An Introduction To Structural Mechanics For Architects

Structural Integrity 4
Series Editors: José A. F. O. Correia · Abílio M. P. De Jesus
Elías Cueto
David González
An Introduction
to Structural
Mechanics for
Architects
Structural Integrity
Volume 4
Series editors
José A. F. O. Correia, Faculty of Engineering, University of Porto, Porto, Portugal
Abílio M. P. De Jesus, Faculty of Engineering, University of Porto, Porto, Portugal
Advisory editors
Majid Reza Ayatollahi, School of Mechanical Engineering, Iran University of
Science & Technology, Tehran, Iran
Filippo Berto, Department of Mechanical and Industrial Engineering, Faculty of
Engineering, NTNU, Trondheim, Norway
Alfonso Fernández-Canteli, Faculty of Engineering, University of Oviedo,
Gijón, Spain
Matthew Hebdon, Virginia State University, Virginia Tech, Blacksburg, VA, USA
Andrei Kotousov, School of Mechanical Engineering, University of Adelaide,
Adelaide, SA, Australia
Grzegorz Lesiuk, Faculty of Mechanical Engineering, Wrocław University of
Science and Technology, Wrocław, Poland
Yukitaka Murakami, Faculty of Engineering, Kyushu University, Higashiku,
Fukuoka, Japan
Hermes Carvalho, Federal University of Minas Gerais, Belo Horizonte, Minas
Gerais, Brazil
Shun-Peng Zhu, University of Electronic Science and Technology of China,
Chengdu, Sichuan, China
The Structural Integrity book series is a high level academic and professional series
publishing research on all areas of Structural Integrity. It promotes and expedites
the dissemination of new research results and tutorial views in the structural
integrity field.
The Series publishes research monographs, professional books, handbooks,
edited volumes and textbooks with worldwide distribution to engineers,
researchers, educators, professionals and libraries.
Topics of interested include but are not limited to:
– Structural integrity
– Structural durability
– Degradation and conservation of materials and structures
– Dynamic and seismic structural analysis
– Fatigue and fracture of materials and structures
– Risk analysis and safety of materials and structural mechanics
– Fracture Mechanics
– Damage mechanics
– Analytical and numerical simulation of materials and structures
– Computational mechanics
– Structural design methodology
– Experimental methods applied to structural integrity
– Multiaxial fatigue and complex loading effects of materials and structures
– Fatigue corrosion analysis
– Scale effects in the fatigue analysis of materials and structures
– Fatigue structural integrity
– Structural integrity in railway and highway systems
– Sustainable structural design
– Structural loads characterization
– Structural health monitoring
– Adhesives connections integrity
– Rock and soil structural integrity
More information about this series at http://www.springer.com/series/15775

Elías Cueto David González
•
An Introduction to Structural
Mechanics for Architects
123
Elías Cueto David González
Aragón Institute of Engineering Research Department of Mechanical Engineering
University of Zaragoza University of Zaragoza
Zaragoza Zaragoza
Spain Spain
ISSN 2522-560X ISSN 2522-5618 (electronic)

Structural Integrity
ISBN 978-3-319-72934-3 ISBN 978-3-319-72935-0 (eBook)
https://doi.org/10.1007/978-3-319-72935-0
Library of Congress Control Number: 2018938396
© Springer International Publishing AG, part of Springer Nature 2018

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, express or implied, with respect to the material contained herein or
for any errors or omissions that may have been made. The publisher remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.
Printed on acid-free paper
This Springer imprint is published by the registered company Springer International Publishing AG
part of Springer Nature
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
To Susana and Mamen, for their patience.
Preface
This book contains a rather concise compilation of the contents of our lectures on
structural mechanics for the students of the second course of architecture at the
School of Engineering and Architecture of the University of Zaragoza. The contents
are intended to be taught during a semester comprising some 15 weeks and 3 h per
week. The student is assumed to have been exposed to introductory university
courses on physics and mathematics, in particular vector and tensor algebra and
differential calculus.
We believe that most of the literature devoted to structural mechanics has a
profound engineering bias. We decided to write this book just because we believe
that the approach followed herein is rather new. This is not a classical book on
strength of materials nor a classical introductory book on structural mechanics.
Instead, it follows a historical perspective and introduces the reader to the funda-
mental problem of architecture (at least, from the mechanical point of view),
namely avoiding bending. It was not until the Industrial Revolution that architects
began to have broad access to iron and steel, i.e., to materials with the same strength
in tension and compression. Until that moment, the history of architectural masters
was that of people trying to avoid bending. This is at the origin of trusses and also
of buttresses in Gothic cathedrals. This book follows this historical pipeline, by
treating the difficulties associated with stone arches as inverted cable structures
subjected to pure compression, then moving to bowstring structures and trusses,
then frames, and finally carefully introducing bending and its associated
complexity.
But we embark on this quest comme il faut, starting from the theory of elasticity,
and introducing properly the various hypotheses that give shape to strength of
materials as a collection of models derived from elasticity, rather than a collection
of engineering recipes with significant professional interest but no apparent links
between them. We repeat throughout the book the famous aphorism by G. Box: no
model is true, but some are useful. This book could be considered, therefore, as a
well-justified collection of lies.
vii
viii Preface
We therefore begin by a rapid review of the theory of elasticity. To this end, the
reader may find useful the appendix on index (also known as Einstein) notation.
This is because strength of materials is no more than a collection of simplifying
hypothesis established on top of the theory of elasticity that we find this approach
rigorous, yet very useful.
We then begin by studying cable structures, the only ones capable of bearing
traction loads in ancient times. It is precisely Hooke’s finding on the equivalence
between a stone arch and an inverted cable that gave rise to the possibility of
studying, from a scientific standpoint, the problem of finding the form of an arch.
Nevertheless, we incorporate in this chapter not only classical results, but also the
most recent developments in the field of masonry structures.
Bowstring structures emerge then as a natural continuation of arches and also a
seamless introduction to the world of trusses. With the eruption of industrialization
and the availability of first iron and then steel, the possibility of designing structures
subjected to bending arose, thus opening the door for sophisticated frame structures.
Every chapter introduces some worked examples. Some others are left deliber-
ately open so as to allow the reader to further investigate the contents of the book.
None of them are especially difficult, so the interested reader may find a solution in
various, well-known references. It is precisely this (re-)search that will enable the
reader to become truly a master of the field.
We hope the reader will find the book useful and apologize in advance for the
possible mistakes it may contain, despite the effort paid in successive lectures.
Zaragoza, Spain Elías Cueto

November 2017 David González
Acknowledgements
We are indebted to the many individuals who have contributed, directly or indi-
rectly, to this book. Particularly, Prof. Alfonso F. Canteli transmitted to us the
tradition, possibly coming from the ETH Zurich, of employing the lower bound
theorem and the stress-field method to achieve a profound understanding of the
behavior of structures. This book takes advantage of this tradition and employs
graphic statics throughout. We truly believe that this method can be employed
successfully to teach undergraduate students and transmit a very intuitive under-
standing of how structures work.
Professor Manuel Doblaré infected us with the passion for rigor, and we have
tried to be the best disciples in this hard task. Sections devoted to beams are
undoubtedly influenced by his work in the field, following the inspiring tradition
of the late Prof. Juan C. Simó.
We must thank, in sum, the faculty at the continuum mechanics division of the
School of Engineering and Architecture of the University of Zaragoza, because it is
by means of sharing good moments over a cup of coffee or around a blackboard that
we finally learn a way of doing things. We are indebted particularly to Profs. Icíar
Alfaro, Begoña Calvo, Fany Peña, and Miguel Ángel Martínez for their help and
support through the years.
Finally, thanks to our students, the interaction with whom provided the most
efficient means of determining the best practices to reach the goal we pursue.
ix
Contents
1 An Introduction to Continuum Mechanics . . . . . . . . . . . . . . . . . . . . 1

1.1 Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.1 Movement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.2 Hypothesis of Small Displacements . . . . . . . . . . . . . . . . . 2
1.1.3 Deformation Gradient . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.1.4 Displacement Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.1.5 Strain Ensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.1.6 Compatibility Equations . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2 Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2.1 Stress Vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2.2 Stress Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2.3 Principal Stresses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.2.4 Equilibrium Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.3 Constitutive Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.3.1 Linear Elastic Materials . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.3.2 Symmetries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3.3 Hooke’s Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.3.4 Limits of Hooke’s Law . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.4 Navier Equations for the Elasticity Problem . . . . . . . . . . . . . . . . . 19
2 Cable Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.2 Main Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.3 Equilibrium in Cables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.3.1 Influence of the Rise . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.3.2 Multiple Loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.3.3 Cables with Arbitrary Loads . . . . . . . . . . . . . . . . . . . . . . . 40
2.4 Funicular Curves for Distributed Loads . . . . . . . . . . . . . . . . . . . . 41
xi
xii Contents
2.5 Catenary Curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

2.6 Sizing of Cables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.7 Deflections in the Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3 Arches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.2 A Brief Historical Perspective . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.3 Main Hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.4 Limit Analysis of Masonry Arches . . . . . . . . . . . . . . . . . . . . . . . 61
3.5 Modern Arch Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.6 Shells, Vaults and Domes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.7 Arch-Cables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4 Trusses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.2 Hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.3 Statically Determinate and Indeterminate Trusses . . . . . . . . . . . . . 78
4.4 Limit Analysis of Statically Determinate Trusses . . . . . . . . . . . . . 81
4.4.1 Graphic Statics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.4.2 An Analytic Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.4.3 Method of Sections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.5 Serviceability Limit-State Analysis of Trusses . . . . . . . . . . . . . . . 87
4.6 Criteria for the Design of Trusses . . . . . . . . . . . . . . . . . . . . . . . . 92
4.6.1 Influence of the Height and the Span . . . . . . . . . . . . . . . . 92
4.6.2 Influence of the Orientation of Diagonals . . . . . . . . . . . . . 93
4.6.3 Various Designs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.7 Space Trusses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
5 Beams (I): Geometric Description and Internal Forces . . . . . . . . . . . 103
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
5.2 Geometrical Description of a Beam . . . . . . . . . . . . . . . . . . . . . . . 105
5.3 Internal Forces and Moments . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
5.4 Diagrams of Internal Forces and Moments . . . . . . . . . . . . . . . . . . 108
5.4.1 An Introductory Example . . . . . . . . . . . . . . . . . . . . . . . . . 109
5.4.2 A Second Introductory Example . . . . . . . . . . . . . . . . . . . . 112
5.5 Gerber Beams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
5.6 Frames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
5.7 Vierendeel Beams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
6 Beams (II). Normal Stresses . . . . . . ....... . . . . . . . . . . . . . . . . . . 127
6.1 Introduction . . . . . . . . . . . . . . . ....... . . . . . . . . . . . . . . . . . . 127
6.2 Euler–Bernoulli–Navier Bending Model . . . . . . . . . . . . . . . . . . . . 127
6.2.1 Main Hypotheses . . . . . . ....... . . . . . . . . . . . . . . . . . . 128
6.2.2 Kinematics . . . . . . . . . . ....... . . . . . . . . . . . . . . . . . . 129
6.2.3 Stresses . . . . . . . . . . . . . ....... . . . . . . . . . . . . . . . . . . 131
Contents xiii
6.2.4 Internal Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131

6.2.5 Stresses as a Function of Internal Forces . . . . . . . . . . . . . . 134
6.2.6 Neutral Axis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
6.2.7 Core of the Cross Section . . . . . . . . . . . . . . . . . . . . . . . . 138
6.3 Principle of Virtual Work for Bending Beams . . . . . . . . . . . . . . . 140
7 Beams (III). Shear Stresses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
7.2 Collignon Theory of Shear Stresses in Beams . . . . . . . . . . . . . . . 148
7.2.1 Hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
7.2.2 Collignon Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
7.3 Shear-Stress Distribution in Thin-Walled Profiles . . . . . . . . . . . . . 153
7.3.1 Shear Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
7.3.2 Shear Center . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
7.4 Deep Beams. Timoshenko Beam Theory . . . . . . . . . . . . . . . . . . . 157
7.4.1 Lower Bound Theorem Analysis . . . . . . . . . . . . . . . . . . . 158
7.4.2 Timoshenko Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
8 Beams (IV). Torsion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
8.2 Saint Venant Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
8.3 Prandtl Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
8.3.1 Torsion in Open, Thin-Walled Profiles . . . . . . . . . . . . . . . 179
8.3.2 Torsion in Closed, Thin-Walled Profiles . . . . . . . . . . . . . . 183
9 Statically Indeterminate Structures . . . . . . . . . . . . . . . . . . . . . . . . . . 191
9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
9.2 Degree of Indeterminacy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
9.3 Methods for the Analysis of Indeterminate Structures . . . . . . . . . . 194
9.3.1 Flexibility Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
9.3.2 Stiffness Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
9.4 Flexibility or Compatibility Methods . . . . . . . . . . . . . . . . . . . . . . 195
9.5 Stiffness or Equilibrium Methods . . . . . . . . . . . . . . . . . . . . . . . . . 197
9.5.1 Slope Deflection Method . . . . . . . . . . . . . . . . . . . . . . . . . 199
9.5.2 Moment Distribution or Cross Method . . . . . . . . . . . . . . . 200
9.6 The Direct Stiffness Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
9.7 The Finite Element Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
Appendix A: Indicial Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
Chapter 1
An Introduction to Continuum
Mechanics
Abstract This chapter revisits the basics of the theory of elasticity upon which
strength of materials is based. The level is kept to the minimum that will enable
reader to understand subsequent chapters.
Although in many good books strength of materials is developed after some of the
basics of classical Newtonian mechanics (equilibrium, basically, is all that you need
most of the times), it is best understood if we introduce, at least briefly, some basic
concepts of continuum mechanics. This will allow the reader to understand the very
basic hypotheses on which this discipline is based. The price to pay is a little bit of
complexity, but, trust us, you will feel much more comfortable from Chap. 2 onwards.
We will begin by acknowledging that the type of structures we, as architects
or engineers, are interested in, are examples of a continuous medium. Even if we
know that matter is discontinuous (it is formed of elementary particles, atoms an
molecules), at the scale of a building it appears to the human eye as continuous.
Therefore, it will be helpful to consider our structure as an homogeneous portion
of R3 . This is true in general, even if our material is clearly heterogeneous, like
reinforced concrete, for instance. We will be interested in its overall macroscopic
properties, rather than those of their components. Only very advanced theories that
fall outside the scope of this book make effective use of the heterogeneities of the
material through, normally, complex multiscale simulations on powerful computers.
We will also accept other fundamental hypothesis like Saint Venant’s principle,
which states that “the difference between the effects of two different but statically
equivalent loads becomes very small at sufficiently large distances from load” [14]. In
addition, we will restrict ourselves to static-load processes and isothermal conditions.
Under these assumptions, the question is: it is necessary to begin describing the
plethora of physical phenomena taking place in a structure, but …how? In this book
we decided to do it by describing first the way a structure moves: displacements.
There are plenty of good books that begin by introducing a concept coined as stress,
but it is unnatural at this moment, needs some advanced mathematics and, above all, it
is not as visible as the displacements! Displacements are a very intuitive concept and
will provide a useful means to introduce the remaining physical processes occurring
in a structure.
© Springer International Publishing AG, part of Springer Nature 2018 1

E. Cueto and D. González, An Introduction to Structural Mechanics for Architects,
Structural Integrity 4, https://doi.org/10.1007/978-3-319-72935-0_1
2 1 An Introduction to Continuum Mechanics
1.1 Kinematics
1.1.1 Movement
If we accept to begin the study of a structure by the description of its movement (as
already mentioned, this is a rather arbitrary choice, and many books prefer to begin
with stresses), let us first consider an arbitrary solid and call it Ω. To place its points
in space, we choose an arbitrary system of reference:
ϕ0 : Ω → ϕ0 (Ω) ⊂ R3 , ϕ0 (P) ≡ X, ϕ0i (P) = X i ,
with i = 1, 2, 3, that assigns to each point P in the solid its coordinates ϕ0 (P) ≡ X.
This ϕ0 function is called the reference configuration of the solid and represents, in
brief, the shape of the solid at rest, prior to the deformation.
After the application of the loads, the solid will inevitably deform (hopefully, in
an almost imperceptible manner, if we speak about structures). Therefore, it will be
necessary to define, at each instant in time, a new configuration that evolves therefore
continuously of the type
ϕ(t) ≡ ϕt : Ω → ϕt (Ω) ⊂ R3 , ϕt (P) ≡ x, ϕti (P) = xi ,
with t ∈ I ∈ R the considered time interval. As you will have noticed, we are denoting
by capital letters, i.e., X, magnitudes in the reference undeformed configuration,
while small letters like x denote magnitudes in the current updated configuration.
The situation can therefore be sketched as in Fig. 1.1. This convention is almost
universal and can be found in every book of continuum mechanics, except, probably,
the oldest ones.
It is important to know if a movement φ, the set of all configurations in time, is
regular or invertible, i.e., if ∀t, ∃φ−1
t : ϕt (Ω) → ϕ0 (Ω). This means that nothing
catastrophic has occurred, like discontinuities in the displacement field (also known
as cracks). Describing this type of phenomena is well beyond the scope of this book.
1.1.2 Hypothesis of Small Displacements
In introductory courses in structural mechanics, it is common to accept that the

displacements are small. This means nothing more than that any function evaluated in
the reference configuration, f (X) is almost identical to the function in the deformed
configuration, viz., f (X) ≈ f (x). In what follows, for instance, we will impose
equilibrium in the undeformed configuration, even if we know well that it may occur
in the deformed one (precisely the one that is unknown a priori). By accepting the
small displacement hypothesis, we are accepting that the difference will be negligible.
1.1 Kinematics 3
x3
x1
x2
ϕt1 (Ω)
ϕt1
φt1
X3
ϕ0
X2
Ω 1
X
ϕ0 (Ω)
ϕt2
φt2
ϕt2 (Ω)
Fig. 1.1 Time-dependent solid configurations. Do not confuse configurations (denoted usually by
the Greek letter ϕ) with movements (denoted by φ)
It is important to note, however, that not all of the structures of practical interest
can be analyzed under the assumption of small displacements. For instance,
Santiago Calatrava, the famous architect, devoted his doctoral dissertation
to the foldability of frames [4], see Fig. 1.2. The Spanish architect Emilio
Pérez Piñero also devoted the majority of his life to this type of structure. The
structural behavior of this type of frame changes drastically depending on the
configuration, from stable to a mechanism or vice versa.
Fig. 1.2 A view of the City of Sciences of Valencia, by S. Calatrava. L’hemisfèric, on the left, has
a foldable structure
Once the movement of a solid has been characterized, it is straightforward to define

its velocity or acceleration, by simply differentiating once or twice, respectively, with
respect to time:
∂φt
v : I × Ω → R3 , v(t, x) = (t, x),
∂t
and, equivalently for the acceleration:
∂ 2 φt
a : I × Ω → R3 , a(t, x) = (t, x).
∂t 2
However, as mentioned before, we will accept that the loading process is quasi-
static and therefore that accelerations are negligible. Phenomena like earthquakes,
for instance, are also beyond the scope of this work.
1.1.3 Deformation Gradient
Once the movement has been defined, it is necessary to ascertain what is the difference
between a rigid solid and a deformable one. Rigid solids also move. Therefore,
1.1 Kinematics 5
the difference should not be the possibility of establishing a movement φ. What

is noticeable here is the dependence of the movement on x, i.e., φ = φ(x, t). In
other words, that every point in the deformable solid can have a movement different
relative to its neighbors (but subject to some restrictions, as mentioned before: the
movement should be invertible and differentiable).
In other words, a movement that can be vary between neighboring points can also
be seen as a variation in movement along the solid. And the best way to see how a
magnitude varies is by computing its derivative along the coordinate axis. It seems
therefore natural that some kind of differentiation is needed to define the concept of
deformation.
Let us make a first attempt by just differentiating the movement with respect to
the reference coordinates, i.e.:
∂φti ∂xi
F : I × Ω ∈ R3 → R3 × R3 , F(t, x) ≡ Fi j = (t, x) = (t, x). (1.1)
∂X j ∂X j
It is important to note that, despite the hypothesis of small displacements, we

are not accepting that x ≈ X (otherwise the previous definition will make no
sense). What we defined as small displacements is the fact that any function
f takes almost the same value if evaluated in the reference or the deformed
configuration: f (X) ≈ f (x).
The magnitude defined in Eq. (1.1) is known as deformation gradient. It is impor-

tant to note that, since it arises after differentiation with respect to x, which has
three components, with respect to each of the components of X, this gives rise to
nine components. This type of magnitudes is known as a tensor of order two (it
has two indices) and also a two-point tensor (a tensor with an index defined on the
undeformed configuration, j, and the other on the deformed one, i 1 ).
However, it is a somewhat strange magnitude. Observe that, if we consider a rigid-
solid movement (and therefore with no deformation), what we obtain is F = I, the
identity matrix. It is strange or, at least, counterintuitive, that, despite that there is no
deformation, F takes on nonvanishing values. Let us try to shed light on the physical
meaning of F.
1 Following the previous criterion, many books employ capital letters for indices of magnitudes
defined in the reference configuration, and therefore F ≡ Fi J , i, J = 1, 2, 3.
1.1.4 Displacement Field
First of all, it is of very high importance to define the displacement field, as the
distance of every point from their reference configuration to the actual one:
u(t, X) = φt (t, X) − X = x(t, X) − X.
In the simplest case of Cartesian coordinates, this reduces to:
u i = xi − X i , i = 1, 2, 3 or, more generally, i = x, y, z.
It is worth noting that we speak of a field since the displacement takes a different
(vector, in this case) value at each of the infinite number of points of the solid and
at each point in time, so we should employ the notation u = u(x, t), see Fig. 1.3.
Sometimes, when there is no risk of confusion, we omit the dependence on x and t.
1.1.5 Strain Ensors
If we write the deformation gradient tensor as a function of the displacement field,

we arrive at:
∂xi ∂(u i + X i ) ∂u i
Fi j = = = + δi j = u i, j + δi j = h i j + δi j ,
∂X j ∂X j ∂X j
Fig. 1.3 Displacement x3

vector x1
x2
ϕt (P )
ϕt (Ω)
u(t, P )
X3
ϕ0 (P )
X2
1
X
ϕ0 (Ω)
1.1 Kinematics 7
where a comma “,” indicates a derivative with respect to the subsequent variable. We
therefore obtain a very clear interpretation: the deformation gradient is composed of:
• The contribution of a movement as rigid solid, i.e., without deformation. This part
is represented by the Kronecker-delta term δi j .
• The displacement gradient, not to be confused with the deformation gradient,
h i j = u i, j , which is a second-order tensor, too.
In turn, the displacement gradient of the solid can be decomposed as the sum of a
symmetric part and an antisymmetric one, viz.,
1 1
Fi j = δi j + u i, j = δi j + (u i, j + u j,i ) + (u i, j − u j,i ) = δi j + εi j + ωi j , (1.2)
2 2
that gives rise to the definition of the Cauchy’s strain tensor, also known as the
infinitesimal strain tensor.
What we have done is to decompose the strain as the contribution of a rotation

tensor
1
ωi j = (u i, j − u j,i ),
2
which is skew-symmetric, and a symmetric part,
1
εi j = (u i, j + u j,i ), (1.3)
2
giving rise to the so-called polar decomposition of the deformation gradient.
In vector notation, ∇u = ω + ε. It is common to denote ε = ∇ s u, to indicate
that it is the symmetric part of the displacement gradient.
Note that various strain measures are possible. For instance, the so-called Green–
Lagrange strain tensor is defined as:
1 T 1
E= (F F − I) = ∇ s u + (∇u · ∇u T ).
2 2
The difference between E and ε is precisely the high-order term 21 (∇u · ∇u T ). If

we accept that the strain is infinitesimal, ∇u · ∇u T ≈ 0, and we recover the Cauchy
strain tensor. This is called the hypothesis of small strains — not to be confused with
the hypothesis of small displacements, even if both are essential parts of the theory
of linear elasticity.
If we work in Cartesian coordinates (which is most often the case), then:
⎧ ⎫ ⎧ ⎫
⎨ u1 ⎬ ⎨ u x ⎬
u ≡ ui = u2 = u y ,
⎩ ⎭ ⎩ ⎭
u3 uz
and ⎛ ⎞ ⎛ ⎞
ε11 ε12 ε13 εx εx y εx z
ε ≡ εi j = ⎝ ε21 ε22 ε23 ⎠ = ⎝ ε yx ε y ε yz ⎠ .
ε31 ε32 ε33 εzx εzy εz
This leads to the most frequent expression for the components of the Cauchy strain
tensor:
∂u x
εx = ,
∂x
∂u y
εy = ,
∂y
∂u z
εz = ,
∂z
1 ∂u y ∂u x
εx y = ε yx = + ,
2 ∂x ∂y
1 ∂u z ∂u x
εx z = εzx = + ,
2 ∂x ∂z
1 ∂u z ∂u y
ε yz = εzy = + .
2 ∂y ∂z
These components have a clear physical interpretation. Those in the diagonal

entries of the matrix representation of ε represent the increment in length, per unit
length, of a segment starting from the point under consideration and oriented along
the coordinate axes, see Fig. 1.4. In turn, those entries out of the diagonal represent
half the angle described after deformation by segments oriented along the coordinate
axes.
There are also other magnitudes with a clear physical interpretation, such as the
volume increment per unit volume,
ΔV (1 + εx )d x(1 + ε y )dy(1 + εz )dz − d x d y dz

θ= = = εx + ε y + εz = εii .
V d x d y dz
1.1.6 Compatibility Equations
It is important to note that not every imaginable displacement field is valid from
a physical point of view (and especially under the conditions established before).
In particular, for this theory to be valid, every field defined in the solid must be
continuous and single-valued. Continuity refers to the fact that nothing catastrophic,
such as a crack, occurs. These can be imagined as the result of an interior point
splitting in two, thus giving rise to two new free surfaces. This type of phenomena
cannot be cast easily in the just introduced kinematic framework.
1.1 Kinematics 9
∂uy
uy + ∂y
dy
Y Y
∂ux
ux ux + ∂x
dx
dy dy
uy
dx X dx X
∂ux
ux + ∂y
dy
Y Y
ux
dy
dy
∂uy
uy + ∂x
dx
uy
X X
dx dx
Fig. 1.4 Physical interpretation of the strain components: a longitudinal; b transversal
The field must also be single-valued. Otherwise the results will be simply non-
sense: imagine for a moment a solid having, at the same time, two different displace-
ment fields. This is equivalent to obtaining a displacement field by integrating the
strains. Equivalently, observe that there are six strain-displacement relations, while
the number of components of the displacement field is three. For the resulting system
of equations to have a unique solution, it is enough to satisfy that:
R = ∇ × (∇ε) = 0,
where R is the so-called Saint Venant or compatibility tensor. Under large strain
conditions, an equivalent condition can be established for the deformation gradient,
∇ × F = 0.
1.2 Dynamics
Of course, once the description of the movement has been completed, it is important
to study the origins of these movements in terms of forces. We will consider the
possibility of volumetric forces b(x, y, z) (such as the force exerted by gravity,
inertial accelerations, …) and punctual ones, which we will call t̄. Here, the bar
indicates a prescribed known value.
1.2.1 Stress Vector
Imagine for a moment that we split the solid into two parts, by cutting it through a
plane, as in Fig. 1.5. Then, in the (infinitesimal) neighborhood of every point at the
newly-created free surfaces, a force per unit area will appear. This is called the stress
vector and measured in pressure units. These forces per unit area are equilibrated
at both sides of the slicing plane, but they have, in principle, an arbitrary direction,
dependent solely on the applied forces b and t̄ (they are not, in principle, parallel nor
perpendicular to the plane, …). What is remarkable about this fact is that we can split
the solid into two pieces by employing an infinite number of planes passing through
a given point. These planes will be defined by their normal vectors, n. For each value
of n, a different distribution of stress vectors t will be obtained. It seems that the
stress is something that shows a different aspect depending on the point from which
it is viewed. We will soon demonstrate that this is precisely one of the characteristics
of a second-order tensor.
Fig. 1.5 Stress vector

1.2 Dynamics 11
1.2.2 Stress Tensors
To ascertain the relationship between the chosen plane (defined by its normal, n) and
the resulting stress vector, consider Fig. 1.6. Here we see a tetrahedron whose three
faces are aligned with the coordinate axes, while the fourth is oriented in an arbitrary
direction. At every face, a stress vector appears. If we equilibrate all these vectors
along the axes’ directions, we obtain:
−txn d Sn + txx d Sx + txy d S y + txz d Sz = 0,

−t yn d Sn + t yx d Sx + t yy d S y + t yz d Sz = 0,
−tzn d Sn + tzx d Sx + tzy d S y + tzz d Sz = 0.
Among the four faces of the tetrahedron, we can establish the following relation-
ship by simply projecting onto the coordinate planes:
d Sx = d Sn n x , d S y = d Sn n y , d Sz = d Sn n z ,
such that
j
tin = ti n j .
What we observe in this equation is that a new magnitude with two indexes
j
appears, ti . We will see it in most textbooks as:
⎛ ⎞
σx τ yx τzx
σ = ⎝ τx y σ y τzy ⎠ .
τx z τ yz σz
Fig. 1.6 Obtention of the Z

Cauchy’s stress tensor
tn
ty n
tx
tz
X
σ is precisely the Cauchy stress tensor, named after Augustin Luis Cauchy, who
first established the relationship between the orientation of the cutting plane and the
resulting stress vector:
tin = σi j n j , or, simply, t = σn.
If we come back for a moment to the case of large displacements and strains,
several stress tensors can be defined. The first possibility is to relate the force
applied in the deformed configuration per unit area in the undeformed con-
figuration. This gives rise to the so-called first Piola–Kirchhoff (or nominal)
stress tensor, P = J σ F −T , where J = det F is the volume change during
deformation.
The second Piola–Kirchhoff stress tensor is defined as S = J F −1 σ F −T .
Note that while P is not symmetric, S is.
1.2.3 Principal Stresses
To clearly characterize the stress state of a point, it is very useful to look for some
direction n in which only normal stresses appear (those in the diagonal of the tensor).
This is equivalent to the classical procedure of diagonalizing a matrix. To this end,
consider the eigenvalue problem:

σi j − σδi j = 0,
which can easily be solved through the solution of a cubic equation expressed in terms
of the three invariants of the stress tensor (values that do not change, regardless of
the chosen normal n, thus representing a valid measure of the stress state in the solid)
σ 3 − I1 σ 2 + I2 σ − I3 = 0,
with
I1 = σ x + σ y + σ z ,

σ τ σ τ σ τ
I2 = y zy + x zx + x yx ,
τ yz σz τ x z σz τx y σ y
and
σx τ yx τzx

I3 = τx y σ y τzy .
τx z τ yz σz
1.2 Dynamics 13
1.2.4 Equilibrium Equations
To complete the picture, we need to identify what is the relationship between the
applied loads and the stresses developed by the solid. This relationship comes from
equilibrium. Consider the elementary volume of Fig. 1.7. By the simple application
of equilibrium of forces along each coordinate axes—remember the force per unit
area units of stresses—we have:
∂σx ∂τ yx
(σx + d x − σx )dydz + (τ yx + dy − τ yx )d xdz
∂x ∂y
∂τzx
+ (τzx + dz − τzx )d xd y + bx ρ d xd ydz − ρ d xd ydz ax = 0,
∂z
with ρ the specific weight of the solid and ax the acceleration along x-axis. After
discarding high-order terms, we have:
∂σx ∂τ yx ∂τx x
+ + + ρbx = ρax . (1.4)
∂x ∂y ∂z
σz + dσz
dx τzy + dτzy
τzx + dτzx σx
τxy
τyx τyz + dτyz
σy τxz
τxz + dτxz σy + dσy
τyx + dτyx
dz τyz τxy + dτxy
σx + dσx
τzx Y
τzy
σz
X
dy
Fig. 1.7 Equilibrium at an elementary volume element

A similar equation can be obtained by applying equilibrium along each one of the
coordinate axes.
The same can be done by applying equilibrium of moment with respect to an
arbitrary point. However, this will lead simply to:
σi j = σ ji , (1.5)
or, in other words, to the symmetry of the stress tensor. Remember that this does not
hold for the first Piola–Kirchhoff stress tensor, for instance, and is only valid under
small strain and displacement assumptions.
Equation (1.4) can be written alternatively as:
σ ji, j + ρbi = ρai , (1.6)
or, even simpler, in vector notation,
∇ · σ + ρb = ρa.
1.3 Constitutive Equations
So far, we have found what are the equations governing the relationship between
displacements and strains (compatibility equations) and equations governing the
relationship between external loads and stresses (equilibrium equations). To close
the circle, the reader will readily notice that we need to find the relationship between
strains and stresses. That is what we call constitutive equations. This relationship
depends on the particular material of interest and must be obtained experimentally.
A general form for this equation would be:
σ(X, t) = σ(X, φt0 →t , Θ, t),
i.e., any suitable form that relates stresses with the movement experienced by the
solid, possibly time dependence, temperature, time and, of course, the particular
material point under consideration, X, since the material may be nonhomogeneous.
We emphasize that the behavior of the material can be history-dependent, and that is
why we have included the dependence on the history of the movement, φt0 →t . This
is the typical case of plasticity, for instance. If this behavior is not history-dependent,
it can take a simpler form of the type:
σ(X, t) = σ(X, ε, ε̇, Θ, t),
which explicitly draws our attention to the fact that the behavior depends on strain,
but also on strain rate, that is, the velocity at which we impose deformations. This
1.3 Constitutive Equations 15
is the typical example of viscous materials. Simplifying further, the material can be
simply elastic, in which case its strain state depends solely on the strain state:
σ(X, t) = σ(X, ε, Θ, t).
For large strain applications, this can be simply stated as [3]
P = P(F, t),
i.e., the stress state (here represented by the first Piola–Kirchhoff stress tensor)
depends solely on the strain state, here represented by the deformation gradient
F.
1.3.1 Linear Elastic Materials
By far, the most used materials in construction—particularly, metals—show linear

elastic behavior during an important range of strains. This means that, upon appli-
cation of a load, they develop instantaneously a deformation, which disappears after
removal of the load, also instantaneously. This deformation is, in turn, a linear func-
tion of the stress, i.e.:

σi j = (σi j )0 + Ci jkl εkl − ε0kl − γi j Θ − Θ 0 , (1.7)
where values affected by the superscript (·)0 indicate some nonnull, reference value
for the variable of interest. Here, it is important to note that, due to the fact that
both strain and stress are second-order tensors, their constitutive relationship exists
in the form of a fourth-order tensor, C ≡ Ci jkl . Note that we employ, as in many
specialized books in the field, sans serif fonts for fourth-order tensors, to distinguish
from boldface second-order tensors.
Equation (1.7) receives the name Duhamel–Neumann’s law and first established
the origin of thermoelastic stresses in solids.
It is important to note that, for elastic solids, there is a strain density functional
from which we can derive the constitutive law. The existence of such a functional
guarantees that the stress-strain relationship does not depend on the history of the
movement. In this case, this functional takes the form:
1 1
V − V0 = Ci jkl (εi j − εi0j )(εkl − ε0kl ) − γi j (Θ − Θ 0 )(εi j − εi0j ) + (σi j )0 (εi j − εi0j ),
2 2
such that
∂V
σi j = .
∂εi j
The most frequent form of linear elasticity is expressed, however, as:
σi j = Ci jkl εkl , (1.8)
by simply neglecting thermal dependence, if the problem can be cast under isothermal
conditions, and also neglecting reference values.
In vector notation,
σ = C : ε,
where “:” indicates the twice-contracted tensor product, i.e., a product that, applied
to matrices, generates a scalar, and applied to this fourth-order times second-order
tensor product, generates a second-order tensor.
In the case of large strain settings, the so-called hyperelastic materials are such
that:
∂V
S= .
∂E
In this case, the second Piola–Kirchhoff stress tensor and the Green–Lagrange
strain tensor E are said to be conjugate.
1.3.2 Symmetries
The constitutive tensor C has four indices, i, j, k, l = 1, 2, 3. Therefore, the number

of entries is 3 × 3 × 3 × 3 = 81. However, not all of them are different, since mate-
rials usually exhibit various symmetries, and therefore so does C. For instance, since
the strain tensor is symmetric, Ci jkl = Ci jlk . Identically, since the stress tensor is sym-
metric, it must hold Ci jkl = C jikl . Finally, if the strain energy is path-independent
(something typical of elastic materials), then it must hold that:
∂2 V ∂2 V
= ,
∂εi j ∂εkl ∂εkl ∂εi j
or, equivalently:
Ci jkl = Ckli j ,
thus needing only 21 constants to be define C completely.

The situation is often even easier, and, for a material that is homogeneous (thus,
not dependent on X) and isotropic (this means that it shows the same characteristics
along any direction in space), it can be demonstrated that only two constants are
enough.
1.3 Constitutive Equations 17
1.3.3 Hooke’s Law
Hooke was the first to establish the linear relationship between strain and stress, so
this very simple constitutive equation was named after him. Its most frequent form
is:
σi j = λθδi j + 2Gεi j , (1.9)
with θ = εii = εx + ε y + εz . Here, λ and G are the two Lame’s coefficients, defined
as a function of the Young’s modulus E and Poisson coefficient ν as
Eν E
λ= , G= .
(1 + ν)(1 − 2ν) 2(1 + ν)
1.3.4 Limits of Hooke’s Law
Every constitutive law has its limits. For instance, no existing material presents linear
behavior independent of the level of stress. Some of them plastify, which means that
they acquire permanent deformations, or simply break in a brittle manner. However,
it is not evident how to know the precise stress state under which this will occur. Note
that stresses are indeed a tensorial magnitude, so several questions arise: do all six
different stress tensor components contribute equally to the failure of the material?
Does this occur at the same value?
Of course, the answers are no and no. The only way to ascertain this is by means
of costly experimental tests. For instance, if we subject a steel bar to a traction
test, it shows a load-displacement curve shown in Fig. 1.8. We first notice a clearly
linear part of the curve that, at some 15 kN, begins to loose linearity and, for small
increments in load, experiences large increments in displacement. This is the plastic
regime of the curve. The last part of the curve (the vertical tray at the end) represents
the register of the machine when the material is already broken and does no longer
provides any useful information.
For a single stress state it is therefore easy to determine the precise value at which
the constitutive equations is no longer valid. But what happens if we experience a
complex non-homogeneous stress state? Many researchers (Tresca, Rankine, et al.)
devoted efforts to try to understand this phenomenon and to supply a practical answer
to professionals.
Among their findings, we can mention:
• They found that the limits of Hooke’s law do not depend on hydrostatic pressure,
i.e., that pieces subject to stress states, such that the principal stresses σ I = σ I I =
σ I I I are all equal, do not produce any yielding.
Fig. 1.8 Load-displacement 20

curve for a low-carbon steel
sample
15
Applied force, kN
10
0 0.5 1 1.5 2 2.5 3

Strain, ε, %
• They also found that plastification depends solely on the second invariant of the
deviatoric stress tensor, in other words, on:
(σ I − σ I I )2 + (σ I I − σ I I I )2 + (σ I I I − σ I )2
J2 = .
6
Von Mises,2 who in 1913 established a failure criterion that has been widely
accepted since then for metals (particularly, steel and aluminium). This criterion
states that, for a complex stress state, the material fails when:
(σ I − σ I I )2 + (σ I I − σ I I I )2 + (σ I I I − σ I )2 = 2σe2 ,
where σe is the uniaxial stress at which the material also fails. The interest relies
precisely on the ease of determining this value with a single laboratory experiment.
In this way, Von Mises established a stress indicator to compare with the yield
stress σe arising from the lab tests. This stress, in an arbitrary coordinate system,
reads:
1
σVM
2
= (σ11 − σ22 )2 + (σ22 − σ33 )2 + (σ33 − σ11 )2 + 6(σ23
2
+ σ312
+ σ122
) ,
2
(1.10)
so that we simply need to verify that, under our design conditions, it holds that:
σVM ≤ σe ,
usually under some suitable safety factors.
2 Do not confuse with Ludwig Heinrich Edler von Mises, the famous Austrian economist advocate
of the free market, who was actually his brother. We refer to Richard Edler von Mises (1883–1953).
1.4 Navier Equations for the Elasticity Problem 19
1.4 Navier Equations for the Elasticity Problem
So far, we have explored all the equations arising in the definition of the elastic
problem, even their limits of validity. We have described the kinematic equations,
the dynamics and the constitutive equations. And, somehow inherently, we have
accepted that the load variables, b and t̄, are the data in the problem. However, we
have not even defined the unknowns of the problem.
One possibility is to decide that the stresses are the most important unknowns in
the problem (in view of the just introduced von Mises criterion) and to substitute
them in the kinematic equations with the help of constitutive equations. This leads
to the so-called Beltrami–Michell3 form of the elasticity problem. This is however
of little practical interest, in general, even though, at the end of this book, we will
develop a method for the practical analysis of structures based on this approach.
The most frequent form of the elastic problem is the so-called Navier form. In it,
we decide to choose the displacements as the unknowns of the problem. Therefore,
if we substitute the displacement-strain Eq. (1.3) into the constitutive Eq. (1.8) and
then into the equilibrium Eq. (1.6), we arrive at:
1
Ci jkl (u k,l + u l,k ) , j + ρbi = ρai in Ω,
2
u i = ū i on δΩiu ,
1
Ci jkl (u k,l + u l,k )n j = t¯i on δΩit ,
2
u i (x, 0) = ū i (x)
on Ω,
u̇ i (x, 0) = v̄i (x)
where all magnitudes affected by a bar indicate the prescribed values (boundary
and initial values for the problem). If the considered material is homogeneous (its
constitutive tensor is constant everywhere in Ω), we have:
1
Ci jkl (u k,l j + u l,k j ) + ρbi = ρai ,
2
and if it were, in addition, isotropic:
Ci jkl = λδi j δkl + 2Gδik δ jl ,
3 You can find a very interesting NASA report on the topic at https://ntrs.nasa.gov/archive/nasa/
casi.ntrs.nasa.gov/19950012494.pdf.
leading finally to:
λδi j u k,k j + G(δik u k,ll + δi j u k, jk ) + ρbi = ρai .
After some algebraic manipulation, we obtain finally:
(λ + G)u k,ki + Gu i,kk + ρbi = ρai ,
which is precisely the expression of the Navier form of the elastic problem. In vector
notation:
(λ + G)∇(∇ · u) + G∇ 2 u + ρb = ρa. (1.11)
This resulting equation is complex. Except in some very simple and academic
cases, no solution is known for it, even if existence and uniqueness theorems exist
that prove the validity of the approach followed therein. Nowadays, powerful compu-
tational methods, such a the finite element method, exist that are based on somewhat
similar forms of the problem. However, architects and engineers have been forced
historically to deal with these complex equations to design buildings. Their approach
is, however, a different one.
Traditionally, engineers have tried to simplify to some extent the Navier equations.
This has been done by accepting some simplifying assumptions and, at the same time,
working in a somewhat different manner. This gave rise to what we know as strength
of materials or mechanics of materials. Surprisingly, strength of materials has nothing
to do with the strength of the materials! It is a technological discipline composed by
a collection of hypotheses that allows one to simplify the elastic problem so as to be
solved (at least, approximately) easily and fast. To be employed in architectural and
engineering practice, in sum.
Very often, strength of materials is studied before the theory of elasticity, and
this hides the true hypothesis behind the resulting formulas, making it a sometimes
awkward collection of technical recipes with no apparent connection among them.
This is why we decided to begin this book by a (fast, very fast) overview of the
theory of elasticity, that will enable us to understand all the mechanisms behind the
well-known formulas and will allow us to dimension our structural members. This
fascinating journey begins in the next chapter.
Worked Examples
1.1. Let us consider the reference configuration of a three-dimensional solid Ω in a

Cartesian coordinate system. For each point on the body, {X, Y, Z }, the coordinates
on the deformed configuration are given by {x, y, z} defined by:
⎛ ⎞ ⎛ 2 ⎞
x X −Z
⎝y⎠ = ⎝ Y 3 − 2 ⎠
z X2 + Y 2
(a) Compute the deformation gradient, its inverse and its transpose.
(b) Obtain the deformation gradient as a function of the displacement field.
(c) For this movement, and taking into account the small displacements and small
strains hypothesis, determine the strain and rotation tensors.
(d) Particularizing the strain tensor expression for the point with Cartesian coordi-
nates P = [0, 0, 0]T , compute the principal strains and their associated principal
directions.
(e) Demonstrate that the volume increment is independent of the coordinate system
used to express the strain tensor.
(f) Given the value of the Young’s modulus E = 2.1 · 106 and Poisson’s coefficient
ν = 0.3, compute the Cauchy stress tensor σ(P) at point P = [0, 0, 0]T .
(g) Compute the principal stresses and their associated principal directions at P.
Solution 1.1. (a) Applying the definitions introduced in this chapter, it is clear that:
⎛ ⎞
2X 0 −1
∂xi
Fi j (X, Y, Z ) = = ⎝ 0 3Y 2 0 ⎠ .
∂X j 2X 2Y 0
The representative matrix of the transposed tensor is the transpose matrix of F,

FT .
The reference configuration in terms of the deformed one will be:
⎛ ⎞
⎛ ⎞
X z − 3
(y + 2) 2
⎜ √ ⎟
⎝Y ⎠ = ⎜ + ⎟
⎝ ⎠
3
y 2
Z
z − 3 (y + 2)2 − x
so that:
⎛ ⎞
0 −1√ √ 1
√
⎜ 3 y+2 z− (y+2) 2 z− (y+2) ⎟
3 3 2 3 2
∂X j ⎜ ⎟
(F −1
)i j (x, y, z) = =⎜
⎜0
√1 0 ⎟
⎟
∂xi ⎝
3 3 (y+2)2
⎠
−1 √−2 1
3 3 y+22
Notice that, by substituting the values of (x, y, z) as a function of (X, Y, Z ), we

obtain the inverse matrix of the deformation gradient:
⎛ ⎞
0 −2Y 3Y 2
1 ⎝ 0
F −1 (X, Y, Z ) = 2X 0 ⎠.
6X Y 2 −6X Y 2 −4X Y 6X Y 2
(b) We assume the origin of the coordinate system to be located at the point (0, 0, 0).
By taking into account that the reference and deformed configurations use Carte-
sian coordinate systems, we have
⎛ ⎞
X2 − Z − X
u = ⎝ Y3 − 2 − Y ⎠.
X2 + Y 2 − Z
Given the definitions above, the displacement gradient—not to be confused with

the deformation gradient—is:
⎛ ⎞
2X − 1 0 −1
h = ui, j =⎝ 0 3Y 2 − 1 0 ⎠ .
2X 2Y −1
so that: ⎛⎞
2X 0 −1
Fi j = δi j + h i j = ⎝ 0 3Y 2 0 ⎠ ,
2X 2Y 0
which provides, indeed, the same tensor previously obtained.

(c) Recall again the displacement field:
⎛ ⎞
X2 − Z − X
u = ⎝ Y3 − 2 − Y ⎠.
X2 + Y 2 − Z
The Cauchy strain tensor is given by:

⎛ ⎞
∂u x ∂u x ∂u y ∂u x ∂u z
∂x
1
∂y
+ ∂x
1
∂z
+ ∂x
⎜
2 2
⎟
⎜ 1 ∂u y ⎟
ε=⎜ ⎟.
∂u x ∂u y ∂u y ∂u z
⎜ 2 ∂x + +
1
∂y ∂y 2 ∂z ∂y ⎟
⎝ ⎠
1 ∂u z ∂u x ∂u y ∂u z ∂u z
2 ∂x
+ ∂z
1
2 ∂z
+ ∂y ∂z
The rotation tensor was defined as:

⎛ ⎞
1 ∂u x ∂u y ∂u x ∂u z
0 ∂y
− ∂x
1
∂z
− ∂x
⎜
2 2
⎟
⎜ 1 ∂u y ⎟
ω=⎜ ⎟.
∂u x ∂u y ∂u z
⎜ 2 ∂x − ∂ y −
1
0 2 ∂z ∂y ⎟
⎝ ⎠
1 ∂u z 1 ∂u y ∂u z
2 ∂x
− ∂u
∂z
x
2 ∂z
− ∂y
0
Therefore, by particularizing values:

⎛ ⎞
2X − 1 0 − 21 (2X + 1)
⎜ ⎟
ε=⎝ 0 3Y 2 − 1 Y ⎠,
− 21 (2X + 1) Y −1
⎛ ⎞
0 0 − 21 (1 + 2X )
⎜ ⎟
ω=⎝ 0 0 −Y ⎠.
1
2
(1 + 2X ) Y 0
(d) The strain tensor particularized at point [0, 0, 0]T is given by:
⎛ ⎞
−1 0 − 21
ε(P) = ⎝ 0 −1 0 ⎠ ,
− 21 0 −1
By solving the characteristic polynomial in order to obtain the principal strains,

we arrive at:
ε − εI = 0
so that we obtain the following eigenvalues (in increasing order),
−1 −3
εI = , ε I I = −1, ε I I I = .
2 2
To determine the principal directions, i.e., the eigenvectors, we must solve the
following system of equations:
(εi j − εδi j )n i = 0
where the three components of each principal direction must satisfy:
n 21 + n 22 + n 23 = 1,
i.e., the mandatory unity norm of each eigenvector. This gives rise to the following
diagonalized strain tensor:
⎛ 1 ⎞
−2 0 0
εprinc = ⎝ 0 −1 0 ⎠ ,
0 0 − 23
T T
where n I = √12 , 0, − √12 , n I I = (0, 1, 0)T and n I I I = √12 , 0, √12 .
Finally, it is easy to see that:
1ε X,Y,Z = (n I |n I I |n I I I ) ε I,I I,I I I (n I |n I I |n I I I )T

⎛ 1 ⎞⎛ ⎞ ⎛ √1 −1
⎞ ⎛ ⎞
√ 0 √1 − 21 0 0 0 √ −1 0 − 21
⎜ 2 2
⎟⎝ ⎜ 2 2
⎟
= ⎝ 0 1 0 ⎠ 0 −1 0 ⎠ ⎝ 0 1 0 ⎠ = ⎝ 0 −1 0 ⎠ .
−1
√
2
0 −3 2
0 0 − 23 √1 0 −3
2 2
− 21 0 −1
(e) The volume increment per unit volume or, equivalently, the trace of the principal
strain tensor is:
ΔV 1 3
= ε I + ε I I + ε I I I = − − 1 − = −3,
V 2 2
that is the same result obtained in a different system of reference, viz.,
ΔV
= ε X + εY + ε Z = −1 − 1 − 1 = −3.
V
Obviously, this same result can be obtained in any system of reference. This is
the meaning behind the name invariants of the strain tensor.
(f) Applying the Lamé constitutive equations, we can find the Cauchy stress ten-
sor expression for the point P, where we know the strain tensor ε. The Lamé
constitutive equation is:
σi j = 2Gεi j + λ(εx + ε y + εz )δi j ,
where
Eν
λ=
(1 + ν)(1 − 2ν)
and
E
2G =
1+ν
are the Lamé parameters.

Taking into account the values of the Young’s modulus E = 2.1 · 106 kp/cm2
and the Poisson’s coefficient ν = 0.3, we have:
λ = 1.212 · 106 kp/cm2
and
2G = 1.615 · 106 kp/cm2 .
With this result in mind, the stress tensor will be:

⎛ ⎞
−5251 0 −807.5
σ(P) = ⎝ 0 −5251 0 ⎠ · 103 kp/cm2 .
−807.5 0 −5251
(g) The stress tensor at [0, 0, 0]T has been obtained in (f):
⎛ ⎞
−5251 0 −807.5
σ(P) = ⎝ 0 −5251 0 ⎠ · 103 kp/cm2 .
−807.5 0 −5251
Solving the characteristic polynomial in order to obtain the principal stresses,

we obtain:
σ − σ I = 0,
which provides the next sorted eigenvalues,
σ I = −4.4435 · 106 kp/cm2 , σ I I = −5.251 · 106 kp/cm2 ,
σ I I I = −6.0585 · 106 kp/cm2 .
Solving the following system of equations for each principal strain, we obtain
the principal directions, i.e., the eigenvectors of the system:
(σi j − σδi j )n i = 0,
where the three components of each principal direction must satisfy again the
unity norm condition of every eigenvector
n 21 + n 22 + n 23 = 1.
Therefore:
⎛ ⎞
−4.4435 0 0
σ princ =⎝ 0 −5.251 0 ⎠ · 106 kp/cm2 ,
0 0 −6.0585
T T
where n I = √12 , 0, − √12 , n I I = (0, 1, 0)T and n I I I = √12 , 0, √12 . The
same directions that the same obtained for the principal strains. This follows
from the linearity of constitutive equations.
1.2. Two blocks, made of different materials but with the same dimensions 50 ×
50 × 50 cm3 , are placed between two infinitely rigid plates. These materials have
elastic, linear and isotropic behaviors with various Young’s moduli and Poisson
coefficients. Determine the minimum value of the forces P1 and P2 that allow the
top plate to remain horizontal, and allow, at the same time, the blocks to come into
contact. We assume that the bottom plate is fixed.
The initial distance between blocks is d = 1 cm, and there is no restriction for the
movement along the perpendicular direction to the plane of the problem. Neglect the
friction between various surfaces, and assume no temperature variations. Assuming
a uniform stress state for each block (produced by the parallel descent of the top
plate and the absence of friction), determine the:
(a) Condition to be fulfilled by both normal strains along x so as to produce contact
between the blocks.
(b) Condition to be fulfilled by both normal strains along y that allow the top plate
to remain horizontal.
(c) Stress and strain tensors for each block.
(d) P1 and P2 that make the contact between blocks possible, while keeping the top
plate horizontal (Fig. 1.9).
Data: E 1 = 1.1E6 kp/cm2 , ν1 = 0.34, E 2 = 7E5 kp/cm2 , ν2 = 0.33 and d =
1 cm.
50 cm 50 cm 1 cm 50 cm 50 cm
P1 P2
y
50 cm
x E1 , ν1 E2 , ν2
Fig. 1.9 Sketch for Example 2. Dashed lines reflect the final, deformed configuration of the top
plate and the two blocks
Solution 1.2. (a) Due to the descent of the top plate that remains always horizontal,
each block will increase its horizontal dimension by magnitude ΔL (i)
x , i = 1, 2.
This magnitude depends, obviously, on the particular constitutive material of
each block. Therefore, for them to come into contact, the necessary condition
will be:
ΔL (1) ΔL (2)
x
+ x
= d = 1 cm.
2 2
In the development of this condition, we have assumed that the center of gravity
of each block remains at its original position.
If we elaborate a little bit, the just obtained expression, by expressing the length
increments in terms of the strain at each block, we have
ε(1)
x · 50 ε(2) · 50
+ x =1
2 2
so that
2
ε(1) (2)
x + εx = = 0.04,
50
which is precisely the condition sought to be satisfied by both normal strains.
(b) If the upper plate remains horizontal, it is clear that we must satisfy ΔL (1)
y =
(2)
ΔL y .
In others words, since L (1) (2)
y = L y , it must hold
ε(1) (2)
y = εy ,
which is the second necessary condition.

(c) Due to the absence of friction, there will be no tangential stress. The stress and
strain tensors for each block will be, consequently:
• Block 1:
⎛ ⎞
σx(1) 0 0
σ 1 = ⎝ 0 σ (1)
y 0 ⎠
0 0 σz(1)
⎛ (1) ⎞
εx 0 0
ε1 = ⎝ 0 ε(1)
y 0 ⎠
0 0 ε(1)
z
• Block 2: ⎛ ⎞
σx(2) 0 0
σ 2 = ⎝ 0 σ (2)
y 0 ⎠
0 0 σz(2)
⎛ (2) ⎞
εx 0 0
ε2 = ⎝ 0 ε(2)
y 0 ⎠
0 0 ε(2)
z
In x and z directions, there is no external force applied to any block. This implies
that:
σ 1 · n = t = 0.
Particularizing along x-direction,

⎛ ⎞ ⎛ ⎞ ⎛ ⎞
σx(1) 0 0 1 0
⎝ 0 σ (1)
y 0 ⎠ · ⎝0⎠ = ⎝0⎠ ,
0 0 σz(1) 0 0
thus obtaining σx(1) = 0.

Particularizing in z-direction and also repeating this reasoning for the second
block, we obtain:
σz(1) = 0, σx(2) = 0, σz(2) = 0.
We thus arrive to a much more detailed expression for the stress tensors, while
the strain tensors remain unknown for the moment:
⎛ ⎞
0 0 0
σ 1 = ⎝0 σ (1) ⎠
y 0 ,
0 0 0
and ⎛ (1) ⎞
εx 0 0
ε1 = ⎝ 0 ε(1)
y 0 ⎠,
0 0 ε(1)
z
while, for the second block:

⎛ ⎞
0 0 0
σ 2 = ⎝0 σ (2) ⎠
y 0 ,
0 0 0
and ⎛ (2) ⎞
εx 0 0
ε2 = ⎝ 0 ε(2)
y 0 ⎠.
0 0 ε(2)
z
These results, together with the already obtained conditions
ε(1) (2)
x + εx = 0, 04,
and
ε(1) (2)
y = εy ,
and applying Hooke’s law for both blocks:
(1 + νi ) (i) ν (i) (i)

εi(i)j = σi j − σ δi j , i = 1, 2,
Ei E i kk
so that we have 8 equations for 8 unknown variables. Along the y-direction, we

have
σ (1)
ε(1)
y
y = ,
E1
and
σ (2)
ε(2)
y
y = .
E2
But we also know from question (b) that ε(1) (2)

y = ε y , so that
E 1 (2)
σ (1)
y = σ .
E2 y
Along the x-direction we have

ν1 (1)
ε(1)
x =− σ = ε(1)
z ,
E1 y
and ν2 (2)
ε(2)
x =− σ = ε(2)
z .
E2 y
And we know from question (a) that ε(1) (2)

x + εx = 0.04, so
ν1 (1) ν2 (2)
− σ − σ = 0.04.
E1 y E2 y
Substituting the expression for σ (1)

y , we arrive at
0, 04 · E 2
σ (2)
y =− = −41791.04 kp/cm2 .
ν 1 + ν2
Finally,
E 1 (2)
σ (1)
y = σ = −65671.64 kp/cm2 .
E2 y
This gives rise to the following stress and strain states:

⎛ ⎞
0 0 0
σ 1 = ⎝0 −65671.6 0⎠ kp/cm2 ,
0 0 0
⎛ ⎞
0.020 0 0
ε1 = ⎝ 0 −0.059 0 ⎠ .
0 0 0.020
For the second block,

⎛ ⎞
0 0 0
σ 2 = ⎝0 −41791.04 0⎠ kp/cm2 ,
0 0 0
⎛ ⎞
0.019 0 0
ε2 = ⎝ 0 −0.059 0 ⎠ .
0 0 0.019
(d) To find the value of forces P1 and P2 that allow the blocks to come into contact
while keeping the top plate horizontal, we impose an equilibrium of forces along
the vertical direction first, see Fig. 1.10,

Fy = 0,
so that
P1 + P2 = σ (1) (2)
y · 50 · 50 + σ y · 50 · 50.
50 cm 25 cm 25 cm 1 cm 25 cm 25 cm 50 cm
P1 P2
(1) (2)
Fy Fy
Fig. 1.10 Equilibrium of the top plate under the action of the two forces and the reactions coming
from the two blocks
This provides us with the first equation that will allow determining the value of
both forces:
P1 + P2 = 268656716.4 kp.
To impose the horizontal position of the top plate in the deformed configuration,
we impose the equilibrium of moments, so that no resulting rotation appears,

Mz (Q) = 0,
where Q is arbitrarily chosen as the point where P1 is applied (to make it disappear
from the second resulting equation and make the final solution simpler—other choices
will obviously produce the same result):
0 = 2500 · 65671.64 · 75 + 2500 · 41791.04 · 126 − P2 · 201.
By solving for P2 = 126754288.3 kp and substituting back in the vertical equi-

librium equation we arrive to P1 = 141902428.1 kp.
Chapter 2
Cable Structures
Abstract This chapter introduces the basic aspects underlying the structures made
of cables.
2.1 Introduction
In the first chapter, we have just seen that the elastic problem is complex. So complex
that we still do not know the solution for a general problem, but only for simple
cases of little practical interest. The discipline we are going to begin to study takes
a different route. As will readily be seen, a new magnitude is to be defined that will
greatly help to alleviate our burden. This is the concept of internal force.
Cables are usually represented as a straight or curved line, depending on the type
of load they support, see Fig. 2.1, with hinges at each end (they are often supported
so that rotations are allowed, see Fig. 2.2, although the true reasons will become
clear when we study beams). This is also typical of this discipline: structures are
represented by a simplified diagram, by employing signs whose meaning will be
explained throughout this book. From now on, a cable will be a thin line, given its
relative slenderness. Even if they are actually thick, such as, for instance, cables in
bridges, see Fig. 2.3, its area to length ratio is always very low.
2.2 Main Hypotheses
Solving the Navier Equation (1.11) for cable structures is indeed very difficult. Even
direct numerical simulation with the most recent codes on the market is not self-
evident. However, cable structures are in general easy to understand under some
classical hypotheses:
1. Cables do not support any form of bending, due to their slenderness. Therefore,
under the action of a load such as P, see Fig. 2.1, they acquire an equilibrium
configuration that depends on the type of load they are supporting. This is in

34 2 Cable Structures
Fig. 2.1 A typical cable

structure. The span, , and
the rise, f , are marked
sharp contradiction to the hypothesis of small displacements introduced in the

previous chapter, but it will produce no special difficulty, as we will see.
2. According to the previous hypothesis, the only component of the stress tensor
that will be non-zero is σx (x, y, z) = σ(x), that is, it is assumed to be constant
throughout every cross-sectional area of the cable (y, z coordinates), but not
necessarily along its longitudinal direction (x-coordinate).
3. Therefore, an internal force will appear, of the form:

N (x) = σx d A = σ(x)A, (2.1)
A
where A is the area of the cross section of the cable and is assumed to be constant.
N (x) is the axial internal force and can depend, in its most general case, on the
x, axial, coordinate of the cable.
Once the axial force has been determined, every magnitude can thus be obtained
by simply applying the simplified form of the constitutive and kinematic equations
seen in the previous chapter:
σ(x) N (x)
εx (x) = = , (2.2)
E AE
where εx is the only non-vanishing component of the strain tensor, and E is the
Young’s modulus of the material under consideration. Once the strain has been deter-
mined, we can obtain the displacement at any point by just integrating the kinematic
equations, viz.,
x=l x=l
N (x)
Δl = εx (x)d x = d x.
x=0 x=0 AE
If, as usual, both the area of the cross section and the Young’s modulus are constant,
we obtain: x=l
1
Δl = N (x)d x. (2.3)
AE x=0
2.2 Main Hypotheses 35
Fig. 2.2 Support apparatus for cable structures often allow free rotation
What we have done in a very subtle way is the usual form of work in this
discipline. Observe that we have defined a new magnitude: the internal force.
This may seem a waste of time, since the problem is already complex enough.
However, throughout this book (and, in general, any book devoted to strength
of materials), we will operate in this same manner: from the geometry and
loads of the structure, we will compute the distribution of internal forces and,
from them, we will compute stresses, then strains and, finally, displacements.
This can seem a weird form of working, but it is, in general, much simpler
than the integration of Eq. (1.11).
Fig. 2.3 Pont de l’Assut de l’Or, Valencia, Spain (S. Calatrava), a structure that employs cables
2.3 Equilibrium in Cables
The true advantage of this way of working is that internal forces can be determined
without solving the Navier equations. In fact, there are some simple cases—called
isostatic—in which simple force-equilibrium considerations will suffice to obtain
the internal forces acting on our structure. Structures formed by one single cable will
always be isostatic. In the example of Fig. 2.1, imagine for a second that we cut the
cable, as we did in the previous chapter, isolating a region around the applied load
P, see Fig. 2.4.
Since the internal forces must follow the geometry of the cable, by hypothesis 1
above, we know in advance the direction of the loads and can establish the equilibrium
2.3 Equilibrium in Cables 37
N1 N2
P
N1
P
N2
Fig. 2.4 Equilibrium of the load in the structure of Fig. 2.1. This is the so-called free body diagram.
On the right is the closed-force polygon ensuring equilibrium. To obtain the force polygon, just
draw every force one after the other, until a closed polygon appears. The particular ordering is not
relevant for the result
Fig. 2.5 Equilibrium of the

R1
load in the structure of
Fig. 2.1 around each support.
Since every structure must be
in equilibrium, but also every
portion of it, we denote by a
dashed circle the portion of
the structure on which we are
imposing equilibrium
N1
equations. Classically, equilibrium was applied by employing graphical techniques.

In particular, for the forces to be in equilibrium, the polygon of forces must be closed,
see Fig. 2.4 (right).
If we perform the same cutting around each one of the attachment points or
supports, we arrive at a similar situation, see Fig. 2.5. In this case, we find that the
reaction is identical to the applied load: R1 = N1 .
It is important to note that, apparently, the sign of N1 has changed with respect
to Fig. 2.4. However, if we isolate the central part of the cable, see Fig. 2.6, we are
now ready to perceive that the cable is actually subject to a pair of internal forces
that produce traction in it.
This is, indeed, the only possibility since, as mentioned before, due to the slender-
ness of cables, axial internal forces of compressive character cannot be supported.
They will lead to some form of buckling, a phenomenon that will not be studied in
this book.
We readily notice, then, that a member of our structure is never subject to a force,
but rather pairs of forces in equilibrium. This pair of forces, in the case of cables,
cannot be anything but traction.
R1
N1
N1
N1
N2
P
N1
Fig. 2.6 Equilibrium diagram of the cable in Fig. 2.1
2.3.1 Influence of the Rise
The so-called Cremona diagram of Fig. 2.4 allows us to analyze the influence of the
rise in the resulting internal forces, see Fig. 2.7. Thus, changing the rise from f to
f /2, produces an increase in the internal forces.
It is important to note that the appropriate way of designing a structure made

of cables is not by fixing the length of the cable. As mentioned before, due
to the lack of fulfillment of the small displacement hypothesis, it is extremely
difficult, in general, to determine the final configuration of the cable.
Instead, it is much more practical to begin by fixing either the slope or the
rise of the cable, then determining the internal forces in them, so as to be able
to obtain the needed area to hold up these forces, and then, eventually, the
increase in length produced by these internal forces.
2.3.2 Multiple Loads
A similar way of reasoning leads us in a straightforward manner to the analysis

of an arbitrary number of loads, load direction etc. It is now evident that the true
N1 N1
f /2
P
f /2
N2 N2
Fig. 2.7 Influence of the rise in the resulting internal forces in the cable. The new Cremona diagram
is depicted in dashed lines. Notice the increase in the traction of both cables, N1 and N2

P
N1
N2
P P N2
N1
f
N3
N2 P
N3
Fig. 2.8 A structure with two loads
advantage of working with internal forces is precisely the possibility of, by applying
simple equilibrium equations, obtaining every magnitude in the cables. For instance,
consider the case with two loads, see Fig. 2.8.
Here, we notice that the Cremona diagram introduces internal force N2 twice. In
fact, since equilibrium is applied at the nodes (those where hinges have been placed),
N2 appears twice: once for the node at the left, once for the node at right. Of course,
both equilibrate (in fact, N2 is an internal force) so that only externally applied forces
from the diagram (i.e., applied loads plus reactions at supports) are not cancelled in
the Cremona diagram.
2.3.3 Cables with Arbitrary Loads
We now consider the case of a cable loaded with two loads acting along arbitrary
directions, see Fig. 2.9. The case can be further generalized in a straightforward
manner, see [16].
First, consider two loads of 10 and 15N , respectively. The procedure begins by
computing the resultant force, see Fig. 2.9, left. For the design of the structure, we
can choose either the slope of the cable or the rise, f¯. We put a bar on the f , since
this will not be the final rise of the cable. In this case, we fix f¯, although the case of
choosing the slope at the hinges is identical. Once chosen, this allows us to determine
the slope of the cable near the anchoring points, see the dashed line.
The intermediate part of the cable will have a slope that we can find easily from
the Cremona diagram, by just joining the intermediate points, between the end of N1
and the end of the 10N force, see the red arrows N2 . The resultant shape of the cable
10N
f¯
N3
10N N2
R
15N
R
15N
N1
Fig. 2.9 A structure with two arbitrary loads. Concept of resultant cable
Fig. 2.10 A structure with

two arbitrary loads.
Resultant cable
10N
15N
is shown in Fig. 2.10. Notice the slope of the intermediate section, parallel to the red
force N2 . What we have obtained is actually the funicular polygon of the loads.
When there are three or more loads acting along arbitrary directions, of course,
they need not cross at a single point. In that case, the previous procedure can be
generalized by computing the resultant of the two first loads, and then computing
the resultant between the one just computed and the third load, and so on.
2.4 Funicular Curves for Distributed Loads
The case of a cable subjected to a distributed load is even more interesting, although
it can be considered as a generalization of the previous cases, see Fig. 2.11. This can
be seen as a particular case in which a series of infinitesimal loads are applied to
the cable. Therefore, instead of a sequence of straight lines, the configuration of the
cable will be a curve, and hence the name of funicular curve rather than polygon.
In this case, the procedure is identical to the previous section. First, compute the
resultant force but note that this type of load acts per unit length, i.e., their units will
be force per unit length (usually, kN/m) in horizontal projection, not per unit cable
length. The resultant, see Fig. 2.12, will have a value q · and will be applied at the
center of gravity of the load distribution (imagine the case of a triangular distribution,
for instance). In this case, the center of gravity coincides with the middle of the span,
since the distribution is uniform. The naive approach to the problem will thus be to
divide the distributed load into small, concentrated loads that will “discretize” the
curve, but will render a close approximation to it.
The resulting funicular curve for a cable under constant, distributed load is a
parabola. This can be easily demonstrated. Let us imagine a cable portion of elemen-
tary length ds, see Fig. 2.13.
By applying equilibrium along horizontal and vertical directions, respectively, we
find,
(N + d N ) cos(θ + dθ) − N cos θ = 0

(N + d N ) sin(θ + dθ) − N sin θ = dp,
Fig. 2.11 A cable with a

distributed load
q
N1
q·
N2
q·
Fig. 2.12 A cable with a distributed load
ds N + dN
dy N
dp
dx
x
Fig. 2.13 Equilibrium at a portion of the cable
where θ and θ + dθ represent, respectively, the angle with the horizontal where n
and N + d N are applied, respectively (not included in Fig. 2.13 for clarity). This is
equivalent to
d(N cos θ) = 0 (2.4)

d(N sin θ) = dp. (2.5)
2.4 Funicular Curves for Distributed Loads 43
From Eq. (2.4), we notice that the horizontal projection of N is actually constant over
the cable, i.e., N cos θ = H = const. In turn, from Eq. (2.5), we obtain, by noting
that tan θ = y ,
d(N sin θ) = d(H tan θ) = H dy = dp,
or, equivalently,
1
dy = dp,
H
from where,
1 dp
y = ,
H dx
which is the differential equation of the so-called funicular curve.
Note that, by virtue of Eq. (2.4), if the horizontal projection of the internal
force in the cable, H , is constant, the steeper you design your cable, the greater
internal forces it will suffer.
By integrating the differential equation of the funicular curve, we will arrive at a

general expression of the type
y = y(x, C1 , C2 ),
with C1 and C2 as two constants whose precise value is to be found with the help
of the attachment conditions. If the curve is subjected to a distributed load, then
dp = q · ds, and by integrating the funicular curve,
q
dy = d x,
H
which gives, in turn,
q 2
y= x + C1 x + C2 ,
2H
which is in fact the equation of a parabola. For the coordinate axis defined in Fig. 2.14,
we have,
q 2
y= x −
2
. (2.6)
2H 4
This is the curve being described by suspension bridges, such as the Golden Gate
bridge, see Fig. 2.15, for instance. In this case, if we neglect the self-weight load, the
main forces come from the hangers, vertical cables hanging from the load-bearing
cable.

y
Fig. 2.14 Parabolic funicular curve
Fig. 2.15 Golden gate bridge in San Francisco, U.S.A., a typical example of a suspension bridge
2.5 Catenary Curve
Of particular importance, for reasons that will become evident in the next chapter,
is the funicular curve resulting when the cable is subject to a load proportional to its
length (like its own weight, for instance, even if we usually neglect it).
In that case, dp = q ds instead of dp = q d x, as in the previous section. Therefore,

dp = q ds = q 1 + y 2 d x,
2.5 Catenary Curve 45
so that we arrive at a funicular curve of the type

q
dy = 1 + y 2 d x.
H
By grouping terms in y and x, respectively,
dy q
= d x,
1+ y 2 H
which is a differential equation that, after integration, gives,

q
y = sinh x + C1 ,
H
and, by integrating again,
H q
y= cosh x + C1 + C2 .
q H
This curve is known as a catenary, and is the curve described by cables along railways,
for instance.
The difference between the parabola of Eq. (2.6) and this catenary curve is
subtle. Both curves are very similar in shape, although the catenary has its
center of gravity slightly lower.
2.6 Sizing of Cables
Cables, as for any structure, are subject to limitations in both their ultimate limit state
(i.e., capable of producing a catastrophic failure) and serviceability-limit state (i.e.,
capable of making the structure deflect or vibrate up to a limit in which it looses its
practical interest). For the first case, it suffices to ensure that the stress in the cable,
σ, is lower than the yield stress of the material, σe , sometimes also denoted as σ y . To
avoid large deformations, it is a practical criterion, see [16], to limit the stress due to
permanent loads to a value
σ ≤ 0.45 σe .
As seen before, the internal force, and therefore the stress, is higher where the
slope of the cable is maximal. This occurs, generally, near the anchoring points. It is
precisely there where we must ensure the safety of the structure.
It is important to note, however, that even if the internal force is maximal where
the slope is higher, a shallow cable will lead, in general, to higher internal forces.
Recall Fig. 2.7, for instance, where a decrease of the slenderness ratio, / f , produced
an important increase in the internal force, as can be readily noticed in the Cremona
diagram of the structure.
2.7 Deflections in the Structure
As mentioned before, large deflections or vibrations can cause the structure to be

useless —remember the problems caused in the millennium bridge, in London, for
instance, that motivated its closing and subsequent reparation.1
Deflections due to permanent loads can be compensated easily by constructing
cables slightly shorter then necessary or, in other words, by taking into account that
these deflections will always be present.
To have an estimation of the deflection caused by variable loads q (other than dead
loads or permanent loads, g), we assume linearity in the behavior of the structure.
Therefore, the displacement at any point will look like
q
u(q) = u(g + q) .
g+q
If the displacement is judged to be too large, and the span cannot be reduced, then,
the solution comes from reducing the slenderness ratio by augmenting the rise f , or
by reducing the displacement by augmenting the area of the cable, recall Eq. (2.3).
It must be highlighted, however, that in general one single cable is not able to
provide with enough stiffness to the structure, so that a change in the variable
load does not produce a large displacement. Variable loads will always pro-
duce large changes in the shape of the cable. See [16] for some practical
examples.
In order to avoid such drastic changes in shape, other solutions must be taken into
account:
1. One simple, but very practical, solution is to increase the weight of the structure, to
make the relative importance of variable loads smaller. This naive or brute-force
solution is indeed very practical and has been applied successfully frequently,
especially in the construction of bridges.
1 Seehow BBC covered the re-opening of the bridge after costly modifications in it: http://news.
bbc.co.uk/2/hi/uk_news/england/1829053.stm.
2.7 Deflections in the Structure 47
2. To avoid adding more weight, an alternative solution is to exert forces on the

structure by means of other cables. This gives rise to the concept of cable beams,
see Fig. 2.16.
The practical effect of this type of structures, see Fig. 2.17, is an increase on the
degree of hyperstaticity or static indeterminacy of the resulting structure. As a prac-
tical result, the internal forces can no longer be determined by a simple application
of equilibrium conditions (nor by the drawing of a Cremona diagram, for instance).
Fig. 2.16 Cable beam
Fig. 2.17 Example of a structure made of a sequence of cable beams. This sequence gives rise to
a sort of cable membrane. José de Yarza, architect. Zaragoza, Spain
Fig. 2.18 Cable stiffening of the glass facade at Gare de Montparnasse, Paris, France. Baudoin,
Cassan de Marien, López and Saubot, architects
Due to the lightness of cable structures, the Irish engineer P. Rice began in the
1980s to employ cables to stiffen glass structures. This practice is nowadays very
common in many airports, railways stations and any other public building with large
glass facades, see Figs. 2.18 and 2.19.
Fig. 2.19 Glass structures stiffened by cables at Dubai airport

Worked Examples
2.1 Given the loads applied to the cable structure represented in Fig. 2.20, compute
the reaction forces, the total cable length and the final height of point C, h C .
2.5 m 5m 2.54 m
1 kN
1.5 m
A
1.875 m
hC
2 kN
D
B
Fig. 2.20 A cable structure with two loads
Solution 2.1 To obtain the reaction forces at A and D, it is necessary verify global
equilibrium on the cable structure. Therefore, the sum of the vertical reactions in A
and D are equal to 1 kN + 2 kN = 3 kN, so that
V A + VD = 3 kN.
Looking for equilibrium of the horizontal forces and momenta at an arbitrary

point (say, D), we have, see Fig. 2.21,
H A = HD ,

M(D) = 2 × 2.5 + 1 × 7.5 + H A × 1.5 − V A × 10 = 0.
The system has three equations with four unknown variables (V A , VD , H A and
H D ), so we need another equation. This will be obtained taking a look to Fig. 2.22,
where the slope of the cable is known. If, in a cable, there is no other force than an
axial internal force, it must act along the axis of the cable, thus giving a relationship
between V A and H A .
The relative position between points A and B is known, so
VA 1.875
= .
HA 2.5
HA
1kN
VA HD
2kN
VD
Fig. 2.21 Equilibrium of the whole structure
RA
NAB
Fig. 2.22 Equilibrium around the support A
This is precisely the fourth equation needed to solve the equilibrium system of
equations.
After solving, we find H A = 2.08 kN, V A = 1.56 kN, H D = 2.08 kN and VD =
1.43 kN.
The modulus of the reaction at support A will obviously be R A = H A2 + V A2 =
2.60 kN = N AB , identical to the internal force in the first section of the cable, AB.
NAB 1kN
α
β
NBC
Fig. 2.23 Equilibrium at point B
By imposing horizontal and vertical equilibrium at point B, where β is the

unknown angle described by the cable with respecto to the horizontal line, see
Fig. 2.23, we have:
1 + N BC · sin β = N AB · sin α = V A ,
N BC · cos β = N AB · cos α = H A ,
obtaining finally, N BC = 1.61 kN and β = 15.10◦ .

The length of the segment between A and B, is
5
L BC = = 5.17 m.
cos β
The rise at point C will be
h C = 1.875 + L BC · sin β = 3.22 m.
Finally, the total length of the cable will be:
L = L AB + L BC + L C D

= 1.8752 + 2.52 + 5.17 + (3.225 − 1.5)2 + 2.52 = 11.34 m.
Note that, although it is somehow counterintuitive, we cannot design a cable

structure by fixing, at the same time, the value of the applied loads and the
geometry of the cable. We usually think in terms of force equilibrium, but
moment equilibrium must also be taken into account. If we fix at the same
time h B and h C and their respective applied forces, the cable structure—its
different sections, in fact—will rotate until moment equilibrium is reached at
different h B and h C positions.
2.2 If the rise of the cable in Fig. 2.24 is 9 m, compute, for a uniform load q =
300 kp/m,
(a) Equation of the parabola adopted by the cable.
(b) Maximum and minimum values of the axial internal force in the cable.
Solution 2.2 (a) We begin by imposing equilibrium for a portion of the cable around
the lowest point of the cable, which we will denote hereafter O, see Fig. 2.25.
The moment equilibrium gives:
x
M(C) = 0; q · x · − NO · h = 0
2
2
so that h = 2N
qx
O
, that is a te equation of a parabola with vertex at O, precisely the
geometry adopted by the cable under the load q.
The horizontal force equilibrium gives:

Fx = 0; NC · cos θC = N O .
90 m
y
9m
kp
q = 300 m
Fig. 2.24 Parabolic funicular curve for Exercise 2.2
NC
y θC
C(x, y)
h
O
NO
x
q = 300 kp/m
Fig. 2.25 Equilibrium at a portion of the cable from the bottom point
This expression indicate that the horizontal component of the axial force is the same
for all points of the cable—something that we have just introduced in previous
sections. By imposing vertical equilibrium,

Fy = 0; NC · sin θC = q · x.
Dividing the two last expressions,

q·x
tan θC = ,
NO
while, if we exponentiate both by two and the sum,
NC = N O2 + q 2 x 2 ,
where NC is the axial force at each point of the cable, irrespective of its position.
Since the supports are placed at the same height above the lowest point of the
cable, the bottom point of the cable will be in the middle of the span, and the rise is
therefore known: h = 9 m, so that
qx2 300 × 452

NO = = = 33750 kp.
2h 2×9
Therefore, the geometry expression for all point in the cable is
q · x2
y=h= = 0.004 · x 2
2 · NO
The minimum axial force will take place at the lowest point of the cable,
Nmin = N O = 33750 kp.
Finally, the maximum one will take at the supports,
Nmax = N O2 + 3002 × 452 = 36349.86 kp.

Chapter 3
Arches
Abstract This chapter introduces the basic aspects of structures composed of arches.
3.1 Introduction
Soon after human beings began to deal with distances greater than those that can
be spanned by a tree trunk, the major problem of architecture saw the light. The
materials at hand were of a very particular nature. For instance, stone blocks can
resist (very well, in fact) compressive stresses, but without any type of mortar, they
simply detach from each other when subject to traction. Granite, for instance, resists
up to 180 MPa [12], but granite blocks simply do not resist any load if subject to
traction, or a loss of stability, leading to a catastrophic failure under complex load
states.
Therefore, the problem of architecture soon became that of designing stable struc-
tures, rather than that of designing resistant structures. Actual stresses in stone build-
ings are so low that resistance was not actually the problem. We are not aware of any
building that collapsed due to excessive stress, but many (thousands, probably; we
will never know) failed due to lack of stability.
In this chapter, we will study how to solve this problem, which was of primary
importance until the 19th century, when the Industrial Revolution introduced iron
and the steel as building materials. For the first time, steel as a building material had
the same characteristics in traction and compression, thus changing completely the
way of designing structures.
3.2 A Brief Historical Perspective
Without aiming to be very rigorous, we can say that the Romans, for whatever
reason, decided that the “perfect” shape for an arch was a semicircle, see Fig. 3.1.
56 3 Arches
Fig. 3.1 A semicircular arch
They mastered arches so well that could build the Pantheon in Rome from 27 B.C.
to 14 A.D., with a diameter of 43 m, and a perfectly spherical shape.
Later, Romanesque architecture also employed semicircular arches profusely,
although most of the Roman mastery of construction was lost in this epoch. How-
ever, the use of arches was not fully understood until the scientific analysis of Hooke
(1635–1703), who was the first to find an explanation for the good load-bearing capa-
bilities of Roman arches. See Hooke’s portrait in Fig. 3.2. In the painting, Hooke is
represented with a chain in his hands, in memory of, precisely, his findings about the
shape of arches.
In his work “A Description of Helioscopes, and Some Other Instruments”, Robert
Hooke wrote the solution to the shape of the perfect arch (“a problem which no
Architectonick Writer hath ever yet attempted, much less performed”) in the form
of an anagram, see Fig. 3.3.
The solution was written as:
abcccddeeeeefggiiiiiiiiillmmmmnnnnnnooprrsssttttttuuuuuuuux.
After Hooke’s death, the executor was allowed to make public the correct ordering
in the phrase:
3.2 A Brief Historical Perspective 57
Fig. 3.2 A portrait of Robert Hooke manipulating a chain. Rita Greer, 2009
Fig. 3.3 Anagram written by Hooke in 1676 containing the solution to the problem of the perfect
shape of an arch
Ut pendet continuum flexile, sic stabit contiguum rigidum inversum.
Besides having included an extra “e” and an extra “i”, the meaning of the anagram
was “As hangs the flexible line, so but inverted will stand the rigid arch”.
The idea is simple, yet appealing. If a cable supports loads in pure traction, its
inverted image will support loads in compression, see Fig. 3.4. This extends in a
straightforward manner to arches subjected to distributed loads, see Fig. 3.5, where
we discover that the “perfect” shape for an arch under distributed load is a parabola.
58 3 Arches
P
P
f
Fig. 3.4 Hooke’s solution to the problem of a perfect arch by just inverting the shape of the
equivalent cable

q
Fig. 3.5 An arch with a distributed load. Following Hooke, the solution will be an arch of parabolic
shape
Fig. 3.6 Load bearing for a

Roman arch. Funicular curve q
(red, dashed line) of the load
and true geometry
If Hooke’s solution to the problem is correct, why did Romanesque arches never
fall and are still bearing loads today, despite being semicircular? The answer is
easy, see Fig. 3.6. In fact, the dimensions of the stone blocks used are forgiving.
To guarantee stability, quarry workers once produced big blocks. If this is the case,
the funicular curve is contained entirely within the geometry of the stone arch, thus
guaranteeing compression somewhere within the block. But if we design a thin arch,
see Fig. 3.7, the possibility arises for the funicular curve to exceed the limits of the
3.2 A Brief Historical Perspective 59
Fig. 3.7 Load bearing for a thin Roman arch. Funicular curve (red, dashed line) of the load and
true geometry. Note that hinges are produced where the funicular curve crosses the geometry of the
arch, thus causing a loss of stability
Fig. 3.8 Experimental design of the Sagrada Familia Cathedral in Barcelona by Antonio Gaudi.
The picture has been turned upside-down so as to appreciate the design of the arches. The small
sand bags in the picture are estimated so as to be proportional to the true loads in the structure
arch. Where the funicular curve crosses the arch a hinge appears, with up to three
being possible. With three hinges, the structure is actually a mechanism, and any
horizontal load will result in a collapse.
60 3 Arches
Therefore, what actually happened is that Romans and artisans during the Middle
Ages learned by doing. And they learned that, for an arch to be stable, its dimen-
sions must be large enough. Without knowing it, what they were doing was actually
enabling the space for the funicular curve to develop entirely within the geometry
of the arch. Probably many arches have fallen since the time of early Romans, but
those that have survived are really stable.
This discussion, far from being old-fashioned, as one might think, is really impor-
tant. Materials of substantial architectural importance, such as concrete, for instance,
behave not very differently than stone. In fact, the tensile stress that concrete can bear
is 10–15% the value of its compressive stress, and is often neglected for practical
dimensioning.
Note that, as mentioned in Chap. 2, if permanent loads predominate in the dimen-
sioning, a catenary curve should be preferred as the shape of the arch. This motivates
the chain in Hooke’s portrait (it suggests that Hooke was actually giving shape to
an arch). This same technique was used by Antonio Gaudi to study the arches in the
Sagrada Familia Cathedral in Barcelona (when visiting, do not miss the exhibition
in the crypt of his model of the structure and the loads applied to each arch, see
Fig. 3.8).
3.3 Main Hypothesis
After the historical perspective presented in the previous section, we are now in the
position to establish the modern hypothesis commonly accepted for the design of
masonry structures, in particular, arches [10]. It is generally accepted that masonry
structures can bear compressive internal forces very well and that, in general, the
resulting stresses are one or two orders of magnitude below their limit values. Failure
by compression is in fact a very rare event.
In addition, masonry structures are not able, in general, to resist internal traction
forces. Traditionally, this has been simply neglected. Furthermore, because com-
pressive forces are predominant, slipping between stone blocks or bricks occurs also
very rarely. So, masonry arches are dimensioned to accommodate solely the axial,
compressive forces.
In view of this, we will accept the following hypothesis:
1. The arch possesses an infinitely great limit to stress under compression, i.e.,
σe+ = ∞.
2. Masonry arches do not have any tensile capacity, i.e., σe− = 0.
3. Failure by slipping between blocks or bricks is impossible,
where symbols + and − have been used to indicate compression and traction, respec-
tively. This convention is, nevertheless, purely arbitrary.
3.4 Limit Analysis of Masonry Arches 61
3.4 Limit Analysis of Masonry Arches
The design process for masonry (stone, brick, adobe) arches under the hypothesis
1–3 above is limited by the fact that, under pure equilibrium conditions, there is no
unique solution to the problem. This can readily be understood by noting that there
exist two limit funicular curves for every arch, see Fig. 3.9.
Observe that, in both limit cases as depicted in red and blue, respectively, the
keystone of the arch will present a hinge at the extrados and intrados, respectively.
Any other funicular curve between these two is equally valid, since it provides a
purely compressive line of thrust and, under hypothesis 1 above, will lead to a stable
arch. Practical methods exist for the design of limit-thrust lines, such as the one
proposed by an anonymous author but commonly attributed to Fuller [11].
The limit analysis of masonry arches, therefore, cannot be achieved by simple
equilibrium considerations, since an infinite number of solutions exist, as we have
just seen. This is caused by the hyperstatic or statically indeterminate character of
these structures. The history of arch design in the 19th century is the quest for the
“true” thrust line in the arch.
Moseley (1843, see [11] and references therein) established that the true thrust line
is the one with a minimal-force value (a sort of naive principle of minimum potential
energy). However, Poncelet (1852) suggested that, to find the right solution, it is
necessary, as in every statically indeterminate structure, for the rest of equations of
the elastic problem, particularly the constitutive equations, to come into play. Spanish
architect Eduardo Saavedra (1860) was however the first, soon before Castigliano, to
state that, to find the right solution for the thrust line in arches, the complete elastic
problem must be solved first.
Heyman, however, was the first to study systematically the effect of the perturba-
tions in arches [10]. By perturbations, we mean the effects of formwork removal, see
Fig. 3.10. After removal, the loads begin to act on the arch, so that it suffers a descent
that causes a horizontal thrust on the springer (lowest voussoir on each side of an
arch). Inevitably, three hinges form at the keystone and the springers, see Fig. 3.11.
The arch with these three hinges is now statically determinate.
Fig. 3.9 Limit-thrust curves for an arch of particular geometry

62 3 Arches
Fig. 3.10 Formwork for the construction of a stone arch
Fig. 3.11 Effect of

perturbations after formwork q
removal on the arch. Three
hinges are formed so that the
thrust line adopted is that of
minimal internal force. An
arch with three hinges is
always statically determinate
It is extremely important at this point is to recall here the lower bound theorem
of the theory of structures.
The lower bound theorem states that

“A load system, based on a statically admissible stress field which
nowhere violates the yield condition is a lower bound to the collapse load.”
[17].
3.4 Limit Analysis of Masonry Arches 63
Fig. 3.12 A wooden, three-hinged arch for the construction of a pedestrian bridge at Mazcuerras,
Cantabria, Spain a A general view of the bridge. b Hinge at the top. c Hinge at the supports
Under this framework, therefore, it is clear that the design of an arch is limited
to the verification that, under any possible load, the thrust line (funicular curve of
loads) passes within the intrados and extrados of the arch. We will then obtain a sort
of envelope of funicular curves that will provide us with the right shape for the arch.
64 3 Arches
Fig. 3.13 Musée d’art moderne et d’art contemporain, also known as MAMAC, in Nice, France.
Yves Bayard and Henri Vidal, architects. Henri Vidal also acted as structural engineer, his primary
degree, prior to becoming an architect
3.5 Modern Arch Structures
The dependence of the behavior of the arch on the perturbations, i.e., on the way
formwork is removed, or in the relative stiffness of the terrain below the foundation,
for instance, is always troublesome and somehow unpredictable. That is the reason
why most architects once chose statically determinate arches to design their struc-
tures. How to design a statically determinate arch? Simple: just force the thrust line to
3.5 Modern Arch Structures 65
Fig. 3.14 Model of the shell roof of the British Museum. Beatriz Moya, M.Sc. dissertation. School
of Engineering and Architecture, Universidad de Zaragoza, 2017
Fig. 3.15 Shell under purely compressive internal forces. Miguel Angel Campos, M.Sc. disserta-
tion. School of Engineering and Architecture, Universidad de Zaragoza, 2014
pass through three chosen points by placing three hinges at these points. Thus, three-
hinged arches became a very popular structural typology that continues to capture
undeniable interest, see Fig. 3.12.
Of course, there is in reality no strict need for a true mechanical device that allows
the two half-arches to rotate freely—this is the strict definition of a hinge. In general,
and especially for reinforced concrete structures, a simple abrupt change of section
is enough for the structure to behave as hinged.
66 3 Arches
Fig. 3.16 A pavilion vault. Tower of the old building of the Technical University of Vienna
Fig. 3.17 Hyperbolic paraboloid surface
In general, there is no practical limitation on the employment of arch structures

with materials of modern architectural interest, other than stone. Arches and funicular
geometries in general continue to provide the architect with a very effective load-
bearing capability. For instance, Yves Bayard and Henri Vidal designed the Museum
of Modern Art and Contemporary Art (MAMAC) in Nice, France, see Fig. 3.13. Its
main structure is composed of funicular structures, from which the floor slabs hang.
3.6 Shells, Vaults and Domes 67
3.6 Shells, Vaults and Domes
The concepts just introduced can be extended in a straightforward manner to

three-dimensional structures, thus forming shells, vaults and domes. At present,
however, it might be surprising to learn that this procedure was not extended to three-
Fig. 3.18 Principal stresses of the four hypars forming the famous umbrella by Felix Candela.
Notice how the third principal stress (relative to bending, or out-of-plane stress) does not appear in
the finite element model of the shell
Fig. 3.19 Structure of the Oceanogràfic in Valencia, Spain. F. Candela and S. Calatrava, architects.
The shell is formed by the intersection of five hypars
68 3 Arches
dimensional, rather than plane, two-dimensional structures, until 2007 [2]. Indeed,
it is not self-evident to find the funicular surface, given the shape of the plant of the
building and an interval in height for every node in the model. See Fig. 3.14 for a
simplified model of the dome of the British Museum following this theory.
Under the so-called Thrust Network Analysis (TNA) theory, a network of com-
pression forces is sought so as to guarantee equilibrium within an interval of pre-
scribed intrados–extrados surfaces. This is a powerful technique that guarantees
structures bearing load under purely compressive forces resulting extremely thin
shells, see Fig. 3.15.
There exist, however, some surfaces whose structural behavior is well-known
now for some centuries. One of these is the so-called groin vault, the result of the
intersection of two mutually perpendicular barrel vaults. The pavilion vault can
be seen as the dual of the groin vault, see Fig. 3.16. In this case, arches rely on
the arrises (intersections of the four barrel vaults, known as spindles). Following
the just introduced theory, arches rest on the arrises, exerting a thrust that is in
principle equilibrated (at least for self-weight, permanent loads; not for wind loads,
for instance) under purely compressive internal forces.
Another example of an interesting surface, well-known due for its inherent struc-
tural properties, is that of hyperbolic paraboloids, hypars, for short, see Fig. 3.17.
Fig. 3.20 A groin vault with cables to avoid excessive horizontal thrusts. Cividale del Friuli, Udine,
Italy
3.6 Shells, Vaults and Domes 69
The Spanish architect Felix Candela (1910–1997) developed a method of analysis

for this type of surface which are incorporated in the vast majority of his works,
mainly in Mexico.
This type of surface possesses double curvature and presents a very interesting
property: hypars of straight sides, supported solely at their corners, resist pressure
loads by developing tension-compression internal forces, without bending. Thus, for
instance, they are ideally suited for reinforced concrete. Reinforcing steel bars are
aligned with the tensile forces, while compressive forces are well supported by the
concrete itself, see Fig. 3.18.
Felix Candela mastered the construction of hypars so as to be able to develop very
interesting shapes, without the need for straight edges, see Fig. 3.19.
Fig. 3.21 A tied-arch structure. Top, sketch of the structure. Bottom, an example: the Third Mil-
lennium Bridge in Zaragoza, Spain
70 3 Arches
3.7 Arch-Cables
As already analyzed, the load in an arch or a barrel vault always implies horizontal
thrust in the springer. This is the reason for the need, in some cases, of attaching a
cable to absorb these forces, see Fig. 3.20.
This idea leads us to the concept of bowstring or tied-arch bridges (this concept has
its greatest utility in the design of bridges), see Fig. 3.21. In the right support, a new
symbol is employed that represents a type of support (in many books it also include
one, two or even three wheels between the two horizontal lines) which indicate that it
can move horizontally and rotate freely, but that the vertical displacement is impeded.
Fig. 3.22 A bowstring structure at heathrow terminal 5 London airport
Fig. 3.23 A tie-and-strut

cantilever structure
3.7 Arch-Cables 71
The goal is that the horizontal thrust is absorbed by the cable, so that no horizontal
force is transmitted to the soil or to the remainder of the structure.
This is also the key main structural element in the roof of the new London
Heathrow Terminal 5 airport, see Fig. 3.22.
Taking this idea to the limit, a very efficient way of designing cantilevered struc-
tures is by joining an arch and a cable or, more properly, a tie and a strut, see Fig. 3.23.
This is the basic constructive element of the famous Hong Kong and Shanghai
Bank building by N. Foster, see Fig. 3.24. The previous bowstring structure and the
strut-and-tie concept will naturally lead us to the fabulous world of trusses in the
following chapter.
Fig. 3.24 Hong Kong and Shanghai Bank building by N. Foster

72 3 Arches
Exercises
3.1. Estimate the position of the hinges after formwork removal for a horseshoe
arch, see Fig. 3.25.
3.2. Do the same for a Gothic arch. Estimate the necessary span-to-thickness ratio
for this type of arch. Analyze the differences between them in a purely structural
sense.
3.3. Figure 3.26 represents a staircase for access to the Carcassonne castle walls.
As can be seen, it consists of two semi-arches, both supported by a board column
where they meet in the middle. With the help of graphical statics, draw a sketch of
the catenary curve of both arches and estimate the width of the central support such
that the resulting force lies within its section (thus avoiding failure).
3.4. Investigate Blondel’s rule (1682) in the dimensioning of arches and try to justify
its validity. This was a practical recipe employed at that time by master architects and
which proved to work quite well, judging by the number of cathedrals still standing.
3.5. Justify the moon-like shape of the arch of the Maria Pia bridge in Porto,
Fig. 3.27. Is there any structural reason for this particular shape?
Fig. 3.25 Horseshoe (also Moorish or keyhole) arch

3.7 Arch-Cables 73
Fig. 3.26 Staircase in the Carcassonne castle, France
Fig. 3.27 Maria Pia bridge in Porto, Portugal. In the background, the Sao Joao bridge is also visible
Chapter 4
Trusses
Abstract This chapter introduces the basic aspects underlying structures made of
trusses.
4.1 Introduction
We have seen that cable structures can be considerably stiffened if additional bars are
added, thus preventing excessive displacements under variable loads, see Fig. 4.1.
This simple idea is at the origin of trusses, although not from a historical perspec-
tive, see Fig. 4.2, where a sort of truss lacking some diagonals is depicted. This is
typical of some well-known historic buildings. They can also be viewed as a means
to avoid mechanisms, see Fig. 4.3. The addition of the diagonal bar is precisely what
prevents the structure from collapsing as a mechanism.
The truss can also be seen as an evolution of the bowstring or arch-cable structure.
Let us assume that the deck of the structure bears a critical load so that we decide to
hang it from the arch, see Figs. 4.4 and 4.5. If, in addition, we add some diagonal bars
to prevent large motion due to variable loads, we arrive at a classical truss structure.
Trusses, similar to cables and arches, bear only traction and compression internal
forces. That is why they became very popular, especially after the Industrial Rev-
olution in the 17–18th centuries and the widespread availability of steel, Fig. 4.6.
Despite their conceptual simplicity, trusses continue to be widely employed both in
industry and architecture, notably for buildings including long spans.
4.2 Hypothesis
As in previous chapters, it is necessary to highlight that the popularity of trusses is due

to both their reliable structural behavior and the ease of their design and limit analysis.

76 4 Trusses
P P
N1
N3
N2
Fig. 4.1 A cable structure with an additional cable to improve stiffness
Fig. 4.2 A sort of primitive truss with a clear lack of diagonal bars. This type of truss is called a
queen-post truss
Therefore, it is essential to establish clearly the hypothesis and simplifications that

will help us to design trusses easily.
First of all, at this point, you probably have noticed that we represent every bar in
a truss with hinges at the ends, but this is not because they must be constructed this
way. Hinges indicate that bars can rotate freely around them, so the internal force
must be axial and pass through the nodes where bars meet. If bars are straight, this
axial internal force will act along the axis of the bar. No bending occurs in this case.
In fact, the absence of bending is not a consequence of the hinges, since many
trusses are not actually constructed this way, see Fig. 4.7. It is precisely because of
4.2 Hypothesis 77
Fig. 4.3 A truss used as a means to avoid collapse through a mechanism. The dashed line represents
the mechanism-type of displacement experienced by the structure in the absence of the diagonal
bar
(a)
(b)
Fig. 4.4 The truss as an evolution of a tied arch or bowstring structure. a After the addition of the
vertical struts and b after the addition of the diagonals
the triangular configuration of bars that this hypothesis is, to a very high degree of
accuracy, true.
Therefore, in the design and analysis of trusses we will accept the following
hypothesis:
1. Trusses will be composed of straight bars.
2. We will consider joints between bars as being hinges (even if they are not literally
so).
3. A load is assumed to be applied at the nodes only (otherwise it will inevitably
produce bending).
Under these very basic hypothesis, trusses will bear loads by developing axial
internal forces only, and therefore Eqs. (2.1)–(2.3) still hold. To see confirmation of
this, consider the structure in Fig. 4.8.
This structure was analyzed under the hypothesis of free rotation at the nodes
and again under the rigid-node assumption, just to see the difference in the results.
78 4 Trusses
Fig. 4.5 Railway bridge at Mandelieu-La Napoule, France, over the Siagne River (1862), a typical
example of a bowstring bridge. The arches were added to an existing bridge to reinforce the structure
in 1936
These are summarized in Table 4.1. As can be seen the differences are minimal, often
under 1%.
Of course, if nodes must rotate freely, so do the supports. Symbols in Fig. 4.8, for
instance, indicate that rotation is allowed (left symbol) and that, in addition, horizon-
tal displacement is free (right symbol). Examples of these supports are also shown
in Fig. 4.9. These supports closely resemble the small symbols that we normally
include in our models, but in fact they do not need to be so similar. To behave as a
pinned support, it suffices that, to a reasonable extent, the support allows a certain
degree of rotation. The symbol on the right of the structure of Fig. 4.8 is a roller.
It must allow some degree of horizontal displacement (see the right support of the
truss in Fig. 4.10).
4.3 Statically Determinate and Indeterminate Trusses
Consider for a moment the truss in Fig. 4.11. It is similar to the one in Fig. 4.3, but we
have added one more bar. Now, every node has three bars arriving to it. It is impossible
to find a suitable funicular polygon to determine every force. An analytic approach to
4.3 Statically Determinate and Indeterminate Trusses 79
Fig. 4.6 Bridge over the Duero River in Soria, Spain
the problem will give the same result: with only two equilibrium equations (vanishing
sum of forces in vertical and horizontal directions, for instance) it is impossible to
determine three unknowns. The situation could be even worse if we consider two
pinned supports, for instance. In that case not even the reactions at the support could
have been determined by equilibrium considerations.
We know from Chap. 1 that we have enough equations to solve the problem.
There is no question at all! However, it is important to known whether we must make
them all work or not. Up to now, due to the particular mechanisms of bearing loads,
masonry arches have been able to develop statically determinate configurations,
as noticed by Heyman [10]. With one single cable, there is always a way to find
their internal forces by means of a simple Cremona diagram. This is not the case,
however, for cable beams, as we have seen. Now, for trusses, this will not be the case,
in general, and it is interesting to have a tool to know in advance the degree of static
indeterminacy of the truss.
By isolating a particular region around a node of the truss and analyzing the
resulting free-body diagram, we have noticed that we are able to find two forces at
most. This is equivalent to the two force equilibrium equations mentioned before. In
principle, there should be a third one: the equilibrium of momenta. This is equivalent
to making all the forces cross at a certain point of space (otherwise, they will provoke
80 4 Trusses
Fig. 4.7 An example of truss with rigid nodes
rotations). But they always cross at one point, precisely the node. Therefore, there is
a simple rule to know in advance if a truss is statically determinate or not:
• if n reactions + n bars < 2n nodes , the systems is unstable (a mechanism),
• if n reactions + n bars = 2n nodes , the systems is statically determinate and,
• if n reactions + n bars > 2n nodes , the systems is statically indeterminate.
See Fig. 4.12 for an example of this simple rule. Note that in the case of Fig. 4.12a
the structure is actually stable but critical: any perturbation in the horizontal direction
will activate the collapse mechanism. Under the forces depicted in the figure, how-
ever, the truss is in equilibrium, with vertical bars bearing an internal force of value
P. We see, in addition, that there are various sources of indeterminacy. While in
4.3 Statically Determinate and Indeterminate Trusses 81
(a)
(b) (c)
Pinned support. Roller support.
Fig. 4.8 a An example truss. b An old pinned support apparatus. c Old roller supports
Fig. 4.12c this indeterminacy is provoked by one extra (often called superabundant)
bar, in Fig. 4.12d, the origin of this issue is the presence of two pinned supports, and
therefore four reactions.
4.4 Limit Analysis of Statically Determinate Trusses
For the time being, we will concentrate on the limit analysis of statically determinate
trusses. The indeterminacy is much more complex, and two large families of methods
exist for solving these cases. These are beyond the scope of this book, even if the
solution has been mentioned before: to add constitutive and compatibility equations
to the well-known equilibrium equations.
4.4.1 Graphic Statics
The first way to analyze trusses, already employed in cables and arches, is to employ
the Cremona diagram. The method is sketched with the help of the example in
Fig. 4.13.
82 4 Trusses
Table 4.1 Results for the truss structure in Fig. 4.8 with and without the hypothesis of hinges at
the nodes
Bar # N w/hinges (kp) N w/o hinges (kp)
1 0.000E+00 3.73E−01
2 1.50E+02 1.50E+02
3 1.50E+02 1.50E+02
4 0.00E+00 3.73E−01
5 −2.00E+02 −2.00E+02
6 2.12E+02 2.11E+02
7 −1.50E+02 −1.49E+02
8 7.07E+01 7.03E+01
9 −1.00E+02 −9.97E+01
10 7.07E+01 7.03E+01
11 −1.50E+02 −1.49E+02
12 2.12E+02 2.11E+02
13 −2.00E+02 −2.00E+02
14 −1.50E+02 −1.50E+02
15 −2.00E+02 −2.00E+02
16 −2.00E+02 −2.00E+02
17 −1.50E+02 −1.50E+02
Fig. 4.9 Examples of pinned supports. Left: Brussels train station. Right: Madrid Airport Terminal 4
4.4 Limit Analysis of Statically Determinate Trusses 83
Fig. 4.10 An examples of a roller support (the one on the right of the central support—the one of
the left is pinned). Paris metro line near Glaciere station. Boulevard Auguste Blanqui
To begin with, we must ensure that the entire structure is not able to rotate as a
rigid solid. This is guaranteed if the three external forces that act on the structure,
P, R1 and R2 meet at one single point. The line of action of P is known, and so it
is the one of R2 , since the support is a roller and therefore is not able to react with
an horizontal component. R1 will certainly have two components, and they can be
found by forcing it to pass through the crossing point of R2 and P, see Fig. 4.13. On
the right of the figure the resulting Cremona diagram is shown. In it, P is known,
both in direction and magnitude. We know only the lines of action of R1 and R2 ,
from the sketch in Fig. 4.13 (right), but this is enough to, by tracing parallels at both
ends of P, determine their intersection, thus fixing the modules of R1 and R2 .
Note that the above procedure is possible just because the reaction R2 is vertical
(i.e., because its support is a roller). If the support is changed to a pinned one,
then the structure would be statically indeterminate, and the line of action of
R2 would be unknown. The Cremona diagram would have no solution.
Once the reactions have been determined, the procedure continues node by node,
by imposing equilibrium or, in other words, a closed-force polygon. For instance, for
the rightmost node in the structure, we have the Cremona diagram shown in Fig. 4.14.
84 4 Trusses
P P
Fig. 4.11 Statically indeterminate trusses
P P P P
(a) 3 + 4 < 2 × 4, mechanism. (b) 3 + 5 = 2 × 4, statically determinate.
P P P P
(c) 3+6 > 2×4, statically indeterminate (d) 4+5 > 2×4, statically indeterminate
(internal). (external).
Fig. 4.12 Analysis of static indeterminacy
If the structure is statically determinate, we will not encounter any problem by pro-
ceeding node by node, always looking for some node with only two unknown forces.
Once all internal forces have been determined, the limit analysis of the structure is
simply to verify that
Ni
σi = ≤ σe , i = 1, 2, . . . , n bars ,
Ai
R2
P
R1
R1 R2

Fig. 4.13 Analysis of a statically determinate truss

N1
R2
N2
R2 N1
N2
Fig. 4.14 Cremona diagram for the rightmost node in Fig. 4.13
holds. Conversely, if one wants to design the structure, to determine the necessary
area of the cross section for each bar, it suffices to compute
Ni
Ai ≥ , i = 1, 2, . . . , n bars .
σe
Very often, for aesthetic criteria, all areas are taken as equal (and therefore the design
will be governed by the bar subjected to the maximum Ni ).
4.4.2 An Analytic Approach
The same equilibrium procedure can, of course, be done analytically. Again, we

begin by imposing equilibrium on the whole structure, so as to impede movement as
86 4 Trusses
a rigid solid. To this end,

FH = 0,

FV = 0,

M = 0,
where H and V indicate “horizontal” and “vertical” directions, respectively (although

any other mutually perpendicular direction could have been chosen). The sum of
moments can be taken with respect to any point in the plane of the truss. Usually it
is a good idea to take a reference point at the intersection of a maximum number of
external loads, to eliminate the contributions of these forces to the sum.
In the example of Fig. 4.13, we have,

FH = 0 → P cos 60 + R1H = 0,
FV = 0 → P sin 60 + R1V + R2 = 0,
M = 0 → R2 × − P sin 60 × = 0,
where moments have been taken with respect to the pinned support. Interior angles in
the triangles are assumed to be 60◦ . We have, therefore, three equations to determine
R1H , R1V and R2 . Again, as in the graphical analysis, if we have a pinned support at
the support on the right, then a fourth unknown R2H would appear, therefore making
the system of equations undetermined.
Once the three reactions have been found, we can continue node by node by
equilibrating their respective internal forces with the externally applied ones. Again,
for the node with a roller support,

FH = 0 → N2 + N1 cos 60 = 0,
FV = 0 → R2 − N2 sin 60 = 0,
where both N1 and N2 have been (arbitrarily) assumed to act towards the right. A
minus sign in the result will indicate that they actually act in the reverse direction.
Observe that it is nonsense to try to apply the equilibrium of moments again for
each node. This is due to the fact that the implied forces cross by definition at the
node, thus vanishing the moment trivially.
4.4.3 Method of Sections
There is no need to proceed node by node if we are interested in a particular internal

force, for instance. We have seen from Chap. 1 that all solids must be in equilibrium,
not only globally but also locally, by isolating every possible portion of the solid.
This is obviously true also for trusses.
P
N5
N4
R1 R2 R1 N3
Fig. 4.15 An illustration of the method of sections
Consider again the truss in Fig. 4.13 and assume we want to obtain the value of
the internal forces N3 , N4 and N5 in Fig. 4.15. By performing an imaginary cut (red,
dashed line in the figure), and keeping only the left portion of the structure, we can
equilibrate solely this part of the structure.
Notice how reaction forces R2 and loads such as P do not intervene in this equi-
libration procedure. Now it is mandatory to equilibrate both forces and moments,
since N3 , N4 and N5 do not cross at a single point. Therefore, we have,

FH = 0 → R1H + N5 + N4 cos 60 + N3 = 0,
FV = 0 → R1V − N4 sin 60 = 0,
M = 0 → N4 cos 15 × (/2) + N5 cos 15 = 0,
where moments have been taken with respect to the pinned support (taking advantage
of the fact that two forces pass through this support). Notice also that the length of
each bar is /2. Again, we have three unknowns and three equations. Statically
indeterminate trusses will lead at some point to a system of three equations with
more than three unknowns.
4.5 Serviceability Limit-State Analysis of Trusses
As already commented in previous sections, structures only must bear the forces
they are designed for, but, notably, their deflections must be small enough so as to
not compromise their serviceability. Therefore, it is of utmost importance to have a
method for easy determination of these displacements. Several such methods exist,
such as Mohr’s method. But perhaps the most useful one, and undoubtedly the one
that will have more implications in the solution of statically indeterminate structures,
is the Principle of Virtual Work.
88 4 Trusses
To begin with, let us consider a single truss bar (axial rod) under a distributed load
per unit length, p, see Fig. 4.16.
As already seen from Eq. (2.2), for traction–compression processes, we have
σ (x) N (x)
εx (x) = = ,
E AE
so that
N (x) = AEεx (x).
where A is the cross-sectional area of the bar and E the Young’s modulus of the
material. The product AE is often referred to as the axial stiffness of the bar. As
always, x is assumed to be the longitudinal axis of the bar. If we isolate it, as we did
repeatedly, but noteworthy in the method of sections, a portion of the bar of length d x
(see the detail in Fig. 4.16, right), the only forces acting on it are precisely this axial
internal force that suffers some variation d N between both ends of the differential
segment and the external load pd x (even if we rarely consider this type of load). In
this case,
N (x) + d N − N (x) = − pd x,
and therefore,
d
(AEεx (x)) = − p.
dx
Since
du x
εx (x) = ,
dx
we arrive at
d du x
AE = − p, (4.1)
dx dx
which is the differential equation governing the displacement of the bar. Obviously,
the first possibility to determine the displacement law of the structure is to integrate,
bar by bar, this differential equation. But imagine the process of doing it, for instance,
for the Eiffel tower! This is simply out of reach for many structures of practical inter-
est. The procedure we are about to develop enables one to obtain one particular nodal
displacement without the need of integrating the differential equation throughout the
entire structure.
To that end, let us multiply both sides of Eq. (4.1) by an arbitrary function ψ,
du x
ψ AE = −ψ p,
dx
and then let us integrate both sides all along the bar,
4.5 Serviceability Limit-State Analysis of Trusses 89
Nb
p pdx
N (x) + dN
N (x)
Na
Fig. 4.16 Equilibrium in a single truss bar. On the right, equilibrium of a differential portion of the
bar

du x
ψ AE dx = − ψ pd x.
0 dx 0
If we now integrate by parts this equation, we obtain,

ψ AE du x − dψ
AE
du x
dx = − ψ pd x,
d x 0 dx dx
0 0
ψ
|ψ N |0 0 εx N d x
ψ
where εx represents some virtual deformation coming from the arbitrary “displace-
ment” field ψ. This is precisely the expression of the Principle of Virtual Displace-
ments:
ψ
|ψ N |0 − εx N d x = − ψ pd x. (4.2)
0 0
What is important about the virtual-displacement field ψ is that it is entirely

arbitrary, with the sole requirement of fulfilling compatibility, i.e., taking into account
the prescribed displacements of the true structure (supports).
Even perhaps more important is the fact that we can exchange the roles of dis-
placements and internal forces in Eq. (4.2) so as to obtain the Principle of Virtual
Forces,
ψ
u N = ε x N ψ
d x − up ψ d x, (4.3)
0
0 0
where we have grouped on the left-hand side the term relative to external loading and
on the right-hand side the contribution of internal forces to the virtual work. Here,
the sole requirement for the virtual-force field is to be in equilibrium. Of course, it is
“virtual” in the sense that it does not correspond to the true solution of the structure.
90 4 Trusses
It is precisely Eq. (4.3) that is the one we will use for the determination of dis-
placements at a given point in the structure. The freedom to choose the external
loading allows us to do so.
Hence, to calculate the displacement (respectively, rotation) of a given node in
the truss, we will follow these steps:
1. Apply one single external force (respectively, moment) of unity value, at the
node whose displacement is sought, say Q.
2. This virtual force will be oriented in the direction in which we are interested
(typically, vertical).
3. Under these assumptions, Eq. (4.3) will look like

N Nψ
1 · uQ = εx N ψ d x = d x,
nbars 0 nbars 0 AE
where u Q is precisely the sought displacement at node Q and nbars the number
of bars in the structure. Note that, due to the freedom provided by the Principle
of Virtual Forces, we have chosen p ψ = 0.
Note that there is one exception to the criterion of “one single virtual load”.
If the original structure is symmetric, and is subjected to symmetric loading,
it is often preferable to apply two symmetric loads so as not to destroy the
symmetry. In that case, the fact of having two loads must be taken into account,
of course, since the sought displacement will be affected by a factor of two.
Note also that the application of the Principle of Virtual Work needs, for the
solution of the structure (determination of internal forces), two things: one for the
original, actual loading, and a second one for the virtual loading. This last item is,
in general, very simple, since the applied load is composed in general by one single
force. Once the practitioner has acquired some dexterity, it will be evident for him
or her how best to solve the resulting system.
As an example of the just proposed technique, consider the following structure:
We are going to compute the vertical displacement at point E of the structure.
To that end, we must first analyze the structure of Fig. 4.17 and obtain its internal
forces:
√ √
3 3
NC D = − P NE D = − P
2 2
√ √
3 1 3 3
NC E = P NE B = + P
2 2 4
√ √
1 3 3
N AE = + P N AC = − P
2 4 2
4.5 Serviceability Limit-State Analysis of Trusses 91
Fig. 4.17 An illustration of

the Principle of Virtual Work
in the determination of C P D
displacements
P/2 A E B P/2
P
VA VB
HA
Fig. 4.18 Virtual loading for C D

the example of Fig. 4.17
A E B
√
5 3
NDB =− P
6
Once the true internal forces have been determined, we proceed to establish the
virtual loading for the structure. In this case, of course, it will consist of a single load
at node E, see Fig. 4.18.
In this case, the internal forces result in
ψ ψ ψ ψ ψ
√
N AC = NC D = −NC E = −N E D = N D B = − 3/3,
and √
N AE = N E B = 3/6,
where, again, we highlight the virtual character of these forces by using a superscript
ψ. Therefore, the vertical (downwards) displacement at node E will be
√ √ √ √ √ √
1 3 3 3 3 3 5 3
u v (E) = P+ P+ P+ P+ P
AE 3 2 2 2 2 6
√ √ √ √
3 1 3 1 3 3 20 + 3P
+ + P+ + P = .
6 2 4 2 4 6AE
92 4 Trusses
Fig. 4.19 Influence of the orientation of the diagonals. Note how a change in their orientation
changes drastically the sign in the internal forces. Red lines indicate traction while blue, dashed
lines indicate compression
4.6 Criteria for the Design of Trusses
So far, we have studied how to ensure an adequate behavior so as not to reach limit
states. It is interesting, however, to develop some intuitive knowledge about how
trusses work and how their different parameters affect their serviceability.
4.6.1 Influence of the Height and the Span
For instance, it is easy to verify that the horizontal internal forces in the bars are
roughly proportional to the span of the truss. This means that doubling the span also
doubles these forces. Conversely, halving the rise will lead to doubling the forces
again.
Indeed, it is easy to understand from the Cremona diagram that internal forces in
the truss only depend on the direction of the bars. Therefore, for constant external
loads, doubling both the rise and the span of the truss leads to exactly the same
Cremona diagram.
4.6 Criteria for the Design of Trusses 93
Fig. 4.20 Long-type truss
Fig. 4.21 K-diagonal truss
4.6.2 Influence of the Orientation of Diagonals
While in general the upper chord will be subjected to compression and the lower one
to tension, the orientation of diagonals completely determines its internal forces. This
is more easily seen in the so-called Howe (or N-) trusses,1 see Fig. 4.19. In it, tension
forces have been represented by a red line with a thickness roughly proportional to
the intensity of the force. Compression forces are represented by a dashed, blue line,
with thickness proportional again to the intensity of the force.
It is worth noting also that the force in the diagonals is inversely proportional to
the internal angle, θ , see Fig. 4.19. Therefore, the smaller the angle, the bigger the
internal force. However, approaching θ to 90◦ makes the force in the struts (vertical
bars) greater.
1 Actually first proposed by Andrea Palladio in 1570 in his four books on architecture.
94 4 Trusses
Fig. 4.22 Poncelet truss
4.6.3 Various Designs
There are many various designs of trusses (most of them patented during the 19th
century), with different characteristics, but whose overall behavior obeys the just-
introduced criteria.
Apart from Howe trusses, just described, the so-called Long system (or X diago-
nals), see Fig. 4.20, is very popular. Note that this type of truss is statically indeter-
minate, and therefore will require some more sophisticated method of analysis than
the ones considered herein.
Another very popular type of trusses is the so-called K-diagonal truss, see
Fig. 4.21. Despite its more or less complex appearance, it is statically determinate.
During the 19th century, the Poncelet truss became very popular due to its very
light appearance, and it was profusely employed in train stations across Europe, see
Fig. 4.22.
In general, albeit obvious, care must be taken when reversing the sign of the loads.
This is very often the case for suction-wind loads. Therefore, nothing guarantees that
a diagonal bar will always be under traction. This is dangerous since compression
could eventually lead to buckling, something that is beyond the scope of this book,
but that must definitely be taken into account.
4.7 Space Trusses
In modern architecture, trusses are widely employed for the construction of

long-span, three-dimensional structures. The naive approach to construct a
three-dimensional truss is simply to put various trusses in parallel, making them
4.7 Space Trusses 95
Fig. 4.23 An example of three-dimensional truss obtained by juxtaposition of two-dimensional

ones. Charles de Gaulle Airport, Paris
Fig. 4.24 Three-

dimensional lattice
truss
work together just by putting secondary beams between them. This is the system
pursued in the Charles de Gaulle Airport in Paris, for instance, see Fig. 4.23.
A second, more elaborate, possibility, is to construct the three-dimensional struc-
ture by creating a lattice, see Fig. 4.24.
96 4 Trusses
Worked Examples
4.1 Given the truss in Fig. 4.25, compute the internal forces N6 and N19 , taking into
account that the vertical and horizontal dimensions of the bars are L.
Solution 4.1 First of all, and with an eye towards the minimization of work, note
that the structure is symmetric, and so is the load. Therefore, we will have,
P
RL = R R = .
2
(This same result could have been obtained by enforcing the vertical equilibrium of
loads plus the equilibrium of moments.)
To avoid the enforcement of equilibrium equations node by node, until we arrive
at bar 6, we can employ the method of sections. We thus perform an imaginary cut
through the structure (dashed line in the figure) and, by keeping only the right part
of the structure, we proceed to enforce global equilibrium for this part (Fig. 4.26).
Enforcing the equilibrium of moments at points A and B, respectively, we obtain,
P
Mz (A) = · 5L + N6 · L = 0,
2
P
Mz (B) = · 6L + N19 · L = 0.
2
P
1 2 3 4 5 6 7 8 9 10 11 12
24 23 22 21 20 19 18 17 16 15 14 13
RL RR
Fig. 4.25 Truss for Example 4.1
B
N6
NAB
N19
A
P
2
Fig. 4.26 Enforcement of equilibrium through the method of sections

2 kN
2 kN 2 kN
H
1 kN 1 kN
2m
F G
1m
A E
B C D
RA RE
2m 2m 2m 2m
Fig. 4.27 Example 4.2
Solving the resulting system of equations,
5
N6 = − P (compression),
2
N19 = 3P (traction).
One can arrive, in this case, at exactly the same result, by enforcing force equi-
librium.
4.2 Determine the vertical displacement of node C in the truss in Fig. 4.27.
Solution 4.2 (a) We begin by enforcing equilibrium in the structure globally. This
will allow us to determine the reaction forces in the supports.
Given the symmetry of the structure and the lack of horizontal loads, vertical
reactions can be found by simply noting that the sum of vertical forces is 1 + 2 +
2 + 2 + 1 = 8 kN, so that
8kN
R A = RE = = 4 kN.
2
By next applying the equations of (internal) equilibrium at the nodes of the truss,
we are in the position to find the internal forces at each bar. In this case, the application
of equilibrium of moments is simply nonsense, since every force acting on a node
passes inevitably through it, thus generating no moment. Therefore, we must proceed
by isolating and equilibrating nodes for which at most two forces are unknown.
Otherwise, we will not be able to find their values.
98 4 Trusses
Fig. 4.28 Equilibrium at 1 kN F

node A
NAF
α B
A
NAB
4 kN
By assuming an outgoing direction (from A to B and F) for the unknown internal

forces, we are in fact assuming traction. If, after solving the resulting system of
equations, any of the internal forces appear with a negative sign, this means that our
initial assumption was false and that there is actually compression instead of traction
in that particular bar.
The angle α between N AB and N AF verifies (see Fig. 4.28) sin α = √15 and cos α =
√2 , such that
5
3·2 1 √
FY = 0 ; + N AF · √ = 0 ; N AF = −3 5 kN (Compression),
2 5
2
FX = 0 ; N AB + N AF · √ = 0 ; N AB = 6 kN (Traction).
5
Working now on B (Fig. 4.29), it is straightforward to obtain N BC = 6 kN (Trac-

tion) and N B F = 0 kN. By next equilibrating node F, see Fig. 4.30, we obtain the
following system of equations:
Fig. 4.29 Equilibrium of F

node B
NBF C
A
6kN NBC
B
Fig. 4.30 Equilibrium at 2Kn

node F H
NF H
F α
α
√ α
3 5
NCF
A C
B
2 √
FX = 0 ; √ NC F + N F H + 3 5 = 0,
5
1 √
FY = 0 ; √ − NC F + N F H + 3 5 = 2,
5
that, after solving, provides,

√
N F H = −2 5 kN (Compression),
√
NC F = − 5 kN (Compression).
Analogously for node H :
Fig. 4.31 Equilibrium at 2 kN

node H
H
α α
√
2 5 NHG
F
G
NCH
Solving the equilibrium equations at this point gives

√
N H G = −2 5 kN (Compression),
NC H = 2 kN (Traction).
By simply imposing the symmetry of the structure, we find the rest of forces in
the structure (Fig. 4.31).
(b) To determine the vertical displacement at point C, we need to define an appro-
priate virtual loading system. It will be composed of a single load, of unity value,
applied at point C, precisely the one whose displacement is to be computed. For that
purpose, we consider the system in Fig. 4.32.
By using the same methodology as in the previous section, we obtain
ψ ψ 1
R A = RE = kN.
2
Equilibrium at node A is shown in Fig. 4.33:
√
ψ 5
N AF =− kN (Compression),
2
100 4 Trusses
2m
F G
1m
A E
B C D
ψ ψ
RA RE
1 kN
2m 2m 2m 2m
Fig. 4.32 Virtual system to compute the vertical displacement of node C in Example 4.2
ψ
N AB = 1 kN (Traction).
Again, equilibrium at node B provides (see Fig. 4.34):

ψ
N BC = 1 kN (Traction),
ψ
N B F = 0 kN.
Equilibrium at node F is shown in Fig. 4.35:

√
ψ 5
NC F =− kN (Compression),
2
ψ
N F H = 0 kN.
Fig. 4.33 (Virtual) F

Equilibrium at node A ψ
NAF
α B
A
ψ
NAB
0.5 kN
Fig. 4.34 Equilibrium at F

node B ψ
NBF C
A
1 kN ψ
B NBC
Fig. 4.35 Equilibrium at

ψ H
node F NF H
F α
√
α α
5
2
ψ
A NCF C
B
Fig. 4.36 Equilibrium at H

node H
√ α α
5 ψ
F 2
NHG
G
ψ
NCH
Table 4.2 Internal forces in the structure

Bars L (m) N (kp) L · E A (kp · m) N ψ (kp)
AB and DE 2 6 12 1
BC and CD 2 6 12 1
√ √ √
AF and EG 5 −3 5 −15 − 5
√ √ √2
FH and HG 5 −2 5 −10 − 25
BF and DG 1 0 0 0
√ √
CF and CG 5 − 5 −5 0
CH 2 2 4 1
Equilibrium at node H gives, see Fig. 4.36:

ψ
NC H = 1 kN (Traction),
√
ψ 5
NH G =− kN (Compression).
2
In brief, the just obtained virtual internal forces are summarized in Table 4.2.
To determine the vertical displacement in C, it suffices to apply the Principle of
Virtual Work:
ψ ψ
Fext · δr eal = Nint · L r eal ,
thus giving
√ √
1 5 5 1 √
1 · u y (C) = 4 · 12 + 2 · 15 + 2 · 10 +4 = 52 + 25 5 m.
EA 2 2 EA
Chapter 5
Beams (I): Geometric Description and
Internal Forces
Abstract This chapter introduces the basic aspects underlying structures made of
beams.
5.1 Introduction
In general, a beam can be considered as a member of a structure subjected to bending.

Although other members, such as supports and columns, can also be subjected to
bending, the underlying physics is essentially the same, and we will consider them
under the general class of beams.
Understanding this physics can be approached from various points of views. To
begin with, consider again the Lower Bound Theorem introduced in Chap. 3. It
states, remember, that “a load system, based E. Cueto and D. Gonzálezon a statically
admissible stress field which nowhere violates the yield condition is a lower bound
to the collapse load”. Therefore a simple and instructive way of analyzing beams is
to locate in their interiors a truss in equilibrium with exterior loads, see Fig. 5.1.
If, following the method of sections, we isolate some portion of the beam, see
Fig. 5.2, we discover the presence of a pair of forces, one compression (corresponding
to the blue thrust in Fig. 5.1) and one traction (the red thrust in the same figure).
These internal forces T (which are not the actual ones, as we will see later in
this chapter, but are equivalent under the assumption of the lower bound theorem)
generate a new type of internal moment M = T · h, which we will refer to as bending
moment. The presence of bending moments is precisely what makes beams different
from the already studied elements, such as cables, arches and trusses.
There is another, alternative, way of understanding the presence of bending
moments. Recall Chap. 3 and the fact that under a geometry dictated by the funicular
polygon, they are subjected to compression only. But, what happens if the funicular
polygon does not pass through the section? See Fig. 5.3 for a response.

104 5 Beams (I): Geometric Description and Internal Forces
P P
Fig. 5.1 Lower bound theorem for analysis of a beam
Fig. 5.2 Lower bound

theorem for analysis of a P T
beam (continued)
M
h
Fig. 5.3 Portion of an arch

whose funicular polygon
does not pass through the
section
N
N
N
Let us assume that the internal force passes at a distance e from the center of
gravity of the cross section. By adding a pair of mutually equilibrating forces at the
center of gravity (in red and blue in Fig. 5.3), the equilibrium is guaranteed. It is
clear then that the original situation is equivalent to having an axial internal force N
(in red in the picture) plus a bending moment M = N · e.
In what follows we introduce a rigorous mathematical description of the geometry
of a beam that will help us to introduce, also in a very rigorous way, the series of
internal forces and moments that a beam may suffer.
5.2 Geometrical Description of a Beam 105
5.2 Geometrical Description of a Beam
For the geometrical description of a beam, we follow classical references in the field
such as [1, 13] or [19]. For these authors, a beam is an example of a Cosserat medium,
i.e., a flexible line whose points join the centers of gravity of all the beam’s sections
that can rotate around these center points. More precisely, a beam will be described
by:
1. A curve ϕ(t) that joins the centers of gravity of all the beam’s cross sections. We
will employ a parameter t to travel along this line:
ϕ : I ⊂ R → R3 .
Of course, this parameter will nearly always be the arc-length parameter of the
curve.
2. A family of flat cross sections A(t).
It is then clear
that the volume of the bar will be the union of all the cross-
sectional areas, A(t). Also, the directrix coincides with the line of centroids, as
t
just introduced, so that
xd A = ϕ(t)A(t).
A(t)
Very often, intrinsic coordinates are employed in the description of the move-
ment, see Fig. 5.4. These correspond to the so-called Frenet–Serret trihedron. In this
classical mathematical parametrization of curves, the first axis (tangent) is always
oriented along the tangent of the curve, the second one (normal) is oriented towards
the derivative of the curve with respect to the arc-length parameter t and the third
one (binormal) results from the cross product of both. These form, in any case, an
orthonormal axis of reference.
Fig. 5.4 Intrinsic

coordinates of a beam. A
cross section at a generic
position t, A(t), is also
highlighted
A(t)
y
x
5.3 Internal Forces and Moments
As already mentioned in Chap. 2, the distinctive characteristic of the strength of

materials is precisely the fact of working with an intermediate variable, distinct from
the stress. These are the so-called internal forces (and moments). So far, for cables,
arches and trusses, no other internal forces other than the axial one existed. However,
as we are about to discover, in the case of beams, up to six different internal forces
can appear. Their description is the objective of this section.
A general analysis procedure will consist, therefore, of the determination of the
internal forces1 that result from the external loads applied to the structure. Once
determined, these will lead, under some kinematic and static assumptions that will
be deeply analyzed in the following chapters, to the determination of displacements,
strains and, of course, stresses.
Given the six distinct components of the stress tensor, recall Eq. (1.5), we can
define the following vector of internal forces as

T (t) = td A = σn(t)d A,
A(t) A(t)
where n(t) is the vector normal to the section of the bar and t the stress vector, which
is a function of the particular point in the section, x.
Note that, if we employ the intrinsic coordinate system and perform an

imaginary cut across the bar, perpendicular to the x-axis, then n = e1 (t) =
(1, 0, 0)T , the vector tangent to the x axis. In that case, the product σn gives,
in a straightforward manner:
⎞ ⎛ ⎛ ⎞
σx N
T (t) = σn(t)d A = σe1 (t)d A = ⎝τ x y ⎠ d A = ⎝ V y ⎠ ,
A(t) A(t) τ Vz
A(t) xz
providing a much more useful expression of the internal force vector. In this
case, N is the already well-known axial internal force, while Vy and Vz are the
shear internal forces. Many textbooks also employ the notation Vy = Q y and
Vz = Q z .
In a similar manner, we will define the vector of internal moments as the result of
integrating over the section the stress vector at each point, multiplied by its distance
to the centroid of the section, i.e.,
1 When there is no risk of confusion, we will speak of “internal forces” to refer to both internal
forces and moments.
5.3 Internal Forces and Moments 107

M(t) = (x − ϕ(t)) × td A = (x − ϕ(t)) × σnd A
A(t) A(t)
⎛ ⎞
σx
= (x − ϕ(t)) × ⎝τx y ⎠ d A,
A(t) τx z
with x an arbitrary point of the beam’s cross section and ϕ(t) the point of the line
of centroids corresponding to A(t).
Again, when the identity n = e1 holds, the vector of internal moments will be:
⎞ ⎛ ⎞ ⎛
σx Mx
M(t) = (x − ϕ(t)) × σnd A = (x − ϕ(t)) × ⎝τx y ⎠ d A = ⎝ M y ⎠ ,
A(t) A(t) τx z Mz
where Mx is the so-called torsion or twisting moment—sometimes denoted as

T —and M y and Mz are the bending moments.
In sum, the complete picture of internal forces in a bar can be sketched as:

T1 = N = σx d A,
A(t)

T2 = Vy = τx y d A,
A(t)

T3 = Vz = τx z d A,
A(t)

M1 = M x = (τx z x y − τx y x z )d A,
A(t)

M2 = M y = σx x z d A,
A(t)

M3 = M z = − σx x y d A.
A(t)
These internal forces are depicted in Figs. 5.5 and 5.6, respectively. Note the right-
hand convention for the rotation of the moments.
Fig. 5.5 Internal forces at a

beam’s cross section
Vy
Vz
y
N
x
Fig. 5.6 Internal moments

at a beam’s cross section
Mz
y
My
x
Mx
5.4 Diagrams of Internal Forces and Moments
We have already mentioned the fact that introducing a new variable, the internal
forces, in the problem, constitutes in fact a true advantage for the solution of some
problems. This will appear clear in subsequent chapters of this book. For the time
being, one of these advantages is the possibility of drawing internal force and moment
diagrams. These will allow us to obtain an overall description of the degree of loading
of each member of the structure.
To obtain these very useful diagrams, it is enough to proceed as in Sect. 4.4.3
and apply the method of sections, by enforcing equilibrium on arbitrary portions of
the structure (say, from x = 0 to x = ). This will provide us with an equilibrium
equation for internal forces of the type
T () − T (0) + pe = 0, (5.1)

5.4 Diagrams of Internal Forces and Moments 109
Fig. 5.7 An introductory p

example
B
A

with pe the vector of external applied loads in the bar portion x ∈ [0, ], i.e.,

pe = ds Xd A + ds td∂ A,
0 0
A(t) ∂ A(s)
with X the force per unit volume and t the force per unit length applied to the bar in
the portion x ∈ [0, ]. ∂ A(s) represents the boundary (external surface) of the bar in
this region. Note the vector character of these equations.
For the moment diagram, we proceed in exactly the same manner, so as to obtain:

ds (x − ϕ()) × Xd A + ds (x − ϕ()) × td∂ A
0 A(s) 0 ∂ A(s)

+ (x − ϕ()) × td A + (x − ϕ()) × td A = 0. (5.2)
A(0) A()
It is then clear that, knowing the internal forces at some point in the structure
(reactions in one of the supports, for instance) will allow us to apply successively
Eqs. (5.1)–(5.2) to obtain, directly from the external loads, and by equilibrium con-
siderations only, the internal force and moment diagrams.
5.4.1 An Introductory Example
Consider the simple structure depicted in Fig. 5.7.

To obtain the internal force diagram, we proceed by first determining the reactions
at the supports, see Fig. 5.8.
Note that, in this case and always where a free, cantilevered beam end appears,
it is not necessary to begin by determining the value of reactions at the sup-
ports. This is due to the fact that a free beam end (with no punctual load nor
prescribed moment) has no internal forces or moments. Thus, we can begin
our calculations by assuming all internal forces to be identically zero at B.
Fig. 5.8 Reactions at the

supports
RAH B
MA
RAV A

Fig. 5.9 Application of the

method of sections to
determine internal forces
V (x) B
M (x)
N (x)
x
At point A, three reactions will appear, namely, a vertical reaction R AV , a horizon-

tal reaction, R AH and a moment M A . There are no other internal forces or moments
since the structure is two-dimensional. We have placed them in Fig. 5.8 pointing
in arbitrary directions. The application of equilibrium equations will tell us if our
choice was correct or not by assigning a “+” or “−” sign to the results.
Therefore, by applying equilibrium along two arbitrary directions (say, vertical
and horizontal, for simplicity):

FH = 0, → R AH = 0,

FV = 0, → R AV = pd x = p,
0

2
MA = 0 → MA − pxd x, → M A = p ,
0 2
where we have taken A as the reference point with respect to which we compute
moments. This is, of course, an arbitrary choice. The sign for the moments has been
taken, also in an arbitrary manner, as positive in the counter-clockwise sense. The
method will work equally well with completely opposite sign conventions.
Once we have determined the reactions at the support points, we can proceed
by the method of sections to equilibrate arbitrary portions of the structure. Thus,
by cutting the structure at a point at an arbitrary distance from the free end, B, we
unveil (see Fig. 5.9) three internal forces: N (x), V (x) and M(x). We highlight the
dependence of these internal forces on the particular, but arbitrary, position x. These
represent the effect of the just eliminated portion of the structure on the conserved
one.
By applying again equilibrium equations on the portion of length x, we obtain:
B B
A A

Fig. 5.10 Internal force diagrams. Left, bending moments. Right, shear force

FH = 0, → N (x) = 0,
x
FV = 0, → V (x) = pd x = px,
0
x
x2
Mx = 0 → M(x) − pxd x, → M(x) = p ,
0 2
where an abuse of notation has been made by naming x as both the variable to
integrate and the integration limits. Note that we have just determined that all the
internal forces are positive. This means nothing more than that we have correctly
assumed their sense.
It is worth noting that, since we are interested in computing the bending moment
at the cutting point x, moments must be computed with respect to precisely this point,
and not with respect to an arbitrary point, as in the global-equilibrium enforcement
for the computation of the reactions.
Note also that we have just determined that the diagram of shear internal force
varies linearly with x, while the diagram of bending moments varies quadratically
with x, see Fig. 5.10.
In the diagrams in Fig. 5.10, we include a small symbol indicating the sense
of both internal forces. These represent what is physically happening at the
beam. For instance, in the bending moment diagram, Fig. 5.10, left, the symbol,
composed by a small portion of the bar and two mutually equilibrating arrows
showing the sense of bending, indicates that the bending moment is actually
applying traction at the upper fiber of the beam and compression to the lower
one. How do we know the precise sense of the arrows? It is simple: just take
a glimpse to the equilibrium of the beam portion in Fig. 5.9 and notice that,
on the left side of the cut, there is a counter-clockwise bending moment. Place
an equivalent arrow on the left side of the small rectangle and complete the
symbol with the opposite arrow on the right.
Moments applying traction on the upper fiber of a beam are traditionally
considered negative, but this is only an old convention, and the contrary is
equally valid. We prefer to indicate clearly the sense of the forces and moments
by employing these small symbols.
Fig. 5.11 A case with a

punctual load
RAH P
MA
RAV A B

Fig. 5.12 Application of the

method of sections to
determine internal forces
V (x) P
M (x)
N (x) B
x
5.4.2 A Second Introductory Example
Consider now what happens if, instead of a distributed load p, a single punctual load
P is acting at point B in the structure, see Fig. 5.11. We begin by determining the
reactions at the clamped support A.
In this case,

FH = 0, → R AH = 0,

FV = 0, → R AV = P,

M A = 0 → M A = P.
By employing again the method of sections, see Fig. 5.12, we arrive at:

FH = 0, → N (x) = 0,

FV = 0, → V (x) = P,

Mx = 0 → M(x) = P x,
B B
A A

Fig. 5.13 Internal force diagrams. Left, bending moments. Right, shear force
In this second example, we can readily see that the shear internal force is constant
(and equal to P), while the bending moment varies linearly with x. See the resulting
diagrams in Fig. 5.13.
Again, we have plotted internal-force diagrams at the top of the beam. The same
tradition that indicates that bending moments are positive if they apply traction
on the upper fiber of the beam indicates that this diagram should be placed
on the “positive” side of the beam, i.e., in the upper part of it. Nevertheless, it
is nonsense to attribute any particular sign to the diagram, since it should be
valid for any of them. Therefore, you can place the diagram above or below
the beam, but always accompanied by the symbol of the moments that actually
take place.
In this case, the change from a distributed load per unit length to a punctual load
has necessitated also a change in the force diagrams. The bending moment diagram
has changed from a quadratic one (Fig. 5.10) to a linear one (Fig. 5.13), while the
shear-force diagram has changed from linear to constant. Of course, this behavior is
completely general and will happen every time we consider a punctual load in our
structure.
5.5 Gerber Beams
Static indeterminacy of beams means in practice that several bending-moment dia-

grams comply with a condition of equilibrium, see Fig. 5.14. Therefore, kinematic
admissibility and constitutive laws must be taken also into account to obtains the
correct solution to the problem.
There is a very clever way to design statically determinate beam structures that
are very efficient since they impose the equality of bending moments on both sides
of the beam (the so-called positive and negative moments, albeit this distinction
is nonsense, as we have mentioned earlier). These are called Gerber beams and
are widely employed in pre-cast concrete structures, as well as steel structures, see
Fig. 5.15.
Fig. 5.14 Various bending-moment diagrams that verify equilibrium

MM
0.85
Fig. 5.15 Principle of Gerber beams. Hinges are located at = 0.15
5.6 Frames
A frame is a structure composed of one or more beams and columns or side walls,
see Fig. 5.16. Essentially, their functioning is the same as an arch whose funicular
polygon does not pass through the bars, thus causing bending.
Three-hinged arches or frames (see also Fig. 3.12) are widely used due to their
static determinacy. For such a system, the funicular polygon must pass through the
three hinges, and it is therefore well-known in advance, see Fig. 5.17.
To analyze this frame, first note that the number of unknown reactions at the
supports (two at A, two at E) will not allow, in principle, to determine them through
Fig. 5.16 A sequence of frames for a concert hall at Palacio de las Artes Reina Sofía. Valencia,
Spain. S. Calatrava, architect
5.6 Frames 115
B D
C
h
A E
HA HE
VA VE
Fig. 5.17 An example of three-hinged frame. In blue, the funicular polygon of the distributed load,
that must pass through the three hinges. When the funicular polygon extends beyond the bar sections
(almost everywhere in this case), bending appears
equilibrium equations only (three equations). However, there is an additional hinge

at C, that incorporates an additional equation (vanishing sum of moments for the
C − D − E or C − B − A half arches). By applying equilibrium, we have:

FH = 0, → H A = HE , (5.3)

FV = 0, → V A + VE = q (5.4)

→ q 2 = VE ,
2
MA = 0 (5.5)
where we have taking moments with respect to point A; so that the line of action of
reactions V A , H A and HE passes through point A and therefore produces no moment.
We have three equations for four unknowns, so we need one more equation. This
will come from the hinge at C:
2
MC = 0, → q + HE h − VE = 0.
4 2
This last equation takes C (the hinge) as the reference point and includes forces and
moments coming from one of the bars starting at C. If we include the whole structure,
we will obtain an equation linearly dependent on Eq. (5.5).
From the moments equilibrium equation at A we obtain

VE = q ,
2
so that, see Eq. (5.4),

VA = q ,
2
which also verifies the symmetry of the structure and its applied loads. From the
moment equilibrium equation at C we obtain
1 2 2
HE = H A = q −q = 0.
h 4 4
Once the reactions have been determined, we can begin to obtain the internal
force diagrams by applying the method of sections. To that end, a suitable option is
to begin by performing an imaginary cut into the bar A − B, see Fig. 5.18.
Under these assumptions, we have:

FH = 0, → N (x) = −V A = −q ,
2
FV = 0, → V (x) = H A ,

Mx = 0 → M(x) = H A x,
where positive moments have been assumed to rotate in the counter-clockwise sense,
but exactly the opposite convention could have been adopted as well. Note also that
the lines of action of V A , V (x) and N (x) pass through the cutting point, so that they
produce no moment from x. These expressions are valid for any x ∈ [0, h], so for
any point in the bar A − B.
Fig. 5.18 First imaginary M (x)

cut into the bar A − B
V (x) N (x)
x
HA
VA
5.6 Frames 117
Fig. 5.19 Second imaginary

cut, along B − C, for the
determination of the M (x)
internal-force diagrams N (x)
B
V (x)
h
A
HA
VA x
To obtain the diagrams along the bar B − C, we perform a second imaginary cut
at an arbitrary position x.2 See Fig. 5.19 for more details.
In this case,

FH = 0, → N (x) = −H A ,

FV = 0, → V (x) + q x = V A ,
x2
Mx = 0 → M(x) = q − V A x + H A h.
2
It is worth noting that now both the internal axial force N (x) and the shear force
V (x) adopt a different direction than those of the bar A − B. Internal forces are
always relative to local coordinate axis, i.e., they depend on the orientation of the
particular bar.
Since the just-obtained diagrams are valid for any point x ∈ [0, /2], and taking
into account that the structure is symmetric and is subjected to symmetric loads, there
is no need to continue our quest for the internal-force diagrams. We have completed
our task. The only work remaining is to plot these diagrams, see Fig. 5.20.
2 Note that we should have taken a second coordinate x , different from x. Again, when there is no
risk of confusion, we adopt x as the default nomenclature for our through-the-bar coordinate.
B D
C
A E
Fig. 5.20 Resulting bending-moments diagram for the structure in Fig. 5.17
It is extremely important to note that hinges produce a point at which bending

moments vanish. This is so since they allow (to some extent) free rotation of
the bars around it. It is not mandatory to construct joints as those in Fig. 4.10,
for instance. Very often, a simple change of stiffness in the bars behaves as a
hinge.
5.7 Vierendeel Beams
If we eliminate the diagonals in a truss, we obtain a structure that closely resembles

a frame, see Fig. 5.21. One or more frames arranged in a horizontal manner is
able to carry vertical loads and is named a Vierendeel beam in honor of Arthur
Vierendeel (1852–1940), a Belgian engineer. He thought that “when constructing in
metal the dimensions must be determined a priori by aesthetic considerations and
only afterwards should mathematical formulas be used”, which seems to be in line
with modern architecture in many well-known examples, but has often led to many
problems in practice.
5.7 Vierendeel Beams 119
Fig. 5.21 A segment of Vierendeel beam at Incheon airport, Seoul. Note that the U-shaped, three-
dimensional truss is properly a truss. However, the segment in the center of the picture has no
diagonals, thus developing strong bending moments at the nodes
Worked Examples
5.1 Determine the internal force diagrams for this structure (Fig. 5.22).
100 kp
B
C
1m
500 kp/m
1m
A
2m
Solution 5.1 At the clamped support C, reaction forces can be obtained by enforcing
equilibrium:
Rx = −500 kp (←)
R y = 100 kp (↑)
Mz = −500 · 1.5 − 100 · 2 = −900 kp · m (clockwise)
If we perform a virtual cut in the bar A − B, at an arbitrary location where the load
q is applied, say x ∈ [0, 1], the resulting free-body diagram is depicted in Fig. 5.23.
For these forces, we enforce equilibrium, see Fig. 5.23, so as to yield
N (x) = 0 kp, ∀x ∈ [0, 1].
For the horizontal equilibrium,
V (x) = 500 · x kp, ∀x ∈ [0, 1].
By equilibrating moments at the section where the virtual cut is performed, x,
x2
M(x) = −500 · kp · m, ∀x ∈ [0, 1].
2
Fig. 5.23 First imaginary M (x)

cut in the bar A − B
V (x) N (x)
500 kp/m
x
A
A second virtual cut in the bar A − B, this time on a section after the load q,
x ∈ [0, 1], provides the free-body diagram depicted in Fig. 5.24.
Again, we enforce equilibrium. First, in the vertical direction, so as to obtain:
N (x) = 0 kp, ∀x ∈ [0, 1] in local coordinates (see Fig. 5.4).
Fig. 5.24 Second imaginary M (x)

cut in the bar A − B
N (x)
V (x)
x
500 kp/m
1m
Along the horizontal direction,
V (x) = 500 · 1 kp = 500 kp (constant).
Finally, equilibrating moments with respect to the x position gives:
M(x) = −500 · 1 · (x + 0.5) kp · m, ∀x ∈ [0, 1].
Finally, a third cut along B − C, see Fig. 5.25, is performed. In this case
(Figs. 5.26, 5.27 and 5.28):
N (x) = −500 · 1 kp, ∀x ∈ [0, 2],
V (x) = −100 kp, ∀x ∈ [0, 2],
M(x) = 500 · 1.5 + 100 · x = 750 + 100x kp · m, ∀x ∈ [0, 2].

Fig. 5.25 The third cut, 100 kp

along B − C, enables
determination of the internal- N (x)
force diagrams M (x)
B
V (x)
1m
500 kp/m
1m
A
x
Fig. 5.26 Axial internal

forces in Example 5.1 B
C
500 kp
1m
500 kp
1m
A
2m
Fig. 5.27 Shear internal 100 kp 100 kp

forces in Example 5.1
B
500 kp
C
1m
500 kp
1m
0 kp A
2m
950 kp · m
750 kp · m
B
750 kp · m
C
1m
250 kp · m
1m
0 kp · m A
2m
Fig. 5.28 Bending-moments diagram in Example 5.1
5.2 Determine the distribution of internal forces for the structure in Fig. 5.29.
Solution 5.2 As we have done repeatedly, we first equilibrate the whole structure
so as to determine the reactions at A:
RxA = 0 kp,
R yA = 0 kp,
and, in the support C,
A
C
2m
500 kp/m
B
2m D 2m E

R Cy = 2000 kp (↑).
Since no horizontal load exists, the sum of vertical reactions must equilibrate the
vertical loads, 2000 kp. By taking moments at A, for instance, we find
R Cy · 2 = 500 · 4 · 2.
Performing now a virtual cut at an arbitrary position along A − B, first in the zone
where no load exists, we conclude that no internal force exists along A − B.
Given that C − D possesses two hinges at the ends (it is, therefore, a cable), it
will suffer only axial internal forces. This force is identical, obviously, to the reaction
at support C.
To determine the internal forces at B − D we perform a new cut, as sketched in
Fig. 5.30.
From the horizontal equilibrium of forces, we have
N (x) = 0 kp, ∀x ∈ [0, 2].
In the vertical direction,
V (x) = −500 · x kp, ∀x ∈ [0, 2].
Finally, by taking moments at x,
x2
M(x) = 500 · kp · m, ∀x ∈ [0, 2].
2
Fig. 5.30 Third cut in the A

bar B − D
2m
500 kp/m
M (x)
N (x)
B
V (x)
x
To obtain the internal forces in the bar D − E we perform again a virtual cut as
depicted in Fig. 5.31. By equilibrating the forces in the horizontal directions, we
obtain:
N (x) = 0 kp, ∀x ∈ [0, 2].
Fig. 5.31 Fourth cut, along 500 kp/m

D−E
V (x)
E
N (x)
M (x)
x
From the vertical equilibrium we find:
V (x) = −500 · x kp, ∀x ∈ [0, 2].
And, finally, by equilibrating moments at x,
x2
M(x) = 500 · kp · m, ∀x ∈ [0, 2].
2
The resulting internal-force diagrams are depicted in Figs. 5.32, 5.33 and 5.34.
A 2000 kp
C
2m
0 kp 2000 kp
B 0 kp
2m D 2m E
Fig. 5.32 Resulting axial internal-force diagram for the structure in Example 5.2
A
2m C 0 kp
1000 kp
0 kp
B 0 kp
D E
1000 kp
2m 2m
Fig. 5.33 Shear internal forces in Example 5.2
A
C 0 kp · m
2m
1000 kp · m
0 kp · m
B 0 kp · m
2m D 2m E
Fig. 5.34 Bending moments in Example 5.2
Exercise
5.3 Determine the internal forces in the structure in Fig. 5.35.

Fig. 5.35 Determine its P
internal forces
q
A
B C

Chapter 6
Beams (II). Normal Stresses
Abstract This chapter introduces the determination of normal stresses produced by

bending.
6.1 Introduction
We have repeatedly insisted throughout the book that the classical analysis of struc-
tures involves the determination of intermediate variables not initially present in the
Navier formulation of the elasticity problem. These variables are, of course, internal
forces and moments. They are appealing since they can be determined in many cases
from equilibrium alone and, notably, they simplify the subsequent determination of
stresses, deformations and displacements, the ultimate goal of our analysis.
However, as in previous chapters, the reader will notice how, to determine normal
stresses—remember, the σx component in our stress tensor σ—some simplifications
or hypothesis need to be taken into account. These will be the objective of this chapter.
6.2 Euler–Bernoulli–Navier Bending Model
The Euler–Bernoulli–Navier1 theory is the most useful and practical theory for long
beams. By “long” we mean bars whose length is more than, say, fifteen times the
average size of the cross-sectional area.
1 Named in honor of Jacob Bernoulli (1654–1705, who first discovered that the curvature of an
elastic beam at any point is proportional to the bending moment at that point), his nephew Daniel
Bernoulli (1700–1782, the first to formulate the differential equation of motion of a vibrating beam),
Leonhard Euler (1707–1783, who later validated their theory) and, of course, Claude-Louis Navier
(1785–1836), who first established the zero line of normal stress in 1819 and established in 1826
the elastic modulus as a property of materials independent of the second moment of area.
128 6 Beams (II). Normal Stresses
6.2.1 Main Hypotheses
Coming back to the geometrical description of a bar—recall Fig. 5.4—, we have, see
Fig. 6.1, two main hypotheses.
1. A first kinematic hypothesis: every cross section moves as if it were a rigid solid.
2. A second kinematic hypothesis: cross sections initially perpendicular to the
beam’s axis (such as A(t) in Fig. 6.1) remain perpendicular after bending.
3. A static hypothesis: normal stresses σ y and σz , as well as shear stress τ yz are
negligible.
Hypotheses 1 and 2 are easily verified experimentally. See Fig. 6.2 for an example
in which the rigid solid-like movement of flat cross sections of the beam is clear.
Hypothesis 1 is equivalent to neglecting ε y , εz and γ yz components of the strain
tensor, meaning that there is no distortion of the cross section. For the physical
interpretation of these strain terms, the reader is referred to Fig. 1.4. This is also
equivalent to neglecting the Poisson effect in the section [9]. In other words, in spite
of the simultaneous existence of traction and compression stresses in the section—
as we will readily discover—the effect of the Poisson coefficient of contraction
and expansion, respectively, along perpendicular directions, is also negligible. Even
simpler: a rectangular cross section will always remain rectangular and of the same
dimensions. Remember always that we accept the hypothesis of small strains!
Hypothesis 3 is, surprisingly, inconsistent with hypothesis 1—recall Eq. (1.9)—
but it can be experimentally measured that these stresses are at least two orders of
magnitude lower than σx , and therefore they can be neglected if the bars show a
moderate radius of curvature.
Fig. 6.1 Intrinsic

coordinates of a beam. A
cross section at a generic
position t, A(t), is
highlighted
A(t)
y
x
6.2 Euler–Bernoulli–Navier Bending Model 129
Fig. 6.2 Experimental verification of the kinematic hypothesis of the Euler–Bernoulli–Navier beam
theory. Notice how the black, straight thick lines remain straight after the application of the load
6.2.2 Kinematics
Let us translate the just introduced hypothesis into equations, useful for the design
and analysis of structures subjected to bending. We restrict ourselves to straight
bars of constant cross section. Other, more elaborate cases, can easily be treated
numerically by means of, for instance, finite element methods [8].
The starting point for our developments will be hypothesis 1, i.e., that every cross
section moves under bending as a rigid solid. It is a common practice, as already
mentioned, to take the center of gravity of the section as the origin of our system
of coordinates, see Fig. 6.1. The line joining every section’s center of gravity will
therefore define our x axis in a natural way. Under this rationale, the movement of
any point in the beam will follow, under bending, the following expression:
u = uo − r × θ, (6.1)
or, in the various components,
u x (x, y, z) = u ox (x) + θ y (x)z − θz (x)y,

u y (x, y, z) = u oy (x) − θx (x),
u z (x, y, z) = u oz (x) + θx (x).
Here, uo is the vector of displacements for the center of gravity of the section, the
point (x, 0, 0)). In turn, θ(x) is the rotation vector of the section.
For the time being, we are considering bending alone, so that no torsion phenom-
ena exists. This implies that we will restrict ourselves to the case in which θx = 0.
Navier hypothesis (number 2) states that the rotation of the cross section is equal
to the rotation of the beam’s axis. Under the small displacement assumption, accepted
throughout this book, this angle of rotation will be approximated by its tangent and
the tangent as the slope of the beam’s axis. In other words:
du oz du oy
θy = − , θz = .
dx dx
Let us do some gymnastics at this moment. Consider again Eq. (6.1), and introduce
it into the expression of the Cauchy strain tensor, Eq. (1.3). This gives:
du ox dθ y dθz
εx = +z −y ,
dx dx dx
o
1 du y
εx y = − θz ,
2 dx

1 du oz
εx z = + θy ,
2 dx
ε y = εz = ε yz = 0.
Taking into account that, under Navier’s hypothesis:
du oz du oy
= −θ y , = θz ,
dx dx
we arrive at:
εx = εn + zκ y − yκz , (6.2)
εx y = 0,
εx z = 0,
ε y = εz = ε yz = 0,
with
du ox dθ y d 2uo dθz d 2 u oy
εn = , κy = = − 2z , κz = = ,
dx dx dx dx dx2
where the terms κα , (α = y/z), correspond to the first-order approximation of the

beam curvatures along the planes x z and x y, respectively. Indeed, it is straightforward
to obtain that:
d 2 (y+u )
y y d2u 2
κz = dx2
= dx2 ∼ d uy ,
3 =
2 23
2 2 dx2
d(y+u y ) du y
1+ dx
1 + dx
with κz = 1/ρz the curvature of the deformed beam’s axis and ρz its radius of cur-
vature. Remember that we restrict ourselves to straight bars.
6.2.3 Stresses
To obtain the expression for the stresses in the bars, let us introduce in Eq. (6.2) the
constitutive equations for an elastic solid, i.e., Eq. (1.9).
It is extremely important to understand the Euler–Bernoulli–Navier beam

model to be aware of its apparent inconsistencies. Given that εx = 0, and by
virtue of Eq. (1.9), σ y , σz = 0 (recall that θ = εx + ε y + εz ). This sharply con-
tradicts hypothesis 3 of the model. However, as mentioned before, for straight
bars, these stresses are actually several orders of magnitude below σx , and are
therefore neglected.
The stress state in the bar can thus be considered as given by the simple one-
dimensional state:
σx = Eεx = Eεn − E yκz + E zκ y , (6.3)
with τx y = τx z = 0.
Again, a contradiction appears, since it is well-known that shear stresses are not
vanishing in the cross section. It is thus necessary to introduce in the model an
additional static hypothesis (the fourth one) to make the model consistent with the
experimental facts. This will be done in the next chapter, the one in which we will
study the form of shear stresses for beams under bending processes.
6.2.4 Internal Forces
The final objective of this chapter is to obtain a convenient way to establish the
relationship between internal forces and stresses. Otherwise, all the work developed
so far would have been useless, and the establishment of internal forces as a new
variable in the problem is nothing but an exercise of style with no practical relevance.
Thus, coming back to the definition of internal forces in Sect. 5.3, we have,

N = σx d A = (Eεn − E yκz + E zκ y )d A
A A
= E εn (x)A(x) + κ y (x)S y (x) − κz (x)Sz (x)) ,
where A(x) represents the area of the cross section at a distance x along the beam’s
axis, and S y , Sz the first area moments of this cross section, i.e.,

Sy = zd A, Sz = yd A.
A A
But, since we have chosen the center of gravity of the section as the origin of coor-
dinates, these vanish:

Sy = zd A = Sz = yd A = 0,
A A
so that
du ox
N = E Aεn = E A , (6.4)
dx
or, equivalently,
N
εn = ,
AE
where AE, the product of the area of the cross section times the Young’s modulus
of the material is often referred to as the axial stiffness of the bar.
Following a similar procedure, we can obtain the expression of the internal bend-
ing moments in terms of the strain field, viz.:

M y (x) = zσx d A = E εn (x)S y (x) + κ y (x)I y (x) − κz (x)I yz (x) ,
A

Mz (x) = − yσx d A = −E εn (x)Sz (x) + κ y (x)I yz (x) − κz (x)Iz (x) ,
A
where, this time, I y , I yz , and Iz represent the second area moments of the cross
section,
Iy = z 2 d A, Iz = y 2 d A, I yz = yzd A,
A A A
and again
Sy = zd A = Sz = yd A = 0,
A A
since we have established our origin of coordinates at the section’s center of gravity.
Therefore,
d 2 u oz d 2 u oy
M y (x) = E κ y (x)I y − κz (x)I yz = −E I y + I yz , (6.5)
dx2 dx2
d 2 u oz d 2 u oy
Mz (x) = −E κ y (x)I yz − κz (x)Iz = E I yz + Iz , (6.6)
dx2 dx2
and, by solving for the curvatures,
1 M y Iz + Mz I yz 1 Mz I y + M y I yz
κy = , κz = .
E I y Iz − I yz
2 E I y Iz − I yz
2
If we extend the previous procedure for the expression of shear internal forces,
we arrive at:
Vy = τx y d A = 0,
A
and, of course,
Vz = τx z d A = 0.
A
Obviously, this is consistent with a general case of loading such as the one
in Fig. 5.13, for instance. This motivates again the correction of the model that
will be done in the next chapter. Remember the famous aphorism by George
Box: “all models are wrong, but some are useful” (Cf. Box, G. E. P. (1976),
Science and Statistics, Journal of the American Statistical Association, 71:
791–799). This model constitutes a paradigmatic example of this assertion.
Equations (6.5) and (6.6) represent an example of constitutive equations for the
Euler–Bernoulli–Navier model. Substituted into the equilibrium equations give rise
to:
d 2 u ox
EA + px = 0
dx2
d 4 θz d 4θy dm z
E Iz 4 + E I yz + + py = 0
dx dx4 dx
d 4θy d 4 θz dm y
E Iy 4
+ E I yz 4 − + pz = 0, (6.7)
dx dx dx
which are the differential equations describing the Euler–Bernoulli–Navier model.

m y and m z represent the applied moments per unit length, while p y , pz represent the
applied loads per unit length.
6.2.5 Stresses as a Function of Internal Forces
Let us come back for a moment. Do not forget that:
1 M y Iz + Mz I yz 1 Mz I y + M y I yz
κy = , κz = . (6.8)
E I y Iz − I yz
2 E I y Iz − I yz
2
If we recall Eq. (6.3) and introduce it in Eq. (6.8), we arrive at:
N M y Iz − Mz I yz Mz I y + M y I yz
σx = − z+ y,
A I y Iz − I yz
2 I y Iz − I yz
2
which relates the normal stress σx and the internal forces in the beam. This is,
therefore, the final objective of our quest and the true interest in defining internal
forces as a variable in the problem.
If we choose to employ doubly-symmetric profiles for our beams (for instance,

I -shaped profiles, as in Fig. 6.3), in which case I yz = 0, the previous expression
simplifies notably, to render
N My Mz
σx = − z+ y.
A Iy Iz
If, in addition, the structure lies in a plane, and so does its loading, then the
situation becomes even simpler, so as to give:
N My
σx = − z.
A Iy
In this last case, with solely one moment Mα , (α = y/z), the maximum value for
the normal stress and therefore the critical point of the section (at least in absence
of other stresses) will happen at the point farthest from the center of gravity of the
section, i.e., for maximal β, with β = z/y, so as to give:
Mα(α=y/z) Mα(α=y/z)
σxmax = β max (β = z/y) = ,
Iα(α=y/z) Wα(α=y/z)
Fig. 6.3 An example of plane frames built with doubly-symmetric profiles. Fish market in Luanco,
Spain
y
1 2 100 kp
3 cm
3 300 kp z
10 cm
4m 2m
Fig. 6.4 An example of structural design
where we have defined a new parameter:
Iα(α=y/z)
Wα(α=y/z) = ,
|β (β = z/y)|
max
known as the elastic section modulus. This modulus, along with the section area,
first and second area moments, are usually included in tables by steel-profile manu-
facturers so as to simplify the work of structural designers.
Let us solve a simple example of structural design. Consider the beam in Fig. 6.4,
with rectangular section of dimensions 10 × 3 cm2 .
Let us determine the location of the greatest stress in the structure (and therefore
the point at which the failure will begin, according to the classical elastic design
of structures). As we did previously, we begin by finding the reaction forces at the
supports in points 1 and 2:

Fx = 0 ⇒ Rx1 − 300 = 0 ⇒ Rx1 = 300 kp
Fy = 0 ⇒ R y1 + R y2 − 100 = 0 R y1 = −50 kp
⇒
Mz = 0 ⇒ 4 · R y2 − 6 · 100 = 0 R y2 = 150 kp
Once determined, these reactions forces will help to find the expression of the
internal force diagrams. The axial internal force will be constant throughout the bar,
since there is no applied force per unit length:
N (x = 0) = N (x = 6 m) = −300 kp ⇒ N (x) = −300 kp
while the shear force will be:
Vy (x = 0) = 50 kp ⇒ V1 = 50 kp,
Vy (x = 6 m) = −100 kp ⇒ V2 = −100 kp.
Finally, for the bending moment, we have:
Mz (x = 0) = 0 ⇒ Mz = −50x ; 0 < x < 4 m,
Mz (x = 6 m) = 0 ⇒ Mz = 100 (x − 4) − 200 ; 4 m < x < 6 m.
Graphically, these internal forces look as depicted in Figs. 6.5, 6.6 and 6.7.
Fig. 6.5 Bending moments 200 kpm

in the structure of Fig. 6.4
1 2 3
Fig. 6.6 Shear forces in the 50 kp

structure of Fig. 6.4
1 2 3
100 kp
Fig. 6.7 Axial internal 300 kp

forces in the structure of
Fig. 6.4
1 2 3
The interest of internal-force diagrams comes from the fact that they make it
possible to determine, at a glance, the critical section in the structure. In this case, we
notice clearly that this point is point 2. It is subjected to N = 300 kp (in compression)
and M = 200 kpm. For the time being, we omit the effect of the shear forces, which
will be studied in the next chapter.
The geometric characteristics of the cross section of the bar are:
1 10 · 33
A = b · h = 10 · 3 = 30 cm2 , Iz = b · h3 = = 22.5 cm4 ,
12 12
so that the distribution of normal stresses will be:
N Mz −300 −20000
σx = − y= − y = −10 + 888.89y kp/cm2 .
A Iz 30 22.5
This means that the maximum stress in traction will take place along the upper fiber
of the beam:
σx,trac
max
= −10 + 888.89 · 1.5 = 1323.33 kp/cm 2 ,
while the maximum stress in compression will take place along the lower fiber of
the section, namely,
σx,comp
max
= −10 + 888.89 · (−1.5) = −1343.33 kp/cm 2 .
The final distribution of stresses is depicted in Fig. 6.8.
6.2.6 Neutral Axis
As can be noticed from Fig. 6.8, while the stresses produced by bending moments
alone (labelled as σx (M)) vanish at an y coordinate corresponding to that of the
center of gravity of the cross section—in other words, at the x-axis—the addition of
the stresses produced by the axial internal force has caused upward displacement of
the neutral axis.
y
−10 kp 1333.3 kpm 1323.3 kpm
3 cm
σx (N ) σx (M ) σx (N + M ) −1343.3 kpm
Fig. 6.8 Normal stresses at the critical section of the structure of Fig. 6.4 (side view). Diagrams are
not to scale. Note that the sum of the stresses produced by the axial internal force plus the contribution
of the bending moment gives rise to the actual stress distribution in the section (rightmost diagram
in the figure)
The neutral axis is concept of utmost importance and is defined as the locus
of the points of the cross section that have no normal stress:
N M y Iz − Mz I yz Mz I y + M y I yz
− z+ y = 0.
A I y Iz − I yz
2 I y Iz − I yz
2
It can be noticed that this axis passes through the center of gravity of the
section if there is no axial internal force acting on it.
6.2.7 Core of the Cross Section
The neutral axis of the section drives the regions of tension and compression stresses.
Sometimes it is necessary to ensure that the section is exclusively subjected to com-
pressive stresses. This is the case, for instance, in foundations, where it is important
that no tension appears, otherwise detachment of the foundation from the soil can
occur. In concrete structures, a material that does not support tension, it is also impor-
tant to ensure compression, although this is rarely achieved in practical cases, leading
to the need for steel bar reinforcement.
Let us introduce the eccentricity of the bending moments as e y = −Mz /N , ez =
M y /N . Let us now substitute the axial force (applied at the section’s center of gravity)
and the (possibly) two bending moments by the statically equivalent axial force N
applied at a position (e y , ez ) in the section-centered coordinate system.
We can then put the expression of the neutral axis in terms of the eccentricities,
so that:
1 ez Iz + e y I yz e y I y − ez I yz
− z− y = 0,
A I y Iz − I yz
2 I y Iz − I yz
2
where no dependence on N is explicit. Equivalently,

ey ez
1+ y + 2 z = 0,
i z2 iy
where i α is the radius of gyration of the section along the α-axis, defined as:

iα = Iα /A, α = y, z.
The problem thus reduces to finding the pairs (e y , ez ) that produce pure tension
or compression within the section.
To further explain this concept, let us determine the core of a rectangular section,
see Fig. 6.9.
To obtain the location of the core, it is enough to determine the eccentricity of
a load that produces the neutral axis to fall along each side of the rectangle. In this
case: ⎫
A = bh ⎪
⎬ Iy 1 3
12 b h b2
I y = 12 b h ⇒ i y = A = 1 bh 3 = 122
2
1 3
⎪ bh
bh 3 ⎭ i z2 = IAz = 12bh = 12
h
Iz = 12
1
Should the neutral axis be located along the bottom side of the rectangle, we have:

ey h ez ez = 0
1+ − + b2 z = 0 ∀y ⇒
h2 2 e y = h6
12 12
Equivalently, for the neutral axis located along the right vertical edge of the
rectangle:
Fig. 6.9 Determine the core y

of a rectangular section
h
b
Fig. 6.10 Core of a b/3

rectangular section
h/3
h
z

ey ez b ez = b6
1+ y+ − = 0 ∀z ⇒
h2 b2 2 ey = 0
12 12
By applying symmetry, we obtain the shape of the core, depicted in Fig. 6.10.
6.3 Principle of Virtual Work for Bending Beams
In Eq. (4.2) we established the principle of virtual work for trusses. Now, we are in
the position of establishing a much more general expression that includes bending.
Even if the model is not yet complete, i.e., we have not yet defined the relationship
between shear stresses and their corresponding internal forces, these are well-known
to have little influence on the final displacement of the structure, if important bending
moments are present.
The most general expression for this principle is:
L
ψ
σi j εi j d V = dx (σx εψx + τx y γxψy + τx z γxψz )d A, (6.9)
0
V
where the index ψ indicates a virtual field.

From the reasoning above, it is clear that, under bending phenomena, the only
non-vanishing term here is:

σx εψx d A = σx (εψn + κψy z − κψz y)d A =
A
6.3 Principle of Virtual Work for Bending Beams 141

= εψn σx d A + κψy zσx d A − κψz yσx d A =
A A A
= εψn N + κψy M y + κψz Mz .
Of course, this term, equivalent to the virtual work produced by internal forces, must
equal the work done by external forces, i.e.:

ψ ψ
ρbi u i d V + ti u i d A
V δV
⎡ ⎤ ⎡ ⎤
L L
⎣ ψ ψ ψ
= ρbi u i d A⎦ d x + ⎣ ti u i dl ⎦ d x + ti u i d A,
0 A 0 δA A f −Ai
where A f and Ai represent, respectively, the final and initial cross sections of the
bar, where possibly some punctual forces or moments could be applied. b represent
the body forces per unit volume and t the applied tractions on the boundary of the
beam.
Worked Examples
6.1. We consider bars whose cross section is depicted in Fig. 6.11, for the structure
in Example 5.2. Determine
(a) the critical section (i.e., the one that will fail the first) for the loads in Fig. 5.29,
(b) the normal stress distribution for that section,
(c) the neutral axis of that section.
Fig. 6.11 Bar cross section y

2 cm
30 cm
z
2 cm
20 cm
Solution 6.1. (a) Section D of bars B − D and D − E is the critical section of the
structure, as one can readily notice from the observation of the diagrams in the
previous chapter. At that section, we have:
Mz = −1000 kp · m,
Vy = 1000 kp (in bar B − D),
and
Vy = −1000 kp (in bar D − E).
(b) Since the axial internal force vanishes, normal stresses take the form
Mz
σx = − · y.
Iz
From the geometrical dimensions of the sections, we easily determine
1
Iz = 30 · 303 − 16 · 263 = 21565, 333 cm4 .
12
Thus,
100000 kp · cm
σ= · y = 4, 63 · y kp/cm2 .
21565, 33 cm4
(Fig. 6.12).
(c) The neutral axis is the locus of the points satisfying σx = 0, so that we have in
this case,
σx = 0 ⇔ y = 0 cm.
In other words, every point with y = 0 belongs to the neutral axis.
y
69.55 kp/cm2
30 cm
σx (Mz )
Fig. 6.12 Normal stresses at the critical section of the structure, point D
6.2. Determine the distribution of normal stresses for the section in Fig. 6.13, by
assuming that the following internal forces are acting on it:
N = 100 kp, M y = 10 kp · m Mz = 500 kp · m.
Fig. 6.13 Beam cross y

section for Example 6.2
9 mm
hG
z
80 mm
6 mm
80 mm
Solution 6.2. As a first step, we determine the first and second area moments of the
section, A, I y y Iz , to be able to compute stresses properly.
• Area. It is easy to see that the area can be computed as a sum of rectangular areas:
A = 80 · 9 + 71 · 6 = 1146 mm2 = 11.46 cm2 .
• Center of mass. Prior to determining the value of the second area moments, it is
necessary to find the position of the axes or, in other words, the center of gravity of
the section. For the z coordinate, its position is straightforward, given the symmetry
of the section.
For the y coordinate, remember that the center of gravity is such that it vanishes
the first area moments of the section,
i
y · Ai
yG = G ,
Ai
where i is an index over the various rectangles composing the section. In this case,
there are only two rectangles. Taking moments from the upper edge of the section,
we have
80 · 9 · 4.5 + 71 · 6 · 44.5
yG = = 19.37 mm = 1.937 cm = h G ,
1146
which is the distance from the center of mass to the upper edge of the section.
• Second area moments. The second area moments of a section composed of the
addition of several simpler geometries is just the sum of the respective moments
with respect to the same axis. However, sometimes it is necessary to apply the
Steiner theorem so as to obtain this second area moment from the global axis of
the section and not the one of the simpler geometry (rectangle) itself:
1 1
Iy = 9 · 803 + 71 · 63 = 385278 mm4 = 38.52 cm4 ,
12 12
1 1
Iz = 80 · 93 + 80 · 9 · (19.37 − 4.5)2 + 6 · 713 + 6 · 71 · (44.5 − 19.37)2
12 12
= 4860 + 159204.16 + 178955.5 + 269026.19 = 612045.86 mm4
= 61.20 cm4 .
• Given the symmetry of the cross section, the cross moment vanishes, I yz = 0.
y y
8.726 −1582.4
z x
80 mm
80 mm 8.726 4953.06 [kp/cm2 ]

σx (N ) σx (Mz )
−103.82
σx (My )
103.82 [kp/cm2 ]
x
Fig. 6.14 Stress distribution in the cross section of Fig. 6.13
With these values, we are in the position of computing the stress distribution:
N Mz My
σx = − ·y+ · z,
A Iz Iy
100 50000 1000

σx = − y+ z = 8.72 − 816.93 · y + 25.95 · z kp/cm2
11.46 61.20 38.53
(Fig. 6.14).
The neutral axis is the locus of the points with vanishing normal stresses, σx = 0.
In this case:
100 50000 1000
σx = − y+ z = 8.726 − 816.93 · y + 25.95 · z = 0
11.46 61.20 38.52
Therefore, these point satisfy the relationship
z = 31.47 · y − 0.33.
To represent a straight line, it is enough to take two values (Fig. 6.15),
If y = 0 cm ⇒ z = −0.33 cm.
If z = 4.1 cm ⇒ y = 0.13 cm.
Fig. 6.15 Neutral axis y
Compression
z L.N.
Traction
Chapter 7
Beams (III). Shear Stresses
Abstract This chapter introduces the determination of shear stresses produced by

bending.
7.1 Introduction
It has already been mentioned that the hypotheses giving rise to the Euler–Bernoulli–
Navier beam model have some inconsistencies. Particularly, they predict no shear
stresses, something that is well known to be false.
To better understand this apparent lack of coherence and to motivate the proper
way to solve it, let us consider a beam under non-uniform bending moments, see
Fig. 7.1. Let us isolate a portion of it. Of course, every possible portion should be in
equilibrium, as is well-known from the first chapter of this book.
The 1 − 2 portion of the beam is obviously in equilibrium as can be noticed by
glancing at to the stress distribution. However, apply this rationale again and imagine
isolating a portion 3 − 4, see Fig. 7.2.
In this case, it is evident that there is an out-of-balance stress distribution that
would eventually result in a force uniformly accelerating the 3 − 4 portion to the
right. There is no possible equilibrium unless we postulate some kind of shear stresses
(τxy ) as the ones depicted in Fig. 7.3. In this chapter, we will study how to solve these
inconsistencies in an elegant (and rigorous!) way. This is usually achieved by means
of the Collignon theory of shear stresses.

148 7 Beams (III). Shear Stresses
1 2 σx (M (1 )) σx (M (2 ))
2
1 x

Fig. 7.1 A beam subjected to non-uniform bending moments. We isolate the central portion
obtained by slicing along planes 1 − 2 and verify equilibrium in it
3 4
x
Fig. 7.2 The 1 − 2 portion of the beam is isolated again, and a new portion obtained after slicing
by the 3 − 4 plane is put under scrutiny
Fig. 7.3 The beam portion

delimited by planes
1 − 2 − 3 − 4 needs some σx (1 ) σx (2 )
form of shear stress τxy to τxy (postulated)
attain equilibrium
7.2 Collignon Theory of Shear Stresses in Beams
Edouard Collignon1 , along with D. J. Jourawski2 , established the most widely

accepted theory of shear stresses for bending beams.
We have just seen that the Euler–Bernoulli–Navier hypothesis, that accepted that
the shear strains εxy = εxz = 0 vanish, gives rise to important inconsistencies. In
particular, it predicts also vanishing shear stresses τxy = τxz = 0. Jourawski observed
that sometimes timber beams broke longitudinally by developing longitudinal cracks
precisely around the center of gravity, where bending (normal, σx ) stresses vanish.
Therefore, this must be due to a different phenomenon, other than the well-known
bending stresses, already known one century ago at that moment.
1 Édouard Charles Romain Collignon (1831–1913) was a French engineer who worked for the
Russian railways. You can consult his 1865 work “Théorie élémentaire des poutres droites:
ponts métalliques, pont américains, combles” in the internet archive https://archive.org/details/
thorielmentaire00collgoog.
2 Jourawski, D. J. “Sur le résistance dun corps prismatique et dúne Piece Composée en Bois ou on
Tôle de Fer à une Force Perpendiculaire à leur Longeur.” Annales des Ponts et Chaussées. Vol. 12.
1856.
7.2 Collignon Theory of Shear Stresses in Beams 149
Fig. 7.4 Collignon- y

Jourawski hypothesis on the
distribution of shear stresses
on a cross section (note that
we are in the yz plane). Note
that, as pointed our in
Fig. 7.3 and also in Fig. 1.7,
shear stresses also act at a
horizontal plane parallel to
xz to attain equilibrium at the τxy
corner. V represents the
shear internal force acting on
z
the cross section, which
ultimately produces the
stresses τxy V
7.2.1 Hypothesis
The Euler–Bernoulli–Navier model hypothesis is thus valid from a kinematic point

of view, but not from a static (equilibrium) one. To address this inconsistency, the
Collignon model introduces a new static hypothesis:
1. Shear stresses along each of the beam’s fibers are constant.
See Fig. 7.4 for a clearer discussion on the meaning of this hypothesis. Under this
rationale, we accept that τxy = τxy (x, y) ≈ τxy (y).
7.2.2 Collignon Formula
Let us derive, in the plane bending case (one bending moment, say Mz ), the expression
for these shear stresses. For a more general case in which both Mz and My act in the
cross section, a simple superposition of their effects will be enough under the linearity
assumptions established so far.
Consider again a differential slice of a beam of length dx, as depicted in Fig. 7.5.
As a consequence of the applied load py , the bending moment evolves from Mz at
the left section to Mz + dMz at the right one. Equivalently, the shear internal force
changes from Vy to Vy + d Vy . Simple equilibrium considerations allow us to obtain
the following very useful equivalences:

Fy = 0 → Vy + d Vy − Vy + py dx = 0,
Fig. 7.5 Equilibrium of a y

differential portion dx of a
beam under transverse
loading py per unit length py
(positive along the positive
Vy + dVy
direction of the y-axis)
Mz Mz + dMz
Vy
dx
in other words,
d Vy
= −py ,
dx
and, similarly,
dx
M(x=0) = 0 → Mz + dMz − Mz + py dx + (Vy + d Vy )dx = 0.
2
By neglecting higher-order derivatives, we obtain,
dMz
= −Vy .
dx
It is important to clearly understand what these formulas mean. The first one,
d Vy
= −py ,
dx
states that the presence of a vertical load causes the shear internal force to
vary along the beam (i.e., along x). Otherwise it would stay constant along the
beam. The second one,
dMz
= −Vy ,
dx
states that the shear internal force is responsible of the evolution of the bending
moment. If no shear internal force appears in the beam, the bending moment
remains constant.
Let us consider now the equilibrium of a portion of the beam above the center of
gravity, see Fig. 7.6.
If we verify equilibrium in the blue-dashed region, we have, see Fig. 7.7,

F= σx dA.
A
Since normal stresses are due to bending moments, we have
Mz
σx = − y,
Iz
y y
py
A
x z
dx b(y)
Fig. 7.6 Equilibrium of a differential portion dx of a circular cross-section beam under transverse
loading py per unit length (positive along the positive direction of the y-axis). In this case, only a
portion above the center of gravity, at a generic position y, is equilibrated. The cross-sectional area
is A, while the area of the dashed portion is A
py
V V + dV
F F + dF
R = τxy bdx
dx x
Fig. 7.7 Equivalent forces due to shear stresses. Note the presence of a resultant force R = τxy bdx.
Do not confuse V with Vy , the internal shear force in the section. V is the resultant of the shear
stresses on A , see Fig. 7.6. Conversely, F is not the axial internal force, but the result of integrating
σx on A
so that
Mz Mz
F =− ydA = − Sz ,
A Iz Iz
with Sz the first area moment of the section,

Sz = ydA.
A
Very simple reasoning mechanisms will lead us to
dF Sz dMz Sz
=− = Vy = τxy b(y),
dx Iz dx Iz
so that we arrive at the so-called Collignon (or Collignon–Jourawski) formula:
1 Sz
τxy = Vy .
b Iz
If, as a practical example of this theory, we calculate the shear stress distribution
on a rectangular section, like the one in Fig. 7.4, we have,
h/2 h/2
y2 b h2
Sz = ydA = b ydy = b = −y ,
2
A y 2 y 2 4
which gives rise to

1 Sz 1 b h2 12 6Vy h2
τxy = Vy = − y2 Vy = − y 2
.
b Iz b2 4 bh3 bh3 4
Notably, this is a quadratic expression on y that attains its maximum at y = 0, i.e.,

at the height of the center of gravity of the section, see Fig. 7.8.
The just-introduced Collignon formula extends straightforwardly to a general case
of shear internal force along an arbitrary direction, see Fig. 7.9.
In this case, the most general form of the Collignon formulas reads:
Fig. 7.8 Distribution of y

shear stresses on a
rectangular section (side max 3 Vy
τxy = 2 bh
view, note the axis of the
beam) x
Fig. 7.9 Extension of the η y

Collignon formula to a
general case. Arbitrary
section and load acting along
an arbitrary direction η. τxη
Shear stress along the
dashed, blue line
z

S (y)I −S (y)I S (y)I −S (y)I
τxy (y) = 1
Vz y Iy zIz −Iz2 yz + Vy z IyyIz −Iy2 yz ,
b(y)

yz yz
Sz (z)Iy −Sy (z)Iyz Sy (z)Iz −Sz (z)Iyz

τxz (z) = b(z) Vy Iy Iz −I 2
1
+ Vz Iy Iz −I 2 .
yz yz
For a doubly-symmetric cross section, in which Iyz = 0, we have:

1 Sy (y) Sz (y) 1 Sz (z) Sy (z)
τxy (y) = Vz + Vy , τxz (z) = Vy + Vz .
b(y) Iy Iz b(z) Iz Iy
For a general case, in which both normal and shear stresses exist, recall the
definition of the equivalent, von Mises stress: Eq. (1.10),

σVM = σx2 + 3τxy
2.
Nevertheless, it is important to note that, for long beams, normal stresses σx

are usually much more important than shear stresses τxy and will, in general,
determine the true resistance of the bar.
7.3 Shear-Stress Distribution in Thin-Walled Profiles
If the cross section of the beam is thin-walled, see Fig. 7.10, it is commonly accepted
that we can assume the arc-length parameter, s, parameterizes the shear stress.
Fig. 7.10 Shear stresses in y

open, thin-walled cross
sections. The thickness of
the wall, e = e(s) may vary, e(s)
in general, along the section ds
z
Fig. 7.11 Calculation of y

shear stresses in open,
thin-walled cross sections
s s0
In this case, we assume that, given the small thickness value, s, shear stresses are
approximately constant at each s-coordinate, so that:
−1 Sy (s, s0 )Iz − Sz (s, s0 )Iyz

τxs (s) = Vz
e(s) Iy Iz − Iyz2
Sz (s, s0 )Iy − Sy (s, s0 )Iyz
+ Vy ,
Iy Iz − Iyz2
where Sy/z (s, s0 ) represent the first area moment with respect to y or z, respectively,
between s0 and s. s0 is usually taken as a free end of the cross section, so that
Sy/z (s0 ) = 0, see Fig. 7.11.
In this formula, it is worth noting that the minus sign comes from the fact that s0
takes its minimum value where y is maximum and vice versa. For doubly-symmetric
cross section profiles, this formula reduces to the more familiar expression:

−1 Sy (s, s0 ) Sz (s, s0 )
τxs (s) = Vz + Vy .
e(s) Iy Iz
7.3.1 Shear Flow
Given the results above, it is common to define a new magnitude coined as shear
flow as
q = τxs e(s),
such that:
Sy (s, s0 )Iz − Sz (s, s0 )Iyz Sz (s, s0 )Iy − Sy (s, s0 )Iyz

q = − Vz + Vy .
Iy Iz − Iyz
2 Iy Iz − Iyz2
7.3 Shear-Stress Distribution in Thin-Walled Profiles 155

shear flows in open,
thin-walled cross sections
s0
s1
z s2
The value of the shear flow at any point can easily be determined for its value at
another point, see Fig. 7.12, given that:
s2 s1 s2
Sy2 = zeds = zeds + zeds = Sy1 + Sy1→2 ,
s0 s0 s1
such that:
Sy (s2 , s1 )Iz − Sz (s2 , s1 )Iyz Sz (s2 , s1 )Iy − Sy (s2 , s1 )Iyz

q2 = q1 − Vz + Vy ,
Iy Iz − Iyz
2 Iy Iz − Iyz2
where the simplified notation
Sy1→2 = Sy (s2 , s1 )
has been employed, for the sake of clarity.
In the case of hollow profiles, it is not easy in principle to determine a point

for which q = 0 to start the computation of shear flows. In this case, points
lying at symmetry planes should be employed since they also obey this q = 0
condition, because τxs = 0 at these symmetry planes, see Fig. 7.13.
7.3.2 Shear Center
The shear center is the point of the cross section at which an internal shear force could
be applied without causing torsional deformation—twisting—of the cross section.
This concept is particularly important in C-shaped profiles which are, for instance,
widely employed in lightweight construction. The shear center does not coincide, in
general, with the center of gravity of the cross section, see Fig. 7.14.
The application of a shear force out of the shear center is equivalent to a shear
force at the shear center plus a twisting moment

shear flows in hollow,
thin-walled cross sections.
Given the symmetry with
respect to the applied shear V q=0
force, the highlighted points
present a vanishing shear
flow z
Fig. 7.14 Shear center. The y

application of a shear force
V at the center of gravity is
equivalent to its application
at the shear center plus a Shear center
twisting moment Mx = V · d
z
d
Center of gravity
Fig. 7.15 Determination of y

the position of the shear
center
Shear center
Mx = V · d .
In some cases, the determination of the position of the shear center is straightfor-
ward. For instance, for cross sections formed by rectangular segments intersection
at a given point, this is precisely the shear center, see Fig. 7.15. For a cross section
possessing symmetry axes, the shear center must lie within these axis. Therefore, is
the profile is doubly-symmetric, the position of the shear center coincides with the
origin of coordinates or center of gravity, see Fig. 7.16.
7.4 Deep Beams. Timoshenko Beam Theory 157
Fig. 7.16 Determination of y

the position of the shear
center for doubly-symmetric
profiles
Shear center
7.4 Deep Beams. Timoshenko Beam Theory
Deep beams—those with a span-to-depth ratio lower than, say, three— are very often
employed to construct walls, such as in precast concrete structures, for instance, see
Fig. 7.17. In this case, the hypothesis of “long” beams, established for the Euler–
Bernoulli–Navier theory, is not satisfied. Of particular importance is the fact that flat
cross sections do not stay perpendicular to the beam’s line of centroids, nor stay flat.
To understand this assertion, try this simple exercise: take a thick book and hold it
tight, so that the friction between pages allows us to consider it as “continuous”,
Fig. 7.17 An example of a deep beam or wall in precast concrete

Fig. 7.18 An intuitive example of shear deformation of the cross section. The dashed line represents
the line of centroids, x, while the thick solid line represents the cross section of the “beam”. The
effect of the shear force on the section makes the section deviate from its initial perpendicular
position
rather than a mere juxtaposition of paper sheets. Then, bend it as in Fig. 7.18. You
will readily notice how bending provokes the cross section to abandon its initial
perpendicular position to the (dashed) line of “centroids”.
7.4.1 Lower Bound Theorem Analysis
The Euler–Bernoulli–Navier theory being no longer valid, employment of the lower

limit theorem (see Chap. 3) remains an appealing alternative, as proposed in [16],
for instance. In particular, the analysis made in Fig. 5.1 for long beams is known not
to be a valid interpretation of what is going on, so that an alternative explanation is
necessary. In Fig. 7.19, a bowstring-like scheme for the thrusts taking place in deep
beams is proposed that is well-known to be a better fit with experimental results.
Employment of the lower bound theorem is particularly useful for the correct
placement of holes such as windows in the walls, see Fig. 7.20. The analysis must
ensure that the proposed mechanism is able to carry not only the dead (self-weight)
load but also variable loads such as wind.
Fig. 7.19 Deep beam

analysis by the lower bound q
theorem. A bowstring-like
thrust line is proposed to
understand the transmission
of the vertical load to the
supports
Fig. 7.20 Opening holes in

deep beams q
These structures are particularly efficient when the walls span several stories. In
Fig. 7.21, an example is depicted of a deep beam with several openings in which their
positioning has been optimized by means of the lower bound theorem.
7.4.2 Timoshenko Theory
Stepan Timoshenko3 established the theory that carries his name by adding the cross
section the possibility to rotate around the line of centroids due to shear stresses
(Fig. 7.22).
Therefore, the main contribution of Timoshenko’s work is the addition of a rotation
degree of freedom θz , independent of uy (respectively, a rotation θy , independent of
uz ). We enumerate the hypothesis necessary to define Timoshenko’s theory:
1. A first kinematic hypothesis: every cross section moves as if it were a rigid solid.
3 Stepan Prokopovych Tymoshenko (1878–1972) was a Ukrainian engineer who moved to the U.S.
He joined the University of Michigan in 1927 and Stanford in 1936. His theories are, therefore,
relatively recent. His books on the theory of elasticity have been translated into some 36 languages.
Slabs
Fig. 7.21 Opening holes in deep beams spanning several stories
Fig. 7.22 Hypothesis of the

Timoshenko beam model. duy
Cross sections remain flat dx
but not perpendicular to the y
θz
line of centroids. An
initial—undeformed—and
deformed cross section are
highlighted with solid thick
lines
2. A second kinematic hypothesis: cross sections initially perpendicular to the

beam’s axis (such as A(t) in Fig. 6.1) remain flat but not perpendicular after
bending.
3. A static hypothesis: normal stresses σy and σz are negligible.
Hypothesis number 2 can be stated mathematically in terms similar to Eq. (6.1)

as:
ux (x, y, z) = ux0 (x) − θz y + θy z,

uy (x, y, z) = uy0 (x),
uz (x, y, z) = uz0 (x),
with uy0 , uz0 the displacements at the cross section’s center of gravity.
As a consequence of the above hypothesis, the strain component γxy does no
longer vanish, i.e.,
duy0
γxy = − θz (x) = βy (x).
dx
This has a direct consequence, in turn, over the value of the shear stress, that would
be
τxy = Gβy (x).
Again, we found an inconsistency between kinematic and static assumptions: this is

in sharp contrast to the conclusions obtained after the development of Collignon’s
formula. However, Timoshenko’s hypothesis number 2 implies the need for a new
degree of freedom, namely the rotation(s), θy , θz . This implies the need for a new
constitutive equation, other than those employed in Euler–Bernoulli–Navier theory.
This new constitutive equation will be obtained by imposing the equivalence between
the strain energy due to Collignon stresses and the work done by the internal shear
force (for simplicity, we restrict ourselves to the xy terms, similar expressions arise
for the xz ones):
τxy γxy dA dx = Vy βy dx,
A
so that
τxy
2
dA = Vy βy .
A G
By substituting Collignon formula,

Vy2 A Sz2
dA = Vy βy ,
GIz2 A b2
giving rise finally to

Vy
βy = ,
GAcy
with
Acy = κy A.
κy = 1
fcy
is known as the Timoshenko shear coefficient, and

A Sz2
fcy = dA.
Iz2 A b2
We thus arrive at the following constitutive equations for the Timoshenko theory:
Vy
βy = ,
GAcy
and
d θz Mz
= .
dx EIz
These constitutive equations give rise to the following model for the Timoshenko
beam:
d duy0
GAcy − θz + py = 0, (7.1)
dx dx
where py is the load per unit length pointing along the y direction, and
d d θz
EIz + Vy + mz = 0,
dx dx
or, equivalently,

d d θz duy0
EIz + GAcy − θz + mz = 0, (7.2)
dx dx dx
where mz is the bending moment per unit length, a type of loading found very rarely
(think, for instance, of the possibility of heating a beam on one side only, so that
it suffers dilation along the heated side, while the others remains at the reference
temperature).
Remember that analogue equations exist for the xz plane.
Worked Examples
7.1. Determine, for the critical cross section of Example 5.1, the distribution of shear
stresses, by following the just-developed theory of thin-walled profiles. We consider
a shear force Vy = −100 kp acting on the section, which is represented again in
Fig. 7.23.
Fig. 7.23 Critical section of y

the structure 2 cm
30 cm
z
2 cm
20 cm
Solution 7.1. The distribution of shear stresses is given, as just developed, by:
q
τxs = ,
e(s)
where
−Vy
q= Sz (s).
Iz
We begin by computing the first area moment of the section between the points
A − B in Fig. 7.24,
s s s
SzA→B = Sz (A; B) = ydA = e · yds = 2 · 14ds = 28 · s,
0 0 0
and, from it, the shear flow,
Vy A→B 100
qA→B = qA − · Sz =0+ · 28 · s = 0.13 · s kp/cm,
Iz 21565.333
which corresponds to a linear distribution along horizontal segments of the section.

Particularizing at the beginning and the end of the segment,
qA = 0 kp/cm ⇒ τxz = 0 kp/cm2 ,
qB (s = 10cm) = 1.3 kp/cm ⇒ τxz = 0.65 kp/cm2 .

B A
C
z
Vy = −100 kp
Fig. 7.24 Shear flow corresponding to a shear force Vy
Remembering that where segments meet, there is always a balance of flows,

joint
joint
q in = q out ,
we proceed in an analogous way for the vertical segment of the cross section, between
B − C:
s s s
SzB→C = Sz (B; C) = ydA = e · yds = 2 · (13 − s)ds = 26 · s − s2 .
0 0 0
Therefore,
Vy B→C 100
qB→C = qB − · Sz = 1, 3 + · (26 · s − s2 )
Iz 21565.333
= 1.3 + 0.12056 · s − 0.004637 · s2 kp/cm,
which corresponds to a quadratic distribution along the vertical segment. Particular-

izing again at the beginning and the end of the segment, we have,
qB (s = 0) = 1.3 kp/cm ⇒ τxy = 0.65 kp/cm2 ,
qC (s = 13 cm) = 2.08 kp/cm ⇒ τxy = 1.04 kp/cm2 .
These values are represented on the section in Fig. 7.25.

0.65 y
0.65 0.65
B A
0.65
C
1.042 z 1.042
0.65
0.65 0.65
[kp/cm2 ]
0.65
Fig. 7.25 Shear stress distribution
7.2. For the critical section of Example 6.2, compute the shear-stress distribution
predicted by assuming a thin-walled profile. We consider a shear force Vy = 300 kp
acting on the profile of Fig. 7.26.
Solution 7.2. The shear stress distribution is given by
Fig. 7.26 Compute the y

distribution of shear stresses
for this section
9 mm
hG
z
80 mm
6 mm
80 mm
Fig. 7.27 Shear flow y

produced by a shear force Vy
E
A B
z Vy = 300 kp
C
q
τxs = ,
e(s)
where
−Vy
q= Sz (s).
Iz
We begin by determining the shear-flow distribution between the points A − B in

Fig. 7.27. The upper part of the section is located at 19, 37 mm from the center of
gravity, such that:
s s
SzA→B = Sz (A; B) = ydA = e · yds
s 0 0
= 9 · (19.37 − 4, 5)ds = 133.83 · s mm3 = 1.34 · s cm3 .

0
Therefore:
Vy A→B 300
qA→B = qA − · Sz =0− · 1.34 · s = −6.56 · s kp/cm.
Iz 61.21
This value corresponds to a linear distribution (notice the dependence on s) along

horizontal segments of the section. Particularizing at the beginning and the end of
these segments,
qA = 0 kp/cm ⇒ τxz = 0 kp/cm2 ,
qB (s = 4cm) = −26.24 kp/cm ⇒ τxz = −29.15 kp/cm2 ,
Remember that the balance of shear flows at the joints of rectangular segments of
the section, joint joint
q in = q out ,
B
such that, qBD = qAB
B
+ qEB
B
= 2 · (−26.24) = −52.48 kp/cm.
Proceeding in an analogous way between B − D:
s s s
SzB→D = Sz (B; D) = ydA = e · yds = 6 · (10.37 − s)ds
0 0 0
= 62.22 · s − 3 · s2 mm3 = 0.62 · s − 0.3 · s2 cm3 .
Thus:
Vy B→C
qB→D = qB − · Sz
Iz
300
= −52.48 − · (0.62 · s − 0.3s2 ) kp/cm.
61.20
In this case, the resulting distribution is quadratic. Particularizing at the beginning
and the end of the segment, we have, given that e = 6 mm = 0.6 cm,
qB (s = 0) = −52.48 kp/cm ⇒ τxy = −87.47 kp/cm2 .
Point C represents the center of gravity of the section. Its s-coordinate is s =

19.37 − 9 = 10.37 mm, such that:
qC (s = 1.03 cm) = −54.06 kp/cm ⇒ τxy = −90.10 kp/cm2 ,
qD (s = 7.1 cm) = 6.57 · 10−03 ∼

= 0 kp/cm ⇒ τxy = 0 kp/cm2 .
The exercise ends by plotting the resulting distribution on the section, see
Fig. 7.28.

Fig. 7.28 Distribution of y

shear stresses under Vy
29.15
A
E
87.46 B 29.15
z
C
90.10
[kp/cm2 ]
0 D
Chapter 8
Beams (IV). Torsion
Abstract This chapter introduces the determination of shear stresses produced by

torsion.
8.1 Introduction
Torsion occurs rarely in buildings. However, there are cases in which it may occur, and
therefore those affected elements must be carefully designed to bear these internal
moments. Torsion occurs, for instance, in spandrel beams that support beams at points
other than the pillars, see Fig. 8.1 or in floor beams firmly welded to girders, in the
case of steel structures, for instance.
A careful analysis of the internal forces and moments in this structure reveals that
bending moments in the b − e beam give rise to a different type of internal moment
in a − b − c and d − e − f , see Fig. 8.2.
In this chapter we will introduce the most common models for torsion, namely,
the Saint Venant and Prandtl models.
8.2 Saint Venant Model
In the development of the Saint Venant model, we will assume the following kine-
matic hypothesis:
1. Each cross section rotates as a rigid solid (see Fig. 8.3).
2. The rate of twist, i.e., the angle rotated by the bar per unit length, κx , is constant.
3. Cross sections are free to warp in the x direction (along the longitudinal axis of
the bar), but this warping is the same for all cross sections, see Fig. 8.4. In other
words: u x = u x (y, z).

170 8 Beams (IV). Torsion
c
f
Fig. 8.1 An example of beams suffering torsion. Beams a − b − c and c − d − e are subjected to
torsion due to the bending moment at b − e
c
f
Fig. 8.2 Internal moments in the structure of Fig. 8.1. Red lines in the diagrams represent torsion
moments, while blue lines represent bending moments. Pinned supports are here assumed to allow
bending but to resist torsion (no free rotation around the local x-axis of each bar)
Fig. 8.3 Torsion of a

prismatic bar. The Saint
Venant model assumes that
each cross section rotates as
a rigid solid. Caution: in
addition to the movement
represented herein, a
warping movement must be
considered, see hypothesis
#3 of the model
8.2 Saint Venant Model 171
ux
ux ux
Fig. 8.4 Warping of a cross section. This warping movement causes the cross section to loose its
initial flatness. According to Saint Venant, this movement must be superimposed on the rigid-solid
one in Fig. 8.3 to satisfy the kinematic assumptions of his model
The rigid solid-like movement of hypothesis #1 occurs around a point of the

section known as center of torsion. The exact location of this point is a priori
unknown, unless we consider doubly-symmetric cross sections. In that case, the
center of torsion will coincide with the center of mass.
These hypotheses have limited ranges of validity. First, they are valid for straight
bars (or at least those with a large radius of curvature). Otherwise, bending and torsion
will be coupled, influencing each other and giving rise to a much more complex
situation. This theory is also valid for bars with constant cross section. Again, this
hypothesis can be relaxed if the variation of the cross section is smooth along the local
x axis. Third, the model considers constant moments. If the moment varies along
the bar, a non-constant warping will arise, thus producing a more complex stress
state along x. Finally, for this theory to be valid, supports (boundary conditions)
must be such that no restriction to warping is introduced. Do not apply this theory
in regions close to supports where reasonable doubts about their behavior exist. In
those regions, rather use a full three-dimensional analysis.
Under the above circumstances, consider a cross section like the one in Fig. 8.5.
The movement of an arbitrary point in the section, (y, z) will be:
u x = u x (y, z),
u y = −θx (x)(z − z T ),
u z = θx (x)(y − yT ).
Once we have postulated the form of the displacement field, by resorting to the
definition of Cauchy’s strain tensor, see Eq. (1.3), we have:
εx = 0,
ε y = 0,
εz = 0,
∂u x dθx
γx y = − z,
∂y dx
(y, z)
u(y, z)
Mx
Center of torsion (yT , zT )
z
Fig. 8.5 Rotation of an arbitrary point around the center of torsion, according to hypothesis #1
∂u x dθx
γx z = + y,
∂z dx
γx z = −θx + θx = 0.
Note that, since warping is assumed to be identical for every section, no normal
stress will be produced. This has been also validated experimentally for regions far
from supports imposing restrictions to warping, as mentioned before. By resorting
to constitutive equations, Eq. (1.7), we have:
σx = 0,
σ y = 0,
σz = 0,
τ yz = 0,

∂u x dθx
τx y = G − z ,
∂y dx

∂u x dθx
τx z =G + y .
∂z dx
Where the aforementioned hypothesis could not be verified, the so-called theory of
non-uniform torsion must be applied. This theory falls beyond the scope of this book
and will not be detailed herein. In practice, a detailed three-dimensional analysis
should be preferred, probably with the help of finite element software, for instance.
If we substitute the just-obtained stresses into the equilibrium equations (1.6), we
arrive at:
∂σx ∂τ yx ∂τx z
+ + = 0, → ∇ 2 u x = 0,
∂x ∂y ∂z
2
∂τx y ∂σ y ∂τ yz ∂ θx ∂ 2 θx
+ + = 0, → −G z =0 → (x) = 0, ∀z,
∂x ∂y ∂z ∂x 2 ∂x 2
2
∂τx z ∂τ yz ∂σz ∂ θx ∂ 2 θx
+ + = 0, → −G y =0 → (x) = 0, ∀y.
∂x ∂y ∂z ∂x 2 ∂x 2
These last two expression prove that the rotation of the cross section is a linear
function of x, the longitudinal coordinate of the bar. In turn, the first one imposes u x
to be an harmonic function of x. Therefore, a suitable hypothesis for the form of θx
could be
θx = κx x + B,
such that
dθx
κx = = constant.
dx
In these circumstances, it is a common strategy to assume that the warping takes
the form
u x = θx · φ(y, z).
Obviously, for the harmonicity condition ∇ 2 u x = 0 established above, it immediately

follows from the need of the warping function also to be harmonic:
∇ 2 φ(y, z) = 0.
This is equivalent to assume that the stress tensor takes the form:
⎛ ⎞
∂φ
0 Gθx ∂y
− z Gθx ∂φ
∂z
+y
⎜ ⎟
⎜ ⎟
σ = ⎜ Gθx ∂φ − z 0 0 ⎟ (8.1)
⎝ ∂y ⎠
Gθx ∂φ
∂z
+y 0 0
To determine the precise expression for the warping function, it is mandatory

to impose boundary conditions. For instance, due to the absence of any externally
applied load on the lateral surface of the bar, see Fig. 8.6, vanishing stresses must
be enforced. In the same spirit, at the right- and left-most sections, we must enforce
vanishing shear internal forces and a resulting internal twist moment equal to the one
externally applied.
Lateral surface, n = (nx , ny , 0)T
End surface, n = (1, 0, 0)T

x
End surface, n = (−1, 0, 0)T
Fig. 8.6 Determining the form of the warping function φ(y, z)
In other words, along the lateral surface of the bar we must enforce:

∂φ ∂φ
t = σ · n = Gθx − z ny + + y n z = 0,
∂y ∂z
thus giving
dφ
= zn y − yn z .
dn
We follow a similar procedure for the right and left surfaces in Fig. 8.6 so as to
obtain:

Vy = τx y d A = 0,
A
Vz = τx z d A = 0.
A
These expressions are satisfied identically due to the harmonic character of φ(y, z).
However, a more interesting result is obtained after imposing the internal equilibrium
of stresses with the externally imposed twisting moment:

∂φ ∂φ
MT = τx y z − τx z y d A = Gθx y +y −z −z dy dz,
A A ∂z ∂y
which is equivalent to

∂φ ∂φ
MT = Gθx y −z + y 2 + z 2 dy dz = Gθ J,
∂z ∂y
A
J
where J is the so-called torsion constant. It can be readily noticed that J depends
only on the shape of the cross section. The product G J is known as the torsional
stiffness of the section. This comes from the fact that:
MT
θx = . (8.2)
GJ
The general procedure to solve the Saint Venant torsion model for a given bar
thus comprises the following steps:
1. Obtain the warping function by solving the Laplace problem
∇ 2 φ(y, z) = 0 in A,
dφ
dn
= zn y − yn z on ∂ A.
2. Obtain now the torsion constant

∂φ ∂φ
J= y −z + y + z dy dz.
2 2
A ∂z ∂y
3. Obtain the rate of twist

MT
θ= .
GJ
4. Finally, obtain the displacement field, if needed, and the associated stresses
u x = u x (y, z),
u y = −θx (x)(z − z T ),
u z = θx (x)(y − yT ),
and

∂φ
τx y = Gθ −z ,
∂y

∂φ
τx z = Gθ +y .
∂z
One last ingredient in the Saint Venant theory is to express it in the form of
a boundary value problem (amenable to be solved by finite element methods, for
instance). This can be achieved by introducing the constitutive Eq. (8.2) arising from
Saint Venant theory into the differential equation of equilibrium of twisting moments,
so as to give:
obtain θx such that

d dθx
GJ + m x = 0, in Ω,
dx dx
θx = θx on Γu ,
dθx
MT = G J = MT on Γt ,
dx
where m T is the twisting moment per unit length applied to the bar (a type of loading
which is indeed very rare), Γu is the so-called Dirichlet boundary (the portion of the
boundary where the twisting angle θx is imposed) and, finally, Γt is the Neumann
boundary, i.e., the portion of the boundary where moments MT are imposed.
8.3 Prandtl Model
Ludwig Prandtl (1875–1953) was a German engineer who developed an alternative

theory for the description of uniform torsion phenomena. While Saint Venant the-
ory employs a displacement-based formulation (actually, a rotation-based one), the
Prandtl formalism employs a stress function, called the Prandtl function. Stresses
derive from this function and must also satisfy compatibility and equilibrium, of
course.
To verify static equilibrium, in the absence of body forces, one has to satisfy
∇ · σ = 0,
or, equivalently,
σi j, j = 0,
For the particular case of torsion, a simple choice is enough to satisfy these require-
ments. Indeed, check the validity of the following choice:
σx = σ y = σz = τ yz = 0,
∂ψ ∂ψ
τx y = G J , τx z = G J . (8.3)
∂z ∂y
Obviously, function ψ(y, z) is precisely the Prandtl function.1
1 Do not confuse the greek letter φ, i.e., phi, employed for the Saint Venant warping function, with
ψ, psi, Prandtl’s function.
8.3 Prandtl Model 177
It can be demonstrated, based on the employment of the so-called Beltrami–

Michell formulation of the elasticity problem (see [18], something that exceeds the
scope of this book), that an additional condition for ψ to satisfy equilibrium is
∇ 2 ψ = constant.
By noting that Eqs. (8.1) and (8.3) represent the same magnitude, we have:
∂ψ ∂φ
= − z,
∂z ∂y

∂ψ ∂φ
=− +y ,
∂y ∂z
so that we can thus determine the value of the constant,
∂2ψ ∂2ψ ∂2φ ∂2φ

∇2ψ = + = −1 − − 1 = −2.
∂z 2 ∂ y2 ∂z 2 ∂ y 2
We therefore face a partial differential equation to find ψ:
∇ 2 ψ = −2.
To be able to find a solution, a set of boundary conditions is mandatory. As in the

case of the quest for the warping function, we have
∂ψ ∂ψ
t = σ · n = Gθx ny + n z = 0,
∂z ∂y
or, equivalently
dψ
= 0.
ds
In other words,
ψ = constant
along the outer surface of the bar, ∂ A. s represents here the coordinate along this
surface. A simple yet practical solution consists in taking
ψ = 0 on Γ = ∂ A.
At the end sections of the bar, we impose the equilibrium of the externally applied
moments with the internal distribution of stresses, as we did in the previous section
for the Saint Venant theory when looking for the warping function:

∂ψ ∂ψ
MT = τx y z − τx z y d A = Gθx −y −z dy dz,
A A ∂y ∂z
so that

∂ψ ∂ψ
J =− y +z dy dz.
A ∂y ∂z
Equivalently,

∂ yψ ∂zψ
J =− + − 2ψ dy dz.
A ∂y ∂z
By applying the divergence theorem,

J= (yψn y + zψn Z )d∂ A + 2 ψd A.
∂A A
Given that ψ = 0 on ∂ A, we arrive at the useful expression

J =2 ψd A,
A
such that

MT = Gθx 2 ψd A
A
To obtain the Prandtl function, therefore, we must solve the following boundary-
value problem:
∇ 2 ψ = −2 in A
ψ = 0 on ∂ A.
We then obtain the rotation rate

MT M
θx = = T ,
GJ 2G A ψd A
and then obtain the stress distribution

∂ψ ∂ψ
τx y = Gθx , τx z = −Gθx .
∂z ∂y
Finally, the displacement field can be obtained by obtaining the strain field
from the stress field and then by integration.
Given the form of the Prandtl function stated here, it sometimes interpreted as a
sort of bubble over the section whose volume

ψd A
A
is proportional to the applied twisting moment MT .

This explains why, for instance, the torsion stiffness of an open, thin-walled section
is orders of magnitude lower than that of a closed section, see Fig. 8.7.
8.3.1 Torsion in Open, Thin-Walled Profiles
When a profile is made of thin-walled segments, it is reasonable to assume that the

Prandtl function depends only on one coordinate (say, z), since the other (say, y) is
much bigger, see Fig. 8.8. Thus, we will accept that
ψ(y, z) ≈ ψ(z).
If we thus try to solve the Prandtl problem under this new assumption,
∇ 2 ψ = −2 in A,
ψ = 0 on ∂ A.
Actually, this produces an ordinary differential equation of the form
∇ 2 ψ(z) = −2 in A, → ψ(z) = −z 2 + c1 z + c2 .
Imposing vanishing values along the boundary,

e
z=± → ψ = 0,
2
(a) Closed section
(b) Open section
Fig. 8.7 Effect on the torsional stiffness of a closed (a) or open (b) thin-walled profile. You can
make a simple experiment by just taking the cardboard tube from a paper towel roll. Try to twist it
(without causing any local buckling), and you will see no perceptible rotation angle. On the contrary,
cut it longitudinally with a pair of scissors and notice the drastic decay of stiffness. In b you will
notice the warping of the section
Fig. 8.8 Prandtl function for y

a thin-walled profile.
Essentially, this function is
assumed to depend on one
coordinate only
such that
e2
c1 = 0, c2 = ,
4
thus giving

e2
ψ(z) = − z − 2
.
4
This greatly facilitates the computation of the torsional constant of a thin-walled,

rectangular profile:

e2
J =2 ψd A = −2 z −
2
d A,
A A 4
giving finally
1 3
J= be ,
3
with b the height of the section and e the (necessarily small) profile thickness. This
also opens the possibility of an easy determination of the rotation rate of the profile:
MT 3MT
θx = = .
GJ Gbe3
Fig. 8.9 Stress distribution y

for a thin-walled profile.
Since ψ is assumed to
depend on z only, there is one
tangential stress only (τx y )
τ max = Gθx e
This produces, in turn, a distribution of tangential stresses of the type:
∂ψ ∂ψ
τx y = Gθx = −2Gθx z, τx z = −Gθx = 0.
∂z ∂y
This type of distribution is depicted in Fig. 8.9.

Note that, contrarily to what happens for solid profiles under the Saint Venant
theory, for thin-walled profiles, the maximum level of stress is located at the boundary
of the profile and not in the interior. This reflects a maybe counterintuitive behavior
of this type of profiles that should be kept in mind for adequate reasoning when
dimensioning profiles. Remember, nevertheless, that this stress distribution is valid
only at points far enough from the supports.
For thin-walled profiles built after several rectangular profiles such as those in
Fig. 8.10, the torsion constant can be obtained as

J= Ji ,
i
with i the number of rectangular profiles employed to build the whole cross section,
see Fig. 8.10.
In this case, the angle rotated per unit length will be identical for every segment
of the profile (otherwise no compatibility could occur between them) and therefore,
MT i
θxi = ,
G Ji
y y
z z
Fig. 8.10 Examples of open, thin-walled sections
such that
MT Ji
MT i = .
J
Since
(τx y )i = 2Gθxi z → (τx y )imax = 2Gθxi eimax .
It is therefore evident that the maximum stress in the thin-walled profile occurs,
surprisingly, where the thickness is maximal! A practical consequence of this fact is
that one should design profiles by juxtaposition of rectangular profiles of identical
thickness. Otherwise, yield will occur where the thickness is maximal.
8.3.2 Torsion in Closed, Thin-Walled Profiles
For closed profiles, the stress distribution is known to follow different patterns. If
the thickness e is small, the stress distribution can be approximated to a reasonable
extent as being constant throughout the thickness, see Fig. 8.11.
If the thickness varies along the perimeter of the profile, from equilibrium con-
siderations it is evident that
τ1 e1 d x = τ2 e2 d x,
Fig. 8.11 Stress distribution y

in closed, thin-walled
sections
so that
τ1 e1 = τ2 e2 .
We can therefore define a sort of torsion stress flow, in an analogous way as we did
for shear stresses, such that
qT = τxs · e = constant,
where the coordinate s has been preferred, since in general the profile cross section
will not be oriented along the reference axis y, z. It is now evident that

Mt = qT · r ds.
Given that

1
qT · r ds = A,
2
we arrive at
Mt
qT = .
2A
We are now in the position of determining the point of maximum stress within
the section. To this end, it is enough to recall that
qT = (τxs )i · ei ,
so that
qT Mt
(τxs )i = = .
ei 2 Aei
Therefore, we find that the maximum stress will be at the point of minimum thickness
eimin , something more intuitive that those conclusions obtained for open, thin-walled
profiles in the previous section.
To conclude, the torsional rigidity can now be evaluated as
4 A2
J = ds ,
e
which, for the general case of constant thickness, leads to
4 A2 e
J= ,
Stotal
with Stotal the length of the perimeter of the section, see Fig. 8.12.
Fig. 8.12 Length of the y

section perimeter
Stotal
z
Worked Examples
8.1. Use Saint Venant’s model to study a circular cross section under torsion and to
compute its shear-stress distribution (Fig. 8.13).
Fig. 8.13 Circular cross ny = sin α

section under torsion y
nz = cos α
MT
α
z
Solution 8.1. (a) Remember that circular sections do not present warping (i.e.,
φ(y, z) = 0) since the number of planes of symmetry is infinite. This symmetry
would be broken if warping appears.
In a Saint Venant context the following Laplace problem must to be solved to
obtain the warping function φ(y, z)
∇ 2 φ(y, z) = 0 in A,
dφ
= zn y − yn z on ∂ A.
dn
Therefore, for this example,
dφ
= zn y − yn z = R cos α sin α − R sin α cos α = 0.
dn
Summarizing,
∇ 2 φ(y, z) = 0 in A,
dφ
= 0 on ∂ A,
dn
which implies that φ(y, z) = constant ⇒ u x = θφ = constant. This is equivalent to

a rigid solid movement, φ = 0.
(b) To obtain now the torsion constant following the Saint Venant model, we recall
that

∂φ ∂φ
J= y −z + y + z dy dz,
2 2
A ∂z ∂y
which in this case leads to

π 4
J= r 2 d A = I0 = R .
A(y,z) 2
In other words, for a circular cross section the torsion constant is the polar moment
of inertia, I0 .
(c) The rate of twist has been defined as
MT MT 2
θ= = .
GJ G π R4
This rate of twist depends on the inverse of the section radius to the power of four,
so to reduce the rate of twist (without any change in the material) it is enough to
increase the radius of the cross section.
(d) Finally, to obtain the stress field we recall that

∂φ
τx y = Gθ − z = −Gθz,
∂y

∂φ
τx z = Gθ + y = Gθy.
∂z
Therefore (Fig. 8.14),

τ= τx2y + τx2z = Gθr.
Fig. 8.14 Shear stress τ

distribution for a circular
cross section under torsion
MT
8.2. Use Prandtl model to study a elliptic cross section under torsion and compute
the resulting shear stress distribution (Fig. 8.15).
Fig. 8.15 Ellipsoidal cross y

section under torsion
MT
Solution 8.2. When the function that describes the geometry of the boundary of
the cross section has a constant Laplacian, it is easy to obtain the Prandtl function
by scaling this function by a constant. For this particular case, the expression of an
ellipse satisfies this condition,

y2 z2
ψ= + 2 − 1 c,
a2 b
∇ 2 ψ = −2,
so that

a 2 b2 a 2 b2 y2 z2
c=− 2 ⇒ ψ = − + 2 −1 .
a + b2 a 2 + b2 a2 b
Therefore, the resulting rate of twist will be
MT M
θ= = T .
GJ G2 A(y,z) ψd A
Finally, to obtain the stresses field,

∂ψ a2 ∂φ
τx y = Gθ = −2Gθ 2 z = Gθ − z ,
∂z a + b2 ∂y

∂ψ b2 ∂φ
τx z = −Gθ = 2Gθ 2 y = Gθ +y .
∂y a + b2 ∂z
The warping function and the displacement field can be found by integrating the
shear stress distribution, i.e., the above equations. It is then clear that in this case we
will face a non-vanishing warping (Fig. 8.16).
τ
MT
Fig. 8.16 Shear stress distribution for a ellipsoidal cross section under torsion
8.3. Let us consider a bar, clamped in its initial and end sections A and B, respec-
tively. Let us consider two different sections along the bar. On the left, a solid circular
cross section with radius r1 = 50 mm is assumed. On the right part of the bar, a hollow
cross section with internal radius r2 = 40 mm, see Figs. 8.17 and 8.18, is considered.
Compute the distance d where we must apply a twisting moment MT to obtain the
same reaction forces MT (A) = MT (B) at both clamped supports.
MT
x
A B
625 mm 625 mm
Fig. 8.17 Fixed bar with non uniform cross section under torsion
y y
r1 r1
z z r2
Fig. 8.18 Sections for the fixed bar with non uniform cross section under torsion
Solution 8.3. Considering MT (A) = MT (B) = T , the torsion moment diagram is

displayed in Fig. 8.19.
The polar moments of inertia of both sections are given by
T T
A B
T T
d
Fig. 8.19 Resulting torsion diagram for the structure in Fig. 8.17
π 4
Isolid = 50 = 9817477.04 mm4 ,
2
π
Ihollow = (504 − 404 ) = 5796238.45 mm 4 .
2
The angle twisted between A and B must vanish thanks to the clamped supports, so,
625 d 1250
T T T
φ A→B = 0 = dx + dx + d x.
0 G · Isolid 625 G · Ihollow d G · Ihollow
Therefore,
625 d 625 −1250 d

+ − + + = 0 ⇒ d = 7533 mm.
Isolid Ihollow Ihollow Ihollow Ihollow

Chapter 9
Statically Indeterminate Structures
Abstract This chapter introduces the methods for the analysis of statically indeter-
minate structures.
9.1 Introduction
Throughout this book, we have studied both statically determinate and indeterminate
structures. Particularly in Chap. 4, we treated this concept for trusses. In the most
recent ones, we have introduced very intuitive and powerful methods based on the
lower bound theorem. However, they do not provide us with a detailed solution for
the structures at hand.
At this point, it is time to study in a systematic way the various methods available
for the analysis of statically indeterminate structures. Since this might take a long
time, we have preferred to introduce only the most classical method for manual
analysis: the flexibility method (sometimes referred to as the method of consistent
deformations for reasons that will become apparent soon). Stiffness methods will
only be covered very briefly. For analysis with the help of a computer, stiffness
methods are always preferred, in particular the method of finite elements [8]. As you
will notice hereafter, flexibility methods are intuitive, but less systematic—always
some training will be welcomed by the analyst—while stiffness methods are very
systematic—and thus very suitable for computer implementation—but less intuitive.
In fact, one can readily lose track of the physical meaning of the resulting equations
with this type of methods.
But, whatever method you choose, the first step will always be to know in advance
if a structure is statically determinate or not.

192 9 Statically Indeterminate Structures
9.2 Degree of Indeterminacy
To know to what extent a given structure is statically indeterminate and, most impor-
tantly, the degree of indeterminacy it presents, many authors employ simple formulas.
Thus, if a structure is statically determinate, it is just because we can find the value
of internal forces by resorting to equilibrium equations only.
The degree of indeterminacy of a structure is then defined as the difference

between the number of available equilibrium equations and the number of
unknown forces (internal or external). We therefore speak about internal or
external indeterminacy, respectively.
Following the definition above, some simple formulas can help to ascertain the
degree of indeterminacy of a structure. For instance, in a three-dimensional by n bars
bars intersecting at n nodes nodes or joints, we will have three force equilibrium equa-
tions per node (3 × n nodes ) and three moment equilibrium equations (also 3 × n nodes ).
On the other hand, the number of unknowns is equal to six internal forces and
moments per bar (thus, 6 × n bars ), plus the number of reactions at the supports,
n reactions , which varies greatly with the design of the structure. The degree of inde-
terminacy (also known as the degree of hyperstaticity, doh) would be in this case:
doh = (6 × n bars + n reactions ) − (6 × n nodes + n hinges ),
where H represents the number of hinges present in the structure (since each hinge
can be considered as an additional equation of the type M = 0 or V = 0).
For two-dimensional structures, this reduces to:
doh = (3 × n bars + n reactions ) − (3 × n nodes + n hinges ),
since there are only three equilibrium equations per node and also three internal
forces (two forces and one moment, in fact) per bar.
These equations, though often useful, do not take into consideration additional
circumstances that can arise in a general situation, such as symmetry, for instance.
In a symmetric structure, symmetry provides an additional number of equations
by imposing the equality of internal forces on both sides of the symmetry plane.
The number of these additional equations strongly depends on the structure under
consideration.
If a structure possesses a doh > 0, it is said to be hyperstatic, while, if it verifies
doh = 0, it is said to be isostatic. Both are examples of stable structures, in the
sense that they can bear loads. However, if doh < 0, the structure is unstable and is
indeed a mechanism, unable to support any type of load without catastrophic failure.
Examples of such types of structures are given in Fig. 9.1.
The structure in Fig. 9.1b deserves some comments. Despite having do f = 0, it
is not a classical isostatic structure. This is so since, in its undeformed configuration,
9.2 Degree of Indeterminacy 193
P P
(a) doh = −1, mechanism. (b) doh = 0, critical.
P P
(c) doh = 0, statically determinate. (d) doh = 1, statically indeterminate

(external).
Fig. 9.1 Analysis of static indeterminacy
P P
N N
N N
Fig. 9.2 Analysis of equilibrium in a critical structure. Equilibrium is possible only in a deformed
configuration
it can not achieve equilibrium. Since there is a hinge under the point of application
of the load, no bending moment can appear, and thus only axial internal forces could
occur, see Fig. 9.2 (left).
Therefore, a straightforward application of force equilibrium in the vertical direc-
tion would lead to

Fv = 0, → P = 0,
something that is obviously impossible in a general setting. For equilibrium to be

possible, we invoke it in the deformed configuration, see Fig. 9.2 (right). Once the
node under the load has acquired some vertical displacement, such that the bars are
no longer horizontal, the vertical components of the axial internal forces N are able
to equilibrate the vertical force P.
In sum, critical structures could be analyzed by the imposition of equilibrium
alone, the only limitation being the need of verifying it in the deformed configuration.
It is precisely because we do not know anything about the deformed configuration
that we need to iterate in the solution of these structures.
Finally, it is important to know if the degree of hyperstaticity is internal or external,

i.e., due only to an excessive number of reactions at the supports. In the case of
Fig. 9.1d, the “extra” unknown is clearly due to the presence of two pinned joints,
thus causing four reactions at the supports, more than the three available equilibrium
equations.
9.3 Methods for the Analysis of Indeterminate Structures
The number of equations in the problem to solve will always be the same. We always
have kinematic equations, cf. Eq. (1.3), equilibrium equations, see Eq. (1.6) and,
finally, constitutive equations, Eq. (1.7). These equations could eventually take some
simplified forms, if we accept any of the just presented models, but they always
constitute the same three families of equations. When a structure is statically deter-
minate (or isostatic), the only difference is that equilibrium equations are uncoupled
from the others, thus making it possible to find internal forces and external reactions
without invoking the rest of the equations.
Therefore, to fully solve the resulting problem, taking into account that consti-
tutive equations establish a relationship between static and kinematic variables, we
have only two options. Either we begin by selecting (internal or external) forces as
unknowns, or we begin by selecting displacements as unknowns. The former pos-
sibility gives rise to the family of flexibility (or compatibility) methods, while the
second leads to the family of stiffness (or equilibrium) methods.
9.3.1 Flexibility Methods
In a nutshell, the family of flexibility methods can be sketched as follows:

1. Select as many internal or external forces as hyperstatic unknowns (HU) as the
degree of hyperstaticity of your structure. These will be the unknowns of our
problem.
2. By invoking equilibrium, find a suitable expression for these forces (in general,
static variables) as a function of the HU,
F, σ = f (HU ).
3. By applying constitutive equations, find an expression for the kinematic variables

as a function of the HU:
ε, u = g(HU ).
9.3 Methods for the Analysis of Indeterminate Structures 195
4. Finally, by invoking compatibility equations, find the value of the HU unknowns

and return to your previous equations so as to obtain F, σ or ε, u.
Flexibility methods are very intuitive, as will readily be noticed, and very appro-
priate for manual computation up to a few hyperstatic unknowns. For a high degree
of hyperstaticity, stiffness methods should be preferred.
9.3.2 Stiffness Methods
Stiffness methods enforce the available equations the other way around:
1. Select an appropriate number of degrees of freedom (DoF) of the structure so
that they accurately govern the displacement of the (at least, theoretically) infi-
nite number of degrees of freedom of the structure, considered as a continuum
medium. Again, this operation depends strongly on the selected method in the
family of stiffness methods.
2. By invoking compatibility equations, obtain an expression for the kinematic vari-
ables as a function of DoF,
ε, u = h(DoF).
3. By applying constitutive equations, obtain an expression for the static variables

of the problem in terms of the DoF,
F, σ = (DoF).
4. Finally, by imposing equilibrium (hence the name of the method), obtain the value
of the selected DoF.
5. Eventually, come back to your previous expressions so as to obtain ε, u, F
and σ.
Stiffness methods are adequate for computer analysis of structures, since they are
less intuitive but very systematic. The finite element method is the classical example
of this family of methods, but also the slope-deflection method [15], for instance,
belongs to this family.
9.4 Flexibility or Compatibility Methods
To easily understand the rationale behind flexibility methods, let us consider a simple
example, see Fig. 9.3. This structure will resent the reactions at the supports shown
in Fig. 9.4.
A simple counting of the number of reactions, bars, nodes and hinges will provide:
n bars = 1, n reactions = 4, n hinges = 0 and n nodes = 2, so that
Fig. 9.3 Example of a

statically indeterminate q
structure to solve by the
flexibility method
Fig. 9.4 Reactions in the

structure of Fig. 9.3 M
q
R1
R2 R3
doh = (3 × n bars + n reactions ) − (3 × n nodes + n hinges )

= (3 × 1 + 4) − (3 × 2 + 0) = 7 − 6 = 1.
Therefore, the first step should consist of choosing the so-called “hyperstatic
unknown”. In general, our experience indicates that working with cantilever-like
structures is easy and very convenient, so we decide to take reaction R3 as the
“hyperstatic” one, and call it F hereafter. From now on, we proceed as if there were
no sliding support in the structure, by substituting its equivalent force, F for it. Note
that we treat F as a known, but it is not (yet). The problem to analyze will be,
therefore, the one in Fig. 9.5.
A simple static analysis of this isostatic equivalent structure reveals that
q2
R1 = q − F, M = F,
2
so that
( − x)2
M(x) = F( − x) − q .
2
Fig. 9.5 Reactions in the

structure of Fig. 9.3 M
q
R1
R2 F
9.4 Flexibility or Compatibility Methods 197
The x-axis has been assumed to be centered at the clamped (left) support. These
constitute the equilibrium equations mentioned in Sect. 9.3.1, step number 2: F =
f (HU ).
By applying the principle of virtual work or any other available method, we can
notice that the displacement at the right end of the structure will be

1 F3 q4
u y (x = ) = − + .
EI 3 8
If you observe carefully the resulting equation, it is of the form u = g(HU ), i.e., a
constitutive equation.
The only remaining equation is therefore a compatibility equation. But, what does
a compatibility equation mean in this context? The term “compatibility” should be
understood here as a means of restoring the structure to its initial state. In other
words: in Fig. 9.3 there is a sliding support at the right end of the beam. In the
isostatic equivalent structure of Fig. 9.5, there is no sliding support, so, to enforce
the “compatibility”, with the initial design, we must enforce u y (x = ) = 0:

1 F3 q4
u y (x = ) = − + = 0,
EI 3 8
which provides
3q
F= .
8
This last compatibility enforcement provides the value of the hyperstatic unknown
and enables us, finally, to obtain the expressions for the reactions, internal forces and
moments in the structure.
9.5 Stiffness or Equilibrium Methods
To introduce briefly the main ingredients of stiffness methods, let us consider as a

simple problem the structure in Fig. 9.6.
Fig. 9.6 Simple structure to

introduce stiffness methods q
A B

Fig. 9.7 Decomposition of

the structure into simple q
contributions. Blue lines 0
MAB 0
A B MBA
represents the deformed
structure if a nodes do not
rotate and hence external
load is the only source of (a)
deformation; b Node A
rotates, but there is no load
and node B is maintained as
if it were clamped; and c A
Node B rotates and node A MAA
is kept clamped, with no
B
external loading
(b)
B
MBB
(c)
Internal forces arising in the analysis of the structure in Fig. 9.6 can be readily
decomposed into three distinct contributions (and this is always the case, not only
for this simple problem!):
1. Internal forces and moments provoked by the external load q, if we assume that
the end nodes of the structure do not rotate, see Fig. 9.7a. These are called the
fixed-end moments and will be denoted by a “0” superscript.
2. Internal forces and moments caused by the rotation of node A alone, without any
external load nor rotation on node B.
3. Idem for node B if node A is assumed to be “clamped”.
Every moment follows a one-to-one relationship with the rotation produced at its
own node, so that,
MI I = k I I θI , I = A, B,
with k I I some stiffness to be determined.

Since a moment at node I will always produce a secondary moment at node J of
the same bar, it is therefore easy to conclude that the moment at node I of the bar
I J will be
M I J = M I0J + k I I θ I + k I J θ J ∀ I, J = A, B.
9.5 Stiffness or Equilibrium Methods 199
As mentioned above, stiffness methods are specially well-suited for computer

implementations, so let us express this last equation in matrix form:
0
M AB M AB k A A k AB θA
= + · .
MB A 0
MB A kB A kB B θB
Stiffness values k I I and k I J must be computed by resorting to flexibility methods.

This may sound inconsistent, but needs to be done once for life. Indeed, these values
are, for constant cross-section beams:
4E I
kI I = ,
L
and
2E I
kI J = ,
L
so that

0 4E I 2E I
M AB M AB θ
= + L L
· A .
MB A M B0 A 2E I
L
4E I
L
θB
These equations must be considered exactly as constitutive equations, in the spirit

of Sect. 9.3.2, step number 3. They express internal moments as a function of nodal
rotation angles, hence, degrees of freedom.
The last step to apply is equilibrium. The way in which this is done distinguishes
the most popular stiffness methods (other than the finite element method, of course,
which follows a different process).
9.5.1 Slope Deflection Method
The slope deflection method1 imposes equilibrium directly at each node whose rota-
tion has been considered as a degree of freedom. For the structure in Fig. 9.6, we
have therefore, assumed that there is no bending moment in nodes A nor B,
0 4E I
0 M AB 2E I
θA
= + 2EL I L
· ,
0 M B0 A L
4E I
L
θB
from where we can obtain, by the direct resolution of a 2 × 2 system of algebraic

equations, the value of the unknowns, θ A and θ B :
1 The slope deflection method is the work of George A. Maney, who developed it in 1914 [15],
although some authors attribute it to the German professor Otto Mohr.
pL 2 4E I
0
2E I
θA
= 12
+ 2EL I L
· ,
4E I
θB
2
0 − pL L L
12
where moments and angles have been considered positive if they act in the counter-
clockwise direction. The direct solution of the system gives:
pL 3
θA = − ,
24E I
pL 3
θB = .
24E I
We are now in the position of verifying that, for these values, it holds
M AB = M B A = 0.
Note that when there are chord rotations (in frames, for instance), additional
equilibrium equations must be established for these degrees of freedom. We will not
detail this method that itself deserves an entire book. Instead, we will briefly describe
others in the sections that follow.
9.5.2 Moment Distribution or Cross Method
The moment distribution method was developed by the North American engineer
Hardy Cross2 solves the equilibrium equations in an iterative way. In 1930, in the
absence of any computer, and for large systems of equations, it was considerably
easier to solve them by iterative methods rather than by a direct one, as in the slope-
deflection method. This is at the origin of the popularity of the method until the
appearance of the direct stiffness method and, later on, the finite element method.
If we come back to the constitutive equations, it is clear that they can be expressed
alternatively as:
0
M AB M AB kAA k AB
= + · θ A + · θB .
MB A M B0 A kB A kB B
The Cross method operates on moments rather than angles (though the differences
are minimal, since they are related by the stiffnesses, k I J . It is clear then:
0
M AB M AB MAA μB A · M A A
= + + ,
MB A M B0 A MB B μ AB · M B B
2 H. Cross proposed the moment distribution method in 1930 [6]. He was a professor at the University
of Illinois at Urbana-Champaign. His relevant article was reviewed by 38 referees and is most likely
the most reviewed paper ever! Reviewers comments comprised some 146 pages.
9.5 Stiffness or Equilibrium Methods 201
where
kI J
μJ I =
kI I
represents the relationship between a moment that appears at a node J as a conse-

quence of a rotation at node I and that which appears at node I itself. These are known
as carryover factors. The method starts from the fixed-end moments applied at each
node (and therefore out of equilibrium) and proceeds iteratively until equilibrium
has been achieved within a prescribed tolerance:
1. Starting from node B of our structure in Fig. 9.6, if we clamp it such that θ B = 0,
and free node A such that M A = 0, we find that
0
M AB
θ(1)
A =− ,
kAA
which corresponds to a first-order estimate of the rotation angle at A. If we

substitute it into the M B A constitutive equation, we will obtain a first order estimate
of this moment,
kB A 0
M B(1)A = M B0 A − M = M B0 A − μ AB M AB
0
.
k A A AB
At this stage, the moment at A takes a value
(1) kAA 0
M AB = M AB
0
− M = 0,
k A A AB
and the rotation at node B, θ(1)

B = 0.
2. Do the same with node A, by fixing its rotation to θ(1)
A , and free node B. We have
−M B0 A − k B A θ(1) M (1)
θ(2)
B =
A
= − BA .
kB B kB B
This also implies that M B(2)A = 0, while
(2) (1) (1) k AB (1)

M AB = M AB + k AB θ B = M AB − M ,
kB B B A
and θ(2) (1)

A = θA .
We proceed iteratively until the out-of-equilibrium moments at every node reach
a prescribed threshold.
It is important to note that, even if we work exclusively with moments, equilibrium
equations actually involve nodal rotations. Even if Cross argued that the convergence
rate of the method is fast, in practice, this method has been abandoned nowadays,
and it is studied solely from a historical perspective. The finite element method [8],
for instance, is preferred in industry and architecture as the ubiquitous method for the
computation of internal forces by employing a computer. A thorough analysis of the
finite element method is nevertheless beyond the scope of this book. The interested
reader is referred to the plethora of good books available on the subject.
9.6 The Direct Stiffness Method
The previous methods have been introduced in a matrix form, even if some of them
were developed long before matrices acquired popularity.3 As long as the number
of chord rotations is significant in a given structure, these methods become less and
less intuitive and amenable to a computer implementation.
On 6 November 1959, M.J. Turner, who was at that time the head of the Struc-
tural Dynamics Unit at Boeing, presented in a conference at Aachen, Germany, the
first paper on a new method coined the Direct Stiffness Method. Surprisingly, the
development of matrix methods of structural analysis originated in the aeroelastic
community, mainly in aviation companies. This conference paper is now lost, so that
the first paper on the topic is considered to be [20].
The direct stiffness method aims at developing a completely general technique,
in which the internal forces can be expressed as functions of the“element” degrees
of freedom. At this time, the word “element” should not yet be understood in the
“finite element” sense, but as a sort of “ingredient” of the structure, typically, a bar.
Therefore, the final objective is to find an expression of the type:
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
Ni k11 k12 k13 k14 k15 k16 ui
⎜ Vi ⎟ ⎜k21 k22 k23 k24 k25 k26 ⎟ ⎜ vi ⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜ Mi ⎟ ⎜k31 k32 k33 k34 k35 k36 ⎟ ⎜ θi ⎟
⎜ ⎟=⎜ ⎟ · ⎜ ⎟,
⎜ N j ⎟ ⎜k41 k42 k43 k44 k45 k46 ⎟ ⎜u j ⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎝ V j ⎠ ⎝k51 k52 k53 k54 k55 k56 ⎠ ⎝v j ⎠
Mj k61 k62 k63 k64 k65 k66 θj
where i and j represent the initial and final nodes of the considered bar. Here, the
internal forces Ni , Vi and Mi are expressed in a completely general framework
as a function of the bar’s degrees of freedom, its axial displacement u i , its shear
displacement vi and its nodal rotations θi , with i = 1, 2, the initial and end nodes of
the bar.
At this moment, it is well-known that, for a bar of constant cross section A, second
area moment I , length L, and made of material with a Young’s modulus E, this matrix
is:
3 Professor
Carlos Felippa [7] attributes the invention of matrices to Cayley at Cambridge or to
Grassmann in Germany, around 1858, but their use was not common before 1930.
9.6 The Direct Stiffness Method 203
⎛ ⎞ ⎛ ⎞
⎛ ⎞ EA
L
0 0 − ELA 0 0 ui
Ni ⎜ ⎟ ⎜v ⎟
⎜ Vi ⎟ ⎜ 0
12E I
L3
− 6E
L2
I
0 − 12E
L3
I 6E I
L2 ⎟ ⎜ i⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜ Mi ⎟ ⎜ 0 − 6E I 4E I
0 − 6E I 2E I ⎟ ⎜ θi ⎟
⎜ ⎟=⎜ L2 L L2 L ⎟ · ⎜ ⎟, (9.1)
⎜ N j ⎟ ⎜− E A 0 EA ⎟ ⎜u ⎟
⎜ ⎟ ⎜ L 0 0 0 ⎟ ⎜ j⎟
⎝ Vj ⎠ ⎜ L
⎟ ⎜ ⎟
⎝ 0 − 12E L3
I
− 6E
L2
I
0 12E I
L3
− 6E
L2
I ⎠ ⎝v j ⎠
Mj
0 6E I
L2
2E I
L
0 − 6E
L2
I 4E I
L
θj
or, in matrix notation,
P = K · δ.
This matrix equation can be understood, of course, as a matrix constitutive equation.

If we consider the obvious possibility of a structure made of more than one bar,
and that these bars are oriented in different directions, we will need to express these
matrices in a common system of reference. For a plane, two-dimensional structure,
the change of coordinates of a vector to a new system {x , y } from a reference one
{x, y}, is given by:

Px cos α sin α Px
= · ,
Py − sin α cosα Py
or, equivalently,
P = L T · P.
The transpose sign is only due to a convention in which the transformation matrix L
is usually assumed to be the transpose of the one just introduced.
Therefore, if we consider one particular constitutive equation as
P = K · δ ,
by introducing the change of reference, we have
L T P = K · L T δ,
from where
P = (L K L T )δ,
such that the new stiffness matrix will be
K = L K LT .
By just applying equilibrium equations at every node in the structure, we will

obtain a global system of equations. Compatibility equations will also enable us
to express the relationship between each bar’s degrees of freedom δ and the struc-
ture’s degrees of freedom, u. This compatibility enforcement is nowadays known as
assembly of the structure’s stiffness matrix.
9.7 The Finite Element Method
The direct stiffness method arises from a completely discrete origin, given by
Eq. (9.1). To obtain this equation, standard calculations are made, possibly by
employing flexibility methods. Soon after the post-World War II period, the com-
munity realized that the direct stiffness method can be obtained from a completely
continuum setting starting from the differential equation governing beam-bending
phenomena. Moreover, mathematicians soon realized that, in a 1943 paper, Courant
had employed a mesh of triangles to solve, starting from variational principles, vibra-
tion problems [5]. This readily opened the possibility to develop matrix analysis
methods with a sound physical basis. Nowadays, the resulting method, coined as the
finite element method, is the ubiquitous method in engineering analysis.
Without the aim to be completely exhaustive (a simple search on Amazon.com
for the term “finite elements” produced some 6,100 results on December 2017), we
will briefly describe the method here for completeness. The interested reader can
consult some very good introductory references such as [8], for instance.
Let us consider first, for the sake of simplicity, the differential equilibrium equation
of a truss, Eq. (4.1),
d du x
AE = − p.
dx dx
This equation constitutes what is commonly known as the strong form of the truss
problem, if completed with adequate boundary conditions. The weak form will be
obtained by first multiplying both sides of the equation by an arbitrary function w
(the so-called weight function), and integrating along the length of the bar, so that
l l
d du
w AE dx + wp d x = 0, ∀w.
0 dx dx 0
In order to simplify a little bit the previous expression, recall the form of the derivative
of a product of functions,
d df dw df d dw
(w f ) = w + f =⇒ w = (w f ) − f,
dx dx dx dx dx dx
that, after integration, provides
9.7 The Finite Element Method 205
l l
df d dw
w dx = (w f )d x − f d x,
0 dx 0 dx 0 dx
or, equivalently,
l
df dw
w d x = (w f )|0 − f d x,
0 dx 0 dx
which is precisely the well-known formula of integration by parts.

If we apply this formula to the weak form of the problem,

d du du l dw du
w AE dx = w AE − AE d x,
0 dx dx d x 0 0 d x d x
such that
⎛ ⎞

l
⎜ du ⎟ dw du
⎜w A E ⎟ − AE d x + wp d x = 0, ∀w with w(l) = 0.
⎝ d x⎠
0 dx dx 0
σx 0
Simplifying a bit,

dw du
AE d x = (w At¯)x=0 + wp d x = 0, ∀w with w() = 0. (9.2)
0 dx dx 0
Since w is arbitrary, we have chosen w(x = ) = 0 to simplify the resulting expres-

sions. This is an arbitrary but very useful choice. This last formula constitutes the
weak form of the truss bar.
Note that the last expression coincides with the principle of virtual work devel-
oped in Eq. (4.3). Albeit it has been obtained by a very different method, both
expressions are exactly the same.
The problem thus consists in finding a sufficiently smooth function u(x) such
that it satisfies the essential boundary conditions (those related to displacements)
that satisfies Eq. (9.2). This solution is not always easy to find. In fact, we have
transformed a differential equation into an integro-differential problem, but many
difficulties remain the same. What is the interest in all these developments? Precisely
the possibility to approximate the solution to the problem within a prescribed degree
of accuracy. Let us explain this a little bit further.
Polynomials are well-known to have nice approximation properties. Therefore,
let us try to approximate the solution (the displacement field) by means of the best
available polynomials,
(1) (2)
u(1) (x) u(2) (x)
(1) (2)
Fig. 9.8 Approximation of the displacement field within each finite element. Here, a bar is split
into two elements, (1) and (2), and a polynomial approximation is assumed within each one
u e = α0e + α1e x + α2e x 2 + · · · ,
where e indicates that this approximation is made within element e, see Fig. 9.8.
Each bar will be split into one or more elements of finite size (hence the name
finite elements) and a polynomial approximation of the displacement field will be
adopted within each one of these regions. Here, αi represent the coefficients for each
monomial, to be chosen so as to guarantee that the resulting displacement field is
continuous.
To guarantee that the integral Eq. (9.2) exists, the displacement field must be
continuous,
u (1) (x2(1) ) = u (2) (x1(2) ).
An easy way of satisfying this condition is to take the αi coefficient as the nodal
values of the displacement or, in other words, to make the polynomials take unity
value at the element nodes. It is also well-known that, for the finite element method
to provide an accurate solution, it must be able to reproduce, at least piece-wise,
constant strain fields within each element. Therefore, the simplest polynomial to
satisfy this condition needs to be at least linear,
u e (x) = α0e + α1e x.
By putting the nodes at the beginning and the end of the element, we will have the
same number of nodal values as parameters in the polynomial approximation. In
other words:

α0e
u (x) = 1 x
e
e = p(x)α .
e
α1
p(x)

αe
le
Fig. 9.9 A simple linear
finite element
xe1 xe2
1 e 2
To determine the precise values of these αi coefficients, we perform a simple manipu-

lation, by identifying the displacement predicted by the finite element at the beginning
and the end of the element with that of its two nodes, see Fig. 9.9:
e e e
u e (x1e ) ≡ u e1 = α0e + α1e x1 u 1 x1 α0
→ e1 = ,
u e (x2e ) ≡ u e2 = α0e + α1e x2 u2 1 x2e α1e

de Me αe
such that
αe = (M e )−1 d e .
Rearranging this result, we have
u e (x) = N e (x)d e , con N e (x) = p(x)(M e )−1 .

The row matrix N e (x) = N1e (x) N2e (x) = p(x)(M e )−1 is known as the shape func-
tions matrix. To determine its precise value, a simple manipulation provides:
e
1 x2 −x1e 1 x2e −x1e
(M e )−1 = = ,
x2e − x1e 1 1 le 1 1
such that:

N e = N1e (x) N2e (x) = p(x)(M e )−1

x2e −x1e 1 1
= 1x = e x2e − x x − x1e ,
1 1 le l
see Fig. 9.10.

The just-computed shape functions verify
N Ie (x J ) = δ I J ,
or, in other words, they are interpolant,

N1e (x)
1
N2e (x)
xe1 e xe2
Fig. 9.10 A simple linear finite element

n ne
n ne
u (x J ) =
e
N Ie (x J )u eI = δ I J u eI = u eJ .
I =1 I =1
This means that the predicted values for the displacement at both ends of the element
coincide with the nodal displacement values.
In this same spirit, we can compute the derivatives of the just-approximated dis-
placement field,
du d d Ne e d N1e e d N2e e
= (N e d e ) = d = u + u .
dx dx dx dx 1 dx 2
that must be expressed in matrix form, of course, by:
du e u e
d N2e
= ddNx1 1 = Be d e .
dx dx u e2
Proceeding in exactly the same way, one can envisage a quadratic element, for
instance, by increasing the polynomial degree of the approximation, see Fig. 9.11.
Obviously, this quadratic element possesses three nodes (since a parabola is described
by three points!).
For bars subjected to bending moments, the differential equation was presented in
Eq. (6.7). Consider, for the sake of simplicity of the exposition, the two-dimensional
case of a bar subjected to only one moment. In this case, the main novelty with
respect to the just explained truss-bar element is the presence of derivatives up to
order four (it was order two for trusses). Therefore, by following the same procedure
introduced before, we will need to apply integration by parts twice, so as to give

d 2w d 2u y dw
E I dx = wpd x + m̄ + (w s̄)|Γs , ∀w,
Ω dx2 dx2 Ω dx Γm
N1e (x) N2e (x)

1
N3e (x)
1 2 3
Fig. 9.11 A quadratic finite element
where w needs to be smooth enough so as the integral of its second derivative to be

defined. Here, m̄ and s̄ represent, respectively, the externally applied moments and
shear forces, respectively, along portions of the boundary denoted by Γm and Γs .
For the integrals to exist, it is necessary that both w and u y possess a C 1 degree
of continuity, i.e., it is necessary that they are continuous and possess continuous
derivatives. Therefore, the type of linear approximation presented in Fig. 9.10, for
instance, will not be adequate. A very practical solution to this difficulty is to employ
Hermite polynomials for the approximation of both w and u y . This implies the need
to consider both the vertical displacement and the rotation of each node as degrees
of freedom,
T
d e = u y1 θ1 u y2 θ2 ,
which implies that the force vector will be

T
f e = f y1 m 1 f y2 m 2 ,
with m I the nodal moments (note that m I = m(x I )). Hermite polynomials take the
form
1 e
Nu1 = (1 − ξ)2 (2 + ξ), Nθ1 = (1 − ξ)2 (1 + ξ),
4 8
1 e
Nu2 = (1 + ξ)2 (2 − ξ), Nθ2 = (1 + ξ)2 (1 − ξ),
4 8
with ξ = 2xe
− 1, such that −1 ≤ ξ ≤ 1. Again, a fundamental characteristic of the
finite element method is that it holds
Nu I (x J ) = δ I J and that Nθ I (x J ) = δ I J .
Once this Hermite approximation has been substituted into the weak form of the
problem, we arrive at a discrete, algebraic systems of equations
Kd = f + r
where

d2 Ne
Ke = E I B e T B e d x, with B e = ,
Ωe dx2

d NeT

f =
e eT
N pd x + (N eT
s̄)|Γs + m̄ .
Ωe dx

Γm

f eΩ
f eΓ
The element stiffness matrix takes the form:

⎛ ⎞
12 6e −12 6e
EI ⎜ 4e2 −6e 2e ⎟
Ke = B eT E I B e d x = e 3 ⎜
⎝
⎟.
Ωe 12 −6e ⎠
Sym 4e2
This value is exactly the same value predicted by classical, discrete matrix analysis
procedure. In other words, for Euler–Bernoulli–Navier beams, the finite element is
exact and not a mere approximation of the solution.
The reader could at this point ask his- or herself about the benefit of performing all
this complex mathematical analysis to arrive to the solution predicted some 50 years
earlier by classical matrix procedures. Indeed, the benefit consists of the possibility of
extending this procedure not to the case of bars, but to any other differential equation,
including the noteworthy equations of linear elasticity, Eq. (1.11). For instance, in
Fig. 9.12 a finite element model of an arch that employs three-dimensional linear
finite element is shown. More sophisticated models are also equally possible, see
Fig. 9.13, where a shell finite element model of the Los Manantiales restaurant at
Xochimilco, Mexico, by Félix Candela, is shown.
The finite element method is nowadays the ubiquitous method for the simulation
of a wide range of physical phenomena, ranging from solid to fluid mechanics,
spanning various scales, boundary conditions and loading types. This technique has
exceeded by far what we intended with this book. We highly recommend that the
reader learn as much as possible about it, since we are confident that it will continue
to be the reference technique for sophisticated computations in the field of structural
mechanics.
Fig. 9.12 Three-dimensional finite element model of an arch subjected to self weight
Fig. 9.13 Finite element model of the shell roof for the Los Manatiales restaurant in Xochimilco,
Mexico. Félix Candela, arch. M. Sc. Thesis by T. Izuel, Universidad de Zaragoza
Worked Examples
9.1. Consider the following statically indeterminate truss under the loads depicted
in Fig. 9.14. The bar AB is subjected to an increment of temperature ΔT = 20 ◦ C
and was mistakenly manufactured shorter than needed. The magnitude of this defect
is δ AB = 0.01 cm. In addition, the support at D undergoes a displacement to the
right u D = 1 mm due to a failure in the soil under the foundation. Determine the
internal forces in the truss by considering E = 2.1 · 106 kp/cm2 , A = 25 cm2 , α =
1.2 · 10−5 ◦ C−1 .
Fig. 9.14 Truss for Example 10 T

9.1
1T
B C
2m
A D
2m
Solution 9.1. (a) As mentioned in the previous sections, the first step in the analysis
of a structure is always to determine its degree of static indeterminacy. This will
enable us to know the type of problem we are facing. In this particular case, we
have 6 bars −2 × 4 nodes +4 reaction forces on the supports. So this system is
statically indeterminate with a degree 2 of indeterminacy. A careful observation
of the structure reveals that one of the sources of indeterminacy is one “extra”
support (there are two reactions at A and another two at D, while there are only
three global-equilibrium equations). The other comes from the bars B D and AC,
crossing over each other.
(b) We consider an equivalent determinate system, like the one in Fig. 9.15. This
choice is obviously arbitrary, but we must take into account the nature of the
indeterminacy. In other words, we could have chosen any other reaction in A or
D, or the force in bar AC, but never two reactions or two internal forces in any
P
B C
2m
X
A D
HA Y
VA VD
2m
Fig. 9.15 Equivalent statically determinate truss for Example 9.1
bar.
We have named these hyperstatic unknown variables as X and Y , respectively.
Now imposing equilibrium conditions for the free-body diagram depicted in
Fig. 9.15,
X X
Fx = 0 ⇒ H A = Y + P + √ − √ = Y + P ⇒ H A = Y + P,
2 2
X X
Mz (A) = 0 ⇒ 2VD − 2P + 2 √ − 2 √ = 0 ⇒ VD = P,
2 2
X X
Fy = 0 ⇒ V A + V D + √ − √ − Q = 0 ⇒ V A + V D = Q ⇒ V A = Q − P.
2 2
We proceed to equilibrate each node in the truss. For instance, in node D we

have, see Fig. 9.16,
X
Fx = 0 ⇒ N AD = Y − √ ,
2
Fig. 9.16 Equilibrium at NCD

node D X
α D
NAD Y
X
Fy = 0 ⇒ NC D = −P − √ .
2
Following the same procedure for node C we have (Fig. 9.17):
Fig. 9.17 Equilibrium at C

node C NBC
α
NAC
X
−P − √
2

N AC X √
Fy = 0 ⇒ √ − P + √ = 0 ⇒ N AC = P 2 + X,
2 2

N AC N AC
Fx = 0 ⇒ N BC + √ = 0 ⇒ N BC = − P + √ .
2 2
And finally for node B (Fig. 9.18).
Fig. 9.18 Equilibrium at Q

node B
B X
P −P − √
2
α
X
NAB
X
Fy = 0 ⇒ N AB = −Q − √ .
2
(c) To determine the value of redundant forces, let us consider two different virtual
systems associated with the equivalent statically determinate truss.
By solving this first virtual system, we find (Fig. 9.19):
B C
2m
A D
I
HA 1
VAI I
VD
2m
Fig. 9.19 Virtual system I
P = 0; Q = 0; Y = 1; X = 0,
so that,
I
N AB = 0; N AC
I
= 0; N BC
I
= 0; N AD
I
= 1; NCI D = 0.
For the second virtual system, coined as I I , see Fig. 9.20, we have,
P = 0; Q = 0; Y = 0; X = 1,
so that,
−1 −1 −1 −1
II
N AB = √ ; N AC
II
= 1; N BC
II
= √ ; N AD
II
= √ ; NCI ID = √ .
2 2 2 2
Fig. 9.20 Virtual system I I

B C
2m
1
A D
II
HA
VAII II
VD
2m
(d) Applying the principle of virtual work,

ψ
ψ
FN · δ N = Nb · ΔL b .
N b
For this exercise, it is necessary to take into account that, for bar AB, its increment
in length is due not only to its internal force, but also to its temperature increment
and the defective length,
N AB · L AB
Δ AB = + α · ΔT · L AB + δ eAB .
EA
In the application of the principle of virtual work to the virtual system I , we
must take into account that the real horizontal displacement to the right of point
D will be the actual movement of this point, or, in other words,

5
X 200
0.1 cm = u x (D) = NiI · ΔL i = Y − √ · · 1,
i=1
2 EA

X 200
0.1cm = Y − √ · ⇒ 26.25 · 103 kp
2 52.5 · 106
X X
= Y − √ ⇒ Y = 26.25 · 103 + √ .
2 2
Applying again the principle of virtual work to the second virtual system, I I ,
the relative displacement between nodes B and D must be equal to the length
increment of the bar B D (in other words: this relative displacement must be
“admissible”), so,
√
ψ −X · 2 · 200
FN · δN = 1 · δB + 1 · δD = δB + δD = ΔB D = .
N
EA
On the other hand,

NbI I · ΔL b
b
I I · ΔL
= N AB II II II II
AB + N AC · ΔL AC + N AD · ΔL AD + N BC · ΔL BC + NC D · ΔL C D

X 200 −1
= −10−4 − √ + αΔT 200 + 0, 01 · √
2 EA 2
√ √
(1000 2 + X )200 2
+ · [1]
EA
⎡ ⎤ ⎡ ⎤
(Y − √X ) (−1000 − √X )
⎣ 2 ⎦ −1 ⎣ 2 ⎦ −1
+ 200 · √ + 200 · √
EA 2 EA 2
⎡ ⎤
(−1000 − √X )
⎣ −1
200⎦ · √ .
2
+
EA 2
Therefore:
√
−X · 2 · 200
E A
X 200 −1
= −10−4 − √ + αΔT 200 + 0, 01 · √
2 EA 2
√ √
(1000 2 + X )200 2
+ · [1]
EA
(−1000 − √X )
(Y − √X2 ) −1 2 −1
+ 200 · √ + 200 · √
EA 2 EA 2

(−1000 − 2 )X
√
−1
+ 200 · √ .
EA 2
Solving the system with two equations and two unknown variables X and Y , we
have that
X = 4353.07 kp Y = 29328.08 kp,

so that the internal forces result to be.
N AB = −13078.08 kp (Compression),
N AC = 5767.28 kp (Traction),
N AD = 26250 kp (Traction),
N BC = −4078.08 kp (Compression),
N B D = X = 4353.07 kp (Traction),
NC D = −4078.08 kp (Compression).

9.2. Consider the following statically indeterminate structure under the loads
depicted in Fig. 9.21. Compute the bending-moment diagram for the structure.
Fig. 9.21 Structure for q

Example 9.2
B C
h
A D
Solution 9.2. (a) The first step will be, as always, to determine the degree of static
indeterminacy of the structure and to know in advance if it is internal or external.
In this case; we have 3 unknown internal forces in 3 beams and 6 reaction forces
at the supports to determine, so we have 15 unknown variables. In the other hand,
we have 3 equations per node (and a total of 4 nodes) and 2 hinges, where we know
that bending moments vanish. Therefore, this system is statically indeterminate
of degree 15 − 14 = 1 (and this is clearly due to one redundant external reaction).
(b) We consider an equivalent isostatic system depicted in Fig. 9.22. Solving the
external equilibrium to obtain the reaction forces, we find that.
Fig. 9.22 Isostatic system q

equivalent to the initial
structure for Exercise 9.2.
Here, M represents the
redundant variable
B C
h
M M
A M M D
h h
qL qL
2 2
L
M
H A = HD = ,
h
qL
V A = VD = .
2
M A = M.
The bending-moment diagram for the equivalent isostatic system is shown in

Fig. 9.23.
qL qx2
M (x) = 2
x − 2
x
B C
qL2
8
x
M (x) = M (1 − h
)
x
A D
M M
Fig. 9.23 Bending-moment diagrams for the equivalent isostatic system to the initial structure for
Exercise 9.2
(c) The virtual system associated with the redundant variable is shown in Fig. 9.24.
This gives rise to the bending-moment diagram shown in Fig. 9.25.
Fig. 9.24 Virtual system for

the Exercise 9.2 B C
h
1 1
A 1 1 D
h h
L
B C
x
M I (x) = 1 − h
x
A D
1 1
Fig. 9.25 Bending-moment diagrams for the virtual system I
(d) According to the compatibility condition θ D = 0, we must enforce at the real

support,
ψ
L
FN · u N = Mzψ · κz d x + · · ·
N b 0
Therefore,

1 h x x 2M h 2x x2
1 · θD = 0 = 2 · M 1− 1− dx = 1− + 2 dx
EI 0 h h EI 0 h h

2M h 2Mh
= h−h+ = ⇐⇒ M = 0.
EI 3 3E I
Therefore, the resulting bending moment diagram is shown in Fig. 9.26.

Fig. 9.26 Momentum qL qx2
M (x) = 2
x − 2
diagrams for the structure for
Exercise 9.2 x
B C
qL2
8
A D
9.3. Compute the vertical displacement of node B, assuming that all the bars
depicted in Fig. 9.27 are subjected to an increment of temperature ΔT .
3L
2
√
60o
A
C
Fig. 9.27 Trus for Exercise 9.3
Solution 9.3. (a) Again, the first step is to determine the degree of indeterminacy
of the structure. For this truss we have 3 bars −2 × 3 nodes +4 reaction forces
on the supports. So this system is statically indeterminate of degree one (this is
clearly due to one redundant external reaction).
(b) We consider an equivalent statically determinate system, shown in Fig. 9.28,
where X is the redundant unknown variable.
B
3L
2
√
60o
A C
X
Fig. 9.28 Equivalent statically determinate truss for Example 9.3

Solving the equivalent statically determinate system depicted in Fig. 9.28, we

have that the reaction forces are,
H A = X (→); V A = VC = 0.
Obtaining now the internal forces by imposing equilibrium at each node of the
truss,
N AC = −X (Compression),
N AB = N BC = 0.
We suggest two different virtual systems. The first one is aimed at computing
the displacement at point B, while the second one will serve to compute the
hyperstatic variable X . Solving this virtual system I (see Fig. 9.29), the reaction
forces on the supports turn out to be
1
H A = 0 V A = VC = (↓)
2
Fig. 9.29 Virtual system I 1

for Exercise 9.3
3L
2
√
60o
A C
The internal forces for this truss are, therefore,
1
I
N AB = N BC
I
= √ (Compression)
3
−3
I
N AC = √ (Traction)
3
Solving this second virtual system, I I , (see Fig. 9.30), the reaction forces on the
supports turn out to be
H A = 1(→) V A = VC = 0.
Fig. 9.30 Virtual system I I B

for Exercise 9.3
3L
2
√
60o
A C
1
The internal forces for this truss are,

II
N AB = N BC
II
= 0,
II
N AC = −1 (Compression)
See Table 9.1 for a summary of the results.
Table 9.1 Results for the truss structure

Bar L N ΔL NI NII
√
AB L 0 αLΔT 3
0
√3
BC L 0 αLΔT 3
3√ 0
−X L − 3
AC L −X EA + αLΔT 6 −1
Now, by taking into account the virtual work principle,

ψ
F psi δ = ΔL b Nb + Nb ψ(αL b ΔT ).
b
Combining virtual system I with the isostatic equivalent one,

√ √ √
3 3 3 −X L
1 · δB = · αLΔT + · αLΔT − · + αLΔT .
3 3 6 EA
And, by combining virtual system I I with the isostatic equivalent one,

−X L
1 · δCx = −1 · + αLΔT .
EA
Finally, by applying the compatibility condition, i.e., by noting that in the real
system δCx = 0,
X = αE AΔT,
and therefore,
√ √ √
2 3 3 −X L 2 3
δB = αLΔT − + αLΔT ⇒ δ B = αLΔT.
3 6 EA 3
Appendix A
Indicial Notation
A.1 Indicial Notation for Vector and Matrix Operations
In tensor analysis, an extensive use of indicial notation is made. Operations using

Cartesian components of vectors and matrices can be expressed efficiently using this
notation.
A.1.1 Elements of a Matrix
A set of n variables x1 , x2 , . . . , xn is denoted using indicial notation such as as

xi , i = 1, 2, . . . , n.
Let us consider now the equation that describes a plane in a three dimensional
space x1 , x2 , x3 ,
a1 x1 + a2 x2 + a3 x3 = p,
where ai and p are constants. This equation can be expressed alternatively as

3
ai xi = p.
i=1
At this point, we can use the convention usually referred to as Einstein notation
or Einstein summation convention, which implies summation over a set of indexed
terms in a formula when an index appears twice in a single term along the variational
rank of the indexes (1 to 3 in the above equation).
Therefore,
ai xi = p
In the same spirit,

Structural Integrity 4, https://doi.org/10.1007/978-3-319-72935-0
228 Appendix A: Indicial Notation
• Vectorial scalar product:

u · v = u i vi
• Norm of a vector: √ √
||u|| = u·u= ui ui
• Derivative of a function:
n
∂f ∂f
df = d xi = d xi = f ,i d xi
i=1
∂ x i ∂ xi
Let us define the Kronecker Delta δi j as a symbol1 whose values can be,

1 if i = j
δi j =
0 if i = j
In the same way, let us define the permutation symbol εi jk as,

⎧
⎨ 0 if i = j j = k i = k
εi jk = 1 if i, j, k for an even number permutation
⎩
−1 if i, j, k form an odd permutation
A.1. Demonstrate the next properties in indicial notation:

• δi j δi j = 3
• εi jk εi jk = 6
• δi j v j = vi
• The matrix determinant 3 × 3 can be expressed like |A| = εi jk ai1 a j2 ak3
• The three vectorial components of a vector product of two vectors are (v × w)i =
εi jk v j wk
• The Kronecker Delta function and the permutation symbol have the next relation-
ship εi jk εist = δ js δkt − δ jt δks
Solution A.1. • Indexes i, j appear repeated on the equation, so we must sum them
from 1 to 3. Using the definition of the Kronecker Delta function, we have
δ11 δ11 + δ12 δ12 + δ13 δ13 + δ21 δ21 + δ22 δ22 + δ23 δ23 + δ31 δ31 +
+δ32 δ32 + δ33 δ33 = δ11 δ11 + δ22 δ22 + δ33 δ33 = 1 + 1 + 1 = 3
• Considering i = 1 and varying j, k from 1 to 3 we have
1It is a second order tensor, actually.

Appendix A: Indicial Notation 229
ε111 ε111 + ε112 ε112 + ε113 ε113 + ε121 ε121 + ε122 ε122 + ε123 ε123 +
+ε131 ε131 + ε132 ε132 + ε133 ε133 = ε123 ε123 + ε132 ε132 =
= 1 · 1 + (−1) · (−1) = 2
For i = 2,
ε211 ε211 + ε212 ε212 + ε213 ε213 + ε221 ε221 + ε222 ε222 + ε223 ε223 +
+ε231 ε231 + ε232 ε232 + ε233 ε233 = ε213 ε213 + ε231 ε231 =
= (−1) · (−1) + 1 · 1 = 2
Finally, for i = 3, ε3 jk ε3 jk = 2, we have,
εi jk εi jk = 6
• Since δ ji vanishes for i = j then δi j v j = vi ,

⎤ ⎤
δ11 · v1 + δ12 · v2 + δ13 · v3 v1

δ21 · v1 + δ22 · v2 + δ23 · v3 ⎦ = v2 ⎦ = vi

δ31 · v1 + δ32 · v2 + δ33 · v3 v3
• The determinant is defined by:

a11 a12 a13

a21 a22 a23 =

a31 a32 a33
= a11 a22 a33 + a21 a32 a13 + a31 a12 a23 − a11 a32 a23 − a12 a21 a33 − a13 a22 a31
= ε123 a11 a22 a33 + ε231 a21 a32 a13 + ε312 a31 a12 a23 + ε132 a11 a32 a23 + ε213 a12 a21 a33
+ε321 a13 a22 a31 = εi jk ai1 a j2 ak3 ,
which is precisely the suggested expression (summing for i, j, k from 1 to 3).
• If, in the definition of determinant of a matrix, we change the first column by the
unit vectors ei , the second component by v, and the third one by w, the vector
product is given by the following indicial expression
(v × w) = εi jk ei v j wk ,
(v × w)i = εi jk v j wk ,
230 Appendix A: Indicial Notation

î ĵ k̂

v × w = v1 v2 v3 = îv2 w3 + k̂v1 w2 + ĵv3 w1 − k̂v2 w1 − îv3 w2 − ĵv1 w3 =
w1 w2 w3
εi jk êi v j wk .
To obtain the ith component it is enough to multiply by ei , so
(v × w)i = εi jk v j wk .
• Finally, it is easy to prove the last expression by operating component by compo-

nent:
– If j = k, both being free indexes in the expression, and taking into account
that the permutation symbol vanishes when the indexes are the same, the first
member of the equation is equal to zero. As per the second term, if s = j = k
and t = j = k, then the second term is 1 − 1 = 0; conversely, if s = j = k and
t = j = k, then the second term is 0 − 0 = 0; if s = j = k and t = j = k, then
the second term is 0 − 0 = 0.
– If s = t, following the same rationale, we obtain the same result by just changing
s, t by i, j.
For i = 1 for instance:

⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
ε123 ε123 δ22 δ33 − δ23 δ32 1−0 1
⎢ ε132 ε123 ⎥ ⎢ δ32 δ23 − δ33 δ22 ⎥ ⎢ 0 − 1 ⎥ ⎢ −1 ⎥
ε1 jk ε1st ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎣ ε123 ε132 ⎦ = ⎣ δ23 δ32 − δ22 δ33 ⎦ = ⎣ 0 − 1 ⎦ = ⎣ −1 ⎦
ε132 ε132 δ33 δ22 − δ32 δ23 1−0 1
⎡ ⎤
1·1
⎢ (−1) · 1 ⎥
=⎢ ⎥
⎣ 1 · (−1) ⎦ = δ js δkt − δ jt δks .
(−1) · (−1)

References
1. Betsch P, Steinmann P (2002) Frame-indifferent beam finite elements based upon the geomet-
rically exact beam theory. Int J Numer Methods Eng 54(12):1775–1788
2. Block P, Ochsendorf J (2007) Thrust network analysis: a new methodology for three-
dimensional equilibrium. J Int Assoc Shell Spatial Struct 48(3):167–173
3. Bonet J, Wood RD (2008) Nonlinear continuum mechanics for finite element analysis. Cam-
bridge University Press, Cambridge
4. Calatrava S (1981) Zur Faltbarkeit von Fachwerke. PhD thesis, ETH Zurich. https://www.
research-collection.ethz.ch/handle/20.500.11850/137273
5. Courant R (1943) Variational methods for the solution of problems of equilibrium and vibra-
tions. Bull Am Math Assoc 42:2165–2186
6. Cross H (1930) Analysis of continuous frames by distributing fixed-end moments. In: Proceed-
ings of the American society of civil engineers, ASCE, pp 919–928
7. Felippa CA (2001) A historical outline of matrix structural analysis: a play in three acts. Comput
Struct 79(14):1313–1324
8. Fish J, Belytschko T (2007) A first course in finite elements. Wiley, New Jersey
9. Han SM, Benaroya H, Wei T (1999) Dynamics of transversely vibrating beams using four
engineering theories. J Sound Vibr 225(5):935–988
10. Heyman J (1982) The masonry arch. Ellis Horwood
11. Huerta S (2004) Arcos, bóvedas y cúpulas. Geometría y equilibrio en el cálculo tradicional de
estructuras de fábrica, Instituto Juan de Herrera
12. Huerta S (2008) The analysis of masonry architecture: a historical approach: to the memory of
professor henry j. cowan. Architect Sci Rev 51(4):297–328
13. Jelenić G, Crisfield MA (1999) Geometrically exact 3d beam theory: implementation of a
strain-invariant finite element for statics and dynamics. Comput Methods Appl Mech Eng
171(1):141–171
14. Love AEH (1927) A treatise on the mathematical theory of elasticity. Cambridge University
Press, Cambridge
15. Maney GA (1915) Studies in engineering. University of Minnesota, Minneapolis
16. Muttoni A (2011) The art of structures. EPFL Press, Lausanne
17. Muttoni A, Schwartz J, Thürlimann B (1996) Design of concrete structures with stress fields.
Springer Science & Business Media, Berlin
18. Reddy JN (2008) An introduction to continuum mechanics. Cambridge University Press, Cam-
bridge
19. Simo JC, Vu-Quoc L (1988) On the dynamics in space of rods undergoing large motions – a
geometrically exact approach. Comput Methods Appl Mech Eng 66(2):125–161
20. Turner MJ, Clough RJ, Martin HC, Topp LJ (1956) Stiffness and deflection analysis of complex
structures. J Aeronaut Sci 23(9):805–823

Index
A Center of torsion, 171

Arches, 64 Chord rotation, 200, 202
Arris, 68 Collignon formula, see also Collignon–
Axial internal force, 77 Jourawski formula
Axial stiffness, 132 Collignon–Jourawski formula, 152
Compatibility method, see also Flexibility
method
B Compatibility tensor, 9
Barrel vaults, 70 Configuration, 2
Beam, 103 Reference, 2
curvature, 130 Constitutive equations, 14, 161
deep, see also Deep Beam Continuousd medium, 1
Euler–Bernoulli–Navier model, 127 Cremona diagram, 38
floor, 169 Cross method, see also Moment distribution
Frenet–Serret trihedron, 105 method
Gerber, 113 Cross section, 104, 105, 107, 127, 128, 149,
intrinsic coordinates, 105 159
Spandrel, 169 core, 138
Timoshenko model, 157
Vierendeel, 118
Beltrami–Michell form of the elastic prob- D
lem, 19 Dead loads, 46
Bending, 103 Deep beam, 157
Bending moment, 103 Deformation gradient, 4, 5, 15
Bridge Deformed configuration, 193
bowstring, 70, 77 Degree of hyperstaticity, 192
tied arch, 70 Degree of indeterminacy, 192
Buckling, 37 Direct stiffness method, 202
Displacement field, 6
Displacement gradient, 7
C Duhamel–Neumann’s law, 15
Cable beams, 47, 79
Cantilever, 71
Carryover factor, 201 E
Catenary, 45 Eccentricity, 138
Cauchy strain tensor, 7 Elastic section modulus, 135
Cauchy stress tensor, 12 Envelope, 63
234 Index
Equilibrium equations, 85, 115 L

Equilibrium method, see also Stiffness Limit analysis
method arches, 61
External static indeterminacy, 192 Load-displacement curve, 17
Lower bound theorem, 62, 103, 158
F
Field, 6 M
First area moment, 132, 152 Masonry structures, 60
First Piola–Kirchhoff stress tensor, 12, 14, Mechanics of materials, 20
15 Membrane, 47
Fixed-end moments, 198 Method of consistent deformations, see also
Flexibility method, 191, 194 Flexibility method
Force polygon, 37 Method of sections, 86, 96, 103, 108, 116
Foundation, 138 Moment distribution method, 200
Frame, 114 Movement, 2
three-hinged, 114 invertible, 2
Free body diagram, 37, 79 regular, 2
Funicular curve, 41, 43
Funicular polygon, 41, 103
N
G Navier form of the elastic problem, 19
Girder, 169 Neutral axis, 137
Green-Lagrange strain tensor, 16 Nominal stress tensor, 12
Groin vaults, 68
O
H Ordinary differential equation, 179
Hollow profile, 156
Hooke’s law, 17
Hypar, see also Hyperbolic paraboloids P
Hyperbolic paraboloids, 68 Pavilion vaults, 66, 68
Hyperstatic structures, 47, 61, 192 Permanent loads, 45
Hypothesis of small displacements, 2 Perturbations
Hypothesis of small strains, 7 arches, 61, 64
Polar decomposition, 7
Prandtl function, 176
I
Principle of Virtual Displacements, 89
Internal force, 33, 106
Principle of Virtual Forces, 89
axial, 34
Principle of Virtual Work, 87, 140
diagrams, 108, 111, 112
shear, 106
Internal moment
bending, 107 R
diagrams, 108 Radius of gyration, 139
torsion, 107 Rise, 33
twisting, 107 Rotation tensor, 7
Internal static indeterminacy, 192
Invariants of tensor, 12
Isostatic structures, 36, 192, 194 S
Saint Venant’s principle, 1
Saint Venant tensor, see also Compatibility
K tensor
Kronecker delta tensor, 7 Second Piola–Kirchhoff stress tensor, 12, 16
Index 235
Shape function, 207 Prandtl model, 169

Shear center, 155 rate of twist, 169
Shear flow, 154 Saint Venant model, 169
Slenderness ratio, 46 Torsional stiffness, 175, 185
Slope deflection method, 199 Torsion flow, 184
Small displacements, 2, 34 Traction, 37
Small strains, 7 Truss, 71, 75
Span, 33 Howe, 93
Specific weight, 13 lattice, 95
Spindle, 68 long, 94
Springer (of an arch), 61 Poncelet, 94
Stable structure, 192 Queen-post, 76
Statically determinate structure, 62, 80, 114, three-dimensional, 95
115
Statically indeterminate structure, 80, 84
Stiffness U
axial, 88 Unstable structure, 80, 192
Stiffness method, 191, 194, 195
Strength of materials, 20
Stress vector, 10 V
Strong form, 204 Vector of internal forces, 106
Strut, 71 Vector of internal moments, 106
Support, 37 Vierendeel beam, 118
pinned, 78 Virtual displacement, 89
roller, 78 Von Mises stress, 18, 153
Voussoir, 61
T
Thin-walled profiles, 153 W
Thrust Network Analysis (TNA), 68 Warping function, 177
Tie, 71 Weak form, 204
Timoshenko shear coefficient, 161
Torsion, 169
center of torsion, 171 Y
constant, 175 Yield stress, 45

An Introduction To Structural Mechanics For Architects

Uploaded by

Copyright:

Available Formats

An Introduction To Structural Mechanics For Architects

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

An Introduction To Structural Mechanics For Architects

Uploaded by

Copyright:

Available Formats

Structural Integrity 4

Series Editors: José A. F. O. Correia · Abílio M. P. De Jesus

More information about this series at http://www.springer.com/series/15775

ISSN 2522-560X ISSN 2522-5618 (electronic)

© Springer International Publishing AG, part of Springer Nature 2018

Printed on acid-free paper

Zaragoza, Spain Elías Cueto

1 An Introduction to Continuum Mechanics . . . . . . . . . . . . . . . . . . . . 1

2.5 Catenary Curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

6.2.4 Internal Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131

© Springer International Publishing AG, part of Springer Nature 2018 1

ϕ0 : Ω → ϕ0 (Ω) ⊂ R3 , ϕ0 (P) ≡ X, ϕ0i (P) = X i ,

ϕ(t) ≡ ϕt : Ω → ϕt (Ω) ⊂ R3 , ϕt (P) ≡ x, ϕti (P) = xi ,

1.1.2 Hypothesis of Small Displacements

In introductory courses in structural mechanics, it is common to accept that the

Once the movement of a solid has been characterized, it is straightforward to define

1.1.3 Deformation Gradient

the difference should not be the possibility of establishing a movement φ. What

It is important to note that, despite the hypothesis of small displacements, we

The magnitude defined in Eq. (1.1) is known as deformation gradient. It is impor-

1.1.4 Displacement Field

u(t, X) = φt (t, X) − X = x(t, X) − X.

In the simplest case of Cartesian coordinates, this reduces to:

u i = xi − X i , i = 1, 2, 3 or, more generally, i = x, y, z.

1.1.5 Strain Ensors

If we write the deformation gradient tensor as a function of the displacement field,

Fig. 1.3 Displacement x3

What we have done is to decompose the strain as the contribution of a rotation

The difference between E and ε is precisely the high-order term 21 (∇u · ∇u T ). If

These components have a clear physical interpretation. Those in the diagonal

ΔV (1 + εx )d x(1 + ε y )dy(1 + εz )dz − d x d y dz

1.1.6 Compatibility Equations

Fig. 1.4 Physical interpretation of the strain components: a longitudinal; b transversal

1.2.1 Stress Vector

Fig. 1.5 Stress vector

1.2.2 Stress Tensors

−txn d Sn + txx d Sx + txy d S y + txz d Sz = 0,

Fig. 1.6 Obtention of the Z

tin = σi j n j , or, simply, t = σn.

1.2.3 Principal Stresses

1.2.4 Equilibrium Equations

Fig. 1.7 Equilibrium at an elementary volume element

σ ji, j + ρbi = ρai , (1.6)

or, even simpler, in vector notation,

1.3 Constitutive Equations

σ(X, t) = σ(X, φt0 →t , Θ, t),

σ(X, t) = σ(X, ε, ε̇, Θ, t),

σ(X, t) = σ(X, ε, Θ, t).

For large strain applications, this can be simply stated as [3]

1.3.1 Linear Elastic Materials

By far, the most used materials in construction—particularly, metals—show linear

The most frequent form of linear elasticity is expressed, however, as:

σi j = Ci jkl εkl , (1.8)

The constitutive tensor C has four indices, i, j, k, l = 1, 2, 3. Therefore, the number

thus needing only 21 constants to be define C completely.

1.3.3 Hooke’s Law

1.3.4 Limits of Hooke’s Law

Fig. 1.8 Load-displacement 20

0 0.5 1 1.5 2 2.5 3