ADVANCED MECHANICS OF SOLIDS

A Gentle Introduction
ADVANCED MECHANICS OF SOLIDS
A Gentle Introduction

K.B.M. Nambudiripad

α
Alpha Science International Ltd.
Oxford, U.K.
ADVANCED MECHANICS OF SOLIDS
A Gentle Introduction
424 pages

K.B.M. Nambudiripad
Retired Professor of Mechanical Engineering
National Institute of Technology, Calicut, Kerala;
Retired Dean (Amritapuri Campus)
Amrita Vishwa Vidyapeetham, Amritapuri, Kerala;
Currently Visiting Professor of Mechanical Engineering
Vidya Academy of Science & Technology, Thrissur, Kerala.

Copyright © 2018
A L P H A S C I E N C E I N T E R N AT I O N A L LT D .
7200 The Quorum, Oxford Business Park North
Garsington Road, Oxford OX4 2JZ, U.K.

www.alphasci.com

ISBN 978-1-78332-361-6

E-ISBN 978-1-78332-423-1

Printed from the camera-ready copy provided by the Author.

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 prior written
permission of the publisher.
Dedicated to the Memory of
My Parents and Teachers
 iv

Preface
This is a book on Advanced Mechanics of Solids, but presented at a more elementary level,
and hence the qualifier A Gentle Introduction in spite of the obvious self-contradiction.
This is written mainly for the undergraduate students of Civil, Mechanical and Aerospace
branches of the relatively new APJ Abdul Kalam Technological University of Kerala, some-
times abbreviated as KTU. The look of helplessness writ large on the young innocent faces
of the students is the main motivation for venturing to write this book.
This second course on the crucially important Mechanics of Solids has the reputation for
being an extremely difficult, “impossible” subject. Large scale failures are commonplace.
The students have a good reason to have such a hardened opinion. Engineering students
nowadays are not of high academic calibre as in yesteryears. Far too many engineering
colleges have mushroomed all over the place and, consequently, there is a severe shortage
of experienced faculty members. Universities continue to believe that a final three-hour
‘closed book examination’ of the traditional kind is the only correct method of assessment
in every course (subject).
I had the privilege of ‘teaching’ a batch of 35 young faculty members in two ‘innings’,
first in late December 2016, and again in early January 2017. KTU had realised that the
faculty members chosen to handle this course needed to be given intensive training. ICT
Academy of Kerala had arranged these classes on behalf of KTU. This initiative taken
by KTU is doubtless admirable; there is a real need to have such training sessions. The
experience was an eye-opener for me. The young teacher-students had unanimously asked
me to help them teach from ‘the KTU examination point of view’ and to give them suitable
study material. Apparently they found the standard textbooks “too difficult to follow”. I
am not an admirer of ‘teaching from an examination point of view’, because in effect it is
advising the students to resort to selective learning. That is to say, learning a set of worked
out examples, avoiding all theory, and praying for deliverance. It seems to work, because
this is what is happening!
But is this the right way? By doing so, students do not learn anything worthwhile.
Worse still, they hate the subject and will never, never again, learn this subject. Writing
yet another book will not solve all these problems. But perhaps the faculty members and
the students may probably find this book helpful.
I feel that there must be greater emphasis on tutorial classes and home assignments.
Challenging problems must be worked out by all students with partial help from teachers.
They should be able to consult books. Why shouldn’t they? It is not fair to expect them
to commit to memory long equations with the unfriendly symbols of higher mathematics.
The closed book examination, if there must be one, could only be to test their conceptual
understanding of the key concepts. But I am digressing; this is not the place to articulate
my views on examinations.
This book is, in a manner of speaking, a ‘derivative’ of two of my earlier books Cartesian
Tensors and the Equations of Solid Mechanics and Variational Methods in Engineering. As
v
viii  Preface

these two books are written at a slightly higher level, I have made what I consider to be
appropriate changes to suit the intended audience. I do not know if my decision is in the
best interest of the students. Perhaps my experienced learned colleagues will advise me.
As the first book does not contain solution of stress analysis problems, new chapters are
written to cover this important part.
The book opens with an introductory chapter that discusses the prerequisites expected of
the students and the important fundamental concepts upon which the mechanics of solids is
based. This is followed by a chapter on Special Problems in Bending. After a brief review of
the theory of simple bending, various special cases like unsymmetrical bending and curved
beams are discussed. The next chapter introduces the index notation as a preparatory
material for the nature of stress at a point. The theory of the invariant, symmetric, stress
tensor is the most important topic to be assimilated. Without a mastery of this topic, it
is impossible to learn the mechanics of solids, the theory of elasticity, experimental stress
analysis, and machine design. These topics are borrowed for the most part from my earlier
book, though diluted to some extent. As this chapter is fairly long, some more of the
important material is relegated to the next two chapters. The chapters on strains and
constitutive equations that follow are kept at an elementary level so that the students can
learn these without difficulty.
These are the preliminary chapters in one sense; only the governing equations are set
up. Actual stress analysis problems are solved in the subsequent chapters. In an elementary
book of this kind, only two-dimensional problems can be taken up; advanced problems are all
left out. The chapter on energy methods is challenging more for the author than, perhaps,
for the readers. The power and beauty of energy methods can be seen only when presented
on a large canvas. Unfortunately, such a treatment will be abstract and mathematically
demanding. Many sacrifices are made in a spirit of compromise. Some useful matter is
presented at an elementary level. Torsion of non-circular prismatic bars is discussed in
some detail. I hope that the students will find this chapter useful, interesting, and readable
in spite of the mathematics used.
The level of the book seems to be increasing slowly but steadily as it progresses. This
is, I believe, as it should be. After all, there should be a clear difference in the academic
level of the students before and after taking this course.
I have to be deferential to the wishes of my teacher-students. I have, therefore, included
a large chapter containing about 60 worked out examples. I hope this chapter will please
them as well as the young students. In some places they supplement the material given in
the theory part. A few of the problems are not worked out fully. Not all are numerical
problems, and not all are of the same level of difficulty. In this second course, the emphasis
is not, or ought not to be, on numerical problems.
The place of mathematics in the engineering curriculum is special. It serves two pur-
poses, both vitally important. One is as a tool. We need to acquire the necessary tools
such as, say, the method to find the maximum / minimum. The other, which is even more
important, is that learning mathematics is a mental tonic. It sharpens our brain power; it
helps us in the study of other subjects also.
 vi
Preface ix 

Engineering science courses like this Advanced Mechanics of Solids are not easy; they
should not be. If they are easy, it simply means that they are not intellectually or aca-
demically challenging. We need to read good books, spending time and asking questions.
Questioning is not attacking; it is the expression of an open mind willing to explore other
possibilities. All this can be accomplished only if we enjoy learning. There is a thrill, excite-
ment and joy in learning. When we read a great book written by a master, it is effectively
spending time in his company. Imagine spending some time every day with somebody like
Albert Einstein or Subramaniam Chandrasekhar!
Imitating the saying Child is the Father of Man, let me state that Student is the Teacher.
Yes, my students are / were my best teachers. I am grateful to them. But it is to my teachers
that I owe the most. My early formative years at IIT, Kharagpur were beautiful; I enjoyed
my stay and study there. My teachers at Kharagpur and other places have shown me that
learning is a pleasure.
It was Sri K.J. Veera Raghavan of ICT Academy of Kerala who requested me to ‘teach’ or
lead the training sessions arranged for the student-teachers. I thank him sincerely for giving
me this honour as the resource person. My friends and colleagues have been helpful: “all,
always, in all ways”. My colleagues now at Vidya have been uniformly kind, encouraging,
and highly supportive. Among them special mention must be made about Professors V.N.
Krishnachandran, Sudha Balagopalan, K.V. Leela, Sooraj K. Prabha and N.K. Sudev. The
seniors of the Vidya family, Ers P.K. Asokan (former Chairman), G. Mohanachandran
(Executive Director), Suresh Lal (Finance Director), Dr B. Anil (Academic Director) and
others have always been kind to me. Their very presence and kind words were a constant
inspiration. The real guiding light was, and still is, Dr Gangan Prathap. Sri Sreerag
Srinivasan drew almost all the figures promptly and cheerfully. I thank them all sincerely.
Finally, I thank the publishers, M/s Narosa Publishing House, New Delhi and in particular,
Shri N.K. Mehra, Publisher and Managing Director.
It is unrealistic to expect that there are no mistakes in this book. I shall sincerely
appreciate if the mistakes, if any, are pointed out to me. Comments, criticisms, and sugges-
tions are always welcome. They will certainly be gratefully acknowledged. If some of the
young students and faculty members find this book useful, I shall feel amply rewarded for
the labour of writing this book.

Calicut K.B.M. Nambudiripad

[email protected]
Contents
   PREFACE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii

1 INTRODUCTION 1-1
1.1 WHAT THIS BOOK IS ABOUT . . . . . . . . . . . . . . . . . . . . . . . . 1-1
1.2 WHY SHOULD WE LEARN ADVANCED MECHANICS OF SOLIDS? . . 1-1
1.3 WHAT ARE THE COMPLICATIONS? . . . . . . . . . . . . . . . . . . . . 1-2
1.4 PREREQUISITES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-3
1.5 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-5
1.6 SIMPLIFICATIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-12
1.7 CLOSING REMARKS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-24

2 SPECIAL PROBLEMS IN BENDING 2-1

2.1 SPECIAL PROBLEMS IN BENDING . . . . . . . . . . . . . . . . . . . . . 2-1
2.2 SIMPLE THEORY OF BENDING . . . . . . . . . . . . . . . . . . . . . . . 2-2
2.3 PROBLEMS IN BENDING: FURTHER EXTENSIONS . . . . . . . . . . . 2-5
2.4 UNSYMMETRICAL BENDING . . . . . . . . . . . . . . . . . . . . . . . . 2-12
2.5 SHEAR CENTRE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-14
2.6 CURVED BEAMS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-16

3 INDEX NOTATION 3-1

3.1 INTRODUCTION: WHY INDEX NOTATION? . . . . . . . . . . . . . . . 3-1
3.2 INDEX NOTATION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-1
3.3 KRONECKER DELTA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-3
3.4 AN EXAMPLE: RECTANGULAR ORTHOGONALITY OF AXES . . . . 3-5
3.5 PERMUTATION SYMBOLS . . . . . . . . . . . . . . . . . . . . . . . . . . 3-9
3.6 PERMUTATION IDENTITY (ee − δ IDENTITY) . . . . . . . . . . . . . . . 3-14
3.7 A FEW ILLUSTRATIVE EXAMPLES . . . . . . . . . . . . . . . . . . . . 3-15

vii
 Contents viii
xii  Contents

3.8 A FEW PROBLEMS FOR EXERCISE . . . . . . . . . . . . . . . . . . . . 3-20

3.9 CLOSURE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-22

4 THE STATE OF STRESS AT A POINT 4-1

4.1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-1
4.2 STRESS COMPONENTS AT A POINT . . . . . . . . . . . . . . . . . . . . 4-4
4.3 LAW OF TRANSFORMATION OF THE STRESS COMPONENTS . . . . 4-8
4.4 STRESS TRANSFORMATION EQUATIONS . . . . . . . . . . . . . . . . 4-16
4.5 PRINCIPAL PLANES AND PRINCIPAL STRESSES . . . . . . . . . . . . 4-17
4.6 STATE OF STRESS IN A PRINCIPAL COORDINATE SYSTEM . . . . . 4-22
4.7 TENSOR CHARACTER OF THE STRESS AT A POINT . . . . . . . . . 4-35
4.8 STATES OF STRESS: HYDROSTATIC AND PURE SHEAR . . . . . . . 4-38
4.9 THEORIES OF FAILURE: TRESCA’S AND VON MISES’ CRITERIA . . 4-40
4.10 CLOSING REMARKS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-46

5 MORE ABOUT STRESSES 5-1

5.1 DIFFERENTIAL EQUATIONS OF EQUILIBRIUM . . . . . . . . . . . . . 5-1
5.2 SOME CONCEPTS RELATED TO 2-D STRESS FIELDS . . . . . . . . . 5-16
5.3 STRESS AT A POINT: GEOMETRIC VISUALISATION . . . . . . . . . . 5-22
5.4 LAMÉ’S ELLIPSOID OF STRESSES . . . . . . . . . . . . . . . . . . . . . 5-22
5.5 CAUCHY’S STRESS QUADRIC . . . . . . . . . . . . . . . . . . . . . . . . 5-29
5.6 MOHR’S CIRCLE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-32

6 ANALYSIS OF STRAIN AT A POINT 6-1

6.1 GENERAL INTRODUCTORY REMARKS . . . . . . . . . . . . . . . . . . 6-1
6.2 TO TRACK THE DEFORMATION . . . . . . . . . . . . . . . . . . . . . . 6-1
6.3 A SIMPLIFIED APPROACH FOR THE LINEAR THEORY . . . . . . . . 6-5
6.4 STRAINS AND STRAIN TRANSFORMATIONS . . . . . . . . . . . . . . 6-9
6.5 FURTHER BEYOND . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-12

7 CONSTITUTIVE EQUATIONS 7-1

7.1 HOOKE’S LAW AND THE GENERALISED HOOKE’S LAW . . . . . . . 7-1
7.2 AN ALTERNATIVE APPROACH . . . . . . . . . . . . . . . . . . . . . . . 7-2
7.3 ISOTROPIC MATERIALS . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-10
ix Contents
Contents xiii 

7.4 NUMERICAL VALUES OF SOME IMPORTANT PROPERTIES . . . . . 7-17

7.5 ANISOTROPY AND ORTHOTROPY . . . . . . . . . . . . . . . . . . . . . 7-18

8 GENERAL CONSIDERATIONS 8-1

8.1 COORDINATE SYSTEMS . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-1
8.2 SIMPLE TENSION TEST . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-4
8.3 SAINT-VENANT’S PRINCIPLE . . . . . . . . . . . . . . . . . . . . . . . . 8-5
8.4 PRINCIPLE OF SUPERPOSITION . . . . . . . . . . . . . . . . . . . . . . 8-6
8.5 UNIQUENESS THEOREM . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-8
8.6 SOME MISCELLANEOUS TOPICS . . . . . . . . . . . . . . . . . . . . . . 8-12
8.7 SOME ELEMENTARY CONCEPTS REVISITED . . . . . . . . . . . . . . 8-13
8.8 FROM THEORY TO PRACTICE . . . . . . . . . . . . . . . . . . . . . . . 8-16

9 TWO-DIMENSIONAL PROBLEMS 9-1

9.1 AIRY’S STRESS FUNCTION IN RECTANGULAR COORDINATES . . . 9-3
9.2 AIRY’S STRESS FUNCTION IN POLAR COORDINATES . . . . . . . . . 9-12
9.3 ROTATING DISCS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-24
9.4 THICK CYLINDERS: LAMÉ’S PROBLEM . . . . . . . . . . . . . . . . . . 9-27

10 ENERGY METHODS 10-1

10.1 STRAIN ENERGY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-2
10.2 STRAIN ENERGY: SOME COMMON STRUCTURAL ELEMENTS . . . 10-4
10.3 LOOKING AHEAD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-14
10.4 SIMPLER ASPECTS OF ENERGY METHODS . . . . . . . . . . . . . . . 10-16
10.5 UNIT-LOAD METHOD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-21

11 TORSION OF NON-CIRCULAR BARS 11-1

11.1 TORSION OF BARS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11-2
11.2 A QUICK REVIEW OF COULOMB’S THEORY . . . . . . . . . . . . . . 11-2
11.3 SAINT-VENANT’S THEORY . . . . . . . . . . . . . . . . . . . . . . . . . 11-7
11.4 TORSION OF A BAR OF RECTANGULAR CROSS-SECTION . . . . . . 11-11
11.5 ANALOGIES IN TORSION . . . . . . . . . . . . . . . . . . . . . . . . . . . 11-13
11.6 HOLLOW THIN-WALLED SECTIONS . . . . . . . . . . . . . . . . . . . . 11-21
 Contents x
xiv  Contents

12 FIELD EQUATIONS OF THE THEORY OF ELASTICITY 12-1

12.1 FUNDAMENTAL EQUATIONS . . . . . . . . . . . . . . . . . . . . . . . . 12-1
12.2 DIFFERENTIAL EQUATIONS OF EQUILIBRIUM . . . . . . . . . . . . . 12-3
12.3 STRAIN-DISPLACEMENT RELATIONS . . . . . . . . . . . . . . . . . . . 12-3
12.4 STRESS-STRAIN RELATIONS . . . . . . . . . . . . . . . . . . . . . . . . 12-10
12.5 SOME COMMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-12
12.6 DISPLACEMENT FORMULATION . . . . . . . . . . . . . . . . . . . . . . 12-13
12.7 BELTRAMI-MICHELL EQUATIONS . . . . . . . . . . . . . . . . . . . . . 12-15
12.8 BOUNDARY CONDITIONS . . . . . . . . . . . . . . . . . . . . . . . . . . 12-18

13 A FEW ILLUSTRATIVE EXAMPLES 13-1

14 EARLY HISTORY 14-1

14.1 ANCIENT TIMES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-1
14.2 SIXTEENTH / SEVENTEETH CENTURY: GALILEO . . . . . . . . . . . 14-2
14.3 SEVENTEENTH CENTURY . . . . . . . . . . . . . . . . . . . . . . . . . . 14-4
14.4 EIGHTEENTH CENTURY . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-8
14.5 SUBSEQUENT DEVELOPMENTS . . . . . . . . . . . . . . . . . . . . . . 14-11
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B-1
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . I-1
Chapter 1

INTRODUCTION

Let me offer with humility salutations1 to all my teachers, most of all to the One Real
Innermost Guru, the Guru of all gurus, the real source of everything beautiful and sublime.

WHAT THIS BOOK IS ABOUT

This book is a gentle introduction — an elementary treatment in spite of the obvious
contradiction — to Advanced Mechanics of Solids2 . Mechanics of Solids is arguably the most
important course (subject) for Civil, Mechanical and Aerospace engineers. This is really a
modern version of the traditional Strength of Materials. Our subject is of crucial importance
in Design, Stress Analysis (both experimental and theoretical), and Finite Element Method.
WHY SHOULD WE LEARN ADVANCED MECHANICS OF SOLIDS?
Engineering students, particularly of the three afore-mentioned branches, invariably take
a first course in the mechanics of solids. This is usually at an elementary level, and is
not sufficient at all for a proper understanding. Hence a follow-up course is essential. The
subject matter of this book aims at meeting two objectives: (i) a more mature understanding
of the elementary concepts, and (ii) at least a peep into advanced topics. There are some
technically important problems that cannot be solved by using the theory studied in the
traditional elementary course. Special problems in bending, stresses (and deformations) in
a rotating disc, stress concentration, contact stresses, bending stresses in shells, torsion of
non-circular cross-sections, and assessment of possible failure of a crack (fracture mechanics)
are some of these non-elementary problems that fall outside the scope of a first course3 .

1
It is an Indian tradition to remember with pleasure and gratitude one’s teachers and to offer one’s saluta-
tions. Let me too follow this great tradition.
2
An elementary treatment of Advanced Mechanics of Solids?
3
It is not necessary, nor is it possible for a practising engineer, to be able to obtain solutions for these difficult
problems. It is quite sufficient if he has a sound knowledge and understanding of the fundamentals, and
to be aware of the limitations. If these conditions are met, he will know when he has to consult experts.
Advanced Mechanics of Solids 1-2

We know that these are extremely important technical problems. No engineer can afford
to be totally ignorant of these topics. In addition, there are some fundamental issues to be
settled: are the stresses in a body independent of the material, can there be two different
solutions for the same stress analysis problem, how do we deal with nonlinearity, and how
can we calculate long term creep effects? There are a large number of questions that cannot
be answered within the framework of an elementary first course in mechanics of solids. Even
more important is the need to have greater conceptual clarity. Hence it is essential to have
a second course in the mechanics of solids.
The trouble, however, is that these problems are difficult to solve. They demand a
higher level of maturity and understanding for their solution. The level of mathematics
needed is quite high.
WHAT ARE THE COMPLICATIONS?
Advanced Mechanics of Solids demands a higher or better level of conceptual clarity. Some
of the fundamental notions like stress at a point are abstract; it takes time to absorb and
digest some concepts like invariance, stress tensor, transformations of coordinates and the
induced transformations. These are technical words pregnant with meaning. To discuss
these matters it is desirable (if not essential) to use the index notation (sometimes called
the tensor notation) and concepts from tensors analysis (at least cartesian tensors). A
higher level of mathematics is also needed.
In addition, before we solve a stress analysis problem — stress analysis is said to be the
‘centre of gravity’ of the mechanics of deformable bodies — it is necessary to formulate it.
Often it turns out to be a boundary value problem which is almost the same as a problem in
partial differential equations. The techniques of solution are based on higher mathematics.
It is necessary, when formulating a problem, to change from volume integrals to surface
integrals, and surface integrals to line integrals, and vice versa. To be able to do so, we need
to use Gauss’ and Stokes’ theorems. Although all students nowadays learn these integral
theorems, they are still not comfortable with them. Part of the reason is that these theorems
are generally taught by mathematicians without the support of the physics of the problem.
The physical significance is practically never brought out and discussed. Maximisation
/ minimisation under constraints using Lagrange multipliers is another technique that is
often used. Students seem to have a hardened opinion that mathematics is really useless
for engineers, and that it is only of nuisance value. This seriously mistaken notion is to be
corrected. One cannot get very far in engineering without the support of mathematics.
Such an attitudinal change — with a healthy respect and love for mathematics — is
necessary before one undertakes the study of serious mathematical-analytical courses (sub-
jects) like advanced mechanics of solids. We need to learn several techniques in mathematics;
these tools are always very useful. Furthermore, learning mathematics enhances our ability
to learn other subjects too. Learning is a joy. Enjoy learning!

This situation is not unlike the condition in the primary health centres in rural areas. It is quite sufficient
if the non-specialist general physicians there can identify complications and advise the patients to consult
specialist doctors.
1-3 Introduction

PREREQUISITES
This being an ‘advanced’ course, students need to be proficient in the first course on the
mechanics of solids. A superstructure cannot be built on unsound foundations. It is quite
possible that because of various reasons, some students may have serious gaps and deficien-
cies in the first course. It is essential to remedy these weaknesses, and to have a fairly good
mastery of the subject. There are several excellent books for a first course. Den Hartog
[3],[4], Popov [11], Timoshenko [15], [16] and several other books by him, and Crandall &
Dahl (and others) [2] are excellent4 .
Students are advised to revisit all the topics of their undergraduate syllabus, and be
prepared for an advanced course. Here below are a few comments about some of the topics.

(i) Bending moment, shearing force, axial thrust and twisting moment dia-
grams: Bending moment, shearing force, axial thrust, and twisting moments and
their variations along the length of a (one-dimensional) bar are traditionally taught as
part of Strength of Materials or Mechanics of Solids. Actually these are part of statics
(unless the problem is statically indeterminate).
The beams considered are almost always horizontal. Students tend to classify struc-
tural elements as beams and columns depending on whether they are horizontal or
vertical! It is dangerous to give a large number of similar problems as exercises, be-
cause in course of time students tend to work out the problems mechanically (with
the brain in the switched off mode)! It is necessary to expose the students to different
looking structures. Airplane structures provide a wide variety of such elements.
A simply supported beam with a concentrated moment, a log of wood floating on
water, an L-shaped member, a balcony beam (horizontal in plan), a semi-circular
beam, etc. are examples where students seem to have difficulty. When problems are
assigned, there must be variety, so that every time the students are required to think
and proceed with the calculations. Some examples are shown in the figures below.
Note that the shear force diagram will have a jump discontinuity at the point where
a concentrated force is applied, the magnitude of the jump being exactly equal to the
magnitude of the force applied. A similar remark is applicable for the other force
resultants like an axial force, a bending moment, and a twisting moment also. The
case of an L-shaped member should receive special attention.

(ii) Love-Kirchhoff assumption in the theory of simple bending: This and its
consequences are important. These are discussed later in the book.

(iii) Deflection of beams: It is very convenient to use six simple formulae to obtain the
deflection of beams in many, if not most, cases5 . We may recall the expressions for
the slope and the deflection of a simple cantilever given below.
4
My own personal likes and prejudices may have influenced this choice of mine. There is no doubt at all
that these are all highly respected authors with their affiliations to MIT, U. of C. (Berkeley), Stanford,
Advanced Mechanics of Solids 1-4

(a) A beam with a moment (b) A wing of an aircraft (c) An inclined load on an L-bar

Figure 1.1: Some cases to be considered. The load P on the L-bar is not perpendicular to
the member BD. It is not vertical; it is not in the vertical plane either.

Ml P l2 wl3
End slope
1 EI 2 EI 6 EI
M l2 P l3 wl4
End deflection
2 EI 3 EI 8 EI

It is easy enough to remember this: 1 2 2 3 6 8, simpler than a telephone number!

Everything else can be written “in each case by dimensional reasoning”. With a
little practice, the final formula can be written down effortlessly as if the formula is
reproduced from memory!
We shall demonstrate the procedure by just one example6 . A cantilever loaded by two
concentrated forces, P1 at C and P2 at B is shown in Fig. 1.2. We desire to calculate
the end deflection. First we shall solve this by superposition. The total deflection at
B = the deflection at B due to the load P1 alone plus the deflection at the same point
B due to the load P2 alone. This gives us

P1 a3 P1 a2 P2 (a + b)3
 
δB = + b+
3 EI 2 EI 3 EI

We may also arrive at the answer differently [Fig. 1.2b]. We note that (i) the shear
force immediately to the left of C is P1 + P2 , (ii) the bending moment at C is P2 b, and

and MIT, respectively.

5
Den Hartog [3] calls this the Myosotis method. He gives this testimony: “After the alphabet and the
tables of multiplication, nothing has proved quite so useful in my professional life as these six expressions,
and undoubtedly some of my students will have the same experience.” See [3] for more details.
6
This example is borrowed from Den Hartog [3].
1-5 Introduction

(a) Deflection at B: superposition (b) Deflection at B: alternative method

Figure 1.2: The deflection at the end B of a cantilever is calculated in two different ways,
first by superposition [Fig. 1.2a], and later differently [Fig. 1.2b].

that (iii) the right part CB of the cantilever can be considered as fixed at an angle
downward. Thus, the total deflection at B can also be calculated as:
(P1 + P2 ) a3 (P2 b) a2 (P1 + P2 ) a2 (P2 b) a P2 b2
     
deflection at B = + + + b+ .
3 EI 2 EI 2 EI EI 3 EI

(iv) Coulomb’s theory of torsion:

This theory is developed with the help of two assumptions: (i) cross-section do not
warp when the twisting moment is applied; and (ii) straight radial lines on the cross-
sections remain straight. That is, the cross-sections do not distort in their own planes7 .
Actually these are not assumptions, but facts. These can be proved by an argument
of rotational symmetry. See Den Hartog [3].

INTRODUCTION
This is a revisit, and not a first time introduction to our subject of Mechanics of Solids.
Still a fairly detailed introduction is given here because most students do not seem to have
a solid foundation.
Opening Remarks
Our subject took shape in the first half of the nineteenth century under the influence of
a strong group of French scientists, largely mathematicians, such as Poisson, Lamé and
Navier8 . Their influence is still sufficiently strong and compelling9 that the original name
7
Straight radial lines, on application of the twisting moment, do not curve in or curve out. Using the
terminology in cricket, the radial lines are neither in-swingers nor out-swingers!
8
Gabriel Lamé (July 1795 - May 1870) was an outstanding French mathematician and engineer. Siméon
Denis Poisson (June 1781 - April 1840) was a French mathematician and physicist. Claude-Louis Navier
(February 1784 - August 1836) was a French engineer, mathematician and physicist. They were all
eminent as engineer-mathematician-scientists. It seems natural that they all studied in the prestigious
École Polytechnique, and that their names are among the 72 great names of engineers, mathematicians
and physicists written in gold letters on the Eiffel tower. See S.P. Timoshenko [14].
9
Readers are advised to read the beautifully written book [3] by the well known author, J.P. Den Hartog.
Advanced Mechanics of Solids 1-6

résistance de matériaux 10 continues to be used by engineers and academicians. In recent

times, however, the approach has changed, dictated mainly by the technological needs of
aerospace applications. The new name that the subject has now taken on, reflecting a more
modern approach, is Solid Mechanics. Mechanics of Materials is another popular name
that is, perhaps, a compromise name accommodating a little bit of both approaches and
points of view.
The traditional name of strength of materials is misleading. It does not deal with the
‘strength’11 of materials. It is the allied subject of Mechanical Testing of Materials that
deals with this aspect. It supplements our subject by giving experimental support to it
principally by providing the numerical values of the various parameters that describe the
mechanical properties of the various materials used. Theory of Elasticity is another re-
lated subject. The approach in this is more mathematical. Nevertheless, in the practical
applications of all these subjects, these are more or less the same. In every case, the prob-
lem is to find out the internal state of a body (stresses, strains and displacements, defined
later) when the geometry (shape) and the composition (material or materials) of the body
and the applied loads are specified. There are other problems also that are addressed by
these subjects. Strength of materials is a simplified engineers’ theory which is empirical in
some places. Theory of elasticity discusses the same problems, but in a more mathemat-
ically thorough style, with frequent excursions into mathematical proofs and (pointless?)
arguments12 . Solid mechanics is the ‘in-between’ subject, concentrating primarily on the
practical applications, yet partially accommodating what elasticians have to say. Thus, to
gain solidity and maturity in this subject, further follow-up courses on Structural Analysis,
Advanced Solid Mechanics, Theory of Elasticity, Theory of Plates & Shells, Continuum Me-
chanics, Experimental Stress Analysis and Computational Mechanics 13 are recommended.
It would also be nice to be bilingual, that is, to be comfortable in the different languages
used by engineers and mathematicians. To learn advanced theory, it is necessary to learn
Tensor Calculus, Differential Geometry, Abstract Algebra and Functional Analysis.
Our subject is of great relevance to Mechanical, Civil and Aerospace Engineering. Next
to Mechanics (Statics and Dynamics of Rigid Bodies), this is arguably the engineering
science subject most relevant to these branches of engineering. Investment of time and
10
“The first mathematician to consider the nature of the resistance of solids to rupture”, writes A.E.H.
Love, a towering figure in the theory of elasticity, “was Galileo”. The first edition of this celebrated
book was published in 1892 - 1893. He goes on to write: “In the history of the theory started by the
question of Galileo, undoubtedly the two great landmarks are the discovery of Hooke’s law in 1660, and
the formulation of the general equations by Navier in 1821”. Ambitious students are advised to read the
history of this subject. Love’s book gives a 31- page historical introduction.
11
‘Strength’ as used here is somewhat ambiguous. It is not only the material, but perhaps even more
importantly, the proper shape and design that contribute to ‘strength’. Surprisingly, there are cases where
the ‘strength’ of a machine part is actually increased if some material is removed! In any case, a deeper
probe into these issues is necessary.
12
Marsden, J.E. & Hughes, T.J.R.: Mathematical Foundations of Elasticity, Courier Corporation, (1994)
state that, among “the three things that every beginner in elasticity theory should know”, “the first is that
Kirchhoff has two h’s in it”, · · · , and “the third is that researchers in elasticity theory are very opinionated,
even when they are wrong.”
13
Fracture Mechanics, Composites, Theory of Plasticity, etc. are also useful.
1-7 Introduction

energy in this crucial subject in the form of clarity of concepts and understanding is sure
to pay rich dividends in future. This subject is part of mechanics; in fact, it can be
appropriately called the Statics of Deformable Elastic Bodies 14 .
Mechanics and Its Various Branches
Mechanics is one branch that received attention even in ancient times. By the time of
Archimedes15 , it had developed considerably. Mechanics is divided into Mechanics of Flu-
ids and Mechanics of Solids. While fluid mechanics is important, we are concerned with
only solid mechanics now. This, in turn, is sub-divided into Mechanics of Rigid Bodies and
Mechanics of Deformable Bodies, which are both important in engineering. We shall first
take up mechanics of rigid bodies before considering the latter.
Rigid bodies:
A body is called rigid if it does not undergo any deformation (change in shape or volume),
no matter how large the applied force is. This means that the distance between any two
points of the body remains unchanged, irrespective of how large the applied load is. A
rigid body is a mathematical abstraction or an idealisation16 . Perfectly rigid bodies do not
exist; all real bodies undergo some deformation when a load is applied. Sometimes the
deformation may be so small, or it may be irrelevant, that we are justified in neglecting it,
and treating the body as rigid.
Particle:
Often the body can be considered as a particle. In this idealisation, the entire mass of the
body (which is really distributed throughout the volume of the body) is regarded as a point
mass located at the centre of mass. The body (which actually has finite dimensions) is
treated as if it were just a particle concentrated at a single point (occupying no volume).
Deformable body:
There are situations where the deformations of a body are of prime concern to us in our
analysis. In such cases, we do not get any result if the body is regarded as a particle or as
a rigid body. The body then is to be treated as deformable. The question is not whether a
body is truly deformable; all real bodies are deformable to varying degrees.
Which among them is correct?
A particle, a rigid body and a deformable body17 are all various simplifications, or ideali-
sations, to solve a problem. The scope of the problem and the method of solution decide
which among these is appropriate. For example, let us say we are analysing the path of
a satellite. If we are interested only in the path of the centre of mass, how long it will

14
If the need arises, the scope can be extended to encompass Dynamics also. Furthermore, plastic, vis-
coelastic, viscoplastic, thermoelastic, and indeed a host of other special cases can also be discussed using
similar methods. In such an event, one would need more advanced concepts and methods which cannot
be adequately covered in a first introductory course.
15
Archimedes of Syracuse (287 BC - 212 BC) was a renowned Greek mathematician, astronomer, physicist,
engineer and inventor, all rolled into one. He was born and assassinated in Syracuse, Italy.
16
Ideal in the sense of being hypothetical, and not in the sense of being the most desirable.
17
There is no simplification, or idealisation, in treating a body as deformable; this is really the case. What
is intended is that these are all models meant to solve a given problem for the limited purpose of obtaining
the desired answer in a given context.
Advanced Mechanics of Solids 1-8

take to reach a certain altitude, etc., it is sufficient to treat it as a particle. The governing
equations and their solution will now be relatively simple. The price that we pay is that
this idealisation, or this model, gives us only partial information. We would know all about
the motion of the centre of mass, but we would have no information at all of the motion
of the body about (that is, the motion of other points relative to) the centre of mass. We
would not be able to know anything about the orientation and the rotation of the body.
If, on the other hand, we need additional information on the motion of the body about
the centre of mass, the same body, the very same body, has now to be treated as a rigid
body (and not as a particle any more). If, additionally, we wish to predict the stresses
inside, and estimate its safety and structural integrity, the same body has to be treated
(modelled) as a deformable body. A rigid body model can give no information about the
internal states of stress in exactly the same way as a particle model is unable to provide
information on the motion of the body about the centre of mass.
In some problems, there is a strong interaction, or coupling, between the forces acting on
a body and its shape. An example is a falling rain drop. The aerodynamic forces on the rain
drop are strongly influenced by its shape; the shape, in turn, of such a highly deformable
body is decisively influenced by the forces. A far better and technically important example
is the flutter of aircraft wings. The aerodynamic force causes bending of the wings; this
bending is accompanied by twisting of the wings. The twist changes the angle of attack
of the fluid stream, thereby changing the flow pattern and, thus, the aerodynamic force.
Problems with such strong cross-effects (coupling) are difficult to solve. In many problems,
however, the coupling is not very strong. It then suffices to regard the body as rigid (in the
first phase of the solution of the problem) for the determination of the forces. Thereafter
(in the second phase of the solution of the problem), the same body is now considered as
deformable to determine the stresses, strains and displacements inside. By way of abundant
caution, such a two-step procedure can be iterated. The enormous computational capacity
and speed of modern computers make it easy to do such iterations18 .
In general, we should choose the simplest model that gives the information that we seek,
because we would like to keep the analysis as simple as possible without losing the heart of
the problem.
Scope of Our Subject
Engineers are often concerned with the design, construction or manufacture, and assembly
of engineering structures. The design of the various components of the structure entails
estimation of the loads (often by a separate elaborate calculation that itself can be complex
and challenging as to be the subject matter of one or more separate courses), choosing a

18
Such a procedure can be tried for problems where there is strong coupling also. But then such a procedure
may or may not converge. Additional investigation is necessary before one can be sure that such a scheme
will converge, that it will converge to the correct answer, and that it will converge to the correct answer
irrespective of the starting point. Engineers and physicists try out these procedures and see if they are
successful, even before convergence is assured. This then becomes a research problem for mathematicians
to investigate. The motivation for several problems in mathematics, we can see from the long history,
comes from the rich and varied problems in physics and engineering!
1-9 Introduction

proper material, and arriving at the proper dimensions19 . This process of ‘dimensioning’
involves calculation of the stresses, strains and displacements at the interior points of the
component bodies20 . It is here that our subject is relevant and important. Thus, a large
class of problems that fall within the scope of strength of materials21 (or mechanics of
materials, or solid mechanics) and the theory of elasticity, and which are of great importance
to engineers is the following.
Given the shape (geometry) and composition (material or materials) of a body in static
equilibrium, and the loading and the support conditions, to determine the stresses, strains
and displacements at any (i.e., every) point inside the body.
This problem as stated above is so general and complex, that even the best mathe-
maticians have not been able to solve it so far. Some of the best brains have thought and
laboured on this and similar problems for a long, long time. Yet, the solution of this prob-
lem is still elusive. The demands from practising design engineers, in the meantime, made it
imperative that at least partial solutions of at least simplified cases should be found. Thus,
while we wait patiently (perhaps for another 100 years?) for the mathematicians to solve
the problem in all its generality, it makes sense to make drastic simplifications for expe-
diency. Thus it is, that our simplified engineers’ theory of strength of materials emerged,
playing a crucial role in the design calculations of engineering structures.
Relatively recently, more refined solutions have become necessary, particularly in aero-
space applications. This is because the simplified theory is unable to give even an approxi-
mate answer in more and more cases. This situation has forced engineers to have a closer
second look at the theory, and to revise it making it more and more rigorous. The engineer-
ing science content has, thereby, increased, with a corresponding decrease in empiricism.
This new development places greater emphasis on the underlying assumptions, and has thus
led to the new avatar in the form of solid mechanics.
What this means is that the emphasis has shifted to the fundamentals, so that the
necessary changes can be made, and a revised theory built to deal with new situations.
Thus, if new situations arise where one is called upon to use special materials, perhaps

19
The loads may be surface and body forces. These act on the surface(s) and the volume, respectively, of the
body. An example of a surface force is the water pressure acting on all the wetted areas of a water tank;
another is the soil thrust on a retaining wall. These forces are sometimes called tractions. Additionally,
there could also be body forces that act over the volume of a body. The self-weight of a structure is
the most common example of body forces. Other examples are the ‘inertia force’ and the magnetic force
of attraction. ‘Inertia forces’ can be of decisive importance in dynamic problems. (‘Inertia forces’ are
imaginary forces assumed to be acting on a body so as to convert a problem in dynamics to one in statics.)
20
If the calculated values exceed the permissible ones, the procedure may have to be iterated.
21
To quote from the preface of Frocht, M.M.: Photoelasticity, Vol.1, John Wiley & Sons, Inc., New York,
(1941): “There are several aspects to the subject of strength of materials. Its center of gravity may,
however, be properly be said to lie in the science of stress analysis.” Frocht quotes from another book:
“The fact deserves emphasis that only one tiny spot need be repeatedly stressed above the endurance limit
to make the whole piece fail from the crack which starts at that point. The most highly stressed spot is
the Achilles heel of the whole. Any spot must fail when it has had unbearably high local stress, no matter
how harmless the applied stresses have been to the rest of the piece. Hence we must focus attention on
the actual local stresses and not be misled by nominal average or calculated stresses.”
Advanced Mechanics of Solids 1-10

smart materials in intelligent structures, one may have to develop the theory right from
the beginning. A careful review of the classical theory, mutatis mutandis22 , making the
necessary departure dictated by the new situation, but otherwise following the classical
course, may be the path to chart. After all, originality is said to be clever imitation. In
this way, the gulf between the engineers’ theory and the more sophisticated mathematical
theory of elasticity is narrowing. These new demands23 have also led to the development
of more and more tools which are the Finite Element Method (FEM) and the Boundary
Element Method (BEM) on the one hand, and Computer Algebraic Systems (CAS) such as
maple and mathematica, on the other.
Also of concern to engineers in solid mechanics is the analysis of stability. Long, slender
rods subjected to compressive loads can buckle out of shape with disastrous, fateful conse-
quences. Thus, plates and shells in compression are also prone to buckling. It is essential
at the stage of design itself to ensure that no buckling occurs24 .
It is difficult at this early stage to define the scope of our subject and elaborate on all
the aspects. We, however, hope that the ‘importance of being earnest’ in learning solid
mechanics is already established beyond doubt.
Stresses, strains and displacements at a point:
We had stated [p. 1-9] that the general problem is to determine the components of stress,
strain, and displacements at every point inside a given loaded body. We are already familiar
with these terms, but we need to have greater clarity about the components of stresses and
strains. This is because stresses and strains at a point are examples of (second order)
tensors, and they are consequently more difficult to understand with conceptual clarity.
For this reason, the concept of tensors is explained in greater detail later. Here we shall see
what is meant by the stress at a point.
Stress at a point:
The pressure or stress25 is sometimes defined and regarded as the force per unit area. This
is fine as long as we consider only uniform distributions of stress, or the average stress, over
a finite area. When we focus our attention to a point and zero in, and ask for the stress at
a point, this definition is inadequate inasmuch as the area of a point is zero. How shall we
get over this impasse?

22
with the necessary appropriate changes
23
Side by side with these developments is the emergence of research engineers. “Mechanical engineering is
not nut and bolt engineering” any more, as the late Professor B.R. Seth was fond of saying. (Professor Bhoj
Raj Seth was a noted applied mathematician, for long at Indian Institute of Technology, Kharagpur. A
large number of applied mathematicians now working both in India and abroad are either his students, or
his students’ students.) They are called upon to play the mathematician’s game, but are often inadequately
trained for the job. The generally higher calibre of these people makes partial amends for this inadequate
training. Thus, young teachers aspiring to go deeper into this branch will do well to invest a good part of
their time and energy in mathematics, both pure and applied. It would be nice to draw inspiration from
the rich academic traditions of some non-English speaking countries also.
24
Students will do well to associate thin walled structures with the possibility of buckling. Whenever the
thought of thin walled structures arises in the minds of engineers and engineering students, an alarm
should ring inside warning them, caution: check to make sure that there will be no buckling!
25
These two words are interchangeable. However, the term stress is almost universally used in our subject.
1-11 Introduction

(a) A stressed body cut

into two parts A and B (b) Resultant force and moment vectors on the small areas

Figure 1.3: On the cut section, consider a small area ∆1 enclosing the point P . There will
be a resultant force vector F 1 and a moment vector acting on this area ∆A1 . The area is
reduced progressively to ∆A2 , ∆A2 , · · · . Let us then look for the limits of F n and M n . If
a limit exists, that can be used as the definition of stress at a point on a plane. We shall
assume that both limits exist, but that the limit of the ratios M n /∆An = 0. The latter
assumption means that there is no locked-in moment inside the stressed body.

We note that the stress at a point depends decisively on the plane. Let us, therefore,
define the stress not at a point, but at a point on a given (i.e., specified) plane passing
through the point. See Fig. 1.3. The plane XX cuts the body into two parts A and B.
The effect of Part B on Part A is to introduce a stress distribution on Part A; the effect of
Part A on Part B, in turn, is to have an identical stress distribution (action and reaction,
Newtons third law) on Part B. The stress distribution exists within the body even before
the cut. This, however, is an internal state for the whole body. For any one part, say Part
B, when isolated, this becomes an external distribution of stress. This is a basic concept in
continuum mechanics.
We desire to define and clarify the notion of stress at a point on a plane (here on the plane
XX). The stresses are distributed on the cut surfaces. Let us consider a small area ∆A1 en-
closing the point P. The stress distribution is equivalent to a force F1 and a moment M1 . Let
us calculate the ratios F1 /∆A1 and M1 /∆A1 . These are the resultant force vector and the
resultant moment vector averaged over the small area ∆A1 . We propose to use the concept
of limits borrowed from the calculus. Accordingly, let us consider smaller and smaller ar-
eas ∆A2 , ∆A3 , · · · , ∆An , and the corresponding force and moment vectors F2 , F3 , · · · , Fn
and M2 , M3 , · · · , Mn . Calculating the ratios F1 /∆A1 , F2 /∆A2 , · · · Fn /∆An and M1 /∆A1 ,
M2 /∆A2 , · · · , Mn /∆An , we proceed to the limit as n → 0.
At this stage, following Cauchy, we make two assumptions.
i) The limit Fn /∆An as n → 0 exists. This is the resultant stress vector at the
point P on the plane XX.
ii) The limit Mn /∆An as n → 0 exists and is equal to 0. That is, there is no
locked-in moment.
We have now defined the stress vector at a point on a plane. The nature of stress at a
point is conceptually more difficult to understand. Hence this topic is discussed in greater
detail later [p. 4-1]. The general problem of stress analysis is extremely difficult. Thus, it
is absolutely necessary to make simplifying assumptions. Some of them are indicated in the
next section below.
Advanced Mechanics of Solids 1-12

It was stated that several simplifying assumptions are unavoidable, especially if we are
attempting to obtain analytical, closed form solutions. These simplifications are at various
levels. Some of them are examined below.
SIMPLIFICATIONS

(a) Highly discontinuous matter (b) A hypothetical continuum (c) An anisotropic body

Figure 1.4: Simplification as a continuum: the actual discontinuous body is replaced by a

hypothetical continuum, and calculations are made on this continuum. Such an idealisation
makes calculations easy or, at least, possible. Fig. 1.4c shows an anisotropic body; the
elastic properties are different in different directions.

Continuum
This concerns the composition of the body. We know that physicists in several situations
find it convenient to conceive of matter as highly discontinuous at the micro level: particles
with large volumes of empty space [Fig. 1.4a]. If our body were really of such hopelessly
discontinuous composition, it would make little sense to refer to its macro properties like
density. It is to get over difficulties of this kind that we assume that the body is a continuum
at the macro level. (Engineers, in such applications as we have in mind here, are interested
only at the macro level.) Technically, therefore, the actual body is replaced by a hypothetical
continuum, and calculations are made on this continuum [Fig. 1.4b]. One hopes that the
results of these calculations are valid for the real actual body. It seems reasonable to suppose
that, when the smallest volume of interest at the macro level still contains a large, very
large number of micro particles with voids, etc., there would be some kind of statistical
averaging over these large, very large number of micro particles, etc., and that it would
physically make sense to treat the body as a continuum for all practical (macro) purposes.
This assumption can, therefore, be taken as reasonable and valid in all situations, except
possibly when we have to deal with cases such as highly rarefied gases (when concepts and
methods of analysis from statistical mechanics, molecular dynamics and other branches may
have to be borrowed).
Linearity
This concerns the mathematical nature of the equations that govern the behaviour of the
body. In developing a theory we are, as indicated earlier, interested in calculating (or
predicting) the response of the body (stress, strain and displacement components at every
point inside the body) when the stimulus or command (loads, etc.) is specified. If we regard
the load as the input, and the resulting response (in the form of stress, etc.) as the output,
1-13 Introduction

(a) Linear (proportional) stress-strain (b) Linear (proportional) load-elongation

relationship relationship

Figure 1.5: Linear (proportional) stress-strain and load-elongation (load-deflection) rela-

tionships: linearity implies that superposition is valid.

and the body itself as the system, then we have the classical system theory as a possible
means of representation. If the governing equations are all linear, then it can be shown that
the output is linearly related to the input. The relationship between the loading (input)
and the response (output) can be conceived of as a straight line passing through the origin
[Figs 1.5a, 1.5b]. This would imply that, say, twice the strain would result in twice the
stress, twice the load would result in twice the deflection, etc., and that in general the effect
of two or more loads is the sum, superposition, of the separate effect of each of the loads
acting alone, etc. This is the principle of superposition26 .
The implications of linearity are deep and are of far reaching consequences. We do not
wish to go into these details at this stage. In reality, most systems in nature are nonlinear27 ;
linearity is often a simplifying assumption used to make the calculations possible or easy28 .
Linearity is usually associated with ‘smallness’ in some sense. In our context, it is often in
the restricted case of small strains and small displacements. It can also be in the (linear,
that is, proportional) relationship between stress and strain components (which is a material
property)29 . We shall briefly point out these at the appropriate places and indicate here
and there the consequences of nonlinearity. Here we shall assume that our systems in this
course on solid mechanics are all linear. [This is not always true. For example, consider a
beam-column [Fig. 2.10]. Here the effects of (i) axial loads P − P and (ii) transverse loads
W1 , W2 cannot be considered separately and then added up to obtain the total effect (such
as the deflection). However, superposition is valid in a restricted sense as shown in Fig. 2.11

26
This principle of superposition is of great fundamental importance, and has far reaching consequences.
Students are advised to spend time to understand this principle and its consequences.
27
Professor R.M. Rosenberg of U. of C., Berkeley, an acknowledged expert of Dynamics, is said to have
remarked that dividing (differential) equations into linear and nonlinear is like dividing all the objects in
the universe as bananas and non-bananas. Poor bananas (linear equations) would hopelessly lose out in
any such classification!
28
Sometimes nonlinearity is undesirable, but unavoidable. Sometimes it is avoided simply because its effect
cannot be easily calculated, or calculated at all. Occasionally, nonlinearity can have beneficial effects
also. This writer has heard Professor J.P. Den Hartog narrating an actual case of a German aircraft
unwittingly protected by nonlinearity by limiting the resonant amplitudes to safe limits (until one fateful
day immediately after it was serviced to take up clearances, etc., three of the four engines failed).
29
The two types of nonlinearity usually encountered in the theory of elasticity are (i) geometric nonlinearity
(in strain-displacement relations), and (ii) material nonlinearity (in stress-strain relationships).
Advanced Mechanics of Solids 1-14

Figure 1.6: A beam-column: superposition as shown is not valid here.

Figure 1.7: A beam-column: superposition as shown is valid here.

as long as the same axial loads P − P are acting. Additionally, there are several nonlinear
cases that can be considered only in advanced courses. Here we limit our considerations to
simple cases within the framework of linear theory.]
The reason why we consider only linearity is primarily because this is relatively simple;
nonlinear systems are far more complex. Furthermore, many technically important problems
can be successfully analysed using only the elementary linear theory30 . The linear theory
is relatively well understood; there are several general results applicable to linear systems.
No such general theorems exist for nonlinear systems. Theorems concerning existence and
uniqueness of solution either do not exist, or have severe restrictions. Besides, there are
several new phenomena in nonlinear theory that do not have any counterpart in a linear
framework. Thus, no apology is needed for leaving out the much harder nonlinear theory.

(a) Causes and effects (b) Causes 1 and 2 together → effects 1 and 2.

Figure 1.8: Cause 1 alone leads to effect 1; cause 2 alone causes effect 2; causes 1 and 2
together lead to effects 1 and 2 together. More generally, (α× cause 1 + β× cause 2) −→
(α times effect 1 + β× effect 2).

30
Professor J.P. Den Hartog, noted not only as a great teacher and author, but also was a famous successful
vibration consultant. He said more than once in his later days that most industrial vibration problems
require, for their effective solution, only elementary theory. He went on to say that “otherwise, I will not
be able to solve such problems at this age”. This writer does not agree with him on the last sentence; it
was stated only out of humility.
1-15 Introduction

(a) Linear (b) A general curve (c) Nonlinear relationships

Figure 1.9: If (x1 , y1 ) and (x2 , y2 ) are two points on the curve [Fig. 1.9b], does the point
(x1 + x2 , y1 + y2 ) fall on the curve? More generally, does the point (αx1 + βx2 , αy1 + βy2 )
fall on the curve? The answer, as we can clearly see, is: (i) yes, if the curve is a straight
line passing through the origin (0, 0); and (ii) no, otherwise.

It is necessary to understand clearly that the relationships represented by the lines

AB, OC, ODE, F G in Fig. 1.9c are all nonlinear. The fact that these are straight line
relationships does not make them linear; they are nonlinear, though piece-wise linear.
[Perhaps it is desirable to give a brief explanation of linearity in the context of differential
equations. A differential equation, we know, is defined as linear if the dependent variable
and its derivatives are in the first degree and there are no products of these.
First let us refer to the figure given [Fig. 1.9] and ask the following question. If (x1 , y1 )
and (x2 , y2 ) are two points on the curve, does the point (x1 + x2 , y1 + y2 ) fall on the curve?
More generally, do the points (αx1 + βx2 , αy1 + βy2 ) (for all values of α and β) fall on the
curve? The answer, as we can clearly see, is: (i) yes, if the curve is a straight line passing
through the origin (0, 0); and (ii) no, otherwise.
Now in the light of this understanding, let us consider a differential equation

d2 y dy
2 2
+3 + 4y = 0 (a < x < b).
dx dx
Let us ask the question similar to the one that we asked: if y = y1 (x) and y = y2 (x) are two
solutions (i.e., if these satisfy the differential equation), is y1 (x) + y2 (x) a solution? Or more
generally, are αy1 (x) + βy2 (x) also solutions (for all values of α and β)? Let us examine.
d2 y1 dy1
As y = y1 (x) and y = y2 (x) are two solutions, we have 2 2
+3 + 4y1 = 0 and
dx dx
d2 y2 dy2
2 2
+3 + 4y2 = 0.
dx dx
Multiplying the first equation above by α and the second one by β, and adding them,
we obtain
d2 y1 dy1 d2 y2 dy2
α2 + α3 + α4y 1 + β2 + β3 + β4y2 = 0.
dx2 dx dx2 dx
d2 (αy1 + βy2 ) d(αy1 + βy2 )
That is, 2 +3 + 4(αy1 + βy2 ) = 0,
dx2 dx
showing that αy1 (x) + βy2 (x) is indeed a solution for all values of α and β.
Advanced Mechanics of Solids 1-16

Why did, or does, this superposition hold here? This is because all the ‘operators’, viz.,
d2 d
, , y L(αy1 + βy2 ) = αL(y1 ) + βL(y2 ), L(0) = 0.
dx2 dx
are linear. Now if, on the other hand, we have a term like y 2 , (d/dx)1.5 , (d2 /dx2 )2.5 , or
y(d/dx) in the differential equation, superposition will not work!
Let us also realise that equations like, say,
d2 y dy
x2 2
+x + (x2 − n2 )y = 0
dx dx
are linear variable coefficient differential equations. The independent variable does appear in
the second degree, and there are products of the independent variable and the derivatives31 .
Integral operators are also linear. Note also that all straight line relationships are not linear.
Fig. 1.9b shows nonlinear relationships; some of them — (How many of them, and which are
they? Examine and find out.) — are nonlinear, but piece-wise linear. Linearity corresponds
to proportionality; the curve must pass through the origin; recall L(0) = 0. Note further
that the linearly independent solutions of a (homogeneous) linear differential equations form
a basis, and that they span the entire solution space.
The concept of linearity is profound; it has far reaching consequences. Electrical engi-
neering students are generally better exposed to these concepts in their course on Signals
and Systems. If the response of a system (linear, time invariant) due to an impulse is
known, the response to any excitation can be obtained. Some key words that act as a
memory trigger — no more than that — are response to a unit step function, convolution,
Faltung integral, Boltzmann superposition, etc.]
Elastic Materials
We assume that all the bodies that we consider are of elastic materials32 . We had emphasised
that it is not the materials that are elastic; ‘elastic material’ is the model33 that we employ
in this book. Physically speaking, a large number of materials fall in this category. When a
load is applied, a deformation results. When the load is removed, the deformation disappears
entirely, and the body return to its original undeformed configuration. If this is the case, we
say that the material is loaded within its elastic range. Most of our interest is in technical
applications34 ; the physical materials (and, therefore, bodies) can be treated as entirely
elastic when the load (and, therefore, the stress) is small35 . When the loads are large, the
31
Students are advised to spend some time, examine all the relevant aspects, understand clearly the roles
of the independent and dependent variables, and form a mature understanding of linearity. It is not
uncommon to see even great learned authors calling linear, variable coefficient differential equations as
nonlinear. As it is not a good idea to pick on learned famous authors, the actual mistakes are not pointed
out. Perhaps it also serves to illustrate that even learned famous authors do make mistakes now and then.
32
Some exceptional cases will be considered here and there in this book.
33
We may consider in a few places thermoelastic models also. This is used when we have to consider strains
because of temperature differences.
34
in the limited scope of this book
35
The stress can be high even if the load is small. An example is the stress in the immediate neighbourhood
of the point of application of a concentrated load. Another is that at a crack tip.
1-17 Introduction

limit of elasticity is crossed. When the limit of elasticity is reached, some materials called
ductile materials (for example, mild steel) start yielding. This is plastic flow. The state of
stress in the material (and, loosely speaking, the material itself) is now said to have entered
the plastic region. If the yield point is crossed, and then the load is removed, the body does
not return to its original configuration. There is a residual deformation which is called the
permanent set [Fig. 1.11b]. Furthermore, when the material is loaded beyond the elastic
limit and unloaded, and loaded again, the elastic limit goes up. This is due to the so-called
strain-hardening, sometimes called work-hardening also36 . We can see more details when
we discuss the simple tension test37 .
Associated with yielding, there are internal structural changes. The details of these are
the concern of the vast area called deformation processes in metals studied in metal physics
and metallurgy. We shall not go into these details, even though it is desirable to have some
familiarity with them. Our study is purely phenomenological38 ; we are concerned only with
the macro level descriptions and calculations. Courses in metallurgy and materials science
give the details and the micro level explanation of several phenomena associated with a
simple tension test to failure of a ductile material such as a mild steel specimen. When

(a) Loading and unloading (b) Loading and unloading: hysterisis loop

Figure 1.10: With loading and unloading, a small amount of energy is dissipated. With
several such cycles, quite some energy is dissipated which accounts for structural damping.

operating within the elastic range, but not outside it, the work done by the external load
during the deformation is stored inside the body as internal energy. None of it is dissipated
irreversibly. During unloading, this stored energy is entirely given back39 [Figs 1.10a, 1.11a].
The elastic limit and the proportional limit are conceptually different, but they are often,
though not always, close to each other40 . Some materials even within the elastic range
36
This phenomenon itself is sometimes called strain-hardening or work-hardening.
37
It is important to know about this test in detail.
38
phenomenological study
39
The case of hysteresis can be of decisive influence in some situations. It is this phenomenon that is
responsible for damping, or energy dissipation, when a body vibrates. In many places, however, this is
not important and can be disregarded. For a hypothetical elastic material, no energy is dissipated at all.
For real materials a small amount of energy is indeed dissipated as hysteresis loss. This amount is small
and negligible in most cases, but it assumes significance when there are repeated loading and unloading
as when a body undergoes vibration. It is these little droplets making an ocean that are responsible for
the energy dissipation. When a thin bar is bent and unbent several times, we can feel an increase in
temperature. This is due to the hysterisis loss manifesting as thermal energy.
40
Hooke’s law states that, within the elastic limit, the stress is proportional to the strain. This implies that
Advanced Mechanics of Solids 1-18

(a) Loading and unloading below the elastic (b) Loading beyond the elastic limit and unload-
limit: no permanent set ing: permanent set

Figure 1.11: Loading and unloading below and beyond the elastic limit. Note the permanent
set on unloading. Note that the stress and the strain are not necessarily proportional for
an elastic body (elastic behaviour).

may not have a proportional relationship between the load (or stress) and the resulting
deformation (or strain). Concrete is one such material; copper is another. Because of the
great importance of concrete as a structural material, special terms like secant modulus
and tangent modulus are sometimes used to bring the calculations within the framework of
linearity (proportional relationship between stress and strain).
Homogeneity, Crack-free Body in Its ‘Natural’ State
The body is assumed to be homogeneous, (i.e., the composition of the body is the same
at all points) even though this is not usually the case in several real life situations. In
reinforced concrete, for example, there are two materials, steel and concrete, steel in the
form of bars surrounded by concrete. Concrete, by itself, too is inhomogeneous; it is made
up of aggregates of varying sizes bonded together by cement paste. Nevertheless, this is a
convenient assumption to make, particularly when considering only simple cases. Thus, we
shall have only homogeneous bodies for our analysis almost everywhere.
The body is also assumed to be crack-free. The classical theory is developed for this
case. A realisation has come in recent years that micro-cracks or some faults can hardly be
avoided in several situations. These micro-cracks are not by themselves a cause for alarm,
and should not be taken as manifestations or indications of failure or impending failure.
A relatively new vigorously growing subject called Fracture Mechanics has techniques and
methods of analysis to judge whether these micro-cracks are ‘benign’ (harmless) or ‘ma-
lignant’ (harmful) (in the sense that these cracks open up, progress further, and lead to
catastrophic failure). The design philosophy is accordingly different. We, however, do not
concern ourselves in this book with these relatively recent developments.
There are also cases where the body might be prestressed or prestrained. Stresses can,
and do, occur inside a body even when there is no external load. Residual stresses (say, left
in a body after it is machined) are a case in point. These situations present complications
that we do not wish to consider now. Thus, the bodies that we consider are in their ‘natural’
state: when there is no external load, there is no stress inside, and when there is no stress,

the proportionality is maintained throughout the elastic region.

1-19 Introduction

there is no strain. The case of thermal strain, however, does occur frequently, and is
sometimes considered in some books at this level. [It is better in such situations to consider
the ‘material’ (really model) as thermoelastic, and not as elastic. Thermoelasticity may be
regarded as a different, though allied, discipline governed by a different set of constitutive
equations like plasticity. We do not discuss thermoelasticity in this book.]
Isotropy
Just as materials are often considered as homogeneous for simplicity even though they
are often not so in practice, these materials are also assumed to be isotropic. Isotropy
means that the properties are the same is all directions. Crystals have strong directional
properties: elastic, optical, piezoelectric, etc. There are cases of practical importance when
this assumption of isotropy is not justified; anisotropy may have to be taken into account
in the calculations. This can happen because of two different reasons or situations. One
is that the material that is used may have directional properties (different properties along
different directions). Wood is one example in which the properties are different along the
directions parallel and perpendicular to the grains. Another is cold rolled copper where
the rolling process introduces directional properties in the direction of rolling. The second
situation arises because of the constructional features. One example is a concrete floor slab
with parallel beams in only one direction. Another is a cast iron slab (or a steel plate) with
stiffening ribs in only one direction. These are examples where the material is, or can be
considered as, substantially isotropic. But the constructional features introduce anisotropy
(in the examples cited, ‘orthotropy’, which is a simpler case of anisotropy compared to the
general anisotropy) in the structure41 . In these situations, the structure, say, the floor slab
can be replaced by another slab with anisotropic (orthotropic) properties.
These are technically important cases. Nevertheless, a large number of technically im-
portant practical problems can be solved by considering only the simple case of isotropy.
In recent years, technological demands have made composite materials or constructions (or
‘composites’ as they are called) more popular. These introduce the unavoidable complica-
tion of anisotropy. Thus, any course that aims at building technical competence cannot
neglect anisotropy.
Simplifications: One-dimensional and Two-dimensional
All physical bodies are really three-dimensional. However, sometimes there are strange
situations. The (two-dimensional) curved surface of a sphere in a three-dimensional space
has features that complicate matters. The ordinary familiar Euclidean geometry is not
applicable to this case. It is desirable, especially for advanced students and teachers to be
aware of the complications lurking in the background.
We often model structural elements and machine components as either one-dimensional
or two-dimensional for simplicity of analysis. When two of the dimensions are small in
comparison to the third, it seems appropriate to model the element as one-dimensional. As

41
Then there can be unusual materials like bio-membranes. Such membranes have to be treated as or-
thotropic. However, these are exceptional situations. We will not discuss such cases in this book.
Advanced Mechanics of Solids 1-20

an example, the two ‘small’ dimensions may be the cross-sectional dimensions of a beam,
while the third may be its length (span). Beams, columns, shafts, members in a truss, etc.
are examples of such bodies (or elements). It is mainly such bodies (or elements) that are
analysed in a first course. We can, thus, obtain much simplification in our analysis.
In a similar way, we can have two-dimensional bodies also. A flat plate is, thus, a two-
dimensional (and, thus, simplified) model of a really three-dimensional structural element
like a roof slab. The analysis will generally be more difficult, though still much simpler than
the three-dimensional analysis. We generally consider only one-dimensional bodies (or more
appropriately, one-dimensional simplified approach), and thus one-dimensional problems in
elementary courses.

(a) A truss (b) One-dimensional bodies

Figure 1.12: A (plane or space) truss is made up of one-dimensional bars. These are usually
considered to be in pure tension or compression (i.e., one-dimensional states of stress). A
beam also is treated as a one-dimensional body.

As another example, let us consider the analysis of a hydrodynamical bearing. The

heart of the problem is the analysis of the flow of the lubricant fluid as the journal rotates
in the bearing. If the bearing is very short [Fig. 1.13b], the flow analysis becomes a two-
dimensional problem (as explained earlier). On the other hand, if the bearing is ‘infinitely
long’ [Fig. 1.13a], and all the cross-sections identical, we can neglect the flow in the z-axis
(axial direction), and again the problem is simplified.

(a) A ‘long’ bearing (b) A ‘short’ bearing

Figure 1.13: An ‘infinitely long’ (long) and an ‘infinitely short’ (short) bearings are shown
in the figure. The analysis of both ‘very long’ and ‘very short’ bearings is greatly simplified.

Strange and paradoxical as it may seem, two-dimensional approximations are possible

under either of these two extreme conditions: infinitely long (long) [Fig. 1.13a], or infinitely
short (short)[Fig. 1.13b]. Let us see another situation also. Shown in Fig. 1.14a is a
1-21 Introduction

(a) A cylindrical body (b) A circular cross-section

Figure 1.14: A cylindrical body (cross-section circular) is shown. Now for this geometry it
is more convenient to choose a cylindrical polar (r, θ, z) coordinate system.

cylindrical body subjected to some radial loads. (These loads are such that the entire body
is in static equilibrium.) The geometry of the problem makes it convenient to choose, not
a rectangular cartesian coordinate system, but a cylindrical polar one (r, θ, z). Now all the
quantities of interest (stresses, strains and displacements) are functions of r, θ, z. The above
three-dimensional problem becomes simpler under special situations.

(i) If the cylinder (in the z-direction) is very short compared to the radius, the problem
can be treated as two-dimensional with (r, θ) as the only two independent variables).

(ii) If the body is very long, and all the cross-sections identical (shape, properties, loading),
the problem is again simplified as two-dimensional with (r, θ) as the two independent
variables.

(iii) We can have simplifications that result from symmetry. If, for example, we have a
homogeneous cylindrical body (with the same cross-section and the same loading all
along its length in the z direction) with a circular cross-section loaded by a radial load
that is distributed symmetrically around the circumference (Fig. 1.14a), we have a
case of circular symmetry (axisymmetry, symmetry about an axis). A thick cylinder
subjected to an internal fluid pressure as shown in the figure is a case in point. We
can now infer that all the quantities of interest inside the body (such as stresses,
strains and displacements) are independent of the coordinate θ. The problem is thus
simplified: all the quantities are now dependent only on r, z.

(iv) If we have the conditions (iii) and either (ii) or (i) satisfied in a given problem, all
the quantities of interest are now functions of only r. Thus, the problem has become
one-dimensional.

In the above case (iv), the problem can be simplified by exploiting the inherent circular
symmetry. If one fails to notice it, and chooses a rectangular cartesian x, y coordinate
system, this convenience is lost, and the problem is no more one-dimensional.
Like circular symmetry, there can be spherical symmetry in some cases. An example
is the problem of finding out the stresses, strains and displacements in a spherical vessel
subjected to fluid pressure (internal, or external). Here the only independent variable is the
radial coordinate r; thus, the problem is one-dimensional.
Advanced Mechanics of Solids 1-22

We may also inquire into questions of the kind: how small (and compared to what) shall
the dimensions be before they can be justifiably treated as ‘small’ ? There are theoretical
issues involved in its answer. Furthermore, from a practical point of view, engineering
judgement is needed in making such assumptions properly.
Perhaps it is necessary to clarify that it is really not the bodies, but the analyses, that
are one-, two- or three-dimensional. A one-dimensional analysis takes into account the
variation along the (only) one direction. Thus, for example, a one-dimensional analysis
of fluid flow in a pipe can give information about the variation (of, say, the pressure, the
velocity, etc.) along the length42 . Variations of those variables across the cross-section
cannot be obtained from this analysis43 .
It looks plausible that when the body is ‘one-dimensional’, a one-dimensional analy-
sis would give good results. That is to say, the results of such a simplified analysis agree
very well with those of a more detailed, rigorous three-dimensional analysis. It is perfectly
possible and sensible in appropriate situations to carry out a one-dimensional (say, con-
duction heat transfer) analysis of a large, thick slab, even though the body is no more
‘one-dimensional’.
It is interesting to note that it is also perfectly possible to carry out an analysis which
is one-dimensional for one variable (say, the pressure), while it is two-dimensional, or even
three-dimensional, for another variable (say, the velocity). These are all various (mathe-
matical) models used to simplify a given problem that may be too difficult to solve with all
the complications retained. Usually the motivation for effecting these simplifications is the
desire to get rid of complicated (often nonlinear) mathematical terms44 . The guiding light
must still be engineering judgement45 .

42
Some people prefer to classify only those problems as one-dimensional, in which all the field variables
change only in the (only) one direction that is considered.
Consider the so-called Hagen-Poisseuille flow, which is a viscous (Newtonian) flow in a uniform, (circular)
cylindrical pipe. Among the velocity components vr , vθ , vz , only the axial component vz exists; the other
two are zero. This component of velocity, vz , however, varies with r, but not with z; i.e., vz = vz (r). The
pressure p, on the other hand, varies only along the axial direction, z; i.e., p = p(z). This problem is,
thus, two-dimensional in this sense. Some authors call this one-dimensional. Opinion is divided on this,
as indeed on everything in life!
43
We note that, in the elementary theory of pure bending of beams, the variation of stress across the cross-
section can be obtained as a result of a ‘one-dimensional’ analysis. This may appear to contradict the
statement made above. This is only an apparent anomaly; really there is no contradiction. Readers are
advised to review the development of the theory of bending, and understand the implications in the above
statement.
44
An order of magnitude estimate enables one to drop some of the terms of the governing equations, which
contribute but little to the final result. Such estimates are widely employed in fluid mechanics.
45
We may recall from our study of fluid mechanics that the governing (Navier-Stokes) equations are nonlinear
and notoriously difficult to solve. Prandtl was able to get over several difficulties and come up with his
boundary layer theory. He, and he alone, had the intuition and judgement to neglect all the complicating
features, retaining only the essential terms. Thus, the analysis became relatively simple — boundary layer
theory is still not very easy — without losing the crucial aspects (call them the heart of the problem, if
you like). This was in spite of the fact that there were scientists in Göttingen at that time who had far
greater mathematical prowess than Prandtl. It is often such abilities that mark out a genius from the rest.
1-23 Introduction

As an example, let us consider the problem of testing an airfoil in a wind tunnel that
seeks to simulate the actual conditions during flight of an aircraft. If the scaled-down
model46 of the wing is small in comparison with the cross-sectional dimensions of the wind
tunnel, we may assume that the conditions of flow over an object in an infinite fluid medium
are met at least approximately.
Infinite and Semi-infinite Bodies / Regions

(a) An ‘infinite’ beam (b) A ‘semi-infinite’ beam

Figure 1.15: ‘Infinite’ and ‘semi-infinite’ beams: rail-road tracks extending in both left and
right directions [Fig. 1.15a], and with a dead end, the track extending only to the right

When modelling physical problems we often use concepts like infinite and semi-infinite
regions. ‘Infinite’ in mathematical modelling is only ‘sufficiently large’ in the physical
sense. ‘Sufficiently large’ means that far-field conditions prevail. One example was indicated
earlier. As another, we might consider a rail-road track that can properly be modelled as as
an infinite beam (supported on elastic foundations) [Fig. 1.15a]. When a train moves, there
is a certain distance beyond which this load has no effect. When a train moves on a rail-track
in Calicut, it is obvious that there would be no consequent stress, strain or displacement in,
say, Chennai. Beyond a certain distance (say, 3 times the value of a certain parameter47 ),
the effects are negligibly small. If the beam is longer than this distance, it is treated as an
infinite beam.

(a) An ‘infinite’ region (b) A ‘semi-infinite’ region

Figure 1.16: ‘Infinite’ and ‘semi-infinite’ regions: fluid all around a flying aircraft [Fig.
1.16a], and soil below a tall building [Fig. 1.16b]

46
In a practical sense, an infinite medium is to be understood in this sense: when a body is introduced in a
fluid flow field, there would be local disturbances. These disturbances become less and less, and eventually
die down as we go farther and farther from the body. We can conceive of certain distances beyond which
the undisturbed conditions (the so-called far field conditions) can be assumed to prevail. These distances
are the limits. When the domain has dimensions that are larger than these distances, we may assume that
the domain to be an infinite region.
47
See the topic of beams on elastic foundations in any good book, ideally, M. Hetéyni: Beams on Elastic
Foundations, University of Michigan Press, Ann Arbor, Mich. (1944).
Advanced Mechanics of Solids 1-24

At the end, the dead end, of a rail-track, the track extends only in one direction. Such
cases are referred to as semi-infinite bodies [Fig. 1.15b]. Other examples are: a flying
aircraft may be modelled as an object moving in an infinite region (sky) of fluid (air) [Fig.
1.16a]; a moving ship may be modelled as an object moving in a semi-infinite region (ocean)
of fluid (sea water); a tall building may be modelled as a static48 load on a semi-infinite
region of elastic or viscoelastic, viscoplastic material (soil) [Fig. 1.16b]; etc.
CLOSING REMARKS
We have pointed out some simplifications. It is on the basis of these that the (simplified)
engineers’ theory of strength of materials or mechanics of solids is developed. We shall
consider only simple cases in this book. However, there may be some exceptional cases
(when one or more simplifying assumptions do not hold) that we may come across in this
book. We shall call attention to these cases at the appropriate places.
In the first course, which is at an elementary level, we take a simple-minded approach.
Thus, everything is simple. The geometry is regular: perfectly straight, perfectly circular,
perfectly spherical, etc.; the body is homogeneous, and free from cracks, defects and inclu-
sions; it is moreover in its ‘natural’ state (in the sense that there is no stress or strain when
there is no external load).
After discussing simple cases in a first course, students are generally required to take
a second course, where slightly more complicated cases are discussed. Special problems in
bending (like curved beams, unsymmetrical bending, beam-columns and beams on elastic
foundations), torsion of non-circular prismatic bars, thick cylinders and rotating discs and
energy methods are of such great importance that they cannot be left out of the curriculum.
Real life problems seldom fall into these cases (elementary and slightly more complicated
cases) referred to above. The geometry is often complex49 ; there can be several materials, as
say, in reinforced concrete; there can be cracks. One can guess that these complex problems
can usually be solved only by numerical methods like the finite element method (FEM).
However, before the numerical methods are used, it is necessary to formulate the problem.
To be able to do this properly, we have to learn the governing equations and the relevant
boundary conditions consistent with the physical conditions that actually prevail during
the operating conditions. Thus, just because there are powerful numerical methods now
available, and access to powerful computers including supercomputers, it does not follow50
that we need not learn the underlying theory properly with conceptual clarity.
48
For studying earthquake responses, the model is to be revised; the problem is now recast as a dynamic
one of a tall structure with support excitation (movement).
49
An example might be the calculation of the thermal stresses in a cylinder block of an automobile. The
geometry is admittedly far from simple. There can be strains (thermal strains) even when there is no
stress. In other cases, there can be locked in ‘residual stresses’ as when a component is machined. We
then have a case of internal stresses even when no external load is applied. There can be many more
complications. These examples are cited to emphasise the fact that only classroom exercises are simple,
and that real life problems are usually difficult, and fall outside the scope of a first course.
50
Of late there is a clear shift in emphasis: rigour seems to have disappeared altogether. If ability to learn
whatever is needed in later life is the objective, it is essential, absolutely essential, to learn at least six or
seven courses rigorously.
1-25 Introduction

New technological challenges have forced scientists and engineers to adopt more scien-
tific approaches. Our subject which used to be called strength of materials traditionally
has changed its style and methods in its new avatar as mechanics of solids. Structural
mechanics, theory of elasticity and the more abstract continuum mechanics at a higher
level are closely associated with our subject. Fracture mechanics and composites have be-
come important. Advances in aerospace engineering have made these topics to be of great
relevance . Experimental stress analysis and powerful numerical methods like the finite
element method (FEM) are also important allied subjects. When these new problems and
approaches are examined, it becomes necessary to learn abstract mathematics, etc. on the
one hand, and experimental methods on the other. There is no end to this already long
list. Ambitious readers of this book will do well to learn these topics slowly and steadily51 .

S.P. Timoshenko
One of the greatest names in the area of Strength of Materials or Mechanics of Solids is
Timoshenko. It is necessary for us to be familiar with his contributions. The following few
lines are meant to introduce this outstanding engineer-scientist to the young readers.
Stephen Prokofyevich Timoshenko (Dec. 22, 1878 - May 29, 1972) was a Ukranian by
birth. He is the author of several highly rated books, and is known as the father of modern
engineering mechanics. He was arguably the greatest ever teacher in the broad area of

The École Polytechnique has already shown us the way to proceed. It will be a folly not to take notice of
such guiding lights.
51
I presume my readers would permit me to make a few informal comments. I am reminded of a doctor
friend of mine who had once told me about his professor in the medical college. The learned professor
began his lecture: “There are three important things in the successful practice of medicine”. And then
he would pause and wait for attention before he would resume. “Number One: diagnosis”. Again he
would pause, and continue. “Number Two: diagnosis”, and pausing once again dramatically, “Number
Three: diagnosis”. I so much liked this that I am persuaded to imitate him and state: there are three
important things in mechanics: Number One: draw the free-body diagram; Number Two: draw the
free-body diagram; and Number Three: draw the free-body diagram!
Talking about free-body diagrams, I am also persuaded to add this. Let us take the example of a simply
supported beam loaded by a concentrated load. We draw the free-body diagram and find the reactions
easily. However, before we spring into the detailed calculations, we would do well to spend some and
examine the problem. How many unknowns do we have, and how many equations? If it is a concurrent
coplanar system of forces, how many separate (linearly independent) equations do we have? If we take
moments about four or five different points, will we or will we not obtain four or five linearly independent
equations? It pays to have a few quiet moments, reflect on such issues, and come to a mature understand-
ing. Such regular exercises would take us to a higher level of understanding. It is good to form the habit
of examining or analysing the problem before the detailed calculations are carried out.
During our students days some of my friends and I used to think that for each problem in mechanics,
there was one (and only one!) point about which the moment is to be taken. And we thought that it
was important to know beforehand which point for each problem. We followed, or tried to follow, this
practice until a faculty member came and told us that we could take moments about any point. That
was a revelation! Yes, we can take moments about any point. But if the point is chosen cleverly, the
subsequent work becomes simpler. It is not a question of correctness; it is a question of convenience in
choosing the proper point about which the moment is taken. Years have rolled by, but I remember his
words with much pleasure and gratitude.
Advanced Mechanics of Solids 1-26

engineering. There are an incredibly large number of outstanding scholars who are either
his students, or his students’ students. Many of them consider him to be their teacher in
an extended sense.
Timoshenko graduated in 1901 from the St. Petersburg Ways and Communication
Institute, and continued to teach there. He had a meteoric rise to become a Professor
& Chair of Strength of Materials in 1906 and Dean (Dean of the Division of Structural
Engineering) in 1909 in Kiev Polytechnic Institute. In 1922 he moved over to the US
where he worked for the Westinghouse Electric Company during 1923 - 1927. He joined
the University of Michigan, but in 1936 he moved to Stanford University where he taught
until he was 75. He has also written Technical Education in Russia, As I Remember (his
autobiography) and the famous History of Strength of Materials. Students are advised to
read his autobiography.

In the next chapter, we shall discuss special problems in bending.

Figure 1.17: Stephen P. Timoshenko

Chapter 2

SPECIAL PROBLEMS IN
BENDING

We already know the simple theory of bending of beams. Beams are one-dimensional
structural elements with transverse forces as the primary loading1 . This theory leading
to the Euler-Bernoulli formula2 is of much fundamental importance. However, there are
several technically important cases that are not covered by this theory. Thus, in this closer
second look at the theory of bending, we need to consider these special problems.
SPECIAL PROBLEMS IN BENDING
Some of these special problems are the following.

1. Unsymmetrical bending

2. Curved beams

3. Bending beyond the elastic limit

4. Beams of two materials

5. Beams with wide cross-sections

6. Beams on elastic foundation

7. Beam-columns

Among these special cases, we shall take up only the first two, viz., (i) unsymmetrical
bending and (ii) curved beams. However, we shall make brief comments about the other
1
Students are often tempted to call horizontal elements as beams. This mistaken idea arises because the
beams that they are exposed to are all, or almost all, horizontal. It is necessary to consider a variety of
situations where the beams are vertical or inclined. Aircraft structures provide several interesting cases.
In fact, it is the demands of the aerospace industry that led to great advances in our subject.
2
The Euler-Bernoulli equation governing simple bending is (with the usual notations) σ/y = M/I = E/R.
Advanced Mechanics of Solids 2-2

four cases also. To have a proper appreciation of these various cases, it is necessary to have
a review of the elementary theory of bending which we shall take up now.
SIMPLE THEORY OF BENDING
The actual body though really three-dimensional, we recall, is treated as one-dimensional
in the following analysis. When the (one-dimensional) beam is bent on application of a
bending moment, there are the resulting stresses, strains and displacements. We must
know these at all points when a known (or specified) bending moment is applied.
We consider the problem of pure bending, that is, bending in the absence of shearing
forces. In pure bending, the bending moment M is constant. That is to say, the same
bending moment M acts on both sides of the beam element. It means equivalently that
there is no shear force on the cross-sections of the beam element because dM/dx = F .
The starting point is the Love-Kirchhoff3 assumption: plane cross-sections continue
to be plane even after bending. Let us consider a small beam element before and after
deformation [Fig. 2.1]. The two cross-sections AA and BB before deformation rotate to
form an angle dθ and become A0 A0 and B 0 B 0 . By the Love-Kirchhoff assumption, the cross-
sections do not become crooked like A00 A00 . The bottom fibres are stretched out, while
the top ones are shortened. As we go from the bottom fibre to the top one, there is a
fibre, somewhere in between, that is neither elongated nor compressed [Fig. 2.1c]. Thus,
CD = C 0 D0 = R dθ. Examining the geometry of the deformed shape of the element, we
can develop the theory of bending.

(a) Before deformation (b) After deformation (c) Geometry of deformation

Figure 2.1: The elementary theory of bending is developed on the basis of the Love-Kirchhoff
assumption. The geometry of deformation is shown in the figures.

What do we mean by developing a theory of bending, or by solving the elementary

bending problem? What do we want to know after solving the problem of bending? Well,
we desire to obtain the stresses, the strains and the displacements when a known bending
moment acts on a beam element. As we consider only pure bending (bending in the absence
of cross-shear on the cross-section; this is, when there is no shear force), the bending moment
M is constant. The same bending moment M acts on both sides of the element. It implies
that there is no shear force on the cross-sections because dM/dx = F .
3
The same assumption is made in the theories of plates and shells also.
2-3 Special Problems in Bending

(a) Before deformation (b) After deformation (c) Strain and stress distributions

Figure 2.2: Examining the deformation it is possible to obtain the strain distribution and,
hence, the stress distribution across the cross-section.

Let us consider a typical fibre EF that, after deformation, becomes E 0 F 0 . The strain
can now be calculated as:
 E 0 F 0 − EF (R + y) dθ − R dθ y
strain, e = = = . (2.1)

y EF R dθ R

Having obtained the (expression for) strain of a fibre distant y from the neutral axis —
we shall show below that the neutral axis is the same as the centroidal axis4 — we can
calculate the stress easily as:
  Ey σ E
stress σ  = E e = −→ = . (2.2)
 
y y R y R

Both the strain and the stress vary linearly across the cross-section as shown in Fig. 2.2c.The
consequence of this fact is that the material farthest from the neutral axis is stressed the
most and that, therefore, the material is most effectively utilised when placed farthest from
the neutral axis.
Let us now consider the equilibrium requirements. These are the conditions to be
satisfied.
Z
σ dA = 0 (no net axial force) (2.3a)
ZA
(σ y) dA = 0 (no bending moment about the x axis) (2.3b)
ZA
(σ x) dA = 0 (no net bending moment about the y axis) (2.3c)
A

Using the expression for σ [Eq. (2.2] in these equilibrium equations, we are led to the
following results.
Z   Z
E
y dA = 0 −→ y dA = 0 (first moment = 0!) (2.4a)
A R A

4
These two axes do not always coincide: (i) a curved beam, and (ii) even a straight beam when accompanied
by an axial force are two cases in point.
Advanced Mechanics of Solids 2-4

Z   Z
E E E Mx
y y dA = Mx −→ y 2 dA = Mx −→ = (2.4b)
R R R I
ZA   A
Z
E E
y x dA = My = 0 −→ xy dA = 0 (x, y are principal axes!) (2.4c)
A R R A

Eq. (2.4a) shows that the neutral axis passes through the centroid!
Let us observe that we have used (i) the strain-displacement equations, (ii) the consti-
tutive equation (material law, Hooke’s law), and (iii) the equations of equilibrium. As the
displacements were assumed, there is no role for the compatibility equations (or, equiva-
lently, the compatibility equations are automatically satisfied). We shall refer to the last
equation (2.4c) when we discuss unsymmetrical bending later in this chapter.
This is a recapitulation of the simple pure bending of beams. This forms the basis of
the various special problems of bending that we shall discuss one by one below. As the
readers would surely have learned this, we do not discuss this theory any further except to
point out the amazing consequences of the Love-Kirchhoff assumption.
The Amazing Consequences of the Love-Kirchhoff Assumption
The classical theory of small deflection of thin plates is developed as a two-dimensional
generalisation of the theory of bending of (one-dimensional) beams. The starting point
for both is the Love-Kirchhoff assumption5 . The consequence of this simple assumption
is amazing. The theory of bending of beams is intimately related to the geometry of the
bent (deflected) shape of the middle line. This means that all the physical quantities of
engineering interest such as the bending moment and the shearing force; and the stress,
the strain, and the displacement at every point are decided entirely by the geometry of
deformation of the middle line. A similar assumption, a similar procedure, and a similar
conclusion are applicable for thin plates and shells also. For a plate now we have a middle
surface (instead of a middle line for a beam), and when the plate bends, the originally flat
middle surface goes into a curved one. The consequence this time also (as before for a beam)
is that the geometry of this curved middle surface is related to all the physical quantities
of engineering interest. The Love-Kirchhoff assumption, thus, converts the engineering
problem of the mechanics of solids to a mathematical one of the geometry of the curved
middle surface6 . The differential geometry of the curved middle surface now takes over; the
curvature tensor7 plays a decisive role in the theory of plates.
This time, however, the geometry is a little more complicated because we have to deal
with a two-dimensional (curved) surface w = w(x, y), where w is the (small) deflection of
the middle surface. Now there is bending in both the directions. Thus, we now have in both
5
Cross-sections which are plane before bending remain plane even after bending; they only rotate.
6
In a voice, loud and clear, the Love-Kirchhoff assumption announces triumphantly: “tell me all about the
geometry of the curved middle surface, and I can tell you all that you want to know about the plate such
as (i) the bending moment, the shearing force, and the twisting moment; and (ii) the stresses, the strains,
and the displacements.”
7
The radii of curvature and the twist, we can see, are analogous to the normal and shearing stresses,
σxx , σyy and the shearing stress τxy . The analogy is based on the fact that the curvature tensor and the
2-5 Special Problems in Bending

directions (i) the slopes ∂w/∂x, ∂w/∂y; the (approximate) curvatures ∂ 2 w/∂x2 , ∂ 2 w/∂y 2 ;
and (iii) the twist ∂ 2 w/(∂x∂y) = ∂ 2 w/(∂y∂x) pertaining to the geometry. On the physical
side, we can expect to have (i) the bending moments Mx , My ; the vertical shear forces
Qx , Qy ; and (iii) the twisting moments Mxy , Myx corresponding to the two directions x, y.
(For a shell, the theory is far more complicated.)
Now we are ready to take up the special problems one by one. First we shall dispose
of items 3 - 7 with brief comments for each before we take up the cases of unsymmetrical
bending and curved beams.
PROBLEMS IN BENDING: FURTHER EXTENSIONS
These are problems when one or more of the simplifying assumptions made in the elementary
theory are dropped, only one at a time. We shall have a cursory glance at some of them
with a few helpful suggestions. First we shall examine what happens if or when the elastic
limit is exceeded.
Bending Beyond the Elastic Limit

(a) Fully elastic region (b) Elastic-plastic region (c) Fully plastic region

Figure 2.3: The stress distributions in the three cases: (i) fully elastic [Fig. 2.3a], (ii)
elastic-plastic [Fig. 2.3b], and (iii) fully plastic [Fig. 2.3c].

We begin with the simple theory of pure bending when the entire body is in the elastic
regime (that is, the stresses are below the elastic limit, which for all practical purposes
can be taken to be the proportional limit). Now what happens when the applied bending
moment is progressively increased? Well, the maximum bending stress (normal stress) is
correspondingly increased until the elastic limit is reached. On further loading plasticity
‘creeps in’ and the stress distribution is modified from the linear variation [Fig. 2.3a] to
a nonlinear one [Fig. 2.3b]. Now we have an elastic core sandwiched between two plastic
regions shown shaded. The figure is drawn on the (simple-minded) assumption that the

(two-dimensional) stress tensor are examples of a second order tensor.

 1 1
  
1
0
   
Rxx Rxy R11 σxx τxy σ11 0
 −→   analogous to  −→ .
     

1 1 1
Ryx Ryy
0 R22
τyx σyy 0 σ22
Advanced Mechanics of Solids 2-6

yield point σp is the same in both tension (bottom) and compression (top)8 . The bending
moment (or resisting moment) that corresponds to this stress distribution [Fig. 2.3b] can
be calculated. On further loading, ‘plasticity’ permeates deeper into the cross-section from
both the bottom and the top ends. The maximum, the limit, is when ‘plasticity’ creep in
and covers the entire region. This, of course, is physically impossible; the stress cannot
obviously change abruptly from +σy (tensile) to −σy (compressive), as we cross the neutral
axis from the bottom to the top of the cross-section. Now a plastic hinge is formed at
that cross-section, because the section cannot develop any further resistance. The limit of
resistance is reached when the entire cross-section goes into yield, the bottom half with a
tensile stress σy , and the top half with a compressive stress σy .
Let us note that we still follow the Love-Kirchhoff assumption, and that the strain
variation is still linear. The constitutive equation is now changed: from elastic through
elastic-plastic to fully plastic.
Beams of Two Materials

(a) Two materials (b) Distribution of strain (c) ‘Transformed area’

Figure 2.4: The cross-section shows two materials [Fig. 2.4a]. The strain distribution is, as
in the earlier cases, linear [Fig. 2.4b]. The stress distribution, however, not linear, because
the moduli of elasticity of the two materials are different [Fig. 2.4c]. It is also possible
to have the concept of a ‘transformed area’ and to pretend that the material is the same
throughout.

For this case also, the strain distribution is the same, viz., linear. The stress distribution
is different, because the Young’s modulus of elasticity has different numerical values, E1
and E2 for the two materials. Once the stress distribution is known, the expression for the
bending moment can be worked out.
It is advantageous to have a large quantity of a low modulus material in the middle
low-stressed area and a small quantity of a high modulus material in the farthest regions
of high stress. For example, we can have wood9 as the low modulus material in the middle,
and steel as the high modulus one at the ends (shown shaded in Fig. 2.4a). Instead of
considering two materials, it is possible to have the concept of a ‘transformed area’ and
treat the beam as if it were only of one material. Thus, if E1 , E2 (with, say, E2 > E1 ) are
8
If this is not so, it is not at all difficult to deal with the case. The corresponding figure will not be skew-
symmetric any more. The consequences are not difficult to explore. We, however, do not undertake this
exercise here.
9
This used to be inexpensive, but is not so any more!
2-7 Special Problems in Bending

the Young’s moduli, respectively, the area A2 can be transformed as A1 × (E2 /E1 ). Now
the transformed A2 will be larger than the real, or original, A2 . The modified calculations
can thus be brought into the framework of the more common or convenient beams with
only one material.
One of the commonest examples of a beam with two materials is a reinforced concrete
beam. Steels rods are placed in the region of tensile stresses. Concrete is weak in tension,
and it is common practice in design to neglect or disregard the presence of concrete in the
tensile region. If the bonding between the steel rods and the surrounding concrete is proper
— the bonding must be proper, so that the steel rods do not ‘slip’ — the stress in the
steel bar is Esteel /Econcrete times the stress in the concrete. The position of the neutral
axis and the bending moment (or the resisting moment) can be worked out from the stress
distribution across the cross-section. There are several important aspects to discuss, but we
cannot do so in this book. (All this will be discussed in great detail in courses on reinforced
concrete design and analysis.)
This approach if often employed in the calculations on Reinforced Cement Concrete
(RCC) beams. Fig. 2.5 refers to such a case. The figures refer to an RCC beam. Steels

(a) Cross-section with two materials (b) Distribution of strain

Figure 2.5: A simplified cross-section of an RCC beam and the corresponding strain distri-
bution are shown in Fig. 2.5a. The stress distribution is shown in [Fig. 2.5b]. The tensile
stress in the steel bars is much higher than that in the concrete, because Esteel is much
higher than Econcrete . Generally the tensile stress in the concrete is neglected. The neutral
axis is not at a depth of d/2.

rods are placed in the tensile region (at the bottom of the cross-section in this figure)10 .
The distribution of strain is as shown. The stresses are, however, not linearly distributed.
The neutral axis is shifted upwards [Figs 13.4, p. 13-18].
Beams with Wide Cross-sections
In the simple bending problem that we have seen, the cross-section is as in Fig. 2.7a. If,
instead, we have cross-sections as in Fig. 2.7b with the width b >> than the depth d, there
are some changes to be made. In the former cases as in Fig. 2.7a, the cross-sections are
free to have lateral contractions (in the top layers where the bending stress is tensile), and
extensions (in the bottom layers where the bending stress in compressive). However, when

10
In other examples, the steel bars could be at the top; it depends on where tensile stresses are developed.
There are many details that cannot be explained here. For students of Civil Engineering, reinforcement
concrete design is a subject of great importance.
Advanced Mechanics of Solids 2-8

(a) Two materials (b) Distributions of strain and stress (c) ‘Transformed area’

Figure 2.6: A cross-section of a beam with two materials [Fig. 2.6a] and the corresponding
strain and stress distributions are shown in Fig. 2.6b. An ‘equivalent cross-section’ based
on the concept of a ‘transformed area’ is shown in Fig. 2.6c.

(a) A usual cross-section (b) Beam with a wide cross-section

Figure 2.7: Fig. 2.7a shows an example of a usual cross-section for which the simple theory
of bending is worked out. Fig. 2.7b shows the case of a wide cross-section. Now the lateral
strain is inhibited and, consequently, the theory is to be modified.

the cross-section is very wide (with b >> d), the lateral strain is restricted or inhibited.
The physical effect of this is that the beam will now be stiffer.
If we assume that ezz = 0, that is, the lateral strain is completely prevented — actually
the lateral strain may not be completely prevented, only reduced considerably —
1
ezz = [σzz − ν(σxx + νσyy )] = 0.
E
Further, we may assume that σyy = 0 — the various layers do not press on one another —
σzz = ν σxx .
1
If this is substituted in the first of the generalised Hooke’s law exx = [σxx − ν(σyy + σzz ]
E
with σyy = 0, we obtain
1 − ν2
exx = σxx .
E
The bending stress σxx given by the Euler-Bernoulli equation is
M
σxx = y.
Izz
2-9 Special Problems in Bending

Thus, we have
1 − ν2
exx =
M y. (2.5)
EIzz
From the usual Euler-Bernoulli equation, we find that
My My
σxx = −→ exx = . (2.6)
Izz EIzz
On comparing Eqs (2.6) (applicable for the usual simple cases) and (2.5) (applicable for the
case of wide cross-sections), we note that the strain exx is reduced by the factor (1 − ν 2 ).
With this correction, the changes to be made can be understood. For example, the governing
differential equation for bending becomes
d2 y
EIzz = (1 − ν 2 ) M. (2.7)
dx2
Beams on Elastic Foundations

(a) A beam on an elastic foundation (b) An element of the beam

Figure 2.8: A beam AB, resting on an elastic foundation with the Winkler constant (spring
stiffness) k, is acted upon by an external load distribution w(x). A concentrated load P
and a bending moment M are applied at the ends. The left end A is fixed, while the right
end B is free.

Shown in Fig. 2.8a is a beam AB on an elastic foundation (continuous elastic support).

The foundation is assumed to be a Winkler foundation11 with a Winkler constant (spring
stiffness) k. In addition to an external distributed load w(x), there is a concentrated load
P and a bending moment M applied at the right end B. The left end A is fixed.
The governing differential equation is
d2 d2 y
 
EI 2 + ky = w(x) (0 < x < l). (2.8)
dx2 dx
11
It is common practice to assume a Winkler foundation when beams on elastic foundations are treated. A
railroad track, a beam cast integrally on a grid of other structural members, and even some problems in
shells are cases where the theory of beams on elastic foundations is used. Objections can be raised on how
such a foundation that is assumed to produce a reaction proportional to the deflection can realistically
represent all these cases. Yet it is quite firmly established that in almost all of these applications such an
assumption (as reaction = ky, where k is a constant called the Winkler constant) is eminently satisfactory.
Hetéyni [7] contains pretty much everything that can be said on beams on continuous elastic support.
Advanced Mechanics of Solids 2-10

[We know from the Euler-Bernoulli theory of bending that the shear force, F ; the bending
moment, M ; and the rate of loading, w; are related by the equations
dF dM d2 y
= −w; = F; EI = −M.]
dx dx dx2

For our problem, the beam is fixed at the left end A (x = 0). Thus, the deflection and
the slope at A are both zero. These are the prescribed boundary conditions.

y(x) ≡ y(0) = 0 No deflection at A Prescribed b.c. (2.9)

x=0
dy 
≡ y 0 (0) = 0 No slope at A Prescribed b.c. (2.10)
dx x=0


The differential equation, let us note, is of the fourth (4th ) order; it will, therefore, take four
(4) boundary conditions. Two of them are the prescribed boundary conditions given above
[Eqs (2.9), (2.10)]. The other two boundary conditions are the natural boundary conditions
at the right end B (x = l). Here we know neither the deflection, nor the slope. Accordingly,
the natural boundary conditions prevail at B.
d2 y 
 
EI 2  = −M Applied B.M. at B Natural b.c. (2.11)
dx x=l
d2 y 
 
d
EI 2  = −P Applied concentrated force at B Natural b.c. (2.12)
dx dx x=l

Let us also note that the primary variables y and dy/dx are specified at the left end A (x =
0). With the specification of these four boundary conditions — two prescribed boundary
conditions at the left end A (x = 0), and two natural boundary conditions at the right end
B (x = l) — the above equation (2.8) can be solved. The actual calculations, though simple
and straightforward in principle, can be quite a burden.
[We have referred to two kinds of boundary conditions, prescribed and natural. These
must be understood in the light of the calculus of variations. Students are advised to learn
at least the most elementary aspects of this extremely useful topic.]
Beam-Columns
Beam-columns are one-dimensional structural elements that carry transverse loads W1 , W2 ,
· · · , etc., like a beam, and also axial compressive forces like P − P like a column [Fig. 2.9a].
Such structural members are not uncommon, particularly in airplane structures. Beam-
columns are special in the sense that superposition (of the separate effects of (i) W1 , W2 ,
· · · and of (ii) P − P ) is not valid. The reason is not hard to understand. The axial forces
introduce bending moments; their magnitudes depend decisively on the deflection. Thus,
there is a cross effect — a coupling, as it were — between the effects of (i) the axial loads
P − P , and of (ii) the transverse loads.
The bending moment in the two sections (0 ≤ x ≤ a) and (a ≤ x ≤ l) is given by
Wb
M =− x − Py (0 ≤ x ≤ a) (2.13a)
l
2-11 Special Problems in Bending

(a) A beam-column (b) A beam-column with a single load W1

Figure 2.9: A beam column with the relevant details is shown in Fig. 2.9a. The same
beam-column but with a single concentrated load W1 is shown in Fig. 2.9b.
Wa
= −W a + x − Py (a ≤ x ≤ l) (2.13b)
l
[From our earlier knowledge of the theory of beams, we know these equations already (with
only marginal appropriate changes). M is the bending moment, EI the flexural rigidity,
and P the axial compressive force. y = y(x) is the vertical downward deflection. Any set
of consistent units and sign conventions may be used. The main difference here is that the
axial force P together with the deflection y contributes to the bending moment.]
The governing differential equation is, therefore,

d2 y Wb
EI 2
+ Py = x (0 ≤ x ≤ a) (2.14a)
dx l
Wa
= Wa − x (a ≤ x ≤ l) (2.14b)
l
If we call P/(EI) = k 2 , these equations can be rewritten as

d2 y Wb
2
+ k2 y = x (0 ≤ x ≤ a) (2.15a)
dx EI l
Wa Wa
= − x (a ≤ x ≤ l) (2.15b)
EI EI l
The solution of this equation defined differently in the two intervals is of the form

y = C1 sin kx + C2 cos kx + (particular integral)1 (2.16a)

= C3 sin kx + C4 cos kx + (particular integral)2 (2.16b)

There are four boundary conditions to determine these four arbitrary constants of integra-
tion. They are:
(a) at x = 0, y = 0;
(b) at x = l, y = 0;
Advanced Mechanics of Solids 2-12

(c) the slope dy/dx computed from Eqs (2.16a) and (2.16b) must be equal; and
(d) the deflection y computed from Eqs (2.16a) and (2.16b) must be equal.
The four constants can be evaluated using these boundary conditions. The procedure is
simple and straightforward, though the algebra is tedious. These steps are not shown here.
The deflection at mid-span for the simple special case when a = l/2 (that is, if W acts at
mid-span) is
r
W l3 3(tan u − u)
 
 kl l P
δ ≡ y = , where u = = . (2.17)

x=l/2 48 EI u3 2 2 EI
When u = π/2, the deflection becomes infinite!
r
π l P π 2 EI
When u = = , i.e., when P = = Pcr ,
2 2 EI l2
the deflection becomes infinite! The parameter u is a measure of how close the axial load
P is to the Euler buckling load Pcr = π 2 EI/l2 . The expression [· · · ] in Eq. (2.17) may
be regarded as a magnification factor: the mid-span deflection of W l3 /(48 EI) is magnified
when there is an axial compressive load P − P !
We can next compute the deflections when (i) there is only W1 , and P − P ; (ii) there
is only W2 , and P − P ; and (iii) there are both W1 and W2 , and P − P . The procedure is
similar, though more tedious. The result is insightful: superposition now holds, if the same
axial load P − P acts. Thus, we see the following important result.

Figure 2.10: A beam-column: superposition as shown is not valid here.

Figure 2.11: A beam-column: superposition as shown is valid here.

With this we close the discussion of beam-columns, and take up the important topic of
unsymmetrical bending in the next section.
UNSYMMETRICAL BENDING
Let us consider the bending of a cantilever with two examples of cross-section: (i) an unequal
angle iron, and (ii) a Z-section. Note that x and y are not principal axes (that is, Ixy 6= 0).
The principal directions uu and vv are shown. The line of load — or more precisely, the
2-13 Special Problems in Bending

(a) A loaded cantilever (b) An angle iron cross-section (c) A Z as the cross-section

Figure 2.12: A cantilever with an end load P is shown in [Fig. 2.12a]. Two possible cross-
sections, an angle iron [Fig. 2.12b] and a Z-section [Fig. 2.12c], along with the principal
axes of inertia Iuu and Ivv , are shown. These are both cases of unsymmetrical bending as
the line of the load — trace of the plane of the bending moment — is not along a principal
axis of the second moment of area (principal axes of inertia).

trace of the plane of the applied bending moment — is along the y axis (non-principal axis),
and not along the principal axis (uu or vv). The bending moment applied on a cross-section
AA is M = P l. This Mx acts in the vertical plane, that is, about the axis xx. Let us try to
obtain the bending stresses using the Euler-Bernoulli formula σ/y = Mx /Ixx = E/R. We
now have the equilibrium equations as before.
Equilibrium:
Z
σ dA = 0 (no net axial force)
ZA
(σ y) dA = Mx (bending moment about the x axis)
ZA
(σ x) dA = 0 (no net bending moment about the y axis)
A

Using the expression for σ [Eq. (2.2] in these equilibrium equations, we are led to the
following results.
Z   Z
E
y dA = 0 −→ y dA = 0 (first moment = 0!)
R
ZA   A
Z
E E E Mx
y y dA = Mx −→ y 2 dA = Mx −→ =
A R R A R I
Z   Z
E E
y y dA = My = 0 −→ xy dA = 0 (x, y are principal axes!)
A R R A

Let us examine the last equation closely. What does it state? Physically it is an equation of
equilibrium stating that there is no net bending moment My . But it leads to the conclusion
that the (area) product of inertia Ixy = 0, which means that the x and y axes are principal!
But we know that x and y axes are not principal axes. Where is the anomaly? Did we
make any mistake?
Advanced Mechanics of Solids 2-14

Sure, we did. We applied the simple theory (of symmetrical bending) to these cases of
loading. This demonstration forcibly tells us that the simple theory of bending is not valid
here. This is a case of unsymmetrical bending.
What do we do? One way to deal with the problem is this. (i) To find the principal axes
of inertia uu and vv, and the principal second moments of area (principal area moments of
inertia) Iuu and Ivv (using Mohr’s circle, or the equivalent transformation equations related
to the inertia tensor); (ii) to resolve the bending moments Mu and Mv about the principal
axes; and finally (iii) to change this one (relatively difficult) problem of unsymmetrical
bending to two problems, each of the simple case of symmetrical bending. An important
consequence is that the neutral axis is not xx, nor perpendicular to the line of loading, any
more. We shall see this below. Referring to the three figures [Fig. 2.13], the (difficult)

(a) Vertical end load P (b) Load resolved along vv (c) Load resolved along uu

Figure 2.13: A (relatively difficult) problem in unsymmetrical bending [Fig. 2.13a] is con-
verted into two (simple) problems of symmetrical bending [Figs 2.13b and 2.13c] by resolving
the load P along the two principal axes of inertia uu and vv.

problem of unsymmetrical bending [Fig. 2.13a] is converted into two (simple) problems,
each of symmetrical bending. The total stresses are obtained by adding up as

Mu v Mv u (P cos θ) v (P sin θ) u
σ= + = + . (2.20)
Iuu Ivv Iuu Ivv

We can obtain the equation to the neutral axis by setting equal to zero the above expression
for the total stress giving us

(P cos θ) v (P sin θ) u
+ = 0. (2.21)
Iuu Ivv

This neutral axis is not horizontal, nor is it perpendicular to the line of action of the load
P ! This equation (2.21) can, of course, be written in terms of x and y. An illustrative
example is worked out later [p. 13-24] which will be helpful to understand the procedure.
SHEAR CENTRE
Shear centre is of much importance to Civil, Mechanical and Aerospace engineers. For
every cross-section used for a beam, there is a point called the shear centre, through which
the external load must pass for the beam to bend without twisting. As an example, when
2-15 Special Problems in Bending

the wing of an aircraft bends due to the air pressure from below, it also twists12 . It is
necessary (i) to understand the theory behind this concept, (ii) to know the location of
the shear centres of common cross-sections, and (iii) to be able to calculate and locate the
shear centre for a given cross-section. Finally it is necessary to have a good qualitative
understanding of bending and twisting13 .
To understand the concept of shear centre, we need to know the distribution of shear
stresses (cross-shear) on the cross-section because of bending accompanied by a (transverse)
shear force. We have studied this in our earlier first course on the mechanics of solids (or
even in strength of materials), but we would still have a quick review of this, omitting
several details. Students are advised to revise this topic using one or two good books.
Shear Stress (Cross-shear) on the Cross-section

(a) Unequal bending moments (b) Unequal bending stresses on (c) Shear stress τ on the under-
M and M + dM the two sides side for force balance

Figure 2.14: A beam element of length dx is considered [Fig. 2.14a]. Because the bending
moment is not a constant, the corresponding bending stresses are also different on the two
faces of the element [Fig. 2.14b]. To keep the shaded portion in equilibrium [Fig. 2.14c], a
shear stress τ — on the y plane in the x direction — must develop on the underside. The
counterpart τyx — shear and complementary shear — will then act on the cross-section in
the vertical direction. It is this shear stress τxy on the cross-section that we are after.

First we note that there is a (vertical) shear force on the cross-section, say, F . This
means that the bending moment M = M (x) is not a constant, because dM/dx = F . If we
have a beam element [Fig. 2.14a], the bending moments are M and M + dM on the two
faces. It follows that the bending stresses are different on the two faces or cross-sections.
If we consider a small portion shown shaded in Figs 2.14b and 2.14c of the beam element,
we can see that the bending stresses are different on the two faces: to the left at x and to
12
This twist changes the angle of attack causing changes in the aerodynamics forces acting on the wing.
The changed forces, in turn, lead to changes in the bending of the wings and the consequent twisting
of the wing. There is thus a strong coupling between the bending and the associated twisting, and the
associated aerodynamic forces acting on the wing. This situation can, under certain situations, lead to
serious problems of flutter.
13
Books on the theory theory of elasticity treat bending and twisting of bars together, twisting being a
sub-problem of a larger general problem of bending.
Advanced Mechanics of Solids 2-16

the right at x + dx. Inasmuch as the bending moment M + dM at x + dx is larger than M at

x, the bending stress also is correspondingly larger [Fig. 2.14b]. The shaded part must, of
course, be in (horizontal) equilibrium. This requirement makes it essential that some shear
stress τ is developed on the underside [Fig. 2.14c] to have equilibrium (in the horizontal
direction). The counterpart of this shear stress τyx on the underside is τxy acting on the
cross-section in the vertical direction. It is this vertical shear stress on the cross-section —
the cross-shear on the cross-section due to bending — that we are after.
Once this argument is well understood, it is easy enough to make the calculations and
obtain an explicit formula for this shear stress τxy .
Consider the horizontal equilibrium of the shaded portion abcd. To the right acts the
bending stress on the cross-section x over the area ad, and to the left acts the similar, but
larger, bending stress on the cross-section x + dx over the area bc. These two are not equal.
The resultant horizontal (longitudinal) force F on the face ad [Fig. 2.14b] acting to the
right is less than the resultant horizontal (longitudinal) force F + dF on the face bc. The
difference dF can be computed by integrating the expressions for the bending stress on
both faces. This difference dF is balanced by the shear stress τ (actually τyx ) acting on the
underside ab on the area ab× width = dx × b. For equilibrium, τ × b × dx = dF where b is
the width of the cross-section at the level (height) of the line ab. We can compute dF as
Z
dM dM dM
dF = q dx = y dA = Aabcd × ȳ = Q
I abcd I I
Q is the moment of the area abcd about the neutral axis. This dF acting to the left must
be balanced by the shear force developed on the underside ab. Thus, dF = τ × b × dx where
b is the width of the cross-section at the level (height) of the line ab. The shear stress τ on
the cross-section — this is the sought after cross-shear — can be calculated as
dV dM Aabcd ȳ VQ q
τ= = = = .
b dx dx Ib Ib b
Q, we repeat, is the moment of the area above the line ab about the neutral axis, and V
the vertical shear force on the cross-section.
 
VQ q FQ q
τ= = = if the vertical shear force is designated by F . (2.22)
Ib b Ib b
gives us the distribution of the shear stress on the cross-section. This we can take as the
operating equation for such shear stress calculations. Although we have already studied
this earlier, we shall clarify this further by two illustrative examples in a later chapter.
CURVED BEAMS
The theory of pure bending leading to the Euler-Bernoulli equation was developed for beams
that are initially straight. When the beam is curved even before the bending moment is
applied, there are some major departures in the theory. Crane hooks, chain links, C-clamps
and curved frames of punching machines are some examples where the curved members are
to be treated using the curved beam theory. We shall develop this theory leading to the
so-called Winkler-Bach solution.
2-17 Special Problems in Bending

General Remarks

(a) A straight beam! (b) A chain link

Figure 2.15: ‘Straight’ and curved beams. The first [Fig. 2.15a] is technically a straight
beam! The second [Fig. 2.15b] is indeed a curved beam. What decides the issue is whether
the depth of the cross-section is small (straight beam) or not (curved beam) compared to
the radius of curvature. Thus, the Winkler-Bach theory is to be used for curved beams.

We again use the Love-Kirchhoff assumption: that cross-sections which are plane before
bending continue to remain plane; they only rotate, but do not become crooked or wavy.
We shall see that the curvature introduces qualitative differences. The most important
of these differences is that (i) the distribution of strains and, therefore, of stresses is not
linear across the cross-section, (ii) the neutral axis does not pass through the centroid of
the cross-section. The distribution of the bending stresses is hyperbolic, and not linear,
and the neutral axis is shifted from the centroidal axis towards the centre of curvature. We
shall also see that these differences arise only if the depth of the beam is comparable to the
radius of curvature. Thus, a thin circular bar (with the radius of curvature several times,
say, 20 times, the depth of the cross-section) behaves like a straight beam in spite of its
obvious appearance as curved.

Fig. 2.16 shows a crane hook. The depth of the beam

is not small compared to the radius of curvature. A
few possible cross-sections are shown alongside.

The calculation of the property Z is a little compli-

Figure 2.16: A crane hook cated for such cross-sections.

We shall see that a certain property of the area of cross-section defined by

Z
1 y
Z=− dA
A A R+y
is important in curved beam calculations. This is somewhat, but not quite, analogous to I,
the second moment of the area of cross-section in straight beam theory. Z is dependent not
only on the geometric property of the cross-section, but also on the radius of curvature R.
Its numerical calculations can sometimes be tricky; unless special care is taken, the error
Advanced Mechanics of Solids 2-18

(a) A C-clamp (b) A crane hook

Figure 2.17: Two examples of a curved beam: a C-clamp and a crane hook

in the result can be serious. Additionally, in curved beam calculations the distance y is
usually reckoned from the centroidal axis, and not from the neutral axis. This is because
the neutral axis is not known a priori; it is known only after calculations. The centroidal
axis, on the other hand, is known as soon as the cross-section is specified.
These are some of the departures from the straight beam theory. These facts will
emerge as consequences of the theory that we are about to develop below known as the
Winkler-Bach theory14 .
Winkler-Bach Theory

(a) An element before and after deformation (b) Geometry of deformation

Figure 2.18: A curved beam element is shown. Plane sections before deformation continue
to remain plane even after deformation. The details of the deformation, the cross-section,
and the stresses on it can be seen.

Let us consider a curved beam element. As always we assume that cross-sections which
were plane before bending continue to remain plane even after bending. They do not become
crooked; they only rotate. Thus, the line A1 B1 rotates to become A0 B 0 [Fig. 2.18a]. The
bottom fibres are stretched out, and the top ones shortened. As we go up from the bottom

14
named after the German engineers E. Winkler and C. Bach
2-19 Special Problems in Bending

fibres to the top ones, there is one fibre that is neither elongated nor compressed. This is
the neutral line N N1 . The cross-section AB, because of symmetry, stays where it is without
any movement (rotation). Thus, the fibre P P1 is stretched to become P P10 , P1 P10 being the
stretch. This fibre is at a distance of y measured from the centroidal axis (and not from
the neutral axis) [Fig. 2.18a]. The cross-section is taken, to be specific and for convenience,
as a rectangle [Fig. 2.18b]. The coordinate axes x, y, z are as marked in Fig. 2.18b passing
through the centroid O of the cross-section.
Now we shall consider the deformations. As in the simple straight beam theory, the
deformation (displacement, elongation or compression) is again proportional to the distance
from the neutral axis, but the strains are not! This is because the fibres are not of equal
lengths: strain = change in length ÷ original length; the original lengths of the various
fibres are different! This is the first major departure from the straight beam theory. We
shall now take a closer look at the geometry of deformation, and work out the expression
for the strain [Figs 2.18].
The general procedure is natural and straightforward. Once the variation of the strain e
with y (measured from the centroidal axis) [Figs 2.18] is obtained, we can use the constitutive
equation σ = E e (simple Hooke’s law) and obtain the expression for the variation of the
stress σ with y. Next we use the equilibrium requirements and proceed further.
General procedure:
Our plan is to look at the geometry of deformation clearly [Fig. 2.17a] and to work out an
expression for the strain e at any fibre distant y from the centroidal axis. We then obtain
the corresponding expression for the stress σ using the constitutive equation σ = E e. Next,
if we use the equilibrium requirements (equilibrium equations) we can obtain the expression
for the bending moment M . By processing these expressions we can obtain the Winkler-
Bach formula expressing the bending stress σ in terms of the bending moment and the
properties of the cross-sectional area. (This formula takes the place of the Euler-Bernoulli
equation of the straight beam theory.) This formula will be derived below.
Detailed calculations:
The strain of the fibre at the centroidal axis serves as some kind of reference. Let us call it
ec . Note once again that this would be zero if we had considered the neutral axis!

change in length OO10 − OO1 O1 O10

ec = = = −→ O1 O10 = ec ds = ec R dθ
original length OO1 OO1
The strain of a fibre at a distance of y from the centroidal axis is now calculated as
 change in length P P10 − P P1 P1 P10 P1 H + HP10 O1 O10 + HP10
e = = = = = .

y original length P P1 P P1 P P1 P P1

Let us note that O1 O10 = ec R dθ; HP10 = O10 H(∆ dθ) = y ∆ dθ; and P P1 = (R + y) dθ.
The expression for the strain e at any distance y from the centroidal axis is, thus,
 ec R dθ + y ∆ dθ R ec + y(∆ dθ/dθ)
e = = .

y (R + y) dθ (R + y)
Advanced Mechanics of Solids 2-20

If we call (∆ dθ/dθ) as ω for convenience, we obtain

 R ec + y ω R ec + (y ec − y ec ) + y ω ec (R + y) + (ω − ec )y
e = = =

y R+y R+y R+y
y
= ec + (ω − ec ) . (2.23)
R+y
This is the variation of the strain with the distance y. Accordingly, the variation of the
bending stress σ is obtained as
 
 y
σ  = E e = E ec + (ω − ec ) . (2.24)

y R+y
Now let us consider the equilibrium requirements. They are
Z X
σ dA = 0 (no net axial force) Fx = 0; (2.25a)
ZA
(σ y) dA = Mz (bending moment about the z axis) = Mz . (2.25b)
A

Substituting the expression for σ [Eq. (2.24], these equilibrium requirements give us
Z Z  
y
σ dA = E ec + (ω − ec ) dA = 0 (2.26a)
R+y
ZA A
Z  
y
M = (σ y) dA = E y ec + (ω − ec ) dA = 0. (2.26b)
A A R+y
As the Young’s modulus E is a constant — this is so if there is only one material with
the usual simplifying assumptions (i) the behaviour is linear; and (ii) the values of E in
both tension and compression are the same — the above two equations [(2.26a) and (2.26b]
appear as follows.
Z Z
y
ec dA = −(ω − ec ) dA; and (2.27a)
A A R + y
y2
 Z Z 
M = E ec y dA + (ω − ec ) dA . (2.27b)
A A R+y

We can simplify these equations by using a notation Z and noting the following.
Z Z Z
1 y
Z=− dA; dA = A; y dA = 0 (first moment = 0.) (2.28)
A A R+y A A
y2
Z Z   Z
y y
dA = y−R dA = −R dA = Z A R (2.29)
A R+y A R+y A R+y

With these Eqs (2.27a) and (2.27b) are simplified as

ec = (ω − ec )Z; (2.30a)
M = E(ω − ec )Z A R. (2.30b)
2-21 Special Problems in Bending

Solving for ec and ω — that is, expressing these in terms of M, E, A and R, we obtain
 
M M 1 M M
ω − ec = ec = ω= + .
E Z AR E AR EA R RZ

If we substitute these expressions in Eq. (2.24), we arrive at the equation

   Z 
M 1 y 1 y
σ= 1+ where Z=− dA . (2.31)
AR Z R+y A A R+y

This is the Winkler-Bach formula giving the bending stress in terms of the applied bending
moment and the properties of the cross-section15 . [See the illustrative examples given in a
later chapter.]
Calculation of Z:
The calculation of the numerical value of Z is tricky even for simple sections like a rectangle
of a circle16 . For trapezia, and combinations of rectangles, analytical methods are possible.
For more complicated shapes, numerical integration will have to be used.
If we use the formula Z
1 y
Z=− dA
A A R+y
and try to evaluate this by integration — which is essentially a summation procedure —
the process involves addition of a series of terms. These are positive where y is positive,
and negative where y is negative. The procedure leads to finding the (small) difference
between two (relatively large) numbers. This can lead to serious errors unless special efforts
are taken to obviate the difficulty17 . The powerful and efficient Gaussian quadrature is
much better than the trapezoidal rule or even the Simpson’s rule for numerical integration.
Graphical integration used to be popular some time ago, but not any more, for a good
reason.
What happens when R is very large?
When R is very large, the curvature effects fade away as it were. It is then sufficient to use
the simple straight beam theory. (This is what we do in cases like the beam shown in Fig.
2.15a, where in spite of the obvious curvature of the beam, the straight beam theory can
be applied!) We shall now show that this is so.

15
The influence of [12] can be seen throughout this section on the topic of Curved Beams. This was one
of the books that my teacher, the distinguished professor C.N. Lakshminarayana of IIT Kharagpur, used
when we were students. The help is gratefully acknowledged. That first love for the book still remains
with me. There is a later edition of the book, but I am still proud of my old copy. Fond memories!
16
Calculation of Z for these cross-sections is shown later on pp.13-29, 13-30.
17
One method which is successfully used is to rewrite Z as

y2
Z
1
Z= dA
AR A R + y
where, unlike the earlier case, the integrand is always positive. Now in carrying out the integration process,
there is no more the difficulty indicated above. We do not propose to give further details here.
Advanced Mechanics of Solids 2-22

To do so, let us begin with the Winkler-Bach formula, and process it as shown below.
 
M 1 y M M y
σ= 1+ = +
AR Z R+y A R ZAR R + y
M M y M My
= + R y2 = + y
R y2 dA
AR dA R + y AR 1+
A R+y R A 1+y/R

My My
If R −→ ∞, σ −→ R 2
= ,
A y dA I
which is the straight beam formula.
Further beyond:
What we have obtained is the formula for bending (circumferential) stress. In addition,
there are radial stresses. These are zero on the outer (convex) and the inner (concave)
surfaces. They are small inside for solid cross-sections like trapezoidal, rectangular and
circular ones. However, they can be large and serious in the web for cross-sections like I,
H and T. These can be analysed without much difficulty, but we do not propose to do this
here18 .

In the next chapter, we shall discuss the index notation.

18
There are many details that must be understood by an engineer when he designs curved beams. But
we will not discuss them here, because this is only a gentle introduction to some advanced topics in the
mechanics of solids.
Chapter 3

INDEX NOTATION

In this chapter, we shall consider the index notation. The summation convention, which is
extensively used, is part of it. First we shall see why we need to use the index notation.
INTRODUCTION: WHY INDEX NOTATION?
The index notation and the summation convention are so much a part of tensor analysis
that they are sometimes referred to as the tensor notation. We shall see how these, together
with the Kronecker delta and the permutation symbols (defined later below), help us write
several important equations concisely. Using these methods we can quickly arrive at results
that are otherwise long and complicated. If we follow some systematic procedures when
manipulating these symbols and indices, we can work out complicated expressions correctly
without straining ourselves; good notations seem to be able to guide us to correct results
without much intellectual effort.
Let us now see how the index notation, together with two conventions, enables us to
pack a large number of similar looking equations in a capsule.
INDEX NOTATION
Let us consider a set of linear algebraic simultaneous equations:

a11 x1 + a12 x2 + a13 x3 = c1

a21 x1 + a22 x2 + a23 x3 = c2
a31 x1 + a32 x2 + a33 x3 = c3 .

These long equations (in matrix notation Ax = c) can be abbreviated as

3
X 3
X 3
X
a1j xj = c1 , a2j xj = c2 , a3j xj = c3 , or as
j=1 j=1 j=1
3
X
aij xj = ci , i = 1, 2, 3.
j=1
Advanced Mechanics of Solids 3-2

[Generally only subscripts such as aij are used when we work with cartesian tensors; su-
perscripts such as aij , or mixed superscripts and subscripts such as aij are used only when
dealing with general tensors. When superscripts are used such as, for instance, P 1 , P 2 , P 3 ,
we should realise that these numbers 1, 2, 3 are superscripts, and not powers. Thus, P 2 and
P 3 are not P squared and P cubed. This fact would be clear from the context. Thus, we
will use only subscripts in this book.]
Such sets of equations appear often enough that it is found convenient to introduce
two conventions. The same repeating pattern makes it possible to abbreviate the above
equations further as
aij xj = ci , (3.1)

where it is understood, according to the two conventions, that

(i) there is summation over the repeated index j, and that
(ii) the equation is valid for every (applicable) possible value of the index i.
In the above example, the index j is repeated in the term aij xj — let us call this a
term, even though it is really the sum of three separate terms — and, thus, according to the
convention (i) above, there is implied summation over the repeated index j. Furthermore,
the equation (3.1) — let us call this an equation, even though it is really three separate
equations1 — holds, according to the convention (ii) above, for every possible value2 of the
index i. The repeated index j is called a dummy index, because aij xj and, say, aik xk mean
the same3 . The index i is called the free (identifying) index. These conventions are called
Einstein’s conventions. To avoid ambiguity, we must be sure to see that the same index
does not appear more than twice in any ‘term’.
Generally Latin letters are used for the indices as subscripts (and superscripts). Some-
times Greek letters are also used4 . The letters at the end of the alphabet such as x, y, z
and u, v, w are not generally used for these indices serving as subscripts (and superscripts).
This is not a rule; it is a common practice just as we generally use the letters x, y, z, · · ·
for variables, and a, b, c, · · · for constants.
We can see readily that a matrix A can be represented conveniently by its components
aij , and its transpose AT by aji . The (row by column) multiplication of two matrices
Am×n (aij , i = 1, 2, 3, · · · , m; j = 1, 2, 3, · · · , n) and B n×p (bij , i = 1, 2, 3, · · · , n; j =

1
Shall we call this a three-in-one, if it is not blasphemous to consider such a trinity?
2
Applicable in the given situation. If there are three equations, i = 1, 2, 3.
3
In this regard, it is analogous to the (dummy) variable of integration in a definite integral

Z b Z b
f (x) dx = f (y) dy.
a a

4
For example, when we discuss the theory of shells, we encounter a situation of a two-dimensional surface
(the curved middle surface of the shell) embedded in the surrounding three-dimensional flat surface, R2
embedded in E 3 . In such cases, to distinguish clearly between R2 and E 3 , the Greek letters such as α and
β are used to refer to the Riemannian space R2 (curved surface of the shell) (α, β = 1, 2), while the Latin
letters such as i, j, k refer to the Euclidean space E 3 (i, j, k = 1, 2, 3).
3-3 Index Notation

1, 2, 3, · · · , p) can be indicated easily and naturally as aij bjk which is convenient for im-
plementation in a computer program.
Symmetry and skew-symmetry of (square) matrices can be defined and represented as

eij = eji (symmetric), and ωij = −ωji (skew-symmetric).

The elements on the leading diagonal (where i = j) of a skew-symmetric matrix like ωij
must necessarily be zero.
The well known result that any (square) matrix can be written as the sum of a symmetric
and a skew-symmetric one is easily demonstrated as follows.

1 1
aij = (aij + aji ) + (aij − aji ) ≡ eij + ωij ,
2 2

where
1 1
eij = (aij + aji ) = (aji + aij ) = eji (symmetric); and (3.2)
2 2
1 1
ωij = (aij − aji ) = (−aji + aij ) = −eji (skew-symmetric). (3.3)
2 2

A useful result that we often need to use when manipulating symmetric and skew-symmetric
matrices in index notation is eij ωij = 0.

eij ωij = eji ωij (symmetry, eij = eji ),

= −eji ωji (skew-symmetry, ωij = ωji ),
= −emn ωmn (i, j are both dummy indices; replace them by m, n)
= −eij ωij (for the same reason, replace them by i, j),
= 0 (something is equal to its own negative; each has to be zero). (3.4)

KRONECKER DELTA
We shall see that it is very convenient to use Kronecker5 delta defined as
(
0 if i 6= j
δij =
1 if i = j.

5
Leopold Kronecker (Dec. 1823 - Dec. 1891) was a wealthy and influential German mathematician. His
remark “Die ganzen Zahlen hat der liebe Gott gemacht, alles andere ist Menschenwerk” (“God created
the integers; everything else is the work of man.”) is well known. His major contributions are in elliptic
functions and the theory of algebraic numbers. He had unbelievably unusual, and sometimes seriously
wrong, academic convictions, and had bitter antagonism with several famous mathematicians including
Georg Cantor. He was the leader of the so-called ‘intuitionists’ (intuitionism stressing that intuition has
priority over logic in mathematics) in the battle of wits against the ‘formalists’ led by David Hilbert.
Hilbert referred to him as ‘Verbotsdiktator’ (forbidding dictator).
Advanced Mechanics of Solids 3-4

We begin to see the convenience immediately. This δij represents a unit matrix in index
notation. For a unit matrix (3 × 3, or 2 × 2, which is usually the case for our applications),
 
1 0 0  
1 0
I =  0 1 0 ; I=
0 1
0 0 1

For all the elements on the leading diagonal where i = j, the value is 1; for the off-
diagonal elements (i 6= j), the value is 0. We note in passing that a hydrostatic state of
stress is represented in matrix form as
   
−p 0 0 1 0 0
 0 −p 0  = −p  0 1 0  = −pI,
0 0 −p 0 0 1

and in index notation as −p δij .

The Kronecker delta often plays the role of a ‘substitution operator’. To see this, let
us consider a ‘term’ aij δjk . As the index j is repeated, there is running summation over j.
Thus,
Xn
aij δjk = aij δjk = ai1 δ1k + ai2 δ2k + · · · + ain δnk .
j=1

All these terms except one vanish because δ = 0 except when its subscripts are equal. As
k is a free index, it can take up different values in different expressions. Thus, we cannot
state which term (the first, second, · · · , etc.) survives. However, we can be sure that only
when j hits the value of k (it does not matter what value k takes), that term alone survives.
We can, therefore, conclude that
aij δjk = aik . (3.5)
The effect of ‘multiplying’ aij with δjk is simply this: the dummy index j in aij is substituted
by the free index k in the Kronecker delta. Hence the name substitution operator. The
matrix equivalent of Eq.(3.5) is AI = A.
Let us see one more simple example which might appear as contrived and trivial at first
sight. If x1 , x2 , x3 are independent variables, it is obvious that ∂x1 /∂x2 , ∂x2 /∂x3 , ∂x3 /∂x1 ,
etc., are all zero. It is also obvious that ∂x1 /∂x1 , ∂x2 /∂x2 , ∂x3 /∂x3 are all equal to 1. That
is, ∂xi /∂xj vanishes if the two indices i and j are different, but is equal to 1 when i and j
have the same value (1, 2 or 3). Thus, ∂xi /∂xj = δij
When we discuss tensors in general, transformation of the coordinates and the induced
transformation of the components of a tensor in these various coordinate systems are of
fundamental importance. When the axes are changed from the ‘old’ ones xi to the ‘new’
ones x0i , we would need to consider functional relations of the kind x0i = x0i (x1 , x2 , x3 ) and
xi = xi (x01 , x02 , x03 ). It would be necessary in this context to compute the partial derivatives
∂x0i /∂xj (and/or ∂xi /∂x0j ), and during simplifications the results in the example considered
will be useful.
3-5 Index Notation

AN EXAMPLE: RECTANGULAR ORTHOGONALITY OF AXES

We shall see, as an example, how the index notation, along with the two conventions
mentioned above, can be used to advantage in writing the conditions of orthogonality of
two sets of rectangular cartesian coordinate axes6 . This example is of importance in the
study of our subject (as in the transformation of stress components). But first we need to
recapitulate two simple, but important results relating to the direction cosines.
Direction Cosines: Two Simple Results

cos (Ox1 , OP ) = l1 , cos (Ox2 , OP ) = m1 , cos (Ox3 , OP ) = n1 (3.6a)

cos (Ox1 , OQ) = l2 , cos (Ox2 , OQ) = m2 , cos (Ox3 , OQ) = n2 (3.6b)

The following two equations are very useful.

Result (a): l2 + m2 + n2 = 1
Result (b): l1 l2 + m1 m2 + n1 n2 = cos θ

Figure 3.1: The direction cosines enjoy these two useful relations.

Result (a): l2 + m2 + n2 = 1
The components of OP along the coordinate directions are, accordingly,

OP cos (Ox1 , OP ) ≡ OP l1 (3.7a)

OP cos (Ox2 , OP ) ≡ OP m1 (3.7b)
OP cos (Ox3 , OP ) ≡ OP n1 . (3.7c)

If these three components are resolved back again along OP and added, we must obviously
obtain the same original length OP. Thus,

(OP l1 ) l1 + (OP m1 ) m1 + (OP n1 ) n1 = OP, (3.8)

showing that l12 + m21 + n21 = 1. Thus, the direction cosines (l, m, n) satisfy the relation
l2 + m2 + n2 = 1.
Result (b): l1 l2 + m1 m2 + n1 n2 = cos θ
Next, let us resolve the components [See Eqs (3.7a), (3.7b) and (3.7c).] along the direction
OQ and add up. Then we obtain

(OP l1 ) cos (Ox1 , OQ) + (OP m1 ) cos (Ox2 , OQ) + (OP n1 ) cos (Ox3 , OQ), (3.9)
6
Only such transformations from one rectangular cartesian coordinate system to another (rectangular carte-
sian coordinate system) are relevant when we work with only cartesian tensors, and not with general
tensors.
Advanced Mechanics of Solids 3-6

Old coordinates
x1 x2 x3
New coordinates x01 a11 a12 a13
x02 a21 a22 a23
x03 a31 a32 a33

Table 3.1: Table of direction cosines

which must be the same as the projection of the length OP along OQ, which is obviously
OP cos (OP, OQ). We, thus, obtain the result

(OP l1 ) l2 + (OP m1 ) m2 + (OP n1 ) n2 = cos (OP, OQ) ≡ cos θ, (3.10)

showing that
l1 l2 + m1 m2 + n1 n2 = cos θ ≡ cos (OP, OQ). (3.11)

There are two important, special cases of this equation (3.11): (i) when θ = 0 (OP and
OQ coincide), and (ii) when θ = π/2 (OP and OQ are perpendicular — orthogonal — to
each other). In these two special cases, we have

l2 + m2 + n2 = 1, when l1 = l2 = l; m1 = m2 = m; n1 = n2 = n; and (3.12a)

l1 l2 + m1 m2 + n1 n2 = 0, when θ = π/2. (3.12b)

These are the two simple, but important results mentioned in the caption of this subsection.
Next we shall see the matrix of the direction cosines.
Matrix of the Direction Cosines
Let us consider two sets of rectangular cartesian coordinate systems, the ‘old’ xi and the
‘new’ x0i (i = 1, 2, 3). (See the figure by the side of Table 3.1.) The orientation of the new
coordinate system x0i with respect to the old one xi is given by the direction cosine matrix
given in Table 3.1.
We shall now see how the index notation can be conveniently used to write down the
orthogonality relations7 .
(a) x10 , x20 , x30 are mutually orthogonal.
These ‘new’ axes are mutually orthogonal: their direction cosines w.r.t. the ‘old’ axes
x1 , x2 , x3 should satisfy the orthogonality conditions. This case is examined below.
7
These are simple to understand. However, experience shows that they are also confusing unless we are
mentally alert, at any rate, when we are learning this for the first time. Therefore, we shall develop the
equations slowly and patiently.
3-7 Index Notation

Step 1:
Let us observe that the axis x01 , with the direction cosines (a11 , a12 , a13 ) w.r.t. the ‘old’ axes
(x1 , x2 , x3 ), is perpendicular to the axis x02 , with the direction cosines (a21 , a22 , a23 ) w.r.t.
the ‘old’ axes (x1 , x2 , x3 ). Thus, we see from Eq: (3.12b) that

a11 a21 + a12 a22 + a13 a23 = 0. (3.13)

This is the same as Eq. (3.12b) with the direction cosines suitably interpreted so as to be
applicable to this context. Note that the second subscripts8 in the three terms are 11, 22, 33.
Recalling the summation convention, this equation in longhand written as
3
X
a11 a21 + a12 a22 + a13 a23 ≡ a1j a2j = 0
j=1

may be written in index notation as a1j a2j = 0 (repeated index j; implied running sum-
mation over all values of j; here j = 1, 2, 3).
In exactly the same way, we argue that the axes x02 with the direction cosines (a21 , a22 ,
a23 ) w.r.t. the ‘old’ axes (x1 , x2 , x3 ), and x03 with the direction cosines (a31 , a32 , a33 ) w.r.t.
the ‘old’ axes (x1 , x2 , x3 ) are perpendicular to each other. This observation enables us to
write
X3
a21 a31 + a22 a32 + a33 a23 ≡ a2j a3j = 0.
j=1

Thus, we note the following equations.

3
X
a11 a21 + a12 a22 + a13 a23 ≡ a1j a2j = 0 (axes 1 and 2 perpendicular), (3.14a)
j=1
3
X
a21 a31 + a22 a32 + a23 a33 ≡ a2j a3j = 0 (axes 2 and 3 perpendicular), and (3.14b)
j=1
3
X
a31 a11 + a32 a12 + a33 a13 ≡ a3j a1j = 0 (axes 3 and 1 perpendicular). (3.14c)
j=1

Using the index notation, these three can be written in capsule form as

aij akj = 0 (i 6= k),

where the dummy index j runs through the entire set of values 1, 2, 3. Furthermore, this
equation is valid for each of the values of 1, 2, 3 for each of the free indices i and k, (i 6= k).
8
Such underlining is not resorted to in actual practice. Here we have done so to attract attention, and,
thus, to have greater clarity. Underlining like aii or as ai i is sometimes used if we wish to indicate that
the summation is suspended.
Advanced Mechanics of Solids 3-8

They together tell us that the axes x01 , x02 , x03 are mutually perpendicular. Notice that
the running sum (repeated index) is on the second subscript. Notice further that the first
subscript refers to the axis concerned, 1 for x01 , 2 for x02 , 3 for x03 .
Step 2:
Now if we had taken the x01 axis, and had computed the sum of the squares of its direction
cosines, we would have obtained the result 1. See Eq. (3.12a).

a211 + a212 + a213 = 1.

A similar result holds for the other two ‘new’ axes, x02 and x03 , also.

a221 + a222 + a223 = 1, and

a231 + a232 + a233 = 1.

Step 3:
Combining the results shown above in Steps 1 and 2, we obtain the following.
The first set of three equations below, Eqs (3.15a), (3.15b), (3.15c), refer to two different
axes; the result is 0 then. The three equations in the second set, Eqs (3.16a), (3.16b), (3.16c),
refer to the same axis; the result is 1 then.
3
X
a11 a21 + a12 a22 + a13 a23 ≡ a1j a2j = 0 (different axes 1 and 2), (3.15a)
j=1
3
X
a21 a31 + a22 a32 + a23 a33 ≡ a2j a3j = 0 (different axes 2 and 3) (3.15b)
j=1
3
X
a31 a11 + a32 a12 + a33 a13 ≡ a3j a1j = 0.(different axes 3 and 1), (3.15c)
j=1

3
X
a11 a11 + a12 a12 + a13 a13 ≡ a1j a1j = 1 (same axis, 1 and 1), (3.16a)
j=1
3
X
a21 a21 + a22 a22 + a23 a23 ≡ a2j a2j = 1 (same axis 2 and 2) (3.16b)
j=1
3
X
a31 a31 + a32 a32 + a33 a33 ≡ a3j a3j = 1 (same axis 3 and 3). (3.16c)
j=1

Let us look at these equations carefully and understand what they mean. It is, in words,
this: (i) when the direction cosines of two different axes such as x01 and x02 are multiplied
together and added, we get the result 0 (as demonstrated in Step 1, Eqs (3.14a), (3.14b),
(3.14c)), and (ii) when the direction cosines of the same axis are squared and added, we
get the result 1 as demonstrated in Step 2, Eqs (3.16a), (3.16b), (3.16c)). To repeat for
3-9 Index Notation

further clarification, when the free indices (i and k) in aij akj , shown in bold here to attract
attention, are different, the result is 0. When, instead, these are the same, the result is 1.
We can exploit the conveniently defined Kronecker delta, and express the result together in
one equation as shown below in Eq. (3.17).
Thus, we can combine all the six (6) equations and display them together in a small
compact form using the index notation as

aij akj = δik . (3.17)

Now we turn around and notice that the ‘old’ axes (x1 , x2 , x3 ) are mutually orthogonal too.
These orthogonality conditions are similar, except that now the summation is on the first
subscript. We shall examine this case below, though not in as great detail as before.
(b) x1 , x2 , x3 are mutually orthogonal.
As the axes x1 and x2 axes are perpendicular, their direction cosines satisfy the equation

a11 a12 + a21 a22 + a31 a32 = aij aik = 0. (3.18)

This equation is similar to Eq. (3.13), except that the running sum now is on the first
subscripts, underlined as before to attract attention.
Orthogonality relations:
Thus, in conclusion, the orthogonality relations are

aij akj = δik (x01 , x02 , x03 are mutually orthogonal), and (3.19a)
aij aik = δjk (x1 , x2 , x3 are mutually orthogonal). (3.19b)

These are important results. We can see that, while these are not difficult, they can be
confusing. Some practice is needed before the notation and its message become familiar
and effortless to understand.
The index notation becomes more useful and powerful only when accompanied by Levi-
Civita’s permutation symbols. We shall discuss them in the next section.
PERMUTATION SYMBOLS
Permutation symbols are a system of 3-index symbols. They are also known as an e-system,
Levi-Civita symbols, anti-symmetric symbols, or alternating symbols. These are part of the
index notation, and are very useful when dealing with it.
Definition
This permutation symbol eijk is defined in 3 dimensions as follows.

+1 if i, j, k is an even permutation of 1, 2, 3;

eijk = −1 if i, j, k is an odd permutation of 1, 2, 3; and (3.20)

0 in other cases when two or three indices are the same.


We shall clarify this further. This permutation symbol eijk is equal to

Advanced Mechanics of Solids 3-10

ˆ +1 if these three indices are in an even permutation of the sequence 1, 2, 3 (cyclic

sequence 1, 2, 3; or 2, 3, 1; or 3, 1, 2);
ˆ −1 if these three indices are in an odd permutation of the sequence 1, 2, 3 (anti-cyclic
sequence 1, 3, 2; or 2, 1, 3; or 3, 2, 1);
ˆ 0 in other cases, that is, if any two or all the three indices are the same (equal) (not
cyclic; not anti-cyclic; acyclic 1, 1, 2; or 1, 2, 1; or 3, 3, 3; 3, 2, 2; etc.). See Figs 3.2a
and 3.2b.

(a) Even permutation: +1 (b) Odd permutation: −1

Figure 3.2: Even permutations of 1, 2, 3 are shown on the left hand side (Fig. 3.2a), and
odd ones on the right (Fig. 3.2b). Thus, for example, e123 = e231 = e312 = +1, while
e132 = e213 = e321 = −1.

They are also defined in 2 dimensions as follows.


+1 if i, j is an even permutation of 1, 2;

eij = −1 if i, j is an odd permutation of 1, 2; and (3.21)

0 in other cases when the two indices are the same.


This definition can be shown in matrix form as

   
e11 e12 0 1
= (3.22)
e21 e22 −1 0
[They can be defined similarly in higher dimensions also. For example, in 4 dimensions,

+1 if i, j, k, l is an even permutation of 1, 2, 3, 4;

eijkl = −1 if i, j, k, l is an odd permutation of 1, 2, 3, 4; and (3.23)

0 when any two, any three, or all the indices are the same.


We hardly ever use these symbols in higher dimensions. Our interest is usually only in 2
and 3 dimensions.]
Expansion of Determinants Using eijk
The long expanded form of any determinant may be recast in abbreviated form using these
symbols. We shall be concerned mostly with 3×3, and occasionally with 2×2, determinants
in the context of applications to our subject.
3-11 Index Notation

A 3 × 3 determinant  
 a11 a12 a13 
 
a =  a21 a22 a23  ≡ |aij | (3.24)

 a31 a32 a33 
may be written in the form

a = ai1 aj2 ak3 eijk (3.25)

= a11 × [its cofactor] + a12 × [its cofactor] + a13 × [its cofactor]

which is expansion of the determinant by column, here by the first column (as here a11 , a12 , a13
are the elements in the first column).
It can also be written in an alternative form

a = a3i a2j a1k eijk (3.26)

= a31 × [its cofactor] + a32 × [its cofactor] + a33 × [its cofactor]

which is expansion of the determinant by row, here by the third row (as here a31 , a32 , a33
are the elements in the third row).
[It would be nice if this expression is patiently written out in full so that the pattern
becomes clear. Another possibility is to write a similar expression for the simpler case of a
2 × 2 determinant, and work it out fully. In any case, it is necessary to convince ourselves
that out of the 3 × 3 × 3 = 27 terms (i, j, k = 1, 2, 3), only the expression shown above
survives. We shall indicate below the kind of steps to be taken and the kind of arguments
to be used. The full expansion is not completely worked out.

a3i a2j a1k eijk = a31 × [a2j a1k e1jk ] + a32 × [a2j a1k e2jk ] + a33 × [a2j a1k e3jk ]
= a31 × [{a21 a1k e11k } + {a22 a1k e12k } + {a23 a1k e13k }]+
a32 × [· · · ] + a33 × [· · · ]

Let us note that

(i) the first ‘term’ {a21 a1k e11k } in the last equation need not be further expanded because
the factor e11k = 0 for all values of k;
(ii) the second ‘term’ {a22 a1k e12k } is to be expanded over all values of k = 1, 2, 3. How-
ever, only e123 is non-zero; e121 = e122 = 0; and that
(iii) similarly the ‘third’ term {a23 a1k e13k }, when expanded, leads to the only non-zero
term {a23 a12 e132 }, which is equal to −{a23 a12 }, because e132 = −1 (anti-cyclic
sequence of 1, 2, 3).]

If, in either of the two forms (of expansion), two subscripts are interchanged, the value
of the determinant becomes −a. This is not difficult to understand; we know that the sign
of a determinant is changed when two columns (rows) are interchanged. Thus,

a = ai1 aj2 ak3 eijk

Advanced Mechanics of Solids 3-12

−a = ai2 aj1 ak3 eijk (interchange of two columns, the first and the second)
= a3i a2j a1k eijk (interchange of two rows, the first and the third).

We can use such expressions, manipulate them, and simplify them by using the rules to
prove the results of operation on determinants. For example, the determinant of a product
matrix P of two component square matrices A and B (that is, P = AB), is the product of
the determinants. Thus, if

P = AB, then |P | = |A| |B|,

where
|A| ≡ a ≡ |aij |, |B| ≡ b ≡ |bij |, and |P | ≡ p ≡ |pik | = |aij bjk |.

When any two columns (rows) of a determinant are the same, the value of the determi-
nant is zero. This can be seen easily from Eqs (3.25) and (3.26), because eijk = 0 whenever
any two indices are the same.
Cross Product of Vectors Using eijk
Cross products of vectors can be conveniently expressed in index notation using these per-
mutation symbols. We first begin by noting the following relationships among the unit
vectors (i, j, k) in a rectangular cartesian coordinate system.

i × j = k; j × k = i; k × i = j;
j × i =−k; k × j =−i; i × k = −j;
i× i = 0; j × j = 0; k × k = 0;

or equivalently,

e1 × e2 = e3 ; e2 × e3 = e1 ; e3 × e1 = e2 ;
e2 × e1 =−e3 ; e3 × e2 =−e1 ; e1 × e3 = −e2 ;
e1 × e1 = 0; e2 × e2 = 0; e3 × e3 = 0;

which are all condensed into the following ‘single’ equation that states that the three coor-
dinate axes are orthogonal: ei × ej = eijk ek .
We are now ready to consider the cross product of any two vectors A and B using the
above result.

P ≡ A× B = (Ai ei ) × (Bj ej )
= (Ai Bj )(ei × ej ) = Ai Bj eijk ek .

If P is written as Pk ek , we obtain P = Pk ek = Ai Bj eijk ek , which is the same in

content as
Pk = Ai Bj eijk . (3.27)
3-13 Index Notation

Triple Scalar Product

The triple scalar product (A × B) . C of three vectors A, B and C can be represented in
index notation as
A × B . C = Ai Bj Ck eijk .
This can also be written as any one of the expressions shown below without any ambigu-
ity. Let us note that the positions of the dot and the cross in a triple scalar product are
immaterial as long as the sequence A, B, C is maintained as an even sequence. When the
sequence is broken, the sign of the triple scalar product is reversed.

A × B . C = Ai Bj Ck eijk = A . B × C = B . C × A = C . A × B. (3.28)

Cross Product and Triple Scalar Product: Physical Meaning

(a) Cross product of two vectors (b) Triple scalar product

Figure 3.3: Cross product A × B of two vectors [Fig. 3.3a] represents the shaded area
OADB; the cross product is along the normal to the shaded area. The triple scalar product
A × B ) . C represents the volume of the parallelepiped shown in dotted lines [Fig. 3.3b].
(A

These products, the cross product and the triple scalar product, can be given physical
or geometrical meanings.
The cross product represents the area of a parallelogram defined by the two vectors A
and B as shown in Fig. 3.3a. The surface area can be given a direction also, viz., the
direction of the normal to the surface. This observation will be useful when we discuss the
area on which a stress component (or a stress vector) acts.
The triple scalar product (A × B) . C represents the volume of the block in Fig. 3.3b.
It is clear from the figure also that the above triple scalar product may be written in any
one of the forms given in Eq. (3.28), because all of them represent the same volume.
If a triple scalar product vanishes, the three vectors concerned are coplanar. That is, if
(A × B) . C = 0, it means that A , B , C are coplanar. Then, any one of the three vectors
can be written as a linear combination of the other two as A = α B + β C .
Moment of a Force
The moment M of a force F about a point P (Fig. 3.4a), we may recall from elementary
mechanics, is given by M = r × F, where r is the position vector. The equation may be
Advanced Mechanics of Solids 3-14

(a) Moment of a force F about a point P = (b) Moment of a force F about a point P in a
r ×F given direction AB = (rr × F ) . n

Figure 3.4: Moment of a force F about a point P = r × F [Fig. 3.4a], where r is the position
vector. The component along the line AB of the moment of the force F about the point P
is (rr × F ) . n [Fig. 3.4b], where n is the unit vector along AB.

written in index notation as

Mk = ri Fj eijk .
The component of this moment M along an axis, say, AB is a scalar, and is represented by
the triple scalar product (r × F) . n, where n is a unit vector along AB [See Fig. (3.4b).]
This equation may be written as

MAB = ri Fj nk eijk . (3.29)

We may, if desired, set up a coordinate system, work out the moments from first principles,
and convince ourselves of the correctness of these equations.
PERMUTATION IDENTITY (ee − δ IDENTITY)
The permutation symbols and the Kronecker deltas are related by an identity known as
the permutation identity or e − δ identity. This is useful when manipulating expressions
and arriving at relationships in index notation. It is a convenient tool for simplification of
expressions in index notation. It reads

eijk eilm = δjl δkm − δjm δkl . (3.30)

Two other results involving these permutation symbols are

eilm ejlm = 2δij and (3.31)

eijk eijk = 6. (3.32)

The corresponding results in 2 dimensions are

eij emn = δim δjn − δin δjm ; (3.33)

eij ein = δjn ; and (3.34)
eij eij = 2. (3.35)
3-15 Index Notation

A FEW ILLUSTRATIVE EXAMPLES

We shall discuss and work out a few illustrative examples below for further clarification.
We hope these will be helpful for beginners.
Example 1
When we substitute one expression in another, we must be sure to avoid having the same
index appearing more than twice in any ‘term’. For example, if we have an equation
aij xi yj = 0, and if we wish to replace the variables xi ’s by a new set of variables zi ’s linearly
related by xi = bij zj , a straightforward substitution gives us an equation aij bij zj yj = 0,
which is meaningless, the index j appearing too many times in the same ‘term’. In such
cases, the proper way to deal with this is to write the first (original) equation in terms of
some other dummy indices, say k and l, as akl xk yl = 0, where xk = bkj zj . Thus, we obtain
akl bkj zj yl = 0, which is a valid equation.
Example 2
Solve a system of linear, simultaneous, algebraic equations aij xj = ci .

aij xj = ci ; Ax = c.

Multiply both sides by bkl , the inverse of the matrix aij (and sum over the index i by setting
l = i). Thus,

bkl aij xj = bkl ci ; BA x = Bc; A−1 A x = A−1 c.

That is, by setting l = i, bki aij = δkj . Hence we have

δkj xj = bki ci ; I x = B c ≡ A−1 c;

xk = bki ci ; x = B c = A−1 c.

Note that we used the substitution property of the Kronecker delta. We finally obtain the
solution as xk = bki ci . We further note that the inverse of a (square) matrix exists9 only if
its determinant does not vanish10 .
9
Given a (square) matrix (corresponding to such sets of linear, simultaneous, algebraic equations), how
its inverse can be calculated, and questions of the existence and uniqueness of the solution, and related
matter are discussed in great detail in books on Linear Algebra.
10
When solving technologically important problems as, say, when using the finite element method, the size of
the matrix is usually very large. In such cases, calculating the inverse to solve a set of linear, simultaneous,
algebraic equations is not computationally wise; it is never done in practice. To demonstrate this in the
simplest, almost trivial case, let us solve the equation 3x = 6 in one unknown. Calculating the inverse as
1/3 involves one operation, and solving for x involves another operation, viz., 1/3 × 6. On the other hand,
a straightforward solution (by Gaussian elimination) gives us the answer as x = 6/3 = 2 in 1 operation
(compared to 2 operations in the previous case).
There are several details that are of great importance when a large number of algebraic, linear, simultaneous
equations are to be solved. These cannot be discussed here. Advanced books like Bathe, K.J., Finite
Element Procedures, 2nd ed., Watertown, MA, (2014), and Datta, B.N., Numerical Linear Algebra and
Applications, 2nd ed., SIAM, (2000) have to be consulted.
Advanced Mechanics of Solids 3-16

Example 3
Write down the following equation in full longhand. The comma (,) denoted partial differ-
entiation11 with respect to the space variable xi .

σji,j + Xi = 0.

This is sometimes written as σij,j + Xi = 0, because σji = σij (symmetry of the stress
matrix).
(Summation over the repeated (dummy) index j; hence there is summation over the
entire set of values for j : (j = 1, 2, 3)). Thus,

σ1i,1 + σ2i,2 + σ3i,3 + Xi = 0.

(Free index i; hence separate equations corresponding to all (possible) values of i : (i =

1, 2, 3)). Thus, we have

i = 1; 1 ≡ x1 ≡ x; X1 ≡ X σ11,1 + σ21,2 + σ31,3 + X1 =0; that is,

∂σxx ∂σyx ∂σzx
+ + + X=0;
∂x ∂y ∂z
i = 2; 2 ≡ x2 ≡ y; X2 ≡ Y σ12,1 + σ22,2 + σ32,3 + X2 =0; that is,
∂σxy ∂σyy ∂σzy
+ + + Y =0;
∂x ∂y ∂z
i = 3; 3 ≡ x3 ≡ z; X3 ≡ Z σ13,1 + σ23,2 + σ33,3 + X3 =0; that is,
∂σxz ∂σyz ∂σzz
+ + + Z =0.
∂x ∂y ∂z

We shall see later (p. 5-4) that these are the differential equations of equilibrium12 in the
mechanics of solids or the theory of elasticity.
Example 4
Write down the following equation in full13 . The comma (,) again stands for partial differ-
entiation with respect to the space variable indicated immediately after the comma.

1
eij = (ei,j + ej,i )
2
11
Here in this book we deal with only cartesian tensors; general tensors are not considered. For general
tensors, a comma stands for what is called covariant differentiation. For the simplified case of cartesian
tensors, this is greatly simplified to be just the usual partial differentiation with respect to the space
variable indicated immediately after the comma.
12
The shear stresses are written as τxy instead of as σxy , etc.. We, engineers, know that when the subscripts
are different, the stress components are shear stresses.
13
This is quite similar to the previous example. Yet we write it out in full, because these two examples 3
and 4 are of much importance to the mechanics of solids and / or the theory of elasticity.
3-17 Index Notation

There is no repeated index here; thus, there is no running summation. The free indices i
and j take on different values 1, 2, 3 in the nine (9) equations.
1 ∂u
i = 1; j= 1; e11 = (u1,1 + u1,1 ) ; exx =
2 ∂x 
1 1 ∂u ∂v
i = 1; j= 2; e12 = (u1,2 + u2,1 ) ; exy = +
2 2 ∂y ∂x
 
1 1 ∂u ∂w
i = 1; j= 3; e13 = (u1,3 + u3,1 ) ; exz = +
2 2 ∂z ∂x
 
1 1 ∂v ∂u
i = 2; j= 1; e21 = (u2,1 + u1,2 ) ; eyx = +
2 2 ∂x ∂y
 
1 ∂v
i = 2; j= 2; e22 = (u2,2 + u2,2 ) ; eyy =
2 ∂y
 
1 1 ∂v ∂w
i = 2; j= 3; e23 = (u2,3 + u3,2 ) ; eyz = +
2 2 ∂z ∂y
 
1 1 ∂w ∂u
i = 3; j= 1; e31 = (u3,1 + u1,3 ) ; ezx = +
2 2 ∂x ∂z
 
1 1 ∂w ∂v
i = 3; j= 2; e32 = (u3,2 + u2,3 ) ; ezy = +
2 2 ∂y ∂z
 
1 ∂w
i = 3; j= 3; e33 = (u3,3 + u3,3 ) ; ezz =
2 ∂z
Again, we shall see later that these are the strain-displacement relations. (Although there
are 3 × 3 = 9 equations here, only six (6) of them are independent because of the symmetry,
eij = eji ).
Example 5
Carry out the partial differentiation indicated below, and obtain the result in index notation.
The C’s are all constants.
∂
(Ckl xk xl )
∂xi
 
∂ ∂xl ∂xk
(Ckl xk xl ) = Ckl xk + xl
∂xi ∂xi ∂xi
= Ckl (xk δli + xl δki )
= Cki xk + Cil xl
= Cki xk + Cik xk , changing the dummy index l to k
= (Cki + Cik ) xk

∂xl
Note that we have used the result = δli above. We have also made use of the substitution
∂xi
property of the Kronecker delta.
Advanced Mechanics of Solids 3-18

Example 6
(a) The dot product of two vectors A and B is given by
A . B = Ai Bi = Ai δij Bj = δij Ai Bj .
(b) δii = δ11 + δ22 + δ33 = 3
(c) δij δjk = δi1 δ1k + δi2 δ2k + δi3 δ3k ,
which is equal to 1 if i = k, and 0 if i 6= k. Thus, it is equal to δik . It does not matter
whether i = j = 1 or 2 or 3. Only one of the three terms above survives.
This result is more readily obtained by invoking the substitution property of the Kro-
necker delta: δij δjk = δik .
(d)

βij ai bj 6= βij bi aj
(βij + βji ) ai bj 6= 2βij ai bj
βij (ai + bj ) 6= βij ai + βij bj

(e) βimn (ai + bm ) cn 6= βimn ai cn + βimn bm cn (No free index on the left hand side!)
(f)

βij (aj + bj ) = βij aj + βij bj

βij ai bj = βij bj ai
βij ai aj = βij aj ai

(g)

(βij + βji ) ai aj = 2βij ai aj

(βij − βji ) ai aj = 0

(h) δij eijk = 0 (Whenever i 6= j, δij = 0, and whenever i = j, eikj = 0, two indices of the
permutation symbol being equal.)

Example 7
If eijk σjk = 0, show that σij = σji .
To see this, all that is needed is to write down this equation for each value of i. For
example, if i = 1, we obtain e1jk σjk = 0, which means

e11k σ1k + e12k σ2k + e13k σ3k = 0.

Recalling that the permutation symbols eijk vanish whenever two indices are equal (the
same), we see from the above equation that

e12k σ2k + e13k σ3k = 0; that is,

3-19 Index Notation

(e121 σ21 + e122 σ22 + e123 σ23 ) + (e131 σ31 + e132 σ32 + e133 σ32 ) = 0.

Again noting that only the third term in the first set of braces ( ), and the second in the
second set survive, we obtain
e123 σ23 + e132 σ32 = 0,
which implies that σ23 = σ32 , (because e123 = 1 and e132 = −1). The other two similar
sets of equations, viz., σ13 = σ31 and σ21 = σ12 , can similarly be obtained by setting the
values of i to 2 and 3. We shall see later that these are the consequences, (the shear and
the complementary shear being equal), of the equations of moment equilibrium.
Example 8
The vector triple product A × (B
B × C ) = (A
A . C ) B − (A
A . B)C.
A typical component, say, the first, of the left hand side, is

A × (B
[A B × C )]1 = A2 (B1 C2 − B2 C1 ) − A3 (B3 C1 − B2 C3 )
= B1 (A2 C2 + A3 C3 ) − C1 (A2 B2 + A3 B3 )
= B1 (A1 C1 + A2 C2 + A3 C3 ) − C1 (A1 B1 + A2 B2 + A3 B3 )
A . C ) − C1 (A
= B1 (A A . B)

If the second and the third terms are also processed similarly, and all the results combined
together, we obtain the result A × (B
B × C ) = (A
A . C ) B − (A
A . B ) C . Let us not fail to notice
that the brackets are important here, because A × (B B × C ) 6= (AA × B) × C.
Example 9
A linear transformation yi = aij xj from the variables xi ’s to the variables yi ’s is followed
by another linear transformation zi = bij yj . Write down the product transformation (after
the two successive transformations) from xi ’s to zi ’s.
If we substitute yi = aij xj directly into the equation zi = bij yj , we would have the index
j too many times in the same ‘term’. This, as explained earlier, becomes quite meaningless.
Thus, the equation yi = aij xj is first to be rewritten as yj = ajk xk before the substitution
is made. Then we obtain zi = bij yj = bij ajk xk . The equivalent matrix equations are
Y = AX , Z = B Y , and Z = B AX . We note that the indicated order is to be maintained,
as the products B A and AB are, in general, different.
Example 10
Expand a determinant |aij | in terms of these elements and their cofactors.
We know that a determinant |aij | ≡ a may be expanded row-wise or column-wise in
terms of its elements and their cofactors. Thus, if Aij is the cofactor of the element aij ,

|aij | ≡ a = aij Aij (row-wise Laplace expansion, no sum on i);

|aij | ≡ a = aij Aij (column-wise Laplace expansion, no sum on j)
Advanced Mechanics of Solids 3-20

We further know that, if the elements of any row / (column) are multiplied by the cofactors
of any other row / (column) and added up, the result is zero. Thus,

aij Akj = δik a (row-wise Laplace expansion), and

aij Ajk = δjk a (column-wise Laplace expansion).

Example 11
Use the above result to derive Cramer’s rule for the solution of a system of linear, simulta-
neous, algebraic equations aij xj = ci .
Multiply both sides of the equation by Aik (and sum over the index i). This gives us

Aik aij xj = Aik ck ;

i.e., a δkj xj = Aik ci −→ a xk = Aik ci ,
Aik ci
giving us the result, xi = .
a

Example 12
A × B ) . (C
Show that (A C × D ) = (A B . D ) − (A
A . C )(B A . D )(B
B . C ).
We recall that

A . B = (Ai ei ) . (Bj ej ); and

A × B = (Ai ei ) × (Bj ej ) = Ai Bj eijk ek .
Thus, A × B ) . (C
(A C × D ) = [(Ai e i ) × (Bj e j )] . [(Ck e k ) × (Dl e l )]
= (Ai Bj eijk e k ) . (Ck Dl eklm e m )

Now when the two (· · · ) and (· · · ) are taken together, there are too many k’s as indices.
The indices in the second (· · · ) are, therefore, replaced by another set of indices, say, p, q
and r. Accordingly,

A × B ) . (C
(A C × D ) = [(Ai e i ) × (Bj e j )] . [(Ck e k ) × (Dl e l )]
= (Ai Bj eijk e k ) . (Cp Dq epqr e r )
= Ai Bj Cp Dq eijk epqr δkr
= Ai Bj Cp Dq eijk epqk because epqr δkr = epqk
= Ai Bj Cp Dq (δip δjq − δiq δjp ) using e − δ identities

A FEW PROBLEMS FOR EXERCISE

To gain mastery over the manipulations with the index notation, it is necessary to work
out some problems. A few problems are given below for exercise.
3-21 Index Notation

1. The strain-displacement relations when the displacements are large are given by
1
eij = (ui,j + uj,i + uk,i uk,j )
2
Write these equations in full in terms of the variables x, y, z. (Note the presence of the
nonlinear terms uk,i uk,j . Such terms introduce geometric nonlinearity and consequent
difficulties in the mathematical theory of elasticity.)

2. The stress-strain relations for an isotropic, linearly elastic material are given by the
so-called generalised Hooke’s law

σij = λ δij ekk + 2 µ eij .

How many separate (independent) equations are contained in this? Write down these
equations in full in terms of the variables x, y, z.

3. Show, using the index notation and operations, that the following scalar triple products
are equal (each representing the volume of a parallelepiped defined by the three vectors,
as explained earlier).
A × B .C = A.B × C

4. Evaluate eijk eijk and eijj .

5. Expand the expression aij xi xj for a bilinear form in full. How many terms does this
have?

6. The stress components in a Newtonian fluid are given by the relation

2
σij = −p δij + 2 µ eij − δij ekk .
3
Write this out in full to obtain all the equations explicitly. Also express the above
equation in the usual Gibbs’ notation of boldface for vectors.

7. Expand in full the expressions given below (which define the ‘metric properties’ of a
metric space).
(ds)2 = gij dxi dxj ; (ds)2 = δij dxi dxj

8. Write down, in index notation, the following quadratic form:

Q ≡ X T AX,

where X is the row vector [xi x2 x3 ]. Under what condition is this quadratic form
positive definite? What is the matrix A that corresponds to the following quadratic
form?
4x2i + 6x22 + 3x23 + 6x1 x2 + 5x2 x3 − 2x3 x1
Is the matrix unique? Can we consider the matrix A as symmetric without any loss
in generality? Diagonalise the matrix into its diagonal canonical form, and write the
quadratic form in terms of the new variables y1 , y2 , y3 .
 Advanced Mechanics of Solids 3-22

9. The quadratic forms

1 1
T = akl q̇kl ql and V = bkl qk ql
2 2
represent, respectively, the kinetic and potential energies of a certain dynamical system
that has the n generalised coordinates qk ’s and the n generalised velocities q̇k ’s. The
Lagrange’s equation of motion for this (conservative) system are

akl q̈l + bkl ql = 0.

The constants akl and bkl are symmetric. Write down the Lagrange’s equations in full
explicitly.

10. Calculate
∂
(bklm yk yl ym )
∂yk
and show that it is equal to

(bmkl + bkml + bklm ) yk yl ,

if bklm are all constants.

A . B ) C + A × D = E in Gibbs’ vector
11. Recast in the index notation the equation (A
notation.

12. Show, using the index notation, that (A B)T = B T AT .

CLOSURE
We hope we have demonstrated at least some of the advantages of using the index notation.
We shall follow up on these ideas by applying these techniques to the ideas and the basic
equations of solid mechanics.

We shall discuss in the next chapter the important topic of the state of stress at a point.
Chapter 4

THE STATE OF STRESS AT A

POINT

The state of stress at a point is a topic of fundamental importance. It is absolutely essential

to understand all the aspects (both physical and mathematical) related to this clearly. This
is a most crucial topic to master. We cannot learn basic subjects like the mechanics of solids,
the theory of elasticity, experimental stress analysis and machine design without a sound
background of this topic. We shall discuss the physical aspects first even though several
concepts are abstract. It is desirable to learn the index notation, sometimes called the
tensor notation. It is necessary to build the mathematical tools before we can effectively
understand the contents of this chapter. Side by side with the analytical-mathematical
treatment, some physical feel also is necessary. Geometrical visualisation is of great help
to understand the nature of stress at a point. Some concepts are somewhat abstract, and
have to be absorbed and digested slowly.
INTRODUCTION
We begin by pointing out the difference in the state of stress at a point between fluids at
rest and solids. We may recall Pascal’s1 law from our earlier study2 . This law states that
for all fluids at rest, the resultant pressure on every plane passing through a given point
has the same magnitude (compressive), and is always normal to the plane3 .

1
Blaise Pascal (June 1623 - Aug. 1662) was a French mathematician and physicist with a “multiplicity of
gifts” in other fields too. Pascal’s law of pressure is explained in his book Treatise on the Equilibrium of
Liquids (1653).
2
In fluid statics. It is valid for all fluids at rest, real (viscous) and ideal (inviscid, without any viscosity;
and not in the sense of being the most desirable kind of fluid), but only for ideal fluids in motion.
3
The numerical value is the same on all planes at the same point. In some books this is stated as “the
same in all directions”. We consider that such a statement is conceptually in error. A picture often
given alongside, like rays shooting out radially outwards all around from a centrally located circular (or
spherical) sun, reinforces the error. We believe that this distinction must be emphasised: the pressure is
the same on all planes (and not “· · · the same in all directions”). We shall see a little later that pressure
at a point does not have, or cannot be assigned, any direction without specifying the plane.
Advanced Mechanics of Solids 4-2

Pressure at a Point in a Fluid at Rest

We know that the pressure in a fluid at rest varies linearly

with the depth as shown. We also know that it remains
constant, the same, at all points on the same horizontal line.
But we are not discussing this here. Our purpose is now to
understand the nature of pressure at a point.

We shall return to the question of how the stress components

vary inside a body (typically a machine part). We will not
Figure 4.1: Variation of discuss fluid statics further except to point out later that the
pressure in a fluid at rest pressure at a point is an isotropic tensor.

(a) Stress vector (b) Stress vector (c) Stress vector

on the x-plane on the y-plane on the z-plane

Figure 4.2: In a fluid at rest, the pressure vectors on different planes passing through the
same point are all normal to the plane concerned. Furthermore, the magnitudes are all the
same. It is necessary to emphasise that the two parallel planes shown are one and the same.
They are shown as separated in order to show clearly that the pressure is compressive.

Let us discuss the pressure at a point in a fluid at rest. We know from our earlier studies
that the pressure varies linearly with the depth [Fig. 4.1], and that it has the same value
at all points on the same horizontal line. But we are not discussing now how the pressure is
distributed from point to point4 . The pressure vector on any plane is always normal to the
plane concerned. Furthermore, it has the same magnitude. This is symbolically represented
in Fig. 4.2. Let us note that the two parallel planes are one and the same.
Stress at a Point in a Solid in Equilibrium
The resultant stress would vary, of course, from point to point. Often the interest in
the solution of problems is to obtain this distribution, i.e., variation, from point to point.
However, we are not discussing such variations here. We need to know in great detail the

The pressure may, in general, vary from point to point, but we are not concerned about such variations
here in this chapter.
4
Our interest does lie eventually when discussing design, etc. in how the pressure is distributed at different
points, but now we are focussing on the nature of pressure at a point.
4-3 The State of Stress at a Point

(a) Stress vector on (b) Stress vector on (c) Stress vector on (d) Stress vector on the
a plane the x-plane the y-plane z-plane

Figure 4.3: Unlike a fluid at rest, the stress vectors on different planes passing through the
same point in a solid, in general, are not normal to the plane concerned. Their magnitudes
also are different. There can be exceptional cases where the state of stress is similar to that
of a fluid at rest. We shall call such a special state as a hydrostatic state of stress in a
figurative sense even though there is no fluid.

For solids, however, the situation is different. The resul-

tant pressure (more commonly referred to as the resultant
stress) at a point, even at the same point, is differenta on
different planes in both magnitude and direction. Now the
resultant stress at the same point depends decisively on the
plane considered. It is, in general, no more normal to the
plane in question.
Figure 4.4: Stress vector a
This is so in general. There can surely be special cases where it may
and its components be the same, but these are exceptions.

nature of stress at a point, before we can embark on the discussion of the larger problem.
Thus, for the rest of this chapter, and even beyond, we shall focus our attention on the
state of stress at the same point; we emphasise: at a point, at a point, at a point5 . This
statement holds until we make a further announcement waiving or superseding this.
We shall now proceed to define the stress at a point on a plane, and then to describe
the state of stress at a point. We shall state here, in anticipation of what we shall see, that
the stress (i) at a point is a (second order) tensor, (ii) at a point on a (given) plane is a
vector, and at a point on a plane in a (given) direction is a scalar6 . The stress at a point
is to be defined using concepts from the calculus such as the limit. This has already been

5
Repetition is said to produce wonderful end results, even though it may appear to be monotonous and
pointless! Besides, Sanskrit aestheticians have emphasised that repetition is not only permissible but is
even desirable in teaching and love making!
6
It is not quite correct or precise to state that the stress is a tensor. Strictly speaking we should state
that the state of stress is an example of a tensor, · · · . Once the concept is unmistakably clear, we may
occasionally make such imprecise statements to avoid circumlocution (long drawn out sentences).
Note the steady climbing down step by step: stress at a point; stress at a point on a plane; stress at a
point on a plane in a direction.
Advanced Mechanics of Solids 4-4

done in our earlier classes and briefly reviewed in the previous chapter [p. 1-11]. Here we
shall not repeat it. We shall assume that we are all sufficiently familiar with it.
Surface and Volume Forces
There are two kinds of forces, surface forces and volume forces. As the qualifications
indicate, these act on the (relevant) surface and volume, respectively, of a body. Examples
of the former are the water pressure acting on the bottom and side surfaces of a water tank,
and the soil pressure on a retaining wall, while those of the latter are weight (gravitational
forces), ‘inertia forces’, and forces of attraction. These latter ones act on the volume of a
body. In this book we represent surface force components by X, Y, Z or Xi per unit area,
and body force components by Fx , Fy , Fz or Fi per unit volume. The body forces may be
reckoned on a ‘per-unit-mass’ basis also. The total body force will be Fi (dV ) or Fi (ρ dV ),
(ρ dV ) being the mass of the volume element of the body. The external surface force applied
on the (part or the entire external) surface of the body is also referred to as the (applied)
surface traction. In addition, there can be couples acting on the entire volume. There may
be locked-in moments. The presence of such locked-in couples (moments) leads to couple
stresses and considerable complications in the resulting theory. As this theory is hard and
abstract, we do not discuss couple stresses7 at all in this book.
STRESS COMPONENTS AT A POINT
Let us discuss a stress vector at a point on a plane and its three components along the coor-
dinate axes. Figs 4.5a and 4.5b show the stress vectors in dotted lines and the components
in firm lines on the x- and y-planes, respectively8 .

(b) Stress components (c) Stress components

(a) Stress components at a point on the x-plane (ABCD) on the y-plane (EF CD)

Figure 4.5: Stress components at a point (the subscripts 1, 2, 3 refer, respectively, to x, y, z);
stress components on the x- and y-planes (ABCD) and (EF CB), respectively

7
The two brothers Eugéne-Maurice-Pierre Cosserat (Mar. 1866 - May 1931) and François Cosserat (Oct.
1852 - March 1914) had proposed a theory as far back as in 1909, and developed it to some extent. It was
not taken much notice of until about 1954 when R.D. Mindlin and J.P. Nowacki revived and developed
it much further. The Cosserat continuum theory is now known as the theory of micro-polar elasticity.
This helps us understand certain situations like dislocation theory, a reduction in the stress concentration
factor around holes and cracks, and the wave speed of plane dilatational waves. This is very hard. One
of the complications is that the stress tensor (stress matrix) is not symmetric σij 6= σji any more with all
its complicating consequences.
8
It is not proper to draw both (i) the resultant and (ii) the three components in firm lines.
4-5 The State of Stress at a Point

Stress Vector on a Plane and Its Three Components

We may represent the state of stress at a point as σ (Gibbs’ coordinate-free representation)
or by its components w.r.to to a coordinate system. Choosing the latter method, we set up
a right handed cartesian coordinate system (x, y, z) at a typical point P [Fig. 4.5]. If we
choose a plane, we obtain a stress vector on this plane [Fig. 4.3a]. This stress vector on
this plane is represented as shown in the figure and as explained below.
(ν)
The stress vector T is resolved first into two components: (i) normal and (ii) tangential.
The dependence of the stress vector on the plane is explicitly shown by (ν) on top of the
letter T . The tangential part (that is, the resultant shear stress at the point on the plane ν,
marked as τ(ν) in Fig. 4.3a) is further resolved along the two coordinate directions, thereby
obtaining the three stress components at the point on this plane. See Fig. 4.3a.
(x)
Similarly, the stress vector T on the x-plane — a plane is defined by its normal; the
x-plane is that plane for which the x-axis is the normal — is resolved9 into a normal and a
tangential components. The normal stress is σxx . The tangential stress (resultant or total)
shear stress is further resoled along the coordinate directions y- and z- directions as τxy
and τxz , respectively [Fig. 4.5a]. All these three stress components have two subscripts10
each, the first subscript (here x) denoting the plane, and the second the direction of the
stress component concerned (y for the shear stress component τxy and z for the shear stress
components τxz ). When the subscripts are the same, the stress component is a normal
stress; when there are different, it is a shear stress11 .
Stress Components on an Inclined Plane
(y) (z)
In exactly the same way, the stress vectors T and T on the other two planes (the y- and
the z-planes) can be represented and shown as in Figs 4.5c and 4.6a. These are all shown
in Fig. 4.6b. There are thus, 3 × 3 = 9 stress components together describing the state of
stress at the point P when referred to the chosen rectangular cartesian coordinate system
(x, y, z). These nine (9) components together give a complete description of the state of
stress at the point P . These components are displayed in the form of a matrix as shown
below. The diagonal elements are normal stresses, and the others shear stresses. Let us
note that for the elements on the leading diagonal, the subscripts are the same (normal
stresses), while for the off-diagonal ones the subscripts are different (shear stresses). [For
the elements on the leading diagonal, the row number and the column number are obviously
the same. This is why these elements with the same subscripts represent normal stresses.]

9
Strictly speaking, a stress cannot be resolved into components like a force. But if the reference is to the
same plane, then resolving the stress in effect is the same as resolving the force. When different planes
are involved, however, this statement takes on special importance, and should serve as a caution.
10
Two subscripts, because these are the components of a tensor of rank 2.
11
Mathematicians use only one letter, either σ or τ . Actually, the same letter can represent both normal and
shear stresses: normal when the two subscripts are the same, and shear otherwise. Engineers generally
prefer to use σ to denote normal stresses and τ for shear stresses. Whether a stress component is a normal
stress or a shear stress makes much difference to engineers.
Advanced Mechanics of Solids 4-6

(a) Stress components (b) Stress components (c) Stress components marked
on the z-plane (GDCF ) on all the three planes on the other three planes

Figure 4.6: Stress components on all the three planes (ABCD, EF CB and GDCF )

(x)
−→
 
σxx τxy τxz the three components of T in the three directions
  (y)
 τyx σyy τyz  −→
 
the three components of T in the three directions
 
(z)
τzx τzy σzz −→ the three components of T in the three directions

Caution: If is perhaps tempting to add all the three elements in the same column as, say,
σxx + τyx + τzx and call it the resultant stress in the x direction. But, no; a firm no. They
are acting on different planes and, it is meaningless to add stress components on different
planes even if they are acting in the same direction12 .
Let us call this as the stress matrix, that is, the matrix representation referred to the
rectangular cartesian coordinate system (x, y, z) of the invariant13 symmetric stress tensor.
There are three key words here: invariant, symmetric, (stress) tensor. They are pregnant
with meaning. We shall explain these terms slowly and carefully. See p. 4-38.
Figs 4.6b, 4.6c together show all the stress components14 on all the six faces. The three
pairs of planes are shown here. It is important to realise that the parallel planes are not
physically separated. Thus, the parallel planes ABCD, OEF G are one and the same, both
being the same x-plane. Similarly the parallel planes EF CB, OGDA (and GDCF, OABE)
are also one and the same, both being the same y-plane (z-plane). They are shown as

12
Strictly, we can add only force components and not stress components. But if the stress components
are all acting on the same plane, adding the stress components comes to the same thing as adding the
corresponding force components. But adding stress components acting on different planes is an entirely
different matter.
13
The word invariant is used with two entirely different meanings: one in the sense that it is a scalar; and
the other in the sense that the tensor itself is invariant (unchanging) even when the components change.
The same word with these two different meanings may create misunderstanding for some students. In
addition this same word is used to refer to certain combinations of stress components like I1 , I2 , I3 [p.
4-30].
14
Actually all the 9 × 2 = 18 components are to be shown in the same figure. As such a figure with too
many entries will look rather cluttered, we have shown them as two sets separately.
4-7 The State of Stress at a Point

separated in order to understand the proper directions of the various stress components15 .
[When we derive the equations of equilibrium, we should consider the changes in the stress
components as we move from one point to a neighbouring point. Thus, we are required to
consider two different, though neighbouring, points.]
Sign Conventions for Stress Components
The stress components are reckoned as positive or negative according to the sign convention
that we choose. Figs 4.6b, 4.6c show the positive directions of all the stress components.
For normal stresses it is traditional to choose tension as positive and compression, therefore,
as negative. For shear stresses, however, such a sign convention cannot be used, because
there cannot be any such classification as tensile or compressive for shear stresses. Thus, it
becomes necessary to adopt a somewhat artificial convention. This is explained below and
illustrated for the cases (a) [Fig. 4.7a] and (b) [Fig. 4.7b].
To decide whether a (given) shear stress component on a (given) plane (say, the com-
ponent τxy ) is positive or negative, first draw a tensile stress on that plane (that is, the
x-plane). Is it acting along the positive direction of the coordinate axis concerned (the
x-axis)? If it is, then the positive direction of the shear stress (τxy ) is along the positive
direction of the coordinate axis concerned (the y-axis). This is illustrated with reference to
Figs 4.7a and 4.7b.

(a) Positive normal and shear stresses (b) Negative normal and shear stresses

Figure 4.7: Sign conventions for stress components, positive (Fig. 4.7a) and negative (Fig.
4.7b) stress components are shown. For example, σyy = 5, τyx = 3 [Fig. 4.7a]; and σyy =
−5, τyx = −3 [Fig. 4.7b], all in MPa.

To carry the explanation further, let us refer to the figures [Figs 4.8a, 4.8b] shown.
Let us see how to mark τxy = τyx = −4 MPa. The stress component τxy acts on the
x-plane (ABCD, OEF G) [Fig. 4.8a]. To find the positive direction of this component
on the ABCD plane, consider a tensile stress on this face. This acts along the positive
x-direction (coming towards us from the page). Hence, according to the sign convention
for shear stresses, the positive direction of τxy on this plane ABCD is along the positive
15
We shall see a similar figure later [Fig. 5.3, p. 5-5]. There the planes are physically separated by the
distances dx, dy and dz; after writing the stress components and, thus, the equations of equilibrium, we
proceed to the limit as dx, dy, dz → 0 to obtain the differential equations of equilibrium at the point P .
Advanced Mechanics of Solids 4-8

(a) τxy = τyx = −4 MPa (b) τyz = τzy = 2 MPa (c) σzz = −3 MPa

Figure 4.8: Sign conventions: another example for stress components: positive (Fig. 4.8a)
and negative (Fig. 4.8b) stress components are shown. For example, σxx = σyy = σzz =
0, τxy = τyx = −4, τyz = τzy = 0, τzx = τxz = 0, all in MPa are shown in Fig. 4.8a;
σxx = σyy = σzz = 0, τxy = τyx = 0, τyz = τzy = 2, τzx = τxz = 0, all in MPa are shown in
Fig. 4.8b; and σxx = σyy = 0, σzz = −3, τxy = τyx = −2, τyz = τzy = 0, τzx = τxz = 0, all
in MPa are shown in Fig. 4.8c. We shall see later that Fig. 4.8a and Fig. 4.8b represent
cases of pure shear, and that the z-plane in Fig. 4.8c is a principal plane.

y-coordinate direction (to the right). Here τxy = −4 MPa. Thus, this stress component
acts to the left. Having already reversed the direction to take care of the negative sign, the
magnitude is to be marked as 4 MPa (and not as −4 MPa). On the plane OEF G which is
also the same x-plane, the direction of τxy = −4 MPa is in the opposite direction, i.e., to
the right, as shown in Fig. 4.8a.
We know this. However, we can also arrive at it arguing out as before. On the plane
OEF G, a tensile stress acts in the negative x-direction (going away from us). Hence the
positive direction of the shear stress component τxy on this face OEF G is along the negative
y-direction (to the left). Here τxy = −4 MPa; hence it is marked as 4 MPa (and not as −4
MPa) (to the right).
LAW OF TRANSFORMATION OF THE STRESS COMPONENTS
The nine (9) stress components, arranged in the form of a matrix following the stated
pattern, constitute the stress matrix with reference to the (x, y, z) coordinate system. These
are independent16 ; they together give a complete description of the state of stress at the
point P . That is, the stress components at the same point, when referred to other coordinate
systems, are all indirectly contained in this set of nine components.
Thus, we are naturally led to examine this question: given the nine (9) stress components
at a point w.r. to a coordinate system (x, y, z), how shall we obtain the stress components at
the same point w.r. to a different coordinate system (x0 , y 0 , z 0 )? The ‘new’ stress components
in the ‘new’ system are obviously related to the ‘old’ ones in the ‘old’ system in a way that

16
In all the cases that we deal with, the stress matrix turns out to be symmetric, in which case only six
(6) of them are independent. The conclusion that it is symmetric follows from the moment equations of
equilibrium. We shall call attention to this fact when we discuss the equations of equilibrium.
4-9 The State of Stress at a Point

depends decisively on how the ‘new’ axes are oriented relative to the ‘old’ ones. This
exercise leads to the important law of transformation of the nine stress components. We
shall examine this question, but before that we need some simple results which are of the
nature of prerequisites17 .
Complete Description of the State of Stress
We stated above that these nine (9) stress components describe the state of stress at a point
completely. If this is so, it should be possible for us to extract from these nine components
all the information that we seek, and answer questions such as the following.

(i) How can we obtain the stress components on an inclined plane?

(ii) How can we find the stress components referred to a different set of (right handed
cartesian) coordinate system, say, (x0 , y 0 , z 0 )?

(iii) How can we find the maximum normal stress and locate its plane?

(iv) Are there planes on which there is no shear stress at all (even in the general case)?

(v) How can we find the maximum shear stress and identify the corresponding plane?

(vi) Is there a coordinate system referred to which there are only normal stresses? Can
there be several such coordinate systems? If so, how many in general?

(vii) · · ·

We need to answer all these questions. We shall take them one by one. First let us obtain
the stress components on an inclined plane. If this set of nine (9) stress components gives a
complete description of the state of stress, then all the information pertaining to the state
of stress at P is contained, directly or indirectly, in this set. In particular, it should be
possible to find the normal and shear stress components on any inclined plane with the
direction cosines (l, m, n) w.r.to the coordinate axes passing through the same point P .
Stress Components on an Inclined Plane
(ν)
The general approach is this. Let us choose a plane ABC [Fig. 4.9a]. Let T be the stress
vector on this inclined plane that has a normal ν with its direction cosines (l, m, n). We
desire to find this stress vector (and its components) and express it (them) in terms of the
known (given) nine (9) stress components σij , (i, j = 1, 2, 3). This is the problem. How
can we obtain these sought after relationships?
We consider the tetrahedron OABC and mark all the stress components acting on
all its faces. The stress components acting on the flat surfaces are known (given). We
17
Similar observations are equally valid in other coordinate systems also. For example, in a (2-dimensional)
polar coordinate system, the stress components are σrr , σθθ , τrθ = τθr . As we are concerned only about
cartesian tensors, we consider only rectangular cartesian coordinate systems.
Advanced Mechanics of Solids 4-10

(ν) (ν)
(a) T on ABC (normal ν) (b) T on ABC resolved along the axes

Figure 4.9: To calculate the stress components on an inclined plane ABC

desire to know the normal and shear stresses on the inclined surface ABC. The body is in
equilibrium, and hence there must be force balance in each of three coordinate directions
x, y and z. The stress components multiplied by the respective areas give us the forces; it is
the equilibrium of this system of forces that we consider. The unknown stress components
on the inclined plane are thus written in terms of the known nine (9) stress components
σij (i, j = 1, 2, 3). After obtaining these relationships, let us move the inclined plane ABC
parallel to itself making the tetrahedron smaller and smaller, until in the limit the inclined
plane passes through the point P . This is the general plan. Now we shall carry out the
various steps in accordance with this plan.

(x) (y) (z)

(a) T on OBC (normal x) (b) T on OCA (normal y) (c) T on OAB (normal z)

Figure 4.10: To calculate the stress components on an inclined plane ABC

(ν) (ν)
T (with its components Ti , (i = x, y, z) along the coordinate directions) is the resultant
stress vector on the plane ABC shown [Fig. 4.9a] shaded for clarity. This is, in general,
not along the normal. The components along the coordinate directions, are shown in Fig.
4.9b. The normal to this inclined plane has the direction cosines (l, m, n). Let us resolve
each of the three stress vectors in terms of the base vectors i , j , k .

(x)
stress vector on OBC T = −σxx i − τxy j − τxz k ; (4.1a)
(y)
stress vector on OAC T = −σyx i − σyy j − τyz k ; and (4.1b)
(x)
stress vector on OAB T = −σxx i − τxy j − τxz k . (4.1c)
4-11 The State of Stress at a Point

The tetrahedron OABC is in equilibrium. Hence, there must be force balance in every
direction. Let us consider the equation of equilibrium in, say, the y-direction. Let ∆ be the
area of the inclined face ABC. Then, the area OBC is its projection on the x-plane. Its
magnitude is, therefore, l ∆. Similarly, the areas OAC and OAB are, respectively, m ∆ and
n ∆. The force components that appear in the equation of equilibrium are the following.
(ν) (ν)
Face ABC: stress component Ty ; area ∆; force Ty ∆
Face OAC: stress component σyy ; area m ∆; force σyy m ∆
Face OBC: stress component τxy ; area l ∆; force τxy l ∆
Face OAB: stress component τzy ; area n ∆; force τzy n ∆
Volume OABC: body force Fy ; volume small; force Fy × small

(ν)
Force balance in the y-direction: Ty ∆ + body force = τxy l ∆ + σyy m ∆ + τzy n ∆.
Now let us make the tetrahedron smaller and smaller, and watch how this equation of equi-
librium behaves as ∆ → 0. During this process the plane ABC must retain its orientation.
Thus, it stays parallel to the original ABC with the same direction cosines (l, m, n). Let
us note that, as ∆ becomes small, the surface forces become smaller and smaller, while the
body force become smaller and smaller and smaller. Thus, the body force term → 0 faster
and, consequently does not come into our reckoning. The inclined plane in the limit passes
through the point P , and the equation of equilibrium in the y-direction appears as
(ν)
Ty ≡ pνy = τxy l + σyy m + τzy n.

Along with its companion equations in the x- and z- directions, the three equations of
equilibrium are:
(ν)
Tx ≡ pνx = σxx l + σyx m + τzx n (4.2a)
(ν)
Ty ≡ pνy = τxy l + σyy m + τzy n (4.2b)
(ν)
Tz ≡ pνz = τxz l + τyz m + σzz n (4.2c)

(a) Stress vectors on all the four faces (b) All the stress components in the y-direction

Figure 4.11: To write down the equation of equilibrium (in the y-direction)
Advanced Mechanics of Solids 4-12

This set of equations appears in index notation as

(ν)
Ti = σji nj or Ti = σji nj (i, j = 1, 2, 3). (4.3)

This is known as Cauchy’s formula or Cauchy’s18 result. We shall see this again. In view
of the importance of this result, let us honour this by enclosing it in a box.

(ν)
Cauchy’s result: Ti = σji nj (i, j = 1, 2, 3)

Now how do we calculate the normal and shear stress on this inclined plane? That is not
difficult. If we resolve pνx , pνy , pνz back again on the normal ν to the plane, and add them
up, we would obtain the normal stress. Thus,

(ν)
σνν = T . ν = pνx l + pνy m + pνz n
= [lσxx + mτyx + nτzx ] l + [lτxy + mσyy + nτzy ] m + [lτxz + mτyz + nσzz ] n
= l2 σxx + m2 σyy + n2 σzz + 2lmτxy + 2mnτyz + 2nlτzx , (4.4)

where we have assumed the symmetry of the stress matrix (τij = τji ).
Having obtained the normal stress, we can calculate the (total, resultant) shear stress
on the inclined plane as follows.
s
(ν)2  2 q
τ(ν) =  T  − σνν  = [p2νx + p2νy + p2νz ] − [σνν ]2
   

2
τ(ν) = [lσxx + mτyx + nτzx ]2 + [lτxy + mσyy + nτzy ]2 + [lτxz + mτyz + nσzz ]2
2
− l2 σxx + m2 σyy + n2 σzz + 2lmτxy + 2mnτyz + 2nlτzx

(4.5)

We shall work out a small numerical example to illustrate this.

A Numerical Example:
The state of stress at a point is given by the matrix given on p. 13-2, Sec. 2 w.r.to the
axes (x, y, z). We desire to calculate the normal and shear stresses on a plane defined by
the direction cosines (0.42, 0.50, 0.76).
Note first of all that the direction cosines (l, m, n) satisfy the relation

l2 + m2 + n2 = 0.422 + 0.502 + 0.762 = 1.

18
Augustine-Louis Cauchy (Aug. 1789 - May 1857) was an outstanding French mathematician. Students of
engineering would be interested to know that Cauchy graduated in Civil Engineering and that he was a
civil engineer in the earlier part of his career. His fame is, however, as a great mathematician. He had
established several important results in the theory of elasticity.
4-13 The State of Stress at a Point

Referring to the stress matrix given on p. 13-2, Sec. 2 w.r.to the axes (x, y, z), we recognise
that the entries in the same (horizontal) row are the stress components acting in the same
direction. Thus,
(x)
the stress vector on the x-plane, T = 10 i + 12 j + 8 k ;
(y)
the stress vector on the y-plane, T = 12 i + 15 j − 5 k ; and
(z)
the stress vector on the z-plane, T = 8 i − 5 j − 7 k .
(ν)
The components of the stress vector T are
(ν)
Tx ≡ pνx = lσxx + mτyx + nτzx
= 0.42 × 10 + 0.50 × 12 + 0.76 × 8 = 16.28 MPa;
(ν)
Ty ≡ pνy = lτxy + mσyy + nτzy
= 0.42 × 12 + 0.50 × 15 − 0.76 × 5 = 8.74 MPa; and
(ν)
Ty ≡ pνz = lτxz + mτyz + nσzz
= 0.42 × 8 + 0.50 × (−5) + 0.76 × (−7) = −4.46 MPa.
The magnitude of the stress vector on the inclined plane is, therefore,
r
(ν) q (ν) (ν) (ν)
 T  = p2νx + p2νy + p2νz ≡ [Tx ]2 + [Ty ]2 + [Ty ]2
 
p
= 16.282 + 8.742 + 4.462 = 19.01 MPa.
The normal stress on the inclined (ν) plane = lpνx + mpνy + npνz
= 0.42 × 16.28 + 0.50 × 8.74 − 0.76 × 4.46 = 7.82 MPa,
√
and the shear stress on the same inclined (ν) plane = 19.012 − 7.822 = 17.33 MPa.
Having learned how to calculate the stresses on an inclined plane, we are now ready to
tackle the problem of stress transformation.
Transformation of Stress Components
Let us now revert to the question that we raised earlier: given the nine (9) stress components
at a point P with reference to a rectangular orthogonal coordinate system, how can we
obtain the new stress components when the axes are changed from the ‘old’ ones (x, y, z),
that is, (x1 , x2 , x3 ), to the ‘new’ ones (x01 , x02 , x03 )? As (x1 , x2 , x3 ) −→ (x01 , x02 , x03 ),
 
σx0 0 x0 τx0 0 y0 τx0 0 z 0
   
σxx τxy τxz ? ? ?
     
 0 0 0

τ σ τ −→ τ σ τ = ? ? ?
   
 yx yy yz   y 0 x0

y0 y0 y0 z0 
  
     
τzx τzy σzz (x ,x ,x ) τz0 0 x0 τz0 0 y0 σz0 0 z 0 0 0 0 ? ? ? (x0 ,x0 ,x0 )
1 2 3 (x1 ,x2 ,x3 ) 1 2 3
Advanced Mechanics of Solids 4-14

Old coordinates
x1 x2 x3
New coordinates x01 a11 a12 a13
x02 a21 a22 a23
x03 a31 a32 a33

Table 4.1: Table of direction cosines

First we note that the nine (9) ‘new’ stress components σi00 j 0 depend on the ‘old’ ones in a
way that depends decisively on how the ‘new’ coordinates are disposed (oriented) relative
to the ‘old’ ones. Thus, we must know or specify the various direction cosines. Table 4.1
and the figure alongside specify the orientations of the ‘new’ axes relative to the ‘old’ ones.
Shown here is a table of direction cosines that defines or specifies the orientation of
the ‘new’ coordinate system relative to the ‘old’ one. Thus, for example, a12 ≡ a10 2 ≡
cos(x01 , x2 ). Note further that a12 ≡ a10 2 6= a120 ≡ a21 ; that is, this matrix of direction
cosines is not symmetric in general. We know that the direction cosines satisfies the two
sets of orthogonal conditions. The procedure is explained below.

(a) Two coordinate systems, ‘old’ and ‘new’ (b) Choose an inclined plane ABC

Figure 4.12: ‘Old’ (x1 , x2 , x3 ) and ‘new’ (x01 , x02 , x03 ) coordinate systems: an inclined plane
ABC is chosen with its normal along the x01 axis.

(i) First calculate the stress components on the x01 plane, viz., (σx0 0 x0 , τx0 0 x0 , τx0 0 x0 )19 .
1 1 1 2 1 3
For this, construct a triangle ABC with its normal along the (x01 ) axis [Fig. 4.12].
(ν) (ν) (ν)
(ii) Next calculate the stress components pνx ≡ T1 , pνy ≡ T2 , pνz ≡ T3 . These are given
by Eqs (4.2a, 4.2b, 4.2c) (Cauchy’s result) as
(ν)
Tx ≡ pνx = σxx l + σyx m + τzx n
(ν)
Ty ≡ pνy = τxy l + σyy m + τzy n
0
19
Hereafter we will drop the prime marks on the subscripts and write only as σ11 for σx0 0 x0 . The primes are
dropped only for the number subscripts, but not for the letter subscripts.
4-15 The State of Stress at a Point

(ν)
Tz ≡ pνz = τxz l + τyz m + σzz n

(iii) We now note that x01 is normal to the plane ABC and, therefore, x02 and x03 being
perpendicular to x01 lie along the plane ABC. Thus, the stress components on the
plane ABC along these directions are shear stresses.
(ν) (ν) (ν)
(iv) We have already calculated Tx ≡ pνx ; Ty ≡ pνy ; Tz ≡ pνz in terms of the given stress
components w.r.to the ‘old’ coordinate system (x, y, z).
To obtain the ‘new’ stress components, all we have to do is to resolve the above
components in the direction concerned and add up. Thus,

σx0 0 x0 =pνx cos(x0 , x) + pνy cos(x0 , y) + pνz cos(x0 , z)

=l [lσxx + mτyx + nτzx ] + m [lτxy + mσyy + nτzy ] + n [lτxz + mτyz + nσzz ]
= cos(x0 , x) cos(x0 , x) σxx + cos(x0 , y) τyx + cos(x0 , z) τzx
 

+ cos(x0 , y) cos(x0 , x) τxy + cos(x0 , y) σyy + cos(x0 , z) τzy

 

+ cos(x0 , z) cos(x0 , x) τxz + cos(x0 , y) τyz + cos(x0 , z) σzz

 

=a11 σxx a11 + a11 τyx a12 + a11 τzx a13

+a12 τxy a11 + a12 σyy a12 + a12 τzy a13
+a13 τxz a11 + a13 τyz a12 + a13 σzz a13 (4.7)

In the same way, we can project pνx , pνy , pνz on (i) on the y 0 axis and add them
together to obtain τx0 y0 ; and (ii) on the z 0 axis to obtain τx0 z 0 . These are not difficult
at all, but they can be confusing. The pattern, once recognised, would help us.

τx0 0 y0 =pνx cos(y 0 , x) + pνy cos(y 0 , y) + pνz cos(y 0 , z)

=a21 [a11 σxx + a12 τyx + a13 τzx ]
+a22 [a11 τxy + a12 σyy + a13 τzy ]
+a23 [a11 τxz + a12 τyz + a13 σzz ]
=a21 σxx a11 + a21 τyx a12 + a21 τzx a13
+a22 τxy a11 + a22 σyy a12 + a22 τzy a13
+a23 τxz a11 + a23 τyz a12 + a23 σzz a13 (4.8)

τx0 0 z 0 =pνx cos(z 0 , x) + pνy cos(z 0 , y) + pνz cos(z 0 , z)

=a31 [a11 σxx + a12 τyx + a13 τzx ]
+a32 [a11 τxy + a12 σyy + a13 τzy ]
+a33 [a11 τxz + a12 τyz + a13 σzz ]
=a31 σxx a11 + a31 τyx a12 + a31 τzx a13
+a32 τxy a11 + a32 σyy a12 + a32 τzy a13
+a33 τxz a11 + a33 τyz a12 + a33 σzz a13 (4.9)
Advanced Mechanics of Solids 4-16

(v) We have thus far obtained the three stress components on the x0 plane. To obtain the
stress components of the y 0 , we can proceed similarly by considering an inclined plane
with y 0 as its normal. After obtaining the three stress components on this y 0 plane,
start afresh and consider an inclined plane with z 0 as its normal. The procedure is
entirely similar. In this way all the nine ‘new’ stress components can be obtained. All
these nine ‘new’ stress components — really only six because of the symmetry of the
stress matrix — are now expressed in terms of the given ‘old’ stress components and
the set of nine direction cosines aij , (i, j = 1, 2, 3). These are the stress transformation
equations.

STRESS TRANSFORMATION EQUATIONS

The final results of the above exercise are displayed below in full.
0
σx0 0 x0 ≡ σ11 =a11 σxx a11 + a11 τyx a12 + a11 τzx a13
+a12 τxy a11 + a12 σyy a12 + a12 τzy a13
+a13 τxz a11 + a13 τyz a12 + a13 σzz a13 (4.10a)
τx0 0 y0 ≡ 0
σ12 =a21 σxx a11 + a21 τyx a12 + a21 τzx a13
+a22 τxy a11 + a22 σyy a12 + a22 τzy a13
+a23 τxz a11 + a23 τyz a12 + a23 σzz a13 (4.10b)
τx0 0 z 0 ≡ 0
σ13 =a31 σxx a11 + a31 τyx a12 + a31 τzx a13
+a32 τxy a11 + a32 σyy a12 + a32 τzy a13
+a33 τxz a11 + a33 τyz a12 + a33 σzz a13 (4.10c)
0
τy0 0 x0 ≡ σ21 =a11 σxx a21 + a11 τyx a22 + a11 τzx a23
+a12 τxy a21 + a12 σyy a22 + a12 τzy a23
+a13 τxz a21 + a13 τyz a22 + a13 σzz a23 (4.10d)
σy0 0 y0 ≡ 0
σ22 =a21 σxx a21 + a21 τyx a22 + a21 τzx a23
+a22 τxy a21 + a22 σyy a22 + a22 τzy a23
+a23 τxz a21 + a23 τyz a22 + a23 σzz a23 (4.10e)
τy0 0 z 0 ≡ 0
σ23 =a31 σxx a21 + a31 τyx a22 + a31 τzx a23
+a32 τxy a21 + a32 σyy a22 + a32 τzy a23
+a33 τxz a21 + a33 τyz a22 + a33 σzz a23 (4.10f)

τz0 0 x0 ≡ σ31
0
=a11 σxx a31 + a11 τyx a32 + a11 τzx a33
+a12 τxy a31 + a12 σyy a32 + a12 τzy a33
+a13 τxz a31 + a13 τyz a32 + a13 σzz a33 (4.10g)
τz0 0 y0 ≡ 0
σ32 =a21 σxx a31 + a21 τyx a32 + a21 τzx a33
+a22 τxy a31 + a22 σyy a32 + a22 τzy a33
+a23 τxz a31 + a23 τyz a32 + a23 σzz a33 (4.10h)
4-17 The State of Stress at a Point

σz0 0 z 0 ≡ σ33
0
=a31 σxx a31 + a31 τyx a32 + a31 τzx a33
+a32 τxy a31 + a32 σyy a32 + a32 τzy a33
+a33 τxz a31 + a33 τyz a32 + a33 σzz a33 (4.10i)

Let us also note that τx0 0 y0 = τy0 0 x0 τy0 0 z 0 = τz0 0 y0 τx0 0 z 0 = τz0 0 x0 .
0 0
(that is, σij = σji symmetry of the stress matrix).

These formidable equations [Eqs 4.10a - 4.10i] — nine (9) long equations each of which has
nine (9) terms on the right hand side! — can be encapsulated in one small equation using
the index notation as

σi00 j 0 = ai0 k aj 0 l σkl more commonly written as 0

σij = aik ajl σkl . (4.11)

If we take this equation (4.11) and carry out the implied summations, we can obtain the
full length formulae shown above [Eqs (4.10a - 4.10i]. We shall demonstrate this for one
special case, say, for i = 1, j = 2.
3
X
0
i = 1, j = 2 σ12 ≡ τxy = [a11 a2l σ1l + a12 a2l σ2l + a13 a2l σ3l ]
k=1
= a11 [a21 σ11 + a22 σ12 + a23 σ13 ]
+ a12 [a21 σ21 + a22 σ22 + a23 σ23 ]
+ a13 [a21 σ31 + a22 σ32 + a23 σ33 ]
= a11 σxx a21 + a11 τxy a22 + a11 τxz a23
+ a12 τyx a21 + a12 σyy a22 + a12 τyz a23
+ a13 τzx a21 + a13 τzy a22 + a13 σzz a23

which agrees with Eq. (4.10b and 4.10d).

We may also write Eq. (4.11), which is the same as the set of equations (4.10a - 4.10i,
in the matrix form as
[σ 0 ] = [a][σ][a]T (4.12)
where [a] is the matrix of the direction cosines [Table: 4.1, p. 4-14]. More explicitly, the
transformation law is given by the matrix equation
 
σx0 0 x0 τx0 0 y0 τx0 0 z 0
 
  σxx τxy τxz  

 0
 a11 a 12 a13   a11 a21 a 31
 τ 0 0 σ 0 0 0 τ 0 0 0  = a21 a22 a23  

 τyx σyy τyz  a12 a22 a32  . (4.13)

 yx yy yz 
a31 a32 a33 a13 a23 a33
   
τz0 0 x0 τz0 0 y0 σz0 0 z 0 τzx τzy σzz

PRINCIPAL PLANES AND PRINCIPAL STRESSES

Principal planes and principal (normal) stresses are of fundamental importance in the sci-
ence of stress analysis. What are principal planes? Well, they are those planes on which
Advanced Mechanics of Solids 4-18

there is no shear stress20 . That is, the stress vector on a principal plane is entirely normal
to it. There are a number of associated questions: (i) are there such planes at every point
for the general state of stress; (ii) how many such planes can we have in general; etc. We
shall answer all such questions in due course. But first let us formulate the problem.
Formulation of the Problem
(ν)
If ABC is a principal plane, then the stress vector T on
this plane is perfectly normal to the plane. Let us call
this magnitude by σ. This means that there are no shear
stresses on this plane. It also means that pνx , pνy , pνz
are the components of σ. Thus,

pνx = l σ; pνy = m σ; pνz = n σ;

Figure 4.13: ABC: where σ is the magnitude of the principal stress (normal
a principal plane stress on the principal plane ABC) vector.

We have already obtained the expressions for pνx , pνy , pνz [Eqs (4.2a, 4.2b, 4.2c), p.
4-12]. Substituting these in the above equation, we obtain

l σxx + m τyx + n τzx = l σ;

l τxy + m σyy + n τzy = m σ;
l τxz + m τyz + n σzz = n σ.

This set of equations may be written in matrix form as

    
σxx τyx τzx  l  l 
 
 τxy σyy τzy  m = σ m ; i.e., σ x = σ x . (4.14)
 
  
n 
τxz τyz σzz  n  

What are known, and what are unknown, in this equation? The stress components are
known (or given); σ, a parameter, is unknown at this stage. What are to be determined?
The non-trivial solutions for l, m, n and the values of the parameter σ for which these
non-trivial solutions exist.
We recognise this immediately as an eigenvalue problem (often written as A x = λ x ,
where λ is the eigenvalue, and x the eigenvector). We have thus formulated the physical
20
We shall see that the normal stress reaches its stationary values on the principal planes. We shall explain
this clearly a little later. We will see that we can have two equally acceptable possibilities of defining
principal planes: (i) as those planes on which the shear stress is zero; and then we can show that on these
special planes, the normal stress reaches its stationary values; or (ii) as those planes on which the normal
stress reaches its stationary values; and then we can show that on these special planes the shear stress
is zero. Both are equally tenable. However, stationary values are a little more difficult conceptually to
understand. Hence we shall define principal planes as those on which the shear stress is zero.
4-19 The State of Stress at a Point

problem of determining the principal planes and the corresponding principal stresses as an
eigenvalue problem. Here for our physical problem, the matrix A is the (given) stress matrix
σ , the eigenvalue λ is the principal stress σ, and the eigenvector x is the direction cosines.
The eigenvector, we recall, is the non-trivial solution which exists when the parameter
σ takes on one of the special values called the eigenvalues. The physical meanings here
are: (i) the matrix A referred to above is here the (given) stress matrix σ ; (ii) σ, the
eigenvalue, stands for the principal stresses; and (iii) x, the eigenvector for the direction
cosines (direction ratios)21 .
When written as a system of linear, algebraic, simultaneous, homogeneous equations,
this appears as
    
(σxx − σ) τyx τzx l 
   0 
τxy (σyy − σ) τzy  m = 0 ; (4.15)
 

   
τxz τyz (σzz − σ)  n   0 

that is, as
σ − σII ) x = 0 .
(σ (4.16)
We shall now continue with further analysis and discussion of this problem.
Further Analysis and Discussion
We notice that l = m = n = 0 is always22 a solution of Eq. (4.14)). This is the trivial
solution. But from a physical point of view, these are the direction cosines and, therefore,
these must satisfy the condition l2 + m2 + n2 = 1. Thus, the trivial solution is not physically
acceptable. The question that we now ask is: is there a non-trivial solution and, if so, under
what conditions?
We know from our earlier study23 of linear algebra that, for a system of linear, algebraic,
simultaneous, homogeneous24 equations, a trivial solution always exists, and that for a non-
trivial solution to exist the determinant of the coefficients must vanish.
Non-trivial solution, characteristic equation, eigenvalues:
Physically, if ABC is indeed a principal plane, then there must be at least one set of
physically possible direction cosines. The trivial solution l = m = n = 0 will not do; this
(trivial solution) cannot represent a physically meaningful set of direction cosines, because
the direction cosines must satisfy the condition l2 + m2 + n2 = 1. Thus, we can conclude
that there must be a non-trivial solution. The condition for the existence of a non-trivial
solution is that the determinant of the coefficients of the homogeneous set of equations shall
21
It is clear from Eq. (4.14) that we can obtain only the direction ratios. The direction cosines may
be obtained from these direction ratios by normalisation using the result l2 + m2 + n2 = 1, that is,
x21 + x22 + x23 = 1.
22
By ‘always’ we mean ‘for all values of σ’.
23
Readers are advised to refresh their knowledge of linear algebra. Such questions arise often enough in
applied mathematics or mathematical engineering that it is wise to invest some time and energy to learn
linear algebra fairly well.
24
the ‘right hand side’ equal to zero
Advanced Mechanics of Solids 4-20

vanish. Thus, for ABC to be a principal plane,

 
σxx − σ τyx τzx 

 τxy σyy − σ τzy  = 0. (4.17)
 
 
 τxz τyz σzz − σ 
This, we note, is the characteristic equation of the stress matrix σij (σ σ ). When this is
written out in full in terms of the unknown parameter σ, we obtain
σ 3 − [σxx + σyyu + σzz ] σ 2 + σxx σyy + σyy σzz + σzz σxx − τxy
2 2 2
 
− τyz − τzx σ
2 2 2
 
− σxx σyy σzz − σxx τyz − σyy τzx − σzz τxy + 2τxy τyz τzx = 0. (4.18)
This is conveniently written as
characteristic equation: σ 3 − [I1 ]σ 2 + [I2 ]σ − [I3 ] = 0, (4.19)
where I1 , I2 , I3 are the coefficients as defined by Eq. (4.19). This, we repeat, is the
characteristic equation of the stress matrix. Its roots are the eigenvalues. Physically they
represent the principal stresses.
I1 = [σxx + σyy + σzz ] (4.20a)
2 2 2
 
I2 = σxx σyy + σyy σzz + σzz σxx − τxy − τyz − τzx (4.20b)
2 2 2
 
I3 = σxx σyy σzz − σxx τyz − σyy τzx − σzz τxy + 2τxy τyz τzx (4.20c)
The roots of the characteristic equation [Eq. (11.18) which is the same as Eq. (4.19)]
are the eigenvalues. Being a cubic equation in σ, this characteristic equation has exactly
three roots. There are all necessarily real; a theorem in linear algebra guarantees this25 .
Let us call them σ11 , σ22 , σ33 . These are the eigenvalues representing the three principal
stresses. Let us emphasise that some of them may be repeated roots. For example, we
may have (i) σ11 6= σ22 6= σ33 (all distinct); (ii) σ11 = σ22 6= σ33 (two of them are equal);
and (iii) σ11 = σ22 = σ33 (all the three of them are equal). Usually the algebraically
largest is designated by σ11 , and the algebraically smallest by σ33 . That is, if the principal
stresses are, say, −20 (that is, 20 compressive), 10 and 5 (all in MPa), we call them as
σ11 = 10, σ22 = 5, σ33 = −20 (all in MPa).
Eigenvectors:
The eigenvectors corresponding to these eigenvalues represent the planes of the principal
stresses. They can be found as explained below.
At this stage, the three eigenvalues are known. Let us first determine the eigenvector
corresponding to the first eigenvalue σ = σ11 which is now known. Substitute this known
value in the set of equations [Eq. (11.16)] to yield
    
(σxx − σ11 ) τyx τzx l 
 0 
  
τxy (σyy − σ11 ) τzy  m = 0 . (4.21)
 

   
τxz τyz (σzz − σ11 )  n   0 

25
The eigenvalues of a real symmetric matrix — and of a Hermitian symmetric matrix — are always real.
We do not encounter Hermitian symmetric matrices in our discussions here.
4-21 The State of Stress at a Point

Let us examine this set of equations. If (l1 , m1 , n1 ) is a solution, then (αl1 , αm1 , αn1 ) is
also a solution for all values of α. What does this mean? Well, this means that there are
several solutions; there is no unique solution. We will be able to obtain only the ratios,
and not the absolute values of (l1 , m1 , n1 ). This realisation — that we can obtain only the
ratios, and not the absolute values of (l1 , m1 , n1 ) — leads us to conclude that there is no
harm in giving any value to any one of (l1 , m1 , n1 ). Let us, then, put l1∗ = 1, and work
out the corresponding values of m∗1 and n∗1 . These starred quantities stand for the direction
ratios.
Eq. (4.21) is the same as

(σxx − σ11 ) l1∗ + τyx m∗1 + τzx n∗1 = 0

τxy l1∗ + (σyy − σ11 ) m∗1 + τzy n∗1 = 0
τxz l1∗ + τyz m∗1 + (σzz − σ11 ) n∗1 = 0.

On putting l1∗ = 1 — we can obtain only the ratios as argued out above — these equations
appear as

τyx m∗1 + τzx n∗1 = σ11 − σxx (4.22a)

(σyy − σ11 ) m∗1 + τzy n∗1 = −τxy (4.22b)
τyz m∗1 + (σzz − σ11 ) n∗1 = −τxz . (4.22c)

Here are now three equations in only two unknowns, viz., m∗1 and n∗1 ! But there arises
the question: are these equations consistent? The condition for the consistency of these
equations — three equations in only two unknowns — is that the rank of the coefficient
matrix is equal to the rank of the augmented matrix. The coefficient matrix is 3 × 2, which
can have at the most a rank of 2. The augmented matrix is 3 × 3, but we know that its
determinant is zero. Thus, the augmented matrix cannot have a rank 3!
Thus, any two equations from Eqs (4.22a, 4.22b, 4.22c) may be used to solve for m∗1 and
n∗1 .
We can be sure that the third equation (which was not used) will surely be satisfied.
How can we be so sure? Well, we have discussed the consistency of these equations26 .
Having thus obtained the direction ratios (l1∗ , m∗1 , n∗1 ), we can get the direction cosines
by normalising them as
l1∗ m∗1 n∗1
l1 = p ; m1 = p ; n 1 = p .
(l1∗ )2 + (m∗1 )2 + (n∗1 )2 (l1∗ )2 + (m∗1 )2 + (n∗1 )2 (l1∗ )2 + (m∗1 )2 + (n∗1 )2

We have now calculated the eigenvector (l1 m1 n1 )T corresponding to the first eigenvalue
σ11 . The physical significance is, let us repeat, that the eigenvalue represents the (first)
principal stress, and the eigenvector the direction cosines of the normal to this principal
plane.
26
A special case arises when the set of three equations is of rank 1. Now we can set, say, m∗1 = 2 and solve
for n∗1 . Recall an important result in linear algebra: if the rank of a set of n × n is r, there are exactly
(n − r) linearly independent solutions. This means that we can choose at will the values of two of the
unknowns, say, l1∗ and m∗1 and calculate the value of the third unknown, here n∗1 .
Advanced Mechanics of Solids 4-22

Other eigenvectors:
The other eigenvectors also are calculated in the same way. Consider again the equation
(11.16) in which we substitute σ22 for σ, and solve for the unknown direction ratios. Let
us put l2∗ = 1 and work out the direction ratios (l2∗ , m∗2 , n∗2 ), and calculate the direction
cosines (l2 , m2 , n2 ) by normalising them. In exactly the same way, to calculate the third
eigenvector, let us go back to Eq. (11.16) and substitute σ33 , and then calculate the corre-
sponding direction ratios. As before, let us put l3∗ = 1, calculate the direction ratios, and
finally obtain the direction cosines by normalising them.
l2∗ m∗2 n∗2
l2 = p ; m2 = p ; n 2 = p ;
(l2∗ )2 + (m∗2 )2 + (n∗2 )2 (l2∗ )2 + (m∗2 )2 + (n∗2 )2 (l2∗ )2 + (m∗2 )2 + (n∗2 )2
l3∗ m∗3 l3∗
l3 = p ∗ ; m3 = p ; n 3 = p .
(l3 )2 + (m∗3 )2 + (n∗3 )2 (l3∗ )2 + (m∗3 )2 + (n∗3 )2 (l3∗ )2 + (m∗3 )2 + (n∗3 )2

Orthogonality of Eigenvectors
These eigenvectors have the important property of orthogonality. If the eigenvectors are
distinct, the corresponding eigenvectors are orthogonal.

l1 l2 +m1 m2 +n1 n2 = 0; l2 l3 +m2 m3 +n2 n3 = 0; l3 l1 +m3 m1 +n3 n1 = 0. (4.23)

Physically this means that, if the principal stresses are distinct, the corresponding principal
planes are mutually orthogonal (perpendicular). This result is an example of an important
result in linear algebra: the eigenvectors corresponding to distinct eigenvalues are mutually
orthogonal.
If perchance the principal stresses are not distinct, the corresponding principal planes
may not be orthogonal. However, it is still possible always to find a set of three mutually or-
thogonal (perpendicular) principal planes. This is achieved by the so-called Gram-Schmidt
procedure.
How many principal planes are there in general? Three in most cases when the principal
stresses are all distinct. When two or three principal planes are equal, there are infinite
number of principal planes. In fact, when all the principal stresses are equal, that is a case
of hydrostatic stress represented by an isotropic tensor. Now every plane is a principal
plane. It is obvious that all of them cannot be mutually orthogonal. We can, as we stated
above, find a set of three mutually orthogonal (perpendicular) principal planes by the Gram-
Schmidt procedure27 . The Lamé’s ellipsoid of stresses helps us to understand these in a
way that we will never, never forget these.
STATE OF STRESS IN A PRINCIPAL COORDINATE SYSTEM
All expressions involving stress components are simplified in a principal coordinate system.
This is obvious; the shear stresses are all zero on the principal planes.
27
It is essential to understand all this clearly. However, students generally do not have a sound background
of linear algebra making it difficult to absorb all this readily. Perhaps we should be realistic; these ideas
must be digested slowly. We shall call attention to these facts again and again. These ideas would become
clearer when we discuss the geometrical representation of stress at a point and Lamé’s stress ellipsoid.
4-23 The State of Stress at a Point

Stress Matrix Referred to the Principal Coordinate System

How does the stress matrix appear in the principal coordinate system (1, 2, 3)?

I1 = [σxx + σyy + σzz ] (4.24a)

2 2 2
 
I2 = σxx σyy + σyy σzz + σzz σxx − τxy − τyz − τzx (4.24b)
2 2 2
 
I3 = σxx σyy σzz − σxx τyz − σyy τzx − σzz τxy + 2τxy τyz τzx (4.24c)

As (x1 , x2 , x3 ) −→ (1, 2, 3),

   
σxx τxy τxz σ11 0 0
   
 τyx σyy τyz  −→  0 σ22 0 
   
   
τzx τzy σzz (x 0 0 σ33 (1,2,3)
1 ,x2 ,x3 )

When referred to the principal coordinate system (1, 2, 3), there are entries only on the
diagonal; there are no shear stresses. (On the principal planes there cannot be any shear
stresses.) Thus, the stress matrix is now diagonalised into its diagonal canonical form!
Given a stress matrix, the problem of finding the principal stresses is exactly the same as
that of diagonalising the matrix.
The characteristic equation corresponding to the above diagonal matrix is

(σ − σ11 )(σ − σ22 )(σ − σ33 ) = 0 (4.25)

so that the roots, the eigenvalues which we have seen are the principal stresses, are (σ11 , σ22 , σ33 ).
The coefficients I1 , I2 , I3 are [compare with Eqs (4.24a), (4.24b), (4.24c)]:

I1 = [σ11 + σ22 + σ33 ] ; (4.26a)

I2 = [σ11 σ22 + σ22 σ33 + σ33 σ11 ] ; (4.26b)
I3 = [σ11 σ22 σ33 ] . (4.26c)

These coefficients (I1 , I2 , I3 ) are called the first, the second and the third invariants, re-
spectively, of the stress matrix. [See later p. 4-30, Sec. 4.6.6 for more explanation.]
Stress Components on an Inclined Plane
The calculation of the stress components becomes simpler when we work with the principal
planes. The reason is obvious; there are no shear stresses on the principal planes.
We shall now work out the expressions for the stress components on an inclined plane.
The general approach is the same as in 4.3.1, p. 4-5. Fig. 4.14 shows a tetrahedron. We shall
mark all the forces acting on the various faces, write down the equation(s) of equilibrium,
proceed to the limit as the tetrahedron becomes smaller and smaller, and thereby obtain
the required expressions. From the figure it is clear that
(1) (2) (3)
T = −σ11 i ; T = −σ22 j ; T = −σ33 k . (4.27)
Advanced Mechanics of Solids 4-24

Shown in the figure 4.14 is a small tetrahedron

OABC. The coordinate axes are along the principal
axes 1, 2, 3. The stress vectors on all the planes are
shown. The normal to the inclined plane is ν with
its direction cosines l, m, n referred to the principal
axes 1, 2, 3. We can see that
(1) (2) (3)
T = −σ11 i ; T = −σ22 j ; T = −σ33 k . (4.28)

Obviously there are no shear stresses on the flat

Figure 4.14: Octahedral plane
faces, because they are principal planes.

The body, the tetrahedron, is in equilibrium. Hence

(1) (2) (3) (ν)
∆1 T + ∆ 2 T + ∆ 3 T + ∆ T = 0 . (4.29)

We have omitted the body force28 . The areas ∆1 , ∆2 , ∆3 are the projections of the area ∆
on the 1, 2, 3 planes, respectively. Accordingly, ∆1 = l ∆; ∆2 = m ∆; ∆3 = n ∆. When
these and the results of Eq. (4.28) are substituted in Eq. (4.29), we obtain
(ν) (1) (2) (3)
T = −l T − −m T − n T
= l2 σ11 i + m2 σ22 j + n2 σ11 k .

From this we can find the expressions for the normal and shearing stresses as
(ν)
σνν = T . ν = l2 σ11 + m2 σ22 + n2 σ33 ; and (4.30)
" #1
2
(ν) h
2 i 21
= l2 σ11 + m2 σ22 + n2 σ33 − l2 σ11 + m2 σ22 + n2 σ33


τ(ν) =  T  − σνν (4.31)

These ideas may appear abstract and difficult. Perhaps it would help if a few numerical
problems are worked out as illustrative examples.
Numerical Example 1:
Let the stress matrix referred to the principal coordinates (1, 2, 3) be
 
40 0 0
 0 20 0  (all in MPa). (4.32)
0 0 −10
28
The reason is that the body force term goes to zero faster as the tetrahedron is made smaller and smaller.
We had seen this earlier.
4-25 The State of Stress at a Point

We shall work out two different problems. First, let us refer this matrix to a non-principal
set of axes. After obtaining this new stress matrix, let us pretend that we do not know
the principal stresses and the corresponding principal planes. We shall, we must, obtain
the principal stresses (the elements on the diagonal of the original diagonal matrix), viz.,
σ11 = 40; σ22 = 20; σ33 = −10, all in MPa.
Part I:
Let the matrix of the direction cosines be:

Old coordinates
x1 ≡ 1 x2 ≡ 2 x3 ≡ 3
New coordinates

x01 0.2666 −0.3928 0.8801

x02 −0.5653 0.6759 0.4729
x03 0.7806 0.6236 0.04187

Table 4.2: Table of direction cosines

Stress transformation - matrix referred to a non-principal set of axes:

Let us transform this matrix [Eq. (4.32)] to a non-principal set of axes, viz., (x01 , x02 , x03 )
whose orientation is defined by Table 4.2.
(a) Stress components on the x01 plane:
Following the procedure and terminology used earlier, let us consider an inclined plane (a
triangle ABC, normal along the x01 -axis), and calculate the stress components px01 .
px01 1 = σ11 × cos(x01 , 1) = 40 × 0.2666 = 10.664 MPa;
px01 2 = σ22 × cos(x01 , 2) = 20 × (−0.3928) = −7.856 MPa;
px01 3 = σ33 × cos(x01 , 3) = (−10) × 0.8801 = −8.801 MPa.
The three stress components on this x01 plane are calculated as shown below.
σx01 x01 = px01 1 × cos(x01 , 1) + px01 2 × cos(x01 , 2) + px01 3 × cos(x01 , 3)
= 10.666 × 0.2666 + (−7.856) × (−0.3928) + (−8.801) × 0.8801 = −1.817 MPa;
τx01 x02 = px01 1 × cos(x02 , 1) + px01 2 × cos(x02 , 2) + px01 3 × cos(x02 , 3)
= 10.664 × (−0.5653) + (−7.856) × 0.6759 + (−8.801) × 0.4729 = −15.500 MPa;
τx01 x03 = px01 1 × cos(x03 , 1) + px01 2 × cos(x03 , 2) + px01 3 × cos(x03 , 3)
= 10.664 × 0.7806 + (−7.856) × 0.6326 + (−8.801) × 0.4729 = 3.055 MPa.
Let us now proceed to the next plane x02 .
(b) Stress components on the x02 plane:
Let us now choose a different inclined plane and a different triangle such that x02 is its
normal. We proceed exactly as before.
px02 1 = σ11 × cos(x02 , 1) = 40 × (−0.5653) = −22.612 MPa;
Advanced Mechanics of Solids 4-26

px02 2 = σ22 × cos(x02 , 2) = 20 × 0.6759 = 13.518 MPa;

px02 3 = σ33 × cos(x02 , 3) = (−10) × 0.4729 = −4.729 MPa.

Resolving these along the relevant directions and adding up the components we obtain the
required stress components on this plane.

σx02 x02 = px02 1 × cos(x02 , 1) + px02 2 × cos(x02 , 2) + px02 3 × cos(x02 , 3)

= −22.612 × (−0.5653) + 13.515 × 0.6759 + (−4.729) × 0.4729 = 19.683 MPa;
τx02 x01 = px02 1 × cos(x01 , 1) + px02 2 × cos(x01 , 2) + px02 3 × cos(x01 , 3)
= −22.612 × 0.2666 + 13.518 × (−0.3928) + (−4.729) × 0.8801 = −15.500 MPa;
τx02 x03 = px02 1 × cos(x03 , 1) + px02 2 × cos(x03 , 2) + px02 3 × cos(x03 , 3)
= −22.612 × 0.7806 + 13.518 × 0.6236 + (−4.729) × 0.04187 = −9.419 MPa.

Now we calculate the stress components on the x03 plane.

(c) Stress components on the x03 plane:

px03 1 = σ11 × cos(x03 , 1) = 40 × 0.7806 = 31.224 MPa;

px03 2 = σ22 × cos(x03 , 2) = 20 × 0.6236 = 12.472 MPa;
px03 3 = σ33 × cos(x03 , 3) = (−10) × 0.04187 = −0.4187 MPa.

Again we resolve these along the relevant directions and add up the components to obtain
the required stress components on this plane.

σx03 x03 = px03 1 × cos(x03 , 1) + px03 2 × cos(x03 , 2) + px03 3 × cos(x03 , 3)

= 31.224 × 0.7806 + 12.472 × 0.6236 + (−0.4187) × 0.04187 = 32.133 MPa;
τx03 x01 = px03 1 × cos(x01 , 1) + px03 2 × cos(x01 , 2) + px03 3 × cos(x01 , 3)
= 31.224 × 0.2666 + 12.472 × (−0.3928) + (−0.4187) × 0.8801 = 3.055 MPa;
τx03 x02 = px03 1 × cos(x02 , 1) + px03 2 × cos(x02 , 2) + px03 3 × cos(x02 , 3)
= 31.224 × (−0.5653) + 12.472 × 0.6759 + (−0.4187) × 0.4729 = −9.419 MPa.

Now the task is completed. However, it is a good habit to check the correctness to see
if (i) τx01 x02 = τx02 x01 ; τx02 x03 = τx03 x02 ; τx03 x01 = τx01 x03 and (ii) the numerical values of the three
invariants. Here we have

τx01 x02 = τx02 x01 = −15.500; τx02 x03 = τx03 x02 = −9.419; τx03 x01 = τx01 x03 = 3.055; all in MPa.

The first invariant is I1 = −1.817 + 19.683 + 32.133 = 40 + 20 − 10 = 50 MPa. The second

and third invariants are not quite as easy to compute, and so they are not checked here.
The stress matrix in the new coordinate system appears as
 
 0  −1.817 −15.500 3.055
σi0 j 0 = −15.500 19.683 −9.419 , all values in MPa. (4.33)
3.055 −9.419 32.133
4-27 The State of Stress at a Point

We have completed the first part of the example. Now for the second part.
Part II:
Our starting point is the given stress matrix [Eq. (4.33)], and we shall ask for the eigen-
values and eigenvectors of this stress matrix. We know, of course, that the eigenvalues are
40, 20, −10, all in MPa. We shall pretend that we do not know this, and we shall obtain the
principal stresses and the principal planes as if nothing else is known about this problem.
Numerical Example 2:
Let us take the stress matrix [13-2, Sec. 2] and compute the principal stresses and the
corresponding principal planes. All the numerical values are in MPa.
 
10 12 8
12 15 −5
8 −5 −7

Characteristic equation and eigenvalues:

The characteristic equation of this stress matrix is [Eq. (4.34)]
 
10 − σ 12 8 

 12 15 − σ −5  = 0. (4.34)
 
 
 8 −5 −7 − σ 

I1 = [σxx + σyy + σzz ] = 10 + 15 − 7 = 18 MPa; (4.35a)

2 2 2
 
I2 = σxx σyy + σyy σzz + σzz σxx − τxy − τyz − τzx
= 10 × 15 + 15 × (−7) + (−7) × 10 − 122 − (−5)2 − 82 = −258 (MPa)2 ; (4.35b)
2 2 2
 
I3 = σxx σyy σzz − σxx τyz − σyy τzx − σzz τxy + 2τxy τyz τzx
= 10 × 15 × (−7) − 10 × (−5)2 − 15 × 82 − (−7) × 122 + 2 × 12 × (−5) × 8 = −2212
= −2212 (MPa)3 . (4.35c)

Thus, the characteristic equation is

σ 3 − I1 σ 2 + I2 σ − I3 = 0 −→ σ 3 − 18σ 2 − 258σ + 2212 = 0. (4.36)

The (three) roots of this characteristic equation are the eigenvalues, which are the principal
stresses. The three roots29 are 24.806, 6.635, −13.441 (all in MPa). We can thus, or in
29
Nowadays there are easy methods of obtaining the roots. In the earlier days of hand computation, solving
a cubic equation like f (x) = x3 + ax2 + bx + c = 0 was not as easy. Here this problem is relatively simple,
because we have an assurance beforehand that all the roots are real. How do we know this? A result in
linear algebra guarantees this: the eigenvalues of real symmetric matrices are always real. One method is
to guess the value of the largest root. Another method is to plot the value of the function f (x) for various
of x and see where the curve cuts the x-axis. That gives us an approximate value of one root. Here we
can invoke the physical meaning of the largest root: it is the largest principal stress. It is the largest value
Advanced Mechanics of Solids 4-28

some other way, obtain the roots. We shall call them as σ11 = 24.806 (algebraically largest),
σ22 = 6.635, and σ33 = −13.441 (algebraically smallest), all in MPa.
Having calculated the eigenvalues, let us proceed to calculate the eigenvectors. We shall
discuss the calculation of the first eigenvector (the eigenvector corresponding to the first —
largest — eigenvalue).
Calculation of the first eigenvector:
Let us now determine the eigenvector30 corresponding to σ = σ11 . Let us now substitute
the known value of σ11 = 24.806 in Eq. (4.21), p. 4-20, to yield
    
(σxx − 24.806) τyx τzx l 
   0 
τ (σ − 24.806) τ m = 0 . (4.37)
 
 xy yy zy 
   
τxz τyz (σzz − 24.806)  n   0 

This is the same as

(σxx − 24.806) l + τyx m + τzx n = 0;

τxy l + (σyy − 24.806) m + τzy n = 0;
τxz l + τyz + (σzz − 24.806) n = 0.

An examination of any of these sets of equations reveals that the solution is not unique. If,
for example, (l1 , m1 , n1 ) is a solution, then (αl1 , αm1 , αn1 ) is also a solution for all values
of α. What does this mean? Well, it means that we cannot obtain the absolute values of
(l1 , m1 , n1 ); we can obtain only the ratios. This observation leads us to conclude that there
is no harm in specifying the value of any one unknown. Thus, we may set, say, l1 = 1 and
compute the corresponding values of m1 and n1 . We would then obtain not the direction
cosines, but only the direction ratios.
In view of this realisation, shall we put31 l1∗ = 1? Substituting the known numerical

of the normal stress; thus, it cannot obviously be 15 MPa or less. Thus, a probable value is, say, 22 MPa.
If this is an approximate value, we can obtain the ‘correction’ h by using the Newton-Raphson method.
This involves expansion in a Taylor series.

To solve f (x) = 0; f (x0 ) 6= 0; f (x0 + h) = 0;

f (x0
f (x0 ) + hf 0 (x0 ) + · · · = 0 −→ h0 = − .
f 0 (x0 )
If the assumed value x0 is a good approximation, the value of h0 calculated thus will be small. If it is not,
revise the assumed approximate value as x1 − x0 + ho and go through the procedure again. Once one root
is obtained, we can divide the original cubic equation and obtain a quadratic equation, etc.
30
Is this the eigenvector (meaning that the eigenvector is unique), or only an eigenvector (implying that
there can be several — certainly more than one — eigenvectors)? We can see that the eigenvector can
be calculated only to within an arbitrary factor. In other words we can obtain only the ratios m1 /l1 and
n1 /l1 . These are not the direction cosines, but only the direction ratios. The direction cosines can be
calculated from the direction ratios by normalising them. We have stated above how this is done, but we
shall show this below again.
31
We shall use the starred letters to represent the direction ratios. Accordingly, the triplet (l1∗ , m∗1 , n∗1 )
stands for the direction ratios.
4-29 The State of Stress at a Point

values of the stress components, we obtain

12m∗1 + 8n∗1 = 14.806; (4.38a)

9.806m∗1 + 5n∗1 = 12; (4.38b)
5m∗1 + 31.806n∗1 = 8. (4.38c)

Here are now a set of three equations in only two unknowns, viz., m∗1 and n∗1 ! There arises
the question of consistency. Are these equations consistent? Sure, they are. The condition
for the consistency, we recall from linear algebra, is that the rank of the coefficient matrix
is equal to the rank of the augmented matrix. Here the augmented matrix cannot have
the rank 3, because the eigenvalues were found by setting the determinant of the coefficient
matrix equal to zero. We may, therefore, take any two of these three equations, solve for m∗1
and n∗1 , and check that the third is satisfied. There is really no need to check, but this is an
illustrative example. Therefore, we shall check and convince ourselves that the equations
are indeed consistent, and that all is well.
From the first two equations [Eqs (4.38a), (4.38b)] we obtain m∗1 = 1.1909152, n∗1 =
0.0643771. These satisfy the third equation (4.38c). If instead, we choose the second and
the third equations, we again obtain the same set of values. Having obtained the direction
ratios, we can readily work out the direction cosines as l1 = −0.642, m1 = −0.765, n1 =
−0.041.
Other eigenvectors:
The other two eigenvectors can also be determined by the same procedure. To calculate the
eigenvector corresponding to the second eigenvalue (of 6.635 MPa), we substitute σ = 6.635
in Eq. (4.21), p. 4-20 and follow the same procedure. It is long; there are no new lessons
to learn. Therefore this part of the example is omitted. The result is displayed as below.
Eigenvectors:
The corresponding eigenvectors are

eigenvalue 1: σ11 = 24.806 eigenvector: (l1 , m1 , n1 ) = (−0.642, −0.765, −0.041);

eigenvalue 2: σ22 = 6.635 eigenvector: (l2 , m2 , n2 ) = (0.617, −0.549, 0.564);
eigenvalue 3: σ33 = −13.441 eigenvector: (l3 , m3 , n3 ) = (0.454, −0.337, −0.825).

Verification - orthogonality:
We can verify — it is always a good habit to check for the numerical correctness and our
own certitude — that these eigenvectors are orthogonal. The eigenvalues are distinct and,
hence, the eigenvectors will be orthogonal.

1 and 2: (−0.642) × (0.617) + (−0.765) × (−0.549) + (−0.041) × (0.564) = 0;

2 and 3: (0.617) × (0.454) + (−0.549) × (−0.337) + (0.564) × (−0.825) = 0;
3 and 1: (0.454) × (−0.642) + (−0.337) × (−0.765) + (−0.825) × (−0.041) = 0.

Verification - invariants:
We can also verify that [Compare with Eq. (4.36).]

σ11 + σ22 + σ33 = 24.806 + 6.635 + 13.441 = 18 MPa

Advanced Mechanics of Solids 4-30

= σxx + σyy + σzz = 10 + 15 − 7 = 18 MPa = I1 ;

σ11 σ22 + σ22 σ33 + σ33 σ11 = 258 (MPa)2 = I2 ;
σ11 σ22 σ33 = 24.806 × 6.635 × (−13.441) = 2212 (MPa)3 = I3 .

The Case of Repeated Eigenvalues

When two or even three eigenvalues are equal, the corresponding eigenvectors may not be
orthogonal. Now the calculations can be a little tricky unless we have the support of clear
understanding32 . There are no difficulties in the calculation of the eigenvalues (principal
stresses). However, the calculation of the eigenvectors can pose some difficulties.
Now that the eigenvalues are not all distinct, the eigenvectors may not be orthogonal.
Now there are infinite principal planes. However, it is still possible to find a set of three
mutually perpendicular principal planes by using a scheme called the Gram-Schmidt or-
thogonalisation procedure. These facts can be seen readily and effortlessly from Lamé’s
stress ellipsoid. Seeing is believing!
These calculations have to be performed as part of our learning process. Once the
concepts are clearly understood, any software tool like maple or matlab may be used to
relieve the tedium of the calculations.
Invariants of the Stress Matrix
Let us realise that both the stress matrices [σij ] and [σi00 j 0 ] represent the state of stress at the
same point, though the coordinate systems are different. Both are matrix representations
of the same invariant, symmetric stress tensor at the same point. Now the principal stresses
and the corresponding principal planes at the same point cannot depend upon the choice of
the coordinate system. Thus, whether we work with the stress matrix [σij ] or with [σi00 j 0 ],
we must necessarily obtain the same principal stresses. This means that the characteristic
equations shown below must necessarily have the same coefficients.
σ 3 − [· · · ]σ 2 + [· · · ]σ − [· · · ] = 0; in the xi system
3 2
σ − [· · · ]σ + [· · · ]σ − [· · · ] = 0; in the x0i system
This observation leads us immediately to the conclusion that the coefficients must be the
same. That is,
I1 [in the xi system] = I1 [in the x0i system] first invariant I1 of the stress matrix;
I2 [in the xi system] = I2 [in the x0i system] second invariant I2 of the stress matrix;
I3 [in the xi system] = I3 [in the x0i system] third invariant I3 of the stress matrix.
On transformation of the coordinate systems, the stress components [σij ] do change to
[σi00 j 0 ], but these three combinations do not change; these are the three invariants of the
32
Routine calculations with the brain in the switched-off mode may not be possible. On the mathematical
side, the rank of the matrix, the fact there are n − r linearly independent solutions, etc. have to be
understood. A geometrical picture like Lamé’s stress ellipsoid would greatly help. We shall revisit this
topic later after our discussion of the geometrical representation of the state of stress at a point.
4-31 The State of Stress at a Point

stress matrix.

I1 = [σxx + σyyu + σzz ] = σx0 0 x0 + σy0 0 y0 + σz0 0 z 0

 

= [σ11 + σ22 + σ33 ] (4.39a)

2 2 2
 
I2 = σxx σyy + σyy σzz + σzz σxx − τxy − τyz − τzx
= σx0 0 x0 σy0 0 y0 + σy0 0 y0 σz0 0 z 0 + σz0 0 z 0 σx0 0 x0 − τx020 y0 − τy020 z 0 − τz020 x0
 

= [σ11 σ22 + σ22 σ33 + σ33 σ11 ] (4.39b)

2 2 2
 
I3 = σxx σyy σzz − σxx τyz − σyy τzx − σzz τxy + 2τxy τyz τzx
= σx0 0 x0 σy0 0 y0 σz0 0 z 0 − σx0 0 x0 τy020 z 0 − σy0 0 y0 τz020 x0 − σz0 0 z 0 τx020 y0 + 2τx0 0 0y0 τy0 0 z 0 τz0 0 x0
 

= [σ11 σ22 σ33 ] (4.39c)

These three invariants play important roles in theoretical developments. When numerical
problems are worked out, it is helpful to check the values of these invariants.
Comments:
The stress matrices [σij ] and [σi00 j 0 ] are similar33 matrices. Similar matrices have the same
eigenvalues, but not the same eigenvectors. How do we understand this statement in the
context of the state of stress at a point? Well, the principal stresses are the same, and so
are the principal planes. However, the same principal planes will be represented by two
different sets of direction cosines, because the coordinate systems are different. Thus, the
eigenvalues are the same, but the eigenvectors are — will have to be — different such that
the same principal planes are referred to.
Principal Stresses as Stationary Values
Let us recall that we defined the principal stress as the (normal) stress on a principal plane
which, in turn, was defined as that on which there is no shear stress. We may also seek the
stationary values of the expression for the normal stress and, thereby, obtain the principal
stress. We shall presently undertake this exercise.
The expression for the normal stress, we have seen, is given by Eq. (4.4) which is
reproduced below.

f (l, m, n) ≡ σνν = l2 σxx + m2 σyy + n2 σzz + 2lmτxy + 2mnτyz + 2nlτzx . (4.40)

Let us try to find the stationary values of this expression. This is a function of three
variables (l, m n). Being direction cosines, they must satisfy the condition

g(l, m, n) ≡ l2 + m2 + n2 − 1 = 0. (4.41)

Stationary values with a constraint: Lagrange multiplier method:

When we are required to find the maximum / minimum / stationary values of a function,
say, f (x, y, z) subject to a constraint condition, say, g(x, y, z) = 0, there is a beautiful
33
Similar is a technical word with a specific technical meaning; it should not taken as an ordinary word.
Advanced Mechanics of Solids 4-32

method34 called the Lagrange multiplier method35 . We define a new function h(x, y, z) =
f (x, y, z) − λg(x, y, z) and find the stationary values of h(x, y, z) subject to no constraint.
Extremise36 f (l, m, n) subject to the constraint g(l, m, n) = 0. To do this, define a
new function h(l, m, n) = f (l, m, n) − λg(l, m, n) and extremise this function subject to no
constraint. Thus, we have
∂h ∂h ∂h
= 0; = 0; = 0. (4.42)
∂l ∂l ∂l
The function h(l, m, n) here is [See Eqs (4.40), (4.41).]

h(l, m, n, ) ≡ l2 σxx + m2 σyy + n2 σzz + 2lmτxy + 2mnτyz + 2nlτzx − λl2 + m2 + n2 − 1. (4.43)

Applying the (necessary) conditions [Eq. (4.42)] to this function h(l, m, n), we obtain

2 [lσxx + mτyx + nτzx ] − 2λl = 0; (4.44a)

2 [lτxy + mσyy + nτzy ] − 2λm = 0; (4.44b)
2 [lτxz + mσyz + nσzz ] − 2λn = 0; (4.44c)

which are the same as

    
σxx τyx τzx  l
  l 
 
 τxy σyy τzy  m = λ m ; i.e., σ x = λ x . (4.45)
 
  
n 
τxz τyz σzz  n  

If we compare this equation with Eq. (4.14), we find that λ introduced as a Lagrange
parameter now appears as the eigenvalue! Note particularly the dual role37 played by the
Lagrange parameter!
The statement that we made earlier that the principal planes may be defined differently
(viz., (i) the planes on which the shear stress is zero, and / or (ii) the planes on which the
normal stress becomes stationary) stands vindicated.
Maximum Shear Stresses as Stationary Values
The maximum shear stresses also may be obtained as the stationary values of the expression
for the shear stress on an inclined plane. The latter, expressed in terms of the principal
stresses (σ11 , σ22 , σ33 ), is given by
h i1
τ(ν) = l2 m2 (σ11 − σ22 )2 + m2 n2 (σ22 − σ33 )2 + n2 l2 (σ33 − σ11 )2 .
2
(4.46)

Let us find the stationary values of this expression subject to the constraint l2 +m2 +n2 = 1.
A similar problem was discussed just now. We have seen that to extremise f (l, m, n)
34
Named after the famous French mathematician Joseph-Louis Lagrange (Jan. 1736 - April 1813)
35
Readers who are not familiar with this method are advised to read this from some good books and
understand it clearly. This is a common technique used extensively in applied mathematics.
36
Find the maximum / minimise / stationary value(s).
37
We should not fail to note and appreciate the beauty of applied mathematics in action.
4-33 The State of Stress at a Point

subject to the constraint g(l, m, n) = 0, it is sufficient to extremise h(l, m, n) ≡ f (l, m, n) +

λ g(l, m, n) subject to no constraint, where λ is a parameter, unknown to begin with, called
as the Lagrange parameter. Thus, this problem is formulated38 as:
h i
extremise f (l, m, n) ≡ l2 m2 (σ11 − σ22 )2 + m2 n2 (σ22 − σ33 )2 + n2 l2 (σ33 − σ11 )2

subject to the constraint g(l, m, n) ≡ l2 + m2 + n2 − 1 = 0.

We proceed by defining h(l, m, n) ≡ f (l, m, n)+λg(l.m.n) and extremising h(l, m, n) subject

to no constraint. This procedure gives us
∂h
=0 −→ lm2 (σ11 − σ22 )2 + ln2 (σ33 − σ11 )2 − λl = 0; (4.47a)
∂l
∂h
=0 −→ l2 m (σ11 − σ22 )2 + mn2 (σ22 − σ33 )2 − λm = 0; (4.47b)
∂m
∂h
=0 −→ m2 n (σ22 − σ33 )2 + l2 n (σ33 − σ11 )2 − λn = 0. (4.47c)
∂n
These are accompanied by the constraint condition l2 + m2 + n2 − 1 = 0.
Case (a) Minimum values:
We know from our understanding of the physical problem even without any calculation
that the shear stress on the principal planes is zero. The inclined plane with its normal (ν)
becomes a principal plane when

(i) (l = ±1, m = n = 0); (ii) (m = ±1, l = n = 0); (iii) (n = ±1, l = n = 0).

Case (b) Maximum values:

The maximum values are reached corresponding to three different sets of the direction
cosines. These are the following.
1 1
(i) (l = ± √ , m = n = 0); corresponding τ(max)1 = ± (σ22 − σ33 ) ; (4.48a)
2 2
1 1
(ii) (m = ± √ , l = n = 0); corresponding τ(max)2 = ± (σ33 − σ11 ) ; (4.48b)
2 2
1 1
(iii) (n = ± √ , l = m = 0); corresponding τ(max)3 = ± (σ11 − σ22 ) . (4.48c)
2 2

The maximum shear at the point is the grand maximum among these three maxima.
 
τgrandmax = maximum of τ(max)1 ; τ(max)2 ; τ(max)3
 
1 1 1
= maximum of (σ22 − σ33 ) ; (σ33 − σ11 ) ; (σ11 − σ22 ) .
2 2 2

What are these maxima? How can there be three maxima? Fig. 4.15 shows three planes on
which the shear stress reaches its stationary values. These three planes are shown shaded
38 2
Extremising τ(ν) is the same as extremising τ(ν) . This is done merely to simplify the algebra.
Advanced Mechanics of Solids 4-34

[Figs 4.15a, 4.15b, 4.15c]. These planes, as we can see from these figures, perpendicular to
the 12, 23, and 31 planes. The maxima on these planes are, respectively,

1 1 1
(σ11 − σ22 ) ; (σ22 − σ33 ) ; (σ33 − σ11 ) .
2 2 2

(a) Maximum shear stress (b) Maximum shear stress on (c) Maximum shear stress on
on the shaded plane the shaded plane the shaded plane

Figure 4.15: Planes of maximum shear stress (shaded): perpendicular to the planes 12,
23, 31. The maxima are (σ11 − σ22 )/2, (σ22 − σ33 )/2, (σ33 − σ11 )/2.

Octahedral Planes and Stresses on These Planes

Planes equally inclined to the three principal planes

are called octahedral planes. There are eight (8)
such planes. Their direction cosines referred to the
principal axes 1, 2, 3 are
1 1 1
±√ ; ±√ ; ±√ .
3 3 3
The stresses on such planes are important because
they are related to the theories of failure.
Figure 4.16: Octahedral plane

Planes which are equally inclined to the principal planes are called octahedral planes
[Fig. 4.16]. Being equally inclined, their direction cosines must satisfy the condition

1 1 1
l2 + m2 + n2 = 1; l = m = n; thus, ±√ ; ±√ ; ±√ .
3 3 3

There are thus eight (8) such planes. The stresses, or the expressions for the stresses, on
these octahedral planes have special significance because they are related to the theories of
failure of ductile materials.
4-35 The State of Stress at a Point

If (σ11 , σ22 , σ33 ) are the principal stresses, the stress components, both normal and shear,
can be obtained by stress transformation. We can see that the expressions for the stress
components, resultant poct , normal σoct and shear τoct , are given by
r
2 + σ2 + σ2
σ11 22 33
poct = ; (4.49a)
3
σ11 + σ22 + σ33
σoct = ; (4.49b)
3
1
q
τoct = (σ11 − σ22 )2 + (σ22 − σ33 )2 + (σ33 − σ11 )2 (4.49c)
3
1
q
= 2 (σ11 + σ22 + σ33 )2 − 6 (σ11 σ22 + σ22 σ33 + σ33 σ11 ). (4.49d)
3
These expressions are in terms of the principal stresses. These will, naturally enough, look
differently when expressed in terms of the (non-principal) stress components. There will
now be some shear terms also which will be absent when the expressions are written in
principal coordinate systems. They are [See Example 2, p. 13-3.]
σxx + σyy + σzz
σoct = (4.50a)
3
1
q
2 (σxx + σyy + σzz )2 − 6 σxx σyy + σyy σzz + σzz σxx − τxy


τoct = 2 − τ2 + τ2 (4.50b)
yz zx
3
1
q
(σxx − σyy )2 + (σyy − σzz )2 + (σzz − σxx )2 + 6 τxy


= 2 + τ2 + τ2 (4.50c)
yz zx
3

TENSOR CHARACTER OF THE STRESS AT A POINT

We have seen some aspects of the stress tensor. We shall now see the number of (indepen-
dent) components for scalars, vectors and tensors, and gain further insights into the tensor
character.
We know that the stress at a point can be specified by four (4) components, viz.,
σxx , τxy ; τyx , σyy in two dimensions, and σxx , τxy , τxz ; τyx , σyy , τyz ; τzx , τzy , σyy in three
dimensions. A little reflection would convince us that, in one dimension, there can be only
one component, viz., σxx for the stress at a point39 .
Next let us consider a vector such as, say, the velocity (of flow of a fluid) at a point.
There are now one (1) component, viz., Vx in one dimension; two (2) components, viz.,
Vx , Vy in two dimensions; and three (3) components, viz., Vx , Vy , Vz in three dimensions.
A scalar such as, say, the temperature at a point has just one component T in one, two and
three dimensions. We can see that Tx , Ty do not make sense.
Let us display the results of these observations systematically in the form of a table
[Table 4.3] below. The pattern is unmistakably clear. A scalar, a vector, and a tensor (of
39
These are with respect to a rectangular cartesian coordinate system. In other coordinate systems also,
the number of (independent) stress components still remains the same, although they are now represented
differently like σrr , τrθ , etc. The fact that τxy = τyx (to satisfy the equations of equilibrium in the absence
of body couples) does not make any difference in principle like 4 − 1 = 3, or 9 − 3 = 6.
Advanced Mechanics of Solids 4-36

rank two) have n0 , n1 , n2 components in n dimensions. Thus, they have, respectively

scalar: 10 = 1, 20 = 1, 30 = 1;
vector: 11 = 1, 21 = 2, 31 = 3; and
tensor: 12 = 1, 22 = 4, 32 = 9

components in one, two, and three dimensions. We may generalise further and state that
in an n-dimensional space, a tensor of rank r has nr (independent) components. This
observation also may lead us to regard a scalar as a tensor of rank 0, and a vector as a
tensor of rank 1. Judged in this light, a scalar is one step (generation?) lower than a vector
which, in turn, is one step (generation) lower than a tensor of rank two40 .

Table 4.3: Number of components of a scalar, a vector, and a tensor

n (dimensions) scalar vector tensor

(e.g., temperature) (e.g., velocity) (e.g., stress)
One-dimensional: n = 1 1 (= n0 ) 1 (= n1 ) 1 (= n2 )
Two-dimensional: n = 2 1 (= n0 ) 2 (= n1 ) 4 (= n2 )
Three-dimensional: n = 3 1 (= n0 ) 3 (= n1 ) 9 (= n2 )
··· ··· ··· ···
n-dimensional: n 1 (= n0 ) n (= n1 ) n2 (= n2 )

Scalars, Vectors and Tensors

By now we have had some familiarity with scalars, vectors and (second order) tensors. Let
us have a few quiet moments to look at them, reflect upon them, and gain an appreciation
of their relationships to one another. We shall take the nature of stress at a point as an
example for this exercise. Let us note:

(i) the stress at a point is (an example of) a second order tensor with 1, 4 and 9 compo-
nents in 1, 2 and 3 dimensions;

(ii) the stress at a point on a (given) plane is a vector with 1, 2 and 3 components in 1, 2
and 3 dimensions; and

(iii) the stress at a point on a (given) plane in a (given) direction is a scalar with 1 com-
ponent only in all the three cases of 1, 2 and 3 dimensions.
40
Shall we then state, if we will not be taken too literally and seriously, that the relationship between a
tensor (of rank 2) is exactly the same as that between a vector and a scalar? Perhaps we might point out
that the relationship between a (paternal) grandfather and a father is the same as that between the father
and his son!
4-37 The State of Stress at a Point

Thus, a scalar can be considered as a tensor of order zero, and a vector as a tensor of order
one. Furthermore,

(i) a second order tensor associates a vector with every direction — a given direction
specifies a plane which has a stress vector acting on it — just as
(ii) a vector associates a scalar with every direction — a given direction defines or specifies
the stress component in that direction on that plane, the component being a scalar.

The order and the rank of a tensor are the same; these two words may be used interchange-
ably.
Tensors: Transformation Law
We have pointed out a few aspects of tensors, but the crucial qualifying aspect is not any of
the above-mentioned facts. It is the transformation property (of the components of vectors
and tensors) that is of decisive importance41 . We do not discuss this any further, because
this aspect is dealt with in much greater detail elsewhere in the book.
Some Comments
We may also note that, just as a vector P can be written in terms of the base42 vectors
e 1 , e 2 , e 3 as
P = P1 e 1 + P2 e 2 + P3 e 3 ,
a second order tensor also may be written as
σ = σ11 e 1 e 1 + σ12 e 1 e 2 + σ13 e 1 e 3 + σ21 e 2 e 1 + σ22 e 2 e 2 + σ23 e 2 e 3 +
σ31 e 3 e 1 + σ32 e 3 e 2 + σ33 e 3 e 3
= σij e i e j ,
where e 1 e 1 , e 1 e 1 , etc. are referred to as dyads. A unit second order tensor (a 2 × 2 matrix
I with transformation properties), in particular, may be written as
I = δij e i e j .
[To digress a little bit, let us return to the state of pressure at a point in a fluid at rest. The
pressure, we had seen, is the same on all planes passing through the point, and is entirely
normal to the plane.
     
−p 0 0 −p 0 0 −p 0 0
     
 0 −p 0  −→  0 −p 0  −→  0 −p 0 
     
     
0 0 −p (xyz) 0 0 −p (x0 y0 z 0 ) 0 0 −p (x00 y00 z 00 )
41
It is when judged from this correct point of view that ‘three-index entities’ like the Christoffel symbols are
not tensors, even though they too have the relevant number of components.
42
These are also the unit vectors in our context of only rectangular cartesian coordinates. When we use the
general curvilinear coordinates, we shall see that the base vectors are not unit vectors! It is only in some
special cases that the base vectors are unit vectors.
Advanced Mechanics of Solids 4-38

There is no tangential component; that is, a fluid at rest cannot sustain any shear stress.
Let us note that this matrix has the same appearance in all coordinate systems! This is a
very special case of the general state of stress at a point in a solid, and is an example of an
isotropic tensor.
We repeat for emphasis: the pressure at a point in a fluid at rest is an example of a
second order tensor, but being an isotropic tensor it can be regarded as a scalar. Just one
number (say, −p = −4 kPa) will do to specify the pressure at a point, which is a feature of
a scalar.]
We shall close this section with the following statements.
The stress at a point −→ a second order tensor;
the stress at a point on a plane −→ a vector;
the stress at a point on a plane in a direction −→ a scalar;
the stress at a point −→ invariant, symmetric.

STATES OF STRESS: HYDROSTATIC AND PURE SHEAR

We have seen earlier the two special states of (i) hydrostatic stress and (ii) pure shear.
They are also referred to as the spherical and deviatoric parts of the stress tensor. These
qualifiers, spherical and deviatoric, are appropriate. For the case of a hydrostatic stress,
the three principal stresses are equal, and that the Lamé’s ellipsoid of stresses becomes a
sphere. Furthermore, for a (two-dimensional) curvature tensor, when the two eigenvalues
— the principal curvatures — are equal, the surface there is indeed spherical!
Hydrostatic and Deviatoric Parts in Stress Space
It is also insightful to look at these special states of stress in a Westergaard stress space
[Fig. 4.17]. The line OA being equally inclined to the principal stress axes σ11 , σ22 , σ33
represents all hydrostatic states of stress. AB is a measure of how far the state of stress A
deviates from a state of hydrostatic stress — that is, how far it deviates from the line OP .
OB and BA represent, respectively, the spherical and the deviatoric parts of the state of
stress A. The hydrostatic part corresponds to the spherical part, and the deviatoric part
to the pure shear part of the stress tensor (state of stress represented by the point A in the
Westergaard stress space).
Hydrostatic State of Stress (Spherical Part of the Stress Tensor)
We have seen that a hydrostatic state of stress is represented by the stress matrix
 
p 0 0
0 p 0 −→ p I −→ p δij .
0 0 p
This matrix has the same appearance in every coordinate system. The corresponding Lamé’s
ellipsoid of stresses becomes a sphere.
4-39 The State of Stress at a Point

A point A is marked in the Westergaard stress space.

The line OB is equally inclined to the three princi-
pal axes σ11 , σ22 , σ33 . Accordingly, every point on
this line (and on its extension) represents a state of
hydrostatic state of stress. The distance AB is a
measure of how far the state of stress represented
by the point A deviates from the hydrostatic state
of stress. Thus, OB and BA represent, respectively,
Figure 4.17: A point A in the the spherical and the deviatoric parts of the state of
Westergaard stress space stress A.

A State of Pure Shear (Deviatoric Part of the Stress Tensor)

It seems natural to define a state of pure shear as explained below. If the stress matrix
appears in the form  
0 τyx τzx
τxy 0 τzy 
τxz τyz 0
with no normal stress components — that is, only zeros on the leading diagonal — it can
be regarded as representing a state of pure shear. However, referred to the principal axes,
even this matrix representing a state of pure shear will appear as a diagonal matrix:
   
0 τyx τzx σ11 0 0
τxy 0 τzy  −→  0 σ22 0  .
τxz τyz 0 xyz 0 0 σ33 123

If the first matrix (the one on the left) represents a state of pure shear, the second matrix
(the one on the right) too must represent the same state of pure shear. After all, they both
represent the same invariant stress tensor! Hence it becomes necessary for us to revise our
definition of the state of pure shear. Just because there are non-zero elements on the leading
diagonal, the stress matrix need not necessarily cease to represent a state of pure shear. We
know that for the two matrices shown, the first invariant43 must be the same. Thus, we
are led to the conclusion that as long as the trace of the matrix (the first invariant) I1 = 0,
that matrix will represent a state of pure shear. It is possible to show that a necessary and
sufficient condition for a matrix to represent a state of pure shear is that

the first invariant, I1 = σxx + σyy + σzz = σ11 + σ22 + σ33 = 0. (4.51)

The first part — the necessity — is obvious. The second part — the sufficiency — is
not quite as easy to prove. That is, we are required to prove44 that, if the first invariant
vanishes, there exists a rectangular cartesian coordinate system referred to which the stress
matrix appears with only zero elements on its leading diagonal.
43
Not only the first invariant, but the other two invariants too
44
Such proofs are important for mathematicians. They are not so important for engineers. A plausibility
argument is quite sufficient for us. It is always nice to have the support of a rigorous proof, though.
Advanced Mechanics of Solids 4-40

Any (general) state of stress can always be broken up into two special states of stress,
viz., (a) a hydrostatic state of stress, and (b) a state of pure shear. But why do we need to
do this? When we learn theories of failure (also called theories of strength), we come across
a famous criterion of failure45 called the von Mises’ criterion. This criterion is associated
with the strain energy of the pure shear component. Thus, it is necessary for us to split the
general stress matrix (at the most vulnerable point; that is, wherever we desire to check the
safety of the machine part) into the two matrices, and then compute the strain energy of
the pure shear part. This part of the strain energy is called the distortion strain energy.
     
σxx τxy τxz σ 0 0 σxx − σ τxy τxz
     
τ σ τ = 0 σ 0 + τ σ − σ τ yz  . (4.52)
     
 yx yy yz     yx yy
     
τzx τzy σzz 0 0 σ τzx τzy σzz − σ

We desire that the first matrix in Eq. (4.52) should represent a hydrostatic state of stress,
and that the second one a state of pure shear. The first matrix is now in good shape to
represent a state of hydrostatic stress; any value of σ will do. But if the second matrix
should represent a state of pure shear, only one value of σ can be used. The condition of
pure shear, we recall, is that the first invariant should be zero. Thus, we require that

1st invariant (of the second matrix): [σxx − σ] + [σyy − σ] + [σzz − σ] = 0.

This gives us the required value of σ as σ = 13 [σxx +σyy +σzz ]. With this choice for the value
of σ, the desired decomposition is obtained. It is of theoretical and practical importance to
do this. Theory of plasticity and theories of strength are two areas of application among
others where such a decomposition is necessary.
THEORIES OF FAILURE: TRESCA’S AND VON MISES’ CRITERIA
Theories of failure (also called theories of strength) are suggested to predict in advance
when a material (say, a machine part) subjected to a complex state of stress fails. This is
to be done with the limited information obtained from a simple tension test.
When does a material or a machine part fail? We discuss failure only from the point
of stress analysis (and not from other considerations like bad appearance and unsuitability
to serve the very purpose of the machine part). Perhaps it is natural to postulate that a
material fails when the maximum stress exceeds the safe limit. The safe limit is, of course,
to be obtained experimentally in a material testing laboratory. The experiment is the simple
tension test usually performed on a universal testing machine. This is for ductile materials.
(For a material like concrete, a compression test is used.)

45
Named after the Austrian-German-American applied mathematician Richard Edler von Mises (April 1983
- July 1953) — not to be confused with Ludwig von Mises, the well known Austrian theoretical economist.
He had made a mark as an applied mathematician. It was he who founded the prestigious journal ZAMM
(Zeitschrift für Angewande Mathematik und Mechanik) (Journal of Applied Mathematics and Mechanics),
the first of its kind.
4-41 The State of Stress at a Point

Simple Tension Test

The usual tension test is the most common and fundamentally important test. This test
is under a uniaxial state of stress46 [Fig. 4.18]47 . With this (limited) information, we are
required to judge when a (vulnerable, critically ‘stressed’) point in a machine part subjected
to complex three dimensional stresses fails. This is what a theory of failure or a theory of
strength is concerned about48 .
One-dimensional state of stress: only σ11 = σy exists.
The stress matrix corresponding to the simple tension test is
 
σy 0 0
 0 0 0 ,
0 0 0

where σy , the yield stress — the stress at which the specimen

is deemed to have failed — is obtained experimentally. With
this meagre information, we are required to judge when a ma-
chine part will fail in a complex — in general, a 3-dimensional
Figure 4.18: A simple — state of stress. This is the relevance of a theory of failure,
tension test also called a theory of strength.

What Causes Failure?

How are we sure that it is the high stress that is responsible for the failure? It could very
well be the high strain that accompanies the high stress that causes failure? How do we
know for sure? To settle this issue, we have to be innovative. We should devise a loading so
that the stress exceeds the ‘safe’ limit, but the strain does not, or vice versa, if that would
be possible. But there are other situations that force us to abandon both these theories.
For example, we know that at the bottom of the sea, there are rocks that are perfectly
intact. If we bring a specimen of the same rock and test it in the laboratory, it gets crushed
at a much, much lower stress! How can we explain this? Well, we have to come up with
some plausibility arguments49 . Usually there are extra supporting experimental facts, or
established theories, that lend credence to such hypotheses.
46
The state of stress is complex and no more uniaxial in the later part of the test after the phenomenon
of ‘necking’, but we disregard all these complexities and consider the ‘simple’ tension as a uniaxial test
producing a uniaxial state of stress in the test specimen.
47
There are many details pertaining to this important test. The specimen is not of uniform cross-section as
shown here. The figure is in this sense somewhat misleading.
48
Readers not exposed to the details of this topic should consult two or three good books. Some books on
machine design also give some aspects of this topic fairly well.
49
To come up with some credible stories is given the more honorable name of proposing a new theory.
This is said to be science, but it is no more than suggesting a possibility that has no inherent internal
contradictions and that is able to explain the observed phenomenon or phenomena and the experimental
evidence. A hypothesis in science is really a bed time story except that there are pompous technical words
and several unfriendly symbols of higher mathematics! No irreverence is meant by this statement.
Advanced Mechanics of Solids 4-42

(a) Tresca’s criterion (b) Von Mises’ criterion

Figure 4.19: The figures represent in a 2-dimensional stress space the ‘safe’ regions (shaded).
Tresca’s criterion is slightly more conservative than the von Mises’ one.

On the physical side we know — it is our experience — that brittle and ductile materials
behave differently. Thus, we may have to understand the internal mechanism50 that is
responsible for failure before we can propose a theory of strength convincingly51 . These
considerations take us away into the realm of materials science (deformation processes in
metals, dislocation theory, etc.), a vast area in its own right. We cannot discuss these here.
Nevertheless, we may perhaps point out that yielding (in ductile materials) is the result of
slippage of certain crystal planes. This happens along the surface of maximum shear stress.
Thus, there is some physical ground to assume that the maximum shear stress is of greater
relevance.
Shall we, then, postulate that it is the maximum shear stress or some entity dependent
on this that causes failure? Are there further arguments supporting this point of view?
Is Shear Stress the Villain?
A brilliant idea is to suppose that it is the shear stress, or some entity dependent on the
shear stress, that causes failure. This would explain why rocks deep down at the bottom of
the sea remain safe; the state of stress that the rocks are subjected to is one of hydrostatic
stress. No matter how high the water pressure is, there is no shear stress at all on any plane!

50
Siavouche (Sia) Nemat-Nasser, the leading expert in mechanics at (U. of California at) San Diego, states
in the preface of his book, Plasticity: A Treatise on the Finite Deformation of Heterogeneous Inelastic
Materials, Cambridge University Press, New York, (2004): “· · · It became clear to me that true advances
in the basic understanding of the mechanics of materials, and particularly the inelastic deformation of
materials and geomaterials, can be achieved only by moving beyond the traditional phenomenological
approach to plasticity models · · · ” combining “· · · the mathematical rigor of solid mechanics with the
physics-based micro-structural understanding of the materials science · · · ”.
51
To see how hard it is to understand the phenomenon of failure, we suggest that we try to define when a
person can be said to have died. Even with our perfect ignorance in matters related to life and death, we
might be persuaded to propose this in different ways. When can a person be said to have died? When his
brain fails, heart fails, kidney fails, all the vital organs fail? We shall try to avoid further complications by
restricting ourselves to clinical death. (Has not Shakespeare said that fools die several times?) We could
even state that a person is dead when his wife becomes a widow! This is fine, except that we must also
address issues like the following. (i) Is there an independent way of finding out if and when a (married)
woman becomes a widow? (ii) What about death of women, unmarried men, children, etc.? (iii) Yogis are
said to be able to enter a state of samādhi when all breathing is suspended. They look, it is said, lifeless
for all untrained people.
4-43 The State of Stress at a Point

This is why, this theory postulates, rocks can stand very, very high52 hydrostatic stresses
while they crumble at much, much lower non-hydrostatic states of stress. In this way two
important theories emerged: (i) the maximum shear stress theory, and (ii) the maximum
shear energy (related to the shear stress part only) theory. The first is called the Tresca’s53
criterion of failure, while the second is known as the von Mises’ criterion. These are the
most commonly used theories of failure to check whether or not a machine part is safe. We
shall discuss these two theories briefly.
Tresca’s Criterion
Tresca’s criterion of failure, known also known as Guest’s criterion, states that the (ductile)
material fails when the maximum shear stress (at the most vulnerable, critically stressed
point) in a complex 3-dimensional state of stress reaches the critical value of the maximum
shear stress in a simple 1-dimensional tensile test54 . If the material fails — yielding is
considered to be failure — at a yield stress of σ11 = σy which corresponds to the maximum
shear stress τmax = (σ11 − 0)/2 = (σy − 0)/2 in a simple tension test, the limit of safety is,
as this theory suggests, when the maximum shear stress in a complex 3-dimensional state
of stress reaches this value of (σy − 0)/2. We have seen that the maximum shear stress
is half the value of the difference between any two principal stresses. Thus, the onset of
yielding is predicted by this theory as: the maximum shear in the complex state of stress =
the maximum shear stress at failure in a simple tension test, i.e., σy /2. The limit of failure
is thus given by Eq. (4.53) below.
σy
simple tension test: σ11 = σy ; τmax = ;
 2 
|σ11 − σ22 | |σ22 − σ33 | |σ33 − σ11 |
complex stress: τmax = max. of ; ; ;
2 2 2
 
|σ11 − σ22 | |σ22 − σ33 | |σ33 − σ11 | σy
onset of yielding: max. of ; ; = ;
2 2 2 2
i.e., max. of [|σ11 − σ22 |; |σ22 − σ33 |; |σ33 − σ11 |] = σy . (4.53)

If σ33 = 0 (2-d), the above criterion or prediction of the onset of yielding becomes55

max. of [|σ11 − σ22 |; |σ22 − 0|; |0 − σ11 |] = σy . (4.54)

Tresca’s criterion is popular, but there is another criterion, called von Mises’ criterion, that
is also just as popular, if not more popular. We shall see this below.
52
If we recognise that the sea is in some places as deep as 10 km or even deeper, and that the density of salt
water is slightly more than that of water, we can see how high the water pressure is at the bottom of the
sea. We suggest that the young readers do carry out this numerical calculation. The value is, we venture
to state, much larger than what we might imagine.
53
Henri Édourd Tresca (Oct. 1814 - June 1885), French, professor of mechanical engineering
54
Named after H.É. Tresca. Also called Guest’s criterion.
55
Let us note clearly that in this case of σ33 = 0 if, say, σ11 = 30 and σ22 = 20 (both in MPa), the maximum
shear stress at the point is not (30 − 20)/2 = 5 MPa, but (30 − 0)/2 = 15 MPa!
Advanced Mechanics of Solids 4-44

Von Mises’ Criterion

This theory56 postulates that it is the strain energy associated with the pure shear part
— the deviatoric part — that decides failure. When this component of the strain energy,
called the distortion energy reaches its limit equal to the distortion energy at the point of
failure (yielding) in a simple tension test, the material fails.
To develop the equation that predicts failure according to this theory, we again start
from the result of the simple tension test. The stress matrix for this case of one-dimensional
loading is broken or decomposed into the hydrostatic (spherical component) and the pure
shear (deviatoric component) parts. We then calculate the strain energy associated with
the pure shear part. This is the limiting value at failure of the distortion energy.
For the simple tension test:
We need to split up the stress tensor into (i) the hydrostatic part and (ii) the pure shear
part as shown below. We can readily recognise that the two matrices on the right hand
side below represent, respectively, (i) the hydrostatic part (the three principal stresses are
all equal, spherical part, isotropic tensor) and (ii) the pure shear part (sum of the three
elements on the diagonal = 0, deviatoric part).
   σy   2σy 
σy 0 0 3 0 0 3 0 0
     
σy σy
 0 0 0 =  0 0 + 0 − 0 (4.55)
     
   3 
 
 3 

σy σ
0 0 0 0 0 3 0 0 − 3y
The distortion energy (the strain energy corresponding to the second matrix, the pure shear
part) per unit volume is calculated57 . It is equal to
1+ν 2
distortion energy (simple tension test): σ . (4.56)
3E y
For the complex 3- dimensional loading:
For this case, we need to split up the matrix representing the state of stress into two parts
as before. The distortion energy corresponding to the second matrix — the pure shear part
— is to be calculated again as before.
     
σ11 0 0 p 0 0 σ11 − p 0 0
     
 0 σ22 0  = 0 p 0 +  0 σ22 − p 0 , (4.57)
     
     
0 0 σ33 0 0 p 0 0 σ33 − p
56
Also known as the Maxwell-Huber-Hencky-von Mises theory. It seems to have been formulated by James
Clark Maxwell as early as 1865, but is usually attributed to Richard Edler von Mises (1913). Tytus
Maksymilian Huber is said to have anticipated this (1904).
57
The calculation of strain energy presupposes some knowledge of the strain components related to the stress
components through the constitutive equations. Thus, to understand this properly, we need to wait until
we discuss the strain components and the constitutive constants (the elastic constants), ν (Poisson’s ratio)
and E (Young’s modulus of elasticity). As the readers have already had some familiarity with these ideas,
we let this sentence stay here as it is.
4-45 The State of Stress at a Point

where
[σ11 + σ22 + σ33 ]
p= .
3
The distortion energy (the strain energy corresponding to the second matrix, the pure shear
part) per unit volume is calculated. It is equal to
1+ν 
(σ11 − σ22 )2 + (σ22 − σ33 )2 + (σ33 − σ11 )2 .

distortion energy (general case):
6E
(4.58)
Equating these two expressions [Eqs 4.56, 4.58], we find that the criterion for yielding
(failure) is
1+ν   1+ν 2
(σ11 − σ22 )2 + (σ22 − σ33 )2 + (σ33 − σ11 )2 = σ ;
6E 3E y
1 h i
i.e., (σ11 − σ22 )2 + (σ22 − σ33 )2 + (σ33 − σ11 )2 = σy2 . (4.59)
2
Onset of yielding:
The onset of failure (yielding) is thus predicted by this theory by the equation
1h i
(σ11 − σ22 )2 + (σ22 − σ33 )2 + (σ33 − σ11 )2 = σy2 . (4.60)
2
In two dimensions with σ33 = 0 (the case of plane stress), the corresponding equation —
the von Mises’ criterion — is
2 2
|σ11 − σ11 σ22 + σ22 | = σy2 . (4.61)

It should be obvious that for the material to be safe, Eqs (4.60, 4.61) will appear slightly
modified, respectively, as
1h i
(σ11 − σ22 )2 + (σ22 − σ33 )2 + (σ33 − σ11 )2 < σy2 ; and (4.62)
2
2 2
|σ11 − σ11 σ22 + σ22 | < σy2 . (4.63)

Tresca’s and von Mises’ Criteria: A Comparison

Both these criteria are widely used. Tresca’s criterion is slightly more conservative than the
von Mises’ one. We can see this from the fact the Tresca’s yield surface is circumscribed by
the von Mises’ one.
Many scientists have performed experiments to check the validity of these criteria. The
experiments by Taylor & Quinney58 are very famous. They applied various combinations of
58
G.I. Taylor & H. Quinney (1932). Geoffrey Ingram Taylor (Mar. 1886 - June 1975) was an outstanding
British mathematician-scientist-engineer who was for the most part at the prestigious Cavendish Labora-
tory. The three great engineer-mathematician-scientists, the other two being S.P. Timoshenko (Stanford)
and Th. von Kármán (CalTech) dominated the scene of mathematical engineering. They all had long and
active professional lives and lived at about the same time. Taylor’s student, G.K. Batchelor, a leading light
in Fluid Mechanics in his own right, writes glowingly about his teacher. See his book, G.K. Batchelor:
The Life and Legacy of G.I. Taylor, Cambridge University Press, (2008).
Advanced Mechanics of Solids 4-46

(a) A thin (copper) tube subjected to axial and (b) Experimental results and the predictions as
torsional loading per the two criteria

Figure 4.20: Taylor & Quinney: Experiments on thin tubes subjected to combined axial
tension and torsion. Fig. 4.20a shows the specimen, a thin tube of circular cross-section
subjected to (a pair of) axial forces and to (a pair of) twisting moments. Their results are
shown in Fig. 4.20b. It can be seen that the experimental points (marked by heavy dots)
lie between the curves aaaa (the lower one) and bbbb which are the predictions as per the
Tresca’s and von Mises’ criteria, respectively.

axial tension and torque on thin-walled tubes and, thus, obtained several cases of combined
stresses. Their results59 show that the experimental points fall between the curves predicted
by the two criteria. The von Mises’ prediction is slightly more accurate (because the exper-
imental points are closer), but it must be conceded that it is harder to apply compared to
the Tresca’s criterion. The Tresca’s criterion is about 15% more conservative than the von
Mises’ one. These two are of importance in the theory of plasticity also.
CLOSING REMARKS
This topic of stress at a point is of great importance. This serves as a foundation upon
which are built the mechanics of solids, the theory of elasticity, continuum mechanics,
experimental and theoretical stress analysis and finite element methods. To be able to solve
stress analysis problems we need to discuss all the governing equations also. We shall see
them in the subsequent chapters.

In the next chapter, we shall have some more discussion of stresses.

59
Thin-walled tubes may be subjected to various combinations of internal pressure and axial tension. This is
another possibility. Such experiments were carried out and experimental results obtained by Lode (1928)
and confirmed by other scientists.
Chapter 5

MORE ABOUT STRESSES

So far we confined ourselves to examining the nature of stress at a point. In a stress analysis
problem, our interest is not limited to a single point; we are concerned with how the various
stress components change (vary) from point to point. For example, if we look for the
most critically stressed point, we have in our minds the entire stress field, and we locate
the critical (vulnerable) point where the normal stress (or shear stress, or the distortional
energy per unit volume) is the largest. Thus, the topics treated in the remaining part of
this chapter still refer to a point, but have reference to several points in the stress field.
We need to discuss all the governing equations before we can solve stress analysis prob-
lems. These governing equations will be discussed one by one later in this book. The most
important among these topics is the differential equations of equilibrium.
DIFFERENTIAL EQUATIONS OF EQUILIBRIUM
We have so far been discussing the state of stress at a point. We shall now start thinking
about how the stress components can change from point to point. Such variations are
governed by some fundamental equations known as the field equations. Among them the
ones that are of the greatest importance are the (differential) equations of equilibrium.
These differential equations also are local equations, in the sense that they must hold at
each point inside the domain. In this sense, it is not inappropriate to discuss the differential
equations of equilibrium as part of the nature of stress at a point.
The physical requirement that these equations demand is that every part of the body
must be in equilibrium. Fair enough! If we consider an arbitrary block, small or large, of
(stressed) material in the domain, there are stress components acting at all points of the
block. The components of stress may change from point to point, but they may change only
in such a way as to keep the block, small or large, in equilibrium. This basic requirement
places some restrictions on how the stress components may change from point to point.
This requirement is finally expressed as a set of differential equations to be satisfied at all
points in the domain (inside the body). We shall see the details below.
Advanced Mechanics of Solids 5-2

Introduction
We shall now consider the differential equations of equilibrium. The basic physical idea
behind these equations, as stated above, is simple: the entire body, adequately supported,
is in equilibrium. Furthermore, every tiny little bit of the body also is in equilibrium.
We shall invoke the (necessary and sufficient) conditions of equilibrium and arrive at these
important equations. These equations can be modified by including the so-called ‘inertia
forces and inertia moments’ so that they are now applicable for dynamical problems also
(using D’Alembert’s principle).
Stress Components on a Block

(a) The stress components acting on a block (b) Stress components in the y-direction

Figure 5.1: The stress components acting on the various faces of the block are shown in
Fig. 5.1a. Only the stress components acting along the y-direction are marked in Fig. 5.1b.
These are used to derive the differential equation of equilibrium in the y-direction.

As the forces are the externally applied forces and those associated with the stress
components, let us consider the stress components acting on a block. To reduce the writings
in the figure so as not to clutter them as far as possible, the following notations are used.

11 : σxx 22 : σyy 33 : σzz

∂σxx ∂σyy ∂σzz
11 : σxx + dx 22 : σyy + dy 33 : σzz + dz
∂x ∂y ∂z
12 : τxy 23 : τyz 31 : τzx
∂τxy ∂τyz ∂σzx
12 : τxy + dx 23 : τyz + dy 31 : τzx + dz
∂x ∂y ∂z
23 : τyz 31 : τzx 12 : τxy
∂τyz ∂τzx ∂σxy
23 : τyz + dy 31 : τzx + dz 12 : τxy + dx
∂y ∂z ∂x
31 : τzx 12 : τxy 23 : τyz
∂τzx ∂τxy ∂σyz
31 : τzx + dz 12 : τxy + dx 23 : τyz + dy
∂z ∂x ∂y
5-3 More About Stresses

The entire body is in equilibrium under the influence of surface and body forces. Every
part of it, small or large, is also in equilibrium. Let us consider a block and mark all the
stress components on the various faces of the block [Fig. 5.1a]. The various faces, their
areas, and the stress components concerned are shown below.
∂σxx
OEF G dy dz σxx ABCD dy dz σxx + dx
∂x
∂τxy
OEF G dy dz τxy ABCD dy dz τxy + dx
∂x
∂τxz
OEF G dy dz τxz ABCD dy dz τxz + dx
∂x
∂τyx
OGDA dx dz τyx EF CB dx dz τyx + dy
∂y
∂σyy
OGDA dx dz σyy EF CB dx dz σyy + dy
∂y
∂τyz
OGDA dx dz τyz EF CB dx dz τyz + dy
∂y
∂τzx
OABE dx dy τzx EF CB dx dy τzx + dz
∂z
∂σzz
OABE dx dy σzz GDCF dx dy σzz + dz
∂z
∂τzy
OABE dx dy τzy GDCF dx dy τzy + dz
∂z
These are the various stress components acting on the various faces of the block. These are
all acting on surfaces and they contribute to surface forces. In addition, there can be body
forces acting on the volume of the block. If Fx , Fy , Fz are the body forces per unit volume,
they lead to the forces Fx dV = Fx dx dy dz; Fy dV = Fy dx dy dz; Fz dV = Fz dx dy dz in
the x, y, z directions, respectively. These too have to be taken into account.
Fig. 5.1a is too cluttered. Hence for greater clarity, the stress components are shown
separately in three figures [5.2a, 5.2b, 5.2c]. These are the stress components in the x-, y-,
and z-directions relevant when the equations of equilibrium are written.
Equations of Equilibrium: Forces
Having considered all the forces on an elemental block, we can now write down the equations
of motion. Let us refer to Fig. 5.1b where the relevant stresses that contribute to forces in
the relevant direction — here the y-direction — are shown. To these we must add the body
force Fy dx dy dz also. Thus, we have
       
∂τxy ∂σyy
−τxy dy dz + τxy + dx dy dz + −σyy dx dz + σyy + dy dx dz +
∂x ∂y
   
∂τzy
−τzy dx dy + τzy + dz dx dy + Fy dx dy dz = 0.
∂z
On cleaning this up, we obtain the differential equation of equilibrium in the y-direction.
Its companion equations in the x- and y-directions can be obtained similarly (by changing
Advanced Mechanics of Solids 5-4

(a) Stress components in the x- (b) Stress components in the y- (c) Stress components in the z-
direction direction direction

Figure 5.2: The stress components acting on the various faces of the block, but all in the
same direction. are shown in the three figures, the first [Fig. 5.2a - x direction] and so on.
These are convenient when we derive the differential equations of equilibrium in the three
(x, y, z) directions.

x to y, y to z, and z to x). These three equations are displayed below, and enclosed in a
box because of their importance.

Differential equations of equilibrium:

∂σxx ∂τyx ∂τzx

+ + + Fx = 0; (5.1a)
∂x ∂y ∂z
∂τxy ∂σyy ∂τzy
+ + + Fy = 0; (5.1b)
∂x ∂y ∂z
∂τxz ∂τyz ∂σzz
+ + + Fz = 0. (5.1c)
∂x ∂y ∂z

In index notation these are written as

σji,j + Fi = 0 which is the same as σij,j + Fi = 0, (i = 1, 2, 3). (5.2)

The differential equations of equilibrium may be presented in matrix form as

 

 σxx 

∂
0 0 ∂
0 ∂

 σyy 
    
  Fx   0 
 
∂x ∂y ∂z 
 
∂ ∂ ∂  σzz
 0 ∂y 0 ∂x ∂z 0  + Fy = 0 . (5.3)

∂ ∂ ∂ τxy 
Fz 0
    
0 0 ∂z 0 ∂y ∂x   
τ 
 
 yz 

 
τzx


We may write this in the form

LT τ + F = 0 , (5.4)
5-5 More About Stresses

where L is a linear (differential) operator operating here on the stress matrix τ written1
slightly differently as
{ττ } = [σxx σyy σzz τxy τyz τzx ]T (5.5)
We shall see this operator L again later in a different context2 . We shall obtain this result
again later following a different procedure.

(a) Moment about the x-axis: (b) Moment about the y-axis: (c) Moment about the z-axis:
τyz = τzy τxz = τzx τxy = τyx

Figure 5.3: To show that τij = τji : the three moment equations of equilibrium lead to the
result that the stress matrix is symmetric. If there is a locked-in moment, the stress matrix
will not be symmetric, which leads to a lot of complications as in micro-polar elasticity.

Equations of Equilibrium: Moments

We have seen the equations of equilibrium of forces. We shall now examine the equations
of moment equilibrium. We shall wee that the moment equations of equilibrium lead to the
conclusion that the stress matrix is symmetric, i.e., τij = τji .
We shall first consider the moment equation of equilibrium about the x-axis. Referring
to, say, Fig. 5.1b, the relevant shear stresses that contribute to this moment are:

face OGDA: τyz area: dx dz;

∂τyz
face EFCB: τyz + dy area: dx dz;
∂y
face OABE: τzy area: dx dy;
∂τyz
face GDCF: τzy + dz area: dx dy;
∂z
1
The stress matrix τ is written here as a column matrix. It is displayed here as the transpose of a row
matrix to save space. A notation {ττ } different from [σσ ] (which is a 3 × 3 matrix) is used here.
2
Strain-displacement relations connecting the strain components and the associated displacement ones. It
looks surprising that the same operator should appear in these two contexts which are entirely different
from the physical point of view: (i) equilibrium of forces associated with stress components (no geometrical
considerations here), and (ii) the strain components related to the associated displacements based on
geometry (no equilibrium considerations here)! Nature seems to have interconnections at a deeper level.
Is everything connected to everything else, as Buddhists are said to believe?
Personal discussions with Dr Gangan Prathap are gratefully acknowledged. How delightful and insightful!
Advanced Mechanics of Solids 5-6

If we take moments of these associated shear forces about the x-axis, we find that
   
∂τyz ∂τzy
τyz + dy × (dx dz) × dy − τzy + dz × (dx dy) × dz = 0,
∂y ∂z
from which we conclude that τyz = τzy . Quantities of higher order of smallness can be,
and are, neglected. We may also mention that the contribution to this moment (about
the x-axis) of (i) the normal stresses on the faces OGDA, EF CB, OABE, GDCF and of
(ii) the shear stresses on the faces OEF G, ABCD is only negligibly small (higher order of
smallness). During the limiting process, the quantities of higher order of smallness → 0
much faster and, therefore, drop out of the resulting equation.
In the same way we can show that3 (i) τzx = τxz , and (ii) τxy = τyx . We shall obtain
this result again later following a different procedure4 .
Two-dimensional Case

(a) Stress components on a block (b) To show that τxy = τyx

Figure 5.4: The stress components acting on a rectangular block are shown in Fig. 5.4a. In
Fig. 5.4b are shown the shear stresses acting on the various faces. These are meant to be
of help in deriving the equations of equilibrium, and in showing that τxy = τyx .

Two-dimensional cases are of special importance. The general three-dimensional prob-

lem is so difficult to solve that often simplifications are made. Among them perhaps the
most important one is the two-dimensional simplification. In experimental stress analysis
it is essential to know the two-dimensional case thoroughly.
A rectangular block with the stress components is shown [Fig. 5.4]. We shall take the
help of Fig. 5.4a to derive the two differential equations of equilibrium, and of Fig. 5.4b
to show that τxy = τyx (which is really the result obtained from the moment equation of
equilibrium). As in earlier cases, we consider the stress components and the associated
areas, and write down the equations of equilibrium in the x- and y- directions. The body
forces are Fx and Fy per unit area. The various faces, their areas and the stress components
acting on them are identified below. The thickness of the block is taken as 1.
face CO: area: dy × 1 σxx ; τxy ;
3
Sometimes this result τij = τji is stated as: ‘shear and complementary shear are equal’.
4
Physically it is the same procedure of computing the moments; mathematically the method used is different.
5-7 More About Stresses

face OA: area: dx × 1 σyy ; τyx ;

∂σxx ∂τxy
face AB: area: dy × 1 σxx + dx; τxy + dx;
∂x ∂x
∂σyy ∂τyx
face BC: area: dx × 1 σyy + dy; τyx + dy.
∂y ∂y

These are all surface forces. Additionally there are, in general, body forces. If these are Fx
and Fy per unit area5 in the x- and y- directions, respectively. With this information, we
can write down the equations of equilibrium in the x- and y-directions. They are:
  
∂σxx
x-direction: σxx + dx (dy × 1) − σxx (dy × 1) +
∂x
  
∂σyx
σyx + dy (dx × 1) − τyx (dx × 1) + Fx (dx × dy × 1) = 0;
∂y
  
∂σyy
y-direction: σyy + dy (dx × 1) − σyy (dx × 1) +
∂y
  
∂τxy
τxy + dx (dy × 1) − τxy (dy × 1) + Fy (dx × dy × 1) = 0.
∂x

Cleaning these up, we obtain the differential equations of equilibrium. They are honoured
by placing them below in an enclosure.

Differential equations of equilibrium:

∂σxx ∂τyx
+ + Fx = 0; (5.6a)
∂x ∂y
∂τxy ∂σyy
+ + Fy = 0. (5.6b)
∂x ∂y

In index notation these are written as

σji,j + Fi = 0 which is the same as σij,j + Fi = 0, (i = 1, 2). (5.7)

The differential equations of equilibrium may be presented in matrix form as

" # σ  
∂ ∂
0 ∂y  xx 
  
∂x Fx 0
∂ ∂ σyy + = . (5.8)
0 ∂y ∂x
Fy 0
τxy
 

We may write this in the form

LT τ + F = 0 , (5.9)
5
per unit area is the same as per unit volume because the thickness is 1.
Advanced Mechanics of Solids 5-8

where L is a linear (differential) operator operating here on the stress matrix τ written6
slightly differently as
{ττ } = [σxx σyy τxy ]T (5.10)

A Little More Rigorous Derivation

(a) A more rigorous analysis (b) A simplified analysis

Figure 5.5: The stress components acting on a rectangular block are shown. The variations
along both x− and y−directions are taken into account [Fig. 5.5a].

Figure 5.6: To show that a simplified analysis is sufficient to obtain the correct equations

Here we shall derive the same equations of equilibrium a little more rigorously. If (σxx )O
is the stress component σxx at the point O, there is a variation of this component along OA
and also along OC [Fig. 5.5a]. Thus,

∂(σxx )O
(σxx )A = (σxx )O + dx;
 ∂x 
∂σxx
(σxx )C = (σxx )O + dy.
∂y O

6
The stress matrix τ is written here as a column matrix. It is displayed here as the transpose of a row
matrix to save space. A notation {ττ } different from [σ
σ ] (which is a 3 × 3 matrix) is used here.
5-9 More About Stresses

As the distances OA, OC, etc. are small — we are considering an elemental rectangle which
would shrink to a point in the limit — the variations may be assumed to be linear as shown
in Fig. 5.5a. Similarly, considering the variation of σxx along AB,

∂(σxx )A
(σxx )B = (σxx )A + dy
∂y
 
∂(σxx )O ∂ ∂(σxx )O
= (σxx )O + dx + (σxx )O + dx dy
∂x ∂y ∂x
∂(σxx )O ∂(σxx )O ∂ 2 (σxx )O
= (σxx )O + dx + dy + dx dy. (5.11)
∂x ∂y ∂x ∂y
The net force in the x-direction corresponding to this stress component σxx is
1 1
net force = [(σxx )B + (σxx )A ] × dy × 1 − [(σxx )C + (σxx )O ] × dy × 1
2  2
∂ 2 (σxx )O

1 ∂(σxx )O ∂(σxx )O
= (σxx )O + dx + dy + dx dy +
2 ∂x ∂y ∂x ∂y
    
1 ∂(σxx )O 1 ∂(σxx )O
(σxx )O + dx dy − (σxx )O + dy + (σxx )O dy
2 ∂x 2 ∂y
∂(σxx )O ∂ 2 (σxx )O
= dx dy + dx dy. (5.12)
∂x ∂x ∂y

For the simplified case [Fig. 5.5b] we have

   
∂σxx
net force = σxx + dx  − σxx  dy
 
∂x AB OC
∂σxx
= dx dy. (5.13)
∂x
A comparison of Eqs (5.12) and (5.13) shows that the result is the same except for a term of
a higher order of smallness which can be neglected. We are looking for the limiting values
as the rectangle is made smaller and smaller until it degenerates to a point. The conclusion
is this: it is quite sufficient to consider the simplified case represented by Fig. 5.5a. The
same conclusion will be arrived at when we consider the net force in the x-direction because
of the shear stresses. We have here as before two cases, the more rigorous one allowing for
the variation of the shear stress components τxy and τyx [Fig. 5.7a], and the simplified case
represented by Fig. 5.7b. The conclusion is the same.
The net force in the x-direction associated with the shear stress τyx [Fig. 5.7a] is

1 1
[(τyx )B + (τyx )C ] dy × 1 − [(τyx )B + (τyx )C ] dy × 1,
2 2
where we write as before
∂(τyx )O
(τyx )A = (τyx )O + dx;
∂x
Advanced Mechanics of Solids 5-10

(a) More rigorous analysis (b) A simplified analysis

Figure 5.7: The shear stress components acting on a rectangular block are shown in Fig.
5.7a where the variations along both x− and y−directions are taken into account. In Fig.
5.7b a simplified case is shown. By comparing the two cases we can conclude that the
simplified case is sufficient to obtain the equations of equilibrium correctly.

∂(τyx )C
(τyx )B = (τyx )C + dx,
∂x
etc. We assume, again as before, that the variations are linear. The net force in the x-
direction associated with the shear stress τyx using the simplified stress distribution [Fig.
5.7b] is  
∂τyx
τyx + dy dx × 1 − [τyx ] dx × 1. (5.14)
∂y
A detailed calculation and comparison with Eq. (5.14) lead us to the same conclusion: a
simplified calculation is sufficient! Obviously, similar conclusions will be arrived at in the
case of the net force in the y-direction also.
When we calculate the moments, the normal stresses also contribute to the moments.
However, if we make detailed calculations, we can convince ourselves that they really do
not have to be reckoned for the same reason: quantities of a higher order of smallness can
be left out. Fig. 5.6 also leads us to the same conclusion.
One-dimensional Case: An Axial Bar
We shall now take up the simplest case, perhaps, of a one-dimensional body. A (one-
dimensional) bar [Fig. 5.8a] is subjected to an axially distributed load F . We desire to
obtain the differential equation of equilibrium.
Let us consider the various forces on the elemental block. The stress components are
σ on the left end at x = 0, and σ + (dσ/dx) dx. The area of cross-section A is variable:
A = A(x). The force components acting on the end cross-sections are marked in Fig. 5.8b.
In addition there is an applied force F per unit length which is variable and distributed
along the length. The equation of equilibrium is, thus,
 
d(σ A)
−(σ A) + (σ A) + dx + F dx = 0;
dx
5-11 More About Stresses

(a) An axial bar with a distributed load (b) Force components on an element

Figure 5.8: To derive the equations of equilibrium for an axial bar

d
i.e., (σ A) + F = 0, (0 < x < l). (5.15)
dx
[This differential equation of equilibrium is in terms of the stress components. This is the
natural approach. However, there are occasions when it is desirable to recast this equation
in terms of the displacements. We shall explain this later. (See the section on displacement
formulation on p. 12-13.) We shall see this problem again on p. 12-22 where this differential
equation is rewritten in terms of the (unknown) displacement u = u(x). At that time our
focus will be on writing the boundary conditions in terms of the displacement u and its
derivatives. We shall see that this is a mixed boundary value problem in the sense that one
boundary condition is on the displacement u, while the other is on du/dx. This is to be
taken notice of; the techniques used may be different for such situations.]
Equations of Equilibrium in Other Coordinate Systems
It is often convenient, and sometimes even necessary, to have the equations of equilibrium7
written in other coordinate systems. The reasons are that (i) sometimes the (boundary
value) problems are easier to formulate, and (ii) sometimes the boundary conditions are
far easier to write in one coordinate system than in another mainly because of the geom-
etry of the problem. Besides, even more importantly, some problems when formulated in
one coordinate system can be easily solved, while they can be incredibly difficult, or even
impossible, in another8 . An example is, say, the problem of determining the stresses in a
thick cylinder subjected to internal fluid pressure. The geometry makes it convenient —
well, almost essential — to choose polar coordinates in order to exploit the inherent circular
symmetry (axisymmetry) of the problem. Thus, there is a strong case to use coordinate
systems other than the simple rectangular coordinate system that we have used so far.
In addition to the above factors, there is the natural academic curiosity of examining
the possibility of formulation and solution of problems in a more general setting. For these
reasons, it is extremely important to consider other coordinate systems also. However,
there are several complications when a general (curvilinear) coordinate system is used. In
7
and, indeed, the other governing equations such as (i) the strain-displacement relations, (ii) the constitutive
equations
8
Readers familiar with the dynamics of rigid bodies may recall how easy the problem of torque-free motion
of a gyroscope is to solve when Euler angles are used, and how difficult it is otherwise.
Advanced Mechanics of Solids 5-12

spite of the complications, it seems appropriate to give here an elementary derivation of the
differential equations of equilibrium in cylindrical and polar coordinates. However, before
we undertake this task, we need to have a glimpse of some of the complications in coordinate
systems other than the rectangular cartesian coordinates.
Governing Equations in Other Coordinate Systems
Let us say we desire to obtain the differential equations of equilibrium in a cylindrical co-
ordinate system. One way is to draw an elemental block, mark all the stress components
acting on the various faces and the components of the body force, and consider the equi-
librium of the block. The necessary condition is that the net resultant force acting in any
direction — more specifically, in the r, θ, z (radial, tangential, and axial directions, respec-
tively) directions — should vanish. After writing these equations, we proceed to the limit
as the element is made smaller and smaller, and we obtain the required equation.
A second method is to change the equation from rectangular cartesian to cylindrical
polar coordinates by a mathematical transformation changing the independent variables
from (x, y, z) to (r, θ, z). This is not difficult, but we would not have the physical feel
of the various stress components and their directions, etc. However, this method is quite
effective in spite of the various mathematical manipulations involved.
A third method is by using general tensors, but it is too difficult for us now.
Equations of Equilibrium in Cylindrical Coordinate Systems

(a) A cylindrical polar coordinate system (b) An elemental block with stress components

Figure 5.9: To derive the equations of equilibrium in cylindrical coordinates

Figs 5.9, 5.10 show an elemental block in a cylindrical polar coordinate system. The
various stress components and the corresponding areas are listed below. As usual, we
calculate the net force acting in each of the r, θ, z directions, and set it equal to zero to
obtain the equations of equilibrium in the radial (r), tangential (θ) and axial (z) directions.
The curved surfaces (ABF E, DCGH) are in the vertical plane. (They do not appear
to be so in the figure shown.) The various faces, their areas, and the stress components
concerned are shown below. We can calculate the forces concerned and write down the
5-13 More About Stresses

(a) An elemental block (b) An elemental block with its faces

Figure 5.10: The faces of an elemental block and the stress components are shown. The
edges (BF, CG, AE, DH) are vertical, and BC, F G, AD, EH) horizontal.

(three) equations of equilibrium in the radial (r), tangential θ and axial (z) directions.
∂σrr
AEF B r dθ dz σrr DHGC (r + dr) dθ dz σrr + dr
∂r
∂τrθ
AEF B r dθ dz τrθ DHGC (r + dr) dθ dz τrθ + dr
∂r
∂τrz
AEF B r dθ dz τrz DHGC (r + dr) dθ dz τrz + dr
∂r
∂σθθ
BF GC dr dz σθθ AEHD dr dz σθθ + dθ
∂θ
∂τθr
BF GC dr dz τθr AEHD dr dz τθr + dθ
∂θ
∂τθz
BF GC dr dz τθz AEHD dr dz τθz + dθ
∂θ
∂σzz
EF GH r dr dθ σzz ABCD r dr dθ σzz + dz
∂z
∂τzr
EF GH r dr dθ τzr ABCD r dr dθ τzr + dz
∂z
∂τzθ
EF GH r dr dθ τzθ ABCD r dr dθ τzθ + dz.
∂z
These are the various stress components acting on the various faces of the block. These
are all acting on surfaces and they contribute to surface forces. In addition, there can be
body forces acting on the volume of the block. If Fr , Fθ , Fz are the body forces per unit
volume in the radial (r), tangential (θ) and axial (z) directions, they lead to the forces
Fr dV = Fr r dr dθ dz; Fθ dV = Fθ r dr dθ dz; Fz dV = Fz r dr dθ dz in the r, θ, z directions,
respectively. These too have to be reckoned.
∂σθθ
Let us note that the tangential stresses σθθ and σθθ + dθ are not parallel; there is
∂θ
a small angle dθ between them. This means that there is a contribution from these stress
components in the radial direction (radially inwards, towards the centre).
Advanced Mechanics of Solids 5-14

If these equations are written (noting the correct direction of each stress component) and
cleaned up which involves some algebraic manipulations, we obtain the (linear differential)
equations

∂σrr 1 ∂τθr ∂τzr σrr − σθθ

+ + + + Fr = 0; (5.16a)
∂r r ∂θ ∂z r
∂τrθ 1 ∂σθθ ∂τzθ 2τrθ
+ + + + Fθ = 0; (5.16b)
∂r r ∂θ ∂z r
∂τrz 1 ∂τθz ∂σzz τrz
+ + + + Fz = 0. (5.16c)
∂r r ∂θ ∂z r

Polar Coordinates
These equations are simplified when (two-dimensional) polar coordinates (r, θ) are used.
Obviously, we will have two equations of equilibrium. The derivation is along the same
lines; there is much simplification. The final equations are shown below.

∂σrr 1 ∂τrθ σrr − σθθ

+ + + Fr = 0; (5.17a)
∂r r ∂θ r
1 ∂σθθ ∂τrθ 2 τrθ
+ + + Fθ = 0. (5.17b)
r ∂θ ∂r r

The Simplified Axisymmetric Case

There are important cases where there is much simplification because of axisymmetry (also
called rotational symmetry). A cylinder, thin or thick, subjected to a fluid pressure9 and a
rotating disc are two examples. All the stress components acting on the element are shown

(a) Deformations (b) Stress components

Figure 5.11: To explain the simplified case of an axisymmetric problem

9
This problem is relevant even when there is no fluid at all! If we wish to make interference fit calculations,
the stresses are calculated using the well known Lamé’s equations for the stresses in a thick cylinder
subjected to fluid pressure. Some of the equations of fluid mechanics become relevant even when there is
no fluid. The techniques used in fluid mechanics are applicable in several other situations quite unrelated
to fluids. This is so in the case of most engineering science courses. This is the reason why a rigorous
study of engineering science courses is emphasised.
5-15 More About Stresses

here. Because of the axisymmetry10 , there is no equation of equilibrium in the tangential

direction11 . The only equation of equilibrium is in the radial direction. The relevant stress
components in the r-direction are (the thickness t is taken to be 1):

dσrr
AB r dθ × 1 σrr CD (r + dr) dθ × 1 σrr + dr
dr
BC dr × 1 σθθ AD dr × 1 σθθ

Because of axisymmetry, BC and AD and, hence, AB and CD are principal planes; there
are no shear stresses on them. Furthermore, all the variables of interest are independent of
θ; the derivatives are ordinary, and not partial.
The tangential stress components are not quite tangential; they have a component in
the radial direction, because of the small angle dθ between the edges BC and AD. The
radial force component is

dθ
2 × σθθ × dr × 1 × = σθθ dr dθ radially inwards.
2
Additionally, there can be a body force component Fr per unit volume12 in the r-direction.
Collecting all these terms, the (only) equation of equilibrium (in the r-direction) is
 
dσrr
σrr + dr (r + dr) dθ − σrr r dθ − σθθ dr dθ + Fr r dr dθ × 1 = 0;
dr
dσrr σrr − σθθ
i.e., + + Fr = 0. (5.18)
dr r

Other Cases
There are several more cases of interest and practical use in stress analysis problems; we
cannot obviously discuss all of them here.
Concluding Remarks
We have presented some information about one prime requirement, viz., the (differential)
equations of equilibrium shall be satisfied. The same requirement may be met by demanding
that the principle of virtual work shall hold. We can show that the principle of virtual work
is both a necessary and sufficient condition for equilibrium.
10
rotational symmetry about the axis passing through O perpendicular to the plane of the paper. Ax-
isymmetry leads to considerable simplifications, because all the variables will now be independent of the
independent variable θ.
11
What is meant is not that there is no equation of equilibrium, but that it is a trivial one. This equation
is a correct, but useless equation being of the form: [something] = [the same something].
12
Is this per unit volume, or is it per unit area? The thickness being 1, do these two come to the same thing?
Let us examine and settle the issue, and not accept this statement. Let us learn to question. Questioning
is not attacking or challenging. It is an expression of an open mind willing to explore other possibilities.
Gullibility is not a virtue!
Advanced Mechanics of Solids 5-16

The differential equations of equilibrium are, traditionally and almost always, written
in terms of the stress components. However, it is possible to write these in terms of strain
components (using the stress-strain relations), and / or in terms of the displacement com-
ponents using the strain-displacement relations. One of the motivations for doing so is to
express all the unknowns in terms of, say, the strain components or, better still, in terms
of displacements. There may be some convenience. Some problems can be solved by this
approach. However, we must realise that the equations, though expressed in terms of dis-
placements, are still equilibrium equations. Furthermore, inasmuch as we have used the
constitutive equations, the resulting equations of equilibrium are now restricted to elastic
materials only!
We shall see that the governing equations are generally written in terms of different
(unknown, dependent) variables: (i) the equations of equilibrium in terms of the stress
components σij , (i, j = x, y, z); (ii) the strain-displacement relations in terms of the strain
and displacement components eij , (i, j = x, y, z); ui (i = 1, 2, 3) (i.e., u, v, w); and the
constitutive equation in terms of σij and eij (i, j = x, y, z). This situation is admittedly
inconvenient. Can we recast them all in terms of the same unknowns, the displacement
components u, v, w? Sure enough, this can be done. We shall do this later [p. 12-13].
SOME CONCEPTS RELATED TO 2-D STRESS FIELDS
There are some important terms and concepts related to two-dimensional stress fields —
not just at a point — that have special relevance and significance in photoelasticity. Ex-
perimental stress analysis is a separate discipline in own right13 . It is not possible to deal
with it here. Nevertheless we shall discuss briefly a few terms. There are many aspects
to be discussed in detail before we can appreciate, or even understand, the significance of
these technical terms. We shall use p and q to refer to the two principal stresses just for
this section only.
Experimental Stress Analysis: Photoelasticity
Most of the methods of experimental stress analysis are based on measurement of strain.
Photoelasticity is an exception where stress, and not strain, is directly measured (or calcu-
lated from experiments). The following has special relevance for photoelasticity.
Dependence of the Stresses on the Material Properties
We use photoelasticity to determine the stresses in actual structures like a dam and machine
parts. This is essentially a two-dimensional analysis (although three-dimensional photoelas-
tic analysis also is possible). Behind this procedure, there lies a fundamental question: are

13
Photoelasticity is an important part of experimental stress analysis. The method is based on the phe-
nomenon of double refraction. When polarised light passes through a stressed material, a phase angle is
introduced between the two rays, called the ordinary and the extraordinary. Thus, fringes are formed.
These fringes are obtained experimentally using a photoelastic bench and polarised light. Fig. 5.12 shows
two fringe patters and two special curves of special importance in photoelasticity. Students are urged
to be familiar with this extremely useful topic. Most colleges have facilities at least as demonstration
experiments in photoelasticity.
5-17 More About Stresses

(a) Some important curves (b) A fringe pattern (c) Another fringe pattern

Figure 5.12: Photoelastic fringes and curves related to photoelasticity

the stresses in an actual structure or a machine part the same as those in the photoelastic
model? This question leads us to the examination of an even more fundamental question:
are the stresses in a body independent of the material? Let us examine.
Let us consider a few specific cases and examine if the stresses are dependent only on
the applied loads but independent of the elastic properties of the material.

(i) A uniform bar loaded by a pair of (tensile) loads P − P :

The (tensile) stress is given by σ = P/A which is clearly independent of the material
(constitutive) properties.
(ii) A uniform circular shaft loaded by a pair of twisting moments T − T :
The (shear) stress is given by τ = T r/J which is again independent of the material.
(iii) A uniform beam in pure bending loaded by M − M :
The bending stress is given by σ = M y/I which is independent of the material.
(iv) Lamé’s thick cylinder problem: a thick cylinder subjected to internal pressure
The stresses are given by σrr = A + B/r2 and σθθ = A − B/r2 (independent of the
material).
With these examples in mind, we may be tempted to conclude that the stresses depend
only on the load applied and never on the elastic (constitutive) properties. But wait a
minute. Let us not jump into hasty conclusions. Let us consider two more problems.
(v) Rotating discs: the stresses now depend on the elastic properties!
B 3+ν
σrr = A + 2
− ρω 2 r2 ,
r 8
B 1 + 3ν
σθθ =A− 2 − ρω 2 r2 ,
r 8
(vi) The free end of a uniform (horizontal) cantilever lifted by a specified amount of dis-
placement, δ
3 EI
Now the load P needed to bring about the end deflection = 3 δ. The stresses are
l
dependent on the material properties!
Advanced Mechanics of Solids 5-18

From the examination of these few cases, what can we understand? In all the problems
(i) - (iv), the displacements do depend on the elastic properties. In (v), even the stresses
depend on the constitutive properties. Problem (v) is the only one where a body force is
present. What does all this lead us to?
Well, the general problem is very difficult to tackle. We will not attempt to attack the
problem in all its generality. Instead we shall merely obtain a partial answer applicable to
two-dimensional (plane stress and plane strain) problems.
The differential equations of equilibrium and the traction boundary conditions are:

∂σxx ∂τyx ∂τxy ∂σyy

+ + Fx = 0 + + Fy = 0; (5.19)
∂x ∂y ∂x ∂y
l σxx + m τyx = pνx l τxy + m σyy = pνy . (5.20)

Now there are three unknowns (the three stress components), but only two equations. We
need an additional equation which is the compatibility equation

∂ 2 exx ∂ 2 eyy ∂ 2 γxy

+ = . (5.21)
∂y 2 ∂x2 ∂x ∂y

The constitutive equations for the two cases of plane stress and plane strain are:
    
 exx  1 −ν 0 σxx 
1  
eyy = −ν 1 0  σyy (plane stress); (5.22)
E
exy 0 0 (1 + ν) τxy
   
     
 exx  1 + ν 1 − ν −ν 0  σxx 
eyy =  −ν 1 − ν 0  σyy (plane strain); (5.23)
E
exy 0 0 1 τxy
   

Substituting for the stress components in Eq. (5.21) using the constitutive equations Eq.
(5.22, plane stress), we obtain

1 ∂ 2 σxx ∂ 2 σyy ∂ 2 σyy ∂ 2 σxx 2(1 + ν) ∂ 2 τxy

 
− ν + − ν = . (5.24)
E ∂y 2 ∂y 2 ∂x2 ∂x2 E ∂x ∂y

We now desire to get rid of τxy in this equation. Towards this end, let us differentiate Eq.
(5.19) w.r.to x and Eq. (5.20) w.r.to y and add to yield

∂ 2 σxx ∂ 2 σyy ∂Fx ∂Fy ∂ 2 τxy

+ + + = −2 .
∂x2 ∂y 2 ∂x ∂y ∂x ∂y

From here if we substitute for τxy on the right hand side of Eq. (5.24), we obtain on
simplification
 
2 ∂Fx ∂Fy
5 (σxx + σyy ) = −(1 + ν) + (plane stress). (5.25)
∂x ∂y
5-19 More About Stresses

In the same way we can obtain for the case of plane strain
 
2 1 ∂Fx ∂Fy
5 (σxx + σyy ) = − + (plane strain). (5.26)
(1 − ν) ∂x ∂y

If the body forces are absent or constant, the right hand sides of Eqs (5.25, 5.26) vanish and
we obtain an important result: the sum of the principal stresses (σ11 + σ22 ) = (σxx + σyy )
satisfy the Laplace’s equation
52 (σxx + σyy ) = 0.

Now the equations of equilibrium, the traction boundary conditions, and the compatibility
conditions are all independent of the elastic properties. Thus, in conclusion we note that
under the afore-mentioned conditions, the stresses inside the body depend only on the
loading and not on the elastic properties of the material.
A similar result holds for the three-dimensional case also. However, the general problem
is far more complex; we do not discuss this at all.
Isochromatics
These are the lines along which the maximum shear stress has the same value. This maxi-
mum value will change from fringe to fringe. These isochromatics are obtained experimen-
tally. [There are several details that must be understood before a photoelastic experimental
analysis can be successfully completed. For example, along with the isochromatics, there
will be isoclinics also. How to distinguish between these different curves is an important
aspect that must be addressed. Would it help if the magnitude of the applied load is slightly
increased or decreased without changing the pattern of loading? Would one set of lines, but
not the other, change? Would this technique help us to distinguish between the two sets?
We regret our inability to discuss all these aspects here.]
Isotropic Point
This is a special point where both the principal stresses are equal: p = q. The Mohr’s circle
degenerates to a point; every plane is a principal plane and every stress a principal stress.
There is no shear stress14 on any plane15 passing through the isotropic point.
Principal Stresses and Principal Directions
The principal stresses and their planes (directions) are given by
s 2
σxx + σyy σxx − σyy 2 ;
σ11 ≡ p = + + τxy (5.27a)
2 2

14
In photoelasticity, there are fringes which are formed depending on the maximum shear stress reaching
some specified values. At an isotropic point, the maximum shear stress is always zero. Consequently,
in a standard photoelastic image, there is a dark spot at an isotropic point. Readers not familiar with
photoelasticity are advised to read about it from some good books on Experimental Stress Analysis.
15
It must be understood clearly that on planes that are inclined to the xy-plane, there will be shear stresses.
Advanced Mechanics of Solids 5-20

Fig. 5.13 is a pictorial representation of the (2-

dimensional) state of stress at a point. The principal
stresses p ≡ σ11 and q ≡ σ22 and their directions are
marked in the figure. The angle θ defines the direction of
one of the principal stresses with respect to the x-axis.

Traditionally the angle θ is measured anticlockwise from

Figure 5.13: Stress
the x-axis to the nearer principal direction. The value of
components at a point
θ is never negative, and never larger than 90 ◦ .
Figure 5.14: Stress components at a point

s 2
σxx − σyy σxx − σyy 2 ;
σ22 ≡ q = + + τxy (5.27b)
2 2
2τxy
tan θ = . (5.27c)
(σxx − σyy )

The third principal stress is zero, because we are considering only the 2-d case now. The
angle θ often serves as a parameter which, in general, varies from point to point.
Isoclinics

(a) An arbitrary curve aaaa (b) An isoclinic bbbb (θ = 30 ◦ ) (c) A stress trajectory cccc

Figure 5.15: The first curve aaaa [Fig. 5.15a] is an arbitrary curve; naturally enough, the
principal directions will be different at different points. The second curve bbbb [Fig. 5.15b]
is an isoclinic of parameter θ = 30 ◦ ; one of the principal directions makes an angle of 30 ◦
with the x-axis. The third curve cccc [Fig. 5.15c] is a stress trajectory; its tangent is along
a principal direction.

Let us consider an arbitrary curve aaaa in the stress field [Fig. 5.15a]. The value of θ
will, naturally enough, vary from point to point. Now there arises the question: is there a
(continuous) curve along which θ has the same value, say, 30 ◦ ? The answer is yes. This
curve bbbb in the stress field [Fig. 5.15b] is called an isoclinic of parameter θ = 30 ◦ . In
other words, an isoclinic of parameter θ is a (continuous) curve in the stress field such that
at every point on it, one of the principal directions is always in the same direction as shown
5-21 More About Stresses

in Fig. 5.15b. It is obvious that, if θ is the parameter of an isoclinic, then θ + 90 ◦ also is

a parameter of the same isoclinic. Traditionally, the parameter is measured anticlockwise
from the x-axis to the nearer principal direction (of p or of q). The parameter is never
negative, nor is it ever larger than 90 ◦ . [There are many details that must be discussed
before we can understand the significance of these terms.]
Through each point in the stress field, there always exists a curve (like bbbb) along
which the angle θ (which means that both θp and θq are constant) remains constant. Thus,

∂θp ∂θp
= = 0 along an isoclinic like bbbb.
∂s ∂s

Furthermore, in general16 there is only set of principal directions at each point. Thus, there
can be only one isoclinic passing through a given point (as long as it is not an isotropic
point). At an isotropic point, however, the situation is different, because

1 0
σxx = σyy = p = q; τxy = 0. Thus, from Eq. (5.27c), θp, q = tan−1 (indeterminate).
2 0

We, therefore, conclude that there can be isoclinics of different parameters passing through
an isotropic point. Isoclinics can be obtained experimentally by photoelasticity techniques.
At a free boundary, the parameter of the isoclinic must match the known direction of the
principal stresses at a boundary point. There are several details which must be understood
by an experimental stress analyst if he wishes to conduct serious studies using photoe-
lasticity. White light is generally used to improve the quality of the isoclinics obtained
experimentally.
Isoclinics are not easy to obtain analytically, but they can be obtained experimentally
quite easily from photoelastic measurements. They are usually drawn (sketched or obtained)
at regular intervals of 10 ◦ or 5 ◦ . Isoclinics of different parameters can converge at the point
of application of a concentrated load as at an isotropic point. [There are several details that
must be understood. Furthermore, in recent years the face of photoelasticity has changed
greatly, almost out of recognition, from the days of Coker and Filon17 .]
Stress Trajectory
Stress trajectories are (continuously differentiable) curves like cccc [Fig. 5.15c] such that
the tangent at any point to the curve is along one of the principal directions. They are
also known as the principal stress trajectories and isostatics. Written along such stress
trajectories are the Lamé-Maxwell equations of equilibrium. We do not discuss these here.
We shall now discuss how a better understanding of the nature of stress at a point can
be had by geometric visualisation.

16
As an isotropic point there can be several; in fact, every direction (in the xy-plane) is a principal direction.
17
The early investigators on photoelasticity, Ernest George Coker (April 1869 - April 1946), British mathe-
matician and engineer (a noted expert of stress analysis and photoelasticity) and Louis Napolean George
Filon (Nov. 1875 - Dec. 1937), a French-born English applied mathematician
Advanced Mechanics of Solids 5-22

STRESS AT A POINT: GEOMETRIC VISUALISATION

The nature of stress at a point can be abstract and difficult to understand. It pays to have
visual support. Mohr’s circle, Cauchy’s stress quadric, and Lamé’s ellipsoid of stress18 are
widely used19 . Their value is not as computational tools; nobody uses graphical methods to
obtain numerical values. Their primary use is to help us understand the qualitative features
clearly. Seeing is believing. A picture is far more appealing and effective than elaborate
explanations. With this introduction we shall first see Lamé’s stress ellipsoid.
LAMÉ’S ELLIPSOID OF STRESSES

(ν)
(a) The stress resultant T on a plane (b) The tips of the stress resultants

Figure 5.16: The tips of the stress resultants on the various planes fall on an ellipsoid.

We know that the stress vector on different planes passing through a point are, in
general, different in both magnitude and direction. If the tips of these various stress vectors
on various planes are joined together, we obtain a closed surface. This, in general, will be
the surface of an ellipsoid known as Lamé’s ellipsoid of stresses [See Fig. 5.16b.]. Let us
first orient the x, y, z axes along the principal axes 1, 2, 3. Then the equations expressing
the normal stresses on an inclined plane (whose normal ν has the direction cosines l, m, n)
[Eqs (4.2a), (4.2b), (4.2c), p. 4-12 reproduced here] are simplified as shown below.
(ν) (ν)
Tx ≡ pνx = σxx l + σyx m + τzx n −→ Tx ≡ pνx = l σ11 (5.28a)
(ν) (ν)
Ty ≡ pνy = τxy l + σyy m + τzy n −→ Ty ≡ pνy = m σ22 (5.28b)
(ν) (ν)
Tz ≡ pνz = τxz l + τyz m + σzz n −→ Tz ≡ pνz = n σzz (5.28c)

But we know that the direction cosines (l, m, n) [Eqs (5.28a), (5.28b), (5.28c)] should satisfy
the condition
pνx 2 pνy 2 pνz 2
     
2 2 2
l +m +n =1 −→ + + = 1. (5.29)
σ11 σ22 σ33
18
Named after the German engineer, C.O. Mohr; the French mathematician (also a civil engineer), A.L.
Cauchy; and the French mathematician and elastician, G. Lamé
19
The influence of the book [5] is gratefully acknowledged.
5-23 More About Stresses

This, we can see, is an ellipsoid with σ11 , σ22 , σ33 as its semi-axes. If we consider various
planes passing through the point O, and mark off the lengths OP1 , OP2 , · · · , OPn , etc. the
tips of these lengths fall on this ellipsoid. Let OP be one such line. The locus of P for
various planes is the stress ellipsoid.
If we desire, the equation to the ellipsoid may be written in the more familiar form

x2 y2 z2
2 + 2 + 2 = 1.
σ11 σ22 σ33

If the coordinates of the tip P of a typical line OP are (x, y, z), the lengths Ox, Oy, Oz
are the projections of the length OP which represents the magnitude of the resultant stress
vector on some plane. [This plane is not known just at present; some auxiliary construction
is needed to locate it. We shall see this later.] The projections of the resultant stress vector
(on this unknown plane) are the stress components in the coordinate directions. Thus,

(ν) (ν) (ν)

x = Tx = pνx ; y = Ty = pνy ; z = Tz = pνz .

Now Eq. (5.29) appears in the form

x2 y2 z2
2 + 2 + 2 = 1. (5.30)
σ11 σ22 σ33

(ν)
We do not have to know this plane ν on which this resultant stress vector T acts. But if
we so desire, we can locate it by an auxiliary construction.

(a) The stress ellipsoid and a radius (c) The stress ellipsoid and a tangent
vector P1 (b) A typical plane plane

Figure 5.17: In Fig. 5.17a the radius vector OP1 represents the resultant stress vector p(ν)
on an unidentified plane. In Fig. 5.17c it is OQ (and not OP ) that represents the stress
resultant. The tangent plane at P is shown. Fig. 5.17b shows a typical plane, its normal,
a typical radius vector, and the stress resultant on this plane.

Features of Lamé’s Stress Ellipsoid

In connection with the Lamé’s stress ellipsoid, let us note the following.
Advanced Mechanics of Solids 5-24

(i) The stress ellipsoid is visualised and drawn in a Westergaard stress space. [See the
various figures of Lamé’s stress ellipsoid in this subsection with the axes along the
principal axes σ11 , σ22 , σ33 .]

(ii) The tips of the resultant stress vectors on the various planes passing through the
point fall on the surface of this ellipsoid [Fig. 5.16b]. However, as indicated earlier,
we cannot quickly know the plane on which the resultant stress vector, say, OP acts.

(iii) The resultant stress vector acts along the normal of the plane — that is, the normal
component of the stress resultant is all we have; there are no shear components —
only when the plane is a principal plane. [The point Q defined above in the auxiliary
construction falls on the surface of the ellipsoid only at six points in general. In
general, there are only three principal planes and principal stresses.]

(iv) The first invariant I1 is proportional to the sum of the principal radii of the ellipsoid.
This is easy to understand; the semi-axes are the three principal stresses σ11 , σ11 , σ11 .

(v) The second invariant I2 is proportional to the three principal areas of the ellipsoid.
(On each plane the projection of the ellipsoid gives an ellipse; the areas of these three
ellipses are called the principal areas of the ellipsoid.)

(vi) The third invariant I3 is proportional to the volume of the ellipsoid.

(vii) The stress ellipsoid remains the same no matter what the coordinate system is. If the
coordinate system is chosen along the principal directions, the equation to the ellipsoid
contains no cross-terms like xy, yz, zx, but if the coordinate system is chosen along
non-principal axes [Fig. 5.18], the equation will contain some cross-term like xy.
To understand this clearly, let us consider a quadratic form 4u2 + 9v 2 . The equation
4u2 + 9v 2 = 36 clearly represents an ellipse with its semi-axes 3 and 2.
 u 2  v 2
4u2 + 9v 2 = 36 −→ + = 1.
3 2
If the variables u, v are changed to x, y by a linear transformation, we obtain the
equation of the same ellipse in terms of the new variables x, y. If u = c11 x + c12 y; v =
c21 x + c22 y, the above equation to the ellipse will now have the appearance

4 (c11 x + c12 y)2 + 9 (c21 x + c22 y)2 = 36. (5.31)

This equation contains a term in xy. What this example demonstrates is the following.
As a special case, when the new axes x, y are oriented at 45◦ to the old u, v axes, the
associated rotation matrix is
  " √1 √1
#
sin θ cos θ 2 2
= .
− cos θ sin θ − √12 √12
5-25 More About Stresses

We know that this rotation matrix is an orthogonal matrix; that is, (i) its determinant
is equal to 1, and (ii) its inverse is the same as its transpose. Thus, the equation (5.31)
to the ellipse now appears in terms of x, y as
13 2 13 2
x + y − 5xy = 36. (5.32)
2 2
The ellipse referred to is the same, but now it is referred to a set of non-principal axes
x, y. The equation has a cross-term −5xy. [See Figs 5.18a, 5.18b.]

(a) Lamé’s stress ellipse referred to the principal (b) The same stress ellipse referred to a set of
axes u, v non-principal axes x, y

Figure 5.18: Lamé’s stress ellipse referred to principal and non-principal axes. With respect
to a principal set of axes u, v [Fig. 5.18a], there is no cross-term uv in its equation but w.r.to
a set of non-principal axes like x, y [Fig. 5.18b], the equation contains a cross-term xy.

(viii) The quadratic forms and the matrices with and without the cross-term, are
  
2 2
 6.5 −2.5 x
cross-term −5xy, 6.5x + 6.5y − 5xy= x y (5.33a)
−2.5 6.5 y
  
4 0 u
no cross-term uv, 4u2 + 9v 2

= u v (5.33b)
0 9 v

When there is a cross-term as in Eq. (5.33a), the matrix is non-diagonal. If there is

no cross-term as in Eq. (5.33b), the corresponding matrix is a diagonal one. Thus, to
remove the cross-term(s) in the appearance of a quadratic form, the (square) matrix
associated with the quadratic form is to be diagonalised!
[Why should we remove the cross-term(s) in a quadratic form? Well, there are several
reasons. One of them is to check, or know for sure, if the quadratic form is positive
definite or not, so necessary in the discussion of stability of dynamical systems. It is
the presence of the cross-terms that makes it difficult to decide this. If the associated
(square) matrix is rendered into its diagonal canonical form, the issue is settled. The
diagonal elements are the eigenvalues, and we can conclude that a quadratic form is
positive definite if all the eigenvalues of its associated matrix are positive.]

(ix) If the three principal stresses are all equal, i.e., if σ11 = σ22 = σ33 , it is obvious
that the ellipsoid becomes a sphere [Fig. 5.20c]. Now every plane is a principal
plane; there is no shear stress on any plane. This represents a state of hydrostatic
Advanced Mechanics of Solids 5-26

stress (appropriately called a case of spherical stress tensor). The stress matrix now
represents an isotropic state of stress.

(b) Lamé’s stress ellipse of revolution when

(a) Lamé’s stress ellipsoid when σ11 6= σ22 6= σ33 σ11 6= σ22 = σ33

Figure 5.19: When two principal stresses are equal, Lamé’s stress ellipsoid degenerates into
an ellipse of revolution (in this case about the 11 axis) [Fig. 5.19b].

(x) If two principal stresses are equal, say, σ11 6= σ22 = σ33 , the stress ellipsoid degenerates
into an ellipse of revolution about the principal axis 11 [Fig. 5.19b]. All normal cross-
sections (normal to the 11 axis) are, thus, circles. One such circle projected on the
yz, i.e., 23 plane, is shown in this figure. Let us note that y, z; y 0 , z 0 ; y 00 , z 00 are all
principal planes. There are just three (3), no more and no less, principal planes for the
first case when the three principal stresses are distinct [Fig. 5.19a], infinite principal
planes in the second case [Fig. 5.19a]. Note that even though y and y 0 are both
principal planes, they are not normal (perpendicular) to each other. We recall from
linear algebra that the eigenvectors are guaranteed to be orthogonal to one another
only if the eigenvalues (principal stresses) are distinct. These facts must be understood
clearly.
We hope that the figures [Figs 5.20a, 5.20b, 5.20c] throw further light on what happens
when two of the principal stresses are equal. Among all the planes normal to the 23
plane, for example, the shaded plane [Fig. 5.21a] is the one on which the shear stress
reaches a maximum; its value is (σ11 − σ22 )/2.

(xi) Referring to the two figures 5.17a, 5.17b, we note that the point Q is quite different
from the point P , in general. The points Q and P coincide only in very special cases.
We know that the resultant stress vector on a plane ν is normal to it only when ν is
a principal plane.
Thus, Q touches the surface of the ellipsoid only in six places in general where the
surface cuts the coordinate axes. These, we know, correspond to the principal planes.

All these aids for geometric visualisation have their counterparts in a two-dimensional set-
ting when there are considerable simplifications. We shall review them briefly below.
5-27 More About Stresses

(a) Ellipse of revolution when σ11 = (b) Ellipse of revolution when (c) Sphere when σ11 = σ22 =
σ33 6= σ22 σ11 = σ22 6= σ33 σ33

Figure 5.20: As before when two principal stresses are equal, Lamé’s stress ellipsoid degen-
erates into an ellipse of revolution (in this case about the 22 axis) when σ11 = σ33 6= σ22 [Fig.
5.20a] and about the 33 axis when σ11 = σ22 6= σ33 [Fig. 5.20b]. When σ11 = σ22 = σ33
[Fig. 5.20c], the ellipsoid degenerates into a sphere.

Two-dimensional Case: Lamé’s Ellipse of Stresses

There are obviously simplifications when we consider the above topics in the more limited
context of two dimensions. Now among the nine (9) / six (6) stress components, there are
only four (4) / three (3), viz., σxx , σxy = σyx , σyy . In other words, σyz = σzy = 0, σzx =
σxz = 0, σzz = 0. This corresponds to the so-called plane stress simplification20 . Readers
are urged to refer to the earlier (more general) three-dimensional case repeatedly, and realise
that all the results and explanations given earlier are relevant here too.
Let us consider a general two-dimensional state of stress represented by the two principal
stresses σ11 and σ22 [Fig. 5.22a]. If the coordinate axes x, y are oriented along these principal
axes, we have
σxx = σ11 ; σyy = σ22 ; σxy = 0.

The normal and shearing stresses and the resultant stress on an inclined plane ν [Fig. 5.22b]
are given by

pνx = σ11 cos (ν, x) = l σ11 ; (5.34a)

20
Readers who are not exposed to these concepts are advised to consult any good book on the Theory
of Elasticity, say, Timoshenko & Goodier [16]. There are other simplifications such as plane strain and
generalised plane stress. If the plane stress problem is solved, the plane strain solution also (and vice
versa) can be obtained by making a few changes:
E ν
plane stress solution −→ plane strain solution: E −→ ; ν −→ ;
1 − ν2 1−ν
E(1 + 2ν) ν
plane strain solution −→ plane stress solution: E −→ ; ν −→ .
(1 + ν)2 1+ν
Advanced Mechanics of Solids 5-28

(a) Plane(s) normal to the 23 (b) Plane(s) normal to the 13 (c) Plane(s) normal to the 12
plane plane plane

Figure 5.21: Planes of maximum shear stress: the figures show the planes on which the
maximum shear stress occurs. The maximum values, we know, are (σ11 − σ22 )/2, (σ22 −
σ33 )/2, (σ33 −σ11 )/2. The maximum shear stress at the point is the grand maximum among
these three values.

(a) A two-dimensional state of stress (b) Resultant stress on an inclined plane

Figure 5.22: A two-dimensional state of stress is shown in Fig. 5.22a. The resultant stress
vector on an inclined plane ν is shown in Fig. 5.22b.

pνy = σ22 cos (ν, y) = m σ22 ; (5.34b)

q q
p(ν) = pνx + pνy = l2 σ11
2 2 2 + m2 σ 2 ; (5.34c)
22
−(σ11 − σ22 )
τ(ν) = −σ11 sin θ cos θ = sin 2θ. (5.34d)
2

[The last equation is a special case of Eq. (4.10b), p. 4-16, simplified for this context with
the direction cosines appropriately interpreted. It can also be obtained by considering the
equilibrium of a small rectangular block acted upon by the relevant stress components on
all its faces. When working in two dimensions it is often more convenient to use sin θ and
cos θ instead of the direction cosines.]
Let us consider the resultant stress vectors on all the inclined planes passing through
the point under consideration. The tips of all these resultant stress vectors will lie on an
ellipse. This is the Lamé’s stress ellipse [Fig. 5.23a]. [Compare with the explanation of
Lamé’s ellipsoid of stresses on p. 5-22.]
5-29 More About Stresses

To obtain the equation to the ellipse, we calculate l2 , m2 (that is, cos2 (ν, x), cos2 (ν, y))
from Eqs (5.34a, 5.34b) and use the result that l2 + m2 = 1 (that is, sin2 θ + cos2 θ = 1).
This gives us
p2νx p2νy
2 + σ 2 = 1.
σ11
(5.35)
22
Noting from Fig. 5.22b that the tip of the resultant stress vector p(ν) has the coordinates
(x, y) (that is, pνx = x and pνy = y), we obtain the equation to the ellipse [Fig. 5.23a] as
x2 y2
2 + 2 = 1. (5.36)
σ11 σ22

(a) Lamé’s ellipse of stresses (b) Ellipse: non-principal coordinates (c) Special case: circle

Figure 5.23: Lamé’s ellipse of stresses [Fig. 5.23a] referred to principal axes; there is no
‘cross-term’. The same ellipse when referred to non-principal axes appears as in Fig. 5.23b.
The ellipse is the same; it remains unchanged (invariant). Its equation will now contain
a ‘cross-term’. The ellipse becomes a circle when both the principal stresses are equal:
σ11 = σ22 (a two-dimensional state of hydrostatic stress).

There are two more special cases of the ellipse of stresses. One is when σ11 = −σ22
which is a state of pure shear. The first invariant I1 = σ11 + σ22 = 0. As |σ11 | = |σ22 |,
the stress ellipse is a circle. The magnitude of the resultant stress vector |p(ν) | is the same
on every plane (perpendicular to the xy plane). Let us not fail to note that we are now
discussing only two-dimensional states of stresses. The other special case is when σ11 alone
is present. Now σ22 = 0; the stress ellipse degenerates into a straight line of magnitude
2|σ11 |.
CAUCHY’S STRESS QUADRIC
Next, we shall discuss Cauchy’s stress quadric. We repeat that these geometrical con-
structions are never used for actual computation. Referring to a set of principal axes
(x, y, z) = (1, 2, 3) for convenience, the quadric surface

σ11 x2 + σ22 y 2 + σ33 z 2 = ±1 (5.37)

represents, in general, either an ellipsoid or a hyperboloid [Fig. 5.24]. The hyperboloid

may have one or two sheets. [If the reference axes are non-principal, the equation will have
Advanced Mechanics of Solids 5-30

(a) An ellipsoid (b) A hyperboloid of one or two sheets

Figure 5.24: Cauchy’s stress quadric: this is either an ellipsoid [Fig. 5.24a] or a hyperboloid
of one or two sheets [Fig. 5.24b] depending on whether the principal stresses are of like or
unlike signs. This is explained further in the text.

‘cross-terms’ like xy, yz and zx.] OP any radius vector from the origin O touching the
quadric surface at the point P (x, y, z) has the direction cosines

l = cos (ν, x); m = cos (ν, y); n = cos (ν, z).

Then the length OP is inversely proportional to the square root of the absolute value of
the normal stress |σνν |. This acts on a plane ν through O; that is, the normal to this plane
is along OP [direction cosines: l = cos (ν, x); m = cos (ν, y); n = cos (ν, z)]. We shall
prove this for a general point P . But let us first verify this for the special cases where P is
A, B, C, · · · where the principal axes cuts the coordinate axes (x, y, z) = (1, 2, 3).

1 1 1
OA = √ = OB; OC = √ = OD; OE = √ = OF.
σ11 σ22 σ33

If the coordinates of such a typical point P are (x, y, z), we can see that

x = l OP = OP cos (ν, x); y = m OP = OP cos (ν, y); z = n OP = OP cos (ν, z).

The equation (5.37) to the quadric surface now takes the form

σ11 (OP )2 cos2 (ν, x) + σ22 (OP )2 cos2 (ν, y) + σ33 (OP )2 cos2 (ν, z) = ±1;
     

1
that is, l2 σ11 + m2 σ22 + n2 σ33 = ± ; (5.38)
(OP )2
1
or, σ11 cos2 (ν, x) + σ22 cos2 (ν, y) + σ cos2 (ν, z) = ± . (5.39)
(OP )2

But we know — we have seen this — that

σνν = l2 σ11
2
+ m2 σ22
2
+ n2 σ33
2
= σ11 cos2 (ν, x) + σ22 cos2 (ν, y) + σ33 cos2 (ν, z). (5.40)
5-31 More About Stresses

A comparison of Eqs (5.39), (5.40) shows that

1 1
σνν = ± ; i.e., OP = p . (5.41)
(OP )2 |σνν |

Fig. 5.24a is an ellipsoid somewhat similar to the Lamé’s ellipsoid of stresses. But there are
important differences. In Lamé’s ellipsoid, a radius vector OP represents the magnitude
of the resultant stress vector. Its plane is not immediately apparent; it is to be found by
an auxiliary construction. Here, however, a radius vector OP represents the reciprocal of
the magnitude of the normal stress σνν ; its normal ν is also known. But here an auxiliary
construction is needed to determine the plane on which the resultant stress vector acts. The
other difference is that here the ellipsoid is the shortest along the x-axis p (1-axis) and
p the
longest along the z-axis (3-axis), because |σ11 | > |σ33 | and, therefore, 1/ |σ11 | < 1/ |σ33 |.
Fig. 5.24b shows the hyperboloid of (i) one sheet, and (ii) of two sheets. These corre-
spond, respectively, to

(i) one sheet: σ11 x2 + σ22 y 2 + σ33 z 2 = + 1 shaped like a cooling tower;
2 2 2
(ii) two sheets: σ11 x + σ22 y + σ33 z = − 1 shaped like two bowls.

The radius vectors OP1 , OP2 , · · · with their tips on the hyperboloid of one sheet represents
positive (tensile) normal stresses. On the other hand, OP3 , OP4 , · · · with their tips on
the hyperboloid of two sheets represent negative (compressive) stresses. If P falls on the
asymptotic cone, it is the in-between case. [One of these two asymptotic cones is shown in
Fig. 5.24b.] As Pn −→ ∞ (asymptotically), the normal stress −→ 0. As there are infinite
number of points on the asymptotic cone, P may fall on any one of the infinite generators
of this cone corresponding to the planes on which the normal stress is zero. [We recall that
the length of the radius vector
p OP is the reciprocal of the square root of the magnitude of
the normal stress: OP = 1/ |σνν |.]
The cross-section, shown shaded in Fig. 5.24b, of the hyperboloid of one sheet is an
ellipse whose equation — Eq. (5.37) in which z is set equal to 0 corresponding to the xy
plane — is
σ11 x2 + σ22 y 2 = +1.

[To have a better idea of the shape of these surfaces, let us take the help of maple. For
the numerical values chosen, the hyperboloid of one and two sheets are drawn in the same
figure and displayed in Figs 5.25a, 5.25b, 5.25c. It is the same figure looked at from three
different ‘perspectives’. The governing equations to draw these figures are

50x2 + 30y 2 − 10z 2 = 25 (−5 < x < 5; −5 < x < 5; −2 < x < 2);
2 2 2
50x + 30y − 10z = −25 (−5 < x < 5; −5 < x < 5; −3 < x < 3).]
Advanced Mechanics of Solids 5-32

(a) Stress quadric: one view (b) Stress quadric: a second view (c) Stress quadric: a third view

Figure 5.25: Cauchy’s stress quadric drawn by maple is shown here.

MOHR’S CIRCLE
Mohr’s circle is the most popular and best known pictorial representation of the state of
stress at a point. Readers of this book are expected to have had a good understanding of this
topic for the two-dimensional case21 . They would also have seen that such a visualisation
is possible for the state of strain at point also. Indeed, it is equally valid for all examples
of a second order tensor (inertia tensor, curvature tensor, etc.).
We shall not discuss Mohr’s circle in two dimensions as it is elementary. Here we shall
consider Mohr’s circles in three dimensions. Now there are three (or rather, three pairs
of) Mohr’s circles. As in the two-dimensional case, these are also drawn on the normal
and shear stress axes, σ(ν) and τ(ν) , respectively. If the principal stresses (σ11 , σ22 , σ33 ) are
known or given, these circles tell us quickly, though not as quickly as in two-dimensions,
the normal and shear stress components — σ(ν) and τ(ν) — respectively of the resultant
stress vector on any plane ν passing through the point.
We assume σ11 > σ22 > σ33 . The centres and the radii of the three pairs of circles are
σ11 + σ22 σ11 − σ22 σ11 + σ22
(i) σ11 and σ22 centre E: ; radii: ; σ33 − ; Fig. 5.26a;
2 2 2
σ22 + σ33 σ22 − σ33 σ22 + σ33
(i) σ22 and σ33 centre F : ; radii: ; σ11 − ; Fig. 5.26b;
2 2 2
σ33 + σ11 σ11 − σ33 σ33 + σ11
(i) σ33 and σ11 centre D: ; radii: ; σ22 − ; Fig. 5.26c.
2 2 2
These pairs of circles are shown, one pair at a time, in these figures 5.26a, 5.26b, 5.26c.
Here we have displayed the circles22 , but we have not carried out the calculations that lead
us to these. We shall now undertake this exercise.
Let us begin from the following three equations that we already know. The first one
follows from Eqs (5.28a, 5.28b, 5.28c), while the second one expresses the normal stress on an
21
This construction, Mohr’s circle, is treated quite well in almost all good books on the Mechanics of Solids
and its previous avatar as Strength of Materials.
22
The circles are drawn for a general case where all the three principal stresses are distinct and positive:
σ11 > σ22 > σ33 > 0. A set of numerical values σ11 = 50, σ22 = 30, σ33 = −10, all in MPa is considered
in the example shown later [p. 13-34].
5-33 More About Stresses

(a) The first pair of circles (b) The second pair of circles (c) The third pair of circles

Figure 5.26: Shown here are three pairs of circles. These are the limiting circles correspond-
ing to the limiting values of 0 and 1 that each direction cosine can have. When all of these
pairs of circles are considered together, only the common area (shown shaded in Fig. 5.29b)
is of interest to us.

inclined plane ν in terms of the direction cosines l ≡ cos (ν, x), m ≡ cos (ν, y), n ≡ cos (ν, z)
and the (given) principal stresses. The third one is obvious.
(ν)
2 2 2
σ11 l + σ22 m2 + σ33
2
n2 = p2(ν) ≡ | T |2 ; (5.42a)
2 2 2
σ11 l + σ22 m + σ33 n = σνν ; (5.42b)
l2 + m2 + n2 = 1. (5.42c)
The solution is, using Cramer’s rule,
 2 2 2 2 
  2 2 2 2 
  2 2 2 2 

(σνν + τ(ν) ) σ22 σ33 σ11 (σνν + τ(ν) ) σ33 σ11 σ22 (σνν + τ(ν) )
    

 σνν σ22 σ33  σ11
 σνν σ33  σ11
 σ22 σνν 

2
 1 1 1   1 1 1   1 1 1 
l = ; m2 = ; n2 =
|D| |D| |D|

where the denominator determinant |D| is given by

 2 2 2 

σ11 σ22 σ33
 
|D| = σ11 σ22 σ33  .
 1 1 1 

At this stage, we have to branch off to three different cases, because there are qualitative
differences among them.

(a) The principal stresses are distinct: now the denominator determinant D 6= 0. A unique
non-trivial solution exists.

(b) Two principal stresses are equal: now the denominator determinant D = 0. For a
non-trivial solution to exist, all the three numerator determinants have to be separately
zero. We have two cases to consider:
(i) σ11 = σ22 > σ33 ; and (ii) σ11 > σ22 = σ33 .
Advanced Mechanics of Solids 5-34

(c) All the principal stresses are equal: σ11 = σ22 = σ33 .

(i) All the principal stresses are positive: σ11 > σ22 > σ33 > 0. Now the positive (+)
sign is to be taken in Eq. (5.37). The quadric surface is an ellipsoid [Fig. 5.24a]. We
can see, on comparing Eq. (5.37) with the equation to an ellipsoid in the standard
√ √ √
form [Eq. (5.29)], that the semi-axes are 1/ σ11 , 1/ σ11 , 1/ σ11 . For this case, the
normal stress on every plane is positive, i.e., tensile.
(ii) All the principal stresses are negative: σ33 < σ22 < σ11 < 0. Now the negative (−)
sign is to be taken in Eq. (5.37). The quadric surface is again an ellipsoid [Fig. 5.24a].
The normal stress on every plane will obviously be negative, i.e., compressive. For
these two cases [items (i), (ii)] there is no plane on which there is only shear stress.
(iii) As remarked earlier, when all the principal stresses are equal, the quadric surface
becomes a sphere, representing a state of hydrostatic stress (spherical state of stress).
Every plane is a principal stress; there is no shear stress on any plane.
(iv) If two of the principal stresses are equal, the quadric degenerates into an ellipse of
revolution. The conclusions of this paragraph and of the one just above this, are
similar to the ones mentioned for Lamé’s stress ellipsoid.
(v) If one principal stress is negative and the others are positive, that is, if σ11 ≥ σ22 >
0 > σ33 , both the signs in Eq. (5.37) are to be taken. Corresponding to the positive
(+) sign, we have a hyperboloid of one sheet (shaped in the form of a cooling tower).
Corresponding to the negative (−) sign, we have a hyperboloid of two sheets. These
are shown in Figs 5.24b, 5.25.
(vi) If two principal stresses are equal, that is, if σ11 = σ22 > 0 > σ33 , the quadric surfaces
are hyperbolae of revolution.
(vii) If one principal stress, say σzz is zero (σ11 ≥ σ22 > σ33 = 0), Eqs (5.37) tell us what
happens. Now this equation takes the form
σ11 x2 + σ22 y 2 = +1 (σ11 , σ22 both positive, only +1),
which is independent of z. This obviously represents a cylinder — prismatic, every
cross-section is the same, independent of z — which is an ellipse if σ11 > σ22 , and a
circle if σ11 = σ22 .
If σ11 > σ22 = 0 > σ33 , the two principal stresses are unlike, and both the signs (+)
and (−) are to be taken. Eq. (5.37) now takes the form
σ11 x2 + σ33 z 2 = ±1 (σ11 > 0, σ33 < 0, both sign, both +1 and −),
which represents two hyperbolic cylinders [Fig. 5.27a drawn for the numerical val-
ues (units irrelevant) of σ11 = 3; σ33 = −1]. These are parallel to the y-axis and
asymptotic to the plane s
|σ33 |
x=± z.
σ11
5-35 More About Stresses

When the tip P of the radius vector OP falls on any one of these asymptotic planes,
the corresponding normal stress is zero. Why? Because

1
as OP −→ ∞, σ(νν) = √ −→ 0.
OP

If σ11 = 0 > σ22 ≥ σ33 , this is similar to the case discussed above. The only difference
is that the changed version of Eq. (5.37)

σ22 y 2 + σ33 z 2 = −1 (σ22 , σ33 < 0, both negative, only −),

is now independent of x. All the normal stresses are negative (compressive). The
cylinder — prismatic, everywhere the same — is independent of x. The cross-section
is an ellipse if σ22 > σ33 , and a circle if σ22 = σ33 . This elliptic / circular cylinder is
parallel to the x-axis [Fig. 5.27b drawn for the numerical values (units irrelevant) of
σ22 = −2; σ33 = −1].

(a) Graph of σ11 x2 + σ33 z 2 = (b) Graph of σ22 y 2 + σ33 z 2 = (c) Graph of σ22 y 2 + σ33 z 2 =
±1; σ11 = 3; σ33 = −1. ±1; another view −1; σ22 = −2; σ33 = −1.

Figure 5.27: Two special cases of Cauchy’s stress quadric drawn by maple

Figure 5.28: The same figures drawn by mathematica

(viii) What happens when two principal stresses are zero? Well, now we have a one-
dimensional state of stress, the cases of uniaxial tension (σ11 > σ22 = σ33 = 0),
and uniaxial compression (uniaxial) (σ11 = σ22 = 0 > σ33 ). For these cases, Eq.
(5.37) become, respectively,
r
2 1
σ11 x = +1 (σ11 > 0, only +1) −→ x=± ;
σ11
Advanced Mechanics of Solids 5-36

s
2 1
σ33 z = −1 (σ33 < 0, only −1) −→ x=± .
|σ33 |

These represent two infinite planes parallel, respectively, to the yz and xy planes.

This list pretty much covers all situations. The plane ν of the normal stress σνν and
its direction are readily defined by the radius vector OP . But what is the direction of the
resultant stress vector on this plane ν? This is to be found out by an auxiliary construction.
Auxiliary construction:
We have seen that the radius vector OP [Fig. 5.24a] touches the quadric surface at the point
P . Let us consider the normal ν 0 to the quadric surface at the point P and calculate its
direction ratios. This calculation is similar to what we have done earlier. The equation to
the Cauchy’s stress quadric is f (x, y, z) = σ11 x2 + σ22 y 2 + σ33 z 2 ∓ 1 = 0, and the direction
ratios, given by the partial derivatives, are
∂f
= 2σ11 x = 2σ11 OP cos (ν, x);
∂x
∂f
= 2σ22 y = 2σ22 OP cos (ν, y);
∂y
∂f
= 2σ33 z = 2σ33 OP cos (ν, z).
∂y
Using the equations (5.28a, 5.28b, 5.28c), we obtain
∂f (ν)
= 2σ11 OP cos (ν, x) = 2 OP Tx ;
∂x
∂f (ν)
= 2σ22 OP cos (ν, y) = 2 OP Ty ;
∂y
∂f (ν)
= 2σ33 OP cos (ν, z) = 2 OP Tz .
∂z
(ν) (ν) (ν) (ν)
We, thus, find that the components Tx , Ty , Tz of the resultant stress vector T are propor-
tional to the direction ratios of the normal ν 0 . This fact leads to the conclusion that the
resultant stress vector is along the normal ν 0 to the stress quadric at P .
Three different cases are identified here; we have already seen and discussed these cases
earlier. We repeat them here emphasising this geometric picture when the normal ν 0 is
along the radius vector OP .

(i) The principal stresses are distinct (σ11 6= σ22 ; σ22 6= σ33 ; σ33 6= σ11 ): ellipsoid.
The radius vector OP is along the direction ν 0 to the stress quadric at P only when
OP is along a coordinate direction. This is understandable; the resultant stress vector
on a plane can be entirely normal to it, only if there are no tangential (shear) stresses.
This can happen only for the principal stresses (on the principal planes, when the
principal directions are along the coordinate directions). There are three and only
three principal directions now.
5-37 More About Stresses

(ii) Two of the principal stresses are equal (σ11 6= σ22 = σ33 ): ellipse of revolution.
The surface now becomes an ellipse of revolution about the x-axis (1-axis). The
normal ν 0 is along OP if, and only if, OP is perpendicular to the x-axis (in addition,
of course, to the case when OP is along the x-axis). Now the x-direction and all
directions perpendicular to the x-direction are principal directions. The cases when
two of the other principal stresses are equal (σ22 6= σ33 = σ11 , ellipse of revolution
about the y-axis) and (σ33 6= σ11 = σ22 , ellipse of revolution about the z-axis) are also
similar.

(iii) All the principal stresses are equal (σ11 = σ22 = σ33 ): sphere.
The quadric surface for this case is a sphere. Now the normal ν 0 is along the radius
vector OP for all points P . This means that for this case, every direction is principal;
the resultant stress vector on any plane is always normal to this plane.

(iv) If (x, y, z) are non-principal axes, there will be ‘cross-terms’ (xy, yz, zx), and the
equation to the stress quadric appears in the more complex but equivalent form

σxx x2 + σyy y 2 + σzz z 2 + 2τxy xy + 2τyz yz + 2τzx zx = ±k 2 ,

x2 y2 z2
corresponding to  + +  = ±1.
k2 k2 k2
σ11 σ22 σ33

There is nothing special to discuss for the simplified case of a two-dimensional state of stress
(plane stress, σ33 = 0).
We shall take up these cases (a), (b), and (c) separately.
Distinct Principal Stresses: σ11 > σ22 > σ33
If the principal stresses are distinct, that is, if σ11 > σ22 > σ33 , the denominator determinant
|D| 6= 0. The solution for l2 , m2 , n2 is now unique, and is given by
2 + (σ
2
τ(ν) (ν) − σ22 )(σ(ν) − σ33 )
l ≡ cos (ν, x) = ; (5.43a)
(σ11 − σ22 )(σ11 − σ33 )
2
τ(ν) + (σ(ν) − σ33 )(σ(ν) − σ11 )
m2 ≡ cos (ν, y) = ; (5.43b)
(σ22 − σ33 )(σ33 − σ11 )
2
τ(ν) + (σ(ν) − σ11 )(σ(ν) − σ22 )
n2 ≡ cos (ν, z) = . (5.43c)
(σ33 − σ11 )(σ33 − σ22 )

Equation for l2 (5.43a):

The first equation (5.43a) is recast as
2 2
τ(ν) + σ(ν) − σ(ν) (σ22 + σ33 ) + σ22 σ33 − cos2 (ν x)(σ11 − σ22 )(σ11 − σ22 ) = 0;
σ22 + σ33 2 σ22 − σ33 2
   
2
i.e., τ(ν) + σ(ν) − = + cos2 (ν, x)(σ11 − σ22 )(σ11 − σ33 ).
2 2
Advanced Mechanics of Solids 5-38

This equation, after some algebraic manipulations, takes the form

σ22 + σ33 2 σ22 + σ33 2

   
2
τ(ν) + σ(ν) − = σ11 − cos2 (ν, x)
2 2
σ22 − σ33 2 
 
1 − cos2 (ν, x) .

+ (5.44)
2

This equation represents a family of circles in the σ(ν) τ(ν) plane. The principal stresses
σ11 > σ22 > σ33 are all known (given). The direction cosines may be regarded as a parameter
that can take all values between [0, 1].
 
σ22 + σ33
Family of circles, all with the same centre: , , 0 ≤ cos2 (ν, x) ≤ 1;
2
σ22 + σ33 2 σ22 − σ33 2
   
2
for cos (ν, x) = 0 τ(ν) + σ(ν) − = ; (5.45a)
2 2
σ22 + σ33 2 σ22 + σ33 2
   
2
for cos (ν, x) = 1 τ(ν) + σ(ν) − = σ11 − . (5.45b)
2 2

These are the smallest and the largest limiting circles [Fig. 5.26a]. The operating area
between these limiting circles is shown shaded.
Equations for m2 (5.43b) and n2 (5.43c):
These two equations, on exactly similar treatment, lead to the following.
 
σ33 + σ11
Family of circles, all with the same centre: , 0 , 0 ≤ cos2 (ν, y) ≤ 1;
2
σ33 + σ11 2 σ33 − σ11 2
   
2
for cos (ν, y) = 0 τ(ν) + σ(ν) − = ; (5.46a)
2 2
σ33 + σ11 2 σ33 + σ11 2
   
2
for cos (ν, y) = 1 τ(ν) + σ(ν) − = σ22 − . (5.46b)
2 2

These are the smallest and the largest limiting circles [Fig. 5.26b]. The operating area
between these limiting circles is shown shaded.
 
σ11 + σ22
Family of circles, all with the same centre: , 0 , 0 ≤ cos2 (ν, z) ≤ 1;
2
σ11 + σ22 2 σ11 − σ22 2
   
2
for cos (ν, z) = 0 τ(ν) + σ(ν) − = ; (5.47a)
2 2
σ11 + σ22 2 σ11 + σ22 2
   
2
for cos (ν, z) = 1 τ(ν) + σ(ν) − = σ33 − . (5.47b)
2 2

These are the smallest and the largest limiting circles [Fig. 5.26c]. The operating area
between these limiting circles is shown shaded.
5-39 More About Stresses

(a) All the three pairs of circles (b) Shaded part: common (operative) part

Figure 5.29: All the pairs of Mohr’s semi-circles are shown in Fig. 5.29a. The shaded part,
which alone is the operative region, is shown in Fig. 5.29b. Some of the important or special
points are also marked here.

Further Discussion
We have seen that only the shaded regions in the three pairs of circles is operative. Now
when all the principal stresses are acting, a typical point P with its coordinates (σ(ν) , τ(ν) )
in the σ(ν) , τ(ν) plane can lie only the shaded region common to all the shaded regions shown
earlier [Fig. 5.29b].
What are the maximum and minimum normal stresses at the point P ? What are the
maximum and minimum shear stresses at P ? Will there be a normal stress on these planes?
These and a number of similar questions can be answered readily from the Mohr’s circle
construction. From Figs 5.29b, 5.30c we can note the following. [If the circles are drawn to
scale, the (approximate) numerical values can be measured from these Mohr’s circles.]

Point A: σmax = σ11 Point B: σmin =σ33

σ11 − σ33 σ11 + σ33
Point C: τmax = Point E: σ(max τ ) =
2 2
Points A, B, D: τmin = 0

On which plane(s) do these maximum and minimum shear stresses act? To answer this, all
we have to do is to substitute these values in Eqs (5.43a, 5.43b, 5.43c). In this way, the
direction cosines are obtained which define the plane that we seek.
We can now make the following statements about the nature of stress at a typical point
P [coordinates σ(ν) , τ(ν) ] in the shaded region.

(i) The state of stress at P is defined completely by the three principal stresses (and their
directions).

(ii) The maximum and minimum normal stresses are, respectively, σ11 and σ33 .

(iii) The minimum shear stress, τmin = 0, acts on the principal planes.
Advanced Mechanics of Solids 5-40

(iv) The maximum shear stress, τmax = (σ11 −σ33 )/2 acts on two planes. These two planes
are perpendicular to the σ33 principal direction, and inclined at 45◦ to the σ11 and
σ22 principal directions.

(v) The normal stress σ(max τ ) on these two planes is (σ11 + σ33 )/2.

Mohr Circle Construction

(a) Mohr’s circle: stage 1 (b) Mohr’s circle: stage 2 (c) Mohr’s circles: stage 3

Figure 5.30: The three circles aaaa, bbbb, cccc are shown. The procedure of completing the
construction is explained in the text.

The Mohr’s circles are constructed as described below. We shall assume that the princi-
pal stresses σ11 , σ22 , σ33 are known. We shall see how the normal stress σ(ν) and the shearing
stress τ(ν) on a plane ν of given direction cosines l ≡ cos (ν, x), m ≡ cos (ν, y), n ≡ cos (ν, z)
can be obtained from this graphical construction.

(i) Draw (to a convenient scale) the horizontal σ(ν) and the vertical τ(ν) axes, and locate
the points A (σ11 ), B (σ22 ), C (σ33 ) [Fig. 5.30a].

(ii) Choose these values pairwise, and draw the three circles aaaa, bbbb, cccc [Fig. 5.30a].
σ11 + σ33 σ − σ 
 11 33 
circle aaaa: centre D: radius  ;
2 2
σ − σ 
σ22 + σ11  22 11 
circle bbbb: centre E: radius  ;
2 2
σ − σ 
σ33 + σ22  33 22 
circle cccc: centre F : radius  .
2 2

(iii) Calculate the angles corresponding to the given direction cosines l ≡ cos (ν, x), m ≡
cos (ν, y), n ≡ cos (ν, z). [Actually only two angles are sufficient, just like only two
of the direction cosines need be given, the third one being automatically specified or
known because of the relation l2 + m2 + n2 = 1.]

θl ≡ (ν, x) ≡ cos−1 (ν, x); θm ≡ (ν, y) ≡ cos−1 (ν, y); θn ≡ (ν, z) ≡ cos−1 (ν, z).
5-41 More About Stresses

(iv) Draw two lines CG, at the calculate angle θl ≡ (ν, x) ≡ cos−1 (ν, x), and AG through
the points C and A to touch the circle aaaa at the point G [Fig. 5.30b].

(v) With D as centre, draw a circular arc connecting A and G.

(vi) In the same way, draw two lines AI, at the calculated angle θn ≡ (ν, z) ≡ cos−1 (ν, z),
and CI through the points A and C to touch the circle aaaa at the point I [Fig. 5.30c].

(vii) With E as centre, draw a circular arc through the points I and J. These two circular
arcs intersect at the point P .

(viii) Now the point P represents the point at which we are considering the state of stress.
Thus, its two coordinates σ(ν) and τ(ν) give us the desired normal stress and the shear
stress on the plane ν.

The construction can be carried out in the reverse direction also; that is, if the normal and
shearing stresses are given, and we are required to identify the plane on which these act.
We have described a method to draw the two circular arc (shown in dotted lines in Fig.
5.30c). The radii of these circular arcs can be obtained analytically also23 from Eq. (5.44)
and its companion equations (which are not explicitly shown). The three circular arcs —
we have shown only two of them — will necessarily pass through the unique point P (with
its coordinates σ(ν) , τ(ν) ).
We are now required to prove that the angles that we have set out and marked as (ν, x),
(ν, y) and (ν, z) are indeed the given direction cosines. We shall carry out this exercise
presently as shown below.
To show that the angles marked (ν, x), (ν, y), (ν, z) are the direction cosines
Referring to Fig. 5.30c, let us call for convenience the angle EAI (marked (ν, z) = φ in the
figure) φ. We need to prove that φ = (ν, z). From the triangle EAI (not fully drawn in the
figure), we have
(EI)2 = (AE)2 + (AI)2 − 2(AE)(AI) cos φ.
EI is the radius R12 that we have calculated. From the following equations,
 2
2 2 σ11 − σ22
(EI) = R12 = + cos2 (ν, z)(σ33 − σ11 )(σ33 − σ22 );
2
23
These can be calculated readily for the given values of the direction cosines. These equations, we can see,
are of the form x2 + y 2 = r2 . Thus, the radii of these three arcs can be seen to be:
r
 σ − σ 2
22 33
R23 = + cos2 (ν, x)(σ11 − σ22 )(σ11 − σ33 );
2
r
 σ − σ 2
33 11
R31 = + cos2 (ν, y)(σ22 − σ33 )(σ22 − σ11 );
2
r
 σ − σ 2
11 22
R12 = + cos2 (ν, z)(σ33 − σ11 )(σ33 − σ22 ).
2
Advanced Mechanics of Solids 5-42

σ11 − σ22 2
 
2
(AE) = ;
2
(AI) = (CA) cos φ = (σ11 − σ22 ) cos φ.
Solving for cos φ, we find that cos φ = cos (ν, z).
Two Equal Principal Stresses: σ11 = σ22 > σ33
Let us return to Eqs (5.42a,5.42b,5.42c). We recall that the denominator determinant |D|
is given by  2 2 2 

σ11 σ22 σ33
 
|D| = σ11 σ22 σ33  ,
 1 1 1 
which now vanishes because two of the principal stresses are equal. For this case, a non-
trivial solution for (l2 , m2 , n2 ) [Eqs (5.42a, 5.42b, 5.42c)] can exist only if all the numerator
determinants are zero. This condition gives us
 2 2 ) σ2 2 
  2 2 + τ 2 ) σ2 
  2 2 2 + τ 2 )

(σνν + τ(ν) 22 σ33 
σ11 (σνν (ν) 33 
σ11 σ11 (σνν (ν) 
  

 σνν σ22 σ33  = σ11 σνν σ33  = σ11 σ11 σνν  = 0.

 1 1 1   1 1 1   1 1 1 

The third determinant is zero here, because its first two columns are the same (σ11 = σ22 ).
The first and the second ones when equated to zero give us the same equation
σ11 + σ22 2 σ11 + σ22 2
   
2
τ(ν) + σ(ν) − = ,
2 2
which represents a circle. As σ11 = σ22 , the two circles bbbb and cccc coalesce; the shaded
area degenerates to the semi-circle connecting the points C and A. This now represents
a simpler state of stress (axi-symmetry about the 3 (z)-axis)24 . The normal σ(ν) and the
shear stresses τ(ν) are the same for all (consistent, physically possible) values of the two
direction cosines l ≡ cos (ν, x) and m ≡ cos (ν, y).
The other case of (two principal stresses being equal) σ11 > σ22 = σ33 is similar to the
one considered above and it is, therefore, left out.
All Principal Stresses Equal: σ11 = σ22 = σ33
When (σ11 = σ22 = σ33 ), all the circles degenerate to the single point [σ(ν) , 0]. Now every
plane is a principal plane and that every stress a principal one. There is no shear stress on
any plane. This represents a case of hydrostatic state of stress.
σ(ν) = σ11 = σ22 = σ33 ; τ(ν) = 0 on all planes ν.

We shall discuss in the next chapter the analysis of strain at a point.

24
Lamé’s ellipsoid becomes an ellipse of revolution. The consequences or the conclusions should be obvious
by now.
Chapter 6

ANALYSIS OF STRAIN AT A
POINT

We have seen the analysis of stress at a point. Next, we have to carry out a similar
exercise on the nature of strain at a point. To do this, we have to examine the details
of the deformation in the neighbourhood of a typical point. A few remarks are made to
understand the broad issues involved. Later only a simple-minded analysis is used.
GENERAL INTRODUCTORY REMARKS
There is a great deal of common ground between the analyses of stress and strain at a point;
they are both very similar mathematically. They are examples of a second order tensor.
The mathematical equations are almost similar. All the topics in the analysis of stress carry
over to the analysis of strain also with very minor changes. However, the physical basis of
the analysis of strain is different: the geometry of deformation instead of the equilibrium of
the body under the influence of the forces1 . This is a major difference.
TO TRACK THE DEFORMATION
The body, to begin with, is unloaded and stress-free. When an external load is applied,
the body is deformed. We need to track this and examine this in detail. The body may
undergo rigid body motion; it may also be deformed or strained. The rigid body motion
may be a translation and / or a rotation. We need to identify and filter out the rigid body
motion. It is only the remaining part, the deformation part, that concerns us. Only this
part is associated with stresses and strains.
To explain this, let us take an example. When a number of policemen are deployed to
provide security support, the Chief of Police may wish to know where each policeman is at
every instant. Here the policemen have identity tags, their names or identity numbers. For
a deformable solid / fluid, the situation is similar but harder; the material particles have
no such identification tags. How shall we address this difficulty? This is our first concern.

1
I am grateful to Dr Gangan Prathap for his suggestion to refer to these two as kinematics vs kinetics.
Advanced Mechanics of Solids 6-2

When the undeformed body is loaded, it is

deformed. Fig. 6.1 shows the deformation in
the neigbourhood of a point. We can establish
a mapping that relates every point in the
undeformed body to a corresponding point
in the deformed body. In particular, a brick
element before and after deformation is shown.

We need to consider its deformation, and exam-

ine the changes in its lengths and angles as the
Figure 6.1: Deformation in body deforms. We shall carry out this impor-
the neigbourhood of a point tant exercise a little later.

A material particle cannot possibly occupy two (or more) places (points), nor can two
(or more) material particles occupy the same place (point) at the same instant of time.
Labelling of Material Particles

(a) Deformation (b) Initial configuration (c) Current configuration

Figure 6.2: To assign an identification label to each particle: the coordinates (X1 , X2 , X3 )
of a particle at a reference time, say t = t0 , serve as the identification label of this particle.

Let us try to label the various material particles so that their movements can be con-
veniently tracked. Fig. 6.2 shows a typical material particle P in its initial (time t = t0 )
and current configurations, P (X1 , X2 , X3 , 0) and P ∗ (x1 , x2 , x3 , t) at a general instant of
time t, respectively. The coordinates of P at time t = t0 are (X1 , X2 , X3 ). These num-
bers (X1 , X2 , X3 ) uniquely specify a material particle. They are, in a manner of speaking,
the ‘place of birth’ of the particle! When the particle moves, its position is changed from
P (X1 , X2 , X3 ) to P ∗ (x1 , x2 , x3 ) at time t. Thus,

xi = xi (X1 , X2 , X3 , t) (i = 1, 2, 3) (6.1)

specifies the position (place) xi (that is, x1 , x2 , x3 ) at time t of the particle (X1 , X2 , X3 ).
(X1 , X2 , X3 ) serves as the particle identification label, and (x1 , x2 , x3 ) the place that it
occupies at time t. This is the same as stating that the Mumbai man — ‘place of birth’:
6-3 Strain at a Point

Mumbai (X1 , X2 , X3 ) — is now at time t at Paris (x1 , x2 , x3 ), etc. No matter where he

goes, his identity label (X1 , X2 , X3 ) remains the same, the Mumbai man!
This one-to-one mapping given by Eq. (6.1) can be inverted to yield

Xi = Xi (x1 , x2 , x3 , t) (i = 1, 2, 3). (6.2)

Let us note that the independent variables are (X1 , X2 , X3 ) in Eq. (6.1) and (x1 , x2 , x3 ) in
Eq. (6.4). These form the basis of the two approaches, the Lagrangian and the Eulerian.
Lagrangian and Eulerian Approaches
We note that a quantity of importance, say, the velocity may be expressed either in terms
of (X1 , X2 , X3 ) as the independent coordinates as

Vi = Vi (X1 , X2 , X3 , t) (i = 1, 2, 3), (6.3)

or in terms of (x1 , x2 , x3 ) as the independent variable as

vi = vi (x1 , x2 , x3 , t) (i = 1, 2, 3). (6.4)

It is not a good idea to use the same dependent variable Vi (or vi ) for both these represen-
tations. Why is this so? Because the functional forms are different2 .
We have seen that (X1 , X2 , X3 ) serve as the particle identification labels. If these are the
independent variables, we refer to this as the Lagrangian approach (also called the material
or substantial approach). On other hand, if the position (or place) variables (x1 , x2 , x3 )
are the independent variables, this is called the Eulerian approach3 (also called the local
approach). (X1 , X2 , X3 ) are called the Lagrange’s coordinates, and (x1 , x2 , x3 ) the Euler’s
coordinates4 .
The above explanation is from the mathematical point of view. Physically what is the
basic difference between the two? Let us examine.
In the Lagrangian approach, we focus on a material particle, and watch the different
places that it visits at different times. This is the physical meaning of the equation (6.1).
Obviously all the material particles are covered. This equation is a satisfactory method of
describing the motion by a mathematical / analytical equation. The actual functional form
2
To explain this point clearly, let us take an example.
Let y = y(x) = x2 and let x = x(t) = 1 + t. If we substitute x = 1 + t in the first equation, we obtain
y = y(x) = x2 = (1 + t)2 = 1 + 2t + t2 . This is not the same as y(t) because y(t) means t2 . Thus, the
correct way to write is y = y(x) = y(x(t)) = (1 + t)2 = 1 + 2t + t2 = y1 (t). Note that we now have a
different functional form y1 (t), which must be distinguished carefully from y(t).
v = v(x1 , x2 , x3 , t) = v(x1 (Xi ), x2 (Xi ), x3 (Xi ), t) 6= v(X1 , X2 , X3 , t)!
3
Actually both approaches are due to the great Leonard Euler (April 1707 - Sept. 1783). Euler is known
for his extraordinary magnanimity in letting others take the credit for his original work.
4
Not to be confused with the Euler angles φ, θ, ψ used in rigid body dynamics when discussing the motion
of a gyroscope
Advanced Mechanics of Solids 6-4

cannot be known at this time of formulating the problem; it can be known only after the
(boundary value) problem is solved.
In the Eulerian approach, we focus on a place, and watch the various material particles
that visit this chosen place. Again, all the places are covered. Either of these two approaches
gives a complete description of the motion. In both cases, we need to be able to derive the
mathematical expressions for the velocity, momentum, kinetic energy, etc. This is part
of the development of the theoretical foundation of continuum mechanics that seeks to
encompass both solid and fluid mechanics in its scope5 .
Representation of a Function and Its Differentiation
We shall now see how functions are represented, and differentiation carried out, in the two
approaches. Let a function, to be specific, the velocity Vi be represented in the Lagrangian
coordinate system as
Vi = Vi (X1 , X2 , X3 , t) (i = 1, 2, 3). (6.5)
If we differentiate this w.r.to time, we obtain
dVi ∂Vi 
=  , (6.6)
dt ∂t L
where the special notation is used to indicate the Lagrangian point of view. The displace-
ment vector is
ui = xi (X1 , X2 , X3 , t) − Xi (i = 1, 2, 3), (6.7)
so that
∂ui ∂xi ∂Xi ∂xi
= − = − δij (i, j = 1, 2, 3). (6.8)
∂Xj ∂Xj ∂Xj ∂Xj
In the Eulerian approach,
vi = vi (x1 , x2 , x3 , t) (i = 1, 2, 3), (6.9)
which represents the velocity of an unidentified material particle now at time t occupying
the place (x1 , x2 , x3 ). On differentiation, we obtain
dvi ∂vi ∂vi
= ẋj + , (summation on the repeated index j). (6.10)
dt ∂xj ∂t
5
Perhaps an analogy might help clarify this difference.
In a coed college, let us assume that one youngster follows a particular girl and notes down the various
places she visits at various instants of time. For example, he follows a girl, say, Usha, and notes down
that at 8:50 a.m. she reaches the college gate, that at 8:55 a.m. she visits the girls’ common room, that
at 8:58 a.m. she enters her classroom, etc. Another boy similarly follows another girl, say, Lakshmi, and
similarly notes down the places along with the instants of time when she visits these places. If all such
records are collected together, we get a complete picture of where every girl was at every instant of time.
Let us now turn around and take a different way to do this. One boy stations himself at a given fixed
point, say, the canteen and notes down the various girls who reach there: Usha at 10:15 a.m., Lakshmi at
11:20 a.m., and so on. If every girl is covered in the first approach, and every place in the second, we have
a complete picture of the movement of every girl at every instant of time.
“tarun.astāvad tarun.īsaktah”!
6-5 Strain at a Point

From our earlier study of fluid mechanics, we recognise the left hand side to be the sub-
stantial, or material, derivative6 . The two terms on the right hand side are the convective
derivative and the local derivative, respectively. Thus, as vi represents the velocity of an
unidentified material particle, and as the acceleration is to be tied down to a material parti-
cle — its identity is not revealed directly, but is indicated indirectly by the place (x1 , x2 , x3 )
that it occupies at time t — we recognise them as the convective and local accelerations,
respectively.
The displacement vector

ui = xi − Xi (x1 , x2 , x3 , t) (i = 1, 2, 3) (6.11)

may be differentiated w.r.to the space variable xj as

∂ui ∂xi ∂Xi ∂Xi

= − = δij − . (6.12)
∂xj ∂xj ∂xj ∂xj

Simplified Case for Solids Only

These discussions are equally valid for solids and fluids. Such a treatment is necessary in
continuum mechanics where we try to build a theory equally applicable for both solids and
fluids. As we do not consider flow, time is no more a factor. Thus, we disregard the variable
time t in all further discussions and rewrite Eqs (6.1) and (6.7) as

xi = xi (X1 , X2 , X3 ) (i = 1, 2, 3) (6.13)
ui = xi (X1 , X2 , X3 ) − Xi (i = 1, 2, 3), (6.14)

respectively. Similarly, Eq. (6.11) is also recast as

ui = xi − Xi (x1 , x2 , x3 ) (i = 1, 2, 3), (6.15)

A SIMPLIFIED APPROACH FOR THE LINEAR THEORY

There several cases where the analysis can be complicated: a general (curvilinear) coordi-
nate system, convected coordinates, large deformation theory, etc. We shall drop all such
complications and discuss the analysis of deformation sufficient for the linear infinitesimal
theory of elasticity staying in a rectangular coordinate system.
Deformations in the Neighbourhood of a Typical Point
We have only one rectangular cartesian coordinate system Oxyz now; one is sufficient. The
deformations are so small (unlike the general case of large deformations) that it is no more
necessary to distinguish between the undeformed R and the deformed R∗ configurations.
As before, our interest is in the analysis of deformation in a neighbourhood of a typical
point P . By doing so, we obtain the relevant information. Equilibrium of forces is not
considered here; only geometry is relevant.
6 Dvi
often written as Dt
Advanced Mechanics of Solids 6-6

(a) ‘Small’ deformation of a body (b) The details near a typical point

Figure 6.3: A rectangular cartesian coordinate system Oxyz is chosen to discuss the nature
of deformation in the neighbourhood of a typical point P (x0 , y0 , z0 ). The figure on the right
up is a blow up of the neighbourhood of (i) a point P (x0 , y0 , z0 ) and its map P ∗ (x∗0 , y0∗ , z0∗ ),
and (ii) a point Q(x, y, z) and its map Q∗ (x∗ , y ∗ , z ∗ ).

A typical point P (x0 , y0 , z0 ) and its neighbour Q(x, y, z) are carried to P ∗ (x∗0 , y0∗ , z0∗ )
and Q∗ (x∗ , y ∗ , z ∗ ), respectively. Comparing the coordinates of (i) P (x0 , y0 , z0 ) and its map
P ∗ (x∗0 , y0∗ , z0∗ ), and (ii) Q(x, y, z) and its map Q∗ (x∗ , y ∗ , z ∗ ) [Fig. 6.3a], we see that

dx = x − x0 ; dy = y − y0 ; dz = z − z0 ; (6.16a)
∗ ∗
dx = x − x∗0 ; ∗
dy = y − ∗
y0∗ ; ∗
dz = z −∗
z0∗ ; (6.16b)
∗
x = x + u(x, y, z) = x + u(x0 + dx, y0 + dy, z0 + dz); (6.16c)
∗
y = y + v(x, y, z) = y + v(x0 + dx, y0 + dy, z0 + dz); (6.16d)
∗
z = z + w(x, y, z) = z + w(x0 + dx, y0 + dy, z0 + dz). (6.16e)

Expanding these equations in a Taylor series about the base station P (x0 , y0 , z0 ), we have
     
∂u ∂u ∂u
x∗ = x + u(x0 , y0 , z0 ) + dx + dy + dz; (6.17a)
∂x P ∂y P ∂z P
     
∗ ∂v ∂v ∂v
y = y + v(x0 , y0 , z0 ) + dx + dy + dz; (6.17b)
∂x P ∂y P ∂z P
     
∗ ∂w ∂w ∂w
z = z + w(x0 , y0 , z0 ) + dx + dy + dz. (6.17c)
∂x P ∂y P ∂z P

The series may be terminated with the terms explicitly shown for two separate cases: (i)
if the investigation is restricted to the immediate neighbourhood of the point P , so that
dx, dy, dz are small, and (ii) if the higher derivatives of the displacements components like
∂ 2 u/∂x2 and ∂ 2 v/∂y 2 are zero, or at least very small. The latter condition means that the
derivatives like ∂u/∂x and ∂v/∂y — these are strain components — are either constants
or nearly so. If the strain components are nearly constant, it means that all parts of the
body are subjected to nearly the same strain, which corresponds to the special case of
homogeneous deformation.
It is sometimes more convenient to write these equations (6.17a - 6.17c) differently so
that the position (that is, the coordinates) of Q∗ depends only on the positions (that is, the
6-7 Strain at a Point

coordinates) of Q and P . This can be done by using Eq. (6.16a) to eliminate dx, dy, dz.
On doing so, the above equations (6.17a - 6.17c) appear as
     
∗ ∂u ∂u ∂u
x = u(x0 , y0 , z0 ) − x0 − y0 − z0
∂x P ∂y P ∂z P
       
∂u ∂u ∂u
+ 1+ x+ y+ z; (6.18a)
∂x P ∂y P ∂z P
     
∗ ∂v ∂v ∂v
y = v(x0 , y0 , z0 ) − x0 − y0 − z0
∂x P ∂y P ∂z P
       
∂v ∂v ∂v
+ 1+ y+ x+ z; (6.18b)
∂y P ∂x P ∂z P
     
∂w ∂w ∂w
z ∗ = w(x0 , y0 , z0 ) − x0 − y0 − z0
∂x P ∂y P ∂z P
       
∂w ∂w ∂w
+ 1+ z+ x+ y. (6.18c)
∂z P ∂x P ∂y P

Let us examine these equations. For a given base station P (x0 , y0 , z0 ), the first four terms
are constants; they depend only on the values at the base station P . Thus, for these
equations that define the mapping from Q to Q∗ , these four terms being constant represent
a translation. They do not affect the stretching and rotation of the line element. Calling
them (Cu )P , (Cv )P , (Cw )P , let us write the above three equations (6.18a - 6.18c) as
       
∗ ∂u ∂u ∂u
x = (Cu )P + 1 + x+ y+ z; (6.19a)
∂x P ∂y P ∂z P
       
∗ ∂v ∂v ∂v
y = (Cv )P + x+ 1+ y+ z; (6.19b)
∂x P ∂y P ∂z P
       
∗ ∂w ∂w ∂w
z = (Cw )P + x+ y+ 1+ z. (6.19c)
∂x P ∂y P ∂z P

These equations have only linear terms; there are no quadratic or higher degree terms. Such
transformation equations are known as affine transformations or linear transformations. Let
us emphasise that the linearity7 arises because we confined ourselves to the behaviour in
the immediate neighbourhood of the base station.
These equations can also be recast into a form so that we can see how the elemental
lengths dx, dy, dz are mapped to dx∗ , dy ∗ , dz ∗ . We may use Eqs (6.16c - 6.16e) to make
the necessary changes.

dx∗ = x∗ − x∗0 ; dy ∗ = y ∗ − y0∗ ; dz ∗ = z ∗ − z0∗ ;

x∗ = x0 + u(x0 , y0 , z0 ); y ∗ = y0 + v(x0 , y0 , z0 ); z ∗ = z0 + w(x0 , y0 , z0 ).
7
Even in the case of a general (nonlinear) transformation, the transformation equations are locally lin-
ear. Hence linear transformations have great importance, even though nearly all transformations may be
nonlinear.
Advanced Mechanics of Solids 6-8

Thus, we have
      
∗ ∂u ∂u ∂u
dx = 1 + dx + dy + dz; (6.20a)
∂x
P ∂y P ∂z P
       
∗ ∂v ∂v ∂v
dy = dx + 1 + dy + dz; (6.20b)
∂x P ∂y P ∂z P
       
∂w ∂w ∂w
dz ∗ = dx + dy + 1 + dz. (6.20c)
∂x P ∂y P ∂z P

These also represent an affine transformation; these may be inverted to give dx, dy, dz in
terms of dx∗ , dy ∗ , dz ∗ which will also represent an affine transformation. Because of their
importance, we shall briefly discuss below affine transformations.
Affine Transformations
Eqs (6.19a, 6.19b, 6.19c) and (6.20a, 6.20b, 6.20c) are examples of affine transformations.
They have some important properties. We shall see some of them below.
Let us define8 an affine transformation (also known as an affine mapping, a homogeneous
mapping, and a linear mapping) from the variables xi (i = 1, 2, 3) to the variable yi (i =
1, 2, 3) defined by the equation

yi = ci0 + (δij + cij ) xj (i, j = 1, 2, 3), (6.21)

where ci0 and cij are constants. For this to be a one-to-one mapping, the determinant of the
coefficients shall not vanish: det (δij + cij ) 6= 0. Then the equation (6.21) may be inverted
to yield another affine mapping, this time from the variables yi (i = 1, 2, 3) to the original
variables xi (i = 1, 2, 3), as

xi = bi0 + (δij + bij ) yj (i, j = 1, 2, 3). (6.22)

Such an affine transformation has several — important and useful in applications — prop-
erties. Some of them relevant for us are discussed below.
Properties of Affine Mappings
(i) Planes are mapped onto planes.

(ii) Straight lines are mapped onto straight lines.

(iii) Equal vectors — directed line segments — are mapped onto equal vectors.

(iv) Parallel planes are mapped onto parallel planes.

(v) Identical polygons are mapped onto identical polygons.

8
A general definition acceptable to mathematicians is not attempted here. The definition is limited to the
context of deformation of a body and to a discussion of strains and deformations of a solid.
6-9 Strain at a Point

(vi) Two successive affine transformations: This is the product transformation, the result
of two transformations, one followed by the other. Let us now restrict the coeffi-
cients ci0 , cij ; dk0 , dki to small values. The mappings are now infinitesimal affine
transformations.

(vii) Infinitesimal affine transformations: The coefficients are small; their products are
neglected. This is the case in most engineering structures and machine parts.
Thus, such product terms can be dropped in the product transformation (the result of
successive transformations). Then the coefficients in the product transformation are
simply the sum of the corresponding coefficients in the two separate transformations.

(viii) The order of the two separate transformation is unimportant. Note particularly that
we obtain the same product mapping irrespective of the order of the two separate
component transformations.
This has consequences in the linear, infinitesimal, theory of elasticity. The final result
of combined loading on a body is independent of the order of application of the
two separate component loadings. This property is an important consequence of the
linearity of the transformation and the associated principle of superposition.

(ix) Several successive infinitesimal affine transformations. The same result is true for
more number of successive, infinitesimal, affine transformations also.

Having discussed affine transformations, let us return to Eqs (6.20a)-(6.20c).

STRAINS AND STRAIN TRANSFORMATIONS

(a) A rectangular parallelepiped (b) Shape after deformation

Figure 6.4: A brick shaped block [Fig. 6.4a] is deformed to the shape shown in Fig. 6.4b

Fig. 6.4 shows a small block before and after deformations: AB, AP, AD, · · · →
A∗ B ∗ , A∗ P ∗ , A∗ D∗ , · · · . It is clear that parallel planes such as AB, CD are carried, on
deformation, to the parallel planes such as A∗ B ∗ , C ∗ D∗ . Small parallelepipeds — why
small? Otherwise the transformation may not be affine (linear) — are deformed as par-
allelepipeds. The sides of the undeformed parallelepiped [Fig. 6.4a] are chosen along the
coordinate axes of lengths dx ≡ ∆x, dy ≡ ∆y, dz ≡ ∆z. These lengths, on deformation,
become dx∗ , dy ∗ , dz ∗ , respectively.
Advanced Mechanics of Solids 6-10

Linear Strain
We now calculate the square of the deformed length (P ∗ Q∗ )2 = (dx∗ )2 + (dy ∗ )2 + (dz ∗ )2
using the equations (6.20a - 6.20c) and dropping quantities of higher order of small (because
of the assumed ‘small’ deformations). This gives us
           
∗ ∗ 2 ∂u 2 ∂v 2 ∂w
(P Q ) = 1 + 2 (dx) + 1 + 2 (dy) 1 + 2 (dz)2 +
∂x P ∂y P ∂z P
         
∂u ∂v ∂v ∂w
2 + (dx)(dy) + 2 + (dy)(dz)+
∂y P ∂x P ∂z P ∂y P
    
∂w ∂u
2 + (dz)(dx) (6.23)
∂x P ∂z P
If l, m, n are the direction cosines of the undeformed line element P Q,
dx dy dx
l= ; m= ; n= ; −→ (P Q)2 = (dx)2 + (dy)2 + (dz)2 ,
PQ PQ PQ
the above equation (6.23) can be recast as
          
∂u ∂v ∂w ∂u ∂v
(1 + eP Q )2 = 1+2 l2 + 2 m2 + 2 n2 + 2 + lm
∂x P ∂y P ∂z P ∂y P ∂x P
         
∂v ∂w ∂w ∂u
+2 + mn + 2 + nl.
∂z P ∂y P ∂x P ∂z P
Expanding the left hand side, leaving out the square of the strain eP Q , and simplifying, we
obtain the expression for the strain in a given direction specified by the direction cosines
(l, m, n) in terms of the strain components w.r.to the x, y, z coordinates.
          
∂u ∂v ∂w ∂u ∂v
eP Q = l2 + m2 + n2 + + lm
∂x P ∂y P ∂z P ∂y P ∂x P
         
∂v ∂w ∂w ∂u
+ + mn + + nl. (6.24)
∂z P ∂y P ∂x P ∂z P

Shearing Strain
To obtain the shearing strain, or rather the expression for the shearing strain, we need to
calculate the change is the angle between (i) the lines P Q (l1 , m1 , n1 ) and P R (l2 , m2 , n2 ),
and (ii) their images P ∗ Q∗ (l1∗ , m∗1 , n∗1 ) and P ∗ R∗ (l2∗ , m∗2 , n∗2 ). The direction cosines are,
as we can readily see, the following.
dx1 dy1 dz1
l1 = m1 = n1 =
PQ PQ PQ
dx2 dy2 dz2
l2 = m2 = n2 =
PR PR PR
∗ dx∗ ∗ dy ∗ ∗ dz ∗
l1 = ∗ 1 ∗ m1 = ∗ 1 ∗ n1 = ∗ 1 ∗
P Q P Q P Q
6-11 Strain at a Point

dx∗2 dy2∗ dz2∗

l2∗ = m∗2 = n∗2 =
P ∗ R∗ P ∗ R∗ P ∗ R∗
The first angle is easy to calculate; it is given by

cos(P Q, P R) = l1 l2 + m1 m2 + n1 n2 .

The second angle is similarly given by

dx∗1 dx∗2 + dy1∗ dy2∗ + dz1∗ dz2∗
cos(P ∗ Q∗ , P ∗ R∗ ) = l1∗ l2∗ + m∗1 m∗2 + n∗1 n∗2 =
(P ∗ Q∗ )(P ∗ R∗ )
We have already worked out the expressions for dx∗ , dy ∗ , dz ∗ [Eqs (6.20a) - (6.20c)]. It is
merely a question of substituting these (rather long) expressions, neglecting quantities of a
higher order of smallness, and cleaning up the expressions. When these steps are carried
out, we obtain the expression for the second angle (P ∗ Q∗ , P ∗ R∗ ).

γ(P Q, P R) = (1 − eP Q − eP R ) [(l1 l2 + m1 m2 + n1 n2 ) + 2(exx l1 l2 + eyy m1 m2 + ezz n1 n2 )]

+ (1 − eP Q − eP R ) [γxy (l1 m2 + l2 m1 ) + γyz (m1 n2 + m2 n1 ) + γzx (n1 l2 + n2 l1 )]

But this analysis is too general. The procedure outlined above tells us the change in angle
between two elemental line elements in two arbitrary directions. We do not need results with
such generality. It is sufficient if we can compute the change in angle when the elemental
lines (P Q, P R) are at right angles. For this case, there is considerable simplification. Now
in the last equation (l1 l2 + m1 m2 + n1 n2 ) = 0, because P Q and P R are at right angles.
Thus, collecting all the information that we have gathered, the strain transformation
equations are obtained as follows. We would do well to change over from γxy , etc. to 2exy ,
etc. (to preserve the tensor character in the transformation equations).
If (l1 , m1 , n1 ); (l2 , m2 , n2 ); (l3 , m3 , n3 ) are the direction cosines of the orthogonal
x0 , y 0 , z 0 axes w.r.to the (x, y, z) axes, the strain transformation laws are:
γxy γyz γzx
e0x0 x0 = l12 exx + m21 eyy + n21 ezz + 2l1 m1 + 2m1 n1 + 2n1 l1 (6.25)
2 2 2
γx0 0 y0 γxy
= l1 l2 exx + m1 m2 eyy + n1 n2 ezz + (l1 m2 + l2 m1 ) (6.26)
2 2
γyz γzx
+ (m1 n2 + m2 n1 ) + (n1 l2 + n2 l1 ) (6.27)
2 2
The companion equations can be written down using cyclic changes: x → y; y →
z; z → x. They may be written compactly as

e0ij = aik ajl ekl and / or [e0 ] = [a][e][a]T .

We can see that all this is analogous to the stress transformation equations. Let us repeat
for emphasis: use exy , etc. instead of γxy , etc. In the equations shown above, γxy /2, etc.
are used so that the equations are in good shape to be rewritten in terms of exy , etc. We
may remark here that all the transformation equations valid for stresses are equally valid
for strains also with this one difference indicated about the factor of 1/2.
Advanced Mechanics of Solids 6-12

FURTHER BEYOND
Having established that the strain is another example of a second order tensor, we can
borrow all the results from the analysis of stress at a point. Thus, for example, principal
stresses and principal planes will now, in the context of strains, be referred to as principal
strains and the corresponding principal directions9 . A few minor changes may have to be
made to suit the context and the physical meanings of the terms. It is the geometry of
deformations that is considered here, whereas it was the equilibrium of forces that was the
basis for deriving the equations in the theory of stresses. An example is given below.
Spherical and Deviatoric Parts of the Strain Tensor
Recall that the stress tensor represented as a matrix may be written as the sum of two
matrices, the first one representing a hydrostatic stress (all the eigenvalues are equal, spher-
ical part) and the other a state of pure shear (the first invariant vanishes, deviatoric part).
Here they correspond to the two components of the deformation: (i) only volume change,
no change in shape — spherical part, there are no off-diagonal terms representing (half) the
shearing strains — and (ii) no volume change, only change in shape or distortion (deviatoric
part, the first invariant representing volumetric strain is zero.)10

volumetric strain ev = eii = e11 + e22 + e33 = 0.

Some minor changes may have to be made to suit the context. The decomposition into the
pure shear (deviatoric part) and the hydrostatic (spherical part) is, in index notation
   
1 1
eij = eij − δij ekk + δij ekk .
3 3

Principal Strains and Principal Directions

Eq. (6.25) tells us the linear strain in a direction defined by the direction cosines l, m, n.
Now we might enquire what its maximum / minimum (in general, stationary values) value
is. This leads us to the principal strains and the corresponding directions to the associated
principal directions. We can begin with Eq. (6.25) and find the maximum / minimum
under the constraint (l2 + m2 + n2 − 1 = 0). This problem is exactly similar to what we
discussed earlier on p. 4-31, and for exactly the same reason, we do not proceed further.
There are two important topics that follow, or ought to follow, the analysis of strain
at a point. These are the strain-displacement relations and the compatibility conditions.
These are discussed later in the book.

We shall discuss constitutive equations in the next chapter.

9
Recall that we had stated as an aside in a lighter vein that it is the same story — the same events, the
same ending, everything is the same — except the names of the characters. This is particularly true here.
10
Incompressibility is an important consideration in several places. Plastic flow is associated with no volume
change.
Chapter 7

CONSTITUTIVE EQUATIONS

This chapter refers to the constitutive equations, known also as the material laws. These
are the equations that refer to the properties of the material1 . Elaborate theories have been
developed. Our scope of investigation is limited to elastic materials. Several approaches
are possible, but we shall first see the most natural and simplest of them. The appropriate
equations relevant to our context are Hooke’s law2 , and its extension to three dimensions
called the generalised Hooke’s law.
HOOKE’S LAW AND THE GENERALISED HOOKE’S LAW
It seems to be easiest to begin with Hooke’s law which when applied to our context of one-
dimensional stresses and strains is e = σ/E, where E is the Young’s modulus3 of elasticity.
When this is extended to three dimensions, we begin with the equations
σxx σyy σzz
exx = ; eyy = ; ezz = .
E E E
But then there is the Poisson’s effect and the associated Poisson’s ratio. Thus, the strains
would be modified as
σxx σ σzz  σyy σ σxx  σzz σ σyy 
yy zz xx
exx = −ν + ; eyy = −ν + ; ezz = −ν + .
E E E E E E E E E
1
strictly, not so much to the material, but to the model that we choose
2
Named after Robert Hooke (July 1635 - March 1703), British scientist and natural philosopher. To quote
from Den Hartog [3]: “· · · This law was first enunciated by Robert Hooke (1635 - 1703) in Elizabethan
England in connection with his invention of applying a hairspring to a watch or clock”. · · · “Hooke, after
the manner of his age, published his invention in the shape of a riddle, or “anagram” ceiiinosssttuv, which,
when unscrambled, was supposed to read Ut tensio sic vis, Latin for “ as the extension, so is the force”.
Hooke had three entirely different stages in life: (i) the early years when he had no money, (ii) the middle
years when he had lots and lots of money and fame, and (iii) the later years when he had indifferent health.
3
Named after Thomas Young (June 1773 - May 1829), English physicist and physician. It appears that the
concept was used 25 years earlier by G. Riccati, and that the idea can be traced to one of Euler’s papers
80 years before Young’s publication in 1807. Some people considered, or perhaps still consider, Young to
be the last person “who knew everything” although that honour is usually given to Leibnitz.
Advanced Mechanics of Solids 7-2

Just as the linear stress and linear strain are related by the Young’s modulus of elasticity
(E), the shear stress and the shear strain are related by the modulus of rigidity (G). Thus,
τxy τyz τzx
γxy = 2 exy = ; γyz = 2 eyz = ; γzx = 2 ezx = .
G G G
There is no Poisson’s effect4 here. Thus, combining the above equations, we obtain the
generalised Hooke’s law as
1
exx = [σxx − ν (σyy + σzz )] ; (7.1a)
E
1
eyy = [σyy − ν (σzz + σxx )] ; (7.1b)
E
1
ezz = [σzz − ν (σxx + σyy )] ; (7.1c)
E
1 1 τxy 1
exy = γxy = = γyx = eyx ; (7.1d)
2 2 G 2
1 1 τyz 1
eyz = γyz = = γzy = ezy ; (7.1e)
2 2 G 2
1 1 τzx 1
ezx = γzx = = γxz = exz . (7.1f)
2 2 G 2
The numerical values of these (elastic) constants are to be obtained experimentally in a
material testing laboratory5 . This is one approach to obtain the generalised Hooke’s law.
But this is not entirely satisfactory. Other approaches may shed more light on the topic on
hand. We shall examine an alternative approach.
AN ALTERNATIVE APPROACH
We shall start by stating that the stress is a function of strain. This function obviously
must be continuous. Thus we may write σ = σ(e). Now we know that a continuous function
may be written in the form

σ = σ(e) = ao + a1 e + a2 e2 + a3 e3 + · · · + an en ,

where the larger the number of terms, the better will be the approximation. We can have
any degree of approximation. This fact is intuitively clear, but there is a formal proof.
4
Named after Siméon Denis Poisson (June 1781 - April 1840), the famous French mathematician and physi-
cist. He obtained (1829) the value of Poisson’s ratio, not by experiments in the lab, but a mathematical
/ physical argument.
Mild steel has a Poisson’s ratio of about 0.3. Concrete has a Poisson’s ratio between 0.1 and 0.2 and
aluminium about 0.32. Rubber has a value nearly equal to 0.5, while cork has a value that is almost zero.
Poisson’s ratio cannot be less than −1.0 or greater than 0.5. There are some materials that have negative
values of Poisson’s ratio. Such materials are called auxetic materials. Recently strange cases have been
reported.
5
There are several details concerning such experiments. We cannot discuss them here. Most of the readers,
we hope, would have performed such experiments, at least the straightforward simple ones, as part of their
earlier studies.
7-3 Constitutive Equations

(There is a theorem called the Stone-Weierstrass theorem6 that justifies this.) Next we
argue that the body is in a ‘natural state’. That is, when there is no strain, there is no
stress7 . This means that the constant a0 vanishes (a0 = 0). Thus, the above equation is
simplified to the form

σ = σ(e) = a1 e + a2 e2 + a3 e3 + · · · + an en ,

where n is usually limited to one or two. (The departure from linearity is usually only
slight; there is generally no need to take higher powers.) The consequence is that the stress
strain curve now passes through the origin.
Here we are discussing only the one-dimensional case, when both σ and e are but
numbers. Such a material law can be extended to the general three-dimensional case by
simply regarding the above as a matrix equation, when it reads as

σ = σ (ee) = a1 e + a2 e 2 + a3 e 3 + · · · + an e n .

Here σ and e are, in a three-dimensional setting, 3 × 3 matrices. There is a well known

theorem called the Cayley-Hamilton theorem8 that permits the higher degree terms to be
expressed in terms of e and e 2 . When this is done, the above equation appears as

σ = σ (ee) = c1 e + c2 e 2 .

Thus, we do not need to consider higher orders of e even in the general case. This con-
stitutive equation is nonlinear. This kind of nonlinearity is called material nonlinearity in
contrast to the other kind called geometric nonlinearity. The latter refers to the nonlinear
relationship in the strain-displacement equations.
Let us decide to consider only the simple case of linear theory9 , when the above equation
simplifies further as
σ = σ (ee) = D e , (7.2)
where σ , e , and D are matrices. We shall take Eq. (7.2) as the starting point for further
discussions.
6
Named after Marshall H. Stone (April 1903 - Jan. 1989), American mathematician and Karl Weierstrass
(Oct. 1816 - Feb. 1897), a German mathematician regarded as the father of modern rigour in mathematics.
Weierstrass’ approximation theorem — every continuous function defined on a closed interval [a, b] can be
uniformly approximated as closely as desired by a polynomial function — was generalised by Stone. He
also simplified the proof.
7
There can be situations where (i) there is stress when there is no strain, and (ii) there is no stress, there
is strain. These are, shall we say, exceptional cases that we need not take into account in our context.
8
Named after the British lawyer-mathematician Arthur Cayley (Aug. 1821 - Jan. 1895) and the Irish
mathematician and astronomer William Rowan Hamilton (Aug. 1805 - Sept. 1865). The theorem states
that every square matrix satisfies its own characteristic equation in the matrix sense. Thus, if A is a 3 × 3
matrix,
A 3 − (...) A 2 + (...) A − (...) I = 0 ,
making it possible to express A 3 in terms of A 2 , A and I .
9
Neither material nonlinearity nor geometric nonlinearity is permitted in this simpler theory.
Advanced Mechanics of Solids 7-4

Further Discussion
We begin by taking the above equation (7.2) in matrix form as
6
X
{σ} = [D] {e}, i.e., σi = Dij ej (j = 1, · · · , 6), (7.3)
j=1

where [D] is called the stiffness matrix. This may be inverted to yield
6
X
{e} = [C] {σ}, i.e., ei = Cij ej (e = 1, · · · , 6), (7.4)
j=1

where [C] is known as the compliance matrix. The matrices [C] and [D] are related to each
other as
[C] = [D]−1 and / or [D] = [C]−1 ,
which makes it appropriate to refer to [D] as the stiffness matrix. Both D and C may be
called constitutive matrices. [Some authors use other notations also, which is but natural.]
There are a number of ‘constants’ (or coefficients, or moduli) in these matrices. How many,
that is the question. We shall examine.
How Many Elastic Constants?
In the light of what we have seen so far, let us begin by assuming that each stress component
is linearly related to each of the strain components. Then we will have
    

 σxx  D11 D12 D13 · · · D18 D19   exx  
 σyy 

   D21 D22 D23 · · · D28 D29  
  eyy 
 

     
 σzz   D31 D32 D33 · · · D38 D39   ezz 

 
   
 

··· =  · · · · · · · · · · · · · · · · · · 
 · · · , (7.5)
τ D D D · · · D D e
  
yz  71 72 73 78 79   yz 

 
   
 

  
 τzx    D81 D82 D83 · · · D88 D89    ezx 

  
   

D91 D92 D93 · · · D98 D99
   
τxz exz
where there are 9 × 9 = 81 elastic constants. These are far too many; we cannot deal
with so many in the solution of any real problem. Furthermore, how do we evaluate the
numerical values of these elastic constants by experiments? We must, thus, look for reasons,
arguments and procedures to cut them down to a manageably small number. How shall we
proceed?
Let us note that both the stress and strain matrices are symmetric. Then the number
of stress and strain components will be reduced from 9 to 6. Then we will have 6 × 6 = 36
constants.
    

 σxx 
 D11 D12 D13 D14 D15 D16   exx 



 σyy 



 D21 D22 D23 D24 D25 D26  
 


 eyy 


σzz D D D D D D e
    
31 32 33 34 35 36 zz

= D41 D42 D43 D44 D45 D46   exy  ,
 (7.6)

 τxy 
    
τ  D51 D52 D53 D54 D55 D56   e 
    
 yz   yz 

   
 
τzx

D61 D62 D63 D64 D65 D66 ezx

7-5 Constitutive Equations

But this also is too large a number. Can we reduce this further? Yes, this is possible, but
only if we make further assumptions.
Further Reduction
To have further reduction in the number of elastic constants, we need to make use of the
concept of a strain energy density function, U = U(eij ). This relates the stress components
and the strain components (constitutive equation)10 by

∂U
σij = (i, j = 1, 2, 3).
∂eij

The strain energy U is the integral of the strain energy density function U integrated over
the entire volume V concerned:
ZZZ
U = U (eij ) = U(eij ) dV.
V

This strain energy function U is expressed as

3 3 3 3 3 3
1 XXXX 1 XX 1
U= Dijkl eik ekl = Dαβ eα eβ = {e}T [D] {e},
2 2 2
i=1 j=1 k=1 l=1 α=1 β=1

from which we note that

∂2U ∂2U ∂2U

= Dαβ = Dβα , since = .
∂eα ∂eβ ∂eα ∂eβ ∂eβ ∂eα

Thus, we note that the stiffness matrix (constitutive matrix) D is symmetric. The number
of elastic constants is, thus, reduced from 36 to 21. We have so far made no assumption
at all about isotropy. Hence, we conclude that the most general anisotropic material —
anisotropic model — has 21 elastic constants11 .
We have so far reduced the number of elastic constants in stages as

81 −→ 36 −→ 21.
10
This can be considered as a special case of the more general thermoelastic constitutive equation. Thermo-
dynamics also enters here. For two special cases, (i) isothermal and (ii) reversible adiabatic (isentropic)
processes, this function can be identified or interpreted as (i) Gibbs’ free energy, and (ii) the internal
energy function. To understand these we need to consult advanced books.
This strain energy function U enjoys the property of positive definiteness which has important conse-
quences. Some of them are the uniqueness theorem of linear theory of elasticity, and the principles of
minimum total potential (energy) and minimum complementary energy.
11
Lekhnitskii, S.G. Theory of Elasticity of an Anisotropic Body, Mir Publishers, Moscow, translated from
Russian, (1981), states “In (the) general case of anisotropy the number of elastic constants is 21, but
among them the independent constants are fewer.” · · · “Consequently, even in the most general case, the
number of independent elastic constants is not 21, but fewer, viz., 18.” He has more to say on this.
Advanced Mechanics of Solids 7-6

    

 σxx 
 D11 D12 D13 D14 D15 D16 
 exx 




 σyy 




 D12 D22 D23 D24 D25 D26 


 eyy 



σzz D13 D23 D33 D34 D35 D36 ezz
    
=  , (7.7)

 τxy 


 D14 D24 D34 D44 D45 D46 
 exy 

τ D15 D25 D35 D45 D55 D56 eyz
   
 yz

  
 
 

   
τzx

D16 D26 D36 D46 D56 D66

ezx


Crystals are the most important real bodies that exhibit anisotropy12 . Composite, plywood,
wood and rock have decidedly anisotropic elastic properties. Besides, metallic components
can have cylindrical anisotropy because of the technological processes of producing them
like extrusion and wire drawing. But we will not discuss such matters here.
We have seen rectilinear anisotropy. Here all parallel directions are elastically equivalent.
But this is not the case always. One case of wood is mentioned later when we discuss
transtropy. Some bodies may have curvilinear anisotropy. Among various possibilities,
Saint-Venant long, long ago had studied (i) cylindrical and (ii) spherical anisotropy. These
are of great practical interest.
The generalised Hooke’s law for a material exhibiting curvilinear anisotropy can be
written as Eq. (7.7) with reference to a rectangular cartesian coordinate system (x, y, z).
But the trouble is that the constitutive coefficients (constitutive moduli) are now not con-
stants. Their values change from point to point even if the body is homogeneous due to
changes in the coordinate directions. In a general curvilinear coordinate system (ξ, η, ζ),
the constitutive equation can be written in the form

    

 σξξ 
 D11 D12 D13 D14 D15 D16 
 eξξ 




 σ ηη





 D12 D22 D23 D24 D25 D26 


 eηη 



σζζ D13 D23 D33 D34 D35 D36 eζζ
    
=  ,

 τξη 


 D14 D24 D34 D44 D45 D46 
 eξη 

 τηζ D15 D25 D35 D45 D55 D56 eηζ

 
   

 
 
 


τζξ

D16 D26 D36 D46 D56 D66

eζξ


as before. There are 21 elastic constants13 . The above constitutive equation can be simpli-
fied exactly as in the case of rectilinear anisotropy.
One Plane of Elastic Symmetry
Further, if we assume that there is one plane of elastic symmetry, the number of elastic
constants is further reduced to 13. We shall see this reduction below.

12
It appears that all natural crystals have one or the other of 32 kinds of symmetry. Their elastic properties
can be only in one of the nine types. But we are not interested in crystals.
13
See the footnote on the previous page.
7-7 Constitutive Equations

Old coordinates
x1 x2 x3

New coordinates
x01 a11 a12 a13

x02 a21 a22 a23

x03 a31 a32 a33

The figure shows a set of (‘old’) axes (x, y, z). These axes are changed to a new set
of axes (x0 , y 0 , z 0 ) by rotating (x, y, z) through 180 ◦ about the vertical z (also z 0 ) axis.
Elastic symmetry about one plane implies that the constitutive matrix [D] will still remain
the same. Referred to the new axes, the stress and strain coordinates will obviously change.
These changes can be calculated; we have already seen such calculations when the axes are
changed from the ‘old’ (x, y, z) system to the ‘new’ (x0 , y 0 , z 0 ) one. Here these results can
be obtained merely by inspection. Alternatively, these can be calculated using the stress
and strain transformation equations. In either case, the result is

(x, y, z) −→ (x0 , y 0 , z 0 ) : [σ 0 ] = [a] [σ] [a]T ; [e0 ] = [a] [e] [a]T ;

σx0 0 x0 = σxx ; σy0 0 y0 = σyy ; σz0 0 z 0 = σzz ; τx0 0 y0 = τxy ; τy0 0 z 0 = −τyz ; τz0 0 x0 = −τzx ;
e0x0 x0 = exx ; e0y0 y0 = eyy ; e0z 0 z 0 = ezz ; e0x0 y0 = exy ; e0y0 z 0 = −eyz ; e0z 0 x0 = −ezx .

Let us realise that the constitutive matrix relating (i) σij and eij , and (ii) σi00 j 0 and e0i0 j 0 is
the same; this is what elastic symmetry (about one plane) implies. Thus, we have
    0 
σ 0 e
 xx 
 0 0  D 11 D 12 D 13 D 14 D 15 D 16 
 0
xx  0 

σ 0  
e 0 


 y 0 y0 

  D 12 D 22 D 23 D 24 D 25 D 26  

 y 0 y0 


 σ0
    
 e0 0 
D D D D D D
  
zz0 0 13 23 33 34 35 36
 z z
0 =
 D14 D24 D34 D44 D45 D46   e 0 0  ;
 0 (7.8)

 τx0 y0     x y 

 τ 0 
  D15 D25 D35 D45 D55 D56    e0y0 z 0 

 yz 
 0 0 
 
 

 τ0
  D D D D D D

 e0 0  
z 0 x0 16 26 36 46 56 66 zx
    

 σxx   D11 D12 D13 D14 D15 D16   exx  



 σ yy



 
 D12 D22 D23 D24 D25 D26  



 e yy




σzz D D D D D D e
  
13 32 33 34 35 36 zz
 
i.e., =
  ; (7.9)

 τxy    D14 D24 D34 D44 D45 D46  

 −exy  
−τyz   D15 D25 D35 D45 D55 D56   −eyz 

   

 
 
 

−τzx
 
D16 D26 D36 D46 D56 D66

ezx

    

 σ xx 
 D 11 D 12 D13 D 14 −D 15 −D 16 
 e xx 




 σ yy




 D12
 D 22 D23 D 24 −D 25 −D 26 



 e yy




σzz D D D D −D −D e
    
13 23 33 34 35 36  zz

i.e., =
 ; (7.10)

 τxy    D14 D24 D34 D44 −D45 −D46   exy  
τ    −D15 −D25 −D35 −D45 D55 D56   e 
   
 yz   yz 

  
 
τzx −D16 −D26 −D36 −D46 D56

D66 ezx

Advanced Mechanics of Solids 7-8

Comparing Eqs (7.7) and (7.10), we conclude that

D15 = D25 = D35 = D46 = D16 = D26 = D36 = D46 = 0.
The constitutive matrix for a material — really a model — with one plane of symmetry is
thereby simplified, and the constitutive equation appears as
    

 σ xx 
 D 11 D 12 D 13 D 14 0 0 
 exx 




 σ yy




 D12 D22 D23 D24
 0 0 



 e yy




σzz D D D D 0 0 e
    
13 23 33 34 zz

=  , (7.11)

 τxy   D14 D24 D34 D44 0 0 
 exy 
τ    0 0 0 0 D55 D56   e 
   
 yz   yz 

  
 
τzx

0 0 0 0 D56 D66 ezx


There are now only 13 elastic constants.

Two Planes of Elastic Symmetry
By an argument similar to the one used above, it is possible to show that there are now
nine (9) elastic constants. The constitutive equation looks as shown below.
    

 σxx  D11 D12 D13 0 0 0 
 exx 



 σ yy



 
 D12 D22 D23 0 0 0 


 e yy




σzz D D D 0 0 0 e
   
13 23 33 zz
 
=  , (7.12)

 τxy   0 0 0 D44 0 0   exy 
τ    0 0 0 0 D55 0  e 
   
 yz   yz 

  
 
τzx

0 0 0 0 0 D66 ezx


Three Planes of Symmetry: Orthotropy

Two planes of symmetry implies three planes of symmetry! A little reflection would perhaps
help us to have some kind of (intuitive?) understanding. There is a formal proof14 . This
is the case of orthotropic elasticity. The constitutive equation has the same appearance as
above; there are now nine (9) elastic constants. Orthotropy has taken on special significance
in relatively recent years when composites are used more and more.
    

 σxx  D11 D12 D13 0 0 0 
 exx 



 σ yy




 D12 D22 D23
 0 0 0 



 e yy




σzz D D D 0 0 0 e
    
13 23 33 zz

=   , (7.13)

 τxy   0 0 0 D44 0 0  exy 
τ    0 0 0 0 D55 0  e 
   
 yz   yz 

  
 
τzx

0 0 0 0 0 D66 ezx


Transverse Isotropy (Transtropy)

Consider a material at each point of which there is a principal direction of elastic symmetry
(along the z axis in the figure). In a plane (xy) normal to the principal direction (z-axis)
14
Symmetry in a general sense is a deep and rather abstract concept. These concepts are all exploited in
studying crystal symmetry. Group theory enters into these discussions. We cannot discuss all this here.
7-9 Constitutive Equations

(a) Transtropy (b) Wood

Figure 7.1: To discuss transtropy: The principal direction is the z axis. On the plane normal
to this principal direction, that is, on the xy plane (shown shaded), there is isotropy. That
is, the elastic properties are the same in all directions in this xy plane. Wood [Fig. 7.1b] is
sometimes modelled as a transtropic material. The cross-section of wood shown corresponds
to the xy plane where the elastic properties are the same in all directions. [Cylindrical polar
coordinates can be used with advantage in such cases.]

— here xy is the plane normal to the principal direction (z axis) — there is isotropy. Such
a special simplified case of anisotropy is referred to as transverse isotropy, or transtropy,
and the material is said to be transversely isotropic or transtropic.
Wood is an example of a transtropic material. It is sometimes modelled as transtropic.
The direction of the grains is the one principal direction. There are an infinite number of
principal directions in the plane normal to the direction of the grains. The material has the
same elastic properties, and hence isotropy, in all directions on the cross-section shown in
Fig. 7.1b and on the shaded cross-section xy in Fig. 7.1a. [It is advantageous to exploit
this property by using cylindrical polar coordinates in such cases.]
Now the constitutive equation can be seen to be
    

 exx 
 C11 C12 C13 0 0 0 
 σxx 




 eyy 




 C12 C11 C13 0 0 0 


 σyy 



ezz C13 C13 C33 0 0 0 σzz
    
=  .

 e xy 


 0 0 0 (C11 − C12 ) 0 0 
 τxy 

e 0 0 0 0 C44 0 τyz
   
 yz

 
  
 

   
ezx

0 0 0 0 0 C44

τzx


There are now only five (5) independent elastic ‘constants’.

This is as far as we can go unless we assume isotropy. Isotropy means that the elastic
properties are the same in all directions. We shall now demonstrate how the number of
elastic constants can be further reduced if we assume that the material is isotropic. We
shall see that there are three elastic constants now, but that even these three are not
independent. There are only two independent elastic constants if the material is isotropic.

81 −→ 36 −→ 21 −→ 13 −→ 9 −→ 5 −→ 2

In the next section, we shall discuss this most important case of isotropic materials.
Advanced Mechanics of Solids 7-10

(b) Another coordinate trans- (c) Yet another coordinate

(a) A coordinate transformation formation transformation

Figure 7.2: To discuss isotropy: three pairs of coordinate systems, and the corresponding
transformations, are considered.

ISOTROPIC MATERIALS
What does isotropy imply? Well, it means that the elastic properties are the same in all
directions at the same point. We shall discuss this in detail.
First Possibility
We may write the constitutive equation in tensor form as

σij = Dijkl ekl (7.14)

where Dijkl is a fourth order tensor. [We may write this, as some authors do, as

D = Dijkl e i e j e k e l

emphasising the difference between the tensor D and its components Dijkl .] Accordingly,
these components transform, on change of coordinates as
0
Dijkl = aip ajq akr als Dpqrs . (7.15)
0
When the material is isotropic, Dijkl = Dijkl . We shall not proceed with this any further.

Second Possibility
We shall consider five (5) sets of coordinate transformations. The basic idea is what we
have employed earlier.
Step 1:
Let us change the coordinate axes (x, y, z) to (x0 , y 0 , z 0 ), and see how (i) the stress compo-
nents, (ii) the strain components, and (iii) the constitutive relationships change [Fig. 7.2a].
We can see just by inspection or, if we are too dependent on transformation equations, the
following.

σx0 0 x0 = σxx σy0 0 y0 = σyy σz0 0 z 0 = σzz

τx0 0 y0 = τxy τy0 0 z 0 = −τyz τz0 0 x0 = −τzx
e0x0 x0 = exx e0y0 y0 = eyy e0z 0 z 0 = ezz
e0x0 y0 = exy e0y0 z 0 = −eyz e0z 0 x0 = −ezz
7-11 Constitutive Equations

    

 σxx 
 D11 D12 D13 D14 D15 D16 
 exx 




 σyy 




 D21 D22 D23 D24 D25 D26 


 eyy 



σzz D31 D32 D33 D34 D35 D36 ezz
    
=  (7.16)

 τxy 


 D14 D24 D34 D44 D45 D46 
 exy 

τ D15 D25 D35 D45 D55 D56 eyz
   
 yz

  
 
 

   
τzx

D16 D26 D36 D46 D56 D66

ezx

    

 σx0 0 x0 
 D11 D12 D13 D14 D15 D16 
 e0x0 x0  
σy0 0 y0 e0y0 y0 
   





  D21 D22 D23 D24 D25 D26 

 

 σ0
 
  
e0z 0 z 0 

z0 z0
 D31 D32 D33 D34 D35 D36 
0 = (7.17)
e0x0 y0 


 τx0 y 0 


 D14 D24 D34 D44 D45 D46 
 
τy0 0 z 0 D15 D25 D35 D45 D55 D56 e0y0 z 0 

 
   

 
 
 

 τ0
 
 D16 D26 D36 D46 D56 D66

 0
ez 0 x0 

z 0 x0

The same stiffness matrix appears in both equations (7.16 and 7.17) because the material
is isotropic. As
  0    0 
σ x0 x0  e x0 x0 
 
 σ xx  
   e xx  
 
σy0 0 y0  e0y0 y0 
       


 σ yy





 



 e yy





 

   σ0
      e0
  
σzz e
   
z 0 z 0 zz z 0 z 0
= and = ,

 τxy  
 τx0 0 y0  
 exy  
 e0x0 y0  
τ   −τy0 0 z 0  eyz  −e0y0 x0 
       
 yz 

    
  
 
       
τzx 0 0
  
e
  
 −τz 0 x0 
zx
 −ez 0 x0 
it follows, on comparing the two stiffness matrices after making the necessary changes, that
D15 = D16 = D25 = D26 = D35 = D36 = D45 = D46 = 0;
D51 = D52 = D53 = D54 = D61 = D62 = D63 = D64 = 0.
With these simplifications, the constitutive equation becomes
    

 σ xx 
 D11 D12 D13 D14 0 0 
 exx 




 σyy 


 D12 D22 D23 D24 0 0 

 eyy 


   
σzz  D13 D23 D33 D34 0 0 ezz
  
=  , (7.18)

 τxy 
 D14 D24 D34 D44
 0 0 
 exy 

τ  0 0 0 0 D55 D56 eyz
   
yz

 
 
 


   
τzx

0 0 0 0 D56 D66

ezx


Step 2:
We shall now consider another coordinate transformation. The coordinate system is rotated
about the Ox axis by 180◦ . Proceeding as before, we can conclude that
D14 = D24 = D34 = D41 = D42 = D43 = D56 = 0.
Step 3:
A rotation of the coordinate system about the Ox axis by 90◦ leads, by the same argument,
to the result
D12 = D13 ; D21 = D31 ; D22 = D33 .
Advanced Mechanics of Solids 7-12

Step 4:
A similar rotation of the coordinate system by 90◦ , but this time about the Oz axis gives
us the following simplification:

D13 = D23 ; D11 = D22 ; D44 = D55 .

The generalised Hooke’s law has now become simpler. They now read as

σxx = D11 exx + D12 (eyy + ezz ) (7.19a)

σyy = D11 eyy + D12 (ezz + exx ) (7.19b)
σzz = D11 ezz + D12 (exx + eyy ) (7.19c)
τxy = D44 exy (7.19d)
τyz = D44 eyz (7.19e)
τzx = D44 ezx (7.19f)

Step 5:
As the fifth and last step, let us rotate the coordinate system about the Oz axis through
45◦ [Fig. 7.2c]. From the stress / strain transformation laws we can work out how the ‘new’
stress / strain components are related to the ‘old’ ones. The results are:
   1 1

σ 0 0 1 0 0


 0
xx  0  2 2  σxx 

 σy0 0 y0 
  1 1
0 −1 0 0 

 σ



 
  2 2  
 yy 

 σ0
   0

0 1 0 0 0
 
σ
 
zz0 0 zz
= −1 1 0 0 0   τxy  . (7.20)
 
0
τx0 y0  0 
  2 2
  1 1
0 0 0 √2 − √2  
 
τy0 0 z 0   0 τ 
     
 yz 

   
  
 τ0
  1 1
0 0 0 0 τ

 √ √ zx
z 0 x0 2 2

This means that Dijkl is a fourth order isotropic (elasticity) tensor. We have discussed
elsewhere some features of isotropic tensors. We shall take up such an approach, and
proceed further later [p. 12-11].
   1 1

e0x0 x0  0 1 0 0


  2 2  exx 
e0y0 y0   1 1
2 0 −1 0 0 

  
 e  

 
  2  yy 

 e0
   0

0 1 0 0 0
 
ezz
  
z0 z0 = 0   exy  . (7.21)
 
1 1
e0x0 y0   −2 2 0 0 0
 

0 0 0 √12 − √12 

 0 
  0  
e y 0 x0  eyz 
  

 
  
 

 e0
  1 1
0 0 0 0

e

 √ √ zx
z 0 x0 2 2

As we have isotropy, the constitutive equations written above in the ‘old’ coordinate system
will appear in the ‘new’ also in exactly the same form. Thus, we have

σx0 0 x0 = D11 e0x0 x0 + D12 (e0y0 y0 + e0z 0 z 0 ) (7.22a)

σy0 0 y0 = D11 e0y0 y0 + D12 (e0z 0 z 0 + e0x0 x0 ) (7.22b)
σz0 0 z 0 = D11 e0z 0 z 0 + D12 (e0x0 x0 + e0y0 y0 ) (7.22c)
7-13 Constitutive Equations

τx0 0 y0 = D44 e0x0 y0 (7.22d)

τy0 0 z 0 = D44 e0y0 z 0 (7.22e)
τz0 0 x0 = D44 e0z 0 x0 . (7.22f)
If Eqs (7.20, 7.21) are substituted into Eq. (7.22a), we obtain
   
1 0 1 1 1 1
(σxx + σyy ) + τxy = σx0 x0 = D11 exx + eyy + exy + D12 exx + eyy − exy + ezz .
2 2 2 2 2
The stress components σxx and σyy are given by Eqs (7.19a, 7.19b). When these are
substituted in the above equation, we obtain
τxy = (D11 − D12 ) exy .
Comparison of this expression for τxy with Eq. (7.19d) yields
D44 = (D11 − D12 ).
Replacing the constitutive (stiffness) coefficients by the physically meaningful Lamé’s con-
stants λ and G, we arrive at the generalised Hooke’s law as given below.
σxx = (2G + λ) exx + λ(eyy + ezz ) (7.23a)
σyy = (2G + λ) eyy + λ(ezz + exx ) (7.23b)
σzz = (2G + λ) ezz + λ(exx + eyy ) (7.23c)
τxy = 2G exy (7.23d)
τyz = 2G eyz (7.23e)
τzx = 2G ezx (7.23f)
These equations may be inverted to express the strain components in terms of the stress
components as
(λ + G) λ
exx = σxx − (σyy + σzz ) (7.24a)
G(3λ + 2G) 2G(3λ + 2G)
(λ + G) λ
eyy = σyy − (σzz + σxx ) (7.24b)
G(3λ + 2G) 2G(3λ + 2G)
(λ + G) λ
ezz = σzz − (σxx + σyy ) (7.24c)
G(3λ + 2G) 2G(3λ + 2G)
1
exy = τxy (7.24d)
2G
1
eyz = τyz (7.24e)
2G
1
ezx = τzx (7.24f)
2G
We thus conclude that for isotropic materials, there are two and only two independent
elastic constants. Mathematicians sometimes prefer to use these Lamé’s constants G and λ,
while engineers generally work with the Young’s modulus E, Poisson’s ratio ν and modulus
of rigidity G. It is repeated for emphasis that other elastic constants such as bulk modulus
can be defined, but only two of them are independent.
Advanced Mechanics of Solids 7-14

Physical Significance of These Elastic Constants

The elastic constants (stiffness coefficients, compliance coefficients, together referred to as
constitutive coefficients) were introduced as some numbers. But they stand for physically
meaningful material properties. To see this, we shall apply simple loadings like (i) a simple
uniaxial stress and (ii) a simple shearing stress, and work out the corresponding strain
components. First let us apply a uniaxial stress, say, σxx = p, all the other stress components
being zero. The resulting strain components can be found out from the above equations. We
know what to expect: a strain in the x-direction only (exx = p/E and eyy = ezz = −νp/E
representing the Poisson’s effect). Next we shall apply τxy = τyx , a simple pair of ‘shear
and complementary shear’ and see that the only resulting shearing strain is exy = eyx =
τxy /(2G) = τyx /(2G), showing that this G is nothing other than the familiar modulus of
rigidity.
Constitutive Equations
These are also known as the material law, Hooke’s law, or the generalised Hooke’s law in
the restricted case of the classical linear theory of elasticity.
The nine (9) stress components (9 reduced to 6 because of symmetry) are related to the
nine (9) strain components (9 reduced to 6 because of symmetry again). In the simplest
case of infinitesimal, linear, isotropic, elastic materials (which alone we consider for the
most part here), these are given by the following equations.
Generalised Hooke’s Law
The generalised Hooke’s law is
1
exx = [σxx − ν (σyy + σzz )] , (7.25a)
E
1
eyy = [σyy − ν (σzz + σxx )] , (7.25b)
E
1
ezz = [σzz − ν (σxx + σyy )] , (7.25c)
E
τxy
γxy ≡ 2exy = 2eyx ≡ γyx = , (7.25d)
G
τyz
γyz ≡ 2eyz = 2ezy ≡ γzy = , (7.25e)
G
τzx
γzx ≡ 2ezx = 2exz ≡ γxz = . (7.25f)
G
The above equations may be inverted to give the stress components in terms of the strain
components as

νE E
σxx = e+ exx , (7.26a)
(1 + ν)(1 − 2ν) 1+ν
νE E
σyy = e+ eyy , (7.26b)
(1 + ν)(1 − 2ν) 1+ν
7-15 Constitutive Equations

νE E
σzz = e+ ezz , (7.26c)
(1 + ν)(1 − 2ν) 1+ν
τxy = Gγxy , (7.26d)
τyz = Gγyz , (7.26e)
τzx = Gγzx , (7.26f)

where e ≡ exx + eyy + ezz = eii is the volumetric strain (the first invariant of e ).
In Terms of Lamé’s Constants
It is sometimes convenient to write these in the form

σxx = λe + 2Gexx , (7.27a)

σyy = λe + 2Geyy , (7.27b)
σzz = λe + 2Gezz , (7.27c)
τxy = Gγxy = 2Gexy , (7.27d)
τyz = Gγyz = 2Geyz , (7.27e)
τzx = Gγzx = 2Gezx , (7.27f)

in terms of the Lamé’s constants λ and G, which are related to the Young’s modulus of
elasticity, E and the Poisson’s ratio, ν by the equations

νE E
λ= and G= .
(1 + ν)(1 − 2ν) 2(1 + ν)

We note that there are only two (2) independent elastic constants for a linear, elastic,
isotropic material. These are usually taken as E and ν by engineers. Applied mathemati-
cians and elasticians sometimes prefer to work in terms of the Lamé’s constants λ and G15 .
The constitutive equations, that is, the generalised Hooke’s law in this case, may be written
in index notation as
σij = λδij ekk + 2Geij , (7.28)
where δij is the Kronecker delta16 .
From Eq. (7.28) we obtain, on setting i = j

σii = (3λ + 2G) eii . (7.29)

Substituting for ekk in Eq. (7.28) from Eq. (7.29) — (eii ≡ ekk and σii ≡ σkk ) — and
solving for eij we have  
1 λ
eij = σij − δij σkk . (7.30)
2G 3λ + 2G
15
Some people use λ and µ for the Lamé’s constants. The constant, µ, of course, is the same as G, the
modulus of rigidity.
16
Cauchy had proposed this form of the generalised Hooke’s law (not in index notation, of course!) in 1822.
Advanced Mechanics of Solids 7-16

In direct notation, these equations (7.28) and (7.30) appear as

σ = λ (tr e ) I + 2G e
 
1 λ
e= σ− (tr σ ) I .
2G 3λ + 2G
These may also be written as
   
E ν E ν
σij = eij + δij ekk σ= e+ (tr e ) I
1+ν 1 − 2ν 1+ν 1 − 2ν
1+ν ν 1+ν ν
eij = σij − δij σkk e= σ − (tr σ ) I .
E E E E
The constitutive equations in index notation for an isotropic (linearly elastic) material
are displayed below in a box.

Constitutive equations:

σij = λ δij ekk + 2G eij ; and / or (7.31a)

1 λ
eij = σij − ekk δij . (7.31b)
2G 2G

See p. 12-10 for more details.

Principal Stresses and Principal Strains
We have already discussed the topic of principal stresses and the associated principal planes.
We have also seen principal strains and the corresponding principal directions. Now the
question arises: are the principal axes of stress coincident with the principal axes of strain?
The answer is: yes, they are coincident for isotropic materials.
We have established the following relationship between the components of the stress
and the strain matrices.
   
−


 σ xx 
 1 ν ν ν 0 0 0 
 exx 




 σ yy




 ν
 1 − ν ν 0 0 0 



 eyy 



σzz E ν ν 1 − ν 0 0 0 ezz
     
=   ,

 τxy   0
(1 + ν)(1 − 2ν)  0 0 1−ν 0 0   exy 

τ 0 0 0 0 1 − ν 0 eyz
   
yz

 
  
 

   
1−ν

τzx

0 0 0 0 0

ezx


This is written in a general rectangular cartesian coordinate system (x, y, z). In particular,
if the axes are taken along the principal axes of strain, we will have exx ≡ e11 , eyy ≡ e22 ,
ezz ≡ e11 . We will then obtain the corrsponding (above) stress matrix in the diagonalised
form  
σ11 0 0
σ =  0 σ22 0  .
0 0 σ33
7-17 Constitutive Equations

We know that if the stress matrix is diagonalised (into its diagonal canonical form), the
components on the leading diagonal are, indeed, the principal stresses. Thus, we note that
the principal axes of strain are automatically the principal axes of stress also.
If we carry out a similar exercise expressing the strain components in terms of the stress
components, and if we choose the axes (x, y, z) along the principal axes of stress, we can see
that the strain matrix is diagonalised too. Thus, we note that the principal axes of stress
are automatically the principal axes of strain also.
These two sets of principal axes are always the same (coincident) as long as the material
is isotropic. This may not always be the case of more general constitutive relationships.
[As a relatively simple example where these two sets of principal axes are not the same
(coincident), we can consider a material with one plane of elastic symmetry. If we go
through a similar exercise starting from a matrix equation relating the stress components
and the strain components, and if we choose the axes along the principal axes of strain, we
will find that the corresponding stress matrix is not in the diagonal canonical form. What
does this show? Well, it shows that principal axes of strain are not always the principal
axes of stress!]
NUMERICAL VALUES OF SOME IMPORTANT PROPERTIES
As engineers it is desirable for us to have an idea of the approximate values of some useful
properties of common materials. This cannot be discussed in a few lines because there is
so much to explain. But this is here for a limited purpose. With this caution and warning,
some representative values are given. For serious purposes such as for design, and for solving
real problems, these values have to be taken from reliable sources.
Here the table 7.1 gives the approximate numerical value of (a) Young’s modulus of
elasticity, E; (b) Poisson’s ratio, ν; (c) modulus of rigidity, G; (d) yield stress, σy.p. ; and
(e) ultimate stress, σult , for four common materials, (i) structural steel, (ii) medium carbon
steel, (iii) aluminium, (iv) brass, and (v) concrete (M30).

Table 7.1: Numerical values of some properties for a few common engineering materials

Material E (GPa) ν G (GPa) σ y.p. (MPa) σult (MPa)

Structural steel 200 0.30 77 275 410
Medium carbon steel 200 0.30 77 310 560
Aluminium 70 0.33 26 270 310
Brass 97 0.31 37 220 400
Concrete (M20, M30) 25-30 0.20 10.5-12.5 - 20-40
Advanced Mechanics of Solids 7-18

ANISOTROPY AND ORTHOTROPY

Materials cannot always be treated as isotropic. Some materials have intrinsically different
properties along different directions. Cold rolled copper and wood are two cases in point.
The properties are different along the directions of rolling. Similarly, the properties along
the grains are distinctly different from those perpendicular to this direction. Sometimes
the method of construction introduces anisotropy in a structure. If ribs are provided in one
direction, but not in the other, a slab or a plate, made up of essentially an isotropic material,
will exhibit anisotropy. More significantly, composites, which have in recent times become
an important structural material, are decidedly anisotropic. To deal with such materials,
we need to consider anisotropic elasticity17 also.
If the nine (9) stress components are related to the nine (9) strain components by the
linear equations eij = Cijkl σkl , we can see that there are 9 × 9 = 81 elastic constants
Cijkl (i, j, k, l = 1, 2, 3; 34 = 81). If the symmetry conditions σij = σji and eij = eji are
invoked, this number 81 reduces to 36 (6 × 6 = 36). If, furthermore, the existence of a strain
energy density function is assumed, this number 36 reduces to 21. Thus, in the general case
of anisotropy, there are 21 elastic constants.
To continue with such simplifications or reductions, if there is one plane of elastic sym-
metry, this number will be reduced further to 13. If the material has three planes of elastic
symmetry, there will be only nine (9) elastic constants. These are called orthotropic mate-
rials. (Many types of bio-membranes such as cell walls should be modelled as orthotropic.)
For transtropic materials there are five (5) elastic constants. And finally, when the material
is isotropic, this number five (5) reduces further to three (3). Even these three (3) are
not independent. Thus, we may conclude that there are just two (2) independent elastic
constants for a linear, elastic, isotropic material. These, as indicated earlier, are usually
taken as E and ν, or as the Lamé’s constants λ and G.
In this chapter on the constitutive equations, we have restricted ourselves for the most
part to isotropic materials. Most of the materials for the traditional engineering structures
could be considered as isotropic (although materials like wood, concrete slabs with stiffeners
in one, but not in both, directions, and plates with ribs running all along the length in one
direction were used in the past). We have discussed only aspect of the constitutive (material)
properties. We have not considered the physical aspects at all. These are important too for
engineers. These, we hope, will be discussed in other courses like deformation processes in
metals.

In the next chapter we shall see some important general considerations.

17
The classical book that deals with this topic is Lekhnitskii, S.G.: Theory of Elasticity of an Anisotropic
Body, Holden-Day (1963), translated from Russian.
Chapter 8

GENERAL CONSIDERATIONS

In this chapter we shall consider some useful concepts, results and theorems. They are
applicable for nearly all the topics that we discuss in this book. So far we have discussed
only the fundamentals; we have not considered the actual solution of problems in stress
analysis. We are building the necessary tools for the solution of problems.
COORDINATE SYSTEMS
It is obvious that we need a suitable coordinate system to work with. There are several
possibilities. Which system shall we use? There are some considerations that decide the
most suitable system in a given context.
Coordinate Systems
The easiest is the familiar rectangular cartesian system. This has the great advantage of
being the simplest among all the choices. We can appreciate the convenience and simplicity
of this only when we are exposed to the complications in other coordinate systems.
A suitable coordinate system for a specific problem:
The choice of a coordinate system suitable in a given context is relatively easy. In a two-
dimensional (plane) boundary value problem (b.v.p.) [Fig. 8.1a], a rectangular coordinate
system as shown in Fig. 8.1b is convenient. The axes (x, y) should be as shown to exploit
the symmetry inherent in the problem. The problem here is to find the stresses inside for
the given boundary conditions.
Let us consider two more problems that have circular symmetry: (i) a circular disc in
diametral compression [Fig. 8.2a] and a thick cylinder subjected to an internal fluid pressure.
[Fig. 8.2b]. It is obvious that a rectangular coordinate system, or a polar coordinate system
with the origin elsewhere would be a poor choice. (However, when a general theory is set
up, or a software developed, it is versatility that is of prime concern.)
Choice of a suitable coordinate system - two choices:
We have two choices for a suitable coordinate system. In one we use a coordinate system
fixed in space. In the other, we shall have a coordinate system fixed in the body and,
Advanced Mechanics of Solids 8-2

(a) A b.v.p. that has symmetry (b) A coordinate system (x, y) for the problem

Figure 8.1: A boundary value problem (b.v.p.) and a convenient coordinate system to ex-
ploit the symmetry. Choosing the origin elsewhere with inclined axes would be inconvenient.
So is a polar coordinate system here.

(a) A circular disc in compression (b) A thick cylinder subjected to a fluid pressure

Figure 8.2: Both these b.v.p. have circular symmetry. It is natural to choose a polar (r, θ)
system for both these problems with circular symmetry.

therefore, moving and deforming with the body. This second choice is called a convected
coordinate system. When we take up advanced studies, it will be necessary for us to
understand all this and use general tensor analysis. But we shall be simple minded; we will
not use convected coordinate systems at all in this book.
Coordinate systems other than the rectangular cartesian one:
We have seen that the geometry of the problem makes it necessary, or at least convenient,
to use coordinate systems other the simple rectangular cartesian system. These are more
difficult to handle because of various complicating features. A sound knowledge of general
tensor analysis is necessary to understand and deal with the complicating features.
A polar coordinate system:
Fig. 8.3 shows a polar coordinate system. The relevant base vectors are shown. Note also
the metric: it is not of the form (ds)2 = (dr)2 + (dθ)2 ; it is (ds)2 = (dr)2 + (r dθ)2 .
Some features of curvilinear coordinate systems:
Some of the features of a curvilinear system are indicated below. Readers who may have
worked only with the simple rectangular coordinate system should note the following.

(i) A (two-dimensional) polar (r, θ) coordinate system is shown in Fig. 8.3. The concentric
circular arcs r = constant, and the radial lines θ = constant are shown. These are
the curvilinear coordinate lines. A typical point P has the coordinates (x, y) and
8-3 General Considerations

(a) A polar coordinate system (b) The same polar coordinate system

Figure 8.3: A polar (r, θ) coordinate system with the relevant base vectors is shown. The
elemental area dA = J dr dθ = r dr dθ, not the more familiar dA = dx dy valid in a
rectangular cartesian system. Note also the metric; it is (ds)2 = (dr)2 + (r dθ)2 .

(r, θ) referred to the rectangular cartesian and polar coordinates, respectively. The
familiar relationships connecting the two sets of coordinates are shown later [p. 9-
14]. We can also see that the elemental area dA is given by the expressions dx dy
(rectangular cartesian), and J dr dθ (polar), respectively, where J is the Jacobian of
the transformation. A vector F — to be specific, a force vector — is shown; its
physical components Fr , Fθ along the radial and tangential directions along with the
base vectors concerned er , eθ are also shown. The elemental distance ds, we can see,
is given by (ds)2 = (dr)2 + (r dθ)2 . The base vector e θ here is not a unit vector!
(ii) The force vector F may be written as F = Fr e r + Fθ e θ where Fr , Fθ are the compo-
nents of F , and e r (radial) and e θ the base vectors. What is the unit of Fθ ? What is
the length of the base vector e θ ? Examine and find out.
(iii) If this expression is to be differentiated w.r.to a space variable, say, r, we should
realise that the base vectors themselves are not constants; they change with position.
Accordingly, this force vector is to be differentiated as indicated below:
F
∂F ∂Fr ∂eer ∂Fθ ∂eeθ
= e r + Fr + e θ + Fθ ; and
∂r ∂r ∂r ∂r ∂r
F
∂F ∂Fr ∂eer ∂Fθ ∂eeθ
= e r + Fr + e θ + Fθ .
∂θ ∂θ ∂θ ∂θ ∂θ
(iv) Note that the base vectors are all unit vectors, and are constants in both magnitudes
and directions in rectangular cartesian coordinates. Here the situation is different;
the base vectors have different magnitudes and different directions at different points
(spatial locations).
(v) In Fig. 8.3b are shown some force vectors and some force components at three points
P1 , P2 , P3 in addition to the point P in Fig. 8.3a. If the force is the same, its com-
ponents, Fr , Fθ are different. When a vector is a constant (at different points), its
components are not constant! Conversely, when the (radial and tangential) compo-
nents are constants, the vector is not a constant!
Advanced Mechanics of Solids 8-4

(vi) We further note that at the origin r = 0 where the Jacobian J = 0 (dx dy →
J dr dθ = r dr dθ; J = r = 0), the mapping is not one-to-one! The origin represented
by (x = 0, y = 0) in rectangular coordinates is mapped into (i) (r = 0, θ = 0), (ii)
(r = 0, θ = π/4), (iii) (r = 0, θ = π/2), etc. in polar coordinates!

(vii) Let us write Fr dr + Fθ dθ in imitation of the expression Fx dx + Fy dy. Physically this

latter expression may represent a virtual work. The units of the various elements are:
(i) (Fx , Fy ) (forces) N, and (ii) (dx, dy) (length) m. Now dθ is dimensionless in the
expression Fr dr + Fθ dθ, forcing Fθ , as it were, to have the unit Nm. The component
Fθ has to have the unit Nm, making it impossible for Fθ to be a physical component
of a force!
The moral of this demonstration is this: the vector component may not be the physical
component in a general curvilinear coordinate system1 ! The elemental area dA and
the elemental length ds are given by dA = r dr dθ and (ds)2 = (dr)2 + (r dθ)2 .

These are some of the complicating features that must be considered when we work with
coordinate systems other than the simple rectangular cartesian one.
SIMPLE TENSION TEST

(a) A stress-strain curve (b) A typical specimen (c) A stress-strain curve again

Figure 8.4: Shown in the figure are two stress-strain curves and a typical specimen. These
curves are obtained by performing a (uniaxial) tension test on a universal testing machine.

To solve practical engineering problems we need to obtain the solutions in terms of

numbers. For example, when we design a beam, or a column, or a shaft, we need to
know the dimensions. We depend upon experiments to know the properties such as the
ultimate stress, the yield stress and the Poisson’s ratio. Although there are several useful
and even indispensable experiments, perhaps the most fundamental or basic among them
is the simple tension test. There are several details to be understood. All we do here is
to show the strain-strain curve of a ductile material like mild steel. Further details are not
given here. Students will do well to learn at least some of the details.
1
To take this discussion further, let us rewrite the expression as Fr dr + Fθ dθ = (Fr ) dr + (Fθ /r) (r dθ) and
establish the relationship between the vector components and the physical components of a vector. But
we will stop at this, and hope that some ambitious students would inquire further into such issues.
8-5 General Considerations

SAINT-VENANT’S PRINCIPLE

(a) Loaded by P − P (b) Loaded by u.d.l. (c) Stress distribution at various sections

Figure 8.5: A bar is loaded by P − P [Fig. 8.5a]. In Fig. 8.5b is shown the same bar,
but this time loaded by a uniformly distributed load at the two ends. If these loadings
are statically equivalent, the stresses will be the same at all remote points. After a certain
distance (say, the linear dimension of the bar) the stress distribution is almost the same.
Fig. 8.5c shows how the stress distribution tends to the value σ = P/A (uniform) as we
move farther and farther from the ends.

Barre de Saint-Venant proposed in 1855 an important principle. If, in an elastic body

in equilibrium, the actual forces and moments acting on a small part of the surface are
replaced by their static equivalents, the replacement causes changes only in the immediate
vicinity of where the replacement is made (that is, there will be changes in the stress field2
only locally); there is no great difference in the stress field at any point remote from where
the replacement is made. After a short ‘Saint-Venant distance’ from each end, the stress
distribution is almost the same. This distance is, say, the linear dimension of the cross-
section in the case of the bar referred to for practical purposes.
This principle is applicable in a number of situations3 . We shall see two or three
examples to clarify the statement.
Example (i)
If in the bar AB loaded by two concentrated forces P −P [Fig. 8.5a], the forces are replaced
by a statically equivalent force system (which may include moments also) — uniformly
distributed forces at the ends [Fig. 8.5b], this replacement caused differences in the stress
distribution near the ends where the replacement is made. However, in all parts of the
body (bar) remote from where the replacement is made, that is, away from the ends, there
is no great difference in the stress field. A practical guide may be to leave a ‘Saint-Venant

2
and strain and displacement fields also (except the rigid body displacements)
3
This Saint-Venant’s principle is very useful and easy to apply in practical situations. However, there are
deeper issues. Several investigators have called attention to some of them. Fung [6] points out several of
them. He makes a general statement that “the justification of the principle is largely empirical and, as
such, its interpretation is not entirely clear.” Following some well known investigators, he points out how
Saint-Venant’s principle may be formulated with mathematical precision. Timoshenko [16] states that
“· · · vanishing of the resultant is not an adequate criterion for the degree of localisation”.
Advanced Mechanics of Solids 8-6

distance’ equal to the linear dimension (shown shaded in the figures 8.5a and 8.5b).

(a) A body: loads (a) (b) The same body: loads (b)

Figure 8.6: The applied loads on one small part (shown shaded) is replaced by a statically
equivalent system of forces. There will be some difference in the stress, strain and displace-
ment fields on a nearby cross-section like X − X, but practically no difference on remote
sections like X 0 − X 0 and X 00 − X 00 .

Den Hartog [3] states: “· · · Kelvin made use of a common-sense proposition known as
Saint-Venant’s principle, which states that: If the loading on a small part of the boundary
of an elastic system is replaced by a different loading, which is statically equivalent to the
original loading, then the stress distribution in the system will be sensibly changed only in
the neighborhood of the change; the stresses at a distance from the disturbance equal to
the size of the disturbance itself will be changed by a few per cent only.”
Consider a cantilever with an end load P . The simple strength of materials solution
gives us a normal stress and a shear stress distributions at the fixed end. Actually, however,
there is some uncertainty. It is difficult to achieve these ‘idealised distributions’. The
practical implication of this principle is that the stress distribution is unaltered because of
these uncertainties.
PRINCIPLE OF SUPERPOSITION
We had mentioned this earlier [p. 1-15]. Here we shall discuss this in a general setting. The
validity of superposition, as we had pointed out earlier, is based on the linearity of all the
governing equations. Thus, this theorem is valid only in the linear theory of elasticity. When
the more general nonlinear theory (large deformation theory, also called finite elasticity as
distinct from infinitesimal elasticity) is taken up, superposition is no more valid because
some of the governing equations are no more linear. (There can be different kinds of
nonlinearity and difficulties, but we do not discuss them in this book.)
To be specific, let us consider the all stress formulation (all the governing equations
being expressed in terms of the stress components). The compatibility equations (the
Beltrami-Michell equations) play a decisive role. The governing equations are, therefore: (i)
equations of equilibrium; (ii) equations of compatibility; and (iii) the boundary conditions.
[The constitutive equations are also to be included, but they are already used in (ii).]
Let us consider an elastic body in equilibrium and two sets of solutions. Each of them
must, and does, satisfy (i) the equations of equilibrium, (ii) the equations of compatibility
8-7 General Considerations

expressed in terms of the stresses, and (iii) the boundary conditions. That is, each of these
two sets satisfies all the governing equations and, thus, qualifies to be a solution. We need
to prove that the sum (and / or the difference) also satisfies (satisfy) all these governing
equations and, therefore, qualifies (qualify) to be a solution (solutions). We proceed to
demonstrate this.
(i) Let [(σxx )1 , (σyy )1 , · · · , (τzx )1 ] be the stress components obtained as one solution.
The surface forces and the body forces are, respectively,
ν ν ν
[(Tx )1 , (Ty )1 , (Ty )1 ] and [(Fx )1 , (Fy )2 , (Fz )3 ].

(ii) Let [(σxx )2 , (σyy )2 , · · · , (τzx )2 ] also be the stress components obtained as another so-
lution. The surface forces and the body forces are, respectively, the
ν ν ν
[(Tx )2 , (Ty )2 , (Ty )2 ] and [(Fx )2 , (Fy )2 , (Fz )2 ].

Here are two solutions. We need to show that the sum / difference also is a solution. This
is the idea of superposition. We shall show this.
As [(σxx )1 , (σyy )1 , · · · , (τzx )1 ] are the stress components obtained when the problem is
solved, these must satisfy the differential equations of equilibrium. Thus,

(∂σxx )1 (∂τyx )1 (∂τzx )1

+ + + (Fx )1 = 0;
∂x ∂y ∂z
(∂σyy )1 (∂τzy )1 (∂τxy )1
+ + + (Fy )1 = 0;
∂y ∂z ∂x
(∂σzz )1 (∂τxz )1 (∂τyz )1
+ + + (Fz )1 = 0.
∂z ∂x ∂y

For the same reason, [(σxx )2 , (σyy )2 , · · · , (τzx )2 ] also must satisfy these equations. Thus,

(∂σxx )2 (∂τyx )2 (∂τzx )2

+ + + (Fx )2 = 0;
∂x ∂y ∂z
(∂σyy )2 (∂τzy )2 (∂τxy )2
+ + + (Fy )2 = 0;
∂y ∂z ∂x
(∂σzz )2 (∂τxz )2 (∂τyz )2
+ + + (Fz )1 = 0.
∂z ∂x ∂y

It is clear that the sum / difference also satisfies the equations of equilibrium.

[(∂σxx )1 + (∂σxx )2 ] [(∂τyx )1 + (∂τyx )2 ] [(∂τzx )1 + (∂τzx )2 ]

+ + + [(Fx )1 + (Fx )2 ] = 0;
∂x ∂y ∂z
[(∂σyy )1 + (∂σyy )2 ] [(∂τzy )1 + (∂τzy )2 ] [(∂τxy )1 + (∂τxy )2 ]
+ + + [(Fy )1 + (Fy )2 ] = 0;
∂y ∂z ∂x
[(∂σzz )1 + (∂σzz )2 ] [(∂τxz )1 + (∂τxz )2 ] [(∂τyz )1 + (∂τyz )2 ]
+ + + [(Fz )1 + (Fz )2 ] = 0.
∂z ∂x ∂y
Advanced Mechanics of Solids 8-8

This is because of the linearity of the equations. Similarly, the boundary conditions also
can be written, and a similar conclusion can be arrived at. (i) Let l (σxx )1 + m (τyx )1 +
n (τzx )1 = (px )1 and (ii) l (σxx )2 +m (τyx )2 +n (τzx )2 = (px )2 be the two boundary conditions
corresponding to the two solutions. Then it follows, again from the linearity of the equations,
by adding these two equations, that the sum (iii) l[(σxx )1 + (σxx )2 ] + m[(τyx )1 + (τyx )2 ] +
n[(τzx )1 + (τzx )2 ] = [(px )1 + (px )2 ] satisfies the boundary condition for the ‘superposed’ sum
problem. We see that (i) and (ii) lead to the conclusion (on adding up / subtraction) (iii).
The third set of equations is the Beltrami-Michell equations which are the compatibility
equations (six in number) in terms of the stress components. We note that in the all stress
formulation, these compatibility equations are indeed the third set of governing equations.
These [Eqs (12.22), (12.23), p. 12-17, and their companion equations] are also linear equa-
tions and, therefore, superposition does hold. The same argument can be used for all such
linear governing equations. The conclusion is that the sum (as well as the difference) is also
a solution.
There is, however, one major restriction. The result is true only if the deformations are
small, so that the geometry is not affected when the second set of loads is applied. This is
often, but not always, the case. An important exception arises in the case of beam-columns.
This fact is pointed out at the appropriate place.
UNIQUENESS THEOREM
We shall now examine if a stress analysis problem can have more than one solution. We
shall that this is impossible; the solution is unique. This was established by Kirchhoff in
1859, and is known as the uniqueness theorem. For a linear elastic solid with a positive
definite strain energy function, there is always a one-to-one correspondence between the
forces acting on the body and the resulting elastic deformation.
If possible4 , let there be two solutions5 represented by
(i) (σxx )1 , (τyx )1 , · · · , (σzz )1 ); (exx )1 , (eyx)1 , · · · , (ezz )1 ); (u)1 , (v)1 , (w)1 ; and
(ii) (σxx )2 , (τyx )2 , · · · , (σzz )2 ); (exx )2 , (eyx)2 , · · · , (ezz )2 ); (u)2 , (v)2 , (w)2
for the same problem with the same (a) fixity conditions, (b) surface tractions, and (c)
displacements specified on the boundary
(ν) (ν) (ν)
Fx , Fy , Fz T x, T y , T z u, v, w.
(i) (a) The differential equations of equilibrium, (b) the stress-strain relations, (c) the strain-
displacement relations, (d) the compatibility conditions, and (e) the boundary conditions
will be satisfied by the first set (i). Thus,
∂(σxx )1 ∂(τyx )1 ∂(τzx )1
(a) + + + Fx = 0;
∂x ∂y ∂z
4
We are trying to prove that this is not possible.
5
These two solutions can possibly differ by rigid body displacements that do not affect the deformations
and the stresses.
8-9 General Considerations

··· ··· ··· ··· ;

··· ··· ··· ··· ;
··· ··· ··· ··· ;
(ν)
(e) l(σxx )1 + m(τyx )1 + n(τzx )1 = T x ;
··· ··· ··· ··· .

(ii) All these equations are satisfied by the stress, strain and displacement components of
the second solution also. Thus,

∂(σxx )2 ∂(τyx )2 ∂(τzx )2

(a) + + + Fx = 0;
∂x ∂y ∂z
··· ··· ··· ··· ;
··· ··· ··· ··· ;
··· ··· ··· ··· ;
(ν)
(e) l(σxx )2 + m(τyx )2 + n(τzx )2 = T x ;
··· ··· ··· ··· .

Using superposition, by subtraction, we find that

(ν) (ν) (ν)

Fx = Fy = Fz = 0 Tx= Ty= Tz=0 (u1 − u2 ), (v1 − v2 ), (w1 − w2 )

also satisfy all the governing equations. Thus,

∂[(σxx )1 − (σxx )2 ] ∂[(τyx )1 − (τyx )2 ] ∂[(τzx )1 − (τzx )2 ]

(a) + + = [Fx − Fx ] = 0;
∂x ∂y ∂z
··· ··· ··· ··· ;
··· ··· ··· ··· ;
··· ··· ··· ··· ;
(ν) (ν)
(e) l[(σxx )1 − (σxx )2 ] + m[(τyx )1 − (τyx )2 ] + n[(τzx )1 − (τzx )2 ] = [ T x − T x ] = 0;
··· ··· ··· ··· .

Here is a new stress distribution — the difference between the two solutions — that corre-
spond to zero body forces and zero surface forces (tractions)! The work done by these zero
body forces and zero surface forces is, therefore, zero. This fact leads to the conclusion
ZZZ
U dx dy dz = 0 −→ [(exx )1 − (exx )2 ] = · · · = [(ezz )1 − (ezz )2 ] = 0.

This proves that the two strain distributions and, consequently, the two stress distributions
are the same, proving the uniqueness theorem.
Advanced Mechanics of Solids 8-10

An Alternative Proof
An alternative proof, which is mathematically more appealing, is indicated below. We know
— we have seen this when we discussed constitutive equations — that the stress components
σij are related to the strain energy density function U = U(eij ) by

∂U
σij = (i, j = 1, 2, 3),
∂eij

and that the strain energy function U = U (eij )

ZZZ
U = U (eij ) = U(eij ) dV.
V

This function enjoys the property of positive definiteness which has important, very impor-
tant, consequences. We use this property here to prove the uniqueness theorem.
The differential equations of equilibrium σji,j + Fi = 0 in terms of the stress components
can be recast in terms of U as  
∂U
+ Fi = 0.
∂eij ,j
(ν)
The surface traction T i is prescribed on part of the boundary (S1 ). The displacement ui
is prescribed on the other part (S2 ) of the boundary. We know, of course, that that both
the displacement and the surface traction cannot be specified at the same point on the
boundary.
The boundary conditions on the surface S = S1 + S2 are
(ν) ∂U
(i) the traction T i = νJ on S1 ; and
∂eij
(ii) the displacements ui are prescribed on S2 .

We shall consider, as before, two strain fields (ui )1 and (ui )2 that, being two solutions if
that would be possible6 , satisfy the above equations. Let us call this ‘difference solution’ ui
as [(ui )1 − (ui )2 ] ≡ ui . Now for this ‘difference solution’ the equations of equilibrium
 
∂U
= 0; (8.1)
∂eij ,j

and the boundary conditions

(ν) ∂U
Ti= νj = 0 on S1 ; ui = 0 on S2 ; (8.2)
∂eij
6
We are going to prove that it is impossible to have two different solutions. We must realise, however,
that two solutions can possibly differ by rigid body displacements, because rigid body displacements are
related neither to strains nor to stresses.
8-11 General Considerations

are, must be, satisfied. From Eq. (8.1) we obtain

ZZZ  
∂U
ui dV = 0.
V ∂eij ,j

This, on integration by parts and changing one volume integral to a surface integral, yields
ZZ ZZZ
∂U ∂U
ui νj dS − ui,j dV = 0.
S ∂eij V ∂eij

The first, the surface integral, vanishes because of the boundary conditions. The second,
the volume integral, is
ZZZ ZZZ ZZZ ZZZ
∂U ∂U
ui,j dV = 2 eij dV = 2 U dV = 0.
V ∂eij V V ∂eij V

As U is positive definite, this cannot vanish unless U = 0 which leads to the conclusion
that (ui )1 = (ui )2 . This means that the two solutions are really the same. The uniqueness
theorem is thus proved.
We can see that it is awkward to do these manipulations without using the index
notation. Readers are advised to be convinced of this fact and to learn the index notation.
Comments
There are deeper issues involved in this. This theorem is proved only in the neighbourhood
of the natural state. If the strain energy function is not positive definite, several solutions
or a multi-valued solution may be possible. The uniqueness theorem may not hold if (i) the
strain energy function U = U (eij ) fails to be positive definite (which can happen when the
material becomes unstable when flow or yielding takes place) and (ii) when there are finite
(in contrast to infinitesimal) deformation, or when the forces are non-conservative and / or
when the forces are dependent on — functionals of — the deformation history.
There are some of these complicating features that make the theory not easy and
straightforward. In these cases, uniqueness may not necessarily be violated. Further careful
examination is needed.
When we state that the solution is unique, there can still be some limitation. Even when
the stresses are uniquely known, the displacements may not still be unique. When the body
has rigid body movements — translations along the three axes, and rigid body rotations
about the three axes — the stresses and the strains are unchanged. Thus, when the stresses
and strains are known, and we seek the displacements, we have to identify the rigid body
movements and filter them out. How do we do that? Well, first we need to understand
the problem. With this understanding, we can arrest the rigid body displacements. For
example, in a two-dimensional problem, we can specify that one point of the body is fixed;
that is, u(0, 0) = v(0, 0) = 0. Now the translation is arrested; the body cannot move in the
x and y directions. It can still rotate. This too is to be arrested. There are several ways.
One method is to specify that an element along the x direction does not rotate. Another is
Advanced Mechanics of Solids 8-12

to arrest the rotation about the y axis. In two dimensions, this is sufficient. If the physical
problem is well understood, there will be no difficulty.
On the mathematical side, this situation can manifest in this way. Imagine that we are
trying to determine the displacements in a finite element formulation. Now the determinant
may vanish; the (linear) algebraic equations may not be linearly independent. How do we
deal with this situation? It is, first, by understanding the physical problem and, then, by
deleting the linearly dependent rows (and columns).
SOME MISCELLANEOUS TOPICS
We shall include here some important topics that could not be included anywhere else.
These topics are somewhat jumbled; there is often no logical order. We nevertheless hope
that these remarks will enrich our readers’ understanding.
Role of Thermodynamics in the Mechanics of Solids
A discussion on the mechanics of solids will be incomplete if no reference at all is made
to the role of thermodynamics. Classical thermodynamics imposes certain restrictions on
the mechanics of solids. Some information relevant to our subject can be deduced from
thermodynamical restrictions7 .
Unfortunately, not much can be gained if the discussions are held at such an elementary
level as can be permitted in a first level course. Thermodynamics is a major branch of science
in its own right. In relatively recent years, axiomatic thermodynamics and irreversible
thermodynamics have appeared. These have proved to be of great help and a great step
forward. However, these cannot be discussed here.
The existence of a strain energy density function can be established by thermodynamic
arguments. For two simple processes, isothermal and adiabatic, this is easier. What this
means is that the stress-strain equation (constitutive equation, or material law), can be
written with no reference to the temperature. Physically speaking, sacrificing some technical
correctness for the sake of intuitive understanding, if the loading (straining) process is very
slow, there would be sufficient time for heat transfer to occur, and for thermal equilibrium to
be established; thus, nearly isothermal conditions would prevail. On the other hand, if the
loading (straining) process is very fast, there would be little time for heat transfer to occur,
and we would then have nearly adiabatic conditions. For the in-between cases between these
two extremes of extremely slow and extremely rapid loading, non-isothermal, non-adiabatic
conditions would prevail. For such realistic cases of loading as occur in practice, the situation
is more complex. It is also possible to obtain relationships among the specific heats, the
modulus of elasticity, latent heats of change of stress and strain at constant temperature,
etc. However, for the cases of practical engineering situations, the thermal effects can
7
Perhaps it is well to point out that classical thermodynamics refers to equilibrium conditions. Thus,
thermostatics would be a technically more correct name for this branch of science. Somehow, thermostatics
continues to be called erroneously as thermodynamics. We may recall that statics deals with equilibrium
conditions, while non-equilibrium conditions fall in the domain of dynamics. We may consider calling this
equilibrium thermodynamics in spite of the obvious self-contradiction, if we wish to continue to use the
traditional time-honoured name without sacrificing technical correctness.
8-13 General Considerations

be almost always disregarded, and not much will be lost if we disregard thermodynamics
altogether. This is why thermodynamics is hardly ever mentioned in books of mechanics of
solids (except, of course, in the advanced books).
However, when dealing with plasticity, viscoelasticity, etc., dissipation of energy cannot
be neglected at all. Thus, thermodynamics and even irreversible thermodynamics would
now enter with greater relevance.
Even in the case of elasticity, thermodynamics is important in theoretical considerations.
Thermodynamic arguments can be used to establish the inequalities λ > 0; G > 0; (λ, G are
the Lamé’s constants; these differ slightly, but not sufficiently to be significant in practical
cases, under isothermal and adiabatic conditions); E > 0 and −1 < ν < 0.5. Real materials
with negative values of ν are not of concern to us here in this book8 . Some materials have
such low values of ν that they are sometimes considered as zero for simplifying calculations.
More useful for theoretical considerations is the observation that the strain energy func-
tion is positive definite. From this observation follow important conclusions such as (i) the
Saint-Venant’s principle (with some restrictions), (ii) the uniqueness theorem (meaning that
there is only one solution), (iii) the minimum total potential (energy) theorem, and (iv) the
minimum complementary energy theorem.
Each of the topics mentioned above requires considerable amplification and explanation.
However, these can hardly be attempted here. The limited purpose of this small section
is to draw attention to the fact that the role of thermodynamics cannot be neglected in
advanced studies in the mechanics of solids.
SOME ELEMENTARY CONCEPTS REVISITED
In this section we shall consider a couple of elementary concepts in the nature of a revisit. It
is always a good idea to have a few quiet moments to reflect on ideas and concepts that we
may have unwittingly taken for granted. Sometimes there are tricky situations lurking in
the background. A closer second look sometimes reveals to us some of them. The reward is
in the form of fresh insights and conceptual clarity of fundamentals, and is certainly worth
the effort.
We shall examine if the simple formula σ = P/A, perhaps the first formula (that some,
perhaps most, of us may have studied in our first course in strength of materials or mechanics
of solids) has aspects that lie hidden.
Stress σ = P/A
P/A?
Let us review a simple problem, perhaps the simplest problem that we have studied in the
spirit of a revisit.
Definition of the problem:
Consider a uniform bar loaded axially by two tensile loads P − P [Fig. 8.7a]. We know, we
have learned this in our first course in the mechanics of solids, that the stress is uniform9

8
In relatively recent years, however, there have been references to strange situations.
9
at all sections sufficiently far from the ends. Near the ends the stress will not be uniform.
Advanced Mechanics of Solids 8-14

(a) A uniform bar in tension (b) Several stress distributions (c) Deformed shapes

Figure 8.7: The stress can be seen to be uniform given by σ = P/A, but this conclusion
can be arrived at only after considering the compatibility of displacements also.

and is given by σ = P/A. This answer is correct, but there is a conceptual error in its
derivation as given in many books. Apparently we can arrive at the correct result using
only the equation of equilibrium, viz., σ × A = P −→ P = σ/A. One gets the impression
that this is a statically determinate problem, as the correct answer was obtained by invoking
only the equation of equilibrium.
Is this a statically determinate problem?
But is this true? No, of course, not. All problems in the mechanics of solids / the theory of
elasticity are statically indeterminate internally.
A few stress distributions are shown in Fig. 8.7b. All of them satisfy the equation of
equilibrium as long as the resultant of each of the stress distributions is equal to P . Among
them only one can be correct. Why? Because the uniqueness theorem permits only one
solution. Which is that one correct stress distribution?
Compatibility of displacements:
To answer this question, we have to consider the deformed shapes and examine which
corresponds to a compatible displacement field. We can see almost immediately that only
the uniform stress distribution corresponds to a strain distribution that corresponds to a
displacement field that is compatible. Thus, among these various possibilities shown in Fig.
8.7b — there can be several other possibilities also — which can be regarded as eligible
or qualified candidates, the only successful one is that shown in Fig. 8.7a (uniform stress
distribution)! We conclude that the answer is indeed correct. But we are trying to learn
with emphasis on strong conceptual understanding of the fundamentals. Thus, we must be
aware of the requirement of compatibility of displacements also.
An assumption is sometimes effective:
Engineers often employ an assumption to simplify the problem. In this case, if we assume
that the stress distribution is uniform across the section, we can obtain the correct answer
using only the equation of equilibrium. Other examples are the calculation of the tangential
(hoop) stresses in a thin cylinder subjected to an internal pressure, or more generally, in
the calculation of membrane stresses in shells.
8-15 General Considerations

A Simple Lever
Next, we shall consider a simple lever and look at the equation of equilibrium. This may
also give us fresh insights. Let us examine. Let us consider a simple lever AOB supported

(a) A simple lever AOB (b) Some interesting relationships

Figure 8.8: A kinematically admissible virtual displacement field is imposed [Fig. 8.8a].
The principle of virtual work may be written. In Fig. 8.8b some interesting relationships
are shown.

at the fulcrum O, and acted upon by two vertical forces F1 and F2 at the two ends. We
know the condition for static equilibrium: the net moment at the point O = 0.
Equilibrium: net moment at the point O = F1 l1 − F2 l2 = 0.
If a virtual displacement dθ, consistent with the constraint conditions of the problem (tech-
nically stated as a kinematically admissible displacement field) is given, the system moves
to the configuration (position) shown by the dotted line. From the geometry, we note that
θ = ∆1 /l1 = ∆2 /l2 .
∆1 ∆2
kinematics / geometry: θ = = .
l1 l2
The bar AOB is assumed to be rigid. That is, the possible bending of the bar is not ad-
missible. By the principle of virtual work
principle of virtual work: −F1 ∆1 + F2 ∆2 = 0.
These three equations are displayed below. Let us not fail to notice the role of each one
of these three equations.

equilibrium: F1 l1 − F2 l2 = 0; (8.3a)
∆1 ∆2
kinematics: θ= = ; (8.3b)
l1 l2
virtual work: −F1 ∆1 + F2 ∆2 = 0. (8.3c)

We can see that (8.3a) + (8.3b) −→ (8.3c), and that (8.3c) + (8.3b) −→ (8.3a).
What this means is that the principle of virtual work is a necessary and sufficient condition
for the equilibrium of a system. It is emphasised that the virtual displacement field shall
be kinematically admissible. [The result of this simple demonstration — that the principle
of virtual work is a necessary and sufficient condition for equilibrium — carries over to the
more important and relevant case of deformable solids also.]
Advanced Mechanics of Solids 8-16

Let us observe the following relationships.

   
∆1 −l1
= θ; i.e., {∆} = L θ (kinematics); and (8.4)
∆2 l2
 
−F1
; i.e., {F } = LT M

M = −l1 l2 (kinetics / equilibrium). (8.5)
F2
Referring to Fig. 8.8b, let us note:
e = L u (strain-displacement); σ = D e (material law); LT σ = f (equilibrium).
The same operator in two guises L and LT appear in two apparently unrelated roles. At
first it may appear that kinematics and kinetics / equilibrium are two entirely indepen-
dent domains. But no; the same operator appears in both10 . Nature seems to have some
undercurrents or interconnections that are revealed only on close examination!
FROM THEORY TO PRACTICE
We have by now been exposed to certain aspects of the mechanics of solids. These should
serve to give us an enlightened view of the practical aspects of design and construction. If
the science of stress analysis is digested and absorbed, we would become better engineers. A
couple of examples are given below to illustrate how a sound theoretical training is helpful
in the practice of engineering profession.
Thin Walled Structures: Buckling
Whenever we deal with a thin structural member an alarm must ring inside: beware of the
possibility of buckling! Thin walled members and buckling are associated ideas11 . There are
examples in industries, particularly in chemical engineering, where large cylinders containing
fluids are required to boil at low temperatures. Boiling at low temperatures is possible
only if the pressure is low. When, therefore, there is (partial) vacuum inside the cylinder,
the atmospheric pressure can cause buckling, particularly if the wall thickness becomes
effectively less as a result of prolonged exposure to harmful chemicals12 .
Stress Concentration
Stress concentration is the reason for failure in several practical problems. After learning
the mathematical theory, an engineer can safely forget all the mathematics and even the
formulae. He must, however, have a qualitative understanding of where stress concentration
plays a decisive role. In this context, engineers will do well to distinguish between harmful
and harmless corners. Shall we call them malignant and benign corners? Figs 8.9a and
8.9b show these. A practical example is shown in Fig. 8.10 where the corners C can be the
sources of cracks.
10
It was Dr Gangan Prathap who gave me these insights. This help is gratefully acknowledged.
11
like the problem of ragging in the season of admission of new students every year!
12
This author, having been the consultant, is personally aware of one such massive collapse of a large
cylindrical vessel. On measurement it was seen that the actual wall thickness had become 2.8 mm compared
to the design value of 8.00 mm! When the cylinder buckled, it appeared that all hell had broken loose.
Further details are not given here to protect the privacy.
8-17 General Considerations

(a) A harmless corner (b) A harmful corner

Figure 8.9: Reentrant corners are harmful; provide generous fillets at such corners.

(a) An unsatisfactory design (b) Stress reversal cycles (c) An improved design

Figure 8.10: Vibration leads to stress reversal cycles. A point like C, where there is a high
stress concentration, can be a potential source of trouble. Cracks are possible. To solve
this problem all we have to do is to understand the problem clearly and to reduce the stress
concentration at the reentrant corners by providing generous fillets.

The figure shown [Fig. 8.10] refers to a bus. When the bus runs, particularly on rough
roads, the body vibrates. This vibration causes stress reversals: tension to compression and
back several times. This situation can lead to fatigue. A reentrant corner like C where there
is stress concentration can be the source of cracks [Fig. 8.10a]. The solution is very simple:
understand the problem and reduce the stress concentration at the reentrant corners by
providing generous fillets [Fig. 8.10c]!
Crack in an R.C.C. Roof Slab
When a crack is noticed, or even suspected, the intuitive solution that occurs to untrained
minds may be to provide a prop. But we engineers know that if this is attempted, a moment
of the opposite sign would occur at the point of support. If such a measure is adopted, a
crack which was only a suspicion will now be a confirmed reality! We engineers should not
make such mistakes.
After a (RCC) shell is constructed, the supports must be removed. When doing so,
several moments are called into play. This must be done strictly as directed by the designer,
because several moments are called into play during this removal process13 .
13
There was such a failure years ago in Chennai. A detailed enquiry revealed that the supports were removed
without complying with the designer’s clearly written directions.
Advanced Mechanics of Solids 8-18

Why Should We Calculate the (Small) Displacements?

Why do we attach so much importance to the calculation of displacements? For example,
what does it matter whether the deflection at the end of a long structural member is two
(2) mm or three (3) mm? Well, in some cases, it does not matter at all, but there are cases
when it does matter. For example, in the analysis of statically indeterminate cases, the
(‘small’) displacements lead to ‘large’ changes in the forces (and stresses).
Fracture Mechanics

(a) Stress near a crack tip (b) A progressing crack

√
Figure 8.11: The stress at a point P near a crack tip C is proportional to 1/ r. Fig. 8.11b
shows a crack progressing. The tip is progressing from C to C1 and further to C2 .

Fracture mechanics is a relatively new branch that has come of age, and is now repre-
sented by a large body of literature. This has introduced new notions of the basis of design.
The question that we were used to asking was this: what is the maximum safe load that
can be applied so that the stress does not cross its safe limit of so much? We know that
there are always micro-cracks in all real bodies. Fracture mechanics has shown that the
√
stress near a crack tip is proportional to 1/ r where r is the distance of the point from
the crack tip [Fig. 8.11a]. Thus, there is a singularity at the crack tip. The consequence is
that no matter how small the applied load is, the stress will be infinitely high! In practice,
however, infinitely high stresses cannot be developed, because the material would yield.
Thus, near a crack tip there is a (small) region of plastic zone. The region in the immediate
neighbourhood of a crack tip has assumed importance.
This situation makes it necessary for us to revise our notion of a safe design, and to
ask a different question. Concepts like stress intensity factors (SIF) and crack opening
displacement (COD) have assumed importance. We are also persuaded to classify cracks as
harmless (benign?) and harmful (malignant?). If the crack opens by a small distance, there
is some surface energy released. If this is more than the energy necessary to drive the crack
further, then there could be a progressive catastrophic failure. On the other hand, if the
energy released is less than that necessary to drive the crack forward, then there is perhaps
no great harm if the existing crack sleeps there harmlessly! The so-called J-integral is very
important in fracture mechanics, but we cannot discuss these advanced concepts here.

In the next chapter, we shall take up two-dimensional problems.

Chapter 9

TWO-DIMENSIONAL
PROBLEMS

The general three-dimensional problems are generally very difficult to solve. Hence sim-
plifications are made so that the problems may be treated as two-dimensional. These are
approximations, but if the physical facts of the problems justify the simplifying assump-
tions, the error will be small1 . The two main two-dimensional simplifications are (a) plane
stress and (b) plane strain conditions (or assumptions)2 .
Plane Stress Condition
If the body is very thin, and if the loading is only in the plane of the thin body [Fig. 9.1a],
plane stress conditions prevail. It is clear that on the two flat (stress-free) end surfaces
(xy plane) σzz = τzx = τzy = 0. Now if the thickness is small, we may assume that these
stress components are zero not only on the free surfaces at the ends, but also everywhere
inside. Actually they may not be strictly zero, but they will be small. Why? Because on
the left flat surface they are zero, and they can build up only slowly and gradually as we
1
The error is estimated when the approximate answer so obtained is compared to (i) the exact solution, if
it is known, (ii) the experimental results, and (iii) the numerical solutions obtained by the finite element
method. Theoretical methods of analysis and approximation theory also help us in assessing the goodness
of approximations. We should realise that the solution is not just one number. Thus, comparison with
the exact solution is not as simple as we might imagine. These considerations can be quite abstract.
Caution: Do not give an inferior status to approximate solutions. These may be approximate in the sense
that the governing mathematical equations are not satisfied exactly at all points. However, the so-called
exact solutions are almost always of an approximate problem where all kinds of simplifying assumptions
are made. Approximate solutions are, on the other hand, (from the physical point of view) of a more
realistically formulated problem. Thus, these may often be superior to the exact ones.
2
Some people consider generalised plane stress as another approximation. We do not discuss this. Great
advances have been made in the relatively recent past for two-dimensional problems. Several mathematical
techniques that are thus used cannot be extended to three-dimensions. Complex variable theory, including
conformal mapping, so effectively used by the great Russian elastician N.I. Mushkhelshvili — Some Basic
Problems in the Mathematical Theory of Elasticity, P. Noordhoff, Groningen (1953); this book is a marvel
— and his group is obviously not applicable for three-dimensional problems.
Advanced Mechanics of Solids 9-2

(a) A thin plate with only in-plane loading (b) σzz = τzx = τzy on the two flat end surfaces

Figure 9.1: Fig. 9.1a shows a thin plate with only in-plane loading. The stress components
σzz = τzx = τzy are all zero on the two flat (stress-free) end surfaces. They may be regarded
as very small inside the body also. Thus, plane stress conditions prevail for this case.

move inside (as demanded by continuity requirements), and then they would have to be
zero again as we move to the right and reach the right flat free surface. Thus, only three
stress components need be considered, because the others are all negligibly small.
   
σxx τxy τxz σxx τxy 0
   
 τyx σyy τyz  −→  τyx σyy 0 .
   
   
τzx τzy σzz 0 0 0
Note that the stress matrix has effectively become of size 2 × 2. Note further that ezz 6= 0.
1 ν
ezz = [σzz − ν(σxx + σyy )] = − (σxx + σyy ) 6= 0.
E E

Plane Strain Condition

(a) An infinitely long body (b) An infinitely long body, a dam

Figure 9.2: Two examples where plane strain conditions prevail. The bodies are ‘infinitely’
long — sufficiently long — and all cross-sections identical and identically loaded.

Let us now go to the other extreme, as it were: the body is infinitely long, and the
geometry, the material, and the loadings are identical on all cross-sections. Figs 9.2a, 9.2b
9-3 Two-dimensional Problems

are two such examples of a long body with identical cross-sections and loadings all along the
length which is infinite (sufficiently long). The cross-sections (except possibly at the ends)
being identical, the strain ezz can be regarded as zero. The strain matrix now becomes
   
exx exy exz exx exy 0
   
eyx eyy eyz  −→ eyx eyy 0 .
   
   
ezx ezy ezz 0 0 0
Note that the strain matrix has effectively become of size 2 × 2. Note further that σzz 6= 0.
Why is this so? Because
1
ezz = [σzz − ν(σxx + σyy )] = 0 −→ σzz = ν(σxx + σyy ) 6= 0.
E
For both these simplifications, it is sufficient to determine the stress components σxx , σyy ,
and τxy = τyx . We can show that, if one of these cases is solved, then the other too is solved
by making a few changes as shown below.
E ν
plane stress solution −→ plane strain solution: E ; −→ ν −→ ;
1 − ν2 1−ν
E(1 + 2ν) ν
plane strain solution −→ plane stress solution: E −→ ; ν −→ .
(1 + ν)2 1+ν

The best brains have laboured over our subject for 300 years or more. Thus, it is but
natural that great advances have already been made. Several advanced concepts, methods
and techniques have been developed and used over the years. We shall see only a few of the
simpler methods. First we shall see the use of Airy’s stress function.
AIRY’S STRESS FUNCTION IN RECTANGULAR COORDINATES
We know that there are three fundamental governing equations — the three pillars of the
theory of elasticity or the mechanics of solids. Among them the equations of equilibrium are
usually written in terms of the stress components. As all problems in the mechanics of solids
are statically indeterminate internally, these equations of equilibrium alone cannot give us
the solution. They have to be supplemented by the constitutive equations and the strain-
displacement relations. The former are in terms of the stress and strain components, the
latter in terms of the strain and displacement components. This situation makes it difficult
to handle them effectively. If we try to have an all stress formulation, the constitutive
equations can be used to write the strain components in terms of the stress components.
Now the compatibility equations step in; they prevent stress distributions (satisfying the
equations of equilibrium) that correspond to strain distributions which, in turn, lead to
incompatible (impossible) displacement distributions. Thus, when we begin with a stress
field, the compatibility condition is indeed a genuine governing equation3 .
3
See the contrasting situation: if the displacements are assumed as, for example, in the Saint-Venant’s
theory of torsion, the compatibility equations have only a passive role. That is, the compatibility conditions
are automatically satisfied, because impossible (incompatible) displacements will not be assumed.
Advanced Mechanics of Solids 9-4

Equations of Equilibrium
For the two-dimensional case, the two differential equations of equilibrium are:
∂σxx ∂τxy ∂σyy ∂τxy
with body forces: + + Fx = 0 + + Fy = 0 (9.1)
∂x ∂y ∂x ∂y
∂σxx ∂τxy ∂σyy ∂τxy
only self-weight: + =0 + + ρg = 0 (9.2)
∂x ∂y ∂y ∂x
∂σxx ∂τxy ∂σyy ∂τxy
no body force: + =0 + =0 (9.3)
∂x ∂y ∂y ∂x
There are three unknowns and two equations. In such situations, it is a usual mathematical
technique to use an auxiliary function, say, φ = φ(x, y) defined suitably so that both these
equations (of equilibrium) are automatically satisfied.
∂2φ ∂2φ ∂2φ
σxx = − ρ g y; σyy = − ρ g y; τxy = − . (9.4)
∂y 2 ∂x2 ∂x∂y
We can see by direct verification that Eqs (9.2) are satisfied. The advantage is that we need
not consider the equations of equilibrium at all as long as we use such a cleverly defined
auxiliary function φ = φ(x, y). This auxiliary function is called Airy’s stress function4 .
Compatibility Equation
The compatibility equation connecting the three strain components exx , eyy , exy = eyx is

∂ 2 exx ∂ 2 eyy ∂ 2 γxy ∂ 2 exy

 
+ = =2 . (9.5)
∂y 2 ∂x2 ∂x∂y ∂x∂y
We desire to express this (restrictive5 ) equation in terms of the stress components and,
thereby, in terms of φ. Such an equation — essentially the compatibility condition — in
terms of φ is the governing equation to be satisfied by the Airy’s stress function φ. We shall
now derive this equation, first in terms of the stress components and later in terms of φ.
We shall do this for the two cases of plane stress and plane strain.
(a) For the plane stress case:
For the plane stress case,
1 1 1 (1 + ν)
exx = [σxx − νσyy ] ; eyy = [σyy − νσxx ] ; exy = τxy = τxy .
E E 2G E
If we substitute these expressions (in terms of the stresses) in Eq. (9.5)6 we obtain
∂2 ∂2 ∂ 2 τxy
2
(σxx − νσyy ) + 2 (σyy − νσxx ) = 2(1 + ν) . (9.6)
∂y ∂x ∂x∂y
4
Named after the English mathematician and astronomer Sir George Biddel Airy (July 1801 - Jan. 1892).
5
This equation puts restrictions on possible stress fields; hence the word restrictive.
6
For plane stress problems, there are other compatibility conditions that are violated by the assumptions
made. However, it can be shown that the methods presented here will still give good results. The theory
of elasticity is full of such tricky situations. It is not possible to discuss all this here.
9-5 Two-dimensional Problems

For the case when the self-weight is the only body force, we have (on differentiating the
first of Eqs (9.2) w.r.to x, and the second w.r.to y, and adding up)

∂ 2 τxy ∂ 2 σxx ∂ 2 σyy

2 =− − .
∂x∂y ∂x2 ∂y 2
When the right hand side of Eq. (9.6) is replaced using this equation, we obtain
 2
∂2

∂
+ (σxx + σyy ) = 0.
∂x2 ∂y 2

For the general case of body forces [Eqs (9.5)], the same procedure yields
 2
∂2
  
∂ ∂Fx ∂Fy
+ (σxx + σyy ) = −(1 + ν) + . (9.7)
∂x2 ∂y 2 ∂x ∂y

This is an important result: the sum of the stresses, the first stress invariant (σxx + σyy +
σzz = σ11 + σ22 + σ33 ) satisfies the Laplace’s / Poisson’s equation!
Here we reach the conclusion: the compatibility equation expressed in terms of the
stress components is
 2
∂2
  
∂ ∂Fx ∂Fy
+ (σxx + σyy ) = −(1 + ν) + . (9.8)
∂x2 ∂y 2 ∂x ∂y

(b) For the plane strain case:

1
For the plane strain case σzz = ν(σxx +σyy ) because ezz = [σzz − ν(σxx + σyy ] = 0. Using
E
Hooke’s law
1 
(1 − ν 2 )σxx − ν(1 + ν)σyy ,

exx =
E
1 
(1 − ν 2 )σyy − ν(1 + ν)σxx ,

eyy =
E
(1 + ν)
exy = τxy ,
E
we find that the compatibility equation expressed in terms of the stress components is
 2
∂2
  
∂ 1 ∂Fx ∂Fy
+ (σxx + σyy ) = − + . (9.9)
∂x2 ∂y 2 (1 − ν) ∂x ∂y
Plane stress and plane strain cases:
On comparing Eqs (9.7) and (9.9), we find that the plane stress and plane strain problems
are similar with only minor changes. (See p. 9-3.)
For the case when the self-weight is the only body force the governing equations are the
following.
∂σxx ∂τxy
equilibrium, x direction: + = 0;
∂x ∂y
Advanced Mechanics of Solids 9-6

∂σyy ∂τxy
equilibrium, y direction: + + ρ g = 0;
∂y ∂x
 2
∂2

∂
compatibility: + (σxx + σyy ) = 0.
∂x2 ∂y 2
(ν) (ν)
boundary conditions: T x ≡ px = l σxx + m τyx ; T y ≡ py = l τxy + m σyy .

Compatibility equation in terms of the Airy’s stress function:

We shall now see the compatibility equation in terms of the Airy’s stress function φ =
φ(x, y).
(a) When there is no body force, the stress components are related to the Airy’s stress
function by
∂2φ ∂2φ ∂2φ
σxx = ; σ yy = ; τ xy = − . (9.10)
∂y 2 ∂x2 ∂x∂y

(b) When the self-weight is the only body force, the stress components are related to the
Airy’s stress function by Eq. (9.4) which is reproduced below.

∂2φ ∂2φ ∂2φ

σxx = − ρ g y; σyy = − ρ g y; τxy = − .
∂y 2 ∂x2 ∂x∂y
The compatibility condition is

∂4φ ∂4φ ∂4φ

54 φ ≡ 52 (52 φ) ≡ + 2 + = 0, (9.11)
∂x4 ∂x2 ∂y 2 ∂y 4
which is the well known biharmonic equation! [This equation plays a major role in the
theory of plates and other areas of the theory of elasticity. The Laplace’s and the bihar-
monic (double Laplacian) equations and the consequences cover a lot of ground in applied
mathematics.]
(c) More generally when the body forces are derivable from a potential, say V = V (x, y),

∂2φ ∂2φ ∂2φ

σxx − V = − ρ g y; σyy − V = − ρ g y; τxy = − .
∂y 2 ∂x2 ∂x∂y
The compatibility condition is

∂4φ ∂4φ ∂4φ ∂2V ∂2V

 
4 2 2
5 φ ≡ 5 (5 φ) ≡ + 2 + = −(1 − ν) + . (9.12)
∂x4 ∂x2 ∂y 2 ∂y 4 ∂x2 ∂y 2
An analogous equation holds for the case of plane strain.
In conclusion the biharmonic equation (with appropriate changes depending on the
context) is the compatibility condition. It is the governing equation when we use the Airy’s
stress function to solve two-dimensional problems. The boundary conditions also will have
to be written in terms of the partial derivatives of φ. We shall solve some simple problems
using the Airy’s stress function.
9-7 Two-dimensional Problems

(a) A rectangular block with its dimensions (b) Stress components on the boundary

Figure 9.3: Fig. 9.3a shows a rectangular block and Fig. 9.3b the boundary conditions.
Note that the stress components σxx = 2c, σyy = 2a, τxy = τyx = −b are all constant
everywhere. Note further that the shear stress is marked in the negative direction and that,
consequently, it is not marked as −b.

Solution of Problems Using Airy’s Stress Function (Inverse Method)

Some simple problems can be solved using the Airy’s stress function. However, this is an
inverse method. That is, we choose a function, typically a polynomial in x and y, and find
out what problem it solves. This is called an inverse method because we go backward as
it were, from the answer to the problem, instead of going forward from the problem to its
solution (answer). We shall demonstrate the method with a simple example.
A second degree polynomial:
Let us choose a second degree polynomial φ = φ(x, y) = ax2 + bxy + cy 2 . We know that
for every choice of φ, the equations of equilibrium are automatically satisfied; there is no
need to check even. Every choice of φ is fine as far as the equations of equilibrium are
concerned. But the choice of φ may not correspond to compatible displacements. The
compatibility equation is the master of the situation. Not every function φ will do; only
those satisfying the biharmonic equation (which is the compatibility equation expressed in
terms of φ) are acceptable. Here the governing equation is 54 φ = 0. Here, however, this
equation is obviously satisfied. The stress components are
∂2φ ∂2φ ∂2φ
σxx = = 2c; σyy = = 2a; τxy = − = −b.
∂y 2 ∂x2 ∂x∂y
These are all constant. If the boundary conditions are satisfied, this is a solution. The
uniqueness theorem guarantees that this is the only solution. A rectangular block with the
boundary conditions shown [Fig. 9.3] is the problem that this chosen polynomial solves.
Airy’s Stress Function — Solution by a Polynomial
Let us choose a stress function φ in the form
a b c d e
φ = φ(x, y) = x4 + x3 y + x2 y 2 + xy 3 + y4. (9.13)
4×3 3×2 2 3×2 4×3
This φ must satisfy the compatibility condition 54 = 0. Substituting φ = φ(x, y) in the
compatibility condition 54 = 0, we obtain e = −(2c + a).
Advanced Mechanics of Solids 9-8

We desire to solve this problem by using the Airy’s stress

function. However, this is an inverse method. We shall
examine some simple cases like this, and see how a given
problem can be solved by superposition.

Figure 9.4: A block We shall assume a polynomial with unknown coeffi-

with stresses applied cients, and try to determine these coefficients so that
the relevant conditions are satisfied.

The stress components are, thus,

∂2φ
σxx = = c x2 + d xy − (2c + a)y 2 ;
∂y 2
∂2φ
σxx = = c x2 + d xy − (2c + a)y 2 ;
∂y 2
∂2φ b d
τxy =− = − x2 − 2c xy − y 2 .
∂x∂y 2 2
We can choose these constants as a = b = c = e = 0; d 6= 0. Now we find — this corresponds
to the problem shown in Fig. 9.5a — that
d
σxx = d xy; σyy = 0; τxy = − . (9.14)
2

Bending of a cantilever? Comments:

Now we may wonder if this corresponds to the problem of bending of a cantilever. If a
cantilever of length l is fixed at the end x = l and loaded by an end load P [Fig. 9.5b], the
stresses are as follows.
(a) The bending stresses (i) are linearly distributed across the cross-section, and (ii) grow
linearly with the distance x starting from zero at the free end x = 0. This indeed is what
happens: the bending moment diagram shown confirms this fact.
(b) The shear stresses (cross-shear in bending) across the cross-section is parabolic with
zero at the extreme fibres and a maximum at mid-section. The distribution is the same on
all cross-sections. This is understandable; the shear force is constant along the length of
the cantilever.
(c) However, this solution [Eq. (9.14)] indicates that there is a shear stress τxy = τyx on
the edges AB and DC. There are no such shear stresses in a cantilever in bending.
Thus, the solution [Eq. (9.14)] corresponds to the problem of a cantilever except for
the shear stresses τyx on the edges AB and DC. No such shear stresses can occur because
these are free edges (free surfaces).
Now the question arises: is it possible to have another solution that gives us such a shear
stress distribution? Would it be possible to superpose this stress distribution so that the
problem of bending of a cantilever can be solved? After all, these are all linear equations,
9-9 Two-dimensional Problems

and superposition should be possible. The answer is: yes, this is possible. This technique
of superposition is often employed. We shall demonstrate this below.
Bending of a Cantilever

(a) A body with stress components (b) A cantilever with an end load P

Figure 9.5: A rectangular block subjected to some stress components. Fig. 9.5b shows a
cantilever loaded by an end load. From the solution of the problem defined by Fig. 9.5a
and superposing a shear stress, the cantilever problem can be solved.

Let us try to solve the problem of bending of a uniform cantilever ABCD loaded by an
end load P [Fig. 9.5b]. It is clear that the edges AB and DC (representing free surfaces)
are free of normal stresses σyy and shear stresses τyx (= τxy ), and that the shear stresses
τxy distributed on the end (AD) add up to the applied load P . Accordingly, to the stresses
given by Eq. (9.14), we superpose τxy = −b and choose the various constants so that all
the required conditions are satisfied. Thus,
d
σxx = d xy; σyy = 0; τxy = −b − . (9.15)
2
To ensure that the edges AB and DC are free of shear stresses, we should have
  d 2b
τxy  = τxy  = −b − c2 −→ d=− 2.
 
y=+c y=−c 2 c
Similarly, to ensure that the shear stress τxy on every cross-section adds up to the applied
load (shear force) P , we should have
Z +c Z +c  
b 2 3P
− τxy dy = b− 2 y dy = P −→ b=− .
−c −c c 4c

With these values of b and d now determined, the stress components [Eq. (9.15)] now appear
as
y2
 
3 P 3P
σxx = − 2 xy; σyy = 0; τxy = − 1− 2 . (9.16)
2c 4c c
As the second moment of area I = [1 × (2c)3 ]/12 = 2 c3 /3, these may be rewritten as
P xy P1 2
c − y2 .


σxx = − ; σyy = 0; τxy = −
I I 2
Advanced Mechanics of Solids 9-10

This solution is in perfect agreement — P x is the bending moment at any cross-section —

with the result of the Euler-Bernoulli theory of bending. The shear stress τxy at the fixed
end x = l should be distributed as demanded by this equation. In practice, it is seldom
possible to ensure this, but there is no cause for alarm. Saint-Venant’s principle can be
used, and we can regard that the stresses are still given by Eq. (9.16) at all cross-sections
a little away from the fixed end. (See also the examples worked out in a later chapter 13.)
Having obtained the stress components [Eq. (9.16)], it is now necessary to determine
the displacements. We shall illustrate the procedure for this problem.
Determination of Displacements
From the stress components [Eq. (9.16)] we can obtain the strain components, and then
the displacements by direct integration.
 
∂u σxx ∂v σxx ν P xy 1 ∂u ∂v τxy P
exx = = ; eyy = = −ν = ; exy = + = =− (c2 − y 2 ).
∂x E ∂y E EI 2 ∂y ∂y 2G 4 IG

On integration of the first two of these equations, we obtain

P xy ν P xy 2
u=− + f (y); v= + g(x) [f (y), g(x) arbitrary (unknown) functions].
2 EI 2 EI
(9.17)
If we substitute these in the third of the equations, we have

P x2 ν P y2 P
− + f 0 (y) + + g 0 (x) = − (c2 − y 2 );
2 EI 2 EI 2 IG
that is,
P x2 ν P y2
     
0 P 2 0 P 2
− + g (x) + − y + f (y) = − c .
2 EI 2 EI 2 IG 2 IG
The expressions within the three square brackets [· · · ] are, respectively, (i) a function of x
only, (ii) a function of y only, and (iii) a constant. What does this equation imply? Well,
each of them must be a constant, the equation being of the form A + B = C. Thus,

P c2 P x2 ν P y2 P y2
A+B =C =− ; g 0 (x) = + B; f 0 (y) = − + + A. (9.18)
2 IG 2 EI 2 EI 2 IG
Direct integration of the last two equations yields

ν P y3 P y3
f (y) = − + + Ay + D;
6 EI 6 EI
P x3
g(x) = + Bx + E.
6 EI
If we substitute the expressions for f (y) and g(x) in Eq. (9.17), we obtain

P x2 y P x2 y ν P y 3 P y3
u=− + f (y) = − − + + Ay + D; (9.19)
2 EI 2 EI 6 EI 6 IG
9-11 Two-dimensional Problems

ν P xy 2 ν P xy 2 P x3
v= + g(x) = + + Bx + E. (9.20)
2 EI 2 EI 6 EI
There are four (4) constants; their values are still unknown. At this stage we should
realise that the displacements can be determined only to within a rigid body movement.
Why is this so? This is because rigid body movements are not associated with strains or
stresses. Only if the rigid body movement is arrested (prevented) can the displacements be
determined uniquely. The rigid body motions are (i) translation along, say, the x direction,
(ii) translation along the y direction, and (iii) rotation in the xy plane. The three conditions
arresting these three rigid body motions and the first of Eqs (9.18) give us the necessary
four equations to determine the four (4) constants A, D, B and E. Now we shall proceed
to determine these constants.
Determination of the four constants A , D , B and E :
We can arrest the translation in both x− and y−directions by fixing any point, say, the
point O. Then,

u =0 −→ D = 0;

x=l,y=0
 P l3
v =0 −→ E=− − Bl.

x=l,y=0 6 EI
With these values, the v displacement (which is really the deflection of the central line of
the beam) [Eq. (9.20) with y = 0], is given by
 P x3 P l3
v = − − B(l − x). (9.21)

y=0 6 EI 6 EI
Again we can arrest the rotation of the beam in the xy plane by demanding either (i)
the horizontal element is fixed (at x = l, y = 0), or (ii) the vertical element is fixed (at
x = l, y = 0). These two physical conditions correspond to

∂v 
(i) =0 (no rotation of a horizontal element at the fixed end);
∂x x=l,y=0

∂u 
(ii) =0 (no rotation of a vertical element at the fixed end).
∂x x=l,y=0


If we use the first condition (i), we have from Eq. (9.21) and the first of Eq. (9.18),

P l2 P l2 P c2
B=− ; A= − .
2 EI 2 EI 2 IG
Thus, we obtain the displacement field as

P x2 y ν P y 3 P y3 P l2 P c2
 
u=− − + + − y;
2 EI 6 EI 6 IG 2 EI 2 IG
ν P xy 2 P x3 P l2 x P l3
v= + − + .
2 EI 6 EI 2 EI 3 EI
Advanced Mechanics of Solids 9-12

The deflection is obtained from this expression for v by setting y = 0 (because we are
looking for the deflection of the middle line y = 0) as
 P x3 P l2 x P l3  P l3
deflection: v  = − + −→ end deflection: v  = !
 
y=0 6 EI 2 EI 3 EI x=0,y=0 3 EI
This is the answer that we are all familiar with: the deflection at the end of a cantilever
loaded by an end load P ! A similar result can be obtained by using the condition (ii).
AIRY’S STRESS FUNCTION IN POLAR COORDINATES
Several problems of technical importance can be solved by using the polar coordinate (r, θ)
coordinate system. The first job is to obtain the governing equations in this system. These
equations are (i) the equations of equilibrium, (ii) the strain-displacement relations, and (iii)
the constitutive equations — the three pillars of elasticity theory — are these equations.
Additionally, (iv) the compatibility equation also plays a vital role7 . We shall consider them
one by one.
Equations of Equilibrium

(a) To derive the equations of equilibrium (b) To derive the strain-displacement relations

Figure 9.6: To derive (i) the equations of equilibrium, and (ii) the strain-displacement
relations

We have already derived the equations of equilibrium in the cylinder polar (r, θ, z)
system. The equations that we seek here are the two-dimensional simplification and can be
obtained from Eqs (5.16a, 5.16b, 5.16c, p. 5-14). Here, however, we shall derive the two
equations afresh.
The various stress components on an elemental block are shown [Fig. 9.6a]. We can
now write down the equations of equilibrium in both (i) the radial, and (ii) the tangential
(circumferential) directions. Summing up the forces in the radial (r) direction, we have
       
∂σrr ∂τrθ dθ dθ
σrr + dr (r + dr)dθ + τrθ + dθ dr cos − τrθ dr cos − σrr r dθ−
∂r ∂θ 2 2
7
We have seen that when we work with Airy’s stress functions, the equations of equilibrium are automati-
cally satisfied. Now we try to obtain some solutions by the inverse method. When we use an Airy’s stress
function, we begin from all possible stress fields. All these are not the correct ones; only one of them is.
Only this correct one corresponds to compatible displacements. Thus, the compatibility equation plays
the real role of a governing equation.
9-13 Two-dimensional Problems

     
∂σθθ dθ dθ
σθθ + dθ dr sin − σθθ dr sin + Fr (r dr dθ) = 0.
∂θ 2 2
The tangential stresses σθθ do contribute a little bit to the radial forces, because the former
are not quite perpendicular to the radial direction; a small angle dθ is included. Cleaning
up this, we arrive at the differential equation of equilibrium in the radial direction.
By following a similar procedure, we can obtain the companion equation in the tangential
direction also. These two equations are shown below8 .
∂σrr 1 ∂τrθ σrr − σθθ
+ + + Fr = 0; (9.22a)
∂r r ∂θ r
1 ∂σθθ ∂τrθ 2 τrθ
+ + + Fθ = 0. (9.22b)
r ∂θ ∂r r
Next, we shall see the strain displacement relations.
Strain-displacement Relations
We are trying to establish the relations connecting the radial (u) and tangential (v) dis-
placements, and the corresponding strains (err ) (radial) and (eθθ ) (tangential). We shall do
this by referring to Fig. 9.6b and working out the following expressions using geometry.
A typical line ab before deformation moves to (a0 b0 ) after deformation. The length
(a0 b0 )
can be worked out in terms of both (i) u (radial) and v (tangential) of a; and (ii)
u + (∂u/∂r) dr (radial) and v + (∂v/∂r) dr (tangential). Thus,
 2  2
0 0 2 ∂u ∂v
(a b ) = dr + dr + dr
∂r ∂r
The radial and tangential strains are expressed in terms of the displacements. This is what
we desire to obtain.
p
a0 b0 − ab (· · · )2 + (· · · )2 − dr ∂u
err = = ≈
ab dr ∂r
∂v
va = v = r dφ1 ; vc = v + dθ = r dφ2
∂θ
a0 c00 − ac 1 ∂v u ∂v 1 ∂v v
eθθ = = + ; 2 erθ ≡ γrθ = + − .
ac r ∂θ r ∂r r ∂θ r
In this way we arrive at the following strain-displacement relations.
∂u
err = (9.23a)
∂r
1 ∂v u
eθθ = + (9.23b)
r ∂θ r
∂v 1 ∂v v
2 erθ = + − (9.23c)
∂r r ∂θ r
8
See p. 13-31 also where these equations are obtained by coordinate transformation and stress transforma-
tion equations.
Advanced Mechanics of Solids 9-14

Constitutive Equations
We shall now consider the constitutive equations (generalised Hooke’s law, suitably simpli-
fied for our context of two-dimensional stress and strain fields). There is nothing to discuss
here. These are
σrr σθθ 1
err = −ν = (σrr − νσθθ ) (9.24a)
E E E
σθθ σrr 1
eθθ = −ν = (σθθ − νσrr ) (9.24b)
E E E
τrθ 2(1 + ν)
2 erθ ≡ γrθ = = τrθ . (9.24c)
G E
These relations are more conveniently written, (i) strains in terms of stresses, and (ii)
stresses in terms of strains, in matrix form as
         
 err  1 −ν 0  σrr   σrr  1 ν 0  err 
1 E
eθθ = −ν 1 0  σθθ ; and σθθ = ν 1 0  eθθ .
 E  1 − ν2 1−ν
γrθ 0 0 2(1 + ν) τrθ τrθ 0 0 γrθ
     
2

We need the compatibility equation also in terms of the polar coordinates (r, θ). We
shall obtain this as shown below.
Compatibility Equation: Biharmonic Equation in Polar Coordinates
We have seen that the compatibility equation in the context of using Airy’s stress function
is the biharmonic equation [Eq. (9.11), p. 9-6]. It is
∂4φ ∂4φ ∂4φ
 2
∂2
 2
∂ φ ∂2φ

4 ∂
5 φ≡ +2 2 2 + 4 ≡ + + 2 ≡ (52 )(52 φ) = 0.
∂x4 ∂x ∂y ∂y ∂x2 ∂y 2 ∂x2 ∂y
All we have to do is to express this equation in polar coordinates. This is done by recasting
the equation with (r, θ) as the independent variables. The usual coordinate transformation
using the chain rule of partial differentiation is the tool to be employed for this exercise.
The two systems of coordinates (x, y) and (r, θ) are related by
p
x = r cos θ; r = x2 + y 2 ;
−1 y
 
y = r sin θ; θ = tan .
x
The derivatives are calculated as
∂r x ∂θ y sin θ
= = cos θ; =− 2 =− ;
∂x r ∂x r r
∂r x ∂θ x cos θ
= = sin θ; = 2 = .
∂y r ∂y r r
The terms are calculated one by one as shown below.
∂ ∂ ∂r ∂ ∂θ ∂ sin θ ∂
= + = cos θ −
∂x ∂r ∂x ∂θ ∂x ∂r r ∂θ
9-15 Two-dimensional Problems

∂2
   
∂ ∂ sin θ ∂ ∂
= cos θ −
∂x2 ∂r ∂x r ∂θ ∂x
∂2 sin 2θ ∂ 2 sin2 θ ∂ 2 sin2 θ ∂ sin 2θ ∂
= cos2 θ 2 − + + + (9.25a)
∂r r ∂r∂θ r2 ∂θ2 r ∂r r2 ∂θ
∂2 ∂2 sin 2θ ∂ 2 cos2 θ ∂ 2 cos2 θ ∂ sin 2θ ∂
2
= sin2 θ 2 + + 2 2
+ − (9.25b)
∂y ∂r r ∂r∂θ r ∂θ r ∂r r2 ∂θ
∂2 ∂2 cos 2θ ∂ 2 sin θ cos θ ∂ 2 sin θ cos θ ∂ cos 2θ ∂
= sin θ cos θ 2 + − 2 2
− −
∂x∂y ∂r r ∂r∂θ r ∂θ r ∂r r2 ∂θ

If θ = 0, we realise that σrr = σxx , σθθ = σyy , etc., we can find how the stress components
are related to the Airy’s stress function in polar coordinates. See Eqs (9.29a, 9.29b, 9.29c)
below. The equations above become simplified to read From these equations we find, on
adding Eqs (9.25a) and (9.25b), that

∂2 ∂2 ∂2 1 ∂2
   
2 1 ∂
5 ≡ + −→ + + (9.26)
∂x2 ∂y 2 ∂r2 r ∂r r2 ∂θ2

The Laplacian operator appears, in polar coordinates (r, θ), as

∂2 1 ∂2
 
2 1 ∂
5 ≡ + + . (9.27)
∂r2 r ∂r r2 ∂θ2

[We should note carefully that it is not quite correct to use the same φ in both represen-
tations, because φ(x, y) and φ(r, θ) have different functional forms. Strictly, we must write
φ(r, θ) = φ(r(x, y), θ(x, y)) which cannot be the same as φ(x, y). However, this distinction
is important only if we use both (x, y) and (r, θ) as the independent variables at the same
time. That does not happen, so that in a practical sense it does not hurt to use φ as the
notation for Airy’s stress function in both sections of this chapter.]
The compatibility equation, which is the biharmonic equation when Airy’s stress func-
tion is used, appears in polar coordinates (r, θ) as

∂2 1 ∂2 ∂ 2 φ 1 ∂φ 1 ∂2φ
  
1 ∂
54 φ ≡ 52 52 φ ≡ + + + + = 0. (9.28)
∂r2 r ∂r r2 ∂θ2 ∂r2 r ∂r r2 ∂θ2

Airy’s Stress Function in Polar Coordinates

We know how the stress components are related to the Airy’s stress function [Eqs 9.10]. We
also know how the operators ∂ 2 /∂x2 , ∂ 2 /∂y 2 and ∂ 2 /∂x ∂y appear in the polar coordinate
system. From these we can see that the stress components are related to the Airy’s stress
function as

1 ∂φ 1 ∂2φ
σrr = + 2 ; (9.29a)
r ∂r r ∂θ2
∂2φ
σθθ = ; (9.29b)
∂r2
Advanced Mechanics of Solids 9-16

1 ∂φ 1 ∂ 2 φ
τrθ = − . (9.29c)
r2 ∂θ r ∂r∂θ
We have now set up all the equations that we need at the operational level to solve (two-
dimensional) problems using Airy’s stress function. These are Eqs (9.29a, 9.29b, 9.29c)
(stress components in terms of φ), and Eq. (9.28) (compatibility equation — biharmonic
equation). Next we shall see how problems are solved. We shall examine a few of the more
technically important problems.
Simpler Problems: Axisymmetric Cases
Problems become much simpler if there is rotational symmetry (axisymmetry). We shall
first take up these simpler cases. Now all the variables are independent of the angle θ. On
the mathematical side, the variable θ drops out, and the partial derivatives become ordinary
derivatives. The Airy’s stress function φ, now dependent only on r, is the solution of
 2  2
d4 φ 2 d3 φ 1 d2 φ

d 1 d d φ 1 dφ 1 dφ
2
+ 2
+ ≡ 4
+ 3
− 2 2
+ 3 = 0. (9.30)
dr r dr dr r dr dr r dr r dr r dr
This is a linear, variable coefficient, differential equation of order four (4). This is one of
the easier equations to solve. (See the footnote on p. 9-26.) This can first be converted,
if desired, into a linear differential equation with constant coefficients by changing the
independent variable from r to, say, s by the equation r = exp s. This is not necessary. In
any case, the general solution of Eq. (9.30) involving four arbitrary constants is
φ = A log r + B r2 log r + C r2 + D. (9.31)
This general form of solution is applicable to all (two-dimensional) problems with rotational
symmetry with no body forces.
[We can realise immediately that at r = 0, that is, at the centre there is going to be some
kind of difficulty. We could have anticipated this; the differential equation has a singular
point at r = 0. A detailed analysis is needed before further definite conclusions can be
reached. Students are advised to pay attention to such theoretical matters. It is desirable
to be aware of the qualitative theory of differential equations.] The stress components
corresponding to this general solution are given by
1 ∂φ A
σrr = = + B(1 + 2 log r) + 2C; (9.32a)
r ∂r r2
∂2φ A
σθθ = 2
= − 2 + B(3 + 2 log r) + 2C; (9.32b)
∂r r
τrθ = 0. (9.32c)
Let us choose B = 09 . Thus, we arrive at Lamé’s equations! We shall discuss how the
other constants can be determined from the boundary conditions of the problem. This is
explained in detail later (thick cylinders).
9
When we calculate the corresponding displacements, it turns out that the displacements will be multi-
valued if B 6= 0. While multi-valued displacements can be accepted when discussing dislocations (disloca-
tion theory), generally displacements have to be single-valued. Thus, B must necessarily vanish.
9-17 Two-dimensional Problems

Axisymmetric Case: Pure Bending of Curved Bars

Fig. 9.7 shows a curved beam subjected to pure

bending. The rectangular cross-section is shown
alongside.
1 ∂φ A
σrr = = + B(1 + 2 log r) + 2C;
r ∂r r2
∂2φ A
Figure 9.7: A curved beam σθθ = = − 2 + B(3 + 2 log r) + 2C;
∂r2 r
τrθ = 0.

Fig. 9.7 shows a curved beam subjected to pure bending. From our previous experience
and examination of the problem, we realise that all the cross-sections are identical and
identically loaded. Thus, this is an axisymmetric problem. Consequently, the solution
given by Eqs (9.32a, 9.32b, 9.32c) is applicable here also. All we need to do here is to
determine the constants A, B and C from the conditions of the problem. We shall show
below how this is done.
The conditions that we know are the following.
(a) The curved surfaces AB and CD are traction free; there is no normal stress on them.
(b) There is no net axial (circumferential) force on the cross-sections like AD and BC.
(c) The net (bending) moment applied on the cross-sections AD and BC is M .
(d) There is no shear stress τrθ on the boundary; that is, no tangential forces are applied
on the boundary.
These facts give us the equations needed to determine the constants. They are:
at r = ri , σrr = 0; at r = ro , σrr = 0; (9.33)
Z ro Z ro
σθθ dr = 0; σθθ r dr = −M ; (9.34)
ri ri
at r = ri , τrθ = 0; at r = ro , τrθ = 0. (9.35)

There is no radial stress σrr on the curved boundaries. This condition leads to
A
+ B(1 + 2 log ri ) + 2C = 0; (9.36a)
ri2
A
+ B(1 + 2 log ro ) + 2C = 0. (9.36b)
ro2
Whenever we use Airy’s stress functions, the condition of equilibrium is guaranteed. If
there is any net resultant axial (circumferential) force on the cross-sections AD and BC,
equilibrium will be upset. So we are sure that the condition (b) is satisfied. The moment
condition gives us the third equation. This is obtained as shown below.
Z ro Z ro 2
∂ φ
σθθ r dr = 2
r dr = −M.
ri ri ∂r
Advanced Mechanics of Solids 9-18

Integrating this by parts, we obtain

Z ro 2  ∂φ ro Z ro ∂φ
∂ φ  ∂φ ro  ro
r dr =  r − dr =  r  −  φ .
     
∂r 2 ∂r r ∂r ∂r r ri
ri i ri i

From this equation we conclude that10

 ro  ∂φ ro
φ = M,
(because  r = 0.)
   
ri ∂r ri
Substituting the expression for φ here, we obtain the third equation for the determination
of the constants A, B and C as
ro
A log + B(ro2 log ro − ri2 log ri ) + C(ro2 − ri2 ) = M. (9.37)
ri
The three equations (9.36a), (9.36b), (9.37) are used to solve for A, B, C as
4M 2 2 ro 2M 2
A=− ri ro log ; B=− (r − ri2 );
N ri N o
ro 2
 
M 2
C= [r − ri2 + 2(ro2 log ro − ri2 log ri )], where N = (ro2 − ri2 )2 − 4ro2 ri2 log .
N o ri
We can now obtain the stresses [Eqs (9.32a), (9.32b), (9.32c)] as:

4 M ro2 ri2
 
ro 2 r 2 ri
σrr = − log + ro log + ri log ;
N r2 ri ro r
 2 2 
4M ro ri ro 2 r 2 ri 2 2
σθθ = − − 2 log + ro log + ri log + ro − ri ;
N r ri ro r
τrθ = 0.

We do not discuss the consequences of the solution, plot the distribution of the stresses,
work out the corresponding displacements, etc. We shall merely point out that, if the
stresses are not distributed as given above at the ends, this will not be the exact solution.
However, for practical purposes, we can use Saint-Venant’s principle and find that a little
away from the ends, the stresses are indeed as given above.
Stress Concentration: Kirsch’s Solution
The problem of stress concentration is of the utmost technical importance in machine de-
sign11 . This is one important problem that cannot be solved by the methods used in the
10
Note that the expressions in Eqs (9.36a) and (9.36b) represent the radial stress σrr on the curved boundaries
(surfaces) AD and BC. Multiplying the first of these by ri2 and the second by ro2 , and subtracting one
from the other we obtain
 ∂φ ro
ro  
σrr ro2 − σrr ri2 = σrr  =  r = 0.

ri ∂r ri
11
This was solved by Ernst Gustav Kirsch (Sept. 1841 - Jan. 1901), German engineer and professor in 1898,
and is sometimes referred to as the Kirsch problem.
9-19 Two-dimensional Problems

(a) A plate with a small hole in uniaxial tension (b) Distribution of stress

Figure 9.8: Stress concentration around a circular hole. A large plate with a small hole in
uniaxial tension [Fig. 9.8a]. The stress distribution obtained is shown in Fig. 9.8b
.
traditional Strength of Materials approach. Solving this problem and obtaining the stress
concentration factor is of great interest to mechanical and aerospace engineers. We shall
solve this problem here using the Airy’s stress function, but we will not discuss the practical
implications of the solution further.
Consider a small circular hole in a large plate12 in uniaxial tension (uniform tensile
stress S) [Fig. 9.8a]. We know that the stress is very high in the immediate vicinity of the
hole [Fig. 9.8b]. Our interest is to solve this problem using Airy’s stress function.
We know that at locations far from the (small) hole, the stress will be uniform (Saint-
Venant’s principle). At a typical point P sufficiently far away from the hole, the stress
components are σxx = S, σyy = 0, τrθ = 0. This state of stress, when referred to the polar
coordinates (r, θ), is given by the stress transformation matrix.
     
σrr τrθ cos θ sin θ S 0 cos θ − sin θ
=
τθr σθθ − sin θ cos θ 0 cos θ sin θ cos θ
S cos2 θ
 
−S cos θ sin θ
= .
−S cos θ sin θ S sin2 θ

The stress components at P (at r = rO , large compared to the radius of the hole ri ) are
1
σrr = S cos2 θ = (S + S cos θ);
2
1
σθθ = S sin2 θ = (S − S cos θ);
2
1
τrθ = − S sin 2θ.
2
Let us break this up into two parts (a) and (b). The total solution (c) is the superposition
of the two parts: (c)=(a)+(b).
1 1 1
(a) σrr = S (b) σrr = S cos 2θ (c) σrr = (S + S cos 2θ);
2 2 2
12
This is sometimes differently worded as ‘a plate with a pin hole’ or as ‘a hole in an infinitely wide plate’.
Advanced Mechanics of Solids 9-20

1 1 1
σθθ = S σθθ = S cos 2θ σθθ = (S + S cos 2θ);
2 2 2
1 1 1
τrθ = S τrθ = S cos 2θ τrθ = (S + S cos 2θ).
2 2 2
Part (a) is the Lamé’s thick cylinder problem. We shall return to this after we solve the
Part (b) problem.
Part (b) problem:
Assume an Airy’s stress function in the form φ = φ(r, θ) = f (r) cos 2θ, where f (r, θ)
is unknown. We can determine it by first substituting this function in the biharmonic
equation, thereby obtaining an ordinary differential equation. On solving this, we obtain
the general solution. As usual, the boundary conditions are used to determine the four
arbitrary constants . The Airy’s stress function is now known, and the expressions for the
stress components are worked out to obtain the solution. This is the procedure; we shall
follow this.
Substituting φ = f (r) cos 2θ in the biharmonic equation, we obtain a fourth order,
linear, variable coefficient, ordinary differential equation.
 2  2 
d 1 d 4 d f 1 df 4f
52 52 φ = 0 −→ + − + − = 0.
dr2 r dr r2 dr2 r dr r2
This equation may be solved directly, or by first converting this into a differential equation
with constant coefficients. Either way, the general solution is seen to be
 
2 4 C 2 4 C
f (r) = A r + B r + 2 + D −→ φ(r, θ) = A r + B r + 2 + D cos 2θ.
r r
The stress components are readily worked out as
1 ∂2φ
 
1 ∂φ 6C 4D
σrr = + 2 = − 2A + 4 + 2 cos 2θ;
r ∂r r ∂θ2 r r
2
 
∂ φ 6C
σθθ = 2
= − 2A + 12Br2 + 4 cos 2θ;
∂r r
   
∂ 1 ∂φ 2 6C 2D
τrθ = − = 2A + 6Br − 4 − 2 sin 2θ.
∂r r ∂θ r r
The boundary conditions used to determine the constants A, B, C, D are:
(a) at r = ri (‘inner’ radius ri , radius of the hole), the radial stress, σrr = 0;
(b) at r = ri (‘inner’ radius ri , radius of the hole), the shear stress, τrθ = 0;
(c) at r = ro (at a sufficiently large distance), the radial stress, σrr = (1/2) S cos 2θ; and
(d) at r = r0 (at a sufficiently large distance), the shear stress, τrθ = −(1/2) S sin 2θ.
These conditions lead to the following four equations.
6C 4D
(a) 2A + + 2 = 0;
ri4 ri
6C 2D
(b) 2A + 6Bri2 − 4 − 2 = 0;
ri ri
9-21 Two-dimensional Problems

6C 4D S
(c) 2A +2
+ 2 =− ;
ro ro 2
6C 2D S
(d) 2A + 6Bro2 − 4 − 2 = − .
ro ri 2

These equations enable us to solve for the four constants under the condition that the hole
is small (or, equivalently, that the plate is very large), i.e., ri /ro → 0.
Multiplying the third equation (c) by ri2 using the condition ri /ro → 0, we obtain
A = −S/4. Similarly multiplying the last equation (d) by ri2 /ro2 and letting ri /ro → 0,
we conclude that B = 0. Now we obtain from the first two equations (a) and (b) the results
for the constants C and D as C = −(ri4 S)/4 and D = (ri2 S)/2.
With these values of the constants A, B, C, D, the stress components corresponding to
case (a) are:

3ri4 4ri2
 
S
σrr = 1 + 4 − 2 cos 2θ;
2 r r
4
 
S 3r
σθθ = − 1 + 4i cos 2θ;
2 r
3ri4 2ri2
 
S
τrθ = − 1 − 4 + 2 sin 2θ.
2 r r

This completes the solution corresponding to case (b). Now we shall take up the case (a).
Part (a) problem:
This is the Lamé’s problem discussed elsewhere in detail. The present problem here corre-
sponds to a thick cylinder of radii ro (outer radius) and ri (inner radius) subjected to an
external hydrostatic pressure po and no internal pressure pi = 0. This solution is worked
out in detail (p. 9-31). Borrowing the result from there, we obtain the expressions for the
stress components as

S ri2 ro2 S ro2

σrr = − + ;
2(ro2 − ri2 )r2 2(ro2 − ri2 )
S ri2 ro2 S ro2
σθθ = + ;
2(ro2 − ri2 )r2 2(ro2 − ri2 )
τrθ = 0.

In the limit as ri /ro → 0, these become

S r2 ri2
 
S S
σrr =−  i +  1− 2 ;
=
ri2 2 r2 r
2r2 1 − r2 r
2 1−
i
2
o o

S ri2 ri2
 
S S
σθθ = 
r2
+ 
r2
= 1+ 2 ;
2r2 1 − ri2 2 1 − ri2 2 r
o o

τrθ = 0.
Advanced Mechanics of Solids 9-22

Having solved the two sub-problems (a) and (b), we can superpose these two and obtain
the solution of the stress concentration problem. The strain components thus obtained are

ri2 ri4 ri2

   
S S
σrr = 1− 2 + 1 + 3 4 − 4 2 cos 2θ;
2 r 2 r r
2 4
   
S r S r
σθθ = 1 + i2 − 1 + 3 i4 cos 2θ;
2 r 2 r
4 2
 
S r r
τrθ =− 1 − 3 i4 + 2 i2 sin 2θ.
2 r r

The interesting part is in the immediate neighbourhood of the (small) hole. At r = ri , the
stress components are

at r = ri , σrr = 0, τrθ = 0; and σθθ = S − 2S cos 2θ.

There is nothing new in the first two; we already know that this is so. But the last one
is of vital importance. It shows that the vulnerable points are A and B, where there is
severe stress concentration. The peak value of σθθ is 3× the nominal stress S. The stress
concentration factor13 is thus 3! At the points C and D, σθθ = −S. We can also see that
the magnitude of σθθ quickly drops to smaller values when we move away from A and / or
B, showing that this stress concentration is a local phenomenon. Stress concentration is
a very important topic having several practical applications. We, however, do not discuss
this topic any further.
A Concentrated Force on a Semi-infinite Half-space

(a) A force P on an elastic half-space (b) ‘A simple radial solution’

Figure 9.9: A concentrated force P acting on an elastic half-space [Fig. 9.9a] produces a
stress distribution, sometimes known as ‘a simple radial solution’ [Fig. 9.9b].
13
Books on machine design give more details about stress concentration factors. Elaborate charts are
available for designers to help them. Books like Savin, G.N., Stress Concentration Around Holes, Pergamon
Press, (1961) give the solution for several cases. The case of an elliptical hole is of much importance.
Timoshenko [16] states: “a very slender hole (a/b large) perpendicular to the direction of the tension
causes a very high stress concentration. This explains why cracks transverse to applied loads tend to
spread. The spreading can be stopped by drilling holes at the ends of the crack to eliminate the sharp
curvature responsible for the high stress concentration.”
9-23 Two-dimensional Problems

We shall now take up the problem of determining the stresses in a semi-infinite elastic
half-space due to a concentrated (vertical) force [Fig. 9.9].
Consider a stress function φ = φ(r, θ) = Ar θ cos θ, where A is a constant. The corre-
sponding stress components are

1 ∂φ 1 ∂2φ Aθ cos θ Ar(−2 sin θ − θ cos θ) −2A cos θ

σrr = + 2 2
= − 2
= ;
r ∂r r ∂θ r r r
∂2φ
σθθ = = 0;
∂r2  
∂ 1 ∂φ
τrθ =− = 0.
∂r r ∂θ

When we examine the solution, we find that σrr is the only non-zero stress component
and, what is more, that it is independent of θ! It is, thus, a constant at any point on any
semi-circular line as shown in Fig. 9.9b. For this reason, it is sometimes known as ‘the
simple radial solution’. The constant A in the stress function can be evaluated in terms of
the concentrated applied load — really a line load over a unit thickness — by considering
the equilibrium of semi-circular element (like the shaded part in Fig. 9.9b. Thus, we have
from equilibrium requirement
Z π
2 P
σrr r sin θ dθ + P = 0 −→ A= .
0 π

The radial stress is, thus,

2P sin θ 2P
σrr = − =− ,
πr πd
the negative sign implying that it is compressive.
Geotechnical engineers may desire to calculate the stresses at a certain depth, marked
h in the figure. Using the given solution and the stress transformation equations, we can
work them out as shown below.

2P 2P
σxx = σrr cos2 θ =− sin θ cos2 θ = − sin2 θ cos2 θ;
πr πh
2P 2P
σxx = σrr sin2 θ =− sin3 θ =− sin4 θ;
πr πh
2P 2P
σxx = σrr sin θ cos θ = − sin2 θ cos θ = − sin3 θ cos θ.
πr πh

Using this solution, the stresses due to several loads, concentrated and uniformly distributed,
can be worked out by superposition. It is also of interest to work out the corresponding
displacements. Engineers would like to know the displacements, particularly the (vertical)
settlement. But these displacement calculations, though not difficult in principle, are harder.
We do not propose to discuss them here.
Advanced Mechanics of Solids 9-24

ROTATING DISCS
Let us consider the problem of determining the stresses in a rotating disc14 . This is of
great technical importance because rotating discs are a crucial component of steam and gas
turbines. The discs are not generally flat15 , but the theory is simpler.
The disc is shrunk on the shaft with an interference fit. What is the speed at which the
disc runs loose on the shaft making it impossible to transmit power? If we provide too much
interference, the stresses at standstill will be very high. We need to calculate these. The
boundary conditions are in terms of the radial displacement u (related to the interference)
and the radial stress (related to the interference pressure between the shaft and the disc).
Thus, we need to obtain the analytical expressions and the numerical values of σrr , σθθ and
u. As this is not a book on machine design, technically important details are left out. The
shrink allowance is of the order of 1 × 10−3 .
This is an axisymmetric problem. The equation of equilibrium was derived earlier [Eq.
(5.18), p. 5-15]. We borrow it and present it here as
dσrr σrr − σθθ
+ + Fr = 0.
dr r
The body force here is the ‘centrifugal force’ (‘inertia force’ introduced to convert the
problem from dynamics to statics using D’Alembert’s principle). It is
Fr = ρ ω 2 r per unit volume;
Fr × r dr dθ × 1 = ρ ω 2 r2 dr dθ (total radial body force).
Thus, the differential equation of motion is obtained after a little manipulation as
d
(r σrr ) − σθθ + ρ ω 2 r2 = 0. (9.38)
dr
Fig. 9.10a shows an element ABCD. Its deformed position is A0 B 0 C 0 D0 . Each point
moves out only radially. A typical point P moves by u in the radial direction; there is no
tangential displacement, v. Even so, a tangential strain is called into play16 .
The line AB moves out radially to be the line A0 B 0 . This radial displacement is u = u(r).
The radial displacement of the line CD (separated from the line AB by the distance dr) to
be the line C 0 D0 is, therefore, u + (∂u/∂dr) dr. Hence, the radial strain er is
change in length B 0 C 0 − BC u + du
dr dr − u du
er = = = = .
original length BC dr dr
14
This topic, like all others, is treated beautifully in Den Hartog [4]. Readers are advised to read this if
they are not sufficiently familiar with this topic. The most comprehensive, thorough coverage, of nearly
all topics in Engineering Dynamics is in C.B. Biezeno & R. Grammel: Engineering Dynamics, translation
of Technische Dynamik in German; nothing can match this book.
15
Flat discs, that is, discs of uniform thickness are also used.
16
Den Hartog [4] remarks, after quoting these two strain-displacement equations: “ · · · the first one is fairly
obvious, and the second one refers to the feelings of a middle-aged gentleman who lets out one notch of his
belt after his daily good dinner.” The belly expands radially outwards; there is no tangential movement.
Yet there is a tangential strain making it necessary to loosen the belt!
9-25 Two-dimensional Problems

(a) Deformations (b) Stress components

Figure 9.10: To explain the simplified case of an axisymmetric problem

The line AB of length r dθ, after deformation, becomes A0 B 0 of length (r + u) dθ. Hence,
change in length (r + u) dθ u
tangential strain eθ = = = .
original length r dθ r
These are the strain-displacement relations for the special case of axisymmetry.
du
radial displacement, u = u(r) radial strain, er =; (9.39a)
dr
u
tangential displacement, v = 0 tangential strain, eθ = . (9.39b)
r
We have here only the two stress components σrr and σθθ ; the others are all zero. Out of
the three displacement components, we have here only the radial displacement, u. Let us
review the situation and note the unknowns and the available equations to determine them.
The unknowns are (i) the stress components σrr , σθθ (2), (ii) the strain components
err ; eθθ (2); and (iii) the radial displacement component u (1). The available equations are
(i) the equation of equilibrium [Eq. (7.10)] (1); (ii) the strain-displacement relations [Eqs
(9.39a), (9.39b)] (2); and (iii) the stress-strain equations [Eqs (9.40a), (9.40b)] (2). There
are, thus, 5 (2 + 2 + 1 = 5) unknowns, and 5 (1 + 2 + 2 = 5) equations available.

1
err = [σrr − νσθθ ]; and (9.40a)
E
1
eθθ = [σθθ − νσrr ]. (9.40b)
E
We can eliminate any four of them and obtain an equation in terms of the remaining one.
As the boundary conditions are usually expressed in terms of σrr or u, one of these two is
preferred. Choosing σrr here, the steps of the elimination procedure are shown below.

i) The strain components are eliminated by combining the strain-displacement relations

[Eqs (9.39a), (9.39b)] and the constitutive equations [Eqs (9.40a), (9.40b)] as
du 1
= [σrr − νσθθ ]; and (9.41a)
dr E
u 1
= [σθθ − νσrr ]. (9.41b)
r E
Advanced Mechanics of Solids 9-26

ii) Differentiate Eq. (9.41b) and substitute into the result Eqs (9.41a, 9.41b). The variable
u is thus eliminated. We thereupon obtain the equation
d d 1+ν
(σθθ ) − ν (σrr ) + (σθθ − σrr ) = 0. (9.42)
dr dr r
iii) Differentiate σθθ [Eq. (9.38)] w.r.to r, and substitute the result in Eq. (9.42).

We would, thus, obtain the final governing equation in terms of (r σrr ) as

d2 d
r2 2
(r σrr ) + r (r σrr ) − (r σrr ) + (3 + ν) ρ ω 2 r3 = 0; (9.43)
dr dr
d2 d
i.e., r2 2 (r σrr ) + r (r σrr ) − (r σrr ) = −(3 + ν) ρ ω 2 r3 . (9.44)
dr dr
This is a linear, but variable coefficient, differential equation with a ‘right hand side’. We
know that the solution consists of a complementary solution and a particular integral17 . The
complementary function is of the form (r σrr ) = rn , as argued out in the footnote18 . We
obtain the auxiliary equation by substituting this in the (‘left hand side’ of the) differential
equation.
[n(n − 1) + n − 1] rn = 0, −→ n = ±1.
When a particular integral is added to the complementary function, we obtain the general
solution as
B 3−ν
r σrr = Ar + − ρ ω2 r3 .
r 8
The required expressions (for σrr , σθθ , u) are obtained as
B 3+ν
σrr = A + 2
− ρ ω2 r2 ; (9.45a)
r 8
B 1 + 3ν
σrr = A − 2 − ρ ω 2 r2 ; and (9.45b)
r 8
1 − ν2
 
r B 2 2
u= (1 − ν) A − (1 + ν) 2 − ρω r . (9.45c)
E r 8
17
Some authors write rather carelessly the particular integral. There can be several particular integrals.
Two particular integrals can differ by a complementary function.
18
Generally variable coefficient differential equations, even if they are linear, are hard to solve. A closed
form solution is more an exception than the rule. Several special functions, which have been investigated
extensively, are the solutions of such linear, variable coefficient differential equations. Many of them
have similar interesting and useful properties like orthogonality. Bessel functions, Legendre polynomials,
Hermite polynomial, etc. are some of the well known, well explored special functions.
Our equation here is quite easy to solve. To see this, let us assume that the unknown dependent variable,
here (r σrr ), is equal to rn . If we differentiate it once, it goes one step down, and becomes n r(n−1) . If
it is multiplied by r, it goes one step higher to become n rn . Two differentiations, two steps down, but
multiplication by r2 , and thus two steps up, and we are back again to the same level rn . This is true
with every term if the equation is ‘homogeneous’ in this sense. (The word ‘homogeneous’ is used in the
context of linear differential equations also in the sense that the right hand side is zero.) Thus, we can
understand that the solution is of the form rn . This equation is known variously as Cauchy’s equation,
Euler’s equation, etc.
9-27 Two-dimensional Problems

The two arbitrary constants A and B are to be determined from the usually known (some-
times unknown also) boundary conditions. [For example, when we write down the boundary
condition to determine the interference pressure, it is unknown. It can be obtained later.]
Some more calculations and explanations are needed for a complete solution of the problem.
We do not propose to show them here.
The thick cylinder problem (the famous Lamé’s problem) — to determine the stresses
in a thick cylinder subjected to fluid pressure — is the same as this, except that there is no
body force now. The two stresses, radial and tangential (or hoop) stresses are given by
B B
radial stress: σrr = A + ; hoop stress: σθθ = A − .
r2 r2

THICK CYLINDERS: LAMÉ’S PROBLEM

We have seen this above as a special case of rotating disc when there is no rotation and,
consequently, no body force. [On the mathematical side, the consequence is that there is
no particular integral19 .] This problem is of much technical importance. We shall discuss
this in greater detail.

(a) A thin cylinder (b) A thick cylinder

Figure 9.11: The figures show a thin [Fig. 9.17a] and a thick [Fig. 9.17b] cylinders.

Thick Cylinders
When the wall thickness is large, or when the inner pressure is very high, thick cylinder
calculations are necessary. Some examples are in the design of pipes carrying high pressure
fluids, gun barrels for tanks and warships, high pressure steam machinery, and hydraulic
press. Another case in point is the (high pressure) fuel injection systems for diesel engines.
In all these calculations calculations based on thick cylinder theory are to be used. Inter-
ference fits are another important area of application. In short, no engineer can afford not
to learn thick cylinder theory.
19
Is it better to discuss this simpler problem first before taking up the more difficult case of rotating discs
following the general principle of ‘from simple to complex’ ? Or, is it better to discuss the general problem
first with rigorous details, and later obtain the simpler solutions as special cases of the general theory?
Opinion is divided on this matter as in most other things in life. There are advantages and disadvantages
in both approaches. It seems best to be exposed to both approaches so that each person can choose as he
likes.
Advanced Mechanics of Solids 9-28

Thin and Thick Cylinders

Fig. 9.17 shows a thin and a thick cylinders subjected to an internal pressure pi . Both are
axisymmetric problems; that is there is circular symmetry. All the variables are independent
of θ. We shall compare and contrast the stresses in the two cases.
Thin cylinders:
We know — we have studied this topic earlier — that the tangential (also called hoop),
axial, and radial stresses (σθθ , σzz ≡ σaxial , σrr , respectively) in the case of a thin cylinder
are given by σθθ = pi r/t; σzz = pi r/(2t). The radial stress σrr is small in comparison and is
neglected. Because of the circular symmetry (axisymmetry) these are the principal stresses;
the tangential, axial, and radial directions are the principal directions.

Figure 9.12: A thin cylinder subjected to an internal pressure pi is shown. As the thickness t
is small, the stress σθθ may be regarded as substantially constant across the small thickness.

For a thin20 cylinder t << r and, therefore, the tangential (or hoop) stress σθθ , and the
axial stress σzz (if the cylinder is closed at the ends) are several times the applied internal
pressure. The radial stress σrr , being very small, is neglected. Thus,
pi r
σθθ = tensile :
t
pi r
σaxial ≡ σzz = if the cylinder is closed, tensile;
2t
= 0 if the cylinder is open;
σrr = small, between −p and 0 at the boundaries.

From the equation of equilibrium in the radial direction [Fig. 9.17a], we have
pr
pi × 2r × 1 = 2 × σθθ × t × 1 −→ σθθ = .
t
Similarly, from the equation of equilibrium in the axial direction [Fig. 9.12], we have
pi r
pi × π r2 = σzz × 2 π r t −→ σzz = .
2t
20
As an example let us note that the diameter of the fuselage of an aircraft is very, very large compared to
the thickness — typically about 1.4 mm — of the outer skin. Here t << r. We can surely treat the skin
as a thin cylinder! The internal pressure pi arises because the outside pressure when the aircraft flies is
much less than the atmospheric pressure maintained in the cabin. This is perhaps an extreme case.
9-29 Two-dimensional Problems

As the thickness t increases, the tangential stress σθθ becomes less and less. In the case of
a thick cylinder, this becomes comparable in magnitude to σrr . Thus, we cannot neglect
σrr now. This is the difference from the point of view of stress analysis.
[Here arises a question: how could we obtain the stresses using only the equations of
equilibrium? We know that all problems in stress analysis are statically indeterminate;
the equations of equilibrium alone cannot give us the stress components. Where is the
anomaly? Let us realise that we made an assumption that the stress component σθθ is
uniformly distributed across the thickness. This is why we were able to obtain the stress
components without using the strain-displacement relations and the constitutive relations.]

(a) A thin layer of a thick cylinder (b) Stress distribution in a thick cylinder

Figure 9.13: The figure show a thin layer of a thick cylinder. This is used to derive the
stress distribution [Fig. 9.13b] when an internal pressure is applied in a thick cylinder.

Thick cylinders:

(a) Stress components and the stress resultants (b) A part of the thick cylinder

Figure 9.14: The figure show a thin layer of a thick cylinder. The stress components and the
resultants are shown. Plane cross-sections continue to remain plane even after the (fluid)
pressure is applied. From this assumption, we can derive the equation of equilibrium.

We desire to determine the stress distribution in a thick cylinder. This is an axisym-

metric problem and, therefore, all the stress components (and strain and displacement
components too) are functions of only the variable r. Everything is independent of the
variable θ. The derivatives are all ordinary derivatives, not partial ones.
Advanced Mechanics of Solids 9-30

A thin half ring of thickness dr with the stress components is shown 9.14a. The resul-
tants are shown in dotted lines. Considering the equilibrium of the half ring, we have
 
dσrr
σrr × 2 r × 1 = σrr + dr × 2(r + dr) × 1 + (σθθ × dr × 1) × 2,
dr

which, on dropping quantities of a higher order of smallness and cleaning up, reads as
dσrr
σθθ = −σrr − r . (9.46)
dr

There are two unknowns, but only one equation of equilibrium. We, thus, need to use
the other governing equations also to solve for the stress components. Recall the statement
that we made that all problems in stress analysis are statically indeterminate internally.
How shall we proceed? Where will the other equation come from?
Plane strain assumption
At this stage, let us make the assumption that cross-sections that are plane before the
internal (fluid) pressure is applied, continue to remain plane [Fig. 9.14b]. [In this figure
‘yes’ shows that plane cross-sections remain plane, and ‘no’ shows that the cross-sections
do not go crooked.] This implies that eaxial ≡ ezz = 0. Thus,
1
eaxial ≡ ezz = 0 = [0 − νσθθ + νσrr ]
E
ν
= [σrr − σθθ ].
E

From this last equation we can note that σθθ − σrr = a constant, say, 2A. Thus, the
two equations that we have are

σθθ − σrr = a constant, say, 2A;

dσrr
σθθ = −σrr − r .
dr
Multiplying the first of these two equations by r and using the second equation, we obtain
dσrr
σθθ − σrr = −2σrr − r
dr
dσrr
− 2A = −2σrr − r
dr
dσrr d
− 2Ar = 2(rσrr ) + r2 = (r2 σrr ).
dr dr
On integration, we obtain the expression / formula for σrr as −A + B/(r2 ). Now the
expression / formula for σθθ also can be readily worked out. They are (with a slight change
in the definition of the constants)
B B
σrr = A + ; and σθθ = A − .
r2 r2
9-31 Two-dimensional Problems

(a) A thick cylinder subjected to internal (pi ) (b) Axial stresses in the thick cylinder (if, and only
and external fluid pressure (po ) if, the cylinder is closed)

Figure 9.15: The figure [Fig. 9.15a] shows a thick cylinder subjected to internal (pi ) and
external (po ) pressures. [Fig. 9.15b] shows the axial stresses due to the fluid pressure if,
and only if, the cylinder is closed.

Thick Cylinders Subjected to Internal and External Pressure

An important technical problem is to determine the stresses in a thick cylinder subjected
to (i) and internal pressure pi , and / or (ii) an external pressure po . (Sometimes these are
fluid pressures. However, there are problems when there is no fluid at all as in interference
fit calculations21 . The inner and outer radii are ri and ro respectively.
We know that the radial and tangential stresses are given by22

B B
σrr = A + ; σθθ = A − .
r2 r2
The constants A and B are determined using the known boundary conditions.

B
(i) at r = ri , σrr = −pi −→ A+ = −pi
ri2
B
(ii) at r = ro , σrr = −po −→ A + 2 = −po .
ro

Solving for A and B, we obtain — the slightly ‘dirty’ algebra is left out —

pi ri2 − po ro2 (ro ri )2 (po − pi )

A= , B= .
ro2 − ri2 ro2 − ri2
21
The pressure inside a gun barrel, and the pressure difference between the atmospheric pressure inside an
aircraft and the low pressure outside when the aircraft flies to high altitudes are other examples.
22
We may also write these differently, but equivalently, as
D D
σrr = C − ; σθθ = C + .
r2 r2
There is no real difference; the constants are now different (A = C; B = −D). However, it is desirable to
stick to one set of formulae. There is also a case to break this habit occasionally, if not for anything else,
at least to remind ourselves that both sets of formulae are equally correct.
Advanced Mechanics of Solids 9-32

Figure 9.16: Superposition

The expressions for the stress components are

2
2
pi ri2 − po ro2 + rirro (po − pi ) pi ri2 − po ro2 − rirro (po − pi )
σrr = ; σθθ = .
ro2 − ri2 ro2 − ri2

We can solve this problem by superposition also as indicated below. Fig. 9.16 represents
the problem that we solved just now. It can also be solved by superposition. The two cases
shown on the right hand side of Fig. 9.16 — internal pressure pi alone applied, and external
pressure po alone applied — can be superposed to be the case shown on the left hand side
of Fig. 9.16.
Case (a) A thick cylinder subjected to an internal pressure p i only:
Let us consider case (a). The governing equations are, as always, in the form

B B
σrr = A + ; σθθ = A − .
r2 r2
The constants A and B are determined using the known boundary conditions:

B
(a) at r = ri , σrr = −pi −→ A+ = −pi ;
ri2
B
(a) at r = ro , σrr = 0 −→ A + 2 = 0.
ro

Solving for A and B we obtain

pi ri2 −pi ro2 ri2

A= ; B= .
ro2 − ri2 ro2 − ri2

The radial and tangential stresses are

h i
ro ri 2
pi ri2 −


B pi ri2 pi ro2 ri2 r
σrr = A + = − = ;
r2 ro2 − ri2 r2 (ro2 − ri2 ) ro2 − ri2
h
2 i
B pi ri2 pi ro2 ri2 pi ri2 + rorri
σθθ = A − = + = .
r2 ro2 − ri2 r2 (ro2 − ri2 ) ro2 − ri2
9-33 Two-dimensional Problems

Additionally, if the cylinder is closed at the ends,

pi ri2
σzz = ; (σzz = 0 if the cylinder is open at the ends).
ro2 − ri2
Case (b) A thick cylinder subjected to an external pressure p o only:
Now we shall consider case (b). It is always better to begin from the fundamental equations
B B
σrr = A + ; σθθ = A − .
r2 r2
The boundary conditions this time are
B
(a) at r = ri , σrr = 0 −→ A+ = 0;
ri2
B
(a) at r = ro , σrr = −po −→ A + 2 = −po .
ro
Solving for A and B from these equations, we obtain the stress components.
−po ro2 po ro2 ri2
A= ; B= .
(ro2 − ri2 ) (ro2 − ri2 )
 2 2
ro2 ri2
  
ro ri 2
B p o r 2 − ro B −po r2
+ ro2
σrr =A+ 2 = 2 ; σθθ = A − = .
ri 2
(ro − ri ) ri2 (ro2 − ri2 )
Closed cylinder:
In addition, there will be an axial stress σaxial ≡ σzz which is uniform on the cross-section
if, and only if, the cylinder is closed [Fig. 9.15b]. The cross-sectional area is π(ro2 − ri2 ).
The force due to the (fluid) pressure acting on the end plate is pi πri2 . Hence, if the axial
stress is uniformly distributed23 , it is
pi πri2 ri2
σaxial ≡ σzz = 2 = pi .
π(ro2 − ri ) ro2 − ri2
We may also note that the axial strain eaxial ≡ ezz = constant. In other words, plane
cross-sections continue to be plane cross-sections even after the fluid pressure is applied.
This fact can be seen readily.
  
1 1 B B
eaxial ≡ ezz = [σzz − ν(σrr + σθθ )] = σzz − ν A + 2 + A − 2 .
E E r r
This shows that, if ezz = constant, then σzz = constant, and vice versa.
Open cylinder:
If the cylinder is open, there is no end plate, and there is no axial stress. Thus, σzz = 0.
23
It appears that Lamé had first derived these formulae assuming that the axial stress is uniformly dis-
tributed. Objections were raised when he presented his work. A sad Lamé went home and reworked the
problem assuming this time that plane cross-sections remain plane (that is, assuming that ezz = constant).
Lo, and behold! The result was the same! A beaming Lamé presented his work again to the complete
satisfaction of everybody. All is well that ends well!
Advanced Mechanics of Solids 9-34

(a) A closed cylinder (b) An open cylinder

Figure 9.17: The figures show a closed and an open cylinders. There will be an axial stress
σaxial ≡ σzz in a closed cylinder, but not in an open one.

Strains and maximum strain:

The circumferential strain eθθ is given by (the generalised Hooke’s law)
1
eθθ = [σθθ − ν(σrr + σzz ].
E
The maximum strain — it is this that we are concerned with — is at the inner fibre (r = ri ).
Thus,  1
emax = eθθ  = [σθθ − ν(σrr + σzz ]r=ri .

max E
Substituting the expressions for σθθ , σrr , σθθ at the inner fibre (r = ri ), we can obtain
the maximum (circumferential or hoop) strain. Note that we have already worked out the
expression for the axial stress σzz , and that it is zero (σzz = 0) for an open cylinder. When
these are used, we obtain the expression for the maximum strain24 .
 1
emax = eθθ  = [σθθ − ν(σrr + σzz ]r=ri .

max E
The simpler case of only an internal pressure pi is of special interest. Now we have
 h 2 2
i
  pi ro +(1−ν)r i
+ ν (closed cylinders - Clavarino’s equation)
E ro2 −ri2 i
emax = eθθ  = p h r2 +r (9.47)

2
max  i o
E r2 −r2 + ν (open cylinders - Birnie’s equation)
i
o i

These are sometimes called the Clavarino’s and Birnie’s equations.

Displacements:
We often need for interference calculations and in other places the expressions for the
displacements (especially radial displacements). The tangential displacement, of course, is
zero because of the circular symmetry — axisymmetry — in our problem. For this we use
the strain-displacement relations that we have already seen. In the context of axisymmetry,
24
The numerical value, if desired, can be worked out for the specific case of interest. This expression is
useful, among other places of analysis and applications, in the maximum strain theory of failure.
9-35 Two-dimensional Problems

there is only the u displacement; v = 0 because of this circular symmetry; w = 0 because

this is a two-dimensional problem. We need the following result, so crucial in interference
fit calculations25 .
du u
err = ; eθθ = .
dr r

The Case of a Solid Disc Subjected to an (External) Pressure p o

Whatever we have discussed so far carries over to the case of a solid disc (or shaft) also.
Now, of course, it can be subjected only to an external pressure (say, po ), and not to an
internal pressure! Now also, as before we have
B B
σrr = A + ; σθθ = A − .
r2 r2
The boundary conditions this time are different. One condition that is obvious is that
B
at r = ro , σrr = −po −→ A+ = −po .
ro2

What is the other boundary condition? Where does it come from? Well, we note that,
if B survives (and 6= 0), then the stress at the centre becomes infinite. This cannot be
permitted. (In other words, the singularity at r = 0 cannot be allowed to exist because the
stresses would be infinite.) Hence B = 0. The stresses are now σrr = σθθ = 0.
Both the stresses are equal and, what is more, they are both constant everywhere and
equal to −po ! There is no shear stress at any point on any plane26 ! Now the two-dimensional
stress tensor is isotropic! Every plane — every plane normal to the cross-section — at every
point is a principal plane!
We can arrive at the same conclusion by arguing differently. Let us write down the
expression for the tangential strain eθθ . This gives us
u 1
eθθ = = [σθθ − ν σrr ].
r E
Now we can argue that the centre r = 0 must remain as the centre even after deformation: u
must be zero at the centre r = 0. We arrive at the same result: B = 0. Thus, in conclusion,
the stresses inside the solid disc (or solid shaft) are σrr = σθθ = −po .
A Numerical Example
A cylinder has an internal diameter of 380 mm and an external diameter of 240 mm.
Calculate the maximum pressure that can be applied so that the maximum stress does not
25
After the stresses are determined using Lamé’s equations, we can calculate the strains err and eθθ from
the generalised Hooke’s law. From eθθ , we can obtain the radial displacement. The principle of consistent
deformation is in terms of the radial displacements.
26
By any plane, we mean any plane normal to the cross-section. There is no shear stress on any plane. There
can be shear stresses like τxz and τyz . Every point in the cross-section is a (two-dimensional) isotropic
point!
Advanced Mechanics of Solids 9-36

exceed 400 MPa for the two cases: (a) no external pressure; po = 0; pi =?; and (b) no
internal pressure; pi = 0; po =?
The dimensions tell us clearly that this is a thick cylinder.
Case (a)
We know that the maximum stress (which is σθθ occurs at the inner boundary and that its
value is:
 r2 + r2
σθθ  = pi o − 2 i = 400 MPa.

max ro ri
1902 − 1202
pi = 400 × = 171.88 MPa.
1902 + 1202
Thus, the maximum internal pressure that can be applied, pi = 171.88 MPa.
Case (b)
When only an external pressure is applied,

ro2 ri2
 
σθθ = −po 2 1+ 2 .
ro − ri2 r

The maximum of σθθ occurs at the inner boundary where r = ri . Hence,

 ro2
σθθ  = −2po ;

max ro− ri2
1902
= −2po = 400 MPa;
1902 − 1202
400 × 21700
po = = 120.22 MPa.
2 × 36100
Thus, the maximum external pressure that can be applied = 120.22 MPa.
[As a rough guide to decide if a given cylinder is to be considered as thin or thick, we
may perhaps say this: if the ratio of the outer and inner radii (ro /ri ) is greater than 1.1, it is
necessary to treat the cylinder as thick. This is not an absolute command. With experience
and engineering judgement, such prescriptions can be ignored, overruled, or modified.]
Interference Fit Calculations
One of the important applications of theory discussed so far is for interference fit calcu-
lations. If a sleeve is to be mounted on a (solid) shaft by an interference fit, the inner
diameter of the sleeve is made a little less or smaller than the diameter of the shaft. The
sleeve is forced on the shaft, and for this reason, such a fit is also referred to as a force fit or
a shrink fit [Figs 9.18, 9.19]. In practice, a slightly oversize part is slowly pushed through
by a hydraulic ram. See p. 13-57 also.
Let us take the example of a brass sleeve force-fitted on a steel shaft [Fig. 9.19]. The
outer brass sleeve has its outer and inner radii of ro and ri , respectively. The nominal radius
of the steel shaft is the same as ri . If the interference on the radius δ is specified or given, we
9-37 Two-dimensional Problems

Figure 9.18: Interference fit: a sleeve mounted on a shaft. A sleeve is force-fitted on a shaft.
The inner diameter of the outer sleeve is made a little less than the diameter of the shaft
on which the sleeve is assembled.

(a) An outer brass sleeve (b) The inner solid steel shaft (c) The assembly

Figure 9.19: An interference fit of a sleeve on a shaft. The outer diameter of the shaft
is actually a little larger, though nominally the same. When the sleeve is fitted on the
shaft, the inner boundary of the sleeve moves up slightly, and the diameter of the shaft
correspondingly reduced. The figure shows the diameters, before and after the assembly.

can find out the interference pressure. This pressure (which may also be called the shrinkage
pressure ps ) can be calculated, and once it is known the stresses in the shaft and in the
sleeve can be calculated using the standard thick cylinder formulae (Lamé’s equations).
Case (a): Outer (brass) sleeve
The (unknown) interference / shrinkage pressure ps is an external pressure acting radially
inwards on the (outer) brass sleeve. The circumferential (also called hoop) and radial
stresses at the interface r = ri are given by:
 r2 + ri2
σθθ (ri ) = ps o2 (tensile);

sleeve ro − ri2

σrr (ri ) = −ps (compressive).

sleeve

The inner radius moves radially outwards. The amount of radial displacement, u can be
calculated as shown:
u 1 r
tangential strain, eθθ = = [σθθ − νσrr ] −→ u = [σθθ − νσrr ].
r E E
Advanced Mechanics of Solids 9-38

Thus,
ro2 + ri2
 
 ri ri
u(ri ) ≡ usleeve = [σθθ − νbrass σrr ] = ps + νbrass . (9.48)

sleeve Ebrass Ebrass ro2 − ri2
We have, thus, calculated δsleeve . In the same way, we can calculate δshaf t also.
Case (b): Inner (steel) shaft
The inner steel shaft is subjected to the same interference pressure ps , but radially inwards.
This produces circumferential and radial stresses, σθθ and σrr of the same magnitude, both
compressive. The values at the interface are:
 r2 + ri2
σθθ (ri ) = −ps o2 (compressive);

shaf t ro − ri2

σrr (ri ) = −ps (compressive);

shaf t
 ri ri
u(ri ) ≡ ushaf t = [−ps + νsteel ps ] = − ps [1 − νsteel ]. (9.49)

shaf t Esteel Esteel
The radius moves radially inwards. The radial interference, δ = usleeve − ushaf t = |usleeve | +
|ushaf t | [Fig. 9.19c]. Thus, the principle of consistent deformations gives us

δ = usleeve − ushaf t = |usleeve | + |ushaf t |;

 2
ro + ri2
  
1 1
= ps ri + νbrass + (1 − νsteel ) .
Ebrass ro2 − ri2 Esteel
giving us
δ
ps = h 
ro2 +ri2
 i. (9.50)
1 1
ri Ebrass ro2 −ri2
+ νbrass + Esteel (1 − νsteel )
This is the required expression for the interference pressure. Once the interference pressure
ps is known, the stresses inside the outer sleeve and the inner shaft can be readily worked
out using the thick cylinder equations (Lamé’s formulae). If the numerical values are given,
we can always work out the numerical values of the stresses. See p. 13-57 later.
Methods to Enhance the Elastic Strength of Thick Cylinders
In some industrial applications, the internal pressure is very large; then the cylinder would
be too thick. This is not cost effective. The outer layers are not stressed much, and so
the material is not utilised effectively. Can we enhance the elastic strength with a view to
reducing the thickness, or to utilising all parts of the material more effectively?
Yes, we can. One way is to shrink a hollow cylinder (called a jacket, or a hoop, or a
shell) on the cylinder. This will help reduce the maximum circumferential stress, and we
can have a better distribution of circumferential stress σθθ . The problem, we have seen,
is that much of the material is under-utilised. This suggested measure will improve the
distribution of stress [Figs. 9.20a, 9.20b].
Another way is to have several layers, one wrapped around the other, to have a laminated
cylinder. This method will produce a more uniform stress distribution [Fig. 9.21b].
9-39 Two-dimensional Problems

(a) Stresses due to the initial shrinking process (b) Total stresses after pi is applied

Figure 9.20: The stress distributions (i) when a jacket is shrunk on the cylinder, and the
total stress distribution when an internal pressure pi is applied. The stress distribution is
now more favourable.

(a) Distribution of hoop prestress, σθθ (b) Total stresses after pi is applied

Figure 9.21: The figures show the stresses in a multi-layer (laminated) cylinder. Fig. 9.21a]
shows the distribution of the circumferential prestress. Fig. 9.21b shows the distribution
of the total circumferential stress σθθ , ie., after the internal pressure is applied to the
prestressed cylinder. We can see that the distribution of σθθ is now vary favourable; it is
now almost uniform across the thickness.

A third method is autofrettage. A very high pressure is applied initially leading to

yielding at the inner fibre. With still higher pressure, the region of yielding penetrates
from the inner fibre even deeper into the cylinder. When this applied internal pressure
is released, there would be beneficial residual stresses inside the ‘loaded and unloaded’
cylinder. Now such a prestressed cylinder is capable of withstanding much larger internal
pressures compared to the virgin cylinder.
Closing Comments
Two-dimensional problems are an important part of the theory of elasticity. It, therefore,
deserves a far more comprehensive treatment than what is given here. There are several
topics and methods of solution which are not even mentioned. We regret our inability to
give this topic the honourable and well deserved important place that this deserves. We
hope that the ambitious students will recognise that what is presented here is to be regarded
as no more than an introduction to two-dimensional problems. With these words of apology
or explanation, we close this chapter here.
Advanced Mechanics of Solids 9-40

(a) Autofrettage: a schematic figure (b) Autofrettage: stresses

Figure 9.22: A schematic figure of autofrettage. An enormous pressure is applied so that

plasticity creeps into the cylinder to a depth of d (autofrettage depth). When the pressure is
released there are internal compressive residual (hoop) stresses. When an internal pressure is
applied to such a cylinder, the total stresses are reduced because of the residual compressive
stresses left in the cylinder. Fig. 9.22b shows the stress distributions.

In the next chapter, we shall discuss the important topic of energy methods.
Chapter 10

ENERGY METHODS

Energy methods are extremely important in the mechanics, particularly advanced mechan-
ics, of solids. Several problems can be solved effectively by these methods, that are not
otherwise easy to tackle. Even more importantly, these methods help us to formulate im-
portant physical problems. Thus, energy methods may be regarded as the starting point,
first of setting up a general framework of concepts and methodology; next of formulating
key notions, concepts and methods; and finally of solving actual physical problems of much
interest to engineers at the operational level. These are intimately linked to variational
methods based on the calculus of variations. In fact, energy methods and variational meth-
ods applied to deformable bodies are almost synonymous. Finite Element Method (FEM),
which has become so powerful and indispensable in the analysis and design of engineering
structures and machine components, is heavily dependent on these variational methods.
The starting point is Johann Bernoulli’s1 principle of virtual work. This is to be written
for the case of a deformable body. We shall do this later in this chapter. We are immediately
led to the more special principle of stationary potential energy. A mathematical technique
called the Legendre transformation takes us to the complementary energy and the principle
of complementary energy. Castigliano’s theorems follow.
Unfortunately energy methods are not always easy. Beginners2 find them difficult; they
are often confused. We shall, therefore, be simple-minded and present only the relatively
easier part. We hope that the ambitious students will learn the advanced material later3 .
1
Johann (also called Jean I or John I) Bernoulli (July 1667 - Jan. 1748) was a famous mathematician of
the illustrious Bernoulli family. He was regarded as the greatest mathematician in Europe at that time.
2
And who are beginners? Nearly everybody except a few, very few, real ustads or grand masters.
3
What makes these difficult? Well, these principles are formulated in terms of triple integrals often using
the index notation. The minimum principles make it necessary to minimise such triple integrals. The
technique of minimisation uses the calculus of variations. These are frightening for those who may not
have studied the calculus of variations. Actually triple integrals are nothing to be afraid of. In many
problems the evaluation of these triple integral is very simple. For example, when applied to the members
of a truss, the stresses and the dimensions are often constants, and the triple integral crumbles down to
the evaluation of the volume of the truss member. But the real reason is that students of engineering
hardly ever see mathematics used in their engineering subjects.
Advanced Mechanics of Solids 10-2

The concepts of strain energy and total potential energy4 play a crucial role in these
methods. The reason is that there are powerful theorems like the two Castigliano theorems
and the theorem of stationary total potential energy.
As energy plays a central role, we shall begin by discussing strain energy. We shall then
use it to obtain the deflection of beams, etc.
STRAIN ENERGY
We have studied and used the concept of strain energy5 in our earlier studies. Here we shall
discuss this a little more in detail.

(a) An element: only σxx (b) A coordinate system (c) U and U ∗

Figure 10.1: An element subjected only to uniaxial stress σxx is shown. The area OAB
under the stress-strain curve (straight line) OA is the strain energy density U. The area
OAC (which ‘completes the area’) is the complementary energy density U ∗ .

We shall now approach the concept of strain energy differently. Let us first consider a
simple block [Fig. 10.1] acted upon by a uniaxial stress σxx . The force σxx dy dx produces
an elongation exx dx in the x−direction. The work done, therefore, is
1 1 1 1
dW = force × elongation = (σxx dy dz)(exx dx) = σxx exx dx dy dz = σxx exx dV,
2 2 2 2
where dV is the elemental volume. This is the area under the line OA [Fig. 10.1c]. This
work done is stored in the body as internal energy called the strain energy dU . This is
related to the strain energy density function U by the relation

dU = U dV = U dx dy dz.

The strain energy density function, we recall, is the strain energy per unit volume. Now if
all the stress components are simultaneously present,
1
dU = U dV = (σxx exx + σyy eyy + σzz ezz + 2τxy exy + 2τyz eyz + 2τzx ezx + 2τxy exy ) dV,
2
4
It is better to use the term potential instead of the more popular one potential energy. Thus, we shall
prefer to use the terms total potential instead of total potential energy, and the theorem of stationary total
potential instead of the theorem of stationary total potential energy, respectively.
5
We may recall that the existence of a strain energy density function was assumed when we reduced the
number of elastic constants (elastic moduli) from 36 to 21.
10-3 Energy Methods

where 2 exy = γxy , 2 eyz = γyz , 2 ezx = γzx are the shear strains.
This expression for U may be written either (i) in terms of the stress components only,
or (ii) in terms of the strain components only using the constitutive equations (stress-strain
relations, the generalised Hooke’s law). Expressed in terms of the stress components, it
appears as
1 ν 1 2
U= (σ 2 + σyy
2 2
+ σzz ) − (σxx σyy + σyy σzz + σzz σxx ) + 2
(τ + τyz 2
+ τzx )
2E xx E 2G xy
1 2
= [I − 2(1 + ν)I2 ], (10.1)
2E 1
where I1 and I2 are, respectively, the first and the second invariants. If, instead, U is
expressed in terms of the strain components, the expression appears as
1
U = λ e2 + G(e2xx + e2yy + e2zz ) + 2G(e2xy + e2y + e2zx ), (10.2)
2
where
Eν
λ= and e = exx + eyy + ezz .
(1 + ν)(1 − 2ν)
This expression for U tells us that the strain energy U is always positive. This property
of being positive definite is of theoretical importance. Thus, U = 0 only if all the strain
components (and hence all the stress components) are zero. This fact has important, very
important, consequences. Some of them are the uniqueness theorem of linear elasticity, and
the principles of minimum total potential (energy) and minimum complementary energy.
We emphasise that
∂U
= σij (i, j = 1, 2, 3).
∂eij
Here we can verify this result using Eq. (10.2). For example,
∂U
= λ e + 2Gexx = σxx .
∂exx
Strain energy for the case of plane stress
Sometimes it is useful to have the (simplified) expression for the (simplified) case of plane
stress. Now the three stress components σzz , τxz , and τyz vanish. With σzz = τxz = τyz = 0,
we can obtain the expression for U in terms of the stress components as
1 ν 1 2
U= (σ 2 + σyy
2
) − σxx σyy + τ (10.3)
2E xx E 2G xy
and in terms of the strain components as
E
U= (e2 + e2yy + 2ν exx eyy ) + 2G e2xy . (10.4)
2(1 − ν 2 ) xx
R
The expression for the total strain energy U stored in the body is U = U dV. This is a
crucial concept in the mechanics of solids. The energy theorems are based on the strain
energy and its dual (say, its extension in a sense), the complementary energy U ∗ .
Advanced Mechanics of Solids 10-4

(a) U and U ∗ : linear case (b) U and U ∗ : nonlinear case

Figure 10.2: The strain energy and the complementary energy for the linear [Fig. 10.2a]
and nonlinear [Fig. 10.2a] cases. Note that U = U (δ), while U ∗ = U ∗ (P ).

Complementary Energy
The upper triangle OAC Figs 10.2 completes the rectangle OBAC; hence the area enclosed
by OAC is a ’complementary’ quantity. It has the dimension of energy. It is, therefore,
called the complementary energy, even though it is not an energy6 . In this simple case, the
complementary energy U ∗ is related to the strain energy U by the relation

U ∗ = −U + σ1 e1 . (10.5)

The complementary energy U ∗ is obtained from the strain energy U by what is known
as a Legendre transformation7 . In the simple case that we consider here, the relationship
[Eq.(10.5)] is given in this simple form. There are such dual transformations. These are
both interesting and useful. However, we cannot discuss them here. Note that in the linear
case [Fig. 10.2a] (but not in the nonlinear one [Fig. 10.2b]), U = U ∗ numerically. But it is
not quite correct to write U = U ∗ , because U = U (δ) while U ∗ = U ∗ (P ).
STRAIN ENERGY: SOME COMMON STRUCTURAL ELEMENTS
We shall define strain energy U in a general form, with the usual notations, as
ZZZ
1
U= (σxx exx + σyy eyy + σzz ezz + τxy γxy + τyz γyz + τzx γzx ) dx dy dz. (10.6)
2 V

Let us simplify this expression to suit the special case of a ‘one-dimensional body’ such as
an axially loaded bar, a circular shaft in torsion and a beam.
Strain Energy for ‘One-Dimensional’ Bodies
For such a case of a ‘one-dimensional’ body (such as an axially loaded bar, a circular shaft in
torsion, or a cantilever loaded by an end load) [Fig. 10.3], there are only two components of
the stresses, σxx and τxy = τyx . For this special simplified case, the expression [Eq. (10.6)]
6
Named after Harold Malcom Westergaard (Oct. 1888 - June 1950), born in Copenhagen and educated in
Germany and other places in Europe, well known as a professor, first at the University of Illinois and later
at Harvard.
7
See [9] or, better still, Langhaar [8] for more details.
10-5 Energy Methods

(a) xyz system (b) An axial bar (c) A circular shaft (d) A cantilever

Figure 10.3: Three cases of ‘one-dimensional’ bodies are shown: an axially loaded uniform
bar [Fig. 10.3b], a uniform circular shaft subjected to a pair of torques T − T [Fig. 10.3c],
and a cantilever loaded by an end load, P [Fig. 10.3d].

for the strain energy gets simplified as

ZZZ
1
U= (σxx exx + τxy γxy + τyz γyz ) dx dy dz. (10.7)
2 V

There is really no need at all to integrate w.r.to the variable z, because there is no variation
(change) along the z-axis (perpendicular to the plane of the paper). The integration is
essentially only along one axis, say the x-axis; hence the qualification as ‘one-dimensional
bodies’. We can replace the strain components in Eq. (10.7) in terms of the stress compo-
nents using the constitutive equation (Hooke’s law) and obtain
!
ZZZ 2 2
τxy
1 σxx
U= + dx dy dz. (10.8)
2 V E G

This equation is now in good shape for use for all ‘one-dimensional’ bodies such as the ones
shown in Fig. 10.3. First we shall demonstrate how Eq. (10.8) may be used to obtain the
elongation of the bar shown in Fig. 10.3b.
An axially loaded bar:
For this case, there is only one component of stress, viz., σxx = P/A. Accordingly, the
expression for the total strain energy, U stored in the bar expressed as a function of the
axial load P in a form convenient for differentiation w.r.to P (to obtain the ‘work absorbing
component of the displacement’, which is the elongation of the bar) is

1
ZZZ 2
σxx
U = U (P ) = dx dy dz
2 V E
l
1 P 2
Z   ZZ
1
= dx × A, because dy dz = A
2 0 E A
l
P2 P 2l
Z
= dx = . (10.9)
0 2AE 2AE
Advanced Mechanics of Solids 10-6

The elongation is obtained by differentiation of this expression w.r.to the load P to yield
dU Pl
δ= = , (10.10)
dP AE
which we know is correct. [Some students may find this confusing. Which P are we
differentiating with respect to? And which elongation, the full elongation, or the elongation
of the half bar, is this δ? It is perhaps better to consider only one half (say, the top half) of
the bar. Because of symmetry, the middle cross-section X −X does not move; it stays where
it was with no movement. It can, therefore, be considered as fixed as shown in Fig. 10.3b.
Now the strain energy is only half of what is given by Eq. (10.9). Differentiating this w.r.to.
P 2l
dU1/2 Pl
we obtain Uhalf bar ≡ U1/2 = 4AE . Elongation of the top half of the bar = = .
dP 2AE
Pl
As there are two halves, the total elongation is δ = . This explanation, we hope, is
AE
clearer.] Next we shall consider a uniform circular shaft subjected to a torque.
A circular shaft in torsion:
Shown in Fig. 10.3c is a uniform circular shaft — really a cylindrical shaft; by a circular
shaft is meant a shaft with a circular cross-section — subjected to a pair of torques T − T .
We shall now derive the strain energy for this member.
We know from our earlier study of Coulomb’s theory of torsion (applicable only to
circular shafts) that the torsional shear stress τ on the cross-section is given by8 τ = T r/Jp .
With this, the expression for the strain energy, U works out to be
!
ZZZ 2
τxy
1
U= dx dy dz
2 V G
Z l
T2
Z
= dx, because dy dz = dA, and Jp = r2 dA. (10.11)
0 2GJ p A

The angle of twist, which is the ‘work absorbing component of the displacement’, is obtained
by differentiating this expression given in Eq. (10.11) w.r.to the ‘load’ T to yield
l l
T2 T2
Z Z  
dU d d Tl
θ= = dx = dx = , (10.12)
dT dT 0 2GJp 0 dT 2GJp GJp

as T, Jp and G are all constants here. We know that this is the correct formula.
The third and last structural member that we mentioned above is a cantilever. We shall
presently obtain the expression for the strain energy for this case.
A cantilever with an end load:
Consider a cantilever loaded by an end load P . We desire to obtain the expression for the
strain energy U and, thereby, to find the deflection at the free end.
8
Students are advised to review the theories of pure bending of beams, and of torsion of circular shafts, and
to see the similarity between the Euler-Bernoulli formula and the torsion formula in Coulomb’s theory of
torsion (with the usual notation), σ/y = M/I = E/R and σ/y = M/I = E/R and τ /r = T /Jp = Gθ/l.
10-7 Energy Methods

The expression for the strain energy [Eq. (10.8)] is reproduced here for convenience as
!
ZZZ 2 2
τxy
1 σxx
U= + dx dy dz.
2 V E G

First, let us calculate the strain energy associated with the bending moment. The bending
stress is given by the Euler-Bernoulli equation as σxx = M y/I. From this expression the
strain energy due to the (normal) bending stress is calculated as
!
ZZZ 2 2
τxy
ZZZ 2
1 σxx σxx
U= + dx dy dz = dx dy dz
2 V E G V 2E

My 2 M 2y2
ZZZ   ZZZ
1
= dx dy dz = 2
dx dy dz,
V 2E I V 2EI
Z l
M2
Z Z
= dx, because dy dz = dA, and y dy dz = y 2 dA = I.
2
0 2EI
Z l
M2
Thus, U = dx. (10.13)
0 2EI

An element of the beam. Two neighbouring cross-

sections, which were parallel before deformation, have
now rotated (still remaining as plane), through an an-
gle dθ. The distance between the two cross-sections
along the neutral axis is dx. This small length dx is
shown as fairly large for clarity.

In general, on the cross-sections, there are not only bending moments M − M 9 , but also
a direct thrust N and a shear force S. Let us note further that dx = R × dθ.
Eq. (10.13) may be obtained in a different way also. We know that the strain energy
stored is the work done by the bending moment on the ‘work absorbing component of the
displacement’, which is the rotation of the cross-section. The figure above shows two neigh-
bouring cross-sections of a bent beam. These, which were parallel before the deformation,
now after the deformation subtend an angle dθ. Thus, from the geometry, we have
dx dx M E EI
dθ = = EI , as = −→ R= .
R M
I R M
9
Actually the bending moments cannot be the same if there are shear forces, because the shear force is
equal to the rate of change (w.r.to the variable x) of the bending moment. However, here this introduces
no error. (Why?)
Advanced Mechanics of Solids 10-8

1 1
strain energy: dU = × work done by B.M. = × M × angle of rotation
2 2
1 1 dx M2
= M dθ = M = dx.
2 2 R 2EI
Z l
M2
Thus, the strain energy: U= dx.
o 2EI

Next, let us calculate the strain energy associated with the shear force. The shear stress
(cross shear) distribution across the cross-section (here rectangular for this example), we
know, is given by the equation as
Z d " 2 #
S 2 S d2 2
τxy = yb dy = −y
Ib y Ib 2

with the usual notation. Using this expression we can calculate the strain energy due to
the shearing stress (cross shear) as
h i2
Z Z Z S 2 d
2 − y 2
3 l S2
Z
1 2
strain energy: U = dx dy dz = dx. (10.14)
2 V 4I 2 G 5 0 AG
With these expressions for the strain energy, the deflection of the cantilever can be readily
calculated. We shall calculate separately the deflection at the free end due to (i) the bending
moment alone, (ii) the shear alone, and (iii) both the bending moment and the shear.
(i) Deflection due to bending moment alone
The bending moment along the beam [Fig. 10.3d] is given by
M = M (x) = P x (0 ≤ x ≤ l).
The strain energy UM stored in the beam due to the bending moment alone is
Z l Z l
M2 (P x)2
UM = UM (P ) = dx = dx.
0 2AE o 2AE

The deflection at the free end due to the bending moment alone is
Z l Z l
(P x)2 (P x)2 P l3
 
dUM d ∂
δM = = dx = dx = , (10.15)
dP dP 0 2EI 0 ∂P 2EI 3EI
which we know to be correct.
(ii) Deflection due to the shear force alone
We shall now calculate the deflection due to the shear force alone. We have seen that the
shearing force along the beam [Fig. 10.3d] is given by
S = S(x) = P (0 < x < l).
The strain energy US stored in the beam due to the shearing force alone is
3 P 2l
US = US (P ) = .
5 AG
10-9 Energy Methods

The deflection at the free end due to the shearing force alone is
Z l Z l
3 P 2l 3 P 2l 6 P l2
 
dUS d ∂
δS = = dx = dx = . (10.16)
dP dP 0 5 AG 0 ∂P 5 AG 5 AG

(iii) Deflection due to both the bending moment and the shear force
If both the effects (due to the bending moment and the shear force) are considered together,
the total deflection is
P l3 P l3 d2
  
6 Pl
δ = δM + δ S = + = 1 + 0.6(1 + µ) 2 . (10.17)
3EI 5 AG 3EI l

Comments
A few comments seem to be appropriate here. In Eq. (10.17), the first term is the deflection
due to the bending moment alone. The term within the double brackets { · · · } is the
‘correction’ to be added to account for the effects of shear. If we take some realistic figures
and calculate the numerical values, we can see that this second term is very small indeed
in almost all of the usual cases when the length (span) l of the beam is much larger than
its depth d.
A close look at this equation (10.17) rewards us with invaluable insights. In the case
of a long, slender beam, the deflection due to the bending moment is all that matters; the
contribution of the shear is very, very small. On the other hand, if we go to the other
extreme of a short, stubby beam, the contribution of the bending moment towards the total
deflection (which is admittedly much smaller than in the earlier case) is very, very small;
the deflection due to the shear force is all that matters. For the in-between case, both the
bending moment and the shear force do contribute to the total deflection. In other words,
while for most beams that we see in practice, the contribution of the bending moment to
the total deflection may still be the lion’s share, the effect of the shear force may not be
negligible if the dimensions of the beam are such as to make it fall in between the two
extreme cases pointed out10 . The in-between cases of beams, where the effect of the shear
force cannot be neglected in the calculation of the total deflection, are called Timoshenko
beams in honour of that outstanding engineer-scientist S.P. Timoshenko, because it was he
who showed how the effect of shear forces may be reckoned.
Let us also look at the same equation, and examine what it entails. We have added
two components of the deflection to obtain the total deflection. It is equivalent to adding
the strain energies UM and US , calculated one at a time, to get the total strain energy.
This raises an important question: are we permitted to add the strain energy due to each
component of the load, and add them to obtain the total strain energy? In other words,
can such strain energies be superposed? We shall examine this question.
10
Students are urged to do these calculations. The experience will improve our intuitive understanding of
what, and what not, are important in engineering calculations. We learn some topics, and forget the
details; still there is some residue left over. It is this residue that adds to our qualitative understanding.
This collective understanding gained over the years forms what can vaguely be referred to as engineering
judgement. Engineering judgement is no small matter; it is a component of overriding importance in the
professional competence of engineers. This is to be consciously cultivated.
Advanced Mechanics of Solids 10-10

The answer (in anticipation of the result of this inquiry) is: superposition of strain
energies is not permitted in general, but in certain special situations, it is permissible.
What we have seen above in Eq. (10.17) is one of such special situations.
Superposition of Strain Energies
We shall take up this matter in some detail, and come to a proper conclusion. First we
shall consider as an example a simple, special case of two axial (tensile) loads applied on a
uniform bar, one at a time, and later both together.
Why strain energies cannot be obtained by superposition:

(a) Tensile load P1 − P1 (b) Tensile load P2 − P2 (c) Loads (P1 + P2 ) − (P1 + P2 )

Figure 10.4: The strain energies, U1 due to the load P1 and U2 due to the load P2 cannot
be added (superposed) to obtain the total strain energy, U due to (P1 + P2 ): U 6= U1 + U2 !

Let us consider the case of a uniform bar loaded by P1 − P1 as shown in Fig. 10.4a. The
elongation of the bar and the strain energy stored are,
P1 P1 P1 L
stress: σ1 = ; strain: 1 = ; axial elongation: δ1 =
A AE AE
P12 L
strain energy stored: U1 = . (10.18)
2AE
In exactly the same way, if an axial load P2 alone acts on the bar, the strain energy U2
due to this load P2 alone is
P 2L
U2 = 2 . (10.19)
2AE
Now, if the two loads P1 and P2 act together, that is, if the load is P1 + P2 , the strain
energy U is similarly
(P1 + P2 )2 P 2L P 2L
U= 6= 1 + 2 (10.20)
2AE 2AE 2AE
showing that the total strain energy due to two loads is not equal to the sum of the strain
energies due to each load. Why is this so? Mathematically, this is because (P1 + P2 )2 =
10-11 Energy Methods

P12 + P22 + 2P1 P2 6= P12 + P22 , which is almost a silly explanation. The expression for the
strain energy is nonlinear (the square of the load appearing in the expression); we know
that superposition is not valid when there is nonlinearity.
This explanation is fine as far as it goes. But it is insightful when we examine this
from a physical point of view. The ‘cross-term’ 2P1 P2 is the villain; it is the presence
of this ‘cross-term’ that violates or spoils the property of superposition. The cross-term
corresponds to the work done by the first load during the additional displacement caused
when the second load is applied!
The moral of this demonstration is this. The total strain energy corresponding to two
(or more) loads cannot be obtained by calculating the strain energy corresponding to each
load, acting only one at a time, and adding them together.
If this is so, what shall we do to obtain the total strain energy?
Fortunately, the situation is not as hopeless as this demonstration seems to show. When
an axial load acts, there is an axial displacement, and there is a resulting strain energy.
Now if a bending moment is additionally applied, there is a corresponding ‘work absorbing
displacement’ which, for the case of a bending moment, is the rotation of the cross-section.
During this rotation, the axial load does not do any work. There is no ‘cross effect’ now,
and it is, therefore, perfectly legitimate to add the two strain energies, calculated one at a
time, to obtain the total strain energy! Superposition then does hold if the loads are such!
The reason is, to repeat: when the second load acts, and a corresponding ‘work absorbing
displacement’ results, the first load does not do any work!
Thus, in spite of the objection raised, it is indeed permissible to calculate the strain
energy because of each of the separate loads, and add them up (superpose them) to obtain
the total strain energy if there are no ‘cross-effects’.
We have answered the question raised above. We shall proceed to the next topic in the
next section. This concerns the potential of the external loads.
Potential of the External Loads (External Potential Energy)
We have discussed the topic of strain energy above. Let us note once again that, when the
external forces do work on a conservative system — bars in tension, shafts in torsion, beams
in bending, etc. are conservative systems; there is no energy dissipation — the entire work
done is stored as internal energy inside the body. The strain energy is the internal energy
stored in the structural members that we discussed above.
Strain energy is an important topic; some very useful theorems are based on this. There
are other theorems that are based on the total potential. The total potential (energy) is the
sum of internal potential energy (strain energy), and the potential (energy) of the external
loads. Therefore, we shall now see what the potential of the external loads is.
We shall explain the concept of the external potential energy, or the potential of the
external forces, with reference to the example of a cantilever with an end load P producing
an end deflection δ. There is only one applied external load, P . This load moves through the
distance δ (deflection of the free end). The external potential is −P δ. For this cantilever,
Advanced Mechanics of Solids 10-12

the various terms are the following.

L
M2
Z
strain energy (internal energy): U= dx
0 2EI
external potential (energy): V = −P δ
L
M2
Z
total potential (energy): π = V + U = −P δ + dx
0 2EI
Now there are a few questions here11 .
Influence Coefficients

(a) F1 and F2 acting at 1 and 2 (b) Several forces and moments

Figure 10.5: Two forces F1 and F2 are acting on a body at 1 and 2 producing deflections
(total) δ1 and δ2 [Fig. 10.5a]. Fig. 10.5b shows several forces and moments on the body.

Consider a linearly elastic body in equilibrium acted upon by a set of forces and mo-
ments, say, F1 , F2 , · · · ; M1 , M2 , · · · [Fig. 10.5b]. These produce linear and angular dis-
placements. To express these, it is convenient to use influence coefficients αij .
The influence coefficient αij is defined as the deflection (linear or angular, as the case
may be) at the point i when a unit load (force or moment, as the case may be) acts at
the point j. Thus, if a force F1 acts at the point 1, and another force F2 at the point
2, the total deflections at the points 1 and 2 are expressed as δ1 = α11 F1 + α12 F2 and
δ2 = α21 F1 + α22 F2 .
We must realise that the δ1 is the ‘work absorbing component’ of the force F1 . That is,
δ1 is the component (of the total displacement at 1) in the direction of F1 . The work done
by a force F corresponding to the ‘work absorbing component’ δ of the displacement is F δ1
if the full force moves through the full displacement δ. If, on the other hand, the force F
is gradually applied, then the work done is (1/2) × F × δ. Whether or not the factor 1/2 is
present in the expression must be understood clearly. These influence coefficients enjoy the
important property of symmetry, i.e., αij = αji , called the reciprocal relations.
11
Firstly, why the minus sign? After all the force P and its displacement δ are in the same direction.
Secondly, it would appear that the work done by the external force is (1/2)P δ. Some explanation seems
to be necessary to understand these confusing issues. The trouble arises because we call this the work
done. If we use the term potential, we do not have these difficulties. There is a strong case to use the
term (gravitational) potential instead of the more popular term potential energy.
10-13 Energy Methods

Betti-Maxwell Reciprocal Relation

We shall give presently a simple proof of this Betti-Maxwell reciprocal theorem12 .
Proof:
First apply a load F1 at a point 1 on the body [Fig. 10.5a]. Calculate the work done by this
load during the resulting displacement δ1 which is stored inside the body as elastic strain
energy. Next apply a second load F2 at another point 2 [Fig. 10.5a]. Calculate the work
done by the load F2 and also by F1 , and the total strain energy U1 stored in the body.
Now start all over again, and apply first a load F2 at 2, and then the load F1 at 1.
Calculate the total work done and hence the strain energy stored. Now we argue that the
total strain energy stored must be the same independent of the order of application of the
loads. Equating these two expressions for the total energy in the body, we arrive at the
desired result α12 = α21 . This is the procedure. We shall now carry out the calculations.
Step 1
When the first load F1 alone acts, the displacement is δ1 , and hence the work done (equal
to the strain energy stored) is 1/2 × F1 × δ1 because the load F1 is applied gradually.
1 1 1
U1 = F1 δ1 = F1 (α11 F1 ) = α11 F12
2 2 2
Now the second load F2 is applied gradually at the point 2. This produces a displacement
of (i) α22 F2 at the point 2, and (ii) α12 F2 at the point 1. During these displacements, the
work done by F2 is (1/2)F2 (α22 F2 ), while that done by F1 (which has its full presence, not
gradual) is F1 (α12 F2 ). Thus,
1
U2 = F2 (α22 F2 ) + F1 (α12 F2 ).
2
(Note the presence and absence of the factor 1/2 in the first and second terms, respectively.)
Step 2
Now start all over again. Apply on the virgin unloaded body, first F2 at the point 2, and
then F1 at the point 1. The work done during the first loading is
1 1
U10 = F2 δ2 = F2 (α22 F2 ).
2 2
The work done during the application of F1 is
1
U20 = F1 (α11 F1 ) + F2 (α21 F1 ).
2
(Note again the presence and absence of the factor 1/2 in the first and second terms,
respectively.)
Step 3
12
Opinion is divided about the priority. A.E.H. Love attributes this to Enrico Betti (1872), but states
that this is a special case of a more general theorem due to Lord Rayleigh (1873). Lord Rayleigh himself
acknowledges the reciprocal theorem to Betti. Neither Love not Rayleigh mentions Maxwell’s (1864) name.
The theorem is now generally known as the Betti-Maxwell reciprocal theorem.
Advanced Mechanics of Solids 10-14

Now arguing that the total elastic energy stored in the body cannot depend on the order
of application of the two loads F1 and F2 , we write

U = U1 + U2 = U10 + U20 = U 0 ;
1 1
i.e., α11 F12 + α22 F20 + α12 F1 F2 = α22 F22 + α11 F10 + α21 F1 F2 ,
2 2
from which we conclude that α12 = α21 .
More generally, when we have several loads some of which may be moments, αij = αji .
We emphasise that the displacements concerned are the work absorbing displacements. [We
must be careful not to be trapped into concluding the following. The angular displacement
(rotation) θ1 at the point 1 when a moment M2 is applied at the point 2 = the deflection
at the point 2 when a force F1 is applied at the point 1. This is completely wrong. Why?
The angular displacement θ is not a work absorbing component displacement of the force!]
LOOKING AHEAD
So far we have seen some basic concepts. From here on the subject becomes abstract and
difficult. We shall, therefore, just indicate the dual formulation, and then quickly pass on
to a simpler treatment.
Dual Formulation

(a) Linear load-deflection diagram (b) Nonlinear load-deflection diagram

Figure 10.6: Strain energy, U and complementary energy, U ∗ . Load-deflection diagrams:

linear [Fig. 10.6a] and nonlinear [Fig. 10.6b]. Note that P dδ = δU , and that δ dP = δU ∗
for both the linear and nonlinear cases.

We shall discuss this dual formulation briefly. In energy methods this is very important.
For example, corresponding to the principle of virtual work, there is its dual, the principle
of complementary virtual work. There are more of such pairs. We shall see the simplest
of them, namely, strain energy and complementary energy. The complementary quantities
are shown with stars / asterisks (∗ ).
A load-extension diagram is shown in Fig. 10.6. In general it is nonlinear even though
the material is still elastic. The area under the curve is the strain energy, U . The area above
the curve is called the complementary energy, U ∗ . For a linear material, U = U ∗ , because
the two areas are equal. Still it is not quite appropriate to write U = U ∗ , even though the
10-15 Energy Methods

two areas are numerically equal. The strain energy U is a function of the deflection d, while
the complementary energy U ∗ is a function of the load, P . Let us also note that U ∗ is not
an energy, even though it has the same dimension as an energy term.
Referring to the above load-extension diagram [Fig. 10.6] (which may be linear or
nonlinear as we have shown here in the two diagrams), we note that

dU
δU = P δd −→ = P ; and (10.21)
dd
dU ∗
δU ∗ = d δP −→ = d. (10.22)
dP
Note particularly that, to apply Eq. (10.21), U is to be expressed as a function of the
deflection (extension) d, while, to apply Eq. (10.22), U ∗ is to be expressed as a function
of the load P . It is not quite appropriate to write U = U ∗ , even in the linear case (when
the load-deflection curve is a straight line passing through the origin), and even if the two
areas are equal. This is because of the different functional relationships. These are (special,
simple, cases of) Castigliano’s first and second theorems.
Let us consider a relationship like W = P × d. We can (i) keep P unchanged and vary
d, or (ii) keep d unchanged and vary P , giving us

δW = P δd, or δW ∗ = d δP.

We shall call them the (variations of the) virtual work and the complementary work. Sim-
ilarly, in the expression
ZZZ
1
U= (σxx exx + σyy eyy + σzz ezz + τxy γxy + τyz γyz + τzx γzx ) dx dy dz,
2 V

we can keep the stress components σij unchanged, and apply a strain field variation as
ZZZ
1
δU = (σxx δexx + σyy δeyy + σzz δezz + τxy δγxy + τyz δγyz + τzx δγzx ) dx dy dz,
2 V

or, keep the strain components ij unchanged, and apply a stress field variation as
ZZZ
1
δU ∗ = (δσxx exx + δσyy eyy + δσzz ezz + δτxy γxy + δτyz γyz + δτzx γzx ) dx dy dz.
2 V

We can call them (variations of) the strain energy and the complementary energy13 .
13
To understand these matters properly we need to learn the Calculus of Variations and Legendre trans-
formations. We can see that the complementary energy U ∗ is obtained from the strain energy U by a
Legendre transformation. Corresponding to the principles of virtual work and total potential energy, we
can have the principles of complementary virtual work and total complementary energy. These are very
powerful methods. Several approximate methods are based on these.
Ambitious readers are advised to see [9], where several good books are referred to.
Advanced Mechanics of Solids 10-16

Comments
In every problem, two conditions must be uncompromisingly satisfied: (a) the forces /
stresses must satisfy the equations of equilibrium, and (b) the displacements / strains must
satisfy the compatibility conditions. Correspondingly, we can have two approaches. (a)
We can have several force distributions, all of them satisfying the equations of equilibrium,
but among them only one can be correct (a consequence of the uniqueness theorem), the
correct one corresponding to displacements that are compatible. Alternatively, (b) we can
have different displacement distributions, all of them satisfying the compatibility conditions,
but only one of them can be correct (again, a consequence of the uniqueness theorem), the
correct one corresponding to forces / stresses that satisfy the equations of equilibrium.
Energy theorems are of great help to us to choose the correct one directly without having
to compute (a) the displacements, or (b) the forces / stresses. If we keep this in mind, we
can have a better appreciation of the essential difference between the force methods and
the displacement methods. Corresponding to these two approaches, we have the dummy
displacement method and the dummy load method, and the two Castigliano’s theorems.
SIMPLER ASPECTS OF ENERGY METHODS
We shall now abandon these advanced concepts and theorems and take up the simpler
aspects of energy methods. Fortunately, many technically important problems can be solved
by using these relatively simpler methods.
We shall see three important results: (i) Castigliano’s14 first theorem; (ii) Castigliano’s
second theorem; (iii) the theorem of minimum strain energy; and (iv) the theorem of mini-
mum total potential15 .
Castigliano’s First Theorem
If, in an elastic body in equilibrium under the influence of several (generalised) forces Qi ,
the displacement field is changed (varied), the work done by these generalised forces during
these generalised displacements δdi is equal to the increase in the strain energy dU . The
theorem states that
n
X ∂U (di )
dU = Qi δdi −→ = Qi . (10.23)
∂di
i=1
14
Carlo Alberto Castigliano (Nov. 1847 - Oct. 1884) was an Italian engineer and mathematician.
15
The terminology used in the literature is not consistent; different authors use different names for these
theorems. Theorem of work, theorem of least work, etc. are commonly used, though often inconsistently.
Some authors associate the names of Menebrea, Engesser and Crotti-Engesser with these theorems. There
were also priority disputes between Castigliano and Menebrea.
To see the logical reasoning, we need to go the starting point, which is the principle of virtual work applied
to a deformable body, and proceed step by step. We come to the principle of minimum total potential,
and then to Castigliano’s first theorem ∂U (d)/∂dJ = Qj , on the one hand, and on the other, using the
dual formulations, start from the principle of complementary virtual work and come to the principle of
minimum total complementary energy, and then to Castigliano’s second theorem ∂U ∗ (Q)/∂QJ = dj . Here
all we can do is to be aware that different authors use different names, but to stick to our own terminology
consistently.
10-17 Energy Methods

The strain energy U is now to be expressed in terms of the displacements. [See Figs 10.6a,
10.6b, Eq. 10.21].
Castigliano’s Second Theorem
This is confusingly similar to the first theorem. But as we have explained the difference
earlier, we shall merely state the theorem without much explanation.
If the strain energy stored U is expressed as a function of the (generalised) forces Qi ,
then the work absorbing components of the corresponding displacements di are related by
∂U (Qi )
di = .
∂Qi
To apply this theorem, the strain energy is to be expressed in terms of the (generalised)
forces Qi .
Minimum Strain Energy Theorem

(a) A continuous beam (b) The beam without RC (c) Load RC on the beam AB

Figure 10.7: A continuous beam with a uniformly distributed load. If we remove the support
at C, we will have a simply supported beam as shown in Fig. 10.7b. RC can be considered
to be an applied load of the right magnitude as to make the total deflection at C to be zero.

This powerful theorem may be stated in a simple way as follows. Among all the stress /
load distributions that satisfy the equations of equilibrium, but not necessarily the equations
of compatibility, the true or compatible stress16 / load distribution is that which makes the
strain energy a minimum, the non-compatible stress / load distributions containing higher
strain energy than the true or correct one.
Let us take a concrete example to explain this. Consider a continuous beam with a
uniformly distributed load [Fig. 10.7a]. This is a statically indeterminate problem; the
equations of equilibrium alone cannot give us the solution17 . This means that the equations
of equilibrium are to be supplemented by an equation of consistent deformations. How shall
we obtain this?
Well, one way to obtain this is the following. Imagine that the support RC is removed.
Now the central deflection is (5 wl4 )/(384 EI) [Fig. 2.13b]. The support reaction is such as
16
By the term a compatible stress distribution is meant a stress distribution that leads to a compatible
displacement field.
17
If we draw a free-body diagram, we can see that this is a concurrent, coplanar system of forces. There
are three unknowns, RA , RB , RC , but only two separate (linearlyP independent) equations of equilibrium
available. Sum of the forces in the horizontal direction is zero ( Fx = 0) is a correct, but useless equation!
Advanced Mechanics of Solids 10-18

to make the net deflection at C = 0. Thus,

5 wl4 Rc l3
=δ= −→ RC = 0.625 wl.
384 EI 48 EI
Having obtained RC it is easy to obtain the other reactions. Each of them turns to be
(1 − 0.625)/2 = 0.188 wl.
Let us now argue this way. Let us set RC = 0. We can now calculate RA , RB and
the total strain energy U1 . Now set RC = 1. Calculate again RA , RB and the total strain
energy U2 . Thus, we set different values for RC and calculate the strain energy U in each
case. If we compare them, all these load distributions satisfy the equations of equilibrium.
If we calculate the net deflection at C, we will get the correct answer of δ = 0 only in
one case. That case, that case alone, is the correct one. That alone satisfies the condition
of compatibility, or the condition of consistent deformation. Here are, thus, various load
distributions. All of them satisfy the equations of equilibrium. Among them only one leads
to the correct displacement of zero at the point C. According to the theorem, this correct
load distribution has the minimum strain energy!
This, then, is a pointer. Let us call the central load as RC and calculate the reactions
and the strain all in terms of the (unknown) reaction RC . Thus, we obtain an analytical
expression for U in terms of RC : U = U (RC ). Now if we invoke the theorem, RC is such as
to make U a minimum, leading to dU (RC )/dRC = 0. The energy theorem has an in-built
mechanism to make sure that the compatibility condition is satisfied.
[Let us have a slight digression. In the same way, we can conceive of different displace-
ment distributions, all of them satisfying the compatibility condition. Among them, one
and only one, is correct. Which one is that? Physically, the one that leads to the equi-
librium requirement — the equation(s) of equilibrium must be satisfied. This corresponds
to the minimum of the complementary energy! Corresponding to every result, there is a
result in the dual representation also! We can see, understand, appreciate, and admire the
beauty and the structure of the energy methods. Unfortunately, that calls for a higher level
of sophistication. Some other time perhaps!]
Here is another interesting way of looking at the result. We used the third theorem here.
We could equally well have looked at the problem, argued that the strain energy U = U (RC ),
and applied the Castigliano’s second theorem which says that dU (RC )/dRU = δC = 0 (the
deflection at C in the work absorbing vertical direction is zero!) The equation that we use
is the same, but our argument is different! In the first case, we minimise the expression for
the strain energy, while in this second case, we calculate the expression for the deflection
using Castigliano’s second theorem and set it equal to zero!
A similar situation arises when we try to determine the reaction of a propped cantilever
[Figs 12.6a, 12.6b, 12.6c]. If we call the (unknown) prop reaction as R, and if we desire
to determine its value, we can write down the total bending moment, and hence the total
strain energy U in terms of this unknown R as U = U (R). We can write dU (R)/dR = 0.
The equation is the same, but we can argue differently. One way is to understand that
U = U (R) reaches its minimum value for the correct value of R, using the minimum strain
10-19 Energy Methods

energy theorem. The other way is to find the deflection at the end using Castigliano’s
second theorem. In either case, the equation that we obtain is the same. The case of a
yielding support also is not difficult to handle.
Minimum Total Potential Theorem
The strain energy and the potential of the external loads make up the total potential. The
stress components are generated from the strain energy, U (which is the potential of the
internal stresses), and the external loads from the potential, V of the external loads. In this
way the total potential of the system, π = U + V is introduced into the formulation. Then
we obtain the principle of total potential ready and convenient for applications18 .

π =U +V δπ = δU + δV = 0.

We may emphasise here that (i) the external forces must be statically compatible — this
is obvious; otherwise the body will not be in static equilibrium — and (ii) the constitutive
equation — elastic, not necessarily linear — is incorporated into this formulation.
A Simple Example of a Simple Linear Spring

(a) Linear load-deflection diagram (b) Nonlinear load-deflection diagram

Figure 10.8: Strain energy, U and complementary energy, U ∗ . Load-deflection diagrams:

linear [Fig. 10.8a] and nonlinear [Fig. 10.8b]. Note that P dδ = δU , and that δ dP = δU ∗
for both the linear and nonlinear cases. A spring also is shown alongside.

We shall consider a simple linear spring with a spring constant k [Fig. 10.8b — this is a
linear spring for this example] so that the load P acting on it, and the resulting compression
x are related by the relationship P = kx. We shall demonstrate how the theorems are used
in this simple case, perhaps the simplest case that can be imagined.

18
This follows from the principle of virtual work, which is valid for all materials — the constitutive equation
never enters into the principle of virtual work — restricted to elastic materials.
We can also argue backwards and show, using the calculus of variations, that the principle of total potential
is both necessary and sufficient for equilibrium. Actually Castigliano’s first theorem follows from the
principle of minimum total potential. As the dual formulation, we can define a complementary potential
π ∗ = U ∗ + V ∗ also, and we obtain the principle of minimum total complementary energy (δπ ∗ = 0). From
this follows Castigliano’s second theorem.
Advanced Mechanics of Solids 10-20

The strain energy in the spring, U = 12 kx2 . The load P moves down by the amount
of the deflection x. Thus, the potential of the external load, π = −P δ. (Note carefully
that (i) there is no factor 1/2, and that (ii) there is a negative sign.) To this system we
shall apply the various energy theorems. We shall obtain the (expression for) the deflection
(compression) of the spring, x.
Castigliano’s first theorem
To apply this, the strain energy, U is to be expressed in terms of x. Thus, we have
1 2
U= kx ;
2
dU P
= P; −→ kx = P ; −→ x = .
dx k
Castigliano’s second theorem
To apply this, the strain energy, U is to be expressed in terms of P . Thus, we have
 2
1 2 1 P 1 P2
U = kx = k = ;
2 2 k 2 k
dU P P
= x; −→ = x; −→ x= .
dP k k
The expression for the strain energy U expressed as a function of x is U = U (x) = kx2 .
[When this is expressed as a function of P (so as to be in good shape for differentiation
w.r.to P ), it is U = P 2 /(2k). Thus, U = U (x) = U [x(P )] = U1 (P ). We write U1 instead of
U here, because the functional form is different! As U = U (x) = kx2 /2, U = U (P ) stands
for U = kP 2 /2 which is clearly wrong!]19
Minimum strain energy theorem
In the context of this problem, this theorem does not give us any fresh insight. Thus, the
application of this theorem is not discussed here.
Minimum total potential theorem
The total potential, π of the system is the sum of (i) the internal potential which is the
strain energy, U and (ii) potential of the external load, V . They are
1 2 1 2
U= kx ; V = −P x; π = U + V = −P x + kx .
2 2
dπ P
=0 −→ −P + kx = 0; −→ x= .
dx k
Castigliano’s Second Theorem: An Example
The figure [Fig. 10.9a] shows a uniform (EI constant) cantilever of length l with a uniformly
distributed load w/unit length. It is supported by a rigid prop. We desire to calculate the
reaction R at B. We shall use Castigliano’s second theorem to solve this problem.
19
Even some learned authors do not distinguish between the two different functional forms. Granted that
they have no confusion. When they use it, we hesitate to call it a mistake. It is only a carelessly chosen
notation. But we unsuspecting readers may be misled.
10-21 Energy Methods

(a) A propped cantilever (b) Support removed and (c) A concentrated

loaded by a u.d.l. replaced by a reaction R load R at the end

Figure 10.9: A propped cantilever AB with a uniformly distributed load, w per unit length
propped by a rigid support. We shall use Castigliano’s second theorem to calculate the
support reaction, R.

To apply Castigliano’s second theorem, we need to calculate the strain energy U ex-
pressed as a function of the (unknown) reaction R. The strain energy U in the cantilever
is given by
2
l [Rx − w2x ] dx
Z l
M 2 dx
Z
U= = ;
0 2 EI 0 2 EI
1 R2 l3 w2 l5 Rwl4
 
= + − .
E 6 40 8

As the deflection at the rigid support B = 0, we obtain on applying Castigliano’s second

theorem,

dU (R) Rl3 wl4 3

deflection at B = 0 = = − −→ R = wl.
dR 6 EI 8 EI 8
This completes the solution.
Among these four theorems stated here, the last one is the most general and powerful
one. The other fact is that these four results — or, theorems if we like to call them so
— are not independent. To see, understand and appreciate these facts, we have to go to
the fundamentals, which is difficult with the limitations that we have. The power of the
theorem can be demonstrated only when it is applied to more important and more difficult
problems. We regret our inability to pursue these matters further.
UNIT-LOAD METHOD
The unit-load method is one of the useful energy methods by which the deflection and the
slope of a bent beam can be calculated relatively easily. (This method is applicable for
other structures also.) We shall first derive the basic working formula, and then illustrate
the method by a few illustrative examples in a later chapter.
We shall take up the case of a simply supported beam to explain this method. Shown in
Figs 10.10a, 10.10b and 10.10c is a beam simply supported at A and B. We shall consider
three sets of loading and calculate the corresponding (i) work done by the external loads,
and (ii) the internal energy (strain energy).
Case (a) Loads P1 , P2 , P3 only
Three vertical loads are gradually applied on this beam at the points 1, 2, 3 respectively.
Advanced Mechanics of Solids 10-22

(a) Three loads (b) A unit load at C (c) Four (3 + 1) loads

Figure 10.10: A simply supported beam with three sets of loading is shown: (a) three actual
loads; (b) a unit load at C; and (c) three actual loads and the unit load.

The corresponding deflections — the work absorbing deflections — are ∆1 , ∆2 and ∆3 [Fig.
10.10a]. The work done by these external forces on the beam are

3
1 X 1 1 1
Pi ∆i ≡ P1 ∆1 + P2 ∆2 + P3 ∆3 . (10.24)
2 2 2 2
i=1

What happens inside the beam? Well, stresses are developed inside the beam. P There are
strain energies associated with these stresses. These are of the form 1/2 S dL where
S is the axial force and dL is the associated elongation / contraction. As the system is
conservative, the work done by the externally applied forces will be, must be, equal to the
internal energy (which is the same as the strain energy) stored in the beam. Thus, we have

1 1 1 1 X
P1 ∆1 + P2 ∆2 + P3 ∆3 = S dL. (10.25)
2 2 2 2
Case (b) Unit load only
If, in the beginning, only a unit load 1 is applied gradually and vertically at the point C,
what happens now? This unit load produces a vertical deflection at C, and δ1 , δ2 , δ3 at the
points 1, 2, 3, respectively [Fig. 10.10b]. Again, as before, the work done by the externally
P unit load is 1/2 × (1) × dl. The corresponding internal energy (strain energy) is now
applied
1/2 s dl. Conservation of energy demands that

1 1X
× (1) × δ = s dl. (10.26)
2 2
Case (c) The unit load first, and then the three loads P1 , P2 , P3
Let us consider the case (b) (only the unit load applied), and additionally apply the external
loads P1 , P2 , P3 at the points 1, 2, 3, respectively [Fig. 10.10c]. The total deflections now
are:

at 1: (δ1 + ∆1 ) at 2: (δ2 + ∆2 ) at 3: (δ3 + ∆3 ) at C: (δ + ∆).

10-23 Energy Methods

The total work done by the external forces including the unit load is
       
1 1 1 1
×δ + × P1 × ∆1 + × P2 × ∆2 + × P3 × ∆3 + (1 × ∆) . (10.27)
2 2 2 2

Note carefully that there is no factor 1/2 in the last term. This is because the full unit load
was acting before P1 , P2 , P3 were applied. thus, the unit load performs work equal to (i)
(1/2) × 1 × δ, and (ii) (1 × ∆) when the loads P1 , P2 , P3 are applied. The total internal
energy (strain energy) is

1X 1X X
s dl + S dL + s dL.
2 2
Note again that there is no factor 1/2 in the last term for the same reason as explained
above. Equating the total external work done to the total internal energy, we obtain
       
1 1 1 1
×δ + × P1 × ∆1 + × P2 × ∆2 + × P3 × ∆3 + (1 × ∆)
2 2 2 2
1 X 1 X X
= s dl + S dL + s dL. (10.28)
2 2
[The alert attentive reader would have noted that the small letters such as δ, δ1 , δ2 , · · · and
u, dl correspond to the unit load, and that the capital letters such as ∆1 , ∆2 , · · · and S, dL
correspond to the externally applied loads P1 , P2 , P3 .]
From Eqs (10.25), (10.27), and (10.28) we obtain [by subtracting the sum of Eqs (10.25)
and (10.27) from Eq. (10.28)],
X
1×∆= s dL. (10.29)

S and s are the axial forces on an element; that is, S = σ dA. We shall now apply this
equation (10.29) to the problem of determining the deflection of beams.
To Find the Deflection of Beams
Let us apply this unit-load method to find the deflection of beams. To be specific, we
consider the problem of finding the vertical deflection at the point C when three given
loads P1 , P2 , P3 act on a simply supported beam [Fig. 10.10a]. Apply a unit load at C in
the vertical direction (that is, in the direction of the required deflection) [Fig. 10.10b]. We
begin with Eq. (10.29), and process in further to bring it to a ready-to-use form.
Let M and m be the bending moments at a typical cross-section XX for (i) the actual
loading [Fig. 10.10a], and when (ii) the unit load alone is applied, respectively. An element
M N of original length dx is shown on the cross-section (length exaggerated). The deformed
lengths (shown elongated; they are in the tension zone) are dx + dL and dx + dl respectively
for the cases (i) [Fig. 10.10a] and (ii) [Fig. 10.10b]. Consider Eq. (10.29) and work out the
expressions for s and dL as
my
s = σ dA = dA, and
I
Advanced Mechanics of Solids 10-24

stress ×dx force dx

dL = strain × dx = = ×
 E  area E
S dx My dx Mm
= × = dA × = dx
dA E I E dA EI

Substituting these expressions for s and dL in Eq. (10.29), we obtain

X X m y   M y  Z l Z M m y2
∆= s dL = dA dx = 2
dA dx
I EI 0 A EI
Z l Z Z l
Mm 2 Mm
= dx y dA = dx.
0 EI 2 A 0 EI

Thus, we arrive at the working formula for all such deflection calculations as
Z l
Mm
∆= dx (unit-load applied). (10.30)
0 EI

The procedure and the formula remain unchanged when we desire to obtain the slope
(rotation) also. Instead of the unit load, we now apply a unit moment at the desired point
in the desired direction. (The work absorbing displacement is the mantra!)
Z l
Mm
θ= dx (unit moment applied). (10.31)
0 EI

See a later chapter for a few worked out illustrative examples.

In the next chapter, we shall discuss the topic of torsion of non-circular bars.
Chapter 11

TORSION OF NON-CIRCULAR
BARS

Torsion of bars is a topic of great relevance to engineers. Torsion of circular prismatic bars
(shafts) is the most important topic, but that of non-circular bars also is of much interest
to engineers. We shall discuss the simpler aspects of this theory in this chapter.

(a) A grid of horizontal beams (b) A section loaded by P (c) The wing of an aircraft

Figure 11.1: Shown in the figure are three examples of non-circular cross-sections subjected
to twisting moments. A slope in the beam AB [Fig. 11.1a] appears as a twist in the beam
CD, and vice versa. In Fig. 11.1c, if the load P does not pass through the shear centre C,
the channel section will twist. Fig. 11.1c shows the cross-section of an aircraft wing which
is multi-cellular subjected to a twisting moment.

Non-circular prismatic bars are not used for transmitting power1 . But torsion is called
into play not only in shafts transmitting power. In a grid of integrally cast (horizontal)
beams [Fig. 11.1a], the slope of the beam AB manifests as a twist of a (horizontal) beam
like CD. Further, when a verticl load acts on a structural member, if the load does not pass
1
When a keyway is provided, the cross-section of a circular shaft becomes non-circular. It is of interest to
study the effect of a keyway in a circular shaft. The stresses will become very large at the two re-entrant
corners, and the torsional rigidity is slightly reduced.
Advanced Mechanics of Solids 11-2

through the shear centre C [Fig. 11.1b], bending of the beam is accompanied by twisting
(the twisting moment being equal to P e clockwise). Torsion of wings of an aircraft [Fig.
11.1c] as in torsional vibrations is another example when the cross-section is non-circular
and multi-cellular. Again, in the case of a balcony beam, the cross- sections are subjected
not only to bending moments, but also to twisting moments. The cross-sections in these
examples are non-circular. Thus, these examples should suffice to convince ourselves that
there is a strong case to learn this important topic of non-circular prismatic bars.
TORSION OF BARS
In general, we have the following cases to consider.

(i) Circular prismatic bars (The cross-section is the same everywhere.)

(ii) Non-circular prismatic bars (The cross-section is the same everywhere.)

(iii) Circular non-prismatic bars (For example, conical shafts)

(iv) Non-circular non-prismatic bars

The first is the easiest; Coulomb’s theory deals with this case. Analytical solutions are
available for the second and the third also. No analytical solution is available for the last
case. We shall take up the second case here which the topic of this chapter. However, it is
desirable to have a quick review of the Coulomb’s theory.
A QUICK REVIEW OF COULOMB’S THEORY

(a) A circular prismatic bar subjected to

twisting moments T − T (b) To explain Coulomb’s theory

Figure 11.2: A circular prismatic bar in torsion. Fig. 11.2b shows the details of the
deformation. A generator like BA deforms to the position BC when the cross-section
rotates through the angle θ. This introduces a shear strain φ as marked.

Fig. 11.2a shows a circular shaft in torsion.The cross-sections on the left rotate (twist)
in one direction, while those on the right rotate (twist) in the other direction. Somewhere
in the middle there is a cross-section XX that does not rotate. Let us hold on to this
11-3 Torsion of Non-circular Bars

cross-section and consider it a fixed2 . A generator BA, on deformation, becomes BC. The
straight radial line OA, on deformation, becomes a straight radial line OC. θ is the angle
of twist. Small cubes at A and C are shown enlarged.
The Coulomb’s theory, we recall, is developed on the basis of two facts, often stated
as assumptions3 : (i) cross-sections do not warp; and (ii) straight radial lines such as OA
remain, on deformation, as straight radial lines such as OC; they do not become curved.
From Fig. 11.2b we note that the arc length AC = R θ = l φ −→ φ = (R θ)/l,
and that DE = r θ = l φ −→ φ = (r θ)/l, where φ this time refers to shear strain
(corresponding to a general radius r).
This equation φ = (r θ)/l relates the displacement (angular displacement θ) and the
shearing strain φ. Using now the constitutive equation, we can obtain the expression for
the shear strain τ at any radius on the cross-section — this is really τzθ ) as
Grθ τ Gθ
τ = Gφ = −→ = . (11.1)
l r l
Here we have used the constitutive equation. Now we can consider equilibrium. The
condition of equilibrium demands that the applied twisting moment T is related to the
resisting shear stress τ by the equation
Z Z
τ 2
(τ ) × (2 π r dr) × (r) = dT −→ T = τ r dA = r dA (11.2)
A A r
Z Z
Gθ 2 Gθ Gθ
= r dA = r2 dA = J. (11.3)
A l l A l
Combining Eqs (11.1) and (11.3, we obtain the fundamental equation governing torsion of
circular bars (shafts) as4
τ Gθ T
= = , (11.4)
r l J
where J is the polar second moment of area (the polar area moment of inertia)=π d4 /32.
Note in particular that the points farthest from the axis Oz are stressed the most. We
emphasise that τ refers to the (torsional) shear stress on the cross-section (τzθ ).
Argument Based on Circular Symmetry
We had stated that, for the case of torsion of circular (prismatic) shafts, (i) cross-sections
do not warp, and that (ii) straight radial lines remain straight radial even after deformation.
2
Fixed in this limited sense. This is the cross-section marked XX in Fig. 11.2b. Actually when the
shaft rotates — rigid body rotation — every cross-section, including this ‘fixed’ cross-section XX, rotates.
Cross-section near XX rotate — we do not mean rigid body rotation — only a little bit; the farther the
cross-section is from XX, the more it rotates. We repeat for emphasis: we are not talking about the rigid
body rotation of the shaft as a whole.
3
Actually these are facts and not assumptions. Using an argument based on axisymmetry (rotational
symmetry), we can see the truth of these statements. See below [Fig. 13.7] and Den Hartog [3] for details.
4
This is analogous to the Euler-Bernoulli equation governing the bending of beams. The consequences are
also analogous: the material far from the axis is more effectively utilised. For this reason, hollow shafts
are much better. (In some cases, the gain may not be worth the extra effort and cost of having a hollow
cross-section.)
Advanced Mechanics of Solids 11-4

(a) Such warping is impos- (b) Such axisymmetric (c) Compare the two identical parts A and
sible. Why? warping also is impossible. B identically loaded.

Figure 11.3: Such warping as in Fig. 11.3a is clearly impossible. Even the axisymmetric
warping as shown in Fig. 11.3b is impossible. Why? In Fig. 11.3c are shown two identical
parts identically loaded. It follows that the deformations must be identical. This argument
shows us that cross-sections cannot warp in such axisymmetric cases.

Let us examine these facts in the light of the axisymmetry (circular symmetry) that this
problem enjoys.
Fig. 11.3a shows a circular (prismatic) shaft subjected to T − T represented by double-
headed arrows. A typical cross-section like P P cannot possibly warp and become pppppp.
Why? Because there is axisymmetry, all the points like a, b, c, d, · · · can deform only iden-
tically. If a moves up, e cannot move down. However, it may appear that the cross-section
can warp axisymmetrically and become like P qqqqqP which surface is still axisymmetric
[Fig. 11.3b]. If we imagine that the cylindrical shaft is cut into two parts A and B, and
take the lower part B, rotate it and keep it by the side of part A [Fig. 11.3c], we find
an anomaly. Here are two identical parts A and B, identically loaded. Can they deform
differently? This is impossible. Hence we conclude that cross-sections cannot warp!

(a) Straight radial lines like OA (b) Straight radial lines like OA (c) Straight radial lines like OA
remain straight like OB. cannot curve in like OA0 . cannot curve out like OA00 .

Figure 11.4: Straight radial lines like OA on the cross-section remain straight even after
deformation; they do not curve in like OA0 [Fig. 11.4b] or curve out like OA00 [Fig. 11.4c].

A similar argument leads us to conclude that straight radial lines like OA [Fig. 11.4a]
can only rotate to become another straight radial line like OB, but cannot curve in like OA0
11-5 Torsion of Non-circular Bars

[Fig. 11.4b]5 , or curve out like OA00 [Fig. 11.4c]. We can, thus, infer that cross-sections do
not distort in their own planes. It means that there is no shear stress τxy producing a shear
strain γxy ; that is, the right angle in the shaded small block in Fig. 11.4a will be preserved.
Torsion of Non-circular Prismatic Bars
Some of the early investigators tried to develop this theory based on the assumption that
cross-sections do not warp trying to imitate Coulomb’s theory, but they ran into difficulties.
We now know that this is only natural. For the present case of non-circular prismatic bars,
cross-sections are required to warp; they will, they must, warp6 .
Cross-sections Will Have to Warp!

(a) Shear stress components on the (b) The coordinate sys- (c) Shear stress components on the
cross-section at a boundary point P tem xyz used cross-section at a corner point C

Figure 11.5: Our objective is to show that cross-sections will warp. As a first step, we
show that the resultant shearing stress on the cross-section at a boundary point like P is
tangential to the boundary. We are referring to the shear stress component τzx . Fig. 11.5c
is to show that C is a dead corner with no shear stress like τzx or τzy .

We shall show this in three stages taking the example of a rectangular cross-section.
Step (1)
First we note that the resultant shearing stress on the cross-section at a boundary point
must be entirely tangential to the boundary. P is a boundary point on the cross-section [Fig.
11.5a]. The cross-section is the z−plane [Fig. 11.5b]. If possible, let the shear stress τ on
the z−plane (cross-section) be inclined. Then it will have two components, the tangential
component τzx and the normal component τzy . The normal component, if it exists, will be

0
5
The straight radial line OA cannot curve in like OA0 (inswinger), or curve out like OA0 (outswinger), to
use the popular cricket terminology!
6
Straight radial lines will still remain straight radial lines. There is no difference here. We say that cross-
sections do not distort in their own planes. In other words, there is no shearing strain like τrθ or τxy
where the z − axis is along the axis of the prismatic bar. But we can make this statement only after
the problem is solved. (In the case of a circular non-prismatic bar like a conical shaft, cross-sections do
not warp — they cannot; remember the argument based on axisymmetry — but straight radial lines will
become curved. In other words, cross-sections do distort in their own planes. This problem can be solved
analytically, but we do not do this here. Den Hartog [3] gives a delightful treatment of this case including
the so-called razor blade analogy.
Advanced Mechanics of Solids 11-6

accompanied by its complement τyz (‘shear and complementary shear’) acting on the y−
plane. But the y−plane is a free surface. On a free surface, there cannot be any shear
stress7 . Thus, τyz does not exist and, consequently, τzy cannot exist either. Thus, we
conclude that the resultant shearing stress on the cross-section at a boundary point will be,
must be, tangential!
Step (2)
With this understanding, let us consider a corner point, say, C on the cross-section. If there
is a shear stress on the cross-section (on the z−plane) at C, it will have the components τzx
and / or τzy . These, if they exist, will be accompanied by their complementary shear shear
stresses τxz and / or τyz which cannot exist because the x−plane and the y−plane are free
surfaces. Thus, we conclude that there cannot be a shear stress on the cross-section at a
corner point!
Let us not fail to notice this important conclusion. Coulomb’s theory seems to indicate
that the shear stress on the cross-section is a maximum at the farthest point. But no, it
is not a maximum; it is, in fact, zero! Thus, we have the eleventh commandment: “Thou
shalt not apply Coulomb’s theory to non-circular cross-sections!” It is not a minor mistake,
nor is it an approximation, but a major seriously wrong mistake to apply Coulomb’s theory
to non-circular cross-sections!

(a) Shear stress components on the cross-section (b) The shear strain at M and N is zero. Cross-
at a corner point C sections must necessarily warp!

Figure 11.6: To show that cross-sections must necessarily warp: careful examination of the
deformation leads us to conclude that the cross-sections have no choice except to warp!

Step (3)
We have seen that the corners are free of shear stresses. That is, τzy = τyz = 0 at the corner
M [Fig. 11.6a]. Now if the shear stress on the cross-section at M = 0, the corresponding
shear strain also is zero. The twisting introduces an angle φ as shown in Fig. 11.6b. If the
shear strain is zero, the angle of 90◦ will have to be preserved. Thus, the cross-section will
have to tilt up by an angle φ. A similar situation arises at the corner N too. The angle at
N will have to be preserved as 90◦ , making it necessary for the surface to tile down by the
angle φ. Thus the cross-section has no choice except to warp!

7
unless somebody is rubbing vigorously on the free surface as in massaging!
11-7 Torsion of Non-circular Bars

SAINT-VENANT’S THEORY
It was Barre de Saint-Venant who solved the problem correctly in 1855. He used a semi-
inverse method8 . That is, after examining the deformation, he makes a guess, an intelligent
guess, leaving a function as unspecified. We shall see this method below. The prismatic

(a) A cross-section at z from the ‘fixed’ cross-section (b) The u and v displacements of P

Figure 11.7: To make ‘an intelligent guess’ about the displacements u, v, w

bar is ‘fixed’9 at the cross-section10 z = 0. A typical point P moves to the point P 0 .

Because the cross-section warps, the point P 0 moves out of the plane. This is the warping
displacement w which is a function of x and y, but not of z: w = w(x, y). The fact that w
does not depend on z means that all cross-sections warp identically. The projection of P 0
on the cross-section is the point P 00 [Fig. 11.7b]. The angle through which the cross-section
rotates is β which is very small; cross-sections twist only by a small, very small, angle. This
is assumed to be proportional to z. That is, the farther the cross-section is from the ‘fixed’
cross-section, the larger is the angle of twist: β = θ z, where θ is the angle of twist per
unit length. The angle α may, or may not, be small. (It is small for points like P near
the x axis, but is not small for points farther from the x axis.) We do such an exercise
analysis to make ‘an intelligent guess’ about the displacements. If the cross-sections rotate
anti-clockwise, the u displacement is negative (i.e., to the left), while the v displacement is
positive (i.e., upwards).
Let us calculate the expressions for the displacements u and v.

Magnitude of u = OB − OA = OP cos α − OP 00 cos(α + β)

= x − OP (cos α cos β − sin α sin β)
= x − OP cos α + (OP sin α)β = θ yz.

Magnitude of v = P 00 A − P B = OP 00 sin(α + β) − OP sin α

8
The inverse method is to begin with an assumed solution and later to find out what problem it solves.
Here Saint-Venant employs this method, but with some freedom in the form of an unspecified function,
so that he can choose this function later in order that the equations of equilibrium are satisfied.
9
Recall what fixed means. See Fig. 11.2a and its explanation to understand what is meant by a fixed
cross-section.
10
Or rather, z is measured from this ‘fixed’ cross-section.
Advanced Mechanics of Solids 11-8

= OP (sin α cos β + cos α sin β) − OP sin α

= OP sin α + (OP cos α)β − OP sin α = θ yz.

(Note that OP 00 is almost equal to OP , because the warping displacement w is very small.
The angle of twist β per unit length is very small; hence sin β ≈ β and cos β ≈ 1.)
These expressions for the displacements do not seem to be unreasonable. Following
Saint-Venant, let us assume the displacements as:

u = −θ yz (11.5a)
v = θ xz (11.5b)
w = θ φ(x, y) (11.5c)

Saint-Venant at this stage leaves φ = φ(x, y) as unspecified. He still retains the freedom to
choose this function φ so that the equations of equilibrium are satisfied. Hence this method
is known as a semi-inverse method. The assumption behind this — φ being independent of
z — is that all cross-sections warp identically.
First Formulation
Let us begin by assuming the displacements as

u = −θ yz; v = θ xz; w = θ φ(x, y).

The strains, obtained from the strain-displacement relations eij = (1/2)[ui,j + uj,i ], are

exx = eyy = ezz = 0; γxy = 0;

   
∂φ ∂φ
γzx = γxz =θ − y ; γzy = γyz = θ +x .
∂x ∂y
The stress components are computed from the constitutive equations as

σxx = σyy = σzz = 0; τxy = 0;

   
∂φ ∂φ
τzx = τxz = Gθ − y ; τzy = τyz = Gθ +x .
∂x ∂y
Substituting these expressions in the equations of equilibrium σij,l = 0, we find that
∂σxx ∂τyx ∂τzx
+ + =0 (satisfied);
∂x ∂y ∂z
∂τxy ∂σyy ∂τzy
+ + =0 (satisfied);
∂x ∂y ∂z
∂τxz ∂τyz ∂σzz ∂2φ ∂2φ
+ + =0 −→ + 2 = 0 in R.
∂x ∂y ∂z ∂x2 ∂y
Thus, the governing differential equation and the associated boundary condition are
∂2φ ∂2φ
Laplace’s equation: + 2 = 0 in R, (11.6)
∂x2 ∂y
11-9 Torsion of Non-circular Bars

   
∂φ ∂φ
l −y +m + x = 0 on the boundary C. (11.7)
∂x ∂y
Saint Venant thus succeeded in converting the torsion problem to a boundary value problem
(a problem in partial differential equations to solve). Once the function φ is obtained, the
torsion problem is as good as solved, because the stresses, the strains and the displacements
are known. The twisting moment T and, thus, the torsional rigidity GJ, are also easily
calculated; they are only one integration away as given below [Fig. 11.8a].
ZZ  
T
torque: T = (x τzy − y τzx ) dx dy; torsional rigidity: GJ = .
R θ
ZZ  
∂φ ∂φ
= Gθ x −y + x2 + y 2 dx dy;
∂y ∂x
Z Z R 
∂φ ∂φ 2 2
J= x −y +x +y dx dy.
R ∂y ∂x

(a) The region R (b) ds, normal and l, m

Figure 11.8: The region R; a small part of the arc length (elemental arc length ds, normal
and the direction cosines l, m. Let us note that l = cos(n, x) = ∂y/∂s; m = cos(n, y) =
−∂x/∂s. The shear stresses τzx and τzy are marked on an elemental area dx dy; we can
then calculate the torque and thereby the torsional stiffness GJ.

Second Formulation
In the first formulation, the governing differential equation is the well known Laplace’s
equation (11.6); that creates no problem. But the associated boundary condition (11.7)
is not easy to satisfy. Thus, we can have another, a second, formulation in terms of an
auxiliary function ψ = ψ(x, y), so that the boundary condition is easier to satisfy.
Towards this objective, let us change over from the function φ to ψ such that φ + i ψ is
an analytic function of the complex variable x + i y. These two functions are related by the
Cauchy-Riemann equations
∂φ ∂ψ ∂φ ∂ψ
= ; and =− . (11.8)
∂x ∂y ∂y ∂x
Advanced Mechanics of Solids 11-10

Replacing φ by ψ as given above by Eqs (11.8), we obtain the governing differential equation
∂2ψ ∂2ψ
Laplace’s equation: + = 0 in R. (11.9)
∂x2 ∂y 2
[We can write down the above equation without any calculation, as we know that the real
and imaginary parts of an analytic function separately satisfy the Laplace’s equation.]
We shall see in a minute that the boundary condition in terms of ψ will be much simpler
and easier to deal with. In Eq. (11.7), let us make the following replacements.
∂y ∂x
l = cos(n, x) = ; m = cos(n, y) = − ;
∂s ∂s
∂φ ∂ψ ∂φ ∂ψ
= ; =− .
∂x ∂y ∂y ∂x
With these replacements, the boundary condition (11.7) becomes
   
∂y ∂ψ ∂x ∂ψ
−y − − + x = 0 on C;
∂s ∂y ∂s ∂x
 
∂ψ ∂x ∂ψ ∂y ∂x ∂y
i.e., + − x +y = 0 on C;
∂x ∂s ∂y ∂s ∂s ∂s
∂ψ ∂x ∂y 1 ∂ 2
i.e., =x +y = (x + y 2 ) on C;
∂s ∂s ∂s 2 ∂s
1
i.e., ψ = (x2 + y 2 ) + constant on C.
2
Thus, the governing differential equation in this formulation is
∂2ψ ∂2ψ
Laplace’s equation: + = 0 in R, (11.10)
∂x2 ∂y 2
and the associated boundary condition is
1 2
ψ= (x + y 2 ) + constant on the boundary C. (11.11)
2
If there is only one boundary — that is, if there are no holes, or which comes to the
same thing as stating that the region is simply connected — this constant can be chosen
conveniently as 0. This is because the slopes of the ψ hill are all that matter, and the slopes
are unaffected by adding a constant to ψ. With reference to the physical picture of the ψ
hill, we can readily see that the effect of adding a constant is to lift or lower the ψ hill,
which process will never change the slopes. However, if there are two or more boundaries
— i.e., if the region is multiply connected — the constant on any one can be chosen at
will (say, as zero), but the constants (different values on the different boundaries) are to be
chosen so that all is well11 . [This formulation is the basis of an electrical analogy12 .]
11
We avoid discussing the difficulties by stating that all is well. Additional conditions are necessary to
make sure that the displacements are single valued. There are special difficulties associated with multiply
connected regions.
12
We can exploit this analogy and obtain experimentally the torsional shear stresses on the cross-sections
11-11 Torsion of Non-circular Bars

Third Formulation
We have seen [page 11-8] that the first two of the three differential equations of equilibrium
are automatically satisfied, and that the third one becomes
∂τxz ∂τyz
+ = 0.
∂x ∂y
In such cases, it is a usual technique to define a function F = F (x, y) so that
∂F ∂F
τzx = τxz = ; and τzy = τyz = − .
∂y ∂x
Then we have
   
∂F ∂φ ∂F ∂φ
= −τzy = −Gθ +x ; and = +τzx = +Gθ −y ,
∂x ∂y ∂y ∂x
from which we see that 52 F = −2Gθ.
Thus, the governing differential equation and the associated boundary condition are
∂2F ∂2F
Poisson’s equation: + = −2Gθ in R, (11.12)
∂x2 ∂y 2
F = constant on the boundary C. (11.13)
If there is only one boundary, this C can be chosen as zero. Otherwise, the constant value
on any one can be chosen at will (as, say, zero), but the other ones have to be chosen so
that the displacements are single valued.
Alternative Poisson’s equation:
The problem may also be formulated by the Poisson’s equation and the boundary condition
∂2Φ ∂2Φ
Poisson’s equation: + = −1 in R, (11.14)
∂x2 ∂y 2
Φ = constant on the boundary C. (11.15)
This equation (11.14) can be brought to the form [Eq. (11.12)] by the change of variables
F = 2Gθ Φ. See the illustrative examples given in a later chapter 13.
TORSION OF A BAR OF RECTANGULAR CROSS-SECTION
The solution of this problem using a double infinite series is not easy. We shall not discuss
this. But we can obtain very satisfactory results by approximate methods. One such method
is shown, but not worked out in detail, below. This is not very easy either.
Let us obtain an approximate solution of the torsion problem — torsion of non-circular,
prismatic bars — for a rectangular cross-section [Fig. 11.9], governed by the Poisson’s
equation13
and the torsional rigidities. This is done by applying, on an electrically conducting sheet cut in the shape
of the cross-section to be investigated, voltages as demanded by the boundary condition (11.11), and
measuring the voltage gradients. These are interesting, but we do not propose to discuss here this and
other beautiful analogies in torsion.
13
This is based on the third formulation just given above.
Advanced Mechanics of Solids 11-12

The rectangular region inside ABCD is the domain

concerned D; C is its boundary. The coordinate
system is chosen as shown in order to exploit the
inherent symmetry in the problem. To solve the
torsion problem, it is necessary to solve the Pois-
son’s equation subject to the boundary condition
F (x, y) = 0 on C.

Figure 11.9: A rectangular cross-section

∂2F ∂2F
+ = −2Gθ in D (−a < x < a; −b < y < b)
∂x2 ∂x2
subject to the boundary condition
F = F (x, y) = 0 on C.
The corresponding functional to be minimised is
Z Z ( " # )
∂F 2 ∂F 2
 
1
I[F (x, y)] = + − 2Gθ F dx dy (11.16)
D 2 ∂x ∂y
We shall assume the solution in the form
XX
F = F (x, y) = (x2 − a2 )(y 2 − b2 ) cmn xm y n , (11.17)
m n

where the constants cmn are unknown. As we desire an approximate solution, we retain
only a few terms. Let us note that all these coordinate functions (i) satisfy the boundary
conditions, and (ii) are linearly independent. Because of the inherent symmetry in the
problem, m and n are to be even integers.
The method to determine the unknown coefficients cmn is as follows. Let us substitute
Eq. (11.17) in Eq. (11.16) and obtain an approximate functional as
  !2 !2  
ZZ 
1  ∂ F̃ ∂ F̃  
I[F̃ (x, y)] = + − 2Gθ F̃ dx dy. (11.18)
D 2 ∂x ∂y 

Now I is a function of the unknown coefficients cmn . To minimise the expression (11.18),
we need to differentiate the expression with respect to each of these coefficients cmn , and
set each of the derivatives to zero. Although we retain several terms in the assumed form
of the solution, in practice, we retain only a few terms, one, two or three. That is usually
quite sufficient.
To take a simple case for illustration, let us take a two-term approximation14 as
F̃ (x, y) = (x2 − a2 )(y 2 − b2 ) c0 + c1 (x2 + y 2 )
 
(11.19)
14
Actually, there are three terms; but it is perhaps less misleading to refer to this as a two-term approxima-
tion, because there are only two constants, c0 and c1 to be determined.
11-13 Torsion of Non-circular Bars

The two (necessary) conditions to minimise the functional (11.18) are

∂ ∂
I[c0 , c1 ] = 0; I[c0 , c1 ] = 0.
∂c0 ∂c1
Carrying out the calculations using maple, we obtain the results easily. For our purposes
here, the problem is solved completely15 .
There are some beautiful analogies in torsion. It is necessary for us to be familiar with
at least a few of them. We shall discuss them below.
ANALOGIES IN TORSION
There are a few analogies that are relevant here. The basis of the analogies is the mathemat-
ical similarity of the governing differential equations and the boundary conditions. These
analogies are useful in several ways. It is always academically exciting to see common
threads that run through apparently unrelated areas. Apart from this intellectual thrill
and pleasure, they are also useful in a practical sense. Sometimes an analogy helps us to
visualise the solution of a problem. For example, there is no way we can visualise the vari-
ation of the shear stresses or the torsional rigidity of a cross-section directly. Φ = Φ(x, y)
(third formulation) can, however, be conceived of an a Φ hill by appealing to the membrane
analogy. Additionally, theses analogies help us to have new experimental techniques to
solve torsion problems. Electrical analogy, fluid flow analogies (there are different analo-
gies), membrane analogy (soap film analogy), sand heap analogy, and razor blade analogy
are well developed. Several important results were obtained in the early years using exper-
imental techniques based on these analogies. We shall see some of them briefly.
Electrical Analogy
Let us refer to the second formulation. The governing differential equation and the boundary
conditions are
∂2ψ ∂2ψ
+ = 0 in R,
∂x2 ∂y 2
15
However, if we are interested in the topic of torsion, it would be nice to calculate the maximum shear
stress on the cross-section, and the torsional rigidity of the cross-section, and compare them with (i) the
exact values and with (ii) one-term and three-term approximations.
We recall from our earlier studies that the torque, T is given by
ZZ
T =2 F̃ (x, y) dx dy.
D

Although detailed calculations and comparisons cannot be made here, we shall still indicate that the error
is only of the order of 1 %, which is eminently satisfactory for such engineering calculations.
Even a one term approximation Φ(x, y) = c1 (a2 − x2 )(b2 − y 2 ) gives us
5 T
c1 = ; = ···
4(a2 + b2 ) θ
excellent results. This is a measure of how good the approximation is.
Advanced Mechanics of Solids 11-14

1
ψ = (x2 + y 2 ) + C, a constant on the boundary C.
2
We shall consider only simply connected domains16 , that is, cross-sections that have only
one boundary. Then the only constant can be chosen at will as zero. The effect of this is only
to raise or lower the ψ hill. The absolute value of ψ has no importance; only its derivatives
(that stand for the slopes) are important, being related to the shear stress components.
Once ψ = ψ(x, y) is known, all the relevant information such as the shear stress, the torque,
and the torsional rigidity can be computed.
The basis of the analogy is the realisation that that the voltage distribution in an elec-
trically conducting sheet (homogeneous and isotropic) also satisfies the Laplace’s equation.
Thus, if an electrically conducting sheet is cut in the shape of the cross-section to be investi-
gated, and the voltages applied on the boundary in accordance with the boundary condition
V = (1/2)(x2 + y 2 ), and if the voltage gradients ∂V /∂x) and ∂V /∂y are measured with a
two-point probe, we can experimentally solve the torsion problem.
∂2V ∂2V 1
+ = 0 in R, V = (x2 + y 2 ) on the boundary C.
∂x2 ∂y 2 2
Electrolytic tanks and commercially available conducting sheets were used in the past.
There are several details that cannot be discussed here.
Fluid Flow Analogy
We shall discuss a fluid flow analogy. The contour lines of the stress function for twist [Fig.
11.10]17 were interpreted to be analogous to the streamlines of an incompressible fluid18 .
From fluid mechanics we know that the velocity components vx and vy are related to the
stream function Ψ = Ψ(x, y) by the relation
∂Ψ ∂Ψ
vx = − ; vy = , (11.20)
∂y ∂x
∂vx ∂vy
so that the continuity equation + = 0 is automatically satisfied. If the expression
∂x ∂y
for the vorticity ω of the fluid  
1 ∂vx ∂vy
ω= −
2 ∂y ∂x
16
Simply connected domains have only one boundary. They correspond to cross-sections with no holes. In
general, there can be holes and, therefore, several closed boundaries. The constant C will have to be
different on each boundary and in such a way that all is well (that is, the displacements are single-valued).
Some additional conditions are to be satisfied. These are not always easy. Consideration of multiply
connected regions entails complexities that we cannot handle here. The ambitious reader can begin by
examining the behaviour of a function like log z, where z is a complex number.
17
These figures and the following explanatory caption are taken from Den Hartog [3]. “Three cross sections
with the contour lines of the stress functions for twist. In the corner points marked A the stress is zero;
these corners can be pared away without changing the stiffness of the section; the points marked B are
those of maximum stress; the bottom of the keyway has a stress which depends vitally on the fillet radius,
and is sure to reach the fatigue limit for even a small alternating torque in the case of a sharp corner.”
18
This analogy was proposed by William Thompson (June 1824 - Dec. 1907), better known as Lord Kelvin.
He was a famous British mathematician, mathematical physicist, and engineer of Irish origin.
11-15 Torsion of Non-circular Bars

Figure 11.10: Three cross-sections with the contour lines of the stress function for twist.
The corners A in all three cases are dead or stress-free. The ‘mid-points’ B on the boundary
‘closest to the centre’ have the maximum shear stresses. At the reentrant corners B of the
keyway the stresses are theoretically infinite, but actually very large. They are potential
sources of failure unless sufficiently generous fillets are provided. It is insightful to relate
these facts to the velocities in the fluid flow analogy.

is used along with Eq. (13.43), we obtain

∂2Ψ ∂2Ψ ∂2Φ ∂2Φ

+ = −2 ω which is analogous to + = −2 Gθ
∂x2 ∂y 2 ∂x2 ∂y 2
governing torsion (third formulation). We also know from fluid mechanics that the lines
Ψ = constant are the streamlines. It is thus clear that Ψ = constant along the boundary.
Thus, the differential equations and the associated boundary conditions are similar, and the
Kelvin fluid flow analogy is established. Its primary use is not as a computational tool or
as a basis to perform experiments, but as a means of visualisation. Where the streamlines
are crowded together, and where they are well separated, · · · [What happens? Complete
the sentence.]
Prandtl’s Membrane Analogy (Soap Film Analogy)
This is by far the most important and useful analogy. This is due to Prandtl19 . We shall
discuss this in some detail. This is based on the third formulation.
The governing equation and the boundary conditions are

∂2Φ ∂2Φ
+ = −2 Gθ in R, and (11.21)
∂x2 ∂y 2
Φ = constant on C; = 0 if there is only one boundary. (11.22)

If there is only one boundary (simply connected region, cross-section with no holes), the
constant may be chosen at will, say, 0 for convenience. However, if there are several bound-
aries, the constant C will, in general, be different for different boundaries. These different
constant values on the different boundaries (multiply connected regions, cross-sections with
19
The great Ludwig Prandtl (Feb. 1875 - Aug. 1953) of Göttingen, often considered as the father of
aerodynamics, and famous for his work in fluid dynamics (boundary layer theory)!
Advanced Mechanics of Solids 11-16

holes) have to be chosen so that all is well (the displacements are single valued). As these
calculations are not always easy, we shall consider only singly connected regions (cross-
sections with no holes) for the most part.

(a) Holes cut out on a flat plate (b) A schematic arrangement for the soap film

Figure 11.11: The cross-section to be analysed (say, an elliptical hole) and a circular hole
are cut out from a flat plate [Fig. 11.11a]. A small air pressure is applied below the same
soap film stretched across the two holes [Fig. 11.11b]. Optical methods are used to make
the necessary measurements (slopes, and volumes under the soap films).

Our first job is to establish the analogy. A hole is cut out on a flat plate in the shape
of the section to be investigated, and a membrane — a soap film — is stretched across the
hole, and a slight air pressure applied from below [Fig. 11.11b]. The soap film bulbs up.
Alongside the (say, elliptical) cross-section, a circular hole also is cut out and the same soap
film is stretched across this circular cross-section also.
We shall show that (i) the differential equation governing the shape of the soap film and
(ii) the boundary condition are, respectively,
∂2h ∂2h p
2
+ 2 = − , and h = 0 on the only boundary. (11.23)
∂x ∂y S
The boundary condition is obvious; the soap film cannot have any height on the boundary.
We shall now derive the above equation (Poisson’s equation) for the shape of the soap film.
Considering an element of the membrane (soap film) [Fig. 11.12], we note that it is acted
upon by (i) a surface tension T per unit length on all the edges, and (ii) an air pressure
p. If we assume that the surface tension T is relatively large, and that the height of the
soap film h = h(x, y) is small, we can see that the surface tension T can be regarded as a
constant. However, the slopes are not equal on all the edges and, therefore, the equation
of equilibrium in the vertical direction can be written down taking into account the slopes
also. The terms in the equation of equilibrium are [Fig. 11.12b]
 
∂h ∂h ∂ ∂h
−(T dy) + (T dy) + dx .
∂x ∂x ∂x ∂x
A similar pair of terms also is present. This is
 
∂h ∂h ∂ ∂h
−(T dx) + (T dx) + dy .
∂y ∂y ∂y ∂y
11-17 Torsion of Non-circular Bars

(a) Forces acting on an element (b) Forces on an element of the soap film

Figure 11.12: The forces on an element of the soap film are shown. There is an air pressure
p dx dy acting upwards. The surface tension has vertical components. These are the two
forces to be reckoned when we derive the (differential) equation for the shape of the film.

Additionally, there is the upward force acting on the element due to the air pressure p dx dy.
When all these terms are reckoned, the equation of equilibrium in the vertical direction is
  
∂h ∂h ∂ ∂h
−(T dy) + (T dy) + dx +
∂x ∂x ∂x ∂x
  
∂h ∂h ∂ ∂h
−(T dx) + (T dx) + dy + p dx dy = 0,
∂y ∂y ∂y ∂y
which on cleaning up appears as
∂2h ∂2h p
2
+ 2 =− . (11.24)
∂x ∂y S
This is the governing equation defining the shape of the blown up or bulbed up soap film.
The boundary condition is h = h(x, y) = 0 on the boundary. The soap film cannot have
any height on the (only) boundary. [When there are several boundaries as in a multiply
connected region corresponding to a cross-section with holes, there are changes as explained
earlier. We shall for the time being consider only simply connected regions.] Eq. (11.24)
may be interpreted as: the sum of the curvatures (approximate expressions, because the
height h and the slopes are small) is a constant everywhere on the membrane.
The analogy is established if the differential equations and the boundary conditions in
the two cases [the torsion problem, Eq. (11.21), and the soap film, Eq. (11.24)] are similar.
The analogous quantities are
p
(i) → 2 Gθ (ii) h → Φ (iii) slope of the soap film → shear stress.
S
The slope of the soap film in any direction is analogous to the shear stress (on the cross-
section) in the perpendicular direction. For example, the slope in the x direction ∂h/∂x →
τzy and the slope in the y direction ∂h/∂y → τzx . In addition, (iv) twice the volume
under the soap film → torque, T . We know that
ZZ ZZ
T =2 Φ dx dy which is analogous to 2 h dx dy,
R R
Advanced Mechanics of Solids 11-18

(a) Soap film (elliptical section) (b) Soap film (circular section)

Figure 11.13: The soap films corresponding to (a) an elliptical and (b) a circular cross-
sections. The slope of the soap film is analogous to the shear stress in the perpendicular
direction, and twice the volume to the torque. The same soap film is used for both sections
so that the surface tension and the air pressure are the same for both. The circular section
serves as a reference for comparison, because its analytical solution is easy and known.

which is twice the volume under the soap film.

This analogy helps us greatly in visualising the slope of the Φ hill, and to have an
excellent understanding of the qualitative nature of the shear stresses and the torsional
rigidity even without making any calculation. [For example, we know that the resultant
shearing stress on the cross-section at a boundary point must be tangential to the boundary.
This fact can be readily seen. Along the boundary, the slope ∂h/∂s = 0 which, when
translated to the torsion problem gives us the aforementioned result.] Additionally simple
calculations on the shape of the soap film give us valuable information on the corresponding
torsion problem. In the early days20 , several important cases were solved experimentally.
Sand Heap Analogy
This refers to the case of plasticity when the entire cross-section has gone into yield. This
analogy is not difficult to understand. It is an extension of the idea of the membrane analogy.
The cross-section to be investigated is cut on a plate — this section is not removed as for
the membrane — and sand is heaped on the cross-section until the sand spills over to the
ground or to a base plate placed below. The slope of the sand heap will be always the same
everywhere; that is a property of all soils including sand. The slope corresponds to the
yield stress in shear. If the same experiment is repeated on a circular plate, the ratio of the
weights of the sand collected on the two cross-sections gives the ratio of the torques and,
thus, of the torsional rigidities of the two cross-sections. As the torsional rigidity of the
circular cross-section is known, that of the given cross-section, say, elliptical, can be thus

20
There are several experimental details which are of decisive importance when actual experiments are
conducted. The usual soap that we use when bathing is not suitable; the film is not stable and it soon
collapses. Suitable substitutes were developed. The slopes are measured by optical methods. Experiments
were developed to cover the more difficult cases of multiply connected regions.
Those were the early days of aeronautics when Prandtl was in Göttingen, the days of furious academic
activity, the golden years of Göttingen. G.I. Taylor & Co. in England also contributed substantially to
these developments.
11-19 Torsion of Non-circular Bars

experimentally determined. This is known as Nadai’s21 sand heap analogy. It is possible

to cover the case of partially plastic bars in torsion also22 .
A Long Narrow Rectangular Cross-section

(b) The air pressure and the surface

(a) The narrow long rectangle and the soap film tension on the soap film

Figure 11.14: A long narrow rectangular cross-section and the shape of the corresponding
soap film are shown. All the cross-sections, except at the two ends, are identical. On the
right is shown [Fig. 13.8b] a section to enable us to write the equation of equilibrium.

We shall demonstrate how the membrane analogy may be made use of to analyse the
torsion problem of a long narrow rectangular cross-section. The solution of this problem is
of great practical importance. Most structural steel cross-sections (except those in the form
of a box) can be analysed using the results of this problem.
We wish to examine a long narrow rectangle [Fig. 13.8a] (length l and breadth (thick-
ness) t with l >> t) using the membrane analogy. The shape of the corresponding soap
film is shown shaded. Two views of this are shown, both shaded. We can see that the soap
film has the same cross-sectional shape everywhere except at the two ends. This fact makes
the problem one-dimensional (if we disregard the two ends), and the governing differential
equation becomes simpler: from a partial differential equation [two independent variables
x and y; h = h(x, y)] to an ordinary differential equation [only one independent variable y;
h = h(y)]. We shall now derive this (ordinary) differential equation.
Consider a part of the soap film between the sections P P and QQ of width y + y =
2y separated by a unit distance. We note that this element of the soap film of length
unity along the x−axis (between the sections P P and QQ) and of width 2y is acted
upon by (i) the air pressure to the right [Fig. 13.8b] (actually upwards), and (ii) the
component of the surface tension S per unit length acting on the two edges to the left
[Fig. 13.8a] (actually downwards). Writing down the equation of equilibrium, we obtain
dh dh
p × 2y × 1 = −2S × ×1 −→ = − Sp y.
dy dy
21
Named after Arpad Ludwig Nadai (April 1883 - 1963) — born in Hungary, higher education in Germany,
settled down in the US — was a Hungarian American mathematician and engineer.
22
“It is only necessary to erect a fixed roof of constant slope over the membrane with the outer boundary
of the section as its base”, writes Rodney Hill, the famous author of the famous book, The Mathematical
Theory of Plasticity, Oxford University Press, (1950).
Advanced Mechanics of Solids 11-20

Figure 11.15: The theory developed for a long narrow rectangle is applicable to all open
sections (with appropriate minor modifications depending on the context). Five examples
are shown: I, angle iron, channel, Z and a thin ring with a small slit.

On integrating this, we have h = −(p/S) y dy = −(py 2 )/(2S)+ a constant. This

R

constant is determined using the boundary condition: at y = ±(t/2), the height h = 0.

Thus, the constant is evaluated as p/(2S) (t2 )/(4). The shape is a parabola, obtained as
the solution
p t2
 
2
h= −y . (11.25)
2S 4

The maximum slope / the maximum shear stress is clearly at the edges y = ±t/2.
 
dh p t
= −→ τmax = Gθ t, (11.26)
dy max S 2

invoking the membrane analogy to refer to the torsion problem. The volume under the soap
film =(2/3)× base × height (parabola) × length is

p t2 2Gθ t2
   
2 T 2
= t l −→ = t l.
3 2S 4 2 3 2 4

The torsional rigidity (or torsional stiffness) C is, accordingly,

T t3 l
C= =G . (11.27)
θ 3
Eliminating Gθ using Eqs (11.26) and (11.27), we obtain
3T 3T
τmax = Gθ t = −→ τmax = . (11.28)
t3 l t3 l

Application to Structural Members

We had stated that the above problem of torsion of a long narrow rectangle is of much
practical importance. It is applicable for all open sections such as I, angle iron and channel
sections used widely as structural members [Fig. 13.29], but not for closed nor for box
sections. The length l referred to above is now to be interpreted as the total length of all
the legs comprising the open section. Visualisation of the corresponding soap film will help
us understand all this. There will surely be stress concentration, and a corresponding slight
reduction in the height and volume of the membrane (soap film) at the ends, but these
11-21 Torsion of Non-circular Bars

factors do not prohibit using the theory developed to all such open sections. [While on
this topic let us remark in passing that we should distinguish between two kinds of corners:
(i) reentrant corners where the stress is very high (severe stress concentration), and (ii)
dead corners which are free of stresses23 . We are talking about the shear stresses on the
cross-section due to torque. At the ‘dead corners’ there can be bending stresses (which are
normal stresses, tensile or compressive). The fluid flow analogy also helps us understand
the difference between reentrant and dead corners. The velocity at a reentrant corner is
very high; the streamlines close in there. In contrast to this situation, the velocity at a
‘dead corner’ is zero. It is also possible, on the basis of some reasonable assumptions, to
analyse the state of stress when a fillet is provided at a reentrant corner.]
What happens if the various legs in the sections shown in Fig.13.29? Well, Eq. (11.27)
still applies, but the torque (as well as the torsional rigidity) has to be calculated separately
for each leg and added up. A little reflection on the shape of the soap film will tell us what
to do in all such cases. The concept of shear flow based on the fluid flow analogy also will
help us.
The results C = T /θ = (G l t3 )/3 and τ = 3 T /(l t2 ) are significant, and are pointers to
a better qualitative understanding of the problem. It is the points nearest to the centre that
have the largest shear stress, and not the farthest points from the centre! This conclusion is
completely at variance with the implications of the Coulomb’s theory that it is the farthest
points that are stressed the most! Equally important is the fact that the torsional rigidity
(stiffness) C = T /θ = G lt3 /3 increases only as the first power of l. The polar second
moment of area increases as the cube of l, which fact warns us not to apply or extrapolate
the conclusions from the Coulomb’s theory to long rectangular cross-sections.
HOLLOW THIN-WALLED SECTIONS
From a practical point of view, hollow thin-walled sections are very important. Steel box
girders, thin-walled closed tubes, wings and fuselages of aircrafts are but a few examples.
The theory now is relatively simple. Fig. 11.16a shows a thin-walled closed section and the
corresponding membrane. The plate is to be horizontal because the height h (analogous
to the function Φ in the third formulation) must be a constant on every boundary. As the
wall is thin, we may regard the slope of the soap film to be almost constant across the wall
thickness. Thus, it follows that τ × t = constant. (Why? The slope ≈ h/t → τ .) This
relationship leads us to conclude immediately: that where the thickness is small, the shear
stress τ is large, and vice versa. This is analogous (similar) to the case of an incompressible
fluid flowing in a channel (of constant depth) when the continuity equation demands that
the velocity, v × t, thickness = constant. Thus arises the concept of shear flow.
Considering the (vertical) equilibrium of the horizontal plate [Fig. 11.16a], we have
I I
h p h
pA = S ds −→ = ds
C t S C t
(S × ds is the surface tension; h/t is the slope.)
23
This book is not on design; it is on analysis. Even so, let us point out that designers shall be careful about
reentrant corners, and that generous fillets at such vulnerable places should be provided.
Advanced Mechanics of Solids 11-22

(a) A thin-walled closed section (b) To derive the governing equation

Figure 11.16: A thin-walled closed section and the corresponding membrane are shown in
Fig. 11.16a. The plate is maintained horizontal. The expression for the elemental shear
force is [Fig. 11.16b] τ × ds, and its moment about O is τ × ds × n. Let us also note that
the area of the elemental triangle shown shaded is half the base multiplied by the relevant
perpendicular distance = 1/2 × ds × n.

This equation, when translated to the torsion problem, yields

I
1
2 Gθ = τ ds. (11.29)
A C
The torque T is [Fig. 11.16b] the moment of the elemental force (τ × t × ds) about the
point O integrated along the cell. Accordingly,
I I
T = (τ × t) × n ds = (τ × t) n ds = (τ × t) × (2A).
C C

Although τ and t may vary from point to point, the product (τ × t) is a constant (equal to
the height of the horizontal plate). Further (n ds) is twice the area of the shaded triangle
[Fig. 11.16b]. Thus,
T
T = (τ × t) 2A −→ τ= , (11.30)
2 At
where A is the area of the plate. Infinitesimal areas like (n × ds) represented by the shaded
triangle, when integrated (added up), give us the area of the plate.
Using Eqs (11.29) and (11.30), we find
4 G A2
I
T ds T
θ= −→ C= = H ds . (11.31)
4 G A2 C t θ t
If the wall thickness t is a constant which is often the case, this equation is simplified as
TL T 4 G A2 t
θ= −→ C = = , (11.32)
4 G A2 t θ L
where L is the total length — perimeter – of the cell. These equations suffice to solve torsion
problems of hollow thin-walled sections24 .
24
The result τ = T /(2 At) is known as Bredt’s formula, named after the German engineer Rudolph Bredt
(April 1842 - May 1900).
11-23 Torsion of Non-circular Bars

Sections with Two or More Cells (Multicellular Sections)

(a) A two-cell closed section (b) A symmetrical two-cell section

Figure 11.17: A multicellular (two cells) cross-section. The shape of the corresponding
membrane and the shear flows are shown. Fig. 11.17b shows a symmetrical cross-section.
Now there is no shear flow, and no shear stress in the middle leg (wall).

Let us now investigate the case of multicellular cross-sections. We shall consider a two-
cell section (triply connected region; there are three boundaries) [Fig. 11.17]. The (two)
horizontal weightless plates are now at different heights25 . This is because there are three
boundaries for the section, as the region is multiply (triply) connected; the constant values
on the boundary, we recall from the boundary condition in the third formulation, will be
different. The shear stress in the three walls are
h1 h2 h2 − h1
τ1 = , τ1 = , and τ3 = , (11.33)
t1 t2 t1
where h1 and h2 are the unknown heights of the two plates. (The constant value on the
outermost boundaries is, as always, conveniently taken as zero.)
The torque T given by twice the volume under the membrane including the horizontal
plates is

T = 2 × volume = 2(A1 h1 + A2 hh 2)
= 2(A1 τ1 + A2 τ2 ), using Eq. (11.33). (11.34)

The equation I
τ ds = 2 Gθ (11.35)

may be applied around (i) the first cell A1 , (ii) the second cell A2 , and (iii) the full two-cell
section (A1 + A2 ). When applied to the paths enclosing (A1 + A2 ) and A1 , we obtain the
following equations.

τ1 s1 + τ2 s2 = 2 G(A1 + A2 ) θ around A! + A2 (11.36)

τ2 s2 + τ3 s3 = 2 G A2 θ around A2 (11.37)
25
If the section is symmetrical as in Fig. 11.17b, both the plates will be at the same level. We can then
conclude that the middle leg (the middle wall, the wall that separates the two cells) carries no shear stress.
Advanced Mechanics of Solids 11-24

Subtracting the latter from the former, we obtain

τ2 s2 + τ3 s3 = 2 G A2 θ (11.38)

which is what we obtain when Eq. (11.35) is applied around the first cell A1 . Notice the
negative sign and the shear flow in the wall between the two cells A1 and A2 . Appealing
to the fluid flow analogy, we can see that the directions of the velocity in this wall (middle
leg) are opposite to each other. In other words,

τ3 = τ2 − τ1 . (11.39)

We may formulate this problem alternatively in terms of the two unknown heights h1 and h2 .
The net shear flow τ3 t3 is the difference between the shear flows τ3 t3 = τ2 t2 (downwards)
and τ3 t3 = τ1 t1 (upwards).
The above equations are sufficient to solve the problem. See the worked out examples
on thin-walled closed sections.
It is interesting to see what happens when the cross-section is symmetrical as shown
in Fig. (11.17b). Now because of symmetry, h1 = h2 (= h), and τ1 = τ2 (= τ ). The
consequence is that the middle leg (the wall between the two cells) is stress-free! This leg
is entirely useless to resist torsional moments. Still such a leg (wall) is provided, because
it serves other functions. The cross-section in general is subjected to loads other than a
torsional moment.
Closing Remarks
We have discussed some aspects of the theory of torsion of non-circular prismatic bars.
Actually it is better to discuss the general bending problem, and to treat torsion as a sub-
problem in the theory of bending. We know that when the load does not pass through
the shear centre of the cross-section, bending is accompanied by torsion. However, such
an approach is more difficult mathematically. There are many more aspects that we could
not discuss here; this is but natural. With these few remarks we come to the end of this
chapter.

In the next chapter, we shall see the field equations of the theory of elasticity.
Chapter 12

FIELD EQUATIONS OF THE

THEORY OF ELASTICITY

The field equations of elasticity are the basis of solution of problems, not only of mathe-
matical / analytical solution, but also of approximate and numerical methods like the finite
element method FEM. These play a crucial role in the formulation (and later solution also)
of all stress analysis problems. Thus, it is necessary to discuss these equations in fair detail.
We have now seen (i) the nature of the state of stress at a point, (ii) some details of
the state of strain and deformation at a point, and (iii) the constitutive equations (material
laws) for an elastic material. These are the fundamental prerequisites before the topic of
this chapter can be discussed.
FUNDAMENTAL EQUATIONS
We shall present in this chapter all the fundamental equations necessary for the formulation
of a general theory. Stress analysis problems are formulated and later solved, exactly or
approximately, using these equations. With these few words we shall begin the discussions
on the three pillars of the theory of elasticity.
The Three Pillars of the Theory of Elasticity
The fundamental equations are

(i) the equations of equilibrium;

(ii) the strain-displacement relations (also known as the kinematic relations); and
(iii) the constitutive equations (material laws).

These are the three pillars on which the edifice of the theory of elasticity and the science of
stress analysis is built. Thus, these three sets of equations are of great importance.
The first set depends only on the equilibrium of forces (associated with the stress com-
ponents and the applied external forces); it has nothing to do with the material properties.
Advanced Mechanics of Solids 12-2

The second set depends only on the geometry of deformation; it has nothing to do with
the material properties either. Thus, these two sets are equally applicable for other ma-
terials also. Inasmuch as these two sets are independent of the material properties, these
are applicable in the theories of plasticity1 , viscoelasticity, thermoelasticity, etc. Only the
third set involves the material properties (constitutive equations). It is only the third set of
equations that restricts the analysis to the particular material represented by the material
law (constitutive equation)2 .
Other Equations
In addition, there are other important equations: the compatibility equations; Navier’s
equations; Beltrami-Michell equations, etc. Some of the best brains have laboured and
done great work in these areas during the last three hundred years or more3 . There are
many, very many, equations that are derived and used. Only the most important ones can
be considered here.
The principle of virtual work applied to the case of a deformable body is also of great
importance. The variational methods of solid mechanics are based on this principle. Thus,
we shall discuss this also, though only briefly.
Some of these equations referred to above are differential equations. Thus, they must
be accompanied by the appropriate boundary conditions. Even though the boundary con-
ditions are not among the field equations, we shall still discuss them in this chapter.
Number of Unknowns and Equations Available
In a typical stress analysis problems, it is desired to determine (i) the stress components, (ii)
the strain components, and (iii) the displacement components at any point inside the body.
These are the unknowns: the six (6) stress components σij (i, j = 1, 2, 3); the six (6) strain
components eij (i, j = i, 2, 3); and the three (3) displacement components ui (i = 1, 2, 3)
making up a total of 6 + 6 + 3 = 15 unknowns. Here we have used the symmetry of both the
stress and the strain matrices. The available equations to determine them are: the three
1
Even though these remarks are valid, some changes may have to be made depending on the context. For
example, in the theory of elasticity, we can almost always assume that the strains are small. In the theory
of plasticity, however, the strains may not be small; we may have to account for large strains. Accordingly,
the actual strain-displacement equations used may have to be modified accordingly.
2
Let us emphasise that by a material we mean not quite the material, but the model we choose in a given
context. For example, concrete may be considered as an elastic material for many applications. However,
it can be considered as perfectly rigid in some situations where the deformations of the other components
are so large in comparison that the concrete foundation on which these are erected may be regarded as
rigid. On the other hand, when creep is to be calculated — long term creep effects are quite significant for
concrete — the concrete, the very same concrete, is to be regarded as viscoelastic. Similarly, the (clayey)
soil on which buildings are erected may be treated as elastic in some cases. But, if the tall buildings settle
down sinking more and more into the ground — the soil in Mexico city is said to be so notoriously yielding
that one or two storeys of buildings are said to sink down into the clayey soil — it may be necessary to
treat the soil as viscoelastic. Only viscoelasticity, and not elasticity, can account for such time dependent
displacements as in this example and creep.
3
Readers are advised to be familiar with the historical development of our subject. Timoshenko’s History
of the Strength of Materials is recommended.
12-3 Field Equations of the Theory of Elasticity

(3) differential equations of equilibrium σji,j + Fi = 0 (i, j = 1, 2, 3); the six (6) strain-
displacement relations eij = (1/2)(ui,j + uj,i ) (i, j = 1, 2, 3); and the six (6) stress-strain
relations (generalised Hooke’s law) σij = λ δij ekk + 2G eij (i, j, k = 1, 2, 3); making up a
total of 3 + 6 + 6 = 15 equations.
Looked at differently, we can regard that there are nine (9) stress components, in which
case there are six (6) equations of equilibrium (the three moment equations of equilibrium
also are available). Now there are 9 + 6 + 3 = 18 unknowns and 6 + 6 + 6 = 18 equations
available.
These are far too many equations to be solved. Thus, the general problem in the theory
of elasticity is a hopelessly difficult problem. Let us consider first set of equations, viz., the
(differential) equations of equilibrium.
DIFFERENTIAL EQUATIONS OF EQUILIBRIUM
This topic was taken up earlier and discussed in some detail in the earlier chapters. These
will not be repeated here. Generally these equations are linear, but it is possible to have
nonlinearities and other complications when complicated problems or more general formu-
lations are considered.
As we have remarked elsewhere, these equations alone are not sufficient to solve any
problem in stress analysis because there are more unknowns than there are equations. In
other words, all problems in the mechanics of solids and / or the theory of elasticity are
statically indeterminate internally. Hence we must have other equations to supplement.
These are the constitutive equations and the strain-displacement relations. We shall
take the latter for discussion below.
STRAIN-DISPLACEMENT RELATIONS
We shall now take up the next important set of equations, viz., the strain-displacement
relations. These are also called the kinematic relationships. Although we had seen these
earlier, let us derive these equations in a slightly different way.
Shear Strains and Rotations
Fig. 12.1 shows the details of the deformation in a neighbourhood of a point. The point
P moves to P ∗ . The displacement components are u, v, w in the x, y, z directions, respec-
tively. A neighbouring point E moves to E ∗ . Similarly, the point G moves to G∗ . The
displacements concerned are:

of P along the y axis: = v;

∂v
of E along the y axis: =v+ dy;
∂y
∂v
of G along the y axis: =v+ dz;
∂z
of P along the z axis: = w;
Advanced Mechanics of Solids 12-4

(a) Deformation in the neighbourhood of a point P (b) The shear strain and the rotation

Figure 12.1: The details of deformation in the neighbourhood of a typical point P are shown
in Fig. 12.1a. Fig. 12.1b helps us understand how the shear strain and the rotation are
separated. Some numerical values are taken to understand this. The values of 30 ◦ , etc. are
unrealistically high. The changes in the angles are very small indeed.
∂w
of E along the z axis: =w+dy;
∂y
∂w
of G along the z axis: = w + dz.
∂z
Fig. 12.1b shows the shaded wedge shaped element after deformation4 . We can notice that
the change in the angle is 30 + 10 = 40 ◦ , while the rotation of the element is (30 − 10)/2 =
10 ◦ . This example will probably help us understand that
   
1 1 ∂v ∂w 1 ∂v ∂w
exy = γyz = + ; ωyz = − .
2 2 ∂z ∂y 2 ∂z ∂y
From Fig. 12.1a, it is clear that the angles θ1 , θ2 and the shear strain, γyz and the rotation
ωyz are given by
∂w ∂v
θ1 = dy; θ2 = dz;
∂y ∂z
   
1 1 ∂v ∂w 1 ∂v ∂w
eyz ≡ γyz = + ; ωyz = − .
2 2 ∂z ∂y 2 ∂z ∂y
Let us also note that the linear strain along the y direction Fig. 12.1a is
 
∂v
v + ∂y dy − v ∂v
eyy = = .
dy ∂y
Writing the companion equations (from cyclic permutation and by noting the pattern), we
obtain all the strain displacement relations.
∂u ∂v ∂w
exx = ; eyy = ; ezz = ;
∂x ∂y ∂z
4
The angles are unrealistically high; in a real engineering structure, these angles can never be as high as
30 ◦ ; the strains, we know, are only of the order of 1 part in 1000.
12-5 Field Equations of the Theory of Elasticity

     
1 ∂u ∂v 1 ∂v ∂w 1 ∂w ∂x
exy = + ; eyz = + ; ezx = + ;
2 ∂y ∂x 2 ∂z ∂y 2 ∂x ∂z
ωxx = 0; ωyy = 0; ωzz = 0;
     
1 ∂u ∂v 1 ∂v ∂w 1 ∂w ∂x
ωxy = − ; ωyz = − ; ωzx = − .
2 ∂y ∂x 2 ∂z ∂y 2 ∂x ∂z
1 1
Using the index notation we can write as eij = (ui,j + uj,i ); ωij = (ui,j − uj,i ).
2 2

(a) An element before (b) No shear strain, (c) No rotation, only (d) Shear strain and ro-
deformation only rotation shearing strain tation

Figure 12.2: An undeformed element [Fig. 12.2a may undergo either a shearing alone, or a
rotation alone, or both a shearing strain and a rotation.
   
1 1
We can write ui,j = (ui,j + uj,i ) + (ui,j + uj,i ) = eij + ωij . Let us emphasise
2 2
that the strain matrix eij is symmetric (eij = eji ), while the rotation matrix ωij is skew-
symmetric (ωij = −ωji ). Such a decomposition into a strain tensor and a deformation
tensor can be understood physically with the help of Fig. 12.2. A solid element shown in
Fig. 12.2a may undergo a change in shape and / or a rotation. Fig. 12.2b shows the same
element after it rotates, but without any shear strain, while Fig. 12.2c shows the element
with only a shear strain, but no rotation. The general case is represented by Fig. 12.2d
involving both a shearing strain and a rotation.
These rotations are not related to the stress components in Solid Mechanics and, there-
fore, they have only a secondary role here. In Fluid Mechanics, on the other hand, they
play a crucial role.
The reason why we define exy as half the corresponding shear strain is this. Let us
compare the following two matrices.
 
exx γxy γxz
 γyx eyy γyz  (does not have the required transformation property.)
γzx γzy ezz
 
exx exy exz
eyx eyy eyz  (does have the required transformation property.)
ezx ezy ezz
The first one does not transform as a (second order) tensor; hence it is not acceptable, while
the second one does and is, therefore, acceptable. On comparison with the stress matrix,
Advanced Mechanics of Solids 12-6

we notice that the strain matrix is the matrix representation of the invariant, symmetric
strain tensor, and that the analogous quantities are
exx −→ σxx ; eyy −→ σyy ; ezz −→ σzz ;
γxy −→ 2 × τxy ; γyz −→ 2 × τyz ; γzx −→ 2 × τzx .
As a further vindication of this position, we recall the formulae for the principal stresses
and principal strains (for the simplified two-dimensional case).
s
σxx − σyy 2

σxx + σyy
σ11 , σ22 = ± + (τxy )2 ;
2 2
s
exx − eyy 2  γxy 2

exx + eyy
e11 , e22 = ± +
2 2 2
We can see that the formulae are entirely similar or analogous, if τxy is replaced by γxy /2!
Engineers’ Strain and True (or Natural) Strain
The simplest strain-displacement relation is the engineers’ definition of strain,
change in length lf − li ∆l
e= = = .
original length li li
This is fine as long as the strains are small. This is usually the case. The strain of, say,
steel at the yield point is of the order of 10−3 . This is quite small indeed (compared to 1).
However, when large strains are encountered as, say, in plastic flow there are difficulties.
This definition is not adequate. Let us take a simple, hypothetical example of a (uniform)
bar of initial length 100 mm when it is stretched to the final length of lf = 200 mm. By
the engineers’ definition given above, the strain is (200 − 100)/100 = 1.0 = 100 %. Let us
imagine that the elongation has taken place in two stages, say, first (i) from 100 mm to 150
mm, and next (ii) from 150 mm to 200 mm. What are the strains, in the first stage, second
stage, and in the final state? Well, in the first stage, e(1) = (150 − 100)/100 = 0.5 = 50 %,
while in the second stage, e(2) = (200 − 150)/150 = 0.333 = 33.3 %. When we add up,
e(1) + e(2) = 50 % + 33.3 % 6= 100 %.
What we have done is only imagination. We can also imagine, just as well and justifiably,
that the two stages are (i) from 100 mm to 120 mm, and in the next stage, e(2) = (200 −
120)/120 = 0.666 = 66.6 %. We again see not only that (a) they do not add up to 100 %,
but also that (ii) the sum is now different: 20 % + 66.6 % = 86.6 %. We can equally well
imagine that the total elongation has taken place in three stages. The answer will then be
again different.
There is no consistency. This lack of consistency is not only inconvenient, but is more
serious. What shall we do to clean up the resulting confusion?
We can imagine that the elongation has taken place not in one, two or three stages, but
in infinite stages. Then the total strain will be
Z lf    
dl lf lf − li ∆l
e= = log = log 1 + = log 1 + . (12.1)
li l li li li
12-7 Field Equations of the Theory of Elasticity

This is called the true strain or natural strain. We can see that at least the inconsistency is
now removed; everybody will arrive at the same answer! We can also see that this definition
(or measure or index) of strain is consistent with the engineers’ definition when the change
in length ∆l is very small5 compared to the initial length li .
This definition was popular at one time. Really, this is only one of several possibilities.
There is nothing particularly true about this definition, nor anything particularly false about
other measures of strain. For example, we can surely define engineers’ strain as (change
in length)/(final length), which agrees with the traditional definition whenever the change
in length is very small. [Recall the definitions of the Lagrangian and Eulerian measures of
strain.]
Compatibility Equations: Integrability
The strain-displacement relations, we have seen, are the following.

∂u ∂v ∂w
exx = ; eyy = : ezz = ; (12.2a)
∂x ∂y ∂z
     
1 ∂u ∂v 1 ∂v ∂w 1 ∂u ∂w
exy = + ; eyz = + ; ezx = + . (12.2b)
2 ∂y ∂x 2 ∂z ∂y 2 ∂z ∂x

The equations (12.2a) relate the normal strains exx , eyy and ezz to the (partial derivatives
of) the displacement components u, v and w, while the ones (12.2b) relate half6 the shearing
strains to the (partial derivatives7 of) the displacement components.
The six strain components have common parentage; they have all come from the three
displacement components (u, v, w). Thus, the six eij ’s are not independent; they are cousins
or half brothers. The eij ’s are, therefore, related to one another. These relationships among
the six strain components are the compatibility equations. Mathematically speaking, the
above six strain-displacement relations may be regarded as a system of six partial differential
equations for the determination of only three unknown functions (displacement components)
u, v and w. Thus, this is an overdetermined system, and consequently this system cannot
have any solution in general; certain conditions must be satisfied so that there can be
admissible (single-valued, continuous) functions u.v, w. These integrability condition are
the compatibility equations (or conditions).
It is clear from the six (6) strain-displacement relations [Eqs (12.2a), (12.2b)] that the
six strain components cannot be independent, because they have come from only three
three displacement functions u, v, w. This fact shows that the six strain components are
 
5 ∆l
We can see this by expanding log (1 + x) ≡ log 1+ in an infinite series.
li
6
The learned professor, Dr Bhoj Raj Seth (B.R. Seth), for long at IIT, Kharagpur, used to emphasise the
absolute necessity of introducing the factor of half in these equations. We should realise that without this
factor of half, the strain matrix will not have the transformation properties enjoyed (and required) by the
strain tensor. A comparison of the transformation equations of stress components and strain components
would give us this insight. Let me pause here to pay homage to this great teacher of ours with much
pleasure and gratitude. What a great inspiration even his mere presence was!
7
‘cross’ derivatives: u with y, v with x, and so on
Advanced Mechanics of Solids 12-8

related to one another. We shall now obtain these relations. These equations are called the
compatibility equations. There are six of them which are independent. They are of two
kinds, each of three equations.
Three equations of one (the first) kind:
These can be obtained as follows. Starting from the strain displacement relations
∂u ∂v ∂u ∂v
exx = ; eyy = ; γxy ≡ 2 exx = + ,
∂x ∂y ∂y ∂x
we obtain on performing the indicated differentiations,
∂ 2 exx ∂2u ∂ 2 eyy ∂2v ∂ 2 γxy ∂2u ∂2v
2
= 2 ; 2
= ; = + 2 ,
∂y ∂y ∂x ∂x ∂x2 ∂y ∂x ∂y ∂x ∂y 2 ∂x ∂y
which leads to the first of the three compatibility equations.
∂ 2 exx ∂ 2 eyy ∂ 2 γxy
+ = .
∂y 2 ∂x2 ∂x ∂y
The other two companion equations of this kind are obtained by cyclic change:
x → y; y → z; z → x.
Three equations of the other (second) kind:
The other three equations of the other (second) kind can be obtained as shown below.
Starting from the indicated derivatives
∂ 2 exx ∂3u ∂γyz ∂2v ∂ 2 w ∂γxz ∂2u ∂ 2 w ∂γxy ∂2u ∂2v
= ; = + ; = + ; = + ,
∂y ∂z ∂x ∂y ∂z ∂x ∂x ∂z ∂x ∂y ∂y ∂y ∂z ∂x ∂y ∂z ∂y ∂z ∂x ∂z
we note that
∂ 2 exx
 
∂ ∂γyz ∂γxz ∂γxy
2 = − + + . .
∂y ∂z ∂x ∂x ∂y ∂z
As before, the two companion equations are obtained by cyclic change:
x → y; y → z; z → x.
All these six (6) compatibility equations, which are all independent, are displayed below.
Equations of compatibility:
∂ 2 exx ∂ 2 eyy ∂ 2 γxy
+ = ; (12.3a)
∂y 2 ∂x2 ∂x ∂y
∂ 2 eyy ∂ 2 ezz ∂ 2 γyz
+ = ; (12.3b)
∂z 2 ∂y 2 ∂y ∂z
∂ 2 ezz ∂ 2 exx ∂ 2 γzx
+ = ; (12.3c)
∂x2 ∂z 2 ∂z ∂x
∂ 2 exx

∂ ∂γyz ∂γxz ∂γxy
2 = − + + ; (12.3d)
∂y ∂z ∂x ∂x ∂y ∂z
∂ 2 eyy
 
∂ ∂γzx ∂γyx ∂γyz
2 = − + + ; (12.3e)
∂z ∂x ∂y ∂y ∂z ∂x
∂ 2 ezz
 
∂ ∂γxy ∂γzy ∂γzx
2 = − + + . (12.3f)
∂x ∂y ∂z ∂z ∂x ∂y
12-9 Field Equations of the Theory of Elasticity

There is a physically more meaningful way of looking at the compatibility equations.

Let us imagine that the undeformed body is divided into a grid-work of rectangular blocks
(or brick elements). If arbitrary strains are imposed on these elements, they would go
out of shape as demanded by the strain field imposed. If the strain field is recklessly
specified, the deformed shapes will not close to form the deformed shapes with no tears
or cracks. The jig-saw puzzle cannot be solved. The deformed shapes cannot combine
to be a continuous (deformed) body any more. This makes it abundantly clear that the
strains imposed cannot be arbitrary; some conditions will have to be satisfied. These
conditions are the compatibility conditions. These, as stated earlier, may also be regarded
as the integrability conditions for the overdetermined system of six (6) partial differential
equations for the determination of the three (3) displacements.
In index notation these compatibility conditions are written as

eij,kl + ekl,ij − eik,jl − ejl,ik = 0, (i, j, k, l = 1, 2, 3). (12.4)

Although there are 3 × 3 × 3 × 3 = 81 equations here, only six (6) of them are independent.
[The explicit forms of the compatibility equations in a general curvilinear system are long
and complicated. They can be worked out explicitly by interpreting the comma (,) as
covariant differentiation using the methods of the general tensor analysis.]
We can verify the correctness of Eq. (12.4) by calculating the indicated derivatives. Let
us calculate the various derivatives which appear in Eq. (12.4). From eij = (1/2)[ui,j + uj,i ],
we obtain
1
eij,kl = [ui,jkl + uj,ikl ] ;
2
1
ekl,ij = [uk,lij + ul,kij ] ;
2
1
eik,jl = [ui,kjl + uk,ijl ] ;
2
1
ejl,ik = [uj,lik + ul,jik ] .
2
The order of differentiation is immaterial; i.e., uk,lij = ukijl , etc. Then we can see that

1
eij,kl + ekl,ij − eik,jl − ejl,ik = [ui,jkl + uj,ikl + uk,lij + ul,kij ]
2
1
− [ui,kjl + uk,ijl + uj,lik + ul,jik ] = 0.
2

The Role of Compatibility Equations

The role of the compatibility equations is merely to guarantee the existence of admissible
displacements. By admissible we mean continuous displacements that are compatible. If the
displacements are specified (that do not violate the obvious conditions of continuity, etc.),
the compatibility conditions have no further role to play. In other words, the compatibility
conditions will now be surely satisfied.
Advanced Mechanics of Solids 12-10

On the other hand, if we work with stresses, we have no direct information on whether
the associated displacements (corresponding to the strain components that correspond to
these stress components) will turn out to be compatible. Now the compatibility conditions
do step in to stop or prevent inadmissible stress distributions (that correspond to strain
distributions that correspond to impossible displacement fields).
Even at the risk of repetition, let us emphasise the following very important fact. Both
the equations of equilibrium and the compatibility of displacements must be simultaneously
satisfied. The equations of equilibrium are usually specified in terms of stress components,
while the compatibility refers to the displacements8 . For example, in solving the torsion
problem (of non-circular, prismatic bars), the displacements are assumed in the Saint-
Venant’s theory. Then there is no further role for the compatibility equations here. On the
other hand, we may decide to work in terms of Airy’s stress function. When we choose an
Airy’s stress function, effectively, we choose the stress field. The equations of equilibrium
are automatically satisfied, because that is how the Airy’s stress function is cleverly defined.
Now the compatibility condition does step in. This condition leads to the governing equation
which is the biharmonic equation.
In conclusion, the compatibility conditions have only a passive role — their presence
is not even felt or recognised at the operational level — when the displacement field is
assumed. On the other hand, they have an active role when a stress field is assumed9 .
STRESS-STRAIN RELATIONS
We shall take up this topic again to give more details using the index notation. The nine
stress components are related to the nine strain components by the equations

σij = Dijkl ekl (D: stiffness tensor), eij = Cijkl σkl (C: compliance tensor),

where D and C are tensors of order four (4), called the stiffness tensor and the compliance
tensor, respectively.. There are, thus, 3 × 3 × 3 × 3 = 81 elastic constants10 . We may
mention, in passing, that being tensors, both the stiffness tensor and the compliance tensor
— both these qualify to be called constitutive tensors — obey the transformation laws
0
Dijkl = aip ajq akr als Dpqrs ; and
8
There is, thus, a case for formulating the equations of equilibrium also in terms of displacements so that
we can work entirely with displacement fields. This is the motivation for considering the Navier’s (or
Lamé-Navier) equations of equilibrium in terms of the displacements. On the other hand, we can decide
to work entirely with the stress components, in which case the compatibility equations play an active role.
This is the motivation for developing the Beltrami-Michell equations of compatibility in terms of the stress
components.
9
Readers who have difficulty to follow these statements are advised to read a good book on the theory of
elasticity.
10
If we assume that the body is homogeneous, the elastic properties are the same; they do not change from
point to point. They are, thus, constants everywhere. More generally, they may be functions of position.
Assumptions of homogeneity and isotropy are often made together, sometimes tacitly, in some books
written in the strength of materials style. But we should not fail to see that these are quite different, and
that the consequences are also, naturally enough, different.
12-11 Field Equations of the Theory of Elasticity

0
Cijkl = aip ajq akr als Cpqrs .

Reduction from 81 to 36 Elastic Constants: Symmetry

The stress tensor, we know, is symmetric. Hence D and C are symmetric w.r.to the first
two indices of D, and w.r.to the last two indices of C. The strain tensor is symmetric too.
Hence D and C are symmetric w.r.to the last two indices of D, and w.r.to the first two
indices of C. The number of elastic constants thus becomes 36.

σij = σji −→ Dijkl = Djikl and eij = eji −→ Dijkl = Dijlk

Strain Energy Density Function: 36 to 21 Elastic Constants

A strain energy density function U = U(eij ) can be defined for a given state of stress and
strain such that the stress and strain components are related by the equation
∂U
σij = , (i, j = 1, 2, 3). (12.5)
∂eij
From U = U(eij ), we may also write
∂U
δU = δeij = σij δeij . (12.6)
∂ δeij
On comparing Eqs (12.5) and (12.6),

∂2U ∂2U
   
∂ ∂U ∂σkl ∂ ∂U ∂σij
= = = = = .
∂eij ∂ekl ∂eij ∂ δekl ∂eij ∂ekl ∂eij ∂ekl ∂ δeij ∂ekl
The equation
∂σkl ∂σij
=
∂eij ∂ekl
implies that the 6 × 6 constitutive matrix is symmetric. Putting, for example, i = 1, j =
2, k = 3, l = 1, we obtain
∂σ31 ∂σ12
= .
∂e12 ∂e31
Referring to Eq. (7.6) [p. 7-4], we can recognise the derivative on the left hand side to
be the elastic constant relating σ31 ≡ τzx and e12 ≡ exy which is D64 . The derivative on
the right hand side can similarly be seen to be the elastic constant relating σ12 ≡ τxy and
e31 ≡ ezx which is D46 . Thus, D64 = D46 . In this way, the number of elastic constants is
further reduced to 21. The general anisotropic elastic material has 21 elastic constants.
Isotropy
Further reduction for the most important case of isotropic elasticity can be achieved in the
following way. Firstly, let us note that the constitutive tensor, being of order four (4), will
transform as
0
Dijkl = aip ajq akr als Dpqrs . (12.7)
Advanced Mechanics of Solids 12-12

Secondly, let us realise that isotropic elasticity demands that the constitutive tensor is not
affected at all by any rotation of the reference coordinate axes. That is to say,
0
Dijkl = Dijkl . (12.8)
Eqs (12.7) and (12.8) lead to the requirement
Dpqrs = aip ajq akr als Dpqrs . (12.9)
Eq. (12.9) can be satisfied only if
Dpqrs = λ δpq δrs + µ δpr δqs + γ δps δqs , (12.10)
where λ, µ and γ are elastic constants. We can verify the correctness of this statement by
substituting Eq. (12.10) in Eq. (12.9).
[Here is another instance where the use of the index notation is not only convenient, but
also almost indispensable in discussions of the kind shown above. Thus, there is a strong
case for the young readers of this book to be comfortable with the index notation. The
effort taken will not be wasted.]
SOME COMMENTS
We note that every problem in the mechanics of solids (and in the theory of elasticity) is
statically indeterminate internally. That is to say, the equations of equilibrium alone will
not be able to give us the solution11 . Two requirements must always be uncompromisingly
met: (i) each and every part of the body must be in equilibrium12 ; (ii) furthermore, the
displacements (as a result of the deformations13 ) must be compatible, so that the deformed
body stays as a whole without tears or cracks, or without one part of the body intruding
into another.
The first condition (of equilibrium) is expressed naturally and conveniently in terms
of stress components. The second condition (of compatibility of displacements) is on the
displacements that correspond to the strain components (related by the strain-displacement
relations) that correspond to the stress components (related by the stress-strain relations).
One wonders if the solution of problems in the mechanics of solids would be less difficult
if both conditions are written either (a) all in terms of the stress components, or (b) all in
terms of the displacement components. Would this approach help?
11
We may be tempted to state that only the equation of equilibrium is used to obtain the (uniformly
distributed) stress σ = P/A in the case of a one-dimensional uniform bar subjected to end loads P −
P . Actually, among the various possibilities of stress distribution across the cross-section, all of which
satisfying the equations of equilibrium (such as, say, (i) triangular, (ii) parabolic, (iii) sinusoidal, (iv)
uniform stress on only part of the cross-section), only the uniform distribution is correct, because that is
the only case for which the resulting displacements are compatible. It is edifying to relate the two prime,
essential requirements, and see how these are met directly or indirectly by the various methods used. The
educated reader would probably like to link the force and displacement methods of structural analysis,
the strain energy and complementary energy methods, and the load bounding (upper and lower bound
theorems in the theory of plasticity) techniques to what is stated above.
12
It is not sufficient if the equilibrium condition is globally satisfied; that is, if the body as a whole is in
equilibrium. Every tiny bit of the body must be in equilibrium.
13
or when a displacement field is prescribed in advance
12-13 Field Equations of the Theory of Elasticity

DISPLACEMENT FORMULATION
In the displacement formulation, the unknown field variables are the displacements. Thus,
in this approach, we have to recast the differential equations of equilibrium in terms of
displacements. The broad general procedure is as follows.

(i) As the first step, let us replace the stress components in the equations of equilibrium
by the strain components (using the constitutive equations).
(ii) In the second step, we shall replace the strain components by the displacement com-
ponents (using the strain-displacement relations).

Thus we shall obtain the differential equations of equilibrium in terms of the displacements,
known as Navier’s (or Lamé-Navier) equations. We shall now carry out this procedure.
Navier’s (Lamé-Navier) Equations
As outlined above, let us first substitute the stress-strain relations (constitutive equations,
generalised Hooke’s law) σij = λ δij ekk + 2µ eij into the differential equations of equilibrium
σji,j + Xi = 0. This gives us
(λ δij ekk + 2µ eij ),j + Xi = 0,
which is rewritten as
λ δij ekk,j + 2µ eij,j + Xi = 0.
That is, using the substitution property δij ekk,j = ekk,i ,
λ ekk,i + 2µ eij,j + Xi = 0, (12.11)
which are the differential equations of equilibrium in terms of the strain components.
Now the strain components are all replaced by the displacement components using the
1
strain-displacement relations eij = (ui,j + uj,i ) to give us the desired equation in the
2
desired form after a few simplifications shown below.
1 1
λ (uk,k + uk,k ),i + 2µ (ui,j + uj,i ),j + Xi = 0; that is,
2 2
λ uk,ki + µ (ui,jj + uj,ij ) + Xi = 0.
Changing the dummy index k in the first term to j, this equation appears as
λ uj,ji + µ ui,jj + µ uj,ij + Xi = 0,
and as uj,ji = uj,ij , the order of differentiation being immaterial, the well known Navier’s
equation is obtained in the form
(λ + µ) uj,ij + µ ui,jj + Xi = 0. (12.12)
This, in content, is the set of 3 (the free index i = 1, 2, 3) equations of equilibrium, but in
terms of the displacements ui which are functions of position.
Advanced Mechanics of Solids 12-14

Equations of Equilibrium in Longhand Notation

We shall obtain the equations of equilibrium in terms of the displacement components
u, v, w. We shall now carry out this exercise below.
Consider the differential equations of equilibrium and the associated boundary condi-
tions in terms of stresses (Cauchy’s result).

∂σxx ∂τyx ∂τzx (ν)

+ + + Fx = 0; Tx = lσxx + mτyx + nτzx ;
∂x ∂y ∂z
∂τxy ∂σyy ∂τzy (ν)
+ + + Fy = 0; Ty = lτxy + mσyy + nτzy ;
∂x ∂y ∂z
∂τxz ∂τyz ∂σzz (ν)
+ + + Fz = 0; Ty = lτxz + mτyz + nσzz .
∂x ∂y ∂z

Let us substitute the expressions (from the constitutive relations) for the stress components
σxx , σyy , σzz

σxx = λe + 2Gexx ; σyy = λe + 2Geyy ; σzz = λe + 2Gezz (e = exx + eyy + ezz )

   
∂u ∂v ∂w ∂u
τxy = τyx = Gγxy = G + ; τxz = τzx = Gγxz = G +
∂x ∂y ∂x ∂z
Substituting these expressions for the stresses, the first equation of equilibrium becomes
 2
∂ u ∂2u ∂2u

∂e
(λ + G) +G + 2 + 2 + Fx = 0.
∂x ∂x2 ∂y ∂z

Navier’s (Lamé-Navier) equations:

The two companion equations also may be arrived at similarly. Thus, the three Navier’s
(or Lamé-Navier) equations — the equations of equilibrium in terms of the displacement
components — are
 2
∂ u ∂2u ∂2u

∂e
(λ + G) +G + 2 + 2 + Fx = 0; (12.14a)
∂x ∂x2 ∂y ∂z
 2
∂2v ∂2v

∂e ∂ v
(λ + G) +G + + + Fy = 0; (12.14b)
∂y ∂x2 ∂y 2 ∂z 2
 2
∂ w ∂2w ∂2w

∂e
(λ + G) +G + + + Fz = 0. (12.14c)
∂z ∂x2 ∂y 2 ∂z 2

Cauchy’s result:
Cauchy’s result usually written in terms of the stress components now appears in terms of
the displacement components as

(ν)
   
∂u ∂u ∂u ∂u ∂v ∂w
Tx = λe l + G l +m +n +G l +m +n ; (12.15a)
∂x ∂y ∂z ∂x ∂x ∂x
12-15 Field Equations of the Theory of Elasticity

(ν)
   
∂v ∂v ∂w ∂u ∂v ∂w
Ty = λe m + G l +m +n +G l +m +n ; (12.15b)
∂x ∂y ∂z ∂y ∂y ∂y
(ν)
   
∂w ∂w ∂w ∂u ∂v ∂w
Tz = λe n + G l +m +n +G l +m +n . (12.15c)
∂x ∂y ∂z ∂z ∂z ∂z
Special simplified case, no body force:
The Laplacian in the above equations may be written as 52 . If the body forces are absent,
these equations are simplified as
∂e
(λ + G) + G 52 u = 0; (12.16a)
∂x
∂e
(λ + G) + G 52 v = 0; (12.16b)
∂y
∂e
(λ + G) + G 52 w= 0. (12.16c)
∂z
An important relationship:
We may also arrive at an important relationship. The volume dilatation (volume expansion)
e satisfies the (3-dimensional) Laplace’s equation

∂2e ∂2e ∂2e

(λ + 2G) 5 e = 0 −→ 5 e ≡ + + = 0. (12.17)
∂x2 ∂y 2 ∂z 2
We can see this by differentiating Eq. (12.16a) w.r.to x, Eq. (12.16b) w.r.to y, Eq. (12.16c)
w.r.to z, and adding them together. The result is valid not only if the body forces are
absent, but also when they are constant throughout the volume of the body.
BELTRAMI-MICHELL EQUATIONS
We shall now see the Beltrami-Michell equations which are the compatibility equations
expressed in terms of the stress components, convenient in an all-stress formulation.
Introduction
We had stated that, among others, there are two possibilities: (i) recast all the governing
equations so that all the unknowns are in terms of (i) the displacements u, v, w, and (ii)
the stress components. We have seen just now the first possibility (i). We shall presently
examine the second one (ii).
The equations of equilibrium are written in terms of the stress components. There
is nothing more to be done about them; they are already in good shape. However, the
compatibility conditions, which are in terms of the strain components, are to be processed
and to be recast in terms of the stress components. The Beltrami-Michell equations, the
topic here, are the compatibility equations in terms of the stress components.
Compatibility Equations in Terms of the Stress Components
Our objective is to obtain the compatibility equations in terms of the stress components.
We can accomplish this task in the following way.
Advanced Mechanics of Solids 12-16

The compatibility equations are usually written in terms of the strain components.
However, it is possible to recast them in terms of stresses. We shall take up one compatibility
equation, viz.,
∂ 2 eyy ∂ 2 ezz ∂ 2 γyz
+ = , (12.18)
∂z 2 ∂y 2 ∂y ∂z
and show how it is processed. In this equation, we replace the strain components by the
stress ones (stress components) using the constitutive relations

1
eyy = [(1 + ν)σyy − νΘ] ; Θ = σxx + σyy + σzz ;
E
1
ezz = [(1 + ν)σzz − νΘ] ;
E
2(1 + ν)τyz
γyz = ,
E
where Θ ≡ I1 , is the first stress invariant. On substitution, the above compatibility equation
appears as
 2
∂ 2 σzz ∂ 2 τyz
 2
∂ Θ ∂2Θ
 
∂ σyy
(1 + ν) + − ν + = 2(1 + ν) . (12.19)
∂z 2 ∂y 2 ∂z 2 ∂y 2 ∂y ∂z

We have achieved what we set out to do. However, we proceed further and present this in
the more familiar convenient form. Towards this end, let us process the right hand side of
this equation using the equations of equilibrium.

∂σzz ∂τyz ∂τxz ∂τyz ∂σzz ∂τxz

+ + + Fz = 0 −→ =− − − Fz ;
∂z ∂y ∂x ∂y ∂z ∂x
∂σyy ∂τxy ∂τzy ∂τyz ∂σyy ∂τxy
+ + + Fy = 0 −→ =− − − Fy .
∂y ∂x ∂z ∂z ∂y ∂x

Differentiating the first w.r.to z, and the second w.r.to y, and adding them, we obtain

∂ 2 τyz ∂ 2 σzz ∂ 2 σyy

 
∂ ∂τxz ∂τxy ∂Fz ∂Fy
2 =− 2
− 2
− + − −
∂y ∂z ∂z ∂y ∂x ∂z ∂y ∂z ∂y
2 2 2
∂ σxx ∂ σyy ∂ σzz ∂Fx ∂Fy ∂Fz
= 2
− 2
− 2
+ − − .
∂x ∂y ∂z ∂x ∂y ∂z

In the last step we have used the first equation of equilibrium (in the x-direction). Now on
substituting this expression in the right hand side of Eq. (12.19), we obtain

∂2Θ ∂2Θ
     
2 2 2 ∂Fx ∂Fy ∂Fz
(1 + ν) 5 Θ − 5 σxx − −ν 5 Θ− = (1 + ν) − − (12.20)
∂x2 ∂x2 ∂x ∂y ∂z

To make the equation short, we have used the notation

∂ ∂ ∂
52 ≡ 2
+ 2 + 2.
∂x ∂y ∂z
12-17 Field Equations of the Theory of Elasticity

Two more equations of the same kind, the companion equations, can be obtained on similar
lines. If we add up these three equations of the kind of Eq. (12.20), we obtain
 
2 ∂Fx ∂Fy ∂Fz
(1 − ν) 5 Θ = −(1 + ν) + + . (12.21)
∂x ∂y ∂z
These correspond to the three compatibility conditions of the first kind. This last equation
(12.21) gives the expression for 52 Θ. If we substitute this expression for 52 Θ in Eq.
(12.20), we obtain

1 ∂2Θ
 
2 ν ∂Fx ∂Fy ∂Fz ∂Fx
5 σxx + =− + + −2 (12.22)
(1 + ν) ∂x2 (1 − ν) ∂x ∂y ∂z ∂x
There are, of course, the two companion equations. These three correspond to the three
compatibility conditions of the first kind. The other three compatibility conditions (of the
second kind) can also be processed similarly, and recast in terms of the stress components.
We obtain
∂2Θ
 
2 1 ∂Fz ∂Fy
5 τyz + =− + , (12.23)
(1 + ν) ∂y ∂z ∂y ∂z
and its two companion equations. If the body forces are absent, or if they are constant
throughout the volume of the body, there is some simplification. The six equations of
compatibility, now expressed in terms of the stress components appear as follows.

∂2Θ
(1 + ν) 52 σxx + = 0; (12.24a)
∂x2
∂2Θ
(1 + ν) 52 σyy + = 0; (12.24b)
∂y 2
∂2Θ
(1 + ν) 52 σzz + = 0; (12.24c)
∂z 2
∂2Θ
(1 + ν) 52 τyz + = 0; (12.24d)
∂y ∂z
∂2Θ
(1 + ν) 52 τxz + = 0; (12.24e)
∂x ∂z
∂2Θ
(1 + ν) 52 τxy + = 0. (12.24f)
∂x ∂y

Comments
In this all-stress formulation, the governing equations are (i) the equations of equilibrium
and (ii) these six (6) compatibility equations in terms of stresses. The (traction) boundary
conditions (Cauchy’s results) must be satisfied. With these the stress components in a
linearly elastic, isotropic body can be determined. These equations are generally, but not
always14 , sufficient.
14
These exceptional cases where they are not sufficient cannot be discussed here; the discussion will be at a
much higher level, too high for this book.
Advanced Mechanics of Solids 12-18

Let us note further these compatibility conditions contain no higher derivatives of the
stress components than the second. What this means is that the stress field (satisfying
the equations of equilibrium and the boundary conditions) gives the correct solution if the
stress components are either constants or linear functions of the coordinates.
BOUNDARY CONDITIONS
So far we have seen the field equations. These are the basis of the formulation of a stress
analysis problem. However they are not all; they are only one part of the story. These
equations are to be supplemented by the relevant boundary conditions which are almost as
important. We shall discuss the boundary conditions here.
A comprehensive coverage of this topic may not be possible. We will be able to discuss
only simple cases at an elementary level. First we shall see one-dimensional problems like
a beam.
One-dimensional Problems: Beams

(a) A simply supported beam (b) A cantilever

Figure 12.3: The beam in Fig. 12.3a is simply supported at both ends, while that Fig.
12.3b, a cantilever, is fixed at the left end and free at the other. The deflection at a (i)
supported end and (ii) a fixed end shall be zero. The slope at a fixed end is zero.

(a) A beam fixed at both ends (b) An element of a beam

Figure 12.4: A fixed-fixed (fixed at both ends) beam is shown in Fig. 12.4a. The deflection
and slope at each fixed end is zero. Fig. 12.4b shows a beam element with the load, the
shear forces, and the bending moments marked on it in their respective positive directions.

Let us consider a beam loaded by a transverse load. Three cases are shown in Figs 12.3a
(a simply supported beam), 12.3b (a cantilever), and 12.4a (a fixed-fixed beam). They are
12-19 Field Equations of the Theory of Elasticity

all governed by the same linear differential equation15 .

d2 y
EI = −M, (0 < x < l). (12.25)
dx2
The two boundary conditions of this second order differential equation are:

simply supported: [Fig. 12.3a] y(0) = 0; y(l) = 0; (12.26a)

0
cantilever: [Fig. 12.3b] y(0) = 0; y (0) = 0; (12.26b)
0 0
fixed-fixed: [Fig. 12.4a] y(a) = 0; y(b) = 0; y (0) = 0; y (l) = 0.. (12.26c)

The physical meaning of these boundary conditions are:

at a support (which is assumed to be rigid, non-yielding) the deflection y = 0; and
at a fixed end, the deflection and the slope are both zero: y(0) = 0; y 0 (0) = 0.
We know from our earlier study of the mechanics of solids that

dF dM d2 y
= −w; = F; EI = −M, (12.27)
dx dx dx2
and that the bending of the beam is governed by the equation

d2 y d2 y
 
EI 2 = w, (12.28)
dx2 dx

where w, F, M are, respectively, the rate of loading, the shear force, and the bending
moment. The differential equation (still linear, but could be a variable coefficient one) is
now of order four (4). It can now take four (4) boundary conditions. What are these
four boundary conditions? We shall discuss these with reference to a beam on an elastic
foundation [Fig. 12.5].
Fig. 12.5 shows a beam on an elastic
foundation. The left end A (x = 0)
is fixed. Thus, the deflection and the
slope there are both zero. Thus, these
are two of the (four) boundary condi-
tions, viz.,

y(0) = 0; y 0 (0) = 0.

The other two conditions refer to the

bending moment (M ) and the shearing
Figure 12.5: A beam on an elastic foundation force (F ) at the free end B (x = l).

15
The right hand side may be written as M or as −M , depending on the sign convention used. Here we
choose −M . This is consistent with the marking in Fig. 12.4b. The deflection is positive downwards.
Advanced Mechanics of Solids 12-20

The differential equation governing bending is

d2 y d2 y
 
EI 2 + ky = w(x) (0 < x < l), (12.29)
dx2 dx
where k is the ‘spring constant’ (‘Winkler constant’ of the ‘Winkler foundation’).
Now we shall consider the four boundary conditions. These boundary conditions at the
end B (x = l) are:
d2 y 
 
EI 2  = −M applied B.M. at B (12.30)
dx x=l
d2 y 
 
d
EI 2  = −P applied concentrated force at B (12.31)
dx dx x=l

These four (4) boundary conditions can, thus, be written down easily using the physical
facts of the problem. We used only elementary facts to obtain these conditions.
There is a more insightful way of obtaining these boundary conditions and of looking at
them. We can see that they are of two different kinds: (i) on the displacement (deflection)
y and its derivative (slope) y 0 , and (ii) on the bending moment (M ) and its derivative
(shear force) (F ). These two kinds may be regarded as the primary and the secondary
variables. To have this more enlightened view, we need to have a fair understanding of the
calculus of variations where the two kinds of boundary conditions are referred to as (i) the
prescribed and (ii) the natural boundary conditions. [There is nothing natural about these,
nor anything artificial about those of the other kind; they are both technical terms. See [9]
and [10], or better still, the series of books by the learned Professor J.N. Reddy.]
A Propped Cantilever: Rigid and Yielding Props

(a) A propped cantilever (b) Support removed and (c) A concentrated

loaded by a u.d.l. replaced by a reaction R load R at the end

Figure 12.6: A propped cantilever AB with a uniformly distributed load, w per unit length
propped by a rigid or yielding support. We use the principle of consistent deformations
(compatibility of deformations) to calculate the support reaction, R.

It is not possible, or permissible, to specify both the applied force and the correspond-
ing (work absorbing) displacement at the same point. In the language of the calculus of
variations, both the prescribed and the corresponding natural boundary conditions cannot
be specified at the same point. On the physical side this may be understood in the following
way.
We have before us, let us say, a horizontal beam loaded by its self-weight. At a certain
point, say at its midpoint, we can apply an upward load of, say, 1 kN from below to prop
12-21 Field Equations of the Theory of Elasticity

it up. This may result in a reduced deflection. Instead of applying the specified 1 kN
load, we may lift the beam at the same point by, say, 2 mm. This may require a force
of, say, 0.5 kN. We cannot, can we, specify both the force and the displacement at the
same point. If a load of 1 kN is applied, the stiffness of the structure decides how much
the resulting displacement is. We cannot specify or realise both the applied load and the
resultant displacement separately and independently. This is very clear from the physics of
the problem; every engineer can understand it easily. But the interesting part of this is that
these facts will be obtained automatically (by applied mathematical techniques) without
appealing to the physical understanding of the problem. Another example of the beauty
of applied mathematics in action! [An engineer can write down the boundary conditions
correctly from his physical understanding of a problem. An applied mathematician also
will be able to arrive at these conditions correctly. We who are bilingual — we speak
both languages, the languages of both engineers and applied mathematicians — can not
only understand both, but can really enjoy and appreciate how applied mathematics and
engineering science support each other.]
The case of a propped cantilever is taken up in this sub-section. Fig. 12.6a shows a
cantilever AB of length l, fixed at the end A and loaded by a uniformly distributed load w
per unit length. It is propped at the end B. There are two cases to consider: (a) a rigid
(unyielding) support, and (b) a yielding support which is the same as a spring of stiffness
(spring constant) k. We need to discuss the boundary conditions at the ends.
Actually, this is not a question of writing the boundary conditions. Both cases are
statically indeterminate. We need to make use of the principle of consistent deformations
(compatibility of displacements) to solve the problem. Let us remove the support and
introduce a concentrated force, which is the support reaction R.
Case (a) Rigid support:
Remove the force and calculate the end deflection due to the uniformly distributed load w
per unit length. This is w l4 /(8 EI) downwards. Now the force R is such as to produce an
end deflection w l4 /(8 EI) upwards, so that the total deflection is zero, as demanded by the
condition of an unyielding support. Recalling that the end deflection of a cantilever due to
an end load R is R l3 /(3 EI), we obtain
w l4 R l3 3
= −→ R = wl.
8 EI 3 EI 8
This is the principle of consistent deformation. The problem is, thus, solved.
Case (b) Yielding support (spring support):
Let the end deflection be δ. This is now unknown. It is equal to R/k, where k is the
spring constant. This deflection δ is the difference between the downward deflection due to
the u.d.l., and the upward deflection because of the unknown support force R. Thus, the
principle of consistent deformation is
w l4 R l3 R
− =δ= ,
8 EI 3 EI k
from which the support force R can be readily calculated. This completes the solution of
the problem stated.
Advanced Mechanics of Solids 12-22

Our last example in this set refers to the axial deformation of a one-dimensional bar.
Axial Deformation of a One-dimensional Bar

Figure 12.7: A one-dimensional bar with a spring S (stiffness k) at the left end L (x = 0),
and fixed at the right end R (x = l). The area of cross-section A (or more generally, EA)
is variable. An external axial load F = F (x) acts on the bar. An example of such an axial
load might be the self-weight if the bar is vertical with its support at R (x = l).

We had discussed this problem earlier [p. 5-10]. We desire to formulate this problem in
terms of the axial displacement u(x), and discuss the boundary conditions of the problem.
The differential equation of equilibrium (5.15), we recall, was obtained as
d
(σ A) + F = 0, (0 < x < l). (12.32)
dx
We desire to recast this differential equation of equilibrium16 in terms of the displacement
u. We know that
du
stress, σ = E × strain = E .
dx
If the stress σ in the above equation (12.32) is replaced by this expression in terms of the
strain, we obtain the governing differential equation in the desired form as
 
d du
AE + F = 0, (0 < x < l). (12.33)
dx dx

 
du
Natural boundary condition: AE − ku(0) = 0 at x = l (12.34)
dx
Prescribed boundary condition: u(l) = 0. (12.35)

The prescribed boundary condition (12.35) states that the right end of the bar is fixed and,
consequently, there can be no (axial) displacement there; that is, u(l) = 0. The natural
boundary condition (12.34) states that, at the left end of the bar, the force on the bar is
the spring force.
du du
spring force: = ku(0); axial force: = AE × strain = AE ; AE = ku(0).
dx dx
16
Recall that we had called attention to such cases where the differential equation(s) is / are recast in terms
of the displacement(s). This is one such case.
12-23 Field Equations of the Theory of Elasticity

As indicated elsewhere these problems can be formulated using variational methods. The
prescribed boundary conditions are easy to write, but the natural boundary conditions,
which can be hard to write sometimes, emerge beautifully and effortlessly from a variational
formulation.
Next we shall see the boundary conditions in two-dimensional problems.
Two-dimensional Problems

(a) Steady-state heat transfer in a plate (b) A plate ABCD subjected to stresses

Figure 12.8: A rectangular plate ABCD is shown. In the first case [Fig. 12.8a] a steady-
state heat transfer is considered, while in the second [Fig. 12.8b] the rectangular plate is
subjected to tensile stresses as shown. The biharmonic equation is to be solved in the region
shown.

Two-dimensional boundary values problems are associated with two-dimensional re-

gions. There are several cases to be considered here, but we do not propose to be compre-
hensive. We shall consider only two examples of how the boundary conditions are written
for such problems.
Biharmonic Equation: Two-dimensional Stress Analysis
We know from our earlier study of two-dimensional elasticity problems that the Airy’s stress
function is sometimes useful. The problem that we take up is to determine the stresses in
a rectangular plate ABCD subjected to a given stress distribution on the edges. [Fig.
12.8b] gives the details as a stress analysis problem, while [Fig. 12.8a] is to explain the
corresponding boundary value problem in terms of the biharmonic function φ = φ(x, y) in
R when the boundary conditions are specified on C.
We also know that the Airy’s stress function φ = φ(x, y) satisfies the biharmonic equa-
tion, and that the stress components are related to φ by

∂2φ ∂2φ ∂2φ ∂2φ

σxx = ; σyy = ; τxy = τyx = = ;
∂y 2 ∂x2 ∂x ∂y ∂y ∂x
∂4φ ∂4φ ∂4φ
52 (52 φ) = 54 φ ≡ + 2 + = 0.
∂x4 ∂x2 ∂y 2 ∂y 4
Advanced Mechanics of Solids 12-24

The stress analysis problem is thus converted into, or reformulated as, a boundary value
problem for the determination of φ = φ(x, y) in the region R.
∂4φ ∂4φ ∂4φ
differential equation: + 2 + = 0 in R :
∂x4 ∂x2 ∂y 2 ∂y 4
boundary conditions: edge AB φxx = 0 (σyy = 0);
edge AB φxy = 0 (τyx = 0);
edge BC φxy = 0
(τxy = 0);
y2 y2
   
edge BC φyy = S 1 − 2 (σxx = S 1 − 2 );
b b
edge CD φxy = 0 (τxy = 0);
edge CD φxx = 0 (σyy = 0);
edge DA φxy = 0 (τxy = 0);
y2 y2
   
edge DA φyy =S 1− 2 (σxx = S 1 − 2 );
b b
We have written all the boundary conditions. This exercise is over at this stage17 .
Traction and Displacement Boundary Conditions
Stress analysis problems are generally formulated as boundary value problems. Thus, the
boundary conditions expressed in terms of the unknown dependent variable are to be spec-
ified. In several, perhaps most, cases either the traction or the displacement is specified on
the boundary. Sometimes, traction is specified on part of the boundary, and displacement(s)
on the remaining part. It is not possible, or permissible, to specify both the traction and
the (corresponding work absorbing) displacement at the same point.
If, for example, part of the boundary, say C1 is fixed, then the displacement boundary
condition there is u = v = w = 0 on C1 = 0. If the traction is specified on another part, say
C2 , the traction boundary condition there is written using Cauchy’s result as
(ν)
Tx specified = lσxx + mτyx + nτzx ;
(ν)
Ty specified = lσxy + mσyy + nτzx ;
(ν)
Tz specified = lτxz + mτ yz + nσzz .

A Sector of a Circle: Polar Coordinates

Next we shall consider a sector of a circle [Fig. 12.9a]. We shall use convenient polar
coordinates (r, θ). The edge OB is fixed. The loadings on the straight edge OA and the
curved edge AB are shown. The boundary conditions are written as follows.
17
The boundary conditions are not all homogeneous; this is a little inconvenient in some cases as, for
example, in seeking approximate solutions. We can make all the boundary conditions homogeneous by
defining some quantities differently. We do not discuss them here as they are unnecessarily distracting.
12-25 Field Equations of the Theory of Elasticity

(a) A sector OAB of a circle (b) An elastic-plastic interface

Figure 12.9: A circular sector OAB is shown [Fig. 12.9a]. The edge OB is fixed. The
loadings on the other edges are shown. In the second case [Fig. 12.9b] the circular cylinder
is partly, but not fully, in the plastic regime. There is an inner elastic core surrounded by
a plastic region with an elastic-plastic interface.

i) On the edge OB: The edge OB is fixed. Hence the displacements u and v are both
zero at all points on the edge.
  π
u(r, θ) π ≡ u r, = 0 for all r (0 ≤ r ≤ R);

θ= 4 4
  π
v(r, θ) π ≡ v r, = 0 for all r (0 ≤ r ≤ R).

θ= 4 4

ii) On the edge OA: The normal stress σθθ is linearly distributed as shown. The shear
stress τrθ = b, a constant.
 r
σθθ (r, θ) ≡ σθθ (r, 0) = a for all r (0 ≤ r ≤ R);

θ=0 R
τrθ (r, θ) ≡ τrθ (r, 0) = b for all r (0 ≤ r ≤ R).

θ=0

iii) On the edge AB: The normal stress σrr = c, a constant. The shear stress τrθ = b, a
constant.
 π
σrr (r, θ) ≡ σrr (R, θ) = c for all θ (0 ≤ θ ≤ );

r=R 4
π
τrθ (r, θ) ≡ τrθ (R, θ) = b for all θ (0 ≤ θ ≤ ).

r=R 4

All the boundary conditions have now been written; this exercise is over here.
Special Cases and Concluding Remarks
There can be special cases not covered so far. One is when a concentrated load, say, P is
applied at a point (x = a). For this case, the rate of loading w can be written as
Z +∞
w = P δ(a), δ = 0 for x 6= a, δ(x) dx = 1,
−∞
Advanced Mechanics of Solids 12-26

where δ(a) is the Dirac delta ‘function’ (also called an impulse function). [This is not a
function at all in the traditional sense. Mathematicians continued to object, while physicists
continued to use it. How can a mathematicians accept the validity of this integral?
Z +∞
δ(x) dx = 1, δ = 0 for x 6= a.
−∞

Now the use of the Dirac delta function is mathematically legitimatised by the relatively
recent theory of distributions.]
Here is a second case [Fig. 12.9b] of an elastic-plastic interface. A circular, prismatic
shaft is subjected to a torque T . When T is increased, the maximum shear stress on the
cross-section at the outermost fibre reaches its elastic limit. If T is increased, plasticity will
creep in; on further increase plasticity creeps further inside (with an outer plastic region
surrounding an inner elastic core). For such a boundary value problem with an (unknown)
interface, we have two sub-regions: (i) an outer ring where the material has already yielded
and, consequently, plasticity conditions prevail, and (ii) an inner elastic core. Now what
are the boundary conditions at the interface?
There can be other complications when we have an inclusion surrounded by an outer
matrix with different properties. Yet another case is the problem of contact stresses (referred
to as Hertzian stresses). We cannot obviously discuss all these cases here. At this stage we
close this chapter.

In the next chapter we shall discuss some illustrative examples.

Chapter 13

A FEW ILLUSTRATIVE
EXAMPLES

We shall work out several illustrative examples1 . In some cases, only the conceptual part
is discussed, and long laborious calculations are partially, or even entirely, left out. Use of
matlab, maple and / or mathematica is encouraged.
A FEW WORKED OUT EXAMPLES
A few examples — not all numerical ones — are worked out below for illustration. These
are somewhat jumbled, not always in a logical order. It may help if the full procedure is
explained and the steps shown before the detailed calculations are taken up.

1. Example: Deflection of a Cantilever

Calculate the deflection at A when a concentrated force P acts at B on a cantilever.
We know from the Betti-Maxwell reciprocal theorem2 that this is equal to the deflection
at B when the force P acts at A [Fig. 13.1b]. This deflection has two components δ1 and
1
Students, when writing the all important(!) ‘university examinations’ seem to be obsessed with the idea
that numerical correctness of the final answer is all that matters. They have a good reason to think so.
Over the years examiners often appear to attach too much importance for the correctness of the final
answers, and too little for the correctness of the intermediate steps, and nothing at all for the conceptual
understanding of the fundamentals. Furthermore, these questions seem to be a speed test. Actually, speed
within reasonable limits should not be a factor at all. All this has created a warped thinking on the part
of most students. Students are advised to pause, think, and write down the various steps before they
begin the detailed calculations. It is natural for everybody, particularly the young students, to be nervous
in the examination hall. They often spend too much time struggling with the detailed calculation of a
relatively unimportant first step, and end up losing time. Students are advised to write down first the
overall procedure before undertaking the detailed calculations. In this way they can guard against the
pressure created by lack of time. Some of the problems take too long to work out fully. We hope that the
examiners will be kind and considerate, and remember the days when they were students.
2
The deflection at A due to a force acting at B is equal to the deflection at B due to the same force acting
at A. The word force is used in a general sense; it can include moments also just as in the statement Man
is mortal, the word man includes woman also! If the force is a moment, the corresponding work absorbing
displacement is a rotation.
Advanced Mechanics of Solids 13-2

(a) A cantilever loaded at B (b) The same cantilever loaded at A

Figure 13.1: The deflection at A of a cantilever loaded at B [Fig. 13.1] is obtained, using
the Betti-Maxwell reciprocal theorem, by calculating the deflection at B when the load acts
at A [Fig. 13.1b].

δ2 . The second component arises because of the slope at A. The deflected cantilever has
no curvature in the region AB — it is a straight line set at an angle equal to the slope
at A — because the bending moment in the region between A and B is zero3 . Thus, the
P a3 P a2
required deflection is δ = δ1 + δ2 = + b. With a little experience these steps
3 EI 2 EI
are gone through in the mind almost effortlessly, and the final expression can be written
in just about the same time as it takes to write your name!

2. Example: Hydrostatic and Pure Shear States

Let us split the following stress matrix into two matrices representing (a) a hydrostatic
state of stress, and (b) a state of pure shear. All the values are in MPa (N/mm2 ).
 
10 12 8
12 15 −5
8 −5 −7

Let us decompose this as

     
10 12 8 p 0 0 10 − p 12 8
12 15 −5 = 0 p 0 +  12 +15 − p −5  .
8 −5 −7 0 0 p 8 −5 −7 − p

The first matrix represents a state of hydrostatic stress for all values of p, but the second
one represents a state of pure shear only if its first invariant is zero. Thus, we require
that [10 − p] + [15 − p] + [−7 − p] = 0, giving us p = 6. The required decomposition is
     
10 12 8 6 0 0 4 12 8
12 15 −5 = 0 6 0 + 12 9 −5  .
8 −5 −7 0 0 6 8 −5 −13

This completes the solution. But there are some important comments to make. Let us
note that the first matrix represents a state of hydrostatic stress4 , and verify that the
second one a state of pure shear: its first invariant is [4 + 9 − 13 = 0]. Let us note further
3 M E 1 M
Recall the formula = −→ = . Where M is zero, the curvature 1/R is zero.
I R R EI
4
Here it is tensile; we still use the word hydrostatic stress, because it is used in a figurative sense.
13-3 A Few Illustrative Examples

that normal stress elements are present in the second matrix even though it represents
a state of pure shear. We can see — we can prove this if we desire, although we will not
do it here — that there exists a (right handed cartesian) coordinate system (u, v, w) in
which the second matrix appears with no normal stress as
 
0 τuv τuw
 τvu 0 τvw  with τuv = τvu , τvw = τwv , τwu = τuw .
τwu τwv 0

3. Example: Expressions for σ and τ on Octahedral Planes

Obtain the expressions for the normal and shear stresses on the octahedral planes and,
thereby, derive the expressions [Eqs (4.49a - 4.49d), (4.50a - 4.50c), p. 4-35]. We leave
this for the readers to work out.

4. Example: Tresca’s and von Mises’ Criteria

Let us calculate the maximum torque that may be applied on a steel shaft of uniform
diameter d = 20 mm based on the two criteria, Tresca’s and von Mises’. The stress at
which yielding — the onset of yielding is reckoned to be the threshold of failure — in
a simple tension test is 150 MN/m2 . [We shall demonstrate through this example that
the Tresca’s criterion is about 15% more conservative than the von Mises’ one.]
Tresca’s criterion:
We know that the yield stress in shear, τy is one half the yield stress σy in tension.
Equating the maximum shear stress in torsion to the yield strength in shear, we have

16 T 1
τmax = = σy .
π d3 2
Solving for the desired maximum torque, T , we obtain

π d3 π(0.020)3
T = σy = × 150 × 106 = 117.81 Nm (Tresca).
32 32
Von Mises’ criterion:
The state of stress in torsion is
   
0 (τxy = τy ) 0 τy 0 0
(τyx = τy ) 0 0 −→ 0 0 0 
0 0 0 0 0 −τy

The three principal stresses are σ11 = τy ; σ22 = 0; σ33 = −τy . With these values of the
principal stresses, Eq. 4.59 gives us

3τy2 = σy2 τy = 0.577σy . (13.1)

Let us note that the curve bbbb in Fig. 4.20b cuts the vertical axis (y-axis, τ axis) at
this value of 0.577. This is a little more that 15% higher than the value of 0.500, showing
that the Tresca’s criterion is about 15% more conservative than the von Mises’ one.
Advanced Mechanics of Solids 13-4

The maximum torque T is, accordingly,

(0.577) π 3 0.577 π(0.020)3
T = d σy = × 150 × 106 = 117.81 Nm (von Mises).
16 16
This vindicates the statement made earlier that the Tresca’s criterion is about 15% more
conservative than the von Mises’ one.
5. Example: Stress Vector on an Inclined Plane
The state of stress at a point P in a stressed body is given by the stress matrix
 
40 20 30
20 10 −30 ,
30 −30 −10
all in MPa. Calculate the stress vector (magnitude and direction) on a plane passing
through P parallel to the plane x1 + 2x2 − x3 = 2.
From the equation x1 + 2x2 − x3 = 2, we obtain the direction ratios as (1, 2, −1), and
hence the direction cosines as
1 2 −1
n1 = p = 0.408; n2 = √ = 0.816; n3 = √ = −0.408.
2 2
1 + 2 + (−1) 2 6 6
[We do not really need to check if these are orthonormal (n21 + n22 + n23 = 0), but it is
always a good habit to check to be sure that there are no mistakes in the calculations.]
(ν) (ν) (ν)
The stress resultants T1 , T1 , T1 are, using Eq. (4.3) (Cauchy’s result),
 
 (ν) 
 T1 
 
     
40 20 30  0.408   20.412 

 
(ν)
T2  = 20 10
 −30 0.816 = 12.247 all in MPa.
30 −30 −10 −0.408 −8.165
    
 (ν) 

 


T3


(ν)
Thus, the stress vector on the plane is T = 20.412 e 1 + 12.247 e 2 − 8.165 e 3 MPa.
This defines the resultant stress vector completely. Even so, let us calculate its magnitude
and the angle that this makes with the normal to the plane. The magnitude is
(ν) p
 T  = 20.4122 + 12.2472 + (−8.165)2 = 25.166 MPa.
 

The angle between this resultant stress vector and the normal to the plane is given by
(ν)
T .ν (20.412 e 1 + 12.247 e 2 − 8.165 e 3 ) . (0.408 e i + 0.816 e 2 − 0.408 e 3 )
cos−1   = cos−1
(ν) 25.166
 T |νν |
 

= 0.861 −→ the angle = 30.57 ◦ .

The resultant stress is not along the normal. This fact does not surprise us at all.
13-5 A Few Illustrative Examples

6. Example: Normal and Shearing Stresses on an Inclined Plane

The state of stress at a point P in a stressed body w.r.to the xyz coordinate system is
given by the stress matrix  
30 15 −10
 15 20 5 ,
−10 5 10
all in MPa. Determine the normal stress, the shear stress, and the magnitude of the
resultant stress vector at this point on the plane 3x − y + 2z = 5.
Procedure:
(a) First calculate the direction ratios and then the direction cosines l.m, n of the plane.
(ν) (ν) (ν)
(b) Calculate T x , T x , T x using Cauchy’s result.
(c) Calculate the magnitude of the resultant stress vector.
(d) Calculate the normal stress on the ν plane.
(e) Calculate shearing stress on the ν plane.
We shall now carry out the calculations.
(a) If the equation to the plane is x − 2y + z = 5, we know that the direction ratios are
(3, −1, 2). The direction cosines l, m, n of the normal ν to the plane are
3 −2 2
l= p = 0.802; m = √ = −0.535; n = √ = 0.267.
32 + (−1)2 + 22 14 14
(b) Using Cauchy’s result
(ν)
T x = lσxx + mτyx + nτzx = [0.802 × 30] + [(−0.267) × 15] + [0.535 × (−10)] = 16.699 MPa;
(ν)
T y = lτxy + mσyy + nτzy = [0.802 × 15] + [(−0.267) × 20] + [0.535 × 5] = 9.354 MPa;
(ν)
T z = lτxz + mτyz + nσzz = [0.802 × (−10)] + [(−0.267) × 5] + [0.535 × 10] = −4.009 MPa.

(c) Magnitude of the resultant stress vector

v" #
u (ν) 2
u " #2 " #2
(ν) (ν) p
= t T x + T y + T z = (16.699)2 + (9.354)2 + (−4.009)2 = 17.878 MPa.

(d) Magnitude of the normal stress σνν = l2 σxx + m2 σyy + n2 σzz + 2lm τxy + 2mn τyz +
2nl τzx = [0.8022 × 30] + [(−0.267)2 × 20] + [0.5352 × 10] = 7.150 MPa.
√
(e) Magnitude of the shear stress τ(ν) = 17.8782 − 7.1502 = 16.386 MPa.
7. Example: Principal Stresses
The state of stress at a point P in a stressed material is given by the stress matrix
   
σxx τyx τzx 45 15 15
 τxy σyy τzy  = 15 0 30 MPa.
τxz τyz σzz 15 30 0
Advanced Mechanics of Solids 13-6

Determine the principal stresses at the point.

Procedure:
(a) Evaluate the three stress invariants I1 , I2 , I3 .
(b) Write down the characteristic equation and find its roots.
(c) Arrange them in the algebraically decreasing order. These are the principal stresses.
(d) Calculate the sum of the principal stresses and verify that it is an invariant.
(e) Solve using maple and verify that the answers obtained are correct.
We shall now carry out the calculations.
(a) The three stress invariants are

I1 = σxx + σyy + σzz = 45 + 0 + 0 = 45 MPa;

2 2 2
I2 = σxx σyy + σyy σzz + σzz σxx − τxy − τyz − τzx
= (45 × 0) + (0 × 0) + (0 × 45) − 152 − 302 − 152 = −1350 (MPa)2 ;
2 2 2
I3 = σxx σyy σzz − σxx τyz − σyy τzx − σzz τxy + 2τxy τyz τzx
= (45 × 0 × 0) − (45 × 30 × 30) − (0 × 15 × 15) − (0 × 15 × 15) − (2 × 15 × 30 × 15)
= 27000 (MPa)3 .

The characteristic equation of the stress matrix is

σ 3 − I1 σ 2 + I2 σ − I3 = 0;
i.e., σ 3 − 45σ 2 − 1350σ + 27000 = 0 −→ (σ − 15)(σ 2 − 30σ − 1800) = 0.

The roots of this characteristic equation (which are the eigenvalues of the stress matrix)
are 15, 60 and −30, all in MPa.
The principal stresses are the eigenvalues of the stress matrix. These are

σ1 = 60 MPa; σ2 = 15 MPa; σ3 = −30 MPa; (algebraically descending order).

(Generally, the algebraically largest — here it also happens to be the numerically largest
— principal stress is designated as σ1 , and the algebraically lowest designated as σ3 .)
8. Example: Principal Planes
Find the principal planes of the principal stress of the problem above.
The problem of locating the principal planes is the same as finding the eigenvectors of
the given stress matrix. This is the same as finding the non-trivial solutions for l, m, n
from the linear, simultaneous, algebraic, homogeneous equations
    
(σxx − σ) τyx τzx l 
   0 
τ (σ − σ) τ m = 0 . (13.2)
 
 xy yy zy 
   
τxz τyz (σzz − σ)  n   0 

Equivalently,

(σxx − σ) l + τyx m + τzx n = 0;

13-7 A Few Illustrative Examples

τxy l + (σyy − σ) m + τzy n = 0;

τxz l + τyz + (σzz − σ) n = 0.

For our problem, the eigenvalues have already been calculated as

σ1 = 60 MPa; σ2 = 15 MPa; σ3 = −30 MPa; (algebraically descending order).

Eigenvector corresponding to σ1 = 60 MPa.

The relevant linear, simultaneous, algebraic, homogeneous equations are

(45 − 60) l + 15 m + 15 n = 0;
15 l + (0 − 60) m + 30 n = 0;
15 l + 30 m + (0 − 60) n = 0.

First note that we can obtain only the ratios. Why? Because now the solution is not
unique. We can know this from linear algebra in a proper way. A less satisfactory, though
effective, method is to note that, if a, b, c is a solution, 2a, 2b, 2c also is a solution. And
so is any multiple of a, b, c. If we can obtain only the ratios, we can assume l = 1, and
find the corresponding values of m and n.
If we substitute l = 1 in these equations, we obtain

−15 + 15 m + 15 n = 0;
15 − 60 m + 30 n = 0;
15 + 30 m − 60 n = 0.

Here are three equations in only two unknowns m and n. Now there arises the question
of consistency. Are these equations consistent5 ? From the first two equations we find

−15 + 15 m + 15 n = 0;
15 − 60 m + 30 n = 0.

These equations, when solved (which is very easy), give us m = n = 1/2. Thus, the
direction ratios are (1, 0.5, 0.5), which when normalised give us the first eigenvector as
[0.816 0.408 0.408]T .
Eigenvector corresponding to σ2 = 15 MPa.
The relevant linear, simultaneous, algebraic, homogeneous equations now are

(45 − 15) l + 15 m + 15 n = 0;
15 l + (0 − 15) m + 30 n = 0;
15 l + 30 m + (0 − 15) n = 0.
5
Sure, they are? How can we be sure? Consider the condition for consistency. The rank of the coefficient
matrix must be equal to the rank of the augmented matrix. Readers are advised to revise linear algebra.
Such situations occur again and again. It, therefore, pays to learn linear algebra in the right royal way.
Advanced Mechanics of Solids 13-8

Again, as before, let us put l = 1 in these equations. Taking, say, the first two equations
we find
30 + 15 m + 15 n = 0;
15 − 15 m + 30 n = 0.
These equations, when solved (which is very easy), give us m = n = −1. Thus, the
direction ratios are (1, −1, −1), which when normalised give us the second eigenvector
as [−0.577 0.577 0.577]T .
Eigenvector corresponding to σ3 = −30 MPa.
The relevant linear, simultaneous, algebraic, homogeneous equations now are
(45 + 30) l + 15 m + 15 n = 0;
15 l + (0 + 30) m + 30 n = 0;
15 l + 30 m + 30 n = 0.
This time, let us put n = 16 . Taking, say, the first two equations we find
75 l + 15 m = −15;
15 l + 30 m = −30.
These equations, when solved (which is very easy), give us l = 0, m = −1. Thus, the
direction ratios are (0, −1, 1), which when normalised give us the third eigenvector as
[0 − 0.707 0.707]T .
Now that the eigenvalues (principal stresses) are distinct (σ1 6= σ2 6= σ3 ), the correspond-
ing eigenvectors are mutually orthogonal. Physically, this means that now the principal
planes are mutually perpendicular7 .
It is desirable — soon it would become essential — for students to learn how to use
Computer Algebraic Systems like mathematica and maple.
As a motivation to become familiar with Computer Algebraic Systems, the solution by
maple is given below.
Solution (of the Last Two Examples) by maple
If we use maple, the solution of the last two examples — finding the principal stresses
and the principal planes, which is the same as finding the eigenvalues and eigenvectors
of the given stress matrix — is only one click away. The answers, given below, are
displayed in a screen shot [Fig. 9.20]. The numbers are to be interpreted properly.
What is obtained are the direction ratios; the direction cosines can be easily calculated
from them.
σ1 = 60 : {2 1 1}T ; σ2 = 15 : {−1 1 1}T ; σ3 = −30 : {0 −1 1}T .
6
We can give any value to any one of l, m, n. What would happen if we put l = 1? Examine and find out.
7
If the eigenvalues are not distinct, the corresponding eigenvectors may not be orthogonal. However, it
is always possible (by the Gram-Schmidt procedure) to find a set of three mutually orthogonal principal
directions. These are important; students are advised to look up all this in two or three good books.
Geometrical visualisation (say, using Lamé’s ellipsoid) will also be insightful. See [10] or, better still, [5].
13-9 A Few Illustrative Examples

Fig. 13.2 shows a the screen display of the

maple solution. The eigenvectors are the
direction ratios; they are to be normalised
to obtain the direction cosines. Further, as
the eigenvalues are distinct (are all differ-
ent) (σ1 6= σ2 6= σ3 , arranged in the alge-
braically decreasing order), the correspond-
ing eigenvectors (representing the principal
planes) are mutually orthogonal. That is,
Figure 13.2: maple solution: screen the principal planes are mutually orthogonal!
display. The eigenvalues are: σ1 = 60; Let us also note that (0 − 1 1)T is the
σ2 = 15, σ3 = −20, all in MPa. same eigenvector as (0 1 − 1)T .

The direction cosines are the normalised direction ratios. Thus, we have
2
√ = 0.816, etc. σ1 = 60 : {0.816 0.408 0.408}T ;
22
+ 1 2 + 12
σ2 = 15 : {−0.577 0.577 0.577}T ; σ3 = −30 : {0 − 0.707 0.707}T .

As the principal stresses are distinct, the corresponding direction ratios will be mutually
orthogonal. Thus, we verify that

(a): 1.2 : −→ [(0.816) × (−0.577)] + [0.408 × 0.577] + [0.408 × 0.577] = 0;

(b): 2.3 : −→ [(−0.577) × 0] + [0.577 × (−0.707)] + [0.577 × 0.707] = 0;
(c): 3.1 : −→ [0 × 0.816] + [(−0.707 × 0.408] + [0.707 × 0.408] = 0.

9. Example: Strain Transformation Equations

If the coordinate axes are changed from (x, y) to (x0 , y 0 ) — θ is the angle between the x
and the x0 axes — show by coordinate transformation that

ex0 x0 = exx cos2 θ + eyy sin2 θ + exy sin 2θ.

The old and the new axes related by8

x = x0 cos θ − y 0 sin θ; u0 = u cos θ + v sin θ;

y = x0 sin θ + y 0 cos θ; v 0 = −u sin θ + v cos θ.

The strain-displacement relations are exx = ∂u/∂x, e0x0 x0 = ∂u0 /∂x0 , etc. Thus, by the
chain rule of partial differentiation, we have

∂u0 ∂u0 ∂x ∂u0 ∂y

ex0 x0 = = + ;
∂x0 ∂x ∂x0 ∂y ∂x0
8
Draw two sets of axes (x, y and (x0 , y 0 ) and obtain the given equations. (u, v) and (u0 , v 0 ) are the displace-
ment components in the old and new coordinate systems. The results follow from the geometry.
 Advanced Mechanics of Solids 13-10

∂v 0 ∂v 0 ∂x ∂v 0 ∂y
ey 0 y 0 = = + ;
∂y 0 ∂x ∂y 0 ∂y ∂y 0
 0
∂u0 1 ∂v 0 ∂x ∂v 0 ∂y 1 ∂u0 ∂x ∂u0 ∂y
    
1 ∂v
ex0 y0 = + = + + + .
2 ∂x0 ∂y 0 2 ∂x ∂x0 ∂y ∂x0 2 ∂x ∂y 0 ∂y ∂y 0
Thus, we obtain
   
∂u ∂v ∂u ∂v
e x0 x0 = cos θ + sin θ cos θ + cos θ + sin θ sin θ
∂x ∂x ∂y ∂y
 
∂u ∂v ∂v ∂u
= cos2 θ + sin2 θ + + sin θ cos θ
∂x ∂y ∂x ∂y
= exx cos2 θ + eyy sin2 θ + exy sin 2θ.

10. Example: Strains from Stresses

Calculate the strain matrix corresponding to the given stress matrix (all in MPa), if
E = 200 GPa and ν = 0.3.
   
150 50 −60 0.720 0.325 −0.390
 50 100 −70 −→  0.325 0.395 −0.455 × 10−3 .
−60 −70 −80 −0.390 −0.455 −0.775
The strain components are calculated from the generalised Hooke’s law. G = E/[2(1 +
ν)] = 200/2.6 = 76.923 GPa.
1 1
exx = [σxx − ν(σyy + σzz )] = [150 − 0.3(100 − 80)] = 0.72 × 10−3 ;
E 200 × 103
1 1
eyy = [σyy − ν(σzz + σxx )] = [100 − 0.3(150 − 80)] = 0.395 × 10−3 ;
E 200 × 103
1 1
exx = [σzz − ν(σzz + σxx )] = [−80 − 0.3(150 + 100)] = −0.775 × 10−3 ;
E 200 × 103
1 τxy 50
exy = γxy = = = 0.325 × 10−3 ;
2 2G 2 × 76.923
1 τyz −70
eyz = γyz = = = −0.455 × 10−3 ;
2 2G 2 × 76.923
1 τzx −60
ezx = γzx = = = −0.390 × 10−3 .
2 2G 2 × 76.923
The strain matrix, obtained from the generalised Hooke’s law, is displayed alongside the
given stress matrix.
11. Example: Stresses from Strains
Given the strain matrix shown in the example above, calculate the stress matrix. If our
calculations are correct, we must obtain the stress matrix given in the previous example.
Now we know the strain components. We again use the generalised Hooke’s law, but in
the form given below, and calculate the stress components. Let us first calculate λ, e,
and G.
νE
λ= = 115.385 MPa;
(1 + ν)(1 − 2ν)
 13-11 A Few Illustrative Examples

e = exx + eyy + ezz = 0.720 = 0.395 − 0.775 = 0.340 × 10−3 ;

E
G= = 76.923 GPa.
2(1 + ν)
With these values, the stress components are worked out as below.

σxx = λe + 2Gexx = 150 MPa;

σyy = λe + 2Geyy = 100 MPa;
σzz = λe + 2Gezz = −80 MPa;
τxy = 2Gexy = 50 MPa;
τyz = 2Geyz = −70 MPa;
τzx = 2Gezx = −60 MPa.

We obtain the original stress matrix that we started out with in the example above. We
note that e is the first strain invariant, and that λ and G are the Lamé’s constants.

12. Example: Transformation of the Strain Matrix

The strain matrix referred to the (x, y, z) coordinates is given in the previous example.
Find the strain matrix in the ‘new’ coordinate system specified by the angles in the table
below which shows how the new axes are disposed relative to the old ones.

Old coordinates
x1 x2 x3
New coordinates

√ 
x01 30◦ 2
3
60◦ ( 12 ) 90◦ (0)
√ 
x02 120◦ (− 12 ) 30◦ 2
3
90◦ (0)

x03 90◦ (0) 90◦ (0) 0◦ (1)

In three dimensions it is more convenient to work with the direction cosines and not
with the angles. The direction cosines in bold within brackets are shown in the table.

13. Example: Compatibility

Is the given two-dimensional strain field possible?

exx = (x2 + 2y 2 + 3xy + 2) × 10−3 ; eyy = (2x2 + y 3 + x2 y 2 + 3) × 10−3 ;

γxy = [2xy(x2 y + xy 2 ) + 1] × 10−3

To answer this question, we need to check if the following compatibility equation is

satisfied.
∂ 2 exx ∂ 2 eyy ∂ 2 γxy
+ =
∂y 2 ∂x2 ∂x∂y
 Advanced Mechanics of Solids 13-12

Let us calculate the various partial derivatives in the compatibility equation.

∂ 2 exx ∂ 2 eyy ∂ 2 γxy

= 4 × 10−3 ; = (4 + 2y 2 ) × 10−3 ; = (12x2 y + 12xy 2 ) × 10−3 .
∂y 2 ∂x2 ∂x∂y

As the compatibility equation is not satisfied, the given strain field is not possible.

14. Example: Compatibility

Is the strain field given below possible?

exx = 4x2 y; eyy = 2y 2 x; exy = 3xy + 2x3 .

To answer this, we check if the compatibility equation is satisfied. Let us calculate the
terms in the compatibility equation

∂ 2 exx ∂ 2 eyy ∂ 2 γxy

+ = .
∂y 2 ∂x2 ∂x∂y

∂ 2 exx ∂ 2 eyy ∂ 2 γxy

= 0; = 0; = 6.
∂y 2 ∂x2 ∂x∂y
The compatibility equation is not satisfied; it is impossible to have the given strain field.

15. Example: Rigid Body Displacements

The displacement field is given by the following equations. What do the last terms
within the flower brackets {· · · } represent? The C’s and u0 , v0 , w0 are constants. Refer
to Fig. 11.7b, p. 11-7.
P
u= [C13 x + 2 C34 y + 2 C36 z]+ {(+ω2 z − ω3 y) + u0 } ; (13.3a)
A
P
v = [C23 y + 2 C35 z] + {(+ω3 x − ω1 z) + v0 } ; (13.3b)
A
P
w = [C33 ] + {(+ω1 y − ω2 x) + w0 } . (13.3c)
A
The quantities u0 , v0 , w0 in the above equations are constants. They represent transla-
tions of the body in the x, y, z directions, respectively. Consider the term −ω3 y in Eq.
(13.3a). This represents a rotation about the z axis. Refer to Fig. 11.7a, p. 11-7. If we
draw line, say, OP and rotate it about the z axis, we notice that the rotation introduces
u and v displacements. A little reflection would convince us that this ω3 represents a
rotation about the z axis. In the same manner we can see that ω1 , ω2 represent rotations
about the x and y axes, respectively. It is useful to be aware of these facts.

16. Example: Rotation and Strains

When a (deformable) body rotates about an axis (say, the z axis) through an angle
θ — this is a rigid body rotation — there are displacements u and v in the x and y
directions, respectively. Calculate the strains and rotation ωz if the displacements are
u = u(x, y) = (cos θ − 1) x − (sin θ) y; v = v(x, y) = (sin θ) x + (cos θ − 1) y.
 13-13 A Few Illustrative Examples

The strains and rotations are

∂u
exx = = (cos θ − 1);
∂x
∂v
eyy = = (cos θ − 1);
∂y
∂u ∂v
exy = + = − sin θ + sin θ = 0;
∂y ∂x
∂u ∂v
ωz = − = − sin θ − sin θ = −2 sin θ.
∂y ∂x
The shear strain is zero, but the rotation is not.

17. Example: Strains and Rotations

Verify that the displacement field given corresponds to no strain, but only to rotations.

u = 3y + 2z : v = −3x + z; w = −2z − y.

Let us calculate the strain components given by:

∂u ∂v ∂w ∂u ∂v
exx = = 0; eyy = = 0; ezz = = 0; γxy = + = 3 − 3 = 0;
∂x ∂y ∂z ∂y ∂x
∂v ∂w ∂w ∂u
γyz = + = −2 + 2 = 0; γzx = + = 1 − 1 = 0.
∂z ∂x ∂x ∂y
The strains are all zero. We shall presently see that the rotations are not zero.
   
1 ∂u ∂v 1 ∂v ∂w
ωxy = − = 3 6= 0; ωyz = − = 1 6= 0;
2 ∂y ∂x 2 ∂z ∂y
 
1 ∂w ∂u
ωzx = − = −2 6= 0.
2 ∂x ∂z

18. Example: Some Special Cases of Displacement Fields

Four different displacement fields are given below. Calculate the corresponding strains
/ strain fields. What special cases do they represent? (a and λ are constants.)

(a): u = λx; v = w = 0;
(b): u = 2ay; v = w = 0;
(c): u = λx; v = λy; w = λz;
(d): u = u(x, y); v = v(x, y); w = 0.

To answer this question, let us first calculate the various strain components. Then we
can see what special case each of them represents.
Case (a):
∂u ∂v ∂w
exx = = λ; eyy = = 0; ezz = = 0;
∂x ∂y ∂z
 Advanced Mechanics of Solids 13-14

∂u ∂v ∂v ∂w ∂w ∂u
γxy = + = 0 : γyz = + = 0 : γzx = + = 0.
∂y ∂x ∂z ∂y ∂x ∂z
This, then, represents a simple extension in the x direction.
Case (b):

∂u ∂v ∂w
exx == 0; eyy = = 0; ezz = = 0;
∂x ∂y ∂z
∂u ∂v ∂v ∂w ∂w ∂u
γxy = + = 2a : γyz = + = 0 : γzx = + = 0.
∂y ∂x ∂z ∂y ∂x ∂z
This, therefore, represents a simple case of shearing.
Case (c):

∂u ∂v ∂w
exx =
= λ; eyy = = λ; ezz = = λ;
∂x ∂y ∂z
∂u ∂v ∂v ∂w ∂w ∂u
γxy = + = 0; γyz = + = 0 : γzx = + = 0.
∂y ∂x ∂z ∂y ∂x ∂z
What case does this case represent? Well, a uniform dilatation exx + eyy + ezz = 3λ.
Each (normal) strain components is equal to λ/3.
Case (d):

∂u ∂v ∂w
exx = ; eyy = ; ezz = = 0;
∂x ∂y ∂z
∂u ∂v ∂v ∂w ∂w ∂u
γxy = + ; γyz = + = 0 : γzx = + = 0.
∂y ∂x ∂z ∂y ∂x ∂z
Now exx , eyy and γxy are all present in general. The components ezz , γxz , γyz are all
zero. This is the general case of a state stress known as plane strain.

19. Example: Solution of an Elasticity Problem?

Examine if the following stress field represents the solution of a problem in elasticity.

σxx = a y 2 + ν(x2 − y 2 ) ; σyy = a x2 + ν(y 2 − x2 ) ; σzz = aν(x2 + y 2 );

   

τxy = −2aν xy; τyz = τzx = 0.

Let us first check if these stress components satisfy the equations of equilibrium (with
no body forces).

∂σxx ∂τyx ∂τzx

+ + =0 −→ 2aν x − 2aν x + 0 = 0 (satisfied);
∂x ∂y ∂z
∂σyy ∂τxy ∂τyz
+ + =0 −→ 2aν y − 2aν y + 0 = 0 (satisfied);
∂y ∂x ∂y
∂σzz ∂τxz ∂τyz
+ + =0 −→ 0+0+0=0 (satisfied).
∂z ∂x ∂y
 13-15 A Few Illustrative Examples

Yes, the equations of equilibrium are satisfied. However, it is premature to make any
conclusion yet. This is a stress formulation. The compatibility conditions play a vital
role; their role is not passive now. The compatibility conditions are written in terms
of the strain components. To check if the compatibility equations are satisfied, we
therefore need to work out the expressions for the strain components. We shall obtain
the expressions for the strain components using the constitutive equations (generalised
Hooke’s laws).
1
exx = [σxx − ν(σyy + σzz ] = · · · ;
E
1
eyy = [σyy − ν(σzz + σxx ] = · · · ;
E
1
ezz = [σzz − ν(σxx + σyy ] = · · · ;
E
··· = ···

Now we can check if the compatibility equations are satisfied. We can see that all the
compatibility equations are not satisfied. Hence we conclude that the given stress field
cannot represent the solution of any elasticity problem. Readers are advised to complete
the solution and to convince themselves that the statement made above is correct.

20. Example: Pure Bending of a Uniform Beam of Rectangular Cross-section

(a) A uniform beam in pure bending (b) The cross-section

Figure 13.3: A uniform beam of rectangular cross-section in pure bending. Fig. 13.3b shows
the cross-section before and after bending.

Solve the problem of pure bending of a uniform beam of rectangular cross-section [Fig.
13.3]. The usual notations are followed.
From our earlier knowledge of the theory of bending, the stress components are
My
bending stress σxx = ; σyy = σzz = 0; τxy = τyz = τzx = 0.
I
From the generalised Hooke’s law, we find that
My My
exx = ; eyy = ezz = −ν ; exy = eyz = ezx = 0.
I EI
Advanced Mechanics of Solids 13-16

Let us take this as a guess of the displacement field. We can then use the strain-
displacement equations and calculate the displacements.
∂u My ∂v My ∂w My
exx = = ; eyy = = −ν ; ezz = = −ν ; (13.4)
∂x EI ∂y EI ∂z EI
     
1 ∂u ∂v 1 ∂v ∂w 1 ∂w ∂u
exy = + = 0; eyz = + = 0; ezx = + = 0. (13.5)
2 ∂y ∂x 2 ∂z ∂y 2 ∂x ∂z

These equations are now integrated to yield

M
u= xy +{(ω2 z − ω3 y) + u0 }; (13.6a)
EI
M
v=− [x2 + ν(y 2 − z 2 )]+{(ω3 x − ω1 z) + v0 }; (13.6b)
2 EI
νM
w=− yz +{(ω1 y − ω2 x) + w0 }. (13.6c)
EI
We note that the displacements can be determined only to within rigid body displace-
ments that do not contribute to strains and, therefore, to stresses. The terms {· · · } are
these rigid body displacements, u0 , v0 , w0 representing translations, and ω1 , ω2 , ω3 rota-
tions, along / about the x, y, z axes, respectively. This is the general form representing
rigid body displacements. We have seen this before (p. 13-12).
The rigid body translations can be arrested by forcing the point O to be fixed; that
is, by demanding that u = v = w = 0 there at O (x = y = z = 0). Furthermore, we
can assume that (i) an element of the x−axis at the fixed end is zero, and that (ii) an
element of the xy plane is also fixed. These conditions at x = y = z = 0,

∂v ∂w ∂w
u = v = w = 0; = = =0
∂x ∂x ∂y

give the displacement field — the rigid body displacements are now arrested — as

M
u= xy;
EI
M
v=− [x2 + ν(y 2 − z 2 )];
2 EI
νM
w=− yz.
EI
We have thus arrived at a displacement field that appears to be reasonable. We can be
sure that this gives us a correct solution if all the governing equations are satisfied9 . By
the uniqueness theorem we are also sure that this is the only correct solution.
9
(i) Calculate the strain field using the strain-displacement relations.
(ii) Calculate the stress field using the constitutive equations (generalised Hooke’s law).
(iii) Check to make sure that all the equations of equilibrium are satisfied.
The boundary conditions must be satisfied too. The stress distributions at the ends must lead to a moment
M . All these can be seen to be satisfied.
 13-17 A Few Illustrative Examples

21. Strain: Change in Length

A rectangular plate ABCD of length 400 mm (along AB) and width 300 mm (along AD
is subjected to a uniform two-dimensional strain field: exx = 0.002; eyy = 0.003; γxy =
0.001. The x and y coordinate axes are along the edges AB and AD of the plate with
its origin at the lower left corner A of the plate. Calculate the length of the line element
AC (diagonal of the block) before and after deformation.
√
Length of the diagonal AC (before deformation) is obviously 42 + 32 = 500 mm. It
is evident that the direction cosines (l, m, n) of the line element AC are l = cos 30 ◦ =
0.866; m = cos 60 ◦ = 0.500; n = cos 90 ◦ . We use the strain transformation equation to
calculate the strain along AC. This gives us

eAC = l2 exx + m2 eyy + n2 ezz + 2lm exy + 2mn eyz + 2nl ezx
 
2 2 0.001
= 0.866 × 0.002 + 0.500 × 0.003 + 2 × 0.866 × 0.500 ×
2
= 0.00268.
0 0
length A C = 500 × 1.00268 = 501.341 mm.

The length of the line element A0 C 0 = 501.341 mm. This is the length after deformation.
(The line AC before deformation becomes A0 C 0 after deformation.)

22. Example: Taylor & Quinney Experiments

We have already seen the results of the famous Taylor & Quinney experiments [Fig.
4.20b]. The experimental results lie between the predictions of the Tresca’s (curve aaaa)
and the von Mises’ (curve bbbb) criteria. Both these curves aaaa and bbbb are ellipses.
We shall now obtain the equations to these ellipses.
For the test specimen in Taylor & Quinney experiments on thin-walled tubes, if we
take the x- and y-axes along the axis of the tube and perpendicular to it [Fig. 4.20a]
respectively, the stress components are
   
σxx τxy 0 σ11 0 0
 τyx 0 0 −→  0 0 0 .
0 0 0 0 0 σ33

The three principal stresses and the maximum shear stress are
r
σxx σxx 2 2 ;
σ11 = + + τxy
2 2
σ22 = 0;
r
σxx σxx 2 2 ;
σ33 = − + τxy
2 2
r
σ11 − σ33 σxx 2 2 .
τmax = = + τxy
2 2
 Advanced Mechanics of Solids 13-18

(a) Tresca’s criterion:

Thus, the equation that governs the Tresca’s criterion is
r
σxx 2 2 = y,
σ
+ τxy
2 2
which gives the equation to the ellipse as

σxx 2 τxy 2
   
+4 =1 (curve aaaa in Fig. 4.20b). (13.7)
σy σy
(b) Von Mises’ criterion:
The equation that represents the von Mises’ criterion is likewise
" r #" r # " r #2
σxx σxx 2 σ xx σ xx
2  σ 2
xx
+ 2
+ τxy − 2
+ τxy + 2 2
+ τxy = 2σy2 ,
2 2 2 2 2

which, on simplification, is the equation to the ellipse as

σxx 2 τxy 2
   
+3 =1 (curve bbbb in Fig. 4.20b). (13.8)
σy σy

23. Example: A Beam of Two Materials

(a) Cross-section (b) Tension and compression areas (c) Force resultants T and C

Figure 13.4: The cross-section of an RCC beam (dimensions in mm). Four (4) reinforcing
steel rods, the neutral axis, the ‘transformed area’ m As , and the force resultants (T, C)
are shown. The last figure on the right shows that the tension in the concrete is neglected.
The strain distribution across the cross-section is still linear [Fig. 2.5b, p. 2-7].

Examine the behaviour of a reinforced cement concrete (RCC) beam in bending10 . The
dimensions of the cross-section are d = 500 mm (effective depth) and b = 230 mm
(width). The concrete is M20; there are four (4) steel bars, each of 16 mm diameter.
The bending moment applied is M = 50 kNm.
10
I am grateful to Professors K.V. Leela of Vidya and Devdas Menon of IIT Madras, both Professors of
Structural Engineering for their help in this example.
13-19 A Few Illustrative Examples

The beam (cross-section) is clearly of two materials, viz., concrete and steel. The cross-
sectional area As of the steel bars is multiplied by a factor m = Es /Ec to obtain the
‘transformed area’ (or the ‘equivalent area’) reckoned from the point of view of that
there is only one material11 . Generally the presence of tensile stress in the concrete (in
the tension zone) is disregarded. The neutral axis will not pass through the mid-section;
it will be at a depth of kd as shown in the figure.
As always the equilibrium requirements are:
Z
i) σ dA = 0 −→ (compressive force=tensile force; net axial force = 0)
Z A
ii) σ y dA = M −→ (moment = applied bending moment)
A

Equating the moments about the neutral axis, we obtain

kd
b×kd× = m As (d − k d), (13.9)
2
where m As is the ‘transformed area’ (or the ‘equivalent area’), the value of m as per
codal provisions being
280
m= ,
3 (σc )bc
where (σc )bc is the allowable stress in concrete in bending compression. For M 20 concrete
this is (from IS 456) (σc )bc = 7 N/mm2 .
280
m= = 13.33 (which may be taken as 13 as per the relevant code.)
3×7
k = 0.3485 which defines where the neutral axis is.

The equilibrium condition, total compressive force = total tensile force, gives us
σc
× k d × b = σs × As (13.10)
2
σc kd
= . (13.11)
(σs /m) d−kd
The equation, moment of resistance = M , gives us
 
kd
σ s × As d − = M = 50 kNm. (13.12)
3

From these equations (13.9) or [(13.10) and (13.11)], the factor k can be found out. Its
value is obtained as k = 0.3485. Using Eq. (13.12) we obtain the stresses as

σc = 5.78 N/mm2 , σs = 140.68 N/mm2 .

11
This ratio is nominally m = Es /Ec . In reinforced cement concrete calculations, some modification is made
280
as specified by the code: m = . See the comment at the end of this example.
(σc )bc
 Advanced Mechanics of Solids 13-20

Comments: Reinforced concrete calculations are not quite the same as those in the
mechanics of solids. There are several details that cannot be discussed here. There
are codal provisions. These codes are arrived at, recommended, and incorporated as
mandatory provisions by official bodies with the support of subject experts based on (i)
deep analytical studies, (ii) experimental results, (iii) professional experience, and (iv)
sound engineering judgement.
If we use the concept of a transformed area, its width will exceed the width of the
concrete beam. The empirical formula for m is to accommodate for creep in concrete
which effectively enhances the modular ratio with time. In structural design practice,
this working stress design approach is outdated. A limit state design approach at collapse
involving nonlinear stress-strain behaviour of concrete and steel is used.
For example, we know that the stress-strain relationship of concrete is nonlinear. That
is to say, the slope of the stress-strain curve representing the modulus of elasticity is dif-
ferent at different points (at different stress levels). The factor m, which is theoretically
the ratio m = Es /Ec is, thus, dependent on the stress level. These aspects are all ac-
knowledged and looked at by the experts. Although there are several such modifications
specified by the codes, the science of the mechanics of solids is still the guiding light and
the fundamental basis of these calculations.

24. Example: Cross-shear in a Rectangular Cross-section

Find the shear stress distribution on the rectangular cross-section of a beam in bending.

(a) A beam element with a b × d section (b) Parabolic shear stress distribution

Figure 13.5: The figures show how the shear stress in bending (sometimes called cross-shear)
occurs. Fig. 13.5a shows a rectangular (b × d) cross-section. Fig. 13.5b shows the parabolic
shear stress distribution across the rectangular cross-section.

Fig. 13.5a shows the rectangular cross-section of the beam. An element abcd is shown.
From Eq. (2.22, p. 2-16) we borrow the result
Z Z d/2
FQ F F
τ= = y dA = y (b dy)
Ib I b abcd I b y1
"  #
F y 2 d/2 F d 2
=  = − y12
I 2 y1 2I 2
 13-21 A Few Illustrative Examples

This is the distribution of the shear stress (parabolic). The maximum is at the neutral
axis. To find this value, we note that I = bd3 /12 and y1 = 0. The maximum is thus
 2
F d 3F 3F
τmax = = = . (13.13)
2 × bd3 /12 4 2 bd 2A

F/A is the ‘average’ shear stress on the cross-section. The maximum is 50% higher than
this average value12 . This result is important.

25. Same Example: Alternative Method

We shall solve this problem again by an alternative method. We begin with the (differ-
ential) equation of equilibrium with no body force.

∂σxx ∂τyx ∂τyx ∂σxx

+ =0 −→ =− . (13.14)
∂x ∂y ∂y ∂x

But we already know how σxx , the bending stress, is distributed. It is given by (the
Euler-Bernoulli equation)
 
My ∂σxx ∂M y Fy ∂M
σxx = −→ = = = F, the shear force .
I ∂x ∂x I I ∂x

Substituting this in Eq. (13.14), we obtain

∂τyx ∂σxx y Fy
=− = =− . (13.15)
∂y ∂x I I

On integration, this yields

F y2
τyx = −
+ C. (13.16)
2I
This constant of integration C is evaluated using the boundary condition
 
τyx  d = τyx  = 0, (13.17)
 
d
(y= 2 ) (y=− 2 )

because at a boundary point — (at the end y = d/2 and also at y = −d/2) — the shear
stress τyx complementary to τxy cannot exist on a free surface. Applying the boundary
condition in Eq. (13.16), we obtain
 F (d/2)2 F d2
τyx  =0 −→ 0=− +C −→ C= .

(y=± d2 ) 2I 8I
12
Popov [11] remarks: “Eq. (13.13) is very useful. It is widely used in the design of wooden beams since
the shear strength of wood on planes parallel to the grain is small. Thus, although equal shear stresses
exist on mutually perpendicular planes, wooden beams have a tendency to split longitudinally along the
neutral axis. Note that the maximum shear stress is 1.5 times as great as the average shear stress F/A.
Nevertheless, in the analysis of bolts and rivets, it is customary to determine their shear strengths by
dividing the shear force F by the cross-sectional area A. · · · ”.
 Advanced Mechanics of Solids 13-22

The final answer is that the shear stress on the cross-section τxy = τyx is distributed as
"  #
F d 2 2
τxy = τyx = −y (parabola)
2I 2

with its maximum at the neutral axis as

3F
τmax = as before (I = bd3 /12; A = b d).
2 A
Comment: We know that all problems in the theory of elasticity are statically in-
determinate internally. But here we obtained the solution using only the equation of
equilibrium. How was this possible? Where is the mistake? What is the explanation?
Here we are not solving the problem in the right royal way of the mathematical theory
of elasticity. Instead we are following an engineers’ approach. We recall that we made
a crucial assumption that plane cross-sections remain plane even after bending. This is
the Love-Kirchhoff assumption which is an assumption on the kinematics of deformation.
Such assumptions simplify the solution of problems and are, therefore, often employed
in developing an engineers’ theory. Thus, the apparent anomaly is amicably settled.

26. Example: Shear Centre

Locate the shear centre of a channel section shown in Fig. 13.6a.

(a) A channel section (b) The shear stress distribution

Figure 13.6: Fig. 13.6a shows a channel section with a uniform thickness t. C is the
centroid, and S is the shear centre to be located. Fig. 13.6b shows the distribution of the
shear stress in the two horizontal flanges and in the vertical web.

The figure shows a channel section with a uniform thickness t. C is the centroid, and S
the shear centre to be located. If V is the vertical shear force on a typical cross-section,
there will be shear stresses on the cross-section in the two horizontal flanges and in the
vertical web. [If there is no shear force (that is, if this is a case of pure bending in the
absence of any shear force), the shear centre has no significance.] The shear stress in the
flanges and in the web are distributed [Fig. 13.6b], growing from 0 at A and reaching a
maximum at D on the neutral axis and then gradually13 reducing to 0 at D.
13
If the thickness changes abruptly, the shear stress also changes abruptly. Really, however, the stress cannot
13-23 A Few Illustrative Examples

The method to locate the shear centre is (i) first, to recognise the distribution of the shear
stress in the flanges and in the web, and (ii) then to find the magnitude of the moment
due to this shear stress distribution. If the external vertical load passes through the
shear centre S at a distance of e from the web as marked in Fig. 13.6a, the net moment
shall be zero if the beam should only bend without twisting. This condition enables us
calculate the moment arm of the external load and, thus, to locate the shear centre S.
We shall now undertake the detailed calculations.
We recall how the shear stress (cross-shear) distribution is calculated. Consider the
shaded area in the top flange of length s in Fig. 13.6a. The moment of this area about
the neutral axis is area × moment arm = (s × t)d/2. The shear flow q is given by
VQ V (s × t) d2
q= = .
I I
Here V is the vertical shear force on the cross-section, and I the second moment of the
area about the neutral axis.
With this information we are ready to calculate the total horizontal shear force V1 (to
the left on the top flange).
Z b
V (s × t × d) V td b V t d b2
Z
V1 = ds = s ds = .
0 2I 2I 0 4I
An equal force V2 acts to the right on the bottom flange. Further, the vertical shear
force V2 on the web (acting downwards) is the same as V ; there is no need for any
calculation. Now this force system of V1 , V2 and V3 leads to (i) no horizontal resultant,
(ii) the vertical force V = V3 (downwards), and (iii) an anti-clockwise moment V1 × d.
If the beam should only bend without any twisting, there should be no net twisting
moment. Thus, this anti-clockwise moment V1 × d should balance the clockwise moment
of the vertical shear force V = V3 . Thus, we have
V1 d V t d b2 d t d2 b2
V × e = V1 × d −→ e=
= = . (13.18)
V 4I V 4I
The second moment of the area of this channel section is
"  2 #
t d3 b t3 d
I = Iweb + 2 If lange = +2 + bt
12 12 2
t d3 1 1
≈+ t b d2 = t d2 (6 b + d). (13.19)
12 2 12
The first term within the brackets [· · · ] is negligibly small in comparison to the other
term (the cube of a small quantity t being very small). Substituting this expression for
for I in Eq. (13.18), we obtain the distance e as
3 b2 b
e= = . (13.20)
6b + d 2 + 3db
change discontinuously; there will be local redistribution of stresses. In a simplified theory as we consider
now, these finer details cannot be taken into account.
 Advanced Mechanics of Solids 13-24

This locates the shear centre. If the numerical values of the dimensions b and d are given,
we can find the location of the shear centre. For example, if d = 160, b = 90, t = 3 (all
in mm), the distance e works out to
b 90
e= d
= = 34.71 mm.
2 + 3b 2 + 160/270

We may note that the uniform thickness does not figure in the final result.

27. Example: Unsymmetrical Bending

Fig. 13.7 shows a cantilever with an end load P = 3, 000 N as shown. The dimensions

Figure 13.7: A beam in unsymmetrical bending

are given. Locate the neutral axis and find the maximum tensile stress.
Procedure:
We shall first see the steps in fair detail before we undertake the detailed calculations.

(i) Understand the problem. The line of the load is not along, nor parallel to, any
principal axis. Hence this is a problem in unsymmetrical bending.
(ii) Convert this one problem in unsymmetrical problem into two problems, each of
which is a case of symmetrical bending.
(iii) To do this, the load P is resolved along uu and vv which are the principal axes
of inertia. (Why? Because axes of symmetry are automatically principal axes of
inertia.)
(iv) Note that the maximum stress occurs on the cross-section that has the largest
bending moment. This is at the fixed end.
(v) The bending moments at the fixed end are

Mu = P cos θ × l Mv =P sin θ × l
◦
= 3000 cos 35 × 1 = 2475.5 Nm; =3000 sin 35 ◦ × 1 = 1720.7 Nm.

(vi) The principal second moments of area (area moments of inertia) are
1 1
Iuu = (30)(603 ) = 0.54 × 10−6 m4 ; Ivv = (60)(303 ) = 0.135 × 10−6 m4 .
12 12
(vii) Use the Euler-Bernoulli equation to calculate the stresses.
 13-25 A Few Illustrative Examples

(viii) Recognise that A is the point where the tensile stress is the largest. Why? A is in
the tension zone for both Mu and Mv , and farthest from the neutral axis.
(ix) Let us note that the neutral axis is characterised by the fact that the bending stress
at all points on it is zero.
M v u Mu v u Mu Ivv
σ= + −→ =− .
Iv Iu v Mv Iuu
The angle α that defines the neutral axis is given by
u Mu Ivv
tan α = =− .
v Mv Iuu
We can now complete the solution by working out these quantities numerically.
The largest (maximum) tensile stress is at A; its numerical value is
Mu 60 Mv 30 2457.5 × 0.03 1720.7 × 0.015
σmax = × + × = + = 327.6 MPa.
Iu 2 Iv 2 0.54 × 10−6 0.135 × 10−6
tan α = 0.357; α = 19.6 ◦ .

Note particularly that α 6= 35 ◦ ; that is, the neutral axis is not perpendicular to
the line of the applied load! Note further that, if Iuu = Ivv , the angle α = θ = 35 ◦ .
Why is this so? If Iuu = Ivv (as, for example, for a square cross-section, then every
direction is a principal direction! The inertia tensor then becomes an isotropic
tensor! Note further that, if Iuu = Ivv (as for example, for a square cross-section),
then every direction is a principal direction! The inertial tensor then becomes an
isotropic tensor!

28. Example: Bending and Twisting of a Channel Section

(a) A channel section: inclined load (b) Deflections at the end of a cantilever

Figure 13.8: A load P acts at the end of a cantilever as shown [Figs 13.8a, 13.8b]; it does
not pass through the shear centre.

A load P acts at the end of a cantilever with a uniform channel section as shown [Figs
13.8]. We desire to calculate (i) the stresses, and (ii) the deflection at the end.
Let us discuss the procedure before we do the calculations. First we must understand
the problem. A few points are raised below.
Advanced Mechanics of Solids 13-26

(i) uu and vv are the principal axes of inertia passing through the centroid C.
(ii) As the line of the load is not parallel to uu nor to vv, this is a case of unsymmetrical
bending.
(iii) We can replace this one problem of unsymmetrical bending by two sub-problems,
each of symmetrical bending. This is done by resolving the force P along uu and vv
(P cos θ and P sin θ, respectively). The bending moments at the fixed end (which
is the vulnerable cross-section because it carries the largest bending moments) are
Mvv = (P cos θ) l, and Muu = (P sin θ) l. As these sub-problems are of symmetrical
bending, the bending stresses can be calculated using the Euler-Bernoulli equation
σ/y = M/I = E/R interpreting M and I correctly. Thus,
Mvv (P cos θ) l Muu (P sin θ) l
= and = .
Ivv Ivv Iuu Iuu
(iv) The vertical and horizontal deflections (δv and δh , respectively) at the end of the
cantilever can be calculated as
P cos θ l3 P sin θ l3
δv = downwards, and δh = to the left.
3 EIvv 3 EIuu
(v) Notice that the load does not pass through the shear centre. We need to locate the
shear centre S by a separate calculation.
(vi) As the load does not pass through the shear centre S, the cross-section is subjected
to a clockwise torque of magnitude P ×e, where e is the eccentricity to be calculated
as in item (v) above.
(vii) Recall the topic of torsion of open cross-sections based on the solution for long
narrow rectangular sections.
(viii) Calculate the torsional shear stresses on the cross-section, and the torsional rigidity
of the channel section. As the thicknesses of the legs are different in general (the
two flanges are identical), the torsional rigidity of each leg is to be calculated, and
added up.
(ix) The total stresses can be found by superposition. The nature of the stress compo-
nents must be understood clearly before they can be ‘added up’ (superposed).
(x) The critical section (the most vulnerable) is at the support. There are uncertainties
at the fixed support. Besides and more importantly, the warping of the cross-section
is prevented at the fixed end. Here the theory of torsion is to be modified14 and
applied. For the limited purpose of working out this illustrative example at this
level, we need not be concerned above these refinements. These must be regarded
as limitations of the method used.
14
In the early days of Timoshenko’s teaching career, he used to spend his vacation in Germany. On one
such occasion when he was in Göttingen, he requested Prandtl (who had by then left this area and turned
to fluid mechanics and his famous boundary layer theory) to suggest a research topic. Prandtl suggested
the topic of torsion when warping is thus restrained. A little later when someone asked Prandtl what he
thought of young Timoshenko, Prandtl is said to have remarked that “he must be good, because after
getting the problem, he never came to me.”
 13-27 A Few Illustrative Examples

To carry out the subsequent calculations, let us take the following numerical values:
width, b = 90; depth, d = 160; thickness, t = 3 (uniform throughout) (all in mm); end
load, P = 1.0 kN; span, l = 1000 mm.
The calculations are simple and straightforward. Nevertheless, we shall identify the
various steps.
Steps
1. Find locate the centroid of the cross-section.
2. Next calculate the second moments of area Ixx and Iyy about the centroidal axes
(horizontal x, and vertical y).
3. Resolve the load P along the horizontal and the vertical directions, Px and Py , re-
spectively.
Py l 3 Px l 3
4. Calculate the end deflections δvert = and δhoriz = .
3EIxx 3EIyy
5. Calculate the twist at the end. After calculating the twist θ per unit length, this is
to be multiplied by the length, viz., l. The theory of a long narrow rectangle is to
be used here as this (channel) is an open section. The length is the sum of the three
legs; here it is 2b + d. The twisting moment is equal to the load P multiplied by the
distance between the shear centre and the line of application of the load P . (When
the load does not pass through the shear centre, a twisting moment is called into play.
This is why the cross-section twists.)
6. Finally we should recognise that there will be some error, but that this will be small.
The error arises because we did not use Timoshenko’s method when the end is fixed.
The warping is prevented at the fixed end, but we disregarded this in our solution. In
any case, there is some approximation in using the theory of a long narrow section.
The distance e, given by e = 3b2 /(6b + d) = 34.71 mm [Eq. (13.20) as calculated in an
earlier example. The remaining part is left as an exercise for the young readers.
29. Example: A Given Channel Section

Work out the above problem for a beam with the channel section shown. The load is
P = 1 kN, and the span l = 1.5 m. E = 200 GPa; θ = 30 ◦ .
[The one difference is that we can, if we desire, include the self-weight also. This is a
uniformly distributed load. Thus, the vertical deflection will be increased by the amount
wl4 /(8EIxx ). The calculation of the torque is a little tricky; it changes from section to
section along the length of the cantilever. However, if we take a specific case like, say,
the channel section ISJC100, we can see that the self weight is only 56.90 N per metre.
This is negligibly small. Hence the self-weight is neglected.]
Preliminaries
Location of the centroid: We can break the cross-section into three rectangles A1 , A2 , A3
and locate the centroid as shown below.
A1 = 90 × 3 = 270 mm4 ; A2 = 154 × 3 = 462 mm4 ; A3 = 90 × 3 = 270 mm4 ;
 Advanced Mechanics of Solids 13-28

A1 x1 + A2 x2 + A3 x3 270 × 45 + 462 × 1.5 + 270 × 45

x̄ = = = 24.94 mm;
A1 + A2 + A3 270 + 462 + 270
90 × 33
 
ȳ = 80 mm; Ixx = + 270 × (80 − 1.5 ) + · · · = 4.241 × 106 mm4 ;
2
12
Iyy = · · · = 0.8357 mm4 .
Calculations
Vertical deflection, δv
Py l 3 1000 × cos 30 ◦ × 15003
δv = =
3 EIxx 3 × 200 × 109 × 4.241 × 106
= 1.147 mm.
Horizontal deflection, δh
Px l 3 1000 × sin 30 ◦ × 15003
δh = =
3 EIyy 3 × 200 × 109 × 0.8357 × 106
= 3.365 mm.
Rotation, θ
This is the twist because of the twisting moment. We note that the load P does not pass
through the shear centre. The twisting moment, T = P × 34.71 × cos 30 ◦ = 30059.7 N
mm. As we had explained earlier, this is an open section and, therefore, is a case of a
long narrow rectangle in torsion. The torsional rigidity, we recall, is given by
T Gt3 l
= ,
θ 3
where this l is the length of the open section = (90 − 1.5) + (160 − 3) + (90 − 1.5) = 334
mm, and t = 3 mm (thickness of the ‘long, narrow rectangle’ in torsion).
3T
θ= = [2.864 × 10−3 ] ◦ /mm.
Gt3 l
This is the twist per unit length. Hence the total twist of the end section = 2.864×1.5 =
4.296 ◦ . Fig. 13.8b shows what these deflections and twist.
30. Example: Z for a Rectangular Cross-section
Calculate Z for a rectangular cross-section.
Fig. 13.9a shows a rectangular cross-section. The relevant dimensions are all marked in
the figure. The property Z is defined as
Z d/2
1 d/2 y
Z Z
1 y 1 y
Z=− dA = − b dy = − dy.
A A R+y bd −d/2 R + y d −d/2 R + y
The fraction 1/(R + y) may be written and expanded in an infinite series using the
binomial theorem, and the expression for Z may be obtained by direct integration.
1 1  y −1
= (R + y)−1 = 1+
R+y R R
13-29 A Few Illustrative Examples

(a) A rectangular cross-section (b) A circular cross-section

Figure 13.9: Two cross-sections, (rectangular and circular) with the relevant dimensions are
shown. We wish to calculate Z for these two cross-sections.

As |y/R| < 1, we may use the binomial expansion. Then by direction integration and
substituting the limits, we can work out the expression for Z as shown below.
1 d/2  y   y 2  y 3  y 4
Z  
Z=− − + − + · · · dy
d −d/2 R R R R
d 2 1 d 4 1 d 6 1 d 8
       
1
= + + + + ···
3 2R 5 2R 7 2R 9 2R
This infinite series converges fast; there is no difficulty in finding the answer with suffi-
cient accuracy.
To get a feel for numbers, it is desirable to work out the numerical value of Z if, for
example, b = 10, d = 14, and R = 17, all in mm. The value of Z is
1 14 2 1 14 4 1 14 6 1 14 8
       
Z= + + + + · · · = 0.0630.
3 34 5 34 7 34 9 34
Note that this series converges fast. More importantly, note that Z is dimensionless; its
numerical value is the same in all units.
An alternative method:
An alternative method is to integrate directly without the infinite series expansion. This
gives us the value of Z as
R+y−R
Z Z
1 y 1
Z=− dA = − dA
bd A R + y bd A R + y
1 d/2
Z  
R 1 d/2
=− 1− dy = − [y − R ln(R + y)]−d/2
d −d/2 R+y d
!
d  
R R+ 2 R 2R + d
= −1 + log = −1 + ln .
d R − d2 d 2R − d
For the numerical values of b = 10, d = 14, and R = 17, all in mm,
 
17 17 + 7
Z = −1 + ln = 0.0630.
14 17 − 7
 Advanced Mechanics of Solids 13-30

31. Example: Z for a Circular Cross-section

The circular cross-section and the relevant dimensions and markings are shown in Fig.
13.9b. The diameter of the circular cross-section is d, and the radius of curvature is R.
Then,
Z
1 y
Z=− dA
A A R+y
Z π/2 d  2
1 2 sin θ d
= − d2 d
2 cos θ dθ
π 4 −π/2 R + 2 sin θ 2
d π/2 cos2 θ sin θ
Z
=− dθ.
π −π/2 R + d2 sin θ

As before, let us expand

 −1  −1
1 d d
d
= R + sin θ =R 1+ sin θ
R+ 2 sin θ 2 2R

in an infinite series. The result is that the expression for Z is obtained as

d 2 1 d 4 d 6 d 8
       
1 5 7
Z= + + + + ···
4 2R 8 2R 64 2 R 128 2 R
For the numerical values of d = 14, and R = 17, all in mm,

1 14 2 1 14 4 5 14 6
       8
7 14
Z= + + + + · · · = 0.04640.
4 34 8 34 64 34 128 34

32. Example: Stresses in a Curved Beam

(a) A curved beam with a load W (b) Cross-section and stresses

Figure 13.10: A vertical load W acts as shown. The dimensions of the cross-section and
the stress distributions are shown alongside.

A curved beam in the shape of a crane hook with the relevant dimensions is shown in
Figs 13.10. Calculate the maximum load so that the stress does not exceed MPa.
 13-31 A Few Illustrative Examples

The critical cross-section is X −X. There are (i) bending stresses and (ii) direct stresses.
The bending stresses are given by the Winkler-Bach formula. These (circumferential)
stresses are distributed hyperbolically: largest at the inner point A (tensile). At B
the compressive bending stress is a maximum. The direct stresses on the cross-section
is tensile, and is (assumed to be) uniform, [W/(area of c.s.)]. The total stresses are
obtained by adding up the stresses due to (i) bending, and (ii) direct tension. Thus, on
the most vulnerable cross-section X − X, the maximum stress (tensile) is at A.
Now we can make the detailed calculations. The value of Z for this case was already
worked out earlier; it is Z = 0.0630. The area of the cross-section A = b × d = 140 mm2 .
W = 10 kN. The bending moment = −W R = −170 Nm. The distance of the point A
from the centroidal axis is yA = 14/2 = 7 mm = 7 × 10−3 m.
Total stresses at A and B on the cross-section are
 
W M 1 −yA
σA = + 1+ ;
A AR Z R + (−yA )
  
10 × 1000 170 1 −7
= − 1+
140 × 10−6 140 × 10−6 × 17 × 10−3 0.0630 17 − 7
= 71.428 + 722.22 = 793.65 MPa.

 
W M 1 yB
σA = + 1+ ;
A AR Z R + yB
  
10 × 1000 170 1 7
= − 1+
140 × 10−6 140 × 10−6 × 17 × 10−3 0.0630 17 + 7
= 71.428 − 402.116 = −330.69 MPa.

Actually it is not necessary to calculate the stress at B, σB . This is done merely to

clarify matters.

33. Example: Equations of Equilibrium by Coordinate Transformation

Starting from the differential equations of equilibrium in rectangular coordinates, obtain
the corresponding equations in polar coordinates.
The two systems of coordinates (x, y) and (r, θ) are related by
p
x = r cos θ; r = x2 + y 2 ;
y
y = r sin θ; θ = tan−1 .
x
The derivatives are calculated as
∂r x ∂θ y sin θ
= = cos θ; =− 2 =− ;
∂x r ∂x r r
∂r x ∂θ x cos θ
= = sin θ; = 2 = .
∂y r ∂y r r
Advanced Mechanics of Solids 13-32

From the stress transformation laws we obtain the two following sets of results.

(a) σrr = σxx cos2 θ + σyy sin2 θ + τxy sin 2θ;

σθθ = σxx sin2 θ + σyy cos2 θ − τxy sin 2θ;
τrθ = (σyy − σxx ) sin θ cos θ + τxy cos 2θ.

(b) σxx = σrr cos2 θ + σθθ sin2 θ − τrθ sin 2θ;

σyy = σrr sin2 θ + σθθ cos2 θ + τrθ sin 2θ;
τxy = (σrr − σθθ ) sin θ cos θ + τrθ cos 2θ.

Let us now compute the three terms in the differential equations of equilibrium [Eqs
5.6a, 5.6b; p. 5-7].
∂σxx ∂ 
σrr cos2 θ + σθθ sin2 θ − τrθ sin 2θ −

= cos θ
∂x ∂r
sin θ ∂ 
σrr cos2 θ + σθθ sin2 θ − τrθ sin 2θ ;

r ∂θ
∂σyy ∂ 
σrr sin2 θ + σθθ cos2 θ + τrθ sin 2θ +

= sin θ
∂y ∂r
cos θ ∂ 
σrr sin2 θ + σθθ cos2 θ + τrθ sin 2θ ;

r ∂θ
∂σxy
=··· ;
∂x
∂σxy
=··· .
∂y
When these expressions are substituted in the two equations of equilibrium [Eqs 5.6a,
5.6b; p. 5-7], we obtain
   
cos θ ∂τrθ ∂σrr ∂τrθ sin θ ∂σθθ
σrr − σθθ + + cos θ − sin θ − + 2τrθ = 0;
r ∂θ ∂r ∂r r ∂θ
   
sin θ ∂τrθ ∂σrr ∂τrθ cos θ ∂σθθ
σrr − σθθ + + sin θ + cos θ + + 2τrθ = 0.
r ∂θ ∂r ∂r r ∂θ
These two equations are simplified, and we arrive at the desired form as
∂σrr 1 ∂τrθ σrr − σθθ
+ + = 0;
∂r r ∂θ r
1 ∂σθθ ∂τrθ 2 τrθ
+ + = 0.
r ∂θ ∂r r
If the body forces are also to be included, the equations are
∂σrr 1 ∂τrθ σrr − σθθ
+ + + Fr = 0;
∂r r ∂θ r
1 ∂σθθ ∂τrθ 2 τrθ
+ + + Fθ = 0.
r ∂θ ∂r r
 13-33 A Few Illustrative Examples

34. Example: Airy’s Stress Function

Examine what problems are solved by the following Airy’s stress functions.

(i) φ = C1 ;
(ii) φ = C1 x 2 ;
(iii) φ = C2 x2 + C3 xy + C1 y 2
(iv) φ = C2 x2 + C3 x2 y + C1 xy 2 + C4 y 3

C1 , C2 , C3 C4 are all constants.

First of all, let us check and make sure that the governing equation

∂4φ ∂4φ ∂4φ

54 φ ≡ + 2 + =0
∂x4 ∂x2 ∂y 2 ∂y 4

(which is the compatibility equation) is satisfied by all these four functions φ = φ(x, y).
We shall now discuss these four cases one by one.
(i) φ = C1 . This case is trivial; all the stress components are zero everywhere.
(ii)φ = C1 x2 .

∂2φ ∂2φ ∂2φ

σxx = = 0; σyy = = 2C1 ; τxx = − = 0.
∂y 2 ∂x2 ∂x∂y

This can represent a homogeneous (uniform) state of stress (tensile if C1 > 0, compressive
if C1 < 0) in a bar or a plate [Fig. 13.11a].
(iii) φ = C2 x2 + C3 xy + C1 y 2 .

∂2φ ∂2φ ∂2φ

σxx = = 2C1 ; σyy = = 2C2 ; τxx = − = −C3 .
∂y 2 ∂x2 ∂x∂y

(a) A plate loaded by σyy = 2C1 (b) A plate loaded as shown

Figure 13.11: The figures are drawn for the numerical values of C1 = 10, C2 = 15, C3 = 5,
all in MPa. Note particularly that the shear stress τxy = −5 is shown as marked. The
negative sign is already reckoned by reversing the direction; the magnitude, therefore, is to
be marked as 5 (and not as −5).
 Advanced Mechanics of Solids 13-34

This can represent a loading as shown in Fig. 13.11b.

(iv) φ = C2 x3 + C3 x2 y + C1 xy 2 + C4 y 3 .

∂2φ ∂2φ
σxx = = 2C 1 x + 6C 4 y; σ yy = = 6C2 x + 2C3 y;
∂y 2 ∂x2
∂2φ
τxy = − = −2C3 x − 2C1 y.
∂x∂y
All the stress components have linear variations w.r.to x and y. If C1 = C2 = C3 = 0
and C4 6= 0, that is, if φ = C4 y 3 , this stress function solves the problem of bending of a
beam. We have seen this earlier.

35. Example: Mohr’s Circle

Given that σ11 = 50, σ22 = 30, σ33 = −10 (all in MPa), let us use the Mohr’s circle
construction to find out the normal stress σ(ν) and the shear stress τ(ν) on an octahedral
plane — equally √ inclined to the three principal planes — with the direction cosines
l = m = n = 1/ 3. We shall also calculate these analytically for comparison.
Graphical construction:
Following the steps given on p. 5-40, let us proceed as follows [Fig. 13.12]. All numerical
values of stresses are in MPa.

(i) Draw to a convenient scale the σ(ν) τ(ν) axes, and mark off the principal stresses
(σ11 = 50, point A), (σ22 = 30, point B), (σ33 = −10, point C, 10 units to the left
of the origin O).
(ii) Draw three semi-circles aaaa, bbbb, cccc with the centres and radii as follows.
σ11 + σ33 σ − σ 
 11 33 
circle aaaa: centre D: = 20 radius   = 30;
2 2
σ22 + σ11  σ22 − σ11 
 
circle bbbb: centre E: = 40 radius   = 10;
2 2
σ33 + σ22  σ33 − σ22 
 
circle cccc: centre F: = 10 radius   = 20.
2 2
√ √ √
(iii) The angles (ν, x), (ν, z) corresponding
√ to the direction cosines 1/ 3, 1/ 3, 1/ 3
−1
are calculated as cos (1/ 3) ≡ (ν, x) = 54.74 . ◦

(iv) Draw a straight line CG at this calculated angle of (ν, x) = 54.74 ◦ , and the straight
line AG to meet the circle aaaa at G. This line cuts the circle bbbb at the point
H [Fig. 5.30b].
(v) With F as the centre, draw a circular arc through G. This will necessarily pass
through E.
(vi) In the same way draw a straight line AI at the calculated angle of (ν, z) = 54.74 ◦ ,
and the straight line CI to meet the circle aaaa at J. This line cuts the circle cccc
at the point J.
(vii) With E as the centre, draw a circular arc through I. This will necessarily pass
through J.
13-35 A Few Illustrative Examples

(viii) These two circular arcs intersect at the point P . The coordinates of P give us the
sought after values of the normal σ(ν) and shearing τ(ν) stresses. These values are
obtained by measuring the coordinates. The values thus obtained [σ(ν) = 23.33
MPa, τ(ν) = 24.94 MPa] can be compared with the values obtained by calculations
shown below. It will be seen that the values match if due allowance is given for
unavoidable errors in any graphical construction15 .

(a) Mohr’s circles: step 1 (b) Mohr’s circles: step 2

Figure 13.12: The various steps to solve this numerical problem by drawing the Mohr’s
circles are explained in the text with reference to this figure.

Answers using analytical / mathematical equations:

(ν)
The resultant stress  T , the normal stress σoct , and the shear stress τoct , are given by
 
the following expressions. We obtain the answers by substituting the numerical values.
r r
(ν) 2 + σ2 + σ2
σ11 502 + 302 + 102
22 11
|T | = = = 34.16 MPa;
3 3
σ11 + σ22 + σ33
σoct =
3
50 + 30 − 10
= = 23.33 MPa;
3
1p
τoct = (σ11 − σ22 )2 + (σ22 − σ33 )2 + (σ33 − σ11 )2
3
1 2
= (20 + 402 + 602 ) = 24.94 MPa.
3
(ν)
2 2
verify: σoct + τoct = | T |2 −→ 544.44 + 622.21 ≈ 1166.67.

15
This is usually the case, but here the values obtained are remarkably and unbelievably the same as the
ones obtained by analytical methods!
 Advanced Mechanics of Solids 13-36

36. Example: Strains from Displacements

The displacement field in a body is given as u = u i +v j +w k = (2x+y 2 ) i +(x3 +2z 2 ) j +
(xy 2 −xz) k , find (i) the initial and final position vectors and (ii) the displacement vector
u and the strain components at the point P (1, 2, 3). All are in 10−3 mm.

Initial position vector r = 1 i + 2 j + 3 k

Displacement vector u at P (1, 2, 3) = 1 i + 2 j + 3 k
= (2 × 1 + 22 ) i + (13 + 2 × 32 ) j + (1 × 22 − 1 × 3) k
= 6 i + 19 j − k
Final position vector r ∗ = r + u = (1 i + 2 j + 3 k ) + (6 i + 19 j − k )
= 7 i + 21 j + 2 k

∂u ∂v ∂w
exx = = 2; eyy = = 0; ezz = = −x = −1;
∂x ∂y ∂z
 
1 ∂u ∂v 1 1
exy = + = (2y + 3x2 ) = (2 × 2 + 3 × 12 ) = 3.5;
2 ∂y ∂x 2 2
 
1 ∂v ∂w 1 1
eyz = + = (4z + 2xy) = (4 × 3 + 2 × 1 × 2) = 8;
2 ∂z ∂y 2 2
 
1 ∂w ∂u 1
ezx = + = (2y − z + 0) = 0.5, all in 10−3 .
2 ∂x ∂z 2

37. Example: Theories of Failure

Determine the diameter of a steel shaft on the basis of (i) the maximum shear stress
theory, and (ii) the distortion energy theory if the shaft is subjected to a combined load
of a torque T = 20, 000 Nm and a bending moment M = 10, 000 Nm. Assume that
the yield stress, σy.p. = 400 MPa (obtained from a simple tension test), and a factor of
safety of 1.8.
(a) Preliminaries

π d4
Second moment of area (for bending), I = ;
64
π d4
Polar second moment of area (for bending), J = .
32
As there are only the normal stress σ due to bending, and the shear stress τ due to
twisting, the principal stresses — this is a two-dimensional case — and the maximum
shear stress are:
r 
σ σ 2
σ1 , σ 2 = ± + τ 2;
2 2
r 
1 σ 2 1p 2
τmax = (σ1 − σ2 ) = + τ2 = σ + 4τ 2 .
2 2 2
 13-37 A Few Illustrative Examples

M ( d2 ) M d 32 M
normal stress in bending, σ = = πd4 =
I 2 πd3
64
T ( d2 ) T d 16 T
shear stress in twisting, τ = = πd4
= .
J 2 πd3
32

(b) Maximum shear stress theory √

Maximum shear stress, τmax = (1/2) σ 2 + 4τ 2 . The maximum shear at the point of
yielding in a simple tension test = (1/2)σy.p. . Thus, the design equation is
s
32 M 2 16 T 2 1 σy.p.
  
1
+ 4 = ,
2 πd3 πd3 2 N

where N is the ‘factor of safety’16 . Thus,

s
32 M 2 16 T 2 1 σy.p.
  
1 16 p 2 1 σy.p.
3
+ 4 3
= ; i.e., 3
M + T2 = ;
2 πd πd 2 1.8 πd 2 1.8
160000 × 1000 p 2 1 (400)
3
1 + 22 = ; d = 100.8 mm.
πd 2 1.8
We shall see that this theory is a little more conservative than the distortion energy
theory which we shall see below.
(c) Distortion energy theory
The design equation is
r
1 σy.p.
[(σ1 − σ2 )2 + (σ2 − σ3 )2 + (σ3 − σ1 )2 ] = ;
2 N
p σy.p. 16 p σy.p.
i.e., σ 2 + 3τ 2 = ; i.e., 3
4M 2 + 3T 2 = ;
N πd N
160000 × 1000 p 400
3
4(1)2 + 3(2)2 = ; d = 97.1 mm.
πd 1.8
The remark that we made above (that the maximum shear stress theory is a little more
conservative than the distortion energy theory) is vindicated.

38. Example: Maxwell Reciprocal Theorem

The displacement at a point i because of a unit force applied at a point j (another point
or even the same point) is equal to the displacement at the point j because of a unit
force applied at the point i. It is emphasised that (i) the displacement may be linear or
angular (rotational), and that the force may be a force or a moment; and that (ii) the
displacement and the force correspond to each other reckoned at each point in the same
direction — work absorbing component!
16
This is called a factory of safety because, as Den Hartog [3] remarks, ‘factor of ignorance’ sounds too
cynical.
 Advanced Mechanics of Solids 13-38

(a) A load applied at 1 (b) A force and a moment applied at 2

Figure 13.13: Fig. 13.13 shows a cantilever. The loadings (force / moment) and the
displacements (deflection / slope) are shown. We can see verify the Maxwell reciprocal
theorem.

(i) When the forces are applied in the order F 1 , F 2 — first F 1 and then F 2 — the work
done W (and, therefore, the strain energy stored U ) is

1 1
F 1 + (α22 F 2 )F
W1 = U1 = (α11 F 1 )F F 2 + α12 F 1 F 2 .
2 2
(ii) When the order of application of the forces is reversed — first F 2 and then F 1 —
the work done W (and, therefore, the strain energy stored U ) is

1 1
F 2 + (α11 F 1 )F
W2 = U2 = (α22 F 2 )F F 1 + α21 F 1 F 2 .
2 2
As W1 (= U1 ) and W2 (= U2 ) are equal — the total work done is independent of the order
of application of the forces — we conclude that

W1 = W2 α12 = α21 .

This symmetrical relationship can also be stated differently17 which is sometimes more
convenient to apply.

39. Example: An Application of Betti-Maxwell Reciprocal Theorem

An elastic body (E, ν) is acted upon by a pair of forces F − F acting at the points A and
B. If d is the distance between A and B, calculate the change in volume of the body.
At first sight this would appear to be a very difficult problem. However, there is a trick.
We can apply the Betti-Maxwell theorem and obtain the solution very easily.
How do we do that? Well, let us subject this body to a hydrostatic state of stress as
shown in Fig. 13.14b. Now we have two systems: (a) the body with the two forces
17
Fung [6] refers to the symmetrical relationship αij = αji as Maxwell’s theorem, and the different form
given below as the Betti-Maxwell reciprocal theorem.
 13-39 A Few Illustrative Examples

(a) A body with two forces F − F (b) A block in a hydrostatic state of stress

Figure 13.14: Fig. 13.14a shows an elastic body acted upon by a pair of forces F − F acting
at the points A and B separated by a distance d. Fig. 13.14b shows a block acted upon by
a hydrostatic state of stress represented by σxx = σyy = σzz = −p.

F − F , and the desired change in volume ∆V ; and (b) the body in a state of hydrostatic
stress, and the change in the distance ∆d between A and B.
p
The strain in the horizontal direction in system, exx = (1 − 2ν). The points A and B
E
p
move towards each other by δd = d (1 − 2ν).
E
Now we apply the Betti-Maxwell reciprocal theorem: the work done by the force system
in (a) on the displacements in (b)= the work done by the force system in (b) on the
displacements in (a). Thus, we have
F Fd
F × ∆d = σ × ∆V ; −→ ∆V = × ∆d = (1 − 2ν).
σ E
If ν = 0.5, the change in volume is zero. This is understandable; because if the body
is incompressible — ν = 0.5 corresponds to incompressibility — there cannot be any
change in the volume.
We can appreciate the power of this theorem. If we are a little innovative, this theorem
helps us to solve problems that may appear to be difficult.
40. Example: Castigliano’s First and Second Theorems
Find the end deflection of a uniform cantilever subjected to and end load P using Cas-
tigliano’s first and second theorems18 .
The strain energy U is given by
Z l Z l
M2 (P x)2 P 2 l3
U= dx = dx = .
0 2 EI 0 2 EI 6 EI
U is now expressed in terms of the load P . Thus, Castigliano’s second theorem can be
applied readily as
dU P l3
end deflection, δ = = .
dP 3 EI
18
Den Hartog [4] calls them the theorem of work or virtual-work theorem and Castigliano’s theorem. He
mentions only one Castigliano’s theorem, which is what we call Castigliano’s second theorem.
 Advanced Mechanics of Solids 13-40

To apply Castigliano’s first theorem, we must express the strain energy U in terms of δ.
Thus, writing P = kδ,

(kδ)2 l3 dU k 2 δl3 P l3
U= ; −→ = =P −→ δ= .
6 EI dδ 3 EI 3 EI

41. Example: Castigliano’s First Theorem

(a) To find the deflection at C (b) Nonlinear case: to find the deflection

Figure 13.15: We desire to calculate the end deflection of a stepped cantilever [Fig. 13.15a].
Fig. 13.15b shows U and U ∗ for a nonlinear material. We shall see that Castigliano’s first
theorem will give a wrong result, but if it is applied to U ∗ we would get the correct result.

We are required to find the deflection at the end C of the stepped cantilever shown.
The strain energy U (neglecting the shear effects which is justified because this is a
slender beam) is given by

1 l M2
Z b Z l
(W x)2 (W x)2 W 2 b3 W2 3
Z
U= dx = dx + dx = + (l − b3 ).
2 0 EI 0 2 EI2 b 2 EI1 6 EI 2 6 EI1

Using Castigliano’s (first) theorem, the deflection at C is calculated as

1 l M2
Z b Z l
(W x)2 (W x)2 W b3
Z
dU W
δC = = dx = dx + dx = + (l3 − b3 ).
dW 2 0 EI 0 2 EI2 b 2 EI1 3 EI2 3 EI1

42. Example: Castigliano’s Theorems

Fig. 13.15b shows the load-extension diagram of a nonlinear spring given by P = αx2 .
The areas below and above the curve are, as we already know, U (strain energy) and U ∗
(complementary energy). Let us note clearly that U = U (x), while U ∗ = U ∗ (P ), and
that now U and U ∗ are not even numerically equal19 .
The strain energy, U and the complementary energy, U ∗ are given by
Z x Z x
kx3
U = U (x) = P dx = kξ 2 dξ = ;
0 0 3
19
We had emphasised that even in the linear case, when U and U ∗ are numerically equal, it is not proper
to write U = U ∗ , because U = U (x), while U ∗ = U ∗ (P ). Some authors write U (x) and U (P ), instead of
U (x) = U [x(P )] = U1 (P ). Note carefully that the functional forms U and U1 are different!
 13-41 A Few Illustrative Examples

Z P Z P  1  
∗ ∗ P 2 1 2 3
U = U (P ) = x dP = dP = 1 P2 −1 ;
0 0 k k2 3
dU ∗ 1 1
= 1 P 2 = x.
dP k2
Expressing U as a function of P , we obtain
3
kx3

k P 2
U = U (x) = U [x(P )] = U1 (P ) = = ,
3 3 k
from which we can compute dU1 /dP , but it does not give the correct answer.
Note the difference between the linear and the nonlinear cases.
Linear case:
Let us consider a linear spring specified by force-displacement relationship P = kx. Now
the strain energy, U and the complimentary energy, U ∗ are
1 1 P2
P = kx; U = U (x) = kx2 ; U ∗ = U ∗ (P ) = k 2 ;
2 2 k
dU dU ∗ P
= kx = P ; = = x;
dx dP k
dU dU ∗
dU = P dx −→ = P; dU ∗ = x dP −→ = x.
dx dP
43. Example: Timoshenko’s Method of Trigonometric Series
We shall now illustrate Timoshenko’s method of trigonometric series of obtaining the
deflection of beams. Let us, for simplicity, consider a uniform cantilever of length l
loaded by a concentrated end load P . We desire to obtain the deflection curve y = y(x).
Assumed deflection curve:
Let us assume the deflection curve in the form of a trigonometric series
y = y(x) = c1 φ1 (x) + c2 φ2 (x) + c3 φ3 (x) + · · · + cn φn (x) + · · · ,
where each of the coordinate functions φi (x) satisfies the (prescribed) boundary condi-
tions
deflection at the fixed end: y(0) = 0;
slope at the fixed end: y 0 (0) = 0.
Let us choose the coordinate functions as
πx 3πx (2n − 1)πx
φ1 = 1 − cos ; φ2 = 1 − cos ;··· ; φn = 1 − cos ;··· .
2l 2l 2l
Boundary conditions:
We can see that the prescribed boundary conditions are satisfied for each of these func-
tions. Furthermore, each one of them satisfies also the condition of zero bending moment
M = EI y 00 at the free end.
φn (0) = 0; φ0n (x) = 0; φ00n (l) = 0.
Advanced Mechanics of Solids 13-42

Thus, we can see that these are excellent choices, and we may anticipate that the result
also will be excellent. Accordingly, the deflection curve is assumed in the form
   
h πx i 3πx (2n − 1)πx
y = y(x) = c1 1 − cos + c2 1 − cos + · · · + cn 1 − cos + ··· ,
2l 2l 2l
(13.21)
where the ci ’s are the unknown coefficients to be determined.
Procedure:
Timoshenko’s method of determining these ci ’s is the following. When any one coeffi-
cient, say, cn is varied — all the other coefficients are kept unchanged — there will be a
corresponding change in the end deflection. Thus, there is some external work done by
the end force P . Correspondingly, there will be some change in the strain energy too.
Equating these two, one obtains the value of this coefficient cn . We shall now carry out
this procedure.
Calculations:
The approximate expression y 00 for the curvature is obtained from Eq. (13.21) as
X  nπ 2 nπx
y 00 = cn cos .
2l 2l
1,2,3,···

The strain energy U is, therefore, given by

Z l
EI 00 2
U= (y ) dx
0 2
 2
Z l X
EI  nπ  2 nπx 
=  cn cos dx
2 0 2l 2l
1,2,3,···
Z l
EI X  nπ 4 nπx
= c2n cos2 dx
2 0 2l 2l
1,2,3,···

X n4 π 4
= EI c2n , (13.22)
64l3
1,2,3,···

where the orthogonality of the cosine functions

Z l (
mπx nπx 0 if m 6= n
cos cos = l
0 2l 2l 2 if m = n,

is exploited and a few steps of simplification are omitted.

We shall now take Eq. (13.22), and change only one coefficient cn to cn + dcn , while
all the other coefficients are left unchanged. Then the strain energy U will change to
U + dU , and the deflection y to y + dy. The end deflection will change slightly, and the
external force P will do some work. This work must be equal to the change dU in U .

cn → cn + dcn −→ U → U + dU −→ y → y + dy.
 13-43 A Few Illustrative Examples

The change dy in the deflection y at the end of the cantilever x = l is

 nπx  
dy = dcn 1 − cos = dcn
2l

x=l

Equating dU and the external work P dcn , we have

∂U n4 π 4
P dcn = dU −→ P = = 2cn EI.
∂cn 64l3
Thus, we obtain
P l3 32
cn = .
EI π 4 n4
Such a result holds obviously for every coefficient. We, therefore, obtain the deflection
of the cantilever as
32 P l3 X 1  nπx 
y= 4 1 − cos .
π EI n4 2l
1,3,5,···

This rapidly converging infinite series gives the exact solution. The end deflection is of
special interest. It can be calculated (if three terms are taken for the calculation) as

32 P l3 P l3
 
 1 1
δ = y = 4 1 + 4 + 4 + ··· = .

x=l π EI 3 5 3.001 EI

44. Example: Curved Beam - Castigliano’s Theorem

(a) A curved beam: a vertical load at the end (b) Free-body diagram: part of the beam

Figure 13.16: A curved beam BF [Fig. 13.16a] — a quarter of a circle — is loaded by a

vertical end load V . It is desired to find the vertical deflection δv at the free end F . The
free-body diagram of a part of the beam is shown [Fig. 13.16b].

We shall see the use of Castigliano’s theorem to calculate the deflection of beams. The
curved beam of uniform cross-section shown [Fig. 13.16a] is a quarter of a circle of radius
R built in at the end B, and free at the end F . It is loaded by a vertical load V at the
end F . We desire to find the vertical deflection δv at the end of the curved cantilever.
Advanced Mechanics of Solids 13-44

Vertical Deflection Due to a Vertical Load

dU
Castigliano’s theorem gives the deflection by the formula δv = , where U is the strain
dV
energy of the beam expressed in terms of the load V in good shape to carry out the
indicated differentiation. Thus, we need to express U as a function of V in the form
U = U (V ). We know from our earlier study of the Mechanics of Solids that the strain
energy stored in a bent beam is given by
Z L
M2
U= ds,
0 2EI

with the usual notation. Once the bending moment is calculated in terms of the applied
load V , the exercise reduces to some elementary mathematical manipulations. We shall
now carry out the calculation of the bending moment as a function of the angle θ.
Referring to the free-body diagram [Fig. 13.16b], we find that the bending moment M
at a typical section X − X at an angular distance θ from the horizontal line is20

M = V (R − R cos θ) = V R(1 − cos θ).

For this problem, we proceed as follows.

Z L Z π
M2 2 M2
U= ds = R dθ (13.23)
0 2EI 0 2EI
Z π Z π  2
dU d 2
2
2 ∂ M
δv = = f racM 2EI R dθ = R dθ (13.24)
dV dV 0 0 ∂V 2EI
Z π Z π
1 2 ∂M 1 2
= M R dθ = [V R(1 − cos θ)][R(1 − cos θ)] R dθ (13.25)
EI 0 ∂V EI 0
π V R3
= .
2 EI
We have obtained the required answer. However, a few remarks seem to be in order. In
the first place, we need to form a habit of checking the dimensions. Let us make sure that
the final answer obtained does, indeed, have the proper dimensions (m, because this is the
expression for the deflection). In addition, let us note that in Eq. (13.24), two operations
are involved, integration (with respect to θ) first, and then differentiation (with respect
to V ). The order of these operations can be interchanged as first differentiation, and
then integration. This is mathematically justified. Such an interchange is convenient,
because we will have simpler expressions for integration. The integrand in Eq. (13.24) is
a function not merely of V , but also of other variables. Thus, it is necessary to write as
a partial derivative as indicated in Eq. (13.25). We may also clarify that the integration
in Eq. (13.23) is along the full length 0 − L of the beam.
20
The sign of M , positive or negative, depends on the sign convention employed. Here this is of no con-
sequence, because we need the expression of only its square (M 2 ) in the above integrand. Even so, it is
better to stick to one’s own convention as a matter of good habit. Yet, it is also desirable to spurn this
once in a way, to emphasise or remind ourselves the fact that it is, after all, only a convention and not a
binding command.
13-45 A Few Illustrative Examples

This exercise is over at this stage. But we can discuss some more aspects regarding such
problems. The following points are worth noting.
(i) Although this is a ‘curved’ beam, we have not used the curved beam theory (the
so-called Winkler-Bach equation) here. The reason is that the radius of curvature
is very large compared to the depth of the beam. Thus, even though this is a curved
beam for external appearances, this is essentially a straight beam; curvature effects
need not be taken into account.
(ii) We can notice that on a typical cross-section X −X, there is a normal stress N , and
a shear force, S, in addition to the bending moment. Thus, there are normal (taken
to be uniformly distributed) stresses due to this N , and shear stresses (cross-shear)
due to the shear force F . Associated with these stresses, there are contributions
to the strain energy, U . We did not take them into account.
This may appear to be a lapse on our part. But, no. For such a this beam as we
have here, the contributions of these two components are very small. We may, thus,
disregard them in this example. However, when thick beams are considered, we
must include the contributions from the direct normal stresses and the cross-shear
stresses. We shall consider this more complicated case in a later example.
(iii) If we desire to calculate the horizontal deflection at the free end21 , we need to
pretend that there is an external load, say, H at the free end, calculate the strain
energy in terms of both V and H, and use the Castigliano’s theorem in the form
∂U
δh = where U = U (V, H)
∂H
and, then after the differentiation is carried out, wake up and realise that no such
horizontal force is acting. Thus, substitute H = 0 in the final formula to obtain
the horizontal deflection δh .
We shall illustrate this case below.
Horizontal Deflection Due to a Vertical Load
Shown in Fig. 13.17a is the same curved beam, but this time we desire to calculate the
horizontal deflection, δh , at the free end F . We pretend that a horizontal force H also
is acting, and follow the same procedure as in the previous example. We need to do
this, because Castigliano’s theorem is concerned with the ‘work absorbing’ components
of forces and displacements. The desired deflection is the horizontal deflection, and the
associated ‘work absorbing’ force is a horizontal force H at the free end F .
The horizontal deflection, δh is given by
Z π Z π
2 M 2R
 
∂U ∂ 1 2 ∂M 
δh = = dθ = M R dθ . (13.26)

∂H H=0 ∂H 0 2EI EI 0 ∂H H=0

The bending moment on a typical cross-section X − X is M = V R(1 − cos θ) − HR sin θ.

The derivative of M w.r.to the horizontal load H is ∂M/∂H = −R sin θ. Substituting
21
Some students seem to have difficulty in realising that there will be a horizontal deflection δh at the free
end, even though the applied load is entirely vertical.
 Advanced Mechanics of Solids 13-46

(a) A curved beam: vertical load at the end (b) Free-body diagram: part of the curved beam

Figure 13.17: A curved beam BF [Fig. 13.17a] — a quarter of a circle — is loaded by a

vertical end load V . It is desired to find the horizontal deflection δh at the free end F . We
pretend that a horizontal load H is also acting at the free end. The free-body diagram of
a part of the beam is shown [Fig. 13.17b].

these expressions for the bending moment, M , and the derivative ∂M/∂H in Eq. (13.26),
we obtain the required horizontal deflection as
π
V R3
Z
1 2
δh = [V R(1 − cos θ) − HR sin θ][−R sin θ] R dθ = .
EI 0 4EI

45. Example: A Truss Problem

We shall solve a truss problem using Castigliano’s theorem22 in its simplest form.
Shown in Fig. 13.18a is a simple, planar, three-member truss OACB. Shown alongside
is a similar truss, but without the member OC. Let us calculate the force P to be
applied to obtain a specified vertical deflection δ of the joint O for both these cases. The
dimensions are AC = CB = 600, CO = 800, OA = OB = 1000, all in mm. The joint
O is brought down to the position O0 . If the vertical deflection OO0 is specified as δ, we
are required to calculate the vertical load P to be applied at O. All the bars have the
same diameter of 1.5 mm; Young’s modulus of elasticity (steel), E = 200 GPa. There is
no member CO for the second truss.
For convenience, let us designate the length OA (= OB) as l, the area of cross-section
as A (the same for all the members), the semi-vertical angle COA as α, and the vertical
length OC as h. As long as δ is small, which is almost always the case, there is no
significant difference between (i) the angles COA and CO0 A, and (ii) the vertical lengths
OC and O0 C. Both the problems (a) and (b) are exactly similar as far as the geometry
of deformation is concerned.
22 ∂U (δ)
Castigliano’s theorem: =P
∂δ
Some authors, for example, J.P. Den Hartog [4], refer to this as the theorem of work. Some others, for
example, I.H. Shames & C.L. Dym, Energy and Finite Elements in Structural Mechanics, Wiley Eastern
Limited,(1950) call this Castigliano’s first theorem.
13-47 A Few Illustrative Examples

(a) A three-member planar truss (b) A similar two-member truss

Figure 13.18: Two similar planar trusses. The only difference is that in the truss on the
right hand side, there is no member OC. The members meet at the point O where a
displacement δ is imposed. It is desired to calculate the load P needed for this.

Solution from first principles:

We can surely solve these simple problems from first principles. If the vertical deflection
OO0 = δ, then the new length AO0 can be calculated in terms of δ, h and cos α by solving
the shaded triangle AOO0 . Then we can calculate the forces in the members — three,
viz., OA, OB and OC in (a); and two, viz., OA and OB in (b) — and obtain the vertical
load to be applied in each of the two cases by considering the equilibrium of the joint.
We, however, desire to work out this problem to demonstrate how Castigliano’s (first)
theorem is used in such cases.
Solution using Castigliano’s first theorem:
To be able to use the Castigliano’s first theorem, we need to express the energy U as
a function of δ. Let us, therefore, calculate the new length O0 A. This (O0 A) can be
seen to be (by dropping a perpendicular from O to the side O0 A of one of the shaded
triangles) stretched by the amount δ cos α. Let us observe a couple of relationships from
the geometry.
h
cos α = ; change in length, O0 A − OA = δ cos α
l
The strain in the bar O0 A — the same everywhere in the bar — is

δ hl


O0 A − OA δ cos α δ cos α δh
= 0
= = = h

= 2.
OA l (h/ cos α) h/ l l

The strain energy in the bar O0 A is calculated as shown as

Z Z Z Z  Z  Z
σ d dv = σ d Al = Al E d = E2 Al.
V   

The triple integral need not scare us; the integrand is a constant at all points in V. All
we need to do is to calculate the volume as Al and multiply with the constant integrand.
Advanced Mechanics of Solids 13-48

Case (b): only two members OA and OB OB:

Let us now consider case (b) first where there are only the two members OA and OB.
Now the total strain energy, U is
 2  
2 δh h
U = U (δ) = 2 × E Al = 2 × E A
l2 cos α
2
2EA δ cos α3
= (13.27)
h
To obtain the desired load, we only need to differentiate this expression (13.27) for
U = U (δ) w.r.to δ.
∂U (δ) 2EA cos3 α
P = = δ.
∂δ h
When the given numerical values are substituted, we obtain the desired result as
2EA cos3 α
P = δ
h
2
2(200 × 109 )( π 1.5
4 × 10−6 ) × (0.83 )
= δ
0.8
= 452.4 δ kN; i.e., 0.4524 kN for every mm of vertical deflection.
Being a linear structure, the load P will be proportional to the deflection δ. Now, to
complete the problem, we shall take up the first case (a).
Case (a): three members OA OA, OB and OC OC:
In this case, the vertical load should, additionally, be able to stretch the vertical member
OC also by the same amount, δ. Thus, we may compute the axial tensile force necessary
to stretch the member OC — a uniform bar of length 800 mm — by δ, and add this
amount to the answer obtained for the case (b) above.
AE
additional load = δ
l
π 1.52 (200 × 109 )
= δ
4×8
= 0.4418 kN per mm of vertical deflection.
The answer now is P = 0.4524 + 0.4418 = 0.8942 kN per mm of vertical deflection.
We can include the additional strain energy in the member OC in Eq. (13.27) and obtain
the same answer using Castigliano’s first theorem.
2EA δ 2 cos3 α
   
AE 2
total strain energy now U = U (δ) = + δ
h 2l
Applying Castigliano’s first theorem,
∂U (δ)
= P.
∂δ
When we substitute numerical values, we again obtain the same answer: P = 0.4524 +
0.4418 = 0.8942 kN per mm of vertical deflection.
 13-49 A Few Illustrative Examples

46. Example: Unit-load Method

(a) To find the slope at A (b) A unit moment applied at A (c) Half the beam: a cantilever

Figure 13.19: Fig. 13.19 shows a simply supported beam with a uniformly distributed load
w /unit length. Fig. 13.19b shows a unit moment applied at the end A where we desire
to find out the slope (rotation). The bending moment diagram for this loading of a unit-
moment is also shown there. Fig. 13.19c shows how this problem may be solved easily by
an alternative method which Den Hartog calls the Myosotis method.

Using the unit-load method, calculate the slope θA of a simply supported beam AB of
length l with a uniformly distributed load w / unit length [Fig. 13.19a].
The bending moment M for this uniformly distributed load is
1 1
M = M (x) = wlx − wx2 (0 ≤ x ≤ l).
2 2
As we desire to determine the slope (rotation) θA at the end A, we apply a unit moment
at the point A as shown in Fig. 13.19b. The bending moment diagram due to this
applied unit moment is shown in the figure. Its expression is given by
x
m = m(x) = (o < x ≤ l).
l
Applying the equation
Z l
m
∆= dx,
0 EI
we obtain
2
l ( wlx wx x Z l
2 − 2 )( l ) wx2 wx3 w l3
Z 
1
θA = dx = − dx = .
0 EI E 0 2 2l 24 EI

A positive value shows that the slope (rotation) is in the direction of the applied moment
(that is, clockwise). Thus, this rotation is clockwise.
(It is interesting to note that this problem can be solved just by inspection using the
1 2 2 3 6 8 method (Myosotis method). Fig. 13.19c shows a cantilever which is really only
half the beam [Fig. 13.19a]. The slope θA (which happens to be equal to θB because
 Advanced Mechanics of Solids 13-50

of the symmetry) can be seen to be the superposition (taking into account the proper
signs) of (i) a uniformly distributed load w / unit length, and (ii) a concentrated load
of wl/2. Accordingly,
3 2
w( 2l ) ( wl )( l ) wl3 wl3 wl3
θA = θ B = − 2 2 = − = .
6 EI 2 EI 48 EI 16 EI 32 EI

47. Example: Cantilever Loaded by an End Load F

Let us illustrate the dummy load method to calculate the end deflection and the end

(a) A cantilever loaded by an (b) A unit dummy load applied (c) A unit dummy moment ap-
end load, F at the end plied at the end

Figure 13.20: A cantilever with an end load, F . To compute (i) the vertical deflection and
(ii) the rotation (slope) at the end (i) a unit dummy load and (ii) a unit dummy moment
are applied, respectively. The corresponding bending moment diagrams are shown.

slope of a uniform cantilever loaded by an end load F . Referring to Fig. 13.20a, the
bending moment M due to the load F is M = F x.
Now let us apply a unit load at the end of the cantilever where the deflection is desired.
The bending moment due to this unit dummy load 1 is m = x.
Z l
(F x)(1 × x) F x3 l F l3
Z
Mm
δ= ds = dx =  = downwards.
EI 0 EI 3EI 0 3EI
Now to calculate the end slope θ, we apply a unit dummy moment 1. The bending
moment m due to this unit moment 1 is m = 1 constant throughout the length.
The required slope θ at the end can be calculated as
Z l
F x2 l F l2
Z
Mm (F x)(1)
θ= ds = dx =  = clockwise.
EI 0 EI 2EI 0 2EI
A positive value for the answers (deflection and rotation) signifies that it is in the direc-
tion of the applied unit load / applied unit moment.

48. Example: Propped Cantilever

Fig. 13.21 shows a uniform cantilever (constant EI) of length l loaded by a uniformly
 13-51 A Few Illustrative Examples

(a) A propped cantilever loaded (b) Support removed and re- (c) A unit dummy load applied
by a u.d.l. placed by a reaction R at the end

Figure 13.21: A propped cantilever with a uniformly distributed load, w per unit length.
We employ a unit dummy load method to calculate the support reaction, R.

distributed load w per unit length and supported at the end as shown. We desire to
obtain the reaction R at the support using the dummy load method.
Referring to Fig. 13.21b, the bending moment is given by
1
M = − wx2 − Rx.
2
The bending moment due to the unit dummy load applied at the end is m = x. Thus,
the vertical deflection δ at the end is
Z l Z l
Mm 1 1 Rl3 wl4
deflection: δ = dx = (− wx3 + Rx2 ) dx = − . (13.28)
0 EI EI 0 2 3EI 8EI

But the vertical deflection at the end is zero. Hence setting this expression (13.28) for
the vertical deflection d = 0, we obtain the desired answer as R = (3/8) wl.

49. Example: A Large Thin Ring

(a) The full large, thin ring (b) Half the ring (c) Free-body diagram

Figure 13.22: A large, thin ring ACBD is loaded by a vertical downward force F [Fig.
13.22a]. Because of the symmetry about the vertical axis AB, only the left half is considered.
It is desired to find the locked-in moment (M1 ) and the axial thrust (N1 ), both at the point
A. The free-body diagram of a part of the ring (beam) is shown [Fig. 13.22c].
Advanced Mechanics of Solids 13-52

(a) A unit dummy load applied at A (b) A unit dummy moment applied at SA

Figure 13.23: To compute the expressions for (i) the vertical deflection and (ii) the rotation
of the cross-section at A, (i) a unit dummy load [Fig. 13.23a] and (ii) a unit dummy moment
[Fig. 13.23b] are applied respectively. By setting these expressions to zero, we obtain the
two equations to obtain the desired locked-in moment M1 and the axial force N1 .

Fig. 13.22a shows a large, thin ring with a vertical downward force F . We are required
to find the locked-in moment (M1 ) and the axial thrust (N1 ) at the topmost point of
the ring where the vertical load is applied.
Let us note that the structure and the loading are symmetrical about the vertical line
AB. Consequently, there will be neither a horizontal deflection, nor a rotation, at A.
Furthermore, the lowest point B may be regarded as fixed (clamped)23 . We, therefore,
need to consider only one half of the ring as shown in Fig. 13.22b. There will be a
locked-in moment (M1 ) and an axial thrust (N1 ) at A24 .
From Fig. 13.22c we note that the bending moment is given by
(
M1 + N1 R(1 − cos θ) − F2 R sin θ for 0 < θ < π2
M= (13.29)
0 for π2 < θ < π.

We know from our previous experience that (i) it is sufficient to use the straight beam,
Euler-Bernoulli theory, and not the Winkler-Bach theory for curved beams, because
the radius of curvature is large compared to the depth of the beam25 , and that (ii) the
bending moment is all that contributes to the total strain energy. [In principle, the shear
force and the axial thrust do contribute to the strain energy, but these contributions are
negligibly small in such cases of large, thin rings.]
Let us note that the two additional equations26 to determine the unknowns M1 and N1
are obtained by calculating the expressions for (i) the horizontal deflection, and (ii) the
23
Actually the point B may have a vertical deflection, but we also know that a rigid-body displacement does
not affect the internal forces or the stresses.
24
Students often fail to recognise the possible existence of these.
25
The beam looks curved and, indeed, it is curved. Yet it behaves like a straight beam! The curvature
effects are negligible.
26
in addition to the equations of equilibrium
 13-53 A Few Illustrative Examples

rotation of the cross-section, both at the point A, and set them equal to zero, because
we know from symmetry considerations that they must both be zero.

50. Example: Dummy Load at A

Let us apply a unit dummy load at A in the horizontal direction [Fig. 13.23a]. The
bending moment m because of the unit dummy load is

m = 1 × R(1 − cos θ).

R Mm R2 R π
Hence, the horizontal deflection at A = 0 = ds = M (1 − cos θ) dθ,
EI EI 0
giving us one equation as Z π
M (1 − cos θ) dθ = 0. (13.30)
0

51. Example: Dummy Moment at A

Let us now apply a unit dummy moment at A [Fig. 13.23b]. Hence,
Z Z π
Mm R
rotation at A = 0 = ds = M dθ,
EI EI 0

giving us the equation as Z π

M dθ = 0.. (13.31)
0

From Eqs (13.30), (13.31), we obtain

Z π
M cos θ dθ = 0. (13.32)
0

These two equations (13.31), (13.32) give, when the expression for M is substituted from
Eq. (13.29),
π
Z π Z π Z Z π
F 2 F
M1 dθ + N1 R (1 − cos θ) dθ − R sin θ dθ − R dθ = 0, and (13.33)
0 0 2 0 2 π
2
Z π Z π Z π
F 2
M1 cos θ dθ + N1 R (1 − cos θ) cos θ dθ − R sin θ cos θ dθ
0 0 2
Z0 π
F
− R cos θ dθ = 0. (13.34)
2 π
2

Solving these two equations (13.33, 13.34), we obtain the required result as

FR F
M1 = , and N1 = .
4 2π
This completes the illustrative example. Next we shall consider a thick curved beam.
 Advanced Mechanics of Solids 13-54

52. Example: A Thick Curved Beam

We had solved one problem earlier treating the beam as thin [Sec. 44, page 13-43]. We
had mentioned that, when the radius of curvature is large compared to the depth of the
beam, the curved beam theory need not be used, and that the strain energies due to (i)
the direct thrust and (ii) the shear force may be neglected. Let us demonstrate this. Let
us solve this problem here in this section, and show that the contributions of (i) and (ii)
are indeed negligibly small when the radius of curvature is large, thereby vindicating the
position that we took in Sec. 44. Here we shall solve the problem treating the beam as
thick. Referring to the free-body diagram, and using the sign conventions as per the

(a) A thick curved beam loaded by a horizontal load at (b) Shear force, thrust and bending
the end moment on a beam element

Figure 13.24: A curved beam BF [Fig. 13.24a] — a quarter of a circle — is loaded by a

horizontal end load H. It is desired to find the horizontal deflection δh at the free end F .
The free-body diagram of a part of the beam is shown alongside. At the right is shown
a beam element along with the shear force S, direct thrust N and bending moment M
[13.24b].

positive directions marked [Fig. 13.24b, page 13-54], we arrive at the expressions for the
bending moment, M ; the direct thrust, N ; and the shear force, S. These are

bending moment: M = HR sin θ; (13.35)

direct thrust: N = H sin θ; and (13.36)
shear force: S = −H cos θ. (13.37)

The total strain energy, U is given by

U = UM + UN + US
Z s 
M2
  2
αS 2
  
N MN
= + − + ds
0 2AEeR 2AE AER 2AG
Z π 
M2
  2
αS 2
  
2 N MN
= + − + R dθ.
0 2AEeR 2AE AER 2AG
 13-55 A Few Illustrative Examples

Substituting the expressions for M, N and S [Eqs (13.35, 13.36 and 13.37)], we obtain
the expression for the total strain energy, U as
Z π  2 2
H R sin2 θ H sin2 θ H 2 R sin2 θ
  2
αH 2 cos2 θ
  
2
U= + − + dθ.
0 2AEeR 2AE AER 2AG
Carrying out the differentiation of U w.r.to the horizontal load, H to obtain the hori-
zontal deflection of the free end, we have
Z π 
R sin2 θ
  
dU HR 2 2

αE 2
δh = = − sin θ + cos θ dθ
dH AE 0 e G
 
πHR R αE
= + −1
4AE e G
πHR 12R2
 
= + 2.12 (13.38)
4AE h2
We can see that in Eq. (13.38) the terms 2.12 is really negligible in comparison to
the first term within the pair of square brackets, viz., 12R2 /h2 , whenever h is small
compared to R. This vindicates the position that we had taken: that the effects of the
direct thrust and the shear forces are negligible for a thin beam.
53. Example: Thick Cylinders - Lamé’s Problem

Figure 13.25: The figures show a thick cylinder subjected to pressures. The solution for
the case (i) of both an internal pressure pi and an external pressure po can be obtained by
superposition of the solutions for the cases of (ii) an internal pressure pi only, and of (iii)
an external pressure po only.

Find the stresses in a thick cylinder subjected to an internal pressure pi and an external
pressure po . The inner and outer radii are ri and r0 respectively.
We know that the radial and tangential stresses are given by
B B
σrr = A + and σθθ = A − .
r2 r2
The constants A and B are determined using the known boundary conditions:
B
(a) at r = ri , σrr = −pi −→ A+ = −pi ;
ri2
Advanced Mechanics of Solids 13-56

B
(b) at r = ro , σrr = −po −→ A+ = −po .
ro2
Solving for A and B — the slightly ‘dirty’ algebra is left out —

pi ri2 − po ro2 (ri ro )2 (po − pi )

A= ; B= .
ro2 − ri2 ro2 − ri2

The expressions for the stress components are

2
pi ri2 − po ro2 + rirro (po − pi )
σrr = ;
ro2 − ri2

2
pi ri2 − po ro2 − rirro (po − pi )
σθθ = .
ro2 − ri2

[We may also write Lamé’s equations differently, but equivalently, as σrr = C − D/r2
and σθθ = C + D/r2 . The constants are now different. It is desirable for us to stick to
one set of formulae, but once in a while this habit can be broken to remind ourselves
that both sets of formulae are equally correct.]

(a) A closed cylinder (b) An open cylinder

Figure 13.26: Closed and open thick cylinders subjected to internal pressure. There will be
an axial stress σzz for a closed cylinder, but not for an open one.

Closed cylinder:
In addition, there will be an axial stress σzz which is uniform on the cross-section if the
cylinder is closed [Fig. 13.26a]. The cross-sectional area is π(ro2 − ri2 ). The fluid pressure
acting on the end plate is pi π ri2 . Hence the axial stress σzz is

pi π ri2 pi ri2
σzz = = .
π(ro2 − ri2 ) ro2 − ri2

Open cylinder:
If the cylinder is open [Fig. 13.26b], there is no axial stress: σzz = 0.
We may also calculate the stresses for each of the two loadings, viz., (i) pi only, and (ii)
po only, and add up the stresses to obtain the stresses for the combined loading of pi
and po . As explained earlier, this is the idea of superposition.
 13-57 A Few Illustrative Examples

Numerical values:
We can work out the numerical values and the distribution of stresses by giving some
values, say, ri = 100 mm, ro = 140 mm, pi = 2 MPa, po = 1 MPa.

54. Example: A Thick Cylinder Subjected to an Internal Pressure

A thick cylinder of steel is subjected to an internal pressure of 2 MPa. Calculate the
stresses inside the cylinder and the radial expansion of the inner radius. (The inner and
outer radii are ri = 100 mm, ro = 140 = mm; E = 200 GPa; ν = 0.3.)
We can use the formula to obtain the hoop (circumferential) stress σθθ :

ro2 + ri2 1402 + 1002

σθθ = p = 2 = 6.17 MPa.
ro2 − ri2 1402 − 1002

σrr at the inner radius = 2 MPa.

Closed cylinder
If the cylinder is closed, there will be tensile axial stress. Its magnitude is

ri2 1002
σaxial ≡ σzz = p = 20 × = 2.08 MPa.
ro2 − ri2 1402 − 1002

55. Example: Interference Fit Calculations

(a) An outer brass sleeve (b) The inner solid steel shaft (c) The assembly

Figure 13.27: Figs 13.27 show an interference fit (shrink fit) of a brass sleeve on a steel shaft.
Thus, the outer diameter of the solid steel shaft [Fig. 13.27b] is actually a little larger,
though nominally the same. When the sleeve is fitted on the shaft, the inner boundary of
the sleeve moves up slightly, and the diameter of the shaft correspondingly reduced. Fig.
13.27c shows the various diameters, before and after the assembly.

A brass sleeve is shrunk fitted on a steel shaft. The outer and inner diameters of the
sleeve are 280 mm and 200 mm. The shaft has a nominal diameter of 200 mm and
an actual diameter of 200.40 mm. (The interference on the radius is 0.40/2 = 0.20
mm.) Calculate the interference pressure and the stresses in the shaft and in the sleeve.
The Young’s moduli of elasticity E and the Poisson’s ratios ν are Ebrass = 100 GPa;
Esteel = 200 GPa; νbrass = 0.33; and νsteel = 0.30.
Advanced Mechanics of Solids 13-58

Procedure:
(i) Understand the problem: The diameter of the inner steel shaft (nominal diameter
200 mm) is a little more (0.40 mm more) than the inner diameter of the outer brass
sleeve. The two are force fitted. There is an interference pressure ps which acts
radially inwards on the inner steel shaft [Fig. 13.27b], and radially outwards on the
inner boundary of the outer brass sleeve [Fig. 13.27a]. The interference pressure
ps is unknown.
(ii) This statically indeterminate interference pressure ps (also called shrinkage pres-
sure) is unknown. We need to use the principle of consistent deformation — the
compatibility of displacements — to determine this ps .
(iii) This is the same problem that was worked out earlier, except that the numerical
value of the pressure ps is now unknown. Thus, we can obtain the (expressions for
the) stress components σrr and σθθ at the interface in both the brass sleeve and
the steel shaft.
(iv) The (expression for the) tangential strain
u 1
eθθ = = [σθθ − ν σrr ] (the axial stress σzz = 0)
r r
can be worked out for both the steel shaft and the brass sleeve in terms of the
unknown interference pressure ps . The appropriate E, Es (steel) or Eb (brass), is
to be used for these calculations.
(v) From the above, ushaf t and usleeve can be calculated at the common boundary
(interface) in terms of ps . The former ushaf t is radially inwards, while the latter
usleeve is radially outwards.
(vi) Now the principle of consistent deformation is used. It says [Fig. 13.27c]
|ushaf t | + |usleeve | = δ (interference on the radius = 0.40/2 = 0.20 mm)
As ushaf t and usleeve are of different signs, it is less confusing to consider their
numerical values only.
(vii) This equation enables us to determine the interference pressure ps .
(viii) Having obtained the numerical value of ps , the (numerical values of the) stresses
can be calculated at the interface in both the steel shaft and in the brass sleeve.
(ix) It will add to our understanding to see the nature and distribution of both σrr and
σθθ in (i) the steel shaft, and in the brass sleeve.
Now we can undertake the detailed calculations.
Interference pressure
First we shall calculate the interference (shrinkage) pressure ps . We have already worked
out the expression for the interference pressure ps earlier in the section on Thick Cylin-
ders. Let us borrow the result:
δ
ps = h  2 2
r +r
 i.
1 o 1
ri Ebrass 2
r −r
i
2 + νbrass + Esteel (1 − νsteel )
o i
 13-59 A Few Illustrative Examples

For the numerical values given, viz., ro = 140 mm, ri = 100 mm, δ = 0.20 mm, Esteel =
200 GPa and Ebrass = 100 GPa, the above expression gives us ps = 26.57 MPa.
Next, we shall work out the resulting (maximum) stresses in the sleeve and the shaft.
Stresses in the sleeve and the shaft
The maximum stress in the sleeve is the circumferential stress σθθ at the interface radius.
This is
r2 + ri2 1402 + 1002
σθθ = ps o2 = × = 81.92 MPa.
ro − ri2 1402 − 1002

σθθ = 81.92 MPa, tensile.

The stress in the shaft is easy to find; in fact, no calculation is needed. Both the
circumferential and the radial stresses are equal and, what is more, the same everywhere,
and equal to the interference pressure. Thus,

σθθ = σrr = −ps = −26.57 MPa, compressive.

56. Example: Torsion of an Elliptical Cross-section

Let us solve the torsion problem for a prismatic bar of elliptical cross-section [Fig.

(a) An elliptical cross-section (b) A triangular cross-section

Figure 13.28: An elliptical and a triangular cross-sections. The maximum shear stress on
the cross-section is at the boundary points nearest from the centroid.

13.28b]. By solving the torsion problem, we mean to obtain the shear stresses on the
cross-section, and the twist θ per unit length, when a known torque T acts on the bar.
We use the third formulation.
The governing equation and the boundary conditions are

∂2F ∂2F
+ = −2 Gθ in the region R,
∂x2 ∂y 2

subject to the boundary condition

x2 y 2
F = 0 on the only boundary given by the equation + 2 = 0.
a2 b
Advanced Mechanics of Solids 13-60

One convenient and clever technique of satisfying this boundary condition is to assume
the desired function F = F (x, y) in the form
 2
y2

x
F =A + 2 −1 ,
a2 b
where A is an unknown constant. We can obtain the value of this constant by substituting
this assumed form of the solution Φ in the governing differential equation 52 F = −2 Gθ,

∂F 2 Ax ∂2F 2A ∂F 2 Ay ∂2F 2A
= 2 ; 2
= 2; = 2 ; 2
= 2 .
∂x a ∂x a ∂y b ∂y b
 2 2 
2A 2A a b
52 F = −2 Gθ −→ + 2 = −2 Gθ −→ A = −Gθ .
a2 b a2 + b2
Thus,
a2 b2 x2 y 2
 
F = F (x, y) = − 2 Gθ + 2 −1 (13.39)
a + b2 a2 b
is the solution of the differential equation. The problem is solved in principle. All that
remains to be done is to calculate (i) the shear stresses τzx and τzy on the cross-section
and, in particular, the maximum values and their locations, and (ii) then torsional
rigidity. We shall calculate them below.
The shear stresses are merely the partial derivatives of F . Thus,
∂F ∂F
τzx = ; and τzy = − .
∂y ∂x
These give us the shear stress components in terms of the twist θ per unit length.
However, it is more convenient to obtain the results in terms of the applied torque T .
To calculate the torsional rigidity C = T /θ also, we need to work out the expression for
T in terms of Φ. The torque T is given by
ZZ
T =2 F dx dy
R
2 a2 b2
 ZZ ZZ ZZ 
1 2 1 2
=− 2 Gθ 2 x dx dy + 2 y dx dy − dx dy (13.40)
a + b2 a R b R R

The three integrals within the square brackets [· · · ] represent, respectively, the second
moments of area (of the elliptical cross-section) about the y− and x−axes, and the area
of the ellipse.
ZZ ZZ
2 1 3 1
Iyy = x dx dy = πa b; Ixx = y 2 dx dy = πab3 ; area = πab.
R 4 R 4
If we make these substitutions in Eq. (13.40), we obtain

2(a2 + b2 )) T
−2 Gθ = − .
πa3 b3
 13-61 A Few Illustrative Examples

Thus, we have
x2 y 2
 
T
F =− + 2 −1 . (13.41)
πab a2 b
The stresses are
∂F 2Ty ∂F 2Tx
τzx = =− ; τzy = − =− 3 . (13.42)
∂y πab3 ∂x πa b
The resultant shear stress τ is, thus,
r
q
2 2
2T x2 y 2
τ= τzx + τzy = + 4. (13.43)
πab a4 b
The maximum shear stress occurs at the boundary points A and B which as closest to
the origin (centre of the section), and not at the farthest points! The maximum shear
stress τmax and the torsional rigidity C are obtained as
2T T
τmax = and C= = πa3 b3 . (13.44)
πab2 θ
Warning! This technique will not work for all shapes. If, on substitution in the
governing equation 52 F = −2Gθ (Poisson’s equation), A turns out to be not a constant,
this method will fail. The idea of using the equation to the boundary to satisfy the
boundary condition F = 0 on the boundary is applicable in all cases. This idea is
exploited in several places, including the methods to obtain approximate solutions.

57. Example: Torsion of a Triangular Cross-section

As another example, let us solve the torsion problem for a triangular cross-section shown
in Fig. 13.28b. As before we use the third formulation to solve the problem.
To satisfy the boundary condition, let us write the equations to the three sides OA, OB
and AC of the triangle. These are, respectively,
√ √
x + 3y = 0; x − 3y = 0; and x − a = 0.
 √ √ 
Let us, therefore, choose F in the form F = A (x + 3y)(x − 3y)(x − a) , so that
F = 0 at all points on the boundary. Substituting, as before, in the governing differential
equation 52 F = −2Gθ (Poisson’s equation), we find that

∂2F ∂2F −Gθ −Gθ 

(x − a)(x2 − 3y 2 ) .

+ = 4 Aa = −2Gθ → A= → F =
∂x2 ∂y 2 2a 2a
The torque is calculated as
√
−Gθ x=a y=x/ 3
ZZ Z Z
2 2
T =2 Φ dx dy = √ (x − a)(x − 3y ) dx dy
R a x=0 y=−x/ 3
Z x=a
4 Gθ Gθ a4
=− √ x3 (x − a) dx = √
3 3 a x=0 15 3
 Advanced Mechanics of Solids 13-62

The angle of twist θ per unit length and the torsional rigidity C are
√
15 3 T G a4
θ= T ; and C = = √ .
G a4 θ 15 3
The shear stresses are
∂F 3 Gθ ∂F Gθ
τzx = = (x − a)y; and τzy = − = (3x2 − y 2 − 2ax).
∂y a ∂x 2a
The maximum shear stress on the cross-section √ is at P 3 which is the boundary point
closest to the centre. Its value is τmax = (15 3 T )/(2a ). [As remarked earlier, this
technique will not always work.]
58. Example: Torsion of a Circular Prismatic Bar
Obtain the Saint-Venant’s solution for the torsion of a circular prismatic bar.
We have already obtained the solution for an elliptical prismatic bar. A circle is a special,
simplified, case of an ellipse. Thus, we can surely obtain the required solution by setting
a = b in Eq. (13.39), p. 13-60. However, we shall begin afresh. We follow the third
formulation.
Let us assume F in the form F = C(x2 + y 2 − r2 ). The boundary condition F = 0
is obviously satisfied by this function. We can try27 to obtain C by substituting this
assumed solution in the governing equation 52 = −2Gθ. This gives us 2C +2C = −2Gθ,
i.e., C = −Gθ/2. Thus, the solution is F = −(1/2)Gθ(x2 + y 2 − r2 ). The torque T is
given by
ZZ ZZ
1
T =2 F dx dy = 2 − Gθ(x2 + y 2 − r2 ) dx dy
2
Z Z ZZ ZZ 
2 2 2
= −Gθ x dx dy + y dx dy − r dx dy

= −Gθ(Iyy + Ixx − πr4 ).

For our cross-section, Ixx = Iyy and, therefore, the polar second moment of area J =
Ixx + Iyy = πr4 /2. Thus,
T
T = −Gθ(J − 2J) = GθJ −→ = Gθ.
J
We arrive at the solution as
1T 2
F =(x + y 2 − r2 ).
2 J
The shear stress components on the cross-section are obtained by differentiating F w.r.to
x and y. Hence we have
∂F Ty Tr
τzx = =− −→ when y = r, τzx = − ;
∂y J J
27
If, on substitution, C turns out to be a constant, then all is well; this method works. On the other hand,
if we do not obtain C as a constant, this method fails.
 13-63 A Few Illustrative Examples

∂F Tx Tr
τzy = − =− −→ when x = r, τzy = ;
∂x J J
These answers agree with the results of Coulomb’s theory.

59. Example: Membrane Analogy

We know that a thin-walled tube would lose nearly all of its torsional rigidity if we

(a) A thin-walled closed section (b) The same cross-section but with a thin slit

Figure 13.29: A thin-walled closed tube and the shape of the corresponding membrane are
shown [Fig. 13.29a]. Fig. 13.29b shows the same section, but with a marked difference: a
longitudinal slit is made. Now there is only one boundary; the section has changed from a
doubly connected region to a simply connected one. The soap film has all but collapsed.
If we compare the volumes in the two figures we can see that there is a drastic reduction
in the volume under the membrane. This tells that almost all the torsional rigidity is lost
when we introduce a longitudinal slit.

introduce a longitudinal slit [Fig. 13.29]. Let us try to explain and understand this fact
in the light of the membrane analogy.
Fig. 13.29a shows a thin-walled cross-section and the corresponding soap film. As the
plate is at a certain distance, the soap film has quite some volume under it. When a
longitudinal slit is made, the cross-section now becomes singly connected! There is only
one boundary. The volume under the soap film is now only a tiny little bit (shown
shaded) in Fig. 13.29b telling us unmistakably that the section has lost nearly all of its
torsional rigidity. This is our experience, but it is always nice to have a sound theoretical
basis and a convincing technical argument supporting our common-sense understanding!

60. Example: Torsion of a Thin-walled Section

Shown in Fig. 13.30 are two hollow thin-walled rectangular box sections, Fig. 13.30a
with the same thickness throughout, and Fig. 13.30b with different thicknesses. If a
torque of 1.5 kN m is applied, find the shear stress τ and the twist θ per unit length.
The dimensions are given in the figures, all in mm. Assume the material to be an
aluminium alloy for which the modulus of rigidity G is 26 GPa.
Advanced Mechanics of Solids 13-64

(a) A cellular section (same thickness) (b) A cellular section (different thicknesses)

Figure 13.30: Two cellular cross-sections

Case (a) Uniform thickness t [Fig. 13.30a]:

The shear stress τ , we know, is given by Eq. (11.30), p. 11-22.

T 1.5 × 1000 × 1000

τ= = = 37.5 N/mm2 .
2 At 2(100 × 50) × 4

The angle of twist is given by Eq. (11.32) as

TL 1.5 × 1000 × 1000(100 + 50 + 100 + 50)

θ= 2
= = 43.27×10−6 rad = 2.48×10−3 ◦ .
4 GA t 4 × 26 × 103 × (100 × 50)2 × 4

Case (b) Different thicknesses t [Fig. 13.30b]:

As the height of the plate is a constant, the shear stress τ in the various legs will be
different. Another way of looking at the situation is to invoke the fluid flow analogy:
the velocity in a narrow channel will be higher than that in a wider one. Accordingly,
the shear stress τ in the portion AE is given by Eq. (11.30) as

T 1.5 × 1000 × 1000

τ in the leg AE = = = 50.0 N/mm2 ;
2 AtAE 2(100 × 50) × 3
3 3
τ in the leg EB = × τ in the leg AE = × 50.0 = 30.0 N/mm2 ;
5 5
3 3
τ in the leg BC = × τ in the leg AE = × 50.0 = 37.5 N/mm2 ;
4 4
3 3
τ in the leg CD = × τ in the leg AE = × 50.0 = 60.0 N/mm2 ;
2.5 2.5
τ in the leg DA = τ in the leg AE = 50.0 N/mm2 .

The angle of twist per unit length θ is given by [Eq. (11.31)

I  
T ds T ds ds ds
θ= = + + + ···
4 GA2 t 4 GA2 t1 t2 t3
 
T AD DC CB BE EA
= + + + +
4 GA2 3 2.5 4 5 3
 13-65 A Few Illustrative Examples

 
1.5 × 1000 × 1000 50 100 50 30 70
= + + + +
4 × 26 × 103 3 2.5 4 5 3
= 56.827 × 10−6 rad/mm = 56.827 × 10−3 rad/m = 3.26 ◦ /m.

(Such a section with the thickness varying like this is unlikely to be used in practical
applications. This case is taken up merely to illustrate the procedure.)

61. Example: Torsion of Cellular Cross-sections

(a) A cellular cross-section (b) A two-cell cross-section

Figure 13.31: Two cellular cross-sections. The thicknesses of the various walls are different.
The membrane corresponding to the two-cell cross-section is shown in Fig. 13.31b. The
heights of the plates are h1 and h2 .

The cellular cross-section shown [Fig. 13.31a] is subjected to a twisting

R moment of 5,000
Nm. Find the shear stress in each wall. Evaluate the line integral ds/t, and obtain
the torsional rigidity of the cross-section. Assume that the modulus of rigidity of the
material G = 27 GPa.
  
6 5 7 8
Area A enclosed by the middle line = 200 − − 150 − −
2 2 2 2
= 27, 716.25 mm2 .
T 5000 × 1000
shear stress in the wall AB τAB = = = 11.27 MPa.
2 A tAB 2A × 8
8
shear stress in the wall BC τBC = τAB × = 18.03 MPa.
5
8
shear stress in the wall CD τCD = τAB × = 12.88 MPa.
7
8
shear stress in the wall DA τDA = τAB × = 15.03 MPa.
6
R ds
The line integral is calculated as:
t
(200 − 62 − 52 ) (150 − 82 − 72 ) (200 − 26 − 52 ) (150 − 82 − 72 )
Z
ds
= + + + = 104.34.
t 8 5 7 6
 Advanced Mechanics of Solids 13-66

The torsional rigidity C can now be calculated as:

T 4 G × A2 4 × (26 × 109 ) × (27.72 × 10−3 )2

C= = R ds = = 0.765 × 106 Nm/(rad/m).
θ t
104.34

62. Example: Torsion of Multi-cellular Cross-sections

Discuss the problem of torsion of multi-cellular sections.
The general approach indicated in the text [p. 11-23] is explained here in a slightly
different way for better clarity of understanding. If there are n cells, the heights of
the weightless plates held horizontal by pure couples only are, in general, different:
hi (i = 1, 2, 3, · · · , n), i.e., (hi , h2 , · · · , n). The slopes of the membranes on the outside
walls are h/t. For the walls inside, the slopes are ∆h/t. The thickness t is the appropriate
wall thickness which need not be a constant; it can be different at different points. ∆h
is the difference in the heights of the central horizontal plates concerned. The equation
of equilibrium of the ith plate is
I
∆h
(equation of equilibrium in the vertical direction:) pAi = S ds. (13.45)
t

The integral is around the path surrounding the ith plate. As explained earlier, ∆h is
the difference in the heights of the two adjacent horizontal plates concerned.
There are n such equations (i = 1, 2, · · · , n), all linear, algebraic, non-homogeneous
equations for the n unknown heights h1 , h2 , · · · , hn . They can be solved. After these
heights are determined, we can change over from the membrane to the torsion problem
by invoking the membrane analogy:
∆h p
slope = → τ (shear stress); → 2 Gθ; 2 × volume → T.
t S
Now the problem is solved in principle. However, the procedure is rather difficult when
there are three or more cells. Several special methods and shortcuts are available in the
published literature, but we are not concerned with them here. Let us note that we are
not required to perform any experiment. The membrane analogy is used for visualisation
and a qualitative understanding to work out the solution.

63. Example: Two-cell Cross-sections

What are the shear stresses in the various portions AB, BC, · · · , etc. in each of the two
cross-sections when a twisting moment of 30 kNm is applied? What is the twist θ per
unit length in each of two cases? Some of the dimensions in mm are given in the figures.
For the section on the left, the thickness t = 3 mm is the same for all the legs. For the
other section (on the right), t1 = 3, t2 = 4, t3 = 3.5, t4 = 2.5, t5 = 5, t6 = 4, t7 = 4.5,
all in mm. Take the modulus of rigidity G = 27 GPa.
Case (a) Uniform thickness [Fig. 13.32a]
Consider the equilibrium of the two plates held horizontal. For the first plate, there is
the air pressure p×A1 acting upwards, and the vertical component of the surface tension
13-67 A Few Illustrative Examples

(a) A two-cell section (same thickness) (b) A two-cell section (different thicknesses)

Figure 13.32: Two two-cell cross-sections. The thickness of the various walls is the same in
Fig. 13.32a, while the thicknesses are different in Fig. 13.32b. The membranes correspond-
ing to the two-cell cross-section are also shown in the two figures. The heights of the plates
are h1 and h2 .

all around the plate. Similar is the case for the second plate also. Thus, we obtain for
the two plates
S
pA1 = [100 × h1 + 120 × (h1 − h2 ) + 100 × h1 + 120 × h2 ] ;
t
S
pA2 = [150 × h2 + 120 × h2 + 150 × h2 + 120 × (h2 − h1 )] .
t
In each of the cases we go around the plate from the left lower point in the anti-clockwise
direction. The thickness t, being the same for all the terms, is taken out as a common
factor. Note that the slope is (h1 − h2 )/t in one case, and (h2 − h1 )/t, in the other.
[It is also possible for us to consider the equilibrium of both plates together by going
around both A1 and A2 . But this will not give us a third separate (linearly independent)
equation; this would merely be the sum of the two equations already obtained. The
contribution from the leg BE common to both cells will be zero.]
Here are two equations — linear, simultaneous, algebraic — in the two unknowns h1
and h2 . If we solve for them, we obtain
p p
h1 = 116.13 ; h2 = 125.81 .
S S
The shear stress is the same in the legs AB, EF , and F A. The shear stress has the same
value in the legs BC, CD, and DE (but different from those in the other legs). Which
is larger depends on which, between h1 and h2 , is larger. Here we see that h1 < h2 . The
volume V of the membrane is
p
V = A1 h1 + A2 h2 = [116.13 × A1 + 125.81 × A2 ] .
S
Now that we have obtained all the relevant information of the membrane, we can use
the membrane analogy and pass on to the torsion problem.
∆h p
slope = → τ (shear stress); → 2 Gθ; 2 × volume → T.
t S
Advanced Mechanics of Solids 13-68

p
V =A1 h1 + A2 h2 = (100 × 120 × 110.11 + 150 × 120 × 120.02)
  S
6 p P
= 3.66 × 10 −→ 2 Gθ
S S
p
T = 2V = 2 × 3.66 × 106 = 7.32 × 106 × 2 Gθ
S
T 30 × 103 × 103
2 Gθ = = = 4.10
7.32 × 106 7.32 × 106
4.10
θ= = 75.9 × 10−6 rad/mm = 75.9 × 10−3 rad/m = 4.35 ◦ /m.
2 × 27 × 103
The shear stress is the same in the legs AB, EF and F A.
h1 116.13 p 116.13 116.13
τAB = = = × 2Gθ = × 4.10 = 158.71 MPa;
t 3 S 3 3
h2 − h1 125.81 − 116.13 p 9.68 9.68
τBE = = = × 2Gθ = × 4.10 = 13.22 MPa;
t 3 S 3 3
h2 125.81 p 125.81 125.81
τBC = = = × 2Gθ = × 4.10 = 171.94 MPa.
t 3 S 3 3
The shear stress is the same in the legs BC, CD and DE.
Case (b) Different thicknesses [Fig. 13.32b]
This case is also similar. Again, as before we write down the equation of equilibrium
for each of the two plates. For the first plate, we go around the first plate taking the
path AB - BE - EF - F A. This gives us the first equation. For the second plate, the
path around it is BC - CD - DE - EB. This gives us the second equation. These two
equations (of equilibrium) are the following. (Now that the thicknesses of the various
legs are different, the thickness cannot be taken outside as a common factor.)

 
100 120 100 120
pA1 = S × h1 + × (h1 − h2 ) + × h1 + × h1 ;
3 4.5 5 4
 
150 120 150 120
pA2 = S × h2 + × h2 + × h2 + × (h2 − h1 ) .
4 3.5 2.5 4.5
Here are two equations — linear, simultaneous, algebraic — in the two unknowns h1
and h2 . If we solve for them, we obtain
p p
h1 = 142.44 ; h2 = 137.57 .
S S
Here we find that h1 > h2 . Now the heights of the plates are as shown in Fig. 13.31b.

p
V = A1 h1 + A2 h2 = (100 × 120 × 142.44 + 150 × 120 × 137.57)
  S
p P
= 8.37 × 106 −→ 2 Gθ
S S
13-69 A Few Illustrative Examples

p
T = 2V = 2 × 4.185 × 106 = 8.37 × 106 × 2 Gθ
S
30 × 103 × 103
2 Gθ = = 3.58
8.37 × 106
3.58
θ= = 66.29 × 10−6 rad/mm = 66.29 × 10−3 rad/m = 3.79 ◦ /m.
2 × 27 × 103

The shear stresses in the various legs can be calculated as explained earlier and as shown
below. Usually we are interested in the maximum shear stresses. As h1 and h2 are different,
we cannot state just by inspection where the maximum shear stress occurs.

h1 142.44 p 142.44 142.44

τAB = = = × 2Gθ = × 3.58 = 169.98 MPa;
t1 3 S 3 3
h1 142.44 p 142.44
τEF = = = × 3.58 = 101.98 MPa;
t5 5 S 5
h1 142.44
τF A = = × 3.58 = 127.48 MPa;
t6 4
h1 − h2 142.44 − 137.57
τBE = = × 3.58 = 3.87 MPa;
t7 4.5
h2 137.57
τBC = = × 3.58 = 123.13 MPa;
t2 4
h2 137.57
τCD = = × 3.58 = 140.71 MPa;
t3 3.5
h2 137.57
τDE = = × 3.58 = 197.00 MPa.
t4 2.5

The maximum shear stress of 197.00 MPa occurs in the leg DE. The stress in the leg BE
separating the two cells is, expectedly, very low. This is because the shear flow there is the
difference between the two shear flows surrounding the two cells.
What are the comments to be added? Unless there are numerical mistakes (which is
indeed a clear possibility!), the shear stresses and the angle of twist seem to be too large
for the material suggested by the given value of G. What is a probable material? (Usually
we know the material beforehand. Here we are asking this question just to provoke the
curiosity of the readers. Solution of numerical problems should not seen as merely plugging
in numbers in a set of given formulae. There are deeper issues.) What then? This section
seen as a possible design is unsatisfactory; it is to be revised so that the stresses become
less and the torsional rigidity is increased.
A Few Suggestions
When we work out problems we may make mistakes. Examiners will, let us hope, under-
standingly give marks for the various steps and the procedure. They are advised to give
about three times the time it takes for them to work out a problem. Yet these suggestions
are not always followed. It may be a good habit to check how long it takes to work out a
problem.
Advanced Mechanics of Solids 13-70

When we get unreasonable numerical values, we can protect ourselves by writing a

comment. For example, we may get a value of, say, 0.6 for the Poisson’s ratio. This can be
because the values (data) given in the problem are wrong or unrealistic, or more probably,
because of the mistakes that we make. Such a value for the Poisson’s ratio is clearly
impossible. In such cases, it is absolutely necessary to add a comment: that the value of
the Poisson’s ratio cannot exceed 0.5, and that even this value corresponds to incompressible
materials or conditions (as, for example, in plasticity, we assume incompressibility).
It is also a good habit to have an idea of the order of magnitude for each physical
quantity. For example, the deflection of an ordinary beam cannot be as large as, say, 2 m.
Again, if we obtain an expression for the deflection of a beam as, say,

P1 l 3 P2 l 2 wl3
deflection δ = + + ,
48 EI 3 EI 8 EI
one close look at the formula tells us that it is wrong. Why? The dimensions do not match.
(Technically we say that the equation must be dimensionally homogeneous.) Let us be
sensitive to these matters. These are good habits to cultivate.
Finally we need a little bit of luck. Dear young students, best of luck!

In the next and last chapter we shall see a little bit of the early history of our subject.
Chapter 14

EARLY HISTORY

We had made a few indications earlier about the history of our subject. It is desirable, very
desirable, to have a historical perspective1 of the development. This is necessary to have
a healthy interest in a subject that is not dead, but vibrantly alive and kicking. However,
the scope is too vast and, therefore, it is decided to limit this to the early history.
ANCIENT TIMES

(a) Great Wall of China (b) The Pont du Gard (c) A Step-well of Ancient India

Figure 14.1: The Great Wall of China, more than 2300 years old; the Pont du Gard, an
ancient (40 − 60 AD) aqueduct in Southern France, 48.8 m tall; one of the famous step-wells
(200 AD?) of ancient India

It is known that even long, long ago people did constructions of various kinds: roads,
buildings, bridges, canals, towers, etc. Three examples are shown in Fig. 14.1 above. These
are clearly impossible without some knowledge of strength of materials. How can people
have arrived at the dimensions of the various members in these constructions? It was
perhaps only empirical rules gained with experience. Egyptians certainly must have had
some knowledge; otherwise it would be impossible to construct not only the great pyramids,
but even the less impressive monuments and other structures. The Greeks too possessed
considerable knowledge in the art of building. The fundamental science upon which this

1
Part of the material for this section is taken from Timoshenko’s well known book [14]. Readers are advised
to read this book.
Advanced Mechanics of Solids 14-2

knowledge was developed is statics. Archimedes2 knew a great deal of science. He knew the
conditions of equilibrium — he had given rigorous proofs — of levers. He also knew, among
many other things, how to locate the centres of gravity of bodies. He had used all this
knowledge in constructing different types of hoisting devices. Considering that all this was
accomplished in the dim distant past, he must surely rank as one of the greatest scientists.
The Romans also were not far behind in construction activities. Many of their bridges,
temples and monuments are a testimony to their knowledge. They may not have developed
the science of stress analysis. Nevertheless, they did use arches even though the spans
were not too large and the shapes mainly semi-circular. It seems reasonable to assume
that great civilisations such as we had in India, China and Mesopotamia could not have
evolved and matured without considerable construction activities of various kinds: bridges,
canals, columns, dams, domes, forts, fortifications, monoliths, obelisks, roads, roofs, tunnels,
· · · 3 . This, in turn, lends credence to supposing that they too knew the essence of what is
needed to undertake such building work4 . There are several cases where we seem to have
underestimated the knowledge and skills that our forefathers possessed!
It appears that all these advances were lost. European engineers in the sixteenth century
had great difficulty to accomplish engineering tasks that the ancients like the Egyptians must
have done rather routinely: to carry heavy stones to the Nile, carrying obelisks to other
sites, etc. It was only after the Renaissance that the broken thread was picked up again
and progress made.
SIXTEENTH / SEVENTEETH CENTURY: GALILEO

(a) Archimedes of Sicily (b) Leonardo da Vinci (c) Galilei Galileo

Figure 14.2: Three great early scientists

2
Archimedes of Syracuse (287 BC - 212 BC) born in Sicily, Italy was an ancient Greek scientist, perhaps
the greatest among the ancient scientists. He was a mathematician, physicist, astronomer, engineer and
inventor.
3
This list in alphabetical order is borrowed from the internet. This help is gratefully acknowledged.
4
There probably is no clearly documented evidence. Perhaps whatever was available is lost for ever. Ravages
of time and carelessness in preservation are possible reasons. Western scholars are either wholly ignorant
of such non-European contributions, or are reluctant to admit that intelligence and creativity are not their
monopoly. It is also possible that the colonial powers suppressed and even destroyed such evidence so that
they can dominate over the subject race in every sphere.
Finally let us note that absence of proof is not proof of absence!
14-3 Early History

Leonardo da Vinci5 was an exceptional genius. His fame is perhaps more as the leading
artist of his times, but he was also an outstanding engineer-scientist as revealed in his note-
books. He knew a great deal of mechanics. He was able to obtain the correct solutions of
several problems in statics using the method of moments. He had carried out experiments
to obtain the strength of iron wires. He had also examined the strength of beams and had
enunciated certain general principles. Columns also came within the purview of his investi-
gations. All put together, one wonders whether he was primarily an engineer-scientist who
knew a thing or two about art. His statement “Mechanics is the paradise of mathematical
science because here we come to the fruits of mathematics” is very significant and relevant
even to this day. A truly remarkable person indeed!
The next great name that we find is that of Galileo6 . In 1586 Galileo measured the
density of several substances using a hydrostatic balance that he made. His work on the
centre of gravity attracted attention, and he became a professor of mathematics at Pisa
when he was not even 26 years old!
It was during the years 1589 - 1592 that he performed the famous experiments on freely
falling bodies using the leaning tower of Pisa! His findings were is sharp contrast to the
prevailing ideas based on Aristotelian mechanics, and he fell out of favour with the powers
that be. He, therefore, had to leave Pisa, but he could move over to the University of Padua.
The position was kept vacant until a suitable person like Galileo could be found.
Galileo was very active during the first few years at Padua7 . He did a tremendous
amount of great work there. He became famous; his lectures attracted students in large
numbers from several European countries. A very large room had to be used to accommo-
date the very large number of students.
He wrote the famous treatise Della Scienza Meccanica in 1594. The principle of virtual
work was used in this to treat several problems in statics. This book gained considerable
circulation; copies of the manuscript were widely circulated. At that time, there was ship-
building activity, and it became necessary to examine problems in strength of materials in
this perspective. Galileo became interested in such studies. It was not long before he got
5
Leonardo di ser Piere da Vinci, better known as Leonardo da Vinci (April 1452 - May 1519)
6
Galileo Galilei (Feb. 1564 - Jan. 1642) was an Italian astronomer, physicist, engineer, philosopher and
mathematician. He played a key role in the scientific revolution of the seventeenth century.
He was born in Pisa. His native place was Florence. He studied Latin, Greek and logic early in his career,
and started studying medicine in Pisa University. But he became interested in mathematics, and studied
the work of Euclid and Archimedes. He was also attracted by Leonardo da Vinci’s work in mechanics.
But for want of money he was forced to leave Pisa in 1585 without a degree, but he gave private tuition
in mathematics and mechanics. He also continued his own scientific studies. When he became well known
for his scientific work, he became a professor, first at Pisa and later at Padua. His scientific output during
the early years at Padua was amazing. However, as he supported the Copernicus’ theory based on his
investigations in astronomy, the Church was not amused; he was summoned to Rome by the Inquisition.
He was under house arrest for the last eight years of his life. It may be of interest to the readers to know
that Pope Paul II did express, though much belatedly in 1992, regret for the wrong position taken by the
Church. Vatican also released two stamps in his honour as an acknowledgement and admission of guilt.
7
The University of Padua was founded in 1222. This was one of the oldest universities in the world and
the second oldest in Italy.
Advanced Mechanics of Solids 14-4

involved with astronomy also. Galileo states in a letter to Kepler that “many years ago
I became a convert to the opinions of Copernicus · · · ”. Galileo took serious notice of a
rumour of the invention of a telescope. This was in 1609. He proceeded to make his own
telescope. His telescope had a magnification of 32. Using his telescope, he made important
discoveries in astronomy. He could establish important results which had a great impact on
further developments in astronomy. These also provided powerful support to Copernicus’
theory. It became possible to have direct visual confirmation of several facts. These few
years of his work at Padua made him famous. The grand duke of Tuscany nominated him
as the ‘philosopher and mathematician extraordinary’.
He now left Padua to return to his native place Florence. His new appointment gave
Galileo plenty of time to pursue his scientific work, because it entailed no other official duties.
For him this was the time of intense activity in astronomy and astronomical observations.
Among his achievements are the discovery of the peculiar shape of Saturn, observation of
the phases of Venus, and descriptions of the spots on the sun. The Copernicus’ theory now
received more and more support from Galileo’s findings and vigorous writings. Galileo was
indeed a towering figure.
Galileo’s Work in Strength of Materials
Galileo’s book Two New Sciences is very famous and well known. The first two ‘dialogues’
of this book contains his work in the mechanics of materials. He makes several observations
about geometrically similar structures. He considers the strength of a bar in direct tension,
and concludes that the strength (which he calls the ‘absolute resistance to fracture’) is
proportional to the area of the cross-section, and independent of the length. He gives the
numerical value of the ultimate strength of copper. He also considers a cantilever loaded
at the end, and states where it will fracture, should the beam break. He examines cases of
bending and makes important conclusions.
SEVENTEENTH CENTURY
Our subject can properly be considered to have been born in the seventeenth century.
The early investigators were primarily French mathematicians. In those days the Church
controlled most of the universities, and often stepped well outside their legitimate domain
of matters spiritual. Accordingly, the sublime duty of the universities of being centres of
knowledge, promoting independent research in a free and fearless atmosphere, could not be
fulfilled. Instead scientific societies and national academies of science sprang up in several
places in Europe. Italy took the lead in this matter. The Accademia Secretorum Naturae
was set up in Naples as early as 1560. Rome had its famous Accademia dei Lincei in 1603
with Galileo as one of its members. After Galileo died, the Accdemia del Cimento came up in
Florence. The publications of this academy had discussions on scientifc matters pertaining
to thermometers, barometers, and pendula, and various experiments.
In England at about 1645, a set of interested persons used to meet in London every
week and discuss the ‘New Philosophy’ or ‘Experimental Philosophy’. These philosophical
inquiries were related to Physics, Anatomy, Geometry, Astronomy, ‘Natural Experiments’,
etc. There were similar activities in France, Germany, and other countries. The famous
14-5 Early History

Royal Society came into being on July 15, 1662. The prestigious French Academy of Sci-
ence also had its beginnings at about the same time. or shortly afterwards. The Russian
Academy of Science (1725) and the Berlin Academy of Science (1770) were opened later.
All these academies and their published transactions played a most significant role in the
growth of science in the 18th and the 19th centuries.
Robert Hooke (July 1635 - Mar. 1703) was a British experimentalist and skilled me-
chanic. He studied in Oxford and became the curator of the experiments of the Royal
Society. There his inventive skill was of much help to the Royal Society. He used springs,
and using his innovative ability, carried out the famous experiments that led to the Hooke’s
law. In 1664 he became Professor of Geometry in Greesham College, but he continued to
present to the Royal Society his inventions, descriptions and experiments. His paper De
Potentiâ Restitutiva (or “Of Spring”) published in 1678 was the first published paper that
discussed the elastic properties of materials. The linear relation between the force and the
resulting elongation is the famous Hooke’s law which is one the three pillars of the mechan-
ics of deformable elastic bodies. Hooke also has a clear idea of universal gravitation, and
came close to the laws of attraction and motion.
Edme Mariotte (c. 1620 - May 1684) was one of the earliest (1966) members of the
famous French Academy of Science. Experimental methods became part of French science
mainly because of Mariotte. It was he who studied impact. He used balls suspended
by threads, and demonstrated that the momentum is conserved in impacts. The ballistic
pendulum was also invented by him.
As he had to design pipe lines for water supply, he studied the bending strength of
beams. He developed his own theory of bending taking into account the elastic properties
of materials. Starting with simple tension tests, he performed several experiments.
The next important contributions are of the Bernoullis. Euler was closely associated
with the Bernoullis; they all belonged to Basel, which is almost where France, Germany
and Switzerland meet. But Euler lived in the eighteenth century.
Jakob and Johann Bernoullis

(a) Jakob Bernoulli (b) Bernoulli family tree (c) Johann Bernoulli

Figure 14.3: Jakob [Fig. 14.3a] and Johann [Fig. 14.3c] Bernoullis, the first two mathe-
matician / scientists of the famous Bernoulli family [Fig. 14.3b]
Advanced Mechanics of Solids 14-6

Jakob (Jacob) (Dec. 1654 - August 1705) was the first famous person in the illustrious
Bernoulli family of mathematicians and scientists. Johann (John, also called Jean) (July
1667 - Jan. 1748) was his brother and pupil. They were both elected in 1699 by the
French Academy of Sciences as the foreign members. The Bernoulli family was always
represented in the Academy until 1790. The calculus8 progressed rapidly in Europe roughly
in the period 1670 - 1720 largely by these two Bernoulli brothers. Johann, the younger
Bernoulli was regarded as the greatest mathematician of his time. His lectures led to
the publication in 1696 of the first book on the calculus by L’Hôpital9 . The calculus of
variations was born when these two Bernoulli brothers quarrelled about a curve called the
brachistochrone10 which has become historically famous for this reason. Jakob investigated
the shape of the deflected curve of a beam. Although he made some erroneous assumption
— which is understandable — he did state correctly that the curvature of the deflected
curve is proportional to the bending moment. This fact was made use of by Euler and other
mathematicians. To Johann Bernoulli must go the credit of formulating the principle of
virtual work.
Euler and Daniel Bernoulli
Johann’s pupil Euler and son Daniel made far more significant contributions to our subject.
Although Daniel Bernoulli’s fame rests largely on his book Hydrodynamica, he did contribute
to the problem of determining the elastic curve. In a letter to Euler, Daniel suggested that
Euler should apply the calculus of variations to obtain the deflection curve of loaded beams.
It was Daniel who obtained the differential equation for the lateral vibration of prismatic
bars. He studied some particular modes of vibration. Euler integrated this differential
equation. Daniel was also good at doing experiments. He performed several experiments for
verification, and he was happy that “· · · I have performed a great many experiments, · · · ”.
Some of Daniel’s experiments were the feeding ground for some of Euler’s mathematical
problems.
Euler
Euler’s contributions are large. He was an exceptional mathematician. He was very ver-
satile, and his output prodigious. It is said that Euler calculated “without any apparent
mental effort, just as birds fly and men breathe”. He entered the University of Basel, a
most important research centre of mathematics, in 1720 at the young age of 13, obtained his
master’s degree at 16, and published his first scientific research paper and participated in an
international competition for a prize offered by the French Academy of Sciences before he
was 20. Euler benefitted enormously not only by Johann Bernoulli’s brilliant lectures but
also from his private tuition classes once a week. When the Russian Academy of Sciences

8
There have always been controversies about the priority of Leibniz — Gottfried Wilhelm Leibniz (July
1646 - Nov. 1716), German mathematician; he is often regarded as the last of the universalists (who
‘knew everything’) — in continental Europe and Newton in England. These are not settled yet perhaps.
Nevertheless it is generally accepted that the calculus was invented by these two scientists independently.
9
Marquis de l’Hôpital (often spelt as L’Hospital) (Month? 1661 - Feb. 1704).
10
This curve turns out to be a cycloid.
14-7 Early History

(a) Daniel Bernoulli (b) Leonard Euler (c) Joseph-Louis Lagrange

Figure 14.4: D. Bernoulli, L. Euler and J.L. Lagrange

was opened in 1725 at St. Petersburg, and the Bernoullis, Nicholas and Daniel, both sons
of Johann, invited to be members of the new institute, they found a position for Euler there
as an associate. Euler went to St. Petersburg and was furiously at work. While working
there, he wrote his celebrated book on mechanics Mechanica sive motus scientia exposita, 2
vols, St. Petersburg (1736), the first book in which the calculus was extensively used. The
traditional geometrical methods of Newton were abandoned; instead differential equations
were derived and solved to discuss problems in dynamics. Later developments in mechanics
were driven by the tremendous influence that this book had on the growth of mechanics.
Euler turned his attention to the elastic curve, the deflected shape of a loaded beam.
Based on the suggestions of Daniel Bernoulli through correspondence, he investigated the
problems of bending and of lateral vibration of beams and the associated differential equa-
tions. Euler is even more famous for his work on the buckling of columns.
When Frederick II, known as Frederick the Great, became the king of Prussia in 1740, he
wanted to promote scientific studies and persuaded Euler to move to Berlin. Euler came to
Berlin in 1741 and became associated with the Prussian Academy and the Berlin Academy.
For the next 25 years he stayed in Berlin. It was during this period that he wrote first his
famous book Methodus inveniendi lineas curves · · · (1944), which was the first ever book
written on the calculus of variations. This was followed by Introduction to Calculus (1748),
Differential calculus, 2 vols (1755) and Integral calculus, 3 vols (1768 - 1770). All these
books were the leading lights for the mathematicians from the last part of the eighteenth
and the early part of the nineteenth centuries. In this sense, all the mathematicians may
be said to have been Euler’s students!
Catherine II became the empress of Russia in 1762, and persuaded Euler to return to St.
Petersburg in 1766 by giving him a much better offer. During the last phase of his career,
he became completely blind. He had lost one eye as early as in 1735. But it is surprising
that his mathematical prowess only increased. During this last phase of his career (1766 -
1783), he published more than 400 papers!
Euler was prolific. He wrote full research papers between the first (“dinner is ready”)
and the second (“dinner is getting cold”) calls from his wife! It is not possible to discuss
in detail his many contributions. Above all, he was great as a human being: modest, self-
effacing, very appreciative of others and always letting others take the credit of his work!
Let us pause here, remember Euler with pleasure, gratitude and admiration.
Advanced Mechanics of Solids 14-8

Lagrange
It is almost a crime not to bring up Lagrange’s name when Euler’s life and work are men-
tioned. Lagrange was born in Turin (now in Italy). He became a professor of mathematics
when he was only 19! Euler was highly impressed with Lagrange. He helped Euler in
building up the calculus of variations. Based on Euler’s recommendation, Lagrange was
nominated as a foreign member of the Berlin Academy. Lagrange is best known for his
famous book Mécanique analytique in the preface of which he declares that there are no
figures in this book. This book was published in Paris only in 1788. This book exerted
a great influence on subsequent developments in mechanics. Here is an approach quite
different from the Newtonian mechanics. Lagrangian mechanics, based on generalised coor-
dinates, generalised velocities and forces, and the recently developed variational methods,
gave mechanics an entirely different face.
He, in his capacity as the judge of the entries of a prize problem, was the one to present
the governing equation (the biharmonic equation) for the bending of plates. Sophie Germain
competed for this prize, but there were mistakes, and it was Lagrange who corrected them.
In spite of the mistakes Sophie Germain was finally given the prize on her fourth attempt!
Lagrange was in Berlin until 1787 when, consequent on the death of Frederick the Great,
the conditions in Berlin became not as congenial, he moved to Paris where he was given
a warm welcome. His “most important contribution to the theory of elastic curve”, states
Timoshenko [14], “is his memoir Sur la figure des columnes”. Lagrange considers an axially
loaded column, and obtains the critical loads. Further, he examines the deflection when the
load exceeds the critical load. He considers columns of variable cross-section also.
For a short while because of overwork, he temporarily lost interest in mathematics.
It is said that his printed book Mécanique analytique lay unopened for two years. He
was attracted by other sciences. He also served on a commission that was discussing the
possibility of introducing the metric system in France. France then was going through
tumultuous times. Following the French revolution, the new government tended to look
upon scientists with suspicion. It removed some members of the aforesaid commission.
Lavoisier, a chemist and Bailly, an astronomer were executed. Lagrange got disgusted with
all this, and was about to leave France. But he was asked to lecture on the calculus at the
newly established École Polytechnique. Towards the end of his career, Lagrange tried to
revise his book, but he died in 1813 before the revision could be completed.
Hamilton11 calls Lagrange as the Shakespeare of mathematics!
EIGHTEENTH CENTURY

We are already familiar with the work of Galileo, Hooke and Mariotte in the 17th century.
The scientific work of the previous century found applications. Scientific methods were
gradually introduced into the various fields of engineering. New advances in military and
structural engineering demanded not only practical knowledge, but also the scientific ability

11
William Rowan Hamilton (Aug. 1805 - Sept. 1865), Irish mathematician who became a professor of
mathematics when he was still an undergraduate student!
14-9 Early History

to analyse new problems. Thus, in this background several engineering schools were opened,
and the first books on structural engineering published to serve as the textbooks in these
schools. The mechanics of solids and structural engineering got a big boost and progressed
considerably. France was well ahead of all other countries.
In 1720 several military schools were opened in France to train qualified engineers for
artillery and fortifications. Belidor12 published a book on mathematics for use in these
schools. This book contained not only mathematics, but also its applications to artillery,
geodesy and mechanics. Belidor’s book contains only elementary mathematics, but he
recommends the book Analyse des infiniment petits by L’Hôpital. Mathematics was liberally
used in more and more engineering problems, and this tendency grew rapidly. This was the
general scene in France.
Parent13 brought forth two memoirs in 1713, both relating to the bending of beams. It
took more than fifty years before further progress was made. The reason was perhaps that
Parent’s writings were not easy to read, and moreover that he was not too popular as he
was harshly critical of the work of others.
A great deal of experimental work of direct practical use was carried out in the 18th
century. This information is crucial to assessing the strength of structural materials. Among
these scientists mention must be made about Réaumur14 and Mussechenbroek15 . There was
much progress in the theory of retaining walls and arches in the 18th century.
Coulomb
Coulomb16 had his early education in Paris. Thereafter he joined the military corp of engi-
neers and was sent to an island where he was given the responsibility of various construction
activities. This job led him to study the mechanical properties of materials. Various struc-
tural engineering problems also engaged his attention. He worked in this island for nine
years. During this time he presented a famous paper to the French Academy of Sciences.
This was in 1773. After he returned to France, he worked as an engineer. He first won a
prize in 1779, jointly with another person (Van Swindes) from the Academy, for the con-
struction of a compass. Again in 1781 he won the Academy prize for his memoir Theoérie
des machines simples. After this Coulomb was permanently at the Academy in Paris when
he was elected as a member of the Academy. The facilities for scientific work there were
excellent.
He now turned his attention to electricity and magnetism. He invented a very sensitive
torsion balance to measure small electric and magnetic forces. It was this work that led to
his investigations on torsion (of circular prismatic bars, Coulomb’s theory). He devised an
experimental setup to study torsional vibrations.

12
Bernard Forest de Bélidor (1697 or 1698 - Sept. 1761), French engineer
13
Antoine Parent (Sept. 1666 - Sept. 1716), French engineer
14
René Antoine Ferchault de Réaumur (Feb. 1683 - Oct. 1757), Dutch engineer
15
Petrus (Pieter) van Mussechenbrock (Mar. 1692 - Sept. 1761), French scientist
16
Charles-Augustine de Coulomb (June 1736 - Aug. 1806), French scientist and engineer
Advanced Mechanics of Solids 14-10

When the French revolution17 erupted in 1789, Coulomb retired to his place. The
Academy was closed in 1793, but was reopened two years later, but with a different name,
L’Institut National des Sciences et des Arts. Coulomb was elected as a member. There
he continued his scientific work. His last papers dealt with the viscosity of fluids and with
magnetism. These were published in the memoirs of the new institute in 1801 and 1806. In
1802 he was appointed as one of the inspectors of study which entailed a lot of travel. He
was now concerned with the improvement of education.
The greatest contribution in the eighteenth century to the mechanics of solids was by
Coulomb. It is remarkable that his theories of friction, torsion and strength of structural
materials still stand, and are still used.
École Polytechnique

Figure 14.5: École Polytechnique: three pictures

The École Polytechnique18 has played a crucial role in the development of our subject.
The new institute had a vision quite different from that of the existing technical institu-
tions. Admission was made open to students from all classes of society. To get the best
students an entrance examination was introduced. The most accomplished scientists were
appointed as the faculty members. Almost all of the famous French scientists were either
students, or teachers, or both of this institute which immediately became famous. There
were marked revolutionary changes from the existing practices. Great emphasis was given
to basic, fundamental sciences. Students were given a very strong dose of mathematics,
physics, chemistry and mechanics, and not much of anything else19 . All were mathematical
/ analytical subjects taught by the most outstanding scientists. Lectures were delivered to
large classes, but they were supplemented by small groups of twenty students supervised by
younger teachers. This is the same as the small tutorial classes where the students do work
supervised by teachers. The idea is to develop the ability to learn. Learning how to learn
17
1789 - 1799; the revolution ended when Napolean Boneparte took power in 1799. King Louis XVI (1754
- 1793) on January 21, 1793 and later the queen, his wife, Marie Antoinette on October 16, 1793 were
executed. According to the available records, over 17,000 people were officially tried and executed, 16,594
with the guillotine, and an unknown number of others, perhaps more than 40,000, died in prison or without
trial.
18
Founded on March 11, 1794 by Gaspard Monge. It was right in Paris. In 1976 it was moved to a Parisian
suburb, Palaiseau, about 30 km from Paris. The Laboratorie de Méchanique des Solides (LMS) is perhaps
the part of the École Polytechnique that has the greatest relevance to us in this context.
19
It must be conceded that in those days there was no electrical engineering nor computer science. There
were no ‘muck’ subjects.
14-11 Early History

(whatever else will be needed in later life) was the mantra. École Polytechnique had this
unique stamp. History records that the products of this institute became highly successful
engineers. The institute was a great training ground not only for engineers, but also for
teachers and professors. This general style of focussing on developing the ability to learn
has continued to this day. Timoshenko [14] remarks: “It seems that this was the first time
that laboratory work was introduced as part of a teaching program.” This institute played
a pivotal role in the growth of engineering science.
Needless to state, several universities like the Polytechnical Institute in Vienna and the
Polytechnical Institute in Zurich followed suit. Russia copied this French system. Many
German universities also took the best out of this and adapted this to suit their conditions
of giving far greater participation to private industries.

(a) Claude-Louis Navier (b) Gustav Robert Kirchhoff (c) Gabriel Lamé

Figure 14.6: Three great scientists who contributed a great deal to our subject

SUBSEQUENT DEVELOPMENTS
Subsequent developments were very rapid, varied and almost explosive20 . We shall, there-
fore, refrain from carrying on with the story. We had stated right in the beginning of these
historical notes that we propose to have a broad look only of the early progress. Our subject
is almost an inseparable mix of strength of materials, structural engineering and the the-
ory of elasticity. We can see that France had played the leading role in the early days; the
French scientists were clearly far ahead of other countries. By about the nineteenth century,
the scene shifted gradually to Germany. This does not mean that there was nothing of note
in France, England and other countries.
Again in recent times, the face of our subject has changed and is still changing. Con-
tinuum mechanics, rational mechanics, nonlinear theory of elasticity, micropolar elasticity,
thermoelasticity, etc. along one direction; theory of plasticity, viscoelasticity, hyperelas-
ticity, etc. in another; powerful numerical methods like Finite Difference Methods, Finite
Element Methods of various colours and hues in yet another; composites, fracture mechan-
ics, etc. in a new direction; and so on. All these place heavy demands on the mathematics
needed which becomes more and more abstract. And then there are methods of experimen-
tal stress analysis and sophisticated experimental techniques. The subject has opened up

20
Timoshenko [16] divides into separate chapters the history of strength of materials between 1800 - 1833
and 1833 - 1867, and considers in a separate chapter the history of the theory of elasticity.
Advanced Mechanics of Solids 14-12

immense possibilities. It is vigorously growing like all branches of science. Some familiar-
ity with these various disciplines will be beneficial, although it is clearly impossible to be
abreast in all these.
In a sense it is unfair to gloss over the great developments that happened in the later
part of the 18th century and later. But one has to make harsh decisions when leaving out
important matters and topics. With these words of apology, let us see the names of a
few outstanding scientists below and just a few lines about them. Navier21 , Lamé22 and
Kirchhoff23 were these three outstanding scientists who contributed much to our subject.
Navier
Although Coulomb had given in his famous memoir (1773) the correct solutions of several
problems of technical importance, engineers had to wait for a long period of forty years to
understand them and use them for practical purposes. Further progress was made by Navier.
When he was 14, he lost his father, and then he lived with his uncle Gauthey who was a
famous French engineer. The uncle gave his nephew much practical knowledge about bridge
and channel construction. Thus, when he graduated first from École Polytechnique in 1804,
and then from École des Ponts in 1808, he was in an ideal position to study theoretical
matters related to practical engineering problems. When his uncle died in 1807, Navier
completed his uncle’s incomplete book in three volumes in 1809, 1813 and 1816. He had
added several important notes to update his uncle’s book. His own (Navier’s) famous book
on strength of materials appeared in print in 1826. The book expectedly had a profound
influence on the course of development of our subject. In this book he had discussed, among
many other topics, retaining walls, arches, plates and trusses also. In 1820 he presented
a memoir on the bending of plates. In 1821 appeared his famous paper in which all the
fundamental equations of the theory of elasticity were given. In 1830 he became a professor
of the calculus and mechanics at the École Polytechnique.
Kirchhoff
Gustav Robert Kirchhoff was born in Königsberg. He entered the Königsberg University,
where he became a student of Franz Neumann, a great name in those days. He was in
Heidelberg along with the chemist Bunsen and the great Helmholtz, and instrumental in
establishing Heidelberg as the centre of great scientific activity. In 1845 he announced his
famous Kirchhoff’s laws, known sometimes as KVL and KCL, which are of fundamental
importance in electrical engineering.
His contributions to our subject are many: Love-Kirchhoff assumption, proof of the
uniqueness theorem, obtaining the correct boundary conditions for a plate using variational
methods (‘Kirchhoff shear’). This is only one part of his scientific achievements.
21
Claude-Louis Navier (Feb. 1785 - Aug. 1836), French engineer and physicist who specialised in mechanics.
His name, sure enough, is included in the 72 names inscribed on the Eiffel Tower.
22
Gabriel Leon Jean Bapiste Lamé (July 1795 - May 1870) “was the consummate engineer-mathematician.”
“French mathematicians considered him to be too practical, and French scientists too theoretical.”
23
Gustav Robert Kirchhoff (Mar. 1824 - Oct. 1887), German physicist of Königsberg, Kingdom of Prussia
(now in Russia)
14-13 Early History

He moved to Berlin University in 1857. In 1859 he published some of his well known
papers. His famous book on mechanics, which is the first volume of his lectures on theoretical
physics Forlesungen über mathematische Physik, Mechanik appeared in 1876. He was an
excellent teacher, very strong in both theoretical physics and experimental methods. His
health was seriously affected when he sustained a leg injury in an accident and had to use
a wheel-chair. His collected papers were published in 1882. His contributions are too many
to mention here. A great scientist indeed!
Lamé
Gabriel Lamé, like several distinguished French scientists and engineers, studied first at
École Polytechnique (1813 - 1817). He wrote research papers even when he was an under-
graduate student there. Later he studied at École des Mines graduating in 1820. At this
time, he published his new method of calculating the angles between the faces of crystals.
On a request from the Emperor of Russia, Alexander I, France sent Navier and Émile
Clapeyron to work in St. Petersburg. The events in France had convinced the Emperor of
the importance of scientific research. He went to Russia along with his colleague in 1820.
Lamé was appointed as professor and engineer. During his 12 years there, he published
several papers in French and Russian, some jointly with Clapeyron. He returned to Paris
in 1832, he and Clapeyron joined an engineering firm, but he was also Professor of Physics
at École Polytechnique. He was appointed as Chief Engineer in 1836. He was involved in
building railway lines in France. In 1836 he moved to the prestigious Sorbonne as Professor
of Mathematical Physics and Probability. He was elected to Academie Sciences in 1842.
Lamé had made several outstanding contributions including the extensive use of curvi-
linear coordinates. He was considered as the leading French mathematician by no less a
person than the great Gauss and others. He was a great engineer. Perhaps he was even
greater as a scientist in engineering mathematics and elasticity.

(a) C.A. Coulomb (b) Thomas Young (c) B.P.E. Clapeyron

Figure 14.7: Three great engineer-scientists who contributed a great deal to our subject.

Theodore von Kármán (May 1881 - May 1963), Geoffrey Ingram Taylor (March 1886
- June 1975) and Stephen Prokofyevich Timoshenko (Dec. 22, 1878 - May 29, 1972) are
three engineer-mathematician-scientists of relatively recent times.
At this stage we close our excursion to the past. The purpose of these few pages on the
historical development is not to produce a complete and comprehensive document giving all
Advanced Mechanics of Solids 14-14

(a) A.L. Cauchy (b) J.V. Poncelet (c) G. Monge (d) B. Saint-Venant

Figure 14.8: Four great engineer-mathematicians

(a) Th. von Kármán (b) G.I. Taylor (c) S.P. Timoshenko

Figure 14.9: Three giants of relatively recent times

the details (who did what, and when). It is merely to take the readers to a bygone era, and
to be in the company of the great scientists. Some of the young readers may be inspired
to attain higher academic levels. One hopes, therefore, that no apology is needed for not
covering important topics, eras, persons and achievements. It is not easy to achieve the
right balance in such a discussion.
We have come to the end of the book. At this stage, this author seeks the permission
of his readers to sign off.
Best wishes and happy learning!

“I keep six honest serving men

They taught me all I know;
Their names are
What and Why and When
And Where and How and Who.”
(Rudyard Kipling)
Bibliography

[1] Ameen, M.
Computational Elasticity, Narosa Publishing House, New Delhi, (2005).

[2] Crandall, S.H., Dahl, N.C., and Lardner, T.J.

An Introduction to the Mechanics of Solids, 3rd edition, McGraw-Hill Book Company,
Inc., New York, (2003).

[3] Den Hartog, J.P.

Strength of Materials, Dover Publications, Inc., New York, (1961).

[4] Den Hartog, J.P.

Advanced Strength of Materials, McGraw-Hill Book Company, Inc., New York, (1952).

[5] Durelli, A.J., Phillips, E.A., and Tsao, C.H.

An Introduction to the Theoretical and Experimental Analysis of Stress and Strain,
McGraw-Hill Book Company, Inc., New York, (1958).

[6] Fung, Y.C.

Foundations of Solid Mechanics, Prentice-Hall of India Pvt. Ltd, New Delhi, (1968).

[7] Hetéyni, M.
Beams on Elastic Foundations, University of Michigan Press, (1946).

[8] Langhaar, H.L.

Energy Methods in Applied Mechanics, John Wiley & Sons, Inc., New York, (1962).

[9] Nambudiripad, K.B.M.

Variational Methods in Engineering, Narosa Publishing House, New Delhi, (2015).

[10] Nambudiripad, K.B.M.

Cartesian Tensors and the Equations of Solid Mechanics, Narosa Publishing House,
New Delhi, (2017).
Bibliography B-2

[11] Popov, E.A.

Engineering Mechanics of Solids (2nd Edition), Pearson (1998).

[12] Seely, F.B., and Smith, J.O.

Advanced Mechanics of Materials, 2nd ed., John Wiley & Sons, Inc., New York, (1962).

[13] Srinath, L.S.

Advanced Strength of Solids, Tata McGraw-Hill Publishing
company, Inc., New Delhi (1980).

[14] Timoshenko, S.P.

History of Strength of Materials, McGraw-Hill Book Company, Inc., New York, (1953).

[15] Timoshenko, S.P., and Gere, J.M.

Mechanics of Materials, Nelson Thornes Ltd; 3rd edition (1991).

[16] Timoshenko, S.P., and Goodier, J.N.

Theory of Elasticity, 3rd ed., McGraw-Hill Book Company, Inc., New York, (1970).
Index

affine transformation, 6-8 curvilinear anisotropy, 7-6

Airy’s stress function, 9-3 cylindrical polar coordinate system, 5-12
anisotropic elasticity, 7-18
anisotropy, 7-18 deviatoric part, 4-38
autofrettage, 9-39 differential equation of equilibrium, 5-3
auxetic material, 7-2 distortion energy, 4-44
axisymmetric problem, 9-24 dummy index, 3-2

base vector, 8-3 eigenvalue, 4-21

Beltrami-Michell equation, 12-15 eigenvalue problem, 4-18
Betti- Maxwell reciprocal relation, 10-12 eigenvector, 4-21
biharmonic function, 12-23 Einstein’s conventions, 3-2
Birnie’s equation, 9-34 elastic symmetry, 7-7
Bredt’s formula, 11-22 ellipse of stresses, 5-29
endurance limit, 1-9
calculus of variations, 12-20 Eulerian approach, 6-3
Castigliano’s first theorem, 10-16
Castigliano’s second theorem, 10-17 first invariant, 4-26
Cauchy’s result, 4-12 free index, 3-2
Cauchy’s stress quadric, 5-29
Gaussian quadrature, 2-21
Cauchy-Riemann equations, 11-9
generalised Hooke’s law, 7-1
Cayley-Hamilton theorem, 7-3
geometric nonlinearity, 7-3
characteristic equation, 4-20
Gram-Schmidt procedure, 4-22
Clavarino’s equation, 9-34
Guest’s criterion, 4-43
compatibility equation, 12-8
complementary energy, 10-4 Hermitian symmetric matrix, 4-20
complementary virtual work, 10-15 homogeneous deformation, 6-6
compliance matrix, 7-4 hydrostatic state of stress, 4-38
compliance tensor, 12-10
constitutive equation, 7-1 impulse function, 12-26
constitutive matrix, 7-4 index notation, 3-1
constitutive tensor, 12-11 integrability conditions, 12-7
constraint condition, 4-31 interference fit, 9-24
coordinate function, 11-12 interference fit calculations, 9-36
Coulomb’s theory, 11-2 invariant, 4-6
couple stress, 4-4 inverse, 3-15
curvature tensor, 4-38 isotropic, 7-18

I-1
Index I-2

isotropic elasticity, 12-12 orthogonal matrix, 5-25

isotropic material, 7-9 orthotropy, 1-19
isotropic point, 5-19
isotropic tensor, 4-22 permutation identity, 3-14
isotropy, 1-19 permutation symbol, 3-9
plane strain, 5-27
Jacobian, 8-3 plane strain problem, 5-18
plane stress problem, 5-18
Kelvin fluid flow analogy, 11-15 plastic hinge, 2-6
Kirsch’s solution, 9-18 Poisson’s equation, 11-11
Kronecker delta, 3-3 positive definite, 8-13
Lagrange multiplier method, 4-32 prescribed boundary condition, 12-20
Lagrange parameter, 4-32 principle of consistent deformations, 12-21
Lagrange’s coordinates, 6-3 principle of minimum total potential, 7-5
Lagrangian approach, 6-3 principle of superposition, 8-6
Lamé’s constants, 7-13 propped cantilever, 12-21
Lamé’s ellipsoid of stresses, 4-22 pure shear, 4-39
Lamé’s stress ellipse, 5-28 Saint-Venant’s principle, 8-5
Lamé’s thick cylinder problem, 9-27 sand heap analogy, 11-19
Lamé-Navier equations, 12-10 secondary variable, 12-20
Laplace’s equation, 5-19 semi-infinite region, 1-24
Legendre transformation, 10-4 semi-inverse method, 11-7
Levi-Civita symbol, 3-9 shear centre, 2-14
linearly independent, 11-12 shear flow, 11-21
local approach, 6-3 simply connected, 11-10
local derivative, 6-5 spherical part, 4-38
Love-Kirchhoff assumption, 2-2 spherical symmetry, 1-21
Maple, 11-13 stiffness matrix, 7-4
material approach, 6-3 stiffness tensor, 12-10
material nonlinearity, 7-3 strain energy, 10-2
metric, 8-3 strain energy function, 8-8
metric space, 3-21 stress concentration, 9-18
minimum complementary energy, 7-5 stress invariant, 13-6
minimum total potential theorem, 8-13 substantial approach, 6-3
Mohr’s circle, 5-32 substitution operator, 3-4
Myosotis method, 1-4 summation convention, 3-1

natural boundary condition, 12-20 Taylor & Quinney experiment, 4-46

natural strain, 12-7 theorem of work, 13-46
Navier’s equation, 12-2 theory of failure, 4-40
Navier-Stokes equation, 1-22 Timoshenko beam, 10-9
Timoshenko, S.P., 1-25
octahedral plane, 4-34 traction boundary condition, 5-18
one-to-one mapping, 6-8 transtropic material, 7-18
I-3 Early History

transtropy, 7-6 variational method, 12-2

transverse isotropy, 7-9 virtual work, 10-15
Tresca’s criterion, 4-43 von Mises’ criterion, 4-44
triple scalar product, 3-13
true strain, 12-7 Westergaard stress space, 4-38
Winkler constant, 2-9
uniqueness theorem, 8-8 Winkler foundation, 2-9
unit-load method, 10-21 Winkler-Bach formula, 2-21