Fundamental of Stress Waves

Foundations of Stress Waves
This page intentionally left blank

ddddd
ISBN: 7–118–04015–0
Author: WANG Li-li (Li-Lih WANG)
Reviewer and Corrector: Zhu Zhao-xiang (Z.-X. ZHU)
(First Edition in Chinese, 1985)
(Second Edition in Chinese, 2005)
Author: WANG Li-li (Li-Lih WANG)

English Translators: YANG Li-ming, ZHOU Feng-hua, WANG Li-li
For the First Edition in English (2007)
Amsterdam • Boston • Heidelberg • London • New York • Oxford

Paris • San Diego • San Francisco • Singapore • Sydney • Tokyo
Elsevier
The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, UK
Radarweg 29, PO Box 211, 1000 AE Amsterdam, The Netherlands
First edition 2007
Copyright © 2007 Elsevier BV. All rights reserved
No part of this publication may be reproduced, stored in a retrieval system

or transmitted in any form or by any means electronic, mechanical, photocopying,
recording or otherwise without the prior written permission of the publisher
Permissions may be sought directly from Elsevier’s Science & Technology Rights
Department in Oxford, UK: phone (+44) (0) 1865 843830; fax (+44) (0) 1865 853333;
email: permissions@elsevier.com. Alternatively you can submit your request online by
visiting the Elsevier web site at http://elsevier.com/locate/permissions, and selecting
obtaining permission to use Elsevier material
Notice
No responsibility is assumed by the publisher for any injury and/or damage to persons
or property as a matter of products liability, negligence or otherwise, or from any use
or operation of any methods, products, instructions or ideas contained in the material
herein. Because of rapid advances in the medical sciences, in particular, independent
verification of diagnoses and drug dosages should be made.
British Library Cataloguing in Publication Data

A catalogue record for this book is available from the British Library
Library of Congress Cataloging-in-Publication Data

A catalog record for this book is available from the Library of Congress
ISBN–13: 978-0-08-044494-9
ISBN–10: 0-08-044494-6
For information on all Elsevier publications

visit our web site at books.elsevier.com
Printed and bound in Great Britain

07 08 09 10 10 9 8 7 6 5 4 3 2 1
CONTENTS
Preface xi
1 Introduction 1
2 Elementary Theory of One-Dimensional Stress Waves

in Bars 7
2.1 Material Coordinate System and Spatial
Coordinate System 7
2.2 Governing Equations of Longitudinal Waves in Bars
in Material Coordinates 10
2.3 Characteristic Lines and the Compatibility Relationships
along the Characteristic Lines 13
2.4 Elastic–Plastic Longitudinal Loading Waves in
Semi-Infinite Bars 17
2.4.1 Linear elastic waves 17
2.4.2 Elastic–plastic loading waves 22
2.5 Governing Equations of Longitudinal Waves in Bars in
Spatial Coordinates 26
2.6 Strong Discontinuity and Weak Discontinuity; Shock Waves
and Continuous Waves 29
2.7 Conservation Conditions across Wave Front,
Rankine–Hugoniot Relations 35
2.8 Dispersion Effects Induced by the Transverse Inertia 42
2.9 Torsion Waves in Cylindrical Bars 51
v
vi Contents
3 Interaction of Elastic Longitudinal Waves 57

3.1 Coaxial Collision of Two Elastic Bars 57
3.2 Interaction of Two Elastic Longitudinal Waves 58
3.3 Reflection of Elastic Longitudinal Waves at Fixed
End and Free End 60
3.4 Coaxial Collision of Two Elastic Bars with Finite Length 62
3.5 Reflection and Transmission of Elastic Longitudinal Waves
at the Interface of Two Different Bars 66
3.6 Reflection and Transmission of Elastic Waves in Bars with
Varying Cross Sections 69
3.7 Hopkinson Pressure Bar and Flying Piece 74
3.8 Split Hopkinson Pressure Bar 76
3.9 Dynamic Fracture Induced by Reflective Unloading Waves 87
4 Interaction of Elastic–Plastic Longitudinal Waves

in Bars 95
4.1 Interaction of Two Longitudinal Elastic–Plastic Loading
Waves in Bars 95
4.1.1 Head-on loading interaction of two strong-discontinuous
stress waves 96
4.1.2 Head-on loading interaction of two weak-discontinuous
stress waves 99
4.2 Reflection of Longitudinal Elastic–Plastic Loading Waves
at a Fixed End 102
4.3 Governing Equations and Characteristic Lines of
Unloading Waves 105
4.4 Pursuing Unloading by Strong-Discontinuous Unloading
Disturbances 110
4.4.1 Sudden unloading of strong-discontinuous loading
wave in a linear hardening bar 110
4.4.2 Sudden unloading of continuous loading waves
in a linear hardening bar 116
4.4.3 Sudden unloading of centered plastic loading waves 120
4.5 Pursuing Unloading by Weak-Discontinuous Unloading
Disturbances 126
4.5.1 Continuously unloading of centered plastic waves 128
4.5.2 Attenuation of shock waves in linearly hardening
materials 131
4.6 Shock Wave Attenuation Due to Pursuing Unloading 135
4.7 Propagating Properties of Elastic–Plastic Boundaries
in Semi-Infinite Bars 139
4.7.1 An analysis by the characteristic line method 139
4.7.2 Local linearization method 146
Contents vii
4.8 Head-On Unloading 148

4.9 High-Speed Impact of a Finite Bar onto a Rigid Target 153
4.9.1 Linear hardening bar 154
4.9.2 Increasingly hardening bar 161
4.9.3 Decreasingly hardening bar 164
4.10 General Properties of the Loading–Unloading Boundary
Propagation 167
4.10.1 Loading boundary and unloading boundary 167
4.10.2 Elastic–plastic boundary as a singular interface 169
4.10.3 Strong-discontinuous elastic–plastic boundary 170
4.10.4 First order weak-discontinuous boundary 171
4.10.5 Second order weak-discontinuous boundary 176
4.10.6 Discussions on higher-than-second order
weak-discontinuous boundaries 180
4.10.7 Higher order isolated points on elastic–plastic boundary 183
4.10.8 Supplementary conditions on loading boundary 190
5 Rigid Unloading Approximation 197

5.1 Rigid Unloading in a Semi-Infinite Bar 197
5.1.1 Rigid unloading in linear hardening plastic bars 197
5.1.2 Rigid unloading in linear elastic–linear hardening
plastic bars 200
5.1.3 Rigid unloading in linear elastic–decreasing
hardening plastic bars 202
5.2 Rigid Unloading in Finite Bars 207
5.3 Rigid Unloading Analyses for Shock Wave Propagation 212
5.3.1 Rigid unloading of shock wave propagating in
semi-infinite bars 212
5.3.2 Rigid unloading of shock wave propagating
in finite bars 215
6 One-Dimensional Visco-Elastic Waves and

Elastic-Visco-Plastic Waves 219
6.1 Linear Visco-Elastic Constitutive Relationship 220
6.1.1 Maxwell model 222
6.1.2 Kelvin–Voigt model 224
6.1.3 Standard linear solid model 226
6.2 Stress Waves Propagating in Linear Visco-Elastic Bars 227
6.2.1 Longitudinal visco-elastic waves propagating
in Kelvin–Voigt bars 227
viii Contents

in Maxwell bars 229
in standard linear solid bars 230
6.2.4 Solutions of linear visco-elastic longitudinal
waves propagating in bars by the
characteristics method 233
6.2.5 Split Hopkinson visco-elastic bar 237
6.3 Nonlinear Visco-Elastic Constitutive Relationship 246
6.4 Stress Waves Propagating in Nonlinear Visco-Elastic Bars 251
6.5 Elastic-Visco-Plastic Constitutive Relationship 255
6.6 Stress Waves Propagating in Elastic-Visco-Plastic Bars 260
7 One-Dimensional Strain Plane Waves 265

7.1 Governing Equations 265
7.2 One-Dimensional Strain Elastic Waves 267
7.3 Elastic–Plastic Constitutive Relationship in 1D
Strain Condition 270
7.4 One Dimensional Strain Elastic–Plastic Waves 276
7.5 Influence of Reverse Yield on the Propagation of 1D
Strain Elastic–Plastic Wave 279
7.6 State Equation of Solids under High Pressures 284
7.7 Shock Waves in Solids under High Pressures 292
7.7.1 Shock jump conditions 292
7.7.2 Shock adiabat 297
7.8 Interaction of Shock Waves in Solids under High Pressure 308
7.9 Plane Waves in Hydro-Elasto-Plastic Media 317
7.10 Attenuation of Shock Waves in Hydro-Elasto-Plastic Media 325
7.11 One-Dimensional Strain Elasto-Visco-Plastic Waves 333
8 Spherical Waves and Cylindrical Waves 337

8.1 Continuity Equation and Motion Equation 337
8.2 Elastic Spherical Waves and Cylindrical Waves 339
8.3 Elasto-Plastic Spherical Waves 347
8.4 Approximate Analysis for the Fragmentation of
Spherical Shells 354
8.5 Elasto-Visco-Plastic Spherical Waves and Cylindrical
Waves 357
Contents ix
9 Elastic–Plastic Waves Propagating in Flexible Strings 363

9.1 Governing Equations 364
9.2 Semi-Infinite Straight String under Abrupt
Constant-Velocity Oblique Impact 372
9.3 Infinite Straight String under Abrupt Constant-Velocity
Oblique Point-Impact 377
9.4 Prestretched Strings Subjected to Transverse Impact 388
9.4.1 Experimental investigation of wave velocity in a
prestretched string 391
9.4.2 Experimental study on constitutive relationship of
string materials 394
9.4.3 Determination of transverse wave velocity from
longitudinal wave velocity measurements 396
10 Elastic–Plastic Waves Propagating in Beams under

Transverse Impact (Bending Wave Theory) 399
10.1 Basic Assumptions and Governing Equations 399
10.2 Elastic Bending Waves 405
10.3 Plastic Bending Waves (Elastic–Plastic Beams) 409
10.4 Rigid-Plastic Analysis 432
10.5 Shear Failure of Beams under Transverse Impact 442
11 General Theory for Linear Elastic Waves 453

11.1 Linear Elastic Waves in Infinite Media 453
11.2 Oblique-Incidence, Reflection, and Transmission of
Elastic Plane Waves 457
11.3 Elastic Surface Waves 467
12 Numerical Methods for Stress Wave Propagation 475

12.1 Characteristics Numerical Method 477
12.1.1 Characteristics numerical method for 1D waves 477
12.1.2 Characteristic surfaces numerical method for
2D waves 482
12.2 Finite Difference Method 490
12.2.1 Establishment of finite difference format 491
12.2.2 Convergence of difference formats 494
x Contents
12.2.3 Stability of difference formats 495

12.2.4 Artificial viscosity 497
12.3 Finite Element Method (FEM) 500
12.3.1 Basic procedures of finite element method 500
12.3.2 Examples 503
Appendices 511
References 519
Index 529
Preface
Due to its practical and scientific importance, the theory of stress waves has been the subject
of numerous investigations conducted over the past several decades. It is found that the
propagation of stress waves is an important factor in earthquake, high velocity impact,
explosion, and weapon effects and describing the responses of materials and structures
under impact also needs the theory of stress waves.
Applications of the theory of stress waves in natural exploration and exploitation and in
technology development are relatively common, as knowledge that was once confined to a
few specialists is becoming more widespread. For instance, the inner structure of the earth
is imaged or natural resources (i.e. oil, coal bed) are prospected by using the propagation,
reflection, and refraction of wave, which is produced by a blast, underground. Stress waves
can also be used to examine the damage in materials or structures without effects on their
properties, to be a loading resource to investigate the dynamic properties of materials, to
rivet metal plates, to produce the spalling fracture. It can even be used in medicine to
destroy calculi in human body. These applications have furthered the in-depth study on
the theory of stress waves, and have shown that the investigation on the stress waves is
currently an important subject in solid mechanics and the theory of stress waves is the
basic of many science and technology branches.
The theory of stress waves was developed based on the study on the elastic waves. Unex-
pectedly, the investigation on the elastic waves is driven by the study on what nature light
was. In 1821, physicist Fresnal first proposed a new concept that transverse wave could
propagate in a medium, when he was studying the interference of polarized lights. Two
mathematicians Cauchy and Poisson were interested in the new concept, which resulted in
Cauchy building up the general governing equations for elastic body and Poisson finding
out that there were two kinds of waves, longitudinal wave and transverse wave, prop-
agating in elastic medium. Thus, the theory of stress waves was originated by both the
mathematicians.
A collision between two elastic solids was investigated, in earlier days, by using the
theory of vibration. However, it was not successful. Boussinesq (1883) and Saint-Venant
(1883) first began to apply the concept of stress waves to study the collision between elastic
straight bars (and between rigid body and straight bar). The investigation on the earthquake
had further driven the development of stress wave theory. In the Second World War, the
theory of stress waves had been developed quickly by Taylor (1942), Karman (1942) and
(Rakhmatulin) (l945), who had independently built up the theories of plastic
waves to satisfy the needs of military technology. This was followed by post-war period,
xi
xii Preface
when the theory of stress waves underwent holistic development, which can be referred
to in this book.
In the early 1960s, Professor Wang had lectured the course “Dynamics of Plasticity” in
University of Science and Technology of China. Based on his lecture script, “FOUNDA-
TION OF STRESS WAVES” (ddddd), the first edition in Chinese, was published by
National Defense Industry Press (Beijing, China) in 1985. Considering the overwhelm-
ing response from readers and satisfying their needs, the second edition in Chinese was
published in 2005. This book is a translation from the second edition of “FOUNDATION
OF STRESS WAVES” in Chinese. For convenience of studying, this book introduces the
theory of stress waves with a gradual build-up of complexity, which is its track of develop-
ment, from simple to complex and from specific to general. Thus, in this book, the theory
of the linear elastic wave is first studied, then the nonlinear elastic wave and the plastic
waves are discussed; the simple waves are studied earlier than the complex waves; the
waves in rate independent materials are analyzed before the rate dependent materials; one
dimensional problems are dealt with first, followed by the multi-dimensional wave; and
based on the results of study on continuous waves, the propagation of discontinuous wave
surfaces with different orders, such as shock wave (first order discontinuous wave) and
acceleration wave (second order discontinuous wave), are analyzed.
This book includes five topics. In the first one, discussions are as follows: one-dimensional
elastic waves, plastic waves, shock waves, unloading waves, then going with a grad-
ual build-up of complexity to linear visco-elastic waves, nonlinear visco-elastic waves
and visco-elastic-plastic waves. In the second, multi-dimensional waves, such as one-
dimensional strain plane waves, spherical waves, and cylindrical waves, where shock
waves under high pressure, elastic–plastic waves, and visco-elastic-plastic waves are men-
tioned. In the third, the theory of elasto-plastic waves in flexible strings and beams under
transverse impact is studied. The analyses include the propagation in strings, where longi-
tudinal waves and transverse waves couple each other, and the propagation in beams where
bending moment disturbances and shear disturbances couple each other also. After that,
the general three-dimensional theory of stress wave propagation is introduced. Finally, the
numerical methods used to solve the problems of stress wave propagation, including the
characteristics method, the finite difference method, and the finite element method, are
briefly presented.
This book comprehensively describes the fundamental knowledge of the stress waves
propagating in solids, which should appeal to a fairly wide audience. It is important for
studies on explosions, earthquake, and the dynamic responses of materials and structures
under impact loading. The book is intended to be understandable without being superficial.
The fundamental concepts are first described clearly, with a minimum of higher level
mathematics. This enables a reader to gain physical insight into the subject. Theoretical
analyses and practical applications are covered in detail. However, the studies presented
in the book draw on several scientific and engineering disciplines, and require some
familiarity with the basic principles of continuum mechanics and constitutive relationship
of materials. This book is not only suitable as an undergraduate and graduate text, but also
as a reference for engineers and researchers, particularly because it reflects the research
results originally published in Chinese by the author and other Chinese colleagues.
Preface xiii
If this book is used as a college text, covering all the content in a single semester would be
unlikely. The lecturer should select parts of the book that suit the needs and background
of the students. The first three chapters should be the foundation of courses. For an under-
graduate, part of Chapter 4 could be included. As a graduate text, part of Chapters 4–6
can be selected in addition. Other chapters in this book provide a useful reference for the
more advanced readers.
At last, we would like to express our appreciation to Mr. K. Qin, Miss J. Zhu,
Mr. Z.J. Sun and Mr. Y.G. Wang in Ningbo University, who assisted us in the translation
and the illustrations in this book.
YANG Li-Ming (Li-Ming YANG)

CHAPTER 1
Introduction
In a wide range of engineering sciences, military technology science, research, and even in
daily life, people are more and more faced with various problems in relation to explosion
or impact loading. It has been observed that the mechanical response of materials and
structures under explosive or impact loading differs markedly from that under static load-
ing. For example, when a stone hits onto a glass plate, the fracture usually occurs at the
back surface of the plate, called spalling. High-explosive plastic projectile destroying tank
armor is similar to this. For another example, when a metal bar is subjected to axial load
at one end, in the static case the deformation is almost uniformly distributed along the bar
length; while in the impact case (like hammering a drill rod or a pile), the deformation is
extremely non-uniformly distributed and the residual deformation is concentrated near the
hammered end of the bar. The mushroom shape of the projectile head formed when the
projectile impacts against a target is similar to this. The rapid development of dynamic
theory of solid mechanics is closely connected with the urgent requirement to solve such
sorts of mechanics problems.
Why those unusual phenomena mentioned above only occur under explosive or impact
loading? Why those phenomena are unable to be satisfactorily explained and solved on
the basis of the static theory of solid mechanics? What are the main differences between
the dynamic theory and the static theory of solid mechanics?
First of all, it is known that the subjects studied in the static theory of solid mechanics
are in states of static balance, with the prerequisite that the inertia of infinitesimal element
of media could be neglected. This is valid only when the applied load does not change
markedly with time. However, an explosive or impact loading is characterized by short
duration. The load itself and consequently other mechanical variables all markedly change
with time in the scale of milliseconds (ms), microseconds (µs), and even nanoseconds (ns).
Some typical examples are as follows. The pressure at the core zone of a nuclear explosion
dramatically increases to the order of 103 ∼ 104 GPa in few microseconds. The pressure
induced by a contact explosion between the explosive and the solid surface suddenly
increases to 10 GPa in a few microseconds. When a bullet is shot on a target at a speed
of 102 ∼ 103 m/s, the total duration of loading is about several tens of microseconds
1
2 Foundations of Stress Waves
and the pressure at the contact interface is as high as about 1 ∼ 10 GPa. Under such
intensive dynamic loading, the infinitesimal elements of media are in quickly changed
dynamic processes. Such problems are classified as dynamic problems where the inertia
of infinitesimal elements should be taken into account, and thus the problems are finally
attributed to studies of the propagation of stress waves.
In fact, when a part of a surface of a deformable solid is subjected to an external load,

at the beginning only those surface particles directly loaded move away from their initial
equilibrium positions. As there is relative motion between surface particles and their
neighboring particles (deformation), these two particles are loaded with each other, and
the acting force and the reacting force take place between them (stress). This makes
the neighboring particles begin to move away from their initial equilibrium positions.
However, due to inertia of particles of media, the motion of neighboring particles starts
later than surface particles. On this analogy, the disturbances caused by external load
on surface gradually propagate from the near to the distant in the media, and the stress
waves are formed. The boundary between a disturbed area and an undisturbed area is
named as wave front; its propagating speed is named as wave velocity. Wave velocity
of common materials is about 102 ∼ 103 m/s. It should be noticed that the wave velocity
must be distinguished from the particle velocity. The former is the speed that a signal of
disturbance propagates in media, while the latter is the speed of particle motion. If their
directions are identical, the wave is named as longitudinal wave. If their directions are
perpendicular, the wave is named as transverse wave. According to different geometric
shapes of wave front, there are plane wave, cylindrical wave, and spherical wave, etc.
Earthquake waves, sound waves, and ultrasonic waves in solids, shock waves in solids as
well are all typical examples of stress waves.
All solid materials have inertia and are deformable. Under an external load varied with
time, the deformation process of solid is substantially a process containing the propagation,
reflection, and interaction of stress waves. In the static problems of deformable solids, the
inertia of particles is neglected, or in other words, the propagation and interaction process of
stress waves before achieving the static equilibrium is disregarded or neglected, since only
the motion after achieving the static equilibrium is focused on. In rigid-body mechanics,
the ability to deform the material is generally neglected. This is, if the wave velocity
approaches infinity, then wave propagation is not considered. However, in the dynamic
problems of deformable solids under explosive or impact loadings, the wave propagating
process must be considered, since the time scale measuring the external load change is
in the same or lower order of the time scale, measuring the wave passing through the
characteristic length of the subject. In such cases, our main concern is the motion process
of deformable solid, and then the propagating process of stress waves must be considered.
Second, for an intense impulsive loading, since the load changes dramatically in a very
short duration, it implies a high loading rate or high strain rate. The strain rates in
ordinary static tests are about 10−5 ∼ 10−1 s−1 ; while the strain rates in impact test where
the propagation of stress waves must be considered, are about 102 ∼ 104 s−1 , even to
107 s−1 , namely they are much higher than that in static tests. A number of experimen-
tal results have shown that material behavior is generally different under various strain
rates. From the view-point of deformation mechanism of materials, except the ideal elas-
tic deformation which can be regarded as an instantaneous response, various inelastic
Introduction 3
deformation and fracture are non-instantaneous response all of which develop at finite
rate (like the process of dislocation motion, stress-induced diffusion, damage evolution,
crack growth, and propagation, etc.). Thus, mechanical behaviors of materials are essen-
tially rate-dependent, which generally are displayed as follows: with increasing strain
rate, the increase in yield strength and ultimate strength, the decrease in elongation, and
the delay of yielding and fracture, etc. Therefore, in addition to the particle inertia effect
mentioned earlier, the rate-dependent character of material constitutive relation is another
important reason why the mechanical response of structures under explosive or impact
loading much differs from that under static loading. In the view of thermodynamics, the
stress–strain process under static loading can be regarded as an isothermal process, and
the corresponding stress–strain curves can be regarded as isothermal curves. While the
dynamic stress–strain process under high strain rates can be regarded as an adiabatic
process accompanied with temperature change, consequently a thermo-mechanical cou-
pled process, and the corresponding stress–strain curves can be regarded as adiabatic
curves.
Therefore, to distinguish the dynamic response of structure under explosive or impact

loading and that under static loading, it should consider both the inertia effect of media
particles and the strain rate effect of material constitutive relation. When we treat dynamic
problems of solids under explosive or impact loading, we actually face two kinds of
problems. One is to study the motion of media (structures) under given external loadings
when the dynamic mechanical behavior is known; this is attributed to the study on prop-
agation of stress waves (direct problems). Another is to study the dynamic behavior of
materials under high strain rates by measuring and analyzing the propagation of stress
waves (inverse problems). The complexity of the problem lies in the coupling between the
inertia effect and the strain rate effect. In fact, on the one hand, no wave propagation can
be analyzed without knowing the corresponding dynamic constitutive relation of material,
and consequently the basic characteristics of wave propagation inevitably depend on the
strain-rate dependence of mechanical behavior of materials. On the other hand, the study
on dynamic mechanical behavior of materials generally depends on the analysis based on
the theory of stress wave propagation, namely wave propagation effects should not be
neglected in the study of dynamic constitutive relations and failure criteria of materials
under high strain rates. Therefore, studies on stress waves propagating in structures and
studies of dynamic mechanical behavior of materials are interrelated closely.
Essentially, all material constitutive relations are more or less rate-sensitive, but the rate-
sensitivity for different material is much different and depends on the stress range and the
strain rate range. Under certain conditions, material constitutive relation can be approxi-
mately regarded as rate-independent in a certain range of strain rate. Those theories of stress
waves based on such approximate assumption are named as rate-independent theories.
According to whether the stress–strain relation is linear elastic, nonlinear elastic, or plas-
tic, etc., the related theories are named as the theory of linear elastic waves, the theory
of nonlinear elastic waves, and the theory of plastic waves, respectively. On the con-
trary, theories considering rate-dependence of material constitutive relation are named as
rate-dependent theories of stress waves. According to the different rate-dependent stress–
strain relations such as visco-elastic, visco-elasto-plastic, or elasto-visco-plastic, etc., the
related theories are named as the theory of visco-elastic waves, visco-elasto-plastic waves
and elasto-visco-plastic waves, etc., respectively.
Research of stress waves was originated from the rate-independent theory. The theory of
linear elastic waves was developed in the twenties of the nineteenth century by Poisson,
, Stokes, et al. and later Rayleigh et al. in connection with the studies on
elastic vibration (refer to Kolsky, 1953).
Foundation of the theory of plastic waves is 100 years later than that of elastic wave theory.
There are two key difficulties: one is the nonlinear stress–strain relationship induced by
plastic deformation, which needs to develop nonlinear loading wave theory. The other is
that the stress–strain relation in unloading is different in loading due to the irreversibility
of plastic deformation, which needs to develop theory of unloading waves and to take
account of the interaction between the loading waves and the unloading waves. Looking
from this angle, the earliest research on plastic loading waves may trace to Donnell’s
work (1930). But the full theories of plastic waves were founded and developed by von
Karman (1942, in United States), Taylor (1942, in Great Britain) and , (1945,
in Soviet Union), respectively, due to the requirement of military technology during World
War II, although their researches were published only after World War II. It is worthwhile
to note that the plastic wave theory mentioned above was developed following such an
approach, namely, to generalize the linear theory of elastic waves under low stresses to
the nonlinear theory of plastic waves under high stresses, including the unloading waves.
On the other side, during World War II, the theory of plastic waves was developed following
another approach, namely, to generalize the hydrodynamics theory of solids under very
high pressure to the hydro-elastic-plastic theory of solids under medium pressure. In fact,
if the pressure applied to the solids is higher enough so that the shear strength of solids
can be ignored, then the solid under such high pressure can be regarded as compressible
fluids without shear strength. Thereby, the theory of shock waves in compressible fluids
was extended as the theory of shock waves in solids under high pressure (hydrodynamic
approximation). Typical works in this field include a number of works completed in the
forties by Rice et al. (1958). Subsequently, this theory was generalized to sub-high pressure
or medium pressure in which shear strength of solids is considered (Zheng and Xie, 1965;
Lee, 1971; Chou and Hopkins, 1972), and consequently the theory of one-dimensional
strain plastic waves (theory of hydro-elasto-plastic waves) was founded and developed.
As for the rate-dependent theory of visco-plastic waves, it was originated from the
one-dimensional theory of elasto-visco-plastic waves developed by
(Sokolovskii) (1948) and Malvern (1951), and then generalized to three-dimensional
theory by Perzyna (1963).
In the recent 70 years, studies and applications of stress wave theories develop rapidly.
These theories have been extensively applied in many areas, such as seismic research,
engineering blasting (mining, road construction, dam construction, etc.), explosive work-
ing (forming, bonding, combining, welding, hardening, etc.), explosive synthesizing
(synthetic diamond, synthetic BN, synthetic nano-materials, etc.), ultrasonic and sound
emission technology, dynamic response and impact strength of structures (buildings,
machines, constructions, etc.), ordnance effects (fragmentation of shells, penetration of
armor plate, high-explosive plastic projectile destroying tank armor, blast effects of explo-
sives or nuclear weapon and the corresponding protection, etc.), aero-vehicle impacted by
dispersed foreign objects, studies of meteorite craters on the earth and the moon, studies
Introduction 5
of dynamic response of materials under high strain rates and high pressure (including state
equations of solids, phase transformation, and changes of other physical properties),
studies of dynamic fracture, and the interaction between the materials and the particle
beams with high energy intensity like electron beams, X-rays, laser, etc.
In this book, the primary one-dimensional theory of stress waves in bars is first discussed
in Chapter 2. Then, from the shallower to the deeper, elastic waves (Chapter 3), elasto-
plastic loading and unloading waves (Chapter 4), rigid unloading (Chapter 5), visco-elastic
waves and visco-elasto-plastic waves (Chapter 6) are discussed in sequence. Although each
chapter is not all-inclusive, its content is elaborately arranged into a system, and described
from a consistent logic and unified treatment. Lagrange system is mainly used in this book,
while Euler system is also discussed. We sincerely hope that this book will help readers
to obtain a primary but systematic knowledge on the theory of stress waves.
For those reading the theory of stress waves for the first time, the first five chapters are
elementary. Although contents in those chapters are limited to one-dimensional motion,
the mathematical analyses are relatively simple, the basic concepts are readily compre-
hensible and the problem-solving methods are easy to be mastered. The theory of stress
waves mainly focuses on the inhomogeneous and nonsteady motion of media varying with
coordinates and time, emphasizes the analyses of local effects and early effects in media
induced by dynamic loading. In analyses of stress waves, it stresses to pay more attention
to the coupling effect between loads and media and the interdependence between stress
waves and dynamic constitutive behavior of materials. These are the main differences
between the dynamic theory and the static theory of solids, and can be understood from
the easy to the difficult and complicated in the first five chapters.
In Chapter 7, the theory of one-dimensional strain plane waves is discussed. Here, although
the strain state is one dimensional, while the stress state is three dimensional; the constitu-
tive relation of material is much complicated. The problems in this chapter are discussed
from two sides. On one side, along with increase of stress, the analyses on one-dimensional
strain elastic waves under low stresses are extended to the one-dimensional strain elasto-
plastic waves under high stresses. On the other side, first the shock wave theory of solids
under high pressures, in which the shear strength of materials is neglected, is discussed,
and then along with decrease of pressure, the one-dimensional strain elasto-plastic waves
(hydro-elasto-plastic model) are discussed. Because of this, the state equations of solids
under high pressures are discussed briefly in this chapter.
In Chapter 8, spherical waves and cylindrical waves are discussed together, since they
are similar not only in their mathematical description but also in their character of wave
propagation. Here, the discussion also begins from elastic waves, then elasto-plastic waves,
and elasto-visco-plastic waves in the last. Besides, an example of rigid-plastic analysis is
presented.
As an example of interaction of longitudinal waves and transverse waves, the theory of

elasto-plastic waves in flexible strings under transverse impact is discussed in Chapter 9.
A flexible string is such a simple structure component that only bears tension (stretch)
while its bending resistance is disregarded. Under transverse impact, there are two kinds
of wave propagating in strings: one is the longitudinal wave which induces the change
of stress and strain but no change of the string shape, and another is the transverse wave
which induces the change of string shape but no change of the stress and strain, however,
they influence each other.
Based on the theory of elasto-plastic waves propagating in flexible strings and considering
bending resistance of structural element, the theory of elasto-plastic waves propagating
in beams under transverse impact (bending waves) are discussed in Chapter 10. Herein,
the propagation of bending moment disturbances and shear disturbances are coupled each
other. Although these discussions are limited to the primary theory based on the assump-
tions of “plane cross-section” of beams and rate-independence of constitutive relations, it
is expected that they are helpful to beginners.
To obtain elementary knowledge of the general three-dimensional theory of stress wave

propagation, the linear elastic waves in infinite homogeneous media are discussed in
Chapter 11, including the wave reflection and transmission under oblique incidence and
the Rayleigh surface waves.
Finally, to adapt to the rapid development and extensive application of numerical sim-
ulation by computer, the numerical methods used to solve the problems of stress wave
propagation, including the characteristics method, the finite difference method and the
finite element method, are briefly presented in Chapter 12.
Surely, the contents of stress wave theory should be much more than those mentioned
above. The present book only provides a basis for further deeper studies. More specific
contents can be found in references.
The trends of current studies of stress waves include the following aspects: further extend-
ing the one-dimensional theory to the two-dimensional and three-dimensional theory;
developing from the stress wave theory for small deformation to the theory for large
deformation; from the rate-independent theory to the rate-dependent theory; from the
pure mechanical theory to the thermo-mechanical coupled theory; from the theory for
isotropic materials to the theory for anisotropic materials; from the theory for homoge-
neous materials to the theory for heterogeneous materials (such as composites, porous, and
multi-phase materials); wider application of computers in numerical simulation of stress
wave propagation; and developing new experimental technologies.
CHAPTER 2
Elementary Theory of One-Dimensional Stress

Waves in Bars
2.1 Material Coordinate System and Spatial Coordinate System
A coordinate system should be chosen first for investigating the motion of a bar.
In continuum mechanics there exist two coordinate systems (methods) to study the move-
ment of the medium, namely the material coordinate system (Lagrange method) and
the spatial coordinate system (Euler method).
One of the basic concepts of continuum mechanics is that the material’s realistic
(microscopic) structure is not taken into account in the analysis. Instead a medium is math-
ematically idealized as a system constituted of continuously distributed mass points.
In other words, a medium is a continuous collection of mass points. A mass point is
described by its location in the space. Different mass points have different locations at a
time. At a specified time, the locations of all mass points within a medium constitute a
configuration. To distinguish the mass points, each point must be given a name. Simul-
taneously each point has a spatial location at a given time. A reference spatial coordinate
system is used to specify the location of mass points.
Take the one-dimensional movement of a bar as an example. Consider a mass point spec-
ified by X (X is the name of the point), the spatial location of the mass point is expressed
by x. The motion of the media is generally represented by a different spatial location x
occupied by an arbitrary mass point X at different time t, namely, x is the function of X
and t:
x = x(X, t) (2.1)
For a fixed mass point X, Eq. (2.1) defines the spatial location varying with time, namely,
the motion of this mass point with time. At a fixed time t, on the other hand, Eq. (2.1)
defines the spatial locations of all mass points. In general, at a fixed time each mass
point has a unique location; conversely at each location there exists only one mass point.
Therefore at a specified time we can distinguish the mass points by their spatial locations.
7
In other words, as long as the media motion is continuous and single-valued, we can
reverse Eq. (2.1) to give:
X = X(x, t) (2.2)
A simple method to name a mass point is to use its spatial location x0 at a reference time t0 ,
and denote x0 as X. In this case Eqs. (2.1) and (2.2) give the conversion relationships of
mass point locations at time t0 and time t. We may chose t0 = 0, i.e. the initial time as
the reference time. Other proper reference times are also possible. For the convenience of
investigation, the reference spatial coordinate system at time t0 used to define the mass
point can be identical to, or different from the coordinate system used to describe the mass
point motion.
Therefore, there are two methods to study the motion of media. In the first method, we
follow the fixed mass point to observe the motion of media, and what we investigate are
how the physical quantities for this fixed mass point vary with time and how the physical
quantities vary with different mass point. Mathematically, any physical quantity ψ is
viewed as the function of mass point X and time t, ψ = F (X, t). This method is called
Lagrange method. The independent variable X is called Lagrange coordinates or the
material coordinates.
In the second method, we stay on a spatial point to observe the motions of material.
What we investigate is how the physical quantities of different mass points arriving at this
observation location at different time vary with time and how the physical quantities vary
with different spatial location. Mathematically, any physical quantity ψ is viewed as the
function of the spatial location x and time t: ψ = f (x, t). This method is called Euler
method. The independent variable x is called Euler coordinate or the spatial coordinate.
Making use of Eqs. (2.1) and (2.2), namely the conversions between the material coordinate
and the spatial coordinate at time t, we can transform the physical quantity ψ expressed
as a material coordinate function F (X, t) into a spatial coordinate function f (x, t), as:
f (x, t) = F [X(x, t), t]
or conversely:
F (X, t) = f [x(X, t), t]
Accordingly there exist two types of time derivatives. One is the time derivative of quantity
ψ at a fixed spatial location x, denoted as:

∂ψ ∂f (x, t)
= (2.3)
∂t ∂t x
which is called the spatial derivative (or Euler derivative). The other one is the time
derivative of quantity ψ observed by following a fixed particle X, denoted as:

dψ ∂F (X, t)
≡ (2.4)
dt ∂t X
Elementary Theory of One-Dimensional Stress Waves in Bars 9
which is called the material derivative (or Lagrange derivative). We can consider
the function F (X, t) in Eq. (2.4) as a compound function of variables (x, t), namely
f [x(X, t), t]. Using the chain rule to deduce the derivative of compound function, we
obtain:

dψ ∂f [x(X, t), t] ∂f [x(X, t), t] ∂x
= +
dt ∂t x ∂x t ∂t X

∂f (x, t) ∂f (x, t) ∂x
= +
∂t x ∂x t ∂t X
here (∂x/∂t)X is the material derivative of the spatial location x of particle X, namely the
velocity of particle X, denoted as v:

∂x dx
v= ≡ (2.5)
∂t X dt
Neglecting the subscripts, we obtain the following relationship:
dψ ∂ψ ∂ψ
= +v (2.6)
dt ∂t ∂x
If ψ represents particle velocity v, its material derivative is the acceleration of the
particle, a, expressed as:

∂v dv
a= ≡ (2.7)
∂t X dt
From Eq. (2.6) we have:
∂v ∂v
a= +v (2.8)
∂t ∂x
The first term in the right-hand side of (2.8) is the time derivative of particle velocity at
location x with regard to time t. This term is called the local acceleration, which is equal
to zero in the case of steady field. The second term in the RHS of (2.8) is the change rate
of particle velocity with regard to time due to the location change of the mass point, called
the migration acceleration, which is equal to zero in the case of uniform field.
It is noteworthy to notice that the wave velocity definition is closely related to the coordinate
system being used. If we observe the stress wave propagation in a material coordinate
system, at time t the wave front arrives at particle X, and X = Φ(t) represents the locus
of wave front propagation in material coordinate system, then

dX
C= = φ̇(t) (2.9a)
dt W
is called the material wave velocity (Lagrange wave velocity), or the intrinsic wave
velocity. If we observe stress wave propagation in a spatial coordinate system, at time t
the wave front arrives at spatial point x, and x = ϕ(t) represents the locus of wave front
propagation in spatial coordinate system, then

dx
c= = ϕ̇(t) (2.9b)
dt w
is called the spatial wave velocity (Euler wave velocity). These two wave velocities
describe the propagation behavior of one identical wave front. Because they are measured
in two different coordinate systems, their values are generally different. An exception is
the case when the wave front propagates into a media that is stationary and undeformed
before the wave front arrives, then the material wave velocity equals to the spatial wave
velocity.
After the definitions of wave velocity, we can discuss a third type of time derivative that
is frequently used in stress wave study. This is called the wave derivative, defined as
the total derivative of an arbitrary physical quantity ψ with regard to time following the
wave, namely (dψ/dt)w . Similar to the expression for the material derivative in a spatial
coordinate system (2.6), the wave derivative in the spatial coordinate system is:

dψ ∂ψ ∂ψ
= +c (2.10a)
dt W ∂t x ∂x t
and in the material coordinate system is:

dψ ∂ψ ∂ψ
= +C (2.10b)
dt W ∂t X ∂X t
Eqs. (2.10a) and (2.10b) again are descriptions of a same physical quantity from different
coordinate systems. If the quantity ψ specifies the spatial location of a mass point x(X, t),
and noticing that in case of one-dimensional material motion,

∂x
= (1 + ε)
∂X t
where ε is the engineering strain of the material, then we can deduce the relationship
between the spatial wave velocity c and the material wave velocity C of a plane wave
front:
c = v + (1 + ε)C (2.11)
Clearly for the plane wave propagating in a medium with zero initial speed and zero initial
strain, its material wave velocity and spatial wave velocity are the same.
2.2 Governing Equations of Longitudinal Waves in Bars in Material Coordinates
In this section we use the material coordinate system (Lagrange system) to study the
longitudinal motion of a bar with uniform cross section. The spatial coordinate of material
R S
P(X) P(X+dX)
X
dX
X X+dX
Fig. 2.1. An infinitesimal element of bar with uniform cross section in a Lagrange coordinate
system.
before deformation (t = 0) is taken as the material coordinate (Lagrange coordinate).

The axis of the bar is taken as the X axis (Fig. 2.1). Before deformation, the initial bar
cross-sectional area A0 , density ρ 0 and other initial material parameters are independent
of the coordinate. In the current problem no specific limitation on the bar’s cross-sectional
shape is necessary.
A basic assumption is: during deformation the bar’s cross section is kept as planar, and
the only axial stress is uniformly distributed in the cross section. With this assumption
all physical quantities are functions of X and t, and the motion of the bar is reduced to a
one-dimensional problem.
In the following discussions, the displacement u, strain ε = (∂u/∂X), particle velocity

v = (∂u/∂t), and stress σ denote the corresponding components in X direction (except for
special cases we do not use the subscript X). Note that the stress and strain used here are
engineering stress and engineering strain, and for the one-dimensional case, the Lagrangian
formulation strain ∂u/∂X is no longer limited to small strain.
The governing equations in the current problem include the kinematic condition (called the
continuous equation or the mass conservative equation), the dynamic condition (called the
motion equation or the momentum conservative equation), and the material’s constitutive
relationship (material equation). The mathematical forms of these equations are deduced
below.
Having noted that the strain ε and the particle velocity v are the first-order derivatives of
displacement u with respect to X and t, from the single-value, continuous property of u
we obtain a compatible relationship between ε and v. This is the continuous equation, as:
∂v ∂ε
= (2.12)
∂X ∂t
Consider a small segment dX of the bar (Fig. 2.1): on the left cross section R the segment
is subjected to force P (X, t), on the right cross section S it is subjected to force
∂P (X, t)
P (X + dX, t) = P (X, t) + dX
∂X
From Newton’s law we have:
∂v ∂P
ρ0 A0 dX = P (X + dX, t) − P (X, t) = dX
∂t ∂X
Substitute in the engineering stress σ = P/A0 , the motion equation of this segment is:
∂v ∂σ
ρ0 = (2.13)
∂t ∂X
Note that under the current Lagrange coordinate system, the derivative ∂/∂t in Eqs. (2.12)
and (2.13) by default means the material time derivative on a fixed X value. We do not
use d/dt expression further.
Here we discuss the material’s constitutive relationship within the framework of the rate-
independent theory. A second basic assumption made for the current analysis is: stress
is the function of strain. Mathematically the material’s constitutive relationship can be
written as:
σ = σ (ε) (2.14)
Because the stress wave velocity is very high, in the short time, stress wave propagates
through a segment, heat exchange of the segment with the neighboring regions is very
small. As a result we can consider the material’s deformation process to be adiabatic.
The constitutive equation listed here, is therefore an adiabatic stress–strain relationship.
Based on this concept we do not need to introduce other physical equations such as the
energy conservative equation, and the governing Eqs. (2.12)–(2.14) are closed for variables
σ, ε, and v. Given adequate initial conditions and boundary conditions we can solve this
group of equations for the three unknown functions σ (X, t), ε(X, t), and v(X, t).
In the discussions here and after, the stress and strain are positive for tensile case, and
negative for compressive case. The particle velocity is positive when the particle moves
in X axis, and vice versa.
In general the function σ (ε) is continuously differentiable. Introducing
1 dσ
C2 = (2.15)
ρ0 dε
we can eliminate the variable σ from Eqs. (2.13) and (2.14), obtaining:
∂v ∂ε
= C2 (2.16)
∂t ∂X
or, eliminate the variable ε from Eqs. (2.12) and (2.14), obtaining:
∂σ ∂v
= ρ0 C 2 (2.17)
∂t ∂X
Therefore the problem is reduced to solving the first-order partial differential Eqs. (2.12)
and (2.16) for unknown functions ε and v, or solving the first-order partial differential
Eqs. (2.13) and (2.17) for unknown functions σ and v.
If the expressions of ε and v in terms of displacement u are substituted into (2.16), we get
a second-order partial differential equation for the unknown function u, which is called
the wave equation:
∂ 2u 2
2∂ u
− C =0 (2.18)
∂t 2 ∂X 2
In the process of deducing the governing equation, we made two basic assumptions. In the
first assumption, we approximately neglected the inertial effect of the particles due to
their transverse motion. In other words, the kinetic energy corresponding to the transverse
expansion or contraction of the bar is omitted. In fact, the transverse motion of particles
will result in the non-uniform axial stress distribution on the cross section, and hence the
original planar cross section will be warped. Problem involving this deformation is no
longer a one-dimensional one, and is much more difficult to solve. However, according to
the known exact solutions for elastic waves, as long as the ratio of the wavelength to the
transverse size of the bar is large enough, the approximation neglecting the transverse
inertial effect is acceptable (refer to the Section 2.8). In this chapter all discussions on
stress wave propagations are based on this assumption. Theories developed are called the
primary theory or the engineering theory.
The second basic assumption is common in all rate-independent stress wave theories.
At first glance, this theory seems to be applicable only to the elastic deformation regime,
since elastic parameters are generally considered to be strain-rate independent, or to the
elastic–plastic materials that are insensitive to the strain-rate. However, as the strain-rate
under impact loading, is in general, several orders of magnitude higher than the quasistatic
case, and material properties under impact loading, is generally different from that under
quasistatic loading, the rate-independent assumption is better understood in the following
way: the material behavior under impact loading in a certain strain-rate region is described
by a unique stress–strain relationship, which is the average constitutive relationship of the
material within this strain-rate region, and may be different from the quasistatic one.
In this meaning, the strain-rate effect has been included in the constitutive relationship,
but not explicitly. Therefore the second assumption of rate-independence is applicable
practically to broader cases, and the rate-independent stress wave theories are useful in
many engineering applications. In Chapter 6 we will discuss the rate-dependent stress
wave theory.
2.3 Characteristic Lines and the Compatibility Relationships along the

Characteristic Lines
Below we discuss further the governing Eq. (2.18).
First, according to the assumption that “stress is a single-valued function of strain”,

C 2 = (1/ρ0 ) (dσ /dε) is also a function of strain ε(= ∂u/∂X), so Eq. (2.18) is linear
with respect to the second-order derivatives of u, and belongs to the type of second-order
quasi-linear partial differential equation of two independent variables. Under the special
case that stress is a linear function of strain, C 2 is constant, and (2.18) is a linear partial
differential equation.
Second, as long as we exclude the special case of unstable plastic deformation, stress must
be a monotonic increasing function of strain, namely dσ/dε > 0. Also the material density
ρ 0 is positive. Therefore the quantity C 2 is a positive value (C 2 > 0). According to the
categorization of second-order partial differential equations (one may refer to the textbooks
of mathematical physical equations), Eq. (2.18) is a hyperbolic partial differential equation
(PDE) (the so-called wave equation). This equation has two groups of real characteristic
lines, i.e. on the plane of the independent variables (X, t), there exist two different real
characteristic lines passing through any point.
Not only is the concept of the characteristic lines useful for categorizing PDEs, it is also
the fundamentals of one basic PDE solution technique, namely the method of charac-
teristics. Therefore, the concept of the characteristic line is very important to study the
propagations of waves, specifically the one-dimensional wave motion. Using the method
of characteristics, the problem to solve a PDE with two independent variables is reduced
to solving the ordinary differential equations (ODEs) along the characteristic lines.
The characteristic lines can be defined with different but equivalent approaches. There are
two major methods for doing this. The first is called the directional derivative method. If a
second-order PDE (or, equivalently, the linear combinations of a group of first-order PDEs)
can be reduced to the directional derivative along a certain curve C on the independent
variable plane (X, t), this curve C is called a characteristic line. The second method is
called the indeterminate line method. According to this definition, along a certain curve
C on the independent variable plane (X, t) the initial values are prescribed; if with the
prescribed initial values we still cannot determine all partial derivatives, the curve C is a
characteristic line. The two definitions from different angles of view reflect the essential
property of the characteristic line, respectively. Either method applied, we will get the same
results. In the following, we will mainly use the directional derivative method to discuss the
wave Eq. (2.18).
Consider a curve C (X, t) on the (X, t) plane. Along this curve the directional derivative
of the first-order derivatives of displacement u, namely v and ε, are:
∂v ∂v ∂ 2u ∂ 2u
dv = dX + dt = dX + 2 dt (2.19)
∂X ∂t ∂X∂t ∂t
∂ε ∂ε ∂ 2u ∂ 2u
dε = dX + dt = dX + dt (2.20)
∂X ∂t ∂X 2 ∂t∂X
In the above equations dX and dt are the components of a curve segment dS, in X and
t axes, so that dX/dt is the slope of curve C at point (X, t). If curve C is a characteristic
line of Eq. (2.18), the left side of (2.18) shall be reduced to containing only the directional
derivatives along the curve, namely a linear combination of Eqs. (2.19) and (2.20):
∂ 2u ∂ 2u ∂ 2u
dv + λdε = dt + (λdt + dX) + λ dX = 0 (2.21)
∂t 2 ∂X∂t ∂X 2
where λ is a coefficient to be determined. Comparing (2.21) with (2.18), the two equations
are equivalent under following conditions:
1 0 C2
= =− (2.22)
dt λdt + dX λdX
From the first equation, we have λ = −(dX/dt), then by substituting it into the second
equation, we obtain the characteristic direction dX/dt, as (dX/dt) = ±C. Rewrite the
relationship as:
dX = ±Cdt (2.23)
which is the characteristic line equation in differential form. An integration of (2.23)

leads to the characteristic line. Substitute (2.23) back into (2.22) we obtain λ = ±C.
So Eq. (2.18), or equivalently (2.21), is reduced to the ordinary differential equations that
only contain the directional derivatives along the characteristic lines:
dv = ±Cdε (2.24)
Equation (2.24) regulates a compatible relationship between v and ε on the characteristic

line, and is therefore called the compatibility equations on characteristic lines. There-
fore, solving the quasi-linear PDE (2.18) is equivalent to solving the characteristic line
Eq. (2.23) and the corresponding compatibility Eq. (2.24), both are ODEs.
The Eq. (2.23) defines characteristic lines in (X, t) plane. Similarly Eq. (2.24) can be
viewed as a characteristic line equation on (v, ε) plane, and the integration of which
leads to the characteristic lines on (v, ε) plane. In some occasions the (X, t) plane is
called the physical plane, and the (v, ε) plane is called the velocity plane. Therefore,
Eq. (2.23) geometrically represents two groups of characteristic lines in (X, t) plane, and
Eq. (2.24) geometrically represents two groups of characteristic lines in (v, ε) plane. The
characteristic lines in different planes are correlating (mapping) to each other. As shown
in Fig. 2.2, the region G in (X, t) plane and the region G in (v, ε) plane maps to each
other, and the same for the characteristic lines C and C , as well as the intersections of
characteristic lines, Q and Q . Such corresponding relationships provide the fundamentals
of the method of characteristics.
In the above discussions we used the directional derivative method to deduce the equations
of characteristics. We can also use the indeterminate line method to conduct the same
deductions. Note that the first-order PDEs (2.12) and (2.16) are equivalent to the second-
order PDE (2.18). We can rewrite all the first-order governing Eqs. (2.12), (2.16), (2.19),
X v
Q
Q'
G
C C' G'
o o
t e
Fig. 2.2. A region G in (X, t) plane is mapped to a region G in (ν, ε) plane.
and (2.20) as below:
⎧
⎪ ∂v ∂ε
⎪
⎪ − =0
⎪ ∂X
⎪ ∂t
⎪
⎪
⎪
⎪
⎪
⎪ ∂v 2 ∂ε
⎪
⎨ ∂t − C ∂X = 0
⎪
⎪ ∂v ∂v
⎪
⎪ dX + dt = dv
⎪
⎪ ∂X
⎪
⎪ ∂t
⎪
⎪
⎪
⎪
⎩ ∂ε dX + ∂ε
dt = dε
∂X ∂t
This group of equations can be viewed as an algebraic equation group of unknown par-
tial derivatives ∂v/∂X, ∂v/∂t, ∂ε/∂X, and ∂ε/∂t. In the matrix form the equations are
written as:
⎡ ∂v ⎤
⎡ ⎤⎢
⎢ ∂X
⎥
⎥ ⎡ ⎤
1 0 0 −1 ⎢ ∂v ⎥ 0
⎢ 0 −C 2 ⎥⎢ ⎥
⎢ 1 0 ⎥⎢ ∂t ⎥ ⎢
⎥=⎣
0 ⎥
⎣ dX dt 0 0 ⎦⎢
⎢ ∂ε ⎥ dv ⎦
(2.25)
⎢ ⎥
0 0 dX dt ⎢ ∂X ⎥ dε
⎣ ∂ε ⎦
∂t
If curve C is a characteristic line, the above equations have indeterminate solutions,

namely,
∆0 = ∆1 = ∆2 = ∆3 = ∆4 = 0
where

1 0 0 −1 0 0 0 −1

0 1 −C 2 0 −C 2
∆0 = , ∆1 = 0 1 0 , ∆2 = · · · , ···
dX dt 0 0 dv dt 0 0

0 0 dX dt dε 0 dX dt
After the expansions of these determinations, we can obtain again the characteristic line
Eq. (2.23) and the characteristic compatibility Eq. (2.24).
If we use the stress σ and velocity v as unknown functions, and start from the first-
order PDEs (2.13) and (2.17), we can arrive at the characteristic line Eq. (2.23), and the
corresponding compatibility equation as:
dσ = ±ρ0 Cdv (2.26)
Equation (2.26) is equivalent to (2.24). Actually, substituting (2.15) into (2.24), we obtain
Eq. (2.26) immediately.
In the following discussion, we will show that the characteristic line Eq. (2.23) physically
represents the propagation of disturbances. In other words, the characteristic lines on
(X, t)√plane are the propagation loci of the disturbances (wave fronts). The quantity
C = (1/ρ0 ) (dσ /dε) is the material velocity of disturbance propagation. The plus
symbol (+) in Eq. (2.23) represents the positive waves that are propagating to the right,
and the minus symbol (−) represents the negative waves propagating to the left. Equations
(2.24) and (2.26) determine the compatible relationships between the particle velocity v
and strain ε or stress σ in the process of disturbance propagation. The quantity ρ 0 C is called
the wave impedance. Note that in Eqs. (2.24) and (2.26) the (+) symbol corresponds to the
rightward wave and the (−) symbol corresponds to the leftward wave. This is converse to
the symbols in the compatibility relationships across a wave front [which will be discussed
in subsequent sections, see Eq. (2.63) in 2.7].
2.4 Elastic–Plastic Longitudinal Loading Waves in Semi-Infinite Bars
In this section, we discuss the longitudinal stress waves propagating in a semi-infinite

long bar. The bar extends from X = 0 to X = ∞. All stress waves propagate along
the positive X direction and no wave reflection is involved. An example of this case is
wave propagating unidirectionally in a finite length bar before reflections occur. Also in
this section we will consider only the loading case, namely the stress level in the bar is
monotonically increasing. At the boundary of the bar, the external loading can be prescribed
either in terms of boundary stress (∂ |σ |)/(∂t ) ≥ 0, or boundary speed (∂ |v|)/(∂t) ≥ 0.
2.4.1 Linear elastic waves
First, we consider the case when the impact loading is not high so that the bar deforms
elastically. In this case the stress–strain relationship is described by Hooke’s law, and the
constitutive relationship (2.14) is simplified as:
σ = Eε (2.27)
where E is Young’s modulus. In this case the wave Eq. (2.18) is simplified as a linear
wave equation:
∂ 2u ∂ 2u
2
− C02 2 = 0 (2.28)
∂t ∂X
where C0 is a constant that is determined from material parameters ρ 0 and E:

E
C0 = (2.29)
ρ0
From Eqs. (2.23) and (2.24) we see that the characteristic lines and the compatibility
equations are two groups of straight lines with slope ±C0 , on (X, t) plane and (v, ε) plane,
respectively:
dX = ±C0 dt
dv = ±C0 dε (2.30a)
By introducing the integration constants ξ 1 , ξ 2 , R1 , and R2 , equations in (2.30a) can be

represented by:

X − C0 t = ξ1
for rightward waves
v − C0 ε = R1

X + C0 t = ξ2
for leftward waves (2.30b)
v + C0 ε = R2
The constants R1 and R2 are sometimes called Riemann’s invariables.
Assume that the bar is at a stationary, stress-free state before loading. At time t = 0,
an impact is applied at the bar end X = 0. If, for example, the boundary condition
is prescribed as a particle velocity v0 (τ ), the problem is to solve the partial differential
Eq. (2.28) under the initial condition:
v(X, 0) = ε(X, 0) = 0, 0<X≤∞ (2.31a)
and the boundary condition:
v(0, t) = v0 (τ ), t ≥0 (2.31b)
or, to solve the characteristic Eqs. (2.30) for v and ε that satisfy the same boundary and
initial conditions. Actually, in solving this problem we deal with two kinds of initial-
boundary-value-problems (IBVP), namely the Cauchy problem and the Picard problem.
t v
B(X, t)
A
D
C(0, t) a
P
O Q R X O e
Fig. 2.3. Left: On (X, t) plane, a positive characteristic line and a negative one pass through an
arbitrary point; Right: On (v, ε) plane, the line Oa represents the simple wave relationship Eq. (2.33).
As shown in Fig. 2.3, passing through an arbitrary point on (X, t) plane, there exist
one positive and one negative characteristic line. The line OA in Fig. 2.3 is the positive
characteristic line passing through the origin (0, 0). We consider the region AOX below
line OA. Along the OX axis the values of v and ε are known as the initial conditions. Since
the two characteristic lines (QP and RP) passing through an arbitrary point (P ) in the
AOX region eventually intersect with the OX axis, the Riemann invariables along these
two characteristic lines R1 and R2 are determined by the initial conditions:
Along QP v − C0 ε = v(Q) − C0 ε(Q)

Along RP v + C0 ε = v(R) + C0 ε(R)
so that at the intersection point P of lines QP and RP, the values v(P ) and ε(P ) can be
solved from the above equations, as:
1
v(P ) = {[v(R) + v(Q)] + C0 [ε(R) − ε(Q)]}
2
1
ε(P ) = {[v(R) − v(Q)] + C0 [ε(R) + ε(Q)]} (2.32)
2C0
For the current case with zero initial conditions (2.31), v(Q) = ε(Q) = v(R) =
ε(R) = 0, we therefore get v(P ) = ε(P ) = 0. Since P is an arbitrary point in the AOX
region, this means that the whole AOX region is a constant-value region with v = ε = 0.
Indeed as long as the initial values are constant, namely v(Q) = v(R) = const.,
ε(Q) = ε(R) = const., the region AOX is a constant-value region with values:
v(P ) = v(Q) = v(R) and ε(P ) = ε(Q) = ε(R).
In the above discussions, the initial value curve OX is not a characteristic line. From any
point on this line, two characteristic lines grow with time into the region being considered.
All curves have similar properties (not necessarily parallel to the X axis) and are called the
space-like curve. If on a segment of space-like curve QR the velocity v and the strain ε
are prescribed, then the curvilinear triangle bounded by QR and two characteristic lines
QP and RP, namely the region QRP can be exclusively solved. This kind of IBVP is called
the initial value problem or the Cauchy problem.
Now we consider the region above OA line, namely the AOt region. Observe an arbitrary
point B in this region, the negative characteristic line passing through point B, BD,
eventually interacts with the OA line. Since we have known that v = ε = 0 along OA,
we have R2 = 0 in the AOt region. Namely,
σ
v = −C0 ε = − (2.33)
ρ0 C 0
On the other hand, the positive characteristic line passing through B always intersects with
the time axis Ot. Notice that on Ot the velocity v is prescribed as the boundary condition
(2.31b), so that along CB we have:
R1 = v − C0 ε = 2v = −2C0 ε = 2v0 (τ )
Note that the mathematical expression of the positive characteristic line CB is:
X = C0 (t − τ )
where C0 τ is the integration constant ξ1 , and τ is the intercept of the characteristic line on
t axis. Therefore, the v and ε values at any point B(X, t) in the AOt region are determined as:

X
v = −C0 ε = v0 t− (2.34)
C0
This means that the boundary velocity disturbance v0 (τ ) propagates into the bar with
velocity C0 , and at time t arrives at the cross section X. Therefore the characteristic lines
are physically the propagation loci of the disturbances (wave fronts). C0 is the longitudinal
elastic wave speed in the bar determined by the material parameters ρ 0 and E (2.29).
Similar to the concept of space-like curve, the non-characteristic curves such as Ot axis
are called the time-like curve. From any point on such curve only one characteristic line
grows with time into the region being considered. Consider a region that is bounded by a
characteristic line and a time-like line. On the characteristic line boundary, the quantities
v and ε are given, while on the other boundary, that is a time-like curve, one quantity v
or ε is given. We can solve the region using the approach similar to that used to solve the
AOt region. This kind of problem is called the mixed problem or the Picard problem.
As a result, solution of a semi-infinite bar impacted at the end is reduced to a Cauchy

problem in the AOX region, and a Picard problem in AOt region. The solution of the
Cauchy problem is completely determined by the initial condition. This means that the
region is not influenced by the boundary disturbances. The solution of the Picard problem
is determined jointly by the initial condition and the boundary condition. This means that
any point B in the AOt region is influenced by the initial disturbance propagating leftwards,
and the boundary disturbance propagating rightwards. Under the current initial-boundary
conditions (2.31), the initial disturbances are zero. Therefore the bar cross section X is at
rest until the first boundary disturbance (propagating along the characteristic line OA with
velocity C0 ) arrives, namely before t = X/C0 . The state of region AOX is mapped as the
origin point in (v, ε) plane. After the time t = X/C0 , the boundary disturbances v0 (τ )
arrive at the bar cross section X successively. Because the initial disturbances propagating
leftwards are all zero, the boundary disturbances propagating rightwards are not disturbed
and reserve the original boundary state (2.34). Such kind of undisturbed waves are called
the simple waves. For the elastic simple waves propagating into an initially stationary,
undeformed region, the mass velocity v, strain ε, and stress σ satisfy the Eq. (2.33), which
is called the simple wave relationship. Note that the physical states of all the points on a
leftward characteristic line DB in the AOt region are mapped onto the (v, ε) plane as line
Oa, and the simple wave equation is the mathematical equation of line Oa. Obviously,
if the bar has a uniform initial velocity v0 , initial strain ε0 , and initial stress σ 0 , then
Eq. (2.33) should be rewritten as:
σ − σ0
v − v0 = ∓C0 (ε − ε0 ) = ∓ (2.35)
ρ0 C0
where (−) corresponds to the rightward wave and (+) corresponds to the leftward wave.
This equation determines the basic relationship between particle velocity and strain or
stress, and is one of the most important equations in studying elastic wave propagation.
The quantity ρ 0 C0 is usually called the wave impedance or the sound impedance, and is
one of the most important parameters describing the mechanical properties of materials
under impact loading. Table 2.1 collects the elastic longitudinal wave speed C0 and the
wave impedance ρ 0 C0 of several common materials (Kolsky, 1953).
Note that, as shown in Fig. 2.3, the constant value region AOX on (X, t) plane is mapped
onto the (v, ε) plane as one point O, and the simple wave region AOt on (X, t) plane is
mapped onto (v, ε) plane as a line segment Oa. In other words, any characteristic line in
the simple wave region on (X, t) plane, representing a non-zero (un-trivial) disturbance, is
mapped as a point on (v, ε) plane. An important conclusion can be drawn from the above
discussions: A simple wave region is always neighboring a constant wave region.
The above discussions are based on the prescribed velocity boundary condition. If the
boundary condition is prescribed as a strain condition:
ε(0, t) = ε0 (τ ) t ≥0
or, as a stress condition:
σ (0, t) = σ0 (τ ) t ≥0
we can obtain the similar results.
Table 2.1. The elastic longitudinal wave velocity C0 and the wave impedance ρ0 C0 of several
common materials.
Steel Copper Aluminum Glass Rubber
ρ0 (103 =
kg/m3 g/cm3 ) 7.8 8.9 2.7 2.5 0.93
E (GPa = 1010 dyne/cm2 ) 210 120 70 70 2.0 ×10−3
C0 (km/s) 5.19 3.67 5.09 5.30 0.046
ρ 0 C0 (MPa/m/s = 106 kg/m2 /s) 40.5 32.7 13.7 13.3 42.8 ×10−3
t t t
6′
5′ 6′
A
v(X, t 1 4′ 5′
6 5 4 3 2 1
)
s 0 (t
t1
1′ 2′ 3′ 4′
5 6
t6
)
3′
(t) or
t5
2′
3 4
t4
e 0
1′
or
t3
v 0 (t)
2
t2
1
t1
v(s, e) o o X1 X v(s, e) o
v(X, t
v(s, e) 6 1)
5 4 3 2
1
o X
Fig. 2.4. The wave profiles (waveform curves) at any time, or the wave histories (temporal curves)
on any cross section of a bar can be determined by using the method of characteristics.
Based on the method of characteristics, a simple and convenient drawing method can be
used to determine the stress (or strain, or particle velocity) distributions at any time, or
the stress (or strain, or material velocity) history curves on any cross section. As shown
in Fig. 2.4, we can draw a horizontal line t = t1 on (X, t) plane. The line intersects with
each characteristic line at points 1, 2, 3, 4, 5, and 6. Since along each characteristic line,
v (or ε, or σ ) equals the known boundary value v0 (t) (or ε0 (t), or σ 0 (t)) depending on the
type of the boundary condition, therefore the velocity distribution at the time t = t1 , and
the corresponding strain distribution and stress distribution can be determined, which are
called the waveform curves as shown in the bottom of Fig. 2.4. Similarly, we can draw a
vertical line X = X1 on (X, t) plane, which intersects with the characteristic lines at points
1 , 2 , 3 , 4 , 5 , and 6 . The velocity, strain, and stress at each point (representing different
times) can be solved; hence we obtain the temporal curves at the cross section X1 , as
shown in the right part of Fig. 2.4. Using the waveform curves at a sequence of times,
or the temporal curves at different points, we can imaginarily describe the propagation
process of stress waves. The waveforms of the linear elastic waves are not changing during
propagation, because the wave velocity C0 is a constant.
2.4.2 Elastic–plastic loading waves
Consider an initially stationary and unstressed long bar that is subjected to an impact
loading at its end. By the simple wave relationship v = − (σ/ρ0 C0 ) , (2.33), we know
that the stress magnitude of the stress wave is proportional to the impact velocity. If the
dynamic uni-axial yielding stress of the material is denoted as Y , then once the impact
velocity v exceeds the so-called yielding velocity vY , namely,
Y
|v| > vY = (2.36)
ρ 0 C0
the material enters into the plastic deformation region and plastic waves will propagate
along the bar.
We now use the method of characteristics √ to solve the problem of elastic–plastic wave
propagation. As the wave speed C = (1/ρ0 ) (dσ /dε) is a function of strain ε, the
characteristic Eq. (2.23) and the compatibility Eq. (2.24)
dX = ±Cdt
dv = ±Cdε
do not represent straight lines on (X, t) plane and (v, ε) plane. However, by introducing a
variable
ε σ
dσ
ϕ= Cdε = (2.37)
0 0 ρ 0C
the compatibility equations on the characteristic lines, either Eq. (2.24) or Eq. (2.26), can
be written as the following unified form:
dv = ±dϕ (2.38)
which means that on (v, ϕ) plane the characteristic lines are two groups of orthogonal
straight lines that intersect with the coordinate axes by ±45◦ :

v − ϕ = R1
(2.38 )
v + ϕ = R2
Obviously, if the relationship σ = σ (ε) is known, the function C = C(ε) and ϕ = ϕ(ε)
are also known functions (Fig. 2.5).
To solve the problem of elastic–plastic wave propagation in a semi-infinite bar with the
initial-boundary conditions (2.31), the procedures previously applied for solving the elas-
tic wave problem are still available, namely, the problem is reduced to solve a Cauchy
problem in AOX region and a Picard problem in AOt region (Fig. 2.5). The elastic wave
parts in the constant value region AOX and in the simple wave region AOt are exactly iden-
tical with that discussed previously (Fig. 2.4). The plastic wave part corresponding to the
boundary condition |v0 (τ )| ≥ vY , however, should be handled separately. Since all neg-
ative characteristic lines will intersect with X-axis, the zero disturbance initial condition
(2.31a) eventually leads the Riemann invariables R2 in Eq. (2.38) to be zero. Therefore in
the plastic simple wave region following conditions exist:
ε σ
dσ
v = −ϕ = − Cdε = − (2.39)
0 0 ρ0 C
6′
t
6′
t t
4′ 5′
5′
4 3 2 1
t1 6 5
1′ 2′ 3′ 4′
6
t6
2′ 3′
t5 plastic zone
5 4
t4
1′
t3
3
elastic zone
t2
2
t1
1
o o
v vy X1 X v o
v 6 5 4
3
2
1
o X
s,C,f
s (e
) v v
f (e
)
y
C(
e)
o ey e ey o e o f
Fig. 2.5. Up: Characteristics lines in a simple wave region on (X,t) plane; Low left: Variances of
stress σ , wave velocity C, and variable φ with strain; Low Middle: States of simple wave region on
(v,ε) plane; Low Right: States of simple wave region on (φ,ε) plane.
and along the positive characteristic lines the Riemann invariable R2 can be determined
via boundary condition (2.31b), as:
R1
v = −ϕ = = v0 (τ )
2
namely, along a positive characteristic line the particle velocity v, strain ε, and stress σ
are invariant, and so C(ε) is invariant either. However, on different positive characteristic
lines, the values of C are different. As a result, in simple plastic wave region the positive
characteristic lines are a group of straight lines with different slopes (Fig. 2.5):
X = C(ε)(t − τ )
s s s
d 2s
=0
de 2
d 2s < 0
Y Y d 2s Y
E1
de 2 <0
de 2
o E o E o E
e e e
(a) (b) (c)
Fig. 2.6. Stress–strain curves showing different hardening properties: (a) decreasing hardening; (b)
increasing hardening; and (c) linear hardening.
where the quantity C physically represents the propagation velocity of the plastic wave.
According to Eq. (2.15) the plastic wave speed C is determined by the material density ρ 0
and the tangent modulus of the material dynamic stress–strain curve (dσ /dε 0 ). Depending
on the different strain-hardening characters of the stress–strain curve σ = σ (ε) there exist
three different cases as shown in Fig. 2.6, wherein Fig. 2.6(a) represents decreasingly
hardening character, (b) represents increasingly hardening character, and (c) represents
linearly hardening character.
Case (a). In this case the material tangent modulus decreases with the increase of
strain (decreasing hardening material), namely the σ ∼ ε curve is convex upwards
(d 2 σ/dε 2 < 0). The plastic wave speed decreases with the increase of strain. This means
that in loading process the propagation velocity of a higher magnitude plastic disturbance
is less than that of a lower magnitude plastic disturbance ahead. Therefore in a pro-
cess of wave propagation, the waveform becomes flatter and flatter (called the divergent
wave).
Case (b). In this case the material tangent modulus increases with the increase of
strain (increasing hardening material), namely the σ ∼ ε curve is concave upwards
(d 2 σ/dε 2 > 0). The plastic wave velocity increases with the increase of strain. This means
that during a loading process the propagation velocity of a higher magnitude plastic dis-
turbance is higher than that of a lower magnitude plastic disturbance ahead. Therefore,
the waveform becomes more and more steep with stress wave propagation (called the
convergent wave). Finally jumps of the particle velocity, stress and strain across a wave
front will form, and the plastic stress wave becomes a shock wave. Such shock wave
problem will be discussed further in Section 2.6.
In a special case, the material tangent modulus can be approximated as a constant E1

(d 2 σ/dε 2 = 0). This kind of material is called the linear hardening material, and
E1 is called
√ the linear hardening modulus. Then the plastic wave velocity is constant
C1 = E1 /ρ0 . In general E1 < E – for example, E1 /E = 0.003 − 0.010 for steel,
El /E = 0.05 − 0.1 for dry soil (Nowaski, 1978) – therefore the propagation of plastic
waves is much slower than that of the elastic wave.
The problem solved in Fig. 2.5 corresponds to the decreasing hardening material. On the
(X, t) plane the characteristic lines in elastic region are parallel straight lines, and the
positive characteristic lines in the plastic simple wave region are divergent straight lines.
The shape of the elastic parts of the temporal curve (before point 3 ) and the waveform
curve (before point 3) is invariable. Nevertheless the shape of the plastic parts is divergent,
and becomes more and more flat during propagation. The plastic waveform is elongated.
If the material is linear hardening, on the (X, t) plane the positive characteristic lines in the
plastic simple wave region is another group of parallel straight lines, whose slope differs
from that in elastic region. In this case, the elastic waveform and the plastic waveform are
invariable during propagation, but the interval between them becomes larger and larger.
The readers can show this by geometric drawing.
In Fig. 2.5 the corresponding v–ε diagram and v–ϕ diagram are also plotted. Same as
previously discussed, on these diagrams the constant value region is mapped to a point,
and the simple wave region to a line segment. It is notable that on the (v, ε) plane the
plastic simple wave region is mapped to a curve segment, which is described by the plastic
simple wave relationship (2.39). However, after introducing a variable ϕ on the (v, ϕ)
plane the region is again mapped as a segment of straight line.
The simple wave relationship (2.39) is satisfied everywhere within a simple wave region.
When applied to the impact end of the bar, this condition leads to the relationship between
particle velocity and strain or stress at the contact boundary. Further it can be deduced
that corresponding to the material’s ultimate strength σb , the critical impact velocity vc
is expressed as:
σb dσ
vc = − (2.40)
0 ρ0 C
Failure will occur if the particle velocity at the impact end of the bar reaches this critical
value.
2.5 Governing Equations of Longitudinal Waves in Bars in Spatial Coordinates
Now we use Euler method to study the elastic–plastic wave propagation. In this method
we consider a specified region in the space, called the controlled volume. In general
the conservation laws of mass, momentum, and energy are applied to a fixed mass of the
material. So when applying these laws to a controlled volume we must account the changes
of the physical quantities within the controlled volume, and the flow of the quantities across
the surfaces of this volume (the controlled surfaces). All physical quantities involved in
the wave propagation are functions of independent Euler variables: the spatial coordinate x
and the time t.
For the current one-dimensional uniaxial stress wave problem, we consider a controlled
volume between x and x + dx, as illustrated in Fig. 2.7. Assuming that during the wave
propagation any cross section of the bar is planar with uniform distributions of all physical
quantities, the mathematical one-dimensional spatial problem is posed below.
m(x) v (x) m(x + dx) v (x + dx)

m(x) v2 (x) m(x + dx) v2 (x + dx)
p(x) p(x + dx) x
dx
Fig. 2.7. An infinitesimal controlled volume of bar with uniform cross section in a Euler coordinate
system.
The mass occupied by a spatial segment of dx is:
M = ρAdx = ρ0 A0 dX
where ρ and A are the actual density and cross-sectional area of the bar, ρ 0 and A0 are the
initial density and cross-sectional area of the bar before deformation, dX is the length of
the current bar segment dx before deformation. The above equation represents the mass
conservation of a bar segment before and after deformation. From the definition of strain
dx = (1 + ε)dX, the line density (defined as mass per unit length) of the bar is:
ρ0 A 0
m = ρA = (2.41)
1+ε
According to the mass conservation of the controlled volume, the mass rate within a spatial
segment dx equals to the difference between the mass flowing into the volume and the
mass flowing out of the volume:
∂(mdx)
= m(x)v(x) − m(x + dx)v(x + dx)
∂t
Simultaneously, the momentum rate within dx equals to the momentum flow difference
into and out of the volume, plus the difference of the external forces:
∂(mvdx)
= m(x)v2 (x) − m(x + dx)v2 (x + dx) − P (x) + P (x + dx)
∂t
Here P = σA0 is the total force applied on a cross section. The above two equations can
be simplified as the following forms:
∂ε ∂ε ∂v
+v − (1 + ε) =0 (2.42)
∂t ∂x ∂x

ρ0 ∂v ∂v ∂σ
+v = (2.43)
1 + ε ∂t ∂x ∂x
which are the continuous equation and the dynamic equation expressed by Euler
coordinates.
Here again we assume that stress is a function of strain:
σ = σ (ε)
and denote the quantity (1/ρ0 )(dσ /dε) as C 2 , (2.15), the Eq. (2.43) can be rewritten as:
∂ε ∂v ∂v
(1 + ε)C 2 − −v =0 (2.44)
∂x ∂t ∂x
Equations (2.42) and (2.44) are a group of the first-order partial differential equations of
the unknown functions v and ε. These equations are similar to the propagation equations of
finite magnitude plane wave in ideal compressible fluid, which were treated by Earnshow
and Reimann (see Courant and Friedrichs, 1948). The continuous plastic wave is therefore
called Riemann wave in some occasions. We now use the method of characteristics to
solve these equations. A linear combination of these equations is:
∂ε ∂ε ∂v ∂v
L + Lv + M(1 + ε)C 2 −M − [L(1 + ε) + Mv] =0
∂t ∂x ∂t ∂x
here L and M are undetermined coefficients. According to the characteristic line defini-
tion, in case of a characteristic line this combination should contain only the directional
derivatives along the characteristic line, namely:
dx Lv + M(1 + ε)C 2 Mv + L(1 + ε)

= =
dt L M
From the second equation we obtain L = ±MC, therefore the characteristic lines in Euler
system must satisfy:
dx = [v ± (1 + ε)C]dt (2.45)
and the corresponding compatibility equations are:
dv = ±Cdε (2.46)
The corresponding equations in Lagrange system are Eqs. (2.23) and (2.24), respectively.
Equation (2.45) implies a relationship between Euler wave velocity c(= dx/dt) and
Lagrange wave velocity C, which is just the previously deduced Eq. (2.11). In the current
discussion the spatial location of particles before deformation (t = 0) are taken as the
material coordinate. When a wave front propagates in a material coordinate system with
velocity C, considering the deformation of the material coordinate itself, in the spatial
coordinate system the relative wave speed with respect to the particle ahead of the wave
front should be (1 + ε)C. This quantity is equivalent to the local sound speed in fluid
mechanics. Consider, again, that the particle itself is moving with velocity v, so that in
the spatial coordinate system the absolute spatial wave velocity should be [v ± (1 + ε)C].
This is the physical meaning of Eqs. (2.11) or (2.45). In these equations, (+) corresponds
to rightward waves and (−) corresponds to leftward waves.
Equation (2.46) is exactly the same as Eq. (2.24) in a material coordinate system, as it ought
to be, since the compatible relationships represent the continuous condition, momentum
conservation, and material constitutive relation, all these should not be dependent on what
kind of coordinate system is chosen.
As a matter of fact, the fundamental equations in Lagrange system (2.12) and (2.16), and
those in Euler system (2.42) and (2.44), are convertible by simply using the coordinate
conversion:

∂ ∂ ∂
= +v
∂t X ∂t x ∂x t

∂ ∂x ∂ ∂
= = (1 + ε)
∂X t ∂X ∂x t ∂x t
Although the appearances of these equations are different in different coordinate systems,
they represent the same physical essence.
In observing stress wave propagation, if the sensor is fixed in the observing space, the
measured wave velocity is a Euler speed; if the sensor is attached to the specimen such
as the bar, then the measured wave velocity is a Lagrange speed. In case the particle
velocity and the strain ahead of the wave front are zero, both measurements give the same
value.
2.6 Strong Discontinuity and Weak Discontinuity; Shock Waves

and Continuous Waves
On a wave front the particle displacement u must be continuous to satisfy the continuity
condition. However, the derivatives of u can be discontinuous. Mathematically this kind
of interface with “derivative discontinuity” is called a singular interface. If the first-order
derivatives, namely the particle velocity v(= ∂u/∂t) and the strain ε(= ∂u/∂x) have a
jump across a wave front, this interface (wave front) is called a first-order singular
interface, or a strong discontinuity. For example, in an increasing hardening material,
as the plastic wave velocity of higher magnitude disturbance is faster than that of a lower
magnitude disturbance, a strong discontinuity will finally form. This kind of stress wave is
called a shock wave. If u and its first-order derivatives are continuous but the second-order
derivative(s) such as the acceleration a(= ∂v/∂t = ∂ 2 u/∂t 2 ) is discontinuous, the interface
is called a second-order singular interface. This kind of stress wave is frequently called
an acceleration wave. And so forth there exist even higher singular interfaces. All second
or higher order singular interfaces are called weak discontinuities. For example, in a
decreasing hardening material the plastic wave of higher magnitude disturbance is slower
than that of lower magnitude disturbance, so that the plastic waves in this material are
weak discontinuous. This kind of waves have continuous waveforms, and therefore are
called the continuous waves.
In the previous discussions on stress wave propagation in a semi-infinite bar, as shown

in Figs. 2.4 and 2.5, we assumed that the impact velocity at the end of the bar increases
with time, namely the boundary loading is the so-called gradually increasing loading.
Mathematically the function v0 (τ ) in the boundary condition (2.31b) is a continuous func-
tion with respect to τ . Assume v0 (τ ) in Fig. 2.4 is a constant when τ > τ6 . As τ 6 approaches
to zero, the boundary loading is equivalent to a suddenly applied constant velocity impact.
The boundary condition is strong discontinuous in this case.
Whether the stress wave propagates in the form of strong discontinuity or weak discontinu-
ity is determined by the material’s stress–strain relationship and the boundary conditions.
We will discuss several different cases as follows (Fig. 2.8):
(a) In the case of linear elastic materials (Fig. 2.8a): If the boundary condition is a
weak discontinuous function of time, the elastic stress wave is a weak discon-
tinuous wave. If the boundary condition is of a strong discontinuity, the elastic
stress wave is a strong discontinuous wave. In other words, the type of stress
wave is determined by the boundary condition. Actually, if the boundary func-
tion v0 (τ ) in Fig. 2.4 is constant for τ ≥ τ6 , the region 6τ 6 t on X–t plane is
s s s
d 2s <0
E1 de 2
o E o E o E
e e e
t t t t
t1 t1 t1
o o
v o x x x
v v v
o x o x o x
t t t t
t1 t1 t1
o o o
v x x x
v v v
o o o
x x x
(a) (b) (c)
Fig. 2.8. Characters of stress waves propagation are different depending on stress–strain curves of
materials and the boundary conditions.
a constant value region. As τ 6 approaches to zero, all the previous positive char-
acteristic lines starting from the points between 0 and τ 6 on t axis overlap on one
line OA. The two regions across OA are constant value regions. Across the OA
line the quantities v, ε, and σ jump. Therefore the characteristic line OA is the
propagation locus of a strong discontinuity. The propagation velocity of this strong
discontinuity is also C0 . Summarizing the above discussions, for a linear wave
system, the type of the boundary discontinuity determines the type of propagation.
This is one of the important properties of the linear hyperbolic partial differential
system.
(b) In the case of linear elastic-linear hardening plastic materials (Fig. 2.8b): If the
boundary condition is a continuous function of time, both the elastic stress wave
and the plastic stress wave propagate with invariable waveforms. The intervals
between these two waveforms, however, become larger and larger. Here we have
assumed that the plastic hardening modulus E1 is less than the elastic modulus E.
If the boundary loading is a suddenly applied velocity (the strong discontinu-
ity), two strong discontinuous waves are created (the double-wave structure),
the first faster wave propagates as the characteristic line with slope C0 , across
which an elastic jump arises. The second slower wave propagates as the char-
acteristic line with slope C1 , across which a plastic jump arises. The distance
between these two abrupt wave fronts (shock waves) increases during wave
propagation.
(c) In case of linear elastic-decreasing hardening plastic materials (Fig. 2.8c): If the
boundary condition is a continuous function of time, as discussed in Fig. 2.5, the
weak discontinuous elastic–plastic waves are generated. Now consider the bound-
ary loading being a suddenly applied constant velocity. This loading is equivalent
to the problem of Fig. 2.5 when the boundary velocity v0 (τ ) is constant as τ ≥ τ6
and τ 6 approaches to zero. Therefore the previous parallel characteristic lines in
the elastic wave region overlap as one line OA, representing a strong discontinuous
elastic wave. The previous divergent characteristic lines in the plastic region now
start from one point O (which is a singular point), forming a so-called centered-
wave structure. In the current situation, the elastic part of the strong discontinuous
boundary disturbance, propagates as a strong discontinuous wave, and the plastic
part propagates as a series of divergent continuous waves, which are of weak discon-
tinuity. The waveform becomes more and more flat during propagation. Therefore,
for material of d 2 σ /dε 2 < 0, even if a strong discontinuity exists in the initial
condition or boundary condition, this discontinuity disappears when waves start
propagating. The strong discontinuity does not proagate in the form of strong
discontinuous wave.
(d) In the case of linear elastic-increasing hardening plastic materials (see Fig. 2.6b).
As pointed out previously, for plastic waves:
dC 1 d 2σ
= >0
dε 2ρ0 C dε 2
namely the velocity of a high magnitude plastic disturbance is larger than that
of a low magnitude plastic disturbance. Therefore the plastic waves are so-called
s
t1 t2 >t1 t3 >t2
C(b)>C(a) C(b)>C(a) C(b)=C(a)
b b b
a a a
o
X
Fig. 2.9. Formation of a shock wave in material with increasing hardening plasticity.
convergent waves. The waveform becomes steeper and steeper during propagation,
and finally the stress, strain, and particle velocity almost jump instantaneously,
forming a shock wave (Fig. 2.9). Note that this strong discontinuous (shock) wave
differs from the above discussed (in a and b) shock waves, which are generated by
the suddenly applied boundary loading. Actually, the strong discontinuous waves
in (a) and (b) originate from the strong discontinuous boundary disturbance, and
thermodynamically no extra entropy production exists in wave propagation. The
current strong discontinuous wave, however, is created by the convergent property
of the plastic waves, whatever the boundary conditions are. In the subsequent
discussions (Section 2.7) we will see that this shock wave accompanies extra entropy
production.
Although in case (d) the shock wave formation originates from the specific nonlinear
constitutive property of the material, the shock wave’s formation time and site are also
dependent on the initial or boundary conditions. Take the rightward simple wave case as
an example. Assume at time t = 0 the initial waveform be expressed as:
σ (X, 0) = f (ξ )
where ξ is the X locations of a stress disturbance σ at t = 0. For rightward waves, the

location of the same disturbance at time t is X = ξ + C(σ )t, therefore the waveform at
time t is expressed by:
σ = f [X − C(σ )t]
Using function F , the inverse function of f , the location of disturbance σ at time t is:
X = F (σ ) + C(σ )t
Condition of the shock wave formation is:

∂σ ∂X
= ∞ or =0 (2.47)
∂X ∂σ
from which the time of shock wave formation t0 , and the corresponding shock wave
location X0 are deduced as:

1 F (σ )
t0 = − = −
C (σ )f

min C (σ ) min
C(σ )F (σ )
X0 = F (σ ) − (2.48)
C (σ )
where {a}min represents the minimum of argument a. Actually any stress disturbance
corresponds to a time that satisfies condition (2.47), the earliest one corresponds to the time
of shock wave formation. If the initial condition already includes a strong discontinuity,
then we have f (ξ ) = ∞, and t0 = 0.
Similar approach is applied to the problem with a given boundary condition:
σ (0, t) = g(τ )
For rightward stress wave propagation, as X = C(σ )(t − τ ), we have:

X
σ =g t−
C(σ )
By representing G as the inverse function of g, the above equation is rewritten as:
X
t = G(σ ) +
C(σ )
Condition of the shock wave formation is:
∂σ ∂t
= ∞ or =0
∂t ∂σ
from which the time of shock wave formation t0 , and the corresponding shock wave
location X0 are deduced as:

C2 C 2 G
X0 = =
gC min C min
CG
t0 = G(σ ) + (2.49)
C
If the boundary condition already contains a strong discontinuity, then g = ∞, X0 = 0,

namely the suddenly applied boundary loading propagates as a strong discontinuous wave
from the very beginning.
It is noteworthy that for a suddenly applied boundary loading, for example, the case
of v(0, t) = v1 , the problem of the elastic–plastic wave propagation in a semi-infinite
bar has a self-similar solution (see Appendix 3). The first plastic wave theory proposed
by von Karman (1942) is for this kind of problem where the material is of decreasing
hardening plasticity. Such a problem does not contain a characteristic length or time.
Applying dimensional analysis to the governing Eqs. (2.12) and (2.16) where ε and v are
unknown variables, from π theorem the strain ε should be expressed as the following
non-dimensional form:

C(ε) X
ε=g ,
v0 v0 t
In other words, the variables X and t are not independent but appear in a combined form
of X/t. Introducing a variable β = X/t, the strain is a function of β:

X
ε=f = f (β)
t
Accordingly, the displacement u is:

X X β
∂u
u= dX = f (β)dX = t f (β)dβ
∞ ∂X ∞ ∞
and the second-order partial derivatives of u with respect to X and t are:
∂ 2u 1
2
= f (β)
∂X t
∂ 2u β2
= f (β)
∂t 2 t
Substituting this into the governing Eq. (2.18), we have

β 2 − C 2 f (β) = 0
then two solutions may exist (Fig. 2.10):
f (β) = 0
β2 = C2
The first solution renders ε = f (β) = const. that corresponds to a constant value region.
The second solution renders X/t = C(ε) that corresponds to a centered wave region.
As long as X/t is constant, the strain ε is invariable, regardless of what the value of X
or t is.
e = f(b )
e1
o
b1 =C(e1) b 0 =C0 b= X
t
Fig. 2.10. Self-similar solution for the elastic-plastic wave propagation in a semi-infinite bar under
constant velocity impact.
From the boundary condition

0
u = u(0, t) = v1 t = t f (β)dβ
∞
we obtain
∞ ∞
v1 = − f (β)dβ = − ε(β) dβ
0 0
which is the area integration under the curve shown in Fig. 2.10, and can be rewritten as
ε1 ε1
v1 = − β(ε)dε = − C(ε)dε
0 0
The above equation is the application of Eq. (2.39) at the end of the bar.
2.7 Conservation Conditions across Wave Front, Rankine–Hugoniot Relations
As a planar wave propagates as a singular surface in an infinite media, the physical quan-
tities across the wave front must satisfy certain conditions. In this section we will discuss
these conditions, in other words, the compatibility equations on the wave front.
Consider a planar wave front propagating along the positive X axis (rightward wave) with
the material wave velocity D = dX/dt, here X denotes Lagrange location of the wave
front at time t. Suppose we stand on the wave front to observe a physical quantity ψ(X, t).
The total rate of ψ with respect to time, namely the wave derivative (2.10) is:
dψ ∂ψ ∂ψ
= +D (2.50)
dt ∂t ∂X
Denote the values of ψ just before and after the wave front by ψ + and ψ − , the difference
between them as [ψ]
[ψ] = ψ − − ψ + (2.51)
Obviously, if ψ is continuous across the wave front [ψ] = 0; otherwise [ψ]

= 0. The
quantity [ψ] therefore represents the jump of ψ across the wave front. Taking the wave
derivative of ψ + and ψ − by (2.50), the difference between them is:

d ∂ψ ∂ψ
[ψ] = +D (2.52)
dt ∂t ∂X
If across the wave front ψ is continuous, but its first-order derivatives are not (namely the
wave front is a first-order singular surface), we have

∂ψ ∂ψ
= −D (2.53)
∂t ∂X
which is the famous Maxwell theorem.
We use ∂ψ/∂t, ∂ψ/∂X instead of the ψ variable in Eq. (2.52), obtaining:

2 2
d ∂ψ ∂ ψ ∂ ψ
= + D
dt ∂t ∂t 2 ∂t∂X
2 2
d ∂ψ ∂ ψ ∂ ψ
= +D
dt ∂X ∂t∂X ∂X 2
Therefore if across the wave front ψ and all of its first-order derivatives are continuous,
but its second-order derivatives are not (namely the wave front is a second-order singular
surface), we have

∂ 2ψ ∂ 2ψ ∂ 2ψ
= −D = D2 (2.54)
∂t 2 ∂X∂t ∂X 2
The Eqs. (2.52)–(2.54) are the kinematic compatibility conditions on wave front, corre-
sponding to cases when ψ itself, its first-order derivatives, and its second-order derivatives
are discontinuous, respectively. Similar deductions can be derived on cases of higher-order
singular interfaces. For cases when the wave front is leftward, we can simply substitute
the quantity D by −D and all equations are true.
Take the quantity ψ as the particle displacement u(X, t). By the condition of continuity,
the displacement must be continuous across a wave front, namely [u] = 0. In case of
a shock wave front, by the compatibility condition of a first-order discontinuity (2.53)
we have:
[v] = −D [ε] (2.55)

In case of an acceleration wave front, by the compatibility condition of a second-order

discontinuity (2.54) we have:
2 2
∂ 2u ∂ u 2 ∂ u
= −D = D
∂t 2 ∂X∂t ∂X 2
or expressed by v and ε:

∂v ∂ε ∂v 2 ∂ε
= −D = −D =D (2.56)
∂t ∂t ∂X ∂X
Equations (2.55) and (2.56) are the kinematic compatibility conditions across a shock
wave front and an acceleration wave front, respectively. These conditions reflect
physically the condition of mass conservation.
Now we consider another type of compatibility conditions based on the law of dynamics.
Consider a strong discontinuous wave front. At time t the front is located at AB (Fig. 2.11),
in a time dt it arrives at location A B . The distance between these locations are connected
to the wave velocity D by dX = D dt. By the momentum conservation of the region
ABA B :

σ + − σ − A0 dt = ρ0 A0 dX(v− − v+ )
which can be simplified as:
[σ ] = −ρ0 D [v] (2.57)
In case the wave front is weak continuous, since [v] = 0, [σ ] = 0, we must consider the
relationships between the derivatives of v and σ . As an example, for an acceleration wave
we can apply the differential momentum conservation Eq. (2.13) to the materials before
and after the wave front; the difference of these two equations leads to:

∂σ ∂v
= ρ0 (2.58)
∂X ∂t
A A′
s− s+
e− Ᏸ= dX e+
v− dt v+
e− e+
B B′
dX
Fig. 2.11. Propagation distance of shock (strong discontinuous) wave front in time dt is dX (= D dt).
Equations (2.57) and (2.58) are the dynamic compatibility conditions across a shock
wave front and an acceleration wave front, respectively, in a continuous medium. These
conditions reflect physically the condition of momentum conservation.
For a shock wave front, from the kinematic Eq. (2.55) and the dynamic Eq. (2.57) we can
eliminate the term [v], giving:
[σ ] = ρ0 D 2 [ε]
from which the shock wave velocity D is related to the stress jump [σ ] and the strain jump
[ε] across the wave front, as:

1 [σ ]
D= (2.59)
ρ0 [ε]
For an acceleration wave front, from the kinematic Eq. (2.56) and the dynamic Eq. (2.58)
we can eliminate the term [∂v/∂t], giving:

∂σ 2 ∂ε
= ρ0 D
∂X ∂X
from which the acceleration wave velocity D is related to the stress gradient jump [∂σ /∂X]
and the strain gradient jump [∂ε/∂X], by:

∂σ

1 ∂X
D = (2.60)
ρ0 ∂ε
∂X
Note that in the deductions of the kinematic and the dynamic compatibility equations,
the material properties are not involved. Therefore, these equations are true for the plane
waves in any continuum. Nevertheless, the determination of the wave speed D is related
to the constitutive relationship of the materials. In general, the relation between [σ ] and
[ε] and the relations between [∂σ /∂X] and [∂ε/∂X] are different, therefore the velocities
of a shock wave and an acceleration wave are different.
Under the framework of the rate-independent theory, we assume that there exists an
exclusive dynamic stress–strain relationship, namely σ = σ (ε). By this condition we
have:

∂σ dσ ∂ε
=
∂X dε ∂X
so that the Eq. (2.60) is reduced to (2.15). This means that the velocity of a con-
tinuous wave is determined by the tangent slope of the σ ∼ ε curve. This conclusion
has been obtained previously in elastic–plastic wave analysis. On the other side,
Eq. (2.59) means that the velocity of a shock wave is determined by the secant slope of
the σ ∼ ε curve, namely the slope of the straight-line linking the pre-shock state and
post-shock states. Similar to the shock wave discussions in fluid dynamics (Readers are
referred to Courant and Friedrichs, 1948), this secant line is called Rayleigh line or shock
secant.
Take the elastic-increasing hardening plastic material for example. The plastic shock wave
velocity is determined by the slope of Rayleigh line AB linking the yield point A (the initial
state prior to shock) and B (the final state after shock), as shown in Fig. 2.12. As long as this
slope is less than the elastic modulus, the plastic shock velocity Dp is always slower than
the elastic wave speed C0 , resulting in a double-wave structure in the media. Conversely,
once Dp ≥ C0 , a single elastic–plastic shock wave is created. If the stress and strain are
linearly related, [σ ]/[ε] = dσ /dε = const., and the shock wave speed is identical with
the continuous wave speed.
Now we discuss the energy conservation on a shock wave front. Consider the region
ABA B in Fig. 2.11, denoting the material’s internal energy in a unit mass by e, the energy
conservation of this region is:
1 − 2 + 2
σ + v+ − σ − v− A0 dt = e− − e+ ρ0 A0 dX + v − v ρ0 A0 dX
2
After simplification, the equation is:
1 !
[σ v] = −ρ0 D [e] − ρ0 D v2 (2.61)
2
which is the energy conservation equation across a shock wave front.
A F
Y
C
o ey e
D E
Fig. 2.12. For the “linear elastic-increasing hardening plastic” material, the shock wave velocity is
determined by the slope of the shock secant AB.
Introducing the material’s internal energy in a unit volume, E = ρ0 e, and substituting

(2.55) into (2.57), we notice that:
1 1 1
[σ v] + ρ0 D v2 = [σ v] + ρ0 D [v] v− + v+ = [σ v] − [σ ] v− + v+
2 2 2
1 " #
= σ − v− − σ + v+ − σ − v− + σ − v+ − σ + v− − σ + v+
2
1 " − # 1
= σ + σ + v− − σ − + σ + v+ = σ − + σ + [v]
2 2
D −
=− σ + σ + [ε]
2
after some mathematical manipulations we obtain another form of the energy conservation
condition:

E− −E+ = 1
2 σ − + σ + ε− − ε+ (2.62)
For the linear elastic, increasing-hardening plastic material shown in Fig. 2.12, after a
suddenly applied loading, two stress waves are generated: an elastic precursor wave fol-
lowed by a plastic shock wave. Assume that the bar is originally at rest and stress-free, its
initial state before the elastic precursor wave is represented by the point O in the figure.
After the elastic wave passes, the material’s state is represented by the dynamic yielding
point A, therefore:
σ − = Y, ε− = εy , E − = 12 Y εy
This means that across the wave front of the elastic precursor, the internal energy jump
[E] is represented by the area of the triangle OAC below elastic path OA. This quantity
is exactly equal to the elastic strain energy. The state prior to the plastic shock wave is at
point A, and the state after the shock is at point B. The location of B is determined by the
boundary condition of the sudden loading. The energy conservation Eq. (2.62) means that
across the shock front the internal energy jump [E ] is represented by the trapezoidal
$ ε area
ABEC below the shock secant AB. This area is bigger than the plastic strain energy εAB σ dε
by an area of region AFB (shaded in the figure). Expression of this part of energy is:
εB εB
1
∆Q = (σB + σA )(εB − εA ) − σ dε = EB − σ dε
2 εA εA
which corresponds to the thermal energy that has been dissipated during the formation
of the shock wave. Indeed the very large velocity gradient across a shock wave front
activates the internal friction of the solid – which has been neglected in a strain-rate
independent theory – and results in irreversible energy dissipation. Therefore, the state
change across a shock wave front by nature is an adiabatic process but not an isentropic
process. This thermodynamic process is frequently called the shock adiabatic process.
Certainly we should point out, the plastic deformation process itself is also an irreversible
process
$ εB and the plastic wave is thus not isentropic. Among the plastic deformation energy
0 σ dε, only the part equivalent to the triangular area BDE is elastically recoverable, and
the remains is transferred to the thermal dissipation during the plastic deformation. In case
of the plastic shock wave, an extra entropy-increase corresponding to ∆Q is produced.
Since the energy dissipation is always positive, there must be
∆Q ≥ 0
For the linear hardening material d 2 σ /dε 2 = 0, the extra energy dissipation ∆Q = 0
and the internal energy increase equals the plastic deformation energy. No extra entropy
increase is produced in a shock wave process for such materials.
Across a shock wave front, the mass conservation Eq. (2.55), the momentum conservation
Eq. (2.57), and the energy conservation Eq. (2.62) are collectively called the shock jump
condition or the Rankine–Hugoniot relationship. The equations listed here are Lagrange
forms for an elastic–plastic bar.
Given the initial state of the material (all quantities with superscript + are known), the
shock jump condition and the material’s constitutive relationship provide four equations
linking any two of the five unknown quantities σ − , ε− , v− , e− , and D . These equations
are called the shock adiabatic lines or the Hugoniot lines. They describe the loci of any
possible final equilibrium states arrived by shock jumping from the known initial state.
These relationships, nevertheless, do not represent all the state points which material
experiences in an actual shock process. For example, a shock adiabatic σ ∼ ε curve is not
material’s constitutive σ ∼ ε curve in adiabatic condition. Note that the rate-independent
theory assumes the existence of a unique dynamic constitutive σ ∼ ε curve, and therefore,
neglects the difference between a shock adiabatic σ ∼ ε relationship and a constitutive
σ ∼ ε relationship. This assumption made it possible for us to avoid using the energy
conservation equation, and from only the mass conservation Eq. (2.55), the momentum
conservation Eq. (2.57) and the material’s constitutive relationship can we determine the
quantities σ − , ε− , v− , and D with the given initial-boundary conditions. In a more strict
theory, the state equation of solid should be used instead of Eq. (2.14), and the energy
conservation equation should be included. We will discuss this topic further in Chapter 7
when discussing the one-dimensional plane strain shock wave problem (see Section 7.7).
As the jumps across a shock wave front decrease from finite values to infinitely small
values, the three conservation Eqs. (2.55), (2.57), and (2.62) become the conservative
conditions across a weak discontinuous wave front:
⎧
⎪
⎨dv = ∓Cdε (2.63a)
dσ = ∓ρ0 Cdv (2.63b)
⎪
⎩de = σ dε/ρ
0 (2.63c)
The minus (−) in the first two equations correspond to rightward wave, and the plus (+) to
leftward wave. These signs are opposite to that of characteristic compatibility relationships
(2.24) and (2.26). The reason is: Eq. (2.63) represents the relationships between state
variables ahead of and behind a wave front, namely the state variables across wave front;
conversely the characteristic compatibility equations represent the relationships between

state variables following the characteristic lines. As a disturbance propagates along a
rightward characteristic line, it is across a series of leftward characteristic lines namely
the leftward wave fronts, therefore the signs in two types of relationships differ by a
negative sign.
In studying the stress wave propagation, we can start from the governing equations, as has
been done in Sections 2.1–2.5; or we can work to establish the conservative conditions on
wave fronts, as the work in this section shows. These two approaches are interlinked and
interchangeable. In the subsequent discussions we may use either or both approaches.
2.8 Dispersion Effects Induced by the Transverse Inertia
In the previous discussions the cross section of the bar is assumed to be planar during
deformation, and the axial stress is assumed to be uniformly distributed on the cross
section. This approximate theory is called the primary theory or the engineering theory,
as it neglects the inertia effects of particle transversal movement, or the contribution of
the transversal expansions or contractions of the bar to the kinetic energy.
Within the range of elastic waves, we now discuss the effects of the transversal inertia.
This study will help us understand the limitations of the primary theory, and make clear
under what conditions such approximate theory is applicable.
Under a uni-axial stress σX (X, t), the axial strain of the bar is:
∂uX σX (X, t)
εX = =
∂X E
Besides this component, the bar also has transversal strain due to Poisson effect:
∂uY
εY = = −νεX (X, t)
∂Y
∂uZ
εZ = = −νεX (X, t)
∂Z
In the above equations, uX , uY , and uZ are the displacement components in X, Y , and Z

axes, respectively, ν is the Poisson ratio. Since the axial stress σX is a function of X and t,
so is the axial strain εX . Integrating the above equations lead to:
∂uX (X, t)
uY = −νYεX = −νY
∂X
∂uX (X, t)
uZ = −νZεX = −νZ (2.64)
∂X
where the center of the cross section is taken as the origin point of Y and Z axes. Therefore
the particle velocity vY and vZ , and the particle acceleration aY and aZ are respectively:
∂uY ∂εX ∂vX

vY = = −νY = −νY
∂t ∂t ∂X
∂uZ ∂εX ∂vX
vZ = = −νZ = −νZ
∂t ∂t ∂X
∂vY ∂ 2 εX ∂ 2 vX ∂aX
aY = = −νY 2
= −νY = −νY
∂t ∂t ∂t∂X ∂X
∂vZ ∂ 2 εX ∂ 2 vX ∂aX
aZ = = −νZ 2 = −νZ = −νZ (2.65)
∂t ∂t ∂t∂X ∂X
The above equations show that on a cross section there exist non-uniformly distributed
transversal particle displacements, velocities, and accelerations. Accordingly, there must
be non-uniformly distributed transversal stresses, resulting in the warping of the cross
section. Therefore attributed to the particle transversal movement of the bar, the stress
state of the bar is not uni-axial, and the cross section is no more planar. Strictly it should
be treated as a three-dimensional problem, or at least an axial-symmetric (for cylindrical
bars or tubes) two-dimensional problem.
From the energy viewpoint, neglecting the particle transversal inertia means that the kinetic
energy related to the particle transversal motion has been neglected. The average kinetic
energy density (energy per unit volume) can be calculated from Eq. (2.65) as:

1 1 2 1 ∂εX 2
ρ0 vY + vZ
2
dXdYdZ = ρ0 ν 2 rg2
A0 dX A0 2 2 ∂t
where rg is the rotational radius of the cross section with respect to X axis:

1
rg = Y 2 + Z 2 dYdZ
A0 A0
The energy conservation Eqs. (2.61) and (2.63c), deduced from the primary theory in the
previous section, account only the longitudinal kinetic energy and neglect this transversal
kinetic energy.
To investigate the effects of this energy, we consider again the bar segment shown in
Fig. 2.1. As the involved quantities are all those with respect to the X axis, we omit the
subscript X for convenience. The bar segment dX is subjected to a balanced force A0 σ
and an unbalanced force A0 (∂σ/∂X)dX. The latter is related to the longitudinal inertia of
the segment. In unit time the work done by this net force equals the increase rate of the
longitudinal kinetic energy of the bar:

∂σ ∂ 1
A0 dX · v = ρ0 A0 dX v2
∂X ∂t 2
The above equation, after rearrangement, leads to the motion equation of the bar
segment (2.13):
∂v ∂σ
ρ0 =
∂t ∂X
Therefore in primary theory, the work conducted by the balanced force is completely
converted into the internal energy of the system, i.e. the strain energy during elastic wave
motion [readers can refer to Eq. (2.63)]. In the current analysis the transversal motion
must be accounted. The work done by the balanced external force, therefore, is converted
into the longitudinal strain energy and the transversal kinetic energy. Approximately in
unit time and unit volume, the energy equilibrium equation is:
% 2 &
∂ε ∂ 1 2 ∂ 1 2 2 ∂ε
σ = Eε + ρ0 ν rg
∂t ∂t 2 ∂t 2 ∂t
from which we obtain:
∂ 2ε
σ = Eε + ρ0 ν 2 rg2 (2.66)
∂t 2
When neglecting the transversal kinetic energy, the above equation becomes the one-
dimensional Hooke’s law. Therefore, Eq. (2.66) can be viewed as a rate-dependent
stress–strain relationship instead of Hooke’s law when the bar’s transversal inertia is taken
into account. Since the transversal inertia term is proportional to the second-order time
derivative of strain, this term becomes important only when the second-order derivative
is significant.
By substituting (2.66) into the motion Eq. (2.13), we obtain the following wave equation
in terms of the displacement:
∂ 2u 4
2 2 ∂ u E ∂ 2u ∂ 2u
− ν rg 2 = = C0 (2.67)
∂t 2 ∂ X∂t 2 ρ0 ∂X 2 ∂X 2
Comparing this equation with Eq. (2.18) in elementary theory, we see that the only dif-
ference here is the second term in the left-hand side of equation. Because of this term,
the elastic longitudinal wave no longer propagates in the bar with a constant velocity C0 .
Instead, the harmonic waves with different frequency f (or wavelength λ) will propagate
with different wave velocity (phase velocity). Actually, substituting the harmonic wave
solution
u(X, t) = u0 exp{i(ωt − kt)}
into Eq. (2.67), we obtain:
ω2 + ν 2 rg2 ω2 k 2 = C02 k 2
where ω is the angular frequency (ω = 2πf ), and k is the wave number (k = 2π /λ).
From this equation, the phase velocity C (C = ω/k = f λ) of a harmonic wave with the
angular frequency ω is:
C02
C 2 = C02 − ν 2 rg2 ω2 = (2.68a)
1 + ν 2 rg2 k 2
In case ν 2 rg2 k 2 1, we have the following approximation:
C 1 2 r 2
g
≈ 1 − νrg k = 1 − 2ν 2 π 2 (2.68b)
C0 2 λ
√
For a solid cylindrical bar with radius a, rg = a/ 2, leading to:
C a 2
≈ 1 − ν2π 2 (2.68c)
C0 λ
The above equation is the so-called Rayleigh approximate solution that includes the
modification of the transversal inertia effect (Rayleigh, Lord, 1887). This approximate
solution gives quite accurate results in the range a/λ ≤ 0.7.
From the above equations, it can be seen that high frequency waves (short waves)
propagate slower, and low frequency waves (long waves) propagate faster. In case of
linear elastic waves, a wave with arbitrary waveform can be regarded as a superposi-
tion of a series of harmonic components by Fourier theorem. Since different components
propagate with different velocity, the waveform is likely to change during propagation.
In other words, the wave will disperse as it propagates. This dispersion phenomenon
originates from the transversal inertial effect. It differs from the nonlinear constitutive
dispersion discussed previously, which originates from the nonlinearity of the mate-
rial’s stress–strain relationship. On the other side, the current dispersion also differs from
the constitutive viscous dispersion (will be discussed in Chapter 6), which originates
from the viscous effect of the material. Because this dispersion is closely related to
the geometry of the bar (such as the bar’s radius), it is sometimes called the geometric
dispersion.
In experiments related to stress wave propagation, for example, in a Hopkinson pressure

bar test (will be discussed in Chapter 3), the measured stress waves more or less exhibit the
geometric dispersion. The local oscillations of a real stress wave shown in Fig. 2.13 resulted
from this dispersion effect. Specifically, a strong discontinuous stress wave contains many
high frequency harmonic components which are prone to dispersion. As a result, the
abrupt front of a strong discontinuous stress wave is no longer kept as a jump during wave
propagation. Actually most so-called shock waves propagating along a bar only have a
relatively steep wave front.
The recent two-dimensional (axi-symmetric) numerical simulations on the elastic wave

propagation in a cylindrical bar (e.g., Liu and Hu 2000a; Wang and Wang, 2005; etc.)
eX
Elementary theory
Experimental data
o
t
Fig. 2.13. A typical waveform recorded during a Hopkinson pressure bar test.
have exposed the following phenomena that are connected to the transversal inertial effect
of a one-dimensional bar:
1. The non-uniform stress distribution on cross section. Different from the elementary
theory that assumes uniform stress distribution on a bar cross section, the transversal
inertia effect introduces non-uniform, two-dimensional stress state. Consider a steel
cylindrical bar of diameter D = 2R = 37 mm, the material’s parameters are: Young’s
modulus E = 200 GPa, density ρ = 7.8 × 103 kg/m3 , Poisson ratio ν = 0.3.
At the end of the bar X = 0 a trapezoidal impulse is applied. The loading parameters
are: magnitude σ0 = 800 MPa, total loading time 120 µs including the rising and
decreasing time 10 µs respectively. Figure 2.14(a) plots the history curves of the
non-dimensional axial stresses at three points (r/R = 0, 0.5, and 1) on a cross
section X1 = 0.5D. Along the radial direction, the axial stress decreases from
center to surface. At the axial center of the bar the axial stress has the maximum
magnitude (the material is in an approximately one-dimensional strain state). At the
surface of the bar the axial stress has the minimum magnitude (the material in an
approximately one-dimensional stress state). The axial stress magnitude at location
0.5R is intermediate. Nevertheless, as the stress pulse propagates, the axial stress
distribution on a cross section gradually becomes uniform, as Fig. 2.14(b) shows,
though the stress histories still retain significant high frequency oscillations.
2. Stress wave oscillations. According to Eq. (2.68c), the phase velocity of a harmonic
wave is dependent on the ratio of the bar diameter to wavelength. As the bar diam-
eter increases, or the wavelength decreases, the waveform oscillations shown in
Fig. 2.13 become more significant. Figure 2.15(a–d) plot the stress wave profiles
on cross sections X = 0, 100, 200, 300, 400, and 500 mm of steel bars with
diameters 5, 14.5, 37, and 74 mm, respectively. The prescribed boundary condition
is same as that stated above.
As can be seen, for a given bar diameter the oscillations of a waveform increase as
the wave propagates; on the other hand the oscillations increase significantly as the
bar diameter increases. Indeed, the trapezoidal wave with 10 µs rising time prop-
agates along the 5 mm diameter bar with negligible oscillations, while the same
1.25
f 37 mm
r =0
1.00 r = 0.5R
r =R
0.75
s/s0
0.50
0.25
0.00
0 50 100 150 200
t (ms)
(a) The stress history curves at r = 0, 0.5R, and R on the cross section 0.5D from end.
1.25
f 37 mm
r =0
1.00 r = 0.5R
r =R
0.75
s/s0
0.50
0.25
0.00
0 50 100 150 200
t (ms)
(b) Stress histories at same radial locations on cross section 2D from end.
Fig. 2.14. Axial stress waveforms in different radial locations of a 37 mm diameter steel bar.
wave propagates along the 74 mm bar with significant distortions. Such waveform
oscillations may pose a severe problem to the experiments of split Hopkinson pres-
sure bar (will be discussed in Section 3.8). If a thick bar is used in the experiments,
adequate measures must be taken to improve the experimental accuracy.
3. Increases of wave front rising time. From Fig. 2.15 it is seen that due to the transversal
inertia effect, the waveform of a stress pulse flattens during propagation. One result
is that the rising time in the front of the stress pulse ts (the duration when stress pulse
increases from starting point to peak) increases with the propagation distance X.
This effect is more marked as the bar diameter increases. Figure 2.16 plots the
ts vs. X curves corresponding to different bar diameters. It is understandable that as
1000
800 1 − X= 0 cm
1 2 3 4 5 6 2 − X= 10 cm
3 − X= 20 cm
600 4 − X= 30 cm
5 − X= 40 cm
s (MPa)
6 − X= 50 cm
400
200
0 50 100 150 200 250

t (µs)
(a) Stress waveforms in 5 mm diameter bar.
1000
800
1 2 1 − X= 0 cm
3 4 5 6 2 − X= 10 cm
600 3 − X= 20 cm
4 − X= 30 cm
s (MPa)
5 − X= 40 cm
400 6 − X= 50 cm
200
0 50 100 150 200 250

t (µs)
(b) Stress waveforms in 14.5 mm diameter bar.
Fig. 2.15. Stress waveforms propagate in bars of different diameters.
a thicker bar has more transversal inertia, the rising time increase in a thick bar is
more significant than that in a thin bar. The rising time increase is larger in the early
propagation stage, and tends to level up as time goes on.
4. Attenuations of stress peaks during propagation. Another phenomenon of the geo-
metric dispersion is: the peak level of a stress pulse decreases during its propagation.
1000
1 − X = 0 cm
800 2 − X = 10 cm
1
2 3 − X = 20 cm
3 4 5 6 4 − X = 30 cm
600
5 − X = 40 cm
6 − X = 50 cm
s (MPa)
400
200
−200
0 50 100 150 200 250
t (µs)
(c) Stress waveforms in 37 mm diameter bar.
1000
1 − X = 0 cm
800 1 2 − X = 10 cm
2 3 3 − X = 20 cm
4 5
600 6 4 − X = 30 cm
s (MPa)
5 − X = 40 cm
400 6 − X = 50 cm
200
−200
0 50 100 150 200 250 300
t (µs)
(d) Stress waveforms in 74 mm diameter bar.
Fig. 2.15. Continued.
As a trapezoidal stress pulse results in wave oscillations so that the attenuation of

wave peak is difficult to quantify, we consider the propagation process of a triangular
stress pulse to illustrate this peak attenuation phenomenon. The triangular stress pulse
applied at the bar end has a peak magnitude σ 0 = 800 MPa. The rising time and the
descending time are 150 µs. Figure 2.17 plots the waveforms of the stress pulse after
propagating different distances along a 37 mm diameter bar. Obvious peak stress
70 f 100 mm
60
f 74 mm
50
ts (ms) 40
f 37 mm
30
20 f 14.5 mm
f 5 mm
10
0 50 100 150 200
X (cm)
Fig. 2.16. The rising times of stress pulse front (ts ) as functions of propagation distance (X). Results
corresponding to different bars.
800
600 1 − X = 0 cm
2 − X = 50 cm
3 − X = 100 cm
s (MPa)
400 1 2 3 4 5
4 − X = 150 cm
5 − X = 200 cm
200
−100
0 100 200 300 400 500 600
t (ms)
Fig. 2.17. Waveforms of triangular stress pulse propagating in a 37 mm diameter steel bar.
800
790
f 37 mm
s (MPa)
780
770 f 74 mm
f 100 mm
760
0 50 100 150 200 250 300
X (cm)
Fig. 2.18. Attenuations of peaks of a triangular stress pulse propagating in different bars.
attenuation phenomenon is seen. Figure 2.18 plots the peak stress values as func-
tions of propagation distance X in 37, 74, and 100 mm diameter bars. Obviously a
larger bar diameter results in more significant attenuation. This is consistent with the
fact that a thicker bar has more marked effect of transversal inertia.
In summarization, if the transversal dimension of the bar is much less than the wave-
length of a stress pulse, then the transversal kinetic energy of the bar is much small
than the longitudinal kinetic energy. In this case the elementary one-dimensional
stress wave theory provides sufficiently accurate results. If the bar’s transversal
dimension is comparable with the wavelength, geometric dispersions resulting from
transversal inertia is should be taken into account. Further study shows, for stress
waves within the range a/l ≤ 0.7, the Rayleigh approximate solution (2.68) gives
quite good transversal inertia modification. However, for the waves that have even
shorter wavelengths, the more exact but complicated Pochhammer–Chree solutions
should be taken into consideration (the readers can refer the books of Kolsky, 1953
or Miklowitz, 1978).
2.9 Torsion Waves in Cylindrical Bars
The discussions up to now are for uni-axial longitudinal stress waves. In concluding this
chapter we briefly discuss the propagation of a simple kind of transverse wave, namely
the elastic torsion wave.
Lagrange coordinate system is used in the current analysis. Under the planar cross section
assumption1 , we investigate the pure torsion motion of a cylinder with uniform cross
section. The cylinder is subjected to a torque M, and twists an angle ϕ (Fig. 2.19). The
angular velocity and the twist per unit length, denoted by ω and θ respectively, are the
first-order derivatives of ϕ:
∂ϕ
ω=
∂t
∂ϕ
θ= (2.69)
∂X
M w M+ M dX
X
j
j+ j dX X
X
dX
Fig. 2.19. Pure torsion movement of a cylindrical bar with uniform cross sections.
1According to the elasticity theory, the planar cross-section assumption is true only to the cylinder bar or tube
with the circular cross sections. For the bar with other shape of cross section, the cross section warps when the
bar is twisted.
The geometric compatibility equation between ω and θ , and the conservation equation of
the angular momentum in a segment dX are:
∂ω ∂θ
= (2.70)
∂X ∂t
∂ω ∂M
I = (2.71)
∂t ∂X
where I is the rotational inertia of a unit length cylinder with respect to its axis X:

I= r 2 dm = ρ0 r 2 dA (2.72)
and m is the line density (dm = ρ0 dA). In case the material is linear elastic, the
relationship between the shear stress τ and shear strain γ is:
τ = Gγ (2.73)
where G is the shear modulus. As the cross sections of the cylinder is not deformed during
twist, we have:
γ = rθ (2.74)

M= τ rdA (2.75)
where r is distance of a point on cross section to the twisting axis and A is the cross
sectional area of the cylinder. Substituting (2.73, 2.74) into (2.75) and considering (2.72)
gives:

GIθ
M= τ rds = Gθ r 2 ds = (2.76)
ρ0
Substituting (2.76) into the angular momentum conservation Eq. (2.71) leads to:
∂ω ∂θ
= CT2 (2.77)
∂t ∂X
where we have introduced a quantity:

G
CT = (2.78)
ρ0
Equations (2.70) and (2.77) constitute a hyperbolic first-order partial differential system
having the unknown functions of the angular velocity ω and the relative angular twist θ .
Formally, these equations are identical to the longitudinal wave Eqs. (2.12) and (2.16).
Substituting (2.69) into (2.77) gives a second-order hyperbolic PDE having the unknown
function of the twist angle ϕ:
∂ 2ϕ 2
2∂ ϕ
− C T =0 (2.79)
∂t 2 ∂X 2
Obviously the quantity CT is the wave velocity of the elastic torsion wave.
As G = E/[2 (1 + ν)], and the value of Poisson ratio ν is generally between 0 and
0.5, the torsion wave
√ velocity is less than √
the longitudinal
√ wave velocity in the bar. The
ratio of C0 /CT = 2(1 + ν) is between 2 and 3. Table 2.2 collects the CT values
of some common materials (Kolsky, 1953). These data can be compared to those listed in
Table 2.1.
In case of a thin-walled-thickness cylindrical tube with average radius a, the shear stress
is uniformly distributed on the tube’s cross section. Equations (2.70) and (2.77) can be
rewritten as:
∂(aω) 1 ∂τ
=
∂X ρ0 CT2 ∂t
∂(aω) 1 ∂τ
= (2.80)
∂t ρ0 ∂X
which in form are identical to the longitudinal wave Eqs. (2.17) and (2.13). The strategy
of using the method of characteristics to solve these equations is similar to that discussed
previously in Eqs. (2.24) and (2.26). It is easy to prove that along the characteristic lines:
dX = ±CT dt (2.81a)
the angular velocity ω and the relative twist angle θ satisfy the compatibility equation:
dω = ±CT dθ (2.81b)
or represented by the shear stress τ and the tangent twist speed (aω), the characteristic
compatibility equation is:
dτ = ±ρ0 CT d(aω) (2.81c)
Table 2.2. Shear modulus and the elastic torsion wave velocity of common
materials.
G (GPa = 1010 dyne/cm2 ) 81 45 26 28 7.0×10−4
CT (km/s) 3.22 2.25 3.10 3.35 0.027
For a simple torsion wave propagating in a tube that is initially at rest and stress-free,
behind the wave front the following simple torsion wave relationships exist:
ω = ∓CT θ
τ = ∓ρ0 CT aω
where (−) corresponds to the rightward wave and (+) corresponds to the leftward wave.
Note that when a tube is subjected to pure torsion deformation, the direction that inclines
to the tube axis X by 45◦ , namely the X direction shown in Fig. 2.20, is the principle
stress direction. Therefore under pure torsion loading, the spiral fiber that is 45◦ inclined
to the tube axis is under a tensile stress of value τ in X direction, and a compressive
stress of value −τ in the direction normal to X . As Johnson (1972) pointed out, a torsion
wave propagating with the velocity CT along the tube axis (the X axis) can be viewed as
a longitudinal wave propagating with the velocity C0 along the 45◦ spiral fiber direction
(the X axis), accordingly:
C
CT = C0 cos 45◦ = √0 (2.82)
2
In X direction, the stress–strain relationship accounting for the transversal compression

effect is:
1 1+ν
εX = (τ + ντ ) = τ
E E
from which the longitudinal wave velocity, expressed by Eq. (2.15), is deduced as:

E
C0 =
ρ0 (1 + ν)
X
t
−t
t
X′
45°
−t
t
−t
t
45°
X
Fig. 2.20. Propagation of torsion waves along the axis of a thin elastic tube is equivalent to that of
longitudinal waves along the spiral fibers on the tube surface 45◦ inclined to the tube axis.
Substituting the above result into (2.82), and noticing G = E/2(1 + ν), the torsion wave
velocity CT is:

C E G
CT = √0 = =
2 2ρ0 (1 + ν) ρ0
This expression is exactly identical with Eq. (2.78).
It is worthwhile to note that although the theory of the longitudinal wave propagation in
a bar, based on invariable plane cross-section assumption, is an approximate theory; the
torsion wave theory based on the same assumption leads solutions that are identical with
Pochhammer’s elastodynamic exact solution (Kolsky, 1953). All harmonic √ torsion waves
of different frequencies propagate with the same phase velocity CT = G/ρ0 , and no
frequency dispersion exists during torsion wave propagation.
As a matter of fact, the elastic torsion wave velocity CT is the isometric wave velocity,
or the distortional wave velocity, in an infinite media. Sometimes this velocity is called
the shear wave velocity. In Chapter 11 we will further discuss such wave velocity.
CHAPTER 3
Interaction of Elastic Longitudinal Waves
3.1 Coaxial Collision of Two Elastic Bars
Before discussing the interaction between two elastic waves, we first discuss the longi-
tudinal coaxial collision of two elastic bars. This is helpful to understand the interaction
between two elastic waves.
Consider two elastic bars B1 and B2 with the same section sizes but different sound
impedance (ρ0 C0 )l and (ρ0 C0 )2 , respectively. Before their collision, the initial stresses in
both bars is zero, while the initial particle velocities are v1 and v2 respectively, and assum-
ing v1 < v2 , as schematically shown in Fig. 3.1(a). On the σ –v plane, such initial states
correspond to the points 1 and 2, respectively as shown in Fig. 3.1(d). When a longitudinal
coaxial collision takes place, originating from the collision interface, a rightward strong
discontinuous elastic wave propagates in bar B1 and simultaneously a leftward strong
discontinuous elastic wave propagates in bar B2 as shown in Fig. 3.1(c). At the colli-
sion interface, the particle velocities of the two bars should be identical (the continuity
condition), as well as the stresses of the two bars should be identical (the action force
and reaction force equality condition). Based on these, and moreover according to the
momentum conservation condition across a strong discontinuous interface [Eq. (2.57)],
we obtain
σ = −(ρ0 C0 )1 (v − v1 ) = (ρ0 C0 )2 (v − v2 )
From this, the particle velocity v and the stress σ in the two bars after collision can be
determined respectively:
⎫
(ρ0 C0 )1 v1 + (ρ0 C0 )2 v2 ⎪
v = ⎪
(ρ0 C0 )1 + (ρ0 C0 )2 ⎪ ⎪
⎪
⎬
v2 − v 1 (3.1)
=− ⎪
⎪
σ ⎪
⎪
1
+
1 ⎪
⎭
(ρ0 C0 )1 (ρ0 C0 )2
57
v2 v1 < v2
B2 B1
L2 L1
(a)
v2 v v v1
(C0)2 (C0)1
B2 B1
(b)
t s
v1 v v2
o 1 2v
X
) 1t
=
(C 0
(C 0
=
) 2t
X
σ
3
o X
(c) (d)
Fig. 3.1. Longitudinal coaxial collision of two elastic bars.
On the σ –v plane, this state corresponds to the intersection point 3 of the leftward char-
acteristics passing point 1 with slope −(ρ0 C0 )1 and the rightward characteristics passing
point 2 with slope (ρ0 C0 )2 as shown in Fig. 3.1(d).
If the material of the two bars are identical, or the sound impedance are identical, namely
if (ρ0 C0 )1 = (ρ0 C0 )2 = ρ0 C0 , then v = (1/2)(v1 + v2 ) while σ = −(1/2)ρ0 C0 (v2 − v1 ).
Further if v1 = v2 , then v = 0 and σ = ρ0 C0 v2 . This is equivalent to such a situation
where bar B2 impacts onto a rigid wall at a velocity v2 . This result coincides with that of
a rigid wall condition where (ρ0 C0 )1 → ∞ and v1 = 0 are applied. On the other hand,
If (ρ0 C0 )2 → ∞, then v → v2 and σ → −(ρ0 C0 )1 (v2 − v1 ), which corresponds to such
a situation that a rigid bar B2 impacts onto an elastic bar B1 . The above results are valid
from the beginning of collision up to the time when the stress waves arrive at the other
end of the bars. Once the stress waves arrive at the other end of the bars, the situation will
be different since the wave reflection occurs (refer to Section 3.4).
3.2 Interaction of Two Elastic Longitudinal Waves
Consider an elastic bar which is initially at rest and in natural state and is suddenly sub-
jected to a constant impact load at both ends as shown in Fig. 3.2(a). Then, originating
from the two ends, two strong discontinuous elastic waves propagate toward each other
(called the primary waves), which are represented by the rightward characteristics OA and
the leftward characteristics LA on the X–t plane as shown in Fig. 3.2(e). The bar part swept
by the leftward primary wave front (namely across the characteristics LA) is in the state of
vl , εl , σ1 . According to Eq. (2.57) and reversing the sign for leftward waves, there exists
Interaction of Elastic Longitudinal Waves 59
s2
s1
v2 v1 C0 C0
X
o L (b)
(a)
s2 s3
C0 C0
s1 s2
s1
(c)
(d )
t s
D 3 B 3 s3
2 A 1 2 s2
s1 1
0
o L X v2 v3 o v1 v
(e)
(f )
Fig. 3.2. Interaction of two head-on propagating strong discontinuous elastic waves.
the relation σ1 = ρ0 C0 v1 . The bar part swept by the rightward primary wave (namely
across the characteristics OA) is in the state of v2 , ε2 , σ2 . According to Eq. (2.57) there
exists the relation σ2 = −ρ0 C0 v2 . On the σ –v plane, these correspond to the state-jump
from point 0 to point 1 and point 2, respectively as shown in Fig. 3.2(f). All these are
entirely the same as what we discussed for the elastic waves in semi-infinite bars. At the
instant that two waves meet at an interface as shown in Fig. 3.2(c), the particle velocity of
the bar on the right side of the interface is v1 , while the particle velocity of the bar on the
left side of the interface is v2 , just similar to a coaxial collision of two elastic bars shown
in Fig. 3.1. So an interaction of two waves in a bar is equivalent to an internal collision
within the bar, sometimes called the internal collision. As a result of such an internal
collision, from the interface where the two primary waves meet, two internal reflective
waves are generated, namely the rightward secondary wave AB and the leftward sec-
ondary wave AD, and propagate towards the two ends of the bar respectively, as shown in
Fig. 3.2(d and e).
The state of the bar swept by the rightward secondary wave front (namely across the char-
acteristics AB) changes suddenly from state v1 , ε1 , σ1 to the state v3 , ε3 , σ3 , and according
to the compatibility condition [Eq. (2.57)] there is,
σ3 − σ1 = −ρ0 C0 (v3 − v1 )
On the σ –v plane, this corresponds to a state-jump from point 1 to point 3 in the direction
coincident to the direction across the characteristics AB on the X–t plane.
The state of the bar swept by the leftward secondary wave front changes suddenly from
state v2 , ε2 , σ2 , to state v3 , ε3 , σ3 , and according to Eq. (2.57), there is,
σ3 − σ2 = ρ0 C0 (v3 − v2 )
On the σ –v plane, this corresponds to a state-jump from point 2 to point 3 in the direction
coincident to the direction across the characteristics AD on the X–t plane. But, according to
the particle-velocity continuity condition and the stress equality condition at the internal
collision interface, there should be v3 = v3 = v3 , and σ3 = σ3 = σ3 . On the σ –v
plane, this corresponds to the determination of the intersection point 3 of the leftward
characteristics passing the point 1 and the rightward characteristics passing the point 2.
Thus the problem is solvable. It should be noticed that the Eq. (2.57) is deduced for the
rightward waves, so it is needed to replace D with –D (reversing sign) for the leftward
waves, and the initial states ahead of the primary wave front and the secondary wave front
are different. Thereby, we obtain
σ3 = −ρ0 C0 v2 +ρ0 C0 (v3 −v2 ) = ρ0 C0 v1 −ρ0 C0 (v3 −v1 )

(rightward primary)+(leftward secondary) = (leftward primary)+(rightward secondary)
* +, - * +, -
left part of bar right part of bar
Then, the particle velocity v3 and the stress σ3 in the bar experiencing the interaction of
two elastic waves can be respectively determined as follows [Fig. 3.2(f)]:

v3 = v1 + v2
(3.2)
σ3 = σ1 + σ2
It is clear that the final state of the bar experiencing the interaction of two elastic waves
can be obtained by a superposition of the results respectively induced by each elastic wave
propagation. This is because the governing equations of elastic waves are linear so that
the principle of superposition must be valid.
Bear in mind, the basic principles and methods to treat the interaction of stress waves used
here will be repeatedly used in the following sections. For whatever be the type of waves
like strong discontinuous waves or the weak discontinuous waves, the loading waves or the
unloading waves, the elastic waves, the elasto-plastic wave or the hydrodynamic waves,
these principles are still valid.
3.3 Reflection of Elastic Longitudinal Waves at Fixed End and Free End
When a stress wave propagating in a bar arrives at the other end of the bar, a wave
reflection occurs. The reflective wave varies depending on the actual boundary condition.
With respect to the incident wave, the boundary condition is a new disturbance to the state
behind the incident wave front, and the propagation of this new disturbance is displayed
as a reflective wave. The state behind a reflective wave depends on the total result of the
incident wave together with the reflective wave, satisfying the given boundary condition.
For the elastic waves, the total result of the incident wave together with the reflective wave
s s
3 3
o v
2 1 2
o v
(a) (b)
Fig. 3.3. Reflection of an elastic wave at a fixed end.
can be determined based on the principle of superposition. Thereby, the reflection of an

elastic wave at the fixed end and at the free end can be viewed as special examples of
the general discussions on the interaction of two elastic waves conducted in the previous
section.
As a special case of that shown in Fig. 3.2, let v2 = −v1 , then we have v3 = 0 and σ3 = 2σ1
as shown in Fig. 3.3(a), namely the final particle velocity is zero but the final stress is
doubled. This satisfies the boundary condition of particle velocity at a fixed end. Thus,
this special case corresponds to the reflection at a fixed end (a rigid wall) of a normally
incident elastic wave. Therefore, if the bar end is imagined as a mirror, when a normally
incident elastic wave arrives at a fixed end, the reflective wave is just a positive image of
the incident wave.
Based on the reflection of an elastic wave at a fixed end, the well-known drop-weight tests
for the impact tension of steel string performed by Hopkinson (1872) can be explained.
From the Hopkinson’s tests as schematically shown in Fig. 3.4, it was shown that when
a steel string is subjected to an impact tension by a dropping weight, it breaks not at
the loaded end A but at the fixed end B, and moreover whether the string breaks or not
is mainly dependent on the height of the weight (or the impact velocity), but almost
independent of the mass of weight. This is because that for a given material (namely the
sound impedance ρ0 C0 is known), the amplitude of the incidence stress wave generated
B
string
weight
Fig. 3.4. Drop tensile test.

at point A only depends on the impact velocity v0 regardless of the mass of weight,
σ = ρ0 C0 v0 [see Eq. (2.33)], and the stress at the fixed end B reaches two times of the
incidence stress σ due to the wave reflection at a fixed end, 2σ = 2ρ0 C0 v0 . A further
analysis shows that the dropping mass will influence the profile of the incident wave,
which will be discussed in the next section, refer to Eq. (3.5).
As another special case of that shown in Fig. 3.2, let v2 = v1 , then we have v3 = 2v1 , σ3 = 0
as shown in Fig. 3.3(b), namely the final stress is zero while the particle velocity is
doubled. This satisfies the boundary condition of stress at a free end. Thus, this special
case corresponds to the reflection at a free end (free surface) of a normal incident
elastic wave. Therefore, if the bar end is imagined as a mirror, when a normally incident
elastic wave is reflected at a free end, the reflective wave is just an inverted image of the
incident wave.
3.4 Coaxial Collision of Two Elastic Bars with Finite Length
In the previous discussion for the coaxial collision of two elastic bars shown in Fig. 3.1,
we temporarily only focused on the primary elastic waves generated by the collision, and
neither the wave reflection nor the wave interaction was involved. Having the knowledge
of the reflection of elastic waves at the fixed end and the free end discussed above, now
we can further discuss the coaxial collision of two elastic bars with finite lengths.
Assume that in Fig. 3.1 the length of bar B2 is L2 and of bar B1 is L1 , and L2 < L1 , namely,
we consider a problem where a short bar B2 impacts onto a long bar B1 . For convenience, let
the long bar B1 be initially at rest, namely v1 = 0. The primary elastic waves generated at
the beginning of collision (t = 0) can be computed according to Eq. (3.1). Subsequently,
at time t = L2 /(C0 )2 , the leftward wave is first reflected at the free end, and at time
t = 2L2 /(C0 )2 , this rightward reflected wave returns to the collision interface. The
subsequent situations depend on the ratio of wave impedances (sound impedances) of
these two bars, and can be categorized into three cases:
(a) (ρ0 C0 )2 /(ρ0 C0 )1 = 1 [Fig. 3.5(a)].

In this case, since wave impedances of the two bars are identical, the rightward
unloading wave reflected from the free end propagates through the collision interface
without any additional disturbance, just as propagating in the same bar, although the
wave velocities in the two bar can be different, namely there may be (C0 )2
= (C0 )1 .
Therefore, at time t = 2L2 /(C0 )2 , the contact interface of two bars is unloaded to
v = 0 and σ = 0. This indicates the termination of collision (no interaction between
the two bars). After this, a stress pulse consisting of a strong discontinuous loading
wave front and a strong discontinuous unloading wave front still propagates in the
long bar. In the case of (C0 )2 = (C0 )1 , the pulse length λ is two times the length
of the short bar, namely λ = 2L2 . By changing the length of the short bar, we can
obtain stress pulses with various pulse lengths. This is a typical method to control
the length of stress pulses in stress wave experiments.
In the special case of L2 /(C0 )2 = L1 /(C0 )1 , the primary elastic waves generated
by the collision simultaneously arrive at each free end, and the reflected elastic
s
t
0
0
2
3 3 0 v
2 0
B2 B1 X 3
L2 0
(a)
s
t
0
4 2
4 0 v
3 3
2 0
B2 B1 X 3
L2 0
(b)
t
s
7
7
5 6 4 2
5 0 v
4 7
5
3 3
2 3
B2 B1 X
L2 0
(c)
Fig. 3.5. Coaxial collision between a short elastic bar and a long elastic bar.
unloading waves also simultaneously return to the collision interface. At time

t = 2L2 /(C0 )2 , the collision is finished, the whole bar B2 is at rest (v = 0 and
σ = 0), and bar B1 is moving as a whole (v = v2 and σ = 0). We advise the
readers to plot these states simultaneously on the X–t plane and the σ –v plane by
themselves. Note that before the collision the particle velocity of bar B2 is v2 while
that of bar B1 is 0, but after the collision the particle velocity of bar B2 is 0 while
that of bar B1 is v2 . It is clear that through the collision, the momentum and kinetic
energy of bar B2 is entirely transferred to bar B1 (readers can compare this process
with the elastic collision of two homogenous balls with the same mass in the text-
book of physics). It should be noticed that in the case of L2 < L1 , although the
whole bar B2 is at rest, bar B1 is not in a state like a rigid body motion (namely, all
its particles move at a same velocity), but in a state of back-and-forth propagating
stress waves.
(b) (ρ0 C0 )2 /(ρ0 C0 )1 < 1 [Fig. 3.5(b)]

In this case, when the leftward incident wave in the short bar B2 is reflected at the
free end, the stress decreases to zero and the particle velocity changes to negative,
as shown by point 4 on the σ –v plane in Fig. 3.5(b), since the increment of particle
velocity (v4 − v2 ) of a reflective wave at a free end is twice the increment of particle
velocity (v3 − v2 ) of an incident wave. When this reflective unloading wave returns
to the collision interface, namely at time t = 2L2 /(C0 )2 , the whole short bar B2 is
at the state of v = v4 (negative!) and σ = 0, while for the long bar B1 both the stress
and the particle velocity at the contact interface are unloaded to zero. At this time,
the velocities of the two bars are different. As v4 < 0, the short bar B2 rebounds
at velocity v4 and is no longer in contact with the long bar B1 . Thus, the collision
is finished. After that, a stress pulse with a length of λ = 2L2 (C0 )1 /(C0 )2 still
propagates in the long bar B1 .
In the special case of L2 /(C0 )2 = L1 /(C0 )1 , at time t = 2L2 /(C0 )2 , the whole
bar B2 rebounds at velocity v4 , and the whole bar B1 moves at velocity 2v3 . At the
end of collision, only a part of the initial momentum and kinetic energy of bar B2 is
transferred to bar B1 . The readers are suggested to scheme the corresponding figure
on the X–t plane and the σ –v plane on their own.
If (ρ0 C0 )1 → ∞, this corresponds to an elastic short bar impacting onto a rigid
target. Obviously, at time t = 2L2 /(C0 )2 , the short bar rebounds at a velocity, the
absolute value of which is the same as the impact velocity v2 but in the opposite
direction (v4 = −v2 ). The readers are suggested to plot the corresponding figure on
the X–t plane and the σ –v plane themselves.
(c) (ρ0 C0 )2 /(ρ0 C0 )1 > 1 [Fig. 3.5(c)]
In this case, when the leftward primary incident wave in the short bar B2 is reflected
at the free end, the stress decreases to zero, and the particle velocity also decreases,
though it is still positive [v4 − v2 = 2(v3 − v2 )], as shown by point 4 on the σ –v
plane in Fig. 3.5(c). When this reflected unloading wave returns to the collision
interface, namely at time t = 2L2 /(C0 )2 , the whole short bar B2 is in the state of
v = v4 (positive) and σ = 0. However, for the long bar B1 , if the stress at the
collision interface decreases to zero, then the particle velocity should also decrease
to zero. Since now v4 > 0, it means that the collision has not yet ended, and now it
is equivalent to such a situation where the bar B2 moving with velocity v4 impacts
onto the bar B1 at rest, and then the secondary waves are generated. The state of such
secondary waves is represented by point 5 on the σ –v plane. Based on this analogy,
before the unloading wave reflected from the free end of the long bar returns to
the collision interface, the collision never ends, but the amplitude of the incident
waves propagating in the bar B1 decreases successively. The readers are suggested
to plot by themselves, the stress wave profile on the σ –X plane for the long bar B1
at a time t (after the wave in the short bar has been reflected back-and-forth several
times) and discuss the relationship between such step-shaped wave profile and the
short bar length L2 .
In the special case of L2 /(C0 )2 = L1 /(C0 )1 , at time t = 2L2 /(C0 )2 , bar B2 moves
as a whole at velocity v4 (0 < v4 < v2 ), while the bar B1 moves as a whole at a much
larger velocity 2v3 (>v2 ), then the collision is finished. After the end of collision,
only a part of the initial momentum and kinetic energy of bar B2 is transferred to
the bar B1 . The readers are suggested to plot the corresponding figure on the X–t
plane and the σ –v plane themselves.
If (ρ0 C0 )2 → ∞, this corresponds to a situation where a rigid body impacts onto
an elastic bar. If the mass of the rigid body is denoted by M and the reflection at the
other end of the elastic bar is disregarded, then according to the particle velocity
continuity condition and the stress equality condition at the collision interface, we
can write the following formulas for the elastic bar and the rigid body respectively
σ = −ρ0 C0 v, (3.3)
M dv
σ = (3.4)
A dt
where A is the section area of the elastic bar. From these equations, we obtain

ρ 0 C0 A Mt
v = v0 exp − t = v0 exp − (3.5a)
M M
where v0 is the initial impact velocity, Mt (= ρ0 C0 At) denotes the mass in the bar
swept by the elastic wave front at time t. This formula indicates that the particle
velocity profile is displayed as a strong discontinuous wave front followed by an
exponentially attenuating wave tail.
If we introduce the initial impact stress σ0 = −ρ0 C0 v0 according to Eq. (3.3), then
corresponding to Eq. (3.5a) we obtain

ρ 0 C0 A Mt
σ = σ0 exp − t = σ0 exp − (3.5b)
M M
which indicates that stress σ attenuates with t exponentially. The readers are sug-
gested to compare this result with the time-histories profile σ –t at a certain X of the
long bar in Fig. 3.5(c).
If Mt M, then v → v0 , then it is reduced to an approximate constant velocity collision.

Recalling the Hopkinson’s impact tension tests on steel strings (Fig. 3.4), obviously,
a more complete analysis is presented here.
From the above discussion, we can further understand that, unlike the static theory of solid
mechanics, the theory of stress waves focuses on the unsteady and non-uniform motion
of media varying with time and coordinates. We can also further understand the cou-
pling between loads and media and the important influence of the mechanical constitutive
behavior of materials on the propagation of stress waves.
Although in the above discussions, the collisions of only two elastic bars were taken as
examples, the principles mentioned are also suitable to analyze the coaxial collision of
three or more elastic bars.
For example, consider three identical elastic bars B1 , B2 , and B3 , the bars B2 and B1 are
originally at rest and in contact with each other, while bar B3 moves at a velocity v0 and
t s
v0
1 0
1 0 v
2 0
2
2 2
0 0 1
−2L −L 0 L X
B1 B2 B3
v=0 v=0 v = v0
Fig. 3.6. Coaxial collision of three identical bars.
impacts onto the bar B2 as shown in Fig. 3.6. It is easy to prove that the collision ends at
time t = 3L/C0 , and at this time bar B1 flies out as a whole at a velocity v0 , while both
the bars B2 and B3 are now at rest. The initial momentum and kinetic energy of bar B3 are
entirely transferred to bar B1 through bar B2 .
3.5 Reflection and Transmission of Elastic Longitudinal Waves at the Interface

of Two Different Bars
In the previous discussion on the collision of two finite length bars with different wave
impedances, when the reflected wave in the short bar returns from the free end to the
collision interface, we have actually dealt with the transmission and reflection of elastic
wave at the interface between two different materials. In the following, we will discuss
the general cases of this sort.
Consider a situation when an elastic wave propagates from one medium (the related param-
eters are denoted by the subscript 1) to another (the related parameters are denoted by the
subscript 2) and the propagating direction of the wave is perpendicular to the interface of
the two media, namely a normal incidence. When an incident elastic wave disturbance
∆σI arrives at the interface, new disturbances not only with respect to the first medium
but also to the second medium, are generated, namely a reflective disturbance ∆σR and a
transmission disturbance ∆σT will propagate into the two media respectively, as shown
in Fig. 3.7. So long as the two media experiencing the reflection–transmission are still
in contact at their interface (namely they are not separated and can undergo both com-
pression and tension), according to the continuity condition and the Newton’s third law,
the particle velocities on both sides and the stresses on both sides should be identical
respectively.
∆vI + ∆vR = ∆vT (3.6)

∆σI + ∆σR = ∆σT (3.7)
where the subscripts I, R, and T denote the parameters relevant to the incident wave,
reflective wave, and transmission wave.
R 2 2 T
1
I
0 0
0 interface X (r0C0)1 < (r0C0)2
s σ
(C0)1 2
1
(C0)2
0 X 0 v
(a)
R 2 2
1
I
0 0
0 interface X (r0C0)1 > (r0C0)2
s σ
1
(C0)1
(C0)2 2
0 X 0 v
(b)
Fig. 3.7. Wave reflection and transmission in bars of different materials.
Based on the momentum conservation condition across wave front [Eq. (2.57)], Eq. (3.6)
can be written as
∆σI ∆σR ∆σT

− = (3.8)
(ρ0 C0 )1 (ρ0 C0 )1 (ρ0 C0 )2
Solving the simultaneous Eqs. (3.8) and (3.7), we obtain

∆σR = F · ∆σI
(3.9)
∆vR = −F · ∆vI

∆σT = T · ∆σI
(3.10)
∆vT = nT · ∆vI
where
⎫
(ρ0 C0 )1 ⎪
n= ⎪
(ρ0 C0 )2 ⎪
⎪
⎪
⎪
⎬
1−n
F= (3.11)
1+n ⎪ ⎪
⎪
⎪
⎪
2 ⎪
⎭
T=
1+n
where n is the ratio of wave impedances of two media, F and T are called the reflection
coefficient and transmission coefficient respectively, which depend on the ratio of wave
impedances n. It is clear that
1+F =T
On the σ –v plane, the leftward σ –v characteristics 0–1 passing point 0 in the first media
corresponds to the incident disturbance ∆σI ; the rightward σ –v characteristics 1–2 passing
point 1 in the first medium corresponds to the reflection disturbance ∆σR ; while the
leftward σ –v characteristics 0–2 passing point 0 in the second medium corresponds to the
transmission disturbance ∆σT . The above approach shows how to determine the state of
the interface after reflection and transmission, namely σ2 and v2 , from the intersection
point 2 of the characteristics 1–2 and 0–2, as shown in Fig. 3.7. It should be emphasized
that, either from the angle of concept or in the actual analyses of wave reflection and
transmission, the reflection disturbance ∆σR itself should not be confused with the stress
state σ2 after the reflection.
It should be noted that T is always positive and consequently the sign of the transmission
wave is always the same as that of the incident wave, while the sign of F depends on the
relative value of the wave impedances of two media. There are two different cases which
will be discussed below:
(a) If n < 1, namely (ρ0 C0 )1 < (ρ0 C0 )2 , then F > 0. In such a case, the reflective
stress disturbance and the incident stress disturbance have the same sign (reflection
loading), while the stress of transmitted wave is larger than that of the incident
wave (T > 1). This is the so-called wave transmission from a “softer” material into
a “harder” material, as shown in Fig. 3.7(a).
In the special case of (ρ0 C0 )2 → ∞(n → 0), which corresponds to the reflection
of elastic waves at the rigid wall (fixed end), there is T = 2, F = 1.
(b) If n > 1, namely (ρ0 C0 )1 > (ρ0 C0 )2 , then F < 0. In such a case, the reflective
stress disturbance and the incident stress disturbance have opposite signs (reflec-
tion unloading), while the stress of transmitted wave is smaller than that of incident
wave (T < 1). This is the so-called wave transmission from a “harder” material into
a “softer” material, as shown in Fig. 3.7(b). Thereby, we can explain the cushion
effect of “softer” materials.
In the special case of (ρ0 C0 )2 → 0 (corresponding to n → ∞), which corre-
sponds to the reflection of an elastic wave at the free surface (free end), there is
T = 0, F = −1.
It should be noticed, for two different media, although their ρ0 and C0 are different, so
long as their wave impedances are the same, namely (ρ0 C0 )1 = (ρ0 C0 )2 (corresponding
to n = 1), there is no reflection (F = 0) at the interface of two media, which is named
the impedance matching. In the cases where no wave reflection is expected or required,
the requirement of impedance matching should be satisfied in the selection of materials.
For instance, in the experimental study of stress wave propagation, if the wave impedance
of gauges matches with that of the tested materials, the unwanted disturbances on the
measured wave signals can be avoided.
3.6 Reflection and Transmission of Elastic Waves in Bars with Varying

Cross Sections
In a bar with different cross sections, when a stress wave passes through a sudden varying
section, a wave reflection and transmission occurs, as shown in Fig. 3.8. The stress distri-
bution in the bar before the reflection is shown in Fig. 3.8(a), and the stress distribution
after the reflection is shown in Fig. 3.8(b). In such a case, the requirement of stress equality
across the interface [Eq. (3.7)] should be replaced by the requirement of total action force
equality across the interface, while the requirement of particle velocity continuity across
the interface [Eq. (3.8)] is still valid, so we have
A1 (∆σI + ∆σR ) = A2 (∆σT ),

∆σI ∆σR ∆σT
− =
(ρ0 C0 )1 (ρ0 C0 )1 (ρ0 C0 )2
Solving these simultaneous equations, we obtain

∆σR = F (∆σI )
(3.12)
∆vR = −F (∆vI )

∆σT = T (∆σI )A1 /A2
(3.13)
∆vT = nT (∆vI )
A1 I A2 R T
(C0)
s s
I R T
(a) X (b) X
Fig. 3.8. Reflection and transmission of stress waves in a bar with sudden varying cross sections.
⎫
(ρ0 C0 A)1 ⎪
n= ⎪
(ρ0 C0 A)2 ⎪
⎪
⎪
⎪
⎬
1−n
F= (3.14)
1+n ⎪
⎪
⎪
⎪
2 ⎪
⎪
T= ⎭
1+n
where (ρ0 C0 A) is called the generalized wave impedance.
If the wave impedances on the two sides of an interface are identical and the wave reflection
and transmission are caused only by a section-area discontinuity of the bar, such as the
case of a stepped bar, then there is n = A1 /A2 . Introducing TA = nT , then ∆σT =
TA ∆σI . Since n and T are always positive, and the same for TA , consequently, the signs of
transmission wave and reflective wave are always the same; while the sign of F depends
on the relative value of A1 and A2 .
When a stress wave propagates from a bar segment with a small section into a bar segment
with a large section (A1 < A2 or n < 1), the signs of the reflected stress and the incident
stress are identical (reflection loading). But, due to TA = 2n/ (1+n) < 1, the transmission
stress is less than the incident stress.
When a stress wave propagates from a bar segment with a large section into a bar segment
with a small section (A1 > A2 or n>1), the signs of the reflected stress and the incident
stress are opposite (reflection unloading). But the transmission stress is larger than the
incident stress (TA > 1).
This is different from the reflection and transmission caused only by the different wave
impedance (ρ0 C0 ). Based on these, it is clear that, if the larger end of a stepped bar is sub-
jected to an impact loading, then the smaller segment acts as a wave trap. However, when
A2 /A1 → 0 (n → ∞), TA → 2. Thereby, for a multi-stepped bar, when a stress wave
passes an interface of section-area discontinuity, the magnification limit of a transmission
stress is 2 for each step.
A taper or conical bar can be approximately viewed as a series of stepped bars with strong
discontinuous interfaces of section-area. As shown in Fig. 3.9, when a compressive pulse
propagates rightward from the bottom of a large section to the top of a small section,
each time the stress wave passes through a strong discontinuous interface, the rightward
transmitted stress disturbance increases TA = [2An /(An + An+1 )] times and simulta-
neously a leftward tension unloading disturbance is reflected at an interface. When this
leftward tension unloading disturbance propagates from the small section to the large
section, a rightward intensified tension disturbance will be reflected. From these, it can
be imagined that with the propagation of a compressive pulse from the large bottom to
the small top of the stepped bar, the compressive stress intensity at the pulse front will
successively increase, while a tension zone following the compression zone of the pulse
will be developed. The tensile stress intensity at the pulse tail zone will also successively
increase. Once the tensile stress is large enough, it will result in a tensile break at a certain
section near the top and the related segment will fly off. Such a dynamic fracture may take
place even before the compressive pulse arrives at the top, so it is different from another
form of dynamic fracture, the so-called spallation, which will be discussed in Section 3.9.
An example with calculated data is also shown in Fig. 3.9 (Johnson, 1972). A conical
bar is approximated by a stepped bar consisting of several segments with equal length l.
The diameter ratios of each segment to the minimum diameter (segment F) are 2, 3, 4, 5,
and so on, or the section-area ratios are 4, 9, 16, 25, and so on. Assume that a rectangle
compressive pulse of unit intensity (σ0 = −1) with pulse length of l propagates rightward
from the segment A to the segment B at a wave velocity of C0 . Introducing T (= l/C0 )
Area Proportion 36 25 16
9 4 1
B C D E F
2 1
4 3
Diameter Proportion 6 5
t
+0.249 −4.08
5T +0.055 +0.47 +0.386 5T
+1.53
4T +0.304 +0.146 +0.71 −2.55 4T
+0.213
3T +0.403 −1.84 3T
2T +0.259 −1.44 2T
+0.18 1T −1.18 1T
o l 2l 3l 4l 5l X
s 0.71
0.304 0.146
t = 4T
X
−2.55
s 1.53
0.47 0.386
0.055
t = 5T
X
−4.08
Fig. 3.9. Reflection and transmission of an elastic stress pulse in a stepped bar.
to denote the time duration required for propagating a distance l, then the computational
results of the wave propagation process before the stress disturbance reaches the top of
the bar, including the stress distribution profiles at t = 4T and t = 5T , are shown in
the bottom part of Fig. 3.9. It can be seen that, at time t = 4T , the intensity of the
compressive pulse arriving at the segment E already reaches σ/σ0 = 2.55, and the tension
stress following it already reaches σ/σ0 = −0.71. At time t = 5T , the intensity of the
compressive pulse arriving at F further increases to σ/σ0 = 4.08, and the tension stress
following it further increases to σ/σ0 = −1.53, namely, the tension stress induced by the
section-area decrease has exceeded the intensity of incident compressive pulse.
Indeed, after introducing the generalized wave impedance, a longitudinal wave propa-
gating in a conical bar can be approximated by a 1D stress longitudinal plane wave. For
an elastic conical bar, assuming that its section area A(X) is a function of coordinate X,
its governing equations consist of the continuity equation, the motion equation, and the
Hooke’s constitutive relation as shown below:
⎫
∂ε ∂v ⎪
= ⎪
⎪
∂t ∂X ⎪
⎬
∂v ∂(σ A) ∂σ ∂A (3.15a)
ρ0 A(X) = =A +σ ⎪
∂t ∂X ∂X ∂X ⎪⎪
⎪
⎭
σ = Eε
Eliminating ε in these equations and introducing the dimensionless parameters below,
= X , t¯ = t , σ v
X σ = , v̄ =
L tL σ0 v0
where L is the characteristic length of the conical bar, σ0 the maximum stress of the
incident wave, tL = L/C0 the characteristic time, v0 = σ0 /(ρ0 C0 ) the characteristic
particle velocity, then Eq. (3.15) can be rewritten as the following first-order differential
equations with respect to the dimensionless parameter σ and v̄:
⎫
∂ v̄ d ln A
= 0⎪
∂σ
− −
σ ⎪
⎬
∂t ∂X dX
(3.15b)
∂ v̄ ∂σ ⎪
⎪
− =0 ⎭
∂X ∂ t¯
According to the theory of characteristics discussed in Chapter 2, it is easy to deduce that
the differential equation of characteristics and the differential equation of characteristic
compatibility relationships are respectively

± d t¯ = 0,
dX
(3.15c)
σ +
d σ d ln A ± d v̄ = 0
where the negative sign corresponds to the rightward waves and the positive sign corre-
sponds to the leftward waves. It is clear that, the cone shape A(X) influences the stress
wave propagation through the item ln A(X) in the characteristic compatibility relationship,
which results in the distortion of wave profile.
Once the A(X) has been given, by using the characteristics numerical method, the problem
is solvable. For example, consider an elastic straight conical bar, its radius and area of
the large end are RL and AL , respectively; and its radius and area of the small end are
RS and AS , respectively. If we introduce the dimensionless area A = A/AL and the

dimensionless radius R = RL /RS , then A(X) can be expressed in a dimensionless form
as below
2 2
= A(X) = 1 − 1 − RS X = 1 − 1 − 1 X
A
AL RL L
R
For a straight conical bar with dimensionless radius R = RL /RS = 3, if its large end
is subjected to a triangle stress pulse with a dimensionless pulse length λ̄ (namely the
dimensionless pulse duration τ̄ ) defined as
τ τ C0 λ
τ̄ = = = = λ̄
tL L L
then the characteristics numerical results of the transmission stress wave at the small end
are shown in Fig. 3.10 for τ̄ = 0.5, 1, 2, 3, 4, and 6 (Wang and Hu, 1988). It is clear
that the conical bar plays a role of a stress amplifier, when a stress wave propagates from
the large end to the small end. On the contrary, the conical bar plays a role of a stress
attenuator, when a stress wave propagates from the small end to the large end. It can also
be seen from Fig. 3.10 that, with the decreasing pulse length of incident stress pulse at the
large end, the magnification of dimensionless stress σ = σS /σ0 at the small end increases,
but its upper limit is σS /σL = RL /Rs , which can be obtained from Eq. (3.15c) and
sS
sL
3
a 1 ab c d e f
b
c
d e f
2 0 2 4 6 t
0 1 2 3 4 5 6 7 t
−1
Fig. 3.10. Triangle stress pulses with various pulse lengths propagate in a straight conical bar,
showing the stress magnification effect.
the dynamic compatibility condition across a strong discontinuous wave front (Eq. 2.57),
namely [σ ] = ±[v] in dimensionless form.
This feature of a conical bar, in modulating the shape and intensity of stress wave, can be
applied in stress wave working and profile-modulation in the split Hopkinson pressure bar
technique, etc. (Refer to Wang and Hu, 1988; Liu and Hu, 2000).
3.7 Hopkinson Pressure Bar and Flying Piece
Utilizing the property that the compressive pulse is reflected as a tension pulse at the
free end of a bar, Hopkinson (1914) presented a crafty test to measure the pressure–
time waveform in a bar when one end of the bar is loaded by an explosion or shot by a
bullet. The equipment used by him is usually named the Hopkinson pressure bars (HPBs),
as schematically shown in Fig. 3.11. Its main component is an elastic cylindrical steel bar,
which is horizontally hung by strings and can swing in the perpendicular plane. One end
of the bar is subjected to a sudden pressure pulse, which is produced by either an explosion
or an impact of a bullet; on the other end the bar connects with a short cylinder (called
timer or flying piece). The connected surfaces of both the bar and the timer have been
grinded as smooth as possible. Then the two objects connect together by a little grease or
magnetic force so that the connection does not influence the propagation of stress wave
on the one hand, but cannot bear tension on the other hand.
Thereby, when a pressure pulse is reflected as a tension pulse at the free end of the timer,
a net tension stress can be formed at the bar–timer interface due to the superposition of
the incident pressure pulse and the reflected tension pulse, and then the timer flies off with
the momentum passing into it. The momentum of the timer can be measured by using a
ballistic pendulum, as schematically shown in Fig. 3.11, while the momentum kept in
the bar can be determined by measuring the swing amplitude of the bar.
Obviously, if the length of the timer is equal to or larger than half of the pulse length of the
incident pressure pulse, the entire momentum of the pressure pulse is passed to the timer;
consequently, the bar is at rest when the timer flies off. Thereby, by changing the length of
Explosive Pressure bar Time-meter Ballistic pendulum
Fig. 3.11. Schematic of Hopkinson pressure bar.

Pressure bar Time-meter Pressure bar Time-meter
(a) (b)
Fig. 3.12. Measuring waveform of a pressure pulse by using timers of different lengths.
the timer, finding the minimum length l0 of the timer when the bar is at rest while the timer
flies off, we obtain the pulse length λ = 2l0 , or the pulse duration τ = λ/C0 = 2l0 /C0 of
the incident pressure pulse. Figure 3.12(a) shows the superposition of the incident wave
and the reflected wave in such a case. If the length of the timer is smaller than l0 , then
only a part of the momentum of the pressure pulse is passed to the timer, as shown in
Fig. 3.12(b). Note that the momentum m corresponds to the area under the pressure–time
curve of stress wave, namely
T
m=A σ (t)dt
0
where A is the section area of the timer (or the bar), T is the time duration that a stress wave
propagates back-and-forth in the timer. Therefore, by using a series of timers of different
length, the waveform of a pressure pulse can be obtained approximately.
Certainly, the precision of the waveform of a pressure pulse measured by using timers of
different length is not as high as that of the σ –t relationship measured directly, and the
connection between the bar and the timer can also make errors. In principle, the maximum
pressure measured cannot exceed the yield strength of the bar material. Moreover, the
bar diameter should be much smaller than the pulse length, otherwise, the assumption
that transverse inertia is negligible, required in the theory of 1D stress waves cannot be
satisfied, in which case there will be a stress dispersion. These limit the application of
Hopkinson pressure bar. However, before the establishment of the modern technology
used to measure the pressure pulse σ (t) directly, this experimental method was very
helpful. Up till now, various kinds of improved Hopkinson pressure bars, especially the
Split Hopkinson Pressure Bar (SHPB) are still being applied and developed, which will
be discussed in the next section.
On the other hand, the flying-plate technology based on the timer flying-off principle has
been widely applied to measure the particle velocity caused by the wave reflection at the
free surface, or applied as a high-velocity impact technique. For example, the velocity
of a flying-plate driven by explosion can reach to up 5 mm/µs. The pressure generated
under such high velocity reaches up to 200 GPa, which is much higher than the velocity
induced by the contact explosion. Thereby, the flying-plate technology is of significance
in the research of state equation of materials under high pressure.
3.8 Split Hopkinson Pressure Bar
Kolsky (1949) first modified the Hopkinson pressure bar discussed in the previous sec-
tion to study the dynamic material behavior at high strain rates and their mathematical
expressions, namely, the dynamic constitutive relations of materials. He established
an experimental technology, called the split Hopkinson pressure bar (SHPB) or
Kolsky bar.
It should be pointed out that, to distinguish from the research on the static behavior of mate-
rials, in the research on the dynamic behavior of materials at high strain rates, in general,
two dynamic effects are considered, namely, the inertia effect in structures and the strain
rate effect in materials. The key to the problem is how to distinguish these two effects.
The main difficulty is that these two effects are usually interconnected, interdependent,
or intercoupled, which make the problem much more complicated. In fact, on the one hand,
no wave propagation in structures can be analyzed without knowing the corresponding
constitutive relation of material, and consequently the basic characteristics of wave prop-
agation inevitably depend on the mechanical behavior of materials, as discussed earlier.
On the other hand, in the study of dynamic constitutive relationships of materials at high
strain rates, the stress wave propagation either in the specimens or in the experimental
equipments should not be neglected. Thus, we have to handle a hot potato – a kind of
recurrence of “whether hen before egg or egg before hen?”
How to solve this circular argument? In the view of the research of dynamic constitutive
behavior of materials, there are two approaches at present. The first one uses specimens
of simple structures that are easy to analyze with the theory of stress wave propagation.
With the known explosion or impact loading (namely the initial and boundary conditions
are known), experimentally measure the dynamic information (such as the wave signals)
or the residual information (such as the residual strain distribution) of the wave propaga-
tion. Based on these, we can deduce the dynamic constitutive relationship of materials.
In principle, this approach is to determine the dynamic constitutive relationship of materials
from the information of stress wave propagation, namely, to solve the so-called “second
kind of reverse problem”. The relevant discussion will be conducted in the following
sections.
The key to the second approach is to try to uncouple the stress wave effect and the strain
rate effect in experiments. Experiments using the split Hopkinson pressure bar (SHPB)
that will be discussed in this section are the typical examples of this approach and has been
extensively used. Its schematic is shown in Fig. 3.13, where the strike bar (bullet), input
bar (incident bar), and output bar (transmission bar) is generally of identical diameter and
material, namely the elastic modulus E, wave velocity C0 , and wave impedance ρ0 C0
are identical, and are required to keep in an elastic one-dimensional stress state. During
an experiment, a short specimen is sandwiched between the long input bar and the long
output bar. When a strike bar of length L0 driven by an air-gun hits the input bar at
velocity v∗ , an incident pulse σI (t) is generated. The amplitude of the incident pulse can
be adjusted by controlling the impact velocity v∗ , while its duration (= 2L0 /C) can be
adjusted by controlling the length L0 of the strike bar. The short specimen is deformed
in high strain rate under the action of incident pulse. Meanwhile, a reflected pulse σR (t)
propagates back to the input bar and a transmission pulse σR (t) propagates to the output bar.
Gas gun Light beams Input bar Specimen Output bar Absorption bar Dash pot
Strain gauge
Strike bar
Oscillator Bridge & amplifier
Electronic counter Transient wave memory Data processing system
Fig. 3.13. Schematic of the split Hopkinson pressure bar (SHPB) set up.
These two pulses reflect the dynamic mechanical behavior of the specimen. The signals of
these two pulses are measured and recorded by a system mainly consisting of two strain
gauges mounted on the surfaces of the two bars respectively, an instrument for super-
dynamic strain measurement and a transient wave memory, etc. while the impact velocity
is measured by a system consisting of a pair of parallel light beams and photoelectric cell,
an amplification circuit, an electronic time counter, etc. The absorption bar works like the
flying plate (timer) as shown in Fig. 3.11. When the transmitted pulse is reflected at the
free end of the absorption bar, the bar flies off with all the momentum passed to it so that it
keeps the output bar at rest. All the transmitted energy is finally dissipated by a dash-pot,
as shown in Fig. 3.13.
It should be emphasized that the SHPB technique is established on two basic assumptions,
namely (1) one-dimensional (1D) stress state in bars, and (2) uniform distribution of
stress–strain along the length of the short bar.
We will examine the first assumption. Under the condition that the 1D stress assumption is
satisfied, once the stress σ (X1 , t) and particle velocity v(X1 , t) at the interface X1 between
the specimen and the input bar (Fig. 3.14) as well as the stress σ (X2 , t) and the particle
velocity v(X2 , t) at the interface X2 between the specimen and the output bar (Fig. 3.14)
are measured, the average stress σs (t), strain rate ε̇s (t) and strain εs (t) of the specimen
Input bar Specimen Output bar
X1 X2
Strain gauge G1 Strain gauge G2
Fig. 3.14. Layout of input bar, specimen, and output bar.

can then be determined according to the equations below,
A A
σs (t) = [σ (X1 , t) + σ (X2 , t)] = [σI (X1 , t) + σR (X1 , t) + σT (X2 , t)]
2As 2As
(3.16a)
v(X2 , t) − v(X1 , t) vT (X2 , t) − vI (X1 , t) − vR (X1 , t)
ε̇s (t) = = (3.16b)
ls ls
t t
1
εs (t) = ε̇s (t)dt = [vT (X2 , t) − vI (X1 , t) − vR (X1 , t)]dt (3.16c)
0 ls 0
where A is the section area of pressure bar, As the section area of specimen, and ls the
length of specimen. In the case of elastic pressure bar, from the previous discussion on
the 1D elastic waves in this chapter, it has been established that there exists the following
linear relationship between the strain and the stress as well as the particle velocity,
⎫
σ1 = σ (X1 , t) = σI (X1 , t) + σR (X1 , t) = E [εI (X1 , t) + εR (X1 , t)]⎪
⎪
⎬
σ2 = σ (X2 , t) = σT (X2 , t) = EεT (X2 , t)
(3.17)
v1 = v(X1 , t) = vI (X1 , t) + vR (X1 , t) = C0 [εI (X1 , t) − εR (X1 , t)]⎪
⎪
⎭
v2 = v(X2 , t) = vT (X2 , t) = C0 εT (X2 , t);
Then, the problem is converted to how to measure the incident strain wave εI (X1 , t) and
the reflected strain wave εR (X1 , t) at the interface X1 and the transmitted strain wave
εT (X2 , t) at the interface X2 . However, so long as the pressure bar is in elastic state, all
waveforms at different positions are the same. In other words, by using the characteristic
that there is no wave profile distortion when an elastic wave propagates in a long bar in the
1D stress state, the incident strain wave εI (X1 , t) and the reflected strain wave εR (X1 , t)
at the interface X1 can be replaced by the incident strain wave εI (X1 , t) and the reflected
strain wave εR (X1 , t) measured by the gauge G1 mounted at XG1 on the input bar, and the
transmitted strain wave εT (X2 , t) at the interface X2 can be replaced by the transmitted
strain εT (XG2 , t) measured by the gauge G2 mounted at XG2 on the output bar. Therefore,
we can determine the dynamic stress σs (t) and strain εs (t) of the specimen through the
signals measured by the gauge G1 and the gauge G2 .
EA EA
σs (t) = εT (XG2 , t) = [εI (XG1 , t) + εR (XG1 , t)] (3.18a)
As As

2C0 t 2C0 t
εs (t) = − εR (XG1 , t)dt = [εI (XG1 , t) − εT (XG2 , t)] dt (3.18b)
ls 0 ls 0
It should be pointed out that when a reflected wave generated at the interface X1 and a
transmitted wave generated at the interface X2 propagate in the input bar and the output
bar respectively, simultaneously there are stress waves propagating back-and-forth in
the specimen between the interface X1 and the interface X2 . We can imagine that if the
specimen is short enough, the stress–strain is quickly distributed uniformly along its length
and then the stress wave effect in it can be neglected. This is the second assumption that
SHPB experiments are based on. According to this assumption of uniform stress–strain
distribution, there is σ1 = σ2 ; or combining the theory of 1D stress waves, there is
σI + σR = σT , εI + εR = εT (3.19)
Then, if any two of the incident strain wave εI (X1 , t), the reflected strain wave εR (X1 , t)
and the transmission strain wave εT (X2 , t) have been measured, we can determine the
dynamic stress σs (t) and strain εs (t) according to Eq. (3.18). After eliminating the time
parameter t, we obtain a dynamic stress–strain curve, σs –εs , of the specimen material at
high strain rate.
From the above discussion, it is known that the art of SHPB technology lies in that it
uncouples the stress wave effect and the strain rate effect. On the one hand, with respect
to the input bar and the output bar which play the roles of both impact loading and dynamic
measuring, since they are always in elastic state, only the stress wave propagation should
be considered while the strain rate effect can be neglected. So long as the bar diameter is
small enough, the theory of 1D stress waves can be applied to analyze these experiments.
On the other hand, if the specimen sandwiched between the input bar and the output bar is
short enough so that the specimen can be viewed as in the state of uniform deformation,
then, only the strain rate effect should be considered for the specimen while the stress wave
effect can be neglected. In fact, if the specimen is so short that the duration for a stress
wave propagating between the two ends of the specimen is much shorter than the loading
duration due to the incident wave, then the specimen can be approximately regarded as in
a state of uniform deformation. Thus, the stress wave effect and the strain rate effect in the
pressure bars and the specimen are cleverly uncoupled respectively, and consequently the
rate-dependent mechanical response of specimen material can be obtained by analyzing
the signals of stress wave propagating in elastic bars. For the specimen, this corresponds
to conducting a “quasi-static” test, but at high strain rates; for the pressure bar, this
corresponds to solving an inverse problem, namely, to deduce the constitutive response of
the adjacent specimen from the signals of wave propagation.
Certainly, the assumption of 1D stress waves in bars and the assumption of uniform
distribution of stress–strain along the specimen length (or henceforth the uniformity
assumption for short) are also the constraints to assure the validity and reliability of
SHPB experiments. Considering how to satisfy the assumption of 1D stress waves in
bars, the key is whether the transverse inertia effects can be neglected; readers can refer
to Section 2.8. In the following, we will examine the factors influencing the uniformity
assumption based on the theory of elastic wave propagation in bars.
Obviously, once the specimen is loaded by an incident pulse, the earlier the uniformity
assumption is satisfied, the more reliable the experimental results will be. Obviously,
it would be perfect if the uniformity assumption could be satisfied when the specimen
is still in an elastic state. Therefore, in the analysis below, we assume that the input
bar–specimen–output bar in Fig. 3.14 are all in elastic state and denote the generalized
elastic wave impedance by (ρCA)B and (ρCA)S for the pressure bar and specimen respec-
tively. Without loss of generality, we also assume that the section areas of pressure bar and
specimen are identical. As discussed in Section 3.5, the wave reflection and transmission in
a system of input bar–specimen–output bar mainly depend on the ratio of (ρC)B and ρCS .
Thus, based on the analysis of elastic waves discussed earlier in this chapter, the process of
reflection and transmission of elastic waves in the input bar–specimen–output bar system,
and the state of stress σ and particle velocity v in each step, can be determined and
graphically displayed on the X–t plane and σ –v plane, as shown in Fig. 3.15(a and b).
In these figures, it has been assumed that the strike bar hits the input bar at velocity v0 ,
and so a strong discontinuous elastic wave (with a rectangle wave front) is generated, the
stress amplitude of which is σA = −(ρC)B v0 /2.
According to Eqs. (3.9)–(3.11), when an incident wave σA propagates with the elastic wave
velocity (C)B in the input bar and arrives at the interface X1 , the primary transmission and
reflection occur. The transmitted wave passes to the specimen at the elastic wave velocity
(C)S and induces a strong discontinuous stress disturbance
∆σ1 = σ1 − 0 = TB−S σA
where the transmission coefficient TB−S can be calculated according to Eq. (3.11)
2 (ρC)B
TB−S = , nB−S =
1 + nB−S (ρC)S
where the subscript “B–S” of the transmission coefficient T and the ratio of wave
impedance n indicates that the stress waves propagate from bar B to the specimen S.
8
7
s
6 v0 /2 v0
B v
0
5
1
4 2
3 3
4
2 5
6
7
2ts
1 8
A 0 sA
A
Ls X
X1 X2
Fig. 3.15. Reflection and transmission of elastic waves in a system consisting of input bar, specimen,
and output bar (rectangle wave front).
At time τS = LS /(C)S , where LS is the length of the specimen, the secondary transmis-
sion and reflection occur. According to Eq. (3.9), the reflected wave propagating back to
the specimen induces a strong discontinuous stress disturbance
∆σ2 = σ2 − σ1 = FS−B ∆σ1
where the reflection coefficient FS−B can be calculated according to Eq. (3.11)
1 − nS−B
FS−B =
1 + nS−B
where the subscript “S–B” of the reflection coefficient F indicates that the stress waves
propagate from the specimen S to bar B. Noticing that nB−S and nS−B are reciprocal, and
rewriting nS−B as β, it can be easily proved that there exist the following relationship
between TB−S and FS−B and can be written respectively as:
⎫
2 2β ⎪
⎪
TB−S = = ⎪
⎪
1 + nB−S 1+β ⎪
⎪
⎪
⎬
1−β
FS−B = (3.20)
1+β ⎪
⎪
⎪
⎪
1−β 2β ⎪
⎪
1 − FS−B = 1 − = = TB−S ⎪
⎭
1+β 1+β
When the reflected wave returns to the interface X1 , the third transmission and reflection
occur. The strong discontinuous stress disturbance induced in the specimen is
∆σ3 = σ3 − σ2 = FS−B ∆σ2 = FS−B

2
∆σ1
On this analogy, when the kth transmission and reflection occur, a strong discontinuous
stress disturbance induced in the specimen is
∆σk = σk − σk−1 = FS−B ∆σk−1 = FS−B

k−1
∆σ1 (3.21)
while the final state of stress σk in the zone k (Fig. 3.15) after kth transmission and
reflection is
.
k
σk = ∆σi = 1 + FS−B + FS−B
2
+ FS−B
3
+ · · · + FS−B
k−1
∆σ1 (3.22a)
i=1
By using the expansion of a binomial below

1 − x k = (1 − x) 1 + x + x 2 + x 3 + · · · + x k−1
and considering the relationship between TB−S and FS−B given by Eq. (3.20) and the
expression of β, Eq. (3.22a) can finally be written as
/ 0
1 − FS−B
k 1 − FS−B
k 1−β k
σk = ∆σ1 = TB−S σA = 1 − FS−B σA = 1 −
k
σA
1 − FS−B 1 − FS−B 1+β
(3.22b)
This indicates that after k-times back-and-forth transmissions and reflections in the speci-
men, the stress σk in the specimen depends on both the number of transmission–reflection
times and the ratio of impedances of specimen and pressure bar. Note, the number of times
k indeed equals the non-dimensional time t¯ = t/τS = t (C)S /LS .
For a given β, when k is an even number (refer to Fig. 3.15), Eq. (3.22) shows how the
transmission stress at the interface X2 varies with the number of transmission–reflection
times k or the dimensionless time t¯ = t/τS = t (C)S /LS ), while when k is an odd number,
Eq. (3.22) shows how the reflective stress at the interface X1 varies with the number of
transmission–reflection times k or the dimensionless time t¯.
As an example, when β = 1/10, according to Eq. (3.22), the variation of dimensionless

stress σ/σA with k (or t¯) at the interface X1 and interface X2 are shown in Fig. 3.16(a and b).
Both of them gradually approach 1, which indicates that the stresses along the specimen
length gradually realize a uniform distribution, namely it is a process depending on both
β and k.
Noticing that Eq. (3.21) just shows the stress difference between that at the interface X1
and at the interface X2 (refer to Fig. 3.15), a dimensionless stress difference (relative
stresses) for the two ends of the specimen can be defined as
∆σk
αk = (3.23)
σk
For a rectangular strong discontinuous incident wave, substituting Eqs. (3.21) and (3.22)
into Eq. (3.23), we obtain
k−1
1−β 1−β
k−1 1−
∆σk FS−B 1+β 1+β 2β (1 − β)k−1
αk = = = k =
σk 1 − FS−B
k
1−β (1 + β)k − (1 − β)k
1−
1 − FS−B 1+β
(3.24)
This formula depicts analytically how the relative stress difference for the two ends varies
with the impedance ratio β and the number of transmission–reflection times k. For different
values of β (= 1/2, 1/4, 1/6, 1/10, 1/25, 1/100), the variation of αk with k calculated
according to Eq. (3.24) is shown in Fig. 3.17. It is clear that the αk increases with decreasing
β for a given k, or in other words, with the decrease of the impedance ratio β, stress waves
must be reflected more times in the specimen in order to satisfy the requirement of the
uniformity assumption.
1.2
at X1
1
s/sA 0.8
0.6
0.4
0.2
0
0 4 8 12 16 20 24 28 32 36
k(= t/ts = tCs /Ls)
(a) At interface X1
1.2
at X2
1
0.8
s/sA
0.6
0.4
0.2
0
0 4 8 12 16 20 24 28 32 36
k(= t/ts = tCs /Ls)
(b) At interface X2
Fig. 3.16. Variation of the dimensionless stress σ/σA with k (or t¯) at the interface X1 and the
interface X2 , when β = 1/10.
0.35
0.5
0.3 0.25
0.16667
0.25 0.1
0.04
0.2 0.01
ak
0.15
0.1
0.05
0
0 2 4 6 8 10 12 14 16 18 20
k
Fig. 3.17. Variation of the dimensionless stress difference αk with the wave impedance ratio β and
the number of reflection times k (for a rectangular incident wave front).
As suggested by Zhou et al (1992a) and Ravichandran and Subhash (1994), if the stress
(strain) distribution along the bar can be regarded as approximately satisfying the uni-
formity assumption when αk ≤ 5%, then it can be seen from Fig. 3.17 that in the case
of β = 1/2 the minimum number of reflective times is kmin = 4, while in the case of
β = 1/100, the minimum number of reflective times increases to kmin = 18.
The above results are obtained for the rectangular strong discontinuous incident wave.
However, in most SPHB experiments, the incident wave with a trapezoidal front, namely
with a certain “rise time” tr , are usually observed. For such forms of waves, the analysis
methods similar to those used in the above discussion are still valid, although the problem
becomes more complicated. Assuming that the rise time of the trapezoidal wave front tr
just equals the time an elastic wave propagates around the specimen, namely tr = 2τS =
2LS /(C)S , Yang and Shim (2005) obtained an analytical solution for the case after an
elastic wave propagates back-and-forth one time in the specimen (k > 2), as
2β 2 (1 − β)k−2
αk = , k >2 (3.25)
(1 + β)k − (1 − β)k−2
For different values of β (= 1/2, 1/4, 1/6, 1/10, 1/25, 1/100), the variation of αk with k
calculated according to Eq. (3.25) is shown in Fig. 3.18. It is clear that, contrary to the case
for a rectangular incident wave (Fig. 3.17), the αk –k curves now move downwards with
decreasing β, namely for a given k the αk decreases with decreasing β. It is worthwhile to
notice that in the range of β discussed in this example, the minimum number of reflection
times for satisfying the uniformity assumption is only 3–4.
Let us now consider a linear ramp wave, namely an incident wave with a longer rise time.
Assume that the rise time tr linearly increases with time, namely, it can be depicted by
σI (t) = σ ∗ t/τS = σ ∗ (C)S t/LS , where σ ∗ denotes the stress when t = τS = LS /(C)S .
0.1
0.5
0.08 0.25
0.16667
0.06 0.1
0.04
ak
0.04 0.01
0.02
0
0 2 4 6 8 10 12 14 16 18 20
k
the numbers of reflection times k (for a trapezoidal incident wave front).
0.3
0.5
0.25 0.25
0.2 0.16667
0.1
0.15
0.04
ak
0.1 0.01
0.05
0
−0.05
0 2 4 6 8 10 12 14 16 18 20
k
the numbers of reflection times k (for a linear ramp incident wave front).
Yang and Shim (2005) presented an analytical solution as below,

/ k 0
1 − β
2β 2 1 − −
1+β
αk = (3.26)
1−β k
2kβ − 1 +
1+β
For different values of β (= 1/2, 1/4, 1/6, 1/10, 1/25, 1/100), the variation of αk with k
calculated according to Eq. (3.26) is shown in Fig. 3.19. It is clear that the αk –k curves for
the linear ramp wave also move downwards with decreasing β, but are accompanied by a
marked oscillation of the curve. In addition, in the range of β discussed in this example,
the minimum number of reflection times for satisfying the uniformity assumption in the
present case is larger than that for the trapezoidal wave.
For the same β, the comparison of αk –k curves between the rectangular wave, the
trapezoidal wave, and the linear ramp wave is shown in Fig. 3.20.
It can be seen from Fig. 3.20 that, when β = 0.5, there is almost no difference between
the rectangular wave and the trapezoidal wave, and both are better than the linear ramp
wave with respect to satisfying the uniformity assumption; but with the decrease of β, the
αk –k curves of the trapezoidal wave and the linear ramp wave move downwards, but
the αk –k curves of the rectangular wave move upwards, consequently the situation
changes. Therefore, when β = 0.1, the rectangular wave becomes the worst one while
the trapezoidal wave becomes the best one with respect to satisfying the uniformity
assumption.
From the above discussion on the propagation of stress waves in the system consisting
of input bar, specimen, and output bar, both the impedance ratio β and the waveform,
b = 0.5
0.3
0.25
0.2 C
aK
0.15
0.1
B
0.05
A
0
0 2 4 6 8 10 12 14 16 18 20
k
b = 0.25
0.25
0.2 A
0.15
aK
0.1
C
0.05
B
0
0 2 4 6 8 10 12 14 16 18 20
k
b = 0.1
0.3 A
0.25
0.2 C
aK
0.15
0.1
0.05
0
0 2B 4 6 8 10 12 14 16 18 20
k
Fig. 3.20. Comparison of αk –k curves of the rectangle, trapezoidal, and linear ramp wave front
(denoted by curve A, B, and C respectively) under an identical β.
particularly the rise time of the incident wave front, markedly influence the minimum
number of reflection times kmin for satisfying the uniformity assumption.
3.9 Dynamic Fracture Induced by Reflective Unloading Waves
When a compressive pulse is reflected at the free surface of a bar (or a plane) as a tension
pulse, the tension stress in a certain zone near the free surface may be quite large. Once
a certain criterion of dynamic fracture (spalling criterion) is satisfied, the material in
such a zone will be cracked. When the crack is large enough, a whole cracked piece
will fly off with the momentum in it. Such a type of dynamic fracture near the back
surface of the bar (or plane) caused by the reflection of compressive pulse at the free
surface is named spalling or scabbing. The fly-off crack piece is called spalling piece or
scab. Figure 3.21(a) schemes the back spalling of a thick steel plant caused by a contact
explosion.
In the above situation, once a spalling takes place, a new free surface is also formed.
The continuously incident compressive pulse will be reflected at this new free surface;
consequently, a second spalling could be formed, if the spalling criterion is satisfied.
On this analogy a multiple spalling may be formed in a certain condition, and a succession
of multilayer scabs will be generated. Figure 3.21(b) schemes a multi-spalling at the free
end of a cement bar caused by a contact explosion at another end.
It is necessary to emphasize that a compressive pulse always consists of a compressive

loading front and an unloading tail following it. Most engineering materials can undergo
quite intense compressive stress waves without fracture but cannot undergo tension stress
waves of the same stress amplitude. The occurrence of spalling is attributed to a tension
stress that is formed by the reflection of compressive pulse at the free surfaces and this
tension stress is large enough to satisfy the dynamic fracture criterion of materials, while
the occurrence of tension stress is attributed to the interaction of the reflective tension
wave which is formed when an incident compressive wave is reflected at free surface, and
the unloading tail of the incident compressive wave; or briefly, the tension stress is caused
by the interaction of the reflected unloading wave and the incident unloading wave.
Therefore, the intensity and shape of incident compressive pulse greatly influence whether
the spalling occurs, where it happens, and how many scabs are formed, etc. Indeed, the
formation of tension stress is only a prerequisite, whether the spalling occurs depends on
whether the tension stress satisfies the dynamic fracture criterion of materials.
5 4 3 2 1
Scab
(a) (b)
Fig. 3.21. Spalling caused by the reflection of compressive pulse at free surface.
The first and simplest dynamic fracture criterion of materials is the maximum tension
stress instantaneous fracture criterion. According to this criterion, once a tension stress
σ reaches or exceeds the critical tension stress of materials, σc ,
σ ≥ σc (3.27)
a spalling occurs instantaneously, where σc is the parameter representing the material

behavior resisting to dynamic tension fracture, called the dynamic fracture strength of
materials. This criterion was formerly a generalization of the maximum normal stress
criterion in the theory of static strength to the dynamic situation, namely, the fracture
is regarded to occur instantaneously when the criterion is satisfied. So it belongs to the
rate/time-independent theory of fracture. However, the σc is measured in dynamic tests
and is generally much higher than the static strength of materials σb ; in this view, the
rate/time-dependency has been taken into consideration in this criterion.
Based on the theory of 1D elastic waves, according to the criterion of instantaneous fracture
at maximum tension stress, how the spalling takes place under the compressive pulses with
various profiles will be discussed below.
Figure 3.22 schemes the wave profiles at five typical times when an incident rectangular
pulse of length λ is reflected at a free surface: (a) an incident rectangular pulse approaches
the free surface; (b) when one-fourth of the incident pulse has been reflected, the sum
of incident compressive stress and reflected tension stress is zero in the distance range
of λ/4 from the free surface; (c) when half of the incident pulse has been reflected, the
stress superposed is just zero, but the particle velocity in the distance range of λ/2 from
the free surface is twice the particle velocity of the incident wave; after this, the tension
stress will be generated due to the interaction between the incident unloading wave and the
reflected unloading wave; (d) when three-fourth of the incident pulse has been reflected,
a tension zone with a length of λ/2 is formed, while the stress in the distance range
of λ/4 from the free surface still is zero; (e) when the reflection has ended, the rightward
l l
4
l l l
2 4
(a) (b) (c) (d) (e)
Fig. 3.22. Wave profiles at five typical times when an incident rectangle pulse of length λ is reflected
at a free surface.
incident compressive pulse has been totally reflected to a leftward tension pulse. Therefore,
for a rectangle pulse, so long as the stress amplitude of the pulse |σm | is larger than the
dynamic fracture strength σc , namely, |σm | > σc , then just after half of the incident wave
is reflected at the free surface, namely at time t = λ/(2C0 ), spalling occurs; the thickness
of the spalling piece δ = λ/2. Since the spalling piece flies off with the total momentum of
incident compressive pulse, on the one hand, the velocity of the flying-off spalling piece
vf can be calculated which is twice the particle velocity of incident wave [σm /(ρ0 C0 )]:
λ
σm
C0 2σm
vf = =
ρ0 δ ρ 0 C0
On the other hand, for a rectangle pulse, no matter how large the amplitude of incident
pulse is, multiple spalling will not occur.
For a saw-toothed compressive stress pulse with a linearly attenuating tail, the situations
are different. When it is reflected at the free surface, the wave profiles at five typical times
are schemed in Fig. 3.23. It can be seen that even at the beginning of reflection, a net
2d
sm
l
d 2
(a) (b) (c)
(d) (e)
Fig. 3.23. Wave profiles at five typical times when an incident saw-toothed pulse of length λ is
reflected at a free surface.
tensile stress zone is simultaneously formed due to the interaction of reflected unloading
wave and incident unloading wave. The net tensile stress at the front is larger than other
parts of the reflected wave, and it continuously increases with the process of reflection.
When half of the incident wave is reflected, it reaches its maximum. However, before this,
once Eq. (3.27) is satisfied, spalling occurs.
When a stress pulse is expressed as a function of time, namely it is represented by a

time-histories curve σ (t at an arbitrary point X, assume that the time when the wave front
arrives at the point is the start of time, t = 0, if the stress at a distance of δ from the free
surface satisfies the instantaneous fracture criterion based on maximum tensile normal
stress [Eq. (3.27)], then spalling occurs and obviously there should be

2δ
σ (0) − σ = σc (3.28)
C0
For the saw-toothed stress pulse with linearly attenuating tail as shown in Fig. 3.23, the
time-histories profile σ (t) can be expressed as

C0 t
σ = σm 1− (3.29)
λ
where σm is the peak value of the stress pulse. According to Eq. (3.28), the thickness of
the first spalling piece δ1 can be determined as,
λ σc
δ1 = ·
2 σm
The time t1 (after the beginning of reflection) that spalling occurs is
δ1 λ σc
t1 = =
C0 2C0 σm
The velocity of the spalling piece can be approximately calculated, according to that the
momentum of the spalling piece equals the impulse of the part of the pulse entrapped in
the spalling piece, namely
2δ/C0
1
vf = σ (t)dt
ρ0 δ 0
Substituting Eq. (3.29) into the above equation, we obtain
2σm − σc
vf =
ρ0 C 0
If |σm | = σc , similar to the case of a rectangular pulse, then spalling occurs at time
t = λ/(2C0 ) after the beginning of reflection, the thickness of spalling piece δ = λ/2,
all the pulse impulse is entrapped in the spalling piece and multiple spalling is impossible
to occur. But the flying-off velocity of the spalling piece is less than that for an incident
sc
sc
sc
d1
2d 3 2d 2 2d1
Fig. 3.24. Multiple spalling for a saw-toothed pulse with exponentially attenuating tail.
rectangular pulse. If |σm | > σc , then a multiple spalling is possible to occur. The readers
are suggested to prove that if nσc ≤ |σm | < (n + 1)σc , then the number of spalling is n
and the thickness of all spalling pieces are identical, while the flying-off velocity of each
spalling piece will successively decrease.
On the analogy of the above analysis, it is easy to illustrate that for a saw-toothed pulse
with exponentially attenuating tail as shown in Fig. 3.24, the thickness of each spalling
piece will successively increase when multiple spalling occurs.
The above discussion is an approximate analysis based on the theory of 1D elastic waves,
practical problems are much complicated. Even so, it can be seen that, because of the
interaction of stress waves, dynamic fracture can occur not only in the loading condition
but also in the unloading condition; the latter is a kind of the so-called unloading failure
due to stress waves.
It should also be pointed out that, except for perfect crystals, any fracture or rupture of
engineering materials does not occur instantaneously, but is a final result of a time/rate-
dependent process developing at a finite velocity; especially at high strain rates, displayed
as the so-called fracture-delay, the decrease of critical stress with increasing duration
of loading, the increase of flow stress with increasing strain rate and so on. This shows
that the material break depends on not only the magnitude of applied stress but also
on the loading duration or the stress–strain rate. Therefore, in principle an instantaneous
fracture criterion of material like Eq. (3.27) should be replaced by a time- or rate-dependent
dynamic fracture criterion. This research topic has attracted a lot of attention. One class
of time/rate-dependent fracture criterion of material is the damage accumulation criterion
in various forms, in which the evolution process of internal damage of material has been
considered.
For instance, a time-dependent spalling criterion was presented by Tuler and Butcher
(1968), which states that when the stress σ (t) as a function of time t at an arbitrary point
satisfies the following equation, spalling occurs

t
(σ − σ0 )α dt = K (3.30)
0
where α, K, and σ0 are all material parameters, σ0 denotes the minimum stress (threshold
stress) for spalling. When σ (t) < σ0 , dynamic fracture does not occur however long the
duration of the stress pulse is. When α = 1, Eq. (3.30) is reduced to the so-called impulse
criterion, which indicates that dynamic fracture occurs only when the impulse reaches a
critical value.
Finally, it should be pointed out that, similar to the back spalling near the free surface of
a plate or bar, caused by the tensile stress induced by the interaction between an incident
unloading wave and an unloading wave reflected from a free surface, if a compressive
Reflective wave
Corner fracture
Incident wave Incident wave
Fig. 3.25. Corner spalling caused by the net tensile stress induced by two unloading waves reflected
at the two free surfaces respectively of a corner.
Reflective wave
Central Central
fracture fracture
Incident wave Spallation

Corner
fracture
Fig. 3.26. Central spalling occurs due to the interaction of two unloading waves generated by the
reflection of a compressive wave at two free surfaces.
wave propagates to a corner enclosed by two free surfaces at a certain angle1 , then net
tensile stress may be induced by the interaction between the two unloading waves reflected
from these two free surface respectively and then cause a spalling at the corner, which is
called the corner spalling, as approximately schemed in Fig. 3.25.
If two unloading tensile waves which are generated by the reflection of an incident com-
pressive wave at two free surfaces respectively, meet at the center of a solid body, the
so-called central spalling may occur, if the net tensile stress satisfies the spalling crite-
rion, as shown in Fig. 3.26. From the view of fracture mechanism, either the ordinary
spalling, or the corner spalling and central spalling are all relevant to the tensile stress
inducted by the reflection of stress waves, so all of them can be called reflective tensile
fracture.
1 Here, the reflection of stress wave at a free surface is in fact an oblique reflection, which will be discussed in
Chapter 11. When a longitudinal wave is incident obliquely, there are both oblique transverse wave and oblique
longitudinal wave reflected; and consequently the real case is much more complicated than that discussed here.
CHAPTER 4
Interaction of Elastic–Plastic Longitudinal

Waves in Bars
When we study the problems of wave interaction in the case of linear elastic stress waves,
as discussed in Chapter 3, there is no need to distinguish whether such a wave interaction
is a loading process or an unloading process. This is because for linear elastic material,
the stress–strain relationship is linear, identical for both the loading and unloading cases,
representing a reversible thermal–mechanical process. Therefore, there is no essential
distinction between a linear elastic loading wave and a linear elastic unloading wave; both
propagate at speed C0 (= E/ρ0 )1/2 . Nevertheless, when we start to discuss the interaction
between elastic–plastic waves, the problem becomes much more complicated. First, the
stress–strain curves are not linear, thus the superimposition principle is no longer held good.
On the other hand, the stress–strain curves of an elastic–plastic material are different in
the loading and unloading cases, representing irreversible thermal–mechanical processes.
As a result, it is necessary to distinguish whether the stress wave interaction is a loading
case or an unloading case, and then to treat the respective case differently.
In the following sections, we will first discuss the loading cases of two elastic–plastic
wave interaction, and then the unloading cases.
4.1 Interaction of Two Longitudinal Elastic–Plastic Loading Waves in Bars
The interaction between two loading elastic–plastic waves can be categorized as two
types: the head-on loading and the pursuing loading. The former type is the interaction
process when two loading waves propagate in the opposite direction and interact with
each other when they meet together. Such waves must have the same signs (namely, both
are compressive waves or tensile waves), to guarantee that the interaction will load the
material further. If the two stress waves have different signs, the interaction will involve
unloading process. The latter type of interaction is the process when two loading stress
waves propagate in the same direction, and the back one pursues the front one. This
process only happens in the material with increasing strain-hardening coefficient, so that
the plastic wave velocity increases with the plastic deformation. Such a pursuing loading
95
interaction is actually the process of shock wave formation and growing, which has been
qualitatively discussed in the earlier chapter.
In this section we will discuss the head-on loading case in detail. It does not matter
whether material is of increasing strain-hardening or decreasing strain-hardening. When
the reader gets familiar with the basic approaches of the analysis method, it will be easy
to treat the pursuing loading case.
The general analytical method is, in principle, the same as discussed previously for elastic
wave interaction. The interaction between two stress waves is considered as an internal
impact process. On the boundaries of interaction between any two stress waves, the equality
of material velocities (condition of material continuation) and the equality of stresses
(principle of reaction force) should be satisfied. Moreover, the momentum-conservation
condition across a wave front, such as Eq. 2.57 for strong-discontinuous wave front or
Eq. 2.63 for weak-discontinuous wave front, should be satisfied on both sides of the
interaction boundary, respectively. The problem then can be solved from these equations.
Different from the linear elastic case, the wave speed D for strong-discontinuous shock
wave, or C(ε) for weak-discontinuous plastic wave, is no longer constant. Consequently,
the principle of superimposition does not hold good any more. Therefore the problem
becomes more complex.
4.1.1 Head-on loading interaction of two strong-discontinuous stress waves
First we assume that√the material is of linear strain-hardening. In √

this case both the elastic
wave speed C0 (= E/ρ0 ) and the plastic wave speed C1 (= E1 /ρ0 ) are constants.
The problem is slightly simpler.
Consider a bar with uniform cross section and length l, which is originally in a rest,
stress-free condition. At time t = 0, its two ends are suddenly loaded by a constant-speed
impact loading. At the right end of the bar (X = l), the boundary condition is v3 > 0.
At the left end of the bar (X = 0), the boundary condition is v4 < 0. Therefore, along the
bar two strong-discontinuous elastic–plastic tensile waves are propagating, as illustrated
in Fig. 4.1. This is a typical case of the interaction between two strong-discontinuous
elastic–plastic waves propagating in the opposite directions, namely a head-on
loading case.
Before the two waves meet, the elastic–plastic wave propagation is the same as that
propagating in an infinite bar, as discussed in the previous chapter. Thus the stress and
velocity states of the zones marked as 0, 1, 2, 3, and 4 on the X–t plane in Fig. 4.1 are all
known. Let the subscript of the state variables be the same as their zone number, we have:
σ0 = v0 = 0
σ1 = σ2 = Y
Y
v1 = −v2 = vY =
ρ0 C0
Interaction of Elastic–Plastic Longitudinal Waves in Bars 97
t
8
t4
d
7 6
σ
t3
4 c 5 b 3
t2
2 1
t1 a
Y E1
0
E
0 t X 0 ε
σ
4σ4 C
C1 σ 3
1 σ2 σ1 σ
t=t1 C0 C0
8
0 6
X 7 3
σ σ3 4 5
σ 4 C1 C C1 C1 2 Y 1
1
t=t2 σ2 σ1 0 v
0 8
X
σ ϕ
σ6 C1 σ
σ C 1 C1 3 6
σ 4 C1 7
3
σ5 7
t=t3
4 5
0
X
σ σ8 C
C1 1 σ 2 1
σ7 6 C1
C1
σ3 0 v
σ4
t=t4
0 X
Fig. 4.1. An uniform cross-sectioned bar is subjected to sudden constant velocity pulling at both
ends: (up left) different zones on X–t plane; (up right) elastic, linear hardening plastic stress–strain
curve; (low left) stress profiles at different times marked in X–t plane; (low right) the corresponding
σ –v states and ϕ–v states of different zones in X–t plane.
σ3 = σ1 + ρ0 C1 (v3 − v1 )
σ4 = σ2 − ρ0 C1 (v4 − v2 )
In the following, we will discuss the situation after the two waves meet. As the elastic wave
is faster than the plastic wave (C0 > C1 ), the two elastic precursor waves first meet at the
time ta at the point a. Since the stress level of both elastic precursor waves has reached
the elastic limit Y , the reflected loading waves after the interaction must be plastic waves.
According to the momentum conservative condition at the wave front Eq. (2.57),
[σ ] = −ρ0 D [v]
on the right side of the internal impact point a, we have:
σ5 − σ1 σ − Y
v5 = v1 − = vY − 5
ρ0 C1 ρ0 C1
Similarly, on the left side of the internal impact point, we have:
σ5 − σ2 σ − Y
v5 = v2 + = −vY + 5
ρ0 C1 ρ0 C1
Moreover, the material velocity and the stress should be the same on this boundary, namely:
v5 = v5 = v5
σ5 = σ5 = σ5
From the above four equations we get:

C1
v5 = 0, σ5 = Y + ρ0 C1 vY = 1 + Y
C0
After this, the rightward internal reflective plastic wave ab will meet the leftward incident
plastic wave lb at the point b, and the leftward internal reflective plastic wave ac will meet
the rightward incident wave oc at the point c. These wave-interactions all result in further
plastic loading. The secondary internal reflective plastic waves, bd and cd, will finally meet
at the point d. Only by this time the complete interaction between the two elastic–plastic
loading waves finishes. The states of zone 6, 7, and 8 can be determined, respectively,
by the approach similar to that applied to determine the state of zone 5. That is, each time
when two waves meet, two stress-velocity relationships can be obtained according to the
momentum conservative conditions [Eq. (2.57)] at both sides of the meeting point. Then,
by using the condition that both the particle velocity and the stress should be equal at the
meeting interface, the problem can be solved.
Figure 4.1 gives the wave profiles of stress distribution at four typical times: t1 , t2 , t3 , and
t4 , respectively. These wave profiles vividly depict the course of the loading interaction
between two strong-discontinuous elastic–plastic waves. In addition to the σ –v diagram,
the corresponding ϕ–v diagram is also given in Fig. 4.1. By introducing the quantity ϕ
[Eq. (2.37)], the nonlinear compatibility relationship along the characteristics lines is now
linearlized.
4.1.2 Head-on loading interaction of two weak-discontinuous stress waves
For the decreasingly

√ hardening plastic material (d 2 σ /dε 2 < 0), the plastic wave speed
C(ε) = (1/ρ)(dσ/dε) is not a constant but decreases with increasing stress. Therefore
the plastic wave propagates in the form of continuous wave. If we view the continu-
ous wave as a series of discontinuous incremental waves, then the loading interaction
between two continuous elastic–plastic waves can be imagined as the interactions of
many discontinuous incremental waves. Each time when two incremental waves meet,
both the rightward and the leftward internal reflective wave result to a further loading,
and in the meantime the wave speed decreases further, just as the situation previously
discussed at the point a in Fig. 4.1. Therefore, it is imaginable that in X–t plane, the
characteristics lines, which represent a series of interaction between two simple con-
tinuous
elastic–plastic
waves, are a set of curves with decreasing, absolute value of its

slope dX
dt .
Now, the problem is solved in the same way as that used to solve the discontinuous
wave interaction. The only difference is that the momentum conservative condition for
continuous waves, Eq. (2.63), should be used instead of the momentum conservative
condition for discontinuous waves, Eq. (2.57).
Take a decreasingly hardening bar with finite length as an example. The initial state of
the bar is: velocity va (>0), stress σa (>0). Assume that the right end of the bar is loaded
by a gradually increasing impact velocity up to vb (vb > va ) and then is held on at a
constant velocity; while the left end of the bar is loaded by a gradually increasing impact
velocity up to vc (vc < va ) and then is held on at a constant velocity. Therefore, two weak-
discontinuous elastic–plastic tensile waves are propagating within the bar in the opposite
direction. Before they meet, both waves are simple waves. The stress-velocity conditions
after the wave fronts are respectively:
σb⎫
dσ ⎪
vb = va + ⎪
ρ C⎬
σaσc 0 (4.1)
dσ ⎪⎪
vc = va − ⎭
σa ρ 0C
Assume that the leftward simple wave and the rightward simple wave begin to meet at
the point a, and the interactions finish at the point d, as shown in Fig. 4.2. The state
variables vd and σd can be determined as follows. According to Eq. (2.63)
dσ = ∓ρ0 Cdv
at the right side of the bar we have:
σd dσ
vd = vb −
σb ρ0 C
t ϕ
d
d
b
c b c
S
R Q S
a
R Q
a
0 X 0 v
Fig. 4.2. Interaction of two sets of weak-discontinuous elastic-plastic tensile waves propagating in
a bar in the opposite direction: (left) X–t figure showing different zones; (right) the physical states
of different zones on ϕ–v plane.
while at the left side of the bar we have:

σd dσ
vd = vc +
σc ρ0 C
Moreover, at the interface (i.e. at the point d), the continuation conditions for stress and
velocity should be held: vd = vd = vd , σd = σd = σd . Thus, the state variables vd and
σd are solved.
Introducing a substitution variable ϕ [defined in Eq. (2.37)], the two stress-velocity

conditions listed above are rewritten as:

vd = vb + ϕb − ϕd
(4.2)
vd = vc − ϕc + ϕd
while Eq. (4.1) is rewritten as:

vb = va + ϕb − ϕa
(4.3)
vc = va − ϕc + ϕa
The variable ϕd can be eliminated from the two equations of Eq. 4.2; and similarly the
variable ϕa can be eliminated from the two equations of Eq. 4.3. Then two equations
that contain the variable (φb –φc ) are obtained. Eliminating this variable again we get an
equation that only contains the variable vd . By similar approach, we can get an equation
that only contains the variable ϕd . The final forms of these two equations are:

vd = vc + vb − va
(4.4)
ϕd = ϕc + ϕb − ϕa
which can be further rewritten as:

(vd − va ) = (vc − va ) + (vb − va )
(ϕd − ϕa ) = (ϕc − ϕa ) + (ϕb − ϕa )
As can be seen, the above equations are actually a generalization of Eq. (3.2) from the case
of elastic waves to the case of elastic–plastic waves. The only difference is that the variable
σ is substituted by the variable ϕ. In other words, if we use the variable ϕ instead of the
old variable σ , then the superposition principle is still applicable. In fact, Eq. (3.2) is only
a special case of Eq. (4.4) under the conditions that va = ϕa = σa = 0 and C(ε) = C0 .
The above discussions are applicable not only to the determination of the state variables
in zone d, where the interactions of two weak-discontinuous elastic–plastic waves fin-
ished completely. They can also be applied to determine the state variables at any point
of the region abcd, representing a specified point of the interaction process. Actually,
the location of point d can be determined only after the state variables and the locations
of all points within the interaction zone have been known. The reason is that in case of
elastic–plastic wave interaction, the tangent slope of a characteristic line at any point of
X–t plane, namely the characteristic line direction, is a function of the state variable at
this point [dX/dt = ±C(ε)]. Unless the state variables at the points along the character-
istic lines bd and cd have been determined, the locations of the characteristic lines and
consequently their intersection point d are impossible to be determined. In the previous
discussions, we only assume the existence of the point d, but its location has not been deter-
mined yet. Compared to a linear wave problem, this reflects the complexity of a nonlinear
wave problem. Nevertheless, since the state variables of any point along the characteristic
lines ab and ac, and therefore its location, are known, we can use the finite differential
approach, where the differential algorithm is performed along the characteristic lines, to
solve the problem approximately. In practice, according to the accuracy requirement, the
characteristic lines ab and ac can be divided into m and n segments respectively (see
Fig. 4.2). The interaction region is then divided by the leftward characteristic lines starting
at the segmental points along ab line and the rightward characteristic lines starting at the
segmental points along ac line into m × n meshes. Within each mesh, the state variables,
such as the particle velocity and the stress–strain, can be regarded as uniform. There-
fore, the slope of the characteristic line segment QS can be approximately determined by
the state of point Q, and similarly the slope of a characteristic line segment RS can be
approximately determined by the state of point R. In this way, the algebraic equations:
XS − XQ = −C(εQ )(tS − tQ )
XS − XR = +C(εR )(tS − tR )
are used instead of the differential equations along the characteristic lines. Therefore the
location of point S, which is the intersection of the line segments QS and RS, is determined
(XS , tS ). The state variables of point S can be calculated from the state variables of points a,
Q, and R, according to Eq. (4.4):
vS = vQ + vR − va
ϕS = ϕQ + ϕR − ϕa
Repeating the same approach we can determine the locations and the state variables at all
mesh points.
In a problem like the previous one, if the state variables v and σ (or ε) along two char-
acteristic lines of the different sets are given, then within the area bounded by these two
characteristic lines and another two characteristic lines starting from the other ends of these
two characteristics lines, there exists a single-valued solution. Such a type of problem is
called a Darboux problem, or a characteristic line boundary value problem. Therefore, the
problem to solve the loading interaction between two weak-discontinuous elastic–plastic
waves is actually attributed to solve a Darboux problem.
Now we have discussed three types of solution-determined problem: the Cauchy problem
(the initial value problem), the Picard problem (the mixed initial-boundary value problem),
and the Darboux problem (the characteristic line boundary value problem).
4.2 Reflection of Longitudinal Elastic–Plastic Loading Waves at a Fixed End
For the problem shown in Fig. 4.2, if the two head-on elastic–plastic waves have the same
stress magnitude, i.e. σc = σb (ϕc = ϕb ), then from Eqs. (4.2) and (4.3) we have:
vd = va
This is equivalent to the case when an elastic–plastic wave reflects from a rigid wall
(a special case is va = 0). After wave reflection, the change of the ϕ value (D ϕ)R is the
two times of the incident ϕ value change (D ϕ)I , namely:
(∆ϕ)R = ϕd − ϕa = 2(ϕb − ϕa ) = 2(∆ϕ)I (4.5)
We know that the stress level doubles when an elastic wave reflects from a rigid wall.
Actually this is only a special case of Eq. (4.5) when C = C0 .
As a typical example of elastic–plastic wave interaction including fixed end reflection, we

consider a decreasingly hardening bar with finite length. The left end of the bar (X = 0)
is fixed. At time t = 0, the right end of the bar (X = L) is suddenly loaded by a constant
velocity v∗ (>0). The elastic–plastic waves will be reflected back-and-forth between the
fixed end and the impact end, and gradually the wave intensity increases. This process is
shown in Fig. 4.3 (von Karman et al., 1942), wherein the characteristic lines are shown not
only in the (X, t) plane and the (ε, t) plane, but also in (φ, v) plane. As stated previously,
in the (φ, v) plane, all characteristic lines are straight lines with an angle ±45◦ to the axes
of coordinates.
Before the stress wave reaches the fixed end (t < L/C0 ), the situation is the same as
that in an infinite bar loaded by a sudden constant velocity impact. A beam of central
simple waves, pioneered by a strong-discontinuous elastic wave, emit from the impact end.
Assume that the bar is at a stress-free, rest state before impact (corresponding to the point 0
on the state plane). Across the precursor elastic wave front LA, the stress, strain, particle
velocity, and ϕ value jump from zero to Y (the elastic limit), εY (= Y /E), vY (= Y /ρ0 C0 ),
ε ϕ t
IX IX
IX VIII
VIII
S S
VIII
Q Q
S VI VI VII'
P R
K N VII P
V'
VI Q VII V R VII
R N
P N M III''
K H
III IV V
V G
DE M
K III' H
M
III H II 4 E
IV G II'
G 3 III
II D E A' 2 II 4 D
A' 3
4 φY 1 A 1 2 3 1'
εY A12 A 0
0 vY v* v 0 v* v 0 L X
FRACTURE
ε
0.05
485µs
450µs
0.04
µs
400
µs
27
35
375
0µ
0µ
s
0.03
s
0µ
0µ
s 30
20 25
0µ
s
0.02 12
13 0µs
0µ
s
s
1µ
15
11
0µ
s
10
0µ
s
0.01 90µs
50
µs
70µs
µs
10µs
30
0 5 10 15 20 X (cm)
Fig. 4.3. Wave propagation in a finite bar of decreasingly-hardening material, the left end of bar
(X = 0) is fixed, and the right end of the bar X = L is suddenly loaded by a constant velocity
v∗ (>0).
and φY (= Y /ρ0 C0 ), respectively. Accordingly, on the state plane, the state point jumps
from the point 0 to the point A (the characteristic line LA on the physical plane corresponds
to the characteristic line OA on the state plane). After this discontinuous elastic wave front,
the continuous plastic waves come. On the physical plane (X–t plane) it is the simple wave
region I bounded by the curves LA and LD. On the state plane (v–ε plane) it corresponds
to the line segment AD. The point D is determined by the boundary condition v = v∗ . The
constant value region I on the X–t plane corresponds to the point E on the state plane.
The point E is identical to the point D.
At the time t = L/C0 , the elastic wave with the stress level of Y reflects at the fixed end.
As mentioned before, this reflection wave must be a kind of loading plastic wave. Since
the material is decreasing hardening as assumed, the incident strong-discontinuous elastic
wave in fact reflects into a beam of weak-discontinuous plastic wave. On state plane this
reflection wave region is represented by the line segment AA . The state variable at point A
is determined by the boundary condition at the fixed end [Eq. (4.5)], namely, ϕA = 2ϕY .
Therefore at the fixed end of the bar (X = 0), at time t = L/C0 , a beam of central
plastic wave should be reflected. Nevertheless, from the very beginning of reflection, those
rightward reflection waves meet the leftward incident plastic waves. Therefore solving
the region II on X–t plane is reduced to a problem that two weak-discontinuous plastic
wave head-on interact and load each other further. As aforementioned, this is a solution-
determined Darboux problem. In fact, the slope of the characteristic line at an arbitrary
point P along the reflective characteristic line AD on X–t plane, with regard to the absolute
value, should be equal to the slope of the characteristic line LP, namely,
dX L−X
=
dt t
Integrating the above function, and using the boundary condition of X = 0, t = L/C0 , we
can find that the characteristic line AD is a segment of a hyperbolic curve expressed by:
t (L − X) = L2 /C0
Therefore, the locations of the characteristic line AD and the state variables at these loca-
tions are known. Correspondingly, on the state plane, the boundary characteristic lines of
Region II, the line AD and the line AA in different sets, are known. Thus the region II can
be solved as a characteristic line boundary value problem (Darboux problem).
After the interaction between the reflective plastic wave tail (AG) and the incident plastic
wave tail (LDG) is completed, the reflective plastic waves will no longer interact with any
other head-on plastic waves and then propagate in the constant value zone I . Therefore,
the region II on X–t plane is a simple wave zone. From this, it is easy to understand that
a simple wave zone is always neighboring a constant value zone.
Since the region II has been solved, one boundary of the region III – the characteristic
line AG is known. Another boundary of the region III, the time axis AK, which is not a
characteristic line, is prescribed by the boundary condition v = 0 (the fixed end boundary
condition). Therefore, the solution of the region III is attributed to solve as a Picard problem
(mixed boundary problem).
On the analogy of these approaches, the solutions of the other regions on the X–t plane
can be attributed to solve either as a Darboux problem, or as a Picard problem, until the
whole field is solved. In the X–t plane shown in Fig. 4.3, three types of regions exist as
listed below:
(a) The constant value regions, which are marked by the superscript ( ). For such
regions, the positive and negative characteristic lines on the X–t plane are all
straight lines, while on the state plane they are correspondingly reduced to a point.
(b) The simple wave regions, which are marked by the superscript ( ). They are always
neighboring a constant value region. On the state plane such a region is reduced to
a line segment. If this line segment is positive, then on the X–t plane the negative
characteristic lines are straight lines, while the other set of the characteristic lines
are curves.
(c) The complex wave regions, which are not marked with any superscript. For such
regions, a one-to-one area correspondence exists on the physical plane and the state
plane. If one boundary of such a region is the t axis, then this region is attributed
to solve as a mixed boundary problem. The regions III, IV, VI, VII, and IX are all
of this type. Otherwise the problem is attributed to solve as a characteristic line
boundary problem. The regions II, V, VIII are all of this type.
In principle, the exact solution can be found for this problem. However, in practice,
because of the complicated function of σ = σ (ε), the problem is usually solved by
a numerical approach. An example solved by the characteristics numerical approach is
given in Fig. 4.3(d), showing the calculated strain wave profiles along the bar at different
times. The calculation indicates that at t = 485 µs, the fracture happens at a point near
the impact end of the bar, because the ultimate tensile strength is reached. Note that a
secondary fracture will happen at another point near the fixed end before the unloading
wave due to the first fracture arrives at the fixed end of the bar.
4.3 Governing Equations and Characteristic Lines of Unloading Waves
We now start to discuss the unloading problem of elastic–plastic stress wave interaction.
It is well known that the stress–strain curves in the loading process and in the unloading
process are the same for an elastic (or even for a nonlinear elastic) material. However,
the stress–strain relationship of an elastic–plastic material differs for the loading and the
unloading processes, and consequently the governing equations for the loading waves and
the unloading waves in an elastic–plastic material are also different.
From the phenomenological point of view, the stress wave propagation in an elastic–plastic
material can be treated as the same as the stress wave propagation in a nonlinear elastic
material, if only the loading process is concerned. In fact, without knowing the unloading
stress–strain curve of a material, it is impossible to distinguish an elastic–plastic material
from a nonlinearly elastic material. In other words, in a loading process, the stress–strain
curve for an elastic–plastic material can be mathematically regarded as that for a nonlinear
elastic material. Consequently, the loading wave propagation for the two materials can
be mathematically regarded as the same. Only when the unloading happens, the two
materials behave in different ways: the elastic–plastic material exhibits irreversible plastic
deformation, while the elastic material exhibits reversible elastic deformation. In this
sense, the essential features of plastic wave propagation, which distinguishes from the
elastic wave propagation, are mainly reflected in the unloading process.
Therefore, when we analyze an elastic–plastic wave problem including both the loading
and unloading processes, it is necessary to identify what the state is for any arbitrary
material particle at any time. Is it in a loading process or in an unloading process? In other
words, on the X–t plane, it is necessary to distinguish between the plastic loading zone
and the unloading zone, and to determine the unknown boundary between these two zones.
Note that the location of such a boundary, X, changes with the propagation and interaction
of stress waves, and so changes with time t. Thus, the problem can now be expressed
as: to determine the propagation of the boundary X = f (t) between the plastic loading
zone and the unloading zone, and meanwhile to solve the plastic loading zone and the
unloading zone, which are bounded by the X = f (t), by different governing equations.
Mathematically, it is a kind of problem with undetermined internal boundaries. This makes
the problem much more complicated.
The governing equations in a plastic loading zone have been stated previously [see
Eq. (2.18)]. Now, we need to discuss the governing equations in an unloading zone. Similar
to the case for loading waves, the governing equations for unloading waves should be com-
posed of the kinematics equation, the dynamics equation and the constitutive equation.
The basic assumptions applied are the same as for the loading waves in bars too. In the
following, we still use Lagrange variables to describe those governing equations.
Consider the unloading stress–strain curve of an elastic–plastic material that has

experienced plastic loading. If before unloading, the material is in a state with stress
σm and strain εm , as shown in Fig. 4.4. No matter whether the material is unloaded or
reloaded again after previous unloading, provided that the stress does not go over σm ,
we assume that the unloading stress–strain curve is a linear line, and its slope is same
as the initial elastic modulus. In other words, the unloading and reloading (below σm )
curve obeys the same Hooke’s law for the elastic loading (Fig. 4.4). This is the so-called
elastic unloading hypothesis. Note that when the material is unloaded from σm and then
reloaded again, the yield limit of the material is raised to σm , representing the so-called
work-hardening effect or the strain-hardening effect.
Using an overbar to indicate the variables after unloading, the elastic unloading stress–
strain relationship for material, in the one-dimensional stress state, is:
σ̄ = σm + E(ε̄ − εm ) (4.6)
The total strain ε can be viewed as the sum of the elastic strain ε e and the plastic strain ε p .
The elastic strain εe is recoverable, and satisfies Hooke’s law, ε e = σ/E. The plastic strain
εp is unrecoverable.
Note that, generally speaking, during the unloading process, the stress value σm may be
different for different points along the bar. However, once the unloading process begins, as
long as plastic re-yielding does not occur again, the value of σm will not change with time.
(dsm /dX)dX
σ σ
b
(σm, εm)
E(dem/dX )dX
a
G
G
D IN g
DIN
LOA
E(de /dX)dX
OA
f e
-L
Y
ING
RE
h
AD
σ dX
LO
X
i j
UN
E E E
0 ε 0 ε
σ dX
X
Fig. 4.4. Elastic-plastic loading and elastic unloading stress–strain relationship.
Therefore, in an unloading region, σm and εm are functions of X and are independent of

the variable t. We then get:
∂ σ̄ ∂ ε̄ ∂ ε̄
=E = ρ0 C02
∂t ∂t ∂t
∂ σ̄ ∂ ε̄ d ∂ ε̄ dσm dεm
=E + (σm − Eεm ) = ρ0 C02 + − ρ0 C02
∂X ∂X dX ∂X dX dX
If at time t, the stresses before unloading, σm , are denoted by the symbol a and b in
Fig. 4.4(b) for the point X and X + dX, respectively, the unloading stress σ̄ for the point
X and X + dX are denoted by the symbol i and h in Fig. 4.4(b), respectively, then each
term in the above equation for ∂ σ̄ /∂X is illustrated in the same figure, showing its physical
meaning respectively.
The kinematics equation and the dynamics equation for the unloading waves are the same
as those for the loading waves, namely Eqs. (2.12) and (2.13). Those equations, together
with the unloading stress–strain relationship, constitute the governing equations of the
unloading waves:
⎫
∂ ε̄ ∂ v̄ ⎪
= ⎪
⎪
∂t ∂X ⎪
⎪
⎬
∂ v̄ ∂ σ̄ (4.7)
ρ0 = ⎪
⎪
∂t ∂X ⎪
⎪
⎪
⎭
σ̄ = σm + E(ε̄ − εm )
We can eliminate the variable σ̄ from the above equations, getting two first order partial
differential equations with regard to the variables ε̄ and v̄:
⎫
∂ v̄ ∂ ε̄ ⎪
= ⎪
⎬
∂X ∂t
(4.8)
∂ v̄ ∂ ε̄ 1 dσm dεm ⎪
⎪
= C02 + − C02 ⎭
∂t ∂X ρ0 dX dX
Or, we can eliminate the variable ε̄ from the Eq. (4.7), getting two first order, homogeneous
partial differential equations with regard to the variables σ̄ and v̄:
⎫
∂ σ̄ ∂ v̄ ⎪
= ρ0 C02 ⎪ ⎬
∂t ∂X
(4.9)
∂ σ̄ ∂ v̄ ⎪ ⎪
= ρ0 ⎭
∂X ∂t
Moreover, we can get an equivalent second order partial differential equation with regard
to the unknown displacement ū:
∂ 2 ū ∂ 2 ū 1 dσm dεm
2
= C02 2 + − C02 (4.10)
∂t ∂X ρ0 dX dX
The characteristic line method can be used to solve these equations. Using the procedures
similar to that stated in Section 2.1, the characteristic line equations and the compatibility
equations along the characteristic lines can be obtained without any difficulty. For example,
corresponding to Eq. (4.8), we have:
⎫
dX = ±C
0 dt ⎬
1 (4.11)
d v̄ = ± C0 d ε̄ + dσm − C0 dεm ⎭
ρ0 C0
And corresponding to Eq. (4.9), we have:

⎫
dX = ±C0 dt ⎪
⎬
1 (4.12)
d v̄ = ± d σ̄ ⎪
⎭
ρ0 C0
The above equations are equivalent to each other. Among them the governing equations
with regard to the variables σ̄ and v̄ [Eq. (4.9)], and their corresponding characteristic line
equations [Eq. (4.12)], are completely identical to those for the elastic waves. It is also
more convenient to use this set of equations to solve the problem with wave interactions,
as will be shown below. Therefore we tend to use this set of equations in the following
discussions.
From Eq. (4.11) or Eq. (4.12) we also know that: the unloading disturbance propagates
with the elastic wave speed C0 . This is an inevitable outcome resulted from the elastic
unloading hypothesis. It should be emphasized that the propagation of unloading distur-
bance and the propagation of the boundary of plastic-loading/elastic-unloading are two
different concepts. The former is the unloading wave, which propagates along the elastic
characteristic line. Under the elastic unloading hypothesis, the speed of the unloading wave
is always C0 . On the other hand, the latter is the spread or propagation of the so-called
elastic-plastic boundary. Generally the elastic–plastic boundary propagates neither along
the elastic characteristic line nor along the plastic characteristic line. Its propagation speed
can only be determined by a detailed analysis of each case. Only under some specific
conditions, the propagation of an elastic–plastic boundary coincides with the propagation
of an unloading wave.
On the X–t plane, if the curve X = f (t) presents the elastic–plastic boundary during
an unloading process, which will be shortly called unloading boundary below, for
convenience, then the slope of this curve:

= dX
C = f (t) (4.13)
dt X=f (t)
represents the propagation speed of this boundary. Note that there were literatures where
this elastic–plastic boundary propagation had been called “unloading wave”, or “Rakhmat-
ulin wave”. Since it may induce confusion in the basic concept, it will not be accepted in
our discussion.
Thus, for an elastic–plastic wave problem including both the plastic loading and the elastic
unloading processes, the problem is reduced to: under the prescribed initial condition and
a boundary condition, simultaneously solve the partial differential equations in an elastic–
plastic loading zone (Eq. (2.18) or other equivalences) and the partial differential equations
in an elastic unloading zone (Eq. (4.10) or other equivalences), and meanwhile on the
boundary of these two zones, X = f (t), the conditions of the displacement continuation
and the momentum conservation must be satisfied. The main difficulty lies in that generally
the elastic–plastic boundary X = f (t) is not known before the whole problem has been
solved.
In the next few sections, we will discuss the problem of unloading wave propagation, first
in a semi-infinite bar. The end of the bar is loaded to the elastic–plastic state, and then
is unloaded. Because the unloading disturbance propagates faster than the plastic loading
disturbance (C0 > C), the unloading disturbances will pursue the proceeding plastic
loading disturbances and then interact each other. This kind of wave interaction is called
a pursuing unloading problem. Afterwards, we will discuss the problem of unloading
wave propagation in a finite bar, wherein an unloading disturbance propagates from one
end, and interacts with a head-on plastic loading disturbance propagated from the other
end, resulting in a head-on unloading. Such kind of wave interaction is called a head-
on unloading problem. Finally, we will discuss the general properties of elastic–plastic
boundary propagation.
4.4 Pursuing Unloading by Strong-Discontinuous Unloading Disturbances
4.4.1 Sudden unloading of strong-discontinuous loading wave in a linear hardening bar
Consider a semi-infinite bar made of linear hardening material. The bar is in a natural, rest
state before loading is applied. At time t = 0, a sudden constant value impact loading σ ∗
is applied on the end of the bar. The impact loading is large enough to cause the bar to
deform plastically. After time t1 the applied load is suddenly released to 0. This problem
is equivalent to the case that two bars impact suddenly and then bounce back suddenly.
In such a case, both the loading disturbance and the unloading disturbance propagate
within the bar in the form of strong-discontinuity. Since the unloading disturbance is
faster than the plastic loading disturbance, it will pursue and interact with the loading
disturbance.
When t < t1 , namely, for the region 1 and region 2 on the X–t plane shown in Fig. 4.5(a),
it is equivalent to solving the elastic–plastic loading wave propagation in a semi-infinite
bar as has been discussed previously, so it is easy to get:
Y Y
σ1 = −Y, ε1 = −εY = − , v1 = = vY
E ρ0 C0

∗ σ∗ 1 1
σ2 = σ , ε2 = + − Y
E1 E1 E

1 1 σ∗
v2 = − Y− = v∗
ρ0 C0 ρ0 C1 ρ0 C1
where√
Y and εY are the
√ yield stress and the yield strain of the material (absolute values).
C0 = E/ρ0 , C1 = E1 /ρ0 , and E1 is the linear hardening modulus [Fig. 4.5(c)].
At t = t1 , the plastic wave front arrives at the location l1 = C1 t1 , while the stress at the
impact end is suddenly decreased to 0, which means that the end of the bar becomes a free
end. So a strong-discontinuous unloading disturbance is propagating in the plastic constant
value region 2, with speed C0 (> C1 )1 . This unloading wave pursues the forerunning plastic
loading disturbance, which propagates with the velocity C1 . The state of the unloading
region 3 can be determined by the prescribed unloading boundary condition, the unloading
stress–strain relationship [Eq. (4.6)], and the dynamic compatibility condition across a
strong-discontinuous wave front [Eq. (2.57)], namely:
⎫
σ̄3 = 0 ⎪
⎪
⎪
σ2 ⎬
ε̄3 = ε2 − = ε̄∗ (4.14)
E ⎪
σ2 σ∗ ⎪
⎪
v̄3 = v2 + = v∗ + = v̄∗ ⎭
ρ0 C0 ρ0 C0
1As discussed in Section 4.3, under the elastic unloading hypothesis, the unloading disturbance propagates by
the elastic wave velocity C0 . This is true no matter whether the unloading disturbance is weak-discontinuous or
strong-discontinuous. Actually when Eq. (4.6) is substituted into Eq. (2.59) we get D = C0 .
v v* v*
0 5 Y
3 v
t
-y 4
1
σ∗
2
5" 5'
(b)
t2' σ
t2 4" 4'
a ε∗ 0
t1' 3
5" 3 E 5' ε
t2
2 X= C 0t
1t
C
1
X=
4'
4" σ4
0 l1 l1' X E1 1 Y
2
(a) (c)
Fig. 4.5. Sudden unloading of strong discontinuous loading wave in a linear-hardening

bar (|v̄ ∗ | < 2|vY |).
As can be seen that although the stress σ̄3 is unloaded to zero, the material still has residual
strain ε̄∗ and the residual particle velocity v̄∗ , which are shown in Fig. 4.5(c and b),
respectively. This feature reflects the irreversible property of plastic deformation. Note
that the concept of residual velocity is important in understanding the internal impact phe-
nomenon when an unloading disturbance has caught up with a plastic loading disturbance.
The internal impact phenomenon will be discussed below.
Assume that at time t = t the unloading disturbance catches up with the plastic loading
disturbance, the value of t can be determined by the condition that on the X–t plane the
line t1 a intersects with the line oa, which gives:
C0 (t1 − t1 ) = C1 t1
Then we have:
C 0 t1 t1
t1 = = (4.15a)
C0 − C 1 1 − C1 /C0
and the location where the two waves interact with each other is:
C1 t1 l1
l1 = C1 t1 = = (4.15b)
1 − C1 /C0 1 − C1 /C0
Therefore, while X < l1 , t < t1 , the unloading disturbance propagates in the plastic con-
stant value zone 2, and behind the strong-discontinuous front of the propagating unloading
disturbance, the stress σ̄3 is unloaded to zero, but the particle velocity v̄3 is not unloaded
to zero, although this residual velocity v̄3 = v∗ is uniformly distributed along the bar seg-
ment X ≤ l1 . However, once the unloading disturbance catches up with the forerunning
plastic loading disturbance, let us imagine that the unloading disturbance now propagates
in the bar segment X > l1 , which is the elastic constant value zone 1, so that behind the
unloading wave front both the stress and the particle velocity are now unloaded to 0. Thus,
across the bar section X = l1 , a particle velocity jump (v̄3 − 0) occurs. In other words, an
internal impact should take place at X = l1 . Therefore, when the unloading disturbance
catches up with the plastic loading disturbance, the interaction between those two waves
will induce the so-called secondary stress waves or the internal reflective waves.
After the interaction of the unloading wave and the plastic loading wave, the concrete state
variables will be determined depending on the magnitude of residual velocity difference
(v̄3 − 0) across the section point X = l1 . It is similar to a case that a bar traveling with the
speed v̄3 longitudinally impacts on a bar at rest. The states of the bars just after the impact
will be determined depending on the level of v̄3 . From our previous discussions about the
elastic–plastic bar impact [e.g. refer to Eq. (3.1)], if the condition
|v̄3 | < |2vY | (4.16)
is satisfied, then the level of the secondary stress wave is less than the yield limit Y . On the
contrary, when |v̄3 | > |2vY |, the secondary stress wave will include plastic wave.
Figure 4.5 shows the case of |v̄3 | < |2vY |. The state variables in region 4 can be determined
as follows. At the left side and the right side of the section point X = l1 , according to the
dynamic compatibility condition across a strong-discontinuous wave front [Eq. (2.57)],
we have:
σ4 = σ̄3 + ρ0 C0 (v̄4 − v̄3 ) = ρ0 C0 (v̄4 − v̄3 )

σ4 = σ1 − ρ0 C0 (v̄4 − v1 ) = −ρ0 C0 v̄4
Moreover, the stress and particle velocity should be continuous at the point X = l1 :
σ̄4 = σ̄4 = σ4 , v̄4 = v̄4 = v4
Therefore the following results can be obtained:

v̄3
v̄4 =
2
v̄3
σ̄4 = −ρ0 C0
2
However, from Fig. 4.5 we note that:
σ̄4
ε̄4 =
E
σ̄4
ε̄4 = ε̄3 + = ε̄3 + ε̄4
E
Therefore, at the point X = l1 , the strain is discontinuous. This strong-discontinuity of

strain will retain at the point X = l1 , unless the so-called secondary plastic loading happens
in the future. Such kind of discontinuity is called a stationary discontinuous interface.
A stationary strain-discontinuous interface, originating from the interaction between a
strong-discontinuous unloading disturbance and a strong-discontinuous plastic-loading
disturbance, reflects the strain history difference at the two sides of the interaction point.
The difference of the plastic strain history causes the difference of the stress–strain rela-
tionships. Therefore, once a stationary strain-discontinuous interface appears in a uniform
bar, the mechanical properties (e.g. at least the yield stress) of the material will be different
at two sides of the interface, and the bar is no longer a uniform bar. Under certain conditions
like those shown in Fig. 4.5, suppose a stress wave with the stress level between σ1 and
σ2 (i.e. Y < |σ | < |σ2 |) propagates in the bar, when this stress wave passes through the
stationary strain-discontinuous interface l1 , transmitting and reflecting phenomena will
appear, just like a wave passing through an interface of different media. Note that the elas-
tic wave resistance, ρ0 C0 , is the same on two sides of the stationary strain-discontinuous
interface. So the propagation of an elastic wave with level |σ | < Y , or an unloading wave,
will not be influenced by the stationary strain-discontinuous interface. For example, in
the case shown in Fig. 4.5, the leftward reflective wave l1 t2 originating at the interface l1
will be reflected again from the free end at time t = t2 , the rightward reflective unloading
wave t2 b will pass through the interface l1 without any interference.
For the problem illustrated in Fig. 4.5, the distributions of the stress, strain, and the particle
velocity at several typical times are shown in Fig. 4.6. As can be seen, once the unloading
process appears in the propagation of plastic wave, the stress wave profile, the strain wave
profile and the particle velocity wave profile become different from each other.
It should be noted that in general the propagation of unloading boundary should be dis-
tinguished from the propagation of unloading disturbance itself. However, in the special
case discussed here, when the propagating locus of unloading boundary t1 a (Fig. 4.5) just
coincides with the propagating locus of strong-discontinuous unloading disturbance itself,
then the propagating velocity of unloading boundary C = C0 .
In the case shown in Figs. 4.5 and 4.6, the problem is relatively simple, because the
boundary impact loading is only high enough to result in a residual particle velocity as
small as (v̄∗ < 2vY ), and consequently the secondary stress wave σ4 created by the internal
impact is less than the yield limit (|σ4 | < Y ), which means that once the unloading wave
catches up with the forerunning plastic loading wave and interacts each other, the plastic
loading wave “disappears” completely.
If the impact loading σ ∗ , or correspondingly v∗ , is high enough to result in a residual

particle velocity v̄4∗ > 2vY , the situation becomes more complex. In this case, the secondary
stress wave originating from the internal impact has a stress level beyond the yield limit Y .
Therefore the secondary stress wave propagating to the right side of the stationary strain-
discontinuous interface l1 is a strong-discontinuous plastic wave, although the stress
level is lower than the initial plastic wave. This means that the initial plastic wave, after
interacting with the unloading disturbance, is weakened. On the other side, the secondary
stress wave propagating to the left side of the stationary strain-discontinuous interface l1
is an elastic wave propagating in the unloading region, because its stress level is lower
ν
2 ν2
= 1 2
1 1 ν∗
1 1
1 ν1
0 0 0
(a)
2 ν ν2
2 3 1 ν3
1 0 1
1' > > 1 1 1 0
ν1
0 0 0 0
(b)
ν
3 ν3
= 1' 1 1 ν1
(c) 3=0
l' l' l'
0 4∋∋ ν
3 ν3
2> > 1' ν1
1 ∋ 1 ν4
4 4
0 0 0 0
(d) 0
l' l' l'
4∋∋ ν
= 2 1 ∋ 1 ν1
4 4 ν4
0
(e)
l' l1' l1'
4∋∋ ν
1 0 1 ν1
2' > > 2
4 5∋∋= 3 4
∋ ν4
0
(f) 0 0
0 0
l' l' l'
5∋∋= 3 ν
> 2'
1 4∋ ν1
(g) 5=0 0 5 ∋= 0 0 ν5 = 0 0
l' l' l'
Fig. 4.6. The distributions of the σ , ε, and v at several typical times in Fig. 4.5.
than the initial plastic wave. At t = t2 this leftward wave reaches the bar end (which is
now a free end), and is reflected again as the secondary unloading disturbance. At the
point X = l2 , this strong-discontinuous, secondary unloading disturbance catches up with
the already weakened, forerunning plastic loading wave. This time a second stationary
strong-discontinuous strain interface is formed at l2 . The plastic loading wave propagating
to the right side of the interface l2 is weakened further.
On the analogy of this unloading process, in each unloading cycle a stationary strong-
discontinuous strain interface is formed and the rightward plastic loading wave is
weakened, until the forerunning plastic wave disappears, that is, unloaded to an elastic
wave. Clearly, higher the initial impact loading σ ∗ (or v∗ ) is, more will be the number of
stationary strong-discontinuous strain interfaces. Figure 4.7 shows an example where three
stationary strong-discontinuous strain interfaces are formed. The detailed explanation for
the processes of wave propagation and interaction in Fig. 4.7 is omitted here. Readers can
easily understand it by referring to the previous discussions for the case in Fig. 4.5. It is
t s
09 7 5 3
v
9
8
8
7 1
6 6
5 1 4
3 4
2
2
0
l1'l2'l3' X
sm
eR s
9(4) 3 9''' 9'' 9' 0 e
M1
M2
M3 s ~X
m 8(4) 8''' 8'' 8' s8
eR~X 6''' 6'' 6' 1 s6
4'' s4
4'
2
X
Fig. 4.7. Sudden unloading of strong discontinuous loading wave in a linear-hardening bar
(|v∗ | > 2|vY |).
recommended that the readers go through the deduction to acquire a better understand-
ing. Besides, it can be easily proved by the readers that if the following dimensionless
parameter β (>1) is introduced,
C0 + C1 1 + C1 /C0
β= = >1 (4.17)
C0 − C 1 1 − C1 /C0

then the length of the (n + 1)th stationary strain-discontinuous interfaces, ln+1 , is β times

the length of the nth stationary strain-discontinuous interfaces, ln ; and, the maximum

plastic stress experienced at point ln+1 , σ2(n+1) , is 1/β times the maximum plastic stress

at point ln , σ2n , that is:

ln+1 = βln
(n = 1, 2, 3, . . .)
σ2n
σ2(n+1) =
β
This means that when a stationary strain-discontinuity forms in each unloading cycle, the
maximum stress decreases by a factor of 1/β, while the distribution length of the maximum
stress increases by a factor of β. Thus the points M1 , M2 , M3 , . . . on the plot of maximum
stress distribution in Fig. 4.7 fall upon a hyperbolic curve expressed by:
σ2 l1 = σ4 l2 = . . . σ2n l2 = σ2(n+1) l(n+1)

= const.
In the same plot the distribution of the residual strain after impact, εR , is also shown.
As can be seen, the residual deformation is concentrated to a small region near the impact
end of the bar, and decreases step by step. The number of the steps is the number of the
stationary strain-discontinuities. This result is completely different from the case under
static loading.
4.4.2 Sudden unloading of continuous loading waves in a linear hardening bar
In the previous discussions, both the unloading disturbance and the plastic loading distur-
bance are strong-discontinuous waves. Now we consider a slightly complex case where the
unloading disturbance is a strong-discontinuous wave but the forerunning plastic loading
disturbances are a series of continuous waves. We need to deal with the wave interactions
between the strong-discontinuous unloading disturbance and a series of continuous plastic
loading disturbances.
To help understand the image of such a kind of wave interaction process, we start the
discussion on the basis of the example shown in Fig. 4.5 for a semi-infinite bar made of
linear hardening material, but with a boundary condition of gradually increasing impact
loading. That is, at time t = 0, the end of the bar is loaded by an increasing external
load, which is linearly changed with time, and at time t = T0 the external load is sud-
denly released [Fig. 4.8(a)]. In such a situation, the unloading disturbance propagates
as a strong-discontinuous wave, while the plastic loading disturbances propagate as a
series of weak-discontinuous waves but with the constant speed C1 . As in our previous
discussions, such continuous plastic waves can be regarded as a series of incremental
waves. Similarly, the linearly increasing boundary load can be regarded as a series of step-
like incremental boundary conditions [Fig. 4.8(b)]. Therefore, the interactions between
a strong-discontinuous unloading disturbance and a series of weak-discontinuous plastic
disturbances can be viewed as the interactions between a strong-discontinuous unload-
ing disturbance and a series of small strong-discontinuous plastic loading disturbances in
succession. Since the interaction between a strong-discontinuous unloading disturbance
and a strong-discontinuous plastic-loading disturbance has been dealt with previously, the
problem can be solved step by step as follows.
From the time t = T0 , the strong-discontinuous unloading disturbance propagates with

wave speed C0 in the constant-value zone 5. Behind the unloading wave front, the stress
drops from σ ∗ to zero, and the particle velocity drops from v∗ to the residual velocity
v̄∗ = v∗ + σ ∗ /(ρ0 C0 ). At the time t = T1 , the unloading disturbance catches up with the
first small-discontinuous plastic wave at the point X = L1 and interact with each other.
This interaction process can be understood as follows: since a particle velocity jump (∆v)1
and a stress jump (∆σ )1 exist across this small-discontinuous interface, if both stresses
are unloaded to zero, then a residual particle velocity jump (∆v̄)1 is produced:

(∆σ )1 (∆σ )1 (∆σ )1 (∆σ )1 1 1
(∆v̄)1 = (∆v)1 + =− + = − (4.18)
ρ0 C 0 ρ0 C1 ρ0 C0 ρ0 C0 C1
t t
T0
5
4
3
2
1
σ σ∗ -Y 0
X
(a)
t t
8 9
T3
7
T2
6
T1
5 T0 5
4 4
3 3
2 2
1
1 0
σ σ∗ -Y 0 L1 L2 L3 X
(b)
σ
(∆ v)1
6 a' U* b'
0 v
s7(∆s)1 7
1 8
9
-Y
2 a'' 3
σ* b'' 4 5 a
b
(c)
Fig. 4.8. Sudden unloading of continuous loading waves in a linear hardening bar.
Therefore, internal impact occurs. The stress jump (∆σ̄ )1 induced by this internal impact
is within the unloading region, thus can be calculated for the elastic waves as:

(∆v̄)1 (∆σ )1 C0 1
(∆σ̄ )1 = −ρ0 C0 = − 1 = − [(∆σ )1 + ρ0 C0 (∆v)1 ] (4.19)
2 2 C1 2
This is corresponding to the point 7 in the σ –v plane as shown in Fig. 4.8(c).

After the first internal impact, on the one hand, the intensity of the plastic wave is weakened
by the unloading disturbance, namely, its stress magnitude drops from |σ5 | to |σ4 | =
|σ5 −(∆σ )1 | = |σ ∗ −(∆σ )1 |. On the other hand, the intensity of the unloading disturbance
is weakened by the plastic wave, namely, its stress magnitude drops from |σ5 −σ̄6 | = |σ ∗ | to
|σ4 − σ̄7 | = |σ ∗ − (∆σ )1 − (∆σ̄ )1 | = |σ ∗ − [(∆σ )1 + (∆σ̄ )1 ]|.
Thus, the intensity of the strong-discontinuous unloading disturbance decreases by

|(∆σ )1 + (∆σ̄ )1 |.
At the time t = T2 , the unloading disturbance catches up with the next small-discontinuous
plastic wave and the second internal impact happens at the point X = L2 . Similar to
Eqs. (4.18) and (4.19), we have:

(∆σ )2 1 1
(∆v̄)2 = −
ρ0 C0 C1
1
(∆σ̄ )2 = − [(∆σ )2 + ρ0 C0 (∆v)2 ]
2
After the internal impact, the intensity of the secondary stress wave in unloading zone is
raised from |σ̄7 | to |σ̄8 | = |σ̄7 + (∆σ̄ )2 | = |(∆σ̄ )1 + (∆σ̄ )2 |, while the intensity of plastic
wave decreases from |σ4 | to |σ3 | = |σ4 − (∆σ )2 |. Therefore the strong-discontinuous
unloading disturbance is weakened to the level:
|σ3 − σ̄8 | = |σ ∗ − {(∆σ )1 + (∆σ )2 + (∆σ̄ )1 + (∆σ̄ )2 }|
Compared to the initial stress level σ ∗ , the intensity of the unloading disturbance drops by
|(∆σ )1 + (∆σ )2 + (∆σ̄ )1 + (∆σ̄ )2 |
On the analogy of those procedures, when the strong-discontinuous unloading disturbance

pursues the ith small-discontinuous plastic wave and the corresponding internal impact
occurs, the difference of the residual particle velocity:

(∆σ )i (∆σ )i 1 1
(∆v̄)i = (∆v)i + = − (4.20)
ρ 0 C0 ρ0 C0 C1
will induce a stress jump:

(∆σ )i C0 1
(∆σ̄ )i = − 1 = − [(∆σ )i + ρ0 C0 (∆v)i ] (4.21)
2 C1 2
so that the secondary stress in the unloading zone is reloaded to a level of:
.
i .
i
(∆σ )k C0 1 C0
σ̄ = (∆σ̄ )k = −1 = − 1 (σ ∗ − σ )
2 C1 2 C1
k=1 k=1 (4.22)
1 1
= (σ + ρ0 C0 v) − (σ ∗ + ρ0 C0 v∗ )
2 2
while the strong-discontinuous unloading disturbance is weakened to |σ − σ̄ |:
.
i .i
∗ (∆σ )k C0
∗
|σ − σ̄ | = σ − [(∆σ )k + (∆σ̄ )k ] = σ − +1
2 C1
k=1 k=1

1 C0 1 C0 (4.23)
= 1+ σ+ 1− σ∗
2 C1 2 C1
1 1
= (σ − ρ0 C0 v) + (σ ∗ + ρ0 C0 v∗ )
2 2
On the σ –v plane shown in Fig. 4.8(c), the intensity of the secondary stress wave, |σ |, is
raised with the internal impact in succession, as illustrated by that the state point moves
correspondingly from point 6 to point 7, 8, . . . in succession along a positive characteristic
line. Obviously, once |σ − σ̄ | = 0, the unloading strong-discontinuity vanishes. The
point 6 in Fig. 4.8(c) represents the case that the initial residual velocity v∗ is less than
2vY [Eq. (4.16)]. In such a case, the strong-discontinuous unloading disturbance can
pass through the whole plastic zone and is satisfied the condition of |σ2 − σ̄ | = Y −
|σ̄ | > 0. In other words, having interacted with all the plastic loading disturbances and
causing the plastic region to be unloaded, the unloading disturbance still retains a strong-
discontinuity. The final unloading disturbance becomes a strong-discontinuous unloading
wave propagating in the elastic loading zone (the point 9). In this case, the locus of the
unloading boundary propagation coincides with the locus of the unloading disturbance
propagation.
However, if a stronger load such as |ν ∗ | > 2vY impacts the bar, then after the strong-
discontinuous unloading disturbance interacts with only a part of the plastic loading
disturbances, its intensity |σ2 − σ̄ | will decrease to zero. Namely, the unloading distur-
bance will be “absorbed” somewhere in the plastic loading zone and cannot pass through
the whole plastic zone. It means that the first phase of unloading is finished. An example
is shown below:
Suppose that the point a in Fig. 4.8(c) presents the loading state of the plastic loading zone
just before the sudden unloading. The initial unloading state is represented by the point a
(v̄a > 2vY ). Along with the interactions between the strong-discontinuous unloading
disturbance and the plastic loading disturbances, the plastic wave intensity |σ | decreases
along the line aa , while the secondary stress wave intensity |σ̄ | increases along the line
a a . At the point a , the intensity of the unloading disturbance |σ2 − σ̄ | finally decreases
to zero, implying that the unloading disturbance is now “absorbed” by the plastic zone. The
plastic waves are only weakened but not yet vanished, and thus they continue to propagate
rightward with the maximum stress level represented by the point a .
Similarly, if the initial loading state is represented by the point b, then the initial unloading
state is represented by the point b , and the strong-discontinuous unloading disturbance
vanishes at the point b . From now on, the plastic waves continue to propagate rightwards
with the intensity of |σ | ≤ |σb |, and will be unloaded again only when the secondary
unloading waves arrive. In fact, at the first phase of unloading, the leftwards internal
reflective waves will be reflected again at the free end (X = 0) of the bar, forming
a set of rightwards “secondary unloading waves”. Once those secondary unloading waves
pursue the forerunning already-weakened plastic loading waves, they will be unloaded
secondary, the so-called secondary phase of unloading. Obviously, if the initial plastic
waves are strong enough, it is possible that the plastic waves be unloaded by the “third
phase unloading”, the “fourth phase unloading”, and so on.
In the above discussions, when the incremental value between the small-discontinuous
plastic loading waves approaches to infinitely small, those plastic disturbances are actually
reduced to continuous waves. Accordingly, when the finite jump value is substituted by
the infinite small increment, and the summation algorithm is substituted by integral, the
above equations should be still effective. In this way, the discussions toward the problem
shown in Fig. 4.8(b) are applied to the problem shown in Fig. 4.8(a). Actually, the X–t
plot and the σ –v plot are identical for both problems.
The above results can be extended conveniently to the problem where the material is
decreasingly hardening plastic, as first analyzed by White and Griffis for the unloading
waves (White and Griffis, 1942, 1948). In such cases, the plastic wave√speed is no longer
a constant value, but a function of strain or stress [Eq. (2.15)], C = (1/ρ0 )(dσ/dε) =
C(σ ), then the Eqs. (4.20)–(4.23) are converted to the following forms, respectively:

dσ 1 1 dσ
d v̄ = dv + = − (4.24)
ρ0 C0 C0 C ρ0

1 C0 dσ
d σ̄ = − (dσ + ρ0 C0 dv) = −1 (4.25)
2 C 2
1
σ ∗ C
1 1
− 1 dσ = (σ + ρ0 C0 v) − (σ ∗ + ρ0 C0 v∗ )
0
σ̄ = (4.26)
2 σ C 2 2
σ ∗
1 C0 1 σ C0
σ − σ̄ = 1+ dσ + 1− dσ
2 0 C 2 0 C
(4.27)
1 1
= (σ − ρ0 C0 v) + (σ ∗ + ρ0 C0 v∗ )
2 2
As can be seen, the final forms for σ̄ and (σ − σ̄ ) do not change.
4.4.3 Sudden unloading of centered plastic loading waves
In the following, we consider an example where a strong-discontinuous unloading distur-

bance pursues a set of weak-discontinuous plastic loading disturbances in a semi-infinite
bar, which is made of decreasingly hardening material. The bar is in a natural, rest state
before the impact load is applied. At the time t = 0, a sudden constant-value impact
loading σ ∗ is applied on the end of the bar (X = 0), while at the time t = T the
applied load is suddenly released to 0. The associated X–t plot and the σ –v plot are
shown in Fig. 4.9 (Bohnenblust, 1942b). Some key points are emphasized below, while
the details of the analysis are omitted herein.
(1) When the unloading process begins, the strong-discontinuous unloading distur-
bance propagates with the wave speed C0 , catching
√ up with the forerunning plastic
waves that propagate with the velocity C(= (1/ρ0 )(dσ/dε)). So long as the
unloading wave still keeps a strong-discontinuous, after having interacted with the
plastic waves, the propagating locus of the unloading disturbance is the same as
the propagating locus of the unloading boundary, namely, the propagating locus
= C0 . Therefore, on the X–t plane, the location
of the elastic–plastic boundary C
of this boundary segment A0 A2 can be determined.
(2) To the right side of the boundary A0 A2 , it is the plastic loading zone. According to
the previous discussions about elastic–plastic loading waves, the state variables σ
and v in this zone can be determined. Particularly, under the situation of the simple
waves propagating in a semi-infinite bar, there exists the following simple wave
relationship in the whole plastic zone:
σ
dσ
v=− (4.28)
0 ρ 0 C0
To the left side of the boundary A0 A2 , it is the unloading zone, where the state
variables need to be calculated. Note the strong-discontinuous feature of this
boundary, the state variables σ̄ and v̄ in the unloading zone should satisfy the
momentum-conservation equation for a discontinuous wave [Eq. (2.57)], which is:
σ̄ σ
v̄ + =v+ (4.29)
ρ0 C0 ρ0 C0
At the same time, the boundary A0 A2 coincides with the rightward characteristic line
in the unloading region that starts from the point A0 . Therefore the state variables
σ̄ and v̄ in the unloading zone should satisfy the compatibility equation along this
characteristic line [Eq. (4.12)]. Moreover, since the state variables at the point A0
after unloading have been known as:
σ∗
σ̄ (A0 ) = 0, v̄(A0 ) = v̄∗ = v∗ +
ρ0 C0
we can get:
σ̄ σ∗
v̄ − = v∗ + (4.30)
ρ0 C0 ρ0 C0
Thus the variables σ̄ and v̄ can be simultaneously solved from Eqs. (4.29) and
(4.30). For example, we can subtract Eq. (4.30) from Eq. (4.29), and the resulting
equation is substituted with Eq. (4.28). This renders:
1 1 1
σ ∗ C
σ̄ = (σ + ρ0 C0 v) − (σ ∗ + ρ0 C0 v∗ ) =
0
− 1 dσ
2 2 2 σ C
t t
C2
ary
nd D0
Bou
tic
las X
ic-P
El ast
C1
C0 B2 Centered Plastic Wave Zone
unloading
B C0
B1
B0 zone sor W
ave)
B0 Ct ecur
E ic Pr
A0 A2 X= t (Ea
lst
A X=C 0
A0 A1
σ σ∗ 0 G X
(a)
εR
(b)
0 X
σ
D0 C0 B0 E A0 v*
0
v
-Y
S D1
C2
D0 C1 B2 B1 A
2 A1
σ∗ C0 B
(c) B0 A A0
Fig. 4.9. Sudden unloading of centered plastic loading waves.

Furthermore, the stress jump along the unloading boundary can be determined:
1 1
σ − σ̄ = (σ − ρ0 C0 v) + (σ ∗ + ρ0 C0 v∗ )
2 2
σ ∗
1 C0 1 σ C0
= 1+ dσ + 1− dσ
2 0 C 2 0 C
Note that the above two equations are just the same as that have been obtained
previously when we discussed the internal impact between an unloading disturbance
and a plastic loading disturbance [Eqs. (4.26) and (4.27)]. Here the equations are
deduced by the characteristics method.
(3) On the plastic loading side of the elastic–plastic boundary, the lowest stress level
is the plastic yield limit Y . If the condition |σ − σ̄ | > 0 is satisfied even though
the loading stress level σ is less than Y , then the strong-discontinuous unloading
disturbance can pass through the whole plastic zone and does not vanish. According
to Eq. (4.27), the condition that the strong discontinuous unloading wave will not
be absorbed by the plastic loading zone, is:
σ∗
v∗ + = v̄∗ < 2vY
ρ0 C0
Note that this formula has also been obtained previously [Eq. (4.16)]. In the contrary
situation, if the strong-discontinuity vanishes at a certain point, say the point A2 ,
then at this point we must have: σ (A2 ) − σ̄ (A2 ) = 0. According to Eq. (4.27)
we get:
σ (A2 ) σ (A0 )
v(A2 ) − = v(A0 ) + (4.31a)
ρ0 C0 ρ0 C0
or
σ (A2 ) C0
σ ∗
C0

1+ dσ = − 1 dσ (4.31b)
0 C 0 C
From this equation we can determine the value of σ (A2 ). Therefore the intersection
point A2 of the straight line OA2 whose slope is C(σ (A2 )) and the straight line
A0 A2 whose slope is C0 can be determined. As a result, all information about
the elastic–plastic boundary A0 A2 in the first phase of unloading is known. This
includes the location of A0 A2 on the X–t plane, and the corresponding state map
of A0 A2 on the σ –v plane. Thus, the state variables at the zones in both sides of
A0 A2 , namely σ, v, σ̄ , and v̄, can all be determined.
(4) With regard to the unloading zone, since the variables σ̄ and v̄ along the charac-
teristic line A0 A2 have been determined and the boundary condition governs that
σ̄ |x=0 = 0 along the time axis, the unloading zone problem is reduced to a mixed
boundary (Picard) problem in the case of |v∗ | < |2vY | (the point A2 is infinitely far
away). However, in the case that |v∗ | > |2vY |, only in the triangle region A0 A2 B̄0 ,
the problem is reduced to a mixed boundary problem, where the point B 0 is the
intersection of the time axis with the leftward characteristic line emitted from the
point A2 . Within this region, along the leftward characteristic lines what propa-
gate are the influences of interactions between the unloading disturbance and the
plastic loading disturbances, while along the rightward characteristic lines what
propagate are the influences of boundary disturbances. For example, at an arbitrary
point E on t-axis the particle velocity v̄E can be calculated by the compatibility
condition along the leftward characteristic line AE and the boundary condition
σ̄E = 0. As the state variables at the point A (σ̄A and v̄A ) have been known, the
variable v̄E is:
σ̄A
v̄E = v̄A + (4.32)
ρ0 C0
(5) If the strong-discontinuous unloading disturbance is absorbed by the plastic loading

zone and vanishes at the point A2 , then after this point the elastic–plastic boundary
is reduced to a weak-discontinuous unloading boundary. Across this boundary, the
stress and the particle velocity should satisfy the continuity conditions:
σ̄ = σ, v̄ = v, along the weak-discontinuous boundary (4.33)
However, the locus of this boundary is not known in advance and this is just what we
need to solve. To determine the location of this boundary, now consider an arbitrary
point on the boundary, say the point B. In the plastic loading zone, the leftward
characteristic line passing through the point B intersects X-axis at the point G.
According to the compatibility condition along BG [Eq. (2.26)] and the zero initial
condition, or according to the simple wave relationship Eq. (4.28), we have:
σB dσ
vB + =0 (4.34)
0 ρ0 C
On the other hand, in the unloading zone, a rightward characteristic line passing
through point B intersects t-axis at the point E. According to the compatibility
condition along BE [Eq. (4.12)], and taking account of Eq. (4.32), we have:
σ̄B σ̄A
v̄B − = v̄E = v̄A + (4.35a)
ρ0 C0 ρ0 C0
Since the state variables at the point A are known, the above two equations, and
Eq. (4.33) applied to the point B, totally provide four equations for the unknown
variables vb , σB , v̄B , and σ̄B . Alternatively, substituting Eqs. (4.33) and (4.29) into
Eq. (4.35a), we get:
σB σA
vB − = vA + (4.35b)
ρ0 C0 ρ 0 C0
Therefore from Eqs. (4.34) and (4.35b) we can solve for the unknown variables vb
and σB . The rightward characteristic line OB in the plastic loading region has a
slope C(σB ). The location of point B can be determined by the intersection of this
line with the rightward characteristic line EB in the unloading region, while the
point E is the point where the leftward characteristic line AE intersects with the
t-axis. Therefore, the state of the arbitrary point B on the weak-discontinuous
unloading boundary during the second unloading phase is completely determined
by the state of the corresponding point A on the unloading boundary during the
first unloading phase. In other words, when A is known, B can be determined. The
particle velocity and the stress at the points A and B are related by Eq. (4.35a)
or Eq. (4.35b). These relationships are called the conjugate relationship. The
point pair on the unloading boundary like A and B are called the conjugate
points.
(6) Since the point B is an arbitrary point, the conjugate relationship is held on through
the whole unloading boundary. Therefore, from the already determined boundary
A0 A2 in the first phase of unloading, the conjugate boundary segment B0 B2 can
be obtained. Particularly, by applying Eq. (4.35b) to the point A0 and comparing
it with Eq. (4.31a), it is clear that the starting point of B0 B2 (the point B0 ) is just
the finishing point of A0 A2 (the point A2 ), and consequently, corresponding to
the constant-value part A0 A1 , the boundary segment B0 B1 is also a constant-value
part. So the boundary segment B0 B1 must coincide with the characteristic line OB1
in the plastic simple wave region. Since B0 B2 has been determined from A0 A2 ,
we can use the conjugate relationship to determine the boundary segment C0 C2
from B0 B2 . By the analogy of this approach, the whole unloading boundary can be
solved until all plastic waves vanish.
(7) With regard to the unloading region, since the stress σ̄ and the particle velocity v̄
along the whole unloading boundary are known, as well as the boundary stress σ̄ or
boundary velocity v̄ along the t-axis has been prescribed by the boundary condition,
the solution within the whole unloading region can be reduced to one of the different
kinds of solution-determinable boundary value problems, like a mixed boundary
problem, or a characteristic line boundary problem, etc. Thus the problem can be
solved accordingly.
(8) On the σ –v plane [Fig. 4.9(c)], the line OS corresponds to the strong-discontinuous
elastic wave, and the curve SA0 corresponds to the centered plastic wave [Eq. (4.28)].
In the whole plastic zone, including the plastic side of the unloading boundary, the
Eq. (4.28) is satisfied. So the curve SA0 is the mapping of the plastic side along the
whole unloading boundary. As for the strong-discontinuous unloading boundary
A0 A2 , the states on the unloading side start from the point A 0 and change along

the line A0 A2 up to the point A2 , where σ − σ̄ = 0, namely the unloading strong-
discontinuity vanishes. Therefore the line A 0 A2 is the mapping of the unloading
side along the strong-discontinuous unloading boundary A0 A2 . As for the weak-
discontinuous unloading boundary, since the continuity conditions for σ and v
[Eq. (4.33)] are satisfied, the line SB0 is precisely the mapping of the whole weak-
discontinuous unloading boundary. It is thus clear that although the location of
the unloading boundary on the X–t plane is not yet known, their mapping on
the σ –v plane can be determined immediately. Besides, an arbitrary point A and
its conjugate point B on the unloading boundary can be easily determined by the
characteristic lines AE and EB on the σ –v plane. The readers are suggested to deduce
the conjugate relationship [Eq. (4.35)] from the geometric relationship of the curved
triangle AEB.
(9) From this example, it is thus clear that if the unloading disturbance is in the form
of a strong-discontinuity from the very beginning, or, if the unloading disturbance
propagates in a constant-value plastic region from the very beginning regardless
of whether it is a strong-discontinuity or a weak-discontinuity, then the locus
of the unloading disturbance propagation coincides with the locus of unloading
boundary, C = C0 . However, in general, the propagating locus of the weak-
discontinuous unloading boundary is different from the propagating locus of the
unloading disturbance.
(10) The distribution of residual strain εR along the bar is shown in Fig. 4.9(b). Each
constant-value plastic zone on the X–t plane is corresponding to a constant-strain
platform on the εR –X plot. Comparing with the strain distribution given in Fig. 4.7,
the stationary strong-discontinuous strain interface in Fig. 4.7 is spread into a
“transition region”, wherein the residual strain gradually changes from one step
level to the next step level. This is because that the plastic loading waves herein is a
continuous wave, so that the residual strain distribution is weak-discontinuous and
thus no stationary strain-discontinuity appears.
Now we summarize the above discussions. The pursuing unloading problem where an
unloading disturbance pursues and interacts with the forerunning plastic loading distur-
bance in the semi-infinite bar has been analyzed. We have shown that when the stress–strain
relationship of material, as well as the initial-boundary conditions are known, this prob-
lem can be solved so long as the unloading disturbance from the very beginning is a
strong-discontinuity, such as the case of sudden unloading, no matter whether the fore-
running plastic loading disturbance is a strong- or a weak-discontinuity. This problem
is classified to solve a kind of direct problem. It is solvable because first of all, the
sudden unloading particularity can be utilized. In the case of sudden unloading, the
initial part of the unloading boundary is a strong-discontinuous boundary, which coin-
cides with the propagating locus of the unloading disturbance. Therefore, this initial
part of the unloading boundary, the strong-discontinuous part, is determinable. Second,
once this strong-discontinuous unloading boundary vanishes, the conjugate relationship
can be utilized to determine the successive parts of the weak-discontinuous unloading
boundary piece by piece. Finally, after the whole unloading boundary is determined, the
rest of the problems within the unloading region are reduced to solving either a mixed
boundary value problem, or a characteristic boundary value problem, both of which are
determinable.
4.5 Pursuing Unloading by Weak-Discontinuous Unloading Disturbances
Now we consider a semi-infinite bar made of decreasingly hardening material. The bar is
in a natural, rest state before the impact load is applied. The boundary loading on the end of
the bar (X = 0) is prescribed as: σ (0, t) = σ0 (t), where the function of σ0 (t) is graphically
shown in Fig. 4.10, namely, first gradually increasing in the duration 0 < t < t0 , then
gradually decreasing down to zero in the duration t0 < t < T0 , and finally keeping zero
t t
T0
X=f (t)
M(0,t*) M2(X2,t2)
t*
M1(X1,t1)
t0
σ 0(t) M(0,t0)
σ Y 0 X
Fig. 4.10. Pursuing unloading of weak-discontinuous loading disturbances by weak-discontinuous

unloading disturbances.
(i.e. the stress-free end condition) for t ≥ T0 . This is a kind of boundary loading what we
frequently encounter when studying the dynamic response of structures under explosive
loading.
Now the problem we will solve is such a problem where the after- running weak-
discontinuous unloading disturbances pursue the forerunning weak-discontinuous plastic-
loading disturbance. The interaction of these two kinds of disturbances results in an
unloading process and consequently a weak-discontinuous unloading boundary. On this
unloading boundary X = f (t) the particle velocity, stress, and strain are all cont-
inuous:
⎧
⎪
⎪v̄(X, t)|X=f (t) = v(X, t)|X=f (t) = vm (X)
⎨
σ̄ (X, t)|X=f (t) = σ (X, t)|X=f (t) = σm (X) (4.36)
⎪
⎪
⎩ε̄(X, t)| = ε(X, t)| = ε (X)
X=f (t) X=f (t) m
However, the derivative of the above variables may be discontinuous across the unloading
boundary X = f (t). The complexity lies in that, in general, the whole weak-discontinuous
unloading boundary X = f (t) is completely unknown at the present stage.
Consider an arbitrary point M2 (X2 , t2 ) on the boundary X = f (t). In the unloading zone,
the rightward characteristic line passing through the point M2 , the line MM2 , intersects
the t-axis at the point M(0, t ∗ ), while the leftward characteristic line passing through the
point M intersects the unloading boundary X = f (t) at the point M1 (X1 , t1 ). According
to the compatibility equation [Eq. (4.12)] along the characteristic line MM2 , we have the
following relationship that links the state variables at points M and M2 :
ρ0 C0 v̄(X2 , t2 ) − σ̄ (X2 , t2 ) = ρ0 C0 v̄(0, t ∗ ) − σ̄ (0, t ∗ )

And, from the compatibility equation along the characteristic line MM1 , we have the
following relationship that links the state variables at points M and M1 :
ρ0 C0 v̄(X1 , t1 ) + σ̄ (X1 , t1 ) = ρ0 C0 v̄(0, t ∗ ) + σ̄ (0, t ∗ )
Eliminating the variable v̄(0, t ∗ ) from the above two equations, and using the continuity
relationship of Eq. (4.36), we obtain:
ρ0 C0 vm (X2 ) − σm (X2 ) = ρ0 C0 vm (X1 ) + σm (X1 ) − 2σ0 (t ∗ ) (4.37a)
This is an extension of the conjugate relationship given in Eq. (4.35) to the case that the
unloading stress at bar end σ0 (0, t) is not zero. Note that as the weak-discontinuous unload-
ing boundary is now propagating in a simple wave region, the simple wave relationship
[Eq. (2.63)] must be satisfied. After introducing the wave speed ratio µ:
C0
µ(σ ) = ≥1
C(σ )
The conjugate relationship Eq. (4.37a) is rewritten as:

σm (X2 ) σm (X1 )
(µ + 1)dσ = (µ − 1)dσ + 2σ0 (t ∗ ) (4.37b)
0 0
which is an extension of Eq. (4.31b) given in the previous section.
From Eq. (4.37), it is clear that any point M2 on the weak-discontinuous boundary
X = f (t) is linked to its conjugate point M1 through the unloading boundary condi-
tion σ0 (t ∗ ). This is the most important relationship for studying the weak-discontinuous
boundary. By this basic equation, if we have known a point M1 on X = f (t), then the
conjugate point M2 on X = f (t) can be calculated. By the same argument, if a small part
of X = f (t), say the segment M0 M1 is known, then the rest of the unloading boundary
can be calculated piece by piece. However, in general, even a very short segment of the
unloading boundary is unknown in advance. Thus, the key is how to determine an ini-
tial short segment of the weak-discontinuous boundary X = f (t). This will be further
discussed in Section 4.7.
4.5.1 Continuously unloading of centered plastic waves
If the end of the bar is loaded suddenly and immediately unloaded gradually, then the
unloading problem is relatively easy to treat. In this case, since the plastic zone is a
simple, centered wave region as shown in Fig. 4.11, at any point on the weak-discontinuous
unloading boundary, we have the following relationship:
X
= C[σm (X)] on X = f (t) (4.38a)
t
t t
X=f(t) or t=g (X)
M2
M X=Ct (centered plastic waves)
t*
M1
σ 0(t)
X=C0 t (strong-discontineous elastic wave)
σ 0 X
Fig. 4.11. Continuously unloading of centered plastic waves.
Using the geometric relationship between the characteristic lines MM1 and MM2 in the
unloading zone, we obtain:
C0 t ∗ C0 t ∗
X1 = , X2 = (4.38b)
µ[σm (X1 )] + 1 µ[σm (X2 )] − 1
If the stress–strain relationship of the material is known, and if the residual strain distri-
bution along the bar εR (X) has also been experimentally measured, which is equivalent
to that, the function σm (X) and the function εm (X) can be determined according to the
unloading stress–strain relationship [Eq. (4.6)]:
σm (X)
εm (X) − = εR (X)
E
meanwhile the unloading boundary location X = f (t) can be determined from Eq. (4.38),
then the loading character at the bar end σ0 (t) can be solved from Eq. (4.37). Such kind of
problem is classified as the so-called inverse problem, namely, to determine the boundary
condition from the known stress–strain relationship of material and the known dynamic
response of the bar. It is clear from the above example that if the boundary loading is
suddenly applied then the inverse problem is solvable.
As to the direct problem for the same example discussed, it can be briefly stated as:
when the stress–strain relationship of the material and the initial-boundary conditions are
known, the demand is to determine the dynamic response of the bar, such as the stress,
strain, and the particle velocity as functions of X and t, or the residual strain distribution
εR (X). Although the accurate solution is almost unable to be found in the case of the
present example, nevertheless the approximate solution is obtainable, as will be discussed
in the following.
First, a progressive approximation method is introduced to solve the direct problem for
the case shown in Fig. 4.11.
Under the boundary condition of suddenly applied loading, Eq. (4.37b) can be rewritten
as below by using Eq. (4.38):

σm (vX) σm (X) (µ[σm (X)] + 1)X
[µ(σ ) + 1]dσ = [µ(σ ) − 1]dσ + 2σ0 (4.39a)
0 0 C0
where
µ[σm (X)] + 1
v= (4.39b)
µ[σm (vX)] − 1
In case that C0 → ∞, v → 1, we have:

X
σm (X) = 2σ0 (4.40)
C
This is equivalent to approximately regarding the unloading part of the bar as a rigid
body, of which the inertial effect is neglected. By this equation we can determine from the
boundary condition σ0 (t), the maximum plastic stress distribution along the bar σm (X)
just before unloading. Simultaneously we can use Eq. (4.38a) to determine the unloading
boundary X = f (t), or its inverse function t = g(X), namely:
X
t = g(X) = (4.41)
C[σm (X)]
Obviously the solution thus obtained does not satisfy the elastic unloading Eq. (4.39a).
Nevertheless, we can use this solution as the 0th order approximate solution, written as
(0)
σm (X) and g (0) (X). This 0th order solution is substituted into the1right
side
of Eq.(4.39a)
(1) (1)
again to obtain the 1st order solution, σm (X) and g (1) (X) = X C σm (X) , from
the left side of Eq. (4.39a). Similarly, substitute the 1st order solutions into the right side
(2)
of Eq. (4.39a), then
1 the left side of
Eq. (4.39a) provides the 2nd order solutions: σm (X)
(2)
and g (2) (X) = X C σm (X) . Repeating this approach, the iterative formula for the
nth order approximate solution can be written as:
⎡ (n−1) ⎤
(n)
σm (vX) σm
(n−1)
(X) µ σm (X) + 1 X
[µ(σ ) + 1]dσ = [µ(σ ) − 1]dσ + 2σ0 ⎣ ⎦
0 0 C0
(4.42)
In substance, the above approximate approach is like the previous approach where the
point M2 is calculated from the point M1 by the conjugate relationship [Eq. (4.37)]. Now
the 0th order solutions (equivalent to the point M1 ) are used to obtain the 1st order solutions
(equivalent to the point M2 ) through the conjugate Eq. (4.39). Again the 1st order solutions
(now equivalent to point M1 ) provide the 2nd order solutions (now equivalent to point M2 ),
and so on. By each iterative operation, the solutions are corrected once again through the
boundary condition at the bar end σ0 (t) and approach the exact solutions. The choice of
the 0th order solution does not matter much to the final results, but can affect the number of
iterations to get the accurate solutions. In the above discussion, Eq. (4.40) was taken as the
0th order solution, which was first suggested by (Buravtsev) (1970). However,
the progressive approximation algorithm used by subjects to the limitation that
1/µ 1, which is no longer needed in the present algorithm modified by the present
author.
4.5.2 Attenuation of shock waves in linearly hardening materials
In the previous example, if the material is linearly hardening plastic instead of decreasingly
hardening plastic, then rather than being a set of continuous waves, the plastic √wave will be
a strong-discontinuous shock wave propagating with constant velocity C1 = E1 /ρ. The
unloading process is a pursuing interaction between weak-discontinuous unloading distur-
bances and a strong-discontinuous plastic loading disturbance propagating with constant
velocity C1 .
Obviously, in this situation, the propagating locus of the unloading boundary coincides
with the propagating locus of the strong-discontinuous plastic loading disturbance, namely
= C1 . On the X–t plane, the unloading boundary is expressed as:
C
= C1 t
X = f (t) = Ct
This result can also be obtained from the previous problem shown in Fig. 4.10. In the
case of linearly hardening material, first the rightward characteristic lines in the plastic
loading region in Fig. 4.10 become a set of parallel lines with constant slope C1 , then let
t0 → 0 (representing the boundary condition of sudden loading), the original problem is
reduced to the present problem, as shown in Fig. 4.12. Note that the strong-discontinuity
t t
(elastic unloading wave)
e)
A1
av
w
ck
ic 1 t
ast X=C
sho
T1 A0
T0
(pl
*) ave)
M(0,t M2(X2,t2) cursor w
c pre
t*
t(elasti
p (t)
M1(X1,t1) X=C 0
−σ −σ 0 0 X
Fig. 4.12. Pursuing unloading of constant-velocity propagating strong-discontinuous plastic load-

ing disturbances by weak-discontinuous unloading disturbances.
along X = C1 t is attributed to the plastic suddenly loading process. The unloading process
is, in nature, still a weak-discontinuity.
Because of the linearly hardening plastic behavior, the conjugate relationship [Eq. (4.37)]
is rewritten as:
(µ1 + 1)σm (X2 ) = (µ1 − 1)σm (X1 ) + 2σ0 (t ∗ )

C0 E (4.43)
µ1 = =
C1 E1
which, by taking account of Eq. (4.38), can be further reduced to:

1 C0 t 1 C0 t
(µ1 + 1)σm − (µ1 + 1)σm = −σ0 (t) = p(t) (4.44)
2 µ1 + 1 2 µ1 − 1
Assume that the function of boundary unloading pressure, p(t), can be expanded in a
power series [ (Rakhmatulin), 1945b]:
∞
.
p(t) = pn t n (4.45a)
n=0
Then from Eq. (4.44) we know that σm (X) should have the following form:
∞
.
σm (X) = bn X n (4.45b)
n=0
where
2pn
bn = n n
C0 C0
(µ1 − 1) − (µ1 + 1)
µ1 + 1 µ1 − 1
Accordingly, the functions of strain εm (X) and the residual strain distribution εR (X) are
expressed as:
σm (X)
∞
1 .
εm (X) = + 1 − µ21 εY = bn X n + 1 − µ21 εY (4.46)
E1 E1
n=0
.
∞
σm (X) 1 1
εR (X) = εm (X) − = − bn X n + 1 − µ21 εY (4.47)
E E1 E
n=0
Therefore the direct problem is solvable. This approach is called a power series
approximate method.
The length of plastic deformation l can be determined from Eq. (4.47) by the condition
εR (l) = 0. For example, suppose that the boundary unloading pressure, p(t), has a form:
n
t
p(t) = pm 1 − (4.48)
T0
where pm is the peak pressure applied on the bar end when t = 0. T0 is the time when
boundary load is unloaded to zero. We can obtain
/ n+1 n+1 0 2 n1
C1 T0 µ1 1 1 Y
l= 1+ − 1− 1− (4.49)
1 2 µ1 µ1 pm
1− 2
µ1
Generally 1/µ = (C1 /C0 ) < 1, when we omit the small term 1/µ2 and those of the
higher order, Eq. (4.49) is reduced to
1
Y n
l = C1 T0 (n + 1) 1 −
pm
It is thus clear that the length of plastic deformation l increases with the plastic wave
velocity of the linearly hardening material, C1 , the maximum boundary-loading pm , and
the boundary loading duration T0 , while it is independent of the material elastic modulus E.
Suppose that the boundary loading is linearly decreasing with time. This means that the
index n of Eq. (4.48) is 1. Or equivalently, from Eq. (4.45a) we take
pm
p0 = pm , p1 = − , pi = 0 (i = 2, 3, . . .)
T0
Then from Eq. (4.45b) we have:
µ21 − 1 pm
b0 = −pm , b1 = , bi = 0 (i = 2, 3, . . .)
2µ1 C0 T0
Therefore we have
% &
µ2 − 1 X
σm (X) = −pm 1− 1 (4.50)
2µ1 C0 T0
From Eq. (4.49) we get

Y
2C1 T0 1 −
pm
l=
1
1− 2
µ1
Of course, the same problem can be solved by the progressive approximation method
based on Eq. (4.39). In case of linearly hardening material, Eq. (4.39) is reduced to:

1 2 (µ1 + 1)X
σm (ν1 X) = σm (X) + σ0 (4.51a)
ν1 µ1 + 1 C0
or

2 (µ1 + 1)λ1 X
σm (X) = λ1 σm (λ1 X) + σ0 (4.51b)
µ1 + 1 C0
where the parameters ν1 and λ1 are known constants:
µ1 + 1 1
ν1 = =
µ1 − 1 λ1
Take the solution when C0 → ∞ as the 0th order approximation, namely from Eq. (4.48)
when n = 1, we have:

X
σm (X) = −pm 1 −
(0)
C1 T0
Substitute this 0th order approximation into the right side of Eq. (4.51), the left side of
Eq. (4.51) provides the 1st order approximate solution:

λ1 X 2 (µ1 + 1)λ1 X
σm (X) = −λ1 pm 1 −
(1)
− pm 1 −
C 1 T0 µ1 + 1 C0 T0
X
= −pm 1 − 2λ1 + λ21 µ1
C0 T0
Similarly, substitute this 1st order solution into the right side of Eq. (4.51), the left side of
Eq. (4.51) provides the 2nd order approximate solution:

2 (µ1 + 1)λ1 X
σm(2) (X) = λ1 σm(1) (X) + σ0
µ1 + 1 C0
X
2 (µ1 + 1)λ1 X
= −λ1 pm 1 − 2λ1 + λ1 µ1 2
− pm 1 −
C0 T0 µ1 + 1 C0 T 0
X
= −pm 1 − 2λ1 1 + λ21 + λ41 µ1
C0 T0
Repeating this approach, the nth order approximate solution can be expressed as:
X
2(n−1)
σm(n) (X) = −pm 1 − 2λ1 1 + λ21 + λ41 + · · · + λ1 + λ2n
1 µ 1
C0 T0
/ 0 2 (4.52)
2λ1 1 − λ2n 1 X
= −pm 1 − + λ 1 µ1
2n
1 − λ21 C0 T0
Table 4.1. The B values calculated by different equations for µ1 = 2.

Calculation Exact solution Approximate solution of Eq. (4.52)
equation [Eq. (4.50)] 0th order 1st order 2nd order 3rd order 4th order
B value 3/4 = 0.750 2.00 0.889 0.765 0.752 0.750
As λ1 < 1, when n → ∞ we get:

2
µ2 − 1 X
lim σ (n) (X) = −pm 1− 1
n→∞ m 2µ1 C0 T0
which is exactly the same as the formula in Eq. (4.50). As can be seen, by two different
methods we get the same results.
In the nondimensional form, the approximate solution given by Eq. (4.52) and the exact
solution given by Eq. (4.50) can be unified as the following linear functions:

σm (X) X
=1−B (4.53)
−pm C0 T0
The differences between the different order approximate solution and the exact solution
are expressed by the coefficient B, the slope of this linear function. Assume µ1 = 2,
the B value calculated by Eqs. (4.50) and (4.52), respectively, are listed in Table 4.1.
As can be seen, comparing the B values calculated up to three significant digits, the
4th order approximation gives the same results as the exact solution. Of course, the speed
of divergence is also influenced by the proper choice of the 0th order approximate form.
It should be pointed out that the above solutions are applicable to the OA0 segment in
Fig. 4.12 corresponding to the unloading stage when |p(t)| > 0. At the time when t ≥ T0 ,
the boundary load becomes zero, p(t) = 0. The solution at t ≥ T0 can be calculated from
the OA0 by the conjugate relationship [Eq. (4.35)]. Note that at the time t = T0 , the first
order derivative of p(t) is discontinuous, the zigzag curve T0 A0 T1 A1 . . . represents the
propagating locus of this weak-discontinuity.
4.6 Shock Wave Attenuation Due to Pursuing Unloading
In the previous section we have discussed the pursuing unloading problem where a weak-
discontinuous unloading disturbance pursues and interacts with a strong-discontinuous
plastic wave (plastic shock wave), but the material is elastic–plastic with linearly hard-
ening character so that the speed of plastic shock wave is constant. If the material is
increasingly hardening plastic, the speed of plastic shock wave is no longer constant but
changes with the intensity of shock wave. The problem can be treated in the similar way as
discussed previously, except that the varied shock wave velocity, D , should be taken into
account.
In such a case, since the unloading boundary still coincides with the propagating locus
of the shock wave, the problem to determine the unloading boundary is equivalent to the
problem to determine the propagating locus of the shock wave which is interacted with
the tail-following unloading disturbance. Alternatively, it is equivalent to investigating the
attenuation of a shock wave caused by the pursuing unloading disturbance. The following
example shows an approach to solve this problem.
Let us consider a semi-infinite bar made of increasingly hardening material. The bar is at a
natural, rest state before the impact loading is applied. At the time t = 0 the end of the bar
(X = 0) is applied a sudden load σm (0) and then is gradually released to zero, as shown
in Fig. 4.13. The present problem differs from the previous problem (Fig. 4.12) in that the
strong-discontinuous plastic loading wave (i.e. plastic shock wave) no longer propagates
with a constant speed. When the intensity of the shock wave decreases, its speed D also
decreases. The unloading boundary, which coincides with the shock wave locus, is an
unknown curve expressed by function X = f (t) or the inverse function t = g(X).
For the convenience of discussion, we assume that the stress–strain curve of the material
can be simplified as shown in Fig. 4.13 (the left diagram). This means that the linear
elastic loading part of the stress–strain curve can be omitted. Soil is a typical material
example that exhibits this type of stress–strain relationship. The above assumption does
not change the basic features of our problem. Adding a linear elastic loading part to the
σ –ε curve only means that a strong-discontinuous elastic precursor wave runs before the
plastic shock wave.
The key of this problem is how to determine the attenuation properties and thus the propa-
gating locus of the plastic shock wave. With the present stress–strain relationship and the
zero initial conditions, the shock wave velocity is:

1 1 σm
D = f (t) = = (4.54)
g (t) ρ0 εm
σ t t
X=f (t)
M2(X2,t2)
*)
M(0,t A4
t*
p (t) M1(X1,t1)
σ 0(t) A3
0 ε
A2
A1
σ σ m(0) 0 X
Fig. 4.13. An increasingly-hardening semi-infinite bar subjected to a suddenly applied loading and
subsequently a gradually unloading.
and the momentum conservation equation across the shock wave front is reduced to
σm = −ρ0 D vm (4.55)
In the unloading zone, an arbitrary point M1 on the unloading boundary is related to the
conjugate point M2 on the same boundary, via the characteristic lines M1 M and MM2 ,
as shown in Fig. 4.13. Referring to Eq. (4.37a), the conjugate relationship between the
point M1 and the point M2 is
ρ0 C0 vm (X2 ) − σm (X2 ) = ρ0 C0 vm (X1 ) + σm (X1 ) − 2σ0 (t ∗ )
Using Eq. (4.55), the above equation is reduced to:

C0 C0
+ 1 σm (X2 ) = − 1 σm (X1 ) + 2σ0 (t ∗ ) (4.56)
D [σm (X2 )] D [σm (X1 )]
By introducing a dimensionless parameter µD representing the velocity ratio of C0 to D
C0
µD = (4.57)
D (σ )
the conjugate relationship can be rewritten as:
{µD [σm (X2 )] + 1} σm (X2 ) = {µD [σm (X1 )] − 1} σm (X1 ) + 2σ0 (t ∗ ) (4.58)
Since the σ –ε relationship is given, the functions D (σ ) and thus µD (σ ) are also known.
On the other side, from the geometric relationship of characteristic lines M1 M and MM2 ,
we can deduce the coordinate relationships between the point X1 , the point X2 and the
point t ∗ :
⎫
X1 X1 X1 ⎪
∗
t = t1 + = g(X1 ) + = {µS (X1 ) + 1} ⎪ ⎪
C0 C0 C0 ⎪⎪
⎪
⎪
⎬
∗ X 2 X 2 X 2
t = t2 − = g(X2 ) − = {µS (X2 ) − 1} (4.59)
C0 C0 C0 ⎪⎪
⎪
⎪
µS (X2 ) − 1 ⎪
⎪
X1 = · X2 = λS X2 ⎪
⎭
µS (X1 ) + 1
where the quantities µS (X1 ) and µS (X2 ) defined at points X1 , X2 , respectively, represent
the ratio of C0 to the slope of straight line OM1 or OM2
C0 g(X)
µS (X) = (4.60)
X
and the quantity λS is defined as:
µS (X) − 1
λS (X) = (4.61)
µS (λS X) + 1
Substitute the above equations into Eq. (4.58), we get:

(µS (X) − 1)X
{µD [σm (X)] + 1}σm (X) = {µD [σm (λS X)] − 1}σm (λS X) + 2σ0
C0
(4.62)
If t = g(X) is known, then the functions µS (X) and accordingly λS (X) are all known.
Therefore, based on Eq. (4.58), we can use the progressive approximation method to
determine the unloading boundary t = g(X).
Imagine that the intensity of shock wave during the unloading process is approximately
equal to the boundary loading, which is equivalent to the situation that C0 → ∞, and
regard this situation as the 0th order approximate solution, we have
σm(0) (t) = σ0 (t) (4.63)
Since the stress–strain relationship of the material has been known, the function D (t) can
be determined from Eq. (4.54). By a further integration, the 0th order locus of the shock
wave can be obtained
t
X = f (0) (t) = D (t)dt (4.64)
0
Equivalently, its inverse function t = g (0) (X) is also obtained. Substituting it into
(0) (0) (0)
Eq. (4.63) we get the function σm (X), and accordingly the functions µS (X) and λS (X),
(0)
where the λS (X) is actually determined by an iteration manipulation of Eq. (4.61).
The 0th order approximate solutions are substituted into the right side of Eq. (4.58). The
(1)
1st order approximation of the function σm (X), namely σm (X) can be got from the left
side of Eq. (4.58) through an iteration manipulation:

(0) (0)
µD σm(1) (X) +1 σm(1) (X) = µD σm(0) λS X −1 σm(0) λS X
⎡ (0) ⎤
µS (X)−1 X
+2σ0 ⎣ ⎦ (4.65)
C0
By Eq. (4.54) the 1st order locus of shock wave, t = g (1) (X), and accordingly the functions
(1) (1)
µS (X) and λS (X) can be determined too. Similarly the 1st order approximate solutions
are substituted into the right side of Eq. (4.62) and the left side of Eq. (4.62) gives the
2nd order approximations. This iteration process is repeated until the solutions satisfy the
accuracy requirement.
Once the shock wave locus OA1 A2 . . . is determined, the intensity of the shock wave σm (X)
is also obtained, which reflects the attenuation of the shock wave during the pursuing
unloading process. The whole unloading zone can be solved as a Picard problem. Thus
this problem is completely solved.
4.7 Propagating Properties of Elastic–Plastic Boundaries in Semi-Infinite Bars
From the discussions up till now, we know that in general the propagating locus of an
elastic–plastic boundary differs from that of an unloading disturbance. The exceptions are
the cases that either the elastic unloading disturbance or the plastic loading disturbance
is a strong-discontinuity. The elastic–plastic boundary is not a wave motion of mechani-
cal disturbance itself. It may either move forward, or move backward, depending on the
interaction of waves. The propagation velocity of an elastic–plastic boundary, C, gener-
ally varies during the interaction process of the elastic–plastic waves. An elastic–plastic
boundary can be a loading boundary, which develops with time from an elastic zone to a
plastic zone, or an unloading boundary, which develops with time from a plastic zone to
an elastic zone.
In case of a semi-infinite bar, because no wave propagates from the other end, the elastic–
plastic loading wave is a simple wave. Therefore, the loading boundary can be easily
determined. The difficulty is to determine the unloading boundary. In the following, we
will use the characteristic line method to investigate the general propagating features of a
weak-discontinuous unloading boundary in a semi-infinite bar. Particularly, we will discuss
what are the major factors which influence the propagating speed of the boundary C.
4.7.1 An analysis by the characteristic line method
Suppose that the weak-discontinuous unloading boundary is expressed by the function

X = f (t) or its inverse function t = g(X). Now consider an arbitrary point M1 (X1 , t1 )
and its neighboring point M2 (X2 , t2 ) on the unloading boundary. Those two points are
connected by the characteristic lines to a third point M(X, t) in the unloading zone. The
characteristic line MM1 in the unloading zone is leftwards, and the characteristic line
MM2 in the unloading zone is rightwards, as shown in Fig. 4.14. In the plastic loading
zone, the leftward characteristic line passing through point M1 interacts with the rightward
characteristic line passing through point M2 at the point N .
From the compatibility relationship along the characteristic line MM1 :
ρ0 C0 v̄(M) + σ̄ (M) = ρ0 C0 vm (M1 ) + σm (M1 )
and from the compatibility relationship along the characteristic line MM2 :
ρ0 C0 v̄(M) − σ̄ (M) = ρ0 C0 vm (M2 ) − σm (M2 )
by eliminating the quantity v̄(M), and taking account of that in the semi-infinite bar there
exists a simple wave relationship between vm and σm [Eq. (4.28)], finally we have:
1
σm (X1 ) C0

1 σm (X2 )

C0

σ̄ (X, t) = 1− dσ + 1+ dσ (4.66)
2 0 C 2 0 C
The above equation is actually an extension of the conjugate relationship [Eq. (4.37b)] to
the case that the point M is not located on the t-axis.
t t
X=f (t)
or t=g(X)
M2
Elastic unloading zone
M M1
N
A3
t2
σ 0(t) plastic loading zone
t1 A2
A1
t0
M0
σ 0 X
Fig. 4.14. An analysis of general propagating features of a weak-discontinuous unloading boundary

in a semi-infinite bar by the characteristics method.
The expressions of the characteristic lines MM1 and MM2 are:

MM 1 : X1 = X + C0 {t − g(X1 )}
(4.67a)
MM 2 : X2 = X − C0 {t − g(X2 )}
differentiating
Taking account of, that X and t are independent variables and g = 1/C,
X1 and X2 in Eq. (4.67) with respect to variable t, we have:
⎫
∂X1 ∂X1 C0 ⎪
= C0 1 − g = ⎪
∂t ∂t C0 ⎪
⎪
⎪
1+ ⎪
⎬

C
(4.67b)
∂X2 ∂X2 C0 ⎪ ⎪
= C0 g −1 = ⎪
⎪
∂t ∂t C0 ⎪
⎪
− 1⎭

C
Therefore from Eq. (4.66) we get:

∂σ 1 C0 dσm ∂X1 1 C0 dσm ∂X2
= 1− + 1+
∂t 2 C dX 1 ∂t 2 C dX 2 ∂t
(4.68)
1 C0 C0 dσm 1 C0 C0 dσm
= 1− · + 1+
2 C C 0 dX 2 C C 0 dX
+1 1 −1 2

C
C
When M2 → M1 , M → M1 , the relationship between the quantities ∂ σ̄ /∂t and dσm /dX
at an arbitrary point M1 on the weak-discontinuous unloading boundary t = g(X) is
obtained
⎧ ⎫
⎪
⎪ C0 C0 ⎪
⎪
∂ σ̄ 1⎨ 1 − C 1+
C
⎬ dσ
m
= + (4.69)
∂t ⎪
2⎪ 1 1 1 1 ⎪
⎪ dX
⎩ + − ⎭ 2
C0
C C0
C
On the other hand, in the plastic loading simple wave zone, along the characteristic line
NM2 we have:
⎫
σ (N ) = σ (M2 ) ⎬
∂σ dσm ∂X2 (4.70)
= ⎭
∂t dX 2 ∂t
while from the expression of the characteristic line NM2 we have:
X2 = X + C{g(X2 ) − t}

∂X2 ∂X2 dC dσm ∂X2
= C g − 1 + (g(X2 ) − t)
∂t ∂t dσ dX 2 ∂t (4.71)
C
=
C dC dσm
− 1 + (g(X2 ) − t)

C dX 2 dX 2
Substitute Eq. (4.71) into Eq. (4.70), and let M2 → M1 , we get the relationship between
the quantities ∂σ/∂t and dσm /dt at an arbitrary point M1 on the boundary t = g(X):
dσm
∂σ
= dX (4.72)
∂t 1 1
−
C
C
Since both C0 and C are known quantities, from Eqs. (4.69) and (4.72) we can eliminate
the quantity dσm /dX and thus obtain the following relationship between the propagation
and the quantities ∂σ/∂t and ∂ σ̄ /∂t at an arbitrary
velocity of the unloading boundary C
point M1 on the unloading boundary t = g(X):
2
∂σ 2 C
2 C 2 1−
C
∂t = 0 − C C
= 2 (4.73a)
∂ σ̄ 2 C 2
C2 − C C
∂t 1−
C
can be solved in the explicit form

Alternatively, the quantity C

∂ ε̄
∂σ ∂ σ̄ C2 ∂ε
− − C2

=
C ∂t ∂t = ∂t
∂ ε̄ ∂ε
∂t (4.73b)
1 ∂σ 1 ∂ σ̄
− −
C 2 ∂t C0 ∂t ∂t ∂t
In the present situation, t = g(X) expresses the unloading boundary, across which the
stress state changes from plastic loading region to the elastic unloading region. As long
as the quantities ∂σ/∂t and ∂ σ̄ /∂t are not simultaneously zero, these quantities must
have different signs, namely (∂ σ̄ /∂t)/(∂σ/∂t) ≤ 0 or (∂σ/∂t)/(∂ σ̄ /∂t) ≤ 0. Thus from
Eq. (4.73a) we have:
≥C
C0 ≥ C (4.74)
This means that in a semi-infinite bar the moving speed of an unloading boundary C is
between the elastic wave velocity C0 and the plastic wave velocity C. For example, in
Fig. 4.14, the unloading boundary passing through the point M2 must fall within the shaded
area shown in the figure.
either reaches its

Under the following four specific situations, as shown in Fig. 4.15, C
upper limit C0 , or reaches its lower limit C:
∂σ ∂ σ̄ ⎫
(a) = 0,
= 0: = C0 ⎪
C ⎪
⎪
⎪
∂t ∂t ⎪
⎪
⎪
⎪
⎪
⎪
∂σ

= ∞,
∂ σ̄
= −∞: ⎪
C = C0 ⎪
(b) ⎪
⎬
∂t ∂t
(4.75)
⎪
⎪
(c)
∂σ

= 0,
∂ σ̄
= 0: =C ⎪
C ⎪
⎪
∂t ∂t ⎪
⎪
⎪
⎪
⎪
⎪
∂σ ∂ σ̄ ⎪
⎪
(d) = ∞,
= −∞: C = C ⎭

∂t ∂t
Note that the Case (b) is in accordance with the situation when a strong-discontinuous
unloading disturbance pursues a weak-discontinuous plastic wave (C = C0 ); the Case (d)
is in accordance with the situation when a weak-discontinuous unloading disturbance
pursues a strong-discontinuous plastic wave (C = C), while the Case (a) coincides with
the situation that the unloading disturbance propagates in the constant-value plastic zone
= C0 ).
(C
In the previous discussions, we have assumed that the derivatives ∂σ/∂t and ∂ σ̄ /∂t are
not simultaneously zero, which means that the unloading boundary is a first order weak-
discontinuity with respect to the stress σ and particle velocity v. In case that ∂σ/∂t =
∂ σ̄ /∂t = 0, the right side of Eq. (4.73a) represents an indefinite quantity. In order to
understand the features of the unloading boundary, we must further consider the second
order derivatives of stress: ∂ 2 σ/∂t 2 , ∂ 2 σ̄ /∂t 2 , etc.
σ σ
σ =0 σ=∞ σ =- ∞
t t t
loading unloading
loading unloading
0 t 0 t
(a) (b)
σ σ
σ=0 σ= ∞ σ =- ∞
σ =0 t
t t t
loading unloading loading unloading
0 t 0 t
(c) (d)
Fig. 4.15. Four specific situations of unloading boundary propagation in a semi-infinite bar.
In the unloading region, from Eq. (4.68), we have the following relationship at the point M:

∂ 2 σ̄ 1 C0 d 2 σm ∂X1 2 1 C0 d 2 σm ∂X2 2
= 1 − + 1 +
∂t 2 2 C dX 21 ∂t 2 C dX 22 ∂t
⎛ ⎞2 ⎛ ⎞2

1 C0 ⎜ ⎟ 2 ⎜
⎜ C 0 ⎟ d σ m + 1 1 + C 0 ⎜ C 0 ⎟ d σm
⎟ 2
= 1−
2 C ⎝ C0 ⎠ dX 2 2 C ⎝ C0 ⎠ dX 2
+1 1 −1 2
C
C
Note that in the above derivation we have used the conditions that (dσm /dX) = 0 when
(∂ σ̄ /∂t = 0) [Eq. (4.69)], so that all the terms with dσm /dX are omitted. When M2 → M1
so that M → M1 , we get the following relationship between the second order derivatives
∂ 2 σ̄ /∂t 2 and d 2 σm /dt 2 at the point M1 on the unloading boundary t = g(X):
⎡ ⎤
C0 C0
∂ 2 σ̄ 1⎢
⎢
1−
C
1+
C ⎥
⎥ d 2σ
m
= ⎢ + 2⎥
(4.76)
∂t 2 2⎣ 1 1 2
1 1 ⎦ dX 2
+ −

C C
C C
Similarly, in the plastic loading region, differentiate Eq. (4.72) with respect to the variable t,
and let M2 → M1 . This way we get a relationship between the second order derivatives
∂ 2 σ/∂t 2 and d 2 σm /dX 2 at the point M1 on t = g(X), as shown in the following:
d 2 σm
∂ 2σ dX 2
= (4.77)
∂t 2 1 1 2
−
C
C
Eliminate the quantity d 2 σm /dX 2 from Eqs. (4.76) and (4.77), we get the follow-
ing relationship at the point M1 on t = g(X), connecting the quantity ∂ 2 σ̄ /∂t 2 and

∂ 2 σ/∂t 2 to C:
∂ 2σ
∂t 2 = 2
⎧ ⎫
2
∂ σ̄ ⎪
⎪ C C ⎪
⎪
2
2 ⎪
⎨ 1 −
0
1 +
0 ⎪
⎬
∂t 1 1 C C
− +
C ⎪
C ⎪ 1 1 2 1 1 2⎪ ⎪
⎪
⎩ + − ⎪
⎭ (4.78a)
C
C C
C
2
C0 − C2 2 C2
= % &· 2
C2 C0
C−C 2 C2 + 2 0 C +C 2
0 C
Introducing
This is an algebraic equation of fourth degree with respect to the variable C.
the following nondimensional parameters:
∂ 2σ
2 C
C
β = ∂t2 , a= , ā =
∂ σ̄ C0 C0
∂t 2
then Eq. (4.78a) can be rewritten in the nondimensional form as following:
2a
β= 2 (4.78b)
ā 1+a 1−a
1− −
a (1 − ā)2 (1 + ā)2
Or the equation can be written as an algebraic equation of fourth degree with respect to
variable ā, as:

1 " #
β − a ā + 2β
2 4
− a ā 3 + β + 2 a 2 − 3β ā 2 + (β − 1)a 2 = 0 (4.78c)
a
Note that for a second order weak-discontinuous unloading boundary, ∂ 2 σ/∂t 2 and
∂ 2 σ̄ /∂t 2 must have the same signs, therefore β ≥ 0.
Under the following four specific situations, the problem can be further simplified and an
can be obtained:
explicit expression of C
= C0 ;
(i) When β = 0, from Eq. (4.78a) we get: C
(ii) When (1/β) = 0, from Eq. (4.78a) we get: C = C;
2
(iii) When β = 1, which means that ∂ σ/∂t 2 = ∂ 2 σ̄ /∂t 2 , because 1 − a 2
= 0,
the Eq. (4.78c) can be reduced to:

2ā 3 2ā
ā +
4
− 3ā = ā ā +
2 2 2
−3 =0
a a
from which the positive real root of C can be solved, which represents the
propagating velocity of the rightward unloading boundary shown in Fig. 4.14:
⎡ ⎤
2
C02 C
=
C ⎣ 1+3 − 1⎦ (4.79)
C C0
This result coincides with that obtained by (Biderman) (1952).

2 2
(iv) When β = a , which means that ∂ ε/∂t = ∂ ε̄/∂t 2 , Eq. (4.78c) can be
2 2
reduced to an cubic algebraic equation with respect to ā:
a a
ā 3 + ā 2 − = 0
2 2
of which the following real positive root can be solved:
% & % &
9 9
3 4 3 4 2

ā = α + 3 α + α 1 + 1 + α + 3 α + α 1 − 1 + α
3 2 3
2 3 2 3
(4.80)
where a parameter α = a/6 is introduced for the convenience of expression.

Similarly, if ∂ 2 σ/∂t 2 = ∂ 2 σ̄ /∂t 2 = 0, Eq. (4.78) becomes indefinite, then we need
to investigate the features of the third derivatives ∂ 3 σ/∂t 3 , ∂ 3 σ̄ /∂t 3 etc. By this analogy,
if the derivatives (∂ i σ /∂t i ) = (∂ i σ̄ /∂t i ) = 0, i = 1, 2, 3, . . . , (n − 1), then in order
to understand the features of the unloading boundary we must investigate the nonzero
nth derivatives ∂ n σ/∂t n and ∂ n σ̄ /∂t n . By the similar mathematical deductions, we get a
relationship that connects the quantities ∂ n σ/∂t n , ∂ n σ̄ /∂t n and C:
∂ nσ
∂t n = 2
⎡ ⎤ (4.81)
∂ n σ̄ C0 C0
n 1 − 1 +
∂t n 1 1 ⎢ ⎢ C + C ⎥
⎥
− ⎣
C
C 1 1 n
1 1 n⎦
+ −
C
C C
C

which is an algebraic equation of 2nth degree with respect to the variable C.
4.7.2 Local linearization method
The above results not only provide a further understanding about the propagating features
of an unloading boundary moving in a semi-infinite bar, but also provide an idea to solve
the problem approximately by a new method, the so-called local linearization method.
The basic ideal is as follows: apply the results obtained in the subsection 4.7.1 to the end
of the bar (X = 0), the initial propagating velocity of the unloading boundary, namely
i = f (t0 ), at the point M0 (the beginning point of the unloading boundary at t = t0 as
C
shown in Fig. 4.14) can be determined from Eq. (4.81), since the boundary loading σ (0, t)
and its derivatives with respect to time t including the least order partial derivatives of
σ and σ̄ which are not simultaneously zero, are known. In other words, the initial slope
of the curve X = f (t) at the point M0 on the X–t plane (Fig. 4.14) can be determined.
If the length of the initial segment of the curve X = f (t), M0 A, is taken so short as it can
be approximated by a short straight line with slope C i , then the location of M0 A is now
approximately known. On the other hand, the stress and the particle velocity along this
line segment can be determined by the physical states known in the plastic loading zone.
Therefore the line segment M0 A is completely solved. As mentioned previously, once a
short initial part of the curve X = f (t) is known, the rest of the curve can be determined
by applying the conjugate relationship [Eq. (4.37)] piece by piece.
An example to solve the unloading boundary by means of the local linearization method
is given in Fig. 4.16 [ (Shapiro), 1946]. In this problem, the semi-infinite bar
is made of linearly hardening plastic material. The bar is at natural, rest state before a
parabolic-shape impact load is applied at its end:

t t
σ0 (t) = −8Y 1−
T T
where Y is the yield limit, T the loading duration. At the time t = T /2 the boundary
load reaches its maximum (σ0 )max = −2Y, and the unloading
starts. Since at this time
(∂σ/∂t) = (∂ σ̄ /∂t) = 0, while ∂ 2 σ/∂t 2 = ∂ 2 σ̄ /∂t 2
= 0, the initial propagating
velocity of the unloading boundary can be calculated from Eq. (4.79). Assuming C1 /C0 =
1/4, we have:
√
0 ) = C0 ( 19 − 4) = 0.359C0 = 1.44C1
C(M
0 ). The stress at the

Now we can draw an initial short straight line M0 A1 with the slope C(M
point A1 , σm (A1 ), can be determined from the boundary loading σ0 (t) at the intersection
of the time axis and the rightward plastic characteristic line passing through the point A1
with slope C1 . Practically, it is very close to the maximum boundary load (σ0 )max .
Once the initial boundary segment M0 A1 , has been determined, we can use the conjugate
relationship to determine the following segments A1 A2 , A2 A3 , . . . until the plastic wave
vanishes at the point K. Figure 4.16(b) shows the stress–time curves at three sections of
the bar. From this figure it is clear that during the unloading process the plastic stress grad-
ually attenuates until it completely vanishes. As can be seen too, the profiles of the stress
impulse, including the quantities ∂σ/∂t and ∂ σ̄ /∂t at the maximum stress point, varies
t t
T T σ
-Y
2
1 X=0
K t
0 0.5 1.0 1.5 2.0 T
4 A4
13 2
AK X=0.5C1t
1
2 A3
1 t
2 1
A 0 0.5 1.0 1.5 2.0 T
A1 2
M0
2
X=1.5C1
1
t
σ0 2 1 0 0 0.5 1.0 1.5 X 0 0.5
t
1.0 1.5 2.0 T
-Y (a) C1T (b)
εR
σm −ε Y
-Y M 15
2 0 10
1 K 5 X
C1T
0 0.5 1.0 1.5
0 0.5 1.0 X
(c) C1T (d)
Fig. 4.16. A linearly-hardening plastic semi-infinite bar subjected to a parabolic-shape impact

loading.
at different locations of the bar, particularly significantly different from those at the bar
end. Figure 4.16(c) shows the maximum stress distribution along the unloading boundary,
σm (X); and Fig. 4.16(d) shows the residual strain distribution along the unloading bound-
ary, εR (X). The length of the plastic deformation region l is 1.32C1 T , which is dependent
on both the plastic wave speed C1 and the loading duration T [Refer to Eq. (4.49)].
Note that on the σ –v plane, the mapping of the weak-discontinuous unloading boundary
coincides with the mapping of the plastic simple waves, namely the line KM0 shown in
0 1 2 3 4 5
4 3 v/vY
1 2
K A4 1
AK A3 A
A2 M0 1
−σ /Y
Fig. 4.17. The mapping of the weak-discontinuous unloading boundary coincides with the mapping
of the plastic simple waves on σ –v plane.
Fig. 4.17. Each characteristic line in the unloading region on the X–t plane is corresponding
to an unloading characteristic line on the σ –v plane. Together with the diagram method,
the locations of points A2 , A3 , A4 , and AK on the σ –v plane as shown in Fig. 4.17 can
be easily determined. The locations of points 1, 2, 3, and 4 shown in the Fig. 4.17 are
determined by the boundary stress levels at the corresponding points on the X–t plane
shown in Fig. 4.16(a).
4.8 Head-On Unloading
So far our discussions about the problem of unloading waves are limited to the semi-
infinite bar. The unloading disturbance and the plastic loading disturbance are propagating
in the same direction (pursuing unloading). With regard to the interactions of two elastic–
plastic waves propagating in the opposite directions, the situations when two head-on
stress waves having the same signs have been discussed in the Section 4.1. In those cases,
the interactions of two head-on waves result in a further plastic loading. In the present
section, we will discuss the cases that the two head-on waves have different signs, and thus
the interaction of these two waves will unload each other, so called a head-on unloading
problem.
To better understand the differences between the head-on unloading and the pursuing
unloading, it is useful to recall some known results in the case of linear elastic wave
propagation. Consider an elastic bar with finite length. The bar is at natural, rest state before
its left end is loaded by a sudden constant-velocity impact (v1 > 0) and simultaneously
its right end is loaded by another sudden constant-velocity impact (v2 > 0), as shown in
Fig. 4.18. Consequently, a compressive shock wave with the stress level σ1 is propagating
to the right direction in the bar, while a tensile shock wave with the stress level σ2 is
propagating to the left direction in the bar. Now let us consider two different unloading
cases that have been discussed previously in Chapter 3. First, if the stress at the left end
of the bar is unloaded to zero, then the particle velocity at the left end should also be
zero. In the σ –v plane this process corresponds to the state-point 1 moving to the point 0.
However, if the rightward compressive stress wave interacts with the leftward tensile stress
wave and thus the stress σ1 is unloaded to zero (under the situation that σ2 = −σ1 ), then
the particle velocity after unloading should change from v1 to v3 = 2v1 . In σ –v plane
this process corresponds to the state-point 1 jumps to the point 3. Therefore, in the above
σ
t 2
3
0 3 0 v
1 2
0 1
0 l X
Fig. 4.18. Head-on unloading in a finite elastic bar.

two situations, although the stress σ1 is unloaded to the same value (zero), the unloading
particle velocity is actually different.
This result can be understood from another way. Assume the bar is in an initial state σ1
and v1 , corresponding to the point 1 on the σ –v plane in Fig. 4.18. If the bar is unloaded
to zero stress by a rightward unloading wave propagating from the left end, the unload-
ing state corresponds to the point 0. However, if the bar is unloaded to zero stress by a
leftward unloading wave propagating from the right end, the unloading state corresponds
to the point 3. Note that the point 3 differs from the point 0 by a velocity difference
[−(2σ1 /ρ0 C0 )]. This difference originates from the fact that when the momentum con-
servation relationship [Eq. (2.35)] is applied to a wave, different signs must be taken
respectively for the leftward wave and the rightward wave.
Now consider an elastic, linearly hardening plastic bar with a finite length and a stress-
free, rest, initial state. When the bar is loaded by a sudden constant velocity v impact at
its left end, a strong-discontinuous elastic wave and a strong-discontinuous plastic wave
simultaneously propagate to the right direction in the bar. The state behind the plastic wave
front corresponds to the point 2 on the σ –v plane in Fig. 4.19. Now if the bar is unloaded
to zero stress by a rightward unloading wave propagating from the left end, by Eq. (2.35)
the unloading state corresponds to the point L in Fig. 4.19 (referring to Fig. 4.5 too), the
residual particle velocity vL is [referring to Eq. (4.14)]:
σ2
vL = v2 + (4.82)
ρ0 C0
However, if the bar is unloaded to zero stress by a leftward elastic unloading wave prop-
agating from the right end, the unloading state corresponds to the point R. By Eq. (2.35)
for a leftward wave, after unloading, the residual particle velocity vR is:
σ2
vR = v2 − (4.83)
ρ0 C0
Thus the latter unloading state differs from the former unloading state by a velocity
difference [−(2σ2 /ρ0 C0 )].
σ2 σ2
σ
ρ0C0 ρ0C0
t
0 L R v
-Y
1
2
L R
2
0 l X
Fig. 4.19. Head-on unloading in a finite elastic-linearly hardening plastic bar.

With the above understanding, we now consider in detail the problem of head-on unloading
when two elastic–plastic stress waves, having different stress signs and propagating in the
opposite direction, interact. Assume a finite bar made of linearly hardening material,
initially in a natural, rest state. Then at its left end a sudden constant velocity compressive
impact (v2 ) is loaded, and simultaneously a sudden constant-velocity tensile impact (v4 )
is loaded at its right end. Consequently, two sets of strong-discontinuous elastic–plastic
waves, having different stress sign and level, are propagating in the opposite direction
within the bar. Before these two sets of waves meet, there exist four zones: the zones
behind the rightward and the leftward elastic wave front respectively, i.e. the zone 1
and the zone 3; and the zones behind the rightward and the leftward plastic wave front
respectively, i.e. the zone 2 and the zone 4, as shown in Fig. 4.20. The two elastic precursor
waves first meet at the point a. As these two waves have the same absolute stress level (Y )
but with different signs, after their interaction the stress is unloaded to zero: σ5 = 0 but
the particle velocity is: v5 = 2vY , a state corresponding to the point 5 on the σ –v plane and
the region 5 on the X–t plane. This stress-velocity state is similar to that after an elastic
wave with stress level Y reflects from a stress-free boundary.
After the internal impact of the two elastic precursor waves, the leftward reflective wave ab,
which is an unloading wave, meets the rightward plastic loading wave ob at the point b.
Imagine that on the left side of the point b if the plastic loading region 2 is also unloaded
to zero stress after the unloading wave ab has passed, then the unloading particle velocity
will be vR , as expressed by Eq. (4.83). If so, on the two sides of the point b, there will be a
particle velocity difference (vR −v5 ), which must result in an internal impact at the point b.
If |vR − v5 | ≤ 2vY , then the two internal reflective waves due to the internal impact are
within the elastic range. The state behind the two reflective wave fronts is corresponding
to the point 6 on the σ –v plane.
Similarly, after the internal impact of the elastic precursor waves, the rightward reflective
wave ac, which is an unloading wave, meets the leftward plastic loading wave lc at the
point c. Again an internal impact occurs and two internal reflective waves are induced.
The state behind these two reflective wave fronts is corresponding to the point 8 on the
σ –v plane, assuming that |vB − v5 | ≤ 2vY .
t
σ 4
3
Y
8
10
6 8 vY 10
b d c 0 5 RB
5 v
2 4 6
1 a 3
0 -Y 1 2
0
l X
Fig. 4.20. Head-on unloading of two elastic-plastic stress waves with opposite stress signs in a
finite elastic-linearly hardening plastic bar if |vR − v5 | > 2vY .
4
σ 9
15 13
t d 3 8 12
14 Y
11 12 13 15
10
7 6 8 9 14
5 10
b 5 c 0 v
2 4
1 3
a 11
0 -Y 1 6
7
0 l X 2
finite elastic-linearly hardening plastic bar if |vR − v5 | > 2vY .
Finally, the rightward reflective wave bd and the leftward reflective wave cd meet at
the point d on X–t plane. Again an internal impact occurs and two reflective waves
originate from this point. The state behind these two reflective wave fronts is corresponding
to the point 10 on the σ –v plane. As the point 10 falls within the elastic range, both
reflective waves emitting from the point d are elastic waves. Having developed to this point,
the process of interaction between two different sign elastic–plastic waves propagating in
opposite directions is completed.
However, if in the previous discussions, we have |vR − v5 | > 2vY or |vB − v5 | > 2vY ,
then after the internal impact (at the points b or c) the induced internal reflective waves
still contain plastic waves, though the intensities of these plastic waves are weakened. This
situation is similar to what has been previously discussed on the pursuing unloading cases.
Figure 4.21 shows one such example. Detailed explanations are omitted. The readers are
suggested to analyze this situation.
Moreover, similar to the discussions about the pursuing unloading, the above results
about the head-on unloading interaction between two strong-discontinuous, elastic–plastic
waves, propagating with constant velocity, can be extended to the other situations, such
as, that the head-on stress waves could be either shock waves (whose wave speed changes
with wave intensity), or weak-discontinuous elastic–plastic waves. In the next section
we will discuss the phenomena when two head-on shock waves unload each other in
an increasingly hardening bar, through a detailed example. Here the head-on unloading
process between two weak-discontinuous elastic–plastic waves is briefly discussed.
As previously discussed on the pursuing unloading problems, a weak-discontinuous

elastic–plastic wave can be approximately considered as a series of incremental loading
waves, each incremental wave is regarded as a strong-discontinuous wave with a small
stress jump. Therefore the head-on unloading between two weak-discontinuous elastic–
plastic waves can be treated as the head-on unloading between two series of incremental
waves. Figure 4.22 shows such an example, which is corresponding to the problem shown
in Fig. 4.20 but now the material is decreasingly hardening. On the X–t plane of Fig. 4.22,
t
σ
h
f
10 Y
d c
6 8
a d
b 5 c 0 v
2 3 4 b
1
a
0 -Y e g
0 l X
finite elastic-increasingly hardening plastic bar.
the regions 0, 2, 4, and 10 are constant-value zones. Their mappings on the σ –t plane
are the point o, g, h, and d, respectively. The regions 1, 3, 6, and 8 on the X–t plane are
simple wave zones. Their mappings on the σ –t plane are the line-segments eg, fh, bd,
and cd, respectively. The unloading zone abcd (the region 5) on the X–t plane, where
the internal reflective waves propagate in both directions, is corresponding to the region
abcd on the σ –t plane. It is noteworthy that stress and particle velocity at the point d,
which indicates the terminating point of the above wave interactions, can be solved
directly without knowing all the intermediate states. Viewing from the left side of the
bar, the state d is arrived by following the loading line oeg (representing a rightward load-
ing wave) and then the unloading line gbd (representing a leftward unloading wave) on
the σ –t plane. By the momentum conservation relationship across the wave front [refer
Eq. (2.63)], and noting that the signs are different for a leftward or a rightward wave,
we have:
σg σd σg
dσ dσ 1 C0 σd
vd = − + =− 1+ dσ +
0 ρ 0 C σg ρ C
0 0 ρ C
0 0 0 C ρ 0 C0
Similarly, viewing from the right side of the bar, we have:

σh σd σh
dσ dσ 1 C0 σd
vd = − = 1+ dσ −
0 ρ0 C σh ρ0 C0 ρ0 C0 0 C ρ 0 C0
Thus the quantities σd and vd can be solved from the above two equations:
σh σg ⎫
1 C0 C0
vd = 1+ dσ − 1+ dσ ⎪⎪
⎬
2ρ0 C0 0 C 0 C
σh σg (4.84)
1 C0 C0 ⎪
⎪
σd = 1+ dσ + 1+ dσ ⎭
2 0 C 0 C
Based on the above discussions for the head-on unloading problems, now we have accu-
mulated experiences to treat different and even more complicated unloading problems,
such as the reflection of an elastic–plastic wave on a free end, the reflection of an elastic–
plastic wave on other surface that may cause unloading, the reflection and transmission
of an elastic–plastic wave on an interface of different media; specifically the unloading
reflection when an elastic–plastic wave propagates from a high wave-impedance media to a
low wave-impedance media. The guiding principles to deal with this kind of elastic–plastic
wave reflection problems are similar to those applied to analyze elastic wave reflections.
Namely, on a boundary surface the prescribed boundary conditions must be satisfied;
or on an internal surface between two media, the stress and particle velocity must be
continuous across the interface. If an unloading disturbance appears and interacts with
other waves, we need to distinguish whether the interaction is a pursuing unloading or a
head-on unloading. Different unloading cases require different treatment. A further dis-
cussion will be given in the following section by analyzing in detail a practical problem,
namely the elastic–plastic wave propagation within a finite length bar impacted on a rigid
anvil.
4.9 High-Speed Impact of a Finite Bar onto a Rigid Target
The case that a flat-head projectile impacts onto a target is a practical problem of impact
dynamics. We can view the projectile as a bar with a finite length, the target as a rigid anvil.
Then the problem is a high-speed impact of a finite length bar on a rigid anvil. Researchers
have used this case as a test method to determine the dynamic mechanical properties of
the materials [Taylor, 1948; Whiffen, 1948; (Lenskii), 1951]. For such kind
of impact problem a head-on unloading takes place, namely elastic–plastic waves will be
reflected at the free end of the bar, and the reflected unloading waves will interact with the
head-on elastic–plastic loading waves.
Suppose that the speed of the projectile is −v∗ . When the projectile impacts on a static
rigid target, the particle velocity at the impact end of the bar is suddenly reduced to zero.
Therefore a set of elastic–plastic waves propagate toward the free end of the projectile.
Now we view the problem in a coordinate system that moves with uniform velocity −v∗ .
In this system, the problem is transferred to another problem where a finite length bar is
originally at a natural, rest state (σ = v = 0) and then is loaded by a sudden constant-value
impact loading v∗ at one end, while the other end of the bar is kept stress-free.
When the rightward elastic–plastic waves with intensity σ ∗ are reflected at the free end,
the final particle velocity at the free end can be determined, by using the momentum
conservation equation across the wave front, as:
σ∗ 0
dσ 0 dσ 1 1 1 σ∗
vfs = − + = + dσ = v∗ − (4.85)
0 ρ0 C σ∗ ρ0 C0 ρ0 σ∗ C C0 ρ0 C0
When C = C0 , we get vfs = 2v∗ , which is the result of the elastic wave reflection. In case
of plastic wave, we know that (1/C0 ) < (1/C), so vfs < 2v∗ . Therefore when the elastic–
plastic waves are reflected at a free end, the end particle velocity is less than twice the
particle velocity of incident wave. It is noteworthy that actually a plastic loading wave
can never reach the free end. Because before it arrives at the free end, the forerunning
elastic precursor wave has arrived at the free end and been reflected back as an unloading
wave, which interacts with the after-running plastic wave and weakens it, until it vanishes
completely. This phenomenon can be understood in another way. We know that the plastic
waves always accompany with residual strains. Since the stress at the free end is always
zero, plastic deformation is impossible to produce at the free end. It reveals that no plastic
wave will ever arrive at the free end. In a finite length bar, this kind of head-on unloading,
where the elastic unloading waves are successively reflected from the free end and repeat-
edly interact with the head-on plastic loading waves, are frequently seen and needed to be
well understood.
4.9.1 Linear hardening bar
Let us first consider the case when the finite length bar is made of linear hardening bar
[ (Lenskii), 1949]. Under this situation both the elastic wave and plastic wave
are strong-discontinuous waves. As shown in Fig. 4.23, before the elastic precursor wave
arrives at the free end (X = l), the situation is same as that in a semi-infinite bar, so the
states of the zones 0, 1, and 2 on the X–t plane are known:
v0 = σ0 = ε0 = 0
Y Y
v1 = vY = , σ1 = −Y, ε1 = −εY = −
ρ0 C0 E

Y C0 εY − v∗
v2 = v∗ , σ2 = ρ0 C1 − v∗ − Y, ε2 = − εY
ρ0 C 0 C1
where Y is the yield stress, εY (=Y /E) is the yield strain.
6*
σ
t 7
vY v* 2vY 5
3
0 6 v
4
8'' 9
F 7'' -Y 1
7' E 2
6'' 6' 5 (b)
C 4'' 4' D
3 σ
2 B 7''7'
1 A 3 5
0 6'' 6' 0 ε
0 XB 4'' 4'
l X
-Y
2 1
v* vY < v* < 2vY
(a) (c)
Fig. 4.23. Impact of a finite linearly-hardening bar onto a rigid target when (4.88) is satisfied.
At the time t = l/C0 , the elastic precursor wave arrives at the free end and is reflected.
The state of the unloading zone 3 is:
2Y
v3 = 2vY = , σ3 = ε3 = 0
ρ0 C0
At the point B, the reflected unloading wave AB meets with the incident plastic wave OB
and unloads it. Imagine that the stress on the left side of the point B is unloaded to zero,
then the particle velocity would be v5 according to Eq. (4.83). However, this velocity v5
differs from the particle velocity v3 on the left side of the point B. Such velocity difference
actually results in an internal impact, which produces two internal reflected waves BC and
BD. The physical state behind the waves BC and BD actually correspond to the point 4
(not the presumed point 5) on the σ –v plane, since in the zone 4 the particle velocity and
the stress should be continuous on both sides of the point B. Therefore, we have:

v5 + v3 1 ∗ C1
v4 = = v − (vY − v∗ ) + vY + 2vy
2 2 C0

C1
3− (vY − v∗ )
C0
= + 2v∗
2

ρ0 C0 (v5 − v3 ) ρ 0 C0 ∗ C 1 ∗
σ4 = − =− v − (vY − v ) + vY − 2vY
2 2 C0
(4.86)
C1
ρ0 C0 1 + (vY − v∗ )
C0
=
2
Note that the strains on both sides of the point B are different. To the right side of the
point B, the strain ε4 is:

C1
1+ (vY − v∗ )
σ4 C0
ε4 = =
E 2C0
while to the left side of the point B, the strain ε4 is:
% &
C1 C12
2+ − 2 (vY − v∗ )
σ2 − σ4 C0 C0
ε4 = ε2 − =
E 2C1
Therefore a stationary strong-discontinuity of strain exists at the section X = XB .
This section is the intersection of the point of lines OB and AB. Namely, (XB /C1 ) =
(2l − XB )/(C0 ). From this condition, the location of the stationary strong-discontinuity
of strain XB can be determined:
2C1 l
XB = (4.87)
C0 + C 1
After the internal impact at the point B, the leftward internal reflective wave BC is an
unloading wave. This wave releases the stress level from the state 2 to the state 4 on the
σ –ε plot in Fig. 4.22(c). On the other side, the rightward internal reflective wave BD is
a weakened loading wave. This wave raises the stress level from the unloading state 3 to
the re-loading state 4 . The concrete value of stress σ4 is dependent on the initial impact
velocity v∗ .
If the initial impact velocity v∗ is not much higher than the yield velocity vY , then the
stress σ4 is below the yield limit, namely −σ4 = |σ4 | ≤ Y = ρ0 C0 vY . From the expression
of σ4 [Eq. (4.86)], we can further get the inequality condition that v∗ must correspondingly
satisfy:
3C0 + C1
vY < v ∗ ≤ vY (4.88)
C0 + C 1
Under this condition, only one plastic zone and consequently one stationary strong-
discontinuity of strain are produced within the bar. After the internal impact at the point B
the rightward reflective wave BD is an elastic wave. Its reflection at the free end is same
as that discussed in the elastic bar problem.
However, when the leftward internal reflective wave BC arrives at the impact end and is
reflected back, the situation becomes complex. Depending on the magnitude of the initial
impact velocity v∗ , we should distinguish the following cases.
First, assume that the impact process still continues, namely the bar is not bounced away
from the target. Therefore, after the unloading wave BC is reflected at the impact end
(the zone 6), the particle velocity should be held as the prescribed one, i.e. v6 = v∗ , the
stress σ6 can be determined by the wave front momentum conservation equation along the
rightward wave CE:
σ6 = σ4 − ρ0 C0 (v6 − v4 ) = ρ0 C0 (2vY − v∗ ) (4.89)
From this equation, it is clear that if
vY < v∗ ≤ 2vY (4.90)
then σ6 ≥ 0, showing a tensile stress state corresponding to the point 6∗ in Fig. 4.23(b).
Note that in practice an impact surface cannot carry any tensile loading, so actually
the stress behind the reflected wave CE can only be unloaded to zero (σ6 = 0). Now the
impacted end of the bar becomes a free end. Corresponding to the zero stress state, the par-
ticle velocity is: v6 = 2vY > v∗ (Note the inequality 4.90). This means that the bar bounces
away from the target, and the assumption of the impact being continuous is untenable. The
practical impact duration is t0 = (2l/C0 ). Rather than the point 6∗ , the physical state of
the zone 6 actually corresponds to the point 6 in the σ –v plane, which coincides with the
point 3.
Now both the waves CE and DE are elastic. They represent an elastic pulse propagating
forward and reflected backward repeatedly in a finite length bar with both free ends.
In other words, the bar is in a free oscillation state. Note that the bar has a stationary strain-
discontinuity at the section B. On the right side of B, the stress–strain state repeatedly
changes between the state 4 to 7 in Fig. 4.23(c), while on the left side of B the stress–strain
state repeatedly changes between the state 4 to 7 [see Fig. 4.23(c)].
We consider another region of the impact velocity. Assume:
3C0 + C1
2vY < v∗ ≤ vY (4.91)
C0 + C 1
Now the stress in the zone 6, σ6 , calculated by Eq. (4.89) is a compressive stress (σ6 < 0).
It means that the impact process continues after the reflection at the point C (as illustrated
in Fig. 4.24), though |σ4 | is still less than Y . Through a discussion similar to the above
one, it can be concluded that the impact process will finish at the point F and the impact
duration is tF = 4l/(C0 + C1 ) = 2tB . There still exists one plastic zone deformed within
the bar, of which the length is also determined by Eq. (4.87).
When v∗ = [(3C0 + C1 )/(C0 + C1 )]vY , we have σ4 = −Y . So if the initial impact

velocity exceeds the value [(3C0 +C1 )/(C0 +C1 )]vY , then the stress level after the internal
impact at the point B must be beyond the yield limit. In this case, a precursor elastic
loading wave BD and a weakened plastic loading wave BB1 propagate from the point
B to the right end of the bar. Afterwards there are two possible cases. In the first case,
the rightward elastic wave BD is reflected as a leftward unloading wave DB1 at the right
end (the free end), and then interacts with the incoming rightward plastic wave BB1 in a
head-on unloading way (Fig. 4.25). In the second case, the leftward elastic wave BC is
reflected as a rightward unloading wave CB2 at the left end (the impacted end), and then
catches up with the forerunning plastic wave BB2 in a pursuing unloading way (Fig. 4.26).
t
σ
8*
10 H
8'' 9
7'' 7' 7
F E
6'' 6' 5 3 v* 9 8
D 0 5 v
C 4'' 4' 6
3 10
B
-Y 1 4
2 1 A
2
0
0 XB l X
3C0+C1
2vY < v* < vY
C0+C1
Fig. 4.24. Impact of a finite linearly-hardening bar onto a rigid target when (4.91) is satisfied.
C0
2+ 5
C1
σ 3C0+C1 5C0+C1
v ν∗ v
17 C0+C1 Y C0+C1 Y
14 16
15 10* 13
13 12
G 16
10 12 vY 3 v* 11
F 0 15
7 10 14
9 11 v
9 17
6 8
5 7 8
C B1 D
4 6
3 -Y 4 5
B 1
2 A
1 2
0
0 XB XB1 l X
Fig. 4.25. Impact of a finite linearly-hardening bar onto a rigid target when (4.92a) is satisfied.
C0
>2+ 5
t C1
19 σ
18 3C0+C1 12*
17 vY <v* < vY
15 16 C0+C1
12 14
11 13
G
9 10 10
F 7 8 19 11
6 3 v* 13 16 8
B2 4 D
0 17 12 v
C 5 14 18
3 15
2 B 6
7
1 A 9
1 4 5
0
0
XB XB2 l X 2
Fig. 4.26. Impact of a finite linearly-hardening bar onto a rigid target when (4.92b) is satisfied.
What actually happens, in practice, depends on the ratio of elastic wave speed to plastic
wave speed, C0 /C1 . A critical situation is that the point B1 in Fig. 4.25 coincides with
the point B2 in Fig. 4. 26. It √
is easy to prove that this critical situation
√ corresponds to the
condition of C0 /C1 = (2 + 5), and thus √ when C 0 /C 1 < (2 + 5), the first case occurs
(Fig. 4.25), while when C0 /C1 > (2 + 5) the second case occurs (Fig. 4.26).
In the case of Fig. 4.25, it can be easily proved that when the initial impact velocity
satisfies:
3C0 + C1 5C0 + C1
vY < v ∗ ≤ vY (4.92)
C0 + C 1 C0 + C 1
the plastic wave vanishes at the point B1 . The two internal reflective waves emitted from
the point B1 are both elastic waves. The second stationary strong-discontinuity of strain
is located at
4C0 C1 l
XB1 =
(C0 + C1 )2
The impact process finishes at the point F . The impact duration is: tF = 4l/(C0 + C1 ).
In the case of Fig. 4.26, it can be easily proved that when the initial impact velocity v∗
satisfies
3C0 + C1
vY < v∗ ≤ 3vY
C0 + C 1
the plastic wave vanishes at the point B2 . The two internal reflective waves emitted from
the point B2 are both elastic waves. No reflected plastic wave will be resulted at the point F
(i.e. the zone 9 is an elastic zone). The second stationary strong-discontinuity of strain is
located at
2C1 l
XB2 =
C0 − C 1
The impact process finishes at the point G. The impact duration is tG = 4l/(C0 + C1 ).
From the above discussions on the problems shown in Figs. 4.23–4.26, we see that in
a finite-length bar loaded by a compressive impact, when two unloading waves interact
in a head-on unloading way, a tensile stress will be induced. For example, the zone 7
in Figs. 4.23 and 4.24, the zone 12 and zone 13 in Fig. 4.25, the zone 10 in Fig. 4.26,
etc., are all tensile stress zones. This phenomenon has been pointed out when discussing
the interactions of elastic waves. Especially, the so-called spalling and other reflective
fracture phenomena have been explained on this basis. Here it is worthwhile to point out
that under a certain condition, the tensile stress induced by the interaction of two unloading
waves can also result in a tensile yield, the so-called yield in the reverse direction or
briefly reverse yield, comparative to the compressive yield occuring previously.
With regard to the tensile stress waves, in the previous analysis on the problems shown in
Figs. 4.23–4.26, we have implicitly assumed that if the absolute value of a tensile stress is
less than the initial yield stress Y or the hardened yield stress in compression, the tensile
stress wave is elastic. In this context we have used the isotropic hardening hypothesis.
This hypothesis assumes that the hardening effect due to compressive plastic deformation
is kept when the material is under tensile loading. However, experimental results have
practically shown that when a material has experienced plastic deformation and hardening
in the positive direction (for example, the compressive direction), its yield stress tends to
be lower under loading in the negative direction (the tensile direction). This is the so-called
Bauschinger effect. The isotropic hardening hypothesis neglects the Bauschinger effect
of the material.
In a material model where the Bauschinger effect is taken into account, it is generally
assumed that the difference between the yield stress in the positive, tensile direction and
the yield stress in the negative, compressive direction is always 2Y , regardless of the
development of plastic hardening. In the three-dimensional stress state this assumption is
called a kinematic hardening hypothesis. Therefore, when the material has been hardened
by compressive plastic stress waves, the tensile stress induced by a head-on unloading wave
interaction may cause the material to yield in the negative direction (tensile direction) even
this tensile stress level is not very high. Another possible situation is when tensile stress
wave passes through a stationary strain discontinuity. Because the tensile yield stress across
this discontinuity is different, when the incident wave is transmitted and reflected at the
strain-discontinuous interface, a reverse yield may be induced. An example is illustrated
in Fig. 4.27. When the leftward unloading elastic wave BD and the rightward elastic
unloading wave CD meet at the point D, a reverse, tensile plastic yield (as shown by the
point 6 on the σ –v plane) takes place due to the Bauschinger effect. Similarly, when the
rightward tensile plastic wave DE passes through the stationary strain-discontinuity XB ,
because the materials across the point XB have different reverse (i.e. tensile) yield stress,
the rightward transmitting wave is an elastic loading wave, while the leftward reflected
wave is a strengthened tensile plastic wave. The zone 8 and the zone 9 on the X–t plane
in Fig. 4.27 are all such kind of reverse (tensile) plastic zones.
Although the discussions so far are about strong-discontinuous elastic–plastic waves with
constant propagating velocities, just as we have pointed out in the previous sections
(Section 4.8 and before), these discussions in principle can be extended to the shock
waves propagating in an increasingly hardening bar, or the weak-discontinuous waves
propagating in a decreasingly hardening bar. We will briefly describe these treatments in
the following sections.
σ
t t
9
9'' 6 8 7
9'
E 10
6 8 7
5 4' 5 3 10
4'' 0
D 3 v
2Y
C B
4
2
1
0 -Y
1
0 XB l X 2
σ
Fig. 4.27. An example showing a reverse, tensile plastic yield due to the Bauschinger effect.
4.9.2 Increasingly hardening bar
Now we consider an increasingly hardening bar impacted onto a rigid anvil with a constant
velocity v∗ (Lee and Tupper, 1954). When the problem is viewed from a coordinate system
that moves with velocity −v∗ , it is identical to the problem that a natural, rest bar is
impacted at the left end by a sudden constant velocity v∗ . The other end of the bar is kept
free. Thus this problem is reduced to the head-on unloading of shock wave propagating
in a finite length bar made of increasingly hardening material. The problem is a little
more complex than the previous one because the speed of the plastic shock wave is now
dependent on its intensity. Several key points to treat this problem are described below.
When the left end of the bar is impacted by a constant velocity v∗ , a strong-discontinuous
elastic precursor wave oa and a plastic shock wave o1 are propagating right with velocity C0
and DJK , respectively, as shown in Fig. 4.28. The shock wave velocity DJK is determined
by the slope of the chord that connects the point J and K on the compressive stress–strain
curve. Here the point J is the yield point, σJ = −Y , while the point K represents the final
state behind the shock wave, which is determined by the impact velocity v∗ . Therefore,
the formula to calculate DJK is:

1 σK + Y
DJK =
ρ 0 εK − ε Y
while the σK should satisfy the momentum conservation equation across the shock wave
front:
σK = −Y − ρ0 DJK (v∗ − vY )
Strictly speaking, the σ –ε relationship used to calculate the velocity of a shock wave should
be the shock adiabat in the form of σ –ε, rather than the usual isothermal or isentropic
stress–strain curve (refer to Section 7.7). However, when the shock wave intensity is not
too high, the entropy-change during a shock process in solids is negligible. In this case,
there are little differences between these three kinds of stress–strain relationships.
The rightward strong-discontinuous elastic loading wave oa is reflected at the free end
(X = L) as a leftward strong-discontinuous elastic unloading wave. This unloading wave
meets the rightward plastic shock wave o1 at the point 1. A head-on unloading process
occurs. This unloading process can be analyzed by the approach stated in the Section 4.8.
The results of this head-on unloading include: a stationary strong-discontinuity of strain
is formed at X1 ; a leftward internal reflective wave 1–A unloads the compressive stress
in the zone K from σK to σN ; while the rightward internal reflective wave, as its intensity
is still higher than the yield limit, has a two-wave structure, namely, behind the rightward
elastic precursor wave 1b, a weakened plastic shock wave 12 propagates in the zone M.
Because the intensity of stress-jump across this shock wave has been lowered to |σN + Y |,
the velocity of the shock wave is thus lowered to DMN , which can be determined by the
slope of the chord MN of the compressive stress–strain curve. (Note that on the σ –ε curve
the point M coincides with the point J , and the stress level at the point N is the same as
C0 t Residual Strain
0.5
40 0.4
c
3
30T 5S 2'V 0.3
MQ
B Q 1' M b
P 2
σ =0
MN
20A N 0.2
L ε =0
1 v=2vY
K σ =Y
10 J ε = εY a 0.1
v=vY
JK
0
0 1 2 3 4 5 6 7 8 9 10 0 2 4 6 X
X
(a) (b)
T'
ST
R enlarged
2vY 4vY v* MN
0 N',S
v
V' K
QV
JK
ST
MQ
P QV T'
Q'' R T
Y MN
J M Q −σ MQ N',S N
V V' V
Q'' Q' Q
S T N J,M
R P' P
K
−σ (c)
0 −ε
(d)
Fig. 4.28. Impact of a finite increasingly-hardening bar onto a rigid target with a constant velocity
v ∗ [Reproduced from Lee & Tupper (1954) by permission of ASME].
that at the point N ):

1 σN + Y
DMN = (4.93)
ρ0 εN − εY
The quantities σN and DMN are calculated by the so-called cut-and-trial method. In fact,
regarding the leftward internal reflective wave 1A, we have:
1
v N − vK = (σN − σK )
ρ0 C0
On the other hand, regarding the rightward reflective plastic shock wave 12, we have:
1
v N − vM = (σN − σM )
ρ0 DMN
Eliminate the quantity vN from the above two equations, we obtain:
1 1
vK + (σN − σK ) = vM − (σN − σM ) (4.94)
ρ0 C0 ρ0 DMN
where vK = v∗ , σK = σ ∗ , vM = 3vY , and σM = −Y are all known quantities. Equa-

tions (4.93) and (4.94) constitute a system of equations with regard to the unknown σN
and DMN . These two equations can be solved by the cut-and-trial method.
The leftward internal reflective wave 1A is reflected at the impacted end. To satisfy the
prescribed boundary velocity condition (v = v∗ ), the reflected wave A2 is an unloading
wave, which unloads the stress σN in the zone N down to σP (in the zone P ). This
unloading wave catches up with the forerunning plastic shock wave 12 at the point 2.
The corresponding wave interaction is a pursuing unloading. After this unloading, the
rightward plastic shock wave 23 is weakened to |σQ + Y |. The leftward internal reflective
loading wave 2B raises the unloading zone stress σP up to σQ . The values of σQ and the
plastic shock wave velocity DMQ can be determined by the similar cut-and-trial method
mentioned above. To the left side of the stationary strain-discontinuity X2 , though σQ is
higher than Y , it is lower than the maximum compressive stress ever reached (|σK | and
|σN |). Therefore the internal reflective wave 2B is still an elastic wave. This wave can
passes through the stationary strain-discontinuity X1 without any change.
As can be seen from the σ –v plot, vQ < v∗ , so that a reflective loading happens when the
internal reflective elastic wave 2B is reflected at the impact-end, raising the compressive
stress σQ in the unloading zone further to σR . However because |σK | > |σR | > |σN |,
when the rightward elastic wave B1 passes through the stationary strain-discontinuity X1 ,
a wave-transmission and -reflection occurs, just like the stress wave that propagates from a
higher yield material to a lower yield material. To the right side of X1 the material re-enters
the plastic deformation zone, namely the transmitted waves are elastic–plastic waves with
two-wave structure. The elastic precursor stress wave 1 2 raises the stress level from σQ
to the yield stress σN = σ5 , while the plastic stress wave 1 5 raises the stress σs in the
zone S up to the stress σT in the zone T . The value of σT and the speed of shock wave
1 5, DST , can be determined by the cut-and-trial method again.
Similarly, when the rightward elastic wave 1 2 reaches the stationary strain-discontinuity
X2 , because the right side of X2 is in a plastic deformation state with stress level σQ
and |σS | > |σQ |, a wave reflection and transmission will occur at point 2 . Behind the
rightward transmitted plastic shock wave the stress level is plastically compressed from
σQ to σV , while behind the leftward reflected elastic unloading wave the stress level is
lowered from σS to σV . The σV and DQV are determined by the cut-and-trial method too.
By the analogy of this approach, we can determine the later wave reflections and
transmissions at the stationary strain-discontinuity X2 , and a series of wave reflection
and transmission at the stationary strain-discontinuity X3 , which is the third strain-

discontinuity induced by the interaction between the reflective unloading wave b3 and
the plastic shock wave 23. These processes are repeated for the third reflective unload-
ing wave reflected from the free end, C4, to interact successively with the plastic shock
wave 34 and the following, until all plastic shock waves are unloaded, weakened, and
finally vanish. At this stage the impact-end losses contact with the target. Afterwards only
the remaining elastic waves propagate back-and-forth within the bar.
After the impact is ended, the residual plastic strain distribution is shown in Fig. 4.28(b).
The plastic deformation notably concentrates at the impacted end of the bar. This reflects
the phenomenon of plastic deformation localization induced by the effect of stress wave
propagation, which is frequently seen in an impact process. Corresponding to the sta-
tionary strain-discontinuities at the points such as 1, 2, 3, the residual strain distribution
displays relatively large jumps at these locations, respectively. Moreover, since an elastic
incident wave may induce a transmitting plastic wave at these strain-discontinuities, some
additional stationary strain-discontinuities may be formed such as at the point 5, but with
relatively small jump of strain. Correspondingly, some relatively small jumps of residual
plastic strain can be seen among the relatively large jumps of residual plastic strain, as
shown in Fig. 4.28(b). Such additional residual plastic strain indicates that the secondary
plastic loading appears in the elastic unloading zone. As has been discussed before, under
a certain condition the material in an elastic unloading state can re-enter into a plastic
deformation state due to the interaction between the elastic wave and the stationary strain-
discontinuity. In the same plot, the dotted line represents the residual strain distribution
that is calculated by the “rigid unloading analysis” method, which will be discussed in the
following Chapter (see Section 5.3).
4.9.3 Decreasingly hardening bar
Finally, we will discuss an example of a finite length bar made of decreasingly hardening
material and loaded by a sudden constant-velocity v∗ at one end, as shown in Fig. 4.29.
t σ
vY v* A' D
0 v
D B'
C B'
B -Y
2 A' A σ* A
1 B
0
0 l X
Fig. 4.29. A finite decreasingly-hardening bar subjected to a suddenly applied constant velocity v ∗ .
It is similar to the problem shown in Fig. 4.23, except that the plastic waves are now all
weak-discontinuous waves.
Obviously, at the time t = l/C0 , the strong-discontinuous precursor elastic loading wave
OA is reflected at the free end as a strong-discontinuous unloading wave AB. On the σ –v plot
this reflection corresponds to the sudden jump from the point A to A . Immediately after the
reflection, the unloading wave AB interacts with a series of rightward weak-discontinuous
centered plastic waves, namely, a series of internal impacts take place. As long as AB
being strong discontinuous, according to the momentum conservation equation across this
leftward strong-discontinuous wave front, we have:
σ̄ − σ = ρ0 C0 (v̄ − v)
Since AB simultaneously is a leftward characteristic line in the unloading zone, we have:

σ̄ + ρ0 C0 v̄ = σ̄ A + ρ0 C0 v̄ A = 2ρ0 C0 vY
Here the point A represents the states behind the strong-discontinuous unloading wave,
which is corresponding to the line AA on the σ –v plot in Fig. 4.29. Eliminating the quantity
σ̄ from the above two equations, and utilizing the simple wave relationship in the loading
zone, we get:
1
σ C0

σ − σ̄ = −Y + 1− dσ (4.95)
2 0 C
As can be seen, with the propagation of the reflected unloading wave from the free end
towards the impact end, the head-on plastic waves are unloaded and simultaneously the
intensity of the unloading wave itself |σ − σ̄ | becomes smaller and smaller. On the σ –v
plot, the line segment AB is the mapping on the plastic side of this unloading wave; and
the line segment A B is the mapping on the elastic unloading side of this unloading wave.
Note that unlike the pursuing unloading case, the intensity of the plastic wave, which
successively interacts with the unloading wave, is stronger in sequence. Suppose that
until the point B, where the intensity of the centered plastic waves reaches the maximum
(σ ∗ , v∗ ), the intensity of the unloading wave, namely the jump |σ − σ̄ |, is still larger than
zero, or according to Eq. (4.95):
σ∗ 1
σ ∗ 1 1

v∗ + = − dσ < 2vY (4.96)
ρ0 C 0 ρ0 0 C0 C
then the strong-discontinuous unloading wave AB can pass through the whole plastic zone.
This condition is the same as Eq. (4.16) in the pursuing unloading case, and is reduced to
Eq. (4.88) when C = C1 .
If the inequality 4.96 is satisfied, then after the strong-discontinuous reflective unloading
wave passes through the point B, its intensity (i.e. the stress-discontinuity across the
unloading wave) is kept as a constant, which corresponds to the line segment BC on the
X–t plot and corresponds to the jump from the point B to the point B on the σ –v plot.
The state variables at the point B can be calculated by Eq. (4.95) under the condition of
σ = σ ∗:
1 ∗
σ̄ B = σ − ρ0 C0 v∗ + 2Y
2

1 ∗ σ∗
v̄ B = v − + 2vY
2 ρ0 C0
Afterwards, the unloading wave BC is reflected at the impact-end denoted by the point C.
The stress σ̄ (C) and particle velocity v̄(C) after reflection is dependent on the magni-
tude of the initial impact velocity v∗ . According to the momentum conservation equation
across a rightward strong-discontinuous wave front, the following relationship between
the quantities σ̄ (C) and v̄(C) is obtained:
" #
σ̄ (C) − σ̄ B = −ρ0 C0 v̄(C) − v̄ B
If the impact process continues, v̄(C) = v∗ , then we have:
σ̄ (C) = 2Y − ρ0 C0 v∗
On the other hand, if the impact process continues, then σ̄ (C) should be in a compressive
state, thus we have,
v∗ > 2vY (4.97)
If the above inequalities are not satisfied, the impact process finishes at the point C. The
impact-end becomes a free end, with σ̄ (C) = 0 and v̄(C) = 2vY. This means that the bar
is bounced away from the target by a velocity-difference, 2vY − v∗ . This situation is same
as that discussed previously in Fig. 4.23 [refer to Eq. (4.90)].
If v∗ simultaneously satisfies Eqs. (4.97) and (4.96), then although the strong-
discontinuous unloading wave AB can pass through the plastic centered wave zone, the
impact process does not finish at the point C. Under a certain circumstance the unloaded
zone may be re-loaded plastically, forming a secondary plastic zone. For example, once
the stress level of the internal reflective wave propagating along the rightward character-
istic line BD is larger than the historical maximum stress σm (X) at a certain section along
the unloading boundary AB, a plastic re-loading will occur to form a secondary plastic
zone. In the present situation, since the σm (X) reaches its maximum at the point B and
decreases with X towards the point A, the possibility of such secondary plastic re-loading
is understandable.
If the impact velocity is so large that the inequality 4.96 is not satisfied, then the strong-
discontinuous reflective unloading wave vanishes at a certain point K within the plastic
loading zone. Namely the unloading wave is absorbed in the plastic zone. The stress level
σK can be calculated by Eq. (4.95) under the condition of σK − σ̄K = 0:
σK
C0
1− dσ = 2Y
0 C
After the unloading wave is absorbed, on the left side of the point K the rightward plastic
waves are continuous to propagate; on the right side of the point K, it is also possible to
form secondary plastic zones.
In order to better understand the secondary plastic zones, and to analyze the propagation of
elastic–plastic boundaries under more general conditions, it is necessary to study further
the general features of elastic–plastic boundary propagation, which will be discussed in
the next section.
4.10 General Properties of the Loading–Unloading Boundary Propagation
4.10.1 Loading boundary and unloading boundary
In general, the boundary between the plastic loading zone and the elastic unloading zone
varies with the propagation and interaction of stress waves. In the case of one-dimensional
bar problem, the location of elastic–plastic boundary, changes with time. In the X–t plane
if the locus of this boundary is expressed by a function X = f (t), then the moving speed
or propagating speed of this boundary, C, is:
= dX = f (t)
C (4.98)
dt
Elastic–plastic boundaries can be categorized as two types. As shown in Fig. 4.30, when
looking at the change of state across an elastic–plastic boundary with the increase of time,
if the state changes from a plastic-loading state to an elastic unloading state, then this part
of the boundary is called an unloading boundary, for which the propagating velocity is
t Unloading Boundary
Secondary Plastic
Loading Zone
Secondary Plastic
Loading Zone
Loading Bounary
Elastic
Unloading Zone Unloading Boundary
Plastic
Loading Zone
0 X
Fig. 4.30. Two categories of elastic-plastic boundaries: loading boundary and unloading boundary.
denoted by C U . On the contrary, if the state changes from an elastic state to a plastic-
loading state, including the change from an elastic unloading state to a secondary-plastic
loading state, then this part of the boundary is called a loading boundary, for which the
propagating velocity is denoted by C L . For a continuous elastic–plastic boundary, once its

slope, namely the moving speed C, changes sign, it corresponds to a transition between
these two types of boundaries.
It is worthwhile to emphasize again, unlike a characteristic line that represents the prop-
agating locus of a mechanical disturbance (stress wave), an elastic–plastic boundary, in
general, is not the propagating locus of a mechanical disturbance itself, but is the linking
or moving locus of such critical points X, wherein due to the interaction of stress waves
a transition occurs at time t either from the plastic loading state to the elastic unloading
state, or from the elastic (either loading or unloading) state to the plastic loading state.
The former is called an unloading boundary while the latter is called a loading boundary.
Under certain specific conditions, as we have discussed before, an elastic–plastic boundary
may coincide with a propagating locus of mechanical disturbance (namely a characteristic
line). Except for those specific cases, in general, it is necessary to distinguish the boundary
propagating velocity C from the elastic wave speed Ce and the plastic wave speed Cp .
With regard to the loading boundary, consider an arbitrary material particle on the bound-
ary. Before the boundary passes, the particle is in an elastic loading state. After the
boundary passes, the particle point is in a plastic loading state. Therefore, at the time
when the boundary passes the particle, the slopes of the stress–time curves on both sides
of the boundary have the same sign, as:
:
∂σ e ∂σ p
≥0
∂t ∂t
Hereafter the superscripts (or subscripts) “e” and “p” stand for elastic and plastic,
respectively.
However, in case that (∂σ e /∂t) = (∂σ p /∂t) = 0, we should consider the signs of the
2nd order derivatives. If, before and after the loading boundary passes the particle, the
curvatures of the stress–time curves changes from nonpositive to nonnegative (tensile
loading), or changes from nonnegative to nonpositive (compressive loading), then:
:
∂ 2σ e ∂ 2σ p
≤0
∂t 2 ∂t 2
With regard to the unloading boundary, consider an arbitrary material particle on the
boundary. Now, when the boundary passes the particle, the state changes from a plastic
loading process to an elastic unloading process, so we have:
:
∂σ e ∂σ p
≤0
∂t ∂t
In case that (∂σ e /∂t) = (∂σ p /∂t) = 0, before and after the unloading boundary passes
the particle, the curvatures of the stress–time curves have the same signs, as:
:
∂ 2σ e ∂ 2σ p
≥0
∂t 2 ∂t 2
Summarizing the above discussions, on a loading boundary we have:

: :
∂σ e ∂σ p ∂ 2σ e ∂ 2σ p
≥0 or ≤0 (4.99)
∂t ∂t ∂t 2 ∂t 2
while on an unloading boundary we have:

: :
∂σ e ∂σ p ∂ 2σ e ∂ 2σ p
≤0 or ≥0 (4.100)
∂t ∂t ∂t 2 ∂t 2
4.10.2 Elastic–plastic boundary as a singular interface
In Section 4.7 we have discussed the propagation features of an unloading boundary

by means of the characteristic line method, wherein the bar length is semi-infinite and
consequently the plastic loading zone is a simple wave zone. Now we are going to introduce
a new method – singular interface method – to discuss the problem in more general cases.
This approach provides the basic relationships to determine the location of a loading or an
unloading boundary. Once the location of an elastic–plastic boundary is determined, the
wave problems that include unloading process and/or secondary plastic loading process
can be solved easily.
When an elastic–plastic boundary passes through a material particle, the particle either
is loaded from an elastic state into a plastic state, or is unloaded from a plastic state into
an elastic state. In both cases, the mechanical properties of material display an essential
change. Correspondingly, as we usually assume, a discontinuity of the slope of stress–strain
curve, dσ/dε occurs, a kind of singularity with respect to the slope of the stress–strain
curve. Consequently, a discontinuity of the wave speed C occurs, namely, we have
[C] = Ce − Cp
= 0 (4.101)

Or, by introducing a nondimensional parameter γ = C/C, the discontinuity of wave
speed can be re-expressed as:
[γ ] = γe − γp
= 0 (4.101 )
Note that because of the mentioned singularity in the material constitutive relationship,
the state variables σ and v or their derivatives are discontinuous across the elastic–plastic
boundary. Therefore the elastic–plastic boundary is actually a kind of singular interface.
This singular interface is different from the wave front. A wave front represents the prop-
agation of a mechanical disturbance in media. It is the interface between disturbed media
and undisturbed media, reflecting the singularities of mechanical variables. On the other
hand, when an elastic–plastic boundary propagates in media, it carries the singularities
of the material constitutive property, as well as the singularities of state variables. Note
that during a propagating process of an elastic–plastic boundary, the singularities of the
material constitutive property are invariable, while the singularities of state variables may
vary. No matter what the situation, as will be proved, the state variable singularities and the
changes of these singularities along the boundary, in the final analysis, are determined by
the discontinuity of the stress–strain curve slope (i.e. the wave speed discontinuity). Thus,
we shall call the wave speed discontinuity across an elastic–plastic boundary [expressed
by Eq. (4.101)] the basic singularity. Accordingly, the hypothesis that the basic singular-
ity exists is called the basic assumption. The basic assumption [Eq. (4.101)] is the basis
for studying the propagation features of an elastic–plastic boundary (Wang et al., 1983).
Otherwise, if across a boundary [C] = 0, namely dσ /dε is continuous regardless of
loading or unloading, then the problem is a simple nonlinear wave problem, which can be
solved by usual approaches stated previously. In such cases, an elastic–plastic boundary
coincides with the corresponding characteristic line, and there is no need to be treated
specially.
Since the elastic–plastic boundary propagation is viewed as the propagation of a specific

singularity interface, we can use the singularity interface method similar to that used in
Section 2.7, in addition to the basic singularity expressed by Eq. (4.101), to deduce the
kinematic compatibility equation and dynamic compatibility equation across an elastic–
plastic boundary.
Thus, consider the discontinuity of an arbitrary physical quantity F across an elastic–plastic

boundary
[F ] = F e − F p
If we follow the propagation of the elastic–plastic boundary to observe the variance of the
discontinuity of F , then similar to the “wave front derivative” in Eq. (2.51), now we have
the “boundary derivative”, namely the total derivative along the boundary:
d[F ] ∂[F ] ∂[F ]

= +C (4.102)
dt ∂t ∂X
This boundary derivative will be applied frequently to discuss the features of an elastic–
plastic boundary.
4.10.3 Strong-discontinuous elastic–plastic boundary
First, we consider the case that the elastic–plastic boundary is a strong-discontinuity.

Replacing the F in Eq. (4.102) with the particle displacement u and noting that the
displacement across the boundary should be continuous, we get the displacement
continuity condition (also called the kinematic compatibility condition) across the
strong-discontinuous boundary, as:

[v] = −C[ε] (4.103)
Additionally, similar to Eq. (2.57), we have the momentum conservation condi-

tion (also called the dynamic compatibility condition) across the strong-discontinuous
boundary, as:

[σ ] = −ρ0 C[v] (4.104)
From Eqs. (4.103) and (4.104), we obtain:

= 1 [σ ]
C (4.105)
ρ0 [ε]
The above equation does not contradict the Eq. (4.101), so a strong-discontinuous elastic–
plastic boundary can exist. Under the specific condition that a jump of elastic unloading
stress exists across the boundary, therefore:
[σ ] = Ce
= E, C
[ε]
which means that the propagating locus of an unloading boundary coincides with that of
the strong-discontinuous unloading disturbance. This is the case that has been discussed
in Section 4.9. Another specific case is that a jump of plastic loading stress exists across
the boundary. In this case the propagating locus of a loading boundary coincides with that
of the strong-discontinuous plastic loading disturbance (or plastic shock wave). In case of
linear hardening material, we have:
[σ ] = C1
= E1 , C
[ε]
This is the case that has been discussed in the problem shown in Fig. 4.12. If across
= 0. This represents a stationary
the boundary, we have [σ ] = 0 but [ε]
= 0, then C
strain-discontinuous interface.
4.10.4 First order weak-discontinuous boundary
Second, we consider the case that the elastic–plastic boundary is a weak-discontinuity.

In this case the stress σ and the particle velocity v themselves are all continuous across
the boundary but their derivatives are discontinuous. It should be emphasized again that,
hereafter, when material is unloaded from a plastic state to an elastic state, or is loaded from
an elastic loading/unloading state to a plastic state, we assume that the stress–strain curve
is continuous, but the slope of stress–strain curve is discontinuous, including the case of
initial yield. Thus, Eq. (4.101) must be satisfied on any elastic–plastic boundary. In fact,
if the slope of stress–strain curve is continuous at the initial yield, then the elastic–plastic
boundary coincides with the elastic characteristic line, and the problem can be easily
solved.
Since σ and v are continuous across an elastic–plastic boundary, we can replace the F in
Eq. (4.102) with σ and v, obtaining:

d[σ ] dσ ∂σ ∂σ
= = +C =0
dt dt ∂t ∂X

d[v] dv ∂v ∂v = 0
= = +C (4.106)
dt dt ∂t ∂X
Moreover, when the first order partial derivative governing equations are expressed in
terms of the dependent variables σ and v, they have the same form as whatever an elastic
zone or a plastic zone is concerned, or whatever a loading process or an unloading process
is concerned [refer to Eqs. (2.13), (2.17), and (4.9)], namely we have:
∂v 1 ∂σ
=
∂X ρ0 C 2 ∂t
(e, p) (4.107)
∂v 1 ∂σ
=
∂t ρ0 ∂X
Here (e, p) means that the equations are suitable to both the elastic zone and the plastic
zone, although the wave speed should be chosen as Ce or Cp , respectively, depending on
the zone concerned. Rewriting the Eq. (4.107) in the form of the discontinuous values (the
jumps) across the elastic–plastic boundary and noting Eq. (4.101), we get:

∂v 1 1 ∂σ
=
∂X ρ0 C 2 ∂t
(4.107 )
∂v 1 ∂σ
=
∂t ρ0 ∂X
From Eq. (4.107), if we define:

∂σ ∂v
K1 ≡ = ρ0 C 2
∂t ∂X
(4.108)
∂v ∂σ
J1 ≡ ρ0 C =C
∂t ∂X
substituting them into Eq. (4.106), then the Eq. (4.107 ) is reduced to:
[K1 + γJ1 ] = 0
(4.109)
1
(J1 + γK1 ) = 0
C
The above two equations are the basic equations that a weak-discontinuous elastic–plastic
boundary must satisfy.
Eliminating the quantity J1 from the above two equations, we get:

1 − γ 2 K1 = 0 (4.110)
which can be rewritten in the following expanded form:
∂σ p 2
C
1−
Ce2
∂t = (4.110 )
∂σ e 2
C
1−
∂t Cp2
Note that because of the basic assumption that across an elastic–plastic boundary the
wave speed is discontinuous, namely Ce
= Cp , therefore from Eq. (4.110 ) we have
(∂σ e /∂t)
= (∂σ p /∂t), namely [K1 ] = [∂σ/∂t]
= 0. Consequently, from the first equa-

tion of Eq. (4.109) we also have: [γJ1 ] = ρ0 C[∂v/∂t]
= 0. Finally from Eq. (4.106) we
know [∂σ/∂X]
= 0, [∂v/∂X]
= 0. In other words, across the elastic–plastic boundary, all
the first order derivatives of σ and v with respect to X and t must be discontinuous, except
in the case that (∂σ e /∂t) = (∂σ p /∂t) = 0 (which will be discussed later). Thus we get
an important Theorem of the elastic–plastic boundary.
Theorem 1: Unless on the whole elastic–plastic boundary (∂σ e /∂t) = (∂σ p /∂t) = 0,
all the first order partial derivatives of σ and v are discontinuous, i.e. the elastic–plastic
boundary is a first order weak-discontinuity of σ and v.
From Eq. (4.110), we can directly get the expression for the propagating velocity of this
first order weak-discontinuity:

∂σ

∂t
C = (4.111)
1 ∂σ
C 2 ∂t
which is the same as the one we deduced in Section 4.7 [Eq. (4.73)] for a semi-infinite
bar by using characteristic line method. Here, the equation is deduced for general sit-
uations, applicable to both the unloading boundary and the loading boundary and no
limitation of whether the plastic zone is a simple wave zone or not, etc. The equation
expressed in the form of Eq. (4.111) was first obtained by Karman, Bohnenblust and Hyers
(Karman et al., 1942).
The first order weak-discontinuous boundary can be determined by the following approach,
depending on the specific situation. Assume that in the X–t plot (Fig. 4.31), the solution
before t ≤ t1 is known. Curve GF is a determined unloading boundary moving from the
right lower direction toward the point F . We now try to determine the boundary point P
at the next time-step t + ∆t.
Starting from the point F , we draw five straight lines with slope of ±Ce , ±Cp , and 0 (the
line FK), respectively. These lines divide the half plane (t ≥ t1 ) into six zones: the zone I,
K
+Cp
–Cp
III IV
V
II +Ce
–Ce
t=t1 I VI
t F Unloading Zone
Plastic Zone
X G
Fig. 4.31. Six possible zones exist for a weak-discontinuous boundary.
II, III, IV, V, and VI. If the next boundary segment FP lies to the left side of the line FK,
meaning that C does not change sign, then FP is still an unloading boundary. On the
changes sign, then FP
contrary, if FP lies to the right side of the line FK, meaning that C
is a loading boundary, forming the secondary plastic zone.
With regard to the unloading boundary, since (∂σ e /∂t)/(∂σ p /∂t) ≤ 0 [Eq. (4.100)], then
must satisfy the following inequality [comparing to Eq. (4.74)]:
from Eq. (4.110 ), C
U | ≥ Cp
Ce ≥ |C (4.112)
which means that the successive boundary segment FP must fall into the zone II. Note
that if GF moves from the left lower direction toward the point F , then FP must fall into
the zone V, which is the mirror symmetry of the zone II.
With regard to the loading boundary, including the secondary plastic loading boundary,
must satisfy either
since (∂σ e /∂t)/(∂σ p /∂t) ≥ 0 [Eq. (4.99)], then from Eq. (4.110 ), C
of the following inequality:
⎧
⎨Cp ≥
⎪ CL > 0
or (4.113)
⎩
⎪
CL ≥ Ce
which means that the successive boundary segment FP must fall into either zone IV, or zone
VI, corresponding to the first inequality or the second inequality of (4.113), respectively.
In case that GF moves from the left lower direction toward the point F , then FP must fall
into the zone III or the zone I.
III
II
P
I
n1
t=t1
F unloading zone
n2 n3
plastic zone
G
(a)
IV
V
P
n1
unlo t=t1
adin
n2 F σm g zon
e
plastic zone
G
(b)
V
VI
t P
n4 n1 t=t1
F unloading zone
σm
X
plastic zone G
(c)
Fig. 4.32. Determination of the successive first order weak-discontinuity boundary in different
zones.
What actually happened will be determined by the specific situation as solved before the
time t1 . Whatever the zone the successive boundary segment FP falls into, all situations
can be solved.
If FP is in zone II, then in the unloading zone a characteristic line n1 can be drawn from
the point P [see Fig. 4.32(a)]:
σe
n1: dX + Ce dt = 0, ve + = β1
ρ0 C0
and in the plastic zone two characteristic lines n2 and n3 can be drawn from the point P :
n2: dX − Cp dt = 0, vp − φ(σ p ) = α2
n3: dX + Cp dt = 0, vp + φ(σ p ) = β3
The Riemann invariables β1 , α2 , β3 in the above expressions are all known. The above
three equations, in addition to the stress continuous condition σ e = σ p and the particle
velocity continuous condition ve = vp , five equations totally are sufficient to solve the
Therefore the point P is determined.
five unknown quantities: σ e , σ p , ve , vp , and C.
If FP is in zone IV, then from the point P a characteristic line n1 in the elastic zone and
a characteristic line n2 in the plastic zone can be drawn [see Fig. 4.32(b)]. These give us
two compatibility equations containing two known Riemann invariables. In addition, at
the point P the stress must be continuous and satisfy the yield condition: σ e = σ p = σm
(two equations), and the particle velocity must be continuous: ve = vp . The five equations
or can be
totally are sufficient to solve the five unknown quantities: σ e, σ p, ve, vp, and C,
reduced to solving the following equation:
ρ0 Ce α2 − ρ0 Ce β3 = σm + ρ0 Ce φ(σm )
which can be solved by a cut-and-trial method.
If FP is in zone VI, then from the point P two characteristic lines n1 and n4 in the unloading
zone can be drawn [see Fig. 4.32(c)]. The expressions of n1 have been given previously
and the expression of the characteristic line n4 is:
σe
n4: dX − Ce dt = 0, ve − = α4
ρ0 C0
where the Riemann invariables α4 is also known. In addition, at the point P we have
conditions: σ e = σ p = σm , ve = vp . The five equations totally are sufficient to solve the
five unknown quantities, or can be reduced to solving the following equation:
ρ0 Ce (β1 − α4 ) = 2σm
which can be solved by a cut-and-trial method.
As to the cases when GF moves from the left side, and thus FP falls into zone I, III, or V,
the problem can be solved in the similar way.
The problem of determining the elastic–plastic boundary for a finite length bar made of
decreasingly hardening material, under the boundary condition that one end is fixed and
another end is loaded by a rectangular impact loading, has been discussed by Bohnenblust
et al. (1942). For the problem of a decreasingly hardening bar impacted onto a rigid anvil,
Lee analyzed the complex phenomenon wherein an elastic–plastic boundary propagates
in the zone II, IV, and V (Lee, 1952). However, at that time they didn’t adequately treat
the situation when ∂σ e /∂t and ∂σ p /∂t are simultaneously zero.
4.10.5 Second order weak-discontinuous boundary
From the previous discussions about the first order weak-discontinuous boundary, we
know that if across the boundary (∂σ e /∂t) = (∂σ p /∂t) = 0, then by Eq. (4.106)
and Eq. (4.107 ),

∂σ ∂σ ∂v ∂v
= = = =0 (4.114)
∂t ∂X ∂t ∂X
Further, from Eq. (4.107), which is a combination of the continuation equation and the
constitutive relationship and is satisfied on both the elastic side and the plastic side across
the boundary, therefore we have (∂ve /∂X) = (∂vp /∂X) = 0. This is another important
property of the elastic–plastic boundary.
Theorem 2: If across the whole elastic–plastic boundary (∂σ e /∂t) = (∂σ p /∂t) = 0,
then all the first order partial derivatives of σ and v are continuous everywhere along the
boundary, and (∂ve /∂X) and (∂vp /∂X) must simultaneously be zero.
In this case, the boundary propagating velocity expressed by Eq. (4.111) becomes an
indefinite 0/0 form. To determine the propagation features of the boundary it is necessary
to consider the higher order derivatives of σ and v.
Taking the boundary derivatives of the 1st order derivatives of σ and v, utilizing Theorem 2,
we obtain:

d ∂σ ∂ 2σ 2
∂ σ = 0, (e, p)
= 2 +C
dt ∂t ∂t ∂t∂X
2 2
d ∂σ ∂ σ ∂ σ
= +C =0
(4.115)
d ∂v ∂ 2v 2
∂ v = 0, (e, p)
= +C
2 2
d ∂v ∂ v ∂ v =0
= + C
dt ∂t ∂t 2 ∂t∂X
where the (e, p) means that the equations are suitable to both the elastic zone and the
plastic zone across the boundary.
Differentiate the first order partial differential governing equations Eq. (4.107), which are
suitable to both the plastic zone and the elastic zone, with respect to variables X and t,
we obtain:

∂ 2v 1 ∂ 2σ 1 ∂σ ∂ 1
= 2 2
+
∂t∂X ρ0 C ∂t ρ0 ∂t ∂tC2

∂ 2v 1 ∂ 2σ 1 ∂σ ∂ 1
= +
∂X 2 ρ0 C 2 ∂t∂X ρ0 ∂t ∂X C 2
∂ 2v 1 ∂ 2σ
=
∂t 2 ρ0 ∂t∂X
∂ 2v 1 ∂ 2σ
=
∂t∂X ρ0 ∂X 2
The above four equations contain six second order derivatives of σ and v. Since we have
assumed that across the boundary (∂σ/∂t) = 0 (e, p), only two second order derivatives
are independent quantities, others can be separated into two groups and interrelated by the
following equations, by defining two parameters K2 and J2 for convenience:
∂ 2σ 2
2 ∂ v
2
2∂ σ
K2 ≡ = ρ0 C = C
∂t 2 ∂t∂X ∂X 2
(4.116)
∂ 2v ∂ 2σ ∂ 2v
J2 ≡ ρ0 C 2 = C = ρ0 C 3 2
∂t ∂t∂X ∂X
Substituting them into Eq. (4.115), we get:
K2 + γ J2 = 0, (e, p)
(4.117)
1
(J2 + γ K2 ) = 0
C
The first equation of Eq. (4.117) is suitable to both sides across the boundary. The total
three equations of Eq. (4.117) are the basic equations that a higher-than-first order weak-
discontinuous elastic–plastic boundary must satisfy.
Eliminate quantity J2 in Eq. (4.117), we get:

!
1 − γ 2 K2 = 0 (4.118)
By the basic assumption, we have [γ ]

= 0 [Eq. (4.101)]. From the above equation
we know that [K2 ] = [γ 2 K2 ]
= 0 (unless [K2 ] = 0 and consequently (∂ 2 σ e /∂t 2 ) =
(∂ 2 σ p /∂t 2 ) = 0). Moreover from Eq. (4.117) we also know that [γ J2 ]
= 0. Finally from
Eq. (4.115) we know that all the secondary partial derivatives of σ and v are discontinuous
across the boundary, except in the case that (∂ 2 σ e /∂t 2 ) = (∂ 2 σ p /∂t 2 ) = 0 (which will be
discussed further). Thus we get another important property of the elastic-plastic boundary.
Theorem 3: If across the whole elastic–plastic boundary (∂σ e /∂t) = (∂σ p /∂t) = 0,
then unless on the boundary ∂ 2 σ e /∂t 2 and ∂ 2 σ p /∂t 2 are continuous, all the second order
partial derivatives of σ and v are discontinuous. In this case the elastic-plastic boundary
is a second order weak-discontinuity of σ and v.
From Eq. (4.118), the propagating velocity of this second order weak-discontinuity can
be determined by the following formula:

2
∂ σ

∂t 2
=
C (4.119)
1 ∂ 2σ
C 2 ∂t 2
This formula was obtained by Clifton and Ting (1968) in a special case of elastic–plastic
wave propagation. Nevertheless, their general formulae for a boundary with higher-than-
second order discontinuity is not likely adequate (refer to the theorem 5 stated below).
should also
From the first equation of Eq. (4.117), the boundary propagation velocity C
simultaneously satisfy the following two equations:
∂ 2σ e ∂ 2σ e
2 2
= − ∂t = − ∂t
C 2
∂ σ e 2
∂ ve
ρ0 2
∂t∂X ∂t
∂ 2σ p ∂ 2σ p
2 2
= − ∂t = − ∂t
C (4.119 )
2
∂ σ p 2
∂ vp
ρ0 2
∂t∂X ∂t
The above equations do not contradict the Eq. (4.119), so it is possible that an elastic–
plastic boundary satisfies both Eqs. (4.119) and (4.119 ). Therefore a second order weak-
discontinuous boundary can exist realistically.
In the X–t plane, the locations of a second order weak-discontinuous boundary are rather
different from that of a first order weak-discontinuous boundary.
With regard to the unloading boundary, since the curvatures of σ –t curve have the
same signs on the elastic side and the plastic side across the boundary, namely
must
(∂ 2 σ e /∂t 2 )/(∂ 2 σ p /∂t 2 ) ≥ 0 [Eq. (4.100)], so from Eq. (4.118), we know that C
satisfy either one of the following inequalities:
⎧
⎪
⎨Cp ≥ |C U | ≥ 0
or
⎪
⎩|C U | ≥ C e
which means that the successive unloading boundary should fall into zone VI or IV, or
their mirror symmetric zones I or III (Fig. 4.33).
With regard to the loading boundary,

because
the curvatures
of σ –t curve change signs
across the boundary, namely ∂ 2 σ e /∂t 2 / ∂ 2 σ p /∂t 2 ≤ 0 [Eq. (4.99)], from Eq. (4.118),
we know that C must satisfy the following inequality:
L | > Cp
Ce ≥ |C
which means that the successive unloading boundary should fall into zone II or its mirror
symmetric zone V (Fig. 4.33).
By comparing the above results with Eqs. (4.112) and (4.113), it is thus clear that the
U , or a loading boundary velocity C
possible scopes of an unloading boundary velocity C L ,
–Cp Cp
III IV
II V Ce
–Ce
t I VI
−∞ M +∞
Fig. 4.33. Determination of the successive second order weak-discontinuity boundary in different
zones.
in the case of a second order weak-discontinuous boundary are just the counterchange of
that in the case of first order weak-discontinuous boundary. The combination of those for
these two cases just cover the whole possible range on the X–t plot from X = −∞ to
X = ∞, as shown in Fig. 4.33, and correspondingly in Table 4.2.
4.10.6 Discussions on higher-than-second order weak-discontinuous boundaries
In the previous discussions, we have deduced the basic equations for the higher-than-first
order weak-discontinuous boundary, Eq. (4.117), as:
K2 + γJ2 = 0, (e, p)

1
(J2 + γK2 ) = 0
C
p
If K2e (≡ ∂ 2 σ e /∂t 2 ) and K2 (≡ ∂ 2 σ p /∂t 2 ) are continuous everywhere along the boundary,
then from Eq. (4.118) we must have (∂ 2 σ e /∂t 2 ) = (∂ 2 σ p /∂t 2 ) = 0. Furthermore
from the first equation of Eq. (4.117) we get (γJ2 )e = (∂ 2 ve /∂t 2 ) = 0 and (γJ2 )p =
(∂ 2 vp /∂t 2 ) = 0. Therefore the second equation of Eq. (4.117) is satisfied automatically.
Table 4.2. The possible scopes of both the first and the second order weak-discontinuous loading-
unloading boundaries.
Elastic-plastic Loading boundary Unloading boundary

boundary
1st order weak L | ≥ Ce, (zone I & VI)
|C U | ≤ Ce , (zone II & V)
Cp ≤ |C
discontinuity L | ≤ Cp , (zone III & IV)
|C
2nd order weak L | ≤ Ce , (zone II & V)

Cp ≤ |C U | ≥ Ce , (zone I & VI)
|C
discontinuity
|CU | ≤ Cp , (zone III & IV)
From the definitions of K2 and J2 [Eq. (4.116)] we can deduce that on both the elastic side
and the plastic side across the boundary the following relations exist:
∂ 2σ ∂ 2σ ∂ 2σ ∂ 2v ∂ 2v ∂ 2v
= = = = = = 0, (e, p)
∂t 2 ∂t∂X ∂X 2 ∂t 2 ∂t∂X ∂X 2
Thus we obtain the fourth theorem of the elastic–plastic boundary as follows.
Theorem 4: If everywhere along the whole elastic–plastic boundary (∂σ e /∂t) =

(∂σ p /∂t) = 0 and simultaneously (∂ 2 σ e /∂t 2 ) = (∂ 2 σ p /∂t 2 ), then all the second order
partial derivatives of σ and v are continuous and are all equal to zero.
In this case, the boundary propagating velocity expressed by either formula expressed in
Eq. (4.119) becomes an indefinite 0/0 form. It seems that we need to consider the third
order derivatives of σ and v in order to determine the propagating velocity of the boundary.
However, we can immediately prove that because of the basic singularity expressed by
Eq. (4.101), the higher-than-second order weak-discontinuous elastic–plastic boundary is
actually impossible to exist, unless this boundary coincides with the characteristic line.
The proof is stated below.
Suppose that this kind of weak-discontinuity exists. According to theorem 4, along the
whole boundary, it must have

d ∂ 2σ d ∂ 2σ d ∂ 2σ d ∂ 2v
= = =
dt ∂t 2 dt ∂t∂X dt ∂X 2 dt ∂t 2
2 (4.120)
d ∂ 2v d ∂ v
= = =0
dt ∂t∂X dt ∂X 2
On the other hand, in both the plastic zone and the elastic zone across the boundary,
differentiating the governing equations [Eq. (4.107)] twice, and utilizing the preconditions
that (∂σ/∂t) = 0 and (∂ 2 σ/∂t 2 ) = 0, it can be shown that among the eight 3rd order partial
derivatives of σ and v, only two of them are independent, others can be separated into two
groups and interrelated by the following equations, by defining two parameters K3 and J3
for convenience:
∂ 3σ ∂ 3v ∂ 3σ ∂ 3v
K3 ≡ 3
= ρ0 C 2 = C2 2
= ρ0 C 4 3
∂t ∂t ∂X ∂t∂X ∂X
∂ 3v ∂ 3σ 3
3 ∂ σ
3
3∂ σ
J3 ≡ ρ0 C = C = ρ0 C = C (4.121)
∂t 3 ∂t 2 ∂X ∂t∂X 2 ∂X 3
Substitute Eq. (4.121) into Eq. (4.120), and recall that the total derivative following the

boundary, namely the “boundary derivative”, is defined as: (d/dt) = (∂/∂t) + C(∂/∂X)
[see Eq. (4.102)], we finally get:
K3 + γJ3 = 0 (e, p)
J3 + γK3 = 0 (e, p) (4.122)
The above equations are suitable to both the elastic side and the plastic side across the
boundary. So Eq. (4.122) is equivalent to a system of homogeneous linear algebraic equa-
p p
tions with respect to four unknown quantities K3e , K3 , J3e , and J3 . The sufficient and
necessary conditions that a nontrivial solution exists are that the determinant of coefficients
should be zero:

1 0 γe 0

0 1 1 γp
= 1 − γe2 1 − γp2 = 0
γe 0 1 0
0 γp 0 1

It is thus known that there exist four possible solutions for the boundary velocity C:
1 − γe2 = 0, namely = ±Ce

C
1 − γp2 = 0, namely = ±Cp
C
This means that a 3rd order weak-discontinuous elastic–plastic boundary, if exists, should
either coincide with an elastic characteristic line, or coincide with a plastic characteristic
line. However, in order to simultaneously satisfy Eq. (4.122) and the basic assumption that
wave speed is discontinuous across the boundary [Eq. (4.101)], if the boundary coincides
p p
with an elastic characteristic line, then must have K3 = J3 = 0; or if coincides with a
plastic characteristic line, then must have K3e = J3e = 0. On the contrary, if the bound-
ary neither coincides with an elastic characteristic line (namely |γe |
= 1), nor a plastic
characteristic line (namely |γp |
= 1), then Eq. (4.122) only have trivial solutions as:
p p
K3e = J3e = K3 = J3 = 0
which means that all the 3rd partial derivatives of σ and v must simultaneously be zero.
In other words, there exists no 3rd order weak-discontinuous elastic–plastic boundary that
does not coincide with a characteristic line. By the analogy of this approach, it is deduced
that any higher order derivatives of σ and v cannot be discontinuous across an elastic–
plastic boundary propagating in any noncharacteristic direction. To conclude, we have
the following most important feature of elastic–plastic boundary (Yu et al., 1981, 1982,
1984).
Theorem 5: Across a weak-discontinuous elastic–plastic boundary characterized by the

basic singularity [Eq. (4.101)], either the 1st order or the 2nd order derivatives of σ and
v must be discontinuous, while any higher-than-second order discontinuity is impossible,
except that the boundary coincides with a characteristic line.
It is worthy to point out that the above important theorem comes from the basic singu-
larity, namely the slope of material stress–strain curve (dσ/dε) and thus the wave speed
being discontinuous across the boundary. In fact, suppose there exists a higher order
weak-discontinuous boundary, say, a 3rd order weak-discontinuity. Then the 2nd order
derivatives across the boundary must be continuous, namely Eq. (4.122) must be satisfied.
Rewriting Eq. (4.122) in the form of jumps across the boundary, as:

K3
+γ =0
J3

J3
+γ =0
K3
from which we obtain:

K3 J3
[γ ] = − =− (4.122 )
J3 K3
On the other hand, note that

⎡ ⎤
K3

J3 ⎢ 1 ⎥ J3
=⎢ ⎥
⎣ K3 ⎦ = K3 e p
K3 K3
J3 J3 J3
then if Eq. (4.112 ) is held then the following equation must be satisfied:

K3
[γ ] = − =0
J3
which is contradictory to the basic assumption [Eq. (4.101)]. In other words under the
basic assumption, the expressions of Eq. (4.122) cannot be held simultaneously. Thus
on the elastic–plastic boundary the 2nd order derivatives of σ and v are not continuous
everywhere, namely the 3rd order weak-discontinuous boundary does not exist. Since
the existence of an nth order weak-discontinuous boundary is based on the prerequi-
site that all the lower-than-n order partial derivatives of σ and v are continuous, any
higher-than-second weak-discontinuous boundary is impossible, unless it coincides with
a characteristic line.
4.10.7 Higher order isolated points on elastic–plastic boundary
Earlier we have proved that an elastic–plastic boundary can only be a first order or a second
order discontinuity with respect to σ and v. This property does not exclude the possibility
that at a certain isolated point of the boundary, say the point M, higher order derivatives
of σ and v are discontinuous. Namely, there may exist the isolated nth order (n ≥ 2)
discontinuous points. Because the existence of an nth order (n ≥ 2) discontinuous point
only requires that the σ and v and their derivatives up to (n − 1)th order are continuous
across the boundary at this point, rather than everywhere of the whole boundary. For
example, on a 1st order weak-discontinuous boundary there may exist 2nd or higher order
isolated discontinuous points, or on a 2nd order weak-discontinuous boundary there may
exist 3rd or higher order isolated discontinuous points, etc.
For a 1st order weak-discontinuous boundary, Eq. (4.109) must be satisfied at any point
of the boundary, namely:

J1
[K1 ] = −C
C

J1 K1
= −C
C C2
Surely they should be satisfied at an arbitrary point M on the boundary. If at and

p
only at this point M: K1e = K1 , or written as [K1 ]M = 0, then at and only at the
point M, we have:

J1 K1
[K1 ]M = = =0
C M C2 M
Therefore at the point M all the first order derivatives of σ and v are continuous. Because
of the basic singularity, the wave speed at the point M is discontinuous, [C]
= 0. This
p
means that K1e (M) = K1 (M) = 0 too. If the second order derivatives of σ and v are
discontinuous at the point M, then M is an isolated 2nd order weak-discontinuity point.
We now use the same approach to consider a 2nd order weak-discontinuous boundary.
We know that on the boundary all the first order derivatives of σ and v are continuous
everywhere, and (∂σ e /∂t) = (∂σ p /∂t) = (∂ve /∂X) = (∂vp /∂X) = 0. Also everywhere
on the boundary, Eq. (4.117) must be satisfied, which can be rewritten as:
J2 , (e, p)
K2 = −C
C

J2 K2
= −C
C C2
Certainly they should be also satisfied at an arbitrary point M on the boundary. If at and
p
only at the point M: K2e = K2 , namely [K2 ]M = 0 (Noting that this is not the case at
other points), then at and only at the point M, we have:
e p
p J2 J2 K2
K2e = K2 , = , [K2 ]M = =0
C C C2 M
Moreover, because of the basic singularity, the wave speed at the point M is discontinuous,
[C]
= 0. This means that:
p p
K2e (M) = K2 (M) = J2e (M) = J2 (M) = 0
Therefore all the second order derivatives of σ and v are continuous at the point M.
If the third order derivatives of σ and v are discontinuous at the point M, then M is an
isolated 3rd order weak-discontinuity point. It is easy to show: at an arbitrary point M
p
on a 1st order weak-discontinuous boundary, if both conditions K1e (M) = K1 (M) and
p p
K2e (M) = K2 (M) are satisfied, but K3e (M) and K3 (M) are not simultaneously zero, then
M is a 3rd order isolated point at which all the second order derivatives of σ and v are zero.
By the analogy of such argument, the nth isolated weak-discontinuous point can exist.
Moreover, because of the basic singularity, we have the following important theorem for
the nth isolated weak-discontinuous point:
Theorem 6: On an elastic–plastic boundary, the existence of an isolated nth (n ≥ 2)

weak-discontinuous point is possible. At this point and only at this point, all the partial
derivatives of σ and v up to the (n−1)th order continuous and all these partial derivatives
are zero except ∂σ/∂X and ∂v/∂t.
By differentiating the governing Eq. (4.107) many times, we know that for an nth order
weak-discontinuous point M, among the (n + 1) partial derivatives of nth order in either
the elastic zone or the plastic zone, only two of them are independent:
⎧ n
⎪
⎪
∂ σ
≡ Kn , (i − even number)
⎪
⎨ ∂t n
∂ nσ M
C i
= n
∂X i ∂t n−i ⎪
⎪ ∂ v
M ⎪
⎩ 0 C n
ρ ≡ Jn , (i − odd number)
∂t M
⎧ n
⎪
⎪
∂ v
≡
Jn
(i − even number)
⎪
⎨ ∂t n ,
∂ nv M ρ0 C
Ci =
∂X i ∂t n−i ⎪ 1 1 ∂ nσ
⎪ Kn
M ⎪
⎩ ≡ , (i − odd number)
ρ0 C ∂t Mn ρ0 C
(i = 0, 1, 2, . . . , n) (4.123)
With regard to the successive boundary following the nth order isolated point M, according
to theorem 5, the boundary only can be either a 1st order weak-discontinuous boundary,
or a 2nd order weak-discontinuous boundary.
First, we discuss the situation that the successive boundary is a 1st order weak-
discontinuous boundary. In such a case, since everywhere on the boundary must have:
[σ ] = 0, [v] = 0
the nth order boundary derivatives (d n /dt n ) of [σ ] and [v] obviously should also be equal
to zero everywhere, including at the point M, namely:

d n [σ ] d n [v]
= 0, =0
dt n M dt n M

Note that by the definition of boundary derivative, (d n /dt n ) = [(∂/∂t) + C(∂/∂X)] n.
Using the binomial expansion, and taking into consideration the theorem 6, the above
equations can be written as:
.
n
∂ nσ i = 0
Cni C M
∂X i ∂t n−i M
i=0
(4.124)
.
n
∂ nv
Cni i
C =0
M
∂X i ∂t n−i M
i=0
where Cni = i!(n−i)!

n!
is the binomial coefficient. Substitute Eq. (4.123) into Eq. (4.124),
at the point M we have:

Cn0 Kn + Cn1 γJn + Cn2 γ 2 Kn + +Cn31 γ 3Jn + · · · + the nth term = 0

1
C Jn + Cn γKn + Cn γ Jn + +Cn γ Kn + · · · + the nth term = 0
0 1 2 2 31 3
C n
Introducing the coefficient φn for the even terms and the coefficient ψn for the odd terms:
1" #
φn = Cn0 + Cn2 γ 2 + Cn4 γ 4 + · · · = (1 + γ )n + (1 − γ )n
2
(4.125)
1" #
ψn = Cn1 γ + Cn3 γ 3 + Cn5 γ 5 + · · · = (1 + γ )n − (1 − γ )n
2
we obtain the basic equations for the nth order isolated point M following which the
elastic–plastic boundary is a 1st order weak-discontinuity:
[φn Kn + ψn Jn ] = 0
(4.126)
1
(φn Jn + ψn Kn ) = 0
C
Note that between coefficients φ and ψ the following recursion relationships exist:
φn = φn−1 + γ ψn−1
(4.127)
ψn = ψn−1 + γ φn−1
Substituting Eq. (4.127) into (4.126) and after some operations, we get:

1 − γ 2 (φn−1 Kn + ψn−1 Jn ) = 0 (4.128)
or
2
C
p p p p 1− M
φn−1 Kn + ψn−1 Jn Ce2
= (4.128 )
e Ke
φn−1 n + ψn−1
e Je
n
2
C
1− M
Cp2
It is worthy to point out that Eqs. (4.126) and (4.128 ) actually can be deduced from the
basic equations for a 1st order weak-discontinuous boundary [Eq. (4.109)]:
[K1 + γJ1 ] = 0

1
(J1 + γK1 ) = 0
C
and the Eq. (4.110), [(1 − γ 2 )K1 ] = 0, rewritten in the form:
2
C
p 1−
K1 1 − (γ e )2 Ce2
= =
K1e 1 − (γ p )2 2
C
1−
Cp2
by directly applying to the point M. The deduction procedures are shown below.
First we expand K1 into a Taylor series of the partial derivatives of σ up to the nth order at
the point M along the elastic–plastic boundary. Because the point M is an isolated nth order
weak discontinuity, all partial derivatives lower than the nth order are zero (Theorem 6),
therefore we finally get:

∂σ 1 d n−1 ∂σ
K1 = = (t − tM )n−1
∂t (n − 1)! dt n ∂t M

(t − tM )n−1 . i
n−1
∂ nσ i
= Cn−1 C
(n − 1)! ∂X i ∂t n−i M M
i=0
(t − tM )n−1 0
= Cn−1 Kn + Cn−1
1
γJn + Cn−1
2
γ nKn + · · ·
(n − 1)! M
(t − tM )n−1
= (φn−1 Kn + ψn−1 Jn )
(n − 1)!
Similarly, for J1 we have:

∂v (t − tM )n−1
J1 = ρC = (φn−1 Jn + ψn−1 Kn )
∂t M (n − 1)!
Substituting these expressions into Eqs. (4.109) and (4.110), taking into consideration
the theorem 6 and the recursion relationship 4.127, we can deduce the Eqs. (4.126) and
(4.128 ). Note that the deduction of Eq. (4.128 ) is actually the same as to calculate the
p
indeterminate form (K1 /K1e ) = (0/0) by the L’hopital’s rule.
Among the three equations of Eqs. (4.126) and (4.128 ), only two, and any two, are inde-
p p M . Among
pendent. These two equations contain five quantities Kne , Kn , Jne , Jn , and C
the four partial derivatives, if we have known three, then from these two equations we can
calculate the fourth derivative, and C M . Or eliminating the fourth derivative from these
two equations, the equation for solving C M can be obtained, which is a (2n)th algebric
p p
equation. For example, if the quantities Kne , Kn , and Jn are known, eliminating the
e
quantity Jn , then the equation for CM is:

Ce + Cp Ce − Cp n p p
(1 − γe ) −
n
(1 − γe )n
1 + γp Kn + Jn
2 2

Ce + Cp Ce − Cp n p p
+ (1 + γe )n − (1 − γe )n 1 − γp Kn − Jn (4.129)
2 2
n
− 2Cp 1 − γe2 Kne = 0
p p
Alternatively, if the quantities Kne , Kn , and Jne are known, we eliminate the quantity Jn ,
and then the equation for CM is:

Ce + Cp n C e − C p n p p
1 − γp − 1 + γp (1 + γe )n Kn + Jn
2 2

Ce + Cp n C e − C p n
(4.130)
+ 1 + γp + 1 − γp (1 − γe )n Kne − Jne
2 2
n
p
− 2Ce 1 − γp2 Kn = 0
p
When the boundary conditions at the end of bar (Kne and Kn ) are known, Eq. (4.129)
can be applied to determine the initial propagating velocity of an unloading boundary,
p
across which a known plastic zone (i.e. Jn is known) is unloaded to an elastic zone.
Similarly, Eq. (4.130) can be applied to determine the initial propagating velocity of a
loading boundary, across which a known elastic zone (i.e. Jne is known) is loaded to a
plastic zone.
If the point M is a second order weak-discontinuity (n = 2) and the successive boundary

is a 1st order weak-discontinuous, then Eq. (4.126) is reduced to:

1 + γ 2 K2 + 2γJ2 = 0
(4.131)
1
1 + γ 2 J2 + 2γK2 = 0
C
while Eq. (4.128) is reduced to:

1 − γ 2 (K2 + γJ2 ) = 0 (4.132)
Of the above three equations, two of them are independent. In discussions on the problem
where the end of the bar is a 2nd order weak-discontinuity and the successive elastic–
plastic boundary is a 1st order weak-discontinuous, Ting first deduced the first and the
third equations although in a different form (Ting, 1971).
Therefore, to a 1st order weak-discontinuous boundary, no matter which order of disconti-

nuity the isolated point M is, there always exists two independent equations which connect
the boundary propagating velocity C M to the four quantities, the ∂ n σ/∂t n and ∂ n v/∂t n at
both the elastic side and the plastic side across the boundary. In general situation, once
three of the four partial derivatives are known, then C M can be calculated. In the case of
p
simple wave propagation, the problem is reduced to that, once Kne and Kn are known then
M can be determined, just as stated in Section 4.7.
C
Similarly, we discuss the situation that the isolated point M is an nth order discontinuity and
the successive boundary is a 2nd order weak-discontinuous. According to the theorem 2,
everywhere on the boundary we have:
∂σ ∂v
= =0
∂t ∂X

∂σ ∂v
=
∂X ∂t
Obviously the (n − 1)th order boundary derivatives (d n−1 /dt n−1 ) of them should also be
equal to zero everywhere, including at the point M, namely:

d n−1∂σ d n−1 ∂v
= =0
dt n−1 ∂t M dt n−1 ∂X M
n−1 n−1
d ∂σ d ∂v
n−1 ∂X
= n−1
=0 (4.133)
dt M dt ∂t M
After expanding the above equations to binomial expansions, taking into consideration
the theorem 6 about the isolated point M, and then substituting Eqs. (4.123) and (4.124)
in the expansions, we get the following basic equations for the isolated nth order weak-
discontinuous point M on a 2nd order weak-discontinuous successive boundary:
φn−1 Kn + ψn−1 Jn = 0, (e, p)

(4.134)
1
(φn−1 Jn + ψn−1 Kn ) = 0
C
Substitute the recursive relationship for φ and ψ [Eq. (4.127)] into Eq. (4.134), and after
some operations, we have:

1 − γ 2 (φn−2 Kn + ψn−2 Jn ) = 0 (4.135)
which may be rewritten as:
2
C
p p p p 1− M
φn−2 Kn + ψn−2 Jn Ce2
= (4.135 )
e Ke
φn−2 n + ψn−2
e Je
n
2
C
1 − M2
Cp
Obviously, among the four equations expressed in Eqs. (4.134) and (4.135), only three of
them are independent.
4.10.8 Supplementary conditions on loading boundary
The above discussions are applicable to both a loading boundary and an unloading
boundary. However, in the case of loading boundary, the yield condition should be added.
Note that when the material is loaded from an elastic state (or unloading state) to a plastic
state, since the slope of stress–strain curve dσ/dε is discontinuous [Eq. (4.101)], the wave
speed under the yield stress is discontinuous too, namely is two-valued. Thus the yield cri-
terion such as Mises criterion or Tresca criterion usually used in the static plasticity is only
the necessary condition but not the sufficient condition for yielding. An example is shown
in Fig. 4.34 wherein a set of elastic-plastic simple waves is propagating in a semi-infinite
bar. When the boundary stress is just equal to the initial yield stress Y0 , as it has been
discussed in Chapter 2 about the two-wave structure, a precursor elastic wave with stress
level σ = Y0 propagates by a faster velocity of Ce , and simultaneously a plastic wave with
the same stress level σ = Y0 propagates by a slower velocity Cp (Y0 ). Between these two
wave fronts, is formed a constant-value zone with σ = Y0 . Note that this constant-value
zone is essentially an elastic zone, since the plastic disturbance, which is propagated by
the successive plastic wave, has not arrived yet. Therefore, much different from the static
t t
ry
nda
Bou
stic
Pla
-
P
astic
El Elastic Constant-Value Zone
σ Y0 0 X
Fig. 4.34. Elastic-plastic boundary in case of a set of elastic-plastic simple waves propagating in a
semi-infinitive bar.
plasticity, in the dynamic plasticity related to elastic–plastic wave propagation we could

not say that any zone that satisfies the usual yield condition of σ = Y0 is a plastic zone.
Strictly speaking, when the slope of stress–strain curve dσ/dε is discontinuous at the yield
point (4.101), the yield condition on a loading elastic–plastic boundary t = g(X) should
be expressed as:
⎧
⎪ ∂|σ p |
⎪
⎨ > 0 (For a first order boundary)
∂t
σ (X, g(X)) = σm (X), and (4.136)
⎪ ∂ 2 |σ p |
⎪
⎩ > 0 (For a second order boundary)
∂t 2
where σm (X) is the initial yield limit, σm (X) = Y0 during the initially yielding, otherwise
σm (X) is the maximum stress ever experienced at the point X.
Another feature can be seen from Fig. 4.34. Because the stress wave velocity has two
values at yield stress, the boundary derivative of stress ∂σ/∂t just before the yielding
(noted by (∂σ b /∂t)) and the boundary derivative of stress ∂σ/∂t just after the yielding
(noted by (∂σ a /∂t)) propagate separately along different characteristic lines by different
propagating velocities. These values do not represent respectively the values of ∂σ e /∂t
and ∂σ p /∂t across an elastic–plastic loading boundary at this point. Namely, we cannot
substitute the mentioned ∂σ b /∂t and ∂σ a /∂t values directly into Eq. (4.111) to calculate
the propagating velocity of an elastic–plastic loading boundary, C L . In fact, in a general

case, according to the inequality 4.113, the varied range of CL has two possibilities:
(1) Cp ≥ |C L | ≥ 0 and (2) |C
L | ≥ Ce .
L | ≤ Cp ), as shown in Fig. 4.35(a), the ∂σ b /∂t actually propagates

In the first case (|C
along the elastic characteristic lines, while the ∂σ e /∂t on the elastic side across the loading
boundary corresponds to the ∂σ m /∂t in the zone m, which needs to be determined.
In the second case (|CL | ≥ Ce ), as shown in Fig. 4.35(b), the ∂σ a /∂t actually propagates
along the plastic characteristic lines, while the ∂σ p /∂t on the plastic side across the loading
boundary corresponds to the ∂σ m /∂t in the zone m, which needs to be determined.
t Cp t t Cp
Cp
P CL P
E Ce Ce Cu Ce
a a P a E
σa
m m
t
E CL m CP
E L
σb
b b E b
t
σ Y0 X 0 X 0 X
(a) (b) (c)
Fig. 4.35. Multiple-valued propagation speeds of a first-order weak-discontinuous elastic-plastic

boundary.
Therefore, physically or mathematically speaking, to determine C L only Eq. (4.109) is

not sufficient and other conditions (namely the yield conditions) must be supplemented.
Thus, for a first order weak-discontinuous loading boundary, the supplemental yield con-
ditions expressed by Eq. (4.136), or equivalently the corresponding differential form
expressed by Eq. (4.137), should be satisfied too:
dσm (X) ∂σ 1 ∂σ 1
= + = (φ1 K1 + ψ1 J1 ) (4.137)
dX ∂X
CL ∂t
CL
or
∂σ
L =
C ∂t (4.137 )
dσm ∂σ
−
dX ∂X
Moreover we have the following compatibility equation along the elastic or plastic
characteristic line, across which are the zone m and the zone i:
m
∂σ m ∂σ i ∂σ ∂σ i
− = −Ci − (4.138)
∂t ∂t ∂X ∂X
where, if it is the case shown in Fig. 4.35(a), the superscript i should be substituted by
superscript a, accordingly Ci takes the value Ce ; while if it is the case shown in Fig. 4.35(b),
the superscript ishould be substituted by superscript b, and Ci takes the value Cp . Thus,
Eqs. (4.137) and (4.138), in addition to the two equations in Eq. (4.109), totally four
equations are enough to solve the four unknown quantities: C L , ∂σ m /∂t, ∂σ m /∂X, and
a
∂σ /∂X. The problem is solvable.
Now consider the case for a second order weak-discontinuous loading boundary (Fig. 4.36).
Similar to the previous case, the yield condition 4.136 in the second order differential
form is:
% &
d 2 σm 1 1 1 ∂ 2σ 2 ∂ 2σ
= 2 (φ2 K2 + ψ2 J2 ) = + 2 + (4.139)
dX 2
C C 2
C ∂t 2 L ∂X∂t
C
L L
Or, considering the Eq. (4.119 ) the above equation can be rewritten as:
% &
d 2 σm 1 1 ∂ 2σ
= − 2 (4.139 )
dX 2 C2
C ∂t 2
L
Moreover, similar to Eq. (4.138), we have the following compatibility equation along the
elastic characteristic line, across which are the zone b and the zone r in Fig. 4.36:
r
∂σ r ∂σ b ∂σ ∂σ b
− = −Ce − (4.140)
∂t ∂t ∂X ∂X
t t
Cp
Cp
Cu
P2 E3 P2 CL
P1 CL
a s r E2 Ce a q s P2 E2 Ce
σa
r
t
E1 E1
b
σb
b
t
Y0 0 X σ 0 X
(a) (b)
Fig. 4.36. Multiple-valued propagation speeds of a second-order weak-discontinuous elastic-plastic

boundary.
and, the following compatibility equation along the plastic characteristic line, across which
are the zone s and the zone a in Fig. 4.36:
a
∂σ a ∂σ s ∂σ ∂σ s
− = −Cp − (4.141)
∂t ∂t ∂X ∂X
Thus, Eqs. (4.139)–(4.141), in addition to the three basic equations in Eq. (4.117), totally
six equations are enough to solve the six unknown quantities: C L , ∂σ r /∂t, ∂σ r /∂X,
s s a
∂σ /∂t, ∂σ /∂X, and ∂σ /∂X. The problem is solvable.
If the point M being considered is a nth (n ≥ 2) order weak-discontinuous point, then by

differentiating Eq. (4.136) n times, we deduce that at the point M the following equation
should be satisfied:
d n σm 1
= n (φn Kn + ψn Jn ) (4.142)
dX n
CL
regardless of whether the successive boundary is a 1st order discontinuity or a 2nd order
discontinuity. Note that during the deduction of Eq. (4.142) we have already assumed:

d i σm
= 0, i = 2, 3, . . . , (n − 1)
dX i M
M = 0.
otherwise we will get: C
Based on the above discussions, it is worthwhile to point out another important theorem
associated with the so-called branching point of boundary. For the 1st order weak-
discontinuous loading boundary in the case of C L ≥ Ce , since the dσm /dX and the
∂σ b /∂t, ∂σ b /∂X in the elastic zone are all known, the propagating velocity of this loading
194
Table 4.3. Basic relationships for determining propagation speeds of elastic-plastic boundaries in different cases.
The point 1st order 2nd order weak-discontinuous point nth order weak-discontinuous point
weak-discontinuous
Successive point
boundary
!
[K1 + γJ1 ] = 0 (1 + γ 2 )K2 + 2γJ2 = 0 [φn Kn + ψn Jn ] = 0

1 1 1
(J1 + γK1 ) = 0 (1 + γ 2 )J2 + 2γ K2 = 0 (φn Jn + ψn Kn ) = 0

1st order C C C
weak-discontinuous ! ! !
boundary (1 − γ 2 )K1 = 0 (1 − γ 2 )(K2 + γJ2 ) = 0 (1 − γ 2 )(φn−1 Kn + ψn−1 Jn ) = 0
Karman et al. (1942) Ting (1971)
K2 + γJ2 = 0 φn−1 Kn + ψn−1 Jn = 0

1 1
2nd order (J2 + γK2 ) = 0 (φn−1 Jn + ψn−1 Kn ) = 0
C C
weak-discontinuous
boundary ! !
(1 − γ 2 )K2 = 0 (1 − γ 2 )(φn−2 Kn + ψn−2 Jn ) = 0
Clifton and Ting (1968)
Supplementary dσm 1 d n σm 1
conditions on = (K1 + γJ1 ) d 2 σm 1 " # = n (φn Kn + ψn Jn )
dX = (1 + γ 2 )K2 + 2γ J2 dX n
loading boundary C dX 2 2
C C
boundary C L can be determined immediately from Eq. (4.137). It means that the determi-
nation of C L is independent of ∂σ a /∂t, even though (∂σ a /∂t) < 0, namely, the zone a
is in an unloading situation. If so, then another unloading boundary can be formed in the
plastic zone behind this loading boundary, as shown in Fig. 4.35(c). The same situation
can appear in the case of a 2nd order weak-discontinuous loading boundary, as shown
in Fig. 4.36(b). Thus there may be more than one elastic–plastic boundaries propagating
through the same point M considered. In other words, the boundary propagating velocity C
has multi-values at the point M considered. Such a point is called the branching point
of boundary, which is a singular point with respect to the elastic–plastic boundary itself
(Yu et al., 1984). Such phenomena, as can be understood from the previous discussion, are
closely related to or originated from the basic singularity of an elastic–plastic boundary
[C]
= 0.
Whether or not a branching point appears on an elastic–plastic boundary, the approaches

stated in this section are sufficient to determine the propagating velocity of an elastic–
plastic boundary under all the possible situations mentioned (Yu et al., 1984). The
corresponding basic relationships are summarized in Table 4.3. The readers, who are
interested in a further study on the discontinuities across elastic/plastic boundary in
elastic–plastic wave propagation, are suggested to read the review paper written by Ting
(Ting, 1990).
CHAPTER 5
Rigid Unloading Approximation
All the elastic–plastic wave propagation containing unloading processes discussed in

Chapter 4 are based on the assumption of elastic unloading. In such cases, unloading
disturbances propagate with elastic wave velocity C0 . If the elastic modulus E is much
larger than the slope of plastic stress–strain curve dσ/dε; or in other words, if the elas-
tic wave velocity is much larger than the plastic wave velocity C (i.e. C0 C), then
relative to the plastic wave velocity C, the C0 can be approximately regarded as infi-
nite (i.e. C0 /C ≈ ∞). When unloading problems are dealt with, this is corresponding
to that of the elastic unloading stress–strain curve which is approximately regarded as a
straight line parallel to the σ axis (Fig. 5.1), or corresponding to that of the elastic–plastic
body once unloading is regarded as a rigid body but the inertia effect of rigid part should
be still taken into account. By such a “rigid unloading” assumption, the elastic–plastic
wave propagation problems containing unloading processes usually can be simplified
a lot.
5.1 Rigid Unloading in a Semi-Infinite Bar
5.1.1 Rigid unloading in linear hardening plastic bars
At the outset, let us discuss a very simple example. Consider a semi-infinite bar with a
stress–strain relationship of the so-called linear hardening plastic loading–rigid unload-
ing, as shown in Fig. 5.1(a), which can be regarded as a simplified model of the linear
hardening plastic loading–elastic unloading stress–strain relationship of soil shown in
Fig. 4.13. Aconstant-velocity impact loading v* is suddenly applied on the bar end (X = 0),
and at time t1 the bar end is suddenly unloaded to make it stress-free (σ = 0), as shown
in Fig. 5.2.
Obviously, before unloading (t < t1 ) the situation is the same as what we have discussed
earlier on plastic loading waves, namely
√ a strong discontinuous wave is propagating with
a constant wave velocity C1 = E1 /ρ0 . In case of zero initial condition, the states in
197
s s s
C Y E1 Y
A
B
E1 E E
O e O e O e
(a) (b) (c)
Fig. 5.1. Some stress–strain relationships under rigid–unloading assumption: (a) linear loading–
rigid unloading; (b) elastic–linear hardening plastic loading–rigid unloading; (c) elastic-decreasing
hardening plastic loading–rigid unloading.
t t
B
t1 A
−s
−r0C1v* O l1 X
Fig. 5.2. In a linear loading–rigid unloading material [Fig. 5.1(a)], a strong discontinuous wave
propagates with a constant wave velocity when a boundary loading is suddenly applied, and unload-
ing disturbances propagate with an infinite wave velocity when a boundary unloading is suddenly
applied.
region 1 are:
σ1
v = v∗ , σ1 = −ρ0 C1 v1 = −ρ0 C1 v∗ , ε1 = (5.1)
E1
At t = t1 , the plastic loading wave front arrives at location l1 = C1 t1 , while an unloading

suddenly takes place at the bar end. According to the rigid unloading assumption, the
unloading disturbance propagates with an infinite wave velocity, so that all particles of the
segment behind the plastic loading wave front (X ≤ l1 ) are instantaneously unloaded and
move as a rigid body, while the amplitude of the plastic loading wave begins to attenuate
due to the interaction with the unloading disturbances. Similar to the elastic unloading
Rigid Unloading Approximation 199
process discussed earlier, such an attenuation process results from the continuous inter-
action between the plastic loading disturbances, which propagate with a velocity of C1 , and
the unloading disturbances, which propagate with infinite velocity. Note that as the prop-
agation of the continuously attenuating plastic wave goes on, the length of the unloaded
rigid segment l continuously increases (l = C1 t). Obviously, at t = t1 the unloading
boundary which propagates with infinite velocity, is represented by the line t1 A in Fig. 5.2
= ∞), while at t > t1 the unloading boundary coincides with the propagating locus
(C
AB of plastic loading wave and so the boundary propagating velocity is equal to the plastic
wave velocity (C = C1 ).
For the strong discontinuous plastic wave, according to the dynamic compatibility con-
dition across a strong discontinuity, we have the following equation to relate its stress
amplitude σm with its particle velocity amplitude vm :
σm = −ρ0 C1 vm (5.2)
the subscript m used here is to denote that these quantities are the maximum that have
been attained in the history of plastic deformation for each particle. In general they are
also the quantities along the unloading boundary.
On the other hand, for the unloaded rigid segment, according to the Newton’s second law,
we have the following motion equation:
dvm
σm = ρ0 l (5.3)
dt
From Eqs. (5.1) and (5.2) and noticing l = C1 t, it is clear that the particle velocity vm (t)
for the rigid segment should satisfy the following ordinal differential equation:
dvm vm
+ =0 (5.4)
dt t
After determining the integration constant by the condition vm (t1 ) = v1 , the solution of
this ordinary differential equation (ODE) can be obtained as:
vm t = v1 t1 (5.5a)
or noticing l/t = l1 /t1 = C1 , it can be rewritten as
vm l = v1 l1 (5.5b)
By substituting Eqs. (5.1) and (5.2) into Eq. (5.5b), we obtain the following hyperbolic
functions which show how σm and εm attenuate with the unloading boundary propagating
location l, or with time t:
σm l = σ1 l1 , σm t = σ1 t1 (5.6)
εm l = ε1 l1 , εm t = ε1 t1 (5.7)
Recall that in the case of elastic unloading, a similar hyperbolic relationship between
σm and the locations of stationary strain-discontinuous interfaces l has been obtained
(see Section 4.4.1 and Fig. 4.7). By comparing these two cases, it is easy to find that
the discrete stationary strain-discontinuous interfaces in the case of elastic unloading now
become continuously distributed in the present case of rigid unloading.
5.1.2 Rigid unloading in linear elastic–linear hardening plastic bars
Consider now a semi-infinite bar with a stress–strain relationship of the so-called linear
elastic–linear hardening plastic loading–rigid unloading, as shown in Fig. 5.1(b).
In such a case, the elastic precursor wave ahead of the plastic loading wave should be
taken into consideration. If the boundary condition at bar end is not sudden unloading but
gradual unloading as shown in Fig. 5.3, then the loads applied on the rigid segment are not
only the σm acting from the unloading boundary (X = l), as given on the left-hand side
of Eq. (5.3), but also the σ0 (t) acting at the bar end (X = 0). Nevertheless, the problem
can be solved by similar procedures mentioned above.
Actually, in such a case, Eqs. (5.2) and (5.3) become respectively to:
1
vm = vY − (σm − Y ) (5.8)
ρ0 C1
dvm
σm − σ0 = ρ0 l (5.9)
dt
Substituting Eq. (5.8) into Eq. (5.9), we have
1 dσm
σm − σ0 (t) = − (5.10)
C1 dt
t t
K
tK
1t
X=C
t1
C 0t
X=
0 t)
s(
s O XK X
Fig. 5.3. A semi-infinite bar with σ –ε curve shown in Fig. 5.1(b) is subjected to a sudden boundary
loading followed by gradual unloading.
Noticing l = C1 t, the solution of this equation can be obtained as:

1 t
σm = σ0 (t)dt (5.11)
t 0
Since σ0 (t) is a known monotonously decreasing function presenting gradual unloading,
Eq. (5.11) shows how σm attenuates with time. If at t ≥ t1 , σ0 (t) = 0, then by Eq. (5.11)
σm at t = t1 is

1 t1
σ1 = σ0 (t)dt (5.12)
t1 0
Therefore, from time t ≥ t1 , the attenuation of σm follows the same rule of Eq. (5.6).
The time tK in which the plastic wave begins to disappear due to the interaction with the
unloading disturbances, and the corresponding location XK = C1 tK can be determined
according to the condition:
σm (tK ) = Y
After that time, the unloading of the bar segment which has undergone elastic deformation
can be dealt with as we have discussed earlier on elastic unloading.
Once the attenuation of the maximum stress σm (t) is obtained, from the dynamic compat-
ibility condition across a strong discontinuity [Eq. (5.8)] and the linear hardening plastic
stress–strain relationship εm = εY + ((σm − Y )/E1 ), the corresponding attenuation of
the maximum particle velocity vm (t) and the maximum strain εm (t) can be obtained
respectively as:

1 1 1 t
vm (t) = 1− Y− σ0 (t)dt (5.13a)
ρ0 C 1 µ1 t 0
/% & 0
1 1 1 t
εm (t) = −1 Y − σ0 (t)dt (5.13b)
E1 µ21 t 0
where

C0 E
µ1 = =
C1 E1
Especially, if σ0 (t) linearly decreases with time, as expressed by

t
σ0 (t) = −pm 1 −
T
where T is the time when σ0 (t) = 0, then from Eq. (5.11) it can be shown that at t ≤ T ,
σm is a linear function of t or X:

t µ1 X
σm (t) = −pm 1 − = −pm 1 − (5.14)
2T 2 C0 T
Recall that in Chapter 4, the following equation has been obtained for the case of elastic
unloading [Eq. (4.50)]
⎛ ⎞
1
µ1 −
µ2 − 1 X ⎜ µ1 X ⎟
σm (t) = −pm 1 − · = −pm ⎜
⎝ 1 − ⎟
2µ1 C0 T 2 C0 T ⎠
Comparing this equation with Eq. (5.14), it can be seen that the difference between
these two equations decreases with increasing µ1 . If we introduce BR = µ1 /2 and
BE = (µ21 − 1)/(2µ1 ) to denote the linear coefficient of the term (X/C0 T ) in these
two equations [see Eq. (4.53)], respectively, then the relative difference of these two
coefficients is:
BR − BE 1
δB = = 2
BR µ1
For metallic materials, the elastic wave velocity is approximately 10 times the plastic
wave velocity, namely µ1 ≈ 10, so that δB ≈ 1%. It means that the error induced by using
rigid unloading simplification where the elastic wave velocity is approximately regarded
as infinite, is small enough, if the elastic wave velocity C0 is much larger than the plastic
wave velocity C1 . Then the complicated elastic unloading problem, which is attributed to
solving simultaneously two different groups of partial differential equations respectively
in the plastic region and the elastic region, can be simplified to solve the problem which
only contains ordinal differential equation.
The example discussed above is a problem where the rigid unloading boundary coincides
with the propagating locus of strong discontinuous plastic loading wave propagating with
constant velocity. If the plastic loading wave is a continuous wave, then the unloading
boundary is a weak discontinuous boundary and no longer coincides with the propagating
loci of plastic waves, so that the problem will be more complicated. Nevertheless, the
problem can still be solved by a similar treatment as shown below.
5.1.3 Rigid unloading in linear elastic–decreasing hardening plastic bars
Consider now a semi-infinite bar with a stress–strain relationship of the so-called

linear elastic–decreasing hardening plastic loading–rigid unloading, as shown in
Fig. 5.1(c). Assume a gradual loading denoted by σ0l (t) and then a gradual unloading
denoted by σ0u (t), without any strong discontinuity, are applied at the bar end, which are
expressed respectively as (Fig. 5.4).
⎧ d|σ0u |
⎪
⎨σ (0, t) = σ0l (t), ≥ 0, 0 ≤ t ≤ t0
dt
⎪
⎩σ (0, t) = σ (t), d|σ0u |
0u ≤ 0, t ≥ t0
dt
t t
X = j(
t)
Unloading zone
M(j(t), t)
ou (t)
s
t0 Plastic zone
τ
s ol
(t)
s O X
Fig. 5.4. A semi–infinite bar with σ –ε curve shown in Fig. 5.1(c) is subjected to a gradual boundary
loading up to a maximum σmax , followed by a gradual unloading.
In such a case, the unloading boundary is a weak discontinuous one X = ϕ(t), as shown
in Fig. 5.4. Correspondingly, the previous compatibility condition across a strong discon-
tinuous interface, Eq. (5.8), now should be replaced by the following differential relation
for the rightward plastic simple waves:
dσm
dvm = − (5.15)
ρ0 C
Noticing that the length of rigid segment l in the motion equation of rigid segment
[Eq. (5.9)] now should be replaced by ϕ(t), then Eq. (5.10) should be rewritten as:
ϕ(t) dσm
+ σm − σ0u = 0, t ≥ t0 (5.16)
C dt
For plastic simple waves, there is one more relation:
ϕ(t) = C(σm ) · [t − τl (σm )] (5.17)
where τl (σ ) is the inverse function of σ0l (t); or we have

ϕ(t)
σm (t) = σ0l t − (5.18)
C(σm )
Substituting Eq. (5.7) into Eq. (5.16), eliminating ϕ(t), σm (t):
dσm σm − σ0u
+ =0 (5.19)
dt t − τl (σm )
Once σm (t) is solved, substitute it into Eq. (5.17), the unloading boundary X = ϕ(t) can
be determined. Alternatively, substituting Eq. (5.18) into Eq. (5.16), eliminating σm (t),
then the problem is summed up to solve a first-order ordinal differential equation with
respect to ϕ(t). Once ϕ(t) is solved, substitute it into Eq. (5.18), σm (t) can be determined.
For example, if σ0l (t) is a linear increasing function of time (Nowaski, 1978):
τ
σ0l (τ ) = σmax ,
t0
then Eqs. (5.18) and (5.19) are respectively:

σmax ϕ(t)
σm (t) = t− , (5.20)
t0 C
dσm σm − σ0u
+ σm − σ0u = 0 (5.21)
dt t−
σmax
Furthermore, assuming C = C1 (linear hardening plastic material), Eq. (5.20) reduces to

σmax ϕ(t)
σm (t) = · t−
t0 C1
(5.22)
dσm σmax ϕ
= · 1−
dt t0 C1
Substituting the above equation into Eq. (5.21), we have

t0 σ0u
ϕϕ = C12 · t − (5.23)
σmax
Thus the unloading boundary X = ϕ(t) can be solved
t 1/2
2t0
ϕ(t) = C1 · t 2 − t02 − σ0u (t)dt (5.24)
σmax t0
where the condition ϕ(t0 ) = 0 has been used to determine the integrate constant.
If σ0u (t) is also a linear increasing function of time, as shown in Fig. 5.5:
t1 − t
σ0u (t) = σmax · (5.25)
t1 − t 0
t t t)
j(
=
X
t1
u t)
so (
t0
s ol
(t)
s smax O X
Fig. 5.5. A semi–infinite bar with σ –ε shown in Fig. 5.1(a) is subjected to a linear increasing
boundary loading up to a maximum σmax followed by a linear decreasing boundary unloading.
then from Eq. (5.24), we have
⎧ 9
⎪
⎨C1 · t1
(t − t0 ), t0 ≤ t ≤ t1
ϕ(t) = t1 − t 0 (5.26)
⎪
⎩ ;
C1 · t 2 − t 0 t1 , t ≥ t1
Thus it can be seen, at t0 ≤ t ≤ t1 , the unloading boundary is a straight line propagating

with a constant velocity C = ϕ (t) = C1 [t1 / (t1 − t0 )]1/2 > C1 , while at t ≥ t1 , the
unloading boundary is a hyperbolic curve propagating with a velocity C which gradually
decreases and approaches C1 .
Substituting Eq. (5.26) into Eq. (5.22), the stress along the unloading boundary σm (t) can
be obtained, and furthermore from Eq. (5.15) the corresponding particle velocity vm (t)
can be obtained, respectively as:
for t0 ≤ t ≤ t1
⎧ 9
⎪
⎪ σmax t1
⎪
⎨σ m (t) = t − (t − t 0 ) ,
t0 t1 − t0
9 (5.27)
⎪
⎪ Y σmax t1
⎪
⎩vm (t) = − (1 − µ1 ) − t− (t − t0 ) ,
ρ0 C0 ρ0 C1 t0 t1 − t 0
for t ≥ t1
⎧ σmax ;
⎪
⎨σm (t) = t
⎪ t − t 2 − t0 t1
0
σmax
; (5.28)
⎪
⎪ Y
⎩vm (t) = − (1 − µ1 ) − t − t 2 − t 0 t1
ρ0 C0 ρ0 C1 t0
where µ1 = C0 /C1 . It should be noted that the particle velocities of the unloading segment
(as a rigid segment) of bar are uniform along the bar at any time, namely in the unloading
region we have
v̄(X, t) = vm (t),
while the stresses are non-uniform along the bar in the unloading region due to the inertia
effect of rigid unloading segment, namely in the unloading region, similar to Eq. (5.9),
we have
d v̄ dvm

σ (X, t) = ρ0 X + σ0u = ρ0 X + σ0u .
dt dt
Substituting Eqs. (5.25), (5.27), and (5.28) into the above equation, we obtain
⎧ 9
⎪
⎪ σmax t1 t1 − t
⎪
⎪
σ (X, t) = − 1 − X + σmax · , t0 ≤ t ≤ t1
⎪
⎨ C t
1 0 t 1 − t 0 t 1 − t0
% & (5.29)
⎪
⎪
⎪
⎪ σmax t
⎩
⎪ σ (X, t) = − 1− ; X, t ≥ t1
C1 t0 t 2 − t1 t0
Moreover, since the strains in the unloading region are entirely determined by the maximum
stresses σm having reached before unloading

σm − Y Y E0 σm
εm = εY + = 1− +
E1 E0 E1 E1
thus the strain distribution in the unloading region can be determined too as
⎧ ⎡ ⎛ ⎞⎤
⎪
⎪
⎪
⎪ ⎢ ⎜ ⎟⎥
⎪
⎪
⎪ ε (X) =
Y
(1 − µ 2) + σm
· ⎢t0 + X ⎜ 9 1 ⎟⎥, t0 ≤ t ≤ t1
⎪
⎪
m
ρ0 C02 1
ρ0 C12 t0 ⎣ C1 ⎝ t1 ⎠⎦
⎨
t1 − t0
⎪
⎪
⎪
⎪ / 0
⎪
⎪
⎪
⎪ Y σm X2 X
⎪εm (X) =
⎩ (1 − µ1 ) +
2 + t 0 t1 − , t ≥ t1
ρ0 C02 ρ0 C12 t0 C12 C1
(5.30)
With regard to the general propagating properties of the rigid unloading boundary in
semi-infinite bars, all can be directly deduced from the corresponding results for the elastic
unloading boundary discussed earlier in Chapter 4 by letting C0 → ∞. For example, for
the one-order weak discontinuous rigid unloading boundary, let C0 → ∞ in Eq. (4.73),
we obtain the following relation
∂σ/∂t 1
= (5.31a)
∂
σ /∂t 2
C
1− 2
C
or

= C 1 − (∂σ /∂t)
C (5.31b)
(∂σ/∂t)
For the unloading boundary, there must be ∂σ/∂t ≥ 0 and ∂

σ /∂t ≤ 0, therefore
≥ C.
C
Obviously, if ∂σ/∂t = ∞, ∂ σ /∂t

= ∞, or if ∂ σ /∂t = 0, ∂σ/∂t
= 0, then we have
= C; while if ∂σ/∂t = 0, ∂
C σ /∂t
= 0, or if ∂
σ /∂t = −∞, ∂σ/∂t
= ∞, then we have
= ∞.
C
5.2 Rigid Unloading in Finite Bars
In this section, we discuss an example for stress wave propagation in a finite bar char-
acterized by rigid unloading (Nowaski, 1978). The bar consists of two segments: one
(0 ≤ X ≤ l) is made of A material with linear hardening plastic loading–rigid unloading
behavior, as shown in Fig. 5.1(a), denoting its density and plastic wave velocity as ρA
and CA , respectively; while another (X ≥ l) is made of B material with linear elastic
behavior, denoting its density and elastic wave velocity as ρB and CB , respectively. In
addition, a rigid mass exists between these two segments, as shown in Fig. 5.6. Originally,
the bar is at rest and stress-free. At time t = 0, the bar end (X = 0) is suddenly loaded
and then gradually unloaded.
The whole problem can be separated into several regions as shown in Fig. 5.6, and the
solution in each region is discussed as follows.
t t t
t1 VI T
III
V
ou (t)
tK K
IV
t0 II
s
I O1 X
s O l X
A M B
rACA rBCB
Fig. 5.6. Stress wave propagations in a finite bar characterized by rigid unloading.
Region I: Before the rightward plastic wave OO1 arrives at the interface X = l, the solution
is the same as that in a semi-infinite bar discussed earlier, namely:
⎧
⎪ $ CA $ X/CA
⎪ 1 t
⎪σm1 (t) = t 0 σ0 (t)dt, σm1 (t) = σ0 (t)dt,
⎪
⎪ X 0
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ 1 $t
⎪vm1 (t) = −
⎪ σ0 (t)dt,
⎪
⎨ ρA CA t 0
(5.32)
⎪
⎪ X $t
⎪
⎪ = −
X
σ0 (t) +
⎪
⎪ σ1 (X, t) 1 σ0 (t)dt,
⎪
⎪ CA t CA t 2 0
⎪
⎪
⎪
⎪
⎪
⎪ $ X/CA
⎪
⎪ 1
⎩εm1 (X) = σ0 (t)dt
ρA CA X 0
The unloading boundary coincides with the propagating locus of the strong discontinuous
plastic loading wave OO1 .
At t = t0 = l/CA , the strong discontinuous plastic wave arrives at the interface, and the
wave reflection and transmission take place. Because a rigid mass exists at the interface,
the interface acceleration is finite, and moreover the B material is linearly elastic, so the
reflected wave O1 K is a strong discontinuous plastic wave, while the transmitted wave
O1 T is a weak discontinuous elastic wave. The propagating locus of the reflected strong
discontinuous plastic wave is expressed as
X = 2l − CA t (5.33)
The region in front of it is the region II that has experienced rigid unloading and is now
plastically reloaded, while the region behind it is the region IV, a new rigid unloading
region caused by the movement of rigid mass M.
Region II: In this region, the stress σ2 (X,t) should satisfy the following motion equation:
dv2
σ2 (X, t) = ρA X + σ0 (t). (5.34)
dt
From this equation, along the wave front O1 K, we have
dvm2
σm2 (t) = σ2 [ϕ(t), t] = ρA (2l − CA t) + σ0 (t). (5.35)
dt
Note that when a rigid unloading state is to be reloaded into a plastic state, the stress should
reach the maximum which is reached before unloading [point A on Fig. 5.1(a)], namely
σm2 (X) = σm1 (X).
Then from Eqs. (5.32) and (5.33), σm2 (X) can be expressed as
((2l/CA )−t)
CA
σm2 (t) = σ0 (t)dt.
2l − CA t 0
Substituting it into Eq. (5.35) and considering

t0
1
v2 (t0 ) = vm1 (t0 ) = − σ0 (t)dt
ρA CA t0 0
we have vm2 (t), namely v2 (t):

t t ((2l/CA )−η)
2
1 σ0 (τ ) 1 CA
v2 (t) = − + σ0 (τ )dτ dη
ρA t0 2l − CA τ ρA t0 (2l − CA η)2 0
t0
1
− σ0 (τ )dτ
ρA CA t0 0
(5.36)
Thus, substituting Eq. (5.36) into Eq. (5.34), the stress distribution in region II, σ2 (X, t),
can be obtained as:
((2l/CA )−t)
X CA X
σ2 (X, t) = 1 − σ0 (t) + σ0 (τ )dτ . (5.37)
2l − CA t (2l − CA t)2 0
The strain distribution in region II is the same as in region I.
ε2 (X) = ε1 (X)
Regions III and IV: The solutions in regions III and IV are coupled since the rigid mass M
should simultaneously satisfy the following two equations, representing the continuity
condition and momentum conservation condition, respectively:
v4 (t¯) = v3 (l, t¯) (5.38)

dv4
M = σ3 (l, t¯) − σ4 (l, t¯) (5.39)
d t¯
here the notation t¯ = t − (l/CA ) has been introduced. The region III is an elastic simple
wave region, hence we have
σ3 (l, t¯) = −ρB CB v3 (l, t¯) (5.40)
On the other hand, the region IV is a rigid unloading region, so the following equation
should be satisfied.
dv4
σ4 (l, t¯) − σm4 (t¯) = ρB CB t¯ (5.41)
d t¯
and according to the conservation condition across the strong discontinuous plastic wave
front O1 K, σm4 (t¯), and v4 (t¯) should furthermore satisfy
σm4 (t¯) = σm2 (t¯) + ρA CA {v4 (t¯) − v2 (t¯)} (5.42)
where σm2 and v2 have been obtained from the solutions in region II. From the five
equations given in Eqs. (5.38)–(5.42), eliminating σ3 (l, t¯), σ4 (l, t¯), σm4 (l, t¯), and v3 (l, t¯),
the following linear ordinal differential equation with respect to v4 (t¯) can be obtained
dv4 (t¯)
+ f1 (t¯)v4 (t¯) = f2 (t¯) (5.43)
d t¯
where
ρA CA + ρB CB
f1 (t¯) =
ρA CA t¯ + M
ρA CA v2 (t¯) + σm2 (t¯)
f2 (t¯) =
ρA CA t¯ + M
The solution of Eq. (5.43) is

/ 0
t¯ t¯
v4 (t¯) = f2 (η) exp − f1 (τ )dτ dη (5.44)
0 η
here the condition v4 (0) = 0 has been used to determine the integrate constant. Once v4 (t¯)
is determined, other variables in region IV can be determined too from Eqs. (5.41) and
(5.42). Consequently, v3 (l, t¯) and σ3 (l, t¯) can be determined from Eqs. (5.38) and (5.40),
and thereby the region III (elastic simple wave region) is solved.
Because of the unloading effect, the strong discontinuous amplitude of the leftward
reflected plastic wave O1 K continuously decreases in its propagation. If at t = tK < t1 ,
the following condition is satisfied
v4 (tk ) = v2 (tk ), σm4 (tk ) = σm2 (tk ) (5.45)
It means that the strong discontinuous plastic wave disappears. Henceforth, the whole bar
segment A is in an unloading process. Since v4 (t) and v2 (t) have been obtained, tK can
be determined from Eq. (5.45).
Region V: The region V still should be simultaneously solved with region III. Similar to
Eqs. (5.38)–(5.41), we have
⎫
v5 (t¯) = v3 (l, t¯), ⎪
⎪
⎪
⎪
dv5 ⎪
⎪
M = σ3 (l, t ) − σ5 (l, t ),⎪
¯ ¯ ⎬
dt (5.46)
σ3 (l, t¯) = −ρB CR v3 (i, t¯), ⎪ ⎪
⎪
⎪
dv5 ⎪ ⎪
⎪
⎭
σ5 (l, t¯) − σ0 (t¯) = ρA l
d t¯
Eliminating σ3 (l, t¯), σ5 (l, t¯), and v3 (l, t¯) from the above four equations, we obtain the
following linear ordinal differential equation with respect to v5 (t¯):
dv5 (t¯)
+ f3 v5 (t¯) = f4 σ0 (t¯) (5.47)
d t¯
ρB CB
f3 =
ρA l + M
1
f4 = −
ρA l + M
of which the solution is

2
t¯
v5 (t¯) = exp[−f3 · (t¯ − t¯K )] f4 · σ0 (η) · exp[f3 · (η − t¯K )]dη + v2 (t¯K )
t¯K
where the condition v5 (t¯K ) = v2 (t¯K ) has been used to determine the integrate constant.
Once v5 (t¯K ) is determined, similar to the situation discussed for region IV, both region V
and region III between time tK and t1 (i.e. tK ≤ t ≤ t1 ) are entirely solved.
Region VI: At t ≥ t1 , except σ0 (t) ≡ 0, all others are the same as discussed on region V.
Therefore, let all σ0 (t¯) in the associated equations in discussions on region V vanish
(σ0 (t¯) = 0), the solution for region VI can be obtained.
In the current example, if the rigid mass M ≡ 0, then the problem is reduced to solve the
stress wave propagation in the laminated media composed by a linear hardening plastic
loading–rigid unloading material (A) and a linear elastic material (B).
Alternatively, if the rigid mass M → ∞, then the problem is reduced to solve the stress
wave propagation in a finite bar made of a linear hardening plastic loading–rigid unloading
material and with a fixed end boundary condition. In such a case, we have f1 = f2 =
f3 = 0, v4 (X,t) = 0, and
((2l/CA )−t) t
CA CA
σ4 (X, t) = σm4 (t) = σ0 (τ )dτ + σ0 (τ )dτ
2l − CA t 0 t0 2l − CA τ
t/ 0
CA2 ((2l/CA )−η) 1 t0
− σ0 (τ )dτ dη + σ0 (t)dt
t0 (2l − CA η)2 0 t0 0
5.3 Rigid Unloading Analyses for Shock Wave Propagation
In the previous two sections, the strong discontinuous plastic loading waves concerned are
those formed in linear hardening plastic materials under the boundary condition of suddenly
applied loading. In such cases, the strong discontinuous wave velocity is constant, so the
problem is relatively easier to deal with. However, if the shock waves form in a gradually
increasing hardening plastic material, wherein the shock wave velocity varies with shock
intensity, the problem is then relatively complicated. In the present section, we discuss the
rigid unloading in such a case.
In the following, we first discuss an example of semi-infinite bar (Kaliski et al., 1967),
and then discuss an example of finite bar (Lee and Tupper, 1954).
5.3.1 Rigid unloading of shock wave propagating in semi-infinite bars
Consider a semi-infinite bar with a stress–strain relationship as shown in Fig. 5.7(a),

namely, for the loading:

σ = E1 ε, σ ≤ σA
ε = εA , σ ≥ σA
therefore if σ ≤ σA , then the plastic velocity is constant C1 = (E1 /ρ0 )1/2 , and if σ > σA ,
then a shock wave will form and its velocity D is:
9
σ
D=
ρ0 εA
while for unloading, it shows a rigid unloading behavior. Such a stress–strain relationship
can be regarded as a simplified model of the actual stress–strain relationship for soils as
shown in Fig. 5.7(b).
Assume that the semi-infinite bar is originally at rest and stress-free, at time t = 0 it is
suddenly loaded by a boundary stress larger than σA , and then gradually unloaded. Thus,
s s
sA A
E1
o eA e o e
(a) (b)
Fig. 5.7. Two stress–strain models that may result in shock waves: (a) linear hardening plastic
loading–rigid unloading; (b) stress–strain relation for soils. Note that (a) can be regarded as a
simplified model of (b).
from the bar end, a gradually attenuating plastic shock wave propagates rightwards. Its
propagating locus X = ϕ(t) is also the rigid unloading boundary, as plotted in Fig. 5.8.
According to the kinematic compatibility condition and the dynamic compatibility condi-
tion across a shock wave front, the plastic stress value σm (t), strain εA , particle velocity
vm (t), and the shock wave velocity ϕ (t) should satisfy the following relations:
vm (t) = −ϕ (t)εA (5.48)

σm (t) = −ρ0 ϕ (t)vm (t) (5.49)
t t
C1
t)
j(
t
)
s0 (t
=
X
s smax o X
Fig. 5.8. Gradual attenuation of a plastic shock wave propagating in a semi–infinite bar due to rigid
unloading.
On the other hand, the rigid unloading segment of bar should comply with
d v̄
ρ0 X =
σ (X, t) − σ0 (t). (5.50)
dt
Applying it along the unloading boundary X = ϕ(t), noticing v̄(t) = vm (t), we have
dvm
ρ0 ϕ = σm (X, t) − σ0 (t).
dt
Substituting Eqs. (5.48) and (5.49) into the above equation and after rearranging, we obtain
2 σ0 (t)
ϕϕ + ϕ = = f (t) (5.51)
ρ 0 εA
or namely

ϕ 2 = 2f (t).

Integrating it twice and using the initial conditions ϕ(0) = 0, (ϕ 2 ) t=0 = 2ϕϕ t=0 = 0
to determine the integrate constants, the unloading boundary can be solved as:
t τ 1/2
ϕ(t) = 2 dτ f (ξ )dξ (5.52)
0 0
Correspondingly, the shock wave velocity, or namely the propagating velocity of unloading
boundary ϕ (t) is
$t
1 t 0 f (ξ )dξ
ϕ (t) = f (ξ )dξ = $ $τ 1/2
ϕ 0 t
2 0 dτ 0 f (ξ )dξ
Substituting it into Eqs. (5.48)–(5.50), the particle velocity vm (t) and plastic stress σm (t)
along the unloading boundary, as well as the stress σ (X, t) of the rigid unloading segment,
which is varied with X and t, can be obtained, respectively.
$t
0 f (ξ )dξ
vm (t) = −εA $ $τ 1/2
t
2 0 dτ 0 f (ξ )dξ
$t
0 f (ξ )dξ
σm (t) = ρ0 εA $ $τ 1/2
t
2 0 dτ 0 f (ξ )dξ

σ (X,t) =
$ $τ 1/2 $ 2 $ $τ −1/2
t t t
f (t) 2 0 dτ 0 f (ξ )dξ − 0 f (ξ )dξ 2 0 dτ 0 f (ξ )dξ
σ0 (t)−ρ0 εA X $t $τ
2 0 dτ 0 f (ξ )dξ
Obviously, the rightward propagating shock wave continuously attenuates due to the rigid
unloading effect. If at time t = t ∗ ,
σm (t ∗ ) = σA
then after that the plastic wave velocity is constant C1 , the problem can be dealt with as
what we discussed earlier for the linear hardening materials.
5.3.2 Rigid unloading of shock wave propagating in finite bars
Finally, consider an example of rigid unloading of shock wave propagating in a finite

bar (Lee and Tupper, 1954). With regard to the example discussed earlier in Section 4.9,
namely, for a finite, increasingly hardening plastic bar impacts on a rigid target as shown in
Fig. 4.28, if the elastic wave velocity, either at loading or at unloading, is relatively much
larger than that of the plastic wave velocity and consequently can be regarded as infinite,
and moreover the elastic deformation is relatively much smaller than the plastic deforma-
tion and consequently can be neglected, then the elastic-plastic stress–strain relationship
shown in Fig. 4.28(d) can be replaced by the rigid-plastic stress–strain relationship shown
in Fig. 5.9(a). This is equivalent to such a problem of stress wave analysis: all the bar
segments where the plastic shock wave front has not arrived yet, can be regarded as a
rigid body without any deformation, while all the bar segments where the plastic shock
wave front has passed also can be regarded as a rigid body without any elastic recovery
deformation.
If the finite bar impacts on the rigid target with a constant velocity v∗ [Fig. 5.9(b)], then
a rigid plastic shock wave propagates leftward with a shock velocity D from the bar end
X = l, then the stress jump across the wave front is (σ − Y ), the strain jump is (ε − 0), and
the particle velocity jump is (0 − v). According to the jump conditions across the leftward
e v*
h lin
y leig
Ra o
D
l X
Y
o e
(a) (b)
Fig. 5.9. Rigid unloading of a plastic shock wave propagation in a finite bar: (a) the loading and
unloading stress–strain curve; (b) a finite bar impacts onto a rigid target with a constant velocity v∗ .
shock wave front, we have
v = −Dε (5.53)
σ − Y = −ρ0 Dv (5.54)
Eliminating the v from these equations, the plastic shock wave velocity D is obtained:
σ −Y
D2 = (5.55)
ρ0 ε
while eliminating D from Eqs. (5.53) and (5.54), we have
ρ0 v2 = (σ − Y )ε (5.56)
which gives the relation between v and σ , when the σ ∼ ε relationship is given. For
example, from the initial impact velocity v*, the corresponding initial shock amplitude σ ∗
can be determined.
The bar segment where the plastic shock wave has not yet arrived, should satisfy the
following motion equation for rigid body:
dv(t)
ρ0 X =Y (5.57)
dt
where X is the length of the bar segment in front of shock wave. Noticing that the propa-
gating locus of the shock wave front X = ϕ(t) or t = ψ(X) is also the locus of the rigid
boundary, Eq. (5.57) can be rewritten as:
dv[ψ(X)]
−ρ0 XD =Y
dX
Eliminating D from Eq. (5.53), we have
dv
ρ0 Xv = Yε
dX
d(v2 )
ρ0 X = 2Y ε
dX
Moreover, eliminating v2 from Eq. (5.55) gives
d[(σ − Y )ε] dX
=2
Yε X
Integrating it, noting σ = σ ∗ at X = l, and introducing
(σ −Y )ε d [(σ − Y ) ε]
F (σ ) =
0 Yε
we obtain
2
X
ln = F (σ ) − F (σ ∗ ) (5.58)
l
After some mathematical calculations, and noticing Eq. (5.55), F (σ ) can be expressed in
the following form:
(σ −Y )ε
d [(σ − Y ) ε] σ dσ ε(σ ) σ −Y
F (σ ) = = + dε
0 Yε Y Y 0 Yε
/ 0 σ 2
1 ε(σ ) 1 D
= σ + ρ0 D dε − 1 =
2
σ+ 2
dσ − 1
Y 0 Y Y C
Since for a given σ ∼ ε curve, we knew D2 = ρ1 σ −Y and C 2 = ρ1 dσ , this is also a known

0 ε 0 dε
function. Thus Eq. (5.58) determines the stress distribution σm (X) and the corresponding
strain distribution εm (X) along the rigid plastic boundary. In addition, it can be seen from
Eq. (5.58) that the distributions of maximum plastic stress and strain are merely a function
of X/ l.
Let σ = Y in Eq. (5.58), the location where the plastic wave vanishes, and consequently
the total length of the plastic deformation distributing in bar is determined. Obviously,
the plastic shock wave must vanish before it arrives at the stress-free end. This is under-
standable: in the process the plastic shock wave propagates towards the stress-free end,
the shock wave continuously interacts with the rigid unloading waves reflected from the
stress-free end with a wave velocity of ∞, till the shock wave vanishes.
The results obtained by both the elastic–plastic wave theoretical analysis and the rigid
unloading analysis for the same problem are compared as shown in Fig. 4.28 (Lee and
Tupper, 1954). In Fig. 4.28(a), the propagating locus of shock wave front given by the
rigid unloading analysis is marked by solid dots, and the locus given by the elastic–plastic
wave analysis is shown by the solid line. As can be seen, the former basically coincides
with the latter. The residual strain distribution is shown in Fig. 4.28(b), wherein the result
given by the rigid unloading analysis is plotted by dashed line while the result given by the
elastic–plastic wave theory is plotted by the solid line. It is shown that the rigid unloading
analysis averages the original discontinuous distribution of residual strain obtained in
the elastic–plastic wave analysis so that its distribution is continuous. Obviously, with
increasing impact velocity, the plastic deformation of bar must increase and the number
of unloading wave reflected from the stress-free end must increase, then the approximate
result obtained by the rigid unloading analysis is closer to the elastic–plastic wave solution.
CHAPTER 6
One-Dimensional Visco-Elastic Waves and

Elastic-Visco-Plastic Waves
The results of elastic–plastic wave propagation discussed so far are all based on the basic
assumption that stress is only dependent on strain and not on strain rate. This kind of
stress wave theory is called rate-independent theory. However, the mechanical behavior
of real material is, more or less, rate dependent or time dependent. For example, the strain
increase with increasing time under constant stress (the phenomenon called creep), the
stress decrease with increasing time under constant strain (the phenomenon called stress
relaxation), the hysteresis loop phenomenon in a loading–unloading or stress cyclic
process, the absorption and dispersion phenomenons in stress wave propagation, and the
strengthening and embrittlement of materials under high strain rates, etc., all show that
the constitutive response of real material is essentially time dependent or rate dependent.
In such a situation, to describe the rate-dependent constitutive relationship of material,
a curve on the (σ, ε) plane, f (σ, ε) = 0, is no longer appropriate; at least a curved
surface in the (σ, ε, ε̇) space, f (σ, ε, ε̇) = 0, should be used. The stress wave theory
based on such rate-dependent constitutive relationship is called rate-dependent theory
or strain-rate-dependent theory.
The mechanical response of rate-dependent materials in general can be separated into

two parts: the time-independent instantaneous response and the time-dependent non-
instantaneous response. Macroscopically, all kinds of noninstantaneous responses may
be attributed to or equivalent to some kinds of viscous responses induced by the so-called
internal dissipative force. Thus, based on the elastic response to take such viscous effects
into account, the visco-elastic wave theory has been correspondingly developed, and has
been extensively applied to study polymer materials. In wave propagation, the viscous
effect mainly results in the dispersion phenomenon (i.e. the wave velocity depends on
the wave frequency) and the absorption phenomenon (i.e. the wave amplitude attenuates
with the propagating distance). Moreover, based on the elastic–plastic response to take the
viscous effects into account for both the elastic part and the plastic part, the corresponding
visco-elastic-plastic wave theory has been developed. However, if the viscous effect is
taken into account only for the plastic part, then correspondingly the elastic-visco-plastic
219
wave theory has been developed, which has become one of the most important rate-
dependent plastic wave theories since the –Malvern model was proposed
around the 1950s.
In the following, we will mainly discuss the one-dimensional longitudinal visco-elastic

waves and elastic-visco-plastic waves.
6.1 Linear Visco-Elastic Constitutive Relationship
The development of visco-elastic theory is closely related to the studies on the mechanical
behavior of high-molecular polymers. It is well known that even in a usual range of
temperature or time change, polymers display a markedly viscous behavior, showing
almost all the behavior patterns of the typical viscous fluid to the typical elastic solid.
The simplest constitutive relationship for describing the elastic solid is the linear elastic
stress–strain law (Hooke’s law), while the simplest constitutive relationship for describing
the viscous fluid is the linear viscous law between the stress and the velocity gradient
(Newton’s law), of which the one-dimensional forms are expressed, respectively, as
σ = Eε (6.1)
∂v
σ =η (6.2)
∂x
where η is the viscous coefficient, and others have been defined previously. Under the
condition of small deformation, the difference between Lagrange variables and Euler
variables can be neglected; then, the velocity gradient and the strain rate are the same:
∂v ∂v ∂ 2u ∂ε
≈ = = = ε̇
∂x ∂X ∂X∂t ∂t
so Eq. (6.2) can be rewritten as:
σ = ηε̇ (6.3)
It means that the relationship between the stress and strain rate is linear.
In the scope of linear constitutive relationship, a medium having both elastic and viscous
character can be imaged as a linear combination of linear elastic solids and linear viscous
fluids, called linear visco-elastic body, of which the constitutive relationship is a kind
of linear combination of Eqs. (6.1) and (6.3). If the linear elastic body is represented by
a spring, as shown in Fig. 6.1(a), and the linear viscous fluid by a dashpot, as shown in
Fig. 6.1(b), then the simplest form of linear combination is either a body connected to a
spring with a dashpot in series, called Maxwell body, as shown in Fig. 6.2, or a body
connected to a spring with a dashpot in parallel, called Kelvin–Voigt body, as shown in
Fig. 6.3.
One-Dimensional Visco-Elastic Waves and Elastic-Visco-Plastic Waves 221
s = h (de /dt)
s = Ee
(a) (b)
Fig. 6.1. (a) Spring model and (b) Dash-pot model.
EM
hM
Fig. 6.2. Maxwell model.
Ev
hv
Fig. 6.3. Kelvin–Voigt model.

6.1.1 Maxwell model
For the Maxwell model (Fig. 6.2), the stress carried by the spring equals that carried by
the dashpot:
σ = σE = ση
while the total strain is the sum of that carried by the spring and the dashpot, respectively:
ε = ε E + εη
Consequently, the total strain rate is the sum of both parts too:
ε̇ = ε̇E + ε̇η
Substituting Eqs. (6.1) and (6.3) into the above equation, we obtain the following
constitutive equation for Maxwell model:
σ̇ σ
ε̇ = + (6.4)
EM ηM
Thus, under the constant strain condition (ε̇ = 0), stress will attenuate with time from the
initial stress σ0 according to the following exponential function:

EM t t
σ (t) = σ0 exp − = σ0 exp − (6.5)
ηM θM
which approximately describes the stress relaxation behavior of materials, as shown in

Fig. 6.4, where θM = ηM /EM represents the time required to let the stress relax from σ0
to its 1/e (36.79%), called the relaxation time. It is an important material parameter used
to characterize visco-elastic materials.
s0
s(
t)=
s
s0 0 e −t
/q
e M
o
qM t
Fig. 6.4. Stress relaxation under constant strain (ε̇ = 0).

A relaxation modulus is defined by
σ (t)
Er (t) = (6.6)
ε
Equation (6.5) can be rewritten as:

σ0 t t
Er (t) = exp − = EM exp − (6.7)
ε θM θM
Therefore, the relaxation character is independent of the value of applied constant strain.
Of course, the Maxwell model is a simplified ideal model. In fact, to describe the practical
stress relaxation behavior of most true materials as close to the experimental results as
possible, a simplified model containing only one exponential attenuation term is insuffi-
cient, and usually a sum of a series of exponential terms with different relaxation times
should be used. In other words, not a single relaxation time but a series of relaxation
times θi (i = 1, 2, 3, . . . n) should be used to describe the true visco-elastic behavior
of materials. Therefore, under the condition of constant strain, the relation of how stress
relaxes with time should be expressed as:
.
n
t
σ (t) = ε Ei exp − (6.8a)
θi
i=1
or, expressed by relaxation modulus, we have
.
n
t
Er (t) = Ei exp − (6.8b)
θi
i=1
This is equivalent to a visco-elastic body that is combined by a series of Maxwell elements

connected in parallel, each with different Ei and θi (or ηi ), called generalized Maxwell
body, as shown in Fig. 6.5. The Ei in the equation can be regarded as the contribution of
the i th Maxwell element with relaxation time θi to the total stress relaxation, the weight.
When n → ∞, the series of discrete relaxation time approaches continuous distribution,
and the Ei is replaced by the weight function E(θ ) dθ . Thus, the above equation in the
sum form transfers into the following equation in the integral form:
∞
t
Er (t) = E(θ) exp − dθ (6.9)
0 θ
E(θ ) is called relaxation-time spectrum, or relaxation spectrum for short. Sometimes,

the distribution of relaxation may be very wide, and it will be more convenient to use
ln(θ ) as an independent variable to investigate the relaxation spectrum. For this reason,
introduce
H (ln θ) d ln θ = E(θ) dθ
Ei−2 Ei−1 Ei Ei+1 Ei+2
hi−2 hi−1 hi hi+1 hi+2
Fig. 6.5. Generalized Maxwell model.
Equation (6.9) is then rewritten as:
∞ ∞
t 1 t
Er (t) = H (ln θ ) exp − d ln θ = H (θ) exp − dθ (6.10)
−∞ θ 0 θ θ
where H (ln θ) is the relaxation spectrum with respect to ln θ .
6.1.2 Kelvin–Voigt model
For the Kelvin–Voigt model (Fig. 6.3), strain experienced by the spring equals strain
experienced by the dashpot:
ε = εE = εη
while the total stress is the sum of both parts:
σ = σ E + ση
Substituting Eqs. (6.1) and (6.2) into the above equation, we obtain the following
constitutive relationship for Voigt model:

∂
σ = Ev ε + ηv ε̇ = Ev + ηv ε (6.11)
∂t

which can be formally obtained from Hooke’s law by using the operator Ev + ηv ∂t∂
instead of the elastic modulus E.
Thus, under the condition of constant stress σ , assuming a zero initial strain, the strain
increases with time according to the following rule:
⎫
ε(t) =
σ
1 − exp −
t ⎪
⎬
Ev τv (6.12)
τv =
ηv
, ⎪
⎭
Ev
which approximately describes the creep behavior of materials, as shown in Fig. 6.6,
where τv = ηv /Ev . If the stress is suddenly unloaded from σ to zero at ε = ε0 , Eq. (6.11)
shows that the strain displays a delayed recovery, according to the following rule:

t
ε(t) = ε0 exp − (6.13)
τv
at t = τv , the strain is recovered to 1/e (36.79%) of its initial value, where τv is called
retardation time. It is another important material parameter used to characterize visco-
elastic materials.
Similar to the definition of relaxation modulus, creep compliance is defined by:
ε(t)
Jc (t) = (6.14)
σ
then Eq. (6.12) is reduced to:

1 t t
Jc (t) = 1 − exp − = Jv 1 − exp − (6.15)
Ev τv τv
where Jv = 1/Ev is the compliance of spring element in Voigt model.
s asymptotic line
Ev
ep
cre
rec
ov
er
s = const. s =0
0
t
Fig. 6.6. Creep under constant stress and a delayed recover of strain when stress is suddenly
unloaded from σ to zero.
Ei
hi
Fig. 6.7. Generalized Kelvin–Voigt model.
Similar to what we have discussed on Maxwell body, a Kelvin–Voigt body with only
a single retardation time is insufficient to precisely describe the creep behavior of true
materials, so we introduce the generalized Kelvin–Voigt body combined by a series of
Kelvin–Voigt elements connected in series, each with different retardation time, as shown
in Fig. 6.7. Then, similar to Eqs. (6.9) and (6.10), we have
∞ ∞
t t
Jc (t) = J (τ ) 1 − exp − dτ = L(ln τ ) 1 − exp − d ln τ (6.16)
0 τ −∞ τ
J (t) and L(ln τ ) are the retardation time spectrum as functions of τ and ln τ , respectively.
6.1.3 Standard linear solid model
This points out that, although Maxwell body can approximately describe stress relaxation,
the strain rate is constant under the constant stress condition, according to Eq. (6.4).
Therefore, this, in fact, just reflects Newton viscous flow and is insufficient to describe the
more complicated true creep phenomenon. Moreover, when it is used to describe stress
relaxation, according to Eq. (6.5), after a sufficiently long time period, the stress will relax
to zero, while in practice the relaxation stress of true materials usually approaches a finite
value under the constant strain condition.
On the other hand, although the Kelvin–Voigt body can approximately describe the creep
phenomenon, the stress, according to Eq. (6.11), is constant under constant strain condition
and is unable to reflect the stress relaxation behavior.
In order to more satisfactorily describe stress relaxation as well as the creep, more compli-
cated models, combined with springs and dashpots, are obviously required. Among these,
the three-element model is attractive and not too complicated, which may be either com-
bined with a Maxwell body connected to a spring Ea in parallel [Fig. 6.8(a)], or combined
with a Kelvin–Voigt body connected to a spring Ea in series [Fig. 6.8(b)]. The constitutive
equations for them are, respectively, expressed as:
EM σ + ηM σ̇ = EM Ea ε + (EM + Ea )ηM ε̇ (6.17)

Ea + Ea σ + ηv σ̇ = Ea Ea ε + Ea ηv ε̇ (6.18)
Obviously, if Ea = Ea + EM , Ev = Ea (1 + (Ea /EM )), and ηv EM Ea = ηM Ev Ea , then

these two equations, or rather these two models, are equivalent to each other. In fact,
E 'a
EM
Ea
hM hv Ev
(a) (b)
Fig. 6.8. Standard linear solid model.
both of them can be regarded as a linear combination of Eqs. (6.4) and (6.11):
a0 σ + a1 σ̇ = b0 ε + b1 ε̇
This model is sometimes called a standard linear solid model. Of course, in order to
more precisely describe the general behavior of true visco-elastic bodies, usually a more
general linear visco-elastic model is required, of which the constitutive equation can be
written as:
.
n
∂ iσ . ∂ iε
n
ai i
= bi i
∂t ∂t
i=0 i=0
where ai and bi (i = 1, 2, 3, . . . n) are all material parameters.
6.2 Stress Waves Propagating in Linear Visco-Elastic Bars
Let us now discuss the stress waves propagating in linear visco-elastic bodies, and emphati-
cally investigate how the viscous terms in constitutive equations influence stress wave
propagation. For convenience sake, in the following, we will mainly discuss the longi-
tudinal waves propagating in bars made of Maxwell body, Kelvin–Voigt body, and the
standard linear solid body.
6.2.1 Longitudinal visco-elastic waves propagating in Kelvin–Voigt bars
The control equations for the longitudinal visco-elastic waves propagating in Kelvin–Voigt
bars consist of Eqs. (2.12), (2.13), and (6.11), namely:
⎫
∂v ∂ε ⎪
= ⎪
⎪
∂X ∂t ⎪
⎪
⎬
∂v ∂σ
ρ0 = ⎪
(6.19)
∂t ∂X ⎪
⎪
∂ε ⎪
⎪
σ = Ev ε + ηv ⎭
∂t
If the displacement u is taken as the unknown variable, then equivalently we have
∂ 2u ∂ 2u ∂ 3u
ρ0 = Ev + η v (6.20a)
∂t 2 ∂X 2 ∂X 2 ∂t
which, by introducing Cv2 = Ev /ρ0 , can be rewritten as:
∂ 2u 2
2∂ u ∂ 3u
− C v − C 2
v τ v =0 (6.20b)
∂t 2 ∂X 2 ∂X 2 ∂t
This is a third-order partial differential equation with respect to u. When the retardation
time τv is so small that the third viscous term could be neglected, the Eq. (6.20b) is reduced
to the second-order partial differential equation.
Because of the existence of the viscous term, Eq. (6.20b) is no longer satisfied by the
elastic solution u = F (X ± C0 t). Let us try the following harmonic wave solution:
u(X, t) = A · exp[i(ωt − k1 X)]
where A is the wave amplitude, ω(= 2πf ) the radial frequency (f the frequency), and k1
the wave number. If the above solution could satisfy Eq. (6.20), then we must have
ρ0 ω2 = Ev k12 + iηv k12 ω (6.21)
This is an expression in complex form. Since it has been assumed that the radial frequency ω
is a real number, then k1 must be a complex number, say
k1 = k + iα
Substituting it into Eq. (6.21), let the real part and the imaginary part on the left side of
the equation equal those on the right side, respectively. We have

ρ0 ω2 = Ev k 2− α 2 −2ηv ωαk
(6.22)
2Ev αk = −ηv ω k 2 − α 2
Thus, the solutions are obtained as follows:

/ 0 <
ρ 0 Ev ω 2 ηv2 ω2 ω2
k = 2
2 1 + + 1 = 1 + ω 2τ 2 + 1
v
2 Ev + ηv2 ω2 Ev2 2Cv2 1 + ω2 τv2
(6.23a)
/ 0 <
ρ 0 Ev ω 2 ηv2 ω2 ω2
α2 = 2 1 + − 1 = 1 + ω 2τ 2 − 1
v
2 Ev + ηv2 ω2 Ev2 2Cv2 1 + ω2 τv2
(6.23b)
where α could be positive or negative. However, the positive root implies that the wave
amplitude increases with X, with no physical meaning (violating the energy conservation);
therefore, α must be negative and for convenience could be rewritten as (−α). Thus, the
solution of Eq. (6.20) is
u(X, t) = A exp(−αX) exp [i(ωt − kX)] (6.24)
It is thus clear that the α describes the wave amplitude attenuation exponential with increas-
ing propagating distance. This phenomenon is called absorption, and α is called the
attenuation factor. Equation (6.23b) shows that α varies with ω. For the low-frequency
waves, of which the radial frequency ω 1/τv (namely, the period T = 2π/ω is much
larger than the retardation time τv ), α is proportional to ω2 .
Equation (6.23a) shows that the phase velocity of disturbance propagation, C(=ω/k),
depends on the radial frequency ω (or the period T = 1/f = ω/(2π)):
/ 0
2πτv 2
2Cv 1 +
2
2Cv2 1 + ω2 τv2 T
C =
2
; = (6.25)
1 + 1 + ω2 τv2 2πτv 2
1+ 1+
T
Thus, the profile of wave discussed must distort in the wave propagation process. This
phenomenon is called dispersion (chromatic dispersion). For the low-frequency waves
satisfying ω 1/τv , the phase velocity equals the longitudinal wave velocity in elastic
bars, C = C0 , while the phase velocity increases with increasing frequency.
6.2.2 Longitudinal visco-elastic waves propagating in Maxwell bars
The longitudinal waves in Maxwell bars can be discussed in a similar way, for which the
governing equations consist of Eqs. (2.12), (3.13), and (6.4), namely:
⎫
∂v ∂ε ⎪
= ⎪
⎪
∂X ∂t ⎪
⎪
⎪
⎬
∂v ∂σ
ρ0 = (6.26)
∂t ∂X ⎪
⎪
⎪
∂ε 1 ∂σ σ ⎪⎪
⎪
= + ⎭
∂t EM ∂t ηM
or, in terms of the displacement u as unknown variable, equivalently we have
∂ 3u ∂ 3u ρ 0 EM ∂ 2 u
ρ0 − E M + =0 (6.27a)
∂t 3 ∂X 2 ∂t ηM ∂t 2
2 = E /ρ , the above equation also can be rewritten as:
Introducing CM M 0
∂ 2u 3
2 ∂ u 1 ∂ 2u
− C M + =0 (6.27b)
∂t 2 ∂X 2 ∂t θM ∂t 2
It is easy to demonstrate that the above equation also is satisfied by the harmonic solution
[Eq. (6.24)], although the wave number k and the attenuation factor α should be determined
by the following equations, respectively:
/ 0
ω2 1
k =
2
2
1+ 2
+1 (6.28a)
2CM 2
ω θM
/ 0
ω2 1
α =
2
2
1+ 2
−1 (6.28b)
2CM 2
ω θM
Thus, similar to what we have discussed on Kelvin–Voigt bars, the waves in Maxwell bars
also display the dispersion and absorption phenomena. However, for the high-frequency
waves, when ω 1/θM , the phase velocity equals the longitudinal wave velocity in
elastic bars, and, in such cases, the attenuation factor α is independent of the frequency:
1 ρ0 CM
α= = (6.29)
2CM θM 2ηM
6.2.3 Longitudinal visco-elastic waves propagating in standard linear solid bars
For the standard linear solid bar shown in Fig. 6.8(a), similarly, from Eqs. (2.12),
(2.13), and (6.17), the governing equation in terms of u as an unknown variable can be
deduced as:
∂ 3u ρ0 ∂ 2 u ∂ 3u Ea ∂ 2 u
ρ0 + = (E a + E M ) + (6.30)
∂t 3 θM ∂t 2 ∂t∂x 2 θM ∂x 2
where θM = ηM /EM . Moreover, it is easy to demonstrate that the above equation is

also satisfied by the harmonic solution [Eq. (6.24)], although the wave number k and
attenuation factor α should be determined by the following equations, respectively
⎧⎡ ⎤1/2 ⎫
⎪
⎪ EM ⎪
⎪⎢
⎨ ⎥ 1+ 1+ ω θM ⎪
2 2 ⎪
⎬
ρ0 ω 2 ⎢ 1 + ω2 θM
2
⎥ Ea
k2 = ⎢ 2 ⎥ + (6.31a)
2Ea ⎪
⎪ ⎣ EM ⎦ EM 2 2 2 ⎪ ⎪
⎪
⎩ 1+ 1+ 2
ω2 θM 1+ 1+ ω θM ⎪
⎭
Ea Ea
⎧⎡ ⎤1/2 ⎫
⎪
⎪ EM ⎪
⎪
⎨⎢ ⎥ 1+ 1+ ω θM ⎪
2 2 ⎪
⎬
ρ0 ω 2 ⎢ 1 + ω2 θM
2
⎥ Ea
α2 = ⎢ 2 ⎥ − (6.31b)
2Ea ⎪
⎪ ⎣ EM ⎦ EM 2 2 2 ⎪ ⎪
⎪
⎩ 1+ 1+ 2
ω2 θM 1+ 1+ ω θM ⎪
⎭
Ea Ea
On the other hand, for the standard linear solid bar shown in Fig. 6.8(b), similarly, from
Eqs. (2.12), (2.13), and (6.18), the governing equation in terms of u as unknown variable
can be deduced as:
∂ 3u ∂ 2u 3
∂ u
2
∂ u
ρ 0 ηv + ρ 0 E a + E v − η v E a − E v E a =0 (6.32)
∂t 3 ∂t 2 ∂X 2 ∂t ∂X 2
Moreover, it is easy to demonstrate that the above equation is also satisfied by the harmonic
solution [Eq. (6.24)], although the wave number k and attenuation factor α should be
determined by the following equations, respectively:
⎡ ⎤
2
ρ0ω2 ⎣ Ea + Ec2 ω2 τv2 Ea+ Ec ω2 τv2 ⎦
k2 = + (6.33a)
2Ec Ea 1 + ω2 τv2 1 + ω2 τv2
⎡ ⎤
2 + E ω2 τ 2
ρ 0 ω 2
⎣ E + E 2 ω2 τ 2
v E α v ⎦
α2 = a c
− a (6.33b)
2Ec Ea 1 + ω2 τv2 1 + ω2 τv2
where Ec is the elastic modulus of the springs Ev and Ea connected in series:
1 1 1
= +
Ec Ea Ev
According to either Eq. (6.31a) or Eq. (3.33a), the phase velocity of disturbance propaga-
tion, C(= ω/k), is dependent on the frequency, which describes the dispersion (chromatic
dispersion) phenomenon. On the other hand, according to Eqs. (6.31b) or (6.33b), the
attenuation factor α is also dependent on the frequency ω, which describes the absorp-
tion phenomenon of visco-elastic waves in standard linear solid bars. In other words, the
visco-elastic waves propagating in the standard linear solid bars must distort according to
the rules expressed by Eqs. (6.31) and (6.33).
For the high-strain-rate loading such as in the cases of explosion and impact, more attention
is paid to the high-frequency waves. In the following, especially such cases are analyzed
in a little detail. For example, for the standard linear solid body shown in Fig. 6.8(a), in
the case of high-frequency waves, because of ω2 1/θM 2 , Eq. (6.31a) is reduced to:
ρ0 ω 2
k2 ≈
Ea + E M
Consequently, the corresponding phase velocity is expressed as:
ω2 Ea + EM
C2 = 2
= = CV2 (6.34)
k ρ0
which shows that the phase velocity for high-frequency waves is independent of the fre-
quency. In other words, it means that for the standard linear solid shown in Fig. 6.8(a), in
the case of high frequency, the dashpot in Maxwell element does not take effect, and then
the standard linear solid body is reduced to an elastic body that consists of a spring Ea
connected with a spring EM in parallel. Therefore, the wave velocity is just equal to the
elastic wave velocity CV in such an elastic body, as expressed by Eq. (6.34).
To discuss the attenuation factor α in the case of high frequency, if we directly start from
Eq. (6.31b) and eliminate the term (ω2 θM 2 ) in both the numerator and the denominator in
the right side of the equation, then an incorrect conclusion (αC0 /ω = 0 in the case of high
frequency) will be reached1 . In order to obtain a correct conclusion, Eq. (6.31b) is first
rewritten in the following form (Wang et al., 1994):
⎧⎡ ⎤1/2 ⎫
⎪
⎪ 1 ⎪
⎪
⎪
⎪ ⎢ ⎥ 1+ ⎪
⎪
⎪
⎪ ⎢ 1 ⎥ EM ⎪
⎪
2
⎪
⎨ ⎢ 1+ 2 ⎥ 1+ 2
ω2 θM ⎪
⎬
2
ρ0 ω ⎢ ω θM ⎥ Ea
α2 = ⎢ ⎥ −
EM ⎪⎪ ⎢ 1 ⎥ 1 ⎪
⎪
2Ea 1 + ⎪
⎪⎢ 1+ ⎥ 1+ 2 ⎪
⎪
⎪
Ea ⎪ ⎣ EM 2 2 2 ⎦ EM ⎪
⎪
⎪
⎩ 1+ ω θM 1+ 2
ω2 θM ⎪
⎭
Ea Ea
(6.35)
For high-frequency waves, 1/ω2 θM 2 1, the following terms in Eq. (6.35) can be
expanded into a series by neglecting the small quantities of higher orders:
% &1/2 % &
1 1 1
1+ 2
≈1+ 2
2
ω θM 2 2
ω θM
/ 01/2 / 0
1 1 1
1+ ≈1−
2
(EM /Ea )2 ω2 θM 2 (1 + (EM /Ea ))2 ω2 θM
2
1 1
≈1−
1+
1 (1 + (EM /Ea ))2 ω2 θM
2
(1 + (EM /Ea ))2 ω2 θ22
We then obtain the attenuation factor for the high-frequency waves in standard linear solid
(denoted as αLS in the following):
% & / 02
ρ0 ω 2 1 1 1 1 1
αLS =
2
− +
2(Ea +EM ) 2 ω2 θM 2 (1+(EM /Ea ))ω2 θM 2 2 (1+(EM /Ea ))2 ω2 θM 2
2 2
ρ0 1 ρ 0 EM
= 1− = 2
4θM (Ea +EM )
2 (1+(EM /Ea )) 4θM (Ea +EM )3
(6.36a)
This means that the attenuation factor for high-frequency waves is neither dependent
on frequency nor equal to zero. Convenient for comparing with the attenuation factor
1 See Hillier (1949), or as cited by Kolsky (1953) and Graff (1975).

for the high-frequency wave in Maxwell body (denoted as αM in the following), the α in
Eq. (6.29), Eq. (6.36a) can be rewritten in the following several different forms by noticing
θM = ηM /EM , CM 2 = E /ρ , and C 2 = (E + E )/ρ (Eq. 6.34):
M 0 V a M 0
EM ρ 0 Cv ρ0 CM αM
αLS = = 2 = 3/2 =
2θM Cv (Ea +EM ) Ea Ea Ea 3/2
2ηM 1+ 2ηM 1+ 1+
EM EM EM
(6.36b)
Since Ea /EM > 0, it follows that αLS < αM . It means that the high-frequency wave
attenuation in the standard linear solid body is slower than that in the corresponding
Maxwell body.
It should distinguish the dispersion due to the transverse inertia in bars, as discussed in
Section 2.8, from the dispersion due to the viscous effect, as discussed in the present
section. The former results from the bar’s geometrical factor and is sometimes called geo-
metrical dispersion, while the latter results from the viscosity of material and is sometimes
called constitutive viscous dispersion. Note that the frequency effect in these two kinds of
dispersion is expressed in two opposite directions. For visco-elastic waves, high-frequency
wave velocity is faster than low-frequency wave velocity, while for geometrical dispersion,
the situation is contrary (see Eq. 2.68).
6.2.4 Solutions of linear visco-elastic longitudinal waves propagating in bars

by the characteristics method
In the following, we discuss the characteristics method used to solve the linear visco-
elastic longitudinal waves propagating in linear visco-elastic bars. It can be regarded as
a development of the characteristics method that we have discussed for rate-independent
elastic and elastic–plastic waves; and, on the other hand, it can be regarded as a starting
step for the further studies on the rate-dependent elastic-visco-plastic waves in Section 6.6.
For convenience, we take the bar made of Maxwell body as a typical example.
Multiplying each equation of the governing equations [Eq. (6.26)], respectively, by the
indeterminate coefficients L, M, and N , and then adding them, the characteristic equa-
tions and the corresponding characteristic compatibility relations are obtained (herein the
subscript M for the Maxwell model is temporarily neglected):

∂ε ∂ ∂ N ∂ ∂ σ
(L + N ) + Mρ0 − L v+ − −M σ −N =0 (6.37)
∂t ∂t ∂X E ∂t ∂X η
For the above equation to contain only the directional derivatives along the characteristics
line, these indeterminate coefficients should satisfy the following conditions:
dX 0 −L ME
= = = (6.38)
dt L+N Mρ0 N
of which the solutions are:
L + N = 0, ρ0 EM 2 = −LN (6.39a)
L = M = 0, N
= 0 (6.39b)
Substituting Eq. (6.39a) into Eqs. (6.38) and (6.37), we obtain two families of real
characteristic line and the corresponding characteristic compatibility relationships:

E
dX = ± dt = ±C0 dt (6.40a)
ρ0
1 C0 1 C0
dv = ± dσ ± σ dt = ± dσ ± dX (6.40b)
ρ0 C0 η ρ0 C0 η
which represent the loci of wave fronts propagating with the wave velocity C0 determined
by the instantaneous response of material.
Substituting Eq. (6.39b) into Eqs. (6.38) and (6.37), we obtain the third set of real
characteristic line and the corresponding characteristic compatibility relationship:
dX = 0 (6.41a)
dσ σ dt
dε − − =0 (6.41b)
E η
which represent the loci of particles movement, and the characteristic compatibility relation
[Eq. (6.41b)] is actually another expression of the material derivative of the constitutive
relation of material [Eq. (6.4)] along the particle motion locus.
Different from the situation for elastic waves, note that the terms having dt or dX appear in
the compatibility conditions, Eqs. (6.40b) and (6.41b). They are the terms that describe the
rate (time) dependence of visco-elastic wave propagation and reflect the dispersion and
dissipation characters of visco-elastic waves. Although the calculation of compatibility
relations along characteristics becomes more complicated due to the existence of these
terms, the corresponding three families of characteristics are all in the form of straight
lines, which are very convenient for numerical calculation.
So, we use the characteristics method to solve the problem of visco-elastic wave prop-
agation. Through an arbitrary point Ni on the X–t plane, there are three characteristic
lines as shown in Fig. 6.9 and three corresponding characteristic relations. If we write
those relations in the difference form instead of the differential form, the three unknown
variables σ, v, and ε at the point Ni can be determined according to the known initial and
boundary conditions (by using the algebra equations). The specific procedures are similar
to what discussed previously. We will illustrate the details by an example, the visco-elastic
wave propagation in a semi-infinite Maxwell bar, as follows.
t t
N3
N1 N2 M3
M2
M1
o
s s* X
Fig. 6.9. Three families of characteristics for the linear visco-elastic longitudinal waves propagating
in bars.
Consider a semi-infinite Maxwell bar, which is originally at rest and in a stress-free state
(v0 = σ0 = ε0 = 0), and suddenly suffers a constant stress σ ∗ at the bar end (X = 0).
Then a strong discontinuous wave propagates with wave velocity D along the line OA,
as shown in Fig. 6.9. According to the kinematic compatibility condition (Eq. 2.55) and
the dynamic compatibility condition (Eq. 2.57) across a strong discontinuous interface
deduced in Chapter 2, we have, respectively,
[v] = −D [ε]
[σ ] = −ρ0 D [v]
Since both are independent of the material characters, they are also tenable for the strong
discontinuous visco-elastic waves. Moreover, since ε̇ → ∞ across a strong discontinuous
interface, only the instantaneous response part of the constitutive relation takes effect.
Thus, for Maxwell body, we have [σ ]/[ε] = E, so the strong discontinuous wave velocity
D in Maxwell bar is

1 [σ ] E
D= = = C0
ρ0 [ε] ρ0
which equals the elastic wave velocity in bars.
Thus, along the line OA, according to the dynamic compatibility condition and the
kinematic compatibility condition, respectively, we have
σ
v=− (6.42a)
ρ0 C0
v = −C0 ε (6.42b)
On the other hand, since the line OA is a characteristic line, according to the corresponding
characteristic compatibility condition (Eq. 6.40b), we have
dσ σ
dv = + dX
ρ0 C0 η
Substituting Eq. (6.42a) into the above equation, it is reduced to a first-order ordinary
differential equation with respect to σ
dσ ρ 0 C0
+ σ =0
dX 2η
Determining the integration constant by using the boundary condition σ |X=0 = σ ∗ =

−ρ0 C0 v∗ , we obtain the solution along the line OA:

ρ0 C0
σ = σ ∗ exp − X (6.43a)
2η
and from Eq. (6.42) we obtain

σ∗ ρ 0 C0 ρ 0 C0
v=− exp − X = v∗ exp − X (6.43b)
ρ 0 C0 2η 2η
∗
σ ρ 0 C0
ε= exp − X (6.43c)
E 2η
It shows that the amplitude of a strong discontinuous wave propagating in Maxwell bar
attenuates exponentially (an absorption phenomenon), and the corresponding attenuation
factor α is
ρ0 C0
α=
2η
which coincides with the result for high-frequency waves expressed by Eq. (6.29).
Since the solutions along the line OA have been given by Eq. (6.43), the remaining solutions
in the region AOt shown in Fig. 6.9 can be further determined. In fact, the σ along another
characteristic line Ot are given by the boundary condition at the bar end, so the problem
in the region AOt is attributed to solve a characteristics boundary-value problem for
visco-elastic waves. However, it should be noticed that the region AOt is no longer a
constant wave region as in the situation for elastic waves, but a region wherein a series of
weak discontinuous visco-elastic waves propagate.
Let us now first discuss the solution at an arbitrary boundary point along Ot, e.g. the
point N1 in Fig. 6.9, where the strain ε(N1 ) and the particle velocity v(N1 ) are unknown,
while the stress σ (N1 ) has been given by the boundary condition. From Eqs. (6.41a) and
(6.40a), it is known that two characteristics ON 1 and M1 N1 can always be found to link
the unknown boundary point N1 with two known points O and M1 . Then, we have two
compatibility conditions Eqs. (6.41b) and (6.40b) along ON1 and MN1 , which, in the form
of finite differences, are
1 σ (M1 )
v(N1 ) − v(M1 ) = − [σ (N1 ) − σ (M1 )] + [X(N1 ) − X(M1 )] (6.44a)
ρ 0 C0 η
1 σ (0)
ε(N1 ) − ε(0) − [σ (N1 ) − σ (0)] − [t (N1 ) − t (0)] = 0 (6.44b)
E η
by which the particle velocity v(N1 ) and strain ε(N1 ) at the point N1 can be found.
Similarly, the unknown stress σ (N2 ), particle velocity v(N2 ), and strain ε(N2 ) at an
arbitrary interior point N2 can be solved by the following three characteristic compati-
bility conditions along the corresponding characteristic lines, N1 N2 , M2 N2 and M1 N2 ,
respectively:
1 σ (N1 )
v(N2 ) − v(N1 ) = [σ (N2 ) − σ (N1 )] + [X(N2 ) − X(N1 )] (6.45a)
ρ0 C0 η
1 σ (M2 )
v(N2 ) − v(M2 ) = − [σ (N2 ) − σ (M2 )] + [X(N2 ) − X(M2 )] (6.45b)
ρ0 C0 η
1 σ (M1 )
ε(N2 ) − ε(M1 ) − [σ (N2 ) − σ (M1 )] − [t (N2 ) − t (M1 )] = 0 (6.45c)
E η
On the analogy of this, the solutions in the whole region AOt can be obtained, so that the
problem of visco-elastic waves propagating in a semi-infinite bar is solvable.
6.2.5 Split Hopkinson visco-elastic bar
In the preceding subsection, it was shown that, on the prerequisite that both the visco-
elastic constitutive equation of material and the initial boundary conditions are known, the
problem of visco-elastic wave propagation is solvable by using the characteristics method.
This kind of problem is the so-called “direct problem”. In the current subsection, taking
the “split Hopkinson visco-elastic bar” as an example, we will discuss how the boundary
condition can be determined from the experimentally measured visco-elastic wave profiles
when the visco-elastic constitutive equation of material and the initial condition are given.
This kind of problem belongs to the category of the so-called “first kind of inverse
problem”. In Section 6.4, we will discuss the so-called “second kind of inverse problem”,
namely, how the visco-elastic constitutive equation of material can be determined from
the experimentally measured visco-elastic wave profiles when the initial and boundary
conditions are given.
The classical split Hopkinson pressure bar (SHPB) technique is based on the principle of
one-dimensional linear elastic wave propagating in bars, as has been briefly discussed in
Section 3.7. Traditionally, the pressure bars used are made of high-strength steel with a
high elastic limit σs > 1 GPa and a high acoustic impedance ρ0 C0 around 40 MPa s/m (see
Table 2.1), so that the specimen of test material that generally has lower strength and lower
wave impedance will deform nonelastically, while the bars are still in the state of elastic
deformation. However, if the wave impedance of specimen is as low as 0.1∼1 MPa/m/s,
much lower than that of bars, the signal of transmitted pulse will become too weak to
be detected accurately. In such a case, it seems that a bar material with lower acoustic
impedance, instead of high-strength steel, should be used. One preferable way is to use
plastic (polymer) bars, such as a polymethyl methacrylate (PMMA) bar, but the dispersion
and dissipation characters of visco-elastic wave propagation in polymer bars should be
taken into account (Wang et al., 1994).
Now, let us first analyze what the differences are between the split Hopkinson elastic bar
and the split Hopkinson visco-elastic bar. In other words, what new problems will happen
when the SHPB technique is generalized to use visco-elastic polymer bars.
As discussed in Section 3.7, the key of SHPB technique is how to determine the stress
σ (X1 , t) and the particle velocity v(X1 , t) at the interface X1 from the signals of incident
wave σi (Xg1 , t) and the reflected wave signal σr (Xg1 , t) measured at Xg1 , where the X1
is the interface of the specimen with the input bar and the Xg1 is the position of strain
gauge G1 on the input bar, as shown in Fig. 6.10; and also the stress σ (X2 , t) and the
particle velocity v(X2 , t) at the interface X2 of the specimen with the output bar from
the transmitted wave signal σt (Xg2 , t) measured at Xg2 , the position of strain gauge G2
on the input bar. In fact, once σ (X1 , t), v(X1 , t), σ (X2 , t), and v(X2 , t) are obtained, the
stress σs (t), strain rate ε̇s (t), and strain εs (t) for the specimen can then be determined
according to Eq. (3.16), respectively, as shown below again:
A A
σs (t) = [σ (X1 , t) + σ (X2 , t)] = [σi (X1 , t) + σr (X1 , t) + σt (X2 , t)]
2As 2As
v(X2 , t) − v(X1 , t) vt (X2 , t) − vi (X1 , t) − vr (X1 , t)
ε̇s (t) = =
ls ls
t t
1
εs (t) = ε̇s (t) dt = [vt (X2 , t) − vi (X1 , t) − vr (X1 , t)] dt
0 ls 0
Recall that in the case of elastic bars, according to the linear propositional relations
between the stress, strain, and particle velocity for elastic waves, we have
⎫
σ1 = σ (X1 , t) = σi (X1 , t) + σr (X1 , t) = E[εi (X1 , t) + εr (X1 , t)]⎪
⎪
⎪
σ = σ (X , t) = σ (X , t) = Eε (X , t) ⎬
2 2 t 2 r 2
v1 = v(X1 , t) = vi (X1 , t) + vr (X1 , t) = C0 [εi (X1 , t) − εr (X1 , t)] ⎪
⎪
⎪
⎭
v2 = v(X2 , t) = vt (X2 , t) = C0 εt (X2 , t)
Fig. 6.10. Layout of input bar, specimen and output bar

then the dynamic stress σs (t) and strain εs (t) of the specimen can be determined only if the
incident strain wave εi (X1 , t) and the reflected strain wave εr (X1 , t) at the interface X1
as well as the transmitted strain wave εt (X2 , t) at the interface X2 are known. However,
due to the nondistortion character of elastic wave propagating in bar, the incident strain
wave εi (X1 , t) and the reflected strain wave εr (X1 , t) at the interface X1 can be replaced
by the incident wave signal εi (XG1 , t), and the reflected wave signal εr (XG1 , t) can be
measured by the strain gauge G1 at the position XG1 on the input bar. Also, the transmitted
strain wave εt (X2 , t) at the interface X2 can be replaced by the transmitted wave signal
εt (XG2 , t) can be measured by the strain gauge G2 at the position XG2 on the output bar.
Thus, from the wave signals measured by the strain gauge G1 at XG1 and the strain gauge
G2 at XG2 , respectively, the dynamic stress σs (t) and strain εs (t) for the specimen can be
finally determined:
EA EA
σs (t) = εt (XG2 , t) = [εi (XG1 , t) + εr (XG1 , t)]
As As

2C0 t 2C0 t
εs (t) = − εr (XG1 , t) dt = [εi (XG1 , t) − εt (XG2 , t)] dt
ls 0 ls 0
However, in the case of visco-elastic bars, due to the dispersion and attenuation char-
acters of visco-elastic waves, the linear propositional relations between the stress, strain,
and particle velocity and, consequently, the nondistortion character no longer exist for
visco-elastic waves. Thus, the keys of the current problem are attributed to: (1) how
to determine the incident stress σi (X1 , t) and incident particle velocity vi (X1 , t) at the
interface X1 from the incident strain wave signal εi (XG1 , t) measured by the strain
gauge G1 at XG1 ; (2) how to determine the reflected stress σr (X1 , t) and reflected
particle velocity vr (X1 , t) at the interface X1 from the reflected strain wave signal
εr (XG1 , t) measured by the strain gauge G1 at XG1 ; (3) how to determine the trans-
mitted stress σt (X1 , t) and transmitted particle velocity vt (X1 , t) at the interface X2
from the transmitted strain wave signal εt (XG2 , t) measured by the strain gauge G2 at
XG2 ; and then finally to determine the dynamic stress σs (t) and strain εs (t) for speci-
men according to Eq. (3.16). Among these, the first one is essentially attributed to solve
the direct problem of visco-elastic wave propagation, similar to what we have discussed
in Subsection 6.2.4; the other two are attributed to solve the so-called second kind of
inverse problem of visco-elastic wave propagation, which will be discussed in detail in
the following.
For convenience, let us discuss the SHPB bar made of PMMA as an example. Since defor-
mation of the pressure bar itself is usually small, the problem can be simplified as a linear
visco-elastic wave propagation analysis. In such a situation, the constitutive relationship
of PMMA bar at high strain rates can be satisfactorily described by the following equation
in the differential form (Wang et al., 1994), which corresponds to the standard linear solid
model, as shown in Fig. 6.8(a):
∂ε 1 ∂σ Ea ε σ
− + − =0 (6.46)
∂t Ea + EM ∂t (Ea + EM )θM (Ea + EM )θM
which, together with the following movement equation and continuous equation,
∂υ ∂σ
ρ0 − =0
∂t ∂X
∂ε ∂υ
− =0
∂t ∂X
constitute the governing equations of the current problem. Similar to what we discussed
in Subsection 6.2.4 for the visco-elastic waves propagating in Maxwell bar by using the
characteristics method, multiplying each of them by undetermined coefficients N , M, and
L, respectively, and then adding together, we have

∂ε ∂ ∂ N ∂ ∂
(L + N ) + Mρ0 − L v− +M σ
∂t ∂t ∂X Ea + EM ∂t ∂X
N
+ (Ea ε − σ ) = 0
(Ea + EM )θM
In order to satisfy the requirement that the above equation consists of only the directional
differential along characteristics C(X, t), the undetermined coefficient, L, M, and N
should satisfy the following relations:

dX 0 L M(Ea + EM )
= =− =
dt c L+N Mρ0 N
Obviously, two sets of solutions for L,M, and N exist. One is determined by the following
equations
L+N =0
ρ0 (Ea + EM )M 2 = −LN (6.47)
Then we obtain two families of characteristics:

dx Ea + E M
=± = Cv (6.48a)
dt ρ0
and, correspondingly, two sets of compatibility conditions along the characteristics

1 σ − Ea ε 1 σ − Ea ε
dυ = ± dσ ± dt = ± dσ + dX (6.48b)
ρ0 C v ρ 0 C v θM ρ 0 Cv (Ea + EM )θM
where the plus and minus signs are for rightward-propagating and leftward-propagating
waves, respectively. Note that the propagation velocity along characteristics is the same
as the phase velocity of high-frequency waves CV [Eq. (6.34)].
Another set of solutions is determined by:
L=M=0
N
= 0 (6.49)
Then, we obtain the third family of characteristics and the corresponding compatibility
condition along it, respectively, as
dX = 0 (6.50a)
dσ σ − Ea ε
dε − − dt = 0 (6.50b)
Ea + E M (Ea + EM )θM
Obviously, Eq. (6.50a) coincides with the particle motion loci, and Eq. (6.50b) is, in fact,
a special form of visco-elastic constitutive equation, Eq. (6.46), along the particle motion
locus.
Thus, we use the characteristics method to solve the problem; similar to what we have
discussed in Section 6.2.4 for the semi-infinite Maxwell bar, through an arbitrary point Ni
on the X-t plane, there are three characteristic lines as shown in Fig. 6.9 and three corre-
sponding characteristic relations. These relations are written in the difference form instead
of the differential form, by which the three unknown variables σ, v, and ε at the point Ni
can be determined according to the known initial and boundary conditions. The specific
procedures are similar to those discussed previously for the Maxwell bar, only noting that
the characteristics propagation velocity for the standard linear solid bar CV should be used
instead of the characteristics propagation velocity for the Maxwell bar C0 .
Now, let us consider a semi-infinite standard linear solid bar, which is originally at rest
and in a stress-free state:
σ (X, 0) = ε(X, 0) = v(X, 0) = 0
and at the bar end (X = 0) suddenly suffered a constant stress σ ∗ . Then, as shown in
Fig. 6.9, a strong discontinuous wave propagates along the line OA with wave velocity D :

1 [σ ] Ea + E M
D= = (6.51)
ρ0 [ε] ρ0
According to the kinematic [Eq. (2.55)] and dynamic [Eq. (2.57)] compatibility condi-
tions across a strong discontinuous interface, as deduced in Chapter 2, and, moreover,
considering that the line OA is also a characteristics so that the characteristic compati-
bility condition, Eq. (6.48b), should be satisfied, then, similar to Eq. (6.43), it could be
easily shown that this strong discontinuous wave attenuates according to the following
exponential:
⎡ ⎤
⎢ ⎥
⎢ ρ0 CV ⎥
σ = σ ∗ exp ⎢−
∗
2 X ⎥ = σ exp(−αa X) (6.52)
⎣ Ea ⎦
2ηM 1 +
EM
Completely similar results can be obtained for the particle velocity v and strain ε along
the line OA. Once the solutions along the line OA are obtained, the remaining solutions
in the region AOt in Fig. 6.9 are determinable, since it is attributed to solve a solvable
characteristics boundary-value problem for visco-elastic waves. This mainly includes
two kinds of basic operations: (1) to determine the solutions at boundary points, and
(2) to determine the solutions at interior points.
For example, at an arbitrary boundary point N1 in Fig. 6.9, the particle velocity v(N1 ) and
strain ε(N1 ) can be determined from two characteristic compatibility conditions along the
characteristics M1 N1 and ON1 , respectively, namely, Eqs. (6.48) and (6.50). Writing in
the finite difference form, we have
1 Ea ε(M1 ) − σ (M1 )
v(N1 ) − v(M1 ) = − [σ (N1 ) − σ (M1 )] + [t (N1 ) − t (M1 )]
ρ0 CV ρ 0 C V θM
1 Ea ε(0) − σ (0)
ε(N1 ) − ε(0) − [σ (N1 ) − σ (0)] + [t (N1 ) − t (0)] = 0
(6.53)
While at an arbitrary interior point N2 in Fig. 6.9, the particle velocity v(N2 ), stress σ (N2 ),
and strain ε(N2 ) can be determined from three characteristic compatibility conditions along
the characteristics N1 N2 , M1 N2 , and M2 N2 , respectively. Writing in the finite difference
form, we have
1 Ea ε(N1 ) − σ (N1 )
v(N2 ) − v(N1 ) = [σ (N2 ) − σ (N1 )] − [t (N2 ) − t (N1 )]
ρ 0 CV ρ0 CV θ M
1 Ea ε(M2 ) − σ (M2 )
v(N2 ) − v(M2 ) = − [σ (N2 ) − σ (M2 )] + [t (N2 ) − t (M2 )]
ρ0 CV ρ 0 C V θM
1 Ea ε(M1 ) − σ (M1 )
ε(N2 ) − ε(M1 ) = [σ (N2 ) − σ (M1 )] − [t (N2 ) − t (M1 )]
(6.54)
Although the examples of Eqs. (6.53) and (6.54) that we deduced above are used to solve
the so-called direct problem, i.e. to find the solution of visco-elastic wave propagation from
the given initial and boundary conditions, the so-called inverse problem can be solved in
a similar way without difficulty, i.e. to find the boundary condition from the known results
of wave propagation and the given initial condition. For example, as shown in Fig. 6.9, if
the stress, strain, and particle velocities at the points M2 and N3 have been known and if
the stress, strain, and particle velocity at point M1 have been given by the initial condition,
then the stress, strain, and particle velocity at the point N2 can be solved from the following
three characteristic compatibility conditions along the characteristics N3 N2 , M2 N2 , and
M1 N2 , respectively:
1 Ea ε(N3 ) − σ (N3 )
v(N2 ) − v(N3 ) = [σ (N2 ) − σ (N3 )] − [t (N3 ) − t (N2 )]
ρ0 CV ρ 0 C V θM
1 Ea ε(M2 ) − σ (M2 )
v(N2 ) − v(M2 ) = − [σ (N2 ) − σ (M2 )] + [t (N2 ) − t (M2 )]
ρ0 CV ρ0 CV θ M
1 Ea ε(M1 ) − σ (M1 )
ε(N2 ) − ε(M1 ) = [σ (N2 ) − σ (M1 )] − [t (N2 ) − t (M1 )]
(6.55)
From the above results, we can now discuss, in the case of the split Hopkinson visco-
elastic bar, how to determine the dynamic stress–strain relationship of a specimen from the
measured incident strain wave signals, the reflected strain wave signals, and the transmitted
strain wave signals.
As mentioned previously, the specimens in SHPB tests are generally short enough so that
the basic assumption of “uniform distribution of stresses along the specimen length” is
satisfied, namely, σ (X2 , t) = σ (X1 , t). Thus, any two measurements of the incident,
reflected, and transmitted waves are sufficient to determine the dynamic stress–strain
relationship of the specimen. In the case of visco-elastic bars, considering that the reflected
wave is generally more difficult to separate from the dispersive tail-part of the incident
visco-elastic wave, the incident signal εi (Xg1 , t) and the transmitted signal ετ (Xg2 , t) are
usually used in practice. Then, the dynamic stress–strain relationship of the specimen can
be obtained by the following four successive subprocedures:
1. To determine the unknown incident stress σi (X1 , t) and particle velocity vi (X1 , t)
at the incident interface X1 from the incident strain signal ετ (Xg1 , t) measured by
the strain gauge G1 at Xg1 .
This subprocedure is attributed to solve a direct problem, i.e. to find the solution
of visco-elastic wave propagation (in the positive X direction) for a given strain
boundary condition and initial condition. Thus, the unknown stress and particle
velocity at a boundary point (Xi , tj ), as shown in Fig. 6.11(a), can be determined
by Eqs. (6.53). Or, in an explicit form, we have:
σ (Xi , tj ) = σ (Xi , tj −2 ) + (Ea + EM )[ε(Xi , tj ) − ε(Xi , tj −2 )]

2∆t
+ [Ea ε(Xi , tj −2 ) − σ (Xi , tj −2 )]
θM
[σ (Xi+1 , tj −1 ) − σ (Xi , tj )]
v(Xi , tj ) = v(Xi+1 , tj −1 ) +
ρ0 CV
[Ea ε(Xi+1 , tj −1 ) − σ (Xi+1 , tj −1 )]∆t
+ (6.56)
ρ0 CV θM
(xi, tj) (xi, tj)
(xi+1, tj−1) (xj−1, tj−1) (xi+1, tj−1)
(xi, tj−2) (xi, tj−2)

a. Boundary point for rightward waves b. Interior point in a direct problem
(xi+1, tj+1) (xi, tj)

(xi, tj) (xj−1, tj−1)
(xi+1, tj−1)
(xi, tj−2)
(xi, tj−2)
c. Interior point in an inverse problem d. Boundary point for leftward waves
Fig. 6.11. The characteristics solutions for different cases in polymer SHPB.
where ∆t = tj − tj −1 is the time difference, i.e. the time step length in numerical
calculation. On the other hand, the unknown stress, strain, and particle velocity at
an interior point (Xi , tj ), as shown in Fig. 6.11(b), can be determined by Eqs. (6.54).
Or, in an explicit form, we have:

1
σ (Xi ,tj ) = σ (Xi+1 ,tj −1 )+σ (Xi−1 ,tj −1 )+ρ0 CV [v(Xi−1 ,tj −1 )
2
−v(Xi−1 ,tj −1 )]+[Ea ε(Xi+1 ,tj −1 )−σ (Xi+1 ,tj −1 )

∆t
+Ea ε(Xi−1 ,tj −1 )−σ (Xi−1 ,tj −1 )]
θM
⎧
⎪
⎪
1 ⎨ σ (Xi+1 ,tj −1 )−σ (Xi−1 ,tj −1 )
v(Xi ,tj ) = +v(Xi+1 ,tj −1 )+v(Xi−1 ,tj −1 )
2⎪⎪ ρ0 CV
⎩
⎛ ⎞⎫
[Ea ε(Xi+1 ,tj −1 )−σ (Xi+1 ,tj −1 )] ⎪
⎪
⎜ −[Ea ε(Xi−1 ,tj −1 )−σ (Xi−1 ,tj −1 )] ∆t ⎟⎬
⎜
+⎝ ⎟
ρ 0 CV θM ⎠⎪⎪
⎭
σ (Xi ,tj )−σ (Xi ,tj −2 )

ε(Xi ,tj ) = ε(Xi ,tj −2 )+
Ea +EM
Ea ε(Xi ,tj −2 )−σ (Xi ,tj −2 ) 2∆t
− (6.57)
Ea +EM θM
2. To determine the unknown transmitted stress σt (Xg2 , t) and particle velocity

vt (Xg2 , t) at the strain gauge G2 from the measured εt (Xg2 , t)
This subprocedure is also attributed to solve a direct problem for a given strain-
boundary condition, similar to that shown in Fig. 6.11(b). Thus, Eqs. (6.56) and
(6.57) can be used again in this subprocedure. Note that this subprocedure is a
necessary preliminary step for the next subprocedure.
3. To determine the unknown transmitted stress σt (X2 , t), strain εt (X2 , t), and particle
velocity vt (X2 , t) at the transmitted interface of specimen, X2 , from the known
σt (Xg2 , t), εt (Xg2 , t), and vt (Xg2 , t) at G2 .
This subprocedure is attributed to solve an inverse problem, i.e. to find the unknown
boundary condition from the given initial condition and the known results of visco-
elastic wave propagation at a given point. Thus, Eq. (6.55) should be used to
determine the stress, strain, and particle velocity at an interior point (Xi , tj ) as
shown in Fig. 6.11(c). Or, in an explicit form, we have

1
σ (Xi ,tj ) = σ (Xi+1 ,tj −1 )+σ (Xi+1 ,tj +1 )+ρ0 CV [v(Xi+1 ,tj −1 )
2
−v(Xi+1 ,tj +1 )]+[Ea ε(Xi+1 ,tj −1 )−σ (Xi+1 ,tj −1 )

∆t
+Ea ε(Xi+1 ,tj +1 )+σ (Xi+1 ,tj +1 )]
θM
⎧
⎪
⎪
1 ⎨ σ (Xi+1 ,tj −1 )−σ (Xi+1 ,tj +1 )
v(Xi ,tj ) = +v(Xi+1 ,tj −1 )+v(Xi+1 ,tj +1 )
2⎪⎪ ρ 0 CV
⎩
⎛ ⎞⎫
[Ea ε(Xi+1 ,tj −1 )−σ (Xi+1 ,tj −1 )] ⎪
⎪
⎜ +[Ea ε(Xi+1 ,tj +1 )−σ (Xi+1 ,tj +1 )] ∆t ⎟⎬
+⎜
⎝
⎟
ρ0 CV θM ⎠⎪⎪
⎭
σ (Xi ,tj )−σ (Xi ,tj −2 )

ε(Xi ,tj ) = ε(Xi ,tj −2 )+
Ea +EM
Ea ε(Xi ,tj −2 )−σ (Xi ,tj −2 ) 2∆t
− (6.58)
Ea +EM θM
Now, σi (X1 , t), εi (X1 , t), and vi (X1 , t) at the incident interface and σt (X2 , t),
εt (X2 , t), and vt (X2 , t) at the transmitted interface of the specimen have all been
determined. Then, by the assumption of “uniform distribution of stress along the
specimen length” (σi + σr = σt ), the reflected stress σr (X1 , t) can be immediately
determined:
σr (X1 , t) = σt (X2 , t) − σi (X1 , t) (6.59)

The rest of quantities of reflected wave at the interface X1 are determined by the
fourth subprocedure, as follows.
4. To determine the reflected particle velocity vr (X1 , t) and strain εr (X1 , t) at the
incident interface of specimen, X1 , from the known σr (X1 , t).
This subprocedure is attributed to solve the direct problem of a visco-elastic wave
propagating in negative X direction for a given stress-boundary condition. For the
boundary points, the equations similar to Eq. (6.53) can be used, except that, due
to the change of wave propagation direction, the corresponding change of sign
in compatibility conditions should be taken into account. Thus, for an arbitrary
boundary point (Xi , tj ), as shown in Fig. 6.11(d), we have
σ (Xi ,tj )−σ (Xi ,tj −2 )

ε(Xi ,tj ) = ε(Xi ,tj −2 )+
Ea +EM
Ea ε(Xi ,tj −2 )−σ (Xi ,tj −2 ) 2∆t
−
Ea +EM θM
[σ (Xi−1 ,tj −1 )−σ (Xi ,tj )]
v(Xi ,tj ) = v(Xi−1 ,tj −1 )+
ρ0 CV
[Ea ε(Xi−1 ,tj −1 )−σ (Xi−1 ,tj −1 )]∆t
+ (6.60)
ρ0 C V θ M
However, for the interior points as shown in Fig. 6.11(b), Eq. (6.54) can still be used,
since they are derived from the compatibility conditions along both rightward and
leftward characteristics and, consequently, are independent of the wave propagation
direction.
Thus, we have obtained all the incident, reflected, and transmitted stresses, strains, and
particle velocities at two interfaces of the specimen. Then, as discussed at the beginning
of this subsection, by Eq. (3.16) in Chapter 3, the dynamic stress σs (t), strain-rate ε̇s (t),
and strain εs (t) of the specimen, and, consequently, the dynamic stress–strain relation of
specimen at a high strain rate can finally be determined.
6.3 Nonlinear Visco-Elastic Constitutive Relationship
The linear visco-elastic constitutive relationship discussed in the previous section

is no longer appropriate to describe the more complicated nonlinear behavior of
visco-elastic materials such as polymers under large deformation. Thus, non-
linear visco-elastic consti-tutive relationships are required and the corresponding
nonlinear visco-elastic wave propagation theories are developed.
The experimental investigation by Zhu, Wang, and their coworkers in the recent 20 years
for a variety of polymers, including polymethylmethacrylate (PMMA), polycarbonate
(PC), polyamide (PA or nylon), acrylonitrile-butadiene-styrene (ABS), polybuthylene
terephthalate (PBT), epoxy, and phenolics thermoset plastics, at strain rates from 10−4
to 103 s−1 showed that the mechanical behavior of polymers is highly sensitive to
strain rates. The nonlinear visco-elastic behavior in uniaxial stress state for those polymers
studied can be well described by the following Zhu-Wang-Tang (ZWT) nonlinear visco-
elastic constitutive equation (Tang et al., 1981; Wang and Yang, 1992), which in the
differential form is:

∂σ dfe (ε) ∂ε ∂σ1 ∂σ2 dfe (ε) ∂ε σ1 σ2
= + + = + E 1 + E2 − − (6.61a)
∂t dε ∂t ∂t ∂t dε ∂t θ1 θ2
and in the integral form is

t
t t −τ t −τ
σ = fe (ε) + E1 ε̇ exp − dτ + E2 ε̇ exp − dτ (6.61b)
0 θ1 0 θ2
where the first term
fe (ε) = E0 ε + αε 2 + βε 3 (6.61c)
describes the nonlinear elastic equilibrium response, and E0 , α, and β herein are the corre-
sponding elastic constants; the next integral term in Eq. (6.61b) describes the viscoelastic
response in low strain rates, and E1 , and θ1 are the elastic constant and relaxation time,
respectively, of the corresponding Maxwell element I; and the last integral term describes
the viscoelastic response in high strain rates, and E2 and θ2 are the elastic constant and
relaxation time, respectively, of the corresponding Maxwell element II. The corresponding
rheological model is shown in Fig. 6.12, consisting of a nonlinear spring (non-Hookean)
and two Maxwell elements in parallel connection.
For some typical engineering plastics such as epoxy, PMMA, and PC, the material
parameters in ZWT equation obtained from the experiments are listed in Table 6.1.
E1 E2
E0 , a, b
h1 = E1q1 h2 = E2q2
Fig. 6.12. The ZWT nonlinear visco-elastic constitutive model.

Table 6.1. Material parameters in ZWT equation for epoxy, PMMA, and PC.
Epoxy PMMA-1 PMMA-2 PMMA-3 PC
ρ0 , kg/m3 1.20 × 103 1.19× 103 1.19 × 103 × 103
1.19 1.20 × 103
E0 , GPa 1.96 2.04 2.19 2.95 2.20
α, GPa 4.12 4.17 4.55 10.9 23
β, GPa −181 −233 −199 −96.4 −52
E1 , GPa 1.47 0.897 0.949 0.832 0.10
θ1 , s 157 15.3 13.8 7.33 470
E2 , GPa 3.43 3.07 3.98 5.24 0.73
θ2 , µs 8.57 95.4 67.4 40.5 140
Note that to avoid any unreal “elastic strain softening” that may be spuriously introduced by
Eq. (6.61c), the nonlinear elastic equilibrium response fe (ε) can be equivalently described
by the following exponential form (Zhou et al., 1992):
/ % n &0
. (mε)i
fe (ε) = σm 1 − exp − (6.62)
i
i=1
where σm denotes the asymptotic maximum, m the ratio of E0 and σm , and the positive
integer n is a material parameter characterizing the initial linearity.
With regard to the relaxation time and in the ZWT equation, which describe the viscous
character of the polymer, it is worthwhile to notice that each corresponds to an effective
strain rate range or the so-called “effective influence domain”. In fact (Chu et al., 1985),
for an arbitrary Maxwell element with elastic coefficient Ej and viscous coefficient ηj ,
according to Eq. (6.4), we have:
σ̇j σj
ε̇ = +
Ej ηj
At constant strain rate (ε̇ = const., ε = ε̇t), the stress relaxation response σj is given as

ε
σj = Ej θj ε̇ 1 −exp − (6.63)
θj ε̇
Obviously, σj approaches its maximum, i.e. the instantaneous response σI = Ej ε, when

ε̇ approaches infinity, and its minimum, i.e. the equilibrium response σE = 0, when ε̇
approaches zero. Introducing the nondimensional stress relaxation response defined as
σj σj σj
σ̄j = = =
σmax − σmin σI − σ E Ej ε
Eq. (6.63) can be rewritten as:

θj ε̇ ε θj t
σ̄j = 1 −exp − = 1 − exp −
ε θj ε̇ t θj
If we regard σ̄j = 0.995 as the beginning of a relaxation process, and σ̄j = 0.005 as
the end of this relaxation process, then the effective influence domain (EID) of Maxwell
element with relaxation time θj can be easily determined from the above equation. When
it is in terms of time t, we have
t
10−2 ≤ ≤ 102.3 (6.64a)
θj
and when it is in terms of strain rate, assuming ε = 1 we have
ε̇
102 ≥ ≥ 10−2.3 (6.64b)
ε
θj
It means that the EID for any relaxation time θj is about 4.3 orders of magnitude in either
timescale or strain-rate scale.
Taking into account the above analyses and the experimental results shown in Table 6.1, it
is worthwhile to notice the following features of ZWT nonlinear visco-elastic constitutive
equation:
1. The constitutive nonlinearity of the tested polymers only comes from the pure
elastic equilibrium response fe (ε), while all the rate/time-dependent responses are
essentially linear. Such a constitutive nonlinearity is a kind of “rate-independent
nonlinearity”, or may be termed weak nonlinearity. Thus, if we name such a
kind of material as ZWT material, it provides a possibility of generalizing what
has been established in the linear viscoelasticity to the rate-dependent response of
ZWT materials without substantial difficulty.
2. The experimental results for typical engineering plastics show that the ratio of α/E0
is on the order of 1 to 10, and the ratio of β/E0 is on the order of 10 to 102 , as
shown in Table 6.1. This means that the nonlinearity should be taken into account
when ε > 1%, while it may be neglected if ε < 1%, depending on the specific
material.
3. The experimental results for typical engineering plastics also show that the relax-
ation time θ1 is generally on the order of 10 to 102 s, while the relaxation time θ2
is on the order of 10−4 to 10−6 s, as shown in Table 6.1. It means that θ1 is respon-
sible for the visco-elastic behavior at low strain rates, while θ2 is responsible for
the visco-elastic behavior at high strain rates. Since θ1 is on the order of 105 –108
times higher than θ2 , and each relaxation time θ has its effective influence domain
of about 4.3 orders of magnitude, thus θ1 and θ2 individually play their respective
roles in each effective influence domain.
4. Consequently, under quasi-static loading conditions, where the timescale is on the
order of 100 to 102 s, the high-frequency Maxwell element II (with relaxation time
of 100 to 102 µs) has already relaxed even at the beginning of loading. The ZWT
equation [Eq. (6.61)] is then reduced to

t t −τ
σ = fe (ε) + E1 ε̇(t) exp − dτ (6.65a)
0 θ1
5. On the contrary, under impact-loading conditions, where the timescale is on the

order of 1 to 102 µs, the low-frequency Maxwell element I (with relaxation time θ1
of 10−1 to 1 s) does not have enough time to relax even until the end of loading.
Then the Maxwell element I is reduced to a single-spring element with an elastic
constant E1 , and, correspondingly, the ZWT equation [Eq. (6.61)] is reduced to

t t −τ
σ = fe (ε) + E1 ε + E2 ε̇(t) exp − dτ
0 θ2
t
t −τ
= σe (ε) + E2 ε̇(t) exp − dτ (6.65b)
0 θ2
σe (ε) = fe (ε) + E1 ε
It means that the nonlinear visco-elastic wave propagation in ZWT polymer bars is
actually controlled by Eq. (6.65b).
Theoretically, ZWT equation [Eq. (6.61)] can be deduced from either the Coleman-Noll
finite linear visco-elastic theory (Coleman and Noll, 1960; 1961) or the Green-Revlin
multiple integral (Green, and Rivlin, 1957). In fact, the one-dimensional form of the
Green-Revlin multiple integral can be expressed as:
t
σ (t) = φ1 (t − τ1 )ε̇(τ1 )dτ1
−∞
t
+ φ2 (t − τ1 , t − τ2 )ε̇(τ1 )ε̇(τ2 )dτ1 dτ2
−∞
t
+ φ3 (t − τ1 , t − τ2 , t − τ3 )ε̇(τ1 )ε̇(τ2 )ε̇(τ3 )dτ1 dτ2 dτ3 + · · · (6.66)
−∞
where φi is relaxation function, the first term on the right-hand side of the equation is the
linear term that obeys the Boltzmann superposition principle, and the rest are nonlinear
terms representing the joint contributions of the strain history to the final stress.
However, Eq. (6.66) is obviously mathematically intractable, particularly for engineering

applications. In fact, even if we only take the first three terms, at least 28 groups of
experiments at different conditions should be completed in order to determine those three
relaxation functions φ1 , φ2 , φ3 (Lockett, 1972).
Practically, from the hint given by the experimental results mentioned above, it is
reasonable to assume that (Wang and Yang, 1992)

t − τ1 t − τ1
φ1 (t − τ1 ) = E0 + E1 exp − + E2 exp −
θ1 θ2
φ2 (t − τ1 , t − τ2 ) = const. = α
φ3 (t − τ1 , t − τ2 , t − τ3 ) = const. = β
Then, the Green-Revlin equation [Eq. (6.66)] immediately reduces to the ZWT equation
[Eq. (6.61)].
6.4 Stress Waves Propagating in Nonlinear Visco-Elastic Bars
Now, let us discuss the stress waves propagating in nonlinear visco-elastic bodies, partic-
ularly to see how the nonlinear terms in the constitutive relation influence the visco-elastic
wave propagation. For convenience, we will mainly discuss the longitudinal propagation
in ZWT bars.
The control equations are composed of the movement equation
∂v ∂σ
ρ0 = (6.67)
∂t ∂x
the continuous equation
∂v ∂ε
= (6.68)
∂x ∂t
and the nonlinear visco-elastic constitutive equation. With regard to ZWT materials, under
impact-loading conditions where the timescale is on the order of 1 to 102 µs, the ZWT
nonlinear visco-elastic constitutive equation [Eq. (6.61)] is reduced to Eq. (6.65)

t t −τ
σ = (E0 + E1 ) ε + αε 2 + βε 3 + E2 ε̇(t) exp − dτ
0 θ2
or equivalently in the differential form:
∂σ σ ∂ε σ (ε)
eff
+ = σeff (ε) + E2 + (6.69a)
∂t θ2 ∂t θ2
where σeff (ε) is effective nonlinear pure elastic response:
dfe (ε)
σeff (ε) = fe (ε) + E1 ε, σeff (ε) = + E1 (6.69b)
dε
Similar to the characteristics method previously discussed for linear visco-elastic waves
propagating in bars, the above partial differential control equations are equivalent to
the following three groups of ordinary differential equations. The first and second
groups are
dx = ±CV dt (6.70a)
/ 0
1 σ − σeff (ε) 1 σ − σeff (ε)
dv = ± dσ ± dt = ± dσ + + E )θ dx (6.70b)
ρ0 CV ρ 0 C V θ2 ρ0 C V (σeff 2 2
where the plus and minus signs are for rightward-propagating and leftward-propagating
waves, respectively, and the propagation velocity along the characteristics, CV , is

1 d(σeff + E2 ) σe (ε) + E1 + E2
CV = = (6.71)
ρ0 dε ρ0
The third group corresponds to the particle motion loci and the material derivative of the
constitutive relation of material [Eq. (6.69)] along the particle motion locus:
dx = 0 (6.72a)
dσ σ − σeff (ε) dt
dε − + E − σ + E θ = 0 (6.72b)
σeff 2 eff 2 2
Thus, if the material parameters are known, the nonlinear visco-elastic waves propagating
in the bar can be solved by the characteristics method.
Note the terms having dt or dX in the compatibility conditions, Eqs. (6.70b) and (6.72b).
They are the terms that describe the dispersion and dissipation characters of nonlinear
visco-elastic waves. Moreover, it is worthwhile to note that the high-frequency relaxation
time θ2 in Eqs. (6.70b) and (6.72b) always appears in the terms with dt or dX, in the form
(σ − σe )dt/θ2 or (σ − σe )dX/(Cv θ2 ). It means that the dispersion and attenuation of
visco-elastic waves, in fact, mainly depend on θ2 and the stress difference between the
overall visco-elastic stress and the pure elastic equilibrium stress [σ (ε) − σe (ε)], called
“over stress”.
In the following, the numerical results by characteristics method are compared between
two typical polymers, namely, a thermoplastic one, PMMA, and a thermoset one, epoxy.
The high-frequency relaxation time θ2 for PMMA is 95.4 µs, while for epoxy is 8.57 µs;
the former is larger than the latter by about one order. The rest material parameters for
numerical calculation are listed in Table 6.1. Under the boundary condition of constant
velocity impact (80 m/s) applied at the end (X = 0) of a semi-infinite bar, the stress
profiles and strain profiles at the different distances from the impact boundary (X = 0,
0.2, 0.4, 0.6, 0.8, 1.0 m) are calculated and given respectively in Figs. 6.13(a and b) for
PMMA, and in Figs. 6.14(a, b) for epoxy (Wang et al., 1995; Wang, 2003). In those figures,
in addition to the wave profiles for nonlinear visco-elastic waves (as shown in solid lines),
the profiles for linear approximation are given too for comparison (as shown by dash lines),
which can be obtained once the nonlinear elastic coefficients vanish (α = β = 0).
300
STRESS (MPa) 200
100
0
0 100 200 300 400 500 600 700
(a) TIME (micro-second)
5
STRAIN (%)
0
0 100 200 300 400 500 600 700
(b) TIME (micro-second)
Fig. 6.13. Nonlinear visco-elastic waves in PMMA bar under constant velocity impact condition
(80 m/s), showing (a) stress profiles and (b) strain profiles, at X = 0, 0.2, 0.4, 0.6, 0.8, 1.0 m,
respectively, (solid lines: nonlinear, dash lines: linear).
As can be seen from Figs. 6.13 and 6.14, either for PMMA or for epoxy, the influences
of constitutive nonlinear elasticity on wave profile shape and wave attenuation cannot be
neglected in the case of high-velocity large deformation; even the visco-elastic response
is linear.
Moreover, from Figs. 6.13 and 6.14, it is worthwhile to point out the following basic
common characters for visco-elastic waves:
First, in contrast to linear elastic waves, no proportional relation exists between the visco-
elastic stress wave, strain wave, and particle velocity wave [see Eqs. (6.70b) and (6.72b)].
One of the most distinct characters of visco-elastic waves is that for the position near and
at the impact boundary, the visco-elastic stress decreases with time, showing a “stress-
relaxation-like” character, while the visco-elastic strain, on the contrary, increases with
time, showing a “strain-creep-like” character. It implies that the visco-elastic dynamic
stress σ (X, t) at an arbitrary position X cannot be determined directly from an experi-
mentally measured dynamic strain ε(X, t) at that position by simply multiplying with the
apparent elastic modulus.
300
STRESS (MPa)
200
100
0
0 100 200 300 400 500 600 700 800 900
(a) TIME (micro-second)
6.0
5.0
STRAIN (%)
4.0
3.0
2.0
1.0
0.0
0 100 200 300 400 500 600 700 800 900
(b) TIME (micro-second)
Fig. 6.14. Nonlinear visco-elastic waves in epoxy bar under constant velocity impact condition
(80 m/s), showing (a) stress profiles and (b) strain profiles, at X = 0, 0.2, 0.4, 0.6, 0.8, 1.0 m,
respectively, (solid lines: nonlinear, dash lines: linear).
Secondly, all visco-elastic waves attenuate with distance X, i.e., the wave amplitude
decreases with the increase of propagating distance. Such attenuation, according to
Eqs. (6.70b) and (6.72b), is mainly dominated by the following θ2 -related term:
σ − σeff (ε)
+ E )θ
(σeff 2 2
In other words, wave attenuation becomes weaker with increasing θ2 .
Finally, recall that each θ has its “effective influence domain” about 4.3 orders of magnitude
in either timescale or strain-rate scale [Eq. (6.64)]. Then, with regard to the propagation
of visco-elastic waves, an “effective influence time” (EIT), teff = θ2 , or an “effective
influence distance” (EID), Xeff = Cv0 /θ2 , should correspondingly exist, and both are
effectively dominated by θ2 . Once t > teff or x > Xeff , θ2 will no longer play a marked
influence. For example, from the material parameters listed in Table 6.1, it can be estimated
that Xeff = 0.214 m for PMMA, and Xeff = 0.0205 m for epoxy. Noticing such a
θ2 -related character is significant either for the studies on visco-elastic wave propagation
(direct problems) or for the studies on visco-elastic constitutive relationships of materials
by means of visco-elastic wave propagation (inverse problems).
6.5 Elastic-Visco-Plastic Constitutive Relationship
It has long been known that the plastic behavior of materials under impact loading is
markedly different from that under quasi-static loading, and it has been understood that
this is caused by the influence of strain rate on plastic deformation. Among researches
of great historical significance, the impact tensile experiments of steel wire (Fig. 6.15),
conducted in succession by Hopkinson (1872), the father, and Hopkinson (1905), the
son, should be specially mentioned. The three most important results obtained from
their investigations are: (1) The tensile rupture of steel wire under impact loading is
controlled by the height of the falling weight, i.e. the impact velocity, but is indepen-
dent of its mass. (2) The position at which the steel wire breaks under impact tension
is at the fixed suspension end B, but not at the impact end A. (3) The dynamic yield
stress measured, Yd , is as high as about twice the static yield stress, Ys , and the steel
wire can withstand a duration of about 100 µs under a stress level of 1.5 Ys without
any evident yield (retard yield phenomenon). The first two results result from stress
wave propagation, which can be understood from the discussions given in the previous
chapters, while the third is attributed to the strain-rate-dependent mechanical behavior of
materials.
Recall what we have discussed in the rate-independent plastic wave theory: on the one
hand, it is emphasized that the dynamic stress–strain curves be distinguished from the
static stress–strain curves, which means that the strain rate effect is recognized; on the
other hand, for convenience, it is assumed that the dynamic stress is a single-valued
function of dynamic strain, and thus the constitutive equation in the rate-independent form
is adopted. Although this theory has succeeded in solving a great deal of engineering
X
B
Falling weight
Steel wire
Fig. 6.15. Impact tensile experiment of steel wire.

problems, people found later that the theory did not completely coincide with experimental
facts in some aspects. For example, when a bar suffers a constant velocity impact, according
to the rate-independent plastic wave theory, a constant plastic strain zone as shown in
Fig. 2.10 will be formed around the impacted bar end. However, it was not consistently
verified by all the known experiments (Lee, 1956; Riparbelli, 1953; Sternglass and Stuart,
1953). In another example, when an elastic–plastic bar, preloaded to a certain plastic
deformation εp , suffers an infinite impact disturbance, according to the rate-independent
plastic wave theory, an incremental wave will propagate with the plastic wave velocity
Cp , depending on the slope of the σ ∼ ε curve at ε = εp (Eq. 2.15). However, several
experimental results have shown that the incremental wave in fact propagates, but with
elastic wave velocity C0 (Bell, 1951; Alter and Cautis, 1956; Bianchi, 1964). These facts
all indicate that the rate-independent plastic wave theory has some limitations. To remove
such limitations, the constitutive relation should take account of the rate-dependent plastic
deformation. Thus, the elastic-visco-plastic wave theory was developed toward the end
of 1940s.
For this purpose, first of all, the elastic-visco-plastic constitutive relation should be estab-
lished. A number of experiments have been conducted to study the influence of strain
rate on the yield stress or flow stress at a given strain (Campbell, 1973). The earliest
experimental research can be traced back to the work of Lüdwik (1909). Based on the
experimental data reported by a number of researchers, two kinds of empirical equations
under one-dimensional stress state were proposed, namely, the power function law and
the logarithmic function law expressed, respectively, as:
n
σ ε̇
= (6.73)
σ0 ε̇0

σ ε̇
= 1 + λ ln (6.74)
σ0 ε̇0
where σ0 is the yield stress or flow stress measured at quasi-static test in strain rate ε̇ = ε̇0 ,
and n and λ are material parameters characterizing the strain rate sensitivity of mate-
rial. Equation (6.73) is geographically represented by a straight line in the ln σ –ln ε̇
plot, while Eq. (6.74) is represented by a straight line in the σ –ln ε̇ plot, as shown in
Fig. 6.16. This means that strain rate displays marked influence only when its value varies
in orders.
Equation (6.74) can be rewritten as

1 σ − σ0
ε̇ = ε̇0 exp (6.75)
λ σ0
Note that (σ − σ0 ) is the difference between dynamic stress and static stress, usually called
over stress or extra stress, and, consequently, (σ − σ0 )/σ0 is the dimensionless over
stress. Thus, Eq. (6.75) means that strain rate is a function of over stress.
Based on these experimental results, the so-called elastic-visco-plastic theory was devel-
oped, which is a generalization of the classical plastic theory to take account of the
⎛ s⎞ s
ln ⎜ ⎟ s0
⎝ s0 ⎠
⎛ e⋅ ⎞ ⎛ e⋅ ⎞
ln ⎜ ln ⎜
⎝ e⋅0 ⎟ ⎝ e⋅0 ⎟
⎠ ⎠
(a) (b)
Fig. 6.16. Two kinds of strain rate dependent equation, (a) the power function law and (b) the
logarithmic function law.
strain rate effect of plastic deformation, but neglecting the strain rate effect of elastic
deformation. In other words, assume that the total strain rate ε̇ is composed of the elastic
component (instantaneous response) ε̇ e and the nonelastic component (noninstantaneous
response) ε̇ p :
ε̇ = ε̇ e + ε̇ p (6.76)
where the nonelastic component ε̇p is a function of over stress, representing the plastic
response coupled with the viscous response.
It was (Sokolovskii)(1948) who, based on the perfect plastic model, first

proposed the following elastic-visco-plastic constitutive relationship:

σ̇ |σ |
ε̇ = + Sign σ · g −1 (6.77)
E Y0
where the ε̇p is regarded as a function of such an over stress, defined as the difference
between the dynamic stress σ and the static yield limit of perfect plastic body Y0 , i.e.
g[(|σ |/Y0 ) − 1]. Under different constant strain rates, Eq. (6.77) is represented by a set of
stress–strain curves, of which the visco-plastic deformation parts are parallel horizontal
straight lines, as shown in Fig. 6.17(a). The function g is generally a nonlinear func-
tion. If g is a linear function of over stress, then Eq. (6.77) is reduced to the following
form:
σ̇ |σ | − Y0 σ̇ |σ | − Y0
ε̇ = + = + Sign σ · γ ∗ (6.78)
E η E Y0
which is similar to the visco-elastic constitutive equation for Maxwell body, except the
viscous term in Eq. (6.4) is now replaced by the over stress viscous term. If the perfect
E
e⋅3p> e⋅2p
e⋅2p> e⋅1p
e⋅1p> e⋅0p
e⋅ 0p
Y0 h
0 e
(a) (b)
Fig. 6.17. The elastic visco-plastic constitutive model described by Eq. (6.78).
plastic body is rheologically represented by a couple of sliding plates with constant

friction force Y0 , then Eq. (6.78) is correspondingly represented by such a three-elements
model as shown in Fig. 6.17(b), which is composed of a spring with a visco-plastic
model (the so-called Binham model) connected in series, and the latter is again com-
posed of a dashpot and a sliding-plate element connected in parallel. The γ (=Y0 /η) in
Eq. (6.78) is a material parameter characterizing the viscous character of the visco-plastic
model.
In order to take account of strain hardening, Malvern (1951) suggested replacing the Y0 in
Eq. (6.77) with the static stress–strain curve σ0 (ε) and, thus, the following elastic-visco-
plastic constitutive equation was proposed:

σ̇ σ
ε̇ = + Sign σ · g −1 (6.79)
E σ0 (ε)
which is equal to regarding the friction force Y0 in the rheological model shown in
Fig. 6.17(b) as a function of plastic strain εp , Y = Y (ε p ). Under different constant strain
rates, Eq. (6.79) is represented in the σ –ε plane by such a set of stress–strain curves that
the visco-plastic deformation parts are all parallel to the static stress–strain curve, namely
the distance between any two visco-plastic parts of curves is kept as constant, as shown
in Fig. 6.18.
According to -Malvern model,only when the over stress is greater than zero
can the visco-plastic deformation happen. When the over stress is less than or equal to zero,
only elastic deformation happens. Mathematically, it can be expressed by the following
s
e⋅3p> e⋅2p
e⋅2p> e⋅1p
e⋅1p> e⋅0p
e⋅0p
s 0(e)
0 e
Fig. 6.18. The elastic visco-plastic constitutive model described by Eq. (6.79).
unified form of constitutive equation:

= > ⎫
ε̇ =
σ̇
+ g
σ
−1 ⎪
⎪
⎪
⎪
E σ0 (ε) ⎪
⎪
⎪
⎬
⎧ σ
⎪ − 1 ≥ 0⎪ (6.80)
⎨0, for ⎪
σ0 (ε) ⎪
⎪
g = σ ⎪
⎪
⎪
⎩g, for − 1 > 0⎪
⎭
σ0 (ε)
Or, similar to Eq. (6.78), introduce a viscous coefficient γ ∗ , defined by the following
equation:
g ξ = γ ∗ φ(ξ )
then Eq. (6.80) can be rewritten in another form as:
σ̇ ⎫
ε̇ = + γ ∗ φ(F ) ⎪
⎪
⎪
⎪
E ⎪
⎪
⎪
⎬
0, F ≤0 (6.81)
φ(F ) = ⎪
⎪
φ(F ) F > 0 ⎪
⎪
⎪
⎪
σ ⎪
⎭
F= −1
σ0 (E)
For the -Malvern elastic-visco-plastic body, it should be noted that its enter-
ing into the visco-plastic region from the elastic region or returning the elastic region from
the visco-plastic region completely depends on weather the over stress is greater than
zero. As long as the over stress is greater than zero,i.e. either the stress is increasing with
s
A C B
Y0
A C
Y0
D
0 0 t
D e
(b)
(a)
Fig. 6.19. The σ –ε curve for the elastic visco-plastic materials described by Eq. (6.78) in the case
of linearly loading and then linearly unloading.
time (∂σ/∂t > 0) or decreasing with time (∂σ/∂t < 0), it follows the same constitutive
relationship, i.e. the visco-plastic deformation still continuously develops, even when the
stress is decreasing with time. For example, if the stress is first linearly increasing with
time, and then linearly decreasing with time, as shown in Fig. 6.19(b), the σ –ε curve
calculated according to Eq. (6.78) is shown in Fig. 6.19(a). As can be seen from the figure,
the visco-plastic deformation still continuously develops when stress begins to decrease.
Only when the stress is unloaded to the value indicated by the point C, showing that over
stress equals to zero, a transition from the visco-plastic state to the elastic unloading state
happens. Note that point C neither corresponds to the beginning of stress decrease or
the beginning of strain decrease. In such aspects, elastic-visco-plastic materials are fully
different from elastic–plastic materials.
6.6 Stress Waves Propagating in Elastic-Visco-Plastic Bars
Let us now discuss the stress waves propagating in the –Malvern elastic-
visco-plastic bars. The control equations are composed of the continuous equation
[Eq. (2.12)], the motion equation [Eq. (2.13)], and the elastic-visco-plastic constitutive
equation [Eq. (6.81)], namely
⎫
∂v ∂ε ⎪
= ⎪
⎪
∂X ∂t ⎪
⎪
⎪
⎬
∂v ∂σ
ρ0 = (6.82)
∂t ∂X ⎪
⎪
⎪
⎪
∂ε 1 ∂σ ⎪
⎪
= + γ φ (F )⎭
∗
∂t E ∂t
which are one-order partial differential equations and can be solved by the characteristics
method. Similar to what we have discussed on visco-elastic wave propagation, in order to
find the characteristics equations and the corresponding characteristic compatibility equa-
tions, multiplying each of them with undetermined coefficients N , M, and L, respectively,
and then adding together, we have:

∂ε ∂ ∂ N ∂ ∂
(L + N ) + Mρ0 − L v− +M σ − N γ ∗ φ (F ) = 0
∂t ∂t ∂X E ∂t ∂X
(6.83)
These undetermined coefficients should satisfy the following conditions:
dX 0 L ME
= =− = (6.84)
dt L+N Mρ0 N
which are fully the same as Eq. (6.38), and, thus, similar to Eq. (6.40) and Eq. (6.41),
three real characteristics differential equations and three corresponding compatibility
differential equations can be obtained, respectively, as:

E
dX = ± dt = ±C0 dt (6.85a)
ρ0
1 1
dv = ± dσ ± C0 γ ∗ φ (F ) dt = ± dσ + γ ∗ φ (F ) dX (6.85b)
ρ0 C 0 ρ0 C0
and
dX = 0 (6.86a)
dσ
dε = + γ ∗ φ(F ) dt (6.86b)
E
The characteristics described by Eq. (6.85a) represent the propagating loci of wave front
on the X–t plane, while the characteristics described by Eq. (6.86a) represents the motion
loci of particles. Since all these are similar to the situations previously discussed for visco-
elastic waves propagating in Maxwell bars, no details need be repeated again, except that
the original σ/η should be correspondingly replaced by γ ∗ φ(F ) now.
As an example, let us discuss a semi-infinite bar, which is originally at rest and in a

stress-free state (v0 = σ0 = ε0 = 0) and, at the bar end (X = 0), suddenly suffers a
constant stress σ ∗ . Similar to the situation shown in Fig. 6.9, since the relation between
stress discontinuity [σ ] and strain discontinuity [ε] across the strong discontinuity wave
front is determined by the instantaneous response of Eq. (6.81), the strong discontinuous
wave propagates along OA with elastic wave velocity C0 . Note that, on the one hand, the
dynamic compatibility condition and the kinematics compatibility condition across the
strong discontinuous wave locus OA, namely Eq. (6.42), should be satisfied; and, on
the other hand, along the characteristics OA, the following characteristics compatibility
condition along the rightward characteristics, namely Eq. (6.85b), should be satisfied too:
1
dv = dσ + γ ∗ φ(F ) dX
ρ 0 C0
Instituting Eq. (6.42a) into the above equation, we have
dσ 1
= − ρ0 C0 γ ∗ φ(F ) = −φ[X, σ (X)] (6.87)
dX 2
Thus, the problem is attributed to solve the following integral equation:

X
σ (X) = σ ∗ − ψ[ξ, σ (ξ )] dξ
0
where σ ∗ has been given by the boundary condition at the bar end. If ε̇ p is a linear function
of the over stress with respect to the yield stress Y0 of perfect plastic material, namely the
constitutive relation is expressed by Eq. (6.78), Then, from Eq. (6.87), the solution along
OA can be obtained as

∗ ρ0 C0
σ = Y0 + (σ − Y0 ) exp − X (6.88)
2η
As can be seen, the attenuation factor in Eq. (6.88) is the same as that in Eq. (6.43a).
In other words, the over stress of the strong discontinuous elastic-visco-plastic wave
exponentially attenuates with propagation distance X – the same as the attenuation of strong
discontinuous Maxwell visco-elastic wave. Once the solution along OA is determined, the
solutions in the region AOt are attributed to solve a solvable characteristics boundary
problem, the same as has been discussed for Fig. 6.9.
If the Y0 in Eq. (6.78) is replaced by the static stress–strain curve σ0 (ε) to take account of
strain hardening effect, the problem is still solvable by a similar method (Malvern, 1951).
Some results are given in Fig. 6.20 (plotted by solid lines). For comparison, the results
calculated according to the rate-independent elastic plastic theory are also given in the
same figure (plotted by dash lines).
It is worthwhile to notice the following important differences between the rate-dependent

elastic-visco-plastic waves and the rate-independent elastic plastic waves:
1. Because the instantaneous response of elastic-visco-plastic material is elastic, under

the same suddenly applied constant velocity impact condition, the initial stress
amplitude of an elastic-visco-plastic strong discontinuous wave is higher than that
of an elastic-plastic wave. This is the so-called “over stress effect”. However, due
to the viscous effect, the stress amplitude of an elastic-visco-plastic strong dis-
continuous wave will continuously attenuate and approach the amplitude of strong
discontinuous elastic precursor wave in the elastic–plastic wave theory, i.e. the over
stress will gradually decrease until zero, as shown in Fig. 6.20(a). But the prop-
agating velocity of the elastic-visco-plastic wave is kept the same as the elastic
wave velocity C0 , regardless of what the value of stress discontinuity is. This coin-
cides with the experimentally observed fact that the incremental wave velocity in a
preplastically deformed bar is C0 .
t
e=0.007
e=0.005
s
e=0.003
along X=C0 t e=0.002
0 0
X X
(a) (b)
s
e X=0
X=0.32
X=0.64
s0(e)
X=1.28
0 X 0 e
(c) (d )
Fig. 6.20. Dynamic response of a semi-infinite bar under constant velocity impact, wherein solid
lines indicate the elastic-visco-plastic solution, and dash lines indicate the elastic–plastic solution.
(a) the stress distribution along the strong discontinuous wave OA, (b) iso-strain lines on X–t plane,
(c) the maximum strain distribution along the bar, (d) the stress–strain curves at different X of the bar.
2. With regard to the weak discontinuous wave region behind the strong discontinu-
ous wave front propagating in a semi-infinite bar under constant initial condition,
according to the elastic-plastic wave theory, it is a simple wave region. Conse-
quently, the σ, ε, and v are all invariable along the characteristics, but, according to
the elastic-visco-plastic wave theory, it is no longer a simple wave region and, con-
sequently, the iso-strain line no longer coincides with the characteristics, as shown
in Fig. 6.20(b). If the slope of iso-strain line dX
dt ε=const is defined as the phase
velocity of strain propagation, then in the case of small strains, the strain propa-
gates quicker in the elastic-visco-plastic bar than in the elastic–plastic bar, because
in such case the instantaneous response plays a dominating role. However in the
case of large strains, the strain propagates slower in the elastic-visco-plastic bar than
in the elastic–plastic bar, because in such case the viscous response with a delayed
effect plays a dominating role.
3. When a bar suffers a constant velocity impact, according to the rate-independent
plastic wave theory, a constant plastic strain zone as shown in Fig. 2.10 will be
formed around the impacted bar end. However, such a constant plastic strain
zone no longer exists according to the elastic-visco-plastic wave theory due to
the viscous effect. Moreover, the maximum strain at the bar end predicted by the
elastic-visco-plastic wave theory is larger than that predicted by the elastic–plastic
wave theory, as shown in Fig. 6.20(c), which is more in conformity with some
known experimental facts.
4. At a different cross-section of the bar, the situation of how the strain rate varies
with time is different, and, thus, the stress–strain curve calculated at the different
cross-section is different, as shown in Fig. 6.20(d). However, the strain rate at
every cross-section gradually decreases to zero, so that the stress–strain curve at
every cross-section finally approaches the quasi-static stress–strain curve σ0 (ε).
Other examples of elastic-visco-plastic waves propagating in bars (including the problems

for finite bars, for the determination of loading–unloading boundaries, i.e. the boundaries
of the elastic zone and the visco-plastic zone, and when the instantaneous response contains
instantaneous plastic response, etc.) all can be studied on the basis of the previous dis-
cussions, or some related monographs can be referred to (e.g. Cristescu, 1967; Nowaski,
1978). In the next two chapters, we will further discuss the one-dimensional strain elastic-
visco-plastic waves (in Section 7.11) and the elastic-visco-plastic spherical and cylindrical
waves (in Section 8.5), respectively.
CHAPTER 7
One-Dimensional Strain Plane Waves
7.1 Governing Equations
In the elementary theory of stress wave propagating in bars, it is assumed that plane cross
sections normal to the longitudinal axis of the bar remain plane in deformation, and on
the cross section, there are only uniformly distributing axial stresses σX . Therefore, the
problem is reduced to a one-dimensional (1D) stress problem. As previously discussed in
Section 2.8, this assumption is valid approximately only when the transverse size of the
bar is much smaller than the wavelength.
As another extreme case, if the transverse size perpendicular to the propagating direction
of longitudinal wave (X axis) is large enough to constrain any transverse motion, namely if
uY = uZ = 0 (7.1a)
∂uY ∂uZ
εY = = 0, εZ = =0 (7.1b)
∂Y ∂Z
∂uY ∂uZ
vY = = 0, vZ = =0 (7.1c)
∂t ∂t
where symbols denote the same variables as those in previous chapters, and the subscriptsY
and Z denote the components along Y and Z direction respectively, then there is only
disturbance of axial strain εX , propagating in the media, which is called 1D strain plane
wave. In such a case, strictly speaking, there is no transverse inertia effect, but there is
constraint from the corresponding lateral normal stress σY and σz , so the media is in the
state of 3D stress. Assuming that the acting σz are uniformly distributed on cross sections,
then all nonzero stress components and strain components are functions of X and t only.
Due to symmetry, there is
σY (X, t) = σZ (X, t) (7.2)
Strictly, only when a semi-infinite body is subjected to uniformly distributing normal

impact, can it be a problem of 1D strain plane longitudinal wave propagation. However, for
265
a practical problem, during the early period when the disturbances reflected from the lateral
boundaries have not arrived at the area discussed yet, it can be treated as a 1D strain plane
wave problem. For example, in a contact explosion on a plate or a high velocity normal
impact onto a plane, if the plate transverse sizes are larger enough than the plane thickness,
during the early period of stress wave propagation, we can view these problems as problems
of 1D strain plane longitudinal waves.
Similar to the situation discussed on 1D stress longitudinal waves in bars, the governing
equations still consist of three aspects: the continuity equation, the momentum conser-
vation equation, and the constitutive equation of material. If the problem is described by
Lagrange variables and the axial stress is assumed to be a single-value function of the axial
strain, then the governing equations in form are the same as that of 1D stress longitudinal
waves in bars discussed in Section 2.2.
∂vX ∂εX
= (7.3a)
∂X ∂t
∂vX ∂εX
= (7.3b)
∂X ∂t
σX = σX (εX ) (7.3c)
Note that the current relationship between stress and strain, σ –ε, is no longer the relation-
ship under the 1D tension or compression stress state. Equation (7.3) can also be expressed
in other forms, for example, by eliminating σX , we obtain the following first-order partial
differential equations with respect to vX and εX as unknown variables
∂vX ∂εX
= (7.4a)
∂X ∂t
∂vX ∂εX
= CL2 (7.4b)
∂t ∂X
By eliminating εX , we obtain the following first-order partial differential equations with

respect to σX and vX as unknown variables
∂vX ∂σX
ρ0 = (7.5a)
∂t ∂X
∂v X ∂σ X
ρ0 CL2 = (7.5b)
∂X ∂t
If we use the longitudinal displacement uX (X, t) as an unknown function, then we obtain

the following second-order partial differential equation
∂ 2 uX ∂ 2 uX
2
− CL2 =0 (7.6)
∂t ∂X 2
One-Dimensional Strain Plane Waves 267
The CL in the above equations is the wave velocity of 1D strain longitudinal plane wave

1 dσX
CL = (7.7)
ρ0 dεX
Thus it can be seen that most of the methods used to analyze the 1D stress waves can be
directly adopted and thus will not be repeated again here. In the following sections, we
will focus on some specific characteristics of 1D strain waves.
7.2 One-Dimensional Strain Elastic Waves
The differences between elastic 1D strain waves and 1D stress waves are mainly caused
by the difference in the stress/strain states, and consequently the different longitudinal
stress–strain relationships. In the former problem, media are in the state of 3D stress,
while in the latter problem, media are in the state of 1D stress. To deduce the relationship
between longitudinal stress and strain in 1D strain problems, we first review the general
Hooke’s law below.
It is well known from elastic mechanics, the general Hooke’s law for isotropic,
homogenous, linear elastic materials can be written as
⎫
1
εX = [σX − ν(σY + σZ )]⎪
⎪
⎪
E ⎪
⎪
⎪
⎪
1 ⎪
εY = [σY − ν(σZ + σX )]⎪⎪
⎪
⎪
E ⎪
⎪
⎪
⎪
1 ⎪
εZ = [σZ − ν(σX + σY )]⎪⎬
E (7.8)
1 ⎪
εXY = τXY ⎪
⎪
⎪
2G ⎪
⎪
⎪
⎪
1 ⎪
⎪
εYZ = ⎪
τYZ ⎪
2G ⎪
⎪
⎪
⎪
1 ⎪
⎪
εZX = τZX ⎭
2G
where strain components are defined below:

⎫
∂uX 1 ∂uX ∂uY ⎪
εX = , εXY = + ⎪
∂X 2 ∂Y ∂X ⎪
⎪
⎪
⎪
∂uY 1 ∂uY ∂uZ ⎬
εY = , εYZ = + (7.9)
∂Y 2 ∂Z ∂Y ⎪⎪
⎪
∂uZ 1 ∂uZ ∂uX ⎪⎪
⎪
εZ = , εZX = + ⎭
∂Z 2 ∂X ∂Z
where E is the Young’s modulus, G is the shear modulus, and v is the Poisson’s ratio. If we
regard stresses as functions of strains, the general Hooke’s law in the tensor form can be
equivalently expressed as
σij = λδij εkk + 2µεij (7.10)
where δij is the Kronecker defined by

0, i
= j
δij =
1, i=j
λ and µ are Lame’s coefficients:
Ev
λ=
(1 + ν)(1 − 2ν)
(7.11)
E
µ=G=
2(1 + ν)
If introducing the volume strain ∆ and the hydrostatic stress p (the negative of the average
normal stress), respectively, defined by
∆ = ε X + εY + εZ (7.12)
−p = 13 σkk = 13 (σX + σY + σZ ) (7.13)
and introducing the deviator strain tensor eij characterizing distortion (shape deformation)
and the corresponding deviator stress tensor sij :

eij = εij − 13 ∆δij
(7.14)
sij = σij + pδij
then the generalized Hooke’s law can be expressed in two parts: the volumetric deformation
law and the distortional deformation law:

−p = K∆
(7.15)
sij = 2Geij
where K is the volumetric modulus, which is related to other elastic coefficients by the
following relation:
2G E
K =λ+ = (7.16)
3 3(1 − 2ν)
Under the current 1D strain condition (εY = εZ = 0), we have

∆ = εX
(7.17)
eXX = εX − 13 εX = 23 εX
Thus the relationship between longitudinal stress and longitudinal strain can be obtained,

σX = −p + sXX = K∆ + 2GeXX = K + 43 G εX (7.18a)
or from Eq. (7.10), we obtain
σX = (λ + 2µ)εX = EL εX (7.18b)
where EL is called the laterally constrained elastic modulus
(1 − ν)E
EL = K + 43 G = λ + 2µ = (7.18c)
(1 + ν)(1 − 2ν)
Substituting Eq. (7.18) into Eq. (7.7), the wave velocity of elastic 1D strain longitudinal
waves, CLe , can be obtained

EL 1 4 1 1 (1 − ν)E
CLe = = K+ G = (λ + 2µ) = (7.19)
ρ0 ρ0 3 ρ0 ρ0 (1 + ν)(1 − 2ν)
The laterally constrained stresses σY and σZ can be obtained from Eqs. (7.8) and (7.10)
v λ
σY = σZ = σX = σX (7.20)
1−v λ + 2µ
Therefore, all the methods used and the results obtained in the previous discussions of
elastic 1D stress longitudinal waves can be in principle adopted to the analyses of elastic
1D strain longitudinal waves, only if that the Young’s modulus E is replaced by the laterally
constrained elastic modulus EL . However, the following points should be noticed.
(1) Since λ > 0, µ > 0, then it is known from Eq. (7.20) that the signs of σY and σZ
are always the same as the sign of σX . So, when a 1D strain plane wave propagates
in the media, it is either in the state of 3D compression or 3D tension.
(2) Since in general 0 < v < 0.5, then it is known from Eqs. (7.18) and (7.19) that
EL > E, CLe > C0 , namely a 1D strain elastic longitudinal wave propagates
faster than a 1D stress elastic longitudinal wave in bars. Data of CLe for common
materials are shown in Table 7.1 (Kolsky, 1953), referring to the data of C0 in the
Table 2.1 and CT in the Table 2.2. For common metals, v is in the range of 1/4–1/3,
and correspondingly EL /E is in the range of 1.2–1.5, CLe /C0 is in the range of
1.1–1.22. When v → 1/2 (namely for incompressible materials), CLe → ∞.
(3) The 1D strain elastic longitudinal wave is actually the irrotational plane wave in
isotropic, homogenous, linear elastic, infinite media, and CLe is actually the velocity
of irrotational wave in infinite elastic media. These will be discussed further in
Chapter 11.
Table 7.1. Lame’s coefficients and the values of 1D strain elastic longitudinal wave velocity CLe
for some common materials.
λ (GPa = 1010 dyne/cm2 ) 112 95 56 28 10
µ (GPa = 1010 dyne/cm2 ) 81 45 26 28 7.0 × 10−4
CLe (km/s) 5.94 4.56 6.32 5.80 1.04
7.3 Elastic–Plastic Constitutive Relationship in 1D Strain Condition
First recall that in 1D stress condition, plastic deformation occurs when the stress reaches
the yield stress Y determined in 1D tension test of materials. The yielding criterion can be
expressed by
σX = Y (7.21)
When the strain-hardening effect is taken into consideration, Y is a function of plastic

p
strain εX :
p
Y = Yp εX (7.22)
or a function of plastic work Wp :

Y = YW Wp (7.23)
where plastic work Wp in 1D stress condition is

p
Wp = σX dεX (7.24)
According to Eq. (7.22), the derivative of stress with respect to plastic strain is
dσX dYp
p = p = Yp (7.25)
dεX dεX
while according to Eq. (7.23), it is
dσX dYW dWp dYW

p = . p = .YW = YW YW (7.26)
dεX dWp dεX dWp
So we have YP = YW YW
.
Assuming that the total strain ε is the sum of the elastic strain ε e and the plastic strain ε p
ε = εe + εp (7.27)
and since in the elastic scope

p
dεX 1
= (7.28)
dσX E
then in the plastic scope, considering Eqs. (7.25) and (7.26), there should be:
p
dεX dε e dε 1 1 1 1
= X + X = + = +
dσX dσX dσX E YP E YW Y W
Therefore, the slope dσX /dεX of the plastic σX –εX curve in 1D stress condition can be
expressed by E and Yp or YW as
EYP EYW YW
dσX
= Ep = = (7.29)
dεX E + YW E + Y W YW
The EP is called the plastic hardening modulus. If Ep is a constant, the material is a

linear hardening material.
In the general 3D condition, assuming that the material is isotropic, the hydrostatic stress
contributes none to yielding, and the Bauschinger’s effect is negligible, then there are
two yield criteria extensively used: the Mises criterion (the maximum distortion energy
criterion) and the Tresca criterion (the maximum shear stress criterion).
According to the Mises criterion, Eq. (7.21) is generalized to the following criterion
√ <
; 2
σi = 3J2 = (σ1 − σ2 )2 + (σ2 − σ3 )2 + (σ3 − σ1 )2 = Y (7.30)
2
where σ1 , σ2 , σ3 are the principle stresses, J2 = sij sij /2 is the second invariable of
deviator stress tensor, and σi is called effective stress or stress intensity (σi = σ1 in
1D condition). The physical meaning of this criterion is that the plastic deformation
begins when the elastic distortion energy Wd = J2 /(2G) reaches its critical value for
yielding.
The Tresca criteria can be expressed as
σ1 − σ3 = ±Y (7.31)
where σ1 and σ3 correspond to the maximum principle stress and the minimum princi-
ple stress respectively. The physical meaning of this criterion is that plastic deformation
begins when the maximum shear stress τmax [=(σ1 − σ3 )/2] reaches its critical value for
yielding.
In the current 1D strain condition, these two criteria have the same form:
σX − σY = ±Y (7.32)
s us ( sY = e X − s asti us
lo X − li s ly p loc
=Y c)
0
sX
ct ld
)
tic
l
fe ie
as
er r y
pl
(p ppe
ly
U
ct
pe 0
n
rfe
B
ld s tatic
s
ro
0
-Y
c
we Hyd
Y=
ie
X−
ry
s
Lo
A
YH
Y0
C Elastic unloading
sY − sY (B)= ν {sX −sX (B)}
D 1−ν
o
sY
–Y0
Fig. 7.1. Upper and lower yield loci on stress plane (σX , σY ).
On the stress plane (σX , σY ), this equation is graphically represented by two parallel
straight lines with slopes equal to 1, which are called the upper and the lower yield
loci, respectively, as shown in Fig. 7.1. The range between these two lines is an elastic
zone. The two loci are parallel and symmetric to the hydrostatic line σX = σY (= σZ ),
in consistence with the assumption that the hydrostatic stress contributes none to yielding.
For the perfectly plastic material, Y = Y0 , then the yield loci are invariable. As for
the isotropic hardening material, since Y is a function of plastic deformation or plastic
work, then the distance between the two yield loci increases with increasing the plastic
deformation or plastic work, as shown by the two dashed lines in Fig. 7.1.
Noticing the relationship between σY and σX in the 1D strain elastic deformation condition
[Eq. (7.20)], substituting this relationship into Eq. (7.32), then the initial yield strength
YH with respect to the axial stress σX in 1D strain is
1−ν λ + 2µ K + 4G/3
YH = Y0 = Y0 = Y0 (7.33)
1 − 2ν 2µ 2G
which is called the limit of laterally constrained yielding or Hugoniot elastic limit,
corresponding to the point A in Fig. 7.1. Obviously, there must be YH > Y0 . For example,
if ν = 1/3, then YH = 2Y0 .
Based on the above discussions, we will further discuss the elastic–plastic relationship,
σX –εX , in the 1D strain condition as follows.
Assume that the plastic deformation contributes none to the volumetric deformation,
namely
p p p
ε X + εY + εZ = 0 (7.34)
then the volumetric deformation law is completely a pure elastic law. Moreover, if we
notice that in the 1D strain condition we have Eq. (7.17) and the relationship below
σXX = σX + p = 23 (σX − σY ) = 43 τ (7.35)
then the elastic–plastic stress–strain relationship in the 1D strain condition can be

written as
⎧ ⎫
⎪
⎪ Volumetric law : −p = KεX ⎪
⎪
⎪
⎨ ⎧ ⎪
⎬
⎨2GεX (Elastic) (7.36)
⎪
⎪ Distortional law : σX − σY = ⎪
⎪
⎪
⎩ ⎩±Y (Plastic) ⎭ ⎪
So, similar to Eq. (7.18), for the loading, the axial relationship of σX –εX in 1D strain
condition can be written as (refer to Fig. 7.2),
⎧
⎨ K + 4 G εX , σX ≤ YH
3
σX = −p + sXX = −p + 23 (σX − σY ) = (7.37)
⎩
KεX + 23 Y, σX ≥ YH
sX
B'
sX −YH =(K+ 34Gp1)(eX −eH)
B
sX=KeX + 32Y0
sX=KeX
2Y0 E
3 sX −sX (B)=(K+ 43G){eX −eX (B)}
2Y0 C sX (C)=sX(B)−2YH
YH
3
sX=(K+43 G)eX
C'
0 e D D' eX
Fig. 7.2. Axial elastic plastic σX –εX curve in 1D strain condition.

The right side of the above equation is a sum of two items. The first item is related to the
volumetric deformation, as graphically represented by the straight line OE with slope K
in Fig. 7.2, while the second item is related to the distortional deformation.
For a perfect plastic material, Y = Y0 , then the plastic part of σX –εX curve parallels to
the line OE with a perpendicular distance of 2Y0 /3, as shown by the line AB in Fig. 7.2.
In such a case, the slopes of σX –εX curve for both elastic loading and plastic loading are
invariable, namely

dσX K + 43 G, σX ≤ YH
= (7.38)
dεX K, σX ≥ YH
For an isotropic hardening plastic material, the situation is a bit complicated. As can be seen
from Eqs. (7.36) and (7.37), the key to determine the slope dσX /dεX lies on determining
d(σX − σY )/dεX ; namely, how to determine the slope of the distortional stress–strain
curve, d(σX − σY )/d(εX − εY ) or dτ/dγ , where τ is the maximum shear stress while
γ is the maximum shear strain. Based on the symmetry of deformation in the 1D strain
p p
condition (εY = εZ ) and the assumption that plastic deformation contributes none to
volume change [Eq. (7.34)], it can be seen that there is only one independent among the
three principle plastic strains, namely
p p p
εX = 2
3 εX − εY = 43 γ p (7.39)
p
Then, if Y is regarded as a function of axial plastic strain εX and taken into account of
strain hardening, then from the distortional law in Eq. (7.36) it is known that the dτ/dγ
in the elastic part and the plastic part can be respectively expressed as
⎫
dτ ⎪
= 2G ⎪
⎬
dγ e
p (7.40)
= 23 Yp ⎪
dτ 1 dYp dεX
= p ·
⎪
⎭
dγ p 2 dεX dγ p
Thus, similar to Eq. (7.25) deduced in the 1D stress condition, assuming that the total
strain is the sum of an elastic part and a plastic part [Eq. (7.25)], we have
dγ dγ e dγ p 1 3 3G + Yp
= + = + =
dτ dτ dτ 2G 2Yp 2GYp
Corresponding to the shear modulus G in elastic distortional law, the half of the slope of
plastic distortional curve is defined as the plastic shear modulus Gp
1 dτ GYp
Gp = = (7.41a)
2 dγ 3G + Yp
Similar to deducing Eq. (7.22b), if Y is regarded as a function of the axial plastic work Wp ,
namely YW (Wp ), and noting that in the 1D strain condition there is
p p p
dWp = σX dεX + 2σY dεY = (σX − σY ) dεX = 83 τ dγ p
then we have
dτ 1 dY W dW p
= · = YW · 23 (σX − σY ) = 23 YW YW
dγ p 2 dW p dγ p
Comparing with Eq. (7.40), it can be seen that Yp = YW YW . This is consistent with the
Eq. (7.26) deduced in the 1D stress condition. Then the plastic shear modulus can be also
written as

GYW YW
Gp = (7.41b)
3G + YW YW
After introducing the plastic shear modulus, according to Eq. (7.37), the slope of axial
stress–strain curve of an isotropic hardening plastic material in the 1D strain condition is
⎧
dσX ⎨K + 43 G,
σ X ≤ YH
= (7.42)
dεX ⎩K + 4 G , σ ≥ Y
3 p X H
It is obvious that the expression of slope of plastic part is formally similar to that of elastic
part, only if the G is replaced by Gp . For a linear hardening plastic material, Gp = Gpl
(constant), the plastic part of the axial stress–strain curve is a straight line, as graphically
presented by the dash-dotted line AB in Fig. 7.2. This corresponds to
Yp = YW YW

= EY (constant)
p
which indicates Yp is a linear function of εX :
p
Yp = YH + EY εX (7.43a)
2 is a linear function of plastic work W :

or equivalently, YW p
2
YW = YH2 + 2EY Wp (7.43b)
where YH is the Hugoniot elastic limit of materials [Eq. (7.33)], EY is a material constant
characterizing the linear hardening, both of which are determined by experiments.
Now, let us look at the loading path and the unloading path on both the plane (σX , εX )
(Fig. 7.2) and the plane (σX , σY ) (Fig. 7.1), and special attention will be focused on what
differs with those in the 1D stress condition. As an example, we discuss the situation for the
perfectly plastic material. When it is loaded from zero-state (σX = εX = 0), the loading
path first follows the line OA representing the linear elastic deformation process. On the
σX –εX plane, according to Eq. (7.37), its slope is (K + 4/3G); while on the σX –σY
plane, according to Eq. (7.20), its slope is v/(1 − v). After the σX reaches the laterally
constrained yield limit YH Eq. (7.23), the loading path follows the line AB representing
the plastic deformation process. On the σX –εX plane, the slope of stress–strain curve is K
according to Eq. (7.37), while on the σX –σY plane, the slope is 1 according to Eqn (7.36).
If the material is unloaded at the point B, following the assumption of elastic unloading,
the unloading path follows the line BC which is parallel to the elastic loading line OA.
On the plane σX –εX , the unloading stress–strain relation is expressed by

σX = σX (B) + K + 43 G {εX − εX (B)} (7.44)
while on the plane σX –σY , it corresponds to
ν
σY = σY (B) + {σX − σX (B)} (7.45)
1−ν
Note that the material is only subjected to hydrostatic pressure when it is unloaded to the
point E. After that, a sign transition of shear stress (and the corresponding distortional
deformation) takes place. With a further unloading of σX , the shear stress increases in the
opposite direction, namely the elastic distortional deformation is loaded in the negative
direction following the line EC. While at the point C, according to Eq. (7.32), the following
reverse yield criterion is satisfied,
σX − σY = −Y
It indicates that a reverse plastic deformation begins in the material. The stress at C, σX (C),
can be determined by the above equation and Eq. (7.45)
2(1 − ν)
σX (C) = σX (B) − Y = σX (B) − 2YH (7.46)
1 − 2ν
After that, with regard to the original axial stress σX , it is an unloading process following
the line CD; while with regard to the yield criteria it is actually a reverse plastic loading
process following the lower yield locus CD, although it is sometimes inadequately called
plastic unloading.
7.4 One Dimensional Strain Elastic–Plastic Waves
The elastic–plastic σX –εX relationship in the 1D strain condition discussed in the last
section is based on the assumption that the elastic deformation follows Hooke’s law.
Based on those discussions, now let us analyze the propagation of elasto-plastic 1D strain
wave. Research in this area is developed from the works of Wood (1952) and others.
In the loading stage, according to Eq. (7.19), the wave velocity of 1D strain elastic wave is

K + 43 G
CLe =
ρ0
According to Eqs. (7.7) and (7.42), the wave velocity of 1D strain plastic wave is

p K + 43 Gp
CL = (7.47)
ρ0
Since G > 0 and Yp > 0 (due to hardening effect), it can be seen from Eq. (7.41) that
Gp /G < 1, and then consequently we obtain
p
CL = CLe
This indicates that, same as in the situation of 1D stress waves, during the propagation of a
1D strain elasto-plastic wave, a faster precursor elastic wave propagates ahead, and then is
followed by a slower plastic wave. Analyses of the whole problem could follow the same
way that was used in analyses of 1D stress waves, including the conservation conditions
across the wave front as discussed in Section 2.7. Those will be no longer repeated here.
However, there are some points to be specially noticed:
(1) For a perfect plastic material, the plastic portion of its stress–strain curve in the 1D
stress condition is a horizontal line σX = Y (Fig. 7.3). So, there is no one-to-one
corresponding, single-valued function relationship between stress and strain. This
violates the second assumption of the rate-independent stress–strain relationship
[Eq. (2.14)] required in establishing the rate-independent theory of stress waves
(Chapter 2). Otherwise, we will deduce that the velocity of a plastic wave is zero,
or no propagation of plastic deformation. However, in the 1D strain condition,
the rate-independent theory of stress waves can still be adopted for perfect plastic
materials. This is because although the distortional δX –εX curve is graphically a
sX, SX
sX
Y0
2Y SX
3 0
2E E
3
0 eX
Fig. 7.3. Stress–strain relation of perfect plastic material in 1D stress condition.

sX SX
1-ν Y K
1-2ν 0
2Y
3 0
K+43 G 4G
3
0 eX 0 eX
(a) (b)
Fig. 7.4. Stress–strain relation of perfect plastic material in 1D strain condition.
horizontal line [Fig. 7.4(b)], the axial stress σX now is a one-to-one corresponding
single-valued function of the axial strain εX , with a linear relation for its plastic
portion whose slope depends on the value of K, as shown in Fig. 7.4(a). Thus,
formally it takes the same form as the linear hardening stress–strain relationship in
the 1D stress condition.
p
(2) Generally, the plastic wave velocity CL in 1D strain condition is a function of
p
plastic strain εX or plastic work Wp [Eqs. (7.47) and (7.41)]. For linear hardening
and perfect plastic materials, the plastic wave velocity is a constant, and if denoted
p
by CL1 and CK respectively, then we have

p 1 4GEY
CL1 = K+ (7.48)
ρ0 3(3G + EY )

K
CK = (7.49)
ρ0
CK is also called the body wave velocity.

(3) In the 1D stress condition, the plastic wave velocity CL is much smaller than
the elastic wave velocity C0 , about one order less. For example, the ratio EL /E
of metals is about 0.01, then we obtain (CL /C0 ) ≈ 0.1. While in the 1D strain
condition, the plastic wave velocity is relatively high, its lower limit is the body wave
velocity CK under the perfect plastic assumption. Since K ≈ E, the body wave
velocity approaches the velocity of elastic longitudinal wave in bars, CK ≈ C0 .
For example, if ν = 1/3, then it can be seen from Eqs. (7.16) √ and (7.11) that
K = E, G = 3E/8, and consequently CK = C0 , CLe = 2/3C0 ≈ 1.22C0 ,
or (CK /CLe ) ≈ 0.82.
The above discussions are for the elastic–plastic loading process.

During the unloading process, first, there is elastic unloading, and according to Eq. (7.44)
the elastic unloading disturbances propagate with CLe , the elastic longitudinal wave veloc-
p
ity in infinite media. Because CLe > CL , namely an elastic unloading wave propagates
faster than a plastic loading wave, then similar to those in 1D stress condition, there
exist the pursuing unloading and the head-on unloading, etc., depending on different
p
actual situations. However, because CLe is not much larger than CL , during the pursuing
unloading, the plastic loading wave in the 1D strain condition attenuates slower than
that in the 1D stress condition. As to the propagation of elastic–plastic boundaries, the
methods used for the 1D stress waves (Section 4.10) can be similarly adopted here. But
the difference in yield criterion between those two conditions and the possibility of the
appearance of reverse plastic loading boundaries should be noticed (Wang et al., 1983).
Comparing with the 1D stress waves, one unique feature of the 1D strain waves is the
propagation of reverse plastic loading waves when σX satisfies the reverse yield criterion
[Eq. (7.32)] even σX has not unloaded to zero yet. During the reverse plastic loading,
p
since CL < CLe , namely the latter formed reverse plastic wave propagates slower than the
earlier formed elastic unloading wave, then the distance between these two wave fronts
increases with wave propagation.
7.5 Influence of Reverse Yield on the Propagation of 1D Strain Elastic–Plastic Wave
As an example, let us consider a semi-infinite media subjected to a rectangular impulse

(Wang, 1982). Initially, the media is in the natural state. At t = 0, the surface of the
media (X = 0) is suddenly subjected to a uniformly distributed, constant pressure σX∗ ;
after t = T , it is suddenly unloaded to zero, as shown in Fig. 7.5.
Thus, a loading dual-wave formation first occurs at t = T , namely, a strong discontinu-

ous elastic 1D strain precursory wave and a strong discontinuous plastic 1D strain wave
simultaneously propagate from X = 0 with constant velocity CLe and CK ; respectively.
They are graphically represented by the line OA and OB in Fig. 7.5(a), respectively. The
constant-valued zones 0, 1, and 2 in Fig. 7.5(a) correspond to the points 0, 1, and 2 on the
σX –vX plane in Fig. 7.5(b). It is easy to determine
σX0 = vX0 = 0
σX1 YH
σX1 = −YH , vX1 = − =−
ρ0 CLe ρ0 CLe

1 CK
σX1 = σX∗ , vX2 = − 1 YH − σ ∗ ∗
X = vX
ρ0 CK CLe
since they are totally similar to those for the linear hardening plastic materials in the 1D
stress condition.
However, from the beginning of unloading the situation is totally different. At t = T , when
the stresses on the surface boundary suddenly unload to zero, an unloading dual-wave
formation occurs, namely, not only a strong discontinuous elastic 1D strain unload-
ing wave TB propagates with a constant velocity CLe , but also a strong discontinuous
t t
30' 30
2627'27'' 29
25 28 19
G 24 19''
20 23 18
21 22 16' 16 tress)
F 17 (Tensile S
14'' 15 A5
13 14' B''
11'' 11' 12' 12
E
10 B' A4
C''
8 8
9'' 9' A3
7
C' 6
D 6'' 6' A2
5' 5
4 C
B A1
3
2
T
1 A
0
sX sX* O XC XB X
(a)
sX
sX
1310 0(7) 4 2930

14 12 vX 28 19
11 18
9 24 22
8 3 16
27 17
15
6
2326
20 13(21) 10 0(7)
5 25 vX
-YH 14
1
11 12
s ∗X
2
(b)
Fig. 7.5. The influence of reverse yield on the propagation of 1D strain wave.
plastic 1D strain reverse loading wave TC simultaneously propagates with a constant

velocity CK . The constant-valued zones 3 and 4 behind these two wave fronts correspond
to the points 3 and 4 on the σX –vX plane. The states in these two zones can be determined
by using the compatibility conditions across discontinuous wave front and the reverse yield
condition:
2YH
σX3 = σX∗ + 2YH , ∗
vX3 = vX −
ρ0 CLe
σX3
σX4 = 0, vX4 = vX3 +
ρ0 CK
For the instant before the strong discontinuous elastic unloading wave TB catches up with
the positive plastic loading wave OB, the wave profiles of σX –X and εX –X are shown as
the solid line and the dashed line respectively in Fig. 7.6(a). During the wave propagating
process, the distance between the positive plastic loading wave and the positive elastic
loading wave increases, while the distance between the elastic unloading wave and the
–sX –eX 2
Stress distribution
Strain distribution
e p
Cl Cl
1 –eX ~X
–sX ~X e
Cl
3
4 p
O Cl
X
(a) t < tB
1
5 e
e Cl
Cl
5'
3 5
4 p
O Cl
XB X
(b) tB <t<tC
1
5 Cl
e
6'' 6' e
Cl 5'
6 5
e
Cl
O 4
XC XB X
(c) tC <t<tD
Fig. 7.6. Distribution of stress (solid lines) and strain (dashed lines) at various times.
positive plastic loading wave decreases. At the time t = tB , TB catches up with OB

and they interact with each other (pursuing unloading). For convenience, assume that
the unloading residual particle velocity vX3 is not too high which satisfy the following
condition
σX3 − σX1
vX3 < vX1 +
ρ0 CLe
in other words, if
⎛ p ⎞
CL
⎜ 1 + 5 Ce ⎟
⎜ ⎟
σX∗ ≥ − ⎜ L
p ⎟ YH
⎝ CL ⎠
1+ e
CL
then the positive loading wave vanishes after their interaction, and a stationary strong
discontinuous strain interface is formed at the point B. The rightward wave from the
point B is an elastic unloading wave BA1 , while the leftward wave from the point B is an
internal reflective elastic wave BC. Note that with regard to the elastic unloading zone 3
the leftward wave BC is an elastic loading wave, reloading the media from σX3 to σX5 ;
while with regard to the reverse plastic loading wave TC, BC is an elastic unloading wave
(head-on unloading wave). In fact, at the time t = tC , BC and TC meet and interact with
each other, and consequently a stationary strong discontinuous strain interface is formed
at C. The rightward wave from the point C is an elastic internal reflective unloading wave
CA2 , further unloading the media from σX5 to σX6 . The states in the zones 5 and 6 and in
the other further formed zones as well, can be obviously determined by the compatibility
conditions across the wave fronts and the continuity requirement that the stresses and
particle velocities across the interfaces where waves interact, should be respectively equal
to each other, as mentioned previously. The results obtained are shown in Fig. 7.5. It is
not difficult for the readers to understand these results by comparing the characteristics
in (X, t) plane and the characteristics in (σX , vX ) plane. However, the following points
should be emphasized:
(1) Corresponding to B and C, two stationary strong discontinuous interfaces with

respect to strain are formed at XB and XC . It should be noted that they are of different
characters. The former is with regard to the positive plastic loading, namely there
is a jump of positive plastic strain at XB ; the strain at left side is larger than that at
right side, as shown in Fig. 7.6(b). The latter is with respect to the reverse plastic
loading, namely there is a jump of reverse plastic strain at XC ; the strain at left side
is smaller than that at right side, as shown in Fig. 7.6(c).
(2) Since the media on the left side of the stationary strong discontinuous strain interface
XC have experienced reverse plastic loading up to the value of σX4 = 0 (zone 4),
their reverse succeeded-yield stress due to strain-hardening is zero, then in the
interval OXC , the disturbance σX = 0 propagating rightward from the unloading
surface, such as DC, is an elastic wave. On the other hand, in the interval XC XB on
the right side of the discontinuous interface XC , the media only experienced posi-
tive plastic loading up to the value of σX2 = σX∗ , then their reverse succeeded-yield
stress is σX3 = σ ∗ + 2YH . Consequently, any unloading disturbance whose abso-

lute value is less than |σX3 | will lead to reverse plastic loading. Therefore, when
DC is incident on to the stationary discontinuous strain interface XC , similar to
the situation when wave reflection and transmission occur when an elastic wave
propagates from a material with high yield limit into a material with low yield limit
(assuming the two materials have identical sound resistance), a rightward reverse
plastic loading wave C B will be transmitted and a leftward elastic unloading wave
will be reflected. On this analogy, all the rightward waves reflected from the unload-
ing surface are elastic waves in the interval OX C , but when they are incident on
to XC , they will transmit reverse plastic loading waves, only if the stationary dis-
continuous interface XC exists. Thereby, the intensity of reverse plastic loading
waves on the right side of XC successively increases as the zone 9 , 11 , and 14
successively appear in Fig. 7.5 (the absolute value of |σX | successively decreases
and approaches zero).
(3) Since the media in the interval XC XB on the left side of the stationary strong
discontinuous strain interface XB experienced positive plastic loading up to the
value of σX2 = σX∗ , while the media on the right side of XB only experienced elastic
deformation, their reverse succeeded-yield stress is σX3 = σX∗ + 2YH on the left
side but is still YH on the right side of XB . Consequently, when a rightward reverse
plastic loading wave, such as C B , propagates to XB , similar to the situation when
a plastic wave propagates from a material with low yield limit into a material with
high yield limit (assuming the two materials have identical sound resistance) and the
amplitude of the plastic wave is lower than the yield limit of the latter material, there
must exist a reflection of a further loaded plastic wave and a transmission of only
an elastic wave, then a leftward reverse plastic loading wave B E will be reflected
and a rightward elastic unloading wave B A4 will be transmitted, where the word
unloading is relative to the original positive loading. Thus it can be seen that the
reverse plastic wave propagation ends at the stationary strong discontinuous strain
interface XB , unless the intensity of the rightward incident reverse plastic loading
wave is higher than the YH (tensile) so that the reverse plastic loading could take
place on the right side of XB .
(4) When the leftward reverse plastic loading wave B E reflected from XB meets the
rightward reverse plastic loading wave C E transmitted across XC , a further reverse
plastic loading must take place (head-on loading), same as the intensifying effect
resulting from the head-on loading of two waves with the same sign. Consequently
the tensile stress zone could be formed. Moreover, on the one hand, because of
the interaction of the leftward internal reflective tensile waves with a series of
successively intensified rightward reverse plastic loading waves transmitted across
XC , enable the magnitude of tensile stress to increase (as in the zone 17). On the other
hand, the reflective loading must take place when the rightward internal reflective
waves arrive at the stationary discontinuous strain interface XB , enable the tensile
stress to increase further (as in the zone 16). Therefore, for the reverse plastic
loading zones in the interval XC XB , the magnitude of tensile stress in those zones
will be successively larger and larger. Meanwhile, the magnitude of tensile stress
in the reverse elastic loading zones (namely the tensile elastic unloading zones)
formed correspondingly on the right side of XB will also be successively larger and
larger.
(5) Since the media on the left side of the stationary strong discontinuous strain interface
XC only experienced reverse plastic loading up to the value of σX4 = 0 (zone 4),
when the reverse plastic loading waves (tensile stress waves) developed in the
interval XC XB are leftward incident on to XC , similar to the situation when a plastic
wave propagates from a material with high yield limit into a material with low yield
limit (assuming the two materials have identical sound resistance), there will be
leftward transmission of an elastic wave and a reverse plastic loading (tensile stress)
wave (as in the zones 20 and 22). Along with the media on the left side of XC getting
into the tensile state of reverse plastic loading, the stationary strong discontinuous
strain interface XC disappears. The afterwards leftward reverse plastic loading
waves with higher and higher tensile stresses (as in the zone 28) are not influenced
by the disappearance of XC . Finally, the tensile stress zone will be extended to the
entire interval OX B .
(6) If before the leftward plastic loading wave (tensile stress wave) FG is incident on to
the unloading surface (X = 0), namely at t < tG , the unloading boundary condition
(σX = 0) in fact plays a role to promote the development of reverse plastic loading,
then after that it contrarily plays an unloading role to the reverse plastic loading,
and then the problem can be treated as the same as what was discussed previously
for the unloading problem of 1D stress plastic waves.
Thus it can be seen, for the problems of one-dimensional strain elastic plastic wave propa-
gation, only because of the reverse yield effect, the tensile stress zone can be formed near
the surface where the impulsive compressive load is subjected to, and the correspond-
ing particle velocity becomes negative. Such a situation neither exists in the problems
of one-dimensional stress wave propagation discussed previously, nor exists in the prob-
lems of one-dimensional strain wave propagation based on the hydrodynamic model to
be discussed in the next section. Once the tensile stress is larger than the critical tensile
stress of the concerned material, the so-called front spalling may occur. This may be of
significance in the study of dynamic fracture, particularly for the brittle materials with
lower tensile strength or/and internal pre-cracks.
7.6 State Equation of Solids under High Pressures
The discussions of 1D strain elasto-plastic waves in the above two sections are based on
the fact that the elastic deformation of materials follows the Hooke’s law. It is clear that
it is valid only when strain is small. In problems of high pressure and larger deformation,
elastic moduli are not constants any more, but the functions of strain or stress. Then, the
elastic components of constitutive equation are nonlinear.
Under high pressure, the distortional resistance (shearing strength) of solids can be
neglected approximately. Then, the distortion part of constitutive relation can be neglected,
and only the volumetric part should be considered. In other words, the constitutive rela-
tion is now simplified to a relationship between the hydrostatic pressure p and the volume
strain ∆ or the specific volume V; which is sometimes called the state equation of solids
under high pressure. It is equivalent to regarding the solid under high pressure as the
compressible nonviscous fluid.
In this section, the state equation of solids under high pressure, obeying the nonlinear
volumetric law, will be discussed. In fact, state equations are not only valid in the studies
of shock waves in solids under high pressure when shear strength can be neglected, but
also valid in studies of nonlinear elasto-plastic waves in solids whose shear strength cannot
be neglected.
With regard to the volume strain, according to its engineering definition, if denoted by ∆,
is expressed as
V − V0
∆= = (1 + εX )(1 + εY )(1 + εZ ) − 1 (7.50)
V0
where V0 is the initial specific volume (V0 = 1/ρ0 ), then the corresponding volumetric
modulus is:
dp dp
K=− = −V0 (7.51)
d∆ dV
This is depicted in Lagrange system, and so is called Lagrange volumetric modulus.
? then it
While following the definition of logarithmic strain (true strain), if denoted by ∆,
is expressed as
? = ln V = ε̃X + ε̃Y +?
∆ εZ , ?
ε = ln(1 + ε) (7.52)
V0
then the corresponding volumetric modulus is
dp dp dp
k=− =− = −V (7.53)
?
d∆ d ln V dV
It is depicted in Euler system, and so is called Euler volumetric modulus. Obviously,

there exists a relationship between k and K
V
k= K = (1 + ∆)K (7.54)
V0
In the 1D strain condition, Eqs. (7.51), (7.53), and (7.54) can be reduced respectively to:
dp dp
K=− , k=− (7.55a)
dεX d ε̃
k = (1 + εX )K (7.55b)
Bridgman equation The studies on state equation of solids under high pressure originated
from the experimental studies on the relationship between the volumetric modulus and the
hydrostatic pressure under static high pressure. Bridgman (1949) studied such relationships
for several tens of elements and compounds under the range of static high pressure p about
104 –105 bars (1–10 GPa). Based on his experimental results, he proposed an empirical
equation:
V0 − V
= −∆ = ap − bp 2 (7.56)
V
where a and b are material constants. This equation is usually called the Bridgman
equation or the isothermal state equation of solids. For most materials tested, when
p is in the unit of bars, a is in the order of 10−6 –10−7 bar−1 and b is in the order of
10−12 bar−2 . For example, based on the data obtained from more than 400 tests for iron
at temperature 24◦ C, there is:
V0 − V
= −∆ = 5.826 × 10−7 p − 0.80 × 10−12 p 2
V
From Eq. (7.56), Lagrange volumetric modulus K as a function of hydrostatic pressure p

can be obtained:
1 1
K(p) = = (7.57)
a − 2bp 2b
a 1− p
a
It is clear that K increases with the increase of p, namely the p–V curve is a concave curve
as shown in Fig. 7.7. This indicates that the resistance of volumetric compression of solid
materials increases with the increase of compressive deformation; in other words, it is
more difficult to compress a solid as its volume is more compressed.
As b/a is in the order of 10−5 –10−6 bar−1 , Eq. (7.57) shows that a pressure in the order
of 103 –104 bars (0.1–1 GPa) is needed at least to obtain a change of 1% in K. This shows
why the nonlinearity of elastic volumetric law is considered only in problems under high
pressure. When 2bp/a < l, extending Eq. (7.57) into a power series and neglecting the
– = ap–bp2
=–
ap
1
o a
v
Fig. 7.7. The p–V curve according to the Bridgman equation [Eq. (7.56)].
small quantities of high-orders, we obtain

1 2b
K(p) ≈ 1+ p (7.58)
a a
Obviously, 1/a is equivalent to the volumetric modulus K0 of the linear elastic volumetric
law under low pressure; while b is the coefficient depicting the change of K with p when
K is approximated by a linear function of p.
The functions in other forms are also used to describe the relationship between p and V;
and the material coefficients in these functions are determined by fitting the data form
Bridgman’s experiments. For instance, based on the theory of solid physics, Pack et al.
(1948) suggested the following p–V relationship:
2 / % 1 &0 2
V0 3 V 3
p=α exp β 1 − −1 (7.59a)
V V0
In the 1D strain condition, as V = (1 + εX )V0 , the above equation can be written as:
2
1

p = α(1 + εX )− 3 exp β 1 − (1 + εX ) 2 − 1 (7.59b)
The values of α and β for several materials were determined based on the Bridgman’s
experimental data and are listed in Table 7.2 (Broberg, 1956).
Murnagham equation Now let us start from Euler volumetric modulus defined by
Eq. (7.52) to discuss the relationship between k and p. If, similar to Eq. (7.58), we assume
that k is a linear function of p, namely
k = k0 (1 + ηp) (7.60)
where k0 and η are material constants, then the p–V relationship can be written as an
ordinary differential equation
dp
−V = k0 + np
dV
where n = k0 η. From initial condition V |p = 0 = V0 , we obtain
n
1 V 1
p+ = (constant) (7.61a)
η V0 η
Table 7.2. The values of α and β for several materials [Eq. (7.59b)].
Materials Iron Aluminum Magnesium Copper Lead Silex
ρ0 (103 kg/m3 ) 7.86 2.72 1.74 8.93 11.34 2.21
α (105 bar) 6.01 1.90 0.639 3.52 1.32 4.7
β 8.4 11.4 15.6 11.7 10.0 2.0
or rewritten as
n
k0 V0
p= −1 (7.61b)
n V
In the 1D strain condition, it is reduced to
k0 " # k0
p= (1 + εX )−n − 1 = {exp(−nε̃X ) − 1} (7.61c)
n n
It should be noticed that Eq. (7.61a) in form is similar to the following isentropic equation
for ideal gases
γ
V
p = constant
V0
And the n in Eq. (7.61a) is similar to the adiabatic exponent γ in the above equation for
ideal gases. For metals, n typically takes 4. Equation (7.61) is usually called Murnagham
equation or the isentropic state equation of solids, where material constant K0 and n
can be obtained by experiments on wave propagation.
The Murnagham equation was originated from research on the elastic theory of finite defor-
mation by performing quite complicated mathematical calculations. As shown above, it is
actually equivalent to the assumption that k is proportional to p [Eq. (7.60)], namely Euler
volumetric modulus increase linearly with the increase of hydrostatic pressure. Obviously,
the p–V curve depicted by Eq. (7.61) is also a concave curve, formally similar to that in
Fig. 7.7. Or in the 1D strain compressive loading condition, the corresponding axial stress–
strain curves σX –εX is a concave curve as shown in Fig. 7.8, representing the nonlinear
elastic volumetric law.
–sX
(p)
o –eX
k0
n
Fig. 7.8. The σX –εX curve corresponding to the Murnagham equation [Eq. (7.61)] in the 1D strain
condition.
Grüneisen equation The Bridgman equation and the Murnagham equation depict the
relationship between p and V in an isothermal process and an isentropic process respec-
tively. Thereby, they are the state equations valid only under certain thermodynamics
condition, and are insufficient to depict the relationships among state parameters under
general thermodynamics conditions involving changes in temperature T and entropy S.
In other words, it is inadequate to represent the state equations of solids under general
conditions by a relationship relating only two state parameters p and V, and so the relation-
ships among p, V, and other thermodynamics parameters must be considered. For example,
we can use a series of Bridgman equations under different temperatures (a series of isother-
mal p–V curves) or a series of Murnagham equations under different entropies (a series
of isentropic p–V curves) to describe the state equations under general conditions. These
lead to respectively the so-called thermal equations of state:
fT (p, V , T ) = 0 (7.62)
and the so-called entropic equations of state:
fS (p, V , S) = 0 (7.63)
They have been usually adopted in the studies on fluid mechanics. Geographically, they
are depicted by a surface in a there dimensional space of p, V, T or a surface in a three
dimensional space of p, V, S respectively. It is clear that the Bridgman equation represents
only the intersection curve of the surface of Eq. (7.62) with a plane of constant temperature;
while the Murnagham equation only represents the intersection curve of the surface of
Eq. (7.63) and a plane of constant entropy. From the slope of these intersection curves
respectively, the isothermal volumetric modulus k can be determined by

∂p
kT = −V (7.64)
∂V T
and the isentropic volumetric modulus kS can be determined by

∂p
kS = −V (7.65)
∂V S
In previous discussions on linear elastic waves, the difference between kT and kS has not
been emphasized.
However, for the research on shock waves in solids under high pressure, neither Eqs. (7.62)
nor (7.63) is sufficient. Because from the equations based on the conservation of mass,
momentum, and energy across a shock wave front (refer to Section 2.7), it can be
understood that if the distortion of materials is neglected then the parameters in these
conservation equations only contain the hydrostatic pressure p, the specific volume V, and
the internal energy E (also refer to Section 7.7), while the temperature T and the entropy S
are not concerned directly. Therefore, it is convenient to write a state equation in internal
energy form which relates p, V, and E:
fE (p, V , E) = 0 (7.66a)
Or equivalently, p can be written as a function of E and V
p = p(E, V ) (7.66b)
The problem now is what the actual function for the state equations of solids in internal
energy form will be.
To conduct a further discussion, similar to defining the volumetric modulus by Eqs. (7.63)
or (7.64), we first introduce a new variable Γ based on Eq. (7.66b):

∂p
Γ≡V (7.67)
∂E V
which represents the change rate of the pressure with the internal energy change in unit
volume under isometric condition, and is usually named Grüneisen coefficient.
Considering the first and second thermodynamics law, we have
dE = TdS − pdV (7.68)
Consequently, in an isometric process, there is

∂E ∂S
=T ≡ CV (7.69)
∂T V ∂T V
where CV is the isometric specific heat. On the other hand, since it is an isometric process,
there should be

∂V ∂V
dV = dp + dT = 0
∂p T ∂T p
then we obtain

∂p ∂p ∂V ∂p 1 ∂V
=− = −V · = kT α (7.70)
∂T V ∂V T ∂T p ∂V T V ∂T p
!
where α ≡ (1/V )(∂V /∂T )p is the isobaric expansion coefficient. Therefore, from
Eqs. (7.69) and (7.70), we obtain

∂p
∂T V kT αV
Γ=V = (7.71)
∂E CV
∂T V
This is a basic formula to determine Grüneisen coefficient by three thermodynamics

parameters kT , α, and CV , which are easy to measure.
Assume that Γ is a function of V only:
Γ = γ (V ) (7.72)
it is named Grüneisen assumption. Integrating Eq. (7.67), we obtain

γ
p= E + Π(V ), (7.73a)
V
or rewrite it as
γ
p = pK (V ) + {E − EK (V )} (7.73b)
V
This is the well-known Mie–Grüneisen state equation, commonly used in internal energy
form in the studies of stress waves in solids under high pressure, up till now.
To highlight the physical meaning of EK (V ), integrate Eq. (7.69), to obtain

T
E = EK (V ) + CV dT = EK (V ) + ET (V , T ) (7.74)
0
This indicates that the internal energy E consists of two parts: the cold energy or the
energy of cold compression EK (V ) independent of temperature and the thermal energy
ET (V , T ) dependent on temperature. The former is the internal energy at 0◦ K, also called
the lattice potential energy (elastic energy), which includes the interaction energy among
ions or neutral molecules (lattice binding energy), the energy of “zero-point oscillation”
and the energy of compression of the gas of valence electrons (in a metal), all independent
of temperature. The latter is also called lattice kinetic energy, which includes the thermal
vibration energy of lattices and the thermal excitation energy of electrons, all dependent
on temperature.
To highlight the physical meaning of pK (V ), integrate Eq. (7.70), to obtain

T T ΓCV
p = pK (V ) + kT αdT = pK (V ) + dT = pK (V ) + pT (V , T ) (7.75)
0 0 V
This indicates that pressure p also consists of two parts: the cold pressure pK (V ) inde-
pendent of temperature and the thermal pressure pT (V , T ) dependent on temperature,
which correspond to the cold energy EK and the thermal energy ET , respectively. In fact,
applying Eq. (7.68) at 0◦ K and noting that the entropy keeps zero at 0◦ K (Nernst theorem),
we then have
dE K (V )
pK (V ) = − (7.76)
dV
while according to Eqs. (7.68) and (7.75), the thermal pressure pT can be expressed as:

∂ET T ΓCV
pT (V , T ) = − = dT
∂V S 0 V
according to the Grüneisen assumption [Eq. (7.72)], then we have

T
γ γ
pT (V , T ) = CV dT = ET (V , T ) (7.77)
V 0 T
This indicates that the ratio of the thermal pressure pT to the thermal energy ET is inde-
pendent of temperature and is a function only of specific volume, γ (V )/V . This is the right
physical meaning of Grüneisen assumption, and is equivalent to the assumption of quasi-
harmonic vibration of lattices in the theory of thermal vibration of lattices. In other words,
it is assumed that the vibration frequencies of lattices are functions of atomic spacing (i.e.
specific volume) and the vibration can be treated as harmonic at each specific volume.
Equation (7.77) is actually the Grüneisen equation. In fact, if pT and ET in Eq. (7.77) are
replaced by (p − pK ) and (E − EK ) respectively, then Eq. (7.73b) is obtained.
7.7 Shock Waves in Solids under High Pressures
Under such high pressure where the shearing strength of solids can be neglected, as
discussed previously, solids can be treated as non-viscous compressible fluids. Then, a
number of research results of isentropic waves and shock waves in hydrodynamics field
can be directly generalized to the studies of stress waves in solids under high pressure;
only the differences in the state equations between solids and gases should be considered.
Such a kind of approximation is usually called the hydrodynamics approximation.
Common solid materials under high pressure show such character that they become more
difficult to be compressed with increase in the volumetric compression, namely volumetric
modulus increases with the increase of pressure. This is schemed in Figs. 7.7 and 7.8
qualitatively. Thereby, similar to the discussions in Section 2.6, it can be understood that
the strong discontinuous shock wave will be formed in the propagation of compression
disturbances in the solids under high pressure, while the weak discontinuous rarefaction
wave will be formed in the propagation of expansion disturbances.
The propagation of rarefaction waves is similar to the propagation of nonlinear elasto-

plastic waves in decreasingly hardening materials discussed previously and so will not
be discussed here again. In the following, only the problems related to the shock wave
propagating in solids under high pressure will be discussed in detail.
7.7.1 Shock jump conditions
In Section 2.7, by using Lagrange variables we discussed the conditions of conservation of

mass, momentum, and energy across a shock wave front in the 1D stress condition, namely
the so-called shock jump conditions. Here, we will extend those discussions to the 1D
strain problems, and the expressions depicted by Euler variables will also be presented.
Lagrange formulation In the material coordinate system, consider a 1D strain plane shock
wave propagating with the material wave velocity D in the X-axis direction. Assume that
the initial state ahead of the wave front and the final state behind the wave front are all in
the thermodynamic equilibrium, then the conservation of mass, momentum, and energy
across the wave front of a shock wave can be expressed in the same form as Eqs. (2.55),
(2.57), and (2.61) or (2.62), and are summarized below:
[v] = −D [ε]
[σ ] = −ρ0 D [v]
[σ v] 1 2 ! 1 −
[e] = − − v = σ + σ + [ε]
ρ0 D 2 2ρ0
Here, we only need to regard the σ , ε, and v in these equations as the X-axial stress, strain,
and particle velocity in the 1D strain condition.
For solids under high pressure, according to the hydrodynamics approximation, the terms
related to the distortion can be neglected. This is equivalent to replacing the σ by (−p)
and the ε by (V0 − V )/V0 in the above three equations. Moreover, consistent with the
traditional notations used in the discussion of shock waves in solids under high pressure,
we change the notation for the particle velocity by u and for the internal energy by E, and
consequently the above equations are reduced to
[u] = −ρ0 D [V ] (7.78a)

[p] = ρ0 D [u] (7.78b)
[pu] !
[E] = − 1
u2 = − 12 p − + p + [V ] (7.78c)
ρ0 D 2
These are Lagrange formulations of the so-called shock jump conditions or the Rankine–
Hugoniot relations (R–H relations for short) of the plane shock waves in solids under
high pressure.
Euler formulation Now we discuss the same problem in the space coordinate system.
Consider a plane shock wave that propagates with the spatial wave velocity D along the
x-axis in the space coordinates, all the state parameters ahead of the wave front are denoted
with superscript “+” while those behind the wave front are denoted with superscript “−”,
u− u+=0
p− p+
1
V −= ρ− 1
V += ρ+
E − E +
T− T+
S− U = ( dx
dt )W S+
x
Shock wave front
Fig. 7.9. A plane shock wave in a space coordinate system.

First, noticing that the particle velocity relative to the wave front is (u − D), according
to the mass conservation, the mass flow rate, m, passing through a wave front should be
conservative, namely

ρ − (u− − D) = ρ + u+ − D = m
or
[ρu] = D[ρ] (7.79a)
Second, according to the conservation of momentum, there should be

ρ − u− (u− − D) − ρ + u+ u+ − D = p+ − p −
or by using the above mass conservation, we have
mu− + p − = mu+ + p +
or

[p] = −m[u] = ρ + D − u+ [u] (7.79b)
Third, according the energy conservation, there should be

2
ρ− u− + E − (u− − D) − ρ + 12 u+ + E + u+ − D = p+ u+ − p − u−
1 2
2
Or by using the above momentum conservation, we have

2
u− + E − + p − u− = m 12 u+ + E + + p + u+
1 2
m 2
or
[pu] 1 2 ! [pu] 1 !
[E] = − − u = + +
− u2 (7.79c)
m 2 ρ D−u 2
If also using the momentum conservation [Eq. (7.79b)], the above equation can be written
in a form independent of the coordinate system,

[E] = − 12 p + + p − [V ] (7.79c )
which is consistent with Eq. (7.78c). Equations (7.79) are then Euler formulation of the
Rankine–Hugoniot relations in the space coordinate system.
If the coordinate system moves with the particles ahead of a shock wave, then the relative
spatial wave velocity U of the shock wave, relative to the particles ahead, is
U = D − u+ (7.80)
Substituting this into Eq. (7.79) and noticing u∗ = 0 in the present coordinate system,
then Euler formulation of the Rankine–Hugoniot relations in such cases are
u− = −ρ + U [V ] (7.81a)
[p] = ρ + U u− (7.81b)
ρ − u− 2
[E] = − 1
u− (7.81c)
ρ+U 2
Equations (7.78), (7.79), and (7.81) are equivalent to each other. They are different
expression forms of the same R–H relations in the different coordinate systems. It should
be noticed that even in the cases of 1D strain plane shock waves where the distortion is
not negligible, since there still exists εX = (V0 − V )/V0 , these relations are still valid,
only need to regard the p in those relations as the axial stress (−σX ).
From these three equation sets, it can be seen that the material wave velocity D , the relative
spatial wave velocity U, and the absolute spatial wave velocity D of a shock wave can be
expressed respectively by

[p]
D = V0 − (7.82a)
[V ]

[p]
U = V+ − (7.82b)
[V ]

[p]
D = u+ + V + − (7.82c)
[V ]
Among these, there obviously exist the following relations
V+
D = U + u+ = · D + u+ (7.83)
V0
This is actually a concrete expression of Eq. (2.11) in the present situation. If the state
ahead of the shock wave is in an uncompressed natural state (V + = V0 ), then U = D ; if
moreover it is in a static state (u+ = 0), then D = U = D . Only in this condition, these
three wave velocities are identical.
When the magnitude of jump of the state parameters across a wave front approaches infin-
ity, and the discontinuity with the differential under the isentropic condition is replaced,
then D , U, and D in Eq. (7.82) are transformed to the corresponding wave velocities of
a rarefaction wave (weak continuous wave), namely the material wave velocity C, the
relative spatial wave velocity a, and the absolute spatial wave velocity c, respectively:

∂p ρ ∂p
C = C0 − = (7.84a)
∂V S ρ0 ∂ρ S

∂p ∂p
a=V − = (7.84b)
∂V S ∂ρ S

∂p ∂p
c =u+V − =u+ (7.84c)
∂V S ∂ρ S
Be reminded, the above three forms of R–H relations are all deduced for the rightward
wave. For the leftward wave, the sign of wave velocity should be reversed, similar to that
discussed previously for the strong discontinuous interfaces in bars.
Therefore, there are totally four equations, the three conservation equations for the shock
jump conditions, together with a state equation in the internal energy form to characterize
material properties. If the initial conditions, namely the initial states before shock loading
are given, then there are five unknowns p, V (or ρ), u, E, and the shock wave velocity
D (or U, D) to characterize the final states but only by four equations. Once any of the
unknowns is given from the boundary conditions, then the other four unknowns can be
determined. Therefore, for a certain material (of which the state equation is known), under
the given initial and boundary conditions, the problem of plane shock wave propagation
is solvable.
For instance, consider the problem that two plane plates A and B collide at high velocity.
There are shock waves generated simultaneously in the two plates, while the state parame-
ters at the impact interface are indeterminate, as shown in Fig. 7.10. Here, there are totally
ten unknowns, five each for the plate A and the plate B. We can list four equations each
for the plate A and the plate B, together with the continuity conditions of particle velocity
pA, VA, uA, EA pB, VB, uB, EB

ᏰB
ᏰA
pA0, VA0, uA0, EA0 pB, VB, uB, EB

X
Material A Material B
Fig. 7.10. Propagation of shock waves during the collision between two planes A and B.
and pressure, respectively, at the interface:
uA = uB , pA = pB
then there are totally ten equations for ten unknowns. Therefore the problem is solvable.
Of course, it should be noticed here, the shock wave velocity should take different sign
when the R–H relations [Eq. (7.78)] or its other equivalent forms) are applied for rightward
waves and leftward waves, respectively.
7.7.2 Shock adiabat
If no concrete boundary conditions are specified, in certain initial equilibrium states, the
R–H relations together with the state equation provide the relationship between any two
of the five unknowns, such as p–V, p–u, etc. The relationships between any couple of
unknowns obtained in this way are named shock adiabat, or Rankine–Hugoniot curve,
or briefly Hugoniot curves. Because the shock jump process across the shock wave front
is an irreversible nonequilibrium process, so the shock adiabat actually only represents
the locus of the possible final equilibrium states through a shock jump started at a certain
initial equilibrium state (called the center point of Hugoniot); it does not represent the locus
of the successive states that the material experienced in the process. This is completely
different with those curves in the equilibrium states, such as the isothermal curve which
represents the relationship between the state parameters in an isothermal process, or the
isentropic adiabatic curve which represents the relationship between the state parameters
in an isentropic adiabatic process, or the quasi-equilibrium adiabatic stress–strain curve of
a nonincreasingly hardening material discussed previously in the 1D stress elastic–plastic
waves. In addition, since Hugoniot curves depend on the initial equilibrium states, so
even for the same material, different Hugoniot curves exist corresponding to different
initial states. The problem that the shock jump process is a nonequilibrium process with
additional entropy-increase will be discussed further, besides it has been discussed a little
in Section 2.7.
Among various types of Hugoniot curves, the following three types are commonly used:
the p–V (or p–ρ) curves, the p–u curves, and the U–u curves.
The p–V Hugoniot curves From the condition of energy conservation [Eq. (7.78c)] and
the state equation in internal energy form [Eq. (7.66)], eliminating the internal energy E,
we obtain the p–V relationship in the shock adiabatic condition, namely the p–V Hugoniot
curves. The shape of p–V Hugoniot curves for ordinary materials are concave upwards.
Figure 7.11 shows the typical p–V Hugoniot curves of materials determined experimentally
by shock wave tests (Jones, 1972).
The chord connecting the start point A and the final point B of a p–V Hugoniot curve, as
shown in Fig. 7.12, is of specific significance, called the Rayleigh line. In fact, it is known
from Eq. (7.82) that the velocity of a shock wave depends on the slope of this Rayleigh
line. Moreover, it is known from Eq. (7.78c) that the area below a Rayleigh line represents
the jump of internal energy across the shock wave front. It should be noticed that the
material velocity D is completely determined by the slope of the Rayleigh line for a given
W
Fe
4
3
p, M bar
Al
2 Mg
PM
MA
0
0.4 0.6 0.8 1.0
V
V0
Fig. 7.11. The p–V Hugoniot curves of some materials.
Rayleigh line
Tangent at point B
Hugoniot
A
V
V0
Fig. 7.12. p–V Hugoniot curve and the corresponding Rayleigh line.
material (namely its density is given), but the special velocity additionally depends on the
initial density of material, ρ + , ahead of the shock wave front.
For a stable shock wave, since every part of wave front propagates with an identical
wave velocity and the wave velocity depends on the slope of the Rayleigh line, this
indicates that the locus of all states that the shock jump experiences is just the Rayleigh
line. In the previous content, it has been emphasized that a Hugoniot curve is a locus of
the possible final equilibrium states corresponding to a certain initial equilibrium state,
but not a locus of the states experienced in a shock jump process. Now we are clear
that the locus of the states experienced in a shock jump process is just the Rayleigh line.
Certainly, the states of the points on a Rayleigh line except the initial one and the final one
are not at thermodynamic equilibrium. In a shock jump process, if the p–V relationship
representing an equilibrium process can be approximated by a Hugoniot curve, then the
difference of pressure between a Hugoniot curve and the corresponding Rayleigh line
for a given specific volume approximately represents the nonequilibrium force, namely
the irreversible viscous force (dissipation force) which results in the irreversible entropy-
increase in a shock jump. Based on such a mechanism of dissipation, a stable shock wave
is realized by adjusting the actual profile shape of shock wave front to make the sum of the
equilibrium forces and the nonequilibrium forces just equal to the corresponding pressure
of Rayleigh line.
Let us discuss the p–V Hugoniot curve from another angle to illustrate again that a
shock jump process is a process having irreversible entropy-increase. Differentiating
Eq. (7.78c), and denoting the initial state by subscript 0 and final state without subscript,
we obtain
dE = 12 (V − V0 )dp − 12 (p + p0 )dV
This shows the differential relationship between two neighboring possible final points
on a p–V Hugoniot curve. On the other hand, since any final state is at thermodynamic
equilibrium, the first and second law of thermodynamics [Eq. (7.68)] should also be
satisfied. Eliminating dE from these two equations, we obtain
⎧ ⎫
p − p0 ⎪
⎪
⎪ ⎪
1⎨
V − V0
⎬
TdS = 2 (V0 − V )dp + 2 (p − p0 )dV =
1 1
1− (V0 − V )dp (7.85)
⎪
2⎪ dp ⎪⎪
⎩ − ⎭
dV
Referring to Fig. 7.12, it is known that the (p–p0 )/(V0 − V ) is the slope of the Rayleigh
line; while the (−dp/dV ) is the tangent slope of the p–V Hugoniot curve at the final state B.
Since the p–V Hugoniot curve for a common material is concave upwards, it means

p − p0
V −V
0 <1
dp
−
dV
then Eq. (7.85) shows that the entropy increases with the increase of pressure along the
Hugoniot curves, dS/dp > 0. Even under the abnormal condition that the Hugoniot curve
is bent back as shown by the dashed line in Fig. 7.12, because dp/dV > 0, we still obtain
from Eq. (7.85) the conclusion that dS/dp > 0.
Now, let us discuss the relationship between the Hugoniot curve, the isentropic adiabatic
curve, and the isothermal curve. If we conduct a total differential of p = p(S, V ) with
respect to V along the p–V Hugoniot curve, then the relationship between the slope dp/dV
of the Hugoniot curve and the slope (∂p/∂V )S of the isentropic p–V curve is obtained:

dp ∂p ∂p dS
= + (7.86)
dV ∂V S ∂S V dV
According to Eqs. (7.69) and (7.70), in the normal cases, we have

∂p ∂p ∂T kT αT
= = >0
∂S V ∂T V ∂S V CV
Since Eq. (7.85) indicates that dS/dV and dp/dV have the same sign during a compressive
process (V < V0 ), then Eq. (7.86) indicates that if dp/dV < 0, then

dp ∂p
− >−
dV ∂V S
This means that the Hugoniot curve AB is above the isentropic curve AS1 passing through
the initial point A; while it is below the isentropic curve BC passing through the final point
B, as shown in Fig. 7.13. If dp/dV > 0, then the difference between the slope of the
Hugoniot curve and the slope of the corresponding isentropic curve further increases, this
conclusion is still valid. In addition, since the area below an isentropic p–V curve represents
B Rayleigh chord
Isentropic line
+ Hugoniot
Iso
the
rm
al
o -
A C V
V0
Fig. 7.13. The Hugoniot curve, Rayleigh line, isentropic curve, and isothermal curve.
the reversible change of internal energy during an isentropic process [see Eq. (7.68)], then
the area encompassed by the Rayleigh line AB and the expansion isentropic curve BC
(as the shadowed area shown in Fig. 7.13) represents the irreversible dissipation of energy
during the process of shock jump, which corresponds to the irreversible entropy-increase
given by Eq. (7.85).
As for the relation between the slope (∂p/∂V )S of an isentropic curve and the slope
(∂p/∂V )T of an isothermal curve, conducting a total differentiation of p = p(V , T ) with
respect to V under the isentropic condition, we have

∂p ∂p ∂p dT
= + (7.87)
∂V S ∂V T ∂T V dV S
Noting that for an isothermal process, there is

∂T ∂T
dT = dV + dS = 0
∂V S ∂S V
and considering Eq. (7.69) and the following Maxwell formula1

∂p ∂S
=
∂T V ∂V T
we obtain

∂T ∂T ∂S T ∂p
− = =
∂V S ∂S V ∂V T CV ∂T V
Thereby, Eq. (7.87) can be rewritten as

2
∂p ∂p T ∂p
− =− +
∂V S ∂V T CV ∂T V
Since T > 0, CV > 0, the above equation indicates

∂p ∂p
− >−
∂V S ∂V T
which means that an isentropic curve is above an isothermal curve (Fig. 7.13).
1 Introducing the Helmholtz free energy A = E − T S, then the first and second thermodynamics law [Eq. (7.68)]
is converted to dA = −SdT − pdV; from this, we have S = −(∂A/∂T )V and p = −(∂A/∂V )T , and
consequently (∂p/∂T )V = −∂ 2 A/∂T ∂V = (∂S/∂V )T .

dS
Since at the initial state (point A), according to Eq. (7.85) there is dV A = 0, then according
to Eq. (7.86) there is

dp ∂p
=
dV A ∂V S A
namely, the slope of a Hugoniot curve and the slope of the corresponding isentropic curve
are identical at the initial state point A. Moreover, after a total differentiation of Eq. (7.86)
and the successive mathematical operations, it can be proved that
2
d 2 p ∂ p
=
dV A2 ∂V 2 S A
3
d 3 p ∂ p

=
dV A3 ∂V 3 S A
namely, the curvature of a Hugoniot curve and the curvature of the corresponding isentropic
curve are also identical, and the difference between these two curves only appear in their
third-order derivatives.
In the above discussion, the influence of the phase-transition of materials on the Hugoniot
curve has not been taken into consideration, and it has been implicitly assumed that the
following relationship is valid,
p 2 − p1 p1 − p0
> (7.88)
V1 − V 2 V0 − V 1
where the subscript 0 denotes the initial state, the subscript 2 denotes the final state, while
the subscript 1 denotes a state between the initial state and the final state on a Hugoniot
curve. It is clear that a single stable shock wave front is formed if Eq. (7.88) is satisfied.
However, since shock waves are usually formed under high pressure and the irreversible
entropy-increase induced in a shock jump process must lead to a dramatic increase in
temperature, then the phase-transition may occur under such conditions of high pressure
and high temperature. Consequently, Eq. (7.88) is no longer satisfied. For example, it has
been found that when iron is under an impact pressure of 13 GPa, a phase-transition from
the phase α body-centred cubic (BCC) to the phase ε hexagonal close-packed (HCP) takes
place, etc.
In Fig. 7.14, if the point 1 is corresponding to a phase-transition, because of the different

material properties before and after the phase-transition, the state equations and the shock
adiabatic curves are correspondingly different. For the p–V Hugoniot curve, its phase-
transition may be displayed as a discontinuity in the curve shape itself (called the first-order
phase-transition) or a discontinuity in the slope of the curve (called the second-order
phase-transition).
As an example, let us discuss a second-order phase-transition as shown in Fig. 7.14. If the

shock pressure is not much larger than the phase-transition pressure p1 , such as if the final
shock state is corresponding to the point 2 in Fig. 7.14, then the absolute value of the slope
4
3
2
Second-order
1 phase-transition
0
o
V
V0
Fig. 7.14. The p–V Hugoniot curve with a phase-transition.
of Rayleigh line 1–2 is smaller than those of Rayleigh line 0–1; consequently, the shock
wave velocity in the high-pressure phase is smaller than that in the low-pressure phase.
Then, Eq. (7.88) is no longer valid, which means that a single shock wave is unstable and
a dual-wave construction will be formed. Only when the shock pressure is high enough
(e.g. as high as p4 at the point 4) so that the slope of Rayleigh line 0–4 is larger than
those of Rayleigh line 0–1, a single shock wave is then stable again. It is clear that
the point 3 in the figure corresponds to the critical condition for the transition from the
dual-wave construction to the stable one-wave structure, which satisfies the following
condition
p 3 − p1 p1 − p0
=
V1 − V 3 V0 − V 1
The critical pressure p3 (corresponding to the point 3) is called the overdrive pressure.
The influence of phase-transition on the propagation of shock waves has been utilized
as an important method to study the phase transition, particularly under high pressure.
On the other side, effects of phase-transition have been utilized to realize the required
phase-transition, such as to synthesize diamond from graphite, etc.
Although the above discussion was with respect to the Hugoniot curves in p–V form, it
in fact shows the common properties of Hugoniot curves in other forms, because we can
convert the Hugoniot curves in p–V form into other forms through the R–H relationship.
Therefore, in the following, the Hugoniot curves in other forms will not be discussed in
detail, but only some additional features will be briefly analyzed for the Hugoniot curves
in another two commonly used forms, namely the p–u and the U–u Hugoniot curves.
The p–u Hugoniot curves From the R–H relationship [Eq. (7.78b)], it is known that the
line connecting the initial point A and the final point B on the Hugoniot curve in p–u forms,
namely the slope [p]/[u] of the Rayleigh line represents the shock wave resistance ρ0 D ,
]
= [p
]
[u
Ᏸ
0
A ρ
o
u
Fig. 7.15. The p–u Hugoniot curve.
In the analyses of the high-velocity collision between two objects or the interaction of
two shock waves, and the reflection or transmission of shock waves propagating from one
material to another material, the continuity of p and u respectively at the interface must
be satisfied and will be utilized to solve the problem. Therefore, it is convenient to use
the Hugoniot curves in p–u forms, which is similar to the stressparticle velocity curve
(σ –v) in the theory of 1D stress elasto-plastic waves in bars. This point will be clear in
the discussion on the reflection or transmission of shock waves in the next section.
The U–u Hugoniot curves In view of the experimental investigation of shock waves
and the related measuring technique, the parameters with a dimension of velocity such
as the particle velocity u and the shock wave velocity U usually are much easy to be
directly measured. In fact, the measurements of these parameters can be decomposed into
the measurements of the distance and the time, and both are easy to conduct now. As to
the other parameters such as the pressure, specific volume and internal energy, etc., they
are relatively difficult to be measured directly in the dynamic conditions. Therefore, in the
experimental determination of Hugoniot curves and in the experimental studies of the state
equations of materials under high pressure by means of the measurements of the Hugoniot
curves, the U–u Hugoniot curve is most commonly used.
A number of experimental results for a lot of materials in a wide range of experimental

pressures show that if no phase-transition takes place then the U–u Hugoniot curves display
a linear relationship as shown in Fig. 7.16.
U = a0 + su (7.89a)
where a0 and s are material constants. The material which satisfies this relationship some-
times is named the “a, s material”. For some materials such as iron, if its experimental
U–u Hugoniot curve is a departure from the linear relationship, then an item of square of
+su
U=a 0
a0
o u
Fig. 7.16. The U–u Hugoniot curve.
u is added, namely
U = a0 + su + qu2 (7.89b)
where q is also a material constant.
Since the nonisentropic shock wave approaches the isentropic sound wave when the par-
ticle velocity u approaches zero, it seems that the a0 in Eq. (7.89) should represent the
sound velocity of materials. In fact, the data of a0 determined in the experiments of shock
waves are indeed much closer to the sound velocity of body wave (k0 /ρ0 )1/2 . For some
typical materials, the data of a0 and s determined by the experiments of shock waves are
shown in Table 7.3 (McQueen, et al., 1970). In addition, the density ρ0 and the Grüneisen
coefficient Γ0 under the normal pressure and temperature are also listed in the table. For
the metallic materials, the Γ0 is usually close to 2.
Table 7.3. Parameters of U–u Hugoniot curves for some typical materials [Eq. (7.89)].
Materials a0 (km/s) s ρ0 (g/cm3 ) Γ0
2024 Al 5.328 1.338 2.785 2.00
921-T Al 5.041 1.420 2.833 2.10
Copper 3.940 1.489 8.93 1.96
Iron (C: 0.18%)∗ 3.574 1.920 7.85 1.69
Stainless Steel 4.569 1.490 7.896 2.17
U–Mo Alloy 2.565 1.531 18.45 2.03
WC 4.920 1.339 15.02 1.50
Polyethylene (low density) 2.901 1.481 0.915 1.644
Polystyrene 2.746 1.319 1.044 1.18
Plexiglass 2.572 1.536 1.185 0.97
Epoxy 2.678 1.520 1.198 1.13
Durite 2.847 1.404 1.370 0.50
∗ For mild steel (C: 0.18%), U = a0 + su + qu2 , while q = −0.068.
If we introduce η = (V0 − V )/V0 , then the parameters u, p, U, E of the final state for the
“a, s materials” can be expressed explicitly by the material constant a0 , s, and η. Thus,
the mass conservation of the R–H relationship [Eq. (7.81a)] is written as
u
η=
U
Combining with Eq. (7.89), we obtain

a0 η
u= (7.90a)
1 − sη
a0
U= (7.90b)
1 − sη
while the momentum conservation [Eq. (7.81b)], when p0 = 0, is written as
ρa02 η
p= (7.90c)
(1 − sη)2
and energy conservation [Eq. (7.81c)], when E0 = 0, is written as
a02 η2
E= (7.90d)
2 (1 − sη)2
Sometimes, the following shock Mach number MS is introduced
U
MS =
a0
then from the R–H relationship [Eq. (7.81)] and the linear U–u Hugoniot relationship
[Eq. (7.89)], we obtain
u Ms − 1
= (7.91a)
a0 s
ρ sMs
= (7.91b)
ρ0 1 + (s − 1) Ms
p − p0 Ms (Ms − 1)
2
= (7.91c)
ρ0 σ0 s
E − E0 (Ms − 1)2
= (7.91d)
a02 2s 2
Before ending the discussion on shock adiabat, it should be pointed out that there are two
types of problems in relation to the shock adiabat. One is to determine the Hugoniot curves
from the given state equations; the other is to determine the state equations from the given
Hugoniot curves. The former is a determinate problem. So far as the problems of stress
wave propagation in solids are discussed, the problems involved are mainly the problems
of this sort. The latter is an indeterminate problem, namely the problems of this sort can
be solved only if a concrete form of state equation has been assumed. When we study the
state equations of solids by means of shock wave experiments, the problems involved are
mainly the problems of this sort. In the case of Grüneisen state equations, the key is how to
determine the Grüneisen coefficients γ (V ), which is the most difficult to be determined.
Fortunately, for the analysis of shock waves the error due to the indetermination of γ (V )
is not large, and in the engineering application, the γ (V ) can be approximated by
γ (V )
= ρ0 Γ0 (constant) (7.92)
V
Or even γ = Γ0 (constant) in a first approximate analysis. The constant Γ0 can be deter-

mined by other known thermodynamics parameters according to Eq. (7.71). The data of
Γ0 for some typical materials are also listed in Table 7.3.
For the so-called weak shock waves whose intensity is rather weak, the shock jump can
be approximately regarded as an isentropic process, and then the corresponding state
equations of solids can be expressed in the form of Murnagham equation [Eq. (7.61)]
approximately:
n n
k0 V0 k0 ρ
p= −1 = −1
n V n ρ0
If we have obtained the linear U–u shock adiabat [Eq. (7.89)] from shock experiments, it
is easy to determine k0 and n from a0 and s. In fact, substituting the Murnagham equation
into the equation for the local sound velocity, namely the definition of the relative spatial
wave velocity a [Eq. (7.84b)], we obtain
n−1
∂p k0 ρ
a =
2
=
∂ρ s ρ0 ρ0
after differentiating, we have
2 a
da = dρ (7.93)
n−1 ρ
On the other hand, since the weak shock wave is approximately regarded as isentropic, then
the Riemann invariable across a weak shock wave is constant. This is equivalent to replac-
ing the jump terms [ ] in the conservation condition across a shock wave front [Eq. (7.79)]
by the differential terms, and correspondingly replacing the shock wave velocity D by the
weak discontinuous wave velocity c. Thus, from Eq. (7.79a) we obtain
dρ a
du = (c − u) = dρ
ρ ρ
Substituting this into Eq. (7.93) and integrating the equation, we have
n−1
a = a0 + (u − u0 )
2
where the a0 and u0 with the subscript 0 represent the initial state quantities ahead of
a shock wave respectively, while the symbols without a subscript denote the final state
quantities behind a shock wave. Consequently, considering this relationship, the absolute
spatial wave velocity D of weak shock waves can be expressed by
n+1 n−3
D = a0 + u0 = a + u = 12 (a + u + a0 + u0 ) = a0 + u+ u0
4 4
This indicates that theoretically a linear relationship between D and u can be deduced
according to the Murnagham equation. If u0 = 0, the above equation is reduced to
n+1
D = U = a0 + u (7.94)
4
Comparing with the linear U–u Hugoniot curves [Eq. (7.89)] obtained from the experi-
mental measurements, n and k0 can be determined by s and a0 according to the following
relationship

n = 4s − 1
(7.95)
k0 = ρ0 a02
7.8 Interaction of Shock Waves in Solids under High Pressure
The general principles to deal with the interaction, reflection, and transmission of shock
waves in solids under high pressure are similar to those used in analyzing the shock waves
in 1D bars discussed previously. The continuous requirement in pressure and particle
velocity respectively at the interface should be satisfied. However, it should be noticed:
(i) when the reflective wave is a reloading shock wave, its final state should be on a new
Hugoniot curve whose central point is the initial state point ahead of the reflected shock
wave. This new Hugoniot curve is different from the original Hugoniot curve whose
central point is the initial state point ahead of the incident shock wave; (ii) when the
reflective wave is an unloading expansion rarefaction wave, its states are determined by
the unloading isentropic curve, namely, the successive states experienced by the media
during the rarefaction wave passing through are all on the expansion isentropic curve
whose start point is the final state point behind the incident shock wave.
Indeed, as mentioned previously, if the state equation is known, the Hugoniot curve and the
isentropic curve passing through any point are determinate. In the case when the Grüneisen
state equation is applied, namely the Grüneisen coefficient γ (V ) is known, it is easy from
a known Hugoniot curve to figure out a new Hugoniot curve or a new isentropic curve
with an arbitrary point on the known Hugoniot curve as its new central point, as shown in
Fig. 7.17. This procedure includes the following steps.
p p
2
1
H
(i-1)
1 (i) Isentropic curve
Hugoniot
H 2
0 o 0
o
V2 V V
V0 V0 V0
(a) (b)
Fig. 7.17. Determining the loading Hugoniot curve (a) and the unloading isentropic curve (b) from
a known Hugoniot curve.
Assuming point 0 is the initial state point and point 1 is the final state point of an incidence
shock wave. The point 1 is also the initial state point of the reflected shock wave. For an
arbitrary point 2 on the Hugoniot curve of the reflected shock wave as shown in Fig. 7.17(a),
Eq. (7.78c) should be satisfied, namely,
E2 = E1 + 12 (p2 + p1 )(V1 − V2 ) (7.96)
Because both points 2 and H on the original Hugoniot curve having the same specific
volume should satisfy the Grüneisen equation [Eq. (7.73b)], there is
(p2 − pH )
E2 = EH + γ (7.97)
V 2
where subscript 2 and H denote the corresponding quantities at points 2 and H, respectively.
Eliminating E2 from the above two equations, we obtain
p2 − pH
E1 − EH = γ − 12 (p2 + p1 )(V1 − V2 ) (7.98)
V 2
On the other hand, both points 1 and H are the points on the original Hugoniot curve with
the point 0 being the central point. According to Eq. (7.78c), they should satisfy:
E1 − E0 = 12 (p1 + p0 )(V0 − V1 )
EH − E0 = 12 (pH + p0 )(V0 − V2 )
Eliminating E0 from the above two equations, we obtain
E1 − EH = 12 (p1 + p0 )(V0 − V1 ) − 12 (ph + p0 )(V0 − V2 ) (7.99)
Then eliminating (E1 − EH ) from Eqs. (7.98) and (7.99), and assuming p0 = 0, we obtain
γ (pH − p1 )(V0 − V2 )
pH − ·
p2 = V 2 2
γ (V − V ) (7.100)
1 2
1− ·
V 2 2
This is the p–V Hugoniot curve with point 1 as the central point, which is required in
analyzing the reflected shock waves.
For the isentropic curve passing through point 1, it can be determined in the following
way [Fig. 7.17(b)]: since along an isentropic curve there must be dS = 0, then from the
thermodynamics law [Eq. (7.68)], there should be
dE = −pdV
or expressed in a different form, we have
Ei = Ei−1 − 12 (pi + pi−1 ) ∆V (7.101)
On the other hand, both the point i and the point H on the original Hugoniot curve having
the identical specific volume Vi should satisfy the Grüneisen equation [Eq. (7.73b)], so
there is
pi − pH
Ei = EH + γ (7.102)
V i
Eliminating Ei from the above two equations, we obtain

γ ∆V

pH − pi−1 · + EH − Ei−1
V i 2
pi = γ ∆V (7.103)
1+ ·
V i 2
Therefore, the states at the point i can be determined from that at the point (i − 1) by a
numerical method, and other points can also be determined point by point in the same
way. Thus the expansion isentropic curve 1–2 is determined.
After the Hugoniot curves of reflected shock waves and the expansion isentropic curves
of reflected unloading waves can be determined, it is easy to deal with the interaction,
reflection, and transmission of shock waves. As mentioned previously, it is convenient
to analyze this sort of problem by using p–u curves. In the following, we highlight this
by analyzing the reflection and transmission of shock waves at the interface between two
materials.
First, we discuss the problem when a plane shock wave Si with pressure p1 is normally
incident from a material A with low wave impedance into a material B with high wave
impedance, as shown in Fig. 7.18. Assume that both materials were undisturbed initially,
which corresponds to the point 0 in the p–u plane. According to Eq. (7.78b), it is known
that the slope of the Rayleigh line connecting the initial state point 0 and the final state
point 1 represents the shock wave impedance (ρ0 D )A of material A. Since the impedances
of these two materials are different, reflection and transmission simultaneously occur
when an incident shock wave arrives at the interface of the two materials. Since the shock
wave impedance (ρ0 D )B of material B is larger than that (ρ0 D )A of material A, from the
p1,u1
Ᏸi
p0=u0=0
A B
Interface
p
Hugoniot A' 1
Hugoniot B
2
᏿r
HugoniotA
᏿t ) A
Ᏸ 1
B
(ρ 0
Ᏸ)
(0
᏿i
o
u
p2
p1 Ᏸr
Ᏸt
p0
A B
Interface
Fig. 7.18. Reflection and transmission of a shock wave which is incident into a high wave impedance
material from a low wave impedance material.
previous discussion on the reflection of elastic waves at the interface of different materials
(Section 3.5), it is clear that the reflected wave Sr is a shock wave that further loads
the material A from state 1 to state 2, and the transmitted wave St is also a shock wave
that compresses the material B from the undisturbed state 0 to state 2. According to the
continuity condition for p and u at the interface respectively, and noting the negative sign
of D in the R–H relationship [Eq. (7.78)] for leftward waves, then point 2 is the intersection
of the positive Hugoniot curve of material B, whose center point is point 0, and the negative
Hugoniot curve of material A, whose center point is point 1. Obviously, the intensity of
transmitted wave St is larger than that of incident wave Si . For instance, when a shock
wave with pressure amplitude of 24 GPa propagates from aluminum with low shock wave
impedance to iron with high shock wave impedance, the intensity of transmitted shock
wave is about 34 GPa.
Now, we discuss the problem that a rightward plane shock wave Si with pressure p1 is
normally incident from a material A with high wave impedance into a material B with
low wave impedance, as shown in Fig. 7.19. Similar to the previous discussion, it can be
deduced that an unloading reflection occurs at the interface of the two materials, namely,
the reflected wave Sr is a expansion rarefaction wave that unloads the material A from
state 1 to state 2 and the transmitted wave is a shock wave that loads the material B from
state 0 to state 2. Point 2 is the intersection point of the negative isentropic curve of material
A passing point 1 and the positive Hugoniot curve of material B, whose center point is
point 0. Obviously, the intensity of transmitted wave St is smaller than that of incident
wave Si . For instance, when a shock wave with pressure amplitude of 24 GPa propagates
from aluminum with high shock wave impedance to polyethylene with low shock wave
impedance, the intensity of transmitted shock wave is about 9.5 GPa.
Figure 7.20 schematically shows the positive Hugoniot curves for several common materi-
als, as well as the negative Hugoniot curves and the negative isentropic curves for several
typical explosives, the initial state point for both negative curves is the C–J (Chapman–
Jouget) detonation state (the dark points in the figure). The intersection point of any
positive curve and any negative curve determine the wave intensity at the interface when
the detonation wave of the corresponding explosive is normally incident into the relevant
material. If the intersection point is located at a position higher than the C–J point (namely
the pressure of the intersection point is higher than the C–J detonation pressure), the wave
reflected into the detonation-product is a shock wave; otherwise is a rarefaction wave.
As pointed out in the discussion on the reflection and transmission of elastic waves in 1D
bars (Section 3.5), the loading reflection at a rigid wall and the unloading reflection at a
free surface can be viewed as examples of transmission into materials with shock wave
impedance of ∞ and 0 respectively. However, due to the difference between the Hugoniot
curve and the isentropic curve as well as the difference between the Hugoniot curves with
different initial state points, the pressure amplitude of wave reflected at a rigid wall is no
longer twice that of incident wave, and the particle velocity of wave reflected at a free
surface is no longer twice that of incident wave.
For nonporous solid materials, especially metals, when the shock pressure is not very high
(e.g. in the order of 10 GPa for metals), the differences between Hugoniot curves and
isentropic curves can often be neglected. This is equivalent to the situation of weak shock
p1,u1
Di
p0=u0=0
A B
Interface
p
Hugoniot A
2
᏿r
Hugoniot B
᏿i ) B 2
Ᏸ
)A
(ρ 0
0Ᏸ
Isentropic A
(ρ
᏿t
o
u
p1
Cr p2
Ᏸt
p
A B
Interface
Fig. 7.19. Reflection and transmission of a shock wave which is incident into a low wave impedance
material from a high wave impedance material.
waves, for which the entropy-increase due to shock jump can be neglected. In such cases,
both the Hugoniot curve of reflected shock wave and the isentropic curve of reflected
rarefaction wave can be approximately regarded as the image of incident shock wave
about the perpendicular line ab passing the final state point (point 1) of the incident shock
wave, as shown in Fig. 7.21. Then the problem can be simplified, and consequently the
discussions on the reflection of shock waves either at the rigid wall or at the free surface
are consistent with the discussions for elastic waves.
It should be noticed that, for elastic waves, the wave impedance ρ0 C0 is a constant; while
for shock waves, the shock wave impedance ρ0 D increases with pressure. Therefore, the
relative value of wave impedance for two materials also changes with pressure. A special
0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0

p(k bar) Fe 600
Au U Cu
Brass
500 500
Ti
Al Gabbro
400 400
Mg
o Lucite
300 o Water 300
(alluvium)
Poly-
Cy ethylene
200 clo 200
o tol
Co
mp
B
100 TN 100
T
Polyur-
ethane
foam
0
0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
u(mm/µs)
Fig. 7.20. The positive Hugoniot curves for several materials, and the negative Hugoniot curves
and negative isentropic curves for several explosives.
p
a
o
b u
Fig. 7.21. Simplification for weak shock waves.
situation may even occur when the pressure is smaller than a critical pressure pK , namely
p < pK , the shock wave impedance of material A is larger than that of material B; while
when p > pK , the shock wave impedance of material A is smaller than that of material
B, as shown in Fig. 7.22, where the critical pressure pK is the intersection point of two
Hugoniot curves of the two materials.
To give an example, let us analyze the process of explosive expansion of an explo-

sive device with dispensing liquids (like an air fuel explosive bomb). For convenience,
B
p
A
pK
o
u
Fig. 7.22. Shock wave impedance increases with pressure.
C
A
E B
Fig. 7.23. Schematic explosive device for liquid dispensing.
the problem is approximately analyzed in the following by replacing spherical waves

with plane waves (Duvall, 1972b), and Lagrange coordinate system will be used here.
Figure 7.23 schemes the explosive device. The explosive region E (the center sphere) and
the liquid to be dispensed B are separated by the inner spherical shell A; all of them are
encased by the outer spherical shell C.
Assume that (ρ0 D )E > (ρ0 D )A > (ρ0 D )B < (ρ0 D )C . First, as shown in Fig. 7.24, when
the detonation wave whose state is denoted by point 1 arrives at shell A, an unload-
ing reflection occurs, whose state corresponds to point 2. Then, when the transmitted
shock wave in A arrives at the interface of A and B, an unloading reflection occurs again,
whose state corresponds to point 3. However, when the reflected wave returns to the
interface of A and E, a loading reflection occurs, whose state corresponds to point 4.
On this analogy, there are a series of leftward rarefaction waves reflected back and a
6 7 7
6 5
4 5
4 3
2 2 3
1
o 0 0 0
E A B X
p
A
1
2
B
4
A A 6
A 7
3 5
o
u
Fig. 7.24. Waves between A and liquid B.
series of rightward shock waves reflected forward in the inner shell A, respectively cor-
responding to the negative isentropic curves 2–3, 4–5, . . . , and the positive Hugoniot
curves 3–4, 5–6, . . .. Simultaneously there are a series of rightward waves transmit-
ted into the liquid B, corresponding to the positive Hugoniot curves 0–3, 3–5, 5–7, and
so on.
Second, as shown in Fig. 7.25, when the transmitted shock wave in liquid B arrives at
the outer shell C, a loading reflection occurs, corresponding to the negative Hugoniot
curve 1–2. Then, when the transmitted shock wave in C arrives at its free surface, an
unloading reflection occurs, corresponding to the negative isentropic curve 2–3. This
indicates that the expanding particle velocity of the outer shell increases, accelerating the
outer shell outward. However, when the reflected wave returns to the interface of B and
C, a rightward shock wave is reflected, corresponding to the positive Hugoniot curve 3–4.
On this analogy, there are a series of leftward rarefaction waves reflected back and a series
of rightward shock waves reflected forward in the outer shell C, respectively corresponding
to the negative isentropic curves 4–5, 6–7, . . . , and the positive Hugoniot curves 5–6,
7–8, . . .. With each reflection at the free surface, the expanding velocity of outer shell
increases once again, corresponding to the state 3, 5, 7 and so on. As a result, the outer
shell successively and acceleratively expands outward until the break of the outer shell,
and the liquid is spread to fog-like drops over a large volume in the order of thousands of
its original volume.
88
66 7
44 5
3
22
1
0 0
o
B C X
p
2 B
1
4 6 8
CC B
o
3 5 7 u
Fig. 7.25. Waves between liquid B and C.
7.9 Plane Waves in Hydro-Elasto-Plastic Media
Historically, the theory of nonlinear elasto-plastic waves was developed from two different
approaches, though the common conclusion was finally reached. On the one hand, based on
the theory of linear elastic waves under relative small loading, with the gradual increase of
loading, the plastic deformation is taken into consideration (Section 7.4), and furthermore
the influence of nonlinear elasticity is taken into consideration (Morland, 1959). This is
the approach extending the theory of linear elastic waves under lower stresses to the theory
of nonlinear elasto-plastic waves under higher stresses. On the other hand, based on the
theory of shock waves in solids under higher pressure neglecting its shear strength (Rice
et al., 1958), with the decrease of loading, the shear strength of materials is taken into
consideration. This is the approach extending the theory of shock waves of solids under
very high pressure (Section 7.7) to the theory of nonlinear elasto-plastic waves under sub-
high pressure. Although the two approaches are different, a common conclusion has been
reached.
First, we discuss the propagation of plane waves in nonlinear elasto-plastic media following
the first approach. Reviewing the discussion on the 1D strain elasto-plastic waves in
section 7.4, it was assumed that the elastic deformation follows Hooke’s law. Obviously,
this assumption is valid only in the problems involving small strain. In the cases of high
pressure and large deformation, the elastic moduli of materials are no longer invariables,
namely, the elastic constitutive relationship is actually nonlinear. However, if the nonlinear
elastic constitutive relationship is written in differential form, it is similar to the differential
form of Hooke’s law [Eq. (7.15)]
volumetric law: − dp = Kd∆

(7.104)
distortional law: dS ij = 2Gdeij
only that the K and G in these equations are generally the functions of strain or stress,
representing the local tangent slope of the corresponding stress–strain curve respectively.
Correspondingly, if the axial stress–strain relationship σX –εX in the 1D strain problems
is written in differential form, similar to Eq. (7.42), then we obtain
⎧
⎨ K + 4 G dεX , σX < YH
=
3
dσX = −dp + dS XX (7.105)
⎩ K + 4 Gp dεX , σX > YH
3
Noting εX = ∆ in the 1D strain problems, K and G generally are the functions of volumetric
strain ∆ or hydrostatic pressure p.
With regard to the nonlinear elastic volumetric modulus K(p), it has actually been discussed
in Section 7.6. If the state equation of solid under high pressure is given, then K(p) can be
determined. In the cases when the pressure is not high enough to neglect the shear strength
of solids, the Bridgman equation [Eq. (7.57)] or Murnagham equation [Eq. (7.60)] are
usually used to determine the nonlinear volumetric modulus for engineering application.
With regard to the nonlinear elastic shear modulus G(p), the related experimental data are
not as much as that of K(p). Based on the experimental data obtained from the supersonic
measurements under the pressure of around 1 GPa, (Voronov) and
(Vereshchakin)(1961) proposed, similar to Eq. (7.60), the following linear relationship:
G = G0 (1 + ηp) (7.106)
where η (=k0 n) is just the material parameter in Murnagham equation [Eq. (7.60)].
Without enough data, the influence of p on G is usually neglected and G is viewed as
a constant for engineering applications.
If the plastic shear modulus Gp is also determined [Eq. (7.41)], then the nonlinear elasto-
plastic stress–strain relationship in the 1D strain problems is completely determined. Based
on this, similar to the discussions in Section 7.4, the problems of nonlinear elasto-plastic
plane waves can be solved. Obviously, according to Eq. (7.7), both the nonlinear elastic
p
wave velocity CLe and the plastic wave velocity CL

K(p) + 43 G(p)
CLe = (7.107)
ρ0

p K + 43 Gp
CL = (7.108)
ρ0
–sX B
eX =K
–d X
(p)
s
–d
E
A
YH
o –eX
D
Fig. 7.26. Perfect plastic stress–strain curve in the 1D strain problems.
depend on the axial strain εX or volumetric strain ∆; or equivalently, vary with the
hydrostatic pressure p.
For perfect plastic materials, Gp = 0. In such cases, the stress–strain curves corresponding
to the compression loading and the expansion unloading are shown in Fig. 7.26. In the
figure, the dash-dotted curve OE corresponds to the nonlinear elastic volumetric law; the
solid curves OA and AB correspond to the elastic compression loading and the plastic
compression loading respectively, while the solid curves BC and CD correspond to the
elastic expansion unloading and the reverse plastic loading. In the situation of perfect
plasticity discussed here, if the pressure is not very high, the curves AB and CD can
be regarded as to keep a constant distance (2Y0 /3) in the σX -axis direction with the
hydrostatic pressure curve OE. Different from the stress–strain curve of linear elasto-
perfect plastic material (Fig. 7.2), during the compression loading in the present case, the
stress–strain curves of both the elastic
loading2 segment
OA and the plastic loading segment
AB are normally concave upward d 2 σX /dεX > 0 . Therefore, the corresponding elastic
shock wave and plastic shock waves are formed respectively. Their wave velocities are
determined by the tangent slope of Rayleigh line OA and OB respectively, namely the
elastic shock wave velocity De is

1 σX (A)
De = ·
ρ0 εX (A)
and the plastic shock wave velocity Dp is

1 σX (B) − σX (A)
Dp = ·
ρ0 εX (B) − εX (A)
In the unloading process, either for the elastic unloading segment BC or for the reverse
plastic loading segment CD, because their slope dσX /dεX decreases with the decrease
of σX , both the elastic unloading waves and the reverse plastic loading waves must be
the dispersive weak-discontinuous waves, and their wave velocities are determined by
Eqs. (7.107) and (7.108) respectively.
It should be pointed out that a process of shock jump is neither isothermal nor isentropic
but an irreversible adiabatic process with extra entropy increase. Thereby the segment
OA and AB which are corresponding to the formation of shock waves actually should
be the shock adiabatic curves taking account of distortion. Strictly, they are not parallel
to the isentropic hydrostatic pressure curve OE. In the process of unloading, the elastic
unloading segment BC is an isentropic adiabatic curve, while the reverse plastic loading
segment CD is an adiabatic curve paralleling the isentropic hydrostatic pressure curve OE
but having entropy increase in relation to the irreversible plastic deformation. Therefore,
even in the cases of perfect plasticity, the segment AB and the segment CD are strictly
not parallel. Only if the pressure is not very high, so that the differences among the
isothermal p–V curve, the isentropic p–V curve, and the shock adiabatic p–V curve are
small enough to be negligible, then the shock adiabatic curve and the isentropic adiabatic
curve can be viewed equivalent approximately. When their differences should be strictly
taken into consideration, the hydrodynamics approximation neglecting the distortion of
solids as discussed in Section7.7, should be extended to take account of distortion. This is
the second approach introduced at the beginning of this section. A further discussion will
be conducted below following this approach.
Reviewing the discussion on shock waves in solids under high pressure in Section 7.7, it
was assumed that the shear strength of solids can be neglected, and then materials can be
viewed as fluids. It is clear that this is valid only when the shock wave pressure is much
higher than the shear strength of materials. In fact, in the 1D strain problems, the axial
stress can be decomposed into two parts: the hydrostatic pressure p and the maximum
shear stress τ ,
σX = −p + sXX = −p + 23 (σX − σY ) = −p + 43 τ
The p increases unlimitedly with increase in the volumetric compression, and it can reach
the order of 103 –108 GPa in a wide range of practical problems. The τ increases with
increase in the shear deformation, and its limit is the shear strength of materials, in the
order of 10 Mpa–1 GPa for most engineering materials. Therefore, if the shock pressure is
two orders or more larger than the shear strength of materials, namely the ratio τ/p ≤ 0.01,
then the term of τ in the above equation can be neglected, and we have approximately
σX ≈ −p. Only under such condition, the hydrodynamics model provides satisfactory
approximate solutions. In contrast, when the shock pressure approaches the shear strength
of materials, the hydrodynamics approximation is not valid anymore, and the modification
should be made to take the shear strength of materials into consideration.
The modification actually includes two aspects:
(i) Since the state equation of solids under high pressure adopted in the hydrodynamics
approximation such as the state equation in the internal energy [Eq. (7.66)], only
depicts the nonlinear volumetric deformation law, it should now be replaced by
the nonlinear elasto-plastic constitutive relationship including both the volumetric
deformation law and the distortional deformation law. If it is assumed that the
plastic deformation contributes none to the volumetric deformation, then Eq. (7.66)
is still valid and an elasto-plastic distortion law should also be added. In the 1D
strain condition, according to Mises criteria or Tresca criteria, similar to Eq. (7.42),
the elasto-plastic distortion law in the differential form can be expressed as

⎧
dsXX ⎨ 43 G, |sXX | < 23 Y
= (7.109)
dεX ⎩ 4 G , |s | = 2 Y
3 p XX 3
Such nonlinear elasto-plastic constitutive model is usually called the hydro-elasto-

plastic model, and the media following such constitutive relationship are usually
called the hydro-elasto-plastic media. In the 1D strain condition, such consti-
tutive relationship can be written in the form of axial stress versus axial strain,
σx –εx , which is completely consistent with the Eq. (7.105) deduced following the
first approach mentioned previously. Thus it can be seen that the results obtained
following the two approaches mentioned above are exactly consistent.
(ii) The shock jump condition Eq. (7.78) deduced in the hydrodynamics approximation
should be correspondingly modified, namely the hydrostatic pressure (−p) in the
original equation now should be replaced by the axial stress σX .2
Therefore, the governing equation set of the plane shock waves in the hydro-elasto-plastic
media includes: three conservation conditions contained in the modified shock jump con-
ditions, two equations contained in the constitutive relationship, an equation connecting
σX , p, and sXX , an equation connecting the axial strain εX and the specific volume V. These
seven equations in total are listed below,
[u] = ρ0 D [V ] (7.110a)
[σX ] = −ρ0 D [u] (7.110b)
[E] = 12 (σX + σX0 ) [V ] (7.110c)
p = p (V , E) (7.110d)
⎧4
dsXX ⎨ 3 G, |sXX | < 23 Y
= (7.110e)
dεX ⎩4
3 Gp , |sXX | = 3 Y
2
σX = −p + sXX (7.110f )
V
εX = −1 (7.110g)
V0
When the initial states (the variables with subscript 0) ahead of the shock wave front are
known, there are eight unknowns in the above seven equations, namely variables σX , p,
sXX , εX , V, u, E, and the shock wave velocity D , which depicts the final states behind the
2 In the deduction of Eq. (7.78), it has been set that the positive sign of p corresponds to the pressure, and
correspondingly εX = 1 − V /V0 , namely the positive sign of εX temporarily corresponds to the compressive
strain. Here, consistent with the denotation used in other parts of this book, the positive sign of σX and εX
corresponds to the tensile stress and strain. So, please note the difference of signs in Eqs. (7.110) and (7.78).
shock wave front. If any unknown variable among these is given in the boundary condition,
the other seven unknown variables can be determined. In other words, the above seven
equations depict the relationships between any two of these unknown variables. These
relationships are the shock adiabatic curves or the Hugoniot curves in the case of hydro-
elasto-plastic model. It should be emphasized again, these curves only represent the loci
of the possible final equilibrium states through a shock jump started at a certain initial
equilibrium state; they do not represent the locus of the successive states that the material
experienced in the shock jump process.
Figure 7.27 shows three forms of Hugoniot curves: the σX – εX , the p–εX , and the sXX –εX ,
of which the initial state is the undisturbed state 0. The p − εX curve reflects the influence
of the volumetric law of solids on the shock jump process, the sXX − εX curve reflects
the influence of the distortion law of solids, and the σX –εX curve reflects the composite
influence of the volumetric law and the distortion law. The difference between the σX –
εX curve and the p − εX curve reflects the difference between a hydro-elasto-plastic
model and a hydrodynamics model. The point A in the figure corresponds to the Hugoniot
elastic limit, |σX (A)| = YH . If |σX | ≤ YH , then single elastic shock wave is formed. If
|σX | > YH , then a dual-wave structure is formed, namely a plastic shock wave follows
an elastic precursor shock wave. With increase in pressure, the plastic wave velocity
increases. When |σX | ≥ σX (K), a stable single shock wave is again formed; here, σX (K)
is the overdrive stress, which satisfies the condition that the slope of the line connecting the
point K and A just equals the slope of the Rayleigh line OA. If |σX | YH , the difference
between σX − εX curve and p − εX curve can be neglected, then the hydrodynamics
approximation can be applied.
Comparing with the analysis based on the hydrodynamics, an important feature of the
analysis based on the hydro-elasto-plastic theory is that the irreversible plastic deformation
is taken into consideration. Therefore, across a shock jump of a plastic shock wave, the
irreversible entropy-increase actually consists of two parts:
dS = dS q + dSp (7.111)
–sX
p
SXX
sX ~eX
p~eX
K
A ~e X
S XX
o
–eX
Fig. 7.27. Three forms of Hugoniot curves for hydro-elasto-plastic model.

–sX
B
Sq C
Td
Sp
Td
o –eX
D
Fig. 7.28. The entropy-increase dS q and dSp of a plastic shock wave in the case of hydro-elasto-
plastic model.
where dS q is the extra entropy-increase caused by the formation of a shock wave as

discussed in Section 7.7, which can approximately be represented by the area enclosed
by the σX –εX Hugoniot curve AB and the Rayleigh line AB, as shown in Fig. 7.28; dSp
is the entropy-increase caused by the irreversible plastic deformation (quasi-equilibrium),
namely,
p
TdS p = dW d (7.112)
p
where Wd denotes the plastic part of distortional work Wd . Since it is assumed that the
p
plastic deformation contributes none to the volumetric deformation, Wd is also the total
plastic work.
For the dS q , similar to the entropy-increase calculated by Eq. (7.85) based on the hydro-
dynamics approximation, in the present situation, the p and V in Eq. (7.85) are replaced
by σX and εX respectively, and so it can be calculated by
⎧ ⎫
σ X − σA ⎪
⎪
⎪
⎨ ⎪
⎬
εX − ε A
TdS q = 2 V0 (εX − εA )dσX · 1 −
1
(7.113)
⎪
⎪ dσX ⎪ ⎪
⎩ ⎭
dεX
p
As for the dSp , it can be calculated through the plastic distortional work dW d . In the case
of 1D strain, the calculation of the plastic work can be simplified by
dW d = V0 SXX dεX = 43 V0 τ dεX (7.114)

where the elastic part can be calculated according to the linear elastic distortional law
(SXX = 4GεX /3), we have
4 4 V0 τ
dWde = V0 · GεX dεX = dτ (7.115)
3 3 G
then the plastic part is

p 4 dτ
dW d = dW − dW ed = V0 τ dεX −
3 G
Substituting it into Eq. (7.112), we obtain

4 dτ
T dS p = V0 τ dεX − (7.116)
3 G
Under the condition of perfect plasticity, dτ ≡ 0, there is

p
T dS p = dW d = dW d = 43 V0 τ dεX
Table 7.4 lists the ratios of dS q and total entropy-increase dS under various pressures
(Duvall, 1972a). It can be seen that under high pressures the plastic deformation con-
tributes little to the entropy-increase. In this sense, the hydrodynamics approximation
is acceptable. However, under sub-high pressure or the so-called medium pressure, the
plastic deformation cannot be neglected. Since the plastic work causes energy dissipa-
tion, it is clear that the energy dissipation induced by the plastic shock wave propagation
based on the hydro-elasto-plastic analysis is larger, corresponding to a larger increase in
temperature.
Comparing with the analysis based on the hydrodynamics approximation, another impor-
tant feature of the hydro-elasto-plastic analysis lies in the process of unloading. If the
state before unloading corresponds to point B in Fig. 7.28, the unloading process begins
in an elastic unloading, so it proceeds first following the isentropic curve BC in the figure.
Correspondingly, the elastic unloading waves are continuous waves (rarefaction waves),
and their wave velocities CLe are determined by the tangent slope of the isentropic curve
BC [Eq. (7.107)]. Generally, the elastic unloading wave velocity CLe are larger than the
p
plastic loading wave velocity CL determined by the slope of Rayleigh line AB. Therefore,
the elastic unloading waves will catch up with the plastic shock waves ahead and interact
with them (pursuing unloading). With the decrease of σX , when the reverse yield condition
Table 7.4. The ratios of dS q and total entropy-increase dS under various pressures.
Aluminum Copper
Y (GPa) 0.72 1.74 2.30 0.51 1.28 2.53
σX (GPa) 9.0 20.2 37.5 16.6 41.3 81.6
dS q /dS 0.75 0.87 0.92 0.88 0.95 0.97
is satisfied (point C in the figure), the reverse plastic deformation occurs following the
curve CD in the figure, which is determined based on the isentropic expansion curve tak-
ing account of distortion. The reverse plastic loading waves are continuous waves, whose
wave velocities are determined by the tangent slope of CD [Eq. (7.108)] and generally
smaller than the elastic unloading wave velocities.
With regard to the determination of elasto-plastic boundaries, in principle, it is similar to

that discussed on the 1D stress elasto-plastic waves in Section 4.10. There are two points
that should be noticed. First, for the σX –εX curve, it should distinguish the loading σX –εX
curve and the unloading σX –εX curve. In the case of compressive loading when a shock
wave is formed, the corresponding σX –εX curve is a shock adiabatic curve; while in the
case of unloading, the σX –εX curve is either an isentropic adiabatic curve (during elastic
unloading), or a quasi-equilibrium adiabatic curve (during reverse plastic loading), which
is a curve based on the isentropic p–V curve but considering plastic deformation. And
they all change with different initial states. These make the determination of elasto-plastic
boundaries much more complicated.
Second, since the reverse plastic loading may occur in the unloading process, the reverse
plastic loading boundaries may appear. Moreover, the internal reflective elastic waves
which propagate back-and-forth between the positive plastic loading boundary and the
reverse plastic loading boundary, not only play an unloading role for the positive plas-
tic loading waves (pursuing unloading), but also play an unloading role for the reverse
plastic loading waves (head-on unloading). Thereby, the determination of the reverse plas-
tic loading boundaries and the determination of the positive plastic loading boundaries
are connected and dependent on each other. Such coupling character brings new prob-
lems to determine the elasto-plastic boundaries. This will become clear in the discussions
below.
7.10 Attenuation of Shock Waves in Hydro-Elasto-Plastic Media
Consider a semi-infinite hydro-elasto-plastic body originally at rest and natural state (no
initial stress/strain). At the time t = 0, on its surface (X = 0), it is suddenly subjected
to a constant load σ ∗ and then gradually unloaded, which is expressed by σ0 (τ ). We now
discuss how the shock waves attenuate due to the pursuing unloading disturbances (Wang
et al., 1983). For convenience, the subscript X of the axial stress σX is temporarily omitted
in this section.
As shown in Fig. 7.29, at the beginning of sudden loading, an elastic precursor shock wave
OH with an amplitude YH (Hugoniot elastic limit) and a plastic shock wave OA2 with an
initial amplitude of stress (σ ∗ −YH ) simultaneously propagate rightward. Since the plastic
shock wave is continuously unloaded by the pursuing unloading disturbances from the
beginning, its amplitude attenuates continuously and its wave velocity Dp also decreases
continuously. In the present example, the unloading boundary by chance coincides with the
propagating trajectory of the plastic shock wave. Then, the determination of the unloading
boundaries is actually the same problem as to determine how the plastic shock wave
attenuates due to the pursuing unloading disturbances.
t C1
)
(X
2
g
t=
ry
da
B2
un
bo
ng
di
A2
a
lo
tic
s
la
ep
)
1 (X
M4
rs
=g
ve
Re
C0
et
av
w
B1
ck
ho
Ce
s
tic
N
2
as
Q0 Cp
Pl
N A1 H
N Cp 0
N L2
1
B′0 M3
a M2
B0 m
b Ce L1
A′0
M Ce A0
P0 M1
s0 s* s0 (B0) 0 X
s0 (P0)
Fig. 7.29. Interaction between a positive plastic loading boundary and a reverse plastic loading
boundary.
When the surface load is unloaded to satisfy the condition of reverse yield (the point
B0 in Fig. 7.29), the reverse plastic loading boundary B0 B2 is formed. First, we discuss
the inter-influence between the development of the reverse plastic loading boundary and
the unloading boundary A0 A2 . For this purpose, we draw the rightward and leftward
characteristics in the elastic unloading zone through the beginning point B0 of the reverse
plastic loading boundary; they intersect the unloading boundary at A0 and A1 respectively.
And we further draw the successive characteristics A1 B1 , B1 A2 , A2 B2 , and so on in the
elastic unloading zone and the characteristics B1 Q0 in the reverse plastic loading zone.
The following points can be seen from Fig. 7.29 (1) The rightward characteristic MM 2
passing an arbitrary point M2 on the segment OA1 in the elastic unloading zone intersects
with the axis t, which indicates that the attenuation of shock wave on the segment OA1 is
determined by the surface unload condition for the duration OB0 . (2) the rightward char-
acteristic N0 M4 passing an arbitrary point M4 on the segment A1 A2 in elastic unloading
zone intersects with the segment B0 B1 of the reverse plastic loading boundary, which indi-
cates that the attenuation of unloading boundary A1 A2 or the attenuation of plastic shock
waves on the segment A1 A2 is determined by the states on the segment B0 B1 of reverse
loading boundary. (3) The leftward characteristic N2 L2 passing an arbitrary point N2 on
the segment B0 B1 in elastic unloading zone intersects with the segment A0 A1 and the
rightward characteristic NN 2 passing an arbitrary point N2 on the reverse plastic loading
zone intersects with the axis t, which indicates that the states on the segment B0 B1 are
determined by both the attenuation of plastic shock waves on the segment A0 A1 and the
surface unloading condition for the duration B0 Q0 .
Thus, in order to determine the segment A1 A2 of the unloading boundary, it is first needed
to determine the segment B0 B1 on the reverse plastic loading boundary. However in order
to determine the segment B0 B1 , it is again needed to determine the segment A0 A1 on
the unloading boundary. Therefore, the determination of the whole unloading boundary
or the whole propagating trajectory of the plastic shock waves OA0 A1 A2 A3 is a com-
plicated problem coupled with the determination of the reverse plastic loading boundary
OB0 B1 B2 B3 .
From the above discussion, it is also known that the whole problem can be dealt with
by dividing the whole elastic–plastic boundary into some segments, namely we can first
determine the segment OA1 independent of the reverse plastic loading boundary, then
we try to determine the segment B0 B1 ; after this, similar to the procedures discussed in
Chapter 4, by utilizing the conjugate relationship [Eq. (4.37)], from the known elasto-
plastic boundary, the corresponding conjugate boundary can be determined. In this war,
the segment A1 A2 on the unloading boundary, the segment B1 B2 on the reverse plastic
loading boundary, etc., can be determined alternately, and then the whole boundary can
be determined. Thus, the main problem remaining is how to determine the initial segment
OA1 of the unloading boundary and the initial segment B0 B1 of the reverse plastic loading
boundary.
In the actual treatment, the elastic unloading zone and the reverse plastic loading zone are
divided into some subzones, and assume that the characteristics in those subzones can be
regarded as straight lines respectively, namely assume that the elastic wave velocity CLe
p
and the reverse plastic loading wave velocity CL are constants in the subzones respec-
tively. Note that for an arbitrary point M2 (X2 , t2 ) on the segment OA1 , there must exist two
associated points connected by two associated characteristics, namely point M(0, τ ) con-
nected by the positive characteristics MM 2 and the so-called conjugate point M1 (X1 , t1 )
connected by the negative characteristics MM 1 . Then, similar to deducing the conjugate
relationship Eq. (4.37), among these three points, the following conjugate relationship can
be obtained:
µD (M2 ) + 1 σ (M2 ) − µD (M2 )YH = µD (M1 ) − 1 σ (M1 )

− µD (M1 )YH + 2σ0 (M) (7.117)
where µD = CLe /D(σ ) is a known function of σ for a given material. Equation (7.117)
shows that if the σ (M1 ) has been known, then σ (M2 ) is entirely determined by the
unloading surface condition σ0 (M).
For an arbitrary point M4 (X4 , t4 ) on the segment A1 A2 , there exist a relevant point
N0 (XN0 , tN0 ) on the reverse plastic loading boundary and a conjugate point M3 (X3 , t3 )
connected by the characteristics N0 M4 and N0 M3 , respectively. Similarly, the following
conjugate relationship can be obtained:
µD (M4 ) + 1 σ (M4 ) − µD (M4 )YH = µD (M3 ) − 1 σ (M3 )

− µD (M3 )YH + 2σB (M0 ) (7.118)
This equation shows that if the σ (M3 ) has been known, then the σ (M4 ) is entirely
determined by the stress σD (N0 ) at the relevant point N0 on the reverse plastic loading
boundary.
For an arbitrary point N2 on the segment B0 B1 , there exist four relevant points N, N1 , L1 ,
and L2 connected by the characteristics N N2 , N N1 , N2 L2 , and N1 L1 , respectively, and
the following conjugate relationship can be obtained similarly:

µp + 1 σ (N2 ) = 2µp σ0 (N ) − µp − 1 σ (N1 ) − µD (L2 ) − 1 σ (L2 )
(7.119)
+ {µD (L1 ) − 1} σ (L1 ) + {µD (L2 ) − µD (L1 )} YH
p
where µp = CLe /CL . The above equation shows that if σ (N1 ) has been known then σ (N2 )
is determined by the free surface condition, the stress σ (L1 ) and σ (L2 ) at the relevant
points L1 and L2 on the unloading boundary A0 A1 .
Thereby, if a small initial segment of OA1 and a small initial segment of B0 B1 are known,
using the conjugate relationships Eqs. (7.117)–(7.119), the whole unloading boundary and
reverse plastic loading boundary can be determined alternately segment by segment. As
for the determination of the initial segments of OA1 and B0 B1 , similar to the discussion in
Chapter 4, we can use the “successive approximations method” based on the conjugate
relationships Eqs. (7.117)–(7.119) or the “local linearization method” when the initial
propagating velocity of elasto-plastic boundary has been determined.
When the “successive approximations method” is used, it should be pointed out that,
although how the zero-order approximation is chosen does not affect the final result,
the number of iteration is affected. In the discussion of the attenuation of shock waves
in bars in Section 4.6, we took the surface load to be the instantaneous stress of shock
wave at each time as the zero-order approximation [Eq. (4.63)]. This is equivalent to
making a rigid unloading assumption (Ce → ∞) and the inertia of rigid body after
unloading is neglected. In the current situation of 1D strain plastic wave, because the ratio
of elastic wave velocity and the plastic wave velocity is not as large as that in the case
p
of 1D stress wave (CLe /CL ≈ 0.8 or larger), the error due to the rigid body assumption
obviously becomes larger, and the corresponding attenuation rate of plastic shock wave
will be overestimated. If considering that in the case of 1D strain wave the change of Dp
due to the attenuation of shock waves along OA1 is actually not large, so we can take
µD (M2 ) = µD (M1 ) approximately, then the conjugate relationship [Eq. (7.117)] can be
simplified to:

µD − 1 σ (M1 ) − σ0 (M)
σ (M2 ) = σ0 (M) 1 + · = σ0 (M) (1 + sD )
µD + 1 σ0 (M)
Since for the 1D strain elasto-plastic wave µD ≈ 1, and consequently sD 1, then a

zero-order assumption better than the rigid unloading assumption can be made:
σ (0) (M2 ) = σ0 (M) (7.120)
This is equivalent to that when the unloading disturbance catches up with the plastic shock
wave at the elastic wave velocity, the stress amplitude of the shock wave immediately
decreases to this value, and the internal reflection due to the interaction of these two
waves is neglected. By the same argument, for the reverse plastic loading boundary B0 B1 ,
when we approximately take µD (L1 ) = µD (L2 ), the conjugate relationship Eq. (7.119)
is simplified to:

µp − 1 σ (N1 ) − σ0 (N ) µD − 1 σ (L1 ) − σ (L2 )
σ (N2 ) = σ0 (N ) 1 − · − ·
µp + 1 σ0 (N ) µD + 1 σ0 (N )
= σ0 (N )(1 + Sp )
then similarly, the following zero-order approximation can be made:
σ (0) (N2 ) = σ0 (N ) (7.121)
The remaining actual steps of iteration computation are basically similar to those discussed
in Chapter 4. Only for the reverse plastic loading boundary t = g2 (X), once the stress
distribution along this boundary σB [X) = σ (X, g2 (X)] is approximately obtained by
iteration computation, the reverse yield condition
σB (X) = σA (X) − 2YH (7.122)
should be applied so that the trajectory of this boundary t = g2 (X) can be determined,
where σA (X) = σ (X, g1 [X)] is the stress distribution along the unloading boundary
t = g1 (X).
It should be pointed out that, in the case of µD ≈ 1 and µp ≈ 1, Eqs. (7.120) and (7.121)
as zero-order approximation often directly provide, although approximate but accurate
enough solutions, and no further iteration computation is needed. Such an approximation
is called the “method of simple waves in subzones”. This is equivalent to, when solving
the segment OA1 , the zone OA1B0 is approximately regarded as a zone of elastic simple
waves, and the internal reflection which is generated when the unloading disturbances
catch up with the plastic shock waves, is neglected; also when solving the segment B0 B1 ,
the zone B0 B1 Q0 is approximately regarded as a zone of plastic simple waves, and the
internal reflection which is generated when the reverse plastic loading disturbances head-
on meet the unloading disturbances, is neglected. On this analogy, for the segment A1 A2
on the unloading boundary, we approximately set
σ (M4 ) = σB (N0 )
and so on, which is equivalent to, when solving the segment A1 A2 , the zone A1 B0 B1 A2
is approximately regarded as a zone of elastic simple waves, and the relevant internal
reflection is neglected.
When the local linearization method is used, the initial propagating velocity Dp (σ ∗ ) of
the strong discontinuous unloading boundary can be easily determined by the slope of
Rayleigh line of σX –εX Hugoniot curve. Difficulties lie in the determination of the initial
propagating velocity C L of the weak discontinuous reverse plastic loading boundary. It
should be first noticed that, for the first-order weak discontinuous interface B0 B1 , same as
that shown in Fig. 4.35(a), the ∂σ/∂t which is given by the boundary condition for the state
just before the reverse yielding (denoted as ∂σ b/∂t) and the ∂σ/∂t which is given by the
boundary condition for the state just after the reverse yielding (denoted as ∂σ a /∂t) cannot
be simply regarded as ∂σ e /∂t and ∂σ p /∂t respectively, and substituted into Eq. (4.111) for
the determination of C L . Actually, ∂σ e /∂t ahead of the reverse plastic loading boundary
is ∂σ m /∂t in the zone m, which is to be solved. Thereby, according to Eq. (4.111), there
should be:
% &2
CL
∂σ a 1−
CLe
∂t =
∂σ m % &2
CL
∂t 1− p
CL
While along the elastic characteristics B0 A1 , there is:

m
∂σ m ∂σ b ∂σ ∂σ b
− = −CLe −
∂t ∂t ∂X ∂X
Meanwhile B0 B1 should also satisfy the reverse yield condition Eq. (7.122), of which the
differential form in the case of perfect plasticity is:
dσB (X) ∂σ m 1 ∂σ m dσA (X)

= + =
dX ∂X C L ∂t dX
Eliminating ∂σ m /∂X and ∂σ m /∂t from the above three equations, and noticing ∂σ b /∂t =
∂σ a /∂t at point B0 , the following formula about C L can be obtained,
2

1 A

A + A + 4 p 2 + e
2
CL CL
1
= (7.123)
CL 2
!
where A = (1/η) − (1/CLe ), η = ∂σ b /∂t / (dσA /dX) − ∂σ b /∂X . In actual com-
putation, to avoid the computation of ∂σ b /∂X, we try to express the η by other known
variables. Therefore using the compatibility relationships along the elastic characteristics
B0 B0 , B0 A0 , and B0 A0 , we obtain

1 CLe @ A
σ (B0 ) − σ (B0 ) = 1− σ (A0 ) − σ (A0 )
2 D (A0 )
while from the geometric relationship of quadrangle B0 B0 A0 A0 , it is known:

1 CLe @ A
X(B0 ) − X(B0 ) = 1+ X(A0 ) − X(A0 )
2 D (A0 )

Dividing the first equation above by the second one, and let B0 → B0 , then the relationship
dσ (B0 )
between the total differential dX of σ along the elastic characteristics B0 A1 at
B0 A1
point B0 and the total differential dσA (A0 )/dX of σ along the unloading boundary A0 A2
at point A0 is obtained

dσ (B0 ) µD − 1 dσA (A0 )
=−
dX B0 A1 µD + 1 dX
while according to the definition of total differential along the characteristics, there is

dσ ∂σ 1 ∂σ
= +
dX B0 A1 ∂X CLe ∂t
Substituting the above results into the definition of η in Eq. (7.123), then the η at point B0 is
dσA (0) µD (A0 ) − 1 dσA (A0 ) σ0 (B0 )

+ · +
1 dX µD (A0 ) + 1 dX CLe
= (7.124)
η(B0 ) σ0 (B0 )
where σ0 = dσ0 (t)/dt. If further we consider that the change of the slope of initial segment
OA0 of the unloading boundary is small, there is approximately µD = µs = constant,
and we consider that the initial unloading segment OP0 of the surface load can be dealt
with as local linear attenuation, namely approximately σ0 = constant, then referring to
Eq. (4.50) it is known:
dσA (0) dσA (A0 ) µ2 − 1 σ0 (0)

= = D ·
dX dX 2µD CLe
Substituting it into Eq. (7.124), we obtain:

1 1 1 σ0 (0) 1
= − e + (7.124 )
η(B0 ) D CL σ0 (B0 ) CLe
or according to the definition of A in Eq. (7.123), there is

1 1 σ0 (0)
A= − e
D CL σ0 (B0 )
For a given material, the CLe , CL , and D (σ ) are all known. Therefore, from the σ0 (0) at
p
point 0 and the σ0 (B0 ) at point B0 given by the surface unloading condition, and according
to Eqs. (7.123) and (7.124 ), the initial propagating velocity of the reverse plastic loading
boundary at point B0 , C L (B0 ), can be determined.
If the surface load σ0 (t) is a linearly attenuating function, then σ0 (0) = σ0 (B0 ), and
A = (1/D − 1/CLe ), therefore Eq. (7.123) is reduced to the result given by Lee and Liu
(1964):

1 2 % &2
1 1 1 2 1 1 4
− e + − e + p + − e
1 D CL D CL CL D C L C e
L
= (7.125)
CL 2
Once the initial propagating velocity of the elasto-plastic boundary is known, the whole
elasto-plastic boundary can be determined by using the corresponding conjugate relation-
ships according to the local linearization approximation method as discussed in Chapter 4.
The details will not be repeated here.
As an example, the computational results of attenuation of plastic shock waves in the

steel plate subjected to a contact explosion by using the above successive iteration method
are shown as the solid line in Fig. 7.30 (Wang et al., 1983). The computation iterations
were conducted until the calculated stress reached to an accuracy of three valid digitals.
The dotted line near the solid line are the results obtained by the “method of simple waves
in subzones” mentioned above, which almost coincide with the solid line with an error less
than 1%. The results calculated by the hydrodynamics approximation, namely neglecting
the shearing strength of material, are shown as the dashed line in the figure. Also the
results calculated by the finite difference numerical method based on the hydro-elasto-
plastic model are shown as the dash-dotted line in the figure (Chu et al., 1981). It is clear
that the result obtained by the approximation methods discussed in this section agrees better
with the result obtained by the finite difference numerical method than the result obtained
by the hydrodynamics approximation. It illustrates that the elasto-plastic distortion of
solid influences the shock waves to attenuate more quickly. In this example, the maximum
stress amplitude of plastic shock waves is 39.3 GPa and the Hugoniot elastic limit of the
sX
s0
1.0
o
o
0.9 o Hydrodynamics approximation
o o
o o o Simple-wave-approximate
o o
0.8 o Successive iteration approx. Hydro-elastic
o
-plastic model
0.7 Finite differential calculation

0 10 20 30 40 50 60 70 X(mm)
Fig. 7.30. Attenuation of plastic shock waves in steel plate under contact explosion.
material is 1.21 GPa, so the influence of shear strength of material on the attenuation of
shock waves is no longer negligible.
7.11 One-Dimensional Strain Elasto-Visco-Plastic Waves
Now, we will generalize the one-dimensional stress (Sokolovskii)–Malvern

elasto-visco-plastic theory discussed in Section 6.5 to the general 3D stress case, based on
the classical plastic-flow theory, and furthermore the 1D strain elasto-visco-plastic waves
will be discussed.
It is well known that the general form of the plastic flow theory (increment theory) is
expressed as
p ∂F
ε̇ij = λ (7.126)
∂σij
p p
where ε̇ij = ∂εij /∂t is the plastic strain rate, λ is the nonnegative scalar function, F is
the plastic potential function. In the frame of the so-called associate theory, F is regarded
as the yield function. If the Mises yield condition is applied, namely the plastic potential
function is in the form below:
F = J2 − K 2 (Wp ) = 0
then Eq. (7.126) is reduced to the form below:

p
ε̇ij = λsij (7.127)
the J2 in the above equations is the second invariable of stress deviator sij ,

J2 = 12 sij sij = 16 (σ1 − σ2 )2 + (σ2 − σ3 )2 + (σ3 − σ1 )2
the K (Wp ) is the strain hardening parameter of the material, a function of plastic work
Wp . For√the perfect plastic materials, K equals the yield limit in pure shear, k; while
k = Y0 / 3, where Y0 is the yield limit in simple tension.
p p
If introducing the second invariable I2 of the plastic strain rate tensor ε̇ij :
p p p
I2 = 12 ε̇ij ε̇ij
p
and defining the following plastic strain rate intensity ε̇i and stress intensity σi
respectively as3 :
<
p p
ε̇i = I2
;
σi = J2
3 The
√
σi defined here is different from the definition of Eq. (7.30) by a factor 3.
then from Eq. (7.127), we obtain

1 p p
2 ε̇ij ε̇ij = λ2 1
2 sij sij
therefore the λ can be expressed as:

<
p p
ε̇i I2
λ= = √
σi J2
and finally Eq. (7.127) is reduced to:
p p sij
ε̇ij = ε̇i √ (7.128)
J2
According to the (Sokolovskii)–Malvern model (Section 6.5), the inelastic

strain rate ε̇ p is a function of over-stress [σ − Y (Wp )]. Generalizing this to the 3D gen-
p
eral
√ situation, the inelastic strain rate intensity ε̇i should be a function of “over-stress”
J2 − K (Wp ) . Thereby, similar to Eq. (6.81), in the 3D situation, there is
= √ >
p J2
ε̇i =γ φ −1 (7.129)
K
where γ is the viscous coefficient. Thus, the constitutive relationship Eq. (7.128), when
the general plastic flow theory is extended to the general visco-plastic theory, becomes
= √ >
p J2 sij
ε̇ij =γ φ −1 √ (7.130)
K J2
p
If further we assume that the inelastic strain ε̇ij contributes none to the volumetric
deformation
p
ε̇kk = 0
p p p
eij = εij − 13 εkk δij = εij
where eij is deviator strain, then the general form of elasto-visco-plastic constitutive
relationship is:
⎧ 1
⎪
⎨ε̇kk = σ̇kk
3K = √ > (7.131)
⎪ 1 J2 sij
⎩ε̇ij = ε̇ije + ε̇ijp = ṡij + γ φ −1 √
2µ K J2
This constitutive relationship was first presented by Perzyna (1963) through a generaliza-
tion of the (Sokolovskii)–Malvern’s theory.
In the 1D strain condition, Eq. (7.131) is reduced to
⎧
⎪ 1
⎨ε̇X = (σ̇X + 2σ̇Y )
3K
⎪
e − ε̇ e + ε̇ p − ε̇ p = 1 σ̇ − σ̇
√ (7.132)
⎩ε̇X = ε̇X Y X Y ( X Y ) + 3γ φ (F )
2µ
where
√
0, F ≤ 0, J2
φ(F ) = F = −1
φ(F ), F > 0, K
Solving the σ̇Y from the first equation of Eq. (7.132) and then substituting it into the second
equation of Eq. (7.132), we obtain the axial elasto-visco-plastic constitutive relationship
in the 1D strain condition:
√
1 4 3µ
ε̇X = σ̇X + γ φ(F ) (7.133)
EL 3EL
where EL is the laterally constrained elastic modulus [Eq. (7.18c)]
4 (1 − v)E
EL = K + G = λ + 2µ =
3 (1 + v)(1 − 2v)
Therefore, the continuity equation [Eq. (7.3a)], the momentum conservation equation
[Eq. (7.3b)], and the elasto-visco-plastic constitutive relationship [Eq. (7.133)] together
formulate the governing equation set of the 1D strain elasto-visco-plastic waves, which
are collected below:
⎧ ∂v ∂εX
⎪ X
⎪ ∂X = ∂t
⎪
⎪
⎪
⎪
⎨ ∂vX ∂σX
ρ0 =
⎪
⎪ ∂t ∂X
⎪
⎪ √
⎪
⎪ ∂ε 1 ∂σ 4 3µ
⎩ X
=
X
+ γ φ (F )
∂t EL ∂t 3EL
Compared with the governing equation set of the 1D stress elasto-visco-plastic waves
[Eq. (6.82)], it is clear that the forms of these two equation sets are almost identical,
what is different is only to replace the Young’s√modulus E by the laterally constrained
elastic modulus EL and to replace the γ ∗ by the 4 3µγ / (3EL ). Therefore, similar to that,
the differential equations of characteristics [Eq. (6.85)] and the compatibility relationship
[Eq. (6.86)] along the characteristics can be obtained from Eq. (6.82), the following three
sets of differential equations for the characteristics and the compatibility relationship in
the 1D strain elasto-visco-plastic wave condition can be obtained, respectively:

⎧
⎪
⎪ K + 43 G
⎪
⎪ dX = ± dt = ±CLe dt
⎪
⎪
⎪
⎪ ρ 0
⎪
⎨ √ (7.134)
1 4 3µγ
⎪ dvX = ± dσX ± φ(F ) dt
⎪
⎪ ρ0 CLe 3ρ0 CLe
⎪
⎪ √
⎪
⎪
⎪
⎪ 1 4 3µγ
⎩ =± e dσX + 3E φ(F ) dX
⎧ ρ 0 C L L
⎨ dX = 0 √
dσX 4 3µγ (7.135)
⎩dεX = + φ(F ) dt
EL 3EL
Under the given initial-boundary conditions, the concrete steps in solving the problem are
the same as those discussed in Section 6.6 for the 1D stress elasto-visco-plastic waves,
and will not be repeated here again.
CHAPTER 8
Spherical Waves and Cylindrical Waves
In the practical underground explosions and engineering blasts, there are often problems of
point explosion, point impact (severely localized impact loading), and spherical or cylin-
drical cavity-walls subjected to an explosion loading. In these problems, it is necessary to
deal with the propagation of spherical or cylindrical waves.
As different from the plane waves discussed, the front of spherical and cylindrical waves
disperses during propagation. Even in an elastic situation, their wave profile keeps
changing. In addition, in the propagation of a compressive wave, tensile stress and
amplitude oscillation can be formed in the region behind the wave front.
Due to the many similar characteristics between a spherical wave and a cylindrical wave,
they can be analyzed together. We will mainly study the propagation of a spherical wave.
Later, it will be pointed out that the analyses on a cylindrical wave can be obtained
following the same way.
8.1 Continuity Equation and Motion Equation
The fronts of a spherical wave are surfaces of concentric spheres. Because of the spherical
symmetry of the motion of particles, the system of spherical coordinates, shown in Fig. 8.1,
is chosen to describe the motion. Thus, there is only non-zero displacement component
at radial direction u(r, t), and all variables are functions of radius of sphere r and time t,
independent of angle θ and ϕ, then
∂u(r, t) ∂u(r, t)
εr (r, t) = , v(r, t) = (8.1a)
∂r ∂t
u(r, t)
εθ (r, t) = εϕ (r, t) = (8.1b)
r
σr = σr (r, t), σθ (r, t) = σϕ (r, t) (8.1c)
337
∂sr
sr + dr
∂r ∂sr
sj sr + dr
∂r
sq sq
sj dq
sq 2
dr
sr
sq
sr
r
r
dq dj
dq
Fig. 8.1. An infinitesimal element in spherical coordinates (dr, dθ, dϕ).
where r, θ , and ϕ are Lagrange (material) coordinates and denote the corresponding com-
ponents when used as subscripts, other symbols denote the same quantities as in previous
chapters. Compared with Eq. (7.2), it can be found that the stress state of a spherical
wave is similar to that of a 1D strain plane wave. If the curvature radius approaches
infinity (r → ∞), the problem of spherical waves is reduced to that of 1D strain plane
waves.
To satisfy the condition that a displacement is a continuous and single value function of
r and t, from Eq. (8.1) it is known that the strains εr , εθ and the radial particle velocity v
should follow the compatible condition below:
∂εr ∂v
= (8.2a)
∂t ∂r
∂εθ v
= (8.2b)
∂t r
This is the continuity equation among the governing equations for a spherical wave,
representing the mass conservation.
Considering the momentum conservation in the radial direction (shown in Fig. 8.1),
we obtain

∂σr dθ dr
σr + dr (r + dr) dθ dϕ − σr r dθdϕ − 2σθ sin
2 2
· r+ dϕdr
∂r 2 2

dϕ dr dr 2 ∂v
− 2σϕ sin · r+ dθ dr = ρ0 r + dϕdθ dr ·
2 2 2 ∂t
Spherical Waves and Cylindrical Waves 339
neglecting high-order small quantities, and noting σθ = σϕ , then
∂σr 2(σr − σθ ) ∂v
+ = ρ0 (8.3)
∂r r ∂t
This is the motion equation of a spherical wave.
Similarly, in analyzing a cylindrical wave whose wave fronts are coaxial cylinders, there
is only non-zero displacement component in radial directions u(r, t), and all variables are
functions of cylinder radius r and time t, independent of angle θ and ϕ, then
∂u ∂u
εr = ,v = (8.4a)
∂r ∂t
u
εθ = , εz = 0 (8.4b)
r
σr = σr (r, t), σθ = σθ (r, t), σz = σz (r, t) (8.4c)
It is shown that the continuity equation of a radial cylindrical wave is totally the same
as that of a spherical wave [refer to Eq. (8.2)]. With respect to the motion equation for
a cylindrical wave, there is a slight difference in the coefficient of the item (σr − σθ )/r
from that for a spherical wave, namely we have
∂σr (σr − σθ ) ∂v
+ = ρ0 (8.5)
∂r r ∂t
Sometimes Eqs. (8.3) and (8.5) can be written in a unified equation as:
⎫
∂σr (σr − σθ ) ∂v ⎪
+ n0 = ρ0 ⎪
⎬
∂r r ∂t
(8.6)
2 (spherical waves) ⎪
⎪
n0 = ⎭
1 (radial cylindrical waves)
Thus, Eqs. (8.2) and (8.6) are the common continuity equation and motion equation of both
spherical waves and cylindrical waves. Hence, both the spherical wave and the cylindrical
wave can be analyzed jointly as a whole.
8.2 Elastic Spherical Waves and Cylindrical Waves
As discussed earlier for the plane waves, the governing equations consist of continuity
equation, motion equation, and material constitutive equation. If the former two have been
obtained shown as Eqs. (8.2) and (8.6), then the constitutive relationship can be discussed.
The constitutive equation dominates the property of a stress wave, namely, whether a stress
wave is elastic, elasto-plastic, or elasto-visco-plastic. At first, we discuss about spherical
waves and cylindrical waves in linear elastic media.
Under the condition shown in Eqs. (8.1) and (8.4), the Hooke’s law expressed in the form
of volumetric deformation law and distortional deformation law Eq. (7.15) is reduced to
σr + n0 σθ + n1 σz = 3K(εr + n0 εθ ) (8.7a)
σr − σθ = 2G(εr − εθ ) (8.7b)
(σr − σθ − 2σz )n1 = 2G(εr + εθ )n1 (8.7c)
where
n0 = 2, n1 = 0, (for spherical waves)

n0 = 1, n1 = 1, (for cylindrical waves)
Then, for a spherical wave, the governing equations consist of five equations, namely
Eqs. (8.2a), (8.2b), (8.6), (8.7a), and (8.7b), that can be used to solve five unknown
functions σr , σθ , εr , εθ , and v. For a cylindrical wave, there is one more unknown function
σz , but there is also one more equation [Eq. (8.7c)]. Therefore, there are six equations for
determining six variables.
As an example, the method of characteristics is used to analyze the propagation of a

spherical wave in elastic media. Using Eqs. (8.2) and (8.7) to eliminate εr and εθ , we obtain
1 ∂σr 2 ∂σθ ∂v 2v
+ − − =0 (8.8)
3K ∂t 3K ∂t ∂r r
1 ∂σr 1 ∂σθ ∂v v
− − + =0 (8.9)
2G ∂t 2G ∂t ∂r r
Equations (8.3), (8.8), and (8.9) build up a group of first-order hyperbolic partial differential
equations with respect to σr , σθ , and v. When solving the differential equations by using
the method of characteristics, the linear combination of the partial differential equations
can be transformed to the equations which only contain the directional derivatives along
the characteristics. So, the above three equations are multiplied by the indeterminate
coefficients L, M, and N respectively, then the summation of the three equation yields:

∂ M N ∂ 2M N ∂σθ ∂ ∂
L + + σr + − − (M + N ) + Lρ0 v
∂r 3K 2G ∂t 3K 2G ∂t ∂r ∂t
1
+ {2L (σr − σθ ) + (N − 2M)v} = 0 (8.10)
r
These coefficients, L, M, and N, should satisfy:
dr L 0 M +N
= = = (8.11)
dt M N 2M N Lρ0
+ −
3K 2G 3K 2G
Then we obtain either
L = 0, M + N = 0 (8.12)
or
2
2M N L K + 43 G
= , = (8.13)
3K 2G M ρ0 K 2
Substitution of the above results into Eqs. (8.10) and (8.11) yields three groups of charac-
teristic equations and the corresponding characteristic relationships. One is resulted from
Eq. (8.12),
dr = 0 (8.14)

1 1 2 1 3v
− dσr + + dσθ = dt (8.15)
3K 2G 3K 2G r
and other two from Eq. (8.13)
dr = ±CL dt (8.16)

2G v dr
dσr = ±ρ0 CL dv − 2 (σr − σθ ) ± K − (8.17)
3 CL r
The first family of characteristics [Eq. (8.14)] depicts the motion loci of particles
(the coordinates of particles are fixed when Lagrange coordinate system is adopted),
while the compatibility condition along this characteristics [Eq. (8.15)] is the differential
form of the constitutive equation [Eq. (8.7)]. Actually, Eq. (8.15) can be obtained by differ-
entiating the difference between Eq. (8.7a) and (8.7b) with respect to t and noticing that the
partial derivative with respect to t here equals the total derivative with respect to t along the
characteristics. The other two families of characteristics then depict the loci of wave fronts
that propagate in the positive (rightward) and negative (leftward) directions. The corre-
sponding compatibility condition, Eq. (8.17) specifies the constraint relationship among
σr , σθ , and v in order to satisfy the continuity equation, motion equation, and material con-
stitutive equation. But, comparing Eq. (8.17) with the characteristic compatibility relation
for the 1D plane strain waves, it is found that there is an additional term which relates to
dr/r. This term represents the dispersive character of spherical waves. When r → ∞, this
term tends to vanish, then the problem is reduced to that of 1D plane strain waves.
Therefore, when solving elastic spherical wave by using the method of characteristics,
it is necessary to solve simultaneously three equations of characteristic compatibility
[Eqs. (8.15) and (8.17)], where three variables σr , σθ , and v are included. This is equivalent
to that by using the governing equations [Eqs. (8.3), (8.8), and (8.9)] to solve the three
variables σr , σθ , and v. The details of solving the problem by the method of characteristics
are similar to that used in analyzing visco-elastic waves in Chapter 6.
For example, consider an elastic media with a spherical cavity of radius a, and assume that
the initial conditions, namely the σr , σθ , and v along r axis are known, and the boundary
b
N3
N2
N1 M3 Q
M2
M1
0 a P1 P2 P3 r
Fig. 8.2. The method of characteristics for solving the problem of elastic spherical waves.
condition, namely the σr on r = a is known. It is also assumed that there is temporarily

no strong discontinuity involved, namely the spherical wave is a kind of continuous wave.
Then the problem is summed up to solve the Cauchy initial-value problem in the rab region
and the characteristic boundary-value problem in the bat region as shown in Fig. 8.2.
For the bat region, considering an arbitrary point Q near r axis, there must be three
characteristics QP1 , QP2 , and QP3 that pass point Q and intersect r axis as shown in
Fig. 8.2. If the distance from Q to three points P1 , P2 , P3 is small enough, we can solve
the differential by the difference in Eqs. (8.15) and (8.17), then we have

2G v
σr (Q) − σr (P1 ) = ρ0 CL {v(Q) − v(P1 )} − 2 (σr − σθ ) − K −
3 CL P1
r(Q) − r(P1 )
·
r(P1 )

2G v
σr (Q) − σr (P3 ) = −ρ0 CL {v(Q) − v(P3 )} − 2 (σr − σθ ) + K −
3 CL P3
r(Q) − r(P3 )
·
r(P3 )

1 1 2 1
− {σr (Q) − σr (P2 )} + + {σθ (Q) − σθ (P2 )}
3K 2G 3K 2G
3v(P2 )
= {t (Q) − t (P2 )}
r(P2 )
As the position of Q, rQ , and tQ , can be determined by Eqs. (8.14) and (8.16) in advance
and σr , σθ , and v at points P1 , P2 , and P3 have been known based on the initial condition,
σr (Q), σθ (Q), and v(Q) can be obtained from the above three equations. On this analogy,
the solutions in the whole region rab can be determined. If the initial condition is a nil-
constant condition (undisturbed state), it is easy to prove that the values of the variables
in the rab region are always zero.
For the bat region, to solve a mixed problem by the method of characteristics for 1D
stress waves in bars, we should first figure out the values of all variables on the so-called
boundary points on the boundary r = a, shown as the point N1 in Fig. 8.2. As long as the
distance between N1 and M1 (and a) is small enough, according to Eqs. (8.15) and (8.17),
there are difference equations along the characteristics M1 N1 and aN1 respectively,

2G v
σr (N1 ) − σr (M1 ) = −ρ0 CL {v(N1 ) − v(M1 )} − 2 (σr − σθ ) + K −
3 CL M1
r(N1 ) − r(M1 )
·
r(M1 )

1 1 2 1
− {σr (N1 ) − σr (a)} + + {σθ (N1 ) − σθ (a)}
3K 2G 3K 2G
3v(a)
= {t (N1 ) − t (a)}
r(a)
As the solutions along the characteristics ab have just been obtained in solving the rab
region, while σ r at point N1 has been specified based on the boundary condition, so
σθ (N1 ) and v(N1 ) can be figured out from the above two equations. After obtaining the
solutions at N1 , following the same steps used to solve the Cauchy initial value problems,
we can obtain the solutions at point N2 according to the known data at the points M1 ,
M2 , and N1 . Solutions at the other so-called interior points can be obtained successively.
On this analogy, the solutions of the whole bat region can be determined.
With regard to the strong discontinuous spherical wave, according to the displacement
continuity condition and the momentum conservation condition across the wave front,
noting that for a spherical wave εθ must be continuous if u is continuous, then the kinetics
compatibility condition [Eq. (2.55)] and the dynamics compatibility condition [Eq. (2.57)]
originally deduced for the strong discontinuous plane waves are still valid. In other words,
the discontinuous jump across the wave front, with respect to the particle velocity v, stress
component σr , and strain component εr in the direction perpendicular to the spherical
wave front, satisfy the following relationships:
2
[v] = ±D[εr ]
(8.18)
[σr ] = ±ρ0 D[v]
In the case of elastic wave discussed, D = CL . It can be found that its mathematic
expression is completely same as that of the 1D strain plane waves. The dispersive effect
of spherical wave is mainly exhibited by the changes in values of the discontinuities [σr ],
[v], and [εr ] during the propagation of a strong discontinuous spherical wave, even if the
region ahead of the wave front is undisturbed.
In Fig. 8.2, if the initial conditions and boundary conditions are assumed to be:
2
u(r, 0) = v(r, 0) = σr (r, 0) = σθ (r, 0) = 0, a<r≤∞
(8.19)
σr (a, t) = −p0 , t ≥0
These initial and boundary conditions correspond to a constant pressure p0 which is sud-
denly applied on the inner wall of a spherical cavity r = a in an undisturbed infinite
body. A spherical wave is caused by the sudden loading and propagates outwards in the
infinite body. Assuming p0 is not high enough to induce plastic deformation, an elastic
strong discontinuous spherical wave front propagates along the line ab. On the one hand,
Eq. (8.18) should be valid along ab. On the other hand, ab is also a characteristic so that
Eq. (8.17) should be satisfied along ab. Considering again that from the displacement
continuity condition along ab there is εθ = u/r = 0, then from Eq. (8.17) it is known that
along ab there is
2G 2G
σr − σθ = 2Gεr = σr = σr (8.20a)
K+ 4
3G
ρ0 CL2
Thus, the dispersion term of spherical wave in the compatibility relationship [Eq. (8.17)]
along the characteristics ab can be written as

2 2G u 2σr
(σr − σθ ) − K − dr = dr
r 3 CL r
while Eq. (8.17) is reduced to:
dσr dr
=− (8.20b)
σr r
After determining the integral constant by using boundary condition σr |a = p0 , the atten-
uation of σr along ab is obtained
p0 a
σr = − (8.21a)
r
Substituting this result into Eqs. (8.18) and (8.20), we obtain the following relationships
for v and σθ along ab
σr p0 a
v=− = (8.21b)
ρ0 CL ρ 0 CL r
K − 23 G λ ν −ν p0 a
σθ = σr = σr = σr = (8.21c)
K + 43 G λ + 2µ 1−ν 1−ν r
It can be seen from Eq. (8.21c) that the stress amplitude σr of a strong discontinuous
spherical wave decreases with the increase of its distance r from the center of the sphere
in an inversely proportional relation. This is due to the effect of spherical dispersion.
Contrarily, the wave intensity will increase if a spherical wave propagates towards the
center point of the sphere (implosion). It can also be seen that the proportional relation-
ships between σr , σθ , and v on the front of a strong discontinuous spherical elastic wave
expressed in Eqs. (8.21b) and (8.21c) are the same as the relationship between those on the
front of a 1D strain plane elastic wave. Therefore, on the one hand, a strong discontinuous
spherical wave differs from a 1D strain plane wave with respect to the attenuation of its
intensity under the initial-boundary conditions shown in Eq. (8.19). On the other hand,
at each time, the stress state of a strong discontinuous spherical wave is the same as that
of a 1D strain plane wave. This means that if σr is a compressive stress, σθ is also a
compressive stress; on the contrary, if σr is a tensile stress, σθ is also a tensile stress.
The region bat swept by a strong discontinuous wave front (behind the front), as shown
in Fig. 8.2, is dominated by weak discontinuous waves. After the solutions along ab have
been obtained by using the relations of a strong discontinuous spherical wave [Eq. (8.21)],
as discussed earlier, the problem is summed up to solve a characteristic boundary-value
problem (Darboux problem) which is solvable. We will not discuss this in detail again.
But it should be emphasized that, as different from the 1D plane strain wave, the spherical
waves are no longer simple waves even under the zero initial condition and constant
loading boundary conditions [Eq. (8.19)]. Furthermore, due to the dispersion of spherical
waves, tensile stress may quickly appear in the region behind the strong discontinuous
compressive wave, especially with respect to σθ , the magnitude of the tensile stress can,
sometimes, even exceed the magnitude of the original compressive stress.
To solve the problem of elastic spherical wave propagation, there are other methods, such
as the Fourier transformation (see Miklowitz, 1978). For convenience in connecting and
comparing with the previous chapters, we use the characteristics method here. On the
other hand, based on the results mentioned above, it is also convenient to further analyze
the elasto-plastic spherical waves in the following sections.
Figure 8.3 shows the stresses vs. time curves at three points (r = a, r = 2a and r/a → ∞)
for the spherical wave propagation under the initial-boundary conditions of Eq. (8.19).
For convenience in comparison, the dimensionless stresses [−(rσr /ap0 )], [−(rσθ /ap0 )],
and [−(rτ /ap0 )] are used as vertical coordinates, where τ is the maximum shear stress
|σ − σθ |/2. The dimensionless time t¯ = [(CL t/a) − (r − a/a)] is used as horizontal
coordinates. These computations are for the case of Poisson ratio v = 0.25, namely λ = µ
(see Selberg, 1952).
Here, we compare the above results with those in the static analysis. It is known from
the analysis of static elasticity (Timoshenko and Goodier, 1951), the well-known Lame
solution shows that the stresses in the wall of a thick-wall spherical vessel subjected to a
static inner pressure (−p0 ) can be expressed by:
⎫
p0 a 3 b3 − r 3 ⎪
⎪
σr = 3 3 ⎪
⎪
r b − a3 ⎪
⎬
⎪ (8.22)
p0 a 3 b3 + 2r 3 ⎪⎪
σθ = − ⎪
⎪
⎭
2r 3 b3 − a 3
1.0
0.8 r=a
0.6
(− ar p r )
0
s
0.4 r=2a
0.2
0
−0.2 r ∞
a
2 4 6
(a)
1
3
λ ∞
2
0
(− a p )
r sr
0
r=2a
−0.4
r=a
−0.8
2 4 6 t
(b)
0.8
r=a
(− ar p r )
0
s
1
3 r=2a
r
a ∞
0
−0.2
2 4 6 t
(c)
Fig. 8.3. Propagation of elastic spherical waves under initial-boundary condition of Eq. (8.19).
where a and b are the inner and outer radius of the spherical vessel, respectively. For a
strong discontinuous spherical wave as just discussed, the dynamic analysis of elastic
spherical wave has indicated that both σr and σθ on the wave front are compressive.
However, according to Eq. (8.22) for the static analysis, σr is compressive while σθ is
tensile. Furthermore, it can be seen that the stress amplitude of a strong discontinuous
spherical wave is inversely proportional to r according to Eq. (8.21a), while the static
stress σr is inversely proportional to r 3. With respect to the region dominated by the

weak discontinuous spherical waves, the dynamic analysis of elastic spherical wave has
shown that, after a strong discontinuous wave has passed, the radial compressive stress σr
decreases, while the hoop stress σθ becomes a tensile stress in a short time. Especially at
the inner wall (r = a), the magnitude of σθ obtained by using wave analysis is larger than
that obtained by static analysis, and consequently the maximum shear stress τ obtained
by dynamic analysis is also larger than that found in static one.
With regard to the radial cylindrical waves, based on the six equations shown in Eqs. (8.2),
(8.5), and (8.7), the problem can be analyzed following the same way. And Eq. (8.18) is
still valid for the strong discontinuous cylindrical waves.
Consider an infinite elastic media with a cylindrical cavity, of which the length is infinite
and the radius is a. Under the zero initial condition and the boundary condition corre-
sponding to a uniform pressure (−p0 ) is suddenly loaded on the cavity, similar to Fig. 8.2,
a strong discontinuous cylindrical wave propagates
√ at velocity CL along the line ab, but
the amplitude of σr attenuates inversely with r, namely, Eq. (8.21) is now replaced by
the following relationships for a strong discontinuous cylindrical wave along ab:
9
a
σr = − p0 (8.23a)
r
9
a p0
v= (8.23b)
r ρ0 CL
9
ν a
σθ = σZ = − p0 (8.23c)
1−ν r
In the region swept by the strong discontinuous wave front (behind the wave front) which
is dominated by the weak discontinuous waves, the stresses vs. time curves at three
points r/a = 1, 2, and ∞, similar to that shown in Fig. 8.3, are shown in Fig. 8.4 (see
Selberg, 1952).
8.3 Elasto-Plastic Spherical Waves
In the example shown in Fig. 8.2, if the impact loading applied on the inner cavity-wall is
intense enough, then plastic deformation will take place. In the following, we will discuss
the propagation of elasto-plastic spherical waves.
Assume that the plastic deformation does not contribute to the volumetric deformation
[Eq. (7.34)] and the elastic deformation follows the Hooke’s law. Obviously, such
assumptions are valid under the conditions that the pressure is not too high and the defor-
mation is small. Due to spherical symmetry [Eq. (8.1)], both von Mises yield criterion
and Tresca yield criterion are reduced to the form of Eq. (7.32). Therefore, to replace
1.0
r=a
(− ar p r )
0
s
r=2a
0 r ∞
a
−0.2
2 4 6 t
(a)
1
3
0
r ∞
a
(− ar p r )
−0.4
0
s
r=2a
−0.6
r=a
−0.8
2 4 6 t
(b)
1
3
(− ar p r )
r
0
∞
s
a
0
r=a
−0.2
2 4 6 t
(c)
Fig. 8.4. Propagation of an elastic cylindrical wave in an undisturbed body suddenly subjected to
a constant loading.
Eq. (8.7), the constitutive equations in the governing equations for elasto-plastic spherical
wave are:
volumetric deformation law: σr + 2σθ = 3K(εr + 2εθ ) (8.24a)

2G(εr − εθ ) (elastic)
distortional deformation law: σr − σθ = (8.24b)
±Y (plastic)
p p
where the yield stress Y can be a function of plastic strain εr , Y = Yp (εr ), alternatively,
a function of plastic work Wp , Y = YW (Wp ). Differentiating Eq. (8.24) with respect to t
and after some operations based on the analogy used in deducing of Eq. (7.41), we obtain

∂σr ∂σθ ∂εr ∂εθ
+2 = 3K +2 (8.25a)
∂t ∂t ∂t ∂t
⎧
⎪ ∂εr ∂εθ
⎪
⎨ 2G ∂t − ∂t (elastic)
∂σr ∂σθ
− = (8.25b)
∂t ∂t ⎪
⎪ ∂εr ∂εθ
⎩ 2Gp − (plastic)
∂t ∂t
where the plastic shear modulus Gp is the same as that in Eq. (7.41), namely
GYp GYW YW

Gp = =
3G + Yp 3G + YW YW
Substituting Eq. (8.2) into Eq. (8.25a) and eliminating εr and εθ , we obtain Eq. (8.8);
and substituting Eq. (8.2) into Eq. (8.25b) and eliminating εr and εθ , we obtain Eq. (8.9).
While substituting Eq. (8.2) into Eq. (8.25c) and eliminating εr and εθ , we obtain
1 ∂σr 1 ∂σθ ∂v v
− − + =0 (8.26)
2Gp ∂t 2Gp ∂t ∂r r
The difference between Eq. (8.26) and Eq. (8.9) is due to the shear modulus only, where
Gp is used to replace G in Eq. (8.9). Then Eqs. (8.3), (8.8), and (8.26) constitute the
governing equations for elasto-plastic spherical waves, that are first-order quasi-linear
hyperbolic partial differential equations with respect to σr , σθ , and v. The difference
from the elastic spherical wave equations is only due to G in the third equation being
replaced by Gp .
Thus, repeating the steps used in deducing the equations since Eq. (8.10) for the elastic
spherical wave, we can obtain three groups of characteristic equations and their corre-
sponding compatibility relationships for solving the elasto-plastic spherical wave by the
characteristic method. The results obtained are similar to Eqs. (8.14)–(8.17), except that
the G is replaced by Gp and the elastic spherical wave velocity CLe

K + 43 G
CLe =
ρ0
p
is replaced by the following plastic spherical wave velocityCL

p K + 43 Gp
CL =
ρ0
Compared to Eq. (7.47), it is known that the plastic spherical wave velocity is the same as
the 1D strain plastic wave velocity. Similar to a 1D strain elasto-plastic wave, the plastic
p
spherical wave velocity (CL ) is generally smaller than the elastic spherical wave velocity
(CLe ). Consequently, a plastic spherical wave propagates behind an elastic spherical wave.
p p
Because CL is generally a function of εr the two families of characteristics describing
the plastic spherical waves on the r–t plane are generally not straight any more. In the
special cases of linear hardening materials (Gp = constant) or perfect plastic materials
p
(Gp = 0), CL constant, and the corresponding characteristics are straight lines. All the
above discussions are similar to those of 1D strain elasto-plastic waves. Similar to the situ-
ations in the 1D strain wave, once the plastic shear modulus Gp is introduced, the analyses
on the elastic spherical wave can be conveniently generalized to study the elasto-plastic
spherical wave.
Comparing with the 1D elasto-plastic waves, it should be noted that both the intensity of
the elastic precursor spherical wave and the subsequent plastic spherical wave attenuates
with the increase in r because of spherical dispersion. This case is similar to that during
the propagating process of the elastic precursor spherical wave and the subsequent plastic
spherical wave, both of which continuously reflect internal reflective waves and moreover
these internal reflective waves interact with each other, thus making the problem more
complicated. For example, in the case of linear elasto-perfect plastic materials or linear
elasto-linear plastic hardening materials, when the inner wall of a spherical cavity is
suddenly subjected to a constant loading, both the elastic precursor spherical wave and
the subsequent plastic spherical wave become strong discontinuous waves. The elasto-
plastic loading boundary coincides with the propagating locus of strong discontinuous
plastic wave until the latter vanishes due to the attenuation effect. In this aspect, it is
similar to that of 1D strain elasto-plastic waves. However, the region between an elastic
strong discontinuous spherical wave and a plastic strong discontinuous spherical wave is
no longer a constant-value region. In this aspect, it differs from that of a 1D strain elasto-
plastic wave. This is because that, on the one hand, the stresses on the wave front of an
elastic strong discontinuous spherical wave continuously attenuate in inverse proportion
to r, and so that the yield criterion is no longer satisfied; and on the other hand, the initial
yield condition in a spherical wave is no longer simply expressed by σX = YH [Eq. (7.33)].
In fact, after eliminating σθ and εr from the three equations of Eq. (8.24), it can be found
that the initial yield condition from an elastic state to a plastic state can be expressed as
σr = 3Kεθ + YH
where
K + 43 G
YH = Y0
2G
is the Hugoniot yield limit. So, the σr ahead of the plastic strong discontinuous wave front
is not equal to YH , but is related to the current value of εθ in the elastic region.
Because of the same reason, for a weak discontinuous spherical plastic wave, the spherical
dispersion causes the attenuation of stress amplitudes, resulting in the stress originally
satisfying the yield condition to no longer satisfy the yield criterion. Consequently, a plastic
wave will be reduced or transformed to an elastic wave during its propagation, and the
propagating velocity of weak discontinuous elasto-plastic boundary (loading boundary)
will be lower than that of a plastic spherical wave. This is another character of spherical
waves that differs from that of 1D strain plane waves.
During unloading, an elastic unloading occurs first. Under the assumption of elastic unload-
ing, the propagating velocity of unloading disturbance is CLe . When the stress decreases
p
further, a reverse plastic loading occurs and the disturbance propagates at a velocity of CL .
These are similar to that of the 1D strain elasto-plastic wave. However, it should be noticed
that when an elastic unloading disturbance pursues a plastic loading disturbance, its stress
amplitude is going to decrease continuously due to the spherical dispersion, so that the
elastic unloading disturbance could be transformed to a reverse plastic loading. Therefore,
the propagating velocity of reverse plastic loading boundary will not only be faster than
the propagating velocity of plastic loading disturbance itself, but also even faster than the
propagating velocity of elastic wave.
Consider an infinite elasto-perfect plastic media (Gp = 0) with a spherical cavity of

radius a, and the inner wall of the cavity is subjected to a compressive pulse expressed by
the following formula

CK t
σr (t) |r=a = −5.0Y0 exp −3.5
a
Figure 8.5 shows the calculated radial stress σr profiles

√ at several dimensionless times
t¯ = CK t/a (Friedman et al., 1965), where CK = K/ρ0 is the 1D strain plastic wave
velocity in an elasto-perfect plastic material. From the figure, it can be seen that the
strong discontinuous elastic precursor wave continuously attenuates, and the stress σr
just ahead of the strong discontinuous plastic wave front also keeps changing (although
the yield condition is satisfied). The stress amplitude of the strong discontinuous plastic
sr (t) r=a= −5.0Y0e−3.5t

5 C t
t= aK , CK = ρ K
0
5Y a
4 t=0.05 sr= r 0 (Solution of elastic waves)
3 0.30
Y0
sr
2 0.60
0.90
0
1 1.25 1.5 1.75 2.0
r
a
Fig. 8.5. Radial stress profiles of an elasto-perfect plastic spherical wave.

wave attenuates much faster. This is caused by the subsequent unloading of exponential
attenuating disturbance on the one hand and the effect of spherical dispersion on the other
hand.
With respect to the determination of weak discontinuous boundary of elastic zone and plas-
tic zone for spherical elasto-plastic waves, it is similar to that mentioned in Section 4.10.
If it is assumed that the plastic deformation makes no contribution to volumetric defor-
mation, then it is known from Eq. (8.25) that the volumetric modulus is continuous at
the elastic–plastic boundary, namely [K] = 0; while the shear modulus is discontinuous,
namely [G]
= 0, and consequently the wave velocity is discontinuous
p
[CL ] = CL − CLe
= 0 (8.27)
This is the essential singularity which dominates the discontinuous properties of elastic–
plastic boundary produced for spherical waves. Other singularities along the elastic–plastic
boundary depend on this essential singularity and are determined by this singularity.
In relation to this, (∂ |σr − σθ |)/∂t or (∂ |εr − εθ |)/∂t should be considered in judging
whether the media is undergoing a plastic loading or an elastic unloading in the study on
the problems of spherical elasto-plastic waves, instead of ∂σX /∂t or ∂εX /∂t in the cases
of 1D stress wave or 1D strain elasto-plastic waves. Or if introducing the maximum shear
stress τ = (σr − σθ ) /2 and shear strain γ = (εr − εθ )/2, and noticing the continuity
equation Eq. (8.2), we have
∂ ∂v v
(εr − εθ ) = −
∂t ∂r r
then we can also equivalently examine ∂τ/∂t, ∂γ /∂t, or (∂v/∂r, v/r), etc. The disconti-
nuity of elasto-plastic distortional deformation law [Eq. (8.25b)] across the elastic–plastic
boundary can be expressed as

1 ∂τ ∂γ 1 ∂v v
= = − (8.28)
2G ∂t ∂t 2 ∂r r
For the first-order weak discontinuous boundary, namely σr , σθ , and v are continuous but
their first-order partial differentials are discontinuous, similar to Eq. (4.106), we have the
following relationships

∂σr ∂σr
= −C (8.29)
∂t ∂r

∂v v ∂v v
− = −C − (8.30)
∂t r ∂r r
On the other hand, from the governing equations for an elastic region [Eqs. (8.3), (8.8),
and (8.9)] and the governing equations for a plastic region [Eqs. (8.3), (8.8), and (8.26)],
it is known that

∂σr ∂v v
= ρ0 − (8.31)
∂r ∂t r

∂σr ∂v v
= ρ0 CL2 − (8.32)
∂t ∂t r
To simplify the analysis, the items related to [v/r] have been added to Eqs. (8.30) and
(8.31). As [v/r] = 0, they have no effect on results.
Thus, Eqs. (8.27), (8.29)–(8.32) are the basic equations that the weak discontinuous elasto-
plastic boundaries of spherical waves should obey. They are corresponding to Eqs. (4.101),
(4.106), and (4.107) for the 1D stress waves. Based on these, we can perform similar dis-
cussions mentioned in Section 4.10. For example, for a first-order weak discontinuous
boundary, by eliminating [∂σr /∂t], [∂σr /∂r], and [∂v/∂t − v/t] from Eqs. (8.29)–(8.32),
we obtain the following relationship for the boundary propagating velocity C
p e 2
vp
C 2 ∂v − v ∂v
− 1−
CL
L
∂r r ∂r r C
C= , or e = % p &2 (8.33)
∂v v ∂v ve
− − C
∂r r ∂r r 1− L
C
Or following Eq. (8.28), we can rewrite Eq. (8.33) in several equivalent forms, such as:
⎫

⎡⎡ C 2 ∂τ ⎤⎤ ⎪
C 2 ∂γ ⎣⎣ L ⎪
⎪
L ⎦⎦ ⎪
⎪
∂t G ∂t ⎪
⎪

C = = / 1 ∂τ 0 ⎪
⎪
∂γ ⎪
⎪
⎪
⎪
⎪
⎪
∂t G ∂t ⎪
⎬
% & (8.33 )
⎛ p ⎞2 p 2 ⎪
⎪
CL CL ⎪
⎪
1−⎝ ⎠ 1− ⎪
⎪
p p C ⎪
⎪
∂γ /∂t C ∂τ /∂t Gp ⎪
⎪
= ⎛ ⎞ , = ⎪
⎪
∂γ e /∂t e
CL
2
∂τ e /∂t G CLe 2 ⎪
⎪
⎝ ⎠ 1− ⎪
⎪
1− ⎭
C C
Referring to Eq. (4.111), Eq. (8.33) can be expressed in terms of ∂ε/∂t as [or refer to
Eq. (4.73b)]

C 2 ∂ε

∂t
C=
∂ε
∂t
then replacing only ε in the above equation by γ , we obtain the corresponding equation
given in Eq. (8.33 ) for spherical waves.
In the above discussions, we assumed that the volumetric modulus K is a constant, which
is only valid in situations of small strain and correspondingly the pressure is not too high.
So the discussions in this section can be regarded as an extension of the theory of 1D strain
elasto-plastic waves in small strain condition (as discussed in Section 7.4) to the spherical
waves. In the case of high pressure, it must be considered that the elastic modulus k is not
a constant any more, which generally increases with the increase in pressure (Section 7.6),
and consequently a shock wave will be formed. In such a case, it is necessary to extend the
theory of 1D hydro-elasto-plastic plane wave (Section 7.9) to the spherical waves, so that
the analysis becomes much more complicated.
8.4 Approximate Analysis for the Fragmentation of Spherical Shells
As discussed in Chapter 5, if plastic deformation is much larger than elastic deformation

and the elastic wave velocity is much faster than the plastic one, especially when the
problem we are mainly concerned with is the situation where the elastic wave after having
propagated back-and-forth many times in the body studied, then we can approximately let
the elastic wave velocity approach infinity. This means that the material is assumed to be
rigid-plastic, so that the study will be simplified.
Consider a thick spherical shell with an inner radius r = a, which is subjected to an explo-
sion pressure pa on its inner wall. To simplify the dynamic analysis, let the elastic modulus
be approximately infinity. This is equivalent to that in the elasto-plastic constitutive rela-
tionship [Eq. (8.24)] discussed in the previous section, the elastic volumetric deformation
law is replaced by the volume-incompressible condition (εKK = 0), while only the yield
condition [the second equation in Eq. (8.24b)] is still valid in the distortional deforma-
tion law. Then, the governing equations for the problem are composed of the continuous
equation [Eq. (8.2)], the motion equation [Eq. (8.3)], and the yield condition [the second
equation in Eq. (8.24b)], that are together re-listed below:
⎧
⎪ ∂ε ∂v
⎨ r =
continuity equation: ∂t ∂r (8.34a)
⎪
⎩ ∂εθ = v
∂t r
∂σr 2(σr − σθ ) ∂v
motion equation: + = ρ0 (8.34b)
∂t r ∂t
incompressible condition: εr + 2εθ = 0 (8.34c)
yield condition: σr − σθ = ±1 (8.34d)
Substituting Eqs. (8.34a) and (8.34b) into the equation obtained by differentiating
Eq. (8.34c) with respect to t, the volume incompressible condition becomes
∂v v
+2 =0
∂r r
or
∂ 2
vr = 0
∂r
Integrating it, we obtain
a2
v= va (8.35)
r2
where va is the radial velocity of particles on the inner wall (r = a). Substituting the above
equation and yield condition [Eq. (8.34d)] into the motion equation [Eq. (8.34b)], we obtain
∂σr 2Y ρ0 a 2 dva
= + 2
∂r r r dt
Integrating this equation and using the boundary condition σr |a = −pa to determine the
integral constant, we obtain
r a dva
σr = −pa + 2Y ln + ρ0 a 1 − (8.36)
a r dt
where the material has been assumed as perfect plastic (Y = constant). Substitution of the
above equation into the yield condition yields
r a dva
σθ = Y − pa + 2Y ln + ρ0 a 1 − (8.37)
a r dt
In general, the explosion pressure pa on the inner wall is much larger than the yield stress
Y of materials. So unless pa decreases to Y or even lower, both σr and σθ on the inner
wall are compressive stresses,
σr |a = −pa , σθ |a = − (pa − Y )
This differs from that obtained in static elastic analysis [Eq. (8.22)].
From Eq. (8.37) it is known that the value of compressive stress σθ decreases with
increasing r. Let σθ = 0 at rh = (b − h), then σθ ≥ 0 at r > rh , namely σθ is tensile at
r > rh . Here, b is the outer radius of the thick spherical shell, and h is the distance from
the outer wall.
Taylor (1963) studied the fragmentation of a cylindrical shell. According to the observa-
tions recorded by a high-speed camera, it was found that axial cracks firstly appear on the
outer wall of the shell, and then grow with the shell expansion. Until the radius of the
shell expands to two times its original radius, cracks will penetrate through the shell wall,
and lead to shell fragmentation. Obviously, only when the compressive stress zone of σθ
disappears, the cracks could penetrate up to the inner wall, so that the cylindrical shell
could be fragmented.
It is known from Eq. (8.37) that the location rh or h, where stress σθ changes from
compressive to tensile, can be determined by the condition of σθ = 0 at r = rh = (b − h),
namely,

b−h a dva
pa − Y = 2Y ln + ρ0 a 1 −
a b−h dt
The condition that compressive stress zone of σθ disappears is h = b − a, which results

when the right-hand side of the above equation equals zero. Therefore, the shell will be
fragmented only till pa decreases with time to the magnitude of Y .
Assuming that the explosive gas within the spherical shell expands adiabatically, then
pV γ = constant
where γ is the exponent of gas adiabatic expansion. The original inner radius of the shell
is denoted by a, and the inner radius becomes A at an arbitrary time t due to expansion.
Noticing V = 4πA3/3, if the initial pressure is p0 , then pressure pa on inner wall at
time t is
a 3γ
pa = p0
A
If the spherical cavity is not fully filled with explosive and the radius of explosive charge
is ac (< a), then
a 3γ
c
p0 = pc
a
Because the expansion velocity of a spherical wall is much smaller than that of a detonation
wave, it is reasonable to assume that the explosion is instantaneous. So pc can be regarded
as the instantaneous detonation pressure (namely the half of C-J pressure), thus
a 3γ a 3γ a 3γ

0 c
pa = pc · = pc
a A A
If the condition that the compressive zone of σθ disappears (namely pa = Y ) as the criterion
of shell fragmentation is taken into consideration, then from the above equation we obtain
the following equation for the fragmentation radius Af :
p 1
c 3γ
Af = ac (8.38)
Y
Al-Hassani and Johnson (1969) analyzed this problem by using Euler variables. In the
present section, we use Lagrange variables to analyze the same problem (Wang, 1983).
8.5 Elasto-Visco-Plastic Spherical Waves and Cylindrical Waves
For the rate-dependent materials, if we adopt the (Sokolovskii)-Malvern-

Perzyna elasto-visco-plastic theory [ref. Eq. (7.131) in Chapter 7], then in solving problems
of spherical waves and cylindrical waves, the following constitutive relationship is used
to replace Eq. (8.7) in the theory of elastic waves,
1
ε̇r + n0 ε̇θ = (σ̇r + n0 σ̇θ + n1 σ̇z ) (8.39a)
3K
= √ >
1 J2 σr − σ θ
ε̇r − ε̇θ = (σ̇r − σ̇θ ) + γ φ −1 √ (8.39b)
2µ K J2
= √ >
1 J2 2σz − σr − σθ
n1 (ε̇r + ε̇θ ) = −n1 (2σ̇z − σ̇r − σ̇θ ) + γ φ −1 · √
2µ K J2
(8.39c)
Together with the continuity equation [Eq. (8.2)] and the motion equation [Eq. (8.6)], the
governing equations can be re-listed below,
⎫
∂εr ∂v ⎪
− =0 ⎪
⎪
∂t ∂r ⎪
⎪
⎪
⎪
∂εθ v ⎪
⎪
− =0 ⎪
⎪
∂t r ⎪
⎪
⎪
⎪
∂v ∂σr (σr − σθ ) ⎪
⎪
ρ0 − − n0 =0 ⎪
⎪
⎪
⎪
∂t ∂r r ⎪
⎪
∂v 1 ∂σr n0 ∂σθ n1 ∂σz v ⎪
⎬
− − − + n0 = 0 (8.40)
∂r 3K ∂t 3K ∂t 3K ∂t r ⎪
= √ > ⎪
⎪
∂v 1 ∂σr 1 ∂σθ v J2 σ r − σθ ⎪
− − − −γ φ −1 · √ =0⎪⎪
⎪
⎪
∂r 2µ ∂t 2µ ∂t r K J2 ⎪
⎪
B C⎪
⎪
∂v n1 ∂σr n1 ∂σθ n1 ∂σz v √ ⎪
⎪
n1 − − + + n1 + n1 γ φ K − 1 ⎪
J2
⎪
⎪
∂r 2µ ∂t 2µ ∂t µ ∂t r ⎪
⎪
⎪
⎪
2σz − σr − σθ ⎪
⎪
· √ =0 ⎪
⎭
J2
where each symbol denotes the same variable used in the previous sections. Thus, for
cylindrical waves there are six unknowns σr , σθ , σz , εr , εθ and v in six equations. For
spherical waves, one unknown σz is reduced while one equation (the last one) is also
reduced.
When the characteristic method is adopted to solve these hyperbolic first-order partial
differential equations, in order to determine the characteristics and the corresponding
characteristic compatibility relationships, multiplying the above six equations by six
undetermined coefficients L, M, N , P , Q, and R respectively and adding them together,
we obtain

∂ ∂ ∂ ∂
L εr + M εθ + Nρ0 + (P + Q + Rn1 − L) v
∂t ∂t ∂t ∂r

∂ P Q Rn1 ∂ P n0 Q Rn1 ∂
− n + + + σr − − + σθ
∂r 3K 2µ 2µ ∂t 3K 2µ 2µ ∂t

P n1 Rn1 ∂ v σr − σ θ
+ − + σz + (P n0 − M − Q + Rn1 ) − N n0
3K µ ∂t r r
γ φ
+ √ {2Rn1 σz − (Rn1 − Q) σr − (Rn1 − Q) σθ } = 0
J2
These coefficients should satisfy
dr 0 0 P + Q + Rn1 − L N
= = = =
dt L M Nρ0 P Q Rn1
+ +
3K 2µ 2µ
0 0
= = (8.41)
P n0 Q Rn1 P n1 Rn1
− + − +
3K 2µ 2µ 3K µ
Then we obtain the following solutions
P n0 θ Rn1 Rn1 P n1
(1) L=M= − + = − =0
3K 2µ 2µ µ 3K
In such a case, we obtain two families of real characteristics. The characteristic equations
and the corresponding compatibility relationships are:

K + 43 µ
dr = ± dt = ± CLe dt, (8.42a)
ρ0

1 2µ
dσr = ± ρ0 CL dv + n0
e
K− v ± CL (σr − σθ )
e
r 3

2µ γ φ
− √ [(1 + n1 ) σr − σθ − n1 σz ] dt
3 J2

1 2µ v
= ± ρ0 CLe dv − n0 (σr − σθ ) ± K −
r 3 CLe

2µ γ φ
± e √ [(1 + n1 ) σr − σθ − n1 σz ] dr (8.42b)
3CL J2
These two families of characteristics represent the propagating loci of the positive and the
negative elasto-visco-plastic waves.
(2) N = P + Q + Rn1 − L = 0
In such a case, we obtain four families of real characteristics in coincidence:
dr = 0 (8.43a)
The corresponding compatibility relationships are:
(i) M
= 0 and other coefficients are all zero, then
v
dεθ = dt (8.43b)
r
(ii) P = L
1 v
dεr − (dσr − n0 dσθ + n1 dσz ) + n0 dt = 0 (8.43c)
3K r
(iii) Q = L

1 v σr − σ θ
dεr − (dσr − dσθ ) − + γ φ √ dt = 0 (8.43d)
2µ r J2
(iv) Rn1 = L

n1 v 2σz − σr − σθ
n1 dεr + (2dσz − dσr − dσθ ) + n1 + γ φ √ dt = 0
2µ r J2
(8.43e)
The family of characteristics described by Eq. (8.43a) represents the motion loci of par-
ticles. The corresponding compatibility relationships [Eqs. (8.43b)–(8.43e)] are actually
the total differential forms of the continuity equations [Eq. (8.40b)] and the constitutive
equations [Eq. (8.39)]. Because when using Lagrange variables, ∂/∂t is the same as the
total derivative with respect to t along dr = 0.
Thus, we have obtained all differential equations of characteristics and the corresponding
compatibility relationships for elasto-visco-plastic spherical waves or elasto-visco-plastic
cylindrical waves. With given initial and boundary conditions, we can solve the prob-
lems of elasto-visco-plastic spherical waves and cylindrical waves by using a numerical
characteristics method.
Consider an infinite elasto-visco-plastic media with a spherical cavity of radius r = a,

which is suddenly subjected to an explosion pressure uniformly distributed on its inner
wall, as shown in Fig. 8.6. In the beginning, there is a strong discontinuous spherical
wave propagating along the characteristics ab. As discussed in Section 8.2, for the strong
discontinuous spherical waves, Eq. (8.18) should be satisfied across the wave front. If the
media are in rest and undisturbed state initially, then there exist the following relationships
t t
II (Elastic)
I (E t=g
p(t)
last (r)
ic-v
isco
pla
stic b
)
sr p0 a r* r
Fig. 8.6. Schematics of a numerical characteristic method to solve a problem of elasto-visco-plastic

spherical wave propagation.
to describe the states on the strong discontinuous spherical wave front, which is the same
as that for the 1D strain waves,
u = εθ + ε ϕ = 0 (8.44a)
σr σr
v=− e , εr = (8.44b)
ρ0 C L K + 43 µ
K − 23 µ λ ν
σθ = σϕ = σr = σr = σr (8.44c)
K+ 4
3µ
λ + 2µ 1−ν
However, as different from the case of 1D strain waves, each variable in the above equa-
tions does not remain constant along the characteristics ab. In addition, the characteristic
compatibility relationship should be satisfied along ab. Because, for an elasto-visco-
plastic wave, there is only instantaneous response across a strong discontinuous
√ wave

front (ε̇ = ∞), which implies Wp = 0, then the K in the term φ ( J2 /K) − 1 of
√
Eq. (8.42b) is reduced as K = k = Y0 / 3 (where Y0 is the yield limit in simple tension).
Moreover, consideration of Eq. (8.44) yields [refer to Eq. (8.20a)]
; σr − σθ 2µ
J2 = √ =√ σr
3 3 K + 34 µ
Then following similar steps in deducing Eq. (8.20b), along ab we have:

D % &E
dσr n0 σ r 2µ 2µσr
=− −√ eγ φ √ 2 − 1 ≡ −ψ (r, σr (r)) (8.45)
dr 2r 3CL 3ρ0 CLe k
Integrating both the sides of the above equation and using the initial condition
σ r (a, 0) = −p0 , we obtain the following nonlinear second-kind Volterra integral equation
r
σ r = p0 − ψ(ξ, σr (ξ ))dξ (8.46)
a
In general, this equation can be solved by successive approximations method. Once

σ r along ab is obtained, the determination of the solutions in region I is a solvable
characteristic boundary-value problem.
Note that when a strong discontinuous spherical wave propagates along ab, its amplitude
decreases with increasing r. This is caused by the effect of spherical dispersion on the
one hand and the effect of visco-plastic dissipation on the other hand. These two effects
correspond to the first and the last items on the right-hand side of the first equality sign
in Eq. (8.45) respectively, which defines ψ. Assuming that the over-stress disappears at
point b(r ∗ ) shown in Fig. 8.6, namely there is

J2 r ∗ = k 2 (8.47)
then the solutions for the region of r ≥ r ∗ are elastic ones [Eq. (8.21)]. By the same argu-
ment, the solutions at points in region I (elasto-visco-plastic zone) that satisfy Eq. (8.47)
can be solved. Then, we can determine the unloading boundary t = g(r) which is locus of
points where the over-stress disappears. In the region II (elastic unloading zone), we can
obtain solutions by the method previously used in the analysis on elastic spherical waves.
Thus, the whole problem can be solved (Bejda and Perzyna, 1964).
CHAPTER 9
Elastic–Plastic Waves Propagating in Flexible

Strings
In the present chapter, the elastic–plastic waves propagating in flexible strings, i.e. the
strings that only bear stretch force, will be discussed.
In the classic theory of elastic waves propagating in flexible strings, only two simple kinds
of elastic wave propagation are dealt with: the longitudinal waves without transverse
displacement, which are in fact in the scope of the longitudinal wave propagation theory
discussed in Chapter 2, and the transverse waves, with only transverse displacement u but
without longitudinal strain.
If it is assumed that the transverse displacement u is always perpendicular to the string

(the X-axis) at any time and it is further limited only to study the small-amplitude vibration
of strings, [assuming (∂u/∂X)2 ≤ 1], then the stretch force of string T is unchanged with
coordinate X and time t, i.e.:
T = T0 = constant
and the transverse wave velocity Ch is also constant:

T0
Ch = = constant (9.1)
ρ0
where ρ0 is the initial density of the string.
The problems of longitudinal and transverse waves simultaneously propagating in strings

have been studied by Cole et al. (1953) and Smith et al. (1958). However, it is limited to
the linear elastic strings.
If the loading exerted on the string is high enough, the string will experience plastic
deformation, and plastic waves will propagate along the string. In general, both plastic
363
longitudinal waves and plastic transverse waves may simultaneously propagate in the
string and then interact. The problem becomes very complicated.
Historically, the studies on elasto-plastic dynamics of strings probably began in the period
of World War II, although the research results were published only after the war. The first
work can probably be traced to that of (Rakhmatulin) (1945), and then the
works of (Cristescu) (1954) and Craggs (1954). Compared to the studies on
elastic–plastic waves propagating in bars, the studies on elastic–plastic waves propagating
in strings are much less satisfactory.
The studies of elastic–plastic dynamics of strings are motivated by a series of practical

problems, such as the impact strength of the strings of a parachute, the stress wave prop-
agation in the fibers of long-fiber-reinforced composites, the impact strength of the rope
used in mines, the impact deformation of yarn in weaving machines, etc. Moreover, with
regard to the stress wave propagation in bars, longitudinal wave propagation is the sim-
plest problem. In more general cases, the bar may be subjected to oblique impact, where
upon the problem of elastic–plastic bending wave propagation in beams should be dealt
with, as will be discussed in Chapter 10. For slender beams, if the bending stiffness is
as small as could be neglected, then the problem is approximately reduced to the study
on elastic–plastic wave propagation in flexible strings. Furthermore, based on the stud-
ies on stress wave propagation in strings, a further study can be made on the impact
problem of membranes, which can be regarded as a generalization of the impact problem
of strings. Finally, the study on elastic-plastic wave propagation in strings has been used
as a dynamic experimental technique to investigate the dynamic stress–strain relation of
materials, as proposed by (Rakhmatulin) (1945b).
Similar to what we discussed on stress wave propagation in bars, the stress wave propa-
gation in strings can be studied either by Lagrange or Euler variables, or a mix of both.
In the following, in order to establish a more closed connection between the related chap-
ters in the present book, and also for convenience, the Lagrange variables will be used
(Wang, 1964).
9.1 Governing Equations
First of all, it should be made clear again that we are limited to the study of flexible strings.
“Flexible” herein means that the stretch force at any point on the string is always acting in
the tangential direction of the instantaneous profile of the string, i.e. the string only bears
stretch force without any resistance to bending.
In a string that keeps its straight line shape unchanged in the motion throughout, the stretch
force acting on the string also keeps its direction invariable throughout. In such a case,
only longitudinal waves propagate in the string, and can thus be dealt with completely by
the longitudinal wave propagation theory discussed in Chapter 2; no further discussion is
needed.
On the other hand, when a transverse wave propagates in a string, a shape change of the
string must take place. In other words, only when a direction change of the stretch force
vector T acting on the string occurs will a transverse disturbance propagate in the string.
Elastic–Plastic Waves Propagating in Flexible Strings 365
If an abrupt change of string shape (break off in shape) takes place, the direction of stretch
force vector T abruptly changes across this break point of the string. In other words, a strong
discontinuity of stretch force direction appears. According to the momentum conservation
theorem, there must correspondingly be an abrupt jump of transverse velocity, and vice
versa. Thus, such a strong discontinuous transverse wave front must propagate in the string
in the form of a “break point”.
In the following, we will first discuss the propagation of such strong discontinuous trans-
verse waves.
Let us use Lagrange variables S0 and t to describe the motion of the string, where S0 is the
original arc length of the string, and t is the time. Denote the displacement vector by u, then
the particle velocity vector v is expressed as v = ∂u/∂t. Note that there exists, in general,
an intersection angle β between the particle velocity vector and the string direction, as
shown in Fig. 9.1, where the subscripts 1 and 2 indicate the quantities in front of and
behind the strong discontinuity, respectively.
Consider the case in which a strong discontinuous transverse wave propagates an infinites-
imal distance along a string segment dS0 in an infinitesimal time interval dt. The
infinitesimal string segment suffers an impulse of
(T1 + T2 )dt
and the corresponding change of momentum is
ρ0 (v2 − v1 )dS0
According to the momentum theorem, we have the following dynamic condition across
the strong discontinuous interface:
ρ0 Ch (v2 − v1 ) = T2 + T1 (9.2)
where Ch = dS0 /dt is the Lagrange propagating velocity of the strong discontinuous
transverse wave and ρ0 the original line density of the string.
v1
v2 g b1
b2 T1
T2
Fig. 9.1. A strong discontinuous transverse wave propagating in the string.

The continuity condition requires the displacements across the strong discontinuous
interface to be equal, i.e.:
u2 (S0 , t) = u1 (S0 , t)
This condition should be always held along with the propagation of strong discontin-
uous interfaces. In other words, the total derivative of u2 and u1 following the strong
discontinuous interface propagation should be equal, i.e.:
∂u2 ∂u2 ∂u1 ∂u1

+ Ch = + Ch (9.3)
∂t ∂S0 ∂t ∂S0
which can be rewritten as

∂u2 ∂u1
v2 − v1 = −Ch − (9.4)
∂S0 ∂S1
where ∂u/∂S0 is the relative displacement vector. If the unit vectors of infinitesimal string
segment before and after deformation are denoted by S0 and S1 , respectively, as shown in
Fig. 9.2.
Then, obviously, we have
∂u
(1 + ε)S1 = S0 + (9.5)
∂S0
dS − dS 0
where ε = is the tensile strain of the string.
dS 0
The strain ε and stretch force T are connected to each other by the known constitutive
relation of string material in the form of
T = T (ε) (9.6)
s 0.
s1
)d
+e
.s 1
=(1
ds u ds 0
s0
s0
s0 d
u
u+
dS0.S0
Fig. 9.2. An infinitesimal string segment before and after deformation.

It has been assumed herein and from now on that the stretch force is only the function of
strain T = T (ε), independent of strain rate, and only the situation of (d 2 T /dε 2 ) ≤ 0 will
be discussed.
Thus, Eqs. (9.2), (9.4), and (9.6) constitute the basic equations of the current problem.
Recalling the Rankine–Hugoniot relations deduced in Section 2.3 for the strong discontin-
uous longitudinal wave propagating in a bar, it can be found immediately that Eqs. (9.2),
(9.4), and (9.6), respectively, correspond to Eqs. (2.57), (2.56), and (2.14).
Without loss of generality, assume that the string is in plane motion, and then the vector
equations (9.2) and (9.4) are equivalent to four scalar equations. For example, if we
resolve the vector into the tangent component and the normal component with respect to
the string segment in front of the strong discontinuous interface, denote the intersection
angle between the string segments in front of and behind the “break point” by γ , and
take account of Eq. (9.5). Then we have the following four scalar equations (note that the
intersection angle between the vectors S0 and S1 equals γ behind the strong discontinuity
but equals zero in front of the strong discontinuity):
⎧
⎪
⎪ ρ0 Ch [v2 cos(β2 + γ ) − v1 cos β1 ] = T2 cos γ − T1
⎪
⎪
(9.7)
⎪
⎪
⎨ ρ0 Ch [v2 sin(β2 + γ ) − v1 sin β1 ] = T2 sin γ (9.8)
(9-I)
⎪
⎪ v2 cos(β2 + γ ) − v1 cos β1 = Ch [(1 + ε2 ) cos γ − 1 − ε1 ] (9.9)
⎪
⎪
⎪
⎪
⎩ v sin(β + γ ) − v sin β = C (1 + ε ) sin γ
2 2 1 1 h 2 (9.10)
From Eqs. (9.8) and (9.10), we immediately obtain the transverse wave velocity:

1 T2
Ch = (9.11)
ρ0 1 + ε 2
On the other hand, from Eqs. (9.7) and (9.9), we have
T2 cos γ − T1 T2
=
(1 + ε2 ) cos γ − (1 + ε1 ) 1 + ε2
which can be reduced to

T1 T2
= (9.12)
1 + ε1 1 + ε2
In order to satisfy this condition, there are two possible solutions, i.e. either
T1 = T2 , ε1 = ε2 (9.13)
or
T1 − T2 T1 T2
= = (9.14)
ε1 − ε 2 1 + ε1 1 + ε2
T
E1
E0
–1 0 e
Fig. 9.3. The extension of linear hardening plastic segment just intersects at ε = −1.
Except in the special case of T = T (ε) shown in Fig. 9.3, only Eq. (9.13) could be
the solution of Eq. (9.12). It means that the transverse wave only brings about shape
change of the string, and not strain disturbance. In the special linear hardening case
of T = T (ε) shown in Fig. 9.3, Eq. (9.14) could be a possible solution of Eq. (9.12).
In such cases, the strong discontinuous longitudinal wave and the strong discontinuous
transverse wave coincide with each other. The jump of T and ε can be considered as the
result induced by the longitudinal wave, while the shape change (break in shape) can be
considered as the result induced by the transverse wave.
If ε 1, the stretch force can be approximately regarded as a constant, and Eq. (9.11) is
reduced to Eq. (9.1) the elastic transverse wave velocity given by the classic elastic theory
under the condition of small-amplitude vibration.
When γ = 0, Eqs. (9.7)–(9.10), i.e. the equation group (9-I), are reduced to:
⎧
⎪
⎪ ρ0 Ch [(v cos β)2 − (v cos β)1 ] = T2 − T1 (9.15)
⎪
⎪
⎪
⎨
(9-II) (v cos β)2 − (v cos β)1 = Ch (ε2 − ε1 ) (9.16)
⎪
⎪
⎪
⎪
⎪
⎩ (v sin β)2 = (v sin β)1 (9.17)
They are, in fact, the Rankine–Hugoniot relations for strong discontinuous longitudi-
nal wave propagating in a moving bar, which moves with a transverse particle velocity
of v sin β = constant, completely coinciding with Eqs. (2.57) and (2.56), and Ch is
reduced to:

1 T2 − T 1
Ch =
ρ0 ε2 − ε 1
When the jump across the front of a strong discontinuous transverse wave gradually
decreases and approaches infinity, its limit is the weak discontinuous transverse wave.
In such cases, use dγ instead of γ , and, correspondingly, let
T2 = T1 dT , ε2 = ε1 + dε, β2 = β1 + dβ, v2 = v1 + dv
the equation group (9-I) is reduced to the following dynamic condition and kinematics
condition across a weak discontinuous transverse wave front:
⎧
⎪
⎪ ρ0 Ch [d(v cos β) − v sin βdγ ] = dT (9.18)
⎪
⎪
⎪
⎪
⎨ ρ0 Ch [d(v sin β) + v cos βdγ = Tdγ (9.19)
(9-III)
⎪
⎪ d(v cos β) − v sin βdγ = Ch dε (9.20)
⎪
⎪
⎪
⎪
⎩ d(v sin β) + v cos βdγ = Ch (1 + ε)dγ (9.21)
The wave velocity of a weak discontinuous transverse wave can be obtained immediately
from Eqs. (9.19) and (9.21) as

1 T (ε)
Ch (ε) = (9.22)
ρ0 1 + ε
On the other hand, from Eqs. (9.18) and (9.20), we have
T dT
= (9.23)
1+ε dε
In order to satisfy this condition, there are two possible solutions, i.e. either
dT = 0, dε = 0 (9.24)
or
ε = εA (9.25)
where εA is the strain at the point A as shown in Fig. 9.4, and the point A is the tangent
point of T = T (ε), with the tangent line starting from ε = −1.
Except in the special case shown in Fig. 9.4, only Eq. (9.24), in general, could be the
solution of Eq. (9.23). It shows again that in the case of weak discontinuous transverse
wave propagation, the transverse wave also only brings to shape change (dγ
= 0)
T
A
–1 0 eA e
Fig. 9.4. The tangent line at the point A of T = T (ε) curve just intersects at ε = −1.
of string but not strain disturbance (dε = 0). In the special case shown in Fig. 9.4,
Eq. (9.25) is a solution of Eq. (9.23). In such a case, the weak discontinuous longitudinal
wave and the weak discontinuous transverse wave coincide with each other. The change
of stretch force and strain, dT and dε can be considered as the result induced by the
longitudinal wave, while the shape change dγ can be considered as the result induced by
the transverse wave.
If dγ = 0, equation group (9-III) is reduced to the following familiar relations previously

discussed on the weak discontinuous longitudinal wave propagation theory:
⎧
⎪
⎪
⎨ ρ0 Cl d(v cos β) = dT
⎪ (9.26)
(9-IV) d(v cos β) = Cl dε (9.27)
⎪
⎪
⎪
⎩ d(v sin β) = 0 (9.28)
9
1 dT
where Ch has been rewritten as Cl = . Equation group (9-IV) is, in fact, equivalent
ρ0 dε
to Eq. (2.63) deduced in Chapter 2.
The four groups of equations obtained above, (9-I) to (9-IV), are the basic differential
equations for elastic–plastic longitudinal and transverse waves propagating in strings,
regardless of strong or weak discontinuous waves. They represent the dynamic conditions
and kinematic conditions across the wave front in different cases, respectively.
√
Obviously, transverse waves propagate with wave velocity √Ch = (1/ρ0 )(T /1 + ε), and
longitudinal waves propagate with wave velocity Cl = (1/ρ0 )(dT /dε), which on the
Lagrange plane (S0 –t plane) are, respectively, represented by the characteristics families
dS 0 = ±Ch dt (9.29)
and
dS 0 = ±Cl dt (9.30)
The basic differential relations (9-III) and (9-IV) give the differential relations across
the corresponding positive characteristics (namely along the negative characteristics).
Similar to what was discussed in Section 2.1, the differential relations across the negative
characteristics (i.e. along the positive characteristics) can be obtained just by exchanging
the corresponding positive and negative signs. Starting from these differential relations,
the problems of elastic–plastic waves propagating in strings can, in principle, be solved by
characteristics method, although they are certainly much complicated than that discussed
in Chapter 2.
It should be particularly noticed that the longitudinal wave only results in strain (dε
= 0)
without shape change (dr = 0), whereas the transverse wave only results in shape change
(dr
= 0) without strain (dε = 0); they are interinfluenced. In fact, for the longitu-
dinal wave, according to the equation group (9-IV), dT and dγ are related with the
change of tangential velocity d(v cos β). Therefore, once the transverse wave induces
a shape change of string, it must influence the longitudinal wave propagation through
the change of tangential velocity. On the other hand, according to Eq. (9.22), the
transverse wave velocity Ch is a function of strain ε; therefore, once the longitudi-
nal wave induces a strain change of string, it must also influence the transverse wave
propagation.
In different cases, either the longitudinal wave or the transverse wave could be weak
discontinuous or strong discontinuous, respectively; and the propagating precedence of the
longitudinal wave and the transverse wave could be different, either Cl > Ch or Cl < Ch .
The situation mainly depends on the stress–strain relation T = T (ε), the original shape
of string, and the initial-boundary
conditions. The
boundary condition could be given in
different forms, such as T boundary = T0 (t), v boundary = v0 (t) or ε boundary = ε0 (t).
With regard to the longitudinal wave, if we are limited to the stress–strain relation
T = T (ε) with the character of (d 2 T /dε 2 ) ≤ 0 and without unloading (∂ε/∂t
√ ≥ 0), then it
has been known from Chapter 2 that the longitudinal wave velocity Cl = (1/ρ0 )(dT /dε)
decreases with increasing strain ε. Therefore, under the general loading condition, the lon-
gitudinal wave is a kind of continuous wave (weak discontinuous wave). Only if, there is a
straight part on the T = T (ε) curve such as a linear elastic or a linear hardening part, and,
moreover, the boundary condition includes strong discontinuity (abrupt loading), then a
shock wave (strong discontinuous wave) can form.
With regard to the transverse wave, the situation is a little different from
√ the longitudinal
wave. When ε ≤ εA (Fig. 9.4), the transverse wave velocity Ch = (1/ρ0 )(T /1 + ε)
increases with increasing ε, so the later disturbances will catch up with the preceding
ones, and finally a strong discontinuous transverse wave will form √ and propagate with the
velocity corresponding to the maximum strain εm , i.e. Ch (εm ) = (1/ρ0 )[T (εm )/1 + εm ]
Because Ch (εm ) ≤ Cl (εm ), the transverse wave will be behind the longitudinal wave if
ε < εA , or will coincide with the longitudinal wave if ε = εA . But when ε > εA , the
transverse wave velocity Ch decreases with increasing ε. Roughly speaking, it seems
that there will be a series of weak discontinuous transverse waves that follow the strong
discontinuous transverse wave that corresponding to the εA . However, this is not the truth.
Because, when ε > εA , we have [T /(1 + ε)] > (dT /dε); in other words, when ε > εA ,
the propagation of transverse waves is faster than the longitudinal wave. Moreover, notice
that the strain disturbance ε results from longitudinal propagation, and the smaller the ε is
the faster the longitudinal wave propagates. Therefore, along with the transverse waves,
successively to catch up with the preceding longitudinal waves, ε ill gradually decrease.
Thus, the transverse wave will propagate faster and faster, and a strong discontinuous
transverse wave propagating with velocity Ch (εA ) will finally form.
It is thus clear that a strong discontinuous transverse wave will finally form. Under the
gradually loading boundary condition [(∂σ0 /∂t) ≥ 0, (∂σ0 /∂t)
= ∞], the strong discon-
tinuous transverse wave forms gradually. As long as the loading applied on the boundary
continuously increases with time (∂σ0 /∂t > 0), the strong discontinuous wave is an
unstable one, i.e. its amplitude and propagating velocity also continuously vary with time.
Only when the boundary condition contains a constant loading (∂σ0 /∂t = 0) can a stable
strong discontinuous transverse wave finally form. Obviously, when a constant loading
is abruptly applied, then a stable, strong discontinuous transverse wave will immediately
form at the beginning.
Finally, it should be noted again that the basic equation group (9-I) in the scalar form is
obtained by resolving the vector equations (9.2) and (9.4) into the tangent component and
the normal component with respect to the string segment in front of the strong discontinuous
interface. Alternatively, by resolving the vector equations (9.2) and (9.4) into the tangent
component and the normal component with respect to the string segment behind the strong
discontinuous interface, we have the following similar equation group:
⎧
⎪
⎪ ρ0 Ch [v2 cos β2 − v1 cos(β1 − γ )] = T2 − T1 cos γ (9.31)
⎪
⎪
⎪
⎪
⎨ ρ0 Ch [v2 sin β2 − v1 sin(β1 − γ )] = T1 sin γ (9.32)
(9-I )
⎪
⎪ v2 cos β2 − v1 cos(β1 − γ ) = Ch [ε2 − (1 + ε1 ) cos γ + 1] (9.33)
⎪
⎪
⎪
⎪
⎩ v sin β − v sin(β − γ ) = C (1 + ε ) sin γ
2 2 1 1 h 1 (9.34)
Obviously, only one of (9-I) and (9-I ) is independent. For example, Eq. (9.31) can be
obtained by first multiplying Eq. (9.7) by cos γ , then multiplying Eq. (9.8) by sin γ , and
then adding them. Moreover, from Eqs. (9.8) and (9.10), and from Eqs. (9.32) and (9.34),
we immediately have
1 T1 1 T2
Ch2 = =
ρ0 1 + ε 1 ρ0 1 + ε 2
which is just Eq. (9.69) obtained previously.
9.2 Semi-Infinite Straight String under Abrupt Constant-Velocity Oblique Impact
Let us now consider a semi-infinite straight string, which is subjected to an abrupt constant
velocity impact, V0 , β0 , at its end S0 . According to the discussions in the preceding
section, obviously, a stable, strong discontinuous transverse wave forms immediately
at the beginning. In such case, neither characteristic length nor characteristic time is
contained, and, consequently, according to the theory of dimensional analysis, only one
dimensionless variable S0 /V0 t is contained. This is the so-called “self-similar problem”.
On the S0 –t plot, it corresponds to a bunch of “centered waves”, as shown in Figs. 9.5
and 9.7.
If the loading is not too high so that εm < εA , then, as shown in Fig. 9.5, first a series of
elastic–plastic longitudinal waves propagates in the semi-infinite straight string, behind
which a “constant wave region” forms. Within this region, a slower, strong discontinuous
transverse wave propagates, accompanied with a break of straight string, but both sides
across the break point are still kept in a straight line, as shown in Fig. 9.6. The stretch
force T and strain ε are no longer varied in the straight line part of the string behind the
break point.
h (em )t
m)
t
l (e
S0 =C
0=
C
S
C 0t
S 0=
0
S0
Fig. 9.5. The elastic–plastic longitudinal waves and strong discontinuous transverse wave propa-
gating in a semi-infinite straight string.
e1 e0
b0
v0
Ch
e1
Fig. 9.6. Both sides across the break point of strong discontinuous transverse wave are straight
lines.
If the initial strain of string is ε0 and the initial particle velocity is zero, then for the straight
part of string in front of the strong discontinuous transverse wave front, we have
εL
v1 = Cl (ε)dε (9.35)
ε0
According to the equation group (9-I), across the strong discontinuous interface
we have
ρ0 Ch [v2 cos(β2 + γ ) − v1 ] = T2 cos γ − T1 (9.36)
ρ0 Ch [v2 sin(β2 + γ )] = T2 sin γ (9.37)
v2 cos(β2 + γ ) − v1 = Ch [(1 + ε2 ) cos γ − (1 + ε1 )] (9.38)
v2 sin(β2 + γ ) = Ch (1 + ε2 ) sin γ (9.39)
For the straight part of the string behind the strong discontinuous transverse wave front,
we have
v cos β = constant, v sin β = constant

By again using the given boundary condition v0 and β0 at the impact end, we have
v2 sin β2 = v0 sin(β0 − γ ) (9.40)

v2 cos β2 = v0 cos(β0 − γ ) (9.41)
where β0 is the intersection angle of the direction of impact velocity v0 with the ini-
tial string, as shown in Fig. 9.6. Since β0 , v0 , ε0 , and T = T (ε) are known, the seven
unknown variables ε1 , v1 , ε2 , v2 , β2 , Ch , and γ can be solved from the seven equations
from Eqs. (9.35)–(9.41).
If the loading is high enough so that εm > εA , then, as shown in Fig. 9.7, after a series
of elastic–plastic longitudinal wave propagation in the semi-infinite straight string, in
coincidence with the longitudinal wave of ε = εA , a strong discontinuous transverse wave
simultaneously propagates, and, consequently, the string breaks, although both sides of
the string across the break point are still kept in a straight line. However, in the straight
line part of the string behind the break point, some longitudinal waves still propagate with
a velocity slower than that of the transverse wave, and thus the stretch force T and strain ε
are still varied. Only after that, finally, there is the “constant wave region”, as shown in
Fig. 9.8.
t
A )t
t
em )
h (e
C(
=C
l
A )t
S0 =
(e
C
l
0=
S
C 0t
S 0=
0
S0
Fig. 9.7. Stress waves propagating in a semi-infinite straight string when εm > εA .
v0 eA e0
b0
m)
(e
Ch
C
l
em
Fig. 9.8. Slower longitudinal waves propagating behind the strong discontinuous transverse wave.
In such cases, Eqs. (9.35)–(9.40) are still held, but Eq. (9.41) should be replaced with the
following equation:
εm
v0 cos(β0 − γ ) = Cl (ε)dε + v2 cos β2 (9.42)
ε2
Moreover, we have an additional equation to represent the condition that the transverse
wave coincides with the longitudinal wave at ε = εA , namely Cl (εA ) = Ch (εA ), or

dT T (ε2 )
= (9.43)
dε ε=ε 1 + ε2
2
Thus, the eight unknown variables ε1 , v1 , ε2 , v2 , β2 , Ch , γ , and εm can be solved by those

eight equations, namely Eqs. (9.35)–(9.40), (9.42), and (9.43).
It can be seen, in the self-similar problem, that the transverse wave and longitudinal wave
can be dealt with separately, so that solving of the problem becomes much simpler.
In engineering applications, the general T = T (ε) relation is sometimes simplified as a

linear hardening one. In such cases, there are three possible situations:
Ts
(1) E1 > (Fig. 9.9)
1 + εs
Ts
(2) E1 < (Fig. 9.10)
1 + εs
Ts
(3) E1 = (Fig. 9.11)
1 + εs
where Ts and εs represent yield stress and yield strain, respectively, and E1 is the linear
hardening modulus.
By using the following relation:

E E1
c0 = and c1 =
ρ0 ρ0
m )t
h (e
T t t
1 v0
=C
C
Ts
S0 =
S 0 C1
C 0t
S 0= Ch
–1 0 es e 0
S0
Ts
Fig. 9.9. For E1 > .
1 + εs
T
t )t
Ct
(e s
1
Ts v0
S0 =
= Ch C1
S 0 C 0t
S 0= Ch
–1 0 e 0
S0
Ts
Fig. 9.10. For E1 < .
1 + εs
T
t
h (e
s )t
v0
C
Ts
1 t=
Ch=C1
C
C 0t
0=
S 0=
S
–1 0 e 0 S0
Ts
Fig. 9.11. For E1 = .
1 + εs
the above three situations can also be correspondingly represented as:
(1) c02 εs < c12 (1 + εs )

(2) c02 εs > c12 (1 + εs )
(3) c02 εs = c12 (1 + εs )
where c0 is the elastic longitudinal wave velocity and c1 is the plastic longitudinal wave
velocity for linear hardening materials, as has been discussed in Chapter 2. Under the
boundary condition of abrupt constant loading, both the transverse wave and the longi-
tudinal wave are strong discontinuous waves. Depending on different situations, in the
case of (1), the plastic longitudinal wave is faster than the transverse wave, and in the case
of (2), the transverse wave is faster than the plastic longitudinal wave, while in the case
of (3), both are in coincidence.
For most engineering materials, the situation (1) is usually satisfied. For example, the
typical data show that c12 /c02 ≈ 0.05, εs ≈ 0.002 for steel, and c12 /c02 ≈ 0.003, εs ≈ 0.001
for copper.
No matter what the situation is, problem solving is much simpler than when including a
general T = T (ε) relation.
9.3 Infinite Straight String under Abrupt Constant-Velocity Oblique Point-Impact
Let us now consider an infinite straight string, which is subjected to an abrupt constant-
velocity oblique impact. If the contact arc between the striker and the string is small
enough, it could be assumed that the loading is exerted at a point on the string. This is the
so-called “point-impact”. In such cases, as was discussed in the preceding section, it is
also a self-similar problem.
However, two different kinds of impact should be distinguished: (1) The impact loading
is exerted on a fixed particle of the string, such as S0 = 0, so that there is no relative
slide between the striker and the string. Such kind of impact is called “impact without
slide”. (2) The position of impact point on the string is varied with time, namely the striker
moves along a certain direction in the space but does not exert the impact loading on a
fixed particle of string. So, there is relative slide between the striker and the string. Such
kind of impact is called “impact with slide”.
When an infinite straight string is subjected to an abrupt constant-velocity oblique point-

impact, a break point appears at the impact point of string. Such a break point itself is also
a strong discontinuous interface. Describing by Lagrange variables, for the impact without
slide, the impact point itself does not propagate along the string, so it is a stationary strong
discontinuous interface; while for the impact with slide, the impact point is a moving
strong discontinuous interface propagating along the string with a certain velocity Ch0 ,
which is in general smaller than the transverse wave velocity Ch , as will be seen in the
following discussion.
Therefore, for the impact without slide, the problem is equivalent to deal with two semi-
infinite straight strings, the left side and the right side of the break point, under impact
loading. Such impact problem of semi-infinite straight string is solvable as has been
discussed in Section 9.2. However, for the impact with slide, since it is a varying boundary
problem, the loading boundary itself is undetermined. So, a further discussion about the
propagation of impact point itself is required.
The main difference between the impact-point strong discontinuity and the transverse
wave strong discontinuity is that there acts external force T0 on the impact point strong
discontinuous interface. Denoting the variables in front of and behind the impact point
by subscripts 3 and 4, respectively, as shown in Fig. 9.12, now the dynamic condition
T'1 v'1 v'2 b0 b'0 v2 v1 T1

g2 g1
b'2 v0 T0
v4 b2
T2 b3 v3 b4 T2
T3
T4
Fig. 9.12. An infinite straight string subjected to abrupt constant-velocity oblique point-impact with
slide.
Eq. (9.2) should be changed to:
ρ0 Ch0 (v4 − v3 ) = T4 + T3 + T0 (9.44)
There is no need for the kinematic condition to change, and, similar to Eq. (9.64), we have

∂u4 ∂u3
v4 − v3 = −Ch0 − (9.45)
∂s0 ∂s0
Noticing the symbols marked in Fig. 9.12, solving the vector equation (Eqs. (9.44) and
(9.45)) into the tangent direction and the normal direction of the vector stretch force T3 ,
we obtain the following four scalar equations:
⎧
⎪
⎪ ρ0 Ch0 [v4 cos(π − β4 − γ1 − γ2 ) − v3 cos β3 ]
⎪
⎪
⎪
⎪
⎪
⎪ = T4 cos(γ1 + γ2 ) + T0 cos(π − β0 − γ2 ) − T3
⎪
⎪ (9.46)
⎪
⎪
⎪ ρ C [v sin(π − β − γ − γ ) − v sin β ]
⎪
⎪
⎪ 0 h0 1 4 1 2 3 3
⎪
⎪
⎨ = −T4 sin(γ1 + γ2 ) + T0 sin(π − β0 − γ2 ) (9.47)
(9-V)
⎪
⎪ v4 cos(π − β4 − γ1 − γ2 ) − v3 cos β3
⎪
⎪
⎪
⎪
⎪
⎪ = Ch0 [(1 + ε4 ) cos(γ1 + γ2 ) − (1 + ε3 )]
⎪
⎪
(9.48)
⎪
⎪
⎪
⎪ v4 sin(π − β4 − γ1 − γ2 ) − v3 sin β3
⎪
⎪
⎪
⎪
⎩ = −Ch0 (1 + ε4 ) sin(γ1 + γ2 ) (9.49)
Or solving the vector equation (Eqs. (9.44) and (9.45)) into the tangent direction and the
normal direction of the vector stretch force T4 , we obtain
⎧
⎪
⎪
⎪ ρ0 Ch0 [v4 cos β4 − v3 cos(π − β3 − γ1 − γ2 )]
⎪
⎪
⎪
⎪
⎪ = −T4 + T0 cos(β0 − γ1 ) + T3 cos(γ1 + γ2 )
⎪
⎪ (9.50)
⎪
⎪
⎪
⎪ ρ0 Ch0 [v4 sin β4 − v3 sin(π − β3 − γ1 − γ2 )]
⎪
⎪
⎪
⎪
⎨ = T0 sin(β0 − γ1 ) − T3 sin(γ1 + γ2 ) (9.51)
(9-V )
⎪
⎪ v4 cosβ4 − v3 cos(π − β3 − γ1 − γ2 )
⎪
⎪
⎪
⎪
⎪
⎪ = −Ch0 [(1 + ε4 ) − (1 + ε3 ) cos(γ1 + γ2 )]
⎪
⎪
(9.52)
⎪
⎪
⎪
⎪ v4 sinβ4 − v3 sin(π − β3 − γ1 − γ2 )
⎪
⎪
⎪
⎪
⎩ = −Ch0 (1 + ε3 ) sin(γ1 + γ2 ) (9.53)
Obviously, among the four equations of dynamic condition, namely Eqs. (9.46), (9.47),
(9.50), and (9.51), arbitrary two are independent. Also, among the four equations of
kinematic condition, namely Eqs. (9.48), (9.49), (9.52), and (9.53), arbitrary two are
independent. Thus, we have totally four independent scalar equations.
Note that, in general, β0

= β0 , namely the T0 and v0 are not necessarily in the same
direction.
From Eqs. (9.47) and (9.49), and from Eqs. (9.51) and (9.53), the expression of Ch0 can
be obtained as
1 T4 sin(γ1 + γ2 ) − T0 sin(β0 + γ2 )
2
Ch0 =
ρ0 (1 + ε4 ) sin(γ1 + γ2 )
(9.54)
1 T3 sin(γ1 + γ2 ) − T0 sin(β0 − γ1 )
=
ρ0 (1 + ε3 ) sin(γ1 + γ2 )
Assuming T3
= T4 , and performing a further mathematical calculation, we obtain
1 T4 sin(β0 − γ1 ) − T3 sin(β0 + γ2 )
2
Ch0 = (9.55)
ρ0 (1 + ε4 ) sin(β0 − γ1 ) − (1 + ε3 ) sin(β0 + γ2 )
T4 (1 + ε3 ) − T3 (1 + ε4 )
T0 = sin(γ2 + γ1 ) (9.56)
(1 + ε3 ) sin(β0 + γ2 ) − (1 + ε4 ) sin(β0 − γ1 )
In addition, according to Eqs. (9.68) and (9.55), we have
T0 sin(β0 + γ2 ) T0 sin(β0 − γ1 )
2
Ch0 = Ch2 (ε4 ) − = Ch2 (ε3 ) −
ρ0 (1 + ε4 ) sin(γ1 + γ2 ) ρ0 (1 + ε3 ) sin(γ1 + γ2 )
Thus, it can be seen, Ch02 < C 2 (ε ), C 2 < C 2 (ε ). It means that the impact point
h 4 h0 h 3
propagates relatively slower in the case of impact with slide.
Eliminating Ch0 and T0 from the equation group (9-V) and (9-V ), two equations to relate
other variables can be obtained. For example, from Eqs. (9.49) and (9.53), we can obtain
v4 sin(β4 + γ1 + γ2 ) − v3 sin β3 v4 sin β4 − v3 sin(β3 + γ1 + γ2 )

= (9.57)
(1 + ε4 ) (1 + ε3 )
and eliminating T0 from Eqs. (9.46) and (9.47), then eliminating Ch0 from Eq. (9.55), we
can obtain
ρ0 [v4 sin(β4 +γ1 −β0 )−v3 sin(β3 +γ2 +β0 )]2 = [T4 sin(β0 −γ1 )−T3 sin(β0 +γ2 )]
·[(1+ε4 )sin(β0 −γ1 )−(1+ε3 )sin(β0 +γ2 )] (9.58)
It should be pointed out that the particle velocities across the strong discontinuous impact
point are discontinuous v3
= v4 , and, moreover v3 and v4 are all unequal to v0 . However,
the normal component of both impact velocity and particle velocity at the impact point
should be equal to each other. This is the so-called contact condition. According to this
condition, similar to the boundary condition (Eq. (9.40)), we have
v4 sin β4 = v0 sin(β0 − γ1 ) (9.59)

v3 sin β3 = v0 sin(β0 + γ2 ) (9.60)
In fact, v̄0 is also the Euler velocity of impact point propagating in spatial coordinates.
Similar to that discussed in Section 2.4, the following relation between Euler wave velocity
and Lagrange wave velocity should be satisfied:
v0 = v3 + (1 + ε3 )Ch0 S3 = v4 + (1 + ε4 )Ch0 S4
where S3 and S4 are unit vectors in the directions of T3 and T4 , respectively. By using
this relation, Eqs. (9.59) and (9.60) can also be obtained immediately.
Thus, in the case of εm > εA , synthesizing all the discussions given above, for the string
on the right side of impact point as shown in Fig. 9.12, we must have
⎧ ε1
⎪
⎪
⎪
⎪ v1 = c(ε)dε (a)
⎪
⎪
⎪
⎪
ε0
⎪
⎪
⎪
⎪ ρ0 Ch [v2 cos(β2 + γ1 ) − v1 ] = T2 cos γ1 − T1
⎪
⎪
(b)
⎪
⎪
⎪
⎪ ρ0 Ch [v2 sin(β2 + γ1 )] = T2 sin γ1 (c)
⎪
⎪
⎪
⎪
⎪
⎪ v cos(β2 + γ1 ) − v1 = Ch [(1 + ε2 ) cos γ1 − (1 + ε1 )] (d)
⎨ 2
(9-VI) v2 sin(β2 + γ1 ) = Ch (1 + ε2 ) sin γ1 (e)
⎪
⎪
⎪
⎪ ε4
⎪
⎪
⎪
⎪ v4 cos β4 = c(ε)dε + v2 cos β2 (f )
⎪
⎪
⎪
⎪
ε2
⎪
⎪
⎪
⎪ dT T (ε2 )
⎪
⎪ = (g)
⎪
⎪ + ε2
⎪
⎪
dε ε=ε2 1
⎪
⎪
⎩
v4 sin β4 = v2 sin β2 (h)
For the string on the left side of impact point, there are eight similar equations, which can
be obtained from (9-VI) by using v1 , ε1 , v2 , ε2 , β2 , T1 , T2 , γ2 , Ch , v3 , β3 , and ε3 to cor-
respondingly replace v1 , ε1 , v2 , ε2 , β2 , T1 , T2 , γ1 , Ch , v4 , β4 , and ε4 , respectively . Since
the relation T = T (ε) is a known function, there are 20 unknown variables in the above
16 equations. At the impact point, after eliminating Ch0 and T0 , there are four more equa-
tions, Eqs. (9.57)–(9.60), although a new unknown variable β0 is introduced. Thus, we
have a total of 21 unknown variables but 20 equations, which are established from the
dynamic conditions, kinematic conditions, and boundary conditions at the impact point.
In order to solve the problem, a supplement equation is required, which should come from
the friction condition at the impact point in the case of impact with slide.
Note that the equation group (9-IV) established is based on a more complicated case of
εm > εA . If in the case of εm < εA , the transverse wave will propagate in a constant wave
region, then the last three equations in (9-IV) can be replaced by ε4 = ε2 , v4 = v2 , and
β4 = β2 , and solving of the problem therefore becomes simpler.
Now, let us discuss the friction condition.
Assume that the friction between the striker and the string follows the Column’s friction
theorem. In other words, the friction force equals the component of external force T0 in
the normal direction of string multiplied by the friction coefficient f . However, in the
case of point impact, the impact point is a break point. Mathematically, neither normal
direction nor tangent direction is meaningful with regard to a break point. Therefore, we
meet difficulty in applying the Column’s friction theorem to the point impact. For this
reason, we first regard the impact point as an infinitesimal arc with radius R and included
angle γ . Of course, either the normal direction or the tangent direction is determinable
at each point of this infinitesimal arc. Meanwhile, we regard the external force T0 as the
resultant force of the external load distributed on this arc. If denote the distributed normal
load by N , the distributed friction force by F , and the friction coefficient by f , then,
according to the Column’s friction theorem, we have
F =f ·N
Let the angular bisector of γ be the Y axis and its normal direction the X axis, the resultant
of N as P = Px i + Py j, and the resultant of F as Q = Qx i + Qy j. Then we have
γ /2
Px = RN sin αdα = 0
−γ /2
γ /2
Py = RN cos αdα = RNγ
−γ /2
γ /2
Qx = RF cos αdα = RFγ = fRNγ
−γ /2
γ /2
Qy = RF sin αdα = 0
−γ /2
Thus, it can be seen that: (1) P = Py j = RNγ j, i.e. the resultant of distributed normal
load is in the direction of Y axis, or the direction of the angular bisector of γ ; (2) Q =
Qx i = fRNγ i, i.e. the resultant of distributed friction force is in the direction of X axis, or
the normal direction of the angular bisector of γ ; (3) |Q| = f |P|, i.e. when the Column’s
friction theorem is expressed in the resultant form herein, the normal direction is equivalent
to the direction of the angular bisector of γ while the tangent direction is equivalent to
its normal direction. When R → 0, the infinitesimal arc approaches a point. This is just
the situation of point impact. Thus, from now on, the normal direction and the tangent
direction at the break point, in the sense of applying the Column’s friction theorem, can
be understood as the direction of the angular bisector of the included angle of string at the
break point and its normal direction, respectively, as shown in Fig. 9.13. Mathematically,
they are just the average of the normal and tangent of the two sides of the included
angle.
When the component of external force T0 in the tangent direction of string is smaller
than friction force, the striker cannot slide along the string. This belongs to the problem
of impact without slide, which has been solved, as discussed previously. Only when the
tangent component of T0 is large enough to overcome the friction force, the striker will
slide along the string. In other words, in the case of impact with slide, the included angle
Angular bisector
p γ1−γ 2
+
2 2
g2 b0' g1
b2
Fig. 9.13. The angular bisector of the included angle of string at the break point and its normal
direction.
between the external force T0 and the normal direction of string [(π/2)+((γ1 −γ2 )/2)−β0 ]
equals the friction angle ϕ, i.e.
π γ 1 − γ2
+ − β0 = ϕ = arctg f (9.61)
2 2
Having such a supplement equation, the problem is solvable.
It seems very complicated to solve 21 unknown variables by 21 equations. But, in practice,

by using Eqs. (9.59) and (9.60), all the other unknown variables in the equation group
(9-VI) can be expressed as the functions of only ε1 and ε1 , and then substituting into
Eqs. (9.57), (9.58), and (9.61), the problem is finally reduced to solve three unknown
variables ε1 , ε1 , and β0 by three equations.
The simplest case is, of course, the case without friction. Then, let f = 0 in Eq. (9.61),
and we have
π γ 1 − γ2
β0 = + (9.62)
2 2
It indicates that the direction of external force T0 is just the direction of angular
bisector the included angle at impact point. Substituting Eq. (9.62) into Eq. (9.54),
we have
T3 = T4 , ε3 = ε4 (9.63)
This indicates that without friction, the stretch forces in the left-side and right-side string
across the impact point are equal, and the same as for the strains. In addition, it could be
imagined that, without friction, any oblique impact under the condition of β0
= (π/2)
is a kind of impact with slide. Having Eq. (9.63), the β0 is not necessary to be taken
into consideration in solving the problem. Because, from the equation group (9-VI) and,
moreover, from Eqs. (9.57), (9.59), (9.60), and (9.63), there are 20 equations, enough to
solve the 20 unknown variables.
The problem of infinite straight string subjected to abrupt constant-velocity oblique point
impact was first studied by (Rakhmatulin) (1945b). However, as one of
the basic equations, Euler equation, which is usually used in mechanical engineering to
describe the friction between the pulley and the rope, was adopted by him to describe the
dynamic condition at impact point, as
T4 − ρ0 u2s = ef (γ1 +γ2 ) (T3 − ρ0 u2s ) (9.64)
Note that Eq. (9.64) is deduced from the static equilibrium condition, while the problem
we discussed herein is a dynamic problem. Only in the case without friction, a result of
T3 = T4 , which occasionally coincides with Eq. (9.63), can be obtained from Eq. (9.64).
Such a coincidence provides us with a possibility to make use of all results obtained by
(Rakhmatulin) under the condition without friction.
In the following, let us discuss more concretely the situation of impact with slide but
without friction. For convenience, it is limited to discuss the situation of εm < εA . In such
cases, the equation group (9-VI), together with Eqs. (9.57), (9.59), (9.60), and (9.62), after
some mathematical calculations, transform to the following group of equations:
ε1
v1 = c(ε)dε
ε0
T2 = T1 , ε1 = ε2

1 T1
Ch =
ρ0 1 + ε 1
v2 cos β2 − v1 cos γ2 = Ch (1 + ε1 )(1 − cos γ1 )
v2 sin β2 + v1 sin γ2 = Ch (1 + ε1 ) sin γ1
v4 = v2
β4 = β2
ε4 = ε2 , T4 = T2
T3 = T4 , ε3 = ε4
v2 sin β2 = v0 sin(β0 − γ1 )
v2 sin β2 = v0 sin(β0 + γ2 )
v2 sin(β2 + γ1 + γ2 ) − v2 sin β2 = v2 sin β2 − v2 sin(β2 + γ1 + γ2 )
The first seven equations are deduced from Eqs. (9-VI) for the right-side string. When we
consider the left-side string, there are seven other similar equations. After a further arrange-
ment, the unknown variables can be reduced to ε(=ε4 =ε2 =ε1 =ε3 =ε2 =ε1 ), v1 (=v1 ),
b(= b ), v2 (=v4 ), β2 (=β4 ), v2 (=v2 ), β2 (=β3 ), and γ2 . Those nine unknown variables
can be solved by the following nine equations:

⎧ ε
⎪
⎪
⎪
⎪ v1 = c(ε)dε (a)
⎪
⎪
⎪
⎪
ε0
⎪
⎪
⎪
⎪
⎪
⎪ 1 T
⎪
⎪ Ch = (b)
⎪
⎪ ρ 0 1+ε
⎪
⎪
⎪
⎪
⎪
⎪ v2 cos β2 − v1 cos γ2 = Ch (1 + ε)(1 − cos γ1 ) (c)
⎪
⎪
⎪
⎪
⎪
⎨ v2 sin β2 + v1 sin γ1 = Ch (1 + ε) sin γ1 (d)
(9-VII)
⎪
⎪ v cos β2 − v1 cos γ2 = Ch (1 + ε)(1 − cos γ1 ) (e)
⎪ 2
⎪
⎪
⎪
⎪ v2 sin β2 + v1 sin γ2 = Ch (1 + ε) sin γ2
⎪ (f )
⎪
⎪
⎪
⎪
⎪
⎪ v2 sin β2 = v0 sin(β0 − γ1 ) (g)
⎪
⎪
⎪
⎪
⎪
⎪
⎪ v2 sin β2 = v0 sin(β0 + γ2 ) (h)
⎪
⎪
⎪
⎪ v2 sin β2 cos(γ1 + γ2 ) + v2 cos β2 sin(γ1 + γ2 ) − v2 sin β2
⎪
⎪
⎪
⎪
⎩
= v2 sin β2 − v2 sin β2 cos(γ1 + γ2 ) − v2 cos β2 sin(γ1 + γ2 ) (i)
Among these equations, from Eqs. (d) and (g) above, we obtain
v0 sin β0 v0 sin β0
tgγ1 = = (9.65a)
Ch (1 + ε) − v1 + v0 cos β0 ch + v0 cos β0
where
ch = Ch (1 + ε) − v1 (9.65b)
Its physical meaning is Euler velocity of transverse wave. Similarly, from Eqs. (e) and (h),
we obtain
v0 sin β0 v0 sin β0
tgγ2 = = (9.66)
Ch (1 + ε) − v1 − v0 cos β0 ch − v0 cos β0
From Eqs. (9.65) and (9.66), the following equations can be solved:
⎧ ⎫
⎪ ch (tgγ2 − tgγ1 ) ⎪
⎪
⎪ v cos β = ⎪
tgγ1 + tgγ2 ⎪
0 0
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ 2c tgγ tgγ ⎪
⎪
⎨v0 sin β0 = h 1 2 ⎬
tgγ1 + tgγ2 (9.67)
⎪
⎪ ⎪
⎪
⎪
⎪ 2tgγ1 tgγ2 ⎪
⎪
⎪
⎪tgβ = ⎪
⎪
⎪
⎪
0
tgγ − tgγ ⎪
⎪
⎪
⎩
2 1 ⎪
⎭
ctgγ1 = 2ctgβ0 + ctgγ2
In practice, for the certain known ε0 and β0 , first presume a γ1 , and calculate γ2 from
Eq. (9.67). For example, some results calculated by such way are listed in Table 9.1
[ (Rakhmatulin), 1945b].
Table 9.1. Values of γ1 .

γ2
β0 10◦ 13◦ 18◦ 23◦ 30◦ 35◦ 40◦ 50◦ 55◦
90◦ 10◦ 00 13◦ 00 18◦ 00 23◦ 00 30◦ 00 35◦ 00 40◦ 00 50◦ 00 —
70◦ 8◦ 50 11◦ 10 14◦ 40 18◦ 00 22◦ 10 24◦ 50 27◦ 30 32◦ 30 —
50◦ 7◦ 40 9◦ 20 11◦ 50 14◦ 00 16◦ 20 17◦ 50 19◦ 10 21◦ 40 —
30◦ 6◦ 10 7◦ 20 8◦ 40 9◦ 40 10◦ 50 11◦ 30 12◦ 10 13◦ 00 13◦ 30
Since γ1 and γ2 are known and v1 and Ch are functions of only ε, then according to
Eqs. (c)–(f), all v2 cos β2 , v2 sin β2 , v2 cos β 2 , and v2 sin β 2 can be expressed as functions
of only ε. Substituting them into Eq. (i), ε can be solved, and thus all other unknown
variables can be determined. Finally, from Eq. (9.67), the impact velocity v0 , which is
required to bring in such value of ε, can be determined. Such a cut-and-trial method was
proposed by (Rakhmatulin), (1945b).
For linear hardening materials, the problem becomes much simpler. Let the stretch force–
strain relationship be expressed as

T = Eε, ε < εs
T = Ts + E1 (ε − εs ) ε > εs
where Ts and εs are yield stretch force and yield strain, respectively, E the elastic modulus,
and E1 the linear hardening modulus. In such case, Eqs. (a) and (b) are deduced to
v1 = c0 (εs − ε0 ) + c1 (ε − εs )

c02 εs + c12 (ε − εs )
Ch =
1+ε
√ √
where c0 = E/ρ0 and c1 = E1 /ρ0 denote the elastic wave velocity and plastic

wave velocity, respectively. Introducing c̄12 = (c1 /c02 ), B(=Ch /c0 ), and v̄0 (=v0 /c0 ), let
εs = 0.002, for two different initial strains ε0 (ε0 = 0 and ε0 = 0.001) and four different
impact angles β0 (β0 = 30◦ , 50◦ , 70◦ and 90◦ ), the numerical results calculated for ε, B,
and v̄0 are given in Table 9.2, and the corresponding plot ε = ε(v0 ) is given in Fig. 9.14
[ (Rakhmatulin), 1945b].
In the case of normal impact, we have

π
β0 =
2
v0 = v2 = v2
γ1 = γ2 = γ
π
β2 = β0 − γ1 = −γ
2
π
β 2 = β0 − γ2 = + γ
2
386
calculated under different conditions.
Table 9.2. Values of ε and B
β0 = 90◦ , ε0 = 0, c̄12 = 0.05, εs = 0.002
v̄0 0.0135 0.0167 0.0192 0.0245 0.0347 0.045 0.0518 0.064 0.0661 0.0742 0.085
ε 0.0020 0.006 0.010 0.020 0.04 0.060 0.0800 0.100 0.120 0.140 0.160

B 0.0426 0.0441 0.0453 0.0481 0.0530 0.05696 0.06045 0.0656 0.0655 0.068 0.0707
β0 = 90◦ , ε0 = 0.001, c̄12 = 0.05, εs = 0.002

v̄0 0.00978 0.1295 0.0164 0.02035 0.0228 0.0283 0.0391 0.0381 0.0423 0.0511
ε 0.002 0.006 0.010 0.016 0.020 0.030 0.040 0.050 0.060 0.080

B 0.0436 0.0491 0.0463 0.0481 0.0491 0.0518 0.054 0.0563 0.0579 0.0614
β0 = 70◦ , ε0 = 0.001, c̄12 = 0.05, εs = 0.002

v̄0 0.0139 0.0182 0.0254 0.0311 0.0376 0.0554
ε 0.006 0.010 0.0219 0.0326 0.0444 0.0787

B 0.0450 0.0464 0.0499 0.0525 0.0551 0.0612
β0 = 50◦ , ε0 = 0.001, c̄12 = 0.05, εs = 0.002

v̄0 0.0113 0.0148 0.0188 0.02426 0.0286 0.0337 0.0423
ε 0.0002 0.0042 0.008 0.0146 0.0207 0.0323 0.0455

B 0.0431 0.0445 0.0456 0.0478 0.0496 0.00525 0.0555
β0 = 30◦ , ε0 = 0.001, c̄12 = 0.05, εs = 0.002

v̄0 0.0142 0.0183 0.0217 0.0263 0.0297 0.0332 0.036 0.0388 0.0425
ε 0.0004 0.0023 0.0058 0.0082 0.0114 0.015 0.0185 0.0223 0.029

B 0.043 0.0439 0.0446 0.0458 0.0472 0.0487 0.049 0.051 0.0517
e, x0.001
b0=90°
70
b0=70°
60
e0=0.001; es=0.002;
50
C 21=0.05
b0=50°
40
30
b0=30°
20
10
v0
0 10 20 30 40 50
Fig. 9.14. The relation ε = ε(v0 ) under different impact angle β0 .
Therefore, Eqs. (9-VII) are reduced to:

⎧ ε
⎪
⎪
⎪
⎪ v1 = c(ε)dε
⎪
⎪
⎪
⎪
ε0
⎪
⎪
⎪
⎪
⎨ 1 T
Ch =
(9-VIII)
⎪ ρ0 1 + ε
⎪
⎪
⎪
⎪ π
⎪
⎪ − γ − v1 cos γ = Ch (1 + ε)(1 − cos γ )
⎪
⎪
v0 cos
⎪
⎪ 2
⎪
⎩ π
v0 sin − γ + v1 sin γ = Ch (1 + ε) sin γ
2
from which four unknown variables ε, v1 , Ch , and γ can be solved. The last two equations,
after some mathematical calculations, can be reduced to
v0 ⎫
tgγ = ⎪
⎬
(1 + ε) − v1
Ch (1 + ε) ⎪ (9.68)
γ= ⎭
Ch (1 + ε) − v1
from which, eliminating γ , we have

2 2
Ch (1 + ε) v0
− =1
Ch (1 + ε) − v1 Ch (1 + ε) − v1
or
v02 = 2Ch (1 + ε)v1 − v12
Substituting the v1 and Ch of the first two equations in (9-VIII) into above equation, for
a given v0 , the ε can be solved, and, consequently, Ch , v1 , and γ can also be determined.
For example, as to elastic string, T = Eε we then have
v1 = c0 (ε − ε0 )
9
ε
Ch = c0
1+ε
< ;
v0
v̄0 = = 2(ε − ε0 ) ε(1 + ε) − (ε − ε0 )2 (9.69)
c0
When ε0 ≈ 0, ε 1, from Eq. (9.69), we have

4/3
v̄0
ε≈ √ 3
(9.70)
4
Consequently,
2/3
v̄0 2/3 1/3
b ≈ c0 √ 3
= 0.8v0 c0 (9.71)
2
Again, from Eq. (9.68), we obtain

9
v0 v0
tgγ ≈ 2/3 1/3
= 1.25 · 3 (9.72)
0.8v0 c0 c0
It is thus clear that while v0 is small, γ is still large. Assume that the elastic limit strain of
material is εs = 0.002, and it can be seen from Eq. (9.70) that the normal impact velocity
required for the beginning of yield in the string is considerably large, v0 ≈ 55 m/s.
9.4 Prestretched Strings Subjected to Transverse Impact
Now, let us discuss the problem of prestretched strings subjected to transverse impact, to
see how the prestretch force influences the propagating character of both the longitudinal
wave and transverse wave in the string (Wang et al., 1992).
Consider an infinite straight string with original line density ρ0 . The string is originally
at rest but prestretched by a force T0 (and a corresponding prestrain ε0 ) and subjected to
a constant-velocity normal impact, as shown in Fig. 9.15, where the string is coincident
with the X axis, and the impact velocity V is coincident with the Y axis (V is positive
in the positive direction of Y axis). Obviously, there must be longitudinal and transverse
A*
Vdt
V B*
A P* T(
V 2 e
2)
Vt
B T1(e1)
g P Qx U Q F
X, x
CS
cS t=[CS (1+e 2 )-U]t cS t=[CS (1+e 2 )-U]dt Udt CL
Fig. 9.15. An infinite straight string with prestretched force under transverse impact.
waves in succession, propagating in the string with material velocity (Lagrange velocity)
of Cl and Ch , respectively. Due to the symmetry of the problem, we only need to discuss
the motion of the half part of the string, X ≥ 0.
When the problem is described by material coordinates (Lagrange coordinates) for the
strong discontinuous longitudinal wave front, we have the following dynamic compatibil-
ity condition (momentum conservation) and kinematic compatibility condition (continuity
of displacement)
[T ] = ρ0 Cl [U ]
[U ] = Cl [ε]
where the stretch force T (=σ A0 ) and strain ε are positive for the tensile case, while the
particle velocity in the X direction U is positive when it directs to the impact point (i.e.
–X direction). Noticing the sign definition for U , it can be seen that the above two equations
are actually the dynamic compatibility condition (Eq. 2.57) and kinematic compatibility
condition (Eq. 2.55), respectively, deduced in Chapter 2 for the strong discontinuous
longitudinal waves in bars.
For the string initially at rest and prestretched by force T0 , if we denote the quantities
behind the strong discontinuous longitudinal wave front in string by subscript 1, then the
above two equations can be rewritten, respectively, as
T1 − T0 = ρ0 Cl U1 (9.73)
U1 = Cl (ε1 − ε0 ) (9.74)
For the strong discontinuous transverse wave front, the general forms of the dynamic con-
dition and the displacement continuity condition across the wave front have been deduced
previously, as expressed by Eq. (9-I). In the current case, if we notice v1 = U1 , v2 = V ,
b1 = 0, β2 +γ = π/2, and denote the quantities behind the strong discontinuous transverse
wave front in string by the subscript 2, then Eqs. (9-I) can be rewritten as
T2 cos γ − T1 = −ρ0 Ch U1 (9.75)

T2 sin γ = ρ0 Ch V (9.76)
−U1 = Ch [(1 + ε2 ) cos γ − (1 + ε1 )] (9.77)
V = Ch (1 + ε2 ) sin γ (9.78)
Obviously, from Eqs. (9.73) and (9.74), the wave velocity of longitudinal Cl can be
immediately obtained:

1 T1 − T 0
Cl = (9.79a)
ρ0 ε1 − ε 0
It coincides with the longitudinal wave velocity in the bar (Eq. 2.59).
From Eq. (9.79a), it can be seen that when the constitutive relation of string material
T = T (ε) is linear, the prestretch force T0 does not influence the longitudinal wave veloc-
ity. However, when the constitutive relation of string material T = T (ε) is nonlinear,
the prestretch force T0 does influence the longitudinal wave velocity. Similar to the situ-
ation discussed on the strong discontinuous longitudinal wave in bars, if (d 2 T /dε 2 ) < 0
[namely, T = T (ε) is a upwards convex curve], the strong discontinuous longitudinal
wave will be transformed into the weak discontinuous longitudinal wave (continuous
wave), and, correspondingly, Eq. (9.79) is reduced to

1 dT A0 dσ
Cl = = (9.79b)
ρ0 dε ρ0 dε
Conversely, if (d 2 T /dε 2 ) > 0 (namely, T = T (ε) is a downwards concave curve),

then the propagating velocity of strong discontinuous longitudinal wave varies with the
prestretch force T0 ., since the strong discontinuous wave velocity depends on the slope of
Rayleigh line, while the slope of Rayleigh line varies with the position of its initial point,
which corresponds to the prestretch force.
For the transverse wave, from Eqs. (9.75) and (9.77), and from Eqs. (9.76) and (9.78), two
expression forms of the transverse wave velocity can be obtained:

1 T2 cos γ − T1
Ch = (9.80a)
ρ0 (1 + ε2 ) cos γ − (1 + ε1 )

1 T2
Ch = (9.80b)
ρ0 1 + ε 2
In order to satisfy both these equations, similar to that discussed for Eqs. (9.12)–(9.14),
in general they must have
T2 = T1 (9.81a)
ε2 = ε1 (9.81b)
It means that the transverse wave only induces string shape change but not strain
disturbance.
Thus, the current problem, including all the momentum conservation conditions and dis-
placement continuity conditions for both longitudinal and transverse waves, can finally be
summed up as solving the following four independent equations:
ρ0 Cl U = T − T0 = Td (9.82a)
U = Cl (ε − ε0 ) = Cl εd (9.82b)
ρ0 Ch V = T sin γ (9.82c)
V = Ch (1 + ε) sin γ (9.82d)
Herein, for convenience, by noticing Eq. (9.81), the subscripts 1 and 2 are omitted.
Obviously, when the line density ρ0 and the constitutive relation T = T (ε) of the string
are known, and the prestretch force T0 (and, correspondingly, the prestrain ε0 ) and the
transverse impact velocity V are given, the above four independent equations contain
five unknown variables: Cl , U , T , Ch , and γ . Therefore, once any one of them can be
experimentally determined, the other four can all be solved.
Now, let us discuss an example of how to measure the wave velocity in a prestretched
string.
9.4.1 Experimental investigation of wave velocity in a prestretched string
The propagation of longitudinal and transverse waves in the prestretched string was
experimentally investigated by Wang (1992). The schematic of experiment is shown in
Fig. 9.16.
A gas gun system was used to fire projectiles at vertically suspended yarns with length L.
The projectile is composed of a nylon sabot with an attached sharp blade, cut from a
razor blade, as shown in Fig. 9.17. A weight is suspended underneath the yarn to apply a
prestretch force T0 , which can be adjusted by changing the weight. The impact velocity
of projectile V is determined by measuring the time difference ∆t0 passing two laser
beams with a fixed distance ∆Y0 , namely V = ∆Y0 /∆t0 . Two piezoelectric sensors are
connected on the top end and lower end of the yarn, respectively. The distance from the
impact point to the top sensor L1 is larger than the distance from the impact point to the
lower sensor L2 = L − L1 , so that the longitudinal wave velocity Cl can be directly
Laser beams
∆Y0 L1
Gun barrel
Sabot Razor
L
Y
V Test yarn
T0
Fig. 9.16. Schematic of the experiment set-up.
n
r
(a) (b)
Fig. 9.17. The projectile attached a sharp razor blade.
measured, Cl = (L1 − L2 )/(t1 − t2 ), where t1 and t2 are the arrival times of longitudinal
wave at top sensor and lower sensor, respectively.
A typical oscilloscope record for the longitudinal wave velocity measurement for Kevlar
yarn under the transverse impact at V = 81 m/s is shown in Fig. 9.18. From the recorded
signal jump corresponding to the arrival of the stress wave, it can be deduced that the
recorded longitudinal wave is a strong discontinuous one. It means that the constitu-
tive relation T = T (ε) of the tested material, Kevlar, is either linear or nonlinear and
downwards-concave. If it is the former, then the longitudinal wave velocity does not vary
with the stretch force; while if it is the latter, the longitudinal wave velocity increases with
increasing stretch force.
The experimentally measured longitudinal wave velocities for Kevlar under the conditions
of different prestretch forces and different impact velocities are collected in Table 9.3.
As can be seen, under the same prestretch force T0 , the longitudinal wave velocity Cl
almost does not vary with the impact velocity V , but under the same impact velocity V ,
500. V/DIV 29.5v0/DIV 500. V/DIV 29.5v0/DIV

CH2
Kevlar
T0 = 2.04 N (208 g)
V0 = 81 m/s
b
CH1
CH2 a
a
CH1
Fig. 9.18. Typical oscilloscope record for the longitudinal wave velocity measurement for Kevlar
yarn under the transverse impact at V = 81 m/s. (Wang, 1992)
Table 9.3. The longitudinal wave velocities Cl for Kevlar under the conditions of different
prestretch forces T0 (prestress σ0 ) and different impact velocities V .
T0 (N) 2.04 3.02 5.96 10.9 10.9 10.9 10.9 20.7 30.5
σ0 (MPa) 32 47 93 170 170 170 170 324 478
V (m/s) 81 79 81 54.5 81 138 170 82 81
Cl (km/s) 9.04 9.13 9.22 9.35 9.35 9.36 9.27 9.52 9.79
the Cl of Kevlar increases with increasing prestretch force, showing an approximately

linear relation, as shown by the oblique line k in Fig. 9.19. In the same figure, the
measured Cl − T0 relations for the polymer yarn Spectra and the Ni-Cr alloy string
are also given, shown by the oblique line s and the horizontal line n, respectively.
Thus, under the present test conditions, the T = T (ε) relation for both polymer yarn
Kevlar and Spectra are nonlinear and downwards-concave, while for Ni-Cr alloy is
linear.
A high-speed image converter camera (Imacon 792), operating at a framing interval of

20 µs, was used to record the break angle γ and the failure processes of the test yarns.
Typical high-speed sequences for impacts on Kevlar yarns with prestress of 170 MPa at
impact velocity of (a) 55 m/s and (b) 170 m/s are given in Figs. 9.20(a and b), respectively
(Wang et al., 1992).
15000
14000 s
13000
12000
11000
C1 (m/s)
10000 k
9000
8000
7000
6000
n
5000
4000
0 100 200 300 400 500 600 700
Pre-tensile Stress (MPa)
Fig. 9.19. Experimental results for Kevlar (curve k), Spactra (curve s), and Ni-Cr alloy (curve n),
showing how Cl varies with T0 .
It is evident that the failure process of a Kevlar yarn under transverse impact is going on
by a series of successive broken process of one fiber by one fiber. When a fiber is broken,
it springs back, as the stored tension is released. Under a given prestretch force, with
increasing the impact velocity, the break angle γ increases, and the yarn is broken in a
shorter time.
9.4.2 Experimental study on constitutive relationship of string materials
Once the longitudinal wave velocities of string with different prestretch forces are exper-
imentally determined, the instantaneous stress–strain relation can be deduced. This is
similar to solving the second kind of inverse problems previously discussed on the stress
wave propagating in bars, i.e. to inversely solve the constitutive relation of materials from
the measured wave signals.
For the polymer yarn Kevlar, the measured data of Cl versus T0 shown in Fig. 9.19,
by means of the least square fitting, can be expressed as the following linear
relation
Cl = 9.10 + 1.56 × 10−3 σ0 (9.83)
where the unit of Cl is km/s and the unit of σ0 is MPa. Substituting the above equation
into Eq. (9.79b) and then integrating it, the instantaneous nonlinear σ = σ (ε) relation
2 4 6 8
s f
1 3 5 7
(a)
s f
2 4 6 8
1 3 5 7
(b)
Fig. 9.20. High-speed sequences for impacts on Kevlar yarns with prestress of 170 MPa at impact
velocity of (a) 55 m/s and (b) 170 m/s. Interframe time 20 µs.
can be expressed as

1
σ = 5.85 × 103 −1 (9.84a)
1 − 20.8ε
or
σ
ε= (9.84b)
20.8σ + 1.216 × 105
which is plotted in Fig. 9.21 by the solid line. In the same figure, the instantaneous nonlinear
σ = σ (ε) relation for polymer yarn Spectra, which is determined by the same method,
is plotted by dash line for comparison.
1.2
Spectra
Kevlar
1.0
Dynamic Stress (GPa)
0.8
0.6
0.4
0.2
0.0
0.000 0.001 0.002 0.003 0.004 0.005
Dynamic Strain
Fig. 9.21. The instantaneous nonlinear σ = σ (ε) relations for Kevlar and Spectra.
9.4.3 Determination of transverse wave velocity from longitudinal wave velocity

measurements
As can be seen in Fig. 9.15, there exists a simple geographical relation between the trans-
verse impact velocity V and the spatial wave velocity (Euler wave velocity) of transverse
wave ch as
V = ch tan γ (9.85)
On the other hand, similar to the derivation of Eq. (2.11), in the current situation, there
should exist the following relation between the wave velocity ch and Lagrange wave
velocity Ch (ref. Eq. 9.65b):
ch = Ch (1 + ε) − U (9.86)
The above two equations together with Eqs. (9.82a,b,c,d) are totally six equations, which
contain 11 variables: ρ0 , Cl , Ch , ch , U, V , γ , T , ε, T0 , and ε0 . In general, ρ0 , T0 , and ε0
are known, and the transverse impact velocity can be measured in the experiment, so the
number of remained unknown variables is seven. Therefore, once the longitudinal wave
velocity Cl can be measured in the experiment, the other six unknown variables can be
determined by the above six equations.
Let us consider Kevlar yarn as an example. According to the measured data of longitudinal
wave velocity Cl , all other variables calculated are collected in Table 9.4. Note that for
the break angle γ , which is an important quantity characterizing the transverse wave
Table 9.4. The measured longitudinal wave velocity Cl and the calculated other variables for
Kevlar under the conditions of different prestretch forces and different transverse impact velocities
(Wang, 1992).
T0 (N) 2.04 3.02 5.96 10.9 10.9 10.9 10.9 20.7 30.5
σ0 (MPa) 32 47 93 170 170 170 170 324 478
ε0 (µε) 262 383 753 1360 1360 1360 1360 2520 3630
V (m/s) 81 79 81 54.5 81 138 170 82 81
Cl (km/s) 9.04 9.13 9.22 9.35 9.35 9.36 9.27 9.52 9.79
ch (m/s) 324 333 373 388 424 519 567 527 610
U (m/s) 10 9.2 8.7 3.8 7.48 18 24.9 6.3 5.35
T (N) 10.4 11.0 13.6 14.4 17.5 27.0 32.8 26.3 35.4
σ (MPa) 165 171 210 221 273 414 513 409 555
ε (µε) 1350 1400 1710 1780 2160 3320 4030 3220 4180
γcal (deg.) 14 13.3 12.3 8.0 10.7 14.9 16.7 8.8 7.6
γexp (deg.) – – 13 8.5 11.1 15 17 9.5 8
propagation, both the γcal calculated from the measured longitudinal wave velocity and
the γexp measured from the high-speed sequences as shown in Fig. 9.20 are given in the
same table. As can be seen, both coincide well.
Moreover, as can be seen from Table 9.4, the stress and the strain of string under transverse
impact are markedly dependant on the prestretch force and the impact velocity.
Finally, it is worthwhile to notice that the transverse wave velocity of the polymer yarn
such as Kevlar is much faster than the transverse wave velocity of metals such as Ni-Cr
alloy. A faster transverse wave velocity is favorable to transmit the impact energy out
faster and to avoid high localization of strain. Thus, it is of significance in impact-resistant
engineering applications.
CHAPTER 10
Elastic–Plastic Waves Propagating in Beams under

Transverse Impact (Bending Wave Theory)
Bending motion of bar occurs when bar is subjected to transverse impact or eccentric
longitudinal impact. The bar bearing bend load is usually called beam. In the elementary
theory of beam which is based on the so-called “plane cross section assumption,” the
bending motion of beam is attributed to the relative rotation between cross sections and
the transverse motion of the neutral axis of beam.
In the previous chapter, we discussed the impact problem of flexible string, neglecting the
stiffness of string. The beam can be regarded as a string with un-negligible stiffness.
The dynamic bending problem of beams can be described either by vibration approach
or by wave propagation approach. The former is usually discussed in the discipline of
Vibration or Dynamics of Structures. In the present chapter, only the latter, namely the
bending wave propagation in beams, will be discussed. First, the elastic bending waves
will be briefly discussed, and then mainly the plastic bending waves will be discussed.
Similar to the situations of plastic longitudinal waves in bars, the plastic bending waves
can be dealt with by elastic–plastic analysis, or by rigid-plastic approximate analysis.
It should be noticed that the current chapter is limited to discuss the rate-independent
bending wave theory, and also limited to discuss the elemental theory based on the “plane
cross section assumption.” Obviously, the precise theory without “plane cross section
assumption” and taking account of strain rate effect will be much more complicated.
10.1 Basic Assumptions and Governing Equations
Basic equations of the present problem consist of three aspects: the motion equation,
the kinematical equation, and the constitutive equation of material. In general cases, the
variables used to describe the motion of beam are the functions of spatial coordinates
x, y, z, and time t. So, mathematically the problem is very complicated and difficult
399
to solve accurately. For this reason, some limitations and basic assumptions are usually
introduced.
First of all, it is limited to discuss the straight beam with constant cross-sectional areas,
and at least one geometrically symmetrical plane of cross section where the applied loads
lie in, exists. Therefore, the deformation of beam axis is always in the symmetrical plane,
and no torsion deformation happens. This is the so-called plane bending problem. The
meaning of such a limitation is to simplify a general three-dimensional problem to a
two-dimensional plane problem.
Then let us make the first basic assumption: the plane beam cross section perpendicular
to the beam axis before deformation remains a plane cross section perpendicular to the
beam axis after deformation. This is the so-called plane cross section assumption. Thus a
beam layer, the length of which is unvaried during the bending, must exist, called neutral
layer (surface). For the beam with rectangular cross section, the neutral layer is another
symmetrical plane which is perpendicular to the external load applied to the symmetrical
plane. The beam fibers above and below the neutral layer are respectively subjected to
tension and compression. The intersection line of the neutral layer with each cross section
is called neutral axis. We choose the beam axis before deformation as x axis, the neutral
axis as y axis, and the external force applied symmetrical to the axis as z axis, as shown
in Fig. 10.1. Obviously, the plane cross section assumption attributes the beam bending
to two basic motions, namely, the transverse displacement w of beam axis (where w is
positive in the positive z direction), and the rotation of each cross section around its neutral
axis (where the rotation angle α is positive in the clockwise direction).
Let us denote the transverse shear force and the bending moment by Q(x, t) and M(x, t)
respectively, the positive directions of which are defined as shown in Fig. 10.2. Consider
an infinitesimal beam element dx. According to the momentum theorem and the moment
of momentum theorem, we have the following two dynamic equations
∂Q ∂v
+ q(x) = ρ0 A0 (10.1)
∂x ∂t
∂M ∂ω
− Q = −ρ0 I (10.2)
∂x ∂t
a c r da
0 0'
x
b d
a' c'
0 0'
y
z .z
b' d'
Fig. 10.1. Plane bending of beam under plane cross section assumption.
Elastic–Plastic Waves Propagating in Beams under Transverse Impact 401
0
x
Q
M
M+dM
a
Q+dQ
dx
z,w
Fig. 10.2. The positive shear force Q and bending moment M acting on an infinitesimal element
of beam.
where q(x) is the distributed external load, ρ0 the initial density of the beam $ before
deformation, A0 the initial cross section of the beam before deformation, I = A z2 dA the
moment of inertia of the beam cross section with respect to the neutral axis, v the transverse
velocity of the beam axis, and ω the rotating angular velocity of the cross section. As can
be seen, the plane cross section assumption is used to simplify the plane bending problem
to a one-dimensional problem, namely, the six dynamic equations in the general situation
are now reduced to only two, Eqs. (10.1) and (10.2).
The kinematical equations under the plane cross section assumption can be easily
expressed as:
∂w
v= (10.3)
∂t
∂w
tgα = (10.4)
∂x
∂α
ω= (10.5)
∂t
∂ 2w
1 ∂α ∂x 2
k= = = −/ 2 03/2 (10.6)
ρ ∂s ∂w
1+
∂x
where k is the curvature, ρ the radius of curvature, and s the arc length of bending beam
axis. Because of the negative sign appearing on the right-hand side of Eq. (10.6), we have
defined the positive sign of M and k as shown in Fig. 10.2, wherein the deformed beam
axis is shown as a convex curve in the w–x coordinates (∂ 2 w/∂x 2 ) < 0.
In the bending wave theory to be discussed in the following, we again make the following
second basic assumption: the deformation of beam is sufficiently small (tan α 1).
Therefore the difference between Lagrange coordinates and Euler coordinates can be
ignored, and then Eqs. (10.4)–(10.6) are reduced to, respectively
∂w
α= (10.4a)
∂x
∂α ∂ 2w
ω= = (10.5a)
∂t ∂t∂x
1 ∂ 2w
k= =− 2 (10.6a)
ρ ∂x
Or, the continuous equations can also be expressed as
∂v ∂α
= =ω (10.7)
∂x ∂t
∂k ∂ω
=− (10.8)
∂t ∂x
On the cross section, the normal strain εx at an arbitrary point, which is at a distance
of z from the neutral axis, can be determined from the geometrical relation as shown in
Fig. 10.1, as
(ρ + z) dα − ρ dα z
εx = = = zk (10.9)
ρdα ρ
Under the assumption of small strain, substituting Eq. (10.6a) into the above equation,
we have
∂ 2w
εx = −z (10.9a)
∂x 2
So the meaning of the plane cross section assumption with respect to the determination of
beam deformation is that once the deformation of beam axis is obtained, the deformation
at any point of beam can be determined.
Finally, let us discuss the constitutive equation of material. In the cases of longitudinal
motion of bars, we are concerned with the simple tensile (compressive) σ –ε relation.
In the general cases of rate-independent problems, we need the stress–strain relation to
relate the stress tensor with the strain tensor. In the current cases of beam bending, M and
Q are used in the dynamic equations instead of stress components, while the variable in
relation to the strain ε is the curvature k [Eq. (10.9)]. By the plane cross section assumption,
the effect of shear force on the beam deformation is substantially ignored. Thus, with regard
to the beam, the constitutive relation of material is transformed to the M–k relationship.
According to the definition of bending moment, M is the resultant moment of the normal
stresses on the cross section with respect to the neutral axis, namely

M= σx z dA (10.10)
A
Under the condition of simple tension (compression), the σ –ε relation is easy to deter-
mine experimentally. However, in the general situation of three-dimensional stress states,
εx is not only dependent on σx , but also on σy and σz , namely, σx should be a func-
tion of εx , εy , and εz . Usually, at the same time by making the plane cross section
assumption, we make the following third basic assumption: there is no interference
between each longitudinal beam fiber, namely, all longitudinal beam fibers are under
unidirectional tension (compression), σx = σx (εx ). If the width and the height of the
rectangular cross section are denoted by b and h, respectively, then Eq. (10.10) can be
written as
h/2
M=b σx (εx )z dz
−h/2
In addition, by making the following fourth basic assumption: the σx –εx curve for
tension is the same as that for compression, the above equation can be written as
h/2
M = 2b σx (kz)z dz = M(k) (10.11a)
0
or
h/2
2b
M= σx (εx ) εx dεx = M(k) (10.11b)
k2 0
So if σx –εx curve is known, then M = M(k) curve can be calculated, and M must be a
function of k. Wherein, we have in fact implicitly made the following fifth basic
assumption: the constitutive equation of material is independent of strain rate. With
regard to this assumption, it should be understood that there approximately exists sole
M = M(k) relation under dynamic loading, although the constitutive equation under
dynamic loading is surely different from that under static loading.
In the range of linear elastic deformation, M = M(k) is also linear. So, Eq. (10.11) is
reduced to
M = EIk (10.12)
However, it should be noted that the linear hardening elastic–plastic σx –εx relation such as:

σx = Eεx εx ≤ εS
σx = (E − E1 )εS + E1 εx εx ≥ εS
does not correspond to a linear hardening elastic–plastic M = M(k) relation. In fact,

if the position of the boundary of the elastic layers and the plastic layers is expressed
by zS = εS /k, then from Eq. (10.11), we have

2
2bkEz3S h 2bkE 1 h3
M= + b(E − E1 )εS − zS2 + − zS3
3 4 3 8
z 3 z
S S
= EIk 1 − λ − 4λ + 3λ (10.13)
h h
ε 3 ε
S S
= EIk 1 − λ − 4λ + 3λ
hk hk
where λ = (E − E1 )/E and I = bh3 /12 (b and h the width and height of the rectangular
cross section, respectively). Obviously, the constitutive relation is of linear elasticity till
the point of kS = 2εS / h, MS = 2EIεS / h, and then is a nonlinear curve. When λ = 0, the
above equation coincides with Eq. (10.12). When λ = 1, it corresponds to the perfectly
plastic situation, and we have
/ 0
3εS 4ε 3
MS = EI − 3 S2
h h k
When k → ∞, we have zS = εS /k → 0 and M → 3EIεS / h = 1.5MS , as shown in

Fig. 10.3. If the full cross section enters into plastic deformation, then curvature k can
increase infinitely, and thus the so-called “plastic hinge” is formed.
Hence one can see that, from now on, when we talk about the linear hardening M–k
relation or the perfectly plastic M–k relation, they do not correspond to the linear hardening
σ –ε relation or perfectly plastic σ –ε relation, respectively.
Thus, we established the total basic equations, including the dynamic equations [Eqs. (10.1)
and (10.2)], kinematical equations (10.3), (10.4a), (10.5a), and (10.6a) [or (10.7) and
(10.8)], and constitutive equation [Eq. (10.11)]. There are altogether seven equations to
solve seven unknown variables M, Q, w, v, α, ω and k. The derivation of these basic equa-
tions is based on the above-mentioned five basic assumptions, and is limited to the plane
bending problems. Moreover, it has been implicitly assumed that there is no longitudinal
tensile (or compressive) loading applied on the beam.
1.5Ms
Ms
0
ks=2es / h k
Fig. 10.3. The M = M(k) relation for the elastic-perfectly plastic material.
Because of the introduction of the above-mentioned approximate assumptions, obviously

these equations are applicable only under some related conditions. For example, the rota-
tion angle of beam is limited to α < 10◦ by the small strain assumption. The shear stress
effect is disregarded due to the “plane cross section” assumption and the “no interfer-
ence between each longitudinal beam fiber” assumption. In the case of pure bending, it
is true that there is no shear stress. However, in the general cases, there actually exist
shear stresses on the cross section of a beam, the result of which is shear force Q. And
according to the “paired shear stresses theorem,” shear stresses should exist too between
the longitudinal beam fibers. Therefore, the beam elements between two original plane
cross sections (which are plane before deformation) must experience shear strain, then
the original plane cross sections must be distorted and are no longer plane. With regard to
the static equilibrium problem, it has been proved in the theory of elasticity that such an
approximate “plane cross section” assumption is acceptable only for long beams. This can
be interpreted too by an estimation of and comparison between the value-order of each
stress component. If the beam length is of order l, the cross-sectional area is of order h2 ,
and the distributed loading (per unit length) is of order q, then the shear force Q is of
order ql, the normal stress between the beam fibers σY is of order q/ l/ h, the shear stress
τ is of order Q/ h2 or order ql/h, and the normal stress on the cross section σX is of order
Ql/h3 or order ql2 / h3. So τ/σX is of order h/ l, and σY /σX is of order h2 / l 2 . Thus, if l/ h is
large enough, then by comparing with σX , τ is negligible, and σY is more negligible. It is
usually recognized that the shear stress effect can be ignored for the beams with the ratio
h:l ≤ 1:5. However, in the dynamic problems of beams, the key factor is not the ratio of
the cross-sectional size of the beam with the length, but the ratio of the cross-sectional size
of the beam with the wavelength. As pointed out by Timoshenko, the shear stress effect can
be disregarded so long as the wavelength is large enough than the radius of gyration (see
Kolsky, 1953). In such cases, the results obtained under the above-mentioned approximate
assumptions satisfactorily coincide with the results obtained from the precise theoretical
analysis, otherwise corrections are required.
10.2 Elastic Bending Waves
In the simplest theory of elastic bending waves, it is assumed that the motion of each beam
element is purely the transverse motion perpendicular to the beam axis; or in other words,
the rotatory inertia of beam cross section is ignored. In addition, let q = 0, the basic
equations are then reduced to the following linear equation set:
⎧
⎪
⎪ ∂Q ∂v
⎪
⎪ = ρ0 A0 (10.1a)
⎪
⎪ ∂x ∂t
⎪
⎪
⎪
⎪ ∂M
⎪
⎪ =Q (10.2a)
⎪
⎪
⎨ ∂x
∂w
⎪ v= (10.3)
⎪
⎪ ∂t
⎪
⎪
⎪
⎪ ∂ 2w
⎪
⎪ k=− 2
⎪
⎪
(10.6a)
⎪
⎪ ∂x
⎪
⎩M = EIk (10.12)
Or, the above equation set can be reduced to solve the following fourth-order partial
differential equation:
∂ 4w ∂ 2w
C02 R 2 + 2 =0 (10.14)
∂x 4 ∂t
√ √
where C0 = E/ρ0 is the wave velocity of longitudinal elastic wave and R = I /A0 is
the rotating radius of cross section with respect to the neutral axis.
Assume that the bending wave solution of Eq. (10.14) can be expressed in the form of
w = D cos( pt − fx) (10.15)
where D is the amplitude, f = 2π/Λ the wave number, p = fC the circular frequency,
while Λ the wavelength, and C the phase velocity of the wave. Substituting Eq. (10.15)
into Eq. (10.14), we have
p2 = C02 R 2 f 4
or
2πRC 0
C= (10.16)
Λ
It is thus clear that, in the elementary theory of bending wave propagation, elastic bend-
ing wave velocity C is inversely proportional to wavelength Λ. Therefore, as different
from the elastic longitudinal waves, where the wave velocity is a constant, a bending
disturbance with an arbitrary profile will propagate with a profile-change (the so-called
dispersion phenomenon); namely, the disturbances with different wavelengths propagate
with different velocities.
However, according to Eq. (10.16), if the wavelength changes infinitesimally, then the
wave velocity tends to approach infinity. Physically, an infinite wave velocity is unaccept-
able. This is because Eq. (10.16) is applicable only when wavelength Λ is much larger than
rotating radius R of beam cross section. When wavelength Λ is comparable with rotating
radius R, the rotatory inertia ρ0 I (∂ω/∂t) should not be disregarded again, and the plane
cross section assumption should be corrected to take account of shear stress effect.
The correction of taking rotatory inertia into account was at first proposed by Rayleigh
(see Kolsky, 1953). In such cases, the basic equation set returns to:
⎧
⎪
⎪
∂Q
= ρ0 A0
∂v
⎪
⎪
⎪
⎪ ∂x ∂t
⎪
⎪
⎪
⎪ ∂M ∂ω
⎨ − Q = −ρ0 I
∂x ∂t
⎪
⎪ 2
⎪
⎪ ∂w ∂ w ∂ 2w
⎪
⎪ v= , ω= , κ=− 2
⎪
⎪ ∂t ∂t∂x ∂x
⎪
⎪
⎩
M = EIk
or to solve the transverse displacement w from the following partial differential equation:
∂ 4w ∂ 4w ∂ 2w
C02 R 2 4
− R2 2 2 + 2 = 0 (10.17)
∂x ∂x ∂t ∂t
As compared to Eq. (10.14), Eq. (10.17) contains an additional term to take account of
rotatory inertia. Let Eq. (10.14) be its solution, we obtain
−1/2
Λ2
C = C0 1 + (10.18)
4π2 R 2
The above equation is physically reasonable, since wave velocity C approaches a finite
value C0 when wavelength Λ approaches zero. When ratio R/Λ is small enough, namely
either R/Λ 1 or Λ/R 1, Eq. (10.18) coincides with Eq. (10.16). It means that the
rotatory inertia can be ignored only when Λ is much larger than R.
As pointed out by Timoshenko, the correction of shear stress effect is as important as

the correction of rotatory inertia (see Kolsky, 1953). When the shear stress is taken into
account, the original plane cross section is distorted due to shear strain, and thus the
transverse displacement w of beam axis is actually composed of two parts: wM which
corresponds to rotating angle α induced by applied bending moment M, and wQ which
corresponds to shear strain γ induced by shear force Q:
w = wM + w Q (10.19)
The following elastic constitutive relation exists between the Q and γ :

Q= τ dA = Gγ (z) dA = Gγ AS (10.20)
A A
where AS is a coefficient which depends only on the shape of beam cross section, γ (z) the
real shear strain which varies with the cross section, and G the elastic shear modulus.
Taking account of such a correction, the basic equations should be correspondingly
corrected to:
∂Q ∂v ⎫
= ρ0 A0 (translation)⎪
⎬
∂x ∂t (dynamic equations)
∂M ∂ω ⎪
⎭
− Q = −ρ0 I (rotation)
∂x ∂t
⎫
w = wM + wQ ⎪
⎪
⎪
∂ 2 wM ∂ 2 wM ⎪
⎬
ω= , k= − (bending) [kinematical Eqs. (I)]
∂t∂x ∂x 2 ⎪
⎪
∂wQ ∂w ⎪
⎪
γ= , v= (shear) ⎭
∂x ∂t
or
∂k ∂ω ⎫
=− (bending)⎪
⎬
∂t ∂x [kinematical Eqs. (II)]
∂γ ∂v ⎪
⎭
+ω = (shear)
∂t ∂x

M = EIk (bending)
(constitutive equations)
Q = GAs γ (shear)
Substituting the kinematical equations and the constitutive equations into the dynamic
equations, respectively, we obtain:

∂ 2 w ∂ 2 wM ∂ 2w
GAS − = ρ0 A 0
∂x 2 ∂x 2 ∂t 2

∂ 3 wM ∂w ∂wM ∂ 3 wM
EI + GAS − = ρ0 I 2
∂x 3 ∂x ∂x ∂t ∂x
Eliminating the terms containing wM from the above two equations, finally we obtain the
following fourth-order partial differential equation with respect to w:
4
EI ∂ 4 w I E ρ0 A0 ∂ w I ρ 0 A0 ∂ 4 w ∂ 2 w
4
− 1 + 2 2
+ · + 2 =0
ρ0 A0 ∂x A0 ρ0 GAS ∂x ∂t A0 GAS ∂t 4 ∂t
Noticing C02 = E/ρ0 and R 2 = I /A0 , and introducing CQ

2 = GA /ρ A , the above
S 0 0
equation can be rewritten as
% &
∂ 4w C02 ∂ 4w R2 ∂ 4w ∂ 2w
C02 R 2 4 −R 2
1+ 2 2 2
+ 2 · 4 + 2 =0 (10.21a)
∂x CQ ∂x ∂t CQ ∂t ∂t
or
% &
∂ 4w 1 1 ∂ 4w 1 1 ∂ 4w 1 ∂ 2w
4
− 2
+ 2 2 2
+ 2 2 4 + 2 =0 (10.21b)
∂x C0 CQ ∂x ∂t CQ C0 ∂t C0 R 2 ∂t 2
which is usually called Timoshenko equation.1
1 If
we substitute the constitutive equation into the kinematical equation, together with the dynamic equation,
then the following first-order partial differential equation is obtained:
⎧ ⎫
⎪ ∂M ∂ω 1 ∂M ∂ω ⎪
⎪
⎨ ∂x + ρ0 I ∂t = Q; − =0 ⎪
⎬
EI ∂t ∂x
⎪ ∂Q
⎪ ∂v 1 ∂Q ∂v ⎪
⎩ − ρ0 A 0 = 0; − = −ω⎪
⎭
∂x ∂t GAS ∂t ∂x
Let Eq. (10.15) be a solution, we obtain

% &
1 1 p4 p2
f −
4
+ 2 f 2 p2 + − 2 =0
C02 CQ 2 2
C0 CQ C0 R 2
or
% &
C2 C2 C4 C2
1− 2
+ 2 + 2 2
− 2 =0 (10.22)
C0 CQ C0 CQ C0 R 2 f 2
The solution for the above equation provides the relation between bending wave velocity C
and wavelength Λ = 2π/f .
For a cylindrical bar with a radius of a, when the material Poisson ratio ν = 0.29, the
dimensionless relations between C/C0 and a/Λ obtained by different theories were given
by Davies, as shown in Fig. 10.4 (see Kolsky, 1953).
The results calculated from Timoshenko’s theory coincide very well with the results calcu-
lated from the precise theory based on the general elastic equations. The different theories
show that bending waves display dispersion phenomenon. When a/Λ < 0.1 (namely, the
wavelength is 10 times larger than the radius of the cylindrical bar), the results given by
all these theories are very close to each other; otherwise both the elementary theory and
the Rayleigh’s correction will result in remarkable errors.
10.3 Plastic Bending Waves (Elastic–Plastic Beams)
Studies on the plastic bending waves were carried out later than the plastic longitudinal
waves. The earliest theoretical and experimental investigation of plastic deformation of
beam under transverse impact may be traced back to the work by Duwez et al. (1943, 1950),
based on which other research works were developed. In the following, we will mainly
discuss their results, the theoretical part of which was established by Bohnenblust.
Consider an infinite beam which is subjected to transverse constant-velocity impact V1 .

In the case of elastic deformation, the problem has been solved by Boussinesq in 1885
(see Duwez et al., 1950), and it was shown that w/t is solely dependent on x 2 /t, namely
the problem possesses similar solution. This result was generalized by Bohnenblust from
the elastic bending problem to the plastic bending problem, namely generalized from the
linear relation M = EIk to the nonlinear relation M = M(k). They verified that in the
If they are solved by the characteristics method, the corresponding characteristics equations are:
dM ± ρ0 IC 0 dω = ±C0 Q dt, along dx ± C0 dt = 0;
dQ ± ρ0 IC Q dv = ρ0 A0 CQ
2
ω dt, along dx ± CQ dt = 0
However, the above two sets of equation are not independent but coupled each other. In fact, M–ω waves depend
on Q, while Q–v waves depend on ω.
elementary theory
1.0
Rayleigh's theory
0.9
0.8
0.7
precise theory
0.6
C0
C
0.5
0.4
Timoshenko's theory
0.3
0.2
0.1
0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
aV
Fig. 10.4. Phase velocities of elastic bending waves for a cylindrical bar with a radius of a
(ν = 0.29).
situation of plastic bending, w/t is still solely dependent on x 2 /t, thus the problem can
be easily solved.
Boussinesq’s theory did not provide a precise solution starting from the general equations,
but an approximate solution starting from the elementary theory, wherein the shear stress
effect and the rotatory inertia are ignored. In such a case, the self-similarity character of
solution can be perceived by the dimensional analysis as follows. The basic equation of
the current problem is Eq. (10.14). Or by introducing a defined as
EI
a4 = = C02 R 2 (10.23)
ρ0 A0
Equation (10.14) can be rewritten as
∂ 4w ∂ 2w
a4 + 2 =0 (10.14a)
∂x 4 ∂t
Since in the above equation a always appears simultaneously with ∂/∂x, by introducing
ξ defined as
x
ξ=
a
Equation (10.14a) can be further reduced to
∂ 4w ∂ 2w
+ 2 =0
∂ξ 4 ∂t
Note that only constant-velocity V1 is included in the conditions used to determine the
solution for the current problem, while neither characteristic length nor characteristic time
are included, then we have
w = f (ξ, t, V1 )
Noticing that the dimension of ξ = x/a is t 1/2 , choose t and V1 as the basic dimensions,
the problem obviously possesses the following similar solution (see Appendix 3)
2 2
w ξ x
=f =f
V1 t t a2t
Or for convenience in practical application, if it is rewritten as
x2
w = tf (η), η= (10.24)
4a 2 t
then we have
∂w
= f (η) − ηf (η) = v (10.25)
∂t
∂ 2w η2
= f (η) (10.26)
∂t 2 t
√
∂w x t√
= 2 f (η) = ηf (η) = α (10.27)
∂x 2a a
∂ 2w 1
2
= 2 [f (η) + 2ηf (η)] = −k (10.28)
∂x 2a
∂ 2w 1
= − √ η3/2 f (η) = ω (10.29)
∂t∂x a t
∂ 3w 1 x
= 4 [3f (η) + 2ηf (η)] (10.30)
∂x 3 4a t
∂ 4w 1 !
4
= 4 3f (η) + 12ηf (η) + 4η2 f (4) (η) (10.31)
∂x 4a t
Substituting Eqs. (10.26) and (10.31) into Eq. (10.14a), the partial differential equation is
then reduced to the following ordinary differential equation
4η2 f (4) (η) + 12ηf (η) + (3 + 4η2 )f (η) = 0 (10.32)
It is worthwhile to note that v(=∂w/∂t) and κ(=−∂ 2 w/∂x 2 ) are only dependent on η,
while α(=∂w/∂x), ω(=∂ 2 w/∂t∂x) and ∂v/∂t (=∂ 2 w/∂t 2 ) are also dependent on x or t.
Bohnenblust adopted all the basic assumptions used in the elementary theory of elastic
beam. He first presumed that the Boussinesq’s self-similar solution [Eq. (10.24)] is still
correct for the elastic–plastic beam. After integrating the basic equation, he found that
the solution satisfies the initial and boundary conditions, and consequently verified that
Eq. (10.24) holds. Actually, we can still use the dimensional analysis to explain it as
follows.
Under the basic limitations and assumptions discussed in Section 10.1, and when the
rotatory inertia is ignored, the basic equations are
∂Q ∂v ⎫
= ρ0 A0 ⎪ ⎬ (10.1a)
∂x ∂t (dynamic conditions)
∂M ⎪
⎭
=Q (10.2a)
∂x
⎫
v=
∂w ⎪
⎪
⎬ (10.3)
∂t
(kinematical conditions)
∂ 2w ⎪
κ=− 2⎪ ⎭ (10.6a)
∂x
M = M(κ) (constitutive equation) (10.11)
Substituting Eqs. (10.3), (10.6a), and (10.11) into Eqs. (10.1a) and (10.2a), we obtain the
following equation set, containing only variables Q and w:
∂Q ∂ 2w
= ρ0 A0 2 (10.33)
∂x ∂t
dM ∂ 3 w
Q=− (10.34)
dk ∂x 3
which can be further reduced to the following quasi-linear fourth-order partial differential
equation with respect to w:
2
1 dM ∂ 4 w 1 d 2M ∂ 3 w ∂ 2w
− + =0 (10.35)
ρ0 A0 dk ∂x 4 ρ0 A0 dk 2 ∂x 3 ∂t 2
It is known that M is a function of k, namely M = M(k) or expressing in the dimensionless

form:

M k
=f
MS kS
where MS and kS are yield bending moment and yield curvature, respectively. Therefore
we have
dM d(M/MS ) MS d 2M d 2 (M/MS ) MS EI
= = EIf1 , = = f
dk d(k/kS ) kS dk 2
d(k/kS )2 kS2 kS 1
Recalling the definition of notation a, namely a 4 = EI/ρ0 A0 , Eq. (10.35) can be

rewritten as
2
∂ 4w 4 f1 ∂ 3w ∂ 2w
a 4 f1 4 −a + =0
∂x kS ∂x 3 ∂t 2
Introducing ξ = x/a and Q = w/kS , then we have

2
∂ 4 Q f1 ∂ 3Q ∂ 2Q
f1 − 2 + =0 (10.35a)
∂ξ 4 a ∂ξ 3 ∂t 2
Noticing that we also have

2
M k 1 ∂ 2w ∂ Q
= f1 = f1 − 2
= f1 − 2
MS kS kS ∂x ∂x
and once more introducing ζ = −Q/a 2 , then Eq. (10.35a) can be further reduced to
2
∂ 4ζ ∂ 3ζ ∂ 2ζ
f1 4 − f1 + =0 (10.35b)
∂ξ ∂ξ 3 ∂t 2
and
2 2
M ∂ Q ∂ ζ
= f1 − 2 = f1 − 2
MS ∂x ∂ξ
as well. The following original boundary condition

∂w
= V1
∂t x=0
is now reduced to

∂ζ V1
=
∂t ξ =0 kS a 2
Under such a boundary condition, the solution of Eq. (10.35b) can be written as

V1
ζ = g1 ξ, t,
kS a 2
The dimensions of each variable in the above equation are, respectively:
[t] = T

x L
[ξ ] = √ = = T 1/2
4
EI/ρ0 A0 LT −1/2

w L
[ζ ] = √ = −1 2 −1 = T
kS EI/ρ0 A0 L ·L T

V1 V1 LT −1
= √ = −1 2 −1 = L0 T 0
kS a 2 kS EI/ρ0 A0 L ·L T
Therefore, obviously we have

2
ζ ξ V1
=g ,
t t kS a 2
or

w x2 V1
√ =g √ , √
kS · EI/ρ0 A0 · t EI/ρ0 A0 · t kS · EI /ρ0 A0
It shows that Eq. (10.24) still holds for elastic–plastic beam.
Thus, according to Eq. (10.28) k is only dependent on η, and consequently according to

Eq. (10.11) M is also only dependent on η, but according to Eq. (10.2a) Q should be
∂M dM ∂η 1 √ dM
Q= = = √ η (10.36)
∂x dη ∂x a· t dη
In order to deal with the functions which are dependent only on η, S is introduced which
is defined as
2a 2 √ 2a 2 √ dM
S= tQ = η (10.37)
EI EI dη
Since we have
∂Q EI ds ∂η EI √
= 3√ = 4 ηS
∂x 2a t dη ∂x 2a t
and moreover taking into account Eqs. (10.26) and (10.30), Eqs. (10.33) and (10.34) can
be correspondingly reduced to
S = 2η3/2 f (η) (10.38)

1√ ! dM
S=− η · 3f (η) + 2ηf (η) · (10.39)
EI dk
If we eliminate S from these equations, then the following fourth-order ordinary differential
equation can be obtained:

dM (4) dM η2 d 2 M

4η2 f + 12η −4 2 − ηf + 3f f
dk dk a dk 2
(10.40)
dM η d 2 M
+ 3 + 4EIη − 9 2
2
f f =0
dk a dk 2
In the case of elastic beam, we have M = EIk, then Eq. (10.40) coincides with Eq. (10.32).
Obviously, substituting Eqs. (10.26), (10.30), and (10.31) directly into Eq. (10.35), we can
obtain Eq. (10.40) too.
Alternatively, eliminating f (η) from Eqs. (10.38) and (10.39), an ordinary differential
equation with respect to S can be obtained. To this end, differentiating Eq. (10.38),
we have
√ !
S = η 3f (η) + 2ηf (η)
Comparing it with Eq. (10.39), we obtain

dk
S + EIS =0 (10.41)
dM
Obviously, Eqs. (10.40) and (10.41) are equivalent to each other. From the former, the
transverse displacement w = tf (η) can be solved and then k, M, and Q can be obtained
by differentiating operations. From the latter, S can be solved and then M, k, and w
can be obtained by integrating operations. Mathematically, solving Eq. (10.41) is sim-
pler than Eq. (10.40), since Eq. (10.41) is a second-order ordinary differential equation.
It was Bohnenblust who just started from this equation to study the plastic bending wave
propagating in beams.
Now we discuss the initial and boundary conditions of the current problem, they are
t = 0, 0 < x ≤ ∞
w=v=α=ω=k=M=0 (10.42)
∂w ∂w ∂ 2w ∂ 2w
or w = = = = =M=0
∂t ∂x ∂t∂x ∂x 2
t > 0, x=∞
w=v=α=ω=k=M=0 (10.43)
∂w ∂w ∂ 2w ∂ 2w
or w = = = = =M=0
∂t ∂x ∂t∂x ∂x 2
t > 0, x=0
∂w
v= = V1 (10.44)
∂t
By introducing the dimensionless independent variable defined as η = x 2 /4a 2 t, the initial

condition (10.42) and the boundary condition (10.43) both correspond to the boundary
condition at η = ∞. From Eqs. (10.24)–(10.29) it can be known that there should be the
following boundary conditions:
η = ∞, w=v=α=ω=k=M=0
(10.45)
or f (η) = ηf (η) = η3/2 f (η) = 0
∂w
η = 0, v= = f (η) − ηf (η) = f (η) = V1 (10.46)
∂t
Note that the following condition has been used in Eq. (10.46).
lim ηf (η) = 0
η→0
This can be explained as follows: at the impact point x = 0, in a finite duration the angle of
rotation of cross section α = ∂w/∂x should also be finite. Then according to Eq. (10.27),
√
we must have limη→0 ηf (η) = A, where A is a finite quantity, thus the above equation
holds. Alternatively, from the conditions of w(0, t) = V1 t = t · f (0), Eq. (10.46) can be
directly obtained and thus it is required to satisfy limη→0 ηf (η) = 0.
In addition, if at x = 0, the rotating angle α = ∂w/∂x is a continuous variable, then

because the deformations of beam on the right and left sides of x = 0 are anti-symmetric,
we must have

∂w
α(0, t) = =0
∂x x=0
Or according to Eq. (10.27), the continuous condition of ∂w/∂x can be expressed as

√
lim ηf (η) = A = 0 (10.47)
η→0
If a strong discontinuity of ∂w/∂x appears at the impact point (e.g. the so-called plastic
hinge appears and then a break in shape of beam axis appears), then in the above equation
A
= 0; but because α(0, τ ) is always finite, A is at least still finite. If at any time t,
∂w/∂x|x=0 remains continuous, then we have

∂α ∂ 2 w
= = − lim η3/2 f (η) = 0 (10.47a)
∂t x=0 ∂t∂x x=0 η→0
When we start from Eq. (10.41) to solve the problem, these boundary conditions should
be rewritten to the corresponding expressions in the form as functions of S(η). Noticing
Eq. (10.38), corresponding to Eqs. (10.45) and (10.46), we have
S (∞) = 0 (10.48)
0
1 0 S (η)
V1 = dη 3/2
dη (10.49)
2 ∞ ∞ η
According to Eqs. (10.27)–(10.29), Eq. (10.38) can also be written in the following different
forms:
√ ∂ 2w √
S (η) = 2η3/2 f (η) = −2a t = −2a tω
∂t∂x
(10.50)
2√ ∂ 2w √ √ a
= 2a η 2 − ηf (η) = −2a 2 ηk − √ α
∂t t
At x = 0, the continuous condition of α, namely Eq. (10.47) or (10.47a), can be

expressed as
S (0) = 0 (10.51)
It is thus clear that, according to Eq. (10.50), S (η) in its physical meaning can be under-
stood as a function of angular velocity ω and similar to Eq. (10.37) depends only on η.
The condition (10.51) can be understood as that of the angular velocity at x = 0 equals
zero, or at x = 0 the rotation angle is continuous (equal to zero) while curvature k is
finite. If a plastic hinge is formed at x = 0, then a discontinuity of α will take place,
and consequently S (0)
= 0, and a discontinuity of S (0) will take place. This will be
mentioned again in the following discussion.
Once S(η) is solved from Eq. (10.41), Q, M, α, k, and w can be obtained from the following
equations.
According to Eq. (10.37), we have
EI
Q= √ S(η) (10.52)
2a 3 t
and then the √external force P (t) at x = 0 can be calculated, P (t) = 2Q(0, t) =
(EIS(0))/(a 3 t)
η ∞
EI S(η) EI S(η)
M= √ dη = − 2 √ dη (10.53)
2a 2 ∞ η 2a η η
wherein the boundary condition (10.45) has been utilized.

1 η S S (η) η S
f (η) = 3/2
dη = − √ + √ dη;
2 ∞ η η ∞ η
wherein the boundary conditions (10.45) and (10.48) have been utilized. Then according
to Eq. (10.27) we have
√ √ ∞
∂w t t√ S
α= = − S (η) − η √ dη; (10.54)
∂x a a η η
and according to Eq. (10.50a) we have

∞
S a 1 S
k=− 2 √ − √√ = 2 √ dη (10.55)
2a η 2a t η 2a η η
Finally according to Eqs. (10.24) and (10.38) we have

t η η S
w= dη dη
2 ∞ ∞ η3/2
wherein the boundary conditions (10.45) have been utilized. With regard to the last double
integral, by changing the integral sequence as shown in Fig. 10.5, it can be rewritten as
∞ ∞ ∞
t η ζ S (ξ ) t S (ξ ) ξ t S (ξ ) t S (ξ )
w= dζ 3/2
dξ = dξ · dζ = √ dξ − η dξ
2 ∞ ∞ ξ 2 η ξ 3/2 η 2 η ξ 2 η ξ 3/4
and finally it can be written as

∞
t η S t S
w= √ dη − η dη (10.56)
2 ∞ η 2 η η3/2
$∞
Noticing that (1/2)η η (S /η3/2 ) dη is similar to −ηf (η), and according to Eq. (10.46)
we have limη→0 ηf (η) = 0, then the boundary condition (10.46) or (10.49) can be
expressed as
∞
1 S
V1 = √ dη (10.57)
2 0 η
z B
z
x=
C
A(x=z=h)
0
x
Fig. 10.5. Change of the integral sequence of double integrate.

In fact, Eq. (10.56) has provided the following relation

∞ ∞ ∞
1 S 1 S 1 S
f (η) = √ dη − η 3/2
dη = √ dη + ηf (η)
2 η η 2 η η 2 η η
Comparing it with Eq. (10.25), we obtain

∞
∂w 1 S
v= = √ dη (10.58)
∂t 2 η η
It means that the boundary condition which should be satisfied is Eq. (10.57).
Thus, the whole problem is solved. The previous discussions can be summed up as
follows: when an infinite beam is subjected to a transverse constant-velocity impact V1 , the
so-called self-similar solutions (10.24) exist regardless of whether the beam deformation
is elastic or elastic–plastic. Then the problem is reduced to solve the ordinary differen-
tial equation (10.40) or (10.41). If we start from Eq. (10.41) to solve the function S(η)
under the boundary conditions (10.48) and (10.57) [namely (10.49) and (10.51)], then
Q, M, α, k, w, and v can be obtained respectively from Eqs. (10.52)–(10.56) and (10.58).
So long as the impact velocity is the same as V1 , the beam deflection curves at differ-
ent times are completely the same w(x, t) if these are expressed in the form of w/t and
x/t 1/2 ; particularly, the point x0 where w = 0 is proportional to t 1/2 . For a certain value
of η1 = x 2 /(4a 2 t), w/t, v, α/t 1/2 , k, M, and Qt1/2 have the same values respectively.
This means that √ these quantities propagate along the beam axis with the wave velocity
√
dx/dt = a η1 / t. In other words, the wave velocity is not a constant but decreases
√
with increasing t. Such a phenomenon coincides with the dispersive character of elastic
bending waves as pointed out earlier.
It should be noted, both the rotatory inertia and the shear stress effect are disregarded in the
above analyses. As pointed out in the analyses on elastic bending waves, such elementary
theory is reliable only when the wavelength is much larger than the size of beam cross
section. We can imagine that a similar situation exists for the plastic bending waves.
Obviously, a more precise solution can be obtained only when the rotatory inertia and
the shear stress effect are taken into consideration. In addition, in the above analyses, the
reflection and interaction of waves, as well as the more complicated situations in relation
to the loading–unloading process, are all not yet discussed.
The key to the whole problem lies in solving Eq. (10.41). The nonlinearity of dM/dk
makes the problem mathematically complicated. In general, if M = M(k) relation is
experimentally determined, the solution is usually obtained by numerical calculation.
However, if M = M(k) relation can be approximately described by linear hardening
plastic or perfectly plastic model, as shown in Fig. 10.6, then it is possible to solve the
problem by analytical calculation. In the following, we will discuss two such special cases
respectively.
We will first discuss the corresponding elastic beam, and determine the critical condition
for the beginning of plastic yield.
M
1
0< x <1
x 2EI
Ms
2
EI
0
ks k
Fig. 10.6. The perfectly plastic and linear hardening plastic M = M(k) relations.
(1) In elastic cases (M = EIk)
In the case of elastic deformation, Eq. (10.41) is reduced to the following second-order
constant-coefficient linear homogeneous ordinary differential equation:
S + S = 0
Its general solution is S = C1 cos η + C2 sin η, and the indeterminate coefficients are
determined by the boundary conditions (10.51) and (10.57):
S (0) = −C1 sin 0 + C2 cos 0 = 0, C2 = 0

9 9
C1 ∞ sin η C1 π 2
V1 = − √ dη = − , C1 = −2 V1
2 0 η 2 2 π
Therefore the solution is

9
2
S(η) = −2 V1 cos η (10.59)
π
According to Eqs. (10.53) and (10.55) we have

9 9
2 EIV 1 ∞ cos η 2 V1 ∞ cos η
M(η) = √ dη, k(η) = √ dη
π a2 η η π a2 η η
√
As can be seen, S(η) is a cosine function of η and its amplitude is 2V1 2/π, while the
absolute amplitudes of the period functions M(η) and k(η) decrease with increasing η,
S(h)
M
0
h h=0
(a)
M(h), k(h)
h2
h4
h= ∞ k
0
h (c) h3
(b) h1
h1=π/2 h3=5π/2
h2=3π/2 h4=7π/2
Fig. 10.7. In the case of elastic deformation (M = EIk).
The maximum values appear at λ = 0:
9
2 EIV 1 V1
S(0) = −2 V1 , M(0) = , k(0) = (10.60)
π a2 a2
At η = (2n + 1)π/2 (n = 0, 1, 2, . . .), S(η) reaches its null point, while according to
Eq. (10.36), these are extreme points for M(η) and k(η), which alternatively reach the
positive extreme and the negative extreme. It means that with the propagation of bending
wave starting from the impact point x = 0 along the beam axis, at each cross section
the beam will be successively convex upwards and concave downwards, and the bending
deformation will gradually increase after each convex–concave alternative. However, the
propagation velocity of bending wave is not a constant. For example, the propagation of
the point w = 0 is proportional to t 1/2 . Let us estimate how x0 which is the first point of
w = 0, varies with time t. Note that now η0 = x02 /4a 2 t is a small quantity. According to
Eq. (10.56), now η0 should satisfy the following equation:
∞ ∞
sin η sin η
√ dη − η0 dη = 0
η0 η η0 η3/2
By using the following series expansion relations:

sin η 1 1 3 1 1 1
√ = √ η− η + η5 − · · · = η1/2 − η5/2 + η9/2 − · · ·
η η 3! 5! 3! 5!

sin η 1 3/2
; = η−1/2 − η + 1/5!η7/2 − · · ·
η3 3!
the above equation is reduced to

0 sin η ∞ 0 ∞
sin η sin η sin η
√ dη + √ dη − η0 dη − η0 dη
η0 η 0 η η0 η3/2 0 η3/2
η0 9
2 3/2 2 1 7/2 2 1 11/2 π
=− η − · η + · η − · · · +
3 7 3! 11 5! 0 2
η0 9
2 1 2 1 π
+ η0 2η1/2 − · η5/2 + · η9/2 − · · · −2 η0 = 0
5 3! 9 5! 0 2
7/2
Ignoring η0 and higher order small quantities, we have
9 9
π 4 3/2 π
2 η0 − η 0 − =0
2 3 2
Thus finally we obtain

< √ EI √
x0 = 4a 2 tη0 ≈ 2.13a t = 2.13 · 4
t (10.61)
ρ0 A0
√
It indicates that the relation between x0 and t is dependent only on beam size and material
properties, but independent of impact velocity V1 .
Since the maximum of M and k occurs at point x = 0 and is proportional to V1 , with

increase of V1 the plastic deformation must occur first at point x = 0. From Eq. (10.60),
let M(0) = MS , the critical impact velocity VS which is corresponding to the beginning
of yield for the elastic beam, is obtained as
a 2 MS 2a 2 εS
VS = = a 2 kS = (10.62)
EI h
where h is the height of the beam cross section and εS the tensile (compressive) yield
stress.
(2) In linear hardening plastic cases
When V1 > VS , the plastic deformation takes place in the beam. If we assume that the
M–k relation is described by the so-called linear hardening model, namely, comprising
two straight lines with the slope of EI and ξ 2 EI (ξ < 1) respectively, as shown in Fig. 10.6,
then in the elastic deformation region and the plastic deformation region, Eq. (10.41) is
respectively reduced to
S + S = 0
1
S + S=0
ξ2
The general solutions are respectively:
S(η) = C1 cos η + C2 sin η = A1 sin(η − A2 ) (10.63)

η η
S(η) = C3 cos + C4 sin = A3 sin(η − A4 ) (10.64)
ξ ξ
Thus, S(η) comprises these solutions in different regions. At the connecting point of these
two kinds of solutions, S as well as S at both regions should be the same, respectively.
As can be seen from Eq. (10.50a), such requirement means that rotating angle α and
curvature k should be continuous at the connecting point of elastic region and plastic
region.
For example, for 0 ≤ η ≤ η0 , since M(η) ≥ MS , the plastic deformation takes place,
corresponding to the solution of (10.64); while for η ≥ η0 , the deformation is still elastic,
corresponding to the solution of (10.63), as shown in Fig. 10.8. Now we have totally
five indeterminate coefficients: C1 , C2 , C3 , C4 , and η0 , which can be determined by the
following five conditions, namely, the boundary conditions (10.51) and (10.57) at η = 0,
the continuous conditions at η = η0 for both S(η) and S (η) calculated respectively from
Eqs. (10.63) and (10.64), and finally M(η0 ) = MS .
Figure 10.8 displays the situation where at each beam cross section the plastic defor-
mation only takes place once and no unloading happens. So the problem is relatively
simple to solve. In such a case, at an arbitrary beam cross section x, at first a series
of positive–negative alternative bending elastic waves pass through, and the amplitude
successively increases. At time t = x 2 /(4a 2 η0 ), the cross section gets into plastic bending
under the action of +M (if V1 is positive). After that the plastic deformation monotonously
goes up. When t approaches infinity, M approaches M(0).
If after the first plastic deformation, an unloading and soon afterwards a second plastic
loading take place, the problem is more complicated. Let us first have a look at how the
cross section comes into the unloading and re-loading.
S(h) M h=0
h0 h2 h4 M3
o h=h0
(a) h1 h3 h
h=h2
h=∞
M(h) o ks k
(c) h=h3
Ms h=h1
(b) o Ms
h
−Ms
Fig. 10.8. The solutions when the plastic deformation only takes place once at beam cross section
and no unloading happens.
S(h) M
h= 0
h0 h3 h=h0
o h1 h2 h=h3
(a) h
h4
h=∞
I II III IV o k
M(h)
(c)
Ms
h= h2
o h= h1
2Ms
h
(b) −Ms
Fig. 10.9. The solutions when the unloading and the secondary plastic loading take place at beam
cross section.
With increasing V1 , η0 in Fig. 10.8 moves rightwards. However, η0 is unable to reach

point η1 , because M reaches its negative extreme at point η1 . So between η0 and η1 , there
must exist a null-point of M, or there must exist an interval |M(η)| < MS . However, it
should be noted that the absolute value of M(η1 ) also increases with increasing V1 . There-
fore, if V1 is large enough, an interval M(η1 ) ≤ −MS will finally appear. Consequently
a new plastic region (η1 ≤ η ≤ η2 ) will be formed in both S–η and M–k plots, and an
accompanying unloading region will appear between η0 and η1 , as shown by regions II and
III in Fig. 10.9, respectively. In such a case, for an arbitrary cross section x of beam, the
motion is in this way: first a series of positive–negative alternative bending elastic waves
pass through; at time t = x 2 /(4a 2 η2 ), the cross section gets into plastic bending under the
action of −MS , and the plastic deformation gradually goes up; then at time t = x 2 /(4a 2 η1 ),
the negative M reaches its extreme; after that the absolute value of M decreases, namely
the unloading of post-plastic deformation takes place. Along with the development of
unloading process, M transforms to positive, and then at time t = x 2 /(4a 2 η0 ) the cross
section gets into the secondary plastic deformation under the action of M. Later on the
plastic deformation monotonously goes up. On this analogy, if V1 is large enough, plastic
deformation and the accompanying unloading can take place one more time. The cross
section finally always gets into plastic bending under the action of +M provided that V1 is
positive. Before this final monotonously increasing plastic bending, how many times the
positive–negative alternative plastic bending and the accompanying unloading will take
place, depends on how large the V1 is. It should be noted that the interior unloading
discussed here should be distinguished from the unloading of external applied load.
In the following, the situation where an unloading and a secondary plastic loading take
place as shown in Fig. 10.9, will be further discussed in detail. For the situations where
the plastic loading takes place many times, the problem in principle can be treated in the
same way.
First, it is assumed for the M–k relation that the unloading curve is parallel to the elastic
straight line, and if the unloading begins at M(η1 ), then when the inverse reloading reaches
M(η0 ), which satisfies the condition |M(η0 ) − M(η1 )| = 2MS , the plastic deformation
takes place once again, as shown in Fig. 10.9(c). In other words, we have
M(η2 ) = −MS (10.65)

|M(η0 ) − M(η1 )| = 2MS (10.66)
This is equivalent to assuming that the unloading process is an elastic one and the plastic
reloading process takes into account the Bauschinger’s effect.
In the regions I–IV in Fig. 10.9, S(η) is obtained respectively as

1
I S = A cos η 0 ≤ η ≤ η0 ; (10.67a)
ξ
II S = −B sin(η − η1 ) η0 ≤ η ≤ η1 ; (10.67b)

1
III S = −C sin (η − η2 ) η1 ≤ η ≤ η; (10.67c)
ξ
IV S = C sin(η − η3 ) η2 ≤ η (10.67d)
where the conditions S (0) = 0 and S(η1 ) = 0 have been utilized, and consequently
the requirement of S(η) in regions II and III should be continuous at point η1 has
been naturally satisfied. The equation set (10.67) contains eight indeterminate coeffi-
cients: A, B, C, D, η0 , η1 , η2 , and η3 . They can be determined by the following eight
equations: two equations to express the requirement that S(η) in different regions should
be continuous at the connecting points (except η1 ), three equations to express the require-
ment that S (η) in different regions should be continuous at the connecting points, and
three equations [Eq. (10.57), (10.65), and (10.66)], namely
⎧ ⎫
⎪
⎪
1
η0 = −B sin(η0 − η1 ) (a)⎪
⎪
⎪
⎪ A cos ⎪
⎪
⎪
⎪ ξ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ A η0 ⎪
⎪
⎪ sin = B cos(η0 − η1 ) (b)⎪
⎪
⎪
⎪
⎪ ξ ξ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ C ⎪
⎪
⎪
⎪ B= (c)⎪
⎪
⎪
⎪
⎪
⎪ ξ ⎪
⎪
⎪
⎪ η2 − η1 ⎪
⎪
⎪
⎪ −C sin = D sin(η2 − η3 ) (d)⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎨ ξ ⎪
⎬
C η2 − η1 (10.68)
⎪ cos = −D cos(η2 − η3 ) (e)⎪
⎪
⎪ ξ ξ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪ ∞ S (η)
⎪
⎪
⎪
⎪
⎪
⎪
⎪ √ dη = 2V1 (f )⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ 0 η ⎪
⎪
⎪
⎪ ∞ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪
S(η)
(g)⎪⎪
⎪
⎪ √ dη = 2VS ⎪
⎪
⎪
⎪ η2 η ⎪
⎪
⎪
⎪ η1 ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪
S(η)
(h)⎪⎪
⎩ √ dη = −4VS ⎭
η0 η
The practical calculation process is as following: choose a pair of values η2 and η3 to

determine the values of η0 , η1 and the ratios A:B:C:D from Eqs. (10.68a–c). Then by
trial and error, find a pair of values η2 and η3 to satisfy Eqs. (10.68g, h). The impact
velocity V1 is then calculated from Eq. (10.68f ). This calculation is repeated until the
given V1 is obtained. The integrals involved in the calculation can be easily reduced to
the Fresnel integrals, the values of which can be found from the integral table (e.g. see
Sparrow, 1934).
(3) In perfectly plastic cases
If the M–k relation of beam is described by the so-called perfect plastic model, as shown
by curve 2 in Fig. 10.6, then the calculation will be much simplified. Such a situation can
be regarded as a special case of linear hardening plastic model when ξ = 0.
Let us first discuss the situation where at each beam cross section the plastic deformation
only takes place once, as shown in Fig. 10.8 but now ξ = 0. Since (η0 /ξ ) < π/2, region I
will shrink to a point η = 0. At this point, M(0) = MS while k can increase infinitely,
namely a plastic hinge is formed. So the solution of the current problem is reduced to
S = −B sin(η − η1 ), 0≤η (10.69)
The indeterminate coefficients B and η1 are determined by the condition of M(0) = MS

and Eq. (10.57):
⎧ ∞ cos(η − η )
⎪
⎪ √
1
dη = −2V1
⎨B η
0
∞ (10.70)
⎪
⎪ sin(η − η1 )
⎩B √ dη = 2VS
0 η
Because
$ ∞ cos(η−η ) = cos$η cos η1 + sin η sin η1√, sin(η−η1 ) = sin η cos η1 −cos η sin η1
√1 ∞ √
and 0 (cos η/ η)dη = 0 (sin η/ η) dη = π/2, from the above two equations we
obtain
⎧ 9
⎪
⎪ π
⎨B 2 cos η1 = VS − V1
⎪
9 (10.70a)
⎪
⎪ π
⎪
⎩B sin η1 = −(VS + V1 )
2
and then we obtain

⎧
⎪
⎪ V12 + VS2
⎪
⎪
⎨B = 2
π
(10.71)
⎪
⎪
⎪
⎪ −1 V1 + V S π −1 VS
⎩η1 = tg = + tg
V1 − V S 4 V1
Note that since a plastic hinge is formed at point x = 0, (∂w/∂x) at x = 0 is no longer

continuous and k at x = 0 could increase infinitely. In fact, according to Eq. (10.54), now
we have
√ √
∂w t t
tan θ = α|x=+0 = = − S (η) = B cos η1 (10.72a)
∂x x
=+0 a
η=0 a
or from Eq. (10.70a) it can be rewritten as
9
21 √ 2kS √
tan θ = α|x=+0 = (VS − V1 ) t = (VS − V1 ) t (10.72b)
πa πVS
√
It indicates that α(0, t) increases in direct proportion to t. Note that with regard to α the
deformation of beam is anti-symmetric with respect to the impact point, so the solution
of α|x=−0 is obviously the reversal of the sign of the solution of α|x=+0 expressed by
Eq. (10.72). It can be noted from Eq. (10.50a) that a discontinuity of a at x = 0 must
correspond to a discontinuity of S(η)|η=0 . In fact, it can be directly seen from Eq. (10.69)
that the condition of S (0) = 0 expressed by Eq. (10.51) no longer holds, and the value of

S (η)η=+0 = − √at α is finite.
x=+0
The above solution holds only when |M(η)| < MS , (η

= 0). With increasing V1 , we will
first have M(η1 ) = −MS . According to Eq. (10.53) we obtain
∞
EIB sin(η − η1 ) EIV S
M(η1 ) = √ dη = −MS = − 2
2a 2 η1 η a
∞ ∞
sin η cos η
B cos η1 √ dη − sin η1 √ dη = −2VS
η1 η η1 η
√
From Eq. (10.70a) we know 2VS = B (π/2)(cos η1 − sin η), and eliminating B from
the above equation we obtain
η1 η1
1 cos η 1 sin η
sin η1 √ √ dη − cos η1 √ √ dη = sin η1 − cos η1
2π 0 η 2π 0 η
The solution of this equation is η1 = 70.6◦ , and correspondingly we have V1 = 2.087VS ,

which is the critical velocity of the above solution. In other words, when V3 < V1 <
2.087VS , we have the solution (10.69). In such a case, only at x = 0 a plastic hinge
is formed and the rotating angle of plastic bending at the impact point is determined by
Eq. (10.72), while remaining part of the beam is in the state of elastic bending.
When V1 = 2.087VS , region III as shown in Fig. 10.9 will appear. However, since
(η2 − η1 )/α < π, when a approaches zero, region III will shrink at point η = η1 .
The solution of the current problem is:

S = −B sin(η − η1 ), 0 ≤ η ≤ η1
(10.73)
S = −D sin(η − η1 ), η1 ≤ η
The indeterminate coefficients B, D, and η1 can be determined by the following three

conditions: Eq. (10.57), M(η1 ) = −MS and –M(η1 ) = 2MS , namely
⎧ η1 ∞ ⎫
⎪ cos(η − η1 ) cos(η − η1 ) ⎪
⎪
⎪ B √ dη + D √ dη = −2V ⎪
⎪
⎪
⎪ η η
1 ⎪
⎪
⎪
⎪ 0 η1 ⎪
⎪
⎪
⎨ ∞ sin(η − η ) ⎪
⎬
1
D √ dη = −2VS (10.74)
⎪
⎪ η ⎪
⎪
⎪
⎪ η
η1
⎪
⎪
⎪
⎪ 1 sin(η − η ) ⎪
⎪
⎪
⎪B 1 ⎪
⎪
⎩ √ dη = 4VS ⎭
0 η
For the same reason as was discussed earlier, because regions I and III now shrink to two
points, the discontinuity of S (η) appears at point η = 0 and point η = η1 , as shown in
Fig. 10.10.
It can be seen from the basic equation (10.41) that these two points are singular points,
S → ∞. Thus, corresponding to η = 0, a stationary plastic hinge is formed at x = 0;
and corresponding to the discontinuity of S (0), α(0, t) is discontinuous at x = 0, namely
the deflection curve of beam possesses a nonzero angle α
= 0 at x = 0, the value of which
is still calculated according to Eq. (10.72).√On the other hand, corresponding to η = η1 ,
another
√ plastic hinge is formed at x1 = 2a η1 t. Noting that x1 is directly
√ proportional to
t, so this plastic hinge moves with a velocity of Cη1 = dx 1 /dt = a η1 /t in the direction
deviating from the impact point. Corresponding tothe discontinuity of S (η), a jump of

curvature ∆k happens at point x1 , although α|x=x1 = ∂w
∂x x=x1 is still continuous. It can
be physically explained as follows. Since a constant bending moment MS acts at the plastic
0
h1 h
Fig. 10.10. The discontinuity of S (η) appears at η = 0 and η = η1 .

hinge, the curvature can increase infinitely, although it needs time for such a process. For a
stationary plastic hinge, a finite discontinuity of a takes place after a finite interval, and the
jump level of discontinuity increases continuously with time, as described by Eq. (10.72).
However, for a traveling plastic hinge, since it only passes instantaneously through each
cross section, there is not enough time to let the discontinuity of α to take place, so α|x=x1
is still continuous. It can be seen from Eq. (10.50), the discontinuity of S (η) now only
means the discontinuity of curvature k or the discontinuity of angular velocity ω, and
obviously the jump level of discontinuity of curvature ∆k is
1 ! kS 1
∆k = k(η1 −0)−k(η1 +0) = √ S (η1 +0)−S (η1 −0) = · √ (B −D)
2a 2 η1 2VS η
(10.75)
Once B, D, and η1 are obtained from Eq. (10.74), θ and ∆k can be calculated from
Eqs. (10.72) and (10.75). When V1 VS , we have the following approximate equations
given by Duwez et al. (1943, 1950)
⎧ 3VS
⎪
⎨η1 ≈ V
⎪
1
9 (10.76)
⎪
⎪ √ 3/2
⎩tgθ ≈ t kS V1
3 VS
In Fig. 10.11, the deflection distribution curves and the bending moment distribution
curves are given for three different situations, namely (a) V1 < VS (elastic beam),
(b) VS < V1 < 2.087VS (one stationary plastic hinge), and (c) V1 > 2.087VS (stationary
and traveling plastic hinges).
Thus, in the case of perfectly plastic M–k relation, the plastic bending of beam is accom-
plished through the stationary plastic hinge and the traveling plastic hinges which propagate
longitudinally along the beam axis.
The deflection curves of rectangular cross section beams under transverse constant-velocity
impact were experimentally measured by Duwez et al. (1943, 1950) too. The tested
beams were made of two different metals, namely, cold-rolled low-carbon steel, and
annealed copper. The actual M–k curve experimentally measured (plotted by solid line)
and the approximate M–k curve used for computation (plotted by dashed line) are shown
in Fig. 10.12 for steel and in Fig. 10.13 for copper, respectively. The comparisons between
the experimental and the computational deflection curves for both metals are given in
Figs. 10.14 and 10.15, respectively.
As can be seen, the shapes of the theoretical curves and the experimental curves are similar.
In the case of the cold-rolled steel beams, the elastic deflection curve agrees rather well
with the experimental curve. However, in the case of the annealed copper beams, the
elastic–plastic curve is in rather good agreement with the experimental curve. Moreover,
it was observed from the experiments that plastic strains are localized in a very small region
near the impact point in the steel beam, while plastic strains extend to some distance from
the impact point in the copper beam. Thus, as explained by the authors, it seems that the
influence of plasticity is greater in copper beam than in steel beam.
−1
1.4 1.6 1.8

1.0 1.2 2.0
0 0.2 0.4 0.6 0.8
h=4 r0 A0 x
EI 2 t
+1 v1/vs=0.3 (a)
W 2 (b)
v1t M 5
+1 Ms 10 } (c)
r0 A0 x
h=4
EI 2 t
−1
P(t) Ms Ms
Ms Ms
v1<vs −Ms −Ms −Ms −Ms

vs<v1<2.087vs v1>2.087vs
(a) (b) (c)
Fig. 10.11. Deflection curves and bending moment curves for an infinite beam under transverse
constant-velocity impact V1 in different situations: (a) V1 < VS , (b) VS < V1 < 2.087VS , and
(c) V1 > 2.087VS .
10000
1
M (in. 1b)
0 5 10 15
k(10−2 in−1.)
Fig. 10.12. M–k curve for cold-rolled steel beam (solid line 1 – experimentally measured curve
and dashed line 2 – approximate curve). [Reproduced from Duwez et al. (1950) by permission
of ASTM]
1000
M (in. 1b)
2 1
0 8 16 24
k (10–2 in–1.)
Fig. 10.13. M–k curve for annealed copper beam (solid line 1 – experimentally measured curve
and dashed line 2 – approximate curve). [Reproduced from Duwez et al. (1950) by permission
of ASTM]
0 5 10 15 25 30 35 40
w (in.)
x (in.)
b a
Fig. 10.14. Deflection curves for cold-rolled steel beam (solid line a – experimentally measured
curve, dotted line b – computed elastic–plastic beam curve, and dashed line c – computed elastic
beam curve). [Reproduced from Duwez et al. (1950) by permission of ASTM]
0 5 10 15 20
w (in.)
a x (in.)
b
2
c
4
Fig. 10.15. Deflection curves for annealed copper beam (solid line a – experimentally measured
curve, dotted line b – computed elastic–plastic beam curve, and dashed line c – computed elastic
beam curve). [Reproduced from Duwez et al. (1950) by permission of ASTM]
With regard to some appreciable differences between the computational and the experi-
mental curves, it can be attributed to the following main reasons: (1) The M–k curve used
for computation is an approximation; (2) The experimental results may be influenced by
the bending waves reflected from the beam ends; and (3) The elementary theory itself
contains a series of basic assumptions, particularly the rotatory inertia and the shear stress
effect are ignored; as well as the computation is not based on the dynamic M–k curve but
on the static M–k curve.
In addition, the above experiments verified that the distance x0 at which w = 0, is directly
proportional to t 1/2 , namely, when the bending wave propagates along the beam axis its
velocity is not a constant but dependent on x 2 /t.
10.4 Rigid-Plastic Analysis
The rigid-plastic analysis of beam was developed by Lee and Symonds (1952). Similar
to the earlier discussion on the plastic waves in bars, if the plastic deformation is much
larger than the elastic deformation, then the elastic deformation could be ignored, and the
problem is reduced to an analysis of the so-called rigid-plastic beam.
Assume that the M–k relation is simplified as shown in Fig. 10.16. It is equivalent to
making such an assumption: when |M| < M0 , the beam behaves as a rigid body, while
when |M| = M0 at any cross section of the beam, the curvature can increase infinitely,
namely, the so-called plastic hinge is formed, as illustrated in Fig. 10.17. Thus, the problem
is reduced to the rigid motion of beam segments which are connected to each other by
M0
0
k
Fig. 10.16. Rigid-plastic M–k curve.
M =M0
M <M0
Fig. 10.17. Plastic hinge at |M| = M0 .

P(t)
0
l (I)
l
(II)
q0
ql
q0 (III)
Fig. 10.18. Three phases of movement in the rigid-plastic analysis for a finite straight beam with
constant section area under mid-point impact.
plastic hinges. It should be noticed that an actual plastic hinge is gradually formed and
occupies a finite plastic deformation region. Only if the plastic deformation is much larger
than the elastic deformation, then we can regard the plastic hinge as a singular cross
section formed suddenly, and the plastic deformation is concentrated on this cross section.
The problem will be much simplified by such an approximation. Not only the problem of
finite beam can be solved, including the unloading of external load, but also the external
load condition is not limited to the constant-velocity impact, these could not be treated in
Section 10.3.
Now let us consider a finite straight beam with a length of 2l, two stress-free ends and
constant cross-sectional area, which is subjected to an impulsive force P (t) at the central
point, as shown in Fig. 10.18.
The movement of the beam can be described by dividing it into three stages or three phases.
In phase I, M < M0 , only translational motion of rigid body is involved; In phase II, at
the impact point M = M0 , there exists the relative rotation of two halves of beam around
the middle plastic hinge; In phase III, at the cross section departing from the impact point
at a certain distance M = −M0 , in addition to the middle stationary plastic hinge, there
are two traveling plastic hinges on the two sides of beam.
In the following, let us discuss each phase in more detail.

PI
Phase I µ = <4 :
M0
When P (t) starts to increase from zero, the beam also begins to move in parallel to itself as
a motion of rigid body. In this stage, the acceleration a of each point is the same, which is
P
a= (10.77)
2ml
where m is the density of beam per unit length (line density). This is equivalent to a static
beam subjected to a uniformly distributed load −ma. So the bending moment distribution is
l−x m(l − x)2 P (l − x)2
M(x) = maξ dξ = a= (10.78)
0 2 4l
Obviously, the maximum bending moment exists at the central point of beam, x = 0,
which is
ml2 Pl
M(0) = a=
2 4
Therefore, when P = 4M0 / l, we have M(0) = M0 , a plastic hinge is then formed at the
central point, and the beam no longer moves in the form of the translational motion of
rigid body. By introducing a dimensionless parameter µ = Pl/M0 , when it satisfies the
following condition
Pl
µ= =4 (10.79)
M0
Phase I ends and Phase II begins.
Phase II (4 ≤ µ < 22.9):
In this stage, there is a constant bending moment M0 acting at the middle plastic hinge.
The two halves of the beam still move in the form of rigid motion, but they all relatively
rotate around the plastic hinge, as shown in Fig. 10.19. Let a0 be the linear acceleration
at the central point of beam x = 0, θ0 the rotation angle of the right half of the beam
around the hinge (which is positive in the anti-clockwise direction), α0 = d 2 θ0 /dt 2 the
corresponding angular acceleration, then the momentum equation, and the moment of
momentum equation (the moment with respect to x = l/2) for the right half of the beam
are expressed respectively as

P l l d 2 θ0 lα0
= ma0 dx − x dx = ml a 0 − (10.80)
2 0 0 dt 2 2
0
x
P
P/2
w q0 a
M0 l
u0
Fig. 10.19. Relative rigid rotations of the two halves of the beam around the hinge.

Pl l/2 d 2 θ0 ml3
− M0 = 2 m x 2 dx = α0 (10.81)
4 0 dt 2 12
From these equations, a0 and α0 can be solved and expressed as

M0 2Pl M0
a0 = 2 − 6 = 2 (2µ − 6) (10.82)
ml M0 ml

M0 3Pl M0
lα0 = 2 − 12 = 2 (3µ − 12) (10.83)
ml M 0 ml
When P (t) is given, both linear acceleration a0 (t) and angular acceleration α0 (t) can be
calculated from the above equations. Then by integrating with respect to time t, the linear
velocity and the angular velocity can be further calculated, and displacement w(0, t) and
angular displacement θ0 can be finally calculated.
Note that even when P (t) decreases, the above equations still hold as long as the plastic
hinge does not disappear. In fact, with decreasing P (t), first the sign of a0 and α0 will
change from positive to negative, while the sign of angular velocity is still positive,
which means that the plastic hinge does not immediately disappear. Even after P (t) has
decreased to zero, the plastic hinge can still exist. This is because of the fact that after the
unloading process of external load begins, the rigid kinetic energy of beam segments can
still transform to the plastic deformation energy, and thus the additional plastic deformation
is further induced. As can be seen from this example, the problem concerning the unloading
process of external load can be solved by the rigid-plastic analysis.
During Phase II, except the central plastic hinge at x = 0, the bending moment at any
other cross section of the beam is smaller than M0 . If P (t) increases further, new plastic
hinge may appear at other cross sections. The critical condition for the end of Phase II will
be discussed as follows.
Let us first look at the bending moment distribution, which can be obviously expressed as
(see Fig. 10.19)

P x x d 2 θ0
M(x) = x − M0 − ma0 x dx + m (x − ξ )ξ dξ
2 0 0 dt 2
(10.84a)
P mx 2 mx 3
= x − M0 − a0 + α0
2 2 6
By substituting Eqs. (10.82) and (10.83) into the above equation, we have
µM0 x 1 x 2 1 x 3
M(x) = − M0 − M0 (2µ − 6) + M0 (3µ − 12)
2 l 2 l 6 l
(10.84b)
µ−4x x 2
= M0 −1 1−
2 l l
The maximum of bending moment appears at the cross section x = x1 , where

(dM/dx)x1 = 0, that is
P 1
− ma0 x1 + ma0 x12 = 0
2 2
or
x 2 4 µ − 3 x1 µ
1
− + =0
l 3µ−4 l 3(µ − 4)
The above equation has two roots. This equation and the corresponding M(x1 ) can be
solved as
⎧ x1
⎪
⎨ = 1, M(x1 ) = 0
l (10.85)
⎪ x µ 9µ2 − 2µ3
⎩ 1 = , M(x1 ) = M0 1 +
l 3(µ − 4) 27(µ − 4) 2
Let us first discuss the second root, which is related to the beginning of Phase II. Due to
x1 / l ≤ 1, the second root exists only when µ ≥ 6. In fact, in the range of 4 ≤ µ ≤ 6,
the bending moment distribution monotonously decreases with x, and the minimum
appears at x = l, i.e. M(l) = 0, which corresponds to the first root. If by substituting 4 ≤
µ ≤ 6 into Eq. (10.85), the value of M(x1 ) is positive, but M(x1 ) decreases with increas-
ing µ or P (t); when µ = 6, M(x1 ) = 0, and when µ > 6 the value of M(x1 ) is negative.
Thus, the critical condition for the end of Phase II is determined by the following equation:

9µ2 − 2µ3
M(x1 ) = M0 1+ = −M0
27(µ − 4)2
This is a cubic equation with respect to µ:
2µ3 − 63µ2 + 432µ − 864 = 0
Its only real root is µ = 22.9, and corresponding to this real root we have
µ
x1 = l = 0.404l (10.86)
3(µ − 4)
Therefore when µ = 22.9, M = −M0 at x = ±0.404l two new plastic hinges appear,
so Phase II ends and Phase III begins.
Phase III (µ ≥ 22.9):
When µ = 22.9, Phase III begins. If the external load increases continuously, then the limit
bending moment M0 will take place at x < 0.404l. In other words, with increasing external
load, the plastic hinge originally at x1 will move (travel) towards the impact point x = 0.
This may not be evident immediately, but can be seen from a further discussion by solving
the motion equations together with the kinematic conditions across the hinge as follows.
First of all, it should be pointed out that as soon as the plastic hinge moves to a new
position, the bending moment M at the original position decreases to a value smaller than
M0 , and consequently the relative rotation no longer exists there. In fact, on the left and
right sides of plastic hinge, there must be M < M0 . Otherwise, a region of constant M
must exist. But according to the relation dM/dx = Q, the shear force must be zero in this
region, and consequently the acceleration of all particles in this region should be zero too.
It means that this region moves in the form of rigid motion, no longer as a plastic hinge.
For convenience, let us consider the right half of the beam. Now imagine that at an
instance t, the traveling plastic hinge is at position xh . Since both the left side and the
right side of xh are rigid bodies, the momentum equations and the moment of momentum
equations for them can be expressed respectively as (see Fig. 10.20)

P xh xh d 2 θ0 α 0 xh
= ma0 dx − m x dx = mx h a 0 − (10.87)
2 0 0 dt 2 2
xh /2
Px h d 2 θ0 mx 3 α0
− 2M0 = 2 m 2 x 2 dx = (10.88)
4 0 dt 12
l l
2
d θl αl (l − xh )
0= mal dx + m 2 x dx = m(l − xh ) al + (10.89)
xh xh dt 2
(1−xh )/2
d 2 θl m(l − xh )2
M0 = 2 m 2 ξ 2 dξ = α1 (10.90)
0 dt 12
where al is the linear acceleration at point x = l, al = d 2 θl /dt 2 is the angular acceleration

of rigid segment on the right side of the traveling plastic hinge. The above dynamic
conditions, totally four equations, are not enough to solve the five unknown variables
a0 , α0 , al , αl , and xh . Thus the kinematical condition of the traveling plastic hinge should
be further considered as follows.
Now let us consider the kinematical equation for the traveling plastic hinge. Denoting the
left side and the right side of the plastic hinge position by xh− and xh+ respectively, the
xh al
al
Wl
x
W0
M0
q0 ql P/2
M0
M0
w a0 x a0 l-xh
h
(a) (b) (c)
Fig. 10.20. The traveling plastic hinge moves towards the impact point (µ ≥ 22.9).
continuity of displacement w at the plastic hinge is expressed as

w xh− = w xh+
which should hold regardless of the time and the position of the plastic hinge. In other
words, for the left and right sides of the hinge, the total derivatives of displacement
following the traveling plastic hinge should be the same:
d ! d !
w xh− = w xh+
dt dt
− −
∂w xh ∂w xh ∂w xh+ ∂w xh+
+c = +c
∂t ∂t ∂t ∂x
where c = dx h /dt is the traveling velocity of the plastic hinge along the beam axis.
Therefore, the following relation between the transverse velocity v = ∂w/∂t and the
rotating angle θ = −(∂w/∂x) exists
F G
v xh− − v xh+ = +c θ xh− − θ xh+ (10.91)
As mentioned in Section 10.3, when a traveling plastic hinge moves along the beam axis,
it instantaneously (in an infinitesimal time duration) passes through each cross section,
there is not enough time to let the discontinuity of rotation
angle to
take place, so the
rotation angle θ across the hinge is continuous. Since θ xh− = θ xh+ , by Eq. (10.91) the
transverse velocity is also continuous. Similarly, since we have
d − d +
θ xh = θ xh
dt dt
d − d +
v xh = v xh
dt dt
correspondingly we can establish the following kinematical relation between angular

velocity ω = ∂θ/∂t and curvature k = ∂θ/∂x = −(∂ 2 w/∂x 2 ):
F G
ω xh− − ω xh+ = −c k xh− − k xh+ (10.92)
and the kinematical equation between angular velocity ω = ∂θ/∂t = −(∂ 2 w/∂t∂x) and
the transverse acceleration a = ∂v/∂t = ∂ 2 w/∂t 2 :
F G
a xh− − a xh+ = c ω xh− − ω xh+ (10.93)
Since angular velocity ω is discontinuous across the plastic hinge, it can be seen
from Eqs. (10.92) and (10.93) that transverse acceleration a and curvature k must be
discontinuous too.
Having the above results, the velocities and the accelerations for both the sides of the
plastic hinge are obtained as following
x
w = w0 − θ dx (10.94)
0
for x < xh ,
x
∂w ∂w0 ∂θ dθ0
v= = − dx = v0 − x (10.95)
∂t ∂t 0 ∂t dt

∂ 2w ∂ 2 w0 x ∂ 2θ
a= 2 = − dx = a0 − α0 x (10.96)
∂t ∂t 2 0 ∂t 2
for x > xh ,

∂w ∂w0 xh ∂θ x ∂θ dx h − !
v= = − dx − dx − θ xh − θ xh+
∂t ∂t 0 ∂t xh ∂t dt
(10.97)
dθ0 dθl
= v0 − xh − (x − xh )
dt dt

∂ 2w dx h dθ0 dθl
a= = a0 − α x
0 h − α l (x − x h ) − −
∂t 2 dt dt dt
(10.98)
dx h
= a0 − α0 xh − αl (x − xh ) − (ω0 − ωl )
dt
From Eqs. (10.95) and (10.97)

it is also
indicated that the transverse velocity is continuous
across the plastic hinge v xh− = v xh+ .
From Eqs. (10.95) and (10.97), we have respectively,

a xh− = a0 − α0 xh (10.99)
dx h
a xh+ = a0 − α0 xh − (ω0 − ωl ) (10.100)
dt
On the other hand, we have

a xh+ = al + αl (l − xh ) (10.101)
then from Eqs. (10.100) and (10.101) we obtain an additional equation (kinematic
condition) to relate a0 , α0 , al , αl , and xh :

dx h dθ0 dθl
a0 − α0 xh − al − αl (x − xh ) = − (10.102)
dt dt dt
By introducing ξ = xh / l and µ = Pl/M0 , Eqs. (10.87)–(10.90) and (10.102) can be

rewritten as

a0 M0 2µ 12
= 3 − 2 (10.103)
l ml ξ ξ

d 2 θ0 M0 3µ 24
α0 = = − (10.104)
dt 2 ml 3 ξ 2 ξ
al −M0 6
= (10.105)
l ml (1 − ξ )2
3
d 2 θl M0 12
αl = = (10.106)
dt 2
ml (13− ξ )3

M0 µ 12 6 dξ dξ dθ0 dθl
− 2 + = − (ω0 − ωl ) = − − (10.107)
ml 3 ξ ξ (1 − ξ )2 dt dt dt dt
The motion of beam in Phase III will be determined from Eqs. (10.103)–(10.107). They
hold regardless of whether the external load increases or decreases from a certain µ > 22.9,
only if dθ0 /dt > dθl /dt (the traveling plastic hinge does not disappear).
The solutions of these equations are complicated due to the presence of the term
(dξ/dt) (ω0 − ωl ). If this term is absent, then a simple relation exists between µ and
ξ , and a0 , α0 , a1 , and αl can be easily expressed as functions of µ. The velocities and the
displacements can be obtained by direct integration. At present, since the relation between
µ and ξ involves unknown quantities, dξ/dt, dθ0 /dt, and dθl /dt, it is impossible to solve
the velocities and the displacements by direct integration. In such a case, the problem can
be solved by either of the following two methods.
One is the method of successive approximations. First, set (dξ/dt) (ω0 − ωl ) = 0, a first
approximate solution can be solved from Eqs. (10.103)–(10.106), and then (ω0 − ωl ) and
dξ/dt can be calculated by integration. Substituting this approximate (dξ/dt) (ω0 − ωl )
into Eq. (10.107), the second approximate solution can be solved from the equation set.
The process should be repeated until the difference (dξ/dt) (ω0 − ωl ) is calculated from
the previous approximation and the present approximation is less than the required error.
Another one is the method of step-by-step integration. First, µ1 and ξ 1 at some time t1 is
used to calculate the accelerations at this instant from Eqs. (10.103)–(10.106). Then the
velocities at time t2 = t1 + ∆t can be calculated, for example
(ω0 )2 = (ω0 )1 + (α0 )1 ∆t
After (ω0 − ωl ) at time t2 is calculated, µ2 and ξ2 at t2 can be solved from Eq. (10.107)
by repeated trials. Then the accelerations at time t2 can be evaluated again from
Eqs. (10.103)–(10.106), and ω0 and ωl at t2 can be re-calculated, for example
(ω0 )2 = (ω0 )1 + 1
2 [(α0 )1 + (α0 )2 ] ∆t
After the new ξ2 and µ2 is re-calculated, such a process can be repeated to obtain the
values at a new time t3 = t2 + ∆t and so forth.
In the motion of Phase III, if the external load begins to decrease (unloading), the problem
can be dealt with by the same principle as discussed in Phase II. Equations (10.104)–
(10.107) still hold so long as the traveling plastic hinge does not disappear. If Pm , the
maximum of P (t), is large enough, there still will be a relative angular velocity (ω0 − ωl ),
even P (t) decreases to zero. In this case, plastic hinge continues to move until ω0 = ωl .
The final position ξmax where the plastic hinge just disappears is obtained from Eq. (10.107)
with µ = 0 and (dξ/dt) (ω0 − ωl ) = 0.
√
ξmax = 2 − 2 = 0.586 (10.108)
After the traveling plastic hinge disappears, the action of the stationary plastic hinge at the
central point still continues. In this case, Eqs. (10.28) and (10.83) deduced for Phase II
should be used to solve the problem in this stage. Eventually the central stationary hinge
disappears (ω0 = 0), the beam thereafter translates as a rigid body without any further
deformation. The final plastic strain is the permanent deformation caused by the impact.
As mentioned earlier, this rigid-plastic analysis is expected to provide satisfactory results

when the plastic deformation is much larger than the elastic deformation, in other words the
plastic work should be much larger than the elastic strain energy. In the current problem,
the plastic work done at the central stationary hinge is M0 θ0 . On the other hand, the
elastic bending strain energy for a beam with length l under the action of bending moment
M is known as (1/2)Mθ = Ml/2P = M 2 l/2EI, and its maximum is M02 l/2EI. The
requirement that the plastic work should be much larger than the elastic strain energy means
M02 l
M0 θ0 (10.109)
2EI
For the beam with two stress-free ends, its fundamental frequency of elastic vibration fe
is known as

(4.73)2 EI
fe = ·
2πl 2 ρ0 A0
or its fundamental period τ is

9
1 2πl 2 ρ0 A0
τ= = ·
fe (4.73)2 EI
Then Eq. (10.109) can be rewritten in a dimensionless form as
θ0 1 τ2
H 3 (10.110)
M0 T 2 ml 2.5 T 2
where T is the impact duration. The above equation can be regarded as a criterion to eval-
uate whether the rigid-plastic analysis could be applied. In addition, the present analysis
is also limited in the situation where θ0 ≤ 10◦ , since it has been assumed in the present
theory that the deformation is sufficiently small, namely θ ≈ −(∂w/∂x).
10.5 Shear Failure of Beams under Transverse Impact
In the previous sections, the bending waves propagating in the beam subjected to transverse
impact were mainly discussed. However, it should be emphasized again, as we have
pointed out in Section 10.2, that the bending waves are in fact resulted from the interaction
of two kinds of waves, M–ω waves and Q–v waves, which are coupled with each other.
It implies that the effect of shear force effects have been taken into account. So, shear
force plays more important role in the dynamic theory than in the static theory of beams,
since the transverse inertia is included in the present governing equation set.
With regard to the dynamic failure modes for beams subjected to transverse impact, a num-
ber of experimental and theoretical investigations revealed (Menkes and Opat, 1973;
Jones, 1989) that there are at least three basic failure modes, namely, large ductile defor-
mation or large deflection (Mode I), tensile tearing (Mode II), and transverse shearing
(Mode III). The former two modes mainly resulted from bending wave propagation, while
the third one is closely related to the transverse shear wave propagation.
In the following, the propagation characters of plastic shear wave in beams will be
discussed (Wang and Jones, 1996).
In accordance with the discussions in Section 10.1, assume that the plane cross section
perpendicular to the beam axis before the deformation remains a plane cross section
perpendicular to the beam axis after the deformation (plane cross section assumption).
Denote the transverse shear force and the transverse displacement on the cross section
of the beam by Q and w, then correspondingly, the shear strain is γ = ∂w/∂X and the
transverse particle velocity is v = ẇ = ∂w/∂t. Introduce m(= ρ0 A0 ) to denote the linear
density of the beam, where ρ0 is the initial density of the beam and A0 is the initial section
area. Similar to the strong discontinuous longitudinal waves propagating in the bar as
discussed in Chapter 2 [see Eqs. (2.55) and (2.57)], for the shear disturbance propagating
in the form of strong discontinuity (shear hinge), obviously there should exist the following
dynamic compatibility condition (momentum conservation)
[Q] = −mξ̇ [ẇ] (10.111)
and the following kinematic compatibility condition (continuity condition)
[ẇ] = −ξ̇ [γ ] (10.112)
where ξ̇ = dξ/dt indicates the propagation velocity of strong discontinuous shear distur-
bances propagating along the neutral axis of beam (X axis) and ξ the position of shear
hinge on the X axis.
From Eqs. (10.111) and (10.112), we have
[Q]
ξ̇ 2 = (10.113)
m[γ ]
It means that the propagation velocity is mainly dependent on Q–γ character of the beam.
Completely similar to the situation discussed in Chapter 2 for the strong discontinuous
longitudinal wave propagating in bars (see Section 2.6), only when the plastic part of
Q–γ curve shows a linear hardening character (dQ/dγ = constant) or an increasing
hardening character (dQ/dγ > 0), can the plastic shear hinge forms.
In the case of rigid-linear hardening plastic beam, as shown in Fig. 10.21, its Q–γ relation
is expressed as
γ =0 if Q < Q0 (10.114a)
Q = Q0 + Gp γ if Q ≥ Q0 (10.114b)
where Q0 indicates the yield shear force and Gp the linear hardening modulus. Then the
propagation velocity of plastic shear hinge is invariable:
9
Gp
CQ = ± (10.115)
m
where the plus and minus signs correspond to the plastic shear hinge propagating in the
positive and negative directions along X axis, respectively.
Note that when Gp approaches zero, Eq. (10.114) is reduced to the rigid-perfectly plas-
tic Q–γ relation, and then ξ̇ = 0 which is corresponding to a stationary plastic shear
hinge. In such a case, it can be seen from the continuity condition across a strong discon-
tinuous interface [Eq. (10.112)] that the transverse velocity across the shear hinge must
Gp
Q0
0
g
Fig. 10.21. The Q–γ curve for rigid-linear hardening plastic beam.
G M
V0
0 X
L L
Fig. 10.22. Simple supported beam with length 2L subjected to a uniformly distributed impulsive
velocity V0 .
be continuous, namely [ẇ] = 0. Hence the plastic hardening character of Q–γ relation is
of significance for the study of shear hinge.
In the following, the shear failure of a simple supported beam subjected to transverse
impact is analyzed as an example.
Consider a simple supported beam with length 2L subjected to a uniformly distributed

impulsive velocity V0 , as shown in Fig. 10.22.
According to the momentum theorem and the moment of momentum theorem [Eqs. (10.1)
and (10.2)], when the rotatory inertia effects are disregarded, we have
∂Q
= mẅ (10.116a)
∂X
∂M
=Q (10.116b)
∂X
The initial and boundary conditions for the problem shown in Fig. 10.22 are
w = 0, ẇ = V0 , for t = 0 (10.117a,b)
Q=0 for X = 0 (10.117c)
M=0 for X = ±L (10.117d)
Only one-half of the beam, 0 ≤ X ≤ L, is considered henceforth owing to the symmetry

about X = 0. If a traveling plastic shear hinge moves from X = L in the negative
direction of X with velocity ξ̇ = −CQ towards the beam mid-span X = 0, then jumps in
the transverse shear force, shear strain, and transverse particle velocity across the plastic
shear hinge, namely, [Q], [γ ], and [ẇ] are, respectively
[Q] = (−Q1 ) − (−Q0 ) (10.118a)

[γ ] = (−γ1 ) − 0 (10.118b)
[ẇ] = 0 − ẇ (10.118c)
where the subscript 1 denotes the quantity just behind the traveling hinge. Thus, now
Eqs. (10.111) and (10.112) are respectively reduced to
Q1 − Q0 = mC Q ẇ (10.119a)
ẇ = CQ γ1 (10.119b)
According to the rigid-linearly hardening plastic model, both parts of the beam on either
side of the traveling hinge are rigid, so the transverse velocity distribution with respect to
X for t > 0, as shown in Fig. 10.23(a), is
ẇ = V = Ẇ for 0 ≤ X ≤ ξ (10.120a)
ẇ = 0 for ξ ≤ X ≤ L (10.120b)
where V = Ẇ (t) denotes the transverse velocity at X = 0, ξ is the transient location of

the plastic shear hinge and in the present case
ξ = L − CQ t (10.121)
The transverse shear force distribution can be obtained from Eq. (10.116a), as shown in
Fig. 10.23(b)
Q = mẄ X for 0 ≤ X ≤ ξ (10.122a)

Q = −Q1 (t) for ξ ≤ X ≤ L (10.122b)
where Ẅ = d Ẇ /dt, which can be obtained from Eq. (10.117c) and by substituting
Eqs. (10.118a) and (10.121) into Eq. (10.122)
Q0
Ẅ = − (10.123)
m(L − CQ t)
Integrating this equation with respect to t, together with the initial condition in
Eq. (10.117b), we have

Q0 CQ t
ẇ(t) = Ẇ (t) = V0 + ln 1 − for 0 ≤ X ≤ ξ (10.124)
mC Q L
Note that Ẇ (t) decreases with time t, since ln(1 − (CQ t/L)) < 0 when ξ < L. Thus,
there exists a time TS which satisfies the condition Ẇ (TS ) = 0 and consequently [ẇ] = 0.
x L
X
0
CQ t
. V
w
(a)
Q0
Q Q1
(b)
w W
(c)
g
g1
(d) g2
M M
(e)
Fig. 10.23. Simply supported beam subjected to impulsive loading at time t < TS . (a) Transverse
velocity profile, (b) shear force distribution, (c) transverse displacement profile, (d) shear strain
distribution, and (e) bending moment distribution.
In other words, when t = TS , the traveling hinge disappears at the corresponding location
X = ξS , and the motion of the whole beam ceases at the same time. TS can be determined
from Eq. (10.124) setting ẇ(t) = 0

L mC Q V0
TS = 1 − exp − (10.125a)
CQ Q0
Note that the beam segment 0 ≤ X ≤ ξS reaches its final state at t = TS , while the motion
of the rigid part behind the traveling hinge (ξ ≤ X ≤ L) ceases immediately upon arrival
of the plastic hinge, as indicated by Eq. (10.120b), namely the motion of the outer regions
stops at time T (X) which is given by
L−X
T (X) = for ξS ≤ X ≤ L (10.125b)
CQ
where ξS = L − LS and LS is the final length of the plastic zone

mC Q V0
LS = CQ TS = L 1 − exp − (10.126)
Q0
Now the transverse displacement can be obtained by integrating Eq. (10.124) and utilizing
the initial condition in Eq. (10.117a) to give

Q0 L CQ t CQ t
w(t) = V0 t − 1− ln 1 − −1 +1 for 0 ≤ X ≤ ξ
mC 2Q L L
(10.127)
This equation together with Eqs. (10.125a,b) predict the final permanent displacement
profile, as shown in Fig. 10.23(c)

V0 L Q0 mC Q V0
wf = Wf = 1+ exp − −1 for 0 ≤ X ≤ ξS
CQ mCQ V0 Q0
(10.128a)

V0 L X Q0 X X
wf = wf (X) = 1− − ln − 1 + 1 for ξS ≤ X ≤ L
CQ L mC Q V0 L L
(10.128b)
Differentiating wf (X) with respect to X, the shear strain distribution γ (X) = ∂w/∂X can
be obtained, as shown in Fig. 10.23(d)
γ (X) = 0 for 0 ≤ X ≤ ξS (10.129a)

Q0 X
γ (X) = −γS 1 + ln for ξS ≤ X ≤ L (10.129b)
mC Q V0 L
the absolute value of which increases with X up to its maximum value γS
V0
γS = (10.129c)
CQ
Note that Eqs. (10.129) can also be derived from Eqs. (10.119b), (10.124), and (10.125a),
showing that the kinematic condition across the traveling plastic hinge is satisfied.
On the other hand, in order to satisfy the dynamic condition across the traveling plastic
shear hinge, it can be seen from Eqs. (10.119a) and (10.124) that the shear force Q1 should
be the following function of t

CQ t CQ t
Q1 (t) = mC Q V0 + Q0 1 + ln 1 − = QS + Q0 ln 1 − (10.130a)
L L
where
QS = Q0 + mC Q V0 (10.130b)
is the maximum value of Q1 (t) at t = 0, since Eq. (10.130a) shows that Q1 (t) decreases
with t because the length of the plastic zone, lS = CQ t, is less than L.
The decrease of Q1 (t) with time means that the beam segment behind the plastic shear
hinge (ξ ≤ X ≤ L) is in a rigid-unloading state, which leads to a non-uniform distribution
of plastic shear strain described by Eq. (10.129b) as shown in Fig. 10.23(d).
The strain rate across a traveling plastic shear hinge (CQ > 0) is theoretically infinite when
[γ ]
= 0. Therefore, in order to estimate the strain rate in a beam from an engineering
viewpoint, the absolute value of average shear strain γav over the localized plastic zone
Ls is first defined as
$
L
L−LS γ (X) dX Wf
γav = = (10.131a)
LS LS
Consequently the average shear strain rate can be approximately defined as
γav Wf
γ̇av = = (10.131b)
TS LS TS
A further analysis is required to determine the bending moment distribution, to see whether
or not the foregoing theoretical analysis satisfies the associated yield condition. From
Eqs. (10.116b), (10.121)–(10.123), and (10.130), together with the boundary condition
for the bending moment at X = ξ = (L − CQ t), the bending moment distribution M(X)
is expressed as

Q0 CQ t X
M(X) = Q1 (L−X) = QS L 1+ ln 1− 1− for (L−CQ t) ≤ X ≤ L
QS L L
(10.132a)

Q0 L CQ t X2
M(X) = Q1 CQ t + 1− 1−
2 L (1−CQ t/L)2 L2

C Q t Q 0 CQ t CQ t 1 CQ t
= QS L + ln 1− + 1−
L QS L L 2 L

X2
× 1− , for 0 ≤ X ≤ (L−CQ t) (10.132b)
(1−CQ t/L)2 L2
According to Eq. (10.132b), the maximum bending moment with respect to spatial
coordinate X occurs at the mid-span (X = 0), as shown in Fig. 10.23(e).

Q0 L CQ t Q0 L Q0
Mx max = M(0, t) = Q1 CQ t + 1− = + Q1 − CQ t
2 L 2 2
(10.133)
which increases monotonously with t.
Equation (10.130) gives Q1 = Q0 when t = TS , so that the maximum bending moment

with respect to both the special coordinate X and the time coordinate t, MMAX , is

Q0 L C Q TS
MMAX = M(0, TS ) = 1+ (10.134a)
2 L
or, in a dimensionless form
MAX = ν(1 + L
M S ) (10.134b)
S , the dimensionless bending

where the following dimensionless final plastic zone length L

moment M, and the well-known dimensionless strength ratio ν have been introduced and
defined as
S = LS = CQ TS ,
L = M,
M ν=
Q0 L
(10.135)
L L M0 2M0
and M0 is the fully plastic bending moment at yield.
If a square yield condition relating Q and M is assumed, then it is evident from

Eq. (10.134b) that when ν(1 + L S ) ≥ 1 the maximum bending moment at the mid-
span (X = 0) satisfies the yield criterion (MMAX ≥ M0 ) and a new plastic bending hinge
is formed there. The problem becomes much more complicated. Therefore, the foregoing
theoretical analysis for the pure shear failure of beams is valid only for
S ) ≤ 0
ν(1 + L (10.136)
For a beam having relatively high yield strength but relatively weak strain-hardening
character, we have
mC Q V0 Gp γ S
= 1 (10.137)
Q0 Q0
and subsequently from Eq. (10.126), we have
S ∼ mC Q V0
L = 1 (10.138)
Q0
This means that the final length of a plastic zone LS is much shorter than L, the one-half
span length of the beam. In this circumstance, most of the equations deduced earlier
can be simplified to facilitate engineering applications. In fact, Eqs. (10.125a), (10.128),

(10.129), and (10.131) now are respectively reduced to:
mLV 0
TS ∼
= (10.139)
Q0
mV 20 L
wf = Wf ∼
= for 0 ≤ X ≤ (L − LS ) (10.140a)
2Q0
/ 2 0
V L X Q X
wf ∼
0 0
= 1− − 1− for (L − LS ) ≤ X ≤ L (10.140b)
CQ L 2mCQ V0 L

Q0 X
γ (X) ∼
= −γS 1 − 1− for (L − LS ) ≤ X ≤ L (10.141)
mCQ V0 L
Q0
γ̇av ∼
= (10.142)
2mCQ L
Equations (10.139) and (10.140) are identical to the well-known Nonaka’s solution for the
rigid-perfectly plastic beam (Nonaka, 1967). Thus, the Nonaka’s solution can be regarded
as an approximate solution of the present analysis when L S ∼
= 0.
Finally, it can be seen from the above analyses that there are at least three possible modes
of shear failure for the beam subjected to impulsive loading:
(a) Excess transverse deflection failure. It occurs when the transverse displacement wf
predicted by Eq. (10.128) exceeds the critical transverse displacement wc , which is
pre-determined from the design requirements, namely
wf ≥ wc (10.143)
(b) Excess transverse shear strain failure. It occurs when the maximum shear strain
γS predicted by Eq. (10.129) exceeds the critical shear strain γc , which is pre-
determined from the design requirements or the fracture shear strain of material,
namely
γS ≥ γc (10.144)
Note that the critical or fracture shear strain γc of a material is generally dependent
on the shear strain-rate γ̇ .
(c) Adiabatic shear failure. This mode of failure is related to the balance between the
stress drop due to thermal softening by adiabatic heating (∂τ/∂T < 0, where
τ is the shear stress and T the temperature) and the stress rise due to strain
hardening (∂τ/∂γ > 0) and strain-rate hardening (∂τ/∂ γ̇ > 0) of a mate-
rial, namely the so-called thermo-viscoplastic constitutive instability of a material
(Wang et al., 1987, 1988). Macroscopically, for a given environmental temperature,
the critical thermo-viscoplastic instability criterion can be described, for example,

by the following equation (Wang, 1986; Wang and Bao He-sheng, 1991)

γ̇ αβG1 τ0
f (γ , γ̇ ) = 1 + g A− +γ −1=0 (10.145)
γ̇0 T0 ρC G1
where G1 , g, and α characterize the strain-hardening, strain-rate hardening, and

thermal-softening properties, respectively, τ0 , γ̇0 , and T0 are the characteristic
stress, strain-rate, and temperature for a quasi-static test, respectively; ρ and C
are the density and the specific heat of the material, respectively; β is the Taylor–
Quinney coefficient, namely the fraction of viscoplastic work converted to heat
(β = 0.9 ∼ 1.0 in general). The A in the equation is a material parameter which
characterizes a certain state of the adiabatic shearing process, for example, the ini-
tiation of a deformed shear band, the initiation of a transformed shear band, or the
initiation of adiabatic fracture, etc.
The mode which dominates the shear failure of a practical beam will generally depend on
which critical condition among Eqs. (10.143)–(10.145) is first satisfied.
For the readers who are interested in the studies on the dynamic elastic–plastic problems of
beams, more detailed analyses and examples can be found from the early review literatures
or books written by, for example, Abramson et al. (1958) and Rakhmatulin and Demyanov
(1961); and found from the recent book written by Jones (1989).
CHAPTER 11
General Theory for Linear Elastic Waves
11.1 Linear Elastic Waves in Infinite Media
In the general three-dimensional cases, when body forces are not considered, the motion
equations (momentum conservation) are expressed as
⎫
∂vX ∂σXX ∂σYX ∂σZX ⎪⎪
ρ0 = + + ⎪
∂t ∂X ∂Y ∂Z ⎪ ⎪
⎪
⎬
∂vY ∂σXY ∂σYY ∂σZY
ρ0 = + + (11.1)
∂t ∂X ∂Y ∂Z ⎪ ⎪
⎪
∂σZZ ⎪⎪
ρ0
∂vZ
=
∂σXZ
+
∂σYZ
+ ⎪
⎭
∂t ∂X ∂Y ∂Z
For homogeneous, isotropic, and linearly elastic media, substituting the Hooke’s law in
the Lame’s form [Eq. (7.10) into Eq. (11.1)] to eliminate the stress components, we obtain
⎫
∂vX ∂ ∂ ∂ ⎪
ρ0 = (λ∆ + 2µεXX ) + (2µεXY ) + (2µεZZ )⎪
⎪
⎪
∂t ∂X ∂Y ∂Z ⎪
⎪
⎬
∂vY ∂ ∂ ∂
ρ0 = (2µεXY ) + (λ∆ + 2µεYY ) + (2µεYZ ) (11.2)
∂t ∂X ∂Y ∂Z ⎪
⎪
⎪
⎪
⎪
(λ∆ + 2µεZZ )⎪
∂vZ ∂ ∂ ∂
ρ0 = (2µεXZ ) + (2µεYZ ) + ⎭
∂t ∂X ∂Y ∂Z
Further, expressing strain components by displacement components uX, uY, and uZ
[Eq. (7.9)], Eq. (11.1) can be re-written as
2 ⎫
∂ 2 uX ∂∆ ∂ ∂2 ∂2 ⎪
ρ0 2 = (λ + µ) +µ + 2 + 2 uX⎪ ⎪
⎪
∂t ∂X ∂X 2 ∂Y ∂Z ⎪
⎪
2 2 2 2 ⎪
⎬
∂ uY ∂∆ ∂ ∂ ∂
ρ0 2 = (λ + µ) +µ + + u Y (11.3)
∂t ∂Y ∂X 2 ∂Y 2 ∂Z 2 ⎪
⎪
⎪
⎪
∂ 2 uZ ∂∆ ∂2 ∂2 ∂2 ⎪
⎪
ρ0 2 = (λ + µ) +µ + + u Z
⎪
⎭
∂t ∂Z ∂X 2 ∂Y 2 ∂Z 2
all symbols in the above equations represent the same variables as those in Chapter 7.
453
Differentiating the three equations in Eq. (11.2) with respect to X, Y , and Z respectively,
and adding together, further utilizing the following continuity conditions
∂vX ∂εXX ∂vY ∂εYY ∂vZ ∂εZZ

= , = , = (11.4)
∂X ∂t ∂Y ∂t ∂Z ∂t
then we obtain
2
∂ 2∆ ∂ ∂2 ∂2
ρ0 = (λ + 2µ) + + ∆ (11.5)
∂t 2 ∂X 2 ∂Y 2 ∂Z 2
where ∆ = εXX + εYY + εZZ represents the volume dilatation. Equation (11.5) is a
linear hyperbolic partial differential equation with respect to ∆, and it means that volume
dilatation ∆ propagates with a wave velocity CLe

λ + 2µ
CLe =
ρ0
Therefore, this wave is sometimes called dilatation wave. Comparing with Eq. (7.19),
it can be seen that the velocity of the dilatation wave is just the velocity of one-dimensional
strain elastic longitudinal wave. Note that when the dilatation wave propagates, it is
actually accompanied with distortion deformation.
Differentiating the third and second equations of Eq. (11.2) with respect to Y and Z,
respectively, and subtracting the latter from the former to eliminate ∆, further utilizing
the following continuity conditions

∂vZ ∂vY ∂ ∂uZ ∂uY ∂ωX
− = − =2
∂Y ∂Z ∂t ∂Y ∂Z ∂t
then we obtain
2
∂ 2 ωX ∂ ∂2 ∂2
ρ0 =µ + 2 + 2 ωX (11.6)
∂t 2 ∂X 2 ∂Y ∂Z
where ωX denotes the rotation with respect to X-axis

1 ∂uZ ∂uY
ωX = −
2 ∂Y ∂Z
The other two components of the rotation vector ω = (1/2)rotu are ωY and ωZ

1 ∂uX ∂uZ
ωY = −
2 ∂Z ∂X

1 ∂uY ∂uX
ωZ = −
2 ∂X ∂Y
General Theory for Linear Elastic Waves 455
Analogous to the deduction of Eq. (11.6), from the first and second equations, and from the
third and first equations of Eq. (11.2), the linear hyperbolic partial differential equations
with respect to ωY and ωZ can be obtained.
2
∂ 2 ωY ∂ ∂2 ∂2
ρ0 = µ + + ωY
∂t 2 ∂X 2 ∂Y 2 ∂Z 2
2
∂ 2 ωZ ∂ ∂2 ∂2
ρ0 =µ + 2 + 2 ωZ
∂t 2 ∂X 2 ∂Y ∂Z
This means that rotation ω propagates with the velocity

9
µ G
CT = =
ρ0 ρ0
Since the propagation of rotation disturbance induces pure distortion or shear deformation,
it is usually named as distortion wave or shear wave. Comparing the above equation for
CT with Eq. (2.78), it can be seen that the velocity of distortion wave is just the velocity
of torsion wave [Eq. (2.78)], which has been discussed in Chapter 2.
It is easy to prove that a dilatation wave is an irrotational wave (ω = 0), while a

distortion wave is an isometric wave (∆ = 0). In fact, the first equation of Eq. (11.3) can
be re-written as
2
∂ 2 uX ∂ ∂2 ∂2 ∂ωZ ∂ωY
ρ0 2 = (λ + 2µ) + + u X + 2(λ + µ) −
∂t ∂X 2 ∂Y 2 ∂Z 2 ∂Y ∂Z
The second and the third equations of Eq. (11.3) can also be re-written in a similar form,
and then it can be shown that the displacement in the irrotational case up satisfies the
following equation
2
∂ 2 up ∂ ∂2 ∂2 1
ρ0 = (λ + 2µ) + + up , ωp = rotup (11.7)
∂t 2 ∂X 2 ∂Y 2 ∂Z 2 2
which indicates that up propagates with velocity CLe . Meanwhile, from Eq. (11.3), it can
be directly deduced that the displacement in the isometric case us satisfies
2
∂ 2 us ∂ ∂2 ∂2
ρ0 2 = µ + 2 + 2 us , ∆s = divus = 0 (11.8)
∂t ∂X 2 ∂Y ∂Z
which indicates that us propagates with wave velocity CT .
So far, it has been established that an arbitrary displacement can be divided into two parts,
an irrotational part and an isometric part.
u = up + us
(11.9)
rotup = 0, divus = 0
Each part satisfies Eqs. (11.7) and (11.8) respectively. Substituting Eq. (11.9) into
Eq. (11.3) and re-writing to a vector form, we have
∂ 2u ∂ 2 up ∂ 2 us
= +
∂t 2 ∂t 2 ∂t 2
2
λ + 2µ ∂ ∂2 ∂2 µ ∂2 ∂2 ∂2
= + + u p + + + us (11.10)
ρ0 ∂X 2 ∂Y 2 ∂Z 2 ρ0 ∂X 2 ∂Y 2 ∂Z 2
This means that in homogeneous, isotropic, and linearly elastic media, the propagation of
an arbitrary displacement disturbance can generally be divided into an irrotational wave
and an isometric wave, irrelatively propagating with velocity CLe and CT , respectively.
Because these waves propagate in the interior of a body, independent of the boundary
effects, they are named as body waves, to distinguish from the surface waves, which are
generated and propagate on surfaces, and will be discussed in Section 11.3.
It is established in vector analysis that there must be a potential in an irrotational field.

In other words, up must be a gradient of a scalar ϕ
up = gradϕ (11.11)
ϕ is called the scalar potential of displacement; while a nondivergent field must be a

solenoidal field. In other words, us must be a curl of a vector ψ.
us = rotψ (11.12)
ψ is called the vector potential of displacement. Then Eq. (11.9) can be re-written as
u = gradϕ + rotψ (11.13a)
Equations (11.7) and (11.8) can be re-written in the following wave equations with respect
to ϕ and ψ, respectively.

∂ 2ϕ λ + 2µ ∂ 2 ∂2 ∂2
− + + ϕ=0 (11.13b)
∂t 2 ρ0 ∂X 2 ∂Y 2 ∂Z 2

∂ 2ψ µ ∂2 ∂2 ∂2
− + 2 + 2 ψ=0 (11.13c)
∂t 2 ρ0 ∂X 2 ∂Y ∂Z
Thus the studies on irrotational waves and isometric waves are transferred to the studies
on the displacement functions ϕ and ψ, which satisfy Eq. (11.13).
It is easy to prove that an irrotational wave is a longitudinal wave while an isometric

wave is a transverse wave. Now let us consider an arbitrary plane wave propagating in
homogeneous, isotropic infinite media with velocity C. Without loss of generality, let this
plane wave propagate along X-axis, then displacements uX , uY , and uZ are functions
of ξ = X−Ct:
uX = uX (X − Ct), uY = uY (X − Ct), uZ = uZ (X − Ct)
Substituting these functions into Eq. (11.3), we obtain
ρ0 C 2 uX = (λ + 2µ)uX
ρ0 C 2 uY = µuY
ρ0 C 2 uZ = µuZ
where ( ) denotes a second order differential with respect to ξ . To obtain at least one
nontrivial solution of uX , uY , and uZ , it is possible only under one of the two conditions
below,
λ + 2µ
C2 = = CL2 , uY = u = 0
ρ0
which means that there is disturbance propagating in the X-direction only; or

µ
C2 = = CT2 , uX = 0
ρ0
which means that there is disturbance propagating in the Y- or Z-directions only. These
results on the one hand prove that an arbitrary plane wave propagating in homogeneous,
isotropic infinite media is either an irrotational wave propagating with velocity CLe or an
isometric wave propagating with velocity CT , in accordance with Eq. (11.10), on the other
hand, they prove that an irrotational wave is a longitudinal wave while an isometric wave
is a transverse wave.
As (λ + 2µ) > µ, a longitudinal wave propagates faster than a transverse one. During
seismic, longitudinal waves are always observed first, followed by the transverse waves.
Because of this, these two kinds of waves are usually named as P(primary)-waves and
S(secondary)-waves. The S-waves are further categorized into SH-waves that propagate
parallel to the free surface (the horizon) and SV-waves that propagate perpendicular to the
free surface.
11.2 Oblique-Incidence, Reflection, and Transmission of Elastic Plane Waves
Although irrotational waves and isometric waves are unrelated during their propagation,
they are often correlated when they are reflected or refracted at the body surface or at the
interface between two media. Only if the direction of an incident wave is perpendicular
to the surface or interface, namely in the case of a normal incidence, then these two kinds
of waves are uncoupled and thus can be solved according to the principles discussed in
Section 3.5. Otherwise, during an oblique incidence, in order to satisfy the given boundary
conditions, whatever be the incident wave i.e. an irrotational wave only or an isometric
Y
Reflective irrotational wave P2
u
2
u4
a2 Free surface
Reflective isometric wave SV2
b2
a1 X
Incident irrotational wave P1

u1
Fig. 11.1. Reflection of an irrotational plane wave obliquely incident at a free surface.
wave only, in general, both an irrotational wave and an isometric wave are simultaneously
reflected and refracted. This is called mode coupling.
As an example of mode coupling, now let us consider an irrotational plane wave which is
obliquely incident onto a free surface. Setting the free surface as Y –Z coordinate plane
as shown in Fig. 11.1, assuming that the propagating direction of the incident irrotational
wave P1 is on the X–Y plane, then the displacement components uX , uY are independent
of Z and uZ = 0. Regarding the situation of uZ
= 0, namely, the reflection of SH-wave
will be discussed in the ensuing sections. Now consider the general possible situation
of an oblique-incidence, namely both an irrotational wave and an isometric wave are
simultaneously reflected. Denote the angle between the direction of P1 and X-axis by α 1 ,
the angle between the direction of the reflected irrotational wave P2 and X-axis by α 2 , the
angle between the direction of the reflected isometric wave SV2 and X-axis by β 2 . They are
named as the incidence angle, the reflective angle of the irrotational wave, and the reflective
angle of the isometric wave, respectively. Thereupon, if we regard the concerned waves
as harmonic waves, then the displacements u1 , u2 of the longitudinal wave P1 , P2 along
their wave-vector direction k1 , k2 , respectively, and the displacement u4 of the transverse
wave SV2 along the normal of its wave vector k4 can be expressed respectively as
⎫
u1 = A1 exp{i(Xk1 cos α1 + Y k1 sin α1 − ω1 t)} ⎬
⎪
u2 = A2 exp{i(−Xk2 cos α2 + Y k2 sin α2 − ω2 t)} (11.14)
⎪
⎭
u4 = A4 exp{i(−Xk4 cos β2 + Y k4 sin β2 − ω4 t)}
where A is the amplitude, ω the circular frequency, k the wave number, the subscripts used
here are identical with the subscripts of u mentioned above. The relationship between ω, k,
and wave velocity is
ω1 ω2 ω4
= = CL , = CT (11.15)
k1 k2 k4
Obviously, the components along X and Y directions of u1 , u2 , and u4 are respectively:
⎫
u1X = u1 cos α1 , u1Y = u1 sin α1 ⎪
⎬
u2X = −u2 cos α2 , u2Y = u2 sin α2 (11.16)
⎪
u4X = u4 sin β2 , u4Y = u4 cos β2⎭
The conditions that stress components σXX , σYY , σXZ on the free surface (X = 0) are all
zero, should be satisfied. Following Hooke’ law [Eq. (7.10)], also recalling that in the
current situation uX , uY are independent of Z and uZ = 0, then the uX and uY on the free
surface should satisfy the conditions below:
⎫
∂uX ∂uY ⎪
σXX |X=0 = (λ + 2µ) +λ = 0 ⎪
⎪
∂X ∂Y X=0 ⎬
⎪
(11.17)
∂uY ∂uX ⎪
σXY |X=0 = µ + = 0⎪⎭
∂X ∂Y X=0
where uX and uY include all the contributions of incident waves and reflected waves,
namely,
uX = u1X + u2X + u4X

uY = u1Y + u2Y + u4Y
Substituting Eq. (9.16) and Eq. (9.14) in Eq. (9.17), we obtain

k1 λ + 2µ cos2 α1 · A1 exp{i(Y k1 sin α1 − ω1 t)} + k2 λ + 2µ cos2 α2
· A2 exp{i(Y k2 sin α2 − ω2 t)} − 2µk4 sin β2 cos β2
· A4 exp{i(Y k4 sin β2 − ω4 t)} = 0
2k1 sin α1 cos α1 A1 exp{i(Y k1 sin α1 − ω1 t)} − 2k2 sin α2 cos α2

· A2 exp{i(Y k2 sin α2 − ω2 t)} − k4 cos2 β2 − sin2 β2
· A4 exp{i(Y k4 sin β2 − ω4 t)} = 0 (11.18)
In order for these boundary conditions to be satisfied for any Y and t, there must be:
k1 sin α1 = k2 sin α2 = k4 sin β2

ω1 = ω2 = ω4 = ω (11.19)
Also considering Eq. (11.15), we obtain the following law of equal in visual velocities,
CL CL CL
= = =C (11.20a)
sin α1 sin α2 sin β2
where C is the so-called visual velocity. Equation (11.20a) is the Snell’s law in optics, and
can equivalently be written as
α1 = α2

sin α1 kT CL λ + 2µ 2 (1 − ν)
= = = = ≡κ (11.20b)
sin β2 kL CT µ (1 − 2ν)
where kT = k4 and kL = k1 = k2 are the wave numbers of SV-wave and P -wave respec-
tively. Equation (11.20b) indicates that the reflective angle α 2 of a reflected irrotational
wave equals the incidence angle α 1 ; the reflective angle β 2 of a reflected isometric wave
is less than the incidence angle α 1 . In general β 2 > 0, except in the case of normal inci-
dence where α1 = 0 and then β2 = 0. Therefore, when an irrotational wave is obliquely
incident on a free surface, in general, both an irrotational wave and an isometric wave are
reflected and coupled together. Readers are suggested to prove by themselves that if only
an irrotational wave is reflected, then in case the first equation of the boundary conditions
[Eq. (11.17)], namely σXX |X=0 = 0 (and consequently α1 = α2 ) is satisfied, the second
equation is no longer satisfied; in order to satisfy simultaneously the second equation,
an isometric wave must be simultaneously reflected.
Substituting the above relationship into Eq. (11.18), we obtain

kL λ + 2µ cos2 α1 (A1 + A2 ) − 2µkT A4 sin β2 cos β2 = 0

2kL sin α1 cos α1 (A1 − A2 ) − kT A4 cos2 β2 − sin2 β2 = 0
Moreover, utilizing Eq. (11.20b), there is
λ sin2 α1 sin2 α1
+ 2 cos2 α1 = − 2 sin 2
α 1 = · cos 2β2
µ sin2 β2 sin2 β2
By operating the above two equations, we obtain
A2 A4
(CL cos 2β2 ) − (CT sin 2β2 ) = −CL cos 2β2
A1 A1
A2 A4
(CT sin 2α1 ) + (CL sin 2β2 ) = CT sin 2α1
A1 A1
They are the algebraic equations set with respect to the displacement reflection coefficients
A2 /A1 and A4 /A1 . On solving the equations, we obtain

−CL cos 2β2 −CT sin 2β2

A2 CT sin 2α1 CL cos 2β2
=
A1 CL cos 2β2 −CT sin 2β2

CT sin 2α1 CL cos 2β2

CL cos 2β2 −CL cos 2β2

A4 CT sin 2α1 CT sin 2α1
=
A1 CL cos 2β2 −CT sin 2β2

CT sin 2α1 CL cos 2β2
Expanding these equations, we have
A2 sin 2α1 sin 2β2 − κ 2 cos2 2β2 tgβ2 tg2 2β2 − tgα1
= = (11.21a)
A1 sin 2α1 sin 2β2 + κ 2 cos2 2β2 tgβ2 tg2 2β2 + tgα1
A4 2κ sin 2α1 cos 2β2 4 sin β2 cos 2β2 cos α1
= = (11.21b)
A1 sin 2α1 sin 2β2 + x 2 cos2 2β2 tgβ2 sin2 2β2 ctgα1 + cos2 2β2
It can be seen from Eq. (11.21b) that the reflection coefficient is a function of Poisson’s
ratio ν and the incident angle α1 . When ν = 1/3 (then CL /CT = sin α1 / sin β2 = 2),
how the reflection coefficients A2 /A1 and A4 /A1 varies with the incident angle α 1
calculated by Eq. (11.21) is shown in Fig. 11.2 (see Kolsky, 1953).
It can be seen from Fig. 11.2 that (a) when the incident angle approaches 48◦ , the amplitude
of the reflective SV-wave approaches its maximum, which is even larger than the amplitude
1.1
1.0
0.9
0.8
0.7
A1 , A1
A2 A4
0.6
0.5
0.4
0.3
0.2 A2 /A1
0.1 A4 /A1
0
20 40 60 80
Incident angle a1 (degree)
Fig. 11.2. Reflection coefficients A2 /A1 and A4 /A1 vary with the incident angle α 1 .
of the incident irrotational wave (A4 /A1 ≈ 1.04); (b) when the incident angle approaches
65◦ , the amplitude of the reflective P-wave approaches its maximum (A2 /A1 ≈ −0.38);
(c) in the case of normal incidence (α 1 = 0), only a P-wave (A2 /A1 = −1) but no SV-wave
(A4 = 0) is reflected, then this problem is the same as that discussed in Section 3.3;
(d) in the case of grazing incidence, namely α1 = π /2, there are A4 = 0 and A2 /A1 = −1
again. But Eq. (11.21) shows u1 = −u2 , which indicates a trivial solution, namely waves
disappear. However, it should be noticed that such solutions are obtained under the plane
wave assumption, namely both incidence waves and reflecting waves are plane waves.
If the reflected waves are not limited to homogeneous plane waves only, then nontrivial
solutions may exist during the grazing incidence of P-waves at a free surface. In such
cases, a nonhomogeneous wave similar to a surface wave will be formed which will be
discussed in Section 11.3.
When the oblique incidence of P-waves or SV-waves is discussed in terms of the scalar
potential defined by Eq. (11.7) and the vector potential defined by Eq. (11.8), under the
current condition of uz = (∂uX /∂Z) = (∂uY /∂Z) = 0, the displacement components
upX , upY of P-wave and the displacement components usX , usY of SV-wave can be
expressed respectively by
∂ϕ ∂ϕ
upX = , upY =
∂X ∂Y
(11.22)
∂ψZ ∂ψZ
usX = , usY = −
∂Y ∂X
then the problem is reduced to the analysis on ϕ and ψ Z , and the subscript Z of ψ Z , can
be temporarily neglected here. If the u1 , u2 , and u4 in Eq. (11.14) are replaced by ϕ 1 ,
ϕ 2, and ψ 4 , respectively, and the amplitudes of the corresponding harmonic waves are
expressed by a1 , a2 , and a4 , namely
ϕ1 = a1 exp{i(Xk1 cos α1 + Y k1 sin α1 − ω1 t)}

ϕ2 = a2 exp{i(−Xk2 cos α2 + Y k2 sin α2 − ω2 t)} (11.23)
ψ4 = a4 exp{i(−Xk4 cos β2 + Y k4 sin β2 − ω4 t)}
then the equal visual velocity law, Eq. (11.20), can be deduced again, and the displacement
reflection coefficients a2 /a1 and a4 /a1 can also be deduced in the same way as Eq. (11.21).
They are related with A2 /A1 and A4 /A1 as
u2X u2Y a2 u4X u4Y a4
=− =− , = = −κ
u1X u1Y a1 u1X u1Y a1
(11.24)
A2 a2 A4 a4
= , = −κ
A1 a1 A1 a1
Figure 11.3 shows the relationship between a2 /a1 (and consequently A2 /A1 ) and incident
angle α1 with various ν (Arenberg, 1948). When ν > 0.26, A2 /A1 is always negative; when
ν < 0.26, there exist such incident angles that A2 /A1 = 0, which means that there is no
P-wave but only SV-wave reflected. In other words, a conversion of wave mode (P → SV )
1.0
v =0
0.8
0.14
0.6
0.19
0.4
0.22
0.2
0.25
A2
A1
0 0.28
−0.2 0.31
−0.4 0.35
−0.6 0.38
−0.8 0.455
−1.0
0 10 20 30 40 50 60 70 80 90
Incident angle a1 (degree)
Fig. 11.3. Relationship between A2 /A1 and incident angle α1 .
occurs through the wave reflection. This phenomenon is called total mode conversion or
polarization conversion. The corresponding incident angle is called mode conversion
angle, denoted as α (P → SV ), which is a function of ν. For example, if ν = 1/4, then
the wave mode conversion of P → SV occurs when α1 = 60◦ and 77.2◦ .
In the same manner, it is easy to prove that when a SV1 wave propagating in the direction
parallel to X–Y plane is obliquely incident onto the free surface (Y –Z plane), in general
two coupled waves are reflected, a reflected isometric wave P2 and a reflected irrotational
wave SV2 , as shown in Fig 11.4.
Reflective irrotational wave P2 Y
Free surface
Reflective isometric wave SV2 a
b2
b1 X
Incident isometric wave SV1
Fig. 11.4. Reflection of an isometric plane wave obliquely incident at a free surface.
Moreover, Snell’s law is still valid
CT CT CL
= =
sin β1 sin β2 sin α2
and the displacement reflection coefficients are:
A4 sin β1 sin 2α2 − κ 2 cos2 2β1

=
A3 sin 2β1 sin 2α2 + κ 2 cos2 2β1
(11.25)
A2 2κ sin 2β1 cos 2β1
=
A3 sin 2β1 sin 2α2 + x 2 cos2 2β1
where A3 is the displacement amplitude of SV-wave, β 1 the incident angle, other symbols
denote the same variables as mentioned previously. In the same way, when analyzing with
respect to ϕ and ψ, similar to Eq. (11.24), we obtain
u2X u2Y 1 a2 u4X u4Y a4

= = , =− =
u3X u3Y κ a3 u3X u3Y a3
(11.26)
A2 1 a2 A4 a4
= , = ,
A3 κ a3 A3 a3
Figure 11.5 shows the relationship between a4 /a3 (and consequently A4 /A3 ) and incident
angle β 1 with various ν (Arenberg, 1948).
As can be seen from Fig. 11.5 that (a) in the cases of both normal incidence (β1 = 0)
and 45◦ angle oblique incidence (β1 = π /4), no P-wave but only SV-wave is reflected;
1.0
v =0
0.8
0.14
0.6
0.19
0.4
0.22
0.2
0.25
A4
A3
0
0.28
−0.2 0.31
−0.4 0.35
−0.6 0.38
−0.8
0.455
−1.0
0 5 10 15 20 25 30 35 40 45
Incident angle b1 (degree)
Fig. 11.5. Relationship between A4 /A3 and incident angle β 1 .

0.50
0.40
bcr
Poisson ratio
0.30
0.20 a
(P SV )
b
0.10
(SV P)
0 30 40 60 80
Incident angle
Fig. 11.6. The relationship between βcr , α (P → SV ), and Poisson’s ratio ν.
(b) if ν < 0.26 (except ν = 0), there exist such incident angles that A4 /A2 = 0, which
indicates that mode conversion of SV → P occurs; the corresponding angle is denoted
as β(SV → P ); (c) since κ = (sin α/ sin β) = (CL /CT ) > 1, there exists such a critical
angle βcr = sin−1 (1/κ) so that if the incident angle β1 > βc , then the value of reflective
angle α2 is no longer a real number. This phenomenon, analogous to optics, is called the
total reflection.
The critical angle of total reflection β cr , similar to the mode conversion angle α(P →SV),
is a function of Poisson’s ratio ν. The relationship between βcr and ν is shown in Fig. 11.6
(Arenberg, 1948).
So far we have discussed the coupling reflection of P-wave and SV-wave when they
are obliquely incident onto the free surface. As for the SH-wave, namely the isometric
wave of which the direction of displacement disturbances are parallel to the free surface
(uX = uY = 0, uZ
= 0), it can be proved by a similar method used previously i.e. when
an SH-wave is obliquely incident onto a free surface, only an equal-amplitude SH-wave,
but no other mode wave, is reflected, and the reflective angle equals the incident angle.
Analogous to this, the problems of reflection and refraction can be treated in the same way
as that when an elastic wave is obliquely incident on an interface between two different
materials A and B. According to the continuity requirement at the interface for the normal
stress, shear stress, normal displacement and shear displacement, it can be proved that
whatever be the obliquely incident wave i.e. an irrotational wave or an isometric wave, in
general, four types of waves are generated at the interface, namely a reflected irrotational
wave, a reflected isometric wave, a refracted irrotational wave, and a refracted isometric
wave, and Snell’s law is still valid. For instance, if the obliquely incident wave is an
Y Y
P3 P3
P2 P2
a2 a3 a2 SV3
SV2 SV3 a3
SV2 b2 b3
b2 b3
a1 X b1 X
SV1
P1
Material A Material B Material A Material B
(a) (b)
Fig. 11.7. Reflection and refraction when an elastic wave is obliquely incident on an interface
between two media.
irrotational wave (P -wave), as shown in Fig. 11.7(a), then we have
sin α1 sin α2 sin β2 sin α3 sin β3

= = = = (11.27)
(CL )A (CL )A (CT )A (CL )B (CT )B
while if the obliquely incident wave is an isometric wave (SV-wave), as shown in

Fig. 11.7(b), then we have
sin β1 sin β2 sin α2 sin β3 sin α3

= = = = (11.28)
(CT )A (CT )A (CL )A (CL )A (CL )B
where the symbols used to denote the related angles are shown in the figure. Subscripts A
and B denote the parameters for material A and B respectively. Similar to Eqs. (11.20) or
(11.23), the relationship between the reflection coefficient and the material constants ρ,
λ, µ, as well as the incident angle can be determined in the same way as the relationship
between the deflection coefficient and the material constants ρ, λ, µ, as well as the incident
angle is determined (see Miklowitz, 1978).
It should be noted that the interactions between the reflected unloading waves and the
rearward portion (unloading portion) of the incident pressure pulse could induce tensile
stress, which may result in the dynamic fracture called spalling, similar to that in the case
of normal incidence as discussed previously in Section 3.8. However, the direction of the
maximum tension stress is now related to the incident angle. The formation of spalling due
to the reflected unloading wave when a pressure pulse (P -wave) is obliquely incident onto
a free surface of an isotropic media is schematically shown in Fig. 11.8 (Rinehart, 1975).
ion
p ort e
v
Fr ard wa e)
on
to e arw ident wav
(U f refl
R nc
i ing
nlo ec of load
ad tive n
ing w (U
) ave ve
s t wa
e n )
cid ve
o f in g wa
t din
on
Micro-fractures Fr (Loa
have formed
Micro-fractures Micro-fractures Free surface
have coalesced begin to form
Fig. 11.8. Schematic of the formation of spalling due to the reflection on a free surface of an
obliquely incident P -wave.
Obviously, the spalling develops with the reflected process of waves, and the orientation
of micro-crack depends on the incident angle.
11.3 Elastic Surface Waves
As discussed previously, when an SV-wave is obliquely incident onto a free surface, if

the incident angle β is larger than the critical angle βcr = sin−1 (1/ κ), then the ordinary
reflection following Snell’s law does not occur. As sin β > 1/κ, and κ = CL /CT > 1,
there is sin α = κ sin β > 1, which indicates that α is not a real number, and consequently
leads to a so-called complex reflection. Such nonordinary reflected P -wave which is a
nonuniform wave generated by grazing incidence at a free surface is classified as a kind
of surface wave, and will be discussed below.
In the coordinate system shown in Fig. 11.4, if the wave reflection is discussed in terms
of the scalar potential ϕ, the reflected P -wave can be expressed by
ϕ2 = A exp{i(−Xk L cos α + Yk L sin α − ωt)} = A exp{ik S (Y − Xctgα − Cs t)}

(11.29)
where we have introduced
ω CL
kS = kL sin α, CS = = (11.30)
kS sin α
As sin α > 1, so
CS < CL (11.31)
On the other side, using the relations of triangle function, from Eqs. (11.30) and (11.31)
we obtain
2
1 CS
ctg α =
2
−1= −1<0
sin2 α CL
namely, ctgα must be an imaginary number. Introducing a positive real number

2
CS
rS = 1−
CL
then ctgα can be written as
ctgα = ±ir S
It is noted that if ctgα = −ir S , then it would be physically meaningless and moreover it
cannot satisfy the convergence requirement that the wave amplitude should not approach
infinity, at an infinite distance from the free surface. Therefore, we only have to take
ctgα = +ir S , and then Eq. (11.29) is re-written as
ϕ2 = A exp(rS kS X)exp(ik S (Y − CS t)) = A exp(rS kS X)exp(i(kS Y − ωt)) (11.32)
which indicates that the complex reflected P -wave is such a nonuniform wave, which
propagates with velocity CS along the free surface (namely the positive Y -direction in
Fig. 11.4), while its amplitude attenuates exponentially with the distance away from the
free surface (namely the negative X-direction in Fig. 11.4), and its wave velocity CS is
less than the wave velocity CL of an ordinary reflected P -wave (refer to Eq. (11.31)).
It can also be proved that when a P -wave is grazing along a free surface (grazing incidence),
the reflected P-wave is a nonuniform wave propagating along the free surface, while the
reflected SV-wave is still an ordinary reflected wave. But when an SV-wave is grazing
along a free surface, both the reflected SV-wave and the reflected P -wave are nonuniform
waves propagating along the free surface.
The formation of a nonuniform wave can be interpreted in the following way. Consider
a semi-infinite media, the free surface of which is now chosen as the Y –Z plane of
coordinate system. A compressive plane wave (P -wave) propagates along the Y direction
with wave velocity CL , as shown in Fig. 11.9 (Rinehart, 1975). The media through which
the compressive plane wave passes are under compression. But the media at the free surface
dilate transversely due to the Poisson’s effect, namely a dilatation wave is correspondingly
exited when the compressive plane wave arrives. In other words, each surface-point swept
by the compressive plane wave can be regarded as a source of a new dilatation wave, which
is a cylindrical dilatation wave propagating with the velocity CL . For example, when a
compressive plane wave successively passes through the points M0 , M1 , M2 , M3 . . . in
the figure, a series of cylindrical dilatation waves are generated at these points. When
the compressive plane wave front propagates from M0 N0 to AB, the series of cylindrical
dilatation waves arrive successively at the positions of arc AD0 , AD1 , AD2 , AD3 . . . in
X M0 M1 M2 M3 A
Y
b
D3
G
SV (reflected)
D2
D1
P (Incident)
P (reflected)
D0
N0
Fig. 11.9. Reflection of nonuniform P -wave due to a grazing incidence of a plane P -wave.
the figure. Because they propagate out from different sources at different times, no envelope
surface with respect to those arcs exists, and thus no real wave front can be formed.
Moreover, considering that the cylindrical dilatation waves attenuate with the propagating
distance due to the cylindrical dispersion effect (see Section 8.2), it can be understood that
the particles at the left side of arc AD0 move not only longitudinally (in Y -direction) but also
transversely (in X-direction), and such motion of particles are nonuniformly distributed
in coordinate X direction. The plane AD 0 in Fig. 11.9, namely the plane having 45◦ with
the free surface, is obviously the trajectories of the particles having maximum transverse
particle velocity. On the other hand, the particles in the area between the arc AD0 and the
plane wave front AB only move longitudinally. As for the ordinary reflective SV wave,
due to the existence of free surface, the position of its wave front is shown as AG in the
figure, when the incident plane P wave propagates from M0 N0 to AB. Thus, Fig. 11.9
schematically provides a physical image to illustrate how the ordinary plane SV-wave and
the nonuniform P -wave (nonplane wave) are reflected when a plane P -wave is grazing
along a free surface (namely in the grazing incidence).
Analogically, it can also be understood that a nonuniform SV wave can be formed along a
free surface.
In the same way, under certain conditions, a nonuniform P -wave and a nonuniform
SV-wave can be coupled to propagate along a free surface with an identical velocity.
This sort of surface waves was first discovered and depicted by Rayleigh (1887), hence
called Rayleigh surface wave. It plays an important role in the research of seismic waves.
Similar to Eq. (11.29), for a couple of P -wave and SV-wave propagating at a same wave
velocity CR = ω/k, describing in terms of displacement functions ϕ and ψ, there should be
ϕ = A exp[ik(X + Zctgα − CR t)] = A exp(−rZ) exp[ik(X − CR t)]

(11.33)
ψ = B exp[ik(X + Zctgβ − CR t)] = B exp(−sZ) exp[ik(X − CR t)]
X
Rayleigh surface wave
Nonuniform P wave
Nonuniform SV wave Coupled
Fig. 11.10. Schematics of Rayleigh surface wave.
where the coordinate system has been changed to that shown in Fig. 11.10 in order to
maintain consistency with the customary notations used in the discussion of Rayleigh
surface waves, namely the Y and X in Fig. 11.4 and Eq. (11.29) are now replaced by X
and (−Z). Note that in the case of nonuniform wave, both ctgα and ctgβ are imaginary
numbers, namely
2
CR
ctg α =
2
−1<1
CL
2 (11.34)
CR
ctg2 β = −1<1
CT
then we have
CR < CT < CL (11.35)
while r and s are all positive real numbers

2 <
CR
r =k 1− = k 2 − kL2 > 0
CL
(11.36)
2 <
CR
s =k 1− = k 2 − kT2 > 0
CL
Using Hooke’s law [Eqs. (7.10) and (11.13a)], the boundary conditions on free surfaces
can be expressed in terms of ϕ and ψ (actually ψ z , while the subscript z has been ignored
for convenience) as

∂uZ ∂uZ
σZZ |Z=0 = (λ + 2µ) +λ
∂Z ∂X Z=0
2 2
∂ ϕ ∂ 2ϕ ∂ ψ ∂ 2 ϕ
= (λ + 2µ) + 2 + 2µ − =0
∂X 2 ∂Z ∂Z∂X ∂X 2 Z=0

∂uX ∂uZ
σZX |Z=0 = µ +
∂Z ∂X Z=0

∂ 2ϕ ∂ 2ψ ∂ 2 ψ
=µ 2 + − =0
∂X∂Z ∂X 2 ∂Z 2 Z=0
Substituting Eqs. (11.33) and (11.34) in the above equations, we obtain

(λ + 2µ)r 2 − λk 2 A − i2µ ksB = 0
(11.37)
i2krA − k 2 + s 2 B = 0
The condition that the nontrivial solution of A and B exists is that determinant of coef-
ficients must be zero. Then we obtain the following Rayleigh equation with respect
to k 2
2 < <
2k 2 − kT2 − 4k 2 k 2 − kL2 · k 2 − kT2 = 0 (11.38a)
or
2 1/ 2 1/ 2
2 − KR2 = 4 1 − b2 kR2 1 − KR2 (11.38b)
where b2 = CT2 /CL2 = 1/κ is determined totally by material constants (refer to Eq. 11.20b),
while KR2 is
CR2 kT2
KR2 = =
CT2 k2
Therefore, if k can be solved from Rayleigh equation [Eq. (11.38)], then it is proved that
there really exists the surface wave expressed by Eq. (11.33).
Squaring Eq. (11.38), we obtain

KR2 KR6 − 8KR4 + 8KR2 3 − 2b2 − 16 1 − b2 = 0
Since the solution of KR2 = 0 is physically meaningless, the problem is now reduced to
solving the following cubic equation with respect to KR2
3 2
f KR2 = KR2 − 8 KR2 + 8 3 − 2b2 KR2 − 16 1 − b2 = 0 (11.39)
Because f (0) < 0 and f (1) = 1, there must be a real root in the interval 0 <
KR < 1. This is consistent with Eq. (11.35). For example, if ν = 1/4, λ = µ, and
√ 1/2
b2 = 1/3, then three roots can be solved from Eq. (11.39): KR1 = 2 − 2/ 3 ,
0.96
0.94
KR =CR / CT 0.92
0.90
0.88
0 0.1 0.2 0.3 0.4 0.5 v
Fig. 11.11. KR varies as a function of ν.
√ 1/2
KR2 = 2 + 2/ 3 , and KR3 = 2. However, the last two roots conflict with
Eq. (11.35). As only the first root is of significance, we obtain
CR = 0.9194CT or k = 1.087kT (11.40)
Figure 11.11 shows how KR varies as a function of ν in the range of 0 ≤ ν ≤ 1/2

(Knopoff, 1952).
The displacement components uX and uZ , according to Eq. (11.13a) in the current

situation, can be expressed as
∂ϕ ∂ψ
uX = −
∂X ∂Z
∂ϕ ∂ψ
uZ = +
∂Z ∂X
Substituting Eq. (11.33) in the above equations and using Eq. (11.37), we have

2rs
uX = iAk exp(−rZ) − exp(−sZ) exp[ik(X − CR t)] (11.41a)
k2 + s 2

r 2rk
uZ = Ak − exp(−rZ) + 2 exp(−sZ) exp[ik(X − CR t)] (11.41b)
k k + s2
Therefore, once KR (or equivalently k) is determined by Eq. (11.39), the displacement

of a Rayleigh surface wave can be calculated by Eq. (11.41). For example, if ν = 1/4,
the real parts of Eq. (11.41) are
uX = D(exp(−0.8475kZ) − 0.5773 exp(−0.3933kZ)) sin(kZ − ωt)

uZ = D −0.8475 exp(−0.8475kZ)−0.8475kZ + 1.4678 exp(−0.3933kZ)
× sin(kX − ωt) (11.42)
where D = Ak. At the free surface (Z = 0), the above equations are reduced to
uX = 0.433D sin(kX − ωt) (11.43a)

uZ = 0.62D cos(kX − ωt) (11.43b)
Therefore, the trajectory of a particle is an anticlockwise ellipse, as shown in Fig. 11.10.
u2X u2Z
+ =1
(0.433D)2 (0.62D)2
At Z = 0.193L (L = 2π/k is the wave length), the horizontal displacement is zero,

namely uX = 0, only vertical displacement exists. At further depths, Z/L > 0.193, the
sign of the wave amplitude reverses (an opposite phase). The trajectory of a parti-
cle is a clockwise ellipse (shown in Fig. 11.10). If we introduce the dimensionless
amplitudes of displacements (ûX /ûZ0 ) and (ûZ /ûZ0 ), where ûZ0 is the amplitude of
vertical displacement of particle at free surface, ûX (Z) and ûZ (Z) are the amplitudes
of uX and uZ , respectively, then Fig. 11.12 shows how the dimensionless amplitudes
of displacements ûX /ûZ0 and ûZ /ûZ0 exponentially decay with the dimensionless depth
Z/L (Viktorov, 1967).
The Rayleigh wave is a kind of surface wave due to the existence of free surface. Besides
this kind of wave, other kinds of surface waves may also exist. For instance, nonuniform
1.0
∧
0.8 uZ
uZ 0
uZ 0 , uZ0
uZ uX
0.6
∧
∧
uX
∧
0.4
uZ 0
0.2
0
−0.2
0 0.2 0.4 0.6 0.8 1.0 1.2 Z/L
Fig. 11.12. The exponential decay of (ûX /ûZ0 ) and (ûZ /ûZ0 ) with the depth Z/L.
waves propagating along the interface between two media can be similarly formed, such
as the so-called Stoneley waves, etc.
Reviewing the previous two sections, comparing with 1D stress wave problems, the com-
plexities of 3D elastic wave problems mainly lie on the conversion of wave mode and
the generation of new nonuniform waves during the reflection of waves arriving at the
interface. It is easy to understand that the 3D problems involving elastic–plastic waves,
visco-elastic waves, or elastic-visco-plastic waves, etc., will be much more complicated.
To solve these problems, we have to work with the aid of numerical methods, which will
be discussed in the next chapter.
CHAPTER 12
Numerical Methods for Stress Wave Propagation
A main difference in the study of structure response under impact loadings and quasi-static
loadings is that the inertia of each infinitesimal element of structure has to be considered,
which leads to the research on propagation of stress waves in various forms, whether using
accurate methods or simplified methods. Mathematically, it results in solving hyperbolic
wave equations. However, only under specially limited conditions, accurate solutions of
such equations can be obtained. Such restricted conditions include: geometry of structures,
spatial dimensions of motion, material constitutive equations, initial-boundary conditions,
etc. For example, even for a relatively simple problem, such as the propagation of waves
in a nonlinear elastic bar with finite length, if the nonlinear elastic constitutive equa-
tion of the bar is relatively complicated, then it could be very difficult, or sometimes
impossible, to obtain the analytical solution of interaction of two simple waves. Then,
numerical simulation is an important method to solve such hyperbolic (wave) equations.
Especially in the last half a century, accompanying rapid development in computer tech-
nology, there are well-developed computational methods, which have been widely used
to provide satisfactory approximate solutions even for the most complicated problems,
neither subjected to restrictions on the form of material constitutive models, nor subjected
to restrictions on initial and boundary conditions. Although mathematicians have only
been able to prove the methods being applicable to linear cases, in practice, such meth-
ods seem to be equally applicable in complicated nonlinear situations too (e.g. Chou and
Hopkins, 1972). At present, there are many numerical methods that can be used to analyze
the instantaneous response of structures involving stress wave propagation, regardless of
how complicated the material constitutive models and the initial-boundary conditions are.
In this chapter, three numerical methods are briefly introduced, including characteristics
method, finite difference method, and finite element method, which have been extensively
used to solve the stress wave problems. As a rudiment, only elementary theories of these
methods are introduced here. In addition, illustrations and features of these methods are
presented by analyzing some simple examples.
Among these three numerical methods, the characteristics method can be regarded
as an “optimized” finite difference method. Once the characteristic equations and the
475
corresponding compatibility equations along the characteristics have been determined by

manipulating the original wave equation set (hyperbolic partial differential equation set),
the problem to solve the wave equation set is transferred to solve these characteristic
equations and characteristics compatibility equations by using finite difference numeri-
cal methods. Generally, the latter is much simpler than the former. For problems of 1D
stress waves, since solving the original hyperbolic partial differential equations with two
independent variables has been converted into solving simultaneously the corresponding
ordinary differential equations, the problem is much more simplified in mathematics.
Moreover, the characteristics method can provide clear physical pictures of the prob-
lem to be solved and is capable to deal with the propagation of strong discontinuous
waves directly. So the characteristics method has been widely applied to solve problems
of 1D stress waves. However, for general multidimensional problems of stress waves,
the corresponding characteristics equations and compatibility equations along character-
istics are still quite complicated, so that the applications of characteristics method are
limited. Actually, only in some very special situations, characteristics method can be
applied to solve 2D or 3D problems. In other words, in the case of 1D stress waves,
the advantage of characteristics method is that the numerical integration of the original
partial differential equations is converted into numerical calculation of the corresponding
ordinary differential equations along characteristics. Obviously the difficulty in using the
characteristics method lies in the manipulation of such a conversion.
The basis of finite difference method is to convert the original partial differential equa-
tions into the corresponding approximate difference equations, namely, continuous time
and space are discretized into small intervals represented by nodes, then using Taylor
series to solve the partial differential equations on each node. Therefore, a group of alge-
braic equations with respect to the unknowns on nodes is obtained, considering the initial
and boundary conditions. Following the order of time and space, this group of alge-
braic equations can be solved step by step. Thus, finite difference method is looking for
the approximate solutions of the original accurate partial differential equations. The key
attention for this method should be focused on numerical errors, convergence, and stability
of this method.
Finite element method is based on the principle of variation. The spatial region (structure),
where the continuous functions described by partial differential equations exist, is divided
into finite small regions (elements). The variation of unknown functions in each element
is depicted by the selected function (shape function). Thus, the unknown functions in
the whole region (structure) are discretized and an approximate mathematical–physical
model is composed to describe the motion governed by the original partial differential
equations. This approximated model is characterized by a group of algebraic equations,
and the equations can be solved numerically by means of a computer. Therefore, finite
element method is, looking for precise solutions of an approximated model. As the partial
differentials with respect to time are included in the wave equation, the finite difference
method is usually used to deal with the time variable in finite element method.
However, readers should also realize the disadvantages of numerical methods. First,
the real influence of some variables on the results obtained may be concealed by such
approximate methods. Second, the numerical errors, convergence, and stability involved
in the numerical methods may encumber us to understand and recognize the physical
Numerical Methods for Stress Wave Propagation 477
essence of the problem studied. Therefore, it is usually necessary to judge and check
physically the numerical results from the view of stress wave theory, so that not only the
physical pictures of the obtained solutions can be correctly understood and explained, but
also the numerical methods can be more reasonably and effectively used to study the wave
propagation. In this sense, the contents of previous chapters on the fundamental theory of
stress wave propagation are also important to the readers who mainly work on numerical
simulations of structural dynamic response.
12.1 Characteristics Numerical Method
Characteristics method has been accepted extensively and has been applied in the study
of elastic and plastic 1D stress waves, because it can be used to simulate accurately,
propagation of strong discontinuous waves and it clearly figures out physical processes.
Actually, the basic theory and application of characteristics methods used in problems of
elastic or elasto-plastic 1D stress waves has been discussed in detail in Sections 2.3 and
2.4, and Sections in Chapter 4. Characteristics methods on visco-elastic and visco-plastic
1D stress waves have been studied in Chapter 6. Characteristics method on elasto-plastic
1D strain waves has been mentioned in Chapter 7. Characteristics method on spherical
waves has been analyzed in Chapter 8. In this section, characteristics method is briefly
discussed in the view of numerical methods. To obtain more details, readers should refer
to the related monographs (e.g. Chou and Hopkis, 1972).
12.1.1 Characteristics numerical method for 1D waves
In governing equations of 1D waves, there are two independent variables, namely spatial
variable X and time variable t. From Chapter 2, it is known that governing equations can
be represented by two first-order partial differential equations with respect to strain ε and
particle velocity v [refer to Eqs. (2.12) and (2.16)]. More generally, assuming f1 (X, t) and
f2 (X, t) are unknown functions of two variables, governing equations of wave propagation
can be written as
∂f1 ∂f1 ∂f2 ∂f2
a11 + b11 + a12 + b12 = R1
∂X ∂t ∂X ∂t
(12.1)
∂f1 ∂f1 ∂f2 ∂f2
a21 + b21 + a22 + b22 = R2
∂X ∂t ∂X ∂t
where aij , bij , and Ri (i,j = 1, 2) could be functions of f1 , f2 , X, and t.Using charac-
teristics method, the numerical integration on Eq. (12.1) is converted into the integration
of four ordinary differential equations, namely two families of characteristics S1 and S2
represented by
√
dX −b ± b2 − 4ac
= = τ1
dt 2a
√ (12.2)
dX −b − b2 − 4ac
= = τ2
dt 2a
where
a = b11 b22 − b12 b21

b = a21 b12 + b21 a12 − b11 a22 − b22 a11
c = a11 a22 − a12 a21
and two compatibility equations along characteristics represented by

< <
df1 C df2
A 1 + τ12 + B− 1 + τ12 = (τ1 b21 − a21 )R1 − (τ1 b11 − a11 )R2
dα τ1 dα
< <
df1 C df2
A 1 + τ2
2 + B− 1 + τ22 = (τ2 b21 − a21 )R1 − (τ2 b11 − a11 )R2
dβ τ2 dβ
(12.3)
where dα and dβ are the increment of arc length along the characteristics S1 and S2 ,
respectively, and
A = a11 b21 − b11 a21

B = a12 b21 − b11 a22
C = a12 a21 − a11 a22
Assuming that functions f1 and f2 are defined and specified on an arbitrary part QR of
a space-like curve as shown in Fig. 12.1, the solutions of partial differential equations
in the curving triangular region QPR which is bounded by the curve QR, the rightward
characteristic QP and the leftward characteristic RP, can be obtained. Such kind of problem
is often named the initial-value problem or Cauchy problem (referring to Section 2.4).
Since the characteristics in the same family cannot intersect, the solutions at P (x, t) depend
only on the variables defined in the curving triangle QPR. Such a domain bounded by the
space-like curve QR, the rightward characteristic QP, and the leftward characteristics RP
t P(X, t)
S1
S2
Domain of
Q
dependence
Fig. 12.1. Domain of dependence for P (x, t).

Region of
influence
S2 S1
Fig. 12.2. Region of influence of P (x, t).
is called the domain of dependence of point P (x, t). The solutions at P (x, t) depend
only on the initial values prescribed on the curve Γ which lies between Q and R.
Similarly, it is easy to understand that the region of influence of point P (x, t) is an infinite
curving triangle which is bounded by the leftward and rightward characteristics passing
the point P (x, t), as shown in Fig. 12.2. Obviously, the solutions in the region of influence
are influenced by the variables at P (x, t).
As discussed in Section 2.4, if functions f1 and f2 are defined and specified on a space-
like curve and are also defined and specified on a time-like curve, then the solutions in the
region bounded by both the curves can be uniquely determined. Such kind of problems
are usually named the mixed problem or Picard problem.
Taking the 1D wave propagation in a semi-infinite linear elastic bar as a simple example,
the steps of numerical integration along characteristics are illustrated as follows. According
to Eqs. (2.12) and (2.16), the governing equations with respect to the unknown functions
ε and v can be written as
∂v ∂ε
− =0
∂X ∂t
(12.4)
∂v ∂ε
− C2 =0
∂t ∂X
where C = (E/ρ0 )1/2 is the elastic wave velocity in bars. Assuming that the bar is initially
in an undisturbed state, namely the initial condition is
v(X, 0) = ε(X, 0) = 0 (12.5a)
To start a motion of the media, a velocity boundary condition at X = 0 is given by
v(0, t) = g(t) (12.5b)

It is clear that if a11 = 1, a22 = −C 2 , b12 = −1, b21 = 1 in Eq. (12.1) and other
coefficients are zero, then Eq. (12.1) is converted to Eq. (12.4). Therefore, the differential
equations of characteristics S1 and S2 can be obtained from Eq. (12.2)
dX dX
S1 : =C and S2 : = −C (12.6)
dt dt
Furthermore, the compatibility equations can be obtained from Eq. (12.3)
dv dε
−C =0 (12.7)
dα dα
dv dε
+C =0 (12.8)
dβ dβ
where Eq. (12.7) is valid along S1 and Eq. (12.8) is valid along S2 .
As the velocity C of an elastic wave in bars is a constant, Eq. (12.6) shows that the two
families of characteristics S1 and S2 are straight lines with slopes ±C (wave velocity),
respectively. Thereby, the grids of characteristics in the X–t plane can be drawn as shown
in Fig. 12.3. The size of the grid ∆X is dependent on the precision required for the problem
studied, while the time step is ∆t = ∆X/C.
Our aim is to obtain numerical solutions at every node. Actually three categories of
computation are involved, namely, (1) the computation of points on the strong discon-
tinuity (such as the nodes 1, 3, 6, 10. . . in Fig. 12.3), (2) the computation of points on the
boundary (such as the nodes 2, 5, 9. . . in Fig. 12.3), and (3) the computation of the interior
points (such as the nodes 4, 7, 8. . . in Fig. 12.3).
Computation of points on the strong discontinuous wave front. If there is a velocity-jump

on the boundary at t = 0, namely if velocity-jump g(0)
= 0 is loaded on the end of a semi-
infinite linear elastic bar, then it can be proved that for this linear system, the velocity-jump
S2 S1 S1, S2− characteristics

V(0,t)=g(t)
discontinuous wave front
⌬t (undisturbed zone)
points on the 9 8 7
boundary 5 4 10 ⌬x
2 6 points on the discontinuity
3 interior points
1
0 X
Fig. 12.3. Characteristics of a 1D wave in a semi-infinite linear elastic bar.

will propagate along a characteristic in simple wave zone (refer to Section 2.5), and keeps
its amplitude value (a constant intensity of the strong discontinuity) during its propagation
along this leading characteristics, namely
[v] ≡ v+ − v− = g(0) = constant
and according to the compatibility conditions across a strong discontinuous wave front
[Eq. (2.55)], strain discontinuity also keeps constant
1 g(0)
[ε] = − [v] = −
C C
As it is known from Eq. (12.5a) that the particle velocity and strain ahead of the strong
discontinuous wave front are all zero, the particle velocity v and strain ε behind the wave
front can be solved by using the above two equations directly. Along the rightward leading
characteristics S1 , namely behind the strong discontinuous wave front, we have
g(0)
v = g(0), ε = −
C
When g(0) approaches zero, the strong discontinuous wave front converts to a weak
discontinuous wave front, which results in the particle velocity v and the strain ε along
the rightward leading characteristic S1 also approaches zero. Thus, it is concluded that
the particle velocity v and the strain ε can be determined uniquely for both a strong
discontinuous wave and a weak discontinuous wave.
Computation of points on the boundary. The solutions at an arbitrary point P on the

boundary can be obtained by using the boundary condition and the compatibility condition
along the leftward characteristics passing this point. In Fig. 12.4, the line QP is the leftward
characteristics S2 connecting the point P and the point Q, at which the solutions have
been obtained in previous computation. The compatibility condition along the leftward
characteristic S2 [Eq. (12.8)] should be satisfied, which in a finite difference form can be
written as
vP − vQ + C(εP − εQ ) = 0
B
S1 S2
A C
P S2
Q
Fig. 12.4. Computation pattern of boundary point P and interior point B.

As the particle velocity at the point P , vp = v(0, tp ), has been determined by the boundary
condition v(0, tp ) = g(tp ), the strain at P on the boundary can be uniquely calculated by
the above equation.
Computation of interior points. For an arbitrary interior point B neither on discontinuous

lines nor on boundaries, there are two characteristics S1 (AB) and S2 (CB) through the
point, as shown in Fig. 12.4. Then, the solutions at the point B can be determined by
the compatibility conditions along these two characteristics S1 (AB) and S2 (CB), namely
Eqs. (12.7) and (12.8), which in the finite difference form are
along S1 : vB − vA − C(εB − εA ) = 0
along S2 : vB − vC + C(εB − εC ) = 0
As solutions at points A and C have been obtained in previous computations, the solutions
of vB and εB , at B are determined uniquely.
Therefore, according to the time order of nodes on characteristics grids shown in Fig. 12.3,
the solutions of an elastic semi-infinite bar can be obtained from left to right in turn.
For an elastic finite bar, the computation of end nodes is involved due to reflection of stress
waves at the bar end. Although it increases the complexity of computation a little bit, the
computation on those nodes is, in principle, similar to the computation on the boundary
points on the loading boundary as discussed earlier.
However, for elasto-plastic bars, due to the nonlinearity in elasto-plastic constitutive equa-
tion, the governing equations are generally quasi-linear. In this situation, the characteristics
equations are nonlinear and the slopes of characteristics depend on the dependent variables
(f1 and f2 ). Consequently, the characteristics equations and the characteristics compati-
bility equations have to be solved together for the determination of unknown functions f1
and f2 . A repeated iteration is sometimes needed to successively approach the solutions
which increases difficulties in characteristics numerical methods. In Chapter 4, actually,
it has been discussed in detail how to solve such problems of elasto-plastic waves prop-
agating in bars, including the analyses of the interaction of elasto-plastic waves and the
propagation of elasto-plastic boundaries, etc. Those are not necessary to be repeated here
again.
12.1.2 Characteristic surfaces numerical method for 2D waves
In the previous section, characteristics numerical method is mainly applied to solve the
problems of 1D wave propagation, and the problems of 2D waves have not been touched.
For 1D problems of wave propagation, the advantages of characteristics method are very
clear. The reason is that only one characteristic direction exists, so the problem of numeri-
cally integrating the original partial differential equations can be reduced to integrating the
ordinary differential equations which involve the directional derivatives along the char-
acteristics. Consequently, the difficulty to obtain the solutions of the governing equations
decreases much more. However, for 2D problems that include three independent vari-
ables, the problem of numerically integrating the original partial differential equations,
under certain conditions, can be simplified to integrating the equations which involve the
directional derivatives along two characteristics directions on the characteristic surfaces.
Unfortunately, these are still partial differential equations, so that the simplification for
solving the equations in 2D situation is not as great as that in 1D situation. Therefore,
relatively complex numerical schemes are required for the integration of these equations.
In some very special situations and strictly limited conditions, the problem of integrating
the original partial differential equations can be converted into integrating the lower-order
partial differential equations which involve the characteristic equations along the so-called
bicharacteristics on a characteristic cone.
In an example of a 2D linear elastic waves below, the elementary numerical method of

characteristic surfaces will be discussed in brief (refer to Chou and Hopkins, 1972).
First, from Eq. (11.13) in Chapter 11, it has been established that the governing equations
of 3D linear elastic waves are

∂ 2ϕ λ + 2µ ∂ 2 ∂2 ∂2
= + + ϕ = Cl2 ∇ 2 ϕ
∂t 2 ρ0 ∂X 2 ∂Y 2 ∂Z 2
2
∂ 2Ψ µ ∂ ∂2 ∂2
= + + Ψ = CT2 ∇ 2 Ψ
∂t 2 ρ0 ∂X 2 ∂Y 2 ∂Z 2
where ϕ denotes the scalar potential of displacement of dilatation waves (irrota-

tional waves), Ψ denotes the vector potential of displacement of distortion waves (isometric
waves), CL and CT are the corresponding wave velocities of dilatation wave and distor-
tion wave respectively and both of them are constant (refer to Chapter 11). Concisely, the
above two equations can be represented by one equation
∂ 2φ ∂ 2φ ∂ 2φ 1 ∂ 2φ
∇ 2φ = 2
+ 2
+ 2
= 2 2 (12.9)
∂X ∂Y ∂Z C ∂t
where C is a constant (C could be CL or CT , depending on different instances).
If a problem is cylindrically symmetric, it can be converted into a 2D problem. In the

cylindrical coordinates (r, θ ,Z), Eq. (12.9) is reduced to
∂ 2φ 1 ∂φ ∂ 2φ ∂ 2φ
+ + = (12.10)
∂r 2 r 2r ∂Z 2 ∂T 2
where T = Ct. In practical problems, the first-order partial derivatives of displacement

potential φ (displacement, etc.) are of much concern. Taking
∂φ ∂φ ∂φ
p= , q= , S= (12.11)
∂r ∂Z ∂t
then Eq. (12.10) is equivalent to the following first-order partial differential equation set
⎧
⎪
⎪
∂S
=
∂p p
+ +
∂q
⎪
⎪
⎪
⎪ ∂T ∂r r ∂Z
⎨
∂p ∂S
= (12.12)
⎪
⎪ ∂T ∂r
⎪
⎪
⎪
⎪
⎩ ∂q = ∂S
∂T ∂Z
Following the “indeterminate line method” used to determine the characteristics in 1D
wave propagation (Section 2.3), the characteristic directions for Eq. (12.12) are determined
below. Multiplying the above three equations (in Eq. 12.12) with indeterminate coefficients
α1 , α2 , and α3 , their summation yields
∂p ∂p ∂q ∂q ∂S ∂S ∂S p
−α1 + α2 − α1 + α3 − α2 − α3 + α1 = α1 (12.13)
∂r ∂T ∂Z ∂T ∂r ∂Z ∂T r
We should determine these indeterminate coefficients to make the above equation include
only the directional derivatives along the characteristic directions on a common surface.
If er , ez , and eT are the unit vectors on axis r, Z, and T respectively, Eq. (12.13) can be
written in a vector form
p
A1 · ∇p + A2 · ∇q + A3 · ∇S = α1 (12.14)
r
where the vectors A1 , A2 , A3 and the gradient operator ∇ are defined below:
A1 = −α1 er + α2 eT
A2 = −α1 eZ + α3 eT
A3 = −α2 er + α3 eZ + α1 eT
∂ ∂ ∂
∇ = er + ez + eT
∂r ∂z ∂T
The first item on the left side of Eq. (12.14) represents the product of the magnitude of
vector A1 and the directional derivative of p taken in the direction of vector A1 . It can be
interpreted similarly for the second and third items.
Similar to the directional derivative method in obtaining the characteristics equations

in the 1D problems, it is required that the three derivative directions A1 , A2 , and A3
must
be on a common plane. If the normal unit vector of this plane is denoted as
λ = λγ er + λz ez + λT eT , then
A1 · λ = A2 · λ = A3 · λ = 0
which in matrix form can be written as

⎡ ⎤⎡ ⎤
−λr λT 0 α1
⎣ −λz 0 λ T ⎦ ⎣ α2 ⎦ = 0 (12.15)
λT −λr −λz α3
A nontrivial solution of indeterminate coefficients α1 , α2 , and α3 will exist if the

determinant of the above coefficient matrix equals zero, namely
λT (λ2r + λ2z − λ2T ) = 0 (12.16)
If λT
= 0 (λT > 0), assuming that θ is the angle between the r-axis and the projection of
λ on the plane r–Z, then Eq. (12.16) is reduced to
λ= √1 (cos θ er + sin θ eZ + eT ) (12.17)

2
This normal vector λ of the characteristic surface determines a family of planes with a
single parameter θ ; their enveloping surface is a cone, which is called the characteristic
cone (Fig. 12.5). Then the characteristic surface passing through the point O (r0 , Z0 , T0 )
is expressed as
(r − r0 )2 + (Z − Z0 )2 = (T − T0 )2
Assuming that the unit vector β is the bicharacteristic direction which is the intersection
between the characteristic plane and the characteristic cone, and the unit vector γ is tangent
to the cone and perpendicular to λ and β, then according to Eq. (12.17), β and γ can be
expressed respectively as
β= √1 (− cos θ er − sin θ eZ + eT )
2
(12.18)
γ= √1 (− sin θ er + cos θ eT )
2
As shown in Fig. 12.5, at the point i, the intersection between the characteristic cone and
the plane parallel to r–Z plane is a circle. It is clear that γ is the tangent unit vector of this
Z
0 (r0, Z0, T0)
r
b l
g
i
θ
Fig. 12.5. The characteristic cone at the point O(r0 , z0 , T0 ).

circle at the point i. Furthermore, from Eq. (12.15), we can also obtain
α2 α3
= cos θ and = sin θ
α1 α1
Thus, dividing each side of Eq. (12.13) by α1 , using the above equations, we obtain
∂p ∂p ∂q ∂q ∂S ∂S ∂S p
− + cos θ − + sin θ − cos θ − sin θ + = (12.19a)
∂r ∂T ∂Z ∂T ∂r ∂Z ∂T r
Or in a vector form, we have
1 p
cos θβ · ∇p + sin θβ · ∇q + β · ∇S = √ (cos θ∇q − sin θ∇p) · γ + (12.19b)
2 r
Equation (12.19) is the compatibility equation on characteristic surface, it contains only

directional derivatives along the bicharacteristics β and γ. If dβ is the length incre-
ment in direction β of bicharacteristics defined by θ , and dγ is the length increment
in direction γ, then according to the definition of directional derivative, the compatibility
equation Eq. (12.19) can be written in the form of ordinary differential equation

dp dq dS 1 dq dp p
cos θ + sin θ + = √ cos θ − sin θ + (12.20)
dβ dβ dβ 2 dγ dγ r
It should be noted, if the λT in Eq. (12.16) equals zero, any plane perpendicular to the
plane r–Z is always a characteristic surface. Following the steps presented in previous
section, the corresponding compatibility equation can be obtained

∂S ∂P ∂q ∂S
sin θ − + cos θ − =0
∂r ∂T ∂T ∂Z
Actually, this equation is only another form of the last two equations in Eq. (12.12). If we
search a characteristic surface from the original second-order partial differential equation
(Eq. 12.9), then λT will not appear. Consequently, the characteristic surface perpendicular
to the plane r–Z can be neglected. The reason is that three new variables p, q, and S were
adopted, which introduced the λT .
Steps to obtain numerical solution of the compatibility equation [Eq. (2.20)] will be dis-
cussed below. In Eq. (12.20), there are two directional derivatives in directions along β
and γ which increase the difficulty of solving numerically. For further simplification, we
will manipulate the compatibility Eq. (12.20). Expanding the right side of Eq. (12.20) and
multiplying them with dβ, we obtain

dβ 2 ∂p ∂q ∂p 2 ∂p p
cos θ dp + sin θ dq + dS = √ sin θ − sin θ cos θ + + cos θ +
2 ∂r ∂r ∂Z ∂Z r
(12.21)
where all differentials denote the increment in the directions of the bicharacteristics defined
by β. Comparing with the partial differential Eq. (12.12), Eq. (12.21) contains neither the
partial differential of the dependent variable S, nor the partial differential with respect
to the variable T , which results in the compatibility Eq. (12.21) for which it is much
easier to obtain numerical solution than Eq. (12.12). Note that the first equation of the
bicharacteristics Eq. (12.18) can also be written as
dr dZ dT
= = (12.22)
− cos θ − sin θ 1
consequently, the length increment of characteristics can be written as
; √
dβ = dr 2 + dZ 2 + dT 2 = 2dT
√
and then the dβ/ 2 in Eq. (12.21) can be replaced by dT. On the other hand, in Eq. (12.21),
the parameter θ is an arbitrary value between 0 and 2π . When θ is set to be a specific
value θI , which means that a bicharacteristics has been selected to connect the points I
and J (shown in Fig. 12.6), the compatibility equation in a finite difference form between
the two points I and J can be written as
cos θI ∆p + sin θI ∆q + ∆S =
/ % & 0
∂p ∂q ∂p ∂p 1 pI pJ
∆T sin θI
2
− sin θI cos θI + + cos θI
2
+ +
∂r ∂r ∂Z ∂Z 2 rI rJ
(12.23)
Where ∆Y = YJ − YI , (Y = p, q, S, T ), and ∂Y /∂r is defined as the average of partial

derivatives along the bicharacteristics between the points I and J . Here, we consider that
the state before time J has been determined, in other words, all the variables of every
point located in the bottom of the cone which passes through the point J , are known.
The rest of the problem is how to determine the average of partial differential in Eq. (12.23).
There are generally two methods (Chou and Hopkins, 1972). One is to assume that
bicharacteristics
J
T
⌬T
Z
I
qΙ
characteristic cone
Fig. 12.6. Characteristic cone and the bicharacteristics connecting points I and J .
the dependent variables are continuous in the region between points I and J , then two
additional equations can be obtained (Sauerwin, 1967),
∂p ∂p ∂p
dp = dr + dZ + dT (12.24a)
∂r ∂Z ∂T
∂q ∂q ∂q
dq = dr + dZ + dT (12.24b)
∂r ∂Z ∂T
Their finite difference forms are
∂p ∂p ∂p
∆p = ∆r + ∆Z + ∆T
∂r ∂Z ∂T
(12.25)
∂q ∂q ∂q
∆q = ∆r + ∆Z + ∆T
∂r ∂Z ∂T
Here, the average of partial differential is defined by (Fig. 12.7)

∂Y 1 ∂Y ∂Y
= +
∂y 2 ∂y I ∂y J
Thus, we can determine the variables at point J based on the known variables at point I
by using Eqs. (12.23) and (12.25). However, only three equations can be obtained based
on one characteristics, while there are nine unknowns pJ , qJ , SJ , ∂p , ∂p , ∂p ,
∂q
∂q
∂q
∂r J ∂Z J ∂T J
∂r J , ∂Z J , and ∂T J . Hence to solve all the unknowns, six equations are still needed.
It is known that every straight line from the bottom of characteristic cone to its top is a
bicharacteristics. According to the numerical scheme shown in Fig. 12.7 (Sauerwin, 1967),
we can select arbitrarily three bicharacteristics to provide nine equations. Consequently,
all nine unknowns can be obtained numerically.
Another numerical method used to determine the average of partial differential in

Eq. (12.23) is to adopt the numerical scheme based on four bicharacteristics, which
is shown in Fig. 12.8 (Butler, 1962; Clifton, 1967). Corresponding to the four
⌬T
3 2
I
1 TJ − ⌬T
Fig. 12.7. Numerical scheme based on three bicharacteristics.

⌬T
3
4 2
5
1 TJ − ⌬T
Fig. 12.8. Numerical scheme based on four bicharacteristics.
bicharacteristics, four compatibility equations (I = 1, 2, 3, 4) can be obtained according

to Eq. (12.23)
cos θI (PJ − PI ) + sin θI (qJ − qI ) + (SJ − SI )

∆T ∂p ∂p ∂q ∂q PJ PI
= sin2 θI + + cos 2
θ + + +
∂r J ∂r I ∂Z J ∂Z I
I
2 rJ rI
(12.26)
where θI = 0, π/2, π, 3π /2. Note that sin θI · cos θI = 0 here. As all relevant quantities
in the cone bottom are known,
Eq. (12.26) includes four equations with five unknowns:
pJ , qJ , SJ , ∂p , and ∂q . To solve these equations, an additional equation is needed.
∂r J ∂Z J
Connecting a perpendicular line from point J to the bottom at the intersect point 5
(Fig. 12.8), and integrating the first equations of Eq. (12.12) from points 5 to J , we obtain
the following equation

∆T ∂p ∂p PJ PS ∂q ∂q
S J − S5 = + + + + + (12.27)
2 ∂r J ∂r 5 rJ rS ∂Z J ∂Z 5
Then, using the five equations [Eqs. (12.26) and (12.27)], five unknowns
can be solved.
The actual steps include to eliminate derivative items ∂p and ∂q first, then using the
∂r J ∂Z J
three equations obtained to solve pJ , qJ , and SJ , the derivatives on the plane TJ − ∆T
of the cone bottom can be determined by using central difference equations.
Therefore, the dependent variables at every point can be computed numerically by using
either of the above two methods. But it should be noted that in order to obtain a stable
numerical result, the Courant–Friedrichs–Lewy criterion must be satisfied; namely, for a
set of first-order linear hyperbolic equations, the domain of dependence of the differ-
ence equations must contain the domain of dependence of the differential equations.
In Butler’s method (numerical scheme based on four bicharacteristics), for example, the
domain of dependence of the differential equations is a cone, and its intersection with
the plane TJ − ∆T is a circle. But the intersection between the domain of dependence of
the difference equations (central difference) and the plane TJ − ∆T is a right rectangle
as shown by the dashed lines in Fig. 12.9. It is stable only when the former is contained
J
T
⌬T ′
3 c
⌬T
3
Z 4 d 2
5 2 4 5 b
1
r 1 a
(a) Unstable (b) Stable
Fig. 12.9. Unstable numerical scheme and critical stable numerical scheme.
in the latter. If time step ∆T is too large, the right rectangle (the domain of dependence
of the difference equations) cannot cover the circular area of the differential equations, as
shown in Fig. 12.9(a), which results in an unstable numerical result. In this case, the time
step ∆T must be reduced to ensure that the rectangular area contains the circular area.
The critical state is shown in Fig. 12.9(b). In the numerical scheme of critical stability,
although each point (1, 2, 3, and 4) on the cone bottom is not a node, its physical quantities
can be obtained by an interpolation from the values at its neighboring nodes (a, b, c, d, 5).
It should be noted that the discussion so far is mainly based on Eq. (12.9). When it is applied
to 2D elastic wave problems, in fact, it is known from Eq. (11.13) that elastic waves are
actually composed of dilatation waves and distortion waves, propagating at velocities CL
and CT respectively. When a 2D characteristic surface method is adopted, there are two
kinds of characteristic cones generated: a volumetric characteristic cone which corresponds
to the propagation of dilatation wave, and a shear characteristic cone which corresponds to
the propagation of distortion wave. The whole computation is more complicated than that
discussed above. It can be imagined that the computation is much more complicated in
multidimensional nonlinear elasto-plastic, and visco-plastic problems. In such cases, other
numerical methods can be adopted, including the finite difference method to be discussed
in the next section or the finite element method to be discussed later.
12.2 Finite Difference Method
To introduce finite difference method (refer to Chou and Hopkins, 1972; Xu, 1985), we
discuss a simplest example of a hyperbolic equation
∂f ∂f
−a =0 (12.28)
∂t ∂X
with an initial condition
f (X, 0) = ψ(X), −∞ < X < +∞ (12.29)

Actually, Eq. (12.28) is in the same form as the following well-known wave equation.
∂v ∂ε
− C2 =0
∂t ∂X
(12.30)
∂ε ∂v
− =0
∂t ∂X
When such an equation is solved by means of finite difference method, it is first needed to
establish the corresponding finite difference format which is compatible with the partial
differential equation. Meanwhile, the convergence and stability of these difference equa-
tions should be checked. For strong discontinuous waves, artificial viscosity should also
be introduced. These aspects will be discussed below.
12.2.1 Establishment of finite difference format
The X–t plane is meshed in rectangular grids by the following two groups of isometric
straight lines parallel to the axis
X = Xk = k∆X k = 0, ±1, ±2, . . .

t = tj = t0 + j ∆t j = 0, 1, 2, . . .
where ∆X and ∆t are step length (spacing) in the X and t directions respectively. From
now on, for convenience, the node coordinate (Xk , tj ) is denoted by (k, j ) and f (Xk , tj )
is denoted by f (k,j ). In finite difference method, partial differential equations are approx-
imated by difference equations. But for the same first-order partial derivative, there are
several difference formats. For example, the so-called forward difference quotient is
defined by

∂f f (k,j + 1) − f (k,j ) ∆t ∂ 2 f (k,t1 )
= −
∂t (k,j ) ∆t 2 ∂t 2
(12.31)
∂f f (k + 1,j ) − f (k,j ) ∆X ∂ 2 f (X1 ,j )
= −
∂X (k,j ) ∆X 2 ∂X 2
and the backward difference quotient is defined by

∂f f (k,j ) − f (k,j − 1) ∆t ∂ 2 f (k,t2 )
= +
∂t (k,j ) ∆t 2 ∂t 2
(12.32)
∂f f (k,j ) − f (k − 1,j ) ∆X ∂f 2 (X2 ,j )
= −
∂X (k,j ) ∆X 2 ∂X 2
while the central difference quotient is defined by

∂f f (k,j + 1) − f (k,j − 1) ∆t 2 ∂ 3 f (k,t3 )
= −
∂t (k,j ) 2∆t 6 ∂t 3
(12.33)
∂f f (k + 1,j ) − f (k − 1,j ) ∆X 2 ∂ 3 f (X3 ,j )
= −
∂X (k,j ) 2∆X 6 ∂X 3
where tj ≤ t1 ≤ tj +1 , Xk ≤ X1 ≤ Xk+1 , tj −1 ≤ t2 ≤ tj , Xk−1 ≤ X2 ≤ Xk ,

tj −1 ≤ t3 ≤ tj +1 and Xk−1 ≤ X3 ≤ Xk+1 . So by taking different difference formats,
the different difference equations are obtained. Therefore, it is necessary to study which
difference format is more suitable to the partial differential equations to be solved. Here, the
“more suitable” means that a shorter computation time is needed under a given accuracy.
The accuracy is generally depicted by errors.
For example, substituting the forward difference quotient format Eq. (12.31) in Eq. (12.28),
we have
f (k,j + 1) − f (k,j ) f (k + 1,j ) − f (k,j )

+a − R1 (∆X,∆t) = 0
∆t ∆X
where
∆t ∂ 2 f (k,t1 ) a∆X ∂ 2 f (X1 ,j )

R1 (∆X,∆t) = + = 0(∆t,∆X)
2 ∂t 2 2 ∂X 2
is called the error item. When ∆t and ∆X are small enough, since R1 (∆X, ∆t) is the
same order of small quantity as ∆t and ∆X, then the difference equations of Eq. (12.28) is
f (k,j + 1) − f (k,j ) + aξ [f (k,j + 1) − f (k,j )] = 0 (12.34a)
where ξ = (∆t/∆X). The R1 (∆X, ∆t) is named the truncation error of Eq. (12.34).
Considering that the initial condition Eq. (12.29) can be expressed as
f (k, 0) = g(k), k = 0, ±1, ±2, . . .
and combining Eqs. (12.28) and (12.29), we obtain the final difference format
f (k,j + 1) = f (k,j ) − aξ [f (k + 1,j ) − f (k,j )]

(12.34b)
f (k,0) = g(k)
If the partial differential with respect to t is still approximated using forward difference
quotient format, but the partial differential with respect to X is approximated using back-
ward difference quotient format, then according to the definition of backward difference
quotient format, namely the second equation of Eq. (12.32), the difference formats of
Eqs. (12.28) and (12.29) are written as
f (k,j + 1) = f (k,j ) − aξ [f (k,j ) − f (k − 1,j )]

(12.35)
f (k, 0) = g(k)
its truncation error R2 (∆X, ∆t) is
∆t ∂ 2 f (k, t1 ) a∆X ∂ 2 f (X2 , j )

R2 (∆X, ∆t) = − = 0(∆X, ∆t)
2 ∂t 2 2 ∂X 2
Similarly, if the partial differential with respect to X is approximated by using central

difference quotient format, the corresponding difference formats are
aξ
f (k,j + 1) = f (k,j ) − [f (k + 1,j ) − f (k − 1,j )]
2 (12.36)
f (k, 0) = g(k)
its truncation error R3 (∆X, ∆t) is
∆t ∂ 2 f (k, t1 ) a∆X 2 ∂ 3 f (X3 ,j )

R3 (∆X, ∆t) = − = 0(∆t, ∆X2 )
2 ∂t 2 6 ∂X 3
In the above three difference formats, the partial derivative with respect to t is all approx-
imated by a forward difference quotient. In such a case, when the initial condition g(k)
is given, the f (k, 1) of grid nodes on the first layer can be calculated. In general, when
the f (k, j ) of the layer j have been known, the f (k, j + 1) of the layer j + 1 can be
calculated. This difference format is named the explicit format.
If we use the backward difference quotient for t, while the central difference quotient
for X, then the difference formats for the same problem mentioned above are
aξ
f (k,j + 1) = f (k,j ) − [f (k + 1,j + 1) − f (k − 1,j + 1)]
2 (12.37)
f (k, 0) = g(k)
its truncation error R4 (∆X,∆t) is
∆t ∂ 2 f (k,t2 ) a∆X 2 ∂ 3 f (X3 ,j )

R4 (∆X,∆t) = − = 0(∆t,∆X 2 )
2 ∂t 2 6 ∂X 3
For the difference formats Eq. (12.37), when the initial value of f (k,0) is known, we
cannot directly calculate f (k, 1) [and successively up to f (k, j )], but must solve a group
of simultaneous equations to obtain the solution of f at t = 1. This difference format
is named the implicit format. As it is needed to solve simultaneous equations, implicit
formats are suitable to the problems with the combining initial-boundary value conditions.
In fact, there are many other difference formats. Readers can refer to the relevant books
(e.g. Xu, 1985).
In the above discussions on the difference formats and their truncation errors, it is clear that
their truncation error approaches zero when ∆X, ∆t → 0. This indicates that the limit
of a difference format is its corresponding partial differential equation. The difference
format which satisfies this condition is called, that it is compatible with the corresponding
partial differential equations. From the truncation error R1 to R4 , it is clear that these
difference formats are of first-order of accuracy with respect to ∆t. On the other hand,
R1 and R2 indicate that Eqs. (12.34) and (12.35) are of first-order of accuracy with respect
to ∆X, while the R3 and R4 indicate that the central quotient formats are of second-order
of accuracy with respect to ∆X.
In numerical simulation, the compatibility of difference formats is necessary but not a

sufficient condition. The convergence and stability of numerical methods should also be
considered and will be analyzed in the following sections.
12.2.2 Convergence of difference formats
Consider a point P (X, t) in the solution region and X = k∆X, t = j ∆t, if a difference
format is used to calculate f (k, j ) at the point P , and when ∆, ∆X → 0, f (k, j ) satisfies
the following relationship
f (k, j ) − fˆ(k, j ) → 0
where fˆ represents the accurate solution while f is the numerical solution, then this
difference format is regarded to be convergent.
For convenience, in the following discussion on the convergence of difference format,

we first investigate the domain of dependence of different difference formats. If we use
Eq. (12.34) to calculate f at point P , as shown in Fig. 12.10(a), the value of f at point P
depends on the quantities at nodes on the initial line segment BC. Then, the segment
BC is called the domain of dependence of the solution of Eq. (12.34) at point P . If we
use Eq. (12.35) to calculate f at point P , as shown in Fig. 12.10(b), the segment BC is
the domain of dependence of point P . In Fig. 12.10(c) the segment BC is the domain of
dependence of the solution of Eq. (12.36) at point P . Here, the domain of dependence
means the initial segment where the grid nodes are located. Figure 12.10 also indicates
that the domain of dependence is not only dependent on the difference formats but also
related to the time being calculated.
Consider the case of a > 0 in Eq. (12.28). It is known that the characteristics through
point P must be located on the left side of the point, as shown in Fig. 12.10. If this
characteristics intersects the line t = 0 at point D, then the line PD is the characteristics
passing through the point P . If the domain of dependence of a difference equation does
not cover the domain of dependence of a differential equation, then the value fp obtained
by difference equation at point P is irrelevant to the quantities at point D. Consequently,
P P P
t=4
t=2
t=1
t=0
D B C BD C B D C
(a) (b) (c)
Fig. 12.10. The domain of dependence of three difference formats.
fp − fˆp → 0 cannot be satisfied always, namely the convergence of difference format

cannot be ensured. Therefore, a necessary condition of convergence of a difference format
is to satisfy the Courant condition, namely the domain of dependence of a difference for-
mat must cover that of a differential equation. When a > 0, thereby, from Fig. 12.10, it
is clear that (a) the difference format shown in Fig. 12.10(a) (Eq. (12.34)) is not convergent,
(b) the Courant condition of convergence for the difference format Eq. (12.35) is
0 ≤ a∆t ≤ ∆X (a > 0) (12.38)
and (c) the Courant condition of convergence for the difference format Eq. (12.36) is
∆X
−∆X ≤ a∆t ≤ ∆X namely |a| ≤ (12.39)
∆t
These conditions should be noticed in difference meshing.
12.2.3 Stability of difference formats
Truncation error is not the only error source in numerically solving initial-value problems
using difference formats, because there is also rounding error in every computing step.
Computation of initial-value problem is conducted layer by layer. Even though rounding
error of each layer is very small, the sum of rounding error accumulates and propagates
with increasing layer number, which may be very large. As this accumulated error grows
larger and larger, the real solution may be “submerged” by the accumulated error, which
results in an instability of numerical results. Thereby, it is very important to analyze the
stability of difference formats.
If the coefficients in differential equations are constants, Fourier analysis can generally
be adopted to study the stability of difference formats, this is the so-called von Neumann
method. The steps of this method will be illustrated by an example of the difference format
of Eq. (12.35).
If the error of Eq. (12.35) is written as δ(k,j ) = f (k,j ) − fˆ(k,j ), it is known that
δ(k, j + 1) should satisfy
δ(k,j + 1) = δ(k,j ) − aξ [δ(k,j ) − δ(k − 1,j )] (12.40)

If δ(k,j ) is a harmonic wave with amplitude Gj and frequency w, then δ(k, j ) is written as
δ(k,j ) = Gj eiwXk (w is an arbitrary real number) (12.41)
Substitution of the above equation into Eq. (12.40) yields
G = 1 − aξ + aξ e−iwX k (12.42)
Replacing the exponent functions with triangle functions, and using the double-angle
relationship of triangle functions, the above equation can be written as
wX k wX k wX k
G = 1 − 2aξ sin2 − i2aξ sin cos (12.43a)
2 2 2
and
wX k
|G|2 = 1 − 4aξ(1 − aξ ) sin2 (12.43b)
2
where |G| is named the growing factor. According to Eq. (12.41), when |G| > 1, error
grows with j exponentially. Therefore, the stability condition is
|G| ≤ 1 (for every w) (12.44)
And Eq. (12.43) also indicates that the stability condition of this difference format
(Eq. (12.35)) is
0 ≤ aξ < 1 (for a > 0)
From this example, it is clear that the main steps in von Neumann’s method are: first,
assuming the solution of error equation is a form of harmonic wave; second substituting
this solution into error equation to obtain the error growing factor; then checking whether
the error growing factor is not larger than one. Through these steps, we can judge that the
difference format Eq. (12.36) is not stable permanently, while the stability condition of
difference format Eq. (12.34) is
a < 0 and − 1 ≤ aξ ≤ 0.
If we regard the coefficient in the above difference equations a as the disturbance propa-
gation velocity, a > 0 represents a wave propagating rightward; a > 0 represents a wave
propagating leftward; and the (1/ξ ) = (∆x/∆t) in difference equation represents the
propagation velocity of difference disturbance, then we know that the difference format
Eq. (12.35) is stable for rightward waves and Eq. (12.34) is stable for leftward waves.
Whether rightward waves or leftward waves, the stability condition requires that the prop-
agating velocity of a difference disturbance should not be slower than the propagating
velocity of a differential disturbance, namely

1
|a| ≤
ξ
This is the so-called Courant condition, which is the necessary condition for stability.
12.2.4 Artificial viscosity
For hyperbolic partial differential equations, smoothness of solutions becomes very promi-
nent once a strong discontinuous wave has been formed. The aim to adopt artificial viscosity
is to smoothen the ramp rising edge of a shock wave front, so that a more smooth solution
can be obtained. The effect of artificial viscosity in numerical methods to solve problems
of shock waves will be discussed as follows by an example of 1D governing equations for
nonlinear flow.
In Lagrange coordinates, the conservation equations of mass, momentum, and energy, and
state equation of 1D nonviscous fluid are
∂V 1 ∂v
= (12.45a)
∂t ρ0 ∂X
∂v 1 ∂p
=− (12.45b)
∂t ρ0 ∂X
∂E p ∂v ∂V
=− = −p (12.45c)
∂t ρ0 ∂X ∂t
p = p(E,V ) (12.45d)
where V = 1/ρ is the specific volume, ρ0 is the initial density, p is the pressure, and E is
the specific internal energy. The difference equations of Eq. (12.45) can be taken as
∆t
V k + 12 , j + 1 = V k + 12 ,j + v k + 1,j + 21 − v k,j + 12
ρ0 ∆X
∆t
v k,j + 12 = v k,j − 12 − p k + 12 ,j − p k − 12 ,j
ρ0 ∆X
1
E k + 12 , j + 1 = E k + 12 , j − p k + 12 , j + 1 + p k + 12 , j
2

V k + 2 , j + 1 − V k + 12 , j
1

p k + 12 ,j + 1 = p E k + 12 ,j + 1 , V k + 12 ,j + 1 (12.46)
The last two equations should be solved simultaneously by iteration for E(k + 12 ,j + 1),
V (k + 12 ,j + 1)], and p(k + 12 ,j + 1). This difference format will give a second-order
accurate approximation to a smooth solution.
But, if a shock wave appears, the numerical result is as shown in Fig. 12.11 (Richtmyer and
Morton, 1967). It can be seen that not only the shock wave velocity is inaccurate, but the
numerical solution fluctuates significantly due to the disturbing spurious high-frequency
signals in the region near the shock wave front. However, this method is not unstable,
since the solution remains bounded when ∆X and ∆t decrease with a constant ratio.
20
a2= 0
15 p = 5.0
g = 2.0
p Lf = 0.2236
10
Shock velocity
10.4% too large
1
0 10 20 30
X
Fig. 12.11. Numerical solution according to Eq. (12.46) without artificial viscosity.
To overcome this difficulty, von Neumann and Richtmyer (1950) introduced the so-called
artificial viscosity. The basic idea of their method is to replace pressure p in the original
differential equations with p + Q, where Q is a function of ∂v/∂X (the spatial derivative
of particle velocity with respect to X) and has the form of viscous force. By providing
additional dissipation, Q makes all approximate solutions smooth. In other words, the
sudden change of pressure p is distributed in several grids so that the solution is smoothed.
The Q is suggested in the following form

(α∆X)2 ∂v ∂v
Q=− (12.47)
V ∂X ∂X
where α is a dimensionless coefficient. It has the effect of smearing the shock wave over
an invariable region with a width of about 2–3 ∆X. Because of that,Q is proportional to
the square of the spatial derivative of velocity, its amplitude attenuated rapidly to vanish
away from the shock wave front, so that the Rankine–Hugoniot conditions can be satisfied
across the region of the shock wave (refer to Chapter 7).
After replacing p with p + Q in Eq. (12.45), we can write the corresponding difference
equations. For various values of the coefficient α, the numerical solutions of the difference
equations considering von Neumann–Richtmyer artificial viscosity are shown in Fig. 12.12
g = 2.0
p = 10.0
Lf = 0.3453
p
Shock wave velocity lower 0.05%
0.15% a 2 = 4.0
0.05% p 1
0 X
p
a 2 = 1.7
1
0 X
a 2 = 0.6
1
0 X
Fig. 12.12. Numerical solution of Eq. (12.45) with artificial viscosity.
(Richtmyer and Morton, 1967). It can be seen that with increasing α, the shock wave is
spread over a wider zone and the wave profile is more smooth with less noise.
For shock waves in solids, the influence of selection of artificial viscosity on the numerical
solutions is relatively large. The numerical results obtained for shock waves in solids are
no longer satisfactory even if von Neumann–Richtmyer’s viscosity item (Eq. (12.47)) is
adopted. A common method, which is accepted widely, is to add a linear viscous item,
namely

(α∆X)2 ∂v ∂v β∆X ∂v
Q=− +
V ∂X ∂X V ∂X
where α = 1.5 and β = 0.06 usually.
In addition, the history of material deformation should sometimes be considered when

using finite difference method. This is because the stress tensor of a solid not only depends
on its present state, but also depends on its deformation history. Thereby, the finite differ-
ence method must carry and update the quantities that record or valuate the deformation
(or loading) history of the material. At present, the most common way is by saving the
historic value of the stress (or strain) tensor. Simultaneously, it is often necessary to carry
additional numbers to indicate, for instance, the amount of work hardening that has been
done, or the extent to which the pores in a porous material have been closed, or whether
or not a crack (or spalling) takes place in the media. In these situations, using Lagrange
coordinates is much more convenient. If other coordinate systems are used, the equations
for these quantities must include convective terms, and assumptions should be introduced
such as “continuity of deformation history” and so on.
In the previous discussions on finite difference method, although we only analyzed the
examples for 1D wave problems, the elementary principles and methods are still valid
for study on multidimensional problems without essential difference. This is because
the partial differential governing equations now are established in a multidimensional
space, the corresponding difference equations are also multidimensional. Consequently,
the dimension-number of grids increases, which results in an enormous increase in the
amount of computation.
12.3 Finite Element Method (FEM)
Finite element method has been widely applied in computation of structural mechanics,
solid mechanics, fluid mechanics, and thermodynamics. It is an important method to solve
partial differential equations. Finite element method is developed on the basis of the vari-
ation principle and by adopting the ideas of finite difference formats and combining the
block polynomial interpolation. This combination makes finite element method keep the
advantages of existing variation methods and the flexibility of finite difference method, and
counteract the deficiency of classic variation methods. Therefore, finite element method
is an innovation and development of the classic variation methods (refer to Xu, 1985).
For example, in ABAQUS/Explicit (a widely used commercial finite element software),
the finite difference formats have been adopted to treat the time derivatives in partial
differential equations.
In this section, we will introduce the finite element method briefly; for detailed discussion
refer to the relevant monographs (e.g. Zienkewicz, 1983; Desai and Kundu, 2001; Shorr,
2004, etc.).
12.3.1 Basic procedures of finite element method
In solid mechanics, we usually need to determine deformation or displacement u of a

structure subjected to external loads. As it is difficult to solve displacement of a structure
by analytic methods, approximate numerical methods are generally applied. The kernel of
finite element method is to segment a continuous object into many small regions, namely
finite elements, and based on governing equations and variation method using polynomial
interpolation, the displacement in each element is solved. The main procedures include
the following steps.
Meshing elements (discretization). Meshing an object into many small elements, the
points connecting the elements are named by nodes. Although the size of elements is not
limited, a good meshing must satisfy the requirements of, for instance, the capacity and
speed of the computer used on the one hand and the accuracy of computation required
on the other hand. In general, the following points must be considered: (1) a vertex of
an element is also a vertex of its neighboring element; (2) big obtuse angles and big
ratio of element edges should be avoided; (3) regions where the gradient of displacement
changes intensively must be meshed with fine grids while regions where the gradient of
displacement changes smoothly could be meshed with relatively coarse grids; (4) in order
to reduce the bandwidth of a stiffness matrix so as to reduce the amount of computing, the
absolute value of the biggest difference between the node-numbers of all adjacent nodes
should be possibly small.
With many decades of development of finite element method, many element types
have been built. Selection of element types can be based on characteristics of object
motion and shapes of object structure. For example, for a 1D object [Fig. 12.13(a)],
line element (two-node element) can be selected. For a 2D object [Fig. 12.13(b)], tri-
angle element (three-node element) or quadrangle element (four-node element) can be
applied. For a 3D object [Fig. 12.13(c)], hexahedron element (eight-node element) can be
adopted.
ui+1
ui
l
i i+1
1D object (a) linear element
quadrangle element
triangle
element
2D object triangle and quadrangle element

(b)
3D object (c) hexahedron element
Fig. 12.13. Various types of finite element, (a) 1D element; (b) 2D element; (c) 3D element.
Choosing interpolation functions (shape functions). It is usually assumed that the depen-
dent variables (such as displacement) in each element changes linearly. Mathematically,
this means that the actual distribution of dependent variables (unknown functions) is
depicted by piecewise linear functions, namely dealt with by a piecewise linear inter-
polation. This linear function describes the distribution shape of dependent variables in
elements, and so is named the shape function. Thus, there are only finite parameters
included in shape functions that are needed to figure out. The problem is discretized into
an algebraic problem with finite degrees of freedom. In other words, when the function
values at nodes of an element are determined, the key problem is converted into how to
depict approximately the distribution of this function in the element by using a polynomial
interpolation function.
For example, if the unknown function (dependent variable) is displacement u, then a

polynomial interpolation function can be written as
u = N1 u1 + N2 u2 + · · · + Nm um (12.48)
where uk (k = 1, 2, . . ., m) are displacements at nodes of this element, Nk (k =

1, 2, . . ., m) are indeterminate interpolation functions (shape functions). Taking the linear
element in Fig. 12.13(a) as an example, and assuming the displacements ui and ui+1 in
the ith element have been given, then the displacement function u(x) in this element can
be expressed by
xi+1 − x x − xi
u(x) = ui + ui+1 , xi ≤ x ≤ xi+1
xi+1 − xi xi+1 − xi
Thereby, the problem is simplified to determining displacements at finite nodes only.
Choosing material models. In analyses by finite element method, material models or

constitutive equations are absolutely necessary. Constitutive equations of materials are
usually determined by mechanical experiments and constitutive theories. For numerical
computation relating to large deformation, material models must satisfy the requirement
of relevant constitutive theories, such as irrelevant to coordinate systems, nonlinear geo-
metric depiction of large deformation, and nonlinear physical depiction of constitutive
relationship, etc. For numerical computation related to dynamic response of structures
under explosion/impact loadings, rate-dependence of constitutive equation should also be
considered.
Deducing equations of finite elements. Governing equations of finite element are deduced
based on the variation principle. Energy method and weighted residual method are the two
common methods in the finite element analysis. Equations depicting mechanical behavior
of finite elements can be obtained by using either of these methods. The obtained equations
are named the equations of finite elements, and they can be written into matrix form as
[k]{q} + [m]{q̈} = {Q(t)} (12.49)
where [k] is the stiffness matrix, {q} is the displacement vector, [m] is the mass matrix,
[q̈] is the acceleration vector, and {Q} is the force vector at nodes.
Governing equations and boundary conditions. Our target is to determine the mechanical
response of the whole object under external loadings. Although Eq. (12.49) is valid for
every element, during the deformation process of the structure, the displacement at each
node is required to be kept continuous in order to satisfy the continuity requirement of
the object. Based on this consideration, Eq. (12.49) is assembled to obtain the following
whole governing equations
[K]{r} = {R} (12.50)
where [K] is the whole stiffness matrix, {r} is the vector of assembly displacement at
nodes, {R} is the vector of assembly force at nodes.
The mechanical behavior of the object loaded is also influenced by boundary conditions.
Considering boundary conditions, the whole governing equations Eq. (12.50) are further
modified to

[K]{r̄}
= {R} (12.51)
Solving the set of linear equations. From Eq. (12.51), the displacements can be figured
out, and consequently, the strain, particle velocity, and stress, etc., can be obtained.
The process of finite element numerical analysis is illustrated as follows by solving the
1D wave equations, as an example.
12.3.2 Examples
As an example, we examine a problem of stress wave propagation in a finite elastic bar

under external loading p(t), as shown in Fig. 12.14 (Desai and Kundu, 2001). The motion
equation under 1D condition is (refer to Chapter 2)
∂σ ∂ 2u
= ρ0 2 + f (t) (12.52)
∂X ∂t
L = 30 cm
F4
1 2 t
2 3 3 4
Fig. 12.14. Meshing of a 1D bar for numerical computation (the number in ∆ labels elements, the
◦
number in labels nodes).
where f (t) is the body force uniformly distributed in the 1D bar. The 1D elastic constitutive
equation is
∂u
σ = Eε = E (12.53)
∂X
where E denotes the elastic modulus of the bar material. Combining the above two
equations, we obtain the governing equation with respect to displacement u(X,t)
∂ 2u ∂ 2u
E = ρ0 + f (t) (12.54)
∂X 2 ∂t 2
Meshing the bar uniformly with three line elements (four nodes) is shown in Fig. 12.14.
The length of each element is l. The displacements at the two ends of the ith element are
denoted by ui and ui+1 . The displacement u of the ith element, by the linear interpolation,
can be expressed as
u = Ni ui + Ni+1 ui+1 = [N] {q} (12.55)
where N1 = 1 − s, N2 = s and s = (x − xi )/ l.
Finite element equations can be obtained from virtual work principle. According to this
principle, the mechanical state of the ith element should satisfy

σ δεA dx = F δuA dx + f δuA dx (12.56)
l l l
where δε and δu denote the virtual strain and virtual displacement, A is the cross-section
area, and
∂ 2u
F = −ρ0 (12.57)
∂t 2
According to Eq. (12.55), we obtain

∂u 1 1 ui
ε= = − , = [B]{q} (12.58)
∂X l l ui+1
{u̇} = [N]{q̇} (12.59)
{ü} = [N]{q̈} (12.60)
Substituting Eqs. (12.57)–(12.60) into Eq. (12.56), and considering that the virtual
displacement δ{u} is arbitrary, we obtain

E[B] [B]Adx{q} =
T
ρ0 [N] [N]Adx{q̈} +
T
[N]T {f}Adx (12.61)
l l l
or
[k]i {q}i + [m]i {q̈}i = {Q(t)}i (12.62)
where the stiffness matrix [k], mass matrix [m], and stress matrix at nodes {Q} are

AE 1 −1
[k] = (12.63a)
l −1 1
1
1−s ρ0 Al 2 1
[m] = ρ0 Al [1 − s,s] ds = (12.63b)
0 s 6 1 2

Alf 1
{Q} = (12.63c)
2 1
Next, we need to assemble the equations of finite element, namely to compose the general
governing equations for depicting the response of the whole bar. Considering that there are
only four nodes in this example, for convenience to sum up, we extend these second-order
matrix to fourth-order matrix, namely
⎛ ⎞
.. ..
⎜ . . ⎟
⎜ ··· ki,i ki,i+1 ··· ⎟
[k]i = ⎜ ⎟
⎜ · · · ki+1,i ki+1,i+1 · · · ⎟
⎝ ⎠
.. ..
. .
⎛ ⎞
.. ..
⎜ . . ⎟
⎜ ··· mi,i mi,i+1 ··· ⎟
[m]i = ⎜⎜ ⎟
⎟
⎝ · · · mi+1,i mi+1,i+1 · · · ⎠
.. ..
. .
Then the general stiffness matrix and general mass matrix can be written as [K] =
I3 I3
i=1 [k]i and [M] = i=1 [m]i respectively, namely
⎡ ⎤ ⎡ ⎤
1 −1 0 0 1 −1 0 0
AE ⎢−1 1 + 1 −1 0 ⎥ AE ⎢−1 2 −1 0 ⎥
[K] = ⎣0 =
−1 1 + 1 −1⎦ ⎣0 2 −1⎦
(12.64a)
l l −1
0 0 −1 1 0 0 −1 1
⎡ ⎤ ⎡ ⎤
2 1 0 0 2 1 0 0
ρ0 Al ⎢ 1 2 + 2 1 0 ⎥ ρ0 Al ⎢1 4 1 0⎥
[M] = ⎣0 =
1⎦ ⎣0 1⎦
(12.64b)
6 1 2 + 2 6 1 4
0 0 1 2 0 0 1 2
Introducing
⎛ ⎞ ⎛ ⎞
u1 1
⎜ u2 ⎟ Alf ⎜1⎟
{r} = ⎝ ⎠ , {R} = ⎝1⎠
u3 2
u4 1
then we obtain the general governing equations below
[K] {r} + [M] {r̈} = {R(t)} (12.65)

(4×4) (4×1) (4×4) (4×1) (4×1)
The above equation contains the term of second-order differential of displacement with
respect to time (particle acceleration), of which the finite difference formats is
3 ∆t
{ṙ}t+∆t = ({r}t+∆t − {r}t ) − 2 {ṙ}t − {r̈}t
∆t 2
(12.66)
6 6
{r̈}t+∆t = ({r}t+∆t − {r}t ) − {ṙ}t − 2{r̈}t
∆t 2 ∆t
Then the general governing equations are further reduced to

[K]{r} t+∆t
t+∆t = {R} (12.67)
where
= [K] + 6
[K] [M]
(∆t)2
(12.68)
t+∆t = {R}t+∆t + 6 1
{R} [M] {r}t + ∆t (ṙ)t + (∆t) {r̈}t
2
(∆t)2 3
If the length of the bar is 30 mm (the length of each element l = 10 mm), the cross-
sectional area A = 1 mm2 , the Young’s modulus is E = 62.5 GPa, the density of the bar
is ρ0 = 2.5 × 103 kg/m3 , then the elastic wave velocity is C = 5 km/s, the corresponding
time step can be taken as
l
∆t = = 2 × 10−6 s (12.69)
C
Assume that the bar is in static state initially, namely the initial condition is
u(x, 0) = u̇(x, 0) = ü(x, 0) = 0 (12.70a)
As to the boundary condition, assume that the bar is subjected to a external load of 1 N at
the bar end with the 4th node (X = 30 mm),
f4 (t) = 1 N (12.70b)
while the other bar end with the 1st node (X = 0) is fixed,
u1 (0, t) = 0 (12.70c)
According to the above conditions, the relevant items in the governing equations
Eq. (12.67) are
⎡ ⎤
3 0 0 0
⎢
= 6.25 × 106 ⎣ 0 6 0 0⎥
[K]
0⎦
(12.71a)
0 0 6
0 0 0 3
⎡ ⎤
2 1 0 0
" # 6⎢1 4 1 0⎥
= {R}t+∆t + 6.25 × 10 ⎣
1⎦
R t+∆t 0 1 4
0 0 1 2

−6 4 × 10−12
× {r}t + 2 × 10 {ṙ}t + {r̈}t (12.71b)
3
And according to the stress conditions at nodes (Eq. 12.70b), it is known

⎛
⎞
0
⎜ 0 ⎟
{R}t+∆t =⎝
0 ⎠
(12.71c)
1
Substituting Eqs. (12.69)–(12.71) into Eq. (12.68), we obtain the general matrix
relationship at t = ∆t
⎡ ⎤ ⎛ ⎞ ⎡ ⎤
3 0 0 0 u1 0
6⎢0 6 0 0 ⎥ ⎜ u2 ⎟ ⎢ 0 ⎥
6.25 × 10 ⎣ =
0 ⎦ ⎝ u3 ⎠ ⎣ 0 ⎦
(12.72)
0 0 6
0 0 0 3 u4 1
As u1 (0, t) = 0, namely u1 = 0, Eq. (12.72) can be simplified to

⎡ ⎤⎡ ⎤ ⎡ ⎤
6 0 0 u2 0
6.25 × 106 ⎣ 0 6 0 ⎦ ⎣ u3 ⎦ = ⎣ 0 ⎦ (12.73)
0 0 3 u4 1
From the above equation, the displacement (its unit is m) of each node at t = ∆t can be
figured out
⎛ ⎞ ⎛ ⎞
u1 0
⎜ u2 ⎟ ⎜ 0 ⎟
⎝u ⎠ = ⎜
⎝ 0
⎟
⎠ (12.74)
3
u4 5.33 × 10−8
Substituting Eqs. (12.74) and (12.70a) into Eq. (12.66), the velocity and acceleration of
each node at t = ∆t can be obtained as
⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞
u̇1 0 ü1 0
⎜ u̇2 ⎟ ⎜ 0 ⎟ ⎜ ü2 ⎟ ⎜ 0 ⎟
⎝ u̇ ⎠ = ⎝ 0 ⎠ and ⎝ ü ⎠ = ⎜ ⎝ 0
⎟
⎠ (12.75)
3 3
u̇4 ∆t 0.08 m/s ü4 ∆t 8 × 10 m/s
4 2
As to the strain and stress of each node at t = ∆t, can be figured out according to
Eqs. (12.58) and (12.53). Repeating these steps, we can obtain solutions of Eq. (12.67) at
t = 2∆t, 3∆t. . . and so on. Consequently, all the unknowns required can be figured out.
The mass matrix shown by Eq. (12.63) represents that mass is uniformly distributed in
the bar. In some finite element software, for convenience, mass is usually assumed to be
concentrated at the nodes. Thereby, the mass Alρ 0 of a line element is equally shared by
two nodes, namely

Alρ0 1 0
[m] = (12.76)
2 0 1
This means that the mass matrix is a diagonal matrix when it is concentrated at the
nodes, which facilitates computation. However, better accuracy can be obtained by tak-
ing a uniform mass distribution described by Eq. (12.63). This is shown in the following
example.
In Fig. 12.14, let a bar of length 500 mm be equally meshed into 50 elements (the length
of an element l = 10 mm, totally 51 nodes), the section area A = 1 mm2 , the Young’s
modulus E = 200 GPa, the density of the bar be ρ0 = 8×103 kg/m3 , then the elastic
wave velocity is C = 5×103 mm/ms. Initial and boundary conditions are
u(x, 0) = u̇(x, 0) = ü(x, 0) = 0

u(0, t) = 0, u̇t (L, t) = 1 mm/ms
In principle, we can obtain the solution to this problem following the same way as the
above example. But, due to the node number being 51, the general governing equations is
of 51-order. So a computer has to be used to perform this finite element computation.
The numerical results of particle velocity at t = 0.08 ms are shown in Fig. 12.15 (Desai
and Kundu, 2001). The time steps used in both Fig. 12.15(a and b) are identical, namely
∆t = l/C = 0.002 ms. On comparison, it is clear that the result of the model with
uniformly distributed mass is better than that of the model with concentrated mass. The
comparison between Fig. 12.15(c and d) also shows this trend, where the identical time
step is ∆t = 0.5l/C = 0.001 ms. In addition, Fig. 12.15 shows that the computation
results for ∆t = l/C has higher accuracy than that for ∆t = 0.5l/C.
Because engineering material usually shows more or less viscosity, there are certain damp-
ing effects in structure response. In finite element method, it is usually assumed that
viscous resistance is proportional to particle velocity (which is equivalent to Newton’s
2.0 numerical solution
v (mm/ms)
accurate solutions
t=0.08 ms
1.0
0
1 11 21 31 41 51
(a) Node number

v (mm/ms)
t =0.08 ms accurate solutions

1.0
0
1 11 21 31 41 51
(b) Node number

v (mm/ms)
accurate solutions
t=0.08 ms
1.0
0
1 11 21 31 41 51
(c) Node number
numerical solution
2.0
v (mm/ms)
t=0.08 ms accurate solutions
1.0
0
1 11 21 31 41 51
(d) Node number
Fig. 12.15. Numerical results obtained by finite element method for 1D elastic waves: (a) ∆t = l/C,
mass uniformly distributed; (b) ∆t = l/C, mass concentrated; (c) ∆t = 0.5l/C, mass uniformly
distributed; (d) ∆t = 0.5l/C, mass concentrated.
viscosity law), namely it is proportional to [c]{q̇}, where [c] is a damp matrix. Considering
this damping, governing equations become
[k]{q} + [c]{q̇} + [m]{q̈} = {Q(t)} (12.77)
The determination of damping property of materials is related to the research on rate-

dependent dynamic behavior of materials, which will not be discussed here. Readers can
refer relevant books (e.g. Desai and Christian, 1972; Shorr, 2004).
Appendices
Appendix I Conversion Factors for Stress and Pressure
Pa dyne/cm2 bar atm kgf/mm2 ksi

1 Pa = 1 10 10−5 9.8692×10−6 1.0197×10−7 1.4504×10−7
1 dyne/cm2 = 10−1 1 10−6 9.8692×10−7 1.0197×10−8 1.4504×10−8
1 bar = 105 106 1 0.98692 1.0197×10−2 1.4504×10−2
1 atm = 1.0133×105 1.0133×106 1.0133 1 1.0133×10−2 1.4696×10−2
1 kgf/mm2 = 9.8067×106 9.8067×107 98.067 96.784 1 1.4223
1 ksi = 6.8947×106 6.8947×107 68.947 68.046 0.70307 1
Appendix II Characteristic Method used to Solve Quasi-Linear Hyperbolic

Partial Differential Equations
We will discuss the following second-order quasi-linear hyperbolic partial differential

equation with two independent variables, such as Eq. (2.18) in Chapter 2
∂ 2u 2
2∂ u
= c (A.1)
∂t 2 ∂x 2
where u is the unknown function, t and x are independent variables, the coefficient c2
depends on (∂u/∂t), (∂u/∂x), u, x, and t, namely
c2 = c2 (ut , ux , u, x, t)
where and hereafter, ut and ux are used to denote (∂u/∂t) and (∂u/∂x), respectively.
A2.1 If u satisfies Eq. (A.1) and the values of ut and ux on the curve c(x, t) are given, it is
required to determine under what conditions there are infinite second-order derivatives of u.
511
Along curve c, we have

⎫
∂ 2u ∂ 2u ⎪
dux = dx + dt ⎪
⎬
∂x 2 ∂x∂t
(A.2)
∂ 2u ∂ 2u ⎪ ⎪
dut = dx + 2 dt ⎭
∂x∂t ∂t
where dx and dt are inter-related by the expression of curve c (dx/dt is the tangent slope
of curve c), ux and ut are known. Therefore, Eqs. (A.1) and (A.2) are algebraic equations
set in respect to (∂ 2 u/∂t 2 ), (∂ 2 u/∂x 2 ), and (∂ 2 u/∂t∂x). Only if these equation set are
interdependent, there would exist infinite solutions. In other words, a linear combination
of two arbitrary equations, e.g. the two equations in (A.2), should be identically equal to
the third equation. The linear combination of Eq. (A.2) is
∂ 2u ∂ 2u ∂ 2u
dut + λdux = λ 2
dx + (λdt + dx) + 2 dt
∂x ∂x∂t ∂t
where λ is an indeterminate coefficient. The condition under which the above equation is
identically equal to Eq. (A.1), is
1 c2 0 0
=− = =
dt λdx λdt + dx dut + λdux
From these, we have

2
dx c2
= c2 , dut = du
dt dx x
dt
For hyperbolic equations, c2 is positive, so there are two different real roots. Thus, when
the following characteristic relations are satisfied
dx = cdt (A.3a)
dut = cdux (A.3b)
or
dx = −cdt (A.4a)
dut = −cdux (A.4b)
there exist infinite second-order derivatives of u(x, t). In the above equations, equation
a is called the characteristics differential equation, while the equation b is called the
compatibility condition along the characteristics, or the characteristics differential
equation on the ut − ux plane.
Appendices 513
A2.2 It is easy to prove that in a certain variables domain, if Eqs. (A.3) or (A.4) are
identical equations, then the secondary differentiable function u(x, t) satisfies Eq. (A.1).
In fact, if there exist the following identical equations
dut ≡ ± cdux , dx ≡ ± cdt
since u(x, t) is secondary differentiable, they can be immediately rewritten as

2
∂ 2u ∂ 2u ∂ u ∂ 2u
dt + dx ≡ ± c dt + dx
∂t 2 ∂t∂x ∂x∂t ∂x 2
∂ 2u ∂ 2u 2
2∂ u ∂ 2u
± c ≡ c ± c
∂t 2 ∂t∂x ∂x 2 ∂x∂t
namely, Eq. (A.1) is satisfied. Similarly, it is easy to prove the inverse inference: if the
function u, in addition to satisfying Eq. (A.1), simultaneously complies with Eqs. (A.3a)
and (A.4a), then Eqs. (A.3b) and (A.4b) along the characteristics are identities.
A2.3 In general, Eqs. (A.3) and (A.4) are not ordinary differential equations. In special
cases, if c2 only depends on ux and ut , then Eqs. (A.3b) and (A.4b) are reduced to ordinary
differential equations. If there exists secondary differentiable single-valued solution for
Eq. (A.1) as well, then simultaneously Eqs. (A.3a) and (A.4a) are reduced to ordinary dif-
ferential equations. Geographically, Eqs. a of (A.3) and (A.4) and Eqs. b of (A.3) and (A.4)
are represented by curves on the x–t plane and the ux –ut plane, which are called char-
acteristics on the corresponding planes. Thus, the problem of solving partial differential
equation (A.1) is reduced to the problem of solving ordinary differential equations (A.3)
and (A.4), namely the problem to obtain the characteristics. These are just the problems
discussed in the present book, since c2 only depends on ux , c2 = c2 (ux ).
A2.4 The geographical representation of the above inference is to map the domain G on
the the x–t plane to the domain G on the ux –ut plane, and to establish the correspondence
between the characteristics C on the x − t plane and the characteristics C on the ux –ut
plane, as shown in Fig. A.1. Therefore, the function u, which establishes the correspon-
dence between the intersections of two different series of characteristics on the x–t plane
and on the ux –ut plane, is the solution of Eq. (A.1).
A2.5 By utilizing such correspondence between the characteristics, usually the following
three classes of boundary-value problems can be solved. (1) The Cauchy problem. If ux
t ux
C C' G'
G
o o
X ut
Fig. A.1. The domain G on ux –ut plane is the mapping of the domain G on x–t plane.
t
C
B
0 x
ux
C'
A'
B'
0 ut
Fig. A.2. Cauchy problem.
and ut are given on the non-characteristics curve, AB, then the single-value solutions in the
curved triangle domain ABC, as shown in Fig. A.2 can be obtained, where AC and BC are
characteristics. (2) The Darboux problem or the Cauchy characteristics boundary-value
problem. If ux and ut are given on the characteristics curves AB and AC, which are not of
the same set of characteristics, then the single-value solutions in the curved tetragonum
domain ABCD as shown in Fig. A.3 can be obtained. (3) The Picard problem or the
mixtured problem. If ux and ut are given on the characteristics curves AB and ux (or ut ) is
given on the non-characteristics curve AC, then by utilizing the correspondence between
the characteristics the single-value solutions in the curved triangle domain ABC as shown
in Fig. A.4 can be obtained. Some examples of those three classes of boundary-value
problems have been given in Chapters 2 and 4.
A2.6 The second-order partial differential equation (A.1) is equivalent to the following
first-order quasi-linear partial differential equation set in respect to ut and ux as unknown
t
C
A
D
B
o x
ux C'
A'
D'
B'
o
ut
Fig. A.3. Darboux problem.

Appendices 515
t C
2
4
A
1
3 B
o x
C'
ux
2' 4'
A' 1' 3' B'

o ut
Fig. A.4. Picard problem.
functions [see Eqs. (2.12) and (2.16) in Chapter 2]

⎧
⎪
⎪ ∂ut ∂ux
⎨ − c2 =0
∂t ∂x
(A.5)
⎪
⎪ ∂u ∂u
⎩ t − x =0
∂x ∂t
Starting from Eq. (A.5) the same Eqs. (A.3) and (A.4) can be obtained. Equations
(A.5) and (A.2) can be regarded as four algebraic equations with respect to
(∂ux /∂t), (∂ut /∂t), (∂ux /∂x), and (∂ut /∂x), of which the solutions obtained by the
well-known determinant method are
∂ux ∆1 ∂ut ∆2
= ,··· , =
∂t ∆ ∂x ∆
where

0 1 −c2 0 0 1 −c2 0

1 0 0 −1 1 0 0 0
∆ = , ∆1 = ,···
dx dt 0 0 dx dt 0 dut
0 0 dx dt 0 0 dx dux
If the curve c is a characteristics, then the conditions under which the derivatives of ux
and ut along this curve are indeterminate (or the solutions are infinite), should be
∆ = ∆1 = ∆2 = ∆3 = ∆4 = 0
From the determinant operation, we obtain

dx ± cdt = 0,
dut ± cdux = 0,
which are just Eqs. (A.3) and (A.4). Notice that if the u here means the displacement, then
the ux and ut mean strain ε and velocity v, respectively, and thus Eqs. (A.3) and (A.4)
here are just the Eqs. (2.23) and (2.24) in Chapter 2.
Appendix III An Introduction to Self-Similar Movement
A3.1 When we study the movement of continuum media, sometimes we may encounter
such a kind of movement, if it is described by dimensionless variables then the time variable
t is unseparatively composed with the coordinate variables x, y, z. No matter how t and
x (or y, z) vary independently, so long as such composed dimensionless variables are
identical, the movement is similar. This is the so-called similar variable problem, or
self-similar problem. Such kind of movement is defined by Sedov, L. E. (
1977) as “if all the dimensionless characteristic variables in the movement of continuum
media are the functions of the only dimensionless variable composed of (x/bt δ ), (y/bt δ )
and (z/bt δ ), then such kind of movement is called a self-similar movement, where δ is a
constant and b is a constant with a dimension of LT −δ .”
Because of one independent variable reduction, such problems are much simplified. In the
one-dimensional spatial problems, a partial differential equation is then reduced to an
ordinary differential equation. So long as the spatial distribution of the characteristic
variable of movement concerned at an instant is known, its spatial distribution at any time
can be inferred. So long as how the characteristic variable of movement concerned at a
spatial location varies with time is known, how it varies with time at all spatial locations
can be inferred.
A3.2 The self-similar movement can be judged by the “Π-theorem” of dimensional

analysis.
The “Π-theorem” is stated as following: if a physical relation consisted of n quantities

with different dimensions described as
f (a1 , a2 , . . . , ak , bk+1 , bk+2 , . . . , bn ) = 0
where the first k quantities are basic quantities which are independent of each other, and
the latter n–k quantities are derived quantities, then this physical relation certainly can be
completely expressed by n–k dimensionless quantities, π1 , π2 , . . . , πn−k , namely
f (π1 , π2 , . . . , πn−k ) = 0
Thereupon, we can prove that if all the characteristic quantities of movement are only
dependent on a, b, x, y, z, t, α1 , α2 , . . . , αn , where the dimensions of a and b are inde-
pendent of each other and different from the length dimension L or the time dimension
T , x, y, and z are the spatial coordinates, t the time, ai the i dimensionless constants (no
limit for the number i), then the movement is self-similar. Usually we set the dimensions
of a and b as [a] = ML K T S and [b] = LT −δ , respectively, where δ
= 0, K and S are arbi-
trary numbers, both a and b could be even zero. This proposition can be very easily proved,
namely, by choosing [a], [b], and T as basic dimensions, and then using the “Π-theorem.”
Appendices 517
This proposition can also be simply stated as following: if no characteristic length and
characteristic time are included in all the variables involved to determine the movement,
then the movement is self-similar. Some examples of such self-similar problems have been
given in Chapters 2 and 9.
A3.3 It should be noted that sometimes the self-similar movement is not easy to judge
immediately from a simple dimensional analysis, but a further physical analysis is needed.
For example, as we discussed in Chapter 10, the deflection of a beam under transverse
impact is usually considered as a function of the following variables
w = f (x, t, a, v0 )
where x denotes the coordinate along the beam axis, t the time, a = (EI/ρ0 A0 )1/4 [see
Eq. (10.23) in Chapter 10], v0 the invariable impact velocity. In such case, by using the
“Π-theorem,” we have
% &
w x a2
=f ,
v0 t v0 t v02 t
Obviously, no conclusion of self-similar movement can be found immediately from the

above equation. However, if we notice the physical equation of the problem, Eq. (10.14),
and further introduce a 4 = (EI/ρ0 A0 ) = c02 R 2 as shown in Eq. (10.23), then the basic
physical equation can be rewritten as [see Eq. (10.14a) following Eq. (10.23) in Chapter 10]
∂ 4w ∂ 2w
a4 + 2 =0
∂x 4 ∂t
By introducing a new variable ξ = x/a, we can finally prove that this problem is self-
similar, as discussed in detail in Section 10.3 of Chapter 10.
References
Abramson, H.N., Plass, H.J. and Ripperger, E.A., Stress Wave Propagation in Rods and Beams,
Advance in Applied Mechanics, Vol. 5, eds. H.L. Dryden and T. von Karman, Academic Press,
New York, 1958, 111.
Al-Hassani, S.T.S. and Johnson, W., The Dynamics of the Fragmentation Process of Spherical
Bombs, Int. J. Mech. Sci., 11 (1969) 811.
Alter, B.E.K. and Cautis, C.W., Effect of Strain-Rate on the Propagation of a Plastic Strain Pulse
along a Lead Bar, J. Appl. Phys., 27 (1956) 1079.
Arenberg, D.L., Ultrasonic Solid Delay Lines, J. Acoust. Soc. Am., 20 (1948) 1.
Bathe, K.J., Finite Element Procedures in Engineering Analysis, Prentice-Hall Inc., 1982.
Bejda, J. and Perzyna, P., The Propagation of Stress Waves in a Rate Sensitive and Work-Hardening
Plastic Medium, Archwm Mech. Stosow., 16 (1964) 5.
Bell, J.F., Propagation of Plastic Waves in Solids, J. Appl. Phys., 30(2) (1959) 196.
Bell, J.F., Propagation of Plastic Waves in Pre-Stressed Bar, Nary Contract N6-ONR-243, VIII,
Johns Hopkins Univ. Tech. Rept. No. 5 (1951).
Bianchi, G., Some Experimental and Theoretical Studies on the Propagation of Longitudinal Plastic
Waves in a Strain-Rate-Dependent Material, Stress Waves in Anelastic Solids, eds. H. Kolsky
and W. Prager, Springer-Verlag, Berlin, 1964, 101.
Birch, F., The Effect of Pressure upon Elastic Parameters of Isotropic Solids, according to
Murnagham’s Theory of Finite Strain, J. Appl. Phys., 9 (1938).
Bohnenblust, H.F., A Note of von Karman’s Theory of Propagation of Plastic Deformation in Solids,
NDRC Memo. A-47M (1942), PB. 32180.
Bohnenblust, H.F., Comments on White and Griffis’ Theory of the Permanent Strain in a Uniform
Bar due to Longitudinal Impact, NDRC Memo. A-47M (1942), PB. 32180.
Bohnenblust, H.F., Propagation of Plastic Waves. A Comparison of Report NDRC A-29 and
R. C. 329, NDRC Memo. A-53M (1942), PB. 20275.
519
520 References
Bohnenblust, H.F., Charyk, J.V. and Hyers, D.H., Graphical Solutions for Problems of Strain
Propagation in Tension, NDRC Report A-131 (OSRD No. 1204) (1943).
Bridgman, P.W., The Physics of High Pressure, G. Bell and Sons, London, 1949.
Broberg, K.B., Shock Waves in Elastic and Elastic-Plastic Media, Avhandling Kungl. Tekn Högsk.,
Stockholm, 1956.
Brown, A.F.C. and Edmonds, R., The Dynamic Yield Strength of Steel and Intermediate Rate of
Loading, Proc. Inst. Mech. Eng., 159 (1948) 11.
Butler, D.S., The Numerical Simulation of Hyperbolic Systems of Partial Differential Equations in
Three Independent Variables, Proc. R. Soc. London, A255 (1962) 232–252.
Campbell, J.D., Dynamic Plasticity: Macroscopic and Microscopic Aspects, Mater. Sci. Eng.,
12 (1973) 3.
Chou, P.C. and Hopkins, A.K., eds., Dynamic Response of Materials to Intense Impulsive Loading,
U.S. Air Materials Lab., Wright Patterson AFB, Ohio, 1972.
Chu Chao-Hsiang, Li Yong-chi and Wang Xiao-jun, Numerical Analyses of Spallation for
Steel Plates under Explosive Loading, Appl. Math. and Mech., 2(4) (1981) 353
(in Chinese).
Chu Chaoshiang, Wang Lili and Xu Daben, A Nonlinear Thermo-Viscoelastic Constitutive Equation
for Thermoset Plastics at High Strain-Rates, Proceedings of the International Conference on
Nonlinear Mechanics, (Oct. 28–31, 1985, Shanghai, China), ed. Chien Wei-Zang, Science
Press, Beijing, 92–97.
Clifton, R.J., A Difference Method for Plane Problems in Dynamic Elasticity, Quarterly of Applied
Mechanics, 25 (1967) 97–116.
Clifton, R.J. and Bodner, S.R., An Analysis of Longitudinal Elastic-Plastic Pulse Propagation,
J. Appl. Mech., 31 (1966) 248–255.
Clifton, R.J. and Ting, T.C.T., The Elastic-Plastic Boundary in One Dimensional Wave Propagation,
J. Appl. Mech., 35 (1968) 812–814.
Cole, J.D., Dougherty, C.B. and Huth, J.H., Constant-Strain Waves in Strings, J. Appl. Mech.,
20 (1953) 519–522.
Coleman, B.D. and Noll, W., An Approximation Theorem for Functionals with Applications in
Continuum Mechanics, Arch. Ratl. Mech. Anal., 6 (1960) 355.
Coleman, B.D. and Noll, W., Foundation of Linear Viscoelasticity, Rev. Mod. Phys., 33 (1961) 239.
Courant, R. and Friedrichs, Supersonic Flow and Shock Waves, Interscience Publishers Inc.,
New York, 1948.
Craggs, J.W., Wave Motions in Elastic-Plastic Strings, J. Mech. Phys. Solids, 2(4) (1954) 286–295.
Craggs, J.W., The Propagation of Infinitesimal Plane Waves in Elastic-Plastic Materials, J. Mech.
Phys. Solids, 5 (1957) 115.
Craggs, J.W., Plastic Waves, Progress in Solid Mechanics, Vol. II, eds. I.N. Sneddon and R. Hill,
North-Holland Pub. Co., Amsterdam, 1961, Chap. IV.
Cristescu, N., Dynamic Plasticity, North-Holland Pub. Co., Amsterdam, 1967.
Davies, R.M., Stress Waves in Solids, Appl. Mech. Rev., 6 (1953) 1.
References 521
Davies, R.M., Stress Waves in Solids, Br. J. Appl. Phys., 7 (1956) 203.
De Juhasz, K.J., Graphical Analysis of Impact of Bars Stressed above the Elastic Range, J. Franklin
Inst., 248 (1949) 113–142.
Desai, C.S. and Christian, J.T., eds. Introduction to the Finite Element Method, Van Nostrand
Reinhold, New York, 1972.
Desai, C.S. and Kundu, T., Introduction of Finite Element Method, CRC press, New York, 2001.
Donnell, L.H., Longitudinal Wave Transmission and Impact, Trans. ASME, 52 (1930) 153–161.
Duvall, G.E., Shock Waves in Condensed Media, Physics of High Energy Density, eds. P. Caldirola
and H. Knoepfel, Academic Press, New York, 1971, 7.
Duvall, G.E., Appications, Dynamic Response of Materials to Intense Impulsive Loading,
eds. P.C. Chou and A.K. Hopkins, U. S. Air Materials Lab., Wright Patterson AFB, Ohio,
1972, 481–516.
Duvall, G.E., Shock Waves and Equations of State, Dynamic Response of Materials to Intense
Impulsive Loading, eds. P.C. Chou and A.K. Hopkins, U. S. Air Materials Lab., Wright Patterson
AFB, Ohio, 1972, 89–122.
Duwez, P.E., Preliminary Experiments on the Propagation of Plastic Deformation, NDRC Report
A-33 (1942), PB. 40536.
Duwez, P.E., Physics of Solid-Plastic Flow, J. Aero. Sci., 23 (1956) 435.
Duwez, P.E. and Clark, D.S., An Experimental Study of the Propagation of Plastic Deformation
under Conditions of Longitudinal Impact, Proc. ASTM, 47 (1947) 502.
Duwez, P.E., Wood, D.S., Clark, D.S. and Charyk, J.N., The Effect of Stopped Impact and Reflection
on the Propagation of Plastic Strain in Tension, NDRC Report A-108 (1942), PB. 202725.
Duwez, P.E., Clark, D.S., Wood, D.S. and Bohnenblust, H.F., The Behavior of Long Beams under
Impact Loading, NDRC Report A-216 (OSRD No. 1828) (1943), PB. 18472.
Duwez, P.E., Clark, D.S. and Bonenblust, H.F., The Behavior of Long Beams under Impact
Loading, J. Appl. Mech., 17(1) (1950) 27–34.
Friedman, M.B., Bleich, H.H. and Parnes, R., Spherical Elastic-Plastic Shock Propagation, Proc.
ASCE, 91 (1965) 189.
Goldsmith, W., Impact, The Theory and Physical Behavior of Colliding Solids, Edward Arnold,
London, 1960.
Graff, K.F., Wave Motion in Elastic Solids, Clarendon Press, Oxford, 1975.
Green, A.E. and Rivlin, R.S., The Mechanics of Nonlinear Materials with Memory, Part 1, Arch.
Rat. Mech. Anal., 1 (1957) 1–21.
Griffis, L., The Behavior of Longitudinal Stress-Waves Near Discontinuities in Bars of Plastic
Materials, NDRC Report A-212 (1943), PB. 18476.
Hillier, K.W., A Method of Measuring Some Dynamic Elastic Constants and its Application to the
Study of High Polymers, Proc. Phys. Soc. B, 62 (1949) 701–713.
Hopkins, H.G., Dynamic Expansion of Spherical Cavities in Metals, Progress in Solid Mechanics,
Vol. I, eds. I.N. Sneddon and R. Hill, North-Holland Publ. Co., Amsterdam, 1960, Chap. III.
Hopkinson, J., On the Rupture of Iron Wire by a Blow, Proc. Man. Lit. Phil. Soc., 11 (1872) 40.
522 References
Hopkinson, B., The Effect of Momentary Stresses in Metals, Proc. R. Soc., A74 (1905) 498.
Hopkinson, B., A Method of Measuring the Pressure Produced in the Detonation of High Explosive
or by the Impact of Bullets, Philos. Trans. R. Soc., A213 (1914) 437.
Johnson, W., Impact Strength of Materials, Edward Arnold, London, 1972.
Jones, N., Structural Impact, Cambridge University Press, Cambridge, UK, 1989.
Jones, O.E., Metal Response under Explosive Loading, Behavior and Utilization of Explosives in
Engineering Design, New Mexico Sec., ASME, 1972, 125.
Kaliski, S., Nowaski, W.K. and Wlodarczyk, E., On Certain Closed Solution for the Shock Wave
with Rigid Unloading, Bull. Acad. Pol. Sci., Serie Sci. Technol., 15 (1967) 5.
Karman, T. von, On the Propagation of Plastic Deformation in Solids, NDRC Report A-29
(OSRD No. 365) (1942).
Karman, T. von and Duwez, P., The Propagation of Plastic Deformation in Solids, J. Appl. Phys.,
21 (1950) 987–994.
Karman, T. von, Bohnenblust, H.F. and Hyers, D.H., The Propagation of Plastic Waves in Tension
Specimens of Finite Length. Theory and Methods of Integration, NDRC Report A-103 (1942),
PB. 18477.
Knopoff, L., On Rayleigh Wave Velocities, Bulletin of the Seismological Society of America,
42 (1952) 307.
Kobayashi, A. and Wang Lili, Quest for Dynamic Deformation and Fracture of Viscoelastic Solids,
Ryoin Publishers, Japan, 2001.
Kolsky, H., An Investigation of the Mechanical Properties of Materials at Very High Rates of
Loading, Proc. Phys. Soc., B62 (1949) 676.
Kolsky, H., Stress Wave in Solids, Clarendous Press, Oxford, 1953.
Lee, E.H., A Boundary Value Problem in the Theory of Plastic Wave Propagation, Quart. Appl.
Math., 10 (1952) 335–346.
Lee, E.H., Wave Propagation in Anelastic Materials, Deformation and Flow of Solids,
ed. R. Grammel, Springer-Verlag, Berlin, 1956, 129.
Lee, E.H., Plastic-Wave Propagation Analysis and Elastic-Plastic Theory at Finite Deformation,
Shock Wave and the Mechanical Properties of Solids, eds. J.J. Burke and V. Weiss, Syracuse
Univ. Press, Syracuse, 1971, 3.
Lee, E.H. and Liu, D.T., Influence of Yield on High Pressure Wave Propagation, Stress Waves in
Anelastic Solids, eds. H. Kolsky and W. Prager, Springer-Verlag, 1964, 239.
Lee, E.H. and Symonds, P.S., Large Plastic Deformations of Beams under Transverse Impact,
J. Appl. Mech., 19 (1952) 308–314.
Lee, E.H. and Tupper, S.J., Analysis of Plastic Deformation in a Steel Cylinder Striking a Rigid
Target, J. Appl. Mech., 21 (1954) 63.
Lee, E.H. and Wierzbicki, T., Analysis of the Propagation of Plane Elastic-Plastic Waves at Finite
Strain, J. Appl. Mech., 34 (1967) 931.
Liu Xiao-min and Hu Shi-sheng, Two-Dimensional Numerical Analyses of the Dispersion Effects in
SHPB with Large Diameter Bars, Experimental Mechanics, 15(4) (2000) 371–376 (in Chinese).
References 523
Liu Xiao-min and Hu Shi-sheng, Wave Propagation Characteristics in Cone Bars Used
for Variable Cross-Section SHPB, Explosion and Shock Waves, 20(2) (2000) 110–114
(in Chinese).
Lockett, F.J., Nonlinear Viscoelastic Solids, Academic Press, London, 1972.
Lüdwik, P., Elemente der technologischen Mechanik, Springer-Verlag, Berlin, 1909.
Malvern, L.E., The Propagation of Longitudinal Waves of Plastic Deformation in a Bar of Material
Exhibiting a Strain-Rate Effect, J. Appl. Mech., 18 (1951) 203.
McQueen, R.G., Marsh, S.P., Taylor, J.W., Fritz, J.N. and Carter, W.J., The Equation of State of Solids
form Shock Wave Studies, High-Velocity Impact Phenomena, ed. R. Kinslow, Academic Press,
New York, 1970, 293.
Menkes, S.B. and Opat, H.J., Broken beams, Exp. Mech., 13 (1973) 480–486.
Miklowitz, T., The Theory of Elastic Waves and Wave-guides, North-Holland Publ. Co., Amsterdam,
1978.
Morland, L.W., The propagation of Plane Irrotational Waves through an Elastoplastic Medium,
Philos. Trans. R. Soc. London, A251 (1959) 341.
Morland, L.W., Spherical Wave Propagation in Elastic-Plastic Work-hardening Materials, J. Mech.
Phys. Solids, 17 (1969) 371.
Murnagham, F., The Compressibility of Media under Extreme Pressures, Proc. Nat. Acad. Sci.,
30 (1994).
Neumann, J. von and Richtmyer, R.D., A Method for the Numerical Calculation of Hydrodynamic
Shocks, J. Appl. Phys., 21 (1950) 232–257.
Nonaka, T., Some Interaction Effects in a Problem of Plastic Beam Dynamics, part 1: Interaction
Analysis of a Rigid-perfectly Plastic Beams, J. Mech., 34(3) (1967) 623–630.
Nowaski, W.K., Stress Waves in Non-elastic Solids, Pergamon Press Ltd., Oxford, 1978.
Pack, D.C., Evans, W.M. and James, H.J., The Propagation of Shock Waves in Steel and Lead,
Proc. Phys. Soc., 60 (1948) 1.
Perzyna, P., The Constitutive Equations for Rate Sensitive Plastic Materials, Quart. Appl. Math.,
20 (1963) 321.
Plass, H.J. Jr., A Theory of Longitudinal Plastic Waves in Rods of Strain-Rate Dependent Material,
Including Effects of Lateral Inertia and Shear, Plasticity, eds. E.H. Lee and P.S. Symonds,
Pergamon Press, Oxford, 1960, 453.
Ravichandran, G. and Subhash, G., Critical Appraisal of Limiting Strain Rates for Compres-
sion Testing of Ceramics in a Split Hopkinson Pressure Bar. J. Amer. Cer. Soc., 77 (1994)
263–267.
Rayleigh, L., On Waves Propagated along the Plane Surface of an Elastic Solid, Proc. London Math.
Soc., 17 (1887) 4.
Rice, M.H., McQueen, R.G. and Walsh, J.M. Compression of Solids by Strong Shock Waves, Solid
State Physics, Vol. 6, eds. F. Seitz and D. Turnbull, Academic Press, New York, 1958, 1.
Richtmyer, R.D. and Morton, K.W., Difference Methods for Initial-Value Problems, Interscience
Publishers, New York, 1967.
524 References
Rinehart, J.S., Stress Transients in Solids, Hyper Dynamics, New Mexico, 1975.
Riparbelli, C., On the Time Lag of Plastic Deformation, Proc. 1st Midwestern Conf. on Solid Mech.,
Univ. of Illinois, Urbana, 1953, 148.
Sauerwin, H., Numerical Calculations of Multidimensional and Unsteady Flows by the Method of
Characteristics, J. Comput. Phys., 1 (1967) 406–432.
Selberg, H.L., Transient Compression Waves from Spherical and Cylindrical Cavities,
Arkiv. for Fysik, 5 (1952) 97.
Shorr, B.F., The Wave Finite Element Method, Springer-Verlag, Berlin, 2004.
Simmons, J.A., Hauser, F. and Dorn, J.E., Mathematical Theories of Plastic Deformation under
Impulsive Loading, Univ. of California Press, Berkeley, California, 1962.
Smith, J.C., McCrackin, F.L. and Schiefer, H.F., Stress-Strain Relationships in Yarns Subjected to
Rapid Impact Loading, Part V, Texitile Res. J., 28 (1958) 288–302.
Sparrow, C.M., Table of Integrals, Edward Brothers, Ann Arbor, Michigan, 1934.
Sternglass, E.J. and Stuart, D.A., An Experimental Study of the Propagation of Transient Longitudinal
Deformation in Elastoplastic Media, J. Appl. Mech., 20 (1953) 427.
Tang Zhi-Ping, Chu Chao-Hsiang, Wang Lili and Tian Lan-Qiao, Mechanical Behavior of Epoxy
Resin under High Strain Rates, Second National Conference of Explosive Mechanics, Chinese
Society of Theoretical and Applied Mechanics, Yangzhou, 1981, 4-1-2 (in Chinese).
Taylor, G.I., The Use of Flat-ended Projectiles for Determining Dynamic Yield Stress, I. Theoretical
Considerations, Proc. R. Soc., A194 (1948) 289.
Taylor, G.I., Propagation of Earth Waves from an Explosion, British Official Report RC. 70(1940);
The Scientific Papers of G.I. Taylor, v.1, Mechanics of Solids, ed. G.K. Batchelor, Univ. Press,
Cambridge, 1958, 456.
Taylor, G.I., The Plastic Wave in a Wire Extended by an Impact Load, British Official Report RC.
329(1942); The Scientific Papers of G. I. Taylor, v. 1, Mechanics of Solids, ed. G.K. Batchelor,
University Press, Cambridge, 1958, 467–479.
Taylor, G.I., Fragmentation of Tubular Bombs, The Scientific Papers of G. I. Taylor, v. III,
Univ. Press, Cambridge, 1963, 559.
Timoshenko, S. and Goodier, J.N., Theory of Elasticity, McGraw-Hill, New York, 1951.
Ting, T.C.T., On the Initial Speed of Elastic-Plastic Boundaries in Longitudinal Wave Propagation
in a Rod, J. Appl. Mech., 38 (1971) 441–447.
Ting, T.C.T., Nonexistence of Higher Order Discontinuities Across Elastic/Plastic Boundary in
Elastic-Plastic Wave Propagation, in Plasticity and Failure Behavior of Solids, Memorial
Volume Dedicated to the Late Professor Yuriy Nickolaevich Rabotnov, eds. G.C. Sih,
A.J. Ishlinsky and S.T. Mileiko, Kluwer Academic Pub., Dordrecht, 1990, 115–136.
Tuler, F.R., Butcher, B.M., A Criterion for the Time Dependence of Dynamic Fracture, Int. J. Fract.
Mech., 4 (1968) 431.
Viktorov, I.A., Rayleigh and Lamb Waves, Plenum Press, New York, 1967.
Wang Li-Lih, On Propagation of Elastic-Plastic Waves in Strings, Acta Mech. Sin., 7(3) (1964)
228–240.
References 525
Wang Lili, Tensile Stress Regions Caused by Reverse Plastic Deformation in One Dimensional
Strain Elastic-Plastic Compressional Wave Propagation, Explosion and Shock Waves, 2 (1982)
39–44 (in Chinese).
Wang Lili, Dynamic Fracture of Spherical Vessel under Explosive Internal Pressure Loading, Chem.
Eng. Machinery, 10(2) (1983) 11–17 (in Chinese).
Wang Li-li, A Criterion of Thermo-Viscoplastic Instability for Adiabatic Shearing, Proc. Interna-
tional Symposium on Intense Dynamic Loading its Effects, eds. Zheng Zhemin and Jing-Ting,
Science Press, Beijing, 1986, 359–792.
Wang Lili, Adiabatic Shearing—the Constitutive Instability of Materials Under Impact Loading,
in Progress in Impact Dynamics, eds. Wang Lili, Yu Tongxi and Li Yongchi, the Press of
University of Science and Technology of China, Hefei, 1992, 3–33 (in Chinese).
Wang Lili (Li-Lih WANG), Stress Wave Propagation for Nonlinear Viscoelastic Polymeric Materials
at High Strain Rates, Chinese J. Mech., Series A, 19(1) (2003) 177–183.
Wang Lili, Influences of Stress Wave Propagation upon Studying Dynamic Response of Materials
at High Strain Rates, J. Beijing Inst. Technol., 13(3) (2004) 225–235.
Wang Lili, Chu Chao-Hsiang and Yu Ji-Lin, On Discontinuous Properties of Elastic-Plastic Bound-
aries in Elastic-Plastic Plane Waves Propagation, Explosion and Shock Waves, 3(1) (1983) 1–8
(in Chinese).
Wang Li-li, Bao He-sheng and Lu Wei-Xian, The Dependence of Adiabatic Shear Bending on Strain
and Temperature, J. Physique, Colloque C3, 49 (1988) 207–214.
Wang Li-li and Bao He-sheng, A Strain-Localization Analysis for Adiabatic Shear Band at
Different Environmental Temperatures, Proc. 6th International Conference, Mechanical
Behavior of Materials, eds. M. Jono and T. Inoue, Pergamon Press, Kyoto, Japan (1991) 1,
479–486.
Wang, Li-Lih (Wang Lili), Field, J.E. and Sun, Q., Dynamic Behaviour of Pre-Stressed High Strength
Polymeric Yarns Transversely Impacted by a Blade, Proceedings of Second International Sym-
posium on Intense Dynamic Loading and Its Effects, June 9–12, 1992, eds. Zhang Guanren and
Huang Shihui, Sichuan University Press, Chengdu, China, 1992, 354–359.
Wang Lili and Hu Shi-Sheng, Wang Xiao-Jun, The Attenuation of Plane Shock Waves Propagation
in an Elasto-Plastic Medium, J. China Univ. Sci. Technol., 13(1) (1983) 90–100 (in Chinese).
Wang Lili and Hu Shisheng, The Amplification of Stress Waves Propagated in a Conical Bar,
Journal of Ningbo University (Natural Science & Engineering Edition), 1(1) (1988) 69–78
(in Chinese).
Wang, L.-L., Huang, D.-J. and Gan, S., Nonlinear Viscoelastic Constitutive Relations and Non-
linear Viscoelastic Wave Propagation for Polymers at High Strain Rates, in Constitutive
Relation in High/Very High Strain Rates, IUTAM Symposium, Oct. 16–19, 1995, Noda, Japan,
eds. K. Kawata, J. Shioiri, Springer-Verlag, Tokyo, 1996, 137–146.
Wang Lili (Li-Lih Wang) and Jones, N., An Analysis of the Shear Failure of Rigid-Linear Hard-
ening Beam Under Impulsive Loading, Acta Mech. Sin. (English Series), 12(4) (1996)
338–348.
Wang, L.L., Labibes, K., Azari, Z. and Pluvinage, G., Generalization of Split Hopkinson Bar
Technique to Use Viscoelastic Bars, Int. J. Impact Eng., 15(5) (1994) 669–686.
526 References
Wang Li-li, Lu Weixian, Hu Shisheng and Tang Zhiping, Study of the Initiation and Development
of Adiabatic Shear Bands for a Titanium Alloy Under High Strain Rates, IUTAM Symposium,
Macro and Micro-mechanics of High Velocity Deformation and Fracture, 1985, Tokyo, Japan,
eds. K. Kawata and J. Shioiri, Springer Verlag, Tokyo, 1987, 359–406.
Wang Lili, Pluvinage, G. and Labibes, K., The Influence of Dynamic Constitutive Relation
of Polymer at Impact Loading on the Viscoelastic Wave Propagation Character, Jour-
nal of Ningbo University (Natural Science & Engineering Edition), 8(3) (1995) 30–57
(in Chinese).
Wang Li-li, Shi Shao-qiu, Chen Jiang-ying and Huang De-jin, Studies on ZWT Non-Linear Thermo-
Viscoelastic Constitutive Relation and its Application, Journal of Ningbo University (Natural
Science & Engineering Edition), 13(Supp.) (2000) 141–149 (in Chinese).
Wang Li-li and Wang Yong-gang, The Important Role Stress Waves Played in the Studies on Dynamic
Constitutive Behavior of Materials by SHPB, Explosion and Shock Waves, 24(1) (2005) 17–25
(in Chinese).
Wang Lili and Yang Liming, The Non-Linear Viscoelastic Constitutive Relation of Solid Poly-
meric Materials, in Progress in Impact Dynamics, eds. Wang Lili, Yu Tongxi and Li Yongchi,
The Press of University of Science and Technology of China, Hefei, 1992, 88–116
(in Chinese).
Wang Yong-gang and Wang Li-li, Stress Wave Dispersion in Large-Diameter SHPB and its Manifold
Manifestations, J. Beijing Inst. Technol., 13(3) (2004) 247–253.
Whiffin, A.C., The Use of Flat-ended Projectiles for Determining Dynamic Yield Stress, II, Tests on
Various Metallic Materials, Proc. R. Soc., A194 (1948) 300.
White, M.P. and Griffis, L., The Permanent Strain in a Uniform Bar Due to Longitudinal Impact,
NDRC Report A-71(1942), PB. 18480; J. Appl. Mech., 14 (1947), 337–343.
White, M.P. and Griffis, L., The Propagation of Plasticity in Uniaxial Compression, J. Appl. Mech.,
15 (1948) 256.
Wood, D.S., On Longitudinal Plane Waves of Elastic-Plastic Strain in Solids, J. Appl. Mech.,
19 (1952) 521.
Xu Cui-wei, An Introduction of Computational Methods, Higher Education Press, Beijing, 1985
(in Chinese).
Yang, L.M. and Shim, P.V.W., An Analysis of Stress Uniformity in Split Hopkinson Bar Test
Specimens, Int. J. Impact Eng., 31 (2005) 129–150.
Yu Ji-Lin and Wang Lili (Wang Li-Lih), Chu Chao-Hsiang, Propagation Velocity of Elastic-Plastic
Boundaries with Weak Discontinuities in a Bar, Kexue Tongbao (AMonthly Journal of Science),
27(5) (1982) 574 (in Chinese).
Yu Ji-Lin, Wang Lili and Chu Chao-Hsiang, Basic Properties of Elastic-Plastic Boundaries in Stress
Wave Propagation in a Bar, Acta Mech. Sol. Sin., (3) (1982) 313–324 (in Chinese).
Yu Ji-Lin, Wang Lili and Chu Chao-Hsiang, Determination of Propagation Velocity of Elastic-Plastic
Boundaries in a Bar, Acta Mech. Sol. Sin., (1) (1984) 16–26 (in Chinese).
Zheng Zhe-min and Xie Bo-min, A Suggestion of Numerical Model for Underground Explosion,
Sci. Tech. Research Report No. 00756, Institute of Mechanics, Chinese Academy of Sciences,
May, 1965; in ed. Zheng Zhe-min, Collected Works of Zheng Zhe-min, Science Press, Beijing,
2004, 166–190.
References 527
Zhou Fenghua, Wang Lili and Hu Shisheng. On the Effect of Non-uniformity in Polymer Specimen
of SHPB Tests. J. Exp. Mech., 7(1) (1992) 23–29.
Zhou Fenghua, Wang Lili and Hu Shisheng, ADamage-Modified Nonlinear Viscoelastic Constitutive
Relation and Failure Criterion of PMMA at High Stain Rates, Explosion and Shock Waves,
12(4) (1992) 333–342 (in Chinese).
Zienkewicz, O.C. and Morgan, K, Finite Elements and Approximation, Jone Wiley & Son Inc.,
New York 1983.
528 References
Index
1D strain plane wave, 265, 266, 269, 338, Complex wave regions, 105
343, 345, 351 Condition of mass conservation, 37
Condition of momentum conservation, 38
Absorption, 77, 219, 229, 230, 231, 236 Conjugate points, 125
Adiabatic shear failure, 450 Conjugate relationship, 125, 128, 328
Annealed copper, 429, 431 Constant value regions, 31, 105
Artificial viscosity, 497–499 Constitutive viscous dispersion, 45, 233
Associate theory, 333 Contact condition, 379
Attenuation characters, 239 Continuity condition, 170
Attenuation factor, 229–232, 236, 262 Continuous waves, 29, 31, 99, 116, 120, 131,
324, 325
Ballistic pendulum, 74 Controlled surfaces, 26
Basic assumption Controlled volume, 26, 27
fifth, 403 Convergent wave, 25, 32
first, 400 Conversion factors
fourth, 403 pressure, 511
second, 12, 13, 401 stress, 511
third, 403 Creep, 219, 225, 226, 252
Basic singularity, 170, 181, 182, 184, Creep compliance, 225
185, 195 Critical impact velocity, 26, 422
Bauschinger effect, 160 Cylindrical wave, 2, 5, 337, 339, 340, 347,
Bending Wave Theory, 399, 401 348, 357, 359
Body waves, 456
Boundary point, 173, 236, 238, 242, 243, 245, Darboux problem, 102, 104, 105, 345, 514
246, 343, 481, 482 Decreasing hardening material, 25, 26, 29
Branching point, 193, 195 Diagram method, 148
Bridgman equation, 285, 286, 289, 318 Dilatation wave, 454, 455, 468, 469,
483, 490
Cauchy problem, 513 Direct problem, 3, 126, 129, 132, 237, 238,
Centeredwave structure, 31 239, 242–245, 254
Central spalling, 92, 93 Directional derivative method, 14, 15, 484
Characteristic direction, 15, 482, 484 Dispersion, 45, 229, 231, 233, 406, 409
Characteristic line equation, 15, 108 Dispersion phenomenon
Characteristics differential equation, 260 constitutive viscous dispersion, 45, 233
Characteristics numerical method, 73, nonlinear constitutive dispersion, 45
477, 482 Dissipation, 40, 41, 234, 238, 252, 299, 301,
Coaxial collision, 57–59, 62, 63, 65, 66 324, 361, 498
Cold energy, 291 Distortion wave, 455, 483, 490
Cold pressure, 291 Divergent wave, 25
Cold-rolled low-carbon steel, 429 Domain of dependence, 478, 479, 489, 490,
Complex reflection, 467 494, 495
529
530 Index
Double-wave structure, 31, 39 Head-on loading

Dual-wave formation, 279 interaction of
Dynamic two strong-discontinuous stress waves
compatibility condition, 38, 74, 110, 112, 69–99
171, 199, 201, 213, 235, 261, 389, 442 two weak-discontinuous stress waves
condition, 11, 304, 365, 369, 370, 377, 378, 99–102
380, 383, 390, 412, 437, 448 Head-on unloading, 109, 148–154, 157, 159,
fracture, 5, 70, 71, 87–89, 91, 92, 284, 466 161, 279, 282, 325
criterion, 87, 88, 91 High loading rate or high strain rate, 2
strength, 88, 89 Hooke’s law, 17, 44, 106, 220, 224, 267,
Effective influence domain (EID), 248, 268, 276, 284, 317, 318, 340, 347,
249, 254 453, 470
Effective stress, 271 Hopkinson pressure bar (HPBs), 45–47, 74–77
Elastic beam, 412, 415, 419, 422, 429, 431 Hugoniot
Elastic bending waves, 399, 405, 410, 419 curves, 297, 298, 300, 303–308, 310, 312,
Elastic strain softening, 248 314, 316, 322
Elastic surface waves, 467 elastic limit, 272, 275, 322, 325, 332
Elastic unloading, 106, 107, 109, 168, 19 lines, 41
Elastic unloading hypothesis, 106, 109, 110 Hydrodynamics approximation, 292, 293,
Elastic-plastic boundary, 109, 169, 170, 320–324, 332
180, 183 Hydro-elasto-plastic media, 317, 321, 325
Elastic-plastic loading waves, 22, 98, 121, 153 Hydro-elastoplastic model, 321, 323, 332
Elastic-visco-plastic theory, 256 Hysteresis loop, 219
Elastic-visco-plastic wave theory, 256,
263, 264
Engineering theory, 13, 42 Impact with slide, 377, 379–383
Entropic equations of state, 289 Impact without slide, 377, 381
Equilibrium response, 247–249 Impedance matching, 69
Euler coordinate(s), 8, 27, 401 Implosion, 345
Euler derivative, 8 Impulse criterion, 92
Euler method, 7, 8, 26 Increasing hardening material, 25, 29
Euler volumetric modulus, 285, 287, 288 Indeterminate line method, 14, 15, 484
Excess transverse deflection failure, 450 Initial value problem, 19, 102, 342, 343,
Excess transverse shear strain failure, 450 478, 495
Initialboundary-value-problems (IBVP)
cauchy problem, 18–20, 23, 102, 478,
Finite element method (FEM), 500 513, 514
First kind of inverse problem, 237 picard problem, 18, 20, 23, 102, 104, 105,
First-order partial differential equation, 28, 123, 138, 479, 514, 515
266, 357, 408, 477, 484 Initial-value problem or cauchy problem, 478
Flexible strings, 5, 6, 363, 364 Instantaneous response, 20, 219, 234, 235,
Fracture-delay, 91 248, 257, 261–264, 360, 475
Friction condition, 380 Interior point, 237, 242, 244–246, 343,
Front spalling, 284 480–482
Internal collision, 59, 60
Generalized Internal reflective waves, 59, 112, 119,
maxwell body, 223 150–152, 159, 283, 350
wave impedance, 70, 72 Intrinsic wave velocity, 9
Geometrical dispersion, 233 Inverse problem, 3, 79, 129, 237, 239, 242,
Grüneisen 244, 245, 255, 394
coefficient, 290, 305, 307, 308 Irrotational wave, 269, 455–458, 460, 462, 463,
equation, 289, 292, 309, 310 465, 466, 483
Index 531
Isometric wave, 55, 455–458, 460, 463, 465, Meshing elements (discretization), 500
466, 483 Mie-Grüneisen state equation, 291
Isotropic hardening hypothesis, 159, 160 Migration acceleration, 9
Mixed problem, 20, 343, 479
Kelvin-Voigt body, 220, 226, 227 Mixed problem or picard problem, 479
Kelvin-Voigt model, 221, 224, 226 Mode conversion angle, 463, 465
Kinematic Mode coupling, 458
compatibility condition, 36, 37, 170, 213, Momentum conservation condition, 57, 67, 96,
235, 389, 442 171, 210, 343, 391
condition, 11, 370, 378, 380, 437, 439, 447 Murnagham equation, 287–289, 307,
hardening hypothesis, 160 308, 318
Lagrange Neutral axis, 399–402, 406, 442

coordinates, 8, 389, 401, 497 Neutral layer (surface), 400
derivative, 9 Nonaka’s solution, 450
method, 7, 8 Nondistortion character, 239
velocity, 389 Noninstantaneous response, 219, 518
volumetric modulus, 285, 286
Lame solution, 345 Optimized finite difference method, 475
Lame’s coefficients, 268, 270 Ordinary differential equations (ODEs), 14, 15,
Laterally constrained elastic modulus, 269, 335 251, 476, 477, 482, 513
Lattice kinetic energy, 291 Over stress or extra stress, 257
Lattice potential energy, 291 Overdrive pressure, 303
Linear elastic waves, 3, 4, 6, 17–22, 45, 253,
289, 317, 483
Linear hardening material, 25, 41, 110, 116, Physical plane, 15, 104, 105
171, 215, 271, 350, 376, 385 Plane bending problem, 15
Linear propositional, 238, 239 Plane cross section assumption, 55, 399–403,
Linear visco-elastic body, 220 405, 406, 442
Loading boundary, 167–169, 180, 190–194 Plane cross-section, 6, 55
Loading state, 119, 156, 167, 168 Plastic Bending Waves, 409
Plastic hardening modulus, 31, 271
Local acceleration, 9
Plastic shear modulus, 274, 275, 318,
Local linearization method, 146, 328, 330
349, 350
Longitudinal visco-elastic waves propagating in
Plastic unloading, 276
kelvin-Voigt bars, 227–229
Poisson’s ratio, 267, 461, 465
maxwell bars, 229–230
Polarization conversion, 463
standard linear solid bars, 230–233
Power series approximate method, 132
Longitudinal wave, 2
Primary theory, 6, 13, 42–44
Principle of superposition, 60, 61
Malvern model, 220, 258, 334 Progressive approximation method, 129, 134,
Material coordinate system, 7, 9, 10, 28, 138
29, 292 Pursuing loading, 95, 96
Material coordinates, 8, 10, 338, 389 Pursuing unloading, 109, 110, 126, 127, 131,
Material 135, 136, 138, 141, 151, 153, 157, 163,
constitutive relationship, 169 165, 279, 282, 324, 325
derivative, 9, 10, 234, 252
wave velocity, 9, 10, 35, 196, 292, 295
wave velocity (Lagrange wave velocity) Quasi-linear hyperbolic partial differential
Maxwell equations, 511
body, 220, 223, 226, 227, 233, 235, 257
model, 221–224, 223 Rankine-Hugoniot curve, 297
theorem, 36 Rankine-Hugoniot relationship, 41
532 Index
Rate-dependent stress-strain relations, 3 Shock adiabat, 161, 297, 306, 307

elasto-visco-plastic, 3 Shock adiabatic process, 40
visco-elastic, 3 Shock jump condition, 41, 321
visco-elasto-plastic, 3 Shock secant, 39, 40
Rate-independent nonlinearity, 249 Short duration, 1
Rate-independent theory, 4, 6, 38, 41, 219, 277 Similar variable problem or self-similar
Rayleigh line, 39, 297, 298 problem, 516
Rayleigh surface wave, 6, 469, 472 Simple
Reflection coefficient, 68, 81, 461, 462, 464, torsion wave, 54
466 wave regions, 105
Reflection loading, 68, 70 wave relationship, 19, 21, 22, 26, 121, 124,
Reflection unloading, 68, 70 128, 139, 165
Reflective fracture, 159 waves, 21, 90, 99, 102, 121, 147, 203, 329,
Reflective tensile fracture, 93 332, 345, 475
Region of influence, 479 Singular interface
Relaxation modulus, 223, 225 first-order singular interface, 29
Relaxation spectrum, 223, 224 second-order singular interface, 29
Relaxation time, 222, 223, 247–250, 252 Small deformation, 6, 220
Relaxation-time spectrum, 223 Sound impedance, 21, 57, 58, 61, 62
Residual particle velocity, 111, 113, 116, 118, Space-like curve, 19, 20, 478
149, 282 Spalling, 1, 87, 466
Residual strain, 76, 111, 116, 126, 129, 132, Spalling piece, 87, 89–91
147, 154, 164, 217 Spatial coordinate, 8, 10, 26, 380, 399, 449, 516
Retard yield phenomenon, 255 Spatial
Reverse plastic loading, 276 coordinate system, 7–10, 28
Reverse yield, 159, 160, 276, 279–281, 284, derivative, 8, 498
324, 326, 329, 330 wave velocity, 10, 28, 293, 295, 296, 307,
R-H relations, 293, 295–297, 303, 306, 312 308, 396
Riemann wave, 28 Split Hopkinson pressure bar (SHPB), 75–77,
Riemann’s invariables, 18 237
Rigid unloading boundary, 202, 207, 213 Split Hopkinson visco-elastic bar, 237, 238
Rigid unloading in, Standard linear solid model, 226, 227, 239
finite Bars 207–212 Stationary discontinuous interface, 113, 283
linear elastic-decreasing hardening plastic Straight beam, 400, 433
bars 202–207 Strain-rate-dependent theory, 219
linear elastic-linear hardening plastic bars Stress intensity, 70, 271, 333
200–202 Stress relaxation, 219, 222, 223, 226, 248
linear hardening plastic bars, 197, 200 Stress wave propagation, 475
Rigid unloading of shock wave propagating in Surface waves, 456
semi-infinite bars 212–215
finite bars 215–217 Temporal curves, 22
Rigid-plastic beam, 432 Thermal energy, 40, 291, 292
Thermal equations of state, 289
Scab, 87 Thermal pressure, 291, 292
Scabbing, 87 Time-like curve, 20, 479
Second kind of inverse problem, 237, 239, 394 Timer or flying piece, 74
Secondary plastic zone, 166, 167, 174 Timoshenko equation, 408
Secondary stress waves, 112 Total mode conversion, 463
Second-order partial differential equation, 13, Total reflection, 465
14, 228, 266, 486, 514 Transmission coefficient, 68, 80
Self-similar problem, 372, 375, 377, 516, 517 Transverse inertial effect, 13
Shear wave, 55, 442, 455 Triangle stress pulses, 73
Index 533
Uniformity assumption, 79, 82, 84, Waveform curves, 22

85, 87 Weak discontinuities, 29
Varying boundary problem, 377

Yielding velocity, 23
Velocity plane, 15
Visco-elastic and visco-plastic1d stress
waves, 477 Zhu-Wang-Tang (ZWT) nonlinear visco-elastic
Visco-elastic wave theory, 219 constitutive equation, 247
Visco-elastic-plastic wave theory, 219 ZWT material, 249
Visual velocity, 460, 462 π-theorem, 516

Fundamental of Stress Waves

Uploaded by

Copyright:

Available Formats

Fundamental of Stress Waves

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Fundamental of Stress Waves

Uploaded by

Copyright:

Available Formats

Foundations of Stress Waves

This page intentionally left blank

Author: WANG Li-li (Li-Lih WANG)

Amsterdam • Boston • Heidelberg • London • New York • Oxford

First edition 2007

Copyright © 2007 Elsevier BV. All rights reserved

No part of this publication may be reproduced, stored in a retrieval system

British Library Cataloguing in Publication Data

Library of Congress Cataloging-in-Publication Data

For information on all Elsevier publications

Printed and bound in Great Britain

2 Elementary Theory of One-Dimensional Stress Waves

3 Interaction of Elastic Longitudinal Waves 57

4 Interaction of Elastic–Plastic Longitudinal Waves

4.8 Head-On Unloading 148

5 Rigid Unloading Approximation 197

6 One-Dimensional Visco-Elastic Waves and

6.2.2 Longitudinal visco-elastic waves propagating

7 One-Dimensional Strain Plane Waves 265

8 Spherical Waves and Cylindrical Waves 337

9 Elastic–Plastic Waves Propagating in Flexible Strings 363

10 Elastic–Plastic Waves Propagating in Beams under

11 General Theory for Linear Elastic Waves 453

12 Numerical Methods for Stress Wave Propagation 475

12.2.3 Stability of difference formats 495

YANG Li-Ming (Li-Ming YANG)

In fact, when a part of a surface of a deformable solid is subjected to an external load,

Therefore, to distinguish the dynamic response of structure under explosive or impact

As an example of interaction of longitudinal waves and transverse waves, the theory of

To obtain elementary knowledge of the general three-dimensional theory of stress wave

Elementary Theory of One-Dimensional Stress

2.1 Material Coordinate System and Spatial Coordinate System

f (x, t) = F [X(x, t), t]

F (X, t) = f [x(X, t), t]

Neglecting the subscripts, we obtain the following relationship:

From Eq. (2.6) we have:

and in the material coordinate system is:

2.2 Governing Equations of Longitudinal Waves in Bars in Material Coordinates

before deformation (t = 0) is taken as the material coordinate (Lagrange coordinate).

In the following discussions, the displacement u, strain ε = (∂u/∂X), particle velocity

From Newton’s law we have:

In general the function σ (ε) is continuously differentiable. Introducing

2.3 Characteristic Lines and the Compatibility Relationships along the

Below we discuss further the governing Eq. (2.18).

First, according to the assumption that “stress is a single-valued function of strain”,

which is the characteristic line equation in differential form. An integration of (2.23)

Equation (2.24) regulates a compatible relationship between v and ε on the characteristic

Fig. 2.2. A region G in (X, t) plane is mapped to a region G in (ν, ε) plane.

and (2.20) as below:

If curve C is a characteristic line, the above equations have indeterminate solutions,

dσ = ±ρ0 Cdv (2.26)

2.4 Elastic–Plastic Longitudinal Loading Waves in Semi-Infinite Bars

In this section, we discuss the longitudinal stress waves propagating in a semi-infinite

2.4.1 Linear elastic waves

constitutive relationship (2.14) is simplified as:

By introducing the integration constants ξ 1 , ξ 2 , R1 , and R2 , equations in (2.30a) can be

The constants R1 and R2 are sometimes called Riemann’s invariables.

v(X, 0) = ε(X, 0) = 0, 0<X≤∞ (2.31a)

and the boundary condition:

Along QP v − C0 ε = v(Q) − C0 ε(Q)