Martin Starzewski Microstructural Randomness

P1: Binod
July 13, 2007 21:36 C4174 C4174C000
Microstructural
Randomness and
Scaling in
Mechanics of
Materials
P1: Binod
July 13, 2007 21:36 C4174 C4174C000
P1: Binod
July 13, 2007 21:36 C4174 C4174C000
P1: Binod
July 13, 2007 21:36 C4174 C4174C000
Chapman & Hall/CRC

Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742
2008 by Taylor & Francis Group, LLC
Chapman & Hall/CRC is an imprint of Taylor & Francis Group, an Informa business
No claim to original U.S. Government works

Printed in the United States of America on acid-free paper
10 9 8 7 6 5 4 3 2 1
International Standard Book Number-13: 978-1-58488-417-0 (Hardcover)
This book contains information obtained from authentic and highly regarded sources. Reprinted
material is quoted with permission, and sources are indicated. A wide variety of references are
listed. Reasonable eorts have been made to publish reliable data and information, but the author
and the publisher cannot assume responsibility for the validity of all materials or for the conse-
quences of their use.
No part of this book may be reprinted, reproduced, transmitted, or utilized in any form by any
electronic, mechanical, or other means, now known or hereafter invented, including photocopying,
microlming, and recording, or in any information storage or retrieval system, without written
permission from the publishers.
For permission to photocopy or use material electronically from this work, please access www.
copyright.com (http://www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC)
222 Rosewood Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-for-prot organization that
provides licenses and registration for a variety of users. For organizations that have been granted a
photocopy license by the CCC, a separate system of payment has been arranged.
Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and
are used only for identication and explanation without intent to infringe.
Library of Congress Cataloging-in-Publication Data
Ostoja-Starzewski, Martin.
Microstructural randomness and scaling in mechanics of materials / Martin
Ostoja-Starzewski.
p. cm. -- (Modern mechanics and mathematics)
Includes bibliographical references and index.
ISBN-13: 978-1-58488-417-0 (alk. paper)
ISBN-10: 1-58488-417-7 (alk. paper)
1. Strength of materials. I. Title. II. Series.
TA405.O88 2007
620.112--dc22 2007000170
Visit the Taylor & Francis Web site at

http://www.taylorandfrancis.com
and the CRC Press Web site at
http://www.crcpress.com
P1: Binod
July 13, 2007 21:36 C4174 C4174C000
Dedication
To Pauline and Michael
v
P1: Binod
July 13, 2007 21:36 C4174 C4174C000
P1: Binod
July 13, 2007 21:36 C4174 C4174C000
Table of Contents
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvii
About the Author . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxv
1 Basic Random Media Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1 Probability Measure of Geometric Objects . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Definitions of Probability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.2 Probabilities on Countable and Euclidean
Sample Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.1.2.1 is a Countable Set: = {1 , 2 , . . .} . . . . . . . . . . . . 4
1.1.2.2 is a 1D Euclidean Space: = R. . . . . . . . . . . . . . . .7
1.1.2.3 is a 2D Euclidean Space: = R2 . . . . . . . . . . . . . . . 8
1.1.3 Random Points, Lines, and Planes. . . . . . . . . . . . . . . . . . . . . . .11
1.1.3.1 Random Lines in Two Dimensions . . . . . . . . . . . . . . 11
1.1.3.2 Planes in Three Dimensions . . . . . . . . . . . . . . . . . . . . 14
1.1.3.3 Straight Lines in Three Dimensions . . . . . . . . . . . . . 14
1.2 Basic Point Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.2.1 Bernoulli Trials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.2.2 ExampleModel of a Fiber Structure of Paper . . . . . . . . . . 15
1.2.3 Generalization to Many Types of Outcomes . . . . . . . . . . . . . 17
1.2.4 Binomial and Multinomial Point Fields . . . . . . . . . . . . . . . . . 17
1.2.4.1 Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.2.4.2 Simulation of a Binomial Point Field
with n Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.2.4.3 Generalization to a Multinomial Point Field . . . . 20
1.2.5 Bernoulli Lattice Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.2.6 Poisson Point Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.2.6.1 Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.2.6.2 Simulation of a Poisson Point Field . . . . . . . . . . . . . 22
1.2.6.3 Inhomogeneous Poisson Point Field . . . . . . . . . . . . 22
1.2.6.4 Inhibition and Hard-Core Processes . . . . . . . . . . . . 23
1.3 Directional Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.3.1 Basic Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.3.2 Circular Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.4 Random Fibers, Random Line Fields, Tessellations . . . . . . . . . . . . . 27
1.4.1 Poisson Random Lines in Plane . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.4.2 Finite Fiber Field in Plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
vii
P1: Binod
July 13, 2007 21:36 C4174 C4174C000
viii
1.4.3 Random Tessellations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

1.4.3.1 Basic Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.4.3.2 Planar (Poisson-)Voronoi Tessellations
and Delaunay Triangulations . . . . . . . . . . . . . . . . . . . 31
1.4.3.3 Modifications of Voronoi Tessellations . . . . . . . . . . 31
1.4.3.4 Random Crack Model . . . . . . . . . . . . . . . . . . . . . . . . . . 34
1.5 Basic Concepts and Definitions of Random Microstructures . . . . 35
1.5.1 General. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .35
1.5.1.1 Germ-Grain and Boolean Models . . . . . . . . . . . . . . . 35
1.5.1.2 Flocs as Continua . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
1.5.2 Toward Mathematical Morphology . . . . . . . . . . . . . . . . . . . . . 38
2 Random Processes and Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

2.1 Elements of One-Dimensional Random Fields. . . . . . . . . . . . . . . . . .45
2.1.1 Scalar Random Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.1.1.1 Basic Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.1.1.2 Homogeneity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.1.2 Vector Random Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.2 Mechanics Problems on One-Dimensional Random Fields . . . . . . 55
2.2.1 Propagation of Surface Waves along
Random Boundaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.2.2 Fracture of Brittle Microbeams . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.2.2.1 Randomness of Microbeams . . . . . . . . . . . . . . . . . . . . 57
2.2.2.2 Strain Energy Release Rate in Random
Microbeams. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .58
2.2.2.3 Stochastic Crack Stability . . . . . . . . . . . . . . . . . . . . . . . 61
2.3 Elements of Two- and Three-Dimensional Random Fields . . . . . . 62
2.3.1 Scalar and Vector Fields. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .62
2.3.1.1 Basic Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.3.1.2 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
2.3.1.3 Properties of the Correlation Function
Anisotropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
2.3.2 Random Tensor Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
2.3.2.1 Second-Rank Tensor Fields . . . . . . . . . . . . . . . . . . . . . 69
2.3.2.2 Fourth-Rank Tensor Fields . . . . . . . . . . . . . . . . . . . . . 71
2.4 Mechanics Problems on Two- and Three-Dimensional
Random Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
2.4.1 Mean Field Equations of Random Materials . . . . . . . . . . . . . 72
2.4.2 Mean Field Equations of Turbulent Media . . . . . . . . . . . . . . . 73
2.5 Ergodicity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
2.5.1 Basic Considerations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .75
2.5.2 Computation of (2.146) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
2.5.3 Conditions for (2.146) to Hold . . . . . . . . . . . . . . . . . . . . . . . . . . 76
2.5.4 Existence of the Limit in (2.146) . . . . . . . . . . . . . . . . . . . . . . . . . 76
P1: Binod
July 13, 2007 21:36 C4174 C4174C000
ix
2.6 The Maximum Entropy Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

2.6.1 Cracks in Plates with Holes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
2.6.2 Disorder and Information Entropy . . . . . . . . . . . . . . . . . . . . . . 80
3 Planar Lattice Models: Periodic Topologies and Elastostatics . . . . . . . 87

3.1 One-Dimensional Lattices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.1.1 Simple Lattice and Elastic Strings . . . . . . . . . . . . . . . . . . . . . . . 87
3.1.2 Micropolar Lattice and Elastic Beams . . . . . . . . . . . . . . . . . . . 88
3.2 Planar Lattices: Classical Continua . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
3.2.1 Basic Idea of a Spring Network Representation . . . . . . . . . . 91
3.2.2 Antiplane Elasticity on Square Lattice . . . . . . . . . . . . . . . . . . . 93
3.2.3 In-Plane Elasticity: Triangular Lattice with
Central Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
3.2.4 In-Plane Elasticity: Triangular Lattice with Central
and Angular Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
3.2.5 Triple Honeycomb Lattice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
3.3 Applications in Mechanics of Composites . . . . . . . . . . . . . . . . . . . . . . 99
3.3.1 Representation by a Fine Mesh . . . . . . . . . . . . . . . . . . . . . . . . . . 99
3.3.2 Solutions of Linear Algebraic Problems . . . . . . . . . . . . . . . . 102
3.3.3 Example: Simulation of a Polycrystal . . . . . . . . . . . . . . . . . . 105
3.4 Planar Lattices: Nonclassical Continua . . . . . . . . . . . . . . . . . . . . . . . . 107
3.4.1 Triangular Lattice of BernoulliEuler Beams . . . . . . . . . . . 107
3.4.2 Triangular Lattice of Timoshenko Beams . . . . . . . . . . . . . . . 110
3.4.3 From Stubby Beams to a Perforated Plate Model . . . . . . . 112
3.4.4 Hexagonal Lattice of BernoulliEuler Beams . . . . . . . . . . . 114
3.4.5 Square Lattice of BernoulliEuler Beams . . . . . . . . . . . . . . . 115
3.4.6 Nonlocal and Gradient Elasticity on a Lattice with
Central Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
3.4.6.1 General Considerations . . . . . . . . . . . . . . . . . . . . . . . 117
3.4.6.2 Local Continuum Model . . . . . . . . . . . . . . . . . . . . . . 118
3.4.6.3 Nonlocal Continuum Model . . . . . . . . . . . . . . . . . . .118
3.4.6.4 Strain-Gradient Continuum Model . . . . . . . . . . . . 119
3.4.7 Plate-Bending Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
3.5 Extension-Twist Coupling in a Helix . . . . . . . . . . . . . . . . . . . . . . . . . . 122
3.5.1 Constitutive Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
3.5.2 Harmonic Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
3.5.2.1 Elastic Helix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
3.5.2.2 Thermoelastic Helix . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
3.5.3 Viscoelastic Helices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
3.5.3.1 Viscoelasticity with Integer-Order
Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
3.5.3.2 Viscoelasticity with Fractional-Order
Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
P1: Binod
July 13, 2007 21:36 C4174 C4174C000
4 Lattice Models: Rigidity, Randomness, Dynamics,

and Optimality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
4.1 Rigidity of Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
4.1.1 Structural Topology and Rigidity Percolation . . . . . . . . . . 133
4.1.2 Application to Cellulose Fiber Networks . . . . . . . . . . . . . . . 137
4.1.2.1 Rigidity of a Graph of Poisson Line
Field Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
4.1.2.2 Loss of Rigidity in a Fiber-Beam Network . . . . . 139
4.2 Spring Network Models for Disordered Topologies . . . . . . . . . . . 141
4.2.1 Granular-Type Media . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
4.2.1.1 Load Transfer Mechanisms . . . . . . . . . . . . . . . . . . . . 141
4.2.1.2 Graph Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
4.2.1.3 Periodic Graphs with Disorder . . . . . . . . . . . . . . . . 145
4.2.2 Solutions of Truss Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
4.2.3 Mesoscale Elasticity of Paper . . . . . . . . . . . . . . . . . . . . . . . . . . 150
4.2.3.1 Dilemma of Special In-Plane Orthotropy
of Paper . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
4.2.3.2 Explanation via Random Fiber Network . . . . . . . 153
4.2.4 Damage Patterns and Maps of Disordered
Composites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
4.3 Particle Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
4.3.1 Governing Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
4.3.1.1 Basic Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
4.3.1.2 Leapfrog Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
4.3.2 Examples. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .162
4.3.2.1 Other Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
4.4 Michell Trusses: Optimal Use of Material . . . . . . . . . . . . . . . . . . . . . 165
4.4.1 Study via Hyperbolic System . . . . . . . . . . . . . . . . . . . . . . . . . . 165
4.4.1.1 Governing Equations . . . . . . . . . . . . . . . . . . . . . . . . . 165
4.4.1.2 Example of an Optimal Layout . . . . . . . . . . . . . . . . 167
4.4.2 Study via Elliptic System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
5 Two- Versus Three-Dimensional Classical Elasticity . . . . . . . . . . . . . . 171

5.1 Basic Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
5.1.1 Isotropic Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
5.1.1.1 Three-Dimensional Elasticity . . . . . . . . . . . . . . . . . . 171
5.1.1.2 Two-Dimensional Elasticity . . . . . . . . . . . . . . . . . . . 172
5.1.2 Plane Elasticity Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
5.1.3 Special Planar Orthotropies . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
5.2 The CLM Result and Stress Invariance . . . . . . . . . . . . . . . . . . . . . . . . 177
5.2.1 Isotropic Materials. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .177
5.2.1.1 Basic Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
5.2.1.2 Two-Phase Composites and Dundurs
Constants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
5.2.2 Anisotropic Materials and the Null-Lagrangian . . . . . . . . 180
P1: Binod
July 13, 2007 21:36 C4174 C4174C000
xi
5.2.3 Multiply Connected Materials . . . . . . . . . . . . . . . . . . . . . . . . . 181

5.2.4 Applications to Composites. . . . . . . . . . . . . . . . . . . . . . . . . . . .183
5.2.4.1 Effective Moduli of Composites . . . . . . . . . . . . . . . 183
5.2.4.2 Plates with Holes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
5.2.5 Extension of Stress Invariance to Presence
of Eigenstrains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
5.2.5.1 Basic Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
5.2.5.2 Planar Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
5.3 Poroelasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
6 Two- versus Three-Dimensional Micropolar Elasticity . . . . . . . . . . . . 191

6.1 Micropolar Elastic Continua . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
6.1.1 Force Transfer and Degrees of Freedom . . . . . . . . . . . . . . . . 191
6.1.2 Equations of Motion and Constitutive Equations . . . . . . . 194
6.1.3 Isotropic Micropolar Materials . . . . . . . . . . . . . . . . . . . . . . . . . 196
6.1.4 Virtual Work Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
6.1.5 Hamiltons Principle. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .199
6.1.6 Reciprocity Relation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
6.1.7 Elements of Micropolar Elastodynamics . . . . . . . . . . . . . . . 202
6.1.7.1 Basic Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
6.1.7.2 Plane Monochromatic Waves . . . . . . . . . . . . . . . . . . 204
6.1.8 Noncentrosymmetric Micropolar Elasticity. . . . . . . . . . . . .204
6.2 Classical vis-a-vis Nonclassical (Elasticity) Models . . . . . . . . . . . . . 205
6.2.1 A Brief History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
6.2.2 The Ensemble Average of a Random Local
Medium is Nonlocal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
6.3 Planar Cosserat Elasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
6.3.1 First Planar Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
6.3.2 Characteristic Lengths in Isotropic and
Orthotropic Media . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
6.3.3 Restricted Continuum vis-a-vis the Micropolar
Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
6.4 The CLM Result and Stress-Invariance . . . . . . . . . . . . . . . . . . . . . . . . 216
6.4.1 Isotropic Materials. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .216
6.4.2 Anisotropic Materials and the Null-Lagrangian . . . . . . . . 218
6.4.3 Multiply Connected Materials . . . . . . . . . . . . . . . . . . . . . . . . . 220
6.4.4 Applications to Composites. . . . . . . . . . . . . . . . . . . . . . . . . . . .221
6.4.4.1 Two-Phase Materials . . . . . . . . . . . . . . . . . . . . . . . . . . 221
6.4.4.2 Effective Moduli of Composites . . . . . . . . . . . . . . . 222
6.4.5 Extensions of Stress Invariance to Presence of
Eigenstrains and Eigencurvatures . . . . . . . . . . . . . . . . . . . . . . 223
6.4.5.1 Basic Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
6.4.5.2 Inhomogeneous Materials . . . . . . . . . . . . . . . . . . . . . 225
6.5 Effective Micropolar Moduli and Characteristic Lengths
of Composites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
P1: Binod
July 13, 2007 21:36 C4174 C4174C000
xii
6.5.1 From a Heterogeneous Cauchy to a Homogeneous

Cosserat Continuum. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .228
6.5.2 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
7 Mesoscale Bounds for Linear Elastic Microstructures . . . . . . . . . . . . . 237

7.1 Micro-, Meso-, and Macroscales. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .237
7.1.1 Separation of Scales . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
7.1.2 Basic Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
7.1.3 The RVE Postulate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
7.2 Volume Averaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
7.2.1 A Paradigm of Boundary Conditions Effect . . . . . . . . . . . . 241
7.2.2 The Hill Condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
7.2.2.1 Mechanical versus Energy Definitions . . . . . . . . . 244
7.2.2.2 Order Relations Dictated by Three Types
of Loading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
7.2.3 Apparent Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
7.3 Spatial Randomness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
7.3.1 Stationarity of Spatial Statistics . . . . . . . . . . . . . . . . . . . . . . . . 249
7.3.2 Ergodicity of Spatial Statistics. . . . . . . . . . . . . . . . . . . . . . . . . .251
7.4 Hierarchies of Mesoscale Bounds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
7.4.1 Response under Displacement Boundary Condition . . . . 251
7.4.2 Response under Traction Boundary Condition . . . . . . . . . 254
7.4.3 Scale-Dependent Hierarchy . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
7.4.4 Homogenization Theory Viewpoint . . . . . . . . . . . . . . . . . . . . 256
7.4.5 Apparent Moduli in In-Plane Elasticity . . . . . . . . . . . . . . . . 257
7.4.5.1 General Considerations . . . . . . . . . . . . . . . . . . . . . . . 257
7.5 Examples of Hierarchies of Mesoscale Bounds . . . . . . . . . . . . . . . . 258
7.5.1 Random Chessboards and Bernoulli Lattices . . . . . . . . . . . 258
7.5.2 Disk-Matrix Composites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
7.5.3 Functionally Graded Materials. . . . . . . . . . . . . . . . . . . . . . . . .263
7.5.4 Effective and Apparent Moduli of Multicracked
Solids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
7.5.4.1 Scale-Dependent Hierarchies of Bounds:
Numerical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
7.5.4.2 Cross-Correlations of the Mesoscale
Moduli with the Crack Density Tensor . . . . . . . . 268
7.6 Moduli of the Trabecular Bone . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
8 Random Field Models and Stochastic Finite Elements . . . . . . . . . . . . 273

8.1 Mesoscale Random Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
8.1.1 From Discrete to Continuum Random Fields . . . . . . . . . . . 273
8.1.2 Scale Dependence via Beta Distribution . . . . . . . . . . . . . . . . 275
8.1.3 Mesoscopic Continuum Physics Due to Muschik . . . . . . . 276
8.2 Second-Order Properties of Mesoscale Random Fields . . . . . . . . 278
8.2.1 Governing Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
P1: Binod
July 13, 2007 21:36 C4174 C4174C000
xiii
8.2.2 Universal Properties of Mesoscale Bounds . . . . . . . . . . . . . 279

8.2.3 Correlation Structure of Mesoscale Random Fields . . . . . 282
8.3 Does There Exist a Locally Isotropic, Smooth
Elastic Material?. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .284
8.3.1 Correlation Theory Viewpoint . . . . . . . . . . . . . . . . . . . . . . . . . 284
8.3.2 Micromechanics Viewpoint . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
8.3.3 Closure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .286
8.4 Stochastic Finite Elements for Elastic Media . . . . . . . . . . . . . . . . . . . 286
8.4.1 Bounds on Global Response . . . . . . . . . . . . . . . . . . . . . . . . . . . 286
8.4.2 Example: Torsion of a Duplex-Steel Bar . . . . . . . . . . . . . . . . 288
8.4.3 An Overview of Phenomenological SFE Studies . . . . . . . . 290
8.5 Method of Slip-Lines for Inhomogeneous Plastic Media . . . . . . . 293
8.5.1 Finite Difference Spacing vis-a-vis Grain Size . . . . . . . . . . . 293
8.5.2 Sensitivity of Boundary Value Problems
to Randomness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
8.5.2.1 Cauchy Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
8.5.2.2 Limit Analysis of a Pipe under
Internal Loading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
8.6 Michell Trusses in the Presence of Random Microstructure . . . . 301
8.6.1 Truss-Like Continuum vis-a-vis Random
Microstructure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
8.6.2 Solution via the Hyperbolic System . . . . . . . . . . . . . . . . . . . . 303
8.6.3 Solution via the Elliptic System . . . . . . . . . . . . . . . . . . . . . . . . 305
9 Hierarchies of Mesoscale Bounds for Nonlinear

and Inelastic Microstructures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311
9.1 Physically Nonlinear Elastic Microstructures . . . . . . . . . . . . . . . . . . 311
9.1.1 General. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .311
9.1.2 Power-Law Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313
9.1.3 Random Formation vis-a-vis Inelastic Response
of Paper . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
9.2 Finite Elasticity of Random Composites . . . . . . . . . . . . . . . . . . . . . . . 318
9.2.1 Averaging Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318
9.2.2 Variational Principles and Mesoscale Bounds . . . . . . . . . . 319
9.3 Elastic-Plastic Microstructures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
9.3.1 Variational Principles and Mesoscale Bounds . . . . . . . . . . 322
9.3.2 Matrix-Inclusion Composites . . . . . . . . . . . . . . . . . . . . . . . . . . 324
9.3.2.1 Case 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
9.3.2.2 Case 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
9.3.3 Geodesic Properties of Shear-Band Patterns . . . . . . . . . . . . 329
9.4 Rigid-Perfectly Plastic Microstructures . . . . . . . . . . . . . . . . . . . . . . . . 333
9.4.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
9.4.2 Bounding on Mesoscales via Kinematic
and Traction Boundary Conditions . . . . . . . . . . . . . . . . . . . . 334
9.4.3 Random Chessboard of Hubervon Mises Phases . . . . . . 336
P1: Binod
July 13, 2007 21:36 C4174 C4174C000
xiv
9.5 Viscoelastic Microstructures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341

9.6 Stokes Flow in Porous Media . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342
9.7 Thermoelastic Microstructures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347
9.7.1 Linear Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347
9.7.2 The Nonlinear Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352
9.8 Scaling and Stochastic Evolution in Damage Phenomena . . . . . . 354
9.9 Comparison of Scaling Trends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357
10 Mesoscale Response in Thermomechanics of Random Media . . . . 359

10.1 From Statistical Mechanics to Continuum
Thermodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
10.1.1 Dissipation Function of the RVE . . . . . . . . . . . . . . . . . . . . . 359
10.1.2 Departure from the Second Law of
Thermodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 364
10.2 Extensions of the Hill Condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 364
10.2.1 The Hill Condition in Thermomechanics . . . . . . . . . . . . . 364
10.2.2 Homogenization in Dynamic Response . . . . . . . . . . . . . . 366
10.3 Legendre Transformations in (Thermo)Elasticity . . . . . . . . . . . . . 367
10.3.1 Elasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 367
10.3.2 Thermoelasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 367
10.4 Thermodynamic Orthogonality on the Mesoscale . . . . . . . . . . . . 370
10.4.1 General . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 370
10.4.2 Homogeneous Dissipation Functions . . . . . . . . . . . . . . . . 371
10.4.3 Quasi-Homogeneous Dissipation Functions . . . . . . . . . 374
10.5 Complex versus Compound Processes:
The Scaling Viewpoint . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 376
10.5.1 General Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 376
10.5.2 Micropolar Plasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379
10.6 Toward Continuum Mechanics of Fractal Media . . . . . . . . . . . . . 380
11 Waves and Wavefronts in Random Media. . . . . . . . . . . . . . . . . . . . . . . . .385

11.1 Basic Methods in Stochastic Wave Propagation . . . . . . . . . . . . . . .386
11.1.1 The Long Wavelength Case . . . . . . . . . . . . . . . . . . . . . . . . . . 386
11.1.1.1 Elementary Considerations . . . . . . . . . . . . . . . . 386
11.1.1.2 Series Expansion . . . . . . . . . . . . . . . . . . . . . . . . . . .387
11.1.2 The Short Wavelength Case: Ray Method . . . . . . . . . . . . 390
11.1.2.1 Fermats Principle . . . . . . . . . . . . . . . . . . . . . . . . . 390
11.1.2.2 Markov Character of Rays . . . . . . . . . . . . . . . . . 392
11.1.3 The Short Wavelength Case: Rytov Method . . . . . . . . . . 393
11.2 Toward Spectral Finite Elements for Random Media . . . . . . . . . 395
11.2.1 Spectral Finite Element for Waves in Rods . . . . . . . . . . . 395
11.2.1.1 Deterministic Case . . . . . . . . . . . . . . . . . . . . . . . . . 395
11.2.1.2 Random Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 396
P1: Binod
July 13, 2007 21:36 C4174 C4174C000
xv
Spectral Finite Element for Flexural Waves . . . . . . . . . . . 398

11.2.2
11.2.2.1 Deterministic Case . . . . . . . . . . . . . . . . . . . . . . . . . 398
11.2.2.2 Random Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 400
11.2.3 Observations and Related Work . . . . . . . . . . . . . . . . . . . . . 402
11.3 Waves in Random 1D Composites . . . . . . . . . . . . . . . . . . . . . . . . . . . 403
11.3.1 Motion in an Imperfectly Periodic Composite . . . . . . . . 403
11.3.1.1 Random Evolutions. . . . . . . . . . . . . . . . . . . . . . . .403
11.3.1.2 Effects of Imperfections on
Floquet Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 404
11.3.2 Waves in Randomly Segmented Elastic Bars . . . . . . . . . 407
11.4 Transient Waves in Heterogeneous Nonlinear Media . . . . . . . . . 408
11.4.1 A Class of Models of Random Media . . . . . . . . . . . . . . . . 408
11.4.2 Pulse Propagation in a Linear Elastic
Microstructure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 410
11.4.3 Pulse Propagation in Nonlinear Microstructures . . . . . 414
11.4.3.1 Bilinear Elastic Microstructures . . . . . . . . . . . . 414
11.4.3.2 Nonlinear Elastic Microstructures . . . . . . . . . . 417
11.4.3.3 Hysteretic Microstructures . . . . . . . . . . . . . . . . . 419
11.5 Acceleration Wavefronts in Nonlinear Media . . . . . . . . . . . . . . . . 422
11.5.1 Microscale Heterogeneity versus
Wavefront Thickness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
11.5.1.1 Basic Considerations . . . . . . . . . . . . . . . . . . . . . . . 422
11.5.1.2 Mesoscale Response . . . . . . . . . . . . . . . . . . . . . . . 425
11.5.2 Wavefront Dynamics in Random Microstructures . . . . 427
11.5.2.1 Model with One White Noise . . . . . . . . . . . . . . 427
11.5.2.2 Model with Two Correlated
Gaussian Noises . . . . . . . . . . . . . . . . . . . . . . . . . . . 429
11.5.2.3 Model with Four Correlated Noises . . . . . . . . 432
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .435
Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 459
Subject Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 465

P1: Binod
July 13, 2007 21:36 C4174 C4174C000
P1: Binod
July 13, 2007 21:36 C4174 C4174C000
Preface
0.1 Randomness versus Determinism

Partial differential equations form basic models of the mechanics of materials.
Thus, typically, we have a deterministic field equation of the form
Lu = f, (0.1)
where L is a differential operator, f is a source or forcing function, and u is

a solution field. This needs to be accompanied by appropriate boundary and
initial conditions. [Throughout this book we shall interchangeably use the
symbolic (u) and the subscript (ui... ) notations for tensors, as the need arises;
also an overdot will mean d/dt.]
There are three basic ways in which the randomness, indicated by the
dependence on an elementary event (a sample space), can be introduced
into (0.1):
1. Randomness of the operator:
L () u = f. (0.2)
2. Randomness of the forcing function:
Lu = f (). (0.3)
3. Randomness of the boundary and/or initial conditions.
Of course, several combinations of these three basic cases are also possible.
However, in this book we shall focus primarily on the first case, which is
naturally dictated by the presence of imperfect, disorderedi.e., random, in
the ensemble sensematerial microstructures. In this case, the coefficients of
L (), such as the elastic moduli C (Cijkl ), form a tensor-valued random field,
and the stochastic equation (0.1) governs the response of a random medium
(or random material) B. The latter is taken as a set of all the realizations B ()
parametrized by sample events of the space
B = {B (); }. (0.4)
xvii
P1: Binod
July 13, 2007 21:36 C4174 C4174C000
xviii
Each of the realizations follows deterministic laws of classical mechanics in

that it is a specific heterogeneous material sample; probability is introduced to
deal with the set (0.4). Case (1), (2), (3), or any combination thereof, constitutes
the stochastic mechanics.
When speaking of the material spatial randomness, the key issue is the
interpretation of some noise-like observation (measurement/signal), taken
over a certain space and/or time domain, as a realization of a random phe-
nomenon. This is, in essence, an epistemological issue: is the physical world,
and the particular phenomenon at hand, a deterministic or a random event?
In principle, any given physical phenomenonsuch as the spatial variability
of, say, elastic moduli on a millimeter scale of the particular sheet of paper on
which this text is typedis unique, and therefore, deterministic, albeit non-
uniform (disordered). We make an extension to the space of many sheets
(i.e., ensemble) to deal with fluctuations in a statistical manner. Thus, in the
view of this authorand consistent with the prevailing thinking in stochastic
mechanicsthe random field of moduli is chosen so as to reflect our inability
to obtain a precise mathematical model of spatial composition of the material
in each and every realization. Our focus is on what physicists call quenched
randomness.
Alternatively, the choice to work with a random field (or just a set of
random variables) may reflect our preference to discard a huge amount of
information on space- or time-dependent fluctuations, and work, instead,
with a statistical model. However, we consider each and every realization, no
matter how complex, to behave according to the laws of deterministic me-
chanics. It is the objective of stochastic mechanics to determine this behavior:
be it -by- or in terms of a distribution over . Thus, a materials evolution
is stochastic in the sense that there is an ensemble of deterministic specimens.
This approach is in line with the foundations of both probability theory and
classical statistical mechanics (Primas, 1999), as well as with the principle
of determinism (or causality) of continuum mechanics, according to which
knowledge of the history of the material up to time t yields knowledge of the
state of stresses at t (Truesdell and Noll, 1965).
0.2 Randomness and Scales in Mechanics of Materials

Material spatial randomness forces one to reexamine various basic concepts
and results of solid mechanics. Figure 0.1 shows three levels: (a) the microscale
d, (b) the mesoscale L, and (c) the macroscale L macro . The reason for introducing
the intermediate level (b)note that meso, just as micro and macro, comes
from Greekis that we cannot claim a priori that the domain sizes involved
in (a) are sufficient to homogenize the actual microstructures depicted there
and replace them by uniform continua in (c), such as typically involved in
P1: Binod
July 13, 2007 21:36 C4174 C4174C000
xix
(a)
(b)
(c)
FIGURE 0.1
(a) A Boolean model of a microlayered material; (b) a mesoscale continuum approximation,
modeling smoothly inhomogeneous medium by placing everywhere a mesoscale window in the
microstructure of (a); (c) a macroscopic body.
conventional solid mechanics. This is the essence of the so-called separation of

scales

d
L L macro . (0.5)
d<
The first inequality allows one to postulate the existence of a representative

volume element (RVE) of continuum mechanics. The second one covers the
range of length scales where conventional continuum mechanics applies
this is the domain of spatial dependence of stress, strain and displacement
fields one is interested in when solving a boundary value problem. This is
P1: Binod
July 13, 2007 21:36 C4174 C4174C000
xx
where (0.1), with L being a deterministic operator, applies; the classical Navier
equation is an example.
When the RVE does not apply, one is faced with fluctuations, and equation
(0.2) applies, perhaps with the mesoscale randomness represented as smooth
noise about the average stiffness tensor in Figure 0.1(b).
Note: Problems lacking separation of scales occur in many situations,
ranging from micro/nano-electromechanical systems (MEMS/NEMS) to
geophysics.
Note: In fluid mechanics one uses the Knudsen number, a dimensionless
number defined as the ratio of the molecular mean free path length to a
representative physical length scale L (e.g., the radius of a body in a fluid)
K n = /L. Generally, when K n is on the order of one, the determinis-
tic continuum model of fluid mechanics should be replaced by a statistical
approach.
Suppose we want to determine the ensemble average response u. By
inverting (0.2), ensemble averaging and inverting it back again, we find that
u is governed by
1
L1 u = f. (0.6)
This begs two questions.
1. How do we determine the operator L1 1 ?

2. Can one replace L1 1 by L? That is, what is the difference
between an average solution of the stochastic problem governed
by (0.6) and a deterministic problem obtained by straightforward
averaging of (0.7)? This replacement, often implied by deterministic
continuum mechanics models, corresponds to
L u = f. (0.7)
This is the basic question of stochastic mechanics for, if there is no dif-

ference in a given problem, then one may safely work with its deterministic
counterpart. However, theoretical and applied mechanics are ripe with exam-
ples to the contrary. Buckling and random vibration are perhaps the most
classical instances where fluctuations cause significant changes from the sit-
uations governed by homogeneous properties and nice inputs equal to the
averages of random ones.
Various other questions arise here:
3. How does one pass from the microscale d to the mesoscale L, so

as to remove the mesoscale fluctuations shown as the grayscale
variability in (b)? In other words, unless we reach the RVE, we are
faced with a continuum-type random field.
P1: Binod
July 13, 2007 21:36 C4174 C4174C000
xxi
4. Supposing we deal with a periodic or disordered lattice-type

microstructure instead of the piecewise-constant continuum in
(a), what homogeneous continuum models are then dictated?
5. If we are below the RVE scale, how is the continuum random field
in (b) defined? Is a locally isotropic random field admissible?
6. What universal statements can be made about the finite sizescaling
and convergence of statistical volume element (SVE) domains, such
as the one in Figure 0.1(b), to the RVE?
7. How can one proceed with the homogenization in the case of mate-
rials having high, or very high, mismatch in phase properties? Note
that in those situations the classical homogenization studies, such
as the VoigtReuss, HashinShtrikman bounds and their improve-
eff
ments, yield very wide bounds on Cijkl . Another dilemma arises
when the material to be studied is of an inhomogeneous type, with-
out clearly distinguishable phases, such as shown in Figure 0.1(c).
Paper on millimeter to centimeter scales is a perfect example of
this for, even though it is well known that paper is made of a cellu-
lose fiber network, the latter displays random grayscale fluctuations
directly seen with the naked eye. However, as we move the sheet far
away from the eye, the fluctuations vanish and the sheet begins to
look homogeneous. Here then we have another motivation to study
mesoscale effects.
8. What is the possibility of solving arbitrary boundary value problems
of random fields of elastic and inelastic properties governed by (0.2)?
9. What techniques and what results apply when one deals with the
inertia term in (0.2)?
10. Can one set up continuum-type field equations for a fractal micro-
structure where there is no separation of scales whatsoever?
11. How can we analyze wave dynamics in random elastic and/or
inelastic media?
0.3 Outline of Contents

Chapter 1 begins with simple aspects of discrete random processes and
random geometry and moves on to construction of stochastic models of
real microstructuressuch as composites, polycrystals, granular, cellular, or
fibrous media. By contrast, Chapter 2 focuses on random processes and fields
having continuous realizations. These two chapters are written as introduc-
tions and guides for nonspecialists who come from deterministic contin-
uum mechanics and need to acquire or recall some tools useful in stochastic
mechanics. A deliberate decision was made to only briefly present probability,
P1: Binod
July 13, 2007 21:36 C4174 C4174C000
xxii
random variables and processes, very classical topics that are expertly and
extensively covered in many other books. Our focus is on a basic introduction
to random geometry, as well as random processes and fields, because these
subjects are scattered in very different places in the literature. Several prob-
lems are given in these chapters so as to develop basic probabilistic skills
in mechanics students and recall some elementary concepts from statistical
physics.
An analogous foundational rolealbeit on the mechanics sideis played
by Chapters 3 and 4, where we outline planar lattice (spring) networks: first
periodic, then disordered. These constructs have their roots in condensed mat-
ter physics and structural mechanics, and over the past few decades provided
computational mechanics models in composite materials and granular me-
dia. A brief presentation of one-dimensional (1D) models is also made. Spring
network models are naturally suited for materials with discrete topology, but
they also offer a powerful tool for the representation of piecewise-constant
continuous media. They work best in two dimensions (2D), and their con-
tinuum counterpart is the planar elasticity discussed in Chapter 5, where we
discuss primarily the so-called stress-invariance. Chapter 6 continues this sub-
ject in the setting of micropolar elasticity; it also covers several other topics
not yet collected in book form, including the passage from a microstructure
to an effective micropolar continuum.
Chapters 3 through 6 also contain problems, so that the book can be used
as a text in a graduate course. Indeed, it is with this combination of sev-
eral different stochastic and mechanics methods that one can approach prob-
lems involving microstructural randomness and scaling. We thus move to
Chapters 7 through 11, which present various applications reflecting our own
interests. However, it is felt that, in spite of this personal tinge and scope in
the presentation, the problems are representative of some important research
directions in (micro)mechanics of random media.
In particular, Chapter 7 elaborates the finite-size scaling to a classical rep-
resentative volume element (RVE) and the coupled dependence of moduli on
scale and boundary conditions in the sense of Hill (1963). The focus is on linear
elastic microstructures. A basic role is played here by a mesoscale window,
which may also be called a statistical volume element (SVE). In Chapter 8 we
use this approach as a basis for mechanics problems lacking the separation of
scales. Thus, we first consider the determination of mesoscale random con-
tinuum fields, which, in turn, provide a stepping stone to micromechanically
based stochastic finite elements (SFE), slip-lines, and optimal trusses. Here,
perhaps, we have a stochastic version of multiscale computational mechanics.
Chapter 9 continues the scaling issues of Chapter 7 in more challeng-
ing areas: nonlinear elastic and inelastic materials, including a generalization
of thermodynamics with internal variables to random media. It is shown
again that the SVE rather than the conventional RVE plays an important
role here. This leads to a stochastic formulation of thermomechanics with
internal variables in Chapter 10. Although the book focuses on nonfractals,
Chapter 10s last section discusses continuum-type equations of fractal media.
P1: Binod
July 13, 2007 21:36 C4174 C4174C000
xxiii
Chapter 11 focuses on selected topics in wave propagation in random

media. Among others, we look at wavefronts, which, given the microscale
material randomness, are seen as windows of finite thickness rather than
classical discontinuity surfacesyet another paradigm for the SVE concept.
We also report some research on homogenization problems for wave propa-
gation, namely: SFE in the frequency domain, waves in randomly perturbed
imperfect periodic structures, and waves in granular media with nonlinear
elastic interactions.
In fact, it may be a mixed blessing of stochastic mechanics that, depend-
ing on the problem, a special subfield of stochastic mathematics is usually
needed. Overall, depending on the readers interests, the book may require
background in several fields:
1. Probability and random variables roughly at the level of Part 1 of

Papoulis (1984).
2. Standard continuum (thermo)mechanics (e.g., Ziegler, 1983).
3. Computational methods (finite differences and elements) at the first
graduate course level.
The book is an outgrowth of graduate courses on micromechanics, me-

chanics of random media, and probabilistic mechanics that I taught at several
universities over the past two decades. The invitation by D.Y. Gao and
R.W. Ogden to contribute to their book series has been an honor and I hope
that the book meets their expectations. The book would not have materialized
in its final form and time frame without the continuing support of Bob Stern
and Theresa Delforn at Taylor & Francis.
It is a pleasure to acknowledge past and present coworkersmany of them
my studentswho, through cooperation with me, contributed to this book.
My understanding of various randomness and scaling issues in mechanics
of materials has been enriched by discussions with many people, including
E. Altus, A. Beaudoin, M.J. Beran, J. Carmeliet, A. Chudnovsky, I. Elishakoff,
J. Engelbrecht, S. Forest, J.D. Goddard, M. Grigoriu, R.J. Hill, C. Huet, J.T.
Jenkins, D. Jeulin, B. Kunin, G.A. Maugin, J.J. McCoy, D.L. McDowell, M.
Micunovic, V. Mizel, V.F. Nesterenko, M. Pfuff, S.L. Phoenix, A. Pineau, M.-J.
Pindera, K. Sab, J. Schicker, S. Schmauder, J. Schulte, K. Sobczyk, D.C. Stahl, A.
S. Torquato, J. Trebicki,
Swiech, N.J. Triantafyllidis, P. Trovalusci, C.L. Tucker,
III, O. Vinogradov, J.R. Willis, A. Zaoui, and A. Zubelewicz. Finally, my special
warm thanks go to Iwona Jasiuk, my wife, who supported the idea of writing
this book from the outset, while our joint research and her input (including
joint work on Chapter 5) substantially helped improve it.
My researchand therefore, indirectly, this bookhas been made possible
by various funding sources over the past decade or so. Principally, these
include the AES, AFOSR, CFI, COREM, NSERC, NSF, ONR, USACE, USDA,
and I am very grateful for their support.
P1: Binod
July 13, 2007 21:36 C4174 C4174C000
xxiv
Bibliography
Cristescu, N.D., Craciun, E.-M., and Soos, E. (2003), Mechanics of Elastic Composites,
Chapman & Hall/CRC Press, Boca Raton, FL.
Hill, R. (1963), Elastic properties of reinforced solids: Some theoretical principles, J.
Mech. Phys. Solids 11, 357372.
Papoulis, A. (1984), Probability, Random Variables, and Stochastic Processes, McGraw-Hill,
New York.
Primas, H. (1999), Basic elements and problems of probability theory, J. Sci. Explor.
13(4), 579613.
Truesdell, C. and Noll, W. (1965), The Non-Linear Field Theories of Mechanics, Hand-
buch der Physik, Vol. III/3, Springer-Verlag, Berlin.
Ziegler, H. (1983), An Introduction to Thermomechanics, North-Holland, Amsterdam.
P1: Binod
July 13, 2007 21:36 C4174 C4174C000
About the Author
Martin Ostoja-Starzewski is a Professor of Mechanical Science and Engineer-

ing at the University of Illinois at Urbana-Champaign. He did his undergrad-
uate studies at the Cracow University of Technology, followed by Masters
and Ph.D. degrees at McGill University, all in mechanical engineering. He has
been a visiting scientist at Cornell University, the Institute for Mechanics and
Materials at UCSD, Ecole des Mines de Paris, Ecole Polytechnique (France),
GKSS Research Centre (Germany), and the University of Stuttgart.
He has published over 100 journal papers in mechanics, materials science,
applied mathematics/physics and geophysics, as well as over 80 conference
proceedings papers and book chapters. He is on the editorial boards of Journal
of Thermal Stresses, Probabilistic Engineering Mechanics, ASME Journal of Applied
Mechanics, Actual Problems of Aviation and Aerospace Systems and International
Journal of Damage Mechanics. He is a Fellow of ASME and W.I.F., an Associate
Fellow of AIAA, and a Member of ISIMM. He has (co-)edited 10 books/journal
special issues and (co-)organized many symposia and conferences. He chairs
the PACAM Committee at the American Academy of Mechanics. He is an
avid sailor and alpine skier.
xxv
P1: Binod
July 13, 2007 21:36 C4174 C4174C000
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
1
Basic Random Media Models
Originating, as legend has it, with the Buffon needle problem

(which after two centuries has lost little of its elegance and ap-
peal), geometric probabilities have run into difficulties culmi-
nating in the paradoxes of Bertrand which have threatened the
fledgling field with banishment from the home of Mathematics.
M. Kac, 1976
In his foreword to Santalos treatise (1976), partly quoted as a motto to this

chapter, Mark Kac pointed out a need for a measure-theoretic formulation of
probability theory in general, and random geometry in particular. We shall
therefore initially devote some space to one of Bertrands paradoxes, and
thus motivate the formulation of random geometry as it was accomplished
in the first half of the twentieth century. This will set the stage for a review of
some basic notions of classical probability theory of engineering and applied
science curricula, as well as for an introduction to the simplest random geo-
metric models of disordered microstructures. The review is not complete
for instance, the conditional probability and description of microstructures
by joint probability distributions are not treated. The focus of this chapter
is on a review of basic concepts of random processes and fields for discrete
systems.
1.1 Probability Measure of Geometric Objects

1.1.1 Definitions of Probability
The classical definition of probability of an event Ahistorically, the oldestis
due to de Moivre (1718):
NA
P( A) = , (1.1)
N
where NA is the number of all outcomes favorable to A, and N is the number of
all possible outcomes. In a simple example of a toss of a fair coin, A = {head},
so that we have NA = 1 and N = 2, and, hence, P( A) = 1/2. However, the
1
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
2 Microstructural Randomness and Scaling in Mechanics of Materials
significance of NA, and N is not always clear, as another example shows: roll
two dice and consider P( A), where A = {sum of the numbers that equals 7}.
Then, basically, we have three ways to proceed:
(a) Consider eleven possible sums: 2, 3, . . . , 12. This leads to NA = 1 so

that P ( A) = 1/11, which, of course, is a wrong way of reasoning.
(b) Consider all possible outcomes of pairs of numbers not distinguish-
ing between the first and the second die: {1, 1}, {1, 2},. . ., {6, 6}. This
leads to NA = 3 and N = 21, so that P ( A) = 3/21, which is again
wrong.
(c) Count all the pairs distinguishing between the first and the second:

11 . . . 16

= . . . . . . . . ., (1.2)
61 ... 66
where by we denote the space of all elementary events. Now,

NA = 6 and N = 36, so that P( A) = 6/36, which is a correct answer.
This is the type of problem that motivated Laplace (1814) to come up

with an improved classical definition of probability: The probability of an event
equals the ratio of its favorable outcomes to the total number of outcomes,
provided that all outcomes are equally likely. There are several drawbacks
to this definition:
1. It contains a logical error: equally likely meaning equally possible

defines a concept in terms of the same concept.
2. It cannot deal with an unfair (loaded) die. The classical definitions
are based on the principle of insufficient reason, according to which,
in the absence of prior knowledge, we must assume that all the
elementary events have equal probabilities.
3. It cannot deal with an infinite set of possible outcomes, as is the
case, for example, with Bertrands paradox below.
One way to deal with the obstacle of an infinite number of outcomes was
offered by a relative frequency definition (von Mises, 1931):
nA
P( A) = lim , (1.3)
n n
where n A is the number of occurrences, and n is the number of trials. Let us
note, however, that the numbers n A and n might be large but not infinite in
any physical experiment. Therefore, this definition can only be accepted as a
hypothesis, and not as an experimentally obtainable number.
Bertrands paradox arises from the following problem: given a circle of
radius r , we wish to determine the probability p that the length l of a randomly
selected cord AB is longer than the side of an inscribed equilateral triangle.
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
Basic Random Media Model 3
(a) (b) (c)
FIGURE 1.1
Three different solutions of Bertrands problem.
There are, in fact, three possible answers:
1. The random radius method: Let be a fixed diameter of the circle,

having a uniform distribution along it. We draw cords normal to it
and find p = 1/2, Figure 1.1(a).
2. The random endpoints method: Let be an angle at one vertex
of the triangle, having a uniform distribution. Considering all the
cords through that vertex, we find p = (/3)/ = 1/3, Figure 1.1(b).
3. The random midpoint method: Let be a circle of radius r in-
scribed in the triangle and the cords center have a uniform distribu-
tion over the circle. Then the probability that the cords center lies in
any region of area A is A/r 2 . If the cords center falls outside a cir-
cle of radius r/2, then it is too small. Hence, p = (r 2 /4)/r 2 = 1/4,
Figure 1.1(c).
Clearly, each of the three solutions is correct in its own right as each one of
them solves a different problem. The problems differ in the specification of the
outcomes of an experiment and the meaning of the term possible (or favorable). The
first notion refers to the sample space of elementary events , and the second
one to a probability measure P (or, equivalently, a probability distribution)
defined on . The basic question to ask is: What is the space in this problem?
The correct framework in which to set up and P is offered by an axiomatic
definition of probability, with which we shall later briefly return to Bertrands
paradox.
The axiomatic definition (Kolmogorov, 1933) requires that, for a space of
elementary events , there is identified a so-called algebra F of subsets of
. The system {, F} is called a measurable space (Rudin, 1974). Then, on F
one can define a real-valued set function P, called a probability measure, which
satisfies three axioms:
Axiom I. For every A F, P is non-negative:
P( A) 0.
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
Axiom II. The probability of a certain event equals 1:

P() = 1.
Axiom III. For any sequence of mutually disjoint events A1 , A2 , . . . (Ai Aj

= for i, j = 1, 2, . . .; i =
j), there holds the equality

P(i=1 Ai ) = P( Ai ),
i=1
and we say that P is a countably additive function.

The sample space , together with F and P defined on it, is called a
probability space and denoted by {, F, P}. Here are the basic properties of
P (Problem 1):
1. The probability of an impossible event is zero: P( ) = 0. Note that
the inverse statement is not true, i.e., P( A) = 0 does not imply
A = .
2. P is an additive function:
n for any system of pairwise disjoint events

n
A1 , A2 , . . . , An , P Ai = P( Ai ).
i=1 i=1
3. Probability of an event Aopposite to Ais given as: P( A) = 1 P( A).
4. For any two events A and B, P( A B) = P( A) + P( B) P( A B).
5. If A B, then P( A) P( B).
6. For every A F, P( A) 1.

7. For any sequence of events of A1 A2 . . . such that A = Ai ,

i
P( A) = limn P( An ).
We now discuss basic cases of and the corresponding ways of setting
up of probability P.
1.1.2 Probabilities on Countable and Euclidean Sample Spaces

1.1.2.1 is a Countable Set: = 1 , 2 , . . .
We define P on all the elementary events {i } in the following way:
P({i }) = pi , (1.4)
where

pi 0 pi = 1. (1.5)
i=1
It follows from this and the system of axioms I to III that, if A and
A = {1 , 2 , . . .}, then
P ( A) = P ({1 , 2 , . . .}) = P ({1 } {2 } . . .)
= P ({1 }) + P ({2 }) + . . . = p1 + p2 + . . . . (1.6)
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
The function P defined by (1.6) satisfies axioms I and II because of (1.5), and
axiom III due to the fact that the sum of a series of non-negative numbers
does not change under arbitrary grouping or change in ordering of the terms
in the series.
A particular case of the countable space is that of a countably finite set
= {1 , 2 , . . . , N }, where
1
P ({1 }) = P ({2 }) = . . . P ({ N }) = , (1.7)
N
where N is the number of all possible elementary events, and n is the number
of the elementary events that favor A. In view of (1.7) and the three axioms,
if A and A = {1 , n , . . . , n }, then
P ( A) = P ({1 , 2 , . . . , 2 }) = P ({1 } {2 } . . . {n })
n
= P ({1 }) + P ({2 }) + P ({n }) = p1 + p2 + pn = (1.8)
N
In (1.8) we recognize Laplaces definition of probability, and its usefulness
when is countably finite and (1.7) holds.
Example 1: Each of m elements may be assigned to one of n different sets
(m < n). Compute the probability that no two elements get assigned to the
same set, assuming each of the assignments is equally probable.
The solution depends on first observing that there are nm different assign-
ments. Thus, each one is given a probability 1/mn . The sought probability
equals [m (m 1) (m 2) . . . (m n + 1)] /mn .
Example 2: A box contains seven white and three black balls. We take two
balls at random from the box, i.e., with (1.7) satisfied. Find the probability of
the event A = {both balls are black}.
The solution follows by first noting that has N = 10 = 45 elements,
2
so that each one has the probability 1/45. The event A has NA = 32 = 3
elements. Thus, P( A) = NA/N = 1/15.
This example is a special case of the so-called FermiDirac statistics. Clas-
sical statistical mechanics offers a number of combinatorial problems of this
type, and we collect three basic ones here for the sake of reference. Note the
difference between a permutation (P), which is an ordering of a set of objects,
with regard to order, and a combination (C), which is an ordering without
regard to order. Besides the aspect of ordering, there is also the aspect of
repetitions versus no repetitions.
1. N > M: Number of permutations of N objects taken M at a time

without repetitions (sequence important): N PM = N!/( N M)! The
special case of this is:
2. N = M: Number of permutations of N objects taken N at a time
without repetitions (sequence important): N PN = N!
3. N = M: Number of permutations of N objects taken N at a time with
repetitions (sequence important): N N . Generalizing this to N or
< M, we get N M .
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
The latter, which actually shows up in the MaxwellBoltzmann statistics,

is very relevant in the mechanics of microstructures: it is the so-called random
chessboard (or checkerboard). Consider a two-phase material in two dimensions
(2D): each square cell of M sites of a square lattice L L = M is occupied,
independently of what happens in other cells, with probability p1 and p2
by phases 1 and 2, respectively. Figure 1.2(a) shows one possible elementary
event , i.e., an assignment of two phases to the lattice. It follows that is
made of all such events. Clearly, || = 2 LL . Given the construction process,
(a)
(b)
FIGURE 1.2
(a) One event (or realization) of the random, two-phase chessboard on an L L lattice, with
L = 11. (b) A 16-phase mosaic.
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
each occurs with probability 1/2 LL . Figure 1.2(b) shows a realization of a

16-phase material taken out of an accordingly much larger space.
1.1.2.2 is a 1D Euclidean Space: = R

Two basic questions arise here: does there exist a function P satisfying the set
of axioms I to III, and if it does, how can it be constructed? We proceed by
introducing an auxiliary, real-valued function F on R having the following
properties:
1o F is nondecreasing.
2o limx F (x) = 0, limx F (x) = 1.
3o F is continuous from the left.
As these simple examples show, there exist many such functions:

1 1 0 for x 0,
F (x) = arctan x + , F (x) =
2 1 for x > 0;
x
0 for x 0,
F (x) = F (x) = F (x) = f (u)du, (1.9)
1 e x for x > 0;

where f 0 and f (u)du = 1.
For the function (1.9)4 , the condition 1o implies that, for a < b,
b
F (b) F (a ) = f (x) dx 0, (1.10)
a
that is, an integral over an arbitrary interval (a , b) is non-negative. Thus, a

function P can be constructed from F by taking, for each a < b,
P (a , b) = F (b) F (a ). (1.11)
This defines P (a probability measure) on the family of all sets of the form
[a , b). One can show that P can be extended in a unique way onto S, the
Borel -field on R, so as to satisfy axioms I to III (e.g., Prohorov and Rozanov,
1969). The function F is called a probability distribution, and the function f
is called a probability density. A classical result is that, if P is a probability
measure defined on R, then F (x) = P ((, x)) possesses properties 1o to 3o
(Problem 2). Basically, there are two possibilities: either to begin with a func-
tion P and construct F , or to begin with a function F and construct P.
Example 3: Consider this probability distribution:

0 for x a,

x a
F (x) = for a < x b, (1.12)

ba

1 for b < x.
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
This function can be shown to have properties 1o to 3o . Now, let a < c < d < b,
so that
d a ca d c
P (c, d) = F (d) F (c) = = , (1.13)
ba ba ba
which demonstrates that the probability of an interval [c, d) depends on its
length, but not on the location of the points c and d in [a , b). It is noteworthy
that the function (1.12) can be written as

0 for x a ,

1 x
F (x) = du for a < x b, (1.14)

ba a

1 for b < x.
In other words, for x [a , b), F is an integral of a constant, and such a

probability distribution is called uniform (or rectangular).
Note: The deterministic case fixed at one value x0 is modeled by a causal
distribution F (x) = H (x x0 ), with H being the Heaviside function. Dirac
delta plays the role of the corresponding causal probability density.
1.1.2.3 is a 2D Euclidean Space: = R2

Let F be defined on R2 and have the following properties:
1o F is nondecreasing with respect to each one of its arguments (while

the other is fixed).
2o x lim y F (x, y) = 0, y limx F (x, y) = 0,
limx, y F (x, y) = 1.
o
3 F is continuous from the left with respect to each argument (while
the other is fixed).
4 x1 x2 and y1 y2 , F (x2 , y2 ) F (x1 , y2 ) F (x2 , y1 )
o
+F (x1 , y1 ) 0.
Similarly as before, an important example of F involves (Problem 3)

x y
F (x, y) = f (u, v)dudv, (1.15)

where f is the probability density satisfying

f 0, f (x, y)dxdy = 1. (1.16)

Now, for every rectangle {(x, y) : x1 < x < x2 , y1 < y < y2 }, P is defined by
P({(x, y) : x1 < x < x2 , y1 < y < y2 }) = F (x2 , y2 ) F (x1 , y2 )

x2 y2
F (x2 , y1 ) + F (x1 , y1 ) = f (x, y)dxdy 0. (1.17)
x1 y1
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
Again, one can show that P can be extended in a unique way onto S, the
Borel -field on R2 , so as to satisfy all the axioms I to III. Furthermore, it is
also possible to prove that every P defined on R2 determines a probability
distribution
F (x, y) = P({(x , y ) : x < x, y < y}) (1.18)
having properties 1o to 4o . Although the proof of properties 1o to 3o is anal-

ogous to the one-dimensional case, it is instructive to demonstrate here the
property 4o . To this end, we introduce four sets (Figure 1.3):
J 1 = {(x, y) : x < x1 , y < y1 }

J 2 = {(x, y) : x1 x < x2 , y < y1 }
J 3 = {(x, y) : x < x1 , y1 y < y2 ,} (1.19)
J 4 = {(x, y) : x1 x < x2 , y1 y < y2 ,}.
The sets J 1 , J 2 , J 3 , and J 4 are disjoint, so that
F (x2 , y2 ) F (x1 , y2 ) F (x2 , y1 ) + F (x1 , y1 )

= P( J 1 J 2 J 3 J 4 ) P( J 1 J 2 ) P( J 1 J 3 ) + P( J 1 )
= P( J 1 ) + P( J 2 ) + P( J 3 ) + P( J 4 ) P( J 1 ) P( J 2 ) P( J 1 ) P( J 3 ) + P( J 1 )
= P( J 4 ) 0. (1.20)
Thus, there is a unique correspondence between the functions P and F with

a precision down to within sets of probability measure zero.
The above formulation may be extended to = Rn , with F having the
same properties as previously. However, the notation becomes quite cumber-
some and this is not written here explicitly. Note that the case Rn may be
treated as a special case of = Rn by taking P (Rn ) = 0.
y y
y2
b
J3 J4
y1
x1 x2 x
k
J2
J1
k b x
(a) (b)
FIGURE 1.3
(a) Domains J 1 , J 2 , J 3 , and J 4 . (b) Domain A of Example 5, bounded by two lines x y = k.
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
Let us consider one more case in R2 .

Example 4: F is given by (1.14) with
1
(ba )(dc)
for a x b, c y d,
f (x, y) = (1.21)
0 for all other x, y.
It is left an exercise to show that F has the properties 1o to 4o above. Now, let
a < a < b < b and c < c < d < d, so that

P( (x, y) : a x b , c y d )
= F (b , d ) F (b , c ) F (a , d ) + F (a , c )
b d b c
= f (x, y)d xd y f (x, y)d xd y

a d a c
f (x, y)dxdy + f (x, y)d xd y

b d
(b a )(d c )
= f (x, y)d xd y = . (1.22)
a c (b a )(d c)
This probability distribution is called uniform in a rectangle ((x, y) : a x b,
c y d).
It is interesting to note that the uniform distributions of this example as
well as that of (1.12) can also be introduced in another way: Consider to
be a subset of Rn , and let be a Lebesgue measure in that space, such that
0 < () < and A is a Borel set. Then we can introduce a probability
P in the following fashion:
( A)
P ( A) = . (1.23)
()
Clearly, the distributions encountered in Examples 3 and 4 are special cases of
(1.23). It is left as Problem 4 to check that P defined by (1.23) satisfies the set
of axioms I to III. F corresponding to it is called a uniform distribution, while
P itself is sometimes called a geometric probability.
Example 5: Let a = c = 0 and b = d in Example 4. Determine P ( A) for
A = {(x, y) : |x y| < k} with k (0, b).
The solution follows by first observing that = {(x, y) : 0 x b, 0
y b}. With () = b 2 and ( A) = b 2 (b k) 2 , P( A) is readily calculated
from (1.23). This example may be used to deal with the following application:
Given that two impulses arrive at random instants at a receiver over a period
of time T, compute the probability that the difference in their arrival times is
smaller than k < T. Another Example is given in Problem 15.
In problems of this type the term random connotes a uniform probability
density in a certain domain. In the case of arbitrary sets of a positive, bounded
Lebesgue measure, this implies that the probability distribution is constructed
from the formula (1.23).
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
1.1.3 Random Points, Lines, and Planes

1.1.3.1 Random Lines in Two Dimensions
Let us consider a set A of geometric objects in a 2D Euclidean space, and re-
quire that its properties remain invariant under the group of transformations
appropriate to such space, that is, all translations, rotations and reflections.
Now, all translations are obtained by transforming the coordinates by addition
of constants, while all rotations and reflections are effected by an orthogonal
transformation. Thus, the coordinates xk are mapped into new coordinates
xk by
xk = xk(0) + a ki xi , (1.24)
where [a ki ] is an orthogonal matrix, xk(0) is the translation, and the set A

is mapped into A . If we now denote by E the set of parameter points of
A, then E will be a new set corresponding to A . If we now introduce a
probability measure on the -algebra of all Es, then it is natural to impose
here a requirement of invariance of this measure under (a group of) all the
transformations (1.24).
For example, considering the points in the 2D Euclidean space, the set E
of these points must be contained in some region R of finite extent (i.e., of
bounded measure), like the inside of a cube or sphere. Then, by ( R) we
denote the Lebesgue measure of R, and by ( E) the Lebesgue measure of E.
It follows that the probability of a point lying in E is
( E)
P ( E) = . (1.25)
( R)
In the case of a random number of points, the simplest model is to consider

the number of these points in E to be a Poisson random variable with mean
( E) = M ( E), where is a constant; a formal definition of a random variable
is given in Chapter 2. If the number of such points is n, the probability that
they lie, respectively, in sets E 1 , E 2 , . . . , E n contained in E, given n, is
( E1) ( E2) ( En)

. (1.26)
[ ( E)]n
Another example concerns straight lines in a 2D Euclidean space. Let each

line be defined by an equation of the form
ux + vy + 1 = 0, (1.27)
so that u and v play the role of the lines coordinates, and the parameter space
of all possible events is the (u, v) plane, with exclusion of the point (0, 0);
this presents no special problem because the associated Lebesgue measure is
zero. The transformation (1.24) takes the form
X = a + x1 cos x2 sin Y = b + x1 sin + x2 cos , (1.28)

P1: Binod
July 11, 2007 20:47 C4174 C4174C001
with a and b being arbitrary constants and [0, 2 ). The new line is speci-
fied by
U X + VY + 1 = 0, (1.29)
with
U cos + V sin U sin + V cos
u= v= . (1.30)
aU + bV + 1 aU + bV + 1
Notice that aU + bV + 1 would be zero for lines passing through the origin,
and so, again, such lines are being excluded.
We now look for a measure M ( E) on the space , defined by an integral
of the form

M ( E) = F (u, v) dudv, (1.31)
E

such that it is equal to M E under all the transformations (1.18). That is, we
want

F (u, v) dudv = M ( E) = M E = F (U, V) dUd V. (1.32)
E E
Observe that from F (U, V) = F (u, v) J we obtain the condition where the
Jacobian is evaluated as
J = (aU + bV + 1) 3 = (u2 + v2 ) 3/2 (U 2 + V 2 ) 3/2 , (1.33)
which shows that F (U, V) is to be taken proportional to (u2 + v2 ) 3/2 , and

dudv
M ( E) = 3/2 . (1.34)
E u + v2
2
Note that (1) for this technique to work, we require that the Jacobian be of
the form J = (u, v) / (U, V) and (2) with the measure M ( E), we can now
establish the probability P ( E) of a single line, under the condition that its
coordinates are not (0, 0).
In the foregoing we began with a standard description of a straight line in
the form y = a x + b. However, we note that a line is also completely specified
by the distance p from the origin of the (x, y)-system and the angle that the
foot of the perpendicular to the line makes with the positive axis x, Figure 1.4.
The ranges of these two parameters are
0 p < , 0 2. (1.35)
In these polar coordinates ( p, ) one can then represent a line by a so-called

Hesse normal form:
x cos + y sin = p. (1.36)

P1: Binod
July 11, 2007 20:47 C4174 C4174C001
0 P D/2
(a)
(b)
FIGURE 1.4
(a) Two steps in generation of a Poisson line within a circle of diameter D. (b) A realization of a
Poisson line field of 100 lines in a unit square.
With reference to Figure 1.4(a), we see that lines parametrized by p and ,

and falling within a circle of diameter D, are represented as points in the strip
S = (0, D/2] [0, 2).
Under the ( p, ) parametrization, and the substitution
cos sin
u= v= (1.37)
p p
the differential element in the (u, v)-system transforms into one in the ( p, )-
system:
dudv
3/2 dpd. (1.38)
u2 + v2
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
Note that this formulation corresponds to the first solution of Bertrands

problem of Section 1.1. Poincare (1912) has shown that dpd is the only differ-
ential element that remains invariant under the group of all translations and
rotations. See also Jaynes (1973).
Note: In continuum physics this property results in the invariance of
material response under superposed rigid body motions with one observer
present, which is unnecessarily restrictive, while the correct path is to require
the invariance of material response under the change of observer (Murdoch,
2003).
One may continue in this fashion to determine probability measures for
straight lines uniformly distributed in 3D space, planes in 3D space, etc. (San-
talo, 1976). Indeed, to quote Kendall and Moran (1963): Once the probability
measure of a geometrical set is constituted, the solution of particular problems
can proceed without the appearance of paradoxes or of difficulties touching
on axiomatization.
1.1.3.2 Planes in Three Dimensions

A plane is represented by the equation
ux + vy + wz = 1, (1.39)
which, similar to the procedure given above, is better represented in terms

of the length of the perpendicular to the plane from the origin, together with
the polar coordinates of these perpendiculars. The corresponding equation is
x sin cos + y sin sin + z cos = p, (1.40)
where 0 < , 0 < , and 0 < p < .
1.1.3.3 Straight Lines in Three Dimensions

This involves a parametrization by four numbers (a , b, p, q ) because a line is
conveniently represented by two equations
x = a z + p, y = bz + q . (1.41)
1.2 Basic Point Fields

1.2.1 Bernoulli Trials
Let us begin with a few basic facts from the probability theory. A Bernoulli
trial is an experiment resulting in a success (event 1 ) or failure (event 2 )
with probability p or q , respectively. Here the sample space = {1 , 2 }
with P ({1 }) = p and P ({2 }) = q . A binomial distribution is a series of n
Bernoulli trials, all of them independent and identically distributed, in which
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
the probability of k successes is given by

n k nk
P{N ( A) = k} = p q k = 0, 1, 2, . . . , n. (1.42)
k
There are two important asymptotic theorems for this binomial distribu-
tion: one due to de Moivre and Laplace, and another due to Poisson. The first
of these states that if p 1 and n is so large that np npq 1, then

n k nk 1
e (knp) /2npq ,
2
p q k = 0, 1, 2, . . . , n, (1.43)
k 2npq
for k being around np. For x = k, the right-hand side in the above equals

x np 1
e (xnp) /npq ,
2
P (1.44)
npq 2
which is the Gaussian density function.

The Poisson theorem applies in the situation where np is of the order of 1,
with k being on the order of np, equal, say, to a ; it states that

n k nk ak
p q e a as n p 0. (1.45)
k k!
1.2.2 ExampleModel of a Fiber Structure of Paper

It is well known that paper is made of a multitude of cellulose fibers arranged
in a disordered fashion, Figure 1.5(a). We now make an assumption very close
to reality, namely, that each fiber lies horizontallyi.e., in a plane of the sheet
of paperand there are some n = 7 to 10 layers of fibers, as is the case in
a typical paper; the fibers are of a rectangular cross-section and uniform in
thickness, Figure 1.5(b). It follows that the process of sampling of voids and
fibers by a vertical line is a sequence of n (number of layers) Bernoulli trials,
with each trial being a success (finding a void) or failure (finding a fiber) with
respective probabilities:
P (void) = p, P (fiber) = q = 1 p. (1.46)
As an example, note that the event corresponding to the vertical line X in

Figure 1.5(b) is {1110101}. Thus, there are no bonds between the third, fifth,
and seventh layer from the top.
A quantity of fundamental interest in paper physics is the relative bonded
area (RBA), which is defined as the fraction of the total area of fibers, in plan
view projection, that is bonded (via hydrogen bonding) to other fibers. A
mat of fibers does not form paper for RBA = 0, and, clearly, elasticity and
strength of paper strongly depend on a positive-valued RBA. Soszynski (1995)
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
(a)
X
X = 1110101
(b)
FIGURE 1.5
(a) A perspective view of a sheet of paper; (b) section through an idealized sheet with n = 7, and
a sample outcome X ={1110101}. (From Soszynski, 1995, with permission.)
has shown that, for a sheet idealized as a system of layers of fibers, RBA is
given by
n1 2
RBA = q . (1.47)
n
With reference to Figure 1.5(b) we note:
(i) The distribution of fibers across the sheet is binomial and formula
(1.42) applies.
(ii) The probability of the occurrence of bond (fiberfiber) sites on n 1
interlayer planes, for an n-layer paper, is
k1 or n(k1)
k1 n (k 1) k i
Pn (k) = q k p nk
i=1
i 1 i n 1
k = 2, . . . , n,
(1.48)
where i is the so-called bond index; i = 1, . . . , k 1 or i = 1, . . . ,

n (k 1).
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
(iii) The probability of finding all bond sites on all the interlayer
surfaces is (Problem 7):

n
Pn (k) = q 2 (1.49)
k=2
which, noting that one bond belongs to two neighboring layers,

leads to (1.48).
1.2.3 Generalization to Many Types of Outcomes

Consider a partition [A1 , A2 , . . . , Am ] of , in a single trial, into m subsets such
that P ( Ai ) = pi and p1 + p2 + + pn = 1. Repeat the experiment n times
(i.e., the sample space becomes an n-fold Cartesian product n ) and denote by
pn (k1 , k2 , . . . , km ) the probability of the event {A1 occurs k1 times, A2 occurs
k2 times, . . . , Am occurs km times}, where k1 + k2 + + km = n. Then
n!
pn (k1 , k2 , . . . , km ) = p k 1 p k 2 . . . pm
km
. (1.50)
k1 !k2 ! . . . km ! 1 2
Next, suppose that each npi tends to some constant a i , as n and pi 0
for all i m 1. In this case
k
a 1k1 a 1 a 2k2 a 2 a m1
pn (k1 , k2 , . . . , km ) e e . . . m1 e a m1 as n . (1.51)
k1 ! k2 ! km1 !
1.2.4 Binomial and Multinomial Point Fields

1.2.4.1 Basics
Let us consider a rectangular window W in the R2 plane. Next, we generate
n points in this window under two conditions:
(i) The positions of n points (x1 , . . . , xn ) are stochastically independent

in the sense that
P (x1 B1 , . . . , xn Bn ) = P (x1 B1 ) . . . P (xn Bn ), (1.52)
where each Bi is an arbitrary Borel set in W. (In general, we can

consider W to be an arbitrary compact set.)
(ii) Each of the points (x1 , . . . , xn ) is uniformly distributed in W, i.e., for
any Borel set B W, we have
( B)
p B = P (xi B) = , (1.53)
(W)
where () is the area of a given set (i.e., its Lebesgue measure). We

observe two basic things from the above. First, the probability that x
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
lies in B is proportional to the area of B, and, second, the probability

that k points in a sequence of n Bernoulli trials lie in B is a binomial
distribution

n
P{N ( B) = k} = p kB (1 p B ) nk k = 0, 1, 2, . . . , n. (1.54)
k
Thus, the number of points in B, N ( B), is called a binomial point process

or a binomial point field. We prefer the latter terminology given the fact that
things happen spatially rather than temporally, while the word process usually
implies some time dependence.
From the consideration of the size of our window W, (W), and the num-
ber of assumed points, n, we immediately obtain the mean number of points
per unit area (or areal point density):
n
= , (1.55)
(W)
and the mean number of points per set B,
N ( B) = ( B). (1.56)
From this one can get the so-called emptiness probability
[ (W) ( B)]n
P{N ( B) = 0} = , (1.57)
(W) n
or the probability of finding no points in B.
1.2.4.2 Simulation of a Binomial Point Field with n Points

There are three basic methods.
Method 1: If W is a unit square [0, 1]2 , proceed by generating x = (x1 , x2 )
with both xi s being uniform random numbers in [0, 1], and repeating it n
times.
Method 2: If W is a square [0, a ]2 , proceed n times by generating x = (x1 , x2 )
with both xi s being uniform random numbers in [0, 1], and scaling each time
by a to get xi , i = 1, 2.
Method 3: If W is an arbitrarily shaped region, there exist three possibilities:
1. Place W in [0, a ]2 and proceed as in the process in method 2 above.

However, accept points falling into W only, and generate up to n
such pointssee Figure 1.6(a), with W being a unit square.
2. Divide W into a disjoint union of k small squares, which together
approximate W. Then, pick each square (i.e., decide whether there
is a random point in it) according to its size relative to A(W), and
generate its coordinates according to a uniform distribution over
that square.
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
(a)
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
(b)
FIGURE 1.6
Samples of (a) the binomial point process and (b) the sequential inhibition process, each with
100 points.
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
3. Providing W has adequate symmetry, apply a transformation of

coordinates. For example, for a circle of radius R, a random point is
described by x = ( p, ). In order to have a probability distribution
F R (r ) = p{r R} proportional to the area (i.e., to r
2
) we need to
use a uniform number u from [0, R], to generate = u. The angle
is uniform in [0, ].
1.2.4.3 Generalization to a Multinomial Point Field

The binomial point field can be generalized to a multinomial one by consi-
dering a partition of the window W into a union of m pairwise-disjoint subsets
B1 , B2 , . . . , Bm with
( Bi )
p Bi = P{ ( Bi )} = . (1.58)
(W)
It follows that the probability of finding k1 points in B1 , k2 points in B2 , . . . ,
and km points in Bm , in a sequence of n trials, is
Pn {N ( B1 ) = k1 , N ( B2 ) = k2 , . . . , N ( Bm ) = km }
n!
= p k1 p k2 . . . p kBmm . (1.59)
k1 !k2 ! + km ! B1 B2
This field may be used to model a composite with m different types of
inclusions.
1.2.5 Bernoulli Lattice Process

Consider a Cartesian lattice of spacing a in R2 , that is
L a = {x = (m1 a , m2 a )}, (1.60)
where m1 and m2 are integers. A Bernoulli lattice process p,a on L a is a

random subset of the lattice where each point of L a is contained in p,a with
probability p independently of all the other points. If the random variable
p,a ( B) is the number of points in B, then it is binomially distributed with
parameters p and n (the number of lattice points that belong to B). Also,
p,a ( B1 ), p,a ( B2 ), . . . , p,a ( Bk ) are independent if B1 , B2 , . . . , Bk are pairwise
disjoint, and we have

n
P{ p,a ( B) = k} = p k (1 p) nk k = 0, 1, 2, . . . , n, (1.61)
k
Simulation of a Bernoulli lattice process: For each point x L a , generate a

random variable zk , uniformly distributed in [0, 1], and accept this point if
zx < p. Then, p,a is the union of all such points.
The asymptotic limit of a Bernoulli lattice process: It is easy to see that the
mean of p,a is

p,a = np, (1.62)
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
where, as before, n is the number of points that are in B. This leads to a concept
of intensity
= p/a 2 . (1.63)
If we consider B to be a unit square divided into m m spacings with each

spacing being of the size a = 1/m, we get = p/a 2 = pm2 . We now let
m2 and p 0 while we keep constant, so that, analogously to
Section 1.2.1, the random variable p,a ( B) has a distribution that is asymp-
totically Poisson of mean ( B)
[ ( B)]k ( B)
P{ p,a ( B) = k} e as m2 p 0. (1.64)
k!
1.2.6 Poisson Point Field

1.2.6.1 Basics
The homogeneous planar Poisson field (or process), , is the cornerstone of the
theory of spatial point fields (or processes). It represents the simplest possible
stochastic mechanism for the generation of spatial random point patterns.
This process is defined by two postulates:
P1: For any > 0, and any finite planar region A, N ( A) is a Poisson
distribution with mean |A|.
P2: Given N ( A) = n, the n events in A form an independent random
sample from the uniform distribution on A.
Note: P1 implies that the intensity does not vary over the plane. P2
implies that there are no interactions among the events, that is, no inhibition
or encouragement of events in the neighborhood of x given an event at x. The
third important property is given in Problem 9.
The intensity (x) of a spatial point process is defined by
N (dx)
(x) = lim , (1.65)
|dx|0 |dx|
where denotes the average. Note that the parameter of P1 is the intensity
of the process, independent of x. The emptiness probability becomes
P{N ( B) = 0} = e |B| . (1.66)
Furthermore, in analogy to the formula (1.59) for the multinomial field
Pn {N ( B1 ) = k1 , N ( B2 ) = k2 , . . . , N ( Bm ) = km }
[ |B1 |]k1 |B1 | [ |B2 |]k2 |B2 | [ |Bm |]km |Bm |
= e e ... e . (1.67)
k1 ! k2 ! km !
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
Variance of the Poisson point field follows from the variance of the Poisson
random variable
Var {N ( B)} = |B|. (1.68)
It follows from P1 that, if B1 and B2 are disjoint,
N ( B1 ) N ( B1 ) = |B1 | |B1 |. (1.69)
It then follows from Problem 10 that, for any two sets B1 and B2 , such that
B1 B2 = D,
N ( B1 ) N ( B2 ) = |B1 | |B2 | |D|. (1.70)
Now, consider two disjoint, infinitesimal disks B1 and B2 , at a distance r ,

and of areas d F1 and d F2 , respectively. Then, because the disks are very small,
N ( B1 ) N ( B1 ) P ( N ( B1 ) = 1, N ( B2 ) = 1). (1.71)
On the other hand,
PN ( B1 ) = 1, N ( B2 ) = 1 = d F1 d F2 = (2) (r ) d F1 d F2 , (1.72)
where we have introduced the second-order intensity function (2) (r ) to charac-

terize arbitrary (not necessarily Poisson) point fields. Of course, in the Poisson
field case, (2) (r ) = 2 , and this is the ideal reference case. On this basis, one
then defines a pair correlation function for point fields
g (r ) (2) (r ) /2 , (1.73)
Further concepts, such as a pair correlation function and radial distribution

function and a k-function are used to deal with point fields (Matheron,
1975; Stoyan et al., 1995).
1.2.6.2 Simulation of a Poisson Point Field

If a Poisson point process is stationary or homogeneous (i.e., invariant with
respect to arbitrary shifts in space), one can consider its restriction to a compact
set W of the plane under the condition that (W) = n. This conditioning
yields a new point process that is actually the binomial point process in W
with n points, simulated according to Section 1.2.5.
1.2.6.3 Inhomogeneous Poisson Point Field

This process is defined by two postulates:

P1: N ( A) has a Poisson distribution with mean A (x) dx.
P2: Given N ( A) = n, the n events in A form an independent random
sample from the distribution on A with the probability density
function proportional to (x).
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
This type of process is very useful for simulating functionally graded

materials; see Problem 8.
As for a homogeneous Poisson point field, E ( N ( B)) = |B|, while N ( B)
= 1 = (x) |B|, if B is an infinitesimal disk of area |B| = d F.
1.2.6.4 Inhibition and Hard-Core Processes

The so-called car parking problem is a point of reference for more complex,
analogous problems in higher dimensions. It involves a sequential coverage
of an interval (0, x) with x > 1 by intervals of unit length without overlap,
until no additional unit length interval can be accommodated. Let N denote
the number of all such intervals placed in (0, x). It was shown by Renyi (1959)
that the ratio of the average of N to x tends to 0.74759 . . . as x , whereas
the ratio of the variance of the occupied portion of the interval to its length is
0.03567.
Various extensions of this problem have been considered in the probability
and statistical physics literature. They involve parking of intervals (or cars)
of random length in one and more dimensions. Of primary interest to us is
the latter case, which is also called a random space-filling problem, and we list
several references. The case of 2D was considered by Palasti (1960) and Gani
(1972). The case of 3D was studied in the setting of liquids by Bernal (1960),
Bernal et al. (1962), and Scott (1962).
Typical inhibition (or hard-core) processes are:
1. Thinning: A Poisson point field of a given intensity is thinned by

the deletion of all pairs of events a distance less than apart.
2. Sequential inhibition process: Throw Poisson points on a plane and
keep only those which fall no closer than dmin to any previous ones.
Figure 1.6(b) shows a realization of this process with 100 points at
dmin = 0.0785, together with disks centered at these points. The
volume fraction of the disks is 0.4, their radius being 0.0357. Note
that the classical car-parking problem is a simple inhibition in 1D.
3. Modified sequential inhibition process: take dmin to be a random variable
for each Poisson point.
All these processes lead to a development of regularity of point patterns.
1.3 Directional Data

1.3.1 Basic Concepts
Many materials involve features whose angular directionssuch as fibers
or cracks in a matrix, crystal orientationsare of interest. Because these are
often of a random character, a basic calculus of angular random variables
is needed. First of all, the mean requires a definition modified from that of
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
conventional linear variables. Thus, suppose we have N measurements of the

angle : {n ; n = 1, . . . , N}. Then the mean angle is determined as a solution
of the equation
S
tan = , (1.74)
C
where
1 1
N N
S= sin n , C= cos n . (1.75)
N n=1 N n=1
Clearly, S and C play the role of center of gravity of all the data in a polar
diagram of coordinates {(cos n , sin n ); n = 1, . . . , N}, the distance from the
2 2
origin being R = S + C . We have

N
sin(n ) = 0, (1.76)
n=1
which shows that the deviation about the mean vanishes; this is analogous to
the linear case:

N
(xn x) = 0. (1.77)
n=1
In fact, a shift of the reference zero direction by some angle in the former
case results in
= ( )mod 2, (1.78)
while a shift of origin by a in the latter case gives
x = x a . (1.79)
These equations show the distinct property of angular measurements as op-

posed to the linear ones, and this is what motivates the modified concepts
below. A full presentation is given in Mardia (1972).
The distribution function of taking values in (0, 2 ] is defined by F () =
P{0 < t }, but this definition is extended to the entire real line, as follows:
F ( + 2) F () = 1, < < . (1.80)
If F is absolutely continuous, it has a probability density function f such that

f ()d = F () F (), < < < , (1.81)

P1: Binod
July 11, 2007 20:47 C4174 C4174C001
and f has these properties:
1o f () 0, < < ,
2o f ( + 2 ) = f (), < < ,
2
3o 0 f ()d = 1.
The Fourier series of f is defined by

1 2
f () = p e i p , p = e i p f ()d, (1.82)
2 p= 0
which, in applications, is often written as

1
f () = 1+2 ( p cos p + p sin p) , (1.83)
2 p=1
so that the distribution function is

1
F () = +2 {( p sin p + p (1 cos p)}/ p . (1.84)
2 p=1
In view of the considerations at the outset of this section, the mean of

a continuous random variable is defined by
sin( ) = 0, (1.85)
and the circular variance is defined as
V0 = 1 cos( ) . (1.86)
It takes values in [0, 1], unlike [0, ] in the linear case.
1.3.2 Circular Models

In this section we collect some most often occurring models.
(i) Point distribution: P{ = 0 } = 1; this may be thought of as an

analog of a causal distribution in the linear case. Its Fourier expan-
sion for 0 = 0 is

1
f () = 1+2 cos k . (1.87)
2 k=1
Here V0 = 0, and, in fact, it can be shown that if V0 = 0, then one

deals with a point distribution.
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
(ii) Lattice distribution:

pr = P{ = 2r/m}, r = 0, 1, . . . , m 1, pr 0, pr = 1.
r
(1.88)
When all pr = 1/m, we get a discrete uniform distribution on the

circle.
(iii) Uniform distribution:
1
f () = , 0 < 2. (1.89)
2
This and the next cases pertain to a continuous random variable .

(iv) Cardioid distribution:
1
f () = [1 + 2 cos( 0 )], 0 < 2, || < 1/2. (1.90)
2
(v) von Mises distribution:
1
f (; 0 , ) = e cos(0 ) , 0 < 2, 0 0 < 2, > 0
2 I0 ()
(1.91)
where I0 () is the modified Bessel function of the first kind and
order zero. 0 is the mean direction, while is the concentration
parameter; the higher is, the stronger is the concentration of
about 0 . See Problem 9.
(vi) Wrapped distribution: it arises when we wrap a certain distribution
F (x) of a linear random variable x on the circumference of a circle
of unit radius. The resulting angular variable xw is
xw = x mod 2, (1.92)
having the distribution function

Fw () = [F ( + 2 k) F (2 k)], 0 < 2, (1.93)
k=
and, if x possess a density function, then the density function of xw is

f w () = f ( + 2 k). (1.94)
k=
Two important examples of wrapped distributions are given below.

P1: Binod
July 11, 2007 20:47 C4174 C4174C001
(vii) Wrapped Gaussian (normal) distribution:

! "
1 ( + 2 k) 2
f w () = exp , 0 < 2, (1.95)
2 k= 2 2
which follows from the preceding equation by wrapping x from

N(0, ).
(viii) Wrapped Cauchy distribution:
1 1 2
f w () = , 0 < 2, 0 < 1, (1.96)
2 1 + 2 2 cos
which is seen to reduce to the uniform one for 0, and to the
one fully concentrated at = 0 for 1. Note that V0 , in contrast
to the linear case, is finite: it equals 1 .
1.4 Random Fibers, Random Line Fields, Tessellations

1.4.1 Poisson Random Lines in Plane
It was shown in Section 1.1.1 that straight lines are best represented in the
( p, )-system, Figure 1.4(a). It follows that under the ( p, ) parametrization,
the set G of all lines g in the (x, y)-plane is equivalent to the semiinfinite strip
S of (1.35). Thus, each point of this strip corresponds to a line, and so, each
subset of S to a subset of G. Now, let G b( O,r ) be the set of all lines that intersect
the disk b ( O, r ), centered at O and of radius r . Clearly, the corresponding
subset of S is
Sb( O,r ) = {( p, ) : 0 < p r, 0 < 2 } . (1.97)
Now, in order to simulate a line from G b( O,r ) , we use two independent, uniform
random variates u and v from [0, 1] and generate
p = r u, = 2 v. (1.98)
Next, consider a point field in R2 . Those points that lie in S form a sub-
point field. Each of these points lies in the (x, y)-plane, and the set of these
lines is called a line field. Now, consider a Poisson point field of intensity in
R2 . The points lying in S correspond to a random set of lines, called a Poisson
line field, , with parameter = 2. An example is shown in Figure 1.4(b).
Basic properties of the Poisson line field are (Miles, 1964):
1o The intersection points with a fixed line form a linear Poisson point
process with intensity = 2. The intersecting angles are inde-
pendent random variables and have a probability density (sin ) /2,
[0, 2).
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
2o The number of lines that intersect a plane set K has a Poisson dis-
tribution with parameter (SK ).
3o The mean total length L A of a line segment of the field in a region of
area 1 is , which is known in random geometry as a line density.
4o The probability that more than two lines intersect at a point is
exactly 0.
5 The Poisson line field leads to a so-called Poisson mosaic of Poisson
o
polygons, the average number of edges in a polygon being four.

With A denoting the area of a polygon, its first, second, and third
moments are
2 4 7
A = A2 = 4 A3 = . (1.99)
2 2 76
6o = 2 is the mean number of lines intersected by a test line segment
of unit length.
o
7 The probability density of diameters D of circles inscribed into
Poisson polygons follows a negative exponential:
f ( D) = Ce D , (1.100)

where C is determined from the normalization condition 0 f ( D)
d D = 1, while is the coverage parameter (= number of lines per
diameter, 2r , of the Poisson line field).
Note: Upon letting the Poisson polygons of the Poisson mosaic be occu-
pied at random by either one of two phases (black or white), one can generate
a two-phase composite, Figure 1.7(a).
The property 7o is relevant in paper technologies, where one studies the
retention of spherical particles by sieving as they flow across a planar mat of
theoretically infinite, straight fibers, having the geometry of the Poisson line
field. Making a correction for the finite width of fibers, d f , amounts to shifting
of the density (1.100) to the left by d f . However, trying to determine the
networks ability to retain the particles leads one to consider the relative area
of the polygon-inscribed circles, rather than the number of these circles alone.
This leads to an area-based probability density having a Gamma function
form
3 D2 D
f A ( D) = e , (1.101)
2
from which we obtain the probability of retention

D 2 D2
P {retention} = F A ( D) = f D d D = 1 1 + D + e D .
0 2
(1.102)
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
(a)
(b)
FIGURE 1.7
(a) A two-phase mosaic generated from a Poisson line field with 100 lines. (b) A finite fiber
field.
In effect, this may be replaced by F A in the function of a new dimensionless

parameter = D, so that

2
P {retention} = F A () = 1 1 + v + e . (1.103)
2
Note: One may consider line fields generated by non-Poissonian point
fields in S. A particular case, again motivated by machine-made papers which
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
display preferential orientation of fibers, is a generalization of the Poisson

line field to an anisotropic model. Here one takes as a non-uniform random
variable given by the Fourier series model (1.83) with all p = 0. Interestingly,
whereas the formula (1.100) of property 7o was long known to hold for an
anisotropic system (Miles, 1964), the area-based formula (1.103) carries over
to that case as well (Castro and Ostoja-Starzewski, 2000).
Note: Dodson and Sampson (1996) derived (1.101) differently.
1.4.2 Finite Fiber Field in Plane

We now consider random fields of finite length fibers (i.e., segments) in plane,
Figure 1.7(b). The simulation proceeds as follows:
1. Generate n Poisson points in a square [0, L]2 .
2. Generate an angle with respect to the x1 -axis, for each Poisson
point, according to a specific probability distribution, again allow-
ing for anisotropic patterns.
3. Generate a random fiber of length oriented at angle , for each
Poisson point, according to a specific probability distribution P (l).
The centers of the fibers are fixed at their respective Poisson points.
Note: There is no continuous passage from this process to the line field.
In other words, by extending the fiber lengths l to infinity, one cannot obtain
the random (Poisson) line field.
Note: One can generate fibers in clusters, see the section on germ-grain
models below.
1.4.3 Random Tessellations

1.4.3.1 Basic Concepts
The Poisson line field above has been our first example of a tessellation. Basi-
cally, a system of polygons in a plane is said to be a tessellation if the constituent
polygons are pairwise disjoint and the union of their closures fills the plane.
Formally, a polygon is a bounded, convex, open, nonempty set in R2 . The poly-
gons are called the cells of the tessellation, the vertices are called the nodes of
a tessellation, and the sides of polygons are called the edges of a tessellation.
These concepts may be generalized to 3D, and one can speak of a spatial
tessellation as a division of R3 into pairwise disjoint polyhedra. A polyhedron
involves a set of vertices, V, a set of edges, E, and a set of faces (i.e. cells), F .
Here we recall the Euler formula for finite polyhedra in R3 :
|V| |E| + |F | = 2, (1.104)
where || stands for the number of elements in a given set. Now, a planar tes-
sellation can be interpreted as the surface of an infinite polyhedron, whereby
(1.104) is used to show that, in the limit of an infinitely large polyhedron,
E = V + F, (1.105)
where V stands for the intensity of V and so on.
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
1.4.3.2 Planar (Poisson-)Voronoi Tessellations and Delaunay Triangulations

Take a realization of points x j ; j = 1, 2, . . . from a Poisson point field
in R2 . Typically one considers in a unit square, so that the set of Poisson
points is finite. Then, define a cell of the tessellation to be a domain D j of all
the points y in R2 closer to a given Poisson point x j than to all other Poisson
points xk ( j =
k):
# # # #
D j = y R 2 : # y = x j # # y = xk # , j =
k . (1.106)
Almost all y in R2 have a unique nearest Poisson point, that is, except the
edges of cells that are equidistant to the two neighboring Poisson points. The
domains Di are called Voronoi (or Dirichlet) cells, and the set of all D j s forms
a PoissonVoronoi tessellation, alternatively called a PoissonDirichlet mosaic.
Another way of viewing this tessellation is to look at the growth of (cir-
cular) disks at a uniform rate, all starting at the same time, until they meet,
which happens along the bisector of a line joining two neighboring Poisson
points. These latter lines form a Delaunay triangulation. There is a one-to-one
relation between the edge set of the Voronoi tessellations and the edge set
of the Delaunay triangulation, and another one-to-one relation between the
vertex set of Voronoi and the triangles of Delaunay. The average number of
Delaunay edges, whether the network is based on a Poisson point field or not,
incident onto a vertex is six (Problem 11).
Two Voronoi tessellations and their Delaunay triangulations correspond-
ing, respectively, to the point patterns of Figures. 1.6(a) and 1.6(b) are shown in
Figure 1.8. Of course, the tessellation of Figure 1.8(b) is not PoissonVoronoi.
Note: While the Voronoi tessellation (and its modifications) is a very pop-
ular model of polycrystals, the Delaunay triangulation is useful in mechanics
of granular media (Goddard, 2001).
Note: A Voronoi tessellation of a 3D space may be carried out similarly to
the 2D construct here, and these two models are very popular in mechanics
and physics of random media. However, one needs to remember here that
a planar intersection of a spatial PoissonVoronoi tessellation is not a planar
PoissonVoronoi tessellation. It is an open question whether there exists a
planar point process whose Voronoi tessellation is identical in distribution to
this intersection.
1.4.3.3 Modifications of Voronoi Tessellations

There are various ways in which the Voronoi tessellation may be modified,
e.g.:
1. Start growing the disks from the Poisson point process as soon as
these points fall onto the plane (i.e., sequentially), so as to get a
JohnsonMehl model.
2. Introduce random rates of growth for each disk.
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
(a)
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
(b)
FIGURE 1.8
(a) Voronoi tessellation and its Delaunay triangulation, generated from the 100 points shown in
Figure 1.6(a); Voronoi tessellation, Delaunay triangulation, and disks centered at the points of
Figure 1.6(b).
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
3. Grow the disks by employing a general metric

n 1/ p

d p y, x = |yi xi | p 1 p < , (1.107)
i
as opposed to the conventional Euclidean d2 metric involved in the

Voronoi tessellation described above (Okabe et al., 1992). Figure 1.9
shows such two networks, one for the d1 (Manhattan) and another
for the d (sup) metric, both grown from the Poissonian point pat-
tern of Figure 1.6(a) in 2D.
(a)
(b)
FIGURE 1.9
Voronoi tessellations, obtained from the point pattern shown in Figure 1.6(a) in the d1 (a) and d
(b) metric, respectively.
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
1.4.3.4 Random Crack Model

The so-called random crack model is a purely geometric construction designed
to simulate some crack patterns encountered in real systems. For one ,
it is described by the following sequence of steps:
1. Drop N points at random in R2 .
2. Assign an angle sampled from some angular distribution, such as
the uniform one (1.89).
3. From each and every point, starting at time t = 0, grow a line
segment in both directions at a uniform ratesuch as shown in
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 0.2 0.4 0.6 0.8 1
(a)
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 0.2 0.4 0.6 0.8 1
(a)
FIGURE 1.10
The intermediate (a) and the final (b) stages in the generation of the random crack model.
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
Figure 1.10(a)until it meets another line or the window boundary,

and then stop.
4. As a result, obtain a tessellation of the plane such as shown in Figure
1.10(b).
Note: This model may be modified to periodic boundary conditions.

Note: Many models presented here can be generalized to 3D. For instance,
by taking random planes according to a parametrization mentioned at the end
of Section 1.1, we generate a spatial tessellation made up of (convex) Poisson
polyhedra.
1.5 Basic Concepts and Definitions of Random Microstructures

1.5.1 General
It should be clear by now that various probabilistic models introduced in
previous sections may be used to represent real materials. They are particular
examples of random media. Now, a random medium, denoted by B, is defined as
B = {B () ; }, (1.108)
where each B () is a particular, spatially heterogeneous realization, follow-

ing laws of deterministic mechanics; recall this books Preface. Equivalently,
we also use the terms random microstructure, random material, and random com-
posite. Basically, these are media whose properties vary randomly from point
to point, and, therefore, their evolution as an ensemble is stochastic. In the
preceding sections we have introduced several random media models, and
in the following we discuss several more types.
Before we proceed, we state that the media we are interested in are not
fractal. That is, their mass (the basic physical property) scales with volume
M ( R) = k R D (1.109)
where M is the mass of fractal medium, R is a box size (or a sphere radius),
and D is the spatial dimension of a given problem (i.e., either 1, 2, or 3).
The scaling (1.109) is understood on average: due to spatial randomness of
the material, there are fluctuations of mass, but they vanish upon ensemble
averaging. We relax the assumption (1.109) in Section 10.6 of Chapter 10.
1.5.1.1 Germ-Grain and Boolean Models

(Flocs of fibers as grains). A germ-grain model is made of grains planted at germs,
whose centers are given by some point field in some Euclidean space. A
Boolean model is a special case of the germ-grain model in the sense that
is the spatially homogeneous Poisson point field. The model is introduced
constructively: we start with = {1 , 2 , . . .}, the points i being germs.
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
Next, on each i we place a grain i , such that 1 , 2 , . . . form a sequence of

independent, identically distributed random compact sets in the x, y, z-space,
that are independent of the process. The Boolean model is the union of
the grains i , translated by the i in
= i (i + i ). (1.110)
As an example, we now develop a germ-grain model of a fibrous struc-

ture of paper, constructed so as to obtain fiber flocculation, commonly seen as
nonuniform grayscale effects in a typical sheet of paper held against light.
This model is based on a hard-core point process = {1 , 2 , . . .}, where
the grains i are flocs of fibers, whereby each fibers center is at a position
r = (r x , r y ) relative to the flocs center i . We focus on realizations of in
windows of size L x L y t in the x, y, z coordinate system. L x , L y , and t
are the two in-plane dimensions and the (much smaller) z-thickness of paper,
respectively. Upon defining a dimensionless mesoscale parameter
= L/ l , (1.111)
where L = L x = L y and is the average fiber length, we focus on a finite-

size random medium B = {B (); }. By using a random variable r0
governed by the one-parameter triangular probability density function
! "
b2 2
p(r0 ) = r0 + b r0 0, , (1.112)
2 b
with b being a floc parameter, we generate coordinates for the location of a

fiber center relative to the floc center:
r x = (1 + a 1 )r0 cos r y = r0 sin . (1.113)
Here is the fibers in-plane orientation angle controlled via (1.90) with 0 = 0
and a 1 being a free parameter.
The model (1.1121.113) is chosen so as to obtain clustering in a finite disk
of radius 2/b if a 1 = 0, or in an ellipse if a 1 > 0. In the latter case, the flocs are
stretched in the machine direction (x) according to the degree of a preferred
orientation of fibers so as to reflect a typical structure of machine-made paper.
Note that as b increases, fibers are clustered into tight flocs of radius tending
to zero as b , and, as b decreases, they are scattered. This is apparent
in (a) and (b) of Figure 1.11 (left column). (We return to the middle and right
columns in Section 4.3 of Chapter 4.)
We define the density d of the network as the total fiber length per unit in-
plane area. The coverage (average number of fibers per point) and sheet basis
weight (weight per unit area) are obviously directly proportional to d. For a
chosen d, a number of fibers to be assigned to the test volume is computed,
and then each fiber is assigned at random, uniformly, to any of the germs.
The fiber centers within each floc are not generated in a common z-plane; the
z-coordinate is sampled from a uniform depth density.
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
By keeping d and t independent, we can simulate papers with the same

coverage but different degrees of compaction, corresponding to different
degrees of pressing during papermaking. The degree of compaction is mea-
sured by the relative bonded area RBA defined as
Abonds
RBA = , (1.114)
Aprojected
where Abonds is the total area of all bond parallelograms, and Aprojected is the
total projected area of all fibers (Deng and Dodson, 1994). This methodology
reflects our understanding that the fiber network model is likely to be realistic
only for papers that have relatively low RBA, so their primary load-carrying
mechanism is the transfer of forces and moments along fiber axes between
nodes, rather than a more complex interaction of plate- or solid-like fiber
segments. Mechanics of thus constructed fiber networks is studied in various
sections of Chapters 4, 7, and 9.
1.5.1.2 Flocs as Continua

If one looks at Figure 1.11(a) from afar, it tends to look like a random contin-
uum of Figure 1.1(b) in the Preface. Thus, motivated by the Boolean model
of fiber network introduced above, and aiming at the effects of floc-type
(a) Floc parameter b = 2.0
(b) Floc parameter b = 0.4
FIGURE 1.11
Highly (top) and weakly (bottom) flocculated networks in undeformed state (left column) and
deformed state amplified for demonstration (center column). Figures of the right column show
differences between the true node displacements and those of the uniform strain assumption.
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
= 10 = 10
1 Layer 13 Layers
FIGURE 1.12
Generation of the Boolean model of paper formation through random placement of disks, with
coverage at a point ranging from 1 up to 13 discs (layers). Also, zooming in with a mesoscale
from = 10 to = 4 is shown.
formation, we now introduce a Boolean model of a random quasi-continuous

paper material. We again focus on realizations B() in 3D windows L x L y t.
In this model each grain i is an elliptical disk of unit thickness t, represent-
ing a floc of fibers. This, of course, is a simplification from the preceding
section where we dealt with fibers scattered randomly around the grains
center.
After generating a number of disks in the window, we have n disks (layers),
each of the same thickness t, stacked at each x,y-location of the plane. Thus, we
obtain a sheet of piece-wise variable thickness, nt, per unit area, which gives
us a distribution of basis weight. Figure 1.12 illustrates an realization of this
simulation process at = 10, and a zooming in with = 4. If the disks were
ellipses and their major axes were to coincide with the x-axis, there would be a
biased orientation in the MD direction, just like in a machine-made paper with
plane orthotropy. With circular disks the simulation represents a laboratory
handsheet with an in-plane isotropy.
A somewhat different paper formation can be obtained by taking as a
hard-core process, that is, one in which the minimum distance between any
two i s is non-zero. This prevents flocs from being arbitrarily close to each
other (i.e., near overlap) as in being a Poisson point process, and then
turns from a Boolean into a germ-grain model.
1.5.2 Toward Mathematical Morphology

The germ-grain models are the cornerstone of the mathematical morphology.
Four examples of microstructures generated by that field of mathematics are
given in Figure 1.13; see Jeulin (1997). In the following we only introduce a few
basic concepts. The covariogram K ( X, h) is a measure (Mes) of the intersection
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
of the set X (such as a surface in 2D, or a volume in 3D) with the set X obtained
by translation of X by h: Xh . Thus,

K ( X, h) = Mes ( X Xh ) = k (x) k (x + h) d x, (1.115)
(a)
(b)
FIGURE 1.13
(a) Sequential alternate random function (modeling mica, or pack ice, with gray scale); (b) Boolean
random function (modeling biological tissue); (c) dead leaves random tessellation of Poisson
polygons (e.g., modeling layered material); (d) Boolean model of Poisson polygons (modeling
tungsten-carbide [black] and cobalt [white]).
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
(c)
(d)
FIGURE 1.13
(Continued).
where

1 if x X,
k (x) = (1.116)
0 if x/ X.
The covariogram has the following properties:
1o For h = 0:
K ( X, 0) = Mes ( X) . (1.117)
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
2o For a bounded set X:
K ( X, ) = 0 K ( X, h) = 0 for h > A, (1.118)
where A is the largest distance between two points in the direction

of h.
3o The integral of the covariogram is

K ( X, h) dh = [Mes ( X)]2 . (1.119)
Rn
Note: The probabilistic version of the covariogram for a stationary set X
is the covariance function C ( X, h):
C ( X, h) = P (x X, x + h X) , (1.120)
which has the property
C ( X, h) = C ( X, h) C ( X, 0) . (1.121)
Whether the set X is in 2D or in 3D, we have
C ( X, 0) = V ( X), (1.122)
which is the areal or volume fraction of X in R2 or R3 , respectively. It follows

that
lim C ( X, h) = [V ( X)]2 , (1.123)

h
which says that the covariance asymptotes to the volume fraction squared.
The function C ( X, h) is indicative of the connectivity of the set X. If the limit
above is reached at some finite h, say, h c (called range of the covariance), the
points of the structure beyond h c are uncorrelated.
The covariance and its range can be estimated from images by means of the
covariogram. This is shown here in terms of the examples of microstructures
from the food industry, due to Kanit et al. (2003). In the case of Figure 1.14(a),
the range is about 37, and in the case of 1.14(b) it is 19. Also note that, with
the horizontal and vertical covariances being very close, both microstructures
are isotropic.
Following the presentation of Kanit et al. (2003), we introduce the notion
of an integral range, which specifies how well the parameters obtained for a
domain of finite size statistically represent the random microstructure as a
whole. That range is defined (Matheron, 1975) as

1
An = [C( X, h) C( X, 0) 2 ]dh. (1.124)
C( X, 0) C( X, 0) 2 Rn
If this is applied to the volume fraction
Mes( X V)
VV = , (1.125)
Mes(V)
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
0.45
Horizontal covariance
0.4 Vertical covariance
0.35 Asymptotic value
Covariance range
0.3
0.25
0.2
0.15
0.1
0 20 40 60 80 100 120 140 160
(a) Distance (m)
0.45
Horizontal covariance
0.4 Vertical covariance
0.35 Asymptotic value
Covariance range
0.3
0.25
0.2
0.15
0.1
0 20 40 60 80 100 120 140 160
(b) Distance (m)
FIGURE 1.14
Two microstructures (a) and (b) with the same volume fraction but different morphologies. Also
shown are the corresponding horizontal and vertical covariances, the asymptotic values, and
covariance ranges. (From Kanit et al., 2003. With permission.)
one finds its variance, for a variance, for a microstructure of covariance

C( X, h), to be

1
P (VV ) = 2
2
[C( X, x y) P 2 ]d xd y. (1.126)
V V V
For a large specimen (V An )
P(1 P) An
P2 (VV ) = , (1.127)
V
where P = C( X,0) in equation (1.122). Having the variance, one can deter-
mine the confidence interval of the average volume fraction P : P 2 P (VV ).
Problems
1. Prove the properties of P given in Section 1.1.1.
2. Prove that, if P is a probability measure defined on R, then F (x) =
P {(, x)} possesses properties 1o to 3o of Section 1.1.2.2.
P1: Binod
July 11, 2007 20:47 C4174 C4174C001
3. Show that F defined by (1.14) satisfies properties 1o to 4o of Section

1.1.2.3.
4. Prove that P defined by (1.23) satisfies axioms I to III of probability
theory.
5. Solve Buffons needle problem: given a board ruled with parallel
lines of uniform spacing a , what is the probability of a needle of
length l a to cross one of the lines?
6. Solve Buffons needle problem by a Monte Carlo method on a com-
puter. Note that this yields an estimate of the number .
7. Prove the formula (1.49).
8. Justify the expression for the emptiness probability of a binomial
field.
9. Show that, for small values of , the von Mises distribution (1.91)
reduces to the cardioid distribution (1.90).
10. Show that P1 and P2 of Section 1.2.6 imply that, for any two disjoint
regions A and B, N ( A) and N ( B) are independent.
11. For a PoissonVoronoi tessellation of intensity , determine the
intensities of sets V (vertices) and F (cells).
12. Write computer programs to generate the modifications of Voronoi
tessellations listed in Section 1.4.3.3.
13. Write a project on the generation of arbitrary random variables from
uniform random variables using the inverse function method and
the acceptance-rejection method; consult any book on Monte Carlo
methods.
14. Consider a random variable having a Weibull distribution. (a) Simu-
late random variables from this distribution via the inverse function
method. (b) How does the scheme of (a) compare to a generation
via the acceptance-rejection method?
15. Two impulses arrive at random instants at a receiver during the time
interval [0, b]. Find the probability that at least one impulse arrives
before the instant k < b.
16. Prove that P (A|B) defined as P( A B)/P( B) satisfies Axioms I-III
of Kolmogorovs definition of a probability.
17. (a) Consider a lottery of 100 tickets in which 5 tickets are good. Find
the probability that exactly one ticket is good using the concept of
conditional probability. Hint: consider the events Ak = {k-th ticket
wins}, k = 1, . . . 3, and note that
P(C1 C2 C3 ) = P(C3 |C1 C2 ) P(C1 C2 )

= P(C3 |C1 C2 ) P(C2 |C1 ) P(C1 ).
(b) Now, solve that problem employing the concept of combinations.

P1: Binod
July 11, 2007 20:47 C4174 C4174C001
18. Consider an inhomogeneous Poisson point process of Section 1.2.6.3.

Generate with a computer a partial realization of this process on A
being a unit square, with N( A) = 100 and ( X) = e xp(2x1 x2 ).
The figure should show that the intensity gradient in x1 is more
pronounced than that in x2
19. Retrace the arguments of Jaynes (1970) to justify the first random ra-
dius method as providing the correct solution of Bertrands problem.
20. Consider a unit vector n aligned with a random fiber lying in the
(x1 , x2 )-plane, and having an orientation with respect to x1 . Then,
with the density function f () of (1.81), one can introduce second
and fourth rank tensors
2
ai j = f ()ni n j d
0
2
a i jkl = f ()ni n j nk nl d.
0
Next can define traceless orientation tensors

1
b i j = a i j i j
2
1
b i jkl = a i jkl (i j a kl + ik a jl + il a jk + jk a il + jl a ik + kl a i j )
6
1
+ (i j kl + ik jl + il jk ).
24
and similarly for higher orders. Show that if tensor basis functions
of n are
1
gi j (n) = ni n j i j
2
gi jkl (n) = ni n j nk nl
1
(i j nk nl + ik n j nl + il n j nk + jk ni nl + jl ni nk + kl ni n j )
6
1
+ (i j kl + ik jl + il jk ),
24
and similarly for higher orders, then the Fourier series representa-
tion of the density f (n) is
1 2 8
f (n) = + b i j gi j (n) + b i jkl gi jkl (n) +
2
See Onat and Leckie (1998) for the 3D case and Advani and Tucker
(1987) for a discussion of many other theoretical and applied aspects
of that model in 2D and 3D.
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
2
Random Processes and Fields
The theory of probability is the only mathematical tool available

to help map the unknown and the uncontrollable. It is fortu-
nate that this tool, while tricky, is extraordinarily powerful and
convenient.
B.B. Mandelbrot, 1983
With reference to the introduction to this book, the methods of Chapter 1 per-
tain more to discrete-type microstructures shown in Figure 1.1(a) than contin-
uous ones of Figure 1.1(b) of the Preface. Therefore, in this chapter, we move
to random processes and fields having continuous realizations. The material
is laid out as a general guide for students of mechanics and materials, who
have no particular background in stochastics. Depending on the application,
one may have to go well beyond a particular section of random processes
or random fields that are sketched here. The objective of our presentation is
not rigor, but communicating the gist of multifarious concepts and models
of stochastics such as stationarity, ergodicity, and entropy. To that end, some
example problems in stochastic mechanics involving random processes and
fields are discussed in Sections 2.2 and 2.4.
2.1 Elements of One-Dimensional Random Fields

2.1.1 Scalar Random Fields
A (scalar) random variable is a function Z assigning to an elementary event
(an outcome in a sample space) a number z on a real line, that is,
Z : R, Z() = z. (2.1)
In some applications it is preferable to define autocorrelation Z as a complex

function from the outset.
Note: The need for the enigmatic may sometimes be discarded by taking
Z as a so-called directly given random variable (Prohorov and Rozanov, 1969).
45
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
This choice would mean that = R and the function (2.1) would be simply
Z(z) = z.
Note: The term random variable is a misnomer, as it is a function in the
first place. The function
F Z (z) = P{Z z} (2.2)
is called the probability distribution of Z. Wherever F Z (z) is differentiable, we

may define the probability density of Z as
d F Z (z)
f Z (z) = . (2.3)
dz
Example 1: Measurements of random paper stiffness. The Young modulus
E of 1 1 specimens, separated by 1.5 , is measured at n (= 500) points along
a paper web, resulting thereby in a string E i , i I = {1, 2, . . ., n}. We index
this sequence of numbers by 1 , so that we deal with a string E i (1 ) = E(1 , i)
shown in Figure 2.1. For another web, indexed by 2 , we obtain another string
of numbers E i (2 ) = E(2 , i), i I , and so on.
FIGURE 2.1
A sample realization of random process E (Youngs modulus), its autocorrelation, and spectrum.
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
Random Processes and Fields 47
Guided by this example of Z = E, we are led to say that the set {Z(, i);
, i I } of all realizations Z(, i) is a (scalar) random field parameterized by
the position i. Clearly, we have two options: to keep a discrete parametrization
by i, or to consider any string as being parameterized by a continuous spatial
coordinate x. In the latter case, the set I is replaced by an appropriate subset
X R. We say that Z is a 1D random field (alternatively called a random (or
stochastic) process in 1D) if it assigns to an elementary event a realization (or
trajectory) over X, that is,
Z : X R, Z(, x) = z. (2.4)
Note: There exist four possible interpretations of Z:

1. A set, or ensemble, of all functions Z(, x): and x are variable
2. A single deterministic function (a realization) Z(): is fixed but x
is variable
3. A random variable Z(x): is variable but x is fixed
4. A deterministic number z: and x are fixed
Note: Conventional literature on random/stochastic processes typically
introduces a parametrization by the time (t) rather than space (x) coordinate.
We prefer x due to our interest in spatially random microstructures, and
hence the term random field. A dependence on time is more prominent in
the chapter on waves (Chapter 11).
With reference to (2.1), at each point x we may introduce a function
F Z (z; x) F1 (z; x) = P{Z(x) e}, (2.5)
called a first-order (or one-point) probability distribution of the process Z; recall

the four interpretations above and note that F Z (z; x) is a function of two
variables. At points where F Z (z; x) is differentiable with respect to z, we
define the corresponding first-order probability density of Z:
d F1 (z; x)
f Z (z; x) f 1 (z; x) = . (2.6)
dz
Given two points x1 and x2 , we consider two random variables Z(x1 ) and
Z(x2 ). Their joint second-order (or two-point) probability distribution is
F2 (z1 , z2 ; x1 , x2 ) = P{Z(x1 ) z1 , Z(x2 ) z2 }, (2.7)
with the corresponding density, if it exists, defined by
2 F2 (z1 , z2 ; x1 , x2 )
f 2 (z1 , z2 ; x1 , x2 ) = . (2.8)
z1 z2
Proceeding in this fashion we may introduce an nth-order (or n-point)

probability distribution of Z:
Fn (z1 , . . ., zn ; x1 , . . ., xn ) = P{Z(x1 ) z1 , . . ., Z(xn ) zn }. (2.9)

P1: Binod
July 5, 2007 16:5 C4174 C4174C002
Note that it is a function of 2n variables. The corresponding nth-order proba-

bility density is
n Fn (z1 , . . ., zn ; x1 , . . ., xn )
f n (z1 , . . ., zn ; x1 , . . ., xn ) = . (2.10)
z1 ...zn
A question naturally arises here: given a family of functions Fn , can we
find a stochastic process such that Fn s are its nth-order probability distri-
butions? The answer is provided by two Kolmogorov conditions: if Fn s are a
family of distributions dependent on n, such that for any n, any xi X, and
any permutation i 1 , . . ., i n of numbers 1, . . ., n, the following conditions hold:
1. Symmetry (i.e., invariance with respect to the permutation)
Fn (zi1 , . . ., zin ; xi1 , . . ., xin ) = Fn (z1 , . . ., zn ; x1 , . . ., xn ); (2.11)
2. Consistency m < n
Fn (z1 , . . ., zm , , . . ., ; x1 , . . ., xn ) = Fn (z1 , . . ., zm ; x1 , . . ., xm ).
(2.12)
Note: Providing the differentiability conditions hold, the above can also
be written in terms of the density function (Problem 1).
The usefulness of the Kolmogorov conditions is illustrated by this simple
example.
Example 2: Consider the case n = 2, and
2
1 1 z1 z1 z2 z22
f (z1 , z2 ; x1 , x2 ) = exp 2 + 2 ,
2 1 2 1 2 2 1 2 12 1 2 2
(2.13)
where 1 = x1 , 2 = x2 x1 , 0 < x1 < x2 , zi R1 , i = 1, 2. Then the marginal

densities are
2
1 z
f (z1 , z2 ; x1 , x2 ) dz2 = exp 12
R1 x1 2 1

1 z22
f (z1 , z2 ; x1 , x2 ) dz1 = exp (2.14)
R1 (x2 x1 ) 2 2 (x2 x1 ) 2
= (x1 , x2 , z2 ) .
Clearly, (x1 , x2 , z2 ) defined by (2.14) depends on x1 as well as on x2 , so that

the function f (z1 , z2 ; x1 , x2 ) does not satisfy the Kolmogorov conditions, and
hence it cannot represent a second-order density function f 2 (z1 , z2 ; x1 , x2 ) of
a stochastic process Z (x).
The mean (or average) of the process is

Z(x) = zd F1 (z; x) = zf 1 (z; x)dz, (2.15)
R R
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
which introduces the symbol

for the operation of ensemble averaging; it is
a function of x. Note that this is ensemble averaging because the integration
runs over the entire space . Sometimes, it is convenient to introduce a special
symbol, such as ( (x)), for the mean.
The autocorrelation of the process is defined by

R(x1 , x2 ) RZ (x1 , x2 ) =
Z(x1 ) Z(x2 ) = z1 z2 f 2 (z1 , z2 ; x1 , x2 )dz1 dz2 ,
R
(2.16)
and the autocovariance of the process is the covariance of two random variables
Z(x1 ) and Z(x2 )
C(x1 , x2 ) C Z (x1 , x2 ) =
[Z(x1 ) (x1 )][Z(x2 ) (x2 )]. (2.17)
It is easily shown that C Z (x1 , x2 ) = RZ (x1 , x2 ) (x1 )(x2 ). By setting x1 =

x2 = x, the variance of Z follows as a special case of the above ( Z(x) is the
standard deviation of Z at x)
Z(x)
2
= C Z (x, x) = RZ (x) 2 (x) =
Z(x) 2
Z(x)2 . (2.18)
Note: The covariance is symmetric C(x1 , x2 ) = C(x2 , x1 ); when the process

is complex-valued, it is Hermitian. Furthermore, by the CauchySchwartz
inequality, the squared modulus of C(x1 , x2 ) never exceeds the product of the
variances 2 (x1 ) and 2 (x2 )
C 2 (x1 , x2 ) 2 (x1 ) 2 (x2 ) = C(x1 , x1 )C(x2 , x2 ). (2.19)
Here is a very simple stochastic kinematics example in space-time (t, x),

whereby we briefly switch to a parametrization by t in place of x:
Example 3: A point is at ( O, P) at time t0 = 0 in the (t, x)-plane, and then
moves with a velocity Q on the straight line. ( P, Q) is a random vector Z. At
time t the point is at (t, X (, t)) where X (, t) = P () + t Q (). Realizations
of the X (, t) process are rays x (t) = p + q t for t 0; p and q are fixed. The
mean and the autocorrelation are found to be (Problem 2)

X(t) =
P + t
QRX (t1 , t2 ) = P 2 +
P Q(t1 + t2 ) + Q2 t1 t2
(2.20)
Example 4: With P and Q as above, we form a differential equation for

t>0
dX
P + QX = 0 (2.21)
dt
with the initial condition X (t) = 2H (t), H (t) being the Heaviside function.
Obviously, the solution is X (t) = 2 exp (q t/ p) H (t) and this stochastic pro-
cess consists of a family of exponentials.
Example 5: Consider a random process Z (x) = Y sin x, where FY ( y) is
given and x is a deterministic parameter x [1/4, 1/2]. Then, the first-order
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
distribution function is
z z
F Z (z) = P {Y sin x z} = P Y = FY . (2.22)
sin x sin x
Example 6: Consider

sin x if = tails
Y (, x) = . (2.23)
2x if = heads
Clearly, the process Yx consists of two very regular curves. Nevertheless, it is

a stochastic process.
These simple examples show that one may have a stochastic process E to be
essentially a function of one, two, or more random variables. If they are known
for one or two values of the parameter t, they are completely determined
for any t. As stochastic processes they are trivial and uninteresting. We are
rather interested in processes where randomness is richer and extends into
infinity. A modest step in that direction may be taken by describing a stochas-
tic process, as sometimes encountered in physical applications, in terms of an
analytic formula as a function of an independent parameter (such as x) and
containing a set of known random variables () = {1 () , . . ., n () , } in
the form E (, x) = h [x, ()] where the deterministic function h is given.
We thus arrive at a classical example of a process specified as
Y (, x) = A() cos kx + B () sin kx, (2.24)
where k > 0, and A, B are independent random variables with standard

Gaussian (or normal) densities N (0, ). It is easily shown that this process is,
equivalently, given by
Y (, x) = C () [cos kx + ()] , (2.25)
and its mean and autocorrelation are (Problem 4)

Y (x) = 0 RX (x1 , x2 ) = 2 cos k (x2 x1 ) . (2.26)
This process is a stepping-stone to a more general concept

n
Y (x, ) = [Ai () cos ki x + Bi () sin ki x] , (2.27)
i=1
where ki > 0, and Ai , Bi , for i = 1, . . ., n, are independent random variables

with N (0, 1). This example leads to a somewhat richer process, a so-called
Rice noise

n
Y (, x) = Ai () cos [i () x + i ()] , (2.28)
i=1
where Ai s, Bi s, i s and i s are random variables with known statistics (Prob-

lem 4). Typical uses of the Rice noise are random noise currents in physical
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
devices and random surfaces (Sobczyk, 1985; Ostoja-Starzewski, 1987) as dis-

cussed in the next section. It was shown earlier that, in principle, we describe
random processes by their nth-order distribution functions. At this point it
should be noted that the knowledge of all such functions does not give the
full information about things like continuity, differentiability, or integrability
of the realizations of stochastic processes.
From the Rice noise we go to a random Fourier series

Y (, x) = Vm () e imgx , (2.29)
m=1
where the coefficients V1 , V2 , ... are mutually uncorrelated, zero-mean random

variables, such that

Wm2 if m = n

Vm = 0
Vm Vn = . (2.30)
0 if m = n
An important condition (Papoulis, 1965) states that, in order for Y (, x) to

(, x) adequately (i.e., to minimize the mean-
represent a certain function Y
square error) for every x [X, X], g should satisfy
gX
4 sin2 1. (2.31)
4
For Y (, x) to represent Y (, x) adequately in the interval [X, X] (i.e.,

to minimize the mean-square error
|Y Y|2 ), g should satisfy
gX
4 sin2 1. (2.32)
4
Another class of random processes is the one whose values e 1 , . . ., e n , at
the respective positions x1 , . . ., xn , are independent random variables. Then

n
f n (e 1 , . . ., e n ; x1 , . . ., xn ) = f 1 (e i ; xi ). (2.33)
i=1
Problems 5 and 6 focus on two examples of this kind: the Bernoulli process and
the binomial process.
2.1.1.2 Homogeneity
Homogeneity (or stationarity) of a random process is an important property;
there are two kinds. First, a process is called strict-sense stationary (SSS) if all
n-order distributions Fn are invariant with respect to arbitrary shifts x , and
for any choice of xi s:
Fn (z1 , . . ., zn ; x1 , . . ., xn ) = Fn (z1 , . . ., zn ; x1 + x , . . ., xn + x ). (2.34)

P1: Binod
July 5, 2007 16:5 C4174 C4174C002
Next, a process is called wide-sense stationary (WSS) if its mean is constant and
its finite-valued autocorrelation depends only on x = x2 x1 :

Z(x) = ,
Z(x1 ) Z(x1 + x) = RZ (x) < . (2.35)
Note: WSS is a much weaker (less restrictive) property than SSS. The latter
implies WSS whenever Fn yields a finite second moment.
Note: If a random process can be entirely specified in terms of its first and
second moments, then it is SSS providing it is WSS. A well-known example
is the Gaussian process.
Note: Roughly speaking, random processes can be classified as stationary
and evolutionaryeach requiring specialized techniques. Given our concern
with random media, we focus here on random processes of stationary type;
evolutionary processes are more common in the case of parametrization by
time (also see Chapter 11).
It is convenient to employ a normalized autocovariance, or correlation coeffi-
cient, which is defined as
C Z (x1 , x2 )
(x1 , x2 ) Z (x1 , x2 ) = . (2.36)
Z1 Z2
Zi denotes the standard deviation of Z at xi (Problem 8). Two examples of
WSS autocovariances ((x1 , x2 ) = (x1 x2 )) are
the Gaussian curve:
(x) = exp[x 2 /2l 2 ]; (2.37)
and the exponential curve:
(x) = exp[x/l]. (2.38)
More information on models of (x) is given in the section on random fields

below.
One defines the spectral density s( ) of (x) as its Fourier transform:

1
s( ) = (x)e i x d x, (x) = s( )e i x d (2.39)
2
Note: For a real-valued random process this gives

(x) = s( ) cos( x)d . (2.40)

By Bochners theorem: every non-negative definite function has a non-

negative Fourier transform, i.e., s( ) 0.
As a simple application of this, consider a correlation function with

0 , |x| < xc
(x) = . (2.41)
0, else
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
Now, since

1 0 xc
0
s( ) = (x)e i x d x = cos( x)d x = sin( xc ), (2.42)
2 2 xc
which may take on negative values, model (2.41) is inadmissible.
For a WSS processin analogy to random processes parameterized by
timeone can define a correlation length (or correlation radius) as follows:

1
lc = 2 R(x)d x = (x)d x. (2.43)
0 0
For the Gaussian autocovariance we find lc = l(/2) 1/2 , whereas for the
exponential one lc = l.
Note: It is possible for the integral above to diverge, as is, for instance the
case with (x) = [1 + x 2 /l 2 ]a , a < 1/2.
Note: For a WSS process one can introduce a Fourier transform of its
autocovarianceit is called a spectral density (or power spectrum). In prac-
tice, autocovariances and spectra are often estimated from single realizations
of random processes (Figures 2.1 and 2.2); this procedure is based on the
ergodic assumption (Section 2.5).
FIGURE 2.2
A sample realization of random process max , its autocorrelation, and spectrum.
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
An important operation that one may want to conduct on a random proc-

ess is that of local averaging: given any realization Z() of random process Z,
consider a new process with realization ZL (), defined for each point x by
x+L
1
ZL (, x) = Z(, x )d x . (2.44)
2L xL
Note: Upon local averaging, the autocorrelations (and hence spectral den-
sities) are changed in function of L.
2.1.2 Vector Random Processes

Example: Returning to our measurements of paper properties, we now report
strength (max ) for the same 500 specimens placed on the paper web, Fig. 2.2.
This figure also shows the autocorrelation and spectral density of max based
on this realization.
Note: The use of for stress/strength and standard deviation is the well-
known clash of notation between solid mechanics and probability theory.
Clearly, we now need to generalize the concept of a scalar random variable
(2.1) to a vector random variable:

E
: R, E() = e, max () = s, (2.45)
max
whereby E and max take values e and s, respectively.

Given the parametrization of our measurements by position (either dis-
crete i or continous x), the set of all realizations forms a two-component vector
random process. In general, we say that a vector random process (or vector
stochastic process) is a vector-valued function:

E
: X R2 , E(, x) = e, max (, x) = s. (2.46)
max
In general, we shall simply write Z for an n-component vector random vari-

able, or a processs (or a field later on), taking values z so that previous relation
leads to Z : X Rn , Z(, x) = z.
As before, the complete specification of a vector random process is given
in terms of all the n-point probability distributions of Z:
Fn (z1 , . . ., zn ; x1 , . . ., xn ) = P{Z(x1 ) z1 , . . ., Z(xn ) zn }, (2.47)
which again are subject to the Kolmogorov conditions. But, such a description
is very difficult to achieve in practice. Therefore, attention focuses on WSS
fields.
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
Taking the random variables Zi and Z j at points x1 and x2 , respectively,

we call the function

Cij (x1 , x2 ) =
[Zi (x1 )
Zi (x1 )][Z j (x2 ) Z j (x2 ) ] (2.48)
joint covariance of the process. We may next generalize the normalized auto-
covariance, or correlation coefficient, as follows:
Cij (x1 , x2 )
ij (x1 , x2 ) = , (2.49)
i j
where i denotes the standard deviation of Zi (x1 ). If i = j, we have auto-

covariance, otherwise crosscovariance, of Z. Putting x1 = x2 = x, we obtain a
covariance matrix, and a correlation coefficient, at x
Cij (x)
ij (x) = . (2.50)
i j
Example: In terms of the paper properties, the covariance between E and

max components of a vector process give basic information about the relation
of the variability of E to the variability of max , both pointwise and globally. In
Ostoja-Starzewski (2001) we reported measurements on two more quantities
besides stiffness (E) and strength (max ): the strain-to-failure (max ) and ten-
sile energy absorption (TEA)all taken in the same sequence of 500 1 1
specimens. We therefore deal with a vector process ZT = [E, max , max , TEA].
2.2 Mechanics Problems on One-Dimensional Random Fields

In continuum mechanics, when we are faced with fluctuating fields and their
averages, we often take a field Z to be a superposition of the constant mean

Z and the zero-mean fluctuation Z (x):
Z(, x) =
Z + Z (, x)
Z = 0 , x. (2.51)
Next, we typically assume that the operations of ensemble averaging and

differentiation with respect to space (as well as time) commute

Z =
Z Z =
Z . (2.52)
x x t t
These concepts are directly generalized to vector random processes and ran-
dom fields in Sections 2.3 and 2.4 below.
Note: In general, formulas (2.52) are not true. This is especially the case
when time differentiation of extensive quantities is involved; see Section 10.1
of Chapter 10.
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
2.2.1 Propagation of Surface Waves along Random Boundaries

The effect of random boundary roughness on the propagation of Rayleigh-
type surface waves can relatively easily be analyzed under the assumption of
the wavelength being much larger than the roughness scale of the boundary
(Brekhovskikh, 1959; Sobczyk, 1985). When that roughness is expressed in
terms of the correlation length lc , and independent of y, we can write for a
random surface profile x of a half-space { x, y ; z (x, y)}

lc d
1 1. (2.53)
dx
That is, the roughness is assumed to be only x-dependent, so that, for the unit
vector n = nx , n y , nz normal to the surface we have
d
nx = /S 1 nz = 1/S 1, (2.54)
dx
1/2
where S = 1 ( /x) 2 . Next, x is taken as a zero-mean WSS process
modeled by the random Fourier series. For a reference, the unperturbed,
harmonic (e i t ) Rayleigh wave, propagating in the direction x, is represented
by two elastodynamic potentials:
= e i pxa z = e i pxbz , (2.55)
where
2
a 2 = p 2 k L2 k L2 =
+ 2
(2.56)
2
b =p
2 2
k T2 k T2 =

In principle, the traction boundary

conditions
should be stated in a ran-
dom (primed) coordinate system x , y , z , associated locally with every
point of the boundary z = (, x). However, in view of (2.54), one can ap-
proximate such conditions by those at the mean surface z = 0 through

zx
1
zx = zx = 0 zz
1
zz = zz = 0, (2.57)
where each field, say, zx , is represented as a superposition of the scattered

1
field (zx ) and the perturbed field (zx ). The latter one is calculated by using
the transformation of stresses from the unprimed to the primed coordinate
system, taking the first expansion of the unprimed stresses

ij
ij (x, ) = ij (x, 0) + , (2.58)
z z=0
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
and rejecting the higher-order terms

0
0
zx zz

zx = 0 d
zz = , (2.59)
z z=0 xx
dx z z=0
where zx 0
, zz
0
correspond to the unperturbed wave of (2.53).
In order to find the scattered fields, one introduces a system of plane waves
{n , n ; n = 1, 2, ...} where
n = e i pn xn z n = e i pn xn z , (2.60)
the wavenumbers
set equal to p + ng for each n, so that we have
pn being
n = k L pn2 and n = k T pn2 .
Thus, we find

zx
1
= zx
1(n)
= 2An pn n + Bn n2 n2 e i pn x , (2.61)
n=1 n=1
and a similar formula is obtained for zz

1
(Problem 9). The coefficients of scat-
tered fields are then computed from (2.57), and hence, the random boundary
effects on the Rayleigh wave can be determined. Although the methodology
outlined here can readily be extended to waves at solidfluid and solidsolid
interfaces (OstojaStarzewski, 1987), it is seen that a number of other prob-
lems in elasticity, involving weakly random boundaries, can also be analyzed
by the perturbation method.
2.2.2 Fracture of Brittle Microbeams

2.2.2.1 Randomness of Microbeams
Linear elastic fracture mechanics involves two material properties: the mate-
rial stiffness tensor C and the surface energy [the same symbol as that used
for frequency in the previous section). In a 1D situation, such as encountered
in a slender microbeam, C is represented by a scalar (e.g., Youngs modulus
E), and we have a pair [E, ]. Both of these are conventionally taken to be
constant, but, given the presence of a randomly microheterogeneous mate-
rial structure, E and are random along the beams span x, and the beam is
then described by a vector random process [E, ]x (Ostoja-Starzewski, 2004).
Assume each component of that process to satisfy (2.52).
The need to consider randomness of E arises when the representative vol-
ume element (RVE) of continuum mechanics cannot be safely applied to the
actual beam. Such a case is shown in Figure 2.3(a), where the microbeam is so
thin that its lateral dimension L (i.e., the very one defining its Youngs mod-
ulus) is comparable to the crystal size d. The comparable aspect signifies
the problem of scaling from a statistical volume element (SVE) to RVE; see
Chapters 7 to 10 for 2D and 3D problems.
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
L
P
d
(a)
(b)
FIGURE 2.3
(a) Fracturing of a microbeam of thickness L off a substrate, where a statistical volume element
imposed by the random microstructure characterized by a scale d is shown. (b) Potential energy
(
1/E) (thick line) and its scatter shown by a parabolic wedge (thin lines), summed with the
surface energy
= 2a
(thick line) and its scatter shown by a straight wedge (thin lines),
results in (
1/E) +
(thick line) and having scatter shown by a wider parabolic wedge (thin
lines). Dashed region indicates the range of a random critical crack length a c ( E ()).
2.2.2.2 Strain Energy Release Rate in Random Microbeams

First, according to Griffiths (1921) theory of elastic-brittle solids, the strain
energy release rate G is given by
W U
G= = 2 (2.62)
A A
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
where A is the crack surface area formed, W is the work performed by the
applied loads, U is the elastic strain energy, and is the energy required to
form a unit of new material surface (e.g., Gdoutos, 1993). If we consider the
case of dead-loading, the force is not random, but the kinematic variable is.
Now, only the second term in (2.62) remains, and, assuming a EulerBernoulli
beam, the strain energy is
a
M2
U(a ) = d x, (2.63)
0 2I E
where a is the crack length, M is the bending moment, I is the beams moment
of inertia, and E is the elastic modulus. Henceforth, we simply work with a =
A/B, where B is the constant beam (and crack) width. In view of Clapeyrons
theorem, the strain energy release rate is G = U/B a . Evidently, U is a
random integral
a
M2 d x
U(a , E ()) = , (2.64)
0 2I E(, x)
which, upon ensemble averaging, leads to an average energy

a
M2 d x

U(a , E) = . (2.65)
0 2I [
E + E (, x)]
In the conventional formulation of deterministic fracture mechanics, ran-

dom microscale heterogeneities E (x, ) are disregarded, and (2.64) is evalu-
ated by simply replacing the denominator by
E, so that
a
M2 d x
U(a ,
E) = . (2.66)
0 2I
E
We observe that this procedure corresponds to replacing the operator

L1 1
by
L as discussed in the preface. It amounts to postulating a priori that the
response of an idealized homogeneous material is equal to that of a random
one on average. Therefore, we are interested in making a statement about

U(a , E) versus U(a ,
E), and about
G( E) versus G(
E).
First, note that, since the random process E is positive-valued almost
surely (i.e., with probability one), Jensens inequality (Rudin, 1974) yields a
relation between harmonic and arithmetic averages of the random variable
E ()

1 1
. (2.67)

E E
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
whereby the x-dependence is immaterial in view of the assumed wide-sense

stationarity of E. With (2.65) and (2.66), this implies that
a a a
M2 d x M2 1 M2 d x
U(a ,
E) = dx = =
U(a , E) ,
0 2I
E 0 2I E 0 2I E(, x)
(2.68)
since the conditions required by Fubinis theorem (Rudin, 1974) are met.
Now, if we define the strain energy release rate G(a ,
E) in a hypothetical
material specified by
E, and the strain energy release rate
G(a , E) properly
ensemble averaged in the random material {E(, x); , x [0, a ]}
U(a ,
E)
U(a , E)
G(a ,
E) =
G(a , E) = , (2.69)
B a B a
and note that the side condition is the same in both cases
U(a ,
E) |a =0 = 0,
U(a , E) |a =0 = 0, (2.70)
we obtain
G(a ,
E)
G(a , E) . (2.71)
This provides a formula for the ensemble average G under dead-load condi-
tions using deterministic fracture mechanics for EulerBernoulli beams made
of random materials. That is, G computed under the assumption that the ran-
dom material is directly replaced by a homogeneous material (E(x, ) =
E),
is lower than G computed with E taken explicitly as a spatially varying
material property. Clearly,
G(a , E) is the correct quantity to be used un-
der dead loading, and this result may be generalized to Timoshenko beams
(Problem 11).
Note: With the beam thickness L increasing, the mesoscale L/d grows, so
1
that E 0. Thus, E 1
E, and (2.71) turns into an equality, whereby
the deterministic fracture mechanics is recovered.
Under the fixed-grip conditions, the displacement is constant (i.e., non-
random), and the load is random. Take this load to be the force P applied at
the tip. Now, only the first term in (2.62) remains, so that
U e (a ) u P
G= = . (2.72)
B a 2B a
For a cantilever beam problem, G can be computed by a direct ensemble
averaging of E under fixed-grip loading.
In the case of mixed loading conditions, both the load and the displacement
vary during crack growth, and there is no explicit relation between the crack
driving force and the change in elastic strain energy. However, we can bound
G under mixed loading (G mi xed ) by the Gs computed under dead-load (G P )
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
and fixed-grip (G u ) conditions, from above and below, respectively:
G u G mi xed G P . (2.73)
2.2.2.3 Stochastic Crack Stability

Recalling the fracture criterion (2.62), we observe that cracking along the x
axis is governed by an interplay of two random fields (parametrized by x):
the elastic property E and the surface energy density . In view of the scaling
arguments concerning the SVE versus the RVE given above, the first one
is a function of the beam thickness L, but the second one is not. Thus, for
statistically stationary and ergodic materials, the randomness of E decreases
to zero as L/d , but the randomness of remains constant. To sum
up, cracking of microbeams is more sensitive to the material randomness of
elastic moduli than cracking of, say, large plates.
Crack stability in any particular microbeam ( ), in a general loading
situation, is governed by the condition of the same form as that in determin-
istic fracture mechanics

< 0 : unstable equilibrium
( () + ())
2
= 0 : neutral equilibrium (2.74)
a 2

> 0 : stable equilibrium
Here, both the total potential energy () and the surface energy ()
are random. Now, under dead-load conditions, the correctly averaged

(shown by a solid line) is bounded from above by the deterministic esti-
mated by a straightforward averaging of E:
(
1/E) =
(
E) . (2.75)
The above follows again from (2.67). Typically, the energy goes like a 3 .
Thus, in Figure 2.3(b), we use two parabolas to indicate a wedge of scatter
associated with the mean
(
1/E) =
. (2.76)
Next, the surface energy () = 2a [

+ ()] for any . Using
two straight lines, we indicate a wedge of scatter about
= 2a
. Con-
sequently, the scatter about the mean of () + () is larger than that of
() or () alone and, at the maximum of their sum, we have a stochastic
(rather than a deterministic) competition between both contributions. Ev-
idently, according to (2.74), the critical crack length a c becomes a random
variable a c ( E ()), whose range is shown by a dashed region in Figure 2.3(b).
In view of (2.75), there is an inequality between the average a c properly cal-
culated from (
1/E):
2 [ (
1/E) +
]
= 0 a c (
1/E) =
a c ( E) , (2.77)
a 2
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
and the deterministic a c simplistically calculated from (

E):
2 [ (
E) +
]
= 0 a c (
E) . (2.78)
a 2
The said inequality is
a c (
1/E) a c (
E) . (2.79)
Note that the equality a c (

1/E) =
a c ( E) in (2.77) follows from (2.76).
Finally, Figure 2.3(b) shows that small random fluctuations in E and (i.e.,
the scatter about the maximum of (
1/E) +
) lead to relatively much
stronger (!) fluctuations in a c .
For other aspects of mechanics of random microbeams, see for example
Altus (2001); Altus and Givli (2003); Beran (1998); Givli and Altus (2003).
2.3 Elements of Two- and Three-Dimensional Random Fields

2.3.1 Scalar and Vector Fields
Let us first return to the problem of spatial variability of paper properties, this
time reporting measurements in two directions x1 and x2 . In fact, Figure 2.4
gives stiffness (E), strength (max ), strain-to-failure (max ), and tensile energy
absorption (TEA)all in a 25 8 array of 7 1 ( L 1 L 2 ) specimens.
Clearly, we have a field realization Z()with an identification of vector
components [Z1 , Z2 , Z3 , Z4 ] [E, max , max ,TEA]over a 2D domain in R2 .
This motivates us to introduce scalar- and vector-valued random fields.
In the first case we say that, say, Z is a random scalar field in D dimensions if it
assigns to an elementary event a realization over X R D , that is,
Z : X R, Z(, x) = z, (2.80)
where x X. In the second case we say that Z is an n-component random

vector field in D parameter dimensions if it assigns to an elementary event a
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
1800
1750
1700
1650
1600
13.5
13.0
12.5
12.0
11.5
1.6
1.4
1.2
1.0
0.11
0.10
0.09
0.08
0.07
0.06
FIGURE 2.4
A map (from the top) of stiffness, strength, strain-to-failure, and tensile energy absorption for
a 25 8 array of 7 1 specimens (After DiMillo and Ostoja-Starzewski (1998); paper web
provided by Champion Corp.)
realization over X, that is,
Z : X Rn , Z(, x) = z. (2.81)
Note: There exist four possible interpretations of E and Z (an ensemble,

a field realization, a random variable at specific position, and a value) just as
with random processes.
Note: One might say that the random processes parametrized by spatial
coordinates, treated in the preceding section, are a special case of random
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
fields; the term process is reserved by some authors for parametrization by

time, and field for parametrization by space.
Note: As before, the complete specification of a random vector (or scalar)
field is given in terms of all the n-point probability distributions of Zthat is
Fn (z1 , . . ., zn ; x1 , . . ., xn ) = P{Z(x1 ) z1 , . . ., Z(xn ) zn }. (2.82)
A classical theorem states that, for any complex number r (r = 1, . . ., m)

and any x1 , x2 Rn the correlation function (x1 , x2 ) has the following
properties (the overbar indicates a complex conjugate):
1. (x1 , x1 ) 0
2. (x1 , x2 ) = (x2 , x1 )
3. |(x1 , x2 )|2 (x1 , x1 )(x2 , x2 )
m
4. (x1 , x2 ) j k 0
j,k=1
Analogous theorems can be set up for the cross-correlation and multiple

correlation functions. This theorem can also be stated for WSS random fields
whereby (1) simplifies to (0) 0and multiply correlated random fields.
We now focus attention on WSS random fields, that is, when

Z(x) = ,
Z(x1 ) Z(x1 + x) = RZ (x) < . (2.83)
The concepts of correlation (and covariance) functions are very simply gen-
eralized by substituting x for x. Thus, taking the random variables Zi and Z j
at points x1 and x2 , respectively, we call the function

Cij (x1 , x2 ) =
[Zi (x1 )
Zi (x1 )][Z j (x2 ) Z j (x2 ) ] (2.84)
joint covariance of the process. We next generalize the normalized covariance

(2.20), or correlation coefficient, as follows:
Cij (x1 , x2 )
ij (x1 , x2 ) = , (2.85)
i (x1 ) j (x2 )
which, for a WSS field, simplifies to
Cij (x)
ij (x) = . (2.86)
i (x) j (x)
Its one-point special case is
Cij (0)
ij (0) = . (2.87)
i (0) j (0)
A special class of so-called isotropic random fields needs now to be defined.
They occur when the autocorrelation (or autocovariance) depends only on
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
the magnitude, but not direction, of the vector x = x1 x2 connecting two

points x1 and x2 :

(x) = (x), x |x| = xi xi . (2.88)
Note: An analog of isotropy property for random fields in 1D is: (x) =

(x), x = x1 x2 .
Note: For WSS isotropic random fieldsjust as for random processes
one can define a correlation length (or radius) by the same definition as in the
previous section.
Two rather wide classes of correlation function for isotropic scalar fields
are
(x) = exp[Ax ], A > 0, 0 < 2; (2.89)
(x) = [1 + Ax ]1 , A > 0, 0 < 2. (2.90)
Note: For an isotropic field on R D , (x) 1/D.

Note: A valid isotropic (x) in R D2 is always a valid isotropic (x) in R D1 ,
where D2 > D1 . However, the converse is not true as this example illustrates:
the tent function

2 (1 |x| /a ), 0 |x| < a
(x) = (2.91)
0, |x| > a
is valid in R1 , but not in R2 .

Using simpler models, one may construct new, more complex correlation
functions. The way to proceed is first to note a few facts from probability
theory:
1. A convex combination of probability distributions is a probability

distribution; the same holds for density functions.
2. A convex combination of correlation functions is a correlation func-
tion.
3. A finite product of correlation functions is a correlation function.
On this basis one obtains

exp[Ax ]
(x) = , A, B > 0, 0 < , 2; (2.92)
1 + Bx

r
(x) = exp[ Ax s ], As > 0, 0 < s 2, s = 1, . . ., r ; (2.93)
s=1

r
(x) = [ (1 + Bs x s )ls ], Bs > 0, 0 < s 2, ls = 1, 2, ...; (2.94)
s=1
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
(x) = exp[Ax ](cosh Bx ) s , A + B(2l s) > 0, 0 < 2,

s = 1, . . ., r ; (2.95)
(cosh Bx ) s
(x) = , A + B(2l s) > 0, 0 < 2, s = 1, . . ., r.
1 + Ax
(2.96)
See Szczepankiewicz (1985) for rigorous derivations of the above, as well as

more advanced models.
Very recently, a new class of so-called Dagum correlation functions has been
developed (Porcu et al., 2006)

(x) = 1 1 + x , < (7 ) (1 + 5) , < 7. (2.97)
Here and act as smoothing parameters and is a scale parameter.
2.3.1.2 Example
In the problem of paper properties stated above, ij (x) is estimated invoking
both stationarity and ergodicity. With the identification E, max , max , TEA

Z1 E
Z
2 max
, (2.98)
Z3 max
Z4 TEA
we compute the following (symmetric) matrix:

E, E

E,max max ,max sym
ij (0)
E,max max ,max max ,max
E,TEA max ,max max ,TEA TEA,TEA
(2.99)
1
0.43
1 sym
= .
0.10 0.56 1
0.14 0.90 0.69 1
From this we note:
1. Cross-correlations between E and inelastic parameters max , max , TEA

are weak, although we note that E,max is greater than E,max or
E,TEA ;
2. Three cross-correlations between max , max , and TEA are about the
same.
P1: Binod
July 5, 2007 16:5 C4174 C4174C002

Next, we present ij for the nearest neighbors separated by L 1 = 7 and
L 2 = 1 , in the x1 and x2 directions, respectively

0.24
0.21 0.27 sym

ij (L 1 ) = ,
0.00 0.10 0.21
0.14 0.23 0.15 0.22
(2.100)
0.33
0.18 0.13 sym

ij (L 2 ) = .
0.09 0.02 0.26
0.05 0.04 0.04 0.04
2.3.1.3 Properties of the Correlation Function Anisotropy

Evidently, the field at hand is not isotropic. To account for preferential direc-
tions, one generally proceeds as follows: introduce a transformation
T(x) = [T1 (x), . . ., T1 (x)], x = (x1 , . . ., xD ) (2.101)
of R D into R D . If the Jacobian of T is different from zero, there exists an inverse

transformation T 1 . Let |x| denote a norm in R D and ||y|| a norm in R D after
T. If a WSS field Z has a property that |x| = ||y|| implies
(x) = (y), (2.102)
then Z is called a WSS quasi-isotropic random field.

Of particular interest are fields with ellipsoidal structure. Let b ij be the matrix
of a positive-definite quadratic form

n
||y||2 = b ij yi y j , (2.103)
i, j=1
whenever y = 0. Consider (||y||). Then the Jacobian of T is the determinant

of matrix b ij . For example, the Gaussian autocovariance of a random process
on R1 can now be generalized as

n
(y) = exp[ b ij yi y j ]. (2.104)
i, j=1
Another model of a quasi-isotropic field on R D is
T(x) = [a 1 x1 , . . .,a D xD ], all a i > 0. (2.105)
Note: One needs to distinguish the anisotropy in terms of the correlation

function from the anisotropy of realizations.
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
Note: Models of the correlation function for isotropic random fields (2.88)
carry over to quasi-isotropic fields provided that we replace x by the norm
||T 1 (y)||. For example, (2.89) leads to
(x) = exp[A||T 1 (y)|| ], A > 0, 0 < 2. (2.106)
Separable structure: An important property that a random field may pos-

sess is that of a fully separable
(x) = (x1 )(x2 )...(xD ), (2.107)
or a partially separable structure, such as in
(x) = (x1 , x2 )...(xD ). (2.108)
A common model in fluid mechanics is the one involving a separation be-

tween space coordinates and time: (x, t) = (x)(t).
Local averaging: This operation, already introduced in (2.44), may be gen-
eralized to random fields: given any realization Z(, x) of random field Z,
consider a new field with a realization ZL (, x), defined for each point x, by
(e.g., Vanmarcke, 1983)

1
ZL (, x) = 2 Z(, x )dx , (2.109)
L DL
where DL is a square-shaped neighborhood NL centered at x. If applied to a

stiffness C (respectively, compliance S) tensor field, this would yield a Voigt-
type (Reuss-type) estimate/bound of the stiffness (compliance) for the do-
main NL . Clearly, the local averaging is a simple operation, but it may yield
very misleading estimates of actual material properties.
Note: DL in (2.108) plays the role of a mesoscale with respect to the ac-
tual microscale where the random fluctuations of C reside. See Chapter 8
for micromechanically based mesoscale random fields that smooth the mi-
crostructures, and the stochastic finite elements based thereon.
2.3.2 Random Tensor Fields

Although the previous section treated a random vector field, the vector there
was not really considered as a tensor of rank oneit was simply an ordered ar-
ray of scalars. This is why a simple generalization of stationarity and isotropy
with respect to each component of Z was possible. Proceeding in the same
way for an actual tensor is not logical because a rotation of the coordinate
system is tantamount to a transformation of its components.
Let us then consider the vector z = Z(, x) to be an actual first-rank tensor,
that is, an object transforming according to the rule
z = Q z, (2.110)
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
where Q is a matrix giving x = Q x upon rotation. Of course, for an nth-rank

tensor we have
z = Q...Q z, (2.111)
where Q appears n times.

This suggests defining a random tensor field of first-rank to be wide-sense
stationary and isotropic whenever the mean
Z(x) and the correlation ik (x)
do not change when subjected to arbitrary shifts and when transforming by
rotation in (2.110) according to

Z = Q
Z , i (x ) = Q i (x),
j j
(2.112)
whereby x is transformed into x according to (2.109).

Note: This definition may be extended to strict-sense stationarity (SSS) by
requiring the above property to hold for all n-order distributions Fn , not just
for the first and second moments.
In (2.112) we employ the correlation coefficient

j
[i (x1 )
i (x1 )][ j (x2 ) j (x2 ) ]
i (x1 , x2 ) = , (2.113)
i (x1 ) j (x2 )
j
which is the same thing as (2.85), but the notation i is superior to ij when
dealing with tensor fields below.
The WSS property means that, for any pair (x1 , x2 ),
j j
i (x1 , x2 ) = i (x), x = |x1 x2 | . (2.114)
Next, the isotropy says

j j
i (x) = i (x), x |x| = xi xi . (2.115)
j
Robertson (1940) showed that i (x) in 3D admits this representation (see
also Lomakin, 1965, 1970; Sobczyk and Kirkner, 2001):
j
i (x) = K 1 (x) xi xk + K 2 (x) ik (2.116)
wherein the K i s are real-valued functions of x = |x|:
K 1 = 11
1
= 22
1
K 1 + K 2 = 33
1
(2.117)
j
and ni = xi /x. Here ij1 stands for the correlation i (x) between Cij (x1 ) and
Ckl (x2 ) in a coordinate system centered at x1 and directed to x2 .
2.3.2.1 Second-Rank Tensor Fields

In general, we say that Z is a random tensor field in D dimensions if it assigns
to an elementary event a realization over X R D , that is,
Z : X Rn , Z(, x) = z (2.118)
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
where x X. In general, for D = 2, n = 4, while for D = 3, n = 9. Of course,

the concept of a random tensor field may be applied to just about any tensor
field encountered in continuum mechanics.
The normalized correlation function of, say, a second-rank tensor field is

[Zij (x1 ) Zij (x1 ) ][Zkl (x2 )
Zkl (x2 )]
ij (x1 , x2 ) =
kl
, (2.119)
ij (x1 )kl (x2 )
where ij (x1 ) and kl (x2 ) are the standard deviations of the pair [zij (x1 ) ,
zkl (x2 )] at respective points. It follows from the tensor property of z that
ijkl is a fourth-rank tensor.
As a special case, take Z as an antiplane stiffness tensor C ( Cij ) of a hyper-
elastic material the conductivity-type tensor. Then, the well-known symmetry
holding at x X of every realization C () of the random field
Cij = C ji (2.120)
implies these symmetries of ijkl
ijkl = klji = ijlk . (2.121)
Now, recall after Robertson (1940), Lomakin (1965, 1970), and Sobczyk and
Kirkner (2001) that ijkl (x) in 3D admits this representation:

ijkl (x) = K 4 (x) ij kl + K 6 (x) ik jl + il jk

+ [K 5 (x) K 6 (x)] n j nk il + ni nl jk + ni nk jl + n j nl ik

+ [K 3 (x) K 4 (x)] ni n j kl + nk nl ij
+ [K 1 (x) + K 2 (x) 2K 3 (x) 4K 5 (x)] ni n j nk nl , (2.122)
wherein the K i s are
K 1 = 1111
1
K 2 = 2222
1
K 3 = 1122
1
(2.123)
K 4 = 2233
1
K 5 = 1212
1
K 6 = 2323
1
,
they satisfy the relation
K 4 + 2K 6 K 2 = 0, (2.124)
and ni = xi /x. Here ijkl

1
stands for the correlation between Cij (x1 ) and Ckl (x2 )
in a coordinate system centered at x1 and directed to x2 . In Chapter 8, we return
to this result in the context of those tensor fields for which each realization
C () is locally isotropic, i.e., x X.
Interestingly, while a second-rank tensor in 3D generally has nine inde-
pendent components, the assumption of isotropy of its realizations reduces
the number of parameters of its correlation function to six.
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
2.3.2.2 Fourth-Rank Tensor Fields

We say that Z is a random tensor field in D dimensions if it assigns
to an
elementary event a realization of a fourth-rank tensor z = zijkl over
X R D , that is,
Z : X Rn , Z(, x) = z. (2.125)
In general, for D = 2, n = 16, while for D = 3, n = 81. For example, in classical

(nonmicropolar) hyperelasticity, where Z stands for the stiffness tensor, we
generally have n = 21. As an example, if we were able to measure not just
the uniaxial stiffness (E) but the entire stiffness tensor (C, or Cijkl ) at each and
every point x of a realization of the random medium, we would deal with a
random field of stiffness tensor.
We focus on Z being WSS. Then, the normalized correlation function of
such a fourth-rank tensor field is

prst
[Zijkl (x1 ) Zijkl (x1 ) ][Z pr st (x2 ) Z pr st (x2 ) ]
ijkl (x1 , x2 ) = , (2.126)
ijkl (x1 ) pr st (x2 )
where ijkl (x1 ) and pr st (x2 ) are the standard deviations of the pair [Zijkl (x1 ),
Z pr st (x2 )] at respective points. It follows from the tensor property of Z, that
prst
ijkl is an eighth-rank tensor.
Analogously, for a random fourth-rank tensor field Z we have

prst
[Zijkl (x1 ) Zijkl (x1 ) ][Z pr st (x2 ) Z pr st (x2 ) ]
ijkl (x1 , x2 ) = , (2.127)
ijkl (x1 ) pr st (x2 )
where ijkl (x1 ) and pr st (x2 ) are the standard deviations of the pair [zijkl (x1 ) ,
z pr st (x2 )] at respective points. It follows from the tensor property of Z, that
prst
ijkl is an eighth-rank tensor.
prst
Since Z is WSS, ijkl has the property
prst prst
ijkl (x1 , x2 ) = ijkl (x), (2.128)
where x = x2 x1 . In the case of isotropy of the correlation function, we have

prst prst
ijkl (x) = ijkl (x). (2.129)
Now suppose that Z stands for a random stiffness tensor C ( Cijkl ) in a lin-
ear hyperelastic material. In that case, these well-known symmetries holding
at x X of every realization C ()
Cijkl = Cjikl = Cijlk = Cklij (2.130)

prst
imply these symmetries of ijkl
prst prst prst prst rpst prts stpr
ijkl = jikl = ijlk = klij = ijkl = ijkl = ijkl . (2.131)
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
Now, let K i (x), i = 1, . . ., 15, be fifteen scalar functions of x. Note from

prst
Lomakin (1965, 1970) that ijkl (x) of a WSS and isotropic random field in 3D
admits this representation
prst
ijkl (x) = F12 + K 13 (x) ij kl pr st

+K 14 (x) ij kl ps r t + ij kl pt r s + ik jl pr st + il jk pr st

+K 15 (x) ij jl ps r t + ik jl pt r s + il jk ps r t + il jk pt r s
wherein F12 is a linear function of all K i (x), i = 1, . . ., 12, each term involving
combinations of the vector x and the Kronecker deltas in pairs from i, j, k, l
and p, r, s, t.
The concepts of separable structure and local averaging introduced ear-
lier apply, in appropriately generalized forms, to random fourth-rank tensor
fields.
2.4 Mechanics Problems on Two- and Three-Dimensional

Random Fields
In this section we discuss two problems of random media: (1) mechanics
of materials with the randomness inherent in its constitutive law, and (2)
turbulence in a fluid with constant Newtonian viscosity, that is, formation of
random velocity and stress fields, due to the inherent fluid instability.
2.4.1 Mean Field Equations of Random Materials

As noted in the preface, in mechanics of materials with spatial randomness,
the field is governed by the equation Lu = f, where L is typically a linear
differential operator, f is the source or forcing function, and u is the solution
field. Using the decompositions
u =
u + u L =
L + L , (2.132)
and first averaging the original equation and then subtracting the result from
it, we obtain an equation governing u

L u + ( I P) L u = L
u . (2.133)
Here P is a so-called projection, basically signifying an averaging operation.

Solving (2.133) for u , we arrive at an equation governing
(
L M)
u = f, (2.134)
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
where
& 1 '
M = L
L + ( I P) L L (2.135)
stands for the so-called mass operator; 1 indicates an inversion.

Following Fishman and McCoy (1981), a more concrete view of this pro-
cedure is given in terms of(the heat conduction ) problem in a medium de-
scribed by a random field K ij (, x) ; , under the action of a source
term f (x) and the boundary temperature field T 0 (x), both slowly varying
on the macroscale. The heat flux is denoted by q i . Thus, for each realization
B () B, we have
q i,i = f (x) x B
q i = K ij (, x) T, j x B (2.136)
T (x) = T 0 (x)
x B,

Note that (2.136)2 dictates K ij (, x) = K ij + K ij (, x), so that, upon
averaging, we obtain this set of equations:

q i ,i = f (x) x B
& '

q i = K ij T, j + K ij (, x) T,j x B (2.137)

T (x) = T 0 (x) x B.
From the equations governing the fluctuations

q i ,i = 0 x B

q i = K ij T,j + ( I P) K ij (, x) T,j + K ij (, x) T,j x B (2.138)
T (x) = 0 x B,
we derive an analog of (2.133)

* +
K ij T, ji + ( I P) K ij (, x) T,j = K ij (, x) T,j . (2.139)
,i
Hence, the constitutive equation for the average fields is

q i (x) = K ij T, j + K ij x, x T, j x ,i
dx . (2.140)
B
Note: That the effective constitutive response (2.140) has a nonlocal char-
acter; this result carries over to random elastic and inelastic materials (Beran
and McCoy, 1970a; Eimer, 1971; McCoy, 1991); Problem 14.
2.4.2 Mean Field Equations of Turbulent Media

Mechanics of random materials focuses primarily on problems driven by
random fields of constitutive properties, whereas studies of turbulence deal
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
with spatiotemporal random velocity fields due to inherent instabilities of

fluids. In the latter subject area, we recall the continuity equation, the linear
momentum theorem, and the energy conservation law of an incompressible,
viscous fluid.
vi,i = 0 vi = i j, j u = ij dij q i,i , (2.141)
where vi is the velocity, is the mass density, dij is the deformation rate tensor,
u is the internal energy, and q i is the heat flux. Now, take the velocity vi to be
the average superposed with the fluctuation according to (2.51). Assuming
the properties (2.52) to hold for all the fields involved, we then find

vi,i = 0
vi = ij, j u = ij dij q i,i

, (2.142)
where

ij = ij vi vj (2.143)
is the so-called Reynolds stress (also denoted Rij ),
1
u =
u + vv (2.144)
2 i i
is the sum of the average internal energy and the kinetic energy of random
(microscale) fluctuations, while
& '
1
q i = qi ij vi + u + vj vj

vi (2.145)
2
is the effective heat flux appearing as the sum of the original heat flux with
(1) the rate of work of the fluctuating stresses on the surface of an elementary
volume and (2) the stochastic energy convection through the boundary.
Note: The average field
vi satisfies the same continuity equation as vi .
The divergence of the Reynolds stress may be interpreted as the force density
on the fluid due to turbulent fluctuations. The Reynolds stress also appears
when analyzing the Euler or NavierStokes equations (Problem 15).
Note: While, formally,
vi in the above equations is the ensemble average,
in practice, this average is sometimes also thought of as a spatial average over
some length scale, or a temporal average. Accordingly, the fluctuation v is
then interpreted not as a statistical one, but a spatial or temporal one. In the
case of a temporal average, one works with a separation of scales: the time
scale of variation of
v is much larger than that of v . The equivalence between
such averages in statistical turbulence is an open problem, but is justified in
the more established field of equilibrium statistical mechanics by the ergodic
theorem, as discussd in the next section.
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
2.5 Ergodicity
2.5.1 Basic Considerations
It is clear by now that random fields (or processes) are sets of realizations over
some space (respectively, time) domain. Of primary importance is a possibil-
ity of determination of probabilistic characteristics (e.g., moments, distribu-
tions) of a random process or field in terms of a set of values {z(x1 ), . . . , z(xn )}
observed over just one realization Z(, x), x Ras is often the case in
practice. At first sight, such an approach appears impossible because it seems
that not much statistics (if any) may be obtained from a single realization.
Nevertheless, one intuitive explanation of such an approach lies in a possi-
bility of treating the realization Z() at hand as typical (in a certain sense,
as discussed below) of the whole field Z, recall Figures 2.1 and 2.2 of this
chapter. Another explanation involves treating the realization observed in
an interval (x0 , xn ) as a set of separate realizations observed in the intervals
(x0 , x1 ), (x1 , x2 ), . . ., (xn1 , xn ).
In the engineering literature this kind of supposition is called an ergodic
property or ergodicity; in mathematics these terms have a narrower meaning
(see below). Now, ergodicity in the mean (or, the process is mean-ergodic) means
that any realization Z(), is sufficient to get the ensemble average

Z(x) at any x from its spatial average Z() for any taken over a
sufficiently large interval:
L
1
Z() lim Z(, x)d x = Z(, x)d P()
Z(x). (2.146)
L 2L L
Several issues arise here.
2.5.2 Computation of (2.146)

The left-hand side of (2.141) may be evaluated only with some accuracy, both
because of the finite scale discretization of measurements and the impossi-
bility of carrying out the limit L . In practice, Z() must be replaced,
respectively, by a spatial average (generally, volume average) from a finite
number of sampling points N (taken over one realization )
1
N
Z() Z(, xn ). (2.147)
N n=1
Additionally, also in a real application,

Z(x) must be computed from the
ensemble average over a finite number M of realizations (taken at a chosen
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
sampling point x)
1
M

Z(x) Z(m , x). (2.148)
M m=1
2.5.3 Conditions for (2.146) to Hold

Assuming the limit in (2.146) exists, its value Z(), in general, depends on .
Under what conditions does it equal the constant ? The answers are provided
by the so-called ergodic theorems. The first one of these states that the process
is mean-ergodic iff its autocovariance is such that
L L
1
lim C(x1 , x2 )d x1 d x2 = 0. (2.149)
L 4L 2 L L
This is proved by noting that the variance L2 of the random variable SL () =

,L
(2L) 1 L Z(, x)d x of the process Z is
L L
1
L2 = |SL L |2 = C(x1 , x2 )d x1 d x2 (2.150)
4L 2 L L
where L is the mean of SL ().
Note: A process may be ergodic without being stationary.
Upon introducing the WSS property of Z, one readily finds that a proc-
ess is mean-ergodic if its autocovariance C (x) = R (x) 2 is such that
(Problem 16):
2L
1 |x|
lim C(x) 1 d x = 0. (2.151)
L 2L 2L 2L
,
One gets sufficiency here if C(x)d x < . There are also other kinds
of ergodicity: correlation-ergodic, distribution-ergodic (Papoulis, 1984), for
example. The former of these is expressed by
L
1
RZ (, x) lim Z(, x1 + x) Z(, x1 )d x1
L 2L L
(2.152)
= Z(, x1 + x) Z(, x1 )d P()
Z(, x1 + x) Z(, x1 ) ,

which, in fact, was the basis for computation of autocorrelations in Figures

2.1 and 2.2.
2.5.4 Existence of the Limit in (2.146)

Under what conditions does the limit Z() in (2.146) exist? This is actually
known as the so-called ergodic problem in the mathematics literature. This
problem of ergodicity has its roots in statistical mechanics, where one is inter-
ested in estimating system properties from a single trajectory over a relatively
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
very long period of time. Because a trajectory occurs in a phase space, this
leads us to the concept of an ergodic flow in the phase space, that is, a flow for
which the integral with respect to a time parameter t converges to a random
-
variable Z()

1 -
lim Z(, t)d x = Z(). (2.153)
L L
More generally, for a random process Z indexed by t, one can define a

flow in phase (or state) space of Z as a transformation mapping this space
onto itself, whereby any event A is transformed into some other one, A , by
means of an operator T( ):
A A = T( ) A. (2.154)
As a result, a function g( A) transforms as
g( A) g( A ) = g[T( ) A]. (2.155)
In classical statistical mechanics one focuses on Hamiltonian flows (i.e.,

those where the total energy of the system is conserved), which are measure-
preserving in the sense that
P( A) = P( A ). (2.156)
In (2.154) A and A stand for an initial set and a set after the transformation
(2.155). The measure is a mathematical term, whose counterpart in physics
is the density (q n , pn , t) in the phase space of coordinates q i and conjugate
momenta pn , n = 1, . . ., N (the number of degrees of freedom); we return
to in Chapter 10. Note that a Hamiltonian system is measure-preserving
because of Liouvilles theorem expressed in terms of the convective derivative
being zero (Problem 17):
d
= + q n + pn = 0, (2.157)
dt t q n pn
where the summation convention is implied. Recall here Hamiltons equa-

tions
H H
q n = pn = . (2.158)
pn q n
Let us now consider three basic types of measure-preserving flows in

phase space, as depicted in Figure 2.5. In each case we let the set A modeling
the state of the system evolve while keeping another set B (akin to a control
window) fixed. In the case (a), the set A moves in a periodic fashion through
the phase space. It visits just a fraction of it without ever entering other regions;
a harmonic oscillator is a very simple example:
p2 1
H (q , p) = + kq 2 (2.159)
2m 2
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
(a) (b) (c)
FIGURE 2.5
Three types of flow in phase space: (a) periodic, nonergodic; (b) ergodic, nonmixing; (c) ergodic,
mixing. (After Balescu [1975]). In each case, a square-shaped set B is shown.
In the case (b), we see the set Aonly slowly being altered during its motion,
while it sweeps the entire space if observed for a sufficiently long (infinite)
timethis is an ergodic flow. It was shown by G.D. Birkhoff in 1931 that, in
this flow, every invariant set has the measure 0 or 1 (meaning it comprises
almost whole space), or, intuitively, no trajectory can be confined to a finite
portion of phase space because it has to wander (ergodos in Greek) through
all of it. Another way to express this is that the trajectory of the process is not
sensitive to initial conditions.
Finally, in the third case (c), the set A not only sweeps the entire space,
but its shape is being altered during its motion so as to fill the entire space
through a multitude of growing branches, subbranches, and so onthis is
a mixing flow. Indeed, Figure 2.5(c) shows three stages: an initial one at time
zero when A is just a small blob, a second one when A has already moved and
diffused somewhat, and a third stage when A begins to diffuse even more. As
time goes on, since the flow is mixing, A begins to intersect B, and tends
to a uniformly mixed situation where the volume fraction of A in B will be
the same as the initial volume fraction of A in the whole space , that is,
lim P( A B) = P( A) P( B). (2.160)

It is easy to show that mixing implies ergodicity: If A is a measure-

preserving set, then A = A, and hence A A = A, so that
P( A A) = P( A). (2.161)
On the other hand, because the flow is mixing, then by (2.160) above, setting
B = A, we find
lim P( A A) = [P( A)]2 . (2.162)

Comparing (2.160) with (2.162), we find
P( A) = [P( A)]2 , (2.163)
which is possible only when P( A) equals 0 or 1. Clearly, there may exist

ergodic flows that are not mixing, Figure 2.5(b).
Note: Evolution of a mixing flow: the set A is painted black so that, fol-
lowing J.W. Gibbs, we can think of ink mixing with water. Some people may
prefer to think in terms of mixing cocktails.
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
The condition (2.160) is sometimes called strong mixing. A weaker case is

the so-called weak mixing described by

1
lim [P( A B) P( A) P( B)] d = 0. (2.164)
0
An even weaker case is specified by the condition

1
lim [P( A B)] d = P( A) P( B). (2.165)
0
Note: (2.160), (2.164), and (2.165) are various extensions of the conven-
tional independence property [P( A B) = P( A) P( B)] when the time evolu-
tion is involved.
Motivated by this simple account of time-dependent flows (processes),
we now consider a transformation of the state space with probability dis-
tribution P in the context of a directly given random field E, defined on R D ,
D being the number of spatial dimensions. In analogy to (2.154), consider a
family of shift transformations
( j)
g( E) g( E ) = g[T E] (2.166)
where
( j)
T g(x1 , . . ., x j , . . ., xD ) = g(x1 , . . ., x j + , . . ., xD ). (2.167)
( j)
Here g(x) = , each of which takes a set S into a set S composed
of the functions of S shifted by at their jth parameter x j . If the random field
is strictly homogeneous, then these transformations are measure preserving
in the sense that
( j)
P(S) = P(S ). (2.168)
( j)
A set S is called invariant if, for every j and , the sets S and S differ at most
by a set of P-measure zero. We say that the strictly homogeneous random field
is ergodic if every invariant set has either probability zero or one. See Adler
(1981) for a rigorous account of this subject.
2.6 The Maximum Entropy Method

2.6.1 Cracks in Plates with Holes
A recent study (Al-Ostaz and Jasiuk, 1997) investigated fracture response
in tension of both elastic-brittle (epoxy) and ductile (aluminum) plates
having some thirty holes punched in a specific disordered pattern in them.
In the case of each material, several macroscopically identical specimens were
tested under the same conditions, and each displayed a different crack
pattern. The situation is depicted in Figure 2.6 for the elastic-brittle case.
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
FIGURE 2.6
Schematic plot of final crack patterns superposed from seven epoxy specimens (in various colors)
under the same uniaxial vertical tensile loading conditions, obtained experimentally. (After Al-
Ostaz and Jasiuk [1997].)
The typical experiment involved a number of nominally identical plates (8.25

cm 33.02 cm), each of a uniform thickness 0.38 cm, and made of epoxy
(PSM-5). Each plate had the same nonperiodic, nonuniform distribution of
thirty-one randomly distributed circular holes of the same size, an inch in
diameter. The locations of the holes centers were generated according to a
hard-core (sequential-inhibition) point field with exclusion. Each plate was
subjected to a uniaxial tensile loading in the y-direction at a constant displace-
ment rate of 0.03 cm/s initially, and then decreased to 0.0017 cm/s in order
to capture more details of the fracture process.
The nonuniqueness of the experimental results may be understood by not-
ing, on purely combinatorial grounds, that there exist a large number of geo-
metrically acceptable (plausible) crack paths cutting the specimen across and
this number grows rapidly as we consider systems with more holes. Whether
we assume a maximum strain energy or a maximum principal stress energy
criterion, the respective values associated with all these paths do not differ
much from one another. Thus, minute material and geometric imperfections
are likely to decide which crack path will actually take place in a particular
specimen that is nominally (on the macroscopic scale) the same as the others.
The material imperfections arise from the intrinsic nature of materials that
are all heterogeneous at a microscopic level. The geometric imperfections in-
clude roughness of the holes surfaces and microscopic damage (microcracks
and other surface flaws) due to drilling, which may give rise to singular stress
fields.
2.6.2 Disorder and Information Entropy

The phenomenon described above can be (and was) tackled by numerical
methods involving finite elements and spring networks (Al-Ostaz and Jasiuk,
1997), resulting in distinct crack paths and distinct overall load-displacement
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
diagrams. However, there arose a basic question: which crack path should
actually be used as guidance in fine-tuning the computational mechanics
model? In principle, by adjusting the elastic parameters and failure crite-
ria, as well as by introducing noise in these parameters, one could repro-
duce all the crack pathsand this is the methodology of the right column in
Figure 2.7.
The best-known variational principles of mechanics are those based on
minimization of energy. They are the basis on which modern continuum me-
chanics, having its roots in classical dynamics, is built. These principles are
deterministic. Let us consider the minimum potential energy principle in the
setting of mechanics of random media: for a given body B() B
min [U () W] (2.169)
where U () is the strain energy and W is the work of body forces and surface
tractions. In light of the basic considerations in the preface to this book, the
above could be further generalized by taking W to be random as well, but
this is not crucial here.
As is well known, (2.169) forms the starting point for, say, a finite element
method (FEM), which leads to a method of quantitatively analyzing each and
every specimen of the ensemble separately. The FEM is set up from a system
MACROSCALE
maximum entropy method (MEM)
ensemble {B(); } estimate P(Wfailure)

W () = work-to-failure
failure
n times (Monte Carlo)

(i) get fu curves MG
(ii) estimate possible crack paths B() for ensemble MG
estimate fracture condition:

surface energy solve for W () from
failure
min [U() W]|constraints
maxH|
constrainsts
postulate failure criterion P()
estimate P()
minimum potential energy method
MICROSCALE
FIGURE 2.7
Philosophies of two variational principles in mechanics of random media, and their roles in
establishing a connection between microscale and macroscale responses. The left column illus-
trates a maximum entropy approach where the microscale probabilities are determined from the
macroscale statements and observations, which represent constraints. The right column illus-
trates a (usually) much more familiar variational method of deterministic mechanics extended
to a representative subset of heterogeneous specimens of a random medium.
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
of equations governing the entire domain of the body B() partitioned into a
number of elements. Once the solution is found, one has to repeat the entire
numerical analysis from the beginning to establish a solution for another
specimen of , see Figure 2.7. Thus, in order to investigate a specific effect
of microscale randomness, one needs to carry out a finite, and necessarily
limited, number of studies (n MC ) on a subset MC .
Of course, if an explicit analytical solution is available, one can proceed di-
rectly to the stochastic stage, and find, say, probability distributions of macro-
scopic response in terms of the microscale noise. Unfortunately, such nice
solutions are rather rare and a bridge from microscale to macroscale may be
too cumbersome or unwieldy. However, Figure 2.7 shows that there is another
way of building this bridge: rather than to proceed from micro to macro, one
can go the opposite way. This, in fact, is accomplished by a so-called maximum
entropy method (MEM): for a statistical ensemble (on which a random vari-
able X = {X(); } has been defined), maximize the possible disorder as
expressed by the information entropy

h ( X) =
ln p (x) = p (x) ln p (x) d x (2.170)

subject to the condition that the expected values =

gi (x) of n known
functions gi (x) of X are given.
In the case of a discrete-type random variable (taking values i = 1, . . ., I ),

I
h ( X) =
ln p (x) = pi (x) ln pi (x) , (2.171)
i=1
one may show that the probability density of X is given by
p (x) = Aexp [1 g1 (x) I g I (x)] . (2.172)
Note: The entropy definition (2.170) represents the measure of uncertainty,

or lack of information, for a continuous valued random variable.
Note: The conditions involved in finding the solution p (x) represent our
limited knowledge of possible constraints. If all the constraints were known
(and introduced into the problem), we would get exactly the same solution
as by conducting a deterministic mechanics study -by- according to the
right column of Figure 2.7there is no philosophical conflict between both
methods. Usually, however, the knowledge of all the constraints is not there,
or their introduction into the analysis is prohibitively complex. Consequently,
we view the MEM as an approach complementary to the one based on con-
ventional variational principles.
The MEM outlined in the left column of Figure 2.7 is ideally suited to
study the microscale material randomness, as expressed by the surface energy
density fluctuations. We proceed in the following steps:
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
1. Fracture occurs at the expense of dissipated energy E(), and the

fracture path has a length L(). The two quantities can be computed
from Figure 2.6.
2. Calculate surface energy per unit length () = E()/L(). Let
us note here that (and hence its probability density p( )) is now
assessed on the mesoscale l, which is the mean hole spacing, rather
than on, say, a molecular scale.
3. Given the results for N specimens, we get { (n ); 1 , . . ., N }, each
of probability
1
P{1 } = ... = P{ N } = . (2.173)
N

4. Compute moments
, 2 , 3 , ... to get, according to (2.172),

p( ) = Aexp 1
2 2 3 3 ... (2.174)
subject to

A exp 1 2 2 3 3 d = 1
0

A exp 1 2 2 3 3 d =
(2.175)
0

A 2 exp 1 2 2 3 3 d =
2 .
0
5. This basic model can be improved to account for:

Orientation of cracks with respect to the macroscopic loading
direction
Local crack (holehole) interactions
Further crack (holehole) interactions, etc.
6. With p( ) one can predict the probability of occurrence of any other
crack path.
Note: The MEM provides a closure method for many nonlinear problems
of stochastic mechanics, for example, fragmentation under dynamic impact
(Englman et al., 1987), turbulence (Frisch, 1995), random vibration (Sobczyk
and Trebicki, 1990, 1993), improved bounds on the effective response of ran-
dom materials (Kreher and Pompe, 1989), or contact forces in granular media
(Goddard, 2004). In Chapter 4 we return to the plate fragmentation experi-
ment with the dynamics taken into account.
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
Problems
1. Express the Kolmogorov consistency condition in terms of the prob-
ability density function.
2. Suppose that P and Q in Example 3 are jointly Gaussian with the
density N (1 , 2 , 1 , 2 , ). Compute the probability distribution
of X (, t).
3. Show the equivalence of (2.24) with (2.25), and prove (2.26). As-
suming P and Q are jointly normal random variables with the den-
sity N (1 , 2 , 1 , 2 , ), find the first-order probability density of
X (t, ); consult Problem 8 below.
4. Show that, in the case of a Rice noise, we have
n n
Y (x) = 0, VarY (x) = i=1 i2 , RY (x1 , x2 ) = i=1 i2 cos i (x2 x1 ) .
5. Make a literature search to define the Bernoulli process and discuss

its properties.
6. Make a literature search to define the binomial process and discuss
its properties.
7. Prove that |(x1 , x2 )| 1.
8. If we drop the dependence of f in (2.13) on x1 and x2 , then f (z1 , z2 ) is
the joint probability density of zero-mean normal random variables
z1 and z2 , commonly denoted N (1 , 2 , 1 , 2 , ), with 1 = 2 = 0
here. Show that is their correlation coefficient.
9. For a Rayleigh wave propagating along a weakly random boundary

(Section 2.2) derive zz and zz
1
.
10. One of the consequences of Jensens inequality is that, for any real-
valued function defined over the probability space,

exp f d P e f d P.

(a) Derive from this the inequality between the geometric and
arithmetic averages of n positive numbers y1 y2 . . . yn ,
1
( y1 y2 . . . yn ) 1/n ( y1 + y2 + + yn ) .
n
Hint: Consider to be a finite set with ( P{i }) = p for all i.
(b) Derive the inequality between the harmonic and geometric
averages

1
1 1 1 1
+ + + ( y1 y2 . . . yn ) 1/n
n y1 y2 yn
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
11. Generalize (2.71) to Timoshenko beams.

12. State the Kolmogorov consistency conditions for a vector random
field.
13. Consider pi to be a potential random vector field (p = S), where
S is a scalar random field with a zero mean and correlation function
C (S) (x), x = x2 x1 . Show that C (S) (x) is related to the correlation
( p)
function Cij (x) of p through
2 C (S) ( p)
= Cij (x)
x1i x2 j
where x1i (x2 j ) denotes a differentiation with respect to the ith

component of x1 ( jth component of x2 ).
14. Following Eimer (1971), discuss the effective constitutive equation
approximations of a two-phase, random material (with one phase
elastic and another viscous) as KelvinVoigt or Maxwell models.
15. Obtain the Reynolds stress by analyzing either the NavierStokes
or Euler equations in the same fashion as in Section 2.4.
16. Prove the statement of equation (2.151).
17. Prove Liouvilles equation (2.157) by taking the continuity equation
for in the phase space, and noting the divergence-free charac-
ter of the velocity field q, p in phase space (recall Hamiltons
equations).
18. Show that the EulerLagrange equations remain valid for gyro-
scopic systems in which the gyroscopic forces G k are derivable from
a potential V according to

d V V
Gk = V q k, q k = f (q k ) q k ,
dt q k q k
and the Lagrangian is taken in the form L = T V V instead of

the classical L = T V.
19. Give an example of a WSS process that is not ergodic.
20. Give an example of an ergodic flow that is not mixing.
21. Consider a collection of particles having a stationary distribution in
a conservative field, whose potential is V (x). Using the MEM, find
the probability density f (x) of positions x, under the assumption
that the mean of V with respect to f is known.
22. Prove properties 14 of (X1 , X2 ) listed in Section 2.3.1.1
23. Consider a collection of particles having a stationary distribution in
a conservative field, whose potential is V(x). Using the MEM, it is
a simple matter to find the probability density P(x) of positions x,
under the assumption that the function g(x) = V(x) and the mean

V of V with respect to p are known.
P1: Binod
July 5, 2007 16:5 C4174 C4174C002
24. Verify (2.134) and (2.135).

25. Check whether the random process of equation (2.24) is mean-
ergodic.
26. The one degree-of-freedom oscillator under random loading pro-
vides a basic model for equation (0.3) of the Preface. Considering the
equation of motion derive the oscillators stationary response un-
der a random forcing whose spectral density is (i.e., a white noise).
Hint: transform the problem to the frequency domain.
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
3
Planar Lattice Models: Periodic Topologies
and Elastostatics
Elasticity, thanks to solid state physics, has been reduced from

primordial and precise to secondary, approximate and derived.
J.A. Wheeler, 1994
Spring network (or lattice) models are based, in principle, on the atomic lattice
models of materials. These models work best when the material may naturally
be represented by a system of discrete units interacting via springs, or, more
generally, rheological elements. It is not surprising that spatial trusses and
frameworks have been the primary material systems thus modeled. Spring
networks can also be used to model continuum systems by a lattice much
coarser than the true atomic one. In engineering mechanics that idea dates
back, at least, to Hrennikoff (1941), if not to Maxwell (1869), in a special setting
of optimal trusses. This obviates the need to work with an enormously large
number of degrees of freedom that would be required in a true lattice model,
and allows a very modest number of nodes per single heterogeneity (e.g., in-
clusion in a composite, or grain in a polycrystal). As a result, spring networks
are a relative of the much more widespread finite element method. In this
chapter we focus on basic concepts and applications of spring networks, in
particular to antiplane elasticity, planar classical elasticity, and planar nonclas-
sical elasticity. The chapter ends with an additional section on the mechanics
of a helix, a 1D nonclassical continuum.
3.1 One-Dimensional Lattices

3.1.1 Simple Lattice and Elastic Strings
Let us begin with a lattice-based derivation of a wave equation for a 1D chain
of particles; see also Askar (1985). The particles (parametrized by i), each of
mass m, interact via nearest-neighbor linear elastic interactions, Figure 3.1.
87
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
i1 i i+1
FIGURE 3.1
A 1 D chain of particles of lattice spacing s, connected by axial springs (thin lines).
For the potential and kinetic energies we have:
1 1 1
U= Fi (ui+1 ui ) = K (ui+1 ui ) 2 T= mui2 , (3.1)
2 i 2 i 2 i
where Fi = K (ui+1 ui ) is the axial force at i, and K is the spring constant

between i and i + 1. Using the EulerLagrange equations for the Lagrangian
L = T U, we arrive at the dynamical equations
K (ui1 2ui + ui+1 ) = mui , (3.2)
which describe a system of coupled oscillators.

By taking the Taylor expansion up to the second derivative for the dis-
placement ui1 u (xi s),
1
ui1
= u|xi u, x xi s + u, xx xi s 2 , (3.3)
2
we find from (3.2) an approximating continuum mechanics model, that is, the
basic wave equation
E Au, xx = Au , (3.4)
where (A being the cross-sectional area of the rod)
Ks m
E= = . (3.5)
A As
Of course, (3.4) can also be obtained from Hamiltons principle for the
Lagrangian L expressed in terms of continuum-like quantities, by first intro-
ducing (3.3) in (3.1)1 with terms up to the first derivative,

AE d 2 A d
U= u, x dx T= ( u) 2 d x . (3.6)
2 0 2 0
3.1.2 Micropolar Lattice and Elastic Beams

We now generalize the preceding lattice model to describe transverse motions
of a 1D chain of dumbbell particles (rigid bars) pin-connected by central-
force (axial) springs, Figure 3.2. The particles (again parametrized by i) each
have mass m and moment of inertia J . We need to consider two degrees of
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
Planar Lattice Models: Periodic Topologies and Elastostatics 89
i1 i i+1
(a) (b)
FIGURE 3.2
(a) A 1D chain of dumbbell particles (vertical rigid bars) of X-braced girder geometry, pin-
connected by axial springs (thin lines); (b) the shear and curvature modes of a single bay.
freedom per particle i: total transverse displacement wi and rotation i . The

constitutive laws for a single bay (between i and i + 1) are
i = K [wi+1 wi si+1 ]
F Mi = K (i+1 i ) . (3.7)
Here Fi is the shear force and Mi the bending moment at i, whereby the
term si+1 is subtracted in (3.7)1 so as to deal with shear only. For this 1D
chain of particles, we write down potential and kinetic energies
1 1
U= K [wi+1 wi si+1 ]2 + K (i+1 i ) 2 T= mwi2 + J i2 .
2 i 2 i
(3.8)
The first term in (3.8)1 accounts for shear deformations, and the second one for
bending. Using the EulerLagrange equations for the Lagrangian L = T U,
we obtain a system of equations
K [wi+1 2wi + wi1 s(i+1 i )] = mwi

(3.9)
K [i+1 2 + i1 ] + K [wi+1 wi1 si ] s = J i .
By introducing the Taylor expansions for wi and i in (3.9)1 with terms up

to the second derivative, and taking the limit s 0, we find
G A[w, x ], x = Aw
(3.10)
E I , xx + G A[w, x ] = I ,
where (A and I , respectively, being the area and second moment of the cross-
section of this beam-like lattice):
Ks m JA K sm
G= = I = E= . (3.11)
A As m JA
Relations (3.10) are recognized as the equations of a Timoshenko beam.
Evidently, this is a 1D micropolar continuum with two degrees of freedom:
displacement w and rotation .
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
As in Section 3.1.1, (3.10) could alternatively be obtained by first introduc-

ing the Taylor series with terms up to the first derivative into (3.8)1 , to first
get

1 d 2 2 1 d

U= G A w , x + E I , x dx T= A ( w) 2 + I ( ) 2 d x ,
2 0 2 0
(3.12)
and then, by employing the Hamilton principle.

A question arises here: Can other, more complex (micro)structures, espe-
cially those made of little beams connected by rigid joints, of a general beam-
like geometrysuch as shown in Figure 3.3be sufficiently well described
by this beam model? The general answer is: No (Noor and Nemeth, 1980).
The basic procedure, however, recommended by those authors is basically
the same as that outlined here:
h x
L L L
(a) Vierendeel girder.
(b) Pratt girder.
(c) X-braced girder.
h x
L L
(d) Warren girder.
FIGURE 3.3
Planar lattices and their repeating elements. (From Noor and Nemeth, 1980. With permission.)
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
1. An equivalent micropolar beam model is set up from the postulate

of the same strain and kinetic energies stored in the original lattice
when both are deformed identically.
2. A typical repeating element is identified and the energies for this
element are expressed in terms of the nodal displacements, joint
rotations, as well as the geometric and material properties of the
individual members;
3. A passage to an effective continuum is carried out via a Taylor
expansion, whereby it turns out that higher-order terms show up
in the governing continuum equations, depending on the actual
microgeometry of the rods making up the structure; see also
Triantafyllidis and Bardenhagen (1993).
It is appropriate to note here that beam-like lattices can also be analyzed

by a cell transfer matrix approachthe eigenvalues of this matrix are the
decay rates relevant in Saint-Venants principle for these discrete, rather than
continuum, systems (Stephen and Wang, 1996). The associated eigenvectors
and principal vectors lead then to equivalent continuum-beam properties.
Finally, we note that continuum approximations of plate-like structures
were also investigated (Noor, 1988). In that review, among the problems re-
quiring new investigations was listed the effect of microstructural material
randomnesswe address this topic, in the context of beam vibrations, in
Chapter 11. That chapter also outlines the transfer matrix approach in dy-
namic problems of periodic, possibly disordered, structures.
3.2 Planar Lattices: Classical Continua

3.2.1 Basic Idea of a Spring Network Representation
As already demonstrated in the setting of 1D models, the basic idea in setting
up 2D and 3D spring network models is based on the equivalence of energies
stored in the unit cell (Figure 3.4) of a given network. In the case of static
problems, to which we will restrict the discussion henceforth, for a cell of
volume V we therefore have
Ucell = Ucontinuum . (3.13)
The unit cell is a periodically repeating part of the network. Two aspects
should be noted here: (1) the choice of the unit cell may be nonunique, see
Figure 3.4(a); and (2) the inner structure of the unit cell is not necessarily
nicely orderedit may be of a disordered microgeometry, with an under-
standing that it repeats itself in space such as the periodic PoissonDelaunay
network (Chapter 4).
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
(a)
x2
2
x1
3 1
4
(b)
(c)
FIGURE 3.4
Three periodic planar lattices: honeycomb, square, and triangular. In each case a possible (not
necessarily unique) periodic unit cell is shown.
The energies of the cell and its equivalent continuum, respectively, are

1 b N
1
Ucell = Eb = (F u) (b) Ucontinuum = : d V. (3.14)
b
2 b 2 V
The superscript b in (3.14)1 stands for the bth spring (bond), and Nb for the
total number of bonds. Our discussion is set in the 2D setting so that, by a
volume we actually mean an area of unit thickness. In the sequel we restrict
ourselves to linear elastic springs and spatially linear displacement fields u
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
(i.e., uniform strain fields ), and (3.14) will become

|b|
1 V
Ucell = (ku u) (b) Ucontinuum = : C : . (3.15)
2 b 2
In (3.15) u is a generalized spring displacement and k its corresponding

spring constant. The next step, depending on the particular geometry of the
unit element and on the particular model of interactions, will involve making
a connection between u and , and then deriving C from (3.13). The cor-
responding procedures and resulting formulas are given below for several
elasticity problems set in the square and triangular network geometries.
3.2.2 Antiplane Elasticity on Square Lattice

Of all the elasticity problems, the antiplane is the simplest one on which to
illustrate the spring network idea. In the continuum setting we thus have the
constitutive law
i = Cij j i, j = 1, 2, (3.16)
where = (1 , 2 ) (31 , 32 ), = (1 , 2 ) (31 , 32 ) and Cij C3i3 j . Upon

the substitution of (3.16) into the equilibrium equation
i,i = 0, (3.17)
we obtain

Cij u, j ,i = 0. (3.18)
Henceforth, we are interested in approximations of locally homogeneous

media, so that this governing equation becomes
Cij u,i j = 0. (3.19)
In the special case of an isotropic medium, (3.19) simplifies to the Laplace

equation
Cu,ii = 0. (3.20)
We now discretize the material with a square lattice network, Figure 3.4(b),
whereby each node has one degree of freedom (antiplane displacement u),
and the nearest neighbor nodes are connected by springs of constant k. It
follows that the strain energy of the unit cell of such a lattice is
1 (b) (b)
4
U= k l l i j . (3.21)
2 b=1 i j
In the above we employed the uniform strain = (1 , 2 ) . Also, l(b) =

(l1(b) , l2(b) )
is the vector of half-length of bond b. In view of (3.13), the stiffness
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
tensor is obtained as
k (b) (b)
4
Cij = l l i, j = 1, 2 , (3.22)
V b=1 i j
where V = 4 if all the bonds are of unit length (|l(b) | = 1). This leads to a
relation between the bond spring constant k and the Cij tensor
k
C11 = C22 = C12 = C21 = 0. (3.23)
2
In order to model an orthotropic medium, different bonds are applied in
the x1 and x2 directions: k (1) and k (2) . The strain energy of the unit cell is now
1 (b) (b) (b)

4
U= k li l j i j , (3.24)
2 b=1
so that the stiffness tensor is
1 (b) (b) (b)

4
Cij = k li l j , (3.25)
V b=1
which leads to relations

k (1) k (2)
C11 = C22 = C12 = C21 = 0. (3.26)
2 2
If one wants to model an anisotropic medium (i.e., with C12 = 0), one may
either choose to rotate its principal axes to coincide with those of the square
lattice and use the network model just described, or introduce diagonal bonds.
In the latter case, the unit cell energy is given by the formula (3.24) with Nb = 8.
The expressions for Cij s are
k (1) k (2)
C11 = + k (5) C22 = + k (6) C12 = C21 = k (5) k (6) . (3.27)
2 2
It will become clear in the next section how this model can be modified to a
triangular spring network geometry.
3.2.3 In-Plane Elasticity: Triangular Lattice with Central Interactions

In the planar continuum setting (see Chapter 5), Hookes law
ij = Cijkm km i, j, k, m = 1, 2, (3.28)
upon substitution into the balance law
ij,j = 0, (3.29)
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
results in a planar Navier equation for the displacement ui
ui, j j + u j, ji = 0. (3.30)
In (3.30) is defined by 12 = 12 , which makes it the same as the classical

3D shear modulus. On the other hand, is the (planar) 2D bulk modulus, that
is defined by ii = ii .
As in the foregoing section, we are interested in approximations of locally
homogeneous media. Consider the regular triangular network of Figure 3.4(c)
with central force interactions, which are described, for each bond b, by
Fi = ij(b) u j where ij(b) u j = (b) ni(b) n(b)

j . (3.31)
Similar to the case of antiplane elasticity, (b) is the spring constant of half-
lengths of such central (normal) interactionsthat is, of those parts of the
springs that fall within the given unit cell, Figure 3.5(a). The unit vectors n(b)
at respective angles of the first three -springs are
(1) = 00 n(1)
1 =1 n(1)
2 =0

1 3
(2) = 600 n(2)
1 = n(2)
2 = (3.32)
2 2
1 3
(3) = 1200 n(3)
1 = n(3)
2 = .
2 2
The other three springs (b = 4, 5, 6) must, by the requirement of symmetry
with respect to the center of the unit cell, have the same properties as b =
1, 2, 3, respectively. All the -springs are of length
l, that is, the spacing of the
triangular mesh is s = 2l. The cell area is V = 2 3l 2 .
Every node has two degrees of freedom, and it follows that the strain
energy of a unit hexagonal cell of this lattice, under conditions of uniform
x2
(3) (2)
3 2 b+1 b
(2)
(b + 1)
(3) (1)
(4)4 1 (1)
(4) (6)
(5)
5 6 (b)
x1
(5) (6)
(a) (b)
FIGURE 3.5
Unit cell of a triangular lattice model; (1) , . . ., (6) are the normal spring constants; (1) , . . ., (6)
are the angular spring constants; in the isotropic Kirkwood model (b) = (b+3) and (b) = (b+3) ,
b = 1, 2, 3. (b) Details of the angular spring model.
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
strain = (11 , 22 , 12 ), is
l 2 (b) (b) (b) (b) (b)

6
U= ni n j nk nm ij km , (3.33)
2 b=1
so that, again by (3.13), the stiffness tensor becomes
l 2 (b) (b) (b) (b) (b)

6
Cijkm = ni n j nk nm . (3.34)
V b=1
In particular, taking all (b) the same, we see that

9 3 3
C1111 = C2222 = C1122 = C2211 = C1212 = , (3.35)
8 3 8 3 8 3
so that there is only one independent elastic modulus, and the modeled con-
tinuum is isotropic.
It is important to note here that the isotropy follows from the triangular
lattice having an axis of symmetry of the sixth order. This, combined with the
fact that (3.34) satisfies the conditions of Cauchy symmetry (Love, 1934) with
respect to the permutations of all the four indices (which is the last equality
here)
Cijkm = Ci jmk = C jikm = Ckmi j = Cik jm , (3.36)
implies that Cijkm is of the form

Cijkm = ij km + ik jm + im jk . (3.37)
In view of (3.35), we obtain the classical Lame constants

3
= = . (3.38)
8 3
The above is a paradigm from the crystal lattice theory that the Cauchy
symmetry occurs when:
1. The interaction forces between the atoms (or molecules) of a crystal

are of a central force type.
2. Each atom (or molecule) is a center of symmetry.
3. The interaction potential in a crystal can be approximated by a har-
monic one.
Note: The Cauchy symmetry reduces the number of independent con-

stants in general 3D anisotropy from 21 to 15. The first case has been called
the multi-constant theory, and the second one the rari-constant theory. Basically,
there is a decomposition of the stiffness tensor into two irreducible parts with
15 and 6 independent components, respectively; see Hehl and Itin (2002) for
a group-theoretical study of these issues.
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
Note: One might try to model anisotropy by considering three different s

in (3.33 and 3.34), but such an approach would be limited given the fact that
only three of those can be varied: one needs to have six parameters in order
to freely adjust any planar anisotropy that involves six independent Cijkm
values. This can be achieved by introducing the additional angular springs,
as discussed below. In fact, angular springs are also the device to vary the
Poisson ratio.
3.2.4 In-Plane Elasticity: Triangular Lattice with Central

and Angular Interactions
We continue with the triangular network, and introduce angular springs act-
ing between the contiguous bonds incident onto the same node. These are
assigned spring constants (b) , and, again by the argument of symmetry with
respect to the center of the unit cell, only three of those can be independent.
Thus, we arrive at six spring constants: { (1) , (2) , (3) , (1) , (2) , (3) }. With ref-
erence to Figure 3.5(b), let (b) be the (infinitesimal) angle change of the bth
spring orientation from the undeformed position. Noting that n n = l,
we obtain
(b) = e ki j j p ni n p , (3.39)
where e ki j is the LeviCivita permutation tensor. The angle change between

two contiguous springs (b and b + 1) is measured by = (b+1) (b) ,
so that the energy stored in the spring (b) is
1 (b) 1 2
E (b) = ||2 = (b) ki j j p ni(b+1) n(b+1)
p n(b) (b)
i n p . (3.40)
2 2
By superposing the energies of all the angular bonds with the energy (3.33),
the elastic moduli are derived as (Grah et al., 1996):
l 2 (b) (b) (b) (b) (b) 1

(b)
6 6
(b) (b) (b)
Cijkm = ni n j nk nm + + (b1) ik n(b)
p n j n p nm
V b=1 V b=1

(b) + (b1) ni(b) n(b) (b) (b) (b) (b+1) (b+1) (b)
j nk nm ik n p n j
(b)
n p nm
+ (b) ni(b) n(b+1)

j n(b+1)
k
(b)
nm (b) ik n(b) (b) (b+1) (b+1)
p n j np nm

+ (b) ni(b+1) n(b) (b) (b+1)
j nk nm , (3.41)
where b = 0 is the same as b = 6.

This provides the basis for a spring network representation of an ani-
sotropic material; it also forms a generalization of the so-called Kirkwood model
(Kirkwood, 1939) of an isotropic material. The latter is obtained by assigning
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
the same to all the normal and the same to all the angular springs:
(b) (b) (b) (b)

6 6
Cijkm = ni n j nk nm + {2ik n(b)
j nm
(b)
2 3 b=1 2
2 3l b=1
2ni(b) n(b) (b) (b) (b) (b+1) (b+1) (b)

j nk nm ik n p n j n p nm + ni(b) n(b+1)
j n(b+1)
k
(b)
nm
(b) (b+1) (b+1)
ik n(b)
p n j np nm + ni(b+1) n(b) (b) (b+1)
j nk nm }. (3.42)
In accordance with the above,

1
9 19 2
C1111 = C2222 = + 2
4
2 3 l 4

1 3 19
C1122 = C2211 = 2 (3.43)
2 3 4 l 4

1 3 19
C1212 = + 2 .
2 3 4 l 4
Condition C1212 = (C1111 C1122 )/2 is satisfied, so that there are only two
independent elastic moduli.
From equation (3.43), the and constants are related to the planar bulk
and shear moduli by

1 3 1 3 19
= = 2 . (3.44)
2 3 2 2 3 4 l 4
It is noted here that the angular springs have no effect on , that is, the presence
of angular springs does not affect the dilatational response. The formula for
planar Poissons ratio (Chapter 5) gives
C1111 2C1212 1 3/l 2

= = = . (3.45)
+ C1111 3 + 3/l 2
From (3.45) there follows the full range of Poissons ratio that can be covered
with this model. It has two limiting cases
1
= if 0, model
3
(3.46)

= 1 if , model.

For the subrange of Poissons ratio between 1/3 and 1/3, one may also use
a Keating model (1966), which employs a different calculation of the energy
stored in angular bonds.
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
A C B

B
C A

y
x
B A C
(a) (b)
FIGURE 3.6
(a) A triple honeycomb lattice made of three different spring types , , and belonging, respec-
tively; to three sublattices A, B, and C; (b) a 42 42 unit cell of a triangular lattice of hexagonal
pixels, with 11 pixel diameter circular inclusions centered on pixels and randomly placed with
periodic boundary conditions. (From Snyder et al. (1992). With permission.)
3.2.5 Triple Honeycomb Lattice

It is recalled from Section 3.2.4 that 1/3 is the highest Poissons ratio of central-
force triangular lattices with one spring constant. An interesting model per-
mitting higher values, from 1/3 up to 1, was introduced by Day et al. (1992)
and Snyder et al. (1992). The model sets up three honeycomb lattices, having
spring constants , , and , respectively, overlapping in such a way that they
form a single triangular lattice, Figure 3.6. The planar bulk and shear moduli
of a single phase are
1
1 27 1 1 1
= ( + + ) = + + . (3.47)
12 16
In the case of two (or more) phases, a spring that crosses a boundary between
any two phases 1 and 2 is assigned a spring constant according to a series
rule = [(21 ) 1 + (22 ) 1 ], where i , i = 1, 2, (i.e., , , or ), is a spring
constant of the respective phase.
Note: Although this chapter is focused on planar lattice models of elastic
solids, there also exist extensions of the lattice approach to 3D and inelastic
materials, e.g., Buxton et al. (2001).
3.3 Applications in Mechanics of Composites

3.3.1 Representation by a Fine Mesh
As discussed in Section 3.2.2, one may employ the square mesh of Figure 3.4(b)
in the x1 , x2 -plane for discretization of an antiplane elasticity problem. Indeed,
this approach may be applied to model multiphase composites treated as
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
C (i)/C (m) Rigid disks
Sti disks Sti ellipses Sti needles
1.0 Homogeneous
medium
Soft disks
Holes Soft ellipses Soft needles Cracks

0.0
(a)
(b)
C (i)/C (m) Rigid inclusions
Sti pixels Sti ellipses Sti needles
Homogeneous
1.0
medium
Soft pixels
Holes Soft ellipses Soft needles Cracks

0.0
(c)
FIGURE 3.7
(a) Parameter plane: aspect ratio of inclusions and the contrast; (b) spring network as a basis
for resolution of round disks, ellipses, pixels, and needles in the parameter plane; (c) another
interpretation of the parameter plane: from pixels to needles.
planar, piecewise-constant continua, providing a lattice or mesh (very) much

finer than a single inclusion is involved, Figure 3.7(b). How much finer it
should actually be needs to be assessed on a reference problem. The governing
equations for the displacement field u u3 are
u (i, j) [kr + kl + ku + kd ] u (i + 1, j) kr u(i 1, j)kl

u(i, j + 1)ku u(i, j 1)kd = f (i, j), (3.48)
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
where f (i, j) is the body force (or source) at node (i, j), while i and j are the
coordinates of mesh points, and kr (right), kl (left), ku (up), and kd (down) are
defined from the series spring model
1
1 1
kr = +
C(i, j) C(i + 1, j)
1
1 1
kl = +
C(i, j) C(i 1, j)
1 (3.49)
1 1
ku = +
C(i, j) C(i, j + 1)
1
1 1
kd = + .
C(i, j) C(i, j 1)
In (3.49), C(i, j) is the material property at node (i, j).

This type of a discretization is equivalent to a finite difference method that
would be derived by considering the expansions

u(i, j) s u(i, j)
2 2
u (i 1, j) = u (i. j) s +
x1 i, j 2! x12
i, j
(3.50)
u(i, j) s u(i, j)
2 2
u (i, j 1) = u (i. j) s +
x2 i, j 2! x22
i, j
in the governing equation (recall [3.20])

2 u 2 u
C + 2 = 0. (3.51)
x12 x2
However, in the case of in-plane elasticity problems, the spring network

approach is not identical to the finite difference method, because the node
node connections of a spring network do really have a meaning of springs,
whereas the finite difference connections do not.
In the case of a composite made of two locally isotropic phases: matrix (m)
and inclusions (i), antiplane Hookes law is
i = Cij j i, j = 1, 2 Cij = C (m) ij or C (i) ij . (3.52)
The above leads to a so-called contrast (or mismatch) C (i) /C (m) . It is clear that
by increasing the contrast we can approximately model materials with rigid
inclusions. Similarly, by decreasing the contrast, we go to very soft inclusions
and nearly reach a system with holes.
Although the disk is the most basic inclusion shape when dealing with
composites, a departure from this is of interest. Thus, another basic parameter
specifying the composite is the aspect ratio of ellipses a /b, where a respectively,
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
(b) is the ellipses major (minor) semi-axis. By varying the aspect ratio from
1 up through higher values we can model systems having disk-type, ellipse-
type, through needle-type inclusions. We are thus led to the concept of a
parameter plane shown in Figure 3.7(a).
Resolution of several different types of inclusions by the spring network
is shown in Figure 3.7(b). That is, we can model disks, ellipses, needles, etc.
Admittedly, this type of modeling is approximate so that a somewhat different
interpretation of a parameter plane is given in Figure 3.7(c). It is seen that
disks may most simply be modeled as single pixels or more accurately as
finite regions; in the latter case arbitrary anisotropies can be modeled. The
former case allows one to deal with very large scale systems, while the latter
allows a much better resolution of local stress/strain fields within and around
inclusions. By decreasing the spring network mesh size, an increasingly better
accuracy can be achieved. Depending on the shape functions employed in
finite element models, somewhat more accurate results may be obtained, but
this comes at a higher price of costly and cumbersome remeshing for each
and every new configuration B () from the ensemble B, which is required
in statistical (Monte Carlo) studies.
It is noteworthy that, in contradistinction to the finite element method, no
need for remeshing and constructing of a stiffness matrix exists in our spring
network method: spring constants are very easily assigned throughout the
mesh, and the conjugate gradient method finds the solution of the equilibrium
displacement field u (i, j). In that manner, a system having 106 million degrees
of freedom (1000 1000 nodes) can readily be handled on a computer with
90 MB of random access memory. For 2000 2000 nodes one requires some
360 MB, and so on, because of the linear scaling of memory requirements with
the number of degrees of freedom.
The quality of approximation of ellipses and needle-type cracks/inclusions
can be varied according to the number of nodes chosen to represent such ob-
jects. Local fields cannot be perfectly resolved, but the solution by the spring
network is sufficient to rapidly establish the elastic moduli of a number of dif-
ferent B () realizations from the random medium B, and the corresponding
statistics with a sufficient accuracy. Spring networks are used in later chapters
to study scaling laws of various planar composites.
Note: Interestingly, the computational method for determining effective
moduli of composite materials with circular inclusions due to Bird and Steele
(1992) would be very well suited for analysis of this type of stationarity and
isotropy.
3.3.2 Solutions of Linear Algebraic Problems

The steady-state conductivity and elastostatics problems on spring networks
always lead to linear algebraic systems
A x = b, (3.53)
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
because they simply are elliptic problems in discretized forms. There are, in
principle, two methods to set up and solve the governing equations. One of
them is exactly the same as that conventionally used in the finite element
methodsinvolving the global stiffness matrix accompanied by the connec-
tivity of all the nodesand will therefore not be elaborated here. The second
one comes, just like the spring networks themselves, from the condensed mat-
ter physics. It is the so-called conjugate gradient method, which involves the
energy of the system as a functional
1
F(x) = xAxbx (3.54)
2
of all the relevant degrees of freedom x, and the gradient of this energy
F(x) = A x b (3.55)
with respect to all these degrees. Once written in an explicit form as two
subroutines, the program is connected with any of the widely available solvers
(e.g., Press et al., 1992). Note that F(x) is minimized when equation (3.55)
equals zero, which is then equivalent to (3.53). Of course, one may also employ
other algebraic solvers.
It is noteworthy that the entire task of mesh generationsuch as typically
required by the finite element methodsis absent. The energy and energy
gradient subroutines are written once and for all for the given mesh of, say,
Figure 3.8. The assignment of all the local spring stiffnessesaccording to
FIGURE 3.8
A functionally graded matrix-inclusion composite with 47.2% volume fraction of black phase is
partitioned into subdomains, corresponding to a 64-processor parallel computer.
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
any chosen lattice model of Section 3.2is done very rapidly in the first stage
of the program. These stiffnesses are stored in the common block (in case
of a Fortran program) and are readily accessible to the conjugate gradient
subroutines that are activated in the second, and main, stage of the program.
Once the energy minimum is reached to within any specified accuracy, this
energy is used to compute the overall, effective moduli of a given domain of
the lattice based on the postulate of the energy equivalence.
Here we list several exact relations that may be used in testing the resulting
computer programs, homogenization methods, or effective medium theories;
see also Garboczi (1998).
1. Suppose we have solutions of two elasticity problems on a certain
domain B, with boundary B, corresponding to the displacement
(d) and traction (t) boundary value problems, respectively. Then we
can check whether the BettiMaxwell reciprocity theorem

(t) (d)
ui ti ds = ui(d) ti(t) ds (3.56)
B B
is satisfied numerically within some acceptable accuracy.
2. Perfect series and parallel systems are well known to result in the
arithmetic and harmonic averages, or the so-called Voigt (C V ) and
Reuss (C R ) bounds

f1 f 2 1
C = f 1 C1 + f 2 C2 C =
V R
+ , (3.57)
C1 C2
where f 1 and f 2 are the volume fractions of phases and 1 and 2,
respectively.
3. The case of small contrast in properties allows an expansion of, say,
effective conductivity to second order in the difference (C1 C2 ) as
follows (Torquato, 1997):
(C2 C1 ) 1
C eff = C1 + f 2 (C2 C1 ) f 1 f 2 + O(C2 C1 ) 3 + ,
C1 D
(3.58)
where D is the dimensionality of the space.

4. There are many exact relations in the 2D conductivity (Milton, 2002).
Perhaps the most well-known one, due to Keller (1964) (also Mendel-
son, 1975) says that, for a two-phase isotropic system in 2D,
C eff (C1 , C2 )C eff (C2 , C1 ) = C1 C2 , (3.59)
where C eff (C1 , C2 ) is the effective conductivity of a given system,

while C eff (C2 , C1 ) is the effective conductivity with the phases 1
and 2 interchanged.
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
5. The so-called CLM result and stress-invariance discussed in

Chapters 5 and 6 may be employed for planar elastic (classical as
well as micropolar) composites with either twice-differentiable or
piecewise-constant properties. In essence, the effective compliance
tensor of the transformed material is given by that of the origi-
nal material plus the same transformation (shift) in bulk and shear
compliances as that for the individual phases; see Section 5.2.4 in
Chapter 5.
When dealing with very large systems, the spring network method is
limited by the available computer memory size. This, for example, is the
case with a functionally graded material. A composite of that type is shown
in Figure 3.8, where the diskmatrix interphase is taken as a finite thickness
zone of two randomly mixed phases of the disk (2) and the matrix (1) material.
Both phases are locally homogeneous and isotropicthey are described by
two constant isotropic conductivities C1 and C2 . We see here three different
length-scales: the fine structure of the interphase region, the size and spacing
of inclusions, and the macroscopic dimension of the composite.
For this type of problem we can also use parallel computing. Thus, in
Figure 3.8 we show a partition of the entire simulated domain of a function-
ally graded composite into 64 = 8 8 subdomains, each of which represents
a spring network that is assigned to a separate processor of a parallel com-
puter. Thus, the boundary value is solved by using 64 processors operating in
tandem. The computational effort is limited by the speed of a single processor
(which goes down with the subdomain size) and the communication between
the processors (which simultaneously goes up); the latter leads to so-called
bottlenecks. Determining an optimal partition is therefore an important task
and remains a major challenge today as we move to computers having thou-
sands and tens of thousands of processors. There are, in principle, two ways
to execute such a parallel scheme: either to write ones own software, or to
adapt an existing solver running on a given parallel computer.
3.3.3 Example: Simulation of a Polycrystal

The generalization of the Kirkwood spring network model outlined in Section
3.2.4 to an anisotropic case was motivated by a need to study micromechanics
of a planar polycrystalline aluminum specimen (Grah et al., 1996). The basic
strategy is as follows. First, an image of crystal domains (i.e., grains), such
as the one showed in Figure 3.9, needs to be scanned and mapped onto a
triangular mesh. Next, every bond is assigned its stiffness depending on the
domain it falls in. And finally, the mechanics problem of the resulting spring
network is solved computationally.
In order to assign spring stiffnesses to any node of the spring network
mesh, the 3D stiffness tensor Ci jkm for each crystal must be found according
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
FIGURE 3.9
Scanned image of a very thin polycrystal aluminum sheet. All the grain boundaries are orthogonal
to the plane of the sheet. (From Grah et al., 1996. With permission.)
to its transformation (rotation) matrix a ij (i, j = 1, 2, 3); the latter is provided

from the Kikuchi surface electron backscattering technique. Thus, to set up the
spring network model, we start from the stiffness matrix of an (anisotropic)
aluminum crystal which is given as

10, 82 6, 13 6, 13 0 0 0
6, 13 10, 82 0
6, 13 0 0

6, 13 6, 13 10, 82 0 0 0
C =
0
104 MPa . (3.60)
0 0 2, 85 0 0

0 0 0 0 2, 85 0
0 0 0 0 0 2, 85
Its corresponding fourth-rank stiffness tensor Cijkm is then set up taking into
account three symmetries Cijkm = Ci jmk = C jikm = Ckmi j . We next use a
transformation formula for a 4th-rank tensor
Cnpqr = a ni a j p a kp a mr Cijkm n, p, q , r = 1, 2, 3 (3.61)
to set up the in-plane (2D) part of Cnpqr at every mesh node. This in-plane

part, consisting of C1111 , C2222 , C1112 , C2212 , C1212 is then mapped into the six
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
spring constants 1 , 2 , 3 , 1 , 2 , 3 according to

C1111 1 1 1 3
3 3
1
4 4

3 16 3 16 3

C2222 9 9 3 3
0 3 4 2

16 3 16 3 4
C1122

3
0 3 3
49 3 3
43

=
16 16 2
.
(3.62)
C1112 1
161 3 3 1
4

0 0

16
4
C 9 2
1212 0 3 3 9
0
16 16 4 3 4 3

C2212 0 3
163
34 0 3 3
16 4
While there is no ambiguity concerning the spring constant of any a-bond

that entirely belonged to any given crystal domain, special care has to be
taken of the bonds that straddle the boundaries of contiguous crystals. The
effective stiffnesses are assigned according to a series rule: a = ( 12 1 + 12 2 ) 1 .
Assignment of the b-springs presents no such ambiguities.
A general finite element mapping procedure for defining spring network
representations in 2D and 3D for isotropic and anisotropic solids has recently
been developed by Gusev (2004).
3.4 Planar Lattices: Nonclassical Continua

3.4.1 Triangular Lattice of BernoulliEuler Beams
In the solid state physics literature the Kirkwood and Keating models are
sometimes referred to as the beam-bending models. This is a misnomer be-
cause there is no account taken in these models of the actual presence of
moments and curvature change of spring bonds connecting the neighboring
nodes. True beam bending was fully and rigorously considered by Wozniak
(1970) and his coworkers, and, considering a limited access to that book, in
this section we give a very brief account of the triangular lattice case. For more
modern developments see Trovalusci and Masiani (1999).
We focus on the deformations of a typical beam, its bending into a curved
arch allowing the definition of its curvature, and a cut in a free body diagram
specifying the normal force F , the shear force F , and the bending moment M,
see Figure 3.10. It follows that, in 2D, the force field within the beam network
is described by fields of force-stresses kl and moment-stresses mk , so that we
have a micropolar medium, see also Chapter 6.
The kinematics of the network of beams is now described by three
functions:
u1 (x) u2 (x) (x), (3.63)

P1: Binod
July 12, 2007 19:23 C4174 C4174C003
~
F (f) F (f)
M (f)
n( f )
~

(a) (b)
FIGURE 3.10
The lattice geometry (a); curvature and internal loads in a single beam element (b).
which coincide with the actual displacements (u1 , u2 ) and rotations at the
network nodes. Within each triangular pore, these functions may be assumed
to be linear, and hence, the local strain kl , and curvature, i , fields are related
to u1 , u2 , and by
kl = ul,k + e lk i = ,i , (3.64)
where e lk is the Ricci symbol.

It follows from geometric considerations that
(b) nk(b) nl(b) kl (3.65)
is the average axial strain, with s (b) (b) being its average axial length change.
Similarly,
(b) nk(b) nl(b) kl = nk(b) nl(b) u(l, k) (3.66)
is the difference between the rotation angle of the beam chord and the rotation
angle of its end node. Thus, the difference between the rotation angles of its
ends is
(b) nk(b) k . (3.67)
The elementary beam theory suggests that the force-displacement and

moment-rotation response laws of each bond (Figure 3.10) are given as
12E (b) I (b) (b)
F (b) = E (b) A(h) (b) F (b) =
M(b) = E (b) I (b) (b) , (3.68)
s2
where A(b) = wh is the beam cross-sectional area, I (b) = 12
1
w 3 h is its centroidal
moment of inertia with respect to an axis normal to the plane of the network,
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
and E (b) is the Young modulus of the beams material. All the beams are of
length s s (b) , which is the spacing of the mesh.
Turning now to the continuum picture, the strain energy of the micropolar
continuum is expressed as (note Chapter 6):
V (1) V
Ucontinuum = ij Cijkm km + i Cij(2) j , (3.69)
2 2
Here Cij(2) stands for the tensor Ci3

(2)
j3 . From the above we find

6
6
(1)
Cijkm = ni(b) nk(b) n(b)
j n(b) (b)
m R + n(b) (b) (b)
j nm R Cij(2) = ni(b) n(b) (b)
j S ,
b=1 b=1
(3.70)
where
2E (b) A(b) 24E (b) I (b) 2E (b) I (b)
R(b) = R(b) = 3 S(b) = . (3.71)
s (b) 3 s (b) 3 s (b) 3
If we assume all the beams to be the same (R(b) = R, etc.), we obtain

(1) (1) (1) (1)
C1111 = C2222 = 38 3R + R C1212 = C2121 = 38 R + 3 R
(1) (1) (1) (1)
C1122 = C2211 = 38 R R C1221 = C2112 = 38 R R (3.72)
(2) (2)
C11 = C22 = 32 S,
with all the other components of the stiffness tensors being zero. That is, we
have
(1)
Cijkm = ij km + ik jm + im jk Cij(2) = ij , (3.73)
in which
3 3 3
== R R = R + 3 R = S. (3.74)
8 8 2
We note from (3.58)1 that this beam lattice is an isotropic continuum.
The micropolar model (refer to Chapter 6) is conveniently expressed in
terms of four compliances
1 2 2 2
A= S= P= M= . (3.75)
+ +
2
+ 3
The effective bulk and shear moduli are now identified on the basis of (3.73
and 3.74) as
3 3
= R = R + R , (3.76)
4 8
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
which are seen to reduce to the formulas of Section 3.2.3 in the special case
of flexural rigidity being absent. Furthermore, the effective Young modulus
and Poisson ratio are
I + R/R 1 R/R
E = 3R = . (3.77)
3 + R/R 3 + R/R
It is noteworthy that the introduction of beam-type effects has a similar in-

fluence on E and as the introduction of the angular -interactions in the
Kirkwood model. However, noting that R/R = (w/s) 2 , where w is the beam
width, we see that, in view of the slenderness assumption of the beam ele-
ments (neither too thin, nor too thick), this model admits between 1/3
and 0.2. With reference to Section 6.1 of Chapter 6, interpreting this lattice
in the plane stress formulation does not change , whereas adopting the plane
strain formulation results in between 0.25 and 0.16.
Finally, the stiffness tensors (3.58) can be inverted to get the compliance
tensors
(1) 1
ij
Sijkm = ij km ( A S) + ik jm (S + P) + im jk (S P) Sij(2) = ,
4
(3.78)
so that, given the definition of the micropolar characteristic length

(Chapter 6),

S+ P s 1 + 3 (w/s) 2
l= = , (3.79)
4M 4 1 + (w/s) 2
where we used the basic facts relating the beams cross-sectional area to its
moment of inertia.
3.4.2 Triangular Lattice of Timoshenko Beams

In the foregoing section we began with the model of BernoulliEuler beams,
which implies slender connections between the lattice nodes. It is well known
that the situation of stubby connections is describable more adequately by
Timoshenko beams. The boundary value problem that needs to be solved is
that of a beam fixed at both ends and subjected to a shear displacement at one
end. That is, with the boundary conditions at the beams left end where v is
displacement and rotation).
(0) = 0 (0) = 0
(0) (0) = 0, (3.80)
and right end
(s) = (b) s (s) = 0

(s) (s) = 0, (3.81)
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
it is readily determined from the beams governing equations
E I
+ G A(
) = 0 G A(
) = 0, (3.82)
that a following relation holds between the shear force F (b) and the displace-
ment s (b)
12E (b) I (b) (b)

F (b) = s . (3.83)
s 3 (1 + )
Here
12E I (b) E w 2
= (b) 2
= (3.84)
GA s G s
is the dimensionless ratio of bending to shear stiffness, with A(b) = ta being

the beams cross-sectional area, and I (b) = ta w 3 /12 its centroidal moment of
inertia.
Two limiting cases are noteworthy:
0, high shear stiffness and, hence, less deflection owing to shear; the
BernoulliEuler slender beam is recovered.
> 1, low shear stiffness and, hence, deflection owing to shear domi-
nates over that due to the Young modulus E; this is the general case of the
Timoshenko beam.
Observing that (3.83) replaces (3.68) 2 , we now proceed to derive the effec-
tive moduli so that (3.71) is replaced by
2E (b) A(b) 24E (b) I (b) 1 2E (b) I (b)

R(b) = R(b) = 3 S(b) = , (3.85)
s (b) 3 s (b) 31+ s (b) 3
whereby we note
R w 2 1
= . (3.86)
R s 1+
Following the same steps as in the previous section, we see that the ef-
fective continuum moduli are given by (3.76 and 3.77) as before. It is now
possible to express them in terms of the beam aspect ratio and . Thus, for
E eff (normalized by the beams modulus E (b) ) and eff we find
E eff w 1 + (w/s) 2 / (1 + )
= 2 3 ,
ta E (b) s 3 + (w/s) 2 / (1 + )
(3.87)
1 (w/s) 2 / (1 + )
eff
= .
3 + (w/s) 2 / (1 + )
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
Considering the geometry of the hexagonal unit cell of the lattice, we can
write the above in terms of the porosity p (i.e., the pores volume fraction):
2
E eff 1+ 1 1 1 p
3(1+)
= 2 1 1 p 2 ,
ta E (b) 3 + 3(1+)
1
1 1 p
2 (3.88)
3 1 1 p 3/ (1 + )
=
eff
2 .
9 + 1 1 p 3/ (1 + )
The above are plotted in Figure 3.11(b) and (c) for the special case E (b) =
(b)
9G . We see that the consideration of lattice connections as stubby (Timo-
shenko) beams has a minor softening effect on E eff relative to the Bernoulli-
Euler beam model.
This may be explained by noting that the admission of the beams angle
of rotation as an independent degree of freedom amounts to G (b) being finite,
rather than infinite, in the Timoshenko beam model. The Poisson ratio falls
off nonlinearly from 1/3 with p increasing in both models. Also here, the
admission of finite shear modulus is weak. Also plotted in Figure 3.11(b) and
(c) are the results for the central-force lattice of Section 3.2.3, the perforated
plate model introduced below, and the Cox model discussed in Chapter 4.
3.4.3 From Stubby Beams to a Perforated Plate Model

As the porosity p goes beyond 50%, the beams aspect ratio w/s increases so
high that one can no longer model the connections between the lattice nodes
as beams. Thus, a basic question arises: can any simple explicit model be
derived for this low porosity range? One avenue is offered by a perforated
plate model. In the limit of p 1, we have a plate with a regular distribution
of triangular-shaped pores, Figure 3.11(a). This is a so-called dilute limit of
a locally isotropic material with holes (in either periodic or disordered ar-
rangements). Following Jasiuk et al. (1994) and Jasiuk (1995), the respective
formulas are
E eff
= 1 (1 p) eff = (b) (b) 0 (1 p) . (3.89)
ta E (b)
The coefficients = 4.2019 and 0 = 0.2312 have been computed in the
above references, and, in fact, analogous coefficients are also available there
for plates with other than triangular holes (squares, pentagons, etc.). It is
noteworthy that:
1. Both formulas are uncoupled from one another.
2. (3.89)1 models the high porosity range much better than the beam
lattice model; E eff is modeled by an upper envelope of all the curves
in Figure 3.11(b), that is, curves 3 and 5.
3. (3.89)2 depends on the Poisson ratio (b) of the plate material; the
latter can be specified only in the Timoshenko beam model.
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
(a)
1
4
3
0.8
2
1
0.6
E e
E (b) 0.4 4
0.2
0
0 0.2 0.4 0.6 0.8 1
p
(b)
0.5
0.4 1,4
2 3
0.3
v e
5
0.2
0.1
0
0 0.2 0.4 0.6 0.8 1
p
(c)
FIGURE 3.11
(a) A decrease in pore sizes (left to right): from large holes (slender beams), through a lattice of
stubby beams, to a plate perforated with small holes; shown at porosities p = 10%, 50%, and 90%.
(b) Effective Youngs modulus E eff , normalized by the beams Youngs modulus, as a function
of p for: (1) the central-force lattice, (2) the Timoshenko beams lattice, (3) the BernoulliEuler
beams lattice, (4) the Cox model, and (5) the effective medium theory for a perforated plate.
(c) Effective Poissons ratio eff as a function of p, models (15) shown.
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
One more question remains in connection with Figure 3.11: What happens
when the p values are too high for a beam lattice model to hold and too low
for the dilute model to be truly dilute? Or, can anything be done to smooth out
the transition between the two curves 3 and 5 at P f around 0.8? One could try
here a usual device of micromechanics: an effective medium theory in any one
of its guises: differential scheme, self-consistent, etc. (Jasiuk, 1995). However,
for the sake of clarity of Figures 3.11(b) and (c), we do not plot these.
Summing up, it is seen from Figure 3.11 that, as p grows, beam bending
tends to increase the effective Young modulus E eff . In other words, bending
effects increase as connections become wider. On the other hand, as they
become slender, one can work with segments carrying axial forces only. Thus,
beam effects gain in influence as the pores volume fraction increases, and lead
to an increase of the effective Young modulus relative to the central-force
model.
Two more things may be said about the beam network model. First, Tim-
oshenko beams, although more sophisticated than BernoulliEuler beams,
remain, in principle, 1D objects, of micropolar type in fact. Therefore, what
they yield is about as far as one gets with a beam model. A better approach
would have to consider beam segments as little plates, that is, 2D objects.
Finally, lattice nodes that are taken as rigid objects in this model could more
realistically be modeled by considering their deformability.
3.4.4 Hexagonal Lattice of BernoulliEuler Beams

Elasticity of a hexagonal (honeycomb) lattice of beams may be handled in
a manner similar to that of the triangular lattice. However, the analysis is
somewhat more complicated, due in part to the setup of a unit cell, whose
choice is nonunique. Wang and Stronge (1999) analyzed a honeycomb in plane
strain, which in a cross section is equivalent to a hexagonal beam lattice;
see also Warren and Byskov (2002). As before, the continuum kinematics
is described by three functions (3.63). The strain energy of the micropolar
continuum is of the general form (3.69), and the stiffness tensors possess the
isotropies (3.73). In particular, the non-zero components are
(b) 3
(1) 3E (b) w 1 + 3 (w/s) 2 (1) 3E w 3 + (w/s) 2
C1111= C1212 =
6s 1 + (w/s) 2 6s 3 1 + (w/s) 2
(b) (b) 3
(1) 3E w 1 (w/s) 2 (1) 3E w 1 (w/s) 2
C1122 = C 1221 = (3.90)
6s 1 + (w/s) 2 6s 3 1 + (w/s) 2
(b) 3
(2) 3E w
C11 = .
36s
The effective bulk and shear moduli are now identified as
E (b) w E (b) w
= = , (3.91)
2 3s 3s 1 + s 2 /w 2
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
which then lead to
4E (b) w 1 (w/s) 2
E= = . (3.92)
3s 3 + s 2 /w 2 1 + 3 (w/s) 2
Again, considering the slenderness constraints of beam elements (neither

too thin, nor too thick), this model admits between (just under) 1 and
0.85. With reference to Section 6.1 of Chapter 6, interpreting this lattice in
the plane stress formulation does not change , whereas adopting the plane
strain formulation results in between 0.5 and 0.45.
From (3.90) we determine the compliances

2 3s 2 3s 3 3s s2 8 3s
A= P= S= 1+ M= , (3.93)
E (b) w E (b) w 3 E (b) w w2 E (b) w 3
so that the micropolar characteristic length defined in (Chapter 6) is given by

S+ P s
l= = (w/s) 2 + 3. (3.94)
4M 4 2
This result compares very well with experiments of Mora and Waas (2000).
3.4.5 Square Lattice of BernoulliEuler Beams

Following the same procedure as in Section 3.4.1, we now analyze a micropo-
lar model of a square lattice network, Figure 3.12. Thus, assuming Bernoulli
FIGURE 3.12
A square beam lattice.
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
Euler beams, by analogy to (3.70), we have

4
4
(1)
Cijkm = ni(b) nk(b) n(b)
j n(b) (b)
m R + n(b) (b) (b)
j nm R Cij(2) = ni(b) n(b) (b)
j S , (3.95)
b=1 b=1
where
E (b) A(b) 12E (b) I (b) E (b) I (b)
R(b) = R(b) = 3 S(b) = . (3.96)
s (b) s (b) s (b)
When all the beams are identical, this leads to

(1) (1) (1) (1) (2) (2)
C1111 = C2222 = R C1212 = C2121 = R C11 = C22 = S, (3.97)
with all the other components of the stiffness tensors being zero. Clearly, this
beam lattice results in a special case of an orthotropic continuum.
With reference to Chapter 6, we now have two micropolar characteristic
lengths

S s S I
l1 = = l2 = =r , (3.98)
R 2 3 R A
where the radius of gyration r is easily recognized. For beams

of a rectangular
cross-section, the second one of these becomes l2 = w/2 3.
In the foregoing derivation, lattice nodes were taken as rigid objects. As
Wozniak (1970) showed, this model may be generalized to a situation of
deformable nodes, in which case we have
!
(1)
4
(b) (b)

4
(b) (b) (bb) (b) (b) (b) (b) (b)
Cijkm = ni n j nk nm R + ni n j nk nm R
b=1 b=1

4
Cij(2) = ni(b) n(b) (b)
j S , (3.99)
b=1
where
!
d E ( I ) ( I ) E ( I )
R (bb)
=
1 ( I ) ( I I ) ( I I ) E ( I I ) E ( I I )
24E (b) I (b) 2E (b) I (b)

R(b) = 3 S(b) = . (3.100)
s (b) 3 s (b) 3
Recently, an extension of such micropolar modelswith an eye for wave

propagation and vibration phenomenahas been carried out through the
introduction of internal variables (Wozniak, 1997; Cielecka et al., 1998). Such
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
models, in contradistinction to the more classical homogenization methods,

do more correctly account for the internal microstructure.
3.4.6 Nonlocal and Gradient Elasticity on a Lattice with Central Interactions

3.4.6.1 General Considerations
Let us now focus our attention on a lattice made of two central force structures,
with structure I (a regular triangular network with short-range interactions)
and structure II (three regular triangular networks with long-range interac-
tions). These structures are superposed in a way shown in Figure 3.13, so that
a typical node communicates with its six nearest neighbors via structure I,
and with its six second neighbors via structure II. Generalizing the develop-
ment of Section 3.2.3, the central-force interaction in the spring
connecting
the nodes r and r
is related to the displacements u(r) and u r
of these nodes
by

Fi r, r
= rr
ij u j r, r , (3.101)
where

rr
ij = ri r j
rr
u j r, r
= u j r
u j (r) ri = ri
ri . (3.102)
Similar to the case of antiplane elasticity, rr is the spring constant of a half-

length of a given nodenode interaction. However, assuming the structures
I and II to be made of two types of springs, we simply have two kinds of
spring constants:
I and I I ; s I = 2l is the lattice spacing of structure I, while
s = s 3 is that of structure II.
II I
3 2
Vo = l
2
I
II
FIGURE 3.13
Two structures, I and II, resulting in a lattice with local (nearest neighbor) and nonlocal (second
neighbor) interactions. Note that structure II consists of three triangular networks having separate
sets of nodes, and that all these nodes coincide with the nodes of structure I. (From Holnicki-Szulc
and Rogula (1979a. With permission.)
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
A question now arises: what continuum model should be set up to ap-

proximate this discrete system? Following Holnicki-Szulc and Rogula (1979a),
three types of continuum models will now be formulated: local, nonlocal, and
strain-gradient.
3.4.6.2 Local Continuum Model

Proceeding as in Section 3.2.3 under conditions of uniform strain, andpostu-
lating the equivalence of strain energy in a unit cell of volume V = 2 3l due
to all the spring constants
1
1 r,r

E= Fi r, r ui r, r
= ij ui r, r
u j r, r
(3.103)
2 r,r
2 r,r
to equal the strain energy of an effective continuum

1
E= ij Cijkm km d V (3.104)
2 V
we determine an effective, local-type stiffness tensor
Cijkm = C Ijkm + Cijkm

II
, (3.105)
where, recalling (3.34),
2 I (b) I (b) I (b)

I
Cijkm = I ni n j nk nmI (b)
3 b=1,2,3
6 I I (b) I I (b) I I (b) (3.106)
II
Cijkm = I I ni n j nk nmI I (b) .
3 b=1,2,3
The niI (b) and niI I (b) are given by (3.32). The above tensors are of the form (3.37),
so that
3 3
I = I = I I I = I I = I I . (3.107)
8 3 8 3
3.4.6.3 Nonlocal Continuum Model

As pointed out in Chapter 6, a non-local model should result in stresses at
a point dependent upon the deformation within the range of interactions
associated with the point. As a result, the more inhomogeneous is the strain
field, the closer is the nonlocal model to grasping the actual strain state of
I II
the lattice. First, we distribute the values of tensors Cijkm and Cijkm at point r
uniformly over the regions of interactions of structures I and II (Figure 3.13),
P1: Binod
July 12, 2007 19:23 C4174 C4174C003

and form a new tensor Cijkm r, r
such that

Cijkm r, r
= Cijkm
I
r, r + Cijkm
II
r, r
(3.108)
I
Cijkm h I r, r

II
Cijkm h I I r, r
I
Cijkm r, r = r, r = II
Cijkm .
AI AI I
2 2
Here AI = s I , AI I = s I I are the areas, and h I r, r
and h I I r, r
are the characteristic functions of the regions of interactions in the neighbor-

hood of r.
Note: Stress in a nonlocal elastic continuum may be determined as a gra-
dient of energy functional in the space of strains, but this stress lacks the
clear meaning of a Cauchy stress in a local continuum. The model of nonlocal
interactions is valid only in an infinite body, and its justification has to be
found in an appropriate limit of the lattice dynamics of an infinite crystal
(Maugin, 2002). However, there is a problem at the boundary: constitutive
properties in the boundary layer of a homogeneous nonlocal continuum are
necessarily different from those in the interior, which contradicts the assump-
tion of homogeneity.
3.4.6.4 Strain-Gradient Continuum Model

Again with reference to Chapter 6, a strain-gradient model is similar to the
nonlocal model in that it resolves the local inhomogeneity of deformation
within the range of interactions associated with a continuum point. One be-
gins here with a series expansion of the relative displacement field involving
two termslinear and quadraticthat is,
1
u j r, r
= ijr r
j r
j + irjk r
j r j rk
rk , (3.109)
2
where
ijr = u(i, j) (r) irjk = (i, j,k) (r) (3.110)
are gradients of the first and second orders, respectively, of the displacement
field.
In view of (3.103), the elastic energy of the structure (3.104) is now
expressed as

1 r,r
r
1 r

E= ik rk rk + ikm rk rk rm rm
2 r,r
ij 2
(3.111)

1

+ jm rm rm + jmn rm rm rn rm .
r r
2
Observing the continuum form of energy

1
E continuum = ij Cijkm kl + i jk Cijklmn lmn d V, (3.112)
2 V
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
one can identify

Cijkl r, r
= Cijkl
I
r, r + Cijkl
II
r, r

(3.113)
Cijklmn r, r
= Cijklmn
I
r, r + Cijklmn
II
r, r ,
where

2E I AI I (b) I (b) I (b) I (b)
I
Cijkl r, r = ni n j nk nl
3s I b=1,2,3
(3.114)

2E I AI s I I (b) I (b) I (b) I (b) I (b) I (b)

I
Cijklmn r, r = ni n j nk nl nm nn ,
3 b=1,2,3
II

I

with completely analogous formulas holding for Cijkl r, r and Cijklmn r, r .
Holnicki-Szulc and Rogula (1979b) present further considerations of these
models, especially in connection with the setup of boundary value problems,
the modeling of surface energy accounting for the heterogeneity of material
properties in the boundary layer of the microstructure, and the determination
of internal forces.
The subject of higher-order gradient theories has been receiving a lot of
attention over the past two decades (e.g., Bardenhagen and Triantafyllidis,
1994). We return to it in the context of composite materials in Chapter 9.
Various mathematical aspects of lattice, truss, and frame systems are studied
by Martinsson and Babuska (2006, 2007); see also Pshenichnov (1993), and
Cioranescu and Saint Jean Paulin (1999).
3.4.7 Plate-Bending Response

We can apply the same approach as that outlined so far for the in-plane prob-
lems, to the determination of effective plate-bending response of a periodic
beam network. We sketch the basic ideas in terms of a triangular lattice, within
the assumptions of a Kirchhoff (thin) plate model. To this end we must con-
sider out-of-plane deformations of a triangular geometry lattice, Figure 3.14.
The kinematics is therefore described by three functions, one out-of-plane
displacement and two rotations (with respect to the x1 , and x2 axes):
u ( x) w1 ( x) w2 ( x) , (3.115)
which coincide with the actual displacement (u) and rotations (w1 , w2 ) at the
lattice vertices. Within each triangular pore these functions may be assumed
to be linear. It follows then that the strain and curvature fields are related to
u, 1 , 2 by
kl = l,k k = u,k + kl l . (3.116)

P1: Binod
July 12, 2007 19:23 C4174 C4174C003
~
P(b) M (b)
M(b)
FIGURE 3.14
A perspective view of a triangular lattice, showing the relevant internal loads in a beam cross-
section.
With reference to Figure 3.14, for a single beam the mechanical (force-
displacement and moment-rotation) response laws are given as
12E (b) I (b) (b)

M(b) = C (b) (b) M(b) = E (b) I (b) (b) P( f ) = , (3.117)
s (b)
where A(b) = ta is a cross-sectional area of the beam; I (b) = ta3 w/12 is a

centroidal moment of inertia of the cross-sectional area of the beam with
respect to an axis normal to the plane of the lattice; s (b) = 2l (b) is the full length
of each beam. The beams parameters are as follows: torsional stiffness C (b) ;
Youngs modulus E (b) ; in-plane twisting moment M(b) ; out-of-plane bending
moment M(b) ; shear force P (b) ; shear deformation (b) ; curvature (b) .
The strain energy of the unit cell is
V V eff
Ucontinuum = ij Cijkl kl + i Aij j , (3.118)
2 2
which is consistent with Hookes law
mkl = Cijkl kl pk = Akl l . (3.119)
Here mkl is the tensor of moment-stresses, pk is the vector of shear tractions.

P1: Binod
July 12, 2007 19:23 C4174 C4174C003
Proceeding in a fashion analogous to the in-plane problems, Wozniak

(1970) found

6
6
Cijkl = ni(b) nk(b) n(b) (b) (b)
j nl S + n(b) (b) (b)
j nl S Aij = ni(b) n(b) (b)
j R , (3.120)
b=1 b=1
where
C (b) E (b) I (b) 12E (b) I (b)

S(b) = S(b) = R(b) = 2 . (3.121)
s (b) s (b) s (b) s (b)
In the case of a triangular lattice made of identical beams (E (b) = E, etc.) we

find
Cijkl = ij kl + ik jl Y + il jk Aij = ij B, (3.122)
in which
3 3 3
== S S Y= S + 3 S B= R
8 8 2
(3.123)
2C 2E I 24E I
S= S = R = .
s 3 s 3 s3 3
The same type of derivation may be conducted for a lattice of either rectan-
gular or hexagonal geometry.
3.5 Extension-Twist Coupling in a Helix

3.5.1 Constitutive Properties
The Timoshenko beam appears several times throughout this book. Interest-
ingly, while it involves a shear force and a moment normal to the beams
axis, both mutually orthogonal, in a helix the axial force and the moment
are parallel to the main axis. The mechanics of this periodic system presents
many interesting phenomena involving at least two scales, and leaves open
challenges. The pitch of the helix provides one unit cell.
Consider a wire rope made of helically shaped strands (helices), wound
along the x-axis (Costello, 1997), Figure 3.15. The helical geometry of strands
gives risealready in the case of a single strandto a coupling between axial
and twisting effects. More specifically, if we replace this helical system by a
rod-like element along x, the axial force F parallel to x is directly related not
only to the axial strain u, x (= ) but also to the rotational strain , x (angle of
twist per unit length). Similarly, the torque M acting along x is a function of
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
F M
FIGURE 3.15
A wire-rope made of several strands, each of the same helix angle.
, x and . Thus, the effective constitutive equation of the helix is

F !) *
C11 C12 u, x
AE
= . (3.124)

M C21 C22 , x

E R3
Here A and R = R1 + 2R2 are the effective cross-sectional area and radius of
the rod-like element, where R1 is the radius of the center straight strand and
R2 is the radius of m helical strands winding around it and thus forming a
wire rope. E is the Young modulus of the strands materials, while its Poisson
ratio appears in the expressions for Cij below.
Here, from a requirement of a positive definite strain energy density, we
obtain two conditions on four constitutive coefficients Cij :
C11 > 0 C22 > 0 C12 = C21 C11 C22 C12 C21 > 0. (3.125)
In the language of continuum mechanics, the wire rope is a 1D micropolar

medium of a noncentrosymmetric type. Basic equations of such continuaalso
called a hemitropic, antisymmetric, or chiral materialare given in Chapter 6.
Costello (1997) analyzed the elastostatics of a bundle of m helical strands (each
of radius R2 ) winding around the center straight strand (of radius R1 ), jointly
forming a wire rope. The following assumptions were involved:
1. The strands are uniformly spaced along the perimeter of a circle of

radius r2 = R1 + R2 , thus forming a ring in a plane perpendicular
to the bundles axis without touching each other.
2. Each strands equilibrium configuration is a helix of constant radius
r2 = R1 + R2 and constant helix angle 2 (the subscript 2 refers to
the deformed configuration).
3. Strands are linear elastic (with an axial modulus and Poissons ratio)
and undergo very small strains.
4. Friction and contact deformations are neglected.
By reworking Costellos analysis, the coefficients Cij have been derived

explicitly (Shahsavari and Ostoja-Starzewski, 2005b), here we reproduce only
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
the first one:
R12 mR22
C11 = +
R12 + mR22 R12 + mR22 r2 + r2 tan2 2 + R2

R22 1 2 sin2 2 sin 2 cos2 2 (3.126)
r2 tan 2 R1 sin 2 +
2

4r2
,

R22 sin3 2 cos2 2 (1 + ) 2 R22 sin 2 cos4 2 R1 + r2 tan2 2
+ +
2r2 4 (1 + ) r2
In Figure 3.16 the strain energies involved in these Cij coefficients are plot-
ted as functions of . These energies are based on 2 /2E and 2 /2G as strain
energy densities for axial and torsional deformations, and are normalized
upon dividing by E. We see that:
1. Strain energy plots involved in C11 and C22 are qualitatively similar
to analogous plots of C11 and C22 themselves. As expected, with
approaching /2, the energy in C11 tends to a maximum. How-
ever, the maximum energy in C22 occurs at value of different
from /2.
103
0.014 1.2
0.012 1
C11 Energy
C21 Energy
0.01
0.8
0.008
0.6
0.006
0.004 0.4
0.002 0.2
0 00
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6

103 104
1 6
0.8 5
C21 Energy
C22 Energy
4
0.6
3
0.4
2
0.2 1
0 0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6

FIGURE 3.16
Strain energies associated with Cij s as functions of , at a helical strands Poisson ratio = 0.25.
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
2. The strain energy plots involved in coupling terms C12 and C21 are
qualitatively similar, but not exactly the same. They are quite close
for > 0.5 rad, with the discrepancies stronger below that value.
The maximum difference between those strain energies is about
20%.
The dependencies on (with = 0 rad signifying the case of strands fully

aligned with the wire axis) are shown in Figure 3.16. In the first place we note
that C12 = C21 , and this points to an open challenge of providing a better
derivation of Cij s.
Of course, in light of the BettiMaxwell reciprocity theorem applied to the
helix on the macro level one should have C12 = C21 . Now, the key assumptions
that cause C12 = C21 are: the product of higher-order terms resulting from the
strain of a single helical strand is neglected; the changes in the curvature and
twist per unit length are linearized; small changes in helix angle (from 1 to 2 )
have been assumed, which then allowed some trigonometric functions to be
simplified accordingly. It is not clear which assumption should be relaxed so as
to arrive at C12 C21 without dealing with yet more complicated derivations
and formulas. Interestingly, an analogous challenge exists in experimental
mechanics: to measure C12 and C21 such that they are equal.
The phenomenological equations (3.124) generally apply to more complex
physical systems than a wire rope, for example a wood fiber made of helically
wound fibrils, or a continuum shell. Indeed, it was shown, in the context
of structural mechanics (Blouin and Cardou, 1989), that either assumption
would lead to a few percent difference (at most 11%) for any of the Cij
coefficients. It remains to be seen, however, what those differences would
be for a shell made of a large number of thin cellulose fibrils winding along
the axis of a cellulose fiber rather than a few thick wires such as shown in
Figure 4.10(d) of Chapter 4. Note that a helical inclusion in an elastic space
was studied by Slepyan et al. (2000).
3.5.2 Harmonic Waves

3.5.2.1 Elastic Helix
Equations (3.124), together with the equations of motion, led Samras et al.
(1974) to derive a system of two coupled wave equations governing the axial-
twisting response of a fiber
.. ..
C11 u, xx + C12 , xx = u C21 u, xx + C22 , xx = J , (3.127)
where is the mass density and J is the mass polar moment of inertia. They
considered a monochromatic wave propagation along the fiber
u (x, t) = Ue ik(xct) (x, t) = e ik(xct) , (3.128)

P1: Binod
July 12, 2007 19:23 C4174 C4174C003
and arrived at a dispersion relation whose analysis indicated two possible

wave speeds:
2 (C11 C22 C12 C21 )
c 1,2 =
1/2 . (3.129)
(C11 J + C22 ) (C11 J C22 ) 2 + 4 J C12 C21
From this it followed, by inspection, that c 1 < c 2 , and, in fact, there may be an
order of magnitude difference between both wave speeds. Actual numbers
depend, of course, on the choice of a theory employed for the derivation of all
Cij s or on the pertinent experiments. Given (3.129), Samras et al. (1974) found
that the axial vibrations of the helix are described by two types of waves
slow and fasteach of which consists of forward and backward traveling
pulses
u (x, t) = U1 e ik(xc1 t) + U2 e ik(x+c1 t) + U3 e ik(xc2 t) + U4 e ik(x+c2 t)

(3.130)
(x, t) = 1 e ik(xc1 t) + 2 e ik(x+c1 t) + 3 e ik(xc2 t) + 4 e ik(x+c2 t) .
Next, by considering the ratio of axial to torsional amplitudes U/ , they

concluded that the waves that are primarily axial in nature (U/ > 1) prop-
agate at speeds c 1 , whereas the waves that are primarily torsional in nature
(U/ < 1) propagate at speeds c 2 . Clearly, by assuming C12 = C21 = 0 one
immediately arrives at two uncoupled wave equations
for purely axial and
torsional waves, respectively, i.e., c 1,2 = C11 / , C22 /J .
The above analysis is the stepping stone for deriving the spectral finite
element of a helical element. This element is specified via a 4 4 spectral
stiffness matrix connecting kinematic quantities (extensions + +i )
ui and twists
and dynamic quantities (forces F +i and moments M + i ), all in the frequency
space, at both ends (i = 1, 2) of the element of length L:
+

F1
+
u1

F
+2 +u
= [K +] 2
. (3.131)

+1
M
+1

+2
M +2

+ ] and a
Explicit formulas for all the elements of the spectral stiffness matrix [ K
study of their dependencies on frequency and the coupling coefficient C12 =
C21 was given by Shahsavari and Ostoja-Starzewski (2005a).
3.5.2.2 Thermoelastic Helix

Recently, we have generalized the elastodynamic helix model to account for
coupled thermoelastodynamic effects (Ostoja-Starzewski, 2003). Introducing
the thermal expansion effect ( E) into (3.124)1 above, we find
T = A1 u, x + A2 , x E
M = A3 u, x + A4 , x (3.132)
Q = AK , x ,
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
where the third equation governs the heat conduction (Q = Aq is the flux in
the whole cross-section) according to the basic model of Fourier; K is the ther-
mal conductivity along the strands axis. This leads to three coupled equations
governing the triplet (u, , )
..
A u = A1 u, x + A2 , x E
..
J = A3 u, x + A4 , x (3.133)
c v
= K , xx E T0 u, x .
To
Upon taking time- and space-harmonic wave forms, we get a system of three
algebraic equations, which results in a characteristic equation for the roots c
in terms of k and all the material parameters

Ac 2 AE A2 i AE/k

A3 J c 2 A4 0
=0. (3.134)
c E T
0 c
0 K
c v
ik
c v
Evidently, the wave motion is not only dispersive, as in the mechanical model
above, but also damped. Equation (3.134) leads to a fifth-order algebraic equa-
tion for the roots c, which cannot be solved explicitly, and a numerical root-
solving method has been employed to assess the trends in change of c in
function of wave number k, in the presence of weak thermal effects expressed
by the dimensionless thermoelastic coupling constant () and the thermal
diffusivity at constant deformation (kv )
2 E T0 K
= kv = . (3.135)
c v c v
Considering the values pertaining to an oceanographic steel cable studied
by Samras et al. (1974), we have found that, with increasing from zero up,
while keeping kv = 0, both c 1 and c 2 increase linearly; with in metals taking
values up to 0.1, those speeds may easily go up by a few percent. On the other
hand, increasing kv from zero up, while keeping = 0, has no effect on c a
and c t .
As is well known, it is more correct in coupled thermoelastodynamics to
use a wave-type equation for heat propagation. Thus, employing the Maxwell
Cattaneo model, we must replace (3.133)3 by
E T0 ..
c v ..
u, x + u = , xx + , (3.136)
K ,x K
where is a time-like parameter. We now have a sixth-order algebraic equation
for the roots c, but the ensuing numerical analysis reveals that they are only
weakly affected by .
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
3.5.3 Viscoelastic Helices

3.5.3.1 Viscoelasticity with Integer-Order Derivatives
Recall from Section 3.5.1 the explicit formulas of Cij in terms of parameters of
the geometry and elastic properties of a helical strand (or a bundle of strands).
Although these formulas are not perfect (recall C12 = C21 ), one may begin
to examine viscoelastic helices, that is, helical strands made of viscoelastic
materials. We consider such materials to be non-micropolar and differential
type, that is,
.. ..
T + P1 T + P2 T + = Q0 + Q1 + Q2 + . (3.137)
Here T and stand for the Cauchy stress and infinitesimal strain, respectively.
We now inquire whether a helical strand made of, say, a Kelvin material will
result in an effective Kelvin type or more complex response on the helix level
(Shahsavari and Ostoja-Starzewski, 2005b).
We proceed by using the correspondence principle of viscoelasticity. That
is, we replace the elastic modulus E and elastic Poisson ratio by the Laplace
transforms of appropriate viscoelastic relaxation functions and viscoelastic
Poissons ratio, multiplied by the transform parameter s:
E s E (s) s (s) . (3.138)
Here we recall that in most viscoelastic materials, Poissons ratio is not a

constant, but rather a function of time or frequency (Hilton, 2001; Tschoegl
et al., 2002). We also need the relations of E (s) and (s) to the transformed
bulk and shear moduli (s) and (s)
9 (s) (s)
E (s) =
3 (s) + (s)
(3.139)
3 (s) 2 (s)
(s) = .
2s [3 (s) + (s)]
Now, assume that the helical strand material is described by two Kelvin
models, one for dilatational and another for distortional response:
Tm = Q0 m + Q1 m T
= q 0
+ q 1
, (3.140)
where m and
denote the isotropic and deviatoric parts of stress and strain
tensors, respectively. It follows that

3 q 0 Q0 + (q 0 Q1 + Q0 q 1 ) s + q 1 Q1 s 2
s E (s) =
2Q0 + q 0 + (2Q1 + q 1 ) s
(3.141)
Q0 q 0 + ( Q1 q 1 ) s
s (s) = .
2Q0 + q 0 + (2Q1 + q 1 ) s
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
By using the above equations together with (3.124), we arrive at relations

linking the Cauchy stress = F /A and the couple-stress = M/A, along
with their time derivatives, with axial and rotational strains (u, x and , x ) and
their derivatives
.. .. ...
D0 + D1 + D2 = E 0 u, x + E 1 u, x + E 2 u +E 3 u +B0 , x
,x ,x
.. ...
+B1 , x + B2 +B3
,x ,x
.. .. ...
D0
+ D1
+ D2
= E 0
u, x + E 1
u, x + E 2
u +E 3
u +B0
, x
,x ,x
.. ...
+B1
, x + B2
+B3
. (3.142)
,x ,x
Here all the constant coefficients Di , E i , Bi , Di

, E i
, and Bi
, i = 1, . . ., 3, are
functions of the strands geometry and the Kelvin model parameters.
Observe that the constitutive equations of the helix (3.142) are qualita-
tively different from those of the Kelvin material itself, i.e., (3.140). Moreover,
one can show that only in the singular case of = 0 do the second- and third-
order derivatives vanish, and then the helix becomes one of a Kelvin type.
Analogous results are obtained upon assuming either Maxwell or Zener mod-
els of the strand. This indicates the general trend: the macro-level viscoelastic
response of the helix is different in type and more complex than that of the vis-
coelastic material at the micro-level (the strand material itself). Consequently,
direct viscoelastic generalizations of effective constitutive equations of he-
lices, not based on analyses such as those presented here, are likely to be
invalid.
3.5.3.2 Viscoelasticity with Fractional-Order Derivatives

The above results are amplified in the case of helices governed by viscoelas-
ticity with fractional derivatives. Although the advantages of such a formu-
lation of rheological relations are well known (e.g., Bagley and Torvik, 1983),
the thermomechanics should be closely connected with the fractal geometry
of the material (Carpinteri and Mainardi, 1997), a subject that still requires
development, see Section 10.6 of Chapter 10.
Let us start with a generalization of (3.140)
Tm = Q0 m + Q1 D m T
= q 0
+ q 1 D
, (3.143)
with time derivatives of order and meant in the sense of Caputo (1967),
e.g.,
t
d f (t) 1 d
D f (t) =
= (t ) f ( ) d. (3.144)
dt (1 ) 0 dt
Proceeding in the same manner as earlier for the model involving integer-
order derivatives, we find
P = Qu, x + S, x P
= Q
u, x + S
, x , (3.145)
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
where
P = c 0 + c 1 D + c 2 D + c 3 D+ + c 4 D2 + c 5 D2 ,
Q = h 0 + h 1 D + h 2 D + h 3 D+ + h 4 D2 + h 5 D2
+h 6 D+2 + h 7 D2+ + h 8 D3 , (3.146)
+
S = l0 + l1 D + l2 D + l3 D + l4 D
2
+ l5 D
2
+2
+l6 D + l7 D
2+
+ l8 D .3
Here value for c, h, and l are constant coefficients depending on the geometry
and the constitutive parameters of the helical strand. The operators P
,Q
,
and S
also comprise five, eight, and eight terms, respectively, analogously
to the three operators in (3.146). When the fractional-order derivative reverts
to the integer-order onethat is, upon setting = = 1 in (3.140)the
results reduce to (3.142). Thus, the fractional-order model leads to even higher
complexity of the helix than the conventional one. The same trend persists
when more realistic models (i.e., with more parameters) are adopted (Ostoja-
Starzewski and Shahsavari, 2007).
The foregoing observation relating to uniaxial helices also provides guid-
ance for admissible vis-a-vis inadmissible models of 3D chiral (i.e., helically
structured) materials. The constitutive relation of a linear elastic chiral mate-
rial in 3D involves two coupled equations linking the stress and couple-stress
tensors with the strain and torsion-curvature tensorssee equation (6.80) in
Chapter 6. In view of the above results, one cannot arbitrarily postulate vis-
coelastic generalizations of such equations in terms of differential tensorial
equations.
Problems
1. The dispersion relation, generally written as = (q ), expresses a
relation between the frequency and the wave number q . Verify
that the dispersion relation for the 1D chain of particles in Figure
3.1 is

= max | sin qs/2|max = 2 k/m.
Draw a plot to compare this dispersion relation with the one corre-
sponding to the equation (3.4). the range qs [, ] is the so-called
1st Brillouin zone. Finally, determine the wave speed c and the group
velocity c g of this 1D lattice, and consider their maximum possible
values.
2. Consider a simple orthorhombic lattice with all the bonds having
thermal conductivities K i ,where i is the bond direction xi .Derive
the effective thermal conductivity of that lattice.
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
3. Outline the derivation of (2.35).

4. Generalize the derivation of effective moduli of a hexagonal lattice
(Section 3.4.4) to the case of Timoshenko beams.
5. Generalize the derivation of effective moduli of a square lattice
(Section 3.4.5) to the case of Timoshenko beams.
6. Determine the effective moduli in plate-bending response of a square
lattice with beams of (a) BernoulliEuler and (b) Timoshenko type.
7. Demonstrate that, in the case of vibrations of an elastic helix, the
waves
that are primarily axial in nature propagate at speeds c 1 =
C11 /, while the waves
that are primarily torsional in nature prop-
agate at speeds c 2 = C22 /J .
8. Derive the spectral finite element of an elastic rod.
9. Derive the spectral finite element of an elastic helix.
10. Study the literature and write a report on the fractional derivative of
Caputo. What is its main advantages over the fractional derivative
of Riemann-Liouville?
P1: Binod
July 12, 2007 19:23 C4174 C4174C003
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
4
Lattice Models: Rigidity, Randomness,
Dynamics, and Optimality
There is a crack in everything: thats how the light gets in.

Leonard Cohen, 1992
This chapter takes lattice models to more general settings. There are four
themes: rigidity, randomness, dynamics, and optimality. The first two of these
involve the introduction of spatial randomness into a lattice as a departure
from an originally periodic geometry of, say, the central-force triangular net-
work of Chapter 3. One path is through a random depletion of bonds, which
leads to a total loss of stiffness, or rigidity, of the lattice. Another way of creat-
ing a random lattice is through a random (instead of a regular) network of, say,
a PoissonDelaunay variety. The third theme considered here is a generaliza-
tion from statics to dynamics, with nodes acting as quasi-particleshere we
have a coarse scale cousin to molecular dynamics, and, at the same time, an
alternative to finite element methods. Finally, the fourth topic is that of opti-
mal use of material for given loading and support conditions, where a special
case of central-force lattices arises, Michell trusses being the basic paradigm.
4.1 Rigidity of Networks

4.1.1 Structural Topology and Rigidity Percolation
When considering a central force (or truss) network, a question of fundamen-
tal importance is whether such a structure is a sufficiently constrained system
or not. In other words: is it an intrinsically rigid body? This is the subject mat-
ter of a field called structural topology. In the following we provide its basic
concepts.
Any central force network is a set of edges (or bars), and vertices
(or nodes acting as frictionless pivots). We immediately have an edge set
E, and a vertex set V, so that the network is represented by a graph G(V, E).
An edge is an unordered pair of two vertices. Structural rigidity can be based
either on statics or on kinematics, and, as we shall see below, they are, in a
certain sense, equivalent; see Table 4.1.
133
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
TABLE 4.1
A Comparison of Various Terminologies from Structural Mechanics,
Structural Topology, and Physics
Field Terminology Terminology
Structural mechanics Minimally constrained truss Mechanism

Structural topology Isostatic structure Nonisostatic structure
statical approach
Structural topology Infinitesimally rigid structure Nonrigid structure
kinematical approach
Condensed matter physics Rigid network Floppy network
The statical approach involves, in the first place, the concept of an equilib-
rium load. A system of forces assigned to the nodes of a network is said to be
an equilibrium load if and only if (iff ) the sum of the assigned vectors is a zero
vector, and the total moment of those vectors about any one point is zero. A
network resolves an equilibrium load iff there is an assignment of tensions
and compressions to all the bars of E, such that the sum at each node is equal
and opposite to its assigned load. A structure is said to be statically rigid iff
it resolves all equilibrium loads.
The kinematic approach involves the concept of an infinitesimal motion,
which is an assignment of velocities to all the nodes of V, such that the dif-
ference of velocities assigned to the ends of any bar is perpendicular to the
bar itself. This means that the motion does not result in any extension or
compression of the bar. Every connected plane structure has at least three
degrees of freedom (two translations and one rotation), and this is called a
rigid motion. A structure is said to be infinitesimally rigid if and only if all its in-
finitesimal motions are rigid motions. These statical and kinematical pictures
are connected by a theorem due to Crapo and Whiteley (1989): A structure
is statically rigid iff it is infinitesimally rigid.
Next, a structure is said to be isostatic iff it is minimally rigid, that is,
when it is infinitesimally rigid but the removal of any bar introduces some in-
finitesimal motion. Clearly, in an isostatic structure all the bars are necessary
to maintain the overall rigidity. In statics this is called a statically determi-
nate structure, as opposed to the indeterminate ones that have more than a
minimally sufficient number of bars for the global rigidity. It is a well-known
result that, in 2D, a determinate structure of |V| nodes has edges numbering
|E| = 2|V| 3, (4.1)
where || denotes the number of elements in a given set. As an example, let

us consider an incomplete triangular lattice shown in Figure 4.1. Although it
satisfies (4.1) it is not at all clear whether it is isostatic.
This example shows that |E| = 2|V| 3 is only a necessary but not a
sufficient condition for rigidity. The latter is provided by this theorem (e.g.,
Laman, 1970; Asimov and Roth, 1978): A planar graph structure is isostatic
if and only if it has 2|V| 3 bars, and, for every m, 2 m |V|, no subset
of m nodes has more than 2m 3 bars connecting it. This, effectively, allows
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
Lattice Models: Rigidity, Randomness, Dynamics, and Optimality 135
FIGURE 4.1
A triangular lattice with 71 edges and 37 vertices; it is generically rigid.
one to check whether the edges of the graph are not distributed spatially in
a uniform manner. If they are crowded locally, than the odds are that the
structure is not isostatic.
The isostatic concept so far discussed falls in the category of generic rigid-
ity, where only the topological information on a graphs connectivity comes
into the picture. However, one may also deal with unexpected infinitesimal
motions when, say, two edges incident onto the same vertex lie on a straight
line. The applicable techniques for analyzing such geometric problems are re-
viewed in Guyon et al. (1990). For a review of problems in 3D see Guest (2000).
When dealing with very large systemssuch as encountered in condensed
matter physicswe need to ask the question: what critical fraction, pr , of
edges of E needs to be kept so as to render the structure isostatic? We note
that we would have |E | = pr |E| new edges of thus modified, or depleted,
set |E |. It follows immediately from |E | = 2|V| 3 that we would have
pr = 2/3. This value is a simple estimate of the so-called rigidity percolation, a
concept also useful in biophysics (Shechao Feng et al., 1985; Boal, 1993; Hansen
et al., 1996). As shown in these references, the actual critical point occurs at
a somewhat different value than 2/3; theoretical methods involved include
effective medium theories and spring network computations. The latter of
these will be demonstrated later on the example of Delaunay networks.
Finally, it is important to keep in mind that the rigidity percolation typi-
cally occurs above the connectivity percolation, that is, pr > pc . For example,
pc = 1/3 in a triangular central-force network, while pr = 2/3. This is demon-
strated in Figure 4.2(a) in terms of the planar bulk and shear moduli in function
of the volume fraction C2 of phase 2, which is 105 times softer than phase
1 of the undepleted network, thus simulating zero stiffness bonds. Note that
C2 = 1C1 , and C1 p. That figure also shows the bulk and shear moduli of a
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
1
Truss shear modulus 1/E1
0.9 Beam shear modulus 2/E1
Truss bulk modulus 1/E1
0.8 Beam bulk modulus 2/E1
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
C2
(a)
1
Truss shear modulus 1/E1
0.9 Beam shear modulus 2/E1
Truss bulk modulus 1/E1
0.8 Beam bulk modulus 2/E1
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
C2
(b)
FIGURE 4.2
Dependence of bulk and shear moduli of triangular truss (central-force) and beam
networks, obtained by computational mechanics on 50 50 lattices at contrasts (a) 105
and (b) 102 .
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
triangular beam network (recall Section 3.4 of Chapter 3) at the same contrast
of phases. Clearly, rigid bonding and beam bending are essential in providing
the rigidity of the network right down to the point of connectivity percola-
tion at C2 = 2/3. The effect of loss of rigidity becomes less dramatic with a
decreasing contrast, see Figure 4.2(b). A much more extensive treatment of
rigidity percolation is in Sahimi (2003).
4.1.2 Application to Cellulose Fiber Networks

4.1.2.1 Rigidity of a Graph of Poisson Line Field Geometry
The planar Poisson line field was introduced in Chapter 1. If one reads the
basic assumptions of the classical article in paper physics/mechanics (Cox,
1952), one arrives at the conclusion that the geometry of a cellulose fiber
network assumed therein must be that of the Poisson line field. Note here
that:
A homogeneous field of infinite lines cannot be obtained from a

random field of straight and finite segments by extending those
segments to infinity.
The connectivity of the field of finite fiber segmentswhich clearly
is a more realistic representation of paperis lower that that of the
Poisson line field.
The Cox model also assumed the network of not interacting (!) infi-
nite lines/fibers to be held by a frame so as to make it solid-like.
Clearly, the last of these assumptions is not acceptable, but one way to
perhaps save the Cox model is first to note that under affine motions (ui =
ij0 x j , ij0 = const), straight lines transform into straight lines, and their original
(Lagrangian) points of intersection in plane are preserved. We now identify
the line segments (between any two consecutive intersections) to be edges of
E, and pivots to be vertices of V. Let us recall that the triple-fiber intersections
occur with probability zero for isotropic and anisotropic distributions of lines.
Thus, we typically have vertices of connectivity 4, that is, V4 .
Now, with reference to Figure 4.3(a), which shows a typical realization of
the Poisson line field, we see that there are two types of edges: those in direct
contact with the square-shaped window, and those entirely in the interior.
Clearly, the square window is needed to prevent these boundary layer bonds
from dangling, and this immediately renders the entire network a mechanism.
However, one may argue that the boundary layer of dangling bonds is very
thin relative to the whole field, and ask the question concerning the isostatic
condition for the graph G(V, E) representing the interior network of edges
not directly in contact with the square window boundary; these are shown in
bold in Figure 4.3(a). Here we observe that, while the V4 vertices occur in the
interior of this graph, its boundary involves V2 and V3 vertices. Now, since
there two vertices to every edge, we can calculate the total number of edges
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
(a) (b)
(c) (d)
FIGURE 4.3
Samples of (a) a planar Poisson line field and (b) a finite fiber field (recall Section 1.4 of Chapter
1), with a 1 = 1 and all other a i = 0. Test windows of size L L are considered. (c) Deformation
of a network of (b), with 195 fibers with originally straight fibers, with fiber bending present,
subjected to axial strain 11 = 1%. (d) The same network, with fiber bending almost absent,
subjected to axial strain 11 = 1%. All displacements in (c) and (d) are magnified by a factor 8 for
clarity. Figure (d) shows large, mechanism-type motions of the network including those of some
fibers which spring outside the original domain of the network.
in the bold drawn graph G(V, E) according to
|E| = |V2 | + 32 |V2 | + 2|V4 |. (4.2)
Evidently, since V = V2 V3 V4 , the total number of all the vertices is
|V| = |V2 | + |V3 | + |V4 | (4.3)
so that |E| < 2|V| 3 and the system is not isostatic, it is underconstrained
(i.e., a mechanism). Given this observation, the Poisson line field of axial force
fiber segments (the so-called Cox model, 1952) is not a valid model of paper,
or any other solid material for that matter. The fact that the Cox model does
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
give finite values for elastic moduli including the shear modulus is easily
explained by the presence of fully stretched fibers, spanning the entire test
window. The situation is analogous to a graph of a square lattice topology,
which, even though it is an obvious mechanism, will give finite axial mod-
uli in two directions, if fully stretched and subjected to kinematic boundary
conditions.
In real networks fibers have finite length, so their ends are loose. When
fiber ends are removed to eliminate their obvious mechanism motions, the
number of vertices in sets V2 and V3 , increases. Consequently, (4.1) is even
further from being satisfied. In order to deal with finite fiber effects, Cox and
others modified the basic model by a so-called shear-lag theory. However,
the latter assumes single-fiber segments to carry axial and shear forces only,
which (see Section 4.2), is not a valid model of a solid element: fiber bending
should also be included.
Paper exhibits finite stiffnesses in 2D as well as in 3D. In the latter case,
condition (4.1) is replaced by an even more stringent one as more constraints
are needed when dealing with the additional degrees of freedom (Asimov
and Roth, 1978).
4.1.2.2 Loss of Rigidity in a Fiber-Beam Network

Besides the foregoing structural topology considerations, there is another fact
that casts doubt on any fiber network model in which fiber segments are
joined by pivots. Namely, any two cellulosic fibers have a finite contact area of
hydrogen bonding (Page et al., 1961), which would be sheared by hinge-type
connections. While it is very difficult to assess experimentally to what extent
this region is deformable, our model will treat it as somewhat deformable in
the sense that bonds are rigid but have no dimension, and fiber segments are
treated as extensible beams from node to node of the graph G(V, E) (Ostoja-
Starzewski et al., 1999).
This modeling of mechanics of fiber networks is similar to that of a cement-
coated wood strands composite (Stahl and Cramer, 1998), as well as to the
one used for highly porous materials (Chung et al., 1996), and is based on the
following assumptions and steps:
1. Generate a system of finite-length straight fibers, such as shown in

Figure 4.3(b) according to specific geometric characteristics: distri-
bution of fiber lengths and widths, distribution of angular orienta-
tions () of fiber chords, etc. The fibers are laid in three dimensions
on top of one another with a possible, slight non-zero out-of-plane
angle. [0, ], that is, the angle a fiber makes with respect to
the x-axis is modeled by the Fourier series-type probability density
function of Chapter 1.
2. Fibers are homogenous, but each fiber may have different dimen-
sions and mechanical properties, all sampled from any prescribed
statistical distribution.
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
3. Each fiber is a series of linear elastic 3D extensible Timoshenko beam

elements. Each of these is described by a stiffness matrix written here
in an abbreviated form setup in a corotational coordinate system
(Cook et al., 1989):
EA
0 0 0 0 0
F l L

0

T
GJ
0 0 0 0
x

l

Ma l +3g
2
l 6g
2

0 0 0 0 a
y
= 4a 2a y
l +3h l 6h a (4.4)
Mz
2 2
z
a
0

0 0 4b
0 2b

My

b 0 0 l 2
6g
0 l 2
+3g
0 yh

b 2a 4a
b
Mz 0 0 0 l 6h
2
0 l +3g
2
z
2b 4a
where
E Iy E Iz E Iy E Iz
g = 12 h = 12 a= b= .
GA GA l(12g + l 2 ) l(12h + l 2 )
(4.5)
Here F and T are the axial force and the twisting moment, while Mya ,
Mza , Myb , and Mzb are the bending moments around the y and z axes
at the a and b ends, respectively. Also, L, x , ya , za , yb , and zb
denote axial elongation, angle of twist, and four angles of rotation.
Finally, l, A, J , Ix , and I y are, respectively, the length, cross-sectional
area, cross-sectional polar moment of inertia, and the moments of
inertia with respect to the x- and y-axes. E and G are the Young
modulus and shear modulus of a fiber-beam.
4. After identifying all the fiberfiber intersection/contact points a
connectivity matrix is set up.
5. Equilibrium is found under the boundary condition ui = ij0 x j .
6. All six effective, in-plane stiffness coefficients are determined from
the postulate of equivalence of strain energy stored in a square-
shaped window of finite thickness with the strain energy of an
equivalent continuum.
The undeformed network, shown in Figure 4.3(b) in its top view, has
the following parameters: window size: 4 4 0.1 mm, a 1 = 1, and other
coefficients in equation (4.4) are zero; fiber length: 2 mm; fiber width: 0.04 mm;
fiber height: 0.015 mm. As a result of a Boolean process of fiber placement
(Chapter 1), we obtain: 195 fibers with an average of 4.8 bonds per fiber, the
whole system having 859 nodes with six degrees of freedom per node.
The state of deformation corresponding to axial strain 11 = 8% is shown
in Figure 4.3(c). The analyzed strain is actually 1%, but displacements are
magnified for clarity. Compare this deformed network to that in Figure 4.3(d),
which shows the same network of fibers subjected to the same strain but with
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
the ratio of fiber flexural stiffness to fiber axial stiffness reduced by a factor of
104 . Note the following:
1. The sharp kinks we see in both figures are only the artifact of
simple computer graphicsthe micromechanical model assumes
fibers deform into differentiable curves. Magnification creates the
appearance of large displacementsactually, an infinitesimal dis-
placement assumption is used in the computational mechanics pro-
gram.
2. The kinks are far more pronounced when fibers have low flexural
stiffness. Portions of the network where connected fibers do not
form triangular pores can generate significant forces in response
to deformation when fibers have high flexural stiffness, but they
cannot do so when fibers rely almost entirely on axial stiffness. These
portions of the network are not stable in the sense of loss of generic
rigidity discussed earlier.
3. We do not study this rigid-floppy transition by turning, in an ad
hoc fashion, all the connections into pivots. Rather, with the model
taking into account all the displacements and rotations of nodes, we
can study it as a continuous function of fiber slenderness; see also
Kuznetsov (1991). Note that this aspect is impossible to
investigate with models based on central-force potentials for single
fiber segments (e.g., Kellomaki et al., 1996). Our model also fills
a gap pointed out in Raisanen et al. (1997) consisting of a need
to set up finite element models of 3D disordered fiber networks,
yet avoids their simplistic mapping into electrical resistor networks
(i.e., second-rank tensor problems) of the same topology.
The fiber network model has also found an application in studies of

mesoscale stiffness and instability (splitting) of fibril networks found in a
bones lamellar structure (Jasiuk and Ostoja-Starzewski, 2004), see Figure 7.15
of Chapter 7.
4.2 Spring Network Models for Disordered Topologies

4.2.1 Granular-Type Media
4.2.1.1 Load Transfer Mechanisms
The spring network models are most natural when applied to systems that
have the same topology as the underlying lattice. One such example has
been discussed above: a cellulose fiber network. Another one is offered by a
granular medium. Here, the principal method of computational mechanics
analyses, dating back to Cundall and Strack (1979), is the so-called discrete
element (DE) model.
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
2 V2
1 V1
3 V3
(a) (b)
V2 V2
kn
ka ka
V1 V1
kn
V3 V3
(d) (c)
FIGURE 4.4
(a) A cluster of three grains (1, 2, 3) showing the three lines of interactions; (b) a discrete element
model showing the normal force, the shear force and the moment exerted by grains 2 and 3
onto grain 1; (c) a most general model showing the same grain-grain interactions as before but
augmented by an internal, angular spring constant k a ; a simplified model showing only normal
(k n ) and angular (k a ) effects.
Let us employ a graph representation of the planar granular medium: a

graph G(V, E), whereby vertices of the set V signify grain centers and edges
of the set E represent the existing graingrain interactions, Figure 4.4(a). We
fix an r polar coordinate system at a grain center. There are several types
of the DE model that one may consider:
Central interactions: the total energy is a sum total of central inter-

actions of all the edges
U = U cental , (4.6)
and this model is a generalization of the basic model of Section 3.2.3
in Chapter 3.
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
Central and angular interactions: the total energy is modified to
U = U cental + U angular , (4.7)
and this model is a generalization of what we saw in Section 3.2.4 in

Chapter 3. Continuum mechanics tells us that equals r 1 u / +
ur /r . This shows that the angular changes / between two adja-
cent edges V1 V2 and V1 V3 in Figure 4.4 correspond to r 1 u /
in the r polar coordinate system fixed at a grain center. This term
does not show up in two other expressions for rr and r . However,
ur /r is due to a radial displacement, and so / does not exactly
equal , which leads us to call it a -type strain. We now adopt
the Kirkwood model (recall Chapter 3) to account for / in addi-
tion to the normal graingrain interactions, and introduce angular
springs of constant k a acting between the edges V1 V2 and V1 V3
incident onto the node V1 , Figure 4.4(d); the edges remain straight
throughout deformation.
Central, shear, and bending interactions: the total energy is
U = U cental + U shear + U moment , (4.8)
and this model is a generalization of what we saw in Sections 3.4.1

to 3.4.2 of Chapter 3. This is a typical DE model, which, of course,
may be termed a locally inhomogeneous micropolar continuum,
with inhomogeneity varying on the scale of grains; see the section
below.
Central, shear, bending, and angular interactions: the total energy is
U = U cental + U shear + U moment + U angular . (4.9)
One may argue that the three-point interaction should be introduced in

the DE models so as to better represent the micromechanics, and to make,
in accordance with Figure 4.4(c), the strain energy stored in a single Voronoi
cell equal to (4.9). However, there exist successful DE models that account for
normal and shear forces only (Bathurst and Rothenburg, 1988a,b); this neglect
of the contact moment is justified by the fact that only small numerical errors
are thus caused in problems of interest in granular materials. In the case of a
regular, triangular array of disks, this model is equivalent to a classical Born
model of crystal lattices, which is known to lack the rotational invariance
(Jagota and Bennison, 1994).
4.2.1.2 Graph Models

Let us pursue the planar graph representation of granular media in some more
detail (Satake, 1976, 1978). First, we list in Table 4.2 a correspondence between
a system of round grains and its graph model. Besides the vertex (V) and edge
(E) sets introduced earlier, we also have a loop set L. With this geometric
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
TABLE 4.2
Granular Structure and Graph Terminology
Assembly of grains Graph Index No. of elements
Grain Vertex v |V|

Contacting point Edge e |E|
Void (in 2D) Loop l |L|
Note: After Satake (1978).
reference, we can then set up an assignment of load quantitiesforces on the

left and corresponding kinematic measures on the rightin Table 4.3.
The connectivity of the graph is described by the incidence matrix Dve .
Let us write down a total of 3|V| scalar equilibrium equations, each one with
respect to a typical grain of radius r v and volume V v
e
F Bv
D ve
+V v
= 0, (4.10)
Me Nv
where

0 Dve
ve
D = v ve e . (4.11)
r D n Dve
Here ne is the unit vector of edge e in the nonoriented graph. The operator
Dve (and its dual Dev mapping from vertices into edges) also plays a key role
in the kinematics of all the edges:

ue uv
= D ev
= 0, (4.12)
we wv
where

Dev ne Devr v
ev =
D . (4.13)
0 Dev
The kinematics is subject to 3|L| compatibility constraints written for all the
loops, where we make a reference to Satakes work.
TABLE 4.3
Load Versus Kinematic Quantities
Load quantity Notation Number of elements Notation Kinematic quantity
Body force Bv |V| uv Grain displacement

Body couple Nv |V| wv Grain rotation
Contact force Fe |E| ue Relative displacement
Contact couple Me |E| we Relative rotation
Note: After Satake (1978).

P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
The above should be augmented by 3|E| constitutive equations connecting

the contact force Fe and contact moment Me with the relative displacement
ue and relative rotation we . Given three global equilibrium conditions, we
have a total of
3(|V| 1 + |E| + |L|) = 6|E| (4.14)
equations. Taking note of the Euler formula |V| |E| + |L| = 1 in 2D (as
opposed to that in 3D in Section 1.4 of Chapter 1), we see that this budget of
equations agrees with the total of 6|E| unknowns: Fe , Me , ue , and we , all
defined on edges of the set E.
Finally, we note a formal analogy of (4.10) and (4.12) to the equilibrium
and strain-displacement equations of Cosserat continua (Chapter 6)

b u
div + =0 = grad . (4.15)
m w
The similarity of compatibility relations for graph and continuum descrip-

tions has also been shown by Satake (1976, 1978). See Goddard (2006) for an
in-depth analysis of these issues.
Applications of graph models to cell biomechanics have been explored
in Hansen et al. (1996, 1997). While this section focuses on classical contin-
uum modeling, multifield models for granular-type materials are analyzed
in Trovalusci and Masiani (2005).
4.2.1.3 Periodic Graphs with Disorder

Randomness may be introduced into the periodic networks in various ways.
Figure 4.5 displays two basic possibilities: substitutional disorder and topolog-
ical disorder. The first of these connotes a variability in properties per vertex
(or node), and the second consists in a departure from the periodic topol-
ogy. There is also a third case, of much more interest in solid-state physics:
geometric disorder, which involves the variability in the geometry of a net-
works structurelike uneven lengths of various bondsbut preserving a
(a) (b)
FIGURE 4.5
Substitutional (a) versus topological disorder (b) of a hard-core Delaunay network.
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
topological periodic structure (Ziman, 1979). The topological disorder is typ-

ically caused by an incompatibility of crystal-like domains in a granular ma-
terial. For example, the material may consist of equisized disks, which are
organized into regular, periodic arrays, but the fact that they happen to be
differently oriented in space causes an irregular structure and network of
domain boundaries.
As observed earlier in connection with the DE model carrying torques,
the topological disorder leads to a locally inhomogeneous polar, or micropo-
lar, continuumdepending on the type of vertexvertex interactionswith
inhomogeneity varying on the scale of grains. Such a continuum model con-
tains a lot of information, but, in the first place, one wants to establish the
eff
effective, in the macroscopic sense, first-order moduli Ceff ( Cijkl ) of the mate-
rial. These are obtained from the so-called periodic boundary conditions on an
L L square B:
ui (x + L) = ui (x) + ij0 L j ti (x) = ti (x + L) x B. (4.16)
Here ij0 is the macroscopic strain, ti is the traction on the boundary B of B,

and L = Lei , with ei being the unit base vector. The periodicity means that
the network topology is modified so as to repeat itself with some periodicity
L in x1 and x2 directions, whereby L is usually taken much larger than the
typical vertex-vertex spacing (or edge length).
Now, the periodic conditions (4.16) require that a periodic network be
set up, and this, in turn requires a periodic Poisson point field on the L L
square; recall Chapter 1. Topologically, our square turns into a torus, but, as
Figure 4.6 shows, there are also three other possibilities.
B B A A
Cylinder
Torus
A A A A
(a) (c)
B A A A
Mbius band
Klein bottle
A B A A
(b) (d)
FIGURE 4.6
Mapping of edges of a a square-shaped domain resulting in (a) a cylinder, (b) a Mobius band,
(c) a torus and (d) a Klein bottle. The torus corresponds to periodic boundary conditions.
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
FIGURE 4.7
A periodic PoissonDelaunay network with 200 vertices.
A typical realization B() of a periodic PoissonDelaunay network, num-

bering 200 vertices, is shown in Figure 4.7. The set B ={B(); } forms
the random medium; a single indicates one realization of the Poisson point
field and a chosen assignment of spring constants. In actual simulations only
a minute subset of the entire sample space can be investigated, but by the
standard Monte Carlo and ergodicity arguments, this subset is representative
of the whole system. Thus, already the response of a single network much
larger than the grain size is sufficient to gain a good estimate of Ceff . The
ensemble average of that tensor is isotropic for a microstructure of space-
homogeneous and isotropic statistics, but, with the number of vertices large,
even one realization of the network should be close to isotropic.
Using the formula for the strain energy of a 2D elastic continuum of volume
V = L 2:

V 1 0 0
U= ii jj + 2 ij ij ii jj ,
0 0 0 0
(4.17)
2 2
leads, with reference to Chapter 5, to planar bulk and shear moduli:
2U(1) 2U(2)
= = 0 2 . (4.18)
0 2
V 11
0
+ 22 V 12
0
Here U(1) and U(2) denote energies under dilatational (11 = 22
0
) and shear
0
(12 ) strains, respectively. The moduli computed this way were compared to
various models in Ostoja-Starzewski et al. (1995):
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
A symmetric self-consistent approximation for elliptical disks with

perfectly bonded interfaces, which treats the medium as a mixture
of N phases without distinguishing any one as a matrix.
An asymmetric self-consistent approximation for elliptical disks
with perfectly bonded interfaces, which treats the medium as a mix-
ture of N 1 phases embedded in a matrix.
An asymmetric self-consistent approximation for circular disks with
perfectly bonded interfaces.
An asymmetric self-consistent approximation for circular disks with
springy interfaces.
Voigt and Reuss bounds for circular disks with springy interfaces.
Comparisons to such continuum models are possible providing one inter-

prets a vertex and all the edges incident onto it as a spider, which allows
interpretation of that object as a continuum-type inclusion. As a result we
have a spider-inclusion analogy of Figure 4.8.
4.2.2 Solutions of Truss Models

A specialized method of determination of apparent moduli applies in the case
of truss-type microstructures. Thus, let us focus on the response of a window,
arbitrarily larger than the average edge length, subjected to displacement
boundary conditions, Figure 4.9 (Ostoja-Starzewski and Wang, 1989). First,
a window is cut out of the network as indicated by a square with dashed
lines in Figure 4.9(a), and treated as an independent body (b)in this case,
given the assumption of central forces in all the Delaunay edges, a planar
truss. At this stage, all the boundary points (on four sides) of the window
are pinned, and this is where the displacement condition is appliedsee
equation (4.23) below. For example, a state of uniaxial extension and shear is
depicted in Figure 4.9(c). The actual mechanical response of this planar truss
is now solved by a structural mechanics method.
FIGURE 4.8
A periodic PoissonDelaunay network with 200 vertices, showing a spider of edges incident
onto a vertex in the figure on the right.
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
(a) (b) (c)
FIGURE 4.9
A finite window (b) cut out of a Delaunay network (a), and subjected to uniform displacement
boundary conditions in (c).
We introduce a vector notation, so that the effective mechanical response

of the window is described by
0
11
C11 C12 C13 11

22 { } = [C]{ } C21 C22 C23 22
0 0
. (4.19)

0

12 C31 C32 C33 12
The structural mechanics of the truss is governed by
[K ]{u} = {F }, (4.20)
where [K ], {u}, and {F } are the stiffness matrix of the system and the dis-
placement and force vectors at the nodes, respectively. Noting that the set of
global equilibrium equations governing all the elements may be partitioned
into those corresponding to the degrees of freedom at the window boundaries
(b) and the ones in the interior (i), we have

K (ii) K (ib) u(i) F (i)
[K ] = , {u} = , {F } = . (4.21)
K (bi) K (bb) u(b) F (b)
The net force on any interior node (and, thus, any interior degree of freedom)
must be zero, F (b) = 0, so that the static condensation gives
[K ]{u(b) } = {F (b) } [K ] = [K (bb) ] [K (bi) ][K (ii) ]1 [K (ib) ]. (4.22)
For any boundary node the displacement can be written as

0
11
u1(b) x1 0 x2 /2 0
= [A]{} 22 , (4.23)
u2(b) 0 y2 x1 /2 0
12
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
while for an interior node we have
[u(i) ] = [K (ii) ]1 [K (ib) ][A]{ 0 }. (4.24)
Now, noting that the strain energy of the window is

1 (b) T (b) 1
U= {u } {F } = {u(b) }T [K ]{u(b) }, (4.25)
2 2
we obtain, by equivalence with the continuum model, the effective moduli
1
[C] = [A]T [K ][A], (4.26)
V
where the strain-displacement matrix [ A] is obtained through the assembly
of the nodal matrices for all the boundary nodes.
In the case of traction boundary conditions
{F (b) } = [H]{ }, (4.27)
a force-based homogenization technique has been proposed (Huyse and Maes,
2001). Here [H] is a matrix that depends on the actual boundarys microge-
ometry. The structural mechanics of the truss is now rewritten as

f (ii) f (ib)
[ f ]{F } = {u} [ f ] = , (4.28)
f (bi) f (bb)
where [ f ] is the flexibility matrix and the strain energy of the window is
1 (b) T (b) 1
U= {u } {F } = {F (b) }T [ f (b) ]{F (b) }. (4.29)
2 2
From the above equations we obtain, by equivalence of the discrete with the
continuum model, the apparent compliances
1
[S] = [H]T [ f (bb) ][H], (4.30)
V
where [ f (bb) ] is obtained through an inversion.
4.2.3 Mesoscale Elasticity of Paper

4.2.3.1 Dilemma of Special In-Plane Orthotropy of Paper
Paper is one of the most challenging engineered materials. The difficulties in
understanding it are due to its complex multiscale structurethis necessitates
random fields of Chapters 1 and/or 2, random network models of this chapter,
(non)classical elasticity of Chapters 5 and 6, as well as micromechanics models
discussed here and also later. Things get still more complicated when the
inelastic behavior is considered, see Chapter 9.
For the sake of reference, because we are dealing with the machine-made
(i.e., oriented) paper, for the linear elastic tensile range, an orthotropic model
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
is adopted (e.g., Uesaka et al., 1979):

E1 E 2 12
11 = 11 + 22
(1 12 21 ) (1 12 21 )
E 1 21 E2
22 = 11 + 22
(1 12 21 ) (1 12 21 )
12 = G12 = 2G12 (4.31)
where the engineering constants are:
E 1 = Youngs modulus in x1 -direction (MD)

E 2 = Youngs modulus in x2 -direction (CD)
22
12 = is the Poisson ratio for strain in x2 -direction (CD) when
11
paper is stressed in x1 -direction (MD) only
11
21 = is the Poisson ratio for strain in x1 -direction (MD) when
22
paper is stressed in x2 -direction (CD) only
G = shear modulus in x1 x2 -plane
Here MD (x1 ) and CD (x2 ) are the so-called machine- and cross-directions,
respectively. To ensure that the stiffness matrix is orthotropic, we require
E 1 21 = E 2 12 . (4.32)
The shear modulus G is very difficult to measure by quasi-static experiments,

but ultrasonic experiments on many paper materials indicate that the follow-
ing relation is approximately satisfied (Horio and Onogi, 1951; Campbell,
1961):
E1 E2
G= . (4.33)
E 2 + E 1 + E 1 21 + E 2 12
Thus, paper has a special in-plane orthotropy property where only three elas-
tic constants are independent. This special orthotropy means that the shear
modulus is (approximately) invariant with respect to rotations of the coordi-
nate system (Horio and Onogi, 1951; Campbell, 1961). Another way to express
this invariance, fully equivalent to (4.33) above, is in terms of the in-plane
compliances
4S1212 = S1111 + S2222 S1122 . (4.34)
Note here that approximately reflects the fact that the compliance S is an
apparent (i.e., mesoscale) property, typically measured on scales of centi-
meters, and being actually random since we are below the RVE.
Note: Given the multiscale structure of paper shown in Figure 4.10, there
is no clear length scale at which homogenization to a perfect RVE can be
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
ZD
(a)
CD MD
(b)
(c)
P S1 S2 S3
35A (d)
100A
(e) H
CH2OH H1 OH
H H
CH2OH H OH
H
O O O O
H OH H H OH H
OH H H OH H H
O H H O O H H O O
H OH CH2OH H OH CH2OH
(f )
FIGURE 4.10
A hierarchy of scales in paper: (a) roll of paper on a paper machine (up to 10 m wide, thousands
of kilometers long) with a possible presence of streaks; (b) paper sheet (scale of centimeters);
(c) random fiber network (scale of millimeters); (d) cellulose fiber with its layers P, S1, S2, S3
(all made of fibrils) and the lumen (microns); (e) fibrils; (f) molecular chains.
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
carried out; see the discussion of RVE in Chapters 7 to 10. Although the con-
stitutive law above applies to length scales of centimeters, specimens display
some scatter there. Going to scales of tens of centimeters, one begins to see
streaks, while on the scales of meters there is a strong dependence on the
cross-direction of the paper machine. There are also complex fluctuations in
the machine direction, which do not qualify as a WSS random field.
Staying on the scale of centimeters, the question becomes: Why does the
relationship (4.34), in fact hardly found in other materials, occur in paper?
4.2.3.2 Explanation via Random Fiber Network

The above question is addressed by Ostoja-Starzewski and Stahl (2000) with
the help of models introduced earlier. First, we employ the germ-grain model
of Chapter 1, constructed so as to obtain nonuniform fiber flocculation, com-
monly seen as spatially nonuniform grayscale effects in a typical sheet of
paper held against light. Next, we take the computational mechanics model
of fiber networks of Chapter 3. In brief, fibers are placed in 3D with possible
non-zero angles to control out-of-plane orientation of the fiber axis and the
roll of the fiber about its own axis. Each fiber, depending on its contacts
with other fibers, is a series of linear elastic three-dimensional Timoshenko
beam elements e, also with torsional response included, between bonds with
other fibers. In the analysis of each and every body B (), having solved for
displacements and rotations of all the bonds under uniform kinematic bound-
ary conditions ui = ij0 x j , we establish the networks effective stiffness tensor
Cijkm () from a postulate of equivalence of the total strain energy of all the
networks elements e, Utot , with that of an effective continuum (V = L x L y t)
V
Utot = ij Cijkm ()km = [Ueaxial + Ueshear + Uemoment + Uetorsion ]. (4.35)
2 eE
It may be argued, by virtue of a qualitative analogy to in-plane conductiv-

ity of networks of identical geometry, that the kinematic boundary condition
allows a much faster asymptotic scaling to the RVE than the uniform traction
boundary condition ti = ji0 n j . The point is that fiber networks are somewhat
analogous to plates with holes, reported in Section 3.4.3 of Chapter 3, recall
Figure 3.11 there.
Let us now go back to Figure 1.11 of Chapter 1. It displays two networks
differing only by the degree of flocculation. That is, parameter b equals 2.0
in (a) and 0.4 in (b), so that we have a highly versus a weakly flocculated
network. Their responses under overall 11 0
-strain are shown in the middle
figures, while the figures on the right show the difference between the result-
ing displacement of each node and what the displacements would be if the
strain field in the interior of the network had been uniform. In the latter case,
the figures would consist simply of dots; thus, the lines represent deviation
from uniformity. Let us first note that the deviation from uniform displace-
ments is apparent in both networks, but it is certainly higher in the network
with b = 2.0 (strong flocculation) than with b = 0.4. (weak flocculation).
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
This is so because fibers belonging to a given floc move together, and this
motion, being different from the affine one, resembles a swirl.
Now, it turns out that the special orthotropy relation (4.34) is satisfied
for a network with b between the values employed here. Various parameters
of the network may be altered, but some degree of flocculationneither too
much nor too littleis always necessary. [In our original paper the factor 4
in equation (4.34) was inadvertently missed, but this had no effect on the
results since we worked with G in computational mechanics.] To sum up,
this analysis shows that random network geometry involving two scales
(1) random fiber placement within a floc with random angular orientation
and (2) flocculation modeled by the germ-grain processtogether with a
network model possessing generic rigidity (thanks to beam-type fibers) offers
an explanation of the peculiar property of many papers.
4.2.4 Damage Patterns and Maps of Disordered Composites

As mentioned in Chapter 3, the lattice method can also be used to simulate
damage of heterogeneous materials. This works particularly well in the case of
elastic-brittle failure of composites, where one uses a mesh (much) finer than
the typical size of the microstructure. In principle, one needs to determine
which lattice spacing ensures mesh-independence, or nearly so. Such a study
has been conducted for a thin aluminum polycrystalline sheet discussed in
Section 3.3 of Chapter 3.
Following Ostoja-Starzewski et al. (1997) and Alzebdeh et al. (1998), we
now focus on two-phase composites in antiplane shear, under periodic bound-
ary conditions and (necessarily) periodic geometries. Now, since both phases
(inclusion i and matrix m) are isotropic and elastic-brittle (Figure 4.11a), the
C i/C m
Sti Sti
i
C weak strong
Cm
icr /m
cr
icr m Sti Soft

cr
weak strong
(a) (b)
FIGURE 4.11
(a) Elastic-brittle stress-strain curves for matrix and inclusion phases; (b) sketch of the damage
plane.
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
10.0
C i/ C m
1.0
0.1
0.1 1.0 10.0

i m
/
FIGURE 4.12
Crack patterns in the damage plane on a scale 4.5 times larger than that of the inclusion diameter.
The center figure of a homogeneous body is not shown as it corresponds to all the bonds failing
simultaneously.
composite can be characterized by two dimensionless parameters
cr
i
/cr
i
C i /C m , (4.36)
where cri
is the strain-to-failure of either phase, and C its stiffness. This leads
to the concept of a damage plane (Figure 4.11b), where we see various combi-
nations of strengths and stiffnesses. While the response in the first and third
quarters of damage plane is quite intuitive, this is not so for the second and
fourth quarters. In those two quarters there is a competition of either high
stiffness with low strength of the inclusions with the reverse properties of the
matrix, or the opposite of that. The damage plane is useful for displaying ef-
fective damage patterns of any particular geometric realization of the random
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
/m
/m
/cr
cr
cr
10.0
/m
cr /m
cr /m
cr
m
/m
/cr
cr
m
1.0
C /C
i
m m
/cr /cr
m
/m
/m
/cr
cr
0.1 cr
m m m
/cr /cr /cr
0.1 1.0 10.0

i m
/
FIGURE 4.13
Damage maps of statistics of constitutive responses for twenty realizations of the random com-
posite, such as that in Figure 4.12.
composite while varying its physical properties (Figure 4.12) as well as other
characteristics, say, statistics of response in the ensemble sense (Figure 4.13).
A number of other issues are studied in the referenced papers:
Stress and strain concentrations

Finite size scaling of response
Function fitting of statistics (where it turns out that the beta proba-
bility distribution offers a more universal fit than either Weibull or
Gumbel)
Effects of disorder versus periodicity
Also, see Ostoja-Starzewski and Lee (1996) for a similar study under in-plane
loading; computer movies of evolving damage are at http://www.mechse.
uiuc.edu/research/martinos.
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
4.3 Particle Models

4.3.1 Governing Equations
Particle models are a generalization of lattice models to include dynamic
effects, and can also be viewed as an offshoot of molecular dynamics (MD).
The latter field has developed over the past few decades in parallel with the
growth of computers and computational techniques. Its objective has been to
simulate many interacting atoms or molecules in order to derive macroscopic
properties of liquid or solid materials (Greenspan, 1997, 2002; Hockney and
Eastwood, 1999). The governing Hamiltonian differential equations of motion
need to be integrated over long time intervals so as to extract the relevant
statistical information about the system from the computed trajectories.
Techniques of that type have been adapted over the past two decades
to simulate materials at larger length scales, whereby the role of a particle
is played by a larger-than-molecular piece of material, a so-called particle or
quasi-particle. The need to reduce the number of degrees of freedom in complex
systems has also driven models of galaxies as systems of quasi-particles, each
representing lumps of large numbers of stars. In all these so-called particle
models (PM), the material is discretized into particles arranged in a periodic
lattice, just like in the spring network models studied in earlier sections, yet
interacting through nonlinear potentials, and accounting for inertial effects,
that is, full dynamics. As shown in Figure 4.14, the lattice may be in 2D or
in 3D.
Note that, by comparison with finite elements (FE), which indeed also
involve a quite artificial spatial partitioning, PMs are naturally suited to in-
volve interparticle potentials of the same functional form as the interatomic
potentials, providing one uses the same type of lattice. The PM can therefore
take advantage of the same numerical techniques as those of MD, and rather
easily deal with various highly complex motions. Thus, the key issue is how
to pass from a given molecular potential in MD to an interparticle potential
in PM. In the case when the molecular interactions are not well known, the
PM may still turn out to be superior relative to the FE. Among others, this
indeed is the case with comminution of minerals where scales up to meters
are involved (Wang and Ostoja-Starzewski, 2006; Wang et al., 2006).
In MD, the motion of a system of atoms or molecules is governed by clas-
sical molecular potentials and Newtonian mechanics. As an example, let us
consider copper. Following Greenspan (1997), its 612 Lennard-Jones poten-
tial is
1.398068 10 1.55104 8
(r ) = 10 + 10 erg. (4.37)
r6 r 12
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
(a)
Z
Y
X
4
Z (cm)
0 2
m)
0
X (c
2
6 4 2 0 2 4 4
6
Y (cm)
(b)
FIGURE 4.14
Particle models and intermediate stages of fracture in (a) 2D and (b) 3D. (After Wang et al. [2006]).
Here r is measured in A. It follows that the interaction force between two

copper atoms is
d(r ) 8.388408 2 18.61248
F (r ) = = 10 + dyn. (4.38)
dr r7 r 13
In (4.38) F (r ) = 0 occurs at r0 = 2.46A, and then attains the minimum:
(r0 ) = 3.15045 1013 erg.
Let us next recall Ashby and Joness (1980) simple method to evaluate
Youngs modulus E of the material from (r ): compute
S0 d 2 (r )
E= where S0 = |r . (4.39)
r0 dr 2 0
With this method, we obtain Youngs modulus of copper as 152.942 GPa,
a number that closely matches the physical property of copper and copper
alloys valued at 120150 GPa. Ashby and Jones also defined the continuum-
type tensile stress
(r ) = NF (r ), (4.40)
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
where N is the number of bonds/unit area, equal to 1/r02 . Tensile strength,

TS , results when d F (r )/dr = 0, that is, at rd = 2.73A (bond damage spacing),
and yields
TS = NF (rd ) = 462.84 MN/m2 . (4.41)
This value is quite consistent with data for the actual copper and copper-based
alloys reported at 2501000 MPa.
In the PM, the interaction force is also considered only between nearest-
neighbor (quasi-)particles and assumed to be of the same form as in MD
G H
(r ) = p
+ q. (4.42)
r r
Here G, H, p, and q , all positive constants, are yet to be determined, and this
will be done below. Inequality q > p must hold so as to obtain the repulsive
effect that is necessarily (much) stronger than the attractive one. Three
examples of interaction force for three pairs of p and q are displayed in
Figure 4.15(a). The dependence of Youngs modulus for a wide range of p
and q is shown in Figure 4.15(b).
The conventional approach in PM, just as in MD, is to take the equation
of motion for each particle Pi of the system as

d 2 ri Gi Hi rji
mi 2 = p + q i=
j, (4.43)
dt j
rij rij rij
where mi is the mass of Pi and rji is the vector from P j to Pi ; summation is

taken over all the neighbors of Pi . Also, is a normalizing constant obtained
by requiring that the force between two particles must be small in the presence
of gravity:

Gi Hi

p + q < 0.001 980mi . (4.44)
D D
Here D is the distance of local interaction (1.7r0 cm in this particular example),

where r0 is the equilibrium spacing of the particle structure. The reason for
introducing the parameter by Greenspan (1997) was to define the interaction
force between two particles as local in the presence of gravity. However, since
setting according to (4.44) would result in a pseudo-dynamic solution,
we set = 1.
According to equation (4.42), different ( p, q ) pairs result in different
continuum-type material properties, such as Youngs modulus E. Clearly,
changing r0 and volume of the simulated material V (= A B C) will
additionally influence Youngs modulus. Therefore, in general, we have some
functional dependence
E = E( p, q , r0 , V). (4.45)
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
20
10
Interaction force (KN)
10
(p, q) = 3, 5
(p, q) = 5, 10
20 (p, q) = 7, 14
30
40
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Equilibrium position r0 (cm)
(a)
15
14
13
12
11
10
9
q
8
7
6
5
4
3
2
1 2 3 4 5 6 7 8 9 10 11 12 13 14
p
Youngs modulus (GPa)
14 37 59 82 104 127 149 172 195 217 240 262 285 307 330
(b)
FIGURE 4.15
(a) The interaction force for pairs of ( p, q ) exponents, at r0 = 0.2 cm. (b) The variability of Youngs
modulus in the ( p, q )-plane.
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
In Wang and Ostoja-Starzewski (2005) we formulated four conditions to deter-

mine continuum-level Youngs modulus and tensile strength, while maintain-
ing the conservation of mass and energy of the particle system and satisfying
the interaction laws between all the particles in the PM model for a given MD
model.
4.3.1.2 Leapfrog Method

Just like in MD, there are two commonly used numerical schemes in parti-
cle modeling: the completely conservative method and the leapfrog method. The
first scheme is exact in that it perfectly conserves energy, linear, and angular
momentum, but requires a very costly solution of a large algebraic prob-
lem. The second scheme is approximate. Since in most problems one needs
large numbers of particles to adequately represent a simulated body, the com-
pletely conservative method is unwieldy and, therefore, usually abandoned in
favor of the leapfrog method (Hockney and Eastwood, 1999). That method
is derived by considering Taylor expansions of positions ri,k+1 and ri,k of the
particle Pi (i = 1, 2, . . . , N) at times tk = kt and tk+1 = (k +1)t, respectively,
about time tk+1/2 = (k + 1/2)t (with t being the time step):
t t 2 t 3
ri,k = ri,k+1/2 vi,k+1/2 + ai,k+1/2 ai,k+1/2 + O(t 4 )
2 8 48
t t 2 t 3
ri,k+1 = ri,k+1/2 + vi,k+1/2 + ai,k+1/2 + ai,k+1/2 + O(t 4 ). (4.46)
2 8 48
Here vi and ai denote velocity and acceleration. Upon addition and subtrac-
tion of these we get the new position and velocity
t 2
ri,k+1 = 2ri,k+1/2 ri,k + ai,k + O(t 4 )
4
vi,k+1/2 = (ri,k+1 ri,k )/t + O(t 2 ), (4.47)
which shows that the position calculation is two orders of magnitude more
accurate than the velocity calculation. However, the error in computation
of velocity accumulates only as fast as that in position because it is really
being calculated from positions. It is easy to see that the leapfrog method
is more accurate than the conventional Euler integration based on vi,k+1 =
vi,k + (t)ai,k and ri,k+1 = ri,k + (t)vi,k .
Often the leapfrog formulas relating position ri , velocity vi , and accelera-
tion ai for all the particles Pi (i = 1, 2, . . . , N) are written as
t
vi,1/2 = vi,0 + ai,0 (starter formula)
2
vi,k+1/2 = vi,k1/2 + (t)ai,k k = 0, 1, 2, . . . (4.48)
ri,k+1 = ri,k + (t)vi,k+1/2 k = 0, 1, 2, . . . .
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
Clearly, the name of the method comes from taking velocities at intermediate
time steps relative to positions and accelerations; it is also known as a Verlet
algorithm.
It can be shown that the global (cumulative) error in position going from
ri,k to ri,k+n (i.e., over T = nt) of Pi is
error(ri,k+n ri,k ) = O(t 2 ). (4.49)
By the argument following (4.46) above, the global error in velocity is also
O(t 2 ).
Stability is concerned with the propagation of errors. Even if the trunca-
tion and round-off errors are very small, a scheme would be of little value if
the effects of small errors were to grow rapidly with time. Thus, instability
arises from the nonphysical solution of the discretized equations. If the dis-
crete equations have solutions that grow much more rapidly than the correct
solution of the differential equations, then even a very small round-off error
is certain to eventually seed that solution and render the numerical results
meaningless. By the root locus method for an atomistic unit of time, the safe
time-step used in the leapfrog method meeting this requirement is
1/2
1 dF
t 2 = . (4.50)
m dr max
We see that, as r 0, d F /dr , which results in t 0. Because this
may well cause problems in computation, we introduce the smallest distance
between two particles according to these conditions:
1. For a stretching problem of a plate/beam, take (dF/dr) max dF/

dr|r =r0 , which with (4.50) dictates t 107 106 s.
2. For an impact problem, one often needs to set up a minimum dis-
tance limiting the spacing between two nearest particles, e.g., rmin =
0.1r0 . It is easy to see from Figure 4.15(a) that, in this case, this suit-
able time increment is greatly reduced because of a rapid increase
in . This leads to t 108 s.
Following the MD methodology (Napier-Munn et al., 1999), we can also

set up a criterion for convergence: t < 2 m/k, where m is the smallest mass
to be considered, k is the same stiffness as S0 in (4.39)2 . An examination of
these two criteria shows there is not much quantitative difference between
them in the case of elastic or elastic-brittle, but not plastic, materials.
4.3.2 Examples
The maximum entropy formalism of Section 2.6 in Chapter 2 is much more
suited to deal with quasi-static rather than dynamic fracture. The dynamic
character of fracture in these experiments, combined with the presence of
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
multiple incipient spots, was also a big challenge in several computational me-
chanics models reviewed in Al-Ostaz and Jasiuk (1997) employing
commercial finite element programs, as well as in an independent study using
a meshless element program (Belytschko et al., 1995). Upon trying various fail-
ure criteria, several numerical methods, and even being forced to initialize the
cracking process by a subjective choice in the meshless model, the modelers
(including this author) have run into uncertainty as to which modeling as-
pect is more critical, and whether there is a way to clarify it. The recent study
by Ostoja-Starzewski and Wang (2006) was motivated by this outstanding
challenge, and offered a way to test the PM vis-a-vis experiments.
In Section 2.6 of Chapter 2 we considered crack patterns in epoxy plates
perforated by holes. Two analysesone based on the minimum potential en-
ergy formulation and one based on the maximum entropy methodrelied on
the assumption of quasi-static response. Strictly speaking, although the load-
ing was static, the fragmentation event was dynamic. Clearly, the preparation
of mineral specimensinvolving measurement of highly heterogeneous and
multiphase microstructuresfor a direct comparison with the model predic-
tion is very hard. Thus, we can apply the model to the experimentally tested
plate with 31 holes and follow this strategy:
1. Decrease the lattice spacing until we attain mesh-independent crack
patterns.
2. Find out whether the lattice of (1) will also result in the most domi-
nant crack pattern of Figure 4.16. Indeed, the crack patterns stabi-
lize as we refine the mesh.
3. Assuming the answer to (2) is positive, introduce weak perturba-
tions in the material propertieseither stiffness or strengthto de-
termine which one of these has a stronger effect on the deviation
away from the dominant crack pattern, that is, on the scatter in
Figure 4.16.
4.3.2.1 Other Models

The PM is but one of the variations on the theme of MD. Here are some other
possibilities:
Molecular statics (MS)by disregarding the inertia forces, it in-
volves a static solution of the system of atoms (Chen, 1995; Vino-
gradov, 2006). Although the MD allows simulations of large systems
with a constraint to very short time scales (transient phenomena of
the order of nanoseconds), the MS allow large (macroscopic type)
time scales albeit with a limitation by the size of a (nonlinear) alge-
braic system one is able to solve and a restriction to 0 K.
Derivation of a continuum model from a microscopic model based
on the assumption that the displacements on the macroscopic level
are the same as those on the molecular level (Blanc et al., 2002).
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
FIGURE 4.16
Final crack patterns for four mesh configurations at ever finer lattice spacings: (a) r0 = 0.1 cm;
(b) r0 = 0.05 cm; (c) r0 = 0.02 cm; (d) r0 = 0.01 cm. (From Ostoja-Starzewski and Wang, 2006.
With permission.)
Introduction of a finite extension and spin for continuum-type par-

ticles (Yserentant, 1997).
Direct incorporation of interatomic potentials into a continuum anal-
ysis on the atomic scale (Zhang et al., 2002).
Computational mechanics of granular media (Hermann and
Luding, 1998).
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
4.4 Michell Trusses: Optimal Use of Material

4.4.1 Study via Hyperbolic System
4.4.1.1 Governing Equations
The classical problem of an optimal truss concerns a minimum-weight design
of a planar truss T that transmits a given load to a given rigid foundation
with the requirement that the axial stresses in the bars of the truss stay within
an allowable range 0 0 (Michell, 1904). This forms the basis for a
problem of layout of a truss whose locally orthogonal members are of a rigid-
perfectly plastic material with tensile and compressive stresses 0 , (Savee
and Prager, 1985; Rozvany et al., 1995, and references therein). The solution
to this problem is provided by a so-called Michell truss-like continuum, whose
members form a dense structure having the geometry of an orthogonal net of
characteristics. That is, as the mesh spacing becomes infinitesimally fine, the
volume (and hence the weight) of the material reaches a minimum.
Allowable stresses in all the truss members are in the range 0 0 .
Given a modulus E, we have a range for strains k k, where k = 0 /E.
As pointed out by Rozvany et al. (1995), equal permissible stresses in tension
and compression are necessary for the Michell (1904) criteria to hold. With
length li and cross-sectional area Ai of bar i, the design variables are the yield
forces Yi = 0 Ai , and the design objective is the minimization of the cost

= Yi li , (4.51)
i
which is proportional to the total volume of the bars V = i Ai li .

As an example let us now consider a problem of optimal layout of a truss
set up on a rigid circular foundation F , which can support a force P acting
at a point A, Figure 4.17. The solution is provided by a so-called field of type
T (Save and Prager, 1985) for which the principal strains have equal absolute
values k, but carry opposite signs, and the principal lines are logarithmic
spirals. Indeed, plots (a)(d) of this figure show a sequence of four ever-finer
trusses providing supports under the same global conditions (the foundation
F and the force P). These trusses are based on, respectively, 2n + 1 (n =
2, 3, 4, and 5) boundary nodes. With n growing, their geometries tend to
an optimal truss-like continuum whose principal strain/stress directions are
mutually orthogonal characteristics of a quasi-linear hyperbolic systemthis
is discussed below. This convergence of trusses (of volume V = V(n), n
finite) to the optimal truss-like continuum (of volume V = V(n)|n ) can be
quantified by the efficiency
E f f = V/V . (4.52)
We indicate Eff in all four cases of Figure 4.17.

P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
P P
A A
F F
(a) E = 0.939 (b) E = 0.965
P P
A A
F F
(c) E = 0.981 (d) E = 0.991
FIGURE 4.17
Successive approximations to the Michell truss, all governed by (3.6) for a homogeneous material,
according to meshes based on, respectively, 2n + 1(n = 2, 3, 4 and 5) boundary points on the rigid
foundation F.
Our presentation of the governing equations follows Hegemier and Prager

(1969). First, we let u and v be the displacement components with respect to
the x, y coordinates in the plane of the truss. Then

u v v u v u
x = y = = + /2 W= /2 (4.53)
x y x y x y
are the strain components and the rotation. If by we denote the angle be-
tween the negative y-direction and arbitrarily assigned positive direction
along the line with the unit extension k, then
x = k cos 2 y = k cos 2 = k sin 2. (4.54)

P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
It now follows from the above that

u u
= k cos 2 = k(2w + sin 2)
x y
v v (4.55)
= k(2w sin 2) = k cos 2,
x y
where w = W/k.
Eliminating u and v from (4.55) by cross-differentiation, we obtain
w
k + k cos 2 + k sin 2 = 0,
x x y
(4.56)
w
k + k sin 2 k cos 2 = 0.
y x y
and, setting = /2, we find equations
d d
(w ) = 0 (w + ) = 0, (4.57)
ds1 ds2
which hold along two characteristics s1 and s2 , at angles specified, respectively,
by
= + /4 = + 3/4. (4.58)
is defined as the angle formed by the positive direction along the foundation
F with the positive x-direction. On this boundary, w = 1.
w = 1. (4.59)
4.4.1.2 Example of an Optimal Layout

In practice, as illustrated by the example of Figure 4.17, the stochastic quasi-
linear hyperbolic system governing the field is solved by finite differences.
At the typical point Q of the foundation F , let the positive direction along
this boundary form the angle with the positive x-direction. As the rigid
foundation is inextensible, its tangent and normal at Q bisect the right angles
formed by the principal axes of strain at Q. As u vanishes along F , cos ux +
sin uy = 0 at Q. From this, along with (4.55) and (4.58) there follows
w = 1 (4.60)
along F ; in Figures 4.1 and 4.3 the upper sign in the above is appropriate.
Thus, we have an inverse Cauchy problem: find the net of characteristics
supporting the given load P at point A, that emanates from the foundation F
with conditions (4.58) and (4.59) specified on it. Figures 4.17(a)(d) display
four deterministic solutions, all governed by (4.57) according to meshes based
on, respectively, 2n +1 (n = 2, 3, 4 and 5) boundary points. In the limit n ,
we arrive at a truss-like continuum with Eff = 1.
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
In case one wants to make (i.e., manufacture) the Michell truss from a metal
platesuch as a polycrystalone encounters the effect of random material
microstructure. In effect, the denser the trussor the finer the mesh spacing
the more significant is the effect of microstructural fluctuations on the plastic
limit. This leads to a question we address in Chapter 8: Can one truly reach
the limit of truss-like continuum as the mesh is refined ad infinitum?
4.4.2 Study via Elliptic System

The foregoing approach applied when one assumed a fixed load and searches
for a structure of optimal (i.e., minimal) weight. Another approach, much
more in line with the conventional methodology of shape optimization of
engineering structures (Bendse and Kikuchi, 1988), is to seek the shape of
a structure with minimum compliance (maximum stiffness), that possesses a
prescribed weight. The minimum compliance problem for a planar (respec-
tively, spatial) body B of volume V in R2 (R3 ) subjected to body forces f and
tractions t takes the form:
min L(u)
(4.61)
Cijkl Uad
subject to a C (u, v) = L(v), all v U, design constraints, where

a C (u, v) = (u)C(x)(v)dx (4.62)
B
is the energy bilinear form, and

L(u) = fvdx+ tvdx (4.63)
B Bt
is the load linear form. That is, we seek the optimal choice of stiffness tensor
C in some given set of admissible tensors Uad . C are generally fields over R2 ,
so that Uad (L (V)) 6 , corresponding to the six independent elements of in-
plane stiffness tensor. By the design constraints we understand constraints
on stresses, strains, displacements, etc., while sizing constraints, volume con-
straints, etc., are accounted for in the choice of Uad . Finally, U is the space of
kinematically admissible displacement fields.
In the case of optimal shape design, elements C( Cijkl ) of Uad take on the
form
Cijkl (x) = (x)C ijkl , (4.64)
where C ijkl is the constant stiffness tensor for the material employed for the
construction of the mechanical element, and (x) is the indicator function.
The discretized formulation of the topology optimization problem can then
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
be stated as follows:
min f ()
s.t. V = j i V (4.65)
i 1, i = 1, . . . , N,
where f represents the objective function, i and i are element densities
and volumes, respectively, V is the target volume, N is the total number
of elements and is the small number that prevents stiffness matrix from
being ill-conditioned. The common objective function is the weighted sum of
compliances under all load cases. Note that the problem in (4.65) is a relaxation
formulation of the topology problem, where the density should only take
value 0 or 1. To force the design to be close to a 0/1 solution, a penalty is
introduced to reduce the efficiency of intermediate density elements, namely,
by a following power law formulation
Ki () = i Ki
p
(4.66)
where Ki and Ki represent the penalized and the real stiffness matrix of the ith
element, respectively, and p is the penalization factor that is bigger than 1. For
a survey of historical development and a summary of theory and techniques
of topology optimization, see Rozvany et al. (1995), while for the penalty
formulation see Allaire and Kohn (1993).
In Chapter 8 we return to Michell trusses in the setting of materials pos-
sessing microstructural randomness.
Problems
1. The extension of the condition (4.1) in 3D is the so-called Maxwells
rule:
|E| = 6|V| 6.
Justify this equation.

2. Verify that the network of Figure 4.1 is generically rigid.
3. Using the graph representation, formulate the virtual work prin-
ciple in granular media relating the work of forces and moments
of grain-grain interactions with the work of forces and moments
acting at the boundary of the granular medium.
4. In the case of iron, the interatomic potential is

(r ) = D e 2a (r r0 ) 2e a (r r0 )
where the binding energy D = 0.4174eV, a = 1.3885( A) 1 and

0 = 2.845 A (Milstein, 1971). Compute E, c and max . To see the
P1: BINAYA DASH
June 1, 2007 12:16 C4174 C4174C004
applicability of linear elastodynamics, compare the latter quantity

with the highest frequencies of about 106 /sec excited in transducers.
Translate these results into wavelengths.
5. Modify the 1D chain of particles of Section 3.1.1 by adding a cubic
term in the equation (3.1)1 :
1 1
U= K (ui+1 ui ) 2 + C(ui+1 ui ) 3 .
i
2 3
Derive the resulting dynamical equation generalizing (3.2). Then,

using a Taylor expansion with terms up to the fourth derivative,
obtain the continuum equation

s2
E u, xx + u, xxxx + a u, x u, xx = u.
12
Identify E, and .
6. Consider two particles (of mass mi , i = 1, 2 located at ri , respec-
tively, separated by a distance rij = rji , and interacting via a central-
force potential (rij ). In the completely conservative method of
particle modeling (PM) one replaces their Newtonian equations of
motion (no summation convention and no tensor index notation)
d 2 ri ri r j
mi = , i, j = 1, 2
dt 2 ri j ri j
by two first order difference equations

ri,n+1 ri,n Vi,n+1 Vi,n
=
t 2
vi,n+1 vi,n (ri j,n+1 ) (ri j,n ) ri,n+1 + ri,n r j,n+1 r j,n
mi = ,
t i, j =r1, 2 ri j,n
i j,n+1 ri j,n+1 + ri j,n
Where the first subscript specifies the particle number while the
second subscript indicates the time t or t+t. These equations can be
solved by the Newtons method. Show that the numerical solution
conserves the total energy, the linear and the angular momentum.
7. Consider the system of three particles interacting via central forces.
Write their Netwonian equations of motion, and then the corre-
sponding first order difference equations.
8. Show that the numerical method discussed in Problem 5 is invariant
with respect to rotations of the frame of coordinates.
9. Verify (4.56), and derive a more general version of the equations
governing w and for the case of k being smooth in x and y.
P1: Binod
May 28, 2007 18:4 C4174 C4174C005
5
Two- Versus Three-Dimensional
Classical Elasticity
Doing physics in Flatland requires no apologies: many areas of

theoretical and condensed matter physics have benefitted greatly
from such studies.
U. Frisch, 1995
Classical (linear) elasticity is a very old subject, and its planar/2D cases (plane
stress and plane strain) equally so. However, the possibility of having the same
stress field in 2D materials whose (generally spatially inhomogeneous) elastic
moduli are different but satisfy certain relations is a relatively new result. This
reduced parameter dependence has consequences for the effective moduli of
composite materials, including the special case of a plate perforated by holes
(especially up to the percolation point), and lends itself to extensions, such as
the presence of body force fields or thermal stresses. These issues, including
a short section on poroelasticity, are reviewed in this chapter.
5.1 Basic Relations

5.1.1 Isotropic Relations
5.1.1.1 Three-Dimensional Elasticity
The constitutive relations for a linear elastic isotropic 3D material are
1 1 + 3D
11 = [11 3D (22 + 33 )] 12 = 12 (5.1)
E 3D E 3D
together with cyclic permutations 1 2 3. Here E 3D and 3D stand for
conventional 3D Youngs modulus and Poissons ratio, and this convention
is followed with respect to other quantities. (5.1) is, of course, equivalent to
1
i j = (1 + 3D ) i j 3D kk i j i, j, k = 1, 2, 3 (5.2)
E 3D
Chapter written jointly with I. Jasiuk.
171
P1: Binod
May 28, 2007 18:4 C4174 C4174C005
or
i j = 3D kk i j + 23D i j , i, j, k = 1, 2, 3 (5.3)
where the classical relations involving the Lame constants and bulk modulus
are well known:
E 3D 23D 3D E 3D 2
3D = 3D = 3D = = 3D + 3D .
2 (1 + 3D ) 1 23D 3 (1 23D ) 3
(5.4)
5.1.1.2 Two-Dimensional Elasticity

In the 2D elasticity there is no x3 direction and thus 11 , 22 , and 12 are the
only strains and 11 , 22 , and 12 are the only stresses. Also, there is just one
compatibility equation
22,11 + 11,22 = 212,12 . (5.5)
For the Hooke law, we have

1 1 + 2D
11 = [11 2D 22 ] 12 = 12 , (5.6)
E 2D E 2D
with cyclic permutation 1 2, where the subscript 2D indicates planar (or
area) material properties. The above is equivalent to
1
i j = (1 + 2D ) i j 2D kk i j i, j, k = 1, 2 . (5.7)
E 2D
Now, in analogy to (5.3) above, we can write
i j = 2D kk i j + 22D i j , i, j, k = 1, 2 (5.8)
and work out expressions for the planar Young modulus and planar Poisson ratio:
2D + 2D 2D
E 2D = 42D 2D = . (5.9)
2D + 22D 2D + 22D
In analogy to the 3D case (e.g., Ziegler, 1983), we can work out the basic
inequalities that hold between these planar moduli. First, consider three tests:
1. Uniaxial stress occurring in a narrow strip subjected to tension or
compression in, say, x1 direction and defined by the condition that
of all three stress components only a single normal stress 11 is non-
zero. The equation (5.6) implies then that
1 2D
11 = 11 22 = 11 12 = 0. (5.10)
E 2D E 2D
2. Hydrostatic stress (= p) which leads to
1
11 + 22 = kk = p, (5.11)
2D
P1: Binod
May 28, 2007 18:4 C4174 C4174C005
Two- Versus Three-Dimensional Classical Elasticity 173
where we identify the planar bulk modulus 2D

E 2D
2D = = 2D + 2D . (5.12)
2 (1 2D )
Of course, in 3D the shear modulus is accompanied by a factor 2/3,
as given in equation (5.4).
3. Simple shear stress defined by the condition 12 = 21 and 11 = 22 =
0, which yields
1
12 + 21 = 12 = 12 , (5.13)
2D
where the planar shear modulus is given by the same formula in
2D as in 3D
E 2D
2D = . (5.14)
2 (1 + 2D )
Stability of the materialthat is, positivity of E 2D , 2D and 2D imposes

three inequalities which follow from the above tests:
2D + 22D > 0 2D + 2D > 0 2D > 0. (5.15)
If 2D obeys the third of these inequalities, the first one is a consequence

of the second and can be dropped. In Figure 5.1 we show the region in the
(2D ,2D )-plane where the two remaining inequalities are satisfied.
2D
2D = 2D
Q
q
2D
FIGURE 5.1
Region where the inequalities (5.15) are satisfied.
P1: Binod
May 28, 2007 18:4 C4174 C4174C005
On its entire left-hand boundary, Poissons ratio given by (5.9)2 assumes the
value 1. Since
2D 2D
=2 > 0, (5.16)
2D [2D + 22D ]2
2D increases monotonically on any vector q parallel to the 2D axis, tending

toward 1 with increasing distance from the point Q, Figure 5.1. Note from
(5.9)2 that 2D is seen to range from 1 through +1, in contradistinction to
3D , which is bounded by 1 and 1/2.
5.1.2 Plane Elasticity Models

There are two classes of plane elasticity problems that can be solved by using
a reduced form of 3D formulation: plane strain (such as a long cylinder) and
plane stress (such as a thin plate).
Plane strain. In that case, one requires u3 = 0 in (5.1) along with the
independence of all the fields with respect to the x3 direction, so that 33 =
31 = 32 = 0, but 33 =
0, which leads to the following equations:
1 1 + 3D
11 = 1 3D
2
11 3D + 3D
2
22 12 = 12 (5.17)
E 3D E 3D
again with the cyclic permutation 1 2. A comparison with (5.6) readily

shows that the relations between the 2D (planar) and the 3D are as follows:
1 1 3D
2
2D 3D + 3D
2
1 + 2D 1 + 3D
= = = . (5.18)
E 2D E 3D E 2D E 3D E 2D E 3D
This is a mapping of two constants onto two constants, so that only two
relations of the above three are independent. Of particular interest is the
relation between the plane strain Poisson ratio and the 3D Poisson ratio
3D
2D = . (5.19)
1 3D
Now, an inspection of (5.17)3 immediately reveals that the 2D shear modulus

has the same form as the 3D one, while applying the concept of bulk modulus
to relations (5.2), we infer the plane strain 2D , which has the form differing
from 3D elasticity (5.4)3 . Thus, as already shown in (5.12) and (5.14),
E 2D E 2D
2D = 2D = . (5.20)
2 (1 2D ) 2 (1 + 2D )
P1: Binod
May 28, 2007 18:4 C4174 C4174C005
Two other very useful relations linking these two-dimensional moduli

E 2D , 2D , 2D , and 2D can readily be inferred
4 1 1 2D 2D
= + 2D = . (5.21)
E 2D 2D 2D 2D + 2D
Plane stress. In that case, one requires 33 = 31 = 32 = 0 (but 33 = 0)

along with the independence of all the fields with respect to the x3 direction,
which leads to
1 1 + 3D
11 = [11 3D 22 ] 12 = 12 , (5.22)
E 3D E 3D
with cyclic permutation 1 2. A comparison with (5.6) readily shows that

the following relationships between these plane-stress and 3D moduli hold
1 1 2D 3D 1 + 2D 1 + 3D
= = = . (5.23)
E 2D E 3D E 2D E 3D E 2D E 3D
Again the third of these relations is redundant, but, collecting all, we see that
E 2D E 2D
E 2D = E 3D 2D = 3D 2D = 2D = . (5.24)
2 (1 + 2D ) 2 (1 2D )
Note that (5.9) holds again. A detailed discussion of all these relationships
is given in Thorpe and Jasiuk (1992); we also reproduce their Table 1 here
(Table 5.1).
Unified treatment. Further down, it will be useful to deal with the Hooke
law in this form:
4i j = 2Si j + ( A S) kk i j , i, j, k = 1, 2 (5.25)
TABLE 5.1
Elastic Constants in Plane Strain and Plane Stress Expressed
in Terms of Conventional 3D Moduli
Compliance Plane Strain Plane Stress 2D Relations
93D 3D E 2D
Bulk modulus 2D = 3D + 3D /3 2D = 33D +43D 2D = 2(12D )
E 2D
Shear modulus 2D = 3D 2D = 3D 2D =
2(1+2D )
Youngs modulus E 2D = E 3D / 1 3D
2 E 2D = E 3D E 2D
Poissons ratio 2D = 3D / (1 3D ) 2D = 3D 2D
Note: From Thorpe and Jasiuk, 1992. With permission.

P1: Binod
May 28, 2007 18:4 C4174 C4174C005
involving two planar compliances: bulk compliance A and shear compli-

ance S:
1 1 1
A= = S= , (5.26)
2D 22D 2D
where = (3 2D ) / (1 + 2D ) is the Kolosov constant for plane strain as
well as plane stress. Note also that in 3D:
plane strain: = 3 43D

3 3D (5.27)
plane stress: = .
1 + 3D
Finally, a uniaxial compliance C
1 +1
C= = (5.28)
E 2D 82D
will also be useful. Observe from (5.21)1 that
A + S = 4C. (5.29)
5.1.3 Special Planar Orthotropies

The constitutive relations for a linear elastic planar material may be written
in a polar form (Jones, 1975; Vannucci, 2002; Vannucci and Verchery, 2001):
C1111 = T0 + 2T1 + R0 cos 4 0 + 4R1 cos 2 1

C1112 = R0 sin 4 0 + 2R1 sin 2 1
C1122 = T0 + 2T1 R0 cos 4 0
(5.30)
C1212 = T0 R0 cos 4 0
C2212 = R0 sin 4 0 + 2R1 sin 2 1
C2222 = T0 + 2T1 + R0 cos 4 0 4R1 cos 2 1
Here T0 , T1 , R0 and R1 are scalars, while 0 and 1 are angles, and these six
quantities are related to the six values for Cijkl through
8T0 = C1111 2C1122 + 4C1212 + C2222

8T1 = C1111 + 2C1122 + C2222
(5.31)
8R0 e 4i 0 = C1111 2C1122 4C1212 + C2222 + 4i (C1112 C2212 )
8R1 e 2i 1 = C1111 C2222 + 2i (C1112 + C2212 )
The six in-plane compliances Sijkl may be expressed through equations of

the same form as (5.30), except that one has to replace T0 through 1 by, say,
P1: Binod
May 28, 2007 18:4 C4174 C4174C005
lower-case letters t0 through 1 ; then a form fully analogous to (5.31) holds

for expressing t0 through 1 in terms of the Sijkl values.
The polar form has several interesting features:
1. In a coordinate system rotated by around the x3 axis the Cijkl s are

expressed by the equations (5.30) with 0 and 1 replaced by 0
and 1 , respectively.
2. The five quantities T0 , T1 , R0 , R1 , and 0 1 are invariant with
respect to the orientation , and are thus invariant.
3. Each component is a Fourier expansion with at most three terms,
so that the period equals or 2 .
4. Orthotropy (C1112 = C2212 = 0) occurs for 0 1 being a multiple of
/4, and then the number of independent parameters is only four:
T0 , T1 , (1) K R0 , R1 , with K being either 0 or 1.
5. For an orthotropic material, the usual thermodynamic restrictions
result in:

T0 > R0 T1 T0 + (1) K R0 > 2R12 (5.32)
6. One special type of orthotropy occurs when R0 = 0, and this cor-

responds to the absence of the 4 harmonic and the shear modulus
being invariant to rotations. However, this does not imply r0 = 0,
so that the shear compliance is dependent on , and the Sijkl tensor
still depends on both harmonics.
7. Another special type of orthotropy occurs when R1 = 0; this corre-
sponds to the absence of the 2 harmonic and implies r1 = 0 for the
Sijkl tensor.
8. Isotropy occurs when R0 = R1 = 0, and then, obviously, there are
just two independent invariants.
9. In the case of Cauchy symmetry (Cijkl = Cik jl ), isotropy reduces
to the stiffness being described by just one elastic constant; recall
Chapter 3.
5.2 The CLM Result and Stress Invariance

5.2.1 Isotropic Materials
5.2.1.1 Basic Result
Consider a planar elastic solid occupying a simply connected domain B in
(x1 , x2 )-plane, in static equilibrium
i j, j (x) = 0 or (x) = 0. (5.33)

P1: Binod
May 28, 2007 18:4 C4174 C4174C005
Elastic moduli are isotropic and assumed to be twice-differentiable. The solid

is subjected to traction boundary conditions
ji (x) n j = ti(n) (x) or (x) n = t(n) (x) x B, (5.34)
in such a way that the global equilibrium is satisfied

t (x) d S = 0
(n)
x t(n) (x) d S = 0. (5.35)
B B
Now, substituting (5.25) into the compatibility condition (5.5), and using
(5.33), we obtain, after some manipulations,
A+ S
2[ (11 + 22 )] S,11 11 + 2S,12 12 + S,22 22 = 0. (5.36)
2
The following question may now be asked: Supposing that A and S are
changed to some A and S, then under what restrictions would the original
stress field (11 , 22 , 12 ) remain unchanged? Denoting by a hat the quantities
pertaining to the new material, the so-called invariance of the stress field is
written as
(x) = (x) (5.37)
An examination of (5.36) implies that we must have

+
A S = m ( A + S)
S,11 = mS,11
S,22 = mS,22
S,12 = mS,12 (5.38)
where m is an arbitrary scalar. Note that this means

=m A + a + bx + cy
A
S = mS a bx cy = mC
C (5.39)
where the third equality comes from (5.29). The constants m, a , b, and c are
subject to restrictions dictating that the compliances be non-negative.
The result that the stress field is unchanged (invariant) under such a shift
of compliances is called the CLM stress invariance or transformation after the
authors (Cherkaev et al., 1992); see also Milton (2002). However, their original
result was that
=A + c
A
S= Sc = C,
C (5.40)
while the shift linear in x and y (5.39) is due to Dundurs and Markenscoff
(1993). Note that in the original paper of Cherkaev et al. (1992)
1 1 1 1 1 1
= + = , (5.41)
2D 2D 2D
2D
where constant = 1/c and m = 1.
The CLM result holds also for materials with two or more distinct phases
(e.g., matrix-inclusion composites) with perfectly bonded or slipping inter-
faces, as well as for anisotropic materials (see below). Several extensions and
P1: Binod
May 28, 2007 18:4 C4174 C4174C005
generalizations of the CLM result are given in this chapter and Chapter 6.
Note that a pair of materials satisfying (5.395.41) is called equivalent mate-
rials. In effect, one obtains a so-called reduced parameter dependence, which is
important in parametric studies of composites, both experimental and theo-
retical. It can be used as a check for numerical and analytical calculations, it
reduces the number of output parameters, and facilitates the presentation of
results, leading to savings in time and space resources.
5.2.1.2 Two-Phase Composites and Dundurs Constants

When dealing with a two-phase, planar composite material (with phases
(1)
1 and 2), we have four material constants, for example, 2D , (1) (2) (2)
2D , 2D , 2D .
(1) (1) (2) (2)
This, of course, is equivalent to A , S , A , and S , which undergo the
CLM shift

A(1) = A(1) + c,
S(1) = S(1) c,
(5.42)
A(2) = A(2) + c,
S(2) = S(2) c,
and the stress field remains the same under traction boundary conditions.
This implies a reduction in the number of independent constants by one.
That result is directly related to the result of Dundurs (1967, 1969).
In general, we may expect the stress field to depend on the magnitude of
the external loading, expressed by 0 , and on three dimensionless parameters
of material constants:

(1) (2)
(x) = 0 g x; 2D /2D , (1) (1) (2) (2)
2D /2D , 2D /2D , (5.43)
where g is some second-rank tensor function. Dundurs (1967, 1969) has also
shown that, for plane problems, only two dimensionless parameters,
C (1) C (2) 1/E (1) 1/E (2)

12 = =
C +C
(1) (2) 1/E (1) + 1/E (2)
A(1) A(2) 1/ (1) 1/ (2)
12 = = , (5.44)
4 C +C
(1) (2) 4 1/E (1) + 1/E (2)
are needed; these are called Dundurs constants. Thus, (5.43) may be replaced
by
(x) = 0 g (x;12 , 12 ) . (5.45)
The choice of 12 and 12 is not unique, and other combinations of elastic

constants are possible. Note that 12 and 12 are invariant under the CLM
shift. Dundurs result holds for two-phase materials with perfectly bonded
and slipping interfaces. A generalization to composites with many phases
was done by Neumeister (1992).
P1: Binod
May 28, 2007 18:4 C4174 C4174C005
5.2.2 Anisotropic Materials and the Null-Lagrangian

Note that the 2D compliance Sijkl of an isotropic material is a following function
of two constants 2D and 2D (here we drop 2D for simplicity of
notation):
1 1
I
Sijkl (, ) = i j kl + ik jl + il jk i j kl , (5.46)
4 4
its associated stiffness tensor is

1
I
Cijkl (, ) = Sijkl
I
(, ) = i j kl + ik jl + il jk i j kl . (5.47)
I
Equivalently, noting (5.26), Sijkl (, ) can also be given in terms of A and S as
A S
I
Sijkl ( A, S) = i j kl + ik jl + il jk i j kl . (5.48)
4 4
If we let = and = in (5.48), it follows that the shift tensor is
1 1 1
I
Sijkl ( , ) = i j kl ik jl + il jk = Rijkl , (5.49)
2 4 2
where
1 1
Rijkl = i j kl ik jl + il jk i j kl (5.50)
2 2
represents the rotation by a right angle of a symmetric second-rank tensor.
In the case of an anisotropic material, the starting point is again pro-
vided by the compatibility relation (5.5). Proceeding in the same manner
as when deriving and analyzing (5.36), we conclude that the stress field
remains unchanged when the material constants are modified from Sijkl to
Sijkl = Sijkl + Sijkl
I
( , ). Thus, we see from (5.47) that the shift (5.46) in
compliances R(1)
ijkl kl is a right-angle rotation of kl .
The local equilibrium equation also holds
(x) = 0, (5.51)
and this may be written in terms of the Airy stress function

4 = 0. (5.52)
Here the stress is expressed by

= R () R, (5.53)
in which

0 1 2 /x12 2 / x1 x2
Rik = = . (5.54)
1 0 2 /x1 x2 2 / x22
Thus, Rik is a two-dimensional analog of the LeviCivita permutation tensor.

P1: Binod
May 28, 2007 18:4 C4174 C4174C005
The foregoing development allows us to present the shift-result in terms

of the null-Lagrangian (Cherkaev et al., 1992). It follows from (5.50) that the
strain energy density
1
W Sijkl = i j Sijkl kl (5.55)
2
may now be written as

I
W Sijkl = Rijkl ,kl Sijmn Rmnpq , pq . (5.56)
First, let us note that the a minimization of (5.53) via the EulerLagrange
equations of (5.52) results in compatibility equations in terms of for a general
anisotropy. On the other hand, the energy density of the shift in compliance
becomes

I
W Sijkl ( , ) = Rijkl ,kl Sijmn
I
Rmnpq , pq . (5.57)
Now, observing that
Rkli j Rijmn Rmnpq = Rklpq Rijkl Rijmn = km ln , (5.58)
we find
1 2
I
W Sijkl ( , ) = ,11 ,22 ,12 . (5.59)
2
The energy (5.56) can also be written as the divergence of a vector field vk
such that

1 1 ,1 ,22 ,2 ,12
vk = ,l Rklpq , pq = . (5.60)
2 2 ,2 ,11 ,1 ,12
It follows now that vk,k = 0, or that the EulerLagrange equations for

I
W(Sijkl ( , )) are satisfied identically, justifying the name of W(Sijkl
I
( , )):
null-Lagrangian. For a study of symmetries of 2D stiffness tensors see He and
Zheng (1996).
5.2.3 Multiply Connected Materials

In the case of multiply connected materials, the compatibility condition (5.4)
must be accompanied by the so-called Cesaro integrals. These are three line
integrals on a closed contour surrounding any one cavity

3 = d3 D1 = du1 D2 = du2 (5.61)
P1: Binod
May 28, 2007 18:4 C4174 C4174C005
where 3 = u(2,1) . Following Mindlin and Salvadori (1950), this implies

12 11 22 12
3 = d x1 + d x2
x1 x2 x1 x2

11 11
D1 + y0 3 = x1 d x1 + d x2
x1 x2

11 22 12
x2 d x1 2 d x2
x1 x2 x2

22 22
D2 x0 3 = x2 d x1 + d x2
x1 x2

22 12 11
x1 d x2 + 2 d x1 . (5.62)
x1 x1 x2
Here D1 and D2 are dislocation vectors and 3 is a disclination, while (x0 , y0 )

are the coordinates of the starting point on the contour. Using (5.25), we
express the above equations in terms of stresses

S S
43 = [( A + S) (11 + 22 )] ds 2 t1 ds 2 t2 ds, (5.63)
n x1 x2

4 ( D1 + y0 3 ) = x2 x1 ( A + S) (11 + 22 ) ds
n s

S S
+2 St2 ds 2 x2 t1 + t2 ds, (5.64)
x1 x2

4 ( D2 x0 3 ) = x1 + x2 ( A + S) (11 + 22 ) ds
n s

S S
2 St1 ds + 2 x1 t1 + t2 ds, (5.65)
x1 x2
where n and s denote the outer unit normal and arc length of the hole bound-
ary. Let us now consider the situation of a continuous displacement field:
the dislocations are not allowed, and the left-hand sides in the above equa-
tions equal zero. A transformation of the elastic compliances according to
(5.395.41) should then leave the stress state invariant, which implies that the
following equations in tractions hold:

t1 ds = t2 ds = 0 (x1 t2 x2 t1 ) ds = 0. (5.66)
These relations, also called Michell conditions, mean that the net forces and
couples applied to each and every cavity must be self-equilibrated. The above
results are due to Dundurs and Markenscoff (1993) for linear transformations
P1: Binod
May 28, 2007 18:4 C4174 C4174C005
[Equation (5.39)]; those authors also give an analysis of discontinuous compli-

ances, bonded versus slipping interfaces, and intrusions. The constant trans-
formation (5.40) requires only the satisfaction of the first two equations in
(5.66)this is in agreement with Michell (1899), who showed that the stress
field in a material with holes, satisfying (5.66)1,2 , is independent of elastic
constants. This result has been used extensively in photoelasticity.
The effect of body forces b on the reduction of material parameters in
multiply connected bodies has been further analyzed by Markenscoff and
Jasiuk (1998). In particular, the Michell conditions (of zero traction on the
hole boundary) have been generalized to account for the presence of b. For
two-phase, multiply connected composites with either perfectly bonded or
slipping interfaces, the requirement of reducibility is the continuity of the
normal component of the body force across the interface, in addition to the
condition that body forces need to be divergence-free:

b n ds = 0 (tx + xb n) ds = 0 ty + yb n ds = 0. (5.67)
5.2.4 Applications to Composites

5.2.4.1 Effective Moduli of Composites
As mentioned in the preface, determination of the effective properties of com-
posites is one of the primary applications of the CLM result. It is thus of in-
terest to investigate the shift in macroscopically effective compliance tensors
(1)e f f
Sijkl connecting the volume averaged (with an overbar) stress and strain
tensors
eff
i j = Sijkl kl . (5.68)
As before, there is a restriction to materials with smooth (class C 2 ) properties.

Now, stresses 11 , 22 , 12 are the same in two equivalent materials (de-
scribed by Sijkl (x) and
Sijkl (x)), so that the strain fields i j and
i j satisfy the
relation
i j =
Sijkl kl = Sijkl kl + Sijkl
I
( , ) kl = i j + Sijkl
I
( , ) kl . (5.69)
I
Carrying out volume averaging of (5.68), and noting that Sijkl ( , ) is
independent of position x, we find
eff eff

i j = Sijkl kl = Sijkl kl + Sijkl
I
( , ) kl , (5.70)
which shows that the effective compliance tensor of the transformed material
(with hat) is given by that of the first material (without hat) plus the shift
given by (5.49), that is,
eff eff
Sijkl = Sijkl + Sijkl
I
( , ) . (5.71)
P1: Binod
May 28, 2007 18:4 C4174 C4174C005
When the original material is isotropic, these results are expressed by

saying that, if the shift of the material properties is governed by (5.41), then
the shift of the effective compliances is governed by
1 1 1 1 1 1 eff = E eff ,
= eff + = eff E (5.72)
eff eff

with the same .
Note: This conclusion holds for simply connected inhomogeneous mate-
rials with twice-differentiable properties mentioned at the beginning of this
section, as well as for composite materials (e.g., matrix-inclusion composites)
with two or more discrete phases, with either perfectly bonded or sliding
interfaces (Thorpe and Jasiuk, 1992; Dundurs and Markenscoff, 1993; Jun and
Jasiuk, 1993; Moran and Gosz, 1994). This exact result applies to any geome-
try of the composite and is independent of the method used (computational
or analytical) to evaluate these elastic constants. Thus, it can serve as a very
useful check of such calculations. It should be noted that, strictly speaking,
this result is valid under traction boundary conditions, but if the domain is
large enough to qualify as the RVE (see Chapter 7), then it will also hold under
displacement or mixed-orthogonal boundary conditions, that is, three basic
loadings satisfied by the Hill condition, because the derived effective moduli
are independent of the boundary conditions.
5.2.4.2 Plates with Holes

Thorpe and Jasiuk (1992) have shown that the effective Young modulus E eff
of a plate containing holes is independent of the Poisson ratio of the matrix
material, where E is the Young modulus of the matrix, and the effective Pois-
son ratio eff flows, as the volume fraction of holes increases, toward a fixed
point that is reached at percolation of holes (i.e., when E eff = 0):

E = E eff /E
eff / E eff eff = E eff /E, (5.73)
This is illustrated in Figure 5.2 for a material with randomly placed circu-
lar holes (Poisson point field of Chapter 1), in which the effective Poisson
ratio eff flows toward the fixed point of value 1/3 as the volume fraction
of holes increases and reaches the percolation point at the volume fraction
2/3. The fixed point and the percolation point depend on the particular
microgeometries and the approximations employed (e.g., different effective
medium theories) to find the effective moduli. These results were shown to
hold by various effective medium theories (Jun and Jasiuk, 1993), and were
first observed numerically by Day et al. (1992).
5.2.5 Extension of Stress Invariance to Presence of Eigenstrains

This section briefly reviews the results of Jasiuk and Boccara (2002). Thus, we
begin with the total strain i j being the sum of the elastic strain e i j and the
P1: Binod
May 28, 2007 18:4 C4174 C4174C005
1.0
(a)
0.8
Youngs modulus E*/E1

0.6
0.4
0.2
1.0
(b)
0.5
Poissons ratio v*
0.5
1.0
0.4 0.6 0.8 1.0
q
FIGURE 5.2
(a) Youngs modulus E eff /E for a sheet containing randomly distributed circular holes for various
values of the planar Poisson ratio of the matrix material, in function of the fraction of material
remaining q = 1 f , where f is porosity. (b) Flow of the effective planar Poisson ratio eff towards
2/3 for a wide range of matrix materials. (From Thorpe and Jasiuk, 1992. With permission.)
eigenstrain ij
i j = e i j + ij where e i j = Sijkl kl . (5.74)
In elasticity with eigenstrains the material is assumed to be free from any

external forces and surface constraints (Mura, 1987). If these conditions of free
surface are not satisfied, the stress fields can be obtained by a superposition
of the stress of a free body and the stress obtained from the solution of a
given boundary value problem with non-zero external forces or boundary
conditions. Following the approach of Mura (1987), by substituting (5.74)
into (5.33) and assuming a homogeneous material, we have

Ci jkl kl, j = Ci jkl kl, j, (5.75)
P1: Binod
May 28, 2007 18:4 C4174 C4174C005
and by substituting (5.74) into (5.34) with ti(n) (x) = 0, we obtain
Ci jkl kl n j = Ci jkl kl n j . (5.76)
Note that in the absence of eigenstrains (kl = 0), the left-hand side of (5.75)
corresponds to i j, j , and the left-hand side of (5.76) to i j n j . Thus, (5.75) is in

the form i j, j = Xi where Xi = Ci jkl kl, j and (5.76) is in the form i j n j = ti

where ti = Ci jkl kl n j . Therefore, the contribution of the eigenstrain ij to the
equations of equilibrium (5.75) is mathematically equivalent to a body force,
and in the boundary conditions (5.76) is similar to a surface force.
We next focus on the planar elasticity with eigenstrains, assuming isotropy
in elastic properties. In addition we admit non-zero tractions ti(n) (x) to make
the formulation more general. This will not change our conclusions on the
reduced parameter dependence.
Note that, for the special case of uncoupled thermoelasticity, the eigen-
strains ij are defined as
ij = i j T i j = 0 if i =
j i, j = 1, 2, 3, (5.77)
where i j is the thermal expansion coefficient and T is temperature change.

We will refer to this special case in examples.
5.2.5.2 Planar Case

In the case of an isotropic 2D material, the constitutive law is generalized
from (5.25) to
4i j = 2Si j + ( A S) kk i j + 4ij + 433

i j , i, j = 1, 2, (5.78)
where is the Kolosov constant defined in (5.27). With the assumption of

both compliances and eigenstrains being smooth functions of position, the
compatibility condition becomes

2 A+S
2
(11 + 22 ) S, 11 11 S, 22 22 2S, 12 12 = 211,22 222,11

+ 412,12 2 2 33 4, 1 33,1 4, 2 33,2 2 2 33 , (5.79)
where, given (5.265.27), 3D = 3D = (1 A/S) /2 for plane strain, and = 0

for plane stress.
Equation (5.79) remains unchanged under the linear shift (5.39), and thus
there is a reduced parameter dependence, in these particular situations:
For the plane stress case and m = 1, no extra condition is needed

For the plane stress case and m = 1, need to satisfy

2 33 + 2,1 33,1 + 2,2 33,2 + 2 33 = 0. (5.80)
For the plane stress case and m =

1, need to satisfy

22,11 212,12 + 211,22 = 0. (5.81)
P1: Binod
May 28, 2007 18:4 C4174 C4174C005
For the case of plane strain and m =

1, need to satisfy

11,22 + 22,11 212,12 + 2 33 + 2,1 33,1

+2,2 33,2 + 2 33 = 0. (5.82)
Besides two special cases: [(1) a homogeneous material with uniform

eigenstrains subject to zero traction boundary conditions, and (2) a homo-
geneous material with nonuniform eigenstrains subject to traction boundary
conditions, Jasiuk and Boccara (2002) have also studied the situation of a two-
phase material with perfectly bonded interfaces. Here each case is governed
by equation (5.79) with all the material and field quantities being indexed by
( p) ( p)
p = 1, 2 (for either phase p), that is, Ap , Sp , p , i j , and i j . The perfect
interfaces S12 , with n and s being unit and tangential directions in the plane,
are modeled via the continuity of normal and tangential tractions
nn
(1)
= nn ns = ns
(2) (1) (2)
, (5.83)
and the continuity of changes in curvature and stretch strains (Dundurs, 1990)
K (1) = K (2) ss
(1)
= ss
(1)
. (5.84)
Note: the latter two equations replace the conventional conditions

involving the continuity of normal and tangential displacements u(1) n = un ,
(2)
us = us .
(1) (1)
The linear shift pertaining to either the phase p = 1 or p = 2 takes on the

form

Ap = m Ap + a + bx + cy
Sp = mSp a bx cy. (5.85)
Observing that only the boundary conditions (5.84) depend on the compli-
ances A1 , ..., S2 , it was found that, in general, shifts (5.85) are possible in both
plane strain and plane stress for either m = 1 or m = 1, providing specific,
additional conditions hold. Moreover, these results carry over to both simply
and multiply connected two-phase materials, whereby for the multiply con-
nected materials with perfectly bonded interfaces, the Cesaro integrals do not
need to be considered (Markenscoff, 1996).
5.3 Poroelasticity
This theory was introduced to incorporate the change of volume fraction as
an additional degree of freedom (Goodman and Cowin, 1972; Cowin and
Nunziato, 1983). The constitutive equations in 3D are:
i j = kk i j + 2i j + i j h k = ,k g = rr , (5.86)
P1: Binod
May 28, 2007 18:4 C4174 C4174C005
where new quantities, relative to the classical elasticity, are the equilibrated
stress vector h k , the intrinsic equilibrated body force g, the change in volume
fraction from the reference volume fraction . The Cauchy stress tensor i j is
symmetric here and the strain-displacement relation is just as in the classical
case (i, j = u(i, j) ). The equilibrium equations are
i j, j = 0 h i,i + g = 0. (5.87)
Clearly, , , and in (5.86) are the new material constants besides the con-
ventional and , and, at this point, one assumes the internal energy density
to be a positive quadratic form, thus implying the inequalities
(3 + 2) 3 2 > 0 >0 >0 > 0. (5.88)
Recently, De Cicco and Guarracino (2004) have formulated plane elasto-

statics of such a material in terms of stress functions. Thus, we have two
functions and of class C 6 and C 4 , respectively, such that the definitions
12 = 12 22 = ,11 12 = ,12
(5.89)
g = 2 h 1 = ,1 h 2 = ,2 .
result in (5.87) being satisfied identically. Clearly, plays the same role as the
familiar Airy stress function of classical elastostatics. It has been shown that
the following four equations result from the above relations:
2u1,1 = ,22 2 (v + ) 2u2,2 = ,11 2 (v + )

(5.90)
2u(1,2) = ,12 =
1 2
G [2 ( + ) ] ,
where

G = 2 ( + ) v = 2 /G = 2/G. (5.91)
Note a certain similarity of (5.90) to (6.108) of Chapter 6. From this,

assuming the function
P1 = 2 [(1 v) ] (5.92)
to be harmonic, one can show that and satisfy the equations

4 2 p2 = 0 2 2 p 2 = 0, (5.93)
which again bear some resemblance to (6.109) of Chapter 6. The coefficient

p 2 = ( + 2) (5.94)
( + 2)
P1: Binod
May 28, 2007 18:4 C4174 C4174C005
appearing in equations (5.93) has dimension of inverse of length squared, so

that there is also a characteristic length in Cowins theory:
1
l= . (5.95)
p
De Cicco and Guarracino (2004) also established a representation of func-

tions and in terms of a pair of complex analytic functions and a real
function that satisfies a homogeneous Helmholtz equation. It now remains
to check whether there is stress invariance in this theory, and, if so, of what
kind.
Problems
1. Prove properties 4. and 5. of Section 5.1.3.
2. What does the Cauchy symmetry imply for the polar representation
of a general anisotropic planar material?
3. Dundurs parameters may be written as
(1 + 1) (2 + 1) (1 1) (2 1)
12 = 12 =
(1 + 1) + (2 + 1) (1 + 1) + (2 + 1)
Where = 2 /1 and 1 , 2 are two Kolosov constants. Show that

they are invariant under the CLM shift, or that they agree with (5.44).
4. Consider a unidirectional fiber-reinforced composite with a dilute
concentration of fibers. The fibers are in the shape of circular cylin-
ders; their volume fraction is c 2 . The effective shear and bulk moduli
are given as follows

1 1 1 1 2(1/1 + 1/1 )
= + c2
eff 1 2 1 1/1 + 1/2 + 2/2

1 1 1 1 2(1/1 + 1/1 )
= + c2
eff 1 2 1 1/2 + 1/1
Show that these effective moduli satisfy the CLM shift.

5. Consider the following boundary value problems involving a lin-
early elastic plate (made of an isotropic, homogeneous material)
with a hole:
(a) the plate subjected to a remote uniaxial tension;
(b) the hole loaded by two concentrated and opposite forces on
its boundary;
P1: Binod
May 28, 2007 18:4 C4174 C4174C005
(c) the hole loaded by one concentrated force;

(d) the hole loaded by a uniformly distributed torque on its
boundary;
(e) the plate subjected to a displacement boundary condition
ui (x) = i0j x j .
Determine in which of these problems is the stress field independent
of the elastic constants.
6. Solve the problem of a large sheet with a circular disk. The sheet
and the disk are linearly elastic, isotropic and homogeneous. They
are made of distinct elastic constants and have different thermal
expansion coefficients. Show that the stress field invariant under
the CLM shift. Hint: the problem is axisymmetric.
7. Demonstrate that the stress invariance does not hold in anti-plane
elasticity.
8. Verify equation (5.36).
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
6
Two- versus Three-Dimensional
Micropolar Elasticity
Of course the continuum theory can yield, in principle, less

information about a material than does that of a correct, detailed
molecular model.
C.A. Truesdell, 1966
It was noted in Chapter 3 that lattices of beams should be modeled by

micropolar, rather than classical, elastic continua. This chapter outlines the
basic theory of such continuum modelsboth unrestricted and restricted
(couple-stress) ones. In particular, we first provide a formulation of basic equa-
tions in 3D, and then, in analogy to Chapter 3, we focus on planar micropolar
elasticity. Special attention is given to a generalization of the CLM result, and
its consequences. Furthermore, we discuss the problem of homogenization of
a heterogeneous Cauchy-type composite by a homogeneous micropolar-type
material. If conducted properly, one may then reduce the number of degrees
of freedom involved in, say, a finite element method, although the method
would have to account for a micropolar nature of the approximating body.
This also provides more physical insight into the so-called characteristic length,
usually an enigmatic concept appearing (and vanishing) in papers on micro-
polar theories. Although several monographs have been written on the subject
of micropolar media, most of the topics discussed in this chapter have never
been collected in a book form.
6.1 Micropolar Elastic Continua

6.1.1 Force Transfer and Degrees of Freedom
Every course on solid mechanics starts out with an introduction of the Cauchy
stress concept. This first involves identification of a finite surface area A
(= L 2 )either in the interior of the body or on its external surfacedefined
by an outer unit normal n, and a force F acting on A, Figure 6.1(a). Next,
191
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
(a) (b)
FIGURE 6.1
(a) Force F and couple M acting on an internal (or external) surface area A (= L 2 ) in a
continuum; A is the area of any face of a cubic element of side L. (b) A porous medium in 2D,
viewed as a beam lattice, with each beam carrying a force and a couple locally. A unit cell of size
L is indicated with dashed lines.
one considers the ratio of F to A, and takes the limit

F(n)
lim = t(n) . (6.1)
A0 A
It is a basic postulate of conventional solid mechanics that such a limit is well

defined, that is, that it is finite except the singularity points in the body, such
as crack tips. In the third step, following Cauchy himself, one introduces his
force-stress tensor as a linear mapping from n into t(n)
t(n) = n. (6.2)
However, any consideration of a finite area A should involve a surface

couple M accompanying F. Thus, in analogy to (6.1), we must consider
M(n)
lim = m(n) , (6.3)
A0 A
and, following Voigt (1887) and the brothers Eugene and Francois Cosserat
(1909), should introduce a couple-stress tensor as a linear mapping from n
into m(n)
m(n) = n. (6.4)
Both t(n) and m(n) are shown acting on a face ABC of an arbitrary orientation
in Figure 6.2.
Note: The explicit consideration of makes nonsymmetric in general,
and that is the reason for using instead of the conventional Cauchy
stress .
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
Two- versus Three-Dimensional Micropolar Elasticity 193
x3 t(n)
m(n)
t(1)
C
m(1) n
x2
m(2)
B
t(2) O
A
m(3)
t(3) x1
FIGURE 6.2
Force traction and moment traction acting on face ABC with outer unit normal n of an infinitesimal
tetrahedron OABC.
If the microstructure is disregarded, we are dealing with an idealized,

homogeneous continuum in which M(n) must vanish in the limit L 0. To
see this, take n to be aligned with n1 , and consider shear stresses 12 and 13 .
The torque caused by them, proportional to L 3 (12 13 ), must disappear as
L 0, because the cubes volume scales as L 3 . Otherwise, we would be left
with a non-zero angular acceleration of a continuum point. This, in fact, is the
case with classical/conventional solid mechanics of Cauchy-type continua.
One then only has displacement u at a point, and assumes that m(n) = 0. But,
if the material intrinsically carries couples, we cannot disregard M(n) . Such a
situation occurs when the material has a discrete-type microstructure, such as
a beam-lattice shown in Figure 6.1(b), which simply precludes one from taking
A 0. Here one needs to take A equal to the area of the elementary cells
cross-section, and the moment traction m(n) in (6.3) is defined at A finite.
For this model of force distribution in a continuous body to be fully con-
sistent with kinematics, each continuum point is endowed with six degrees of
freedom of a rigid body: three displacements ui (i = 1, . . ., 3) (or u) and three
rotations i (i = 1, . . ., 3) (or ), which are, in general, independent functions
of position and time. In particular, this implies that is not the same as the
macrorotation given by the gradient of u:
1
i =
e ijk uk, j (6.5)
2
Here, as before, e ijk is the LeviCivita permutation tensor.
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
When the inequality above is replaced by an equality, the continuum is

kinematically constrainedthis will be discussed at various points in this
chapter. We now give some key facts in the vein of Nowacki, (1986).
6.1.2 Equations of Motion and Constitutive Equations

The equations of motion and constitutive equations of linear elasticity may
be derived from the energy conservation principle. This principle, for an adi-
abatic thermodynamic process and an anisotropic body, has the following
form:

d
(U + K ) = ( Xi vi + Yi wi ) dV + (ti vi + mi wi ) dS, (6.6)
dt V S
where

1
U= ud V K = kd V k= vi vi + Iij wi w j vi = ui wi = i .
V V 2
(6.7)
In the above k is the the kinetic energy density, and u the internal energy,
both referred to a unit volume, while denotes mass density and Iij is the
rotational inertia tensor. Note that the left-hand side of (6.6) represents the
rate of change of kinetic and internal energies, while the right-hand side is
the power of body forces and moments and surface forces and moments.
Let us now assume that the energy balance is invariant with respect to
rigid body motions when Xi , ti , Yi , and mi are kept fixed. Considering a
translational motion first, we substitute (with b i being an arbitrary constant
vector)
vi vi + b i , vi vi (6.8)
into (6.6) to get

u + (vi + b i ) vi + Iij wi w j dV = [Xi (vi + b i ) + Yi wi ] dV
V V

+ [ti (vi + b i ) + mi wi ] dS. (6.9)
S
Subtracting equation (6.6) from (6.9), we obtain

bi ( Xi vi ) dV + b i ti dS = 0, (6.10)
V S
which, noting (6.2) and the GreenGauss theorem, becomes

bi Xi + ji,j vi dV = 0. (6.11)
V
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
Since this has to hold for an arbitrary volume V, we obtain a local form of the
conservation of linear momentum
ji,j + Xi = vi . (6.12)
With the above we can now simplify (6.6) to the form

u + Iij wi w j dV = ji vi + Yi wi dV + [ti (vi + b i ) + mi wi ] dS,
V V S
(6.13)
which, noting (6.3) and the GreenGauss theorem, becomes

udV = ji vi, j + ji wi, j dV + ji,j + Yi Iij w j wi dS. (6.14)
V V S
From this, the local form of conservation of energy may now be written as

u = ji vi, j + ji wi, j + ji,j + Yi Iij w j wi . (6.15)
Let us now postulate the energy balance to be invariant with respect to

rigid body rotations (with k being an arbitrary constant vector):
vi, j vi, j e ijk k wi, j wi, j . (6.16)
Assuming u, Yi , ij , ij , and Iij to be unchanged, and proceeding in a fashion

similar as before, we arrive at a local form of the conservation of angular
momentum
..
e ijk jk + ji,j + Yi = Iij j . (6.17)
This equation brings about a simplification of the energy balance

u = ji vi, j e k ji wk + ji wi, j . (6.18)
It is now convenient to introduce two tensors describing the deformation

of the bodystrain ji and torsion ij as follows:
ji = ui, j e k ji k ji = i, j . (6.19)
Evidently, ji and ij are generally nonsymmetric; ij is also called curvature

tensor, or torsion-curvature tensor. Just like in a classical continuum, we need
compatibility equations, and these are
li,h hi,l e khi ik + e kli hk = 0 li,h = hi,l . (6.20)
Given (6.19), we can write the energy balance (6.18) as
u = ij ij + ij ij . (6.21)
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
Taking u to be a state function of ij and ij , we have
u u
u = ij + ij , (6.22)
ij ij
whereby we also assume ij and ij not to be explicitly dependent on the

temporal derivatives of ij and ij . A comparison of (6.21) with (6.22) then
leads to
u u
ij = ij = . (6.23)
ij ij

Clearly, ij , ij and ij , ij are conjugate pairs. Assuming a micropolar ma-
terial of linear elastic type, its energy density is given by a scalar product
1 (1) 1 (2)
u= ij Cijkl kl + ij Cijkl kl , (6.24)
2 2
so that Hookes law is
(1) (2)
ij = Cijkl kl ij = Cijkl kl , (6.25)
(1) (2)
Here Cijkl and Cijkl are two micropolar stiffness tensors. Note that, due to
the existence of u, we have the basic symmetry of both stiffness (and hence,
compliance) tensors
(1) (1) (2) (2)
Cijkl = Cklij Cijkl = Cklij , (6.26)

but not the two other symmetries since ij , ij and ij , ij are, in general,
nonsymmetric. Indeed, this is the reason for calling this theory an asymmetric
elasticity by Nowacki (1970, 1986). The inverse of (6.25) is written as
(1) (2)
ij = Sijkl kl ij = Sijkl kl . (6.27)
Note: In this chapter we employ and to distinguish them, respectively,

from the symmetric stress and symmetric strain of classical continuum
theory. This convention will become very useful in homogenization of a
heterogeneous Cauchy-type composite by a homogeneous micropolar-type
material, Section 6.4.
6.1.3 Isotropic Micropolar Materials

Focusing henceforth on the centrosymmetric case, for an isotropic material,
(1) (2)
Iij = ij I , while tensors Cijkl and Cijkl of (6.25) become
(1)
Cijkl = ( ) jk il + ( + ) jl ik + ij kl
(2)
Cijkl = ( ) jk il + ( + ) jl ik + ij kl ,
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
where and are the Lame constants of classical elasticity, while , , , and
are the micropolar constants. The free energy density is given by a scalar
product
+
u= ji ji + ji ij + kk nn
2 2 2
+
+ ji ji + ji ij + kk nn , (6.28)
2 2 2
With (6.28), we can write (6.25) in two equivalent forms:
ji = ( + ) ji + ( ) ij + ij kk ij = 2(i j) + 2[i j] + ij kk
ji = ( + ) ji + ( ) ij + ij kk ij = 2 (i j) + 2[i j] + ij kk .
(6.29)
The round and square brackets indicate symmetric and antisymmetric parts
of the tensors, respectively. Of use also will be the inverse forms of this con-
stitutive law, namely,
ij = 2 (i j) + 2 [i j] + ij kk ij = 2 (i j) + 2 [i j] + ij kk , (6.30)
in which
1 1 1 1
2 = 2 = 2 = 2 =
2 2 2 2
(6.31)
2 2
= = =+ = + .
6K 6 3 3
Here we recognize the familiar bulk modulus , and its mathematically anal-
ogous micropolar quantity .
Clearly, there are six material constants: , , , , , . Considering the
fact that u in (6.28) is a positive definite quadratic form, one can show that
the following inequalities should hold:
3 + 2 > 0 >0 3 + 2 > 0 > 0

(6.32)
+ >0 + >0 >0 > 0.
Our constitutive tensors (6.26) may alternatively be expressed in the

notation of Eringen (1966, 1999)
(1)
Cijkl = E jk il + ( E + E ) jl ik + E ij kl
(2)
(6.33)
Cijkl = E jk il + E jl ik + E ij kl ,
where, using the subscript E to denote quantities in Eringens notation, we

have
E = E = 2 E = E = + E = . (6.34)
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
We end this section by noting that, just like in the classical elasticity, we
can express a micropolar field problem in displacements and rotations, or in
stresses and couple-stresses. In the first case, six such equations (for ui and i ,
i = 1, . . ., 3) are obtained by substituting (6.30) into the equilibrium equations
(6.12) and (6.17), and using (6.19):
..
( + ) ui, j j + ( + ) u j, ji + 2e ijk k + Xi = ui
.. (6.35)
( + ) i, j j 4i + ( + ) j, ji + 2e ijk uk + Yi = I .
This generalization of the Navier equations of classical elasticity is to be sup-

plemented by the kinematic boundary conditions on Bk and traction condi-
tions on Bt , where Bk Bt = B, that is,
ij (x, t) n j = ti (x, t) ij (x, t) n j = mi (x, t) x Bt t>0

(6.36)
ui (x, t) = f i (x, t) ui (x, t) = gi (x, t) x Bk t > 0.
Here ti , mi , f i , gi are prescribed functions.

The field equations in stresses are a generalization of the BeltramiMichell
equations. Here we make a reference to a study of Schafer (1967), who gen-
eralized the functions of Morrey and Maxwell and Kessel. We will return to
this topic in the 2D setting in Section 6.3 below.
As pointed out in Section 6.1.1, is an independent kinematic quantity.
However, a special model assuming the equality
1
i = e ijk uk, j (6.37)
2
is sometimes used, and this is the same definition as in classical elasticity. It

is called a pseudo-continuum, a restricted, a couple-stress, or a Koiter-Mindlin
model, Koiter (1963); see also Truesdell and Toupin (1960), Grioli (1960),
Toupin (1962), Mindlin and Tiersten (1962), and Mindlin (1963). In view of
(6.37), the strains ij become symmetric and are (classically) defined as
ij = u(i, j) . (6.38)
A well-known case of such materials is the BernoulliEuler beamindeed,

a 1D continuumin which the cross-section is restricted to rotate according
to the gradient of the transverse displacement.
6.1.4 Virtual Work Principle

Let us consider fields of virtual displacement ui and rotation i both of
them continuous, consistent with boundary conditions, and infinitesimal. If
we multiply equation (6.12) by ui , and equation (6.17) by i , and add the
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
results, we find

.. ..
[( Xi ui )ui + (Yi I i )i ]dV
V

+ [ji,j ui + (e ijk jk + ji,j )i ]dV = 0. (6.39)
V
Upon transformation of the second of the above integrals, we obtain

.. ..
( Xi ui )ui dV + (Yi I i )i dV
V V

+ [ti ui + mi i ] dS = ji ji + ji ji dV, (6.40)
V V
which expresses an equality between the virtual work of external and internal
forces; the latter are conjugate to fields of virtual strain ji and rotation ji .
Upon introduction of (6.29) into (6.39), we set up the variational principle

.. ..
[( Xi ui )ui + (Yi I i )i ]dV
V

+ [ti ui + (mi i ]d S = W (6.41)
S
where

W= (i j) (i j) + [i j] [i j] + kk mm
V 2

+ (i j) (i j) + [i j] [i j] + kk mm dV. (6.42)
2
This principle may be used to derive the energy conservation principle by
comparing the functions u and at a point x and time t with those quantities at
x and time t + dt. Thus, introducing ui = vi dt; i = wi dt; vi = ui ; wi = i
into (6.41), we obtain

d
( K + W) = ( Xi vi + Yi wi ) d V + (ti vi + mi wi ) d S (6.43)
dt V S
This is the starting point for the proof of uniqueness of solutionsthe proce-
dure is analogous to that in classical elasticity.
6.1.5 Hamiltons Principle

Consider now a micropolar elastic body undergoing some motion between
times t = t1 and t = t2 . We now compare the actual displacements u (x, t) and
rotations (x, t) with the virtual u (x, t) + u and (x, t) + , whereby the
latter are chosen so as to satisfy the conditions
u (x, t1 ) = u (x, t2 ) = 0 (x, t1 ) = (x, t2 ) = 0. (6.44)

P1: Binod
May 28, 2007 18:6 C4174 C4174C006
The virtual work principle is now written as

.. ..
L ( ui ui + I i i )dV = W. (6.45)
V
where

L = ( Xi ui + Yi i ) dV + (ti ui + mi wi ) dS. (6.46)
V S
Integrating (6.45) over the interval t1 t t2 ,

t2 t2 t2
.. ..
Wdt = Ldt dt ( ui ui + I i i )dV, (6.47)
t1 t1 t1 V
while introducing the variation of kinetic energy,

.. ..
K = ( ui ui ) dV ui ui dV + I ( i i ) dV I i i dV,
V t V V t V
(6.48)
and integrating it also over t1 t t2 , and taking note of (6.44), we find

t2 t2 t2
.. ..
K dt = dt ui ui dV I dt i i dV. (6.49)
t1 t1 V t1 V
In view of (6.47) and (6.49), we finally obtain Hamiltons principle generalized

to a micropolar medium
t2 t2
(W K ) dt = Ldt. (6.50)
t1 t1
Variation and integration on the right-hand side of (6.50) commute when

the external forces are conservative and derivable from a potential V. In that
case,

V V V V
L = ui + i = ui + i , (6.51)
ui i ui i
and (6.50) becomes

t2
( K ) dt = 0 =W+V (6.52)
t1
in which denotes the total potential energy.
6.1.6 Reciprocity Relation

Consider an isotropic body of volume V and bounding surface B loaded by
a system {X, Y, t, m}, which then results in {u, }. The initial conditions are
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
homogeneous:
ui (x, 0) = 0 ui (x, 0) = 0
i (x, 0) = 0 i (x, 0) = 0 (6.53)

Also, consider loading system X , Y , t , m acting on the same body,

another
resulting in u , , both causes and effects being now denoted by primes.
This is subject to analogous initial conditions as in the first case.
We now apply Laplace transformation to the constitutive equations (6.29)
to get
ji = ( + ) ji + ( ) ij + ij
kk
(6.54)
ji = ( + ) ji + ( ) ij + ij
kk ,
where

ji (x, p) = ji (x, t) e pt dt, etc. (6.55)
0
Proceeding in a similar fashion with the primed quantities ji (x, p), etc., a
statement entirely similar to (6.54) above is obtained.
It is easy to verify that the following is true:
ji ji +
ji ji = ji ji +
ji ;
ji (6.56)
and this, upon a volume integration, becomes

ji ji +
ji ji dV = ji dV.
ji ji + ji (6.57)
V V
Next, carry out the Laplace transformation on the equations of motion (6.12)
and (6.17) corresponding to the first loading system to get
i
ji,j + Xi = p u
2
e ijk
jk +
i .
ji,j + Yi = p I
2
(6.58)
Noting a completely analogous relation corresponding to the second loading

system, (6.57) can be converted to

Xi ui + Yi i p ui p I i dV +
2 2
i i dS
ti ui + m
V S

i dS.
= Xi ui + Yi i dV + ti ui + mi (6.59)
V S
Upon carrying out the inverse Laplace transformation, we arrive at the the-
orem of reciprocity of work of causes and effects in both loading systems
(Sandru, 1966):

Xi ui + Yi i dV + ti ui + mi i dS
V S

= Xi ui + Yi i dV + ti ui + mi i dS, (6.60)
V S
where denotes a convolution operation between Xi and ui .

P1: Binod
May 28, 2007 18:6 C4174 C4174C006
This is one of the most interesting theorems of micropolar elasticity. Its

generality offers a possibility of integration of the equations of elastodynamics
using Greens function.
6.1.7 Elements of Micropolar Elastodynamics

6.1.7.1 Basic Equations
We begin by noting that equations (6.35) can also be written in a vector form
2u + ( + ) grad divu + 2 rot + X = 0

(6.61)
4 + ( + ) grad div + 2 rotu + Y = 0,
where
2 = ( + ) 2 2t 4 = ( + ) 2 4 + I 2t , (6.62)
are the dAlembert and KleinGordon operators, respectively, and 2t indicates

the second derivative with respect to time.
The physics represented by this coupled system of hyperbolic differential
equations can be understood by operating either with divergence or rotation
upon it. In the first case, we find
1 divu + divX = 0
(6.63)
3 div + divY = 0,
where again we introduced two partial differential operators
1 = ( + 2) 2 2t 3 = ( + 2 ) 2 4 + I 2t . (6.64)
On the other hand, upon carrying out a rotation on (6.61), we find
3 r otu + 2 rot rot + r otX = 0

(6.65)
3 r ot + 2 rot rotu + r otY = 0.
If we now operate with 1 4 on (6.61)1 and employ (6.63)1 and (6.65)1 ,

and similarly we operate with 3 4 on (6.62)2 and employ (6.63)2 and (6.65)2 ,
we shall find

2 4 + 4 u = 1 4 grad div X + 2 rot 1 Y
2 2
1
(6.66)
2 4 + 4 = 2 3 grad div Y + 2 rot 1 X,
2 2
3
where
= ( + ) 4 4 2 = ( + ) 2 4 2 . (6.67)
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
Now, let us introduce a representation for displacements and rotations,

employing two vector functions F and G:

u= 1 4 grad div F 2 rot 3 G
(6.68)
= 2 3 grad div G 2 rot 1 F.
With the substitution of (6.68) into (6.66) we find two equations governing F
and G:

1 2 4 + 4 2 2 F + X = 0
(6.69)
4 + 4 G + Y = 0.
2 2
3 2
More insight into what is represented by the equations of motion (6.61)

may be gained by using Helmholtzs theorem for u and :
u = grad + r ot , div = 0
(6.70)
= grad + r ot H, div H = 0.
If we also introduce the same type of decomposition for body forces and
body couples, that is,

X = grad + r ot , div = 0
(6.71)
Y = I grad + r ot , div = 0.
we find that (6.61) reduces to four wave equations
1 + = 0 3 + I = 0
(6.72)
2 + 2 rotH + = 0 4H + 2 rot + I = 0.
The first of these equations represents a longitudinal wave motion, identi-

cal with what is known from classical elastodynamics. The second equation
represents a longitudinal wave of (micro)rotation. The third and fourth equa-
tions represent propagation of two transverse wavesin displacements and
in rotationswhich may be cast in the form

2 4 + 4 2 2 = 2 I rot 4
(6.73)
4 + 4 H = 2 rot I 2 .
2 2
2
These waves were investigated by Ignaczak (1970).

P1: Binod
May 28, 2007 18:6 C4174 C4174C006
6.1.7.2 Plane Monochromatic Waves

Consider a plane monochromatic wave propagating in an infinite body, and
harmonic in time. Let the wavefront at time t = 0 be in the plane p = xi ni ,
where n is the unit vector normal to the plane. The displacements and rota-
tions are now of the form
u j = Aj exp [ik (t nk xk )] j = B j exp [ik (t nk xk )] , (6.74)
where k = / = 2/l is the phase velocity, is the angular velocity, and

l is the wave length. Introducing (6.74) into (6.35), we arrive at a system of
algebraic equations
2i
+ 2 Aj + ( + ) n j xk Ak + e jkl nl Bk = 0,
k
(6.75)
4 2i
+ + 2 I 2 B j + ( + ) n j xk Bk + e jkl nl Ak = 0.
k k
Setting this systems determinant to zero leads to

4
+ 2 2
+ 2 + 2 I 2
k
(6.76)
4 4 2
+ 2 + + 2 I 2 2 = 0,
k k
from which we determine phase velocities of various plane waves. The first
term in (6.76) yields
1/2
+ 2
= (6.77)

The second term in (6.76) yields dispersive (i.e., -dependent) wave propa-
gation
1/2 1/2
2 + 2 4
= 3 1 02 , 3 = , 02 = , (6.78)
I I
which has physical meaning only for > 0 , because this condition ensures
real values of . The third term in (6.76) yields a quartic equation

4 4 2
+ 2 + + 2 I 2 2 = 0. (6.79)
k k
6.1.8 Noncentrosymmetric Micropolar Elasticity

Recall that the constitutive equations relate force stresses with strains on one
hand, and moment stresses with curvatures on the other. In general, however,
there is a possibility of a direct coupling between force-type and couple-type
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
effects in the constitutive model, whereby ij would also be a function of kl ,

and ij a function of kl . A simple 1D case of chirality is the helix already
discussed in Section 3.5 of Chapter 3. Thus, generalizing (6.25), we write
(1) (3)
ij = Cijkl kl + Cijkl kl ,
(4) (2)
(6.80)
ij = Cijkl kl + Cijkl kl .
This is called either non-centrosymmetry or hemitropy (Aero and Kuvshinskii

1964, 1969; Nowacki, 1986), or chirality (Lakes and Benedict, 1982; Lakes, 2001).
Note that all the developments of this chapter preceding this equation per-
tained to centrosymmetric materials. Note that, by the argument of reciprocity,
(3) (4)
Cijkl = Cijkl . (6.81)
As Lakes points out, because the coordinate changes in the centrosymmetric

material result in the transformation matrix
a im = jm , (6.82)
(1)
the stiffness tensor Cijkl satisfies
(1) (1) (1)

Cijkl = a im a jn a ko a lp Cmnop
(1)
= (1) 4 Cijkl = Cijkl . (6.83)
6.2 Classical vis-a-vis Nonclassical (Elasticity) Models

6.2.1 A Brief History
Following its birth, the theory of the Cosserat brothers (1896, 1909) remained
dormant for half a century, apparently the only exceptions being the works
of Somigliana (1910) and Sudria (1935); see also Ball and James (2002). This
hibernation was likely due to the theorys generality (as a nonlinear theory
with finite motions and inelastic interactions) and its presentation as a uni-
fied theory incorporating mechanics, optics, magnetism and electrodynam-
ics. The dynamic growth of continuum mechanics and thermodynamics (e.g.,
Ericksen and Truesdell, 1958; Truesdell and Toupin, 1960) begun in the fifties
and sixties brought the work of the Cosserat brothers back into focus. Funda-
mentals of a general linear Cosserat continuum were given by Gunther (1958),
who discussed in detail the 1-, 2-, and 3D Cosserat models, as well as their
significance in the dislocation theory, and Schafer (1962), who focused on the
planar case. From that period one should also mention several other works.
Thus, Grioli (1960) established the constitutive relations for finite deforma-
tions of perfectly elastic solids. Aero and Kuvshinskii (1960) independently
derived the equilibrium equations and constitutive relations for anisotropic
solids in the linearized theory. Mindlin and Tiersten (1962) established the
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
boundary conditions; see also Kroner (1963), Koiter (1963), and Eringen (1968).
An expression of the growing interest in Cosserat theory was soon found in
symposia (e.g., Kroner 1968) and monographs on the subject (e.g., Nowacki,
1970, 1986; Stojanovic, 1970; Brulin and Hsieh, 1982).
Building on the shoulders of the Cosserats, and to account for increasing
levels of complexity, other, more general theories accounting for higher-order
interactions such as monopolar, multipolar, and strain-gradient were intro-
duced (see e.g., Green and Rivlin, 1964; Toupin, 1962, 1964; Jaunzemis, 1967;
Tiersten and Bleustein, 1974). There are also micropolar, microstretch,
and most generally micromorphic continua (Eringen, 1999, 2001; Mariano,
2001). To clarify the key concepts here, following Goddard (2006), let us con-
sider a series expansion of the velocity (or infinitesimal displacement) field v
v (x) = v0 + L1 r + L2 r2 + , (6.84)
where
r = x x 0 , Ln = 1
n! ( v) 0T , (6.85)
and rn denotes the n-fold symmetric tensor product n r. This allows an

expansion for the global stress-power density in a simple continuum

1
w = : LdV = wn , (6.86)
V V n
with

wn = n : Ln , n := V : rn dV, (6.87)
where L = ( v) T is is the first velocity gradient, its dual being the Cauchy
stress . Furthermore, while we easily see that the higher-order kinematic
quantities Ln are conjugate to the stress moments n , there are two ways to
interpret the Ln :
1. Multipolar: the Ln are identical with the higher gradients of the
velocity field. This viewpoint was advanced by Green and Rivlin
(1964a,b) and Mindlin (1963).
2. Micromorphic: the Ln are intrinsic particulate fields (i.e., pertaining
to generally deformable particles making up the macro-continuum),
which require their own constitutive equations. This approach dates
back to the Cosserat brothers, and was then further pursued by
Eringen (1999) leading to microstretch and micromorphic theories.
The microscopic treatment dictates that, in a multipolar continuum, the
stress moments should satisfy a hierarchy of balances
n+1
T
+ n = Gn+1 for n = 0, 1, . . ., (6.88)
where 0 := 0, 1 := , and the Gs represent extrinsic body moments accu-

mulation of intrinsic multipolar momenta, which effectively vanish in the
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
quasi-static limit. Equation (6.88) leads to the following balance:

V n + x n1 + + x dS = V Gn dV for n = 2, 3, . . .,
n1
(6.89)
In a certain sense, all of these theories can be considered as simpler cases of

nonlocal continuum theories (Eringen and Hanson, 2002), which, according
to these authors, are concerned with material bodies whose behavior at any
interior point depends on the state of all other points in the bodyrather than
only on an effective field resulting from these pointsin addition to its own
state and the state of some calculable external field. Focusing henceforth on
micropolar theories, we would like to note their extensions beyond purely
elastic material behaviors. An extension pertaining to thermoelasticity was
already given by Nowacki (1966) and Tauchert et al. (1968); see Dhaliwal
and Singh (1987) for a review. A micropolar generalization of viscoelasticity,
with a focus on waves, was presented by Maugin (1974). Beginning with
Green and Naghdi (1965), Misicu (1964), and Sawczuk (1967), there has also
been research on (elastic-)plastic continua with microstructure, e.g., Fleck
et al. (1994) and Hutchinson (2000). This has then led to strain-gradient models
(Aifantis, 1987; Zbib and Aifantis, 1989; Fleck and Hutchinson, 1997; Pamin,
2005). Extensive research has also been done on micropolar fluid mechanics
(e.g., Cowin, 1974; Eringen, 2001).
Although the nonclassical theories have become very advanced mathe-
matically and explained effects that could not be brought out by classical
theories, they usually lacked the input of physically based constitutive coef-
ficients. Besides the beam lattices discussed in Chapters 3 and 4, progress has
been made on that front for composite materials; see Section 6.5 below.
Mindlin (1963) found that stress concentrations in the presence of holes
are lowered in Cosserat-type versus those in Cauchy-type solids. This was
followed by studies due to Neuber (1966) Kaloni and Ariman (1967), Cowin
(1970a,b) and Itou (1973). On the other hand, an increase of stress concentra-
tions in the vicinity of rigid inclusions was established by Hartranft and Sih
(1965) and Weitsman (1964). Micropolar effects also allow a better analysis of
localization present in failure of solids then that possible in classical continue
(William, 1995).
The case of holes motivated one of the earliest experimental studies of
couple-stress effects by Schijve (1966), who actually found that effect to be in-
significant. However, given the fact that he used aluminum sheetsa macro-
scopically homogeneous material without, say, reinforcing inclusionshis
investigation pertained to couple-stress effects due to the atomic lattice of
aluminum. This is not surprising in view of the fact that couple-stress effects
vanish on scales much larger than the microscale. Indeed, the situation is much
different in, say, a lattice of beams (which may be interpreted as a material
with large holes), if one looks at dependent fields on scales comparable to the
lattice spacing; recall Chapter 3.
Many studies of wave propagation in the context of harmonic distur-
bances have also been conducted. First, in addition to classical dilatational
and shear waves in an unbounded medium, there also exist rotational waves.
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
Next, it turns out that only the dilatational waves propagate non-dispersively
(Nowacki, 1986; Eringen, 1999). In general, this is indicative of various new
dispersion effects in other wave problems, which are not present in classical
continua. In some cases of Cosserat continua, entirely new phenomena arise
such as, for instance, that a layer on top of an elastic half-plane is not necessary
for the propagation of Love wavesin the classical case, a layer is necessary.
Many results on periodic and aperiodic waves were collected by Nowacki
(1986), see also Eringen (1999).
The recent monograph by Dyszlewicz (2004) on micropolar elasticity
collects many new results, including the general methods of integration of
basic equations (Galerkin, GreenLame, and PapkovitchNeuber type), for-
mulations of problems (displacement-rotation and pure stress problems of
elastodynamics), as well as solutions to various boundary value problems
(stationary 2D and 3D problems for a half-space, singular solutions to 2D and
3D elastodynamics and the thermoelastodynamics problems for an infinite
space).
Several workers, in the 1960s, derived micropolar models explicitly from
the microstructure. The work of theoreticians started from lattice-type models
enriched with flexuralin addition to centralinteractions (e.g., Askar, 1985;
Banks and Sokolowski, 1968; Wozniak, 1970; Bazant and Christensen, 1972;
Holnicki-Szulc and Rogula, 1979a,b; Bardenhagen and Triantafyllidis, 1994).
From the outset, these models adopted Cosserat-type continua in analyses
of large engineering structures such as perforated plates and shells, or lat-
ticed roofs. There, the presence of beam-type connections automatically led to
micropolar interactions and defined the constitutive coefficients. In principle,
such models have their origin in atomic lattice theories (e.g., Berglund, 1982);
see Friesecke and James (2000) for the latest work in that direction.
Several workers (e.g., Perkins and Thompson, 1973; Gauthier and
Jahsman, 1975; Yang and Lakes, 1982; Lakes, 1983, 1986) have provided ex-
perimental evidence of micropolar effects in porous materials such as foams
and bones. In particular, Lakes (1995) was able to infer micropolar constants
from his experiments, both for centrosymmetric and chiral materials. Another
interesting application in the context of biomechanics was due to Shahinpoor
(1978).
It is also to be noted that composite materials may naturally lead to
Cosserat models where the nonclassical material constants can directly be
calculated from the microstructure; this was done in 1D by Herrmann and
Achenbach (1968). But, a similar task in 2D and 3D has only been undertaken
recently, and this is described in Section 6.5. In more recent years, progress
has been made on derivation of effective (homogeneous) Cosserat models
for heterogeneous composite materials of either Cauchy or Cosserat type. We
point out in Chapter 4 that a central-force lattice (truss of two-force members)
is an example of the former material, while a lattice of beams is an example
of the latter one.
All the studies in the area of stress singularities due to cracks were pre-
ceded by Muki and Sternberg (1965), who studied stress concentrations caused
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
by concentrated surface loads or discontinuously distributed surface shear

tractions. Next, Sternberg and Muki (1967) and Bogy and Sternberg (1967)
studied the implications of the couple-stress theory on unbounded concen-
trations of stress and on locally infinite deformation gradients. Basically, it
was found that, depending on a given situation, where the classical elasticity
would predict infinite (singular) stresses, the couple-stress theory may give
either finite stresses or weaker singularities, or have an opposite tendency (see
also Cowin, 1969; Atkinson and Leppington, 1977). This involves a proper
generalization of conservation integrals, which has recently been given in the
setting of couple-stress elasticity (Lubarda and Markenscoff, 2000).
Recently, Griffiths fracture theory has been generalized to rectilinear and
fractal cracks in micropolar solids (Yavari et al., 2002). In particular, two cases
of the Griffith criterion were considered, depending on whether the effects of
stresses and couple-stresses are coupled or uncoupled, the key finding being
that both cases give equal orders of stress and couple-stress singularities,
which is the same result as that in a classical continuum. Also, the effect
of fractality of fracture surfaces on the powers of stress and couple-stress
singularity was studied.
6.2.2 The Ensemble Average of a Random Local Medium is Nonlocal

As pointed out in Chapter 2, the formal solution for the average of a field
problem governed by a linear random (and local) operator on the domain B
is a deterministic nonlocal operator. This was illustrated in terms of a Fourier-
type heat conduction problem, a result that immediately carries over to an-
tiplane elasticity. Moving to a general setting of linear elastostatics on the
random field of a fourth-rank stiffness tensor Cijkl = {Cijkl (, x) ; , x R2 },
the equations governing the average fields in a Cauchy-type continuum are
(Beran and McCoy, 1970a)

ij (x) , j = 0

ij (x) = ijkl x, x kl x dx (6.90)
B

kl = uk,l + ul,k /2.
(Here we employ
ij and kl to denote symmetric stress and strain tensors.) In
(6.90)2 ijkl x, x is an infinite sum of integrodifferential operators, involving
moments of all orders of the random field Cijkl

ijkl x, x = Cijkl + Dijkl x x x + E ijkl x, x , (6.91)

where Dijkl x and E ijkl x, x are functions of the statistical properties of Cijkl
and the free-space Greens function of the nonstatistical problem. Addition
of a deterministic body force field fi does not change
the results. When the
fluctuations in Cijkl are small, Dijkl x and E ijkl x, x may be evaluated explic-
itly, and this was done by Beran and McCoy (1970) in the special case of the
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
realizations Cijkl () being locally isotropic, that is, expressed in terms of a vec-
tor random field of two Lame coefficients {[ (, x) , (, x)] ; , x R2 };
recall Section 2.3 of Chapter 2.
Next, considering this random field to be statistically homogeneous and
mean-ergodic, one may disregard the contributions of this operator for |xx |
> lc (the correlation length). Thus, since only the neighborhood within the
distance lc of x has a significant input into the integral (6.85)2 , one may expand
kl (x ) in a power series about x:

kl x = kl (x) + xm xm kl (x),m

xm xm xn xn
+ kl (x),mn . . . (6.92)
2
so as to obtain

ij (x) = ijkl x, x dx kl (x)
B

+ ijkl x, x xm xm dx kl (x),m + . (6.93)
B
This, in turn, can be rewritten as a sum of local, plus first gradient, plus higher
gradient strain effects:

ij = Cijkl kl + Dijklm kl ,m + E ijklmn kl ,mn + . (6.94)

Thus, B ijkl x, x dx in (6.93) is recognized as the effective stiffness Cijkl ; in-
deed the stiffness of a single realization B () of the random material B. If
one is given the ensemble B of B (), then one may determine the microstruc-

tural statistics, and hence the higher-order approximations Dijklm , E ijklmn , and
so on.
6.3 Planar Cosserat Elasticity

6.3.1 First Planar Problem
There are, in general, two planar problems of Cosserat elasticity (Nowacki,
1986):
1. The so-called first planar problem with u = (u1 , u2 , 0) and =
(0, 0, 3 ), which is a generalization of the classical planar elasticity.
2. The so-called second planar problem u = (0, 0, u3 ) and = (1 , 2 , 0),
which is a generalization of the classical antiplane elasticity.
Focusing on the first planar problem in static setting, from (6.12) and (6.7),
the equilibrium equations become
11,1 + 21,2 = 0 12,1 + 22,2 = 0 12 21 + 13,1 + 23,2 = 0, (6.95)

P1: Binod
May 28, 2007 18:6 C4174 C4174C006
while the kinematic relations (6.19) are
11 = u1,1 22 = u2,2 12 = u2,1 3 21 = u1,2 + 3

(6.96)
13 = 3,1 23 = 3,2 ,
and these satisfy the compatibility equations
21,1 11,2 = 13 22,1 12,2 = 23 23,1 = 13,2 . (6.97)
In the isotropic planar Cosserat medium, compliances of (6.27) become
(1) 1 (2)
Sijkl = (S + P)ik jl + (S P)il jk + ( A S)ij kl Si3k3 = ik M, (6.98)
4
where A, S, P, and M are four independent planar Cosserat constants, defined
in Ostoja-Starzewski and Jasiuk (1995):
1 1 1 1 1
A= = S= P= M= . (6.99)
+ +
Note that A and S define planar bulk and shear compliances of classical
elasticity (Dundurs and Markenscoff, 1993), while P and M are two additional
Cosserat constants; in the couple-stress elasticity P = 0. The restriction that
the strain energy be nonnegative implies the following inequalities:
0 A S 0 P 0 M. (6.100)
In the case of orthotropy for plane Cosserat elasticity, constitutive equa-

tions (6.27) become
(1) (1) (1) (1)
11 = S1111 11 + S1122 22 22 = S2211 11 + S2222 22
(1) (1) (1) (1)
12 = S1212 12 + S1221 21 21 = S2112 12 + S2121 21 (6.101)
(2) (2)
13 = S1313 13 23 = S2323 23 .
In the above, given (6.26), we have

(1) (1) (1) (1)
S1122 = S2211 S1221 = S2112 . (6.102)
Because for the couple-stress formulation 12 = 21 (recall equation 6.39),

we must have
(1) (1) (1) (1)
S1212 = S2112 S1221 = S2121 . (6.103)
This, combined with (6.102)2 above implies

(1) (1) (1) (1)
S1212 = S2112 = S1221 = S2121 , (6.104)
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
so that the constitutive relations (6.101) take on a simpler form

(1) (1) (1) (1)
11 = S1111 11 + S1122 22 22 = S1122 11 + S2222 22
(1)
12 = 21 = S1212 (12 + 21 ) (6.105)
(2) (2)
13 = S1313 13 23 = S2323 23 .
Finally, for the special type of orthotropy (symmetric, referred to in Section

6.5 below) we have two additional simplifications
(1) (1) (2) (2)
S1111 = S2222 S1313 = S2323 . (6.106)
Thus, the constitutive law for such an orthotropic and symmetric planar
(1)
couple-stress model involves four independent compliance components: S1111 ,
(1) (1) (2)
S1122 , S1212 , and S1313 .
6.3.2 Characteristic Lengths in Isotropic and Orthotropic Media

In the early 1960s when the Cosserat models began to undergo a revival fol-
lowing half a century of dormancy after invention by the Cosserats, several
people realized that, contrary to classical elasticity, an intrinsic length scale
was involved in the governing equations. It was denoted l, and called a char-
acteristic length. Let us now see how this l can be arrived at. Following Mindlin
(1963) and Schafer (1962), we employ a stress function formulation, which for
the planar Cosserat (as well as the couple-stress) elasticity involves two stress
functions, and
11 = ,22 ,12 22 = ,11 + ,12

12 = ,12 ,22 21 = ,12 + ,11 (6.107)
13 = ,1 23 = ,2 .
Note that is the Airy stress function known from the classical elastostatics.
Recall also that, for the isotropic planar Cosserat elasticity, the compatibility
conditions in terms of and are given by (e.g., Nowacki, 1986)

P+S 2 A+ S 2 P+S 2 A+ S 2
= ,2 = ,1 .
4M ,1 4M 4M ,2 4M
(6.108)
These are the Cauchy-Riemann conditions for the functions A+S4M

2 and
[ 4M ], so that we actually have two harmonic functions
P+S 2

P+S 2
2 2 = 0 2 = 0. (6.109)
4M
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
The coefficient ( P + S)/4M appearing above has the dimension of length

squared and has thus led to a definition of characteristic length l via
P+S S(1)
l2 = 1212
(2)
. (6.110)
4M S1313
Note: In the couple-stress theory P = 0 in equations (6.1076.109).

For the orthotropic Cosserat elasticity case, the compatibility conditions
(6.97) result in
(1) (1) (1) (1)
[S1111 S1221 S1122 ],122 + S2121 ,111
(1) (1) (1) (1) (2)
[S1122 + S2121 + S1221 ],112 S1111 ,222 = S1313 ,1
(1) (1) (1) (1)

[S2222 S1221 S1122 ],112 + S1212 ,222
(1) (1) (1) (1) (2)
+[S1122 + S1212 + S1221 ],122 + S2222 ,111 = S2323 ,2 , (6.111)
which suggest the following definitions of four characteristic lengths (Bouyge

et al., 2002):

(1) (1)
S S(1) S(1) S
l1 = 1111 1221
(2)
1122
l2 = 2121
(2)
S1313 S1313
(6.112)
(1) (1)
S (1)
S1221 (1)
S1122 S
l3 = 2222 (2)
l4 = 1212
(2)
.
S2323 S2323
In the special case of plane isotropic Cosserat elasticity, the following

relations hold
(1) (1) (1) (1) (1) (1) (1) (1)
S1111 S1221 S1122 = S2121 = S2222 S1221 S1122 = S1212 , (6.113)
and equations (6.112) reduce to a single characteristic length defined in (6.110).

For the planar orthotropic couple-stress case, the compatibility equations
(6.113) yield
(1) (1) (1) (1)

[S1111 S1212 S1122 ],122 + S1212 ,111
(1) (1) (1) (2)
[S1122 + 2S1212 ],112 S1111 ,222 = S1313 ,1
(1) (1) (1) (1)

[S2222 S1212 S1122 ],112 + S1212 ,222
(1) (1) (1) (2)
+ [S1122 + 2S1212 ],122 + S2222 ,111 = S2323 ,2 . (6.114)
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
Thus, the four characteristic lengths are

(1) (1)
S S(1) S(1) S
l1 = 1111 1212
(2)
1122
l2 = 1212
(2)
S1313 S1313

(1) (1)
S (1)
S1212 (1)
S1122 S
l3 = 2222 (2)
l4 = 1212
(2)
, (6.115)
S2323 S2323
whereupon equations (6.114) can be rewritten as
(1) (1) (1)

S1122 + 2S1212 S1111
l12 ,122 + l22 ,111 (2)
,112 (2)
,222 = ,1
S1313 S1313
(1) (1) (1)
S1122 + 2S1212 S2222
l32 ,112 + l42 ,222 + (2)
,122 + (2)
,111 = ,2 . (6.116)
S2323 S2323
For the special case of plane orthotropic couple-stress case with symmetry,
in view of (6.106), there are only two characteristic lengths

(1) (1)
S S(1) S(1) S
l1 = 1111 1212
(2)
1122
l2 = 1212
(2)
. (6.117)
S1313 S1313
6.3.3 Restricted Continuum vis-a-vis the Micropolar Model

It is now recalled that a restricted continuum (couple-stress) Cosserat model,
with the limitation (6.38), was introduced in the past as a simplified, and
somewhat restricted, version of the general micropolar case. First, let us note
that, with the definition (6.117), the counterpart of equations (6.108) is
[ l 2 2 ],1 = 2 (1 ) l 2 2 ,2 [ l 2 2 ],2 = 2 (1 ) l 2 2 ,1 ,
(6.118)
and (6.109) holds with ( P + S) /4M replaced by S/4M. The length l appearing
in (6.118) is
S 2 (1 + ) B
l2 = . (6.119)
4M E
where B is the modulus of curvature and , E, and are the shear modulus,
Youngs modulus, and Poissons ratio of classical elasticity, respectively.
To illustrate the distinction between both models we focus now on the
problem of a hole in an infinite body in plane strain under uniaxial tension p.
Mindlin (1963) found that in a couple-stress material the maximum stress m
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
(or stress concentration factor)
3+ F
m = p (6.120)
1+ F
where, in our notation,
4 (S + A) /S
F = (6.121)
4 + (a /l) + 2 (a /l) K 0 (a /l) /K 1 (a /l)
2
where l is defined by (6.119) and a is the hole radius; also, K 0 and K 1 are
the modified Bessel functions of the second kind of orders zero and one,
respectively.
On the other hand, somewhat later Kaloni and Ariman (1967) solved the
same problem for a micropolar elastic body with the result
4 (S + A) / (S + P)
F = (6.122)
4 + (a /l) 2 + 2 (a /l) K 0 (a /l) /K 1 (a /l)
where l is defined by (6.110).

Finally, we would like to point out that if we set
1
M= P = 0, (6.123)
4B
in (6.121) and other pertinent micropolar formulas, we recover Mindlins
couple-stress result. Also, when the details of the microstructure become
much smaller than the hole radius, i.e. l 0, then F 0, and m 3 p,
which recovers the classical elasticity result.
To see a continuous transition from the classical elasticity to both Cosserat
models, it is convenient at this stage to bring in, after Cowin (1969, 1970a, b),
a nondimensional constant

S
N= = 0 N 1. (6.124)
+ S+ P
Then, following Cowin, F is given by
8 (1 ) N2
F = (6.125)
4 + ( NL) 2 + 2NL KK 01 (( NL)
NL)
whereby, in our notation,

A+ S a 4M
1 = L= =a . (6.126)
2S l S
Indeed, for this example, the case N = 0 leads to the classical elasticity solu-
tion, while N = 1 gives the limiting case of couple-stress theory (6.121). The
micropolar unrestricted theory follows for any value of N between 0 and 1.
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
TABLE 6.1
A Comparison of Various Notations for Micropolar Compliances in the First
Planar Problem
Compliance Our Notation Nowacki Eringen Mindlin
1 1 2(1+)
Shear compliance 1/G S E + E /2 E
1 1 12
Plane strain bulk compliance A + E + E + E /2 G
1 2
Bulk compliance P E 0
1 1 1
Bending or curvature compliance M + E 4B
P+S (+)( +) E ( E + E ) B
Characteristic length (square of) l 2 4M 4 E (2 E + E ) G
Note: Mindlins column refers to the restricted model.
However, in general, the limit N = 0 needs to be used with caution (Lakes,

1985) because the zero value of N does not automatically imply that all the
micropolar stiffnesses vanish (i.e., the compliances go to infinity), and, in fact,
the microstructural degrees of freedom may remain.
Finally, for the sake of reference we provide in Table 6.1 a comparison
between several different notations employed in the works referenced here.
A similar table linking the notation of other references is included in Cowin
(1970a).
6.4 The CLM Result and Stress-Invariance

6.4.1 Isotropic Materials
We now allow the micropolar solid to be inhomogeneous by taking all the
material coefficients in (6.99) to be class C 2 functions of x, and assume it is
simply connected (i.e., no holes are present). Now, note that the compatibility
equations (6.97) of Section 6.3 can be written as
22,11 + 11,22 = (12 + 21 ) ,12

12,22 21,11 = (22 11 ) 12 13,1 23,2 (6.127)
23,1 = 13,2 .
Substituting (6.98) of Section 6.3 into the compatibility condition (6.127)1 ,

and using (6.95)1 and (6.95)2 of Section 6.3, we obtain, after rather lengthy
manipulations,
A+ S
2[ (11 + 22 )] [S,1 11 ],1 [S,2 22 ],2 [S,1 12 ],2 [S,2 21 ],1 = 0.
2
(6.128)
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
Proceeding similarly with respect to the compatibility condition (6.127)2 , we

obtain, again after much algebra,
P+S
2[ (12 21 )] + [S,1 11 ],2 [S,2 22 ],1 [S,1 12 ],1 [S,2 21 ],2
2
+ 2 [M13 ],1 + 2 [M23 ],2 = 0. (6.129)
The third compatibility condition (6.127)3 yields
[M23 ],1 [M13 ],2 = 0. (6.130)
Let us note here that (6.128) simplifies to equation (5.36) of Chapter 5 in

the case of classical elasticity where 12 = 21 . As before, we ask the question:
Supposing that A, P, S, and M are changed to some A, P,
S and M, then
under what restrictions would the original stress field 11 , 22 , 12 , 21 , 13 , 23
remain unchanged? An examination of (6.95) implies that we must have
+
A S = m ( A + S)
S,1 = mS,1
S,2 = mS,2
(6.131)

S,11 = mS,11
S,22 = mS,22
S,12 = mS,12 ,
where m is an arbitrary scalar. Next, note that (6.131) leads to
=m A + b
A
S = mS b, (6.132)
where b is an arbitrary constant restricted by the requirement that the com-

pliances be non-negative.
By a similar reasoning, the compatibility equation (6.127)2 implies
= nP + c
P
S = nS c = nM,
M (6.133)
where c is an arbitrary constant restricted by the requirement that the com-

pliances be non-negative.
Finally, the compatibility condition (6.127)3 involves a micropolar compli-
ance M and its first derivatives, and so we have the following conditions:
= nM
M ,1 = nM,1
M ,2 = nM,2 .
M (6.134)
Considering all the above results, it is seen that they are consistent pro-
viding m = n and b = c. It thus follows that the stress will be invariant if the
following shifts in material compliances are taken:
=m A + c
A = mP + c
P
S = mS c = nM.
M (6.135)
In the terminology of the CLM result (Chapter 5), both materials are equivalent.
Clearly, equations (6.135) represent a constant shift in three out of four material
parameters, and this is a weaker shift than the linear one: (5.39) in Chapter 5.
Note: The second planar problem does not admit a shift.
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
6.4.2 Anisotropic Materials and the Null-Lagrangian

In the case of an anisotropic material the starting point is provided by
relations (6.27). A substitution into the compatibility relations (6.127) and
a subsequent inspection leads us to conclude that the stress field remains
(1) (2)
unchanged when the material constants are modified from Sijkl and Sijkl to
(1) (1) (1) I (2) (2) (1) I
S = S + S (, , ) and S = S , providing S (, , ) is the
ijkl ijkl ijkl ijkl ijkl ijkl
(2)
shift defined by (6.135) for isotropic materials, and Sijkl undergoes no shift.
(1)
Let us first note that the compliance Sijkl
of an isotropic material is a fol-
lowing function of three constants , , and :

(1) I 1 1 1 1 1 1 1
Sijkl (, , ) = + ik jl + il jk + ij kl .
4
(6.136)
Equivalently, noting (6.99), this can also be given in terms of A, P, S as
(1) I 1
Sijkl ( A, P, S) = (S + P) ik jl + (S P) il jk + ( A S) ij kl . (6.137)
4
If we let = , = , and = in (6.136), it follows that the shift tensor

is given as
(1) I 1
Sijkl (, , ) = ij kl il jk . (6.138)
2
Let us now compare this with the CLM shift tensor of classical elasticity
(5.49) with the rotation (5.50) of Chapter 5. This leads to a question: What is
(1) I
the meaning of Sijkl (, , )? The answer is obtained from a consideration
of a new rotation tensor defined as
(1) I 1 (1)
R(1)
ijkl = ij kl il jk Sijkl (, , ) = R , (6.139)
2 ijkl
showing that R(1)ijkl kl is also a right-angle rotation of kl .

The foregoing development allows us to present the shift-result in terms
of the null-Lagrangian. First, let us recall from the theory of micropolar elas-
ticity (Nowacki, 1986) that two stress functions and can be introduced
such that [recall (6.107)]
ij = R(1)
ijkl ,kl Rik ,k j i3 = ,i , (6.140)
where Rik was specified in (5.54) of Chapter 5. The strain energy density
(1) (2) (1) (2)

W(Sijkl , Sijkl ) = ij Sijkl kl + i3 Si3k3 k3 (6.141)
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
may now be written as

(1) (2) (1) (1)
W Sijkl , Sijkl = R(1)
ijkl ,kl Rik ,k j Sijmn Rmnpq , pq Rmp , pm
(2)
+ i3 Si3k3 k3 . (6.142)
First, let us note that the minimization of (6.142) via the EulerLagrange
equations of (6.141) results in the compatibility equations in terms of and
for a general anisotropy. On the other hand, the energy density of the shift
in compliance becomes

(1) I
W Sijkl (, , ) , Oijkl

(1) I (1)
= R(1) ijkl ,kl Rik ,k j Sijmn Rmnpq , pq Rmp , pm , (6.143)
where Oijkl is a null tensor. Now, observing that
R(1) (1) (1)

klij Rijmn Rmnpq = Rklpq
(1)
R(1) (1)
ijkl Rijmn = km ln , (6.144)
we find
1 2
(1) I
W Sijkl (, , ) , Oijkl = ,11 ,22 ,12
2

+ ,12 ,22 ,11 + ,12 ,22 ,11 . (6.145)
It is interesting to note here that:
1. The first term in the square brackets is the same as that in the classical
elasticity [our (5.59) in Chapter 5, or equation (35) of Cherkaev
et al. (1992)].
2. The second and third terms represent the coupled contribution of
and potentials.
3. The energy (6.145) can also be written as the divergence of a vector
field v,k such that
1
vk = ,l R(1)
klpq , pq ,l Rkr ,rl ,i j Rik , j
2

+ , j R jk Rnr Rnk Rjr ,rn

1 ,1 ,22 ,2 ,12 ,1 ,12 ,2 ,22 + ,21 ,1 + ,22 ,2
= ,
2 ,2 ,11 ,1 ,12 + ,1 ,11 + ,2 ,12 ,11 ,1 ,12 ,2
(6.146)
where, again, the first term in each of the square brackets can be
recognized as that of classical elasticity. It follows now that vk,k = 0,
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
(1) I
or that the EulerLagrange equations for W(Sijkl (, , ), Oijkl )
are satisfied identically, implying that it is a null-Lagrangian.
6.4.3 Multiply Connected Materials

The requirement that B be simply connected ensures that the stress field is
single valued. Generalizing the results of Section 5.2.3 of Chapter 5, we have

3 3
3 = d x1 + d x2
x1 x2

21 11 22 12
= d x1 + d x2 , (6.147)
x1 x2 x1 x2

11 11
D1 + y0 3 = x1 d x1 + d x2
x1 x2

11 22 12 21
x2 d x1 d x2 , (6.148)
x2 x1 x2 x2

22 22
D2 x0 3 = x2 d x1 + d x2
x1 x2

22 12 21 11
x1 d x2 + + d x1 , (6.149)
x1 x1 x1 x2
where D1 and D2 are dislocation vectors and 3 is a disclination. (Nowacki,

1986; Takeuti, 1973). If, using (6.98), we express the above equations in terms
of stresses, they take on the following forms:

S S
3 = [( A + S) (11 + 22 )] ds = 2 t1 ds 2 t2 ds
n x1 x2

+ (S + P) (21 12 ) ds + (12 21 ) (S P) ds, (6.150)
s s

4 ( D1 + y0 3 ) = x2 x1 ( A + S) (11 + 22 ) ds
n s

S S
=2 St2 ds 2 x2 t1 + t2 ds
x1 x2

S
2 x2 (21 12 ) ds, (6.151)
s
P1: Binod
May 28, 2007 18:6 C4174 C4174C006

4 ( D2 x0 3 ) = x1 + x2 ( A + S) (11 + 22 ) ds
n s

S S
=2 St1 ds + 2 x1 t1 + t2 ds
x1 x2

S
+2 x1 (21 12 ) ds, (6.152)
s
where n and s denote the outer unit normal and arc length of the hole
boundary.
With reference to Section 6.3.3, for the problem of an infinite plate with
hole, we have these conclusions:
1. For classical elasticity, m as well as the entire stress field are inde-
pendent of elastic constants, say A and S (or and ).
2. For a pseudo-continuum which has three constants A, S, and M
(or , and B)the stress field depends on two combinations of
these constants, such as ( A + S) /S (or ) and l 2 = S/4M, and thus
no shift is possible here.
3. For an unrestricted continuum, which has four constants, A, S, P,
and M, the dependence is on two independent combinations of the
elastic constants ( A + S) / ( P + S) and l 2 = (S + P) /4M, which, in
light of (6.134), allow a shift.
Note: Setting P = 0, we get the pseudo-continuum. In this case S + P
becomes S, and S by itself is not invariant under shift.
6.4.4 Applications to Composites

6.4.4.1 Two-Phase Materials
We continue to generalize the results of Chapter 5. Thus, when the planar
body is made up of two or more phases, we must also consider the inter-
face boundary conditions. Assuming perfect bounding between micropolar
phases (1 and 2), they have the following (classical) form in the curvilinear
coordinate system (n, s, x3 ):
nn
(1)
= nn
(2)
ns
(1)
= ns
(2)
(1) (2)
n3 = n3
(6.153)
n = un
u(1) (2)
s = us
u(1) (2)
3(1) = 3(2) .
Alternately, using the boundary conditions proposed by Dundurs (1996),

we have
nn
(1)
= nn
(2)
ns
(1)
= ns
(2)
(1) (2)
n3 = n3
3
(1)
3
(2) (6.154)
n(1) = n(2) ss(1) = ss(2) = ,
s s
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
where n is the change in curvature of boundary curve

ss
n = (ns + sn ) nn , (6.155)
s n
and ss is the stretch strain. The advantage of using this second set of boundary
conditions is that they can be expressed in terns of stresses, and thus the
dependence of the solution on the micropolar constants can be seen more
easily. Using the constitutive relations (6.98), which remain of the same form
in (n, s, x3 ), and the equilibrium equation
nn 1
+ (ns + sn ) + (ss nn ) = 0, (6.156)
n 2 s
the boundary condition (6.154)4 , which implies the continuity of a change in
curvature, becomes, in view of (6.155) and again (6.98),
(1) (2)
( A2 + S2 ) ss(2) ( A1 + S1 ) ss(1) + sn ( A1 + S1 ) sn ( A2 + S2 )
n n s s
S1 S2 ns
(1)
+ 2sn
(1)
2sn
(2)
+ 2ns
(1)
(S1 S2 ) 2 (S1 S2 )
s s s s

+ nn
(1)
[( A2 A1 ) (S2 S1 ) + 2 ( A2 A1 ) ] = 0. (6.157)
n
Now, taking note of (6.99), the continuity of stretch strain (6.154)5 implies
( A2 + S2 ) ss(2) ( A1 + S1 ) ss(1) + nn
(1)
( A2 A1 ) (S2 S1 ) = 0. (6.158)
Finally, noting 3 /s = s3 , we observe that (6.154)6 implies
M1 (1) (2)
s3 M2 s3 = 0. (6.159)
Note that these boundary conditions are invariant under the shift (6.135).
Thus, if the multiphase material is simply connected (i.e., contains intrusions),
the governing equations in terms of stresses are (6.95) and (6.1286.130) for
each phase, and these are invariant under traction loading and boundary
conditions (6.154), or, equivalently, (6.154)13 and (6.1576.159). However, if
the material is multiply connected, we also need Cesaro integrals that involve
the continuity of displacements.
6.4.4.2 Effective Moduli of Composites

We now move to the determination of effective properties of composites,
which in the case of classical elasticity turned out to nicely involve the CLM
result. We thus examine the shift in macroscopically effective compliance
(1)eff (2)eff
tensors Sijkl and Sijkl connecting the volume averaged stress and strain
tensors
(1) (2)
ij = Sijkl kl ij = Sijkl kl , (6.160)
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
wherein denotes a volume average. We proceed here in a fashion analogous to

that of Cherkaev et al. (1992) with a restriction to materials with continuously
varying (class C 2 ) properties.
Now, stresses 11 , 22 , 12 , 21 , 13 , 23 are the same in two equivalent
(1)
materials (Sijkl (2)
(x), Sijkl (x)) and ( (1)
Sijkl (x), (2)
Sijkl (x)), so that the strain fields ij and
ij satisfy the relation
ij = (1)
Sijkl (1)
kl = Sijkl (1) I
kl + Sijkl (1) I
(, , ) kl = ij + Sijkl (, , ) kl .
(6.161)
(1) I
Volume averaging (6.161), and noting that Sijkl (, , ) is independent of
position, we find
ij =
(1)eff (1)eff (1) I
Si jkl kl = Si jkl kl + Sijkl (, , ) kl . (6.162)
which shows that the effective compliance tensor of the second material (with
hat) is given by that of the first material (without hat) plus the shift given by
(6.138)
(1)eff (1)eff (1) I

Sijkl = Sijkl + Sijkl (, , ) . (6.163)
We conclude (by inspection) that there is no shift in the second effective com-
pliance tensor and
(2)eff (2)eff
Sijkl = Sijkl . (6.164)
As mentioned at the beginning of this section, this conclusion holds for simply
connected inhomogeneous media with twice-differentiable properties.
6.4.5 Extensions of Stress Invariance to Presence

of Eigenstrains and Eigencurvatures
Here we extend the results of Section 5.2.3, Chapter 5. First note that one
can generalize the elasticity with eigenstrains to Cosserat elasticity. Thus, in
analogy to (5.74), the total strain ij is the sum of the elastic strain gij and the
eigenstrain ij ,
ij = gij + ij (6.165)
while the total (generally nonsymmetric) curvature ij is the sum of the elastic
curvature kij and the eigencurvature ij
ij = kij + ij (6.166)
The eigenstrain ij and eigencurvature ij are inelastic strains and curva-

tures, respectively. The total strain ij and total curvature ij must satisfy
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
compatibility equations
li,n ni,l kni lk + kli nk = 0 li,n = ni,l . (6.167)
From (6.165 and 6.166) we have

(1) (2)
(1)
ij = Cijkl gkl = Cijkl kl kl (2)
ij = Cijkl kkl = Cijkl kl kl . (6.168)
The inverses of (6.168) are

(1) (2)
gij = Sijkl kl kij = Sijkl kl . (6.169)
Note that all the quantities may depend on the spatial position x ( xi ).
We extend the assumption that the material is free from any external forces
and surface constraints to Cosserat elasticity with eigenstrains and eigencur-
vatures. If these conditions of free surface are not satisfied, the force-stress
and couple-stress fields can be obtained by a superposition of the force-stress
and couple-stress of a free body and the stress obtained from the solution of
a given boundary value problem with non zero external forces or boundary
conditions.
The force-stresses and couple-stresses must satisfy the equations of equi-
librium [assume no body and inertia forces in (6.12) and (6.17)] in B
ji,j = 0 ijk jk + ji,j = 0 i, j = 1, 2, 3, (6.170)
and zero force-traction or couple-traction free boundary conditions on B
ji n j = 0 ji n j = 0. (6.171)
In analogy to Muras approach in classical elasticity, by substituting (6.168)

into (6.170) and assuming a homogeneous material, we have
(1) (1) (1)

Cjikl kl, j = Cjikl kl, j C (2) (2)
jimn mn, j = C jimn mn, j ijk C jkmn mn mn ,
(6.172)
and by substituting (6.168) into (6.171) we obtain

(1) (1) (2) (2)
Cjikl kl n j = Cjikl kl n j Cjikl kl n j = Cjikl kl n j . (6.173)
Note that in the absence of eigenstrains ( = 0), the left-hand side of

(6.172)1 corresponds to ji,j , and the left-hand side of (6.173)1 to ji n j . Thus,

(6.172)1 , is in the form ji,j = Xi where Xi = Cjikl kl, j and (6.173)1 is in the

form ji n j = ti where ti = Cjikl kl n j . Therefore, the contribution of eigenstrain
to the equations of equilibrium (6.170) is mathematically equivalent to a
body force, and in the boundary conditions (6.171)1 is similar to a surface
force.
Next, in the absence of eigencurvatures ( = 0) the left-hand side of
(6.172)2 corresponds to ijk jk + ji,j , whereas the left-hand side of (6.173)2 to
ji n j . Thus, (6.172)2 is in the form ijk jk + ji,j = Yi where the couple-body
P1: Binod
May 28, 2007 18:6 C4174 C4174C006

force Yi = Cijkl kl, j and (6.173)2 is in the form ji n j = ti where ti = C jikl kl n j .

Therefore, the contribution of eigenstrain ij to the equations of equilibrium
(6.170)2 is mathematically equivalent to a body force, while their contribution
to the boundary conditions (6.171)2 is similar to a surface force.
In the next sections we focus on the planar elasticity with eigenstrains,
assuming isotropy in elastic properties. In addition, we relax the bound-
ary condition (6.171) and admit non-zero tractions to make the formulation
more general. This will not change our conclusions on the reduced parameter
dependence.
Note that the special case of uncoupled micropolar thermoelasticity with
eigenstrains ij is defined as
ij = ij T ij = 0 if i=
j i, j = 1, 2, 3, (6.174)
where ij is a thermal expansion coefficient and T is temperature change.

We will refer to this special case in examples.
We demonstrated the reduced parameter dependence in the in-plane stress
fields in the problems governed by plane Cosserat elasticity with eigenstrains
and eigencurvatures, if eigenstrains and curvatures satisfy certain conditions.
Note that no conditions are needed for the plane stress case for the form
of shift with m = 1. These results can be applied for two-phase materials
to linear planar uncoupled micropolar thermoelasticity, where eigenstrains
are uniform and represent the product of the thermal expansion coefficient
and temperature change. The analysis can also be extended to multiphase
materials and inhomogeneous multiply connected materials.
6.4.5.2 Inhomogeneous Materials

Here we focus on 2D boundary value problems involving applied force- and
couple-tractions
ti = ji n j mi = ji n j on B. (6.175)
It is also useful to write the constitutive laws (recall Section 6.3.1)
A+ S S
11 = (11 + 22 ) 22 + 11 + 33
4 2
A+ S S
22 = (11 + 22 ) 11 + 22 + 33
4 2
S P
12 = (12 + 21 ) + (12 21 ) + 12 (6.176)
4 4
S P
21 = (12 + 21 ) (12 21 ) + 21
4 4

13 = M13 + 13 23 = M23 + 23 ,
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
where
1 1 1
S= P= M= , (6.177)
+
and for plane strain
1 1
A= = , = , (6.178)
plane-strain + 3K
while for plane stress
1 (3 + 2)
A= = , = 0. (6.179)
plane-stress + 2
If we express in terms of Poissons ratio v
2v
= , (6.180)
1 2v
then, for plane strain

1 1 2v 1 A
A= = , =v= 1 , (6.181)
plane-strain 2 S
while, for plane stress
1 1v
A= = , = 0. (6.182)
plane-stress (1 + v)
Now, with the assumption of both compliances and eigenstrains being
smooth functions of position, it follows from the substitution of (6.176) into
(6.127)1 , and in light of (6.171)1 , that the first compatibility condition in
(6.127) is
1 2 1 1 1
[( A + S) (11 + 22 )] [S,1 11 ],1 [S,2 22 ],2 [S,1 12 ],2
4 2 2 2
1

[S,2 21 ],1 = 11,22 22,11 + 12 + 21 ,12
2 33
2

2,1 33,1 2,2 33,2 2 33 . (6.183)
Similarly, the compatibility condition (6.127)2 yields

1 2 1 1 1
[( P + S)] + (S,1 11 ) ,2 (S,2 22 ) ,1 (S,1 12 ) ,1
4 2 2 2
1
+ (S,2 21 ) ,2 = 11,12 + 22,12 12,22 + 21,11
2

( M13 ) ,1 13,1 ( M23 ) ,2 223,2 . (6.184)
Finally, from the compatibility condition (6.127)3 we have

( M23 ) ,1 + 23,1 = ( M13 ) ,2 + 13,2 . (6.185)
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
Next, following Ostoja-Starzewski and Jasiuk (1995), we seek the condi-

tions for invariance in planar force-stresses and couple-stresses with respect
to shift in compliances. We first state the conditions for no eigenstrains and
eigencurvatures. Equations (6.1836.185) remain invariant if the compliances
A, P and S undergo a shift
=m A + c
A = nP + c
P
S = mS c = mM.
M (6.186)
Under such a linear shift the force-stress does not change in the absence of
eigenstrains and eigencurvatures. Next, we investigate what conditions are
needed to be satisfied in the presence of eigenstrains and eigencurvatures. In
this analysis, in addition to the plane stress and plane strain cases, which lead
to different results, the distinction is made between the cases when m = 1 and
m= 1.
For the plane stress case and m = 1, (6.1836.185) remain unchanged under
the linear shift (6.186), i.e., the planar stress components remain unchanged,
and thus there is a reduced parameter dependence.
For the plane stress case and m = 1, (6.1836.185) give the following
conditions on eigenstrains and eigencurvatures:

11,22 + 22,11 12 + 21 ,12
=0

11,12 22,12 + 12,22 + 21,11 + 13,1 + 23,2 =0 (6.187)

23,2 = 13,1 .
For the plane strain case and m = 1, (6.183) remains invariant under the
shift (6.186) when

2 33 + 2,1 33,1 + 233,2 + 2 33 = 0. (6.188)
For the special case of uniform eigenstrains, the condition above is satisfied
provided that

33 =0 or 2 = 0, (6.189)
while for the case of a homogeneous material the condition (6.188) is

satisfied if

2 33 =0 or = 0. (6.190)
For the case of plane strain and m =

1, the following conditions need to be
satisfied

11,22 + 22,11 (12 + 21 ) ,12 + 2 33 + 2,1 33,1 2,2 33,2 + 2 33 =0

(11 22 ) ,12 + 12,22 + 21,11 + 13,1 + 23,2 =0 (6.191)

23,1 = 13,2 .
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
6.5 Effective Micropolar Moduli and Characteristic Lengths

of Composites
6.5.1 From a Heterogeneous Cauchy to a Homogeneous
Cosserat Continuum
Any of the lattices considered in Chapter 3 can be viewed as a planar, two-
phase composite material: one phase (1) is the solid that makes up the beams
and another (2), quite trivially, is the vacuum in the pores. Clearly then, the
mismatch of elastic properties, that is the ratio of moduli E (1) /E (2) , is in-
finite (Ostoja-Starzewski et al., 1999). Starting from this consideration, one
may now consider nontrivial two-phase composites made of two kinds of
solids having a finite mismatch, and generalizing the previously established
method (i.e., that for lattices) for passage from heterogeneous Cauchy to a
homogeneous Cosserat continuum; Fig. 6.3. This passage is done according
to the following equality:

1 V 0 (1) 0 0 (2) 0

ij Cijkl kl dV = C + i3 Ci3k3 k3 , i, j, k, l = 1, 2, (6.192)
2 V 2 ij ijkl kl
where the left-hand side is the total elastic strain energy stored in the unit
cell of the matrix-inclusion composite (a function of Cauchy strain fields ij ),
while the right-hand side is the energy of a Cosserat continuum (a function of
volume-average strains ij0 and curvatures i3 0
of the unit cell). V is the volume
of the unit cell B L . Cijkl is the elastic stiffness of the composites constituents,
(1) (2)
while Cijkl and Ci3k3 are the sought (effective) micropolar stiffnesses.
The key issue concerns the choice of loading on the periodic unit cell. Fol-
lowing Forest (1989, 1999) and Forest and Sab (1998), the appropriate periodic
boundary conditions in terms of displacements are generally nonlinear:
ui (x) = Ai + Bij x j + Cijk x j xk + Dijkl x j xk xl (6.193)
(a) (b) (c)
FIGURE 6.3
(a) A periodic, globally orthotropic, matrix-inclusion composite, of period L, with inclusions of
diameter d arranged in a square array; (b) a periodic unit cell with soft inclusions at corners;
(c) a periodic unit cell with a stiff inclusion at the center.
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
where vi takes an equal value on corresponding points of the boundary. The

field (6.193) satisfies the Hill condition generalized to the Cosserat material
2
ij ij = ij ij + ij ij with ij = ij0 ij = e ikl xk l j . (6.194)
3
In the notation and language of homogenization theory (e.g., Bensoussan
et al., 1978), the Cauchy material in the unit cell is described by a fast co-
ordinate x, while the Cosserat material is described by the slow coordinate y.
Accordingly, the displacement, stress and strain fields in the Cauchy conti-
nuum are
u , (6.195)
while the fields of displacement, rotation, force-stress, couple-stress, strain,

and rotation in the Cosserat continuum are
U M E K. (6.196)
We now rewrite (6.194) as

2
ij ij = ij E ij + Mij K ij with ij = ij0 Mij = e ikl xk l j . (6.197)
3
In the general setting, we have a micromorphic continuum with U (dis-
placement) and (microdeformation) fields, and these are determined from
a minimization problem

(U, ) = min |u (x) U (x X) |2 (6.198)
The solution is
U (X) = u (x) (x) = u (x X) A1 , (6.199)
with A = (x X) (x X) and respective gradients being
U X = u x x = (u x) x A1 U AT . (6.200)
Although none of our boundary conditions were of periodic type, the situ-
ation changes when an unrestricted model is used. Indeed, such a derivation
has been done in Forest and Sab (1998) by extending the homogenization
method (e.g., Sanchez-Palencia and Zaoui, 1987). The loading on B L in 2D is
then effected by boundary conditions involving polynomials of the general
form

u1 (x) = B11 x1 + B12 x2 C23 x22 + 2C13 x1 x2 + D12 x23 3x12 x2

u2 (x) = B12 x1 + B22 x2 C13 x12 + 2C23 x1 x2 D12 x13 3x1 x22 . (6.201)
Upon a comparison of this with equations (6.1936.196), we observe that the

derivation of the restricted model involves a second-order polynomial, while
that of the unrestricted one requires a third-order polynomial. Work has also
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
been done on homogenization of a heterogeneous Cosserat-type continuum

by a homogeneous one (Forest, 1999; Forest et al., 1999); see also Forest et al.
(2000, 2001).
Most recently, a related homogenization procedure was outlined by Onck
(2002) for the derivation of the micropolar model. In particular, a loading
via a skew-symmetric part of the strain tensor, applied in terms of bound-
ary rotations, was proposed to grasp the effect of the difference between the
micro-rotation i and the macrorotation e ijk uk, j /2; recall the restricted models
condition (6.37), where that difference is set to zero.
Note: A major outstanding challenge is to develop boundary conditions for the
homogeneous Cosserat from a heterogeneous Cauchy continuum. This should
be done through a scheme somewhat analogous to that developed so far for
the effective micropolar moduli.
6.5.2 Applications
A few years ago we computed effective micropolar moduli for planar matrix-
inclusion composites arranged in periodic arrays: triangular (Bouyge et al.,
2001) and square (Bouyge et al., 2002), Figure 6.4, using a finite element
method. Several different boundary conditionsranging from displacement-
type to traction-type, and various combinations thereofwere used. For
(1)
example, using displacement boundary conditions, we determine Cijkl from
three tests:
1. Uniaxial extension:
u1 (x) = 0 u2 (x) = 22 x2 x B (6.202)

(1)
gives C2222 (= 2U cell /V when we set 22 = 1). For our composite,
(1) (1)
C1111 = C2222 due to the symmetry of the square arrangement.
2. Biaxial extension:
u1 (x) = 11 x1 u2 (x) = 22 x2 x B (6.203)

(1) (1)
yields 2C1111 + C1122 (= 2U cell /V when we set 11 = 22 = 1).
3. Shear test:
u1 (x) = 12 x2 u2 (x) = 0 x B (6.204)

(1)
yields C1212 (= 2U cell /V when we set 12 = 1).
(2)
4. Finally, to determine Ci3k3 , we conduct the fourth test, the bending
test:
x12
u1 (x) = 13 x1 x2 u2 (x) = 13 x B (6.205)
2
(2) (2) (2)
gives C1313 . Note that in our study C1313 = C2323 due to the symmetry
of square arrangement.
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
u1 (x) = 0, u2 (x) = y22x2
u1 (x) = y11x1, u2 (x) = y22x2
u1 (x) = y12 x2, u2 (x) = 0
x12
u1 (x) = x1x213, u2 (x) =
2 13
FIGURE 6.4
(1) (1) (1) (2)
Tests for the determination of constants C2222 , C1122 , C1212 , and C1313 of a periodic composite with
circular inclusions in a square arrangement under displacement boundary conditions (Bouyge
et al., 2002). Left (right) column corresponds to the inclusion at the corner (center). Inclusions
can be seen from the mesh pattern.
The resulting deformation modes for the above four tests under displace-
ment boundary conditions are shown in Figure 6.4. Two distinct situations are
considered here depending on whether the inclusion is softer or stiffer than
the matrix. In the first case, the inclusion is located at the corner, whereas in
the second it is located at the center. Typical results for effective moduli are
(1)
shown in Figure 6.5 in terms of C1212 for a wide range of Poissons ratio of the
matrix. In the special case of no mismatch in the properties we recover a ho-
mogeneous medium of Cauchy type, whereby the composite microstructure
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
0.6
0.5
0.4
dd
C1212
0.3 dp
tt
0.2
0.1
0
0.0001 0.001 0.01 0.1 10 100 1000 10000
Mismatch
FIGURE 6.5
(1)
The effective moduli C1212 , normalized by E m , from three types of boundary conditions
displacement (dd) displacement-periodic (dp), and traction (tt)plotted as functions of the stiff-
ness ratio E i /E m for the case of Poissons ratios m = i = 0.3, at inclusions volume fraction of
18.4%. (From Bouyge et al., 2002. With permisssion.)
disappears and no Cosserat continuum is to be set up. Note that when the
inclusion is softer, as well as stiffer, than the matrix, the micropolar model
provides a better representation of the mechanics of the composite than the
classical model. Indeed, this was brought out by the experiments of Mora and
Waas (2000) on honeycombs with either porous or very stiff inclusions.
In the case of traction boundary conditions, we use

1 V 0 (1) 0 0 (2)

ij Sijkl kl dV = ij Sijkl kl + i3 Si3k3 0k3 , i, j, k, l = 1, 2, (6.206)
2 V 2
where on the left we have the total complementary strain energy in the unit cell
(a function of Cauchy stresses ij ), and on the right we have the complemen-
tary strain energy of a couple-stress continuum (a function of volume-average
stresses ij0 and couple-stresses i3
0
of the unit cell). Here Sijkl (inverse of Cijkl ) is
(1) (2)
the microscale elastic compliance, while Sijkl and Si3k3 are the sought effective,
micropolar compliances.
Summing up, for the restricted (or Koiter) model of the composite, the
micropolar moduli are bounded from above and below, respectively, by dis-
placement and traction boundary conditions. In fact, as these bounds are
wide, we recommend three mixed types of loadings to get tighter results.
On the other hand, the characteristic lengths are highly insensitive to the
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
L L L
u2 u2 u2 u2 u2 u2
1. E-004
9.0
9.0E
1.0
8
0E-
3 0 3
.0
3
-00 -00 -0
E-
003
E-00
.0E .0E .0E
- 00 4
00
1 1 1
4
4 4
-00 7.0 -00
3
04
8
E-0 8 .0E E-0 .0E

.0 9 04 9
.0 E
9 . 0E
4 004 4
-0
7 -00 -0
0E-
6.0 -00
04
E-00 0E
.0E .0E 7 04
8. 4 8.
-0 8 . 0E
04 -0 5.0
E-00
6
4
04
6 -004
4
6
4
00
-00
04
04
.0E
4.
.0E
00
-0
-0
.0E
-0
0E-
.0E
.0E
E-
.0E
.0 E
4
-00
0
04
04
.0
7
.0E
7
7
6
6
4
-00
4
Cosserat Cauchy Cosserat Cauchy Cosserat Cauchy
/L = 2/100 /L = 2/10 /L = 2/1
FIGURE 6.6
Contour lines of vertical components of displacement in a panel loaded as shown, for various
scale ratios l/L, where l is the brick length and L is the macroscopic load print. (From Trovalusci
and Masiani, 2003. With permission).
mismatch in moduli, especially in the case of stiff inclusions, and this must
be contrasted with the sensitivity of moduli.
An interesting comparison of boundary value problems set up on struc-
tures made of Cauchy vis-a-vis Cosserat materials was conducted by
Trovalusci and Masiani (2003). Their research is motivated by the mechanics
of block-type masonry structures, whose stability (and, therefore, safety) is of
primary concern in places rich with ancient architecture like Italy and Greece.
Figures 6.6 and 6.7 show comparisons of symmetric boundary value problems
L L L
W12 (skw H)12 W12 (skw H)12 W12 (skw H)12
06 06 06
E-0 E-0 E-0
2.00 2.00 06 2.00
2
1.0E-0
.0
.0E
-006 06 -006
E-
1.00E 1.00E-0 1.00E

-0
00
1.0E-0
06
6
007
1.0E-0
E-
5.0
06
Cosserat Cauchy Cosserat Cauchy Cosserat Cauchy

06
1.0E-0
1.0E-0
2.0
5
2.
06 6
06
-0 00
.0E
E-
-0
0E-
0E
E-00
6 6 0 006
E
06
1.0 -00 00
-00
1.0 0E-
00
1.0E
06
0E 0E-
006
-006 2.0
1.
2.0 2.0
7
6
/L = 2/100 /L = 2/10 /L = 2/1
FIGURE 6.7
Contour lines of components of microrotation (Cosserat model) and macrorotation (Cauchy
model) in a panel loaded as shown, for various scale ratios l/L, where l is the brick length and
L is the macroscopic load print.
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
(a)
t
d
(b)
FIGURE 6.8
(a) Microstructure of trabecular bone at the so-called mesoscale level of Figure 7.15 in Chapter 7;
(b) Unit cell of an idealized periodic model following Gibson and Ashby (1988). (From Yoo and
Jasiuk, 2006. With permission).
of a plate-type structure, modeled in either of two fashions. It is clear that,

as the ratio l/L increases, the discrepancies between both models tend to
increase.
Another example of determination of micropolar moduli has recently been
developed in studies of trabecular bone (Yoo and Jasiuk, 2006), see Figure 6.8.
It appears to be a first-ever prediction and evaluation of apparent couple-
stress moduli for a 3D periodic orthotropic material.
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
Problems
1. Derive the local equations of motion of a micropolar continuum
from the global equations of conservation of linear and angular
momenta.
2. Generalize the Mohr circle concept and analysis to plane stress mi-
cropolar elasticity.
3. Prove the inequalities (6.32). Hint: use the Sylvester theorem.
4. Determine the convolution operation involved in the equation (6.60).
5. With reference to equations (6.72), demonstrate that , and H are
dispersive waves.
6. Examine the implications of equation (6.79).
7. With reference to equation (6.80), consider an isotropic, hemitropic
medium. Develop the corresponding form of the free energy func-
tion and obtain restrictions on all the elastic constants, more general
than those of (6.32). Hint: introduce three new elastic constants.
8. Extend Kirchhoffs uniqueness proof from the setting of linear elas-
ticity to linear micropolar elasticity.
9. Outline a theory of micropolar media for finite motions and strains.
10. Formulate the Clausius-Duhem inequality for micropolar elastic-
dissipative solids, and then outline a formulation of thermomechan-
ics with internal variables. Consult other sources as necessary.
P1: Binod
May 28, 2007 18:6 C4174 C4174C006
P1: Binod
May 28, 2007 18:8 C4174 C4174C007
7
Mesoscale Bounds for Linear Elastic
Microstructures
Three statisticians go deer hunting with bows and arrows. They

spot a big buck and take aim. One shoots and his arrow flies off
three metres to the right. The second shoots and his arrow flies
off three metres to the left. The third statistician jumps up and
down yelling; We got him! We got him!
The problem of effective properties of material microstructures has received

considerable, and ever-growing, attention over the past thirty years. By effec-
tive (or overall, macroscopic, global) is meant the response assuming existence
of a representative volume element (RVE). The RVE in the case of disorder
(i.e., lack of microstructural periodicity), implies that there must be some
scale larger than the microscale (e.g., single heterogeneity size) to ensure a
homogenization limit. Overall, most studies of effective properties simply
assume that the RVE is attained and do not specify its sizescarce prescrip-
tions of solid mechanics vaguely state that domains roughly 10 to 100 times
larger than the heterogeneity should be taken. In the late 1980s work began
on the determination of RVE in the sense of Hill (1963) and on continuum
random fields serving as input into stochastic finite element methods. This
then led to bounds that explicitly involve the size of a mesoscale domainthis
domain also being called a statistical volume element (SVE)relative to the
microscale and the type of boundary conditions applied to this domain. In
general, the trend to pass from the SVE to RVE depends on various factors,
and displays certain tendencies. This chapter discusses that issue for linear
elastic materials, thus setting the stage for nonlinear and/or inelastic mate-
rials as well as for continuum random field models and stochastic boundary
value problems, topics dealt with in the subsequent chapters.
7.1 Micro-, Meso-, and Macroscales

7.1.1 Separation of Scales
Continuum mechanics hinges on the concept of a representative volume
element (RVE) playing the role of a mathematical point of a continuum field
237
P1: Binod
May 28, 2007 18:8 C4174 C4174C007
approximating the true material microstructure. The RVE is very clearly de-
fined in two situations only: (1) unit cell in a periodic microstructure, and
(2) volume containing a very large (mathematically infinite) set of microscale
elements (e.g., grains), possessing statistically homogeneous and ergodic
properties. The approach via the unit cell is, strictly speaking, restricted to
materials displaying periodic geometries. When we consider case (2) we in-
tuitively think of a medium with a microstructure so fine we cannot see it
naturally then we envisage a homogeneous deterministic continuum in its place.
This situation, as suggested by Figure 7.1 of the preface, and called the sepa-
ration of scales

d<
L L macro , (7.1)
d
introduces three scales:
1. The microscale d, such as the average size of grain (or inclusion,

crystal, etc.) in a given microstructure; we initially assume the
microstructures to be characterized by just one size d.
2. The mesoscale L, size of the RVE (if so justified see below).
3. The macroscale L macro , macroscopic body size.
In equation (7.1) on the left we do admit two options, because the in-
equality d < L may be sufficient for microstructures with weak geometric
disorder and weak mismatch in properties; otherwise a much stronger state-
ment d L applies. Note also that the first inequality in (7.1)1 could even be
a weak one because we may be considering a microstructure with a nearly
periodic geometry and small mismatch in the properties of the phases. As
opposed to the periodic homogenization which relies on a periodic window
directly taken as the RVE (Figure 7.1[a]), the homogenization in random me-
dia is more complicated.
In any case, the issue of central concern is the trendeither rapid, moder-
ate, or slowof mesoscale constitutive response, with L/d increasing, to the
situation postulated by Hill (1963): a sample that (a) is structurally entirely
typical of the whole mixture on average, and (b) contains a sufficient number
of inclusions for the apparent overall moduli to be effectively independent of
the surface values of traction and displacement, so long as these values are
macroscopically uniform. In essence, (a) is a statement about the materials
statistics, while (b) is a pronouncement on the independence of effective con-
stitutive response with respect to the boundary conditions. Both of these are
issues of mesoscale L of the domain of random microstructure over which
smoothing (or homogenization) is being done relative to the microscale d and
macroscale L macro . These considerations, however, are not rigorous, because
neither spatial statistics nor mechanics (or physics) definitions of properties
have yet been introduced.
P1: Binod
May 28, 2007 18:8 C4174 C4174C007
Mesoscale Bounds for Linear Elastic Microstructures 239
(a)
(b)
FIGURE 7.1
(a) A disordered microstructure of a periodic composite with a periodic window of size L;
(b) one realization of a random composite B L/d of size L.
7.1.2 Basic Concepts

A note on determinism. In principle, any realization B() of the composite
B = {B(); }, while spatially disordered (i.e., heterogeneous), follows
deterministic laws of mechanics. The most preferred approach, dictated by
stochastic mechanics, would be to first ascertain what happens to each and
P1: Binod
May 28, 2007 18:8 C4174 C4174C007
every B() of B, starting from a certain random microstructure model, and

then pass to ensemble setting, by taking the averages or higher moments as
the need arises. In many situations, however, this may generate enormous
amounts of perhaps not very useful information.
Ensemble versus volume averaging. We reserve the overbar for spatial
(volume-type) averages, and for ensemble averages. That is, if we have
a random (n-component, real valued) field defined over some probability
space {, F , P} F being a -field and P a probability measureover some
domain X in R D of volume V
: X Rn , (7.2)
the said averages are

1
() (, x)dV (x) (, x)dP. (7.3)
V v
We assume the conditions necessary for the fulfillment of commutativity of

both operations to be satisfied (i.e., requirements of Fubinis theorem), so that

= . (7.4)
The existence of the integral (7.3)2 is assumed in accordance with the

ergodic theorem, while the interchangeability of both operations in (7.4) is
the subject of so-called ergodicity, or ergodic property, discussed in Chapter 2
and shortly recalled in Section 7.3 below.
7.1.3 The RVE Postulate

The random material onmesoscale, such as shown in Figure 7.1(b), is denoted
B L/d = B L/d (); with B L/d () being one realization. Properties on
mesoscale are also described by an adjective apparent (Huet, 1995), as opposed
to effective. The latter term pertains to the limit L/d as it connotes the
passage to the RVE, while any finite mesoscale involves statistical scatter
and, therefore, describes some statistical volume element (SVE). Note that the
separation of scales is also known in the solid mechanics literature as the
MMM principle (Hashin, 1983).
In the following, it will be convenient to describe the mesoscale by a nondi-
mensional parameter
= L/d (7.5)
in the range [0, ), so that B L/d , a mesodomain, will be written B , etc. Thus,
= 0 signifies the pointwise description of the material, while is the
RVE limit.
The setting is one of quasi-static loading, so that the body is governed
locally by the equilibrium equation
ij,j = 0, (7.6)
P1: Binod
May 28, 2007 18:8 C4174 C4174C007
with ij being the Cauchy stress, body forces being disregarded. The case of
dynamic loading is considered in Chapter 9.
For a mesoscale body B () of volume V , such as the microstructure
shown in Figure 7.1(b), we define volume average stress and strain

1 1
() = (, x)dV () = (, x)dV. (7.7)
V V
Assuming we deal with a linear elastic microstructure, the problem is to pass
from the random field of stiffness with fluctuations on the microscale
= C(, x) : , , xB (7.8)
to some effective Hookes law
= Ceff : , (7.9)
whereby (a) the dependence on (i.e., randomness) would be removed,

(b) the dependence on x (i.e. spatial fluctuations) of strain and stress fields
would also vanish, and (c) the independence of response with respect to
boundary conditions would be attained. It is intuitively expected that needs
to be large, but exactly how large it should be, is the key problem.
Suppose we are tackling a boundary value problem of a body on macro-
scopic length scales. In general situations, we have to use computational
mechanicssuch as, say, a finite element meshing of the bodyand this ap-
proach conventionally assumes that every single finite element is at least as
large as the RVE, although this is rarely verified. Thus, there arises a need
to know the rate of approach of SVE to RVE in function of , mismatch and
microgeometry of the phases, etc., and, whether that rate is too slow in a given
problem. If the latter is actually the case, one then needs to set up a stochastic
finite element (SFE) scheme to account for the microstructure-borne noise on
the mesoscale of any given finite element, see Chapter 8. Let us note here that
the typical recipes of solid mechanics (e.g., Lemaitre and Chaboche, 1994) say
that should be about 10 to 100 for the RVE to apply, but the analyses reported
below show that depends on the type of problem studied.
7.2 Volume Averaging

7.2.1 A Paradigm of Boundary Conditions Effect
Clearly, the attainment of the RVE is a function of the scale as well as the
mismatch in properties of inclusions versus matrix. To illustrate this point,
let us consider boundary distributions of displacement u3 and stress traction
t3 in two boundary value problems of the mesodomain B () of the matrix-
inclusion specimen of Figure 7.2 in antiplane elasticity. The material is piece-
wise uniform with perfectly bonded, isotropic phases, so that the governing
P1: Binod
May 28, 2007 18:8 C4174 C4174C007
FIGURE 7.2
Antiplane elastic responses of a matrix-inclusion composite, with nominal 35% volume fraction
of inclusions, at decreasing contrasts: (a) C (i) /C (m) = 1, (b) C (i) /C (m) = 0.2, (c) C (i) /C (m) = 0.05,
(d) C (i) /C (m) = 0.02. For (bd), the first figure shows response under Dirichlet boundary condi-
tions, while the second shows response under Neumann boundary conditions with 0 equal to
the volume average of stress computed in the Dirichlet problem.
P1: Binod
May 28, 2007 18:8 C4174 C4174C007
equation is
C ( p) 2 u = 0 p = m, i. (7.10)
Here m and i denote matrix and inclusion, respectively, and C3i3 j = C ( p) ij

(ij is the Kronecker delta) is the phase stiffness. Two boundary value problems
are considered: one of Dirichlet type:
u3 (x) = 3i
0
xi x B , (7.11)
where the strain 30 j is prescribed, and the other of Neumann type:
t3 (x) = 30j n j x B , (7.12)
where the stress 30j is prescribed; B denotes the boundary of B .

Figure 7.2(a) treats the situation of no mismatch in the material properties:
C (i) /C (m) = 1. and so we can interpret it as either a uniform displacement field
on the boundary B under 0 = (31 0
, 32
0
), with 32
0
= 0, resulting in a uniform
stress field on B , or a uniform stress field on B under 0 = (31 0
, 32
0
), with
32 = 0, resulting in a uniform displacement field on B . Evidently, both
0
problems are perfectly interchangeable because the microstructure is trivially

homogeneous. This then is the situation of the RVE.
Both boundary value problems become much more interesting when
C (i) /C (m) =

1. In Figures 7.2 (bd) we decrease the mismatch by first setting
it to 0.2, then 0.05, and finally 0.02. In each case, we first solve the Dirichlet
problem under 0 = (31 0
, 0), and find t(x). Next, we compute the volume
average t of t(x) on B , and set t0 = t to run the Neumann problem. We keep
0 identical in all four cases (ad).
Let us now define an apparent stiffness Cd in the displacement con-
trolled (d ) problem (7.11) via equation
= Cd : 0 , (7.13)
and apparent compliance St in the Neumann (t ) problem (7.12) via equation
= St : 0 . (7.14)
The latter allows us to define apparent stiffness in the traction controlled

1
problem as Ct = St .
The following points are noteworthy:
1. The volume average displacement of the resulting u3 (x) distribution

in the problem (7.11) differs from that in the problem (7.12).
2. The apparent stiffness in one boundary value problem is different
from that in the other one; this should not be surprising given the
preceding observation.
P1: Binod
May 28, 2007 18:8 C4174 C4174C007
TABLE 7.1
A Collection of Diverse Physical Problems Governed by the Laplace
Equation
Physical Subject u(= u3 ) (= 3i ) C(= C3i3j ) (= 3i )
Antiplane Displacement Strain Elastic moduli Stress

elasticity
Thermal Temperature Thermal Thermal Heat flux
conductivity gradient conductivity
Torsion Stress function Strain Shear moduli Stress
Electrical Potential Intensity Electrical Current
conduction conductivity density
Electrostatics Potential Intensity Permittivity Electric
induction
Magnetostatics Potential Intensity Magnetic Magnetic
permeability induction
Diffusion Concentration Gradient Diffusivity Flux
3. The degree to which Cd is different from (St ) 1 may be regarded

as an indication of the departure from the effective moduli Ceff in
separation of scales; as a measure of their closeness to Ceff one might
use the deviation of the product Cd : St from unity.
4. Although the governing partial differential equation is linear (and
we could even replace (7.10) by (Cij u3, j ) ,i = 0), the resulting prop-
erty is nonlinear as a function of actual realization , scale , mis-
match C (i) /C (m) , and type of boundary conditions (i.e., Dirichlet or
Neumann); see Markov (2000) for a related discussion.
For the sake of completeness, in Table 7.1 we collect various classical analo-
gies of problems locally governed by the Laplace equation C T,ii = 0 (or by
(Cij T, j ) ,i = 0) in two dimensions; see also (Hashin, 1983). Note here, with
reference to equations (10.39) and (10.40) of Chapter 10, that the thermal dis-
sipation th (which is analogous to the strain energy) is a scalar product of q
and T, divided by the absolute temperature T. However, when looking for
response in a steady-state and for small temperature changes (which is where
the linear conductivity applies), one can approximately treat T as a constant
(effectively, a volume-averaged quantity).
7.2.2 The Hill Condition

7.2.2.1 Mechanical versus Energy Definitions
Let us consider a body B () with a given microstructure, in which, as a
result of some boundary conditions and in the absence of body and inertia
P1: Binod
May 28, 2007 18:8 C4174 C4174C007
forces, there are stress and strain fields and . If we represent them as
a superposition of the means ( and ) with the zero-mean fluctuations
( and )
(, x) = + (, x) (, x) = + (, x), (7.15)
we find for the volume average of the energy density over B ()

1 1 1 1
U (, x) : (, x)dV = : = : + : . (7.16)
2V B () 2 2 2
Thus, we see that for the average of a scalar product of stress and strain fields
to equal the product of their averages
: = : , (7.17)
we need
: = 0. (7.18)
Relation (7.17) is called the Hill condition in the (conventional) volume average
form (Hill, 1963; see also Kroner, 1972, 1986; Huet, 1982, 1990; Sab, 1991). Some
authors (e.g., Stolz, 1986) call it the HillMandel macrohomogeneity condition,
after J. Mandel (1966).
Writing (7.18) in the index notation, we have

1
ij ij = (ij ij )(ij ij )dV
V V

1
= {[(ij ij )(ui ui )], j (ij,j ij,j )(ui ui )}dV
V V

1
= [(ij ij )(ui ui )]n j dS
V V

1
= [(ti ij n j )(ui ij x j )]dS. (7.19)
V V
Now, for an unbounded space domain ( ), the fluctuations are
negligible, but for a finite mesoscale, we find the necessary and sufficient
condition for (7.17) to hold

: = : (t n) (u x)dS = 0. (7.20)
B
This is satisfied by three different types of boundary conditions on the meso-

scale: uniform displacement (also called kinematic, essential, or Dirichlet)
boundary condition (d)
u(x) = 0 x x B ; (7.21)
P1: Binod
May 28, 2007 18:8 C4174 C4174C007
uniform traction (also called static, natural, or Neumann) boundary con-

dition (t)
t(x) = 0 n x B ; (7.22)
uniform displacement-traction (also called orthogonal-mixed) boundary con-

dition (dt)

t(x) 0 n u(x) 0 x = 0 x B . (7.23)
Here we employ 0 and 0 to denote constant tensors, prescribed a priori.

and note, from the strain average and stress average theorems: 0 = and
0 = .
Each of these boundary conditions results in a different mesoscale (or
apparent) stiffness, or compliance tensor; Huet uses the term apparent. Either
of these terms is used to make a distinction from the macroscale (or effective,
global, overall) properties that are typically denoted by eff or , see also Nemat-
Nasser and Hori (1993).
For a given realization B () of the random medium B , taken as a linear
elastic body ( = C(, x) : ), on some mesoscale , condition (7.21) yields an
apparent random stiffness tensor Cd ()sometimes denoted Ce ()with
the constitutive law
= Cd () : 0 . (7.24)
Similarly, the boundary condition (7.22) results in an apparent random

compliance tensor St ()sometimes denoted Sn ()with the constitutive
law being stated as
= St () : 0 . (7.25)
The third type of boundary condition, (7.23), involves a combination of

(7.21) and (7.22); it results in a stiffness tensor Cdt
(). In fact, this condition
may best represent actual experimental setups; the other two are nearly im-
possible to realize physically. For example, (7.23) may signify displacement
boundary conditions on two parallel sides, and traction-free boundary con-
ditions on the remaining two parallel sides. Or, it may signify pure shear-type
loading through boundary conditions (see Figure 7.3):
11
0
= 22
0
12
0
= 0. (7.26)
7.2.2.2 Order Relations Dictated by Three Types of Loading

Note from the discussion above that the use of (7.17) assures the equivalence of
the properties from the mechanical standpointthat is, via apparent Hookes
law (7.21) or (7.22)with the properties from the energy standpoint.
P1: Binod
May 28, 2007 18:8 C4174 C4174C007
FIGURE 7.3
Possible loading under orthogonal-mixed boundary condition (7.23).
Both approaches are equivalent for a homogeneous materialthat is, the

RVEbut not necessarily so for a heterogeneous one, B (), of size finite
relative to the microscale heterogeneity ( < ). In fact, the following relation
ordering the Neumann and Dirichlet apparent moduli holds: [St ()]1
Cd (). A proof of the above in the framework of functional analysis has been
given by Suquet (1986), and another one in the framework of general convex
analysis applied to the nonlinear case by Willis and Talbot (1989), which was
followed by Huet (1990). Later on, Hazanov and Huet (1994) extended it to
the following:
[St ()]1 Cdt

() C (),
d
(7.27)
that is, the modulus of B () obtained under mixed dt-conditions (7.21)

always lies between the moduli obtained under the t-conditions (7.22) and the
d-conditions (7.23). Other consequences of (7.27)especially, in the context
of orthotropic materialswere discussed by Hazanov and Amieur (1995).
Another important result, also due to Huet (1990), is that
1
St () Ceff () Cd () . (7.28)
That is, the effective modulus Ceff () always lies between the harmonic av-
erage of moduli obtained under the Neumann boundary conditions on the
ensemble B and the arithmetic average of moduli obtained under the Dirich-
let conditions on the same ensemble.
Hill (1967) and Mandel (1966) gave a qualitative estimate of the error
between St () and Cd ()
St () : Cd () = 1 + O (1/) 3 (7.29)
P1: Binod
May 28, 2007 18:8 C4174 C4174C007
Many quantitative estimates of -dependenceor what may be called scal-

ingwere computed for many different materials by Huet and coworkers,
this author and coworkers; see also Terada et al. (2000), Kanit et al. (2003),
Vinogradov (2001).
Note: Other definitions of the RVE than that employed here, without the
concept of SVE, have also been considered (e.g., Stroeven et al., 2004). Recently,
Sab and Nedjar (2005) introduced a periodization of random media, which
supports the computational mechanics results of Gusev (1997) involving an
assumption of some finite scale periodicity in random microstructures and
confirmed that analysis.
7.2.3 Apparent Properties

In general, if we consider a body B () of volume V subjected to a volume
average strain by the boundary condition (7.21):
ui (x) = ij0 x j x B , (7.30)
given the continuity of displacements throughout, we have the average strain

theorem:

1 1
ij = ui, j + u j,i dV = ui n j + u j ni dS
2V V 2V S

1 0
= ik xk n j + 0jk xk ni dS = ij0 . (7.31)
2V S
d d
The apparent stiffness Ckli j ( C ) of B () made of a linear elastic microstruc-
ture may now be defined by
ij = Cijkl
d 0
kl , (7.32)
where is the volume average stress. Alternatively, we may consider the

volume average energy density in B ()

1 1 1
U= ij ij dV = ij ui, j dV = ij ui n j dS
2V V 2V V 2V S

1 1 1
= ij ij0 dV = ij ij0 = kl0 Ckli d
j ij ,
0
(7.33)
2V V 2 2
where the equilibrium ij,j = 0 was used. Thus, the apparent stiffness Ckli
d
j may
be defined either from the mean stress or from the mean energy density U
if the boundary condition (7.30) is imposed; see also Willis (1981) and Nemat-
Nasser and Hori (1993).
On the other hand, considering the traction boundary condition (7.22),
that is,
ti (x) = ij0 n j (x) x B , (7.34)

P1: Binod
May 28, 2007 18:8 C4174 C4174C007
we first have the average stress theorem:

1 1 1
ij = ij dV = ik k j dV = ik x j,k dV
V V V V V V

1 1 1
= ik x j ,k dV = ik0 x j nk dS = 0 x j,k dV = ij0 . (7.35)
V S V S V V ik
We can now define the apparent compliance St from this relation
ij = Sijkl
t
ij0 , (7.36)
or from the volume average energy density in B ():

1 1 1
U= ij ij dV = ij ui, j dV = ij ui n j dS
2V V 2V V 2V V

1 1 1
= 0 ui, j dS = ij0 ij = kl0 Skli t
j ij .
0
(7.37)
2V V ij 2 2
j may be defined either from

t
This shows that the apparent compliance Skli
or from U if the boundary condition (7.34) is imposed.
7.3 Spatial Randomness

7.3.1 Stationarity of Spatial Statistics
In the following we revisit the key concepts of spatial homogeneity (stationar-
ity) and ergodicity from the standpoint of what is required by Hills definition
of RVE. With reference to Chapter 2, we shall do this in terms of a material
property (or a vector of properties) Z entering the heterogeneous medium
model as a random field over the D-dimensional physical space
Z : R D R. (7.38)
In the case of an r -phase material microstructure, Z is itself described by a

random indicator function r of all the phases. All that follows can then be
generalized to a vector or tensor random field Z, as may be the case with a
linear elastic microstructure represented by a random stiffness field C. Now,
let us recall two well-known classes of spatial homogeneity:
Strict-sense stationary (SSS) random fields. This requires that all n-order
probability distributions Fn are invariant with respect to arbitrary shifts x ,
and for any n and any choice of xi , they satisfy
Fn (z1 , . . ., zn ; x1 , . . ., xn ) = Fn (z1 , . . ., zn ; x1 + x , . . ., xn + x ). (7.39)

P1: Binod
May 28, 2007 18:8 C4174 C4174C007
Wide-sense (or weak-sense) stationary (WSS) random fields. The ensemble

mean is constant and its finite-valued covariance depends only on the shift h
from x to x + h
Z(x) , [Z(x) Z(x)] [Z(x + h) Z(x + h)] K Z (h) < .

(7.40)
Clearly, SSS implies WSS whenever the first-order distribution F1 yields a

finite second moment. It is through the covariance function K Z (h) that we
introduce the concept of correlation distance lc , which, in turn, gives an indi-
cation of decay of correlations between two different points. If this decay is
such that Z(x) and Z(x + h) become asymptotically uncorrelated according to
lim|h| K Z (h) 0, then, using lc we would rewrite the separation of scales
(7.1) as

lc <
L L macro . (7.41)
lc
In other words, the RVE of size L = V 1/D (D = 1, . . ., 3) should be at least

larger than lc , and the material could be taken as homogeneous beyond L.
This concept can also be applied to the SSS fields.
There also exist two more general classes of spatially homogeneous fields:
(1) intrinsically stationary (locally homogeneous) random fields, and
(2) quasi-stationary random fields. These types of random fields become rele-
vant when, in general, there is no hope of establishing even the WSS property.
In those situations, strictly speaking, there is no certainty of having the RVE
in the sense employed in this chapter, and the SVE is needed.
7.3.2 Ergodicity of Spatial Statistics

Basic considerations. Following Chapter 2, let us recall that the ergodic prop-
erty means that the spatial average over any realization equals the ensemble
average at any specific point x:
Z() = Z(x). (7.42)
Also as noted earlier, this is a property that:
1. Must be checked a priori for a given material microstructure.
2. Is hard to verify because the measurements are necessarily finite in
number both in as well as in B.
3. Is practically impossible to verify because the measurements are
supposed to be carried out in a pointwise sense, on domains of
infinite spatial extent.
The framework we develop below applies to materials satisfying (7.42).
Ergodic response. Any given boundary value problem for an elastic body
B = {B(); } is defined by the equilibrium equation (7.8), Hookes law,
P1: Binod
May 28, 2007 18:8 C4174 C4174C007
compatibility conditions, and some boundary conditions. Now, subject B()

to a uniform boundary condition such as (7.15) or (7.20) and find the ensemble
of solutions
{(, x), (, x), u(, x); , x V}, (7.43)
which would represent a complete solution of the stochastic mechanics prob-

lem. Then, at any point x, we may define effective moduli Ceff (x) as
(x) = C(x) : (x) = Ceff (x) : (x). (7.44)
Note:
1. For any given boundary conditions, Ceff is, in general, a function

of x.
2. Ceff also generally depends on the boundary conditions applied.
3. Assuming B is ergodic in the sense that the stiffness field C is
ergodic, the explicit dependence on the location in (7.44) vanishes
and we can write
(x) = or (x) = , (7.45)
providing (7.29), respectively (7.30), is applied and the displacement

and stress fields are continuous; consequently,
= Ceff : . (7.46)
4. Providing the mesodomain tends to macroscale , write the

Hill condition (7.17) as
: = : . (7.47)
Of course, by analogy to (7.18), we have

: = 0, (7.48)
which means that stresses and strains are statistically uncorrelated.
7.4 Hierarchies of Mesoscale Bounds

7.4.1 Response under Displacement Boundary Condition
We take the 2D square-shaped mesodomain (a mesoscale window) of a com-
posite body B () of Figure 7.4 to be described everywhere by the local
stress-strain relations = C(, x) : . Suppose that this body is now evenly
partitioned into four square-shaped bodies Bs (), s = 1, . . ., 4. We define two
types of uniform displacement boundary conditions, in terms of a prescribed
P1: Binod
May 28, 2007 18:8 C4174 C4174C007
FIGURE 7.4
Partition of the window of Figure 7.1(b) into four subwindows.
constant strain 0 , over the mesodomain with any given microstructure:

unrestricted
u(x) = 0 x x B , (7.49)
and: restricted
ur (x) = 0 x x Bs s = 1, . . ., 4. (7.50)
The superscript r in equation (7.50) indicates a restriction. That is, (7.49) is

given on the external boundary of the mesodomain, whereas (7.50) is given on
the boundaries of each of the four submesodomains, Figure 7.4. Let us note,
by the strain averaging theorem (assuming perfect bonding throughout the
material), that the volume average strain is the same in each sub-mesodomain
and also equals that in the large mesodomain
0 = = s . (7.51)

Let ,
be any kinematically admissible fields: they satisfy everywhere
the local stress-strain relations = C(, x) : and the displacement boundary
condition (7.49), with ij = u(i, j) but is not necessarily in equilibrium.
Now,
there is a minimum potential energy principle for the fields , in B ()
(e.g., Hill, 1950)

1 1
dS
tu :
dV t udS : dV. (7.52)
Bt 2 B Bt 2 B
P1: Binod
May 28, 2007 18:8 C4174 C4174C007
For the displacement boundary condition, Bu () = B () and Bt () = ,

so that

1 1
: dV = U (, ) U ,
= :
dV, (7.53)
2 B 2 B
or
: :
. (7.54)
However, the Hill condition (7.17), combined with the fact that 0 = ,
allows us to write
: :
. (7.55)

Because the solution r ,
r under the restricted condition (7.50) is an admis-
sible distribution under unrestricted conditions (7.49) (but not vice versa), we
have
: r :
r . (7.56)
This in turn implies a weak inequality between the mesoscale stiffness tensors
obtained under unrestricted (Cd ()) and restricted (Cdr
()) conditions
1 d
4
Cd () Cdr
() = C () = /2 (7.57)
4 s=1 s
where it follows from the relation
1 d
4
= Cdr
() :
0
Cd () = C (). (7.58)
4 s=1 s
That is, the effective stiffness of a partitioned domain subjected to (7.50)

involves respective stiffnesses of four subdomains.
Now, in view of the tacitly assumed statistical homogeneity and ergodicity
d
of
d material,
the ensemble
d averaging of (7.57) allows us to replace C () by
dr
C , and C () by C , so that
d d
C C = /2. (7.59)
By applying this inequality

to ever larger mesodomains ad infinitum we get
a hierarchy of bounds on Cd from above
d
C . . . Cd Cd . . . Cd1 CV = /2. (7.60)
In fact, we have

Cd = Ceff
=C
eff
(7.61)
for the macroscopically effective response ( ) because, by the ergodicity

argument, it must be deterministic. On the upper end, the hierarchy stops at
P1: Binod
May 28, 2007 18:8 C4174 C4174C007
the scale of a single heterogeneity. Now, since the single heterogeneitylike

an inclusion or a crystalis homogeneous, the uniform strain is true strain,
so that ensemble averaging gives the Voigt bound CV .
7.4.2 Response under Traction Boundary Condition

The above suggests an analogous procedure in proving a hierarchy bounding
Ceff ( (Seff ) 1 ) from below. The proof is analogous, providing we first replace
the displacement by traction boundary conditions (in terms of a prescribed
constant stress 0 ), again of two types: unrestricted
t(x) = 0 n x B , (7.62)
and: restricted
tr (x) = 0 n x Bs s = 1, . . ., 4. (7.63)
The superscript r indicates a restriction as before. Next, by the stress averag-

ing theorem, the volume average stress is the same in each sub-mesodomain
and also equals that in the mesodomain
0 = = s. (7.64)
The key inequality is obtained from a minimum complementary energy

principle for statically admissible fields ,
in B ()

1 1 t udS,
: dV t udS :
dV (7.65)
2 B B
u 2 B Bu
and noting that Bt () = B () and Bu () = , so that : :

.
Using the Hill condition leads then to
St () Str
(), (7.66)
and, upon ensemble averaging, to

t t
S S = /2. (7.67)
By applying this inequality

toever-larger mesodomains
ad infinitum we get
a hierarchy of bounds on St from above (i.e., on Cd from below)

St . . . St St . . . St1 S R = /2, (7.68)
where

=S
St = Seff eff
= (Ceff ) 1 . (7.69)
P1: Binod
May 28, 2007 18:8 C4174 C4174C007
7.4.3 Scale-Dependent Hierarchy

Combined hierarchy. Combining relations (7.60) with (7.68), we arrive at a
scale-dependent hierarchy of bounds on the macroscopically effective moduli
t 1 1 t 1 d d d
S1 . . . St S . . . C C . . . C 1
. . . Ceff
= /2. (7.70)
These hierarchies were essentially first derived by Huet (1990); see also
(Huet, 1991, 1997). A more rigorous proof using techniques of homogenization
and probability theories was given by Sab (1992), see Section 7.4.4 below.
The decrease of the upper (displacement-controlled) bound with increasing
scale appears to have first been demonstrated on planar random networks of
Delaunay topology by Ostoja-Starzewski and Wang (1989); as mentioned in
Chapter 4, the uniform traction conditions, however, could not be applied in
a unique way to such a disordered discrete system.
Spatial statistics aspects. Considering that the hierarchy (7.70) is stated in
terms of the averages, it suffices to choose the setting of material properties
specified via wide-sense stationary random fields. Furthermore, note that in
the case of C being an isotropic random field, Ceff should become an isotropic
tensor involving two Lame constants eff and eff . On the other hand, for an
orthotropic random field C, Ceff should become an orthotropic tensor; a two-
dimensional example in the setting of machine-made paper is discussed in
Chapter 9.
Extension to noncommensurate partitions. The hierarchy (7.70) has been
shown to hold for commensurate partitions, i.e., = /2. We may, however,
extend these inequalities to an arbitrary pair of mesoscales < not just for
commensurate ones involving partitions in which = n , n being a natural
number. It will suffice to focus on the Dirichlet bounds, because then the
Neumann bounds follow by an analogous argument. To this end, consider
two separate cases of the hierarchy (7.60) for commensurate partitions: one
at an arbitrary 1 and another at 2 , whereby 1 < 2 < 21 . Thus, in the first
case, we have a sequence of inequalities

Cd41 Cd21 Cd1 , (7.71)
while in the second case, we have another sequence

Cd42 Cd22 Cd2 (7.72)
We prove by contradiction: assume that inequalities (7.71) and (7.72) are not
consistent with each other in the following way

> Cd42 < Cd41 > Cd22 < Cd21 > Cd2 < Cd1 > (7.73)
This, however, would imply that the microstructure is characterized by a

Dirichlet bound which, while displaying an overall decreasing behavior
P1: Binod
May 28, 2007 18:8 C4174 C4174C007
with , also fluctuates with a period increasing like n. One can certainly
construct a random, ergodic microstructure with such a scale-dependent fluc-
tuation. This would, however, contradict the assumption of statistical homo-
geneity of the material properties which is always taken in the definition of
the RVE.
Summarizing, we conclude that the hierarchy (7.70) holds for noncom-
mensurate mesoscale sequences. Other than (7.73) types of inconsistencies
between (7.71) and (7.72) may also be considered, but then one is led to similar
contradictory conclusions as above.
7.4.4 Homogenization Theory Viewpoint

We assume the compositesuch as that of Figure 7.2to be made of a finite
number of r phases each of which is linear elastic and elliptic: 0 < e : C : e <
, e =
0. Next, following Sab (1992), let F ( B, ) represent a real functional
of the open-bounded domain B of volume V and random field with the
following five properties:
1. F is a property of the medium invariant with respect to any trans-

lation in the material domain.
2. F or any partition of the domain B into n disjoint subdomains, F
satisfies a subadditivity property
n
Vi
F ( B, ) F ( Bi , ) B = i=1
n
Bi . (7.74)
i=1
V
3. F is a measurable mapping with respect to the sample space of

outcomes .
4. is a statistically homogeneous, ergodic random field (recall
Chapter 2).
5. F is uniformly bounded in B and in the sense that there exists a
real b, such that
|F ( B, )| 0 B, . (7.75)
Let us now take B to be a square-shaped domain, with side of length L,

and which contains some microstructure of characteristic microscale d, see
Figure 7.2. With the conditions 15 satisfied, we can adopt the result that
there exists a nonrandom (i.e., deterministic) constant F hom such that, for all
with probability one, we will have
lim |F ( B, )| = F hom B, , (7.76)

L/d
with the bound
inf( F ( B, )) = F hom B, . (7.77)

L/d
P1: Binod
May 28, 2007 18:8 C4174 C4174C007
This limit is understood in the sense of the homogenization theory: x

F
(x) = F (x/
) F
(y), where x and y are the so-called slow (macroscopic)
and fast (microscopic) variables, respectively, and
is a small parameter,
reciprocal of our = L/d. The limit should be taken here in such a
way that L is kept finite so as to keep the energy finite. If F represents the
volume average elastic energy density (or complementary) energy density,
then stands for the stiffness C (respectively, compliance S) tensor of the
domain B, and from (7.77) one has
sup S1 = (Shom ) 1 = Chom = inf S, (7.78)

L/d L/d
where (Shom ) 1 = Chom is the macroscopic (effective) stiffness tensor Ceff in

the sense of Hill (1963). However, (7.78) does not assert that the averages S1
and C are monotonic functions of L/d (= ).
Finally we note that, for the RVE B (), in the notation of homogeniza-
tion theory, we have the macrostress and macrostrain
() = () E() = (). (7.79)
7.4.5 Apparent Moduli in In-Plane Elasticity

7.4.5.1 General Considerations
When considering the apparent constitutive law of a planar elastic material
ij = Cijkl (, x)kl i, j, k, l = 1, 2, (7.80)
we must, in general, deal with an arbitrary anisotropy. Thus, to determine six

d t
unknown values for Cijkl (or Sijkl ) for B () we need six tests, Figure 7.5. When
seeking Cijkl s, each test is run by applying the affine displacements on B so
d
that the strain energy density is (V is the volume of B ())
V V V 0 2 0 2 0 2
U = ij ij0 = ij0 Cijkl kl0 = C1111 11 + C2222 22 + 4C1212 12
2 2 2

+ 211
0
C1122 22
0
+ 4220
C2212 12
0
+ 412
0
C1211 11
0
. (7.81)
In each and every test, separately, the energy is found by computational

mechanics, and this is set equal to a corresponding special form of (7.81).
FIGURE 7.5
Six tests: #1, #2, . . ., #6 from left to right, to determine the six unknowns of the in-plane stiffness
tensor Cijkl .
P1: Binod
May 28, 2007 18:8 C4174 C4174C007
For example, in test #1,

V 0 2
U = C1111 11 , (7.82)
2
from which we infer C1111 . Similarly, from the test #2 we find C2222 , while test
#3 gives C1211 . We are then in a position to find C1122 from
V 0 2 0 2
U = [C1111 11 + C2222 22 + 11
0
C1122 22
0
], (7.83)
2
and then C2212 as well as C1211 in analogous fashion.
In practice, one can proceed by truly carrying out six tests, or by carry-
ing out only the three tests #1#3 of Figure 7.5, and then combining their
results through a superposition. In any case, the solution necessarily involves
some computational mechanics method for discretization of the compos-
ite, such as spring networks, finite elements, or boundary elementsrecall
Chapters 3 to 5. Indeed, in Chapter 4 we discussed such a method for truss
systems.
In the case of traction boundary conditions, determination of the apparent
compliances follows the same type of approach and one works with the com-
plementary energy, that is,
V V V 0 2 0 2 0 2
U = ij0 ij = ij0 Sijkl kl0 S1111 11 + S2222 22 + 4S1212 12
2 2 2

+ 211
0 0
S1122 22
0
+ 4220 0
S2212 12
0
+ 412 S1211 11
0 0 0
. (7.84)
An extension of this methodology to three dimensions is straightforward.

Examples of bounds obtained in this way are given in the aforementioned
works of Huet and his coworkers (Amieur et al., 1995). A numerical method
for materials having a linear elastic truss microstructure is discussed in Section
4.2 of Chapter 4.
7.5 Examples of Hierarchies of Mesoscale Bounds

7.5.1 Random Chessboards and Bernoulli Lattices
In this section we focus on antiplane elasticity of random microstructures,
which, as recalled in Table 7.1, also gives information on various other physi-
cal problems of material systems having the same morphology. With reference
to Chapter 1, let us now consider the Bernoulli lattice process p,a on a Carte-
sian lattice of spacing a with each point of this lattice being of type 1 (or 2)
with probability p (respectively, q = 1 p) independently of all the other
points. Evidently, p and q define the volume fractions of both types of phases
(1 and 2). Clearly, the local stress and strain concentrations cannot be re-
solved, but the statistics of such a simple system gives an indication of the
P1: Binod
May 28, 2007 18:8 C4174 C4174C007
statistics of random media because this, perhaps, is the simplest setup in

which to investigate the scale and volume fraction dependence of the en-
semble average estimates based on the essential (e) and natural (n) boundary
conditions. The notation e and n is equivalent to d and t, respectively. After

Ostoja-Starzewski and Schulte (1996), in Figures 7.6 through 7.8 we give Ce
1
and Sn at = 4, 10, and 20 (for the highest contrast) and their comparison
with Hashin-Shtrikman) upper and lower bounds
CuH = C2 + f 1 [1/(C1 C2 ) + f 2 /2C2 ]1

ClH = C1 + f 2 [1/(C2 C1 ) + f 1 /2C1 ]1 . (7.85)
Observe that CuH and ClH are outside the mesodomains of size 4. In other
words, relatively very small mesodomains can give tighter mesoscale bounds
than those of Hashin.
Note that the problem of -dependence, especially in the setting of such
binary systems, is akin to the so-called finite-size scaling in statistical physics,
but the attention in that area has always been focused on the phase transition
problems (Cardy, 1988). The approach to such a transition at about 2/3 volume
fraction of the soft phase is shown in Figure 7.8 at contrast 1000. However,
in contradistinction to the terminology of phase transitions, we now have a
different tool to describe the scale dependence. The particular case of p =
q = 0.5 has been studied in Ostoja-Starzewski and Schulte (1996), and it was
found that
Ce = exp[ m ] Sn = exp[ n ], (7.86)
where m and n are functions of the contrast
m = 3.8 0.14 n = 2.4 0.59 . (7.87)
These results were obtained from computations over a range of scales 1

through 1000. While the smallest scale can be calculated explicitly as the
Voigt and Reuss bounds, the largest involved a lattice of 1000 1000 nodes,
that is, having 106 degrees of freedom. The parameter space of contrast and
volume fraction is vast, and therefore only select cases can be run numerically
at these large scales. But, (7.86 and 7.87) give an idea of the functional forms
of scaling laws for other volume fractions.
Now, the Bernoulli lattice at volume fraction below, say, 30% can also be
interpreted as a very crude model of a disk matrix compositeagain with one
degree of freedom per disk. Given the fact that a more realistic spring network
model requires several (at least five) lattice spacings per disk, a lattice of some
50005000 nodes (25106 degrees of freedom) would have to be run. Thus, the
above scaling laws provide the best available indication of finite-size scaling
of both bounds, Ce and Sn , of disk-matrix composites.
P1: Binod
May 28, 2007 18:8 C4174 C4174C007
5
4
4 10
1
C e S n
0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
FIGURE 7.6 1
Bounds on tr (Ceff ) for a random two-phase lattice at contrast 10, showing tr Ce and tr Sn
at = 4 and 10; also shown, by dashed lines, are Hashin upper and lower bounds CuH
and ClH .
50
4
40 10
e 1
C S n
30
20
10
0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
FIGURE 7.7 1
Bounds on tr (Ceff ) for a random two-phase lattice at contrast 100, showing tr Ce and tr Sn
at = 4 and 10; also shown, by dashed lines, are Hashin bounds CuH and ClH .
P1: Binod
May 28, 2007 18:8 C4174 C4174C007
500
4
400 10
20
300 n 1
C e S
200
100
0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
FIGURE 7.8
Bounds on tr (Ceff ) for a random two-phase lattice at contrast 1000, showing tr Ce and tr Sn 1
at = 4, 10 and 20; also shown, by dashed lines, are Hashin bounds CuH and ClH .
7.5.2 Disk-Matrix Composites

The hierarchy of bounds (7.70) is now illustrated on the example of a disk-
matrix composite at volume fraction 20%. These microstructures have been
generated via the planar Poisson point process for disk centers, with sequen-
tial inhibition rule (recall Chapter 1) that prevents any two Poisson points
from coming closer than 110% of diameter, so as to avoid the numerically and
analytically difficult problem of very narrow necks between disks.
We consider two examples of this composite: one with relatively soft
and another with relatively hard inclusions. In the first case, we have a soft
matrix (C (m) = 1) and inclusions C (i) = 102 in Figure 7.9(a), and C (i) = 104 in
Figure 7.9(b). Clearly, as the contrast in the composite increases, the bounds
take larger mesodomains to converge. Basically, in order to attain the equal-
ity Sn : Ce = 1 within, say, 10%, one has to take mesodomains that are some
ten and fifty times larger than a single inclusion, for these two contrasts,
respectively.
In Figure 7.10 we show results for the opposite case: soft inclusions in a
hard matrix, with the same volume fraction 20%. Again, we consider two cases
of contrastC (i) /C (m) = 102 and 104 while keeping the matrix at C (m) = 1.
The first one is shown in Figure 7.10(a), the second one in Figure 7.10(b),
and both were obtained with exactly the same spring network resolution as
above. As before, an increase in the contrast in the composite has the effect of
slowing down the convergence of Sn : Ce to unity with increasing, but, by
comparison with Figure 7.9, this convergence is relatively much slower (!) for
P1: Binod
May 28, 2007 18:8 C4174 C4174C007
15
10
5
C e
1
0
0 10 20 S n 30 40 50

(a)
300
200
e
C
100
0 1
0 10 Se 20 30 40 50
(b)
FIGURE 7.9
A hierarchy of scale-dependent bounds on tr (Ceff ) of the disk-matrix composite at contrasts 102
(a), and 104 (b) (After Ostoja-Starzewski, 1998).
1.0
0.9
0.8
0.7 C e
0.6
0.5
n 1
0.4 S
0.3
0.2
0.1
0.0
0 10 20 30 40 50
(a)
1.0
0.9
0.8
e
0.7 C
0.6
0.5
0.4
0.3
0.2 n 1
0.1 S
0.0
0 10 20 30 40 50
(b)
FIGURE 7.10
A hierarchy of scale-dependent bounds on tr (Ceff ) for the disk-matrix composite at contrasts
102 (a), and 104 (b). (After Ostoja-Starzewski, 1998).
P1: Binod
May 28, 2007 18:8 C4174 C4174C007
extreme contrasts. It follows that one needs to go to very large scales in order
to homogenize such a composite material. This is the principal difference
from the case of high contrasts, and is indicative of all the material systems
with soft inclusions of whatever shape.
Note: The tendency for both bounds to converge onto Ceff increases as the
contrast C (i) /C (m) tends to unity. Thus, for low contrasts, the RVE size may
be taken of the order of just a few inclusion diameters.
Note: In all the systems one has the option of using the orthogonal-mixed
boundary condition, which results in an intermediate mesoscale response,
having a weaker scale dependence than Ce and Sn . The trade-off is this: no
bounding property, but the tendency to asymptote to, and attain Ceff , is most
rapid. This property will be displayed in an elastoplastic composite in a later
chapter.
The above results were explored in more depth by Jiang et al. (2001).
They studied responses under several different boundary conditions: uniform
displacement, uniform traction, periodic, and mixed ones (a combination of
any of the first three) to evaluate the mesoscale moduli.
7.5.3 Functionally Graded Materials

Interfaces in composite materials influence their local fields and effective
properties. Theoretical studies in this area represent the interface as either
a 2D bounding surface, or as a 3D region of certain microstructure, called
an interphase (e.g., Drzal, 1990), Figure 7.11(a). The inhomogeneity of the
interphase may be due, for example, to the chemical reaction(s) or diffusion.
Composites with inhomogeneous interphases have been studied by various
researchers, whereby the interphase region was assumed to be isotropic with
one property, such as Youngs modulus, varying linearly or nonlinearly and
Poissons ratio taken as constant.
Materials with interphasesalso called functionally graded materials
(FGM)present a fundamental challenge to the mechanician not only because
of their spatially varying nature, but also due to the need to account for the
spatially graded random microstructure. In Ostoja-Starzewski et al. (1996a)
we represented the interphase as a zone of two randomly interpenetrating
phases with radially dependent statistics, assuming a linear distribution of
the indicator function
P{(black) = 1} = A1r + A2 . (7.88)
To make things specific, with reference to Figure 7.11(a), we admitted two

models of the microstructure: a fine-grained model with topology of a ran-
dom chessboard, and a coarser-grained model with a geometry of a Voronoi
tessellation, whose cells are occupied at random by either one of two phases.
The analysis of in-plane conductivity Cij (r ) was then based on a mesoscale
window larger than the scale of heterogeneity in the FGM. To define apparent
P1: Binod
May 28, 2007 18:8 C4174 C4174C007
Matrix
Fiber
Interphase
a0 b0 r
(a)
FIGURE 7.11(a)
Sketch of the fiber-interphase matrix system, and the mesoscale window shown in a random
chessboard model (top right) and a two-phase Voronoi model (bottom right).
properties of this window, we employed two boundary value problems: es-

sential and natural. As a result, local (radially dependent) properties of the
interphase turned out to be orthotropic and nonlinear in r ; they could be
approximated by polynomials of the form
Cij (r ) = k0 + k1r + k2r 2 + k3r 3 + (7.89)
These functions (in the r, -coordinate system) represent ensemble aver-

1
ages Ce and Sn for a given window scale , Figure 7.11(b). The temper-
ature field T(r, ) in the interphase zone resulting from the field equation

2 T Crr 1 T 1 Cr T Cr 2 T C 2 T
Crr + + Crr + +2 + 2 =0
r 2 r r r r r r r r 2
(7.90)
was then used as input to the effective medium theory allowing prediction of
the overall transverse conductivity of the composite. At this latter stage we
used the composite cylinders assemblage model to account for the interac-
tions of the fibers.
Let us note that the mesoscale may be set up quite arbitrarily over
a range of values, but, in contrast to other problems discussed earlier, the
finite size of the interphase precludes a passage to infinity. We thus have two
P1: Binod
May 28, 2007 18:8 C4174 C4174C007
L
C(f )
C(m)
0 a a0 b0 b r
(b)
FIGURE 7.11(b)
A typical bi-phasic distribution along the radial coordinate r, where a 0 r b 0 , displays
the heterogeneity of the interphase region. C ( f ) is the fiber phase, while C (m) is the matrix
phase. Four curves, top to bottom, are C(e) (e) (n) (n)
, Crr , C a nd Crr . Also, a mesoscale L = 2(a 0 a ) =
2 (b 0 b) is indicated.
fields bounding the global responsea precursor of two random fields to be

introduced in the next chapter.
7.5.4 Effective and Apparent Moduli of Multicracked Solids

7.5.4.1 Scale-Dependent Hierarchies of Bounds: Numerical Results
Let us again make use of one of the random media models developed in Chap-
ter 1. This time, as we are interested in planar fields of randomly placed nee-
dles, we consider a random fiber field generated from the Poisson point field
Figure 7.12(a) presents its typical realization B(); (Ostoja-Starzewski, 1999a).
The field density chosen here is 0.015. As in the previous section, the mate-
rial has two locally isotropic phasesmatrix (m) and inclusions (i)and we
keep C (m) = 1 and vary C (i) . Hierarchies of bounds are shown in Figure 7.12(b).
Before we proceed further, we note that the work on effective moduli of
materials with microcracks dates back to Vakulenko and Kachanov (1971) and
Budiansky and OConnell (1976), while a comprehensive review was given by
Kachanov (1993). An experimental confirmation of the results of an effective
medium (MoriTanaka) theory for low crack densities, typically employed by
Kachanov and others, has been given in Carvalho and Labuz (1996). The effec-
tive medium theory, however, cannot say anything about the finite-size scal-
ing of (mesoscale) moduli and their statistics; nor is it reliable for higher crack
densities. The importance of mesoscale bounds is now increasingly being rec-
ognized in wave propagation studies in geophysics (Saenger et al., 2006).
P1: Binod
May 28, 2007 18:8 C4174 C4174C007
trC/2 1
0.9 e
C
0.8
0.7 choles
0.6
0.5
1
0.4 S n
0.3
0.2
0.1
0 = L
0 1 2 3 4 5 6 7 8 9 10 d
(a) (b)
FIGURE 7.12
A field of long (1 100) needles in a 1000 1000 square, generated from a Poisson point field
of density = 104 , i.e., there are 100 needles; (b) effect of increasing window scale on the
convergence of bounds (7.70) for soft needles C (i) = 104 ; the effective stiffness Choles 0.65 is
computed by a mean field method of Garboczi et al. (1991).
Note also that these problemsespecially in connection with finite-size

scalingcan be treated by the lattice (spring-network) method of Chapter 3
bearing in mind two approximations: needles have a finite thickness, and the
contrast is finite although it can be made very close to zero.
In Figure 7.13 we display the scaling of Ce and Sn 1 for needle systems
at contrasts 102 and 104 at two mesoscales: = 10 and = 50. Of principal
interest here is the same type of slow approach to the RVE as that in soft disk-
matrix systems noted earlier in Figure 7.10. Now, in the studies of effective
moduli of heterogeneous materials, the resulting Ceff is typically presented
versus the volume fraction x of one of the phases. For our system of short
needles, this is shown in terms of Choles against x = nL 2e f f almost all the way
to the percolation point at 5.9; Choles is computed by the physicists mean
field method (Garboczi et al., 1991) for needle-shaped holes of any aspect ratio
and with arbitrarily strong interactions.
To sum up, this figure displays (1) a very slow approach of Ce and
n 1
S to the RVE (i.e., curve Choles ), and (2) a discrepancy between Choles and
the Dirichlet as well as the Neumann bounds. Note that Choles corresponds to
an effective (macroscopic) response of a very large random system, which is
typically computed under periodic boundary conditions.
For the system at hand, there is an interesting result concerning the
statistics of second invariants of Ce () and Sn (). Namely, within a few
percent, their coefficients of variation are constant for a given type of
boundary conditions (either Dirichlet or Neumann), are independent of the
changing mesoscale , and independent of the contrast in the material,
P1: Binod
May 28, 2007 18:8 C4174 C4174C007
(a)
trC/2 1
0.9 (C e) = 10
0.8
e
0.7 (C ) = 50
0.6
0.5
0.4
0.3
n
0.2 (S )1 = 50 choles
n 1
0.1 (S ) = 10
0 x
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
(b)
FIGURE 7.13
(a) A 1000 1000 window of 2000 randomly placed 10 1 needles with an isotropic distribution;
a subwindow of size ( = 50) is indicated. (b) Normalized overall moduli Ce and Sn 1 , at
= 10 and = 50, and the effective stiffness Choles for a random field of short (1 10) needles
(such as that of Figure 7.12), as functions of the volume fraction x. Data were computed only at
discrete intervals x = 1.21, 2.42, 3.63, and 4.84.
except, of course the singular and trivial case of C (m) /C (i) = 1. We return
to this property, also with reference to other microstructures, in Section 8.2 of
Chapter 8.
7.5.4.2 Cross-Correlations of the Mesoscale Moduli

with the Crack Density Tensor
The aforementioned studies of effective moduli (Ceff ) of materials with cracks
of elliptical (or ellipsoidal) pores rely on a crack density tensor (Kachanov, 1993),
which may also be called a fabric tensor,
2
ij = (a ni n j + b 2 mi m j ) (k) . (7.91)
A k
P1: Binod
May 28, 2007 18:8 C4174 C4174C007
m
b a
n
(a) (b)
FIGURE 7.14
(a) Randomly located ellipses, with overlap permitted: (b) basis for the fabric tensor.
Here a and b are the semi-axes of an ellipse, Figure 7.14. We focus on the
correlation of the fabric tensor ij with the mesoscale response
tensors
C
e
and
S . This is done via the cross-covariance function K ij Ckl = ij Ckl ij Ckl ,
n
where Ckl stands for either Ce or Sn ; we suppress the parameter for simplicity
of notation. It is more convenient to use the correlation coefficient

ij Ckl ij Ckl
ij Ckl = . (7.92)
ij Ckl
Although our computations, given the vast extent of the parameter space,
have so far been restricted, some interesting trends could be observed. For
example, for isotropic systems of moderately soft needles (contrast 0.1) of
aspect ratio either 10 or 20, at window sizes either = 5 or 10, the following
fabric-property cross-correlations hold:

11 C11e > 0 12 C12e < 0 22 C22e > 0

ij Ckle = 11 C e > 0 12 C12e < 0 22 C12e > 0
12 ,
11 C22e > 0 12 C22e < 0 22 C22e > 0
(7.93)

11 S11n < 0 12 S12n > 0 22 S22n < 0

ij Skln = 11 S12n < 0 12 S12n > 0 22 S12n < 0.

11 S22n < 0 12 S22n > 0 22 S22n < 0
This implies that for finite windows the correlation of ij with Ckle tends to be
opposite in sign to that of ij with Skln
ij Ckle = ij Skln . (7.94)

P1: Binod
May 28, 2007 18:8 C4174 C4174C007
Now, Ce and (Sn ) 1 must converge to the macroscopic, deterministic Ceff =

lim Ce , so that the only way that (7.94) may hold is for ij to become
uncorrelated with Ce and Sn in the infinite scale limitthat is, ij Ckle and ij Skln
should tend to 0. This implies that the second-order moments of geometry
of a material with cracks, such as the conventional fabric tensor ij , may
be insufficient in describing macroscopic structure-property relations. This
in turn indicates that one should consider higher moments of ij and other
geometric measures of the crack network connectivity.
This conclusion is supported by a study of the cross-correlation structure
of systems with anisotropy: a 1 = 1 in the Fourier series angular distribution
of the long axis of the needles, and higher a i s equal zero, while all the other
parameters are the same as before. This leads to fabric-property cross-
correlations with these inequality signs:

11 C11e < 0 12 C12e > 0 22 C22e < 0

ij Ckle = 11 C e < 0 12 C12e > 0 22 C12e < 0
12 ,
11 C22e > 0 12 C22e < 0 22 C22e > 0
(7.95)

11 S11n > 0 12 S12n < 0 22 S22n > 0

ij Skln = 11 Sn > 0 12 Sn < 0 22 S12n > 0
12 12 .
11 S22n < 0 12 S22n > 0 22 S22n < 0
Note that, although the inequality signs in (7.95) are not the same as those
in (7.93), they satisfy (7.94).
7.6 Moduli of the Trabecular Bone

The trabecular bone has a multiscale, random 3D structure, Figure 7.15.
Evidently, there are at least five length scales present. Focusing on the meso-
scale of that structure, Wang and Jasiuk (2006) have recently considered a
model of a trabecular bone as having a 2D periodic prismatic structure and
they predicted computationally the apparent elastic moduli of such a model.
They studied a periodic structure in order to avoid the complexity of ac-
counting for the actual random structure of trabecular bone (by using a
micro-CT based finite elements) and to avoid the need to obtain many re-
alizations to obtain the ensemble average of response. In order to explore
the effect of scale and boundary conditions, they varied the size of the win-
dow from one unit cell to larger window sizes (multiple unit cells) and
P1: Binod
May 28, 2007 18:8 C4174 C4174C007
Trabecular bone
Macroscale
Cortical
bone
X 100
Mesoscale
(a) Trabecular
network
X 1,000
Single
Microscale
trabeculae
(b)
X 10,000
Submicroscale
Single
(c) lamella
X 100,000
Collagen fibers
Nanoscale apatite crystals
(d)
FIGURE 7.15
Hierarchical structure of trabecular bone, showing (a) mesostructure (0.510 cm) of trabec-
ular network; (b) microstructure (10500 m) single trabeculae; (c) submicrostructure; and
(d) nanoscale. (After Jasiuk, 2005.)
applied four types of boundary conditions: uniform displacement, traction,

mixed and periodic. The results calculated using periodic boundary condi-
tions give effective response while the remaining three boundary conditions
give apparent moduli. The apparent moduli calculated using the uniform
displacement boundary conditions bound effective moduli from above while
the moduli obtained using the uniform traction boundary condition bound
the effective moduli from below. The larger was the window, the closer were
the bounds.
P1: Binod
May 28, 2007 18:8 C4174 C4174C007
In the analysis the trabecular bone is modeled as a two-phase material

consisting of bone tissue (hard phase) forming a trabecular network and soft
tissue (soft phase) present in pores. Both phases are assumed linear elastic
and isotropic. Bone tissues are assigned Youngs modulus of 13.0 GPa and
Poissons ratio of 0.3 while Youngs modulus of soft tissue is given as 1.3 kPa
(giving moduli mismatch of 106) with the same Poissons ratio of 0.3. The
bone tissues volume fraction is estimated at 20%.
P1: Binod
May 28, 2007 18:8 C4174 C4174C007
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
8
Random Field Models and Stochastic
Finite Elements
Il ny a pas de probabilite en soi; il ny a que des modeles

probabilistes.
G. Matheron, 1978
Chapter 7 began with a discussion of the separation of scales, and then fo-
cused on the problem of material properties below the RVE. With reference
to equation (7.1) there, our concern was with the left part of that inequality:
d < L or d L. The issue we focus on in this chapter is the right part of
(7.1): L L ma cr o . That is, if the RVE on scale L is not justified, only a ran-
dom continuum can be usedrecall Figure 1(b) of the Preface. As a result, we
need to establish some methods to deal with solution of macroscopic bound-
ary value problems having the mesoscale SVE as input. Such problems are
necessarily stochastic, and this leads us to a formulation of random fields of
material properties from the SVE information, and their input into numerical
methods leading then to so-called stochastic finite element (SFE) and stochastic
finite difference methods.
8.1 Mesoscale Random Fields

8.1.1 From Discrete to Continuum Random Fields
As discussed earlier, a random medium is a set of deterministic media: B =
{B(); }. Suppose we deal with the antiplane elasticity of a matrix-
inclusion composite (B() = Bm Bi ) with locally isotropic phases of proper-
ties C (m) (matrix) and C (i) (inclusion), respectively. The most complete
description of this two-phase microstructure is given in terms of an indicator
(or characteristic) function

1 i f x Bm
m (, x) = or m : R2 {0, 1}. (8.1)
0 i f x Bi
Probability in itself does not exist; there are only probabilistic models.
273
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
Ci
Ce
1
Sn

Cm
FIGURE 8.1
The setup of random fields: from a piecewise-constant realization of a composite to two approx-
imating continua at a finite mesoscale.
Of course, the indicator function of C ( Cii /2) gives the local property at any
point
C(, x) = m (, x)C (m) + [1 m (, x)] C (i) . (8.2)
The indicator function of either phase is a random field with discrete-valued

realizations with a continuous parameter x: {m (, x); , x R2 }. Material
properties are given in terms of a random field
C : R2 {IC (i) , IC (m) } or {C(, x); , x R2 }, (8.3)
Clearly, this random field is also discrete-valued with a continuous parameter,

as illustrated in Figure 8.1 in terms of one realization along x.
In view of the considerations leading to the derivation of the -dependent
hierarchy of bounds (7.70) in Chapter 7, at any point x in the material and
at any mesoscale , two estimates of effective properties may be introduced:
Ce and Cn (Sn ) 1 . Consequently, at any mesoscale there are two approx-
imating tensor-valued random fields
Ce : R2 R3 Sn : R2 R3 . (8.4)
These two mesoscale random fields are continuous-valued with continuous

parameter x R2 . The random composite material B is now described, in
an approximate way, by two sets of realizations: {Ce (, x); , x R2 }
and {Sn (, x); , x R2 }, see Figure 8.1. These approximations, as they
directly depend on the choice of , provide two alternate inputs to the field
equation governing the global response on the smoothing mesoscale
[Ci j (, x)T, j ]i = 0. (8.5)
Let us also recall from Chapter 7 that, besides Cd and St we can define the
mesoscale response via uniform orthogonal-mixed boundary conditions. As
a result, we thus have three different approximating random fields
B d = {B(); }, B t = {B(); }, B dt = {B(); }, (8.6)

P1: Binod
May 28, 2007 18:9 C4174 C4174C008
Random Field Models and Stochastic Finite Elements 275
with Cd and St tensors bounding Cdt

. If we generalize this line of thought to a
more general microstructure, we return to Figure 1 of the Preface and conclude
that there is no unique way of setting up a random field with continuous
realizations in (b).
8.1.2 Scale Dependence via Beta Distribution

Assume the statistics of m to be homogeneous and isotropic. Recalling our
definition of the mesoscale parameter , we see that the equation (8.2) is a
pointwise limit 0, described by a probability density

p[C(x)] = f (m) C(x) C (m) + f (i) C(x) C (i) . (8.7)
The Dirac deltas on the right are weighted by the volume fractions f (m) = m
and f (i) = 1 m of phases m and i, respectively.
Suppose now that we sample the local properties not in this pointwise limit
but, rather, with some finite mesoscale 1 , that is smaller than the inclusion
size: 0 < 1 < 1. Figure 8.2 shows that if we take such a finite size window, it
can fall into either of two phases, or on the boundary of inclusions. The former
possibility of the pointwise limit, considered in the preceding paragraph,
corresponds to a Lebesgue measure zero and thus we have simply had the
equation above. This discrete distribution is now replaced by a continuous one
such as shown by the curve p1 . Note that the probability mass is distributed
continuously between C (m) and C (i) , but not outside this finite range. As
grows, we see a redistribution, or flow, of the probability mass away from
the end points of the interval [C (m) , C (i) ] toward some region indicated by
the curve p2 . When finally , p[C(x)] tends to the causal distribution
centered at C eff C the graph p .
These considerations indicate that of all the classical probability densities,
beta is the most convenient one to describe this scale effect while keeping all
admissible values within a finite range. It is given by
C a 1 (1 C) b1
p C, a , b, C (m) , C (i) = f or C (m) < C < C (i) , (8.8)
[C (i) C (m) ]B(a , b)
where
(a + b)
B[a , b] = for C (m) < C < C (i) , (8.9)
(a )(b)
with being the gamma function.
In Figure 8.3 we assess its statistical character by displaying probability
e
densities of tr (Ce ) (a) and C12,max (b) at = 10 under essential boundary
condition. The first of these plots confirms the beta character suggested in
Figure 8.2, while the second one indicates that the radius of the Mohr circle is
strongly positively skewed. When comparing Figure 8.3 with corresponding
plots for Sn , we observed that densities of traces are similar; this conclusion
does not carry over to skewnesses of Ce and Sn (Ostoja-Starzewski, 1998).
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
p(C ) P2 P P1
Cm C e Ci C i
FIGURE 8.2
Sampling of the mesoscale property (trace of apparent tensor C ) of a disk-matrix composite
via windows of different sizes. The beta distribution gives a practical approximation for the
entire range of window sizes, showing four cases: the pointwise limit of equation (8.7); the scale
1 and fit p1 ; the scale 2 and fit p2 ; and the scale and the causal distribution p . (After
Ostoja-Starzewski, 1998.)
8.1.3 Mesoscopic Continuum Physics Due to Muschik

The line of studies by Blenk et al. (1991), Papenfu and Muschik (1998), and
Muschik et al. (2000) introduces the concept of a statistical element as a meso-
scopic distribution function (MDF) f (m, x, t) generated by the different values
of the mesoscopic variables in a volume element
f (m, x, t) f (.) (.) (m, x, t) M R D Rt . (8.10)
The MDF is defined on the mesoscopic space MR D Rt describing

the distri-
bution of m in a volume element at (x, t), normalized by f (m, x, t) dM = 1
as it should. The macroscopic properties are obtained via averaging with
respect to the distribution f (m, x, t).
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
(a)
(b)
FIGURE 8.3
Two planar random microstructures: (a) four-phase PoissonVoronoi mosaics; (b) superposition
of a matrix-disk composite with a matrix-needle composite.
Thus, when trying to integrate Muschiks mesoscopic continuum physics

approach with ours, we first recognize M to be analogous to the random field
on mesoscale L. In terms of an elastic microstructure problem, f (m, x, t) is
analogous to a probability distribution of the mesoscale stiffness. On one
hand, this analogy necessarily forces one to choose L in mesoscopic contin-
uum physics, and on the other, it shows that the mesoscale continuum ther-
modynamics approximation is nonunique for we have three possible choices
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
stemming from three uniform boundary conditions: either displacement, trac-

tion, or orthogonal-mixed. This appears to be a serious dilemma and per-
haps no single choice is universally good for all the applications, although
Muschiks model works well for liquid crystals, for which it was developed.
8.2 Second-Order Properties of Mesoscale Random Fields

8.2.1 Governing Equations
First, consider the equilibrium equation of antiplane statics
j, j = f, (8.11)
where f i represents the negative of the body force b i per unit volume. Take
both fields as superpositions of their averages and zero-mean random fluc-
tuations

j (x) = j + j
(x) f (x) = f + f
(x), (8.12)
where f i represents the negative of the body force b i per unit volume. Mul-
tiplying j, j at x1 by j, j at x2 , and then ensemble averaging, yields (recall
Problem 13 of Chapter 2)
j
2 Ci (x1 , x2 ) 2 i
(x1 ) j
(x2 )
= = f
(x1 ) f
(x2 ) = F (x1 , x2 ). (8.13)
x1i x2 j x1i x2 j
Here we introduced Ci as the correlation function of the stress field

, and
j
F (x1 , x2 ) as the correlation function of the body force field f

.
The same approach may be used in the general case of statics
governed by
i j, j = f i , (8.14)
to obtain the relation
2 Ciklj (x1 , x2 )
= Fik (x1 , x2 ), (8.15)
x1 j x2l
where Ciklj = i
j (x1 )kl
(x2 ) and Fik (x1 , x2 ) = f i
(x1 ) f k
(x2 ).
One can now proceed to statics of a micropolar body with random force-
stress, couple-stress, and body force fields (but zero body moment fields)

ji (x) = ji + ji
(x) ji (x) = ji +
ji (x) f (x) = f + f
(x).
(8.16)
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
Recalling from Chapter 6 the governing equations (6.12) and (6.17), we first
derive
2 Tjilk (x1 , x2 )
= Fik (x1 , x2 ), (8.17)
x1 j x2l
where Tjilk = ji

(x1 ) lk
(x2 ) and Fik = f i
(x1 ) f k
(x2 ) is the same as above.
Next, from the angular momentum balance we obtain
qp
2 Mji (x1 , x2 )
= e iab e pcd Tabcd (x1 , x2 ), (8.18)
x1 j x2k
qp
where Mji =
ji (x1 ) q
p (x2 ) is the correlation function of the couple-stress
field
. Differentiating (8.18) with respect to xa at x1 , and with respect to
xc at x2 , we find
qp
4 Mji (x1 , x2 )
= e iab e pcd Fbd (x1 , x2 ). (8.19)
x1 j x2k x1a x2c
Other relations of that kind remain to be developed.
8.2.2 Universal Properties of Mesoscale Bounds

An interesting statistical property of radius R of Ci j has been found to hold
for four types of microstructures (Ostoja-Starzewski, 2000):
1. Matrix-needle composites with stiff needles

2. Multiphase PoissonVoronoi mosaics, Figure 8.4(a),
3. Matrix-disk composites with circular or elliptical disks,
4. Superpositions of the latter with matrix-disk composites, Figure 8.4(b).
In essence, we deal here with random two-dimensional microstructures

of spatially homogeneous, isotropic, and ergodic statistics, that are generated
from planar Poisson point fields. Thus, besides the PoissonVoronoi mosaics,
we are considering here Boolean models (recall Chapter 1); hard-core point
processes are excluded, and so we do see partial overlaps of inclusions in
Figure 8.4(b). In the first case, the material of each Voronoi cell is sampled at
random from either two, three, or four types of phases; p = 1, . . ., 4, depend-
ing on the actual choice of a p-phase random microstructure. The sampling is
done sequentially, independent of the states of other cells of the mosaic. An
example of a mosaic with four phases present is shown in Figure 8.4(a).
In the case of Boolean models, we generate inclusions sampled at random
from any one of two (or three, or four) types of phases; the matrix is phase
p = 1, and inclusions are or p = 2, 3, or 4. Also here, the sampling is done
sequentiallyone inclusion after anotherindependent of the states of other
cells of the composite. Each phase is locally homogeneous and isotropic, and
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
1 1
0.5 0.5
0 0
0.5 0.5
1 1
0 0 0 0
5 5 5 10 10 5
10 15 15 10 15 15
20 20 20 20
(a) (d)
1
1
0.5 0.5
0 0
0.5 0.5
1 1
0 5 0 0 0
10 10 5 5 10 10 5
15 20 20 15 15 20 15
20
(b) (e)
1
1
0.5 0.5
0 0
0.5 0.5
1 1
0 0 0 0
5 10 15 10 5 5 5
20 20 15 10 15
20 20
15 10
(c)
(f )
FIGURE 8.4
Graphs of the correlation coefficient i jkl (r) Ci j Ckl (r) of the components of Ce : (a) 1111 (r); (b)
1212 (r); (c) 1112 (r); (d) 1122 (r); (e) 1212 (r) under uniform strain; (f) 1111 (r) under uniform strain.
r1 (respectively, r2 ) axis goes to the right (left).
it is characterized by its volume fraction f ( p) and conductivity C ( p) . Thus, the

contrast for a phase p = 1 is ( p) = C ( p) /C (1) .
For each of the mesoscale second-rank tensorsconductivity Ce (essen-
tial) and resistivity Sn (natural), respectivelyfor any specific configuration
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
B() of B we can compute the second invariants

Re () = (C11 C22 ) 2 /4 + C12
2
, Rn () = (S11 S22 ) 2 /4 + S12
2
. (8.20)
Thus, in the ensemble sense, for any scale and any type of boundary
conditions (e or n), we have two random invariants: {Re (); } or {Rn ();
}. We next consider the coefficient of variation of each of these invariants
( Re ) ( Rn )
CV e = , CV n = . (8.21)
( Re ) ( Rn )
In the above, stands for the ensemble average and for the standard devi-
ation of the given invariant.
We have carried out a range of numerical experiments on microstruc-
tures of Voronoi mosaic and Boolean type to determine CV e and CV n , usually
employing a very fine spring network (Chapter 3) for the resolution of the
microstructure and solution of both types of boundary value problems. It has
turned out that, whatever the point in the parameter space, the coefficients
of variation of both invariants (i.e., CV e and CV n ), at any > 1, equal about
0.55 0.1 irrespective of:
1. The window size .
2. The boundary conditions applied to the window (uniform Dirichlet
or uniform Neumann).
3. The contrasts ( p) ( p = 2, . . ., 4), and the shape of the inclusion.
4. The volume fraction f ( p) of any phase p = 1, . . ., 4, providing its
conductivity is not 0 or . This result indicates a universal nature
of CV e and CV n for planar random media generated from Poisson
point patterns.
The fluctuations of up to 0.1 around 0.55 appear to be due to the finite

number of realizations of the random microstructure (generated by a Monte
Carlo method) in any given parameter case.
An exact mathematical analysis and proof of the constancy of these coef-
ficients of variation does not appear possible at the present stage of theories
of random media. However, we offer some observations that may prove vital
to such a proof in the future:
The Poisson point process does not possess any intrinsic length
scale, which fact seems consistent with CV e and CV n being inde-
pendent of the window size .
If our microstructures are generated from hard-core point processes
(i.e., non-Poisson point fields), then CV e and CV n are usually lower
than 0.55 for window sizes on the order of several grains ( 5), and
then rise and stabilize around 0.55 at higher (Jiang et al., 2001a).
Although there are no explicit formulas for the conductivity or
resistivity tensors for heterogeneous domains of finite size, we can
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
argue that these tensors are continuous in three parameters: win-

dow size , contrast ( p) , and volume fraction f ( p) of either phase.
The extreme cases of ( p) = 0 or need to be excluded in order to
avoid discontinuous dependence at percolation points.
The second invariant of the conductivity tensor Ce (as well as the
resistivity tensor Sn ) of the material possessing isotropic statistics
goes to zero as the window size . Thus, the mean ( Re ) and
standard deviation ( Re ) of this invariant also go to zero as .
In view of (8.21), the constancy of CV e (and CV n ) with implies that
the mean and standard deviation remain in the same ratio as they
both go to zero.
Finally, the third invariant C(3) also appears to have similar universal prop-
erties. In fact, observing that, in planar problems (i, j = 1, 2),
C(3) Ci j C jk Cki = C(1) [3C(2) C(1)

2
]/2, (8.22)
C(3) should decrease with due to a decrease of C(1) with . In effect, (8.22)
allows determination of any statistic of C(3) from those of C(1) and C(2) .
8.2.3 Correlation Structure of Mesoscale Random Fields

For a composite having SSS statistics of its propertiessuch as its indicator
function (8.1)the mesoscale random field C is stationary in the same sense
too, that is,
Ci j Ckl (x1 , x2 ) = Ci j Ckl (r) r = x1 x2 . (8.23)
A natural question arises here: For a composite having also an isotropic statis-
tics of its properties, is the C field isotropic in terms of its correlation function?
This isotropy property is expressed by
Ci j Ckl (r) = Ci j Ckl (|r|), (8.24)
and is, of course, different from the isotropy of its realizations. In Ostoja-
Starzewski (1993a,b, 1994) we addressed this issue through Monte Carlo
computations of antiplane mechanics of two systems: binomial fields on
square lattices and disk-matrix composites. For each and every realization
B() B of the given random material, the study necessarily involved find-
ing Ci j and Ckl for two windows: one placed at the origin (r = 0) and another
at some arbitrary position r = (r1 , r2 ). Resulting plots for the Ce tensor, at
= 10, for the binomial field at a nominal 50% volume fraction of either
phase, at contrast 10, are shown in Figure 8.5; surface fluctuations are due to
a finite number of B()s. The computations were discontinued for windows
separated by several disks because then the apparent moduli became
independent. This observation shows that any long-range correlation function
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
FIGURE 8.5
A two-phase material with a Voronoi mosaic microgeometry of a total 104,858 black and white
cells, at volume fraction 50% each.
modelsfor example of exponential typeneed to be substantiated by a long-

range microstructural memory such as, say, reinforcing rods.
Other observations with regard to Figure 8.5 are:
1. The autocovariance C11 C11 is not isotropic as there is a stronger

correlation in x1 than in x2 .
2. The autocovariance C12 C12 is isotropic.
3. The crosscovariance C11 C12 is practically zero.
4. The crosscovariance C11 C22 attains, at the origin, the maximum value
of 0.75 rather than 1.0 as might intuitively be expected.
5. Practically identical plots of Ci j Ckl are obtained under the assump-
tion of uniform strain, as evidenced by Figure 8.4 (e) and (f).
6. Practically identical plots of Ci j Ckl are obtained for the Sn tensor,
with uniform strain in (5) being replaced by uniform stress.
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
8.3 Does There Exist a Locally Isotropic, Smooth Elastic Material?

8.3.1 Correlation Theory Viewpoint
The need to include lower than the RVE length scale phenomena in elasticity,
combined with their nondeterministic characteras discussed at the begin-
ning of this chapterhas motivated a number of researchers to adopt random
stiffness fields, assuming invertibility of the constitutive law
= S(, x): = C(, x): S(, x) = C1 (, x), (8.25)

with and being uniform fields applied to a rather hypothetical and
unspecified RVE (or SVE) of a random medium. When dealing with fourth-
rank tensors of elasticity, typically a locally isotropic form
1
= [(1 + (, x))i j (, x)i j kk ] x B RD (8.26)
E(, x)
is adopted by simply postulating E and (often just E) to be random fields

usually of Gaussian typewith differentiable realizations.
To simplify
the setting somewhat, let us consider random fields of second-
rank tensors Ci j (, x); ; x B , such as those of Section 2.3 in Chapter
2. Assuming the isotropy of realizations, we have
Ci j (x) = C (x) i j or C (x) = C (x) I . (8.27)
However, in view of the developments of Chapter 7, a smooth stiffness tensor

field C (x) is really a mesoscale continuum approximation
Ci j (x, ) = C (x) i j or C (x, ) = C (x, ) I , (8.28)
with expressing the ratio of L to d (heterogeneity size) of the underlying

microstructure, while i j stands for the Kronecker delta. Thus, we should have
1212
1
= 2323
1
= 0, (8.29)
with the obvious symmetries (2.121) of Chapter 2 present, so that (2.123) of

that chapter implies
K 5 = K 6 = 0. (8.30)
Clearly, the 2D setting of antiplane elasticity is a special case of this, and we

come to conclude that, in particular, 1212 (x) = 0; it must be because C12 = 0
everywhere.
At this point we refer to our computational mechanics studies in
Section 8.2.3 above. We see that i jkl is strongly dependent on the particular
pair [Ci j , Ckl ] as well as on the direction x. Here, we make correlation-type
(C) observations:
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
C1: As expected, 1111 transforms into 2222 upon the rotation of x = (x1 , 0)
into x
= (0, x2 ).
C2: Against expectation, 1212 = 0, which indeed can be understood
from physical considerations alone, especially at x = 0, without
any recourse to numerics. This provides an argument via reductio
ad absurdum against the admissibility of (8.27) and, by extension of
the same arguments to higher-rank tensors, of (8.26).
C3: Only 1112 and 2212 turn out to be null.
8.3.2 Micromechanics Viewpoint

From the standpoint of the Hill condition, we have three loadings to choose
from when setting up a random field with continuous realizations, and this
leads to nonuniqueness of response on mesoscales. Now, in place of the dis-
ordered microstructure of Figure 8.1(a) of the Preface, let us consider a spe-
cial case of a smooth elastic continuum (8.27) without any microstructure.
Assume the dependence of C (x) on x1 -direction only. Henceforth, by C(i j)
we denote the components of the mesoscale stiffness C .
We make analysis-type (A) observations:
A1: The mesoscale response C(11) of C of the L L window is calculated
exactly under the assumption of a uniform stress 1 (x) = 1 , x
B L , because for this loading we have a smooth microstructure of
a series-type.
A2: The mesoscale response C(22) of the L L window is calculated
exactly under the assumption of a uniform strain 2 (x) = 2 , x
B L , because for this loading we have a smooth microstructure of
a parallel type.
A3: Loadings dictated by A1 and A2 jointly correspond to the special
case of the orthogonal-mixed boundary condition. Then, assuming
any smooth function C (x) in (8.27), C(i j) on the left of equation
(8.11) can be evaluated via integration of C (x) over the mesoscale
domain B L . This would involve a calculation of compliance S(11)
and of stiffness C(22) , which, given the fact that the axes x1 and
x are oriented along the principal directions, would allow one to
determine all the elements of C(i j) . Next, one can take the limit
lim C(i j) (x) = C (x) i j , (8.31)

0
to recover the original smooth continuum.

Suppose now that a heterogeneous material microstructure with statistics
depending on x is introduced. We then need a mesoscale continuum at > 0,
and, in view of A13 above, Cmix (x, ) must be determined. That tensor is
anisotropic! At this point we recall (7.27) of Chapter 7:
[St ()]1 Cdt

() C (),
d
(8.32)
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
We now make micromechanics-type (M) observations:
M1: Each one of the three tensors in (8.32) is anisotropic.

M2: As the mesoscale , the tensors Cd (, x1 ), Cdt (, x1 ) and
[Sn (, x1 )]1 converge to Ceff , but this is far above the scale of all
the fluctuations, that is, beyond Figure 1(b) of the Preface. In princi-
ple, this is assured by the separation of scales limit (assured in turn
by the ergodic and SSS properties of the microstructure) where the
homogeneous continuum appliesa situation of no interest to us
because fluctuations of Figure 1(b) arise below that limit.
1
M3: The ensemble average tensors CD (x1 ), Cmix (x1 ) and S (x1 )
N
are
orthotropic; a computed example was given in a study of function-
ally graded materials in Section 7.5.3 of Chapter 7.
8.3.3 Closure
The observations C13, A13 and M13 of Sections 8.3.1 and 8.3.2 show that a
locally isotropic, smooth elastic continuum is untenable, unless one is willing
to disregard the underlying microstructure. Here we may quote Truesdell and
Noll (1965): Continuum mechanics presumes nothing regarding the struc-
ture of matter. That is, continuum theories are not only phenomenological,
but some may well have no physical content. This can happen with determin-
istic as well as stochastic models. Evidently, generalizing the notion of a uni-
form isotropic elastic continuum to an inhomogeneous, smooth and isotropic
one does not appear to violate any principles of continuous media. However,
when the micromechanics is brought into the analysis of antiplane elastic
responseindeed, one of the simplest models in continuum mechanicswe
arrive at contradictions above, and these extend to higher-rank tensors and/or
inelastic responses.
8.4 Stochastic Finite Elements for Elastic Media

8.4.1 Bounds on Global Response
The mesoscale random fields introduced in Sections 8.1 and 8.2 provide
inputs to two finite element schemes, based on minimum potential and com-
plementary energy principles, respectively, for bounding the global response
(Ostoja-Starzewski, 1993). As an example of such an approach, a (scalar)
problem of torsion of a bar made of a two-phase microstructure has been
analyzed (Ostoja-Starzewski and Wang, 1999; Ostoja-Starzewski, 1999b).
Assuming a deterministic body force f , its governing field equation is
[Ci j (, x), j ],i + f = 0, x B() = 0, x B(). (8.33)

P1: Binod
May 28, 2007 18:9 C4174 C4174C008
Here is a stress function and C(, x) ( Ci j (, x)) corresponds to one par-

ticular realization B() (of volume V) of the random medium B, so that the
problem is entirely deterministic; recall (8.5). The lower and upper bounds
on global response of B() are obtained, respectively, from two, energy prin-
ciples: the minimum potential energy principle

1 1
inf T C ()d V f d V, (8.34)
H01 (V) 2 V 2 V
and the minimum complementary energy principle,

1
inf T S () d V H = { (L(V) 2 | + f = 0}. (8.35)
H01 (V) 2 V
Here C () and S () are the stiffness and compliance tensor fields on meso-
scale (set by the size L of the finite element relative to the grain size d), while
and stand for and C (), respectively. We thus have two algebraic
problems
[K()] {} = { f } , (8.36)
where [K()] is the global stiffness matrix, and
[L()] {} = {} , (8.37)
where [L()] is the global flexibility matrix. Here and are the respective
vector solutions; see the first reference above for all the details.
The essence of this setup is that these two energy principles ensure a
monotonic convergence of the lower and upper bounds of the energy norm
from below and above, respectively, in terms of the energy norm

1
E = T C()d V . (8.38)
2 V
provided (1) we have a homogeneous material and (2) the mesh resolution
0 (e.g., Brezzi and Fortin, 1991). This is the classical limit of infinite-
simal finite elements solving a deterministic continuum problem without
identifying any microstructure.
The situation, however, is not that straightforward in the case of a hetero-
geneous material. Namely, because the effective stiffness tensor on mesoscales
is nonunique, and (8.34) is set up in displacements, C () determined from
the displacement boundary condition, should enter this principle; the global
stiffness matrix in (8.36) is then built, element by element, from C ()s. On the
other hand, given the fact that (8.35) is set up in stresses, the apparent compli-
ances S () from natural boundary condition should be used as input to the
minimum complementary energy formulation; the global flexibility matrix is
then built, also element by element, from S ()s.
The tendency of global FE methods to converge with decreasingas
pointed out following (8.38)is now hindered by the fact that the mesoscale
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
(i.e. apparent) responses C () and S () tend to diverge as decreases. Evi-

dently, we have a competition of two opposing scaling trends. Nevertheless,
both FE methods provide bounds on the global, macroscopic response for a
given choice of , and, in fact, a possibility of an optimal mesoscale emerges.
To sum up, we are dealing here with the situation lacking separation of scales
alluded to in the Preface.
8.4.2 Example: Torsion of a Duplex-Steel Bar

The ideas discussed above are illustrated in the context of two-dimensional,
two-phase microstructures of linear elastic materials governed locally by
(8.33). The microstructural geometry is specified by a PoissonVoronoi
mosaic. Each cell of the mosaic is being occupied by either phase 1 or 2,
according to a probability equal to the global volume fraction, which is chosen
at 50% in Figure 8.5. Because the Voronoi cells are six-connected on aver-
age, we have a percolating systemstrictly speaking, bipercolatingwhich
clearly lacks any periodicity, and so, a periodic unit cell cannot be setup.
The isotropy of both phases stiffness tensors leads to a contrast =
C (2) /C (1) . When = 1 the material is homogeneous, otherwise it is heteroge-
neous. Note that the two-phase PoissonVoronoi mosaic microstructure cho-
sen here may be applied to model a range of different materialsexamples
are offered by duplex steels for a finite (e.g., Werner et al., 1994), or porous
materials for the extreme cases of = 0 or .
In Figure 8.6(a) we show the case of a homogeneous material ( = 1): as
discussed in connection with equation (8.38), the finer the meshthat is, the
smaller the mesoscalethe closer are both estimates of the global response.
However, in the case of a heterogeneous material ( = 1), there exists an
opposing trend according to the hierarchy of mesoscale bounds, Figure 8.7.
Thus we observe a competition of two opposing trends:
1. The global responses, computed by (8.36) and (8.37), tend to con-

verge as decreases.
2. The mesoscale responses, serving as input to (1), computed respec-
tively from the essential and natural boundary conditions, tend to
diverge as decreases.
The results of this competition are shown in terms of the energy norm
(8.38), in function of the increasing finite element size, for three contrasts
= 10, 100, 1000 in Figures 8.6(b), (c) and (d), respectively. Note that in
the case (b) of relatively weak contrast ( = 10) an optimal finite element
mesh size, or mesoscale opt , can clearly be seen-it gives the closest upper
and lower bounds. As the contrast increases-cases (c) and (d)the bounds
diverge further away from one another and only the crudest meshing of the
entire domain provides a relatively useful estimate of the global response.
Our methodology employing mesoscale finite elements is checked by a
comparison to the response of the same material computed by a finite element
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
8000
2.15 a=1 a = 10 Complementary energy
formulation
2.05 7000
Complementary energy
E
E
formulation
||||
||||
Computational micromechanics solution
1.95
Energy formulation
6000
1.85
Energy formulation
1.75 5000
20 40 60 20 40 60
Number of elements per side Number of elements per side
(a) (b)
5000 5500
a = 103 Complementary energy a = 102 Complementary energy
4000 formulation 4500 formulation
3000 3500
E
E
||||
||||
2000 2500
Computational micromechanics solution
Computational
1000 micromechanics solution 1500
Energy formulation Energy formulation
0 500
0 20 40 60 0 20 40 60
Number of elements per side Number of elements per side
(c) (d)
FIGURE 8.6
Behavior of the energy norm (8.38) with respect to a sequence of self-accommodating finite
element meshes, in terms of the increasing finite element resolution, for: (a) a homogeneous
material domain contrast = 1, and (b) a heterogeneous domain of Figure 8.5 for contrast = 10,
(c) the same domain for = 100, and (d) the same domain for = 1000. In (ad) computa-
tional micromechanics solutions taking account of the entire microstructure of Figure 8.5 are
also shown.
mesh much finer than any single grainit presents an absolute and best avail-
able, albeit very costly, reference solution. Thus, in all four cases of Figure 8.6
this computational micromechanics solution directly taking into account
the entire microstructure of 104,858 black and white cells is also shown. It is
seen that the micromechanics solutions always fall between the bounds based
on the micromechanics moduli.
The mesoscale window, or the SVE, is identified as a mesoscale finite ele-
ment of the global finite element mesh. With the demonstration of the method
for a single realization B() of random microstructure, it is a rather simple
matter to generalize it to an ensemble response. Thus, instead of computing
the mesoscale moduli Cd and St from Dirichlet and Neumann boundary value
problems, one could generate them, by and for any specific , based on the
statistics such as those presented in Chapter 7 and in the preceding sections.
The rest of the procedure would then be identical.
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
2.2
2.1
1.9
Eective moduli
1.8 C e
1.7
1.6 1
Snd
1.5
1.4
0 10 20 30 40 50 60 70 80

FIGURE 8.7
1
Hierarchies of bounds Ce () and Sn () for the two-phase microstructure of Figure 8.5 at
volume fraction and contrast = 10; five data sets are shown.
Finally we must note that, so far, this formulation is suited to handle prob-
lems with Dirichlet-type boundary conditions on the macroscale. In order to
deal with Neumann-type (or, generally, mixed) conditions on the macroscale,
a mixed variational formulation is necessary.
8.4.3 An Overview of Phenomenological SFE Studies

The foregoing development is different from the conventional SFE, which
basically proceed as follows: (1) assume a random field of constitutive co-
efficients, (2) use it as input into the global FE scheme, usually based on
the minimum potential energy principle, and (3) derive the global response
either for the first two moments or for the ensemble in the Monte Carlo sense
(Benaroya and Rehak, 1988; Brenner, 1991; Ditlevsen, 1996; Liu et al., 1995;
Kaminski, 2002); see also Elishakoff and Yongjian (2003). In the following we
review basic tenets of the SFE.
With reference to the preface, given a deterministic field equation in
mechanics
Lu = f, (8.39)
randomness may enter through either the operator (i.e., material properties),
the forcing function (temporal in nature), or the boundary and/or initial con-
ditions. However, the choice of randomness of forcing f in time is fundamen-
tally different from the randomness of the field operator L in physical space.
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
The point is that f(, t) in (8.39) implicitly involves some local averaging in the
time domain

t+t/2
1
f t (, t) = f (, t
)dt
, (8.40)
t tt/2
which is needed to smear out fluctuations in, say, wind forcing on a structure,
over time scales too short to have any influence on the oscillator. Commonly
the subscript t in f on the left-hand side is suppressed, and we simply write
f (, t). On the other hand, the local averaging in physical space is not consis-
tent with the concepts of micromechanics, and, as shown in Chapter 7, should
be replaced by stochastic homogenization, which, by offering three optional
boundary conditions, leads to a nonuniqueness of continuum approximation.
Now, if local averaging is applied to a stiffness (respectively, compliance) ten-
sor field, it yields a Voigt-type (Reuss-type) estimate of stiffness (compliance)
for some spatial domain of the microstructure.
The need for development of finite element methods taking into account
the uncertainty in structural material parameters has been recognized since
the late 1970s (Contreras, 1980); one also needs to mention here stochastic
boundary element methods (Liu et al., 1995; Saigal, 1995). As a result of
the observation that many engineering structures are described by spatially
random material properties, several theoretical methodologies were devel-
oped in the civil engineering literature. Most of these studiesbroadly called
stochastic (probabilistic) finite elements (SFE)are based on a direct generaliza-
tion of Hookes law to random fields such as in (8.25) and (8.26).
It has to be noted, however, that the effort and merit in SFE has been
on the development of efficient numerical methods for solution of boundary
value problems, rather than on development of a connection to the material
microstructure. Moreover, most cases have been restricted to the case of weak
fluctuations in material properties, whereby the stiffness matrix is expressed
as the sum of the mean and noise

[K ()] = [K ] + K
() . (8.41)
Perturbation method. This approach consists in a replacement of the ran-

dom system by a (theoretically infinite) number of identical deterministic
systems each of which depends on the solution for the lower order equations.
Thus, to second order, for the static problem[K ()] {U} = { f }the solution
is expressed as the sum
{U} = {U0 } + {U1 } + 2 {U2 }, (8.42)
where 1. This leads to a system of equations
{U0 } = [K ]1 { f }

{U1 } = [K ]1 K
() {U0 } (8.43)

{U2 } = [K ]1 K
() {U1 } .
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
Neumann series method. This method (Shinozuka and Yamazaki, 1988),

is based on a Neumann series for the inverse of the random operator [K ()],
which takes the following form:
[K ()] = ( I P() + P 2 () P 3 () + ) [K ]1

P() = [K ]1 K
() . (8.44)
The approach was introduced as an avenue for a speedier way of solving the
stochastic problem by a Monte Carlo simulation. To that end also a Cholesky
decomposition of [K ()] is implemented.
Weighted integral method. In contradistinction to the above two methods,
this one focuses on the determination of the random stiffness matrix [K ()].
The idea, in the setting of an elastic plate problem (e.g., Deodatis and Graham,
1997), is to start with a locally isotropic random field of, say, Youngs modulus
and assign it to all the finite elements according to
i
i
E(, x) = i E + 1 +i f (, x) f = 0. (8.45)
Next, the stiffness of each element of i V is calculated as

i
K () = [i B]T [C()] [i B]dx = i K 0 +i X0 () i K 0 () , (8.46)
iV
where [i K 0 ] and [i K 0 ()] are deterministic matrices, and i X0 () is a random

variable given as

i
i
X0 () = f (, x)dx. (8.47)
i A
From a micromechanics standpoint, this approach gives a Voigt-type es-

timate for the effective stiffness of the ith finite element. Similarly, if applied
to the compliance, it would yield a Reuss-type estimate of flexibility.
Spectral method. It is well known that, in a representation of a random
function by a Fourier series, the coefficients of the expansion become, in
general, correlated. In order to retain the uncorrelatedness while obtaining
the desired orthogonality of random coefficients, a KarhunenLoeve expan-
sion (e.g., Papoulis, 1984; Yaglom, 1957) is introduced. This idea has been
employed by Ghanem and Spanos (1991) to represent the spatial variability
of random field of Youngs modulus such as in (8.45). This method claims not
to be limited to weak fluctuations and to avoid the inconsistencies between
various other methods involved in the inversion of the random stiffness ma-
trix [K ()]. Also, it is designed to do away with the problem of dealing with
a large number of random variates resulting from a pointwise representation
of the random field E(, x).
Conclusions. There is: (1) a necessity of a correct link to micromechanics in
setting up of the continuum random fields and of the random stiffness matrix
[K ()]; and (2) a need for a careful interpretation of the variational principles
as a basis for SFE.
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
Thus, although an assumption such as (8.25 and 8.26) is generally incorrect,

it follows that the classical SFE methodologies are amenable to modifications
to incorporate the micromechanical input of the type described above. More
precisely, once continuum random fields of a given material are found, an
existing approachfor example, a Neumann seriesmay be applied to de-
termine the upper and lower bounds on response according to the stochastic
variational formulation given in Section 8.4.1. For applied mathematics trends
in multiscale methods see e.g., E et al. (2005) and Gloria (2006b).
8.5 Method of Slip-Lines for Inhomogeneous Plastic Media

8.5.1 Finite Difference Spacing vis-a-vis Grain Size
The breakdown of separation of scalesrepresented by equation (7.1) of
Chapter 7can also be encountered in mechanics of plastic materials. Let
us now suppose that we deal with a specific boundary value problem of a
rigid perfectly plastic material of Hubervon Mises type by the method
characteristics. This means we want to determine the planar stress field
(11 , 22 , 12 ) from the equilibrium equations combined with the yield con-
dition (e.g., Hill, 1950; Kachanov, 1971)
11,1 + 12,2 = 0 22,2 + 12,2 = 0 (11 22 ) 2 + 412

2
= 4k 2 . (8.48)
As is well known, solution proceeds by setting up a system of two equations

that hold along two mutually orthogonal families of characteristics, and then
employs a finite difference method. If the separation of scales holds, k can be
taken as constant throughout the domain of material. This implies that the
crystal size in a material effectively (on the scale of RVE) described by (8.48)3
is infinitesimal relative to the spacing of the finite difference mesh.
However, if we have the situation depicted in Figure 8.8, we ought to
consider k in (8.48)3 as a smoothing, mesoscale random field k , in the (x1 , x2 )-
plane where is set by the ratio of the window size L to the average crystal size
d. Following Ostoja-Starzewski and Ilies (1996), the random plastic medium
B is now given by the set {B (); } = {k (, x); , x D} with

k (, x) = k + k
(, x) k (, x) = 0, (8.49)
where k
(, x) is zero-mean noise. To make things simple and tractable, we
assumejust like in phenomenological SFE studies of Section 8.3.3 abovea
high signal-to-noise ratio

k (, x) k . (8.50)

If we now introduce a deterministic, homogeneous plastic medium Bdet

defined by kdet = k , the basic question that arises is this: What is the
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
1 2
FIGURE 8.8
Mesoscale windows (or SVEs) involved in finding apparent plasticities k at 1, 2, and N. Evidently,
the separation of scales would be recovered with crystal size becoming infinitesimal relative to
the spacing of characteristics.
difference, if any, between the solution of Bdet and the ensemble average
solution of B ? We do know that in the limit both solutions should coincide,
but, to answer the question we first have to modify the derivation of charac-
teristics. As is usual in the theory of slip-lines (here we follow Szczepinski,
1979), two functions p and are now introduced
11 = p + k cos 2 22 = p k cos 2 12 = k sin 2. (8.51)
Upon substitution of (8.48) into (8.48)1,2 , and setting = p/4 on differenti-

ation, we get
p k p k
+ 2k = 2k = , (8.52)
x1 x1 x2 x2 x2 x1
where the rectangular axes are now along the local slip-line directions. Equa-
tions (8.52) become independent of the orientation of the axes if /x1 and
/x2 are replaced by the tangential derivatives /s1 and /s1 along the s1
and s2 characteristics, respectively. Thus, we find
k k
d p + 2k d = s1 d p 2k d = s2 . (8.53)
s2 s1
These relations represent a system of two quasi-linear hyperbolic equations

driven by the random terms involving k , both on the right- and left-hand
sides. Clearly, the randomness vanishes as k tends to a constant. The basic
setup is completed by corresponding characteristic directions which are, in
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
fact, specified by the same relations as in the homogeneous medium case

d x2 d x2
s1 : = tan( + /4) s2 : = tan( /4). (8.54)
d x1 d x1
Equations (8.53) and (8.54) form the basis for determination of the Hencky
Prandtl network of slip-lines in a given boundary value problem. In cases of
Cauchy and characteristic problems studied below this relies on the method
of finite-differences for finding {x1 , x2 , p, } N at a new point N given the data
{x1 , x2 , p, }i at the two preceding points i = 1, 2. It follows that, due to the
randomness in k at 1 and k at 2 and k N , as well as the possible randomness
in the initial data { p, }1 and { p, }2 two characteristics of the deterministic
problem are replaced here by two cones (or wedges) of forward dependence, which,
in the ensemble sense, contain all the characteristics of the stochastic problem
emanating from points 1 and 2, Figure 8.8.
The new values { p, } N are found explicitly by finite differencing from
(8.53), while {x1 , x2 } N are found either by backward or forward differencing
from (8.54). The latter choice itself follows either the scheme recommended
by Hill (1950),
yN yi = (xN xi ) tan [(i + N )/2 /4] i = 1, 2, (8.55)
or that by Szczepinski (1979),

1
yN yi = (xN xi ) [tan(i /4) + tan( N /4)] i = 1, 2. (8.56)
2
8.5.2 Sensitivity of Boundary Value Problems to Randomness

8.5.2.1 Cauchy Problem
The Cauchy boundary value problem we consider is defined as one in which
the normal stress and the shear stress are specified on a line AB, whereby AB
intersects only once each of the characteristics. Henceforth, we adopt AB to
lie along the x1 -axis over the interval 0.0 to 9.0 so that the normal stress is
22 and the shear stress is 12 , while the domain of influence lies in the first
quadrant of the (x1 , x2 )-plane, Figure 8.9. The particular case we give in that
figure involves an inhomogeneous boundary condition
x1
22 = 1.5 12 = 0.6 +, (8.57)
10
where we take n = 0, 1, . . ., 9 and increments x = 1.0.
The sensitivity of these boundary value problems to the mediums ran-
domness is studied through a comparison of three settings:
1. Deterministic case (medium Bdet )zero noise: k

= 0
2. Random casevery small noise of 0.5% about the mean: k

[0.0025, 0.0025]
3. Random casesmall noise of 5% about the mean: k
[0.025, 0.025]
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
(a) x
x
(b)
y
x
(c)
FIGURE 8.9
The Cauchy problem with boundary data (8.57) for: (a) deterministic homogeneous medium
case of zero noise; (b) random mediumcase of very small noise; (c) random mediumcase of
small noise.
Here k = 1 and k
is a uniform random variate in a given interval.
Forward dependence cones of HenckyPrandtl nets in the boundary value
problem (8.57) in all three cases are shown in Figure 8.9(a), (b) and (c), re-
spectively. The latter two were obtained from 100 simulated realizations of
B().
Studies of such Cauchy as well as characteristic (Goursat) problems lead
to these conclusions regarding the effect of noise k
in the yield limit k :
There is practically no difference between the ensemble average net

of slip-lines of the stochastic problem (i.e., for a random medium
B) and the net of the corresponding deterministic problem (i.e., for
a homogeneous medium Bdet ) for very small noise; however, this
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
difference increases with the growing inhomogeneity in the bound-

ary data.
The scatter about this average has a coefficient of variation which
tends to quickly increase over that of k on the length scales L ma cr o
of the boundary value problem; it increases with the growing inho-
mogeneity in the boundary data such as that modeled by (8.57).
For noise growing above 5% there is an increasing change in the
average, and very amplified scatter, in the random fields of stress
as well as velocity.
There is practically (to within four significant figures) no difference
in resulting slip-line nets between the explicit and implicit integra-
tion methods; here, the explicit formulas were used first as a pre-
dictor and then, without any numerical convergence problems, the
implicit formulas were employed as a corrector.
We have also made a comparison of the uniform versus truncated
Gaussian randomness of k
. As equivalence criterion between both
kinds of probability distributions we took the same standard devi-
ation for the case of very small noise (k = 14.437 105 ), and
the case of small noise (k = 144.37 105 ). With this setup all our
Gaussian-based simulations resulted in practically the same slip-
line networks as the ones shown here.
To sum up, we conclude that only in the case of very small noise
may one safely replace the average solution of a stochastic problem
by the solution of a deterministic problem with ke f f = kdet = k ,
that is Be f f = Bdet .
All these results were obtained under the assumption of independence of

random variables k at all the points of the finite difference nets. By introduc-
ing the correlatedness between neighboring windows, the scatter in slip-lines
tends to decrease, but only for < 10 (i.e., very small mesoscale windows) is
this a significant effect.
8.5.2.2 Limit Analysis of a Pipe under Internal Loading

As a further application of this method we consider limit analysis of a cylin-
drical tube loaded by a uniform traction on its internal boundary. The solution
to this classical problem (e.g., Kachanov, 1971) in the case of a homogeneous
material is based on the solution of a Cauchy problem for an axisymmetric
stress field (rr , , r ) in the (r, ) coordinate system in a plate with circular
hole, under a pressure boundary condition
rr = p < 0 r = 0 at r = a. (8.58)
With the yield condition rr = 2k the stress field is determined explicitly
by the formula
r
rr = p + 2k ln = r + 2k. (8.59)
a
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
The slip-line field is now given by a system of two orthogonal, logarithmic

spirals
r r
+ ln = ln = . (8.60)
a a
It follows from (8.59)1 that, in case of a tube of external radius b, the condition
rr = 0 at r = b gives the limit pressure
b
p = 2k ln . (8.61)
a
Having this formula we may regard p as given and ask for the external radius
b as a function of p . To make the example concrete, we set k = 1.5, p = 3.0

and a = 1, which then results in b( p ) = e p /2k = 2.723.
In the case of a tube made of an inhomogeneous material, however, the net
of slip-lines becomes distorted from such perfect pattern and the above nice
relations no longer apply. The finite differencing outlined above has now to
be used to determine the slip-lines as well as the stress field in any particular
realization of B. Now, the evolution of r along any slip-line emanating from
r = a is a certain random walk in the body domain D starting from the
value p there and stopping at 0 at r = b. Thus, the condition rr = 0
plays the key role in the definition of an excursion set of the random field
rr (r, ) = {rr (); } (e.g., Adler, 1981)
A0 (rr , D) = {(r, ) D|rr (r, ) 0}. (8.62)
This then leads to a set of level crossings
A0 (r , D) = {(r, ) D|rr (r, ) = 0} (8.63)
where A0 (rr , D) is a set of closed contours of the plastic zone, which in the
case of a homogeneous material Bdet under the pressure boundary condition
would be a circle of radius b( p ) = 2.723. Thus, in any given random material
the set of level crossings is a random set in plane, which, assuming spatial
homogeneity and isotropy of field {k (, r, ); , (r, ) D}, is a circle
containing all the possibilities.
Now, let b max and b min be a maximum and minimum distance for a given
realization, respectively, from the origin to the contour. Since b max determines
the minimal amount of material needed for a tube to withstand the internal
pressure p , we ask: Is b max ( p ) smaller, equal to, or larger than b( p ) of the
homogeneous medium problem?
In Figure 8.10(a) we plot patterns of slip-linesactually as wedges of for-
ward dependenceunder condition (8.67) corresponding to 400 realizations
of B with k = 1.5 and k
[0.025, 0.025]. The set of level crossings is
shown as a ring containing all 400 piecewise-constant, noncircular closed
curves. Clearly, b max ( p ) is always larger than b( p ), whereby b max increases
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
(a)
(b)
FIGURE 8.10
Slip-line patterns in a randomly inhomogeneous material with k
[0.025, 0.025], under:
(a) the boundary condition (8.58), and (b) the boundary condition (8.64). In each case 400
realizations of B() are used.
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
as the noise level in k increases. Thus, the principal conclusion is that the
presence of material inhomogeneities requires a thicker tube than what is pre-
dicted by the homogeneous medium theory. Let us also note that the higher
the pressure p, the greater is the external radius b, and hence, the greater is the
spread of the forward evolution cones implying an increase in the scatter of
b max and b min . Interestingly, both random variables b max and b min are symmet-
rically distributed about the deterministic radius b( p ) of the homogeneous
medium problem.
The same qualitative conclusions carry over to the case of a combined
pressure and shear boundary condition
rr = p < 0 r = q =
0 at r = a, (8.64)
The slip-lines are still spirals, albeit no longer of logarithmic type. The stress
field is now given by (Kachanov, 1971)

r 2 q r 2 q
+ +
rr = p k 2 ln
a k a k

q q
1 k
+ 1+ k

a r 4 q 2 q 2
+ 1 , (8.65)
r a k k
a 4
rr = k2 q 2 . (8.66)
r
The general character of spirals is displayed in Figure 8.10(b), which shows the
random slip-line network, also for 400 realizations, corresponding to q = 1.3,
and all other parameters the same as before. These data result in the external
radius b( p , q ) = 3.028. We note here that the addition of shear traction has
a strongly amplifying effect on the scatter of dependent field quantities, and
most notably on the spread of slip-lines.
Note: While this section focused on metal plasticity, similar concepts have
recently been developed for Mohr-Coulomb plasticity of granular materials
(Kamrin and Bazant, 2007). In it a stochastic flow rule has been introduced to
model fluidization (stick-slip) transition along non-deterministic slip-lines.
Note: Another related study focused on the plastic collapse (under a fixed
load) of an elastic/perfectly plastic medium with the yield strength taken as
Gaussian random field (Ku and Nordgren, 2001). Theorems of limit analysis
and methods of reliability theory were used to develop algorithms for the
computation of upper and lower bounds on the probability of plastic collapse.
Significantly, three-dimensional results were found to differ from those of the
plane problem.
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
8.6 Michell Trusses in the Presence of Random Microstructure

8.6.1 Truss-Like Continuum vis-a-vis Random Microstructure
We now return to the issue concerning the actual making of a Michell truss-like
continuum introduced in Section 4.4.1 of Chapter 4, but when the material
to be used for the manufacture has a random microstructure, such as in a
polycrystal plate. First, consider the situation in which we are going to use
the same realization B() to manufacture the trusses of sequence shown in
Figure 8.11, and continuing for ever higher discretizations (n increasing). As
illustrated in that figure, the finer the truss, the shorter and narrower are the
truss members. Associated with this there must be a growing dependence
of the yield limit k on the chosen mesh spacing L. Although we employ our
nondimensional parameter = L/d,where d is the typical size of microscale
imperfections (e.g., grain in a polycrystal), we keep in mind that now L may
strongly vary in space. For the truss problem introduced, L is smallest near
the rigid foundations boundary and largest at the point A of application of
force P.
From the standpoint of plasticity of random heterogeneous media, we
note two effects: (1) the scatter in k grows as decreases, and (2) the statistical
average k may change as decreases. Here we shall assume that the mean
of k is constant while its standard deviation, k , is inversely proportional to
k = const k () 1 . (8.67)
This suggests an impossibility to physically attain the limit of a truss-like

continuuma deterministic continuum with an infinitesimally fine spacing
of truss connectionsdue to competition of two opposing effects (Ostoja-
Starzewski, 2001b): (1) noise in k decreases to zero as the truss spacing L
grows; and (2) classical deterministic solutions tend to hold as the spacing L
decreases.
Similar to the plastic media problem of Section 8.5, k is a random field,
parametrized by location x = (x1 , x2 ) in the truss plane. Relations (8.67)
describe the apparent plastic limit, as a result of local smoothing of the ran-
dom microstructure (one realization B() of which is shown in Figure 8.11),
of a given truss member according to the given mesh spacing L = d. This
leads us again to the concept of a random medium: a set B = {B(); } =
{k(); }, where is a sample space of inhomogeneous continua locally
smoothed according to (8.67).
We can write
k() = k + k
() k
() = 0, (8.68)
where k
is the zero-mean noise in k. In the following we assume:
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
(a)
(b)
FIGURE 8.11
Manufacturing of a truss from one mosaic of a polycrystal using either a coarse (a) or a twice finer
(b) refinement of the mesh. It is understood that the material in the interior of squares is removed,
thus leaving an orthogonal grid of bars. Clearly, the scatter of the effective plastic limit of a bar
on the scale of a single cell (mesoscale) increases as we go from (a) to (b). And, simultaneously,
the thickness of the bars decreases with the mesh refinement. The dash-dot lines are the axes of
bars of the truss (i.e., characteristics of the hyperbolic system).
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
1. The truss spacings L of interest to us are greater than the grain size
d, so that k may be treated as a field of independent random vari-
ables when entering the finite difference formulation generalized to
inhomogeneous media (see below).
2. The underlying material microstructure is space-homogeneous and
ergodic.
8.6.2 Solution Via the Hyperbolic System

The governing equations need now to be modified from the homogeneous
medium case given in Section 4.4 of Chapter 4, following Hegemier and Prager
(1969). The modification consists in recognizing k not to be constant, so that,
with all the prior definitions, we obtain
w 2w + sin 2 k cos 2 k
k + k cos 2 + k sin 2 = +
x1 x1 x2 2 x1 2 x2
w cos 2 k 2w sin 2 k
k + k sin 2 k cos 2 = + . (8.69)
x2 x1 x2 2 x1 2 x2
Setting = /2, we obtain
w w k 1 k w 1 k w k
= + = . (8.70)
x x k x 2k y y y 2k x k y
If k = const, we recover the equations of a Michell truss made of a homoge-

neous material
d d
(w ) = 0 (w + ) = 0, (8.71)
ds1 ds2
which hold along two characteristics s1 and s2 , at angles specified, res-

pectively, by
= + /4 = + 3/4. (8.72)
is defined as the angle formed by the positive direction along the foundation
F with the positive x-direction. On this boundary w = 1.
If k is a random field, the characteristics directions are still given by (8.72)
and w = 1 on the foundation F , but the evolution of w and along the
characteristics is governed by (8.70). Now, if the derivatives /x and / y
are replaced by the tangential derivatives /s1 and /s2 along the s1 and s2
characteristics, equations (8.70) become independent of the orientation of the
axes, and result in
w 1 k w 1 k
dw d = dk ds1 dw d = dk ds2 . (8.73)
k 2k s2 k 2k s1
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
Summarizing, with k being a random function of x and y, we have a stochastic

quasi-linear hyperbolic system governing the field.
Given the presence of randomness, we have a stochastic inverse Cauchy
problem, a set of deterministic inverse Cauchy problems, each one corre-
sponding to one realization k() of random field k: find the net of char-
acteristics supporting the given load P at point A and emanating from the
foundation F with conditions (8.72) and (8.73) specified on it. A solution to
any one of these Cauchy problemsdenoted by a truss T()represents a
departure from that of the deterministic homogeneous medium Bdet . Because
all nets meet at A, we deal with an ensemble T = {T(); } of nets of
backward dependence wedges, which contains all T()s, Figure 8.12(a)(d). In
these figures we show the wedges in black and the solutions of Bdet in white;
the latter are repeated from Figure 4.17(a)(d) of Chapter 4 for a reference.
The net of characteristics for every realization k(), according to meshes based
on, respectively, 2n + 1 (n = 2, 3, 4 and 5) boundary points, was determined
from equations (8.72 and 8.73) by finite differencing.
We recall here that an analogous scatter of HenckyPrandtl nets of char-
acteristics, also due to the resolution by finite mesoscales, was observed in
plasticity problems in Section 8.4 above. Given our interest there in direct (not
inverse) Cauchy problems, we dealt accordingly with forward (rather than
backward) dependence wedges.
In Figure 8.12(d), where truss spacings on the foundation arc F are about
10% of the radius, we take the noise k
to be a zero-mean uniform random
variable having standard deviation 0.015; the mean of k is k = 1.0. All
noises k
away from F in Figure 8.12(d), as well as in Figures 8.12(ac), are
scaled according to (8.67)2 , that is, in inverse proportion to the local truss
spacing L. For the small noise-to-signal ratio adopted here, taking Gaussian
as opposed to uniform noise, while keeping the same variance, made hardly
any difference on the final results.
The efficiencies Eff of four random truss systems are given in Figure 8.13
they are defined by generalizing the concept used for the deterministic truss

Eff = V/ V . (8.74)
Here V() is the volume of T() from T , and V = V(n) (n finite) is the
ensemble average volume. We observe at once that values for Eff are lower
than those obtained under the deterministic, homogeneous medium assump-
tion involved in Section 4.4 of Chapter 4. Most importantly, however, a finer
random truss than that shown in Figure 8.12(d) is not possible (!) because the
noise is too strong for the net of characteristics to continue in a stable manner
up to the point Athe characteristics just tend to intersect prior to A. This
suggests that, should a larger number than 25 + 1 boundary nodes on F be
desired, a different truss topologyone of a disordered typerather than that
of an orthogonal net might be needed to solve the problem. This shows that
the random microstructure prevents attainment of the classically predicted
optimal shape using the homogeneous continuum. Additionally, the efficien-
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
P P
A A
F F
(a) E = 0.93 (b) E = 0.935
P P
A A
F F
(c) E = 0.942 (d) E = 0.95
FIGURE 8.12
Shown in black are successive sets of trusses, all governed by (8.54), according to meshes based
on, respectively, 2n + 1 (n = 2, 3, 4 and 5) boundary points on the rigid foundation F . Shown in
white are the trusses of Bdet , the same as those of Figure 4.17(ad) in Chapter 4.
cies in Figure 8.12 are lower than those of the corresponding Figure 4.17 in
Chapter 4, although that effect might be differentperhaps even opposite
depending on the choice of scaling law such as the one assumed in (8.67).
8.6.3 Solution via the Elliptic System

The elliptic system approach has already been outlined for a deterministic
material in Section 4.4.2 of Chapter 4. Recall that the shape optimization
problem is set up in the space of Ua d of admissible tensor fields of C(x). Any
such field is now taken as a particular realization C(, x) of the random body
{B(); }, and we look at optimization in the ensemble sense. Since even
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
(a)
(b)
FIGURE 8.13
(a) Michell truss problem: design space and boundary conditions. (b) Random distribution of
the Young modulus on the 27 27 mesh.
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
in the deterministic problem a computational solution needs to be employed,

we can only think of employing a Monte Carlo method in the stochastic case
at hand.
The formulation outlined above is also applied to the classical Michell
problem of finding an optimal structure within the unit square domain, sup-
ported by a circular foundation boundary F placed within and subjected to
a loading condition (force down) at A, Figure 8.13(a). This domain is parti-
tioned into a 27 27 mesh of square-shaped finite elements, each one with
a linear elastic and isotropic stiffness tensor C(, x), but of random magni-
tude; the Poisson ratio was = 0.3. That is, the stiffness tensor is taken as a
superposition of a mean C(, x) and of a zero-mean random fluctuating part

C (, x):

C(, x) = C + C (, x) C (, x) = 0, (8.75)

Here the overbar indicates the volume average mean. Tensor C (, x) is gener-
ated, without any dependence
on x (i.e., no spatial memory), by multiplying
each elements mean part C by a random number r sampled from a uniform
distribution [1, 1] and multiplied by a constant scale factor s

C () = rs C (8.76)
As an example, for s = 0.3, the Young modulus is varied from 0.7 to 1.3;
its sample realization is shown in Figure 8.13(b). The scale factor was intro-
duced to account approximately for smoothing and finite-size scaling of a
heterogeneous material microstructure with variability on a smaller scale.
This device, of course, is an oversimplification for we already know that the
mesoscale stiffness of a heterogeneous material should be derived from one
of three boundary value problems.
Figure 8.14 shows the results of the optimization problem at radius of
foundation equal to 0.1 of the square domain size. In (a) we display the
homogeneous medium case (a), and then consecutive realizations of the ran-
dom field with increasing scale factor. In particular, figures of the sequence
(b) through (h) correspond, respectively, to the multipliers s = 0.1, 0.15, 0.2,
0.25, 0.3, 0.6, 0.9. In all these figures immediately evident is the breakdown
of global symmetry of the truss structure due to the spatial nonuniformity of
materiala feature that could not be attained via the hyperbolic system. On
the other hand, we note that the CPU time involved in generation of any sin-
gle truss realization of Figure 8.12 is just a few seconds, as opposed to about
four hours on an identical computer for, say, Figure 8.14(b).
Our other principal conclusions (Liszka and Ostoja-Starzewski, 2003) are:
Our method assumes that the actual stochastic realization is known

a priori during the optimization process. This does not correspond
to a typical engineering process, but rather represents an organic
type optimization, like the growth of veins in leaves on a tree.
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
(a) (b)
(c) (d)
(e) (f )
(g) (h)
FIGURE 8.14
Topology optimization method applied to the mesh of square-shaped finite elements of Fig.
8.13(b) with uniform properties (a) and then increasing random properties: from (b) through (h).
In general, introduction of stochastic properties reduces the quality

of the results, but in some cases stochastic material properties may
improve the compliance of optimal design because the optimization
is performed on a given random distribution, so that the design
process has an opportunity to choose stiffer cells and discard those
with weaker material.
Our study is a preparation to the robust design itself. In particular,
we propose and plan this course of action:
In the course of an optimization procedure, the choice of a
goal function in terms of local variabilities will represent an
intermediate step.
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
As the particular function form is decisive with respect to the

choice of an optimal shape, the sensitivities need to be mod-
ified so as to account for a characteristic of spatial material
randomness and a statistical goal function.
Save and Prager (1985) said: Although a truss-like continuum is not prac-
tical, it uses the smallest possible volume V of structural material for the con-
sidered behavioral constraint and thus furnishes a useful basis for computing
the efficiencies of practical structures. Similarly, the present study uses the
Michell truss-like continuum to provide a better guidance on optimal struc-
tures in the presence of practically unavoidable microscale material random-
ness. One can argue that Michell-type truss-like continua with infinitesimal
spacing are rather theoretical concepts in conventional engineering structures,
and relatively few members can already ensure a high efficiency as indicated
by the trend of Eff in Figure 8.14. This, however, is not necessarily the case
with (very) small-scale systems, such as encountered in nanotechnology and
biostructures. It is here that the practically unavoidable microscale noise may
significantly alter predictions of conventional continuum mechanics in that
it may prevent the realizability of optimal, deterministic truss-like continua.
P1: Binod
May 28, 2007 18:9 C4174 C4174C008
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
9
Hierarchies of Mesoscale Bounds for
Nonlinear and Inelastic Microstructures
Law 0: Your friend is my friend

Law I: You cannot win
Law II: You cannot break even
Law III: You cannot even get out of the game
It should be clear from the derivation of mesoscale bounds for linear

elastic materials that three properties are required for their derivation: (1)
the statistical homogeneity and ergodicity; (2) the Hill condition leading
to admissible boundary conditions; and (3) an extremum (or variational)
principle. It follows that the mesoscale bounds can be generalized to vari-
ous other types of material behavior. More specifically, in this chapter we give
hierarchies of mesoscale bounds for nonlinear elasticity, elastoplasticity, rigid-
plasticity, elastic-brittle damage, viscoelasticity, flow in porous media and
thermoelasticity.
9.1 Physically Nonlinear Elastic Microstructures

9.1.1 General
Consider physically nonlinear elastic materials in the range of infinitesimal
strains, described by the constitutive law
w() w ( )
= () = = ( ) = , (9.1)

where the energy densities are related by w = : w; w is a statistically

homogeneous and ergodic field. As in Chapter 7, we assume the absence of
body and inertia forces. The Hill condition, and its implication for the type of
admissible boundary conditions, is

: d = : d (t n) (du d x)dS = 0, (9.2)
B
311
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
where = 0 and = 0 . This results in the same three types of uniform

boundary conditions on the mesoscale as (7.217.23) in the linear elastic case
of Chapter 7. For a given realization B () of the random medium B on some
mesoscale , the uniform kinematic boundary condition yields an apparent
constitutive law
= ( 0 ). (9.3)
Similarly, the uniform traction condition results in an apparent constitutive

law
wd ( 0 ) wt ( 0 )
= () = = ( 0 ) = . (9.4)
0 0
Next, we have the minimum potential energy principle

1 t u
() )dV
w( dS, (9.5)
2 B Bt
and the minimum complementary energy principle

1 t u
( ) )dV
w( dS. (9.6)
2 B Bu
Here tildes indicate admissible fields.

From this, using the same partition concept as in Chapter 7, the apparent
constitutive responses are shown to be related by
wd (0 , ) wd ( 0 , ) = /2, (9.7)
and
wt ( 0 , ) wt (0 , ) = /2. (9.8)
By passing to the ensemble, we get a hierarchy of bounds (Jiang et al., 2001b)

from above

d
w . . . wd wd . . . w1d = /2, (9.9)
and from below

t
w . . . wt wt . . . w1t = /2. (9.10)
Note: wt =
: 0 wd because the mesoscale response depends on the type
of loading.
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
Hierarchies of Mesoscale Bounds for Nonlinear and Inelastic Microstructures 313
9.1.2 Power-Law Materials

Consider a random medium of the specific constitutive law (Kroner, 1994)

N
w()
N
w ( )
= () = Cn : n = = ( ) = Sn : n = , (9.11)

n=1 n=1
While the above are in mechanical form, the energetic laws are

N
n
N
n
w() = Cn : w ( ) = Sn : . (9.12)
n=1
n+1 n=1
n+1
This leads to apparent responses under uniform displacement and traction

boundary conditions

N
wd ( 0 )
= Cdn () : ( 0 ) n =
0
n=1

N
wt ( 0 )
= Stn () : ( 0 ) n = . (9.13)
0
n=1
and, respectively, the energetic forms

N
n
N
n
w() = Cn : w ( ) = Sn : . (9.14)
n=1
n+1 n=1
n+1
The apparent constitutive laws for B () are next shown to be related by

a partition theorem (Hazanov, 1998, 1999)
(Stn ()) 1 Cn () Cdn (),

eff
(9.15)
and, in view of the statistical homogeneity and ergodicity of the material, we

have hierarchies of bounds from above

Cdn . . . Cdn Cdn . . . Cdn1 = /2, (9.16)
and from below

Stn . . . Stn Stn . . . Stn1 = /2. (9.17)
It is known that, under proportional monotonic loading, strain-hardening

elastoplastic composites may be treated in the framework of deformation the-
ory of plasticity, which is formally equivalent to physically nonlinear, small-
deformation elasticity, such as dealt with here. In (Jiang et al., 2001b) we have
assumed this equivalence to also hold for apparent elastoplastic response,
and have thus obtained energy bounds on random elastoplastic composites.
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
In that reference we have also proposed an approach via tangent moduli.

Finally, the effect of imperfect interfaces on the hierarchies of bounds has
been investigated by Hazanov (1999).
HashinShtrikman-type and more advanced bounds, without scaling laws,
for the effective energy function of more general type than power-law were
investigated by various authors, for example, Gibiansky and Torquato (1998),
Ponte Castaneda and Suquet (1998), Talbot (2000), and Torquato (2002).
9.1.3 Random Formation vis-a-vis Inelastic Response of Paper

The linear elastic response of paper from the standpoint of its random fiber
network structure was discussed in Chapter 4. Going to higher strains in
tensile loadingbe it uniaxial or biaxial in-plane loadingpaper becomes
inelastic without any clear elastic limit, or a yield point. (Compressive loading
has to be treated separately because of an obvious tendency to buckle.) All the
experiments show that papers inelasticity (i.e., plastic deformation) occurs
beyond the linear range as a plastic-hardening response, and the end of this
range is commonly taken as the yield point. Since here we are interested in a
papers response in that regime in loading only, we generalize the continuum
model given in Chapter 4 to a physically nonlinear elastic model (Castro and
Ostoja-Starzewski, 2003):

tanh (b11 )

11 Q11 Q12 0 tanh b 21
1 22
22 = Q12 Q22 0 12 , (9.18)
b
12 0 0 G tanh (b11 )
12
11
where, evaluating at zero strain ( = 0), we have
(1, 0)
d11 (0) E1
Q11 = = ,
d11 (1 12 21 )
(0, 1)
d11 (0) E 1 21
Q12 = = = Q21 , (9.19)
d22 (1 12 21 )
(0, 1)
d22 (0) E2
Q22 = = .
d22 (1 12 21 )
Two experiments on a typical specimen and their fitting by the above
formulas are shown in Figure 9.1. Thus, MD and CD indicate the resulting
responses under applied strains (11 , 22 ), equaling either (1, 0) or (0, 1).
Clearly, for very small strains (below 0.1%), we effectively have a linear
elastic model. Note that the MD(1,0) curve goes up to 1.5% strain, and then
drops due to yield, damage and failure occurring at smaller strains than those
for the CD(0,1). The thing to note is that, for the initial slopes, we have:
Q12 = Q21 .
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
50
45 Q11 tanh (b e)
40
MD (1,0)
35
Stress (MP a)
30
25 CD (0,1)
20
15 Q11
10 Q22 MD (0,1)
5 Q12 CD (1,0)
0
0.0% 0.5% 1.0% 1.5% 2.0% 2.5%
Strain e
FIGURE 9.1
Stress-strain curves obtained from testing
145-mm 145-mm specimens. Loading cases (1, 0)
and (0, 1) indicate the strains 11
0
, 22
0
applied on the biaxial tester.
Curves such as those shown in Figure 9.1 actually display scatter from
specimen to specimen, and so, it is of interest to address the effect of papers
formation on scaling of its mechanical properties. To model formation we
employ the Boolean model of Chapter 1, wherein d represents the papers fiber
floc size, which varies typically between 3 and 6 mm depending on grade and
furnish (i.e., a particular fiber suspension going into the paper manufacture).
The paper sheet is now discretized by a very fine mesh of rectangular finite
elements with material properties of equation (9.19) assigned to each element
proportionally to its basis weight (weight per unit area).
Arbitrary in-plane, biaxial stress-strain tests on such nonhomogeneous
material samples can now be run with the finite element model. In the fol-
lowing, we apply uniform displacement and traction boundary conditions to
realizations of the Boolean model having isotropic statistics and flocs of size
d = 5 mm, introduced in Figure 1.12 of Chapter 1. Thus, we work with two
square-shaped mesoscale domains 20 mm 20 mm and 50 mm 50 mm at
= 4 and = 10, respectively. We resolve the Boolean microstructure by a
fine mesh finite element model, where each finite element is 1 mm 1 mm,
which means that each B4 () and B10 ()sampled, respectively, from the en-
sembles B4 = {B4 (); } and B10 = {B10 (); }is modeled through
meshes of 20 20 and 50 50 elements, respectively.
Figure 9.2 shows mesh deformations of B10 () from the previous figure
under uniform displacement and traction boundary conditions (7.21) and
(7.22) in pure shear loading. In both cases the nonuniform deformations of
the mesh are clearly visible, whereby loading by the uniform displacement
condition results in a more uniform deformation than that under the uniform
traction. Of course, should the material be perfectly homogeneous, all the
mesh lines would be parallel and would deform affinely under both kinematic
and traction boundary conditions.
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
= 10
Traction b. c.
Number of
Layers
Displacement b. c.
FIGURE 9.2
Materials response under pure shear loading (uniform displacement or traction boundary con-
ditions) at mesoscale = 10.
This line of thought is pursued in Figure 9.3, which shows a comparison

of shear stress fields under both types of boundary conditions, and for both
mesoscales considered. The loading is in MD: uniaxial extension (11 0
= 0.5%),
and uniaxial tension (11 = 0.5%). Due to the material inhomogeneity, there is
0
a non-zero 12 field, and the difference between both stress fields is stronger at
= 4 than at 10. Indeed, we note a clear tendency at = 10 to form orthogonal
shear bands running at angles 45o to the horizontal. This suggests also that
the patterns of strain become very alike at this mesoscale, which occurs well
below the limit of the perfect RVE.
With numerical simulations carried out in the Monte Carlo mode involv-
ing ten (at = 10) and twenty (at = 4) realizations, we obtain ensemble
averaged shear stress-strain curves of Figure 9.4. They demonstrate scaling
trends of potential and complementary energies, as well as those on tangent
moduli for this model paper material. On that basis we establish that the RVE
in the sense of Hill is approximately reached on scales about ten times larger
than the floc size.
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
1.00 + 00
3.00E + 00
5.00E + 00
= 4 7.00E + 00
9.00E + 00
1.10E + 01
1.30E + 01
1.00 + 00
3.00E + 00
5.00E + 00
= 10 7.00E + 00
9.00E + 00
1.10E + 01
1.30E + 01
Displacement b.c. Traction b.c.
FIGURE 9.3
Comparison of shear stress fields according to applied uniform boundary conditions (either
displacement or traction) and the mesoscale , computed for two formation fields of Figure 1.12
of Chapter 1.
100,000
90,000 = 4 displ.
80,000 = 10 displ.
Average shear stress (Pa)
70,000
= 10 traction
60,000
50,000 = 4 traction
40,000
30,000
20,000
10,000
0
0 0.002 0.004 0.006 0.008 0.01
Average shear strain
FIGURE 9.4
Stress-strain curves of shear response resulting from uniform displacement (respectively,
traction) boundary conditions on mesoscales = 4 and 10.
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
The inelastic response of paper under both monotonic and cyclic uniax-
ial tension has also been studied via an elastic-plastic fiber network model,
effectively a generalization of a linear elastic model reported in Chapter 7
(Bronkhorst, 2003).
9.2 Finite Elasticity of Random Composites

9.2.1 Averaging Theorems
The key assumption of the finite hyperelasticity theory is the existence of
a strain energy function per unit volume of an undeformed body, which
depends on the deformation of the object and its material properties. Here we
restrict ourselves to the reference configuration, so that the equation of state
of the material takes the form:

Pij = , (9.20)
Fij
where Pij is the first PiolaKirchhoff stress tensor and Fij is the deformation
gradient tensor. With reference to Hill (1972), Nemat-Nasser (1999), Lohnert
and Wriggers (2003), and Costanzo et al. (2004, 2005), the average strain and
stress theorems of the infinitesimal strain case (Chapter 7) generalize to

1
Fij Fij (X)dV = Fij0 , (9.21)
V0 B

1
Pij Pij (X)dV = Pij0 . (9.22)
V0 B
Indeed,
outof several possible pairs for finite motions (Macvean, 1968), the
pair Pij , Fij is dictated by the Hill condition for finite motions
Pij Fij = Pij Fij . (9.23)
Thus, we now have three types of boundary conditions:
1. Uniform kinematic

ui (X) = Fij0 ij x j X B , (9.24)
where Fij0 is prescribed.

2. Uniform traction
ti (X) = Pij0 n j X B , (9.25)
where Pij0 is prescribed.

P1: Binod
May 28, 2007 18:23 C4174 C4174C009
3. Uniform orthogonal-mixed

ti (X) Pij0 n j ui (X) Fij0 ij x j X B , (9.26)
where the averaging theorems (9.219.22) have been used.
In the sequel, following Khisaeva (2006) and Khisaeva and Ostoja-Starzewski

(2006a,b), we use these conditions in setting up mesoscale bounds for random
microstructures in finite elasticity.
9.2.2 Variational Principles and Mesoscale Bounds

Consider the functional

(
u) = (
ui,k )dV ti0
ui dS, (9.27)
B Bt
where u is an admissible displacement field such that u = u on the portion

of the boundary B where displacement is prescribed, and ti0 is the specified
u
boundary traction on the remaining part of B . This is the finite elasticity

counterpart of the principle of minimum potential energy, in that the func-
tional (u) assumes a local minimum for the actual solution u if

2
di,k d p,q dV > 0 (9.28)
B ui,k u p,q
for all non-zero di such that di = 0 on Bu (Lee and Shield, 1980).

Under the uniform displacement boundary condition (9.24) the functional
(9.27) reduces to

(u) = ( ui,k )dV. (9.29)
B
At this point we again consider an alternative loading of the domain

B () B by restricted boundary conditionsin the same sense as (7.50)
vis-a-vis (7.49) of Chapter 7. Thus, if the inequality (9.28) holds, the energy
stored in the body under the restricted boundary condition (
r (F0 , )) is
related to that under the unrestricted one (
(F0 , )) as

(F0 , )
r (F0 , ). (9.30)
Upon passing to the ensemble, we get a hierarchy of bounds on the energy

density of the RVE (
(F0 )) from above

(F0 ) . . .
(F0 )
(F0 ) . . .
1 (F0 ) = /2. (9.31)
We now turn to the derivation of a reciprocal expression for the lower

bounds. It is well known that, in nonlinear elasticity, the strain-energy func-
tion can be nonconvex and therefore noninvertible, that is, strain cannot be
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
expressed in terms of stress. As a result, the principle of minimum comple-

mentary energy cannot be used without certain restrictions placed on the
strain energy function. In order to avoid these difficulties, the following func-
tional was proposed by Lee and Shield (1980):

u) =
Q(
uik dV nk
ui dS, (9.32)
B
uik Bt
uik
where
uik is a trial function, which satisfies the following conditions:

= 0 in B , nk = ti0 on Bt (9.33)
xk
ui,k
ui,k
It was shown that the functional Q is stationary for

uik = uik , where ui , the
actual solution of a problem, assumes a local minimum if

2
dik dpq dV > 0 (9.34)
B ui,k u p,q
for all non-zero dik satisfying the following conditions:

2 2
dpq = 0 in B , dpq nk = 0 on Bt . (9.35)
xk ui,k u p,q ui,k u p,q
Now, under the uniform traction boundary condition (9.25), the functional
Q reduces to

u) =
Q(
uik dV. (9.36)
B
uik
Thus, the following inequality between responses under restricted and unre-
stricted traction boundary conditions holds:

(P0 , )
r (P0 , ). (9.37)

where
(P0 , ) = B (uik /uik ) dV. From this, upon ensemble aver-
aging, we can derive a scale-dependent hierarchy of lower bounds on the
effective property Q (P0 )

(P0 ) . . .
(P0 )
(P0 ) . . .
1 (P0 ) = /2. (9.38)
The bounds (9.31) and (9.38) are illustrated on the example of a planar
matrix-inclusion composite in Figure 9.5, and the corresponding stress-strain
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
FIGURE 9.5
The finite element mesh of a composite in (a) undeformed and (b) deformed (traction boundary
conditions) configurations.
curves are shown in Figure 9.6. In particular, we model the matrix by a strain
energy function of Ogden form:

N
2i N
1 el 2i

(m) = 1 i + 2 i + 3 i 3 + J 1 (9.39)
i=1
i2 i=1
Di
with 1 = 4.095 105 Nm 1 = 1.3 2 = 0.03 105 Nm 2 = 5.0 D1 =

4.733 108 Nm, 3 = 0.01 105 Nm 3 = 2.0 and the inclusions by a
strain energy function of a neo-Hookean form
1 el 2

(i) = C10 I 1 3 + J 1 , (9.40)
D1
with C10 = 2.062 106 Nm D1 = 4.733 108 Nm.

The mismatch between both phases is set at
(i)
N
0
= 10 with (m)
0 = i (i)
0 = 2C 10 . (9.41)
(m)
0 i=1
It is clear that the approach to RVE is rapid, and it is attained at = 16 with

very good accuracy.
For other studies of homogenization in finite elasticity see e.g., Gloria
(2006a) and He et al. (2006).
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
5.E + 06
5.E + 06
4.E + 06
4.E + 06
3.E + 06
Hard phase
Soft phase
3.E + 06 KUBC = 16
KUBC = 8
KUBC = 4
2.E + 06 KUBC = 2
KUBC = 1
SUBC = 16
SUBC = 8
2.E + 06 SUBC = 4
SUBC = 2
SUBC = 1
1.E + 06 MIXED = 16
MIXED = 8
MIXED = 4
MIXED = 2
6.E + 05 MIXED = 1
Voigt bound
Reuss bound
0.E + 00
1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0
FIGURE 9.6
Stress-strain curves of a planar matrix-inclusion composite made of neo-Hookean inclusions
in an Ogden matrix with (i) 0 /0
(m)
= 10 under traction and kinematic boundary conditions at
= 1, . . ., 16; also shown are the individual responses of both phases.
9.3 Elastic-Plastic Microstructures

9.3.1 Variational Principles and Mesoscale Bounds
Let us consider a multiphase elastic-plastic-hardening material with per-
fect bonding between the phases p = 1, . . ., ptot . Each realization B () of B
is described by an associated flow rule (e.g., Hill, 1950)
dij f
dij = + d fp whenever fp = cp and d f 0
2G p ij
dij
dij = whenever fp < cp (9.42)
2G p
1 2 p
d = d everywhere (d = dii /3 d = dii /3).
2G p (1 + p )
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
Here G p (shear modulus), p (Poissons ratio), and c p (yield limit) form a

vector, each component of which (described by its indicator function p ) gives
rise to a scalar random field, such as

ptot
G (, x) = G p p (, x) . (9.43)
p=1
The entire body B = {B () ; } is described by a random vector field

= {G, , c}.
On the microscale we have tangent moduli, CT () or ST (), of the body
B (), which connect stress increments with strain increments applied to it
d = CT () : d d = ST () : d. (9.44)
Consequently, the Hill condition, and its implication for the type of admissible
boundary conditions, is

d :d = d : d (dt d n) (du d x)d S = 0, (9.45)
B
where d = d 0 and d = d 0 . Thus, we have
du(x) = d x x B (9.46)
and
dt(x) = d n x B (9.47)
Next, we recall two extremum principles (e.g., Hill, 1950): one for kine-
matically admissible fields

1 1
dS
dt d u d : d
dV dt dudS d : ddV,
Bt 2 B Bt 2 B
(9.48)
and another for statically admissible fields

1 1
d : ddV dt dud S d : d
dV dt dudS.
2 B Bu 2 B Bu
(9.49)
From these, noting that Bt = under kinematic boundary conditions and

Bu = under traction boundary conditions, we find
() C (),
CTd Td
() S (),
STt Tt
= /2. (9.50)
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
Upon ensemble averaging, and applying this to ever-larger windows ad

infinitum we get a hierarchy of bounds on the macroscopically effective
T 1
tangent modulus CT
= (S )
1 1 Tt 1
STt
1 . . . STt
S . . . C T

. . . CTd
C Td
. . . C Td
1
= /2, (9.51)
1
where STt1 and CTd
1 are recognized as the Sachs (1928) and Taylor (1938)
bounds, respectively. Also, see Suquet (1997) and Ponte Castaneda and
Suquet (1998) for comprehensive reviews of effective (RVE level) properties
of nonlinear composites.
For elastic-perfectly plastic materials, similar results have been obtained
by He (2001) using a mathematically more advanced analysis involving gauge
functions.
9.3.2 Matrix-Inclusion Composites

9.3.2.1 Case 1
Jiang et al. (2001) studied a composite with an elastic-hardening plastic ma-
trix phase and linear elastic inclusions. The latter were of the same modulus
as that of the matrix in the elastic range. The study confirms the bound-
ing character of responses computed under (9.46) and (9.47), respectively.
In particular, patterns of plastic shear bands under these two boundary
conditions are displayed in Figure 9.7, and the corresponding hierarchy of
bounds, computed here on two scales, is shown in Figure 9.8. Also note:
1. At a smaller scale ( = 6), the response under uniform kinematic
(9.46) is much more uniform than that under uniform stress bound-
ary conditions (9.47).
2. At a larger scale ( = 20), the discrepancy due to different types
of loading is much smaller, which fact shows a tendency to homo-
genize with mesoscale tending to .
3. The macroscopic response is quite well approximated by the uni-
form stress (but not the uniform strain) assumption. This situation
would be reversed for an elastic-plastic matrix with soft, rather than
hard, inclusions, and helps explain why Taylor bound works well
for polycrystals with dislocations.
An application of this type of approach to elastoviscoplastic composites of

Perzynas type has been explored by van der Sluis et al. (2000).
9.3.2.2 Case 2
Li and Ostoja-Starzewski (2006) considered an elastoplastic, matrix-inclusion
composite, whose stress-strain response is characterized by a power law
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
PEEQ VALUE PEEQ VALUE

+0.00E00 +0.00E00
+3.89E02 +1.17E02
+7.78E02 +2.35E02
+1.17E01 +3.52E02
+1.56E01 +4.69E02
+1.94E01 +5.87E02
+2.33E01 +7.04E02
+2.72E01 +8.21E02
+3.11E01 +9.39E02
+3.50E01 +1.06E01
+3.89E01 +1.17E01
+4.28E01 +1.29E01
+4.67E01 +1.41E01
+5.06E01 +1.53E01
PEEQ VALUE PEEQ VALUE

+0.00E00 +0.00E00
+5.18E02 +5.22E02
+1.04E01 +1.04E01
+1.55E01 +1.57E01
+2.07E01 +2.09E01
+2.59E01 +2.61E01
+3.11E01 +3.13E01
+3.63E01 +3.65E01
+4.14E01 +4.17E01
+4.66E01 +4.70E01
+5.18E01 +5.22E01
+5.70E01 +5.74E01
+6.22E01 +6.26E01
+6.73E01 +6.78E01
FIGURE 9.7
Two samples (realizations B ()) of a random matrix-inclusion composite at = 6 (left column,
top row) and = 20 (right column, top row); corresponding von Mises strain patterns shown
under traction (middle row) and displacement boundary conditions (bottom row). (After Jiang
et al. (2001b.)
(Dowling, 1993)

if 0 ,

0

= N (9.52)
0

else.
0
The material parameters are given in Table 9.1. The Hubervon Mises
yield criterion, with an associated flow rule, is assumed for each phase. Shear
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
1
f T
0.8 f
T
(Average stress)/0 f m
S m
m S
0.6
kinematic b.c., = 6
traction b.c., = 6
0.4 kinematic b.c., = 20
traction b.c., = 20
f ber
m matrix
0.2 T Taylor bound
S Sachs bound
mixed b.c., = 6
mixed b.c., = 20
0
0 1 2 3 4 5
(Average strain)/0
FIGURE 9.8
Stress-strain curves of a random matrix-inclusion composite, shown in Figure 9.12, with a linear
elastic inclusion (i) phase and a power-law hardening matrix (m) phase: ensemble averages of
eight sample windows at mesoscale = 6, and four sample windows at mesoscale = 20, under
displacement, mixed and traction boundary conditions. (After Jiang et al. (2001b.)
loading is applied through either a uniform kinematic or traction boundary

condition with 0 = 0.04 or 0 = 1.7 102 MPa.
Due to the heterogeneity of the material, stress distribution is nonuniform
under uniform loading, which results in the local stress reaching the yield
stress level somewhere in the material domain, even when the volume average
stress is (much) lower than the yield stress of material. Thus, it would not be
reasonable to define the yield condition of a sample as the stress level when the
first yield occurs in the specimen. We therefore adopt the model of Dvorak
and Bahei-Ei-Din (1987), who considered the overall yield of a sample to
indicate magnitudes of the overall stress , which causes the local volume
average stress to satisfy the yield condition in any phase. Thus, for Huber
von Mises materials, the mesoscale yield condition of the composite in the
space of overall stresses is defined as
F () = inf{ R33 | (x) , = , f p (K p ) = c p , x B , p = 1.2},

(9.53)
where K p is the mechanical stress concentration factor in the form of a 6 6

matrix, with treated as a (6 1) vector. The computational application of
TABLE 9.1
Material Properties of the Matrix-Inclusion Composite
Material Properties 0 0 [MPa] N E [GPa]
Soft phase 0.001036 75 0.25 72.4 0.33

Hard phase 0.001425 295 0.15 207 0.32
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
(9.53) involves an equality between the local stress of the soft phase and the
yield stress of that phase within a 3.5% error.
Patterns of the von Mises plastic strain, as well as those of the ensemble
average stress-strain responses, are similar to those of Case 1. Of particular
interest are the yield surface and flow rule on mesoscales. At this point, let us
introduce two different loading programs: displacement increment control and
traction increment control. In the traction increment control
T = n, (9.54)
where T is the increment of traction, and is the increment of volume

average stress. In the displacement increment control
U = E x, (9.55)
where U is the increment of displacement, E is the increment of volume

average strain. For a specific load (E 11 /E 22 = const or 11 /22 = const) in
both loading programs, we fix the ratio of increment (E 11 /E 22 = const
or 11 /22 = const) and continue to increase it until the volume aver-
age stress satisfies equation (9.53); we then get one specific yield point.
The mesoscale responses are collected in Figure 9.9, while the loading paths
are shown in Figure 9.10(a). Actually, for each sample we apply 17 different
loading paths corresponding to 17 different ratios of loading so as to obtain
17 yield points, whose ratios (11 /22 ) would vary from 1 to 1 (albeit only
approximately so for the displacement control). These 17 points cover quite
densely one quarter of the yield surface, which, by symmetry arguments, is
representative of the entire surface.
The loading paths are always linear for the traction increment control,
but not so for the displacement increment control because there are some
spots becoming plastic due to the local stress concentrations, even when the
volume average stress is still lower than the overall yield stress. (For a random
chessboard composite not shown here this nonlinearity is even stronger.)
Interestingly, the plastic strain rate is not always normal to the yield sur-
face and the shape of the yield surface is not perfectly elliptical. Figure 9.10(b)
shows the ensemble average yield surface on different mesoscales under two
different loading controls. Clearly, with the increasing mesoscale , the yield
surface bounds under displacement and traction boundary conditions be-
come tighter. We find a departure of the mesoscale flow rule from normality
under both uniform kinematic and traction boundary conditions. That depar-
ture is the strongest when the in-plane ensemble averaged principal stresses
are in the ratio of about 0.3 0.9. Given the limitation of available computers,
we cannot establish the expected trend to recover normality as the mesoscale
domain (i.e., SVE) grows and tends to the macroscale (RVE).
Our understanding of normality in plasticity follows the thermomechanics-
based argument of Ziegler (1983), who points out the much more fundamental
role played by the thermodynamic orthogonality in the space of velocities, E,
and notes that only when the dissipation function depends on velocities
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
Ensemble average stress-strain response for matrix-inclusion composite

108
3.5
2.5
Von mises stress (Pa)
1.5
Soft phase
1 Hard phase
Sachs bound
Taylor bound
Displacement B.C., = 6
0.5
Traction B.C., = 6
Traction B.C., = 24
0
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04
Average equivalent strain
FIGURE 9.9
Ensemble average stress-strain responses for different mesoscales under various boundary
conditions for the random matrix-inclusion composite. Also shown are the responses of both
constituent phases, as well as the Sachs and Taylor bounds.
alone in its arguments, the normality carries over to the space of dissipative
stresses. In the situation when depends also on other quantities, say, stresses
or internal variables, the normality gets violated. Now, in the case of a het-
erogeneous material, is also a function of the particular microstructure of
B (), which may roughly be represented by an internal variable . The latter
must be chosen so that, in the case of homogeneity, it becomes null and the
dependence of on vanishes. Perhaps, the simplest candidate for may be
the ratio of yield limits of both phases or the volume fraction of inclusions.
Another viewpoint on the loss of normality is offered by making a ref-
erence to the classical result of nonlinear homogenization where the exis-
tence of a plasticity potential at the micro-level implies the existence of a
macro-potential from which the effective constitutive equations are derived;
the macro-potential is the mean value of the local ones (Suquet, 1997), and
so, the normality is preserved by a scale transition. In our study, according to
(9.53), the macro-yielding takes place as soon as the local plastic flow begins
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
107
4
Stress 22(Pa)
Traction B.C.
Displacement B.C.
4 Matrix
Inclusion
Plastic strain rate
Plastic strain rate
0.0013 2.5013 5.0013 7.5013 10.0013
Stress 11(Pa) 107
(a)
FIGURE 9.10
(a) Loading paths and yield surfaces for one sample of a matrix-inclusion composite at = 6
(After Li and Ostoja-Starzowski, 2006).
for the first time at some point of the heterogeneous material. While such a
macro-yield criterion is not very useful in practical applications, in the case
of a more realistic (tolerant) yield criterion, the loss of normality would also
persist under scale transition.
9.3.3 Geodesic Properties of Shear-Band Patterns

It is clear from the preceding section that the shear bands display geomet-
ric shapes that conform to the actual spatial distribution of the material
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
108
3.5
2.5
1.5
Stress 22(Pa)
0.5
0.5
Traction B.C., = 6
1
Traction B.C., = 12
Matrix
Inclusion
1.5
0.5 1 1.5 2 2.5 3 3.5
Stress 11(Pa) 108
(b)
FIGURE 9.10
(b) A hierarchy of ensemble averaged mesoscale responsed (yield loci and flows) at = 6 and
= 12 (After Li and Osteja-Starzewski, 2006).
microstructure. With the latter being spatially irregular, the shear bands are
irregular too: they take paths of lowest plastic resistance while avoiding
the obstacles of high yield limit. Conceptually, therefore, they may well be-
have as geodesicscurves of shortest path joining two specific points in space;
see also Fermats principle, discussed in Chapter 11. In accordance with this
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
(a) (b)
FIGURE 9.11
(a) A matrix-inclusion composite studied in this paper; (b) shear bands (bright points) obtained
by computational micromechanics.
hypothesis, the geodesics should join the opposite faces of a specimen B (),
such as that in Figure 9.11(a), to which shear loading is applied. The basic
concepts of geodesics for mechanics of materials, especially in the context of
fracture mechanics of random media, were given by Jeulin (1993).
Taking the material model of Case 1, Figure 9.11(b) displays patterns
of equivalent plastic strain, found by computational micromechanics due
to uniform kinematic loading on B: 11 0
= 22
0
= 0, 12
0
= 21
0
.While in a
homogeneous material perfectly straight (horizontal and vertical) shear bands
would form, in the heterogeneous material of Figure 9.11(a) a distortion of
these bands, due to the presence of elastic inclusions, occurs. We postulate
that these distorted shear bands follow the shortest paths in the matrix, and
proceed to estimate, by geodesic propagations, shortest paths on the two-
dimensional configuration of the composite of domain B (). As potential
shear bands, we therefore consider two families of shortest paths obtained
by geodesic propagations in two orthogonal directions that avoid (black)
inclusions. For a horizontal propagation as source S and destination D we
first take the left and right faces, and then invert their roles, obtaining the
set G-hor. For a vertical propagation as source S and destination D we first
take the top and bottom faces, and then invert their roles, obtaining the set
G-ver. We consider propagations obtained on hexagonal lattices, so that we
deal with the so-called hexagonal geodesic distances.
Final results, following Jeulin and Ostoja-Starzewski (2000), are synthe-
sized as follows:
1. Addition of G-hor with G-ver to obtain the set G-add, Figure 9.12(a).
2. Supremum of G-hor with G-ver to obtain the set G-sup,
Figure 9.12(b).
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
(a) (b)
FIGURE 9.12
(a) Shortest paths (in dark) that avoid (black) inclusions obtained by (a) addition or (b) supremum
of horizontal with vertical geodesics. Note that the darkest (brightest) points here correspond to
the brightest (respectively, darkest) ones in (b) of the previous figure.
The grayscale at any given point in Figures 9.12(a) and (b) indicates the
(appropriately normalized) distance it takes to connect the opposite edges
of the square-shaped domain B (): the darker the point, the shorter is this
distance. Given two fields such as {g(i); i = 1, . . ., I } (geodesic) and {e(i);
i = 1, . . ., I } (equivalent strain), both defined on a square lattice of the same
size (256 256), so that I = 2562 , we compare them by using the normalized
covariance (or crosscorrelation coefficient) ge : equation (2.36) of Chapter 2.
We compute under the ergodic assumption (2.152)
I
I I
1
I i=1 g(i)e(i) 1
I i=1 g(i) 1
I i=1 e(i)
ge = ! .
2 " ! 2 "
1 I I I I
I i=1 g 2 (i) 1
I i=1 g(i) 1
I i=1 e 2 (i) 1
I i=1 e(i)
(9.56)
Basically, working with a single realization forces us to invoke the ergodic

assumption.
It turns out that the crosscorrelation of G-sup (Figure 9.12[b]) with the
true solution, e, obtained by finite elements (Figure 9.11[b]) is only 0.2,
and a lower number ( 0.1) is obtained for the G-add geodesic propagation
(Figure 9.12(a)). The advantage of this purely geometric method is an ex-
tremely fast computation of the pattern of plastic deformation as opposed to a
full computational mechanics approach. This may then offer a very rapid way
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
(a) (b)
FIGURE 9.13
The von Mises strain distribution obtained by a computational mechanics simulation (a) versus
the geodesic pattern (b).
of a purely geometric assessment of zones of plastic flow in disordered hetero-

geneous materials, although much more research needs to be done to assess
the appropriate ranges of parameters. One situation where the geodesics may
be very useful occurs in northern latitudes, where a highly heterogeneous ice
field is covering a large body of water. Given only an imagewith the gray
scale roughly indicative of the ice thickness, and hence of its strengthan ice
breaker wants to find the shortest path, yet one of least resistance, to go from
point A to B.
This methodology is also applied to the composite material Case 2 and
Figure 9.13 shows a typical result with ge = 0.85. Mapping out the ranges
of composite material parameters for which geodesics work as well as here
(or even better) and developing a solid theoretical basis is presently an open
problem.
9.4 Rigid-Perfectly Plastic Microstructures

9.4.1 Background
The random rigid-plastic material B = {B(); }, is defined by stating
that, for any grain (i.e., on microscale),

F ij , F ij ,
dij = , ij dij = ij 0. (9.57)
ij ij
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
Thus, in each grain the admissible stress states lies within the set
P1 () = { |F (, ) 0} { |eq Y }, (9.58)
where eq is the equivalent stress; Y is the yield stress.

Next, if on any mesoscale = L/d, we define volume averages

= B1 (x) dV d = B1 d (x) dV , (9.59)
then, on the macroscale, B (), in the notation of homogenization theory

(Suquet, 1993):
= () D = d () . (9.60)
For the RVE ( ), the yield (or extremal) surface of the composite B
delimits the set
P hom = { R33 | (x) with = , div = 0, F (, ) 0, x B},

(9.61)
and the elementary Taylor and Sachs bounds on P hom are expressed via a
hierarchy of inclusion relations
{|eq inf Y (x)}Sachs P hom {|eq Y }Taylor . (9.62)

xB
On any finite mesoscalea scale below RVEthere holds some form of

an apparent yield surface F ( , ), bounding some set
P () = { | (x) , div = 0, F (, ) 0, x B }. (9.63)
In contradistinction to P hom , P () depends on the configuration and the

mesoscale , and these are two issues we investigate in the following.
9.4.2 Bounding on Mesoscales via Kinematic

and Traction Boundary Conditions
Let us start by recalling the upper bound theorem allowing for discontinuities
in the velocity field, according to Kachanov (1971), but in a form more suited
for our purposes. First, we consider an arbitrary kinematically admissible
velocity field v of the body B . If ij is the stress field associated with vi by
(9.57)1 and also satisfying (9.57)2 , the basic energy balance equation can now
be written as

# #
ti vi dS = ij dij dV + Y #[v ]# dS, (9.64)
B B

S[v ]

where S[v ] = m=1
M
Sm[v ] is the set of internal surfaces of discontinuity in v .
Recalling the inequality (e.g., Hill, 1950)
ij dij ij dij , (9.65)

P1: Binod
May 28, 2007 18:23 C4174 C4174C009
we find

# #
ti vi dS ij dij dV + Y #[v ]# dS. (9.66)
B B S[v]
Upon subjecting the body B () on its entire boundary B = Bv to a uniform

kinematic boundary condition (dij is a constant and equals the volume average
of dij )
vi (x) = dij x j x B , (9.67)
noting (9.64), we write this upper-bound theorem as

# #
ij dij dV + |Y | [v] dS = ti vi dS
ij dij dV + |Y | #[v ]# dS.
B S[v] B B

S[v ]
(9.68)
This states that the upper bound on the actual external forces at which plastic
deformations begin may be found by assuming an arbitrary kinematically
admissible deformation mechanism of the body under consideration.
Next, as vi let us take a restricted uniform kinematic boundary condition
vi (x) = dij x j x s=1

S
Bs s = 1, . . ., S, (9.69)
which acts on all the boundaries of the partition. In the context of Figure 9.6(a),
(9.69) applies to the external square-shaped boundary B as well as to the
internal cross. Because the solution under the condition (9.69) on the partition
s=1
S
Bs is a kinematically admissible distribution under the condition (9.67),
but not vice versa, from (9.68) we have

# #
ij dij dV + |Y | [v] dS
ij dij dV + |Y | #[v ]# dS.

B S[v] s=1
S
B S[v ]
s
(9.70)
This says that the actual external forces at which plastic deformations
begin in B under (9.67) are bounded from above by forces at which plastic
deformations begin in the partition s=1
S
Bs under (9.69). If by Pv and Pvs we
denote domains in stress space bounded by volume average yield stresses
Y and Y corresponding to B and s=1
S
Bs , respectively, we can write
Pv () Pvs (). (9.71)
These two domains are bounded by the surfaces Fv ( , ) and Fvs ( , ),

respectively. We can next take ensemble averages of Y and Y so as to

$v and
define mesoscale yield surfaces Fv and Fv , respectively. Thus, if P
v v
$ denote two domains bounded, respectively, by F and F , we can
P v

write
$v P
P $v = /2. (9.72)

P1: Binod
May 28, 2007 18:23 C4174 C4174C009
Here we simply write in place of s , because the information about a par-

ticular part s of the partition is lost. By induction, we obtain a hierarchy of
inclusions
v
P $v P
... P $v . . . P
$v = /2. (9.73)
1
v
where P is the yield locus P hom of RVE, for which the statistical average may
be dropped. On the other hand, P $v is the domain bounded by the average
1
yield locus of a single grain, or the Taylor (1938) bound of (9.62), which results
from prescribing a uniform deformation rate field everywhere.
Although an analytical derivation of a lower hierarchy of bounds under
uniform traction boundary conditions is not available, our computational
results lead us to conjecture that this hierarchy of inclusions holds:
t
P ... $
Pt $
Pt . . . $
P1t = /2, (9.74)
where $
Pt and P$t denote a domain in stress space bounded by ensemble

averaged yield surfaces Ft and Ft .
9.4.3 Random Chessboard of Hubervon Mises Phases

We consider a two-phase (r = 1, 2) material with the microgeometry of a
random chessboard in two dimensions (x1 and x2 ), see Figures 9.6 and 9.7.
This material is a system of phase 1 (light) and phase 2 (dark) square-shaped
grains d d. More precisely, the microgeometry of phase 1 is taken as a
Bernoulli lattice process p,d ( p is a site probability) on a Cartesian lattice of
spacing d in R2
Ld = {x = (m1 d, m2 d)} m1 , m2 N. (9.75)
Any configuration B() of this piecewise-constant material is described by an

indicator function (, x) taking values 1 and 0 according as x falls in phase
r = 1 and outside (i.e., phase 2), respectively.
We take grains to be incompressible, rigid-plastic, isotropic of the Huber
von Mises type (HM), that is, having a yield condition

F (ij , ) = (2) kr2 () = 0, r = 1, 2. (9.76)

Here (2) is the second invariant of the stress deviator ij . The above shows
that randomness appears at the level of yield stress kr () of a homogeneous
grain; prime denotes a deviator. As is well known, in this case, the flow rule
becomes: dij = ij whenever F (ij , ) = 0 and d F 0. Given k1 and k2 of
both phases, we also introduce a mismatch (or contrast) = k2 /k1 .
We set and the nominal volume fraction of light and dark phases to
50%, so that we have 1 ,d . On finite scales ( < ) there are, of course, local
2
fluctuations from one B () to another, and this is reflected in Figures 9.14.
More specifically, we consider two domain sizes = 10 and 50. The response
of any given specimen B (), generated according to the Bernoulli lattice
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
Hard phase Soft phase
FIGURE 9.14
Two samples [realizations B ()] of a random chessboard composite at = 10 (left column, top
row) and = 50 (right column, top row); corresponding von Mises strain patterns shown under
traction (middle row) and kinematic boundary conditions (bottom row). (After Ostoja-Starzewski
(2005b.)
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
process, and subjected to in-plane loadings, is obtained by a computational

mechanics method described below. We study mechanical responses of 50
specimens at mesoscale = 10, and 20 specimens at = 50. These numbers
reflect the fact that the larger the scale, the smaller is the scatter, and hence,
the number of realizations required.
In plane stress the yield condition (9.76) now simplifies to
Fr ( ) = 11
2
11 22 + 22
2
+ 312
2
3kr2 , (9.77)
and the flow rule becomes

d11 d22 d12
= = . (9.78)
211 22 222 11 612
Graphics in Figure 9.14 display plastic responsesin terms of the von
Mises strainof the Hubervon Mises two-phase chessboards at contrast 5,
on mesoscales = 10 (a) and 50 (b). The loading is again of shear type: in
the left column for the kinematic boundary conditions, and in right column
for the traction boundary conditions. As before, in each case, bottom pictures
show the resulting plastic zones superposed onto the original mesh.
The inhomogeneity of strain fields, especially under the kinematic con-
dition, is evident. However, as the scale increases from 10 to 50, there is
a clear tendency for responses under both types of loading to attain spatial
uniformity and similarity.
It is interesting to compare the plastic responses of Figure 9.14 for the two-
phase 10 10 chessboard: there is a strong difference between the kinematic-
controlled responses, but not between the traction-controlled, as we switch
from the plane deformation rate to the plane stress. This comparison may
be continued for the two-phase 50 50 chessboard by considering apparent
plastic responses (always shown in terms of the von Mises strain). Also note
the formation of shear band features under the kinematic condition, which
was brought out at 0.01 of the maximum von Mises strain.
Figure 9.15 gives an ensemble view of apparent yield surfaces in the
(1 , 2 )-plane on mesoscales = 10 and 50, whereby, respectively, 50 and
20 realizations were employed. Hard and soft phase yield loci at contrast 5,
having the classical HM elliptical shape, are shown for reference. Also drawn
is the ensemble average kinematic-controlled yield locus and two most ex-
treme ones for scale = 50; their symmetry with respect to the origin of
(1 , 2 )-plane was noted. Two Hubervon Mises ellipses are drawn in broken
lines for reference so as to display the departure of these apparent (mesoscale)
responses from the HM response.
This departure is made more visible in Figure 9.16 representing a close-up
view of a part of Figure 9.15, focusing on the quarter of apparent yield surfaces
ranging from pure shear up to uniaxial (ex)tension, on mesoscales = 10 and
50. In general, we are led to a model that could describe the ensemble of yield
surfaces
F = {F (); } , (9.79)
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
6.0
2(10^8 Pa)
5.0
4.0
Hard phase
3.0
2.0 Average
1.0 Soft phase
0.0
6.0 5.0 4.0 3.0 2.0 1.0 0.0 1.0 2.0 3.0 4.0 5.0 6.0
1.0 1(10^8 Pa)
2.0
3.0
4.0
5.0
6.0
FIGURE 9.15
An ensemble view of apparent yield surfaces under kinematic conditions in plane stress on
mesoscales = 10 and 50. Hard and soft phase yield loci, as well as an average kinematic-
controlled locus and two most extreme ones on = 50, are shown.
where F is of an elliptical (finite thickness) shell shape. The shell converges to

an infinitesimal thickness in the limit, but bifurcates into two distinct
HM ellipses in the 1 limit. Figure 9.16 also displays the d vectors,
ensemble averaged for each one of four yield surfaces here (at = 10 and 50
under both kinematic and traction conditions). Noteworthy is the departure
of apparent (mesoscale) flow rule, for each case, from the normality.
Observations:
Neither in plane deformation rate, nor in plane stress, is the flow
rule normal for any given realization of the random medium.
In plane deformation rate, the flow rule is normal upon ensemble
averaging, but not so in plane stress.
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
6.0
5.0
4.0
Average 10 10 Traction b.c.
3.0 Average 50 50 Traction b.c.
Average 50 50 Kinematic b.c.
2.0
Average 10 10 Kinematic b.c.
1.0
0.0
0.0 1.0 2.0 3.0 4.0 5.0 6.0
1.0
2.0
FIGURE 9.16
A close-up from Figure 9.15 of the ensemble view of apparent yield surfaces in plane stress,
focusing on the first quarter in the space of principal stresses, on mesoscales = 10 and 50. The
d vectors, shown for ensemble average yield surfaces at = 10 and 50 under kinematic and
traction conditions, display the departure from normality. (After Ostoja-Starzewski (2005b.)
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
In plane deformation rate the yield surface, for any and of ran-
dom medium, consists of two parallel lines. In plane stress, the yield
surface is ellipse-shaped, but, for both boundary conditions (kine-
matic and traction), of a smaller aspect ratio than that of the HM
material. This suggests a convenient form of F ( ) for homogeneous
transversely isotropic materials (Mellor, 1982)
F ( ) = (1 + 2 ) 2 + (1 + 2R ) (1 2 ) 2 2 (1 + R ) Y
2
, (9.80)
where Y is the yield stress in uniaxial tension and R 0.7.

It is not yet clear what scale is necessary to approximately attain
the classical HM material for the RVE. All we can say is that much
larger domains need to be simulated, and these involve much more
computing power.
Scatter of yield loci is decreasing with mesoscale increasing, under
either type of boundary conditions, be it plane strain or plane stress.
9.5 Viscoelastic Microstructures

Here we give a very brief account of research carried out on heterogeneous
viscoelastic materials by Huet (1995, 1997, 1999a), First, the Hill condition
involves strain rates
: = : . (9.81)
Translated to the mesoscale, it implies that
: = : 0 or 0 : , (9.82)
depending on whether the strain rate (0 ) or stress ( 0 ) is prescribed.

On the microscale (i.e., locally) the material is governed by a formula
involving the relaxation modulus tensor (r)
t

(t) = r t t : d t dt , (9.83)
0
or a dual one involving the creep compliance tensor (f)

t

(t) = f t t : d t dt . (9.84)
0
On the mesoscale, under the kinematic boundary condition, the material

domain B () is governed by a formula involving the mesoscale relaxation
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
modulus tensor (r )
t
(t) = r t t : d 0 x, t dt , (9.85)
0
and, under the traction boundary condition, a similar one involving the
mesoscale creep compliance tensor (f )
t

(t) = f t t : d 0 x, t dt . (9.86)
0
On the macroscale the two tensors become dual, and then we have reff (t)
and feff (t). Huet has shown that reff (t) is bounded by the hierarchy
reff (t) . . . r (t) r (t) . . . r1 (t) = /2, t 0, (9.87)
and feff (t) is bounded by the hierarchy
feff (t) . . . f (t) f (t) . . . f1 (t) = /2, t 0 (9.88)
A number of related results on that subject are in Huets papers referenced

above.
9.6 Stokes Flow in Porous Media

Under consideration is a steady, single-phase fluid flow in a random porous
medium with explicit account of a spatially disordered microstructure. In phe-
nomenological, deterministic continuum mechanics, such a flow is
described by Darcys law. With the slow flow in a porous medium also consid-
ered as incompressible and viscous, Darcys law is governed by the equations
(Dullien, 1979)
1
U = Kp U = g (x) (9.89)

where U is the Darcy velocity (volume averaged) velocity, p is the applied
pressure gradient driving the flow, is the fluid viscosity, and K is the per-
meability, a second-rank tensor that depends on the microstructure of the
porous medium. Equation (9.89)2 is the continuity equation and g (x) is the
source/sink term.
In the general studies on property bounds, the variational principles usu-
ally provide the basic starting point (Prager, 1961; Torquato, 2002). Now,
Darcys law (9.89) is not a local constitutive relation, since the slow flow
in the voids is governed by the Stokes equation and the continuity equation
2 v = p v=0 (9.90)
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
In any chosen element dV of a porous medium the energy dissipation can

be expressed in two different quadratic forms: one in the space of pressure
gradient, and another in the space of Darcy velocity
K
( p) = V1 V ij p, j p, j dV (U) = V1 V Rij U j U j dV (9.91)
The Hill condition for Darcys law is
pU = p U, (9.92)
which leads to three possible loadings:
1. A uniform Dirichlet (or essential) boundary condition
p(x) = p,i xi x B (9.93)
2. A uniform Neumann (or natural) boundary condition
Ui (x)ni = U,i ni x B (9.94)
3. A uniform orthogonal-mixed boundary condition

% &
[ p(x) p,i xi ] Ui (x)ni U,i ni = 0 x B (9.95)
At this point, on the mesoscale, we introduce apparent dissipation func-

tionals
K ij
( p) = p, j p, j (U) = Rij U j U j , (9.96)

which, respectively, are homogeneous functions of degree 2 in the volume
average pressure gradient p ( p,i ) and the volume average Darcy velocity
U ( Ui ). Thus, we have (Ziegler, 1983)
1 1
Ui = p,i = (9.97)
2 p,i 2 Ui
With this we have a full analogy between the Stokesian flow in porous
media and the antiplane elasticity in a matrix of finite stiffness, containing
rigid (infinitely stiff) inclusions. Employing the minimum potential energy
principle with the essential (e ) condition
(9.93), we find a scale-dependent
hierarchy of upper bounds on Ke
e
K . . . Ke Ke . . . Ke1 = /2 (9.98)
where

=K
Ke = Keff eff
(9.99)

is the effective (RVE level) permeability. Note that Ke1 pertains to the smallest
scale where Darcys law still applies; by analogy to elasticity of random media,
it may be called a Voigt bound.
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
Similarly, from the minimum complementary energy principle together

with the (n ) natural
condition
(9.94), we obtain a hierarchy of upper bounds
on the resistance Rn
n
R . . . Rn Rn . . . Rn1 = /2 (9.100)
where
1
Rn = Keff . (9.101)
Collecting these inequalities, we have a two-sided hierarchy

1 1 n 1
Rn1 . . . Rn R . . . Keff . . . Ke

Ke . . . Ke1 = /2. (9.102)
Figure 9.17 illustrates the comparison between both kinds of boundary

conditions in terms of the pressure fields. We can clearly see in (b) that the
pressure distributes nonlinearly on side boundaries in order to maintain the
flow within the top and bottom boundary according to the requirment of
Neumann boundary condition. At the same time, the velocity field is non-
uniform in (a). As the mesoscale increase, all the fluctuations becomes so
negligible that the results of the Neumann boundary value problem begin
to coincide with those of the Dirichlet problem (except for the obvious non-
uniqueness of the Neumann problem).
Figure 9.18(a) displays the simulation results of the random medium at
60% porosity in terms of the ensemble average half-traces of the permeability
and resistance tensors. The bounding character of these tensors and their
convergence to the RVE (effective medium) with the increasing mesoscale
are clearly seen. In fact, at = 80, the ensemble averages of both tensors are
approximately equal within 0.4%.
Another viewpoint of that convergence to RVE may be offered in terms of
the departure of one half of the ensemble averaged scalar product of Ke and
Rn from unity, because, at , we must have
1 e
K Rn = 1. (9.103)
2
The statistical isotropy of the underlying Poisson point field with exclusion
dictates the isotropy
of not only
the effective (RVE level) properties but also
the isotropy of Ke and Rn . Thus, we can work with
d = Ke,1 j Rn,1 j 1 (9.104)
to characterize the above-mentioned departure from the RVE properties; here

1 j are the indices of particular components of the Ke and Rn tensors. The
said departure is shown in Figure 9.18(b), both for the preceding case of
porosity, as well as for the entire range: 50% through 80%. Clearly, the scaling
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
Pressure on boundary
(a)
(b)
FIGURE 9.17
Pressure fields in a realization of the random porous medium on mesoscale = 4 under uniform
Dirichlet (a) and Neumann (b) boundary conditions.
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
45
Dirichlet
40 B.C.
<K11>( <R11>1) (109m2)

Neumann
35 B.C.
30
25
Round disks: 40%
20
0 20 40 60 80 100
(L/d)
(a)
2.4
Round disk model, porosity:
80%
2.2
70%
2 60%
KijeRjk n(I = k = 1, or 2)
50%
1.8
1.6
1.4
1.2
1
0 20 40 60 80 100
(L/d)
(b)
FIGURE 9.18
(a) Effect of increasing mesoscale on the convergence of the permeability/resistance tensor hier-
archy. (b) Scaling of Rn11 Ke11 to unity. (After Du & Ostoja-Starzewski, 2006b).
effect of permeability in porous media depends more strongly on the size of

micro-channels than on the disk diameter d itself. The rate of convergence to
RVE decreases with the porosity increasing. For example, at = 32, the RVE
size can be considered to be attained at porosity 50%. However, for a higher
porosity (e.g., 80%) the RVE size is not achieved even at = 80.
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
9.7 Thermoelastic Microstructures

9.7.1 Linear Case
Let us now consider a linear thermoelastic composite material (Rosen and
Hashin, 1970; Christensen, 1979). Each specimen B() B is described locally
(within each phase) by either
ij = Sijkl (, x)kl + ij (, x) or ij = Cijkl (, x)kl + ij (, x),
(9.105)
1
where Sijkl (, x) = Cijkl (, x), is the temperature rise, and
ij (, x) = Cijkl (, x)kl (, x) (9.106)
are the thermal stress coefficients computed from the thermal expansion
coefficients kl . We also recall from thermoelasticity that the stress and
entropy are derivable from the Helmholtz free energy density as

A A (9.107)
ij = S=
ij T T ij
where
1 1 2
A= ij Cijkl kl + ij ij c v , (9.108)
2 2 T0
for a small temperature change = T/T0 , and c v is the specific heat under
constant volume. Noting the Legendre transformation, the potential energy
is defined as

U= Ad V Fi ui dV ti ui dS, (9.109)
V S St
where St is the part of boundary S with traction prescribed on it, and Fi is the
body force.
On the other hand, we have

G G
ij = S= (9.110)
ij T T ij
where the Gibbs free energy is
1 1 2
G= ij Sijkl kl ij ij c p , (9.111)
2 2 T0
with c p being the specific heat under constant traction. Again by the Legendre
transformation, the complementary energy is defined as

U = Gd V + ti ui dS, (9.112)
V Su
where Su is the part of S with displacement prescribed on it.

P1: Binod
May 28, 2007 18:23 C4174 C4174C009
Of interest is the derivation of scale-dependent bounds on eff as

well as c v and c p , using the already available bounds on Ceff C (recall
Chapter 7). Note here that
eff eff eff eff eff eff
ij = Cijkl kl , p c v = T0 C ijkl ij kl .
c eff eff
(9.113)
Of course, the energy relations for a mesoscale above have to be reconsid-

ered in light of having specific applied loading. Thus, it has been shown that
the scale-dependent random functional of potential energy under displace-
ment controlled boundary conditions is
1 1 2
U ij , , = ij0 Cijkl, () kl0 + i j, () ij0 c v, () , (9.114)
2 2 T0
where

1 T0
i j, () = ij + Dijkl kl dV c v, () = c v + E ij ij dV. (9.115)
V V V V
In (9.115) is the local fluctuation, whereas the tensor Dijkl relates the applied
strain to the local elastic strain fluctuation e ij , and the tensor E ij relates the
temperature change to the thermal strain fluctuation e ijth
e ij (x) = Dijkl (x) kl0 e ijth (x) = E ij (x) . (9.116)
It is shown in the works referenced above that (macroscopically) effective

eff
thermal expansion coefficients kl can be derived in terms of the effective
eff eff 1
moduli Cijkl = (Sijkl ) and the information on the distribution of individual
phases. That is, in the case of two phases, 1 and 2,
eff eff
ij = (kl(1) kl(2) ) Pklmn (Smni j Smni
(2) (2)
j ) + ij , (9.117)
where
1
s Smnr s ) = Iklr s = (kr ls + ks lr ).
eff
Pklmn (Smnr (9.118)
2
Alternatively, as pointed out in the aforementioned works, bounds on eff
can be obtained by using bounds on Ceff , and such a result was produced
employing the HashinShtrikman bounds.
A study of the scaling from the SVE to the RVE was recently reported by
Du and Ostoja-Starzewski (2006b). First, the Hill condition is extended so as
to develop the equivalence between the energetic and mechanical formula-
tions of constitutive laws of thermoelastic random heterogeneous materials at
arbitrary mesoscale . Let us note there that, while the potential energy of a
homogeneous material is
1 eff 1 2
UP = Cmni j ij kl + ij ij c v , (9.119)
2 2 T0
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
where c v is the specific heat at constant volume and T0 is the reference temper-
ature, the potential energy of a heterogeneous material in a mesoscale domain
B is

1 eff 1 2
UP = Cmni j ij kl + ij ij c v , (9.120)
2 V 2 T0
Indeed, both energy forms (potential and complementary) at the SVE level
can be cast in the same form as those of the homogeneous material, that is,
(9.108) and (9.111).
Next, taking an ensemble averages and noting the scale dependent
hierarchy (7.70) of Chapter 7, leads to two scale-dependent hierarchies for
the isotropic part t of itj, :
(i) (1) (2) 0 and (1) > (2) :

. . . t t . . . 1t R = /2. (9.121)
(ii) (1) (2) 0 and (1) < (2) :

. . . t t . . . 1t V = /2. (9.122)
where R is the Reuss-type bound on .

Furthermore
Cnni
d
j ij
(2)
ijd = (1) (2) (2) ij . (9.123)
(1) (2)
with the help of (7.70) we derive two hierarchical relations for the isotropic
part d of idj, :
(i) 0 (1) (2) 0 and (1) > (2) :

. . . d d . . . 1d V = /2. (9.124)
(ii) 0 (1) (2) 0 and (1) < (2) :

. . . d d . . . 1d V = /2. (9.125)
where V is the Voigt-type bound on . In view of (9.113)2 , this provides a

two-sided bounding hierarchy on , or, equivalently, on . Both inequal-
ities are obtained under the uniform traction and displacement boundary
conditions, respectively. Figures 9.19 and 9.20 display simulation results un-
der these two loadings.
Note: Due to the presence of a nonquadratic term in energy formulas,
the mesoscale bounds on the thermal expansion are more complicated than
those on the stiffness tensor and heat capacity. In general, upper and lower
bounds correspond to loading of mesoscale domains by essential and natural
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
(a)
(b)
FIGURE 9.19
Numerical results at (a) = 2 and (b) = 32 under the displacement boundary condition (9.44)
at ij0 ; disks do not touch.
boundary conditions. Depending on the property mismatches, the upper and

lower bounds can be provided either by essential boundary conditions or
natural boundary conditions.
Suppose we deal with a two-phase composite material with locally
isotropic phases. To fully characterize it, we need three mismatches between
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
(a)
(b)
FIGURE 9.20
Numerical results at (a) = 2 and (b) = 32 under the traction boundary condition (9.50) at ij0 ;
disks do not touch.
both phases:
E (i) /E (i) (i) /(i) (i) / (i) . (9.126)
Figure 9.21 shows the hierarchies of bounds for a composite with 40% volume
fraction of inclusions.
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
2.8682
i
2.8681 Ei = 3, vi = 1, i = 0.5, cp = 2.
cv 106 (J/m3K)
m
E m v m m cp
ctv
2.868
cdv
2.8679
0 8 16 24 32 40
(L/d)
FIGURE 9.21
Scaling of specific heat capacities from SVE toward RVE at 40% volume fraction of inclusions.
Finally, there exist similar hierarchies on the effective specific heat coeffi-
cient: one at constant volume
d d
c v c v,
d
. . . c v, c v, . . . c v,1
d
c vV = /2, (9.127)
and another at constant pressure

c p c tp, . . . c tp, c tp, . . . c tp,1 c Vp = /2. (9.128)
9.7.2 The Nonlinear Case

We refer back to Section 9.2. By generalizing the derivation of the hierarchy
(9.31) to thermoelastostatics, we obtain

(F0 , T0 ) . . .
(F0 , T0 )
(F0 , T0 ) . . .
1 (F0 , T0 )
= /2, (9.129)
where T0 is the prescribed temperature. Similarly, (9.38) is replaced by

G (P0 , T0 ) . . . G (P0 , T0 ) G (P0 , T0 ) . . . G 1 (P0 , T0 )
= /2, (9.130)
P1: Binod
May 28, 2007 18:23 C4174 C4174C009

where G(P0 , T0 , ) = (V0 ) 1 B g X, P0 , T0 , dV. Utilizing the Legendre
transformation relating Gibbs (g) and Helmholtz () energy densities

g ,T = xi, j , (9.131)
ui, j ui, j
and applying stress-free boundary conditions (i.e., a free expansion) P0 = 0,

the hierarchy (9.130) transforms to

(P0 ) . . .
(P0 )
(P0 ) . . .
1 (P0 ) = /2,
(9.132)
which provides a lower bound on the effective Helmholtz free energy. Note
that under stress-free boundary conditions, the elastic part of
equals zero
for a homogeneous body. In contrast, for a heterogeneous material, the elastic
contribution increases with .
For a numerical example, we consider the simplest form of the potential
given by a neo-Hookean strain-energy function (Dhont, 2004)
1% &
= (T) 21 + 22 + 23 3 + (T) ( J M 1) 2 + T (T) 2 = J 1/3 a .
2
(9.133)
Here (T) and (T) are temperature-dependent initial shear and initial bulk
moduli. The quantity T (T) is the purely thermal contribution to the free
energy and, because it does not change with scale, can be ignored in the
hierarchies above. The Jacobian is decomposed into purely mechanical (J M )
and purely thermal (J T ) parts according to J = J M J T , with J T = (1 + T) 3 .
In general, the free energy of the composite can be written as
% &
= 12 (T) 21 + 22 + 23 3 + (T) ( J M 1) 2 + T (T) 2 = J 1/3 a .
(9.134)
where
'
1
F0 , T0 , = (T) ( 01 ) 2 + ( 02 ) 2 + ( 03 ) 2 3
2
0 2 (
J
+ (T) 1 + T (T) . (9.135)
JT

and X, F0 , T0 , is a local fluctuation of .
Various comparisons of thermal expansion/stress coefficients have been
carried out by Khisaeva (2006); see also Khisaeva and Ostoja-Starzewski
(2007). As an example, in Figure
9.22, we show the scaling trend of the ther-
mal stress coefficient tr ij on the temperature change T for one particular
composite. Note the smooth transition of response into that predicted by the
linear theory as T decreases.
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
0.44
T = 300C
T = 150C
0.42
T = 75C
T = 38C
0.40 T = 5C
/max
0.38
0.36
0.34
0.32
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

FIGURE 9.22
Dependence of the thermal stress coefficient on the temperature change T for the rubber
polystyrene composite.
9.8 Scaling and Stochastic Evolution in Damage Phenomena

Consider a material whose elasticity is coupled to damage state, as described
by the constitutive equation (Lemaitre and Chaboche, 1994)
ij = (1 D)Cijkl kl . (9.136)
Here Cijkl is isotropic, and must be coupled with a law of isotropic damage,
that is,

D = , (9.137)
Y
with Y =
/, being the (Helmholtz) free energy. This formulation is
set within the TIV framework, see Chapter 10. In particular, the scalar D
evolves with the elastic strain = ii , which is taken as a time-like parameter,
according to
)
D (/0 ) s when = D and d = d D > 0,
= (9.138)
0 when < D and d < 0.
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
Integration from the initial conditions D = D = 0 up to the total damage,

D = 1, gives

+1
% &s +1 % &
D = (/0 ) s R = (1 + s )sD = 1 (/ R ) s +1 E, (9.139)
where = ii .
This formulation is understood as the effective law for the RVE, that is,
eff
Cijkl = Cijkl | Deff = D|
eff =
| eff = | , (9.140)
as well as a guide for adopting the form of apparent responses on mesoscales.

Thus, assuming that the same types of formulas hold for any mesoscale , we
have an apparent response for any specimen B ()
= (1 Dd )Cd () : 0 (9.141)
under uniform displacement boundary condition. The notation Dd expresses

the fact that the material damage is dependent on the mesoscale and the
type of boundary conditions applied (d ). In fact, although we could formally
write another apparent response = (1Dd ) 1 St () : 0 , we shall not do so
because the damage process under the traction boundary condition (t) would
be unstable.
It is now possible to obtain scale-dependent bounds on Dd through a pro-
cedure analogous to that for linear elastic materials, providing one assumes
a WSS
d and
ergodic microstructure. One then obtains a hierarchy of bounds
on D Deff from above (Ostoja-Starzewski, 2002b):

Dd Dd . . . Dd = /2. (9.142)
These inequalities are consistent with the much more phenomenological

Weibull model of scaling of brittle solids saying that the larger is the specimen
the more likely it is to fail.
Next of interest is the formulation of a stochastic model of evolution of
Dd with to replace (9.138)1 . Said differently, we need a stochastic process
Dd = {Dd (, ); , [0, R ]}. Assuming, for simplicity of discussion,
just as in Lemaitre and Chaboche (1994), that s = 2, we may consider this
setup:
d Dd (, ) = Dd (, ) + 3 2 [1 + r ()]dt, (9.143)
where r () is a zero-mean random variable taking values from [a , a ],

1/ = a < 1. This process has the following properties:
1. Its sample realizations display scatter -by- for < , that is, for
finite body sizes.
2. It becomes deterministic as the body size goes to infinity in the RVE
limit ( ).
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
3. Its sample realizations are weakly monotonically increasing func-

tions of .
4. Its sample realizations are continuous.
5. The scale effect inequality (9.142) is satisfied, providing we take R
as a function of with a property
R () < R ( ) = /2. (9.144)
Let us observe, however, that, given the presence of a random microstruc-

ture, mesoscale damage should be considered as a sequence of microscopic
eventsshown as impulses in Figure 9.23(a)thus rendering the apparent
damage process Dd one with discontinuous sample paths, having increments
d Dd occurring at discrete time instants, Figure 9.23(c). To satisfy this require-
ment one should, in place of the above, take a Markov jump process whose
range is a subset [0, 1] of real line (i.e., where Dd takes values). This process
would be specified by an evolution propagator, or, more precisely, by a next-
jump probability density function defined as follows: p( , Dd | , Dd )d d Dd =
probability that, given the process is in state Dd at time , its next jump will
occur between times + and + + d , and will carry the process to some
state between Dd + Dd and Dd + Dd + d Dd .
Figure 9.23(b) shows one realization Cd (, ); , [0, R ], of the
apparent, mesoscale stiffness, corresponding to the realization Dd (, );
, [0, R ], of Figure 9.23(c). In Figure 9.23(a) we see the resulting consti-
tutive response (, ); , [0, R ].
Calibration of this model (just as the simpler one above)that is, the spec-
ification of p( , Dd | , Dd )d d Dd may be conducted by either laboratory
or computer experiments such as those discussed in Chapter 4. Note that in
the macroscopic picture ( ) the zigzag character and randomness of
an effective stress-strain response vanish. However, many studies in mecha-
nics/physics of fracture of random media (e.g., Herrmann and Roux, 1990),
indicate that the homogenization with is generally very slow, and
hence that the assumption of WSS and ergodic random fields may be too
strong for many applications.
Extension of the above model from isotropic to (much more realistic)
anisotropic damage will require tensor, rather than scalar, Markov processes.
This will lead to a somewhat greater mathematical complexity, which may
be balanced by choosing the first model of this section rather than the latter.
These issues are quite secondary. Our goal has been to outline a stochas-
tic continuum damage mechanics that (1) is based on, and consistent with,
micromechanics of random media as well as the classical thermomechanics
formalism, and (2) reduces to the classical continuum damage mechanics in
the infinite volume limit.
Over the past two decades numerous studies of damage in both classical
and non-classical as well as deterministic and stochastic settings have been
conducted, e.g., Carmeliet and de Borst (1995), Pamin (2006), Bazant (2007).
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
/E
/R
(a)
C = E (1-D)
/R
(b)
D
/R
(c)
FIGURE 9.23
Constitutive behavior of a material with elasticity coupled to damage where / R plays the role
of a controllable, time-like parameter of the stochastic process. (a) Stress-strain response of a
single specimen B from, B having a zigzag realization; (b) deterioration of stiffness Cd (, ); (c)
evolution of the damage variable. Curves shown in (ac) indicate the scatter in stress, stiffness,
and damage at finite scale . Assuming spatial ergodicity, this scatter would vanish in the limit
, whereby unique response curves of continuum damage mechanics would be recovered.
9.9 Comparison of Scaling Trends

We now recall question 6 of the Preface, which may be rephrased as: on what
mesoscales is the RVE attained with the same accuracy for various types of
random microstructures? Said differently, given a specific mesoscale, what are
P1: Binod
May 28, 2007 18:23 C4174 C4174C009
TABLE 9.2
Mismatch and Discrepancy Values on Mesoscale = 16
Physical Subject Mismatch D [%]
(i) (i)
Linear elasticity = 10 =1 2.28
(m) (m)
(i) (i) (i) 1
Linear thermoelasticity = 10 = 10 = 5.51
(m) (m) (m) 10
h (i) E (i)
Plasticity = 10 =1 2.29
h (m) E (m)
(i)
0 0(i)
Finite elasticity = 10 =1 5.86
(m) 0(m)

0
(i)

tr K
Flow in porous media = 27
tr K(i)
the discrepancies between the bounds obtained from kinematic and traction
boundary conditions for various types of random microstructures?
Following a recent study (Ostoja-Starzewski et al., 2007), a comparison
of such results is given in terms of a discrepancy D (Table 9.2). The latter is
defined as
Re Rn
D= , (9.145)
Re + Rn /2
where, depending on the case, Re and Rn are the responses under kinematic
and traction loadings, respectively. In each case, the same planar random
matrix-inclusion composite is used, where the centers of inclusions are gen-
erated via a Poisson random field with inhibition.
We close this chapter with these observations:
Scaling from SVE to RVE slows down when:

1. We go from elastic to inelastic microstructures
2. The mismatch in properties grows
3. We go from 2D or 3D setting
4. The microscale geometries exhibit nonuniform effects (from
disks to ellipses, spatial clustering of inclusions, etc.)
Scaling from SVE to RVE in linear elastic microstructures:
1. Mesoscale moduli of stiff matrix with soft inclusions con-
verge (much) more slowly to RVE than those of a soft matrix
with stiff inclusions
2. Convergence to RVE is slowest in antiplane, faster in in-
plane, and fastest in 3D elasticity.
P1: Binod
July 5, 2007 16:10 C4174 C4174C010
10
Mesoscale Response in Thermomechanics
of Random Media
. . . the dissipation function is the key to irreversible thermo-

dynamics.
H. Ziegler, 1983
The presence of dissipative phenomena in mechanics of materials neces-

sitates a formulation of continuum mechanics consistent with principles of
thermodynamics, leading to so-called thermomechanics or continuum thermo-
dynamics. As elaborated by Maugin (1999), there are now four classical con-
tinuum thermodynamics theories: thermodynamics of irreversible processes
(TIP); thermodynamics with internal variables (TIV); rational thermo-
dynamics (RT); extended (rational) thermodynamics (ET). All of these are
deterministic, homogeneous continuum theories without clear account of
the underlying random compositions of materialsthat is, they a priori pos-
tulate the existence of the RVE. Strictly speaking, some statistical treatments
were carried out as a bridge from micro to macro levels for select variants of
the above theoriesfor example, by Ziegler (1963, 1970) for TIV (see below),
or by Muschik et al. (2000) for TIP (see Chapter 8)but such studies were
only concerned with providing foundations from the standpoint of statistical
physics directly to the level of the RVE, without making clear what the size of
the RVE actually should be. Given the widespread use of TIV in mechanics of
materials (e.g., Lemaitre and Chaboche, 1990), we now set out to generalize
it so as to provide a link to random microstructures.
10.1 From Statistical Mechanics to Continuum Thermodynamics

10.1.1 Dissipation Function of the RVE
In this section, following Ziegler (1962, 1963, 1972), and consistent with Gibbs
(1902), we consider a mechanical system with a very large number of N
particles (i.e., with 6N degrees of freedom) with generalized coordinates q n
(n = 1, . . ., 3N), pn (n = 1, . . ., 3N), and a k (k = 1, . . ., K ). The q n s and pn s
are the so-called microcoordinates, while a n s are the so-called macrocoordinates.
359
P1: Binod
July 5, 2007 16:10 C4174 C4174C010
The first of these, as they refer to atoms or molecules in an RVE, are too numer-
ous and miniscule in magnitude to be directly measured, as opposed to the
latter, which are perceptible to a macroscopic observer and, therefore, used
to described the systems macrostate. Thus, the a n s are the strain components
of the RVE and possibly the internal parameters necessary to study, say, an
elastic-plastic material.
Ziegler begins with a motion of the microsystem described by Hamiltons
equations [recall (2.158) of Chapter 2], where the Hamiltonian H (q n , pn , a k )
represents the total energy of the system. In view of equations (2.158) of
Chapter 2, the time rate of H is
H
H = a k , (10.1)
a k
which means that H is modified through the macrocoordinates alone.

On a continuum mechanics/thermodynamics level we have a determinis-
tic system, modeling the RVE, which is described by the a k s, which vary slowly
compared with the q n s. The evolution of H in the phase space {(q n , pn ) ; n
= 1, . . ., N} follows Hamiltons equations (2.158) of Chapter 2. As mentioned
there, the discrete distribution, and indeed the motion, of phase points is
described by a so-called canonical density (q n , pn , t). The latter is henceforth
treated as a distribution over a continuum since N 1. Thus, we have a
continuum-like, incompressible phase fluid, for which there must hold a con-
servation of phase points, as expressed by the Liouville theorem
d
= 0. (10.2)
dt
Here we may recall the continuity equation for a real fluid of density in
continuum mechanics
d d
0= + vi,i = , (10.3)
dt dt
where the second equality holds if the fluid is incompressible. Equation (10.3)
allows us to write an integral form

df d
dV = f dV , (10.4)
dt dt
where f is an arbitrary function. Also in a real continuum, the material rate

of change of internal energy u per unit volume due to heat flux q alone is
du
+ q i,i = 0. (10.5)
dt
Returning to the phase space, the role of dV is played by dq 1 , . . ., dq N ,
d p1 , . . ., dp N , and the integration over that space means the ensemble aver-
aging relative to the density , as shown in (10.11) below. Also note the
P1: Binod
July 5, 2007 16:10 C4174 C4174C010
Mesoscale Response in Thermomechanics of Random Media 361
normalization condition

dV = 1. (10.6)
Ziegler assumes the particle dynamics to be holonomic, scleronomic, and

admits gyroscopic forces derivable from a potential (Problem 18 of Chapter 2).
The latter feature allows one to admit, say, Coriolis or Lorentz forces occurring,
respectively, in rotating coordinate systems or magnetic fields. Ziegler also
identifies degrees of freedom of microsystem , and notes that (q n , pn , t)
supplies a canonical probability w (q n , pn , t) of finding the phase point of a
given microsystem in the unit volume centered at (q n , pn ) at the time t. In
general, they are proportional
h
w (q n , pn , t) = (q n , pn , t), (10.7)
N
where h is the unit of actionsuch as Plancks constant in quantum
mechanics.
According to a postulate of Gibbs, the surfaces H = const and = const
coincide in a state of statistical equilibrium, that is, all phases of the microsys-
tem characterized by the same value of H are equally probable. Thus, is a
function of the energy H of the system
= ( H), (10.8)
and, as shown by Gibbs, the simplest function suitable for is
H ()
() = exp , (10.9)
kT
where is the free energy and k is the Boltzmann constant. Hereinafter, we
explicitly indicate the dependence of on the actual outcome , and hence
the random character of H. It now follows that the so-called Gibbs index of
probability (or Gibbs phase) is
H ()
() = ln () = , (10.10)
kT
and, consistent with Section 2.6 of Chapter 2, the information entropy h is
again the ensemble average of ln p

h = = ln dV. (10.11)
While from (10.11) we have
d d
= h, (10.12)
dt dt
P1: Binod
July 5, 2007 16:10 C4174 C4174C010
from (10.10)1 we have d/dt = /, which, in view of (10.6), gives

d
= dV = dV = 0. (10.13)
dt
Comparing this with (10.12), we see that the time differentiation does not
generally commute with the ensemble averaging:

d d
=
. (10.14)
dt dt
This type of result carries over to the Hamiltonian. First, from (1.10)2
we see
H () = Tk (), (10.15)
which, upon ensemble averaging, gives
H = + Tkh = + T S, (10.16)
where the second equality follows from the relation between the information
and thermodynamic entropies
kh = S. (10.17)
From this, upon time differentiation, we obtain

d
+ T S + T S.
H = (10.18)
dt
On the other hand, subjecting (10.15) first to time differentiation, and then
ensemble averaging, we get

H = + T S, (10.19)
where (10.11), (10.13), and (10.17) have been used. Thus,

d d d
H = H + T S. (10.20)
dt dt dt
Comparing (10.20) with (10.4), we see that there is no continuity in the phase
space for nonisentropic processes. In other words, if it were not for the pres-
ence of S, continuity would hold for H. To reflect the fact that the Liouville
theorem is not satisfied, Ziegler replaces (10.2) by a transport equation
d a n b n
+ + = 0, (10.21)
dt q n pn
where the vector pair (a, b), called flux, represents the change of density due
to the creation/annihilation of the phase fluid, called flux. Note that (10.21)
combines two types of transport exemplified by (10.3)1 and (10.5).
P1: Binod
July 5, 2007 16:10 C4174 C4174C010
Now, upon recognizing in the average Hamiltonian the RVEs internal

energy
H = U, (10.22)
where both sides are now dependent on S and the vector a of a k s but not q i s and
pi s, we recall the key relations of (deterministic) continuum thermomechanics
(Ziegler, 1983): the energy conservation law,
U = L + Q = A a + Q , (10.23)
where L is the power of forces A corresponding to kinematic parameters a, and

Q is the heat supply per unit time; and the second law of thermodynamics,

S = S(r ) + S(i) S(r ) = Q /T S(i) = A(d) a /T 0, (10.24)
where we have a decomposition of entropy rates S(r ) (reversible) and S(i)

(irreversible), as well as a decomposition of forces into quasi-conservative
and dissipative parts:
A = A(q ) + A(d) . (10.25)
These forces are determined from the free energy and the dissipation
function
( a) = T S(i) . (10.26)
as
1
(q )
Ak = A(d)
k = = a k , (10.27)
a k a k a k
wherein (10.27)2 expresses Zieglers thermodynamic orthogonality. It leads to

the principle of maximal dissipation rate: Provided the dissipative force A(d) is
prescribed, the actual velocity a maximizes the dissipation rate L (d) = A(d) a
subject to the side condition
( a) = A(d) a = L (d) 0. (10.28)
Ziegler (1970) proves this orthogonality on the basis of statistical mechan-

ics (Ziegler, 1970): Consider variations of a by a during a small time interval.
For variations a that do not affect the various sides in (10.28), the behavior of
the macrosystem during that time interval is the same. It follows that not only
a single velocity a is compatible with the prescribed force A(d) , but also all
those varied velocities a + a for which the dissipation function, and hence
also the scalar product with A(d) remain unchanged. However, this is possible
only if a is infinitesimal and A(d) is orthogonal to the tangential plane passing
through the end point of a.
Note: The application of the maximum entropy method (Chapter 2), sub-
ject to the condition (10.22), results in the exponential form of density (10.9).
P1: Binod
July 5, 2007 16:10 C4174 C4174C010
10.1.2 Departure from the Second Law of Thermodynamics

The second law of thermodynamics applies to macroscopic systems, not
microscopic ones. In the 1990s the so-called fluctuation theorem (FT) was
developed; the proof was given in Evans and Searles (1994). It is a statement
about the probability distribution of the irreversible entropy production, t ,
averaged over a time interval t. The FT states that, in systems not in thermo-
dynamic equilibrium over a time interval t, the ratio of the probability that
t = A and the probability that it takes the opposite value, A, is e At :

P t = A

= e At . (10.29)
P t = A
That is, as the time or system size increases (because t is extensive), the
probability of observing an entropy production opposite to that dictated by
the second law of thermodynamics decreases exponentially. When the FT is
applied to macroscopic systems, the second law is recovered.
The FT has consequences in nanomechanics and biophysics. For example,
very small-scale machines will spend part of their time actually running in
reverse, that is, in a way opposite to that for which they were designed.
Note: With the help of the FT, the MEM has recently been shown to provide
a more fundamental basis for Zieglers orthogonality principle (Dewar, 2005).
10.2 Extensions of the Hill Condition

10.2.1 The Hill Condition in Thermomechanics
In order to extend the continuum thermomechanics to random media at
mesoscale, that is, below the RVE but above the atomic/molecular level of the
previous section, we first need to establish other expressions of the Hill con-
dition. As is well known, the field forms of five basic laws of (deterministic)
continuum thermomechanicsthe conservation of mass, conservation of lin-
ear momentum, conservation of angular momentum, conservation of energy,
and second law of thermodynamics (entropy production inequality)are

+ div (v) = 0
t
div + b v = 0
T = (10.30)
u : d = r divq
r q
s div = si 0.
T T
P1: Binod
July 5, 2007 16:10 C4174 C4174C010
Here is the mass density, v the velocity, the Cauchy stress, d the defor-
mation rate, u the internal energy, r the heat source, q the heat flux, T the
temperature, s the entropy, and si the irreversible part of entropy production.
Noting these rules involving the averaging operation, in addition to
relations (2.52) of Chapter 2, pertaining to field (intensive) quantities

V Z = V Z + V Z
a 0 + a i Zi = a 0 + a i Zi ; a i nonrandom

dZ Z
= + v grad Z + v gradZ , (10.31)
dt t
Huet (1982) developed Hill-type conditions for these laws:

+ div( v) = div v
t

div + b v = v gradv + v b
T =

u : d r + div q = u v grad u + : d

1 1
s r + div q si
T T

1
= v grad s s + r
T

1
div r + v grad si + si
T

si v grad si si

(10.32)
Here indicates a time derivative acting on . These six equations lead
to the so-called assimilation conditions, which ensure homogenization at the
effective continuum level
div( v ) = 0

v gradv + v = b

v grad u + u = : d

v grad si + si = 0

1 1
v grad s + s = r div b . (10.33)
T T
P1: Binod
July 5, 2007 16:10 C4174 C4174C010
10.2.2 Homogenization in Dynamic Response

Moving from quasi-statics to dynamics, one has to admit inertial effects. The
Hill condition of earlier chapters needs to be modified. Starting from the local
equations of motion
ij,j = ui , (10.34)
Wang (1997) found
i j i j i j i j + i i j, j i j ki , k x j =

1
i (X) vi,, j x j nk (X) ki (X) ki0 d S (10.35)
V B
When dealing with strain rates and velocities instead of strains and displace-
ments, the left hand side in the above may also be written as i j i j i j i j +
(i i ) i j i x j .
1 d
2 dt
In the case ( ui = 0) and when either one of three uniform
boundary
conditions of Chapter 7 is prescribed, the Hill condition i j i j = i j i j is
satisfied. However, in the dynamic case, when velocities prescribed on B are
affine, the right hand side of (10.35) equals zero, and that equation reduces to
i j i j = i j i j + i i j, j i j ki , k x j
1 d
= i j i j + (i i ) i j i x j . (10.36)
2 dt
On the other hand, in the dynamic case with uniform tractions prescribed,
one finds a simpler relation
involving
only the rate of strain energy computed
from volume averages i j i j , the volume average of rate of strain energy,
and the rate of kinetic energy
1 d
i j i j = i j i j + i i j, j = i j i j + (i i ) (10.37)
2 dt
Clearly, we see a generalization of the conventional Hill condition. Z.-P.
Wang (1997) has developed dynamic plasticity models of porous materials
on the RVE level with this formulation. A still outstanding challenge is the
development of such models to randomly heterogeneous materials on the
SVE (mesoscale) level, with explicit inclusion of scale effects.
It is also, of interest to note the effective (RVE level) equation of motion,
derived by Wang and Sun (2002):
i j, j + Fi = ui , (10.38)
where Fi may be regarded as an effective body force resulting from the

micro-inertia.
P1: Binod
July 5, 2007 16:10 C4174 C4174C010
10.3 Legendre Transformations in (Thermo)Elasticity

10.3.1 Elasticity
On the microscale we have
( ) + () = : , (10.39)
where () and ( ) are also functions of position x and realization .

Passing to a mesoscale > 0, given the randomness of response of B (), there
are two Legendre transformations linking the potential ((0 )) and comple-
mentary ( ( 0 )) energies, depending on whether strain ( 0 ) or stress ( 0 ) is
controlled via uniform displacement or traction boundary conditions:

( ()) + 0 = () : 0 0 + ( ()) = 0 : ().
(10.40)
Upon ensemble averaging on the SVE level, we obtain

0
( ) + 0 = : 0 + () = 0 : .
(10.41)
Thus, in the first case here, is the ensemble average outcome and it becomes
the argument of . In the second case, is the ensemble average outcome
and it becomes the argument of .
As the mesoscale increases indefinitely ( L/d )or, in other
words, as the SVE turns into the RVEthe relations (10.41)1 and (10.41)2
should coincide and turn into the classical statement of a deterministic con-
tinuum theory:

eff 0 + eff 0 = : 0 (10.42)
whereby the distinction between 0 and , as well as between 0 and , van-

ishes, as we are dealing with the RVE situation. Accordingly, we have dropped
the subscript on eff and eff in the above.
10.3.2 Thermoelasticity
When the material is thermoelastic, the single Legendre transformation is gen-
eralized to a quartet of partial Legendre transformations linking internal energy,
enthalpy, Gibbs and Helmholtz energies (Sewell, 1987, Collins and Houlsby,
1997; Houlsby and Puzrin, 2000). To start with, in analogy to (10.39), con-
sider the case of a homogeneous continuumeffectively the RVE level. Thus,
depending on what we take as the reference, or controllable, loading case, we
have either (0 ) or ( 0 ) playing the role of X(xi , i ) in Figure 10.1; x( xi )
is 0 or 0 , and we have to add ( i ) or ( i ) as temperature or entropy.
P1: Binod
July 5, 2007 16:10 C4174 C4174C010
X(xi, i)
X , = X
yi = i
xi i
X(xi, ) + Y(yi, i) = xi yi X(xi, ) + Y(xi, i) = ii
Y( yi, i) V(xi, i)
Y , = Y V , = V
xi = i yi = i
yi i xi i
Y(yi, i) + W(yi, i) = ii V(xi, i) + W(yi, i) = xi yi
W(yi, i)
W W
x i = , i =
yi i
(a)
X(xi, i)
X, = X
yi = i
xi i
X(xi, i) + Y( yi , i ) = xi yi X(xi, i) + V(xi, i ) = i i
Y( yi , i) V(xi, i )
Y , = Y
xi = i yi = V , i = V
yi i xi i
Y( yi , i) + W( yi , i ) = i i V(xi, i ) + W( yi , i ) = xi yi
W( yi , i )
xi = W , i = W
yi i
(b)
FIGURE 10.1
(a) A quartet of deterministic partial Legendre transformations for pairs xi yi and i i
for the functional X (xi , i ). (After Sewell (1987). (b) A quartet of ensemble averaged, partial
Legendre transformations for pairs xi yi and i i , when the pair (xi , i ) is controllable
and the functional X (xi , i ) is random.
For a homogeneous continuum the complementary energy under a pre-

scribed traction boundary condition and the potential energy under a pre-
scribed displacement boundary condition are the negative of each other:
U P = U C . (10.43)
P1: Binod
July 5, 2007 16:10 C4174 C4174C010
We also have this classical Legendre transformation linking the Helmholtz

and Gibbs energies:
A(, ) = G(, ) + : . (10.44)
Here we simply write A, G, , and . This, of course, is one pair out of all four
possible Legendre transformations in a quartet linking the internal energy,
Helmholtz free energy, enthalpy, and Gibbs energy when the temperature ()
and entropy (s) are kept as passive variables. In a random medium (i.e., for
the SVE B () B ), the quartet must be reinterpreted carefully according as
either uniform displacement or traction boundary conditions are applied. In
the case of the first one of these loadings, we have the Helmholtz energy
1 1 2
A (ij0 , , ) = Cijkl, ()ij0 kl
0
+ ij, ()ij0 c v, () , (10.45)
2 2 T0
and, in the case of traction loading, we have the Gibbs energy
1 1 2
G (ij0 , , ) = Sijkl, ()ij0 kl0 ij, ()ij c p, () . (10.46)
2 2 T0
Upon ensemble averaging, (10.45) and (10.46) become, respectively,
1 1 2
A (ij0 , ) = Cijkl, ij0 kl
0
+ ij, ij0 c v, , (10.47)
2 2 T0
and
1 1 2
G (ij0 , ) = Sijkl, ij0 kl0 ij, ij0 c p, . (10.48)
2 2 T0
Clearly, under displacement boundary conditions (ij0 controlled), the vol-

ume average stress is random (i.e., ij ()), so that
G ( ij (), ) = A (ij0 , , ) ij ()ij0 , (10.49)
and hence, the ensemble average Gibbs energy on mesoscale should be

calculated from A (ij0 , ) according to
G ( ij , ) = A (ij0 , ) ij ij0 , (10.50)
rather than as G ( ij , ).

Similarly, under traction boundary conditions (ij0 controlled), the volume
average strain is random (i.e., ij ()), so that
A ( ij (), ) = G (ij0 , , ) + ij0 ij (), (10.51)

P1: Binod
July 5, 2007 16:10 C4174 C4174C010
and hence, the ensemble average Helmholtz energy on mesoscale should

be calculated from G (ij0 , ) according to
A (ij , ) = G (ij0 , ) + ij0 ij , (10.52)
rather than as A ( ij , ).

The above relations are used in computation of bounds for thermoelastic
microstructures discussed in the previous chapter. When the mesoscale SVE
reaches the RVE, the dependence on the type of boundary conditions vanishes,
and we recover the classical relation (10.44) for a homogeneous material.
Note: When the orthogonal-mixed boundary condition is employed, we
obtain an intermediate responsevery likely with much weaker scale effects
than the other two.
10.4 Thermodynamic Orthogonality on the Mesoscale

10.4.1 General
The version of thermomechanics we adopt here belongs to the category of
thermodynamics with internal variables (TIV) originated by Ziegler (1957);
see also Germain et al. (1983), Ziegler and Wehrli (1987), Maugin (1999). As is
well known, the RVE response in TIV is described by the free energy and
the dissipation function , both of which are scalar products
1 (q ) e
= : = th + intr = T S(i) 0, (10.53)
2
where the ClausiusDuhem inequality expresses the second law of thermo-
dynamics with a summation of th (the thermal dissipation) and intr (the
intrinsic dissipation); S(i) is the irreversible part of entropy production rate S.
The said scalar products are
th = q T/T intr = Y a = (d) : d + (d) : . (10.54)
In the above (q ) is the quasi-conservative stress, (d) is the dissipative stress,

(d) is the internal dissipative stress, e is the elastic strain, d is the deformation
rate, is the rate of internal parameters , and q is the heat flux.
In view of (10.54), we may simply view as the scalar product
= Y a, (10.55)
where Y is the dissipative force vector and a is the velocity vector (rate of a).
If we want to generalize the formulas relating the dissipative force with
the velocity via functions and , we must recognize that the situation
will be analogous to conservative processes on mesoscale, where we encoun-
tered two types of the functional , and two types of , depending on
the prescribed loading. An ensemble representation of dissipation surfaces
P1: Binod
July 5, 2007 16:10 C4174 C4174C010

a, in the velocity space and its dual Y, in the force space is shown
Figure 10.2; see also Ostoja-Starzewski (2002c). First, we have a generalization
of the extremum principles of deterministic thermomechanics (Ziegler, 1983)
to the mesoscale random medium B = {B (); }.
Velocity control: a is prescribed and Y() follows from the ensemble of
random dissipation surfaces ( a, ) according to Yi () = () ( a, )/a i ;
Figure 10.2(a).
The principle of maximal dissipation rate for B reads: Provided the dissipa-
tive force Y is prescribed, the actual velocity a maximizes the dissipation
rate L (d)
= Y a subject to the side condition

( a) = Y a = L (d)
>0 (10.56)
The principle of least dissipative force for B reads: Provided the value ( a)
of the dissipation function and the direction n of the dissipative force Y are
prescribed, the actual velocity a minimizes the magnitude of Y subject to
the side condition (10.56).
Force control: Y is prescribed and a() follows from the ensemble of ran-
dom dissipation surfaces (Y, ) according to a i () = () (Y, )/Yi ;
Figure 10.2(c).
The principle of maximal dissipation rate reads now: Provided the dissipative
force Y is prescribed, the actual velocity a maximizes the dissipation rate
L (d)
= Y a subject to the side condition

a = Y a = L (d) >0 (10.57)
The principle of least dissipative force for a random medium B reads: Pro-
vided the value (a) of the dissipation function and the direction n of
the dissipative force Y are prescribed, the actual velocity a minimizes the
magnitude of Y subject to the side condition (10.57).
10.4.2 Homogeneous Dissipation Functions

We now work with mesoscale dissipation functions for each body B () B ,
such that:
1. The function depends on a alone, and is star-shaped, convex, and

homogeneous of degree r

a,
a i = r a, ; (10.58)
a i
2. The function is star-shaped, convex and homogeneous of degree

s = 1r

Y,
Yi = s Y, . (10.59)
Y i
P1: Binod
July 5, 2007 16:10 C4174 C4174C010
(a)
Y ( Y )
(a)
(b)
Y = Y
(Y )
a = a (a)
(c)
Y
(Y)
a ( a )
a
FIGURE 10.2
Thermodynamic orthogonality in: (a) spaces of velocities a and ensemble average forces Y
on mesoscale , with Y showing the scatter in Y ; (b) spaces of velocities a a and forces
Y Y on the RVE level, where the scatter is absent; (c) spaces of ensemble average velocities
a and forces Y on mesoscale , with a showing the scatter in a .
P1: Binod
July 5, 2007 16:10 C4174 C4174C010
Note that ( a, ) and (Y, ) are almost surely not inverses of one another
because perfectly homogeneous domains of material carry probability zero
in the space.
The joint dependence of material response on the mesoscale and on the
choice of independent variable (i.e., velocity a or dissipative force Y) leads to
two Legendre transformations for any B () B :
1. Case of a being an independent variable

Y, + a = Y () a (10.60)
2. Case of Y being an independent variable

Y + a, = Y a () . (10.61)
Again, depending on how we take the ensemble averages (Ostoja-Starzewski,

1990), we find

Y + a = Y a (10.62)
and

Y + a = Y a . (10.63)
In the limit, (10.6010.63) become

eff eff
Y + a = Y a. (10.64)
It is of interest to note here that the conventional OnsagerCasimir

reciprocity relations, that is, those that apply to Figure 10.2(b), need to be
reconsidered depending on whether we work in the space of a or Y for finite-
sized bodies in Figures 10.2(a) and (c), respectively. Thus, in the first case
we actually have two choices: when we are either on the surface ( a) of
Figure 10.2(a)

Yi Yj
= (10.65)
a j a i
or on the surface (a) of Figure 10.2(c)
Yi Y
= j . (10.66)
a j a i
When working in the space of Y, we also have two choices: when we are
on the surface (Y) of Figure 10.2(a), we have
a i a j
= (10.67)
Y j Yi
P1: Binod
July 5, 2007 16:10 C4174 C4174C010
while on the surface (Y) of Figure 10.2(c), we have
a i a j
= . (10.68)
Y j Y i
10.4.3 Quasi-Homogeneous Dissipation Functions

A wide class of dissipative processes is described by dissipation functions
( a, ) of quasi-homogeneous type (Ziegler, 1963). Following the general
framework given in Ostoja-Starzewski (1990), we now consider the apparent
behavior to be described by dissipation functions of also quasi-homogeneous
type on mesoscale, so that ( a, ) pertains to a finite-sized body B ():
( a, )
a i = f ( ( a, )), (10.69)
a i
where f is arbitrary. This, of course, implies that the mesoscale dissipation

functions (Y, ) in the space of dissipative forces, Y, are quasi-homo-
geneous too, that is,
(Y, )
Yi = g( (Y, )). (10.70)
Y i
Given the nonuniqueness of the mesoscale response, these two functions

are not perfectly dual of each other. Clearly, we have two alternatives:
1. Assume velocity a to be prescribed (controllable) for the body B (),

the result being Y.
2. Assume Y to be prescribed (controllable) for the body B (), the
result being a.
In the first case, on account of (10.70), for any B () we have
( a, ) ( a, )
Yi () = (10.71)
f ( ( a, )) a i
If for every B () we define a function ( a, ) from ( a, ) by

( a, ) = d (10.72)
f ( )
and let the additional constant in (10.72) be fixed by setting ( ( a, ) = 0)

= 0, upon ensemble averaging, we obtain

( a, ) ( a, )
Yi = = . (10.73)
a i a i
Turning now to the space of dissipative forces, we may proceed in an

analogous fashion. That is, we may either consider a random dissipation
P1: Binod
July 5, 2007 16:10 C4174 C4174C010
function (Y, ) in the space of controllable forces resulting in a random a,

or a deterministic (Y) in the space of ensemble averaged forces Y such
that

( Y )
a i = (10.74)
Yi
Relevant to our analysis leading to (10.74) is the latter situation. On account

of (10.70) the connection between a and Y reduces to

( Y ) ( Y ) ( Y )
a i = = (10.75)
g( ( Y )) Yi Yi
where
1
( Y )
= . (10.76)
Yi
If we now define a function (Y) from (Y) by

( Y ) = d (10.77)
g( )
and let ( (Y) = 0) = 0, we can write, instead of (10.75)

( Y )
a i = (10.78)
Yi
whereby

( a = 0) = 0 ( Y ) = 0) = 0. (10.79)
We will now consider two curves: C in velocity space and its image C in
force space. Curve C connects the origin O with a point P with coordinates
a, while C connects the origin O with the image P of P having coordinates
Y. Thus, we have (in index notation)

Y d a + a d Y = d Y a = Y a. (10.80)
C C C
In light of (10.73), (10.78) and (10.79), this leads to a Legendre transformation

corresponding to case (1)

( a) + ( Y ) = Y a = ( Y ). (10.81)
An analogous analysis for case (2) results in a very similar Legendre trans-
formation (duality between the results in the velocity space and those in the
force space)

( a ) + (Y) = Y a = ( a ) (10.82)
P1: Binod
July 5, 2007 16:10 C4174 C4174C010
where

( a )
Yi = (10.83)
a i
and

(Y)
a i = . (10.84)
Y i
The functions (a) and (Y) in the above are defined by

( a ) = d ( a ) (10.85)
f ( )
and, for every

B (), (Y, ) = d (Y, ). (10.86)
g( )
In both cases above we note the feature already made transparent earlier:
when the ensemble average acts on the argument of a given functional, it acts
on the dual functional itself.
10.5 Complex versus Compound Processes:

The Scaling Viewpoint
10.5.1 General Considerations
According to Ziegler (1983) the dissipative processes appearing in the
ClausiusDuhem inequality may be classified as either elementary, com-
pound, or complex. The first of these is characterized by a single coherent
set of velocities a, and governed by the thermodynamic orthogonality. The
compound case (exemplified by the well-known split into mechanical and
thermal processes) is the situation where each of the constituent, elementary
processes (s = 1, . . ., S) is governed separately by its own thermodynamic
orthogonalities
1
(s) (s)
S
Yk(d s) = (s) (s) (s)
= (s)
a (s)
(s) i
= (s) . (10.87)
a i a i s=1
Here the last equality gives the dissipation function of the entire process in
terms of those of elementary processes.
On the other hand, the complex case is the situation where all the
constituent, elementary processes are governed jointly by a single
P1: Binod
July 5, 2007 16:10 C4174 C4174C010
thermodynamic orthogonality
1
s)
Yk(d = (s)
= a (s)
(s) i
, (10.88)
a i a i
whereby there is no splitting of into dissipation functions of elementary

processes.
Suppose we now consider an unrestricted micropolar medium with dis-
placement u and rotation being independent degrees of freedom of each
continuum particle. (The case of a restricted [or couple-stress] medium is
not interesting because, as discussed in Chapter 6, would then be given
in terms of u.) The dissipation is involved in the processes described by
velocities u as well as . On a continuum level, we have two sets of conjugate
variables:

velocity gradient l = vi, j dissipative force stress (d)
.
curvature rate = dissipative couple-stress (d)
i, j
(10.89)
Thus, the question to ask is this: How do we decide whether it is a compound

or a complex process?
For guidance, given the ubiquity of helical structures in biomaterials, let us
now consider a model chiral material made of helices (h) placed randomly in
a matrix (m), Figure 10.3. Furthermore, both the helix and matrix material are
of Cauchy type. There are three sources of dissipation: inelasticity of helices,
FIGURE 10.3
A realization B () of a chiral material made of helices randomly placed in a matrix.
P1: Binod
July 5, 2007 16:10 C4174 C4174C010
inelasticity of the matrix and imperfect interfaces between helices and matrix.
Observe:
On the level of the material making up each helix or matrix:
(symmetric) deformation rate d and symmetric Cauchy stress (d)
are linking for either phase (s = h or m).
1
(d p) ( p) ( p)
ij = = ( p) dij
dij dij

= (h)
d (h)
+ (m)
d(h) . (10.90)
This a compound process.

On the level of a single helix (as if it were isolated from the matrix)
there is a coupling of axial with torsional dissipative actions just
like that of an elastic helix (along its local axis x). Thus, we have a
complex process:
F (d) (h) M(d) (h)

= = , (10.91)
AE u, x E R3 , x
where (h) denotes the dissipation function of the helix. Note that
the constitutive equations in the viscoelastic case are in the category
of a complex process, Section 3.5.3 of Chapter 3.
On the microscale ( 1) through mesoscale:

ij(d) = (d)
ij = , (10.92)
vi, j i, j
where, in principle, mesoscale bounds in the sense of the preceding

section should be used. This would involve ( a, ) and (Y, )
for B() B.
On the macroscale (where is sufficiently large to reach RVE) the
micropolar effects become negligible, so that equation (10.92) is
replaced be a relation between the dissipative Cauchy stress (d)
and the deformation rate d:
1
(d)
ij = = dij , (10.93)
dij dij
where is responsible for the dissipation on all the lower scales.

We conclude:
1. The length scale is a factor deciding between the compound or com-
plex nature of the dissipative process.
2. Even if the helical geometry of inclusions were removed, and we
simply dealt with, say, classical ellipsoidal-shaped inclusions, by
P1: Binod
July 5, 2007 16:10 C4174 C4174C010
the argument of Section 6.5 of Chapter 6 extended to inelastic me-

dia, there would be micropolar effects on mesoscales from 1
and up. The corresponding dissipative process would be complex,
but its micropolar effects would vanish in passage to very large
scales .
10.5.2 Micropolar Plasticity

As another example, let us focus on two possible extensions of the J 2 flow
theory to micropolar materials (Muhlhaus, 1995; Forest and Sievert, 2003).
Following these authors, start from the total strain rate ij as the sum of the
elastic strain rate gij and the plastic strain rate ij , with an analogous split of
the total curvature rate ij into the elastic (kij ) and plastic (ij ) parts
ij = gij + ij ij = kij + ij , (10.94)
On one hand, one may formulate an extension of the Hubervon Mises

criterion via a single equation
f (ij , ij ) = J 2 ( ji , ji ) Y

J 2 (ij , ij ) = a 1 sij sij + a 2 sij s ji + b 1 ij ji + b 2 ij ji , (10.95)
where sij is the stress deviator of ij , while Y, a 1 , a 2 , b 1 and b 2 are material

parameters. The plastic rates follow this flow rule
f (ij , ij ) f (ij ,ij )

ij = ij = ij
. (10.96)
ij
On the other hand, one may set up a multicriterion model
f (ij ) = J 2 ( ji ) Y f (ij ) = J 2 (ij ) Y

(10.97)
! !
J 2 ( ji ) = a 1 sij sij + a 2 sij s ji J 2 ( ji ) = b 1 ij ji + b 2 ij ji ,
whereby the flow rule is
f (ij ) f (ij )
ij = ij = . (10.98)
ij ij
In view of the remarks immediately preceding this section, the model

(10.95 and 10.96) is a complex process, and as such is better justified than
(10.97 and 10.98), which is a compound process.
P1: Binod
July 5, 2007 16:10 C4174 C4174C010
10.6 Toward Continuum Mechanics of Fractal Media

This is the only place in the book where we admit fractal geometries of mate-
rials. While much has been said and written over the past three decades about
mathematics and physics of fractals, field equations in the vein of continuum
mechanics have been lacking. However, the recent work of Tarasov (2005a,b)
allows progress in that direction. We start by relaxing the assumption (1.109)
of Chapter 1: the properties of fractal media like mass m obey a power law
relation
m ( R) = k R D D < 3, (10.99)
where R is a box size (or a sphere radius), D is a fractal dimension of mass, and
k is a proportionality constant. It follows that the power law (10.99) describes
the scaling of mass with R.
Note: In this section we depart from the convention used in the book of D
indicating the physical dimension. That is, d shall denote the latter, while D
is now reserved for the fractal dimension.
Focusing on porous media, the power law relation (10.99) is rewritten as
D
R
m D ( R) = m0 , (10.100)
Rp
where R p is the average radius of a pore, and m0 is the mass at R p = R; this
is a reference case.
At this point, the conventional equation giving mass in a 3D region W of
volume V

m (W) = (r) d 3 r (10.101)
W
has to be generalized

23D (3/2)
m3d (W) = (r) |r r0 | D3 d 3 r, (10.102)
( D/2) W
where D is the measure in the 3D space.

Assuming the fractal medium to be spatially homogeneous
(r) = 0 = const, (10.103)
the equation (10.102) is replaced using a fractional integral

23D (3/2)
m3d (W) = 0 |R| D3 d 3 r, (10.104)
( D/2) W
where R = r r0 . That is, the fractal medium with a noninteger mass

dimension D is described using a fractional integral of order D. This
allows an interpretation of the fractal (intrinsically discontinuous) medium
P1: Binod
July 5, 2007 16:10 C4174 C4174C010
as a continuum. In particular, the next step is Tarasovs reformulation of the

Green-Gauss Theorem

A nk dSd = c 31 ( D, R) div(c 2 (d, R) Av)dV D, (10.105)
w W
Where A is an arbitrary function, v is the velocity, and
dSd = c 2 (d, R)dS2 dV D = c 3 ( D, R)dV 3 (10.106)
On account of (2.9), the " left hand side in (2.8) is a fractional integral, equal to a
conventional integral W c 2 (d, R) AvdS2 . Similarly, the"right hand side in (2.8)
is a fractional integral, equal to a conventional integral W div(c 2 (d, R) Av)dV 3 .
See also (2.14) below.
The above formulation allows the derivation of fractional-type balance
equations of fractal media:
the fractional equation of continuity:

d
= kD k , (10.107)
dt D
the fractional equation of balance of density of momentum:

d
k = f k + lD kl , (10.108)
dt D
the fractional equation of balance of density of energy:

d
u = c( D, d, R)kl k,l kD q k . (10.109)
dt D
In the above kl is the Cauchy stress (symmetric according to the balance
of angular momentum, employed just like in non-fractal media), and the
following operators (or, generalized derivatives) are used

kD A = c 3 ( D, R) [c 2 (d, R) A] c 3 ( D, R)k [c 2 (d, R) A] ,
xk

d A A
A= + c ( D, d, R) k , (10.110)
dt D t xk
where
2 Dd1 ( D/2)
c( D, d, R) = |R|d+1D
(3/2)(d/2)
22d
c 2 (d, R) = |R| D3
(d/2)
23D (3/2)
c 3 ( D, R) = |R|2d
( D/2)
c( D, d, R) = c 31 ( D, R)c 2 (d, R). (10.111)
P1: Binod
July 5, 2007 16:10 C4174 C4174C010
Henceforth, for simplicity of notation, we write c, c 2 , and c 3 . Note that, in

a non-fractal medium (D = 3, d = 2) c (D, d, R) = 1, whereby one recovers
conventional forms of local relations of continuum mechanics. Of further use
will also be the fact that the relation kD AB = AkD B + BkD A does not
hold, and should instead be replaced by
kD AB = AkD B + c Bk A. (10.112)
Proceeding in this vein, the Clausius-Duhem inequality for fractal media

takes the form

d d d T, k q k
0 T s (i) = i(d)
j u(i , j) + (d)
ij i j c( D, d, R) ,
dt D dt D dt D T
(10.113)
where

d
lim u(i , j) = (i j) di j , (10.114)
D3 dt D
which is just the deformation rate. Also, for small strains di j is (d/dt) D i j,
with an analogous limit

d
lim i j i j . (10.115)
D3 dt D
It is most interesting that generalized derivatives appear only for the time
rates of external and internal strains but do not arise in the third term in
(10.113) except for the coefficient c( D, d, R).
One can now identify the term T (d/dt) D s (i) as the dissipation function in
three velocity-like arguments { [(d/dt) D u(i ], j), (d/dt) D i j, q k } and a number
of relations of Zieglers theory, including the thermodynamic orthogonality,
carry over to fractals. We can also generalize the Hill condition to quasi-static
loadings, and on account of previous relations, we have
c : d = : d c : d = 0. (10.116)
Adapting the same route as in Section 7.2.2.1, we find the necessary and
sufficient condition for (10.116)2 to be

(t n) (v d x) c 2 d A2 = 0. (10.117)
W
which dictates possible loadings on the boundary of a fractal body.

Observe that Zieglers thermomechanics is a formalism very suitable for
generalization to random media precisely because it allows scale-dependent
homogenization in the vein of Hill condition where (i) either the applied strain
P1: Binod
July 5, 2007 16:10 C4174 C4174C010
or the applied stress are on an equal footing, and (ii) the energy or power
of dissipation is the key criterion for equivalence between a heterogeneous
structure and a smoothing continuum. By contrast, the rational thermome-
chanics of Truesdell does not jibe with Hills condition because there the stress
is taken as a primary quantity while the energy as a secondary one (Ball &
James, 2002).
P1: Binod
July 5, 2007 16:10 C4174 C4174C010
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
11
Waves and Wavefronts in Random Media
To deal with thermodynamics is to look for trouble.

G.A. Maugin, 1999
Heeding the above warning, we now switch to a field where thermodynam-
ics is not of central importance: waves in random media, also called stochastic
wave propagation. That research activity, fast growing since the middle of the
twentieth century, has primarily been motivated by various problems arising
in acoustics, atmospheric physics, geophysics, and composite materials, see
the reviews and books of Chernov (1960), Frisch (1968), Dence and Spence
(1970), Uscinski (1977), Sobczyk (1985, 1986), Rytov et al. (1987), Papanicolaou
(1998). Mathematical problems in all these applications have typically been
set up as ordinary of partial differential equations on random fields with
either discrete or continuous realizations. A key characteristic of random
fields has been the correlation length, and most studies have focused on
the most tractable situations of wavelengths being either much smaller or
much larger than the typical size of heterogeneity; recall Section 2.2 in Chap-
ter 2. Generally speaking, in stochastic wave propagation we must have three
length scales: (1) the typical propagation distances L macro ; (2) the typical wave-
length or wavefront thickness L; and (3) the typical size of inhomogeneity d.
Most studies have focused on linear elastic waves, and in this chapter we
only give a very brief introduction to these topics, which are expertly covered
in the classical references listed above, and then consider less conventional
problems. Thus, in Section 11.1 we first discuss two basic cases of wavelengths
being either much longer or much shorter than the heterogeneity, and in the
latter case we reconsider random geometric acoustics when, in contradistinc-
tion to the common assumption, the elastic medium is not necessarily locally
isotropic. In Section 11.2 we introduce the concept of stochastic spectral finite
elements, which provides a setting for analyses of steady-state vibrations in
random structural elements, without any assumptions on separation of scales
between L macro , , and d, but requires a numerical solution for obtaining quan-
titative results. Section 11.3, while still set in the harmonic regime, focuses
on a surprisingly unconventional behavior of disordered one-dimensional
(1D) composites; the presentation outlines a little-known stochastic homo-
genization technique (random evolutions) that may also be useful in other
385
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
mechanics problems. Section 11.4 is devoted to transient waves in piecewise

constant random media with linear or nonlinear elastic, and then elastic-
dissipative, responses. It is shown through a generalization of the method of
characteristics to random media setting, that even weak material randomness
may strongly affect conventional solutions of homogeneous media. Finally,
Section 11.5 examines the evolution of acceleration wavefronts whose thick-
ness is not infinitesimal as conventionally done in the singular surface wave
analysis (e.g., Chen, 1976), but finite. This aspect forces us to replace the classi-
cal RVE of deterministic continuum mechanics by the mesoscale SVE, leading
us to uncover various characteristics of blow-up of acceleration wavefronts
in random media.
11.1 Basic Methods in Stochastic Wave Propagation

11.1.1 The Long Wavelength Case
11.1.1.1 Elementary Considerations
The starting point in classical analyses of wave propagation in random media
is offered by the wave equation for a scalar field u in a domain X
1 2 u
2u = , , x X. (11.1)
c 2 (, x) t 2
Here c(, x) is a random field, that is, an ensemble {c(, x), , x X }.

Formally speaking, we have a triple (, F,P), where is the space of ele-
mentary events, F is its -algebra, and P is the probability measure defined
on it. We sometimes write explicitly to show the random character of a
given quantity like c(, x), or else, we suppress it for clarity of notation.
Assuming harmonic time dependence (e i t ), we obtain the scalar stochastic
Helmholtz equation
2 u + k02 n2 (, x)u = 0, , x X, (11.2)
where, writing k02 n2 (, x) for k(, x), we introduce a random wave number
to deal with the spatial randomness of the medium. Thus, k0 = /c 0 is the
wave number of a reference homogeneous medium where c 0 is its phase velocity,
and n(x, ) is a random index of refraction. Hereinafter, we employ for the
frequency, rather than the conventional , which has been reserved to denote
an outcome (i.e., a random mediums realization) from the sample space .
Equation (11.2) is a valid ansatz whenever the time variation in the re-
fractive properties of the medium is much slower than the wave propagation
itself; thus, for example, swirling as rapid as the wave motion violates the
monochromaticity assumption. The random field {n(, x), , x X } is de-
termined from experimental measurements. At this point, it is convenient to
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
Waves and Wavefronts in Random Media 387
consider the model
n2 (, x) = 1 + (, x) (, x) = 0 = O (1) , (11.3)
so that all the randomness is present in the zero-mean random field .

The key role is played by a correlation function of , which, given (11.3)2 , is
C (x, x ) = (x)(x ). (11.4)
Usually, is taken as a wide-sense stationary random field
C (x, x ) = C (x x ) < , x, x , (11.5)
possessing an ergodic property almost surely in and X
(x) = (). (11.6)
Rytov et al. (1987) also discuss more general random field models such as,
say, those with stationary increments.
A special class of so-called isotropic random fields occurs when (x) depends
only on the magnitude, but not direction, of the vector x

C (x) = C (r ), r = |x| = xi xi . (11.7)
A very common model for the correlation coefficient is the Gaussian form (a
special case of (2.89))

C (r ) = 2 exp[r 2 /a 2 ], (11.8)
where a is the so-called correlation radius.

Determination of the random field {(, x), , x X } is made on the
basis of experimental measurements. As Chernov (1960) shows, this function
may present a very good fit,

C (r ) = 2 exp[r/a ], (11.9)
but one must bear in mind that it corresponds to random fields with dis-
continuous, rather than continuous, realizations. Indeed, Hudson (1968) took
precisely this form to model scattering in a granular/cellular structure of a
polycrystal.
11.1.1.2 Series Expansion

Returning to the stochastic Helmholtz equation, one takes 1, and con-
siders its solutions in the form of an expansion with respect to the powers
of
u(, x) = u0 (x) + u1 (, x) + 2 u2 (, x) + (11.10)
Note that the Russian school (Chernov, 1960; Rytov et al., 1987) takes n(, x) =
1 + (, x) in place of (11.3)1 .
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
Upon substituting (11.3)1 and (11.10) into (11.1) and equating terms of the
same order in , one obtains a system of recurrence equations for u0 , u1 , u2 ,
and so on
2 u0 + k02 u0 = f (x)
2 u1 + k02 u1 = k02 (, x)u0 (x) (11.11)
2 u2 + k02 u2 = k02 (, x)u1 (x).
It is seen that the solution u0 to a homogeneous medium problem serves as

forcing to the first correction u1 , which then drives the third equation in (11.11)
governing u2 , and so on. Thus, the perturbation approach reflects a multiple
scattering nature of stochastic wave propagation.
Using Greens function for a free space Helmholtz equation of the homo-
geneous medium
exp[ik0 |x x |]
G 0 x, x = , (11.12)
4 |x x |
the solutions to equations (11.11) may be calculated from

u0 (x) = G 0 (x, x ) f x dx

u1 (x) = k02 G 0 x, x , x u0 x dx

u2 (x) = k04 G 0 x, x G 0 x , x , x , x u0 x dx dx .
(11.13)
Five aspects are important with respect to (11.13):

1. The above solution is nonlinear in the boundary conditions.
2. The ensemble average of u1 (x) is zero.
3. Solution ui (x) represents the perturbation of the
original
wave field
ui1 (x) caused by the inhomogeneity field , x , i = 1, 2, . . ..
This characterizes stochastic wave propagation as a successive
multiple scattering.
4. The above solution may be obtained using an integral equation
formulation. First, write the solution to (11.2) as

u1 (x) = u0 (x) k0 G 0 (x, x )(, x )u x dx
2
(11.14)
and iterate to obtain the perturbation expansion. To find the first

iteration of (11.14) we write the value of the field at x = x

u1 x = u0 x k02 G 0 (x , x )(, x )u x dx (11.15)
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
and substitute this into the right-hand side of (11.14) so as to get

u1 (x) = u0 (x) k0 G 0 (x, x )(, x )u x dx
2

+k04 G 0 (x, x )(, x )u x dx G 0 (x , x )(, x )u x dx .
(11.16)
To find the second iteration we write the value of the field at x = x
and substitute the result into the right-hand side of (11.16). This
leads to an infinite perturbation (or Neumann) series, for the integral
equation (11.14), the first three terms of which are given in (11.13).
Following Frisch (1968), it can be shown that this series convergence
requires
Mk02 D2 /2 < 1, (11.17)
where D is the diameter of the scattering region, and | ()| < M

almost surely.
5. A particular case of (11.13) occurs if one considers single scattering
only by assuming
u (, x) = u0 (x) + u1 (, x) , (11.18)
rather than (11.10)this is called (first) Born approximation. It is pop-

ular in applications as it simplifies the analysis, but it imposes strong
restrictions on the validity of this perturbation approach:
(a) The noise-to-signal ratio must be low: 1 in (11.3).
(b) The size of inhomogeneity must be much smaller than the
wavelength: d .
Given the particular form (11.8) of the correlation function, this would be
equivalent to stating that the correlation length needs to be much smaller than
the wavelength: d . In fact, in that case, under the assumption that the
dimensions of the scattering domain V are much larger than the correlation
radius a , the intensity of the scattered field is computed as

k04 a 3 V 2 exp k02 a 2 sin2 /2
I |u1 ( P)| 2
= . (11.19)
16 R02
Here R0 is the distance from the scattering region to the observation point
P, and is the scattering anglethe angle between the wave vector ki of the
incident wave and the wave vector ks of the scattered field.
The above result may be used to obtain various physically interesting
parameters. For example, considering that the scattered energy in a region
of size L = V 1/3 must be much smaller than the total energy, we obtain the
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
condition
1
2 k 0 L . (11.20)
k0 a 1 e k0 a
2 2
Because k0 a = 2a /, this shows that the correlation radius a must go down

as the domain size L goes up, for the Born approximation to be valid.
11.1.2 The Short Wavelength Case: Ray Method

11.1.2.1 Fermats Principle
It is well known that the case of short wavelengths falls into the realm of geo-
metric acoustics (or optics) in which all the disturbances propagate according
to Fermats principle
t1
dt = min . (11.21)
t0
Here t0 and t1 denote the initial and final instants on a path from points P0 (x)
to P1 (x). Assuming the medium is pointwise (i.e., locally) isotropic, we have
c = c(x), and thence follows a description by a field of circular indicatrices
of Figure 11.1(a). Using the EulerLagrange equations, in a d-dimensional
setting, one obtains the well-known equations of ray dynamics (e.g., Hudson,
1980)

dxi dyi 1
= cyi = i = 1, . . ., d, (11.22)
ds ds xi c
where s is the arc length along the ray x(s), propagating in direction y(s) at a
local speed c(x). Associated with (11.22) there is an eikonal equation: (11.31)
below.
All of the above involves the assumption of a locally isotropic, inhomoge-
neous medium, possibly of a spatially random character, and so c(x) should
be interpreted as c(, x). However, a spatial gradient of smooth elastic moduli
suggests that a constitutive response in one principal direction is very likely
different from that in another. Therefore, turning to a medium of anisotropic
type, we should have c(x, x), and, corresponding to it, a pointwise description
by a field of nonspherical indicatrices of Figure 11.1(b); see also Nye (1957).
Here xi = xi /l, l being a parameter of the path. Indeed, in taking c(x, x)
we are motivated here by the discussion of Huygens principle in Arnold
(1978). The equations governing the dynamics of a raythat is, its position x
and direction y (= c 1 dx/ds)are now found by generalizing the variational
procedure leading from (11.21) to (11.22) (Ostoja-Starzewski, 2001c). Thus, for

t1 t1
ds t1
dl
dt = = = min, (11.23)
t0 t0 c t0 c
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
y p
(a)
y
p
(b)
FIGURE 11.1
Indicatrix envelopes in two realizations of a random medium: (a) locally isotropic; (b) locally
anisotropic; see also Arnold (1978). In both cases, the direction of the wavefronts motion p and
the direction of the ray y are shown.
we identify the Lagrangian as
1
L= = [xi xi ]1/2 i = 1, . . ., d, (11.24)
c(x, x) c(x, x)
where the ray x(s) propagates at a local speed c(x, x).

From the Euler-Lagrange equations, with s (or l) playing the role of time,
we now find a dynamical system of rays for an anisotropic medium

dxi d c 1
= cyi ( yi 2 )= i = 1, . . ., d (11.25)
ds ds c xi xi c
in place of (11.22) in the isotropic case.

Let us now consider the SH wave motion of a generally anisotropic medium
of local elastic property Cij (x) Cikjm (x); i, j = 1, 2; k, m = 3. Clearly, the
relevant governing equation is then
(Cij u, j ) ,i = u. (11.26)
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
For isotropy (Cij (x) C 0 (x)ij ), (11.26) reduces to the classical wave equation
u,ii = u/c 2 , with c(x) = [C0 (x)] / being the local wave speed.
Let us now assume a trial solution of the form
u(x, t) = A(x) exp [i[0 S(x) t]] , (11.27)
where 0 = /c 0 = 2/0 is the wave number in the reference homogeneous

medium (c(x) = c 0 = const), and A(x) and S(x) are to be determined. The
latter two quantities are supposed real as there is no dissipation.
Upon substituting (11.27) into (11.26), we find equations governing,
respectively, the real part
Cij A,ij 02 ACij S,i S, j A,i Cij,j = A2 , (11.28)
and the imaginary part
2Cij A,i S, j + ACij S,ij + ACij,i S,i = 0. (11.29)
As a first approximation for a slowly varying medium, we may ignore the

second derivatives of A as well as the products of first derivatives of Cij with
those of A in (11.28), obtaining the eikonal equation for an anisotropic medium
02 Cij S,i S, j = 2 , (11.30)
which, for Cij (x) C0 (x)ij , reduces to the classical eikonal equation for an
isotropic medium
S,i S,i = n2 or | S|2 = n2 . (11.31)
Here n(x) c 0 /c(x) is the refractive index.

Similarly, ignoring the products of first derivatives of Cij with those of S
in (11.29), we find
2Cij A,i S, j + ACij S,ij = 0, (11.32)
which, for Cij (x) C0 (x)ij , reduces to the classical form (e.g., Elmore and
Heald, 1969)
2A, j S, j + AS,ii = 0 or 2 A S + A 2 S = 0. (11.33)
A stepping-stone to extending this analysis to general elastodynamics is

in Epstein and Sniatycki (1992).
11.1.2.2 Markov Character of Rays

Already in the simplest case when c is a locally isotropic random fieldthat
is, when we have equation (11.22)there arose various analytical problems
and approximations. First, Kharanen (1953) and Chernov (1960) treated the
problem in a dishonest way; that is, they assumed that the direction of
the ray y(s) was Markov in s. This assumption enabled them to formulate a
FokkerPlanck equation in order to find the probability density of the rays
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
p(y; s). On the other hand, Keller (1962) employed a perturbation expansion
to obtain results valid only for short ray paths; this was an honest method in
that no unjustified probabilistic assumptions were made a priori.
An intuitive justification of the Markov property can be provided by an
analogy of evolution of y(s) to the evolution of the velocity of a Brownian
particle that has suffered many independent collisions. Thus, the dishonest
methods are good for long ray paths only. A solution method valid for all ray
paths has been developed by Frisch (1968), who, using a first-order pertur-
bation, cast the system (11.22) for a locally isotropic medium, in the form of
a stochastic Liouville equation for the rays, which led to a kinetic equation
for random geometric optics. Motivated by the limitation of the Markovian
assumption, Perez et al. (2004) introduced a fractional Brownian motion model
so as to account for memory along the rays paths.
Another line of approach to the problem of evolution of stochastic rays in
a random medium has been pursued by Brandstatter and Schoenberg (1975),
whose simplified model was
dx = c (x) edt + g (x) e = cy, (11.34)
where g(x) is a random noise with zero mean, finite covariance and all mo-
ments higher than the second being on the order O(t), e is the unit vector. The
effort of mathematical analysis, in the framework of dynamic programming,
was on finding a policy e (x) such that (11.21) obtains in the ensemble average
sense. Here we observe that, substituting dt/ds = 1/c, (11.22) becomes
dx = c 2 (x) ydt dy = c (x) ydt. (11.35)
Clearly, (11.35)1 or, equivalently, dx = c (x) edtstands in stark contrast to

(11.25), and we conclude that the approach of Kharanen (1953) consisted in
assuming the Markov character for the y variable according to the dynamics of
(11.35)2 without taking account of (11.35)1 , while the approach of Brandstatter
and Schoenberg (1975) consisted in replacing the random system (11.35) by an
equation, with an additive, rather than multiplicative, noise for the position
of the mean ray.
Our equation (11.25) shows that the Markov character of {x, y}s is pre-
served in a locally anisotropic medium, but an explicit solution for the statis-
tics of rays will be more challenging than in the isotropic case. In fact, such
a solution via Ito calculus appears unwieldy, and a recourse to a numerical
solution of the stochastic dynamical system is necessary.
11.1.3 The Short Wavelength Case: Rytov Method

While the Born method is based on the expansion of the wave field u with
respect to a small parameter , the Rytov method relies on the expansion of
ln u. First, the solution to the stochastic Helmholtz equation without forcing
(11.2), with (11.3) as before, is represented as
u (, x) = A0 exp [i (, x)]. (11.36)

P1: Binod
May 28, 2007 18:15 C4174 C4174C011
Here A0 is the wave amplitude in the reference homogeneous medium. Sub-

stituting (11.36) into (11.2), we obtain a nonlinear equation
( ) 2 + i 2 = k02 n2 (, x). (11.37)
If we look for its solution in the form of an expansion
(, x) = 0 (x) + 1 (, x) + 2 2 (, x) + , (11.38)
again under the assumption (11.3), we obtain a recurrent sequence of equa-

tions for 0 , 1 , 2 , and so on. The simplest situation results if we restrict
ourselves to the first two terms of (11.38), (, x) = 0 (x) + 1 (, x), in
which case we arrive at
( 0 ) 2 + i 2 0 = k02
2 ( 0 ) ( 1 ) + i 2 1 = k02 (, x) ( 1 ) 2 . (11.39)
Equation (11.39)1 has a plane-wave solution 0 = k0 x (x x1 ), so that

(11.39)2 becomes
1
2k0 + i 2 1 = k02 (, x) ( 1 ) 2 . (11.40)
x
Using an auxiliary function 1 (x), defined via 1 (x) = 1 (x) e ik0 x , we find
an inhomogeneous Helmholtz equation
2 1 + k02 1 = 2ik02 (, x) e ik0 x . (11.41)
whose solution is

ik02 1 ik0 ( R+ )
1 = e (, , ) d dd. (11.42)
2 R
where R = [(x ) 2 + ( y ) 2 + (z ) 2 ]1/2 is the distance between the

observation point at r and the scattering element at (, , ). This results in

ik02 1 ik0 [R(x )]
1 (, r ) = e (, , ) d dd, (11.43)
2 R
which shows the advantage of the Rytov method: the random wave field
1 (, r ) is expressed as a linear transformation of a given random field
(, r ), and hence, the moments of (, r ) = 0 (r ) + 1 (, r ) can be com-
puted by averaging over and integrating over the physical space.
The Rytov method and the older method of geometric optics apply to
situations in which the size of inhomogeneity is much larger than the wave-
length: d . However, since the geometric optics is also limited by a con-
dition on the length L of the path propagated by the ray, that is, d L,
the Rytov method offers a more powerful avenue. These and other related
issues are discussed at length and in depth in the references listed in this
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
section. See also Sections 46 in a recent review of methods for stochas-

tic Helmholtz equation (Beran, 2001). In particular, he discusses single and
multiple scattering problems, and the wave transmission problem in a 1D ran-
dom medium with back-scattering fully accounted for. Two reviews of elasto-
dynamics of random media, especially from the standpoint of variational
principles, are given in Willis (1997, 2001).
11.2 Toward Spectral Finite Elements for Random Media

11.2.1 Spectral Finite Element for Waves in Rods
11.2.1.1 Deterministic Case
The classical approach via stochastic Helmholtz equation (11.2) grasps,
through the random field of refraction index, the spatial variability in the
mass density but not the variability in elastic moduli or geometric parame-
ters. This is immediately seen by considering the equation governing axial
motions in a rod with a space-dependent mass density , elastic modulus E,
and cross-sectional area A, namely,

2
A(x, ) E(x, ) u(x, t) = (x, ) A(x, ) 2 u(x, t), . (11.44)
x x t
In order to remove the aforementioned restrictionsat least in this 1D

modeland to analyze the relative effects of spatial randomness of mass
density, elastic properties, as well as cross-sectional geometric properties, we
may consider stationary responses of rods in longitudinal vibrations, and of
Timoshenko beams in flexural vibrations. This naturally leads to a spectral
approach, in which, given the randomness of a rod or a beam, we seek a
stochastic spectral finite element (Ostoja-Starzewski and Woods, 2003), which
presents a generalization of a spectral finite element (e.g., Doyle, 1997).
In principle, techniques such as those presented in Section 11.1 could be
employed to tackle equation (11.44). However, the analysis would have to
be restricted to the situation where the separation of scales would at least
approximately be satisfied, that is, where either d or d holds.
For the sake of reference, let us first recall basic concepts of deterministic
spectral finite elements through a paradigm of a rod made of a homogeneous
material. The elastodynamic equation governing the axial response of a rod
(assuming zero external forcing) is well known:
2 u 1 2 u
= . (11.45)
x2 c a2 t 2

Here c a = E/ denotes the phase velocity of axial waves; E being the
elastic axial modulus and the mass density. Assuming u(x, t) = u(x)e i t ,
the Helmholtz equation corresponding to (11.45) set up over the domain X of
P1: Binod
May 28, 2007 18:15 C4174 C4174C011

u1 u2

F1 F2
FIGURE 11.2
Finite element for a rod in longitudinal motion.
size L, is
d 2
u
+ k2
u = 0,
u(0) =
u1 ,
u(L) =
u2 , (11.46)
dx2
where k = /c a . Then, the spectral matrix expresses a connection between
the kinematic and dynamic quantities, that is, { u1 , 1 , F
u2 } with { F 2 }, at both
ends of the rod (1 and 2) in Figure 11.2, where the hat signifies the quantities
in the frequency space. Considering solutions in the form u( ) = Asin k +
B sin k(L ), with the definitions F1 = F (0) and F 2 = F (L), we readily
find the following spectral matrix

1
F k cot k L k csc k L
u1
= AE . (11.47)
2
F k csc k L k cot k L
u2
This representation demonstrates the purely real nature of the spectral matrix.
In Figure 11.3 we plot the k11 -component of this matrix as a continuous
black line, with the peaks of k cot k L representing the resonant frequencies
of the system with A = 104 m, E = 27.4 GPa and = 2, 400 kg/m3 . These
values correspond to a rod made of concrete. We are now interested in the
change from this crisp functional form, and the associated scatter as we go
to the random rod.
11.2.1.2 Random Case

The frequency space version of the stochastic equation (11.44), again with
Dirichlet boundary conditions, is

d d
u
A(x, ) E(x, ) + (x, ) A(x, )
u(x) = 0,
dx dx

u(0) =
u1 ,
u(L) =
u2 . (11.48)
There are many ways to simulate imperfect microstructures (e.g., Jeulin

and Ostoja-Starzewski, 2001), and some definite choices have to be made in
the case of rods, which themselves are simplified 1D models of 3D random
bodies. Focusing on a random field model of a band-limited type, rather
than on trying to approximate some nice function, we therefore assume
the mass density, elastic modulus, and cross-sectional area to be described by
random Fourier series (Chapter 2) with a typical/average characteristic size
of inhomogeneity d, which is either smaller, comparable to, or larger than the
wavelength. The third length scale entering the problem, but kept constant,
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
(a) 5
2
Log10 11
0
0.0E + 00 1.0E + 04 2.0E + 04 3.0E + 04 4.0E + 04 5.0E + 04 6.0E + 04 7.0E + 04
1
3
Frequency (Hz)
(b) 5
2
Log10 11
0
0.0E + 00 1.0E + 04 2.0E + 04 3.0E + 04 4.0E + 04 5.0E + 04 6.0E + 04 7.0E + 04
1
3
Frequency (Hz)
(c) 5
2
Log10 11
0
0.0E + 00 1.0E + 04 2.0E + 04 3.0E + 04 4.0E + 04 5.0E + 04 6.0E + 04 7.0E + 04
1
3
Frequency (Hz)
FIGURE 11.3
Rod vibrations in the case of random modulus E showing k11 (black line) for the reference
homogeneous medium and < k11 > (gray line) for the random case with: (a) g = 0.1, (b) g = 1.0,
(c) g = 10.0.
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
is the rod length. Thus, for A we have

10

(i) (i)
A(x, ) = A0 1 + A a A () cos igx + b A () sin igx , (11.49)
i=1
with completely analogous models for (x, ) (having coefficients a (i) , b (i) )
and E(x, ) (having coefficients a (i) (i)
E , b E ). Here, for i = 1, . . . , 10,
(i) (i)
a A (), . . . , b E () come from a uniform distribution on [1/2, 1/2] and .
The spectral finite element is now given by a relation

1
F k11 () k12 ()
u1
= , , (11.50)
2
F k21 () k22 ()
u2
but, since we deal with a differential equation with inhomogeneous coeffi-

cients (11.49) in (11.48), the kij s cannot be determined explicitly. In the ensem-
ble setting, in view of the spatial homogeneity of the random field (11.49), the
averages satisfy k11 = k22 .
A numerical method has been developed to obtain the kij s, and a sample
of results for k11 , corresponding to a random E for g = 0.1, 1 and 10, res-
pectively, is shown in Figure 11.3. The deterministic case already discussed
above is shown as a crisp, black line, while the random case is shown as a gray
thicker line, possibly overlapping the first one. Thus, whenever we only see
the gray line, there is no difference between the deterministic and the mean
of the stochastic problem. Here we see a strong departure from the reference
case at and around the resonant frequencies. Note that the scatter interval
increases with the increasing frequency. The effects decrease with increasing
g but, regardless of the value of g, the effects are most noticeable at higher
frequencies. Very similar results are obtained for random mass density , and
for and E randomized together. On the other hand, the effect of random
cross-section A is quite differentbasically, it has influence only at g 1.0
and, in particular, for higher frequencies.
11.2.2 Spectral Finite Element for Flexural Waves

11.2.2.1 Deterministic Case
The dynamics of a Timoshenko beam with spatially inhomogeneous proper-
ties is governed by two coupled equations:

v 2 v
G A = ,
x x t 2

v 2
EI + G A + I 2 = . (11.51)
x x x t 2
Clearly, there are two kinds of wave motion in such a beam: flexural and
rotational; it is a 1D micropolar continuum indeed. The spatial inhomogeneity
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
is modeled again by random Fourier series, although it now involves five

(rather than three) independent parameters appearing in the governing equa-
tions: mass density , elastic modulus E, Poissons ratio , area Aand moment
of inertia I of the cross-section. It appears more physical, however, to work
with the cross-sectional height h and cross-sectional width w instead of the
latter two. Thus, the beam is described by a five component random field
[, E, , h, w].
Specializing to the spatially homogeneous case, the frequency space equa-
tions governing the transverse deflection v(x, t) = v(x)e i t and the trans-
verse shearing deformation w (as measured by the difference v/x , with

(x, t) = (x)e i t
) of a Timoshenko beam, assuming zero external forcing,
are:

d 2
v d
G A 2
+ A 2
v=0
dx dx

d 2 dv = 0.
E I 2 + G A + I 2 (11.52)
dx dx
Here G is the shear modulus, Ais the cross-sectional area, is the shape factor
of the cross-section, is the mass density, E is the elastic modulus, and I is
the cross-sectional area moment of inertia.
The spectral stiffness matrix expresses now a connection between the kine-
matic and the dynamical quantities, that is, { 1 ,
v1 , 1 , M
2 } with {V
v2 , 1, V
2 , M
2 },
at both ends of the beam, Figure 11.4. In the derivation of the spectral stiffness
matrix for the Timoshenko beam equations (11.52) one employs the boundary
conditions

v(0) =
v1 ,
(0) 1
=

v(L) =
v2 ,
(L) 2
= (11.53)
set up over a domain of length L, and considers a solution of the form

v(x) = B1 Rt cos k1 x B2 Rt sin k1 x + C1 Rh cosh k2 x + C2 Rh sinh k2 x

(x) = B1 sin k1 x + B2 cos k1 x + C1 sinh k2 x + C2 cosh k2 x. (11.54)

V1 V2
v1 v2

1 2

M1 M2
FIGURE 11.4
Finite element for a beam in flexural motion.
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
Here
G Ak1 G Ak2
Rt = and Rh = (11.55)
A 2 G Ak12 A 2 + G Ak22
are the so-called amplitude ratios and the boundary conditions of (11.53)
specify the constants B1 , B2 , C1 , and C2 . See Doyle (1997) for a complete
derivation.
The k11 -component of the resulting spectral stiffness matrix is plotted in
three plots of Figure 11.5 in black, as the reference case; the same material val-
ues as before are employed, and the peaks indicate the resonant frequencies.
11.2.2.2 Random Case

From (11.52) we easily find the frequency space equations of an inhomoge-
neous Timoshenko beam

d Ehw d
v
+ hw 2 v = 0
dx 2(1 + ) dx

d Eh3 w d Ehw dv h3 w 2
+ + = 0. (11.56)
dx 12 dx 2(1 + ) dx 12
We again employ a random field of a band-limited type, and model the mass
density , elastic modulus E, cross-sectional height h, cross-sectional width
w, and Poissons ratio in the same way as in (11.49). The spectral stiffness
matrix

1
V k11 k12 k13 k14
v1

M1 k21 k22 k23 k24 1

= v (11.57)
V2 k31 k32 k33 k34 2
2
M k41 k42 k43 k44 2

is computed by the numerical method outlined before. Figure 11.5 shows

three cases of k11 corresponding to the random E for g = 0.1, 1, and 10,
respectively, wherein the deterministic case is shown as a crisp, black line,
while the random case is shown as a gray thicker line, possibly overlapping
the first one. Evidently, random E has a significant impact on the averaged
solution for all but the lowest frequencies. Moreover, for g = 0.01, after no
more than 10 kHz, the averaged solution k11 resembles random noise. In
fact, this disordered behavior is a result of shifts in the resonant frequencies
of the solution for various realizations. The conclusion, then, is that for these
low values of g, we can rely on a homogenized solution in only the lowest
frequency ranges. The situation improves as we go to higher g values for , E,
and h. Already at g = 1.0, there is some agreement between the deterministic
and the mean at lower frequencies. As we go to g = 10.0 and 100.0, we tend
to have an excellent agreement.
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
(a) 1.00E + 01
9.00E + 00
8.00E + 00
7.00E + 00
Log10 11
6.00E + 00
5.00E + 00
4.00E + 00
3.00E + 00
2.00E + 00
1.00E + 00
0.00E + 00
0.0E + 00 1.0E + 01 2.0E + 01 3.0E + 01 4.0E + 01 5.0E + 01 6.0E + 01 7.0E + 01
Frequency (Hz)
(b) 1.00E + 01
9.00E + 00
8.00E + 00
7.00E + 00
Log10 11
6.00E + 00
5.00E + 00
4.00E + 00
3.00E + 00
2.00E + 00
1.00E + 00
0.00E + 00
0.0E + 00 1.0E + 01 2.0E + 01 3.0E + 01 4.0E + 01 5.0E + 01 6.0E + 01 7.0E + 01
Frequency (Hz)
(c) 1.20E + 01
1.00E + 01
8.00E + 00
Log10 11
6.00E + 00
4.00E + 00
2.00E + 00
0.00E + 00
0.0E + 00 1.0E + 04 2.0E + 04 3.0E + 04 4.0E + 04 5.0E + 04 6.0E + 04 7.0E + 04
Frequency (Hz)
FIGURE 11.5
Timoshenko beam vibrations in the case of random modulus E showing k11 (black line) for
the reference homogeneous medium and mean < k11 > (gray line) for the random case with:
(a) g = 0.1, (b) g = 1.0, (c) g = 10.0.
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
Interestingly the effect of varying w or for any value of g has almost no

effect on the averaged solution at any of the frequency levels we studied. To
sum up, the diffusion of resonances away from those of homogeneous rod
and beam will always occur. In particular, in the case of rods, the effects of
random mass density and elastic modulusbut not of cross-sectional area
are strong. In the case of beams, the effects of random mass density, elastic
modulus and beams heightbut not of Poissons ratio and beams width
are strong. Another new aspect is the very high level of second, third and
fourth moments of response for a much weaker level of noise in the material.
In all the results reported here the introduction of dissipation in the mate-
rial has a tendency of removing the singularities of mean response at resonant
frequencies.
11.2.3 Observations and Related Work

In the deterministic case, by connecting all the elements according to the
spatial geometry, a global stiffness matrix of a given 2D or 3D structure is
constructed and a global response due to a specified impulse is studiedfirst
by going over all the frequencies and then by transforming to the time domain,
which is conveniently done by the fast Fourier transform (FFT), Doyle (1997).
This procedure still needs to be generalized to the stochastic casethe main
problem being that we do not have explicit forms of spectral finite elements
for random field properties. Note here that the classical static stiffness matrix
is actually obtained from the spectral stiffness matrix in a zero-frequency
limit.
Other related work on elastodynamics of structures described by random
fields has been reported, among others, in Adhikari and Manohar (1999, 2000).
Gupta and Manohar (2002), Manohar and Adhikari (1998). Their approach
follows the stochastic finite element method. This method is a close relative of
the stochastic finite difference method (Kaminski, 2002).
The frequency shift in the dispersion relation for waves on random strings
has been studied by Howe (1971)albeit in the long ( d) and short ( d)
wavelength limits only. Those results could not be verified by our numerical
simulations, and the dispersion at those special as well as the general case of
arbitrary wavelengths still remains an open issue.
The 1D wave motion in the case of variable E and was treated analyti-
cally, using a Liouville transformation, in Belyaev and Ziegler (1994), but the
results could only be obtained in the limits d, or d.
Finally, there is also a method of analysis based on the KramersKronig
relations (Beltzer, 1989). This is based on the primitive causality condition,
which expresses the observation that the output cannot precede the input. It
enables one to evaluate the dispersion curves for the entire frequency interval
(0 ) and yields a bound relating the static and dynamic frequency
responses.
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
11.3 Waves in Random 1D Composites

11.3.1 Motion in an Imperfectly Periodic Composite
11.3.1.1 Random Evolutions
In many problems of stochastic mechanics even a weak material disorder may
have a significant impact on the effective material response. In this section
we report on such an occurrence in a relatively classical setting: harmonic
wave motion in an imperfectly periodic composite. The classical subject of
mechanics of periodic composites has seen a great deal of research, but here
we consider a composite with geometric randomness in a nominally periodic
layered structure. We follow Becus (1978, 1979), who himself employed the
method of random evolutions (Hersch, 1974).
Let us consider a dynamical system evolving in any one of several modes
and whose switching from one mode to another is governed by a certain
random mechanism described through a stochastic process. In general, the
dynamical system is represented by a function f (an element in a Banach
space B), and each mode of operation is specified by a linear operator TA (t)
on B with an infinitesimal generator A . These operators have a semigroup
property, that is,
TA (t1 + t2 ) = TA (t1 ) TA (t2 ). (11.58)
Here the sample space = {1, 2, . . ., n} is a countable index set, and

we assume the switching process to be a Markov chain Z(t) with transition
probabilities pij and an infinitesimal matrix Q = [q ij ] = [ pij (0)]. Also, let j (z)
and N(t, z) denote the time of the jth jump and the number of jumps up to
time t, respectively, for the sample path z(t) of Z(t). Now, a product
M (t) = TZ(0) (1 (z)) TZ(1 (z)) (2 (z) 1 (z)) . . . TZ( N (t,z)) (t N (z)) (11.59)
defines a random evolution on B n . It can then be proved that:

1. The ensemble average system evolution (indicated by a tilde) is
given by a semigroup of expectation (ensemble average) operators

(t) f = M (t) f Z(t)
T f B n , (11.60)
where subscript i indicates the state in which the process began.

2. The Cauchy problem

ui
n
= Ai
ui + q ij
ui u (0) = f
i = 1, 2, . . ., n (11.61)
t
j=1
for the unknown vector

u (t) is solved by (t) f .
u (t) = T
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
3. The evolution is asymptotically Gaussian.

4. There is a diffusion approximation to the random evolutions
(Hersch and Pinsky, 1972).
There is also another result from modern probability theory that has use-
fulness in wave motion in composite materials because it generalizes the law
of large numbers to the products of random matrices (Berger, 1984). To this
end, consider the sequence of the products of real random matrices

K
P K () = M j, K (). (11.62)
j=1
It is assumed that, for K , the matrices M j, K () can be represented by
1
M j, K () = Id + B j, K () + R j ( K , ), (11.63)
K
where B j, K () for j = 1, 2, . . ., K are independent, identically distributed
random matrices, integrable with respect to the probability measure P and
|R j ( K , )| = o( K 1 ) for large K . Under these conditions the law of large
numbers takes place and

lim P K () = exp B j, K () (11.64)
K
in the sense of convergence in distribution of all the vectors obtained

by multiplication of the random matrix by an arbitrary deterministic
vector.
11.3.1.2 Effects of Imperfections on Floquet Waves

11.3.1.2.1 Floquet wave
Let us consider a plane wave propagating in direction x in a periodic com-
posite, of period p, made up of two linear elastic alternating layers 1 and 2.
The equations governing stress and displacement u in each layer
2 u u
= (x) = E (x) (11.65)
x t 2 x
led, in the case of a harmonic motion (u(x, t) = e i t u(x)), to an ordinary

differential equation

d d
u
E (x) + (x) 2
u = 0. (11.66)
dx dx
Given the periodic structure
(x + p) = (x) E (x + p) = E (x), (11.67)

P1: Binod
May 28, 2007 18:15 C4174 C4174C011
according to Floquets theorem (e.g., Stoker, 1950) there exist solutions of the
form

u (x) = e iq x v (x). (11.68)
where q is the Floquet wave number and v is a periodic function v(x + p) =

v(x). The periodicity over one period [x0 , x0 + p] of displacements and stresses
leads then to a statement of quasi-periodic boundary conditions
d
u d
u

u (x0 + p) =
u (x0 ) e iq p (x0 + p) = (x0 ) e iq p . (11.69)
dx dx
The solution of the SturmLiouville problem (11.66) and (11.69) is of the
form
u (x, t) = e i[q x+ t] v (x) (11.70)
and v (x) is called a Floquet wave.
11.3.1.2.2 Transfer matrix approach

Going to the frequency domain, one can rewrite (11.65) as a first-order system
(Ziegler, 1976):
X
= A(x) X, (11.71)
x
where

u 0 1/E (x)
X= A= . (11.72)
(x) 2 0
Now, in each homogeneous layer I , E, and are constant, and so the evolution
over this layer is
X (x + h) = TA (h) X (x), (11.73)
were TA is a transfer matrix

1
cosh kh Ek sinh kh
TA (h) = e Ah
= 1 , (11.74)
Ek sinh kh cosh kh

where k = /E is the wave number in I . We see immediately that TA has
the property (11.58), and, since A is piecewise constant, there is a switching
from one layer to another. The global evolution is given by an equation entirely
analogous to (11.59), namely,
X (x0 + p) = TAn (ln ) TAn1 (ln1 ) . . . TA1 (l1 ) X (x0 ), (11.75)
where li is the length of the ith interval, in which A = Ai .

P1: Binod
May 28, 2007 18:15 C4174 C4174C011
One can again set up a quasi-periodic boundary condition
X (x0 + p) = X (x0 ) e iq p (11.76)
for the Floquet wave X (x) = (x) e iq x and (n + p) = (x), and, from (11.75),
arrive at a matrix eigenvalue problem

iq p 1 0
TAn (ln ) TAn1 (ln1 ) ...TA1 (l1 ) X (x0 ) e = 0. (11.77)
0 1
An eigenvalue e iq p with q real (complex) corresponds to a frequency in

a passing (stopping) band, where q = m/ p (m = 1, 2, ...) is a Floquet wave
number.
11.3.1.2.3 Floquet waves in a random composite

We are now ready to consider a situation more general than that of a periodic
composite: there are two layers made of materials 1 and 2, having deter-
ministic properties (, E) 1 and (, E) 2 , respectively, but randomly varying
thicknesses 1 and 2 ; the average thicknesses are a 1 and a 2 . Clearly, within
the general framework of Section 11.3.1, the sample space has two elements
{1, 2}, and the infinitesimal generators A are

0 E 11 0 E 21
A1 = A2 = . (11.78)
1 2 0 2 2 0
Furthermore, the switching process Z (t) is taken as a generalized tele-

graph process with an infinitesimal matrix
1
a 1 a 11
Q= , (11.79)
a 21 a 21
so that the switching from layer 1(2) to 2(1) takes place over distances dis-
tributed according to a Poisson process with intensity a 11 (a 21 ).
With X1 and X
2 denoting the average solution to (11.71), an application of
(11.61) and (11.77) results in

j
X j
X
d
= Oj d X j , j = 1, 2, (11.80)
dx d X j
dx dx
where D j can be determined explicitly from all the parameters of the
problems. This then leads to an eigenvalue problem, whose solution indi-
cates that there is only one frequency of an average Floquet wave when
(1 2 )( E 11 E 21 ) < 0. On the other hand, assuming a perfectly peri-
odic composite without structural imperfections, one finds a whole spectrum
of frequencies of Floquet waves. A diffusion approximation for this system
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
was developed in Becus (1979), and this may become especially useful when
there are more than two layers.
11.3.2 Waves in Randomly Segmented Elastic Bars

Various studies of wave transmission and reflection phenomena, with a focus
on the spectral response, have been conducted (e.g., Kotulski, 1994). He con-
sidered a more general model composite made of 2K segments (i.e., two-
layered again) with random lengths li () , i = 1, 2, . . ., 2K , whereby also
the material properties of the segments (mass density, elastic modulus and
cross-sectional area) are random vectors ( j = 1, 2, . . ., 2K ):

2 j1 () , E 2 j1 () , A2 j1 ()

2 j () , E 2 j () , A2 j () . (11.81)
The lengths of the segments are assumed to satisfy a special relation:

l2 j1 () l2 j ()
l2 j1 () , l2 j () = , j = 1, 2, . . ., K (11.82)
2K 2K
are independent, identically distributed random variables with the means

l2 j1 () = L 1 l2 j () = L 2 . (11.83)
With this setup, he proceeded to study the asymptotic behavior of the random-
ized equation for the amplitudes of the waves by the law of large numbers for
the products of random matrices stated in Section 11.3.1. The transfer matrix
M j, K ( , ) is not reproduced here explicitly, but once again we note that our
notation of for elementary event and for frequency is the reverse of that
used by Kotulski. It has turned out that the randomization of the bar results in
a much slower homogenization of response, in function of the increasing K ,
than in the deterministic case (i.e., one where all the properties are perfectly
periodic).
Analogous, but more complex, effects of coupled thermoelastic wave
propagation in such composite media were studied in Kotulski and

Pretczy nski (1994). As a starting point they took the well-known equations of
thermoelasticity in 1D
2 u 2 u
= ( + 2) (3 + 2)
t 2 x2 x

2 2
u
c = T0 (3 + 2) , (11.84)
t x2 t x
where and are the Lame constants, T0 is the reference temperature, c is

the specific heat at constant strain, is the heat conductivity coefficient, and
is the thermal expansion coefficient. Passing to the frequency domain, they
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
set up a first-order system of the type (11.71), where

(3+2)

u 0 1
(+2) (+2)
0

2 0 0 0
X=

A=
i (3+2)T0
. (11.85)
1

0 0

0 0 ic 0
Here, each layer was described by two random vectors ( j = 1, 2, . . ., 2K ):

2 j1 (), 2 j1 (), 2 j1 (), 2 j1 (), 2 j1 (), c ,2 j1 (), l2 j1 () ,

2 j (), 2 j (), 2 j (), 2 j (), 2 j (), c ,2 j (), l2 j () (11.86)
and the ensuing analysis was analogous to that outlined above.

In the next section we will disregard the backscattered waves, as our
interest will be in nonlinear (in-)elastic microstructures where the forward
transmitted pulse is of primary importance. The backscattering due to mis-
match at differing layers, however, plays an important role in geophysical
problems and results, for example, in an apparent slowing down of the main
pulse carried forward (Asch et al., 1991). See also Foias and Frazho (1990),
Kennett (1981, 1983), and Papanicolaou (1998). The latter paper reviews a
number of aspects important in geophysical applications: coherent versus
incoherent fields, localization and transport.
11.4 Transient Waves in Heterogeneous Nonlinear Media

11.4.1 A Class of Models of Random Media
In this section we consider the random medium B = {B(); } to be
made of realizations B() (i.e., specimens) that are one-dimensional and semi-
infinite in the physical domain: X X = [0, ]. (With strains being infinitesi-
mal, no distinction is required between the material and spatial coordinates
but here we use X instead of x of the previous sections, because x will be
employed for a value of the random variable in the diffusion model to be
developed below.) Material properties vary in a discontinuous fashion from
grain to grain, but remain constant within each grain. In the stochastic termi-
nology this is a chain process. Let us introduce the following classification of
microstructures.
Linear elastic media: length l, mass density , and elastic modulus

E are random, Figure 11.6(a).
Bilinear elastic media: length l, mass density , and two elastic mod-
uli E 0 , E 1 are random. The stress level separating both linear
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
(a) (b)
<E1 >
<E>
<E0 >

<E0 >
(c) (d)
<E>
<E1 >
FIGURE 11.6
Constitutive laws: (a) linear elastic; (b) soft bilinear elastic; (c) soft nonlinear elastic; (d) linear-
hysteretic. In each case, randomness of modulus, or moduli, is indicated.
elastic ranges is assumed deterministic. The stress-strain law of each

grain is

= E 0 () if | | < ,

= + E 1 () ( 0 ) if | | . (11.87)
Thus, B is described by a vector random process parameterized by X, that

is: {l, E 0 , E 1 } X . Three general types of this bilinear elastic granular medium
may be considered:
1. All the grains are of a soft characteristic, Figure 11.6(b)

2. All the grains are of a hard characteristic
3. Both types of grains are present
Nonlinear elastic granular media: length l, mass density , and

elastic modulus E are random; see Figure 11.6(c). The stochastic
stress-strain law is
= E () n , (11.88)
where either n > 1, or < 1. Three general types of this model

analogous to those of the bilinear model may be considered here.
Linear-hysteretic granular media: length l, mass density , and
two elastic moduli E 0 , E 1 are random; see Figure 11.6(d). The
stress-strain curve is a straight line on initial loading; its slope
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
defines the initial modulus E 0 ,. Upon unloading the stress-strain

curve is another straight line, which defines the second
modulus E 1
m = E 1 () ( m ). (11.89)
If the material is reloaded, it follows that the second line backs up, and
then continues along the initial loading line.
In all these models material properties are generally assumed to have a
Markov property in X, while the noise-to-signal ratio of these properties is
assumed small.
11.4.2 Pulse Propagation in a Linear Elastic Microstructure

In this section we describe a method of analysis of transient waves propa-
gating in a linear elastic microstructure, which forms the basis for study of
nonlinear models. We start by considering a space-time graph of a distur-
bance propagating in a semi-infinite sequence of linear elastic grains. By a
disturbance we understand any single point of the pulse f (t) applied at the
free face X = 0, see Figure 11.7. This pulse results in a wavefront moving
into the material domain. Our strategy is to find the rules of evolution of the
disturbance, and use these to construct the wavefront at any later stage. Now,
disturbance propagation in every grain occurs as a Riemann wave. We follow
the forward propagating disturbance only, that is, we do not keep track of
the waves backscattered from all the grain boundaries. Due to a (random)
variation of the properties of the grains, there are two effects:
1. The amplitude of the disturbance undergoes a change with

passage from grain to grain; denotes either stress or velocity.
This change of is described by the transmission coefficient T (it)
defined by
(t) = T (it) (i) , (11.90)
in which i and t denote the incident and transmitted quantities,

respectively. In case of an ideal (nonslip) grain boundary model,
and with standing for stress, T (it) is given by the well-known
formula
2 (it)
T (it) = , (11.91)
1 + (it)
with the relative impedance being

!
(t) E (t)
(it) = . (11.92)
(i) E (i)
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
t
(X,t)
c(t0) <c1()>1
t0 <c()>
(X,)
t(X,)
c+(0)
Sample path
Fastest path
Slowest path
(t,)
f(t) X(t,) X
FIGURE 11.7
Spacetime graph of pulse propagation. Shown are the backward causality cone C (t0 ) and the
forward causality cone C + (t0 ), the latter being bounded by the slowest and fastest paths, and
1
contains two paths at c 1 and c, as well as a sample (random) path c (, X).
2. The disturbance propagates forward with a phase velocity varying

randomly from grain to grain. Thus every point at (X = 0, t = t0 )
gives rise to a set of all possible characteristics contained within a
so-called forward causality cone C + (t0 ), where each single character-
istic corresponds to a disturbance propagating in a single specimen
B() of B, where t0 = 0. For the assumed piecewise-constant ran-
dom medium model these characteristics are continuous piecewise-
linear. The random walk is therefore a natural stochastic model for
their forward evolution, and it may be described by either t ( X, )
if X is chosen as an independent parameter, or by X (t, ) if t plays
that role. However, in order to grasp the scatter in the arrival times
we may introduce a so-called dispersion time
( X, ) = t ( X, ) t ( X), (11.93)
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
where the mean (ensemble average) characteristic is defined by

t ( X) = c 1 X, (11.94)
while, in order to describe the scatter in distances covered in a given

time t, we may introduce a so-called dispersion distance
(t, ) = X (t, ) X (t), (11.95)
where the mean characteristic is defined by
X (t) = c t. (11.96)
In view of the above points, we have a following sequence of

implications:
Assume (l, , E) X is a Markov process

Then (l, , E) , (it) , c X is a Markov process

Then (l, , E) , T (it) , c X is a Markov process
Then [(l, , E) , , ] X is a Markov process
Then [(l, , E) , , ]t is a Markov process.
It follows that a complete description of the disturbance evolution

hereinafter denoted by W, but not implying the Wiener process per seis
obtained through either one of two vector processes (taking values w in the
state space W)
WX = [(l, , E) , , ] X or Wt = [(l, , E) , , ]t , (11.97)
depending on whether a parametrization with respect to X or t is preferred.

An important and natural property of this approach is that the transition
probability function of WX , or Wt , is derivable from the microstructures statis-
tics. In case these statistics are space-homogeneous, the transition function of
WX is space-homogeneous too, while that of Wt is time-homogeneous. Now,
Markov processes having time-homogeneous transition functions satisfy nat-
urally (without transformation of the state space) the semigroup property
M (t1 + t2 ) = M (t1 ) M (t1 ) M (0) = I, (11.98)
where

M (t) [g (w1 )] = g (w2 ) P (t, w1 , w2 ) dw2 . (11.99)
W
The above represents a stochastic form of Huygens minor principle for the
disturbance evolution (Ostoja-Starzewski, 1989).
Markov processes Wx and Wt that model propagating disturbances may
conveniently be approximated by diffusion processes. In the following we
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
discuss the forward Fokker-Planck (FP) approximation, which is expressed

by
p 1 2
= [Ai (w) p] + Bij (w) p , (11.100)
t wi 2 i, j wi w j
i
in which p is the conditional probability or the probability density of W at time

t. The dependence on time of the drift and diffusion coefficients is present if
the transition function of the (l, , E) X process is not homogeneous in X. In
the special case of a complete mutual independence of the properties of the
grains (-correlatedness on scale l) we find for p(t, z, x) for the [, ] vector
at time t
p 1 2 1 2 2
= A p A p + B p + B p + B p .
t z x 2 z2 2 x 2 zx
(11.101)
Thus, z and x denote values of the random variables and , while the drift
and diffusion coefficients have forms
A = Az, A = 0,
B = Bz , 2
B = D, B = E z, (11.102)
which involve constants A, B, D and Ethese can be computed from random

fluctuations of the material. A glance at (11.101) reveals that the wave process
WX is multiplicative in its amplitude , a property characteristic of all wave
phenomena studied hereinafter.
The forms of drift and diffusion coefficients in (11.102) permit a transfor-

mation of the Wt = (, ) t process into a vector Wiener process , t , as
follows:

(, ) , , ( ) = ln ( ). (11.103)
Accordingly, the new drift and diffusion coefficients become
A = 0, A = 0,
B = B, B = D, B = E, (11.104)
so that the FP equation governing the probability density p = p (t, x, y) is
p B 2 p D 2 p 2 p
= + + E , (11.105)
t 2 y2 2 x2 y x
in which y is a value in the range of the random variable . Thus, the solution
to the above, subject to the initial condition
p (0, x, y) = ( y y0 ) (x), y0 = ln (z0 ), (11.106)

P1: Binod
May 28, 2007 18:15 C4174 C4174C011
has a bivariate Gaussian form with a constant mean and a covariance matrix
proportional to

B E
. (11.107)
E D
The probability density in the original variables is then found as
p (z, y) = e y p ( y, x). (11.108)
It is possible to find from (11.101) the evolution of first and second

moments (covariances)

(t) = z0 e At , (t) = 0, (t) = (z0 ) 2 e 2At e Bt 1 ,
(t) = Dt, (t) (t) = E z0 te At . (11.109)
It follows that every point t0 on the taxis is the origin of a forward

causality cone C + (t0 ) centered about the mean characteristic. The latter is
defined either by c or c 1 1 depending, respectively, on whether time or
position is chosen as a controlling parameter. Thus, the entire space-time is
covered by the C + cones rather than by forward characteristics of the homo-
geneous deterministic linear elastic medium problem. The ensemble average
amplitude at an arbitrary point (x, t = t0 + xc 1 ) in space-time may by
calculated by considering all the characteristics within the backward causality
cone C (t0 ) and conducting an integration of the initial pulse f (t) over its
base at x = 0, see Figure 11.7.
11.4.3 Pulse Propagation in Nonlinear Microstructures

11.4.3.1 Bilinear Elastic Microstructures
Let us now extend the preceding analysis to pulse propagation in a microstruc-
ture of soft bilinear elastic grains (Ostoja-Starzewski, 1991a). First, we discuss
the rules of disturbance propagation, and then proceed to infer from this the
response to an initial pulse at the front end x = 0
f (t) = at, a = const. (11.110)
Recalling Figure 11.6(b), we observe: (1) if (t0 ) < , a disturbance will

propagate in the medium as if it was a linear elastic medium; (2) if (t0 ) > ,
the propagation velocity is initially slowc 1 corresponds to E 1 and upon
reaching , due to attenuation, it becomes fastc 0 corresponds to E 0 .
Strictly speaking, there is a possibility of the propagation becoming slow
in one or several grains due to random fluctuations, but the loss of energy
at the wavefront soon takes over and the propagation is fast. The location
of the ensemble average point O = ( X , t ) of transition from fast to slow
propagation is readily found using the results of Section 11.4.2. Thus, the
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
mean amplitude behavior is described by (11.109)1 , so that
= 0 e At , (11.111)
yields the time t . Thin lines in Figure 11.8(a) show the outlines of scatter about
the mean forward characteristic, that is, of the forward cone C + (t0 ), where
"
s (t) = (t) = t D (11.112)
denotes the standard deviation of t . At the point O we begin the second

forward cone with the mean characteristic corresponding to c 0 (or, in case
of parametrization by x, to the harmonic average of c 0 ). Clearly, there is a
scatter about point O , whose range is characterized by two times t1 and t2 .
The first one is the earliest time of reaching the stress level and is calculated
t t
t'2
t'
t'1
0'
t'2
t' 0'
t'1
X X
(a) (b)
t t
<(t)>
<(t)>
t t
f (t) X X
f (t)
(c) (d)
FIGURE 11.8
Space-time graphs of disturbances propagating in (a) soft and (b) hard bilinear elastic media;
(c) linear pulse and the acceleration wavefront (t) propagating in the soft bilinear medium;
(d) linear pulse and the shock wavefront (t) propagating in the hard bilinear medium.
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
"
from the condition z0 e At1 3 (t1 ) = . On the other hand, t2 is the
latest possible
" time of reaching the stress level , and is to be found from

z0 e + 3 (t2 ) = . These two times define the outlines of the second
At2
forward cone, that are shown in thin lines again.

We are now ready to analyze wavefront propagation due to the pulse
(11.110), see Figure 11.8(c). Thus, if by t we denote the time for which
equals , it becomes apparent that all the disturbances originating at times
t0 < t will propagate fast, while those that originate after t will propagate
initially slowly and laterdue to a decrease of their amplitudes down to
will propagate fast. In the ensemble average sense, this switch to fast
propagation occurs at

1
t = t0 + ln , t0 . (11.113)
A a t0 a
We observe that /t is discontinuous at a point of switch from slow to fast
propagation. Thus, the loci of all the points (c 1 (t t0 ), t ) in the space-time
graph, denoted by X = (t ), represent the ensemble average acceleration
wavefront.
In the case of a microstructure made of grains with hard bilinear elastic
response, we observe a disturbance displaying behavior reverse to that dis-
cussed in the preceding section providing (t0 ) > . That is, the propagation
velocity is initially fastc 1 corresponds to E 1 and upon reaching due
to attenuation, it becomes fastc 0 corresponds to E 0 ; see Figure 11.8(b). The
location of the ensemble average point O = ( X , t ) of transition from fast
to slow propagation, and the range of scatter described by t1 and t2 may be
found with the same type of formulas as (11.11111.113).
Turning our attention to wavefront propagation due to the initial pulse
(11.110), we first recall a solution to a corresponding deterministic homoge-
neous medium problem (Wlodarczyk, 1972), which corresponds to the case
of no randomness in the properties of the grains. It is well known that a shock
wave (t) (made of two parts 1 (t) and 2 (t)) will form. More specifically,
1 (t) is an intersection
of fast and slow characteristics and is straight
with velocity c 0 = E/.
2 (t) corresponds to fast characteristics propagating into an undis-
turbed region and it curves progressively to become faster.
These results serve as a reference basis in the case of the same loading of
a random microstructure. In fact, each of the straight characteristics originat-
ing at t0 is to be replaced by a forward causality cone C + (t0 ). Consequently,
the shocks 1 (t) and 2 (t) are Markov random processes evolving within
a region shown by two broken lines in Figure 11.8(d). They may be approxi-
mated by diffusion processes, with the transition functionsand hence, the
drift and diffusion coefficientsbeing derived from the rules of evolution
in the deterministic problem. The idea is to consider the ratesand their
conditional momentsof change of (t) and (t + t) over an interval t
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
corresponding to the passage of through one grain. This then provides a

framework for analysis of random variation of shock strength [ ] = + +
at the wavefront, where + and refer to the quantities ahead of and behind
(t). The vector W = ( + , [ ], ) t evolves as a Markov process, and the drift
and diffusion coefficients can be derived from the rules of evolution in the
deterministic problem.
11.4.3.2 Nonlinear Elastic Microstructures

It is well known that, in the absence of body forces and for small strains, the
equations of motion and continuity in a homogeneous continuum result in a
system of two first-order quasi-linear hyperbolic partial differential equations
v v 1
= = , (11.114)
X t X c 2 ( ) t
where
!
1
c ( ) = (11.115)

is the propagation velocity. This is the situation of physically nonlinear elastic

grains shown in Figure 11.6(c). Following Ostoja-Starzewski (1995a), if and
c are random fields in X, (11.114) is stochastic; the explicit dependence of ,
v, , and c on is suppressed here for clarity of presentation.
In accordance with our basic formulation, we develop the rules of distur-
bance evolution first, and then use them to study response due to the initial
pulse (11.110). As before, there are two choices: parametrization with respect
to Xformula (11.94)1 or with respect to tformula (11.96)2 . We realize
that is now dependent on in case of the first choice, or analogously, is
dependent on in case of the second choice; itself is being driven by the
(l, , E) x vector process. The change of is described by the transmission
coefficient T (it) of (11.90). Using an ideal (nonslip) grain boundary model
for two nonlinear elastic grains in contact, we derive the following relation
governing T (it) :
# 1 (it) (1+n)/2n $2n/(1+n)
T (it) + (it) T 1 1=0
1/2 (t) (i) 1/2n
(it) = (t) / (i) E /E . (11.116)
This reduces to (11.91 and 11.92) in the special linear elastic case of n = 1. The
fact that (11.116)1 is implicit does not pose a problem in the diffusion formu-
lation that follows. However, before developing it, we note the dependence
of phase velocity on the stress amplitude
c ( ) = (nE/) 1/2 [/E](n1)/2n . (11.117)
The diffusion model of a propagating disturbance is formulated now

for the Wx process in the particular case of space-homogeneous statistics;
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
parametrization with respect to x is used in order to avoid the dependence

on of the passage time t referred to a single grain, where x = l is used.
Thus, we work with Wx = [, ]x for which (11.101) holds, and the drift and
diffusion coefficients are
A = A, A = C (1n)/2n ,
B = B 2 , B = D (1n)/2n , B = E (1+n)/2n . (11.118)
All the constants A through E may be computed, most conveniently, through

a Monte Carlo procedure. We observe that:
The functional forms of drift and diffusion coefficients of the -

process are the same as in the linear (or bilinear) elastic case, that is,
A is linear while B is quadratic in .
The drift and diffusion coefficients of the -process are nonlinear in
and hence the process Wx is nonlinearly multiplicative in ; so
that an analytical solution of (11.101) with (11.118) is unwieldy.
Let us again consider the response of a semi-infinite body, this time made
of soft grains of Figure 11.6(c), due to the pulse (11.110). Figure 11.9(a) shows
the graph of the homogeneous medium, and (b) gives the graph of the random
medium. It is seen in the first case that as the stress increases at X = 0, the
propagation velocity of Riemann waves is successively smaller, and as the
pulse is carried away from the front end it is being washed out in time.
In case (b) this phenomenon is modified by the curving and diffusion of
characteristics within their forward evolution cones, thereby reflecting the
accompanying attenuation of stress.
Finally, we consider the response of a material with hard grains to the
same pulse (11.110). As expected, the effect of an increasing pulse will be
opposite to that observed above: instead of washing out we have a com-
pression of the pulse resulting in a so-called loading shock wave (Nowacki,
1978). Thus, Figure 11.10(a) illustrates the classical homogeneous medium
response. Solution by characteristics is continued here until the slower ones
are overtaken by the faster ones. Using ten initial characteristics we obtain
an envelope of the shock wave propagating into an undisturbed body. In the
random medium case (Figure 11.10(b)) we see a qualitative modification of
this phenomenon due to a replacement of all straight characteristics by the
forward evolution cones. Their curving up leads to a delay in the arrival of
the shock, which actually has a progressively weaker strength than that of the
homogeneous problem, due to the stress attenuation. In both cases presented
in Figure 11.10 the computation started at a very small (non-zero) value of
f (t) in order to avoid the situation of a so-called sonic vacuum, which calls
for a zero propagation speed at zero stress in a medium with Hertzian contacts
(Nesterenko, 2001).
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
(a)
x
t
(b)
FIGURE 11.9
Washing out of a pulse in a medium of soft bilinear elastic grains due to a linear forcing:
(a) homogeneous material, (b) random material. Note the curving down of characteristics in
case (b).
11.4.3.3 Hysteretic Microstructures

The same method as formulated above can be employed to study pulse prop-
agation in a hysteretic microstructure described by Figure 11.6(d). The free
face is now subjected to a square pulse

p0 for 0 t t1
f (t) = . (11.119)
0 otherwise
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
(a)
x
t
(b)
FIGURE 11.10
Formation of a loading shock wave in a medium of hard bilinear elastic grains due to a linear
forcing: (a) homogeneous material, (b) random material. Note the curving up of characteristics.
A problem with this initial condition for a deterministic, homogeneous

medium (Salvadori et al., 1960) forms the reference basis for solution of a
stochastic problem. In the space-time of Figure 11.11(a) we see several re-
gions: I, II, III, IV, V, . . .. Thus, region I is that of an undisturbed body I = 0,
I = 0, v I = 0, while region II corresponds to a material in which I I = p0 ,
I I = I I /E 0 , v I I = c 0 I I and the entire x, t-plane can be analyzed in this
fashion.
Turning now to a random medium problem we first observe that each of
the lines in Figure 11.11(a) representing the discontinuity waves (i.e., shocks)
can be considered as the mean path providing reference for stochastic proc-
esses and . Thus, for example, the leading shock is a reference for a family
of characteristics
X (0, , c 0 ) = X (t, , c 0 ) |t=0 . (11.120)

P1: Binod
May 28, 2007 18:15 C4174 C4174C011
t V
IV
III 0'
t1
II I
p(t)
x
0
(a)
F
G
B 0'
H
D
t1 E
A
p(t)
x
0
(b)
FIGURE 11.11
(a) Spacetime graph of response of a deterministic, linear-hysteretic medium to a square pulse.
(b) Intersection of forward dependence cones in a random medium case showing strong scatter
about O .
Similarly, the line bounding the region II from above is a family of character-
istics
X (t1 , , c 1 ) = X (t, , c 1 ) |t=t1 , (11.121)
and so forth. The cones corresponding to the processes along the two
above-mentioned mean characteristics (paths) are shown in Figure 11.11(b).
Clearly, the point of intersection will be diffused about the reference point
(x = c 0 t2 , t2 ).
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
Following an analysis of characteristics in the spacetime, it is seen that,

in case of dependence between the random variables E 0 and E 1 , the scatter in
the point of intersection of two characteristicsas measured by the distance
between points A and Cincreases for the ratio E 0 /E 1 decreasing to 1. This
indicates that even a weak randomness in the mediums properties may alter
certain aspects of its response in a significant way! On the other hand, in the
case of dependence defined by E 0 () = E 0 if and only if E 1 () = E 1 ,the
scatter in the location of intersection point of two characteristicsas measured
by the distance between points B and Dwould be much weaker.
The foregoing analysis of the intersection of characteristics carries over to
any intersection point, as well as to all such points of other related problems.
We also note that extensions to 2D and 3D problems are possible and, in fact,
very natural for problems with cylindrical or spherical symmetry (of both
loading and materials statistics) where analyses would be conducted in a
spacetime having radius in place of X.
Finally, we observe a possibility of treatment of a nonlinear elastic lam-
inated composite (Chen and Gurtin, 1973), made of alternating two-phase
layers, with each phase being described by three variables: mass density, tan-
gent modulus and second-order modulus. The elastic nonlinearity leads to
a wave amplification, while the layer-to-layer mismatch of properties has an
opposite effect. A study of this type of competition, albeit in the random
continuum setting, is reported in the next section.
11.5 Acceleration Wavefronts in Nonlinear Media

11.5.1 Microscale Heterogeneity versus Wavefront Thickness
11.5.1.1 Basic Considerations
It is a general finding of continuum mechanics of nonlinear elastic/dissipative
media that acceleration waves, that is, moving singular surfaces with a jump
in particle acceleration, are governed by a Bernoulli equation (Coleman and
Gurtin, 1965)
d
= + 2 . (11.122)
dx
Here x denotes position (as in all the previous sections except 11.4.), is the
jump in particle acceleration, and the coefficients and represent, respec-
tively, two effects: dissipation and elastic nonlinearity. The interesting aspect
of acceleration waves uncovered through this equation is that, due to the
competition between these two effects, there is a possibility of blow-up, and
hence, of shock formation in a finite distance x , providing the initial ampli-
tude 0 exceeds a critical amplitude c . x is also called distance to blow-up or
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
distance to form a shock. It is easy to establish that, for a homogeneous medium,

1
c = , x = ln 1 . (11.123)
0
Various other cases of deterministic spatial dependence of and on x

were investigated in the wake of the aforementioned reference; see also Chen
(1973), McCarthy (1975), and Menon et al. (1983). It is interesting that (11.122)
was also derived independently from the continuum thermodynamics with
internal variables (Bland, 1969); see also Engelbrecht (1997). The third setting
in which such a problem appears is rational extended thermodynamics (e.g.,
Wilmanski, 1998), a theory which, basically speaking, sets up all the field
equations from a hyperbolic systems standpoint.
Now, all these studies were set in the context of deterministic mecha-
nics (and thermodynamics), as expressed by the fundamental requirement of
separation of scales
d L L macro . (11.124)
Here d is the characteristic scale of the microstructure, L is the RVE size and
L macro is the macroscopic body size. This says that, in the case of wavefront
propagation, d must be infinitesimal relative to the wavefront thickness L.
Using a fine-grained mosaic, Figure 11.12(a) suggests this with the under-
standing that truly infinitesimal grains cannot really be shown.
Evidently, the RVE limit implied in Figure 11.12(a) corresponds to the
classical concept of the wavefront, treated as a singular surface, as shown in
Figure 11.13. Let us therefore recall that a jump in f (x, t) in the classical case
is defined and denoted by
[[ f ]] = f 2 f 1 , (11.125)
where f 1 and f 2 are, respectively, the quantities immediately ahead of and

behind the wavefront. It is well known from continuum mechanics that, when
f is continuous, we have the first-order kinematical and dynamical compat-
ibility conditions

f f ui
= c , ij p j = c . (11.126)
t x t
Given the limit d , the tractions and displacements on either side are
uniform because we effectively deal with a classical continuum. This means
that the constitutive law of the RVE, in order to assure that mechanically
defined response should be identical to the energetically defined response,
satisfies the Hill condition. When d L, assuming spatially homogeneous
and ergodic statistics of material properties, we have a separation of scales and
all three conditions result in the same (i.e. unique) constitutive response. For
example, if also isotropy applies, then a linear elastic law ij = ij kk + 2ij
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
f |t = 0
t
ron
avef p
W
0 x
x0 x0 + L
L L d L
p p p
(a) (b) (c)
FIGURE 11.12
Propagation of a wavefront f (x, t) in spacetime. The wavefront is a zone of finite thickness
L (between x0 and x0 + L at time t = 0) propagating in the direction p, in a microstructure of
characteristic grain size d. Three cases are distinguished: (a) L d, which shows the trend to a
classical (deterministic) continuum limit, in which fluctuations die out to zero; (b) L finite relative
to d, where spatial fluctuations render the wavefront a statistical mesoscale element; (c) L d,
which leads to a piecewise-constant evolution.
2
D
FIGURE 11.13
A singular surface propagating from region 2 to 1.
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
holds for B L/d , and we simply have

ij = ij [[uk,k ]] + ui, j + u j,i , (11.127)
because jumps in displacements imply jumps in components of ij across the

wavefront surface.
However, when d is not infinitesimal relative to L, we face a nonunique-
ness of constitutive responses of B L/d , and an accompanying statistical scat-
ter. The latter is due to the fact that microstructure changes as the wavefront
mesodomain travels across it. The mesodomain B L/d is the statistical volume el-
ement (SVE). Thus, if the mosaic of Figure 11.12(b) is linear elastic everywhere,
we have a random response law (almost surely anisotropic)
ij = Ci jkl (, x, /d)ij, (11.128)
where Ci jkl (, x, /d) is a random stiffness tensor field. In place of (11.131),

we should then have

ij = Ci jkl (, x, /d) ui, j + u j,i /2. (11.129)
Clearly, the SVE B L/d is set up on a mesoscale L relative to d, and the wave-
fronts evolution is stochastically affected by the random mesoscale fluctu-
ations of the microstructure, Figure 11.12(b). To this end, one must consider
the wavefronts modulation according to the Bernoulli equation (11.122), but
now with material coefficients and taken as random processes in x, that
is, jointly forming a vector random process [, ]x .
Finally, there is also a third possibility, shown in Figure 11.12(c), where the
wavefront thickness L is much smaller than the grain size d; the grain signifies
a layer. In that case, the RVE assumption pertains to the microstructure much
finer than d, not shown here, and the evolution involves transmissions and
reflections at consecutive boundaries, resulting in a jump process for the for-
ward propagating wavefront. This is a special case of what has been discussed
in Section 11.4.
11.5.1.2 Mesoscale Response

Focusing henceforth on the case of Figure 11.12(b), we deal with a stochastic
Bernoulli equation driven by [, ]x , a process having continuous realiza-
tions. The question that arises is how to set up (or specify) such a process.
At this point, we recall the explicit formulas for the dissipation coefficient,
the nonlinear amplification coefficient and the velocity of acceleration wave
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
are (Coleman and Gurtin, 1965, and other references above)

!
G 0 0
E G0
= , = , c= . (11.130)
2G 0 2G 0 c R
Here G 0 is called the instantaneous modulus, G 0 is the coefficient responsible

0 is called the instantaneous second-order tangent modulus
for dissipation, E
and R is the mass density in the reference state. Thus, in the random medium
case, we really have a dynamical system
d G 0 % R
E
= 0 2, (11.131)
dx 2G 0 2G 0 G 0
& '
0, R .
driven by a four-component random process x = G 0 , G 0 , E
x
Clearly, it would be most desirable to specify x , according to the meso-
scale L relative to d, rather than via [, ]x . However, mesoscale properties for
the wavefront, except for the mass density where a straightforward volume
averaging is valid, would require a combination of mathematical morpho-
logy (for generation of realizations of random geometries) with computational
mechanics of nonlinear elastic/dissipative microstructures (for boundary
value problems according to either one of (11.128) through (11.130), the actual
loading in Figure 11.12(b) being, of course, unknown. While this procedure
has been shown to provide mesoscale bounds for various linear and nonlin-
ear elastic as well as some inelastic materials (see previous chapters), here
we would also need to compute the second-moments as well as spatial cross-
correlations of x . Assuming we go ahead with this, we would then be faced
with a differential equation (11.131) driven by x , for which a quite compli-
cated parametric study of various dependencies between the four component
processes would still need to be carried out. Therefore, in our studies of
(11.122) to date, we have considered the three most fundamental cases of
the [, ]x process: full positive, zero and full negative cross-correlation of
with . Note that this approach gives bounds on the stochastic problem at
hand (!) in that any particular situation of the four-component vector process
from the said 4 4 matrix must fall within our bounds, yet our analysis is
much more tractable.
Now, for small mismatches in microscale material parameters or for a
wavefronts thickness L rather large relative to d, we can definitely argue that
any micromechanically based mesoscale model would lead to and being
two random processes with small noise-to-signal ratios. This, in fact, has been
our starting point in the previous analyses of the subject (Ostoja-Starzewski,

1993c, 1995c; Ostoja-Starzewski and Trebicki, 1999, 2003), its most fundamen-
tal feature being a stochastic rather than a deterministic competition between
the elastic nonlinearity and dissipation in (11.122), and the resulting random
character of and . Hence, the question we have been asking: how different
are the averages c and x for the random medium from the values given
by (11.123) in which the random noises in [, ]x are neglected?
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
11.5.2 Wavefront Dynamics in Random Microstructures

11.5.2.1 Model with One White Noise
The simplest way to introduce randomness into the Bernoulli equation (11.122)
is to consider and to take them as constants perturbed by the same zero-
mean white noise process (x) with two strengths S1 and S2 , respectively,
(x) = + S1 (x) (x) = + S2 (x)

, (11.132)
S1 , S2 0 S1 + S2 = S,
where
S1 S2 . (11.133)
Introducing (11.132) into (11.122) we obtain a stochastic differential equation

for
d
= + 2 + S2 2 S1 (x) (x0 ) = 0 , (11.134)
dx
where the initial condition is deterministic.
Now, given the fact that perturbations entering (11.134) are of a parametric
type, we set up a Stratonovich equation in the sense that (x) is treated as
a Stratonovich-type differential d W(S) (x) of the Wiener process (e.g., Schuss,
1980)
d
= + 2 + S2 2 S1 d W(S) (x) (x0 ) = 0 . (11.135)
dx
The Ito equation equivalent to (11.135) is
d = () dx
A() dx + B (x0 ) = 0 , (11.136)
where

1 2 3
A() = S + S1 S2 2 + S22 3
2 1 2
() = S2 2 S1
B (11.137)
are, in fact, the drift and diffusion coefficients of the diffusion Markov
process .
As our interest is in determining the blow-up (or escape) of to , it is
more convenient to study the decay of the inverse (or reciprocal) amplitude
= 1/ to zero, Figure 11.14. This way the problem of blow-up in is con-
verted to the classical problem in evolutionary random processes: crossing
the boundary at = 0. The Ito equation for is
1
d = [b 1 + b 2 ] dx + (S1 S2 ) d W (x) (x0 ) = . (11.138)
0
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
2
Acceleration wave
0
(a)
2.0
1.5
Inverse amplitude
1.0
z0 = zc
0.5
0
2 4 6
x
(b)
FIGURE 11.14
Simulation of ten exemplary evolutions of an acceleration wavefront (a) and its inverse = 1/
(b) originating from the critical amplitude of a reference homogeneous deterministic medium
c(det) = / as functions of distance x in a random medium described by one white noise.
Observe that either a growth to or a decay to 0 occur. Parameters: = 1, = 1, S1 = 0.2,
and S2 = 0.35. After Ostoja-Starzewski and Trebicki (1995), with permission.
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
where
1 2 1
b 1 = + S b 2 = S1 S2 . (11.139)
2 1 2
In a similar fashion one can set up formulas for the moments of . Of primary
interest is the equation governing the first moment
1
= b 1 + b 2 (x0 ) = . (11.140)
0
From this, noting from (11.123) that the critical amplitude c(det) of the
reference deterministic homogeneous medium is

c(det) = , (11.141)

we find the following relationships between the average critical amplitude of

the random medium, c , and c(det)
S2
c < c(det) for > ,
S1
S2
c > c(det) for < . (11.142)
S1
11.5.2.2 Model with Two Correlated Gaussian Noises

A richer, and more realistic model can be constructed when the random proc-
esses and are taken as two separate processes. In the following we sketch
this for Gaussian processes, whereby we note that (1) in reality (and strictly
speaking) and cannot be Gaussian, but (2), for weak randomness levels,
non-Gaussian noises result in effectively the same results for probability dis-
tributions of dynamical systems as the Gaussian ones. Also, working with
and having coefficients of variation not more than a few percent, we deal
with random perturbations having the skewness and flatness parameters very
close to zero, that is, just about the same as for the Gaussian processes them-
selves. With two processes we can study all three casesfully positive, zero
and fully negative cross-correlation between and which cover the full
range of all the possibilities between and .
We generalize the original Bernoulli equation (11.122) by introducing ran-
domness in and according to
d
= ( + 1 (x)) + ( + 2 (x)) 2, (11.143)
dx
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
where 1 (x) and 2 (x) are two zero-mean stationary Gaussian noises having
a correlation matrix

g11 g12
R= , g12 = g21 = g, g11 g22 g 2 . (11.144)
g21 g22
Here gij (i, j = 1, 2) are intensities of both noises. A small level of randomness
in and now implies

g11 g22 . (11.145)
In general, the noises 1 (x, ) and 2 (x, ) are being interpreted here as
stochastic processes equivalent, respectively, to two real processes X1 (x, )
and X2 (x, ) with sufficiently small correlation radii, that perturb the ma-
terial parameters and . The conditions for introducing the equivalent
noises 1 (x, ) and 2 (x, ), as well as their relation to the correlation func-
tions X1 (x, ) and X2 (x, ), were discussed at length in Ostoja-Starzewski

and Trebicki (1999).
Interpreting (11.143) in the Stratonovich sense again, we arrive at the
equivalent Ito equation
d = A() + B () d W (x) (x0 ) = 0 , (11.146)
where

1 3
A() = g11 + g 2 + g22 3
2 2
1
B () = g11 2 g 3 + g22 4 . (11.147)
2
Next, the transformation of variables = 1/ leads to an Ito equation for the
inverse amplitude process
1
d = A( ) + B( )d W (x0 ) = , (11.148)
0
where the drift A(z) and diffusion B(z) coefficients of the Markov process
are
1 1
A( ) = ( + g11 ) ( + g) B(z) = (g11
2
2g + g22 ). (11.149)
2 2
This leads to the average critical amplitude of the random medium
+ 12 S12
c = . (11.150)
+ 12 S1 S2
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
We note here:
1. In general, the case of full positive correlation among the noises

(g = g11 g22 ) corresponds to the situation of equivalence, in a prob-
abilistic sense, of processes 1 (x, ) and 2 (x, ) in the governing
system (11.149), and hence in (11.146), thus leading to a weaker ran-
domness (i.e., closest to a homogeneous medium) than in cases of

zero (g = 0) and of full negative cross-correlation (g = g11 g22 ),
see Figure 11.15.
2. In the case of g = 0, this model reduces to that with two independent
white noises
(x) = + 1 (x) (x) = + 2 (x) . (11.151)
3. In the case of g S1 S2 , g11 S12 , and g22 S22 , the model reduces
to that given in the preceding subsection.
Figures 11.14 and 11.15 also show that x is a random variable, whose
scatter is strongly sensitive to even weak perturbations in the material. In the
language of stochastic processes, the problem of finding x is a nonstationary
stochastic evolution problem with absorbing boundary. Although it is gov-
erned by a linear differential equation, it generally does not have an explicit

analytical solution. In Ostoja-Starzewski and Trebicki (2003) the method of
1.0
g11 = 0.001
g22 = 0.001
0.8
Trajectories of inverse amplitude
Full positive correlation

0.6
0.4 Full negative Deterministic

correlation
0.2 Zero correlation
0.0
0.0 0.4 0.8 1.2 1.6 2.0
FIGURE 11.15
Effect of cross-correlations between the noises 1 and 2 on the evolution of a single trajectory of
the inverse amplitude process. After Ostoja-Starzewski and Trebicki (2002), with permission.
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
maximum entropy was employed to find an approximate solution for the

density p (z, x) of (x).
11.5.2.3 Model with Four Correlated Noises

All the above studies were based on some choice of the wavefront thickness
L relative to the microscale d, and an implicit assumption that the ratio L/d
would not change in the course of evolution. It is more realistic, however, to
admit a change of L as a function of the wavefront amplitude , reflecting
the fact that, as grows, the wavefront thickness L decreasesbecause at
we have the shock or caustic formationshowing that the wave
becomes more and more sensitive to random microstructural details of the
material.
Motivated by this, Ostoja-Starzewski and Trebicki (2006) proposed a gen-
eralization of the original stochastic Bernoulli equation (11.122) by introduc-
0 , R (and simultaneously in processes
ing randomness in processes G 0 , G 0 , E

and , which are functions of G 0 , G 0 , E 0 , R ) in function of the amplitude ,
in the following way:
d
= (U (x, )) + (U (x, )) 2
dx
"
dU = a ()Ud x + U () 2a ()d Wx (x, )
dU
= C 1 m1 (11.152)
dx
da
= C 2 m2 ,
dx
with conditions
U (x0 ) = 1, a (x0 ) = a 0 , C1 , C2, m1 , m2, > 0. (11.153)
Here U stands for the OrnsteinUhlenbeck (O-U) process.

Thus, as grows, so does the standard deviation U of the driving random
process O-U, and its correlation length 1/a decreases. The same will also occur
0, R
to the standard deviations and correlation lengths of processes G 0 , G 0 , E
which, are linear functions of the O-U process.
It is again more convenient to express the entire dynamics in terms of the
1
inverse amplitude = so that, we have

d
= (U (x, )) (U (x, ))
dx
"
dU = a ( ) dx + ( ) a ( )d Wx (x, ) (11.154)
dU
= C1 m1
dx
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
da
= C2 m2 , (11.155)
dx
with
(x0 ) = 0 = 1, a (0 ) = a 0 . (11.156)
Our principal conclusions stemming from this model may be summarized as

follows:
1. The introduction of coupling of four random fields of material prop-

erties to the wavefront amplitude process causes growth of the vari-
ance of x , and, therefore, an even higher probability of blow-up.
2. Overall, the coupling of acceleration wavefront dynamics to mate-
rial randomness has a dominant effect irrespective of whether there
0 , and R ,
is a zero or non-zero cross-correlation among G 0 , G 0 , E
and that is an interesting point on which we end this book.
P1: Binod
May 28, 2007 18:15 C4174 C4174C011
P1: Binod
July 5, 2007 16:11 C4174 C4174C012
Bibliography
Adhikari, S. and Manohar, C.S. (1999), Dynamic analysis of framed structures with
statistical uncertainties, Int. J. Num. Meth. Eng. 44, 11571178.
Adhikari, S. and Manohar, C.S. (2000), Transient dynamics of stochastically parame-
tered beams, ASCE J. Eng. Mech. 126, 11311140.
Adler, R.J. (1981), Random Fields, John Wiley and Sons, New York.
Advani, S.G. and Tucker, C. L. (1987), The use of tensors to describe and predict fiber
orientation in short fiber composites, J. Rheology 31 (8), 751-784.
Aero, E.L. and Kuvshinskii, E. V. (1964), Continuum theory of asymmetric elasticity.
Equilibrium of an isotropic body (in Russian), Mech. Solid Body, 6(9), 2689-2699.
Aero, E.L. and Kuvshinskii, E.V. (1960), Fundamental equations of the theory of elastic
media with rotationally interacting particles, Fizika Tverdogo Tela 2, 1399; English
translation in Soviet Phys. Solid State 2, 1272 (1961).
Aero, E.L. and Kuvshinskii, E.V. (1969), Fundamental equations of the theory of elas-
ticity with rotational particle interactions (in Russian), Mech. Solid Bodies 6(9), 2689
2699.
Aifantis, E.C. (1987), The physics of plastic deformation. Int. J. Plasticity 3, 211247.
Allaire, G. and Kohn, R.V. (1993), Topology optimization and optimal shape design
using homogenization, in Bendsoe, M.P. and Mota Soares, C.A. eds., Topology Design
of Structures, Springer, New York, 207218.
Al-Ostaz, A. and Jasiuk, I. (1997), Crack initiation and propagation in materials with
randomly distributed holes, Eng. Fract. Mech. 58, 395420.
Altus, E. (2001), Statistical modeling of heterogeneous micro-beams, Int. J. Solids Struct.
38(3435), 59155934.
Altus, E. and Givli, S. (2003), Strength reliability of statistical heterogeneous mi-
crobeams, Int. J. Solids Struct. 40(9), 20692083.
Alzebdeh, K., Al-Ostaz, A., Jasiuk, I. and Ostoja-Starzewski, M. (1998), Fracture of
random matrix-inclusion composites: scale effects and statistics, Int, J . Solids Struct.
35(19), 25372566.
Alzebdeh, K. and Ostoja-Starzewski, M. (1999), On a spring network model and ef-
fective elastic moduli of granular materials, ASME J. Appl. Mech. 66, 172180.
Amieur, M., Hazanov, S., and Huet, C. (1995), Numerical and experimental assessment
of the size and boundary conditions effects for the overall properties of granular
composite bodies smaller than the representative volume, in D.F. Parker and A.H.
England, eds., IUTAM Symposium on Anisotropy, Inhomogeneity and Nonlinearity in
Solid Mechanics, Kluwer Academic, Dordrecht, 149154.
Arnold, V.I. (1978), Mathematical Methods of Classical Mechanics, Springer-Verlag,
New York.
Asch, M., Kohler, W., Papanicolau, G.C., Postel, M., and White, B. (1991), Frequency
content of randomly scattered signals, SIAM Rev. 33(4), 519625.
Ashby, M.F. and Jones, D.R.H. (1980), Engineering Materials 1: An Introduction to Their
Properties and Applications, Pergamon Press.
435
P1: Binod
July 5, 2007 16:11 C4174 C4174C012
Asimow, L. and Roth, B. (1978), The rigidity of graphs, Trans. Am. Math. Soc. 245,
279289.
Askar, A. (1985), Lattice Dynamical Foundations of Continuum Theories, World Scientific,
Singapore.
Atkinson, C. and Leppington, F.G. (1977), The effect of couple stresses on the tip of a
crack, Int. J. Solids Struct. 13, 11031122.
Bagley, R.L. and Torvik, P.J. (1983), Fractional calculusa different approach to the
analysis of viscoelastically damped structures, AIAA J. 21, 741748.
Balescu, R. (1975), Equilibrium and Nonequilibrium Statistical Mechanics, John Wiley and
Sons, New York.
Ball, J.M. and James, R.D. (2002), The scientific life and influence of Clifford Ambrose
Truesdell III, Arch. Rational Mech. Anal. 161, 126.
Banks, C.B. and Sokolowski, M. (1968), On certain two-dimensional applications of
the couple-stress theory, Int. J. Solids Struct. 4, 1529.
Baranski, W. and Wozniak, C. (1966), The fibrous body as a simply connected model
of multi-hole disks, Arch. Mech. Stos. 18(3), 273283.
Bardenhagen, S. and Triantafyllidis, N. (1994), Derivation of higher order gradient
continuum theories in 2,3D non-linear elasticity from periodic lattice models,
J. Mech. Phys. Solids 42, 111139.
Bathurst, R.J. and Rothenburg, L. (1988a), Micromechanical aspects of isotropic gran-
ular assemblies with linear contact interactions, ASME J. Appl. Mech. 55, 1723.
Bathurst, R.J. and Rothenburg, L. (1988b), Note on a random isotropic granular mate-
rial with negative Poissons ratio, Int. J. Eng. Sci. 26(4), 373383.
Bazant, Z.P. and Christensen, M. (1972), Analogy between micropolar continuum and
grid frameworks under initial stress, Int. J. Solids Struct. 8, 327346.
Bazant, Z.P. and Pang, S.-D. (2007), Activation energy based extreme value statistics
and size effect in brittle and quasibrittle fracture. J. Mech. Phys. Solids 55, 9131.
Becus, G. (1978), Wave propagation in imperfectly periodic structures: A random
evolution approach, ZAMP 29, 253261.
Becus, G. (1979), Diffusion approximation to wave propagation in imperfectly periodic
structures, ZAMP 30, 724727.
Beltzer, A.I. (1989), The effective dynamic response of random composites and
polycrystalsA survey of the causal approach, Wave Motion 11, 211229.
Belyaev, A.K. and Ziegler, F. (1995), Effective loss factor of heterogeneous elastic solids
and fluids, in Monteiro Marques, M.D.P. and Rodrigues, J.F., eds., Trends in Appli-
cations of Mathematics to Mechanics, Wiley and Sons, New York, 313.
Belytschko, T., Lu, Y.Y., and Gu, L. (1995), Crack propagation by element-free Galerkin
method, Eng. Fracture Mech. 51, 295313.
Benaroya, H. and Rehak, M. (1988), Finite element methods in probabilistic structural
analysis: A selective review, Appl. Mech. Rev. 41(5), 201213.
Bendse, M.P. and Kikuchi, N. (1988), Generating optimal topologies in structural
design using homogenization method, Comp. Meth. Appl. Mech. Eng. 71, 197224.
Bensoussan, A. Papanicolaou, G.C., and Lions, J.L. (1978), Asymptotic Analysis for Pe-
riodic Structures, NorthHolland, Amsterdam.
Beran, M.J. (1998), The use of classical beam theory for micro-beams composed of
crystals, Int. J. Solids Struct. 35(19), 24072412.
Beran, M.J. (2001), Statistical continuum mechanics: An introduction, in D. Jeulin
and M. Ostoja-Starzewski, eds., Mechanics of Random and Multiscale Microstructures,
CISM Courses and Lectures 430, Springer, NewYork, 131.
P1: Binod
July 5, 2007 16:11 C4174 C4174C012
Bibliography 437
Beran, M.J. and McCoy, J.J. (1970a), Mean field variations in a statistical sample of
heterogeneous linearly elastic solids, Int. J. Solids Struct. 6, 10351054.
Beran, M.J. and McCoy, J.J. (1970b), The use of strain gradient theory for analysis of
random media, Int. J. Solids Struct. 6, 12671275.
Berger, M.A. (1984), Central limit theorem for products of random matrices, Trans.
A.M.S. 285, 777803.
Berglund, K., (1982), Structural models of micropolar media, in O. Brulin and R.K.
Hsieh, eds., Mechanics of Micropolar Continua, World Scientific, Singapore, 585.
Bernal, J.D. (1960), A geometrical approach to the structure of liquids, Nature 185,
6870.
Bernal, J.D., Masson, J., and Knight, K.F. (1962), Radial distribution of the random
close packing of spheres, Nature 189, 956958.
Bertrand, J. (1889), Calcul des Probabilite, Gauthier-Villars, Paris.
Bird, M.D. and Steele, C.R. (1992), A solution procedure for Laplaces equation on
multiply connected circular domains, J. Appl. Mech. 59(2), 398404.
Blanc, X., LeBris, C., and Lions, P.-L. (2002), From molecular models to continuum
mechanics, Arch. Rational Mech. Anal. 164, 341381.
Bland, D.R. (1969), Nonlinear Dynamic Elasticity, Blaisdell, Waltham, MA.
Blouin, F. and Cardou, A. (1989), A study of helically reinforced cylinders under sym-
metric loads and application to strand mathematical modelling, Int. J. Solids Struct.
25(2), 189200.
Boal, D.H. (1993), Rigidity and connectivity percolation in heterogeneous polymer-
fluid networks, Phys. Rev. E 47, 46044606.
Bogy, D.B. and Sternberg, E. (1967), The effect of couple-stress on singularities due to
discontinuous loadings, Int. J. Solids Struct. 3, 757770.
Borcea, L. and Papanicolaou, G. (1998), Network approximation for transport proper-
ties of high contrast materials, SIAM J. Appl. Math. 58, (1998), 501539.
Born, M. and Huang, K. (1954), Dynamical Theory of Crystal Lattices, Clarendon Press,
Oxford.
Bouchitte, G. and Suquet, P.M. (1991), in G. Dal Maso and G.F. DellAntonio, eds.,
Composite Media and Homogenization Theory, Birkhauser, Boston, 107133.
Bouyge, F., Jasiuk, I., and Ostoja-Starzewski, M. (2001), A micromechanically based
couple-stress model of an elastic two-phase composite, in Mechanics Pan-America
2000, Int. J. Solids Struct. 38(1013), 17211735.
Bouyge, F., Jasiuk, I., Boccara, S., and Ostoja-Starzewski, M. (2002), A micromechan-
ically based couple-stress model of an elastic orthotropic two-phase composite,
Europ. J. Mech. A: Solids 21(3), 465481.
Brandstatter, J.J. and Schoenberg, M. (1975), Dynamic programming, Fermats princi-
ple and stochastic eikonal equation, J. Franklin Inst. 208, 223230.
Brekhovskikh, L.M. (1959), Propagation of surface Rayleigh waves along the uneven
boundary of an elastic body, Soviet Physics Acoustics 5(3), 288295.
Brenner, C.E. (1991), Stochastic Finite Elements (Literature Review), Internal Working
Report 3591, Institute of Engineering Mechanics, University of Innsbruck.
Brezzi, F. and Fortin, M. (1991), Mixed and Hybrid Finite Element Methods, Springer-
Verlag, Berlin.
Bronkhorst, C.A. (2003), Modelling paper as a two-dimensional elastic-plastic
stochastic network, Int. J. Solids Struct. 40, 54415454.
Brulin, O. and Hsieh, R.K.T. eds., (1982), Mechanics of Micropolar Media, Singapore,
World Scientific.
P1: Binod
July 5, 2007 16:11 C4174 C4174C012
Budiansky, B. and OConnell, R.J. (1976), Elastic moduli of a cracked solid, Int. J. Solids
Struct. 12, 8197.
Buryachenko, V.A. (2001), Multiparticle effective field and related methods in
micromechanics of composite materials, Appl. Mech. Rev. 54, 147.
Buxton, G.A., Care, C.M., and Cleaver, D.J. (2001), A lattice spring model of heteroge-
neous materials with plasticity, Modell. Simul. Mater. Sci. Eng. 9, 485497.
Campbell J.G. (1961), The in-plane elastic constants of paper, Austral. J. Appl. Sci 12,
356357.
Caputo, M. (1967), Linear models of dissipation whose Q is almost frequency
independent, Part II, Geophys. J. Royal Astron. Soc. 13, 529539.
Cardy, J.L. ed., (1988), Finite-Size Scaling. North-Holland, Amsterdam.
Carmeliet, J. and de Borst, R., (1995), Stochastic approaches for damage evolution in
standard and non-standard continua, Int. J. Solids Struct. 32(8/9), 11491160.
Carpinteri, F. and Mainardi, F., eds. (1997), Fractals and Fractional Calculus in Continuum
Mechanics, Courses and Lectures 378, Springer, Wien-New York.
Carvalho, F.C.S. and Labuz, J.F., (1996), Experiments on effective elastic modulus of
two-dimensional solids with cracks and holes, Int. J. Solids Struct. 33(28), 41194130.
De Moivre, A. (1718), The Doctrine of Chances: or, a Method of Calculating the Probability
of Events in Play, W. Pearson, London.
Castro, J. and Ostoja-Starzewski, M. (2000), Particle sieving by percolation through a
random fiber network, Appl. Math. Modell. 24(89), 523534.
Castro, J. and Ostoja-Starzewski, M. (2003), Elasto-plasticity of paper, Int. J. Plast.
19(12), 20832098.
Chen, J.Y., Huang, Y., and Ortiz, M. (1998), Fracture analysis of cellular materials: A
strain gradient model, J. Mech. Phys. Solids 46, 789828.
Chen, P.J. (1973), Growth and decay of acceleration waves in solids, in S. Flugge and
C. Truesdell, eds., Encyclopedia of Physics, VI a/3, Springer-Verlag, Berlin.
Chen, P.J. (1976), Selected Topics in Wave Propagation, Noordhoff International, Leyden
The Netherlands.
Chen, P.J. and Gurtin, M.E. (1973), On the propagation of one-dimensional acceleration
waves in laminated composites, ASME J. Appl. Mech. 40, 10551060.
Cherkaev, A.V., Lurie, K.A., and Milton, G.W. (1992), Invariant properties of the stress
in plane elasticity and equivalence classes of composites, Proc. Roy. Soc. Lond. A 438,
519529.
Chernov, L.A. (1960), Wave Propagation in a Random Medium, McGraw-Hill, New York.
Christensen, R.M. (1979), Mechanics of Composite Materials, John Wiley and Sons, New
York.
Chudnovsky, A. and Kunin, B. (1987), A probabilistic model of brittle crack formation,
J. Appl. Phys. 62(10), 41244129.
Chung, J.W., Roos, A., De Hosson, J.Th.M., and van der Giessen, E. (1996), Fracture of
disordered three-dimensional spring networks: A computer simulation methodo-
logy, Phys. Rev. B 54, 1509415100.
Cielecka, I., Wozniak, C., and Wozniak, M. (1998), Internal variables in macrodynamics
of two-dimensional periodic cellular media, Arch. Mech. 50(1), 319.
Cioranescu, D. and Saint Jean Paulin, J. (1999), Homogenization of Reticulated Structures,
Springer-Verlag, New York.
Coleman, B.D. and Gurtin, M.E. (1965), Waves in materials with memory, Arch. Rational
Mech. Anal. 19, 239265.
Collins, I.F. and Houlsby, G.T. (1997), Application of thermomechanical principles to
the modeling of geotechnical materials, Proc. Roy. Soc. Lond. A 453, 19752001.
P1: Binod
July 5, 2007 16:11 C4174 C4174C012
Bibliography 439
Contreras, H. (1980), The stochastic finite-element method, Comp. Struct. 12,

341348.
Cook, R.D., Malkus M.E., and Plesha M.E. (1989), Concepts and Applications of Finite
ELement Analysis, John Wiley and Sons, New York.
Cosserat, E. and Cosserat, F. (1896), Sur la theorie de lelasticite, Ann. de lEcole Normale
de Toulouse 10, 1.
Cosserat, E. and Cosserat, F. (1909), Theorie des Corps Deformables, A. Herman et Fils,
Paris.
Costanzo, F., Gray, G.L., and Andia, P.C. (2004), On the notion of average mechanical
properties in MD simulation via homogenization, Modell. Simul. Mater. Sci. Eng. 12,
S333S345.
Costanzo, F., Gray, G.L., and Andia, P.C. (2005), On the definition of effective stress
and deformation gradient for use in MD: Hills macro-homogeneity and the virial
theorem, Int. J. Eng. Sci. 43, 533555.
Costello, G.A. (1997), Theory of Wire Rope, Springer-Verlag, New York.
Cowin, S.C. (1969), Singular stress concentrations in plane Cosserat elasticity, Z. Angew.
Math. Phys. 20, 979982.
Cowin, S.C. (1970a), Stress functions for Cosserat elasticity, Int. J. Solids Struct. 6,
389398.
Cowin, S.C. (1970b), An incorrect inequality in micropolar elasticity theory, Z. Angew.
Math. Phys. 21, 494497.
Cowin, S.C. (1974), The theory of polar fluids, Adv. Appl. Mech. 14, 279347.
Cowin, S.C. and Nunziato, J.W. (1983), Linear elastic materials with voids, J. Elast. 13,
125.
Cox, H.L., The elasticity and strength of paper and other fibrous materials, Brit. J. Appl.
Phys. 3 (1952) 7279.
Crapo, H. and Whiteley, W. (1989), The geometry of rigid structures, in Encyclopedia of
Mathematics and its Applications, Cambridge University Press, Cambrisge.
Cristescu, N.D., Craciun, E.-M., and Soos, E. (2003), Mechanics of Elastic Composites,
Boca Raton, FL, Chapman and Hall/CRC Press.
Cundall, P.A. and Strack, O.D.L. (1979), A discrete numerical model for granular as-
semblies, Geotechnique 29(1), 4765.
Day, A.R., Snyder, K.A., Garboczi E.J., and Thorpe, M.F. (1992), The elastic moduli of
a sheet containing circular holes, J. Mech. Phys. Solids 40, 10311051.
de Buhan, P. (1983), Homogeneisation en calcul a la rupture: le cas du materiaux
composite multicouche, C.R. Acad. Sci. Paris II 296(11), 933936.
De Cicco, S. and Guarracino, F. (2004), A complex representation in porous elasticity,
J. Bravo-Castillero, R. Rodriguez-Ramos, and M. Ostoja-Starzewski, M., eds., in
Applied Mechanics in the Americas, ASME Press, New York, 10, 440443.
Dence, D. and Spence, J.E. (1970), Wave propagation in random anisotropic media, in
A.T. Bharucha-Reid, ed., Probabilistic Methods in Applied Mathematics, 2, Academic
Press, New York, 121181.
Deng, M. and Dodson, C.T.J. (1994), Paper: An Engineered Stochastic Structure, TAPPI
Press, Atlanta.
Deodatis, G. and Graham, L. (1997), The weighted integral method and the vari-
ability response functions as part of a SFEM formulation, Uncertainty Modeling
in Finite Element, Fatigue and Stability of Structures, World Scientific, Singapore,
71116.
Dewar, R.C., (2005), Maximum entropy production and the fluctuation theorem, J.
Phys. A: Math., Gen, 38, L371L381.
P1: Binod
July 5, 2007 16:11 C4174 C4174C012
Dhaliwal, R.S. and Singh, A. (1987), Micropolar thermoelasticity, in R.B. Hetnarski ed.,
Thermal Stresses, 2, Elsevier Science Publishers, Amsterdam, 267328.
Dhont, G. (2004), The Finite Element Method for Three-Dimensional Thermomechanical
Applications, John Wiley and Sons, Munich.
Ditlevsen, O. (1996), Dimension reduction and discretization in stochastic problems
by regression method, F. Casciati and B. Roberts, Eds., in Mathematical Models for
Structural Reliability Analysis, CRC Press, Boca Raton, FL, 51138.
Dodson, C.T.J. and Sampson, W.W. (1999), Spatial statistics of stochastic fibre networks.
J. Stat. Phys. 96(1/2), 447458.
Dowling, N.E. (1993), Mechanical Behavior of Materials: Engineering Methods for
Deformation, Fracture, and Fatigue, Prentice-Hall, Englewood Cliffs, NJ.
Doyle, J.F. (1997), Wave Propagation in Structures: Spectral Analysis Using Fast Discrete
Fourier Transforms, Springer-Verlag, Berlin.
Drzal, L.T. (1990), The role of the fiber/matrix interphase on composite properties,
Vacuum 41, 16151618.
Du, X. and Ostoja-Starzewski, M. (2006a), On the size of representative volume element
for Darcy law in random media, Proc. Roy. Soc. Lond. A 462, 29492963.
Du, X. and Ostoja-Starzewski, M. (2006b), On the scaling from statistical to repre-
sentative volume element in thermoelasticity of random materials, Networks and
Heterogeneous Media 1(2), 259274.
Dullien, F.A.L. (1979), Porous Media: Fluid Transport and Pore Structure, Academic Press,
New York.
Dundurs, J. (1967), Effect of elastic constants in certain contact problems, J. Comp.
Mater. 1, 310322.
Dundurs, J. (1968), Discussion, ASME J. Appl. Mech. 36, 650652.
Dundurs, J. (1990), Boundary conditions at interface, in G.J. Weng, M. Taya, and H.
Abe, eds., Micromechanics and Inhomogeneity: The Toshio Mura Anniversary Volume,
Springer-Verlag, Berlin, 109114.
Dundurs, J. and Jasiuk, I. (1997), Effective elastic moduli of composite materials: Re-
duced parameter dependence, Appl. Mech. Rev. 50 (No. 11, Part 2), S39S43.
Dundurs, J. and Markenscoff, X. (1993), Invariance of stresses under a change in elastic
compliances, Proc. Roy. Soc. London A 443, 289300.
Dvorak, G.J. and Bahei-Ei-Din, Y.A.(1987), A bimodal plasticity theory of fibrous com-
posite materials, Acta Mech. 69, 219241.
Dyszlewicz, J. (2004), Micropolar Theory of Elasticity, Springer-Verlag, Berlin.
E, W., Engquist, B. Li, X., Ren, W., Vanden-Eijnden, E. (2005), The Heterogeneous Multi-
scale Method: A Review, http://www.math.princeton.edu/multiscale/review.pdf.
Ehrentraut, H., Blenk, S., and Muschik, W. (1991), Statistical foundations for liquid
crystals in alignment tensor formulation, Physica A 174, 119138.
Eimer, Cz. (1971), The viscoelasticity of multi-phase media, Arch. Mech. 23(1),
315.
Elishakoff, I. and Yongjian R. (2003), Finite Element Methods for Structures with Large
Stochastic Variations, Oxford University Press, London.
Elmore, W.C. and Heald, M.A. (1969), Physics of Waves, McGraw-Hill, New York.
Engelbrecht, J. (1997), Nonlinear Wave Dynamics: Complexity and Simplicity, Kluwer
Academic Publishers, Dordrecht.
Englman, R., Rivier, N., and Jaeger, A. (1987), Fragment-size dsitribution in disinte-
gration by maximum-entropy method, Philos. Mag. B 56(6), 751769.
Epstein, M. and Sniatycki, J. (1992), Fermats principle in elastodynamics, J. Elast. 27,
4556.
P1: Binod
July 5, 2007 16:11 C4174 C4174C012
Bibliography 441
Ericksen, J.L. and Truesdell, C. (1958), Exact theory of stress and strain in rods and
shells, Arch. Rational Mech. Anal. 1, 295323.
Eringen, A.C. (1966), Linear theory of micropolar elasticity, J. Math. Mech. 15, 909.
Eringen, A.C. (1968), Theory of micropolar elasticity, in H. Liebowitz, ed., Fracture
Mechanics, an Advanced Treatise 2, Academic Press, New York, 621729.
Eringen, A.C. (1999), Microcontinuum Field Theories. I. Foundations and Solids, Springer-
Verlag, New York.
Eringen, A.C. (2001), Microcontinuum Field Theories. II. Fluent Media, Springer-Verlag,
New York.
Eringen, A.C. and Hanson, G.W. (2002), Non-Local Continuum Field Theories, Springer-
Verlag, New York.
Evans, D.J. and Searles, D.J. (1994), Equilibrium microstates which generate second
law violating steady states, Phys. Rev. E 50, 16451648.
Ferrari, M., Granik, V.T., Imam, A., and Nadeau, J.C., eds., (1997), Advances in Doublet
Mechanics, Springer-Verlag, New York.
Fishman, L. and McCoy, J.J. (1981), A unified view of bulk property theories for stochas-
tic and periodic media, J. Appl. Math. Phys. (ZAMP) 32, 4561.
FLAC (1999), Itasca Consulting Group, Inc., Minneapolis, MN.
Fleck, N.A. and Hutchinson, J.W. (1997), Strain gradient plasticity, Adv. Appl. Mech. 33,
295361, Academic Press, New York.
Fleck, N.A., Muller, G.M., Ashby, M.F., and Hutchinson, J.W. (1994), Strain-gradient
plasticity: Theory and experiment, Acta Metall. Mater. 42, 475487.
Foias, C. and Frazho, A.E. (1990), The Comutant Lifting Approach to Interpolation Problems,
Birkhauser Verlag, Basel.
Forest, S. (1998), Mechanics of generalized continua: Construction by homogenization,
J. Phys. IV 8, Pr4-3948.
Forest, S. (1999), Aufbau und Identifikation von Stoffgleichungen fur hohere Kontinua
mittels Homogenisierungsmethoden, Technische Mechanik, Band 19(4), 297306.
Forest, S. and Sab, K. (1998), Cosserat overall modeling of heterogeneous materials,
Mech. Res. Comm. 25, 449454.
Forest, S. and Sievert, R. (2003), Elastoviscoplastic constitutive frameworks for
generalized continua, Acta Mech. 160, 71111.
Forest, S., Dendievel, R., and Canova, G. (1999), Estimating the overall properties of
heterogeneous Cosserat materials, Modell. Simul. Mater. Sci. Eng. 7, 829840.
Forest, S., Barbe, F., and Cailleteaud, G. (2000), Cosserat modelling of size effects in
the mechanical behaviour of polycrystals and multi-phase materials, Int. J. Solids
Struct. 37, 71057126.
Forest, S., Pradel, F., and Sab, K. (2001), Asymptotic analysis of Cosserat media, Int. J.
Solids Struct. 38, 45854608.
Friesecke, G. and James, R.D. (2000), A scheme for the passage from atomic to con-
tinuum theory for thin films, nanotubes and nanorods, J. Mech. Phys. Solids 48,
15191540.
Frisch, U. (1968), Wave propagation in random media, in A.T. Bharucha-Reid, ed.,
Probabilistic Methods in Applied Mathematics, 1; Academic Press, New York, 75198.
Frisch, U. (1995), Turbulence, Cambridge University Press, Cambridge.
Gani, J. (1972), A car parking problem, Period. Math. Hung. 2, 6172.
Garboczi, E.J. (1987), Cauchy relations for central force random networks, Phys. Rev.
B 36, 21152120.
Garboczi, E. (1998), Finite element programs and finite difference programs for
computing the linear electric and elastic properties of digital images of random
P1: Binod
July 5, 2007 16:11 C4174 C4174C012
materials, NISTIR 6269, National Institute of Standards and Technology,

Gaithersburg, MD.
Garboczi, E.J., Thorpe, M.F., DeVries, M.S. and Day, A.R. (1991), Universal conductance
curve for a plane containing random holes, Phys. Rev. A 43, 64736480.
Gauthier, R.D. and Jahsman, W.E. (1975), A quest for micropolar elastic constants,
ASME J. Appl. Mech. 42, 369374.
Gdoutos, E.E. (1993), Fracture Mechanics: An Introduction, Kluwer Academic Publishers,
Dordrecht.
Germain, P., Quoc Son, N., and Suquet, P. (1983), Continuum thermodynamics, ASME
J. Appl. Mech. 50, 10101020.
Ghandour, E. (1977), Propagation of disturbances in a random medium, J. Inst. Math.
Appl. 20, 393407.
Ghanem, R.G. and Spanos, P.D. (1991), Stochastic Finite Elements: A Spectral Approach,
Springer-Verlag, Berlin.
Gibbs, J.W. (1903), Elementary Principles in Statistical Mechanics, New York: C. Scribners
Sons.
Gibiansky, L. and Torquato, S. (1998), Effective energy of nonlinear elastic and con-
ducting composites: Approximation and cross-property bounds, J. Appl. Phys. 84,
59695976.
Gibson, L.J. and Ashby, M.F. (1988), Cellular Solids, Pergamon Press.
Givli, S. and Altus, E. (2003), Effect of modulus-strength correlation on the reliability
of randomly heterogeneous beams, Int. J. Solids Struct. 40(24), 67036722.
Gloria, A., (2006a), A direct approach to numerical homogenization in finite elasticity,
Networks & Heterogeneous Media 1(1), 109141. Erratum: 513514.
Gloria, A., (2006b), An analytical framework for the numerical homogenization of
monotone elliptic operators and quasiconvex energies, Multiscale Model. Simul. 5(3),
9961043.
Goddard, J.D. (2001), Delaunay triangulation of granular media, in N. Bicanic,
ed., Proc. ICADD-4, Discontinuous Deformation Analysis, University of Glasgow,
281293.
Goddard, J.D. (2007), On entropy estimates of contact forces in static granular assem-
blies, Int. J. Solids Struct. 41(21), 58515861.
Goddard, J.D. (2006), From granular matter to generalized continuum, in P. Mariano
et al., eds., Mathematical Models for Granular Matter, Lecture Notes in Mathematics,
Springer-Verlag, in press.
Grah, M., Alzebdeh, K., Sheng, P.Y., Vaudin, M.D., Bowman, K.J., and Ostoja-
Starzewski, M. (1996), Brittle intergranular failure in 2D microstructures: Experi-
ments and computer simulations, Acta Mater. 44(10), 40034018.
Green, A.E. and Naghdi, P.M. (1965), Plasticity theory and multipolar continuum
mechanics, Mathematica 12, 2126.
Green, A.E. and Rivlin, R.S. (1964a), Multipolar continuum mechanics, Arch. Rational
Mech. Anal. 17, 113147.
Green, A.E. and Rivlin, R.S. (1964b), Multipolar continuum mechanics, Arch. Rational
Mech. Anal. 17, 113147.
Greenspan, D. (1989), Supercomputer simulation of cracks and fractures by quasi-
molecular dynamics, J. Phys. Chem. Solids 50(12), 12451249.
Greenspan, D. (1997), Particle Modeling, Birkhauser Publishing, Basel.
Greenspan, D. (2002), New approaches and new applications for computer simulation
of N-body problems, Acta Appl. Matha. 71, 279313.
P1: Binod
July 5, 2007 16:11 C4174 C4174C012
Bibliography 443
Griffith, A.A. (1921), The phenomena of rupture and flow in solids, Philos. Trans. Roy.
Soc. Lond. A 221, 163198.
Grioli, G. (1960), Elasticita asimetrica, Ann. Mat. Pura Appl. Ser. IV, 50, 389417.
Guest, S.D., (2000), Tensegrities and rotating rings of tetrahedra: a symmetry viewpoint
of structrual mechanics, Phil. Trans. Roy. Soc. Lond. 358, 229243.
Gupta, S. and Manohar, C.S. (2002), Dynamic stiffness method for circular stochastic
Timoshenko beams: Response variability and reliability analysis, J. Sound Vibration
253(5), 10511085.
Gusev, A.A. (1997), Representative volume element size for elastic composites:
A numerical study, J. Mech. Phys. Solids 45, 144914.
Gusev, A.A. (2004), Finite element mapping for spring network representations of the
mechanics of solids, Phys. Rev. Lett. 93(2), 0343021-034302-4.
Gunther, A.C. (1958), Zur Statik und Kinematik des Cosseratschen Kontinuums, Abh.
Braunschweig Wiss. Ges. 10, 195213.
Hansen, J. C., Chien, S., Skalak, R., and Hoger, A. (1996), An elastic network model
based on the structure of the red blood cell membrane skeleton, Biophys. J. 70,
146166.
Hansen, J. C., Chien, S., Skalak, R., and Hoger, A. (1997), Influence of network
topology on the elasticity of the red blood cell membrane skeleton, Biophys. J. 72,
23692381.
Hartranft, R.J. and Sih, G.C. (1965), The efect of couple-stresses on stress concentration
of a circular inclusion, J. Appl. Mech. 32, 429431.
Hashin, Z. (1983), Analysis of composite materials: A survey, J. Appl. Mech. 50, 481.
Hazanov, S. (1998), Hill condition and overall properties of composites, Arch. Appl.
Mech. 68, 385394.
Hazanov, S. (1999), On apparent properties of nonlinear heterogeneous bodies smaller
than the representative volume, Acta Mech. 138, 123134.
Hazanov, S. and Amieur, M. (1995), On overall properties of elastic heterogeneous
bodies smaller than the representative volume, Int. J. Eng. Sci. 23, 12891301.
Hazanov, S. and Huet, C. (1994), Order relationships for boundary conditions effect
in the heterogeneous bodies smaller than the representative volume, J. Mech. Phys.
Solids 42, 1995-2011.
He, Q.-C., Le Quang, H. and Feng, Z.-Q. (1987), Exact results for the homogenization
of elastic fiber reinforced solids at finite strain, J. Elast., 83, 153177.
He, Q.-C. and Zheng, Q.-S. (1987), On the symmetries of 2D elastic and hyperelastic
tensors, J. Elast. 43, 203225.
He, R. (2001), Effects of size and boundary conditions on the yield strength of hetero-
geneous materials, J. Mech. Phys. Solids 49, 25572575.
Hegemier, G.A. and Prager, W. (1969), On Michell trusses, Int. J. Mech. Sci. 11, 209215.
Hehl, F.W. and Itin, Y. (2002), The Cauchy relations in linear elasticity, J. Elast. 66,
185192.
Herrmann, G. and Achenbach, J.D. (1968), in Kroner, E., ed., Mechanics of Generalized
Continua, Proc. IUTAM Symposium, Springer-Verlag, Berlin, 6979.
Herrmann, H.J. and Luding, S. (1998), Modeling granular media on the computer,
Cont. Mech. Thermodyn. 10(4), 189231.
Herrmann, H.J. and Roux, S. (1990), Statistical Models for the Fracture of Disordered Media,
North-Holland, Amsterdam.
Hersch, R. (1974), Random evolutions: A survey of results and problems, Rocky Mtn.
J. Math. 4, 443477.
P1: Binod
July 5, 2007 16:11 C4174 C4174C012
Hersch, R. and Pinsky, M. (1972), Random evolutions are asymptotically Gaussian,

Comm. Pure Appl. Math. 25, 3344.
Hill, R. (1950), The Mathematical Theory of Plasticity, Clarendon Press, Oxford.
Hill, R. (1963), Elastic properties of reinforced solids: Some theoretical principles,
Hill, R. (1972), On constitutive macro-variables for heterogeneous solids at finite strain,
Proc. Roy. Soc. Lond. A. 326,131147.
Hilton, H.H. (2001), Implications and constraints of time-independent Poisson ratios
in linear isotropic and anisotropic viscoelasticity, J. Elast. 63, 221251.
Hockney, R.W. and Eastwood, J.W. (1999), Computer Simulation Using Particles, Taylor
and Francis, London.
Hollister, S.J. and Kikuchi, N. (1992), A comparison of homogenization and standard
mechanics for periodic microstructure, Comput. Mech. 10, 7395.
Holnicki-Szulc, J. and Rogula, D. (1979a), Non-local, continuum models of large
engineering structures, Arch. Mech. 31(6), 793802.
Holnicki-Szulc, J. and Rogula, D. (1979b), Boundary problems in non-local, continuum
models of large engineering structures, Arch. Mech. 31(6), 803811.
Horio, M. and Onogi, S. (1951), Dynamic measurements of physical properties of pulp
and paper by audiofrequency sound, J. Appl. Phys. 22, 971977.
Hornung, U., ed., (1997), Homogenization and Porous Media, Springer-Verlag, New York.
Houlsby, G.T. and A.M. Puzrin (2000), A thermomechanical framework for constitutive
models for rate-independent dissipative materials, Int. J. Plast. 16, 10171047.
Howe, M.S. (1971), Wave propagation in random media, J. Fluid Mech. 45(4),
769783.
Hrennikoff, A. (1941), Solution of problems of elasticity by the framework method,
ASME J. Appl. Mech. 8, A619A715.
Hudson, J.A. (1968), The scattering of elastic waves by granular media, Quart. J. Mech.
Appl. Math. 21(4), 487502.
Hudson, J.A. (1980), The Excitation and Propagation of Elastic Waves, Cambridge
University Press, Cambridge.
Huet, C. (1982), Universal conditions for assimilation of a heterogeneous material to
an effective medium, Mech. Res. Comm. 9(3), 165170.
Huet, C. (1990), Application of variational concepts to size effects in elastic heteroge-
neous bodies, J. Mech. Phys. Solids 38, 813841.
Huet, C. (1991), Hierarchies and bounds for size effects in heterogeneous bodies, in
G.A. Maugin, ed., Continuum Models and Discrete Systems, 2, Longman Scientific and
Technical, London, 127134.
Huet, C. (1995), Bounds and hierarchies for the overall properties of viscoelastic
heterogeneous and composite materials, Arch. Mech. 47(6), 11251155.
Huet, C. (1997), An integrated micromechanics and statistical continuum thermo-
dynamics approach for studying the fracture behaviour of microcracked hetero-
geneous materials with delayed response, Eng. Fract. Mech. 58(56), 459.
Huet, C. (1999a), Coupled size and boundary-condition effects in viscoelastic hetero-
geneous and composite bodies, Mech. Mater. 31(12), 787829.
Huet, C. (1999b), Strength scaling law for elastic materials with interacting defects, in
G. Pijaudier-Cabot, Z. Bittnar, and B. Gerard, eds., Mechanics of Quasi-Brittle Materials
and Structures, Hermes, Paris, 3138.
Hutchinson, J.W. (2000), Plasticity in the micron scale, Int. J. Solids Struct. 37, 225238.
Huyse, L. and Maes, M.A. (2001) Random field modeling of elastic properties using
homogenization, ASCE J. Eng. Mech. 127 (1), 2736.
P1: Binod
July 5, 2007 16:11 C4174 C4174C012
Bibliography 445
Ignaczak, J. (1970), Radiation conditions of Sommerfeld type for elastic materials with
microstructure, Bull. Acad. Pol. Sci. Ser. Sci. Tech. 18, 251266.
Ishimaru, A. (1978), Wave Propagation and Scattering in Random Media, I and II, Academic
Press, New York.
Itou, S. (1973), The effect of couple-stresses on the stress concentration around an
elliptic hole, Acta Mech. 16, 289296.
Jagota, A. and Bennison, S.J. (1993), Spring-network and finite-element models for
elasticity and fracture, in K.K. Bardhan, B.K. Chakrabarti, and A. Hansen, eds.,
Breakdown and Non-Linearity in Soft Condensed Matter, Lecture Notes in Physics, 437,
Jagota, A. and Benison S.J. (1994), Spring-network and finite-element models for
elasticity and fracture, in K.K. Bardhan, B.K. Chakrabarti, and A. Hansen, eds.,
Non-Linearity and Breakdown in Soft Condensed Matter, Lecture Notes in Physics, 437,
Springer-Verlag, New York, 186201.
Jasiuk, I. (1995), Polygonal cavities vis-a-vis rigid inclusions: Effective elastic moduli
of materials with polygonal inclusions, Int. J. Solids Struct. 32, 407422. [Jasiuk, I
(2005), unpublished notes]
Jasiuk, I. and Boccara, S. (2002), On the reduction of constants in plane elasticity with
eigenstrains, Arch. Mech. 54(516), 425437.
Jasiuk, I. and Ostoja-Starzewski, M. (1995), Planar Cosserat elasticity of materials
with holes and intrusions, Appl. Mech. Rev. (Special issue: Mechanics Pan-America
1995, L.A. Godoy, S.R. Idelsohn, P.A.A. Laura, and D.T. Mook, eds.) 48(11, Part 2),
S11S18.
Jasiuk, I. and Ostoja-Starzewski, M. (2004), Modeling of bone at a single lamella level,
Biomech. Model. Mechanobiol. 3, 6774.
Jasiuk, I., Chen, J., and Thorpe, M.F. (1992), Elastic moduli of composites with rigid
sliding inclusions, J. Mech. Phys. Solids 40, 373391.
Jasiuk, I., Chen, J., and Thorpe, M.F. (1994), Elastic moduli of two dimensional materials
with polygonal holes, Appl. Mech. Rev. 47(1, Part 2), S18S28.
Jaunzemis, W. (1967), Continuum Mechanics, McGraw-Hill, New York.
Jaynes, E.T. (1973), The well-posed problem, Found. Phys. 3, 447493.
Jeulin, D. (1993), Damage simulation in heterogeneous materials from geodesic
propagations, Eng. Comp. 10(1), 8191.
Jeulin, D. (1999), Morphological modeling of random composites, Continuum Models
and Discrete Systems 8, 199206.
Jeulin, D. and Ostoja-Starzewski, M. (2000), Shear bands in elasto-plastic response of
random composites by geodesics, Proc. 8th ASCE Spec. Conf. Probab. Mech. Struct.
Reliability, ASCE Press, University of Notre Dame, IN.
Jeulin, D. and Ostoja-Starzewski, M., eds. (2001), Mechanics of Random and Multiscale
Microstructures, CISM Courses and Lectures, 430, Springer, New York.
Jiang, M., Alzebdeh, K., Jasiuk, I., and Ostoja-Starzewski, M. (2001a), Scale and
boundary conditions effects in elastic properties of random composites, Acta Mech.
148(14), 6378.
Jiang, M., Ostoja-Starzewski, M., and Jasiuk, I. (2001b), Scale-dependent bounds on
effective elastoplastic response of random composites, J. Mech. Phys. Solids 49(3),
655673.
Jones, R.M. (1998), Mechanics of Composite Materials, Taylor and Francis, New York.
Jun, S. and Jasiuk, I. (1993), Elastic moduli of two-dimensional composites with sliding
inclusions: A comparison of effective medium theories, Int. J. Solids Struct. 30,
25012503.
P1: Binod
July 5, 2007 16:11 C4174 C4174C012
Kachanov, L.M. (1971), Foundations of the Theory of Plasticity, North-Holland,

Amsterdam.
Kachanov, M. (1993), Elastic solids with many cracks and related problems, Adv. Appl.
Mech. 30, 259.
Kaloni, P.N. and Ariman, T. (1967), Stress concentration effects in micropolar elasticity,
Z. Angew. Math. Phys. 18, 136141.
Kaminski, M. (2002a), Stochastic perturbation approach to engineering structure
vibrations by the finite difference method, J. Sound Vib. 251(4), 651670.
Kaminski, M.M. (2002b), Computational Mechanics of Composite Materials: Sensitivity,
Randomness and Multiscale Behaviour, Springer-Verlag, New York.
Kamrin, K. and Bazant, M.Z. (2007), A stochastic flow rule for granular materials, to
appear in Phys. Rev. E.
Kanit, T., Forest, S., Galliet, I., Monoury, V., and Jeulin, D. (2003), Determination of the
size of the representative volume element for random composites: Statistical and
numerical approach, Int. J. Solids Struct. 40, 36473679.
Keating, P.N. (1966), Effect of invariance requirements on the elastic strain energy of
crystals with application to the diamond structure, Phys. Rev. 145, 637645.
Keller, J.B. (1962), Wave propagation in random media, Proc. Symp. Appl. Math. 13,
227243.
Keller, J.B. (1964), A theorem on the conductivity of a composite medium, J. Math.
Phys. 5, 548549.
Kellomaki, M., Astrom, J., and Timonen, J. (1996), Rigidity and dynamics of random
spring networks, Phys. Rev. Lett. 77(13), 27302733.
Kendall, M.G. and Moran, P.A.P. (1963), Geometrical Probability, Charles Griffin,
London.
Kennett, B.L.N. (1981), Elastic wave propagation in stratified media, Adv. Appl. Mech.
21, 79167.
Kennett, B.L.N. (1983), Seismic Wave Propagation in Stratified Media, Cambridge
University Press, Cambridge.
Kharanen, V.Y. (1953), Sound propagation in a medium with random fluctuation of
the refractive index, Dokl. Acad. Nauk SSSR 88, 253.
Khisaeva, Z.F. (2006), Scale Effects in Finite Elasticity and Thermoelasticity, Ph.D. thesis,
McGill University.
Khisaeva, Z.F. and Ostoja-Starzewski, M. (2006a), Mesoscale bounds in finite elas-
ticity and thermoelasticity of random composites, Proc. Roy. Soc. Lond. A 462,
11671180.
Khisaeva, Z.F. and Ostoja-Starzewski, M. (2006b), On the size of RVE in finite elasticity
of random composites, J. Elast. 85, 153173.
Khisaeva, Z.F. and Ostoja-Starzewski, M. (2007), Scale effects in infinitesimal and finite
thermoelasticity of random composites, J. Thermal Stresses, in press.
Kirkwood, J.G. (1939), The skeletal modes of vibration of long chain molecules, J. Chem.
Phys. 7, 506509.
Kleiber, M. and Hien, T.D. (1993), The Stochastic Finite Element Method, J. Wiley and
Sons, New York.
Koiter, W. (1963), Couple-stresses in the theory of elasticity, Proc. K. Ned. Akad. Wet. B
67, 1744.
Kolmogorov, A.N. (1933), Grundbegriffe der Wahrscheinlichkeitsrechnung, Springer-
Verlag, Berlin.
Kotulski, Z. (1994), On the effective reflection properties of the randomly segmented
elastic bar, Eur. J. Mech. A Solids 13(5), 677696.
P1: Binod
July 5, 2007 16:11 C4174 C4174C012
Bibliography 447
Kotulski, Z. and Pretczy nski, Z. (1994), Wave pulses in one-dimensional randomly

defected thermoelastic media, Arch. Mech. 46(3), 357373.
Krajcinovic, D. (1996), Damage Mechanics, North-Holland, Amsterdam.
Kreher, W. and Pompe, W. (1989), Internal Stresses in Heterogeneous Solids, Akademie-
Verlag, Berlin.
Kroner, E. (1963), On the physical reality of torque stresses in continuum mechanics,
Int. J. Eng. Sci. 1, 261278.
Kroner, E., ed. (1968), Mechanics of Generalized Continua, Proceedings of the IUTAM
Symposium, Springer-Verlag, Berlin.
Kroner, E. (1972), Statistical Continuum Mechanics, CISM Courses and Lectures,
Springer, New York.
Kroner, E. (1986), The statistical basis of polycrystal plasticity, J. Gittus, J. Zarka, and
S. Nemat-Nasser, eds., in Large Deformations of Solids: Physical Basis and Mathematical
Modelling, Elsevier, Amsterdam.
Kroner, E. (1994), Nonlinear elastic properties of micro-heterogeneous media, ASME
J. Eng. Mat. Tech. 11, 325330.
Kruyt, N.P. and Rothenburg, L., (1998), Statistical theories for the elastic moduli of two-
dimensional assemblies of granular materials, Int. J. Eng. Sci. 36(10), 11271142.
Ku, A.P.-D. and Nordgren R.P. (2001), On plastic collapse of media with random yield
strength, ASMEM J. Mech. 68, 715724.
Kunin, B. (1994), A stochastic model for slow crack growth in brittle materials, Appl.
Mech. Rev. 47, 175183.
Kuznetsov, E.N. (1991), Underconstrained Structural Systems, Springer-Verlag,
New York.
Lakes, R. (2001), Elastic and viscoelastic behavior of chiral materials, Int. J. Mech. Sci.
43, 15791589.
Lakes, R.S. (1983), Size effects and micromechanics of a porous solid, J. Mat. Sci. 18,
25722580.
Lakes, R.S. (1986), Experimental microelasticity of two porous solids, Int. J. Solids
Struct. 22, 5563.
Lakes, R.S. (1995), Experimental methods for study of Cosserat elastic solids and other
generalized continua, in H. Muhlhaus, ed., Continuum Models for Materials with
Micro-Structure, John Wiley and Sons, New York, 122.
Lakes, R.S. and Benedict, R. (1982), Noncentrosymmetry in micropolar elasticity, Int.
J. Eng. Sci. 29(10), 11611167.
Lakhtakia, A. (1994), Beltrami Fields in Chiral Media, World Scientific, Singapore.
Lakhtakia, I.A., Varadan, V.K., and Varadan, V.V. (1989), Time-Harmonic Electromagnetic
Fields in Chiral Media, Lecture Notes in Physics, 335, Springer-Verlag, Berlin.
Laman, (1970), On graphs and rigidity of plane skeletal structures, Eng. Math. 4,
331340.
Laplace, P.S. (1814), Essai Philosophique sur les Probabilites, Paris, V. Couricer.
Lee, S.J. and Shield, R.T. (1980), Variational principles in finite elastostatics, ZAMP 31,
437453.
Lematre, J. and Chaboche, J.-L. (1994), Mechanics of Solid Materials, Cambridge Uni-
versity Press.
Li, W. and Ostoja-Starzewski, M. (2006), Yield of random elasto-plastic materials, J.
Mech. Mater. Struct., 1, 1055-1073.
Liszka, T. and Ostoja-Starzewski, M. (2003), Effects of microscale material random-
ness on the attainment of optimal structural shapes, Struct. Multidisc. Optim. 25,
110.
P1: Binod
July 5, 2007 16:11 C4174 C4174C012
Lohnert, S. and Wriggers, P. (2003) Homogenisation of microheterogeneous materials

considering interfacial delamination at finite strains, Tech. Mechanik 23, 167177.
Lomakin, V.A. (1965), Deformation of microscopically nonhomogeneous elastic bodies
(in Russian), Prikl. Mat. Mekh. (Appl. Math. Mech.) 29(5), 888893.
Lomakin, V.A. (1970), Stochastic Problems in Mechanics of Deformable Solids, Nauka,
Moscow.
Love, A.E.H. (1934), The Mathematical Theory of Elasticity, Cambridge University Press,
Cambridge.
Lubarda, V.A. and Markenscoff, X. (2000), Conservation integrals in couple stress
elasticity, J. Mech. Phys. Solids 48, 553564.
Macvean, D.B. (1968), Die elementararbeit in einem Kontinuum und die Zuordnung
von Spannungs und Verzerrungstensoren, ZAMP 19, 157185.
Mandel, J. (1966), Proc, 11th Cong. Appl. Mech., Springer-Verlag, Berlin.
Manohar, C. S. and Adhikari, S. (1998), Dynamic stiffness matrix of randomly para-
metered beams, Probab. Eng. Mech. 13, 3952.
Mardia, K.V. (1972), Statistics of Directional Data, Academic Press, London.
Mariano, P.M. (2001), Multifield theories in mechanics of solids, Adv. Appl. Mech.
38, 193.
Mark, R.E., ed. (1983), Handbook of Physical and Mechanical Testing of Paper and Paper-
board, 1, Marcel Dekker, New York.
Markenscoff, X. (1996), A note on strain jump conditions and Cesaro integrals for
bonded and slipping inclusions, J. Elast. 45, 3344.
Markenscoff, X. and Jasiuk, I. (1998), On multiple-connectivity and the reduc-
tion of constants for composites with body forces, Proc. Roy. Soc. Lond. A 454,
13571369.
Markov, K. (2000), Elementary micromechanics of heterogeneous media, in Markov,
K. Preziosi, L., eds., Heterogeneous Media, Birkhauser, Basel.
Martinsson, P.G. and Babuska, I. (2007), Mechanics of materials with periodic truss or
frame micro-structures, Arch. Rational Mech. Anal., to appear.
Martinsson, P.G. and Babuska, I. (2006), Mechanics of materials with periodic truss or
frame micro-structures II. Homogenization, Yale Research Report YALEU/DCS/
RR-1267, Yale University Press, New Haven, CN.
Matheron, G. (1975), Random Sets and Integral Geometry, John Wiley and Sons, New
York.
Matheron, G. (1978), Estimer et Choisir, Fascicules du CGMM n 7, Ecole des Mines de
Paris (Estimating and Choosing, 1998, translated from the French by A.M. Hasofer;
Berlin, New York: Springer-Verlag).
Maugin, G. A. (1974), Acceleration waves in simple and linear viscoelastic micropolar
materials, Int. J. Eng. Sci. 12, 143157.
Maugin, G.A. (1999), The Thermomechanics of Nonlinear Irreversible Behaviors: An Intro-
duction, World Scientific, Singapore.
Maugin, G.A. (2002), Private Communication.
Maxwell, J.C. (1869), Scientific Papers II.
McCarthy, M.F. (1975), Singular surfaces and waves, in A.C. Eringen, ed., Continuum
Physics, 2, Academic Press, New York, 449521.
McCoy, J.J. (1991), Conditionally averaged response formulations for two-phase
random mixtures, J. Appl. Mech. 58, 973981.
Mellor, P.B. (1982), Experimental studies of plastic anisotropy in sheet metal, in H.G.
Hopkins and M.J. Sewell, eds., Mechanics of Solids. The Rodney Hill 60th Anniversary
Volume, Pergamon Press, New York, 383415.
P1: Binod
July 5, 2007 16:11 C4174 C4174C012
Bibliography 449
Mendelson, K.S. (1975), Effective conductivity of two-phase material with cylindrical

phase boundaries, J. Appl. Phys. 46, 917918.
Menon, V.V., Sharma, V.D., and Jeffrey, A. (1983), On the general behavior of acceler-
ation waves, Applicable Anal. 16, 101120.
Michell, A.G.M. (1904), The limits of economy in frame structures, Philos. Mag. 8,
589597.
Michell, J.M. (1899), On the direct determination of stress in an elastic solid with
applications to the theories of plates, Proc. Lond. Math. Soc. 31, 100125.
MicroMorph (1999), Ecole des Mines de Paris, France.
Miles R.E. (1964), Random polygons determined by random lines in a plane, Proc.
Natl. Acad. Sci. U.S.A. 52, 901907.
Milstein, F. (1971), Theoretical strength of a perfect crystal, Phys. Rev. B 3, 11301141.
Milton, G.W. (2002), The Theory of Composites, Cambridge University Press, Cambridge.
Mindlin, R.D. (1963), Influence of couple-stresses on stress concentrations, Exper. Mech.
3, 17.
Mindlin, R.D. and Salvadori, M.G. (1950), Analogies, in M. Hetenyi, ed., Handbook of
Experimental Stress Analysis, 700827.
Mindlin, R.D. and Tiersten, H.F. (1962), Effects of couple-stresses in linear elasticity,
Arch. Rational Mech. Anal. 11, 415448.
Misicu, M. (1964), On a theory of asymmetric plastic and visco-plastic solids, Mec.
Appl. 9, 477495.
Mora, R. and Waas, A.M. (2000), Measurement of the Cosserat constant of circular-cell
polycarbonate honeycomb, Philos. Mag. 80, 16991713.
Moran, B. and Gosz, M. (1994), Stress invariance in plane anisotropic elasticity, Modell.
Simul. Mater. Sci. Eng. 2, 677688.
Muki, R. and Sternberg, E. (1965), The influence of couple-stresses on singular stress
concentrations in elastic solids, Z. Angew. Math. Phys. 16, 611648.
Mura, T. (1987), Micromechanics of Defects in Solids, 2nd ed., Martinus Nijhoff,
Dordrecht.
Murdoch, A.I. (2003), Objectivity in classical continuum physics: A rationale for
discarding the principle of invariance under superposed rigid body motions
in favour of purely objective considerations, Cont. Mech. Thermodyn. 15,
309320.
Muschik, W., Ehrentraut, H., and Papenfu, C., Concepts of mesoscopic continuum
physics with application to biaxial liquid crystals, J. Non-Equilib. Thermodyn. 25
(2000), 179197.
Muhlhaus, H., ed. (1995), Continuum Models for Materials with Microstructure, John
Wiley and Sons, New York.
Napier-Munn, T.J., Morrell, S., Morrison, R.D., and Kojovic, T. (1999), Mineral Com-
minution CircuitsTheir Operation and Optimisation, Julius Kruttschnitt Mineral Re-
search Centre, The University of Queensland.
Nemat-Nasser, S. (1999), Averaging theorems in finite deformation plasticity, Mech.
Mat. 31, 493523.
Nemat-Nasser, S. and Hori, M. (1993), Micromechanics: Overall Properties of Heteroge-
neous Solids, North-Holland, Amsterdam.
Nesterenko, V. (2001), Dynamics of Heterogeneous Materials, Springer-Verlag, New York.
Neuber, H. (1966), Uber Probleme der Spannungskonzentration im Cosserat-Korper,
Acta. Mech. 2, 4869.
Neumeister, J.M. (1992), On the role of elastic constants in multiphase contact problem,
J. Appl. Mech. 59, 328334.
P1: Binod
July 5, 2007 16:11 C4174 C4174C012
Noor, A.K. (1988), Continuum modeling for repetitive lattice structures, Appl. Mech.
Rev. 41(7), 285296.
Noor, A.K. and Nemeth, M.P. (1980), Micropolar beam models for lattice grids with
rigid joints, Comp. Meth. Appl. Mech. Eng. 21, 249263.
Nowacki, W. (1966), Couple-stresses in the theory of thermoelasticity. I., Bull. Acad.
Polon. Sci., Ser. Sci. Tech. 14, 97106; II., 14, 203-223; III., 14, 505513.
Nowacki, W. (1970), Theory of Micropolar Elasticity, CISM Courses and Lectures, 25,
Springer, New York.
Nowacki, W.K. (1978), Stress Waves in Non-Elastic Solids, Pergamon Press, Oxford.
Nowacki, W. (1986), Theory of Asymmetric Elasticity, Pergamon Press, Oxford/PWN,
Warsaw.
Nye, J.F. (1957), Physical Properties of Crystals, Oxford University Press, Oxford.
Ogden, R.W. (1972), Large deformation isotropic elasticity: On the correlation of the-
ory and experiment for compressible rubberlike solids, Proc. Roy. Soc. Lond. A 326,
565583.
Okabe, A., Boots, B., and Sugihara, K. (1992), Spatial Tessellations: Concepts and Appli-
cations of Voronoi Diagrams, John Wiley and Sons, New York.
Onat, E.T. and Leckie, F.A. (1988), Representation of mechanical behavior in the pres-
ence of changing internal structure, ASME J. Appl. Mech. 55(1), 110.
Onck, P.R. (2002), Cossorat modeling of cellular solids, C.R. Mec. 330, 717722.
Ostoja-Starzewski, M. (1987), Propagation of Rayleigh, Scholte and Stoneley waves
along random boundaries, Probab. Eng. Mech. 2(2), 6473.
Ostoja-Starzewski, M. (1989), Wavefront propagation in discrete random media via
stochastic Huygens minor principle, J. Franklin Inst. 326(2), 281293.
Ostoja-Starzewski, M. (1990), A generalization of thermodynamic orthogonality to
random media, J. Appl. Math. Phys. (ZAMP) 41, 701712.
Ostoja-Starzewski, M. (1991a), Wavefront propagation in a class of random micro-
structures. I. Bilinear elastic grains, Int. J. Non-Linear Mech. 26, 655669.
Ostoja-Starzewski, M. (1991b), Transient waves in a class of random heterogeneous
media, Appl. Mech. Rev. (Special Issue: Mechanics Pan-America 1991, P.A. Kittl and
D.T. Mook, eds.), 44(10, Part 2), S199S209.
Ostoja-Starzewski, M. (1993a), Micromechanics as a basis of stochastic finite elements
and differences: An overview, Appl. Mech. Rev. 46(11, Pt. 2), S136S147.
Ostoja-Starzewski, M. (1993b), Micromechanically based random fields, in G.I.
Schueller, M. Shinozuka, and J.T.P. Yao, eds., ICOSSAR93 Proceedings, A.A.
Belkema, Rotterdam, 629635.
Ostoja-Starzewski, M. (1993c), On the critical amplitudes of acceleration wave to shock
wave transition in white noise random media, ZAMP 44, 865879.
Ostoja-Starzewski, M. (1994), Micromechanics as a basis of continuum random fields,
Appl. Mech. Rev. 47(1, Pt. 2), S221S230.
Ostoja-Starzewski, M. (1995a), Wavefront propagation in a class of random micro-
structures. II. Nonlinear elastic grains, Int. J. Non-Linear Mech. 30, 771781.
Ostoja-Starzewski, M. (1995b), Transition of acceleration waves into shock waves in
random media, Appl. Mech. Rev. (Special Issue: Nonlinear Waves in Solids), 48(1, Pt. 2),
300308.
Ostoja-Starzewski, M. (1998), Random field models of heterogeneous materials,
Int. J. Solids Struct. 35(19), 24292455.
Ostoja-Starzewski, M. (1999a), Scale effects in materials with random distributions of
needles and cracks, Mech. Mater. 31(12), 883893.
P1: Binod
July 5, 2007 16:11 C4174 C4174C012
Bibliography 451
Ostoja-Starzewski, M. (1999b), Microstructural disorder, mesoscale finite elements,

and macroscopic response, Proc. Roy. Soc. Lond. A 455, 31893199.
Ostoja-Starzewski, M. (2000), Universal material property in conductivity of planar
random microstructures, Phys. Rev. B 62(5), 29802983.
Ostoja-Starzewski, M. (2001), Mechanics of random materials: Stochastics, scale effects,
and computation, in D. Jeulin and M. Ostoja-Starzewski, eds., Mechanics of Random
and Multiscale Microstructures, CISM Courses and Lectures, 430, Springer, New York,
93161.
Ostoja-Starzewski, M. (2001a), Crack patterns in plates with randomly placed hole: A
maximum entropy approach, Mech. Res. Comm. 28, 193198.
Ostoja-Starzewski, M. (2001b), Michell trusses in the presence of microscale material
randomness: Limitation of optimality, Proc. Roy. Soc. Lond. A 457, 17871797.
Ostoja-Starzewski, M. (2001c), On geometric acoustics in random, locally anisotropic
media, Cont. Mech. Thermodyn. 13, 131134.
Ostoja-Starzewski, M. (2002a), Lattice models in micromechanics, Appl. Mech. Rev.
55(1), 3560.
Ostoja-Starzewski, M. (2002b), Microstructural randomness versus representative
volume element in thermomechanics, ASME J. Appl. Mech. 69, 2535.
Ostoja-Starzewski, M. (2002c), Towards stochastic continuum thermodynamics,
J. Non-Equilib. Thermodyn. 27(4), 335348.
Ostoja-Starzewski, M. (2003), Thermoelastic waves in a helix with parabolic or hyper-
bolic heat conduction, J. Thermal Stresses 26, 12051219.
Ostoja-Starzewski, M. (2004), Fracture of brittle micro-beams, J. Appl. Mech., 71,
424427.
Ostoja-Starzewski, M. (2005a), On the admissibility of an isotropic, smooth elastic
continuum, Arch. Mech. 57(4), 345355. Errata, ibid. (2006).
Ostoja-Starzewski, M. (2005b), Scale effects in plasticity of random media: Status and
challenges, Int. J. Plast. 21, 11191160.
Ostoja-Starzewski, M. (2006), Material spatial randomness: From statistical to repre-
sentative volume element, Probab. Eng. Mech. 21(2), 112132.
Ostoja-Starzewski, M. and Castro, J. (2003), Random formation, inelastic response,
and scale effects in paper, Philos. Trans. Roy. Soc. Lond. A 361(1806), 965986.
Ostoja-Starzewski, M. and Ilies, H. (1996), The Cauchy and characteristic boundary
value problems for weakly random rigid-perfectly plastic media, Int. J. Solids Struct.
33(8), 11191136.
Ostoja-Starzewski, M. and Jasiuk, I. (1995), Stress invariance in planar Cosserat elas-
ticity, Proc. Roy. Soc. Lond. A 451, 453470; Errata 452, 1503.
Ostoja-Starzewski, M. and Lee, J.D. (1996), Damage maps of disordered composites:
A spring network approach, Int. J. Fract. 75, R51R57.
Ostoja-Starzewski, M. and Schulte, J. (1996), Bounding of effective thermal conduc-
tivities of multiscale materials by essential and natural boundary conditions, Phys.
Rev. B 54, 278285.
Ostoja-Strazewski, M. and Shahsavari, H. (2007), Response of a helix made of a frac-
tional viscoelastic material, ASME J. Appl. Mech., in press.
Ostoja-Starzewski, M., Sheng, P.Y. and Jasiuk, I. (1997), Damage patterns and consti-
tutive response of random matrix-inclusion composites, Eng. Fract. Mech. 58(5&6),
581606.
Ostoja-Starzewski, M. and Stahl, D.C. (2001), Random fiber networks and special
elastic orthotropy of paper, J. Elast. 60(2), 131149.
P1: Binod
July 5, 2007 16:11 C4174 C4174C012

Ostoja-Starzewski, M. and Trebicki, J. (1999), On the growth and decay of acceleration
waves in random media, Proc. Roy. Soc. Lond. A 455, 25772614.

Ostoja-Starzewski, M. and Trebicki, J. (2003), On the distance to blow-up of acceleration
waves in random media, media, Cont. Mech. Thermodyn. 15, 2132.

Ostoja-Starzewski, M. and Trebicki, J. (2006), Stochastic dynamics of acceleration wave-
fronts in random media, Mech. Mater. 38, 840848.
Ostoja-Starzewski, M. and Wang, C. (1989), Linear elasticity of planar Delaunay
networks: Random field characterization of effective moduli, Acta Mech. 80, 6180.
Ostoja-Starzewski, M. and Wang, C. (1990), Linear elasticity of planar Delaunay
networks. II. Voigt and Reuss bounds, and modification for centroids, Acta Mech.
84, 4761.
Ostoja-Starzewski, M. and Wang, G. (2006), Particle modeling of random crack patterns
in epoxy plates, Probab. Eng. Mech. 21(3), 267275.
Ostoja-Starzewski, M. and Wang, X. (1999), Stochastic finite elements as a bridge
between random material microstructure and global response, Comp. Meth. Appl.
Mech. Eng. 168(14), 3549.
Ostoja-Starzewski, M. and Woods, A.N. (2003), Spectral finite elements for vibrating
rods and beams with random field properties, J. Sound Vib. 268, 779797.
Ostoja-Starzewski, M., Alzebdeh, K., and Jasiuk, I. (1995), Linear elasticity of planar
Delaunay networks. III. Self-consistent approximations, Acta Mech. 110,
5772.
Ostoja-Starzewski, M., Jasiuk, I., Wang, W., and Alzebdeh, K. (1996a), Composites with
functionally graded interfaces: Meso-continuum concept and effective properties,
Acta Mater. 44(5), 20572066.
Ostoja-Starzewski, M., Sheng, P.Y., and Alzebdeh, K. (1996b), Spring network models
in elasticity and fracture of composites and polycrystals, Comput. Mater. Sci., 7
(1 & 2), 8293.
Ostoja-Starzewski, M., Quadrelli, M.B., and Stahl, D.C. (1999), Kinematics, stress trans-
fer mechanisms, and effective properties of quasi-planar random fiber networks, C.
R. Acad. Sci. Paris IIb 327, 12231229.
Ostoja-Starzewski, M., Du, X., Khisaeva, Z.F., and Li, W. (2006), On the size of RVE in
elastic, plastic, thermoelastic, and porous random microstructures, Science, THER-
MEC2006, Vancouver, Canada, (2007) Materials Science Forum, 538543, 201206.
Ostoja-Starzewski, M., Du, X., Khisaeva, Z.F. and Li, W. (2007), On the size of RVE in
elastic, plastic, thermoelastic, and porous random microstructures, Int. J. Multiscale
Comp. Eng., to appear.
Page D.H., Tydeman P.A., and Hunt M. (1961), A study of fibre-to-fibre bonding by
direct observation, The Formation and Structure of PaperTrans. Oxford Symposium 1,
171193.
Palasti, I. (1960), On some random space-filling problems, Publ. Math. Inst. Hung. Acad.
Sci. 5, 355360.
Pamin, J. (2005), Gradient plasticity and damage models: A short comparison Comp.
Mater. Sci. 32(3-4), 472479.
Papanicolaou, G.C. (1998), Mathematical problems in geophysical wave propagation,
Doc. Math. (Extra Volume ICM), 125.
Papenfu, C., and Muschik, W., (1998), Liquid crystal theory as an example of meso-
scopic continuum mechanics, in Trends in Continuum Physics, World Scientific,
Singapore, 277291.
Papoulis, A. (1984), Probability, Random Variables, and Stochastic Processes, McGraw-Hill,
New York.
P1: Binod
July 5, 2007 16:11 C4174 C4174C012
Bibliography 453
Perez, D.G., Zunino, L., and Garavaglia, M. (2004), A fractional Brownian motion
model for the turbulent refractive index in lightwave propagation, Optics Comm.
242, 5763.
Perkins, R.W. and Thomson, D. (1973), Experimental evidence of a couple-stress effect,
AIAA J. 11, 10531055.
Perzyna, P. (1966), Fundamental problems in viscoplasticity, Adv. Appl. Mech. 9,
243377.
Pham, D.C. (2006), Revised bounds on the elastic moduli of two-dimensional random
crystals, J. Elast. 85, 120.
Poincare, H. (1912), Calcul des Probabilites, Gauthier-Villars, Paris.
Ponte Castaneda, P. (2001), Nonlinearity and microstructure evolution in composite
materials, Proc. ICTAM 2000.
Ponte Castaneda, P. and Suquet, P.M. (1998), Nonlinear composites, Adv. Appl. Mech.
34, 171302.
Porcu, E., Mateau, J., Zini, A., and Pini, R. (2007), Modelling spatio-temporal data: A
new variogram and covariance structure proposal, Stat. Probab. Lett, 77, 8389.
Prager, S. (1961), Viscous flow through porous media, Phys. Fluids 4, 14771482.
Press, W.H., Teukolsky, S.A., Vetterling, W.T. and Flannery, B.P. (1992), Numerical
Recipes, Cambridge University Press, Cambridge.
Prohorov, Yu.V. and Rozanov, Yu.A. (1969), Probability Theory, Springer-Verlag, Berlin.
Pshenichnov, G.I. (1993), A Theory of Latticed Plates and Shells, World Scientific,
Singapore.
Raisanen V.I., Alava M.J., and Nieminen R.M. (1997), Failure of planar fiber networks,
J. Appl. Phys. 82(8), 37473753.
Renyi, A. (1958), A one-dimensional problem of random space filling, Comm. Math.
Res. Inst. Hung. Acad. Sci. 3, 109127.
Renyi, A. (1970), Foundations of Probability, Holden-Day, San Francisco, CA.
Robertson, H.P. (1940), The invariant theory of isotropic turbulence, Proc. Camb. Philos.
Soc. 36(2), 209233.
Rosen, B.W. and Hashin, Z. (1970), Effective thermal expansion coefficients and specific
heats of composite materials, Int. J. Eng. Sci. 8, 157173.
Rozvany, G.I.N., Bendsoe, M., and Kirsch, U. (1995), Layout optimization of structures,
Appl. Mech. Rev. 41, 48119.
Rudin, W. (1974), Real and Complex Analysis, McGraw-Hill, New York.
Rytov, S.M., Kravtsov, Yu.A., and Tatarskii, V.I. (1987), Principles of Statistical Radio-
physics, 4, Springer-Verlag, Berlin.
Sab, K. (1991), Principe de Hill et homogeneisation des materiaux aleatoires, C.R. Acad.
Sci. Paris II 312, 15.
Sab, K. (1992), On the homogenization and the simulation of random materials, Europ.
J. Mech., A Solids 11, 585607.
Sab, K. and Nedjar, B. (2005), Periodization of random media and representative vol-
ume element size for linear composites, C.R. Acad. Sci. Paris II 333, 187195.
Sachs, G. (1928), Zur Ableitung einer Fliessbedingung, Z. Ver. Deutsch. Ing. 72,
734736.
Saenger, E.H., Kruger, O.S., and Shapiro, S.A. (2006), Effective elastic properties of
fractured rocks: Dynamic vs. static properties, Int. J. Fract. (Lett.) 139, 569576.
Sahimi, M. (2003), Heterogeneous Materials. I. Linear Transport and Optical Properties,
Saigal, S. (1995), Stochastic boundary elements for two-dimensional potential flow in
non-homogeneous media, Comp. Meth. Appl. Mech. Eng. 121, 211230.
P1: Binod
July 5, 2007 16:11 C4174 C4174C012
Salvadori, M,G., Skalak, R., and Weidlinger, P. (1960), Waves and shocks in locking
and dissipative media, ASCE J. Eng. Mech. 86(EM2, Pt. 1), 77105.
Sampson, W.W. (2001), Comments on the pore radius distribution in near-planar
stochastic fibre networks, J. Mat. Sci. 36 (21), 51315135.
Samras, R.K., Skop, R.A., and Millburn, D.A. (1974), An analysis of coupled
extensional-torsional oscillations in wire rope, ASME J. Eng. Ind. 96, 11301135.
Sanchez-Palencia, E. and Zaoui, A. (1987), Homogenization Techniques for Composite
Media, Springer-Verlag, Berlin.
Sandru, N. (1966), Experimental evidence of a couple-stress effect, Int. J. Eng. Sci.
1, 81.
Santalo, L.A. (1976), Integral geometry and geometric probability, in G.-C. Rota (ed.),
Encyclopedia of Mathematics and its Applications, 1, Addison-Wesley, Reading, MA.
Satake, M. (1976), Constitution of mechanics of granular materials through graph
representation, Proc. 26th Japan Natl. Congr. Theor. Appl. Mech., University of Tokyo
Press, Tokyyo, 257266.
Satake, M., (1978), Constitution of mechanics of granular materials through the graph
theory, in S.C. Cowin and M. Satake, eds., Continuum Mechanical and Statistical
Approaches in the Mechanics of Granular Materials, Gabujutsa Bunfer Fukyukai, Tokyo,
4762.
Save, M. and Prager, W. 1985 Structural Optimization, 1, Plenum Press, New York.
Sawczuk, A. (1967), On yielding of Cosserat continua, Arch. Mech. 19, 471480.
Schafer, H. (1962), Versuch einer Elastizitatstheorie des zweidimensionalen ebenen
Cosserat-Kontinuums. Misz. Angew. Math., pp. 277292.
Schafer, H. (1967), Die Spannungsfunktionen eines Kontinuum mit Moment-
enspannungen. I. Bull. Acad. Polon. Sci., Ser. Sci. Tech. 15, 63; II. 15, 485.
Schijve, J. (1966), Note on couple stresses, J. Mech. Phys. Solids 14, 113120.
Schuss, Z. (1980), Theory and Application of Stochastic Differential Equations, John Wiley
and Sons, New York.
Scott, D.G. (1962), Packing of spheres, Nature 194, 956957.
Sewell, M.J. (1987), Maximum and Minimum Principles: a Unified Approach, with Appli-
cations, Cambridge University Press.
Shahinpoor, M. (1978), Governing equations for simple continuum feathers, Arch.
Mech. 30, 113117.
Shahsavari, H. and Ostoja-Starzewski, M. (2005a), Spectral finite element of a helix,
Mech. Res. Comm. 32, 147152.
Shahsavari, H. and Ostoja-Starzewski, M. (2005b), On elastic and viscoelastic helices,
Philos. Mag. 85(3335), 42134230.
Shechao Feng, Thorpe, M.F., and Garboczi, E. (1985), Effective-medium theory of per-
colation on central-force elastic networks, Phys. Rev. B 31, 276280.
Shen, H.H., Satake, M., Mehrabadi, M., Chang, C.S., and Campbell, C.S., eds. (1992),
Advances in Micromechanics of Granular Materials, Studies in Appl. Mech. 31, Elsevier,
Amsterdam.
Shinozuka, M. and Yamazaki, F. (1988), Stochastic finite element analysis: An
introduction, in S.T. Ariaratnam, G.I. Schueller, and I. Elishakoff, eds., Stochastic
Structural Dynamics: Progress in Theory and Applications, Elsevier, London, 241291.
Slepyan, L.I., Krylov, V.I., and Parnes, R. (2000), Helical inclusion in an elastic matrix,
Snyder, K.A., Garboczi E.J., and Day, A.R. (1992), The elastic moduli of simple
two-dimensional composites: Computer simulation and effective medium theory,
J. Appl. Phys. 72, 59485955.
P1: Binod
July 5, 2007 16:11 C4174 C4174C012
Bibliography 455
Sobczyk, K. (1985), Stochastic Wave Propagation, Elsevier, Amsterdam.

Sobczyk, K. (1986), Stochastic waves: The existing results and new problems, Probab.
Eng. Mech. 1(3), 167176.
Sobczyk, K. and Kirkner, D. (2001), Stochastic Modeling of Microstructures, Birkhauser
Verlag, Berlin.

Sobczyk. K. and Trebicki, J. (1990), Maximum entropy principle in stochastic dynamics,
Probab. Eng. Mech. 5(3), 102110.

Sobczyk, K. and Trebicki, J. (1993), Maximum entropy principle and nonlinear stochas-
tic oscillators, Physica A 193, 448468.
Soize, C. (2004), Random-field model for the elasticity tensor of anisotropic random
media. Comp. Rend. Mec. Acad. Sci. Paris 332, 10071012.
Somigliana, C. (1910), Sopra unestensione della teoria dellelasticita, Rend. Acc. Lincei,
Cl. Sci. Fis. Mat. Nat., 19, 1o sem.
Soszynski, R. (1995), Relative bonded area: A different approach, Nordic Pulp Paper J.
2, 150.
Stahl, D.C. and Cramer, S.M. (1998), A three-dimensional network model for a low
density fibrous composite, ASME J. Eng. Mat. Tech. 120(2), 126130.
Stephen, N.G. and Wang, P.J. (1996), On Saint-Venants principle in pinjointed frame-
works, Int. J. Solids Struct. 33(1), 7997.
Sternberg, E. and Muki, R. (1967), The effect of couple-stresses on the stress concen-
tration around a crack, Int. J. Solids Struct. 3, 6995.
Stojanovic, R. (1970), Recent Developments in the Theory of Polar Continua, Springer,
Wien, New York.
Stoker, J.J. (1950), Nonlinear Vibrations in Mechanical and Electrical Systems, Interscience,
New York.
Stoyan, D., Kendall, W.S., and Mecke, J. (1995), Stochastic Geometry and its Applications,
John Wiley and Sons, New York.
Stroeven, M., Askes, H., and Sluys, L.J. (2004), Numerical determination of rep-
resentative volumes for granular materials, Comp. Meth. Appl. Mech. Eng. 193,
32213238.
Sudria, J. (1935), Laction euclidieenne de deformation et de mouvement, Mem. Sci.
Phys. No. 29, 56.
Suquet, P.M. (1986), Elements of homogenization for solid mechanics, in E. Sanchez-
Palencia and A. Zaoui, eds., Homogenization Techniques for Composite Media, Lecture
Notes in Physics, 272, Springer-Verlag, Berlin, 193278.
Suquet, P.M. (1993), Overall potentials and extremal surfaces of power law or ideally
plastic composites, J. Mech. Phys. Solids 41(6), 9811002.
Suquet, P.M. (1997), Effective properties of nonlinear composites, in P.M. Suquet,
ed., Continuum Micromechanics, CISM Courses and Lectures, 377, Springer-Verlag,
Berlin, 197264.
Szczepankiewicz, E. (1985), Applications of Random Fields (in Polish), PWN-Polish
Science Publishers, Warsaw.
Szczepinski, W. (1979), Introduction to the Mechanics of Plastic Forming of Metals, Sijthoff
and Noordhoff Int. Publ./PWN, Alphen an der Rijn/Warsaw.
Takeuti, Y. (1973), On plane micropolar thermoelasticity in multiply-connected do-
mains and its applications, Arch Mech., 25, 975984.
Talbot, D.R.S. (2000), Improved bounds for the effective properties of a nonlinear two-
phase elastic composite, J. Mech. Phys. Solids 48, 12851294.
Tarasov, V.E. (2005a), Continuous medium model for fractal media, Phys. Lett. A 336,
167174.
P1: Binod
July 5, 2007 16:11 C4174 C4174C012
Tarasov, V.E. (2005b), Fractional hydrodynamic equations for fractal media, Ann. Phys.
318(2), 286307.
Tauchert, T.R., Claus, W.D., and Ariman, T. (1968), The linear theory of micropolar
thermoelasticity, Int. J. Eng. Sci. 6, 3747.
Taylor, G. I. (1938), Plastic strain in metals, J. Inst. Met. 62, 307324.
Terada, K., Hori, M., Kyoya, T., and Kikuchi, N. (2000), Simulation of the multi-
convergence in computational homogenization approaches, Int. J. Solids Struct. 37,
22852311.
Thorpe, M.F. and Jasiuk, I. (1992), New results in the theory of elasticity for two-
dimensional composites. Proc. Roy. Soc. Lond. A 438, 531544.
Tiersten, H.F. and Bleustein, J.L. (1974), Generalized elastic continua, in G. Herrmann,
ed., R.D. Mindlin and Applied Mechanics, Pergamon Press, New York; 67103.
Torquato, S. (1991), Random heterogeneous media: Microstructure and improved
bounds on effective properties, Appl. Mech. Rev. 44, 3776.
Torquato, S. (1997), Effective stiffness tensor of composite media, J. Mech. Phys. Solids
45, 14211448.
Torquato, S. (2002), Random Heterogeneous Materials: Microstructure and Macroscopic
Properties, Springer-Verlag, New York.
Toupin, R.A. (1962), Elastic materials with couple-stresses, Arch. Rational Mech. Anal.
11, 385414.
Toupin, R.A. (1964), Theories of elasticity with couple-stresses, Arch. Rational Mech.
Anal. 17, 85112.
Triantafyllidis, N. and Bardenhagen, S. (1993), On higher order gradient continuum
theories in 1D nonlinear elasticity. Derivation from and comparison to the corre-
sponding discrete models, J. Elast. 33, 259293.
Trovalusci, P. and Masiani, R. (1999), Material symmetries of micropolar continua
equivalent to lattices, Int. J. Solids Struct. 36, 20912108.
Trovalusci, P. and Masiani, R. (2003), Non-linear micropolar and classical continua for
anisotropic discontinuous materials, Int. J. Solids Struct. 40, 12811297.
Trovalusci, P. and Masiani, R. (2005), A multifield model for blocky materials based
on multiscale description, Int. J. Solids Struct. 42, 57785794.
Truesdell, C. and Toupin, R.A. (1960), Classical field theories, in Encyclopedia of Physics,
S. Flugge, ed., III/1, Springer-Verlag, Berlin, 236793.
Truesdell, C. and Noll, W. (1965), The Nonlinear Field Theories of Mechanics, in Hand-
buch der Physik III/3, Springer-Verlag, Berlin.
Tschoegl, N.W., Knauss, W.G., and Emri, I. (2002), Poissons ratio in linear viscoelas-
ticity: A critical review, Mech. Time-Dependent Mat. 6, 351.
Uesaka, T., Murakami, K. and Imamura, R. (1979), Biaxial tensile behavior of paper,
Tappi J. 62 (8), 110114.
Uscinski, B.J. (1977), The Elements of Wave Propagation in Random Media, McGraw-Hill,
New York.
Vakulenko, A.A. and Kachanov, M.L. (1971), Continuum model of medium with cracks
(in Russian), Mech. Solids 4, 5459.
van der Sluis, O., Schreurs, P.J.G., Brekelmans, W.A.M., and Meijer, H.E.H. (2000),
Overall behaviour of heterogeneous elastoviscoplastic materials: Effect of mi-
crostructural modelling, Mech. Mater. 32, 449442.
Vanmarcke, E, (1983), Random Fields: Analysis and Synthesis, MIT Press, Cambridge,
MA.
Vannucci, P. (2002), A special planar orthotropic material, J. Elast. 67, 8196.
P1: Binod
July 5, 2007 16:11 C4174 C4174C012
Bibliography 457
Vannucci, P. and Verchery, G. (2001), Stiffness design of laminates using the polar
method, Int. J. Solids Struct. 38, 92819294.
Vinogradov, O. (2001), On a representative volume in the micromechanics of partic-
ulate composites, Mechanics of Composite Materials 37, No. 3 [English translation of
Mekhanika Kompozitnykh Materialov, Latvian Academy of Sciences, Riga].
Vinogradov, O. (2006), A static analog of molecular dynamics method for crystals, Int.
J. Comp. Meth. 3(2), 153161.
Voigt, W. (1887), Theoretische Studien uber die Elastizitatsverhaltnisse der Kristalle,
Abh. Gess. Wiss. Gottingen 34, 351.
von Mises, R. (1931), Wahrscheinlichkeitsrechnung, Deuticke, Leipzig.
Wang, C.Y. and Jasiuk, I. (2007), Scale and boundary condition effects on elastic moduli
of trabecular bone, submitted.
Wang, G. and Ostoja-Starzewski, M. (2005), Particle modeling of dynamic fragmenta-
tion. I. Theoretical considerations, Comp. Mat. Sci. 33(4), 429442.
Wang, G. and Ostoja-Starzewski, M. (2006), Particle modeling of dynamic fragmenta-
tion. II. Fracture in single- and multi-phase materials, Comp. Mat. Sci. 35(2), 116133.
Wang, X.L. and Stronge, W.J. (1999), Micropolar theory for two-dimensional stresses
in elastic honeycomb, Proc. Roy. Soc. Lond. A 455, 20912116.
Wang, Z.-P. (1997), Void-containing nonlinear mateials subject to high-rate loading,
J. Appl. Phys. 81(11), 72137227.
Wang, Z.-P. and Sun, C.T. (2002), Modeling micro-inertia in heterogeneous mateials
under dynamic loading, Wave Motion 36, 473485.
Warren, W.E. & Byskov, E. (2002), Three-fold symmetry restrictions on two-
dimensional micropolar materials, Europ. J. Mech. A/Solids 21, 779792.
Weitsman, Y. (1964), Couple-stress effects on stress concentration around a cylindrical
inclusion in a field of uniaxial tension, J. Appl. Mech. 32, 424428.
Werner, E., Siegmund, T., Weinhandl, H., and Fischer, F.D. (1994), Properties of random
polycrystalline two-phase materials. Appl. Mech. Rev. 47(1, Pt. 2), S231S240.
Willam, K., Dietsche, A., Iordache, M.-M. and Steinmann, P. (1995), Localization in
micropolar continua, in Continuum Models for Materials with Microstructure, ed. H.
Muhlhaus, J. Wiley, New York, Ch. 1, 297339.
Willis, J.R. (1981), Variational and related methods for the overall properties of com-
posites, Adv. Appl. Mech. 21, 178.
Willis, J.R. (1997), Dynamics of composites, in P. Suquet, ed., Continuum Micro-
mechanics, CISM Courses and Lectures, 377, Springer, NewYork, 264347.
Willis, J.R. (2001), Lectures on mechanics of random media, in D. Jeulin and M. Ostoja-
Starzewski, eds., Mechanics of Random and Multiscale Microstructures, CISM Courses
and Lectures, 430, Springer, NewYork, 221267.
Willis, J.R. and Talbot (1990), in G.A. Maugin, ed., Continuum Models and Discrete
Systems, John Wiley & Sons, New York, 113131.
Wilmanski, K. (1998), Thermomechanics of Continua, Springer-Verlag, Berlin.
Wlodarczyk, E. (1972), Propagation of a plane shock wave in a bilinear elastic bar,
Proc. Vibr. Probl. 13, 319329.
Wozniak, C. (1966), Load carrying structures of dense lattice type, Arch. Mech. Stos.
18(5), 581597.
Wozniak, C. (1970), Surface Lattice Structures (in Polish), PWN-Polish Science Publisher,
Warsaw.
Wozniak, C. (1997), Internal variables in dynamics of composite solids with periodic
microstructure, Arch. Mech. 49(2), 421441.
P1: Binod
July 5, 2007 16:11 C4174 C4174C012
Yaglom, A.M. (1957), Some classes of random fields in n-dimensional space, related
to stationary random processes, Theory Probab. Appl. 11, 273320.
Yang, J.F.C. and Lakes, R.S. (1982), Experimental study of micropolar and couple stress
elasticity in compact bone in bending, J. Biomech. 15, 9198.
Yavari, A., Sarkani, S., and Moyer, E.T. Jr. (2002), On fractal cracks in micropolar elastic
solids, ASME J. Appl. Mech. 69, 4554.
Yoo, A. and Jasiuk, I. (2006), A gradient-dependent flow theory of plasticity: Applica-
tion to metal and soil instabilities. J. Biomech. 39, 22412252.
Yserentant, H. (1997), A new class of particle methods, Numer. Math. 76, 87109.
Zbib, H. and Aifantis, E.C. (1989), A gradient-dependent flow theory of plasticity:
Application to metal and soil instabilities. Appl. Mech. Rev. 42(11) Part 2,
S292S304.
Zhang, P., Huang, Y., Gao, H., and Hwang, K.C. (2002), Fracture nucleation in single-
wall carbon nanotubes under tension: A continuum analysis incorporating inter-
atomic potentials, ASME J. Appl. Mech. 69, 454458.
Zhao, J., Sheng, D., and Collins, I.F. (2006), Thermomechanical formulation of strain
gradient plasticity for geomaterials, J. Mech. Mater. Struct. 1(5), 837863.
Ziegler, F. (1977), Wave propagation in periodic and disordered layered composite
elastic materials, Int. J. Solids Struct. 13, 293305.
Ziegler, H. (1957), Thermodynamik und rheologische Probleme, Ing. Arch. 25,
5870.
Ziegler, H. (1962), Die statistischen Grundlagen der irreversiblen Thermodynamik,
Ing. Arch. 31, 317342.
Ziegler, H. (1963), Some extremum principles in irreversible thermodynamics with
applications to continuum mechanics, in I.N. Sneddon and R. Hill, eds., Progress in
Solid Mechanics, 4, North-Holland, Amsterdam, 91191.
Ziegler, H. (1970), Proof of an orthogonality principle, ZAMP 21, 853863.
Ziegler, H. (1972), Thermomechanics, Q. Appl. Math. 28, 91107.
Ziegler, H. (1983), An Introduction to Thermomechanics, North-Holland, Amsterdam.
Ziegler, H. and Wehrli, Ch. (1987), The derivation of constitutive relations from free
energy and the dissipation functions, Adv. Appl. Mech. 25, 183238.
Ziman, J.M. (1979), Models of Disorder, Cambridge University Press, Cambridge.
P1: Naresh Chandra
July 5, 2007 18:26 C4174 C4174C013
Author Index
A Bernal, J.D., 23
Bird, M.D., 102
Achenbach, J.D., 208 Blanc, X., 163
Adhikari, S., 402 Bland, D.R., 423
Adler, R.J., 79 Blenk, S., 276
Advani, S.G., 44 Bleustein, J.L., 206
Aero, E.L., 205 Blouin, F., 125
Aifantis, E.C., 207 Boal, D.H., 135
Al-Ostaz, A., 7980, 154, 163 Boccara, S., 184, 187
Alava M.J., 141 Bogy, D.B., 209
Allaire, G., 169 Boots, B., 33
Altus, E., 62 Bouyge, F., 230
Alzebdeh, K., 105, 154 Brandstatter, J.J., 393
Amieur, M., 247, 258 Brekhovskikh, L.M., 56
Ariman, T., 207, 215 Brenner, C.E., 290
Arnold, V.I., 390 Brezzi, F., 287
Asch, M., 408 Bronkhorst, C.A., 318
Asimow, L., 134, 139 Budiansky, B., 265
Askar, A., 87, 208 Buxton, G.A., 99
Askes, H., 248 Byskov, E., 114
Atkinson, C., 209
C
B
Campbell, J.G., 151
Babuska, I., 120 Caputo, M., 129
Bagley, R.L., 129 Cardou, A., 125
Bahei-Ei-Din, Y.A., 326 Carmeliet, J., 356
Ball, J.M., 205, 383 Carvalho, F.C.S., 265
Banks, C.B., 208 Castro, J., 30, 314
Bardenhagen, S., 91, 120, 208 Chaboche, J.-L., 241, 354355, 359
Bathurst, R.J., 143 Chen, P.J., 163, 386, 422423
Bazant, M.Z., 300 Cherkaev, A.V., 178, 181, 219, 223
Bazant, Z.P., 208, 356 Chernov, L.A., 385, 387, 392
Beltzer, A.I., 402 Chien, S., 135, 145
Belyaev, A.K., 402 Christensen, M., 208
Belytschko, T., 163 Christensen, R.M., 347
Benaroya, H., 290 Chung, J.W., 139
Bendsoe, M., 165, 168169 Cielecka, I., 116
Benedict, R., 205 Cioranescu, D., 120
Bennison, S.J., 143 Claus, W.D., 207
Bensoussan, A., 229 Cohen, L., 133
Beran, M.J., 62, 73, 209, 395 Coleman, B.D., 422, 426
Berger, M.A., 404 Collins, I.F., 367
Berglund, K., 208 Contreras, H., 291
459
P1: Naresh Chandra
July 5, 2007 18:26 C4174 C4174C013
Cook, R.D., 140 Friesecke, G., 208

Cosserat, E., 205 Frisch, U., 83, 171, 385, 389, 393
Cosserat, F., 205
Costanzo, F., 318
G
Costello, G.A., 122123
Cowin, S.C., 187, 207, 209, Gani, J., 23
215216 Garboczi, E., 99, 104, 266
Cox, H.L., 137 Gauthier, R.D., 208
Cramer, S.M., 139 Gdoutos, E.E., 59
Crapo. H., 134 Germain, P., 370
Cundall, P.A., 141 Ghanem, R.G., 292
Gibbs, J.W., 359, 367
Gibiansky, L., 314
D
Givli, S., 62
Day, A.R., 99, 184 Gloria, A., 293, 321
de Borst, R., 356 Goddard, J.D., 31, 83, 145, 206
De Cicco, S., 188189 Gosz, M., 184
Dence, D., 385 Grah, M., 105
Deng, M., 37 Graham, L., 292
Deodatis, G., 292 Gray, G.L., 318
Dhaliwal, R.S., 207 Green, A.E., 206207
Ditlevsen, O., 290 Greenspan, D., 157, 159
Dodson, C.T.J., 30, 37 Griffith, A.A., 58
Dowling, N.E., 325 Grioli, G., 198, 205
Doyle, J.F., 395, 400, 402 Guarracino, F., 188189
Drzal, L.T., 263 Guest, S.D., 135
Du, X., 348 Gupta, S., 402
Dullien, F.A.L., 342 Gurtin, M.E., 422, 426
Dundurs, J., 179, 182, 184, Gusev, A.A., 107, 248
211, 221
Dvorak, G.J., 326 H
Dyszlewicz, J., 208
Hansen, J.C., 135, 145
Hanson, G.W., 207
E Hartranft, R.J., 207
Hashin, Z., 240, 244, 347
Eastwood, J.W., 157, 161
Hazanov, S., 247, 258, 314
Ehrentraut, H., 276, 359
He, Q.-C., 181, 321
Eimer, C., 73, 85
He, R., 324
Elishakoff, I., 290
Hegemier, G.A., 166, 303
Engelbrecht, J., 423
Hehl, F.W., 96
Englman, R., 83
Herrmann, G., 208
Ericksen, J.L., 205
Herrmann, H.J., 164, 356
Eringen, A.C., 206208
Hersch, R., 403404
Evans, D.J., 364
Hill, R., 237238, 245, 247, 252,
293, 295, 318, 322, 334
F Hilton, H.H., 128
Hockney, R.W., 157, 161
Feng, S., 135 Holnicki-Szulc, J., 118, 120, 208
Fishman, L., 73 Hori, M., 246, 248
Fleck, N.A., 207 Horio, M., 151
Foias, C., 408 Houlsby, G.T., 367
Forest, S., 41, 228230, 248, 379 Howe, M.S., 402
Fortin, M., 287 Hrennikoff, A., 87
Frazho, A.E., 408 Huang, Y., 164
P1: Naresh Chandra
July 5, 2007 18:26 C4174 C4174C013
Author Index 461
Hudson, J.A., 390 Kuvshinskii, E.V., 205

Huet, C., 240, 245, 247, 255, 341 Kuznetsov, E.N., 141
Hutchinson, J.W., 207
Huyse, L., 150
L
Labuz, J.F., 265
I
Lakes, R., 205, 208, 216
Ilies, H., 293 Laplace, P.S., 2
Itin, Y., 96 Le Quang, H., 321
Itou, S., 207 LeBris, C., 163
Leckie, F.A., 44
Lee, J.D., 156
J Lee, S.J., 319
Lemaitre, J., 241, 354355, 359
Jagota, A., 143 Leppington, F.G., 209
Jahsman, W.E., 208 Li, W., 324
James, R.D., 205, 208, 383 Liszka, T., 307
Jasiuk, I., 7980, 112, 114, 141, Lohnert, S., 318
163, 175, 183184, 187, Lomakin, V.A., 6970
211, 227, 230, 269 Love, A.E.H., 96
Jaunzemis, W., 206 Lu, Y.Y., 163
Jaynes, E.T., 14, 44 Lubarda, V.A., 209
Jeulin, D., 38, 331, 396 Luding, S., 164
Jiang, M., 263, 281, 313, 324 Lurie, K.A., 178, 181, 219, 223
Jones, R.M., 176
Jun, S., 184
M
K Maes, M.A., 150

Malkus, M.E., 140
Kac, M., 1 Mandel, J., 245, 247
Kachanov, L.M., 265, 293, 297, 334 Mandelbrot, B.B., 45
Kachanov, M., 265, 267 Manohar, C.S., 402
Kaloni, P.N., 207, 215 Mardia, K.V., 24
Kaminski, M., 290, 402 Mariano, P.M., 206
Kamrin, K., 300 Markenscoff, X., 182184, 187, 209, 211
Kanit, T., 41, 248 Markov, K., 244
Keating, P.N., 98 Martinsson, P.G., 120
Keller, J.B., 104, 393 Masiani, R., 107, 145
Kendall, M.G., 14 Masson, J., 23
Kennett, B.L.N., 408 Mateau, J., 66
Kharanen, V.Y., 392 Matheron, G., 41, 273
Khisaeva, Z.F., 319, 353 Maugin, G.A., 119, 207, 359, 370, 385
Kikuchi, N., 168 Maxwell, J.C., 87
Kirkner, D., 6970 McCarthy, M.F., 423
Kirkwood, J.G., 97 McCoy, J.J., 73, 209
Knauss, W.G., 128 Mellor, P.B., 341
Kohler, W., 408 Mendelson, K.S., 104
Kohn, R.V., 169 Menon, V.V., 423
Koiter, W., 198, 206 Michell, A.G.M., 165
Kotulski, Z., 407 Michell, J.M., 183
Kravtsov, Y.A., 385, 387 Miles, R.E., 27, 30
Kreher, W., 83 Milton, G.W., 104, 178
Kroner, E., 206, 245 Mindlin, R.D., 182, 198, 205207, 212, 214
Krylov, V.I., 125 Misicu, M., 207
Ku, A.P.-D., 300 Mora, R., 115
P1: Naresh Chandra
July 5, 2007 18:26 C4174 C4174C013
Moran, B., 184 Press, W.H., 103

Morrell, S., 162 Pretczynski, Z., 407
Muki, R., 208209 Prohorov, Y.V., 7, 45
Muller, G.M., 207 Pshenichnov, G.I., 120
Mura, T., 185 Puzrin, A.M., 367
Murdoch, A.I., 14
Muschik, W., 276, 359 Q
N Quoc Son, N., 370
Naghdi, P.M., 207 R

Napier-Munn, T.J., 162
Nemat-Nasser, S., 246, 248, 318 Raisanen V.I., 141
Nemeth, M.P., 90 Rehak, M., 290
Nesterenko, V., 418 Rivier, N., 83
Neuber, H., 207 Rivlin, R.S., 206
Neumeister, J.M., 179 Robertson, H.P., 6970
Noll, W., 286 Rogula, D., 118, 120, 208
Noor, A.K., 9091 Roos, A., 139
Nordgren R.P., 300 Rosen, B.W., 347
Nowacki, W., 194, 196, 205208, 210, 212, Roth, B., 134, 139
218, 220, 418 Rothenburg, L., 143
Nunziato, J.W., 187 Roux, S., 356
Nye, J.F., 390 Rozanov, Y.A., 7, 45
Rozvany, G.I.N., 165, 169
O Rudin, W., 3, 5960
Rytov, S.M., 385, 387
OConnell, R.J., 265
Okabe, A., 33 S
Onat, E.T., 44
Onck, P.R., 230 Sab, K., 228229
Onogi, S., 151 Sab, V., 245, 248, 255256
Ostoja-Starzewski, M., 30, 51, 57, 123, 126, Sachs, G., 324
128, 130, 139, 141, 147148, 153154, Sahimi, M., 137
156157, 161, 163, 211, 227228, 255, Saigal, S., 291
259, 263, 265, 275, 279, 286, 293, 307, Saint Jean Paulin, J., 120
314, 319, 324, 331, 348, 353, 355, 358, Salvadori, M.G., 182, 420
371, 373374, 390, 395396, 412, 414, Sampson, W.W., 30
417, 426, 430432 Samras, R.K., 125127
Sanchez-Palencia, E., 229
P Santalo, L.A., 1, 14
Sarkani, S., 209
Page, D.H., 139 Satake, M., 143, 145
Palasti, I., 23 Save, M., 165, 309
Pamin, J., 207, 356 Sawczuk, A., 207
Papanicolaou, G.C., 229, 385, 408 Schijve, J., 207
Papenfu, C., 276 Schoenberg, M., 393
Papoulis, A., 292 Schreurs, P.J.G., 324
Perkins, R.W., 208 Schulte, J., 259
Pinsky, M., 404 Scott, D.G., 23
Pompe, W., 83 Searles, D.J., 364
Ponte Castaneda, P., 314, 324 Sewell, M.J., 367
Porcu, E., 66 Shahinpoor, M., 208
Prager, S., 342 Shahsavari, H., 123, 126, 128, 130
Prager, W., 165166, 303, 309 Sharma, V.D., 423
P1: Naresh Chandra
July 5, 2007 18:26 C4174 C4174C013
Author Index 463
Shield, R.T., 319 U

Shinozuka, M., 292
Siegmund, T., 288 Uscinski, B.J., 385
Sievert, R., 379
Sih, G.C., 207
V
Singh, A., 207
Skalak, R., 420 Vakulenko, A.A., 265
Skop, R.A., 125127 van der Sluis, O., 324
Slepyan, L.I., 125 Vannucci, P., 176
Snyder, K.A., 99, 184 Verchery, G., 176
Sobczyk, K., 51, 56, 6970, 83, 385 Vinogradov, O., 163, 248
Sokolowski, M., 208 von Mises, R., 2
Somigliana, C., 205
Soszynski, R., 15
Spanos, P.D., 292 W
Spence, J.E., 385
Stahl, D.C., 139, 153 Waas, A.M., 115
Steele, C.R., 102 Wang, C., 148, 255, 269
Stephen, N.G., 91 Wang, G., 157, 161, 163
Sternberg, E., 208209 Wang, P.J., 91
Stojanovic, R., 206 Wang, X., 114, 286
Stoker, J.J., 405 Wang, Z.-P., 366
Strack, O.D.L., 141 Warren, W.E., 114
Stroeven, M., 248 Wehrli, Ch., 370
Stronge, W.J., 114 Weitsman, Y., 207
Sudria, J., 205 Werner, E., 288
Sun, C.T., 366 Wheeler, J.A., 87
Suquet, P.M., 247, 314, 324, 328 Whiteley, W., 134
Szczepankiewicz, E., 66 William, K., 207
Szczepinski, W., 294 Willis, J.R., 247248, 395
Wilmanski, K., 423
Wlodarczyk, E., 416
Woods, A.N., 395
T Wozniak, C., 107, 116, 208
Wriggers, P., 318
Takeuti, Y., 220
Talbot, D.R.S., 314
Tarasov, V.E., 380 Y
Tauchert, T.R., 207
Taylor, G.I., 324, 336 Yaglom, A.M., 292
Terada, K., 248 Yamazaki, F., 292
Teukolsky, S.A., 103 Yang, J.F.C., 208
Thomson, D., 208 Yavari, A., 209
Thorpe, M.F., 135, 175, 184, 266 Yongjian, R., 290
Tiersten, H.F., 198, 205206 Yserentant, H., 164
Torquato, S., 314, 342
Torvik, P.J., 129 Z
Toupin, R.A., 198, 205206
Trebicki, J., 83, 426, 430432 Zaoui, A., 229
Triantafyllidis, N., 91, 120, 208 Zbib, H., 207
Trovalusci, P., 107, 145 Zhang, P., 164
Truesdell, C., 191, 198, 205, 286 Zheng, Q.-S., 181
Tschoegl, N.W., 128 Ziegler, H., 172, 327, 343, 359, 363,
Tucker, C.L., 44 370371, 374, 376, 402
Tydeman, P.A., 139 Ziman, J.M., 146
P1: Naresh Chandra
July 5, 2007 18:26 C4174 C4174C013
P1: Naresh Chandra
July 12, 2007 19:27 C4174 C4174C014
Subject Index
A C
Acceleration wave, 422, 425, 428 Cardioid distribution, 26, 43
Acceleration wavefront, 386, 415416, 422, Cauchy-Schwartz inequality, 49
428, 433 Cauchy continuum, 229230
Angular distribution, 34, 269 Cauchy distribution, 27
Anisotropic material, 180181, 218220 Cauchy stress, 119, 128129, 188, 191192,
Antiplane elasticity, 9394 206, 232, 241, 365, 378, 381
square lattice, 9394 Causal distribution, 25, 275
Asymptotic theorem, 15 Central interactions, elasticity, lattice with,
Average/averaging theorem, 246, 252, 254, 117120
318319 Characteristic length, 110, 115116, 189,
212214, 216, 228, 232
B Classical continuum, 91, 145, 195196,
208209, 356, 359, 423
Beam Classical elasticity, 171190, 197199,
BernoulliEuler, 107110, 205210, 217219
114117 Clausius-Duhem inequality, 235, 382
elastic, 8891 CLM, 105, 177179, 183, 189191, 216227
fiber network rigidity, 139141 Complex process, 377379
microbeams, 5762 vs. compound, 376379
stubby, 112114 Compound process, 376, 378379
Timoshenko, 60, 85, 89, 110114, 122, Connectivity percolation, 135, 137
131, 140, 153, 395, 398401 Conservation principle, 194, 199
Bernoulli Constants
lattice, 2021, 258261, 336 Dundurs, 179
process, 84 Lamee constants, 96, 172, 197, 255, 408
trial, 1415, 18 multi-constant theory, 96
BernoulliEuler beam, 107110, rari-constant theory, 97
114117 Continuum
Beta distribution, 275276 Cauchy, 229230
scale dependence, 275276 classical, 91, 145, 195196, 208209, 356,
Bilinear elastic microstructure, 359, 423
414417 Cosserat, 145, 205, 208, 228229, 232
Binomial distribution, 1415, 18 fractal media mechanics, 380383
Bone material, 141, 208, 234, 269271 homogeneous cosserat, 228230
Boundary condition local model, 118
displacement response, 251254 mesoscopic physics, 276278
effect paradigm, 241244 micromorphic, 206, 229
kinematic, 139, 153, 198, 312, 322323, micropolar, 89, 109, 114, 143, 146, 235,
335, 337338, 341 398
mixed-orthogonal, 184 multipolar, 206
traction, 315317, 334336 nonlocal, 118119, 207
uniform, 251, 278, 312, 317, 366 nonlocal model, 118119
465
P1: Naresh Chandra
July 12, 2007 19:27 C4174 C4174C014
pseudo-continuum, 221 Dundurs constant, 179

random, 37, 237, 273, 292293, 422 Duplex-steel material, 288
random fields, 273275 Dynamic response, Hill(-Mandel)
restricted, 214216 condition, 366
strain-gradient, 119
strain-gradient model, 119120 E
thermodynamics, 359364
truss-like, 301303 Eigencurvatures, stress invariance,
Correspondence (or viscoelasticity) 223227
principle, 128 Eigenstrains, stress invariance, 184187,
Cosserat continuum, 145, 205, 208, 223227
228229, 232 Elastic beam, 8891
Couple-stress, 129130, 191192, 198, 207, Elastic beams, 8891
209, 211215, 224, 227, 229, 232, Elastic helix, 125, 131, 378
278279, 377 Elastic media, stochastic finite elements,
286293
Elastic-plastic microstructure, 322333
D Elastic strings, simple lattice, 8788
Elasticity, 367
Damage, 80, 154156, 159, 311, 314,
antiplane, 9394
354357
classical, 171190, 197199, 205210,
map, 156
217219
patterns, 154155
finite, 318322
stochastic (random) evolution,
gradient, 117120
354
in-plane, 9499, 257258
Diffusion process, 412, 416
mesoscale, 150154
Directional data, 23
micropolar, 191231, 233, 235
Discrete distribution, 275, 360
noncentrosymmetric micropolar,
Disk-matrix composite material, 259,
204205
261262, 282
nonlocal, 117120
Displacement response boundary
paper, 150154
condition, 251254
planar Cosserat, 210216
Dissipation function, 371, 374, 377
plane models, 174176
homogeneous, 371374
three-dimensional, 171172
quasi-homogeneous, 374376
two-dimensional, 172174
Distribution
Entropy, 45, 7982, 162163, 347, 363365,
angular, 34, 269
367, 369370, 432
beta, 275276
maximum, 79, 81, 162163, 363, 432
scale dependence, 275276
production rate, 370
binomial, 1415, 18
Ergodic/ergodicity, 45, 53, 61, 66, 7476,
cardioid, 26, 43
7879, 85, 147, 238, 240, 249251,
Cauchy, 27
253, 256, 279, 286, 303, 311, 313,
causal, 25, 275
332, 355357, 387, 423
discrete, 275, 360
Ergodic theorem, 74, 76, 240
lattice, 26
EulerBernoulli beams, 107110, 114117
mesoscopic, 276
Extremum or variational principle,
point, 25
8182, 292, 311, 319, 322323, 342,
Poisson, 2122
371, 395
probability, 1, 3, 710, 20, 30, 48,
65, 82, 84, 156, 277, 297,
364, 429 F
radial, 22
stationary, 85 Fermats principle, 330, 390392
von Mises, 26 Fiber network rigidity beam, 139141
Weibull, 43 Finite elasticity, 318322
wrapped, 26 Hill(-Mandel) condition, 318
P1: Naresh Chandra
July 12, 2007 19:27 C4174 C4174C014
Subject Index 467
Finite elements I
spectral, 126, 131, 395, 398, 402
for flexural waves, 398402 In-plane elasticity, 9499, 257258
for random media, 395402 Inelastic microstructure, 311358
for waves in rods, 395398 Inequality
stochastic Cauchy-Schwartz, 49
for elastic media, 286293 Clausius-Duhem, 235, 382
random field models and, Jensens, 59, 84
273309 Inhibition process, 19, 23
Flexural wave, 398402 Inhomogeneous material, 225227
spectral finite elements, 398402 Insufficient reason, principle of, 2
Floquet wave, 404407 Internal variables, thermomechanics, 235
Fractal, 35, 129, 209, 380382 Inverse amplitude process, 430431
mechanics of, 380 Isotropic material, 177179, 216217
Functionally graded material, 23, 105, Isotropic micropolar material, 196198
263265, 286
J
G Jensens inequality, 59, 84
Gaussian process, 52, 429 K

Geodesic, 329, 331333
Gradient elasticity, 117120 Kinematic boundary condition, 139, 153,
Green-Gauss theorem, 381 198, 312, 322323, 335, 337338, 341
L
H
Lamee constant, 96, 172, 197, 255, 408
Hamiltons principle, 88, 90, 199200 Lattice
Hard-core, inhibition process, 23 Bernoulli, 2021, 258261, 336
Harmonic wave, 125127, 403 with central interactions, elasticity,
Heat conduction, 73, 127, 209 117120
Helix, 87, 122123, 125126, 128131, 205, distribution, 26
377378 hexagonal, BernoulliEuler beams,
elastic, 125, 131, 378 114115
thermoelastic, 126 honeycomb, 99, 114
Hexagonal lattice, BernoulliEuler beams, micropolar, 88
114115 elastic beams, 8891
Hill(-Mandel) condition, 184, 229, 245, models, 133170
251253, 285, 311, 323, 341, 343, 348, one-dimensional, 8791
364365, 382383, 423 periodic, 157
dynamic response, 366 planar, 87, 8999, 101, 103, 105, 107123,
in dynamic response, 366 125, 127, 129, 131
in finite elasticity, 318 planar models, 87131
in thermomechanics, 364 simple, 87
Homogeneous Cosserat continuum, elastic strings, 8788
228230 square, 9394, 115, 131, 139, 282, 332
Homogeneous dissipation function, antiplane elasticity, 9394
371374 BernoulliEuler beams, 115117
Homogenization, 104, 117, 150151, 191, triangular, 9499, 107, 110, 114, 120122,
196, 229230, 237238, 255257, 291, 134135
321, 328, 334, 356, 365366, 382, 385, BernoulliEuler beams, 107110
407 Timoshenko beams, 110112
Honeycomb lattice, 99, 114 triple honeycomb, 99
Huygens principle, 390 Legendre transformation, 347, 353,
Hysteretic microstructure, 419422 367369, 373, 375
P1: Naresh Chandra
July 12, 2007 19:27 C4174 C4174C014
Linear elastic microstructure, 237271, hierarchies of, 251, 258, 311, 313, 315,
410414 317, 319, 321, 323, 325, 327, 329, 331,
Liouville theorem, 360, 362 333, 335, 337, 339, 341, 343, 345, 347,
Locally isotropic, smooth elastic material, 349, 351, 353, 355, 357
284286 property, 151, 276, 426
random fields, 273278
correlation structure, 282283
M second-order properties, 278283
random media, thermomechanic
Markov process, 356, 412, 417, 427, 430
response, 359383
Material
thermodynamic orthogonality, 370376
anisotropic, 180181, 218220
universal properties of, 279
bone, 141, 208, 234, 269271
Mesoscale elasticity, 150154
disk-matrix composite, 259, 261262,
Mesoscopic distribution, 276
282
Mesoscopic physics, 276278
duplex-steel, 288
Microbeams, 5762
fractal, 35, 129, 209, 380382
Micromorphic continuum, 206, 229
mechanics of, 380
Micropolar continuum, 89, 109, 114, 143,
functionally graded, 23, 105, 263265,
146, 235, 398
286
Micropolar elasticity, 191231, 233, 235
isotropic, 177179, 216217
Micropolar lattice, 88
isotropic micropolar, 196198
elastic beams, 8891
locally isotropic, smooth elastic,
Microstructure
284286
bilinear elastic, 414417
matrix-inclusion composite, 103, 178,
elastic-plastic, 322333
184, 228, 242, 320, 322, 324326,
hysteretic, 419422
328329, 331, 358
inelastic, 311358
multiply connected, 181183, 220221
linear elastic, 237271, 410414
null-Lagrangian, 180181
nonlinear, 311358, 414422
optimal use, 165169
nonlinear elastic, 311318, 417419
paper, 1517, 150154, 314318
random, 3544, 301309, 427433
plastic, 162, 165, 168, 207, 293, 298,
rigid-perfectly plastic, 333341
300302, 314, 324, 327333, 335,
thermoelastic, 347354
338, 379
viscoelastic, 341342
power-law, 313314
Mixed-orthogonal boundary condition,
random, mean field equations, 7273
184
random chessboard, 258, 263264, 327,
MMM principle, 240
336337
Multi-constant theory, 96
two-phase, 221222
Multiply connected material, 181183,
Mathematical morphology, 38, 426
220221
Matrix-inclusion composite, 103, 178, 184,
Multipolar continuum, 206
228, 242, 320, 322, 324326, 328329,
331, 358
Maximum entropy, 79, 81, 162163, 363, N
432
Mechanics of fractal, 380 Noncentrosymmetric micropolar elasticity,
Mesoscale, 425426 204205
bounding, kinematic, 334336 Nonlinear elastic microstructure, 311318,
bounds 417419
hierarchies, 251269 Nonlinear media wavefront, 422433
linear elastic microstructures, 237271 Nonlinear microstructure, 311358,
nonlinear, 311358 414422
universal properties, 279282 Nonlocal continuum, 118119, 207
variational principles, 319324 Nonlocal elasticity, 117120
crack density tensor, 267269 Null-Lagrangian material, 180181
elasticity, paper, 150154 Numerical solutions, 170, 385, 393
P1: Naresh Chandra
July 12, 2007 19:27 C4174 C4174C014
Subject Index 469
O conservation, 194, 199

correspondence (or viscoelasticity), 128
One-dimensional composite wave, extremum or variational principle,
403408 8182, 292, 311, 319, 322323, 342,
One-dimensional lattice, 8791 371, 395
Operator, 59, 7273, 144, 209210, 290, 292 Fermats, 330, 390392
random, 292 Hamiltons, 88, 90, 199200
Optimal, 87, 105, 133, 165, 167168, 288, Huygens, 390
304, 307309 of insufficient reason, 2
structure, 307, 309 MMM, 240
truss-like continuum, 165 Saint-Venants, 91
use material, 165169 variational, mesoscale bounds,
Orthogonality, thermodynamic, 327, 363, 319324
370, 372, 376377, 382 virtual work principle, 169, 198200
Probability
definitions of, 14
P distribution, 1, 3, 710, 20, 30, 48, 65, 82,
84, 156, 277, 297, 364, 429
Paper, 1517, 150154, 314318
measure, 1, 7, 9, 11, 14, 42, 386
elasticity, 150154
geometric objects, 114
fiber structure, 1517
Process
in-plane orthotropy, 150153
Bernoulli, 2021, 84
mesoscale elasticity, 150154
complex, 377379
random formation, 314318
complex vs. compound, 376379
Particle model, 157158
compound, 376, 378379
Partition theorem, 313
diffusion, 412, 416
Percolation, 133, 135, 137, 171, 184, 266, 282
Gaussian, 52, 429
connectivity, 135, 137
hard-core, inhibition, 23
rigidity, 133, 135, 137
inhibition, 19, 23
Periodic lattice, 157
inverse amplitude, 430431
Physics, mesoscopic, 276278
Markov, 356, 412, 417, 427, 430
Planar Cosserat elasticity, 210216
point, 1819, 2122, 27, 31, 36, 38, 44,
Planar lattice, 87, 8999, 101, 103, 105,
261, 279, 281
107123, 125, 127, 129, 131
random, 4586, 425427
Plane elasticity, 174
stochastic, 4751, 54, 355, 357, 403, 420,
Plane models elasticity, 174176
430431
Plane monochromatic wave, 204
vector random, 5455
Plane wave, 57, 204, 404
Wiener, 413, 427
Plastic material, 162, 165, 168, 207, 293,
Production rate entropy, 370
298, 300302, 314, 324, 327333, 335,
Pseudo-continuum, 221
338, 379
Point distribution, 25
Point process, 1819, 2122, 27, 31, 36, 38, Q
44, 261, 279, 281
Quasi-homogeneous dissipation
Poisson
function, 374376
distribution, 2122
inhomogeneous, point field, 2223
line field geometry, 137139 R
point field, 2123, 27, 31, 35, 146147,
184, 265266, 279, 344 Radial distribution, 22
simulation of, 22 Random
process, 406 boundaries, 5657
random lines in plane, 2730 chessboard, 258261, 336341
Voronoi tessellations, 31 composite, 406407
Power-law material, 313314 crack model, 3435
Principle evolutions, 403404
P1: Naresh Chandra
July 12, 2007 19:27 C4174 C4174C014
fiber network, 153154 dissipation function, 359363

fibers, 2735 postulate, 240241
field (See Random field) size, 263, 346, 423
line fields, 2735
lines, 1114 S
local medium, 209210
materials, 7273 Saint-Venants principle, 91
mean field equations material, 7273 Scaling
media, 359383, 385433 complex vs. compound processes,
class of models, 408410 376379
mesoscale response, 359383 in damage phenomena, 354357
models, 144 laws, 102, 259, 314
spectral finite elements, 395402 trend, 353
wavefronts, 385433 trends, 357358
waves, 385433 Segmented elastic bars, wave, 407408
media models, 144 Separation of scales, 74, 237, 240, 244,
microbeams, 5861 250, 273, 286, 288, 293294, 385,
microstructure, 3544, 301309, 427433 395, 423
one-dimensional composites, 403408 SFE (Stochastic Finite Elements)
points, 1114 elastic media, 286293
Poisson lines, 2730 phenomenological studies, 290293
process, 4586, 425427 phenomenological study, 68, 237, 241,
processes, 4586 273309
tessellations, 3035, 39 random field models and, 273309
vector processes, 5455 Shear band, 316, 324, 329331, 338
Random chessboard material, 258, Shock wavefront, 415
263264, 327, 336337 Simple lattice, 8788
Random continuum, 37, 237, 273, elastic strings, 8788
292293, 422 Slip-line, 293300
Random field Spectral finite elements, 126, 131, 395, 398,
continuum, 237, 273, 292293 402
discrete to continuum, 273275 for flexural waves, 398402
mesoscale, 273283 for random media, 395402
models, 273309 Square lattice, 9394, 115, 131, 139, 282, 332
one-dimensional, 4555 BernoulliEuler beams, 115117
scalar, 4554, 85, 323 Stationary distribution, 85
three-dimensional, 6272 Stochastic finite elements
two-dimensional, 6272 for elastic media, 286293
vector, 85, 210 random field models and, 273309
Random media Stochastic process, 4751, 54, 355, 357, 403,
spectral finite elements, 395402 420, 430431
thermomechanics, 359383 Stochastic propagation wave, 386395
wave, 385433 Stochastic (random) evolution, 354, 385,
wavefront, 385433 403404, 431
Random microstructure, 3544, 301309, in damage phenomena, 354
416, 427433 Strain-gradient continuum, 119120
Random process, 4586, 425427 Stress
Rari-constant theory, 97 Cauchy, 119, 128129, 188, 191192, 206,
Restricted continuum, 214216 232, 241, 365, 378, 381
Reynolds stress, 74, 85 couple-stress, 129130, 191192, 198,
Rigid-perfectly plastic microstructure, 207, 209, 211215, 224, 227, 229, 232,
333341 278279, 377
Rigidity percolation, 133, 135, 137 Reynolds, 74, 85
Rods, wave, 395398 Stress invariance, 105, 216
RVE (Representative Volume Element) CLM result, 177187, 216227
P1: Naresh Chandra
July 12, 2007 19:27 C4174 C4174C014
Subject Index 471
eigencurvatures, extensions to presence Two-dimensional elasticity, 172174

of, 223227 Two-phase material, 221222
eigenstrains, extension to presence of,
184187, 223227
U
Stubby beam, 112114
Surface wave, 56, 386 Uniform boundary condition, 251, 278,
propagation, 5657 312, 317, 366
Upper bound theorem, 334
T
V
Tessellations, random, 3035, 39
Theorem Variational, mesoscale bounds principle,
asymptotic, 15 319324
average/averaging, 246, 252, 254, Variational principle, 8182, 292, 311, 319,
318319 322323, 342, 371, 395
ergodic, 74, 76, 240 Vector random process, 5455
Green-Gauss, 381 Virtual work principle, 169, 198, 200
Liouville, 360, 362 Viscoelastic microstructure, 341342
partition, 313 Viscoelasticity principle, 128
upper bound, 334 von Mises distribution, 26
Thermodynamic orthogonality, 327, 363,
370, 372, 376377, 382 W
Thermodynamics continuum, 359364
Thermoelastic helix, 126 Wave
Thermoelastic microstructure, 347354 acceleration, 422, 425, 428
Thermomechanics flexural, 398402
Hill(-Mandel) condition, 364365 Floquet, 404407
with internal variables, 235 harmonic, 125127, 403
random media, 359383 one-dimensional composites, 403408
Thickness, wavefront, 422, 432 plane, 57, 204, 404
Three-dimensional elasticity, 171172 plane monochromatic, 204
Timoshenko beam, 60, 85, 89, 110114, 122, random media, 385433
131, 140, 153, 395, 398401 rods, 395398
Topology, 133134, 139, 141, 145146, segmented elastic bars, 407408
168169, 255, 263, 304, 308 stochastic propagation, 386395
structural, 133134, 139 surface, 56, 386
Traction boundary condition, 315317, surface propagation, 5657
334336 transient, 386, 408422
Transformation, Legendre, 347, 353, Wavefront
367369, 373, 375 acceleration, 386, 415416, 422, 428, 433
Transient wave, 386, 408422 nonlinear media, 422433
Trial, Bernoulli, 1415, 18 random media, 385433
Triangular lattice, 9499, 107, 110, 114, random microstructures, 427433
120122, 134135 shock, 415
BernoulliEuler beams, 107110 thickness vs. heterogeneity, 422426
Timoshenko beams, 110112 Weibull distribution, 43
Triple honeycomb lattice, 99 Wiener process, 413, 427
Truss-like continuum, 165, 301303 Wrapped distribution, 26
P1: Naresh Chandra
July 12, 2007 19:27 C4174 C4174C014

Martin Starzewski Microstructural Randomness

Hochgeladen von

Dokumentinformationen

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Martin Starzewski Microstructural Randomness

Hochgeladen von

Copyright:

Verfügbare Formate

P1: Binod

July 13, 2007 21:36 C4174 C4174C000

Chapman & Hall/CRC

No claim to original U.S. Government works

International Standard Book Number-13: 978-1-58488-417-0 (Hardcover)

Library of Congress Cataloging-in-Publication Data

Visit the Taylor & Francis Web site at

To Pauline and Michael

About the Author . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxv

1 Basic Random Media Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.4.3 Random Tessellations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

2 Random Processes and Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

2.6 The Maximum Entropy Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

3 Planar Lattice Models: Periodic Topologies and Elastostatics . . . . . . . 87

4 Lattice Models: Rigidity, Randomness, Dynamics,

5 Two- Versus Three-Dimensional Classical Elasticity . . . . . . . . . . . . . . 171

5.2.3 Multiply Connected Materials . . . . . . . . . . . . . . . . . . . . . . . . . 181

6 Two- versus Three-Dimensional Micropolar Elasticity . . . . . . . . . . . . 191

6.5.1 From a Heterogeneous Cauchy to a Homogeneous

7 Mesoscale Bounds for Linear Elastic Microstructures . . . . . . . . . . . . . 237

8 Random Field Models and Stochastic Finite Elements . . . . . . . . . . . . 273

8.2.2 Universal Properties of Mesoscale Bounds . . . . . . . . . . . . . 279

9 Hierarchies of Mesoscale Bounds for Nonlinear

9.5 Viscoelastic Microstructures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341

10 Mesoscale Response in Thermomechanics of Random Media . . . . 359

11 Waves and Wavefronts in Random Media. . . . . . . . . . . . . . . . . . . . . . . . .385

Spectral Finite Element for Flexural Waves . . . . . . . . . . . 398

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 459

Subject Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 465

0.1 Randomness versus Determinism

where L is a differential operator, f is a source or forcing function, and u is

1. Randomness of the operator:

2. Randomness of the forcing function:

3. Randomness of the boundary and/or initial conditions.

B = {B (); }. (0.4)

Each of the realizations follows deterministic laws of classical mechanics in

0.2 Randomness and Scales in Mechanics of Materials

conventional solid mechanics. This is the essence of the so-called separation of

The first inequality allows one to postulate the existence of a representative

This begs two questions.

1. How do we determine the operator L1 1 ?

L u = f. (0.7)

This is the basic question of stochastic mechanics for, if there is no dif-

3. How does one pass from the microscale d to the mesoscale L, so

4. Supposing we deal with a periodic or disordered lattice-type

0.3 Outline of Contents

Chapter 11 focuses on selected topics in wave propagation in random

1. Probability and random variables roughly at the level of Part 1 of

The book is an outgrowth of graduate courses on micromechanics, me-

About the Author

Martin Ostoja-Starzewski is a Professor of Mechanical Science and Engineer-

Originating, as legend has it, with the Buffon needle problem

In his foreword to Santalos treatise (1976), partly quoted as a motto to this

1.1 Probability Measure of Geometric Objects

2 Microstructural Randomness and Scaling in Mechanics of Materials

(a) Consider eleven possible sums: 2, 3, . . . , 12. This leads to NA = 1 so

where by we denote the space of all elementary events. Now,

This is the type of problem that motivated Laplace (1814) to come up

1. It contains a logical error: equally likely meaning equally possible

Basic Random Media Model 3

(a) (b) (c)

There are, in fact, three possible answers:

1. The random radius method: Let be a fixed diameter of the circle,

4 Microstructural Randomness and Scaling in Mechanics of Materials

1. How do we determine the operator L1 1 ?

L u = f. (0.7)