
Figure 8.3 can be modified to allow general loading scenarios. The separation seg-
ment remains unchanged, but the stick cone is moved by slip displacement v such
that its vertex always lies on the separation line F1 = (k11 /k21 )F2 .
During periods of slip, as the instantaneous point P(F1 , F2 ) moves to Q in
Fig. 8.4a, it ‘pushes’ the vertex of the stick cone along this line. Separation, when
it occurs, is always reached through the vertex of the cone and when a transition
from separation to contact occurs, it always occurs through the vertex, as shown in
Fig. 8.4b. The state realized immediately after contact depends on the direction of the
local force trajectory and hence on the ratio Ḟ1 / Ḟ2 . The case illustrated in Fig. 8.4b
will transition from separation to forward slip.
8.2.2 The Critical Coefficient of Friction
The arguments in the last section essentially establish that the rate problem for
Klarbring’s model has a unique solution for all possible loading scenarios. However,
Figs. 8.3 and 8.4 are drawn for the case f < k11 /k21 . If instead the coefficient of
friction f > k11 /k21 , the diagram has the form of Fig. 8.5, where we see a domain in
which stick, backward slip and separation are all possible. Thus, we lose uniqueness
of solution.
More seriously, if we devise a loading scenario that starts in the stick domain and
then moves through this multi-state region, the stick inequalities are satisfied in the
strict sense so stick will persist until we reach the upper boundary beyond which only
separation is possible. However, the separation state so reached involves a non-zero
value of w, so the system seems to make a discontinuous jump, which violates the
quasi-static assumption.
8.2 The Klarbring Model 143
Fig. 8.5 Domains of stick, F2 k11

slip and separation for F1 = F
f > k11 /k21 k 21 2
separation
forward
F1 = fF2 slip
backward
slip F1
F1= - fF2
stick
A scenario involving discontinuities in displacement was also predicted by Mar-

tins and co-workers2 for the case where damping is included in the system and then
allowed to approach zero asymptotically. These authors also note that existence and
uniqueness theorems can be established with arbitrary coefficient of friction if the
requirement of continuity of displacement is relaxed.
Cho and Barber (1998) showed that if the quasi-static assumption is relaxed—i.e.
if the inertia term M ü is included in Eq. (8.7), the discontinuous jump is resolved
into a dynamic event on the time scale of the local dynamics, and the backward slip
state in the multiple solution range is unstable, tending either to separation or stick
depending on the exact initial conditions. However, in more complex discrete systems
such as finite element models, the replacement of quasi-static by dynamic behaviour
involves a major increase in computational effort. In simple systems such as the
Klarbring model, one can devise modified quasi-static algorithms that capture the
long-time behaviour of the system without resolving the dynamics of the occasional
jumps. However, to the authors’ best knowledge, no such algorithm is yet available
for complex systems above the critical friction coefficient.
8.2.3 Wedging
Frictional systems are capable of becoming ‘wedged’ if the coefficient of friction is

sufficiently high. We define a wedged state as one in which the system can support
internal forces even when all external forces are set to zero. Wedging is important in
automated assembly systems, where it might result in an incorrect assembly.
For the Klarbring model, if a wedged state is accessible, it implies that we can
devise a loading scenario F(t) in 0 < t < t0 such that F(0) = 0, F(t0 ) = 0, w(t0 ) = 0
and v(t0 ) = v0 = 0. In other words, the system is left in a state of stick with non-zero
tractions
2 Martins and Oden (1987), Martins et al. (1992, 1994).

144 8 Friction Laws
q(t0 ) = −k11 v0 ; p(t0 ) = k21 v0 , (8.10)
from (8.7). The stick inequalities then require that k21 v0 > 0 and
k11
k11 < f |k21 | or f > . (8.11)
|k21 |
Notice that in this case, the critical coefficient of friction for wedging coincides
with that at which the existence and uniqueness theorems for the rate problem break
down. However, no general proof exists of this equivalency for more complex discrete
systems, and indeed counterexamples can be devised.
8.3 Multinode Systems
These arguments can be generalized by considering a discrete two-dimensional sys-

tem with N contact nodes i ∈ (1, N ). We can then define N -vectors with components
pi , qi , wi , vi and if the system is linearly elastic, these quantities must be linearly
related through the equations

qi qiw Ai j Bi j vj
= + , (8.12)
pi piw B ji Ci j wj
where piw , qiw are the contact tractions that would be produced by the time-varying
external forces if the nodes were all welded in contact with vi = wi = 0.
It is convenient to think of such a system as a generalization to N nodes of
Fig. 8.2a, but a mathematically equivalent system can be generated by discretizing a
continuum elastic contact problem by the finite element method, provided we restrict
attention to small strains. In this case, nodal forces are non-zero only at the contact
nodes and the externally loaded nodes, so the degrees of freedom at internal nodes
can be eliminated, reducing the system to the form of Eq. (8.12). The sign convention
for contact nodal forces and displacements is then as defined in Fig. 8.6.
Notice that

Ai j Bi j
K= (8.13)
B ji Ci j
Fig. 8.6 Sign convention for

(a) (b)
(a) nodal forces and (b)
nodal displacements in a wj
finite element discretization,
leading to Eq. (8.12)
qj vj
pj
8.3 Multinode Systems 145
is the reduced stiffness matrix for the system and hence must be symmetric and
positive definite. It follows that A and C are also symmetric and positive definite,
but no such restriction applies to the matrix B which defines the coupling between
normal and tangential effects. This coupling matrix is the discrete analogue of the
coupling terms with multiplier β in the continuum problems of Chap. 7, and as in that
case, it has a profound effect on the qualitative behaviour of the frictional system.
8.3.1 The Evolution and Rate Problems
If the external forces qiw (t), piw (t) are known functions of time, we wish to be able to
determine the displacements v j (t), w j (t) such that Eq. (8.12) and conditions (8.8) are
satisfied at all nodes i and all times t. We shall refer to this as the evolution problem.
The presence of time derivatives in conditions (8.8) implies that the evolution problem
can be solved if and only if a corresponding rate problem can be solved. In two
dimensions the rate problem can be stated as:
Given a set of nodal forces and displacements that satisfy Eq. (8.12) and the
conditions
|qi | ≤ f pi ; wi ≥ 0; pi ≥ 0; wi pi = 0 (8.14)
at each node i, and given the loading rates q̇iw , ṗiw , determine the time derivatives
q̇i , ṗi , v˙i , ẇi such that each node satisfies the conditions for one and only one of the
four states
v̇i =0 ẇi = 0 |qi | ≤ f pi stick

q̇i = f ṗi ẇi = 0 v̇i >0 forward slip
(8.15)
q̇i = − f ṗi ẇi = 0 v̇i <0 backward slip
q̇i =0 ṗi = 0 wi >0 separation.
Notice that at any node i where either of the strict inequalities |qi | < f pi or wi > 0
holds, the corresponding state (stick or separation respectively) must continue into
the solution of the rate problem. However, nodes where |qi | = f pi may evolve into
either stick or slip [in the appropriate direction], and nodes where both wi = 0 and
pi = 0 may evolve into any of the four states, depending on the external loading rate.
8.3.2 Algorithms for Two-Dimensional Problems with

Time-Varying Forces
A simple numerical algorithm for the evolution problem is to assume provisionally

that the states holding at each node immediately prior to time t continue to hold
through a finite time increment Δt. We then use (8.12) and the equations in (8.8) to
146 8 Friction Laws
determine the new nodal forces and displacements, and compare the resulting values
and implied rates against the respective inequalities. If any inequality is violated,
the nature of the violation is a clue to the correct state at the node in question. For
example, if we assume node i is in forward slip and we detect that Δvi < 0 during
Δt, then the most likely alternative state is one of stick. Ahn and Barber (2008)
developed an algorithm of this kind in which the nodes are solved for sequentially
in a Gauss–Seidel sense.
An alternative approach is to recognize that as long as the states at all the nodes
remain the same, the evolution of the discrete solution is defined by linear equations.
If the external forces can also be described as piecewise linear functions of time, this
opens up the possibility of a linear complementarity solution. Briefly, we assume
that the states remain the same at all nodes in which case the equality conditions are
linear and we can solve for the nodal forces and displacements as linear functions of
time. We then substitute the resulting expressions into the inequalities and solve for
the earliest time at which a violation occurs at any node. We then jump to this time
and ‘pivot’ about it starting a new time step with an appropriate change of state at
the pivoted node. This method was first applied to the friction problem by Klarbring
and Björkman (1988) and an improved algorithm is given by Bertocchi (2009).
8.3.3 History-Dependence and Memory
The necessity of a rate formulation shows that the instantaneous state of a system
involving friction can depend on the history of loading as well as on the instantaneous
forces. Thus, frictional systems fall into the broader class of systems with ‘memory’.
It is clear from conditions (8.8) that the system memory is associated with the fact
that at a ‘stick’ node, only the velocity v̇i is determined from the instantaneous
conditions, the actual position of the node vi depending on the previous history.
Thus, it is reasonable to consider the memory of the system as residing in those
nodes that are instantaneously stuck, and as being defined in some sense by the
displacements vi at those nodes.
Some insight into this question and into the frictional evolution problem, in gen-
eral, can be gained by considering the motion of the point P(v1 , v2 , . . . v N ) in the
N -dimensional hyperspace of slip displacements. Here we shall restrict attention to
the simpler case where the external loading ensures that all the nodes remain in con-
tact at all times [w(t) = 0]. Equation (8.12) and the frictional inequality |qi | ≤ f pi at
node i then imply that
− f piw − f B ji v j ≤ qiw + A ji v j ≤ f piw + f B ji v j , (8.16)
or
8.3 Multinode Systems 147
Fig. 8.7 Motion of the

instantaneous state P(v1 , v2 )
due to changes in external v2
loading
III
I
P4 P3
II P2
IV P1
v1
(A ji − f B ji )v j ≤ f piw − qiw and (A ji + f B ji )v j ≥ − f piw − qiw ,

(8.17)
where summation is implied over repeated indices. Each of these inequalities [two
for each node] excludes the region on one side of a hyperplane in v-space and
when the node in question is slipping, P lies on this hyperplane [corresponding to
an appropriate equality in (8.17)]. The external forces pw (t), q w (t) appear on the
right-hand sides of these inequalities and hence each hyperplane moves in time whilst
preserving the same inclination, which is determined by the elements of the matrices
A and ± f B. Thus, the hyperplanes ‘push’ the point P about the space, the local
motion being determined by the appropriate direction of slip corresponding to the
active inequalities (Ahn et al. 2008).
Figure 8.7 illustrates this process for a simple two-node system, where there are
four inequality constraints I, II, III and IV governing forward and backward slip at
each of the two nodes. For example, during backward slip at node 1 [v̇1 < 0], P
must lie on I, whereas during forward slip at node 2 [v̇2 > 0], it must lie on IV.
The region excluded by each hyperplane is shown shaded in the figure, so at any
instant P must lie inside the unshaded region. However, the instantaneous position
within this region depends on the history of loading. If [for example] the changes in
pw (t), q w (t) cause IV to move upwards to the point P3 , after which I moves to the
left, the point P would follow the trajectory shown.
8.3.4 Klarbring’s P-Matrix Criterion
Under some circumstances, the point P can become ‘trapped’ between two or
more constraints, so that no further evolution satisfying the friction law can occur.
Figure 8.8 shows such a case, where the constraints I and IV govern slip in directions
v̇1 < 0 and v̇2 > 0 respectively. As the constraints move, both directions of motion
are separately possible, but if P reaches the intersection between the constraints as
148 8 Friction Laws
Fig. 8.8 P becomes trapped

between constraints I and IV
v2 P
I
IV
v1
shown, further reduction in the permissible region leads to a contradiction that cannot
be resolved by a legitimate motion of P.
Similar problems can arise at all edges and corners of the instantaneously admis-
sible hypervolume in the N -node case. At a corner where M hyperplanes intersect,
the condition for a situation analogous to that in Fig. 8.8 to arise can be stated in
terms of a subset of M of the 2N constraints (8.17), involving only M of the slip dis-
placements vi , and with appropriate signs for f at each node to reflect the direction
of slip that is prevented.
The resulting criterion was enunciated by Klarbring (1999), who showed that for
two-dimensional problems, the rate problem is well-posed [i.e. situations analogous
to that in Fig. 8.8 cannot arise], if and only if all matrices of the form ( A+ΛB) are
P-matrices [i.e. all principal minors of the matrix are positive definite], where Λ is
any diagonal matrix each of whose diagonal elements3 is either + f or − f .
Since A is a positive-definite stiffness matrix, this condition is clearly satisfied for
f = 0, and for given matrices A, B [with B non-null] there therefore exists a critical
coefficient of friction f cr above which the P-matrix criterion fails and the rate problem
is ill-posed. In the special case where N = 1, each of the matrices reduces to a single
element A = k11 , B = B T = k12 , C = k22 , and we recover f cr = A11 /|B11 | = k11 /|k12 |,
as in Sect. 8.2.2.
For N = 2, Ahn (2010) has shown that, as in the Klarbring model, a dynamic
solution with damping for f > f cr approximates to a unique quasi-static solution
with displacement jumps. The present author is unaware of any proof of this result
for N > 2.
8.4 Periodic Loading
Many engineering applications involve periodic [cyclic] loading, due to vibration or

repetitive operations, and the energy dissipated in friction under these circumstances
can be an important measure of system performance. It has been estimated that
frictional hysteresis in assembled structures accounts for more energy dissipation
3 Klarbring allows the possibility that the coefficient of friction f may be node-dependent, in which
case the elements of Λ must be taken as ± f i .
8.4 Periodic Loading 149
than internal material damping, but the effective damping in such cases is notoriously
difficult to quantify. Also, the energy dissipated in microslip regions produces a
favourable environment for the initiation of fretting fatigue cracks (Lovrich and Neu
2006; Nowell et al. 2006), which are a primary mode of failure in many nominally
static contacting systems subject to vibration, including notably the blade root contact
in aero engines (Murthy et al. 2004).
We anticipate that the system will eventually reach a steady periodic state under
cyclic loading, but since frictional systems are history-dependent, we might reason-
ably ask whether this steady state is a unique function of the loading, or whether
it also depends on the initial conditions, or on an initial period of transient loading
leading to the steady state. For example, for a bolted joint, there might be a depen-
dence on the bolt-tightening protocol. We shall prove here that important features of
the steady state are indeed unique in uncoupled systems [B = 0], but that in coupled
systems we must anticipate a dependence on initial conditions.
8.4.1 A Uniqueness Proof for Uncoupled Systems
If the problem is uncoupled, the complementarity conditions wi ≥ 0, pi ≥ 0, wi pi = 0

determining the normal contact reactions are independent of the frictional problem
and hence are not history-dependent. Thus, the normal forces are unique functions
of the instantaneous external forces pw (t).
Suppose that there exist two time-varying solutions q 1 (t), v 1 (t) and q 2 (t), v 2 (t),
corresponding to the same external loading scenario pw (t), q w (t), but different initial
conditions v 1 (0), v 2 (0) respectively. The following argument applies to both two-
and three-dimensional problems. In the latter case, the tangential nodal tractions
q i and displacements v i are two-dimensional vectors, whereas in two-dimensional
problems, they are scalars.
Following Klarbring et al. (2007), we define an ‘energy norm’
1
E= (v 1 − v 2 )T ·A(v 1 − v 2 ), (8.18)
2
where the stiffness matrix A is defined in (8.12). The norm E can be regarded as
a scalar measure of the difference between the two solutions. Differentiating with
respect to time, we obtain
Ė = (v̇ 1 − v̇ 2 )T ·A(v 1 − v 2 ) = (v̇ 1 − v̇ 2 )T ·(q 1 − q 2 ), (8.19)
using (8.12) with B = 0 and recalling that q w (t) is the same for both solutions. Now
if a given node i is stuck at time t in both solutions, v̇ i1 = v̇ i2 = 0, and the contribution
150 8 Friction Laws
Fig. 8.9 The nodal force q i2

lies strictly within the
friction cone |q i | = f pi , qi1 qi2
whereas, q i1 lies on the cone qi2
since the node slips in
solution 1
O
qi1 .
v 1i
qi = fpi
of that node to Ė is zero. In the two-dimensional case,4 if the given node is slipping
in the same direction, or separated in both solutions, q i1 = q i2 and we can therefore
conclude that Ė = 0 during any period in which the nodes in the two solutions are in
the same state.
Now consider the case where node i is stuck at time t in solution 2 (v̇ i2 = 0),
but slipping in solution 1. There will then generally be a non-zero contribution to Ė
given by v̇ i1 ·(q i1 −q i2 ). Now the friction law (8.1)–(8.3) demands that |q i2 | < f pi and
|q i1 | = f pi , with direction diametrically opposed to v̇ i1 . This situation is illustrated
in Fig. 8.9, where it is clear geometrically that v̇ i1 ·(q i1 −q i2 ) < 0, and hence that the
contribution to Ė from node i is negative.
Similar arguments can be applied to other combinations of nodal states and all
show that the contribution to Ė is negative from any node at which the states in
solutions 1 and 2 differ (Andersson et al. 2011). Thus, E must continue to decrease
with time during any period in which one or more nodes in the two solutions are in
different states. Eventually, either (i) E = 0 indicating that the two solutions converge
on a unique steady state, or (ii) E converges on a constant non-zero value, implying
a situation where the states of the nodes at any given time are the same in the two
solutions, but the displacements may still differ.
It can then be shown that case (ii) can arise if and only if there exists a set T
of ‘permanent stick nodes’, such that for i ∈ T , v̇ i (t) = 0 for all t in the steady
state. If T is null, the system always converges on a unique steady state. This is
consistent with the concept advanced in Sect. 8.3.3 that the memory of the system
resides in nodes that are stuck, so that long-term memory depends on the existence
of nodes that are permanently stuck. Also, at nodes j ∈ / T , the slip velocities v̇ 1j , v̇ 2j
and tractions q j , q j must be equal, so the energy dissipated in friction at each node
1 2
is unique, which as explained above has important consequences for fretting fatigue
and damping.
4 Forthe three-dimensional case, node i may slip in different directions in the two solutions, in
which case q i1 = q i2 . However, it can then be shown that the contribution to the norm Ė < 0 [see
Problem 8.5].
8.4 Periodic Loading 151
8.4.2 Shakedown
A special case of the above result applies to the case where all nodes belong to
the set T so that there is no slip during the steady state. This is analogous with
the state known as ‘shakedown’ for elastic-plastic structures. We conclude that if
there exists a state such that, once reached, all nodes would lie within the friction
cone for all q w (t), pw (t), then the system will eventually shake down from any
initial condition (Klarbring et al. 2007). This is the frictional equivalent of Melan’s
theorem in plasticity (Melan 1936).
8.4.3 Coupled Systems
The above proof depends on the uncoupled assumption B = 0, and indeed counterex-
amples to Melan’s theorem can be found for all cases where B = 0. We recall that
even in the uncoupled state, different initial conditions can lead to different values of
the locked-in slip displacements in the permanently stuck nodes T . If B = 0, these
will imply different normal reactions pi at the remaining nodes, and consequent
differences in frictional [slipping] tractions and displacements.
If shakedown is possible for a coupled two-dimensional system, there must exist
some ‘safe shakedown space’ S in the slip displacement diagram of Fig. 8.7, com-
prising all points that are never excluded by any of the frictional constraints at any
time t during the steady state. Ahn et al. (2008) have shown that a two-dimensional
N -node system will shakedown from all initial conditions if all the frictional inequal-
ities (8.17) are active constraints in defining S.
If the periodic loading is expressed in the form
pw (t) = p0 + λ p1 (t); q w (t) = q 0 + λq 1 (t), (8.20)
where λ is a scalar load factor, there will generally exist some critical value λ1 such
that for λ < λ1 Ahn’s criterion is satisfied, and a larger value λ2 above which S
becomes null. For λ1 < λ < λ2 , shakedown will depend on the initial conditions,
whereas, for λ > λ2 shakedown is not possible. For the simple two-node case of
Fig. 8.5, S will be a quadrilateral for λ < λ1 , but this region shrinks as λ is increased
and becomes a triangle for λ > λ1 which collapses to a point at λ = λ2 .
8.4.4 Asymptotic Approach to a Steady State
The steady periodic state may be reached after a few loading cycles, or it may be
approached asymptotically, depending on the loading history and other characteris-
tics of the system. For example, Fig. 8.10 shows the trajectory for a two-node system
152 8 Friction Laws
Fig. 8.10 Motion of the

point P to the final stuck
point A by a series of v2 I
geometrically decreasing
slips
A
IV
P
v1
in which, during each cycle, constraint IV moves from the dashed position to the
solid position and then recedes, after which constraint I exercises a similar motion.
Starting from the point P, the system will then follow the trajectory shown, and in
particular, the final ‘safe’ point A will be reached by an infinite series of geometrically
decreasing slips. This figure also illustrates that a node that slips during the transient
period but is stuck in the steady state will generally experience at least one instant
of incipient slip5 per cycle, since the motion of the constraint that pushed it to its
final position will be repeated in each subsequent cycle. We shall find this category
of behaviour exhibited by continuous systems in subsequent chapters. It is also a
commonly observed feature of finite element solutions for frictional systems subject
to periodic loading. Since the slips decrease geometrically during the final asymptotic
approach to the steady state, it is possible to reduce computing time in such cases by
summing the geometric series, based on results from a modest number of cycles.
8.5 A Simple Continuum Frictional System
To illustrate some of the features of frictional contact problems in a simple continuum

context, we consider the one-dimensional system illustrated in Fig. 8.11. A rectan-
gular elastic strip of cross-sectional area A is pressed against a rigid plane surface
by a uniform force w per unit length, after which a monotonically increasing tensile
force F0 is applied to one end. The coefficient of friction between the strip and the
plane is f .
If |F0 | < f wL, the entire contact area cannot slip, and hence there must exist a
stick region in which the sliding velocity u̇ x = 0. For this region, integrating in time,
we obtain u x = 0 and hence the axial strain
5 Meaning that the tractions just reach the limiting condition |q i | = f pi , usually only momentarily,
but no slip occurs.
8.5 A Simple Continuum Frictional System 153
Fig. 8.11 An elastic strip w per unit length

pressed against a rigid plane
surface F0
L
x
du x
ex x = = 0. (8.21)
dx
Neglecting Poisson’s ratio effects, it then follows that the axial stress σx x = 0 and
hence the local value of tensile force in the stick region is
F = Aσx x = 0. (8.22)
Consider now the equilibrium of a small element of the strip of length δx, as
shown in Fig. 8.12. If the frictional traction is q per unit length, we have
F(x + δx) − F(x) + qδx = 0 (8.23)
and hence, dividing by δx and taking the limit as δx → 0,
dF
+ q = 0. (8.24)
dx
Now in the stick region, we have shown that F = 0 and hence q must also be zero.
Thus, paradoxically, there is no frictional traction in the stick region.
In a slip region, the friction force q must oppose the relative motion, so q =
− f w sgn(u̇ x ) and
dF
= f w sgn(u̇ x ). (8.25)
dx
Solving for F, we have
F = f wx sgn(u̇ x ) + C, (8.26)
where C is a constant of integration.
Fig. 8.12 Equilibrium of a

segment of the strip
δx
F(x) F(x+δ x)
q δx
154 8 Friction Laws
Fig. 8.13 Axial force F

distribution for the problem
of Fig. 8.11 fw
F0
1
stick slip x
O x=L-c x=L
For the given loading, it is reasonable to expect stick at the left end of the bar
and slip at the right end, with u̇ x > 0. It follows that the force F must vary with x
according to Fig. 8.13 in order to preserve equilibrium of elements at the ends of the
bar and at the transition between stick and slip.
In particular, the extent of the slip region is
F0
c= , (8.27)
fw
and the tensile force F(x) in the slip region is
F(x) = f w(x + c − L) L − c < x < L. (8.28)
The configuration shown, with a region of stick and a region of microslip, can only
occur as long as c < L and hence F0 < f wL. For larger values of F0 there will be
gross slip and the strip will accelerate to the right.
The displacement u x must satisfy the equation
du x σx x F
= ex x = = , (8.29)
dx E AE
and since u x = 0 in the stick zone 0 < x < L −c, continuity at the transition point
demands that
x x
1
u x (x) = ex x d x = Fd x L − c < x < L. (8.30)
L−c AE L−c
In other words, the displacement u x is proportional to the area under the F(x) curve.
Substituting for F(x) from (8.28) and performing the integration, we obtain
fw [ f w(x − L) + F0 ]2
u x (x) = [x − (L − c)]2 = L −c < x < L
2 AE 2 f w AE
(8.31)
for the case shown in Fig. 8.13. In particular, the end displacement is
F02
u x (L) ≡ u 0 = . (8.32)
2 f w AE
8.5.1 Unloading
The slip velocity can be obtained by differentiating equation (8.31) with respect to
time, giving
[ f w(x − L) + F0 ] Ḟ0
u̇ x = . (8.33)
f w AE
It follows that forward slip (u̇ x > 0) strictly only occurs as long as F0 is increasing
and all points instantaneously stick when the maximum force Fmax is reached.
If the force is then reduced, a region of reverse microslip will be developed near
the end, in which u̇ x < 0, so
dF
= −fw (8.34)
dx
from (8.25), and the tensile force F has the form shown in Fig. 8.14. Axial forces and
hence frictional tractions developed during the loading phase are ‘locked in’ except
in those regions where reverse slip has occurred. In particular, if the force is reduced
to zero, the backward slip zone will be exactly half of the original forward slip zone
and we shall leave a residual axial force field of the form shown dashed in Fig. 8.14.
At this point, the extension of the bar is again proportional to the area under the
F(x) curve and is therefore half that at Fmax , being given by
2
Fmax
u1 = . (8.35)
4 f w AE
Although the force has now been reduced to zero, the strip is left in a state of residual
stress and hence retains some memory of the loading cycle. In particular, strain
measurements in the residual state would enable us to deduce the value of Fmax
during the loading cycle. However, if unloading were continued into negative values
with F0 ≤ −Fmax , memory of the loading phase would be ‘erased’.
Fig. 8.14 Axial force

F
distribution during unloading Fmax
stick
F0
0 x
backward
slip
156 8 Friction Laws
8.5.2 Periodic Loading
Consider now the case where the force oscillates between zero and Fmax . In the steady
state, slip will occur only in the region that slipped backwards during unloading, so
there is a region that slipped only during the first half cycle and thereafter remained
stuck. This is characteristic of such problems—the residual stress developed during
the first cycle tends to reduce the extent of slip during subsequent cycles.
Both loading and unloading phases involve axial force distributions with straight
line segments of slope ± f as in Fig. 8.14 and elementary calculations show that the
displacement increases quadratically with force as in Eq. (8.32). In particular, whilst
the force is increasing from zero to Fmax in the second and subsequent cycles, the
displacement at the end is
F02
u = u1 + (8.36)
4 f w AE
and the work done by the force is

F0 =Fmax Fmax 3
F0 Fmax
W1 = F0 du = F0 d F0 = . (8.37)
F0 =0 0 2 f w AE 6 f w AE
During unloading we have

2
Fmax (Fmax − F0 )2
u= − (8.38)
2 f w AE 4 f w AE
and the work done is
F0 =0 Fmax
(Fmax − F0 ) 3
Fmax
W2 = F0 du = F0 d F0 = − . (8.39)
F0 =Fmax 0 2 f w AE 12 f w AE
Thus, only half of the work done during loading is recovered during unloading and
there is a frictional dissipation
3
Fmax
W = W1 + W2 = . (8.40)
12 f w AE
Notice also that the dissipation varies with the cube of the force amplitude. This is
a fairly common relation for a range of elastic frictional systems, including those in
two and three dimensions.
P P P P P P P
k k k k k F0
...
Fig. 8.15 Discrete model approximating the strip of Fig. 8.11
8.5.3 Discrete Model of the Strip Problem
If the elastic strip of Fig. 8.11 is modelled using the finite element method, the
resulting discrete system is mathematically equivalent to that shown in Fig. 8.15,
where a set of rigid blocks make contact with a rigid plane and are connected by
springs of stiffness k. If the entire strip is represented by N blocks separated from each
other by a distance L/N , the system behaviour will approach that of the continuous
strip if each block is pressed against the plane by a force P = wL/N and the springs
are of stiffness k = AE N /L.
Models of this kind were introduced by Iwan (1966, 1967) to approximate non-
linearities associated with material plasticity and are therefore sometimes known as
Iwan models. In particular, they have been used to explore the behaviour of bolted
joints between engineering components (Song et al. 2004, Quinn and Segalman
2005), which are believed to account for a significant proportion of the damping in
built-up structures.
One of the characteristics of Iwan models, shared by continuum models of fric-
tional microslip, is that the slip displacements at nodes (regions) that are instanta-
neously stuck retain some memory of the loading history, but in most cases, this is
restricted to a memory of the extreme values of the force.
8.5.4 The Inverse Problem
We showed in Sect. 8.5.1 that the system of Fig. 8.11 generally remains deformed
when the tangential force F0 is removed. The residual tractions must satisfy the fric-
tional condition |q(x)| ≤ f p(x), but we might reasonably ask the question ‘Can a
loading history be devised such as to develop any desired self-equilibrated distribu-
tion of tangential traction satisfying this condition?’
The answer is certainly ‘yes’ for the discrete model of Fig. 8.15, since we could
proceed as follows:-
1. If the desired residual friction forces at all the blocks are given, equilibrium
arguments can be used to determine the forces in the springs, and hence also their
extensions Δi .
2. If Δ1 > 0, we increase the force monotonically to
158 8 Friction Laws
Fig. 8.16 Residual force distribution [dashed line] fitting a desired distribution [solid line] at
discrete points
F0 = (N − 1) f P + kΔ1 .
This will cause all the blocks except the first to slip to the right, and will achieve
the required extension Δ1 .
3. We now decrease F0 monotonically until
F0 = −(N − 2) f P + kΔ2 ,
at which point all except blocks 1 and 2 are slipping to the left and the desired
extension is achieved in the second spring.
Repeating this procedure, it is clear that all the desired spring extensions can be
achieved by applying alternating tensile and compressive forces F0 of successively
decreasing amplitude, such that at each step one fewer blocks slip.6
The same technique could be applied to the continuous strip of Fig. 8.11, but
the resulting distribution of residual tensile force F(x) will consist of straight line
segments of slope ± f w, as in Fig. 8.14. The desired distribution is therefore achieved
only at a finite set of points, as illustrated in Fig. 8.16. By increasing the number of
force reversals, we can achieve a closer fit, but the residual stress curve remains non-
differentiable. A related result is that this one-dimensional system ‘remembers’ only
a denumerable set of discrete extreme values of F0 , whereas the desired function
F(x) has a non-denumerable infinity of degrees of freedom.
8.6 More Complex Friction Laws
Amontons’ law of Sect. 8.1 is clearly a considerable oversimplification of the physics

of friction, and numerous more sophisticated models have been proposed as improve-
ments. For example, the friction coefficient may exhibit dependence on sliding veloc-
ity or normal contact pressure, possibly as a result of thermal or viscous effects. Also,
Δ1 < 0, the first loading should be monotonically increasing compression to the value F0 =
6 If
−(N −1) f P +kΔ1 , after which the forces alternate in sign as before.
8.6 More Complex Friction Laws 159
Fig. 8.17 A system prone to

frictional vibrations P
k
M
c
V0
creep at areas of actual contact may lead to a more complex dependence on previous
loading or sliding history. Many of these models have been proposed as explanation
of dynamic instabilities such as ‘squeal’ in automotive brakes, or fault rupture of
tectonic plates, leading to earthquake events.
8.6.1 Instabilities During Steady Sliding
Figure 8.17 shows a dynamic system comprising a mass M, loaded by a normal force
P, and supported by a spring of stiffness k and a damper of coefficient c. The mass
rests on a rigid plane surface that translates to the right at constant speed V0 .
With an appropriate friction law, this system provides a simple explanation of
phenomena such as brake squeal, stick-slip motion of machine-tool slides, creaking
doors and the squeak of chalk on a blackboard.
8.6.2 Velocity-Dependent Friction Coefficient
Experimental measurements of friction forces under steady sliding conditions often

take the form of the Stribeck curve, shown in Fig. 8.18, where the friction coefficient
first decreases and then increases with sliding speed. The rising portion of this curve
is associated with viscous effects, particularly if the contact is lubricated, in which
case the equilibrium lubricant film thickness increases with sliding speed. At very
low speeds, the roughness of the surfaces exceeds the theoretical film thickness and
the frictional behaviour becomes dominated by solid-to-solid contact, resulting in a
higher coefficient of friction.
For small perturbations from the steady-state velocity V0 , we can approximate
this curve by the linear function
f (V ) = f (V0 ) + f (V0 )(V − V0 ), (8.41)
shown by the dashed line in Fig. 8.18.

160 8 Friction Laws
Fig. 8.18 The Stribeck f

curve: relation between
friction coefficient f and
sliding speed V . The dashed
line represents the linear f(V0)
0 V0 V
If the rightward elastic displacement of the mass in Fig. 8.17 is denoted by u(t),
the instantaneous sliding velocity is
V = V0 − u̇, (8.42)
and the equation of motion takes the form
M ü + cu̇ + ku = P f (V ) = P f (V0 ) − P f (V0 )u̇, (8.43)
or

M ü + c + P f (V0 ) u̇ + ku = P f (V0 ). (8.44)
This equation has a steady-state solution in which the mass is stationary at the point
u = P f (V0 )/k, but the transient solution also includes exponentially growing per-
turbations from the steady state if the composite damping term is negative, or
c
f (V0 ) < − . (8.45)
P
In other words, steady sliding is unstable if the negative slope of the Stribeck curve
is sufficiently steep near the steady-state velocity.
In most practical cases, the negative effective damping term is relatively small,
so that the exponential growth is oscillatory at a frequency close to the natural fre-
quency of the system. The system may then reach a non-sinusoidal limit cycle as
the amplitude of vibrations carries the instantaneous speed into regions where the
negative slope of the Stribeck curve is less steep.
Numerous other physical mechanisms have been proposed to account for fric-
tional vibrations (Ibrahim 1994), but in most cases the result is to destabilize a
lightly damped natural resonance of the system (Moirot and Nguyen 2000). One
consequence is that accurate predictions of experimentally observed frequencies do
not provide evidence in favour of any particular theory.
8.6.3 Stick-Slip Vibrations
In some cases, the oscillation may grow until the instantaneous sliding speed falls
to zero at some point in the cycle. Generally, the nonlinearity associated with the
frictional inequalities will then cause the system to achieve a limit cycle involving
periods of stick and slip (Moirot et al. 2003).
Experiments show that the force required to initiate slip is often greater than
that required to sustain motion (Rabinowicz 1995), in which case we can define a
coefficient of static friction f s , and of a coefficient of dynamic friction f d . In this
case, stick-slip vibrations can occur even if f d is independent of sliding speed.
For the system of Fig. 8.17, stick corresponds to u̇ = V0 and, once initiated, it will
persist until
f s P − cV0
cV0 + ku = f s P or u = . (8.46)
k
This equation and u̇ = V0 define the initial conditions for the subsequent slip phase
which is governed by the equation of motion
M ü + cu̇ + ku = f d P. (8.47)
The response depends qualitatively on the dimensionless parameters

ωn ( f s − f d )P c ωn k
λ= ; ζ= where ωn = (8.48)
kV0 2k M
is the undamped natural frequency of the system. Figure 8.19 shows the velocity u̇
during the slip phase for λ = 1 and two values of the damping factor ζ.
For ζ = 0.08, the resulting oscillation decays sufficiently rapidly to ensure that the
mass never again reaches the speed V0 and hence slip continues indefinitely. Thus,
stick-slip oscillations are not possible for this case.
u̇
V0
Fig. 8.19 Slip velocity following a period of stick for the system of Fig. 8.17
162 8 Friction Laws
Fig. 8.20 Stability boundary for the system of Fig. 8.17 with f s > f d . The dashed line corresponds
to the relation ζ = λ2 /4π
For a lower level of damping ζ = 0.05, the mass reaches V0 before the end of
the first cycle of oscillation and then remains stuck until the condition (8.46)2 is
once again satisfied. A steady-state stick-slip oscillation is therefore reached in the
first cycle. However, under these conditions, the equation of motion also supports a
continuous slip solution with u̇ = 0 and this state is stable, so the final state depends
on the initial conditions. Figure 8.20 defines the range of values of λ and ζ in which
stick-slip motion can occur. Notice that if the system is critically damped (ζ > 1),
the velocity u̇ decreases monotonically with time and stick-slip is impossible even
for arbitrarily large λ.
In Fig. 8.19, the initial excursion to negative values of u̇ is of the same order as
λ. Since the average value of u̇ over a steady-state cycle must be zero, this implies
that the duration of the stick phase increases with λ. In particular, when V0 is very
small, the motion will comprise widely spaced slip events of duration slightly longer
than π/ωn . This situation can arise in machine-tool slides. An even more striking
example is the ‘clanking’ of heating pipes supported in frictional clamps, where
gradual temperature changes cause the pipe to experience a set of widely separated
slip events.
8.6.4 Slip-Weakening Laws
If the static and dynamic friction coefficients differ, we must postulate a physical
change of some sort when motion commences, and it is unlikely that this can occur
instantaneously. A more likely postulate is that some finite slip displacement must
take place before the dynamic coefficient is achieved, and this suggests a law of the
form f = f (h), where h is the shift (slip displacement) defined in Eq. (7.1), and
f (0) = f s ; f (h) → f d ; h → ∞. (8.49)

We shall argue in Chap. 18, Sect. 18.6 that when two rough solids slide over each
other, the typical event comprises the interaction of asperities in the two surfaces,
and the shift h occurring during one such event will be of the order of the dimensions
of typical actual contact areas, and hence very small. Rabinowicz (1951) described
experiments in which the energy lost in friction during a frictional impact was mea-
sured and compared with independent measurements of f s and f d under steady-state
conditions. He was therefore able to estimate the length scale Δ defined such that
∞
( f s − f d )Δ = [ f (h) − f d ]dh. (8.50)
0
For the dry sliding of metals, he obtained values of a few micrometers.

Papangelo et al. (2015) have shown that in the limit where Δ is very small com-
pared with the macroscopic contact area dimensions, the transition from f s to f d is
mathematically equivalent to a shear fracture governed by a stress-intensity factor

K II = 2E ∗ ( f s − f d ) pΔ, (8.51)
where p is the local contact pressure. There is therefore a close connection between
slip-dependent friction laws and fault propagation laws based on fracture mechanics
arguments (Abercrombie and Rice 2005; Ben-Zion 2008; Svetlizky and Fineberg
2014).
8.6.5 Rate-State Laws
The slip-weakening law is one of a class of friction models known as rate-state

models, based on the premise that if the contact pressure p or the sliding velocity
V change suddenly, the friction force does not immediately reach its new steady-
state value, but approaches it through some transition process (Ruina 1983; Rice
et al. 2001). These models have been used to eliminate unphysical predictions of
elastodynamic instability of arbitrarily small-wavelength perturbations predicted by
Amontons’ law [see Sect. 19.5], but they also provide a convenient way of quantifying
experimental results that are not well characterized by the simpler law.
We express the magnitude of the frictional traction q as a function
|q| = g1 (S, p, V ), (8.52)
where S is a state variable introduced to describe history-dependent changes in the

local contact condition. This relation is then supplemented by an evolution law for
S, typically of the form
Ṡ = g2 (S, p, V ). (8.53)
164 8 Friction Laws
The state variable S can be identified with a physical property of the system, or alter-
natively, Eqs. (8.52, 8.53) can be regarded as defining a ‘black box’ characterization
of the system, whose parameters are determined by appropriate experiments.
As an example of the former approach, we can recreate the slip-weakening law
of Sect. 8.6.4 by treating the slip distance h as the state variable during slip periods,
so that
g1 (S, p, V ) = f (S) p; g2 (S, p, V ) = |V | V = 0, (8.54)
since in this case Ṡ = ḣ = |V |.

For completeness, we require a further equation defining the rate at which the
static friction coefficient is recovered during periods of stick. For example, if the
static coefficient is approached exponentially on a time scale t0 , we might define
S
g1 (S, p, V ) = f (S) p; g2 (S, p, V ) = − V = 0. (8.55)
t0
Alternatively, we could take the proportion of the nominal contact area in actual
contact as the state variable. During static contact, creep allows this to increase
asymptotically to a limiting value, but during sliding, contacts are continually being
established and broken, so the growth of individual contacts is limited by the average
dwell time (Bar Sinai et al. 2012).
Using the black-box approach, some information about the functions g1 , g2 can be
obtained by (i) conducting steady-state sliding experiments with constant p, V , and
then (ii) measuring the transient that is obtained when p or V is suddenly changed to
a new constant value. Equation (8.52) suggests that there will then be an immediate
discontinuity in |q|, followed by an asymptotic approach to the new steady state
governed by the evolution of the state variable. Experimental results exhibiting this
behaviour were discussed by Ruina (1983). Other experiments suitable for system
characterization include steady sinusoidal variation of p or V about some mean value
(Cabboi et al. 2016), and static friction tests where the traction needed to initiate
motion is measured as a function of the duration of static contact under constant p.
Problems
1. The system of Fig. 8.2a is loaded by the periodic force
F1 = F cos(ωt); F2 = F sin(ωt),
where the frequency ω is sufficiently slow for quasi-static conditions to apply.

Identify the times during each steady-state cycle that transitions occur between
the various possible states, if the coefficient of friction f = 0.5 and
Problems 165

10 4
K= .
4 10
Would the behaviour be qualitatively different if the coefficient of friction were

lower?
2. The system of Fig. 8.2a is subjected to a force F(t) = {C1 t, C2 t} starting at time
t = 0, where f > k11 /k21 > 0 and the constants C1 , C2 are defined such that F moves
through the multiple solution segment in Fig. 8.5.
(i) Determine the quasi-static displacements v(t), w(t) under the assumption of
backward slip.
(ii) Now set up the dynamic equations of motion for backward slip, including the
inertia term −M ü.
(iii) Show that the time-varying displacements from (i) satisfy the equations of
motion, provided the mass has an appropriate initial velocity at time t = 0, and
find this velocity.
(iv) Show that if the initial velocity differs very slightly from the quasi-static value,
this perturbation will then grow without limit. In other words, that the backward
slip state is unstable.
3. The contact stiffness matrix (8.13) for a two-node frictional system is defined such
that
10 3 3 10 20 10
A= ; B= ; C= .
3 20 10 3 10 20
Use Klarbring’s criterion [Sect. 8.3.4] to find the critical coefficient of friction above
which the rate problem is ill-posed.
4. A two-node frictional system is defined by the contact stiffness matrix of Problem

8.3. Find the minimum coefficient for which the system is capable of being wedged
in a state with no external forces [ piw = qiw = 0] but non-zero nodal displacements.
Notice that the critical state may involve both nodes in contact [w1 = w2 = 0], or
only one node in contact.
5. Construct a diagram analogous to Fig. 8.9 for the three-dimensional case where
node i is slipping in both solutions 1 and 2, but not in the same direction. Show that
the contribution of this node to Ė is strictly negative under these conditions. Hence
show that in the steady state, the direction of slip at slipping nodes is independent of
initial conditions.
6. Figure 8.21 shows the extreme positions of the frictional constraints for a two-node
system, so that the central white quadrilateral represents the safe shakedown space
S. The periodic loading scenario causes the constraints to advance to and then retreat
from these positions in the sequence I, III, II, IV, I, III …. Track the motion of the
166 8 Friction Laws
Fig. 8.21 The safe

shakedown space
v2 B
III
A safe I
shakedown
II space
D C
IV
v1
point P from an arbitrary starting point outside S and hence partition this space into
regions corresponding to the final position of P lying on the lines AB, BC, C D and
D A, or in one of the corners A, B, C and D. Also, explain which starting positions
will involve the shakedown state being reached in one or two cycles, and which will
involve an asymptotic approach to the steady state.
7. A cylindrical bar of diameter D and length L ( D) just fits inside a long cylindri-
cal hole in a rigid body. The bar is now heated to a temperature T . If the coefficient
of friction between the bar and the hole is f , find the distribution of contact pressure
p and the mean axial stress σzz as functions of distance z along the bar. The mate-
rial of the bar has Young’s modulus E, Poisson’s ratio ν and coefficient of thermal
expansion α. Assume plane cross sections remain plane.
8. Figure 8.22 shows the geometry of a standard ‘pull-out’ test for fibrous composite
materials, in which a tensile force F0 is applied to the end of an exposed fibre and
increased until failure occurs of the bond between the fibre and the matrix.
An approximate model of the process is shown in Fig. 8.23, where the fibre is
represented by a rod (a) and the matrix by a pair of rods (b). The stiffnesses of the
rods are denoted by ka , kb respectively, defined through the relations
Fig. 8.22 The ‘pull-out’ test

F0
fibre
fiber
matrix
Problems 167
Fig. 8.23 Rod model of w per unit length

fibre pull-out
F0 / 2 (b)
F0
(a)
F0 / 2 (b)
x
L
du x F(x)
ex x = = ,
dx k
where u x is the local displacement of the rod in the x-direction and F(x) is the tensile
force in the rod at the point (x).
We make the further assumption that the transfer of force between the rods is
achieved solely by friction, through a coefficient of friction, f , and that the rods are
pressed together by a normal force w per unit length.
The force F0 is gradually increased from zero. Describe what happens and deter-
mine the location and magnitude of any regions of slip as functions of F0 . Also, give
expressions for the axial forces in the matrix and the fibre as functions of x.
9. Suppose that each block in the discrete model of Fig. 8.15 has mass M. At time
t = 0, all the springs are unstretched, and a force F0 > f P is suddenly applied to
the last block as shown in the figure. Find the displacement u of this block assuming
that all the remaining blocks remain stationary. Hence determine the time t = t1 at
which the second block will just start to move.
10. Suppose that the system of Fig. 8.17 can support a steady-state vibration such
that
u(t) = u 0 + A cos(ωt).
Calculate (i) the energy absorbed in the damper and (ii) the work done by the friction
force on the mass during one complete cycle 2π/ω, assuming the friction coefficient
is given by equation
f = f 0 − f 1 (V − V0 ),
where f 0 , f 1 are positive constants. Hence show that such a state is possible if and
only if f 1 = c/P and that in this case, the amplitude A can take any value.
Repeat the calculation for the case where the friction coefficient is defined by the
cubic relation
f (V ) = f (V0 ) − f 1 (V − V0 ) + f 3 (V − V0 )3 ,
168 8 Friction Laws
where f 3 > 0. Show that the amplitude in the steady state is now determinate and
find its value.7
11. A state-variable friction law for unidirectional sliding at speed V is defined by

the relations

S S
Ṡ = |V | − ; Q = f (S)P; f (S) = f 0 + f 1 exp − ,
t0 L0
where P, Q are the normal and frictional force respectively, S is the state variable
which has dimensions of length, and t0 , L 0 are constants with dimensions of time
and length respectively.
(i) If the bodies slide at constant speed V , what will be the relation between the
coefficient of friction Q/P and V .
(ii) If the sliding speed is held constant at V0 for some time, but then changed
suddenly to V1 , how will the friction force vary during the ensuing transient.
7 Note that this is an approximate solution, since the equation of motion will not generally be satisfied
by the sinusoidal function for all t.

Chapter 9
Frictional Problems Involving Half-Spaces
It should be clear from Chaps. 7 and 8 that problems involving tangential loading
are considerably simplified when they are uncoupled. If the contacting bodies can
be approximated by half-spaces, this condition is satisfied if and only if Dundurs’
constant β = 0, as defined in Eq. (7.8), the most common case being that in which the
two bodies are of similar materials. We shall assume that β = 0 through Sects. 9.1–9.4
of this chapter.
An immediate consequence of the uncoupled assumption is that the normal contact
problem depends only on the surface profile and the instantaneous normal force P.
In other words, there is no history-dependence in the normal problem, and indeed the
resulting contact area and normal tractions will be the same as they would have been
if the contact had been frictionless, as in Chaps. 2–6. It is therefore often helpful to
solve the normal problem first, because the normal tractions p(x, y, t) then define
the conditions for limiting friction in the tangential problem through Eqs. (8.4, 8.5).
9.1 Cattaneo’s Problem
Consider the case, where the two bodies have quadratic surfaces and are loaded by a
purely normal force P0 . Since β = 0, no tangential tractions are developed during this
process, and the normal tractions and the semi-axes a0 , b0 of the elliptical contact area
A0 are defined by the Hertzian equations of Chap. 3. Suppose, we then maintain P0 at
a constant value whilst applying a tangential force Q x that increases monotonically
from zero to a value Q 0 < f P0 — i.e. which is insufficient to cause gross slip. This
loading scenario is illustrated in Fig. 9.1.
This problem was first solved by Cattaneo (1938) and later by Mindlin (1949)
who was most likely unaware of Cattaneo’s earlier publication. The contact area is
determined by the normal force and hence remains unchanged during the tangential
loading phase. However, if we assumed that ‘stick’ conditions applied throughout the
https://doi.org/10.1007/978-3-319-70939-0_9
170 9 Frictional Problems Involving Half-Spaces
Fig. 9.1 Loading scenario

for Cattaneo and Mindlin’s Qx Qx = - f P
problem
Q0
P
O P0
Qx = f P
ellipse, the tangential tractions would be given by Eq. (7.62) and in particular would
be singular at the edge of the contact area, where the normal tractions go to zero.
Thus, no finite coefficient of friction is sufficient to ensure full stick, and we must
anticipate the development of a region of microslip around the edge of the contact
area.
Cattaneo and Mindlin solved the tangential problem by first finding the tangential
displacements due to the tangential traction distribution

3 f P0 x2 y2 x2 y2
qx (x, y) = 1− 2
− 2 + < 1, (9.1)
2πa0 b0 a0 b0 a02 b02
which corresponds to the case of slip in the x-direction over the entire elliptical
contact area A0 , with semi-axes a0 , b0 . Galin’s theorem and symmetry considerations
show that the displacements must take the quadratic form
u x (x, y) = C0 + C1 x 2 + C2 y 2 ; u y (x, y) = C3 x y, (9.2)
where C0 , C1 , C2 , C3 are constants which can be determined by substituting (9.1)

into (7.59, 7.60), and using the results from Appendix B to evaluate the resulting
integrals. We obtain
3(1 + ν) f P0
C0 = [(2 − ν)I0 (0, e) + ν I0 (1, e)]
8π Ea0
3(1 + ν) f P0
C1 = − [(2 − ν)I1 (0, e) + ν I1 (1, e)]
8π Ea03
3(1 + ν) f P0
C2 = − [(2 − ν)I2 (0, e) + ν I2 (1, e)] (9.3)
8π Ea03
3ν(1 + ν) f P0
C3 = I3 (1, e),
8π Ea03
9.1 Cattaneo’s Problem 171
Fig. 9.2 Tangential traction fp

distribution along the major
axis for Cattaneo’s problem
qx
slip stick slip x

-a 0 -a1 a1 a0
where the functions I0 , I1 , I2 , I3 are defined and expressed in terms of complete

elliptic integrals in Appendix B,1 Eq. (B.33).
In particular, notice that the quadratic coefficients C1 , C2 , C3 are independent of
P0 , since P0 ∼ a03 from Eq. (3.32), and the eccentricity e is independent of P0 , since
it depends only on the constants A, B defining the initial gap function (3.2).
We next construct the traction distribution

3 f P0 x2 y2 3 f P1 x2 y2
qx (x, y) = 1− 2 − 2 − 1− 2 − 2, (9.4)
2πa0 b0 a0 b0 2πa1 b1 a1 b1
where a1 , b1 are the semi-axes of the contact ellipse A1 corresponding to a different

normal force P1 < P0 , and the square roots are to be interpreted as zero in any region
in which their respective arguments are negative.
The two superposed distributions share the same values of the coefficients
C1 , C2 , C3 in (9.2), and hence, the combined distribution produces tangential dis-
placements with no dependence on x, y inside A1 . There will remain a non-zero
constant in u x , corresponding to the [different] values of C0 in the superposed solu-
tions, but this is consistent with stick inside A1 and defines a relative rigid-body
translation of the contacting bodies. Also, the tractions in the surrounding elliptical
annulus (A0 −A1 ) are given by (9.1) and hence satisfy the conditions for microslip
in this region. The variation of tangential traction qx along the major axis y = 0 is
illustrated in Fig. 9.2.
The traction distribution (9.4) corresponds to a tangential force Q 0 = f P0 − f P1 ,
and hence, if Q 0 is prescribed, we must choose
Q0
P1 = P0 − . (9.5)
f
Once P1 is determined, the Hertz problem defines the dimensions of A1 , which

constitutes the stick region in the tangential loading problem. Notice that as Q 0
increases, P1 decreases, so the stick region shrinks as we should expect, reaching
zero when Q 0 = f P0 , after which gross slip would occur.
Cattaneo’s solution satisfies the friction conditions (8.4)1 and (8.5) exactly, but
the slip displacements in (A0 −A1 ) include a component in the y-direction, which
violates (8.4)2 , except in the special case where ν = 0. In other words, the assumed
frictional tractions do not exactly oppose the relative motion. This error has been
examined in detail by Munisamy et al. (1994) who compared the predictions with
a numerical solution of the problem for the case where the contacting bodies are
axisymmetric and hence A0 , A1 are circular. The maximum error occurs adjacent to
the stick boundary, where slip displacements are very small. In general, the error in
the Cattaneo tractions was found to be extremely small.
Mindlin and Deresiewicz (1953) extended Cattaneo’s solution to consider cases
where both normal and tangential forces change with time t, not necessarily monoton-
ically. In particular, they showed that there would be no slip during any period when
dP d| Q|
>0 and < f. (9.6)
dt dP
Under these conditions, the contact area increases and the new contact is laid down
in a state of stick.
9.2 The Ciavarella–Jäger Theorem
The Cattaneo and Mindlin superposition can clearly be applied to the two-dimensional
Hertzian problem of the contact of two cylinders loaded by a normal force P0 and
a tangential force Q x = Q 0 or Q y = Q 0 , since this is essentially the limiting case,
where the eccentricity of the ellipse e → 1. However, Ciavarella (1998a) and Jäger
(1998) have shown that the same superposition defines the exact solution for any
uncoupled two-dimensional problem that can be represented as the contact of two
half-planes, with the loading defined in Fig. 9.1. Here, we shall develop the argu-
ment for the case of in-plane loading Q x , but an exactly parallel derivation can be
performed for antiplane loading Q y .
Suppose that the contact area due to a normal force P is denoted by A(P) and
the corresponding contact pressure distribution by p(x, P). Notice that we define
p(x, P) over the entire surface, so that p(x, P) = 0, x ∈ / A(P). We shall show that
the tangential contact traction due to the tangential force Q x = Q 0 > 0 is then given
by the Cattaneo distribution
qx (x) = f p(x, P0 ) − f p(x, P1 ), (9.7)
where the fictitious normal force P1 is given by Eq. (9.5). Also, A(P1 ) defines the
stick region and (A(P0 )−A(P1 )) the slip region. Notice that the case where Q 0 < 0
can be accommodated by replacing f by − f in Eqs. (9.5, 9.7).
9.2 The Ciavarella–Jäger Theorem 173
Proof
Consider, the contact of two elastic half-planes defined by the initial gap function
g0 (x). Since the problem is uncoupled, Dundurs’ constant β = 0, and Eqs. (7.16,
7.17) reduce to

du z 2 p(ξ)dξ du x 2 qx (ξ)dξ
=− ∗ ; =− ∗ . (9.8)
dx πE (x − ξ) dx πE (x − ξ)
Also, if we choose the origin to correspond with the point of the first contact and
assume that this point never slips, the gap g(x) and the slip displacement h x (x) are
given by
x x
du z du x
g(x) = g0 (x) + d x; h x (x) = − dx (9.9)
0 dx 0 dx
respectively, where h x is defined in Eq. (7.1). It follows from (9.8) that if qx (x) =
f p(x) for all x (including places where both are zero), then
du x du z
= f and hence h(x) = f [g0 (x) − g(x)] (9.10)
dx dz
for all x.
These conditions apply to each of the terms in the traction distribution (9.7)
when considered separately, so by superposition, we conclude that the resulting slip
displacement is
h x (x) = − f g(x, P0 ) + f g(x, P1 ), (9.11)
where g(x, P) is the gap g(x) when the normal force is P. Notice that the term
g0 (x) cancels in this expression, since the traction distributions p(x, P0 ), p(x, P1 )
correspond to the same initial gap function.
The Stick Region x ∈ A( P1 )
Points in this region are in contact at both P = P1 and P = P0 , so both terms in
Eq. (9.11) are zero, confirming that there is no slip. Also, since P1 < P0 , Theorem 4
of Sect. 4.2 shows that 0 < p(x, P1 ) < p(x, P0 ) and hence the traction distribution
(9.7) satisfies the frictional inequality |qx (x)| < f p(x).
The Slip Region x ∈ [A( P0 )−A( P1 )]
In this region, p(x, P1 ) = 0 so Eq. (9.7) reduces to qx (x) = f p(x, P0 ), which satisfies
the conditions for slip with qx > 0. We also need to verify that the direction of slip is
consistent with this direction of frictional tractions.
We first note that in [A(P0 )−A(P1 )], g(x, P0 ) = 0, but g(x, P1 ) > 0, so
h x (x) = f g(x, P1 ) > 0, (9.12)

from (9.11). Furthermore, we proved in Theorem 4 of Sect. 4.2 that the gap at any
given point is a monotonically non-increasing function of the normal force. Thus,
∂g(x, P1 ) ∂h(x)
≤0 and hence ≥ 0, (9.13)
∂ P1 ∂ Q0
since Q 0 = f P0 − f P1 , from (9.5). The frictional tractions therefore oppose the

incremental slip displacement, as required by the Coulomb friction law.
Multiply-Connected Contact Areas
The preceding arguments remain valid if one or both of the contact areas A(P0 ),
A(P1 ) comprise two or more segments of the line z = 0, with intervening separation
regions. In particular, Eq. (9.11) shows that the slip displacement will be zero (indi-
cating stick) in any region that would be in contact at a normal force P1 . This result
is of interest in connection with the tangential loading of bodies with rough surfaces,
where contact is generally restricted to locations near the peaks of asperities [see
Chap. 16]. It implies that when a tangential force is applied, the first asperities to slip
are those which were last to come into contact during the preceding normal loading
phase.
9.2.1 Three-Dimensional Problems
The original Cattaneo solution of Sect. 9.1 was developed for the three-dimensional
Hertz problem, for which it is strictly exact only in the special case ν = 0, though
the errors for other values of Poisson’s ratio are relatively small. In this limit (similar
materials and ν = 0), the three-dimensional Green’s functions (7.7) reduce to
Qx Qy P
ux = ∗ ; uy = ∗ ; uz = , (9.14)
πE r πE r π E ∗r
and it is clear that the resulting mathematical similarity between the convolution
integrals for normal and tangential loading permits a similar superposition to that
used in the two-dimensional case (Ciavarella 1998b).
We define the normal contact problem through the functions p(x, y, P), A(P),
such that

1 p(ξ, η, P)dξdη
∗ = Δ(P) − g0 (x, y) (x, y) ∈ A(P),
πE A(P) (x − ξ)2 + (y − η)2
(9.15)
as in Eq. (2.9), where Δ(P) is the rigid-body indentation due to the force P. A similar
convolution on the tangential traction qx (x, y), using (9.14) then shows that

1 qx (ξ, η, P)dξdη
u x (x, y) = , (9.16)
πE∗ A(P) (x − ξ)2 + (y − η)2
9.2 The Ciavarella–Jäger Theorem 175
and hence, if we choose q y (x, y) = 0 and
Q0
qx (x, y) = f p(x, y, P0 ) − f p(x, y, P1 ) where P1 = P0 − , (9.17)
f
the corresponding tangential displacements will be
u x (x, y) = f [Δ(P0 ) − g0 (x, y)] − f [Δ(P1 ) − g0 (x, y)]

= f [Δ(P0 ) − Δ(P1 )] (x, y) ∈ A(P1 ), (9.18)
and this is independent of x, y as required in the stick region. We can always choose
the coordinate system such that the applied tangential force Q 0 acts in the x-direction,
so this provides a general solution to the problem.
9.3 More General Loading Scenarios
The Ciavarella–Jäger applies strictly only to the loading scenario shown in Fig. 9.1,
but with some modifications it can be extended to more general cases.
9.3.1 Constant Normal Force
Suppose first that the normal force P remains constant, but the tangential force Q x
is non-monotonic in time. This case was considered by Mindlin and Deresiewicz
(1953), including the important case where Q x is periodic in time. Suppose the
tangential force Q increases monotonically to a maximum value Q 0 and then reduces
monotonically to Q 1 , where Q 1 > −Q 0 . During the loading phase, the tractions
follow the Cattaneo–Mindlin superposition (9.7), but at Q 0 , the entire contact area
A(P0 ) will stick and then the region of a reverse slip will grow inwards from the
edges.
For this phase, we define a second fictitious normal force P2 with associated
contact area A(P2 ) and contact pressure distribution p(x, P2 ). The tangential traction
at this stage will be given by
qx (x) = [ f p( x, P0 ) − f p(x, P1 )] − 2[ f p(x, P0 ) − f p(x, P2 )]

= − f [ p(x, P0 ) + p(x, P1 ) − 2 p(x, P2 )]. (9.19)
This is the sum of two Cattaneo distributions and hence the region A(P2 ) will remain
stuck during the unloading phase. Notice that the factor of two in the second term
is needed in order to ensure that regions that were slipping forwards [qx = f p(x)]
during the loading phase revert to slipping backwards [qx = − f p(x)] during the
unloading phase.
The total tangential force associated with (9.19) is
Q 1 = f (P0 − P1 ) − 2 f (P0 − P2 ) = Q 0 − 2 f (P0 − P2 ) (9.20)
and hence P2 must be chosen such that
(Q 0 − Q 1 ) (Q 0 + Q 1 )
P2 = P0 − = P1 + , (9.21)
2f 2f
using Eq. (9.5) to eliminate P0 . Since Q 1 > −Q 0 , this implies that P2 > P1 and hence
A(P1 ) ∈ A(P2 ). The status of the various regions is that points in A(P1 ) have never
slipped, points in (A(P2 )−A(P1 )) slipped when Q x was increasing, but not when it
was decreasing, and points in (A(P0 )−A(P2 )) are experiencing backward slip.
If Q x is decreased as far as −Q 0 [completely reversed tangential loading], the
region (A(P2 )−A(P1 )) will shrink to zero and the distribution will simplify to
qx (x) = − f [ p(x, P0 ) − p(x, P1 )]. (9.22)
The memory of the loading phase is then essentially erased.

The solution strategy can be summarized in the following rules:-
1. Adding a Cattaneo distribution ± f [ p(x, P0 ) − p(x, Pi )] will leave the region
A(Pi ) in a state of stick.
2. If Q̇ x changes sign, the added distribution must be doubled in order to change
the direction of slip in (A(P0 )−A(Pi )).
3. As long as the magnitude of the tangential force |Q x | does not reach a previous
maximum or minimum value Q i , then stick zones and traction distributions retain
some memory of these extrema. This characteristic is shared by the Iwan models
discussed in Sect. 8.5.3.
4. If a previous extremum |Q i | is exceeded, the traction distribution can be deter-
mined by defining a Cattaneo distribution to proceed directly from the previous
(unexceeded) extremum to the present state.
9.3.2 Variable Normal Force
Similar techniques can be used to determine the quasi-static response to fairly general
variations in both normal and tangential forces. We shall discuss here the case where
the steady-state loading is of the sinusoidal form
P = P0 + P1 cos(ωt); Q x = Q 0 + Q 1 cos(ωt + φ), (9.23)

9.3 More General Loading Scenarios 177
Fig. 9.3 Steady-state

sinusoidal loading. The Qx
dashed lines are all of slope
±f Qx = f P X
E
T
O C P
B A
Y
Qx = - f P
where P0 , P1 , Q 0 , Q 1 are constants and φ is a constant phase lag.1 We assume that

the tangential force is never sufficient to cause gross slip, so the resulting load path
has the form of an ellipse contained between the limiting lines Q x = ± f P, as shown
in Fig. 9.3.
Notice that since the frictional problem is history-dependent, we must also define
some initial load path O A from zero before the steady state is reached. A question
of some interest is the extent to which this initial loading influences the steady-state
behaviour.
We first note that during any period when the criterion (9.6) is satisfied, the
contact area increases with time and the new contact is established in a state of stick.
In this case, the change in the tangential traction distribution can be determined using
arguments similar to those in Sect. 7.7.1 as

∂
Δqx (x) = p(x, P) ΔQ x . (9.24)
∂P
In Fig. 9.3, for simplicity we assume that this condition is satisfied throughout the
initial loading period, so full stick will persist to the tangent point C, at which

PC
∂ d Qx
qxC (x) = p(x, P) d P. (9.25)
0 ∂P dP
Once we pass C, a region of forward slip must develop at the edge of the contact
area and arguments similar to those used above show that the traction distribution
corresponding to the point X on the loading scenario of Fig. 9.3 is defined by

PY
∂ d Qx
qxX (x) = p(x, P) d P + f p(x, PX ) − f p(x, PY ), (9.26)
0 ∂P dP
1 For more details of these arguments, see Barber et al. (2011).

where the point Y is determined by the construction in Fig. 9.3. The stick zone A(PY )
shrinks monotonically as X passes around the ellipse, reaching a minimum at the
point E. However, calculations of the incremental slip displacements show that the
entire instantaneous contact area comes to rest at E — i.e. we have instantaneous total
stick, and a reversed slip zone then starts to develop from the outside of the contact
area. Once again, the traction distribution can be written down by superposing a
Cattaneo distribution of opposite sign with a stick zone that continues to shrink until
we reach the second tangent point B, where full stick is re-established. After the
first loading cycle, points x ∈ A(PT ) experience no further slip, where the point T is
defined by the intersection of the two tangents BT and E T . This therefore defines
the permanent stick zone T of Sect. 8.4.
It should be emphasized that this procedure provides a solution for any uncoupled
two-dimensional problem with Coulomb friction boundary conditions for which the
normal contact problem can be solved and for which the contacting bodies can
reasonably be approximated by half-spaces. In view of Chap. 6, this defines a very
broad class of systems.
9.3.3 Memory and ‘Advancing Stick’
We recall from Sect. 8.3.3 that the memory of a frictional system resides in the slip
displacements at points in the stick zone, and that this memory is established only
at the time when the point in question makes a transition from separation or slip to
stick. With the loading scenario of Fig. 9.3, this occurs in two qualitatively different
ways:-
1. During periods where the loading history satisfies the conditions (9.6), newly
established contact regions pass directly from separation to stick with no inter-
vening slip phase. Comninou and Dundurs (1982) rather misleadingly refer to
this as ‘the slip zone reced[ing] from a stick zone’, but there is in fact no slip.
It is more accurately described as separation receding from a stick zone. Notice
that when this occurs, the problem can be solved only by using an incremental
formulation, leading to a time or load path integral, as in Eq. (9.25).
2. At the points E and B in Fig. 9.3, the entire slip zone sticks instantaneously, so
the corresponding slip displacements at those instants are captured in memory.
This is analogous with the behaviour of the one-dimensional model of Sect. 8.5.1
at the instant of force reversal, and is characteristic of Iwan-type discrete models.
Aleshin et al. (2015) describe a graphical construction that captures the evolution of
the contact conditions in problems satisfying the conditions of the Ciavarella–Jäger
theorem, with particular reference to memory effects.
9.4 The Effect of Bulk Stress 179
Fig. 9.4 Influence of bulk

P
stresses σ1 , σ2 on the contact
problem
σ2 Qx σ2
x
σ1 Qx σ1
9.4 The Effect of Bulk Stress
In problems that can be approximated by half-spaces, the contacting bodies might

also be loaded by normal stresses σ1 , σ2 ‘at infinity’ as shown in Fig. 9.4. These
bulk stresses are important in fretting fatigue situations, since they contribute to the
initiation and propagation of cracks perpendicular to the interface.
They also cause tensile strains ex x = ∂u x /∂x in the contact region which therefore
influence the tangential contact problem through Eq. (7.3). In particular, Eq. (7.16)
is modified to
a
du x 2β p(x) 2 qx (ξ)dξ σ1 (1 − ν12 ) σ2 (1 − ν22 )
=− − + − , (9.27)
dx E∗ π E ∗ b (x − ξ) E1 E2
and in the special case of similar materials

du x 2 a
qx (ξ)dξ (σ1 − σ2 )
=− ∗ + . (9.28)
dx πE b (x − ξ) 2E ∗
9.4.1 Hertz Problem with Superposed Bulk Stress
To introduce the topic, consider the case of a plane Hertzian contact between similar
materials in which the normal force P0 is applied first and then held constant, after
which a bulk stresses σ is applied to body 1 only (Nowell and Hills 1987). No slip
will occur during the application of the normal force, and the contact pressure will
be
x2 2P0 4P0 R
p(x) = p0 1 − 2 where p0 = ; a= , (9.29)
a πa πE∗
from Eqs. (6.24, 6.25).

Fully Stuck Solution
If we assume provisionally that the bodies remain stuck during the subsequent appli-
cation of the bulk stress, the no slip condition ḣ x = 0 can be integrated in time to give
h x = 0. We then have du x /d x = 0 and hence

a
qx (ξ)dξ πσ
= − a < x < a, (9.30)
−a (x − ξ) 4
from (9.28). The right-hand side is independent of x and hence we can use Eq. (6.53)
with n = 1, obtaining
σa
x σx
qx (x) = − √ T1 =− √ . (9.31)
4 a −x
2 2 a 4 a2 − x 2
Partial Slip Solution

Equation (9.31) defines tractions that are singular at x = ±a, so we must anticipate
symmetric regions of slip b < |x| < a in opposite directions at the two edges. The
problem is then defined by the equations

x2
qx (x) = − f p0 1 − 2 sgn(x) b < |x| < a (9.32)
a a
qx (ξ)dξ πσ
= − b < x < b. (9.33)
−a (x − ξ) 4
Following Nowell and Hills (1987), we separate the integral in (9.33) into three
separate ranges, and use (9.32) to substitute for qx (ξ) in the slip regions, giving
b
qx (ξ)dξ
= F(x) − b < x < b, (9.34)
−b (x − ξ)
where
−b 2
πσ f p0 a
a 2 − ξ 2 dξ a − ξ 2 dξ
F(x) ≡ + − . (9.35)
4 a b (x − ξ) −a (x − ξ)
Equation (9.34) is a Cauchy singular integral equation for the unknown tangential
traction in the stick zone. This traction must be bounded at x = ±b, so the unknown
stick-slip boundaries ±b can be found from the consistency condition (6.19), which
here takes the form b
F(x)d x
√ = 0. (9.36)
−b b2 − x 2
Substituting for F(x) and evaluating the integrals, we obtain

πσ b2
K (k) − E (k) = with k= 1− (9.37)
8 f p0 a2
Fig. 9.5 Slip and stick

regions for Hertzian contact
loaded by a bulk stress σ
(Ciavarella and Macina 2003), where K (k), E(k) are complete elliptic integrals
defined in Eqs. (2.32, 3.24) respectively. The extent of the slip zone therefore depends
only on the ratio σ/ f p0 , as shown in Fig. 9.5.
The bounded–bounded solution of (9.34) can then be obtained as
√
b2 − x 2 b
F(ξ)dξ
qx (x) = − − b < x < b. (9.38)
π2 −b (x − ξ) b2 − ξ 2
9.4.2 Combined Bulk Stress and Tangential Force
When normal loading is followed by the application of bulk stresses σ1 , σ2 only,

slip occurs in opposite directions in the two edge slip zones, but we found earlier
in Sect. 9.1 that the application of a tangential Q x alone leads to edge slip zones in
which the slip is in the same direction. For the more general case, where Q x and
σ1 , σ2 are applied simultaneously and in proportion, the direction of slip depends on
the relative magnitude of Q x and σ1 −σ2 . This problem was investigated by Nowell
and Hills (1987) for the Hertzian case, and more general geometries were discussed
by Ciavarella and Demelio (2001) and Ciavarella and Macina (2003).
Moderate Bulk Stress
Ciavarella and Demelio coined the term ‘moderate bulk stress’ to refer to cases where
the bulk stress difference σ1 −σ2 is sufficiently small relative to the tangential force
Q x for the direction of slip in any slip zone to be determined by the direction of Q x .
In this case, the resulting tangential contact problem can be solved for the general
uncoupled case by an extension of the Ciavarella–Jäger superposition of Sect. 9.2.
We first note that if qx (x) = f p(x) for all x, Eqs. (9.8, 9.28) imply that
du x du z (σ1 − σ2 )
= f + . (9.39)
dx dx 2E ∗
This is a generalization of Eq. (9.10)1 to include the strains due to the bulk stresses.
We also extend the notation of Sect. 9.2 to allow the possibility that the upper
body 2 should have a small anticlockwise rigid-body rotation α in addition to the
rigid-body translation Δ, so that
du z
g(x) = g0 (x) − Δ + αx + u z (x) and g
(x) = g0
(x) + α + . (9.40)
dx
We suppose that the normal contact problem can be solved for all values of α and
the normal force P, and that the resulting contact area and contact pressure are then
defined by known functions A(P, α) and p(x, P, α) respectively. In the contact area
A(P, α), we have g
(x) = 0, and hence
du z
= −g0
(x) − α. (9.41)
dx
Now consider the tangential traction distribution
qx (x) = f p(x, P0 , 0) − f p(x, P1 , α), (9.42)
where the fictitious normal force P1 is defined as
Q0
P1 = P0 − . (9.43)
f
In the region x ∈ A(P0 , 0)∩A(P1 , α), condition (9.41) applies separately for each
term in (9.42) and hence, using (9.39),
du x (σ1 − σ2 ) (σ1 − σ2 )
= − f g0
(x) + f g0
(x) + f α + = fα+ . (9.44)
dx 2E ∗ 2E ∗
It follows that Eq. (9.28) will be satisfied in the stick region A(P1 , α) if we choose
(σ1 − σ2 )
α=− . (9.45)
2 f E∗
Also, the slip condition qx (x) = f p(x) is satisfied in the region A(P0 , 0)\A(P1 , α).
This, therefore defines the complete solution to the problem provided that
A(P1 , α) \ A(P0 , 0) = ∅. (9.46)
In other words, all the points in the assumed stick region A(P1 , α) are also in the con-
tact region2 A(P0 , 0). This condition determines the range of bulk stress difference
σ1−σ2 for which the moderate bulk stress assumption is justified. If it is not satisfied,
2 Inthe absence of bulk stress, this condition is satisfied identically by virtue of Theorem 4 of
Sect. 4.2, and we recover the Ciavarella–Jäger theorem of Sect. 9.2.
slip in the various slip regions will not be all in the same direction, and the super-
position leading to the solution (9.42) is not valid. No general method of solution
is available for this case, but particular problems can be treated as in Sect. 9.4.1. In
other words, we assume the existence of slip regions of unknown extent and opposite
directions of slip, define the corresponding slip tractions in these regions, and then
use them to set up a Cauchy singular integral equation for the tangential tractions in
the remaining stick zone. The consistency condition, and the equilibrium condition

qx (x)d x = Q 0 (9.47)
A(P0 )
then provide two equations for the unknown boundaries of the stick region.
Hertzian Geometry
To illustrate the criterion (9.46), we consider the Hertzian problem where,
∗
E 2 4P R
p(x, P, α) = a − (x + c)2 with a= ; c = Rα, (9.48)
2R πE∗
and
A(P, α) = {−a − c < x < a − c}. (9.49)
The tangential traction distribution is obtained by substituting (9.48) into (9.42) using
(9.45) for α and (9.43) for P1 . We obtain
f E

∗
qx (x) = a0 − x 2 − a1 − (x + c)2 , (9.50)
2R
where

4P0 R 4P1 R Q0 R(σ1 − σ2 )
a0 = ; a1 = = a0 1 − ; c=− . (9.51)
πE∗ πE∗ f P0 2 f E∗
The condition (9.46) is satisfied if and only if a1 +|c| ≤ a0 , from which

|σ1 − σ2 | πR Q0
≤1− 1− . (9.52)
4f P0 E ∗ f P0
This expression is based on the assumption that Q 0 > 0 but it is easily shown that
the more general case leads to the same result with |Q 0 | replacing Q 0 .
9.5 Coupled Problems
We showed in Chap. 7 that if the problem is coupled (β = 0), even the normal loading
problem leads to tangential tractions, and if no slip conditions are assumed, the
oscillatory singularity at the edge of the contact area violates the contact inequalities.
Under these conditions, a physically meaningful solution can be obtained assuming
frictional slip occurs in an edge region, though the mathematics is then generally
rather intractable and often only numerical solutions are practicable.
9.5.1 Indentation by a Two-Dimensional Flat Rigid Punch
The simplest case is that in which a rigid flat punch of width 2a is pressed into an
elastic half-plane by a central normal force P. The solution where slip is prevented
was given in Sect. 7.3 and the violations at the contact edges suggest a modified
solution, where frictional slip occurs in two edge regions −a < x < −b and b < x < a.
The boundary conditions can then be stated as
du z
=0 −a < x <a (9.53)
dx
d ḣ
=0 −b < x <b (9.54)
dx
qx (x) = f p(x)sgn(x) b < |x| < a. (9.55)
As in Sect. 7.3, we can remove the time derivative on h by integration as long as the
loading is monotonic, after which Eqs. (7.16, 7.17) allow us to write (9.53, 9.54) in
the form
a
p(ξ)dξ
− πβqx (x) = 0 −a < x <a (9.56)
−a (x − ξ)
a
qx (ξ)dξ
+ πβ p(x) = 0 − b < x < b. (9.57)
−a (x − ξ)
A formal solution to these equations can be obtained by partitioning the integration

domain in Eq. (9.57) so as to write
−b a
b
qx (ξ)dξ p(ξ)dξ
= −πβ p(x) + f − − b < x < b, (9.58)
−b (x − ξ) −a b (x − ξ)
where we have used (9.55) to substitute for qx (ξ) in b < |ξ| < a. This is a Cauchy
singular integral equation on the domain −b < x < b and we require the bounded
solution, since the frictional tractions cannot be unbounded at the stick-slip boundary
x = ±b. Solving this equation for qx (x), we obtain, after some algebra,
9.5 Coupled Problems 185
√ a
2x b2 − x 2 b
p(ξ)dξ p(ξ)dξ
qx (x) = β − f
π 0 (x 2 − ξ 2 ) b2 − ξ 2 b (x 2 − ξ 2 ) ξ 2 − b2
−b < x < b. (9.59)
If this equation and (9.55) are substituted into (9.56), we obtain an integral equa-
tion for p(x) on the domain −a < x < a. The resulting equation is algebraically
challenging, but Spence (1973) gave a solution in which it was reduced to a homo-
geneous Fredholm equation whose first eigenvalue defines the relationship between
the unknown stick-slip boundary b and the parameters f, β. Nowell et al. (1988)
developed a numerical solution of the corresponding Hertzian contact problem, in
which the tractions p(x), qx (x) were approximated by truncated series of Chebyshev
polynomials.
A Goodman Approximation
A more manageable albeit approximate solution can be obtained using the Goodman
approximation of Sect. 7.5. We neglect the influence of the tangential tractions q(x)
on the normal contact pressure distribution, which is therefore given by
P
p(x) = √ , (9.60)
π a2 − x 2
as in Sect. 7.5. The tangential tractions in −b < x < b can then be obtained directly
by substitution into Eq. (9.59) and the unknown dimension b is determined from the
consistency condition associated with Eq. (9.58) which is
b −b a
p(ξ)dξ dx
−πβ p(x) + f − √ = 0. (9.61)
−b −a b (x − ξ) b − x2
2
Using (9.60) for p(x) and evaluating the resulting integrals, we obtain the condition
b
f K (k
) = β K (k) where k= ; k
= 1 − k2, (9.62)
a
and K (k) is the complete elliptic integral of Eq. (2.31).

Equation (9.62) defines a unique relationship between b/a and the ratio β/ f , and
this is shown by the dashed line in Fig. 9.6. We recall that the maximum value of β
is 0.5, and for a rigid indenter,
(1 − 2ν)
β= , (9.63)
2(1 − ν)
so (for example) if ν = 0.3 for the half-space, β = 0.29. However, relatively signifi-
cant slip zones will be predicted if the coefficient of friction is small.
In Spence’s exact solution, the stick-slip boundary depends separately on β and
f , results for f = 0.3, 0.6 and 0.9 being plotted in Fig. 9.6. Each of these curves is
Fig. 9.6 Effect of Dundurs’ constant β and coefficient of friction f on the location of the stick-slip
boundary x = ±b. The dashed line represents the Goodman approximation (9.62)
truncated at the point corresponding to the maximum value of β = 0.5. The Goodman
approximation gives good results for small coefficients of friction, but significant
errors can occur for larger values.
The Case β/ f < 0.4
We notice from Fig. 9.6 that b/a is close to unity for β/ f < 0.4, implying that
slip is confined to a very small region adjacent to the edges. In this range, Spence
(1973) gives an asymptotic approximation to b/a which in the present notation can
be written
arctan( f )
a − b ≈ 8a exp(2Ψ ) exp − , (9.64)

where

1 (1 − λ − ı) π2 λ π2 2
Ψ = ln ≈ 1 + 0.2326πλ + (λ − ) + ... ,
2
2 (1 − ı) 12 15
(9.65)
1
λ = arctan( f β), (9.66)
π
and is a function of β defined in Eq. (7.28).
Now in Sect. 7.3.2, we showed that the assumption of full stick leads to a violation
of the Signorini inequality at a point near the edge defined by Eq. (7.36). However,
with a finite coefficient of friction, a violation of the stick inequality will occur before
this (i.e. further from the edge) at the first point where |qx (x)| = f p(x). Setting Q x = 0
in (7.30) and noting that the contact zone here is defined by −a < x < a, we have
ı
P x +a
p(x) + ıqx (x) = √ , (9.67)
π 1 − β2 a2 − x 2 a−x
and hence
qx (x) x +a
= tan ln , (9.68)
p(x) a−x
using (7.32). Equating this to the coefficient of friction f , we have

x +a arctan( f )
ln = , (9.69)
a−x
and since solutions for x must be close to a, a binomial approximation yields

arctan( f )
a − x ≈ 2a exp − . (9.70)

Notice that if we set qx (x)/ p(x) = − f (corresponding to the opposite direction of

slip) we would find a symmetric point close to x = −a.
Thus, the extent of the slip zone (a − b) is larger than the distance to the point
of first violation of the frictional inequality in the ‘full stick’ solution by the factor
4 exp(2Ψ ). We shall argue in Chap. 10, Sect. 10.5.3 that this is a universal result for
conformal frictional problems if the length so defined is small compared with the
other dimensions of the contact area.
9.5.2 Normal Loading for More General Geometries
Spence (1973, 1975) showed that the normal contact problem for Hertzian or more
general power-law indenters could be related to the corresponding flat punch case by
considering the incremental loading problem, as in Sect. 7.7. In particular, he showed
that during monotonic loading, although both the total contact area and the stick
area increase, the ratio between these areas remains constant and equal to the value
obtained in the flat punch problem. Results were given for both the two-dimensional
(Spence 1973) and axisymmetric (Spence 1975) geometries.
Storåkers and Elaguine (2005) showed that Spence’s argument is not restricted
to power-law indenters, but can be applied to any problem in which the contact area
comprises a single connected region which increases monotonically with normal
force, whilst retaining the same shape. The incremental normal problem clearly
involves uniform incremental displacement over the instantaneous contact region,
and zero tangential displacement over the instantaneous stick region, both of which
conditions are satisfied by the solution of the flat punch problem.
The solution method is essentially similar to that developed in Sect. 7.7, except that
the convolution is performed on the partial slip flat punch solution, rather than on the
full stick solution. Storåkers and Elaguine specifically discussed the axisymmetric
case, but here we shall develop the argument for the two-dimensional case.
As in Sect. 7.7.3, we denote by P(t) the normal force at which the contact area is
defined by −t < x < t, and we postulate that the incremental problem in which the
force increases by P
δt and the contact area extends to −(t+δt) < x < (t+δt) has the
same solution as that obtained in the flat punch solution for the domain −t < x < t.
In particular, we assume that the instantaneous stick zone is defined by −kt < x < kt,
where k = b/a as shown in Fig. 9.6 and approximated in Eq. (9.62).
Dimensional considerations dictate that the flat punch solution take the parametric
form
P P
p(x) = ψ p (ξ, β, f ) ; qx (x) = ψq (ξ, β, f ) (9.71)
t t
du z P du x P
(x) = ∗ φz (ξ, β, f ) ; (x) = ∗ φx (ξ, β, f ) , (9.72)
dx E t dx E t
where ξ = x/t and ψ p , ψq , φz , φx are dimensionless functions satisfying the

conditions
ψ p (ξ, β, f ) = ψq (ξ, β, f ) = 0 ξ>1 (9.73)

ψq (ξ, β, f ) = f ψ p (ξ, β, f ) k<ξ<1 (9.74)
φz (ξ, β, f ) =0 0<ξ<1 (9.75)
φx (ξ, β, f ) =0 0<ξ<k (9.76)
and appropriate frictional inequalities. With the proposed superposition (convolu-

tion), we then define
a P
(t)dt

x
p(x) = ψp , β, f 0<x <a (9.77)
t t
x a
x P
(t)dt
qx (x) = ψq , β, f 0<x <a (9.78)
x t t
x
P
(t)dt
du z 1 x
(x) = ∗ φz , β, f 0<x <a (9.79)
dx E 0 t t
x/k
P
(t)dt
du x 1 x
(x) = ∗ φx , β, f 0 < x < a. (9.80)
dx E 0 t t
Notice that the convolution is performed over the full range 0 < t < a, but the integral
ranges are truncated by conditions (9.73)–(9.76). For example, Eq. (9.73) shows that
ψ p = ψq = 0 for t < x, and this is reflected in the integral ranges in (9.77, 9.78).
Similarly, (9.75, 9.76) show that φz = 0 for t > x and φx = 0 for t > x/k, leading to
the integral ranges in (9.79, 9.80).
The contact condition du z /d x(x) = −g0
(x) in 0 < x < a then gives
x
x P
(t)dt ∗
φ , β, f = −E g0
(x). (9.81)
0 t t
This is a Volterra integral equation for the unknown function P(t), which can be
solved by power series expansion, as in Eqs. (7.95, 7.96). Once P(t) is determined,
the tractions are obtained by substitution into (9.77, 9.78).
It is easily shown that the solution (9.77)–(9.80) satisfies the frictional boundary
conditions. For example, the integral range in (9.78) corresponds to x/a < x/t < 1,
so if x/a > k the integrand satisfies (9.74) for all t, ensuring that qx (x) = f p(x) as
required in the instantaneous slip zone ka < x < a. Also, for x < ka, the integral in
(9.80) is independent of a and hence the accumulated axial strains are ‘locked in’
and see no further change as the force increases beyond P = P(x/k).
The same method can clearly be applied to the axisymmetric normal indentation
problem, as in Sect. 7.7.2 (see Storåkers and Elaguine 2005).
9.5.3 Combined Normal and Tangential Loading
The preceding results apply specifically to the case of monotonic normal loading.
If time-varying normal and tangential forces P(t), Q x (t) are both applied, we can
again identify an incremental problem, but this will be similar to that for an equivalent
flat punch problem with simultaneously applied force increments P
(t)δt, Q
x (t)δt
if and only if (i) the predicted stick region is non-decreasing, and (ii) the slip in
each of the slip regions is in the same direction as that immediately before the
force increment. Solutions of this kind were obtained by Nowell et al. (1988) using
truncated Chebyshev polynomial series to represent the contact tractions.
9.5.4 Unloading
If the applied forces are increased to some maximum value and then reduced, quite
complex arrangements of stick and slip zones can be developed. Turner (1979) con-
sidered the case of purely normal loading and unloading of an axisymmetric flat
punch against an elastic half-space. Experience with uncoupled problems suggests
that the entire contact area should stick at the beginning of the unloading process, fol-
lowed by the growth of an annulus of reverse slip at the outer edge. However, Turner
found that at the beginning of unloading, slip continues in the same direction as that
during loading in an annulus contained between a central circle and a surrounding
annulus of stick. Then, after the force is reduced below about half of the maximum
value, an outer annulus of reverse slip develops. The radii of all these annuli reduces
as the force is reduced, with the entire contact area being instantaneously in a state
of reverse slip just before the force reaches zero.
We now know from asymptotic analysis3 that stick is not possible at the edge
of the punch, so we must conclude that Turner’s numerical method failed to detect
an extremely small annulus of reverse slip during the early stages of the unloading
process.
9.5.5 Periodic Loading
Numerical solutions for the periodic normal loading of a coupled Hertzian contact
were given by Stingl et al. (2013) for the two-dimensional (cylindrical) case and by
Kim and Jang (2014) for the axisymmetric (spherical) case. Kim and Jang used a
finite element description of the elastic body and the static reduction technique of
Sect. 8.3 to determine the contact stiffness matrix K , followed by Ahn’s algorithm
(Ahn and Barber 2008) for the frictional evolution problem. After first loading, they
cycled the normal force between non-zero minimum and maximum values and found
that the amplitude of slip in the ‘contained’ slip zones decreased during the first few
cycles, so that in the steady state, only an outer annulus of the contact area experiences
cyclic slip.
Stingl et al. used Kalker’s CONTACT algorithm (Kalker 1990), which is based
on the Green’s functions for the half-space from Eq. (7.7), combined with a comple-
mentary energy statement of the frictional and unilateral boundary conditions. They
showed that the energy dissipated in friction is less in the steady state than in the
first few cycles. They also reported a power-law dependence of dissipation on force
amplitude, closer to quadratic than the cubic power predicted (for example) for small
amplitude periodic loading in the uncoupled Ciavarella–Jäger problem (Putignano
et al. 2011).
Problems
1. Use the field-point integration method of Sect. 2.3.1 to determine the tangential
surface displacements u x (r ), u y (r ) due to the traction distribution
∗√ 2
2fE a − r2
qx (r ) = 0≤r <a
πR
in the region r > a, outside the loaded circle.
Hence find the direction of the slip displacements in b <r < a due to the Cattaneo–
Mindlin distribution
2 f E
2
∗
qx (r ) = a − r 2 − b2 − r 2 ,
πR
3 See Sects. 10.2 and 7.3.2.

Problems 191
Fig. 9.7 Indentation by a

semi-cylindrical punch
P
where the square roots are to be interpreted as zero in regions where their respective
arguments are negative. Find the angle between the resultant slip displacement and
the local tangential traction as a function of r and comment on the results.
2. A rigid punch in the form of a half-cylinder of radius R is pressed into an elastic

half-plane such that the plane side of the punch remains vertical, as shown in Fig. 9.7.
Find the relationship between the indenting force P and the width a of the contact
area.
A tangential force Q x is now applied to the punch, with P remaining constant.
Use the Ciavarella–Jäger theorem to find the extent of the slip zone as a function of
Qx .
3. A rigid cylinder of radius R is pressed into an elastic half-plane by a normal force

P leading to the contact pressure distribution of Eq. (6.25). This force is then held
constant whilst a tangential force Q x oscillates between zero and f P/2, where f is
the coefficient of friction. Find the maximum extent of the slip zones in the steady
state.
4. The flat and rounded punch of Fig. 9.8 is loaded by a constant normal force P0 and
a periodic tangential force Q x which alternates between ±Q 0 , where Q 0 < f P0 . Use
the Ciavarella–Jäger theorem to show that the flat region −b < x < b never slips.
Microslip in the region b < |x| < a can be expected to cause wear which will
modify the profile of the punch and hence the contact pressure distribution. Show
Fig. 9.8 A flat and rounded

punch subjected to periodic P0
tangential forces
Q0
radius R
b
a
that whatever the form of the worn profile, the flat region will never slip and hence
will also never wear. Discuss the implications for fretting fatigue problems.
5. Use the field-point integration method of Sect. 2.3.1 to determine the tangential
displacement in the stick zone due to the traction distribution (9.4). Notice that the
origin lies in the stick zone, so it is sufficient to set x = y = 0 in C0 (θ), C1 (θ) of
Eq. (2.26), which considerably simplifies the integration.
Use this result to establish the relation between the tangential force Q x and the
tangential surface displacement u x during periodic loading, and hence determine the
energy loss in friction per cycle as a function of the amplitude Q 0 in periodic loading
between Q 0 and −Q 0 .
6. A cylinder of radius R is pressed into an elastic half-plane by a normal force P

and a tangential force Q x that vary with time according to Eq. (9.23) with Q 0 = 0 and
P1 = Q 1 = 0.2P0 . If the coefficient of friction is 0.4, find the extent of the permanent
stick zone.
7. A symmetrical elastic body is pressed into an elastic half-plane by a normal force

P, whilst the half-plane is subjected to a bulk tension σ. Use Eq. (9.31) as the basis
for an incremental formulation analogous to that in Sect. 6.3.1 to show that the entire
instantaneous contact area will stick as long as

dP dσ 4 f
>0
dt
and d P < πa ,
where t is time and a is the semi-width of the instantaneous contact area.
8. An uncoupled two-dimensional Hertzian contact is loaded initially by a purely

normal force P0 , but is then subjected to a combination of normal loading and bulk
stress increasing in time t according to the equations
P1 t σ1 t
P = P0 + ; σ= 0 < t < t0 ,
t0 t0
where P1 > 0, but the ratio σ1 /P1 is sufficiently large to ensure that full stick is
impossible throughout this second phase.
Write down the boundary conditions defining the difference between the traction
and displacement states at t = 0 and t = t0 and show that this differential problem
has a similar form to that in Sect. 9.4.1. Hence obtain an equation whose solution
defines the extent of the stick zone at t = t0 .
9. The inclined rigid punch of Fig. 6.3 is loaded by a normal force P0 which is then
held constant whilst bulk stresses σ1 , σ2 and a tangential force Q 0 are applied in
proportion. If the problem is uncoupled and P0 is insufficient to ensure full contact,
find the extent of the stick zone under the assumption of ‘moderate bulk stress’ and
Problems 193
hence determine the condition which must be satisfied for this assumption to be
appropriate.
10. An uncoupled two-dimensional Hertzian contact is loaded by a normal force and

bulk tension [in body 1 only] that increase in proportion up to their maximum values
P0 , σ0 . Assume that σ0 is sufficiently small to ensure that this process involves no
slip. Use an incremental formulation to find the resulting distribution of tangential
traction.
These loads are then maintained constant whilst an increasing tangential force is
applied, with maximum value Q 0 . Describe in words, what slip zones you expect
to see developed and set up the equations which must be satisfied by the corrective
solution — i.e. the difference between the final state and that before Q 0 was applied.
11. Use the Goodman approximation to formulate the problem of a two-dimensional

flat rigid punch loaded by normal and tangential forces that increase in proportion
— i.e. P(t) = C N t, Q x (t) = C T t, where t is time and C T < f C N . Find two equations
for the coordinates b, c defining the extent of the stick region −b < x < c from (i)
the consistency condition in the equation analogous to (9.58) and (ii) the equilibrium
condition a
Qx = qx (x)d x.
−a
Do not attempt to solve these equations.
12. Use the potential function representation of Appendix A, Sects. A.1 and A.2 to
formulate the coupled problem of a rigid cylindrical flat punch of radius a pressed
into an elastic half-space by a force P, assuming a finite coefficient of friction f .
Satisfy the boundary conditions outside the contact area identically by choosing
appropriate functions related to that in Eq. (5.4) to represent the potential functions.
Then use the remaining boundary conditions to set up integral equations for the
unknown functions [e.g. h(t) in Eq. (5.4)]. Do not attempt to solve these equations.
13. The flat punch of Sect. 9.5.1 is first loaded by a normal force P, after which a
compressive bulk stress σ1 = −s is applied to the half-plane. Set up the Cauchy inte-
gral equations defining the resulting traction distributions and then use the Goodman
approximation to estimate the extent of the slip zones, if the coefficient of friction is f .
What do you think would happen if instead a tensile bulk stress were applied?
Chapter 10
Asymptotic Methods
The inequalities defining unilateral contact and frictional slip control the partition of
the potential contact surface into regions of slip, stick or separation. The asymptotic
elastic fields at the boundaries between these regions generally exhibit singularities,
either in the stresses or in the stress gradients, and the properties of the corresponding
singular fields often lead to results of some generality. For example, we can show
that in any non-conformal frictionless elastic contact problem, the contact pressure
must tend to zero with the square root of the distance from the edge of the contact
region, provided that the contact surfaces are continuous up to and including the first
derivative.
The method of asymptotic analysis was pioneered by Williams (1952), who inves-
tigated the stress field in the corner of a sharp notch, or in the limiting case, a crack
tip. Here, we shall introduce it by considering the fields at the edge of a contact
region between a frictionless rigid punch and an elastic half-plane.
10.1 Indentation by a Frictionless Rigid Punch
Figure 10.1a shows a frictionless rigid punch with ‘sharp’ corners, pressed into an
elastic body by a force P. We focus attention on the right edge of the contact region
x = a and define a system of polar coordinates (r, θ) with this point as origin and the
line θ = 0 in the direction of the negative x-axis on the undeformed body, as shown
in Fig. 10.1b.
We shall be concerned with the stress and displacement fields very near to the
origin, so we magnify the scale of Fig. 10.1b by an arbitrarily large factor. It follows
that (i) the curvature of the punch base (if any) can be regarded as negligible (since
the local radius will look very large once magnified) and (ii) the other edge x = b of
the contact region in Fig. 10.1a and the remaining boundaries of the elastic body are

https://doi.org/10.1007/978-3-319-70939-0_10
196 10 Asymptotic Methods
(a)
P (b)
rigid
traction-free
b a r elastic
Fig. 10.1 (a) Indentation by a rigid punch, (b) greatly expanded view of the region close to the
point x = a
a long way away relative to the region illustrated in Fig. 10.1b, so the elastic body
can be regarded as an infinite half-plane with contact extending to infinity on the left.
In the polar coordinate system, the half-plane occupies the region 0 ≤ θ ≤ π, with
the boundary θ = 0 corresponding to the [frictionless] contact region and θ = π to the
traction-free separation region. Thus, the stress and displacement fields must satisfy
the boundary conditions
∂u θ
σθr = 0; = 0 all r, θ = 0 (10.1)
∂r
σθr = 0; σθθ = 0 all r, θ = π. (10.2)
Since the problem of Fig. 10.1b has no inherent length scale, we anticipate partic-
ular solutions in which the stresses and displacements have separated-variable form,
with power-law dependence on the radial coordinate r . These solutions are readily
obtained by writing
u r = r λ fr (θ); u θ = r λ f θ (θ), (10.3)
using the strain–displacement relations and Hooke’s law to find the corresponding
stress components σrr , σr θ , σθθ , and then substituting these into the two equilib-
rium equations in polar coordinates, leading to a pair of simultaneous homogeneous
ordinary differential equations for the functions fr (θ), f θ (θ).
We find that the required solutions can be written
σrr = λr λ−1 {−A1 (λ + 1) cos(λ + 1)θ − A2 (λ − 3) cos(λ − 1)θ

−A3 (λ + 1) sin(λ + 1)θ − A4 (λ − 3) sin(λ − 1)θ} (10.4)
σr θ = λr λ−1 {A1 (λ + 1) sin(λ + 1)θ + A2 (λ − 1) sin(λ − 1)θ
−A3 (λ + 1) cos(λ + 1)θ − A4 (λ − 1) cos(λ − 1)θ} (10.5)
σθθ = λr λ−1 {A1 (λ + 1) cos(λ + 1)θ + A2 (λ + 1) cos(λ − 1)θ
+A3 (λ + 1) sin(λ + 1)θ + A4 (λ + 1) sin(λ − 1)θ} , (10.6)
10.1 Indentation by a Frictionless Rigid Punch 197
Eu r
= r λ {−A1 (λ + 1) cos(λ + 1)θ + A2 (κ − λ) cos(λ − 1)θ
(1 + ν)
−A3 (λ + 1) sin(λ + 1)θ + A4 (κ − λ) sin(λ − 1)θ} (10.7)
Eu θ
= r λ {A1 (λ + 1) sin(λ + 1)θ + A2 (κ + λ) sin(λ − 1)θ
(1 + ν)
−A3 (λ + 1) cos(λ + 1)θ − A4 (κ + λ) cos(λ − 1)θ} (10.8)
(Barber 2010, Sect. 11.2.2), where A1 , A2 , A3 , A4 are arbitrary constants and
κ = 3 − 4ν. (10.9)
Substituting (10.4)–(10.8) into (10.1), (10.2), we obtain four homogeneous algebraic

equations for the four constants, which have a non-trivial solution only for certain
eigenvalues of the exponent λ. The eigenvalues are restricted to the range λ > 0 [or
more strictly (λ) > 0] by the requirement that the strain energy in a finite region
containing the origin be bounded (Barber 2010, Sect. 11.2.1).
From (10.1), (10.5), (10.8), we obtain the equations
− A3 (λ + 1) − A4 (λ − 1) = 0; −A3 (λ + 1) − A4 (κ + λ) = 0, (10.10)
which have only the trivial solution A3 = A4 = 0, since λ = −1 is not an acceptable

eigenvalue. Using this result in (10.5), (10.6) and substituting into (10.2), we then
obtain
A1 (λ + 1) sin(λ + 1)π + A2 (λ − 1) sin(λ − 1)π = 0 (10.11)

A1 (λ + 1) cos(λ + 1)π + A2 (λ + 1) cos(λ − 1)π = 0, (10.12)
and these equations have a non-trivial solution if and only if

n
sin(2λπ) = 0 or λ = , (10.13)
2
where n is any positive integer.
10.1.1 Eigenfunction Series
For each value of n, a non-trivial stress field can be constructed satisfying (10.1),
(10.2), and a general solution satisfying these conditions can then be constructed by
superposition over all integer values of n with appropriate multiplying constants. In
effect, the stress field local to the origin is represented as an eigenfunction expansion,
∞

Bn r 2 −1 fn (θ),
n
σ(r, θ) = (10.14)
n=1
Fig. 10.2 Deformation and

pressure distribution at the
edge of a rigid punch
indenting an elastic 1/2
p~r deformed surface
half-space
1/2
uθ ~ r
where the eigenfunctions fn are obtained by using (10.11), (10.12) to express the
constants in terms of a single constant Bn for λ = n/2, and then substituting back
into Eqs. (10.4)–(10.6).
Very close to the origin, only the first non-zero term in this series is important,
√
since the ratio between the (n+1)th term and the nth term is proportional to r as
long as Bn = 0 and hence tends to zero as r → 0. For the first term (n = 1), the contact
pressure and the normal displacement of the free surface take the form
4B1r 1/2
p(r ) = −σθθ (r, 0) = B1r −1/2 ; u θ (r, π) = , (10.15)
E∗
∗
where E = E/(1−ν 2 ) since the indenting body is rigid. We conclude that if the con-
tact pressure is to remain positive near the corner, we must have B1 > 0. It then follows
that the surface displacement will also be positive near the corner and will have the
form shown in Fig. 10.2. Notice in particular that the contact pressure is square-root
singular at the edge of the contact area, whilst the normal surface displacement is
square-root bounded.
10.1.2 More General Frictionless Indentation Problems
We introduced this procedure using the two-dimensional problem of Fig. 10.1a, but
it applies equally to three-dimensional problems. For example, if a half-space is
indented by a flat-ended cylindrical punch as in Sect. 5.1.1, when we focus our
attention on the region very near to a point at the edge of the contact region, the
implied magnification will make the radius a of the punch appear very large, so in
the limit the edge of the punch becomes straight and the local problem becomes
identical to Fig. 10.1b.
It also follows that the local stress field will be one of plane strain, since gradients
in the direction normal to the cross section of Fig. 10.1b will tend to zero with
sufficient magnification. This raises a question regarding problems that might more
naturally be formulated in plane stress, such as an otherwise traction-free thin plate
10.1 Indentation by a Frictionless Rigid Punch 199
with a plane edge, indented by a flat rigid punch. In this case, as long as the distance
r is small compared with the thickness of the plate, conditions will indeed be those
of plane strain, except in a small region which is equally near to the traction-free
sides of the plate. This problem is therefore strictly three-dimensional. Similar three-
dimensional effects occur in problems involving cracks in thin plates and lead to the
development of curved crack fronts during fracture.
10.1.3 Non-conformal Problems
The displacement field of Eq. (10.15) [with B1 > 0] satisfies the contact conditions for
the problem of Fig. 10.1a, but if the origin represents the boundary between contact
and separation in a non-conformal contact problem, it is clear that the square-root
bounded displacements in the separation region in Fig. 10.2 would imply interpene-
tration with any indenter whose slope is continuous.
For the non-conformal case, we need to satisfy the two inequalities
∂u θ
σθθ (r, 0) ≤ 0 and (r, π) ≤ 0, (10.16)
∂r
and Eq. (10.15) shows the only value of B1 that satisfies both conditions is B1 = 0.
The next term (n = 2) in the eigenfunction series (10.14) gives zero contact pres-
sure. In fact, it corresponds to a state of uniform tension or compression parallel to
the contact interface [see Sect. 9.4]. This produces strains eθθ parallel to the inter-
face, but these do not affect the contact problem under frictionless conditions. It is
analogous to the ‘T-stress’ in fracture mechanics (Tvergaard and Hutchinson 1994).
The next non-zero term in the expansion of p(r ) corresponds to n = 3, for which
4B3r 3/2
p(r ) = −σθθ (r, 0) = B3r 1/2 ; u θ (r, π) = − . (10.17)
3E ∗
In this case, the two inequalities (10.16) are both satisfied by the choice B3 > 0,
from which we conclude that in general the contact pressure in a non-conformal
contact problem will be square-root bounded at the edge of the contact region, and
the normal surface displacement will vary with the three-halves power of distance
from this edge.
Fig. 10.3 Non-conformal

contact of two deformable
body 2
bodies θ = −π
θ θ=π
r body 1
10.1.4 Both Materials Deformable
The asymptotic fields for frictionless non-conformal contact between two deformable
bodies are essentially similar to those where one body is rigid. We need to define
separate stress and displacement fields in the two bodies, using Eqs. (10.4)–(10.8),
so there will now be eight unknown constants, four for each body.
If bodies 1 and 2 are taken to occupy 0 ≤ θ ≤ π and −π ≤ θ ≤ 0 respectively, and
if contact is again assumed to occur at θ = 0 as shown in Fig. 10.3, we have the
following eight homogeneous boundary conditions to determine these constants:-
(1) (1) (2) (2)
σθr (r, π) = 0; σθθ (r, π) = 0; σθr (r, −π) = 0; σθθ (r, −π) = 0
(1) (2)
σθr (r, 0) = 0; σθr (r, 0) = 0 (10.18)
(1) (2) ∂u (1)

θ ∂u (2)
θ
σθθ (r, 0) = σθθ (r, 0); (r, 0) = (r, 0).
∂r ∂r
However, the eigenvalues are again given by Eq. (10.13) and as before the first
non-trivial eigenfunction that satisfies the contact inequalities corresponds to n = 3.
The dominant terms in the contact pressure and the gap g(r ) in the separation zone
are
4B3r 3/2
p(r ) = B3r 1/2 ; g(r ) = u (2) (1)
θ (r, −π) − u θ (r, π) = , (10.19)
3E ∗
∗
where E is now given by the more general relation (2.15).
It is worth noting that the boundary value problems defined by Eqs. (10.1), (10.2)
and (10.18) could in principle be formulated using the Green’s function of Eq. (6.6),
and hence the resulting eigenvalues and eigenfunctions depend only on the properties
of this Green’s function. In particular, the eigenvalues (10.13) are dictated by the
dependence of (6.6) on 1/x, and the elastic constant in (6.6) appears in the ratio
between displacement and traction measures in Eqs. (10.15), (10.17), (10.19).
Now the factor 1/x in (6.6) is a consequence of linearity, self-similarity and equi-
librium and, as discussed in Sect. 2.2.2, the same result must be obtained for generally
anisotropic elastic materials. We therefore conclude that the eigenvalues (10.13) also
apply to the frictionless contact of two anisotropic materials. Furthermore, since the
stiffness of any elastic body must be positive definite, the sign of the elastic constant
in the Green’s function cannot be affected by the anisotropy. It follows that for the
most general anisotropic material, the contact pressure must go to zero with r 1/2 at
the edge of the contact region in non-conformal frictionless elastic contact problems,
whereas for conformal indentation by a rigid punch, the contact pressure must be
square-root singular.
10.2 No-Slip Conditions 201
10.2 No-Slip Conditions
If the friction coefficient is sufficient to prevent slip at the edge of the contact area
in Fig. 10.1b, the boundary conditions (10.1), (10.2) are modified to
∂u r ∂u θ
= 0; =0 all r, θ = 0 (10.20)
∂r ∂r
σθr = 0; σθθ = 0 all r, θ = π. (10.21)
Using Eqs. (10.4)–(10.8), we obtain the four homogeneous equations
−A1 (λ + 1) + A2 (κ − λ) = 0
−A3 (λ + 1) − A4 (κ + λ) = 0
A1 (λ + 1) sin(λ + 1)π + A2 (λ − 1) sin(λ − 1)π
−A3 (λ + 1) cos(λ + 1)π − A4 (λ − 1) cos(λ − 1)π = 0
A1 (λ + 1) cos(λ + 1)π + A2 (λ + 1) cos(λ − 1)π
+A3 (λ + 1) sin(λ + 1)π + A4 (λ + 1) sin(λ − 1)π = 0,
which have a non-trivial solution if and only if
(1 + κ2 )
cos(2λπ) = − . (10.22)
2κ
Since κ = 3−4ν ≥ 1, the magnitude of the right-hand side of this equation is greater
than unity except in the special case where ν = 0.5. For all other cases, only complex
solutions exist for λ and we can find these by writing 2λπ = c + ıd and separating
real and imaginary parts in (10.22), obtaining
(1 + κ2 )
cos(c) cosh(d) = − ; sin(c) sinh(d) = 0. (10.23)
2κ
We have already established that the solution cannot be real, so d = 0 and it follows
from the second equation that c = mπ where m is an integer, and hence that cos(c) =
(−1)m . Substitution in the first equation then yields
(1 + κ2 )
cosh(d) = (−1)m+1 , (10.24)
2κ
which has a solution only for odd values of m (= 2n−1). We conclude that the
eigenvalues of Eq. (10.22) are
1
λ=n− ± ı , (10.25)
2
where n is a positive integer and
(1 + κ2 ) ln(κ)
cosh(2π ) = or = . (10.26)
2κ 2π
If both materials are deformable, the boundary conditions (10.18) are modified
for no slip to
(1) (1) (2) (2)
(1) (2) (1) (2)

σθr (r, 0) = σθr (r, 0); σθθ (r, 0) = σθθ (r, 0) (10.27)
∂u r(1) ∂u r(2) ∂u (1)

θ ∂u (2)
θ
(r, 0) = (r, 0); (r, 0) = (r, 0),
∂r ∂r ∂r ∂r
and a similar procedure leads once again to (10.25), except that is now given by
the more general expression

1 1+β
= ln , (10.28)
2π 1−β
where Dundurs constant β is defined in Eq. (7.8) and Appendix D. When one body
is rigid (e.g. E 2 → ∞), we have
(1 − 2ν) κ−1
β= = , (10.29)
2(1 − ν) κ+1
and hence (10.28) reduces to (10.26).

These results show that the asymptotic behaviour we observed for the flat punch
problem in Sect. 7.3 applies to any conformal contact problem under no-slip con-
ditions. In particular, the tractions and displacements oscillate as r → 0, and hence,
the no-slip condition necessarily implies local violation of the inequality condi-
tions except in the uncoupled case β = 0. This issue is discussed in more detail in
Sect. 7.3.2.
10.3 Frictional Slip
In frictional problems, we anticipate that the contact area will contain regions of stick
and microslip, so two additional kinds of asymptotic field might be investigated —
those at the transition from stick to slip, and those from slip to separation.
10.3 Frictional Slip 203
10.3.1 Slip-Separation Transition
In this case for two deformable materials, the appropriate boundary conditions in
Fig. 10.3 are
(1) (1) (2) (2)
(1) (1) (2) (2)

σθr (r, 0) = f σθθ (r, 0); σθr (r, 0) = f σθθ (r, 0) (10.30)
(1) (2) ∂u (1)

θ ∂u (2)
θ
σθθ (r, 0) = σθθ (r, 0); (r, 0) = (r, 0),
∂r ∂r
where the Coulomb friction conditions are appropriate to the case where the upper
body 2 slips to the left relative to the lower body 1. The conditions for the opposite
direction of slip are obtained simply by replacing f by − f .
Defining the stress and displacement fields through Eqs. (10.4)–(10.8), and sub-
stituting into (10.30), we obtain eight simultaneous homogeneous equations for the
coefficients which have a non-trivial solution if and only if
sin(λπ) = 0 or cot(λπ) = f β, (10.31)
where β is defined in Eq. (7.8). The first of these conditions is satisfied by any integer
value of λ, but the resulting stress fields do not involve contact tractions.
The solution of Eq. (10.31)2 is illustrated graphically in Fig. 10.4, where the func-
tion cot(λπ) is plotted as a solid line. For f = 0 (frictionless) or β = 0 (uncoupled),
the solutions have the form λ = n−1/2, where n is a positive integer. The dashed
line in Fig. 10.4 defines a positive value f β = 0.4 and shows that the corresponding
values of λ are reduced relative to the frictionless values. By contrast, λ is increased
if f β < 0. Since −0.5 < β < 0.5 for all material combinations, the factor f β is gen-
erally fairly small, so the intersections A, B are usually quite close 0.5 and 1.5,
respectively.
As in Sect. 10.1.3, the first eigenvalue, corresponding to the point A in Fig. 10.4
violates one or other of the inequalities
∂u θ (1) ∂u θ (2)
σθθ (r, 0) ≤ 0 and (r, π) ≤ (r, −π), (10.32)
∂r ∂r
for non-conformal contact problems, for all non-zero values of the multiplying con-
stant, and hence, the first acceptable eigenvalue corresponds to the point B near
λ = 1.5.
If the upper body 2 is rigid, these results can be applied to the conformal geometry
of Fig. 10.1b, where β is now given by Eq. (10.29) and is positive for all values of ν.
With this geometry, the second (displacement) inequality in (10.32) can be relaxed
and the stress field will generally exhibit a singularity corresponding to the point A
Fig. 10.4 Graphical solution of Eq. (10.31)2
in Fig. 10.4. We recall that f > 0 corresponds to the case where the upper body (now
rigid) slips to the left, so it follows that in a conformal sliding contact problem, the
contact pressure singularity is stronger than square root (λ < 0.5) at the trailing edge
and weaker than square root (λ > 0.5) at the leading edge.
10.3.2 Slip–Stick Transition
A transition between slip and stick can occur only at an interior point of the con-
tact area for both conformal and non-conformal contact problems. If we define the
coordinate system such that we have stick at θ = 0 and slip at θ = ±π, the boundary
conditions are
(1) (2) (1) (2)
σθr (r, 0) = σθr (r, 0); σθθ (r, 0) = σθθ (r, 0)
∂u r(1) ∂u r(2) ∂u (1)

θ ∂u (2)
θ
(r, 0) = (r, 0); (r, 0) = (r, 0)
∂r ∂r ∂r ∂r
(1) (1) (2) (2)
σθr (r, π) = f σθθ (r, π); σθr (r, −π) = f σθθ (r, −π) (10.33)
(1) (2) ∂u (1)

θ ∂u (2)
θ
σθθ (r, π) = σθθ (r, −π); (r, π) = (r, −π),
∂r ∂r
where once again we have chosen the sign of f such that f > 0 corresponds to the
case where body 2 slips to the left (in θ = ±π) relative to body 1. The resulting
equations have a non-trivial solution if and only if
sin(λπ) = 0 or cot(λπ) = − f β. (10.34)

10.3 Frictional Slip 205
This differs from (10.31) only in the sign of the term f β and hence the solutions are
again determined by appropriate intersections in Fig. 10.4. In this case, the leading
term in the eigenfunction series must satisfy the inequalities
σθθ (r, 0) ≤ 0; σθθ (r, ±π) ≤ 0

∂u r (1) ∂u r (2)
σθr (r, π) (r, π) − (r, −π) ≤ 0, (10.35)
∂r ∂r
where the last condition states that the frictional traction σθr must oppose the slip
direction. As in previous cases, the first eigenfunction violates one or other of these
conditions for all non-zero multipliers, so the first non-zero term corresponds to an
intersection near λ = 1.5.
It can be shown that σθθ (r, 0) = 0 for all the eigenvalues of (10.34)2 , and hence
the contact pressure distribution is smooth on the ‘stick’ side of the transition and
exhibits an approximately square-root bounded discontinuity on the ‘slip’ side. This
behaviour is clearly visible in the traction distributions reported by Spence (1975)
for the coupled indentation problem. Of course in the uncoupled case β = 0, the
contact pressure is smooth on both sides of the transition, since it is determined
independently of the frictional conditions. In this case, the tangential traction is
smooth on the slip side [as indeed it must be since it is proportional to the contact
pressure], but we obtain a square-root bounded discontinuity on the stick side. This
behaviour is illustrated in the Cattaneo–Mindlin distribution of Fig. 9.2.
10.4 Indentation by an Elastic Wedge
If an elastic body with a sharp corner is pressed into an elastic half-plane as

in Fig. 10.5a, the same method can be used to determine the nature of the local
stress field. In fact, the only change required is that the traction-free conditions
(2) (2)
σθr (r, −π) = 0, σθθ (r, −π) = 0 in (10.18) are replaced by similar conditions on the
(a) (b)
body 2 traction-free body 2 traction

α free
α
θ θ
r body 1 r body 1
Fig. 10.5 (a) Indentation by an elastic punch, (b) the case α > π/2
Fig. 10.6 Indentation of an

elastic half-plane by a traction-free
right-angle wedge body 2
θ
r body 1
exposed face of the body θ = −α, giving

(2) (2)
σθr (r, −α) = 0; σθθ (r, −α) = 0. (10.36)
If α > π/2, we might anticipate a small region of contact on the inclined face
θ = −α as shown in Fig. 10.5b. An asymptotic solution for this case is given by
Churchman et al. (2006a, b), but in many cases, the indenting body will have a right-
angle corner (α = π/2) as shown in Fig. 10.6, and we shall consider this case in some
detail.
10.4.1 Right-Angle Wedge of the Same Material
If the materials are similar and if we have non-slip conditions as in Sect. 10.2, the
problem is equivalent to that of a monolithic body occupying the region −π/2 < θ <
π, with traction-free boundary conditions at θ = π and θ = −π/2. This is a special
case of Williams’ solution and the eigenvalues are solutions of the equations1

3πλ 3πλ
sin − λ = 0 and sin + λ = 0. (10.37)
2 2
The monolith is symmetric about the plane θ = π/4 and Eqs. (10.37)1 and (10.37)2
correspond to stress and displacement fields that are respectively symmetric and
antisymmetric about this plane. However, the symmetry plane does not coincide
with the contact plane (θ = 0), so both symmetric and antisymmetric eigenfunctions
involve non-zero shear and normal tractions on this plane given by

λ−1 3πλ λ−1 3πλ
σθr (r, 0) = C(1 − λ)r sin ; σθθ (r, 0) = Cλr cos ,
2 2
(10.38)
where C is an arbitrary constant related to the constants A1 , A2 , A3 , A4 in Eqs. (10.4)–
(10.8). We also note that the ratio between the shear traction σθr (ρ, 0) and the contact
pressure p = −σθθ (ρ, 0) is a constant for any given eigenfunction, being
1 Williams (1952). See also Barber 2010, Sect. 11.2.3.

10.4 Indentation by an Elastic Wedge 207

σθr (r, 0) (1 − λ) 3πλ
R= =− tan . (10.39)
−σθθ (r, 0) λ 2
Close to the corner, conditions are dominated by the lowest eigenvalue. This
corresponds to the symmetric solution with λ ≈ 0.5445, for which the ratio R = 0.543.
If the coefficient of friction exceeds this value ( f > 0.543), and if the interface distant
from the corner is stuck, there are only two possibilities regardless of the external
loading and the particular geometry of the contacting bodies. Either the constant C
is positive indicating separation at the corner (since otherwise the normal contact
tractions would be predicted to be tensile), or C is negative indicating stick at the
corner (Churchman and Hills 2006a). In other words, there is no system of external
forces that could cause a localized region of contact with slip in such a corner.
10.4.2 A Slipping Interface
Instead, if we assume that slip occurs adjacent to the corner in Fig. 10.6, the boundary
conditions for the asymptotic problem are
(1) (1) (2) (2)
σθr (r, π) = 0; σθθ (r, π) = 0; σθr (r, −π/2) = 0; σθθ (r, −π/2) = 0
(1) (1) (2) (2)

σθr (r, 0) = f σθθ (r, 0); σθr (r, 0) = f σθθ (r, 0) (10.40)
(1) (2) ∂u (1)

θ ∂u (2)
θ
σθθ (r, 0) = σθθ (r, 0); (r, 0) = (r, 0).
∂r ∂r
This problem was first solved by Gdoutos and Theocaris (1975) and Comninou
(1976) and the resulting characteristic equation is

λπ sin2 (λπ)
sin2 − λ2 cos(λπ) + + f λ(1 + λ) sin(λπ) = 0. (10.41)
2 2
We recall that f has been defined as positive for the case where the upper body
moves to the left in Fig. 10.6, in which case this figure could define the trailing edge
of a rectangular body sliding over the half-plane. We shall therefore refer to the case
f > 0 as trailing edge slip, and f < 0 as leading edge slip.
The lowest eigenvalue of Eq. (10.41) is real for −∞ < f < 0.392 and complex for
f > 0.392. Thus, a slip zone can occur at the leading edge of a contact ( f < 0) for all
values of | f |, but at a trailing edge ( f > 0) the slip assumption leads to an oscillatory
field in the corner for f > 0.392, which therefore necessarily involves unacceptable
regions of tensile contact traction. This situation is analogous to those discussed
in Sects. 7.3.2 and 10.2, but here neither stick nor slip is possible in the corner if
0.329 < f < 0.543. The only remaining possibility is that separation occurs in the
corner, and this has been verified by Karuppanan et al. (2008), who superposed
an appropriate distribution of climb and glide dislocations in order to satisfy the

unilateral inequalities in the corner region. These authors also developed a finite
element solution of the problem of a rectangular elastic block sliding over a similar
elastic half-plane and confirmed the local behaviour predicted by the asymptotic
analysis. It is also interesting to note that the lowest (real) eigenvalue at the trailing
edge for 1/π < f < 0.392 corresponds to a bounded stress field in the corner, whereas
the field is weakly singular for 0 < f < 1/π.
10.5 Local Fields
In many cases, the stress and displacement fields near to the edge of the contact area
will differ qualitatively from those elsewhere. For example, when a rigid flat punch
is pressed into an elastic half-space, frictional slip occurs in a region near the edge,
and we see from Fig. 9.5 that this region is extremely small if β/ f < 0.4. Another
example is the ‘flat and rounded’ indenter of Fig. 6.4, where contact will extend a
small distance around the curved portion of the boundary.
In cases of this kind, a good approximation to the complete solution can often
be obtained by (i) solving the simpler problem where the small edge regions are
neglected, and then (ii) ‘patching in’ a universal local field defined by the asymptotic
parameters obtained in stage (i) and local geometric features, such as the rounding
radius in Fig. 6.4. This procedure is analogous to the assumption of ‘small-scale yield-
ing’ (Rice 1974) near the crack tip in Linear Elastic Fracture Mechanics (LEFM). The
nonlinear fields near the corner extend over such a small region that they have no sig-
nificant effect on the surrounding elastic fields, that are therefore characterized by the
leading term(s) in the asymptotic expansion of the simpler problem. The asymptotic
fields are completely defined once the multiplying constants are known. In the case
of LEFM, these comprise the mode I,II and III stress intensity factors K I , K II , K III
respectively, so it is reasonable to describe the corresponding scalar multipliers in
the asymptotic contact problem as ‘generalized stress-intensity factors’.
As in the earlier sections of this chapter, these arguments can equally be applied to
three-dimensional contact problems, since the local fields are defined over a region
Fig. 10.7 Frictionless

P
indentation of an elastic
half-plane by a rigid flat and
rounded indenter
rigid
radius R
b
a elastic
10.5 Local Fields 209
that is small compared with the dimensions of the contacting bodies and the contact
area. In these cases, the stress-intensity factors may vary around the boundary of the
contact area, but the local patched-in field at any particular point will comprise a
plane strain state appropriate to the local values.
10.5.1 The Flat and Rounded Indenter
To introduce this technique, we consider the indentation of an elastic half-plane by a

frictionless rigid punch which is predominantly flat, but which has slightly rounded
corners, as shown in Fig. 10.7.
If the radius R b, we would expect the stress field to approximate that for a flat
punch of width 2b, everywhere except in the immediate vicinity of the corners. In
particular, the contact tractions will be
P b − |x|
p(x) ≈ √ if 1. (10.42)
π b2 − x 2 R
The traction distribution (10.42) is singular at x = ±b, so focussing on the region

near x = −b, we can define a further approximation
Kp
p(ξ) ≈ √ , (10.43)
ξ
where ξ = x +b is the distance from the point x = −b and
P
Kp = √ (10.44)
π 2b
is the generalized stress-intensity factor.2

Now if R b, the region influenced by the rounding, and in particular the extent
of contact
d =a−b (10.45)
on the rounded portion of the punch will be small compared with b, and hence, the
local perturbation in the stress fields due to rounding will be determined only by
the generalized stress-intensity factor K p . The local field can be determined either
by solving the problem of a flat semi-infinite indenter (b → ∞) with a radius R, or
by examining the local behaviour of any known solution (two-dimensional or three-
2 Definitions of stress-intensity factors for LEFM conventionally include a multiplying factor of

√
2π, but this is not done in Eq. (10.43) or elsewhere in this chapter, since several unconventional
asymptotic forms are involved, and the usual connection to energy release rates and path-independent
integrals has no parallel in contact problems. Also, notice that we have adopted the convention that
compressive tractions are positive.
dimensional) for a problem of a flat and rounded indenter. For this case, Sackfield
et al. (2003) obtained

3π R K p 2/3
d= (10.46)
2E ∗

√

3K p
1 + ζ − 1
p(ζ) = √ 2 ζ + 1 + ζ ln
√
, (10.47)
4 d 1 + ζ + 1
where 2/3
∗
ξ 2E
ζ= = (x + b). (10.48)
d 3π R K p
The contact traction distribution is shown in Fig. 10.8, the dashed line representing
the singular approximation of Eq. (10.43). Notice that the punch is flat in ζ > 0 and
rounded in ζ < 0.
The maximum contact pressure occurs in the rounded section and is
∗ 1/3
1.80K p K 2p E
pmax ≈ √ = 1.073 . (10.49)
d R
This permits us to determine the maximum contact pressure for a fairly general
three-dimensional flat and rounded indenter, given only the stress-intensity factor
K p for the much simpler problem where the indenter has sharp corners. This latter
problem is readily formulated and solved using a finite-selement model, from which
the stress-intensity factor can be extracted by an appropriately graded mesh at the
edge, or alternatively by using ‘crack-tip’ elements.
Fig. 10.8 Effect of rounding on the pressure distribution under a rigid flat frictionless indenter
10.5.2 Fretting in Non-conformal Contact
If a nominally stuck contact is subject to vibration, small localized regions of cyclic

slip can give rise to fretting fatigue failure, in which surface damage in the slip region
eventually leads to the initiation of a surface-breaking crack (Nowell et al. 2006).
If the contact is non-conformal, we know from Chap. 9 that slip zones will tend to
develop at the edges of the contact region, and if these are sufficiently small, the slip
solution will depend only on appropriate generalized stress-intensity factors in the
fully adhered solution. Since fretting fatigue is associated with frictional slip, this in
turn implies that experimental fretting data should correlate with these factors, and
this is indeed found to be the case (Hills et al. 2012a, b).
Here, we restrict attention to the case of similar materials, so the slip-separation
boundary is characterized by a square-root bounded normal traction distribution as in
Sect. 10.3.1, since β = 0 in Eq. (10.31)2 and the first (singular) eigenvalue is excluded
by the unilateral inequalities, as in Sect. 10.1.3.
Suppose that the normal force P and hence the extent A of the contact area are
constant. The contact pressure near a point on the boundary of A can be approximated
as
p(x) ≈ Bx 1/2 , (10.50)
where x is a coordinate measured into A along the local normal to the boundary, and B
is a constant that can easily be found (for example) by curve-fitting results from a finite
element solution. Suppose now that a periodic tangential force is applied, tending to
produce cyclic slip at the interface. This corresponds to back and forth motion along
a short vertical line in Fig. 9.3, and if the bodies can reasonably be modelled by half-
planes, the entire tangential traction distribution could be determined by following
the steps defined in Sect. 9.3.1.
However, if the slip zones are sufficiently small, their evolution is described more
simply in terms of asymptotic fields. We first solve the problem assuming complete
adhesion, obtaining a square-root singular distribution of tangential tractions that
can be approximated in the asymptotic form
q(x) ≈ K II x −1/2 (10.51)
for small values of x, where the mode II stress-intensity factor K II is proportional to

the applied tangential force Q. Taking the limit of any convenient Cattaneo–Mindlin
or Ciavarella-Jäger distribution3 for small values of Q, it can be shown that slip will
occur during the first tangential loading cycle over the range 0 < x < c, where
2K II
c= , (10.52)
fB
and the resulting tangential traction distribution near the corner is then
3 See for example, Dini and Hills (2004).

q(x) = f Bx 1/2 0<x <c

1/2
= f B x − (x − c) 1/2
x > c. (10.53)
As in Chap. 9, if the tangential force reaches a maximum value corresponding to

a stress-intensity factor K IImax and is then reversed, the contact will instantaneously
stick everywhere locking in the traction distribution (10.53), and a reverse slip zone
will then grow from the edge, giving a new distribution
q(x) = − f Bx 1/2 0<x <c

1/2
= − f B x − 2(x − c) 1/2
c < x < cmax (10.54)
1/2
= − f B x + (x − cmax ) − 2(x − c)
1/2 1/2
x > cmax ,
where
2K IImax (K IImax − K II )
cmax = ; c= . (10.55)
fB fB
If the steady-state cycle is defined by the range K IImin < K II < K IImax , the permanent
stick zone T comprises the points
(K IImax − K IImin )
x> . (10.56)
fB
For completely reversed loading, K IImin = −K IImax , T = x > cmax and the residual trac-
tions developed during each loading phase are completely erased during the next. In
all other cases, some residual stresses will be locked into T during the first cycle of
loading.
We have developed this argument under the assumption that the tangential trac-
tions are due to a tangential force Q, but the same results could be used if (for
example) one of the bodies were subjected to a bulk stress σx x parallel with the inter-
face. Once again, we solve the ‘full stick’ contact problem, obtain a time-varying
function K II (t), and then apply Eqs. (10.52), (10.53).
10.5.3 Edge Slip Zones with a Rigid Punch
In Sect. 7.3 we gave the solution for a rigid flat punch pressed into an elastic half-plane
under ‘no-slip’ conditions and showed that when β = 0, the local tractions exhibit an
oscillatory singularity at the edges of the contact area. Spence (1973) showed that
if a finite coefficient of friction is assumed, equal regions of slip develop at the two
edges under purely normal loading, and that these represent a small fraction of the
total contact area if β/ f < 0.4 [see Sect. 9.5.1].
Under these circumstances, it is clear that the region near the edges influenced
by frictional slip is embedded in a surrounding region dominated by the ‘full stick’
asymptotics, and hence that the extent of the slip regions and the local traction
distributions must be functions only of the multiplier on this asymptotic field — the
complex stress-intensity factor. Adams (2016) gives a detailed solution for the inner
stress and displacement fields in this case, using a Mellin transform representation.
Here, we simply determine the length of the slip zone, by using a stress-intensity
factor to relate the solution to that of Spence (1973) where Q x = 0.
If we define s = x −b as the (small) distance from the contact edge x = b, we can
approximate Eq. (7.30) as4
−1/2+ı
s (P + ı Q x )
p(s) + ıqx (s) = K where K = .
a−b π 1 − β 2 (a − b)
(10.57)
Following the same procedure as in Sect. 9.5.1, we then deduce that violations of the
friction law |qx (s)| ≤ f p(s) will occur in the range s < c, where

1 tan φ + f Qx
c = (a − b) exp arctan where tan φ = . (10.58)
f tan φ − 1 P
Since the surrounding asymptotic field completely determines conditions near the
edge [at least for modest values of Q x /P and β/ f < 0.4], we deduce that the actual
extent of the slip zone will be 4c exp(2), where is approximated by Eq. (9.65).
Equation (10.58)1 defines a monotonically decreasing function of Q x which
reduces to (9.70) when Q x = 0. Thus, the extent of the slip zone at x = b decreases if
the punch is pushed to the right, in which case x = b defines the trailing edge of the
contact area. Similar arguments applied to the leading edge x = a [or equivalently
symmetry arguments, with Q x < 0, which would make x = b the leading edge] show
that this zone increases with Q x . Of course, if the tangential force is sufficiently large
for the leading edge slip zone to be more than a small proportion of the total con-
tact area, the asymptotic argument fails and a full solution would then be necessary.
However, Adams (2016) shows that this method gives an acceptable approximation
at least for purely normal loading, even when the slip zones are of the order of 20%
of the punch width.
4 Notice that we express this result in a normalized form to avoid defining expressions, where a
quantity with dimensions of length is raised to a complex power.
10.5.4 Slip Zones in Conformal Contact
We saw in Sect. 10.4.1 that the state at the edge of the contact between a body with
a right-angle corner and a similar half-plane depends on the relative magnitudes of
the coefficient of friction f and the ratio R of Eq. (10.39).
Considerably, more insight into this system can be obtained by including the
second term in the asymptotic series and hence approximating the local asymptotic
solution for the monolithic (fully stuck) problem as
σθθ (r, θ) ≈ K I r λI −1 gθθ

I
(θ) + K II r λII −1 gθθ
II
(θ) (10.59)
σθr (r, θ) ≈ K I r λI −1 gθr
I
(θ) + K II r λII −1 gθr
II
(θ), (10.60)
where
λI = 0.5445; λII = 0.9081, (10.61)
K I , K II are the generalized stress-intensity factors and gθθ

k
(θ), gθr
k
(θ), k = I, II are the
corresponding eigenfunctions.
The two eigenfunctions are respectively symmetric and antisymmetric with
respect to the symmetry plane θ = π/4 in Fig. 10.6 so
π π
gθr
I
= 0; gθθ
II
= 0, (10.62)
4 4
and it is convenient to normalize the eigenfunctions such that
π π
gθθ
I
= 1; gθr
II
= 1. (10.63)
4 4
The tractions on the symmetry plane are then defined by
π π
σθθ r, = K I r λI −1 ; σθr r, = K II r λI −1 , (10.64)
4 4
permitting a convenient estimate of K I , K II to be made by constructing an appropriate
finite element solution of the monolithic problem and plotting the tractions on the
symmetry plane as functions of r on a log–log scale. Notice that, this demands that
the mesh local to the corner include the straight line θ = π/4, but a locally polar mesh
is optimal in such problems.
The normalized eigenfunctions are plotted in Fig. 10.9, and in particular, the values
on the contact plane θ = 0 are
gθθ
I
(0) = 0.7303; gθr
I
(0) = −0.3966; gθθ
II
(0) = 1.0873; gθr
II
(0) = 0.2380.
(10.65)
II
gθθ
I II
gθθ gθr
I
gθr
(degrees) (degrees)
Fig. 10.9 The normalized eigenfunctions gθθ

I (θ), g I (θ), g II (θ), g II (θ)
θr θθ θr
For the edge of the contact to remain in contact, we require σθθ (r, 0) < 0 and this
will be guaranteed for all r in (10.59) provided
K I < 0; K II < 0. (10.66)
The terms in the approximation (10.59), (10.60) each have dimensions of stress
and are individually self-similar, so a unique length scale can be defined as
1/(λII −λI )
KI
d0 = . (10.67)
K II
This enables us to write the contact tractions in the dimensionless form

σθθ (ρ, 0)
≈ ρλI −1 gθθ
I
(0) + ρλII −1 gθθ
II
(0) (10.68)
G0
σθr (ρ, 0)
≈ ρλI −1 gθr
I
(0) + ρλII −1 gθr
II
(0), (10.69)
G0
where ρ =r/d0 and

1/(λII −λI )
|K II |1−λI
G0 = (10.70)
|K I |1−λII
is a measure [with dimensions of stress!] of the combined stress intensity at the

corner (Flicek et al. 2013).
The traction ratio R can now be written
σθr (ρ, 0) g I (0) + ρλII −λI gθrII

(0)
R(ρ) = = − θr . (10.71)
−σθθ (ρ, 0) gθθ (0) + ρ
I λ II −λ I gθθ (0)
II
As in Sect. 10.4.1, R(0) = 0.543, so complete adhesion is expected if the coefficient

of friction f > 0.543. For lower values of f , we can determine a critical value ρ = c0
such that the frictional inequality is violated in 0 < ρ < c0 by writing R(c0 ) = f and
solving for c0 . Using the numerical values from Eqs. (10.61), (10.65), we obtain
2.75
0.3966 − 0.7303 f
c0 = . (10.72)
0.2380 + 1.0873 f
Of course, slip will then occur at least in 0 < ρ < c0 and the resulting redistribution
of stress leads us to expect that the actual slip zone will be larger than this. Churchman
and Hills (2006b) have solved the resulting frictional problem by distributing glide
dislocations over the as yet unknown slip zone 0 < ρ < c on the contact interface in
an otherwise monolithic three-quarter plane. By satisfying the frictional conditions
in the slip zone, and using the asymptotic condition from Sect. 10.3.2 to determine
c, they showed that at least in the range 0.25 < f < 0.543, c ≈ 2.4c0 .
The degree of generality of these results cannot be overstated. They apply to any
contact problem involving a body with a right-angle corner contacting a plane surface
of a similar material under any loading conditions, provided the forces are increased
in proportion — in other words, the load path follows a straight line from the origin
in an appropriate loading parameter space.5 In any such case, if we first solve the
corresponding monolithic [complete stick] problem using a numerical method, we
can extract the generalized stress-intensity factors from a logarithmic plot of tractions
along the interface, after which the length scale d0 , the length of the violation region
(c0 d0 ) and the length of the slip zone (2.4c0 d0 ) can be determined from Eqs. (10.67),
(10.72).
Once the slip zone is determined, the actual contact tractions are also of universal
form and can be scaled by the magnitude G 0 from Eq. (10.70). Detailed results are
given by Churchman and Hills (2006b), who also show that (i) the tractions outside
the slip zone are close to the corresponding power-law monolithic tractions and (ii)
those inside the slip zone are close to an appropriate scaling of the slip asymptotic
of Sect. 10.4.2.
Non-proportional Loading
The solution developed in Sect. 10.5.4 satisfies the frictional inequalities in terms
of the shift h rather than slip velocity ḣ, and is therefore a static solution in the
sense of Sect. 8.1.3. It applies to the corresponding rate problem provided that the
loading remains monotonic and proportional, meaning that the ratio K II /K I remains
constant, in which case the slip zone will remain the same size during loading.
The same results also apply to monotonic non-proportional loading provided that
the extent of the slip zone (2.4c0 d0 ) increases monotonically in time. Since c0 is
independent of K I , K II , it follows that this condition is satisfied if and only if d0 and
hence K II /K I are non-decreasing, from Eq. (10.67).
5 Theastute reader will notice a connection here to the conditions for Dundurs’ receding contact
arguments to apply, see Chap. 11.
If at some instant in the loading history the ratio K II /K I starts to decrease, we

would usually expect the contact to stick everywhere instantaneously, as for example
at the points E and B in Fig. 9.3. However, the resulting increment in contact tractions
has a leading eigenvalue λI = 0.5445 which implies a traction ratio increment of
R = 0.543. This field is more singular than the slip field on which it is superposed,
and the incremental traction ratio exceeds the coefficient of friction, so stick is not
possible in the corner. Also, since the increment ex hypothesi is opposite in sense
to that producing the original slip zone, the only possibility in this corner is the
development of a small separation zone and an adjacent region of backward slip.
Churchman and Hills (2006b) examined the consequences of these results for
the problem illustrated in Fig. 10.10, where an elastic rectangular block is pressed
into a half-plane of the same material by a monotonically increasing force P, after
which a cyclic tangential force of amplitude Q is applied. A cyclic moment M is
also applied as shown, in order to ensure that the effective line of action of Q always
passes along the interface. During the initial normal loading phase, ‘leading edge’
slip occurs at the corners (i.e. the block slides in the direction of the local corner) for
f < 0.543. When the tangential force is applied to the right, the slip zone increases
at the right edge, whilst stick with a small superposed separation and reverse slip
zone is developed at the left edge. When the tangential force changes direction, this
behaviour is reversed, with a significant leading edge slip zone developing at the left
edge and separation and slip at the right edge.
Churchman and Hills discuss their results with reference to the particular geome-
try of Fig. 10.10 and the applied forces P, Q, but it is clear from the arguments in this
chapter that as long as the slip and separation zones predicted are small compared with
the dimensions of the contacting bodies, the behaviour in each corner can be com-
pletely characterized in terms of the time history of the generalized stress-intensity
factors K I (t), K II (t). Thus, similar arguments can be applied to other geometries with
contacts defined by right-angle corners, provided these factors are first determined
from a finite element solution of the monolithic (fully adhered) problem.
Fig. 10.10 Rectangular P

block loaded against an M
elastically similar half-plane Q
with normal and tangential
loads
elastic
elastic
Problems
1. Define new variables s = a −r in Eq. (5.21)1 and t =r −a in Eq. (5.22) and then
expand these equations for small values of s, t respectively. Show that the leading
terms in these expansions have the form of Eq. (10.15), and find the corresponding
value of the constant B1 .
2. Extract the square-root singular term in Eq. (5.14) by (i) integrating by parts and
then (ii) performing the differentiation with respect to r . Hence show that the con-
stant B1 in Eq. (10.15)1 is proportional to h(a). Then show that Eq. (5.17) defines a
square-root bounded expression for displacement of the form (10.15)2 with the same
multiplier B1 .
3. Two similar electrically conducting bodies 0 < θ < π and −π < θ < 0 make perfect
electrical contact over the half-line θ = 0 and are separated [and hence also insulated]
over θ = ±π. Show that the local potential fields can be expressed as an eigenfunction
series of the form
∞
V (r, θ) = Bn r λn f n (θ)
n=1
and find the eigenvalues λ and eigenfunctions f n (θ). Hint: The governing equations
for electrical conduction are (4.13, 4.14) and in polar coordinates

∂ 1 ∂ ∂2 1 ∂ 1 ∂2
∇= , ; ∇2 = 2 + + 2 2.
∂r r ∂θ ∂r r ∂r r ∂θ
4. In Problem 3, suppose that in the contact region, there is a contact resistance R

such that the local current density
(V2 (r, 0) − V1 (r, 0))

i θ (r, 0) = ,
R
where V1 (r, θ), V2 (r, θ) are the electrical potentials in bodies 0 < θ < π and −π <
θ < 0 respectively. Show that a parameter L with the dimensions of length can be
constructed from R and the resistivity ρ, and hence that the asymptotic problem is
not now self-similar. What form would you expect the fields to take in the region
where r L?
5. Two similar thermally conducting bodies at different uniform temperatures are
brought into contact at time t = 0 over an area A. By focussing on a region very close
to the boundary of A, and using symmetry, we can describe the local temperature
field T (r, θ, t) in one of the bodies by the conditions
∂T
T (r, θ, 0) = 0; T (r, 0, t) = T0 ; (r, π, t) = 0,
∂θ
Problems 219
where T0 is a constant and r, θ are defined as in Fig. 10.3. The temperature field must
also satisfy the transient heat conduction equation
∂2 T 1 ∂T 1 ∂2 T 1 ∂T
+ + = ,
∂r 2 r ∂r r ∂θ
2 2 k ∂t
where k is the thermal diffusivity of the material.√ Show (i) that the solution for T
depends only on the dimensionless coordinates r/ kt, θ,√and (ii) that the heat flux
qθ (r, 0, t) in the contact region tends to a constant as r/ kt → ∞. Find the value
of this constant and describe in words the evolution of the heat flux qθ (r, 0, t) with
time.
6. Find the first eigenvalue for the problem of Fig. 10.5a with α = 60◦ , if the materials
of the two bodies are the same and no slip occurs at the interface. Hence find the
minimum coefficient of friction for the no-slip assumption to be reasonable.
7. Using the incremental formulation of Sect. 6.3.1, show that for symmetric non-
conformal problems with contact semi-width a, the constant B in Eq. (10.50) is given
by
F(a) 2
B= ,
π a
where F(a) is defined in Eq. (6.29). Hence show that under ‘Cattaneo’ loading, as
defined by Fig. 9.1, the length of each slip zone for sufficiently small ratios Q/P is
given by
Q dP
c= .
f da
8. A frictionless rigid punch has the form of a truncated wedge, as shown in Fig. 10.11.
Define a coordinate system with origin at the point x = b, and use an asymptotic
method to find the dominant singular term in the contact tractions at this point due
to the discontinuity of slope. Notice that the displacement boundary conditions will
then take the inhomogeneous form
∂u θ ∂u θ
(r, 0) = 0 and (r, π) = α,
∂r ∂r

α
b
a
so this is not an eigenvalue problem. Show that the multiplier on the singular term
depends only on α and is independent of the force P.
9. Use the method of Sect. 10.5.1 to estimate the relation between P and a in the
axisymmetric Problem 5.3, assuming R b.
10. A flat rigid frictionless elliptical punch of semi-axes a, b is pressed into an elastic
half-space by a normal force P. The edges of the indenter are slightly rounded, with
uniform radius R a, b. Use the solution of Sect. 2.3.2 and the asymptotic method of
Sect. 10.5.1 to estimate the magnitude and location of the maximum contact pressure.
11. A two-dimensional Hertzian contact between similar materials is loaded by a
normal force P which is then held constant whilst a bulk stress difference σ and
a tangential force Q x are applied in proportion. Assuming full-stick conditions,
determine the resulting mode II stress-intensity factors at the two edges and hence
estimate the sizes of the two slip zones using an asymptotic method. Check your
results against Eq. (9.37) for the case where the tangential force Q x = 0 and k 1.
Chapter 11
Receding Contact
Filon (1903) investigated the state of stress in an infinite elastic layer of thickness
2h subjected to equal and opposite concentrated forces on the two faces, as shown in
Fig. 11.1a. He showed that the normal traction on the plane of symmetry [the dashed
line in Fig. 11.1a] becomes tensile for |x| > 1.35 h, where x is measured from the
line of action of the forces. He deduced correctly that if a single layer of thickness h
were pressed against a frictionless rigid plane, the layer must lift off from the plane
at the edges, as shown in Fig. 11.1b.
Keer et al. (1972) solved the problem of Fig. 11.1b for the more general case
where both materials are elastic with possibly differing elastic properties, and the
normal force applied to the upper surface of the layer may have an arbitrary (but
symmetric) distribution. They represented the elastic fields in terms of harmonic
potential functions as in Sect. A.3 and used the Fourier transform method of Sect. 14.4
to satisfy the global boundary conditions (i.e. those that are satisfied for all x). If the
contact interface is identified as the plane y = 0, there remain the mixed boundary
conditions
σ yy (x, 0) = 0 |x| > c (11.1)
∂u (1)
y ∂u (2)
y
(x, 0) − (x, 0) = 0 − c < x < c, (11.2)
∂x ∂x
where x = ±c defines the unknown boundaries of the contact region and the super-
scripts 1, 2 refer to the layer and the substrate, respectively. These conditions lead
to a pair of dual integral equations that can be reduced to a single homogeneous
Fredholm equation, the eigenvalue of which defines the relation between α and c,
where α is one of Dundurs’ bimaterial constants.1 Keer et al. (1972) also formulated
the corresponding axisymmetric problem.
Appendix D. It is interesting to note that the contact area in this problem depends only on α,
1 See
whereas frictional problems for the half-plane generally depend only on β.

https://doi.org/10.1007/978-3-319-70939-0_11
222 11 Receding Contact
P P
(a) (b)
x h elastic
2h
rigid
P
Fig. 11.1 (a) An infinite layer loaded by equal and opposite concentrated forces, (b) A layer pressed
against a frictionless rigid plane by a concentrated force
If the substrate is rigid, as in Fig. 11.1b, the bimaterial constant α = 1 and the
numerical results in Fig. 2 of Keer et al. (1972) suggest that in this limit c ≈ 0.84 h.
For elastic substrates, a larger contact area is obtained, and full contact is of course
established at the opposite limit α = −1, where the layer is rigid (and hence moves
as a rigid body) and the substrate is elastic. Kauzlarich and Greenwood 2001 gave a
more extended treatment of these problems, including finite element and analytical
solutions, and showed that the extent of the contact region is always significantly
smaller than the range in which the tractions are compressive in the corresponding
monolithic problem such as that shown in Fig. 11.1a.
11.1 Characteristics of Receding Contact
The problem of Fig. 11.1b belongs to a class described by Dundurs and Stippes
(1970) as receding contact problems. These are characterized by the fact that the
extent of the contact area ΓC under load is included within the contact area Γ0 at zero
force—i.e. ΓC ∈ Γ0 . The special case where ΓC = Γ0 —i.e. where the contact area
remains unchanged during loading—was originally defined as ‘stationary contact’
by Dundurs and Stippes, but in later papers Dundurs abandoned this terminology,
since all the results for receding contacts also apply to stationary contacts. However,
it is rather misleading to refer to a contact area as ‘receding’ if in fact it remains
unchanged, so we shall use this term only in the strict sense in this book.
Contact problems are inherently nonlinear, since the extent of the contact area
is determined by the Signorini inequalities. However, Dundurs and Stippes showed
that when the contact is receding, the extent of the contact area is independent of
the magnitude of the applied force. In effect, the contact area ‘jumps’ to its reduced
value as soon as some infinitesimal force is applied, after which it remains unchanged
as the force increases. This contrasts with non-conformal contact (characterized by
Dundurs and Stippes as ‘advancing contact’), where the contact area is usually a
monotonically increasing function of the applied force. Furthermore, in receding
contact problems, since the contact area remains constant, the complete stress and
displacement fields are linearly proportional to the applied forces. For example, if
11.1 Characteristics of Receding Contact 223
the force P in Fig. 11.1b is increased, the gap in the separation region will increase
everywhere in proportion to P.
Proof
Suppose the complete solution to a particular receding contact problem is defined by
the stresses and displacement fields σ, u, respectively, when the external tractions
are given by t. In particular, the unilateral contact conditions imply that the relative
normal displacement (u1 −u2 )·n ≡ w(s) = 0 in the contact region ΓC , where n is the
local normal to the interface Γ0 . With the sign convention of Sect. 8.3, w(s) > 0 in
the separation region Γ0 −ΓC , where s is a coordinate defining points in Γ0 . Also,
the normal contact pressure p(s) > 0 in ΓC , and p(s) = 0 in Γ0 −ΓC .
Now it is clear that the equations of linear elasticity for the same bodies will
be satisfied by the fields λσ, λu where λ is any positive scalar multiplier, and that
these fields will correspond to a traction distribution λt, and to contact displacements
and pressures λw(s), λ p(s), respectively. Also, the new contact pressure λ p(s) will
satisfy the conditions λ p(s) > 0 in ΓC , and λ p(s) = 0 in Γ0 −ΓC and the new relative
normal displacement will satisfy λw(s) = 0 in ΓC , and λw(s) > 0 in the separation
region Γ0 −ΓC . It follows that the fields λσ, λu satisfy the governing equations and
the frictionless contact inequalities and hence they define the unique solution of the
problem for external tractions λt. Since λ can take any positive value, this establishes
(i) that the stresses are linear functions of the applied tractions and (ii) that the extent
of the contact region ΓC remains unchanged as λ changes.
Notice that this proof is restricted to the case of proportional loading, where the
applied tractions are of the form λt and only the scalar multiplier λ is allowed to
increase. In other words, the external tractions must follow a straight line from the
origin in a multidimensional traction space.
11.1.1 Examples of Receding Contact
Figure 11.2 shows an elastic cylindrical pin of radius R in a cylindrical hole of

the same radius in an infinite elastic matrix. The pin is loaded by a force P and
the ‘bilateral solution’ (i.e. the solution assuming contact at all points around the
circumference) will involve unacceptable tensile contact tractions for |θ| > π/2.
The unilateral solution was obtained by Persson (1964) for the case where the
pin and the matrix are of similar materials, and by Ciavarella and Decuzzi (2001b)
for dissimilar materials. For similar materials, contact occurs in an arc of semi-angle
ψ = 87.46o , and the contact pressure distribution is given by
√
P 2 b2 − y 2 1 b2 + 1 + b2 − y 2
p(y) = √ + ln √ ,
πR 1 + b2 (1 + y 2 ) 2b2 (b2 + 1) b2 + 1 − b2 − y 2
(11.3)
Fig. 11.2 An elastic

cylinder pressed against a
cylindrical hole in an elastic
matrix
ψ
P θ
R ψ
where
ψ θ
b = tan ; y = tan , (11.4)
2 2
(Ciavarella and Decuzzi 2001a).

For the more general case of dissimilar materials, the semi-angle ψ is defined
implicitly by the equation

F(b) ≡ (1 − α) ln(b2 + 1) + 2b4 + 4βb2 (b2 + 1) − 2 = 0, (11.5)
(Ciavarella and Decuzzi 2001b). A minimum value of ψ = 72.35o is obtained when

α = −1, β = 0, corresponding to the case of a rigid pin and an incompressible
matrix (ν2 = 0.5). At the opposite extreme, where the matrix is rigid and the pin is
incompressible, contact must occur throughout the interface, giving ψ = 180o , since
the volume of the pin material cannot change and it is just sufficient to fill the hole
in the matrix.
Figure 11.3 shows a related problem in which the pin is not loaded, but the matrix
is subjected to uniaxial tension σ0 at infinity, leading to separation near the loading
axis and contact at the two sides. This problem was solved by Noble and Hussain
(1969) by expanding the stress and displacement fields in Fourier series in the angle θ,
in which case the mixed boundary conditions lead to a pair of dual series equations
which can be solved numerically. A closed-form solution is possible in the case
where the cylinder and the matrix have the same material properties, in which case
the contact arc semi-angle ψ is determined from the equation

2 + sin2 ψ + 2(1 + sin2 ψ) ln(cos ψ)] E(cos ψ)
= 4(2 + ln(cos ψ)) sin2 ψK (cos ψ), (11.6)
11.1 Characteristics of Receding Contact 225
σ0 ψψ σ0
Fig. 11.3 An elastic cylinder in a cylindrical hole in an elastic matrix loaded in uniaxial tension
where the complete elliptic integrals K (·), E(·) are defined in Eqs. (2.32) and (3.24)
respectively.
The solution of Eq. (11.6) is ψ = 19.63o and the corresponding contact pressure
distribution is then given by

σ0
p(θ) = 3 cos θ cos2 θ − cos2 ψ
2

2 + 3 sin2 ψ cos θ + cos2 θ − cos2 ψ
+ ln (11.7)
2(2 + ln(cos ψ)) cos ψ
and is shown in Fig. 11.4.

A more general solution of this problem, including cases of biaxial remote stress
and different elastic materials, was developed by Keer et al. (1973), who give nu-
merical results for several such cases.
(degrees)
Fig. 11.4 Contact pressure distribution for the problem of Fig. 11.3 and Eq. (11.7)
11.2 Frictional Problems
Dundurs’ original discussion of receding contact was restricted to frictionless prob-

lems, where the only inequalities are those precluding tensile contact tractions and
interpenetration. However, it is clear that an exactly similar proof can be extended
to the Coulomb friction inequalities, provided that the external tractions increase
monotonically as well as being proportional.
The contact area ΓC must now be partitioned into a slip region Γ S and a stick
region ΓC −Γ S . In Γ S , the tangential tractions q(x, y) must satisfy the condition
f p(x, y) ḣ(x, y)
q(x, y) = , (11.8)
| ḣ(x, y)|
from Eq. (8.4), whilst in ΓC −Γ S ,
|q(x, y)| ≤ f p(x, y) and ḣ(x, y) = 0, (11.9)
from Eq. (8.5). If we now postulate that the slip and stick regions remain unchanged
during loading, the problem is linear, so that if tangential tractions q(x, y) and
contact pressures p(x, y) correspond to external tractions t and produce slip dis-
placements h(x, y), then external tractions λt will produce tangential tractions
λq(x, y), contact pressures λ p(x, y) and slip displacements λh(x, y). It follows
that ḣ(x, y) = λ̇h(x, y) and hence
ḣ(x, y) λ̇h(x, y)
= . (11.10)
| ḣ(x, y)| |λ̇||h(x, y)|
Now λ̇ = |λ̇| if and only if λ̇ > 0, in which case conditions (11.8 and 11.9) reduce to
f p(x, y)h(s)
q(x, y) = (x, y) ∈ Γ S (11.11)
|h(x, y)|
|q(x, y)| ≤ f p(x, y) and h(x, y) = 0 (x, y) ∈ (ΓC − Γ S ), (11.12)
which is a formal statement of the conditions for the static problem, as defined in
Sect. 8.1.3. This problem is independent of λ, so the same solution remains valid for
all values of λ, subject to the condition λ̇ > 0—i.e. that the loading is monotonic as
well as proportional.
11.2.1 Frictional Unloading
A striking example of the importance of monotonic loading in frictional problems,

albeit in a ‘stationary’ contact problem, is provided by Spence’s solution for the
11.2 Frictional Problems 227
Fig. 11.5 An elastic block p (t)

pressed against a rigid 0
frictional plane surface
2h
Elastic h
Rigid x
4h
indentation of an elastic half-plane by a flat rigid punch, discussed in Sect. 9.5.

During loading, the slip region remains constant and the solution is linear with the
applied normal force. However, in the unloading problem, a complex array of stick
and slip zones is developed, whose boundaries move during the unloading process
(Turner 1979).
(Ahn and Barber 2008) developed a numerical solution for the receding contact
problem illustrated in Fig. 11.5, in which a rectangular elastic block is pressed against
a rigid frictional surface by a time-varying but spatially uniform traction p0 (t). During
first loading, the two edges of the contact area separate, a central region remains
stuck and this is flanked by symmetric regions of slip in opposite directions. This
problem satisfies the receding contact conditions, so the extent of these regions
remains independent of p0 .
Figure 11.6 shows the division of the initial contact area 4 h into regions of stick,
slip and separation during unloading. As in Turner’s problem, a region of reverse
slip develops near the edges of the contact region, but slip continues in the same
sense in part of the original slip region, these various slip regions being separated
by regions of stick. Notice also that the separation region increases in size during
unloading, though the size of the gap at each point decreases, so that at zero force the
entire separation region closes through the remaining gap reaching zero everywhere
simultaneously.
Fig. 11.6 Regions of stick, 1

separation and slip during
unloading from p0max to zero stick stick stick
for the problem of Fig. 11.5.
The direction of slip in the p0
slip zones (shaded) is p0max
indicated by arrows
sep n sep n
0 x
-1 -0.5 0 2h 0.5 1
(a) 1 (b) 1
stick stick stick
p0 p0 stick
p0max p0max
n
sep n sep n sep n sep
0.33 0.33
x x
-1 -0.5 0 2h 0.5 1 -1 -0.5 0 2h 0.5 1
Fig. 11.7 Regions of stick, separation and slip during (a) unloading and (b) reloading in the steady
state between 0.33 p0max and p0max . The dashed line indicates a set of points that just reach the
incipient slip condition, but do not actually slip
The technologically more interesting case is that where the force oscillates peri-
odically without falling to zero. In this case, the contact tractions continue to evolve
through successive cycles, but eventually a steady state is achieved in which the
extent of the separation and slip regions depends only on the ratio between the mini-
mum and maximum values of p0 . Figure 11.7 shows the division of the contact area
during unloading (a) and reloading (b) for the case where the force oscillates between
0.33 p0max and p0max . The contact area increases during loading and decreases during
unloading, and the instantaneous contact area sticks completely at each force rever-
sal, with the growth in symmetrically disposed slip zones during each monotonic
loading phase. In addition, points lying on the dashed line in Fig. 11.7a just reach
the incipient slip condition, but do not actually slip. This line represents the vestigial
remains of a zone that slips only during unloading, and by an amount that decreases
geometrically with each successive cycle. The reader should notice a parallel here
with the asymptotic progression to the steady state discussed in Sect. 8.4.4.
11.3 Thermoelastic Problems
If the block in Fig. 11.5 is caused to slide over the contacting surface, frictional
heat will be generated at the interface. The resulting thermoelastic distortion will
modify the contact pressure distribution, and this may cause the contact to recede
even in cases where the isothermal solution involves stationary (full) contact. If the
instantaneous contact pressure distribution is p(x, t), the rate of heat generation is
f V p(x, t) and it follows that if the external loading is increased by some scalar
factor λ, the heating and, hence, the temperature field will be increased by the same
factor, and hence the homogeneous contact conditions will be satisfied by a linearly
scaled stress and temperature field. As in Sect. 11.1, the inequalities will then also
be satisfied provided λ > 0, showing that the extent of the contact region remains
unchanged in the scaled solution.
11.3 Thermoelastic Problems 229
Notice that the contact area A in such problems will generally vary with time,
since the temperature field satisfies the transient heat conduction equation from some
uniform temperature initial condition, conveniently taken as T (x, 0) = 0. However,
the function A(t) is independent of the scaling factor λ. The system passes through
the same sequence of contact configurations at the same times for all linearly scaled
magnitudes of the applied external forces. Of course, this also implies that the long-
time or steady-state contact configuration will be independent of λ.
We shall show in Chap. 17, Sect. 17.5 that simple steady states in thermoelastic
contact problems may be unstable. For example, Burton et al. (1973a) analysed a sim-
ple thermomechanical system approximating the geometry of a lip seal and showed
that at sufficiently high sliding speed, a uniform contact pressure distribution satis-
fies all the equations of thermoelasticity, but is unstable. In this case, an arbitrarily
small sinusoidal perturbation on the uniform contact pressure can grow until separa-
tion occurs, after which the system rapidly converges on a limit state with partially
contacting surfaces (Burton et al. 1973b; Burton and Nerlikar 1975; Barber 1976).
In view of the above arguments, it is clear that the stress, displacement and tem-
perature fields in this steady state will scale with the externally applied initial sealing
pressure p0 , and in particular, the gap in the separation regions will increase in pro-
portion with p0 . Now the presence of any gap is clearly undesirable in a seal, whose
sole purpose is to prevent the escape of fluid across a sliding interface. But this simple
result shows that the obvious recourse of increasing p0 in an attempt to improve the
seal performance will have exactly the opposite effect, since the gap will then be
increased in proportion, causing increased leakage!
11.4 Almost Conformal Contact Problems
If the radius of the pin in Fig. 11.2 is slightly smaller than that of the hole, initial
contact will occur only at the point θ = 0 and the problem is strictly non-conformal,
or in Dundurs’ terminology, ‘advancing’. However, if the misfit radius ΔR R, the
∗
contact area is a unique function of the dimensionless load factor P/E ΔR, through
the equation
P π(1 + α)(b2 + 1)b2
∗ =− (11.13)
E ΔR F(b)
(Ciavarella and Decuzzi 2001b), where F(b) is defined in Eq. (11.5). The load factor
can be made large either by increasing P or by reducing ΔR and in the limit where
ΔR = 0, we recover the receding contact solution. It follows that this also represents
the maximum extent of contact at high forces for a pin with an initial clearance. The
same argument allows us to conclude that if the pin is oversized, so that there is an
initial interference fit (ΔR < 0), the contact area will be reduced under load, again
approaching the receding contact value in the limit of large force.
∗
The relation between P/E ΔR and the semi-arc ψ is shown in Fig. 11.8 for the
case of similar materials. Notice that the load factor tends to a finite negative value
Fig. 11.8 Contact semi-arc ψ as a function of load factor P/E ∗ΔR for a pin with clearance (ΔR > 0)
or interference (ΔR < 0) and a lateral force P. Both curves are asymptotic to the vertical line at
ψ = 87.46◦ corresponding to the receding contact solution
at full contact (ψ = 180o ), because the interference fit causes a uniform compressive
∗
traction when P = 0 and a finite value of P = π E ΔR/2 must be applied before this
is reduced to zero at the point θ = 180o .
The contact pressure distribution is again given by Eq. (11.3) for the case of similar
materials, with b being determined from Eq. (11.13). Results for some representative
∗
values of dimensionless clearance Φ = E ΔR/P are given in Fig. 11.9.
A wide range of other pin-in-hole problems, including frictional tractions and the
combined effect of pin loading and remote stress are discussed by Sundaram and
Farris (2010a, b).
Other ‘almost conformal’ contact problems exhibit similar behaviour, notably
that the receding contact solution defines the asymptotic limit as the applied forces
are increased, and that the approach to this limit can be described in terms of an
appropriate load factor (Ciavarella et al. 2006a). For example, Tsai et al. (1972)
considered the problem where the elastic layer in Fig. 11.1b has a slightly curved
bottom surface, so that initial contact is restricted to a point immediately below the
force. As the force P is increased, the contact area grows, but it tends asymptotically
to the receding contact limit c ≈ 0.84 h.
Problems
1. An infinite uniform straight beam of flexural rigidity E I and negligible weight

rests on a unilateral elastic foundation2 of modulus k. In other words, the beam is not
bonded to the foundation, which can therefore transmit compressive tractions only.
2 Governing equations for a beam on an elastic foundation are given in Sect. 14.1.2.
Problems 231
(degrees)
Fig. 11.9 Contact pressure distribution p(θ) for various values of the dimensionless clearance
(interference), Φ = E ∗ΔR/P. The receding contact limit corresponds to the case Φ = 0. Full
contact occurs for Φ < −2/π ≈ −0.64
A force F is now applied to the beam as shown in Fig. 11.10, causing contact in the
region −a < z < a and separation elsewhere.
Show that a is independent of F and find its value.
2. A rigid non-conducting half-plane slides over an elastic conducting half-plane at
speed V . The normal contact force is P and the coefficient of friction is f . The work
done against friction is converted to heat which is conducted into the conducting
body. By considering the dimensions of the governing parameters, show that the
semi-width a of the contact region in the steady state must take the form
CK
a= ,
αfVE
where K , α, E are, respectively, the thermal conductivity, thermal expansion coef-

ficient and elastic modulus of the deformable half-plane, and C is a dimensionless
constant that depends only on Poisson’s ratio.
3. Suppose the non-conducting half-plane in Problem 2 is replaced by a slightly
rounded body, leading to a quasi-Hertzian contact problem. How would you expect
the semi-width of the contact area to vary with the normal force P. In particular,
what happens when P is very large?
4. A steel pin of diameter 20 mm fits in a hole in a steel component with a diametral
clearance of 0.1 ± 0.02 mm, depending on manufacturing tolerances. If the axial
length of the contact is 50 mm and the pin transmits a radial force of 100 kN, estimate
Fig. 11.10 A beam resting

on an elastic foundation and F
subjected to a concentrated a a
force
M0
x
Fig. 11.11 A square block supported by Winkler foundations and subjected to a moment
the maximum contact pressure. How sensitive is your result to variability within the
tolerance bands? Would an estimate based on zero clearance be sufficient for design
purposes? [for steel, E = 210 GPa, ν = 0.3.]
5. A cylindrical pin is an interference fit in a hole in a body of the same material.
Separation just starts when the force P = P1 and exactly half of the interface is
separated (ψ = π/2) when P = P2 . Show that
P2 4
=
P1 ln(2)
and find the corresponding ratio as a function of α, β for the case where the materials
are dissimilar.
6. Figure 11.11 shows a rigid square block of side a clamped between two elastic
layers. Each of these layers can be approximated as Winkler foundations of modulus
k—i.e. the contact pressure p(x) = ku(x), where u(x) is the local indentation.
Initially, the layers are pressed together until the contact pressure is uniform and
equal to p0 . They are then held in this position, whilst a moment M0 per unit length
[into the paper] is applied to the block as shown in the figure.
(i) Show that if p0 = 0 implying that the block just fits between the undeformed
layers, the length of the contact region b on each surface is the same for all
positive values of M0 .
(ii) For p0 > 0, show that b depends only on the ratio M0 / p0 and that it tends to the
value obtained in part (i) as this ratio increases.
Chapter 12
Adhesive Forces
At very small length scales, the forces between molecules (e.g. van der Waals forces)
become significant and must be taken into account in the solution of contact problems.
These are most conveniently described by a scalar force potential V defined such
that the force acting on a given molecule is
F = −∇V. (12.1)
The force between two isolated molecules is commonly approximated using the
Lennard-Jones potential
r 6
r 12
m m
V =C −2 (12.2)
R R
Jones (1924), where R is the distance between two molecules, rm is a length scale,
and C is a material property. The attractive force between the molecules is then

∂V 12C rm 6 rm 12
F = −FR = = − . (12.3)
∂R R R R
Notice that the force contains two components — a repulsive force decaying with
R −13 and an attractive force decaying more slowly, with R −7 . We also see that FR = 0
when R =rm , so rm defines the equilibrium distance between two isolated molecules.
This is not the same as the equilibrium spacing in a regular crystalline lattice, since in
that case, the forces exerted by other surrounding molecules affects the equilibrium
configuration.
The form of the potential and the intermolecular force is illustrated in Fig. 12.1. In
this figure, the dotted line represents just the attractive [R −7 ] term in Eq. (12.3) and
shows that the interatomic forces are adequately approximated by this term alone
when R > 1.4rm .
These results can be used to define a continuum description of a solid by assum-
ing that there exist N equally spaced molecules per unit volume in the solid. The
https://doi.org/10.1007/978-3-319-70939-0_12
234 12 Adhesive Forces
Fig. 12.1 The

Lennard-Jones potential V
and the corresponding
intermolecular force FR
(repulsion positive). The
dotted line represents just the
attractive term in Eq. (12.3)
probability Φ of there being a molecule in a given differential volume dΩ is then

Φ = N dΩ and the potential V (x, y, z) due to the complete body Ω can be obtained
[in an averaged sense] by integration as
12 r 6
rm m
V (x, y, z) = C N −2 dξdηdζ, (12.4)
Ω R R
where
R= (x − ξ)2 + (y − η)2 + (z − ζ)2 (12.5)
is the distance from the point (ξ, η, ζ) to (x, y, z).

In particular, if the body Ω comprises the half-space z > h, the potential at the
origin, a distance h above the surface of the half-space, is
∞ ∞
rm12 2rm6
V (h) = 2πC N − r dr dz
h 0 (r 2 + z 2 )6 (r 2 + z 2 )3
12
rm rm6
= 2πC N − , (12.6)
90h 9 6h 3
and a single molecule at height h above the surface is attracted to the half-space by
a force 6
d V (h) rm rm12
F(h) = = 2πC N − . (12.7)
dh 2h 4 10h 10
Now consider the case where two half-spaces of the same material are separated
by a distance g. Each molecule of one body experiences a force determined by its
distance from the surface of the other half-space, so the attractive force per unit area
will be ∞ 6
r r 12
σ(g) = N F(h)dh = 2πC N 2 m3 − m 9 . (12.8)
g 6g 90g
12 Adhesive Forces 235
The equilibrium spacing ε between the two half-spaces can be obtained by setting
σ(ε) = 0, giving1
ε = 15−1/6 rm ≈ 0.64rm , (12.9)
and it is convenient to use this result to rewrite Eq. (12.8) in the form

ε3 ε9
σ(g) = 5πC N ε 2 3
− 9 . (12.10)
g3 g
Suppose the two half-spaces are at the equilibrium spacing and we now apply
a force sufficient to pull them apart, thus creating two new surfaces. We define the
interface energy Δγ as the work that must be done to separate a unit area of interface,
and hence ∞
15πC N 2 ε4
Δγ = σ(g)dg = , (12.11)
ε 8
using (12.10). We use this result to eliminate the constant C in (12.10), giving

8Δγ ε3 ε9
σ(g) = − 9 . (12.12)
3ε g3 g
For similar materials Δγ = 2γ, where γ is the surface energy of the material. For the
more general case of dissimilar materials, Eq. (12.12) still applies, but with Δγ =
γ1+γ2−γ12 where γ1 , γ2 are the surface energies of the two materials, and γ12 (> 0) is
an interface energy associated with modifications to the atomic array at the interface
due to differences in interatomic spacings between the two materials.
The maximum tensile traction σ(g) [the theoretical tensile strength] occurs when
∂σ/∂g = 0. Differentiating (12.12) with respect to g and equating the result to zero,
we find that the maximum occurs at g = 31/6 ε and is
BΔγ 16
σ0 = where B = √ ≈ 1.0264. (12.13)
ε 9 3
This result can be used to express the force law (12.12) in the alternative form
√
3 3σ0 ε3 ε9
σ(g) = − . (12.14)
2 g3 g9
1 We should not read too much into this result, since the molecules have been smeared out to generate
continuous half-spaces.
12.1 Adhesion Between Rigid Bodies
Suppose a rigid sphere of radius R ε is placed near a rigid half-space such that
the point of closest approach corresponds to a gap g = h. Taking this point as the
origin of a system of cylindrical polar coordinates, we can define an infinitesimal
annulus (r, r+dr ) on the sphere surface of area d A = 2πr dr , where the separation is
g = h+r 2 /2R. The force exerted on this area is σ(g)d A, so summing the contributions
of all such annuli, we obtain a total attractive force
∞
F = 2π σ(h + r 2 /2R)r dr. (12.15)
0
Changing the variable of integration to g = h +r 2 /2R, so r dr = Rdg, we obtain

∞
F = 2π σ(g)Rdg. (12.16)
h
It is clear that this integral will achieve its maximum value if the range includes all
values of g for which σ(g) > 0 and none for which σ(g) < 0. Assuming that the
force law is continuous, this implies that σ(h) = 0 and hence that h represents the
equilibrium spacing ε. The maximum value of F [the ‘pull-off’ force] is then given
by ∞
Fmax = 2π σ(g)Rdg = 2πΔγ R, (12.17)
ε
from the definition (12.11) of interface energy Δγ.

This result was first obtained by Bradley (1932). It is notable that the attractive
force depends only on the interface energy Δγ and not on the specific form of the force
law, so the same result would be obtained for any force law. This is a consequence of
the fact that for a sphere [or more strictly for a paraboloid], the amount of interface
area with a given spacing range (g, g +dg) is the same for all g. More generally, it
applies for any contact problem for which the initial gap function g0 (r, θ) has the
form
g0 (r, θ) = r 2 f (θ), (12.18)
in other words, where the gap increases quadratically with distance from the point
of closest approach.
In all other cases, we must anticipate some dependence on the exact force law as
well as the interface energy, and the pull-off force will not necessarily correspond to
the condition where the point of closest approach is at the equilibrium spacing.
Example: A Rigid Cone
Consider the case of a rigid cone of angle π/2 −α 1, so that the gap is defined
approximately by
g = h + g0 (r ) = h + r α.
12.1 Adhesion Between Rigid Bodies 237
The attractive force is then obtained as

∞
2π ∞
F = 2π σ(r α + h)r dr = 2 σ(g)(g − h)dg.
0 α h
For the special case of the Lennard-Jones law (12.12), we obtain

8πΔγε ε ε7
F= − ,
3α2 h 28h 7
and the maximum occurs when

√
ε 16 3 2πΔγε
h=√ and is Fmax = .
3
2 7α2
For this example, the attractive force at h = ε is 11% lower than the maximum.
12.2 The JKR Theory
If a crack in an elastic solid is loaded in tension — i.e. in a way tending to open the
crack — the theoretical elastic stress field is square-root singular along the ligament
ending at the crack tip. The local fields can be analyzed by the asymptotic methods
described in Chap. 10 and are characterized by a stress intensity factor
√
K I = lim σ yy (x, 0) 2πx, (12.19)
x→0
where the crack is assumed to occupy the half-line x < 0, y = 0. The basis of Linear
Elastic Fracture Mechanics (LEFM) is the thesis that crack extension leading to
fracture must occur when K I = K Ic , where K Ic is a material failure parameter known
as the fracture toughness.
An alternative statement of this criterion, dating back to pioneering work of Grif-
fith (1921) is that crack extension will occur when it is energetically favourable —
i.e. when it would lead to a reduction in the total energy comprised of elastic strain
energy, potential energy of applied forces, and surface energy. In this context, we
can define the energy release rate G as the reduction in elastic strain energy per unit
area of crack extension. Asymptotic analysis then yields the conclusion that for a
crack in a homogeneous material, G and K I are related by the equation
K I2 (1 − ν 2 )
G= . (12.20)
E
The celebrated JKR theory of adhesive contact Johnson et al. (1971) is essentially
an application of Griffith’s theory to the contact problem. Since the materials of the
two contacting bodies might differ, we define G in terms of the composite modulus
∗
E as
K2
G = I∗ , (12.21)
2E
and argue that the separation region (equivalent to a crack) will extend until

G = Δγ implying KI = 2E ∗ Δγ. (12.22)
In other words, in contrast to the ‘smooth’ asymptotic contact traction field defined
in Sect. 10.1.3, there will be a square-root singular local tensile contact traction at
the edge of the contact region, and the multiplier on this term depends only on the
material properties.
12.2.1 Axisymmetric Problems
For axisymmetric problems, the contact area will generally be a circle, whose radius
a is determined by the conditions (12.19), (12.22). This is conveniently achieved by
superposing (1) the contact pressure distribution in the corresponding non-conformal
contact problem without adhesive forces, but with a contact area of radius a, and (2)
the distribution under a flat rigid cylindrical punch of radius a. The first term satisfies
the contact condition u z (r ) = Δ − g0 (r ) in 0 ≤ r < a and the second term merely
modifies the rigid-body displacement term Δ.
The flat punch solution (2) is given in Sect. 5.1.1 as
P2 P2
p2 (r ) = √ ; Δ2 = . (12.23)
2πa a − r
2 2 2E ∗ a
Writing x = a −r, r = a −x and considering the case where x a, we can write
P2 P2
p2 (x) = √ ≈ √ , (12.24)
2πa x(2a − x) 2πa 2ax
so (12.19), (12.22) will be satisfied if we choose

P2 = − 8π E ∗ a 3 Δγ, (12.25)
implying
2E ∗ aΔγ 2πaΔγ
p2 (r ) = − ; Δ2 = − . (12.26)
π(a 2 − r 2 ) E∗
12.2 The JKR Theory 239
The negative sign in these expressions follows from the fact that the condition at the
edge of the contact region involves tensile square-root singular tractions, whereas
the conventional flat punch solution involves compressive tractions.
12.2.2 Indentation by a Sphere
For indentation by a sphere of radius R, the solution without adhesive forces is given
in Sect. 5.2 as
∗√ ∗
2E a 2 − r 2 a2 4E a 3
p1 (r ) = ; Δ1 = ; P1 = , (12.27)
πR R 3R
so including the effect of adhesion, we obtain

∗√ 2
2E a − r2 2E ∗ aΔγ
p(r ) = p1 (r ) + p2 (r ) = − (12.28)
πR π(a 2 − r 2 )
∗ 3
4E a
P = P1 + P2 = − 8π E ∗ a 3 Δγ (12.29)
3R
a2 2πaΔγ
Δ = Δ1 + Δ2 = − . (12.30)
R E∗
The relation between P, Δ and the contact radius a can be simplified by defining the
dimensionless parameters
P β2Δ βa
P̂ = ; Δ̂ = ; â = (12.31)
π RΔγ R R
Maugis (2000), where2

∗ 1/3
E R
β= (12.32)
Δγ
is also dimensionless. We then obtain
4â 3 4â 3/2 √

P̂ = − √ ; Δ̂ = â 2 − 2π â. (12.33)
3π 2π
The relation between P̂ and Δ̂ is plotted parametrically (with â as parameter) in

Fig. 12.2. The maximum tensile force occurs when
√ 1/3
d P̂ 4â 2 6 â 9π
= −√ =0 or â = . (12.34)
d â π 2π 8
2 Here we have slightly modified Maugis’ original notation by omitting some numerical factors.
Fig. 12.2 Relation between

dimensionless compressive
force P̂ and dimensionless
rigid-body indentation Δ̂ for
the JKR solution
This corresponds to the point B in Fig. 12.2, where
3
P̂ ≡ P̂0 = − . (12.35)
2
It follows that in order to separate the sphere from the half-space, we need to
apply a tensile force
3π RΔγ
F = − P̂0 π RΔγ = , (12.36)
2
and the radius of the contact circle at this instant is
1/3
R â 9π R 2 Δγ
a= = , (12.37)
β 8E ∗
from Eqs. (12.31), (12.32), (12.34).

The minimum value of rigid-body displacement (maximum negative value) occurs
when √ π 1/3
d Δ̂ 2π
= 2â − √ = 0 or â = (12.38)
d â 2 â 8
and is
3π 2/3
Δ̂ A = − . (12.39)
4
This corresponds to the point A in Fig. 12.2.
12.2.3 Energetic Considerations and Stability
If the sphere is pressed into the plane by a prescribed force P, points to the left of
B in Fig. 12.2 are unstable. However, if instead the rigid-body displacement Δ is
prescribed, stability is retained until the point A, but points between A and the origin
correspond to unstable equilibria in either scenario. To prove these results we need
to consider the total potential energy of the system comprising elastic strain energy,
surface energy, and [when P is prescribed] potential energy of the applied force.
The elastic strain energy can be determined by devising a loading scenario leading
to the required final state and calculating the work done during loading. A suitable
scenario is suggested by the superposition in Sect. 12.2.1 comprising the two steps
(i) Hertzian loading without adhesive forces until the contact area is a circle of the
given radius a, followed by (ii) rigid-body displacement at constant radius a until
the required stress intensity factor is achieved at the contact edge.
During the Hertzian contact phase,
∗√
4E RΔ3/2
P= (12.40)
3
from Eq. (12.27), so the work done during loading is
Δ1 ∗√ 5/2
8E RΔ1
U1 = P(Δ)dΔ = , (12.41)
0 15
where ∗
a2 8E a 5
Δ1 = so U1 = . (12.42)
R 15R 2
During the second, rigid-body displacement phase, the force is given by
∗
P(s) = P1 + 2E a(s − Δ1 ) (12.43)
from (12.23), where s is the total rigid-body displacement, so the work done is
Δ
∗
U2 = P(s)ds = P1 (Δ − Δ1 ) + E a(Δ − Δ1 )2 (12.44)
Δ1
Substituting for Δ1 and simplifying, we then obtain the total elastic strain energy as
∗ ∗
E a5 2E a 3 Δ ∗
U = U1 + U2 = 2
− + E aΔ2 . (12.45)
5R 3R
Displacement Control
The surface energy is −πa 2 Δγ, so if the displacement Δ is prescribed, the total
potential energy is
Π = U − πa 2 Δγ. (12.46)
The equilibrium position is then determined by the condition

∗ ∗
∂Π E a4 2E a 2 Δ ∗
= − + E Δ2 − 2πaΔγ = 0, (12.47)
∂a R 2 R
leading as before to (12.30). However, we can now examine the stability of the
equilibrium state by computing the second derivative
∗ ∗
∂2Π 4E a 3 4E aΔ 4
= − − 2πΔγ = 2πa 3 Δγ E ∗ − 2πΔγ, (12.48)
∂a 2 R2 R R
after substituting for Δ from Eq. (12.30). For this to be positive, indicating stability,
we require that
π 1/3
â > , (12.49)
8
which excludes points between the origin and A in Fig. 12.2.
Force Control
A similar calculation can be performed if the indenting force P is prescribed. In this
case, the rigid-body displacement Δ is a dependent variable given by
P − P1 a2 P
Δ = Δ1 + ∗ = + . (12.50)
2E a 3R 2E ∗ a
The total potential energy must now include a term −PΔ for the external force P,
so
Π = U − πa 2 Δγ − PΔ. (12.51)
Substituting for U from (12.45) and then for Δ from (12.50), we obtain
∗
4E a 5 Pa 2 P2
Π= − − − πa 2 Δγ. (12.52)
45R 2 3R 4E ∗ a
The condition
∗
∂Π 4E a 4 2Pa P2
= − + − 2πaΔγ = 0 (12.53)
∂a 9R 2 3R 4E ∗ a 2
Fig. 12.3 Hysteresis loop

during loading and unloading
reduces to (12.29) as before, and the second derivative ∂ 2 Π/∂a 2 can be shown to be
positive if and only if
1/3
9π
â > , (12.54)
8
which defines points to the right of B in Fig. 12.2.
12.2.4 Hysteretic Energy Dissipation
If the sphere is initially not in contact, but is gradually brought closer to the half-space
under displacement control, there will be no interaction3 until Δ = 0, at which point
there will be an unstable transition [‘jump-in’] to contact corresponding to the line
OC in Fig. 12.3. Further loading then follows the unique curve to the right of C.
If the sphere is now moved away from the half-space reducing Δ, contact will
be retained until we reach the point A, at which there will be an unstable transition
[‘jump-out’] to the separated state at D in Fig. 12.3. The complete contact–separation
cycle therefore involves a hysteretic loss of energy equal to the area D OC B AD, and
since Π = 0 at D and O, this is given by Π A−ΠC , where Π is defined by Eq. (12.46)
or (12.52). Jump-in and jump-out are unstable dynamic processes, so the energy is
mostly dissipated in the form of elastodynamic waves.
12.2.5 JKR Solution for More General Axisymmetric Bodies
The JKR solution is readily extended to cases where the indenting body is of fairly
general axisymmetric shape, subject only to the condition that the resulting contact
3 We shall show in Sect. 12.4.1 that in practical cases, jump into contact generally occurs before we
reach Δ = 0. The JKR theory therefore overestimates hysteresis losses.

area is a circle, rather than one or more annuli. We use the superposition defined in
Sect. 12.2.2, except that the expressions for Δ1 , P1 in Eq. (12.27) are replaced by

a
g (r )dr ∗
a
r 2 g0 (r )dr
Δ1 = a √0 ; P1 = 2E √ (12.55)
0 a2 − r 2 0 a2 − r 2
from Sect. 5.2, Eqs. (5.27) and (5.28).

Example: The General Power-law Punch
Suppose the indenting body is defined by the power-law gap function
g0 (r ) = Cr n ; g0 (r ) = Cnr n−1 , (12.56)
where C, n are constants. Substituting into (12.55), we obtain

∗
Δ1 = Cna n In−1 ; P1 = 2E Cna n+1 In+1 , (12.57)
where the dimensionless definite integral

1
ρk dρ (k − 1)!!
Ik = = k odd (12.58)
0 1−ρ 2 k!!
π (k − 1)!!
= k even (12.59)
2 k!!
and we recall (2m)!! = 2.4.6...(2m), (2m −1)!! = 1.3.5...(2m −1), where m is an

integer.
The total force is

∗
P = P1 + P2 = 2E Cna n+1 In+1 − 8π E ∗ a 3 Δγ (12.60)
and the pull off force corresponds to the contact radius a at which

dP ∗
= 2E Cn(n + 1)a n In+1 − 3 2π E ∗ aΔγ = 0 (12.61)
da
or 1/(2n−1)
9πΔγ
a= ∗ 2 2 . (12.62)
2E C n (n + 1)2 In+1
2
Substituting back into (12.60), we obtain the pull-off force as

∗ (n+1) 1/(2n−1)
2(2n − 1) E (n−2) 9πΔγ
F= . (12.63)
3 (Cn In+1 )3 2(n + 1)2
It is easily verified that this reduces to (12.36) when n = 2 and C = 1/2R. For
other values of n, the pull-off force depends on the elastic modulus, increasing with
modulus for n > 2 and reducing with modulus for n < 2.
12.2.6 Guduru’s Problem
Guduru (2007) used the JKR methodology to explore the adhesion problem for a
spherical indenter with a superposed axisymmetric waviness defined by the gap
function
r2
g0 (r ) = + A [1 − cos (ωr )] , (12.64)
2R
where the amplitude A is chosen to be sufficiently small to ensure that the contact
area comprises a single circle of radius a for all values of indentation depth Δ.
Substituting (12.64) into Eq. (12.55) and evaluating the integrals, we obtain
a2 π Aωa
Δ1 = + H0 (ωa) (12.65)
R 2
3
∗ 4a
P1 = E + π Aωa H0 (ωa) − π Aa H1 (ωa) ,
2
(12.66)
3R
where 1
2x n
Hn (x) = (1 − t 2 )n−1/2 sin(xt)dt (12.67)
π(2n − 1)!! 0
is the Struve function of integer order n.

Adding the adhesive terms Δ2 , P2 from Eqs. (12.25), (12.26) and extending the
dimensionless notation (12.31) by defining
∗
β2 A Rω βa P β2Δ E R
Â = ; ω̂ = ; â = ; P̂ = ; Δ̂ = ; β3 = ,
R β R π RΔγ R Δγ
(12.68)
we obtain
√ π Âω̂ â
Δ̂ = â 2 − 2π â + H0 (ω̂ â) (12.69)
2
4â 3 4â 3/2
P̂ = − √ + Ââ (ω̂ â)H0 ω̂ â − H1 ω̂ â . (12.70)
3π 2π
Notice that the results are now characterized by two parameters ω̂, Â representing
dimensionless wavenumber and amplitude respectively.
Fig. 12.4 Force-displacement relation for Guduru’s problem with ω̂ = 20 and Â = 0.04. The JKR
curve without waviness is shown dotted
The resulting force-displacement relation is shown in Fig. 12.4 for the case where
ω̂ = 20 and Â = 0.04. As we might expect, the curve oscillates about the JKR-curve
of Fig. 12.2, which is here shown dotted. This has two significant consequences:-
1. The maximum tensile force is greater than that for a smooth surface, showing that
this type of waviness increases the pull-off force.
2. The loading and unloading curve will in practice exhibit a series of small hysteresis
loops, one of which is indicated by the arrows near B in Fig. 12.4. This behaviour
was confirmed experimentally by Guduru and Bull (2007) for the indentation
of a gelatin block. It implies that the contact will exhibit ‘toughness’ meaning
that energy is dissipated in small events before complete separation occurs. By
contrast, the energy dissipation for the smooth sphere calculated in Sect. 12.2.4
occurs only at the transitions from contact to separation and vice versa.
Kesari and Lew (2011) established upper and lower envelopes for oscillatory
curves such as those in Fig. 12.4, and the area between these envelopes defines an
upper bound for the hysteretic energy loss during one loading/unloading cycle. They
also used the results to give a qualitative explanation of the dependence of hysteresis
on surface roughness in AFM (Atomic Force Microscope) experiments Kesari et al.
(2010).
12.3 The Tabor Parameter
The pull-off force (12.36) predicted by the JKR theory is independent of the elastic
∗
modulus E , so we might expect it to apply to the case of rigid spheres. However,
Bradley’s solution (12.17) to this problem involves no approximations and differs
12.3 The Tabor Parameter 247
from the JKR value by a factor of 4/3. The explanation of this apparent contradiction
is that the JKR theory neglects any interaction between the bodies outside the contact
∗
circle and this becomes progressively more important as E increases, since the radius
a then gets smaller, as we see from Eq. (12.37).
Derjaguin et al. (1975) developed an alternative approach [usually known as the
DMT theory] specifically focussing on these forces. Starting from the Hertzian solu-
tion without adhesion of Sect. 5.2, they first computed the size of the gap in the region
r > a. The elastic displacement u z (r ) in this region can be found by substituting h(t)
from Eq. (5.31) into (5.17), giving
(2a 2 − r 2 ) a a √r 2 − a 2
u z (r ) = arcsin + . (12.71)
πR r πR
If we assume that the contact circle is at the equilibrium spacing g = ε, the gap in the
surrounding region r > a will be
r2
g(r ) = ε + − Δ + u z (r )
2R
√
a r 2 − a2 (2a 2 − r 2 ) a
= ε+ − arccos , (12.72)
πR πR r
after using (5.30) for the rigid-body displacement Δ. The tensile force contributed
by this region alone is then obtained as
∞
F1 = 2π σ(g)r dr, (12.73)
a
where the function σ(g) is defined by (12.12), after which the total indenting force
is obtained as the algebraic sum P = PH − F1 , where PH is the Hertzian force from
Eq. (5.30) or (12.27)3 . Derjaguin et al. (1975) showed that under these assumptions
the maximum tensile force actually occurs when a = 0 — i.e. when there is only a
single point at the closest approach ε. The theory therefore predicts the same pull-off
force as Bradley’s rigid-body calculation.
The DMT theory is also approximate, since it neglects the elastic deformation
associated with the tensile tractions σ(g), which will modify the traction distribu-
tions both inside and outside the Hertzian contact area. In fact, Greenwood (2007)
shows that the results of the DMT theory are consistently less accurate than those
when the contacting bodies are assumed to be rigid. He proposes an alternative
‘semi-rigid’ theory in which the contact tractions are calculated based on the shapes
of the undeformed bodies, and the separation Δ is then modified using the elas-
tic deformations due to these tractions. A similar approach was used by Song and
Komvopoulos (2014), but using a finite element solution to calculate the resulting
elastic displacements.
Tabor (1977) argued that the DMT and JKR theories define two limiting cases of
a more general problem, which we characterize by the ‘Tabor parameter’
1/3
R(Δγ)2 R
μ= = , (12.74)
E ∗ 2 ε3 β2ε
where β is defined by Eq. (12.32). The JKR solution represents the limit when μ 1,
whilst the DMT solution [or the semi-rigid solution] applies when μ 1.
An alternative definition of μ can be obtained by using (12.13) to substitute for ε
in Eq. (12.74), giving
1/3
R
μ = σ0 , (12.75)
E ∗ 2 Δγ
where σ0 is the maximum tensile traction in the Lennard-Jones force law, and we
have dropped the numerical factor B as being not significantly different from unity.
12.3.1 An Adhesive Length Scale
Both definitions of the Tabor parameter μ involve the radius R and hence strictly can
only be applied to the spherical indentation problem. But the JKR formalism can in
principle be applied to any contact geometry, and to assess whether it will lead to an
acceptable approximation, we need a more general definition of μ.
The JKR solution involves arbitrarily large tensile tractions near the edge of the
contact area, whereas we know that the Lennard-Jones force law precludes tractions
larger than σ0 . The situation is exactly analogous to that in the fracture mechanics of
elastic-plastic materials, where the elastic stress field implied by the stress intensity
factor necessarily exceeds the yield stress SY at sufficiently small distances from the
crack tip. In this latter case, it is argued that the singular field still characterizes the
conditions at failure, as long as the size of the region in which σ > SY is sufficiently
small, relative to the other geometric dimensions of the body. This is known as the
‘small-scale yielding’ criterion and is used to explain the success of linear elastic
fracture mechanics in predicting the failure of brittle solids (Kanninen and Popelar
1985).
In the analogous case of adhesive contact, the stress intensity factor (12.22) com-
pletely defines the traction field near the edge of the contact, so we can universally
determine the width s0 of the zone in which σ > σ0 by the equation

∗
KI E ∗ Δγ E Δγ
σ0 = √ = or s0 = . (12.76)
2πs0 πs0 πσ02
Thus, we anticipate that the JKR solution will represent a good approximation for an
adhesive contact problem if the smallest linear dimension a of the predicted contact
area satisfies the condition a s0 .
We can test this alternative criterion in the context of the spherical indentation
problem, where at pull off, the contact radius is given by Eq. (12.34) as
12.3 The Tabor Parameter 249
1/3
R â 9πΔγ R 2
a= = , (12.77)
β 8E ∗
using (12.32). The condition a s0 therefore translates to

1/3 ∗ 2/3 1/3
9πΔγ R 2 E Δγ R 1 8
or σ02 ≈ 0.21,
8E ∗ πσ02 E ∗ 2 Δγ π 9π
(12.78)
and the left-hand side is immediately recognizable as μ2 from (12.75).
12.3.2 Limitations on the JKR Solution
In view of (12.76), if a is a typical dimension of the contact area, the condition a s0

implies
a
μ 1 where μ ≡ σ0 ∗ . (12.79)
E Δγ
In effect, μ can be seen as a generalization of the Tabor parameter to non-spherical

geometries. However, it should be emphasized that the condition μ 1 is a necessary
but not sufficient condition for the JKR method to be appropriate. We also need to
make sure that there are no other regions in the contact area (i.e. away from the
edges) where a tensile traction is predicted with σ > σ0 . This question is brought into
focus if we consider the problem of determining the profile g0 (r ) of an axisymmetric
indenting body that maximizes the pull-off force F.
Suppose we start with a spherical body and ‘flatten’ it by removing some of the
material near r = 0. This will reduce the force P1 in Eq. (12.27) for a given contact
radius a, and hence we anticipate a larger pull-off force. An even larger force could
be achieved by using an indenter with a concave region near r = 0, but the traction
distribution may then exhibit a central tensile region and we would need to enforce
the condition σ < σ0 explicitly at all points other than those very near the contact
edge. If this condition is not satisfied, the true JKR solution might involve an annular
contact region at pull off, with the K I condition applying at both inner and outer
boundaries.
In practical cases, even if the condition σ < σ0 is satisfied, a small enclosed region
of decohesion might lead to an unstable separation. Johnson (1995) examines this
question in the context of the Westergaard problem of Sect. 6.5.6 and Fig. 6.6 [see
also Sect. 16.7.2 below]. If adhesive forces are present, full contact is theoretically
possible provided the mean pressure
∗
π E h0
p̄ > − σ0 . (12.80)
L
However, suppose there exists a small region of decohesion in the tensile region,
caused either by an air pocket or by microdefects in the surface profile. This region
is then mathematically equivalent to a crack in a monolithic body and it will propagate
if the corresponding stress intensity factor exceeds the value defined in Eq. (12.22).
The final stable state might then involve periodic contact areas or complete separation
of the bodies.
12.4 Solutions for Finite Tabor Parameter
We can state the axisymmetric adhesive contact problem as the search for a of contact
traction σ(g) defined by the Lennard-Jones force law (12.12), such that the resulting
elastic displacements u z (r ) define the gap
g(r ) = ε + g0 (r ) − Δ + u z (r ). (12.81)
Since the force law is defined for all positive values of separation, there is no need to
distinguish between regions of contact and separation, and indeed, such a distinction
is strictly meaningless (Greenwood 1997). However, the tractions in (12.12) decay
strongly with g, so in practice one can limit the outer radius in which Eq. (12.81) is
to be satisfied without loss of accuracy.
Muller et al. (1980) gave a numerical solution of the problem and showed that
the pull-off force decreases monotonically from the Bradley value to the JKR value
as μ increases, as shown in Fig. 12.5. They used the elasticity solution for a ring of
traction as a Green’s function to relate the tractions and displacements, leading to a
Fig. 12.5 Dependence of the pull-off force P0 on the Tabor parameter μ. The solid line represents
a curve fit to a numerical solution using the Lennard-Jones force law, from Eqs. (22),(23) of Wu
(2008). The dashed line is from the analytical approximation of Maugis (1992) [see Sect. 12.4.3
below]
12.4 Solutions for Finite Tabor Parameter 251
nonlinear integral equation with a logarithmically singular kernel involving complete

elliptic integrals. Greenwood (1997) gives a somewhat clearer exposition of a similar
numerical approach and pays particular attention to the points at which the bodies
jump into contact or out of contact. An alternative approach to the elasticity problem
is to use the methods of Sect. 5.1, with the potential function
∞
ϕ(r, z) = F(r, z, t)h(t)dt, (12.82)
0
in other words, with the upper limit in equation (Sect. 5.4) replaced by infinity. The
surface displacement and the contact pressure are then defined as
r
2 h(t)dt
u z (r, 0) = − ∗ √ (12.83)
E 0 r2 − t2
∞
1 d th(t)dt
p(r ) = √ , (12.84)
r dr r t2 − r2
and a suitable discretization of the function h(t) leads to a set of nonlinear algebraic
equations for a set of nodal values h i = h(ti ).
12.4.1 Jump-In at Large Tabor Parameter
The criterion defined in Sect. 12.3.1 suggests that the JKR solution should represent
a good approximation to the exact solution near the pull-off point B in Fig. 12.2
provided μ 1, but the predicted contact radius a tends to zero as we approach O, so
we should anticipate increasing errors in this range for all values of μ. In particular,
the JKR solution does not generally give a good description of the condition just
before two initially separated bodies jump into contact.
Wu (2010) explored this question in detail using a numerical solution, but he also
showed that at jump-in, separations are sufficiently large for the Lennard-Jones force
to be adequately described by the attractive [van der Waals] term alone, as suggested
by the dashed line in Fig. 12.1. In this case, we can replace Eq. (12.12) by the simpler
expression

8Δγ ε 3
σ(g) = − p(g) = . (12.85)
3ε g
Using this approximation, Wu was able to develop a dimensionless representation

of the spherical contact problem in which the solution is independent of μ. With the
present notation, if we define the dimensionless parameters
g (Δ − ε) r t
g̃ = ; Δ̃ = ; r̃ = √ ; t˜ = √
εμ3/7 εμ3/7 μ3/14 εR μ3/14 εR

hμ15/14 ε pεμ9/7 μ6/7 P uz

h̃ = ; p̃ = ; P̃ = ; ũ = , (12.86)
Δγ R Δγ RΔγ εμ3/7
the governing Eqs. (12.81), (12.83), (12.84), (12.85) reduce to

∞
r̃ 2 1 d t˜h̃(t˜)d t˜
g̃ = − Δ̃ + ũ; p̃ = (12.87)
2 r̃ d r̃ r̃ t˜2 − r̃ 2
∞
r̃
h̃(t˜)d t˜ 4
ũ = −2 ; − p̃ = 3 ; P̃ = 2π p̃(r̃ )d r̃ . (12.88)
0 r̃ − t˜
2 2 g̃ 0
These do not contain μ and hence, Δ̃in at the jump-in point is independent of μ.
It follows that (Δin − ε) ∼ μ3/7 or equivalently [in the notation of Eq. (12.31) and
Fig. 12.2]
β 2 Δin Δin C 1
Δ̂in = = ≈ − 4/7 + . (12.89)
R με μ μ
where C is a constant which Wu (2010) evaluated as C = 2.641, using a numerical

solution of the universal boundary value problem. Figure 12.6 compares this predic-
tion with a numerical solution using the full Lennard-Jones force law4 and shows
that the approximation is extremely accurate5 for μ > 1. This defines most of the
practical range, since P is a unique function of Δ for μ < 0.7 and hence jump-in
does not occur in this range. A similar argument shows that the tensile force −Pin at
jump-in should vary with μ−6/7 and this is also confirmed by the numerical results.
12.4.2 Simplified Force Laws
The algebraic and numerical challenge of the general problem arises mainly from the
strong nonlinearity of the Lennard-Jones force law (12.12). In particular, the tractions
are very sensitive to changes in g in the compressive range and very insensitive to such
changes in much of the tensile range, so iterative schemes designed to handle one of
these ranges are typically ill-suited to working in the other. Various approximations
or alternative force laws have been proposed to overcome this difficulty and generally
to simplify the analysis.
4 courtesyof J.A.Greenwood, private communication.

5 Wu (2010) presents a similar comparison with apparently less good agreement, but this is a result
of his confusing Δ with (Δ−ε) [J. J. Wu, private communication].
Fig. 12.6 The separation −Δin at jump-in from a numerical solution using the full Lennard-Jones
force law [circles] and from Eq. (12.89) [solid line]. The diamonds represent numerical values of
P̂in and these are well approximated by the straight line P̂in = 0.6616µ−6/7
One simplification is to treat the contact as rigid in the compressive range, so that
the force law is replaced by the nonlinear complementarity conditions

8Δγ ε3 ε9
σ= − g>0 (12.90)
3ε (g + ε)3 (g + ε)9
g=0 σ < 0. (12.91)
Notice that here we have also redefined g to measure from the contact condition, so
that the term ε should be deleted in Eq. (12.81).
12.4.3 Maugis’ Solution
Maugis (1992) made the further simplification of using a ‘Dugdale’ cohesive zone
approximation Dugdale (1960), in which the force law is defined by a step function,
as shown in Fig. 12.7. Equation (12.90) is then replaced by

Δγ
σ = σ0 H −g g > 0, (12.92)
σ0
where H (·) is the Heaviside step function, and the location of the step is chosen
to ensure that the work done in separating a unit area of surface (the shaded area
in Fig. 12.7) is equal to Δγ. The constant σ0 can be chosen to correspond to the
maximum tensile traction in the Lennard-Jones force law, defined by Eq. (12.13).
Fig. 12.7 Force law for the σ(g)

Maugis-Dugdale model
σ0
0 Δγ g
σ0
With Maugis’ formulation, we can distinguish a unique contact area where the
gap r
r2 2 h(t)dt
g(r ) = −Δ− ∗ √ (12.93)
2R E 0 r2 − t2
is equal to zero. Assuming this to be a circle of radius a, we obtain

r ∗
h(t)dt E r2
√ = −Δ 0 ≤ r < a, (12.94)
0 r2 − t2 2 2R
with solution6 ∗
E t2
h(t) = −Δ 0 ≤ t < a, (12.95)
π R
as in Sect. 5.2.
The Dugdale force law (12.92) ensures that there will be some finite radius b
beyond which σ = 0, and this in turn requires that h(t) = 0 for t > b from (12.84). To
determine h(t) in the range a < t < b, we have the condition
b
1 d th(t)dt
p(r ) = √ = −σ0 a < r < b, (12.96)
r dr r t2 − r2
with solution √
2σ0 b2 − t 2
h(t) = a < t < b, (12.97)
π
where a constant of integration has been assigned so as to ensure that h(b) = 0 in order
to eliminate a singularity at r = b. We also require the tractions to be non-singular at
6 The perceptive reader will notice that this is identical to the result from the Hertz problem
[Eq. (5.31)], and might [erroneously] deduce that Derjaguin et al. (1975) are correct in assum-
ing the Hertz solution to apply in 0 ≤ r < a. However, the contact pressure distribution is given by
Eq. (12.84), and this will be influenced by the value of h(t) in t > a as well as in t < a, even for
points in 0 ≤r < a.
r = a, which in turn requires that h(t) be continuous at t = a, giving

∗ a2
E − Δ = 2σ0 b2 − a 2 . (12.98)
R
The outer radius b defines the boundary where the tensile traction σ0 drops to zero,
so from Fig. 12.7 and Eq. (12.92), we must have
Δγ
g(b) = , (12.99)
σ0
where the gap g(r ) is defined by (12.93).

For a given value of indentation Δ, the unknown radii a, b can be determined
from (12.98), (12.99). However, a more efficient algebraic solution is obtained by
noting that the rigid-body displacement Δ and the total force
b
P = −2π h(t)dt (12.100)
0
are both monotonic functions of the ratio x = a/b, which can therefore be used
as an alternative loading parameter. Using (12.98) to eliminate Δ in (12.95) and
substituting the result and (12.97) into (12.93), we then find, after some algebraic
simplification, that (12.99) is equivalent to
π
4μb̂ 1−x 2 arccos (x)−(1 − x) + b̂2 (1 − 2x 2 ) arccos (x) + x 1−x 2 = ,
μ
(12.101)
where ∗ 1/3
a βb E R
x= and b̂ = ; β= , (12.102)
b R Δγ
as in the dimensionless form of the JKR solution defined by Eq. (12.31). We also
obtain
P 4b̂3 x 3 2μb̂2
P̂ ≡ = − x 1 − x 2 + arccos(x) (12.103)
π RΔγ 3π π
β2Δ
Δ̂ ≡ = b̂2 x 2 − 2μb̂ 1 − x 2 (12.104)
R
from (12.98), (12.100) respectively.
For given values of x, μ, we can solve the quadratic Eq. (12.101) for b̂ and sub-
stitute the result into (12.103), (12.104) to obtain P̂, Δ̂. This procedure allows us
to make parametric plots of P̂ as a function of Δ̂, for any given value of μ. Some
representative plots are shown as the solid lines in Fig. 12.8, the curve for μ = 5 being
essentially indistinguishable from the JKR solution.
Fig. 12.8 Relation between dimensionless compressive force P̂ and dimensionless rigid-body
indentation Δ̂ from Maugis’ solution, for μ = 5, 1, 0.5, 0.2
Notice that the solid lines terminate away from the origin for x = 0 and hence
a = 0. The curve can be completed by considering the case where there is no contact,
so that σ = σ0 throughout 0 ≤r < b, leading to the relations

1 4μ2 b̂ μb̂2
P̂ = −μb̂ ; Δ̂ =
2
1+ − , (12.105)
μ π 2
and these continuations are plotted as dotted lines in Fig. 12.8. The pull-off force
can be identified as the maximum negative value of P̂. The resulting expression for
pull-off force as a function of μ is represented by the dashed line in Fig. 12.5.
12.4.4 The ‘double-Hertz’ Approximation
Maugis’ solution for the pull-off force tends to the JKR value as μ → ∞, but the
pressure distribution in this limit does not exhibit the classical square-root singularity,
since the tensile tractions are bounded by a maximum value of σ0 . An alternative
approximation that tends exactly to the JKR solution in this limit was proposed by
Greenwood and Johnson (1998), based on the superposition of two Hertzian pressure
distributions, as shown schematically in Fig. 12.9.
Since the Hertzian pressure distribution
∗√ 2
2E a − r2
p(r ) = (12.106)
πR
Fig. 12.9 The

‘double-Hertz’ pressure
distribution. The√term p(r)
proportional to b2 −r 2
alone is shown dotted
0 r a b
σ0
produces a quadratic distribution of displacement u z (r ) in 0 ≤ r < a, it is clear that

the distribution
2E 2
∗
p(r ) = C 1 a − r 2 − C 2 b2 − r 2 (12.107)
πR
with b > a, will produce quadratic displacements in 0 ≤ r < a, and corresponds to a
total force P and indentation depth Δ given by
∗
4E (C1 a 3 − C2 b3 ) (C1 a 2 − C2 b2 )
P= ; Δ = u z (0) = . (12.108)
3R R
The resulting curvature must match the radius R of the indenting sphere, from which
we obtain
C1 − C2 = 1. (12.109)
We note from Fig. 12.9 that the maximum tensile contact traction occurs at r = a,
and this must be limited to σ0 , giving
∗ √
2E C2 b2 − a 2
σ0 = − p(a) = . (12.110)
πR
The assumed pressure distribution (12.107) would be the exact solution of the
contact problem if the force law σ(g) were defined by the relation obtained by
eliminating the radius r between the expressions for σ(r ) = − p(r ) and the gap g(r )
in the annulus a < r < b. Thus, whereas Maugis approximates the force law to
make the contact problem more tractable, Greenwood and Johnson approximate the
pressure distribution and accept the approximation in the force law that this implies.
As in Eq. (12.72), the gap in a <r < b due to the pressure distribution (12.108) is
C1 2 a
g(r ) = a r − a 2 − (2a 2 − r 2 ) arccos (12.111)
πR r
and for particular cases, the implied force law could be plotted parametrically, with
r as parameter. In particular, the interface energy is
∗
b
dg 2E C1 C2 (b − a)2 (b + 2a)
Δγ = − p(r ) dr = , (12.112)
a dr 3π R 2
after substituting for p, g from Eqs. (12.107), (12.111) and performing the integra-
tion.
If the force P is prescribed, (12.108), (12.109), (12.110), (12.112) provide four
equations for the unknowns C1 , C2 , a, b. However, as in Maugis’ solution, it is easier
to treat x = a/b as a parameter, and eliminate C1 , C2 from the last three of these
equations. This yields a quadratic equation for b whose solution can then be sub-
stituted back to determine the values of a, P and Δ for a prescribed x. Greenwood
and Johnson report results for the pull-off force that are close to those defined by
Maugis’ approximation plotted in Fig. 12.5.
12.4.5 More General Axisymmetric Geometries
The superposition used in the double-Hertz method is analogous to the Cattaneo–

Mindlin superposition of Sect. 9.1, and indeed Greenwood and Johnson introduced it
by remarking that the superposition of two solutions of the normal contact problem
for different forces P1 , P2 and contact areas A1 , A2 , with A1 > A2 leads to a uniform
displacement in the smaller contact area A2 , an appropriate multiple of which they
then superpose on the solution P2 . The same argument will clearly extend to more
general axisymmetric geometries, as with the Ciavarella-Jäger theorem of Sect. 9.2.
All we need to do is to replace the two Hertzian distributions in (12.107) by pressure
distributions from this more general problem.
Maugis’ method is also easily extended to more general axisymmetric geometries.
Equation (12.97) remains valid and only the value of h(t) in 0 ≤ t < a will be changed
to ∗ t
E d [g0 (r ) − Δ]r dr
h(t) = √ 0≤t <a (12.113)
π dt 0 t2 − r2
[from (5.13)], where g0 (r ) is the initial gap function in the more general problem.
12.5 Other Geometries
All the methods described above can be extended to two-dimensional problems,

using the relations between contact pressure and surface slope du z /d x defined in
12.5 Other Geometries 259
Chap. 6. In particular, the JKR solution for a fairly general profile can be obtained
as in Sect. 12.2.1 by first solving the non-conformal problem without adhesion, and
then superposing the solution of the flat punch problem [Eq. (6.14)], with the linear
multiplier chosen to give the required stress intensity factor at the two edges.
If the indenting body is symmetrical, so that g0 (x) = g0 (−x) and contact occurs
in −a < x < a, the flat punch solution satisfying the energy release rate condition
(12.22) is defined by

P2
p(x) = √ with P2 = − 2π E ∗ Δγa, (12.114)
π a2 − x 2
and the force required to establish this contact area in the absence of adhesion is
given by Eq. (6.22) — i.e.
∗
E a
ξg (ξ)dξ
P1 = 0 . (12.115)
2 −a a2 − ξ2
We recall that the rigid body displacement cannot be defined in two-dimensional

problems for infinite bodies, but it is still possible to establish a relation between the
applied force P and the semi-width a of the contact area, so the pull-off force can
also be determined from the condition
dP
=0 where P = P1 + P2 . (12.116)
da
For lower values of μ, both Maugis’ method (Baney and Hui 1997) and the double-
Hertz method (Jin et al. 2014) can be applied.
12.5.2 Elliptical Contact Area
Johnson and Greenwood (2005) developed a JKR solution for the non-axisymmetric
Hertzian contact problem, where the contact area is expected to be elliptical.
However, the stress intensity factor under an elliptical flat punch varies around the
periphery, so the condition (12.22) cannot be satisfied by the simple expedient of
superposing an appropriate multiple of the flat punch traction distribution.
Galin’s theorem (Sect. 2.4) demands that the contact pressure distribution [with
an elliptical contact area] have the form
−1/2
x2 y2
p(x, y) = (C1 + C2 x 2 + C3 y 2 ) 1 − 2 − 2 , (12.117)
a b
and the contact condition (3.25)

u z (x, y) = Δ − Ax 2 − By 2 (12.118)
provides three equations for the five unknowns compising C1 , C2 , C3 and the
unknown semi-axes a, b. Johnson and Greenwood used the two remaining free
constants to satisfy (12.22) at the ends of the axes (±a, 0) and (0, ±b), and showed
that K I is then almost but not exactly constant around the ellipse, which of course
implies that the actual contact area in the adhesive problem will not be exactly ellipti-
cal. They also showed that adhesion tends to reduce the ellipticity of the contact area,
relative to the Hertzian case. A similar method was used by Barber and Ciavarella
(2014) to estimate the effect of adhesion in the Hertzian contact of anisotropic elastic
bodies.
12.5.3 General Three-Dimensional Geometries
The adhesive contact problem for a general three-dimensional gap function g0 (x, y)
can be stated as in Eq. (12.81) as
g(x, y) = ε + g0 (x, y) − Δ + u z (x, y), (12.119)
where
1 p(ξ, η)dξdη
u z (x, y) = , (12.120)
πE∗ (x − ξ)2 + (y − η 2 )
from (2.17, 2.18), and the contact pressure

8Δγ ε3 ε9
p(x, y) = −σ(g{x, y}) with σ(g) = − . (12.121)
3ε g3 g9
The domain of integration and of definition of p(x, y) strictly comprises the whole
surface of the half-space, but as in Sect. 12.4 we can generally truncate these domains
to a finite region outside which the contact pressure is assumed to be negligible. We
can also define ‘conformal’ adhesive contact problems in which a finite contact area
is defined by the planform of an indenting body such as a flat punch.
No general analytical techniques are known for the solution of these problems and
numerical solution is complicated by the nonlinear relation (12.121). The general
form of the problem is clearly identical to that of a nonlinear ‘Winkler’ layer on an
elastic half-space, discussed in Sect. 14.3.1, and normally we would expect to use
an iterative strategy, but it is challenging to ensure that the resulting algorithm is
convergent and stable.
An alternative approach that is potentially more stable is to approximate the force
law by the Eqs. (12.90, 12.91). We can then define a contact region A as that in which
g = 0 and where the contact traction can take any compressive value. The solution
12.5 Other Geometries 261
might then be implemented numerically in a Gauss–Seidel algorithm, where nodes

are toggled between states depending on the sign of the resulting tractions or gaps.
Problems
1. Use the Lennard-Jones interatomic potential (12.2) to estimate the attractive force
per unit area between two infinite layers each of finite thickness b. Hence determine
the effective interface energy in separating two such layers from their equilibrium
position for the case where b = 3rm . Notice that the equilibrium position for finite
layers differs from (12.9).
2. If we place two half-spaces at the equilibrium spacing of Eq. (12.9), the composite
body essentially comprises a single stress-free monolithic body. Use this model with
the force law (12.12) to obtain an estimate of Young’s modulus E as a function of
Δγ and ε.
3. Two rigid bodies have quadratic profiles such that the initial gap function has the
Hertzian form
x2 y2
g0 (x, y) = + .
2Rx 2R y
Find the attractive force when they are at the point of closest approach and show that
it reduces to Eq. (12.17) when Rx = R y .
4. A cylindrical rigid punch of radius a has a spherically concave end of radius R
where R a. It is placed above a rigid half-space such that the separation on the axis
r = 0 is h. Assuming the Lennard-Jones force law (12.12), find the attractive force
F between the bodies as a function of h. For the special case where R = a 2 /2ε, find
the value of h at which F is a maximum, and the value of this maximum [the pull-off
force]. Why does this problem not fall into the category defined by Eq. (12.18), where
the pull-off force is independent of the form of the force law?
5. A rigid sphere of radius R is compressed between two identical elastic half-spaces.
The half-spaces are now gradually separated under displacement control. Use the JKR
theory to describe the subsequent behaviour and define the relation between force
and separation. Is this similar to Fig. 12.2? If not explain why not.
6. Use the JKR theory to find the pull-off force for the flat and rounded punch of
Fig. 5.5.
7. Find the hysteretic energy loss implied by the contact-separation cycle illustrated
in Fig. 12.3 for the JKR contact of a sphere on a plane, by evaluating the energy
difference Π A −ΠC .
(a) (b)
k k
radius R
radius R
Fig. 12.10 (a) Atomic force microscope [AFM], (b) idealization as a rigid hemisphere supported
by a spring
8. The atomic force microscope [AFM] in Fig. 12.10a can be idealized as an elastic
body with a spherical tip of radius R supported on a spring of stiffness k, as shown
in Fig. 12.10b.
The fixed end of the cantilever is moved down until the tip jumps into contact
with an elastic plane. It is then moved up until the tip jumps out of contact. Use the
∗
JKR theory with composite modulus E and interface energy Δγ to determine the
contact force just after jump-in and just before jump-out [this may require a numerical
solution of the resulting equations]. Show that when k is small [the cantilever is very
flexible], these force values are independent of k. How small does k need to be [in
∗
terms of E , Δγ, R] for the limiting values to provide good estimates.
9. A cylinder of radius R is pressed against a half-plane by a force P. Using the two-
dimensional solutions from Chap. 6 and the stress intensity factor from Eq. (12.22),
determine the relation between P and the contact semi-width a. Hence determine
the pull-off force for this geometry. Does the pull-off force increase or decrease with
∗
E ?
10. Calculate the work done by the force P in the JKR solution in loading from point
A to point C in Fig. 12.3 by evaluating the integral
ΔC aC
dΔ
W = P(Δ)dΔ = P(a) da.
ΔA aA da
Hence show that this is an alternative way of calculating the hysteresis loss of
Problem 7.
11. A conical indenter with profile g0 (r ) = αr, α 1 is moved under displacement
control towards an elastic half-space. Show (i) that a dimensionless parameter λ
analogous to the Tabor parameter can be defined for this problem, and (ii) that if
the force law is defined by (12.85), a universal dimensionless representation of the
governing Eqs. (12.81), (12.83), (12.84) can be defined that is independent of λ.
Hence deduce an equation analogous to (12.89) showing how the approach Δin at
jump-in varies with λ.
Chapter 13
Beams, Plates, Membranes and Shells
Optimal structural designs often involve thin-walled structures. For example, if we

wish to design a torsion bar to a given level of maximum shear stress, the cross
section that minimizes the weight of the bar is a thin-walled cylinder, limited only
by the requirement to avoid buckling of the walls and keep the envelope of the
structure within externally prescribed dimensions. Thin-walled structures are also
often found in the natural world, presumably the result of evolution-driven structural
optimization.
Considerable savings in computational or algebraic complexity can generally
be achieved by treating such structures as beams, plates or shells. However, the
corresponding contact problems for these elements exhibit a remarkable feature:
instead of finding a continuous contact pressure distribution, as in all the two and
three-dimensional problems considered so far, the interaction between contacting
components will generally comprise one or more concentrated point or line forces.
13.1 Contact of Beams
To introduce this topic, we consider the problem of Fig. 13.1a in which a rigid cylinder
of radius R is pressed down by a force P onto the mid-point of a simply supported
beam of length L and flexural rigidity E I . If a concentrated force P were applied at
the mid-point of the beam as shown in Fig. 13.1b, the two support reactions would be
each P/2 and the bending moment at the mid-point would be M = P L/4. It follows
that the initially straight beam would deform locally to a radius of curvature
EI 4E I
R0 = = . (13.1)
M PL

https://doi.org/10.1007/978-3-319-70939-0_13
264 13 Beams, Plates, Membranes and Shells
P
(a) (b)
P
R
L L
Fig. 13.1 A cylinder pressed against a beam
Now the contact inequalities will be satisfied by this deformation in the problem of
Fig. 13.1a, provided the radius of the cylinder is less than that of the deformed beam.
Thus, there will be only point contact between the cylinder and the beam as long as
4E I
R < R0 or P < P0 ≡ . (13.2)
LR
We might expect that for P > P0 , a finite contact region would be established,
but in this range, the loading actually takes the form illustrated in Fig. 13.2, with two
symmetrically disposed contact points each transmitting a force P/2. This loading
is known as ‘four-point bending’ and is often used in fatigue testing to generate
a state of uniform bending in a central gauge length of the specimen. Elementary
calculations show that the bending moment is constant between A and B and of
magnitude
P(L − a)
M= , (13.3)
4
corresponding to a radius of curvature
4E I
R1 = . (13.4)
P(L − a)
Thus, if we choose R1 = R and solve for

4E I P0
a=L− = L 1− , (13.5)
PR P
Fig. 13.2 Contact tractions P P

for P > P0 2 a 2
A B
L
13.1 Contact of Beams 265
Fig. 13.3 A heavy beam F

lifted by a force at one end
w0 per unit length

Δ B C
A
z
a
the cylinder will just touch the beam between A and B, but no contact pressure will
be developed except for the two concentrated forces at A and B. If any other value
of a were chosen, it is easily shown that the contact inequalities would be violated
either between A and B, or to the left of A and the right of B.
We conclude that, as the force P is increased, we first have a period of central
point contact until P = P0 , after which the contact point bifurcates with the two
forces P/2 moving apart as P is further increased.
13.1.1 A Heavy Beam Lifted from the Ground
As a second example, we consider the problem of Fig. 13.3 in which a heavy beam
AC of weight w0 per unit length rests on a rigid plane surface. A vertical force F is
applied at the end A causing a segment of the beam AB of as yet unknown length a
to be lifted away from the support.
Equilibrium arguments show that the bending moment M and the shear force V
in the segment AB are
w0 z 2
M AB (z) = F z − V AB (z) = w0 z − F, (13.6)
2
where z is measured from the lifted end A. However, in BC, the beam is undeformed,
so VBC (z) = M BC (z) = 0. Now if M AB (a) = 0, we would require the contact to furnish
a concentrated moment at B in order to maintain equilibrium, but this is impossible
since the contact forces are only allowed to be compressive and a concentrated
moment could be construed as the limit of a pair of equal and opposite vertical
forces. We therefore conclude that
w0 a 2 2F
M AB (a) = Fa − =0 and hence a= . (13.7)
2 w0
It then follows that

V AB (a) = F, (13.8)
implying the existence of a concentrated compressive contact force F at B. This

form of ‘contact pressure distribution’ is of course compatible with the findings of
the previous example.
13.1.2 Adhesive Forces
In the problem of Fig. 13.3, we justified the exclusion of a concentrated moment at

z = a on the grounds that the contact pressure cannot be tensile. However, if adhesive
forces are considered, as in Chap. 12, this restriction must be lifted. The problem is
then most easily solved using an energy formulation.
We suppose that there exists some interface energy Δγ per unit length associated
with the separated segment in Fig. 13.3. We can then write the total energy in the
system as Π = U +Ω +Γ , where U is the elastic strain energy, Ω is the potential
energy of the external force F and the weight w0 , and Γ is the interface energy. The
vertical displacement of the beam [upward positive] is obtained from the bending
equation using (13.6) as
F(z 3 − 3a 2 z + 2a 2 ) w0 (z 4 − 4a 3 z + 3a 4 )
u(z) = − , (13.9)
6E I 24E I
so a
F 2a3 Fw0 a 4 w0 a 5
Ω = −Fu(0) + w0 u(z)dz = − + − . (13.10)
0 3E I 4E I 20E I
The strain energy is

1 a
F 2a3 Fw0 a 4 w0 a 5
U= [M AB (z)]2 dz = − + (13.11)
2E I 0 6E I 8E I 40E I
and the interface energy Γ = Δγa, so the total potential energy is
F 2a3 Fw0 a 4 w0 a 5
Π =U +Ω +Γ =− + Δγa − + Δγa. (13.12)
6E I 8E I 40E I
The separated segment length must take the value that minimizes Π , giving
dΠ F 2a2 Fw0 a 3 w2 a 4
=− + − 0 + Δγ = 0, (13.13)
da 2E I 2E I 8E I
or √ √
a 1+ 1−λ 2w0 2Δγ E I
= where λ= , (13.14)
a0 2 F2
and a0 = 2F/w0 is the separation length in the absence of adhesive forces.

The inclusion of adhesive and hence potentially tensile tractions permits a con-
centrated moment to be transmitted at the separation point z = a whose magnitude
can be found by substituting z = a in equation (13.6)1 and using (13.14) for a. After
algebraic simplification, we obtain

M(a) = 2Δγ E I , (13.15)
showing that this moment depends only on the interface energy and the flexural
rigidity of the beam. Majidi and Adams (2009) have shown that this is a general
result for all beam problems with adhesive forces. In other words, the√presence of
adhesion implies that a concentrated reaction moment of magnitude 2Δγ E I is
generated at the boundary between regions of contact and separation in any problem
involving contact between a beam and a rigid body. It is analogous with the use of
a geometry-independent mode I stress intensity factor or energy release rate in the
‘JKR’ solution of three-dimensional adhesive contact problems.1 Majidi and Adams
(2009) also proved a related result for contact problems involving plates.
13.1.3 Piston Ring in a Cylinder
Piston rings are used to improve the seal between a piston and cylinder and are
therefore generally preloaded. A simple piston ring might be idealized as a curved
beam of radius R1 = R0 +Δ, where R0 is the radius of the cylinder and Δ R0 . In
other words, the ring is made to have a slightly larger radius than the cylinder, so
that when installed it presses against the cylinder to improve sealing.
If a bending moment M is applied to the beam, the radius will change from R1 to
R, where
1 1 M
− = . (13.16)
R R1 EI
Thus, we can devise a scenario for assembling the ring as:-

1. Apply a moment to the ends of the ring of magnitude

1 1 EIΔ
M0 = E I − ≈ , (13.17)
R0 R1 R02
which is just sufficient to reduce the radius of the ring to that of the cylinder.
Notice that in this state, the bending moment will be given by M0 at all points
around the ring, so it will be circular of uniform radius R0 .
2. Insert the deformed ring in the cylinder. Since the ring is the same radius as the
cylinder, there will be no contact tractions at this stage.
1 See Sect. 12.2.

Fig. 13.4 Step 3 of the

‘assembly scenario’ cylinder
ring
α
M0
θ
α
M0
3. Release the moment M0 , which is equivalent to applying an equal and opposite

moment on the ends of an unstressed ring that just fits it the cylinder.
The final stage is illustrated in Fig. 13.4 and it will clearly cause contact tractions.
The ends θ = 0, 2π will ‘dig in’ causing local concentrated reaction forces, and
we expect regions of separation between the ring and the wall in 0 < θ < α and
2π −α < θ < 2π. Also, based on earlier examples, we might expect a concentrated
normal force at the points θ = α and θ = 2π−α.
If the ring remains in complete contact in α < θ < 2π −α, then the radius of the
beam is unchanged in this segment, so from Eq. (13.16), the bending moment due to
the corrective moments M0 in Fig. 13.4 must be zero. There can be no concentrated
moment at θ = α, so the boundary conditions for the separated segment 0 < θ < α
can be stated as
u(0) = 0; M(0) = M0 ; u(α) = 0; u (α) = 0; M(α) = 0, (13.18)
where u(θ) is the radial displacement of the beam in Fig. 13.4.

If the reaction force at θ = 0 is denoted by F, the bending moment in the separated
segment is
M0
M(θ) = M0 − F R0 sin θ and hence F= , (13.19)
R0 sin α
from (13.18)5 . The elastic strain energy in the separated segment is

α
1
U = M 2 R0 dθ
2E I 0

R0 F 2 R02 {2α − sin(2α)}
= M02 α − 2M0 F R(1 − cos α) + (13.20)
2E I 4
and hence from Castigliano’s second theorem and the boundary condition u(0) = 0,

∂U R0 F R02 {2α − sin(2α)}
= −2M0 R0 (1 − cos α) + = 0. (13.21)
∂F 2E I 2
Substituting for F from (13.19)2 and simplifying, we then obtain the condition
α
cos α + −2=0 (13.22)
sin α
for the angle α defining the extent of separation. The solution is α = 2.139
radians= 122.6◦ , and we note that it is independent of the magnitude of the moment
M0 , which of course we should have anticipated since Fig. 13.4 defines a receding
contact problem.2 Another consequence of the receding contact geometry is that the
solution varies linearly with the applied forces, in this case, the mismatch moment
M0 . Thus, if the ring is made with a larger value of radial interference Δ with a view
to increasing contact pressure and hence reducing gas blow-by, the radial displace-
ment u and hence the gap opened in the separation region will increase, which would
seem to produce exactly the opposite effect to that desired.
Once α has been determined, the reaction force F can be obtained from equation
(13.19)2 as
1.1866M0 1.1866E I Δ
F= = . (13.23)
R0 R03
Also, since the bending moment is uniform in α < θ < 2π −α, the shear force there
must be zero, implying the existence of a concentrated normal [radial] reaction at
θ = α given by
0.6387M0 0.6387E I Δ
Fα = −F cos α = = . (13.24)
R0 R03
Figure 13.5 shows the forces acting on the separation and contact segments in the
‘assembled’ state [i.e. after the moment M0 has been released]. Notice that equilib-
rium of the contact segment α < θ < 2π−α demands that there be a uniform contact
pressure p per unit length, given by
F sin α M0 EIΔ
p= = 2 = . (13.25)
R0 R0 R04
2 see Chap. 11.

Fig. 13.5 Forces acting on EI Δ

the assembled piston ring 0.6387 3
R0
o EI Δ
122.6 1.1866 3
EI Δ R0
4
R0 o EI Δ
122.6 1.1866 3
R0
EI Δ
0.6387 3
R0
13.1.4 Two and Three-Dimensional Effects
The concentrated forces predicted in the preceding examples are clearly unphysi-
cal, since deformation of the materials local to the contact point will lead to the
establishment of a finite contact area, with a corresponding finite maximum contact
pressure.
If the beam is of relatively thin rectangular cross section, the problem shown in
Fig. 13.1a can be formulated as a two-dimensional elastic contact problem under
conditions of plane stress. Keer and Miller (1983) gave a solution based on a Fourier
transform representation of the displacement field.3 For P P0 , a good approxima-
tion can be achieved by simply patching in the two-dimensional Hertzian solution to
the contact point—i.e. essentially treating the beam locally as a half-plane.
A much better approximation in this range is obtained if we allow for the radius
of the deformed beam in defining the contact kinematics. The radius R0 of the beam
centreline is given by Eq. (13.1), and by definition it achieves the value R when
∗
P = P0 [see Eq. (13.2)]. Thus, the effective Hertzian radius R can be defined as

1 1 1 1 P
= − = 1− (13.26)
R∗ R R0 R P0
from Eq. (3.18), where we note that bending deflections cause the local contact sur-
face of the beam to become concave, so the contribution from R0 is negative in
3 see Sect. 14.4 below.

Fig. 13.6 Maximum contact 1.0

pressure p(0) predicted by 0.9
Eq. (13.27) (solid line)
0.8
compared with finite element
results for h/L = 0.2 0.7
[squares] and h/L = 0.02 0.6
[circles] [after Kim et al. h3 p(0)
0.5
(2014)]. The dashed line is L P0
0.4
the simple Hertzian
approximation, omitting the 0.3
contribution from R0 0.2
0.1
0.0
0.0 0.2 0.4 0.6 0.8 1.0
P / P0
(13.26). The maximum contact pressure is then obtained as

P E∗ P E∗ P
p(0) = = 1− , (13.27)
π R∗ πR P0
where P is here the force per unit thickness of the beam. Equation (13.27) is compared
with finite element results due to Kim et al. (2014) in Fig. 13.6, for a beam of length
L and depth h. The agreement is excellent even up to quite large values of P/P0 if
h/L is sufficiently small, and the approximation captures the fact that the maximum
contact pressure actually decreases with P as P0 is approached.
If the beam in Fig. 13.1a has a circular cross section [say of radius R1 ], we can
∗
use this and R from Eq. (13.26) to define the two orthogonal radii for a three-
dimensional Hertzian calculation. The contact area will then be elliptical with the
major axis aligned with that of the beam.
13.1.5 Matched Asymptotic Expansions
The Hertzian approximation is strictly appropriate only as long as the length of

the contact region in the direction of the beam axis is small compared with the
depth of the beam. If this condition is not satisfied, a more rigorous though still
approximate approach is to use the method of matched asymptotic expansions [see
Sect. 6.7.1], but with the beam bending equation defining the ‘outer problem’. The
mathematical effect of this procedure is to introduce a local contact compliance
between the indenting body and the beam, whose properties are defined by a two-
dimensional analysis in the beam cross section (Castillo and Barber 1997).
If a beam is loaded by lateral tractions that are uniform along the axis, the shear
force V varies linearly with z and the corresponding shear stress distribution in the
cross section is also linear with z, but has a universal form V f (x, y) for a given
beam cross section,4 where f also describes the distribution of shear stress in the
case where V is independent of z. The inner problem is therefore defined by the given
contact tractions balanced by a distribution of body forces of form f . The solution for
a beam of circular cross section is given by Castillo and Barber (1997) [see Fig. 6.11
and Eq. (6.121)], but here for illustration, we discuss the simpler case where the beam
is rectangular and the function f can be approximated by the elementary distribution

3 4y 2
fy = 1− 2 ; f x = 0. (13.28)
2h h
The upper surface of the beam is loaded by a contact pressure p, so if this is balanced
by a body force distribution linearly proportional to (13.28), the in-plane stress field
must satisfy the equilibrium equation

dσ yy h h
+ C fy = 0 with σ yy = − p; σ yy − = 0, (13.29)
dy 2 2
from which
1 3y 2y 3
σ yy = −p + − 3 . (13.30)
2 2h h
We then use Hooke’s law to determine the strains, and hence the in-plane displace-
ments, as
p y 3y 2 y4
uy = − + − 3 + B, (13.31)
E 2 4h 2h
where B is an arbitrary rigid-body displacement. The compliance of the contact in

the inner solution is then defined as
h/2
h
v ≡ U − uy where U = u y (y) f y (y)dy (13.32)
2 −h/2
is an energetically neutral weighted average of u y (Renton 1991). We obtain
13 p 13w w 35Ehd
v= = = with k= , (13.33)
35Eh 35Ehd k 13
where d is the width of the beam and w = pd is the contact force per unit length.
This shows that the inner contact problem in this simple case mimics a linear elastic
foundation with modulus k.
In the contact region, we require
w(z) w(z)
g0 (z) − Δ + + u(z) = 0 and hence u(z) = Δ − g0 (z) − , (13.34)
k k
4 See Barber (2010), Chaps. 17 and 28.

where g0 (z) is the initial gap function. Substituting this into the beam bending equa-
tion d 4 u/dz 4 = w/E I , we obtain
d 4w d 4 g0 k
+ 4λ 4
w = −k where λ4 = . (13.35)
dz 4 dz 4 4E I
The general solution of the corresponding homogeneous equation is
w(z) = C1 cos(λz) cosh(λz) + C2 sin(λz) sinh(λz) + C3 sin(λz) cosh(λz)

+ C4 cos(λz) sinh(λz) + w0 (z), (13.36)
where w0 (z) is any particular solution and C1 , C2 , C3 , C4 are arbitrary constants to be

determined from the boundary conditions. The displacements can then be obtained
by substituting (13.36) into (13.34)2 . There will generally be one or more contact
regions surrounded by separation regions in which u(z) is at most a third-order
polynomial in z, and continuity must be enforced up to the third derivative in u(z) at
the boundaries of the contact regions. The extent of these regions is then determined
from the condition(s) that w(z) → 0 at each such boundary.
For the problem of Fig. 13.1a, g0 (z) = z 2 /2R, so a particular solution of (13.35) is
w0 (z) = 0. For P > P0 , we anticipate contact in two symmetrically disposed regions
−b < z < −a and a < z < b and the solution is conveniently obtained by assuming a to
be given, using the continuity conditions to determine b and the remaining constants,
and then determining the contact force per unit width P by integrating p(z) = w(z)/d
over the contact regions. As P is increased, the pressure distribution rapidly stabilizes
to a constant form and simply moves over the beam surface as predicted by Eq. (13.5).
Figure 13.7 shows the solution from Eq. (13.36) for a representative case, compared
with the finite element results of Kim et al. (2014). The approximate solution is not
able to capture the square-root bounded behaviour at the edges of the contact region,
Fig. 13.7 Contact pressure distribution in one of the two contact regions for P > P0 from
Eq. (13.36), compared with finite element results [FE] due to Kim et al. (2014). The dashed line
indicates the location of the concentrated force in the corresponding beam solution of Fig. 13.2 and
Eq. (13.5)
and it underestimates the extent of this region, but the results are quite good for a
simple analytical approximation.
If the beam is circular rather than rectangular, the contact compliance v is given by
Eq. (6.121), which is a nonlinear function of contact pressure since the contact semi-
width a depends on P. However, the nonlinearity is very weak, so the qualitative
behaviour is not very different from the linear case, and the elastic foundation model is
amenable to iterative solution. For other beam sections, the inner [two-dimensional]
problem would most likely need to be solved by numerical methods, but the resulting
contact compliance relation could then be used in the same way in the analytical outer
solution.
13.2 Contact of Plates
The lateral displacement u(x, y) of a uniform plate of thickness h loaded by a dis-

tributed force p(x, y) satisfies the equation

p
∇2 ∇2u = (13.37)
D
Mansfield (1989), where
Eh 3
D= (13.38)
12(1 − ν 2 )
is the stiffness of the plate, and the positive directions of u and p are aligned. Here,
we shall restrict attention to problems involving circular plates, so that u and p will
be expressed as functions of the polar coordinates r, θ. The Laplace operator ∇ 2 then
takes the form
∂2 1 ∂ 1 ∂2
∇2 ≡ 2 + + 2 2 (13.39)
∂r r ∂r r ∂θ
and corresponding expressions for bending moments M and shear forces V per unit
length are given by Mansfield (1989) as
(a) Mθr (b)

Mrθ Mθθ Mrr Vr
Vθ
Mθr Mrθ Vr Vθ
r Mrr Mθθ r
θ θ
Fig. 13.8 Sign conventions for moments and shear forces per unit length
13.2 Contact of Plates 275

∂2u 1 ∂u 1 ∂2u ∂ 1 ∂u
Mrr = −D +ν + 2 2 ; Mr θ = −D(1 − ν)
∂r 2 r ∂r r ∂θ ∂r r ∂θ

1 ∂u 1 ∂2u ∂2u
Mθθ = −D + 2 2 +ν 2 , (13.40)
r ∂r r ∂θ ∂r
∂ 2 D ∂ 2
Vr = −D ∇ u; Vθ = − ∇ u. (13.41)
∂r r ∂θ
These components are illustrated in Fig. 13.8, where the moment arrows represent
clockwise moments when viewed in the direction of the arrows.
13.2.1 Displacement Due to a Concentrated Point Force
Suppose a circular plate of radius a is built in at r = a and loaded by a concentrated

force P at the origin. Equilibrium considerations and Eq. (13.41)1 then demand that
∂ 2 P
Vr = −D ∇ u=− . (13.42)
∂r 2πr
Since u is a function of r only, this is an ordinary differential equation whose general
solution is
Pr 2 ln(r ) Pr 2
u(r ) = − + A ln(r ) + Br 2 + C, (13.43)
8π D 8π D
where A, B, C are arbitrary constants. These can be determined from the boundary
conditions u(a) = u (a) = 0, u (0) = 0, giving
P 2 a

u(r ) = (a − r 2 ) − 2r 2 ln . (13.44)
16π D r
This result applies strictly only when the force is applied exactly at the centre of
the plate, but it might reasonably be used as a Green’s function analogous to that in
Sect. 2.2.1 as long as typical dimensions of the contact area are much smaller than a.
13.2.2 Indentation by a Rigid Sphere
The displacement (13.44) is bounded at r = 0, but the curvature
∂2u P a

= 1 − ln (13.45)
∂r 2 4π D r
Fig. 13.9 Contact radius b

as a function of applied
force P
is not. It follows that if a rigid sphere of radius R is pressed into the plate by a force
P, there is no value of P sufficiently low for the contact to comprise a single point.
Instead, contact occurs around a circle whose radius b increases with P.
Inside this circle, 0 ≤r < b, the shear force is zero and the displacement must take
the form
u(r ) = C1 + C2 r 2 , (13.46)
where C1 , C2 are as yet unknown constants. This defines a spherical deformed surface
which must conform to the shape of the sphere, since otherwise there would be
violation of the contact inequalities either inside or just outside r = b. We conclude
that C2 = −1/2R. This behaviour is, of course, the analogous to that of the beam of
Fig. 13.1a in the range P > P0 .
Outside the contact circle, b <r < a, equilibrium considerations again give
P
Vr = − , (13.47)
2πr
where P is the total force applied to the sphere, so the general solution is again given
by (13.43). The unknown constants C1 , A, B, C and the unknown contact radius b
are then determined from the conditions u(a) = u (a) = 0 and continuity of u, u , Mrr
at r = b. We obtain 2
PR b a
−1
= 2 − 1 + 2 ln . (13.48)
8π D a b
This relation is illustrated in Fig. 13.9. Notice that the contact radius is very small
[but never zero] at low forces, in which range the expression is dominated by the
logarithmic term, giving

PR a
PR 2
≈ 2 ln or b ≈ a exp − . (13.49)
8π D b 8π D
13.2 Contact of Plates 277
More generally, since concentrated moments cannot occur in a contact region

without adhesion, the moment Mrr must be continuous at any boundary r = c between
contact and separation. If adhesive forces are allowed, Majidi and Adams (2009) have
shown that the minimum total potential energy
√ criterion is equivalent to the inclusion
of a moment discontinuity of magnitude 2Δγ D and appropriate sign at each such
boundary.5
13.3 Membrane Effects
Since beams and plates are by definition ‘thin’, it makes sense to expand the stress
fields as power series in the through-thickness coordinate and retain only the first
few terms. In this context, the uniform axial tensile stress due to a tensile force in a
beam is the first, zeroth-order term of the series, and the second, linear term defines
the bending stress. In plate and shell theory, the zeroth-order term corresponds to a
‘membrane stress’, uniform through the thickness, and the first-order term is again
associated with bending.
In Fig. 13.1, it is clear that the curved length of the beam is greater than the
straight distance between its ends. This will cause the ends of the beam to slide
over the supports, or for the supports to move closer together if this is kinematically
possible, but if horizontal motion at the ends is prevented, an axial force will be
induced which will also stiffen the structure with respect to lateral deflections. This
effect is clearly second order, since changing the sign of the force in Fig. 13.1b would
lead to an equal shortening of the horizontal distance between the ends, so for small
displacements, we are justified in ignoring it. Similar effects arise in plate problems,
but this time the membrane stresses cannot generally be relaxed by leaving the edges
of the plate unrestrained, since the deformed shape of the plate [e.g. in the contact
problem of Sect. 13.2.2] is ‘non-developable’. In other words, it is not a shape that can
be achieved by bending deformations alone, so membrane strains must be developed
regardless of the edge conditions assumed.
Although the membrane strain is second order, the fact that plates and beams are
much stiffer in tension than in bending causes them to become significant at relatively
small deformations—typically when the lateral displacement of a plate exceeds the
plate thickness. Problems where this condition is violated can be formulated using
von Kármán plate theory (Ciarlet 1980), but solutions are difficult to obtain and
numerical solutions are the only practical option in most cases.
5 This assumes that the other body is rigid. For contact between two plates, the appropriate condition
is Δγ = M12 /2D1 +M22 /2D2 , where M1 , M2 are the moment discontinuities in the respective plates
whose stiffnesses are D1 , D2 .
13.3.1 ‘Membrane Only’ Solutions
The relative importance of membrane stresses and bending stresses depends on the
ratio between the lateral displacement and the plate thickness, so that for very thin
plates, an acceptable approximation can be made by neglecting the contribution due
to bending.
To illustrate this procedure, we consider the problem of Fig. 13.10, in which a
circular plane membrane of radius a and thickness h is stretched [like the head of a
drum] by a biaxially isotropic membrane stress σ0 , and then indented at its centre by
a rigid sphere of radius R.
Points in the membrane will experience both radial and lateral displacements
u r , u z , respectively, both of which can be expected to be approximately uniform
through the thickness, and hence be functions of radius r only. The corresponding
membrane strains can be expressed in terms of these displacements as
2
du r 1 du z ur
err = + ; eθθ = (13.50)
dr 2 dr r
Mansfield (1989), Chap. 7, and hence the membrane stresses are

∗ du r
∗ 1 du z 2 νu r
σrr = σ0 + E (err + νeθθ ) = σ0 + E + + (13.51)
dr 2 dr r

∗ ∗ ur du r 1 du z 2
σθθ = σ0 + E (eθθ + νerr ) = σ0 + E +ν + .
r dr 2 dr
(13.52)
These must satisfy the in-plane equilibrium equation
dσrr σrr − σθθ

+ = 0, (13.53)
dr r
Fig. 13.10 A stretched P

circular membrane indented
by a rigid sphere
radius R
ur a
r
uz
13.3 Membrane Effects 279
Fig. 13.11 Free-body P

diagram of the contact region
0 ≤r < b radius R
σb h σb h
and the out-of-plane (lateral) equilibrium equation

σrr σθθ p
+ = , (13.54)
R1 R2 h
where p is the contact pressure and R1 , R2 are the radii of curvature of the membrane
in the radial and circumferential directions, respectively, given by
1 d 2uz 1 1 du z
=− 2 ; =− . (13.55)
R1 dr R2 r dr
The Contact Region

It seems reasonable to assume that contact will occur in some circular region 0 ≤r < b,
in which
du z r d 2uz 1
R1 = R2 = R and =− ; =− . (13.56)
dr R dr 2 R
Substituting these results into (13.51), (13.52) and the resulting stress components
into (13.53), we obtain an ordinary differential equation for u r with solution
(3 − ν)r 3
ur = − + Ar, (13.57)
16R 2
where A is a constant that depends on the membrane stress σrr at the edge of the
contact region r = b. Denoting this value by σb , and substituting the resulting stresses
into Eq. (13.54), we obtain the contact pressure as

h E(b2 − 2r 2 )
p= 2σb + , (13.58)
R 8R 2
corresponding to a total force

b
2πhb2 σb
P = 2π p(r )r dr = . (13.59)
0 R
Thus, although the membrane deformation affects the contact pressure distribution,
it does not affect the total contact force, except insofar as σb differs from the original
tension σ0 . We could have anticipated this result since the membrane must be locally
tangential to the sphere at r = b and equilibrium of the free-body diagram of Fig. 13.11
leads immediately to Eq. (13.59).
The Separation Region
In the separation region b < r < a, the lateral deflection u z (r ) is not known a priori.
Equations (13.53), (13.54) with (13.51), (13.52) provide two conditions for the two
displacement components u r (r ), u z (r ), but the resulting differential equations are
nonlinear and can only be solved by numerical techniques. However, if the initial
biaxial tension σ0 is sufficiently large, we can obtain an approximate solution by
writing σrr = σθθ = σ0 in (13.54), giving

σ0 σ0 d 2uz 1 du z
+ = σ0 + ≈ 0, (13.60)
R1 R2 dr 2 r dr
since there is no contact pressure in this region. In other words, the lateral dis-
placement u z is approximately harmonic. Solving this equation and determining the
resulting arbitrary constants from the condition u z (a) = 0 and continuity of slope
and displacement at r = b, we obtain
b2 a
b2 − r 2
uz = ln + 0≤r <b (13.61)
R b 2R
b2 a
= ln b < r < a. (13.62)

R r
Using this result in (13.51), (13.52), we can then solve (13.53) for u r and hence
determine σrr from (13.51). In particular, we find that the radial stress at r = b is

Eb2 b2 b2 a
σb = σ0 + 3(1 − ν) − (1 − 3ν) + 4(1 + ν) ln ,

16R 2 (1 − ν) a2 a2 b
(13.63)
and if the contact area is small, so b a, this reduces to

3Eb2 2πhb2 3Eb2
σb ≈ σ0 + giving P≈ σ0 + , (13.64)
16R 2 R 16R 2
from (13.59). Equation (13.64)2 can be solved to give the radius of the contact area
b as a function of P, after which the deflection Δ of the sphere is obtained from
(13.61) as

b2 a 1
Δ = u z (0) = ln + . (13.65)
R b 2
13.3 Membrane Effects 281
The initial assumption that the in-plane stresses are approximately equal to σ0 is
justified only if the second term in each of these expressions is small compared with
the first, which in turn requires that
P 32πσ02
. (13.66)
hR 3E
However, we used this assumption only to determine the shape of the deflected mem-
brane and it is likely that the logarithmic shape in (13.62) will be a good approxima-
tion even when (13.66) is not strictly satisfied.
13.4 Contact of Shells
In most cases, contact problems involving shells result in a non-developable deformed

surface, so that both bending and membrane stresses are developed. Very few prob-
lems are tractable analytically, but for the purpose of illustration, we consider the
case of an axisymmetric cylindrical shell of radius a and thickness h reinforced by
a rigid ring of rectangular cross section and length 2b, as shown in Fig. 13.12.
We suppose that the ring has a slightly smaller internal diameter than the outer
diameter of the shell resulting in a radial interference Δ, so that it is assembled onto
the shell by a shrink fit. The anticipated state after assembly is shown in more detail
in Fig. 13.13.
A naïve solution to this problem would assume that contact was maintained over
the entire length of the ring −b < z < b, so that the radial displacement
u r (z) = −Δ − b < z < b. (13.67)
However, the slope and displacement must be continuous functions of z, so the

unloaded segment of shell z > b to the right of the ring [for example] would have as
boundary conditions at z = b
du r
u r (b) = −Δ; (b) = 0. (13.68)
dz
Fig. 13.12 Cylindrical shell 2b

with a shrink-fit
reinforcement ring h
a
rigid
h ring Δ
shell b b
a z
Fig. 13.13 Detail assuming full contact
The shell bending problem so defined is easily shown to require a non-zero bending
moment Mzz at z = b, whereas the segment −b < z < b remains cylindrical and
hence has no bending moment. It follows that full contact could only be sustained
if appropriate concentrated moments were imposed at z = ±b and we argued in
Sect. 13.1 that this is not possible with a non-adhesive unilateral support.
We must conclude therefore that the contact configuration will involve concen-
trated forces at the corners z = ±b and segments of separation adjacent to these
corners. We show two such configurations schematically in Fig. 13.14a, b.
If the half-length b of the reinforcing ring is sufficiently short, there will be
separation throughout −b < z < b as in Fig. 13.14a, in which case the contact forces
comprise two forces F0 per unit circumference at z = ±b. A single radially outward
force of magnitude F per unit circumference applied to an infinite cylindrical shell
at z = 0 can be shown to produce radial displacement u r and axial bending moment
Mzz per unit circumference given by
F F
u r (z) = f 3 (β|z|); Mzz = f 4 (β|z|) (13.69)
8β 3 D 4β
(Barber 2011, Chap. 9), where D is defined in Eq. (13.38),
f 3 (x) = e−x [cos(x) + sin(x)]; f 4 (x) = e−x [cos(x) − sin(x)], (13.70)
and
3(1 − ν 2 )
β= .
4
(13.71)
a2h2
The functions f 3 (x), f 4 (x) are illustrated in Fig. 13.15.

The radial displacement due to radially inward forces F0 at z = ±b can then be
written down by superposition as
F0
u r (z) = − [ f 3 (β|z − b|) + f 3 (β|z + b|)] , (13.72)
8β 3 D
13.4 Contact of Shells 283
(a) (b)
Fig. 13.14 Contact configurations involving separation regions
and the magnitude of these forces is determined by the condition u r (±b) = −Δ,
from which
8β 3 DΔ
F0 = . (13.73)
1 + f 3 (2βb)
The configuration of Fig. 13.14a implies that u r (z) ≤ −Δ everywhere in −b < z < b
and it is clear that the first violation will occur at the mid-point z = 0. Imposing this
condition, we obtain
F0 f 3 (βb)
≥Δ or 2 f 3 (βb) ≥ 1 + f 3 (2βb), (13.74)
4β 3 D
using (13.73). This condition is satisfied if and only if βb < 0.929. Notice that this
result is independent of the radial interference Δ, since the problem is one of the
receding contact [see Chap. 11]. For the same reason, the length of the separation
segment in Fig. 13.14b is also independent of Δ.
If βb > 0.929, the system does not immediately transition to the state illustrated
in Fig. 13.14b, since the deformed shell will have some convex axial curvature at
z = 0 as shown in Fig. 13.14a, so initially we simply get another concentrated force
F1 at a central contact point z = 0.
Once again, the displacements can be written down by superposition giving
F0 F1
u r (z) = − [ f 3 (β|z − b|) + f 3 (β|z + b|)] − 3 f 3 (β|z|), (13.75)
8β 3 D 8β D
and we also record the bending moment
Fig. 13.15 The functions f 3 (x), f 4 (x)

F0 F1
Mzz (z) = − [ f 4 (β|z − b|) + f 4 (β|z + b|)] − f 4 (β|z|). (13.76)
4β 4β
Imposing the condition u r = −Δ at z = −b, 0, b, we obtain
8β 3 DΔ[1 − f 3 (βb)] 8β 3 DΔ[1 + f 3 (2βb) − 2 f 3 (βb)]

F0 = ; F1 = .
1 + f 3 (2βb) − 2 f 3 (βb)2 1 + f 3 (2βb) − 2 f 3 (βb)2
(13.77)
This state will obtain as long as the curvature of the shell is convex outwards at the
contact point z = 0. This in turn requires that the bending moment Mzz (0) remain
positive and hence
2F0 f 4 (βb) + F1 ≤ 0 or βb < 1.187, (13.78)
from (13.76) after substituting for F0 , F1 from (13.77).

If βb > 1.187, there will be a central contact region −c < z < c in which u r = −Δ
as shown in Fig. 13.14b. In this region, u r is constant, implying (i) that there is no
bending moment Mzz and (ii) that the contact pressure is
EhΔ
p= . (13.79)
a2
The bending moment at the ends z = ±c of this region must also be zero and hence
the boundary conditions defining each separation region are the same for all b in
βb > 1.187, including the limiting case βb = 1.187. Thus, the solution for these
regions is identical to that in one half of the solution with a central contact force at
the limiting value βb = 1.187. In particular, the length of each separation region is
1.187/β and the corner forces F0 are given by Eq. (13.77)1 with βb = 1.187. There
will also be concentrated forces F2 at the ends z = ±c of the central continuous
contact region equal to half of F1 in the limiting case. We obtain
F0 = 7.070β 3 DΔ; F2 = 1.193β 3 DΔ. (13.80)
If the reinforced cylinder is now subjected to internal pressure, it is tempting to

conclude that the gaps in the separation regions in [e.g.] Fig. 13.14a would tend to
close. However, this is incorrect. The pressurized problem could be solved by (i)
applying the pressure to the unreinforced cylinder, causing an increase in radius,
followed by (ii) application of the reinforced ring. The radial interference in step (ii)
will be increased by the pressure [beyond the unloaded value Δ], but in other respects,
the correction at step (ii) will be the same as before. We conclude that the extent of
the contact and separation regions will be independent of internal pressure, and more
surprisingly, the gaps in Fig. 13.14a, b will be increased by internal pressure.
This example is illustrative of the fact that comparatively simple contact prob-
lems involving beams, plates or shells can exhibit quite complex and indeed rather
unexpected qualitative behaviour.
13.5 Implications for Finite Element Solutions 285
13.5 Implications for Finite Element Solutions
Only a few relatively simple plate and shell contact problems can be solved analyti-
cally, so most often we shall have recourse to numerical methods, such as the finite
element method. However, it is important to realize that the concentrated forces
exhibited by the solutions in this chapter will correspond to very localized contact
traction distributions in the full continuum solution and these will only be captured
adequately by the finite element method if the mesh in these regions is extremely
fine. This is very difficult to ensure since in many cases, we do not know the location
of the localized contact region a priori. For example, if we wished to develop a finite
element solution of the beam problem of Fig. 13.1 over a range of applied forces P,
we would need to provide a fine mesh at the top surface of the beam at least out to a
point beyond which the contact is not expected to penetrate.
Problems
1. A heavy straight beam of flexural rigidity E I and weight w0 per unit length rests
on a rigid plane foundation. One end is subjected to a moment M0 , as shown in
Fig. 13.16. Find the length a of the segment that separates from the foundation, the
reaction at the end and the concentrated contact force at the separation point.
2. Suppose the moment M0 in Problem 1 and Fig. 13.16 is negative, we then expect
the end of the beam to separate from the foundation. Find the length of the separation
segment and the concentrated contact force at the separation point.
3. A simply supported beam of length L and flexural rigidity E I is indented by a
symmetric rigid body with a fourth power surface defined by the initial gap function
g0 (x) = C x 4 , where C is a constant. Show that contact occurs at two points separated
by a distance a as in Fig. 13.2 and find the relation between a and the indenting
force P.
M0
Fig. 13.16 A heavy beam lifted off a support by an end moment

Fig. 13.17 Two glued

beams separated by a wedge
F α
α
a
b
4. Two identical straight beams each of flexural rigidity E I are glued together. A
frictionless wedge of angle 2α is now driven into the end to separate the joint as
shown in Fig. 13.17. If the cohesion of the glued joint can be defined by an interface
energy Δγ per unit length, use an energy argument to find the relation between the
dimensions a, b where z = a defines the location of the apex of the wedge and z = b is
the separation point. Find the concentrated moment transmitted between the beams
at z = b and show that it is independent of α, a and b. Find also the axial force F
exerted on the wedge.
5. Solve Problem 4 for the case where there is friction between the wedge and the
beams with friction coefficient f .
6. If two carbon nanotubes are almost parallel, van der Waals forces will tend to make
them adhere together. Approximate the solution of this problem by treating each tube
as a curved beam of mean radius R and flexural rigidity E I per unit length [along
the axis of the tube]. Assume that an arc segment 2α in each beam is in contact [and
hence by symmetry is bent into a plane surface] and use an energy argument to find
a relation between α, E I, and the interface energy Δγ per unit area.
7. Use the method of Sect. 13.1.5 to estimate the distribution of contact pressure in
the problem of Fig. 13.3, if the beam cross section is rectangular with height h and
width b and the material has density ρ and Young’s modulus E.
8. A large flexible plate of weight w per unit area rests on a rigid horizontal foundation.
A vertical force P is applied to the plate, causing a circular region of radius a to lose
contact with the foundation as shown in Fig. 13.18. Show that

2P
a= .
πw
Fig. 13.18 A heavy circular

P
plate lifted by a central force
w per unit area
a
Problems 287
Fig. 13.19 Two adhered t

plates separated by a
spherical particle
9. A small rigid spherical particle of radius R is trapped between two identical thin
plates, each of thickness t as shown in Fig. 13.19. Adhesive forces, with interface
energy Δγ cause the plates to adhere to each other at points distant from the particle.
Assuming (i) that the separated region is circular and of radius a, and (ii) that the
contact between the sphere and each plate can be approximated by a concentrated
force P, find the values of P and a in terms of R, Δγ and the stiffness of the plates D.
10. A rigid flat-ended cylinder of radius b is pressed into a circular plate of radius a
by a force P that acts through the point c, θ. The plate is built in at r = a. Assuming
that contact occurs for all θ at r = b, find the force per unit length around this line as
a function of θ and the angle of tilt of the cylinder. What is the maximum value of c
for which this assumption is reasonable?
11. Figure 13.20 shows an elastic membrane of thickness h and modulus E that is
being pulled away from a rigid plane to which it adheres with interface energy Δγ.
If the applied tensile stress is σ and it is applied in a direction inclined at an angle
α to the plane, find the value of σ for which further decohesion can be expected to
occur. Will your solution also apply to the cases (i) α = 0 and (ii) α > π/2?
12. Suppose that the rigid reinforcing ring in Sect. 13.4 has a circular cross section
of radius b Δ. In other words, it is a toroid. Find the extent of the contact area and
the contact pressure distribution, including any concentrated forces that may arise.
13. A frictionless rigid conical wedge is driven into a cylindrical tube by an axial
force F0 , as shown in Fig. 13.21. The wedge angle α 1.
Fig. 13.20 An adhered

membrane pulled away from σ
a plane support
h
α
Fig. 13.21 A frictionless t α

conical wedge pressed into a
cylindrical shell
a
z F0
Assuming that contact occurs only at the end of the tube as shown, find an expres-
sion for the radial displacement of the tube u r as a function of z and F0 . Hence, show
that there is a minimum angle α0 for which contact is restricted to the end and find
its value. The cylinder has a radius a and thickness t and the material has Young’s
modulus E and Poisson’s ratio ν. What will happen if α < α0 ?
Chapter 14
Layered Bodies
There are numerous practical applications in which one or both of the contacting
bodies have a layered structure. For example, a layer of a different material may be
used at the surface of a contacting component in order to enhance surface durability
properties or to prevent oxidation or other chemical degradation, surface layers may
be developed by chemical action as in the case of oxide or contaminant layers in
lubricated systems, manufacturing processes will typically cause differences in the
properties of surface layers due to locally high strain rates and temperatures, and
biological systems often involve thin layers of cartilage or other soft tissue supported
on a harder substrate.
The simplest layered structure comprises a homogeneous layer of constant thick-
ness h and elastic properties El , νl bonded to an infinite substrate of a different elastic
material (E s , νs ), as shown in Fig. 14.1.
For this case, it is instructive to consider three limiting cases depending on the
relative magnitudes of E s , El :-
1. E s El : The substrate is very flexible relative to the layer [e.g. a glass sheet on
a rubber foundation]. An extreme form of this limit is if there is no foundation—
i.e. a plate or a beam supported only at the ends, as in Chap. 13. For small but
non-zero modulus ratios E s /El , the effect of the substrate might be approximated
as a linear elastic foundation in which the normal displacement is proportional to
the local pressure.
2. El = E s , νl = νs : The substrate and the layer have the same elastic properties, in
which case the assembly behaves as a single monolithic block of the same material
and [for example] if the contacting surfaces are quadratic, Hertzian behaviour is
obtained. If the moduli are not identical, but are of the same order, the simple
Hertzian solution will probably still give good results.

https://doi.org/10.1007/978-3-319-70939-0_14
290 14 Layered Bodies
Fig. 14.1 Indentation of a P

layer bonded to a half-space
E l , νl h
E s , νs
z
3. E s El : The substrate is very stiff relative to the layer [e.g. a rubber layer on
a steel surface]. In this case, we can generally get good results by assuming the
foundation to be rigid.
The behaviour of the layered system also depends on the size of the contact
area relative to thickness of the layer. For example, if the force is small enough
for the contact area to be small compared with the layer thickness, there will be an
approximately Hertzian stress distribution in the layer. However, further from the
contact region, we shall see a stress distribution similar to that due to the application
of a point force on the layered elastic system. At the other extreme, if the contact area
is large compared with the layer thickness, the contact pressure will be significantly
modified by the substrate properties.
In this chapter, we shall first consider approximate methods appropriate for the
limiting cases E s El and E s El , respectively, and then discuss ‘exact’ methods
for the more general problem. These approximate solutions are useful for interpreting
qualitative features of more exact solutions even when the modulus ratio is not
sufficiently large or small to justify relying on them alone.
14.1 Es El : Plate on an Elastic Foundation
If a stiff layer of thickness h is supported on a relatively flexible substrate, the contact

force is transmitted laterally through the layer and then gradually to the substrate.
This effect can be approximated by representing the layer as a plate and replacing
the substrate by a linear ‘Winkler’ foundation satisfying an equation of the form
f (x, y)
u z (x, y, 0) = . (14.1)
k
where u z (x, y, 0) is the normal surface displacement of the substrate, f (x, y) is the
compressive traction transmitted to the substrate and k is a foundation modulus. The
plate bending Eq. (13.37) is then modified to
ku p
∇2 ∇2u + = . (14.2)
D D
14.1 E s El : Plate on an Elastic Foundation 291
14.1.1 Choice of Foundation Modulus
Unfortunately, if the substrate is an elastic half-space, there is no unique value for

the appropriate foundation modulus. For example, if we load the half-space by a
sinusoidal load f (x, y) = f 0 cos(mx), the resulting displacement is
2(1 − νs2 ) f 0 cos(mx) Es m

u z (x) = corresponding to k= , (14.3)
Es m 2(1 − νs2 )
from Eqs. (6.65), (6.68). Thus, the appropriate effective modulus for the half-space
approximated as an elastic foundation depends on the range of wavenumbers m of
the loading. The substrate appears stiff to short wavelength loads and flexible to long
wavelength loads.
Notice, however, that if a concentrated line force P per unit length is applied
along the y-axis to a plate on an elastic foundation, the resulting displacement takes
the form
P
u(x) = f 3 (|βx|) , (14.4)
8β 3 D
where f 3 (x) is defined in Eq. (13.70) and β is now given by

k 12El h 3
β= D= .
4
with (14.5)
4D (1 − νl2 )
Recalling that plate and beam problems tend to involve concentrated contact forces,
it therefore seems reasonable to use Eq. (14.3)2 with m replaced by β, which after
some algebra allows us to express the foundation modulus in terms of the material
properties and the layer thickness as

Es E s (1 − νl2 )
k= 3
. (14.6)
4h(1 − νs2 ) 12El (1 − νs2 )
If the contact problem is two-dimensional in the sense of Chap. 6—i.e. if the loading is
independent of the y-coordinate—then Eq. (14.2) reduces to the ordinary differential
equation
d 4u ku p
4
+ = . (14.7)
dx D D
If the contact region A comprises an extended segment b < x < a of the x-axis,
and if the rigid-body indentation d is known so that the normal displacement
u(x) = d −g0 (x) is a known function, then the contact pressure p(x) can be written
down by substitution in Eq. (14.7).
In the adjacent separation regions x > a and x < b, there is no contact pressure
[ p(x) = 0] and hence the displacement must satisfy the homogeneous equation
d 4u ku
4
+ = 0. (14.8)
dx D
The most general solution satisfying these conditions is
u(x) = e±βx [B1 cos(βx) + B2 sin(βx)] , (14.9)
where β is defined in (14.5)1 , B1 , B2 are two arbitrary constants and the sign in
the exponential term must be chosen so as to ensure that u(x) decays with distance
from the loaded region. Writing solutions of the form (14.9) for each of the regions
x > a and x < b gives a total of four arbitrary constants in addition to the as yet
unknown boundaries a, b of the contact area. These can be determined from six
equations obtained by enforcing continuity of displacement, slope and curvature at
each boundary. Notice that continuity of curvature is required because unilateral
contact does not permit the transmission of a concentrated bending moment at the
boundary, as in Sect. 13.1.
In general, we anticipate the occurrence of concentrated forces at the separation
points x = a, b. These will be associated with corresponding discontinuities in shear
force and can be evaluated using the relations
d 2u d 3u
M(x) = −D ; V (x) = −D , (14.10)
dx2 dx3
where M, V are the bending moment and shear force per unit length respectively.
Example: Indentation by a Rigid Cylinder
Consider the problem in which a rigid cylinder of radius R is pressed against a plate
on an elastic foundation by a force P per unit length. As in Sect. 13.1, we anticipate
that initial contact will be restricted to the line x = 0 for 0 < P < P0 , where P0 is the
force required to cause the local curvature of the plate to equal that of the indenter.
Differentiating (14.4) with respect to z, we obtain
d 2u P f 4 (β|x|)
2
=− , (14.11)
dx 4β D
where f 4 (x) is defined in (13.70), and hence the limiting value P0 is given by
P0 1 4β D k
= or P0 = = 3 . (14.12)
4β D R R β R
Fig. 14.2 Plate on an elastic

foundation indented by a
rigid cylinder
For P > P0 , we anticipate the configuration of Fig. 14.2, with concentrated forces
at the ends of the contact region x = ±a, but also a distributed contact pressure p(x)
in −a < x < a. If the rigid-body indentation Δ is known, we have
x2 kx 2
u(x) = Δ − and hence p(x) = kΔ − − a < x < a, (14.13)
2R 2R
from Eq. (14.8). We also note that
a2 a 1
u(a) = Δ − ; u (a) = − ; u (a) = − . (14.14)
2R R R
For the separation region, it is convenient to move the origin to the point x = a [i.e.
x → x−a] so that the corresponding continuity conditions are imposed on Eq. (14.9)
[with the decaying exponential multiplier exp(−βx)] at x = 0. We then have
u(0) = B1 ; u (0) = β(B2 − B1 ); u (0) = −2β 2 B2 (14.15)
and B1 , B2 , a are determined from the three equations
a2 a 1
B1 = Δ − ; β(B2 − B1 ) = − ; −2β 2 B2 = − , (14.16)
2R R R
with solution
1 + 2βa 1 (1 + βa)2
B1 = ; B2 = ; Δ= , (14.17)
2β 2 R 2β 2 R 2β 2 R
the last of which defines the relation between the rigid-body indentation Δ and the
contact semi-width a. There is no shear force in the contact region since u is a
quadratic function of x, so the concentrated force at x = a is
k
P1 = Du (0) = 2Dβ 3 (B1 + B2 ) = (1 + βa), (14.18)
2β 3 R
using (14.5), (14.16). The total applied force can then be written
a
k 2(βa)3
P = 2P1 + p(x)d x = 1 + 2βa + 2(βa) +
2
, (14.19)
−a β3 R 3
after substituting for p(x), Δ from (14.13), (14.17). Notice that P → P0 as a → 0, so

there is a smooth transition from a concentrated force solution to the configuration
with a finite contact area at P = P0 .
Similar techniques can be applied to three-dimensional problems, though the mathe-

matics is generally too complex to make this approach viable. If there is an extended
contact region A where u(x, y) is therefore prescribed, the contact pressure can be
written down using Eq. (14.2). However, we anticipate the occurrence of a concen-
trated line force around the boundary of A and this can only be found by analysing
the displacement in the separation region Ā.
If the loading is axisymmetric, the governing Eq. (14.2) takes the form
d 4u 2 d 3u 1 d 2u 1 du ku p(r )
+ − + 3 + = (14.20)
dr 4 r dr 3 r 2 dr 2 r dr D D
and the solution in Ā [where p(r ) = 0] can be written

√
√
√
√
u(r ) = C1 ber 2βr + C2 bei 2βr + C3 ker 2βr + C4 kei 2βr ,

(14.21)
where β is defined by (14.5) and ber, bei, ker and kei are Kelvin functions.1 If the plate
is infinite in extent and the region r > a is unloaded, the displacement must decay to
zero as r → ∞ and the constants C1 , C2 should then be set to zero. In particular, if
the only loading comprises a concentrated force P at the origin, Eq. (14.21) applies
throughout r > 0 and the remaining constants C3 , C4 can be found by enforcing the
boundary conditions
du ∂ 2 P
(0) = 0; Vr = −D ∇ u→− r → 0. (14.22)
dr ∂r 2πr
We obtain C3 = 0 and
P √
u(r ) = − kei 2βr . (14.23)

4πβ 2 D
1 These
functions are related to Bessel functions and are included in the function libraries of Maple
and Mathematica.
Fig. 14.3 Indentation of a

layer on a rigid foundation
P
G,ν z
h
rigid
Under the force, the displacement is
P
u(0) = , (14.24)
16β 2 D
but the curvature d 2 u/dr 2 is unbounded, so if a rigid sphere of radius R is pressed

against the plate, a contact area of finite radius a must be established for all values
of P, as in Sect. 13.2.2.
Just outside the contact circle, we must have
a2 du a d 2u 1 P
u =Δ− ; =− ; = ; Vr = − r = a+ (14.25)
2R dr R dr 2 R 2πa
and these conditions provide four equations for the contact radius a, the indentation
depth Δ and the constants C3 , C4 in Eq. (14.21).
14.2 Es El : Layer on a Rigid Foundation
We now turn our attention to the opposite extreme, where the substrate is much stiffer
than the foundation and hence can be approximated as rigid, as shown in Fig. 14.3.
Typical engineering applications include rubber vibration-absorbing layers on steel
components and cartilage layers attached to bones.
We measure the normal coordinate z from the free surface of the layer, and denote
the normal displacement of this surface u z (x, y, 0) as w(x, y). We also combine the
in-plane [horizontal] displacement components into a vector u = {u x , u y }.
Johnson (1985) discusses three limiting two-dimensional cases in which the linear
dimensions of the contact area are large compared with the layer thickness:-
1. If the layer is free to slide on the foundation without friction, there will be some in-
plane displacement u that is assumed to be independent of the vertical coordinate
z. The layer is everywhere loaded in uniaxial compression and it therefore acts
as a linear Winkler foundation.
2. If the layer is bonded to the substrate, the in-plane displacement at the interface
z = h is zero for all x, y. If Poisson’s ratio ν is not too close to 0.5, so the material
is ‘compressible’, a simple approximation can be obtained by assuming u = 0

throughout the layer [i.e. for all x, y, z], in which case we simply have constrained
uniaxial compression. The layer again acts as a Winkler foundation, but with a
higher foundation modulus due to the constraint.
Notice that in the exact solution with either case 1 or 2, if we load the layer
with [say] a uniform pressure over some area A, we shall obtain constant normal
surface displacement w(x, y) throughout most of A, but there will be a transition
region around the boundary of A comparable in width to the layer thickness,2
beyond which w = 0. This contrasts with the half-space case, where the surface
displacement decays with 1/r in the unloaded region.
3. If ν → 0.5, the layer material becomes incompressible and the foundation modulus
for case 2 becomes infinite. Indentation of the layer then requires that material
should flow out of the loaded region, but this is only possible if u is non-zero and
hence also varies with z, since u(x, y, h) = 0.
We shall analyse each of these cases in turn and also generalize them to three
dimensions.
14.2.1 Frictionless Unbonded Layer
Following Johnson (1985), we assume that plane sections remain plane and hence
that u is a function of (x, y) only. It follows that the in-plane components of strain

∂u x ∂u y 1 ∂u y ∂u x
ex x = ; e yy = ; ex y = + (14.26)
∂x ∂y 2 ∂x ∂y
are also independent of z, whilst the only non-zero out-of-plane strain is

w
ezz = − . (14.27)
h
The in-plane stress components are therefore

2Gν ∂u x ∂u y w ∂u x
σx x = + − + 2G (14.28)
(1 − 2ν) ∂x ∂y h ∂x

2Gν ∂u x ∂u y w ∂u y
σ yy = + − + 2G (14.29)
(1 − 2ν) ∂x ∂y h ∂y

∂u y ∂u x
σx y =G + , (14.30)
∂x ∂y
2 Johnson’s approximate theory does not capture these transition regions and hence is only appropri-
ate when they constitute only a small fraction of the loaded area—i.e. when the linear dimensions
of the contact area are large compared with h.
14.2 E s El : Layer on a Rigid Foundation 297
where G = E/2(1+ν) is the modulus of rigidity [shear modulus].

Substituting into the in-plane equilibrium equations
∂σx x ∂σx y ∂σx z ∂σ yx ∂σ yy ∂σ yz

+ + = 0; + + = 0, (14.31)
∂x ∂y ∂z ∂x ∂y ∂z
and cancelling some non-zero factors, we obtain

∂ ∂u x ∂u y 2ν ∂w
+ + (1 − 2ν)∇ 2 u x =
∂x ∂x ∂y h ∂x

∂ ∂u x ∂u y 2ν ∂w
+ + (1 − 2ν)∇ 2 u y = , (14.32)
∂ y ∂x ∂y h ∂y
and a particular solution of these equations can be found in terms of a scalar potential
function φ(x, y), such that
∂φ ∂φ νw
ux = ; uy = and hence ∇2φ = , (14.33)
∂x ∂y (1 − ν)h
from (14.32). The contact pressure is then recovered as

2Gν ∂u x ∂u y w w
p(x, y) = −σzz = − + − + 2G
(1 − 2ν) ∂x ∂y h h
2Gν 2G(1 − ν)w 2Gw(x, y)
=− ∇2φ + = , (14.34)
(1 − 2ν) (1 − 2ν)h (1 − ν)h
showing that the layer acts as a linear Winkler foundation, with the contact pressure
being proportional to local indentation w.
The indentation w is also the local interpenetration between the indenter and the
layer in the undeformed shape. Thus, if we assume that the contact area is identical
with the interpenetration area, we shall obtain positive contact pressures throughout
the contact area and positive gaps elsewhere. This therefore defines the correct solu-
tion of the unilateral contact problem. Also, for non-conformal contact, we shall find
that w → 0 at the edge of the contact area and hence (14.34) defines a pressure which
satisfies p → 0 at the edge of the contact area.
Example: Indentation by a Rigid Sphere
If a spherical indenter of radius R is pressed into the layer so as to give a central
indentation Δ, we have
r2
w(r ) = Δ − , (14.35)
2R
and the radius a of the contact area is determined from the condition w(a) = 0 as
√
a= 2RΔ. (14.36)
The contact pressure distribution is

2G r2 G(a 2 − r 2 )
p(r ) = Δ− = (14.37)
(1 − ν)h 2R (1 − ν)h R
from (14.34), and the total force is

a
πGa 4 2πG RΔ2
P = 2π p(r )r dr = = , (14.38)
0 2(1 − ν)h R (1 − ν)h
using (14.36).
14.2.2 Bonded Compressible Layer
If the layer is bonded to the substrate, the in-plane displacement u = 0 at z = h. If the

linear dimensions of the contact area are large compared with the layer thickness, u
will therefore be small throughout the layer and the in-plane strains ex x , ex y , e yy will
generally be negligible (Johnson 1985). It then follows that the contact pressure is
2G(1 − ν)w(x, y)
p(x, y) = −σzz = . (14.39)
(1 − 2ν)h
Once again, we find that the layer acts as a linear Winkler foundation, but with
an increased modulus relative to the frictionless case (14.34). Indeed, the constant
of proportionality in (14.39) becomes unbounded in the limit of an incompressible
material, for which ν = 0.5, so this solution is appropriate only for materials for
which Poisson’s ratio is not too close to 0.5.
14.2.3 Bonded Incompressible Layer
For the incompressible or almost incompressible material [(0.5−ν) 1], indentation

is possible only if the material can flow laterally from under the punch, and the result-
ing in-plane displacement cannot be uniform since it is zero at the layer–substrate
interface.
We assume that the contact is frictionless, so [for example] the shear stress com-
ponent
∂u x ∂u z
σzx = 2Gezx = G + (14.40)
∂z ∂x
G,ν z
h
rigid
Fig. 14.4 In-plane displacement of initially vertical planes [shown dotted] for a bonded incom-
pressible layer
must be zero on the contact plane z = 0. We also assume that the slope of the indenter
∂w ∂u z
= (14.41)
∂x ∂x
is negligibly small and hence, using (14.40),
∂u x
=0 at z = 0. (14.42)
∂z
The in-plane displacement u(x, y, z) must satisfy this condition and a similar equa-
tion for u y , and it is also required to be zero throughout the plane z = h, since the
layer is bonded to a rigid foundation. These conditions are satisfied by the quadratic
approximation u = (z 2 −h 2 )∇ψ, or
∂ψ ∂ψ
u x = (z 2 − h 2 ) ; u y = (z 2 − h 2 ) , (14.43)
∂x ∂y
where ψ is a scalar function of x and y. The form of this displacement profile is

illustrated in Fig. 14.4.
With this approximation, the in-plane equilibrium equations can only be satisfied
in the weak sense [in terms of force resultants] as

h
∂σx x ∂σx y h
∂σ yx ∂σ yy
+ dz + τx = 0; + dz + τ y = 0, (14.44)
0 ∂x ∂y 0 ∂x ∂y
where τx = σzx (x, y, h), τ y = σzy (x, y, h) are the components of shear traction at
the layer–substrate interface and are given by
∂ψ ∂ψ
τx = 2Gh ; τ y = 2Gh , (14.45)
∂x ∂y
from (14.40), (14.43), since u z = 0 at the interface.

The constitutive law for an incompressible elastic material is
σi j = σ̄δi j + 2Gei j , (14.46)
where σ̄ is a scalar function of position, representing a state of hydrostatic ten-

sion. Substituting (14.43) into (14.26) and the resulting expressions and (14.45) into
(14.46) and (14.44), we obtain
4Gh 2
∇ σ̄ − ∇∇ 2 ψ + 2G∇ψ = 0. (14.47)
3
Also, the incompressibility condition demands that
h h
(ex x + e yy )dz = − ezz dz = w, (14.48)
0 0
from (14.27) and hence, using (14.43),
2h 3 2 3w
− ∇ ψ=w or ∇2ψ = − . (14.49)
3 2h 3
We can use this result to substitute for the second term in Eq. (14.47), from which
we deduce that w
σ̄ = −2G ψ + . (14.50)
h
We can then recover the contact pressure as

2Gw 2w 4h 2 2
p(x, y) = −σ̄ + = 2G ψ + = 2G ψ − ∇ ψ . (14.51)
h h 3
In the final expression, the second term is of order (h/a)2 compared with the first.
where a is a representative dimension of the contact area. It can therefore be neglected
in the approximate solution, giving the simple expression
p = 2Gψ. (14.52)
More generally, the stress field is dominated by the hydrostatic term σ̄ ≈ −2Gψ,
where ψ is determined from (14.49) with appropriate conditions at the edge of the
contact area.
Edge Conditions
As before, we anticipate that p → 0 at the edge of the contact area, which here
implies that ψ → 0. However, in this case, an additional boundary condition is
needed because the material displaced from under the indenter will bulge upwards
and hence the edge of the contact area is not coincident with the point where w = 0.
Johnson (1985) argues that there can be no change in the total volume of material
under the indenter, which is equivalent to the assumption that none of the material
under the indenter is pushed into the separation region—i.e. that
∂ψ
u n = −h 2 = 0, (14.53)
∂n
where u n is the component of in-plane displacement normal to the edge of the contact
area. This is not exactly correct, but it can be shown that for non-conformal contact
problems, the correction is of the order of the ratio (h/a).
Spherical Indentation Revisited
For the problem defined by Eq. (14.35), ψ is determined from the equation

∂2ψ 1 ∂ψ 1 ∂2ψ 3 r2
∇ ψ=
2
+ + =− 3 Δ− , (14.54)
∂r 2 r ∂r r 2 ∂θ2 2h 2R
from (14.49). The general axisymmetric solution of this equation is

3 Δr 2 r4
ψ=− − + C1 ln(r ) + C2 , (14.55)
2h 3 4 32R
and the constant C1 must be zero to preserve continuity at the origin. At the edge of
the contact area, r = a, the boundary conditions
∂ψ
ψ = 0; =0 (14.56)
∂r
from (14.52), (14.53) yield
Δa 2 a4 Δa a3
− + C2 = 0; − = 0, (14.57)
4 32R 2 8R
with solution
√ Δ2 R
a = 2 ΔR; C2 = − . (14.58)
2
Substituting into Eq. (14.55) and using (14.52), we then obtain the contact pressure
as
3G(a 2 − r 2 )2 3G(4ΔR − r 2 )2
p(r ) = = , (14.59)
32h 3 R 32h 3 R
and the total force as
a
πGa 6 2πGΔ3 R 2
P = 2π p(r )r dr = = . (14.60)
0 32h 3 R h3
Fig. 14.5 Indentation of a layer on a rigid foundation
Notice that the contact radius for a given indentation Δ is greater by a factor
√
of 2 compared with the ‘compressible’ solution (14.36). This arises because the
material displaced from beneath the punch bulges outwards in the surrounding region
as shown in Fig. 14.4. By contrast, in the Hertzian case [Sect. 5.2 and Eq. (12.71)],
the tractions within the contact area cause indentation outside the contact area, which
is therefore smaller than the area that would interpenetrate in the absence of elastic
deformations.
The contact pressure distribution (14.59) is compared with the compressible case
(14.37) and with the Hertzian distribution in Fig. 14.5. The asymptotic arguments
of Chap. 10 show that in an exact solution, the curves must exhibit a square-root
bounded behaviour at the contact-separation boundary, but this occurs in a region
that is small compared with h [the smallest geometric length scale in the problem]
and hence is not captured by the ‘thin-layer’ approximation.
The contact pressures in Fig. 14.5 have been normalized by the central value p(0)
which for the purposes of comparison can be written

Δ
p(0) = C G , (14.61)
R
where
4
C= Hertz
π(1 − ν)
√
2 RΔ
= frictionless compressible layer (14.62)
(1 − ν) h
√ 3
3 RΔ
= bonded incompressible layer.
2 h
Thus,
√ the contact pressure is strongly dependent on the dimensionless parameter
h/ RΔ, particularly in the bonded incompressible case.
Johnson’s approximate solutions predict very different behaviour for bonded com-
pressible and incompressible layers, but it is clear that this change must occur over
some finite range of Poisson’s ratio. Greenwood and Barber (2012) used a Fourier
transform solution3 to solve the two-dimensional problem of a bonded elastic layer
indented by a rigid cylinder, and found that this transition occurs over a very limited
range of ν, with results for ν = 0.45 being essentially indistinguishable from the
bonded compressible results of Sect. 14.2.2 as long as the thin-layer condition [e.g.
h/a 1] is satisfied.
14.2.4 Flat Punch Problems
We can also use Johnson’s approximations to treat problems where the layer is
indented by a flat rigid punch, in which case the contact area is determined by the
planform of the punch. For the compressible layer [bonded or unbonded], we must
relax the condition that the pressure at the edge of the contact area be zero and the
resulting contact pressure will clearly be uniform throughout the contact area.
For the incompressible bonded layer, we can argue as before that the surround-
ing material offers negligible restraint to lateral expansion, so that the hydrostatic
component σ̄ → 0 at the boundary. It therefore follows that the contact pressure still
tends to be zero at the boundary despite the sharp edge of the punch. However, we
clearly have to relax the assumption that the volume of material under the punch is
constant since otherwise there could be no indentation. In practice, a bulge of the
layer material will be developed just outside the contact region.
The problem therefore reduces to the determination of a function ψ satisfying the
equation
3Δ
∇2ψ = − 3 (14.63)
2h
under the indenter and equal to zero on the boundary. The problem is formally
equivalent to the determination of the Prandtl stress function in the Saint Venant
torsion problem (Barber 2010, Sect. 16.1) and hence the solution to many problems
can be written down. In particular, it can be shown that the total indenting force is

3G K Δ
P = 2G ψdΩ = , (14.64)
Ω 4h 3
where G K is the torsional stiffness of a bar of the same cross section Ω as the rigid
punch.
For the particular case of a cylindrical flat punch of radius a, we obtain
3GΔ(a 2 − r 2 ) 3πGΔa 4
p(r ) = ; P= . (14.65)
4h 3 8h 3
3 See Sect. 14.4 below for details of this approach.

14.2.5 Frictional Problems
We assumed in Sect. 14.2.1 that the unbonded layer was free to slide over the rigid
support, but in practice, we must anticipate some frictional resistance. At the other
extreme, if friction is sufficient to prevent relative motion, the solution will be given
by the bonded solution of Sect. 14.2.2 [or Sect. 14.2.3 if ν = 0.5]. In particular, we
can use these latter solutions to determine the shear tractions at the layer–substrate
interface and hence find the minimum coefficient of friction needed to prevent motion.
For the compressible case, the condition of zero in-plane strains implies that
νσzz 2Gνw(x, y)
σx y = 0; σx x = σ yy = =− (14.66)
(1 − ν) (1 − 2ν)h
from (14.39), and hence substituting into the equilibrium Eq. (14.44)
2Gν ∂w 2Gν ∂w
τx = ; τy = . (14.67)
(1 − 2ν) ∂x (1 − 2ν) ∂ y
The minimum coefficient of friction sufficient to prevent slip is therefore
|τ | νh|∇w|
f ≥ f min = = (14.68)
p (1 − ν)w
using (14.39), and this will be small almost everywhere, since |∇w| is of the order
wmax /a where a is a measure of the linear dimensions of the contact area and a h
ex hypothesi. We conclude that in most practical cases, the bonded [or in this case
‘stick’] assumption of Sect. 14.2.2 is appropriate. In non-conformal contact problems,
we must anticipate a region of slip adjacent to the edge of the contact area where w
and hence p tend to zero, but the width of this region will generally be comparable
with or smaller than the layer thickness h.
14.2.6 Effect of Adhesive Forces
Applications of the preceding results often involve polymer layers [e.g. a rubber
layer attached to a steel substrate] or biological materials, such as a cartilage layer
attached to a [relatively rigid] bone. In these cases, adhesive contact forces can be
significant.
We saw in Sects. 14.2.1 and 14.2.2 that the layer acts as a simple Winkler foun-
dation if the material is compressible, or if the layer is free to slip over the support.
In particular, we have

Ew(x, y)
p(x, y) = , (14.69)
h
where
= 2G = 2G(1 − ν)
E [frictionless] ;E [bonded] (14.70)
(1 − ν) (1 − 2ν)
from Eqs. (14.34), (14.39). In this case, an adhesive force law such as the Lennard-
Jones law of Eq. (12.12) acts merely as a nonlinear spring in series with the Winkler
spring. The Lennard-Jones gap g is given by
hp(x, y)
g(x, y) = g0 (x, y) + ε − Δ + , (14.71)

E
where we have assumed that ε is the equilibrium gap at which the traction is zero.
Since p(x, y) = −σ(x, y), we can then use Eq. (12.12) to eliminate p(x, y), giving
a nonlinear equation for g at each point (x, y).
As in Chap. 12, a simpler solution can be obtained if the range of interatomic forces
is small compared with the elastic displacements of the layer. In this case, Eq. (14.69)
remains unchanged, but the boundary condition at the edge of the contact region for
non-conformal problems is changed from

2 EΔγ
p=0 to p=− (14.72)
h
(Yang 2006, Argatov et al. 2016), where Δγ is the interface energy per unit area.
This is analogous to the JKR solution in half-space problems, where adhesive effects
simply introduce a non-zero stress intensity factor at the edge of the contact region.
Adhesion of a Bonded Incompressible Layer
For a bonded compressible layer, Yang (2002) and Argatov et al. (2016) show that a
similar limiting solution can be obtained, in which the displacements and tractions
are still given by Eqs. (14.43), (14.49) and (14.52), and p → 0 at the boundary of the
contact area, but condition (14.53) is modified to

∂ψ 3Δγ ∂p 6GΔγ
= or equivalently = , (14.73)
∂n 2Gh 3 ∂n h3
where n is the outward normal to the contact area.

If the contact area is a circle of radius a, the effect of adhesion is simply to
superpose a multiple of the flat punch Eq. (14.65). This will satisfy (14.73) if

2 2h 3 Δγ πa 3 6GΔγ
Δ=− and hence P =− . (14.74)
a 3G 4 h3
The complete solution is then obtained by adding the corresponding results for the
problem without adhesion. For example, if the indenting body is a sphere of radius
Fig. 14.6 Force–

displacement relation for a
rigid sphere indenting a
bonded incompressible layer
R, we add results from Eqs. (14.58), (14.60) to obtain

πGa 6 πa 3 6GΔγ a2 2 2h 3 Δγ
P= − ; Δ = − . (14.75)
32h 3 R 4 h3 4R a 3G
As in Sect. 12.2.2, these results can be simplified by defining appropriate dimension-

less parameters as
P β2Δ βa
P̂ = ; Δ̂ = ; â = , (14.76)
π RΔγ R R
where 1/6
3G R 4
β= (14.77)
2h 3 Δγ
is modified from the half-space definition (12.32). The dimensionless results are then
â 6 â 3 â 2 2
P̂ = − ; Δ̂ = − , (14.78)
48 2 4 â
and these are plotted parametrically in Fig. 14.6 to yield the force–displacement
curve. In particular, the pull-off force is obtained as
F = −Pmax = 3π RΔγ. (14.79)
It is interesting to note that this is independent of elastic modulus and layer thickness,
but is exactly twice the corresponding value for the elastic half-space. Of course, the
present asymptotic solution is valid only when a h.
We also note that in contrast to the JKR half-space solution, the contact force P̂ is
a single-valued function of the indentation Δ̂. Thus, there is no equivalent of jump-in
and jump-out of contact under displacement control, and hence also no hysteretic
energy loss during a contact–separation cycle.
14.3 Winkler Layer on an Elastic Foundation
If the layer is very thin relative to the dimensions of the contact area, the elastic
deformation of the supporting half-space may influence the contact pressure distrib-
ution significantly, even with E s El . In such cases, it is reasonable to assume that
layer acts as a Winkler layer for all values of ν, since on the scale of the thickness,
the contact pressure will be a slowly varying function of position and hence lateral
motion of the layer material as in Fig. 14.4 will be negligible.
The normal surface displacement of the upper surface is given by u z (x, y, 0)+
w(x, y), where u z (x, y, 0) is the normal surface displacement of the half-space, so
if the layered body is indented by a rigid body with initial gap function g0 (x, y), the
complete contact problem is defined by the conditions
u z (x, y, 0) + w(x, y) = Δ − g0 (x, y) (x, y) ∈ A (14.80)

p(x, y) = 0 (x, y) ∈
/A (14.81)

p(x, y)d xd y = P, (14.82)
A
where A is determined by the inequalities
u z (x, y, 0) ≥ Δ − g0 (x, y) (x, y) ∈

/A (14.83)
p(x, y) > 0 (x, y) ∈ A, (14.84)
and we have used the fact that, for a Winkler layer, w(x, y) = 0 when p(x, y) = 0.
If the layer is linear with modulus k—i.e. if
p(x, y) = kw(x, y), (14.85)
the problem of Eqs. (14.80), (14.80) reduces to the determination of a function p(x, y)
in A, such that

p(x, y) 1 p(ξ, η)dξdη
+ = Δ − g0 (x, y), (14.86)
k πE∗ A (x − ξ)2 + (y − η)2
∗
where the composite modulus E is constructed from the elastic properties of the
half-space and the indenting body.
If the Winkler stiffness k is not too large, one approach to the solution is to write
Eq. (14.86) in the form

1 p(ξ, η)dξdη
p(x, y) =⇒ k Δ − g0 (x, y) − , (14.87)
πE∗ A (x − ξ)2 + (y − η)2
∗
and iterate on p(x, y) starting from the ‘rigid’ solution E → ∞. Alternatively, with
a suitable discretization of the contact area A, Eq. (14.86) can be reduced to a set of
linear algebraic equations in the unknowns pi , representing the nodal values of the
contact pressure.
14.3.1 Nonlinear Layers
In many applications, the layer material may be a polymer or a biomaterial with

significantly nonlinear characteristics. However, the same iterative technique can be
used, provided we can express the constitutive property of the layer in the generalized
Winkler form p = f (w), where f (·) is a known function. We then simply replace
(14.87) by

1 p(ξ, η)dξdη
p(x, y) =⇒ f Δ − g0 (x, y) − . (14.88)
πE∗ A (x − ξ)2 + (y − η)2
Two other important cases leading to the same mathematical structure are those
in which (i) van der Waal’s or Lennard-Jones adhesive forces are taken into account
in problems for the half-space [see Sect. 12.5.3], and/or (ii) the contacting surfaces
are rough with wavelengths much smaller than the typical contact area dimensions
[see Sect. 16.9.1].
These problems share one feature which complicates the numerical solution of the
integral equation (14.86), in that the incremental stiffness of the equivalent Winkler
layer is orders of magnitude larger under high nominal pressures than under light
pressures [or tensile tractions in the case of adhesion]. This implies that in regions
where the contact pressure p(x, y) is high, the iterative scheme (14.88) may diverge.
In other words, each small correction to the elastic displacements on the right-hand
side may cause a large change in the estimate of p(x, y) and may even carry the
function into physically impossible ranges.
In such cases, one could try inverting the iteration—in other words, find the contact
pressure under the assumption that the layer is rigid, then adjust the layer thickness
based on this pressure distribution and use it to modify the condition for the substrate
contact problem. However, this iteration might itself diverge in regions where the
magnitude of the contact tractions is small.
14.4 Fourier Transform Methods 309
14.4 Fourier Transform Methods
More general problems involving layered linear elastic bodies are most conveniently
formulated in terms of Fourier transforms. If a sinusoidal contact pressure p(x) =
p0 cos(ωx) is applied to the surface, the stresses and displacements in each layer can
be written in terms of the potential function solution in Appendix A, Sect. A.3, with
φ(ω, x, z) = [C1 cosh(ωz) + C2 sinh(ωz)] cos(ωx)

ψ(ω, x, z) = [C3 cosh(ωz) + C4 sinh(ωz)] cos(ωx), (14.89)
where C1 , C2 , C3 , C4 are arbitrary constants. It is easily verified by substitution that

these functions are harmonic for all wavenumbers ω.
More general, two-dimensional potential functions can then be written by super-
position over different wavenumbers in the form
∞
φ(x, z) = [C1 (ω) cosh(ωz) + C2 (ω) sinh(ωz)] cos(ωx)dω
0 ∞
ψ(x, z) = [C3 (ω) cosh(ωz) + C4 (ω) sinh(ωz)] cos(ωx)dω, (14.90)
0
where C1 , C2 , C3 , C4 are now arbitrary functions of ω which can be interpreted as

the Fourier transform parameter.
14.4.1 Elastic Layer Bonded to a Rigid Foundation
We illustrate this technique for the simple case of an elastic layer bonded to a rigid
foundation, which was treated by approximate methods in Sect. 14.2.2. We define the
coordinate system4 so that the layer occupies the region 0 < z < h, with the bonded
interface being z = 0.
If the free surface of the layer z = h is loaded by a normal pressure p(x) =
p0 cos(ωx), the problem is defined by the boundary conditions
u x (x, 0) = u z (x, 0) = 0; σzz (x, h) = − p0 cos(ωx); σzx (x, h) = 0. (14.91)
Substituting (14.89) into Eq. (A.7), we obtain
4 This differs from the coordinate system used in Sect. 14.2.2. It is chosen because the homogeneous
displacement conditions at z = 0 then simplify the resulting algebra.

2Gu x = −ω [C1 cosh(ωz) + C2 sinh(ωz)

+ z {C3 cosh(ωz) + C4 sinh(ωz)}] sin(ωx)
2Gu z = ω [C1 sinh(ωz) + C2 cosh(ωz) + z {C3 sinh(ωz) + C4 cosh(ωz)}
− (3 − 4ν) {C3 cosh(ωz) + C4 sinh(ωz)}] cos(ωx), (14.92)
and the first two [displacement] boundary conditions (14.91) then require that
C1 = 0; ωC2 = (3 − 4ν)C3 . (14.93)
Substituting (14.89) into (A.8) and using the other two [traction] boundary con-
ditions, to solve for the remaining constants, we obtain
2 p0
C3 = [(1 − 2ν) sinh(ωh) − ωh cosh(ωh)]
ω f (ωh)
2 p0
C4 = [2(1 − ν) cosh(ωh) + ωh sinh(ωh)] , (14.94)
ω f (ωh)
where
f (ζ) = (3 − 4ν) cosh(2ζ) + 2ζ 2 + (5 − 12ν + 8ν 2 ). (14.95)
Substituting these results back into (14.92) and setting z = h, we obtain the normal
surface displacement of the layer as
2 p0 (1 − ν 2 )g(ωh)
u z (x, h) = − cos(ωx), (14.96)
Eω f (ωh)
where
g(ζ) = (3 − 4ν) sinh(2ζ) − 2ζ. (14.97)
and we have also used the relation G = E/2(1+ν).

Green’s Function for the Bonded Layer
If the bonded elastic layer is subjected to a more general contact pressure distribution
p(x) [even in x], we can define the Fourier cosine transform p̃(s) and its inversion
through
∞ ∞
2
p̃(s) = p(x) cos(sx)d x; p(x) = p̃(s) cos(sx)ds. (14.98)
0 π 0
Thus, a general even pressure distribution p(x) can be regarded as the superposition
of a set of cosine waves and the surface displacement for each of these waves is
defined by (14.96). It follows that the surface displacement due to p(x) is
∞
4(1 − ν 2 ) g(sh) p̃(s) cos(sx)ds
u z (x, h) = − . (14.99)
πE 0 s f (sh)
For the special case where the half-space is loaded by a concentrated compressive
force P, we have p(x) = Pδ(x) and substitution in (14.98)1 yields5
∞
P
p̃(s) = P δ(x) cos(sx)d x = . (14.100)
0 2
It is clear that the only length scale in this problem is the layer thickness h, so it is
convenient to define the dimensionless parameters ξ = x/ h, ζ = z/ h and t = sh. Sub-
stitution of (14.100) in (14.99) then yields the corresponding surface displacement
as

2P(1 − ν 2 ) ∞ g(t) cos(tξ)dt
u z (ξ, 1) = − , (14.101)
πE 0 t f (t)
(Hannah 1951).
This integral can be evaluated by summing the residues inside an appropriate
contour (Greenwood and Barber 2012), leading to a series representation that is good
for numerical calculations at large ξ. However, the Green’s function for the half-plane
of Eq. (6.5) can be regarded as the limit of Eq. (14.101) as the layer thickness h → ∞,
and hence we must anticipate a logarithmic singularity as ξ → 0. This singularity
is associated with the behaviour of the integrand at large values of z and a simple
term with the appropriate asymptotic form can be subtracted out and integrated in
closed form (Greenwood and Barber 2012), leaving a bounded integral that can be
evaluated numerically or as a power series.
Figure 14.7 shows the dimensionless surface displacement for ν = 0.3 and ν = 0.5.
Notice that for the incompressible case ν = 0.5, the displacement becomes positive
Fig. 14.7 Normal surface displacement of the layer u z (x, h) due to a concentrated normal force P
5 Notice that the method used here assumes that the traction p(x) is even in x and hence only half
of the concentrated force P lies in the range 0 < x < ∞. This complication can be avoided by using
the exponential Fourier transform of Eq. (14.102), but the final result is the same as that given here.
at large ξ showing that an inward force causes the layer to bulge outwards in this
range. This effect is much less pronounced for ν = 0.3. The behaviour at large ξ in
both cases is dominated by a single pole t1 on the imaginary axis, corresponding to
simple exponential decay. This contrasts with the case where the layer is not bonded
to the foundation, where oscillatory decay is obtained, similar to that for beams on
elastic foundations.
More General Pressure Distributions
If the contact pressure p(x) is not even in x, we must use the exponential Fourier
transform defined by
∞ ∞
1
p̃(s) = p(x) exp(−ısx)d x; p(x) = p̃(s) exp(ısx)d x. (14.102)
−∞ 2π −∞
The corresponding normal surface displacement for the bonded layer is then given
by
(1 − ν 2 ) ∞ g(sh) p̃(s) exp(ısx)ds
u z (x, h) = − . (14.103)
πE −∞ s f (sh)
We can also define a general three-dimensional pressure distribution using the

double Fourier transform
∞ ∞
p̃(s, t) = p(x, y) exp {−ı(sx + t y)} d xd y
−∞ −∞
∞ ∞
1
p(x, y) = p̃(s, t) exp {ı(sx + t y)} dsdt, (14.104)
4π 2 −∞ −∞
for which the normal surface displacement is

∞ ∞
(1 − ν 2 ) g(ρh) p̃(s, t) exp {ı(sx + t y)} dsdt
u z (x, y, h) = − , (14.105)
2π 2 E −∞ −∞ ρ f (ρh)
√
where ρ = s 2 +t 2 .
Contact Problems
If the elastic layer is indented by a two-dimensional rigid body of known profile,
the contact problem can be formulated in the Fourier transform domain, leading in
general to a pair of dual integral equations, one for the contact region and one for the
separation region. Methods for solving these equations are discussed by Gladwell
(1980), but the resulting expressions are sometimes difficult to evaluate, particularly
in the limiting case ν = 0.5.
An alternative approach is to use the methods of Chap. 6 with the Green’s function
of Eq. (14.101). If the contact pressure distribution is approximated by a series of
Chebyshev polynomials as in Eq. (6.54), the contribution from the singular part of
the Green’s function can be written down using Eq. (6.55). The contribution from
the remaining bounded integral can then be evaluated numerically, after which the
coefficients in the series can be chosen to define the profile of the indenter to any
desired degree of accuracy (Greenwood and Barber 2012).
14.4.2 Multilayered Bodies
For a layered structure with N layers, the solution of Sect. A.3 and Eq. (14.89) can be
applied separately to each layer, with appropriate constants and elastic properties. For
each sinusoidal term, there are therefore 4N arbitrary constants that are determined
from (i) two boundary conditions at the upper surface, (ii) four continuity conditions
at each of N −1 interfaces and (iii) two conditions at the lower surface. If the lowest
layer is of infinite thickness (a half-plane), these latter conditions are replaced by the
requirement that stresses be bounded as |z| → ∞. The continuity conditions at the
interface z j between the jth and ( j + 1)th layers are
( j) ( j+1) ( j) ( j+1)
σzz (z j ) = σzz (z j ); σzx (z j ) = σzx (z j )
u (x j) (z j ) = u (x j+1) (z j ); u (z j) (z j ) = u (z j+1) (z j ). (14.106)
This procedure enables us to determine the multilayer equivalent of Eq. (14.96)—i.e.

the sinusoidal surface displacement due to a given amplitude p0 of sinusoidal loading.
Generalization to the determination of the Green’s function or the displacement due
to any prescribed traction p(x) then proceeds as in Sect. 14.4.1.
14.5 Functionally Graded Materials
Layered structures can experience stress concentrations due to the discontinuity in

material properties between adjacent layers and hence may be susceptible to delam-
ination and other failure modes. These problems can sometimes be mitigated by
developing a functionally graded material [FGM] with continuously variable prop-
erties, for example, by generating a smooth spatial variation of the concentration of
components in an alloy or mixture. In the present context, this would correspond to a
half-space or a layer in which the elastic properties E, G, ν are continuous functions
of z.
The potentials of Appendix A cannot be used when the elastic properties are
functions of the coordinates, but the linearity of the problem ensures that a sinusoidal
surface load will still generate a stress and displacement field that is sinusoidal in x.
We therefore start by writing the displacements as
u x (x, z) = Ux (z) sin(ωx); u z (x, z) = Uz (z) cos(ωx), (14.107)

corresponding to strain components
∂u x ∂u z dUz
ex x = = ωUx (z) cos(ωx); ezz = = cos(ωx)
∂x ∂z dz

1 ∂u x ∂u z 1 dUx
ex z = + = − ωUz sin(ωx). (14.108)
2 ∂z ∂x 2 dz
Hooke’s law then gives the stress components

2G dUz
σx x = (1 − ν)ωUx (z) + ν cos(ωx)
(1 − 2ν) dz

2G dUz
σzz = (1 − ν) + νωUx (z) cos(ωx) (14.109)
(1 − 2ν) dz

dUx
σx z = G − ωUz sin(ωx),
dz
where we recall that G, ν are functions of z. These stress components must satisfy
the equilibrium Eq. (14.31), leading to two ordinary differential equations for the
two unknown functions Ux (z), Uz (z). In most cases, Poisson’s ratio can be assumed
[at least approximately] independent of z, in which case we obtain [after cancelling
some non-zero factors]

d 2 Ux 2(1 − ν)ω 2 Ux ω dUz dG dUx
G − − + − ωUz = 0
dz 2 (1 − 2ν) (1 − 2ν) dz dz dz

2(1 − ν) d 2 Uz ω dUx
G − ω 2 Uz +
(1 − 2ν) dz 2 (1 − 2ν) dz

2 dG dUz
+ (1 − ν) + νωUx = 0. (14.110)
(1 − 2ν) dz dz
14.5.1 Exponential Variation of Modulus
The solution of these equations is particularly straightforward if the modulus varies

exponentially with depth—i.e.
G(z) = G 0 exp(λz), (14.111)
since the exponential factor then cancels and we obtain

14.5 Functionally Graded Materials 315
d 2 Ux dUx 2(1 − ν) 2 ω dUz

+λ − ω Ux = + λωUz
dz 2 dz (1 − 2ν) (1 − 2ν) dz
2(1 − ν) d 2 Uz 2λ(1 − ν) dUz ω dUx 2νλωUx
+ − ω 2 Uz = − − ,
(1 − 2ν) dz 2 (1 − 2ν) dz (1 − 2ν) dz (1 − 2ν)
comprising two ordinary differential equations with constant coefficients for the
unknown functions Ux (z), Uz (z). The general solution is obtained by writing
U ≡ {Ux , Uz }T = {A x , A z }T exp(βz), (14.112)
from which we obtain the homogeneous algebraic equations

2(1 − ν) 2 1
β + λβ −
2
ω Ax − ωβ + λω A z = 0
(1 − 2ν) (1 − 2ν)

ω 2νλω 2(1 − ν) 2 2λ(1 − ν)
β+ β Ax + β + β − ω A z = 0.
2
(1 − 2ν) (1 − 2ν) (1 − 2ν) (1 − 2ν)
There exist four eigenvalues βi , (i = 1, 4) for which these equations have a non-trivial
solution, and the general solution can then be constructed in the form

4
U(z) = Ci Ai exp(βi z), (14.113)
i=1
where Ai are appropriately normalized eigenvectors and Ci are arbitrary constants.

Notice that the eigenvalues may include complex conjugate pairs, in which case
the corresponding constants Ci will also be complex conjugates, and the resulting
displacements and stresses will exhibit oscillatory exponential variation of with z.
14.5.2 Power-Law Grading
A fairly wide range of contact problems can be solved in closed form if the half-space
z < 0 exhibits power-law grading defined such that the shear modulus
G(z) = G 0 z k where 0 < k < 1, (14.114)
but Poisson’s ratio ν is independent of z. As in the homogeneous case, the point and
line force problems of Figs. 2.2 and 6.1 have no inherent length scale and hence, as
we argued in Sect. 2.2.2, the stress and displacement fields must be expressible in
separated variable form in appropriate polar coordinates. Equilibrium considerations
then show that the stress components in the two-dimensional case of Fig. 14.8a must
take the form
(a) P (b)
δr tθ+δθ
r δθ tr
B A
r +δr θ
r A ( r+δr)δθtr +δr
δθ δr tθ
Fig. 14.8 (a) A functionally graded half-plane loaded by a normal force P, and (b) free-body
diagram of the infinitesimal element A
f i j (θ)
σi j = . (14.115)
r
Figure 14.8b shows an enlarged view of the infinitesimal element A. Equation
(14.115) shows that the vector tractions t r and t r +δr on the curved surfaces must be
proportional to 1/r and 1/(r +δr ) respectively, and hence the corresponding forces
must be equal and opposite. It follows that if the element is to be in equilibrium,
the forces on the straight sides must also form an equal and opposite pair. Now the
upper surface of the element B is part of the traction-free surface of the half-plane,
so the lower straight side of B must also be traction-free. It follows immediately that
the straight sides of all such elements must be traction-free, and hence that the most
general state of stress can be written
f (θ)
σrr = ; σθr = σθθ = 0. (14.116)
r
The corresponding solution (6.2) for the homogeneous half-plane is clearly of this
form, but it is interesting to note that the same result holds for any line force problem
exhibiting self-similarity. For example, it would also apply to the case of a half-space
comprising a material with a nonlinear constitutive law6 of the form σi j = ci jkl (ekl )α ,
where α is any positive exponent.
The strains corresponding to the stress field (14.116) are given by Hooke’s law as
(1 − ν) f (θ) ν f (θ)
err = ; eθθ = − ; eθr = 0, (14.117)
2G 0 r k+1 cosk θ 2G 0 r k+1 cosk θ
and substitution in the strain compatibility equation yields an ordinary differential

equation for f (θ) with solution
6 Butbe warned, the nonlinearity invalidates superposition, so the resulting Green’s function is of
rather limited practical value.

(1 + k)[1 − ν(1 + k)]
f (θ) = cos θ[A cos(βθ) + B sin(βθ)]
k
where β= ,
(1 − ν)
(14.118)
and A, B are arbitrary constants (Giannakopoulos and Pallot 2000). For the normal
loading problem of Fig. 14.8a, B = 0 by symmetry, and A can be determined by
considering the equilibrium of a semicircular region including the origin. We obtain
Ck P cosk θ cos(βθ)
σrr = − , (14.119)
2r
where
2k+2 3+k+β 3+k−β
Ck = Γ Γ . (14.120)
πΓ (3 + k) 2 2
The corresponding normal surface displacements are
Ck β P(1 − ν) sin(βπ/2)
u z (x, 0) = . (14.121)
4G 0 k(k + 1)|x|k
Equilibrium and dimensional considerations show that the antisymmetric term

associated with the term B in Eq. (14.118) must correspond to a tangential force
applied to the half-plane, and the resulting tangential displacements must also vary
with |x|−k . It follows that the Ciavarella–Jäger theorem of Sect. 9.2 applies to prob-
lems involving partial slip, provided the problem is uncoupled.
Giannakopoulos and Pallot (2000) used these results to solve a range of two-
dimensional contact problems analogous to those considered in Chaps. 6 and 9. In
particular, the contact pressure under a frictionless flat punch is given by
(k−1)/2
x2 Γ (1 + k) P
p(x) = p0 1 − 2 where p0 = 2 . (14.122)
a 2k Γ 1+k a
2
More general problems for a power-law-graded half-plane can be solved using an

adaptation of the incremental method of Sect. 6.3, leading to Abel integral equations
with fractional powers (Jin et al. 2013).
Giannakopoulos and Suresh (1997a) obtained the corresponding three-dimen-
sional [point force] Green’s function by applying the Smirnov–Sobolev trans-
formation of Sect. 6.6 to the two-dimensional solution. They also solved several
axisymmetric indentation problems, including those for flat, conical and spherical
indenters (Giannakopoulos and Suresh 1997b). Jin et al. (2013) developed a solution
for the general axisymmetric problem, analogous to that introduced in Chap. 5 for
homogeneous materials.
Equation (14.122) shows that the singularity at the edge of the contact region
has a power (k −1)/2, which is weaker than the square-root singularity which we
obtained in the homogeneous case. The arguments developed in Chap. 10 show that
this asymptotic behaviour should be expected at a sharp corner in any contact problem
involving a power-law-graded half-space. This also implies that if adhesive forces
are included (Jin et al. 2013), and if the energy argument underlying the JKR model
of Sect. 12.2 is adopted, then the generalized stress intensity factor at the edge of the
contact area must have a unique value depending only on Δγ, G 0 , k.
In non-conformal contact problems without adhesion, the contact pressure goes
to zero at the edge of the contact region with a corresponding positive power. For
example, for the indentation by a frictionless rigid sphere, the contact pressure is
proportional to (1−r 2 /a 2 )(1+k)/2 (Giannakopoulos and Suresh 1997b).
14.5.3 Linear Variation of Modulus
Equation (14.122) shows that in the limit k → 1, where the elastic modulus is linearly
proportional to the depth7 —i.e.
G(z) = G 0 z, (14.123)
the pressure distribution under the flat punch becomes uniform. More generally, the
half-space then mimics a Winkler foundation, in which the contact displacement is
linearly proportional to the local pressure. Here, we consider the special case where
the material is incompressible (ν = 0.5), but we allow the contact pressure to be any
function of x, y on the surface z = 0.
A particular solution to the elasticity problem can then be obtained by defining a
potential φ, such that the displacement u = ∇φ. The strains are then
∂2φ ∂2φ ∂2φ

ex x = ; e x y = ; e x z = etc., (14.124)
∂x 2 ∂x∂ y ∂x∂z
and since the material is assumed incompressible, the dilatation [volumetric strain]
ex x + e yy + ezz = ∇ 2 φ = 0. (14.125)
The stress components are
∂2φ ∂2φ ∂2φ

σx x = σ̄ + 2G 0 z ; σx y = 2G 0 z ; σx z = 2G 0 z etc.,
∂x 2 ∂x∂ y ∂x∂z
(14.126)
where σ̄(x, y, z) represents a hydrostatic stress, which is included only in the normal
stress components. Substitution of these results into the first equilibrium equation
7 This is a special case of a ‘Gibson soil’ (Gibson 1967), in which G(z) = C z+D. Models of this kind
are used to approximate the properties of soils which generally show an increase of consolidation,
and hence of elastic modulus, with depth.
yields
∂ σ̄ ∂3φ ∂3φ ∂3φ ∂2φ

+ 2G 0 z 3 + 2G 0 z + 2G 0 z + 2G 0 = 0. (14.127)
∂x ∂x ∂x∂ y 2 ∂x∂z 2 ∂x∂z
However, the condition ∇ 2 φ = 0 ensures that the three middle terms sum to zero, so
we obtain
∂ σ̄ ∂2φ
+ 2G 0 = 0. (14.128)
∂x ∂x∂z
Similar operations on the other two equilibrium equations yields
∂ σ̄ ∂2φ ∂ σ̄ ∂2φ
+ 2G 0 = 0; + 2G 0 2 = 0, (14.129)
∂y ∂ y∂z ∂z ∂z
and all three equations are satisfied if we write
∂φ
σ̄ = −2G 0 . (14.130)
∂z
Finally, we substitute this result back into the stress expressions obtaining
∂φ ∂2φ ∂2φ ∂φ ∂2φ

σx x = −2G 0 + 2G 0 z 2 ; σx y = 2G 0 z ; σ yy = −2G 0 + 2G 0 z 2
∂z ∂x ∂x∂ y ∂z ∂y
∂2φ ∂2φ ∂φ ∂2φ

σx z = 2G 0 z ; σ yz = 2G 0 z ; σzz = −2G 0 + 2G 0 z 2 . (14.131)
∂x∂z ∂ y∂z ∂z ∂z
It follows that the shear tractions σzx , σzy are identically zero on the plane z = 0 for
all functions φ, so this representation is appropriate for the solution of frictionless
contact problems for the half-space z > 0.
The corresponding values of normal surface displacement and contact pressure
are
∂φ ∂φ
u z (x, y, 0) = ; p(x, y) = −σzz (x, y, 0) = 2G 0 , (14.132)
∂z ∂z
so that at all points on the surface,
p(x, y) = 2G 0 u z (x, y, 0). (14.133)
In other words, the half-space acts as a Winkler foundation, with normal displacement
proportional to local contact pressure. Another remarkable property follows from
a comparison of Eq. (14.131) with the potential function solution of Appendix A,
Sect. A.1. If we set ν = 0.5 in the expressions for stress components in Sect. A.1 and
use the condition ∇ 2 ϕ = 0, we find that the results are identical to (14.131) with
2G 0 φ = ∂ϕ/∂z. It follows that the stresses in the linearly graded half-space will be
identical to those in a homogeneous elastic half-space with the same contact pressure
distribution, though the displacements will of course be different.
Problems
1. A beam of flexural rigidity E I supported on an elastic foundation of modulus k is

indented by a symmetrical rigid wedge of semi-angle π2 −α, where α 1. Find the
maximum force P0 for which contact will be restricted to a single point at the apex
of the wedge. Describe in words how you expect the contact configuration to evolve
for P > P0 , but do not attempt to solve this problem.
2. Use the boundary conditions (14.25) and the homogeneous solution (14.21) to
find the relation between the indenting force P, the indentation depth Δ and the
contact radius a when a rigid sphere of radius R is pressed into an infinite elastic
plate of thickness h supported on an elastic foundation of modulus k. Plot a figure
showing P and a as functions of Δ in appropriate dimensionless form. Also, find the
pressure distribution in the contact area and the magnitude of the force per unit length
developed at the edge of the contact area. How does the proportion of P carried in
this concentrated force vary with a?
3. A rigid flat punch of equilateral triangular cross section is pressed by a force P
into an incompressible layer of thickness h and shear modulus G that is bonded to
a rigid foundation. Find an approximate solution for the pressure distribution and
the indentation depth Δ, using the displacement function ψ = C(x −a)(x 2 −3y 2 ) in
Eq. (14.43), where C is a constant.
4. The results in Sect. 14.2.3 follow Johnson’s assumption that the contact is friction-
less, but the analysis of Sect. 14.2.5 suggests that the friction coefficient required to
prevent slip is rather small. Develop a similar analysis under the assumption of no
slip by assuming an in-plane displacement of the form
u = z(z − h)∇ψ.
How much will this increase the indentation force for a flat-ended cylindrical indenter
of radius a?
5. The criterion (14.68) implies that for indentation by a flat punch, an arbitrarily
small coefficient of friction is sufficient to prevent slip. Show the resulting solution
implies the existence of an inwardly directed friction force distributed along the
boundary of the contact area and find the magnitude of this force per unit length as
a function of the indentation Δ. If the friction coefficient is actually finite and equal
to f , how wide a strip of microslip would you anticipate at this edge.
6. Using the loading strategy of Sect. 12.2.3, find the elastic strain energy U (a, Δ)
for the bonded incompressible layer indented to a depth Δ by a rigid sphere of radius
Problems 321
R with contact radius a. The total potential energy is then Π =U −πa 2 Δγ. Find the
equilibrium radius a by imposing the condition ∂Π/∂a = 0 and hence verify that
the resulting traction satisfies the condition (14.73).
7. An elastic layer of thickness h rests on a rigid frictionless half-plane and is loaded
by a compressive traction p0 [1 + cos(ωx)] on the free surface. Use the potential
functions (14.89) to find the corresponding normal surface displacements.
8. A normal point force P is applied to the surface of an elastic layer of thickness
h bonded to a rigid foundation. Find the double Fourier transform of this loading,
using Eq. (14.104)1 . Define polar coordinates (r, θ), (ρ, φ) such that
x = r cos θ; y = r sin θ; s = ρ cos φ; t = ρ sin φ
and use this representation to obtain an integral expression for the axisymmetric
normal surface displacement u z (r ) due to the point force.
9. Use the formulation of Sect. 14.5 to determine the normal surface displacements
due to the sinusoidal pressure distribution p(x) = p0 cos(ωx) acting on the surface
of a half-plane for which ν = 1/4 and G = G 0 exp(λz) with λ > 0.
Chapter 15
Indentation Problems
If an elastic–plastic material is indented by a rigid quadratic indenter such as a sphere,

the contact problem will exhibit an elastic range, in which the stress field and the
force–displacement relation are defined by the Hertzian analysis. The maximum
shear stress associated with the axisymmetric Hertzian problem occurs at a depth
of 0.48a, where a is the radius of the contact circle and this point reaches both the
Tresca and von Mises yield conditions when
21.2SY3 R 2
P = PY = , (15.1)
E∗ 2
where SY is the yield stress in uniaxial tension (Johnson 1985). However, significantly
larger forces are needed to create a visible indentation after the force is removed,
since the region affected is buried beneath the surface. Incidentally, this is one of
very few loading conditions that cause stress extrema to occur in the interior of a
body rather than at a point on the boundary.
As the force is increased further, the plastic zone grows and eventually breaks
through to the free surface, permitting a kinematic mechanism for redistribution of
material and leading to a significant residual indentation.
15.1 The Hardness Test
Indentation of a body by a hard indenter is one of the classical ways of measuring

the properties of materials. Typically, the indenter is pressed into a softer body by a
prescribed force P and then removed, after which the size of the permanent inden-
tation is measured optically. The hardness H is then defined as H = P/A, where A
is the area of the observed permanent indentation.

https://doi.org/10.1007/978-3-319-70939-0_15
324 15 Indentation Problems
Vickers Berkovich Knoop

Fig. 15.1 Indenter shapes for the Vickers, Berkovich and Knoop tests
The Brinell test uses a spherical indenter, but hardness measurements are also
frequently made with pyramidal indenters of various shapes, notably the Vickers,
Berkovich and Knoop tests, for which the corresponding indenter shapes are shown
in Fig. 15.1.
The generation of a permanent indentation implies irreversible plastic deformation
of the material, and hence the hardness typically correlates with the yield stress of the
material. In his classic monograph on the hardness of metals, Tabor (1951) reports
the relation H = 2.8SY for the Brinell test, where SY is the uniaxial yield stress.
If H were a true material constant, P would be proportional to A and hence to a 2 ,
where a is a typical linear dimension of the resulting indentation [e.g. the radius in
the case of the Brinell test, where the indenter is spherical]. However, Meyer (1908)
found that
P ∼ an , (15.2)
where the Meyer index n is typically larger than 2. This is attributed to work hardening
of the material, and Tabor (1951) presents experimental data showing that in the
Brinell test, n ≈ 2+β for materials whose constitutive law in the plastic régime can
be approximated by the power-law form σ ∼ β .
More recently, micro- and nano-indentation tests have also been used to probe
the elastic properties of very thin surface layers (Vlassak and Nix 1994; Oliver and
Pharr 2004), by observing the relation between P and the penetration depth Δ.
15.2 Power-Law Material
Spheres, pyramids and cones all fall under the general category of power-law inden-
ters, where the initial gap function g0 (r, θ) has the form
g0 (r, θ) = r α f (θ), (15.3)

15.2 Power-Law Material 325
where α > 0. These are non-conformal contact problems, so the contact area increases
with the indenting force P, but in many cases it is possible to map these various
states into a universal solution and thereby expose useful relations between the bulk
quantities Δ, P and a.
To illustrate this procedure, suppose that the indented body can be represented by
a homogeneous half-space with the nonlinear constitutive law
β
σi j = ci jkl kl , (15.4)
where
1 ∂u k ∂u l
kl = + (15.5)
2 ∂xl ∂xk
is the strain tensor, u i is the displacement vector and β is a scalar exponent. Laws of
this form are often used to approximate plastic behaviour during monotonic loading.
The contact problem is defined in a formal sense by the boundary conditions
p(x1 , x2 ) ≥ 0; u 3 (x1 , x2 , 0) ≥ Δ − r α f (θ), (15.6)
and the equilibrium condition

P= p(x1 , x2 )d x1 d x2 , (15.7)
A
where p(x1 , x2 ) = −σ33 (x1 , x2 , 0) and x1 =r cos θ, x2 =r sin θ.

We anticipate that the solution for different values of P can be mapped into
a universal solution in terms of suitable scaled variables. In preparation for this
process, we define reduced parameters x̃i , r̃ , σ̃i j , p̃, ũ i , P̃ through
xi = x̃i Δλ1 ; r = r̃ Δλ1
σi j = σ̃i j Δλ2 ; p = p̃Δλ2 ; u i = ũ i Δλ3 ; P = P̃Δλ4 , (15.8)
where λ1 , λ2 , λ3 , λ4 are as yet undetermined exponents.

Introducing these results into Eqs. (15.4)–(15.7), we have
β
λ2 Δλ3 −λ1 ∂ ũ k ∂ ũ l
σ̃i j Δ = ci jkl + , (15.9)
2 ∂ x̃l ∂ x̃k
p̃(x̃1 , x̃2 ) ≥ 0; ũ 3 Δλ3 ≥ Δ − r̃ α Δαλ1 f (θ), (15.10)

P̃Δλ4 = Δλ2 Δ2λ1 p̃(x̃1 , x̃2 )d x̃1 d x̃2 , (15.11)
A
and the factors of Δ will cancel in these conditions if

λ2 = β(λ3 − λ1 ); λ3 = αλ1 = 1; λ4 = λ2 + 2λ1 , (15.12)
with solution

1 1 2−β
λ1 = ; λ2 = β 1 − ; λ3 = 1; λ4 = β + . (15.13)
α α α
The problem defined in the reduced variables then has a unique solution which is
independent of force or indentation depth, so we conclude that the relations between
the dimensional parameters P, σ, a, u and Δ must have the form
a ∼ Δ1/α ; P ∼ Δβ+(2−β)/α ; P ∼ a αβ+2−β

σ ∼ P (αβ−β)/(αβ−β+2) ; u ∼ Δ. (15.14)
Hill et al. (1989) applied this argument to the Brinell test for which g0 is quadratic
and hence α = 2. It follows immediately that P ∼ a 2+β in accordance with Tabor’s
empirical data. They also generated a finite element solution of the indentation prob-
lem in the reduced variables.
Similar reduced variable methods have been used to develop finite element models
for Vickers and Berkovich indentation, where the indenter is pyramidal and hence
non-axisymmetric (Giannakopoulos et al. 1994; Larsson et al. 1996; Giannakopoulos
and Larsson 1997). In all these cases, the indentation problem for the entire loading
history is reduced to the solution of a single nonlinear boundary value problem.
15.2.1 Graded Materials
The same technique is easily extended to power-law graded materials, for which the
constitutive tensor can be written in the form
ci jkl = Ci jkl z γ , (15.15)
where the exponent γ defines the form of the grading in the z-direction. Equation
(15.9) is then modified to
β
λ2 γ λ1 γ Δλ3 −λ1 ∂ ũ k ∂ ũ l
σ̃i j Δ = Ci jkl z̃ Δ + , (15.16)
2 ∂ x̃l ∂ x̃k
and hence (15.12)1 is replaced by
λ2 = λ1 γ + β(λ3 − λ1 ). (15.17)
15.2 Power-Law Material 327
Solving for λ1 , λ2 , λ3 , λ4 , we find that the relations between P, σ, a, u and Δ are

then modified to
a ∼ Δ1/α ; P ∼ Δβ+(2+γ−β)/α ; P ∼ a αβ+2+γ−β

σ ∼ P (αβ+γ−β)/(αβ+γ−β+2) ; u ∼ Δ. (15.18)
These results can be used to interpret experimental indentation results. For example,
if the material is assumed to be linearly elastic (β = 1), a logarithmic plot of P against
Δ will provide information about the grading of the modulus with depth.
15.3 Other Constitutive Laws
If the material is homogeneous (γ = 0) and if the punch is of generalized conical

or pyramidal form (α = 1), the indentation problem exhibits self-similarity for more
general constitutive laws. In particular, the contact area preserves the same shape for
all forces P and its linear dimensions increase linearly with the indentation depth Δ.
The stress field is invariant in dimensionless coordinates x̃i = xi /Δ and hence the total
force P ∼ Δ2 . For example, if the material behaves elastically up to some yield stress
and then subsequently follows a prescribed work-hardening trajectory, the shape of
the plastic zone will be independent of force, with dimensions linearly proportional
to the indentation depth Δ. Notice that there is also a disadvantage associated with
this simple result—it implies that no information about the constitutive law can be
gleaned from the shape of the force–displacement law, since all constitutive models
give a quadratic function for this relation.
Problems of this kind generally require numerical solution, but there is consider-
able advantage in formulating them in the reduced variables since (i) only a single
solution has to be found and (ii) the finite element mesh can be chosen so as to give
resolution appropriate to the size of the contact area. By contrast, if the problem
is formulated in the original variables, solutions must be obtained at a sequence of
values of P to track the history-dependent process and the mesh should ideally be
sufficiently fine to capture all of this sequence of geometries adequately.
Borodich (1993) has developed the self-similarity approach in various indentation
contexts and in particular has shown that the above arguments all remain valid if
Coulomb friction conditions apply at the interface, in which case the partition of the
contact area into slip and stick regions preserves the same form as the force [and
hence the total contact area] is increased. Similar methods can also be used to analyse
creeping materials (Hill 1992; Bower et al. 1993; Storåkers et al. 1997).
Table 15.1 Force P and indentation Δ in an indentation test

P (N) 0.21 0.32 0.53 0.65 1.13 2.00 3.35 4.95 6.70 10.75 23.50
Δ(µm) 2.55 3.06 3.83 4.22 5.31 6.82 8.54 10.11 11.52 14.13 19.81
Problems
1. Show that for indentation of a homogeneous half-space by a rigid conical indenter,

the contact pressure p(r ) is a function of the ratio r/a only, where a is the radius of
the contact area, and does not otherwise depend upon the applied force P.
2. Table 15.1 shows values of applied force P and indentation depth Δ from a series
of microindentation measurements using a pyramidal indenter that can be assumed
rigid. Assuming the material to be linear elastic, use an appropriate logarithmic plot
to comment on the likely form of the variation of modulus with depth.
3. A spherical indenter is pressed into an elastic half-space by a normal force that
cycles between a maximum value of P0 and a minimum value of ρP0 , where 0 < ρ <
1. Frictional microslip at the interface leads to a dissipation of energy W per cycle.
Can we argue that the dissipation has a power-law dependence on P0 [i.e. W ∼ P0λ ]
for a given value of ρ and if so, what is the value of λ?
4. A half-plane is indented by a two-dimensional power-law rigid indenter defined by
the gap function g0 = C|x|α . Assuming the constitutive law is defined by Eq. (15.4),
find the expected power-law form of the relation between the indenting force P and
the contact semi-width a.
5. Brinell hardness tests conducted on the same material with the same maximum
force P, but different indenter diameters D show different values for the hardness
H calculated as P/πa 2 , where a is the radius of the residual indentation.
If the results are a good fit to a relation
λ
H D
= ,
H0 D0
where H0 , D0 , λ are constants, show that they are consistent with a power law con-
stitutive relation (15.4) and find the appropriate value of β as a function of λ.
Chapter 16
Contact of Rough Surfaces
The historical development of the study of rough surface contact is intimately

connected with attempts to explain Coulomb’s law of friction, in particular, the
observation that the tangential force F required to cause sliding is approximately
proportional to the applied normal force N and largely independent of the nominal
contact area in a wide variety of situations. The mathematical statement of this law
and its implications for contact problems are discussed in Chap. 8.
16.1 Bowden and Tabor’s Theory of Friction
The first reasonably satisfactory explanation of Coulomb’s law was advanced by

Bowden and Tabor (1950). They argued that the inevitable roughness of the contact-
ing surfaces would cause intimate contact to be restricted to a few microscopic actual
contact areas, constituting a small fraction of the total nominal contact area. The
contact pressure at these small contact areas will therefore be very high and Bowden
and Tabor likened the local contact state to that in the hardness test [see Sect. 15.1].
They concluded that the total area of actual contact A could be approximated by the
relation
P = H A, (16.1)
where P is the applied normal force and H is the hardness of the softer material.
They also argued that the high local pressure would cause ‘cold welding’ to occur
at the actual contact areas and that tangential motion [sliding] would be opposed by
shear stresses at these ‘adhesive’ junctions. Assuming that a constant shear stress τ
is required to break each junction, then the tangential force Q required for sliding is
Q = τA (16.2)

https://doi.org/10.1007/978-3-319-70939-0_16
330 16 Contact of Rough Surfaces
and hence τ
Q= fP where f = . (16.3)
H
In other words, the coefficient of friction f is predicted to be the ratio of two material
properties τ and H , and is independent of the particular geometry of the problem,
including the nominal contact area.
16.1.1 The Ploughing Force
This theory implies that if one material is significantly harder than the other, the
properties of the softer material determine the friction process. Also, during continued
sliding the highest peaks or asperities of the harder material will plough through the
softer material, also possibly removing material as in a machining process. This has
two implications. There will be an additional frictional force because the normal
pressure at the interface will act on an inclined surface, and the amount of material
removed per unit sliding distance can be estimated, based on the size of the grooves
anticipated. Wear by this mechanism is known as abrasive wear.
For example, suppose a single asperity can be represented by a cone of semi-angle
( π2 − α), and that it ploughs through a softer material leaving a groove of depth h as
shown in Fig. 16.1. The semi-width of the groove will then be
a = h cot α (16.4)
and the volume of material removed when ploughing through a distance L will be
V = ah L = h 2 L cot α. (16.5)
If the traction at the contact interface comprises a purely normal contact pressure
equal to H , the resultant force in any direction n will be H An , where An is the
projection of the actual contact area on the plane normal to n. During ploughing,
contact will occur only on the upstream half of the potential contact area because of
Fig. 16.1 Ploughing of a P

soft material by a hard
conical asperity Q
h a a
α
16.1 Bowden and Tabor’s Theory of Friction 331
the material removal, so the normal indenting force will be
πa 2 H π H h 2 cot 2 α
P= = (16.6)
2 2
and the frictional force
Q = ah H = H h 2 cot α. (16.7)
Thus, if ploughing were the sole cause of friction — i.e. if there were no tangential
tractions at the interface — the coefficient of friction associated with a single asperity
contact would be
Q 2 tan α
f = = . (16.8)
P π
It also follows that the work done by the force Q in moving through a distance
L is
W
W = Q L = H h 2 L cot α and hence V = . (16.9)
H
In other words, the volume of abrasive wear is proportional to the work done against
friction and in particular is independent of the cone angle α. In fact, this relation would
be obtained for an abrasive indenter of any shape, since Q = As H and V = As L,
where As is the projection of the contact area on a plane normal to the sliding
direction.
Bowden and Tabor argued that the actual frictional force would be the sum of
those obtained from Eqs. (16.2) and (16.7) and hence
τ 2 tan α
Q= fP where f = + . (16.10)
H π
16.1.2 Plastic Deformation at an Actual Contact
Of course the supposed material constants τ and H in Eq. (16.10) are not independent.
They both relate to a local state of plastic deformation near the actual contact area.
The theory of plasticity was only in its infancy during the development of Bowden and
Tabor’s theory and a more rigorous examination of this problem leads to difficulties.
Investigations were carried out using rigid/plastic analysis [slip-line field theory] and
model experiments were performed using large-scale models of soft materials such
as lead or plasticine.1 The results showed that the contacting bodies would actually
tend to move closer together on the application of a tangential force, establishing
a larger area of actual contact and eventually causing the relative motion to cease,
rather than permitting sliding.
1 See for example, Greenwood and Tabor (1955), Johnson (1985), p. 233 et seq.
(a) (b)
τ τ
(- H , τ)
τ -σ0
-H -σ0 0 σ -H τ 0 σ
(-σ0 ,- τ)
Fig. 16.2 Mohr’s circle for a point near the surface (a) under purely normal loading, and (b) with
superposed tangential loading
To understand this, consider the state of stress at or near the interface. Follow-
ing Bowden and Tabor, we assume that under purely normal loading the normal
stress σzz = −H , and the hardness H is larger than the uniaxial yield stress because
orthogonal compressive stresses σx x , σ yy [smaller in magnitude than σzz ] will be
developed by the constraint of the surrounding material. We also neglect normal-
tangential coupling, so no frictional tractions will be developed during normal load-
ing and σzz will be a principal stress. Figure 16.2a shows a representative Mohr’s
circle for this state of stress for the case where the contact is axisymmetric and hence
σx x = σ yy = −σ0 , with σ0 < H .
Now suppose that a tangential force is applied, causing a uniform shear traction
σzx = τ over the contact area. Symmetry considerations suggest that no normal
stresses will be developed at the interface during this phase of the loading [at least near
the centre of the contact area], so the Mohr’s circle is modified to that in Fig. 16.2b,
and in particular, the radius of the circle is increased by the shear traction. If the stress
state was already on the yield surface under purely normal loading, both Tresca and
von Mises yield criteria then demand that yielding must continue for an arbitrarily
small applied shear stress. More significantly, the direction of the maximum shear
stress [which governs the direction of any plastic strain increment] is still very close
to that associated with the initial normal loading,2 so the principal effect will be for
the bodies to move closer together, increasing the contact area A. Eventually, this
process can result in a substantial cold weld between the surfaces.
16.1.3 The Effect of Surface Films
Under certain circumstances — notably with metallic surfaces in a vacuum or an inert

atmosphere — this is precisely what happens.3 However, in most other cases, the
2A somewhat similar process occurs with frictional sliding. Place an object on a plane surface and
incline the surface until sliding is just [but not quite] about to start. Then push horizontally in the
direction perpendicular to the line of greatest slope. You will find that the resulting motion is almost
aligned with the line of greatest slope, rather than with the incremental tangential force.
3 For this reason, it is very difficult to develop good tribological systems to operate in a vacuum.
16.1 Bowden and Tabor’s Theory of Friction 333
interface is not as strong as the rest of the material because of the presence of surface
films and contaminants. Thus, Bowden and Tabor’s theory can still be regarded as
meaningful, provided we interpret the ‘material property’ τ as a property of a weaker
interfacial layer, rather than that of the bulk material. Of course, it is then much more
difficult to estimate what might be realistic values of this constant and the values
will depend very much on the way the surface was manufactured and other factors
such as humidity [which affects oxidation], and whether the surface was cleaned of
grease or oil.
16.2 Profilometry
Theories of the contact of rough surfaces have been heavily influenced by the devel-
opment of surface measurement techniques or surface profilometry. The earliest
successful measurements of rough surfaces were made by passing a sharp hard sty-
lus over the surface and measuring its vertical displacement. This technique was
perfected by Taylor Hobson in the ‘Talysurf’ machine, but numerous instruments
based on the same principle are now available. The output from a surface profilome-
ter is an approximation to the cross section of the surface along the line traversed, a
typical profile being shown in Fig. 16.3.
It is customary to use a much larger scale for the vertical dimension than for
the horizontal dimension, since typical surface slopes are quite small [generally less
than 5◦ , at least at relatively coarse scales of measurement]. The datum for height
can be set in the machine, by traversing the stylus carrier along a horizontal line, but
this presents difficulty of alignment, since if the measured surface is not perfectly
horizontal, the output will drift across the measurement range. Most profilometers
therefore measure height relative to a large radius ‘skid’ which rides over the highest
points of the surface, as shown in Fig. 16.4.
If the true profile of the surface is conceived as comprising a set of waves of
various wavelength and amplitude, the stylus profilometer filters the signal at both
ends of the spectrum. A lower limit to the measurable wavelength is provided by the
fact that the stylus is necessarily of finite size and hence it cannot penetrate into very
narrow depressions. At the other extreme, the skid will ride over long wavelength
features and hence not record them, and in addition some truncation is inevitable,
since the length of the measured sample is finite.
Fig. 16.3 Profile of a randomly rough surface, showing the multiscale nature of roughness. This
example has a fractal dimension D = 1.5 [see Sect. 16.5.2 below]
skid
stylus
Fig. 16.4 Profilometer measuring surface height relative to a long wavelength skid
Profilometers have improved considerably over the years and are now supple-
mented by optical measurements such as laser interferometry and by the atomic
force microscope [AFM]. As a result, the range of small wavelengths measurable
has been greatly extended. We shall see later in this chapter that this has challenged
conventional theoretical models of the contact process.
16.2.1 The Bearing Area Curve
The profilometer provides a wealth of information about the measured surface and
one of the earliest ways of presenting this data was to plot the proportion of the
profile above a given height as a function of height. This is known as a bearing area
curve. It can also be interpreted in a statistical sense as the probability that a given
point chosen at random on the surface will be higher than a given height.
One advantage of the statistical interpretation is that it makes it clear that the same
probability applies to three-dimensional surfaces, even though it is measured by a
two-dimensional cross-section through the surface. After all, the points sampled by
the instrument are essentially random except for the fact that they lie on a straight
line.4 This point was often missed by early investigators, some of whom argued that
the three-dimensional surface could be envisaged as a kind of regular rectangular
structure of peaks and valleys, drawing on the ‘unit cell’, or ‘representative vol-
ume element’ concept from studies of composite materials. The problem with this
approach is that it tacitly assumes that the representative profile conveniently passed
over the exact summits of these peaks, whereas the more common event would be
for it to pass over the flanks.
The bearing area is clearly very low at high values and approaches unity for low
values of height, a typical curve being shown in Fig. 16.5. If the surface is generated
4 This does mean that the results will not be representative of the whole surface if the line concerned
has some special significance for the surface. For example, if it is parallel with the ‘machining
marks’ generated during manufacture. However, if the line is chosen to be inclined to these marks
and if a sufficient number of peaks and troughs are traversed, the results for the bearing area will
still be representative even for such a quasideterministic surface.
16.2 Profilometry 335
Fig. 16.5 Form of the bearing area curve
by a random process, it is reasonable to expect a Gaussian [normal] distribution

defined by the equation

1 (h − h 0 )2
Φ A (h) = √ exp − , (16.11)
σ 2π 2σ 2
where h is the height above some datum, h 0 is the mean height, σ is the standard
deviation and Φ A (h)δh is the probability that a given point will have height between
h and h +δh.
Real surfaces are not generally Gaussian, particularly at low values of h [the lower
points of the surface], but Eq. (16.11) often represents a reasonable approximation
to the height distribution at higher values, comprising the points most likely to make
contact with another surface.
The bearing area B(h) is the probability that a given point is higher than h and is
therefore related to Φ A (h) by
∞
dB
B(h) = Φ A (h)dh or = − Φ A (h). (16.12)
h dh
For the Gaussian distribution of Eq. (16.11), we then have

∞
1 (h − h 0 )2 1 h − h0
B(h) = √ exp − dh = erfc √ , (16.13)
σ 2π h 2σ 2 2 σ 2
where ∞
2
erfc(x) = √ exp(−x 2 )d x (16.14)
π x
is known as the complementary error function. In statistical terms, Φ A (h) is the

distribution function and B(h) is the cumulative distribution. In practice, it is much
easier to plot the cumulative distribution, since it is less sensitive to discretization of
the data. To see this, imagine we want to plot the statistical data for the heights of a
group of people. If we try to plot Φ A (h), we shall need to divide the range into a set of
finite ranges [e.g. one-inch intervals between 5 feet and 7 feet]. If the sample size N
is small, there may be no individuals in some of these ranges, so the resulting curve
will not be smooth. By contrast, if we plot the cumulative distribution, we simply
make a step up of magnitude 1/N at the height of each individual and it is relatively
easy to smooth [regularize] the resulting curve.
16.2.2 The Contact Problem
Early investigations of the bearing area curve were motivated by the idea that the
actual contact area would be approximately equal to the ‘interference area’ — i.e. the
area in which the gap g(x, y) would be negative in the absence of local deformation.
If one of the bodies has a plane surface and this is brought down to a level h, the
proportion of the nominal contact area Anom in actual contact would then be given
by
A = Anom B(h). (16.15)
Combining this result with Bowden and Tabor’s contact theory [Sect. 16.1], we can
then estimate the relation between normal force and height and hence the incremental
stiffness of the contact. The normal force is obtained as
P = H A = H Anom B(h), (16.16)
from (16.1). Alternatively, we can define the nominal contact pressure
P
pnom = = H B(h). (16.17)
Anom
The incremental stiffness k is defined as the slope of the force–displacement curve

and hence is predicted to be
dP dB
k=− = −H Anom = H Anom Φ A (h), (16.18)
dh dh
from (16.12).
Contact of Two Rough Surfaces
Of course, in most cases both bodies will have rough surfaces, which we can charac-
terize by probability distributions Φ1 (h), Φ2 (h) respectively, where in each case the
height h is measured from the corresponding mean plane. If these bodies are placed
one above the other, such that the distance between the mean planes is h 0 , then a
point at height s in body 1 will interfere if the corresponding point in body 2 has a
16.2 Profilometry 337
height greater than (h 0 −s). The probability distribution for interference is therefore
∞ ∞ ∞
Φ(h 0 ) = Φ1 (s)B2 (h 0 − s)ds = Φ1 (s)Φ2 (t)dtds. (16.19)
−∞ −∞ h 0 −s
If the surfaces are both Gaussian with standard deviations σ1 , σ2 , respectively, the
integrals can be evaluated to give

1 h2
Φ(h 0 ) =
√ exp − 02 , (16.20)
σ 2π 2σ

where the combined standard deviation σ = σ12 +σ22 .
We recall that when the bodies can be represented as half-spaces, the contact
problem depends only on the initial gap function g0 (x, y) of Eq. (1.1). This implies
that the problem of contact between two rough surfaces can be reduced to the contact
of an equivalent rough surface with a plane. The probability distribution for this
equivalent surface is then Φ(h 0 ) of Eq. (16.19). In the rest of this chapter, we shall
generally make use of this simplification.
16.3 Asperity Model Theories
Contact stresses at actual contact areas are expected to be very high, so it seems
reasonable that local plastic deformation will occur. However, most materials work
harden and also the sharper asperities will be flattened by deformation, so after an
initial contact process it is likely that subsequent loading may be largely elastic
(Archard 1957). Even a ‘new’ surface must have been generated by a manufacturing
process, many of which involve plastic deformation and generate harder surface
layers. Also, the surface has probably made contact with other bodies on numerous
occasions during handling, so a good case can be made for most rough surface contact
problems being elastic, at least on the larger length scales.
Unfortunately in elastic contact, there is no linearity between contact area and
normal force. For example, in Hertzian contact the mean contact pressure increases
with normal force P to the one-third power [see Chap. 3]. It follows that Bowden and
Tabor’s theory of friction fails for elastic contact and the force–compliance relation
cannot be deduced directly from the bearing area curve.
An alternative approach is to argue that contact between rough surfaces occurs
only at the tips of the asperities of the two surfaces, so that the contact problem can
be simplified by replacing the actual body [Fig. 16.6a] by an idealization comprising
a distribution of asperities attached to some datum, as shown in Fig. 16.6b.
For simplicity, consider the case where all the asperities have the same shape, but
are at different heights, such that the probability of the summit of a given asperity
(a) (b)
h
datum
Fig. 16.6 (a) A rough surface profile, and (b) its idealization as a set of asperities
being between h and h +δh is Φ N (h)δh. We consider the case where this surface is
pressed against a rigid plane surface.
Suppose also that we can analyze the micromechanical problem of Fig. 16.7, in
which a single asperity, initially of height h, is compressed against a rigid plane
surface, and in particular determine the normal contact force f P (Δ), where Δ is the
indentation depth. For example, if the asperities were elastic spheres of radius R,
f P (Δ) would be given by Eq. (5.30).
If the rigid plane surface is brought down to the level h 0 , all asperities with h > h 0
will contact it, with a penetration depth Δ = h −h 0 . If the asperities on the model
surface can be regarded as independent, each will experience the force f P (h−h 0 ) and
the total force will be the sum of these separate asperity forces over the distribution.
In other words ∞
P = N0 Φ N (h) f P (h − h 0 )dh, (16.21)
h0
where N0 is the total number of asperities. Notice that the integration starts from
h = h 0 , since asperities with h < h 0 are not tall enough to make contact with the
plane.
We can also use this method to predict other quantities that are summations of
those at individual asperities. For example, the tangential force needed to cause a
single asperity to slide might be expected to be some function f Q (Δ) of the local
indentation depth, in which case the total frictional force during sliding will be
∞
Q = N0 Φ N (h) f Q (h − h 0 )dh. (16.22)
h0
Fig. 16.7 Normal contact of

a single asperity. The dashed
f P (Δ)
line shows the undeformed
rigid plane
position of the asperity, and Δ
the indentation depth is Δ h
asperity h0
datum
16.3 Asperity Model Theories 339
Also, the total area of actual contact will be

∞
A = N0 Φ N (h) f A (h − h 0 )dh, (16.23)
h0
where f A (Δ) is the area of contact for the single asperity contact.
Early asperity model theories concentrated on the micromechanics of the individ-
ual asperity contact [Fig. 16.7] and used quite sophisticated arguments to determine
the functions f P (d), f Q (d), f A (d), since the initial motivation for these theories was
to advance beyond the simple bearing area concept of Bowden and Tabor. Perhaps
as a consequence, the assumptions about the height distribution Φ N (h) were often
quite naïve and unrealistic. For example, asperities were assumed to be uniformly
distributed or to follow a power law distribution below some highest point. This led
to major discrepancies with experimental data and a high degree of scatter, since the
highest point of a surface is itself a statistical quantity.
To understand this, imagine we take a sample of the population and determine the
height of the tallest individual. If we increase the size of the sample substantially,
the chances are we shall find a taller individual. With a very large sample, the result
might stabilize, but it still depends on the chance events defining a single individual.
By contrast, the standard deviation of the distribution is a much more stable measure,
since it is an average over all individuals.
The difficulties caused by basing a distribution on the supposed highest summit
are well illustrated by the work of Ling (1958), who measured the force–compliance
behaviour of a solid contact and tried to extrapolate to zero force to determine the
point of first contact, even though the curve in this region is almost horizontal,
resulting in excessive sensitivity to experimental scatter.
16.3.1 The Exponential Distribution
A major breakthrough in asperity model theories was made by Greenwood and

Williamson (1966), who discovered that the height distribution has much more influ-
ence on the resulting macroscopic predictions than the assumed micromechanical
relations.
To explain this result, we first consider the exponential distribution function, which
leads to remarkable results. In particular, we assume the probability density function

1 h
Φ N (h) = exp − , (16.24)
σ σ
where the distribution applies only in the range h > 0.

Substituting into Eq. (16.21), we obtain
∞
N0 h
P= exp − f P (h − h 0 )dh, (16.25)
h0 σ σ
and making the change of variable z = h −h 0 , this becomes

N0 ∞ (z + h 0 )
P= exp − f P (z)dz
σ 0 σ
∞ z
N0 h0
= exp − exp − f P (z)dz. (16.26)
σ σ 0 σ
By a similar argument using Eq. (16.22), we obtain

∞ z
N0 h0
Q= exp − exp − f Q (z)dz, (16.27)
σ σ 0 σ
and hence the coefficient of friction is

∞ z ∞ z
Q
f = = exp − f Q (z)dz exp − f P (z)dz . (16.28)
P 0 σ 0 σ
Now these two integrals depend on the micromechanics functions f P (Δ), f Q (d =

Δ), but they do not depend on h 0 and hence the same coefficient of friction is obtained
for all h 0 and hence for all values of the normal force P. In other words, if the
asperities are distributed according to the exponential function (16.24), the friction
coefficient is predicted to be independent of normal force regardless of the assumed
micromechanics, which could be elastic, plastic, elastic–plastic with work hardening
or any other complex process. In the same way, any quantity that is obtained by a
summation of effects at individual asperities will have the same form and hence will
be proportional to normal force. In particular, the total actual contact area and the
total electrical or thermal conductance across the interface will be predicted to be
proportional to normal force.
16.3.2 The Gaussian Distribution
Of course, real surfaces do not have exponential distributions of asperities. A more

realistic assumption is the Gaussian distribution of equation (16.11), but the contact
process is largely dominated by the highest summits [large values of h] and in this
range the tail of the Gaussian distribution can be approximated by an exponential
with relatively modest errors over quite a wide range of values. We should not be
surprised therefore to find that predicted relations between P, Q, A, etc., with the
Gaussian distribution are fairly close to linear over a substantial range.
To illustrate this result, we consider the case where the asperities are all elastic
spheres of radius R, in which case the Hertzian solution of Eq. (5.30) shows that the
force on a single asperity is

∗
4E R 1/2 Δ3/2
f P (Δ) = (16.29)
3
and the actual contact area at the same asperity is
f A (Δ) = πa 2 = π RΔ. (16.30)
Substituting into Eqs. (16.21), (16.23) and using the Gaussian distribution

1 h2
Φ N (h) = √ exp − 2 , (16.31)
σ 2π 2σ
we obtain

P 211/4 h0
P̃ ≡ ∗
= √ I3/2
√ √ (16.32)
N0 E Rσ 3 3 π σ 2

A √ h0
Ã ≡ = 2π I1 √ , (16.33)
N0 Rσ σ 2
where
∞ ∞

Iλ (x) ≡ exp −ξ 2 (ξ − x)λ dξ = exp −(s + x)2 s λ ds. (16.34)

x 0
(a) (b)
Ã Ã/P̃
P̃ P̃
Fig. 16.8 (a) Dependence of actual contact area Ã on normal force P̃, using Greenwood and
Williamson’s theory and a Gaussian distribution of spherical asperities, (b) Ratio Ã/ P̃ as a function
of normal force
These integrals can be evaluated in terms of special functions [for example, using
Maple or Mathematica] for integer and fractional values of λ.
Figure 16.8a shows the dimensionless total actual contact area Ã as a function of
dimensionless
√ normal force P̃. This curve was plotted parametrically, using values
of h 0 /σ 2 in the range (0, 3), and shows that the relation is approximately linear,
despite the nonlinearity in the single asperity contact law (16.29).
To quantify the deviation from linearity, we plot the ratio Ã/ P̃ as a function of P̃
in Fig. 16.8b. This ratio varies only from 4 to 3 over four decades of normal force.
If, following Bowden and Tabor, we assume that the friction force is given by
Eq. (16.2), it follows that the friction coefficient will be almost independent of P over
a wide range. This result, first remarked by Greenwood and Williamson (1966), is
one of the most convincing explanations of the experimental observation that friction
coefficient is approximately independent of normal force. Furthermore, Bowden and
Tabor’s assumption (16.2) is not necessary to the argument. A similar result will be
obtained if other relations are used for the microscopic friction law of Eq. (16.22).
16.3.3 The Plasticity Index
The explanation for the results described in the last section is that when the normal
force is increased, the principal effect is to increase the number of asperities in contact,
whilst the average force on each changes relatively little. This may seem paradoxical
in that every asperity contact, once established, must become more heavily loaded
as the applied force increases. However, the same increase in force causes many new
lightly loaded asperity contacts to be established, thus restoring the force distribution.
In the special case of the exponential distribution of Sect. 16.3.1, this distribution
[i.e. the probability of a given (loaded) asperity experiencing a particular value of
compression Δ] is completely independent of h 0 , so all the quantities P, Q, A are
all proportional to the total number of asperities in contact and hence proportional
to each other.
For ductile materials, the individual asperity contacts will be elastic at low values
of Δ, but transition to plastic deformation at some larger value. Thus, for any given
normal force, we expect to see some asperity contacts that are elastic and some that are
plastic, and for the exponential distribution at least, the proportion in each state must
remain constant as the normal force is increased. This observation led Greenwood
and Williamson to characterize the stress level in an average asperity contact and
hence define a plasticity index whose value is a measure of the proportion of the
asperity contacts that are plastically deformed. Various definitions have since been
suggested, mostly based on the ratio between the mean contact pressure P/A and
the hardness of the material. For example, with the simple Gaussian distribution of
identical asperities from Sect. 16.3.2, we obtain
∗
P CE σ 29/4 I3/2
= where C= (16.35)
AH H R 3π I1
and the numerical factor C varies only slightly with normal force, as we see in
Fig. 16.8b. For more general surfaces, Johnson (1985) suggests the two definitions
∗ ∗
E √ E σm
ψ= σ s κs and ψ= , (16.36)
H H
where σs is the standard deviation for asperity heights, κs is the RMS value of
asperity summit curvature and σm is the mean slope of the surface. The first of
these is clearly closely related to (16.35), since curvature is the reciprocal of radius.
Also, the quantities σs , κs , σm are related for random surfaces, so the two definitions
(16.36) are not as different as they seem. As a rule of thumb, the contact process is
predicted to be largely elastic if ψ < 1 and significantly plastic if ψ > 1.
16.4 Statistical Models of Surfaces
Greenwood and Williamson’s results focussed the attention of researchers on the

statistics of real surfaces, and in particular, on methods of deducing the height distri-
bution of asperities and summit curvatures from profilometer measurements. Broadly
speaking, there are two approaches to this problem. One is to discretize the profile,
and then use the relation between the heights of adjacent grid points to identify asper-
ity summits and their curvatures. The other is to fit the data to a statistical model
of the surface — typically by determining the spectral content of the profilometer
signal — and then use continuum arguments to predict the distribution of asperity
properties.
We have already remarked that a peak in the profilometer output does not generally
represent the summit of the corresponding asperity, since in most cases the stylus
will travel across the flank of the asperity. Nayak (1971) highlights this distinction by
reserving the term ‘peak’ for a local maximum in the two-dimensional surface profile,
and ‘summit’ for the true highest point of the asperity on the three-dimensional
surface. This terminology is widely used in the subsequent literature and will also
be adopted here.
16.4.1 Discrete Models
Suppose that the profilometer output h(x) is discretized. In other words, it is repre-
sented by a discrete set of values h i , i = (0, N ), where h i = h(xi ) and usually xi will
define a set of equally spaced points
iL
xi = , (16.37)
N
where L is the length of the measured region. We can then define a peak as a point
xi satisfying the conditions
h i > h i−1 and h i > h i+1 . (16.38)
In other words, the point in question is higher than its two nearest neighbours. We
can also estimate the peak curvature κ p by fitting a parabola through the three points
h i−1 , h i , h i+1 with the result
1 2h i − h i+1 − h i−1 L
κp ≡ = where = (16.39)
Rp 2 N
is the sampling distance.

In the same way, we can construct a digital representation of a three-dimensional
surface by defining a set of heights h i j = h(xi , y j ) with xi = i, y j = j. In principle,
these might be obtained by traversing the surface with a profilometer along a set
of parallel lines separated by a distance , but many other measurement techniques
are available for measuring and digitizing surfaces (Whitehouse 1997; Conroy and
Armstrong 2005; Mathia et al. 2011).
It is then conventional to define a summit as any point (xi , y j ) where
h i j > h(xi−1 , y j ) h i j > h(xi+1 , y j ) h i j > h(xi , y j−1 ) h i j > h(xi , y j+1 ).
(16.40)
In other words, the given point is higher than its four nearest neighbours, labelled
1, 2, 3 and 4 in Fig. 16.9. However, even if this condition is satisfied, there is a non-
zero probability that one of the corner points (5, 6, 7, 8) in Fig. 16.9 may have a height
greater than h i j , so that some of the points identified from the criterion (16.40) may
not actually be summits.
Fig. 16.9 Grid points at

which height h(xi , y j ) is
measured
16.4 Statistical Models of Surfaces 345
Also, we note that the quadratic surface
h(x, y) = c1 + c2 x + c3 y + c4 x 2 + c5 x y + c6 y 2 (16.41)
has six degrees of freedom (c1 , . . . , c6 ) and hence there is no unique way of deter-
mining such a surface from the five height values in (16.40). In particular, if the
origin is taken at point 0, the coefficient c5 cannot be determined since the product
x y = 0 at all the five adjacent points sampled. Now, recalling from Sect. 3.1 that the
curvatures
∂2h ∂2h ∂2h
, ,
∂x 2 ∂y 2 ∂x∂ y
obey the Mohr’s circle transformation relations, we conclude that it is always possible
to choose a sufficiently large value for the undetermined coefficient c5 such that one
of the resulting principal curvatures should be negative, implying that the point in
question is a saddle point, rather than a summit. This ambiguity can be mitigated to
some extent by checking also the corner points in Fig. 16.9, but Greenwood (1984)
gives examples where a false summit might fail even this more stringent test.
16.4.2 Random Process Models
The measured profile h(x) can be regarded as a particular realization of a certain

random process, some of whose characteristics can be defined using statistical argu-
ments. To introduce this method, we must first define some of the mathematical
operations and functions involved.
Expected Value
The expected value of a random variable f (x) is defined as
L
1
f (x) = lim f (x)d x. (16.42)
L→∞ 2L −L
This definition is exact only if we have a realization of the random variable over
the infinite range −∞ < x < ∞, which here would correspond to an infinitely long
profilometer output. However, most of the quantities of interest will be such that the
expected value converges to a stable result over relatively short lengths.
Autocorrelation Function
We have seen that to decide whether a given point on a profile is a peak, we need
to determine whether it is higher than its two nearest neighbours. To this end, it is
useful to define the autocorrelation function
L
1
C(z) = f (x + z) f (x) = lim f (x + z) f (x)d x. (16.43)
L→∞ 2L −L
More generally, if f (x), g(x) are two random variables with zero mean value
f (x) = 0, g(x) = 0, the expected value of the product f (x)g(x) is a mea-
sure of the extent to which they are correlated.
Setting z = 0 in Eq. (16.43), we obtain
L
1
C(0) = lim f (x)2 d x = f (x)2 = σ 2 , (16.44)
L→∞ 2L −L
which defines the variance σ 2 or the standard deviation σ of f (x). We can also define
the normalized autocorrelation function
C(z)
ρ(z) = , (16.45)
C(0)
which is then unity when there is perfect correlation.

If two Gaussian variables x and y each have mean value zero and standard devi-
ation σ, their separate probability distributions are

1 x2 1 y2
Φ(x) = √ exp − 2 Φ(y) = √ exp − 2 . (16.46)
σ 2π 2σ σ 2π 2σ
However, if y is known to have the value y0 and if the normalized correlation between
x and y is ρ, then the conditional
probability, Φ(x | y0 ) of x has a mean value ρy0
and standard deviation σ 1−ρ2 — i.e.

1 (x − ρy0 )2
Φ(x | y0 ) = exp − 2 . (16.47)
σ 2π(1 − ρ2 ) 2σ (1 − ρ2 )
If the normalized autocorrelation function ρ(z) is known, the probabilities of the

inequalities (16.38) being satisfied at a given point can be assessed for a given hypo-
thetical sampling interval using (16.47) with the values ρ(), ρ(2). This approach
was pioneered by Whitehouse and Archard (1970), who also used profilometer data
to estimate ρ(z) for a ground surface. They found that the experimental measurements
could be fitted to an exponential curve of the form

z
ρ(z) = exp − , (16.48)
z 0
where z 0 is a parameter characterizing the horizontal length scale of the surface

which they christened the ‘correlation distance’.
Power Spectral Density [PSD]

To characterize the ‘waviness’ of the profile h(x), it is natural to think in terms of
calculating its Fourier transform
∞
h̃(ω) = h(x)exp(−ıωx)d x. (16.49)
−∞
The resulting expression will be complex and the ratio between the real and imaginary
parts arg(h̃(ω)) for some particular ω is related to the phase of that particular spectral
component. However, since the function h(x) is a random variable, these phases
are random, so that different realizations of the statistics will have different phases.
Furthermore, the phases will not be correlated even with those of the same realization
at infinitesimally different frequency, so the Fourier transform as defined in (16.49)
will generally not be continuous or differentiable.
By contrast, the magnitude of the complex Fourier transform converges on a
unique function of ω for a sufficiently large sample length L, so we define the power
spectral density or PSD as5
2
1 L
P(ω) = lim h(x)exp(−ıωx)d x . (16.50)
L→∞ 4πL −L
Now h(x) is a real function, so the complex conjugate of

L L
h(x)exp(−ıωx)d x is h(y)exp(ıω y)dy.
−L −L
Also, the squared magnitude of a complex quantity A + ı B is the product of the

quantity and its conjugate,
|A + ı B|2 = A2 + B 2 = (A + ı B)(A − ı B), (16.51)
so we conclude that
L L
1
P(ω) = lim h(x)h(y)exp{ıω(y − x)}d xd y. (16.52)
L→∞ 4πL −L −L
Changing variables through x − y = z, x = y +z, we have

L L−y
1
P(ω) = lim h(y + z)h(y)exp(−ıωz)dzdy, (16.53)
L→∞ 4πL −L −L−y
5 The reader should be warned that there is no overwhelming consensus on the correct numerical
multiplier on the definition of P(ω) and indeed similar differences occur in definitions of Fourier
transforms. The definition given here follows Nayak (1971, 1973) but differs by a factor of 2π from
that implied by Whitehouse and Archard (1970).
and changing the order of integration,

2L min(L ,L−z)
1 1
P(ω) = lim h(y + z)h(y)dy exp(−ıωz)dz.
L→∞ 2π −2L 2L max(−L ,−L−z)
(16.54)
In the limit, the quantity in square brackets [·] is recognizable as the autocorrelation
function (16.43), since non-zero contributions to the integral are made only for finite
values of z. It follows that
∞
1
P(ω) = C(z)exp(−ıωz)dz (16.55)
2π −∞
and hence that the power spectral density is the Fourier transform of the autocorre-
lation function. The Fourier inverse of (16.55) is
∞
C(z) = P(ω)exp(ıωz)dω, (16.56)
−∞
and hence, setting z = 0 and using (16.44),

∞
σ2 = P(ω)dω. (16.57)
−∞
Moments of the Distribution

Equation (16.57) is a special case of the integral
∞
mn = ω n P(ω)dω, (16.58)
−∞
which defines the nth moment of the PSD. In particular, it can be shown that the
variance of the profile slope σm2 is equal to the second moment, and that of the profile
curvature σκ2 is equal to the fourth moment, so that
∞
[h(z)] = σ = m 0 =
2 2
P(ω)dω (16.59)
−∞
∞
[h
(z)]2 = σm2 = m 2 = ω 2 P(ω)dω (16.60)
−∞
∞
[h(z)
]2 = σκ2 = m 4 = ω 4 P(ω)dω. (16.61)

−∞
It is also helpful to define the ‘bandwidth parameter’ (Nayak 1971)

m0m4
α= , (16.62)
m 22
which gives an indication of the range of frequencies in the spectrum. If the profile
comprises a single sine wave, α = 3/2, but for practical rough surface profiles, much
larger values are generally obtained.
Cosine Transforms
The definition (16.43) shows that the autocorrelation function C(z) for a random
profile must be real and symmetric [C(−z) = C(z)]. Using this result in (16.55), we
obtain
1 ∞
P(ω) = C(z) cos(ωz)dz. (16.63)
π 0
It follows that P(ω) is also real and symmetric, so from (16.56) [or by Fourier cosine
inversion], ∞
C(z) = 2 P(ω) cos(ωz)dω. (16.64)
0
Notice that although these expressions make use only of positive arguments, the
functions P(ω), C(z) are still defined over the entire range (−∞, ∞) and in particular
the power of the spectrum in some particular frequency band includes components
from both P(ω) and P(−ω).
Similar expressions can be obtained for the moments (16.58) when n is even. For
odd values of n, we obtain m n = 0.
PSD of a Surface
Equations (16.50), (16.55) define the power spectral density of a profile, but real
surfaces are two-dimensional and we can define the surface PSD as
∞ ∞
1
PS (ωx , ω y ) = C(ξ, η)exp(−ı(ωx ξ + ω y η)dξdη, (16.65)
4π 2 −∞ −∞
where
C(ξ, η) = h(x + ξ, y + η)h(x, y) (16.66)
is the correlation between points separated by a distance r = {ξ, η}. If the surface is
isotropic, Nayak (1973) shows that this expression reduces to
∞
1
PS (ω) = C(r )r J0 (ωr )dr, (16.67)
2π 0
where J0 (·) is the Bessel function of the first kind of order zero and

r= ξ2 + η2 ω= ωx2 + ω 2y . (16.68)
Also, since the profile defines the properties along a representative line on the sur-
face, the profile PSD [now distinguished by the symbol PP (ω)] and PS (ω) must be
derivable from the same autocorrelation function C(r ). Using this result, it can be
shown that
∞ ∞
k PS (k)dk 1 d k PP (k)dk
PP (ω) = 2 √ ; PS (ω) = − √ , (16.69)
ω k −ω
2 2 πω dω ω k 2 − ω2
(Nayak 1973).
The height variance can be expressed in terms of the surface PSD as
∞ ∞
σ =
2
PS (ωx , ω y )dωx dω y , (16.70)
−∞ −∞
and if the surface is isotropic,

∞
σ 2 = 2π ω PS (ω)dω. (16.71)
0
We also note that for a surface, the slope ∇h is a vector whose magnitude has variance
2 2 ∞
∂h ∂h
+ = 2m 2 = 2 k 2 PP (k)dk, (16.72)
∂x ∂y −∞
where we emphasize that m 2 is defined in terms of the profile PSD through

Eq. (16.60). Alternatively, using (16.69), we can obtain the equivalent relation
2 2 ∞
∂h ∂h
+ = 2π ω 3 PS (ω)dω. (16.73)
∂x ∂y 0
It is important to be clear which of the profile and surface PSD is involved in a

given calculation, particularly since much of the literature on the theory of random
processes is concerned with time series, where this distinction does not arise. A
narrow wavenumber band of the surface PSD (ω, ω + δω) can be regarded as the
superposition of an ensemble of ‘plane’ wavy surfaces of the same wavenumber
ω, but with different orientations. This form of superposition is analogous to the
Smirnov–Sobolev transformation of Sect. 6.6, except that the phases are here random
and uncorrelated. The resulting surface will be isotropic if the density of the waves is
independent of orientation. If a profile of such a surface is now generated by sampling
it along an arbitrary straight line, the component waves will almost all appear to have
longer wavelengths [smaller wavenumber] than the superposed plane waves, because
the sampling line is inclined to the lines of greatest slope. The resulting profile PSD
will therefore have non-zero content throughout the range (0, ω), though in a practical
measurement, there will necessarily be a lower cut-off associated with the finite length
of the sample.
16.4.3 Determining Asperity Parameters
In order to use an asperity model to predict features of rough surface contact, we need
to be able to determine the statistical distributions of asperity heights and curvatures
from measured surface profiles. Longuet-Higgins (1957a, b) showed that the number
of summits per unit area Ns is given by

1 m4
Ns = √ , (16.74)
6π 3 m2
where the moments of the distribution m 2 , m 4 are defined in Eqs. (16.60), (16.61), An
alternative proof of this result is given by Nayak (1971). Alternatively, the summit
density can be can be found by counting the number of peaks per unit length N p in
the profile and then using the relation
2π N p2
Ns = √ (16.75)
3 3
(Longuet-Higgins 1957b; Nayak 1971; Greenwood 1984).

These summits have a Gaussian height distribution with standard deviation

0.8968 1/2
σs = σ 1 − (16.76)
α
(Bush et al. 1976), where α is given by (16.62). However, since α 1 for most
practical surfaces, σs is very close to the standard deviation σ of the complete set of
surface heights.
The mean curvature κs at the summits can be used to estimate the appropriate
asperity radius R. Bush et al. (1976) and McCool (1987) give

1 8 m4
κs = = , (16.77)
R 3 π
which is approximately equal to 1.5σκ , from (16.61). Whitehouse and Phillips (1978,
1982) established fairly general rules for determining the variances of summit heights
and curvatures, and an alternative proof of these results was given by Greenwood
(1984). In particular, the standard deviations for peak [two-dimensional] curvature
and summit curvature are about 0.6σκ and 0.5σκ respectively. An important conclu-
sion from these investigations is that the resulting parameter values are significantly
influenced by the sampling interval used in digitizing the profilometer output.
Bush et al. (1975) used their results to refine Greenwood and Williamson’s theory
and relate the parameters to the profile PSD. In particular, they predicted that at
sufficiently large separations h 0 in Eqs. (16.21)–(16.23), and hence low nominal
pressures pnom , the proportion of the nominal contact area Anom that is in actual
contact is
A pnom π
≈ , (16.78)
Anom E∗ m2
where m 2 is the mean square profile slope defined in Eq. (16.60). This result provides
some motivation for the definition (16.36)2 of the plasticity index ψ.
16.5 Fractal Surfaces
Modern surface measuring systems give extremely good resolution both horizontally
and vertically and this leads to embarrassing results for classical asperity model
idealizations. As the sampling interval is reduced, more and more fine scale features
become apparent on the surface, so that what might have been characterized as a
single asperity in a coarse description becomes revealed as a collection of smaller
asperities on the foothills of a larger one.
16.5.1 Archard’s Model
This way of thinking was foreshadowed in a remarkably prescient paper by Archard

(1957) in which he considered the contact of a plane surface with a sphere carrying
smaller spherical asperities on its surface, as shown in Fig. 16.10a.
Archard’s motivation for this investigation was similar to that of Greenwood and
Williamson. He was trying to explain how the coefficient of friction could be constant,
when in an elastic contact the force-contact area relation was nonlinear. His model
showed that though for a single sphere we have A ∼ P 2/3 , for the system of spherical
asperities on a spherical base, we obtain A ∼ P 8/9 . He then added a third layer of even
smaller spherical asperities on these asperities as shown in Fig. 16.10b and obtained
A ∼ P 26/27 , so it is easy to see that a linear relationship would be obtained if this
multiscale structure were to be extended indefinitely.
16.5.2 Self-affine Fractals and the Fractal Dimension
What Archard had invented with the model of Fig. 16.10 is what would now be called a
fractal surface. In other words, successively smaller self-similar features are revealed
as we examine the surface on smaller and smaller scales and the process continues
theoretically without limit. Fractal processes were extensively investigated in the
1980s, notably by Mandelbrot (1982, 1985). There are numerous examples in nature
of processes which exhibit fractal characteristics at least over a significant range
16.5 Fractal Surfaces 353
(a) (b)
Fig. 16.10 Archard’s multiscale asperity model: (a) a two-surface, (b) a three-scale surface
of scales. Examples include the coastline of continents or the dendrites generated

during solidification.
Surface profiles typically approximate self-affine fractals, meaning that different
scalings are required on the horizontal and vertical axes, as shown in Fig. 16.11, in
order to make a magnified image of the profile look like itself. Here the profile is
magnified by a factor λ on the horizontal scale, but by λ H on the vertical scale.
The parameter H is known as the Hurst exponent and it generally lies in the range
0 < H ≤ 1. Self-similarity, as in the Archard profile of Fig. 16.10 would correspond
to the limiting case H = 1. For all other cases, the magnification on the vertical scale
is less than that on the horizontal scale, implying that average surface slopes become
larger at finer scales and smaller at coarser scales.
A related parameter is the fractal dimension D, which for a two-dimensional
profile is given by D = 2− H . To give a strict definition of the fractal dimension, we
can imagine creating a piecewise linear approximation to the profile such that each
straight line segment has the same length and the number of such segments over
the total length is N . If the exact total length of the curve is bounded and equal to
L ∞ , we should find that for sufficiently small , the product L ≡ N → L ∞ as → 0.
However, for fractal profiles, we find
N ∼ −D so that L ∼ 1−D , (16.79)
and since D > 1 [except for self-similar profiles], the calculated total length of the
curve increases without limit as is decreased.
Fig. 16.11 Self-affine

scaling needed to make a
segment of the profile
statistically similar to the
original profile
H
λ
λ
Similar arguments can be applied to areas and volumes. If the measure M of an

n-dimensional object varies as
M ∼ λ ; → 0, (16.80)
where is a dimensionless parameter defining the discrete scale at which the mea-
surement is performed, the fractal dimension is
D = n − λ. (16.81)
For self-affine functions, the fractal dimension and the Hurst exponent of Fig. 16.11
are related by the equation
D + H = n + 1, (16.82)
so the fractal dimension of the area [n = 2] of a rough surface is D = 3− H and lies

in the range 2 ≤ D < 3, where D = 2 would correspond to a surface in which the total
area is bounded in the limit → 0.
16.5.3 The Weierstrass Function
A simple expression exhibiting fractal properties is the Weierstrass function defined

by
∞
(D−2)n 2πγ n x
h(x) = h 0 γ cos . (16.83)
n=0
L0
At first sight, this looks like a Fourier series, but the wavenumbers of successive
terms increase geometrically rather than arithmetically — i.e. with γ n rather than n.
In the special case D = 1, the waves are geometrically similar to each other and the
profile is self-similar, but for D > 1 the profile is fractal.
Notice that the cosines in Eq. (16.83) are bounded between ±1 and hence
for all x,
∞

h(x) 1
−c ≤ <c where c= γ (D−2)n = (16.84)
h0 n=0
(1 − γ D−2 )
and since γ > 1 and D < 2, c is bounded. A similar argument shows that the
contribution E(m, x) to the value of h(x)/ h 0 by the terms n > m is bounded in
the range −cγ (D−2)m < E(m, x) < cγ (D−2)m and hence that the series (16.83) is
convergent for all x.
By contrast, the derivative
∞

2πh 0 (D−1)n 2πγ n x
h (x) = − γ sin , (16.85)
L 0 n=0 L0
which corresponds to the slope of the profile, is unbounded for all x, since D > 1
and hence the amplitude of the sine waves in the series increases with n. Thus, the
Weierstrass function has the remarkable property of being bounded and continuous
but non-differentiable at all points x.
The Weierstrass–Mandelbrot Function
The Weierstrass function has no smallest scale, but it does have a largest scale,
corresponding to the first term in the series. Mandelbrot extended the concept to
define the function
∞
2πγ n x
h0 γ (D−2)n cos , (16.86)
n=−∞
L0
now known as the Weierstrass–Mandelbrot function, which also has no smallest or

largest scale. In modelling rough surfaces, we generally do need to have a largest
scale, and this must be significantly smaller than the geometric dimensions of the
contact area if roughness effects and macroscopic deformations are to be separable.
Even when this condition is not satisfied, the dimensions of the contact area provide a
natural upper limit to the length scale, and nothing much is achieved by adding larger
wavelength terms. However, the lack of a largest scale in the Weierstrass–Mandelbrot
function can be advantageous in discussing general features of fractal surfaces.
A Continuous Spectrum
The Weierstrass function comprises a set of waves at discrete wavenumbers and
hence its PSD consists of a set of Dirac delta functions
∞
h 20 2(D−2)n 2πγ n 2πγ n
P(ω) = γ δ ω− +δ ω+ . (16.87)
4 n=0 L0 L0
Real random profiles have a continuous spectrum rather than a series of delta func-
tions, but if γ is not too far from unity we can ‘average’ the delta functions over the
intervening wavenumber ranges to define a profile with the continuous PSD
5−2D
L 0 h 20 2π
PP (ω) = , (16.88)
2π ln(γ) ωL 0
(Berry and Lewis 1980). This demonstrates that the fractal dimension of a profile
can be determined by making a logarithmic plot of P(ω) and equating the [negative]
slope to (5−2D), as shown in Fig. 16.12. The slope can also be expressed in terms
of the Hurst exponent H as (2H +1).

the fractal dimension D and
the slope of the power ln[PP (ω)]
spectral density curve 1
5 - 2D
ln(ω)
A corresponding surface PSD can be determined from (16.69)2 . In particular, we

find that PS (ω) ∼ ω 2DS−8 , where D S (2 < D S < 3) is the surface fractal dimension.
Alternatively, in terms of the Hurst exponent, PS (ω) ∼ ω −2−2H .
16.5.4 Generating Realizations of Fractal Profiles

and Surfaces
The Weierstrass function provides a convenient way to define a fractal profile for
numerical studies, in which case we need to randomize the phases of the various
terms by defining
∞

(D−2)n 2πγ n x
h(x) = h 0 γ cos + φn , (16.89)
n=0
L0
where φn is a random number defined with a uniform distribution on the interval

(0, 2π). Alternatively, we can use the original expression (16.83) but sample it in
a range well away from x = 0, compared with the fundamental wavelength L 0 .
Figure 16.3 in Sect. 16.2 was obtained in this way, using a fractal dimension D = 1.5
and γ = 1.2.
This technique can be generalized to three-dimensional surfaces by superposing
terms of the form (16.83) with random Cartesian orientations. For example, following
Yan and Komvopoulos (1998), we can write
∞
1
M
2πγ n r cos(θ − αm )
h(r, θ) = √ hm γ (D−3)n cos + φmn , (16.90)
M m=1 n=0
L0
where (r, θ) are polar coordinates defining position on the surface, αm = mπ/M,
and φmn are a set of random phases uniformly distributed in the domain (0, 2π). The
coefficients h m provide a means of adjusting the degree of anisotropy of the surface,
an isotropic surface corresponding to the case where h m = h 0 is a constant.
The Random Midpoint Displacement Method [RMD]

The Weierstrass function defines a continuous, albeit non-differentiable, function of
the real variable x, [or variables x, y in the case of Eq. (16.90)]. This can be used to
define the heights h i j at a discrete set of grid points, but an alternative direct method
of generating such heights is to use the Random Midpoint Displacement [RMD]
method (Russ, 1994).
For a profile, the procedure is very simple. We first define a random function
f (i, σ) with a Gaussian distribution and standard deviation σ through the relation
√
f (i, σ) = σ 2 erf −1 (Ni ), (16.91)
where Ni is a random number uniformly distributed in (−1, 1) and erf −1 (·) is the
inverse error function. Usually, the length of the segment modelled will be large
compared with the correlation distance, so the heights h(0), h(L) of the two end
points can be assumed uncorrelated and hence chosen as
h(0) = f (1, σ0 ) h(L) = f (2, σ0 ). (16.92)
We next compute the height at the midpoint x = L/2 as

L h(0) + h(L) σ0
h = + f (3, σ1 ) where σ1 = , (16.93)
2 2 2H
and we recall that H = (2 − D) is the Hurst exponent defined in Sect. 16.5.2 and
Fig. 16.11. In other words, the midpoint height is chosen as the result of a linear
interpolation, perturbed by a Gaussian random variable with a reduced standard
deviation. The procedure is then repeated to find the quarter points etc., starting from
the mean value of the two adjacent points. With each reduction in scale, the standard
deviation is reduced by the factor 2(2−D) , which ensures that the resulting profile will
have a fractal dimension D.
The Diamond-Square Algorithm
The RMD algorithm can be adapted to define three-dimensional fractal surfaces.
Suppose at some stage in the process, we know the heights at the grid points identified
by the solid circles in Fig. 16.13.
We determine the height of the point A by averaging the heights at the four
surrounding points and adding a random perturbation — i.e.
h1 + h2 + h3 + h4
hA = + f (i, σ). (16.94)
4
Once h A and h B have been obtained by this procedure, we then apply a similar
Fig. 16.13 Strategy for the

diamond-square algorithm
argument to the dashed square (or diamond) A2B4 to determine the height at its
centre point P as
h A + h2 + h B + h4 σ
hP = + f ( j, σ1 ) where σ1 = . (16.95)
4 2 H/2
Notice that the standard deviation is here reduced by the factor 2 H/2 , since√the
side of the dashed square is smaller than that of the original square in the ratio 1/ 2.
Now P is the midpoint of the side 24 of the original square, so all the other midpoints
can be found by this procedure, after which the whole algorithm is applied again at
the next smaller scale.
Some adjustment is needed at the edge of the modelled region, since the diamonds
that have an edge as a diagonal will lack one corner. This can be accommodated by
(i) averaging just the three available corner heights in Eq. (16.95), and (ii) discarding
the edge values in the final modelled surface.
The RMD and diamond-square algorithms are widely used in the computer
graphics industry to generate natural features such as mountain ranges and cloud
formations.
16.6 Contact of Fractal Surfaces 359
16.6 Contact of Fractal Surfaces
16.6.1 Majumdar and Bhushan’s Theory
The earliest explicit fractal theories of rough surface contact were developed by
Majumdar and Bhushan (1990, 1991, 1995) and are related to the bearing area con-
cept of Sect. 16.2.1. More precisely, these authors argued for a connection between
the distribution of ‘islands’ [areas above a given height] shown in Fig. 16.14 and an
equivalent asperity model, in which each island represents an asperity, so there are
some big ones and larger numbers of smaller ones.
The surface is then idealized by replacing each island by a paraboloidal asperity
with the same bearing area, and whose volume above ‘sea level’ is equal to that of
the exact surface, as shown in Fig. 16.15. If a rigid plane is now brought down to
this level, and if the contact remains elastic, each asperity contact will be defined by
an appropriate Hertzian solution, and in particular, the contact area will be exactly
one half of the interpenetration area [see Sect. 5.2]. Thus, Majumdar and Bhushan
argued that the total area of actual contact would be one half of the bearing area for
a given indentation depth, and that the contact would comprise a set of areas each
equal to one half of the corresponding island area.
The distribution of island areas for a fractal surface has been extensively studied
in other applications such as geophysics, and in general, it is believed to follow
Korcak’s law (Korčák 1938). This states that the number of areas larger than a given
area A0 is of the power-law form
−q
N (Ac > A0 ) = C A0 , (16.96)
where the exponent q lies in the range 0 < q < 1 and C is a constant. This law is
necessarily modified at large A0 , since there must be some largest island [say of area
A1 ], but the predicted total area is bounded. To prove this, note that the number of
areas between A0 and A0 +δ A0 is
dN −q−1
− δ A0 = Cq A0 δ A0 (16.97)
d A0
Fig. 16.14 ‘Islands’ defined islands

by the intersection of a
horizontal plane [‘sea level’]
with a rough surface
Fig. 16.15 Replacement of equivalent

a typical island by a smooth asperity island
parabolic asperity in
Majumdar and Bhushan’s
theory
and hence the total area is

A1 (1−q)
−q−1 Cq A1
A= Cq A0 A0 d A0 = . (16.98)
0 (1 − q)
However, the total number of contact areas is unbounded, since it would be given by
A1
−q−1
N= Cq A0 d A0 (16.99)
0
and this integral is unbounded for 0 < q < 1. Thus, Majumdar and Bhushan’s theory
predicts that the total contact area is bounded, but the average size of a contact area
is zero. The distribution is dominated by an infinite number of infinitesimal areas.
16.6.2 Elastic Contact for a Fractal Surface
The contact areas predicted by Majumdar and Bhushan’s theory are not uniformly
distributed over the nominal contact area, but instead are clustered. In particular,
numerous small contact areas will be found near larger areas, much as small islands
are typically found off the coast of continents. As a result, it is really not plausible
to treat them as independent elastic contacts, as is implied by the comparison with
the Hertzian theory.
Some indication of what to expect in the elastic contact of fractal surfaces is pro-
vided by a numerical study by Borri-Brunetto et al. (1997). They used the random
midpoint displacement algorithm of Sect. 16.5.4 to define a fractal surface over a
square nominal contact area and then subdivided the nominal area into square ele-
ments over each of which the contact pressure was assumed to be uniform. They
then solved the elastic contact problem using progressively refined meshes. For a
coarse mesh, a few separated actual contact areas were identified at a given total
normal force F, but as the mesh was refined, these were resolved into clusters of
smaller areas as shown in Fig. 16.16. The total area of actual contact was reduced
by each progressive refinement and the results approximately followed an inverse
power law, suggesting that if the process could be extended indefinitely [obviously
an impossibility with a numerical solution], the predicted total contact area would
Fig. 16.16 Effect of mesh refinement on the elastic contact area between a rough surface and a
rigid plane, after Borri-Brunetto et al. (1997)
be zero! This is in sharp contrast to Majumdar and Bhushan’s theory which predicts
a bounded total contact area.
The clustering of contact areas in Fig. 16.16 also highlights the greatest weakness
of asperity model theories, in that each asperity is assumed to act independently. In
practice, the force on one asperity in a dense cluster will contribute to the displace-
ment at neighbouring asperities and hence reduce the contact area and total normal
force associated with a given value of separation h 0 . By neglecting interaction, asper-
ity model theories also tend to overestimate the electrical conductance and hence also
the incremental stiffness |d P/dh 0 | for a given force P.
16.6.3 The Weierstrass Profile
Ciavarella et al. (2000) investigated elastic contact for a profile defined by the function

m
2πγ n x
h(x) = h 0 γ (D−2)n cos , (16.100)
n=0
L0
which represents the Weierstrass function of Eq. (16.83) truncated at n = m. In

particular, they defined the function Φm ( p) such that the probability of an arbitrarily
chosen point x experiencing a contact pressure in the range ( p, p+δ p) is Φm ( p)δ p.
They then considered the effect on this function of adding one extra term in the series.
For a single sine wave [m = 0], the problem reduces to the Westergaard problem
of Sect. 6.5.6. If Eq. (6.97) for contact pressure p(x) is inverted to give x( p) as a
function of p, the cumulative probability — i.e. the probability of contact pressure
exceeding p is then 2x( p)/L, since p is a monotonically decreasing function of
x in 0 < x < L/2. Using this approach, (Ciavarella et al. 2000) determined that
Φ0 ( p) = f ( p, p̄, p ∗ ) where for full contact ( p̄ > p ∗ ),
1
f ( p, p̄, p ∗ ) = p̄ − p ∗ < p < p̄ + p ∗ , (16.101)
π p ∗2 − (p − p̄)2
whereas for partial contact ( p̄ < p ∗ ),

p
f ( p, p̄, p ∗ ) =
π {( p ∗ − p̄)2 + p 2 }{2 p̄[ p ∗ − p̄ + ( p ∗ − p̄)2 + p 2 ] − p 2 }

0 < p < 2 p ∗ p̄, (16.102)
where ∗
π E h0
p∗ = . (16.103)
L0
In each case, outside the stated ranges of p, the probability is zero.

If γ is sufficiently large, so that the pressure pm−1 at scale m−1 changes relatively
little over one wavelength of the next finer scale m, we can use Eqs. (16.101), (16.102)
with p̄ = pm−1 , p ∗ = pm∗ to define the conditional probability that a point x at pressure
pm−1 at scale m −1 will experience a pressure pm at scale m. This is essentially the
same approximation as that used by Archard (1957) in his analysis of the multiscale
surface of Fig. 16.10.
We obtain
Φ( pm | pm−1 ) = f ( pm , pm−1 , pm∗ ), (16.104)
where
∗
π E h 0 γ (D−1)m L0
pm∗ = since h 0 → h 0 γ (D−2)m L 0 → . (16.105)
L0 γm
The probability Φm ( pm ) can then be found by iteration using the relation

∞
Φm ( p m ) = Φ( pm | pm−1 )Φm−1 ( pm−1 )dpm−1 , (16.106)
0
starting from Φ0 ( p0 ).
As m → ∞, the number of contact segments increases and the proportion of the
profile in contact decreases, as found numerically by Borri-Brunetto et al. (1997).
However, except at very low values of m, both of these quantities are linearly pro-
portional to the nominal pressure p̄, as predicted by Archard (1957) [Sect. 16.5.1].
Ciavarella et al. (2000) showed that the fractal dimension of the total actual con-
tact area [strictly the total contact length] is (2− D). Also, each contact represents
progressively smaller proportions of their respective waves, so that the individual
contacts approach more closely to the Hertzian limit as m increases.
Plastic Deformation
As the predicted area of actual contact decreases, the mean contact pressure increases,
so in many cases it is reasonable to assume that at some stage the process will be
limited by plastic deformation.
Gao et al. (2006) used the finite element method to solve the contact problem
for the sinusoidal surface of Fig. 6.6, using an elastic–plastic constitutive law. They
characterize the behaviour in terms of the dimensionless parameter
∗
E h0
ψ= , (16.107)
SY L
where SY is the yield stress of the material. We expect the contact to be predominantly
elastic as long as the maximum contact pressure is less than the hardness H of the
material, which is of the order H ≈ 2.8SY (Tabor 1951). In the elastic solution, full
contact first occurs for p̄ = p ∗ and involves a maximum contact pressure
∗
∗2π E h 0
pmax = 2p = ≈ 2.2ψ H, (16.108)
L
so we anticipate significant modification to the contact pressure in the partial contact
régime due to plasticity for ψ > 0.5. Gao et al. (2006) found that for values of ψ of
order unity, an initially almost Hertzian distribution at each contact evolves to a more
uniform distribution with a mean value close to H . However, for larger values of ψ,
the confining effect of the plastic deformation at adjacent contacts causes an increase
in mean pressure as the proportion of the profile in contact approaches unity, and this
can reach a maximum of about 5.8SY for large ψ.
Gao and Bower (2006) used these results and the techniques described in
Sect. 16.6.3 to investigate the elastic–plastic contact problem for the truncated Weier-
strass profile. If we define a scale-dependent parameter ψn by replacing h 0 , L 0 by the
amplitude h 0 γ (D−2)n and wavelength L 0 /γ n of the n-th term in the series, we obtain
∗
E h 0 γ (D−2)n
ψn = = ψ0 γ (D−1)n , (16.109)
SY L 0 γ −n
and this increases with n, showing that plastic deformation becomes more dominant
as we go to finer scales. In particular, Gao and Bower (2006) found that the total
actual contact area is determined by the condition that the mean contact pressure
be 5.8SY for sufficiently large truncation limit m. However, although necessarily
some of the contacts now reach the full contact state and hence merge with adjacent
contacts, the number of distinct contact areas per unit length still increases without
limit as m is increased.
16.6.4 Persson’s Theory
A serious limitation of both elastic and plastic Weierstrass solutions is that they
depend on the scale-separation parameter γ being sufficiently large for the pressure
at scale m to be approximately uniform over a complete wavelength at scale m +1.
Ideally, we would like to be able to determine the change in Φ( p) associated with
an infinitesimal increment P(ω)δω in a continuous PSD, but this cannot be found
using the Westergaard solution, or the elastic–plastic solution of Gao et al. (2006).
Persson (2001) has developed a theory based on the use of the ‘full contact’
elastic solution (16.101) to determine this incremental change and hence define the
probability function Φ( p) as a function of the upper truncation limit ωh in PS (ω).
Here, we shall discuss the theory in the form developed by Manners and Greenwood
(2006).
We first recall from Eqs. (6.65), (6.68) that under elastic deformation, a sinusoidal
pressure distribution p(x) = p0 cos(ωx) produces a surface slope
∂u z 2 p0 sin(ωx)
(x, y, 0) = − . (16.110)
∂x E∗
Now a rough surface defined by the height function h(x, y) could be represented as
a random superposition of uncorrelated plane sine waves of various orientations, so
if such a surface were flattened by a pressure distribution with mean value p̄, we
would have ∗
E 2
V ≡ [ p(x, y) − p̄]2 = |∇h|2 . (16.111)
4
For an isotropic surface whose PSD is PS (ω), the variance of the surface slope is
given by Eq. (16.73), from which we obtain
∗2 ∞
πE
V = ω 3 PS (ω)dω. (16.112)
2 0
If the PSD is truncated at some upper cut-off wavenumber ωh , we can write

∗2 ωh
πE
V (ωh ) = ω 3 PS (ω)dω. (16.113)
2 0
Manners and Greenwood (2006), following Persson (2001), then consider the effect
of changing ωh by a small increment, leading to a corresponding small change δV
in V (ωh ). Since the waves are uncorrelated, we can write
∞
Φ( p, V + δV ) = Φ( p − δ p, V )Φ(δ p, δV )dδ p, (16.114)
−∞
where in view of the definition (16.111), Φ(δ p, δV ) is a Gaussian random process

of mean value zero [since the mean pressure p̄ is unchanged] and variance δV —
i.e.
1 δ p2
Φ(δ p, δV ) = √ exp − . (16.115)
2πδV 2δV
We expand Φ( p−δ p, V ) as a Taylor series
∂ δ p2 ∂ 2
Φ( p − δ p, V ) = Φ( p, V ) − δ p Φ( p, V ) + Φ( p, V ) − . . . (16.116)
∂p 2 ∂ p2
and substitute this expression into (16.114)). The integrals can then be evaluated6 to
give
δV ∂ 2 2
Φ( p, V + δV ) = Φ( p, V ) + Φ( p, V ) − O δV , (16.117)
2 ∂ p2
and proceeding to the limit δV → 0, we conclude that the probability distribution

function for pressure Φ( p, V ) satisfies the partial differential equation
∂2Φ ∂Φ
=2 , (16.118)
∂ p2 ∂V
which is clearly recognizable as the diffusion equation.
6 Notice that the integrals involving odd powers of δ p evaluate to zero.

The Boundary Condition

So far the derivations are based entirely on the full contact solution, and indeed
Manners and Greenwood (2006) note that in this case one might reasonably appeal
immediately to the fact that the contact pressure distribution under full contact is
Gaussian with mean value p̄ and variance V [see Eq. (16.111)] and hence that

1 ( p − p̄)2
Φ( p, V ) = √ exp − , (16.119)
2πV 2V
which clearly satisfies (16.118).

However, Eq. (16.119) implies a non-zero probability of negative [i.e. tensile]
pressures which are consistent and indeed necessary consequences of the full con-
tact assumption with a ‘sufficiently rough’ surface, but which violate the Signorini
inequality. A key ingredient of Persson’s theory is that the probability Φ( p) must
tend to zero as p → 0, so the boundary condition Φ(0, V ) = 0 must be applied to
Eq. (16.118).
The basis for this condition is that as long as ωh is finite, the profile is differentiable
[‘smooth’], so at the edges of all contact regions, the contact pressure must tend to
zero with square-root bounded form, as shown in Sect. 10.1.3 and Eq. (10.16). The
local relation p = Br 1/2 implies r = ( p/B)2 and hence
dr 2p
= 2 →0 as p → 0. (16.120)
dp B
Notice that this relation also holds for the partial contact Westergaard solution dis-
cussed in Sect. 16.6.3. In particular, Eq. (16.102) shows that f ( p, p̄, p ∗ ) → 0 as
p → 0.
The ‘initial condition’ for Eq. (16.118) is clearly Φ( p, 0) = δ( p− p̄), since when
V = 0 there is no roughness and the pressure everywhere is equal to p̄. Using these
Fig. 16.17 Probability distribution Φ( p) for contact pressure p predicted by Persson’s theory [solid
line]. The chain-dotted line represents the corresponding ‘full contact’ solution and the dashed line
represents the second exponential term in Eq. (16.121)
results, the unique solution of (16.118) is

1 ( p − p̄)2 ( p + p̄)2
Φ( p, V ) = √ exp − − exp − . (16.121)
2πV 2V 2V
The proportion of the surface in contact can then be determined as

∞
A(V ) p̄
= Φ( p, V )dp = erf √ . (16.122)
Anom 0 2V
The distribution (16.121) is illustrated by the solid line in Fig. 16.17. For a fractal
surface, V increases without limit as the upper limit ωh in Eq. (16.113) increases,
and at sufficiently large V [or more precisely, when the dimensionless parameter
p̄ 2 /V 1], we can approximate (16.121), (16.122) as

p̄ p 2 p2
Φ( p, V ) ≈ exp − (16.123)
V πV 2V

A(V ) 2
≈ p̄ , (16.124)
Anom πV
both of which are linearly proportional to the nominal pressure p̄. Substituting for V
from (16.111) and noting that p̄ is the nominal contact pressure pnom , we can write
(16.124) in the form
A 2 pnom
≈ ∗√ . (16.125)
Anom E π m2
which has exactly the same form as Eq. (16.78), obtained by Bush et al. (1975) using
an asperity model. However, the two expressions differ in the numerical multiplying
factor, with Bush et al. predicting a contact area that is greater than that of Persson
by a factor of π/2 ≈ 1.57.
Equation (16.124) implies that the average contact pressure in the actual contact
areas ∗ √
P Anom p̄ E σm π
pave ≡ = ≈ , (16.126)
A A 2
using (16.60). Thus, the plasticity index defined in (16.36)2 is, according to Persson’s
theory, equivalent to
∗
E σm 2 pave
ψ= = √ . (16.127)
H H π
Comparison with Numerical Solutions

Equations (16.121), (16.122) depend on the surface roughness only through V as
defined in Eqs. (16.111), (16.113). For example, under full contact conditions, the
contact pressure for the truncated Weierstrass series (16.100) is
∗
π E h 0 (D−1)n
m
2πγ n x
p(x) = p̄ + γ cos (16.128)
L 0 n=0 L0
and since the sine waves are uncorrelated,

∗ 2
2 1 π E h0
m
V = p(x, y) − p̄ = γ 2(D−1)n . (16.129)
2 L0 n=0
Ciavarella et al. (2006b) used this result and Persson’s equation (16.122) to compare
the predicted contact area with the result from the analysis of Ciavarella et al. (2000)
and Sect. 16.6.3, which uses the Westergaard partial contact solution, and the method
of Archard (1957) for apportioning the contact pressure at successive scales. As
might be expected, the partial contact solution predicts a lower total contact area
at any given m, but both results predict the same fractal dimension for A, which
is (2 − D). Ciavarella et al. (2006b) also performed numerical studies for profiles
defined by the Weierstrass series, using a boundary element method. Their results
for actual contact area exceeded both Persson and ‘Archard’ predictions, but showed
the same fractal dependence on m.
Hyun et al. (2004) used the finite-element method to find a numerical solution for
elastic contact with a fractal surface generated by the diamond-square algorithm of
Sect. 16.5.4. The fractal is of course necessarily truncated by the finite mesh size of the
model. They confirmed that the contact area is approximately linearly proportional
to normal force at low nominal pressures with a multiplier that lies between the
predictions of Persson [Eq. (16.122)] and Bush et al. [Eq. (16.78)].
More specifically, Hyun et al. (2004) found results close to Bush’s predictions
for surfaces with large fractal dimension (D ≈ 2.6, H ≈ 0.4) and closer to Persson’s
predictions for lower fractal dimension [D ≈ 2.1, H ≈ 0.9]. A qualitative explanation
of this trend is that long wavelength components become relatively more significant
as D decreases [H increases], and this favours the generation of clusters of contact
areas, rather than sparse distributions. This, in turn, tends to reduce the total actual
contact area relative to that predicted by a ‘non-interacting’ asperity model theory,
as discussed in Sect. 16.6.2.
16.6.5 Implications for Coulomb’s Law of Friction
Measurements of frictional forces using the atomic force microscope [AFM] have
suggested that, at the nanoscale, Bowden and Tabor’s assumption of a constant shear
traction is a reasonable approximation. In other words, the friction force is approx-
imately proportional to the actual contact area at this scale (Carpick et al. 1996;
Enachescu et al. 1999). This result, coupled with Persson’s equation (16.122) there-
fore provide a possible explanation of Coulomb’s friction law.
However, we saw in Sect. 16.3 that the assumption of an exponential distribution

of asperities implies that the total frictional force Q will be linearly proportional to the
total normal force P, regardless of the relation between tangential and normal force at
the asperity scale. Essentially similar arguments can be applied to multiscale models,
including the Archard model of Fig. 16.10, the Weierstrass model of Sect. 16.5.3, and
Persson’s theory.
For example, suppose that the local conditions at the nanoscale can be character-
ized by a contact pressure p and that the local tangential traction is some arbitrary
function q( p) of p. The total tangential force can then be estimated using the large
V approximation (16.123) as
∞
∞
Anom p̄ 2 p2
Q = Anom Φ( p, V )q( p)dp ≈ p exp − q( p)dp
0 V πV 0 2V
∞
P 2 p2
≈ p exp − q( p)dp, (16.130)
V πV 0 2V
showing that Q is linearly proportional to the total normal force P. Similar arguments
can be used for the case where the normal contact is elastic–plastic (Barber 2013a).
For the Archard and Weierstrass models, linearity depends on there being a sufficient
number of scales, suggesting that Coulomb’s law is a consequence of the multiscale
nature of rough surfaces. However, Persson’s theory predicts linearity whenever V
is sufficiently large for the approximation (16.123) to be appropriate, and hence
p̄ p̄
√ = ∗ 1, (16.131)
2V E σm
using (16.111). Thus, in a sense, linearity of the friction law depends on the nominal
pressure being ‘sufficiently small’ or the surface being ‘sufficiently rough’.
16.7 Adhesive Forces
Fractal descriptions imply the existence of arbitrarily fine scale topographical detail
and this raises the question as to the extent to which van der Waal’s forces, discussed
in Chap. 12, influence the contact of rough surfaces. Experiments (Fuller and Tabor
1975) show that relatively small roughness amplitudes are sufficient to destroy adhe-
sive effects by reducing the area of actual contact, but recent investigations, inspired
by the ability of geckos to walk on vertical surfaces (Autumn et al. 2002), have
suggested that appropriate nanoscale topography can actually enhance adhesion.
16.7.1 Asperity Model Predictions
The effects of adhesion can be incorporated in asperity models merely by changing

the asperity force function f P (z) in Eq. (16.21). For example, if the asperities are
assumed to be spherical with summit radius R, we might use the JKR solution
represented in Fig. 12.2, which is well approximated by the relation
fP 5
fˆP (ẑ) ≡ = − − 1.1 ẑ + 0.43ẑ 3/2 , (16.132)
π RΔγ 6
where
3π 2/3
ẑ = Δ̂ − Δ̂ A and Δ̂ A = − (16.133)
4
is defined by the point A in Fig. 12.2.
If this expression is used with the exponential distribution of Eq. (16.24), the total
force is obtained as
h0
P = π N0 RΔγexp − f (λ), (16.134)
σ
where 1/3
1 R(Δγ)2
λ= (16.135)
σ E∗ 2
and the function f (λ) is illustrated in Fig. 16.18. An alternative expression for λ is
με/σ, where the Tabor parameter μ is evaluated at the asperity scale.
Since the number of asperities N0 must be positive, we conclude that the sign
of the force P is the same for all separations h 0 . Thus, the pull-off force is zero if
λ < 0.48 [i.e. if the surface is sufficiently rough], but for larger values of λ, the force
is always tensile. In a sense, λ can be regarded as the inverse of an ‘adhesion index’
analogous to the plasticity index of Sect. 16.3.3. It determines whether asperities in
Fig. 16.18 The function

f (λ) from Eq. (16.134)
Fig. 16.19 The pull-off

force for a Gaussian
distribution of identical
asperities, from Fuller and
Tabor (1975)
tension or asperities in compression dominate the force integral and it is independent

of the actual applied force.
Equation (16.134) is clearly unrealistic when λ > 0.48, since it predicts that
otherwise unloaded bodies will be attracted to each other until all the asperities are
in contact and the separation h 0 = 0. Fuller and Tabor (1975) resolved this paradox
by using the same approach, but with the Gaussian distribution of asperity heights
(16.31). They obtained a non-zero pull-off force
3π RΔγ
F = N0 f 0 g(λ) where f0 = (16.136)
2
is the pull-off force for a single asperity, and g(λ) is shown as a function of 1/λ in
Fig. 16.19. Here we define λ through Eq. (16.135), but using the standard deviation
of the Gaussian asperity distribution for σ. With this notation, Fuller and Tabor’s
parameter Δc = 3π 2/3 λ/4. Notice that although the parameter σ of the exponential
distribution is also a measure of the amplitude of the surface roughness, it is not
directly comparable with the RMS measure in the Gaussian distribution.
16.7.2 The Sinusoidal Profile
More insight into the effect of surface roughness on adhesion is obtained by consid-
ering the contact of a two-dimensional sinusoidal surface as in Fig. 6.6. If no external
forces are applied to the bodies, the full contact state will be energetically favourable if
the elastic strain energy per unit length U is less than Δγ (Persson and Tosatti 2001).
We know from Eqs. (6.65), (6.58) that a pressure distribution p(x) = p0 cos(ωx)
produces a normal surface displacement u(x) = u 0 cos(ωx), where
2 p0
u0 = . (16.137)
E ∗ω
The strain energy per unit surface area in this deformed state is equal to the work
done during loading and hence
∗
1 L
p0 u 0 L
p0 u 0 E ωu 20
U= p(x)u(x)d x = cos2 (ωx)d x = = . (16.138)
2L 0 2L 0 4 8
For the geometry of Fig. 6.6, u 0 = h 0 and ω = 2π/L, so the condition Δγ >U gives
LΔγ π
∗ 2 > . (16.139)
E h0 4
The JKR Solution

The corresponding partial contact problem can be solved under the JKR approxi-
mation by superposing Westergaard’s solution from Sect. 6.5.6 and the solution for
a periodic array of cracks (Koiter 1959), loaded so as to satisfy the energy release
rate criterion (12.19). The solution was given by Johnson (1995), who found that the
contact half-width a is given by

p̄ 2 πa πa
∗
= sin − α tan , (16.140)
p L L
where ∗
2LΔγ π E h0
α= and p∗ = (16.141)
π 2 h 20 E ∗ L
is the mean pressure needed to ensure full contact in the absence of adhesion.
Figure 16.20 shows the relation between the contact semi-width and the mean
pressure p̄ for four different values of α, including the solution without adhesion
p̄
p∗
Fig. 16.20 Contact semi-width a as a function of mean pressure p̄, from Eq. (16.140)
[α = 0]. If we imagine a horizontal line representing a given value of p̄ in this

figure, intersections where the curve crosses the line with positive slope are always
stable, and those with negative slope are unstable. Thus, for example, for α = 0.4,
points between A and C are stable. If the bodies are brought together, but no external
pressure is applied [ p̄ = 0], the bodies will jump into contact with a contact area
corresponding to the point B. If a positive pressure is now applied, the contact area
will increase in a stable fashion until C is reached, at which point there will be an
unstable jump into complete contact.
If the bodies are in partial contact at point B and a tensile traction is now applied,
the contact area will shrink in a stable fashion as far as A, at which point there will be
a jump out of contact. However, if full contact is once established, no finite traction
is strictly sufficient to cause pull off, according to the JKR theory. Johnson argued
that inevitable defects would act like small cracks to initiate pull off, but the pull-off
traction then depends on the size of these flaws. Alternatively, since the JKR theory
is an approximation to the Lennard-Jones force law, separation must be initiated
as soon as the maximum tensile traction at the troughs of the sinusoid reaches the
maximum traction σ0 . This question is further discussed in Sect. 12.3.2.
Figure 16.20 also shows the limiting case α = 33/4 /4 ≈ 0.57, where the curve is
tangential to the line p̄/ p ∗ = 0. For α > 0.57, the only intersection with this line is
the origin, which is unstable, so if the bodies are brought together with no applied
force, they will spontaneously jump into full contact. The condition α > 0.57 implies
LΔγ
> 1.60, (16.142)
E ∗ h 20
which is significantly more restrictive than (16.139). The reason for this apparent
contradiction is that points like C in Fig. 16.20 impose a potential barrier between
the initial unloaded state and the energetically favourable full contact state. This also
implies that the instantaneous state of the system [in particular, the contact area]
depends on the loading history.
The Rigid-Body Solution
The JKR solution clearly becomes inappropriate if the amplitude h 0 of the sinusoid
is comparable with the length scale ε of the Lennard-Jones force law, since attractive
tractions in the separation regions will then have a significant effect. Wu (2012) used
a numerical method to solve the sinusoidal elastic contact problem with the exact
force law and his results show that the JKR solution is significantly in error when
the modified Tabor parameter

L
μ = σ0 ∗ < 1. (16.143)
E Δγ
In the limit μ 1, a good approximation can be obtained by neglecting elastic

deformation, as in the Bradley solution for the sphere. If the rigid plane is located at
a height d above the peaks of the sinusoid, the gap is given by
g(x) = d + h 0 [1 − cos(ωx)] (16.144)
and the mean traction σ̄ is then

L
1
σ̄ = σ(x)d x, (16.145)
L 0
where σ(x) is obtained by substituting (16.144) into the force law (12.14).
For relatively large values of h 0 /ε, only the tractions near the peaks of the sinusoid
contribute to the mean traction, which is then well approximated by the ‘Hertzian’
value
9σ0 3ε 1 2145
σ̄ ≈ − , (16.146)
16 2h 0 d̃ 5/2 4096d̃ 17/2
where d̃ = d/ε. The pull-off traction corresponds to the maximum value of σ̄, which
occurs when d̃ = 1.1009. We obtain

ε
σ̄max = 0.3824 , (16.147)
h0
which is shown by the dashed line in Fig. 16.21 as a function of h 0 /ε. For smaller
values of h 0 /ε, the maximum σ̄ occurs at larger values of d̃ in the range 1.1 < d̃ < 1.2
and the corresponding pull-off traction is shown by the solid line in Fig. 16.21.
16.7.3 Adhesion of Random Rough Surfaces
If the surface roughness is defined by a PSD spanning several decades, the contact
problem is extremely challenging, since different components of the spectrum lie in
different ranges of Tabor number and hence affect the contact problem in different
ways. Persson and Scaraggi (2014) have proposed a method based on the DMT
approximation of Sect. 12.3. Tensile tractions are first ignored, so that the contact
area is determined using Persson’s theory [Sect. 16.6.4]. The probability distribution
Φ(g) for the gap in regions of separation is then estimated and the nominal pressure
is modified by superposing adhesive tractions in the separation regions determined
by convolution of Φ(g) with the force law σ(g).
An alternative approach due to Pastewka and Robbins (2014) argues that the adhe-
sive tractions will be confined to a thin strip of the nominal contact area surrounding
areas of repulsive [hard] contact, in which the adhesive tractions can be estimated
using an asymptotic approach analogous to those in Chap. 10. The tensile contribu-
tion to the total force is therefore predicted to be proportional to the length of the
perimeter of these contact regions, which the authors argue is a fractal with the same
fractal dimension as the contact area itself.
σ̄max
σ0
Fig. 16.21 Pull-off traction for a rigid sinusoidal surface. The dashed line defines Eq. (16.147) in
which the peaks of the waves are approximated by parabolæ
16.8 Incremental Stiffness and Contact Resistance
If two contacting bodies have rough surfaces, there will be less than complete contact
at the interface and we anticipate that this will impose a resistance to electrical
conduction between the contacting bodies. This phenomenon is clearly of significant
technological importance, and it features prominently in early theories of rough
surface contact, notably in the work of Holm (1958) who established the resistance
associated with a single circular contact between two large bodies, and Greenwood
(1966), who investigated the effect of a cluster of such contacts.
In Chap. 4, Sect. 4.1, we showed that for the contact of two half-spaces with fairly
general profiles, the electrical contact resistance and the incremental stiffness are
related by the equation
I 2 dP
= ∗ , (16.148)
V1 − V2 (ρ1 + ρ2 )E dΔ
where I is the total current flow between the bodies, V1 , V2 are the electrical potentials
of the two bodies distant from the contact area, ρ1 , ρ2 are the resistivities of the
materials and d P/dΔ is the incremental stiffness of the contact — i.e. the variation
in normal force P associated with a small elastic relative normal displacement Δ. The
incremental stiffness is also of interest in its own right, since it governs the dynamic
response of assemblages of contacting bodies. Also, it has been estimated that in
such bodies a significant proportion of the overall elastic compliance is associated
with the compliance of the contact interfaces (Back et al. 1973).
Equation (16.148) can be generalized to the contact problem for two finite rough
elastic bodies, provided the longest wavelength in the roughness is significantly
smaller than any length scale associated with the corresponding smooth contact
problem (Barber 2013b). The effect of roughness can then be characterized as a
resistive layer interposed between the bodies in the contact area, whose resistance is
a function of the local nominal [macroscopic] contact pressure p̄.
The properties of this fictitious layer can be explored in the context of a problem in
which two macroscopically plane rough half spaces are pressed together by a uniform
nominal pressure p̄, where both the potential difference (V1 − V2 ) and the relative
approach Δ are to be interpreted as the difference in these quantities from the values
in the corresponding smooth contact case where contact would be complete and the
pressure uniform. In other words, we need to subtract out the electrical resistance
and elastic compliance of the corresponding smooth bodies to leave the effect due to
roughness. In this case, Eq. (16.148) must be modified by replacing the total normal
force P and the total current I by nominal pressure p̄ and mean current density ī
respectively. In the special case where one body is a perfect conductor, we obtain
ρī 2 d p̄ 1
= ∗ ≡ , (16.149)
V1 − V2 E dΔ L
where we define L as the thickness of an extra layer of material of resistivity ρ that

would interpose the same resistance per unit area as the rough surface contact.
16.8.1 Asperity Model Predictions
In asperity model theories, the individual contact areas are assumed to act indepen-
dently. If these are Hertzian, the contact area at each asperity is exactly one half of the
corresponding contribution to the bearing area of Sect. 16.2.1 and hence summing
overall such contacts, we obtain

A B(h) 1 h
= = erfc √ , (16.150)
Anom 2 4 σ 2
where we have assumed a Gaussian height distribution with standard deviation σ

and measure h from the surface mean plane giving h 0 = 0.
Using Eq. (16.78) from the asperity model theory of Bush et al. (1975) to subsitute
for A/Anom , we then obtain
∗
E σm h
p̄ = √ erfc √ , (16.151)
4 π σ 2

incremental stiffness d p̄/dΔ
and nominal pressure p̄
using the asperity model
theory of Bush et al. (1975).
The dashed line is the
σ dp̄
∗
prediction of Persson’s E σm dΔ
theory [see Sect. 16.8.4
below], with γ = 0.48
p̄
E ∗ σm
and hence ∗
d p̄ d p̄ E σm h2
=− = √ exp − 2 . (16.152)
dΔ dh 2 2πσ 2σ
This relation between d p̄/dΔ and p̄ is plotted parametrically with h as parameter in

Fig. 16.22. A good curve fit in the range plotted is given by the power-law expression
0.81
σ d p̄ p̄
∗ = ∗ . (16.153)
E σm dΔ E σm
16.8.2 Clustering of Actual Contacts
Although it is not visible at the scale of Fig. 16.22, the curve actually approaches
∗
infinite slope as p̄/E σm → 0, and this is of concern since the mean slope σm
generally increases without limit as the upper limit ωh of the PSD is increased.
Asperity model theories overestimate the contact area for a given value of separation
h, since they neglect the displacements at any given asperity due to the forces acting
on neighbouring asperities, and this is particularly significant for fractal surfaces for
which long wavelength terms in the PSD contribute to ‘clustering’ of actual contacts.
Greenwood (1966) has shown that the resistance of a cluster of contact areas is
increased approximately by the resistance associated with perfect contact over the
cluster or ‘contour’ area. For example, if two half spaces each of resistivity ρ make
contact at a group of circular contact areas of radius ai clustered within a larger circle
of radius b, the resistance is approximately
ρ ρ
R= + . (16.154)
2b 2 ai
We shall show in the next section that the incremental stiffness is bounded even in
the fractal limit [i.e. with no truncation of the PSD], and indeed is largely determined
by the long wavelength components.
16.8.3 Bounds on Incremental Stiffness
Figure 16.23 shows a smooth rigid body indenting an elastic body with a rough
surface. The two dashed lines 1 and 2 represent the plane surfaces of two half spaces
such that in the configuration shown g1 (x, y) < g(x, y) < g2 (x, y) for all x, y, where
g(x, y) is the initial gap function for the rough surface. The distance between these
planes is denoted by s.
From Theorem 6 of Chap. 4, we know that the force P(Δ) required to press the
rigid body to a position defined by a rigid-body displacement Δ must satisfy the
inequality
P1 (Δ) > P(Δ) > P2 (Δ), (16.155)
where P1 (Δ), P2 (Δ) are the corresponding forces for the plane surfaces 1 and 2.
Furthermore, the functions P1 (Δ), P2 (Δ) must be concave upwards, from Theorem 5
of Chap. 4, and they must have the same shape, since the surfaces 1 and 2 are both
planes separated by a distance s, so P2 (Δ) = P1 (Δ−s). A representative plot of these
functions is shown in Fig. 16.24a.
The force–displacement relation P(Δ) for the rough surface must also be concave
upwards, and the inequality (16.155) shows that it must lie between the two curves
P1 (Δ), P2 (Δ), as suggested by the dashed line in Fig. 16.24a.
These conditions enable us to place strict bounds on the incremental stiffness
d P/dΔ. For a given force P in Fig. 16.24b, we know that Δ must lie between the
points A and B, and the slope d P/dΔ cannot be less than that of the line BC, tangent
to P1 (Δ), since the curve for lower values of P could then only stay between the
1
s
2
Fig. 16.23 Indentation of a rough body by a smooth body, and two bounding problems with plane
surfaces
(a) (b)
P1 P D
P P2
P1
s P2
A
B
C
Δ Δ
Fig. 16.24 (a) The force–displacement relation for the rough surface is bounded by the two smooth
surface curves P1 (Δ), P2 (Δ). (b) Graphical construction for determining the bounds on incremental
stiffness
two limits if it were convex upwards, which violates Theorem 5. A similar argument
for higher values of P shows that the slope of the tangent line B D defines an upper
bound to d P/dΔ.
Clearly the ‘tightness’ of the bounds — i.e. their effectiveness in bracketing the
incremental stiffness in a limited range — will improve if s is reduced. This will
be the case if the RMS roughness σ is reduced, but so far as the above proof is
concerned, the distinction between roughness and the underlying profile is essentially
arbitrary. Suppose we were to develop a numerical solution of the contact problem
including just the long wavelength [low ω] terms in the PSD. Using this solution for
the functions P1 (Δ), P2 (Δ), we can then place bounds on the incremental stiffness
including the entire PSD, with s now representing the peak-to-valley roughness of
only those terms in the PSD that were excluded from the numerical solution, as shown
in Fig. 16.25. An important consequence of these arguments is that the incremental
stiffness, and hence also the electrical contact resistance, is determined principally
by the long wavelength components in the PSD, so it is not necessary to include
a highly refined roughness description in a numerical model. This also lends some
support to Majumdar and Bhushan’s replacement of ‘rough’ asperities by smooth
ones [see Sect. 16.6.1 and Fig. 16.15], at least in estimates of incremental stiffness.
Greenwood and Wu (2001) argue that the whole concept of an asperity therefore
needs rethinking, relative to the statistical derivations of Sect. 16.4, and they favour
Fig. 16.25 Inclusion of the long wavelength roughness in the bounding solutions
the definition of Aramaki et al. (1993) of an asperity as ‘...one which makes a contact
spot’.
16.8.4 Persson’s Theory of Incremental Stiffness
We have seen that Persson’s theory predicts linearity between total contact area and
nominal pressure when the surface is sufficiently rough or the nominal pressure
is sufficiently low [Eq. (16.124)]. However, the theory does not contain interfacial
separation h [or equivalently, rigid-body approach Δ] explicitly, so the incremental
stiffness cannot be immediately calculated. Persson (2007) argues that the reason for
the linearity in Eq. (16.124) is similar to that explaining linearity in the Greenwood
and Williamson theory — viz. that increasing the normal force increases the num-
ber of contact regions, but does not affect their size distribution. Starting from this
assumption alone, it then follows that the incremental stiffness must be proportional
to nominal pressure (Pastewka et al. 2013) — i.e.

d p̄ p̄ Δ
= and hence p̄ ∼ exp , (16.156)
dΔ Δ0 Δ0
where Δ0 is a constant with dimensions of length.

If an increment δ p̄ in the nominal pressure leads to an additional compliance δΔ,
the work done per unit area is p̄ δΔ and this must lead to an increase in the strain
energy U in the elastic bodies. We therefore have

∂U dU d p̄ d p̄ dU
p̄ = = or = U ( p̄) . (16.157)
∂Δ d p̄ dΔ dΔ d p̄
The strain energy per unit surface area associated with a single sine wave u(x) =
u 0 cos(ωx) under full contact conditions is given by
∗
E ωu 20
U (ω) = , (16.158)
8
from Eq. (16.138). To estimate the total strain energy U with partial contact, Pers-
son reduces this expression in the ratio A(V (ω))/Anom , where V (ω) is defined by
Eq. (16.113) and A(V )/Anom by (16.122), and then sums the result over all the waves
in the PSD. The total energy so calculated is then reduced by an additional empirical
factor. For most surfaces, this leads to a relation
Δ0 ≈ γσ, (16.159)
where σ is the RMS roughness height. Akarapu et al. (2011) obtained numerical
estimates for incremental stiffness and fitted a curve of the form (16.156), (16.159)
with γ = 0.48. This value was used to plot the linear relation (16.156) as the dashed
line in Fig. 16.22 [see Sect. 16.8.1 above]. Notice that though the trend predicted by
Persson’s theory and the asperity model theory are different, the numerical values
obtained are really quite close, at least in the range plotted.
16.8.5 Gaps and Fluid Leakage
A related problem of technological importance concerns the leakage of fluids through

seals, where inevitable surface roughness implies the existence of a network of pas-
sages [cavities] between the contacting bodies (Bottiglione et al. 2009; Paggi and
He 2015). At sufficiently light loads, the contact morphology will comprise a set of
‘islands’ of contact around which fluid can flow, but at larger pressures the islands
may coalesce and eventually form an obstruction to flow from one side of the contact
interface to the other (Hunt 2005; Dapp et al. 2012). The question of determining
under what conditions a superposed set of random objects [here cavities] ‘perco-
late’ [i.e. form a continuous path through a space] is of importance in many fields,
including for example electrical conduction through a random fibrous network.
The average value of the gap between two contacting bodies [or equivalently, the
total volume of these cavities per unit nominal area] can be related to the incremental
stiffness using Betti’s reciprocal theorem. Suppose that the undeformed rough surface
is defined by a height function h(x, y) measured from the mean plane, so that

h(x, y)d xd y = 0, (16.160)
Anom
where Anom is the nominal contact area.

If a rigid plane surface is brought down to a level corresponding to h = h 0 , the
gap g(x, y) will be
g(x, y) = h 0 − h(x, y) + u z (x, y), (16.161)
where u z (x, y) is the elastic displacement. We now apply Betti’s theorem [Eq. (4.45)]
to this elastic state, using as auxiliary solution t 2 , u2 the state where a uniform
pressure p0 produces a uniform displacement u z = u 0 = C p0 , where C is a constant.
We obtain

p0 u z (x, y)d xd y = p(x, y)u 0 d xd y = p̄u 0 Anom , (16.162)
Anom Anom
where p̄ is the mean pressure over Anom . Substituting for u z (x, y) from (16.161) and
using (16.160), we then obtain

1
Δ = h 0 − C p̄ = −ḡ where ḡ ≡ g(x, y)d xd y (16.163)
Anom Anom
is the mean gap.7 In other words, the rigid-body indentation is less than that for
an equivalent smooth surface by a distance equal to the mean gap. The quantity Δ
defined in Eq. (16.163) is the rigid-body compliance due to roughness alone and
hence is the same quantity as appears in Eqs. (16.149), (16.153), (16.156). It follows
that the mean gap is related to the incremental stiffness through the equation

d ḡ d p̄
= −1 . (16.164)
d p̄ dΔ
If Persson’s equation (16.156) is used for incremental stiffness, this implies a log-
arithmic decay of ḡ with increasing pressure (Persson 2007), but if a power law is
assumed, as predicted by Pohrt and Popov (2012) and also by the approximation
(16.153), the dependence will be of the form
ḡ = ḡ0 − Bp c , (16.165)
where g0 is the mean gap when the bodies first make contact and B, c are constants,
with the power c generally in the range 0 < c < 0.5.
The fluid flow through a narrow gap increases with the cube of the gap thickness
for a given fluid pressure gradient, so to make further progress we need to know the
probability distribution Φ(g) for g as well as its mean value. Almqvist et al. (2011)
propose a method in which Persson’s theory is used to estimate the area of actual
contact [and hence also the complementary separation area] and the mean gap as
functions of the truncation limit ωh . Once a point transitions into separation, it is
assumed to experience no further deformation, so the gap probability evolves by the
addition of the change in the mean gap and the variance of the roughness added since
the transition.
16.9 Finite-Size Effects
The theoretical studies so far described all relate to the contact of nominally flat
infinite bodies, but numerical studies necessarily involve finite bodies, unless periodic
boundary conditions are imposed, in which case the surface roughness implies a
prescribed maximum wavelength equal to the size of the modelled region. Pohrt
and Popov (2012) argue that Persson’s prediction of linearity between incremental
stiffness and normal force is a consequence of his assumption that the maximum
roughness wavelength is much smaller than the linear dimensions of the nominal
contact area, described by some authors as the ‘thermodynamic limit’. By contrast,
7 Notice that ḡ is here the average over the entire nominal contact area Anom , in part of which there
is contact so the gap is zero. For fluid leakage calculations, a more useful parameter might be the
mean gap in the separation regions which is ḡ/(1− A/Anom ). However, for quasi-fractal surfaces
the proportion of actual contact A/Anom is generally small, so this correction may be relatively
minor.
Pohrt and Popov report the results of a boundary element study in which wavelengths
were included up to the size of the indenting body, which in this case was a flat-ended
rigid punch of square planform, and they found a power-law dependence significantly
below linearity.
This disagreement brings into focus the broader issue of the effect of the finite
dimensions of real contacting bodies on the rough surface contact problem. It is
clearly possible and indeed likely that the roughness PSD will have some content in
the range associated with the macroscopic geometry, but to include this explicitly in
the solution of the contact problem requires a multiscale model beyond the capacity
of most numerical algorithms. More seriously, since the roughness is stochastic,
the existence of a non-trivial content at macroscopic length scales implies that the
macroscopic problem, including the nominal shape of the contacting bodies, is itself
stochastic.
In practice, most authors assume that the roughness scale is sufficiently small to
be decoupled from the macroscopic geometry, in which case the roughness can be
replaced by an equivalent uniform layer of elastic material which is assumed to act
as a nonlinear ‘Winkler’ layer8 — i.e. the layer compliance depends only on the local
value of the nominal pressure. The force–compliance relation for this layer is then
defined by the study of an appropriate rough surface problem in the thermodynamic
limit.
16.9.1 Integral Equation Formulation
The ‘roughness compliance’ is additive to the elastic deformation of the half space,
so the integral equation formulation (2.9) is modified to

1 p(ξ, η)dξdη
f { p(x, y)} + = Δ − g0 (x, y) (x, y) ∈ A
πE∗ A (x − ξ)2 + (y − η)2
(16.166)
where f ( p) is the compliance of the roughness layer due to a nominal pressure p.
For example, with Persson’s pressure–displacement relation (16.156)2 , we would
use
p
f ( p) = Δ0 ln , (16.167)
p0
where p0 is a reference pressure which essentially defines a reference plane in the

roughness profile.
8 See Sects. 14.3.1 and 16.8.

Fig. 16.26 Voronoi partition

of the nominal contact area
into regions Āi surrounding
actual contact areas Ai
A more rigorous argument for treating the roughness as a Winkler foundation can
be obtained from asperity model arguments. The normal surface displacement at any
point in the nominal contact area can be defined by Eq. (2.17) — i.e.

1 p(ξ, η)dξdη
u z (x, y) = , (16.168)
πE∗ A (x − ξ)2 + (y − η)2

where A = Ai , i = (1, N ) here denotes the union of all the actual contact areas
Ai .
The contribution to this integral from a particular actual contact area Ai will be
almost the same as that of a point force of equal magnitude applied at the centre of
Ai , except for points in or near to Ai . Thus, if the individual actual contacts between
asperities are relatively widely separated [sparse], the normal surface displacement
at a point in the contact area A j can be approximated as
⎡ ⎤

1 ⎣ Pi p(ξ, η)dξdη ⎦
u z (x, y) ≈ +
π E ∗ i= j (x −xi )2 + (y − yi )2 Aj (x −ξ)2 + (y −η)2
(16.169)
where
Pi = p(ξ, η)dξdη (16.170)
Ai
is the total force transmitted through the contact Ai and the summation in (16.169)
takes place over all i except i = j.
In developing Eq. (16.169), we approximated the displacement due to the distri-
bution in Ai by that due to an equal concentrated force. An alternative approximation
would replace this displacement by that due to an equal force uniformly distributed
over a region surrounding the region Ai . For this purpose, we need to partition the
nominal contact areas Anom into regions associated with the separate actual contact
areas. A convenient way to do this is to associate any given point with the nearest
actual contact area. This leads to a Voronoi polygonal partition of the nominal contact
area, as shown in Fig. 16.26.
We shall denote the region9 surrounding Ai as Āi and define the mean pressure
in this region as
Pi
p̄i = (x, y) ∈ Āi . (16.171)
Āi
Equation (16.169) can then be written

1 p̄i dξdη
u z (x, y) ≈
πE∗ Anom −Ā j (x − ξ)2 + (y − η)2

1 p(ξ, η)dξdη
+ (x, y) ∈ A j , (16.172)
πE∗ Aj (x − ξ)2 + (y − η)2
where the first integral term simply replaces the summation in (16.169). The pressure
p̄i is a piecewise constant function of ξ, η, but it can be seen as a smoothed out version
of the actual pressure distribution, or as the expected value of the pressure over the
randomness imposed by the surface roughness.
A simpler result is obtained if we add the term

1 p̄ j dξdη

πE∗ Ā j (x − ξ)2 + (y − η)2
Fig. 16.27 Constriction of

heat flow through actual
contact areas, illustrating the
concept of a unit cell
unit cell
9 Notice that Āi is here interpreted as the polygon in which Ai is situated, including Ai itself.
into the first term in (16.172) and subtract it from the second term, giving

1 p̄(ξ, η)dξdη
u z (x, y) ≈
πE∗ Anom (x − ξ)2 + (y − η)2

1 p(ξ, η) − p̄ j dξdη
+ ∗ (x, y) ∈ A j . (16.173)
πE Ā j (x − ξ)2 + (y − η)2
The first term in this expression is then simply the displacement due to the smoothed
out pressure distribution over the entire nominal contact area and hence corresponds
to the integral term in (16.166). The second term describes the local compliance in
the actual contact area A j due to a self-equilibrated traction distribution. If a set of
such distributions were concatenated for all i using the same value for p̄i , we would
reproduce the conditions in the contact of flat rough half-spaces, so this term defines
the function f ( p) in Eq. (16.166). At low normal forces, the mean pressure p̄ j will
make a negligible contribution to the integrand in this term and the compliance will
be simply the sum of that associated with the mean pressure on a smooth surface and
the local asperity compliance for an isolated asperity.
16.9.2 Unit Cells and the Constriction Alleviation Factor
Studies of thermal10 and electrical contact resistance frequently invoke the concept
of a ‘unit cell’ surrounding each actual contact area, as shown in Fig. 16.27. Far
from the contact interface, the flow lines are parallel, so a partition can be defined
analogous to that in Fig. 16.26, such that flux through contact area i remains for all
time within cell i (Cooper et al. 1969). The complete conduction problem can then
be analyzed as a set of such resistances in parallel.
Many authors approximate the unit cell as a cylinder of radius b whose flat end
contains a concentric contact area of radius a. The constriction resistance R for a
single cell is then defined as the additional resistance associated with the deviation
of the flow lines to pass through the central circle. If a b, R can be approximated
by the value for a half space (i.e. b → ∞), which is11
1
R ≈ R0 = , (16.174)
4K a
where K is the thermal conductivity, but this is clearly in error when a = b since then
the entire end of the cylinder makes contact and there is no constriction resistance.
Hunter and Williams (1969) define the ‘constriction alleviation factor’ as
10 We define thermal resistance, by analogy with electrical resistance, as ΔT /Q, where Q is the
steady-state heat flux due to a temperature difference ΔT .
11 Notice that this is the resistance on one side of the interface. For two dissimilar materials, the
total constriction resistance through the contact would be 1/4K 1 a +1/4K 2 a.

a R
f = = 4K a R (16.175)
b R0
and give an analytical solution which ultimately requires numerical evaluation. How-
ever, a good approximation can be obtained in the spirit of Sect. 16.9.1 by superposing
two half-space solutions corresponding to heat input over a circle of radius a and an
equal heat output over a concentric circle of radius b. It is clear that very little heat
flow will penetrate into the region outside an imaginary cylinder of radius b and the
solution obtained is exact in the limits a/b → 0 and a/b → 1. We obtain
1 1 a a
R= − and f =1− . (16.176)
4K a 4K b b b
This expression is compared with Hunter and Williams’ result in Fig. 16.28. A
good curve fit to the more exact result is given by
f (x) = 1 − 1.41x − 0.183x 2 + 0.59x 3 . (16.177)
16.9.3 Contact of Rough Spheres
Greenwood and Tripp (1967) used the formulation (16.166) to solve the problem of
a rigid sphere of radius R indenting an elastic half space with RMS roughness σ.
The compliance function f ( p) was determined from the Greenwood and Williamson
asperity model theory of Sect. 16.3.2. They showed that the extent of the deviation of
the regularized pressure p from the smooth Hertzian pressure distribution depends
on the parameter
σR σ
α= 2 = , (16.178)
a Δ
Fig. 16.28 The constriction

alleviation factor f (a/b)
from Hunter and Williams
(1969) [solid line]. The
dashed line is the
where a is the radius of the contact area in the smooth [Hertzian] problem and Δ is
the corresponding Hertzian rigid-body indentation.
Some typical results are shown in Fig. 16.29. For α 1, the nominal pressure
[i.e. the contact pressure in the nonlinear layer problem] follows the Hertzian distri-
bution closely except at the edge of the contact area, where the square-root bounded
behaviour is ‘rounded off’, as shown in Fig. 16.29a. However, as α approaches unity
and beyond, there is a major deviation from the Hertz theory, with a larger nominal
contact radius, a smaller maximum nominal pressure and a much smoother transi-
tion to separation [Fig. 16.29b]. Yashima et al. (2015) performed experiments on the
contact of spherical bodies with prescribed distributions of spherical asperities and
obtained results very close to those predicted by Greenwood and Tripp.
A Rough Flat Punch
Since the Hertzian indentation Δ increases with the normal force P, Eq. (16.178)
implies that surface roughness effects are most significant at low values of P and the
nominal pressure approaches more closely to the Hertzian as P is increased.
This result is more starkly demonstrated for the indentation of an elastic half space
by a nominally flat rough punch. If we use (16.167) for the roughness compliance,
the governing Eq. (16.166) can be written in the dimensionless form
˜ η̃)d ξd
˜ η̃
1 p̃(ξ,
ln( p̃) + = Δ̃ (x̃, ỹ) ∈ A, (16.179)
π A ˜ 2 + ( ỹ − η̃)2
(x̃ − ξ)
where
ap x ˜ ξ y η
p̃ = ∗ x̃ = ξ= ỹ = η̃ = , (16.180)
E Δ0 a a a a
(a) (b)
Hertz
Hertz
p(r) p(r)
rough
sphere
rough sphere
0 r a0 0 a0 r
Fig. 16.29 Contact pressure distribution for indentation by a rough sphere, (a) α 1, (b) α ≈ 1
[after Greenwood and Tripp 1967]
a is a representative linear dimension of the punch planform and

∗
Δ E Δ0
Δ̃ = + ln (16.181)
Δ0 ap0
is a dimensionless rigid-body indentation. The corresponding dimensionless applied

force is
P̃ = p̃(ξ, ˜ η̃ = ∗ P .
˜ η̃)d ξd (16.182)
A E Δ0 a
Equation (16.179) is independent of the roughness parameters except insofar as these

enter through Δ̃ which in turn is a monotonically increasing function of P̃. It follows
that increasing the normal force P and reducing the roughness height scale Δ0 have
similar effects on the dimensionless pressure distribution.
Some representative results12 are shown in Fig. 16.30 for the case of a circu-
lar punch of radius a. The contact pressures are normalized by the mean value
p̄ = P/πa 2 . At small values of the dimensionless force P̃, the contact problem is
dominated by the deformation of the roughness layer and the nominal pressure is
approximately uniform. However, as the force is increased, the incremental stiffness
of the layer increases, particularly at the edges r = a, and the pressure distribution
evolves towards the smooth punch solution in the limit P̃ → ∞.
p
p̄
Fig. 16.30 Normalized contact pressure distribution for indentation by a flat rough punch of radius
a. The dotted line represents the ‘smooth’ flat punch solution of Eq. (5.21)
12 These results are reproduced with the courtesy of Dr. J.A.Greenwood.

Problems
1. Suppose the asperity in Fig. 16.1 is replaced by a paraboloid whose radius at the
apex is R. Using Bowden and Tabor’s assumptions, find the corresponding relation
between indentation depth h, normal force P and volume of material removed V ,
and comment on the implications for the friction law at a single asperity.
2. A cylindrical column of diameter 50 mm and length 100 mm carries a compressive
force of 5000 N. It is desired to measure small fluctuations in this force due to
vibration by incorporating a load cell as shown in Fig. 16.31.
All the contacting surfaces are ground to a surface finish of 0.5μm RMS, bearing
area results for which are given in Table 16.1. You can assume that all the parts have
Young’s modulus E = 200 GPa, Poisson’s ratio ν = 0.3 and indentation hardness
[flow pressure] H = 800 MPa.
Fit the bearing area data to a normal distribution and use Eq. (16.1) to estimate
the incremental stiffness of the column with this modification and compare it with
the stiffness of a column of equal size without interfaces. The incremental stiffness
is defined as
dP
k= ,
dΔ
Fig. 16.31 All dimensions 35 20 35

are in mm 5 5
5000 Ν φ50 load 5000 Ν

cell
spacing disks
all dimensions are in mm
Table 16.1 Bearing area data

Height above 0.80 0.85 0.90 0.95 1.00 1.05 1.10 1.15 1.20
datum, (µm)
Bearing area, 1.18 0.81 0.55 0.36 0.24 0.15 0.10 0.06 0.04
(%)
Problems 391
where P is the applied compressive force and Δ is the reduction in length of the
column. Remember that there are several interfaces and each interface involves two
rough surfaces.13
3. At large values of x, the complementary error function can be approximated using
the asymptotic series

e−x
2
1 1 1.3 1.3.5
erfc(x) ≈ √ − 3 + 2 5 − 3 7 + ...
π x 2x 2 x 2 x
Carslaw and Jaeger (1959, Appendix II). Assuming that the normal force can
be approximated by the ‘bearing area’ Eq. (16.16), use the above expression and
Eq. (16.18) to develop an approximation for the ratio k/P, where P is the normal
force and k is the incremental stiffness of the contact. How much does this ratio
change in the range
pnom
0.001 < < 0.1.
H
4. Show that if a surface can be represented by an exponential distribution of identical

asperities, as in Sect. 16.3.1, the incremental stiffness is given by
dP P
=− ,
dh σ
where σ is defined in Eq. (16.24). Use this same equation to find an expression for the
bearing area for the surface comprised of an exponential distribution of asperities.
Then use the bearing area data from Problem 2 to estimate σ, and hence estimate the
incremental stiffness d P/dΔ for the problem of Fig. 16.31.
5. If two large bodies of the same material make contact at a single circular area of
radius a, the electrical resistance imposed by the interface is ρ/2a, where ρ is the
resistivity of the material.
Suppose instead that one of the surfaces is plane and the other comprises an expo-
nential distribution of asperities defined by Eq. (16.24). Find the contact resistance of
the interface if the bodies are pressed together to the point where undeformed regions
of the plane surface are at height h 0 . Assume that the asperities are all spherical with
radius R, so the Hertzian relations apply, including for example (16.29).
Show that the contact resistance is inversely proportional to the normal force P
and find the constant of proportionality.
6. Find the contact resistance as in Problem 5, but for a Gaussian distribution of
asperities defined by Eq. (16.31). Plot a suitably normalized graph of contact con-
13 Remember that the sum of two uncorrelated Gaussian signals constitutes a new Gaussian
signal
whose standard deviation is the root mean square of those of the separate signals — i.e. σ = σ12 +σ22 .
ductance [reciprocal of resistance] as a function of normal force and comment on

the degree of deviation from linearity.
7. A particular surface comprises a single sine wave
h(x, y) = h 0 cos(ω0 x + φ),
so that the peaks and troughs are aligned with the y-direction. Show that the profile
PSD sampled in the x-direction is
h 20 δ(k − ω0 )
PP (k) = .
4
A new surface is now constructed by superposition of sinusoidal surfaces of the same
wavenumber ω and amplitude, but with different angular orientations θ relative to
the x-axis and arbitrary phases φ. Find the profile PSD for this surface assuming
that all angles θ are equally represented [thus ensuring that the resulting surface is
isotropic] and that the height variance is σ 2 .
Then use this expression and Eq. (16.69)2 to determine the corresponding surface
PSD. Comment on your result.
8. Use results from Sects. 16.4.2 and 16.4.3 to express the two definitions of the
plasticity index (16.36) in terms of the moments of the profile PSD (16.57)–(16.61).
What properties of the surface would maximize the difference between the two
definitions? In such cases, comment on which definition might be most appropriate.
9. Use Eq. (16.69) to show that the even moments of the profile PSD can be expressed
in terms of the surface PSD as

2π(2n − 1)!! ∞ 2n+1
m 2n = ω PS (ω)dω.
(2n)!! 0
Use this result to evaluate the bandwidth parameter α for a surface for which
A
PS (ω) = ω0 < ω < λω0
ωm
= 0 ω < ω0 and ω > λω0 ,
where A, m, λ are constants and λ > 1. Hence show that the minimum value of α is
3/2 and comment on likely values for practical surfaces.
10. Use Eq. (16.55) to find the PSD for a profile defined by Archard’s exponential
autocorrelation function (16.48). Hence show that this profile is fractal at large ω
with a fractal dimension D = 1.5.
11. Suppose that a random surface is defined by the surface PSD
PS (ω) = Aω 2D−8 ω1 < ω < ω2 ,

Problems 393
where A is a constant. Outside this range PS (ω) = 0. Evaluate the moments

m 0 , m 2 , m 4 using the integral expression given in Problem 9, and hence determine
the bandwidth parameter α as a function of λ ≡ ω2 /ω1 and the fractal dimension D.
Plot a graph of α as a function of λ for D = 2.1 and 2.5.
12. If a rigid fractal surface is pressed against an elastic half space, we anticipate that
the resulting contact area will also be a fractal. Consider the special case where the
contact area at any scale comprises a number N of circular contact areas, all of the
same radius a. Find the dependence of N on a, assuming that the total contact area
has fractal dimension D A . Use this result to determine the total perimeter S of the
contact area, and hence show that the fractal dimension of the length S is the same
as that of the area A.
13. Find the autocorrelation function C(z) for the Weierstrass profile of Eq. (16.83).
Hence prove the result (16.87).
14. Use Persson’s theory [Eq. (16.122)] to estimate the limiting fractal dimension of
the total actual contact area A at a given nominal pressure, for a surface whose PSD
is PS (ω) = Cω −2−H , where H is the Hurst exponent.
15. The nominal contact pressure pnom can be expressed in terms of the probability
distribution Φ( p, V ) as ∞
pnom = Φ( p, V ) p dp.
0
Use Persson’s differential equation (16.118) and the boundary condition Φ(0, V ) = 0
to show that pnom is independent of V and hence that Persson’s theory defines the
evolution of Φ( p) as roughness is added at constant pnom , for any initial condition
Φ( p, 0).
16. From Sect. 12.2.4 and Fig. 12.3, we expect that a spherical asperity would jump
into contact at Δ = 0, but not pull out of contact until Δ = Δ A , where
β2Δ A 3π 2/3
Δ̂ A ≡ =− ,
R 4
from Eq. (12.39). This implies that a surface defined by the Gaussian distribution of
asperities of Eq. (16.31) will exhibit different force–displacement relations during
loading and unloading. Find expressions for these relations using the approximation
(16.132) and make appropriate normalized plots for the cases where the minimum
approach is h 0 = σ and h 0 = 2σ respectively. [Note: If you cannot devise a univer-
sal normalization for these plots, just use representative engineering values for the
remaining parameters.]
17. Use the Maugis-Dugdale force law of Sect. 12.4.3 to approximate the force–
displacement relation for the rigid sinusoidal surface of Sect. 16.7.2 contacting a
rigid plane. Compare your result with Fig. 16.21.
Suppose we now wish to use the same force law, but include the effects of elastic
deformation. Two cases can be distinguished, depending on whether regions near
the troughs of the sinusoid lose adhesion (i) before or (ii) after the contact region
shrinks to zero as the bodies are pulled apart. Use a suitable superposition with the
Westergaard solution to solve the problem for case (ii), when the nominal pressure
p̄ is sufficient to ensure that the contact region is non-null.
18. Suppose that an elastic half space is indented by a rigid body that contains two
spherical asperities of radius R and the same height, whose peaks are separated by
some distance d. Find the relation between total force P and indentation depth Δ,
assuming that the displacement at one asperity due to the force at the other can be
approximated by the point force solution, as in Sect. 4.1.2.
Hence show that the incremental stiffness d P/dΔ is an increasing function of
d. Can you extend this argument to show that the incremental stiffness is always
overestimated if interaction between asperities is neglected?
Chapter 17
Thermoelastic Contact
When two conformal bodies are placed in contact, the pressure distribution is sensi-
tive to comparatively small changes in surface profile. Thermoelastic deformations,
though generally small, can therefore have a major effect on systems involving con-
tact. Further interesting effects are introduced if the thermal boundary conditions
at the interface are influenced by the mechanical contact conditions. The thermal
and thermoelastic problems are then coupled through the boundary conditions and
the conventional uniqueness and existence theorems for heat conduction and linear
elasticity do not apply to this coupled problem. We shall see that this can lead to the
steady-state solution being non-unique and/or unstable.
Thermoelastic contact problems of this class are found in many applications, one
of the most important being sliding systems such as brakes, clutches and seals, where
thermoelastic effects are driven by frictional heat generation which depends on the
local pressure (Zagrodzki 1990; Lee and Barber 1993b). However, coupled problems
are also obtained for the static conduction of heat across an interface between two
thermoelastic bodies, since the temperature distribution depends on the extent of the
contact area which in turn is affected by thermoelastic distortion.
Even if there is full contact between the two bodies, there will generally be a
thermal contact resistance at the interface which varies with local contact pressure
and this can also be a source of thermoelastic contact instability. Conduction across
a solid/solid interface forms part of the heat flow path in many heat transfer appli-
cations, which can therefore exhibit erratic or non-uniform behaviour as a result of
such effects. For example, in the nominally one-dimensional solidification of a metal
against a plane mould, thermoelastic contact between the partially solidified casting
and the mould can become unstable, leading to significantly non-uniform pressure
distribution and alloy composition (Richmond and Huang 1977; Yigit and Barber
1994).

https://doi.org/10.1007/978-3-319-70939-0_17
396 17 Thermoelastic Contact
17.1 Thermoelastic Deformation
The thermoelastic deformation due to a prescribed temperature field T (x, y, z) is

conveniently defined in terms of a thermoelastic displacement potential φ, defined
such that
2Gu = ∇φ (17.1)
(Barber 2010, Chap. 21). The equilibrium equations are then satisfied provided φ
satisfies the equation
EαT
∇2φ = , (17.2)
(1 − ν)
in which case the stress components take the form
∂2φ ∂2φ
σx x = − ∇ 2
φ; σ x y = ; ... (17.3)
∂x 2 ∂x∂ y
etc. This solution will generally not satisfy the mechanical (e.g. traction or dis-
placement) boundary conditions in the problem, but these can then be satisfied by
superposing an appropriate isothermal elastic solution, for example, using one of the
potential function solutions given in Appendix A.
17.1.1 Fourier Transform Solutions
If the temperature field in the traction-free half-space z > 0 has the two-dimensional
form
T (x, z) = Θ(z) cos(mx), (17.4)
the surface displacement u z (x, 0) can be obtained as

∞
u z (x, 0) = −2α(1 + ν) cos(mx) e−ms Θ(s)ds (17.5)
0
(Barber and Hector 1999). Alternatively, if the half space is pressed against a friction-
less rigid plane with sufficient mean pressure p0 to ensure full contact, the contact
pressure distribution will be
∞
Eαm
p(x) = p0 + cos(mx) e−ms Θ(s)ds. (17.6)
(1 − ν) 0
More generally, if the thermoelastic layer −h < z < h is pressed between two
frictionless rigid planes, the resulting contact pressures are given by
17.1 Thermoelastic Deformation 397
h
Eαm
p(x, ±h) = p0 + cos(mx) cosh {m(s ± h)} Θ(s)ds.
(1 − ν) sinh(2mh) −h
(17.7)
These results can be used as in Sect. 14.4 as the basis of a Fourier transform solution
for half-spaces or layers with more general two-dimensional temperature fields.
17.1.2 Steady-State Temperature
In the special case where the temperature is independent of time, a solution particu-
larly appropriate to contact problems can be constructed in terms of a harmonic poten-
tial function ψ, such that the three stress components σzx , σzy , σzz are everywhere
identically zero, so for example, the surface of the half-space z = 0 is traction-free
(Barber 2010, Chap. 21). The temperature is related to ψ by
(1 − ν) ∂ 2 ψ
T =− (17.8)
G(1 + ν)α ∂z 2
and the remaining stress and displacement components are defined by
∂2ψ ∂2ψ ∂2ψ

σx x = −2(1 − ν) ; σ yy = −2(1 − ν) ; σ x y = 2(1 − ν) (17.9)
∂ y2 ∂x 2 ∂x∂ y
(1 − ν) ∂ψ (1 − ν) ∂ψ (1 − ν) ∂ψ
ux = ; uy = ; uz = − . (17.10)
G ∂x G ∂y G ∂z
All surfaces z = h are traction-free, so this solution also defines the steady-state
thermoelastic stresses and displacements in the layer 0 < z < h, with traction-free
surfaces.
The axial heat flux qz can then be obtained as
∂T (1 − ν)K ∂ 3 ψ
qz = −K = , (17.11)
∂z G(1 + ν)α ∂z 3
and since ψ is harmonic,

∂2ψ ∂2ψ ∂2ψ
=− 2 − (17.12)
∂z 2 ∂x ∂ y2
and we can write

(1 − ν)K ∂2 ∂2 ∂ψ
qz = − + . (17.13)
G(1 + ν)α ∂x 2 ∂ y2 ∂z
Using Eq. (17.10), we then have


∂2uz ∂2uz
+ = δqz , (17.14)
∂x 2 ∂ y2
where
α(1 + ν)
δ= . (17.15)
K
The material property δ is known as the thermal distortivity and it plays an important
rôle in thermoelastic contact problems (Dundurs 1974).
17.1.3 Thermoelastic Distortion Due to a Point Heat Source
Consider the traction-free half-space z > 0, with uniform heat flux qz (r ) = Q/πa 2
in 0 ≤r < a, the region r > a being unheated. Since the problem is axisymmetric,
Eq. (17.14) can be written

d 2uz 1 du z 1 d du z Qδ
+ = r = 0≤r <a (17.16)
dr 2 r dr r dr dr πa 2
=0 r > a. (17.17)
Integrating this expression and imposing continuity of slope1 and displacement at

r = 0 and r = a, we obtain
Qδ r 2
uz = +C 0≤r <a (17.18)
4πa 2
Qδ r Qδ
= ln + +C r > a, (17.19)
2π a 4π
where C represents an arbitrary rigid-body displacement. Notice that the logarithmic
term in (17.19) is unbounded as r → ∞, so it is not possible to assign C to give zero
displacement at infinity. We encountered a similar difficulty in Sect. 6.1 in relation
to the elastic displacements in two-dimensional problems. More generally, the rigid-
body displacement cannot be prescribed for steady-state thermoelastic problems
involving infinite bodies.
The surface displacements due to a steady-state point heat source Q can be
deduced by setting a = 0 in (17.19), giving
Qδ
uz = ln(r ) + B, (17.20)
2π
1 A discontinuity of slope would imply a locally singular value of curvature, which in view of (17.14)
could only occur if there were a concentrated ring of heat input at r = a and/or a concentrated heat
source at r = 0.
where B is another arbitrary constant. The temperature field in this case is spherically
symmetric and given by
Q
T (R) = , (17.21)
2πK R

where R = x 2 + y 2 +z 2 is the distance from the origin.
17.1.4 Dundurs’ Theorem
Equation (17.14) states that in the steady state, the sum of the curvatures of an initially
plane traction-free surface of a half space is proportional to the local heat flux across
it.
Dundurs (1974) proved that for two-dimensional problems, this result applies to
simply-connected bodies of arbitrary shape [i.e. not just for half spaces and layers],
provided there are no internal heat sources. Under these conditions, and assuming
plane strain conditions (u z = 0), the steady-state temperature is two-dimensionally
Fig. 17.1 Distortion due to

thermal expansion
harmonic and the thermal strains
ex x = e yy = α(1 + ν)T ; ex y = 0 (17.22)
are compatible, so there are no in-plane thermal stresses.2 Since the shear strain ex y
is zero, we have
∂u y ∂u x
=− , (17.23)
∂x ∂y
and hence
∂2u y ∂2u x ∂ex x ∂T
= − =− = −α(1 + ν) = δq y , (17.24)
∂x 2 ∂x∂ y ∂y ∂y
proving the desired relation between curvature and transverse heat flux.
If the heat flux across the contact boundaries is known, Dundurs’ theorem allows
us to determine the shape of the deformed surface due to thermal expansion alone.
Figure 17.1 shows some examples. Notice that straight edges that are not heated must
remain straight and corner angles are preserved, since there are no shear strains.
Heated initially straight edges become convex and cooled ones concave.
Once the shape of the body is determined, the contact problem is reduced to the
isothermal contact of bodies of this deformed shape. We shall apply this technique
to solve the axisymmetric thermoelastic Hertz contact problem in Sect. 17.2.
17.1.5 Moving Heat Sources
In steady-state sliding contact problems, the heat flux at the surfaces must move
relative to at least one of the bodies. If a concentrated heat source Q moves at
constant speed V in the positive x-direction over the surface of the half-space z > 0,
the temperature field in a frame of reference moving with the source is

Q V (R − x)
T (x, y, z) = exp − , (17.25)
2πK R 2k
where K , k are the thermal conductivity and diffusivity respectively, and R =

x 2 + y 2 +z 2 is the distance from the source, which is located at the origin of the
moving coordinate system.
In two-dimensional problems, if a source Q moves over the surface of the half-
plane z > 0, the temperature field is

Q Vx Vr
T = exp K0 (17.26)
πK 2k 2k
2 The out-of-plane condition ezz = 0 will induce a stress σzz = −EαT .

√
(Carslaw and Jaeger 1959, Sect. 10.7), where r = x 2 +z 2 is the distance from the
source and K 0 (·) is the zeroth-order modified Bessel function of the second kind.
If the surface of the half-plane is traction-free, the thermoelastic surface displace-
ment due to the temperature field (17.26) is
2δk Q
u z (x, 0) = − x <0 (17.27)
V
2δk Q Vx Vx
=− exp − I0 x >0 (17.28)
V 2k 2k
(Barber 1984), where I0 (·) is the zeroth-order modified Bessel function of the first
kind. Thus, the surface displacement decays rapidly ahead of the moving source, but
remains constant behind it.
The dimensional prefactor in Eqs. (17.27), (17.28) can also be written
2δk Q Q 2δkG
= H (1 − ν) where H = (17.29)
V GV (1 − ν)
is a dimensionless thermomechanical material parameter, whose value is quite close

to unity for a wide range of materials (Hills and Barber 1985).
17.2 The Axisymmetric Thermoelastic Hertz Problem
We consider the case where two large bodies with quadratic axisymmetric surfaces
are pressed together by a force P, as shown in Fig. 17.2. The extremities of the bodies
are maintained at different temperatures T1 , T2 respectively, so that heat is conducted
between them. The isothermal equivalent of this problem was solved in Sect. 5.2.
Most of the heat flow will pass through the contact area and it is therefore conve-
nient to start with the idealized problem in which no heat flows across the exposed
surfaces,3 whilst there is perfect thermal contact (continuity of temperature) through-
out the contact area. It seems reasonable to assume that the contact area will be
circular, so the boundary conditions can be stated as
g(r ) = 0; p(r ) ≥ 0; T (1) (r, 0) = T (2) (r, 0) 0≤r <a (17.30)
g(r ) > 0; p(r ) = 0; qz (r, 0) = 0 r > a, (17.31)
where T (1) (r, z), T (2) (r, z) are the temperatures of bodies 1, 2 respectively, and we
also assume continuity of heat flux qz (r, 0) across the interface.
3 It can be shown that the use of more realistic thermal boundary conditions in the separation region
has no effect on the pressure distribution [see Problem 17.1].

Fig. 17.2 Axisymmetric

Hertzian contact between P T2
two bodies at different
temperatures
E 2 ν2 α 2 K 2
E1 ν1 α1 K1 a
z
T1
Taking the radius a of the contact circle as an independent variable, we first

solve the heat conduction problem; then use Eq. (17.14) to determine the shape of
the thermoelastically deformed bodies, and finally determine the force P and the
corresponding pressure distribution p(r ) required to establish contact between these
bodies over the given circle.
17.2.1 The Heat Conduction Problem
Using the analogy between heat conduction and elastic indentation of Sect. 4.1.1 and
Eq. (4.24), and the flat punch solution of Eq. (5.21), we can show that a heat flux
Q
qz (r ) = √ (17.32)
2πa a 2 − r 2
over the circle 0 ≤r < a on the surface of the half space z > 0 will cause that circle
to achieve a uniform temperature
Q
ΔT = (17.33)
4K a
relative to the temperature ‘at infinity’ (i.e. as z → ∞). It follows that in the problem
of Fig. 17.2, the contact area will be an isotherm at temperature
K 1 T1 + K 2 T2
T0 = , (17.34)
K1 + K2
17.2 The Axisymmetric Thermoelastic Hertz Problem 403
and the heat flux across the contact circle will be given by (17.32), where the total
heat flux
1 1 1
Q = 4K ∗ (T2 − T1 )a and ∗
= + . (17.35)
K K1 K2
17.2.2 Thermoelastic Distortion
As in Sect. 17.1.3, the surface displacement of a traction-free half-space subjected

to the heat flux (17.32) can be written in the form

1 d du z δQ
r = √ 0 ≤ r < a. (17.36)
r dr dr 2πa a 2 − r 2
Integrating and imposing the continuity condition du z /dr = 0 at r = 0, we obtain

√
du z δ Q a − a2 − r 2
= . (17.37)
dr 2πar
This thermoelastic distortion is additive to the initial gap between the spherical
bodies, so we can define a modified gap function g0 (r ) such that
√
r (δ1 − δ2 )Q a − a 2 − r 2
g0 (r ) = + , (17.38)
R 2πar
where
1 1 1
= + (17.39)
R R1 R2
and R1 , R2 are the radii of the contacting bodies.
17.2.3 Solution of the Contact Problem
To complete the solution, we need to determine the force P required to establish

a circular contact area of radius a between two bodies defined by the gap function
whose derivative is (17.38). From Eq. (5.28), we obtain
a
r 2 g0 (r )dr 4E ∗ a 3 2K ∗ (δ1 − δ2 )(T2 − T1 )E ∗ a 2
P = 2E ∗ √ = + , (17.40)
0 a2 − r 2 3R π
after substituting for Q using (17.35).

The contact pressure can be determined from Eqs. (17.38), (5.26), (5.24) as
√ √

2E ∗ a 2 − r 2 2K ∗ (δ1 − δ2 )(T2 − T1 )E ∗ π 2 a − a2 − r 2
p(r ) = + − χ2 √
πR π2 8 a + a2 − r 2
(17.41)
(Barber 1973), where
∞
x 2m−1
1 x
1+t dt
χ2 (x) = ln = (17.42)
2 0 1−t t m=1
(2m − 1)2
is Legendre’s chi function.

Equation (17.40) reduces to the classical Hertzian solution with contact radius
1/3
3P R
aH = (17.43)
4E ∗
if there is no heat flow (T1 = T2 ), or if the materials have equal thermal distortivities
(δ1 = δ2 ). In this latter case, the thermoelastic distortion generates a convex bulge
in the cooler body and a conformal concave depression in the hotter body, so the
mechanical contact problem is unchanged.
Following Johnson (1985), we can write Eq. (17.40) in the form
3 2
a a
+β = 1, (17.44)
aH aH
where
3R K ∗ (δ1 − δ2 )(T2 − T1 )
β= (17.45)
2πa H
is a measure of the effect of thermoelastic distortion.

Figure 17.3 shows the dimensionless contact radius a/a H as a function of β. If the
product (δ1 −δ2 )(T2 −T1 ) > 0—i.e. the heat flows into the material with the higher
Fig. 17.3 Effect of

thermoelastic distortion on
the contact radius in
axisymmetric Hertzian
contact. Thermoelastic
effects are characterized by
the parameter β defined in
Eq. (17.45)
17.2 The Axisymmetric Thermoelastic Hertz Problem 405
distortivity—the bulge in the cooler body is larger than the depression in the hotter
body, so the contact becomes less conformal and the contact radius decreases.
One consequence of these results is that the thermal resistance
T2 − T1 1
= (17.46)
Q 4K ∗ a
depends on the direction and magnitude of the heat flux. Clausing (1966) demon-
strated this effect experimentally using two contacting cylinders with spherical ends,
and predictions based on an approximate form of Eq. (17.40) with allowance for the
constriction alleviation effect4 showed good agreement with Clausing’s measure-
ments (Barber 1971).
Equation (17.44) has a convenient dimensionless form, but it obscures an interest-
ing limiting case—that in which the bodies are both flat, so R → ∞. In this case, the
first term on the right-hand side of (17.40) is zero, and the contact radius is therefore
given by
πP
a= . (17.47)
2K ∗ (δ 1 − δ2 )(T2 − T1 )E
∗
In other words, even if the bodies are initially flat and hence conform everywhere in
the unheated state, the thermoelastic problem involves a finite circular contact.5
17.3 Existence and Uniqueness
Equation (17.40) and Fig. 17.3 show that, when the heat flows out of the more dis-
tortive body, the contact radius exceeds the Hertz radius. This implies a corresponding
reduction in thermal contact resistance and Clausing’s experimental results confirm
this conclusion. However, a closer examination of Eq. (17.41) shows that very near
to r = a, the contact pressure p(r ) is always negative, indicating tension, when
(δ1 −δ2 )(T2 −T1 ) < 0.
Comninou and Dundurs (1979b) have shown that this difficulty arises because of
the idealized thermal boundary conditions (17.30), (17.31). They used the asymp-
totic method of Chap. 10 to determine the leading term in the temperature and stress
fields at a transition between a region of perfect contact (temperature continuity) and
one of separation and perfect insulation. If the profiles of the contacting bodies are
smooth—i.e. if there is no discontinuity in surface slope—the resulting fields violate
the unilateral inequalities on one or other side of the transition whatever sign is taken
on the multiple of the leading term when the local heat flux is directed out of the more
distortive material. Furthermore, if this term is set to zero, the problem becomes over-
4 See Sect. 16.9.2.

5 We might note that the location of this contact circle now becomes indeterminate.
Fig. 17.4 Components of heat conduction

the thermoelastic contact contact area and problem temperature
problem contact pressure distribution
distribution
elastic thermoelastic
contact problem problem
free thermoelastic
distorted shape
constrained and hence ill-posed. Thus the contact problem defined by the boundary
conditions (17.30), (17.31) is well-posed only for the case (δ1 −δ2 )(T2 −T1 ) > 0.
It should be emphasized that classical existence and uniqueness theorems do
not apply to coupled thermoelastic contact problems. Existence and uniqueness of
solution can be proved for each of the three sub-problems identified in Fig. 17.4.
For example, if the contact area is known, Eqs. (17.30), (17.31) define a well-posed
problem from which a unique temperature distribution can be obtained, and if the
corresponding shape of the unrestrained thermoelastically deformed body is known,
the elastic contact problem falls under the existence and uniqueness proof of the
Signorini problem (Fichera 1964).
However, these separate theorems provide no guarantee that the coupled problem
has one and only one solution, and in fact the Hertzian problem of Sect. 17.2 provides
a counterexample to a putative existence theorem.
17.3.1 A One-Dimensional Model
Some insight into the nature of this paradox can be obtained by considering the
simple one-dimensional system of Fig. 17.5 (Barber et al. 1980). A thermoelastic
rod of length L is built into a rigid wall at A and separated from a second rigid wall
at B by a small gap g. The gap is equal to g0 when the temperature of the rod, T = 0.
Fig. 17.5 A one-

dimensional rod that can A B
C
make contact with a rigid
wall TA TB
L g
x
17.3 Existence and Uniqueness 407
If there is no heat flow across the gap, the rod will reach a uniform temperature
equal to that of the of wall A and elementary calculations show that the gap is reduced
to
g = g0 − αL T A . (17.48)
The gap cannot be negative, so the configuration of Fig. 17.5 is possible only for
αL T A < g0 . For higher values of T A we anticipate contact between the rod and the
wall at B. If there is perfect thermal contact, there will be heat flow along the rod and
the steady-state temperature will vary linearly from T A to TB . Elementary calculations
then show that the contact pressure p is given by
pL αL(T A + TB )
= − g0 . (17.49)
E 2
This state is possible only if p > 0, so the system is governed by the two inequalities
αL(T A + TB )
αL T A < g0 (separation); > g0 (contact). (17.50)
2
If TB > T A , there is a range in which both inequalities are satisfied and the steady-
state solution is non-unique. If TB < T A there is a range in which neither inequality is
satisfied and no steady-state solution exists. As in the Hertz problem, non-existence
is associated with heat flow out of the more distortive material (in this case the rod).
17.3.2 Effect of a Thermal Interface Resistance
In practice, it is unreasonable to expect a discontinuous transition from perfect ther-

mal contact to perfect insulation during separation. If the surfaces are very close, but
not touching, there will be some heat exchange by conduction and radiation across the
gap, and we saw in Chap. 16 that in the case of contact, the roughness of the surfaces
causes less than complete contact at the interface resulting in a contact resistance that
is approximately inverse with nominal contact pressure. For the present example, all
we shall assume is that the thermal resistance is a monotonically decreasing function
of contact pressure a result that can be confirmed very easily by touching a hot [or
cold] object, first with light finger pressure and then with a firm grip.
Suppose that the free end of the rod in Fig. 17.5 is at temperature, TC in the steady
state. If there is a thermal interface resistance R, the steady-state heat flux q must
satisfy the equations
TC − TB K (T A − TC )
q= ; q= (17.51)
R L
and we can eliminate q to obtain
K RT A + L TB
TC = . (17.52)
KR+L
The unrestrained thermal expansion of the rod is αL(T A +TC )/2 and elementary
calculations then show that the gap g is defined by the equation
2 [g − (g0 − αL T A )] L
f (g) = where f (g) ≡ . (17.53)
αL(T A − TB ) K R(g) + L
This condition can be generalized to both contact and separation régimes by defining
a generalized gap function g̃ through
g̃ = g g>0 (17.54)
pL
=− p > 0. (17.55)
E
Physical considerations suggest that the contact resistance R(g̃) should be a monoton-
ically increasing function of g̃, tending to a small positive value as g̃ → −∞ [very
large contact pressure] and to infinity as g̃ → ∞ [very large gap]. The corresponding
function f (g̃) defined in Eq. (17.53) must therefore have the general form shown in
Fig. 17.6, constrained between the limits 0 < f (g̃) < 1.
The solution of Eq. (17.53) is defined by the intersection of the function f (g̃)
and a straight line of slope 2/αL(T A −TB ) representing the right-hand side equation
(17.53)1 . The geometry of the figure shows there must be at least one intersection for
all such straight lines, so the problem of existence of solution is resolved. If the contact
resistance and hence f (g̃) is monotonic, only one intersection can occur if the slope
is positive (i.e. T A > TB ). This is illustrated by line I in Fig. 17.6. However, multiple
solutions can occur for sufficiently small negative slopes (T A < TB ) as represented
by line II in Fig. 17.6.
II: TA < T B I: TA > T B

1
(a) ~
f (g)
(b)
(c)
0 ~g
contact ~g < 0 separation ~g > 0
Fig. 17.6 Graphical solution of Eq. (17.53)

Duvaut (1980) proved an existence theorem for the general three-dimensional

thermoelastic contact problem for the special case where the thermal contact resis-
tance varies inversely with the contact pressure. He also proved uniqueness of
steady-state solution under the condition that this pressure dependence be ‘suffi-
ciently weak’. In the context of the rod model, this corresponds to the condition that
the maximum slope of the function f (g̃) be sufficiently small, but it is clear that
the required value depends on the slope of the straight line II and indeed that it is
always possible to define temperatures T A , TB to ensure non-uniqueness of solution.
Duvaut’s boundary condition is a special case of that used in the above treatment
of the rod model, but the idealized condition of perfect thermal contact or perfect
insulation is not, because it exhibits a discontinuity in resistance at the transition
from contact to separation—i.e. when p = 0 and g = 0.
17.3.3 Imperfect Thermal Contact
It is tempting to speculate that existence of solution for general geometries and

boundary conditions is guaranteed provided the resistance function is any continuous
function of p, g [not necessarily monotonic], including at the transition between
contact and separation. However, to the present author’s knowledge no such theorem
has as yet been proved, though neither have any counterexamples been discovered.
The nonlinearity associated with a pressure or gap-dependent resistance makes the
solution of boundary-value problems difficult, even when using numerical methods.
A simpler approximation that achieves the same purpose [continuity of resistance]
and preserves linearity is to suppose that the transition from perfect thermal contact
to perfect insulation occurs over an infinitesimal range of contact pressure and/or
gap. This is equivalent to defining a function f (g̃) in Fig. 17.6 that changes rapidly
from 0 to 1 near g̃ = 0, whilst still remaining continuous and monotonic.
II: TA < T B I: TA > T B

1
perfect ~
f (g)
contact
imperfect perfect
contact insulation
(a)
0 ~g
contact ~g < 0 separation ~g > 0
Fig. 17.7 The function f (g̃) in the limiting case including the state of imperfect thermal contact
This limiting case is illustrated in Fig. 17.7. It can be defined by f (g̃) = H (−g̃),
but the step function must include the vertical segment g̃ = 0, 0 < f (g̃) < 1, which
we define as a state of ‘imperfect thermal contact’. In this figure, line I corresponds
to a case where neither perfect contact nor perfect insulation is possible, since the
line passes through the ‘gap’ between these two segments. However, a solution
exists in the imperfect contact segment, corresponding to the intersection (a). We
can characterize this state by noting that the rod expands under the influence of
heating from wall A until it is near enough to wall B to achieve a contact conductance
sufficient to maintain a steady state.
17.3.4 The Hertz Problem Revisited
Comninou and Dundurs (1979b) have shown that the asymptotic conditions are
satisfied for heat flow out of the more distortive body if a region of imperfect contact
is interposed between regions of perfect contact and separation. Thus, we can restate
the boundary conditions for the Hertzian problem of Sect. 17.2 for this case as
g(r ) = 0; p(r ) ≥ 0; T (1) (r, 0) = T (2) (r, 0) 0≤r <a (17.56)

g(r ) = 0; p(r ) = 0; T (2) (r, 0) − T (1) (r, 0) qz (r, 0) > 0 a<r 0; p(r ) = 0; qz (r, 0) = 0 r > b, (17.58)
where the inequality in (17.57) states that the thermal contact resistance must be
positive.
The solution of this boundary-value problem is given by Barber (1978) and some
results are presented by Johnson (1985). In particular, the outer radius a of the
imperfect contact region exceeds the radius of the incorrect ‘perfect contact’ circle
of Fig. 17.3 and Eq. (17.44) by up to 20%, but the total heat flux Q is underestimated
by (17.35), (17.44) by only 5%.
Imperfect thermal contact boundary conditions of the form (17.56)–(17.58) have
been used in the solution of other thermoelastic contact problems involving heat flow
out of the more distortive material, including the case where thermoelastic contact
occurs between the faces of an interface crack (Barber and Comninou 1983).
17.3.5 Stability
The stability of the steady-state solutions for the thermoelastic rod identified in
Fig. 17.6 can be investigated by performing a linear perturbation analysis about the
steady state. We first write a general transient solution for the temperature T in the
form
T (x, t) = T0 (x) + T1 (x, t), (17.59)
where T0 (x) is the temperature field in the rod in the steady state and T1 (x, t) repre-
sents a time-varying perturbation. The steady state satisfies the governing equations
and the boundary conditions ex hypothesi. Thus, T1 (x, t) = 0 must always be a possi-
ble [though trivial] solution, showing that the equations satisfied by the perturbation
must be homogeneous. All the governing equations of the problem are linear except
for (17.51)1 , where nonlinearity is introduced by the dependence of R on g̃. How-
ever, we can linearize this relation for sufficiently small perturbations in temperature
about the steady state. Writing (17.51)1 in the form
ln(q) = ln(TC − TB ) − ln(R) (17.60)
and differentiating, we find
Δq ΔTC R (g̃0 )Δg̃

= 0 − , (17.61)
q0 TC − TB R0
where q0 , TC0 , R0 , g̃0 are the values of q, TC , R, g̃ at the steady state, Δq, ΔTC , Δg̃
are [small] perturbations in q, TC , g̃, and
∂R
R = . (17.62)
∂ g̃
The equations governing the linearized perturbation permit separated variable solu-
tions of the form
T1 (x, t) = Θ(x)ebt . (17.63)
Substituting this expression into the heat conduction equation
∂2 T 1 ∂T
− = 0, (17.64)
∂x 2 k ∂t
we obtain
d 2Θ b
− λ2 Θ = 0 where λ = (17.65)
dx2 k
and k is the thermal diffusivity. The general solution is
Θ(x) = A cosh(λx) + B sinh(λx), (17.66)
where A, B are arbitrary constants. The temperature at x = 0 is fixed at T (0, t) =

T A , so Θ(0) = 0, giving A = 0. The three perturbations, Δq, ΔTC , Δg̃ can then be
determined as
∂T1
Δq = −K (L , t) = −B K λ cosh(λL)ebt , (17.67)
∂x
ΔTC = B sinh(λL)ebt , (17.68)
L
Bα
Δg̃ = −α T1 (x, t)d x = − (cosh(λL) − 1)ebt . (17.69)
0 λ
When these expressions are substituted into the perturbed resistance Eq. (17.61),
the common factor B exp(bt) cancels. Using results from Sect. 17.3.2 for the steady-
state quantities q0 , R0 , TC0 , and Eq. (17.53) to express R(g̃) in terms of f (g̃), we then
obtain the characteristic equation
(1 − f 0 )z 2 cosh(z) + α(T A − TB ) f 0 (cosh(z) − 1) + f 0 z sinh(z) = 0, (17.70)
where
Lb
z = λL = and f 0 = f (g̃0 ); f 0 = f (g̃0 ). (17.71)
k
The zeros of this equation define the eigenvalues b at which non-trivial perturbations
of the form (17.63) can exist. A more general transient solution can then be written
by superposition as an eigenfunction series
∞

T (x, t) = T0 (x) + Θi (x)ebi t , (17.72)
i=1
where bi is the i-th eigenvalue of (17.70) and Θi (x) is the corresponding eigenfunc-
tion. If this series can be assumed to be complete on the domain x ∈ (0, L), it follows
that the system is unstable if and only if one or more of the eigenvalues bi has positive
real part.
The zeros of Eq. (17.70) were investigated by Barber et al. (1980), who showed
that the condition for there to be no eigenvalues with positive real part is
αL(T A − TB ) f 0 < 2, (17.73)
in which case the corresponding steady state is stable. If the contact resistance varies
monotonically with contact pressure and gap, the function f (g̃) will be monotonically

decreasing, as shown in Fig. 17.6, and f will be everywhere negative. The inequality
Fig. 17.8 One-dimensional

model involving two C TB
different materials TA
A B
L1 L2
x g
(17.73) will therefore always be satisfied if TB < T A —i.e. if the heat flows out of the
rod at the contact interface. For TB > T A , the stability criterion (17.73) has a simple
graphical interpretation. Intersections in Fig. 17.6 represent stable steady states if the
straight line cuts the curve from above with increasing g̃ and unstable solutions if it
cuts from below. Thus, of the three intersections (a, b, c) with line II, (a, c) represent
stable steady states and (b) is unstable. This enables us to conclude that if there is
only one steady-state solution, it is stable, but if there are more than one, they will
be alternately stable and unstable.
17.3.6 Contact of Dissimilar Materials
The rod model discussed above involves a single thermoelastic body contacting a
rigid wall at a prescribed temperature. The stability behaviour becomes considerably
more complex when contact occurs between two thermoelastic bodies of different
materials.
Figure 17.8 shows a generalization of the rod model in which two rods of dissimilar
thermoelastic materials are built into two rigid walls, A, B and make contact or have
a small gap at C. The steady-state behaviour of this system is easily analyzed as in
Sect. 17.3.2 and leads to similar results, but the stability behaviour depends on the
relative values of the two dimensionless ratios

L 2 k1 γ1
and ,
L 1 k2 γ2
where γ = αK /k (Barber and Zhang 1988). As a consequence, it is possible to

choose the lengths of the rods such that the steady state is unique but unstable in
appropriate temperature ranges. In these cases, numerical studies of the transient
behaviour show that the system reaches a limit cycle oscillation, usually involving
periods of contact alternating with periods of separation. Clausing (1963) reported
slow periodic variations in experimental measurements of thermal contact resistance
that may be attributable to this mechanism, as may observations of erratic heat transfer
performance in duplex heat exchanger tubes (Srinivasan and France 1985).
17.3.7 Two-Dimensional Stability Problems
Consider the case where a uniform heat flux qz = q0 is conducted across the common
interface z = 0 between two large bodies z > 0 and z < 0 of different materials, denoted
here by the suffices i = 1, 2 respectively. We assume that the system is locally one-
dimensional, so that temperatures and stresses depend only on the coordinate z. In
particular, the contact pressure is uniform and equal to p0 , and there is a pressure-
dependent contact resistance R( p).
The stability of this simple state can be investigated by postulating the existence
of a small perturbation in the temperature field. As in Sect. 17.3.5, the perturbation
must satisfy a homogeneous linear partial differential equation, so we seek spatially
sinusoidal solutions of the form

T (x, z, t) = T0 (z) + Θ(z)ebt+ımx , (17.74)
where T0 (z) is the one-dimensional steady-state solution. Solutions for Θ(z) satis-
fying the heat conduction equation can then be found as

b
Θ(z) = Ai e±λi z where λi = + m2, (17.75)
ki
ki is the thermal diffusivity of material i and the sign in the exponential is taken to
ensure that the perturbation decays with distance from the interface. The solution
then proceeds as in Sect. 17.3.5 leading to the characteristic equation

∗ δ1 δ2 1 1
E R ( p0 ) q 0 m − + R( p0 ) + + =0
λ1 (λ1 + m) λ2 (λ2 + m) K 1 λ1 K 2 λ2
(17.76)
for the exponential growth rate b, which appears implicitly in λ1 , λ2 . This equation
was investigated by Zhang and Barber (1990), who found that for many material
combinations, the stability boundary corresponds to b = 0, which defines a neutrally
stable steady-state solution and hence is independent of the thermal diffusivities.
In this case, instability occurs if and only if the heat flux is directed into the more
Fig. 17.9 Contact of two q0

thin-walled cylinders on
their end faces
p0
interface
1
p0
q0
distortive material and exceeds a certain critical magnitude. However, there also exist
other material combinations for which instability can also occur for heat flow into
the less distortive material, in which case the growth rate is complex, indicating
oscillatory growth.
A simple system exhibiting this behaviour comprises two stationary thin-walled
cylinders in contact on their end faces, as shown in Fig. 17.9. If the heat flows into
the less distortive material, the one-dimensional steady-state solution with a uniform
contact pressure p0 is unique, but material properties can be chosen so as to make
this unique solution unstable above some critical heat flux. A numerical solution of
the problem by Zhang and Barber (1993) shows that the long-time state then involves
one or more contact and separation regions that move around the interface at constant
speed, along with the associated thermal and mechanical fields.
In all cases, the critical heat flux is proportional to the wavenumber m, suggesting
that it will vary inversely with appropriate linear dimensions of the contact region in
problems with finite geometry.
17.4 Solidification Problems
During the casting process, heat is conducted across a solid/solid interface from the
partially solidified casting into the mould. The thermal resistance at this interface
plays an important rôle in the evolution of solidification and the development of
the final grain structure and residual stress. Ho and Pehlke (1985) deduced values
of thermal contact resistance from temperature measurements during solidification
experiments and found that the resistance generally increases significantly as solid-
ification proceeds.
A possible explanation for this phenomenon is suggested by Richmond and Tien
(1971), who show that thermoelastic shrinkage of the casting will cause air gaps to
form at some locations on the interface. Also, if the melt wets the mould surface, the
solidified material will initially conform at least to the long wavelength roughness,
in which case subsequent relative tangential motion due to thermoelastic distortion
may reduce the extent of intimate contact between the surfaces (Borri-Brunetto et al.
1998).
As in the thermoelastic contact of two solids, we should anticipate the possibility
of instability associated with the pressure dependence of the contact resistance and
indeed there is ample experimental evidence of waviness in the development of the
solidification front in nominally uniform solidification that is probably attributable to
this mechanism (Richmond and Huang 1977; Wray 1981). For example, Fig. 17.10
shows a segment from a partially solidified cylinder manufactured by a continuous
casting process.
This waviness leads to a corresponding non-uniformity in the morphology and
concentrations in the solidification of alloys and can even cause remelting in regions
where air gaps develop. This phenomenon should be distinguished from morpholog-
ical instability, which results from differential solidification rates for the components
Fig. 17.10 Non-uniform

solidification during
continuous casting of an
aluminium cylinder
of an alloy and is characterized by the formation of dendrites. Thermoelastic insta-

bility can occur in the solidification of pure metals and typically has a significantly
larger wavelength (≈50 mm) than that associated with morphological instability.
The simplest problem of this class involves the nominally one-dimensional solid-
ification of a pure metal in contact with a plane mould, as shown in Fig. 17.11. We
consider the possibility of a solution for the temperature field of the form
T (x, z, t) = T0 (z, t) + T1 (z, t) cos(mx), (17.77)
where the perturbation T1 is in some sense small compared with the unperturbed solu-
tion, T0 . The solidification front will also contain a sinusoidal perturbation defined
by
s(x, t) = s0 (t) + s1 (t) cos(mx). (17.78)
Notice that the unperturbed solution is now time-dependent and nonlinear because
of the moving boundary, s0 (t). As long as the perturbation is sufficiently small, we
can linearize about the zeroth-order solution, so the equations determining T1 will be
linear, despite the presence of a perturbation in the moving boundary. For example,
z
2π /m
liquid T= Tm
s 1 (t)
s 0 (t)
solid s(x,t)
O T= 0 x
rigid mould pressure-dependent
contact resistance
Fig. 17.11 Perturbed unidirectional solidification

17.4 Solidification Problems 417
the condition that the temperature on the boundary z = s(x, t) is equal to the melting
temperature, Tm , takes the linearized form
∂T
T (x, s, t) ≈ T (x, s0 , t) + (x, s0 , t)(s − s0 )
∂z
∂T0
≈ T0 (s0 , t) + T1 (s0 , t) cos(mx) + (s0 , t) s1 cos(mx)
∂z
= Tm , (17.79)
where we have dropped product terms in the perturbation. We conclude that
∂T0
T0 (s0 , t) = Tm and T1 (s0 , t) = − (s0 , t) s1 . (17.80)
∂z
As usual in linear perturbation problems, the zeroth-order problem is unaffected by

the presence of the perturbation and the perturbation is governed by homogeneous
equations whose coefficients involve derivatives of the zeroth-order solution. These
coefficients will generally be functions of time, so we cannot assume exponential
time-dependence for the perturbation, as in Sect. 17.3.7. We also need to redefine the
concept of instability, since an arbitrarily small initial perturbation would not have
time to grow to serious proportions during the process.
Algebraic solutions have been obtained for a variety of idealized problems, mostly
involving pure metals (Yigit and Barber 1994; Hector et al. 1996; Yigit 1998). The
method is capable of extension to cases where the zeroth-order problem possesses
fairly general nonlinearities, but the zeroth-order problem must then generally be
solved by numerical methods, leading to a set of linear differential equations for
the perturbation that contains piecewise discrete functions. A typical application of
this kind is that in which the material properties are temperature-dependent (Yavuz
1995). These effects can be important, since properties can vary quite extensively at
temperatures near the melting point.
The most rapidly growing perturbations tend to be those whose wavelength is of
the order of 10 ∼ 20 times the instantaneous mean thickness of the solid layer, and
this result is consistent with the wavelengths visible in Fig. 17.10. Since the mean
thickness itself increases during the process, we see initially short wave disturbances,
but these stabilize and become dominated by longer wavelength disturbances as
solidification proceeds.
17.5 Frictional Heating
Arguably the most technologically important applications involving thermoelastic

contact instability are those in which the thermal effects arise from frictional heating
at a sliding interface. If we assume Amonton’s law of friction with coefficient f , the
heat generated per unit area, per unit time at the interface is given by
q(x, y, t) = f V p(x, y, t), (17.81)
where V is the sliding speed, the coordinates (x, y) define a point in the contact area
and p(x, y, t) is the local contact pressure.
The nature of the potentially unstable feedback process is best captured by
Fig. 17.12, which shows that the thermal and mechanical contact problems are cou-
pled only through the boundary conditions. It is clear that this constitutes positive
feedback, since a high-pressure region will receive more heating and will expand
more, leading to a local increase in the contact pressure. Furthermore, Eq. (17.81),
repeated in Fig. 17.12, shows that the product f V acts as the gain in the feedback
process. Thus, we anticipate that for a given friction coefficient, there will be some
sliding speed V0 , above which the system will be unstable (Dow and Burton 1972).
The corresponding eigenmode will then grow exponentially with time, leading even-
tually to localization of the contact force in a small region of the nominal contact
area, and to high local temperatures, known as hot spots. This phenomenon is known
as frictionally excited thermoelastic instability or TEI (Barber 1969; Dow and Burton
1972) and is of critical importance in the design of brakes and clutches (Kennedy
and Ling 1974; Thoms 1988; Anderson and Knapp 1990).
It is important to distinguish the hot spots due to TEI from flash temperatures in
sliding, associated with the microscopic roughness of the contacting surfaces.6 Flash
temperatures are typically very transient, usually lasting for periods of ms or less and
they are also localized in the vicinity of actual contact areas, whose dimensions are
typically measured in microns. By contrast, typical thermoelastically developed hot
spots have dimensions of the order of millimetres and may last for several seconds
or even minutes. This permits the non-uniform temperature field to penetrate into
the sliding bodies and can result in thermal damage to the surfaces, such as heat
checking (cracking) or surface melting.
Experimental evidence of this scale of hot spotting was first reported in tread-
braked railway wheels by Parker and Marshall (1948), but the explanation in terms
of thermoelastic instability was first advanced by Barber (1969). Confirmation of this
explanation was provided by the observation that regions of the brake block that had
Fig. 17.12 The feedback

Contact pressure p(x,y,t)
process for frictionally
and contact area
excited thermoelastic
instability
Solution of contact Frictional heat generation

problem q(x,y,t) = f Vp(x,y,t)
Unconstrained thermal Solution of heat

distortion conduction problem
6 see Sect. 18.6.1 below.

17.5 Frictional Heating 419
just experienced a thermal excursion were significantly recessed below the prevalent
surface level. The only way material could be removed in such locations is if they
were elevated above the recessed level during the sliding process as a result of local
thermoelastic distortion.
17.5.1 The Rod Model
A more definitive experimental confirmation of this mechanism was provided by the

idealized sliding model of Fig. 17.13. Three rods were clamped into a holder, so as
to constitute a brake block with three separate possible contact areas. Conventional
brake design practice argues that the most highly loaded areas will wear the most,
leading eventually to a state in which the force is shared equally between the three
rods. The actual behaviour is quite different. The rod which carries the greatest force
also experiences the greatest frictional heating and hence expands the most, and the
applied force is soon concentrated on a single rod.
Eventually, sufficient wear accumulates to cause a second rod to make contact.
The new rod being initially cold expands rapidly and takes over the force. A regular
sequence is soon established with the rods taking turns at carrying the force.
Figure 17.14 shows typical experimental results for (a) the temperatures near the
surface of the three rods and (b) the normal separation of the two sliding bodies.
Notice how separation increases as the bodies are forced apart by thermal expansion
during and just after the load-transfer process, whereas towards the end of each cycle,
the rod temperature is approaching a steady state and wear starts to dominate over
incremental thermal expansion.
Fig. 17.13 Schematic applied

representation of an force
experimental model with
three discrete contact areas
moving surface
Fig. 17.14 (a) Temperatures (a)

and (b) normal separation of
temperature C
the bodies for the idealized 300
o
model of Fig. 17.13 200
100
0
(b)
separation (μm)
40
20
0
0 50 100
time (s)
17.5.2 Burton’s Stability Analysis
Theoretical investigations of TEI were pioneered in the early 1970s by the research
group of R.A. Burton, then at Northwestern University. In particular, Burton intro-
duced the concept of a critical sliding speed for instability and developed the linear
perturbation method as a test for instability in idealized geometries (Dow and Burton
1972; Burton et al. 1973a).
The simplest such geometry involves the two half-spaces z > 0 and z < 0 sliding
against each other at their common plane surface, z = 0. The bodies are maintained
in contact by a nominally uniform contact pressure p0 and heat is generated at the
interface according to Eq. (17.81). The partition of this heat between the sliding
bodies is determined by the condition of temperature continuity at the interface. In
other words, we assume that there is no thermal contact resistance.
As in Sect. 17.3.7, we postulate the existence of a two-dimensional, spatially
sinusoidal disturbance in temperature that grows exponentially in time—i.e.

T1 (x, y, z, t) = Θ(z)ebt+ımx . (17.82)
Because of the relative motion, at least one of the bodies will move relative to the
reference frame, thereby introducing convective terms into the heat conduction equa-
tion. If the sliding is in the out-of-plane direction y, the corresponding derivatives will
be with respect to y and hence will have no effect on the perturbation of Eq. (17.82).
However, for in-plane sliding (in the x-direction), the convective terms cause the
eigenvalues b to be complex, corresponding to sinusoidal perturbations that migrate
in the sliding direction at a speed that is usually intermediate between the speeds of
the two contacting bodies.
17.5.3 Out-of-Plane Sliding
For the simpler out-of-plane case, the dominant eigenvalue is real and there is no
migration. The stability boundary therefore corresponds to the first value of sliding
speed V at which there exists a zero eigenvalue b = 0. This defines a steady state
(zero exponential growth), so an alternative statement of the stability criterion is
that there should exist a non-trivial solution to the steady-state perturbation problem.
This concept is analogous to the solution of elastic stability problems by seeking the
forces at which non-trivial states of neutral equilibrium exist.
To illustrate the solution process, we consider the case where one of the bodies
[z < 0] is a rigid non-conductor, so that all of the frictional heat passes into the other
body z > 0. If b = 0, the temperature perturbation (17.82) must satisfy the steady-
state heat conduction equation ∇ 2 T = 0, and it is easily shown that the solution that
decays with distance from the interface requires Θ(z) = C exp(−mz) and hence
T1 (x, y, z) = C exp(−mz) cos(mx), (17.83)
where C is a constant and we have chosen the origin of coordinates to coincide

with the maximum of the sine wave. The perturbation in heat flux at the interface is
therefore
∂T1
q1 (x) = −K (x, 0) = K Cm cos(mx). (17.84)
∂z
The temperature profile (17.83) will cause thermal strains and a perturbation in
contact pressure p(x) will be needed to maintain contact with the rigid plane. We
write
p(x) = p0 + p1 (x), (17.85)
where the perturbation p1 (x) is calculated using Eq. (17.6) as

∞
Eα m cos(mx) EαC cos(mx)
p1 (x) = exp(−ms)Θ(s)ds = . (17.86)
(1 − ν) 0 2(1 − ν)
Finally, we substitute (17.84), (17.86) into the frictional heating equation (17.81)
and cancel a common factor C cos(mx) to obtain
f V Eα 2K m(1 − ν)
Km = and hence V = , (17.87)
2(1 − ν) f Eα
which defines the critical speed.

Notice that the critical speed is proportional to the wavenumber, m and hence the
system is always unstable for sufficiently small m, corresponding to long wavelength
perturbations. In practical systems, the wavelength of the perturbation is limited by
the finite dimensions of the sliding bodies, which therefore play a crucial rôle in
determining the critical speed of the system.
Long wavelength perturbations become unstable at low sliding speeds, but they
have slow exponential growth rates because of the large mass of material involved
in the thermal transient. Azarkhin and Barber (1985) showed that the perturbation
with the highest growth rate corresponds to a wavenumber

f V Eα
m 0 = 0.47 , (17.88)
2K (1 − ν)
which is 47% of the largest unstable wavenumber defined by Eq. (17.87). The effect
of this ‘dominant perturbation’ is seen in transient solutions of thermoelastic contact
problems involving sliding. Figure 17.15 shows the evolution of the contact pressure
distribution for a two-dimensional elastic body with a quadratic profile in contact
with a rigid plane, for various values of the dimensionless sliding time t˜ = kt/a02 ,
where a0 is the half-width of the initial (isothermal) contact area, and k is thermal
diffusivity. The pressures are normalized by the instantaneous value at the centre of
the contact region. The initial temperature is taken as uniform so that the classical
Hertzian pressure distribution is obtained at time t˜ = 0. As sliding proceeds, frictional
heating causes thermoelastic distortion that reduces the size of the contact area, but
1
~
t = 0
∼
λ
0.8 18
p(x)
p(0)
0.6 67
0.4
135
0.2
0
0 0.2 0.4 0.6 0.8 1
x
a0
Fig. 17.15 Evolution of an initially Hertzian contact pressure distribution due to thermoelastic
distortion during sliding. The dominant wavelength is identified as λ̃
waves begin to develop in the contact pressure distribution and eventually, these
dominate the transient process, leading to bifurcation of the contact area.7
Subsequent evolution of the process (not shown here) involves the ‘competition’
between these distinct contact regions until eventually only one connected contact
region remains in the steady state (Azarkhin and Barber 1986). The wavelength
corresponding to the maximum exponential growth rate is shown in Fig. 17.15 for
comparison (as λ̃) and clearly correlates well with the perturbations observed.
17.5.4 In-Plane Sliding
The solution for in-plane sliding was first developed by Burton et al. (1973a). The
perturbation will generally migrate over the surfaces, but the problem is conveniently
formulated by choosing a frame of reference in which the perturbation is stationary
and the two bodies move through this frame with speeds c1 , c2 , respectively. The
sliding speed is then V = |c1 −c2 | and the corresponding temperature field solutions
must satisfy the heat conduction equations including appropriate convective terms.
At the stability boundary, there is no change in the perturbation with time in this
frame of reference and hence
∂ 2 Ti ∂ 2 Ti ci ∂Ti
+ + = 0, (17.89)
∂x 2 ∂z i 2 ki ∂x
where the suffix i refers to bodies 1, 2 respectively and for convenience we have
chosen a local coordinate z i that is directed into body i. Assuming Ti (x, z i ) =
{Θi (z i ) exp(ımx)} and solving the resulting ordinary differential equation for
Θi (z i ), we obtain
Θi (z i ) = Θ0 exp(−λi z i ), (17.90)
where
ıci m
λi = m2 − (17.91)
ki
and Θ0 is the amplitude of the interface temperature perturbation, which must be the
same for both bodies in view of the assumption of temperature continuity.
Substitution into Eq. (17.5) then shows that in the absence of contact pressure, the
distorted surfaces of the half-planes would be defined by

2αi (1 + νi )Θ0 exp(ımx)
u (i)
z (x, 0) = − (17.92)
(m + λi )
7 Itshould be emphasized that these perturbations in the pressure distribution are not the result of
numerical instabilities. Extensive convergence tests and changes of mesh refinement were performed
to ensure they describe real features of the continuum solution.
and to maintain contact between the two surfaces, we therefore require a contact
pressure perturbation8

α1 (1 + ν1 ) α2 (1 + ν2 )
∗
p1 (x) = Θ0 E m + exp(ımx) . (17.93)
(m + λ1 ) (m + λ2 )
The total heat flux from the interface is

∂T1 ∂T2
q(x) = −K 1 (x, 0) − K 2 (x, 0) = {Θ0 (K 1 λ1 + K 2 λ2 ) exp(ımx)},
∂z 1 ∂z 2
(17.94)
and the frictional heating equation q(x) = f |c1 −c2 | p(x) must be satisfied for all x.
This yields the single complex characteristic equation

∗α1 (1 + ν1 ) α2 (1 + ν2 )
(K 1 λ1 + K 2 λ2 ) = f |c1 − c2 |E m + , (17.95)
(m + λ1 ) (m + λ2 )
that can be decomposed into two real equations for the speeds c1 , c2 .
If the migration Peclet number is large—i.e. if
ci
Pei ≡ 1, (17.96)
2ki m
the complex exponential decay rate λi can be approximated by

ci m
λi ≈ (1 − ı) = (1 − ı)m Pei . (17.97)
2ki
The sliding speed V = |c1 −c2 | and hence at least one of the two migration speeds
must be of the same order as the sliding speed, and this implies Peclet numbers of the
order of 104 or 105 in typical tribological applications. The exponential decay in the
corresponding body is therefore extremely steep, causing the thermal disturbance to
be concentrated in a thin ‘thermal skin’ adjacent to the interface. It also follows that
the corresponding thermoelastic displacement from Eq. (17.92) is relatively small,
since λi appears in the denominator of this expression.
When the materials are similar, symmetry arguments can be used to predict the
existence of a solution for which the perturbation speed is intermediate between
those of the sliding bodies, giving c1 = −c2 = V /2. In this case, both migration Peclet
numbers are large and all thermoelastic deformations are therefore small. The system
of two similar half-planes is therefore predicted to be stable at all practical sliding
speeds.
8 This expression is derived from Eq. (6.68) and hence implicitly assumes that the normal and
tangential problems are uncoupled in the sense of Sect. 7.2.3. This approximation was adopted by
Burton et al. (1973a) and is almost universally used in the analysis of TEI. Some justification is
provided by the study of Heckmann and Burton (1977). See also Lee and Barber (1993a). The
assumption is, of course, exact if Dundurs’ constant β = 0.
For dissimilar materials, the disturbance moves more slowly over the surface of
the better conductor (Burton et al. 1973a). Brakes and clutches usually involve a steel
or cast iron disc sliding against a composite friction material whose conductivity may
be as much as 100 times lower than that of the disc. In this case, the migration speed
over the disc (the good conductor) is extremely slow and thermoelastic effects in the
disc are significant, giving critical speeds in the practical operating range. In the limit
where one of the materials has zero thermal conductivity, the migration speed in the
conducting material tends to zero and the analysis simplifies, becoming essentially
identical to that for out-of-plane sliding in Sect. 17.5.3, except that E/(1−ν) must
be replaced by E ∗ (1+ν), where ν is Poisson’s ratio for the conducting body. We
conclude that undesirable TEI effects can be minimized by increasing the thermal
conductivity of the friction material. This has the effect of increasing the migration
speed of the disturbance with respect to the good conductor, hence reducing the
magnitude of the thermoelastic distortions. In automotive disc brakes, two commonly
used friction material categories are non-asbestos organic composites (NAOs) and
semimetallic composites. Semimetallics have significantly higher conductivity than
NAOs because of the metal content. Design experience shows that they are indeed
less prone to hot spots and the associated vibration known as hot judder.
This argument predicts that TEI should not be a problem for sliding bodies of
similar or nearly similar conductivities, but experimental evidence has been reported
that appears to contradict this conclusion (Berry and Barber 1984). A possible expla-
nation of this effect is provided by Burton (1973), who argued that real metallic
surfaces develop thin films of oxides and other non-metallic products at the inter-
face. If we postulate the existence of a perturbation that moves rapidly over one such
body and relatively slowly over the other, the Peclet number for the body with the
high migration speed may be sufficiently large for the thermal skin to be contained
entirely in the oxide film. That body will then behave thermally in the same way
as a half-plane entirely made up of oxide. By contrast, the temperature field in the
body with the slower migration speed will be determined largely by the thermal
conductivity of the substrate metal and the overall system will therefore behave like
an oxide/metal sliding pair.
17.5.5 Limiting Configurations
The equations governing TEI are linear for non-infinitesimal pressure perturbations,
so these continue to grow exponentially as long as the mean pressure p0 is sufficient
to ensure complete contact. Separation regions then develop and the system rapidly
stabilizes at a limit state. The problem of determining these limits is simplified if
one body is a rigid non-conducting plane, since the contact area is then stationary in
the conducting body (Burton et al. 1973b). Problems of this class can conveniently
be formulated by superposing the solutions of Sects. 17.1.2 and A.1 in terms of
harmonic potentials ψ, ϕ. The surface heat flux, contact pressure and normal surface
displacement are then given by

(1 − ν) ∂ 3 ψ ∂2ϕ (1 − ν) ∂ψ ∂ϕ
qz = ; p= ; uz = − + . (17.98)
Gδ ∂z 3 ∂z 2 G ∂z ∂z
The condition qz = f V p applies throughout the surface, since in separation regions

both p and qz are zero. We can therefore satisfy this condition by writing
∂ψ f V Gδ
= γϕ where γ = . (17.99)
∂z (1 − ν)
The remaining boundary conditions u z = d −g0 in the contact region A and p = 0

elsewhere then reduce to

(1 − ν) ∂ϕ
γϕ + = g0 (x, y) − d (x, y) ∈ A (17.100)
G ∂z
∂2ϕ
=0 (x, y) ∈
/ A, (17.101)
∂z 2
which can be configured as a harmonic boundary-value problem of the third kind

(Barber 1976). Similar boundary conditions are encountered in steady-state heat
conduction problems where part of the boundary experiences heat exchange with a
reservoir through a linear heat transfer coefficient h c . However, there is an important
difference in that the constant γ has the opposite sign to h c if α > 0, and hence the
uniqueness theorem for boundary-value problems of the third kind (Sternberg and
Smith 1944) ceases to apply. This should not surprise us, since we have already found
non-uniqueness in other thermoelastic contact problems.
If the initial gap function g0 corresponds to a sphere of radius R, the steady-state
contact area is a circle whose radius a is shown in Fig. 17.16. In interpreting this
figure, notice that P is the applied normal force, and PH = 8Ga 3 /3R(1−ν) is the
force required to establish the same contact radius in the absence of thermoelastic
Fig. 17.16 Contact area dimension for steady-state sliding of a Hertzian contact with heat genera-
tion due to friction (Barber 1976). The dotted line represents the first (linear) term of a power series
approximation to the two-dimensional curve
distortion. Figure 17.16 also presents corresponding results for the two-dimensional
Hertzian problem, for which PH = πGa 2 /2R(1−ν).
If both contacting bodies are plane, so R → ∞, we have PH = 0 and a = 2.006/γ.
In other words, a finite circular contact is established in the steady state even when
the bodies are both plane. We found a similar result for static Hertzian contact
in Sect. 17.2, where the radius of the circle was defined by Eq. (17.47). Burton
and Nerlikar (1975) gave the corresponding solution for the case where one of
the cylinders in Fig. 17.9 is caused to rotate about its axis and the other is non-
conducting. A critical speed can be identified below which full contact is main-
tained at uniform pressure, but above which there is partial contact. As explained in
Chap. 11, Sect. 11.3, when the original contact geometry is completely conformal, the
thermoelastic contact problem is ‘receding’, the extent of the contact area is inde-
pendent of the force P, and the temperature and displacement fields have a unique
form and are simply linearly proportional to P.
17.5.6 Effect of Geometry
As we have seen, the system of two half-planes is theoretically unstable at any

speed for sufficiently long wavelength disturbances. Real systems have finite dimen-
sions and this places a limit on the permissible wavelengths and hence establishes
a non-zero critical speed. A bounded critical speed is also produced for systems
involving one or more sliding layers of finite thickness. Lee and Barber (1993b)
applied Burton’s method to the problem of a layer of thickness 2a sliding between
two stationary half-planes, as shown in Fig. 17.17.
Fig. 17.17 A layer of p0

thickness 2a sliding between
two half-planes
1
a
2
V
1
p0
This system can be regarded as an idealization of a brake disc sliding between

two friction pads. It is symmetric about the mid-plane of the layer and we therefore
anticipate eigenfunctions that are respectively symmetric and antisymmetric with
respect to this plane. The analysis for each case proceeds exactly as in Sect. 17.5.4,
except that results for the layer replace those for the half space.9
The critical speed is shown as a function of dimensionless wavenumber ma in
Fig. 17.18. The material properties used in this figure are those for a cast iron layer
and half-planes of a typical NAO composite. Notice that the critical speed for the
antisymmetric mode is always lower than that for the symmetric mode and hence
the hot spots tend to be antisymmetrically disposed on the disc—i.e. hot spots on
one side of the disc are midway between the hot spots on the other side. This agrees
well with observations of hot spotting in practical disc brakes (Thoms 1988). We
also note that the minimum critical speed occurs around ma = 0.2. Thus, the most
unstable eigenmode has a wavelength
2π
l0 = ≈ 30a, (17.102)
m
which is about 15 times the layer thickness. This dimension is also broadly in agree-
ment with experimental observations of hot spot spacing.
Fig. 17.18 Stability 10 8

boundary for the system of
Fig. 17.17 for a cast iron disc both modes
[thermal diffusivity k2 ] unstable
sliding between NAO
composite half-spaces 10 7
antisymmetric
Va mode unstable
k2
10 6
both modes
105 stable
10 4
10 3
0 1 2 3 4 5 6
ma
9 See for example Sect. 17.1.1.

Zagrodzki et al. (2001) describe a finite element solution of a transient

thermoelastic contact problem similar to Fig. 17.17, except that the half-planes were
replaced by finite bodies. Their results confirm the exponential growth rates and
migration speeds from the analytical solution and show that once separation regions
develop, the migration speed drops significantly, only to recover to near its original
value as the steady-state configuration develops. If the sliding speed is sufficiently
high for several modes to be unstable, the resulting steady-state configuration corre-
sponds to the unstable mode with the longest wavelength, even though shorter waves
may grow faster during the initial transient.
17.5.7 Numerical Solutions
The results of Fig. 17.18 show that the geometry of the sliding bodies has a critical
effect in determining the stability behaviour and the complexity of most practical sys-
tems therefore demands numerical solution. However, direct numerical simulation,
as in Zagrodzki et al. (2001), though practicable with modern computing systems, is
extremely time-consuming, particularly in three-dimensional problems.
An alternative approach is to seek a finite element implementation of Burton’s
perturbation technique. The problem remains linear as long as contact is retained
throughout the nominal contact area and hence perturbations of the form

T1 (x, y, z, t) = Θ(x, y, z)ebt (17.103)
are to be anticipated. Substituting this form into the governing equations of heat
conduction and thermoelasticity permits the exponential term to be cancelled and
leads to a set of modified equations in the spatial coordinates (x, y, z), in which the
exponential growth rate b appears as a linear parameter. Discretizing these equations
by the finite element method and imposing the thermal and mechanical boundary
conditions, including the frictional heat generation condition (17.81), leads to a set
Fig. 17.19 Evidence of

thermal damage to a clutch
disc after a single
engagement, showing
periodic hot spots
Fig. 17.20 Automotive disc

brake
of linear equations for the nodal temperatures in which b appears linearly in the
coefficients. These equations can be configured as a generalized linear eigenvalue
problem for b and the corresponding eigenfunctions then define the mode shapes
corresponding to each eigenvalue. This method was first suggested by Yeo and Barber
(1994), who developed it in the context of the static thermoelastic contact problem
of Sect. 17.3.5.
If the system is axisymmetric, as for example in the case of multidisc clutches,
the eigenmodes must have Fourier form in cylindrical polar coordinates (r, θ, z).
For example, Fig. 17.19 shows a clutch disc after a single high-speed engagement,
showing clear evidence of hot spots in a regular pattern around the circumference.
Each Fourier component can then be investigated separately by writing

T1 (r, θ, z, t) = Θ(r, z)ebt+ınθ (17.104)
Fig. 17.21 Comparison of Experimental result

experimental (◦) and finite 3−D FEA model
600
element (∗) predictions of
critical speed of several
quasiperiodic modes for an 500
critical speed (rpm)
automotive caliper disc brake

(Yi et al. 2002) 400
300
200
100
0
0 2 4 6 8 10 12
hot spot number
and discretizing the function Θ on the two-dimensional domain (r, z) (Yi et al. 2000).
Caliper disc brakes such as that shown in Fig. 17.20 are not axisymmetric and
hence must be discretized in three spatial dimensions. The resulting eigenmodes
show an approximate periodicity with clearly identifiable hot spots, but these cool
down during periods of non-contact, so the associated fields are not strictly periodic.
Yi et al. (2002) used thermal imaging methods to track the transient evolution of
hot spots in a disc brake during a constant speed ‘drag braking’ experiment. Fast
Fourier Transform (FFT) methods were then used to separate the resulting fields into
Fourier components and hence determine approximate exponential growth rates as
a function of the number of waves n. This procedure was applied at several different
speeds and extrapolation of the resulting growth rates allowed the critical speed for
each n to be estimated. Figure 17.21 compares these critical speeds with predictions
based on the discretization of Burton’s perturbation method applied to a finite element
model of the experimental disc brake system.
Problems
Use results from Sect. 17.1 to determine the normal surface displacement of a
traction-free half space due to a steady-state heat source Q per unit length around
the circle r = a, the rest of the surface being unheated.
Hence show that the contact pressure distribution in a steady-state axisymmetric
thermoelastic contact problem is unaffected by the thermal boundary conditions [e.g.
convection or radiation to the environment] outside the circular contact area.
2. A traction-free half-space is subjected to surface heating. Show that the maximum
outward thermoelastic displacement must occur in a heated area and the maximum
inward displacement in a cooled area.
3. The layer 0 < z < h rests on a frictionless rigid foundation at z = 0 and the surface
z = h is traction-free. The foundation is a thermal insulator and the free surface is
subjected to the steady-state heat input
qz (x) = q0 cos(mx).
Use Dundurs’ theorem to show that the layer will not separate from the founda-
tion and find the amplitude of the sinusoidal perturbation in the free surface due to
thermoelastic distortion.
4. Suppose that the thermal resistance for the system in Fig. 17.5 is defined by
Duvaut’s law
C
R( p) = p>0
p
=∞ g > 0.
If the temperature difference (TB −T A ) is specified, find the minimum value of the
constant C if the steady-state solution is to be unique for all values of g0 .
5. Assuming that the two rods in Fig. 17.8 are in a steady state with the gap g
positive, find the equation determining g if the thermal resistance is defined by a
known function R(g). Find a function f (g) in the range 0 < f (g) < 1 that allows
you to express this equation in a form similar to Eq. (17.53) and its solution by an
intersection as in Fig. 17.6. What must be the definition of g̃ if this result is to apply
also to the case of contact at pressure p?
6. An assembly like that in Fig. 17.13, but with only two rods, is sliding at speed V
and transmitting a constant force P. Initially both rods are in contact and they have
identical temperature and stress fields. By postulating the occurrence of exponentially
growing perturbations in these fields, determine the condition for this state to be stable
if wear is governed by Archard’s wear law ẇ(t) = γ f |V | p(t) where w is the depth
of material removed, p is the contact pressure, f is the coefficient of friction and γ is
a constant. The rods are made of a material with properties α, E, K , k and they each
have cross-sectional area A and length L, which you can assume to be sufficiently
large for perturbations in temperature at the non-contact end to be negligible.
7. An elastic half-plane is pressed against a rigid plane by a pressure p0 and slides at
speed V in the y-direction. Wear occurs, governed by Archard’s wear law ẇ(x, y, t) =
γ f |V | p(x, y, t) where w is the depth of material removed, p is the contact pressure,
f is the coefficient of friction and γ is a constant. Find the critical speed above which
an out-of-plane perturbation of the form
T (x, y, z, t) = Θ(z)ebt cos(mx),
is unstable.
8. Find the function Θ(z) such that the temperature perturbation
T (x, z, t) = Θ(z)ebt cos(mx)
satisfies the transient heat conduction equation
1 ∂T
∇2T = .
k ∂t
Hence find the relation between growth rate b [assumed real] and wavenumber m for
the out-of-plane sliding problem of Sect. 17.5.3 and verify that the maximum growth
rate corresponds to the wavenumber m 0 of Eq. (17.88).
Chapter 18
Rolling and Sliding Contact
So far, we have focussed mainly on contact problems in which the bodies are nom-
inally stationary, except for the generally small motions associated with material
deformation. However, in tribology, we often have to do with bodies in relative
motion. This can be accomplished by sliding or rolling at the interface, or by a
combination of the two.
18.1 Rigid-Body Kinematics
Before considering sliding or rolling of deformable bodies, it is helpful to explore

the most general classes of motion permitted between two rigid bodies if they are
constrained to remain in contact. Only the relative kinematics is of importance to the
contact problem,1 so it is convenient to define a frame of reference which moves so
as to ensure that the instantaneous contact point is always located at the origin and
the common normal between the contacting bodies is always vertical.
It follows that the only non-zero velocities of particles adjacent to the contact point
in the two-dimensional example of Fig. 18.1 are the tangential velocities V1 , V2 . If
V1 = V2 , we can define the sliding velocity
VS = V2 − V1 . (18.1)
If there is a tangential force (e.g. a frictional force) Q transmitted between the

contacting bodies, the power dissipated in friction will be |QVS | and the force must
oppose the sliding motion.
1 We are not concerned with dynamic effects for the moment, so there is no need to [e.g.] restrict
attention to an inertial frame.
https://doi.org/10.1007/978-3-319-70939-0_18
434 18 Rolling and Sliding Contact
Fig. 18.1 Two rigid bodies O2

in relative motion
maintaining contact
Ω2
R2
V2
V1
R1
Ω1
O1
If V1 = V2 ≡ VR = 0, the sliding velocity VS = 0 and we have a state of pure rolling.

The tangential force Q will then transmit power between the bodies at a rate |QVR |,
but there will be no frictional losses.
Since the contact point has no vertical component of velocity in Fig. 18.1, the
instantaneous centres of rotation O1 , O2 of the two bodies must lie on the common
normal, and in fact they must be located at the centre of curvature of the surface
immediately adjacent to the contact point, since if this condition were not satis-
fied, material points just upstream of the contact point would move to locations
above or below the stationary contact point. Notice that the bodies do not have to
be cylinders—only the local region at the contact must be cylindrical [see Fig. 18.2]
and the instantaneous centres O1 , O2 will generally move along the common normal
as sliding and/or rolling proceeds.
Fig. 18.2 Bodies of more O2

general shape Ω2
R2
R1
Ω1
O1
18.1 Rigid-Body Kinematics 435
In these figures, the instantaneous motion can be described by the angular veloc-
ities Ω1 , Ω2 , and with the anticlockwise positive convention shown
V1 = −R1 Ω1 ; V2 = R2 Ω2 . (18.2)
18.1.1 Three-Dimensional Motions
When we consider the more general kinematics of three-dimensional contact, we find

that a further condition is needed to determine the reference frame for rigid-body
motions. We again choose a frame such that the contact point is stationary and the
common normal at the contact point is vertical, but the system still has one degree
of freedom, represented by rigid-body relative rotation about the common normal.
The two bodies can have independent angular velocities about this axis and we
could make an arbitrary choice of rotation for the reference frame to bring one body to
rest or to make the corresponding angular velocities equal and opposite. In practice,
we don’t gain much from either procedure, so we shall generally allow rotations
about this axis to take their ‘real world’ values—i.e. we don’t impose any rotations
on the frame of reference other than those needed to bring the common normal to
rest.
Since the rotations about the common normal are independent, we have the pos-
sibility of a new kind of relative motion at the contact point called spin. In elastic
contact problems, where there is contact over a finite area of the surfaces instead of
at a point, spin will generally be opposed by frictional tractions whose resultant is a
moment about the axis of rotation. However, for a small contact area—as in the case
of two contacting spheres [see Fig. 18.3], the moment arm and hence this resisting
moment will be small.
The kinematics of three-dimensional contact can be very difficult to visualize
and hence it is not always obvious when a spin component is present. This is best
illustrated by a simple example.
Fig. 18.3 Relative spin

(Ω1 −Ω2 ) of two contacting
spheres Ω1
Ω1 =/ Ω2
Ω2
Fig. 18.4 Rolling of a cone

on a cylinder
2.
Ω A N B
r α r α
cos α
O 1.
C
Example: A Cone Rolling on a Cylinder

Figure 18.4 shows the cross-sectional view of a cone in contact with a cylinder. This
can be considered as a simplified model of a railway wheel [the cone] in contact with
a rail [the cylinder]. As the vehicle moves, we suppose that the cone rolls without
slip along the cylinder and hence moves out of the paper in the figure.
Clearly the contact point O moves at a constant speed V = Ωr out of the paper,
but the common normal O N remains in a constant direction without rotation. In order
to bring the normal to rest, we therefore impose an equal and opposite rigid-body
translation into the paper. This is equivalent to considering a cone rotating about a
stationary axis AB and pushing the cylinder into the paper at constant speed due to
a non-slip contact at the fixed point O.
In order to make the common normal vertical, we now look at the contact point
in the direction of the arrow C in Fig. 18.4. The [vector] angular velocity Ω has a
component Ω cos α about an axis parallel to C and hence we see the view of Fig. 18.5.
Notice that the cross-section of the cone is actually elliptical in this view, but the
Fig. 18.5 View in

direction C N
Ω cos α
2.
r
cos α
O 1.
V
18.1 Rigid-Body Kinematics 437
(a)
2. (b)
Ω
A C
α B
r0 r
B
C
A
1.
Fig. 18.6 (a) Rolling of a cone on a rectangular bar, (b) frictional traction distribution in the contact
area
point N is instantaneously at rest and the radius of curvature at the contact point is
r/ cos α.
The two velocities at the contact point are clearly
r
V1 = −V = −Ωr ; V2 = −(Ω cos α) = −Ωr (18.3)
cos α
and hence V1 = V2 and there is no slip as required. However, Ω also has another
component Ω sin α in the plane perpendicular to C, which we would see as a relative
rotation if we were to look along the common normal. This represents a spinning
motion of the cone as it travels along the cylinder.
Cone Rolling on a Rectangular Bar
More insight into this situation can be obtained by considering the modified problem
of Fig. 18.6a, where the cylindrical rail is replaced by a rectangular bar.
Different points on the contact surface of the bar correspond to different values
of r and hence it is not easy to know at what speed the cone will roll in response to
an angular velocity Ω. We define an as yet unknown radius r0 such that
V
r0 = − or V = −Ωr0 . (18.4)
Ω
At a general point defined by the radius r , we therefore have
V1 = −Ωr0 ; V2 = −Ωr (18.5)
and there will be a slip velocity
VS = V2 − V1 = Ω(r0 − r ) (18.6)
which is non-zero except at the point C, where r =r0 . The slip velocity will be
positive in the segment BC and negative in C A and in each region will be resisted by
frictional tractions of the appropriate direction. If we assume that the normal force is
uniformly distributed along the contact line between A and B, the frictional tractions
will have the form shown in Fig. 18.6b. If the cone rolls at constant speed V , the
resultant force in the direction of motion must be zero and hence C must be mid-way
between A and B. The frictional tractions then clearly represent a moment about
the common normal—a typical response to spin motions. Furthermore, this moment
has a component about the axis of rotation and hence a torque must be applied to
sustain motion at constant speed. The work done by this torque is of course equal to
the work done by the frictional tractions in the slip regions.
Suppose now that the vehicle is accelerating or braking, or that there is some
external force such as air resistance, so that the tangential force transmitted through
the contact region is non-zero. The ‘no-slip’ point C will then be displaced from
the mid-point of AB, and hence the ratio r0 between V and Ω will be changed. For
example, during acceleration, the rotational speed Ω will be greater than that required
to sustain the same speed with zero acceleration.2 The point C cannot pass beyond the
segment AB, so there is a limit to the tangential force that can be transmitted, equal
to f P, where P is the total normal force. If this is reached the system transitions to
gross slip. The cone then skids along the bar in the case of braking, or spins with
uncorrelated translational motion in the case of acceleration.
18.2 Johnson’s Belt Drive Problem
Johnson (1985) showed that many of the qualitative features of the effect of material
deformation on rolling contact can be exposed in the simple belt drive of Fig. 18.7.
Two identical rigid pulleys of radius R are connected by an elastic belt whose axial
strain e is given by
T
e= , (18.7)
k
where T is the belt tension and k is a constant. If the system is running at constant
speed, the pulleys must be in equilibrium and hence the driving and driven torques
must both be given by
M = (T1 − T2 )R , (18.8)
where T1 , T2 are the tensions in the upper and lower free segments of belt respectively.
These tensions must differ if power is to be transmitted, so the corresponding axial
strains e1 , e2 in the non-contacting belt segments must also differ.
Suppose we were to paint a mark every inch along the belt in the unstretched
state. During subsequent steady-state operation, the number of marks per unit time
passing any given point must be the same in order to preserve conservation of mass.
However, the marks will be further apart at the top of the belt where the tension is
higher, so the actual belt velocity V there must be higher than at the bottom. In fact,
2 We shall see later in this chapter that the deformation of the rolling components can cause a similar
dependence of rotational to translational kinematics on tractive force.

18.2 Johnson’s Belt Drive Problem 439
Fig. 18.7 Two pulleys V1

connected by an elastic belt T1
and transmitting a torque M
θ
R R M
Ω2 Ω1
M L
θ
driven T2 driving
V2
we can define the ‘unstretched belt velocity’ V0 , in terms of which

T
V = V0 (1 + e) = V0 1 + . (18.9)
k
At the belt-pulley interface, equilibrium considerations show that the compressive

contact force p(θ) per unit circumference is given by
T (θ)
p(θ) = (18.10)
R
and if there is no slip, the strain and hence T and p must be independent of θ. In slip
regions, tangential equilibrium yields the additional condition
1 dT dT
= ± f p or =±fT (18.11)
R dθ dθ
using (18.10), where f is the coefficient of friction and the sign taken depends on
the direction of the frictional traction. Equation (18.11)2 can be integrated to give
the well known capstan formula
T (θ) = Ae± f θ , (18.12)
where A is an arbitrary constant. The frictional traction must oppose the direction of
slip and this condition can be used to show that a slip region is possible only at the
trailing edge of the contact. In other words, the belt is laid down on the pulley in a
state of stick, but a transition to slip occurs at some point around the circumference. It
follows that the angular velocity of the pulleys are determined by the linear velocities
of the corresponding incoming belt segments, and hence that

V1 V0 T1 V2 V0 T2
Ω1 = = 1+ ; Ω2 = = 1+ . (18.13)
R R k R R k
Thus, the driven pulley rotates more slowly than the driving pulley in the ratio
Ω2 V2 k + T2 (T1 − T2 ) M
= = ≈1− =1− , (18.14)
Ω1 V1 k + T1 k Rk
from (18.8), where we have assumed that the belt is ‘stiff’ and hence k T1 , T2 , or
equivalently, the strains e1 , e2 1.
The total belt length must remain constant and equal to 2π R +2L during operation,
so the mean belt tension must remain equal to the assembled value T0 . As long as
the slip arcs are a relatively small proportion of the total length, this yields
T1 + T2 M M
= T0 and hence T1 = T0 + ; T2 = T0 − . (18.15)
2 2R 2R
The tension must transition from T1 to T2 or vice versa in the slip region, so the
extent φ of the slip arc on each pulley is given by

T1 1 T1
= e f φ or φ = ln . (18.16)
T2 f T2
This cannot exceed π, so the maximum transmitted torque is

efπ − 1 πf
Mmax = 2T0 R = 2T0 R tanh . (18.17)
efπ + 1 2
The speed differential between driven and driving pulleys implies that there is a
frictional power loss
M 2 Ω1
W = MΩ1 − MΩ2 = , (18.18)
Rk
using (18.14). An alternative, but much more time-consuming way of getting this
result would be to integrate the product of frictional traction and slip velocity in the
two slip arcs.
It is remarkable that the power loss is independent of the initial tension T0 and the
coefficient of friction f , even though these parameters affect the size of the slip arcs.
It follows that the same loss would be obtained for an arbitrarily large coefficient
of friction, in which case the slip arcs would be vanishingly small. In the limit, we
approach the condition that Popov et al. (2015) describe as ‘relaxation damping’,
where the loss appears as a sudden release of strain energy at the trailing edge of the
contact, though we note that for finite but large friction, the loss is still frictional.3
It might be thought that a closer analogue to an infinite coefficient of friction
would be a chain drive engaging with two gears. If the chain is elastic, there will be
frictional losses as each link is pulled out of engagement, but this must translate into
different torques at the driven and driving gears, since the speed ratio is necessarily
defined by the tooth ratio.
3 See Sect. 7.7.4.

18.3 Tractive Rolling of Elastic Cylinders 441
18.3 Tractive Rolling of Elastic Cylinders
Figure 18.8 shows an elastic cylinder of radius R rolling over a half-plane whilst
transmitting a normal force P and a tangential force Q. This situation will arise
when a vehicle is accelerating or braking.
Carter (1926) gave a solution of the steady-state problem under the assumption
that normal and tangential tractions are uncoupled, so the contact pressure and the
extent of the contact area are given by the Hertzian solution
∗ √ ∗√
π E a2 2P a 2 − x 2 E a2 − x 2
P= ; p(x) = = , (18.19)
4R πa 2 2R
of Eqs. (6.24), (6.25). Carter’s solution is remarkable in that the mathematical
approach prefigures that used by Cattaneo (1938) and Mindlin (1949) for the arguably
simpler static contact problem.4
In the steady state, the stress and displacement fields are invariant in a frame
of reference moving with the cylinder and hence depend on x and t only in the
combination ξ = (x −V t). Now we recall from Sect. 7.1 that the condition for stick
is ḣ = 0, where h =U −u x and U is a rigid-body relative displacement. Thus, in stick
regions, we must have |qx | ≤ f V p and
∂u x du x
U̇ − = 0 or U̇ + V = 0, (18.20)
∂t dξ
where U̇ is a constant representing a creep velocity, analogous to the velocity differ-

ence (V1 −V2 ) in the belt and pulley system of Sect. 18.2.
In the slip regions, the frictional tractions must oppose the slip velocity and hence
du x
U̇ > −V and qx (ξ) = f p(ξ). (18.21)
dξ
Fig. 18.8 Rolling with

P
traction of a cylinder on a Q
plane
R V
Ω
x
O
a a
4 See Sect. 9.1.

It is clear that the Cattaneo distribution

2 f P 2
qx (ξ) = a − ξ 2− b 2 − ξ2 (18.22)
πa 2
with U̇ = 0 satisfies both equality conditions, as in Sect. 9.2, but substitution into
(7.16) [with β = 0] and evaluation of the integral gives

du x 4P ξ 2 − b2
=∓ b < |ξ| < a, (18.23)
dξ πa 2 E ∗
with the negative sign applying in the leading slip zone b < ξ < a, and the positive
sign in the trailing zone −a < ξ < b. This satisfies (18.21)1 in the trailing zone, but
not in leading zone.
The situation is exactly parallel to that in the belt drive problem and has the same
resolution. Indeed, for the belt drive, if we assembled the system with a uniform
tension T0 and then applied torques at the pulleys without allowing the driven pulley
to rotate, we would obtain a quasi-Cattaneo distribution on the driven pulley with
equal slip arcs surrounding a central stick arc.
For the problem of Fig. 18.8, we assume that material is laid down in a state of stick
so that the segment c < ξ < a sticks and the segment −a < ξ < c at the trailing edge
slips. In the spirit of the Cattaneo solution, we write the shear traction distribution as
qx (ξ) = f p(ξ) + q ∗ (ξ) (18.24)
so that (18.21)2 implies that q ∗ (ξ) = 0 in the slip zone −a < ξ < c. We note that

du x 2 a
f p(x)d x a
q ∗ (x)d x
=− ∗ + (18.25)
dξ πE −a (ξ − x) c (ξ − x)
from (7.16), and imposing (18.20) in c < ξ < a, we obtain

∗
a
q ∗ (x)d x π E U̇ a
f p(x)d x
= −
c (ξ − x) 2V −a (ξ − x)
∗ ∗
π E U̇ f πE ξ
= − c < ξ < a, (18.26)
2V 2R
which is a Cauchy singular integral equation for the unknown corrective traction
q ∗ (x). The frictional traction must be bounded at ξ = c, a, so we require the bounded–
bounded solution, and the consistency condition (C.10) determines the value of the
creep rate U̇ .
We have given a formal statement of the solution procedure, but it is clear that
Eq. (18.26) is exactly analogous with the Hertzian problem for a cylinder of radius
R/ f , except that the domain A is changed from (−a, a) to (c, a). It follows imme-
diately that
∗√
∗ fE (a − ξ)(ξ − c)
q (ξ) = − (18.27)
2R
and the complete distribution
f E 2
∗

qx (ξ) = a − ξ 2 − (a − ξ)(ξ − c) (18.28)
2R
has the form illustrated in Fig. 18.9. Notice that the superposition is exactly the same
as that in Cattaneo’s solution except that the corrective term is shifted to make it
adjacent to the leading edge. The tangential force Q is given by
∗

f πE a−c 2 f π p0 a−c 2
Q= a − 2
= a −
2
(18.29)
4R 2 2a 2

2
a−c
= f P 1− , (18.30)
2a
and hence the extent of the slip zone is defined by

c Q
=1−2 1− . (18.31)
a fP
The creep velocity U̇ can be found by substituting (18.27) into (18.26) and eval-
uating the integral, which is identical to the Hertzian integral except that the origin
is moved to the mid-point of the segment c < x < a. We obtain
∗ ∗ ∗
f πE a+c π E U̇ f πE ξ
− ξ− = − (18.32)
2R 2 2V 2R
and hence the creep ratio

U̇ f (a + c) f p0 (a + c) Q
ζ≡ = = = ζ0 1 − 1 − , (18.33)
V 2R E ∗a fP
Fig. 18.9 Distribution of V

shear tractions qx (x) for
Carter’s problem f p(ξ)
qx (ξ)
c
O
a a ξ
slip stick
Fig. 18.10 Dependence of creep ratio ζ on the traction ratio Q/ f P, from Eq. (18.33). The dotted
line represents Kalker’s approximation for the three-dimensional problem, using the strip theory of
Sect. 18.3.5 below
where p0 = p(0) is the maximum contact pressure from (18.19) and ζ0 is the maxi-
mum creep ratio that can occur whilst remaining in partial slip and is given by
fa 2 f p0
ζ0 = = . (18.34)
R E∗
The relation between tangential force and creep rate is shown in Fig. 18.10. Notice
∗
that if the contact is to remain predominantly elastic, we must have p0 E , so the
maximum creep ratio ζ0 1.
Full Stick Solution
If the coefficient of friction f is very large, most of the contact area will be in a state
of stick and the stick-slip boundary will tend to the trailing edge c → −a. In the limit
f → ∞, the tangential traction distribution takes the form

Q a−ξ
qx (ξ) = , (18.35)
πa a+ξ
which is singular at the trailing edge, and the creep rate
Qa Qp0
ζ= = (18.36)
2P R P E∗
increases linearly with Q.
As in the pulley problem, there is finite creep even for an arbitrarily large coeffi-
cient of friction, and the resulting energy loss then mimics that due to the propagation
of a mode II crack [at the trailing edge].
18.3.1 Dissimilar Materials
The preceding solution is restricted to the case where the normal and tangential prob-
lems are uncoupled, and hence Dundurs’ constant β = 0. The more general coupled
case is considerably more challenging and usually involves a more complex arrange-
ment of stick and slip zones. Also the normal traction deviates from the Hertzian
case.
Bentall and Johnson (1967) gave a numerical solution of the problem of ‘free
rolling’, where the tangential force Q = 0 in Fig. 18.8. They used piecewise-linear
representations of the normal and tangential contact tractions p(x), qx (x) and found
that slip zones develop at both contact edges with the same slip direction, but that
there is also an interior zone of slip in the opposite direction.
Nowell and Hills (1988) used the Goodman approximation of Sect. 7.5 to inves-
tigate the tractive rolling (Q = 0) of dissimilar bodies. In other words, they assumed
that the contact pressure distribution is still given by Eq. (18.19), but the tangential
displacement derivative is now given by

du x 2 a
qx (x)d x 2β p(ξ)
=− ∗ − , (18.37)
dξ πE −a (ξ − x) E∗
from (7.51). If slip occurs only in one direction, so that (for example) qx (ξ) = f p(ξ)
in all slip regions, we can define q ∗ (ξ) as in (18.24), and determine it from (18.20)
in the stick region Astick , giving
∗

q ∗ (x)d x πE U̇ fξ β a2 − ξ2
= − − ξ ∈ Astick . (18.38)
Astick (ξ − x) 2 V R R
The arrangement of stick and slip zones depends on the material mismatch and
the tangential force, but many of the cases considered by Nowell and Hills involve
an interior stick zone, with zones of slip adjacent to each edge of the contact area, in
which case the final integral equation has the same form as (C.2) in Appendix C.
Nowell and Hills (1988) found that the Goodman approximation gives results
for the contact tractions close to those obtained with direct numerical simulation,
but there is a discrepancy between the estimates of frictional power loss from the
work done by the frictional tractions and those from the work done by the externally
applied forces, which indeed in certain cases can appear negative.
Munisamy et al. (1991) resolved this paradox by relaxing the Goodman approx-
imation, in which case the normal traction deviates slightly from the Hertzian, and
more specifically, the resultant normal force no longer passes through the centre of
the cylinder. If this resultant is decomposed into an equal central force and a moment,
it is clear that the moment does work during rolling, and this contribution restores
the energy balance.
18.3.2 Antiplane Loading
If the force Q in Fig. 18.8 is directed into the paper instead of to the right, the cylinder
will tend to creep in the antiplane y-direction and tractions q y (ξ) will be developed at
the interface. The analysis proceeds exactly as before, except that qx , u x are replaced
by q y , u y respectively, and Eq. (18.25) takes the form

du y 2 a
f p(x)d x a
q ∗ (x)d x
=− + , (18.39)
dξ πE −a (ξ − x) c (ξ − x)
from (7.18), where the composite antiplane shear modulus E is defined in (7.13).
∗
The traction distribution is then defined by (18.28) [also with E replaced by E ],
and the stick-slip boundary by (18.31). However, U̇ , which now represents an axial
creep velocity of the cylinder, is given by

∗
U̇ f aE Q
= 1− 1− . (18.40)
V R E fP
We recall that antiplane displacements are always uncoupled from in-plane tractions,
so within the limits of the Goodman approximation, these results apply to all material
combinations. For the case of similar materials, (18.40) reduces to

U̇ fa Q
= 1− 1− . (18.41)
V R(1 − ν) fP
18.3.3 Rolling of Misaligned Cylinders
An important application leading to antiplane creep is that in which two nomi-

nally parallel cylinders are actually misaligned by a very small angle φ 1, due to
manufacturing or assembly errors (Engel and Adams 1980). The misalignment
implies an axial creep velocity U̇ = φV and hence an axial force Q per unit length
will be generated if creep is prevented, given by

Q φR(1 − ν) φG
1− 1− = = , (18.42)
fP fa 2 f p0
where p0 is the maximum contact pressure from (18.19) and we have assumed the
materials are similar in the interests of simplicity, with shear modulus G. Since Q
cannot exceed f P, we deduce that the entire contact area will slip if
2 f p0
φ> (18.43)
G
and this defines a very small angle if the normal contact is to remain in the elastic
range.
18.3.4 Three-Dimensional Rolling Contact Problems
If the cone and cylinder in Fig. 18.4 are elastic, the contact area will be elliptical and
given by the Hertzian contact theory. The rigid-body analysis shows that spin occurs at
the contact area, so frictional tractions will be induced even if no tangential forces are
transmitted by the contact. Further complication is introduced if the contact transmits
a longitudinal or lateral tangential force. Problems of this class are a critical ingredient
in vehicle dynamics models, but generally they can only be treated numerically
(Kalker 1990).
Here we shall discuss the simpler case in which a general Hertzian contact trans-
mits only a constant longitudinal tangential force and there is no spin. This would
arise for example if a cylindrical wheel rolls with traction over a cylindrical rail of
the same material. The normal contact problem is defined by the Hertzian theory of
Chap. 3 and in particular, we anticipate an elliptical contact area of semi-axes a, b,
in which the contact pressure is given by

x2 y2
p(x, y) = p0 1 − − . (18.44)
a2 b2
Fig. 18.11 A ‘Carter’ y

superposition for a
three-dimensional rolling
Hertzian contact. The rolling
violation
direction is from left to right
b zone
stick
O x
trailing
edge
slip
leading
edge
We showed in Sect. 9.1 that the tangential traction distribution qx (x, y) = f p(x, y)
will produce quadratic tangential displacements inside the contact ellipse, and that
a ‘Cattaneo’ superposition satisfies the condition of stick inside the smaller ellipse.
The two-dimensional Carter solution of Eq. (18.28) and Fig. 18.9 suggests that we
might use a similar superposition, but with the ‘stick’ ellipse moved to the leading
edge of the contact, as shown in Fig. 18.11.
This satisfies the stick condition inside the smaller ellipse (Johnson 1985,
Sect. 8.4), but the only point on the leading edge of the contact area that is then
in a state of stick is (a, 0). Elsewhere there exists a region of slip between the leading
edge of the contact ellipse and that of the stick ellipse in which the slip displacements
are opposite in direction to those implied by the frictional tractions. However, this
violation region is narrow, particularly if Q/ f P 1, so the resulting solution might
still be regarded as a first approximation for the tangential tractions in the rest of the
contact region.
18.3.5 Kalker’s Strip Theory
If b a, so that the contact ellipse is thin in the rolling direction and elongated in
the transverse direction, we can reasonably approximate the conditions at any given
value of y as being locally two-dimensional (Kalker 1967). For each such ‘strip’, we
can then assume a traction distribution of the form of Fig. 18.9 and use the equations
from the Carter solution, but with

y2 y2
a → a1 (y) = a 1 − 2 ; p0 → p1 (y) = p0 1 − . (18.45)
b b2
Kinematic considerations demand that the creep ratio ζ be the same for each strip,
and hence
f p1 (a1 + c1 ) f p0 (a1 + c1 ) 2 f p0 d
ζ= = = (18.46)
E ∗ a1 E ∗a E ∗a
from (18.33), where x = c1 (y) defines the stick-slip boundary and x = d = (a1 +c1 )/2
defines the mid-point of the stick zone. This equation shows that d is independent
of y and hence the stick-slip boundary is a reflection of the leading edge boundary
about the line x = d as shown in Fig. 18.12 (Johnson 1985, Sect. 8.4).
Full slip will occur throughout the contact area when d = a and hence
d ζ 2 f p0
= where ζ0 = (18.47)
a ζ0 E∗
is the maximum creep rate in partial slip.

Fig. 18.12 Stick and slip y

regions predicted by
Kalker’s strip theory a
c(y)
b
stick
L
O x
L
trailing leading
edge slip edge
For d < a, slip occurs throughout the region |y| > L, where

d2
L =b 1− (18.48)
a2
defines the intersection of the line x = d with the contact ellipse as shown in Fig. 18.12.
The force per unit length Q(y) in the y-direction in |y| < L is given by (18.29) with
the substitutions (18.45), and hence

f π p0 a1 − c1 2
Q(y) = a1 −
2
−L<y<L (18.49)
2a 2
f π p0 a12
= L < |y| < b. (18.50)
2a
Substituting for c1 from (18.46) and integrating Q(y) over the range −b < y < b, we
obtain the total tangential force Q x as

Qx 3ζ ζ ζ2 ζ2
=1+ cos−1 − 1+ 2 1− 2. (18.51)
fP 2ζ0 ζ0 2ζ0 ζ0
Notice that this expression is independent of the eccentricity of the contact ellipse.
It is plotted as a dotted line in Fig. 18.10.
The strip solution is strictly appropriate only when b a, in which case it is rea-
sonable to assume that conditions vary ‘slowly’ in the transverse direction. Some
measure of the approximation involved can be obtained by using the same approx-
imation to predict the contact pressure in the three-dimensional case, for which the
exact solution is given by the Hertzian theory of Chap. 3. A comparison of the two
expressions shows that they agree in the limit b a as we should expect, whereas
the strip theory underestimates the maximum contact pressure p0 for a given contact
radius by about 28% when the contact area is circular (e = 0). Progressively larger
errors are to be anticipated when a > b in Fig. 18.12.
18.3.6 The Incipient Sliding Solution
One case that can be solved exactly, at least within the approximation examined by
Munisamy et al. (1994), is that in which the stick area has shrunk to a point so that
the bodies are at the limit of incipient sliding. For tractive rolling in the x-direction,
the last point to slip will be (a, 0) and the creep ratio in this limiting state [assuming
only one body is deformable] is
∂u x
ζ0 = (a, 0) = 2C1 a, (18.52)
∂x
where u x is given by Eq. (9.2). If both bodies are deformable, we must add their
separate tangential displacements, which will simply increase the creep ratio by a
factor of 2 if the materials are similar. For this case, if the ellipse is elongated in the
rolling direction, so that a > b, we have
3(1 + ν) f P
ζ0 = [(2 − ν)I1 (0, e) + ν I1 (1, e)] (18.53)
2π Ea 2
from (9.3), where the integrals I1 (m, e) are defined in Appendix B. In many cases,
the contribution of the second term in this expression is small, in which case the
tangential displacements are equal to the normal displacements multiplied by the
factor RT of Eq. (7.58) and an acceptable approximation to ζ0 is
(2 − ν) f a
ζ0 ≈ , (18.54)
2(1 − ν)Rx
where Rx is defined in (3.18).
18.3.7 Transient Problems
Kalker (1971a, b) developed a numerical solution for transient rolling contact prob-
lems with friction, and in particular investigated the case where the two cylinders are
first loaded normally, then tangentially with no rolling motion, after which rolling
commences. The initial shear traction distribution is that due to Cattaneo (1938)
described in Sect. 9.1, and the final steady state is that due to Carter (1926) and
described in Sect. 18.3. Kalker found that the steady state is almost completely estab-
lished by the time the rollers have moved through a linear distance 2a—i.e. the extent
of the contact area. We recall from Sect. 8.3.3 that the system memory resides in the
slip displacements at points that are instantaneously stuck, and after t = 2a/V all
initially stuck points will have left the contact area. Thus, any residual memory can
result only from the influence of these initial slip displacements on the corresponding
displacements locked in to contact regions newly established by the rolling motion.
18.3.8 Rail Corrugations
Railway tracks can develop non-uniform wear patterns known as corrugations, which
cause noise and vibration during traversal by a train. This phenomenon is generally
believed to result from an unstable interaction between the dynamics of the rail and
vehicle, connected through the dynamic characteristics of the rolling contact process.
Rail corrugations have been observed for over 100 years and numerous efforts have
been made to explain and therefore hopefully avoid them (Grassie 2009).
If a small amplitude corrugation already exists, we should in principle be able to
model the dynamic process describing the transit of a loaded wheel. If we assume that
the resulting wear volume is proportional to the frictional energy dissipated in the
slip regions (Archard 1953), we can then find the change in the rail profile due to the
transit, and in particular, determine whether the amplitude of the initial corrugation
is increased or decreased (Knothe and Groß-Thebing 2008). The dynamic response
can be expected to depend on the wavelength of the initial corrugation and the train
speed, but if prevailing speeds are relatively similar, we might then hope to identify
a dominant wavelength corresponding to the most rapidly growing perturbation.
We have seen similar wavelength dependence in the phenomenon of thermoelastic
instability in Sect. 17.5.
The most challenging part of this procedure is the analysis of the response of
the contact to periodic variations in both normal and tangential force. As in the
Hertzian impact theory of Sect. 20.1 [below], it is conventional to assume that the
contact interaction can be approximated by a quasi-static analysis, but both normal
and tangential forces will contain small periodic perturbations. If the spatial wave-
length of the corrugation is long compared with the length of the contact area 2a,
Kalker’s results (Kalker 1971a, b) show that Carter’s steady-state solution provides
an acceptable approximation to the instantaneous state in the two-dimensional tran-
sient case. We can then perform a linear perturbation on the steady-state equations
and use the results in a dynamic analysis of the problem. Notice however that in
the two-dimensional case, rigid-body displacements are strictly undefined, so it is
necessary to introduce a further length scale to characterize the normal and tangen-
tial stiffnesses for the contact. In the real problem, the contact area will generally
be elliptical with the major axis perpendicular to the direction of motion, so a rel-
atively straightforward, albeit approximate approach is to use Kalker’s strip theory
[Sect. 18.3.5].
Unfortunately, many observations of corrugations involve wavelengths that are

too short to justify the use of the steady-state solution, and hence a more rigorous
treatment of the contact problem is required. In particular, the slip displacements
locked into the leading stick zone remain in place until the slip boundary is reached,
and hence the position of this boundary oscillates (Afferante et al. 2011a, b). This
introduces a time delay into the system, analogous to that encountered in machine
tool vibration (chatter). Kalker (1990) describes various numerical algorithms for the
general transient problem, including more accurate three-dimensional models that
capture the effect of spin discussed in Sect. 18.1.1, when elastic deformation of the
bodies is included. Groß-Thebing (1989) used Kalker’s numerical code CONTACT
to examine the effect of small perturbations on steady rolling conditions and hence
develop a set of creep coefficients or receptances for use in the dynamic analysis.
18.4 Steady Sliding
In many tribological systems, one elastic body slides over another at constant speed
V , which we here assume to be sufficiently small for elastodynamic effects to be
neglected.
We start by considering the two-dimensional problem of an elastic indenter defined

by an initial gap function g0 (x) sliding to the right over the elastic half-plane z > 0,
as shown in Fig. 18.13.
Notice that in accordance with Sect. 18.1, we move the reference frame so as to
ensure that the contact area appears stationary, which in this case implies that the
half-space moves to the left at speed V . We assume Coulomb’s friction law applies,
so that the tractions comprise a contact pressure p(x) and a proportional tangential
traction q(x) = f p(x), where f is the coefficient of friction. The normal displacement
2.
x
1.
V z
Fig. 18.13 A contact problem involving steady sliding

18.4 Steady Sliding 453
derivative is then
du z 2 p(ξ)dξ 2β f p(x)
=− ∗ + , (18.55)
dx πE A (x − ξ) E∗
from Eq. (7.17), and the contact condition (6.8) demands that
∗
1 p(ξ)dξ E g0 (x)
− f β p(x) = x ∈ A. (18.56)
π A (x − ξ) 2
This is a Cauchy singular integral equation of the second kind [Sect. C.2] with
1
λ = cot(πγ) = − . (18.57)
fβ
The Singular Solution

If the contact region A comprises the single strip b < x < a, the general solution of
Eq. (18.56) is
∗ ∗ a
E λg0 (x) w(x) E λ2 g0 (ξ)dξ
p(x) = + P cos(πγ) −
2(1 + λ2 ) π 2(1 + λ2 ) b w(ξ)(x − ξ)
(18.58)
from (C.14), where
a
P= p(x)d x (18.59)
b
is the total normal force, and
1
w(x) = . (18.60)
(a − x)1/2+γ (x − b)1/2−γ
Equation (18.58) defines the solution for the case where the contact area is prescribed,
which is strictly only possible if the upper body is rigid. In this case, γ < 0 from
Eqs. (7.8), (18.57), and the singularity is stronger than square root at the trailing
edge x = b and weaker than square root at the leading edge x = a. This result was
previously established using asymptotic arguments in Sect. 10.3.1.
The Bounded Solution
If both bodies are smooth and elastic, the solution must satisfy the consistency con-
dition (C.21), which serves to determine a relation between the contact boundaries
a, b. For the Hertzian problem g0 (x) = x/R, the coordinate transformation (C.1) and
(C.21) yield the condition
γ
1
1−s (s + ζ)ds a+b
√ = 0 where ζ = . (18.61)
−1 1+s 1−s 2 a −b
The semi-width of the contact area is (a −b)/2 and its mid-point is at x = (a +b)/2,
so ζ is a dimensionless measure of the offset of the mid-point of the contact area
from x = 0. Evaluating the integral (18.61) and solving for ζ, we obtain ζ = 2γ.
A second equation for a, b is obtained from equation (C.20)2 , which can be
evaluated as ∗
π E (1 − 4γ 2 )(a − b)2
P= , (18.62)
16R
(Hills and Sackfield 1985). Also, the contact pressure distribution can be found from
equation (C.17) as
8P cos(πγ)(a − x)1/2−γ (x − b)1/2+γ

p(x) = . (18.63)
π(a − b)2 (1 − 4γ 2 )
Solutions for bodies of more general profile are given by Sackfield and Hills (1988).
In three-dimensional sliding contact problems, it is conventional to retain the assump-

tion that the frictional tractions oppose the rigid-body relative motion and hence
qx (x, y) = f p(x, y); q y (x, y) = 0, (18.64)
for sliding in the x-direction as in Fig. 18.13. However, in general these tractions will
cause relative tangential displacements in the transverse y-direction, and hence the
assumed tractions do not exactly oppose the local relative motion. We also encoun-
tered this error in the slip region in the Cattaneo solution in Sect. 9.1 and we recall
that Munisamy et al. (1994) showed that its effect is generally extremely small.
A formal solution of the resulting contact problem can be obtained by combining
the harmonic potential function solutions of Sects. A.1, A.2. On the plane z = 0, we
then have
∂2χ ∂2ψ
qx (x, y) = −σzx (x, y, 0) = − (x, y, 0) − (x, y, 0)
∂x∂z ∂ y∂z
∂2χ ∂2ψ
q y (x, y) = −σzy (x, y, 0) = − (x, y, 0) + (x, y, 0) (18.65)
∂ y∂z ∂x∂z
∂2ϕ
p(x, y) = −σzz (x, y, 0) = (x, y, 0),
∂z 2
and the conditions (18.64) will be satisfied if the harmonic functions ϕ, χ, ψ are
related through
18.4 Steady Sliding 455
∂2χ ∂2ψ ∂2ϕ ∂2χ ∂2ψ

− − = f 2; = . (18.66)
∂x∂z ∂ y∂z ∂z ∂ y∂z ∂x∂z
A convenient way to achieve this is to define a new harmonic potential function Φ,

in terms of which
∂Φ ∂Φ ∂Φ
ϕ= ; χ= f ; ψ= f (18.67)
∂z ∂x ∂y
[remember that Φ is harmonic, so ∂ 2 Φ/∂x 2 +∂ 2 Φ/∂ y 2 = −∂ 2 Φ/∂z 2 ]. The normal

surface displacement u z (x, y, 0) is then given by
∂ϕ ∂χ
2Gu z (x, y, 0) = −2(1 − ν) − (1 − 2ν)
∂z ∂z
∂2Φ ∂2Φ
= −2(1 − ν) 2 − (1 − 2ν) f . (18.68)
∂z ∂z∂x
If this formalism is used for each of the bodies in Fig. 18.13, the contact problem
(1.4) is reduced to the search for a harmonic function Φ satisfying
∗
∂2Φ ∂2Φ E
+ f β = (g0 (x, y) − Δ); (x, y) ∈ A (18.69)
∂z 2 ∂z∂x 2
∂3Φ
= 0; (x, y) ∈ / A. (18.70)
∂z 3
However, there are no algebraically simple solutions to this boundary value problem,
except in the uncoupled case β = 0, or that in which the fields are independent of y,
in which case the problem reduces to that treated in Sect. 18.4.1.
18.5 Wear
If sliding persists, we anticipate that material will be removed by wear and the
resulting small changes in surface profile can have a significant influence on the
contact pressure distribution.
18.5.1 Archard’s Wear Law
It seems reasonable to assume that the detachment of a wear particle is an irreversible

process that requires a certain amount of energy [for example, work done in plastic
deformation leading eventually to exhaustion of ductility], and hence that the volume
of material removed will be proportional to the work done against frictional tractions
(Archard 1953). In particular, if the depth of material removed is defined as w(x, y),
ẇ(x, y) = α f |V | p(x, y) or w(x, y) = α f L p(x, y), (18.71)
where V is the sliding speed, L is the total sliding distance and α is a wear coefficient.
Archard (1953) reached this conclusion by analyzing a contact involving a distrib-
ution of individual asperity contacts, with a small but finite probability of a wear
particle being detached in any one interaction, or equivalently, that a certain (large)
number of interactions are needed to weaken the asperity to the point of detachment.
However, other micromechanical wear theories lead to the same equation, as we saw
for example in Sect. 16.1.1 and Eq. (16.9).
18.5.2 Long-Time Solution
Equation (18.71) shows that wear will be greatest in regions of high contact pressure,
and this will tend to reduce such non-uniformity over time. If the relative normal
displacement of the bodies is prescribed, the total contact force and the total wear
rate will tend asymptotically to zero over time. However, if a constant normal force is
prescribed, wear must continue indefinitely, but we might expect to reach a limiting
state where the contact pressure distribution is independent of time.
Figure 18.14a shows an elastic brake block pressed by a normal force P against
a rigid plane surface moving at speed V . Frictional tractions will also induce a
reaction Q = f P as shown. Figure 18.14b shows a possible configuration after some
wear w(x) has occurred. Notice that the block has two rigid-body degrees of freedom,
corresponding to translation towards the sliding surface and rotation about the loading
pin.
If the contact pressure distribution p(x) is independent of time, the elastic defor-
mation of the block must also be independent of time, so the wear rate ẇ(x) must
be compatible with a rigid-body motion. In other words we must have
ẇ(x) = A + Bx and hence p(x) = C + Dx (18.72)
(a) (b)
P
Q
b
a a
w(x)
V x x
Fig. 18.14 A brake block sliding on a plane surface: (a) initial condition, (b) after some wear has
occurred
18.5 Wear 457
from (18.71), where A, B, C, D are constants.

Equilibrium considerations require that
a a
p(x)d x = P; x p(x)d x = Qb = f Pb, (18.73)
−a −a
where the dimension b is defined in Fig. 18.14a. This provides two equations for the
constants C, D, from which

P 3 f bx
p(x) = 1+ , (18.74)
2a a2
after which the wear rate can be obtained from (18.71). Of course, the dimension b
decreases as wear proceeds, but if the elastic deformation is relatively modest, we
might anticipate that the contact pressure will have the form (18.74) [with b = b(t)]
once an initial ‘running-in’ period is completed.
More generally, the procedure used in this example shows that the long-time
pressure distribution must be such as to produce an incremental wear state that is
compatible with the rigid-body relative motion of the sliding bodies.
18.5.3 Transient Problems
The initial contact pressure distribution will be the solution of an elastic contact
problem and will generally differ from the steady state. To investigate the transient
process from this initial state to the steady state, we first consider a simple example
in which an elastic half-space with a slightly sinusoidal profile defined [at time t = 0]
by
g0 (x, 0) = u 0 cos(mx) (18.75)
slides at speed V against a rigid plane. We suppose that the bodies are pressed together
by a mean pressure p0 which is sufficient to ensure full contact.
In view of Eq. (18.71), we assume tentatively that the wear w(x, t) has the sinu-
soidal form
w(x, t) = w0 (t) + w1 (t) cos(mx), (18.76)
so that the original gap function g0 (x, 0) is modified with time to
g0 (x, t) = u 0 cos(mx) + w1 (t) cos(mx). (18.77)
Notice that there is no need to include the uniform term w0 (t) in this equation, since
it represents merely a rigid-body translation in response to uniform wear.
If Dundurs’ constant β = 0, so that there is no coupling between frictional tractions
and normal contact, we can then deduce from Sect. 6.5.1 that the contact pressure
distribution will be
∗
E m
p(x, t) = p0 − [u 0 + w1 (t)] cos(mx), (18.78)
2
and substitution into (18.71) shows that the functions w0 , w1 must satisfy the equa-
tions ∗
dw0 dw1 α f |V |E mw1 (t)
= α f |V | p0 ; =− . (18.79)
dt dt 2
The solution of these equations satisfying w(x, 0) = 0 is

w(x, t) = α f |V | p0 t − u 0 1 − e−λt cos(mx), (18.80)
where ∗
α f |V |E m
λ= . (18.81)
2
Thus, the initial sinusoidal profile is eventually worn away, and the time scale of
this transient process is inverse with the wavenumber m. In other words, short wave
perturbations (large m) are worn away more quickly.
Effect of Normal-Tangential Coupling
If β = 0, the frictional tractions qx (x) = f p(x) will produce relative normal displace-
ments which modify the contact pressure distribution p(x). This coupling causes the
sinusoidal perturbation to migrate in the x-direction as well as decaying.
To investigate this effect, we first use Eq. (7.17) and the results of Sect. 6.5.1 to
find that the traction distribution
p(x) = {exp(ımx)} ; q(x) = f {exp(ımx)} (18.82)
produces surface displacements
2
u z (x) = {(1 − ı f β) exp(ımx)} . (18.83)
E ∗m
Notice that the traction distribution (18.82) implies that the lower body 1 is sliding
to the left relative to the upper body. For the opposite direction of sliding, f should
be replaced by − f in the following analysis.
We next generalize (18.76) by writing
w(x, t) = w0 (t) + {w1 (t) exp(ımx)} (18.84)
and arguments exactly similar to those in the uncoupled case then yield the differential
equation
dw1
+ Λw1 = −Λu 0 (18.85)
dt
18.5 Wear 459
with solution
w1 (t) = −u 0 1 − exp(−Λt) , (18.86)
where
λ
Λ = λβ (1 + ı f β); λβ = , (18.87)
(1 + ( f β)2 )
and λ is defined in (18.81). Using this result, Eq. (18.84) can be written
w(x, t) = α f |V | p0 t − u 0 exp(−λβ t) cos(m(x − ct)), (18.88)
where ∗
f βλβ α f |V |E f β
c= = (18.89)
m 2(1 + ( f β)2 )
is the migration speed. Notice that the direction of migration depends on the sign of
the coupling constant β. Also, the approach to the steady state is slightly slower than
in the uncoupled case, since λβ < λ.
18.5.4 Galin’s Eigenfunction Method
Figure 18.15 shows a rigid initially flat punch of width 2a that is pressed against an
elastic half-plane by a force P. The half-space moves at speed V causing wear on the
punch. We shall assume that the wear of the half-space that occurs during a single
pass under the punch is small enough to be neglected.
Suppose that there is no coupling (β = 0), in which case the initial contact pressure
distribution [before wear occurs] is given by Eq. (6.14) as
P
p(x) = √ . (18.90)
π a2 − x 2
Fig. 18.15 A rigid flat P

punch sliding against a
half-plane
a a
V x
However, wear will occur preferentially near the edges x = ±a until after a sufficiently
long time, the contact pressure is uniform and equal to P/2a. The shape of the punch
in the steady state must then be such as to conform with the displacement of the half-
plane under uniform normal and frictional forces, and was calculated by Dundurs
and Comninou (1980).
Galin (1976), introduced a method in which the transient solution is constructed
as an eigenfunction series. Guided by the elementary solutions of Sect. 18.5.3, we
first consider the possibility of a solution in which the pressure distribution has the
form
p(x, t) = p̃(x)e−λt and hence ẇ(x, t) = α f |V | p̃(x)e−λt , (18.91)
from (18.71). The normal displacement u z (x) is then defined by Eq. (6.7) as

∂u z 2e−λt a
p̃(ξ)dξ
=− . (18.92)
∂x πE∗ −a (x − ξ)
Integrating with respect to x, we obtain

2e−λt a x − ξ
u z (x) = −
ln p̃(ξ)dξ + C(t), (18.93)
πE∗ −a a
where C(t) represents a time-dependent rigid-body displacement which we omit

because the punch is free to move vertically, and at this stage we seek only a particular
solution.
If the punch is to remain in contact throughout −a < x < a, we must have
∂u z
ẇ(x) + =0 − a < x < a, t > 0, (18.94)
∂t
and hence

2λ a x − ξ
α f |V | p̃(x) +
ln p̃(ξ)dξ = 0 − a < x < a, (18.95)
πE∗ −a a
which is a homogeneous Fredholm integral equation of the second kind (Tricomi

1970) for the function p̃(x). Non-trivial solutions exist only for certain eigenvalues
λi , and when these have been determined, a general solution can be constructed
as the superposition of the steady-state solution p(x) = P/2a and an eigenfunction
series (Galin 1976, 2008), the coefficients of which are determined from the initial
condition (18.90). Galin and Goryacheva (1977) applied the same technique to the
problem of an axisymmetric rigid flat punch sliding against an elastic half-space.
There are clear mathematical parallels between Galin’s procedure and Burton’s
solution of the thermoelastic instability (TEI) problem, discussed in Sect. 17.5.2.
However, in the case of wear, it can be shown that the eigenvalues λi all correspond
18.5 Wear 461
to decaying perturbations (Galin 2008). As with TEI, more complex geometries can
be discretized by the finite element method, leading to a linear eigenvalue problem
(Liu et al. 2014).
18.5.5 Non-conformal Contact Problems
These techniques are not easily extended to non-conformal contact, since the varia-
tion of the contact area renders the resulting integral equation nonlinear. However,
some indication of the effect of wear in such cases can be obtained by considering the
simpler case where the elastic deformation can be approximated by a linear Winkler
foundation defined such that
p(x, y, t)
u z (x, y, t) = , (18.96)
k
where k is the foundation modulus (Menga and Ciavarella 2015). The contact con-
dition can then be written
p(x, y, t)
g(x, y, t) = g0 (x, y) + + w(x, y, t) − Δ(t) = 0, (18.97)
k
where g0 (x, y) is the initial gap and Δ(t) is the rigid-body approach. Since the contact
is non-conformal, both pressure and wear must be zero at a point that is just coming
into contact, so the instantaneous contact area A(t) is defined by the condition
g0 (x, y) < Δ(t) (x, y) ∈ A(t). (18.98)
Differentiating (18.97) with respect to time and then substituting for ẇ(x, y, t) from
the Archard wear law (18.71), we obtain
ṗ(x, y, t) dΔ
+ α f |V | p(x, y, t) = . (18.99)
k dt
Integration over the instantaneous contact area A then yields
Ṗ dΔ
+ α f |V |P = A . (18.100)
k dt
Since the total force P is generally a known function of time and A is a known
function of Δ from (18.98), Eq. (18.100) can be solved for Δ(t). For example, if
the indenter is a cylinder of radius R, we have g0 (x) = x 2 /2R and the√condition
(18.98) shows that contact will occur in −a(t) < x < a(t),√where a(t) = 2 2RΔ(t).
The contact area per unit length is therefore A = 2a = 2 2RΔ and if the force per
unit length P is constant, the solution of (18.100) is obtained as
3α f |V |Pt
Δ3/2 = √ + C, (18.101)
4 2R
where C is a constant of integration which is determined from the condition at time

t = 0 where no wear has yet occurred. For the cylindrical indentor, it is easily verified
that the initial indentation corresponding to the force P is
2/3
a(0)2 1 3P R
Δ(0) = = (18.102)
2R 2R 2k
and hence 2/3

1 3P R
Δ(t) = (1 + α f |V |kt) . (18.103)
2R 2k
Once Δ(t) is known, Eq. (18.99) can be solved to determine the evolution of the
contact pressure distribution. Notice that (18.99) is a first order ordinary differential
equation for p(x, y, t), since x, y appear only as parameters. The initial condition for
points inside the initial contact area (x, y) ∈ A0 is obtained by setting w(x, y, 0) = 0
in (18.97) and solving for p(x, y, 0). For points outside this range (x, y) ∈ / A0 , the
initial condition is p(x, y, t0 ) = 0, w(x, y, t0 ) = 0, where the time t0 at which contact
is first established is determined from the condition g0 (x, y) = Δ(t0 ) for each point
(x, y).
18.6 Sliding of Rough Surfaces
For the static contact of two rough surfaces, it is usual to ‘transfer’ all the roughness
to one body by defining a composite gap function g0 (x, y), as in Eq. (1.1). However,
during sliding, at least one body must be moving, so that its gap function is defined in
a moving coordinate system. For example, if body 1 is stationary and body 2 moves
at speed V in the positive x-direction, we have
g0 (x, y, t) = g1 (x, y) + g2 (x − V t, y), (18.104)
and this clearly varies with time t if body 2 is rough.

This fact is often overlooked in the literature involving sliding rough surfaces,
but it has critical consequences for the resulting contact processes. For example, if
we represent the roughness of each body by appropriate asperity models, the typical
contact event will comprise an asperity in body 1 making a transient contact with an
asperity in body 2, during which both normal and frictional forces rise to a maximum
value and then fall back to zero. These events will be governed by the statistics of the
asperity distributions and will on average lead to random variations in the contact
forces and/or the rigid-body separation of the bodies. We can therefore expect the
process to be noisy, as in fact it generally is in dry sliding.
18.6 Sliding of Rough Surfaces 463
By contrast, if we model the process by first transferring all the roughness to

one body, the contact forces will then be predicted to remain constant during sliding
and the typical contact event will comprise an asperity in the ‘rough’ body making
continuous contact with a plane surface. This is of course correct if one surface really
is plane, but it also may be a reasonable approximation if one body has a relatively
low yield stress, so that the asperities on the harder surface continuously plough
through the softer body, as discussed in Sect. 16.1.1. In this context, it is important to
recognize that the motion must be defined relative to the microscopic actual contact
areas, rather than to the macroscopic sliding components. For example, if a relatively
soft pin slides against a rough hard rotating disc, the asperities on the disc will plough
through the surface of the pin, creating actual contact areas that are stationary with
respect to the disc, even though on the macroscopic scale, it is the disc which appears
to move.
18.6.1 Flash Temperatures
The frictional heat generated at areas of actual contact can give rise to very large
local temperatures (Guha and Chowdhuri 1996). If the asperity interactions are tran-
sient, these will have very short duration, and they are therefore known as ‘flash
temperatures’. In some cases, the temperatures reached may be sufficient to cause
local melting or softening on the microscale, and this generally imposes a limit on
the maximum achievable flash temperature.
Continuous Contact
Blok (1937) and Jaeger (1942) considered the steady-state problem in which two
half-spaces make contact at a single area that is stationary with respect to one of the
two sliding bodies. Before discussing their solutions, we first consider the problem of
determining the surface temperature due to a known distribution of heat flux q(x, y),
moving over the surface of the half-space z > 0.
The steady-state surface temperature due to a heat source Q moving at speed V
in the positive x-direction over the surface of the half-space is given by Eq. (17.25)
as
Q V (r − x)
T (x, y, 0) = exp − , (18.105)
2πK r 2k
where K , k are respectively the thermal conductivity and diffusivity of the material.
Thisresult is expressed in a coordinate system that moves with the source, so that
r = x 2 + y 2 is the distance from the moving source. In effect, the material of the
half-space moves at speed V through this reference frame in the negative x-direction.
Equation (18.105) implies that the temperature ‘at infinity’ (i.e. far from the source)
is zero, but other cases can be included by adding a constant uniform temperature,
which involves no additional heat flow.
If a heat flux distribution q(x, y) moves over the surface of the body, the corre-
sponding surface temperature can be evaluated by convolution as

1 V (r − x + ξ) q(ξ, η)dξdη
T (x, y, 0) = exp − , (18.106)
2πK A 2k r
where A is the heated area and

r= (x − ξ)2 + (y − η 2 ). (18.107)
Alternatively, using the field-point integration method of Sect. 2.3.1, we can write
π
1 S2
V r (1 − cos θ)
T (x, y, 0) = exp − q(r, θ)dr dθ, (18.108)
2πK 0 S1 2k
where S1 , S2 are defined in Fig. 2.4. This equation can also be written in the dimen-
sionless form
π S2
a
T (x, y, 0) = e−Pe ρ(1−cos θ) q(ρ, θ)dρdθ, (18.109)
2πK 0 S1
where a is a characteristic dimension of the heated area, Pe = V a/2k is the Peclet

number, and ρ =r/a. As in Sects. 17.5.4 and 17.5.6, the Peclet number plays a critical
rôle in determining the nature of the resulting temperature field.
Limiting Solutions at Large and Small Peclet Number
If Pe 1, heat conduction is largely restricted to the z-direction, and the surface
temperature at a point on the surface fixed in the moving body is approximately

1 k t
q(t − τ )dτ
T = (18.110)
K π 0 τ 1/2
(Carslaw and Jaeger 1959, Sect. 2.9), where q(t) is the heat flux at that particular
material point as a function of time t.
If the heat flux in the contact area A is uniform, then a material point in A
experiences this uniform flux for a period t = /V , where is the distance in the
sliding direction from the leading edge of A, as shown in Fig. 18.16. The integral in
(18.110) can then be evaluated to give

2q kt 2q k
T = = . (18.111)
K π K πV
Fig. 18.16 Definition of in

Eq. (18.111)
leading
edge
max
At the opposite extreme, where Pe 1, the exponential term in Eq. (18.106) can
be approximated as unity and Eqs. (18.106), (18.108) reduce to
π
1 q(ξ, η)dξdη 1 S2
T (x, y, 0) = = q(r, θ)dr dθ, (18.112)
2πK A r 2πK 0 S1
which are of the same form as Eqs. (2.17), (2.21). This is equivalent to neglecting
the motion of the heat source.
Circular Heated Area
For a circular heated area of radius a with uniform heat flux q, the maximum value
of is 2a, so the maximum temperature at large Peclet number is

2q 2ka qa 2
Tmax = = √ . (18.113)
K πV K π Pe
Similarly, the average temperature is
a 2√a 2 −y 2
2q k Cqa
Tavg = d dy = √ , (18.114)
πa 2 K −a 0 πV K Pe
where √ 2
16 2 3
C= 2
Γ ≈ 0.689. (18.115)
5π 4
This is approximately 61% of the maximum temperature given by Eq. (18.113).

For Pe 1, the integrals (18.112) yield
qa 8qa
Tmax = ; Tavg = . (18.116)
K 3πK
Approximate Solutions for Intermediate Peclet Numbers

For intermediate values of Peclet number, the integral (18.106) can only be performed
numerically, but Tian and Kennedy (1994) have shown that extremely good curve
fits to the numerical data over the entire range can be constructed in the form
1
T (Pe) = √ , (18.117)
A + B Pe
where the constants A, B are chosen to give the correct limiting solutions as Pe → 0
and Pe → ∞. For example, for uniform heating over the circle, we obtain
2qa qa
Tmax = √ ; Tavg = √ . (18.118)
K 4 + π Pe K 1.388 + 2.11 Pe
Results for different distribution functions q(x, y) and for square and elliptical
heated areas are given by Tian and Kennedy (1994).
Contact of Two Sliding Bodies
Suppose that the contact area is stationary with respect to body 1 and moves at speed
V with respect to body 2. Suppose also that far from the contact area, the two bodies
have equal temperatures, which can therefore be taken as zero.
We assume that the temperature is continuous across the interface inside the
contact area A, so the heat conduction problem comprises the determination of the
two functions q1 , q2 such that
T1 (x, y) = T2 (x, y); q1 (x, y) + q2 (x, y) = f V p(x, y) (x, y) ∈ A. (18.119)
The second equation in (18.119) simply states that the sum of the local heat flux
into the two bodies must be equal to the rate of heat generation by friction. In
addition, T2 is given by (18.108) with {q, K , k} = {q2 , K 2 , k2 } and T1 by (18.112)
with {q, K } = {q1 , K 1 }, since the contact area is stationary with respect to body 1.
No simple analytical solutions exist to this problem, but Blok (1937) and Jaeger
(1942) obtained approximate solutions by assuming that the functions q1 , q2 are
uniform (independent of x, y), and matching only the maximum temperature T1max =
avg avg
T2max or average temperature T1 = T2 in the contact area. A related approximation
was developed by Archard (1959), based on Holm’s solution for the heat flux required
to generate a uniform temperature inside a circular contact area (Holm 1958).
To illustrate this procedure, suppose that the contact area is a circle of radius a.
Using the approximation (18.118)1 for body 2 and (18.116)1 for body 1, we have
q1 a 2q2 a
T1max = ; T2max = √ , (18.120)
K1 K 2 4 + π Pe2
Fig. 18.17 Maximum flash temperature (solid line) as a function of Peclet number V a/2k for
the continuous sliding of similar materials. The dashed line represents a numerical solution of the
maximum flash temperature in a transient asperity interaction, from Lee et al. (2016) [see Sect. 18.6.3
below]. Notice that K /k = ρc p , where ρ, c p are respectively the density and specific heat of the
material
where Pe2 = V a/2k2 . These must be equal, so we conclude that the frictional heat
partition ratio √
q2 K 2 4 + π Pe2
λ≡ = . (18.121)
q1 2K 1
At low sliding speeds, this reduces to the ratio of conductivities K 2 /K 1 , but at high
speeds λ increases with Pe1/2 and most of the frictional heat flows into the moving
body. This occurs because the moving body is continually presenting new unheated
surface to the contact area, whereas the stationary body achieves a steady thermal
state.
If the average contact pressure is p, corresponding to a total force P = πa 2 p in
the actual contact area, we have
f Vp
q1 + q2 = q1 (1 + λ) = f V p and hence q1 = . (18.122)
(1 + λ)
Substituting this result into Eq. (18.120), we obtain the flash temperature as
f V pa 4 f pk2 Pe2
Tmax = = √ . (18.123)
K 1 (1 + λ) (2K 1 + K 2 4 + π Pe2 )
Figure 18.17 shows this relationship as a function of Pe = V a/2k for the case
where the two materials are similar. Notice how the curve modulates between linear
[slope unity] and square-root [slope 1/2] behaviour with increasing Peclet number.
18.6.2 Bulk Temperatures
The preceding calculations are based on the assumption that the two bodies have
equal temperatures ‘at infinity’—i.e. at points whose distance from the contact area
is large compared with a. We shall refer to these as bulk temperatures and denote
them by T1∞ , T2∞ . At high Peclet number, most of the frictional heat is predicted
to flow into the moving body, and this can be expected to cause a bulk temperature
differential to develop which will modify the heat partition ratio λ. Ling and Simkins
(1963) observed such temperature ‘jumps’ experimentally in sliding systems with
frictional heating.
Blok and Jaeger’s method can be extended to this case by simply adding uniform
bulk temperatures into the temperature fields (18.106), (18.112). For the circular
contact area, Eq. (18.120) would then be modified to
q1 a 2q2 a
T1max = + T1∞ ; T2max = √ + T2∞ , (18.124)
K1 K 2 4 + π Pe2
and the modified heat partition ratio can then be obtained as before by equating these
expressions.
An alternative and arguably simpler approach is to regard the heat conduction
problem as the superposition of two sub-problems:-
(i) The bulk temperatures are set to zero and there is frictional heating f V p per
unit area. This is the problem solved [approximately] in Sect. 18.6.1.
(ii) There is no frictional heating, so q1 +q2 = 0, but the correct bulk temperatures
T1∞ , T2∞ are imposed, so the condition of continuity of temperature in the contact
area causes heat to flow between the bodies.
For sub-problem (ii) with a circular contact area, Eq. (18.118) define the steady-
state thermal resistance from the contact area to infinity in each body, and when a bulk
temperature difference T2∞ −T1∞ is imposed, these resistances are in series. They are
best approximated using the average rather than the maximum temperatures, so we
write
− T1∞ = q1 R1 ; − T2∞ = q2 R2 ,
avg avg
T1 T2 (18.125)
where
8a a
R1 = ; R2 = √ , (18.126)
3πK 1 K2 1.388 + 2.11 Pe2
avg avg
from (18.116)2 , (18.118)2 respectively. The conditions T1 = T2 and q1 +q2 = 0
then yield the solution
T2∞ − T1∞ avg avg R1 T2∞ + R2 T1∞

q1 = −q2 = ; T1 = T2 = . (18.127)
R1 + R2 R1 + R2
Notice that the temperature of the contact area is a weighted average of the two bulk
temperatures, and this should be added to the flash temperature predicted in sub-
problem (i). However, usually the bulk temperatures are substantially lower than the
flash temperature, so this is a relatively small correction. More significant is the fact
that the superposed heat flux changes the partition of heat predicted in sub-problem
(i).
18.6.3 Transient Asperity Interactions
All the results in Sects. 18.6.1, 18.6.2 are based on the assumption that actual contact
areas persist and are stationary with respect to one of the two sliding bodies. If the
typical contact event is a transient interaction between asperities on the opposing
surfaces, the results for both flash temperature and heat transfer will be close to
those in the steady sliding case, only if the transient process is of sufficient duration
for a steady state to be established.
This condition is satisfied in the limit Pe 1. However, at the opposite extreme
Pe 1, the transient problem is considerably more challenging, since the relative
motion will generally involve points on the surface of one body making first contact
with points on the other that have already experienced some period of contact. An
order of magnitude approximation can be obtained by neglecting this effect and hence
analyzing a process in which the two asperities are pressed into contact for a period
equal to the transit time, during which the appropriate frictional heat is generated
(Holm 1958; Liu and Barber 2014). However, this will generally overestimate the
flash temperature and underestimate the heat conduction between the two bodies due
to a bulk temperature difference.
Smith and Arnell (2013) used a finite element model to estimate the resulting
flash temperature under the assumption that the contact area is the intersection of
two circles in relative motion. A similar approach was used by Lee et al. (2016),
except that the asperities were assumed to be spherical and the interactions elastic,
so that the contact area and pressure distribution are defined by the time-varying
Hertzian solution. In both cases, the flash temperature follows a pattern similar to
that in Fig. 18.16 above, with proportionality with Pe for Pe 1 and with Pe1/2 for
Pe 1. Lee et al. (2016) show that an excellent curve fit to their results is defined by
the equation
f p0 π Pe
Tmax = √ , (18.128)
ρc p 2 2 + Pe
where ρ, c p are respectively the density and specific heat of the material, p0 is
the maximum Hertzian pressure and the Peclet number5 is defined based on the
maximum contact radius achieved during the interaction. This expression is plotted
5 Leeet al. (2016) define the Peclet number as V a/k. Here we have modified their expression to
accord with the definition used in this chapter.
as the dashed line in Fig. 18.16, based on the average contact pressure in the Hertzian
solution p = 2 p0 /3. The two curves are almost identical, despite the very different
kinematics involved.
By contrast, the heat partition ratio λ is very dependent on the kinematics, since
the transient interaction is symmetrical with respect to the two bodies and hence
[for example] with similar materials we obtain λ = 1 for all sliding speeds. Lee et al.
(2016) also give the expression
√
π 2 + 1.8 Pe
Q = ρc p a 3
(T2∞ − T1∞ ) (18.129)
Pe
for the total heat exchange between the sliding bodies through a single asperity
interaction with maximum contact radius a. This result can be used, in combination
with an asperity model theory of rough surface contact, to predict the effective thermal
contact resistance between two sliding bodies (Liu and Barber 2014).
Problems
1. Figure 18.18 shows the cross-section of a spherical roller bearing. The inner sur-
face of the outer race has a spherical profile so that the whole inner assembly—inner
race + rollers—can rotate about the bearing centre in the plane of the paper, per-
mitting the bearing to act as a ‘pin joint’. Spherical bearings are therefore tolerant
Fig. 18.18 Self-aligning

bearing
outer
inner race
R
race
shaft
x
roller a
c
Problems 471
of misalignment and are used in situations where manufacturing tolerances or shaft

and housing deflections make it difficult to maintain exact alignment.
The spherical inner surface demands that the rollers be barrel-shaped as shown
in the figure and there is therefore always slip at some parts of the contact area.
The problem is to determine the extent of this slip and hence estimate the rolling
resistance of the bearing due to frictional dissipation.
(i) Choose a frame of reference in which the rollers rotate about fixed axes at a
given speed Ω, in which case the outer race rotates in the same direction at Ω1
and the inner race rotates in the opposite direction at Ω2 . Taking Ω as given,
determine the slip speed v S as a function of the distance x from the centre of
the roller.
(ii) Assume that the radial force N carried by one roller is uniformly distributed
over the roller width c, in which case there will be a tangential frictional force
f N /c per unit length into or out of the paper, depending on the sign of v S ,
where f is the coefficient of friction. Determine Ω1 from the condition that the
resultant frictional force at the contact should be zero.
(iii) Repeat steps 1, 2 for the inner race speed Ω2 .
(iv) The rate of energy dissipation at the roller/outer race interface is ( f N /c)(v S )
per unit length. Calculate the total dissipation E 1 at this interface and the cor-
responding dissipation E 2 at the inner race/roller interface.
(v) Finally, estimate the effective coefficient of friction for the bearing, defined as
(E 1 + E 2 )
f∗ = .
(Ω1 + Ω2 )R N
This is the coefficient of friction in a plane journal bearing of radius R which

would lead to the same dissipation of energy for the same relative rotational
speed between the shaft and the housing.
2. Repeat the analysis of Sect. 18.2 for the case where an elastic belt is used as a
speed changing mechanism between two rigid pulleys of different radii, R1 , R2 as
shown in Fig. 18.19. In particular, determine:-
(i) the angular velocity Ω2 of the driven pulley, if the angular velocity Ω1 of the
driving pulley and the input torque M1 are given.
(ii) what percentage of the input power M1 Ω1 is lost due to frictional slip.
Which of the following parameters influence the energy loss:-
• the initial belt tension, T0 ,
• the stiffness of the belt, k,
• the coefficient of friction, f ,
• the distance between pulley centres, L.
How should we choose these parameters to minimize the power loss?
M1 R1
R2 M2
Ω2
L Ω1
driven driving
Fig. 18.19 Belt drive involving pulleys of different radii
3. Equation (18.53) for the creep ratio at incipient sliding applies only for the case
where the ellipse is elongated in the rolling direction. Use the change of coordinates
x ↔ y, a ↔ b and Eqs. (9.2), (9.3) to derive a corresponding expression for the case
where b > a. Plot the ratio between the resulting expressions and that from Kalker’s
strip theory (18.47) over the entire range 0 < a/b < ∞ and show that this ratio tends
to unity as a/b → 0.
4. Estimate the maximum percentage error involved in dropping the second term in
(18.53) and hence using (18.54) for the maximum creep ratio.
5. A cone of angle α = 5 deg rolls at constant speed on a rail as shown in Fig. 3.7. A
Hertzian analysis of the normal contact problem shows that the maximum contact
pressure is p0 and the contact area is an ellipse of semi-axes a = , b = 5, where
b is transverse to the rolling direction, as in Fig. 18.11. At the centre of the contact
ellipse, the effective rolling radius [r in Fig. 18.4] is R.
Use Kalker’s strip theory and kinematic arguments from Sect. 18.1.1 to estimate
the shear traction distribution in the contact area and the energy dissipated in frictional
microslip, if the coefficient of friction is f and both bodies have elastic properties
E, ν. Remember that if there is no acceleration, the tangential force Q = 0.
Fig. 18.20 Brake block with

an off-centre pivot P
d b
a a
V x
Problems 473
V L
Fig. 18.21 Wear of a saw-tooth indenter
6. The non-uniform wear of the brake block of Fig. 18.14a can be reduced by moving
the pivot point to the left as shown in Fig. 18.20. Determine the optimal distance it
should be moved, if the unworn thickness corresponds to b = b0 and the fully worn
state with uniform thickness of wear for all x corresponds to b = b1 (0 < b1 < b0 ).
Assume that the contact pressure can be approximated in the form p(x) = C(t)+
D(t)x at all times t.
7. An indenter with the periodic saw-tooth profile shown in Fig. 18.21 slides against
an elastic half-plane at speed V . Expand the initial gap function g0 (x) as a Fourier
series and use results from Sect. 18.5.3 to describe the resulting evolution of the
profile due to wear, assuming that the applied pressure p̄ is sufficient to ensure full
contact at all times and that β = 0.
Show that the full contact assumption is actually unrealistic here because the
required mean pressure is infinite. What do you think will really happen? Can you
think of any way to estimate the time at which the last points in the profile will first
make contact?
8. Use the Winkler approximation of Sect. 18.5.5 to estimate the evolution of the
pressure distribution p(r, t) for a sphere of radius R sliding against a plane at speed
V and loaded by a constant normal force P. Show that the distribution becomes
approximately uniform at large times, and characterize the approach to this state by
an appropriate dimensionless time parameter.
9. In a particular sliding system, the distant boundaries of body 2 are cooled to
maintain a bulk temperature T2∞ = 0, but those of body 1 are insulated, so no heat
can flow out of the body. The two materials have the same thermal properties. If
there is a single circular contact area of radius a which moves relative to body 2 and
is stationary in body 1, find the resulting steady-state bulk temperature T1∞ in body
1 as a function of f, V, p, a, K , k, where these symbols are defined in Sect. 18.6.1.
Would your answer be changed if there were many contact areas with a distribution
of contact radii, assuming the mean contact pressure p is the same for all?
Chapter 19
Elastodynamic Contact Problems
If the loads on an elastic body change in time, the resulting elastic displacements
u(t) will imply local accelerations ü(t), which should be taken into account in the
equations of motion. We obtain
∂σx x ∂σ yx ∂σzx ∂2u x

+ + =ρ 2
∂x ∂y ∂z ∂t
∂σx y ∂σ yy ∂σzy ∂2u y
+ + =ρ 2 (19.1)
∂x ∂y ∂z ∂t
∂σx z ∂σ yz ∂σzz ∂2uz
+ + =ρ 2 ,
∂x ∂y ∂z ∂t
where ρ is the density of the material. In many problems of elasticity, including most
elastic contact problems, we make the assumption that the loading rate is sufficiently
slow for the acceleration terms to be negligible, in which case the solution is described
as ‘quasi-static’. In other words, the body is assumed to pass through a sequence of
equilibrium states.
If the acceleration terms are retained in Eq. (19.1), the solution lies in the field
of elastodynamics, and since the stresses are themselves defined in terms of strains
and hence displacements, the problem must necessarily be solved in a space–time
context. This introduces serious complications into both analytical and numerical
solutions, so the first question to address is ‘What parameters determine whether the
quasi-static solution represents a satisfactory approximation?’
Two approaches can be suggested for answering this question. If the loading is
periodic, the quasi-static solution will generally be acceptable if and only if the
frequency of loading ω is significantly below the first natural frequency ω1 of the
elastic system. Alternatively, we can compare the rate of change of quasi-static elastic

https://doi.org/10.1007/978-3-319-70939-0_19
476 19 Elastodynamic Contact Problems
displacements in the body with the speed at which waves propagate through the
material. This is particularly useful if the problem involves a moving contact area,
as in rolling or sliding contact.
19.1 Wave Speeds
The equations of motion (19.1) can be expressed in terms of displacements alone by

using Hooke’s law in the form

2G ∂u x ∂u y ∂u z ∂u x ∂u y
σx x = (1 − ν) +ν +ν ; σ yx = G +
(1 − 2ν) ∂x ∂y ∂z ∂y ∂x
(19.2)
etc., with the result
(1 − 2ν)ρ ∂ 2 u
∇∇ · u + (1 − 2ν)∇ 2 u = . (19.3)
G ∂t 2
The wave speeds c in a large monolithic body can then be obtained by seeking
solutions in which (for example) u = f (x −ct). Two such states can be identified,
depending on whether the resulting displacements are aligned with, or orthogonal to
the direction of propagation. In the former case, known as irrotational or dilatational
waves or P-waves, we write
u x (x, y, z, t) = f (x − c1 t); u y = u z = 0, (19.4)
and the only non-trivial components of Eq. (19.3) reduces to
(1 − 2ν)ρc12
2(1 − ν) f (x − c1 t) = f (x − c1 t). (19.5)
G
This is satisfied for all functions f (·) as long as

2G(1 − ν)
c1 = , (19.6)
(1 − 2ν)ρ
which defines the dilatational wave speed.

If the displacements are orthogonal to the direction of propagation, the resulting
waves are known as shear waves or transverse waves or S-waves. Writing
u y (x, y, z, t) = f (x − c2 t); u x = u z = 0, (19.7)

19.1 Wave Speeds 477
Equation (19.3) reduces to
ρc22
f (x − c2 t) = f (x − c2 t) (19.8)
G
which is satisfied for all functions f (·) as long as

G
c2 = , (19.9)
ρ
which defines the shear wave speed. For all admissible values of ν, c1 > c2 .
19.1.1 Rayleigh Waves
If a large elastic body has a traction-free plane surface, disturbances can propagate
near the surface in the form of Rayleigh waves. These are most conveniently investi-
gated by considering the special case where the disturbance has the sinusoidal form

u = f (z) exp ıω(x − ct) , (19.10)
where f (z) is a possibly complex in-plane1 vector function of z. Substitution into

Eq. (19.3) yields the equations
d 2 fx d fz
c22 2
− (c12 − c2 )ω 2 f x = −ı(c12 − c22 )ω
dz dz
2
d f z d fx
c12 2 − (c22 − c2 )ω 2 f z = −ı(c12 − c22 )ω ,
dz dz
which have two linearly independent solutions that decay exponentially with z. Tak-
ing a linear combination of these solutions and enforcing the condition that the sur-
face be traction-free, we obtain two homogeneous algebraic equations which have a
non-trivial solution if and only if

R(c) ≡ (2 − M22 )2 − 4 (1 − M12 )(1 − M22 ) = 0, (19.11)
where c c
M1 = ; M2 = (19.12)
c1 c2
are Mach numbers (Achenbach 1984). Only one solution c = c R of Eq. (19.11)
leads to physically meaningful waves and the corresponding wave speed lies in the
1 So f y (z) = 0 for all z.

range 0 < c R < c2 < c1 for all values of ν. Explicit expressions for c R are given by
Rahman and Barber (1995). Notice that Eq. (19.11) does not contain the wavenumber
ω and hence all waves propagate with the same speed—i.e. the propagation is non-
dispersive. Expressions for more general Rayleigh wave forms can be obtained by
using Eq. (19.10) as the kernel of a Fourier transform, which of course is a form of
linear superposition.
19.2 Moving Contact Problems
Consider the problem of a rigid body pressed against the elastic half-plane z > 0 by a
force P, and sliding without friction at constant speed V in the positive x-direction.
It seems reasonable to expect that eventually we shall reach a steady state in which
the stress and displacement fields are invariant in a frame of reference moving with
the body, and this limiting case can be formulated, as in the quasi-static case, using
an appropriate Green’s function.
19.2.1 The Moving Line Force
Cole and Huth (1958) gave the two-dimensional steady-state solution for a concen-
trated line force moving over the surface of a half-plane at constant speed V . In
particular, if the force P is normal to the surface, the normal surface displacements
are
P

u z (x, 0) = F1 (V ) ln |x| + F2 (V )H (−x) , (19.13)
G
where

M22 1 − M12
F1 (V ) = 0 < V < c2
π R(V )

M22 (2 − M22 )2 1 − M12
= c2 < V < c1
π[(2 − M22 )4 + 16(1 − M12 )(M22 − 1)]
=0 V > c1 (19.14)
F2 (V ) = 0 0 < V < c2

4M22 (1 − M12 ) M22 − 1
= c2 < V < c1
(2 − M22 )4 + 16(1 − M12 )(M22 − 1)

M22 M12 − 1
= V > c1 , (19.15)
(2 − M22 )2 + 4 (M12 − 1)(M22 − 1)
19.2 Moving Contact Problems 479
Fig. 19.1 The functions F1 (V ), F2 (V ) from Eqs. (19.14), (19.15), for the case ν = 0.3
and R(V ), M1 , M2 are defined as in (19.11), (19.12) with V replacing c.

Figure 19.1 shows F1 , F2 as functions of V for the case where Poisson’s ratio
ν = 0.3. Notice that F1 is unbounded at the Rayleigh wave speed V = √c R , and
changes sign for V > c R . Also, F1 is tangent to zero at a single point V = 2c2 , but
is otherwise positive in the range c2 < V < c1 .
We also note that the x-derivative of Eq. (19.13) is

P F1 (V )
u z (x) = − F2 (V )δ(x) , (19.16)
G x
where δ(·) is the Dirac delta function.
19.2.2 Integral Equation Formulation
If the moving body has a shape defined by the initial gap function g0 (x), arguments
similar to those in Sect. 6.2 show that the pressure distribution p(x) must satisfy the
integral equation

p(ξ)dξ
F1 (V ) − F2 (V ) p(x) = −Gg0 (x) x ∈ A, (19.17)
A (x − ξ)
where the contact area A must be chosen so as to satisfy the auxiliary conditions

p(ξ)dξ = P (19.18)
A
and
p(x) ≥ 0 x ∈ A; g(x) ≥ 0 x∈
/ A. (19.19)
19.2.3 The Subsonic Problem
If the speed V < c2 , the function F2 (V ) = 0 and (19.17) reduces to

p(ξ)dξ Gg (x)
=− 0 x ∈ A, (19.20)
A (x − ξ) F1 (V )
which has the same form as (6.9) with
∗ 2G
E replaced by − . (19.21)
π F1 (V )
In the limit V → 0, F1 (V ) → −(1−ν)/π and the expressions in (19.21) become

identical [since the moving body is rigid]. Thus from the form of the function F1 (V )
in 0 < V < c R [see Fig. 19.1], we see that the steady motion has the same effect
as a reduction in the elastic modulus of the material.2 Solutions in this speed range
can be simply written down by using the substitution (19.21) in the corresponding
quasi-static solution.
Example
If a rigid cylinder of radius R is pressed against the half-plane by a force P and
rolls or slides without friction at speed V < c R , contact will occur in the range
−a < x < a, where x is measured from the point instantaneously below the centre
of the cylinder, and the contact semi-width a is obtained by using the substitution
(19.21) in Eq. (6.24), giving

2P R F1 (V ) 1/2
a= − . (19.22)
G
The contact pressure is then given by Eq. (6.25) as

√
2P a 2 − x 2
p(x) = . (19.23)
πa 2
This solution was first given by Craggs and Roberts (1967).
As the speed V is increased, the contact area gets larger and theoretically increases
without limit as V → c R . This behaviour is analogous to the excitation of a dynamic
system at a frequency approaching resonance, since the propagation of a Rayleigh
wave is a solution of a homogeneous boundary value problem (the surface is traction-
free) and hence a form of resonance. Of course, before this limiting speed is reached,
nonlinear effects will become important and necessitate a large deformation solution.
2 Notice that R(V ) < 0 and hence F1 (V ) < 0 in the range 0 < V < c R .
19.2.4 The Speed Range c R < V < c2
If the sliding speed lies in the range c R < V < c2 , we still have F2 (V ) = 0, and hence
the integral equation (19.20) defines a formal solution to the problem. However
F1 (V ) is now positive, so in the example problem, Eq. (19.22) defines an imaginary
length and hence the solution is unphysical.
Contact pressure distributions satisfying (19.17)–(19.19) can be found, but only
if we relax the continuity condition that p(x) → 0 at the edges of the contact region.
We recall from Sect. 10.1.3 that under quasi-static conditions, the first (singular)
term in an asymptotic expansion of the fields near the contact edge violates one
or other of the inequalities (19.19) for either sign of the multiplying constant, and
hence both inequalities can be satisfied only if this constant is set to zero. However,
the substitution (19.21) shows that in the range c R < V < c2 , the relative motion
is mathematically equivalent to a change in effective modulus to a negative value.
A similar asymptotic argument then shows that there exists a choice of sign of the
singular asymptotic term that will satisfy both inequalities (Georgiadis and Barber
1993).
The solution of the resulting contact problem is therefore non-unique, since both
boundaries of the contact area can be chosen arbitrarily within certain limits. Fur-
thermore, the moving singularities constitute concentrated energy sources and sinks,
which are generally regarded as unphysical (Freund 1972; Comninou and Dundurs
1978).
Energy Sources
To explore the conditions under which an energy source or sink might be acceptable,
it is instructive to consider the problem of a flat and rounded punch sliding over
the surface of a half plane, as shown in Fig. 19.2. In the sub-Rayleigh speed range,
V < c R , the punch indents the surface as shown in Fig. 19.2a and the contact area is
determined by the requirement that p(x) → 0 at each end, x = ±a. Notice that there
are small regions of contact b < |x| < a at the rounded ends of the punch. As the
punch moves to the right, work is done by the contact pressure in b < x < a, where
the surface is being pressed down, and work is done against the contact pressure in
−a < x < b.
If we now allow the radius of the ends to tend to zero, these edge contact regions
will shrink to zero, but the work done in them will tend to a constant, proportional to
the strength of the square-root singularity in the limiting solution for a sharp-ended
flat punch. The half-plane will see an energy source at the leading edge x = a, and
an equal energy sink at the trailing edge x = −a and these are physically reasonable
when considered as a limiting case of the problem of Fig. 19.2a.
Figure 19.2b shows the configuration implied by a typical solution of Eq. (19.17)
in the speed range c R < V < c2 . The length a defining the contact area is now arbitrary,
and singularities in contact pressure occur at both edges x = ±a. Because of the sign
change in the Green’s function, the leading edge x = a is now an energy sink, whereas
the trailing edge x = −a is an energy source. Furthermore, no limiting arguments
P
P
(a) (b)
V
V
x
b b
a a x a a
Fig. 19.2 Contact configuration for the moving flat and rounded punch (a) for V < c R , (b) for
c R < V < c2 , if singular fields that satisfy the inequality conditions are regarded as acceptable
such as those implied by Fig. 19.2a can be adduced to explain the mechanism of
these sources and sinks.
19.2.5 The Solution of Slepyan and Brun
This paradox, which also applies to the solution of Eq. (19.17) in the transonic range
c2 < V < c1 , remained unresolved for 45 years, until a unique and physically meaning-
ful solution was presented by Slepyan and Brun (2012). The basis of their argument
is
(i) that energy sinks are physically acceptable, because the impact of material at the
leading edge can be inelastic, whereas energy sources are unphysical, and
(ii) that the asymptotics at the trailing edge of the contact area require both the value
and the spatial derivative of the contact pressure to be zero.
The asymptotic fields are central to this argument, since without the extra end con-
dition, the solution would still be non-unique.
Asymptotic Fields
We recall from Sect. 10.1 and Eqs. (10.15), (10.19) that the first few terms in the
asymptotic expansion give
4B1r 1/2 4B3r 3/2

p(r ) = B1r −1/2 + B3r 1/2 + . . . ; g(r ) = − ∗ + + . . . , (19.24)
E 3E ∗
so the corresponding expression for the gap g(r ) when the boundary is moving
steadily at speed V is
2π F1 (V )B1r 1/2 2π F1 (V )B3r 3/2

g(r ) = − + ..., (19.25)
G 3G
using (19.21). If c R < V < c2 , we have F1 (V ) > 0 and hence both the inequalities
p(r ) > 0, g(r ) > 0 are satisfied by the singular term with B1 > 0, and this term
dominates the expansion, so no additional conditions are obtained. However, at the
trailing edge, the singular term defines an energy source which must be excluded,
giving B1 = 0. The second term, B3 is then dominant in the expansion, and this
will violate one or other inequality whichever sign is taken for B3 , so we must also
enforce B3 = 0. This implies that both the value and the spatial derivative of the
contact pressure must be zero at the trailing edge of the contact area.
Example: The Hertz Problem
We illustrate this argument with the two-dimensional Hertz problem, for which
g0 (x) = x 2 /2R. Because the edge conditions are not symmetric, we anticipate that the
contact area will also not be central, so we assume that contact occurs in b < x < c,
where b, c are unknowns to be determined. However, to use the methodology in
Chap. 6 and particularly Sect. 6.4, we move the origin to the mid-point of the contact
area by defining x , a, d, such that
b+c c−b
x = x − d; d = ; a= , (19.26)
2 2
so that contact now occurs in −a < x < a. In this coordinate system, the initial gap
is given by
(x + d)2 (a cos φ + d)2
g0 (x ) = = , (19.27)
2R 2R
where x = a cos φ, so
dg0 (φ) (a cos φ + d)(−a sin φ) ad sin φ a 2 sin(2φ)

= =− − , (19.28)
dφ R R 2R
corresponding to
ad a2
g1 = − ; g2 = − (19.29)
R 2R
in Eq. (6.39)2 . The contact pressure is then given by
p0 + p1 cos θ + p2 cos(2θ) x
p(θ) = where cos θ = (19.30)
sin θ a
and
Ggn P
pn = − ; n = 0; p0 = (19.31)
π F1 (V )a πa
from (6.41), (6.43), where P is the total force and we have used the substitution
(19.21) in (6.41).
Fig. 19.3 Sliding of a rigid

P
cylinder over an elastic
half-plane for sliding speeds
in the range c R < V < c2
V
b 2b
x
The energy criterion requires p(x ) to be non-singular at x = −a which corre-

sponds to θ = π, and hence
p0 − p1 + p2 = 0. (19.32)
Using this condition to eliminate p0 in (19.30) and expressing the result as a function
of x , we obtain
a + x 2 p2 (a − x )
p(x ) = p 1 − . (19.33)
a − x a
This expression will contain a square-root bounded term at the trailing edge x = −a,
unless the term in brackets tends to zero there, leading to the additional smoothness
criterion p1 = 4 p2 . Using (19.31) and (19.29) in this relation, we then have
d = 2a which implies c = 3b, (19.34)
from (19.26).
The resulting contact configuration is illustrated in Fig. 19.3, where we note that
contact is restricted to the ‘leading’ side of the cylinder. The remaining unknowns
can then be determined from Eqs. (19.29), (19.31), (19.32), from which we determine
the contact pressure distribution as
1/2
2P(x − b)3/2 2P F1 (V )R
p(x) = where b= . (19.35)
3πb2 (3b − x)1/2 3G
The contact pressure distribution p(x) (19.35)1 is illustrated in Fig. 19.4. Notice
how it exhibits a dependence on (x−b)3/2 at the trailing edge, in contrast to the square-
root bounded behaviour in quasi-static or sub-Rayleigh elastodynamic problems.
19.2.6 The Transonic Solution c2 < V < c1
In the speed range c2 < V < c1 , both F1 (V ) and F2 (V ) are non-zero,3 so Eq. (19.17) is
a singular integral equation of the second kind, whose solution is given in Appendix
C. In particular, if the contact segment is defined as −a < x < a, the singular solution
√
3 Except for the particular speed V = 2c2 , where F1 (V ) = 0.
Fig. 19.4 Contact pressure

distribution in the sliding
Hertz problem for sliding
speeds in the range
c R < V < c2
has the form

p(x) = (a − x)−1/2−γ (a + x)−1/2+γ f (x), (19.36)
where f (x) is a bounded function and
F2 (V ) 1 1
tan(πγ) = − − <γ< . (19.37)
π F1 (V ) 2 2
Asymptotic analysis of the fields near the end points yields results very similar to
those in the speed range c R < V < c2 . The inequalities are satisfied by the singular
terms at both contact edges, but we rule out the singularity at the trailing edge because
it represents an energy source.4 The second term in the local asymptotics must then
also be ruled out because of the inequalities, as in Sect. 19.2.5. so the permissible
contact pressure distribution will actually take the form
p(x) = (a − x)−1/2−γ (a + x)3/2+γ F(x), (19.38)
where F(x) is bounded. In this speed range F1 (V ), F2 (V ) are both positive, so

−1/2 ≤ γ ≤ 0, implying that the leading singularity is never stronger than square
root. The contact configuration is qualitatively similar to Fig. 19.3, but with different
coordinates for the contact end points. More details of the resulting solution are given
by Slepyan and Brun (2012).
19.2.7 The Superseismic Solution V > c1
If V > c1 , F1 (V ) = 0 and (19.17) reduces to the simple form
Gg0 (x)
p(x) = x ∈ A. (19.39)
F2 (V )
4 Because of the enhanced exponent, such an energy source if permitted would actually be infinite!
Fig. 19.5 Contact

P
configuration for V > c1 . The
surface is undisturbed in
x > a and contact terminates
at the lowest point of the V
punch
a
x
The contact configuration for the Hertzian case g0 (x) = x/R is shown in Fig. 19.5.
Notice that the surface is undisturbed ahead of the sliding punch, as indeed it must be,
since no waves can travel faster than the punch. Contact terminates at the lowest point
of the punch, and behind it, the surface remains depressed by a constant distance.
Substituting for g0 (x) into (19.39), we obtain
Gx
p(x) = 0 < x < a, (19.40)
F2 (V )R
where the distance a defining the right extent of the contact area is determined from
the equilibrium condition

a
Ga 2 2F2 (V )P R
P= p(x)d x = so a= , (19.41)
0 2F2 (V )R G
where P is the total normal force. This solution was first given by Craggs and Roberts
(1967).
Equation (19.39) can be used for a punch of any profile, but it should be noted
that separation will occur in any region where g0 (x) < 0 and contact will then be
re-established at the next point at the same level. The implication of this for a punch
with a wavy profile is shown in Fig. 19.6.
We notice from Eq. (19.13)
√ and Fig. 19.1 that F1 (V ) is zero in the transonic range
at the special value V = 2 c2 , so the pressure distribution (19.39) and the configu-
rations of Figs. 19.5, 19.6 also hold at this speed.
Fig. 19.6 A contact

configuration for V > c1 P
leading to two separate
contact regions
V
Three-dimensional steady-state sliding contact problems for the half-space can be

formulated in terms of integral equations, as in Chap. 2, provided we first obtain the
corresponding Green’s function, defining the normal surface displacement due to a
point force P moving at constant speed V over the surface.
We define polar coordinates (r, θ, z) relative to the instantaneous position of the
force, such that the force moves in the direction θ = 0. As in Sect. 2.2.2, similarity
considerations dictate that the surface displacements vary inversely with distance r
from the force, but the motion introduces a degree of anisotropy into the results,
which can be described by a function f (θ) where
P f (θ)
u z (r, θ, 0) = . (19.42)
Gr
Speeds in the Range 0 < V < cR
For 0 < V < c R , the function f (θ) is given by
F1 (V sin θ)
f (θ) = − , (19.43)
2
where F1 (·) is defined in Eq. (19.13). This result was first obtained by Churilov
(1977), by applying the Smirnov–Sobolev transform of Sect. 6.6 to the two-
dimensional solution. Notice that points along a line perpendicular to the direc-
tion of motion [θ = ±π/2] experience greater displacement than those along the line
of motion [θ = 0, π].
Speeds in the Range c R < V < c2
Equations (19.42), (19.43) also provide a formal solution for the displacements due
to a force moving at a speed V in the range c R < V < c2 , but the results exhibit Cauchy
singularities at θ = ±γ and θ = π ± γ, where sin(γ) = c R /V . If the moving force is
conceived as starting at some finite time and then moving for a long enough time to
establish a steady state, we would expect to see the ‘trailing’ Rayleigh singularities
at θ = π ± γ, but not those at θ = ±γ that advance beyond the force.
This problem can be resolved by noting that for V > c R , the solution is non-
unique, since the combination of two Rayleigh waves of appropriate form moving
in the directions θ = ±γ define a standing wave moving with the force that can
be superposed without changing the boundary conditions of the problem (Barber
1996). In particular, we can choose the form of these Rayleigh waves to eliminate
the ‘leading’ singularities, giving
1
f (θ) = − F1 (V sin θ) − g(V, θ) c R < V < c2 , (19.44)
2
where
√ √
M2 cos θ 1 − Λm 1 (2 − m 1 )2 + 4 (1 − m 1 )(1 − Λm 1 )
g(V, θ) =
π(m 3 − m 1 )(m 1 − m 2 ) m 1 − M22 sin2 θ M22 − m 1
V c2 (1 − 2ν)
M2 = ; Λ = 22 = , (19.45)
c2 c1 2(1 − ν)
and m 1 , m 2 , m 3 are the three non-zero factors of the Rayleigh function

4
D = 2 − M22 − 16 1 − M12 1 − M22

= M22 M22 − m 1 M22 − m 2 M22 − m 3 , (19.46)
with m 1 = (c R /c2 )2 . Closed-form expressions for m 1 , m 2 , m 3 are given by Rahman

and Barber (1995).
Transonic and Superseismic Speeds
Churilov’s method can also be used to obtain the corresponding Green’s function for
V > c2 (Barber 1996). For the transonic range c2 < V < c1 , we have

1
f (θ) = − 2G 1 (V sin θ)F3 (V sin θ)H (− cos θ)
2π

3
Ai M2 cos θ mi − 1
+G 2 (V sin θ)F4 (V sin θ) + 2 2
i=2
M2 sin θ − m i m i − M22
⎫
√ ⎬
M2 cos θ 1 − Λm 1 (2 − m 1 )2
− ⎭
, (19.47)
(m 3 − m 1 )(m 1 − m 2 ) m 1 − M22 sin2 θ M22 − m 1
where

G 1 (V ) = H (c2 − V ) 1 − M22 ; G 2 (V ) = H (c1 − V ) 1 − M12 (19.48)
2
4M22 1 − M12 M22 2 − M22
F3 (V ) = ; F4 (V ) = , (19.49)
D D
the Rayleigh function D is defined in (19.46) and
4(1 − Λm 2 ) 4(1 − Λm 3 )
A2 = − ; A3 = − . (19.50)
(m 1 − m 2 )(m 2 − m 3 ) (m 2 − m 3 )(m 3 − m 1 )
Fig. 19.7 Normal surface displacements due to a moving point force for ν = 0.2 and V /c2 ≡ M2 =
0.80, 0.95, 1.3, 2.5
For supersonic speeds V > c1 , we have
f (θ) = −F1 (V sin θ)H (− cos θ). (19.51)
The function f (θ) is shown for ν = 0.2 and some representative speeds in Fig. 19.7.
For this case, the Rayleigh wave speed c R = 0.911 c2 . Notice how the curve for
M2 = 0.8 (V < c R ) is positive for all θ with a maximum at θ = 90◦ , normal to
the direction of motion. For M2 = 0.95 (c R < V < c2 ), the displacement is negative
(i.e. a compressive force causes the surface to move outwards) for θ < 106◦ , at
which limit there is a singularity, with points in the trailing sector exhibiting positive
displacement.
In the superseismic case M2 = 2.5, there is no displacement in the sector −139◦ <
θ < 139◦ , since disturbances at the moving force have not had time to reach points
in this range. The displacement immediately behind these wave fronts is negative
(outward), but as in all the other super-Rayleigh cases, there is a Cauchy singularity
associated with a sign change along two radii at θ = ±(π−γ).
The above results were extended to include both normal and tangential displace-
ments due to a force with arbitrary direction by Georgiadis and Lykotrafitis (2001),
using the Radon transform method.
Implications for Non-conformal Contact Problems
Willis (1966) showed that with a Green’s function of the form (19.42) [in his case
derived from anisotropic elastostatics], the contact area in the Hertzian problem will
remain elliptical. This result is also a consequence of the anisotropic Galin’s theorem
(Willis 1967). Rahman (1996) used Eq. (19.43) to solve the elastodynamic Hertzian
contact problem for V < c R and showed that the ratio of the axes a/b increases
with V , where a is the axis in the direction of motion. As the Rayleigh wave speed
is approached, this ratio increases without limit and the body essentially rides in a
self-sustaining Rayleigh wave ‘groove’ perpendicular to the direction of motion.
For V > c R , a Hertzian contact pressure distribution can be found that satisfies
the boundary conditions inside an elliptical area, but as in the two-dimensional case,
there is no choice for the dimensions of this area that satisfy the inequality conditions.
The author is unaware of any solutions, numerical or analytical, to three-dimensional
steady-state elastodynamic contact problems in this speed range.
19.3 Interaction of a Bulk Wave with an Interface
Wave speeds in metals are of the order of thousands of metres per second, so it might
be argued that the preceding results are somewhat academic, though much lower
speeds are obtained for low modulus materials such as rubber (around 3 m/s) and
plastics (typical values around 25 m/s). However, even for high modulus materials,
contact problems in the elastodynamic range can arise if an elastic wave or pulse
impinges on a plane interface.
19.3.1 SH-Waves Transmitted Across a Frictional Interface
Figure 19.8 shows a simple case in which a shear wave propagating in the x -direction
is defined by the displacement field

u y (x, z, t) = u 0 sin ω(x − c2 t) = u 0 sin ω(x sin α − z cos α − c2 t) . (19.52)
Notice that the displacement is normal to the x z-plane, so this is an antiplane defor-
mation. Waves of this kind are referred to as SH-waves.
Fig. 19.8 SH-wave incident x

on the plane z = 0
O
x
α
c2
z
incident wave
19.3 Interaction of a Bulk Wave with an Interface 491
The shear stress component

∂u z ∂u y
σzy = G + = −Gu 0 ω cos α cos ω(x sin α − z cos α − c2 t) ,
∂y ∂z
(19.53)
and hence the traction on the plane z = 0 is

c2 t
q y (x, t) = −σzy (x, y, 0, t) = Gu 0 ω cos α cos ω sin α x − . (19.54)
sin α
Now consider the case where we actually have two half-spaces z > 0 and z < 0 of
identical materials making frictional contact at the common plane z = 0. Equation
(19.54) then defines the frictional tractions transmitted between the half-spaces under
the assumption that no slip or separation occurs. In other words, the solution is the
same as if the two half-spaces were bonded together. We shall refer to this as the
bilateral solution. The tractions (19.54) appear to move over the interfacial plane at
a speed
c2
c= , (19.55)
sin α
which we note can take any value in the range c > c2 , depending on the angle of
incidence α of the original wave.
Suppose that the two half-spaces are loaded by a uniform normal compressive
traction −σzz = p0 and sheared by a uniform shear traction σzy = q0 . The no-slip
assumption then requires that |q y (ξ)| < f p0 for all ξ, where
q y (ξ) = q0 + Gu 0 ω cos α cos (

ω ξ) , (19.56)
f is the coefficient of friction, and
ξ = x − ct;
ω = ω sin α (19.57)
define respectively a coordinate moving with the wave at the interface and the pro-
jected wavenumber of this wave.
The no-slip condition will be violated at some parts of the interfacial plane unless
|q0 | + Gu 0 ω cos α < f p0 (19.58)
and these violation regions will appear to move along the interface at speed c (> c2 )
given by Eq. (19.55).
The Corrective Solution
If the condition (19.58) is not satisfied, we must anticipate slip regions Aslip which
move along the interface at speed c and in which q y = ± f p0 . We can satisfy this
Fig. 19.9 Bilateral shear qy

tractions (sine wave) and
frictional limits ± f p0 . The fp0
actual interfacial tractions
will be truncated at these q0
limits as shown, so the q0
corrective solution removes 0
O ξ
the shaded regions from the
distribution
fp0
condition by superposing an appropriate distribution of shear tractions q y∗ (ξ) on the

lower body z > 0 with equal and opposite reactions on the upper body, also moving
at speed c, so as to satisfy the conditions
q y + q y∗ = ± f p0 ξ ∈ Aslip (19.59)

∂u y ∂u y
− =0 ξ ∈ Astick , (19.60)
∂ξ 1 ∂ξ 2
where the suffices 1, 2 refer to the surface displacements of the bodies z > 0 and
z < 0, respectively.
As a preliminary to this process, we note that an antiplane concentrated force Q y
moving at speed c over the body z > 0 produces a surface displacement given by
Q y ln(ξ)
u y (ξ, 0) = − √ M <1 (19.61)
πG 1 − M 2
Q y H (−ξ)
= √ M > 1, (19.62)
G M2 − 1
where c
M= . (19.63)
c2
This is the tangential antiplane counterpart of Eqs. (19.13)–(19.15) for a moving

normal force P.
In the present case, c > c2 from (19.55) so the corrective traction distribution
q y∗ (ξ) produces a surface displacement such that
∂u y q y∗ (ξ)
=− √ , (19.64)
∂ξ G M2 − 1
where we recall that the derivative of a step function is a Dirac delta function. It
follows that the condition (19.60) is satisfied by choosing q y∗ (ξ) = 0 in the stick areas
Astick and hence the corrective solution comprises merely the difference between the
bilateral tractions and the limiting frictional tractions in Aslip .
Figure 19.9 shows the tractions in a particular case, where q0 and u 0 are such as
to require slip regions at the maxima of q y (ξ), but not at the corresponding minima.
The slip zones Aslip are defined by
2nπ 2nπ f p0 − q 0
− ξ0 < ξ < + ξ0 where ω ξ0 ) =
cos( (19.65)

ω
ω Gu 0 ω cos α
and n is any integer. The corrective tractions in these zones are
q y∗ (ξ) = f p0 − q0 − Gu 0 ω cos α cos(

ω ξ). (19.66)
These tractions will cause slip displacements in z > 0 given by (19.64) and an equal
and opposite displacement in the upper body z < 0. It follows that after the passage
of one such slip zone past a given point, there will be an accumulated shift h given
by
ξ0

2 4( f p0 − q0 ) tan(
ω ξ0 ) −
ω ξ0
h0 = − √ q y∗ (ξ)dξ = √ , (19.67)
G M2 − 1 −ξ0 Gω M2 − 1
where we have adopted the convention that positive shift involves the upper body
moving in the positive y-direction relative to the lower body [so that the applied
traction q0 does positive work]. This increment of shift accumulated during a time
period t0 = 2π/ωc2 , so the upper body appears to creep over the lower body at a
creep velocity
h0 ω ξ0 ) −
2( f p0 − q0 )c2 tan( ω ξ0
Vc = = √ , (19.68)
t0 πG sin α M 2 − 1
where ξ0 is defined by Eq. (19.65).
Dissimilar Materials
We have chosen a very simple example to illustrate this process, but similar techniques
can be used if the materials of the two half-spaces are different (Chez et al. 1978).
The bilateral solution will now generally involve a reflected wave

(1)
u (1)
y = u 1 sin ω1 x sin β1 + z cos β1 − c2 t (19.69)
and a refracted wave

(2)
u (2)
y = u 2 sin ω 2 x sin β2 − z cos β2 − c2 t , (19.70)
as shown in Fig. 19.10, where c2(1) , c2(2) are the shear wave speeds for materials 1, 2,
respectively.
Fig. 19.10 SH-wave

(2)
reflected and refracted at a c2
plane interface β2
2.
x
1. α β1
c (1) c (1)
2
2
z
incident wave
The angles β1 , β2 and wavenumbers ω1 , ω2 must be such as to preserve the same

velocity c and wavenumber ω at the interface for all wave components, giving
c2(1) c(2)
β1 = α; = 2 ; ω1 = ω; ω2 sin β2 = ω sin α ≡
ω. (19.71)
sin α sin β2
The displacements and tractions at the interface are now obtained as
u y = (u 0 + u 1 ) sin(
ω t); σzy = G 1 (u 0 − u 1 ) cos α cos(
ω ξ) (19.72)
in body 1, and
u y = u 2 sin(
ω t); σzy = G 2 u 2 cos β2 cos(
ω ξ) (19.73)
in body 2, so the bilateral (bonded) condition requires that
u0 + u1 = u2 and G 1 (u 0 − u 1 ) cos α = G 2 u 2 cos β2 , (19.74)
from which the amplitudes u 1 , u 2 can be determined.

The corrective solution is obtained exactly as in the similar material case, except
that we require the bimaterial equivalent of Eq. (19.64), which is
⎛ ⎞

∂u y ∂u y 1 1
− = −q y∗ (ξ) ⎝ + ⎠, (19.75)
∂ξ 1 ∂ξ 2 2
G 1 M(1) − 1 G 2 M(2)
2
−1
where c
M(k) = k = 1, 2. (19.76)
c2(k)
More General Waveforms
We notice that the wavenumber does not appear in the bilateral equations (19.74)
and hence the reflection and refraction process is non-dispersive, meaning that wave-
forms of more general shape, including isolated pulses are preserved in the process
(Comninou and Dundurs 1980). The same technique can therefore be used for such
problems. In particular, if an incident pulse described by a displacement of the form
u y (x, z, t) = f (x − c2(1) t) = f (x sin α − z cos α − c2(1) t) (19.77)
impinges on the bimaterial interface, the bilateral reflected and refracted pulses are
given by

(1)
u (1)
y (x, z, t) = C 1 f x sin α + z cos α − c2 t (19.78)
$
(1)
c
u (2)
y (x, z, t) = C 2 f
2
x sin β2 − z cos β2 − c2(2) t , (19.79)
c2(2)
where β2 is given by (19.71) and the constants C1 , C2 are determined from continuity
of displacement and traction, as in Eq. (19.74).
If the bilateral tractions exceed the limiting friction condition in some range of
values of ξ, the corrective solution is chosen simply to cancel the excess tractions,
after which the resulting slip can be determined using (19.75). Notice that a single
pulse will cause a relative tangential shift h 0 of the two half-planes, as in Eq. (19.67).
These results are of interest in connection with the relative motion of tectonic plates
during an earthquake.
Evanescent Waves
Equation (19.71) has a real solution for β2 if and only if
c2(2) sin α
sin β2 = ≤1 or equivalently c ≥ c2(2) , (19.80)
c2(1)
where c is given by (19.55) and represents the velocity of the incident wave projected
on the interface. Condition (19.80) is always satisfied for similar materials, but it
places a restriction on the range of incidence angles α if c2(2) > c2(1) .
When c < c2(2) , the tractions on the interface are subsonic with respect to material
2 [z < 0] and the corresponding surface displacements can be determined using the
subsonic Green’s function of Eq. (19.61). In particular, we note that this has the same
[logarithmic] form as in the quasi-static case [M = 0] and as in the in-plane normal
loading case [Eq. (6.5)]. It follows that the moving sinusoidal tractions will cause a
displacement field that decays exponentially with −z, as in Sect. 6.5.1. Displacement
fields of this form are referred to as evanescent waves.
Fig. 19.11 P-wave reflected (2)

(2) c2
and refracted at a plane β2
interface (2)
c1
(2)
β1
2.
x
1. α
c (1)
0 (1) c 1(1)
β1
(1)
incident wave β2 c (1)
2
z
In the bilateral solution, the reflected wave preserves the same amplitude as the
incident wave, but with a phase lag θ, where

θ G2 G 1 ρ2
tan = sin2 α − . (19.81)
2 G 1 cos α G 2 ρ1
If the upper body is relatively rigid, so that G 2 G 1 and c2(2) c2(1) , evanescent
waves are generated for almost all incidence angles, and θ → π implying simply a
sign change in the reflected wave. The phase lag decreases monotonically with G 2 ,
reaching zero when c = c2(2) . This state represents a resonance for body 2, which
therefore mimics a traction-free surface.
Notice that although the phase lag is independent of wavelength, the spatial or
temporal delays are so dependent, and hence non-sinusoidal waves or pulses are
generally distorted by reflection. In other words, the subsonic reflection process is
dispersive.
For a frictional interface, the corrective solution can be obtained by imposing the
boundary conditions (19.59), (19.60), using the subsonic Greens function (19.61) for
body 2 and the supersonic function (19.61) for body 1. This will lead to a Cauchy
singular integral equation of the second kind (see Appendix C) over a domain com-
prising a series of segments of the x-axis (Chez et al. 1978). This in turn can be
transformed to a Cauchy equation on a single segment using the change of variable
(6.76) [see Sect. 6.5.3].
19.3.2 In-Plane Waves
If the incident wave is a dilatational wave or a shear wave involving in-plane dis-
placements (known as an SV-wave), both normal and shear tractions (σzz , σzx ) will
be generated at the interface, and in the bimaterial case, both kinds of reflected and
refracted waves will usually be generated.
Figure 19.11 shows the case where the incident wave is a P-wave, but similar
results apply to the SV-wave case, so we have labelled the incident wave velocity
c0(1) to preserve generality. As in Sect. 19.3.1, the angles must be chosen to preserve
a unique value of velocity along the interface, giving
c0(1) c1(1) c2(1) c1(2) c2(2)

= = = = . (19.82)
sin α sin β1(1) sin β2(1) sin β1(2) sin β2(2)
If any of the sines so determined exceed unity, the corresponding body wave will be
suppressed and an evanescent wave will be generated.
Problems of this kind can be treated by the same methods as in Sect. 19.3.1,
though with considerably more mathematical complexity, since now four bound-
ary conditions must be satisfied at the interface. Comninou and Dundurs (1977a)
investigated the case where the interface is frictionless [ f = 0] and the bodies are
pressed together by a uniform external pressure p0 . The amplitudes of the reflected
and refracted waves in the bilateral solution must then be chosen to satisfy the four
conditions
u (1) (2)
z (x, 0, t) = u z (x, 0, t);
(1)
σzz (2)
(x, 0, t) = σzz (x, 0, t) (19.83)
(1) (2)
σzx (x, 0, t) = σzx (x, 0, t) = 0. (19.84)
If the wave amplitudes are sufficiently large for regions of tensile contact stress to
appear in the bilateral solution, we expect regions of separation Asep to move along
the interface at speed c. We then define a corrective traction p ∗ (ξ), chosen so as to
satisfy the conditions
p(ξ) ≡ p ∗ − σzz = 0 ξ ∈ Asep (19.85)

∂u z ∂u z
− =0 ξ ∈ Acontact , (19.86)
∂ξ 1 ∂ξ 2
where Acontact is the region where contact is retained. In addition, we must satisfy a
closure condition
∂u z ∂u z
− dξ = 0 (19.87)
Asep ∂ξ 1 ∂ξ 2
in each distinct separation region, since the gap must be zero at both ends of each
such region.
Comninou and Dundurs (1977a) described the corrective solution using Fourier
series, leading to a set of dual series equations, which they then converted to inte-
gral equations. Alternatively, an integral equation can be obtained directly, using
Eqs. (19.13)–(19.15) in each body as the appropriate Green’s function, or equiv-
alently, by representing the gap in the separation zones by a distribution B(ξ) of
moving climb dislocations (Weertman 1963; Dundurs and Comninou 1979).
Fig. 19.12 Contact pressure

p(ξ)
for the frictionless problem
in the superseismic range.
Separation starts at ξ = b and p0
extends to ξ = a, where there p0
is a discontinuous jump to 0
the bilateral (sine wave) O a b ξ
value
As in Sect. 19.3.1, the corrective problem is particularly straightforward when

the surface wave is supersonic with respect to both bodies, and hence there are no
evanescent waves. In this case, the Green’s function is described by a step function,
which is zero ahead of the force, so separation must start exactly at the leading edge of
the tensile zone in the bilateral solution. However, in contrast to the SH-wave case,
separation does not cease at the corresponding trailing edge, because the closure
condition (19.87) translates [in the superseismic case only] to the condition

p ∗ (ξ)dξ = 0. (19.88)
Asep
This implies that the extent of the separation zone a < ξ < b in Fig. 19.12 is determined
by the requirement that the two shaded areas be equal. It also follows that contact is
restablished at ξ = a at a non-zero contact pressure, and hence (in view of the Green’s
function) that the closure velocity is non-zero. In other words, the bodies slam into
contact at the trailing edge of the separation zone. This argument can also be used to
determine the extent of the separation zone and the corresponding corrective solution
for a pulse or wave of arbitrary form (Dundurs and Comninou 1979).
If one or more of the waves in the bilateral solution is evanescent, the Green’s
function involves logarithmic as well as step function terms and we generally obtain
a Cauchy singular integral equation of the second kind for p ∗ (ξ) or the equivalent
distribution of dislocations B(ξ). The boundaries a, b of each separation zone are
then determined from (i) the closure condition (19.87) and (ii) the requirement that
the contact pressure be bounded at the trailing edge ξ = b of the previous contact
zone. If this latter condition is not enforced, the solution would generally have a local
singularity exceeding square root, implying an unphysical infinite energy source.
Comninou and Dundurs (1979a) used similar methods to extend the above argu-
ments to problems involving a finite friction coefficient, in which case regions of
in-plane slip as well as possible separation regions are to be expected.
19.4 Interface Waves
We saw in Sect. 19.1.1 that Rayleigh waves can propagate without dispersion along a
traction-free plane surface. Stoneley (1924) considered the more general case where
19.4 Interface Waves 499
two elastic half-spaces are bonded together at a plane interface and showed that,
depending on the relations between the material properties, waves may propagate
along the interface without the necessity for an incident bulk wave. In fact, the
Rayleigh wave could be regarded as a Stoneley wave in the limit where the elastic
modulus and density of one material approaches zero. However, Stoneley waves can
exist only under some fairly restrictive relations between the material properties. In
particular, it is clear that no such wave could exist if the material properties of the
two half-spaces are the same, since in that case there is in essence no interface.
19.4.1 Slip Waves
The situation is changed if the interface between two similar half-spaces is friction-
less, since in this case it is easy to show that equal and opposite Rayleigh waves on
the surfaces of the two bodies would maintain normal contact without involving any
moving surface tractions. There would then exist a relative tangential motion (slip)
between the surfaces, so this form of interface wave is known as a slip wave.
For the more general case of dissimilar materials, the propagating wave will
usually involve normal tractions at the interface, but these must be chosen so as to
maintain contact, so
∂u (1)
z ∂u (2)
− z =0 (19.89)
∂x ∂x
for all x. If the wave speed c is below the shear wave speed in both materials, so that
F2 (c) = 0 in Eq. (19.16), we obtain

∂u (1) ∂u (2) p(ξ)dξ
z
− z = f (1) (c) + f (2) (c) , (19.90)
∂x ∂x (x − ξ)
where
F1 (c)
f (c) = (19.91)
G
for each material, and F1 (c) is defined in Eq. (19.14). It follows that Eq. (19.89) is
satisfied for a wave of arbitrary form if c = c A , where
f (1) (c A ) + f (2) (c A ) = 0 (19.92)
(Achenbach and Epstein 1967; Murty 1975). In order to satisfy (19.92), the functions
f (1) , f (2) must have opposite signs, and it is clear from (19.91) that this will be the
case if and only if F1(1) (c A ), F1(2) (c A ) also have opposite signs. This in turn implies
that
c(1)
R < cA < cR ,
(2)
(19.93)
as can be seen for example from the function F1 (V ) in Fig. 19.1. In other words,
the slip wave speed is intermediate between the Rayleigh wave speeds of the two
materials.
To preserve the contact condition (19.89), the half-planes must be pressed together
by a uniform external pressure p0 sufficient to ensure that p0 + p(ξ) > 0 for all ξ,
so that no regions of tensile traction are generated at the interface. If p0 is too
small, or the amplitude of the slip wave too large, we might expect propagating
regions of separation to be developed. This possibility was investigated by Comninou
and Dundurs (1977b), who found that such solutions could be found if and only if
square-root singular tractions were permitted at both ends of each contact region.
Furthermore, in contrast to slip waves which propagate at a unique speed given by
Eq. (19.92), the waves with separation can then propagate with any value in a certain
speed range. In view of the discussion in Sect. 19.2.4, these singular tractions must
be regarded as of questionable validity, but if they are ruled unphysical, it is not
clear what would happen if a slip wave (for example at the interface between two
tectonic plates) were to propagate into a region where the underlying normal traction
is insufficient to maintain contact.5
19.4.2 Slip Waves at a Sliding Interface
Consider the case where two half-planes of the same material are pressed together
by a uniform pressure p0 and the upper body is caused to slide to the right over
the lower body at some speed V c R . This problem has a trivial solution in which
the shear traction is uniform and given by q0 = −σzx (x, 0, t) = f p0 , where f is the
coefficient of friction.
Suppose we now ‘superpose’ the stresses and displacements corresponding to a
frictionless slip wave of very small amplitude, such that the slip velocity is nowhere
reduced below zero. This superposition is strictly incorrect, because the Rayleigh
wave in the moving body will now have an absolute velocity c R ± V and will not
match that in the stationary body, but for V c R this error will be small.
Since there are no shear or normal tractions associated with the slip wave, the
frictional condition q(x) = f p(x) is still satisfied, so we conclude that such a wave
can propagate at a frictional sliding interface. In effect, the slip wave represents a
neutrally stable dynamic state.
If the two half-planes are of different materials, the slip wave will perturb the
normal tractions and the above superposition will not satisfy the frictional bound-
ary condition, except in the frictionless case f = 0. However, Adams (1995) has
shown that slip waves with exponentially growing amplitudes can then propagate
5 It
is possible that this problem might be resolved using the asymptotic arguments of Slepyan and
Brun (2012) [see Sect. 19.2.5], but to the present author’s knowledge, no such solution has as yet
been proposed.
at the interface. In effect, the normal-tangential coupling due to material mismatch

destabilizes the slip wave.6
We postulate the existence of displacement fields of the form
(1)
u(1) (x, z, t) = u(1)
0 (x, z) + U exp ω(ıξ + at − λ1 z) (19.94)
(2)
u(2) (x, z, t) = u(2)
0 (x, z) + U exp {ω(ıξ + at + λ2 z)} , (19.95)
where u(1) (2)

0 , u0 are the unperturbed uniform fields due to p0 , q0 and ξ = x−ct. Calcu-
lating the corresponding stress components from Hooke’s law and substituting them
into the equations of motion, we obtain characteristic equations from which λ1 , λ2
can be determined. Since the perturbation is assumed to decay with distance from
the interface, we select in each case the values of λ with positive real part. Finally,
the tractions and displacements at the interface are substituted into the interface
conditions
u (1) (2)
z (x, 0, t) = u z (x, 0, t);
(1)
σzz (2)
(x, 0, t) = σzz (x, 0, t) (19.96)
(1) (2) (1)

σzx (x, 0, t) = σzx (x, 0, t) = f σzz (x, 0, t), (19.97)
which then provide four homogeneous equations for the four unknowns Ux(1) , Uz(1) ,
Ux(2) , Uz(2) . These equations have a non-trivial solution only for certain eigenvalues
of the complex parameter (a −ıc), whose imaginary part defines the wave velocity
c, and whose real part defines the exponential growth rate through the multiplier
exp(ωat).
Adams (1995) showed that if the material combination supports frictionless slip
waves, the magnitude of a varies approximately linearly with f and its sign depends
on the direction of propagation of the wave, so there always exists an unstable solu-
tion. In fact, the unstable wave always propagates in the direction of relative motion
of the body with the lower shear wave speed c2 .
If the material combination does not support frictionless slip waves—i.e. if
Eq. (19.92) has no real solutions—there generally exists a non-zero critical coef-
ficient of friction above which unstable waves can propagate at a sliding interface.
19.4.3 Slip–Stick Waves
Adams’ results show that in many situations, steady sliding between dissimilar mate-
rials is unstable, but the resulting unstable waves cannot grow without limit. Ulti-
mately, they must lead to the prediction of regions of either tensile traction or reverse
6 Notice
that this is another instance of the importance of normal-tangential coupling in frictional
problems, first discussed in Chap. 8.
slip. In both cases, we anticipate some kind of limiting solution involving propagating
regions of separation and/or stick.
Adams (1998) gives a solution for the case where slip and stick zones propagate
along an interface, leading to steady relative tangential motion. It is perhaps mislead-
ing to refer to this as slip, since at all times there are regions of the interface that are
stuck. The resulting waves propagate at the speed defined by Eq. (19.92) and hence
are possible only when that equation has a real root.
To formulate the problem, we suppose that external tractions p0 , q0 are applied
such that q0 < f p0 , so slip throughout the interface is not possible. We then write
the complete stress state as the superposition of a uniform state of stress and that
due to a frictionless slip wave with normal tractions p ∗ (ξ) moving at speed c A . Each
component of this superposition satisfies the condition that the surfaces remain in
contact everywhere, and it leads to tractions defined by
p(ξ) = p0 + p ∗ (ξ); q(ξ) = q0 . (19.98)
The corrective traction p ∗ (ξ) is determined from the conditions

q0
q(ξ) = f p(ξ) and hence p ∗ (ξ) = − p0 ξ ∈ Aslip (19.99)
f

∂u x ∂u x
− =0 ξ ∈ Astick . (19.100)
∂ξ 1 ∂ξ 2
To apply condition (19.100), we need to know the tangential displacement u x due

to the moving line force of Sect. 19.2.1. This is given by Eringen and Suhubi (1975)
Sect. 7.11 and in the subsonic range V < c2 is
u x (x) = Pg(V ) sgn(x), (19.101)
where
1
g(V ) = M2 − 2 (1 − M1 )(1 − M2 ) .
2 2 2
(19.102)
2G R(V )
We recall that the derivative of sgn(x) is 2δ(x), so for a distributed contact pressure
p(ξ) moving at speed V , we have
∂u x
= 2g(V ) p(ξ). (19.103)
∂ξ
Condition (19.100) therefore requires that

2 g1 (c A ) − g2 (c A ) p ∗ (ξ) = 0 ξ ∈ Astick , (19.104)
where g1 (c A ), g2 (c A ) are constants obtained by substituting V = c A and the material

properties of bodies 1, 2 respectively into (19.102). Since the materials are different,
these constants will also differ and hence (19.104) can be satisfied if and only if
p ∗ (ξ) = 0 ξ ∈ Astick . (19.105)
Equations (19.99)2 and (19.105) show that the stick–slip wave has a square wave
form, with a constant reduction in normal traction (and hence also a constant slip
velocity) in the slip zones.
Since the tractions in the stick regions are lower than those required for slip, the
system as a whole appears to have a lower coefficient of sliding friction than that
defining local conditions at the interface. In fact, this apparent friction coefficient is
found to decrease as the apparent sliding speed increases, even if the local friction
coefficient is speed-independent. This mechanism for a velocity-dependent friction
coefficient was earlier suggested by Weertman (1980), based on the solution for a
dislocation moving at the interface between two half-planes.
19.5 Stability of Frictional Sliding
The existence of unstable slip waves at a sliding interface has serious consequences
for the numerical solution of elastodynamic sliding problems, such as those involved
in earthquake studies. We notice from Eqs. (19.94), (19.95) that the exponential
growth rate is given by ωa, which therefore increases with the wavenumber ω. The
heuristic basis for using a discrete [e.g. finite element] solution for continuum prob-
lems is that the continuum solution is approached asymptotically as the mesh is
refined, but in the present case, this will introduce the possibility of shorter and
shorter waves with correspondingly larger growth rates, and these waves will always
be triggered by the approximations inherent in the discretization. Thus the elastody-
namic frictional problem is essentially ill-posed (Simões and Martins 1998; Cochard
and Rice 2000).
The problem can be regularized by introducing a length or time scale into the
friction law, typically using a rate and state model, as defined in Sect. 8.6.5. Ranjith
and Rice (2001) showed that a simple law of this form defined by
|V |
q = q0 sgn(V ); q̇0 = − (q0 − f p) (19.106)
L
is sufficient to regularize the elastodynamic frictional contact problem, where p, q are
the normal and tangential tractions respectively, V is the instantaneous sliding speed,
q0 performs the function of state variable and L is a length scale. During steady sliding
at constant pressure p, we recover the Coulomb friction law, but sudden changes in
contact pressure do not cause instantaneous changes in frictional traction. Instead,
the new value is approached asymptotically over a sliding distance related to L. This
law was proposed based on a series of elegant experiments by Prakash (1998), in
which sudden changes in normal and shear tractions were imposed by causing a
shock wave to impinge on an oblique surface.
19.6 Transient Elastodynamic Contact Problems
Most of the discussion so far has concerned elastodynamic problems in which the
stress and displacement fields are invariant in an appropriate moving frame of refer-
ence. More general transient problems can be formulated using as Green’s function
the fields due to an impulsive concentrated force. This solution was given by Lamb
(1904) for both two and three-dimensional geometries.7
19.6.1 Impulsive Line Force
If the surface of an elastic half-plane is subject to the transient pressure distribution
p(x, t) = Pδ(t)δ(x), (19.107)
the subsequent surface displacement can be written

P |x|
u z (x, 0, t) = − F1 t > 0, (19.108)
Gt t
where the function F1 (·) is defined in Eq. (19.14) and shown in Fig. 19.1.
Notice that the argument of this function is |x|/t, which can be interpreted as the
uniform speed that a disturbance would need to have in order to travel a distance x in
time t. Thus if c1 < |x|/t, or equivalently |x| > c1 t, a dilatational wave starting from
the impulsive force has insufficient time to reach the point x, so F1 (|x|/t) = 0 and
the surface is still undisturbed. Points in the range c1 > |x|/t > c2 , will experience
displacements due to the dilatational wave, but not due to the shear wave. Also, we
note that a singularity associated with the Rayleigh wave will pass the point x at time
t = |x|/c R .
19.6.2 A Uniform Pressure Suddenly Applied
Before discussing more general problems, it is instructive to consider the case in

which an initially quiescent half-plane is subject to a uniform pressure p(x, t) =
p0 H (t), over the entire surface z = 0. This problem is clearly one-dimensional, and
7 See also Eringen and Suhubi (1975), Sects. 7.16 and 8.6.
19.6 Transient Elastodynamic Contact Problems 505
it is easily shown as in Sect. 19.1 that the displacement and velocity fields are defined
by
p0 p0
u z (x, z, t) = − (z − c1 t)H (z − c1 t); u̇ z (x, z, t) = H (z − c1 t). (19.109)
ρc12 ρc1
Suppose now that the uniform pressure is applied only over some finite portion
A of the surface. At time t, the fields at some point x will be influenced only by
the pressures at points within a distance c1 t of x—in other words, regions that are
accessible to a dilatational wave starting from x at time t = 0. If for some particular
x, t, all such regions are within A (and hence experiencing uniform pressure p0 ),
then the fields at x are identical with those due to a uniform pressure over the entire
surface. In other words, at sufficiently small time t, Eq. (19.109) will be exact for all
x ∈ A except in a region adjacent to the boundary of A of width c1 t.
Similar arguments show that if a non-uniform pressure p(x, t) = p0 (x)H (t) is
applied to the surface, the initial surface velocity is proportional to the pressure and
is given by
p0 (x)
u̇ z (x, 0, 0) = , (19.110)
ρc1
since at t = 0, c1 t = 0 and the local field is influenced only by the local pressure.
19.6.3 Integral Equation Formulation of the Transient

Contact Problem
Suppose now that a more general time-dependent contact pressure p(x, t), t > 0 is
applied to an initially quiescent half-plane over some area A(t) which itself might be
time-dependent. We generalize the pressure function to the entire surface by defining
p(x, t) = 0, x ∈
/ A(t), after which the normal surface displacement for t > 0 can be
written down by convolution of (19.108) as
t ∞
1 p(ξ, τ ) |x − ξ|
u z (x, 0, t) = − F1 dξdτ t > 0, (19.111)
G 0 −∞ (t − τ ) t −τ
or equivalently
t ∞
1 |x − ξ| dξdτ
u z (x, 0, t) = − p(ξ, τ ) f t > 0, (19.112)
G 0 −∞ t −τ |x − ξ|
where we define
f (s) ≡ s F1 (s). (19.113)
Also, noting that

∂ |x − ξ| |x − ξ|
=− , (19.114)
∂t t −τ (t − τ )2
the surface velocity can be written

t ∞

1 |x − ξ|
dξdτ
u̇ z (x, 0, t) = p(ξ, τ ) f
G 0 −∞ t − τ (t − τ )2
∞
1 |x − ξ| dξ
− p(ξ, t) f . (19.115)
G −∞ t −τ |x − ξ| τ =t
The kernel of the integral in the second term has a delta function form in the limit
τ = t, since the function f (|x −ξ|/(t −τ )) is then zero everywhere except at ξ = x.
Using the change of variable s = |x −ξ| and integrating the first term by parts, we
obtain the simpler expression
t ∞
1
u̇ z (x, 0, t) = − p (x + s(t − τ ), τ ) − p (x − s(t − τ ), τ ) f (s)dsdτ
G 0 0

2 p(x, t) ∞
− F1 (s)ds, (19.116)
G 0
where
∂ p(x, t)
p (x, t) = . (19.117)
∂x
The integral in the last term of (19.116) could be evaluated using Eq. (19.14), but a
simpler procedure is to consider the special case p(x, t) = p0 H (t) for which the first
term in (19.116) is zero. Using u̇ z (x, 0, t) from (19.109) and cancelling a factor of
p0 , we then obtain ∞
G c2
F1 (s)ds = − =− 2 . (19.118)
0 2ρc1 2c1
19.6.4 Normal Indentation by a Rigid Body
Equation (19.116) can be used to formulate problems in which a rigid punch is

pressed at prescribed speed into an elastic half-space. For example, for a flat punch
of width 2a indenting at constant velocity V , we require u̇ z (x, 0, t) = V , −a < x < a
and hence
t ∞
c22
p(x, t) = GV + p (x + s(t − τ ), τ ) − p (x − s(t − τ ), τ ) f (s)dsdτ
c1 0 0
(19.119)
for −a < x < a, t > 0 and p(x, t) = 0 for |x| > a.
Equation (19.119) resembles a Volterra integral equation in the time domain for
the unknown pressure distribution p(x, t) and in a formal sense it defines the pressure
at time t in terms of the corresponding expressions for τ < t. For example, if the
domain is discretized through a set of nodal points xi , t j and if p(x, t) is written in
the discrete form

M
N
p(x, t) = Pi j vi (x)v j (t), (19.120)
i=1 j=1
where vi (x), v j (t) are suitable [e.g. linear] shape functions, Eq. (19.119) will define
a set of explicit equations for the nodal forces Pi N in terms of Pi j , 1 ≤ j ≤ N −1.
If the indenting body is not plane, the contact area A(t) will change with time, but
the satisfaction of the unilateral inequalities is simplified by the hyperbolic nature
of the governing equations. Suppose the indenting body is defined by an initial gap
function g0 (x), so that the instantaneous gap is
g(x, t) = g0 (x) − V t + u z (x, t) (19.121)
and the contact problem is defined by the conditions
g(x, t) = 0; p(x, t) ≥ 0 x ∈ A(t) (19.122)

p(x, t) = 0; g(x, t) > 0 x∈/ A(t). (19.123)
Suppose the problem is solved up to some time t, and we make the tentative assump-
tion that the contact area remains unchanged during the next small time increment
δt. We can then determine p(x, t + δt) in A(t) from (19.119) and g(x, t + δt) for
x∈ / A(t) from Eqs. (19.121), (19.112). Any point where p(x, t +δt) < 0 must then
transition to separation, and any point where g(x, t+δt) < 0 must transition to contact
during δt. Since elastic waves have a finite propagation speed, these determinations
are purely local—i.e. independent of the conditions elsewhere at the interface.
19.6.5 Superseismic Indentation
If the contact area expands at a speed exceeding the dilatational wave speed, the
material outside the contact area will be uninfluenced by the contact pressure and
hence will remain undeformed. In this case8 the contact area is coincident with the
‘interference area’—i.e., the area in which Eq. (19.121) defines a negative gap with
u z (x, t) = 0.
For example, if a cylindrical body of radius R indents the half-plane at constant
velocity V , we have g0 = x 2 /2R and hence the contact area is defined by −a(t) ≤
x ≤ a(t), where
a2 √
= Vt or a(t) = RV t. (19.124)
2R
8 Thereader will notice a parallel here with the superseismic moving punch problem [Sect. 19.2.7],
where material ahead of the punch is undisturbed and the leading edge of the contact area is
determined from the interference area.
In this case, the boundary velocity remains superseismic as long as

1 RV RV
ȧ = > c1 or t< . (19.125)
2 t 4c12
If this condition is satisfied, the normal surface displacement u z (x, 0, t) is known

for all x, t and hence the solution can be written down as a convolution integral of
the solution to the problem where u̇ z (x, 0, t) = δ(x)H (t). Thompson and Robinson
(1977) use this method to solve three-dimensional superseismic contact problems
and give explicit expressions for the corresponding Green’s function.
19.6.6 Self-Similar Indentation Problems
If, the displacement field u(t) is assumed to take the form

x y z
u(x, y, z, t) = t n U (ξ, η, ζ) where ξ= ; η= ; ζ= , (19.126)
t t t
substitution into Eq. (19.3) shows that the powers of t cancel, leaving a homogeneous
partial differential equation for U(ξ, η, ζ). Fields of this form are described as self-
similar (Willis 1973), since their states at any given time can be mapped into those at
any other time with an appropriate length scaling. In effect, the fields ‘expand’ at a
uniform rate whilst retaining the same form. Lamb’s solution (19.108) is a particular
example of a self-similar field.
Indentation problems are self-similar if the kinematic boundary conditions can be
expressed in the form (19.126). A simple example is the rigid wedge of Fig. 19.13,
pressed into an elastic half-plane at constant velocity V , for which the boundary
conditions are
u z (x, t) = V t − α|x|; p(x, t) ≥ 0 − a(t) < x < a(t) (19.127)

p(x, t) = 0; u z (x, t) ≥ V t − α|x| |x| > a(t), (19.128)
where we have assumed that α 1.
Fig. 19.13 The rigid wedge

indenting at constant speed V V
α
a x
Defining U as in (19.126), with n = 1, we have
Uz (ξ, 0) = V − α|ξ|; p(ξ) ≥ 0 − ã < ξ < ã (19.129)

p(ξ) = 0; Uz (ξ, 0) ≥ V − α|ξ| |ξ| > ã, (19.130)
where a(t) = ãt increases linearly with time (Robinson and Thompson 1974). An
exactly parallel argument can be applied to indentation by an axisymmetric cone.
If V > αc1 , both two-dimensional and axisymmetric problems are superseismic as
defined in Sect. 19.6.5, and ȧ = ã = V /α. In this case, the contact pressure distribution
exhibits a singularity at the ‘Rayleigh point’ x = c R t. Bedding and Willis (1973,
1976) considered both two-dimensional and axisymmetric problems for the case
where there is no slip in the contact region.
19.6.7 Three-Dimensional Transient Problems
The discussion so far has focussed mainly on two-dimensional problems in the inter-
ests of mathematical simplicity, but many of the results carry over to the three-
dimensional case. Lamb (1904) gives the solution for an impulsive point force
Pδ(x)δ(y)δ(t) and simpler forms of his results, including the extension to a point
force that is a step function in time Pδ(x)δ(y)H (t) were given by Pekeris (1955) and
Richards (1979). As with the moving point force of Sect. 19.2.8, the development of
a closed-form solution depends on the factorization of the Rayleigh function (19.46).
Figure 19.14 shows the normal surface displacement u z for the step function force
Pδ(x)δ(y)H (t), using closed-form expressions from Pekeris (1955) for the case
ν = 1/4. The displacements are here normalized by the corresponding quasi-static
result of Eq. (2.7) in order to expose that for distances r < c R t from the force,
the quasi-static profile has already been established. Richards (1979) shows that
this result applies for arbitrary Poisson’s ratio, and also gives expressions for the
Fig. 19.14 Displacements due to a concentrated force P starting at time t = 0, for the case ν = 1/4.
The dashed vertical line identifies the singularity at r = c R t
horizontal and vertical surface displacements due to a both horizontal and vertical
concentrated forces.
More generally, we deduce that if a half-space is subjected to the step function
loading
p(x, y, t) = p0 (x, y)H (t) (x, y) ∈ A, (19.131)
the normal surface displacements in A will achieve their quasi-static values at time
t = L/c R , where L is the length of the longest straight line that can be inscribed
completely within A.
At short times, we can also generalize the result from Sect. 19.6.2 to state that
if a rigid flat punch of arbitrary planform A is pressed into an elastic half-space at
velocity V (t), the contact pressure is initially given by
p(x, y, 0, t) = ρc1 V (t), (19.132)
for all points for which the shortest distance s from the boundary of A satisfies the
condition s > c1 t. Similarly, if a non-uniform pressure p(x, y, t) = p0 (x, y)H (t) is
applied to the surface, the initial surface velocity is given by p0 (x, y)/ρc1 .
Problems
1. If a displacement field u is irrotational [∇×u = 0], it can be represented as the

gradient of a scalar potential
u = ∇φ.
Show that a general irrotational wave moving in the x-direction can be described by
a function φ(x −c1 t, y, z) satisfying the equation
∂2φ ∂2φ
+ 2 = 0,
∂ y2 ∂z
where c1 is given by Eq. (19.6).

2. Table 19.1 shows representative values for the elastic constants E, ν and density ρ
for a selection of materials. Estimate the corresponding values of the dilatational and
shear wave speeds c1 , c2 . Comment on the implications of your results for practical
applications. For example, would you expect elastodynamic effects to be significant
in the interaction between a car tyre and the road, or the impact between a ship and
an iceberg.
3. A rigid body with the sinusoidal profile of Fig. 6.6 is pressed against an elastic
half-plane by a mean traction p̄ that is just sufficient to ensure that half of the interface
Problems 511
Table 19.1 Physical properties of various materials

Material Carbon Brass Compact Nylon Glass Granite Ice Aluminium Rubber
steel bone alloy
E 210 110 13.8 3.0 70 49 9.0 72 0.10
(GPa)
ν 0.3 0.33 0.42 0.4 0.24 0.28 0.33 0.31 0.49
ρ 7700 8500 1300 1650 2500 2700 990 2800 1100
(kg/m3 )
is in contact—i.e. a = L/4. The body now slides without friction over the surface at
speed V . Find the minimum value of V < c R for which full contact will occur.
4. Use the Smirnov–Sobolev transform of Sect. 6.6 to prove Churilov’s result (19.43)
for the steady-state surface displacement due to a point force P moving at speed
V < c R over the surface of an elastic half-space.
5. An SH-pulse described by the displacement
2
u y = C a 2 − x − c2 t − a < x − c2 t < a
impinges on a plane interface between two half-planes of identical materials, where

the coordinate direction x is defined in Fig. 19.8 and the angle of incidence is α. The
half-planes are pressed together by a uniform normal pressure p0 , but no external
shear tractions are applied. If the coefficient of friction is f , find the maximum value
of C [= C0 ] for which there is no slip. Then find the shift h 0 caused by the passage
of the pulse, if C = 2C0 .
6. Two granite rock masses can be approximated by half-spaces that are pressed
together by a uniform pressure of 10 MPa. An SH-pulse impinges on the interface
with an angle of incidence of 55◦ . The stresses at a given point associated with the
incident pulse have the step function form
σx y (t) = 45 [H (t) − H (t + 0.5)] MPa
in the coordinate system x y aligned with the direction of propagation of the pulse
[see Figure 19.8], where t is time in seconds. Find the shift at the interface due to
the pulse if the coefficient of friction is 0.6. The mechanical properties of granite
are given in Table 19.1 above. How would your answer change if the interface also
transmits a uniform tangential traction of 3 MPa.
7. Use the Green’s function (19.61) to find the displacement derivative ∂u y /∂x due
to the traction distribution

q y = q0 exp ı
ω (x − ct)
moving at subseismic speed c < c2(2) over the surface of the half-plane z < 0.
Use this result and the arguments of Sect. 19.3.1 to obtain the bilateral solution
for an incident wave defined by Eq. (19.52), and hence verify (i) that the reflected
wave has the same amplitude as the incident wave, and (ii) that the phase lag is given
by (19.81).
8. A frictionless rigid power-law punch defined by the gap function g0 (x) = C|x|m
indents an elastic half-plane at a speed V (t) which is a function of time t. State the
kinematic boundary conditions and hence show that the problem is self-similar if
and only if V (t) ∼ t m−1 .
9. Bedding and Willis (1976) report that the maximum ratio of tangential to normal
tractions in the ‘no slip’ solution to the superseismic wedge indentation problem is
0.16. Suppose that the coefficient of friction f < 0.16, so that we must anticipate
regions of slip and stick. State the boundary conditions for the problem, including
inequalities, and hence show that the resulting problem is also self-similar.
Chapter 20
Impact
Numerous engineering applications involve the impact of two bodies (Goldsmith

1960; Stronge 2000). In the field of dynamics, bodies are generally assumed to be
rigid, so that the impact occurs during an infinitesimal period of time. This implies
a delta function loading P(t) = I δ(t). In other words, the contact force is strictly
infinite at the moment of impact, but an impulse

I = P(t)dt (20.1)
is transmitted corresponding to the exchange of momentum between the bodies.

If one or both of the bodies are deformable, the contact force P(t) must necessarily
be always finite, and the impact takes a finite amount of time. A simple case would
be a rigid body of mass M impacting on a spring of stiffness k, for which the motion
is sinusoidal in time. In particular, the bodies remain in contact for a half-period

M
t0 = π , (20.2)
k
which is independent of the incident velocity V0 . The maximum contact force Pmax
in the spring occurs when the velocity and hence the kinetic energy is zero, so that
M V02 P2 √
= max or Pmax = V0 k M. (20.3)
2 2k
Notice that there is no dissipation in this problem, so the rebound velocity is equal to
the incident velocity. In most practical cases, some energy is lost, and this is described
by dynamicists through the concept of a coefficient of restitution.
There are many possible sources of energy dissipation. Stresses associated with
contact problems are generally high, particularly in impact problems, so plastic

https://doi.org/10.1007/978-3-319-70939-0_20
514 20 Impact
deformation is likely. Energy may also be lost due to viscous effects, either at the
interface or internally to the materials, and there may be losses associated with fric-
tional slip at the interface. Finally, waves will propagate through the elastic bodies
as a result of the impact, so that these are generally left in a state of vibration. For
these reasons, the coefficient of restitution is really little more than a ‘cooking factor’
since its value will change from problem to problem.
A rigorous statement of the impact problem requires that it be formulated in the
context of elastodynamics. Elastic waves will emanate from the contact region and
later reflect from the distant boundaries of the body, so the resulting problem can
be very challenging, both in analytical and numerical formulations. To avoid these
difficulties, we can develop approximate solutions in the time domain by replacing
the elastic continuum by a system of rigid bodies connected by springs, or in the
frequency domain, by restricting the possible deformations of the body to a super-
position of a finite set of normal modes (Stronge 2000).
20.1 Hertz’ Theory of Impact
Hertz (1882) gave an approximate solution to the problem of impact of two spheres
by treating the spheres as rigid bodies and using the Hertzian contact theory to define
the characteristics of a nonlinear spring through which they interact. A heuristic jus-
tification for this approximation is that the elastic contact area is very small compared
with the dimensions of the body, so the mass of material in the region experiencing
significant deformations is small compared with that of the impacting bodies and
hence might reasonably be neglected.
Here, we shall discuss the case where a single elastic sphere of radius R impinges
on a rigid plane surface with initial normal velocity V0 . The more general case of
spheres of mass M1 , M2 and radii R1 , R2 can be subsumed under the following
analysis by defining
1 1 1 1 1 1
= + ; = + (20.4)
M M1 M2 R R1 R2
and interpreting u, V as the relative approach and velocity of the spheres, respec-
tively.
We recall from Sect. 5.2 and Eqs. (5.30), (5.32) that a sphere of radius R pressed
into a rigid plane by a force P will generate an indentation Δ, a circular contact
region of radius a and a contact pressure distribution p(r ) defined by
∗√ 2 ∗
2E a − r2 a2 4E a 3
p(r ) = ; Δ= ; P= , (20.5)
πR R 3R
where the composite modulus for a sphere impacting a rigid plane reduces to
20.1 Hertz’ Theory of Impact 515
∗ E 2G
E = = . (20.6)
(1 − ν )
2 (1 − ν)
Eliminating a between the last two of Eq. (20.5), we obtain

∗
4E R 1/2 Δ3/2
P= , (20.7)
3
which defines the characteristic of the nonlinear ‘contact spring’.
If the bulk of the sphere is regarded as a rigid body, the displacement of its centre
of mass in the direction perpendicularly towards the plane is u = Δ, relative to its
location at the beginning of the impact, so we can use (20.7) to write the equation of
motion for the sphere as
∗
4E R 1/2
M ü = −P = − u 3/2 . (20.8)
3
Writing V = u̇ for the instantaneous velocity, so that ü = V d V /du, we can integrate

this equation, obtaining
∗
16E R 1/2
V 2 = V02 − u 5/2 , (20.9)
15M
where we have used the initial condition V (0) = V0 to determine the constant of
integration. The same result can of course be obtained by requiring
that the sum of
the kinetic energy of the sphere and the strain energy U = Pdu of the spring be
constant.
The maximum indentation occurs when V = 0, at which point
2/5 ∗2 1/5
15M V02 125M 3 V06 E R
u max = ; Pmax = , (20.10)
16E ∗ R 1/2 36
and the maximum contact pressure is

∗ 1/5
E 30M V02
pmax = . (20.11)
π E ∗ R3
20.1.1 Duration of the Impact
During the first half of the impact, V > 0 and Eq. (20.9) can be written
u
V = u̇ = V0 1 −
u 5/2 where
u= , (20.12)
u max
516 20 Impact
from which
u max d
u u max
t= √ = ũ F 25 , 21 ; 75 ; ũ 5/2 , (20.13)
V0 1 −u 5/2 V0
where F(α, β; γ; x) is the hypergeometric function (Gradshteyn and Ryzhik 1980).

In particular, the time taken to reach the maximum indentation is
u max 2 1 7
1.4716 u max
t0 = F 5, 2; 5; 1 = . (20.14)
V0 V0
During unloading, we must take the negative sign of the square root in (20.12), but
the trajectory is exactly the same as during the loading phase, so the total period of
contact is 1/5
M2
tc = 2t0 = 2.868 , (20.15)
E ∗ 2 V0 R
after substituting for u max from (20.10).

Figure 20.1 shows the normalized displacement u/u max as a function of time. The
dashed red line in this figure represents a sine wave with the same amplitude and
period, which is clearly a very good approximation to the exact curve. One might
therefore be tempted to develop an approximate solution based on the assumption of
a constant contact stiffness k. However, if Eqs. (20.15), (20.2) are used to define an
effective value of k, it will be found to increase with V0 because of the nonlinearity
of the actual contact stiffness. Another consequence of this is that the contact time
decreases with incident velocity [it would be independent of velocity in the linear
case], as can be seen from Eq. (20.15). However, this dependence is very weak. In
his original 1882 paper, Hertz calculated rather whimsically that if two steel spheres
of the size of the Earth were to impact at an initial relative velocity of 10 mm/s, the
impact would last for 27 h!
Fig. 20.1 Trajectory of the

sphere during impact. The
dashed red line represents a
sine wave with the same
amplitude and period
20.1 Hertz’ Theory of Impact 517
20.1.2 Homogeneous Sphere
If the sphere is homogeneous and of density ρ, the mass
4π R 3 ρ
M= (20.16)
3
and the duration of the contact can be written
1/5
ρ2
tc = 5.087R . (20.17)
E ∗ 2 V0
∗
Using Eqs. (19.6), (20.6) to eliminate E , ρ, we can write
2/5 1/5
tc (1 − ν)2 c1
= 5.087 , (20.18)
t1 1 − 2ν V0
where
t1 = c1 R (20.19)
is the time taken for a dilatational wave to traverse the distance from the surface to
the centre of the sphere. Using the same change of variables on Eq. (20.11), we can
write the maximum contact pressure as
1/5 2/5
pmax 40(1 − ν)2 V0
∗ = . (20.20)
E π (1 − 2ν)
4 c1
20.1.3 Range of Validity of the Theory
Hertz’ theory of impact combines a quasi-static treatment of the contact problem with
a rigid-body dynamics solution for the impacting sphere(s). We argued in Chap. 19
that the quasi-static approximation is reasonable if the forces vary slowly relative to
the elastodynamic timescale of the problem, which here might be associated with
the period of the lowest natural frequency of a vibrating elastic sphere, or the time
taken for an elastic wave to traverse the body from side to side and back, here given
by 4t1 .
This criterion and Eq. (20.18) suggest that the Hertz theory of impact is reasonable
as long as the approach velocity V0 is much lower than the dilatational wave speed c1 .
This condition is likely to be satisfied in most practical cases and indeed if V0 /c1 is not
small compared with unity, the maximum contact pressure (20.20) will imply large
strains and hence plastic deformation in ductile materials. The numerical multiplying
factor in this equation is of order unity, so (for example) a velocity V0 = 10−5 c1
518 20 Impact
∗
implies pmax ∼ 0.01E and tc /t1 ∼ 50. However, large strains are admissible and
indeed likely in the impact of a rubber ball against a plane.
Johnson (1985) points out that the ‘wave propagation’ argument fails if a rigid
sphere impacts an elastic half-space, since in that case no reflections can ever occur.
A more direct criterion was suggested by Hunter (1957), who argued that the quasi-
static solution is appropriate if the period of loading tc is long compared with the time
taken for the surface displacements due to a suddenly applied pressure p0 (x, y)H (t)
to approximate their quasi-static values. We showed in Sect. 19.6.7 that the displace-
ments due to a suddenly applied force satisfy this condition as soon as there is time
for a Rayleigh wave to traverse the contact area, so the quasi-static solution should
give a good approximation to the elastodynamic surface displacements in the contact
region as long as tc 2amax /c2 . Using the parameters from the Hertzian solution,
this can be shown to lead to the criterion V0 c2 independently of the radii of the
impacting spheres.
20.1.4 The Superseismic Phase
We saw in Sect. 19.6.5 that if a cylindrical rigid body indents an elastic half-plane
at constant speed V , the contact remains superseismic as long as t < RV /2c12 . In
the impact problem, the approach velocity is not constant, but there must still be a
period during which the contact is superseismic. During
√ this period, the contact area
is defined by the ‘interference’ condition, so a(t) = 2Ru(t) and the normal surface
displacement
r2
u z (r, 0, t) = u(t) − 0 ≤ r < a(t) (20.21)
2R
=0 r > a(t), (20.22)
where u(t) is the as yet unknown displacement of the sphere centre. This phase of the
contact was analysed by Borodich (2000), who gave expressions for the contact force
and indentation velocity as functions of time. Argatov (2008) proposed a modification
to the Hertz theory in which Borodich’s results were used for the superseismic phase,
followed by a quasi-static solution for the subsequent motion. His results show that
the effect of the superseismic phase is significant only if the parameter
3/2
V0 πρR 3
α= (20.23)
c1 M
is relatively large compared with unity. For a homogeneous sphere, this again requires
an impact velocity comparable with or greater than c1 .
20.2 Impact of a Cylinder 519
20.2 Impact of a Cylinder
Similar methods can be used to analyse the two-dimensional problem of an elastic

cylinder impacting a rigid plane. However, we note at the outset that the highly
stressed material near the contact region represents a larger proportion of the total
volume of the body than in the axisymmetric case, so we must anticipate a smaller
range of validity for the quasi-static theory.
The corresponding Hertz theory does not provide an expression for the nonlin-
ear stiffness of the contact, since quasi-static displacements for the half-plane are
unbounded. However, we can use the methods discussed in Sects. 6.1 and 6.7 to
develop an approximate load–displacement relation. Arguably the most appropriate
approximation would balance the contact force by a uniform body force, but here
we shall use the simpler expression (6.114), since the exact choice of reaction only
affects the constant term in the approximation. We have

P 4R
u= 2 ln − 1 , (20.24)
πE∗ a
where R is the radius of the cylinder,

∗
π E a2
P= (20.25)
4R
from Eq. (6.24), and a is the semi-width of the contact area. Eliminating a between
these equations, we obtain
u
û ≡ = 4 P̂ − ln P̂ − 1 , (20.26)
R
where
P
P̂ = . (20.27)
4π E ∗ R

contact force and
displacement for contact of a
cylinder [Eq. (20.26)]
520 20 Impact
We also note that the maximum contact pressure p0 is given by
p0 a
= = 2 P̂, (20.28)
E∗ 2R
from (6.25), (20.27), so practical values of P̂ for metals cannot exceed about 10−4 ,
∗
for which p0 /E = 0.02. The relation (20.26) is shown in this range in Fig. 20.2, from
which we conclude that the contact spring is close to linear, despite the increase of
contact semi-width with P. The maximum contact force can be determined as in the
three-dimensional case by equating the work done during loading W to the initial
kinetic energy M V02 /2. Using the result
u max Pmax
W = Pdu = Pmax u max − ud P (20.29)
0 0
and Eq. (20.24), we obtain the expression

MV 2
− 8π P̂max
2
2 ln P̂max + 3 = ∗ 02 , (20.30)
E R
which is plotted in Fig. 20.3.
The duration of the impact can be found as in Sect. 20.1.1, with the result
P̂max
M d û d P̂
tc = , (20.31)
E∗ 0 d P̂
2π f P̂max − f P̂
where û is given by (20.26) and

f P̂ = P̂ 2 −2 ln P̂ − 3 . (20.32)
Fig. 20.3 Maximum

dimensionless contact force
P̂max as a function of initial
kinetic energy
20.2 Impact of a Cylinder 521
The integral in (20.31) can be evaluated numerically and it varies only in the range
−7 −4
for ∗10 < P̂max < 10 . Thus, a good estimate of the contact period is tc ≈
6±1
6 M/E .
For a homogeneous cylinder of density ρ, we obtain tc ≈ 6R/c2 , so reflected waves
start to impinge on the contact area during the rebound phase of the impact. Thus the
quasi-static solution can only be regarded as a first approximation in this case. The
full elastodynamic solution, including wave reflections from the curved surfaces, is
not analytically tractable, but the deformations in the contact region before reflected
waves arrive have been analysed by Adda-Bedia and Llewellyn Smith (2006).
20.3 Oblique Impact
Figure 20.4 shows a sphere of radius R impacting a half-space with an incident

velocity V 0 and an angle of incidence α. If Dundurs’ constant β = 0, the normal
and tangential contact problems are uncoupled and the normal velocity Vz (t), the
normal contact force P(t) and the normal displacement u z (t) will be defined by the
equations in Sect. 20.1, but with the initial normal velocity V0 replaced by |V 0 | cos α.
If the contact is frictionless, the tangential velocity Vx of the sphere will remain equal
to |V 0 | sin α throughout the impact. However, if Coulomb friction conditions apply
at the interface, we anticipate some tangential tractions and hence a time-varying
tangential force Q(t).
If Q(t) were known, we could define the trajectory in P Q-space, as in Sect. 9.3
and use the Ciavarella–Jäger theorem to determine the resulting stick and slip zones,
and the tangential elastic displacement u x (t). The inversion of this procedure, if
feasible, would define a contact mechanics relation between Q(t) and u x (t).
In general, the results of Sect. 9.3 show that at any given time the contact will
comprise one of the three states
Fig. 20.4 Oblique impact of

a sphere on a plane
R
V0
x
z
522 20 Impact
Fig. 20.5 Free-body

diagram of the sphere under
tangential contact loading Q Ω
R
U
u
Q
1. Complete adhesion if d|Q|/d P < f .

2. Partial slip with a central circle of stick and a surrounding annulus of microslip.
3. Slip throughout the contact circle (gross slip), implying Q = ± f P.
20.3.1 The Equation of Motion
A second relation between Q(t) and u x (t) is defined by the tangential equation of
motion. We must first distinguish between the tangential displacement of the centre
of the sphere U and that of the point of contact u as shown in Fig. 20.5.
If the sphere were rigid, the sliding velocity at the contact point would be
Vx = U̇ + RΩ, (20.33)
where Ω is the angular velocity. An opposing tangential force Q produces

accelerations
Q QR
Ü = − ; Ω̇ = − , (20.34)
M I
where I is the moment of inertia of the sphere about its centre, so

1 R2
V̇x = −Q + . (20.35)
M I
In particular, for a homogeneous solid sphere,
2M R 2 7Q
I = and hence V̇x = − . (20.36)
5 2M
If the sphere is elastic, the force Q will produce tangential displacements u x in
the contact zone, so that the sliding velocity is modified to (Vx − u̇ x ). In any stick
20.3 Oblique Impact 523
zone Astick , this must be zero, as must its time derivative, so

1 R2
ü x = −Q + (x, y) ∈ Astick . (20.37)
M I
20.3.2 The Tangential Contact Problem
The tangential tractions are conveniently defined using the incremental approach of
Sect. 7.7.1. We recall from Sect. 7.6.1 that the tangential traction
Q
qx (r ) = √ 0 ≤ r < a, (20.38)
2πa a 2 − r 2
corresponds to a total force Q and produces a rigid-body translation of the loaded

circle given by
RT Q
ux = , (20.39)
2E ∗ a
where we have used (7.58), (20.6) to write this result in the same notation as (20.5).
A more general traction distribution can then be defined by superposing distri-
butions of the form (20.38), treating the radius a as a transform parameter s. A
convenient form is
∗ a(t) ∗ a(t)
2E f sg(s, t)ds 4E f
qx (r, t) = √ ; Q(t) = s 2 g(s, t)ds. (20.40)
πR r s2 − r 2 R 0
(Barber 1979), where the function g(s, t) is allowed to vary with time as well as s.
We also define a normalized tangential velocity ψ as
Vx
ψ= , (20.41)
f RT V0
where the rigid-body velocity Vx is defined by (20.33). The value of ψ at incidence

is denoted by ψ0 .
20.3.3 Complete Stick
If g(s, t) = g(s) is independent of time, the time derivatives of (20.40) take the form
∗ ∗
2E f ag(a)ȧ 4E f a 2 g(a)ȧ
q̇x (r ) = √ ; Q̇ = . (20.42)
πR a2 − r 2 R
524 20 Impact
The corresponding time derivative of the normal force P is obtained from (20.5)3 as
∗
4E a 2 ȧ
Ṗ = (20.43)
R
and hence the stick condition d|Q|/d P < f is satisfied during the approach phase
(ȧ > 0) if and only if |g(a)| < 1.
The contact radius a is given by (20.5)2 , from which
a 2 = Ru z , (20.44)
where the normal displacement of the centre of the sphere [or equivalently the normal
elastic displacement at r = 0] is here denoted by u z to avoid confusion with the
tangential displacement u x . Differentiating this expression with respect to t, we
obtain

a 5
2a ȧ = R u̇ z = RV = RV0 1 − u = RV0 1 −
5/2 (20.45)
amax
from (20.44), (20.12), (20.10), where

1/5
15M V02 R 2
amax = . (20.46)
16E ∗
Substituting these results into (20.42), we obtain

∗ 5
E f V0 g(a) a 5 ∗ a
q̇x (r ) = √ 1− ; Q̇ = 2E f V0 ag(a) 1 − .
π a2 − r 2 amax amax
(20.47)
and in particular, the incremental tractions have the form of (20.38). It then follows
from (20.39) that the corresponding incremental displacements (displacement rate
or velocity u̇ x ) are uniform throughout 0 ≤r < a and given by

5
a
u̇ x = f RT V0 g(a) 1 − . (20.48)
amax
We conclude that the representation (20.40) satisfies the kinematic condition for
complete stick within the instantaneous contact area 0 ≤r < a(t), provided g(s) is
independent of time. Also, since u̇ x = Vx during stick periods, we can write

5
a
ψ = g(a) 1 − , (20.49)
amax
from (20.41).
The governing equation for g(s) can then be found by (i) differentiating the equa-
tion of motion (20.37) with respect to time, (ii) substituting for Q̇ from (20.42)
and (iii) equating the resulting expression for ∂ 3 u x /∂t 3 to the second derivative of
(20.48) with respect to time. In each differentiation, we note that during the first
half of the impact process, a is a monotonically increasing function of t, so we can
express everything in terms of a through the operation
d d
= ȧ , (20.50)
dt da
where ȧ is given by (20.45). We obtain
d2h dh
10ξ(1 − ξ) + 3(2 − 7ξ) + 3(2χ − 1)h = 0, (20.51)
dξ 2 dξ
(Barber 1979), where

5
a 1 M R2
ξ= ; h(ξ) = g(a); χ = 1+ . (20.52)
amax 2RT I
The two independent solutions of (20.51) are
h 1 (ξ) = F(α, β; γ; ξ); h 2 (ξ) = ξ 2/5 F(α − γ + 1, β − γ + 1; 2 − γ; ξ),

(20.53)
where √
11 ± 1 + 240χ 3
α, β = ; γ= , (20.54)
20 5
and F(α, β; γ; ξ) is the hypergeometric function (Gradshteyn and Ryzhik 1980).
The impact commences with a period of full stick if and only if the angle of
incidence is such that u̇ x (0) < f RT V0 or equivalently ψ0 < 1, where we have defined
the coordinate system to ensure that ψ0 > 0, without loss of generality. Consideration
of the conditions when a amax then show that the multiplier on h 2 (ξ) is zero and
the solution is given by
h(ξ) = ψ0 F(α, β; γ; ξ). (20.55)
This expression is unbounded as ξ → 1 and hence a → amax . It follows that the stick
condition |g(s)| < 1 will be violated when the contact radius has reached some value
a < amax and this violation will originate from r = a, indicating that an annulus of
microslip will be developed.
Substituting (20.55) into (20.40)2 , using (20.52) and performing the integral, we
find that the tangential force Q can be written
526 20 Impact
Q
= ψ0 ξ 3/5 F(α, β; 85 ; ξ), (20.56)
f Pmax
where α, β are defined in (20.54).
20.3.4 Gross Slip
Gross slip (sliding) corresponds to the case where g(s) = ±1 for all s. We then have
Q = ± f P and the tangential velocity can be determined by integrating the equation
of motion (20.37).
The impact starts with a period of gross slip if ψ0 > 1, in which case the subsequent
the horizontal velocity Vx is given by

M R2
Vx = Vx (0) − f (V0 − V ) 1 + , (20.57)
I
or equivalently
V
ψ = ψ0 − 2χ 1 − , (20.58)
V0
where χ is defined in (20.52), and the normal velocity V is given by Eq. (20.9). The
corresponding tangential elastic displacements are given by (20.48) with g(a) = 1,
so the time derivative

a 5
u̇ x = f RT V0 1 − = f RT V (20.59)
amax
and is uniform throughout the instantaneous contact area 0 ≤r < a. Gross slip termi-
nates when
V V 2χ − ψ0
u̇ x = Vx and hence ψ = or = . (20.60)
V0 V0 2χ − 1
This happens simultaneously throughout the contact area, so the system transitions
instantaneously from gross slip to complete stick. The minimum value of V is −V0
at the end of the impact, so no such transition is possible if ψ0 > 4χ−1. In this case,
the entire impact takes place in gross slip.
20.3.5 Partial Slip
During the partial slip phase, we anticipate an annulus of microslip b(t) <r < a(t),
in which
∗ √
2E f a2 − r 2
qx (r ) = ± , (20.61)
πR
and a central circle 0 ≤r < b(t) in which u̇ x is independent of r . These conditions
are satisfied if
g(s) = ±1 b<s<a (20.62)

ġ(s) = 0 0 ≤ s < b. (20.63)
The radius b of the stick circle decreases monotonically with time, so (20.63) implies
that g(s) retains its value from the preceding stick phase, even if this was only
momentary. Thus, the only unknown is the radius b(t), an equation for which can
be derived from the equation of motion (20.37). However, the resulting equation is
highly nonlinear, and probably not analytically tractable.
20.3.6 The Complete Trajectory
Maw et al. (1976) developed a numerical solution for the general problem by super-
posing a finite set of tangential tractions of Hertzian form over a set of concentric
circles. This is exactly equivalent to using a piecewise constant approximation to the
function g(s) in Eq. (20.40). The problem is completely characterized by the dimen-
sionless parameters ψ, Q/ f Pmax , which are unique functions of t/t0 for a given value
of χ.
Figure 20.6 shows the variation of the normalized tangential force Q/ f Pmax for
various angles of incidence, for the case χ = 1.44, which corresponds to a homoge-
neous sphere with ν = 0.3. The dashed lines for ψ0 = 0.2, 0.5 correspond to periods
of full stick and are plotted from Eq. (20.56). The remaining trajectories are taken
from Maw et al. (1976). Notice that when the contact starts in full stick, the tangential
force passes through almost a complete quasi-sinusoidal cycle during the impact. A
simple ‘explanation’ of this is that the normal and tangential stiffnesses are of the
same order, but the effective mass for the tangential problem is significantly reduced
due to the rotational term in the equation of motion (20.36).
For 1 < ψ0 < 2χ, the impact starts in gross slip, but transitions to complete stick
before the maximum normal force is achieved. The corresponding value of Q can
be determined using the general solution of Eq. (20.51) with appropriate initial con-
ditions. This segment is shown dashed in Fig. 20.6 for the case ψ = 1.2. It terminates
when Q̇ = − f Ṗ, at which point an annulus of slip develops with the opposite sign
to that during the initial gross slip phase.
If 2χ < ψ0 < 4χ − 1, gross slip terminates in the second half of the impact. Com-
plete stick then occurs but a reverse slip annulus starts to develop immediately. The
reader may recall that similar results were obtained during the normal unloading
phase of a quasi-static contact under periodic loading in Sect. 9.3.2.
528 20 Impact
Fig. 20.6 Variation of normalized tangential force during impact for χ = 1.44 and various angles
of incidence [from Maw et al. (1976)]. The impact commences with gross slip if |ψ0 | > 1, and with
complete stick if |ψ0 | < 1
For all cases in ψ0 < 4χ − 1, the microslip phase gives way to reverse gross slip,
with Q = − f P before the end of the impact.
20.3.7 Rebound Conditions
If gross slip persists throughout the impact, the tangential velocity at rebound is given
by Eq. (20.58) as
ψ1 = ψ(tc ) = ψ0 − 4χ. (20.64)
In the simpler rigid-body theory, only two states can be distinguished—gross

slip [|Q| = f P], and stick [|Q| < f P, Vx = 0] and contact always starts in gross slip
[except in the special case ψ0 = 0]. The assumption of gross slip at the end of the
impact is then reasonable if and only if the sphere is still sliding in the same direction,
implying that ψ1 > 0 and hence ψ0 > 4χ. For lower values of ψ0 , gross slip ceases at
some point during the impact and the rigid-body theory predicts that ψ1 = 0.
This idealized solution is shown as the dashed red line in Fig. 20.7. By contrast,
when Maw’s elastic solution is used to describe the contact, we obtain the solid
line in Fig. 20.7, which shows that for ψ0 < 4χ, the rebound is usually in the opposite
direction from the incident velocity. Also we note that the range of incident velocities
supporting gross slip throughout the impact is increased to ψ0 > 4χ − 1.
Fig. 20.7 Tangential velocity at rebound as a function of incident velocity for χ = 1.44. The dashed
red line represents the prediction of the rigid-body theory of impact
Maw et al. (1981) describe a series of experiments in which a disk cut from a
solid sphere is supported on an air table and caused to impinge on a steel flat. Their
experimental results agree closely with the above predictions.1
20.4 One-Dimensional Bar Problems
Most of the preceding discussion is restricted to quasi-static treatments of the contact

problem, and this is generally justified to some extent by the argument that the contact
region, and hence the mass of the stressed region, is small. When this condition is not
satisfied, exact analytical approaches are not usually practicable, and even numerical
solutions are very challenging because the moving wave fronts imply non-smooth
displacement fields.
One exception concerns problems in which the displacements are functions of
only one space variable. A simple example is illustrated in Fig. 20.8, in which a
uniform elastic bar of cross-sectional area A, length L and material properties E, ρ
strikes a rigid plane surface whilst moving at speed V . We assume that the horizontal
elastic displacement of the bar u x is a function of x, t only and that ‘plane stress’
conditions apply, so that the only non-zero stress component is
Fig. 20.8 A thin elastic bar V

x
impacting a rigid wall
1 Note that the disk has a different radius of gyration from the sphere and hence a different value of
χ is appropriate to these experiments.
530 20 Impact
∂u x
σx x = E . (20.65)
∂x
Substituting this relation into the first equation of motion (19.1), we obtain
E ∂2u x ∂2u x
= , (20.66)
ρ ∂x 2 ∂t 2
whose general solution is

E
u x (x, t) = f 1 (x − c0 t) + f 2 (x + c0 t) where c0 = . (20.67)
ρ
It then follows that the local velocity u̇ x and the axial stress σx x are given by
u̇ x = −c0 f 1 (x − c0 t) + c0 f 2 (x + c0 t) (20.68)
σx x = E f 1 (x − c0 t) + E f 2 (x + c0 t). (20.69)
The astute reader will notice that c0 is not one of the wave speeds identified
in Sect. 19.1, and this results from the approximation inherent in the plane stress
assumption. In fact, waves propagating down a bar with traction-free surfaces are
generally dispersive, meaning that their form becomes modified as propagation pro-
ceeds. Non-dispersive waves can be found in the sinusoidal form
u(x, y, z, t) = f ω (y, z) exp {ıω(x ± cω t)} , (20.70)
but the propagation speed cω depends on the wavenumber ω and only approaches
c0 in the limit where ωa → 0, where a is a representative dimension in the cross
section of the bar (Kolsky 1963). Thus, the approximation involved in Eq. (20.67)
is reasonable if the length scale associated with the functions f 1 , f 2 is very long
compared with a.
20.4.1 The Semi-infinite Bar
Before considering the problem of Fig. 20.8, it is instructive to consider the case of
the semi-infinite bar x > 0, initially at rest with u x (x, t) = 0, but with a step change
in velocity u̇ x (0, t) = V H (t) imposed on the end x = 0 at time t = 0. It is clear that a
wave of the form u x (x, t) = f (x −c0 t) will propagate along the bar at speed c0 and
the end condition and (20.68) require that
V
V H (t) = −c0 f (−c0 t) and hence f (x − c0 t) = − H (c0 t − x). (20.71)
c0
Substitution in (20.69) then shows that the stress field is defined by

20.4 One-Dimensional Bar Problems 531
Fig. 20.9 A semi-infinite F

σ xx = − σxx = 0
bar suddenly loaded at the x A
end F
c 0t
VE
σx x (x, t) = − H (c0 t − x). (20.72)
c0
In other words, the bar segment 0 < x < c0 t is in uniform compression, whilst the rest
of the bar is stress free. It follows that a force
VEA
F(t) = −Aσzz (0) = H (c0 t) (20.73)
c0
must be applied at the end, so Eq. (20.71) defines the solution of the problem of
Fig. 20.9, where a force F is applied to the end of the bar at time t = 0 and then
maintained constant.
Equation (20.71) shows that all particles in 0 ≤ x < c0 t are moving to the right
at speed V , whilst those in x > c0 t are not moving. We can therefore integrate this
equation in time to find the displacements as

x F(c0 t − x)
u x (x, t) = V t − H (c0 t − x) = H (c0 t − x). (20.74)
c0 EA
20.4.2 The Infinite Bar
Suppose two semi-infinite bars x > 0 and x < 0 respectively are each loaded by end
forces of magnitude F/2 in the positive x-direction. A compression wave will prop-
agate into the bar x > 0 as in Sect. 20.4.1, but the bar x < 0 is loaded in tension, so
a corresponding tension wave will propagate to the left, again at speed c0 . The two
problems are exactly similar except for sign and coordinate changes, so the velocities
of the ends of the two bars are the same and they will remain in contact.
σxx = 0 σxx = F σ xx = − F σxx = 0

2A x 2A
F
c 0t c 0t
Fig. 20.10 An infinite bar suddenly loaded at x = 0

532 20 Impact
0 1
-1 0 x /c 0t -1 0 x / c 0 t 1 -1 0 x /c 0t 1
-1 .
2A σxx 2EA ux 2EA ux
F c0F F
Fig. 20.11 Form of the expressions (20.75) for t > 0
It follows that the same solution applies to the infinite bar of Fig. 20.10 loaded by
a force F H (t) at x = 0. The state of the bar is then defined by
F
σx x (x, t) = − sgn(x)H (c0 t − |x|)
2A
Fc0
u̇ x (x, t) = H (c0 t − |x|) (20.75)
2E A
F(c0 t − |x|)
u x (x, t) = H (c0 t − |x|),
2E A
these expressions being shown schematically in Fig. 20.11. We shall use this solution
in other problems, so it is convenient to characterize equations (20.75) as F I(x, t).
20.4.3 Reflections
The preceding solutions relate to infinite or semi-infinite bars, but they can also be
applied to finite bars as long as the propagating waves have not reached the end of
the bar. Once this happens, a wave will generally be reflected from the end, the form
of which depends on the end condition. In particular, we can use this condition to
define a relationship between the functions f 1 , f 2 in Eq. (20.67).
Fixed End
If the bar is fixed at x = L, we require f 1 (L −c0 t) = − f 2 (L +c0 t) for all times t. This
in turn implies that f 2 (y) = − f 1 (2L − y), giving the general solution
u x = f 1 (x − c0 t) − f 1 (2L − x − c0 t)
u̇ x = −c0 f 1 (x − c0 t) + c0 f 1 (2L − x − c0 t) (20.76)
σx x = E f 1 (x − c0 t) + E f 1 (2L − x − c0 t).
The incident wave is reflected from the fixed support with the same form in σx x , but
with a sign change in u x and u̇ x .
Free End
If the end x = L is traction-free, we require f 1 (L −c0 t) = − f 2 (L +c0 t) for all t,
implying f 2 (y) = f 1 (2L − y). We then have
u x = f 1 (x − c0 t) + f 1 (2L − x − c0 t)
u̇ x = −c0 f 1 (x − c0 t) − c0 f 1 (2L − x − c0 t) (20.77)
σx x = E f 1 (x − c0 t) − E f 1 (2L − x − c0 t).
The reflected wave has the opposite sign in σx x —a tension wave is reflected from
a free end as a compression wave and vice versa.
20.4.4 The Impact Problem
We are now in a position to write down the solution to the impact problem of Fig. 20.8.
It is clear that if we adopt a frame of reference moving to the left at speed V , the
fields for the bar are all null until the impact time t = 0, and for t > 0, we must have
u̇ x (0, t) = V , as long as the bar remains in contact with the wall. The solution is
therefore given by Eq. (20.74), or equivalently (20.67) with
Vy
f 1 (y) = − H (−y). (20.78)
c0
This state continues until the wave reaches the right end of the bar at t = L/c0 , after
which the reflection from the free end leads to the solution (20.77). During this
second phase of the impact, the left end remains in contact with the wall, at which
the reaction is constant and equal to V E A/c0 .
When the tensile reflected wave reaches the support, the reaction force drops to
zero and the whole bar has a velocity 2V to the right relative to our moving frame of
reference, which is equivalent to a velocity V to the right relative to the fixed wall.
Thus, the bar leaves the wall with the same velocity as at incidence and no elastic
waves remain within it. In other words, the coefficient of restitution is unity.
20.4.5 A Rigid Mass Impacting an Elastic Bar
Figure 20.12 shows a rigid mass M, initially moving at speed V0 , which impacts an
elastic bar of length L at time t = 0. The bar is fixed at x = L. During the contact
period, we have
dV
u̇ x (0, t) = V (t); M = −F, (20.79)
dt
534 20 Impact
Fig. 20.12 A rigid mass M x

impacting an elastic bar
M
L
V0
where V (t) is the time-varying velocity of the mass and F(t) is the compressive
contact force.
When the mass makes contact with the bar, an elastic wave u x (x, t) = f 1 (x −c0 t)
will emanate from the contact point, and we deduce from Eqs. (20.68), (20.69) that
E E AV (t)
σx x (0, t) = − u̇ x (0, t) and hence F(t) = . (20.80)
c0 c0
Substituting this expression into the equation of motion (20.79)2 , we obtain
dV EA
+ λV = 0 where λ = , (20.81)
dt Mc0
and the solution of this equation is
V0 E A −λt
V (t) = V0 e−λt ; F(t) = e , (20.82)
c0
where we have used the initial condition V (0) = V0 . The propagating wave is then
described by the expression

λ
u̇ x (x, t) = V0 exp (x − c0 t) , (20.83)
c0
and its reflection from the fixed support can be determined from Eq. (20.76) as

λ λ
u̇ x (x, t) = V0 exp (x − c0 t) − exp (2L − x − c0 t) . (20.84)
c0 c0
This solution persists until the reflected wave reaches the contact point at time t = t0 ,
where
2L 2E AL 2Mbar
t0 = or equivalently λt0 = 2
= , (20.85)
c0 Mc0 M
where Mbar = ρAL is the mass of the elastic bar.

In the next phase of the process t > t0 , the velocity and force at the contact can be
written in the form

V (t) = V0 e−λt − e−λ(t−t0 ) + f (t) (20.86)
V0 E A −λt
F(t) = e + e−λ(t−t0 ) + f (t) , (20.87)
c0
where f (t) defines an additional rightward moving wave of as yet unknown form. To
determine f (t), we substitute (20.86), (20.87) into the equation of motion (20.79)2 ,
solve the resulting ordinary differential equation and impose the initial condition that
V (t) is continuous at t = t0 , obtaining
f (t) = [1 − 2λ(t − t0 )] e−λ(t−t0 ) . (20.88)
Substituting into (20.86), (20.87) and simplifying, we then obtain

V (t) = V0 e−λt − 2λ(t − t0 )e−λ(t−t0 ) (20.89)
V0 E A −λt
F(t) = e + 2 {1 − λ(t − t0 )} e−λ(t−t0 ) , (20.90)
c0
for t0 < t < 2t0 . These expressions apply until the reflection of the wave f (t) once
again reaches the contact point at t = 2t0 , unless the force F(t) falls to zero during
this interval, in which case contact is lost and the impact terminates at t = t1 , where
e−λt0
λt1 = 1 + λt0 + . (20.91)
2
For this scenario to apply, we require t1 < 2t0 , or equivalently λt0 > 1.157. We can
then determine the coefficient of restitution as V1 /V0 , where the rebound velocity
V1 = −V (t1 ) is obtained from Eqs. (20.89), (20.91). It is shown in Fig. 20.13 as a
function of λt0 . The energy deficit remains as vibrational energy in the bar after the
impact.
Fig. 20.13 Coefficient of

restitution V1 /V0 for the
problem of Fig. 20.12 with
λt0 > 1.157. The dashed line
represents the limiting value
at large λt0 , which is
V1 /V0 = 2/e
536 20 Impact
Small Values of Mbar /M

The solution of Fig. 20.13 is appropriate if and only if λt0 > 1.157, since for smaller
values, the wave associated with the term f (t) in Eqs. (20.86), (20.87) will have
returned to the contact interface before separation occurs. The solution procedure
can be continued for this case, but becomes increasingly algebraically complex as
the number of wave reflections during the impact increases (Goldsmith 1960). How-
ever, an approximate solution can be obtained at the opposite extreme λt0 1, or
equivalently Mbar M. In this case the exponential decay in Eq. (20.82) is extremely
slow and at any given time, the stress is almost uniform along the bar. The solution
then approximates the quasi-static solution in which the bar is treated as a massless
spring, and the coefficient of restitution approaches unity.
20.4.6 Frictional Problems
In some applications, the bar support may permit some relative motion, governed
by a friction condition. Examples include an elastic pile driven into the ground by
an impact applied at the free end, or an elastic belt wrapped around a pulley and
subjected to dynamic tension forces.
As a simple example of this class, we consider the semi-infinite bar in Fig. 20.14
that rests on a frictional support at the point x = L. At time t = 0, a force F is applied
to the free end and thereafter is maintained constant. We suppose that the support
can exert a force Q, such that
u̇ x (L) = 0; |Q| ≤ Q 0 (20.92)

u̇ x (L) = 0; Q = Q 0 sgn (u̇ x (L)) . (20.93)
The initial phase of the process is described in Sect. 20.4.1. A compression wave
emanates from the free end and just reaches the support at t = L/c0 . The subsequent
behaviour then depends on the relative magnitude of F and Q 0 .
F < Q 0 /2
In this case, when the wave reaches the support, it is reflected as a compression wave
and a support reaction 2F is generated, but since this is less than Q 0 , no slip occurs.
When the reflected wave reaches the left end, it is reflected as a tension wave, since
the applied force must remain unchanged. Waves continue to bounce between x = 0
Fig. 20.14 A semi-infinite

L
bar resting on a frictional F
support
x
x
L c 0t - L
0
F c 0t - L Q0
2
Q0
2
Fig. 20.15 Axial force (tensile positive) for the case Q 0 /2 < F < Q 0 and L < c0 t < 2L
and x = L following the rules in Sect. 20.4.3, but no slip occurs and the bar remains
quiescent in x > L.
Q 0 /2 < F < Q 0
In this case, slip starts at the support at t = L/c0 and a frictional reaction force Q 0
is generated. This reaction causes a compression wave to move to the left from
the support, and a tension wave to move to the right, each of magnitude Q 0 /2, as
described in Sect. 20.4.2 and illustrated in Fig. 20.11a. These waves are additive to
that generated by the force F, so for L/c0 < t < 2L/c0 the axial force distribution is
as shown in Fig. 20.15, with an attenuated wave propagating beyond the support.
When the leftward-moving compression wave reaches the end x = 0, it is reflected
as a tension wave, and when this in turn reaches the support, slip ceases. A fraction
F − Q 0 /2 propagates to the right of the support and the remaining wave of amplitude
Q 0 − F is reflected as a compression wave as shown in Fig. 20.16. A compression
wave of fixed length 2L propagates off to infinity on the right and it leaves behind a
growing segment L < x < c0 t −3L that subsequently remains quiescent. The reflected
wave of amplitude Q 0 − F remains trapped in the segment 0 < x < L with x = L
acting now as a fixed support. The reaction at the support alternates between Q 0 and
2F − Q 0 . In other words, the support experiences incipient slip conditions in part of
the steady-state cycle.
F > Q0
x
L 2L
0 c 0 t - 3L c 0 t - 3L
F Q0
F-
Q0 - F 2
Fig. 20.16 Axial force (tensile positive) for the case Q 0 /2 < F < Q 0 and c0 t > 3L
538 20 Impact
x
L c0 t
F
A X2 O X1 B C
Fig. 20.17 A bar partly embedded in a frictional support
The first part of the process for F > Q 0 is described by Fig. 20.15. However, slip
does not cease when the reflected wave reaches the support for the second time at
t = 3L/c0 . The friction force does not change at this time, so no wave is reflected
and the entire wave propagates beyond the support with amplitude F − Q 0 /2 in
c0 t −2L < x < c0 t and F − Q 0 in L < x < c0 t −2L.
20.4.7 Continuous Frictional Supports
Figure 20.17 shows a semi-infinite bar supported on a continuous frictional support

in the region x > 0 and unloaded in the region −L < x < 0. This could represent an
elastic pile driven into the ground, or [with F < 0] a fibre pulled suddenly out of
the matrix in a composite material (Nikitin and Tyurekhodgaev 1990; Sridhar et al.
2003; Yang et al. 2006).
Suppose a compressive force F H (t + L/c0 ) is applied at A, so that the resulting
compression wave has just reached O at time t = 0. We suppose that the bar is initially
unstressed, though we note that with a frictional support, this is not a necessary
condition. The bar might have been left in a state of residual stress by a previous
loading sequence. We assume that the frictional tractions are q per unit length, where
u̇ x (L) = 0; |q| ≤ q0 (20.94)

u̇ x (L) = 0; q = q0 sgn (u̇ x (L)) . (20.95)
Some slip is inevitable, since the assumption of full stick would require the gener-
ation of a concentrated force at B, with an implied infinite frictional traction. It seems
reasonable to start with the assumption that slip occurs in all regions accessible to
the incident wave, in which case the tractions are defined by
q(x) = −q0 H (c0 t − x). (20.96)
The force acting on a small element of bar Δξ is ΔQ = q0 Δξ to the left (opposing

the motion) starting at time ξ/c0 and the wave generated by this can be described by
−ΔQ I(x − ξ, t − ξ/c0 ), where the function I is defined in Eq. (20.75). Consider
the case where the wave has reached the point C in Fig. 20.17, a distance c0 t from
O. The axial force at X 1 will be decreased by ΔQ/2 for all elements in O X 1 and
increased by ΔQ/2 for all elements in X 1 B, where B is midway between X 1 and C.
The frictional tractions for elements in BC start too late for a wave emanating from
them to reach X 1 at time t. We conclude that the axial force at X 1 is

q0 c0 t − x q0 (c0 t − 3x)
F(x) = F − x− =F+ 0 < x < c0 t. (20.97)
2 2 4
The axial force at the point X 2 to the left of O will be increased by ΔQ/2 for all
elements between O and the mid-point of X 2 C, giving
q0 (c0 t + x)
F(x) = F + − c0 t < x < 0. (20.98)
4
This distribution is sketched as the solid line in Fig. 20.18. When c0 t = 2F/q0 ,
the wave reaches D and propagation to the right ceases, but slip continues at points
to the left of D until the time when a leftward-moving wave emanating from D has
just arrived. The dotted line shows the axial force distribution at the instant when
this reflected wave has just reached the point E.
The last point to slip is O and the axial force distribution at this instant is illustrated
by the dashed line in Fig. 20.18. The force at the support is then equal to 2F, exactly
as in the case of a fixed support, and a distributed wave [one with a linearly varying
traction of amplitude F] is reflected back towards the free end. This wave will then
be reflected at A and some additional slip is to be anticipated when the reflected wave
again reaches O. In fact, even in the quasi-static problem, there is some frictional
energy dissipation whenever the tension varies in time, so the system must tend
asymptotically to a quiescent state.
E
q0 c 0t
F 2 x
c 0t
O C D
Fig. 20.18 Axial force distribution for the bar of Fig. 20.17
540 20 Impact
Problems
1. A body of mass M impinges with initial velocity V0 on a massless platform sup-

ported by a spring of stiffness k and a viscous damper of coefficient c. Find the exit
(rebound) velocity of the mass V1 , and hence show that the coefficient of restitution
V1 /V0 is independent of V0 .
2. Two homogeneous spheres of the same material (with properties E, ν, ρ) and of
radius R1 , R2 respectively approach each other with a normal relative velocity V0 .
Write the equations of motion for each sphere and hence show that the duration of
the impact is given by Eq. (20.15) with the substitutions (20.4).
3. Find an expression for the maximum contact radius amax in the Hertzian theory of
impact and hence show that the criterion tc 2amax /c2 for the applicability of this
theory is satisfied provided that V0 c2 .
4. By comparing Eqs. (20.15), (20.2), determine the value of the stiffness k in a linear
model of impact, if the approximation is to define the correct value for the impact
period tc . Use this value to determine the maximum displacement u max and force
Pmax . Do the results agree with the Hertzian impact theory, and if not what is the
nature and magnitude of the error?
5. Extend the analysis of Sect. 20.4.5 to the case where duration of the impact t I
lies in the range 2t0 < t I < 3t0 . In particular, find the coefficient of restitution as a
function of λt0 and the range of values of λt0 for which your solution applies.
6. Figure 20.19 shows a composite body comprising a rigid mass M bonded to an
elastic bar of length L, cross-sectional area A and elastic modulus E. The body
strikes a fixed rigid support when travelling to the left at speed V0 . Describe the
subsequent motion and determine the coefficient of restitution for cases where the
duration of the impact is less than 4L/c0 , where c0 is the plane stress wave speed.
What restriction does this condition impose on the ratio M/Mbar ?
7. The force F in Fig. 20.14 lies in the range Q 0 /2 < F < Q 0 . Find the distance that
the bar slips through the support in terms of F, A, E, Q 0 .
8. The bar in Fig. 20.14 is loaded by a sinusoidal force
F(t) = F0 sin(ωt) t > 0,
where F0 > Q 0 . Describe the resulting wave propagation and determine the frictional
energy dissipation per cycle in terms of F0 , ω, A, c0 , E, Q 0 . Assume that ωL c0
Fig. 20.19 An elastic bar

x V0
with an end mass striking a
rigid wall M
L
Problems 541
x
L d
F
A O C
Fig. 20.20 An embedded elastic bar loaded by a transient end force
and consider only the period 0 < t < 2L/c0 in which reflected waves have not had
time to reach the support.
9. The pile driving problem is approximated in the form of the bar in Fig. 20.20, with
a frictional support defined by Eqs. (20.94), (20.95). The pile is loaded by impact of
a mass on the free end, which can be approximated as generating the force
F(t) = F0 [H (t) − H (t − t0 )] ,
where c0 t0 F0 /q0 . Determine the condition that must be satisfied if the further end
of the bar x = d is to slip during the initial wave propagation. Also, find the distance
that the pile slips into the support at O. Assume that d is sufficiently large to ensure
that the wave reflected from the end attenuates to zero before reaching O.
Appendix A
Potential Function Solutions for Elasticity
Problems
Elastic contact problems are greatly facilitated by representing the stress and dis-
placement fields in terms of scalar potential functions. These can be tailored so
as to give relatively simple expressions for the tractions and displacements at a
plane surface, starting from the Papkovitch–Neuber solution in terms of harmonic
potentials.1
A.1 Frictionless Problems
For problems in which the elastic half-space z > 0 is subjected to purely normal
tractions, the elastic fields can conveniently be expressed in terms of a potential
function ϕ where
∂2ϕ ∂2ϕ ∂2ϕ
∇2ϕ ≡ + + 2 = 0. (A.1)
∂x 2 ∂ y2 ∂z
The stress and displacement components in Cartesian coordinates (x, y, y) are given
by
∂2ϕ ∂ϕ ∂2ϕ ∂ϕ
2Gu x = z + (1 − 2ν) ; 2Gu y = z + (1 − 2ν)
∂x∂z ∂x ∂ y∂z ∂y
∂2ϕ ∂ϕ
2Gu z = z − 2(1 − ν)
∂z 2 ∂z
∂3ϕ ∂2ϕ ∂2ϕ ∂3ϕ ∂2ϕ

σx x = z + + 2ν ; σ x y = z + (1 − 2ν) (A.2)
∂x 2 ∂z ∂x 2 ∂ y2 ∂x∂ y∂z ∂x∂ y
∂3ϕ ∂2ϕ ∂2ϕ

σ yy = z + + 2ν 2
∂ y ∂z
2 ∂y 2 ∂x
1 For more details of this procedure, see Barber (2010), Chaps. 21,22.
https://doi.org/10.1007/978-3-319-70939-0
544 Appendix A: Potential Function Solutions for Elasticity Problems
∂3ϕ ∂3ϕ ∂3ϕ ∂2ϕ

σx z = z ; σ yz = z ; σzz = z − 2,
∂x∂z 2 ∂ y∂z 2 ∂z 3 ∂z
where G, ν are the shear modulus and Poisson’s ratio respectively for the material
(Green and Zerna 1954; Barber 2010, pp. 339–341).
Corresponding expressions in cylindrical polar coordinates (r, θ, z) are
∂2ϕ ∂ϕ z ∂2ϕ (1 − 2ν) ∂ϕ

2Gu r = z + (1 − 2ν) ; 2Gu θ = +
∂r ∂z ∂r r ∂θ∂z r ∂θ
∂2ϕ ∂ϕ
2Gu z = z − 2(1 − ν)
∂z 2 ∂z

∂3ϕ ∂2ϕ ∂2ϕ ∂2ϕ
σrr = z + 2 − 2ν + 2 (A.3)
∂r ∂z
2 ∂r ∂r 2 ∂z

z ∂3ϕ z ∂2ϕ (1 − 2ν) ∂2ϕ 1 ∂ϕ
σr θ = − 2 + −
r ∂r ∂θ∂z r ∂θ∂z r ∂r ∂θ r ∂θ
∂2ϕ ∂2ϕ ∂3ϕ ∂3ϕ

σθθ = −(1 − 2ν) − − z − z
∂r 2 ∂z 2 ∂r 2 ∂z ∂z 3
∂3ϕ z ∂3ϕ ∂3ϕ ∂2ϕ

σr z = z ; σ zθ = ; σzz = z − 2.
∂r ∂z 2 r ∂θ∂z 2 ∂z 3 ∂z
These expressions satisfy the condition that the shear tractions σzx , σzy or σzθ , σzr
be zero on the surface z = 0 for all harmonic functions ϕ and on this surface, the
normal traction and the normal displacement are given by the simpler expressions
∂2ϕ (1 − ν) ∂ϕ
σzz (r, θ, 0) = − ; u z (r, θ, 0) = − (A.4)
∂z 2 G ∂z
respectively.
A.2 Problems with Tangential Tractions
If tangential tractions are also applied to the surface z = 0, the preceding solution
should be supplemented by two additional harmonic potential functions χ, ψ, defin-
ing the additional stress and displacement components
Appendix A: Potential Function Solutions for Elasticity Problems 545
∂χ ∂2χ ∂ψ ∂χ ∂2χ ∂ψ
2Gu x = 2(1 − ν) +z +2 ; 2Gu y = 2(1 − ν) +z −2
∂x ∂x∂z ∂y ∂y ∂ y∂z ∂x
∂χ ∂2χ
2Gu z = −(1 − 2ν) +z 2
∂z ∂z
∂2χ ∂3χ ∂2χ ∂2ψ

σx x = 2(1 − ν) + z − 2ν + 2 (A.5)
∂x 2 ∂x 2 ∂z ∂z 2 ∂x∂ y
∂2χ ∂3χ ∂2ψ ∂2ψ

σx y = 2(1 − ν) +z + −
∂x∂ y ∂x∂ y∂z ∂ y2 ∂x 2
∂2χ ∂3χ ∂2χ ∂2ψ

σ yy = 2(1 − ν) + z − 2ν − 2
∂ y2 ∂ y 2 ∂z ∂z 2 ∂x∂ y
∂2χ ∂3χ ∂2ψ ∂2χ ∂3χ ∂2ψ ∂3χ

σx z = +z + ; σ yz = +z − ; σ zz = z ,
∂x∂z ∂x∂z 2 ∂ y∂z ∂ y∂z ∂ y∂z 2 ∂x∂z ∂z 3
or in cylindrical polar coordinates (r, θ, z),
∂χ ∂2χ 2 ∂ψ 2(1 − ν) ∂χ z ∂ 2 χ ∂ψ
2Gu r = 2(1 − ν) +z + ; 2Gu θ = + −2
∂r ∂r ∂z r ∂θ r ∂θ r ∂θ∂z ∂r
∂χ ∂2χ
2Gu z = −(1 − 2ν) +z 2
∂z ∂z
∂2χ ∂3χ ∂2χ 2 ∂2ψ 2 ∂ψ

σrr = 2(1 − ν) + z − 2ν + − (A.6)
∂r 2 ∂r 2 ∂z ∂z 2 r ∂r ∂θ r 2 ∂θ

2(1 − ν) ∂2χ 1 ∂χ z ∂3χ z ∂ 2 χ 1 ∂ψ ∂ 2 ψ 1 ∂ 2 ψ
σr θ = − + − 2 + − 2+ 2 2
r ∂r ∂θ r ∂θ r ∂z∂r ∂θ r ∂z∂θ r ∂r ∂r r ∂θ
∂2χ ∂2χ z ∂2χ z ∂3χ 2 ∂2ψ 2 ∂ψ

σθθ = −2(1 − ν) − 2 + + − + 2
∂r 2 ∂z 2 r ∂z∂r r ∂z∂θ
2 2 r ∂r ∂θ r ∂θ
∂2χ ∂3χ 1 ∂2ψ 1 ∂2χ z ∂3χ ∂2ψ

σr z = +z + ; σ zθ = + −
∂r ∂z ∂r ∂z 2 r ∂θ∂z r ∂θ∂z r ∂θ∂z 2 ∂r ∂z
∂3χ
σzz = z .
∂z 3
For axisymmetric problems, only the function χ is required.

546 Appendix A: Potential Function Solutions for Elasticity Problems
A.3 Two-Dimensional Problems
Two-dimensional plane strain solutions can be expressed in terms of two harmonic

potential functions φ, ψ through the relations
∂φ ∂ψ ∂φ ∂ψ
2Gu x = +z ; 2Gu z = +z − (3 − 4ν)ψ (A.7)
∂x ∂x ∂z ∂z
∂2φ ∂2ψ ∂ψ ∂2φ ∂2ψ ∂ψ

σx x = + z − 2ν ; σ x z = + z − (1 − 2ν)
∂x 2 ∂x 2 ∂z ∂x∂z ∂x∂z ∂x
∂2φ ∂2ψ ∂ψ
σzz = + z − 2(1 − ν) . (A.8)
∂z 2 ∂z 2 ∂z
Appendix B
Integrals over Elliptical Domains
In the elastic contact of bodies with quadratic surfaces, the contact area A is an ellipse
defined by
x2 y2
+ < 1. (B.1)
a2 b2
Problems of this class often require the evaluation of integrals of the form
n−1/2
ξ2 η2 H (θ)dξdη
Jn (x, y) = 1− 2 − 2
A a b r
π S2 2 n−1/2
ξ 2
η
= 1− 2 − 2 H (θ)dr dθ, (B.2)
0 S1 a b
where r, θ is a set of polar coordinates centred on the field point P(x, y), and S1 , S2
are defined in Fig. B.1,
Fig. B.1 Elliptical contact

area y,η
S2
Q
b r
P
θ
O a x,ξ
S1

https://doi.org/10.1007/978-3-319-70939-0
548 Appendix B: Integrals over Elliptical Domains
The function H (θ) might arise from anisotropy of the material, as in Sect. 2.2.2,
or from the Green’s function for tangential loading, as in Eq. (7.6) and Sect. 7.6.2.
In both cases, H (θ) satisfies the condition H (θ + π) = H (θ) and hence can be
expanded in the Fourier series
∞
∞

H (θ) = am cos(2mθ) + bm sin(2mθ). (B.3)
m=0 m=1
Some results of this kind were given in Chaps. 2 and 3 in connection with Galin’s
theorem and the Hertzian theory of contact. Here, we give a more unified treatment
of the procedure and show how it can be extended to other problems such as the
contact of anisotropic materials, or tangential loading of Hertzian contacts.
B.1 Mathematical Preliminaries
We note from Fig. B.1 or Eq. (2.24) that ξ = x +r cos θ; η = y +r sin θ, and hence
ξ2 η2
1− − = C0 − C1 (θ)r − C2 (θ)r 2 , (B.4)
a2 b2
where
x2 y2 x cos θ y sin θ
C0 = 1 − 2 − 2 ; C1 (θ) = 2 + (B.5)
a b a2 b2
cos2 θ sin2 θ (1 − e2 cos2 θ)

C2 (θ) = + = , (B.6)
a2 b2 b2
where the eccentricity e is defined by
b2
e2 = 1 − . (B.7)
a2
The integral (B.2) can then be written
π S2 n−1/2
Jn (x, y) = H (θ) C0 − C1 (θ)r − C2 (θ)r 2 dr dθ, (B.8)
0 S1
where S1 , S2 are the two points at which the quadratic function {C0 − C1 B(θ)r −
C2 (θ)r 2 } = 0, from (B.4), (B.1).
Writing
C1 C0 C2
r =t− ; D2 = + 12 , (B.9)
2C2 C2 4C2
Appendix B: Integrals over Elliptical Domains 549
we can evaluate the inner integral as

S2 n−1/2 n−1/2
D n−1/2
C0 − C1 (θ)r − C2 (θ)r 2 dr = C2 D2 − t 2 dt
S1 −D
n−1/2
(2n − 1)!! πC2 D 2n
= , (B.10)
(2n)!!
where (2n−1)!! = 1.3.5...(2n−1), (2n)!! = 2.4.6...(2n). We then have

π
(2n − 1)!!π n−1/2
Jn (x, y) = H (θ)C2 D 2n dθ. (B.11)
(2n)!! 0
B.1.1 The Singular Field n = 0
If n = 0, Eq. (B.2) corresponds to a traction distribution that is singular at the bound-

aries of A. Setting n = 0 in (B.11) and using (B.6) for C2 , we obtain
π
H (θ)dθ
J0 (x, y) = πb √ . (B.12)
0 1 − e2 cos2 θ
If the function H (θ) is defined by (B.3), we obtain

∞

J0 (x, y) = πb am I0 (m, e), (B.13)
m=0
where I0 (m, e) is defined in Eq. (B.31). Notice that the sine terms in (B.3) are anti-
symmetric about θ = π/2 and hence make no contribution to the integral.
B.1.2 The Hertzian Field n = 1
If n = 1,
π π
J1 (x, y) = H (θ)D 2 C2 dθ (B.14)

2 0
and substituting for C2 , D from Eqs. (B.6), (B.9), (B.5), we obtain

π
πb H (θ)dθ x 2 π H (θ) sin2 θdθ
J1 (x, y) = √ − 2
2 0 1 − e2 cos2 θ a 0 (1 − e2 cos2 θ)3/2
π
y2 H (θ) cos2 θdθ 2x y π H (θ) sin θ cos θdθ
− + 2 . (B.15)
a2 0 (1 − e2 cos2 θ)3/2 a 0 (1 − e2 cos2 θ)3/2
If H (θ) is defined by (B.3), we have

∞
πb x2 y2
J1 (x, y) = am I0 (m, e) − 2 I1 (m, e) − 2 I2 (m, e)
2 m=0
a a
∞

xy
+ 2 bm I3 (m, e) , (B.16)
a m=1
where I1 (m, e), I2 (m, e), I3 (m, e) are defined in (B.31, B.32).
B.2 Applications
We now apply these results to several traction distributions arising in three-dimen-

sional contact problems.
B.2.1 Normal Loading of an Isotropic Half-Space
We showed in Sect. 2.3 and Eq. (2.17) that the normal surface displacement u z (x, y)
of an isotropic half-space due to a contact pressure distribution p(x, y) is

1 p(ξ, η)dξdη
u z (x, y) = . (B.17)
πE∗ A r
It follows that the displacement due to the distribution

n−1/2
x2 y2
p(x, y) = p0 1 − 2 − 2 (B.18)
a b
∗ ∗
is given by (B.2) with H (θ) = p0 /π E . Thus, a0 = p0 /π E and the remaining
coefficients in (B.3) are zero.
For the flat punch (n = 0), Eq. (B.13) then gives
p0 bI0 (0, e) 2 p0 bK (e)

u z (x, y) = = (x, y) ∈ A. (B.19)
E∗ E∗
For the Hertzian pressure distribution
1/2
x2 y2
p(x, y) = p0 1 − 2 − 2 , (B.20)
a b
n = 1 and (B.16) yields

p0 b x2 y2
u z (x, y) = I0 (0, e) − 2 I1 (0, e) − 2 I2 (0, e) (x, y) ∈ A. (B.21)
2E ∗ a a
B.2.2 The Anisotropic Half-Space
If the elastic material is anisotropic, the Green’s function will generally depend on θ
as defined in Eq. (2.9) and hence the normal surface displacements due to the pressure
distribution p(x, y) are

h(θ) p(ξ, η)dξdη
u z (x, y) = , (B.22)
A r
where h(θ) has the form of Eq. (2.11). It follows that the displacement due to the
singular pressure distribution
−1/2
x2 y2
p(x, y) = p0 1 − 2 − 2 (B.23)
a b
is
∞

u z (x, y) = π p0 b Am I0 (m, e), (B.24)
m=0
where the constants Am are defined in (2.11). Also, the displacements due to the
Hertzian distribution (B.20) are
∞
π p0 b x2 y2
u z (x, y) = Am I0 (m, e) − 2 I1 (m, e) − 2 I2 (m, e)
2 m=0
a a
∞

xy
+ 2 Bm I3 (m, e) . (B.25)
a m=1
Notice that if the coefficients Bm = 0, the axes of the ellipse will be inclined to the
principal axes of the initial gap function g0 (x, y).
B.2.3 Tangential Loading of an Isotropic Half-Space
If tangential tractions
−1/2 −1/2
x2 y2 x2 y2
qx (x, y) = q1 1 − 2 − 2 ; q y (x, y) = q2 1 − 2 − 2 (B.26)
a b a b
are applied to the surface of an isotropic half-space, the tangential surface displace-
ments are given by Eqs. (7.59), (7.60) and (B.12) as
(1 + ν)(2 − ν)q1 b ν(1 + ν)q1 b

ux = I0 (0, e) + I0 (1, e) (B.27)
2E 2E
(1 + ν)(2 − ν)q2 b ν(1 + ν)q2 b
uy = I0 (0, e) − I0 (1, e), (B.28)
2E 2E
and are independent of x, y.
Alternatively, if
1/2 1/2
x2 y2 x2 y2
qx (x, y) = q1 1 − 2 − 2 ; q y (x, y) = q2 1 − 2 − 2 (B.29)
a b a b
we obtain
u x (x, y) = L 0 − L 1 x 2 − L 2 y 2 + L 3 x y; u y (x, y) = M0 − M1 x 2 − M2 y 2 + M3 x y,
(B.30)
where
(1 + ν)q1 b
L0 = [(2 − ν)I0 (0, e) + ν I0 (1, e)]
4E
(1 + ν)q1 b
L1 = [(2 − ν)I1 (0, e) + ν I1 (1, e)]
4Ea 2
(1 + ν)q1 b
L2 = [(2 − ν)I2 (0, e) + ν I2 (1, e)]
4Ea 2
ν(1 + ν)q2 b
L3 = I3 (1, e)
4Ea 2
(1 + ν)q2 b
M0 = [(2 − ν)I0 (0, e) − ν I0 (1, e)]
4E
(1 + ν)q2 b
M1 = [(2 − ν)I1 (0, e) − ν I1 (1, e)]
4Ea 2
(1 + ν)q2 b
M2 = [(2 − ν)I2 (0, e) − ν I2 (1, e)]
4Ea 2
ν(1 + ν)q1 b
M3 = I3 (1, e).
4Ea 2
B.3 Evaluation of Integrals
Following Barber and Ciavarella (2014), we define the integrals

π π
cos(2mθ)dθ sin2 θ cos(2mθ)dθ
I0 (m, e) = √ ; I1 (m, e) = (B.31)
0 1 − e2 cos2 θ 0 (1 − e2 cos2 θ)3/2
π π
cos2 θ cos(2mθ)dθ sin(2θ) sin(2mθ)dθ
I2 (m, e) = ; I3 (m, e) = .
0 (1 − e2 cos2 θ)3/2 (1 − e2 cos2 θ)3/2 0
(B.32)
The integral I0 (m, e) can be performed in Maple or Mathematica for any given
m, the first few being
I0 (0, e) = 2K (e)
4 [K (e) − E(e)]
I0 (1, e) = − 2K (e)
e2
32 [K (e) − E(e)] 16 [E(e) − 2K (e)]
I0 (2, e) = + + 2K (e) (B.33)
3e4 3e2
512 [K (e) − E(e)] 256 [3K (e) − 2E(e)]
I0 (3, e) = −
15e6 15e4
4 [79K (e) − 23E(e)]
+ − 2K (e).
15e2
Higher order terms can also be obtained from the recurrence relation

4m 2 2m − 1
I0 (m + 1, e) = − 1 I 0 (m, e) − I0 (m − 1, e). (B.34)
(2m + 1) e2 2m + 1
The remaining integrals can then be found by elementary algebraic operations as
(1 − e2 ) d
I1 (m, e) = I0 (m, e) − I0 (m, e) = I0 (m, e) − (1 − e2 )I2 (m, e)
e de
1 d
I2 (m, e) = I0 (m, e)
e de
1
I3 (m, e) = [I1 (m − 1, e) + I2 (m − 1, e) − I1 (m + 1, e) − I2 (m + 1, e)] .
2
(B.35)
Appendix C
Cauchy Singular Integral Equations
The Green’s function for two-dimensional elastic contact problems generates sin-
gular integral equations with Cauchy kernels. Here, we shall collect results for the
various forms of these equations and their general solutions. These results are all pre-
sented in the normalized form, where the range of integration is −1 < t < 1. Equations
involving dimensional contact boundaries such as b < x < a can be normalized by
making the linear coordinate transformation
a + b (a − b)t
x= + . (C.1)
2 2
C.1 Integral Equations of the First Kind
The normalized form of the Cauchy singular integral equation of the first kind is
1
1 F(t)dt
= f (s) − 1 < s < 1, (C.2)
π −1 (s − t)
where f (s) is a known function. The form of the solution depends on the asymptotic
behaviour of the function F(t) near the end points, as discussed in Chap. 10. For
example, for a frictionless normal contact problem, the contact pressure will be
square-root singular at the end point if the indenting body has a sharp corner, but
will be square-root bounded if the indenter is smooth and the contact boundary is
determined by the Signorini inequalities.
If F(t) is singular at t = ±1, the solution of Eq. (C.2) is
1
w(t) f (s)ds
F(t) = P− − 1 < t < −1, (C.3)
π −1 w(s)(t − s)

https://doi.org/10.1007/978-3-319-70939-0
556 Appendix C: Cauchy Singular Integral Equations
where
1
w(t) = √ (C.4)
1 − t2
is the characteristic function, and

1
P= F(t)dt (C.5)
−1
can take any value. Notice that the term involving P remains, even if f (s) = 0,
so it also defines the general solution of the corresponding homogeneous integral
equation. In most of the applications considered in this book, F(t) will represent a
traction distribution, so P will represent the corresponding normalized total force.
For all other cases, the solution of (C.2) can be written

w(t) 1
f (s)ds
F(t) = − − 1 < t < −1, (C.6)
π −1 w(s)(t − s)
where
• F(t) singular at t = −1 and bounded at t = 1

1−t 1
1+s
w(t) = ; P= f (s)ds. (C.7)
1+t −1 1−s
• F(t) bounded at t = −1 and singular at t = 1

1
1+t 1−s
w(t) = ; P=− f (s)ds. (C.8)
1−t −1 1+s
• F(t) bounded at t = ±1

1
s f (s)ds
w(t) = 1 − t 2 ; P= √ , (C.9)
−1 1 − s2
and in this case, we must also satisfy the consistency condition

1
f (s)ds
√ = 0, (C.10)
−1 1 − s2
if a solution is to be possible.
Appendix C: Cauchy Singular Integral Equations 557
C.2 Integral Equations of the Second Kind
The Cauchy singular integral equation of the second kind is

λ 1
F(t)dt
F(s) + = f (s) − 1 < s < 1. (C.11)
π −1 (s − t)
We define a parameter γ such that if λ is real,
1 1
cot(πγ) = λ − <γ< , (C.12)
2 2
and if it is complex,

1 λ+ı 1 1
γ= ln − < (γ) < . (C.13)
2πı λ−ı 2 2
If F(t) is singular at t = ±1, the solution is

1
f (t) w(t) λ f (s)ds
F(t) = + P cos(πγ) − , (C.14)
(1 + λ2 ) π (1 + λ2 ) −1 w(s)(t − s)
where
1
w(t) = (C.15)
(1 − t)1/2+γ (1 + t)1/2−γ
and 1
P= F(t)dt (C.16)
−1
can take any value.

For all other cases

f (t) λw(t) 1
f (s)ds
F(t) = − , (C.17)
(1 + λ ) π(1 + λ2 )
2
−1 w(s)(t − s)
and w(t), P are defined as follows:
• F(t) singular at t = −1 and bounded at t = 1
1/2−γ 1/2−γ
1−t 1
1+s
w(t) = ; P = sin(πγ) f (s)ds (C.18)
1+t −1 1−s
558 Appendix C: Cauchy Singular Integral Equations
• F(t) bounded at t = −1 and singular at t = 1
1/2+γ 1/2+γ
1+t 1
1−s
w(t) = ; P = − sin(πγ) f (s)ds (C.19)
1−t −1 1+s
• F(t) bounded at t = ±1
w(t) = (1 − t)1/2−γ (1 + t)1/2+γ

1
s f (s)ds
P = sin(πγ) (C.20)
−1 (1 − s) (1 + s)1/2+γ
1/2−γ
and we must also satisfy the consistency condition

1
f (s)ds
= 0. (C.21)
−1 (1 − s)1/2−γ (1+ s)1/2+γ
Appendix D
Dundurs’ Bimaterial Constants
If an isotropic elastic body is subjected to prescribed surface tractions, it follows from

dimensional analysis that the resulting stress field is independent of Young’s mod-
ulus, since the elasticity problem can be formulated in terms of normalized stress
components of the form σ/E. However, if the body is simply connected and the
geometry and loading are two-dimensional, it also follows that the stresses are inde-
pendent of Poisson’s ratio ν, since the terms containing ν cancel when Hooke’s law
is used to express the compatibility equation in terms of stresses (Barber 2010). This
result can also be extended to multiply connected two-dimensional bodies subject to
the restriction that the resultant of the tractions acting on each separate hole of the
body be zero.
A similar reduction in dependence on material properties applies to two-
dimensional problems involving two different isotropic elastic materials with prop-
erties E 1 , ν1 and E 2 , ν2 respectively. Dimensional analysis again shows that the
resulting stress field can at most depend on the three dimensionless parameters
E 1 /E 2 , ν1 , ν2 , but Dundurs (1969) has shown that the dependence can be further
reduced to the two independent bimaterial parameters

κ1 + 1 κ2 + 1 κ1 + 1 κ2 + 1
α= − + (D.1)
G1 G2 G1 G2

κ1 − 1 κ2 − 1 κ1 + 1 κ2 + 1
β= − + (D.2)
G1 G2 G1 G2
where G is the modulus of rigidity (shear modulus) and κ = 3−4ν for plane strain

https://doi.org/10.1007/978-3-319-70939-0
560 Appendix D: Dundurs’ Bimaterial Constants
Fig. D.1 Range of values of

α, β if ν is restricted to the
range 0 ≤ ν ≤ 0.5 for both
materials
and κ = (3−ν)/(1−ν) for plane stress. In the plane strain case, these expressions
can be written in the terms of E 1 , E 2 , ν1 , ν2 as

1 − ν12
∗ 1 − ν22
α=E − (D.3)
E1 E2

∗ (1 − 2ν1 )(1 + ν1 ) (1 − 2ν2 )(1 + ν2 )
β=E − (D.4)
2E 1 2E 2
∗
where the composite modulus E is defined by
1 1 − ν12 1 − ν22
∗ = + . (D.5)
E E1 E2
If Poisson’s ratio is restricted to the range 0 ≤ ν1 , ν2 ≤ 0.5, the constants α, β must

lie within the parallelogram shown in Fig. D.1.
References
Abercrombie, R. E., & Rice, J. R. (2005). Can observations of earthquake scaling constrain
slip weakening? Geophysical Journal International, 162(2), 406–424. https://doi.org/10.1111/j.
1365-246X.2005.02579.x.
Achenbach, J. D. (1984). Wave propagation in elastic solids. New York: Elsevier.
Achenbach, J. D., & Epstein, H. I. (1967). Dynamic interaction of a layer and a half space. ASCE
Journal of the Engineering Mechanics Division, EM5, 27–42.
Adams, G. G. (1979). A rigid punch bonded to a half plane. ASME Journal of Applied Mechanics,
46(4), 844–848. https://doi.org/10.1115/1.3424665.
Adams, G. G. (1995). Self-excited oscillations of two elastic half-spaces sliding with a constant
coefficient of friction. ASME Journal of Applied Mechanics, 62(4), 867–872. https://doi.org/10.
1115/1.2896013.
Adams, G. G. (1998). Steady sliding of two elastic half-spaces with friction reduction due to
interface stick-slip. ASME Journal of Applied Mechanics, 65(2), 470–475. https://doi.org/10.
1115/1.2789077.
Adams, G. G. (2016). Frictional slip of a rigid punch on an elastic half-plane. Proceedings of the
Royal Society of London, A 472(2191), 20160352. https://doi.org/10.1098/rspa.2016.0352.
Adda-Bedia, M., & Llewellyn Smith, S. G. (2006). Supersonic and subsonic stages of dynamic
contact between bodies. Proceedings of the Royal Society of London, A 462, 2781–2795. https://
doi.org/10.1098/rspa.2006.1709.
Afferante, L., Ciavarella, M., & Sackfield, A. (2011a). Rolling cylinder on an elastic half-plane
with harmonic oscillations in normal force and rotational speed. Part I: Solution of the partial
slip contact problem. International Journal of Mechanical Sciences, 53(11), 989–999. https://
doi.org/10.1016/j.ijmecsci.2011.08.004.
Afferante, L., Ciavarella, M., & Dell’Orco, M. (2011b). Rolling cylinder on an elastic half-plane
with harmonic oscillations in normal force and rotational speed. Part II: Energy dissipation recep-
tances and example calculations of corrugation in the short-pitch range. International Journal of
Mechanical Sciences, 53(11), 1000–1007. https://doi.org/10.1016/j.ijmecsci.2011.08.005.
Ahn, Y. J. (2010). Discontinuity of quasi-static solution in the two-node Coulomb frictional system.
International Journal of Solids and Structures, 47(21), 2866–2871. https://doi.org/10.1016/j.
ijsolstr.2010.06.007.
Ahn, Y. J. (2017). Relaxation damping and friction. International Journal of Mechanical Sciences,
128–129, 147–149. https://doi.org/10.1016/j.ijmecsci.2017.04.020.

https://doi.org/10.1007/978-3-319-70939-0
562 References
Ahn, Y. J., & Barber, J. R. (2008). Response of frictional receding contact problems to cyclic
loading. International Journal of Mechanical Sciences, 50(10–11), 1519–1525. https://doi.org/
10.1016/j.ijmecsci.2008.08.003.
Ahn, Y. J., Bertocchi, E., & Barber, J. R. (2008). Shakedown of coupled two-dimensional discrete
frictional systems. Journal of the Mechanics and Physics of Solids, 56(12), 3433–3440. https://
doi.org/10.1016/j.jmps.2008.09.003.
Akarapu, S., Sharp, T., & Robbins, M. O. (2011). Stiffness of contacts between rough surfaces.
Physical Review Letters, 106(20), 20430199. https://doi.org/10.1103/PhysRevLett.106.204301.
Aleshin, V., Bou Matar, O., & Van Den Abeele, K. (2015). Method of memory diagrams for
mechanical frictional contacts subject to arbitrary 2D loading. International Journal of Solids
and Structures, 60–61, 84–95. https://doi.org/10.1016/j.ijsolstr.2015.02.016.
Almqvist, A., Campaña, C., Prodanov, N., & Persson, B. N. J. (2011). Interfacial separation between
elastic solids with randomly rough surfaces: Comparison between theory and numerical tech-
niques. Journal of the Mechanics and Physics of Solids, 59, 2355–2369. https://doi.org/10.1016/
j.jmps.2011.08.004.
Andersson, L.-E., Barber, J. R., & Ponter, A. R. S. (2014). Existence and uniqueness of attractors in
frictional systems with uncoupled tangential displacements and normal tractions. International
Journal of Solids and Structures, 51(21–22), 3710–3714. https://doi.org/10.1016/j.ijsolstr.2014.
07.004.
Anderson, A. E., & Knapp, R. A. (1990). Hot spotting in automotive friction systems. Wear, 135(2),
319–337. https://doi.org/10.1016/0043-1648(90)90034-8.
Aramaki, H., Cheng, H. S., & Chung, Y.-W. (1993). The contact between rough surfaces with
longitudinal texture — Part I: Average contact pressure and real contact area. ASME Journal of
Tribology, 115(3), 419–424.
Archard, J. F. (1953). Contact and rubbing of flat surfaces. Journal of Applied Physics, 24(8),
981–988. https://doi.org/10.1063/1.1721448.
Archard, J. F. (1957). Elastic deformation and the laws of friction. Proceedings of the Royal Society
of London, A 243(1233), 190–205. https://doi.org/10.1098/rspa.1957.0214.
Archard, J. F. (1959). The temperature of rubbing surfaces. Wear, 2(6), 438–455. https://doi.org/
10.1016/0043-1648(59)90159-0.
Argatov, I. I. (2008). Asymptotic modeling of the impact of a spherical indenter on an elastic half-
space. International Journal of Solids and Structures, 45(18–19), 5035–5048. https://doi.org/10.
1016/j.ijsolstr.2008.05.003.
Argatov, I. I., Mishuris, G. S., & Popov, V. L. (2016). Asymptotic modelling of the JKR adhsion
contact for thin elastic layer, Quarterly Journal of Mechanics and Applied Mathematics, 69(2),
161–179. https://doi.org/10.1093/qjmam/hbw002.
Autumn, K., Sitti, M., Liang, Y. C. A., Peattie, A. M., Hansen, W. R., Sponberg, S., et al. (2002).
Evidence for van der Waals adhesion in gecko setae. Proceedings of the National Academy of
Sciences of the United States of America, 99(19), 12252–12256. https://doi.org/10.1073/pnas.
192252799.
Azarkhin, A., & Barber, J. R. (1985). Transient thermoelastic contact problem of two sliding half-
planes. Wear, 102(1–2), 1–13. https://doi.org/10.1016/0043-1648(85)90086-9.
Azarkhin, A., & Barber, J. R. (1986). Thermoelastic instability for the transient contact problem of
two sliding half-planes. ASME Journal of Applied Mechanics, 53(3), 565–572. https://doi.org/
10.1115/1.3171812.
Back, N., Burdekin, M., & Cowley, A. (1973). Pressure distributions and deformations of machined
components in contact. International Journal of Mechanical Sciences, 15(12), 993–1010. https://
doi.org/10.1016/0020-7403(73)90109-4.
Baney, J. M., & Hui, C.-Y. (1997). A cohesive zone model for the adhesion of cylinders. Journal of
Adhesion Science and Technology, 11(3), 393–406. https://doi.org/10.1163/156856197X00778.
Barber, J. R. (1969). Thermoelastic instabilities in the sliding of conforming solids. Proceedings of
the Royal Society of London, A 312(1510), 381–394. https://doi.org/10.1098/rspa.1969.0165.
References 563
Barber, J. R. (1971). The effect of thermal distortion on constriction resistance. International Journal
of Heat and Mass Transfer, 14(6), 751–766. https://doi.org/10.1016/0017-9310(71)90105-0.
Barber, J. R. (1973). Indentation of the semi-infinite elastic solid by a hot sphere. International Jour-
nal of Mechanical Sciences, 15(10), 813–819. https://doi.org/10.1016/0020-7403(73)90070-2.
Barber, J. R. (1974). Determining the contact area in elastic indentation problems. Journal of Strain
Analysis for Engineering Design, 9(4), 230–232. https://doi.org/10.1243/03093247V094230.
Barber, J. R. (1976). Some thermoelastic contact problems involving frictional heating. Quarterly
Journal of Mechanics and Applied Mathematics, 29(FEB), 1–13. https://doi.org/10.1093/qjmam/
29.1.1.
Barber, J. R. (1978). Contact problems involving a cooled punch. Journal of Elasticity, 8(4), 409–
423. https://doi.org/10.1007/BF00049190.
Barber, J. R. (1979). Adhesive contact during the oblique impact of elastic spheres. Zeitschrift für
angewandte Mathematik und Physik, 30(3), 468–476. https://doi.org/10.1007/BF01588891.
Barber, J. R. (1984). Thermoelastic displacements and stresses due to a heat source moving over
the surface of a half plane. ASME Journal of Applied Mechanics, 51(3), 636–640. https://doi.org/
10.1115/1.3167685.
Barber, J. R. (1996). Surface displacements due to a steadily moving point force. ASME Journal of
Applied Mechanics, 63(2), 245–251.
Barber, J. R. (2003). Bounds on the electrical resistance between contacting elastic rough bodies.
Proceedings of the Royal Society of London, A 459(2029), 53–66. https://doi.org/10.1098/rspa.
2002.1038.
Barber, J. R. (2010). Elasticity (3rd ed.). Dordrecht: Springer.
Barber, J. R. (2011). Intermediate mechanics of materials (2nd ed.). Dordrecht: Springer.
Barber, J. R. (2013a). Multiscale surfaces and Amontons’ law of friction. Tribology Letters, 49(3),
539–543. https://doi.org/10.1007/s11249-012-0094-6.
Barber, J. R. (2013b). Incremental stiffness and electrical contact conductance in the contact of
rough finite bodies. Physical Review E, 87(1), 013203. https://doi.org/10.1103/PhysRevE.87.
013203.
Barber, J. R., & Billings, D. A. (1990). An approximate solution for the contact area and elastic
compliance of a smooth punch of arbitrary shape. International Journal of Mechanical Sciences,
32(12), 991–997. https://doi.org/10.1016/0020-7403(90)90003-2.
Barber, J. R., & Ciavarella, M. (2014). JKR solution for an anisotropic half space. Journal of the
Mechanics and Physics of Solids, 64, 367–376. https://doi.org/10.1016/j.jmps.2013.12.002.
Barber, J. R., & Comninou, M. (1983). The penny-shaped interface crack with heat flow: II Imper-
fect contact. ASME Journal of Applied Mechanics, 50(4A), 770–776. https://doi.org/10.1115/1.
3167144.
Barber, J. R., & Hector, L. G. (1999). Thermoelastic contact problems for the layer. ASME Journal
of Applied Mechanics, 66(3), 806–808. https://doi.org/10.1115/1.2791759.
Barber, J. R., & Sturla, F. A. (1992). Application of the reciprocal theorem to some problems for
the elastic half-space. Journal of the Mechanics and Physics of Solids, 40(1), 17–25. https://doi.
org/10.1016/0022-5096(92)90212-K.
Barber, J. R., & Zhang, R. (1988). Transient behaviour and stability for the thermoelastic contact of
two rods of dissimilar materials. International Journal of Mechanical Sciences, 30(9), 691–704.
https://doi.org/10.1016/0020-7403(88)90096-3.
Barber, J. R., Dundurs, J., & Comninou, M. (1980). Stability considerations in thermoelastic contact.
ASME Journal of Applied Mechanics, 47(4), 871–874. https://doi.org/10.1115/1.3153805.
Barber, J. R., Davies, M., & Hills, D. A. (2011). Frictional elastic contact with periodic loading.
ijsolstr.2011.03.008.
Bar Sinai, Y., Brener, E. A., & Bouchbinder, E. (2012). Slow rupture of frictional interfaces. Geo-
physical Research Letters, 39, L03308. https://doi.org/10.1029/2011GL050554.
Bedding, R. J., & Willis, J. R. (1973). The dynamic indentation of an elastic half space. Journal of
Elasticity, 3(4), 289–309. https://doi.org/10.1007/BF00045744.
564 References
Bedding, R. J., & Willis, J. R. (1976). High speed indentation of an elastic half-space by con-
ical or wedge-shaped indentors. Journal of Elasticity, 6(2), 195–207. https://doi.org/10.1007/
BF00041786.
Ben-Zion, Y. (2008). Collective behavior of earthquakes and faults: continuum-discrete transitions,
progressive evolutionary changes, and different dynamic regimes. Reviews of Geophysics, 46(4),
RG4006. https://doi.org/10.1029/2008RG000260.
Bentall, R. H., & Johnson, K. L. (1967). Slip in the rolling contact of dissimilar rollers. International
Journal of Mechanical Sciences, 9(6), 389–404. https://doi.org/10.1016/0020-7403(67)90043-
4.
Berry, G. A., & Barber, J. R. (1984). The division of frictional heat — A guide to the nature of sliding
contact. ASME Journal of Tribology, 106(3), 405–415. https://doi.org/10.1115/1.3260948.
Berry, M. V., & Lewis, Z. V. (1980). On theWeierstrass-Mandelbrot fractal function. Proceedings
of the Royal Society of London, A 370(1743), 459–484.
Bertocchi, E. (2009). Selected topics on the plane elastic contact with friction, Ph.D. dissertation,
Department of Mechanical and Civil Engineering, University of Modena and Reggio Emilia,
Italy.
Block, J. M., & Keer, L. M. (2008). Periodic contact problems in plane elasticity. Journal of
Mechanics of Materials and Structures, 3(7), 1207–1237. https://doi.org/10.2140/jomms.2008.
3.1207.
Blok, H. (1937). Theoretical study of temperature rise at surfaces of ctual contact under oiliness
conditions, Institution of Mechanical Engineers (London), General Discussion of Lubrication,
Vol. 2 (pp. 222–235).
Borodich, F. M. (1993). The Hertz frictional contact between non-linear elastic anisotropic bodies
(the similarity approach). International Journal of Solids and Structures, 30, 1513–1526. https://
doi.org/10.1016/0020-7683(93)90075-I.
Borodich, F. M. (2000). Some contact problems of anisotropic elastodynamics: integral character-
istics and exact solutions. International Journal of Solids and Structures, 37(24), 3345–3373.
https://doi.org/10.1016/S0020-7683(99)00135-3.
Borri-Brunetto, M., Carpinteri, A., Chiaia, B. (1997). Lacunarity of the contact domain between
elastic bodies with rough boundaries, In G. Frantziskonis (Ed.)Probamat 97, Probabilities and
Materials, Dordrecht: Kluwer.
Borri-Brunetto, M., Carpinteri, A., & Chiaia, B. (1998). Contact, closure and friction behaviour of
rough crack concrete surfaces. In H. Mikashi (Ed.) Framcos 3, Fracture of Concrete Structures,
Gifu, Japan, Freiburg: Aedificatio Verlag.
Bottiglione, F., Carbone, G., Mangialardi, L., & Mantriota, G. (2009). Leakage mechanism in flat
seals. Journal of Applied Physics, 106(10), 104902. https://doi.org/10.1063/1.3254187.
Bowden, F. P., & Tabor, D. (1950). The friction and lubrication of solids. Oxford: Clarendon Press.
Bower, A. F., Fleck, N. A., Needleman, A., & Ogbonna, N. (1993). Indentation of a power law
creeping solid. Proceedings of the Royal Society of London, A. 441(1911), 97–124. https://doi.
org/10.1098/rspa.1993.0050.
Boyer, L. (2001). Contact resistance calculations: Generalizations of Greenwood’s formula includ-
ing interface films. IEEE Transactions on Components, Packaging and Manufacturing Techology,
24(1), 50–58. https://doi.org/10.1109/6144.910802.
Bradley, R. S. (1932). The cohesive force between solid surfaces and the surface energy of solids.
Philosophical Magazine, 13(86), 853–862. https://doi.org/10.1080/14786449209461990.
Burton, R. A. (1973). The role of insulating surface films in frictionally excited thermoelastic
instabilities. Wear, 24(2), 189–198. https://doi.org/10.1016/0043-1648(73).
Burton, R. A. (1975). Large disturbance solutions for initially flat, frictionally heated thermoelasti-
cally deformed surfaces. ASME Journal of Lubrication Technology, 97(3), 539–544. https://doi.
org/10.1115/1.3452666.
Burton, R. A., Nerlikar, V., & Kilaparti, S. R. (1973a). Thermoelastic instability in a seal-like
configuration. Wear, 24(188), 177. https://doi.org/10.1016/0043-1648(73)90230-5.
References 565
Burton, R. A., Kilaparti, S. R., & Nerlikar, V. (1973b). A limiting stationary configuration with par-
tially contacting surfaces. Wear, 24(2), 199–206. https://doi.org/10.1016/0043-1648(73)90232-
9.
Bush, A. W., Gibson, R. D., & Thomas, T. R. (1975). The elastic contact of a rough surface. Wear,
35(1), 87–111. https://doi.org/10.1016/0043-1648(75)90145-3.
Bush, A. W., Gibson, R. D., & Keogh, G. P. (1976). The limit of elastic deformation in the contact of
rough surfaces. Mechanics Research Communications, 3(3), 169–174. https://doi.org/10.1016/
0093-6413(76)90006-9.
Cabboi, A., Putelat, T., & Woodhouse, J. (2016). The frequency response of dynamic friction:
Enhanced rate-and-state models. Journal of the Mechanics and Physics of Solids, 92, 210–236.
https://doi.org/10.1016/j.jmps.2016.03.025.
Carpick, R. W., Agraït, N., Ogletree, D. F., & Salmeron, M. (1996). Variation of the interfacial shear
strength and adhesion of a nanometer-sized contact. Langmuir, 12(13), 3334–3340. https://doi.
org/10.1021/la9509007.
Carslaw, H., & Jaeger, J. C. (1959). The conduction of heat in solids (2nd ed.). Oxford: Clarendon
Press.
Carter, F. W. (1926). On the action of a locomotive driving wheel. Proceedings of the Royal Society
Castillo, J., & Barber, J. R. (1997). Lateral contact of slender prismatic bodies. Proceedings of the
Royal Society of London, 453(1966), 2397–2412. https://doi.org/10.1098/rspa.1997.0128.
Cattaneo, C. (1938). Sul contatto di due corpi elastici: distribuzione locale degli sforzi. Rendiconti
dell’Accademia Nazionale dei Lincei, 27, 342–348, 431–436, 474–478.
Chez, E. L., Dundurs, J., & Comninou, M. (1978). Reflection and refraction of SH waves in presence
of slip and friction. Bulletin of the Seismological Society of America, 68(4), 999–1011.
Cho, H., & Barber, J. R. (1998). Dynamic behavior and stability of simple frictional sys-
tems. Mathematical and Computer Modeling, 28(4—-8), 37–53. https://doi.org/10.1016/S0895-
7177(98)00107-1.
Churchman, C. M., & Hills, D. A. (2006a). General results for complete contacts subject to oscil-
latory shear. Journal of the Mechanics and Physics of Solids, 54(6), 1186–1205. https://doi.org/
10.1016/j.jmps.2005.12.005.
Churchman, C. M., & Hills, D. A. (2006b). Slip zone length at the edge of a complete contact.
International Journal of Solids and Structures, 43(7—-8), 2037–2049. https://doi.org/10.1016/
j.ijsolstr.2005.06.099.
Churchman, C. M., Dini, D., & Hills, D. A. (2006a). Closure at the root of a sharp notch. International
Journal of Mechanical Sciences, 48(10), 1063–1071. https://doi.org/10.1016/j.ijmeesci.2006.05.
014.
Churchman, C. M., Sackfield, A., & Hills, D. A. (2006b). A semi-infinite chamfered contact solution
and its application to almost complete contacts. International Journal of Solids and Structures,
43(22—-23), 7048–7060. https://doi.org/10.1016/j.ijsolstr.2006.03.030.
Churilov, V. A. (1977). On the effect of a normal load moving at constant velocity along the boundary
of an elastic half-space. Journal of Applied Mathematics and Mechanics, 41(1), 134–142. https://
doi.org/10.1016/0021-8928(77)90095-8.
Ciarlet, P. G. (1980). A justification of the von Kármán equations. Archive for Rational Mechanics
and Analysis, 73(4), 349–389. https://doi.org/10.1007/BF00247674.
Ciavarella, M. (1998a). The generalized Cattaneo partial slip plane contact problem. I–Theory,
II–Examples. International Journal of Solids and Structures, 35(18), 2349–2362. https://
doi.org/10.1016/S0020-7683(97)00154-6, [II] pp. 2363–2378. https://doi.org/10.1016/S0020-
7683(97)00155-8.
Ciavarella, M. (1998b). Tangential loading of general three-dimensional contacts. ASME Journal
Ciavarella, M., & Decuzzi, P. (2001a). The state of stress induced by the plane frictionless cylindrical
contact. I. The case of elastic similarity. International Journal of Solids and Structures, 38(26—
-27), 4507–4523. https://doi.org/10.1016/S0020-7683(00)00289-4.
566 References
Ciavarella, M., & Decuzzi, P. (2001b). The state of stress induced by the plane frictionless cylindrical
contact. II. The general case (elastic dissimilarity). International Journal of Solids and Structures,
38(26—-27), 4525–4533. https://doi.org/10.1016/S0020-7683(00)00290-0.
Ciavarella, M., & Demelio, G. (2001). A review of analytical aspects of fretting fatigue, with
extension to damage parameters, and application to dovetail joints. International Journal of Solids
and Structures, 38(10—-13), 1791–1811. https://doi.org/10.1016/S0020-7683(00)00136-0.
Ciavarella, M., & Macina, G. (2003). New results for the fretting-induced stress concentration on
Hertzian and flat rounded contacts. International Journal of Mechanical Sciences, 45(3), 449–
467. https://doi.org/10.1016/S0020-7403(03)00061-4.
Ciavarella, M., Demelio, G., Barber, J. R., & Jang, Y. H. (2000). Linear elastic contact of the
Weierstrass profile. Proceedings of the Royal Society of London, A 456(1994), 387–405. https://
doi.org/10.1098/rspa.2000.0522.
Ciavarella, M., Baldini, A., Barber, J. R., & Strozzi, A. (2006a). Reduced dependence on loading
parameters in almost conforming contacts. International Journal of Mechanical Sciences, 48(9),
917–925. https://doi.org/10.1016/j.ijmecsci.2006.03.016.
Ciavarella, M., Murolo, C., & Demelio, G. (2006b). On the elastic contact of rough surfaces:
Numerical experiments and comparisons with recent theories. Wear, 261(10), 1102–1113. https://
doi.org/10.1016/j.wear.2006.02.001.
Clausing, A. M. (1963). Thermal contact resistance in a vacuum environment, Ph.D. Thesis, Uni-
versity of Illinois.
Clausing, A. M. (1966). Heat transfer at the interface between dissimilar metals – the influence of
thermal strain. International Journal of Heat and Mass Transfer, 9(8), 791–801. https://doi.org/
10.1016/0017-9310(66)90006-8.
Cochard, A., & Rice, J. R. (2000). Fault rupture between dissimilar materials: Ill-posedness, regu-
larization, and slip-pulse response. Journal of Geophysical Research, 105(B11), 25,891–25,907.
https://doi.org/10.1029/2000JB900230.
Cole, J. D., & Huth, J. H. (1958). Stresses produced in a half-plane by moving loads. ASME Journal
of Applied Mechanics, 20(5), 433–436.
Collins, W. D. (1959). On the solution of some axisymmetric boundary value problems by means
of integral equations, II: Further problems for a circular disc and a spherical cap. Mathematika,
6(2), 120–133. https://doi.org/10.1112/S0025579300002023.
Collins, W. D. (1963). On the solution of some axisymmetric boundary value problems by means of
integral equations. VIII Potential problems for a circular annulus. Proceedings of the Edinburgh
Mathematical Society, 13(3), 235–246. https://doi.org/10.1017/S0013091500010889.
Comninou, M. (1976). Stress singularity at a sharp edge in contact problems with friction. Zeitschrift
für angewandte Mathematik und Physik, 27(4), 493–499. https://doi.org/10.1007/BF01594906.
Comninou, M., & Dundurs, J. (1977a). Reflexion and refraction of elastic waves in presence of
separation. Proceedings of the Royal Society of London, A 356(1687), 509–528. https://doi.org/
10.1098/rspa.1977.0148.
Comninou, M., & Dundurs, J. (1977b). Elastic interface waves involving separation. ASME Journal
Comninou, M., & Dundurs, J. (1978). Singular reflection and refraction of elastic waves due to sepa-
ration. ASME Journal of Applied Mechanics, 45(3), 548–552. https://doi.org/10.1115/1.3424359.
Comninou, M., & Dundurs, J. (1979a). Interaction of elastic waves with a unilateral interface.
Proceedings of the Royal Society of London, A 368(1732), 141–154. https://doi.org/10.1098/
rspa.1979.0120.
Comninou, M., & Dundurs, J. (1979b). On the Barber boundary conditions for thermoelastic contact.
Comninou, M., & Dundurs, J. (1980). Interface slip caused by an SH-pulse. International Journal
of Solids and Structures, 16(3), 283–289. https://doi.org/10.1016/0020-7683(80)90081-5.
Comninou, M., & Dundurs, J. (1982). An educational elasticity problem with friction Part 2: Unload-
ing for strong friction and reloading. ASME Journal of Applied Mechanics, 49(1), 47–51. https://
doi.org/10.1115/1.3162069.
References 567
Conroy, M., & Armstrong, J. (2005). A comparison of surface metrology techniques. Journal of
Physics Conference Series, 13, 458–465. https://doi.org/10.1088/1742-6596/13/1/106.
Cooper, M. G., Mikic, B. B., & Yovanovich, M. M. (1969). Thermal contact conductance. Inter-
national Journal of Heat and Mass Transfer, 12(3), 279–300. https://doi.org/10.1016/0017-
9310(69)90011-8.
Copson, E. T. (1947). On the problem of the electrified disc. Proceedings of the Edinburgh Mathe-
matical Society, 8, 14–19. https://doi.org/10.1017/S0013091500027644.
Craggs, J. W., & Roberts, A. M. (1967). On the motion of a heavy cylinder over the surface of
an elastic half-space, ASME Journal of Applied Mechanics, 34(1), 207–209. https://doi.org/10.
1115/1.3607626.
Dapp, W. B., Lücke, A., Persson, B. N. J., & Müser, M. H. (2012). Self-affine elastic contacts: Perco-
lation and leakage. Physical Review Letters, 108, 244301. https://doi.org/10.1103/PhysRevLett.
108.244301.
Derjaguin, B. V., Muller, V. M., & Toporov, Yu P. (1975). Effect of contact deformations on the
adhesion of particles. Journal of Colloid and Interface Science, 53(2), 314–326. https://doi.org/
10.1016/0021-9797(75)90018-1.
Dini, D., & Hills, D. A. (2004). Bounded asymptotic solution for incomplete contacts in partial
slip. International Journal of Solids and Structures, 41(24—-25), 7049–7062. https://doi.org/10.
1016/j.ijsolstr.2004.05.058.
Dow, T. A., & Burton, R. A. (1972). Thermoelastic instability of sliding contact in the absence of
wear. Wear, 19(3), 315–328. https://doi.org/10.1016/0043-1648(72)90123-8.
Dugdale, D. S. (1960). Yielding of steel sheets containing slits. Journal of the Mechanics and
Physics of Solids, 8(2), 100–104. https://doi.org/10.1016/0022-5096(60)90013-2.
Dundurs, J. (1969). Discussion on Edge bonded dissimilar orthogonal elastic wedges under normal
and shear loading. ASME Journal of Applied Mechanics, 36(3), 650–652. https://doi.org/10.1115/
1.3564739.
Dundurs, J. (1974). Distortion of a body caused by free thermal expansion. Mechanics Research
Communications, 1(3), 121–124. https://doi.org/10.1016/0093-6413(74)90001-9.
Dundurs, J., & Comninou, M. (1979). Interface separation caused by a plane elastic wave of arbitrary
form. Wave Motion, 1(1), 17–23. https://doi.org/10.1016/0165-2125(79)90022-2.
Dundurs, J., & Comninou, M. (1980). Shape of a worn slider. Wear, 62(2), 419–424. https://doi.
org/10.1016/0043-1648(80)90183-0.
Dundurs, J., & Stippes, M. (1970). Role of elastic constants in certain contact problems, ASME
Journal of Applied Mechanics, 37(4), 965–970. https://doi.org/10.1115/1.3408726.
Dundurs, J., Tsai, K. C., & Keer, L. M. (1973). Contact between elastic bodies with wavy surfaces.
Journal of Elasticity, 3(2), 109–115. https://doi.org/10.1007/BF00045817.
Duvaut, G. (1980). Free boundary problem connected with thermoelasticity and unilateral contact.
Free boundary problems (Vol. II, pp. 217–236). Francesco Severi, Rome: Istituto Nazionale di
Alta Matematica.
Enachescu, M., van den Oetelaar, R. J. A., Carpick, R. W., Ogletree, D. F., Flipse, C. F. J., &
Salmeron, M. (1999). Observation of proportionality between friction and contact area at the
nanometer scale. Tribology Letters, 7(2—-3), 73–78. https://doi.org/10.1023/A:1019173404538.
Engel, P. A., & Adams, C. E. (1980). Rolling wear study of misaligned cylindrical contacts. Wear,
60(1), 39–59. https://doi.org/10.1016/0043-1648(80)90248-3.
Erdogan, F., & Gupta, G. D. (1972). On the numerical solution of singular integral equations.
Quarterly of Applied Mathematics, 29(4), 525–534.
Eringen, A. C., & Suhubi, E. S. (1975). Elastodynamics (Vol. II). New York: Academic Press.
Fabrikant, V. I. (1986). Flat punch of arbitrary shape on an elastic half-space. International Journal
of Engineering Science, 24(11), 1731–1740. https://doi.org/10.1016/0020-7225(86)90078-9.
Fichera, G. (1964). Problemi Elastostatici con Vincoli Unilaterali: il Problema di Signorini con
Ambingue Condizioni al Contorno, Memorie dell’Accademia Nazionale dei Lincei, Ser. 8, 7
(91–140) (in Italian).
568 References
Filon, L. N. G. (1903). On an approximate solution for the bending of a beam of rectangular cross-
section under any system of load. Philosophical Transactions of the Royal Society of London, A
201(331–345), 63–155. https://doi.org/10.1098/rsta.1903.0014.
Flicek, R., Hills, D. A. & Dini, D. (2013), Progress in the application of notch
asymptotics to the understanding of complete contacts subject to fretting fatigue,
Fatigue and Fracture of Engineering Materials and Structures 36 (1), 56–64. https://doi.org/
10.1111/j.1460-2695.2012.01694.x
Freund, L. B. (1972). Energy flux into the tip of an extending crack in an elastic solid. Journal of
Elasticity, 2(4), 341–349. https://doi.org/10.1007/BF00045718.
Fuller, K. N. G., & Tabor, D. (1975). The effect of surface roughness on the adhesion of elastic
solids. Proceedings of the Royal Society of London, A 345(1642), 327–342. https://doi.org/10.
1098/rspa.1975.0138.
Galin, L. A. (1961). In Sneddon, I. N. (Ed.), Contact problems in the theory of elasticity. Raleigh:
North Carolina State College Department of Mathematics.
Galin, L. A. (1976). Contact problems of the theory of elasticity in the presence of wear. Jour-
nal of Applied Mathematics and Mechanics, 40(6), 931–936. https://doi.org/10.1016/0021-
8928(76)90132-5.
Galin, L. A. (2008). In Gladwell, G. M. L. (Ed.), Contact problems, the legacy of L. A. Galin.
Dordrecht: Springer.
Galin, L. A., & Goryacheva, I. G. (1977). Axisymmetric contact problems of the theory of elasticity
in the presence of wear. Journal of Applied Mathematics and Mechanics, 41(5), 826–831. https://
doi.org/10.1016/0021-8928(77)90164-2.
Gao, Y. F., & Bower, A. F. (2006). Elastic-plastic contact of a rough surface with Weierstrass profile.
2005.1563.
Gao, Y. F., Bower, A. F., Kim, K.-S., Lev, L., & Cheng, Y. T. (2006). The behavior of an elastic-
perfectly plastic sinusoidal surface under contact loading. Wear, 261(2), 145–154. https://doi.
org/10.1016/j.wear.2005.09.016.
Gdoutos, E. E., & Theocaris, P. S. (1975). Stress concentrations at the apex of a plane indenter
acting on an elastic half-plane. ASME Journal of Applied Mechanics, 42(3), 688–692. https://doi.
org/10.1115/1.3423663.
Geike, T., & Popov, V. L. (2007a). Reduction of three-dimensional contact problems to one-
dimensional ones. Tribology International, 40(6), 924–929. https://doi.org/10.1016/j.triboint.
2006.02.035.
Geike, T., & Popov, V. L. (2007b). Mapping of three-dimensional contact problems into one dimen-
sion. Physical Review E, 76(3), 036710. https://doi.org/10.1103/PhysRevE.76.036710.
Georgiadis, H. G., & Barber, J. R. (1993). On the super-Rayleigh/subseismic elastodynamic inden-
tation problem. Journal of Elasticity, 31(3), 141–161. https://doi.org/10.1007/BF00044967.
Georgiadis, H. G., & Lykotrafitis, G. (2001). A method based on the Radon transform for three-
dimensional elastodynamic problems of moving loads. Journal of Elasticity, 65(1–3), 87–129.
https://doi.org/10.1023/A:1016135605598.
Giannakopoulos, A. E., & Larsson, P.-L. (1997). Analysis of pyramid indentation of pressure-
sensitive hard metals and ceramics. Mechanics of Materials, 25(1), 1–35. https://doi.org/10.
1016/S0167-6636(96)00051-8.
Giannakopoulos, A. E., Larsson, P.-L., & Vestergaard, R. (1994). Analysis of Vickers indentation.
International Journal of Solids and Structures, 31(19), 2679–2708. https://doi.org/10.1016/0020-
7683(94)90225-9.
Giannakopoulos, A. E., & Pallot, P. (2000). Two-dimensional contact analysis of elastic graded
materials. Journal of the Mechanics and Physics of Solids, 48(8), 1597–1631. https://doi.org/10.
1016/S0022-5096(99)00068-X.
Giannakopoulos, A. E., & Suresh, S. (1997a). Indentation of solids with gradients in elastic prop-
erties: Part I. Point force. International Journal of Solids and Structures, 34(19), 2357–2392.
https://doi.org/10.1016/S0020-7683(96)00171-0.
References 569
Giannakopoulos, A. E., & Suresh, S. (1997b). Indentation of solids with gradients in elastic prop-
erties: Part II. Axisymmetric indenters. International Journal of Solids and Structures, 34(19),
2393–2428. https://doi.org/10.1016/S0020-7683(96)00172-2.
Gibson, R. E. (1967). Some results concerning displacements and stresses in a non-homogeneous
elastic half-space. Géotechnique, 17(1), 58–67. https://doi.org/10.1680/geot.1967.17.1.58.
Gladwell, G. M. L. (1980). Contact problems in the classical theory of elasticity. Alphen aan den
Rijn: Sijthoff and Noordhoff.
Goldsmith, W. (1960). Impact. The theory and physical behaviour of colliding solids. London:
Edward Arnold.
Goodman, L. E. (1962). Contact stress analysis of normally loaded rough spheres. ASME Journal
Gradshteyn, I. S., & Ryzhik, I. M. (1980). Tables of integrals, series and products. New York:
Academic Press.
Grassie, S. L. (2009). Rail corrugations. Characteristics, causes, and treatments. Journal of Rail and
Rapid Transit, Proceedings of the Institution of Mechanical Engineers Part F, 223(6), 581–596.
https://doi.org/10.1243/09544097JRRT264.
Green, A. E., & Zerna, W. (1954). Theoretical ElasticityD (pp. 167–174). Oxford: Clarendon Press.
Greenwood, J. A. (1966). Constriction resistance and the area of real contact. British Journal of
Applied Physics, 17(12), 1621–1632. https://doi.org/10.1088/0508-3443/17/12/310.
Greenwood, J. A. (1984). A unified theory of surface roughness. Proceedings of the Royal Society
Greenwood, J. A. (1997). Adhesion of elastic spheres. Proceedings of the Royal Society of London,
A 453(1961), 1277–1297. https://doi.org/10.1098/rspa.1997.0070.
Greenwood, J. A. (2007). On the DMT theory. Tribology Letters, 26(3), 203–211. https://doi.org/
10.1007/s11249-006-9184-7.
Greenwood, J. A., & Barber, J. R. (2012). Indentation of an elastic layer by a rigid cylinder.
ijsolstr.2012.05.036.
Greenwood, J. A., & Johnson, K. L. (1998). An alternative to the Maugis model of adhesion between
elastic spheres. Journal of Physics, D: Applied Physics, 31(22), 3279–3290. https://doi.org/10.
1088/0022-3727/31/22/017.
Greenwood, J. A., & Tabor, D. (1955). Deformation properties of friction junctions. Proceedings
of the Physical Society, B 68(9), 609–619. https://doi.org/10.1088/0370-1301/68/9/305.
Greenwood, J. A., & Tripp, J. H. (1967). The elastic contact of rough spheres. ASME Journal of
Applied Mechanics, 34(1), 153–159. https://doi.org/10.1115/1.3607616.
Greenwood, J. A., & Williamson, J. B. P. (1966). The contact of nominally flat surfaces. Proceedings
of the Royal Society of London, A 295(1442), 300–319. https://doi.org/10.1098/rspa.1966.0242.
Greenwood, J. A., & Wu, J. J. (2001). Surface roughness and contact: an apology. Meccanica, 36(6),
617–630. https://doi.org/10.1023/A:1016340601964.
Griffith, A. A. (1921). The phenomena of rupture and flow in solids. Philosophical Transactions of
the Royal Society of London, A 221(163–198), 582–593. https://doi.org/10.1098/rsta.1921.0006.
Groß-Thebing, A. (1989). Frequency-dependent creep coefficients for three dimensional
rolling contact problems. Vehicle System Dynamics, 18(6), 357–374. https://doi.org/10.1080/
00423118908968927.
Guduru, P. (2007). Detachment of a rigid solid from an elastic wavy surface: Theory. Journal of the
Mechanics and Physics of Solids, 55(3), 445–472. https://doi.org/10.1016/j.jmps.2006.09.004.
Guduru, P., & Bull, C. (2007). Detachment of a rigid solid from an elastic wavy surface: Experiments.
Journal of the Mechanics and Physics of Solids, 55(3), 473–488. https://doi.org/10.1016/j.jmps.
2006.09.007.
Guha, D., & Chowdhuri, S. K. R. (1996). The effect of surface roughness on the temperature
at the contact between sliding bodies. Wear, 197(1–2), 63–73. https://doi.org/10.1016/0043-
1648(95)06833-3.
570 References
Hannah, M. (1951). Contact stress and deformation in a thin elastic layer. Quarterly Journal of
Mechanics and Applied Mathematics, 4(1), 94–105. https://doi.org/10.1093/qjmam/4.1.94.
Hay, J. C., Bolshakov, A., & Pharr, G. M. (1999). A critical examination of the fundamental relations
used in the analysis of nanoindentation data. Journal of Materials Research, 14(6), 2296–2305.
https://doi.org/10.1557/JMR.1999.0306.
Hector, L. G., Kim, W. S., & Richmond, O. (1996). Freezing range on shell growth instability
during alloy solidification. ASME Journal of Applied Mechanics, 63(3), 594–602. https://doi.
org/10.1115/1.2823339.
Heckmann, S. R., & Burton, R. A. (1977). Effects of shear and wear on instabilities caused by
frictional heating in a seal-like configuration. ASLE Transactions, 20(1), 71–78. https://doi.org/
10.1080/05698197708982819.
Hertz, H. (1882). Über die Berührung fester elastischer Körper (On the contact of stiff elastic solids).
Journal für die reine und angewandte Mathematik, 92, 156–171 (in German). https://doi.org/10.
1515/crll.1882.92.156. See also http://home.uni-leipzig.de/pwm/web/download/Hertz1881.pdf.
Hill, R. (1992). Similarity analysis of creep indentation tests. Proceedings of the Royal Society of
London, A 436(1898), 617–630. https://doi.org/10.1098/rspa.1992.0038.
Hill, R., Storåkers, B., & Zdunek, A. B. (1989). A theoretical study of the Brinell hardness test.
1989.0056.
Hills, D. A., & Barber, J. R. (1985). Steady motion of an insulating rigid flat-ended punch
over a thermally conducting half-plane. Wear, 102(1—-2), 15–22. https://doi.org/10.1016/0043-
1648(85)90087-0.
Hills, D. A., & Sackfield, A. (1985). Sliding contacts between dissimilar elastic cylinders. ASME
Journal of Tribology, 107(4), 463–466. https://doi.org/10.1115/1.3261109.
Hills, D. A., Paynter, R. J. H., & Dini, D. (2012a). An overview of the quantification of fretting
fatigue lives of complete contacts. Engineering Fracture Mechanics, 80(SI), 3–12. https://doi.
org/10.1016/j.engfracmech.2011.06.008.
Hills, D. A., Thaitirarot, A., Barber, J. R., & Dini, D. (2012b). Correlation of fretting fatigue
experimental results using an asymptotic approach. International Journal of Fatigue, 43, 62–75.
https://doi.org/10.1016/j.ijfatigue.2012.02.006.
Ho, K., & Pehlke, R. D. (1985). Metal-mold interfacial heat transfer. Metallurgical Transactions,
B 16(3), 585–594. https://doi.org/10.1007/BF02654857.
Holm, R. (1958). Electrical contacts handbook. Berlin: Springer.
Hunt, A. G. (2005). Percolation theory for flow in porous media. New York: Springer.
Hunter, A., & Williams, A. (1969). Heat flow across metallic joints — constriction alleviation factor.
International Journal of Heat and Mass Transfer, 12(4), 524–526. https://doi.org/10.1016/0017-
9310(69)90148-3.
Hunter, S. C. (1957). Energy absorbed by elastic waves during impact. Journal of the Mechanics
and Physics of Solids, 5(3), 162–171. https://doi.org/10.1016/0022-5096(57)90002-9.
Hyun, S., Pei, L., Molinari, J.-F., & Robbins, M. O. (2004). Finite-element analysis of contact
between elastic self-affine surfaces. Physical Review E, 70(2), 026117. https://doi.org/10.1103/
PhysRevE.70.026117.
Ibrahim, R. A. (1994). Friction-induced vibration, chatter, squeal, and chaos, Part II: Dynamics
and modeling. ASME Applied Mechanics Reviews, 47(7), 227–253. https://doi.org/10.1115/1.
3111080.
Iwan, W. D. (1966). A distributed-element model for hysteresis and its steady-state dynamic
response. ASME Journal of Applied Mechanics, 33(4), 893–900. https://doi.org/10.1115/1.
3625199.
Iwan, W. D. (1967). On a class of models for the yielding behaviour of continuous and composite
systems. Journal of Applied Mechanics, 34(3), 612–617. https://doi.org/10.1115/1.3607751.
Jaeger, J. C. (1942). Moving sources of heat and the temperature of sliding contacts. Journal of the
Proceedings of the Royal Society of New South Wales, 76, 203–224. www.ewp.rpi.edu/hartford/
~ernesto/F2007/FWM/Papers/Jaeger1.pdf.
References 571
Jäger, J. (1998). A new principle in contact mechanics. ASME Journal of Tribology, 120(4), 677–
684. https://doi.org/10.1115/1.2833765.
Jain, D. L., & Kanwal, R. P. (1972). Three-part boundary value problems in potential and generalised
axially symmetric potential theories. Journal d’Analyse Mathématique, 25(1), 107–158. https://
doi.org/10.1007/BF02790033.
Jin, F., Guo, X., & Zhang, W. (2013). A unified treatment of axisymmetric adhesive contact on a
power-law graded elastic half-space. ASME Journal of Applied Mechanics, 80(6), 061024. https://
doi.org/10.1115/1.4023980.
Jin, F., Zhang, W., Zhang, S., & Guo, X. (2014). Adhesion between elastic cylinders based on the
double-Hertz model. International Journal of Solids and Structures, 51(14), 2706–2712.
Johnson, K. L. (1985). Contact mechanics. Cambridge: Cambridge University Press.
Johnson, K. L. (1995). The adhesion of two elastic bodies with slightly wavy surfaces. Inter-
national Journal of Solids and Structures, 32(3—-4), 423–430. https://doi.org/10.1016/0020-
7683(94)00111-9.
Johnson, K. L., & Greenwood, J. A. (2005). An approximate JKR theory for elliptical contacts.
Journal of Physics, D: Applied Physics, 38(7), 1042–1046. https://doi.org/10.1088/0022-3727/
38/7/012.
Johnson, K. L., Kendall, K., & Roberts, A. D. (1971). Surface energy and the contact of elastic
solids. Proceedings of the Royal Society of London, A 324(1558), 301–313. https://doi.org/10.
1098/rspa.1971.0141.
Jones, J. E. (1924). On the determination of molecular fields. II. From the equation of state of a gas.
Proceedings of the Royal Society of London, 106(738), 463–477. https://doi.org/10.1098/rspa.
1924.0082.
Kalker, J. J. (1967). A strip theory for rolling with slip and spin. Koninklijke Nederlandse Akademie
van Weteschappen, B 70(1), 10–62.
Kalker, J. J. (1971a). A minimum principle for the law of dry friction, Part 1: Fundamentals –
Application to steady rolling. ASME Journal of Applied Mechanics, 38(4), 875–880. https://doi.
org/10.1115/1.3408969.
Kalker, J. J. (1971b). A minimum principle for the law of dry friction, Part 2: Application to non-
steadily rolling elastic cylinders. ASME Journal of Applied Mechanics, 38(4), 881–887. https://
doi.org/10.1115/1.3408970.
Kalker, J. J. (1972). On elastic line contact. ASME Journal of Applied Mechanics, 39(4), 1125–1132.
https://doi.org/10.1115/1.3422841.
Kalker, J. J. (1977). The surface displacement of an elastic half-space loaded in a slender, bounded,
curved surface region with application to the calculation of the contact pressure under a roller.
Journal of the Institute of Mathematics and its Applications, 19(2), 127–144. https://doi.org/10.
1093/imamat/19.2.127.
Kalker, J. J. (1990). Three-dimensional Elastic Bodies in Rolling Contact. Dordrecht: Kluwer.
Kanninen, M. F., & Popelar, C. H. (1985). Advanced fracture mechanics. Oxford: Clarendon Press.
Karuppanan, S., Churchman, C. M., Hills, D. A., & Giner, E. (2008). Sliding frictional contact
between a square block and an elastically similar half-plane. European Journal of Mechanics
A/Solids, 27(3), 443–459. https://doi.org/10.1016/j.euromechsol.2007.09.001.
Kauzlarich, J. J., & Greenwood, J. A. (2001). Contact between a centrally loaded plate and a rigid
or elastic base, with application to pivoted pad bearings. Journal of Mechanical Engineering
Science, 215(6), 623–628. https://doi.org/10.1243/0954406011523965.
Keer, L. M., Dundurs, J., & Tsai, K. C. (1972). Problems involving a receding contact between a
layer and a half-space. ASME Journal of Applied Mechanics, 39(4), 1115–1120. https://doi.org/
10.1115/1.3422839.
Keer, L. M., Dundurs, J., & Kiattikomol, K. (1973). Separation of a smooth circular inclusion from
a matrix. International Journal of Engineering Science, 11(11), 1221–1233. https://doi.org/10.
1016/0020-7225(73)90086-4.
Keer, L. M., & Miller, G. R. (1983). Smooth indentation of finite layer. ASCE Journal of Engineering
Mechanics, 109(3), 706–717. https://doi.org/10.1061/(ASCE)0733-9399(1983)109:3(706).
572 References
Kellogg, O. D. (1929). Foundations of potential theory. New York: Springer.

Kennedy, F. E., & Ling, F. F. (1974). A thermal, thermoelastic and wear simulation of a high energy
sliding contact problem. ASME Journal of Lubrication Technology, 96(3), 497–507. https://doi.
org/10.1115/1.3452024.
Kesari, H., Doll, J. C., Pruitt, B. L., Cai, W., & Lew, A. J. (2010). Role of surface roughness in
hysteresis during adhesive elastic contact. Philosophical Magazine Letters, 90(12), 891–902.
https://doi.org/10.1080/09500839.2010.521204.
Kesari, H., & Lew, A. J. (2011). Effective macroscopic adhesive contact behavior induced by small
surface roughness. Journal of the Mechanics and Physics of Solids, 59, 2488–2510. https://doi.
org/10.1016/j.jmps.2011.07.009.
Kikuchi, N., & Oden, J. T. (1988). Contact problems in elasticity: a study of variational inequalities
and finite element methods. Philadelphia: SIAM.
Kim, J. H., Ahn, Y. J., Jang, Y. H., & Barber, J. R. (2014). Contact problems involving beams.
International Journal of Solids and Structures, 51(25–26), 4435–4439. https://doi.org/10.1016/
j.ijsolstr.2014.09.013.
Kim, J. H., & Jang, Y. H. (2014). Frictional Hertzian contact problems under cyclic loading using
static reduction. International Journal of Solids and Structures, 51(1), 252–258. https://doi.org/
10.1016/j.ijsolstr.2013.09.028.
Klarbring, A. (1990). Examples of non-uniqueness and non-existence of solutions to quasi-
static contact problems with friction. Ingenieur-Archiv, 60(8), 529–541. https://doi.org/10.1007/
BF00541909.
Klarbring, A. (1999). Contact, friction, discrete mechanical structures and discrete frictional systems
and mathematical programming. In P. Wriggers & P. Panagiotopoulos (Eds.) New developments
in contact problems (pp. 55–100). Wien: Springer.
Klarbring, A., & Björkman, G. (1988). A mathematical programming approach to contact problems
with friction and varying contact surface. Computers & Structures, 30(5), 1185–1198. https://
doi.org/10.1016/0045-7949(88)90162-9.
Klarbring, A., Ciavarella, M., & Barber, J. R. (2007). Shakedown in elastic contact problems
with Coulomb friction. International Journal of Solids and Structures, 44(25—-26), 8355–8365.
https://doi.org/10.1016/j.ijsolstr.2007.06.01.
Knothe, K., & Groß-Thebing, A. (2008). Short wavelength rail corrugation and non-steady-
state contact mechanics. Vehicle System Dynamics, 46(1—-2), 49–56. https://doi.org/10.1080/
00423110701590180.
Koiter, W. (1959). An infinite row of collinear cracks in an infinite elastic sheet. Ingenieur Archiv,
28(1), 168–172. https://doi.org/10.1007/BF00536108.
Kolsky, H. (1963). Stress waves in solids. New York: Dover.
Korčák, J. (1938). Deux types fondamentaux de distribution statistique. Bulletin de l’Institut Inter-
national de Statistique, 3, 295–299.
Krenk, S. (1975a). Use of interpolation polynomial for solutions of singular integral equations.
Quarterly of Applied Mathematics, 32(4), 479–484. http://www.jstor.org/stable/43636712.
Krenk, S. (1975b). Quadrature formulas for singular integral equations of 1st and 2nd kind. Quarterly
of Applied Mathematics, 33(3), 225–232. http://www.jstor.org/stable/43636752.
Lamb, H. (1904). On the propagation of tremors over the surface of an elastic solid. Philosophical
Transactions of the Royal Society of London, A 203, 1–42. https://doi.org/10.1098/rsta.1904.
0013.
Larsson, P.-L., Giannakopoulos, A. E., Söderlund, E., Rowcliffe, D. J., & Vestergaard, R. (1996).
Analysis of Berkovich indentation. International Journal of Solids and Structures, 33(2), 21–248.
https://doi.org/10.1016/0020-7683(95)00033-7.
Lee, K., & Barber, J. R. (1993a). The effect of shear tractions on frictionally-excited thermoelastic
instability. Wear, 160(2), 237–242. https://doi.org/10.1016/0043-1648(93)90426-M.
Lee, K., & Barber, J. R. (1993b). Frictionally-excited thermoelastic instability in automotive disk
brakes. ASME Journal of Tribology, 115(4), 607–614. https://doi.org/10.1115/1.2921683.
References 573
Lee, Y., Liu, Y., Barber, J. R., & Jang, Y. H. (2016). Thermal considerations during transient asperity
contact. Tribology International, 94, 87–91. https://doi.org/10.1016/j.triboint.2015.08.008.
Lekhnitskii, S. G. (1963). Theory of elasticity of an anisotropic elastic body. San Francisco: Holden-
Day.
Ling, F. F. (1958). On asperity distributions of metallic surfaces. Journal of Applied Physics, 29(8),
1168–1174. https://doi.org/10.1063/1.1723395.
Ling, F. F., & Simkins, T. E. (1963). Measurement of pointwise juncture conditions of temperature
at the interface of two bodies in sliding contact. ASME Journal of Basic Engineering, 85(3),
481–487. https://doi.org/10.1115/1.3656655.
Liu, Y., & Barber, J. R. (2014). Transient heat conduction between rough sliding surfaces. Tribology
Letters, 55(1), 23–33. https://doi.org/10.1007/s11249-014-0328-x.
Liu, Y., Jang, Y. H., & Barber, J. R. (2014). Finite element implementation of an eigenfunction
solution for the contact pressure variation due to wear. Wear, 309(1—-2), 134–138. https://doi.
org/10.1016/j.wear.2013.11.004.
Longuet-Higgins, M. S. (1957a). The statistical analysis of a random moving surface. Philosophical
Transactions of the Royal Society of London, A 249(966), 321–387. https://doi.org/10.1098/rsta.
1957.0002.
Longuet-Higgins, M. S. (1957b). Statistical properties of an isotropic random surface. Philosophical
Transactions of the Royal Society of London, A 250(975), 157–174. https://doi.org/10.1098/rsta.
1957.0018.
Lovrich, N. R., & Neu, R. W. (2006). Effect of mean stress on fretting fatigue of Ti-6Al-4V on
Ti-6Al-4V. Fatigue and Fracture of Engineering Materials and Structures, 29(1), 41–55. https://
doi.org/10.1111/j.1460-2965.2006.00959.x.
Majidi, C., & Adams, G. G. (2009). A simplified formulation of adhesion problems with elastic
plates. Proceedings of the Royal Society of London, A 465(2107), 2217–2230. https://doi.org/10.
1098/rspa.2009.0060.
Majumdar, A., & Bhushan, B. (1990). Role of fractal geometry in roughness characterization and
contact mechanics of surfaces. ASME Journal of Tribology, 112(2), 205–216. https://doi.org/10.
1115/1.2920243.
Majumdar, A., & Bhushan, B. (1991). Fractal model of elastic-plastic contact between rough sur-
faces. ASME Journal of Tribology, 113, 1–11. https://doi.org/10.1115/1.2920588.
Majumdar, A., & Bhushan, B. (1995). Handbook of Micro/Nano Tribology. Characterization and
modeling of surface roughness and contact mechanics (pp. 109–165). New York: CRC Press.
Mandelbrot, B. B. (1982). The fractal geometry of nature. San Francisco: Freeman.
Mandelbrot, B. B. (1985). Self affine fractals and fractal dimension. Physica Scripta, 32(4), 37–60.
https://doi.org/10.1088/0031-8949/32/4/001.
Manners, W. (1998). Partial contact between elastic surfaces with periodic profiles. Proceedings of
the Royal Society of London, A 454(1980), 3203–3221. https://doi.org/10.1098/rspa.1998.0298.
Manners, W. (2003). Methods for analysing partial contact between surfaces. International Journal
of Mechanical Sciences, 45(6—-7), 1181–1199. https://doi.org/10.1016/S0020-7403.
Manners, W., & Greenwood, J. A. (2006). Some observations on Persson’s diffusion theory of
elastic contact. Wear, 261(5—-6), 600–610. https://doi.org/10.1016/j.wear.2006.01.007.
Mansfield, E. H. (1989). The bending and stretching of plates (2nd ed.). Cambridge: Cambridge
University Press.
Martins, J. A. C., Montiero Marques, M. D. P., Gastaldi, F., & Simões, F. M. F. (1992). Contact
Mechanics International Symposium. A two degree-of-freedom “quasistatic” frictional contact
problem with instantaneous jumps (pp. 217–228). Switzerland: Lausanne.
Martins, J. A. C., Montiero Marques, M. D. P., & Gastaldi, F. (1994). On an example of non-
existence of solution to a quasistatic frictional contact problem. European Journal of Mechanics
A/Solids, 13(1), 113–133.
Martins, J. A. C., & Oden, J. T. (1987). Existence and uniqueness results for dynamic contact
problems with non-linear normal and friction interface laws. Non-linear Analysis-Theory Methods
and Applications, 11(3), 407–428. https://doi.org/10.1016/0362-546X(87)90055-1.
574 References
Mathia, T. G., Pawlus, P., & Wieczorowski, M. (2011). Recent trends in surface metrology. Wear,
271(3—-4), 494–508. https://doi.org/10.1016/j.wear.2010.06.001.
Maugis, D. (1992). Adhesion of spheres: The JKR-DMT transition using a Dugdale model.
Journal of Colloid and Interface Science, 150(1), 243–269. https://doi.org/10.1016/0021-
9797(92)90285-T.
Maugis, D. (2000). Contact, adhesion and rupture of elastic solids. New York: Springer.
Maw, N., Barber, J. R., & Fawcett, J. N. (1976). The oblique impact of elastic spheres. Wear, 38(1),
101–114. https://doi.org/10.1016/0043-1648(76)90201-5.
Maw, N., Barber, J. R., & Fawcett, J. N. (1981). The role of elastic tangential compliance in
oblique impact. ASME Journal of Lubrication Technology, 103, 74–80. https://doi.org/10.1115/
1.3251617.
Maxwell, J. C. (1892). Electricity and magnetism (3rd ed.). Oxford: Clarendon Press.
McCool, J. I. (1987). Relating profile instrument measurements to the functional performance
of rough surfaces. ASME Journal of Tribology, 109(2), 264–270. https://doi.org/10.1115/1.
3261349.
Melan, E. (1936). Theorie statisch unbestimmter Systeme aus ideal-plastischen Baustoff. Sitzungs-
berichte der Akademie der Wissenschaften in Wien, 145, 195–218. (in German).
Menga, N., & Ciavarella, M. (2015). A Winkler solution for the axisymmetric Hertzian contact prob-
lem with wear and finite element method comparison. Journal of Strain Analysis for Engineering
Design, 50(3), 156–162. https://doi.org/10.1177/0309324714567489.
Meyer, E. (1908). Untersuchungen über Härteprüfung und Härte Brinell Methoden. Zeitschrift des
Vereines Deutscher Ingenieure, 52, 645–654.
Mindlin, R. D. (1949). Compliance of elastic bodies in contact. ASME Journal of Applied Mechanics,
16(3), 259–268.
Mindlin, R. D., & Deresiewicz, H. (1953). Elastic spheres in contact under varying oblique forces.
ASME Journal of Applied Mechanics, 20(3), 327–344.
Moirot, F., & Nguyen, Q. S. (2000). Brake squeal: a problem of flutter insta-
bility of the steady sliding solution? Archives of Mechanics, 52(4–5), 645–662.
http://am.ippt.pan.pl/am/article/viewFile/v52p645/pdf
Moirot, F., Nguyen, Q. S., & Oueslati, A. (2003). An example of stick-slip and stick-slip-separation
waves. Eurpoean Journal of Mechanics, A-Solids, 22(1), 107–118. https://doi.org/10.1016/
S0997-7538(02)00004-9.
Mossakovskii, V. I. (1954). The fundamental general problem of the theory of elasticity for a
half space with a circular curve determining boundary conditions. Prikladnaya Matematika i
Mekhanika, 18(2), 187–196. (in Russian).
Mossakovskii, V. I. (1963). Compression of elastic bodies under conditions of adhesion (axisym-
metric case). Journal of Applied Mathematics and Mechanics, 27(3), 630–643. https://doi.org/
10.1016/0021-8928(63)90150-3.
Muller, V. M., Yushchenko, V. S., & Derjaguin, B. V. (1980). On the influence of molecular forces
on the deformation of an elastic sphere and its sticking to a rigid plane. Journal of Colloid and
Interface Science, 77(1), 91–101. https://doi.org/10.1016/0021-9797(80)90419-1.
Munisamy, R., Hills, D. A., & Nowell, D. (1991). Brief note on the tractive rolling of dissimilar
elastic cylinders. International Journal of Mechanical Sciences, 33(3), 225–228. https://doi.org/
10.1016/0020-7403(91)90048-8.
Munisamy, R. L., Hills, D. A., & Nowell, D. (1994). Static axisymmetrical Hertzian contacts subject
to shearing forces. ASME Journal of Applied Mechanics, 61(2), 278–283. https://doi.org/10.1115/
1.2901441.
Murthy, H., Harish, G., & Farris, T. N. (2004). Efficient modeling of fretting of blade/disk contacts
including load history effects. ASME Journal of Tribology, 126(1), 56–64. https://doi.org/10.
1115/1.1540125.
Murty, G. S. (1975). Wave propagation at an unbonded interface between two elastic half spaces.
Journal of the Acoustical Society of America, 58(5), 1094–1095. https://doi.org/10.1121/1.
380771.
References 575
Muskhelishvili, N. I. (1953). Singular Integral Equations. Groningen: P. Noordhoff.

Nakamura, M. (1995). Computer simulation for the constriction resistance depending on the form of
conducting spots. IEEE Transactions on Components, Packaging and Manufacturing Techology,
18(2), 382383. https://doi.org/10.1109/95.390320.
Nayak, P. R. (1971). Random process model of rough surfaces. ASME Journal of Lubrication
Technology, 93(3), 98–407. https://doi.org/10.1115/1.3451608.
Nayak, P. R. (1973). Some aspects of surface roughness measurement. Wear, 26(2), 165–174.
https://doi.org/10.1016/0043-1648(73)90132-4.
Nikitin, L. V., & Tyurekhodgaev, A. N. (1990). Wave propagation and vibration of elastic rods
with interfacial frictional slip. Wave Motion, 12(6), 513–526. https://doi.org/10.1016/0165-
2125(90)90022-.
Noble, B., & Hussain, M. A. (1969). Exact solution of certain dual series for indentation and
inclusion problems. International Journal of Engineering Science, 7(11), 1149–1161. https://
doi.org/10.1016/0020-7225(69)90081-0.
Nowell, D., & Hills, D. A. (1987). Mechanics of fretting fatigue tests. International Journal of
Mechanical Sciences, 29(5), 355–365. https://doi.org/10.1016/0020-7403(87)90117-2.
Nowell, D., & Hills, D. A. (1988). Tractive rolling of dissimilar elastic cylinders. International
Journal of Mechanical Sciences, 30(6), 427–439. https://doi.org/10.1016/0020-7403(88)90016-
1.
Nowell, D., Dini, D., & Hills, D. A. (2006). Recent developments in the understanding of
fretting fatigue. Engineering Fracture Mechanics, 73(2), 207–222. https://doi.org/10.1016/j.
engfracmech.2005.01.013.
Nowell, D., Hills, D. A., & Sackfield, A. (1988). Contact of dissimilar elastic cylinders under normal
and tangential loading. Journal of the Mechanics and Physics of Solids, 36(1), 59–75. https://
doi.org/10.1016/0022-5096(88)90020-8.
Oden, J. T., & Pires, E. (1983). Nonlocal and nonlinear friction laws and variational principles for
contact problems in elasticity. ASME Journal of Applied Mechanics, 50(1), 67–76. https://doi.
org/10.1115/1.3167019.
Oliver, W. C., & Pharr, G. M. (2004). Measurement of hardness and elastic modulus by instrumented
indentation: Advances in understanding and refinements to methodology. Journal of Materials
Research, 19(1), 3–20. https://doi.org/10.1557/jmr.2004.0002.
Paggi, M., & He, Q.-C. (2015). Evolution of the free volume between rough surfaces in contact.
Wear, 336—-337, 86–95. https://doi.org/10.1016/j.wear.2015.04.021.
Panek, C., & Kalker, J. J. (1977). A solution for the narrow rectangular punch. Journal of Elasticity,
7(2), 213–218. https://doi.org/10.1007/BF00041093.
Papangelo, A., Ciavarella, M., & Barber, J. R. (2015). Fracture mechanics implications for apparent
static friction coefficient in contact problems involving slip-weakening laws. Proceedings of the
Royal Society of London, A 471(2180), 20150271. https://doi.org/10.1098/rspa.2015.0271.
Parker, R. C., & Marshall, P. R. (1948). The measurement of the temperature of sliding surfaces, with
particular reference to railway blocks. Proceedings of the Institution of Mechanical Engineers,
158(1), 209–229. https://doi.org/10.1243/PIME-PROC-1948-158-026-02.
Pastewka, L., & Robbins, M. O. (2014). Contact between rough surfaces and a criterion for macro-
scopic adhesion. Proceedings of the National Academy of Sciences, 111(9), 3298–3303. https://
doi.org/10.1073/pnas.1320846111.
Pastewka, L., Prodanov, N., Lorenz, B., Müser, M. H., Robbins, M. O., & Persson, B. N. J. (2013).
Finite-size scaling in the interfacial stiffness of rough elastic contacts. Physical Review E, 87(6),
062809. https://doi.org/10.1103/PhysRevE.87.062809.
Pekeris, C. L. (1955). The seismic surface pulse. Proceedings of the National Academy of Sciences
of the United States of America, 41(7), 469–480. https://doi.org/10.1073/pnas.41.7.469.
Persson, A. (1964). On the stress distribution of cylindrical elastic bodies in contact, Dissertation,
Chalmers Tekniska Högskola, Göteborg.
Persson, B. N. J. (2001). Theory of rubber friction and contact mechanics. Journal of Chemical
Physics, 115(8), 3840–3861. https://doi.org/10.1063/1.1388626.
576 References
Persson, B. N. J. (2007). Relation between interfacial separation and load: A general theory of con-
tact mechanics. Physical Review Letters, 99(12), 125502. https://doi.org/10.1103/PhysRevLett.
99.125502.
Persson, B. N. J., & Scaraggi, M. (2014). Theory of adhesion: Role of surface roughness. Journal
of Chemical Physics, 141(12), 124701. https://doi.org/10.1063/1.4895789.
Persson, B. N. J., & Tosatti, E. (2001). The effect of surface roughness on the adhesion of elastic
solids. Journal of Chemical Physics, 115(12), 5597–5610. https://doi.org/10.1063/1.1398300.
Pohrt, R., & Popov, V. L. (2012). Normal contact stiffness of elastic solids with fractal rough
surfaces. Physical Review Letters, 108(10), 104301. https://doi.org/10.1103/PhysRevLett.108.
104301.
Popov, V. L., & Heß, M. (2015). Method of dimensionality reduction in contact mechanics and
friction. Heidelberg: Springer.
Pohrt, R., Popov, V. L., & Filippov, A. E. (2012). Normal contact stiffness of elastic solids with
fractal rough surfaces for one- and three-dimensional systems. Physical Review E, 86(2), 026710.
https://doi.org/10.1103/PhysRevE.86.026710.
Popov, M., Popov, V. L., & Pohrt, R. (2015). Relaxation damping in oscillating contacts. Scientific
Reports, 5, 16189. https://doi.org/10.1038/srep16189.
Prakash, V. (1998). Frictional response of sliding interfaces subjected to time varying normal pres-
sures. ASME Journal of Tribology, 120(1), 97–102. https://doi.org/10.1115/1.2834197.
Putignano, C., Ciavarella, M., & Barber, J. R. (2011). Frictional energy dissipation in contact of
nominally flat rough surfaces under harmonically varying loads. Journal of the Mechanics and
Physics of Solids, 59(12), 2442–2454. https://doi.org/10.1016/j.jmps.2011.09.005.
Quinn, D. D., & Segalman, D. J. (2005). Using series-series Iwan-Type models for understanding
joint dynamics. ASME Journal of Applied Mechanics, 72(5), 666–673. https://doi.org/10.1115/
1.1978918.
Rabinowicz, E. (1951). The nature of the static and kinetic coefficients of friction. Journal of Applied
Rabinowicz, E. (1995). Friction and wear of materials (2nd ed.). New York: Wiley.
Rahman, M. (1996). Hertz problem for a rigid punch moving across the surface of a semiinfinite
elastic solid. Zeitschrift für angewandte Mathematik und Physik, 47(4), 601–615. https://doi.org/
10.1007/BF00914874.
Rahman, M., & Barber, J. R. (1995). Exact expressions for the roots of the secular equation for
Rayleigh waves. ASME Journal of Applied Mechanics, 62(1), 250–252. https://doi.org/10.1115/
1.2895917.
Ranjith, K., & Rice, J. R. (2001). Slip dynamics at an interface between dissimilar materials.
Journal of the Mechanics and Physics of Solids, 49(2), 341–361. https://doi.org/10.1016/S0022-
5096(00)00029-6.
Renton, J. D. (1991). Generalized beam theory applied to shear stiffness. International Journal of
Solids and Structures, 27(15), 1955–1967. https://doi.org/10.1016/0020-7683(91)90188-L.
Rice, J. R. (1974). Limitations to the small-scale yielding approximation for crack tip plastic-
ity. Journal of the Mechanics and Physics of Solids, 22, 17–26. https://doi.org/10.1016/0022-
5096(74)90010-6.
Rice, J. R., Lapusta, N., & Ranjith, K. (2001). Rate and state dependent friction and the stability of
sliding between elastically deformable solids. Journal of the Mechanics and Physics of Solids,
49(9), 1865–1898. https://doi.org/10.1016/S0022-5096(01)00042-4.
Richards, P. G. (1979). Elementary solutions to Lamb’s problem for a point source and their relevance
to three-dimensional studies of spontaneous crack propagation. Bulleting of the Seismological
Society of America, 69(4), 947–956. http://bssaonline.org/content/69/4/947.
Richmond, O., & Huang, N. C. (1977). Interface stability during unidirectional solidifcation of a
pure metal. In: Proceedings of the 6th Canadian Congress of Applied Mechanics, Vancouver, pp.
453–454 .
References 577
Richmond, O., & Tien, R. H. (1971). Theory of thermal stresses and air-gap formation during the
early stages of solidification in a rectangular mold. Journal of the Mechanics and Physics of
Solids, 19(2), 273–284. https://doi.org/10.1016/0022-5096(71)90013-5.
Robinson, A. R., & Thompson, J. C. (1974). Transient stresses in an elastic half-space resulting
from the frictionless indentation of a rigid wedge-shaped die ZAMM. Zeitschrift für Angewandte
Mathematik und Mechanik, 54(3), 139–144. https://doi.org/10.1002/zamm.19740540303.
Ruina, A. (1983). Slip instability and state variable friction laws. Journal of Geophysical Research,
88(NB12), 10,359–10,370. https://doi.org/10.1029/JB088iB12p10359.
Russ, J. C. (1994). Fractal surfaces. New York: Plenum Press.
Sackfield, A., & Hills, D. A. (1988). Sliding contact between dissimilar elastic bodies. ASME
Journal of Tribology, 110(4), 592–596. https://doi.org/10.1115/1.3261698.
Sackfield, A., Mugadu, A., Barber, J. R., & Hills, D. A. (2003). The application of asymptotic
solutions to characterising the process zone in almost complete frictionless contacts. Jour-
nal of the Mechanics and Physics of Solids, 51(7), 1333–1346. https://doi.org/10.1016/S0022-
5096(03)00020-6.
Schmueser, D., & Comninou, M. (1979). The periodic array of interface cracks and their interaction.
International Journal of Solids and Structures, 15(12), 927–934. https://doi.org/10.1016/0020-
7683(79)90022-2.
Segedin, C. M. (1957). The relation between load and penetration for a spherical punch. Mathe-
matika, 4, 156–161. https://doi.org/10.1112/S0025579300001236.
Shield, R. T. (1967). Load-displacement relations for elastic bodies. Zeitschrift für angewandte
Mathematik und Physik, 18(5), 682–693. https://doi.org/10.1007/BF01602041.
Simões, F. M. F., & Martins, J. A. C. (1998). Instability and ill-posedness in some friction prob-
lems. International Journal of Engineering Science, 36(11), 1265–1293. https://doi.org/10.1016/
S0020-7225(98)00024-X.
Sivashinsky, G. I. (1975). The problem of a slender die. Journal of Elasticity, 5(2), 161–166. https://
doi.org/10.1007/BF01390077.
Slepyan, L. I., & Brun, M. (2012). Driving forces in moving-contact problems of dynamic elasticity:
Indentation, wedging and free sliding. Journal of the Mechanics and Physics of Solids, 60(11),
1883–1906. https://doi.org/10.1016/j.jmps.2012.06.011.
Smith, E. H., & Arnell, R. D. (2013). A new approach to the calculation of flash temperatures in dry,
sliding contacts. Tribology Letters, 52(3), 407–414. https://doi.org/10.1007/s11249-013-0224-
9.
Sneddon, I. N. (1947). Note on a boundary value problem of Reissner and Sagoci. Journal of Applied
Sneddon, I. N. (1951). Fourier transforms. New York: McGraw-Hill.
Song, Y., Hartwigsen, C. J., McFarland, D. M., Vakakis, A. F., & Bergman, L. A. (2004). Simulation
of dynamics of beam structures with bolted joints using adjusted Iwan beam elements. Journal
of Sound and Vibration, 273(1—-2), 249–276. https://doi.org/10.1016/S002-460X(03)00499-1.
Song, Z., & Komvopoulos, K. (2014). Adhesive contact of an elastic semi-infinite solid with a rigid
rough surface: Strength of adhesion and contact instabilities. International Journal of Solids and
Structures, 51(6), 1197–1207. https://doi.org/10.1016/j.ijsolstr.2013.10.039.
Spence, D. A. (1968). Self similar solutions to adhesive contact problems with incremental loading.
1968.0105.
Spence, D. A. (1973). An eigenvalue problem for elastic contact with finite friction. Pro-
ceedings of the Cambridge Philosophical Society, 73(Jan), 249–268. https://doi.org/10.1017/
S0305004100047666.
Spence, D. A. (1975). The Hertz problem with finite friction. Journal of Elasticity, 5(3—-4), 297–
319. https://doi.org/10.1007/BF00126993.
Sridhar, N., Yang, Q. D., & Cox, B. N. (2003). Slip, stick, and reverse slip characteristics during
dynamic fibre pullout. Journal of the Mechanics and Physics of Solids, 51(7), 1215–1241. https://
doi.org/10.1016/S0022-5096(03)00035-8.
578 References
Srinivasan, M. G., & France, D. M. (1985). Non-uniqueness in steady-state heat transfer in pre-
stressed duplex tubes - Analysis and case history. ASME Journal of Applied Mechanics, 52(2),
257–262. https://doi.org/10.1115/1.3169037.
Sternberg, W. J., & Smith, T. L. (1944). The theory of potential and spherical harmonics. Toronto:
University of Toronto Press.
Stingl, B., Ciavarella, M., & Hoffmann, M. (2013). Frictional dissipation in elastically dissimilar
oscillating Hertzian contacts. International Journal of Mechanical Sciences, 72, 55–62. https://
doi.org/10.1016/j.ijmecsci.2013.03.012.
Stoneley, R. (1924). Elastic waves at the surface of separation of two solids. Proceedings of the
Royal Society of London, 106(738), 416–428. https://doi.org/10.1098/rspa.1924.0079.
Storåkers, B., Biwa, S., & Larsson, P.-L. (1997). Similarity analysis of inelastic contact. Inter-
national Journal of Solids and Structures, 34(24), 3061–3083. https://doi.org/10.1016/S0020-
7683(96)00176-X.
Storåkers, B., & Elaguine, D. (2005). Hertz contact at finite friction and arbitrary profiles. Journal
of the Mechanics and Physics of Solids, 53(6), 1422–1447. https://doi.org/10.1016/j.jmps.2004.
11.009.
Stroh, A. N. (1958). Dislocations and cracks in anisotropic elasticity. Philosophical Magazine,
3(30), 625–646. https://doi.org/10.1080/14786435808565804.
Stroh, A. N. (1962). Steady-state problems in anisotropic elasticity. Journal of Mathematics and
Physics, 41(2), 77–103. https://doi.org/10.1002/sapm196241177.
Stronge, W. J. (2000). Impact mechanics. Cambridge: Cambridge University Press.
Sundaram, N., & Farris, T. N. (2010a). The generalized advancing conformal contact problem with
friction, pin loads and remote loading Case of rigid pin. International Journal of Solids and
Structures, 47(6), 801–815. https://doi.org/10.1016/j.ijsolstr.2009.11.019.
Sundaram, N., & Farris, T. N. (2010b). Mechanics of advancing pin-loaded contacts with friction.
Journal of the Mechanics and Physics of Solids, 58(11), 1819–1833. https://doi.org/10.1016/j.
jmps.2010.08.004.
Sveklo, V. A. (1964). Boussinesq type problems for the anisotropic half-space. Journal of Applied
Mathematics and Mechanics, 28(5), 1099–1105. https://doi.org/10.1016/0021-8928(64)90012-
7.
Svetlizky, I., & Fineberg, J. (2014). Classical shear cracks drive the onset of dry frictional motion.
Nature, 509(7499), 205–208. https://doi.org/10.1038/nature13202.
(1951). The hardness of metals. Oxford: Clarendon Press.
Tabor, D. (1959). Junction growth in metallic friction – the role of combined stresses and surface
contamination. Proceedings of the Royal Society of London, A 251, 378–393. https://doi.org/10.
1098/rspa.1959.0114.
Tabor, D. (1977). Surface forces and surface interactions. Journal of Colloid and Interface Science,
58(1), 2–13. https://doi.org/10.1016/0021-9797(77)90366-6.
Thompson, J. C., & Robinson, A. R. (1977). Exact solution for the superseismic stage of dynamic
contact between a punch and an elastic body. ASME Journal of Applied Mechanics, 44(4), 583–
586. https://doi.org/10.1115/1.3424139.
Thoms, E. (1988). Disc brakes for heavy vehicles. Institution of Mechanical Engineers International
Conference on Disc Brakes for Commercial Vehicles, C464/88, 133–137.
Tian, X., & Kennedy, F. E. (1994). Maximum and average temperatures in sliding contacts. ASME
Journal of Tribology, 116(1), 167–174.
Ting, T. C. T. (1996). Anisotropic elasticity. New York: Oxford University Press.
Tricomi, F. G. (1970). Integral equations. New York: Interscience.
Tsai, K. C., Dundurs, J., & Keer, L. M. (1972). Contact between an elastic layer with a slightly
curved bottom and a substrate. ASME Journal of Applied Mechanics, 39(3), 821–823. https://doi.
org/10.1115/1.3422798.
Turner, J. R. (1979). Frictional unloading problem on a linear elastic half-space. Journal of the
Institute of Mathematics and its Applications, 24(4), 439–469. https://doi.org/10.1093/imamat/
24.4.439.
References 579
Tvergaard, V., & Hutchinson, J. W. (1994). Effect of T-stress on mode I crack growth resistance in
a ductile solid. International Journal of Solids and Structures, 31(6), 823–833. https://doi.org/
10.1016/0020-7683(94)90080-9.
Vlassak, J. J., & Nix, W. D. (1994). Measuring the elastic properties of anisotropic materials by
means of indentation experiments. Journal of the Mechanics and Physics of Solids, 42(8), 1223–
1245. https://doi.org/10.1016/0022-5096(94)90033-7.
Weertman, J. (1963). Dislocations moving uniformly on the interface between isotropic media of
different elastic properties. Journal of the Mechanics and Physics of Solids, 11(3), 197–204.
https://doi.org/10.1016/0022-5096(63)90052-8.
Weertman, J. (1980). Unstable slippage across a fault that separates elastic media of different
elastic constants. Journal of Geophysical Research, 85(NB3), 1455–1461. https://doi.org/10.
1029/JB085iB03p01455.
Westergaard, H. M. (1939). Bearing pressures and cracks. ASME Journal of Applied Mechanics, 6,
49–53.
Westergaard, H. M. (1964). Theory of elasticity and plasticity. New York: Dover.
Westmann, R. A. (1965). Asymmetric mixed boundary value problems of the elastic half space.
Whitehouse, D. J. (1997). Surface metrology. Measurement Science and Technology, 8(9), 955–972.
https://doi.org/10.1088/0957-0233/8/9/002.
Whitehouse, D. J., & Archard, J. F. (1970). Properties of random surfaces of significance in their
contact. Proceedings of the Royal Society of London, A 316(1524), 97–121. https://doi.org/10.
1098/rspa.1970.0068.
Whitehouse, D. J., & Phillips, M. J. (1978). Discrete properties of random surfaces. Philosophical
Transactions of the Royal Society of London, A 290(1369), 267–298. https://doi.org/10.1098/
rsta.1978.0084.
Whitehouse, D. J., & Phillips, M. J. (1982). Two-dimensional discrete properties of random surfaces.
Philosophical Transactions of the Royal Society of London, 305(1490), 441–468. https://doi.org/
10.1098/rsta.1982.0043.
Williams, M. L. (1952). Stress singularities from various boundary conditions in angular corners
of plates in extension. ASME Journal of Applied Mechanics, 19(4), 526–528.
Willis, J. R. (1966). Hertzian contact of anisotropic bodies. Journal of the Mechanics and Physics
of Solids, 14(3), 163–176. https://doi.org/10.1016/0022-5096(66)90036-6.
Journal of the Mechanics and Physics of Solids, (1967). Boussinesq problems for an anisotropic
half-space. 15(5), 331–339. https://doi.org/10.1016/0022-5096(67)90027-0.
Willis, J. R. (1970). The distribution of stress in an anisotropic elastic body containing an exterior
crack. International Journal of Engineering Science, 8(7), 559–574. https://doi.org/10.1016/
0020-7225(70)90041-8.
Willis, J. R. (1973). Self-similar problems in elastodynamics. Philosophical Transactions of the
Royal Society of London, A 274(1240), 435–491. https://doi.org/10.1098/rsta.1973.0073.
Wray, P. J. (1981). Geometric features of chill-cast structures. Metallurgical Transactions, 12B(1),
167–176. https://doi.org/10.1007/BF0267477000.
Wu, J.-J. (2008). Easy-to-implement equations for determining adhesive contact parameters with
the accuracy of numerical simulations. Tribology Letters, 30(2), 99–105. https://doi.org/10.1007/
s11249-008-9315-4.
Wu, J.-J. (2010). The jump-to-contact distance in atomic force microscopy measurement. Journal
of Adhesion, 86(11), 1071–1085. https://doi.org/10.1080/00218464.2010.519256.
Wu, J.-J. (2012). Numerical simulation of the adhesive contact between a slightly wavy surface and
a half-space. Journal of Adhesive Science and Technology, 26(1—-3), 331–351. https://doi.org/
10.1163/016942411X576527.
Yan, W., & Komvopoulos, K. (1998). Contact analysis of elastic-plastic fractal surfaces. Journal of
Applied Physics, 84(7), 3617–3624. https://doi.org/10.1063/1.368536.
580 References
Yang, F. (2002). Adhesive contact between a rigid axisymmetric indenter and an incompressible
elastic thin film. Journal of Physics D, Applied Physics, 35(20), 2614–2620. https://doi.org/10.
1088/0022-3727/35/20/322.
Yang, F. (2006). Asymptotic solution to axisymmetric indentation of a compressible elastic thin
film. Thin Solid Films, 515(4), 2274–2283. https://doi.org/10.1016/j.tsf.2006.07.151.
Yang, Q. D., & Rosakis, A. Cox, B. N. (2006). Dynamic fibre sliding along debonded, frictional
interfaces, Proceedings of the Royal Society of London, A 462 (2068), 1081–1106. https://doi.
org/10.1098/rspa.2005.1602.
Yashima, S., Romero, V., Wandersman, E., Fretigny, C., Chaudhury, M. K., Chateauminois, A.,
et al. (2015). Normal contact and friction of rubber with model randomly rough surfaces. Soft
Matter, 11(5), 871–881. https://doi.org/10.1039/c4sm02346c.
Yavuz, G. (1995). Instability Problems in Unidirectional Solidification Process. Ph.D. Thesis, Uni-
versity of Michigan.
Yeo, T., & Barber, J. R. (1994). Finite element analysis of thermoelastic contact stability. ASME
Journal of Applied Mechanics, 61(4), 919–922. https://doi.org/10.1115/1.2901578.
Yi, Y.-B., Barber, J. R., & Zagrodzki, P. (2000). Eigenvalue solution of thermoelastic instability
problems using fourier reduction. Proceedings of the Royal Society of London, A 456(2003),
2799–2821. https://doi.org/10.1098/rspa.2000.0641.
Y-B. Yi, J. R. Barber and D. L. Hartsock (2002), Thermoelastic instabilities in automotive disc
brakes — Finite element analysis and experimental verification, in J. A. C. Martins and M. D. P.
Monteiro Marques eds., Contact Mechanics, Kluwer, Dordrecht, pp. 187–202 [396, 397].
Yigit, F. (1998). Effect of mold properties on thermo-elastic instability in unidirectional
planar solidification. Journal of Thermal Stresses, 21(1), 55–81. https://doi.org/10.1080/
01495739808956135.
Yigit, F., & Barber, J. R. (1994). Effect of Stefan number on thermoelastic instability in unidirectional
solidification. International Journal of Mechanical Sciences, 36(8), 707–723. https://doi.org/10.
1016/0020-7403(94)90087-6.
Zagrodzki, P. (1990). Analysis of thermomechanical phenomena in multidisc clutches and brakes.
Wear, 140(2), 291–308. https://doi.org/10.1016/0043-1648(90)90091-N.
Zagrodzki, P., Lam, K. B., Al-Bahkali, E., & Barber, J. R. (2001). Nonlinear transient behaviour of
a sliding system with frictionally excited thermoelastic instability. ASME Journal of Tribology,
123(4), 699–708. https://doi.org/10.1115/1.1353180.
Zhang, R., & Barber, J. R. (1990). Effect of material properties on the stability of static thermoelas-
tic contact. ASME Journal of Applied Mechanics, 57(2), 365–369. https://doi.org/10.1115/1.
2891998.
Zhang, R., & Barber, J. R. (1993). Transient thermoelastic contact and stability of two
thin-walled cylinders. Journal of Thermal Stresses, 16(1), 31–54. https://doi.org/10.1080/
01495739308946215.
Index
A Bearing area, 334, 336

Abel integral equation, 65, 69, 98, 120, 317 Belt drive, 438
Abrasive wear, 330 Berkovich indentation, 326
Actual contact area, 329, 340, 351, 367 Bilateral solution, 491, 497
clustering of actual contact areas, 361, Boundary-value problem, 4, 46
377, 384 Bowden and Tabor’s theory, 329, 336, 337
Adhesive forces, 233, 266, 304, 369 Boyer’s approximation, 48
stability, 241 Brakes, 418, 425, 428, 431
Advancing contact, 222 Brake squeal, 159
Advancing stick, 178 Brinell indentation, 324
Almost conformal contact, 9 Bulk stress, 179
Amontons’law, 137 Burton’s stability analysis, 420
Anisotropy, 16, 200, 551
Antiplane loading, 114
Archard’s fractal model, 352 C
Archard’s wear law, 455 Carter’s problem, 441
Asperities, 174, 330, 337, 463, 469 Castigliano’s second theorem, 269
asperity parameters, 351 Cattaneo’s problem, 169, 174, 211, 258
asperity summits, 337, 340, 343, 351 Cauchy singular integral equation, 103, 116,
summit curvature, 343, 351 123, 180, 442, 453, 555, 557
Asperity models, 337, 370, 376 Chebyshev polynomials, 89
exponential distribution of asperities, Ciavarella-Jäger theorem, 172, 175, 178,
339 181, 211, 258
Asymptotic fields, 5, 195 Clutches, 418, 425, 430
Atomic Force Microscope [AFM], 246, 368 Coefficient of restitution, 513, 535
Autocorrelation function, 345, 348 Complete contact. See conformal contact
Axisymmetric problem, 63, 67, 69, 238, 245, Composite elastic modulus, 17
250 Concentrated normal force, 15
annular contact, 68 Conditional probability, 346
flat punch, 65, 119 Conformal contact, 1
Hertz problem, 67 almost conformal contact, 229
Consistency condition, 83
Constriction alleviation factor, 386
B Contact area, 1, 4, 5, 19, 20, 52, 53, 367
Beam, 263 actual contact area, 329, 340, 341, 368
adhesion moment, 267 nominal contact area, 164, 329, 336
on an elastic foundation, 272 Contact inequalities, 5, 52, 195, 199
https://doi.org/10.1007/978-3-319-70939-0
582 Index
Contact resistance, 375, 386 Fourier series, 86, 92

Contact stiffness matrix, 145 Fourier transform, 309, 347, 396
Correlation distance, 346 Four-point bending, 264
Cotangent transform, 93 Fractal dimension, 333, 352
Coulomb friction, 137, 140, 327, 340, 368 Fractal surfaces, 352
Creep ratio, 444, 448 Fracture toughness, 237
Creep velocity in rolling, 441 Fretting fatigue, 150, 179, 211
Creep velocity in wave propagation, 493 Friction
Cyclic loading, 148, 155, 175, 189 Bowden and Tabor’s theory, 329
Cylindrical shell, 281 coefficient of friction, 137
existence and uniqueness, 139, 143
frictional asymptotics, 202
D frictional dissipation, 148, 156, 211
Developable surfaces, 277 frictional inequalities, 139, 142
Diamond-square algorithm, 357 frictional instability, 159
Dilatational wave, 476, 504 frictional memory, 146, 150, 178
Dimensionality Reduction (MDR), 70 frictional shakedown, 151
Discrete problem, 7, 144, 157 frictional vibrations, 160
DMT theory, 247 rate problem, 139, 145
‘Double-Hertz’ solution, 256 rate-state laws, 163, 503
Dundurs’ bimaterial constants, 112, 202, 559 static friction, 161
Dundurs’ theorem, 399 static problem, 139
stick cone, 142
stick-slip transition, 204
E velocity dependence, 159, 503
Earthquakes, 159 Frictional heating, 417, 418, 421, 424, 429,
Eigenfunction series, 197 432
Eigenvalues, 197 Frictional support, 536, 538, 540
Elastic strain energy, 241, 266, 268 Functionally Graded Material (FGM), 313,
Elastodynamic indentation, 506, 508 326
Elastodynamic pulse, 495 exponential grading, 314
Electrical conduction, 44 linear grading, 318
interface conductance, 340, 361, 375 power-law grading, 315
Electrostatic analogy, 47
Elliptic integrals, 22, 34, 38, 181
Energy release rate, 237 G
Equilibrium, 316 Galin’s theorem, 23, 89, 125, 259, 548
Error function, 335, 357 Gap function, 2
Evanescent wave, 495 Gaussian distribution, 335, 340, 357, 371,
Existence theorems, 5, 139, 408 376
Expected value, 345 Gauss–Seidel algorithm, 146, 261
Exponential distribution of asperities, 339 Gibson soil, 318
Goodman approximation, 121, 185
Green–Collins method, 63, 68
F Green’s function, 19, 93, 111, 174, 200, 250,
Fabrikant’s approximation, 49 275, 310, 478, 481, 489, 495, 497,
Fibre pull-out, 538 504
Field-point integration, 20, 123, 464, 547 Greenwood and Williamson’s theory, 339
Flash temperatures, 418, 463 Gross slip, 110
Flat and rounded indenter, 84, 209 Guduru’s problem, 245
Flat elliptical punch, 20
Flexural rigidity, 263
Fluid leakage, 381 H
Force potential, 233 Half-space approximation, 13, 43
Index 583
Hardness, 324, 329, 332, 363 Inverse problem, 157

Hardness test, 39, 323 Irrotational wave, 476
Harmonic functions, 14, 26, 44, 51 Iwan models, 157
Heat conduction, 46
interface conductance, 340
Hertzian contact, 29, 31, 34, 36, 39 J
adhesive forces, 238 JKR adhesion theory, 237, 267, 307, 318,
axisymmetric, 67 372
bulk stress, 179 Jump in and jump out, 243, 251, 373
contact pressure, 35
eccentricity of the contact area, 35
elastodynamic, 480, 484, 489 K
first yield, 39 Kalker’s line contact theory, 102
highly elliptical contacts, 38 Kalker’s strip theory, 449
indentation depth, 39 Klarbring model, 140
tangential loading, 128 Klarbring’s P-matrix criterion, 147
thermoelastic, 402 Knoop indentation, 324
two-dimensional, 83 Korcak’s law, 359
Hertz’ theory of impact, 514, 516, 517
range of validity of, 517
History-dependence, 146, 178 L
Hot judder, 425 Laplace equation, 14
Layered bodies, 289
Hot spots (in sliding), 418, 428
adhesive forces, 304, 305
Hurst exponent, 353
bonded compressible layer, 298
Hysteresis, 148, 243, 246
bonded incompressible layer, 298
elastic layer on a rigid foundation, 309
frictional problems, 304
I frictionless unbonded layer, 296
Impact, 513–521, 525–529, 533–536 layer on a rigid foundation, 295
duration, 515, 520 multilayered bodies, 313
oblique, 521–529 non-linear layers, 308
of a bar, 529–534, 536 Lennard-Jones force law, 237, 248, 252, 305,
of a cylinder, 519 373
Imperfect thermal contact, 410 Lennard-Jones potential, 233
Impulse, 513 Linear Complementarity (LCP), 4
Impulsive force, 504, 509 Linear Elastic Fracture Mechanics (LEFM),
Incipient sliding, 450 208, 237, 248
Incipient slip, 152, 228 Line contact theory, 102
Incomplete contact. See non-conformal con- Line force, 78, 79, 113
tact
Incompressible material, 298, 311
Incremental solution, 85, 127, 129, 131, 132, M
187 Mach number, 477
Incremental stiffness, 45, 54, 336, 360, 375, Majumdar and Bhushan’s theory, 359
378 Matched asymptotic expansions, 104, 271
bounds on, 378 Mathematical analogies, 44, 46, 47
Indentation problems, 323 Maugis-Dugdale force law, 250, 253
elastodynamic indentation, 504, 506, 507 Maximum surface displacement, 52
Integral equation formulation, 18, 80 Melan’s theorem, 151
Interface energy, 235, 266 Membrane effects, 277
Interface wave, 498 Meyer index, 324
Interference area, 336 Microslip, 110, 171, 191
Interior stress fields, 25 Mindlin’s problem, 170
584 Index
Moving dislocation, 497, 503 Pulse, elastodynamic, 494

Moving heat sources, 400, 463 P-wave, 476
Moving line force, 478, 502
antiplane tangential force, 492
Moving point force, 487, 489 Q
Multilayered bodies, 313 Quasi-static solution, 475, 517
N R
Nominal contact area, 329, 336 Radon transform, 24, 489
Nominal contact pressure, 336 Rail corrugations, 451
Non-conformal contact, 1, 66, 127, 199 Random Midpoint Displacement (RMD),
Non-conformal contact problems, 7 357
Non-dispersive propagation, 478, 494 Random process models, 345
Nonlinear kinematics, 8 Rate problem, 139, 145
Normal point force, 15 Rate-state friction laws, 163
Normal-tangential coupling, 114, 136, 184, Rayleigh function, 488, 509
185, 187, 189, 501 Rayleigh wave, 477, 487, 490
Receding contact, 5, 221, 269
thermoelastic contact, 228
O with friction, 286, 425
Oblique impact, 521––528 Reciprocal theorem, 17, 52
Oscillatory singularities, 117 Reflected wave, 493, 496, 532
Refracted wave, 493––495
Reinforcing ring, 281
P Relaxation damping, 132, 133, 440
Peak curvature, 344 Residual tractions, 155, 157
Peak (of a profile), 343 Restitution, coefficient of, 513, 535
Peclet number, 424, 464––470 Rigid-body displacement, 3, 100, 519
Periodic contact problems, 91, 93 Rigid-body rotation, 87
Manners’ solution, 93 Rolling contact, 433––438, 441, 445–448,
Westergaard’s problem, 96 450–452
Periodic loading, 148, 156, 176, 190 creep velocity, 441
Permanent stick zone, 150, 178 tractive rolling, 441
Persson’s theory, 364, 374, 380
Pin-in-hole problem, 223
interference fit, 229 S
Piston ring, 267 Self-affine fractals, 352
Plane strain, 77, 198 Self-similarity, 16, 316, 327, 508
Plane stress, 77 Separation region, 4
Plastic deformation, 331, 337, 342, 363 Shakedown, 151
Plasticity index, 342 Shape factor, 46, 48–50
Plate, 274, 290 Shear wave, 476
adhesion moment, 277 Shells, 281
stiffness, 274 Shift, 109
Potential function methods, 14, 44, 69, 120, SH-wave, 490
543, 544, 546 Signorini problem, 5, 406
Power-law material, 317, 324 Singular contact tractions, 6, 198
Power Spectral Density (PSD), 347 Sliding, 111, 452–460, 462–470
relation between surface and profile PSD, instability of, 503
350 of rough surfaces, 462
Profile curvature, 348 Sliding velocity, 433
Profilometry, 333 Slip–stick wave, 501
Pull-off force, 236, 244, 249, 250, 370, 375 Slip velocity, 109
Index 585
Slip wave, 499 Thermoelastic Instability (TEI), 418–419,

unstable slip wave, 503 421, 422, 424, 426, 428, 429, 431
Slip-weakening, 162 critical speed, 420
Small-scale yielding, 208, 248 Thermoelastic rod models, 406–413, 419
Smirnov-Sobolev transform, 97, 317, 350 Tilted flat punch, 56
Solidification instability, 415–417 Toughness, 246
Spin, 435 Transonic contact problem, 484
Standard deviation, 335 Transverse wave, 476
State variable, 163 T-stress, 199
Steady-state temperature, 397 Two-dimensional flat punch, 82
Stick, 110
Stick-slip vibrations, 161
Stoneley wave, 499 U
Strain energy, 241, 266 Uniqueness theorems, 5, 139
Stress-intensity factor, 163, 208, 237, 259 Unloading, 155, 175, 189, 226
Stribeck curve, 160
Subsonic contact problem, 480
Summit, 337, 340, 343, 344 V
Summit curvature, 343 Van der Waals forces, 233
Superseismic contact problem, 485, 507, 518 Variance, 346
Surface displacement, 3 Vickers indentation, 324
Surface energy, 235 Volterra integral equation, 130, 189, 506
Surface profile, 333 Voronoi partition, 384
Surface slope, 343, 349, 350, 364
S-wave, 476
W
Wave speeds, 476
T Wave transmitted across an interface, 490–
Tabor parameter, 246 498
Tangential compliance tensor, 126 Wear, 455–457, 459, 462, 473
Theoretical tensile strength, 235 Galin’s eigenfunction method, 459
Thermal contact resistance, 340, 375, 386, Wedge indentation, 205
407 Wedging, 143
Thermal distortivity, 398 Weierstrass function, 354, 362, 364
Thermal rectification, 405 Westergaard’s problem, 96, 249, 362, 372
Thermoelastic contact, 228, 395 Winkler foundation, 70, 260, 290, 295, 307,
imperfect thermal contact, 409 319, 383, 461
Thermoelastic contact stability, 410, 413,
415
Thermoelastic deformation, 396, 403 Y
Thermoelastic Hertzian contact, 402 Yield criteria, 39, 323, 332

2018 Book ContactMechanics PDF

Uploaded by

Copyright:

Available Formats

You might also like

2018 Book ContactMechanics PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

2018 Book ContactMechanics PDF

Uploaded by

Copyright:

Available Formats

Solid Mechanics and Its Applications

More information about this series at http://www.springer.com/series/6557

ISSN 0925-0042 ISSN 2214-7764 (electronic)

© Springer International Publishing AG 2018

Printed on acid-free paper

This Springer imprint is published by Springer Nature

Ann Arbor, USA J. R. Barber

3.1.2 More General Cases . . . . . . . . . . . . . . . . . . . . . . . . . . 32

6.5.5 Manners’ Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

8.3 Multinode Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144

10 Asymptotic Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195

12.4.2 Simpliﬁed Force Laws . . . . . . . . . . . . . . . . . . . . . . . . 252

14.5.1 Exponential Variation of Modulus . . . . . . . . . . . . . . . . 314

16.8 Incremental Stiffness and Contact Resistance . . . . . . . . . . . . . . 375

18 Rolling and Sliding Contact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433

19.4 Interface Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 498

Appendix A: Potential Function Solutions for Elasticity Problems . . . . . 543

© Springer International Publishing AG 2018 1

Fig. 1.1 (a) Conformal and (b) non-conformal contact

1.1 Reference Frame and the Initial Gap Function

g0 (x, y) = g1 (x, y) + g2 (x, y), (1.1)

Fig. 1.2 Coordinate system 2.

1.2 Establishment of a Contact Region

g(x, y) = g0 (x, y) − Δ + u (1) (2)

Fig. 1.3 Establishment of a P

Fig. 1.4 Effect of

1.2.1 Definition of Contact

1.2.2 The Boundary Value Problem

traction at any point on the surface of a deformable body, or to impose a prescribed

1.2.3 Signorini Problems

1.2.4 Asymptotic Arguments

1.2.5 The Discrete Problem

1.3 Nonlinear Kinematics

u z (x, 0) = Δ − x tan α. (1.5)

u z (x, 0) = Δ − [x + u x (x, 0)] tan α or u z (x, 0) + u x (x, 0) tan α = Δ − x tan α.

1.4 Almost Conformal Contact

t(r) = − p(r)n(r), (1.9)

g1 (x) = h 0 [1 + cos(mx)] ; g2 (x) = −h 0 (1 − )[1 + cos(mx)],

(i) the strains are everywhere small in comparison with unity;

2.1 The Half-Space Approximation

© Springer International Publishing AG 2018 13

2.2 Normal Loading of the Half-Space

∂2ϕ ∂2ϕ ∂2ϕ

2.2.1 The Point Force Solution

∂3ϕ ∂2ϕ 3Pz 3

which is independent of h. Thus to maintain equilibrium of the layer, there must be

Fig. 2.2 Concentrated

Fig. 2.3 Equilibrium of the P

where E = 2G(1+ν) is Young’s modulus.

2.2.2 Similarity, Equilibrium and Anisotropy

σi j = f (R)Si j (θ, β); u i = g(R)Ui (θ, β), (2.8)

Fig. 2.4 If a normal force is C(a, θ)

h(θ + π) = h(θ), (2.10)

involving even multiples of θ only, where Am , Bm are constants.

2.2.3 The Composite Elastic Modulus

Fig. 2.5 Surface

Ph (1) (θ) Ph (2) (θ)

2.3 Integral Equation Formulation

g1 (x) = h 0 [1 + cos(mx)] ; g2 (x) = −h 0 (1 − )[1 + cos(mx)],

to the field point in Fig. 2.8b. The transformation then