Angelillo Mechanics of Masonry Structures PDF

CISM International Centre for Mechanical Sciences 551
Courses and Lectures
Maurizio Angelillo
Editor
Mechanics of
Masonry Structures
International Centre
for Mechanical Sciences
CISM Courses and Lectures
Series Editors:
The Rectors
Friedrich Pfeiffer - Munich
Franz G. Rammerstorfer - Wien
Elisabeth Guazzelli - Marseille
The Secretary General

Bernhard Schrefler - Padua
Executive Editor
Paolo Serafini - Udine
The series presents lecture notes, monographs, edited works and

proceedings in the field of Mechanics, Engineering, Computer Science
and Applied Mathematics.
Purpose of the series is to make known in the international scientific
and technical community results obtained in some of the activities
organized by CISM, the International Centre for Mechanical Sciences.
International Centre for Mechanical Sciences
Courses and Lectures Vol. 551
For further volumes:

www.springer.com/series/76
Maurizio Angelillo
Editor
Mechanics of Masonry
Structures
Editor
Maurizio Angelillo
Università di Salerno, Fisciano, Italy
ISSN 0254-1971
ISBN 978-3-7091-1773-6 ISBN 978-3-7091-1774-3 (eBook)
DOI 10.1007/ 978-3-7091-1774-3
Springer Wien Heidelberg New York Dordrecht London
© CISM, Udine 2014
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. Ex-
empted from this legal reservation are brief excerpts in connection with reviews or scholarly
analysis or material supplied specifically for the purpose of being entered and executed on
a computer system, for exclusive use by the purchaser of the work. Duplication of this pub-
lication or parts thereof is permitted only under the provisions of the Copyright Law of the
Publisher’s location, in its current version, and permission for use must always be obtained
from Springer. Permissions for use may be obtained through RightsLink at the Copyright
Clearance Center. Violations are liable to prosecution under the respective Copyright Law.
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.
While the advice and information in this book are believed to be true and accurate at
the date of publication, neither the authors nor the editors nor the publisher can ac-
cept any legal responsibility for any errors or omissions that may be made. The publish-
er makes no warranty, express or implied, with respect to the material contained herein.
All contributions have been typeset by the authors

Printed in Italy
Printed on acid-free paper
Springer is part of Springer Science+Business Media (www.springer.com)

PREFACE
It is maybe a trivial remark saying that the vast majority of masonry

structures (excluding tall towers) exhibit an extraordinary stability
under the effect of age and settlements and even under the repeated
action of strong winds and heavy earthquakes. Someone may be skep-
tical about this statement since, lately, collapses of masonry structures
are not so infrequent. The point is that, almost invariably, failure is
caused by some unaware but diabolic alteration of the natural and
pacific equilibrium of masonry. The reason of this pacific stability
stems from the so called strength by shape that is typical also of other
structures carrying axial forces, such as a cord or a membrane, that is
unilateral structures. If pacific-unilateral stability for masonry, does
not mean plainly dumb or boring stability, this is actually due to the
curved structural elements (arches, vaults and domes) that started
to appear systematically in masonry Architecture since the ancient
Rome. The unilateral model, which appears as the clue of structural
interpretation behind the design of the great Architecture masterpieces
of the past, was first rationally introduced in the scientific community
by Heyman in 1966, with his mile-stone paper the stone skeleton.
Since then it has been the Italian school of Structural Mechanics to
carry the torch of the old masonry tradition, with the contribution of
a number of individuals dragged by the charismatic leaderships of Sal-
vatore Di Pasquale, and of whom Lucchesi, Šilhavý and myself are, in
some sense, modern followers. The fire of the unilateral model, still
burning in Naples in the late seventies, was poked by the unlucky event
of the Irpinia earthquake of 1980. At that time I was a young Archi-
tect working under the guidance of Giovanni Castellano (a friend and
former co-worker of Di Pasquale), and I had the occasion not only to
eyewitness the, sometimes turbulent, discussions on the No Tension
model for masonry, but also to see the model at work in the wounded
body of many masonry buildings and monuments of Naples and of its
battered neighbourhoods. But how comes that the unilateral model for
masonry, that has been part of the traditional scientific heritage since
Mery divulgated the thrust line approach of Moseley in 1840, had to
be rediscovered again (and with scant success) in the second half of
the twentieth century? Indeed, though the traditional unilateral ap-
proach to masonry equilibrium has had an outstanding mentor and
divulgator in the person of Jaques Heyman, who, after writing the
mentioned paper, in 1995 published a crystalline book with the same
inspiring title (a book in which the author succeeds in explaining the
stone behaviour to the stones themselves by using barely a few equa-
tions), it seems that the message of the traditional masonry design
has not been welcomed by the modern structural engineers. A reason
for this state of affairs is given by Santiago Huerta, in his paper by the
provoking title Galileo was wrong “... any engineer or architect with
some formation in structural theory feels more comfortable within the
frame of the strength approach of Galileo and the classical theory of
structures. It requires an effort, and some study, to overcome our own
prejudices and to accept that, for example, the medieval master ma-
sons, knowing nothing of mathematics, elastic theory and strength of
materials, had a deeper understanding of masonry architecture than
we engineers and architects of the twenty-first century do.”
The presentation given by Lucchesi, Šilhavý and myself in the first
part of this book represents a modern update of the unilateral model
for masonry and a step forward toward the goal of obtaining a useful
practical tool for the analysis of masonry structures.
Though we believe that the unilateral model can be useful to prac-
titioners and applied engineers, since it captures the essence of ma-
sonry mechanics, still the limits of such a crude model are apparent
and there are aspects of masonry behaviour that need to be understood
such as damage, degradation, friction, heterogeneity and particularly
the role of the interface behaviour in the overall response of masonry.
In order to appreciate the limits of validity of the simplified unilat-
eral approach, it is important to study and interpret the experimental
results with the “eyes” of more sophisticated models. Actually, all I
have said until now refers to the phenomenological modelling of old
masonry for which the assessment of the material properties in the
detail required by fancy models is virtually impossible. The case of
new masonries for which the nature of the blocks and of the mortar
and of their arrangement is known and reliable, is a complete different
story. A typical case is that of brick-works studied in the present book
by Lebon, Sacco and Lourenco & Milani. In the end, these new ma-
sonry structures are nothing else than composite structures to which
sophisticated techniques of homogenization can be applied. The the-
oretical and experimental study of these peculiar structures with this
more in depth focus, is not only useful for the closer simulation of
their mechanical behaviour, but can put light on the mechanical phe-
nomena that are behind the crude approximations of the Heyman’s
model, namely the unilateral and the no-sliding assumptions. Unilat-
erality is an extreme approximation for the brittleness of the mate-
rial under tensile loads, brittleness being responsible for the softening
behaviour of masonry at the macroscopic level. No-sliding is equiva-
lent to assume infinite friction, and friction and sliding are the basic
mechanisms in brick-brick, and brick-mortar-brick interactions. Un-
derstanding toughness and friction is then obviously a necessary step
toward the goal of obtaining a detailed masonry description. Anyone
working at some depth in material engineering knows that fracture
and friction are still the most difficult challenges of modern Mechan-
ics; the main strength of the simplified unilateral model of Heyman
which assumes zero toughness and infinite friction is indeed its abil-
ity, while excluding these two tough guys, of being still able to make
sound predictions on masonry behaviour.
CONTENTS
1. Masonry behaviour and modelling

by M. Angelillo, P. B. Lourenço and G. Milani . . . . . . . . . . . . . 1
Premise 1
Basic behaviour of masonry and simplified unilateral

models 2
Local failure modes . . . . . . . . . . . . . . . . . . . . 2
Structural failure mechanisms . . . . . . . . . . . . . . . 3
Experimental observations: results of typical tests . . . 6
Simplified uniaxial models . . . . . . . . . . . . . . . . . 8
Masonry behavior of regularly arranged masonry struc-

tures. Homogenization and refined models 13
Introduction . . . . . . . . . . . . . . . . . . . . . . . . 13
Mechanical Behaviour of Masonry, Observations and Nu-
merical Data . . . . . . . . . . . . . . . . . . . . . . . . 17
Example of application . . . . . . . . . . . . . . . . . . 23
Bibliography 24
I Simplified Models 27
2. Mathematics of the Masonry-Like model and Limit
Analysis
by M. Šilhavý . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
Introduction 29
Constitutive equations 31
Vector valued measures 38
Loads 40
The existence of equilibrium states 46
Limit analysis 53
Families of measures and the weak compatibility of

loads 58
Integration with gravity 62
Bibliography 67
3. A numerical method for solving BVP of masonry-like

solids
by M. Lucchesi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
Introduction 71
Explicit expression for the stress function for isotropic ma-
sonry-like materials and its derivative . . . . . . . . . . 73
The equilibrium problem and a numerical method for its
solution . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
Comparison between the explicit and numerical solutions 82
Example applications . . . . . . . . . . . . . . . . . . . 87
A numerical method for dynamic analysis of slender ma-
sonry structures . . . . . . . . . . . . . . . . . . . . . . 93
Bibliography 104
4. Practical applications of unilateral models to Ma-

sonry Equilibrium
by M. Angelillo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
Basic tools 109

Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . 110
Equilibrated stress fields, regularity of T . . . . . . . . 110
Compatible displacement fields, regularity of u . . . . . 112
Dirac deltas: a familiar example . . . . . . . . . . . . . 113
Singular stress and strain as line Dirac deltas . . . . . . 115
Model zero (RNT) 118

Constitutive restrictions and equilibrium problem . . . 119
Statically admissible stress fields . . . . . . . . . . . . . 120
Fundamental partition . . . . . . . . . . . . . . . . . . . 121
Concavity of the Airy’s stress function . . . . . . . . . . 122
Kinematically admissible displacement fields . . . . . . 124
Compatibility of loads and distortions . . . . . . . . . . 126
Incompatibility of loads and distortions . . . . . . . . . 128
Limit Analysis . . . . . . . . . . . . . . . . . . . . . . . 130
Simple applications of the theorems of Limit Analysis . 132
Derand’s rule . . . . . . . . . . . . . . . . . . . . . . . . 153
Model one (NENT) 157

A Premise on Minimum problems and the peculiarity of
NT materials . . . . . . . . . . . . . . . . . . . . . . . . 157
The Boundary Value Problem for NENT materials . . . 158
Strain energy density in 2d . . . . . . . . . . . . . . . . 161
Function spaces for the potential energy functional. . . 162
Complementary energy functional. . . . . . . . . . . . . 164
Examples of non-existence. . . . . . . . . . . . . . . . . 165
Elementary analytical solutions . . . . . . . . . . . . . . 165
Masonry-like panels under flexure, shear and compression:
Mansfield-Fortunato semi-analytical solutions . . . . . . 176
Model two (ML) 187

The equilibrium problem for ML materials . . . . . . . 188
Numerical minimization strategy . . . . . . . . . . . . . 191
Numerical examples . . . . . . . . . . . . . . . . . . . . 196
Bibliography 207
II Refined Models 211

5. Modeling the interfaces in masonry structures
by F. Lebon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
Introduction 213
Some comments on experimental results 216

Brick-brick interactions: dry friction . . . . . . . . . . . 216
Brick-mortar interactions . . . . . . . . . . . . . . . . . 217
Masonry structures without mortar: dry friction mod-

eling (model 1) 219
Unilateral contact . . . . . . . . . . . . . . . . . . . . . 220
Dry Friction . . . . . . . . . . . . . . . . . . . . . . . . 220
Formulations . . . . . . . . . . . . . . . . . . . . . . . . 220
Masonry blocks with mortar 226

A phenomenological model taking the adhesion into ac-
count (model 2) . . . . . . . . . . . . . . . . . . . . . . 226
Deductive models: linear (and non linear) multi-scale mod-
els . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
Conclusion 237
Bibliography 237
6. Micro, Multiscale and Macro Models for Masonry
Structures
by E. Sacco . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
Introduction 241
Micromechanical modeling 242

Interface model . . . . . . . . . . . . . . . . . . . . . . 245
Numerical applications . . . . . . . . . . . . . . . . . . 249
Multiscale modeling 255

Macro-level model . . . . . . . . . . . . . . . . . . . . . 259
Micro-level model . . . . . . . . . . . . . . . . . . . . . 260
Homogenization technique . . . . . . . . . . . . . . . . . 263
Macromechanical modeling 272

Plastic-nonlocal damage model . . . . . . . . . . . . . . 274
Plastic model . . . . . . . . . . . . . . . . . . . . . . . . 278
Plastic-damage rate problem . . . . . . . . . . . . . . . 278
Conclusions 284
Bibliography 286
7. Homogenization and Seismic Assessment: Review

and Recent Trends
by P. B. Lourenço and G. Milani . . . . . . . . . . . . . . . . . . . . . . . . . . 293
Homogenization Theory, Basic Assumptions 293

Closed-form and simplified solutions in the linear
elastic range . . . . . . . . . . . . . . . . . . . . . . . . 296
Simplified stress assumed FE discretization . . . . . . . 301
Linear elastic case . . . . . . . . . . . . . . . . . . . . . 304
Constant Stress Triangular elements discretization of the
unit cell . . . . . . . . . . . . . . . . . . . . . . . . . . . 306
Homogenized failure surfaces . . . . . . . . . . . . . . . 309
Out-of-plane homogenized failure surfaces . . . . . . . . 322
Structural Level 323

In-plane loads: shear wall with central opening . . . . . 324
Out-of-plane loads: windowed panel in two-way bending 325
Dynamics and Seismic Behaviour 327

Masonry structures with box behaviour . . . . . . . . . 328
Masonry structures without box behaviour . . . . . . . 331
Conclusions 335
Bibliography 336
Masonry behaviour and modelling
Maurizio Angelillo‡ , Paulo B. Lourenço *

and Gabriele Milani †
‡
Department of Civil Eng., University of Salerno, Italy
*
Institute for Sustainability and Innovation in Structural Engineering (ISISE),
Department of Civil Engineering, University of Minho, Portugal
†
Department of Architecture, Built environment and Construction engineering
(A.B.C.), Politecnico di Milano, Italy
Abstract In this Chapter we present the basic experimental facts

on masonry materials and introduce simple and refined models for
masonry. The simple models are essentially macroscopic and based
on the assumption that the material is incapable of sustaining ten-
sile loads (No-Tension assumption). The refined models account
for the microscopic structure of masonry, modeling the interaction
between the blocks and the interfaces.
1 Premise
The first basic question that any course on Masonry Structures should ad-
dress is: what we consider as masonry material?
Masonry structures can be built with a large variety of materials, ma-
sonry blocks can be of different types and assembled in many different ways;
mortar, if present, can also be of various kinds, and the way it interacts with
the blocks depends on workmanship. There is old masonry, new masonry
and a peculiar place is taken by brickworks.
There are essentially two ways of approaching the modelling of masonry:
the first one is rather ambitious and aims at the modelling of large classes
of masonry buildings (e.g. old masonry structures). The second one is
more pragmatic and restricts to the mechanical description of very specific
types of masonry (masonry structures of regularly arranged blocks, e.g.
brickworks of known geometry). Here Silhavi, Lucchesi and myself adopt
the first approach and Sacco, Lebon, and Lourenco & Milani propose the
second one (also if Sacco has had experiences and papers where the first
approach was considered).
M. Angelillo (Ed.), Mechanics of Masonry Structures, CISM International Centre for

Mechanical Sciences, DOI 10.1007/ 978-3-7091-1774-3_1, © CISM, Udine 2014
2 M. Angelillo, P. B. Lourenço and G. Milani
It is evident that with the second approach the models adopted can
be very sophisticated and more close to reality, whilst the first approach
asks for very crude material assumptions and produces predictions on real
constructions that are affected by large approximations. The point is that,
often, the real geometry and material behaviour of the building is not known
in the detail required by the second approach, the definition of even the
most primitive material parameters, such as strength and stiffness, being
generally difficult and affected by an elevated randomness and uncertainty.
The most basic assumption that can be made, in view of the small and
often erratic value of the tensile strength of masonry materials, is that the
material behaves unilaterally, that is only compressive stresses can be trans-
mitted (No-Tension assumption). It is generally recognized (since the pio-
neering work of Heyman (1966)) that such an assumption is the first clue
for the interpretation of masonry behaviour; on adopting and applying it,
we acquire the eyes to appreciate and interpret the fracture patterns, that is
the masonry most peculiar manifestation, representing, in a sense, its breath
(that is the way in which the masonry buildings relieve and can survive also
to radical and, sometimes, dramatic changes of the environment).
We call the models based on the No-Tension assumption simple models
and the models accounting for more sophisticate stress-strain laws (i.e. ex-
hibiting damage, softening, brittleness) or based on the micro/meso-scopic
structure of the material, refined models. The book is divided into two in-
terconnected but separate parts: Part I, where the simplified models are
studied, Part II where the refined models are described.
In the present Chapter we discuss the basic experimental facts on ma-
sonry materials justifying the introduction of the simple and refined models
for masonry.
2 Basic behaviour of masonry and simplified

unilateral models
M. Angelillo
2.1 Local failure modes

There are basically three failure modes that are visible locally in masonry
structures.
1. The first one is the one associated to the brittleness of the material
and that manifests itself with detachment fractures, such as those
reported in Figure 1. Such fractures consist in cracks that usually
separate neatly two parts of seemingly intact material and are usually
Masonry Behaviour and Modelling 3
the “good” ones, that is those contributing to the accommodation and

release of stress.
Figure 1. Fracture of detachment in brick walls at different scales
2. The second one is a kind of mixed mode in which fractures of de-

tachment alternate to lines of sliding, such as those appearing in the
examples of in-plane shear shown in Figure 2. This mode of failure
presents usually itself in walls subjected to high compressive loads and
shears.
Figure 2. Detachment and sliding due to combined compression and shear.
3. The third failure mode is the so–called crushing of the material (Fig-
ure 3) and occurs essentially under compression. By looking closely to
this failure mode one can see again that it consists of finer detachment
fractures, close together and separated by damaged material, having
sometimes the consistence of powder.
The first type of fractures is the most frequent and usually irrelevant.
The second and third modes often occur when the load is critical or close to
become a collapse load. The third one is the most dangerous since failure
under compression is usually sudden.
2.2 Structural failure mechanisms

Besides crushing of compressed members, such as those shown in Fig-
ure 3, there are basically other two structural failure mechanisms through
Figure 3. Crushing due to compression.
which a masonry structure (or a part of it) may collapse. The most fre-
quent one, under seismic loads, is out of plane rocking as shown in Figure 4.
Such a mechanism can be due to the effect of the self load solely, or can be
favoured by the pushing of the roof, or the hammering of a heavy floor or
ceiling.
Figure 4. Out of plane rocking.
Both crushing failure and out of plane rocking are usually the result
of a poor design, or of unwise modifications of the original construction.
To avoid out of plane rocking many regulations prescribe the maximum
distance between two consecutive transverse walls. The demolition of such
transverse walls is one of the most common examples of risky modifications.
The third failure mechanism, that is in-plane shear, is the one proper
of well designed buildings, that is structures sustaining the horizontal ac-
tions through the harmonized cooperation of the shear resistant structures
(Figure 5), i.e. with local failure modes of their masonry units in their own
planes, of the type shown in Figure 2.
Figure 5. In plane failure due to compression and shear.
Figure 6. Incipient failure of a shallow cross vault due to excessive spread of the
abutments (aisle vaults of a XVI century chapel in Nocera Inferiore (SA), Italy). Courtesy
of Enrico Sicignano.
The reader must be warned that also the detachment fractures, due to
settlement or accommodation of the structure to new loads or to environ-
mental changes, usually physiological and not necessarily entailing an immi-
nent danger, may be the precursors of an incipient collapse: if the amplitude
of the displacements and the size of the cracks become comparatively large
with respect to the overall size of the structure, they could allow for the
structure to become unstable. As an example, an arch or a vault may be
perfectly comfortable under the action of their own weight, in a fractured
and heavily distorted configuration, until the displacements grow to be so
large that a mechanism becomes possible and their overall stability is sud-
denly lost. An example of such extreme conditions is reported in Figure 6.
The message is that detachment fractures are usually unwary, but it is wise
checking the size of cracks and displacements and their evolution in time.
2.3 Experimental observations: results of typical tests
(a) (b)
Figure 7. Typical stress-displacement plot for a masonry material. (a) compression,
(b) tension. In (b) the shaded area labelled G is the toughness, that is the surface energy
per unit area that must be expended to open a crack. The plots are only qualitative;
for any kind of masonry, usually, the values of σc and σt differ of at least one order of
magnitude.
Old master masons could perfectly build without the help of tests on
the construction materials. Nowadays the situation, compared to ancient
times (when the predominance of metaphysics on physics was absolute, see
Benvenuto (1991)), is entirely different and no work on buildings can proceed
in absence of an experimental assessment on material performances.
Masonry is a composite material and accurate tests can be performed
on the component materials, on masonry assemblies and on small masonry
structures. The detailed description of some of these tests will be touched
elsewhere in this book (see Sect.3, Ch.1), here I consider the results of
typical tests on small masonry walls, in order to extract the main aspects
(a) (b)
Figure 8. Simple compression tests on masonry walls made of tuff blocks. (a) test set
up, (b) stress strain plots corresponding to four different tests on walls having identical
geometry and material components. Courtesy of A. Prota et al. (2008).
of masonry behaviour that can be captured by the simple models.

In Figure 7 the typical aspect of the stress-displacement plots corre-
sponding to simple compression and tension load tests on masonry walls
are reported. The strength of the wall depends on the strength of the in-
dividual components (masonry units/blocks and mortar): some regulations
give formulas to relate the compressive strength of masonry to the com-
pressive strength of the components, also if the reliability of such rules is
questionable on many grounds.
As the example reported in Figure 8, relative to compression tests on
tuff masonry walls (50 cm × 60 cm × 20 cm) made of the same kind of
stones and mortar, show, the results are affected by a strong variance, both
for strength and stiffness. Such a state of affairs is even worse in the case
of tension where a reliable value of strength is hardly identifiable with the
standard statistical methods. Therefore we point out that the graphs of
Figure 7 have to be interpreted as the uniaxial stress-displacement plots of
a highly idealized masonry material.
The main feature of masonry materials is that the tensile strength σt
is much lower than the compressive strength σc : the ratio σt /σc is usually
lower than 0.1 and can be as low as 0.01 or even, locally, vanishingly small.
Masonry behaves essentially as an elastic material in compression up
to 80-90 % of the strength, also if due to early microcracking and damage
the stress-strain plot is definitely nonlinear. In the post critical phase the
material undergoes irreversible deformations, showing a sort of plastic be-
haviour. Therefore the stress-strain plot of a typical masonry-like bar can
Masonry Type σc ( daN

cm2 ) τs ( daN
cm2 ) E( daN
cm2 ) ρ( Kg
m3 )
Disarranged masonry 10-18 0.20-0.32 6900-10500 1900
of cobbles/boulders
Masonry of 20-30 0.35-0.51 10200-14400 2000
roughhewed stones
Masonry of 26-38 0.56-0.74 15000-19800 2100
cut stones
Masonry of 14-24 0.28-0.42 9000-12600 1600
soft stones
Masonry of 60-80 0.90-1.20 24000-32000 2200
squared stone blocks
Brickwork of solid 24-40 0.60-0.92 12000-18000 1800
bloks and lime mortar
Brickwork of semisolid 50-80 2.40-3.20 35000-56000 1500
blocks and cem. mortar
Brickwork of 40-60 3.00-4.00 36000-54000 1200
air bricks (45 %)
Brickwork of 30-40 1.00-1.30 27000-36000 1100
air bricks (< 45 %)
Masonry of concrete 30-44 1.80-2.40 24000-35200 1400
air-blocks (45-65 %)
Masonry of concrete 15-20 0.95-1.25 12000-16000 1200
air-blocks (< 45%)
Table 1. Densities and reference mechanical strength and stiffness of different types of
coarse masonry with poor mortar. Strength in compression and shear: σc , τs , E: Young
modulus, ρ: density. Source: Italian Code for Constructions (DM 14.1.2008).
be represented by the graph of Figure 9a.

For book keeping the reference strength and stiffness of some common
masonry materials are listed in Table 1. Notice that σc is the compressive
strength and that the shear strength τs is roughly correlated to the tensile
strength σt through the relation τs ≈ σt /2.
2.4 Simplified uniaxial models

In Part I of the present book, basically three simplified models for ide-
alizing the uniaxial masonry-like behaviour will be used. I call them model
zero, one and two, these names coming from the number of parameters that
are required for their definition. The order in which the models are put, is
an order of gradual improvement.
We must observe explicitly that the first model is not, as one usually
expects for any canonical solid, the linear elastic material. There is of
course the possibility of modelling masonry structures as linearly elastic:
masonry material is definitely an elastic brittle material for very small stress
and strains, but the point is that the levels of stress and strains at which
masonry materials work in real structures, are usually higher.
Model zero (RNT). As a first approximation to the behaviour of
Figure 9a, the Rigid No-Tension material has been proposed (Figure 9b).
This crude unilateral model that describes the material as indefinitely strong
and stiff in compression but incapable of sustaining tensile stresses, was
first rationally introduced by Heyman (1966) and divulgated and extended
in Italy thanks to the effort of Salvatore Di Pasquale (1984) and other
distinguished members of the Italian school of Structural Mechanics (see
below for an extensive reference). This material is rigid in compression and
can elongate freely, a positive deformation of the bar being interpreted as
a measure of fracture into the material (either smeared or concentrated).
It must be observed that, though the material has a limited repertoire of
admissible stresses and strains and exhibits fractures, its uniaxial behaviour
in elongation is elastic.
a b c d
Figure 9. Typical uniaxial behaviour, (a), and simplified models: (b) model zero, (c)
model one, (d) model two.
This statement, that may appear “paradoxical”, derives from the primi-
tive definition of elasticity: stress determined by strain, and the stress has
actually a definite value (zero) if the bar elongates. The behaviour is in-
deed perfectly reversible in elongation, also if deformation occurs without
accumulation of elastic energy. We may also notice that there is a degen-
erate elastic energy associated to this material (Φ = 0 if is non negative,
Φ = +∞ otherwise) and that the constraint on strain makes the compres-
sive stress σ a sort of reaction to this constraint, that is negative values of σ
are non-constitutive (in the same way in which pressure is non-constitutive
in incompressible materials). This model requires no material parameters

since strength and stiffness in compression are assumed to be infinite whilst
they are completely neglected in tension.
Model one (ENT). As a first step into the closer modelling of ma-
sonry behaviour one can consider to add a finite stiffness in compression: a
linear ratio between stress and strain in compression is assumed. Another
possibility would have been to add a limited strength in compression still
assuming an infinite stiffness (rigid-plastic material), but in the vast major-
ity of applications, the first ingredient to add appears to be elasticity since
it allows to study the behaviour of the structure under working conditions,
before collapse takes place. Most masonry constructions are indeed in a
state of average stress that is well below the crushing strength, but the level
of stress is such to trigger not only microscopic but also macroscopic brittle
fracture.
Model one, described for a 1d bar by the plot of Figure 9c, is the so
called Elastic No-Tension material, thoroughly studied since the late 70ties
by the Italian school of structural mechanics (Romano and Romano (1979),
Baratta and Toscano (1982), Como and Grimaldi (1985), Romano and Sacco
(1985), Castellano (1988), Del Piero (1989), Angelillo (1993)).
a. b. c.
Figure 10. In (a) elastic energy for model one, in (b) elastic energy for NT hyperelastic
model with threshold. (c) elastic energy for model two.
Now the strain can be positive or negative, positive strain being the
fracture part of deformation and negative strain the elastic part. The ENT
material is globally elastic, in the sense that strain determines stress for any
value of strain: now compressive stress is constitutive. The material is even
hyperelastic since there exists a stored elastic energy density Φ such that σ
is the derivative of Φ with respect to . The elastic energy corresponding
to model one is depicted in Figure 10a. Notice that such energy, though
convex, is not strictly convex (major source of mathematical and numerical
troubles, see (Giaquinta and Giusti, 1985)).
This model requires only one material parameter: the elastic modulus
E, since strength in compression is assumed to be infinite, whilst strength

and stiffness are completely neglected in tension.
Model two (ML). A further step can be taken by adding to model
one the assumption of a limited strength σc in compression. In this way
the failure modes due to crushing (such as those shown in Figure 3, but
also, in a 2-3d formulation, the failure modes and mechanisms represented
in Figures 2, 5), can be modelled. Here we shall refer to model two, whose
uniaxial behaviour is reported in Figure 9d, as the Masonry-Like material.
Any attempt to enrich this model for real applications is usually frustrated
by the lack of sufficient confidence on the material properties of the real
materials and of their assemblages.
As depicted in Figure 9d, it is assumed that the material behaves as
perfectly plastic in compression, therefore the constitutive response becomes
incremental and the actual stress state is path dependent, being determined
by the whole strain history. In Figure 10b the elastic energy of a path in-
dependent material having the same stress-strain plot of model two under
monotone loading (that is a No-Tension hyperelastic material with thresh-
old) is shown.
The incremental model is obviously not hyperelastic, though a sort of
“path dependent” elastic energy can still be defined (as shown in Figure 10c
for a special loading-unloading cycle). Now the anelastic part of deformation
is further decomposed into a reversible fracture part and in a irreversible
crushing part. We have to remark that this is a peculiar perfectly plastic
material, since, due to the different behaviour in tension (elastic fracture)
and compression (incremental plasticity), the plastic deformations cannot
be cancelled by reversing the strain. This model requires the setting of two
material parameters: the elastic modulus E and the strength in compression
σc , strength and stiffness being still completely neglected in tension.
Extension to 3d.For real applications the three simplified models need
to be extended to 3d. The NT hypothesis translate in 3d into an assump-
tion that the stress T belongs to the cone Sym− of negative semidefinite
symmetric tensors.
The next step is to introduce convenient rules for the latent part of the
deformation, that is for the strain sustaining the unilateral constraint on the
stress. The usual simplifying assumption is that there is no sliding along
the fracture lines, that is the total fracture strain satisfy a law of normality
with respect to the cone Sym− of admissible stresses (equivalent to require
that the stress do no work for the anelastic strain and that the anelastic
strain is positive semidefinite).
The law of normality allows for the simple application of the theorems
of limit analysis in 3d: the static and the kinematic ones.
For what concerns model one the linear elastic assumption in compres-
sion is easily generalized if one restricts to isotropy; in such a case the
definition of a further parameter: the Poisson ratio ν (a value that is diffi-
cult to be assessed through tests and is usually set between 0.1 and 0.2) is
required.
The law of normality and the usual restrictions considered on the elastic
constants, make the global response of model one in 3d elastic, and even
hyperelastic (see Del Piero (1989)), that is path independent. Also for model
two the restriction to isotropy simplify things, but now, besides the material
parameter σc , it is required to define a material function f, that is the limit
surface in compression.
The flow rule for the increments of crushing strain must be also intro-
duced; for simplicity one can choose to adopt an associated flow rule, also if
the frictional nature of sliding under compression would require the adop-
tion of a non-associated law.
Remark 1. Though the RNT model (model zero) appears as rather
rudimentary (notice that it allows only to apply the theorems of limit analy-
sis), it is the opinion of many master masons (among which Heyman (1995),
Como (2010), and Huerta (2006)) that the RNT model is the only choice
for old masonry constructions, the main motivation to this opinion being
the impossibility for the elastic models to define correctly the initial state of
the structure, due to the uncertainties about the boundary conditions and
on the previous history.
Any elastic solution, that, in principle, would allow the definition of the
stress state under the given loads, is indeed extraordinarily sensitive to very
small variations of the boundary conditions, particularly to the change of
the given boundary constraints such as those produced by the unknown
settlements of the foundation. The idea for applying the elastic assumption
(models one and two) is that the given settlements have been accommodated
by means of a small displacement mechanism, that is a kind of rigid body
relative displacement of some parts of the structure. The stress produced
during the nucleation and growth of the fracture necessary to activate the
mechanism are, in the end, almost completely released, and the final state
that we see, can be used as an essentially stress free reference state. In the
analysis of a real masonry artefact, it is up to the sensitivity of the analyzer,
based on the signs that the structure exhibits, to judge if the construction
seats or not in this comfortable state.
Remark 2. We have given reasons for adopting the simplified models,
but it is obvious that doing so we forget about many mechanical properties
of real masonry materials. Such properties can have a more or less fun-
damental role in the modelling and it is important to know them to fully
appreciate the limits of validity of the simplified models. They are recorded
in a “subjective” order of importance in the following list.
• There are signs of damage since the early stages of loading. Effect:
non linear σ − plot and decline of stiffness.
• The behaviour in tension is brittle. Effect: energy is expended to open
a crack.
• The anelastic strains and strain rates are not purely normal, sliding
is ruled by friction.
• Ductility, if any, is rather limited.
• The elastic and anelastic response is anisotropic.
• The anelastic behaviour, initially of hardening type, turns soon into
softening.
• The cyclic response is hysteretic and the stress-strain plots depends
on the rates: viscoelastic behaviour is implied.
• The material shows signs of degradation (strength reduction) under
long term loading.
• Sometimes the displacements are very large and demand for theories
accounting for geometric nonlinearities.
Nowadays there would not be any difficulties, either analytical or numer-
ical, to include all this effects (except brittleness, softening and friction) into
a FEM simulation of the masonry structure. The point is that, as already
remarked, the knowledge of these fine properties of masonry materials is
often lacking, especially for old masonry.
3 Masonry behavior of regularly arranged masonry

structures. Homogenization and refined models
P.B. Lourenço and G. Milani
3.1 Introduction
Masonry is a heterogeneous assemblage of units and joints. Units are
such as bricks, blocks, ashlars, adobes, irregular stones and others. Mortar
can be clay, bitumen, chalk, lime/cement based mortar, glue or other. The
huge number of possible combinations generated by the geometry, nature
and arrangement of units as well as the characteristics of mortars raises
doubts about the accuracy of the term “masonry”. Still, much information
can be gained from the study of regular masonry structures, in which a peri-
odic repetition of the microstructure occurs due to a constant arrangement
of the units (or constant bond).
The difficulties in performing advanced testing of this type of structures

are quite large due to the innumerable variations of masonry, the large scat-
ter of in situ material properties and the impossibility of reproducing it all in
a specimen. Therefore, most of the advanced experimental research carried
out in the last decades concentrated in brick block masonry and its relevance
for design. Accurate modelling requires a comprehensive experimental de-
scription of the material, which seems mostly available at the present state
of knowledge (Rots, 1997; Lourenço, 1998a). The behaviour of masonry is
much dependent not only on the composition of units and joints, but also
on how they are arranged and treated. Figure 11 shows two results of a
set of tests on dry stone masonry joints under shear testing. The adopted
stone is granite and the stone has been polished, sawn or artificially made
rough with a random impact of a spike. It is clear that the same surface
with different treatments has different capacities, with an initial tangent of
the friction angle for the polished stone joint (equal to 0.18) much lower
than the other surface treatments (about 0.6). The situation is even more
dramatic for the initial dilatancy angle, which is about zero for the polished
and sawn joints, but provides a dramatic volume loss for each load reversal
due to compaction of the joints. This phenomenon contributes to interlock-
ing loss between irregular masonry units and masonry out of plane collapse
in case of an earthquake. Another interesting example on how the ma-
sonry meso-structure (i.e. the arrangement of units, usually referred to as
micro-scale in masonry) influences the response is given in Vasconcelos and
Lourenço (2009). Here, three different types of stone masonry walls with
the same external geometry are tested under in plane cyclic shear, namely
regular dry stone masonry, irregular mortared joints masonry and rubble
masonry. Not only the strength and stiffness degradation of the walls is
rather different but also the strength envelop found is much different, with
a tangent of the friction angle varying between 0.4 (for dry stone masonry),
0.3 (for irregular masonry) and 0.2 (for rubble masonry).
The examples given demonstrate the relevance of the micro-structure of
masonry for the structural response. Therefore, the global field of structural
analysis of masonry structures encompasses several different approaches and
a comprehensive review is given in Lourenço (2002), with a recent update
for seismic analysis in Marques and Lourenço (2011) for masonry with box
behaviour and in Lourenço et al. (2011) for masonry without box behaviour.
Depending on the level of accuracy and the simplicity desired, usually the
following representations are possible: (a) micro-modelling, where the ge-
ometry of units and joints is directly considered and the constitutive laws
are obtained experimentally; (b) macro-modelling, where units and joints
are smeared out in the continuum and the constitutive laws are obtained
P series 1.50
S series
W = -0.618V R series 1.25
r2 = 0.975
Shear stress (N/mm )

Linear (P series)
2
Linear (S series) 1.00
W = -0.561V Linear (R series)
r2 = 0.985
0.75
0.50
W = -0.179V
r2 = 0.866 0.25
0.00
-2.00 -1.50 -1.00 -0.50 0.00
Normal stress (N/mm2)
(a)
0
-1.5 -1 -0.5 0 0.5 1 1.5
-0.05
-0.1
-0.15
-0.2
-0.25
-0.3
-0.35
-0.4
(b)
Figure 11. Testing of dry stone masonry joints under cyclic shearing,
Lourenço and Ramos (2004): (a) Coulomb friction law for polished “P”,
sawn “S” and rough “R” surfaces; (b) dilatancy of the joints for “R” surface
under reversed cycles, with the horizontal and vertical displacements of the
joints in the respective axes, in mm.
experimentally; (c) homogenization, where the micro-structure is handled

mathematically in terms of geometry and material data to obtain a smeared
continuum model; (d) structural component models, where constitutive laws
of structural elements are directly provided in terms of internal forces such
as shear force or bending moment (and related generalized displacements),
instead of stresses and strains, see Figure 12. The present chapter focuses on
masonry behaviour and numerical data, on the analysis of masonry struc-
tures making use of homogenization techniques, which has been receiving
a growing interest from the scientific community, see also Lourenço et al.
(2007), and on the seismic analysis of masonry structures.
Mortar Unit Interface “Unit”

Unit/Mortar “Joint” Composite
(a) (b) (c)
(d)
(e)
Figure 12. Modelling approaches for masonry: (a) representation of reg-

ular staggered or running bond masonry; (b) micro-modelling; (c) macro-
modelling; (d) homogenisation; (e) illustrative structural component mod-
els, with beam elements or macro-blocks.
3.2 Mechanical Behaviour of Masonry, Observations and Numer-

ical Data
A basic modern notion in the mechanical behaviour of masonry is soften-
ing, which is a gradual decrease of mechanical resistance under a continuous
increase of deformation forced upon a material specimen or structure, see
Figure 13. Softening is a salient feature of brick, mortar, stone or concrete,
which fail due to a process of progressive internal crack growth. For tensile
failure this phenomenon has been well identified. For shear failure, a soften-
ing process is also observed, associated with degradation of the cohesion in
Coulomb friction models. For compressive failure, experimental data seems
to indicate that both local and continuum fracturing processes govern the
behaviour.
ft V
Gf
V
G
(a)
V
fc
V
Gc
G
(b)
Figure 13. Softening and the definition of fracture energy: (a) tension;
(b) compression. Here, ft equals the tensile strength, fc equals the com-
pressive strength, Gf equals the tensile fracture energy and Gc equals the
compressive fracture energy. It is noted that the shape of the non-linear
response is also considered a parameter controlling the structural response.
Nevertheless, for engineering applications, this seems less relevant than the
other parameters.
Properties of unit and mortar The properties of masonry are strongly

dependent upon the properties of its constituents. Compressive strength
tests are easy to perform and give a good indication of the general quality
of the materials used, but very few results exist in the uniaxial post-peak
behaviour of compressed bricks and blocks, or mortar. The values pro-
posed for concrete in the Model Code 90 (CEB-FIP, 1993) are a peak strain
of 0.2% and a compressive fracture energy given by a parabolic best fit
Gf,c = 15 + 0.43fc − 0.0036fc2 , with fc in N/mm2 and Gf,c in N/mm This
curve is only applicable for compressive strength value of fc between 12 and
80 N/mm2 . The average ductility index in compression du,c (ratio between
fracture energy and strength) is 0.68 mm, even if this value changes signif-
icantly and it is recommended to use the above expression. For fc values
lower than 12 N/mm2 , a du,c value equal to 1.6 mm is suggested and for fc
values higher than 80 N/mm2 , a du,c value equal to 0.33 mm is suggested.
These are the limits obtained from Model Code 90.
It is difficult to relate the tensile strength of the masonry unit to its
compressive strength due to the different shapes, materials, manufacture
processes and volume of perforations. For the longitudinal tensile strength
of clay, calcium-silicate and concrete units, Schubert (1988) carried out an
extensive testing program and obtained a ratio between the tensile and
compressive strength that ranges from 0.03 to 0.10.
Extensive information on the tensile strength of masonry units and their
fracture energy is available, see e.g. van der Pluijm (1999), Lourenço et al.
(2005) and Vasconcelos et al. (2008). The ductility index du , given by the
ratio between the fracture energy Gf and the tensile strength ft , found for
brick was between 0.018 and 0.040 mm. The recommended ductility index
du , in the absence of more information is the average, 0.029 mm.
For stone granites, a non-linear relation given by du = 0.239ft−1.138 was
proposed in Vasconcelos et al. (2008), with du in mm and ft in N/mm2 . For
an average granite tensile strength value of 3.5 N/mm2 , the du value reads
0.057 mm, which is the double of the suggested value for brick, possibly due
to the larger grain structure of granite when compared to clay.
Finally, Model Code 90 (CEB-FIP, 1993) recommends for concrete (max-
imum aggregate size 8 mm), the value of Gf = 0.025(fc /10)0.7 , with Gf in
N/mm and the compressive strength fc in N/mm2 . Assuming that the re-
lation between tensile and compressive strength is about 5%, the following
expression is obtained Gf = 0.04ft0.7 .
For the mortar, standard test specimens are cast in steel moulds and the
water absorption effect of the unit is ignored, being usually not represen-
tative of the mortar inside the composite. Moreover, the interface between
mortar and unit controls the behaviour of the joint in a large extent. For the
tensile fracture energy of mortar, very few results are available, see Paulo-
Pereira (2012), and the average value of ductility found for different mortar
compositions is 0.065 mm.
Properties of the interface The research on masonry has been scarce

when compared with other structural materials and experimental data which
can be used as input for advanced non-linear models is limited. The bond
between the unit and mortar is often the weakest link in masonry assem-
blages. The non-linear response of the joints, which is then controlled by
the unit-mortar interface, is one of the most relevant features of masonry
behaviour. Two different phenomena occur in the unit-mortar interface, one
associated with tensile failure (mode I) and the other associated with shear
failure (mode II).
Different test set-ups have been used for the characterisation of the ten-
sile behaviour of the unit-mortar interface. These include (three-point, four-
point, bond-wrench) flexural testing, diametral compression (splitting test)
and direct tension testing. For the purpose of numerical simulation, direct
tension tests is the one to be adopted as it allows for the complete repre-
sentation of the stress-displacement diagram and yields the correct strength
value, see Figure 14.
V
V
V
Crack displacement
(a) (b)
Figure 14. Tensile bond behaviour of masonry: (a) test specimen (direct
tension); (b) typical experimental stress-crack displacement results for solid
clay brick masonry (the shaded area represents the envelope of four tests),
van der Pluijm (1999).
The parameters needed for the tensile mode (Mode I) are similar to the
previous section, namely the bond tensile strength ft and the bond fracture
energy Gf . The factors that affect the bond between unit and mortar are
highly dependent on the units (material, strength, perforation, size, air

dried or pre-wetted, etc.), on the mortar (composition, water contents, etc.)
and on workmanship (proper filling of the joints, vertical loading, etc.). A
recommendation for the value of the bond tensile strength based on the unit
type or mortar type is impossible, but an indication is given in Eurocode 6
(CEN, 2005) for the characteristic value (95% fractile), in the range of 0.1
to 0.4 N/mm2 . The value for Mode I fracture energy Gf,I found for different
combinations of unit and mortar was between 0.005 mm and 0.035 mm, van
der Pluijm (1999). This value is not dependent on the bond strength and
the recommended fracture energy is the average value of 0.012 N/mm, in
the absence of more information.
An important aspect in the determination of the shear response of ma-
sonry joints is the ability of the test set-up to generate a uniform state
of stress in the joints. This objective is difficult because the equilibrium
constraints introduce non-uniform normal and shear stresses in the joint.
Different test set-ups have been used for the characterisation of the shear
behaviour of the unit-mortar interface. These include direct shear or cou-
plet test and triplet test, see Figure 15. The triplet test can hardly be used
to obtain the post-peak characteristics because the joints do not fail simul-
taneously and a rotation is obtained, Lourenço et al. (2004), so a couplet
test is recommended for this purpose. It is also noted that a key issue for
obtaining the post-peak characteristics is to keep constant the stress normal
to the bed joint during testing.
(a) (b)
Figure 15. Possible test set-ups for shear strength: (a) couplet test;
(b) triplet test.
Experimental results yield typically an exponential shear softening dia-

gram with a residual dry friction level, see Figure 16a. The envelop of the
shear strength for different normal stress values provides the cohesion and
the friction angle for a Coulomb type friction model, see Figure 16b. A rec-
ommendation for the value of the bond shear strength (or cohesion) c based
on the unit type or mortar type is impossible, but an indication is given

in Eurocode 6 (CEN, 2005) for the characteristic value (95% fractile), in
the range of 0.1 to 0.4 N/mm2 . The ductility index du,s , given by the ratio
between the fracture energy Gf,II and the cohesion c, found for different
combinations of unit and mortar was between 0.062 mm and 0.147 mm, van
der Pluijm (1999). The recommended ductility index du,II is the average
value of 0.093 mm. It is noted that the Mode II fracture energy is clearly
dependent of the normal stress level and the above values hold for a zero
normal stress.
W
|V3| > |V2|
|V2| > |V1|

V1 | 0
Shear displacement
(a)
W
Gv
\ arctan
I0 Gh
Ir
Gv
c
Gh
V
(b) (c)
Figure 16. Shear bond behaviour: (a) typical stress-displacement diagram

for different normal stress levels (the shaded area represents the envelope
of three tests), van der Pluijm (1999); (b) cohesion c and friction angle φ,
defining the envelop obtained with the shear strength for different compres-
sive stresses; (c) dilatancy angle ψ is the uplift of neighbouring units upon
shearing
The friction angle of the joint φ, associated with a Coulomb friction

model, ranges from 0.7 to 1.2 for different unit-mortar combinations, van
der Pluijm (1999), but different values are found in the literature. A value
of 0.75 is recommended in the absence of more information. The dilatancy
angle ψ measures the uplift of one unit over the other upon shearing, see
Figure 16c. Note that the dilatancy angle decreases to zero with increasing
vertical stress and with increasing slip, so a zero value is recommended.
The effect of the dilatancy angle is only relevant for applications in which
masonry is confined, due to the interlocking effect of the units and the
associated stress built-up.
Properties of the masonry The compressive strength of masonry in

the direction normal to the bed joints has been traditionally regarded as
the most relevant structural masonry property. Since the pioneering work
of Hilsdorf (1969) it has been accepted by the masonry community that the
difference in elastic properties of the unit and mortar is the precursor of
failure. Lourenço and Pina-Henriques (2006) demonstrated that continuum
models are inadequate to justify the observed strength values in solid ma-
sonry and that there is a need to consider more refined models. Different
formulas to predict the compressive strength of masonry based on the prop-
erties of the components are available and are compared by these authors.
The formula available in Eurocode 6 (CEN, 2005) provides the characteris-
tic value of masonry under uniaxial compression as fk = Kfb0.7 fm 0.3
, where
K is factor about 0.5 for solid units, fb is the unit compressive strength and
fm is the mortar compressive strength.
Uniaxial compression tests in the direction parallel to the bed joints have
received substantially less attention from the masonry community. How-
ever, regular masonry is an anisotropic material and, particularly in the
case of low longitudinal compressive strength of the units due to high or
unfavourable perforation, the resistance to compressive loads parallel to the
bed joints can have a decisive effect on the load bearing capacity. According
to Hoffman and Schubert (1994), the ratio between the uniaxial compressive
strength parallel and normal to the bed joints ranges from 0.2 to 0.8.
For traditional masonry, information is available in PIET-70 (1971) for
the compressive strength of stone masonry, varying between 8.0 and 0.5
N/mm2 depending on the quality of the mortar, the type of stone and the
masonry bond, and for different properties of different masonry types in
OPCM 3431 (2005), varying between 6.0 and 0.6 N/mm2 .
The relation E = αfc between the Young’s modulus E and the com-
pressive strength fc is rather variable for masonry, with values of α ranging
between 200 and 1000 according to Tomazevic (1999), even if the proposed
value in Eurocode 6 (CEN, 2005) is 1000. For dry stone masonry, PIET-70
(1971) proposes a value of α = 500, which is in the mid-range of the interval
of Tomazevic (1999) and is possibly more adequate for traditional masonry.
3.3 Example of application

Advanced non-linear analyses are not particularly sensitive to the input
data, providing that the changes in the material properties are reasonable,
see Lourenço (1998b) and Mendes (2012). Therefore, numerical results can
be expected to reasonably replicate experimental tests and, in general, to
provide a single response to a given combination of geometry, loading and
materials. Next, an example of calculation of advanced material properties
from basic data of the masonry constituents is given taking as an example
a concrete block with two-cells and 10 N/mm2 compressive strength (fb )
and a mortar with 6 N/mm2 compressive strength (fm ). The geometry of
the block is 0.14 × 0.39 × 0.19 mm3 (t × l × h) with a thickness ts for the shell
walls equal to 25 mm.
The properties required for a micro-modelling approach can be estimated
as:
1. The bond tensile strength ft can be estimated at 0.2 N/mm2 which
is a minimum value for modern masonry, with the exception of very
smooth units. The recommended bond tensile fracture energy is
0.012 N/mm.
2. The cohesion or shear strength can be estimated as 1.5 ft or 0.3
N/mm2 The recommended ductility index du,II is 0.093 mm, provid-
ing a fracture energy for mode II of 0.028 N/mm. The recommended
values for the tangent of the friction and dilatancy angle (in case of a
non-associated model) are 0.75 and zero, respectively.
3. The masonry compressive strength (perpendicular to the bed joints)
for this type of units can be safely estimated as 0.7fb , providing a
value of 7 N/mm2 . If the Eurocode 6 (2005) formula is used instead, a
lower bound value would be 0.45 × 100.7 × 60.3 /0.8 = 6 N/mm2 , where
the 0.8 value is adopted to change from the characteristic to the mean
value. The recommended ductility index du,c is 1.6 mm, providing a
fracture energy for compression of 9.6 N/mm.
The properties required for a macro-modelling approach using an isotro-
pic model are the same as above, adopting the tensile bond strength and
compressive strength. If a macro-modelling approach with an anisotropic
model is used, additional material properties are needed:
1. The masonry tensile strength perpendicular to the joints is controlled
by the bed joint, being the same as above.
2. The masonry tensile strength parallel to the bed joints can be ob-
tained as indicated next. Assuming a void ratio of 50%, the compres-
sive strength of concrete fconc is 20 N/mm2 . The tensile strength for
a straight crack through head joints and blocks can be approximated
to the strength of the block given by 10% × fconc × 2ts × h/(t × 2h) =
0.35 N/mm2 . The tensile strength for a stepped crack can be approx-
imated by c × l/2/h = 0.29 N/mm2 . The value to be used is the min-
imum of the two values (0.29 N/mm2 ), being this response in shear
fully plastic. Note that the tensile strength of the head joints was not
used, as it is rather low and will not reach the peak value simultane-
ously with the other phenomena involved.
3. The masonry compressive strength perpendicular to the joints is the
same as above.
4. The masonry compressive strength parallel to the bed joints for this
type of masonry can be estimated as 30% of the vertical strength, as
a lower bound. This value can also can be calculated from Eurocode
6 as 0.35 × (fconc × ts × 2/t)0.7 × 60.3 /0.8 = 3.0 N/mm2 . The recom-
mended ductility index du,c is 1.6 mm, providing a fracture energy for
compression of 4.8 N/mm.
Bibliography
M. Angelillo. Constitutive relations for no-tension materials. Meccanica,
28:195–202. 1993.
A. Baratta and R. Toscano. Stati tensionali in pannelli di materiale non
resistente a trazione. In Atti del VII Congresso Nazionale AIMETA,
1982 (in Italian).
E. Benvenuto. An Introduction on the History of Structural Mechanics Part
II: Vaulted Structures and Elastic Systems, Springer-Verlag. Springer
Verlag, 1991.
G. Castellano. Un modello cinematico per i materiali non resistente a
trazione. In Cinquantenario della Facoltà di Architettura di Napoli:
Franco Jossa e la sua opera, pages 241–256, 1988 (in Italian).
CEB-FIP. Model Code 90. Thomas Telford Ltd., UK, 1993.
CEN. Eurocode 6: Design of masonry structures, Part 1–1: General rules
for reinforced and unreinforced masonry structures. European Commit-
tee for Standardization, Belgium, 2005.
M. Como. Statica delle costruzioni storiche in muratura, Aracne. Aracne,
2010 (in Italian).
M. Como and A. Grimaldi. A unilateral model for the limit analysis of
masonry walls. In Unilateral problems in structural analysis, edited by
G. Del Piero and F. Maceri, CISM Courses and Lectures 288, Springer.,
pages 25–45, 1985.
G. Del Piero. Constitutive equation and compatibility of the external loads
for linear elastic masonry–like materials. Meccanica, 24:150–162, 1989.
S. Di Pasquale. Statica dei solidi murari: teorie ed esperienze. Reports
Dipartimento di Costruzioni, Univ. Firenze, 27, 1984 (in Italian).
M. Giaquinta and E. Giusti. Researches on the equilibrium of masonry

structures. Arch. Rational Mech. Analysis, 88:359–392, 1985. ISSN 0950-
2289.
J. Heyman. The stone skeleton. Int.J.Solids,Struct., 2:249–279, 1966.
J. Heyman. The stone skeleton: structural engineering of masonry archi-
tecture, Cambridge University Press. Cambridge University Press, 1995.
H.K. Hilsdorf. Investigation into the failure mechanism of brick masonry
loaded in axial compression. In F.H. Johnson, editor, Designing, en-
gineering and constructing with masonry products, pages 34–41. Gulf
Publishing Company, Houston, Texas, 1969.
G. Hoffman and P. Schubert. Compressive strength of masonry parallel
to the bed joints. In N.G. Shrive and A. Huizer, editors, Proc. 10th
Int. Brick and Block Masonry Conf., pages 1453–1462, Calgary, Alberta,
1994.
S. Huerta. Galileo was wrong: the geometrical design of arches. Nexus
Network Journal, 8 (2):25–52, 2006.
P.B. Lourenço. Experimental and numerical issues in the modelling of the
mechanical behaviour of masonry. In P. Roca and et al., editors, Proc.
Structural analysis of historical constructions II, CIMNE, pages 57–91,
Barcelona, 1998a.
P.B. Lourenço. Sensitivity analysis of masonry structures. In Proc. 8th
Canadian Masonry Symp., pages 563–574, Jasper, Canada, 1998b.
P.B. Lourenço. Computations on historic masonry structures. Progress in
Structural Engineering and Materials, 4(3):301–319, 2002.
P.B. Lourenço and J.L. Pina-Henriques. Validation of analytical and con-
tinuum numerical methods for estimating the compressive strength of
masonry. Computers & structures, 84(29):1977–1989, 2006.
P.B. Lourenço and L.F. Ramos. Characterization of cyclic behavior of dry
masonry joints. Journal of Structural Engineering, 130(5):779–786, 2004.
P.B. Lourenço, J.O. Barros, and J.T. Oliveira. Shear testing of stack bonded
masonry. Construction and Building Materials, 18(2):125–132, 2004.
P.B. Lourenço, J.C. Almeida, and J.A. Barros. Experimental investigation
of bricks under uniaxial tensile testing. Masonry International, 18(1):
11–20, 2005.
P.B. Lourenço, G. Milani, A. Tralli, and A. Zucchini. Analysis of masonry
structures: review of and recent trends in homogenization techniques
this article is one of a selection of papers published in this special issue
on masonry. Canadian Journal of Civil Engineering, 34(11):1443–1457,
2007.
P.B. Lourenço, N. Mendes, L.F. Ramos, and D.V. Oliveira. Analysis of
masonry structures without box behavior. International Journal of Ar-
chitectural Heritage, 5(4-5):369–382, 2011.
R. Marques and P.B. Lourenço. Possibilities and comparison of structural

component models for the seismic assessment of modern unreinforced
masonry buildings. Computers & Structures, 89(21):2079–2091, 2011.
N. Mendes. Seismic assessment of ancient masonry buildings: Shaking table
tests and numerical analysis. PhD thesis, University of Minho, Portugal,
2012.
OPCM 3431 . Technical regulations for the design, assessment and seis-
mic adaptation of buildings. Gazzetta Ufficiale Serie Generale n.107 del
10/05/2005, Italy, 2005.
M. Paulo-Pereira. Assessment of the seismic performance of building en-
closures (in Portuguese). PhD thesis, University of Minho, Portugal,
2012.
PIET-70. P.I.E.T. 70 Masonry work. Prescriptions from Instituto Eduardo
Torroja. Consejo Superior de Investigaciones CientÃŋficas, Madrid (in
Spanish), 1971.
A. Prota, G. Manfredi, F. Nardone. Assessment of formulas for in plane
FRP strengthening of masonry walls. ASCE J. of Composite for Constr.
12, 6, 643-649, 2008.
G. Romano and M. Romano Sulla soluzione di problemi strutturali in pre-
senza di legami costitutivi unilaterali. Rend. Accad. Naz. Lincei, 67:
104–113, 1979 (in Italian).
G. Romano and E. Sacco. Sul calcolo di strutture non resistenti a trazione.
In Atti del VII Congresso Nazionale AIMETA ., pages 221–233, 1985 (in
Italian).
J.G. Rots, editor. Structural masonry: An experimental/numerical basis for
practical design rules. Balkema, Rotterdam, the Netherlands, 1997.
P. Schubert. The influence of mortar on the strength of masonry. In J.W.
de Courcy, editor, Proc. 8th Int. Brick and Block Masonry Conf., pages
162–174, London, 1988. Elsevier Applied Science.
M. Tomazevic. Earthquake-resistant design of masonry buildings. Imperial
College Press, London, 1999.
R. van der Pluijm. Out-of-plane bending of masonry: Behavior and Strenght.
PhD thesis, Eindhoven University of Technology, The Netherlands, 1999.
G. Vasconcelos and P.B. Lourenço. In-plane experimental behavior of stone
masonry walls under cyclic loading. Journal of structural engineering,
135(10):1269–1277, 2009.
G. Vasconcelos, P.B. Lourenço, C.A.S. Alves, and J. Pamplona. Experimen-
tal characterization of the tensile behaviour of granites. International
journal of rock mechanics and mining sciences, 45(2):268–277, 2008.
I Simplified Models
Mathematics of the Masonry–Like model and
Limit Analysis
M. Šilhavý
Institute of Mathematics of the AV ČR
Žitná 25
115 67 Prague 1
Czech Republic
1 Introduction
These notes present a brief introduction to the mathematics of equilibrium
of no–tension (masonry–like) materials. We first review the constitutive
equations using the idea that the stress of the no–tension material must be
always negative semidefinite. The strain tensor is naturally split into the
sum of the elastic strain and fracture strain. The stress depends linearly
on the elastic strain via the fourth–order tensor of elasticities. Then we
consider a body made of a no–tension material, introduce the loads and the
total energy of the deformation with is the sum of the internal energy and
the energy of the loads. Then we examine the question whether the total
energy is bounded from below. That brings us to the important notion
of the strong compatibility of loads. The loads are strongly compatible
if they can be equilibrated (in the sense of the principle of virtual work)
by a square integrable negative semidefinite stress field. The total energy
is bounded from below if and only if the loads are strongly compatible.
The notion of strong compatibility of loads is central in the limit analysis
and in a strengthened form in the theory of existence of equilibrium states.
Roughly speaking, if the loads are strongly compatible, then the body is
safe, while otherwise strongly incompatible loads lead to the collapse of
the body. To determine whether the loads are strongly compatible, it is
not necessary to solve the full displacement problem, it suffices to find the
negative semidefinite square integrable stress field, which is independent of
the constitutive theory.
The considerations concerning the limit analysis and strong compati-
bility of loads are based on the displacements that belong to the Sobolev
space of square integrable maps with square integrable gradients. Roughly

30 M. Šilhavý
speaking, this means that the fracture part of the strain is always without
macroscopic cracks. To obtain the existence theory of equilibrium states, it
is necessary to enlarge the class of deformations in which the fracture part
of the deformation is a measure in the mathematical sense of that word.
This introduces discontinuous displacements with macroscopic cracks. We
give a brief introduction to the mathematical theory of such displacements,
called displacements with bounded deformation. The full theory of equilib-
rium states is nontrivial and many assertions are presented without proofs.
However, the basic line of thought is preserved.
Then the limit analysis for no–tension materials is presented. The loads
of the limit analysis are assumed to be linearly dependent on the loading
parameter. The ideal goal of the limit analysis is to determine the largest
value of the loading parameter for which the loads are strongly compatible.
This value is called the collapse multiplier; the loads corresponding to the
loading multiplier bigger than the collapse multiplier lead to the collapse of
the body. The loading parameters for which the loads are strongly compat-
ible are called statically admissible loading parameters. By the above, they
are characterized by the existence of a square integrable negative semidef-
inite stress field equilibrating the corresponding loads. For concrete loads,
it is often easier to find a stress field represented by a negative semidefinite
tensor valued measure equilibrating the loads. We call such loads weakly
compatible. The difference between the square integrable stress fields and
the stress fields represented by measures is that the latter can contain sin-
gular part which is concentrated on surfaces and curves in the body. Of
course the strong compatibility implies the weak compatibility but not con-
versely: there are weakly compatible loads that are not strongly compatible.
If the loads happen to be weakly compatible on some interval of the load-
ing parameters, then the averaging procedure to be described in Section 7
may lead to square integrable equilibrating stress fields and hence to the
strong compatibility. The last section presents an example of the averaging
procedure which leads to an explicit determination of the square integrable
averaged stress field.
The mechanical tools to be employed include the notion of stress, the
virtual power principle, the notion of weak solution, and the notion of the
total energy of the body. These notions are established in detail in the
treatment below.
The mathematical tools necessary for the understanding include in par-
ticular the notions of the convex cones and orthogonal projection upon
them, some elements of the convex analysis, vector valued measures, Sobolev
spaces, families of measures, and the basic notions associated with the space
of displacements of bounded deformation. The basic definitions of the math-
Mathematics of the Masonry-Like Model and Limit Analysis 31
ematical notions are given in the text below and the basic properties are
stated without proof.
2 Constitutive equations
Throughout we use the conventions for vectors and second order tensors
identical with those in (Gurtin 1981). Thus Lin denotes the set of all sec-
ond order tensors on Rn , i.e., linear transformations from Rn into itself,
Sym is the subspace of symmetric tensors, Skew is the subspace of skew
(antisymmetric) tensors, Sym+ the set of all positive semidefinite elements
of Sym; additionally, Sym− is the set of all negative semidefinite elements of
Sym . The scalar product of A, B ∈ Lin is defined by A · B = tr(ABT ) and
| · | denotes the associated euclidean norm on Lin. We denote by 1 ∈ Lin
the unit tensor. If A, B ∈ Sym, we write A ≤ B if B − A ∈ Sym+ .
To describe the stress, we assume that C : Sym → Sym is a given linear
transformation, the fourth order tensor of elastic constants, such that

E · CE > 0 for all E ∈ Sym, E = 0,
(1)
E1 · CE2 = E2 · CE1 for all E1 , E2 ∈ Sym .
For example, the tensor of elastic constants of an isotropic material is of the
form
CE = λ(tr E)1 + 2μE
for each E ∈ Sym where λ and μ are constants, Lamé moduli, satisfying
μ > 0, nλ + 2μ > 0
where n = 2, 3 is the dimension of the underlying space.
In the case of a general C, we introduce the energetic scalar product
(·, ·)E and the energetic norm | · |E on Sym by setting

(A, B)E = A · CB, |A|E = (A, A)E
for each A, B ∈ Sym .
We are now going to review briefly some concepts of convex analysis. We
refer to (Rockafellar 1970) and (Ekeland & Temam 1999) for more details.
A nonempty subset K of a vector space V is called a convex cone if
tx + sy ∈ K for each x, y ∈ K and each t ≥ 0, s ≥ 0.
Let K be a closed convex cone in a vector space V with scalar product
(·, ·) and norm | · |. We say that a point y ∈ K is the (orthogonal) projection
of a point x ∈ V if y makes the distance |z − x| minimal among all z ∈ K,
i.e., if
|y − x| ≤ |z − x|
32 M. Šilhavý
for all z ∈ K. The projection onto a closed convex cone exists and is uniquely
determined.
Let K be a nonempty set in a vector space V with scalar product and
x ∈ V. We define the normal cone Norm(K, x) to K at x by
Norm(K, x) = {y ∈ V : (z − x, y) ≤ 0 for all z ∈ K}.
Remark 2.1. Let K be a closed convex cone in a vector space V with

scalar product (·, ·) and norm | · | and let x ∈ K. Then
Norm(K, x) = {y ∈ V : (y, z) ≤ 0 for each z ∈ K and (x, y) = 0}.
Proof. If y ∈ Norm(K, x) then (z − x, y) ≤ 0 for all z ∈ K. Replacing z

by tz where t > 0 in the last inequality, dividing the resulting inequality
by t and letting t → ∞ we obtain (z, y) ≤ 0. Similarly, taking z = 0 in
(z − x, y) ≤ 0 we obtain (x, y) ≥ 0 and as x ∈ K we have also (x, y) ≤ 0 by
the preceding part of the proof. Thus we have (x, y) = 0. Thus y ∈ {y ∈
V : (y, z) ≤ 0 for each z ∈ K and (x, y) = 0}. Conversely, if y ∈ {y ∈ V :
(y, z) ≤ 0 for each z ∈ K and (x, y) = 0} then the inequality (y, z) ≤ 0 and
the equality (x, y) = 0 provide (z − x, y) ≤ 0 for all z ∈ K.
Remark 2.2. Let K be a closed convex cone in a vector space V with

scalar product (·, ·) and norm | · | and let x ∈ V. Then a point y ∈ K is
the projection of x onto K if and only if the following two conditions are
satisfied:
(i) (w, x − y) ≤ 0 for all w ∈ K;
(ii) (x − y, y) = 0.
Equivalently, a point y ∈ K is the projection of x onto K if and only if
x − y ∈ Norm(K, y).
Proof. Assume that y is the projection of x onto K. Then for every z ∈ K

we have
|z − x| ≥ |y − x| (2)
which can be rewritten as
|z − y + y − x|2 ≥ |y − x|2
which in turn implies
|z − y|2 + 2(z − y, y − x) ≥ 0. (3)

We now put z = y + tw where w ∈ K and t > 0 to obtain
t2 |w|2 + 2t(w, y − x) ≥ 0.
Dividing by t and letting t → 0 we obtain (i). Next we put z = (1 + t)y

where t ≥ −1 to obtain
t2 |y|2 + 2t(y, y − x) ≥ 0.
Dividing by t > 0 and letting t ↓ 0 we obtain (y, y − x) ≥ 0; dividing by

t < 0 and letting t ↑ 0 we obtain (y, y − x) ≤ 0 and hence we have (ii).
Conversely, let (i) and (ii) hold. If z ∈ K then summing the relations
2(z, y − x) ≥ 0, −2(y, y − x) = 0 and |z − y|2 ≥ 0 we obtain (3) which in
turn implies (2) and hence y is the projection of x onto K.
This completes the proof of the characterization by (i) and (ii). The
equivalent characterization in terms of the normal cone follows from (i) and
(ii) via Remark 2.1.
Proposition 2.3. Assume (1). If E ∈ Sym, there exists a unique triplet

(T, Ee , Ef ) of elements of Sym such that the following three equivalent char-
acterizations hold:
(i) we have ⎫
E = E e + Ef , ⎪
⎪
⎬
T = CEe ,
− + (4)
T ∈ Sym , Ef ∈ Sym , ⎪ ⎪
⎭
T · Ef = 0;
(ii) we have equations (4)1,2 and

T ∈ Sym− ,
(5)
(T − T ) · E ≥ 0 for each T∗ ∈ Sym− ;
∗ f
(iii) we have equations (4)1,2 and
Ee is the projection of E onto C−1 Sym− with respect to (·, ·)E . (6)
Proof. Let us first show that the three characterizations of the triplet (T,
Ee , Ef ) are equivalent.
Proof of (iii) ⇒ (i). Assume that Characterization (iii) holds. Let Ee be
the projection of E onto the convex cone C−1 Sym− with respect to the ener-
getic scalar product and let Ef and T be as in (4)1,2 . Since Ee ∈ C−1 Sym− ,
we have T = CEe ∈ Sym− and by Remark 2.2(i) (Ef , C−1 T∗ )E = (E −
Ee , C−1 T∗ )E ≤ 0 for all T∗ ∈ Sym− which can be rewritten as Ef · T∗ ≤ 0,
34 M. Šilhavý
which in turn implies that Ef ∈ Sym+ . Thus we have (4)3 . Finally, by Re-
mark 2.2(ii) we have (Ef , C−1 Ee )E = (E − Ee , C−1 T)E = 0 which can be
rewritten as Ef · T = 0. Thus we have (4)4 . This proves that (iii) ⇒ (i).
Proof of (i) ⇒ (ii). Assume that Characterization (i) holds. Then we
have (5)1 . Furthermore, if T∗ ∈ Sym− then T∗ · Ef ≤ 0 since Ef ∈ Sym+
by (4)3 . Combining with (4)4 we obtain (5)2 . Thus (i) ⇒ (ii).
Proof of (ii) ⇒ (iii). Assume that Characterization (ii) holds. Using
(4)1,2 we can rewrite (5) as
Ee ∈ C−1 Sym−
(E∗ − Ee , Ef ) ≤ 0 for each E∗ ∈ C−1 Sym− ;.

Thus
Ef (≡ E − Ee ∈ Norm(C−1 Sym− , Ee ),
and Remark 2.2 asserts (6). Thus we have shown that (ii) ⇒ (iii).
Summarizing, we have shown that the three characterizations of the
triplet (T, Ee , Ef ) are equivalent. Characterization (iii) shows that the
triplet exists and is unique, because Ee , being the projection of E on
C−1 Sym− is unique, and the uniqueness of Ef and T follows from (4)1,2 .
We define the elastic stress T̂ : Sym → Sym and stored energy ŵ :

Sym → R of a masonry material by
T̂(E) = T, ŵ(E) = 12 T̂(E) · E = 12 |PE|2E (7)
for any E ∈ Sym where (T, Ee , Ef ) is the triplet associated with E as

in Proposition 2.3 and where P : Sym → C−1 Sym− denotes the projec-
tion from Sym onto C−1 Sym− with respect to the energetic scalar product
(·, ·)E . The tensors Ee and Ef are called the elastic and fracture parts of
the deformation E.
The no–tension materials have been introduced in the eighties in the
papers (Di Pasquale 1984), (Anzellotti 1985), (Giaquinta & Giusti 1985),
(Del Piero 1989), (Angelillo 1993). The explicit form of the response func-
tion T̂ and its further analysis have been given in the case of C isotropic in
(Anzellotti 1985), (Giaquinta & Giusti 1985) in dimension 2 and in (Luc-
chesi et al. 1994), (Lucchesi et al. 1996) in dimension 3; see also (Lucchesi
et al. 2008a). We also note for the reader’s curiosity that the membranes
with continuously distributed wrinkles differ from the no–tension materials
by the exchange of the roles of the cones Sym− and Sym+ (Epstein 2002),
(Barsotti & Vannucci 2013).
If F : V → R̄ := R ∪ {∞, −∞} is a function on an inner product space

V then F ∗ : V → R̄ is the convex conjugate function defined by
F ∗ (y) = sup{(x, y) − F (x) : x ∈ V },
y ∈ V, (Ekeland & Temam 1999, Part One).
Proposition 2.4. The map T̂ is monotone and Lipschitz continuous and

the function ŵ is continuously differentiable, convex and Dŵ = T̂; in fact,
we have the following inequalities:
(T̂(F) − T̂(E)) · (F − E) ≥ k|T̂(F) − T̂(E)|2 , (8)
|T̂(F) − T̂(E)| ≤ k −1 |F − E|, (9)

ŵ(F) ≥ ŵ(E) + T̂(E) · (F − E) + 12 k|T̂(F) − T̂(E)|2 (10)
for any E, F ∈ Sym where
k := inf{A · C−1 A : A ∈ Sym, |A| = 1} > 0. (11)
We have

∗
1
2T · C−1 T if T ∈ Sym− ,
ŵ (T) = (12)
∞ if T ∈ Sym \ Sym− .
Cf. Del Piero (Del Piero 1989, Proposition 4.4 and Lemma 5.1) for (8)–(10).
Proof. Let E, F ∈ Sym and put T = T̂(E), U = T̂(F). From (5)2 we obtain
(T − U) · (E − C−1 T) ≥ 0, (U − T) · (F − C−1 U) ≥ 0; (13)
summing these two inequalities and rearranging we obtain
(T − U) · C−1 (T − U) ≤ (T − U) · (E − F);
using (11) we obtain (8). Using the Schwarz inequality on the left hand side
of (8) we obtain (9). To prove (10), one finds that
ŵ(F) − ŵ(E) − T · (F − E) − 12 (T − U) · C−1 (T − U) = (U − T) · (F − C−1 U);
the last expression is nonnegative by (13)2 and hence
ŵ(F) − ŵ(E) − T · (F − E) − 12 (T − U) · C−1 (T − U) ≥ 0;

36 M. Šilhavý
a reference to (11) then yields (10) and hence also the convexity of ŵ. To
prove that ŵ is continuously differentiable and T̂ is its derivative, we note
that using (10) twice we obtain
T̂(F) · (F − E) ≥ ŵ(F) − ŵ(E) ≥ T̂(E) · (F − E)
for any E, F ∈ Sym; dividing by |E−F|, letting F → E, using T̂(F) → T̂(E)

and invoking the definition of the Fréchet derivative we obtain Dŵ(E) =
T̂(E). To prove (12), let ĥ : Sym → R ∪ {∞} be the function defined by
the right hand side of (12). We calculate the convex conjugate ĥ∗ (E) of h
at E ∈ Sym . We note that if (T, Ee , Ef ) is the triplet associated with E
as in Proposition 2.3, then algebraic manipulations show that (5)2 can be
rewritten as
T · E − ĥ(T) ≥ S · E − ĥ(S) + 12 (T − S) · C−1 (T − S) (14)
for every S ∈ Sym− with the equality if S = T. Since (14) holds also if
S∈ / Sym− as the right hand side is −∞ in that case, we have
T · E − ĥ(T) ≥ S · E − ĥ(S)
for all S ∈ Sym and thus the definition gives ĥ∗ (E) = T · E − ĥ(T)(≡ ŵ(E).
Then ŵ∗ = ĥ∗∗ = ĥ by (Fonseca & Leoni 2007, Theorem 4.92(iii)) since
ĥ is lowersemicontinuous, convex and bounded from below by an affine
(continuous) function. The proof of (12) is complete.
Proposition 2.5. The stored energy ŵ is decreasing in the sense that
ŵ(E + P) ≤ ŵ(E) (15)
for any E ∈ Sym and any P ∈ Sym+ . Moreover, the function ŵ is com-
pletely characterized by the following two equivalent requirements:
(i) ŵ is the largest decreasing function such that
ŵ(E) ≤ 12 E · CE (16)
for every E ∈ Sym;

(ii) we have
ŵ(E) = inf{ 12 (E − P) · C(E − P) : P ∈ Sym+ }. (17)

Proof. To prove (i), we invoke (10) in which we omit the last term on the
right hand side to obtain
ŵ(E + P) − T̂(E + P) · P ≤ ŵ(E).
Noting that T̂(E + P) · P ≤ 0 since T̂(E + P) ∈ Sym− and P ∈ Sym+

completes the proof of (15).
Next let ŵ : Sym → R be any function and let us prove that if it is given
by (17) then it is the largest decreasing function satisfying (16). Clearly, ŵ
satisfies (16). Let E ∈ Sym and Q ∈ Sym+ . Then
ŵ(E − Q) = inf{ 21 (E − Q − P) · C(E − Q − P) : P ∈ Sym+ }
and noting that Q + P ∈ Sym, we see that S := {Q + P : P ∈ Sym+ } ⊂

{P ∈ Sym+ } thus
ŵ(E − Q) = { 12 (E − P ) · C(E − P ) : P ∈ S }
≥ { 21 (E − P) · C(E − P) : P ∈ Sym+ }
= ŵ(E).
Thus ŵ is decreasing. Next assume that ŵ is a decreasing function satisfy-

ing (16) and prove that ŵ (E) ≤ ŵ(E) for every E ∈ Sym . Let P ∈ Sym+ .
We have
ŵ (E) ≤ ŵ (E − P) ≤ 12 (E − P) · C(E − P).
Taking the infimum over all P we obtain
ŵ (E) ≤ inf{ 12 (E − P) · C(E − P)} = ŵ(E).
This proves the equivalence of (i) and (ii).

Let now ŵ be the stored energy of a no–tension material and prove that
it satisfies (i). We have already seen that ŵ is decreasing. To prove (16),
we note that
E · CE = E · CEe + E · CEf
= Ee · CEe + (Ee + Ef ) · CEf
= 2ŵ(E) + Ee · CEf + Ef · CEf
≥ 2ŵ(E).
Thus ŵ satisfies (16). Furthermore, we have
ŵ(E) = 12 (E − Ef ) · C(E − Ef )
38 M. Šilhavý
and noting that Ef ∈ Sym+ we see that

ŵ(E) ≥ inf{ 12 (E − P) · C(E − P) : P ∈ Sym+ }.
Thus ŵ is a decreasing function satisfying (16) that is larger than the func-
tion defined by the right hand side of (17). However, since this right hand
side defines the largest function with this property, we have the equality
(17).
3 Vector valued measures

In this section we introduce the main object which will represent stresses
in masonry bodies and loads applied to the bodies. These two classes of
objects will be represented by tensor valued measures and vector valued
measures, respectively. Among measures there are ordinary functions (up
to an identification), i.e., stresses and loads in the usual sense, but most
importantly, measures can represent stresses or loads concentrated on lower–
dimensional objects–surfaces or curves. The main goal in this section is to
introduce the terminology and notation for measures with values in a finite
dimensional vector space. We refer to (Ambrosio et al. 2000, Chapter 1)
for further details.
Definition 3.1. Let V be a finite-dimensional vector space. By a V valued
measure in Rn we mean a map m from a system of all Borel sets in Rn to
V which is countably additive in the sense that if B1 , B2 , . . . is a disjoint
family of Borel sets in Rn then
∞ ∞

m Bi = m(Bi ).
i=1 i=1
Below we need the choices V = Sym and V = Rn . We call the Sym

valued measures tensor valued measures; this particular case is used to
model the stress fields over the body. We call the Rn valued measures
vector valued measures. These are used to model the loads applied to the
body.
We say that a function φ defined on the system of all Borel sets in Rn is a
nonnegative measure if it takes the values from the set [0, ∞] of nonnegative
numbers or ∞ which is countably additive in the sense that if B1 , B2 , . . .
is a disjoint family of Borel sets in Rn then
∞ ∞

φ Bi = φ(Bi )
i=1 i=1
and
φ(∅) = 0.
If Ω is a Borel subset of R and m a V valued measure or a nonnegative
n
measure, we say that m is supported by Ω if m(A) = 0 for any Borel set

A such that A ∩ Ω = ∅. We denote by M(Ω, V ) the set of all V valued
measures supported by Ω.
If m ∈ M(Ω, V ) and if α : Ω → V is a bounded Borel function then we
have a well defined integral
α · dm,
Ω
which is a real number.
We denote by Ln the Lebesgue measure in Rn (Ambrosio et al. 2000,
Definition 1.52) and if k is an integer, 0 ≤ k ≤ n, we denote by Hk the k-
dimensional Hausdorff measure (“k dimensional area”) in Rn (Ambrosio et
al. 2000, Section 2.8). If φ is a nonnegative measure or a V valued measure,
we denote by φ A the restriction of φ to a Borel set A ⊂ Rn defined by
φ A(B) = φ(A ∩ B)
for any Borel subset B of R . Thus if N is an n − 1 dimensional surface in
n
Rn then Hn−1 N is the area measure on N .

If φ is a nonnegative measure, we denote by f φ the product of the measure
φ by a φ integrable V valued function f on Rn ; one has

(f φ)(A) = f dφ
A
n
for any Borel subset A of R .
The operations of restriction and multiplication of measures are em-
ployed to construct tensor valued measures concentrated on surfaces as fol-
lows:
Examples Consider a body Ω ⊂ Rn and an n − 1 dimensional surface

N ⊂ Ω and let E : Ω → Rn be a bounded continuous function, interpreted
as a field of strain over Ω.
(i) The measure Hn−1 N is supported by N and thus if Ts : N → Sym
is an Hn−1 integrable tensor field on N , then the measure
Ts := Ts Hn−1 N
is a tensor valued measure in M(Ω, Sym) concentrated on N . One has

E · dTs = E · Ts dHn−1 .
Ω N
40 M. Šilhavý
(ii) If Tr : Ω → Sym is an Ln integrable tensor field, then the tensor

valued measure
Tr := Tr Ln Ω
belongs to M(Ω, Sym) and faithfully represents Tr ; the measure is
distributed over Ω. One has

E · dTr = E · Tr dLn .
Ω Ω
The measures of the type Ts and Tr and their combinations T =

Tr + Ts will be employed in Sections 7–8 where we deal with weakly
compatible loads. Vector valued measures will be employed in the
following section to define the loads of the body.
The polar decomposition of measures (cf. (Rudin 1970, Theorem
6.12)) says that if m ∈ M(Ω, V ), there exists a pair (r, |m|) consisting
of a Borel function r : Ω → V and of a nonnegative measure |m| on Ω
such that
m = r|m|
and
|r(x)| = 1 for |m| almost every x ∈ Ω.
The measure |m| is unique and the function r is unique up to a change
on a |m| null set. The measure |m| is called the total variation measure
of m, and r the amplitude. We denote by M(m) the mass of m, defined
by M(m) = |m|(Rn ).
If Ω is an open subset of Rn , we denote by C0 (Ω, V ) the space of
all continuous V valued functions on Rn with compact support that is
contained in Ω, and denote by |·|C0 the maximum norm on C0 (Rn , V ).
4 Loads
We consider a continuous body represented by a Lipschitz domain
(Adams & Fournier 2003) Ω ⊂ Rn and assume that D, S are two disjoint
subsets of ∂Ω such that D ∪ S = ∂Ω, to be identified below as the set of pre-
scribed boundary displacement and prescribed boundary force. We assume
that D is a closed set.
We put
V0 = {v ∈ C 1 (cl Ω, Rn ) : v = 0 on D}
and
V = {v ∈ W 1,2 (Ω, Rn ) : v = 0 almost everywhere on D};
here C 1 (cl Ω, Rn ) is the set of all continuously differentiable mappings v :

Ω → Rn such that v and its derivative ∇v have a continuous extension to
the closure cl Ω of Ω and W 1,2 (Ω, Rn ) is the Sobolev space of all Rn valued
maps such that v and the weak gradient ∇v of v are square integrable on
Ω, i.e.,
|v|2 dLn < ∞, |∇v|2 dLn < ∞,
Ω Ω
(Adams & Fournier 2003). We assume that V0 is a dense subset of V. For
any v ∈ V we define the infinitesimal strain tensor Ê(v) of v by
Ê(v) = 12 (∇v + ∇vT ).
We assume that the body is subjected to loads which consist of a body

force acting in the interior of Ω and of the surface force acting on S. We rep-
resent both the body and surface forces as vector valued measures supported
by Ω and S, respectively. Thus we assume that we are given b ∈ M(Ω, Rn )
and s ∈ M(S, Rn ) with the following meaning. For any Borel subset P of
Ω the value b(P ) ∈ Rn is the body force acting on P from the exterior and
for any Borel S ⊂ S the value s(S) ∈ Rn is the force acting on the surface
S from the exterior of the body. We call the pair (s, b) the loads acting on
the body. We note that we define the loads as measures, which allows for
the concentration of the body force and more importantly surface tractions.
We define for each u ∈ V the internal energy of the body by

E(u) = ŵ(Ê(u)) dLn
Ω
and for each u ∈ V0 the energy of the loads by

l, u = u · db + u · ds.
Ω S
The total energy of the deformation u ∈ V0 is defined by
F(u) = E(u) − l, u .
An important special case arises when there are square integrable func-
tions b ∈ L2 (Ω, Rn ), s ∈ L2 (S, Rn ) such that
b = bLn Ω, s = sHn−1 S. (18)
Here L2 (Ω, Rn ) is the set of all Ln measurable maps b : Ω → Rn such that

|b|2 dLn < ∞
Ω
42 M. Šilhavý
and L2 (S, Rn ) is the set of all Hn−1 measurable maps s : S → Rn such that

|s|2 dHn−1 < ∞.
S
In the case (18) one can define the potential energy of the loads l, u for
each u from the larger space V by

l, u = u · b dLn + u · s dHn−1
Ω S
u ∈ V.
Given the loads (s, b) and u ∈ V, we say that u is an equilibrium state
of Ω under the given loads if

T̂(Ê(u)) · Ê(v) dLn = l, v (19)
Ω
for each v ∈ V0 .
We note that if the loads are of the special form (18) with b ∈ L2 (Ω, Rn ),
s ∈ L2 (S, Rn ) and if T̂(Ê(u)) ∈ C 1 (cl Ω, Sym) then the variational equation
(19) is equivalent to the strong form
div T + b = 0 in Ω and Tn = s on S
where n is the outer normal to ∂Ω. We note that in general the existence of
the equilibrium state is not guaranteed. The existence theory of equilibrium
state requires the extension of the states to admit fracture. See Section
5, below. On the other hand, the given loads may admit more than one
equilibrium state u.
Remark 4.1. Assume the loads of the special form (18) with b ∈ L2 (Ω, Rn ),
s ∈ L2 (S, Rn ). Then u ∈ V is an equilibrium state under the given loads if
and only if u is a minimizer of the total energy under the given loads.
Proof. Let u be an equilibrium state under the given loads. Let v ∈ V0 .

Then

F(u + v) = Ω ŵ(Ê(u) + Ê(v)) dLn − l, u − l, v

≥ Ω [ŵ(Ê(u)) + T̂(Ê(u) · Ê(v)] dLn − l, u − l, v

= F(u) + [ Ω T̂(Ê(u) · Ê(v) dLn − l, v]
by (10). Since u is an equilibrium state, the square bracket on the last line
vanishes and we obtain
F(u + v) ≥ F(u) (20)
for each v ∈ V0 . Since V0 is dense in V and the energy a continuous func-

tional, we have (20) for each v ∈ V. Thus u is a point of minimum energy.
Conversely, assume that u is a point of minimum energy. Let v ∈ V0
and t ∈ R. We have
F(u + tv) ≥ F(u)
for all t ∈ R with the equality sign for t = 0 and thus the derivative of the
function t → F(u + tv) at t = 0 vanishes. One has

F(u + tv) = ŵ(Ê(u) + tÊ(v)) dLn − l, u − t l, v
Ω
and differentiating under the integral sign we obtain

d
F(u + tv)t=0 = T̂(Ê(u)) · Ê(v)) dLn − l, v = 0.
dt Ω
Hence u is an equilibrium state of Ω under the given loads.
Let (s, b) be given loads of Ω and let T ∈ L2 (Ω, Sym) where 2

L 2(Ω, Sym) is
the set of all L measurable maps T : Ω → Sym such that Ω |T| dLn < ∞.
n
We say that T equilibrates the loads (s, b) if

T · Ê(v) dLn = l, v
Ω
for all v ∈ V0 . We say that T is admissible if T(x) ≤ 0 for Ln almost every

x ∈ Ω. We say that the lods (s, b) are strongly compatible if there exists an
admissible stress field equilibrating the loads. Using this terminology we
can say that if u is an equilibrium state of Ω under the given loads then the
stress field corresponding to u is admissible and equilibrates the loads. Thus
the loads must be strongly compatible for an equilibrium state to exist.
Theorem 4.2. Let u ∈ V0 be an equilibrium state under the loads (s, b).
Then the stress field S := T̂(Ê(u)) is a minimum point of the complemen-
tary energy functional

G(T) = 2 1
T · C−1 T dLn
Ω
among all admissible stress fields T equilibrating the loads.
We call G the complementary energy.

44 M. Šilhavý
Proof. Let T be an admissible stress field equilibrating the loads and let Ee
and Ef be the elastic and fracture parts of the strain corresponding to u.
Using the convexity of the function U → 12 (C−1 U · U) we find

G(T) − G(S) ≥ Ω C−1 S · (T − S) dLn

= Ω Ee · (T − S) dLn

= Ω Ee · T dLn − l, u

= [ Ω Ê(u) · T dLn − l, u] − Ω Ef · T dLn

= − Ω Ef · T dLn ≥ 0
since u is an equilibrium state and because the square bracket vanishes as
T is a stress field equilibrating the loads.
As a corollary to Theorem 4.2 we have that while the displacement corre-

sponding to equilibrium may be nonunique, the equilibrium stress is unique
as the complementary energy has a unique minimum point. We note also
that the complementary energy may admit a minimum among the admissi-
ble stress fields equilibrating the given loads, and yet the equilibrium state
need not exist.
Let (s, b) be the loads of Ω. We put
I0 = inf{F(u) : u ∈ V0 }.
In general,
−∞ ≤ I0 < ∞
and we have I0 > −∞ if and only if the total energy F is bounded from
below. Let H : L2 (Ω, Sym) → R be defined by

H(A) = ŵ(A) dLn
Ω
for each A ∈ L2 (Ω, Sym). Let
H ∗ (T) = sup{A · T − H(A) : A ∈ L2 (Ω, Sym)}
for any T ∈ L2 (Ω, Sym). Then (Ekeland & Temam 1999)

H ∗ (T) = ŵ∗ (T) dLn
Ω
and hence
1

T · C−1 T dLn if T is negative semidefinite,
H ∗ (T) = 2 Ω
∞ otherwise,
T ∈ L2 (Ω, Sym), by (12).

Proposition 4.3. Consider the general loads (s, b). Then the loads are
strongly compatible if and only if
I0 > −∞.
Proof. Let Y := L2 (Ω, Sym) and X0 := {Ê(v) : v ∈ V0 } so that X0 ⊂

Y . Assume that c = I0 ∈ R. Prove a preliminary result: if v1 v2 ∈ V0
satisfy Ê(v1 ) = Ê(v2 ) then l, v1 = l, v2 . Indeed, let t ∈ R and put
v = (1 − t)v1 + tv2 . Then v ∈ V0 , Ê(v) = E := Ê(v1 ) = Ê(v2 ) and thus

F(v) = ŵ(E) dLn − (1 − t) l, v1 − t l, v2 ≥ c.
Ω
Assuming t > 0, dividing the inequality by t and letting t → ∞ we obtain

l, v1 − l, v2 ≥ 0; similarly, assuming t < 0, dividing by t and letting
t → −∞ we obtain l, v1 − l, v2 ≤ 0 and thus l, v1 = l, v2 which
completes the proof of the preliminary result. Let L0 : X0 → R be defined
by
L0 (Ê(v)) = l, v (21)
for each v ∈ V0 , where we use the preliminary result to see that the right
hand side of (21) depends only on Ê(v). Then
L0 (A) ≤ H(A) − c for all A ∈ X0 .
The convexity of ŵ implies the convexity of H and hence by the version

of the Hahn Banach theorem (Fonseca & Leoni 2007, Theorem A.35) there
exists a linear extension L : Y → R of L0 such that
L(A) ≤ H(A) − c for all A ∈ Y. (22)
The continuity of H on Y , which follows from the properties of ŵ, implies

that H is bounded on the unit ball in Y and hence L is bounded on the
unit ball and hence continuous. Thus it can be represented by an element
T ∈ Y as a scalar product in Y, i.e., there exists a T ∈ Y such that

L(A) = T · A dLn
Ω
for each A ∈ Y. Taking in particular A ∈ L2 (Ω, Sym+ ) and noting that

then ŵ(A) = 0, we find from (22) that
L(A) ≤ −c.
46 M. Šilhavý
Repalcing A by tA where t > 0, dividing by t and letting t → ∞ we obtain

L(A) ≤ 0 which implies that T ≤ 0 for almost all points of Ω. Further,
relation (21) gives
T · Ê(v) dLn = l, v
Ω
for each v ∈ V0 and thus T strongly equilibrates the loads (s, b).
To prove the converse part of the statement, we let T be a stress field
strongly equilibrating the loads (s, b). Since T is negative semidefinite and
square integrable, we have

H ∗ (T) = 12 T · C−1 T dLn < ∞
Ω
and hence

∞ > H ∗ (T) := sup{ T · A dLn − H(A) : A ∈ Y }
Ω
from which

H(A) − T · A dLn ≥ −H ∗ (T) for all A ∈ Y ;
Ω
taking A = Ê(v) where v ∈ V0 , this is rewritten as

F(v) ≥ c
for all v ∈ V0 [with c = −H ∗ (T)].
5 The existence of equilibrium states

In this section we outline the theory of existence of equilibrium for masonry
materials. This theory is due to G. Anzellotti (Anzellotti 1985) and M.
Giaquinta & E. Giusti (Giaquinta & Giusti 1985). The presentation below
follows (Anzellotti 1985). The reader is referred to the cited paper for fur-
ther details and proofs. The theories of Anzellotti and Giaquinta & Giusti
are based on the existence of uniformly negative definite stress field, see
Definition 5.10, below. A theory based on an alternative assumption of the
strong absence of collapse mechanism is presented in (Šilhavý 2013).
Definition 5.1. Let Ω ⊂ Rn be an open set. We denote by BD(Ω) the set
of all u ∈ L1 (Ω, Rn ) such that there exist a measure Ê(u) ∈ M(Ω, Sym)
such that
u · div T dLn = − T · dÊ(u) (23)
Ω Ω
for all T ∈ C0∞ (Ω, Sym). Here C0∞ (Ω, Sym) is the set of all T : Rn → Sym
such that the support
spt T = cl{x ∈ Rn : T(x) = 0}
is contained in Ω and is compact. We denote by |u|BD(Ω) the BD norm

defined by
|u|BD(Ω) = |u|L1 (Ω,Rn ) + M(Ê(u))
where we recall that M(Ê(u)) is the mass of Ê(u) defined in Section 3. We

call the elements of BD(Ω) displacements of bounded deformation.
In other words, the strain tensor, being generally a distribution defined by
Ê(u) = 12 (∇u + ∇uT ),
is a Sym valued measure if u ∈ BD(Ω). If ϕ ∈ C0∞ (Ω, R) and if we apply

(23) with Tij = Tji = 12 ϕ for some pair (i, j) of indices and Tkl = 0 otherwise
we obtain
1

n
2 ϕ ,i u j + ϕ ,j u i dL = − ϕ dÊij (u),
Ω Ω
which is the ‘index form’ of the definition of the space BD(Ω).

The space BD(Ω) endowed with the norm | · |BD(Ω) is a Banach space.
We refer to (Temam & Strang 1980), (Temam 1983) and (Ambrosio et
al. 1997) and the references therein for further details and proofs of dis-
placements with bounded deformation.
Example 5.2 (Fracture in BD(Ω)). Let Σ ⊂ Ω be a surface of dimension

n − 1 which divides Ω into two open sets Ω1 and Ω2 . Let u be a function
such that its restriction uk , k = 1, 2, onto Ωk belongs to C 1 (cl Ωk , Rn ).
Then u ∈ BD(Ω) and
Ê(u) = {Ê(u)}Ln Ω + [u] mHn−1 Σ (24)
where {Ê(u)} is the function equal to Ê(uk ) on Ωk (k = 1, 2), [u](x) =

u2 (x) − u1 (x) (x ∈ Σ), m is the normal to Σ pointing from Ω1 to Ω2 , and
a b = 12 (a ⊗ b + b ⊗ a)
for every a, b ∈ Rn .
48 M. Šilhavý
P roof Let T ∈ C0∞ (Ω, Rn ). Applying the divergence theorem to Ω1 and

Ω2 , we obtain

Ω
u · div T dLn = Ω1 u1 · div T dLn + Ω2 u2 · div T dLn

= − Ω1 Ê(u1 ) · T dLn + Σ Tm · u1 dHn−1

− Ω2 Ê(u2 ) · T dLn − Σ Tm · u2 dHn−1

= − Ω {Ê(u)} · T dLn − Σ T · ([u] m) dHn−1 .
Theorem 5.3. Let Ω ⊂ Rn be an open set with Lipschitz boundary. There

exists a linear map γ0 : BD(Ω) → L1 (∂Ω, Rn ) such that we have

u · div T dL +
n
T · dÊ(u) = Tn · γ0 (u) dHn−1
Ω Ω ∂Ω
for each u ∈ BD(Ω) and T ∈ C 1 (cl Ω, Sym). One has
|γ0 (u)|L1 (∂Ω,Rn ) ≤ c|u|BD(Ω)
for each u ∈ BD(Ω) and some c ∈ R.
The function γ0 (u) represents the boundary values of u. We often simplify

the notation and write u for γ0 (u). With this notation we have

u · div T dL +
n
T · dÊ(u) = Tn · u dHn−1 .
Ω Ω ∂Ω
Theorem 5.4. Let Ω ⊂ Rn be an open set with Lipschitz boundary. If

u ∈ BD(Ω) then one has u ∈ Ln/(n−1) (Ω, Rn ) and there exists a c ∈ R such
that
|u|Ln/(n−1) (Ω,Rn ) ≤ c|u|BD(Ω)
for all u ∈ BD(Ω).
Theorem 5.5. Ω ⊂ Rn be an open set with Lipschitz boundary. The

operator imbedding BD(Ω) into Lp (Ω, Rn ), 1 ≤ p < n/(n − 1), is com-
pact; i.e., if uj ∈ BD(Ω) is a sequence bounded in the | · |BD(Ω) norm and
1 ≤ p < n/(n − 1) then there exists a subsequence of uj , still denoted by uj ,
such that
uj → u in Lp (Ω, Rn )
for some u ∈ BD(Ω).
Theorem 5.6. Let Ω ⊂ Rn be an open connected set and u ∈ BD(Ω).

Then Ê(u) = 0 if and only if u is of the form
u(x) = Wx + a (25)
for all x ∈ Ω where a ∈ Rn and W ∈ Skew .
If n = 3 and b is the polar vector of W, we can write
u(x) = b × x + a.
We call any u of the form (25) a rigid body displacement and denote by
R(Ω) the linear space of all rigid body displacements of Ω.
Theorem 5.7. Let Ω ⊂ Rn be an open connected set. There exists a linear

map u → r(u) from BD(Ω) to R(Ω) such that r(u) = u for u ∈ R(Ω) and
|u − r(u)|Ln/(n−1) (Ω,Rn ) ≤ cM(Ê(u))
for all u ∈ BD(Ω) and some c ∈ R.
Recall from Section 2 that P : Sym → C−1 Sym− denotes the orthogonal
projection from Sym onto C−1 Sym− with respect to the energetic scalar
product (·, ·)E . Endow Sym with the energetic scalar product. Let u ∈
BD(Ω) and write
Ê(u) = D|Ê(u)|
for the polar decomposition of the measure Ê(u), with |Ê(u)| a nonnegative
measure on Ω and D : Ω → Sym a function satisfying |D|E = 1 for |Ê(u)|
almost every point of Ω. We denote by P Ê(u) the measure defined by
P Ê(u) = (PD)|Ê(u)|.
Definition 5.8. We denote by U(Ω) the set of all u ∈ BD(Ω) such that the
measure P Ê(u) is absolutely continuous with respect to the Lebesgue mea-
sure, with the density P̃ Ê(u) such that P̃ Ê(u) ∈ L2 (Ω, Sym). We call the
elements of U (Ω) admissible displacements. We define the internal energy
of the admissible displacement u by

E(u) = |P̃ Ê(u)|2E dLn .
Ω
Example 5.9 (Fracture in U (Ω)). The set U (Ω) is the basic set of com-
petitors for the equilibrium problem. Note that U (Ω) is not a linear space
50 M. Šilhavý
since if u ∈ U (Ω) then it may happen that −u ∈ / U (Ω). As an example, let

u ∈ BD(Ω) be as in Example 5.2. Let us show that u ∈ U (Ω) if and only
if there exists a function λ : Σ → [0, ∞) such that
[u] = λm on Σ, (26)
i.e., the jump in u across Σ is positively proportional to the normal to Σ.

This seems to be in agreement with the observation of fractured masonry
structures.
Proof. Assume that there exists a point x on Σ such that [u] and m are not
positively proportional. Then by the continuity there exists a neighborhood
N of x in Σ such that [u] and m are not positively proportional. One easily
finds that in this case the tensor [u] m has a nonzero negative definite
part, which further implies
P([u] m) = 0 on N,
since otherwise [u] m would be positive semidefinite. Equation (24) then

gives
P Ê(u) = P{Ê(u)}Ln Ω + P([u] m)Hn−1 Σ
and thus P Ê(u) has a nonzero singular part, in contradiction with the defi-
nition of U(Ω). Therefore we have (26) with a nonnegative λ. The converse
implication: under (26) we have P([u] m) = 0 and thus
P Ê(u) = P{Ê(u)}Ln Ω
and hence P Ê(u) is absolutely continuous with respect to Ln with a square

integrable density.
We shall deal with the existence of solution for the Neumann problem.
Thus we assume that D = ∅. We call a pair (s, b) loads for the system if
s ∈ L∞ (∂Ω, Rn ), b ∈ Ln (Ω, Rn ). In view of the fact that for any u ∈ BD(Ω)
we have γ0 (u) ∈ L1 (∂Ω), u ∈ Ln/(n−1) (Ω, Rn ), by the Hölder inequality we
have a well defined energy of the loads

l, u = u · s dHn−1 + u · b dLn
∂Ω Ω
for any u ∈ BD(Ω) and in particular for any u ∈ U (Ω). We define the total
energy of u ∈ U(Ω) by
F(u) = E(u) − l, u .

We say that Ω ⊂ Rn is an admissible domain if for any u ∈ U (Ω) there

exists a sequence uj ∈ C 1 (cl Ω, Rn ) such that
uj → u in Ln/(n−1) (Ω, Rn ),
M(Ê(uj )) → M(Ê(u)),
P̃(uj ) → P̃(u) in L2 (Ω, Sym).
It turns out that all Lipschitz domains in R2 are admissible and that for
any n all star shaped Lipschitz domains are admissible.
Definition 5.10. We say that a stress field T ∈ L2 (Ω, Sym) is safe if there
exists an α > 0 such that
−T(x) · A ≥ α|A|
for all A ∈ Sym+ and Ln almost every x ∈ Ω.

In other words, T is uniformly negative definite over Ω, which in particular
implies that s · n ≤ −α < 0 on ∂Ω, i.e., the body must be uniformly com-
pressed on the boundary. As before we say that a stress field T equilibrates
the loads (s, b) if
l, v = T · Ê(v) dLn
Ω
for every v ∈ C 1 (cl Ω, Rn ).
Theorem 5.11. Let Ω be an admissible Lipschitz domain in Rn and con-
sider loads (s, b). If there exists a safe stress field T equilibrating the loads
then the functional F is coercive on U (Ω) in the sense that

F(u) ≥ c1 |P̃(Ê(u))|2E dLn + M(Ê(u)) + c2
Ω
for some constants c1 , c2 with c1 > 0 and all u ∈ U (Ω).
We say that a sequence uj ∈ U(Ω) converges weakly to u ∈ BD(Ω) if we

have the relations
⎫
uj → u in L1 (Ω, Rn ), ⎪
⎪
⎪
⎪
uj u in L n/(n−1)
(Ω, R ), ⎬
n
(27)
M(Ê(uj )) ≤ M, ⎪
⎪
⎪
⎪
⎭
Ω
|P̃(Ê(uj ))|2E dLn ≤ M
for some M ∈ R.
52 M. Šilhavý

sider loads (s, b). Assume that there exists a stress field S ∈ L2 (Ω, Sym− )
and a function c ∈ Ln (Ω, Rn ) such that S is bounded and S equilibrates the
loads (s, c). Then the functional F is weakly sequentially lowersemicontinu-
ous in the sense that for any sequence uj ∈ U (Ω) which converges weakly to
u ∈ U(Ω), we have
lim inf F(uj ) ≥ F(u).
j→∞
Proposition 5.13. The set U (Ω) is closed under the weak convergence of
sequences, i.e., if uj ∈ U (Ω) converges weakly to u ∈ BD(Ω) then u ∈ U(Ω).

sider loads (s, b). Assume that there exists a safe stress field equilibrating
the loads and moreover there exists a stress field S as in Theorem 5.12.
Then there exists in U (Ω) a minimizer of F on U (Ω).
Proof. There exists a sequence uj ∈ U(Ω) such that
lim F(uj ) = inf{F(u) : u ∈ U(Ω)}.

j→∞
Since the energy functional F is coercive by Theorem 5.11, the boundedness

of the sequence F(uj ) implies that there exists a M ∈ R such that
M(Ê(uj )) ≤ M,

|P̃(Ê(uj ))|2E dLn ≤ M
Ω
If u → r(u) is the map from Theorem 5.7, we have
|uj − r(uj )| ≤ M(Ê(uj )) ≤ M
and thus the sequence vj := uj − r(uj ) is bounded in Ln/(n−1) (Ω, Rn ) and

hence it contains a subsequence, again denoted by vj , such that
vj v in Ln/(n−1) (Ω, Rn ) (28)
for some v ∈ Ln/(n−1) (Ω, Rn ). Since Ê(vj ) = Ê(uj ), we also have
M(Ê(vj )) ≤ M, (29)

|P̃(Ê(vj ))|2E dLn ≤ M. (30)
Ω
Since the sequence vj is bounded in Ln/(n−1) (Ω, Rn ), it is also bounded in

L1 (Ω, Rn ). Thus we conclude that |vj |BD(Ω) is bounded. The compactness
of the imbedding of BD(Ω) into L1 (Ω, Rn ) by Theorem 5.5 implies that we
have
vj → v in L1 (Ω, Rn ). (31)
We thus summarize (28)–(31) by saying that the sequence vj ∈ U (Ω) con-
verges weakly to v ∈ BD(Ω). The weak closedness of U (Ω) (Proposition
5.13) then says that v ∈ U (Ω). Moreover, one easily finds that F(vj ) = F(uj )
and thus
lim F(vj ) = min{F(u) : u ∈ U(Ω)}.
j→∞
On the other hand, the functional F is sequentially weakly lowersemicontin-

uous (Theorem 5.12) and thus
lim F(vj ) ≥ F(v).

j→∞
This gives
F(v) ≤ inf{F(u) : u ∈ U (Ω)}
and thus
F(v) = min{F(u) : u ∈ U (Ω)}.
6 Limit analysis
The limit analysis deals with the loads that depend linearly (affinely) on a
scalar parameter λ ∈ R, see (Del Piero 1998). Accordingly, we thus assume
(Lucchesi et al. 2008b), (Lucchesi et al. 2011) that the body and surface
forces bλ ∈ M(Ω, Rn ) and sλ ∈ M(S, Rn ) corresponding to λ are given by
bλ = b0 + λb1 , sλ = s0 + λs1 (32)
where
b0 , b1 ∈ M(Ω, Rn ) s0 , s1 ∈ M(S, Rn ).
We call (s(λ), b(λ)) = (sλ , bλ ) the loads corresponding to λ. If v ∈ V0 then
the work of the loads (s(λ), b(λ)) corresponding to v is

l(λ), v = v · ds +
λ
v · dbλ .
S Ω
If the loads have square integrable densities, i.e., if
b0 = b0 L n , b 1 = b1 L n , s0 = s0 Hn−1 , s1 = s1 Hn−1 ,
54 M. Šilhavý
where
b0 , b1 ∈ L2 (Ω, Rn ), s0 , s1 ∈ L2 (S, Rn ),
then one can extend the definition of l(λ) to elements v of V.
In the general context of loads represented by measures we define the
total energy F(v, λ) of the body corresponding to the loads (s(λ), b(λ)) and
displacement v ∈ V0 by
F(v, λ) = E(v) − l(λ), v
so that F(·, λ) : V0 → R. Central to our considerations is the infimum energy

I0 (λ) ∈ R ∪ {−∞} of the loads (s(λ), b(λ)) defined by
I0 (λ) = inf{F(v, λ) : v ∈ V0 }.
We denote by A(λ) the set of all admissible stress fields equilibrating the
loads (s(λ), b(λ)). Recal that the loads (s(λ), b(λ)) are strongly compatible
if A(λ) = ∅.
We now follow (Lucchesi et al. 2010).
Proposition 6.1.
(i) The loads (s(λ), b(λ)) are strongly compatible if and only if I0 (λ) >
−∞.
(ii) The function I0 : R → R∪{−∞} is concave and uppersemicontinuous,
i.e.,
I0 (αλ + (1 − α)μ) ≥ αI0 (λ) + (1 − α)I0 (μ)
for every λ, μ ∈ R and α ∈ [0, 1] and
I0 (λ) ≥ lim sup I0 (λk )

k→∞
for every λ ∈ R and every sequence λk → λ. Hence the set
Λ = {λ ∈ R : I0 (λ) > −∞}(≡ {λ ∈ R : A(λ) = ∅} (33)
is an interval.
Since the notion of compatibility of loads is independent of the tensor of
elastic constants C, also the finiteness of I0 (λ) is independent of C [within
the class specified by (1)], even though the concrete value of I0 (λ) depends
on C. We emphasize the role of the square integrability requirement of the
stress field in the definition of strongly compatible loads; there are loads
(s(λ), b(λ)) with I0 (λ) = −∞ and yet with (s(λ), b(λ)) being weakly equili-
brated by a stress field T ∈ L1 (Ω, Sym) \ L2 (Ω, Sym) with values in Sym− .
Proof.
(i): This follows from Proposition 4.3.
(ii): The affine dependence of l(λ) on λ implies that the function λ →
F(u, λ) is affine for each u ∈ V0 ; thus the function λ → I0 (λ), being the lower
envelope of the family of affine continuous functions over the parameter set
{u ∈ V0 }, is concave and uppersemicontinuous (Ekeland & Temam 1999,
Chapter I, Section 2).
Definition 6.2. Let Λ be given by (33). A loading multiplier λ ∈ R is said

to
(i) be statically admissible if λ ∈ Λ; otherwise λ is said to be statically

inadmissible;
(ii) be a collapse multiplier if it is a finite endpoint of Λ;
(iii) be
kinematically
admissible if there exists a v ∈ V0 such that Ê(v) ≥ 0,
l̄, v = 1 and
λ = − l0 , v ; (34)
(iv) admit a collapse mechanism if λ is kinematically admissible and λ ≤

sup Λ.
Remark 6.3.
(i) The collapse multiplier can be statically admissible as well as statically
inadmissible.
(ii) The notion of collapse multiplier can be given a dynamical meaning.
The paper (Padovani et al. 2008) considers no–tension bodies in dynamical
situations with a viscous perturbation of the equations of motion. It turns
out that if I0 (λ) > −∞ then the motion with arbitrary initial data stabilizes
in the sense that the kinetic energy satisfies K(t) → 0 as t → ∞ while if
I0 (λ) = −∞ then the total energy, given by the sum of the total potential
energy and the kinetic energy, T (t) = F (t) + K(t), satisfies T (t) → −∞ as
t → ∞.
a collapse mechanism then there exists a v ∈ V with
(iii) If λ admits
Ê(v) ≥ 0, l̄, v = 1 and l(λ), v = 0; each such a v is said to be a collapse
mechanism for the loads (s(λ), b(λ)).
(iv) If λ admits a collapse mechanism and if additionally λ is statically
admissible then each admissible equilibrating stress field for (s(λ), b(λ)) is
called a collapse stress field. A stronger version of the definition of collapse
mechanism v in (Del Piero 1998) requires that v be as in (iii) and that
additionally λ be statically admissible.
56 M. Šilhavý
Remark 6.4. If T ∈ L2 (Ω, Sym), we denote the normal cone to the set
L2 (Ω, Sym− ) at T by Norm(L2 (Ω, Sym− ), T) and applying the definition
from Section 2 we obtain
Norm(L2 (Ω, Sym− ), T)

= {D ∈ L2 (Ω, Sym) : Ω
D · (T − S) dLn ≥ 0
for each admissible stress field S}
= {D ∈ L2 (Ω, Sym) : D · (T − U) ≥ 0
for every U ∈ Sym− and Ln almost every point of Ω}.
Let λ ∈ R. If v ∈ V satisfies

Ê(v) ∈ Norm(L2 (Ω, Sym− ), T), l̄, v = 1 (35)
and
λ = − l0 , v (36)
for some T ∈ A(λ) then v is a collapse mechanism for the loads (s(λ), b(λ)).
Indeed, the second characterization of Norm(L2 (Ω, Sym− ), T) implies that
Ê(v) ≥ 0 almost everywhere on Ω.
The number of collapse multipliers ranges from 0 to 2. In applications,

one is interested in the larger of the possibly two collapse multipliers. Mo-
tivated by this, we introduce the multiplier
λ+
c := sup{λ ∈ R : λ is statically admissible } (37)
−∞ ≤ λ+ + +
c ≤ ∞; thus if λc is finite, then λc is a collapse multiplier, and if
+
there are two collapse multipliers, then λc is the larger of these two. Also,
we consider the multiplier
λ̄+
c = inf{λ ∈ R : λ is kinematically admissible }. (38)
Remark 6.5. The above definitions of λ+ +

c and λ̄c are based on the square
+
integrability: in the definition of λc the admissible equilibrating stresses are
square integrable and in the definition of λ̄+c we consider mechanisms that
are square integrable with the square integrable gradients. The definitions
of the analogs of λ+ +
c and λ̄c using different function spaces is treated in
detail in (Lucchesi et al. 2012). However, it must be emphasized that the
definitions based on the square integrability are well motivated by Proposi-
tion 4.3.
Remark 6.6. It turns out (Lucchesi et al. 2012) that the definition of the
kinematic multiplier (38) can be reformulated to the format of the varia-
tional problem by Ekeland & Temam (Ekeland & Temam 1999) and then
the static multiplier (37) takes the form of the dual problem in the sense of
the cited reference.
Our first result shows that our definition of the collapse multiplier gen-
eralizes that based on the collapse mechanism:
Theorem 6.7.
(i) We have λ+ +
c ≤ λ̄c .
(ii) If λ ∈ R admits a collapse mechanism then λ = λ+ +
c = λ̄c .
Proof.
(i): Let λ ∈ R be kinematically admissible, there exists a v ∈ V0 such
i.e.,
that Ê(v) ≥ 0 almost everywhere on Ω, l̄, v = 1, and l(λ), v = 0. Then
λ = − l0 , v and thus

λ̄+
c = inf{− l0 , v : v ∈ V0 , Ê(v) ≥ 0, l̄, v = 1}. (39)
Let λ ∈ R be statically admissible with the admissible equilibrating stress

fieldT and let v ∈ V0 be such that Ê(v) ≥ 0 almost everywhere on Ω and
l̄, v = 1. Then we have

0≥ T · Ê(v) dLn = l(λ), v = λ + l0 , v ,
Ω
Thus
λ ≤ − l0 , v .
Taking the infimum over all v with the indicated properties and using (39)
we find λ ≤ λ̄+ c and taking the supremum over all T with the indicated
properties, we obtain the inequality in (i).
(ii): Assume that λ ∈ R admits a collapse mechanism. Prove first that
λ = λ+c . Since λ admits a collapse mechanism, λ is kinematically admissible
and hence there exists a v ∈ V0 with

Ê(v) ≥ 0, l̄, v = 1 and l(λ), v = 0. (40)
Prove that I0 (μ) = −∞ for all μ > λ. We have E(tv) = 0 and hence
F(tv, μ) = − l(μ), tv

= − l(λ), tv − (μ − λ) l̄, tv = −(μ − λ)t l̄, v = −(μ − λ)t.
58 M. Šilhavý
Letting t → ∞ we thus obtain F(tv, μ) → −∞ as t → ∞ for all μ > λ.

Hence λ+ +
c ≤ λ; on the other hand, λ ≤ λc as part of the definition of the
multiplier admitting a collapse mechanism. This completes the proof of
λ = λ+c .
On the other hand, λ is kinematically admissible as part of the definition
of the property of λ admitting a collapse multiplier. Thus λ̄+ c ≤ λ = λc
+
and Item (i) completes the proof.
7 Families of measures and the weak compatibility of

loads
In this section we introduce the generalized compatibility of loads called
weak compatibility. This involves balancing the loads by a stressfield rep-
resented by a measure T from M(Ω, Sym). Thus in contrast to the strong
compatibility, which is based on balancing by a square integrable function,
we here admit concentrations of stress on objects of dimension lower than
the dimension n of the physical space Rn , see Example 3(i). In many con-
crete cases, it is easier to prove the weak compatibility than the strong
compatibility. However, Proposition 4.3 on the boundedness below of the
total energy requires strong compatibility. In the case of the limit analysis,
where we deal with the loads depending on the loading parameter λ, we
have the balancing measures depending on λ as well. Then we can use a
procedure of integrating the balancing stress measures with respect to the
loading parameter to smear out the singularities of the stress measure to
obtain a square integrable function.
We say that T ∈ M(Ω, Sym) is admissible if T takes the values in the set
Sym− of the negative semidefinite symmetric tensors, i.e., if T(A)a·a ≤ 0 for
any Borel set A ⊂ Ω and for any a ∈ Rn . We say that T weakly equilibrates
the loads (s, b) if

Ê(v) · dT = v · db + v · ds
Ω Ω N
for any v ∈ V0 . We say that the loads (s, b) are weakly compatible if there
exists an admissible T ∈ M(Ω, Sym) which weakly equilibrates them. The
reader is referred to (Šilhavý 2008) for the general properties of stresses
represented by measures.
If the loads are strongly compatible then they are weakly compatible;
however, there are examples of loads that are weakly compatible but not
strongly compatible.
Example 7.1. Consider a stress measure of the form
T = Tr + Ts , where Tr = Tr Ln Ω, Ts = Ts Hn−1 N
where Tr : cl Ω → Sym is a smooth Ln integrable function over the body

and Ts : cl N → Sym is a smooth Hn−1 integrable function over a smooth
surface N contained in Ω. Assume furthermore given the loads (s, b) of
the form (18) with b and s continuous integrable functions over Ω and S,
respectively. Then T weakly equilibrates the loads (s, b) if and only if the
following two conditions hold:
(i) we have
div Tr + b = 0 in Ω \ N , (41)
where div is the classical divergence operator;
(ii) the stress field Ts is superficial in the sense that Ts m = 0 where m
is the normal to N and we have
[Tr ]m + div N Ts = 0 on N (42)

N
where [Tr ] = T+ −
r −Tr is the jump discontinuity of Tr on N and div
is the superficial divergence (Lucchesi et al. 2006, Section 4);
(iii) we have
Tr n = s on S
where n is the outer normal to ∂Ω.
We now pass to the details of the integration procedure.
Definition 7.2.
An integrable parametric measure (see (Lucchesi et al. 2008b)) is a family
{mλ : λ ∈ Λ} of V valued measures on Rn where Λ ⊂ R is a L1 measurable
set of parameters such that
(i) for every continuous
V valued function f on Rn with compact support
the function λ → Rn f · dm is L1 measurable on Λ;
λ
(ii) we have
c := M(mλ ) dλ < ∞.
Λ
We note that parametric measures similar to those defined above occur

in the context of disintegration (slicing) of measures (Ambrosio et al. 2000,
Section 2.5) and, what is related, in the context of Young’s measures (Müller
1999, Chapter 5).
60 M. Šilhavý
Proposition 7.3. If {mλ : λ ∈ Λ} is an integrable parametric measure then

there exists a unique V valued measure m on Rn such that

f · dm = f · dmλ dλ (43)
Rn Λ Rn
for each continuous V valued function f on Rn with compact support.

We write
m= mλ dλ (44)
Λ
and call m the integral of the family {mλ : λ ∈ Λ} with respect to λ.
Proof. We note that for each continuous V valued function f on Rn with

compact support the right hand side of (43) is a well defined real number:
indeed

f · dmλ dλ ≤ Λ Rn |f | d|mλ | dλ
Λ Rn

≤ max{|f (x)| : x ∈ Rn } Λ M(mλ ) dλ
≤ c max{|f (x)| : x ∈ Rn }.
Thus by the Riesz representation theorem (Ambrosio et al. 2000, Theorem

1.54) there exists a measure m such that (43) holds.
The following two propositions give two important examples of inte-

grable parametric measures. In both cases the corresponding integral (44)
is absolutely continuous with respect to the Lebesgue measure.
Proposition 7.4. Let {Tλr : λ ∈ Λ} be a family of Sym valued functions
on Ω ⊂ Rn defined for all λ from a L1 measurable set Λ ⊂ R such that the
mapping (x, λ) → Tλr (x) is Ln+1 integrable on Ω × Λ, i.e.,

|Tλr (x)| dx dλ < ∞. (45)
Λ Ω
If we define a Sym valued measure Tλr by
Tλr = Tλr Ln Ω
then {Tλr : λ ∈ Λ} is an integrable parametric measure and we have

Tλr dλ = Tr Ln Ω
Λ
where
Tr (x) = Tλr (x) dλ
Λ
for Ln almost every x ∈ Ω.
Proof. This follows directly from Fubini’s theorem.
Proposition 7.5. Let Ω0 ⊂ Rn be open, let ϕ : Ω0 → R be locally Lipschitz

continuous and let Ts : Ω0 → Sym be Ln measurable on Ω0 , with

|Ts ||∇ϕ| dLn < ∞. (46)
Ω0
Then for L1 almost every λ ∈ R the function Ts is Hn−1 ϕ−1 (λ) inte-
grable; denoting by Λ the set of all such λ we define the measure mλ by
Tλ := Ts Hn−1 ϕ−1 (λ)
for each λ ∈ Λ. Then {Tλ : λ ∈ Λ} is an integrable parametric measure and

we have
Tλ dλ = Ts |∇ϕ|Ln Ω0 . (47)
Λ
Proof. Let T be given by

T= Tλ dλ. (48)
Λ
If E ∈ C0 (Rn , Sym) then by the coarea formula (Ambrosio et al. 2000, Sec-
tion 2.12) we have

Ω0
E · Ts |∇ϕ| dLn = R ϕ−1 (λ) E · Ts dHn−1 dλ

= Λ Rn E · dTλ dλ

= Rn E · dT.
We now assume that the loads depend affinely on the loading parameter
as in (32), thus we have loads (sλ , bλ ) defined for each λ ∈ R. It may happen
that there exists an interval (λ−, λ+) such that each load (sμ , bμ ), with μ
from this interval, is weakly equilibrated by a stress field Tμ ∈ M(Ω, Sym)
in such a way that
{Tμ , μ ∈ Λ} (49)
is an integrable parametric measure. In this situation, we have
62 M. Šilhavý
Remark 7.6. The measure U defined by

1 λ+ μ
U= T dμ
2 λ−
weakly equilibrates the loads (sλ , bλ ). However, depending on the nature of

the family (49), it may happen that the measure U is such that
U = ULn Ω U ∈ L2 (Ω, Sym− ),
i.e., the loads (sλ , bλ ) are strongly equilibrated by the stress field U. Such a
situation arises when the family (49) satisfies the hypothesis of Proposition
7.4 or Proposition 7.5, or is a sum of families such that one satisfies Propo-
sition 7.4 and the other Proposition 7.5. In many concrete cases, it is hard
to evalueate U explicitly but for the analysis it suffices to know that U is
represented by a square integrable function.
8 Integration with gravity

In this section we consider a rectangular panel made of a no-tension material
that is fixed at its base and subject to both the weight b which constitutes
the permanent part of the loads, and a horizontal compressive load sλ of
intensity λ, which is uniformly distributed on the right lateral side of the
panel and constitutes the variable part of the loads. Then, for every λ in an
appropriate interval (0, λc ), we determine a negative semidefinite and square
integrable stress field Tλ which is in equilibrium with the given loads and
we conclude that every λ ∈ (0, λc ) is statically admissible (see Definition
6.2). Let
Ω = (0, B) × (0, H) ⊂ R2
be the rectangular panel. We introduce the coordinate system x, y in R2
with the origin in the upper right corner of the panel and with the x axis
pointing to the left and the y axis pointing downward. We denote a general
point of Ω by r = (x, y) and the coordinate vectors along the axes x, y by
i, j, respectively. We put
D = (0, B) × {H}, S = ∂Ω \ D
and consider the loads (sλ , bλ ) where, for b > 0, bλ = bj in Ω, and for
r = (x, y) ∈ S and λ > 0,

λ
λi on {0} × (0, H),
s (r) = (50)
0 elsewhere.
The stress field Tλ will be constructed in two steps. Firstly, for every
λ ∈ (0, λc ) we determine a measure stress field Tλ ∈ M(Ω, Sym− ) that is
in equilibrium with the loads, i.e., such that

Ê(v) · dTλ = b v · j dL2 + λ v · i dH1
Ω Ω {0}×(0,H)
for every v ∈ V = W 1,2 (Ω, R2 ) (we recall that L2 stands for the Lebesgue
measure and H1 for the 1-dimensional Hausdorff measure in R2 ). This
expression is well defined because the loads are of the special form (18).
Secondly, we determine a square integrable negative semidefinite and equi-
librated stress field Tλ for every λ ∈ (0, λc ) by the integration procedure
described in Section 7, i.e. Tλ ∈ A(λ). This will prove that the loads
(sλ , bλ ) are strongly compatible for every λ ∈ (0, λc ) (see (6.2)).
To determine the stress measure Tλ , we use the results of the paper
(Lucchesi et al. 2009). We consider a smooth curve N λ which starts at the
upper right corner and ends at the bottom of the panel, and which is the
graph of an increasing function ω λ : [0, tλ ] → [0, H] to be specified below.
In this way, Ω is divided into the regions Ωλ+ (on the left) and Ωλ− (on the
right) by the curve N λ . We are looking for a weakly equilibrated measure
stress field Tλ which is absolutely continuous with respect to the Lebesgue
measure in Ωλ+ and Ωλ− with a piecewise continuously differentiable density
Tλr and has a concentration on N λ with a continuously differentiable density
Tλs , i.e.,
Tλ = Tλr L2 Ω + Tλs H1 N λ .
The equilibrium condition (41) implies that
div Tλr + bλ = 0 in Ω \ N λ . (51)
Furthermore, Tλs is superficial by Item (ii) of Example 7.1, which means
here
Tλs = σ λ tλ ⊗ tλ , (52)
where tλ is the unit tangent vector to N λ and σ λ is a scalar function on
N λ . By (42) we have
[Tλr ]m + div N Tλs = 0 on N λ (53)
where [Tλr ] is the jump discontinuity Tλr on N λ , m is the unit normal to
N λ pointing toward Ωλ+ .
Following the method presented in (Lucchesi et al. 2009) we obtain (see
(Lucchesi et al. 2009, Eqs. (4.6), (4.7) and (3.1)))

λ −byj ⊗ j if r ∈ Ωλ+ ,
Tr (r) = 2 2 (54)
−λi ⊗ i − 2bxi j − b λx j ⊗ j if r ∈ Ωλ−
64 M. Šilhavý
which satisfies the equilibrium equation (51) and the boundary conditions
(50). From (53) we deduce the equation of N λ (see (Lucchesi et al. 2009,
Eq. (4.16)) with λ = βH and p0 = 0)
√
ω λ (x) = cbx2 /λ, c = 1/2 + 3/6 (55)
which has the unit tangent vector
xi + 2yj
tλ (r) = (56)
x2 + 4y 2
Moreover, from (55) and (53) we obtain (see (Lucchesi et al. 2009, Eq.
(2.19)), with s0 = 0, and f (x, y) at the end of page 229)
√
3 2
σ (r) = −
λ
bx x + 4y 2 (57)
6
r ∈ N λ . If λ ∈ (0, λc ), with λc = cbB 2 /H, then N λ is contained in Ω,
except for the endpoints and the measure stress field Tλ is well defined by
relations (54)–(57).
The parametric measure Tλr L2 Ω is of the form considered in Propo-
sition 7.4 and the integrability condition (45) is satisfied because we have
λc
|Tλr (r)| dr dλ < ∞
0 Ω
Hence for 0 < λ < λc and > 0 such that

Λ = (λ − , λ + ) ⊂ (0, λc ) (58)
the measure
1
T̄λr = Tμr dμ (59)
2 Λ
is an absolutely continuous measure with respect to L2 Ω,
T̄λr = Uλr L2 Ω.
To compute Uλr , let us put
A = {r = (x, y) : bcx2 /y ∈ Λ}. (60)
We obtain
⎧
⎨ −byj ⊗ j
⎪ if r ∈ Ωλ+ \ A,
−1 2 2
Ur (r) = −λi ⊗ i − 2bxi j − (2) b x ln λ− j ⊗ j if r ∈ Ωλ− \ A,
λ λ+
⎪
⎩
(2)−1 (ξ1 (r)i ⊗ i + 2ξ2 (r)i j + ξ3 (r)j ⊗ j) if r ∈ A,
(61)
where
b2 c2 x4 1
ξ1 (r) = 2
− (λ + )2
2y 2
bcx2
ξ2 (r) = bx( − λ − ),
y

y(λ + )
ξ3 (r) = by(λ − ) − b2 x2 c + ln .
bcx2
In order to verify the first two regimes in (61) we note that if r ∈ Ωμ+ \ A or
r ∈ Ωμ− \A then, for all values of μ in Λ, the expression of Tμr (r) is that given
by (54)1 and (54)2 , respectively. Thus, (61)1 and (61)2 can be immediately
obtained from (55). For r ∈ A, we have r ∈ Ωμ+ for μ ∈ (λ − , bcx2 /y) and
r ∈ Ωμ− for μ ∈ (bcx2 /y, λ + ). Therefore
λ+
Uλr (r) = (2)−1 Tμr dμ
λ−
bcx2 /y
= (2)−1 λ− −byj ⊗ j dμ
λ+ 2 2

+ bcx2 /y (−μi ⊗ i − 2bxi j − b μx j ⊗ j) dμ
2 2 4 2
= (2)−1 b 2y c x
2 − 12 (λ + )2 i ⊗ i + 2bx bcxy −λ− ij

+ by(λ − ) − b2 x2 c + ln y(λ+)
bcx2 j⊗j .
The density Uλr is bounded in Ω (we note that for r∈ A we have λ − <
cbx2 /y < λ + by (56)).
Next we consider the measures Tλs . Let ϕ : Ω → R be defined by
ϕ(r) = cbx2 /y, (62)
r = (x, y) ∈ Ω. Then, for any λ ∈ (0, λc ), the curve N λ is the level set of ϕ
corresponding to the value of λ, i.e.
N λ (r) = {r ∈ Ω : ϕ(r) = λ}.
Moreover, ϕ is continuously differentiable and

cbx x2 + 4y 2
|∇ϕ(r)| = (63)
y2
Firstly we note that
√
3 2 2 x4
|∇ϕ(r)||Tλs (r)| = |∇ϕ(r)||σ (r)| =
λ
c b (4x2 + 2 )
6 y
66 M. Šilhavý
is bounded in
Ω0 = {r = (x, y) : cbx2 /y ∈ (0, λc )} = ϕ−1 (0, λc ),
in view of (63), (52) and (57). Then, Proposition 7.5 says that for any
interval Λ as in (58) the measure

1
T̄λs = Tμ dμ
2 Λ s
is L2 absolutely continuous over Ω, i.e.,
T̄λs = Uλs (r)L2 Ω0 ,
with the density given by (47), i.e.

λ (2)−1 Tλs (r)|∇ϕ(r)| if r ∈ A,
Us (r) = (64)
0 otherwise.
Note that ϕ(r) ∈ Λ if and only if r ∈ A, by (62) and (60). In the present
case we have
Tλs (r) |∇ϕ(r)| = σ λ (r)tλ (r) ⊗ tλ (r) |∇ϕ(r)|
√
3cb2 x2
=− 6y 2 (xi + 2yj) ⊗ (xi + 2yj)
√ 2 2
3cb x
=− 6y 2 (x2 i ⊗ i + 4xyi j + 4y 2 j ⊗ j)
r ∈ A, by (63), (52) and (57).
Finally, we obtain the negative semidefinite and square integrable (in
fact bounded) stress field Uλ = Uλr + Uλs ,
⎧
⎨ −byj ⊗ j
⎪ if r ∈ Ωλ+ \ A,
−1 2 2
U (r) = −λi ⊗ i − 2bxi j − (2) b x ln λ− j ⊗ j if r ∈ Ωλ− \ A,
λ λ+
⎪
⎩
s(r) if r ∈ A,
where
2 4
1
s(r) = −(2)−1 − 12y
b x
2 + 2 (λ + )
2
i⊗i
2 3
+ 2 − b3yx + bx(λ + ) i j
√
+ ( 23 + 56 + ln y(λ+) 2 2
bcx2 )b x − by(λ − ) j ⊗ j
by (61) and (64). It is an easy matter to verify that, for every λ ∈ (0, λc ),
Uλ verifies the equilibrium equation div Uλ + bj = 0 in Ω and the boundary
conditions (50), so that λ is statically admissible and the loads (sλ , bλ ) are
strongly compatible.
Acknowledgment. The research of M. Šilhavý was supported by RVO:

67985840. The support is gratefully acknowledged.
Bibliography
[Adams & Fournier 2003] Adams, R. A.; Fournier, J. J. F. 2003. S obolev
spaces (Second edition). New York: Academic Press.
[Ambrosio et al. 1997] Ambrosio, L.; Coscia, A.; Dal Maso, G. 1997. Fine
properties of functions with bounded deformation. Arch. Rational
Mech. Anal., Vol. 139: 201–238.
[Ambrosio et al. 2000] Ambrosio, L.; Fusco, N.; Pallara, D. 2000. F unctions
of bounded variation and free discontinuity problems. Oxford: Claren-
don Press.
[Angelillo 1993] Angelillo, M. 1993. Constitutive relations for no-tension
materials. M eccanica (Milano), Vol. 28: 195–202.
[Anzellotti 1985] Anzellotti, G. 1985. A class of convex non-corecive func-
tionals and masonry-like materials. Ann. Inst. Henri Poincaré, Vol. 2:
261–307.
[Barsotti & Vannucci 2013] Barsotti, R.; Vannucci, P. 2013. Wrinkling of
Orthotropic Membranes: An Analysis by the Polar Method. J. Elas-
ticity. To appear. DOI 10.1007/s10659-012-9408-z
[Del Piero 1989] Del Piero, G. 1989. Constitutive equations and compat-
ibility of the external loads for linear elastic masonry-like materials.
M eccanica, Vol. 24: 150–162.
[Del Piero 1998] Del Piero, G. 1998. Limit analysis and no-tension materi-
als. I nt. J. Plasticity, Vol. 14: 259–271.
[Di Pasquale 1984] Di Pasquale, S. 1984. Statica dei solidi murari teorie ed
esperienze. Dipartimento di Costruzioni, Pubblicazione n. 27
[Ekeland & Temam 1999] Ekeland, I.; Temam, R. 1999. C onvex analysis
and variational problems. Philadelphia: SIAM.
[Epstein 2002] Epstein, M. 2002. From Saturated Elasticity to Finite Evo-
lution, Plasticity and Growth. M athematics and mechanics of solids,
Vol. 7: 255–283.
[Fonseca & Leoni 2007] Fonseca, I.; Leoni, G. 2007. M odern Methods in
the Calculus of Variations: Lp Spaces. New York: Springer.
[Giaquinta & Giusti 1985] Giaquinta, M.; Giusti, E. 1985. Researches on
the equilibrium of masonry structures. Arch. Rational Mech. Anal.,
Vol. 88: 359–392.
[Gurtin 1981] Gurtin, M. E. 1981. An introduction to continuum mechanics.
Boston: Academic Press.
68 M. Šilhavý
[Lucchesi et al. 1994] Lucchesi, M.; Padovani, C.; Pagni, A. 1994. A numer-
ical method for solving equilibrium problems of masonry–like solids.
M eccanica, Vol. 24: 175–193.
[Lucchesi et al. 1996] Lucchesi, M.; Padovani, C.; Zani, N. 1996. Masonry–
like materials with bounded compressive strength. I nt J. Solids and
Structures, Vol. 33: 1961–1994.
[Lucchesi et al. 2008a] Lucchesi, M.; Padovani, C.; Pasquinelli, G.; Zani,
N. 2008. M asonry Constructions: Mechanical Models and Numerical
Applications. Berlin: Springer.
[Lucchesi et al. 2010] Lucchesi, M.; Padovani, C.; Šilhavý, M. 2010. An en-
ergetic view on the limit analysis of normal bodies. Quart. Appl. Math.,
Vol. 68: 713–746.
[Lucchesi et al. 2006] Lucchesi, M.; Šilhavý, M.; Zani, N. 2006. A new class
of equilibrated stress fields for no–tension bodies. J ournal of Mechanics
of Materials and Structures, Vol. 1: 503–539.
[Lucchesi et al. 2008b] Lucchesi, M.; Šilhavý, M.; Zani, N. 2008. Integration
of measures and admissible stress fields for masonry bodies. J ournal of
Mechanics of Materials and Structures, Vol. 3: 675–696.
[Lucchesi et al. 2009] Lucchesi, M.; Šilhavý, M.; Zani, N. 2009. Equili-
brated divergence measure stress tensor fields for heavy masonry bod-
ies. E uropean Journal of Mechanics A/Solids, Vol. 28: 223–232.
[Lucchesi et al. 2011] Lucchesi, M.; Šilhavý, M.; Zani, N. 2011. Integration
of parametric measures and the statics of masonry panels. Annals of
Solid and Structural Mechanics, Vol. 2: 33–44.
[Lucchesi et al. 2012] Lucchesi, M.; Šilhavý, M.; Zani, N. 2012. On the
choice of functions spaces in the limit analysis for masonry bodies.
J ournal of Mechanics of Materials and Structures, Vol. 7: 795–836.
[Müller 1999] Müller, S. 1999. Variational Models for Microstructure and
Phase Transitions. Pp. 85–210 in S. Hildebrandt, M. Struwe (ed.),
In Calculus of variations and geometric evolution problems (Cetraro,
1996) Lecture notes in Math. 1713 Springer, Berlin.
[Padovani et al. 2008] Padovani, C.; Pasquinelli, G.; Šilhavý, M. 2008. Pro-
cesses in masonry bodies and the dynamical significance of collapse.
M ath. Mech. Solids, Vol. 13: 573–610.
[Rockafellar 1970] Rockafellar, R. T. 1970. C onvex analysis. Princeton:
Princeton University Press.
[Rudin 1970] Rudin, W. 1970. Real and complex analysis. New York:
McGraw-Hill.
[Šilhavý 2008] Šilhavý, M. 2008. Cauchy’s stress theorem for stresses rep-
resented by measures. C ontinuum Mechanics and Thermodynamics,
Vol. 20: 75–96.
[Šilhavý 2013] Šilhavý, M. 2013. Collapse mechanisms and the existence of

equilibrium solutions for masonry bodies. To appear in Mathematics
and Mechanics of Solids.
[Temam 1983] Temam, R. 1983. P roblémes mathématiques en plasticité.
Paris: Gauthier–Villars.
[Temam & Strang 1980] Temam, R.; Strang, G. 1980. Functions of
Bounded Deformation. Arch. Rational Mech. Anal., Vol. 75: 7–21.
A numerical method for solving BVP of
masonry-like solids
Massimiliano Lucchesi*
*
Dipartimento di Ingegneria Civile e Ambientale
University of Florence
via S. Marta 3, 50139 Florence
1 Introduction
In this chapter, we shall first explicitly determine the solution to the con-
stitutive equation of isotropic masonry-like materials (i.e., find the stress
tensor T and the fracture strain Ef corresponding to a prescribed strain
tensor E, so as to satisfy relations (2.4)). The problem can be solved by
observing that the stress in the isotropic case is coaxial not only with the
fracture strain, but with the elastic strain as well. This enables representing
all these tensors with respect to the same principal system and then express-
ing the constitutive equation as a function of their eigenvalues. In this way,
(2.4) can be transformed into a linear complementarity problem (7), (27),
(28) whose solution is unique because the tensor of the elastic constants has
been assumed to be positive definite. As the solution to the constitutive
equation depends on the number of principal stresses that vanish, in order
to construct the stress function we need to consider a partition of the strain
space into four different regions, each of which corresponds to different ma-
terial behavior. We then calculate the derivative of the stress function with
respect to the strain, as this will be used to construct the tangent stiffness
matrix when dealing with numerical solution of the equilibrium problem.
The derivative of the stress function turns out to be smooth in each region
and its jump has no tangential component at the interfaces between the
different regions.
In Section 2 we consider the equilibrium problem of a body made of
masonry-like material. We limit ourselves to considering loads that are
square integrable functions (i.e. measures that are absolutely continuum
with a square integrable density). Thus, we can suppose that the displace-
ment field belongs to the Sobolev space V of all vector fields which are
square integrable with their first weak derivative and satisfy the prescribed

72 M. Lucchesi
boundary conditions. Then, for given loads, solving the equilibrium prob-
lem involves finding an equilibrium state, i.e., a displacement field in V
which satisfies the equation of virtual work for every displacement field in
V (see (22) below). As already established in the previous chapter, for gen-
eral loads the equilibrium state cannot exist and, if it does exist it cannot
be unique; on the other hands, the stress fields corresponding to different
equilibrium states coincide. We say that the loads are admissible if there
exists at least one equilibrium state, and that all such equilibrium state
are regular admissible if they are admissible and the corresponding stress
field is continuously differentiable. Because numerical applications require
assigning the loads incrementally, we define a (regular) admissible loading
process, that is, a piecewise continuously differentiable path in the space of
loads which starts at the origin, ends at the assigned loads and whose points
are (regular) admissible loads. Because masonry-like material is hyperelas-
tic, it turns out that the solution is independent of the choice of the process,
at least for the stress. Next, we briefly describe a numerical method for
solving the equilibrium problem. The approach is based on a displacement
formulation of the finite element method (1), (5), (6) and uses the Newton
Raphson method to solve the non-linear algebraic system that is obtained
by the discretization. (See also (8) and (2) for different approaches).
In Section 3, in order to make a comparison between the explicit and
numerical solutions, we consider a trapezoidal panel that is fixed at its base
and whose horizontal displacements are prevented at its vertical lateral side.
The panel undergoes normal and tangential loads that are distributed on its
top (see Fig. 1 below). Due to the constraint conditions used, in this case
the solution is unique also in terms of displacement and strain. This makes
the comparison between explicitly and numerically determined displacement
fields meaningful (see Figures 2 and 3 below).
With the aim of illustrating the possible applications of the constitutive
equation of masonry-like materials and the numerical method presented in
the foregoing, Section 4 reports on a case study. The cross vault of the
Church of San Ponziano in Lucca was analysed using the NOSA code (16)
at the Laboratory of Mechanics of Materials and Structures of the ISTI-
CNR Institute (25).
Lastly, Section 5 addresses the study of the dynamic behavior of slen-
der masonry structures. In order to overcome the difficulties inherent in
performing dynamic analyses of continuous bodies made of masonry-like
materials, a constitutive equation for masonry structures that can be rep-
resented by one-dimensional elements has been formulated (4), (30). By
assuming that the longitudinal fibers of the beam have no resistance under
tension and bounded compressive strength, the generalized stresses (normal
A Numerical Method for Solving BVP… 73
force and bending moment) are expressed as functions of the generalized

strains (extensional strain and the change of curvature). It has been proved
that, under suitable hypotheses on the range of the corresponding stress
field, the equilibrium state turns out to be unique (21). In order to study
the dynamic behavior of age-old masonry towers, the constitutive equation
has been formulated for hollow, rectangular cross-sectional beams (19). As
the application example shows, the effects of horizontal and vertical acceler-
ations applied to the base of the tower can be evaluated in terms of crushed
and cracked volume.
1.1 Explicit expression for the stress function for isotropic ma-
sonry-like materials and its derivative
In this Section we study the constitutive equation of masonry-like mate-
rials in order to characterize the mechanical behavior of a class of isotropic
materials that are incapable of withstanding tensile stresses and have in-
finite compressive strength. See (20), Chap. 2 for a constitutive equation
that takes into account the limited compressive strength of the material.
We say that a masonry-like material, whose constitutive equation is de-
fined in (2.4), is isotropic if the tensor of the elastic constant C is isotropic,
i.e., if

C QEQT = QC (E) QT
for all orthogonal tensors Q. In the following we suppose that the material
is isotropic and that (2.1) is satisfied. Thus, we have (13) p. 235
T = CEe = 2μE e + λ(trEe )1 (1)

where the Lamé moduli of the material satisfy the conditions
μ > 0, 2μ + 3λ > 0. (2)

We say that two tensors A, B ∈Sym are coaxial if they have the same
eigenvectors; as has been proved in (13), p. 12, this is the case if and only
if they commute, i.e. if AB = BA.
The following proposition is the basis for explicit solution of the consti-
tutive equation (2.4); its simple proof can be found in (20), p.22.
Proposition 1.1. If C is isotropic, then E, T, Ee and Ef are coaxial.
Let {q1 , q2 , q3 } be orthonormal eigenvectors of E, T, Ee and Ef and
e= (e1 , e2 , e3 ), t= (t1 , t2 , t3 ), ee = (ee1 , ee2 , ee3 ) and ef = (ef1 , ef2 , ef3 ) the lists
of the corresponding eigenvalues. Therefore, system (2.4) can be rewritten
as
74 M. Lucchesi
t = Dee ,
t ≤ 0,
ef ≥ 0,
t · ef = 0, (3)
where
⎛ ⎞
2μ + λ λ λ
D=⎝ λ 2μ + λ λ ⎠. (4)
λ λ 2μ + λ
This is a linear complementarity problem which, given e, has a unique
solution t and ef , because D is positive definite by (2) (see (7)).
In order to write the explicit solution of system (3), we define
R = {e = (e1 , e2 , e3 ) ∈ R3 : e1 ≤ e2 ≤ e3 } (5)
and consider the partition of R made up of the following four subsets
R1 = {e ∈ R : 2μe3 + λ(e1 + e2 + e3 ) ≤ 0},

R2 = {e ∈ R : e1 ≥ 0},
R3 = {e ∈ R : e1 < 0, λe1 + 2(λ + μ)e2 ≥ 0},
R4 = {e ∈ R : λe1 + 2(λ + μ)e2 < 0, 2μe3 + λ(e1 + e2 + e3 ) > 0}. (6)
If e∈ R1 , the normal component of the stress is compressive in all direc-
tions and the material behaves in a linear elastic manner. In R2 the stress
tensor is zero and the material can fracture in all directions. In R3 and R4
there are two and one principal stresses that are zero, respectively, and the
material can fracture orthogonally to these directions. Therefore, we have
if e∈ R1 then
ti = 2μei + λ(e1 + e2 + e3 ) efi = 0, (7)

if e∈ R2 then
ti = 0, efi = ei , i = 1, 2, 3; (8)
if e∈ R3 then
μ(2μ + 3λ)
t1 = e1 , ef1 = 0,
λ+μ
λ
ti = 0, efi = ei + e1 , i = 2, 3; (9)
2(λ + μ)
if e∈ R4 then
2μ
t1 = (λe2 + 2(λ + μ)e1 ), ef1 = 0,
λ + 2μ
2μ λ
t2 = (λe1 + 2(λ + μ)e2 ), ef2 = 0, t3 = 0, ef3 = e3 + (e1 + e2 ).
λ + 2μ 2(λ + μ)
(10)
%3
Thus, given a symmetric tensor E= i=1 ei qi ⊗qi , we can calculate the
corresponding stress
3

&
T = T(E) = ti q i ⊗ q i (11)
i=1
and the corresponding fracture strain

3

Ef = efi qi ⊗ qi . (12)
i=1
& : Sym →Sym− is isotropic, i.e.,

Note that stress function T
&
T(QEQ T &
) = QT(E)Q T
. (13)
1&
&
Moreover, for the stored energy function w(E) = 2 T(E) · E we have
(20), p. 30
if e ∈ R1 then
2 1
&
w(E) = μ E + λ(trE)2 ; (14)
2
if e ∈ R2 then
&
w(E) = 0; (15)
if e ∈ R3 then
1 μ(2μ + 3λ) 2
&
w(E) = e1 ; (16)
2 λ+μ
if e ∈ R4 then
2μ
&
w(E) = ((λ + μ)(e21 + e22 ) + λe1 e2 ). (17)
λ + 2μ
76 M. Lucchesi
& with respect

In order to compute the derivative of the stress function T
to E we need to compute the derivatives of the principal stresses ti with
respect to E. To this end, we first state the following proposition, whose
proof can be found in (20) , p. 16. Let Sym∗ be the subset of Sym made of
all symmetric tensors having distinct eigenvalues. Consider the functions
ai : Sym∗ → R, A → ai (A), i = 1, 2, 3
∗
which associate each A∈Sym with its eigenvalues, with a1 < a2 < a3 ,
denote by g1 , g2 , g3 a triad of corresponding normalized eigenvectors, and
put
1
Gij = (gi ⊗ gj + gj ⊗ gi ), i, j = 1, 2, 3.
2
Proposition 1.2. We have
DA ai = Gii i = 1, 2, 3,
2
DA Gii = Gij ⊗ Gij , i = 1, 2, 3.
ai − aj
i=j
Now we are in a position to give the explicit expression for the derivative
DE T& of T
& with respect to E in the interior R̊i of each region Ri . Let I and
O be the identity and the null fourth order tensor, respectively. Moreover,
let us put
1
Oij = (qi ⊗ qj + qj ⊗ qi ), i, j = 1, 2, 3.
2
& is differen-
Proposition 1.3. If e belongs to R̊i , i = 1, 2, 3, 4, then T
tiable with respect to E and we have
if e ∈ R1 then
& = 2μI + λ1 ⊗ 1;
DE T (18)
if e ∈ R2 then
& = O;
DE T (19)
if e ∈ R3 then

& = 2μ(2μ + 3λ) O11 ⊗ O11 − e1 O12 ⊗ O12
DE T
λ+μ e2 − e1
e1
− O13 ⊗ O13 ; (20)
e3 − e1
if e ∈ R4 then
& = 4μO12 ⊗ O12 − 4μ 2(λ + μ)e1 + λe2

DE T O13 ⊗ O13
λ + 2μ e3 − e1
4μ 2(λ + μ)e2 + λe1 2μ(3λ + 2μ)
− O23 ⊗O23 + (O11 +O22 )⊗(O11 +O22 )
λ + 2μ e3 − e2 λ + 2μ
+2μ(O11 − O22 ) ⊗ (O11 − O22 ). (21)
For the proof, which follows from (1.11), (1.7)-(1.10) and Proposition
1.2, see [20, p. 45].
& can be found in (20). In particular, it can be
Further properties of DE T
proved that on the interfaces between the different regions Ri , i = 1, 2, 3,
4 the jump of the derivative DE T& has no tangential component (see (20),
Prop. 2.22).
1.2 The equilibrium problem and a numerical method for its

solution
Let Ω ⊂ R3 be a continuous body represented by an open and bounded
set with Lipschitz boundary (9), p. 38 and let D, S be two disjoint subsets of
∂Ω such that D ∪ S = ∂Ω, which will be interpreted as the set of prescribed
boundary displacements and prescribed boundary forces. Moreover, let us
put
V = {v ∈ W 1,2 (Ω, R3 ) : v = 0 a.e. on D},

and for any v ∈ V let
& 1
E(v) = (∇v + ∇vT )
2
be the corresponding infinitesimal strain tensor. We assume that the body
force b acting in Ω and the surface force s acting on S are square integrable
functions as in (2.18), i.e.,
b ∈ L2 (Ω, R3 ), s ∈ L2 (S, R3 )
and call (b,s) the load applied to Ω. Thus, the internal energy of the body

E(u) = w( &
& E(u))dL 3
,
Ω
the energy of the load

78 M. Lucchesi

3
< l, u >= b · u dL + s · u dH2
Ω S
and the total energy of deformation

F(u) = E(u)− < l, u >= &
& E(u))dL
w( 3
− 3
b · udL − s · u, dH2
Ω Ω S
can be defined for each u∈ V .

The equilibrium problem for the masonry-like body Ω with stress func-
& under the load (b,s), is to find a displacement field u∈ V such
tion T,
that

& E(u))
T( & &
· E(v)dL3
=< l, v > (22)
Ω
for each v ∈ V . Such a displacement field, if it exists, is called an equilibrium

state of Ω.
As already noted in Chapter 2, Sec 3, if u satisfies (22) and it happens
& E(u))
that T( & ∈ C 1 (clΩ,Sym), then we have
divT + b = 0 in Ω and Tn = s on S (23)

& E(u))
where T = T( & and n is the outer normal to ∂Ω. Indeed, from formula
(13)
T · ∇v = div(TT v) − v · divT,
the symmetry of T, and because v= 0 on D, by using the divergence theo-
rem, we obtain

& 3
T · E(v)dL = T · ∇vdL3 =
Ω Ω

div(Tv)dL3 − v · divTdL3 =
Ω Ω

2
Tn · vdH − v · divTdL3 . (24)
S Ω
By comparing (22) and (24) we obtain equation (23) due to the arbi-
trariness of v.
In general, the existence of an equilibrium state is not guaranteed for
equilibrium problems of masonry-like solids (see Chapter 2, Sec 3 for the
existence theory). Moreover, it may happen that there more than one equi-
librium state exists. However, if u1 and u2 are two equilibrium states then
& E(u
T( & E(u
& 1 )) = T( & 2 )). (25)
This follows from the fact that if u is an equilibrium state, then T( & E(u))
&
is a minimum point of the complementary energy functional among all ad-
missible stress fields equilibrating the loads, and this minimum point is
unique, as the complementary energy functional is convex. This has been
proved in Chapter 2, Theorem 3.3 (see also (20), Prop. 3.1 for a simple
direct proof of (2.4)). We say that the load (b, s) is admissible if there
is at least one equilibrium state, that is, a displacement field u which
satisfies (2.1); we moreover say that (b, s) is a regular admissible load if
& E(u))
T( & ∈ C 1 (clΩ,Sym).
In order to solve the equilibrium problem by using the finite element
method, we have to assign the load incrementally. Therefore, given a regular
admissible load (b, s), we consider a piecewise continuously differentiable
loading process τ →(b(τ ), s(τ )), where τ ∈ [0, τ̄ ] is a loading parameter,
such that (b(0), s(0))= (0, 0), (b(τ̄ ), s(τ̄ ))=(b, s), (b(τ ), s(τ )) is a regular
admissible load for each τ ∈ [0, τ̄ ] and, moreover, the map τ → T( & E(u(τ
& )))
is piecewise continuously differentiable. We call a map τ →(b(τ ), s(τ )) with
such characteristics a regular loading process.
Given a regular loading process, from (22) we deduce that

& E(u(τ
T( & &
))) · E(v)dL3
= b(τ ) · vdL + 3
s(τ ) · vdH2 (26)
Ω Ω S
for each v ∈ V and τ ∈ [0, τ̄ ]. We assume the initial conditions
u(x, 0) = 0, E(x, 0) = 0, T(x, 0) = 0 on Ω.

By differentiating (26) with respect to τ , we obtain the incremental
equilibrium equation

& E(u))[
DE T( & &
Ė(u)] · E(v)dL 3
= ḃ · v, dL3 + ṡ · vdH2 (27)
Ω Ω S
where a dot denotes differentiation with respect to τ and Ė(u) = 12 (∇u̇ +

∇u̇T ). Note that, in view of (23), variational equation (27) is equivalent to
& E(τ
Ṫ = DE T( & )[Ė(τ )],
80 M. Lucchesi
divṪ + ḃ = 0 on Ω
u̇ = 0 on D,
Ṫn = ṡ on S.
It is easy to see that, given a regular admissible load (b, s), the solution
to the equilibrium problem does not depend on the (regular) loading process,
at least with regard to the stress.
By using the finite element method (see e.g., (3), (5), (6), (24)), the
equilibrium equation (26) and the incremental equilibrium equation (27)
are transformed into the non-linear algebraic equilibrium system
[K{u}]{u} = {f } (28)
and into the non-linear evolution system
[KT [{u}]{u̇} = {ḟ } (29)

where {u} and {f } are the arrays of the nodal displacements and the equiv-
alent nodal forces, respectively, and {u̇} and {ḟ } are their derivatives with
respect to τ . K and KT are the stiffness and tangent stiffness matrix,
which, as usual, are obtained from the assembly of the corresponding ele-
mental matrices. The elemental tangent stiffness matrices are computed by
using the derivatives of the stress function (18)-(21) (see also (20) for the
plane cases). For each load increment, equilibrium is reached by following
a suitable iterative procedure. In the following we describe the Newton-
Raphson procedure. Let us suppose that in correspondence to τ ∈ (0, τ̄ ),
equation (26) holds and the body is thus in equilibrium under the load
(b(τ ), s(τ )). We assign a nodal load increment {Δf } and have to solve the
system
[KT ]{Δu} = {Δf }. (30)

(i) (i)
Let {u }, {Δu } be the nodal vectors corresponding to the i − th
iteration, and denote by {f (i) } the nodal forces corresponding to {Δf } for
i = 0, and the vector of residual load if i > 0. We solve the linear system
[KT {u(i) }]{Δu(i) } = {f (i) } (31)

in order to determine the displacement {u(i+1) } = {u(i) } + {Δu(i) } corre-
sponding to the (i + 1) − th iteration, and the tangent matrix [KT {u(i+1) }]
that can be used in the next iteration or in the next load increment.
In order to avoid the occurence of null pivot during solution of system
& E(u))
(31), at the integration points at which DE T( & vanishes (see (19)) we
use (18), with the right-hand side multiplied by a small positive number
(see (1) for a slightly different technique).
After solving system (31), we calculate the vector of the residual load
{f (i+1) } = {f̄ } − [K{u(i+1) }]{u(i+1) }

where {f̄ } is the vector of overall loads assigned up to the current iteration,
including the reaction forces. We then perform a convergence check
' '
' (i+1) '
'{f }'
≤ξ (32)
{f̄ }
where ξ is a suitable parameter. If (32) is verified, we assign a new load
increment, otherwise we solve system (31) with the vector of residual load
{f (i+1) }. The convergence check (32) guarantees that the calculated stress
field is (negative semidefinite) and in equilibrium with the assigned load.
Before closing this section we briefly describe how this numerical tech-
nique can be applied to the study of masonry vaults. Let us consider a
vault with (not necessarily constant) thickness h and denote by S the mean
surface with Gaussian coordinates η1 and η2 , corresponding unit tangent
vectors g1 and g2 and normal unit vector n. Let ζ ∈[−h/2, h/2] be the
coordinate in the normal direction n. We assume that the stress tensor
T = T(η1 , η2, ζ) is negative semidefinite and satisfies the condition
T(η1 , η2 , ζ)n(η1 , η2 ) = 0. (33)

For each (η1 , η2 ) ∈ S we define the tensors
h/2 h/2
N(η1 , η2 ) = T(η1 , η2, ζ)dζ, M(η1 , η2 ) = T(η1 , η2, ζ)ζdζ. (34)
−h/2 −h/2
Moreover, for each γ ∈[−π/2, π/2) let g(γ) be the unit tangent vector
to S which forms an angle γ with g1 . Thus,
N (η1 , η2 , γ) = g(γ) · N(η1 , η2 )g(γ),

M (η1 , η2 , γ) = g(γ) · M(η1 , η2 )g(γ) (35)
are the normal force and bending moment per unit length corresponding to
the direction γ. Because T is negative semidefinite, it is an easy matter to
verify that N is non-positive and that M satisfies the condition
h
|M (η1 , η2 , γ)| ≤ |N (η1 , η2 , γ)| . (36)
2
82 M. Lucchesi
Now let us consider the subset S of S , where N= 0, and for each
γ ∈[−π/2, π/2) let us consider the corresponding eccentricity
M (η1 , η2 , γ)
e(η1 , η2 , γ) = . (37)
N (η1 , η2 , γ)
If, for some γ, we have N (η1 , η2 , γ) = 0, the corresponding eccentricity
is not well-defined, but it can be proved that in this case e(η1 , η2 , γ) = α is
a constant function of γ. Thus, e(η1 , η2 , γ) can be extended by continuity
(20), Prop. 5.1. Let (η1 , η2 ) ∈ S and ē(η1 , η2 ) be such that
|ē(η1 , η2 )| = sup{|e(η1 , η2 , γ)| : γ ∈ [−π/2, π/2)}.

The surface
M = {p : p = (η1 , η2 ) + ē(η1 , η2 )n, (η1 , η2 ) ∈ S }

is called the maximum modulus eccentricity surface (m.m.e.s.). It can be
proved that (20), Prop. 5.2
⎧
⎪
⎨ω1 if detN(η1 , η2 ) = 0 and |ω1 | ≥ |ω2 | ,
ē(η1 , η2 ) = ω2 if detN(η1 , η2 ) = 0 and |ω2 | ≥ |ω1 | , (38)
⎪
⎩
α if detN(η1 , η2 ) = 0,
where, for β =tr(N−1 M)(detN),

β− β 2 − 4 (detN) (detM) β 2 − 4 (detN) (detM)
β+
ω1 = , ω2 = .
2detN 2detN
(39)
Note that, in view of inequality (36), the m.m.e.s corresponding to a
negative semidefinite stress field is entirely contained within the vault. The
m.m.e.s. allows concise, effective rendering of the results of finite elements
analyses. In dealing with masonry vaults the m.m.e.s. plays a role analogous
to that of the line of thrust for arches (15), (16).
1.3 Comparison between the explicit and numerical solutions

Let b > 0 and α > 0. In the Cartesian reference system shown in
Figure 1, with i and j the unit vectors corresponding to the x and y axes,
respectively, let us consider a trapezoidal masonry panel
Ω = {r = (x, y) ∈ R2 : 0 < x < αy, b/2α < y < b/α}. (40)

x o
y
b
2α
p
αp
b
2α
b/2 b/2
Figure 1. Trapezoidal masonry panel.

84 M. Lucchesi
The panel is fixed at its base (0, b) × {b/α} and, moreover, horizontal
displacements are prevented on its right-hand side {0} × (b/2α, b/α). In
the absence of gravity, the panel is subjected to a uniform vertical load p
and a linear tangential load q(x) = 2αpx/b which are distributed on its
top (0, b/2) × {b/2α}. We now try to solve the BVP with a stress field
T : Ω →Sym− ,
T = σx i ⊗ i + τxy (i ⊗ j + j ⊗ i) + σy j ⊗ j, (41)
such that detT = 0. Thus, if we denote by κ the cotangent of the angle
between the active isostatic lines and the x axis, and σy by σ, we can write
(17)
T = κ2 σi ⊗ i + κσ(i ⊗ j + j ⊗ i) + σj ⊗ j, (42)
where κ and σ have to satisfy the system of PDE

κκ,x + κ,y = 0,
(43)
κσ,x + σκ,x + σ,y = 0
with the boundary conditions
κ(x, b/2α) = 2αx/b and σ(x, b/2α) = −p, (44)

for 0 < x < b/2.
From (43)1 and (44)1 we obtain
x
κ= (45)
y
and then, in view of (43)2 and (44)2 , we have
x 1
σ,x + σ + σ,y = 0 (46)
y y
whose solution is
pb
σ=− (47)
2αy
Therefore, we can write
pb
T=− (x2 i ⊗ i + xy(i ⊗ j + j ⊗ i) + y 2 j ⊗ j) =
2αy 3
pb
− r ⊗ r, (48)
2αy 3
with r = (x, y).

If we denote by
μ(2μ + 3λ) λ
E= and ν=
μ+λ 2(μ + λ)
the Young’s modulus and Poisson’ s ratio of the material, from (2) and (48)
we obtain the elastic strain
pb
Ee = − (r ⊗ r − νr⊥ ⊗ r⊥ )
2αEy 3
with r⊥ = (y, −x). Moreover, because the inequalities Ef ≥ 0, T ≤ 0
and the condition T · Ef = 0 imply TEf = 0, from (48) we deduce the
expression for the fracture strain
1
Ef = a(r)r⊥ ⊗ r⊥
y2
with a : Ω → R+ and we have
pb 1
E = Ee + Ef = − 3
(r ⊗ r − νr⊥ ⊗ r⊥ ) + 2 a(r)r⊥ ⊗ r⊥ . (49)
2αEy y
In order that the strain E satisfy the compatibility condition a, once it
has been expressed as a function of y and κ, has to satisfy the ODE (17)
2 φ
a,yy + a,y = 3 , (50)
y y
where
2pb(1 + 3κ2 )
φ= .
αE
From (50) we obtain
1
a(y, κ) = − (φ(1 + lny) + a0 (κ)) + a1 (κ) (51)
y
with a0 and a1 arbitrary functions.
Let u(r) = (u(r), v(r)) be the displacement field. If we require
u(b/α, x) = v(b/α, x) = 0, for 0 < x < b,
b b
u(0, y) = 0, for <y< ,
2α α
86 M. Lucchesi
Figure 2. Vertical displacement for y = 2 m and y = 3 m.
and, for the sake of simplicity we put ν = 0, then from (49) and (51) we
obtain
pbx(x2 + y 2 ) αy αy
u(r) = 3
(−2ln( ) + − 1),
αEy b b
pb(x2 + y 2 ) 1 αy αy
v(r) = 4
( (3x2 − y 2 )ln( ) + x2 (1 − )). (52)
αEy 2 b b
In order to compare the explicit and numerical solutions, the panel has
been discretized into 6400 four-node isoparametric rectangular finite ele-
ments and the following parameter values have been used.
b = 2m, α = .5, p = 103 P a, E = 2 · 109 P a, ν = 0. (53)

Figures 2 and 3 show the behavior of u and v, respectively, for y = 2 m
and 3 m.
Figure 3. Horizontal displacement for y = 2 m and y = 3 m.
1.4 Example applications

In the early 1990s, the numerical method described in Section 2 was
implemented in the finite element code named NOSA (16), (20), p. 145
by the Mechanics of Materials and Structures Laboratory of the ISTI-CNR
Institute. Since then, the code has been applied to the study of several
historical monuments (see (20), Ch. 7 and the bibliography for an account of
these applications). The results of these numerical analyses were visualized
using the graphic code MSC Mentat . More recently, within the framework
of the NOSA-ITACA project, funded by the Regione Toscana (23), the code
has been integrated with the open-source platform SALOME (29).
Herein, we briefly describe an analysis of the cross vault of the Church
of San Ponziano in Lucca. This study has been conducted by researchers
at the ISTI-CNR Institute, who kindly made their results available to us, a
contribution which we gratefully acknowledge (25).
The church (Fig. 4), which dates back to the 9th century, has recently
undergone a complex series of restorations and functional adjustments. In
88 M. Lucchesi
particular, the roofing structures, which had deteriorated over the centuries,
were consolidated and partially reconstructed. Structurally, the church is
constructed of load-bearing masonry with a rectangular plan divided into
three aisles. The central aisle (nave) in correspondence to the presbytery
area is covered by a constant-thickness cross, or groin, vault (Fig. 4), which
is the focus of the present study. Sometime in the past a number of small
solid-brick walls were built to support the beams, which therefore unloaded
the weight of the roofing, the beams themselves and associated joists onto
the cross vault, thereby giving rise to an asymmetrical distribution of con-
centrated loads.
The NOSA code has been used to study the compound structure made up
of the cross vault overlying the altar, which represents the intersection of two
toroidal surfaces, the arches delimiting it from the nave and lateral aisles,
and the wall overlying the barrel vault of the apse area. The dimensions of
the compound structure under examination are 9 m x 13.8 m in plan and 6.4
m in height. Static analyses of the structure were conducted by discretising
it into 6040 nonconforming “shell” elements (12), subjected to their own
weight and the concentrated loads on the extrados due to the walls, while
at the same time accounting for the presence of reinforcement chains in the
arches (fig. 7). The relevant values of Young’s modulus and Poisson’s ratio
have respectively been assumed to be 5 · 109 Pa and 0.1, while the specific
weight has been taken to be 2 · 104 N/m3 , and the thickness of the vault
and arches 0.25 m and 0.60 m, respectively.
Regarding the boundary conditions, the vault’s springers were considered
to be immovably wedged onto the supporting pillars, and no movement of
the arch extremities was moreover allowed in the direction of the generatices.
This last condition was imposed by assuming that the stiffness of the vault’s
supporting and surrounding structures is such that any movement of the
constraining bonds is severely limited.
Initially, an analysis was made of the structure subjected to its own
weight alone. Analysis of the displacement field reveals a lowering of the
vault in its central area and, regarding the arches, a lowering at the keystone
accompanied by bulging in the area of the haunches. In any event, the
relative displacement did not exceed 1.25 mm. An analysis of the entire
compound structure enables concluding that the compressive stresses are
below 13 · 105 Pa, a value considered acceptable. Analysis of the fracture
strain components moreover excludes the presence of cracking in the cross
vault.
Figures 8, 9 and 10 show plots of the maximum modulus eccentricities
surface respectively for the arch towards the nave, for one of the arches
towards a lateral aisle and for the cross vault. In all three cases, the surface
Figure 4. Transverse section of the Church of San Ponziano.
Figure 5. The roof vault of the presbytery area.

90 M. Lucchesi
Figure 6. Extrados of the cross vault.
remains well within the thickness of the considered structural elements,

thereby confirming the good overall static conditions of the vault.
Subsequently, a further analysis was conducted in order to evaluate the
effects on the displacement and stress fields in the structure of the metal
reinforcement chains in the arches toward the nave and the apse. The finite
elements analysis reveals that the presence of the chains, modelled via 3 x
3 cm cross-sectional steel “beam” elements, has no significant influence on
either the displacement or the stress field. The negligible effect of the chains
is also highlighted by the modest degree of tension that they exert, below
11000 N.
Lastly, an analysis was conducted to account for both the self-weight of
the structure, as well as the concentrated loads transmitted by the roof’s
load-bearing beams set on the walls arranged along the vault extrados. The
arrangement of the walls is shown in the plan view illustration in figure 11.
The analysis of the loads showed that each of the small walls transmits a
vertical load of 37500 N to the vault. All told, the load due to the presence
of the walls is about 12% of the weight of the entire structure, which is
equal to 1.59 106 N.
Figures 12 and 13 show plots of the vertical displacement of the vault
Figure 7. Finite element model of the cross vault and connected arches.
Figure 8. Maximum modulus eccentricities surface of the arch toward the

nave.
92 M. Lucchesi
Figure 9. Maximum modulus eccentricities surface of the arch toward a

lateral aisle.
Figure 10. Maximum modulus eccentricities surface of the cross vault.

Figure 11. Plan of the cross vault with the walls sustaining the roof beams.
subjected to its self-weight alone, and when subjected to its self-weight plus
the loads transmitted by the walls. A fair degree of variation is evident,
especially in correspondence to the group of walls indicated as muretto 1,
muretto 2 and muretto 3 in figure 11. In any case, the displacement values
are very small and their variation quite unlikely to cause any significant
changes in the fracture stress and strain fields. Moreover, once again under
these loading conditions, the maximum modulus eccentricities surface is
contained well within the thickness of the arches and the cross vault.
The study shows the potentials of the constitutive equation and nu-
merical techniques described in the foregoing for analysing the structural
behaviour of historical buildings and modelling any consolidation operations
before undertaking them.
1.5 A numerical method for dynamic analysis of slender masonry

structures
Due to the difficulties inherent in performing dynamic analyses of con-
tinuous masonry-like bodies (10), (11), recent research efforts have been
directed at developing simplified constitutive models for masonry structures
by representing them through one-dimensional elements. Such efforts have
94 M. Lucchesi
Figure 12. Vertical displacement (m) under self-weight alone.
Figure 13. Vertical displacement (m) under both self-weight and loads
transmitted by the walls.
given rise to a nonlinear constitutive equation for beams, which furnishes the
generalized stress (normal force and bending moment) as a function of the
generalized strain (extensional strain and curvature change of the beams
longitudinal axis), by making the assumption that the material does not
withstand tensile stresses in the longitudinal direction; variants of the model
take into account the cases of unbounded, as well as bounded compressive
strength (21).
This constitutive equation has revealed to be suitable for formulating and
refining a simple numerical model which, implemented in the finite-element
code MADY, developed at the Dipartimento di Costruzioni e Restauro,
University of Florence (22), render almost effortless to perform non-linear
dynamic analyses of slender masonry structures which exhibit primarily a
flexural behaviour. Specifically, the model has been developed for both solid
and hollow rectangular cross-sections in order to study masonry arches as
well as free-standing masonry towers (19), (26).
From a practical standpoints, the proposed model has shown to be partic-
ularly suitable for the seismic analysis of slender towers: although it repre-
sents towers as one-dimensional continuum, it is refined enough to reveal the
effects of the higher vibration modes that often affect them. Moreover, cou-
pling phenomena between transverse and axial vibrations, which are widely
recognized as an important factor in the seismic behaviour of slender tow-
ers, are also taken into account by the model, as they are embedded in the
constitutive equation. The vertical components of earthquake excitations,
often significant sources of additional damage to these structures, can also
be introduced as input ground motions.
Furthermore, since the model accounts for the material’s non-linear be-
haviour in all sections along the structure’s height, it can be useful to obtain
measures of local and global damage. Because the damage to such struc-
tures tends to be distributed along their height, rather than concentrated
at the base alone, such measures can be meaningful, even if the assumed
elasticity of the material does not allow for assessing the irreversibility of
the damage process.
Firstly, we introduce the constitutive equation formulated for hollow, rect-
angular cross-section beams made of a non-linear elastic material with no
resistance to tension and limited compressive strength σo .
By making the usual assumption of plane sections, according to the classi-
cal Euler-Bernoulli hypothesis, and accounting for axial stresses alone, the
relation between generalized stress and strain can be determined as follows.
For each longitudinal fiber, the considered constitutive equation is
96 M. Lucchesi
⎧
⎪
⎨σ0 if (ȳ) ≤ 0 ,
σ(ȳ) = E(ȳ) if 0 ≤ (ȳ) ≤ 0, (54)
⎪
⎩
0 if (ȳ) > 0,
where 0 = σ0 /E with E the Young modulus of the material. Thus, the

pattern of the axial stress component σ in any given transverse section of
the beam may be one of those described - with reference to the case M < 0,
- in Fig. 14, plus the trivial cases for which σ is equal to zero or to σo over
the entire section. The various cases differ from each other with regard to
the position of the neutral axis - denoted as yn - and the position of the axis
where the stress reaches its limit value σo - denoted as ys -, with respect to
the section geometry. In Case 1, neither axis crosses the section, and the
constitutive behaviour is linear elastic. In each of the Cases 2, 3 and 4, for
which yn crosses the section differently with respect to the hollow, cracking
of the section occurs. Conversely, in Cases 5, 6 and 7, ys crosses the section
in three different positions, and the corresponding portion of the section is
crushed. Lastly, the remaining cases represent all the possible situations in
which cracking and crushing occur at the same time.
By the Euler-Bernoulli hypothesis, the ordinate of both yn and ys can
be simply expressed as a function of and κ, that is
(yn ) = + κyn = 0, (ys ) = + κys = o . (55)

In order to define the constitutive relations, it is helpful to introduce the
non-dimensionalized strains η and χ, given by
κh
η= , χ= , (56)
o o
and to describe the geometric properties of the sections via the non-dimensionalized
parameters
h−t b−s
α= , β= (57)
h b
where b and 2h denote the section’s dimensions, while (b − s) and 2(h − t)
denote those of the hollow in the x and y directions, respectively (Fig. 14,
Case 1). Analogously, regarding the generalized stresses, it is useful to refer
to the non-dimensionalized quantities
2N 6M
n= m= . (58)
bhσo bh2 σo
Figure 14. σ patterns over the cross-section.

98 M. Lucchesi
Figure 15. Partition in the half-plane (η, χ).
In view of (55), we can define different regions in the plane (η, χ), in
which each of the illustrated σ patterns occur (Fig. 15). Due to the different
distributions of σ, distinct constitutive relations expressing n and m as a
function of η and χ hold for each region.
Thus, since from (55) and (56) we have
η 1−η
yn = − h, ys = h, (59)
χ χ
in view of relations

yn ≤ −h
(60)
ys ≥ h
identifying Case 1, the region of linear elastic behaviour is
E1 = {χ ≥ 0, χ ≤ η, χ ≤ 1 − η} .
In the same way for each of the twelve regions of non-linear behaviour
Ei , i = 2, ..13, we obtain

−h ≤ yn ≤ −αh η
Case 2 : E2 = χ ≥ 0, χ ≥ η, χ ≤ ,χ ≤ 1 − η
ys ≥ h α

−αh ≤ yn ≤ αh η η
Case 3 : E3 = χ ≥ 0, χ ≥ ,χ ≥ − , χ ≤ 1 − η
ys ≥ h α α

αh ≤ yn ≤ h η
Case 4 : E4 = χ ≥ 0, χ ≥ −η, χ ≤ − , χ ≤ 1 − η
ys ≥ h α

yn ≤ −h 1−η
Case 5 : E5 = χ ≥ 0, χ ≤ η, χ ≤ , χ≥1−η
αh ≤ ys ≤ h α

yn ≤ −h η−1 1−η
Case 6 : E6 = χ ≥ 0, χ ≤ η, χ ≥ ,χ≥
−αh ≤ ys ≤ αh α α

yn ≤ −h η−1
Case 7 : E7 = χ ≥ 0, χ ≤ η, χ ≥ η − 1, χ ≤
−αh ≤ ys ≤ αh α

−h ≤ yn ≤ −αh
Case 8 :
αh ≤ ys ≤ h

η 1−η
E8 = χ ≥ 0, χ ≥ η, χ ≤ , χ ≤ , χ≥1−η
α α

−h ≤ yn ≤ −αh
Case 9 :
−αh ≤ ys ≤ αh

η 1−η η−1
E9 = χ ≥ 0, χ ≥ η, χ ≤ , χ ≥ ,χ≥
α α α

yn ≥ −h η−1
Case 10 : E10 = χ ≥ 0, χ ≥ η, χ ≤
ys ≤ −αh α
100 M. Lucchesi

−αh ≤ yn ≤ αh
Case 11 :
αh ≤ ys ≤ h

η η 1−η
E11 = χ ≥ 0, χ ≥ , χ ≥ − , χ ≤ ,χ ≥ 1 − η
α α α

yn ≥ −αh η 1−η
Case 12 : E12 = χ ≥ 0, χ ≥ ,χ≥
ys ≤ αh α α

yn ≥ αh η
Case 13 : E13 = χ ≥ 0, χ ≤ − , χ ≥ 1 − η
ys ≤ h α
The partition of E − = {(η, χ) | χ ≤ 0} that corresponds to M ≥ 0 is

symmetric to that of E + with respect to η-axis.
For (η, χ) belonging to each region, n and m can be evaluated by means
of simple equilibrium considerations, and the following cases can be distin-
guished:
(η, χ) ∈ E1 : n = 4η(1 − αβ); m = 4χ(1 − α3 β).

η 2 +2ηχ (1−2αβ)+χ2 2χ3 (1−2α3 β)+3ηχ2 −η 3
(η, χ) ∈ E2 : n = χ ; m= χ2 .
η 2 (1 − β) + 2ηχ(1 − αβ) + χ2 (1 − α2 β)
(η, χ) ∈ E3 : n = ;
χ
2χ3 (1 − α3 β) + 3ηχ2 (1 − α2 β) − η 3 (1 − β)
m= .
χ2
(η + χ)2 2χ3 + 3ηχ2 − η 3
(η, χ) ∈ E4 : n = ; m= .
χ χ
(η − 1)2 + 2χ(2αβη − η − 1) + χ2
(η, χ) ∈ E5 : {n = − ;
χ
2χ3 (1 − 2α3 β) + 3χ2 (1 − η) + (η − 1)3
m= .
χ2
(η − 1)2 (β − 1) + 2χ(η + 1)(1 − αβ) − χ2 (1 − α2 β)
(η, χ) ∈ E6 : n = ;
χ
2χ3 (1 − α3 β) + 3χ2 (α2 β(η − 1) − η + 1) + (1 − β)(η − 1)3
m= .
χ2
2 2
−(η − 1) + 2χ(η + 1 − 2αβ) − χ
(η, χ) ∈ E7 : n = ;
χ
2χ3 + 3χ2 (1 − η) + (η − 1)3
m= .
χ2
2η(1 − 2αβχ) + 2χ − 1
(η, χ) ∈ E8 : n = ;
χ
−4α3 βχ3 + 3χ2 − 3η 2 + 3η − 1
m= .
χ2
βη 2 − 2η(αβχ + β − 1) + α2 βχ2 + 2χ(1 − αβ) + (β − 1)
(η, χ) ∈ E9 : n = ;
χ
−2α3 βχ3 + 3χ2 (α2 β(η − 1) + 1) − β(η − 1)3 − 3η 2 + 3η − 1
m= .
χ2
2η − 4αβχ + 2χ − 1 3χ2 − 3η 2 + 3η − 1
(η, χ) ∈ E10 : n = ; m= .
χ χ2
−βη 2 − 2η(αβχ − 1) − α2 βχ2 + 2χ − 1
(η, χ) ∈ E11 : n = ;
χ
−2α3 βχ3 − 3χ2 (α2 βη − 1) + βη 3 − 3η 2 + 3η − 1
m= .
χ2
2η(1 − β) − 2χ(αβ − 1) − (1 − β)
(η, χ) ∈ E12 : n = ;
χ
3χ2 (1 − α2 β) + (β − 1)(3η 2 − 3η + 1)
m= .
χ2
2η + 2χ − 1 3χ2 − 3η 2 + 3η − 1
(η, χ) ∈ E13 : n = ; m= . (61)
χ χ2
Once all the expressions for χ > 0 have been defined, the corresponding
ones for χ < 0 can be easily obtained by using the appropriate symmetry
properties.
It should be noted that for α or β equal to zero, i.e., when the section is solid,
the constitutive equation defined above corresponds to that formulated in
(30) for a rectangular cross-section beam. Moreover for σo → ∞, we ob-
tain the constitutive equation for no-tension beam with infinite compressive
strength.
Regarding the admissible domain in the plane (n, m), it is a simple matter
to verify that in order for the axial component σ of the stress to satisfy
the inequality σo ≤ σ ≤ 0 at each point in the section, n must belong to
102 M. Lucchesi
Figure 16. Admissible range in the (n, m) plane.
the interval 0 ≤ n ≤ 4(1 − αβ). For each value of n, m must take on

an expression that depends on the relation between the σ pattern and the
section’s geometry. Precisely, it assumes three different expressions, defining
curves Γ1 , Γ2 , and Γ3 which, as shown in Fig. 16, bound the admissible
region Ω+ = {(n, m) 0 ≤ n ≤ 4(1 − αβ), m ≥ 0}.
Furthermore, the partition defined in region E + determines a partitioning of
Ω+ into thirteen regions, where the stress distributions are those identified
by the thirteen different Cases in Fig. 14.
Equations of curves which bound regions shown in Fig. 16 have been omit-
ted here for the sake of brevity.
Now, we briefly describe the method used to numerically integrate equa-
tions of motion considering the constitutive relation presented in the fore-
going.
The well-known equations for transverse and axial vibrations of a beam
(which as formulated herein includes the axial-force effects) are
∂2v ∂ ∂M ∂v ∂ 2 u ∂N
m 2
+ ( + N ) − q = 0,m 2 − −p=0 (62)
∂t ∂z ∂z ∂z ∂t ∂z
where z denotes the abscissa along the beam’s axis, t the time; M (z, t)
and N (z, t) denote the bending moment and axial forces, v(z, t) and u(z, t)
the transverse and longitudinal displacements, q(z, t) and p(z, t) the trans-
verse and axial distributed loads, and m the mass per unit length of beam.
The dynamic problem of the beam is governed by the equations (62) to-
gether with the strain-displacement relations
∂2v h ∂u 1
χ=− ,η = (63)
∂z 2 o ∂z o
and the constitutive relations (61). Equations (62) and (63) together with
(61) are solved numerically. Beams are discretized into finite elements and
each node has three degrees of freedom: axial and transverse displacement
plus rotation. The flexural problem is addressed by using Hermite shape
functions, which guarantee the continuity of both the transverse displace-
ment and rotation, while for the axial displacement linear shape functions
are adopted (24). Moreover, the Newmark and the Newton-Raphson meth-
ods are used to obtain the numerical solution. As most of these techniques
are standard, a detailed explanation has been omitted. Nevertheless, it
should be noted that defining the element stiffness matrix K̂E requires cal-
culating the derivatives of the generalized stress with respect to the gener-
alized strain in each of the domain regions. Moreover, it should mentioned
that two different contributions are accounted for defining K̂E : the tangent
stiffness matrix of the element K̂Es and the so-called geometric-stiffness ma-
trix K̂E
g , which represents the tendency toward buckling induced by axial
loads, with consequent reduction of the total stiffness matrix. In this re-
gard, it is interesting to point out that, while the stiffness matrix K̂E s is
symmetrical in all regions of the admissible domain, K̂E g is not, as detailed
in (26).
Lastly, regarding the effects of viscous damping, these have been ac-
counted for by including in the motion equation a constant viscous damping
matrix, C, which is obtained as a linear combination of M and the initial
elastic stiffness matrix K, as per the Rayleigh assumption.
Finally, an example application of the proposed model is presented. By

referring to some of the typical characteristics of age-old, free-standing tow-
ers in Italy, the constant squared cross sectional tower is 40 m in height and
5.5 m in width, with walls of constant 1.5 m thickness. For the Youngs mod-
ulus, the mass density and the maximum compressive stress the values of E
= 3000 MPa, ρ = 1900 kg/m3 and σo = - 2.0 MPa have been chosen. The
first flexural and axial periods of the structure in the linear elastic range
are 1.19s and 0.12s, respectively. By way of example, the tower is subjected
to the (main) horizontal and vertical components of the classical El Cen-
tro earthquake: the horizontal accelerogram (El Centro S00E, 18/05/1940)
has a duration of 53.40 s and a PGA of 3.41 m/s2 , while the vertical com-
104 M. Lucchesi
ponent exhibits a PGA of 2.02 m/s2 . The dynamic non-linear analysis is

performed by assuming a viscous damping coefficient ν of 0.02 over the two
flexural modes. The results are given in terms of y - and z-displacements
at the tower’s top, as well as two damage parameters Fv /V and Cv /V as
a function of time. Such parameters represent, respectively, the volumes of
cracked and crushed portions, non-dimensionalized to the tower’s total vol-
ume. Figures 17 and 18 show comparisons of the response for the different
cases: (i) the heavy line represents the response to the earthquake’s actions
obtained via the proposed model (LCS masonry), (ii) the dashed line in-
dicates the response obtained via the proposed model when an unbounded
compressive strenght is considered (UCS masonry).
Non-resistance to tension leads to amplifications of the horizontal displace-
ments and elongation of the lateral period, which are caused by formation
of the first cracked zones and the consequent loss of stiffness. When lim-
ited compressive strength is also considered, further amplification of the
displacements occurs, together with elongated transverse and axial vibra-
tion periods. It is also interesting to note that significant coupling occurs
between the axial and transversal vibrations in the non-linear range, as
evinced by the time histories of the vertical displacements. Lastly, Figure
19 shows the distribution of the areas subjected to craking and crushing
along the tower height at the instant of time at which the maximum of the
Cv /V parameter occurrs.
Acknowledgement This research has been supported by the Regione
Toscana (project tools for modelling and assessing the structural behaviour
of ancient constructions: the NOSA-ITACA code, PAR FAS 2007-2013).
This support is gratefully acknowledged. I would also like to thank C.
Padovani, A. Pagni and G. Pasquinelli of the ISTI-CNR Laboratory of Me-
chanics of Materials and Structures for providing the study reported on in
Section 4, as well as B. Pintucchi and N. Zani for their precious collaboration
on drafting Section 5.
Bibliography
[1] Alfano G., Rosati L., Valoroso N., A numerical strategy for finite el-
ement analysis of no-tension materials. Int. J. Numer. Meth. Engng,
vol. 48, pp. 317-350, 2000.
[2] Angelillo M., Cardamone L., Fortunato A., A new numerical model for
masonry structures. J Mech. Mater. Stuct. 5, 583-615 (2010).
[3] Bathe K. J., Wilson E. L., Numerical methods in finite element analy-
sis, Prentice-Hall, Inc., Englewood Cliffs, New Jersey, 1976.
[4] Bennati S., Barsotti R.: Optimum radii of circular masonry arches.
y−displacement
0.25
UCS masonry
0.2 LCS masonry
0.15
0.1
0.05
[m]
−0.05
−0.1
−0.15
−0.2
−0.25
0 5 10 15 20
t [s]
x 10
−3 z−displacement
4
UCS masonry
LCS masonry
2
−2
[m]
−4
−6
−8
−10
0 5 10 15 20
t [s]
Figure 17. Time history of the trasversal and longitudinal displacements.
Proc. Third Int. Arch Bridges Confer., Paris (2001).

[5] Brenner S. C., Scott L. R. The mathematical theory of finite element
method, Second edition, Springer 2002.
[6] Ciarlet P. G., The finite element method for elliptic problems. North-
Holland Publishing Company, 1978.
[7] Cottle R. W., Pang J. S., Stone R. E., The linear complementarity
problem. Academic Press Inc. 1992.
[8] Cuomo M., Ventura G., A complementary energy formulation of no
tension masonry-like solids. Computer Methods in Applied Mechanics
and Engineering 189, pp. 313-339, 2000.
106 M. Lucchesi
F /V
v
0.4
UCS masonry
LCS masonry
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
0 5 10 15 20
t [s]
C /V
v
0.05
LCS masonry
0.045
0.04
0.035
0.03
0.025
0.02
0.015
0.01
0.005
0
0 5 10 15 20
t [s]
Figure 18. Time history of the damage parameters Fv /V and Cv /V.
[9] Dacorogna B.: Introduction to the Calculus of Variations. Imperial

College Press, (2004).
[10] Degl’Innocenti S., Lucchesi M., Padovani C., Pagni A., Pasquinelli G.,
Zani N.: Dynamical analysis of masonry pillars. Proc. Third Int. Cong.
on Science and Technology for the Safeguard of Cultural Heritage in
the Mediterranean Basin, Alcalá de Henares, (2001).
[11] Degl’Innocenti, S., Padovani, C., Pagni, A., Pasquinelli, G.: Dynamic
analysis of age-old masonry constructions. Proc. Eighth Inter. Conf.
on Computational Structures Technology, B.H.V. Topping, G. Montero
Craked/Total area Crushed/Total area

40 40
35 35
30 30
25 25
H [m]
H [m]
20 20
15 15
10 10
5 5
0 0
0 0.2 0.4 0.6 0 0.2 0.4 0.6
Figure 19. Distribution of the areas subjected to craking and crushing

along the tower height.
and R. Montenegro(Editors), Civil-Comp Press, Stirlingshire, Scotland

2006. Las Palmas de Gran Canaria 12-15 Sept., (2006).
[12] Guidotti P., Lucchesi M., Pagni A., Pasquinelli G., Application of shell
theory to structural problem using the finite element method. CNR-
Quaderni de ”La Ricerca Scientifica”, 115, 1986.
[13] Gurtin M. E., An introduction to continuum mechanics. Academic
Press, Boston, 1981.
[14] Lucchesi M., Padovani C., Pagni A., A numerical method for solving
equilibrium problems of masonry-like solids. Meccanica, vol. 29, pp.
175-193, 1994.
[15] Lucchesi M., Padovani C., Pasquinelli G., Zani N., The maximum mod-
ulus eccentricity surface for masonry vaults and limit analysis. Mathe-
matics and Mechanics of Solids, 4, pp. 71-87, 1999.
[16] Lucchesi M., Padovani C., Pagni A., Pasquinelli G., Zani N: COMES-
NOSA A finite element code for non-linear structural analysis. Report
CNUCE-B4-2000-003 (2000).
108 M. Lucchesi
[17] Lucchesi M., Zani N., Some explicit solution to plane equilibrium prob-
lem for no-tension bodies. Structural Engineering and Mechanics, vol.
16, pp. 295-316, 2003.
[18] Lucchesi M., Padovani C., Pasquinelli G., Zani N., Static analysis of
masonry vaults, constitutive model and numerical analysis. Journal of
Mechanics of Materials and Structures, 2(2), pp. 211-244, 2007.
[19] Lucchesi M., Pintucchi B.: A numerical model for non-linear dynamic
analysis of slender masonry structures. Eur. J. Mech. A. Solids, 26,
88-105, (2007).
[20] Lucchesi M., Padovani C., Pasquinelli G., Zani N.: Masonry Con-
structions: Mechanical Models and Numerical Applications. Lectures
notes in applied and computational mechanics, Vol. 39, Springer Verlag
(2008).
[21] Lucchesi M., Silhavy M., Zani N.: Equilibrium problems and limit
analysis of masonry beams. J. Elasticity, 106, 165-188, (2012).
[22] Lucchesi M., Pintucchi B. Zani N.: MADY, a finite element code for
dynamic analysis of slender masonry structures: Mathematical specifi-
cations and user manual (In preparation) (2013).
[23] http://www.nosaitaca.it/en/
[24] Oden J. T., Carey G. F., Finite elements-Special problems in solid
mechanics, Vol. 5. Prenctice-Hall, Inc., Englewood Cliffs, New Jersey,
1984.
[25] Padovani C., Pagni A., Pasquinelli G.: Un codice di calcolo per l’analisi
statica e il consolidamento di volte in muratura. Proc. WONDERma-
sonry - Workshop on design for rehabilitation of masonry structures,
145 - 153, P. Spinelli (ed.), (2006).
[26] Pintucchi B, Zani N.: Effects of material and geometric non-linearities
on the collapse load of masonry arches. Eur. J. Mech. A. Solids, 28,
45-61, (2009).
[27] Romano G., Sacco E., Sulla proprietà di coassialità del tensore di fes-
surazione. Atti Ist. Scienza delle Costruzioni, Facoltà di Ingegneria,
Napoli, n. 351, 1984.
[28] Sacco E., Modellazione e calcolo di strutture in materiale non resistente
a trazione. Rend. Mat. Acc. Lincei, s. 9, vol. 1, pp. 235-258, 1990.
[29] http://www.salome-platform.org
[30] Zani N.: A constitutive equation and a closed form solution for no-
tension beams with limited compressive strength. Eur. J. Mech. A.
Solids, 23, 467-484, (2004).
Practical applications of unilateral models to
Masonry Equilibrium
Maurizio Angelillo‡
‡
Dipartimento di Ingegneria Civile, University of Salerno, Italy
Abstract This Chapter is devoted to the application of unilateral

models to the stress analysis of masonry structures. Some 2d ap-
plications of what we call the simplified models for masonry, are
discussed and studied. Though the essentially unilateral behaviour
of masonry is largely recognized, some prejudices still persist on the
possibility of making the No-Tension (NT) assumption a practical
model for designing engineers. The results here presented demon-
strate that the unilateral model for masonry can be a useful tool
for modeling real masonry structures. In the exposition the critical
points are emphasized and strategies to handle them are suggested,
both for the most primitive model (namely the Rigid NT material),
and for the more accurate Normal Elastic NT and Masonry-Like
(ML)materials. The first tool here introduced for applying the No-
Tension model to structures is the systematic use of singular stress
and strain fields. Next a number of closed form solutions for NENT
and ML materials is discussed. Finally a numerical approach based
on descent is proposed for handling the zero-energy modes typical
of unilateral materials. Some numerical solutions and comparisons
with analytical solutions and test results are also presented.
This Chapter is dedicated to Giovanni Castellano who inspired
most of my work on masonry since my early steps.
1 Basic tools
In this section the main notation and the basic notions of equilibrium and
compatibility, in presence of singular stress and strain fields, are introduced.
Singular strains are usually considered in perfect plasticity, and the use
of singular stress fields (though in a mathematically unconscious way) has
been around since the nineteenth century (see Mery (1840)). It is only fairly
recently that Šilhavý, Lucchesi et al (see Lucchesi and Zani (2005)) , have
put forward a rigorous mathematical formulation of stress field singularities.
Chapters 2 and 3 of the present book are partly devoted to the mathematical
exposition of these clever concepts within the theory of measures.
M. Angelillo (Ed.), Mechanics of Masonry Structures, CISM International Centre for Me-
chanical Sciences, DOI 10.1007/ 978-3-7091-1774-3_4, © CISM, Udine 2014
110 M. Angelillo
The formulation that is given here, instead, is rather informal and based
mainly on geometrical arguments. A substantial knowledge of the mathe-
matical theory of linear elasticity, such as that given in the monograph by
Gurtin (1972), to which I refer for notations, is presumed. Familiarity with
functional analysis is not strictly required, though some previous experi-
ence with elementary functional analysis in Sobolev spaces and variational
methods (as can be found, for example, in the books by Kreyszig (1989)
and Dym and Shames (1973)) would be of help.
The matter treated and analysed here is not entirely new. Much of
what is reported, apart from the classical and more recent sources cited
throughout the text, leans on a number of papers recently published, or
under print or review, by myself or my reserch group. In particular, on
singular stress: Angelillo et al. (2012) and Angelillo et al. (2013); on Limit
Analysis for masonry: Angelillo and Fortunato (2013); on semianalitical
solutions for panels: Fortunato (2010); on numerical methods for unilateral
materials Angelillo et al. (2010).
1.1 Preliminaries
It is assumed that the body, a domain Ω ∈ n (here n=2), loaded by
the given tractions s on the part ∂ΩN of the boundary, and subject to given
displacements u on the complementary, constrained part of the boundary
∂ΩD , is in equilibrium under the given surface displacements, tractions and
body loads (u; s,b)1 and undergoes displacements u and local deformations,
so small that the infinitesimal strain E(u) is a proper strain measure.
Vectors and tensors are represented in Cartesian components, in a fixed
frame (0; x1 , x2 ). Summation convention is adopted throughout the text.
1.2 Equilibrated stress fields, regularity of T

A stress field T is said to be equilibrated with (s, b), if it satisfies the
equilibrium equations
divT + b = 0 ,
and the traction boundary conditions
Tn = s , on ∂ΩN ,
n denoting the unit outward normal to ∂Ω.
1
Other possible data are the eigenstrains E; here I omit them from the analysis to
simplify the exposition, though some special eigenstrains will be considered as data in
some of the examples which follow.
Practical Applications of Unilateral Models… 111
T is a tensor function of x ∈ Ω, for which some kind of regularity must

be assumed. If the differential equations of equilibrium are considered in
a strong sense, the stress field T must be differentiable and its divergence
must be continuous.
On adopting a variational formulation, if the material is linearly elastic,
the minimal request for T is to be square summable, that is
(
T · Tda < ∞ .
Ω
For some rigid perfectly plastic materials (such as rigid unilateral mate-
rials), less regular and even singular stresses may be admitted. The minimal
request for such materials is that T be summable
√
T · Tda < ∞ .
Ω
If one admits stress fields that are only summable, the set of competing
functions enlarges to bounded measures, that is to summable distributions
)
T:

)
T < ∞ ,
Ω
which, in general, can be decomposed into the sum of two parts
) =T
T )s ,
)r + T
where T ) r is absolutely continuous with respect to the area measure (that

)
is Tr is a density per unit area) and T) s is the singular part.
In the examples, the analysis will be restricted to bounded measures T )
whose singular part is concentrated on a finite number of regular arcs, that
is bounded measures admitting on such curves a density T ) s with respect
to the length measure (that is special bounded measures with void Cantor
part; for reference to these function spaces see Ambrosio et al. (2000)).
Remark 1. If the stress field is summable (and also if it is square
summable), it is not differentiable in strong sense, and the equilibrium equa-
tions have to be reformulated in variational form (e.g. through the Virtual
Work equation). Singular stresses require also special modifications of the
boundary conditions; the trace of the stress T on the loaded part of the
boundary is not given by Tn if T is singular. I shall come to this point
later.
112 M. Angelillo
1.3 Compatible displacement fields, regularity of u

The displacement field u is said to be compatible if, besides being regular
enough for the corresponding strain E(u) to exist 2 , u satisfies the boundary
conditions on the constrained part ∂ΩD of the boundary
u = u , on ∂ΩD .
For linearly elastic bodies, on adopting a variational formulation, the
usual assumption is that E be square summable, that is
(
E · Eda < ∞ .
Ω
For some rigid, perfectly plastic (or rigid unilateral) materials, it is suf-
ficient to assume that E be summable
√
E · Eda < ∞ .
Ω
As before, the set of competing functions enlarges to bounded measures,

that is to summable distributions E; ) then the displacement u can admit
finite discontinuities, i.e. u can be a function with bounded variation. If
E were the whole gradient of u, the summability of E would entail: u∈
BV (Ω), exactly. Since E is only the symmetric part of ∇u, u must belong
to a larger space: BD(Ω). The strain corresponding to u is again a bounded
measure

)
E < ∞ ,
Ω
which, in general, can be decomposed into the sum of two parts
) =E
E )s ,
)r + E
where E) r is absolutely continuous with respect to the area measure (that is

)
Er is a density per unit area) and E) s is the singular part.
)
Es has support on the union of a set of linear 1d measure (the jump set
of u) and a set of fractional measure.
For simplicity, in the examples, I shall restrict to bounded measures E )
whose singular part is concentrated on a finite number of regular arcs, that
is bounded measures admitting on such curves a density E ) s with respect
2
Recall that, here, E(u) is the infinitesimal strain
to the length measure (that is special bounded measures with void Cantor
part).
Remark 2. If u∈ BD(Ω), that is u can be discontinuous, the b.c. u
=u on ∂ΩD makes no sense. A way to keep alive the b.c. of Dirichelet type
is to identify the masonry body rather than with the domain Ω (usually an
open set) with the set Ω ∪ ∂ΩD and to assume that u must comply with
the constraint u =u on the skin ∂ΩD , admitting possible singularities of
the strain at the constrained boundary. Then, from here on, I shall deviate
from standard notation referring to Ω as to the set Ω ∪ ∂ΩD .
Given the displacement field u of x, by taking the gradient of u, in a
classical sense if u is regular, and in a generalized sense if u is singular,
the strain E(u) is derived. Vice versa, if E of x is given, the possibility of
integrating the components Eαβ to get the (possibly discontinuous) compo-
nents uα of u, is submitted to the necessary compatibility conditions (also
sufficient if Ω is simply connected)
E11,22 + E22,11 − 2E12,12 = 0 ,

where a comma followed by an index, say α, means differentiation with
respect to xα .
The reader will see in what follows, that, on admitting discontinuous
displacements, this condition can be reinterpreted in a generalized sense
and applied (with some care), also to discontinuous, and even singular,
strains.
1.4 Dirac deltas: a familiar example

For the unilateral models that here are adopted for masonry materials,
it makes sense to admit singular stresses and strains, that is stress fields
T and strain fields E that can be concentrated on lines (line Dirac deltas).
In mathematical terms these are not functions in a strict sense, since they
assign finite values to all points x∈ Ω, except to those belonging to a set
of lines of Ω ∪ ∂ΩD , to which infinite values are associated. Anyway, these
infinite values must be such that these stresses or strains be summable, that
is

|T| < ∞ , |E| < ∞ ,
Ω Ω
or, in other words, T and E must be bounded measures. Here I call M (Ω)
the set of bounded measures on Ω ∪ ∂ΩD . Line Dirac deltas are special
bounded measures; a simple example of a Dirac delta in 1d, is the concen-
trated load on a beam, that is a point Dirac delta (Figure 1).
114 M. Angelillo
a b c
Figure 1. Concentrated transverse force Q on a beam: (a). Corresponding internal
shear force T : (b), and bending moment M : (c).
From elementary beam theory the internal moment M, the shear force
T and the transverse load per unit length q, are related by the differential
equilibrium relations
M = T , T = −q ,
where prime denotes differentiation with respect to z.
The previous equilibrium conditions admit also the following integral
form
z z
T (z) = T (0) − q , M (z) = M (0) + T .
0 0
The second set of equations have sense also if q is a concentrated force,
that is a Dirac delta:
q(z) = Qδ(z o ) ,
Q being the intensity of the load, δ(.) the unit Dirac delta, zo the point of
application of the force.
The unit point Dirac delta applied at zo is defined as follows
z o +
δ(z o ) = 1 , ∀ > 0 .
z o −
The differential equations can be extended to the case of singular loads by
interpreting -q in Figure 1a as the generalized derivative of T in Figure 1b.
More generally, in 1d, the generalized derivative of a piecewise constant
function f is a distribution whose regular part is zero and whose singular
part has support on the jump set of f. Therefore a piecewise constant
function is a special BV function whose singular part consists of point Dirac
deltas applied at the points of discontinuity of f. The integral of such Dirac
deltas across any point of discontinuity gives the value of the jump of f, at
that point.
a b
Figure 2. Discontinuous displacement alog a straight line Γ, unit tangent t and normal
m to Γ: (a). Graph of the generic component uα , a BV function: (b).
1.5 Singular stress and strain as line Dirac deltas

In what follows I will restrict to consider stress and strain fields that
are special bounded measures, namely Dirac deltas with support on a fi-
nite number of regular arcs, and look at the restrictions imposed on these
singular fields by equilibrium and compatibility, respectively.
Strain. In what follows, special displacement fields of bounded variation
will be considered. In particular, restricting to discontinuous displacement
fields u having finite discontinuities on a finite number of regular arcs Γ, the
strain E(u) consists of a regular part Er , that is a diffuse deformation over
Ω − Γ, and a singular part Es in the form of a line Dirac delta, concentrated
on Γ.
The jumps of u along Γ, can be interpreted as fractures. In Figure 2a,
to which I refer for notations, a crack separating the body Ω into two parts,
℘1 and ℘2 , along a straight interface Γ, is represented. On such a line, the
jump of u:
[[u]] = u+ − u− .
due to a relative translation of the two parts, is considered. Here u+ is the
displacement on the side of Γ where m points
The displacement field is a piecewise constant vector field, discontinuous
on Γ; the graph of a generic Cartesian component of u, is depicted in
Figure 2b.
The jump of u can be decomposed into normal and tangential compo-
nents
Δv = [[u]] · m , Δw = [[u]] · t ,
where t, m are the unit tangent and normal to Γ, represented in Figure 2a.
Notice that, on any crack, incompenetrability of matter requires Δv ≥ 0 (a
unilateral restriction).
116 M. Angelillo
The strain E corresponding to the piecewise constant field u depicted in

Figure 2b, is zero everywhere on Ω − Γ and is singular on Γ:
1 1
E(u) = δ(Γ)(Δvm ⊗ m + Δwt ⊗ m + Δwm ⊗ t) .
2 2
Stress. If the stress field T is non-singular (say T ∈ L2 (Ω)), on a
possible discontinuity line Γ, for equilibrium, the stress emerging on Γ must
be continuous. Then at any regular point of Γ, denoting m the unit normal
to Γ, the stress T must satisfy the condition
(T+ − T− )m = 0 ,
T+ being the stress on the side of Γ where m points. Then, if T ∈ L2 (Ω),
the possible jumps of T must be restricted to the part of T non-emerging
on Γ.
If T is singular, say a Dirac delta on Γ, also the part of T emerging on
Γ can be discontinuous. The unbalanced emerging stress
q = (T+ − T− )m ,
in equilibrium, must be balanced by the stress concentrated on Γ (Figure 3).
Referring for notations to Figure 3, the representation of the singular part
Ts of T on Γ, is
Ts = N δ(Γ)t ⊗ t .
For equilibrium, calling p, q the components of q in the tangential and
normal directions, and denoting ρ the curvature of Γ, the following equations
must hold
N +p=0, N ρ+q =0.

Therefore q must be zero if Γ is straight.
Kinks. Though the singularity lines Γ, for the stress T, that I consider
are a.e. smooth, they can have kinks and multiple points. At such nodes
the equilibrium of forces transmitted to the nodes must be satisfied; then
if the node is inside the body and there are no concentrated external forces
applied to the node there must be at least a triple junction.
Airy’s stress function and singular stresses. In absence of body
forces (b=0), the equilibrium equations admit the following solution in
terms of a scalar function F:
T11 = F,22 , T22 = F,11 , T12 = −F,12 .

qs
Ns
ps ''
Figure 3. Stress singularity: forces acting on the curve Γ.
a b c
Figure 4. Square panel under uniform pressure: (a). Corresponding boundary value
m(s): (b), and normal slope n(s): (c).
This is the general solution of the equilibrium equations, if the loads are
self-balanced on any closed boundary delimiting Ω (see Gurtin (1972)).
The b.c. T n = s on ∂ΩN , must be reformulated in terms of F. Denoting
x(s) the parametrization of ∂ΩN with the arc length, the b.c. on F are:
dF
F (s) = m(s) , = n(s) , on ∂ΩN ,
dν
in which dF/dν is the normal derivative of F at the boundary (that is the
slope of F in the direction of n) and m(s), n(s) are the moment of contact
and the axial force of contact produced by the tractions s(s), on a beam
structure having the same shape of ∂Ω, and cut at the point s=0.
A simple example is shown in Figure 4.
Regular and singular equilibrated stress fields can be derived by stress
functions meeting the prescribed b.c. on F and dF/dν. A regular stress
field is represented by a smooth F (see Figure 5a), a singular stress field by
a continuous but folded F (Figure 5b). The projection of a fold of F on Ω is
118 M. Angelillo
a b c
Figure 5. Square panel under uniform pressure. Smooth Airy’s stress function giving
homogeneous pressure inside the body: (a); folded Airy’s function: (b); representation
of the uniaxial and singular stress field corresponding to the folded Airy’s function: (c).
called folding line and is denoted Γ. On a fold of F , the second derivative of

F , with respect to the normal m to the folding line Γ, is a Dirac delta with
support on Γ. Therefore, along Γ the Hessian H(F) of the stress function
F is a dyad of the form
H(F ) = Δm F δ(Γ)m ⊗ m ,
Δm F denoting the jump of slope of F in the direction of the normal m to

Γ (see Figure 5c). Recalling the Airy’s relation, the corresponding singular
part of the stress is
Ts = N δ(Γ)t ⊗ t ,
where the axial contact force N is given by
N = Δm F .
2 Model zero (RNT)

In this section the main ingredients of the theory concerning the most primi-
tive model for masonry materials, namely the Rigid No-Tension (RNT) ma-
terial, are presented. After introducing the constitutive assumptions, the
definitions of statically admissible and kinematically admissible fields are
given, and the compatibility of loads and distortions is discussed. The RNT
model allows for the application of the theorems of Limit Analysis: the for-
mulation of the static and kinematic theorems for rigid-unilateral materials
is given and a number of illustrative examples is developed.
2.1 Constitutive restrictions and equilibrium problem

It is assumed that the body Ω ∈ n (here n = 2), loaded by the given
tractions s on the part ∂ΩN of the boundary, and subject to given displace-
ments u on the complementary, constrained part of the boundary ∂ΩD , is in
equilibrium under the action of such given surface displacements and trac-
tions, besides body loads b and distortions E (the set of data being denoted:
(u, E; s, b)), and undergoes small displacements u and strains E(u)3 .
I point out that here the masonry structure is identified with the set:
Ω = Ω ∪ ∂ΩD , i.e. it is considered closed on ∂ΩD and open on the rest of
the boundary.
I consider that the body Ω is composed of Rigid No-Tension material,
that is the stress T is negative semidefinite
T ∈ Sym− , (1)
∗
the effective strain E = E(u) − E is positive semidefinite
E∗ ∈ Sym+ , (2)
∗
and the stress T does no work for the corresponding effective strain E
T · E∗ = 0 . (3)
∗
The effective strain E is a positive definite tensor field doing no work
for the corresponding stress, and representing detachment fractures (that is
type 1. fractures, see Sect.2, Chap.1). E ∗ is a sort of “reaction” deformation
associated to the constraint on stress (1), and, therefore, is also called latent
strain. In order to avoid trivial incompatible loads (s,b), it is assumed that
the tractions s satisfy the condition
s · n < 0 , or s = 0, ∀x ∈ ∂ΩN . (4)

Notice that in the plane case (n=2) conditions (1), (2), can be rewritten
as
tr T ≤ 0 , det T ≥ 0 , (5)
tr E∗ ≥ 0 , det E∗ ≥ 0 . (6)
3
When eigenstrains are considered, under the small strain assumption, the total strain
E(u) is decomposed additively as follows: E(u) = E∗ +E, E∗ being the effective strain
of the material.
120 M. Angelillo
2.2 Statically admissible stress fields

An equilibrated stress field T (that is a stress field T balanced with the
prescribed body forces b and the tractions s given on ∂ΩN ) satisfying the
unilateral condition (1) (that is conditions (5)), is said statically admissible
for a RNT body. The set of statically admissible stress fields is denoted H
and is defined as follows

H = T ∈ S(Ω) s.t. divT + b = 0 , Tn = s on ∂ΩN , T ∈ Sym− , (7)
S(Ω) being a function space of convenient regularity. Since for RNT ma-
terials, discontinuous and even singular stress fields will be considered, one
can assume S(Ω) = M (Ω), that is the set of bounded measures.
For Elastic No-Tension (ENT) materials a sensible choice is S(Ω) =
L2 (Ω), that is the function space of square summable functions.
Actually the space M (Ω) contains L2 (Ω) and is much larger than it, that
is the set of functions which compete for equilibrium is richer for RNT than
for ENT materials; this fact makes easier for RNT materials the search of
s.a. stress fields.
The differential equations of equilibrium must be reformulated for non
smooth T, since the derivatives of T do not exist in a classical sense4 . One
way to do it is to impose equilibrium in a variational form, namely by using
the Virtual Work Principle.
On introducing the set of virtual displacements
δ K = {δu ∈ S ∗ (Ω) s.t. δu = 0 on ∂ΩD } , (8)
the stress field T is balanced with (s,b) if and only if

s · δu + b · δu = T · E(δu) , ∀δu ∈ δ K . (9)
∂ΩN Ω Ω
S ∗ (Ω) is a function space of convenient regularity. If T ∈ L2 (Ω), then

S ∗ (Ω)= H 1 (Ω) guarantees the finiteness of the internal work. If T ∈ M (Ω),
the choice S ∗ (Ω)= C 1 (Ω) ensures the possibility of computing the internal
virtual work.
4
For singular stress fields, even the b.c. on the emerging tractions must be changed,
since the Cauchy argument leading to them, is restricted to absolutely continuous
stress vectors.
2.3 Fundamental partition

To any statically admissible stress field T one can associate the following
partition of the domain Ω = Ω ∪ ∂ΩD :
Ω1 = {x ∈ Ω s.t. trT ≤ 0 , detT ≥ 0} , (10)
Ω2 = {x ∈ Ω s.t. trT ≤ 0 , detT = 0} , (11)
Ω3 = {x ∈ Ω s.t. T = 0} . (12)
On introducing the spectral decomposition of T:
T = σ1 k1 ⊗ k1 + σ2 k2 ⊗ k2 ,
in Ω1 the stress is of biaxial compression, that is σ1 < 0, σ2 < 0; in Ω2 the
stress is of uniaxial compression, that is T=σ k ⊗ k, σ < 0; Ω3 is inert.
Notice that the form and the regularity of these regions depend on the
smoothness of T, and that such regions can be rather weird if one admits
T ∈ M (Ω). We will see in what follows that, on admitting only special
bounded measures (i.e. Dirac deltas with support on a finite number of
regular arcs), the regions Ωi can degenerate, but the fundamental partition
can be still easily identified.
Remark 3. In the Ω2 regions, that is where the stress is of uniaxial com-
pression, a classical theorem of Tension Field Theory (see Subsection 3.8
and Remark 9), states that (as the intuition suggests, see below and Fig-
ure 6) the lines of principal compression (tension in the case of TFT) form
a family of straight lines if the body forces vanish.
A simple geometrical proof of this statement can be obtained by observ-
ing that, the equilibrium of a material tube contained in between two such
infinitesimally spaced principal lines, is possible only if the tube is straight
and the internal axial force of contact N is constant (see Figure 6).
If the two contiguous lines are parallel the stress itself is constant. If the
two lines converge, on denoting R the distance between the point P and
the point of intersection of the two lines, and introducing the arc length s
along the ray from the origin P , calling σ(s) and σ ◦ the stress at Q and the
stress at P , for equilibrium, the stress along the ray takes the form
R
σ(s) = σ◦ .
R−s

122 M. Angelillo
Figure 6. Stress along a compression ray.
2.4 Concavity of the Airy’s stress function

In absence of body forces, a statically admissible stress field can be
expressed5 in terms of a scalar function F (called Airy’s solution, see sub-
section 1.5).
The constraint (5), translated in terms of F, reads
2
tr T = F,11 + F,22 ≤ 0 , det T = F,11 F,22 − F,12 ≥0, (13)
then the Hessian H(F) of F, is negative semidefinite and the stress function
F must be concave. Therefore, in absence of body forces b, the equilibrium
problem for a NT material, can be formulated as the search of a concave
function F, taking on the part ∂ΩN of the boundary, a specified value and
a specified slope.
Example. As a simple example of an equilibrium problem, I consider
the traction problem depicted in Figure 7a. Smooth and singular statically
admissible stress fields can be easily derived from simple stress functions
matching the given boundary data.
A smooth solution can be derived from the stress function (here L=1 is
assumed):

− 23 p + 2px2 , x2 < 12 (1 − x21 ) ,
F = (1−x2 )2
− 21 px21 − 2p 1+x22 , x2 ≥ 12 (1 − x21 ) .
1
This F is a composite surface, flat in the region denoted Ω3 in Figure 7c,

and strictly concave in Ω1 . The graph of such F is depicted in Figure 7b. I
5
Univocally, if the body is simply connected or loaded by self balanced tractions on any
closed boundary.
a b c
Figure 7. Wall beam under uniform transverse load: (a). Graph of the Airy’s stress
function, corresponding to the smooth solution: (b). Domain partition and one of the
families of compression lines corresponding to the smooth solution: (c).
a b c
Figure 8. Graph of the folded Airy’s function, showing the intersection of the gener-
ating surfaces: (a). Corresponding domain partition and principal lines of compression:
(b). Support of the singular stress (solid lines 3, 4, 5): (c).
leave to the reader to verify that the corresponding T is statically admissible,

that is that such T matches the boundary data and belongs to Sym− .
A singular statically admissible stress field is derived from the stress
function F depicted in Figure 8a. F is a continuous non-smooth function:
the surface F, making (concave) folds along the lines indicated with 3, 4,
5 in Figure 8c, can be easily produced by prolongating the datum F|∂Ω
with ruled surfaces having the prescribed slope dFdν |∂Ω at the boundary. The
intersections of the four ruled surfaces emanating from the boundary (see
Figure 8a) give the folding lines; the jump of slope hortogonal to the folding
line gives the value of the axial force along the line:
Ts = N δ(Γ)t ⊗ t .
Since the fold is concave the jump of slope is negative, then
N = Δm F < 0 .
124 M. Angelillo
In Figure 8b the principal lines of uniaxial compression corresponding

to the non smooth solution are reported.
2.5 Kinematically admissible displacement fields

A compatible displacement field u, that is a displacement u matching
the given displacements u on ∂ΩD for which (E(u) - E) ∈ Sym+ , i.e. such
that the effective strain satisfies the unilateral conditions (6), is said to be
kinematically admissible for a RNT body.
The set of kinematically admissible displacement fields is denoted K and
is defined as follows:

K = u ∈ T (Ω) s.t. u = u on ∂ΩD , (E(u)-E) ∈ Sym+ , (14)
where Ω = Ω ∪ ∂ΩD and T(Ω) is a function space of convenient regularity.

Since for RNT materials, discontinuous displacements can be considered,
one can assume T(Ω) = BV (Ω), that is the set of functions of bounded
variation (the functions whose gradient belongs to M (Ω), i.e. functions u
admitting finite discontinuities). I restrict to the subset of BV (Ω), consist-
ing of displacement fields u having finite jumps on a finite number of regular
arcs. Actually, as we shall see, I will need only to consider discontinuous
functions u whose jump set is the union of a finite number of segments.
The differential relation between E and u, likewise the displacement b.c.,
must be reformulated in a weak form, since the derivative of u does not exist
in a classical sense, and the trace of u on ∂ΩD is not well defined. One way
to do it is to impose compatibility in a variational form, namely by using
the Complementary Virtual Work Principle.
On introducing the set of virtual stress fields
δ H = {δT ∈ T ∗ (Ω) s.t. divδT = 0 , δTn = 0 on ∂ΩN } , (15)
the displacement field u is compatible with (u, E) if and only if

(δTn) · u − δT · E = δT · E(u) , ∀δT ∈ δ H . (16)
∂ΩD Ω Ω
T ∗ (Ω) is a function space of convenient regularity. If u ∈ H 1 (Ω) (as it is

assumed in linear elasticity), then T ∗ (Ω) = L2 (Ω) guarantees the finiteness
of the internal virtual work. If u ∈ BV (Ω), the choice T ∗ (Ω) = C ◦ (Ω)
ensures the possibility of computing the internal virtual work.
a b
Figure 9. Examples of kinematical problems. Wall loaded by uniform vertical load
at the top and subjected to a given uniform settlement of the right foot: (a). Masonry
panel subject to uniform thermal expansion of the right half: (b).
a b c
Figure 10. Compatible solution for the problem of Figure 9a: (a). Corresponding
(singular) strain components: (b), (c).
Examples. As a simple illustration of typical kinematical problems,

I construct some admissible deformations for the two example problems
reported in Figure 9. In (a) the effect of a given settlement η of the right
foot, is considered. In (b) the constraints are fixed and the effect of the
distortion E =αΔT I, due to the uniform, positive increment of temperature
ΔT, applied to the right half of the strip, is studied.
A kinematically admissible displacement, compatible with the given set-
tlement, is shown in Figure 10a; in Figures 10b,c the strain components
E11 , E22 are graphically represented.
A kinematically admissible displacement for the second example is

αΔT x1 (1 − x22 /L2 ) , x1 < L ,
u1 =
αΔT x1 , x1 ≥ L ,

αΔT x2 x21 /L2 , x1 < L ,
u2 =
αΔT x2 , x1 ≥ L .
126 M. Angelillo
a b c
Figure 11. Compatible solution for the problem of Figure 10b: (a). Corresponding
(regular and singular) strain components: (b), (c).
The corresponding deformation and the strain components E11 , E22 , are
graphically represented in Figure 11. I leave to the reader to verify that the
effective strain (E(u)-E) belongs to Sym+ 6 and that the strain E, whose
non-vanishing components are depicted in Figures 11b c, satisfy the com-
patibility conditions (6) in a generalized sense.
2.6 Compatibility of loads and distortions

The data of a general BVP for a RNT body can be split into two parts
↔ (s, b) ≈ loads ,
∗ ↔ (u, E) ≈ distortions .
The equilibrium problem and the kinematical problem for RNT materials,
namely the search of admissible stress or displacement fields for given data,
are essentially independent, in the sense that they are uncoupled but for
condition (3).
It has to be pointed out that, for RNT bodies, there are non-trivial
compatibility conditions, both on the loads and on the distortions; that is
the existence of statically admissible stress fields for given loads, and the
existence of kinematically admissible displacement fields for given distor-
tions, is submitted to special conditions on the data (for a thorough study
of compatibility conditions on the loads see Del Piero (1989) and Angelillo
and Rosso (1995)).
6
The assumption that the effective strain has to belong to Sym+ , implies that, on a
crack Γ, the form of the singular strain be
Es = δ(Γ)Δvm ⊗ m , with Δv > 0 ,

that is shearing discontinuities are forbidden.
Figure 12. Statically admissible solution reconcilable with the compatible mechanism
of Figure 11a, and corresponding to the BVP depicted in Figure 10a. An arch carrying
a concentrated axial force is formed, springing from the two hinges of Figure 10a. The
stress field is regular and uniaxial above the arch.
The definition of compatible loads and distortions is rather straightfor-

ward:
{ is compatible} ⇔ {H = ∅} , (17)
{∗ is compatible} ⇔ {K = ∅} . (18)
Therefore the more direct way to prove compatibility, both for loads and
distortions, is to construct a s.a. stress field or a k.a. displacement field, as
done in the previous examples.
To prove the existence of a solution to the BVP for a No-Tension body,
the compatibility of and ∗ is necessary but not sufficient, since the further
condition
T · E∗ (u) = 0 ,
must be satisfied (this is the material restriction (3)). Then one can say
that a possible solution to the BVP is given, if there exist a s.a. stress field
and and a k.a. dispalcement field, which are reconcilable in the sense of
condition (3).
As a simple example in Figure 12, a s.a. stress field giving, together
with the mechanism of Figure 10a, a possible solution to the simple problem
depicted in Figure 9a, is reported.
Examples. In the examples of Figure 13, the possible solution of two
classical mixed BVP for RNT materials is pictorially presented. Notice that
both the stress solution and the displacement solution present singularities.
128 M. Angelillo
a b
Figure 13. Examples of solutions of typical BVP’s for NENT materials. Arch loaded
by a uniform (per-unit projection) vertical load at the extrados and subjected to a given
uniform settlement of its abutments. In (a) the supports spread, in (b) they get closer.
The strain is singular at the supports and at the key-stone; two different arches form in
the two cases.
The examples reported in Figures 13, that can be found also in the mile-
stone book by Heyman (1995), testify the need, in order to solve a BVP
for Rigid No-Tension materials, to consider at the same time singular stress
and strain fields, and call for an extended formulation of the theorems of
Limit Analysis.
2.7 Incompatibility of loads and distortions

The way to verify the incompatibility of the data is less straightforward,
requiring the definition of two new sets

H◦ = T◦ ∈ S(Ω) s.t. divT◦ = 0 , T◦ n = 0 on ∂ΩN , T◦ ∈ Sym− ,
and

K◦ = u◦ ∈ T (Ω) s.t. u◦ = 0 on ∂ΩD , E(u◦ ) ∈ Sym+ .
◦ ◦ ◦◦ ◦◦
Both H and K can reduce to the sets H and K corresponding to
null stress and strain fields, depending on the geometry of the boundary, of
the loads and of the constraints.
◦ ◦◦ ◦ ◦◦
Remark 4. The fact that H − H can be void and that K − K
can be non-void is kind of peculiar of RNT materials; indeed we are used to
think to 2d continua as overdetermined and deprived of rigid, zero-energy,
internal modes.
One way to see overdeterminacy is to add to any s.a. stress field a, non
◦ ◦◦
zero, self balanced stress field T◦ . The fact that H − H can be void,
a b
Figure 14. Example of incompatible loads: in (a) wall loaded by uniform vertical
loads at the top and bottom bases. K.a. displacement field for which the load performs
positive work: (b).
means that overdeterminacy depends on the loads. There also can be loads
for which the structure becomes statically admissible. The fact that the
overdeterminacy/underdeterminacy of the structure depends on the load is
typical also of discrete structures with unilateral constraints.
The absence of degrees of freedom is proved, for discrete structures,
◦ ◦◦
by denying the possibility of zero energy mechanisms. u◦ ∈ K − K is
indeed a non trivial mechanism requiring, for the RNT body, zero energy
expended. The underdeterminacy of the structure, descending from the
◦ ◦◦
fact that K − K can be non-void, demands for non trivial compatibility
conditions on the loads.
The incompatibility of the data can be assessed as follows
◦
{ incompatible} ⇐ ∃u◦ ∈ K s.t. , u◦ > 0 , (19)
◦
{∗ incompatible} ⇐ ∃T◦ ∈ H s.t. ∗ , T◦ > 0 , (20)
where , u◦ , ∗ , T◦ represent the work of the loads and distortions for
u◦ , T◦ , respectively.
Examples. The load of Figure 14a is incompatible, since it makes pos-
itive work for the mechanism u◦ depicted in Figure 14b.
The distortion represented in Figure 15a is incompatible, since it makes
positive work for the self stress T◦ depicted in Figure 15b.
Remark 5. The incompatibility of a given set of loads means that
equilibrium is not possible and that acceleration of the structure must take
place7 . The incompatibility of a given set of distortions means that the given
7
A trivial compatibility condition for all kinds of bodies, under pure traction conditions,
130 M. Angelillo
a b
Figure 15. Example of incompatible distortions: in (a) panel subject to a given
displacement of the left constraint. S.a. stress field for which the distortion performs
positive work: (b).
kinematical data cannot be accomodated with a zero energy mechanism

and demand for more complex material models (i.e. elastic NT, elastic
NT-plastic, etc).
2.8 Limit Analysis

We have seen in the preceding sections that, for RNT bodies, both force
and displacement data are subject to compatibility conditions, that is the
existence of a statically admissible stress field and the existence of a kine-
matically admissible displacement field, are subordinated to some necessary
or sufficient conditions on the given data. Here I concentrate on necessary
or sufficient conditions for the compatibility of a given set of loads (s, b),
restricting to the case of zero kinematical data (u, E). The definition of
safe, limit and collapse loads are given first, and the propositions defining
the compatibility of the loads, that are essentially a special form of the
theorems of Limit Analysis (LA), are then discussed.
Theorems of Limit Analysis. Recalling the definition of RNT materi-

als, we can observe that the restrictions (2), (3) are equivalent to a rule of
normality of the total strain to the cone of admissible stress states. Nor-
mality is the essential ingredient allowing for the application of the two
theorems of Limit Analysis (see Del Piero (1998)). In order to avoid the
possibility of trivial incompatible loads (and simplify the formulation of the
two theorems), assumption (4) (i.e. that the tractions s applied at the
is load balance. Load balance is only a necessary compatibility condition for unilateral
bodies.
boundary are either compressive or zero) is made.

Admissible fields. The rigorous proof of the two theorems of Limit
Analysis requires to set the problem in proper functions spaces. For RNT
materials is appropriate and convenient to define the sets of statically ad-
missible stress fields H and kinematically admissible displacement fields K,
as follows


K = u ∈ T (Ω) s.t. u = 0 on ∂ΩD , E(u) ∈ Sym+ , (22)
where a convenient choice for the function spaces S(Ω) and T (Ω) is
S(Ω) = SM F (Ω) ,
T (Ω) = {u s.t. gradu ∈ SM F ∗ (Ω)} ,

SM F being the set of Special Measures (that is measures with null Cantor
part) whose jump set is Finite, in the sense that the support of their singular
part consists of a finite number of regular (n − 1)d arcs8 .
With SM F ∗ I denote the subset of SM F for which the support of the
singular part is restricted to a finite number of (n − 1)d segments.
Notice that, depending on the geometry of the structure Ω = Ω̇ ∪ ∂ΩN
and on the given loads, the set H can be void. If H is void the load (s, b) is
incompatible, in the sense previously specified (no possibility of equilibrium
with purely compressive stresses).
Strictly admissible stress fields and load classification. In or-
der to formulate the theorems of Limit Analysis, I need to introduce the
following definitions.
On denoting, u the work of the load = (s, b) for the displacement
u, the load can be classified as follows:
1. ( is a collapse load)⇔ (∃u∗ ∈ Ks.t. , u∗ > 0),
2. ( is a limit load)⇔ (, u ≤ 0, ∀u ∈ K and ∃u∗ ∈ K −K ◦◦ s.t. , u∗
= 0),
3. ( is a safe load)⇔ (, u < 0 , ∀u ∈ K).
8
We suggest the reader to consult the book (Ambrosio et al., 2000)for a complete essay
on SBV functions and measure spaces.
132 M. Angelillo
I now introduce a useful definition. A stress field T ∈ H such that

trT < 0 and detT > 0 , ∀x ∈ Ω, is said to be strictly admissible.
Notice that, if T is strictly admissible, then at each point of Ω (that is
the open set Ω̇ to which the fixed part of the boundary ∂ΩD is added) it
results:
σ 1 < 0 , σ2 < 0 ,
σ1 , σ2 being the eigenvalues of T at the point x.
Kinematic Theorem. If is a collapse load (in the sense of item (1)
above) then H is void.
Static Theorem. If a strictly admissible stress field T exists, then the
load is safe (in the sense of item (3) above).
Limit Theorem. If H is not void and there exists u∗ ∈ K − K ◦◦
, u∗ = 0, then the load is limit (in the sense of item (2) above).
For the proof of these theorems I refer to the paper (Del Piero, 1998).
The reader must be warned that the proofs given by Del Piero refer to
a similar function space for the displacement but to a different functional
setting for the stress (namely L2 (Ω)). In the present paper I assume that
these theorem are still valid in the present larger setting for the stress and
smaller setting for the displacement9 .
2.9 Simple applications of the theorems of Limit Analysis

Example 1. Compressed wall/pier. The two simple BVP depicted in
Figure 16 are considered. The first one (Figure 16a) refers to a rectangular
wall compressed at the two bases by uniform normal tractions. The second
example (Figure 16b) is a wall compressed at the top base by a uniform
pressure load and fixed at the bottom base. By employing the Airy’s repre-
sentation and by using the static and kinematic theorems, it can be shown
that in case (a) the load is limit and in the second case the load is safe.
The data for F and dF/dν at the boundary, in case (a), are shown in
Figure 16c. In this case the only concave surface that can possibly satisfy
9
For general stress and strain fields that can be line Dirac deltas on a finite number

of regular arcs the internal work Ω T · E is not defined. Considering the restrictions
which define the sets H and K, that is taking into account the constraints on T and E,
the only case in which there are troubles in computing the internal work (if T and E
are so restricted) is when both the stress and the strain are singular on the same line
Γ, the line is curved and there is a stress discontinuity in the direction of the normal m
to Γ. A way to avoid this is to allow stress singularities on curved lines but to assume
that the support of the jumps of u is a segmentation, that is a line formed by the
union of a finite number of straight arcs.
a b c d
Figure 16. Compressed pier (wall). Pure traction problem: (a), Mixed problem: (b).
Corresponding data for F and dF/ν for the two cases.
the data shown in Figure 16c is the parabolic cylinder defined by
px21
F =− .
2
The uniqueness of this F can be proved by observing that the surface
defined by it coincides with the upper part of the convex hull of the curve
carrying the boundary datum for F. The properties of minimality of the
convex hull ensure the uniqueness of F and of the corresponding stress (see
Angelillo and Rosso (1995)), that is of the uniform uniaxial stress

0 0
{T} = .
0 −p
Since this is the only statically admissible stress field, H is a singleton
and we can say that the structure, with this load, is statically determined
(see Remark 4). Notice that based on the definition (3) above and on the
theorems of LA, the load is not safe. It is actually limit (see Limit The-
orem above) since by splitting the panel into two parts along any vertical
line Γ with a normal crack, the strain corresponding to this mechanism is
a horizontal uniaxial Dirac delta whose intensity has the value of the dis-
placement jump along Γ: the work of the load for this non-zero mechanism
is zero. Notice that the strain corresponding to this mechanism and the
unique statically admissible stress field are reconcilable in the sense of con-
dition (3), that is they represent a possible solution for the BVP. The fact
that, under these conditions, strain can increase indefinitely at constant
load is a typical feature of limit loads.
In case (b) the previous stress function can be corrected by adding a
term to it. Notice that the boundary is loaded only on the lateral sides and
on the top base (with the same load of case (a)), and that both the value
and the slope of the stress function can be modified along the bottom base
134 M. Angelillo
of the panel (see Figure 16d). The simplest correction with polinomia one
can think of, is
p x21 (a2 − x21 )2 x22

F =− −β .
2 a4
The corresponding stress is

2(a2 −x21 )2 8(a2 −x21 )x1 x2
−β −β
{T} = a4
8(a2 −x21 )x1 x2
a4
4(a2 −3x21 )x22
.
−β a4 −p + β a4
The trace and the determinant of T are then

4a2 x21 + 4a2 x22 − 2x41 − 12x21 x22
trT = −p + β −2 + ,
a4
2(a2 − x21 )2
4
detT = β 8
pa − β(4a2 + 20x21 )x22 .
a
√ √
2 2
If h ≤ 2 a then trT is always negative. If h > 2 a then trT is negative
on Ω if
pa2
β< ,
4h2 − 2a2
and detT is positive on Ω if
pa2
β< .
24h2
Then T is negative definite on Ω, and the load is safe, on the base of
the Static Theorem of LA, if the second inequality holds. For example, for
p
a square panel, if one takes β < 96 , then the stress given by the above
expression is strictly admissible and the load is safe. In Figure 17 the
stress functions emploied for cases (a) and (b) are shown side by side for
p
comparison, in the special case h = 2a, and putting for case (b), β = 400 .
Notice that the negative definiteness of T in case (b) is not uniform, since
on the lateral sides we must have detT = 0 and one of the two eigenvalues
of T must tend to zero as that part of the boundary is approached.
Remark 6. The bounds found on β give values of β vanishingly small
with respect to p, as the ratio h/a increases; if one takes β/p as a sort of
measure of the safety level of the load with respect to collapse: then slender
walls, under this kind of loading, tend to become less and less safe, as the
ratio h/a is increased.
a b
Figure 17. Compressed pier (wall): in (a) Airy’s stress function for the traction
problem. In (b) Airy’s stress function adopted for the mixed case.
a b
Figure 18. In (a) Airy’s folded stress function for the mixed problem. In (b) a scheme
of the corresponding stress is reported.
Remark 7. It is worth pointing out that the existence of a strictly

admissible stress field does not imply that the actual state of stress in the
body be of biaxial compression. If the material is rigid in compression,
any statically admissible stress has the same dignity and is theoretically
admissible for equilibrium.
The choice among these fields requires the introduction of more advanced
constitutive restrictions, allowing for shortening strains. When elasticity is
assumed it happens that s.a. stress fields which are not strictly admissible
are preferable, on an energetic ground, to strictly admissible ones. Therefore
the material exhibits both biaxial and uniaxial stress states (and fractures)
despite the existence of a strictly admissible stress field (see Exact solution
4, Subsection 3.7).
By using the Airy’s formulation singular s.a. stress fields can be easily
generated. By modifying, with some care, these singular fields, strictly
136 M. Angelillo
a b c
Figure 19. Smoothing of the folded stress function of Figure 18. Sections of the folded
surface and its smoothed transformation at x2 = 0: (a); the section of the parabolic
cylinder corresponding to the solution of the traction problem is reported for reference.
In (b), (c) a 3d view of the folded and smoothed surfaces is shown
admissible stress fields can be obtained. A non smooth F satisfying the b.c.
for case (b) of Figure 16 is depicted in Figure 18a.
The stress field corresponding to the non smooth Airy’s function depicted
in Figure 18a, is reported schematically in Figure 18b. The way in which
such a composite surface can be generated by the boundary data is explained
in detail in the next example. Notice that the half-span αa and the rise βh
of the arch-like structure depicted in Figure 18b can be chosen arbitrarily
provided that α, β vary in the interval [0, 1].
The process through which the singular statically admissible stress field
depicted in Figure 18 can be smoothed out and transformed into a strictly
admissible stress field is not discussed here for brevity. A 3d view of the
folded F and of its strictly admissible modification (together with the stress
function of case (a) used as a reference surface) is shown pictorially in Fig-
ure 19.
Example 2. Rocking of a rectangular panel. This is perhaps the

most classical problem for unilateral masonry-like materials, for which the
determination of the limit load under a horizontal force is trivial; therefore
this is the ideal example to understand how the theorems of limit analysis
can be applied to NT materials and how one can take advantage of singular
stress fields and of the stress function formulation.
The vertical load is fixed to the value q = PL and the intensity of the
concentrated horizontal force is taken as H = αP , α being a load parame-
ter. A trivial upper bound for α ( α◦ = 13 ) in the case shown in Figure 20a,
is found by considering the rocking mechanism of the panel (shown in Fig-
ure 20b) and applying the kinematic theorem. If one can find a statically
admissible stress field for the same value of α then α◦ is the limit value of
the load parameter.
x2
ΑP q
3
2
L
x1
L
a b
Figure 20. Rectangular panel loaded by vertical and horizontal forces. Loading
scheme: (a). Rocking mechanism: (b)
x2
1 s 0 s 4
s
ms ns
s
x1
2 s 3
a b c
Figure 21. Rectangular panel loaded by vertical and horizontal forces. Parametriza-
tion of the boundary: (a). Boundary value for F (moment m(s)): (a). Boundary value
for the normal derivative of F (axial force n(s)): (c)
138 M. Angelillo
x2
t
2
1.5
0.0
0.1 3
0.2 1.0
0.3
0.4
0.5
0.5
0.4
0.2
0.0 x1
0.2
0.4 0.0
a b
Figure 22. Rectangular panel loaded by vertical and horizontal forces for α = 1/4.
Composite stress function: (a). Corresponding support of the singular stress on the
planform and domain partition: (b).
For α < 13 , say α = 14 , a strictly admissible stress field can be con-

structed. To use the Airy’s formulation the loaded boundary is divided into
five parts (see Figure 21a where L = 1 is set).
1. The boundary is parametrized with the arc length s. The data m(s)
and n(s) are defined in terms of the load on the segments 0-1 and 1-2 (see
Figure 21b, c). From them the curve carrying the datum for F and the
gradient gradF at the boundary are computed.
Segment 0-1.
Parametrization: x(s) = {−s, 32 };
2
Moment: m(s) = − qs2 ;
Axial force: n(s) = 0 ;
2
Curve of boundary data: X(s) = {−s, 32 , − qs2 };
Gradient of F at the boundary: gradF ◦ (s) = {−m (s), n(s)} = {qs, 0}.
Segment 1-2.
Parametrization: x(s) = {− 13 , 32 − s};
Moment: m(s) = − 8q − qs 4 ;
Axial force: n(s) = − 2q ;
Curve of boundary data: X(s) = {− 12 , 32 − s, − 8q − qs
4 };
Gradient of F at the boundary: gradF ◦ (s) = {−n(s), m (s)} = { 2q , 4q }.
2. The data are extended inside the body with a uniaxial prolonga-
tion. Such a prolongation is obtained by constructing on each part γi of
the boundary, a ruled surface having as generating curve, the curve X(s)
carrying the Dirichelet data, and formed by the straight lines r directed as
the given loads and whose slope is specified by gradF ◦ (s). If the load is
zero, the direction of the line r is taken as the inward normal -n.
Segment 0-1
Load direction: k = {0, −1};
Propagation vector: v(s) = {k1 , k2 , gradF · k} = {0, −1, 0};
2
Parametric form of F 1 : y(s, ν) = X(s) + νv(s) = {−s, 32 − ν, − qs2 };
qx2
F 1 in terms of x1 , x2 : F 1 (x1, x2 ) = − 21 .
Segment 1-2
Load direction: k = {1, 0};
Propagation vector: v(s) = {k1 , k2 , gradF · k} = {1, 0, 2q };
Parametric form of F 2 : y(s, ν) = X(s) + νv(s) =
{− 12 + ν, 32 − s, − p8 (1 − 2qs + 4νq)};
F 2 in terms of x1 , x2 : F 2 (x1, x2 ) = − 4q (1 − 2x1 − x2 ).
3. The two surfaces F 1 and F 2 intersect along a curve (represented in
Figure 22a) whose projection Γ on the “planform” is determined, in explicit
form. By solving the equation F 1 = F 2 , one obtains the following equation
for Γ:
x2 = 1 − 2x1 − 2x21 .
The curve Γ is a parabola, passing through the point {− 12 , 32 } and inter-

√
secting the base at x◦1 = 12 ( 3 − 1). Γ splits the rectangle into two parts
Ω2 , Ω3 (Figure 22b).
4. Along Γ the stress is singular: it is a concentrated axial force N
whose intensity is determined by the jump of slope ΔF of the composite
surface F in the direction of h (Figure 22ab). Since the gradient jump
across Γ is gradF 1 − gradF 2 = −q{ 12 + x1 , 14 } and the unit normal h to Γ

is h = {2 + 4x1 , 1}/ 5 + 16x1 + 16x21 , we have
*
1
ΔF = (gradF 1 − gradF 2 ) · h = − q 5 + 16x1 + 16x21 .
4
The stress corresponding to this composite F is zero in Ω1 , uniaxial

inside Ω2 , and a Dirac delta stress of intensity N = ΔF , balancing the
stress jump on Γ due to the stress discontinuity, at the interface between
Ω
+ 1 1and Ω2 (Figure
√ , 23a). The graph of N along the curve Γ, in the interval
− 2 , 12 ( 3 − 1) , is shown in Figure 23b.
Remark 8. Notice that the concentrated force emerging at the bottom
edge is the trace of the singular stress field T(x1 ) = N (x1 )δ(Γ)t(x1 )⊗t(x1 ),
t being the unit tangent vector to Γ: {t} = {h2 , −h1 }. The value of such
force is R = N (x◦1 )δ(x◦1 )t(x◦1 ), i.e. it is not given by the Cauchy formula
s(n) = Tn, valid for absolutely continuous stress vectors s.
140 M. Angelillo
pL
p L4 p
T0
p
3
p L
2
p L4
a b
Stress corresponding to the stress function of Figure 22: (a). Graph of the axial force
along the arch: (b).
A s.a. stress field for α = 13 is obtained by following the same steps as

before. The two functions F 1 and F 2 are in this case:
x21 1
F 1 (x1 , x2 ) = −q , F 2 (x1 , x2 ) = − q(9 − 12x1 − 8x2 ) .
2 24
They intersect along the curve Γ depicted in Figure 24a, whose projection
on the planform is given by the equation:
3
x2 = (3 − 4x1 − 4x21 ) .
8
Γ is a parabola passing through the points [−1/2, 3/2], [1/2, 0] (thus is
the two opposite top and bottom corners of the rectangle: Figure 24b). The
jump of slope along Γ is
*
1
ΔF (x1 ) = (gradF 1 − gradF 2 )|Γ · h(x1 ) = − q 13 + 36x1 + 36x21 .
6
the corresponding axial force N along Γ, in the interval [−0.5, 0.5], is re-
ported in Figure 24d.
In Figure 24c the stress field corresponding to the composite surface
F 1 , F 2 is reported. Such a field is not strictly admissible and does zero
work for the mechanism of Figure 20b, therefore α = 1/3 is the limit load.
x2
m
t
2
0.0
1.5
0.2 3
0.4 1.0
0.6
0.5
0.4
0.2
0.0 x1
0.2
0.4 0.0
a b
pL
p L3 p
T0
pL
p L3
c d
Strees function: (a). Support of the singular stress on the planform and domain partition:
(b). Stress corresponding to the stress function depicted in (a): (c). Graph of the axial
force along the arch: (d).
The stress field that was constructed for α = 1/4 is not strictly admissible
(actually the stress that we derived from the folded stress function is on the
boundary of the cone Sym− all over Ω), but, for any value of the load
parameter α < 31 , the previous construction can be used to generate a
strictly admissible stress field. The method proceeds as follows:
Assume that a value of the parameter, say α ) < 13 , is given. If one
removes a portion p = q(1 − 3) α) from the given uniform load, the load
becomes limit, that is the parabola hits the right corner of the rectangle
and, as before, a stress field T1 can be constructed. Now I consider the
panel under the action of the uniform vertical load p applied at the top
base, and take the strictly admissible solution, say T2 , constructed for the
142 M. Angelillo
previous example (the compressed wall). The sum of the two stress fields
T1 + T2 is strictly statically admissible, since Sym− is convex, the second
stress state is strictly inside the cone, and then the sum of the two stress
fields is also strictly inside the cone for any x ∈ Ω.
Remark 9. In the previous example statically admissible stress fields
were produced through the stress function formulation. The overdetermined
problem of equilibrium was reduced to a determined problem by restricting
the search to a special class of stress functions, namely that of ruled surfaces
(that is surfaces composed of straight lines). The stress fields corresponding
to these surfaces are either uniaxial or null, that is the stress is limit almost
everywhere in the body and the equilibrium problem is determined, since to
the two differential equations of equilibrium the algebraic condition that the
stress is uniaxial (detT = 0) is added10 . Any statically admissible uniaxial
stress field balanced with zero body forces b, has one family of isostatic lines
(the ones corresponding to the negative eigenvalue) composed of straight
lines (see Remark 3 and Figure 6). These straight lines, being actually the
projections of the straight lines generating the ruled stress surface associated
to T, are called compression rays 11 .
The differential problem is parabolic, and the stresses inside Ω are deter-
mined by the boundary data. By propagating the loads from opposite parts
of the boundary the corresponding stress fields are usually unbalanced at
the interface between the two fields: singular stress fields must be admitted
along the interface. Both the value of the singular stress and the shape
and location of the interface is determined by equilibrium. With the stress
function formulation both the form and the location of the interface, besides
the intensity of the axial contact force along Γ, are derived by intersecting
two contiguous stress surfaces.
Example 3. Lintel under vertical and horizontal loads. A rect-

angular wall beam, supported at the bottom corners A, B, submitted to
vertical and horizontal loads applied along its top edge, is shown in Fig-
ure 25a to which I refer for notations. This element can be representative
of lintels, that is the transverse structures connecting the piers in masonry
portals or in sequences of arches, when the effect of the loads transmitted
to the arch from other parts of the structure, prevails on the self load, and
10
This assumption is the basic hypothesis of the Tension Field Theory (see (Mansfield,
1969)), a simplified unilateral model for thin elastic membranes, analogous to the NT
model for masonry structures (see Remark 3 and Subsection 3.8).
11
This is a terminology similar to that adopted in Tension Field Theory, see Fortunato
(2010).
the diffuse effect of body forces will be neglected. In Example 5 below the
effect of uniform vertical body forces is considered.
The lintel’s lower edge is actually often curved (see dashed line in Fig-
ure 25a), this feature being in keeping with the kind of stress state that
I wish to consider in the element, as we shall see below. The presence of
this arched intrados is necessary for equilibrium, if vertical body forces are
considered.
By adopting the previous approach, in the case at hand I formulate the
equilibrium problem of the lintel as follows.
The loads acting on Ω consists of a distributed load q, applied along
the top edge of Ω and having two components {q1 , q2 } (see Figure 25b).
The supports A, B reacts with two forces RA , RB , whose components are
denoted {H(A), V (A)}, {−H(B), V (B)} ; the lateral and the lower edges
are unloaded.
Restricting to at most uniaxial stress fields and denoting g = q1 /q2 the
slope of the applied load with respect to x2 , the stress field, in the upper
part of the domain, is a uniaxial field in the direction of the compression
rays emanating from the top edge, whose slope with respect to x2 is g.
Calling τ the length along the (straight) top edge of the domain, measured
from O, I also assume that the slope g is so restricted:
τ L−τ
g(0) = 0 , g(L) = 0 , − ≤ g(τ ) ≤ , |g(τ )| ,
h◦ h◦
that is the initial and final slopes are zero (then the two extreme compression
rays run along the lateral edges), and the compression rays go from base to
base and do not cross each other inside the rectangle enclosing Ω.
The stress field in the lower part of the domain is zero. The upper and
the lower regions are separated by a common boundary Γ, passing through
A and B, that is parametrized in terms of τ as follows
Γ = {{xΓ } = {τ + g(τ )y(τ ), y(τ )} , τ ∈ {0, L}} ,

carrying a concentrated axial force.
We denote H and V the horizontal and vertical components of the axial
contact force N, arising on Γ in order to equilibrate the stress jump 12 . By
imposing the equilibrium of the piece ℘ of Ω represented in Figure 25a, the
following set of differential equations is obtained
y
H = q1 , V = q2 , V = H ,
1 + gy + g y
12
For the positive sign of these two components I refer to the choice reported in Fig-
ure 25a.
144 M. Angelillo
q x
q

f°
HA A B HB
VA VB A B
y
a b
Span Load
q1
Τ
Q
q2
c d
Span Load
q1
q2 Q
e f
Figure 25. Lintel loaded by vertical and horizontal forces. Geometry of the panel:
(a). Forces acting on the panel, compression rays and arch: (b). Horizontal and vertical
load for two special cases (c, e) corresponding to the equilibrium solutions (d, f).
to be solved for H, V, y with the boundary conditions
y(0) = h◦ , y(L) = h◦ , y(τ ◦ ) = 0 , y (τ ◦ ) = 0 ,
τ ◦ being an unknown position along the upper edge. In the special case in
which the horizontal and vertical loads have the form
6q ◦ (L − τ )2 τ
q1 (τ ) = , q2 (τ ) = q ◦ ,
L3
that is the vertical load is uniform, the horizontal load is parabolic, and the
horizontal load resultant is half of the vertical resultant, the solution is

◦ (h◦ − L)2 6L2 τ 2 − 8Lτ 3 + 3τ 4
H=q + ,
8h◦ 2L3
1 ◦ ◦
V = q (h − L + 2τ ) ,
2
h◦ L3 (h◦ − L + 2τ )2
y= ,
L5 + (h◦ )2 a(τ ) − 2h◦ b(τ )
where
a(τ ) = L3 − 24L2 τ + 48Lτ 2 − 24τ 3 ,

b(τ ) = L4 − 12L3 τ + 36L2 τ 2 − 44Lτ 3 + 18τ 4 .
The solution of this special case for h◦ = L
2 is reported in Figure 25.
Example 4. Lintel under a vertical uniform load and a horizontal

concentrated force. Here I apply the stress function method to solve a
problem similar to the previous one, namely the equilibrium of a rectangular,
simply supported panel under a vertical uniform load applied along the top
edge and a horizontal concentrated force acting at the left upper corner
(Figure 26a). Again it is assumed that the supports in A and B react with
two concentrated forces passing through A and B. These two forces are
expressed in the form

q ◦ L F h◦
{R(A)} = H, − ,
2 L

q ◦ L F h◦
{R(B)} = −H − F, + ,
2 L
The corresponding boundary data for the stress function are reported
pictorially in Figure 26b. Prolongating the data with ruled surfaces having
the prescribed slope and direction at the boundary, a surface composed by
three parts, a parabolic cylinder and two planes, is generated. Notice that
the value of the thrust H could be chosen arbitrarily, within some limits,
and that, in Figure 26b, it is set to a value such that the intersection of
the lower plane with the parabolic cylinder, touches the upper boundary
(minimal thrust).
The intersection of the three surfaces (Figure 26c) determines the 1d
structure depicted in Figure 26d. The stress is uniaxial between the top
edge and the structure Γ, null below the estrados of the arch; the singular
part of it is concentrated on the support of the folds of F , that is on the
structure Γ.
146 M. Angelillo
F q
A B
L
a b
' ''
'''
c d
Figure 26. Lintel loaded by vertical and horizontal forces: (a). Boundary data: (b).
Composite stress function: (c). Support of the singular stress, arch structure: (d).
Example 5. Lintel under homogeneous vertical body forces. A

rectangular wall beam of heigth h◦ and width L, supported at the bottom
corners A, B and submitted to homogeneous vertical body forces, is shown
in Figure 27a, to which I refer for notations. This element is representative
of lintels for which the effect of body forces is not negligible. The parameter
◦
p◦ = 2hL defines the aspect ratio of the lintel.
The lintel’s lower edge is actually considered as symmetric (with respect
to the center line) and curved (see dashed line in Figure 27a). For now the
shape of this symmetric curve Γ is not specified, whilst its rise is fixed and
called f ◦ = (1 − α)h◦ , α being a parameter ranging in the open interval
(0, 1). Notice that αh◦ is the thickness of the arch at the keystone.
The curve is described parametrically, in the Cartesian frame depicted
in Figure 27a, as follows
Γ = {{xΓ } = {x, f (x)} , x ∈ [−L/2, L/2]} ,
with
f (L/2) = f (−L/2) = 0, f (0) = (1 − α)h◦ .

9 1
Α Α
10 100
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0.0 0.0
0.4 0.2 0.0 0.2 0.4 0.4 0.2 0.0 0.2 0.4
a b c
Figure 27. Lintel loaded by uniform body forces. Geometry of the panel: (a). Lower
arch for α = 9/10: (b) and for α = 1/100: (c).
I construct two different statically admissible solutions for the equilib-

rium problem of the lintel as described in what follows.
First solution: maximal thrust. With the first solution I assume that
inside the body the stress field is a.e. uniaxial and vertical, and an arch, on
which the normal stress is concentrated, forms along the bottom boundary
curve Γ. The loads acting on the wall produce a resultant vector Q, having
components {0, −Q}, with Q = −γ area(Ω), and located at the center of
the wall. The supports A, B react with two forces RA , RB , whose com-
ponents are denoted {H(A), V (A) = Q/2}, {−H(B), V (B) = Q/2}, where
H(A) = H(B) = H, H being the unknown thrust of the arch; the rest of
the boundary is unloaded.
The uniaxial stress field of simple compression in the vertical direction,
has a non-vanishing component of the form
σ = −γ(h◦ − y) .
Therefore the arch Γ is subjected to the vertical distributed load (per unit
horizontal length) q = −γ(h◦ − f (x)), f (x) being the vertical coordinate of
the arch Γ. For equilibrium the shape of the arch must satisfy the equation
h◦ − f
f = −γ .
H
that one can solve (for f and H) with the boundary conditions
f (L/2) = f (−L/2) = 0, f (0) = (1 − α)h◦ .

The solution is
√γ (L−2x)
√γ (L+2x)
√γ
1 + eL H −e 2 H −e 2 H h◦
f= √γ ,
1 + eL H
148 M. Angelillo
L2
H=γ √ .
Log2 2−α2 −2 1−α2
α2
The value of the force resultant is then
2

√

√
◦ 2L 1 − α2 + 1 − α2 Log 2 − α2 − 2 1 − α2 − 2Log(α)
Q = γp
√ √
−1 + 1 − α2 Log2 2−α +2
2 1−α2
α 2 .
In Figures 27b,c the shape of the lower curve determined through equi-
librium is compared to the actual shape of the intrados of the lintel, for
two extreme special cases (α = 0.9, α = 0.01). From these two pictures
one can see that, in the first case, the statically admissible stress field here
constructed can be accepted as an approximate equilibrated solution (by
neglecting the slight geometrical difference between the two curves); whilst,
in the second case, there is a large portion of the domain (located below the
arch) that remains out of the picture.
It is to be pointed out that for α → 1, the thrust force tends to +∞.
Second solution: reducing the thrust.The domain Ω is divided into
two parts: the part Ω comprised between the curve Γ and the curve
Γ = {{xΓ } = {{x, g(x)} , x ∈ [−L/2, L/2]}} ,

with
1
g(x) = f (x) ,
1−α
and the part Ω , that is the part of Ω located above Γ (see Figure 28). A
solution of the equilibrium problem for the region Ω is constructed first;
the problem of equilibrium of the region Ω is solved in a second phase, and
superimposed to the previous stress field.
Again I restrict to at most uniaxial stress fields, assuming that the stress
field is convected to one of the two families of curvilinear lines ϑ1 = x, ϑ2 =
λ, defined by
x1 = x , x2 = h(x, λ) ,
with
h(x, λ) = (1 − α + λα) g(x) ,

where, x ∈ [−L/2, L/2], λ ∈ (0, 1) and
Figure 28. Lintel loaded by uniform body forces: geometry of the panel and domain
partition for the second solution.
1
g(x) = f (x)
1−α
is the function describing the upper curve Γ .
The natural and reciprocal base vectors associated to this curvilinear
system, in components in the Cartesian reference depicted in Figure 28, are
{a1 } = {1, h,1 } , {a2 } = {0, h,2 } ,

1 h,1 1
a = {1, 0} , a2 = − , .
h,2 h,2
The uniaxial stress field here considered has the form
T = σa1 ⊗ a1 ,
σ being a function of (x, λ), describing the intensity of the stress field, to
be found, together with f , by solving the equilibrium equation
∂
(σa1 ⊗ a1 ) aα + b = 0 .
∂ϑα
By projecting this vector equation along the natural bases, after some
algebra, the following system of second order differential equations, is ob-
tained:
h,12
σ,1 + σ =0,
h,2
γ h,11
− +σ =0,
h,2 h,2
150 M. Angelillo
∂h
Recalling that h,2 = ∂λ = αg, since h,2 is always different from zero

inside Ω , integrating the first equation, one obtains
m(λ)
σ= ,
h,2
m(λ) being an unknown function of λ. Substituting into the second equation
one obtains
h,11
γ = m(λ) .
h,2
∂h ∂2h
Therefore, taking into account that h,2 = ∂λ = αg and h,11 = ∂x2 =

(1 − α + αλ)g , the following two conditions are obtained:
α
g = −kg , m(λ) = −γ ,
k(1 − α + αλ)
k being a constant that I will assume positive.
By solving the first equation with the boundary conditions g(−L/2) =
0, g(0) = h◦ the following solution is obtained
- -√ . .
√ kL √
◦ ◦
g = h cos kx + cot sin kx .
2
By imposing the condition g (0) = 0 the value of the constant k is

determined:
π2
k= ,
L2
and then the form of the curve Γ is given by
πx
g ◦ = h◦ cos ,
L
that is
πx / 0
◦ L L
Γ = {x} = x, h cos , x∈ − , .
L 2 2
The physical uniaxial stress component

a1 a1 L2 γSec πx
σ = T(11) = T · ⊗ =− ◦ 2 L
,
|a1 | |a1 | h π (1 + α(−1 + λ))
along the ϑ1 curves, is depicted in the graphic of Figure 29.
Α 4
5
0.5
0.4
0.3
0.2
0.1
0.0
0.4 0.2 0.0 0.2 0.4
Figure 29. Contour plot of the stress σ in the region Ω .
It is assumed that the load which is given above the curve Γ , is taken by
an arch Γ , that is by a concentrated stress with support on Γ . The form
of this curve, springing from the points A, B, determined by the form of the
load through equilibrium, should be located below the load (that is below
the curve Γ ) and contained inside the masonry (that is above the curve Γ).
In this way, the load can be transmitted to the arch by compressive uniaxial
vertical stresses, linearly varying inside Ω , and constant outside it.
The vertical load that I consider acting on Γ is then
q = γ(s + h◦ − g ◦ ) ,
γs being the effect of a given uniform over-load. By solving the equilibrium

equation for the arch, with the conditions that the arch passes through the
points A and B, and through the point P of coordinates {0, hy }, one obtains

hy π 2 (Lp◦ + 2s) L2 − 4x2 − 8L3 p◦ cos πx
L
Γ = x, ,
L2 (L (−8 + π 2 ) p◦ + 2π 2 s)

x ∈ (−L/2, L/2) ,
◦
where p◦ = 2hL is the aspect ratio of the lintel. In Figure 30 the form of the
arch for p◦ = 1 , s = L/8 and three special cases is reported.
Finally the values of the vertical reactions and of the thrust forces in A
and B, due to the compound effect of the first and the second equilibrium
solutions are:
(π − 4p◦ + 2πp◦ )γ p◦ αγ
Q/2 = + , (23)
8π 2π
152 M. Angelillo
2 1
Α hy
3 5
0.5
0.4
0.3
0.2
0.1
0.0
0.4 0.2 0.0 0.2 0.4
1 1
Α hy
3 3
0.5
0.4
0.3
0.2
0.1
0.0
0.4 0.2 0.0 0.2 0.4
1 21
Α hy
6 50
0.5
0.4
0.3
0.2
0.1
0.0
0.4 0.2 0.0 0.2 0.4
Figure 30. Lintel loaded by uniform body forces: geometry of the arch Γ carrying
the load of the part Ω for three special cases: from top top to bottom, α = 2/3, α = 1/3,
α = 1/6.
l L l
D h C
h0
f b
A B E
h1
G F
Figure 31. Typical masonry portal. L: span of the arch, f : rise, h: thickness. ABCD:
lintel, h◦ : height of the lintel. GF EB: pier, h1 : height of the pier, : width of the pier.

−16p◦ + π 2 (1 + 2p◦ ) γ γLog(1 − α)
H= 2
− . (24)
32hy π π2
2.10 Derand’s rule

Based on the unilateral model the safety of the structure is a matter of
geometry rather than of strength of materials, in keeping with the spirit of
the “rules of proportion” used by the ancient architects for masonry design.
The essential characteristic of all these rules is that they are proportional
and that they control the overall form of the structure of the building,
regardless of its size.
Most of these rules, being not written, have faded and progressively been
forgotten, but it is evident that the great buildings of the past could not
have been built without some kind of rich and complex knowledge, because
its application resulted in astonishing realizations such as the Pantheon and
the Gothic cathedrals. One of the rules that is survived, is the Derand’s rule
(see Benvenuto (1991) and Huerta (2008)), and here I try to assess the safety
margins it assures on the basis of singular stress fields for the masonry-like
model. The results here obtained confirm that this often criticized rule is
actually a sharp technical tool.
The rule applied to a portal. In 1643 François Derand published his

monumental work on vaults (Derand (1643)), where the rule for determin-
ing the thickness of the walls, needed to support the thrust of the vaults,
appears. Here I apply the rule to a simple masonry portal.
If Figure 31, a masonry panel Ω, having the form of a portal, fixed at
the base and unloaded on the rest of the boundary, is depicted. The panel
154 M. Angelillo
0.8
0.6
0.4
0.25
0.20 0.2
0.15
0.10
0.05
0.4 0.2 0.2 0.4 0.4 0.2 0.2 0.4
a b
Figure 32. Derand’s rule for shallow and pointed arches.
is loaded by its own weight b, considered uniform and acting vertically

downward, and by a small uniform vertical load applied along the top edge.
The main nomenclature concerning such a typical structure is reported in
Figure 31, to which I refer for notations. For simplicity I take the unit of
length and force in such a way that L = 1 and |b| = γ = 1.
Derand’s rule applies both to segmental and pointed arches as illustrated
in Figure 32a,b, to which I refer for notations. Based on this rule the thick-
ness of the wall is the segment c indicated as a thick line in Figure 32a,b,
that is given by the projection of the arch segment, reported in the same
figure, on the horizontal line. The rule is that the arch segment whose
projection determines c , is 13 of the total length of the intrados.
On introducing the angle ϑ = artan(2(1 − α)h◦ ) (in which α is the
parameter introduced in the previous example) and noticing that the radii
for the segmental (ϑ < π/4) and pointed (ϑ > π/4) arches are
L L
r = , r = ,
2sin2ϑ 2(1 + cos2ϑ)
the rule gives

2 −
L
r sin 23 ϑ ,
c =
(25)
r
1 − cos 23 (π − 2ϑ) .
Limit Analysis solution We analyze the equilibrium of the portal con-

sidering separately the equilibrium of the lintel ABCD and of the pier
BEFG, and then assembling the two parts.
Lintel. The effect of the dead load b on the lintel depicted in Figure 33a
is analyzed by adopting the second s.a. stress field, introduced in the pre-
vious Section. Based on this solution the forces transmitted to the piers at
H
B E
G F
Figure 33. Forces acting of the pier. Q, H are the vertical and horizontal forces
trasmitted by the lintel; the uniformly distributed load at the top base is the effect of
the weight of the part BEC; b is the body force.
the points A, B, have the vertical and horizontal components given by (23)
and (24).
The concentrated and distributed forces acting on the right pier of the
portal, are depicted in Figure 33a.
Pier. These forces and their slope can be used as data for the equilibrium
of the pier, as shown in Figure 33. In the pier BEF G I consider an arch
Γ springing from the point B with the slope given by 2H/Q. The arch Γ
is represented by its graph z(x) in a right handed reference {O; x, y} with
origin in B and y directed vertically, downward. The arch Γ carries the
inclined thrust force coming from the lintel and the weight of the part of
the wall above it, besides the over-load s. A uniaxial stress field, linearly
varying with y, and balancing the, constant, vertical load b (|b| = γ = 1),
is considered both above and below the arch: the upper part is sustained by
the arch; the part below is supported by the soil. The equilibrium conditions
give, in this case, the following equation for z:
1
z = (s + h◦ + z) .
H
This equation can be integrated with the conditions
2H
z(0) = 0, z (0) = .
Q
156 M. Angelillo
The solution is

◦ 1 ◦L −βx 2H
z = (p + s)(1 − e βx
)(−1 + e ) + (1 + eβx ) ,
2 2 Qβ 2

where β = 1/H. The form taken by the arch Γ depends on the aspect
ratio, p◦ , of the lintel, and on the parameter α defining the thickness of the
arch at the key stone. To compare the results with the rule of Derand, I
introduce the parameter
λ = tan(ϑ) ,
and notice that
(1 − α)2h◦
λ= = (1 − α)p◦ .
L
Then z ◦ can be expressed as a function of p◦ and λ and the limit length

c , that is the intersection of the arch Γ with the base, can be determined
for different values of the heigth h.
Effect of the heigth: Gil’s rule. Derand’rule makes no reference to the

heigth of the pier, though other rules of proportion relating the shape of the
arch to the heigth of the piers are known. One of these is the seventh rule
of Gil (Rodrigo Gil de Hontanon, see Huerta (2004)) for which the heigth
of the pier is approximately twice the rise of the arch.
In Figure 34 a comparison between the values of c given by Derand’s
rule with that obtained with the RNT model under Gil’s prescription (that
is by putting for the heigth: h = 2f ), is presented. The values of c are
plotted against the parameter λ, for a few values of the parameter p◦ , in the
1 9
range of practical interest: { 10 , 10 }. The different curves refer to different
◦
values of p .
The values of c predicted with the unilateral model (which being ob-
tained with the static theorem of LA are safe and then upper bounds to the
limit length c ) are always smaller than the values prescribed by Derand in
the range of values of the aspect ratio p◦ here explored. Notice that the
range considered covers both the cases of shallow and pointed arches and
that, based on the NT theory, Derand’s prescription appears as a rather
sharp rule.
0.5
0.4
0.3
0.2
0.1
0.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0
Figure 34. Comparison of Derand’s (and Gil’s) rule with LA predictions. The con-
tinuous upper curve represents the graphic of Derand’s rule (25). The sequence of curves
represents the LA predictions of c for (from left to right) p◦ = { 18 , 14 , 12 , 1, 2, 3}.
3 Model one (NENT)

In this section the main ingredients of the theory concerning model one for
masonry materials, namely the Normal Elastic No-Tension (NENT) mate-
rial, that is the n > 1 version of the 1d model introduced in Chapter 1,
are presented. The constitutive assumptions, the balance equations and
the boundary conditions are introduced and the boundary value problem
for NENT materials is formulated. On introducing a proper form of stored
elastic energy, the minimum problem for the potential energy functional
provides the existence of the solution for this boundary value problem. The
main issues connected with energy minimization and a number of illustrative
exact solutions and examples are discussed.
3.1 A Premise on Minimum problems and the peculiarity of NT

materials
Several problems in physics and Engineering can be formulated as a
minimum search: a functional describing the energy of the system and de-
pending on an unknown function has to be minimized over the set of all
admissible functions. In the context of elasticity, denoting u the displace-
ment vector describing the deformation from a given configuration Ω, of a
material body subject to a system of applied forces and given boundary
displacements, the energy can be written in the form
E(u) = − , u + U (u) , (26)

158 M. Angelillo
where -, u is the linear form describing the potential energy of the load,
that is minus the work of the load = (s, b) for the displacement u, and
U(u) is the stored energy functional. The simplest case in which existence
theorems for the minimum problem
min E(u) , (27)

u∈K
K being the subset of a convenient Banach space T (Ω̃), occurs when E(u) is
lower semicontinuous and coercive, in the sense that E(u) → ∞ as uS →
∞. Unfortunately, as we shall see, the energy functional for masonry-like
materials is not coercive (see also the form that the strain energy density
takes in the 1d case (see Figure 10.a of Chapter 1) and therefore some
supplementary hypotheses have to be added in order to get existence. For
instance, existence of solution in the case of Normal Elastic No-Tension
materials, under small strains and in the 2d context, has been established
by Giaquinta and Giusti (in Giaquinta and Giusti (1985)), under a so called
safe load condition. Specifically the form of energy to be minimized in the
case of NENT materials is

E(u) = − s·u− b·u+ Φ(E(u)) , (28)
∂ΩN Ω Ω
where E(u) is the infinitesimal strain associated to u, Φ(E(u)) is the elastic

energy density, and the stress T is related to E through the relation
∂Φ
T= . (29)
∂E
The minimizer of E(u) is searched for u ∈ S(Ω) (a Banach space) and
u = u on ∂ΩD . Concerning the nature of such Banach space, that is the
regularity of u, actually it seems reasonable to expect, based on the non
coercivity of Φ (see Figure 10.a of Chapter 1) and on the at most linear
growth of the energy E(u), that u be possibly discontinuous and the cor-
responding deformation not to be absolutely continuous. A popular choice
for S(Ω) is the space BD(Ω) of bounded deformations, though one expects
that, apart from very special cases and away from the case of collapse loads,
the minimizers should be much more regular. We shall return to the vari-
ational formulation of equilibrium for masonry materials after the BVP in
its strong form is considered.
3.2 The Boundary Value Problem for NENT materials

Constitutive restrictions. It is assumed that the structure Ω ∈ n (here
n = 2), loaded by the given tractions s on the part ∂ΩN of the boundary, and
subject to given displacements u on the complementary, constrained part of

the boundary ∂ΩD , is in equilibrium under the action of the given surface
and body loads (s, b) and distortions E, and undergoes small displacements
u and strains E(u)13 .
Notice again that Ω is considered closed on ∂ΩD and open on the rest of
the boundary.
We consider that the structure Ω is composed of Normal Elastic No-
Tension material, that is the stress T is negative semidefinite
T ∈ Sym− , (30)
the effective strain, that is the total infinitesimal strain E(u) minus the
eigenstrains E, is decomposed additively into the sum of an elastic part Ee
and an anelastic part Ea , namely:
E(u) = Ee + Ea + E , (31)
the elastic part being linearly related to the stress T:
Ee = A [T] , (32)
the latent anelastic part (a measure for fracture) being positive semidefinite
Ea ∈ Sym+ , (33)
and the stress T doing no work for the corresponding latent strain Ea
T · Ea = 0 . (34)
Notice that in the plane case (n=2) conditions (30), (33), can be rewrit-
ten as
tr T ≤ 0 , det T ≥ 0 , (35)
tr Ea ≥ 0 , det Ea ≥ 0 . (36)
Remark 10. It is to be pointed out that a NENT material, that is a
material defined by the restrictions (30) through (34), is elastic in the sense
that, given the total strain E(u), the stress can be univocally determined.
The material is actually hyperelastic as we shall see later in what follows.
13
When eigenstrains are considered, under the small strain assumption, the total strain
E(u) is decomposed additively as follows: E(u)=E∗ + E, E∗ being the effective strain
of the material.
160 M. Angelillo
Equilibrium problem. In order to avoid trivial incompatible loads (s,b),

I assume again that the tractions s satisfy condition (4).
In equilibrium, the stress field T must be balanced with b, that is
divT + b = 0 , (37)
and the stress T and the displacement u must comply with the boundary
conditions
Tn = s , on ∂ΩN , u = u , on ∂ΩD . (38)
Boundary Value Problem. The boundary value problem for NENT

materials can be formulated as follows:
Given a bounded open set Ω and the partition ∂Ω = ∂ΩD ∪ ∂ΩN , find
the fields u, T defined over Ω ∪ ∂ΩD , such that the material restrictions
(30) through (34), the balance equations (37) and the b.c. (38) are satisfied.
Peculiar features of the fracture field for NENT materials, in the

three sets of the Fundamental Partition. As it was already observed
in the previous Chapter concerning the RNT model, to each statically ad-
missible stress field T defined over Ω ∪ ∂ΩD , is associated a Fundamental
Partition of the domain: Ω = Ω1 ∪ Ω2 ∪ Ω3 , of biaxial, uniaxial and zero
stress (see Section 2.3).
In Ω2 the equilibrium equations and the condition detT = 0, form a
system of three equations in the three unknown independent components
of T. The differential problem is parabolic and the stress is determined by
equilibrium regardless of the material response.
In Ω1 fractures are not possible, that is Ea = 0.
In Ω3 , where T = 0, any positive semidefinite fracture field is possible.
Also in Ω2 the material can be fractured. The necessity of fractures is
naturally produced by the problem being statically determined: the elastic
strain associated to the statically determined stress is generally not compati-
ble and fracture strains are required to restore compatibility. The normality
condition requires that in Ω2 the fractures must open up orthogonally to
the isostatic compression lines, therefore, as a consequence of normality, on
a crack line (that is a line where the strain is a line Dirac delta) the jump
of displacement must be purely orthogonal to the crack line. If the strain
is regular, in a curvilinear frame with natural bases a1 , a2 coincident with
the eigenvectors of T, T admits the representation
a1 ⊗ a1
T=σ ,
a1 · a1
σ being the only non zero, negative eigenvalue of T. Normality implies the
following form for the fracture strain Ea :
a2 ⊗ a2
Ea = λ ,
a 2 · a2
λ being the only non zero, non negative eigenvalue of Ea . Then, regardless
of the possible elastic anisotropy of the material, the principal directions
of stress and anelastic strain of a NENT material are always coincident all
over Ω.
3.3 Strain energy density in 2d

Most of the features illustrated by the 1d ENT model (that is Model 1
introduced in Chapter 1) are transposed to the 2d case, except that to obtain
hyperelasticity there is a price to pay: the assumption of normality (that is
the constituitve assumptions (33), (34)) must be made on the total latent
strain Ea . Such an assumption implies that on a discontinuity line Γ for the
displacement u, that is on the support of a concentrated fracture, the jump
of displacement must be orthogonal to Γ. Therefore sliding is forbidden
on fracture lines. It is shown by Del Piero in Del Piero (1989) that, for
NENT materials, the major symmetry of A is necessary and sufficient to
get existence of an elastic energy function. In the isotropic case the form
of the energy density Φ can be constructed explicitly14 in terms of the
eigenvalues e1 , e2 of E (here e1 < e2 is assumed). In the case of generalized
plane stress, one has (see Figure 35)
⎧
⎪
⎨0 , e1 ≥ 0 and e2 ≥ 0 ,
Φ = 12 Ee21 , e < 0 and e2 ≥ −νe1 ,
⎪
⎩1 E
1
2 1−ν 2 e21 + e22 + 2νe1 e2 e1 < 0 and e2 < −νe1 ,
E, ν being the Young modulus and the Poisson ratio. Notice that the stress
T derived from Φ satisfies identically the no-tension restriction (30), that
is there is no need to impose it as a constraint.
14
In the general anisotropic case the explicit symbolic form of Φ is not known and must
be constructed numerically, case by case.
162 M. Angelillo
0
0 e2
e1 0
Figure 35. Strain energy density of isotropic NENT materials: zero energy region:
light grey; parabolic energy region: grey; elliptic energy region: dark grey.
3.4 Function spaces for the potential energy functional.

For hyperelastic materials equilibrium states of the body can be searched
as minimizers of the total potential energy (28)
min E(u) , (39)

u∈K
K being the set of kinematically admissible displacement fields for NENT

materials, defined as follows
K = {u ∈ T (Ω) s.t. u = u on ∂ΩD } , (40)
where Ω = Ω̇∪∂ΩD and T(Ω) is a convenient Banach space. As it is evident

from the picture of Figure 35, the strain energy function characterizing
NENT materials is not coercive.
Coercivity of the total potential energy can be restored by considering
the following supersafe load condition on the applied forces (here I consider
for simplicity the case of fixed boundary constraints: u = 0, on ∂ΩD )
Supersafe load condition. The load p, b is said to be supersafe if
there exists at least one stress field T ∈ H, H being the set of statically
admissible stress fields for NT materials defined in (7):

such that
(T + βI) ∈ Sym− , (42)
for some constant β > 0. In other words the load is supersafe if T is strictly
statically admissible over the set Ω in a uniform way, that is independently
of x. The definition of supersafe loads depends on the choice of the function
space S(Ω). For NENT materials one can assume
S(Ω) ≡ L2 (Ω) . (43)
If the load is supersafe then the potential energy associated to the exter-
nal forces can be expressed in terms of T and of any displacement u ∈ K,
in the form

− s·u− b·u=− T · E(u) .
∂ΩN Ω Ω
Then the total potential energy can be rewritten in terms of this super
safe s.a. stress field T as

1
E(u) = − T · E(u) + C[Ee (u)] · Ee (u) .
Ω 2 Ω
Recalling that E(u) = Ee (u) + Ea (u), on using the safe load assumption
one can write

1
E(u) ≥ − T · Ee (u) + C[Ee (u)] · Ee (u) + β |Ea (u)| ,
Ω 2 Ω Ω
that is the energy has at least a linear growth with respect to the norm of the
space BD(Ω): the space of functions u whose corresponding infinitesimal
deformation E is a bounded measure. For full information on this function
space I refer to the paper by Temam and Strang (1994). Here I notice only
that, since the infinitesimal strain E can be a bounded measure, then u can
be discontinuous and E can be decomposed in its absolutely continuous and
singular parts with respect to the 2d Lebesgue measure:
E(u) = Er + Es .
Recalling the decomposition of E into its elastic and fracture parts,
since the potential energy grows quadratically with respect to the elastic
part of the deformation, then only the fracture part Ea can be singular,
164 M. Angelillo
that is only fracture discontinuities are admitted. In other words the elastic
deformation must be regular and the anelastic deformation can be either
regular or singular. In Giaquinta and Giusti (1985) (theorem 6.8, p 381),
the authors show the existence of the solution for the minimum problem
with T (Ω) = BD(Ω) under the supersafe load condition (and some sup-
plementary technical conditions) in the special case of traction problems
and isotropic elastic behaviour. Since the energy is not strictly convex the
solution is in general non unique.
3.5 Complementary energy functional.

Also a dual energy principle based on the Complementary Energy can
be proved (see Giaquinta and Giusti (1985)). The stress state T◦ that cor-
responds to the solution of the boundary value problem for NENT materials
can be characterized as the minimizer of the energy functional

1
Ec (T) = − Tn · u + A[T] · T , (44)
∂ΩD 2 Ω
over the set H of statically admissible stress fields

S(Ω) being a convenient Banach space. S(Ω) can be assumed as the Hilbert
space L2 (Ω): in other words H is represented by the symmetric second order
tensors T of L2 (Ω), such that T is negative semidefinite and balanced with
s, b. Obviously on considering T ∈ L2 (Ω) the balance conditions must be
considered in a generalized sense (see (8), (9)).
The choice of L2 (Ω) as the function space for the stress field seems natu-
ral considering the quadratic term which represents the stress energy in the
complementary energy (44). Since the Complementary Energy functional is
strictly convex over the convex set H, the existence and the uniqueness of
the minimizer T◦ of such functional is guaranteed whenever H is not void
(that is there exist at least one square summable stress field T such that
T is negative semidefinite and balanced with s, b, or, in other words, the
loads are compatible in the sense of definition (17)). Therefore though the
solution u◦ may be non unique the elastic part Ee of the strain solution is
unique. Non uniqueness is restricted to the anelastic part Ea of the defor-
mation E, and to special arrangements of the boundary conditions. This
circumstance makes the displacement and stress approach to the equilib-
rium of NENT materials non symmetric, in the sense that existence of the
minimizer T◦ for the complementary energy. is not sufficient for the exis-
tence of the minimizer u◦ of the potential energy E. The existence of T◦
requires only the existence of an admissible stress field, existence of u◦ , with

the known theorems, requires instead the existence of a uniformly strictly
admissible stress field. There are indeed counterexamples for the existence
of u◦ in the case the loads do not satisfy the safe load condition, in which
a s.a. stress field T◦ can be found; two of them are reported in the next
Subsection.
3.6 Examples of non-existence.

To my knowledge there are no examples of non existence, for traction
problems, in the case in which the loads do not satisfy the safe load con-
dition but there exists a strictly admissible stress field T. The only known
counterexamples refer to the case in which the loads do not satisfy the safe
load condition and there exists a balanced and admissible, but not strictly
admissible, stress field T. In the known examples there are parts of the do-
main that can be taken away and transplaced rigidly, without paying any
energy price. Therefore E = 0 as |u| → +∞, and I can say that the loads
are collapse loads, in the sense that the deformation can increase indefinitely
at constant load. Some trivial examples of non-existence are considered in
what follows.
Two examples of non existence. The NENT material is a pecu-
liar model for a continuum, some care in the imposition of the data must
be exercised to avoid nasty results, that is lack of solution or very large
displacements for small loads. The idea is that the material is rather del-
icate and loads and displacements must be applied cautiously to allow for
the material to accept them. This aspect is somehow in keeping with the
behaviour of real masonry structures which suffer from the application of
concentrated loads or from the abrupt changes of loads or given displace-
ments. The message is that the data of a BVP relative to a body made of
NENT material must be specially disposed and coordinated.
The two following examples are two, more or less famous, examples of
non existence. The first case (Figure 36a) refers to a traction problem, the
second (Figure 36b) to a mixed BVP.
These two examples suggest that a finite normal crack develops on a
line: the assumption of infinitesimal displacements and strains is violated.
3.7 Elementary analytical solutions

A number of closed form solutions for NENT materials were first pre-
sented in Angelillo and Giliberti (1988); here I report some of them.
166 M. Angelillo
a b
Figure 36. Examples of non existence of the solution for the NENT BVP. (a) : the
displacement corresponding to the only statically admissible state cannot possibly verify
the normality condition at the interface Γ (the vertical center line), unless the normal
displacement jump on Γ tends to +∞. (b): since the only statically admissible stress
field is T = 0, any k.a. displacement field cannot possibly satisfy the normality condition
at the base for the given tangential displacements, unless the normal displacement jump
at the base tends to +∞.
a b
Figure 37. Pure shear: (a). Stress field solution: (b).
Exact solution 1: Pure shear. The very first simple, non trivial solu-
tion (i.e. distinct from elementary solutions of linear elasticity), for NENT
materials is pure shear (Figure 37).
The solution, in terms of displacements, is identical to the universal
solution for homogeneous linearly elastic materials, that is
u1 = γx2 , u2 = 0 .
The corresponding strain decomposition into elastic and anelastic strains:

E = Ee + Ea , for a homogeneous and isotropic NENT material, is

0 γ
− γ4 γ γ γ γ γ
2 = 4 +ν 4 4 + (1 − ν) 4 4 ,
γ
2 0 γ
4 − γ4 γ
4
γ
4
γ
4
γ
4
where
γ γ

{E } = (1 − ν)
a 4
γ
4
γ .
4 4
The associated stress field, trivially balanced with zero body loads inside
Ω is (see Figure 37b)
γ
−4 γ
{T} = E 4 .
γ
4 − γ4
With this solution the whole domain is of the Ω2 type, the material
being uniformly compressed along a family of compression rays parallel to
a diagonal of the square panel; uniformly distributed fractures open up in
the direction of the other diagonal. Therefore, in a real masonry panel, one
may expect a pattern of parallel cracks in the direction of the compression
rays.
Exact solution 2: Simple flexure. The second problem I consider is

the flexure of a rectangular strip. The geometry and the boundary condi-
tions are depicted in Figure 38a to which I refer for notations. A simple
solution of the problem exists if the material is isotropic and one assumes
ν = 0 (an assumption that, though far from being realistic, produces easy
analytical solution that one can use as benchmark problems for numerical
approximations).
In this case the displacement field u = uê1 + vê2 with
ϕ
u= (L − 2x1 )x2 ,
L
ϕ
v = − (L − x1 )x1 ,
L
solves all the field and boundary equations, for a homogeneous and isotropic
NENT material, with ν = 0. Indeed the corresponding total strain is

−2 ϕx 2
0
{E} = L ,
0 0
that is easily decomposed into the elastic and fracture parts:
168 M. Angelillo
a b
Figure 38. Pure flexure of a rectangular strip. Boundary data: (a), stress and domain
partition corresponding to the first solution: (b).
⎧
⎪
⎪ −2 ϕx 2
0
⎪
⎪
L , x2 > 0 ,
⎪
⎪ 0 0
⎨
{Ee } = (46)
⎪
⎪
⎪
⎪ 0 0
⎪
⎪ , x2 ≤ 0 ,
⎩
0 0
⎧
⎪
⎪ 0 0
⎪
⎪ , x2 > 0 ,
⎪
⎪
⎨ 0 0
{Ea } = (47)
⎪
⎪
⎪
⎪ −2 ϕx 2
0
⎪
⎪ L , x2 ≤ 0 .
⎩
0 0
The corresponding stress field is
⎧
⎪
⎪ −2 ϕx 2
0
⎪
⎪
L , x2 > 0 ,
⎪
⎪ 0 0
⎨
{T} = E (48)
⎪
⎪
⎪
⎪ 0 0
⎪
⎪ , x2 ≤ 0 .
⎩
0 0
This stress field is obviously balanced with the prescribed body load
b = 0 at the interior and with the surface tractions s = 0 given at the
loaded part of the boundary (x = ∓H/2). Based on this solution for the
stress, the domain Ω is divided into two zones of the type Ω2 , Ω3 , as depicted
in Figure 38b.
For this example I can also give a different solution, namely the same
stress field but a different fracture field, under the same boundary condi-
tions. The second solution (see Figure 39) to the same problem is defined
by the displacement field u = uê1 + vê2 , as follows
⎧
⎪
⎨ L (L − 2x1 )x2
ϕ
, x2 > 0 ,
u = ϕx2 , x2 < 0 and, x1 < L/2 ,
⎪
⎩
−ϕx2 , x2 < 0 and, x1 > L/2 ,
⎧
⎪
⎨− L (L − x1 )x1
ϕ
, x2 > 0 ,
v = −ϕx1 , x2 < 0 and, x1 < L/2 ,
⎪
⎩
ϕ(x1 − L) , x2 < 0 and, x1 > L/2 .
The corresponding total strain is composed of absolutely continuous and

singular parts: E = Er + Es with
⎧
⎪
⎪ −2 L
ϕx2
0
⎪
⎪ , x2 > 0 ,
⎪
⎪ 0 0
⎨
{Er } =
⎪
⎪
⎪
⎪ 0 0
⎪
⎪ , x2 ≤ 0 ,
⎩
0 0
⎧
⎪
⎪ −2ϕx 2 0
⎪
⎪ δ , x1 = L/2 and x2 ≤ 0 ,
⎪
⎪ 0 0
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎨ 0 0
{Es } = ϕx21 δ , x1 < L/2 and x2 = 0 ,
⎪
⎪ 0
⎪
⎪ L
⎪
⎪
⎪
⎪
⎪
⎪
⎪ −2ϕx2
⎪ 0
⎪
⎩ δ , x1 > L/2 & x2 = 0 .
ϕ(x1 −L)2
0 L
where δ denotes the unit Dirac delta. The total strain is then decomposed
in its elastic and anelastic parts as follows:
Ee = E r , E a = E s .
170 M. Angelillo
λ11 λ22
x
y
y
a b c
Figure 39. Deformed configuration corresponding to the second solution: (a). Graph
of the non zero Cartesian components of Ea :(b). Flexure test on a masonry panel
(courtesy G. Castellano): (c).
a b
Figure 40. Heap under uniform body forces resting on a smooth foundation: (a).
Mechanism for which the load does zero work: (b).
Therefore the stress T coincides with that of the first solution and de-
termines the same partition of the domain Ω, as described in Figure 38b.
The fracture strain is singular and describes the cracks exhibited by the
deformed configuration depicted in Figure 39a. A graph of the E11 and E22
components of Ea is reported in Figure 39b. In the absence of any energy
price to pay to open up fractures, the two solutions reported are perfectly
equivalent and the body can choose any of the two. It could be of some
interest to look at the result of the flexure test performed on a masonry-like
material (a mixture of lime and gypsum with a ratio between tensile and
1
compressive strength of 20 ) shown in Figure 39c.
Exact solution 3: 2d heap of masonry stones on a smooth founda-

tion. The third problem I consider concerns a triangle of NENT material,
simply supported at the base on a rigid, perfectly smooth interface (see Fig-
ure 40). The only external forces I consider are represented by a uniform
gravitational load b = −γê2 directed vertically.
A solution to this problem can be easily found since there exists only
one admissible stress field for this geometry and for these data.
The load is a collapse load since the resultant stress across any vertical
section of the triangle (see Figure 40b) is zero and any two parts divided by
a vertical section, can be separated horizontally with a vertical crack, cor-
responding to a zero energy mode and consisting into a horizontal uniaxial
strain, concentrated on the line of separation.
Setting the length scale in such a way that L = 1, the one and only
s.a. stress field, negative semidefinite and in equilibrium with the load, is
written as follows

0 0
{T} = ,
0 σ(x1 , x2 )
with

γ(x2 − x1 ) , x1 ≤ 1 ,
σ(x1 , x2 ) = .
γ(x2 + x1 − 2) , x1 ≥ 1
To this stress field (considering for simplicity the case ν = 0) is associated
the elastic strain

0 0
{E } =
e
,
0 (x1 , x2 )
with

E (x2 − x1 ) , x1 ≤ 1 ,
γ
(x1 , x2 ) = .
E (x2 + x1 − 2) , x1 ≥ 1
γ
Therefore, taking into account the normality condition T.Ea , the anelas-
tic strain takes the form

λ(x1 , x2 ) 0
{Ea } = ,
0
the total strain being

λ(x1 , x2 ) 0
{E} = .
0 (x1 , x2 )
Such total strain is compatible if and only if
∂ 2 λ(x1 , x2 ) ∂ 2 (x1 , x2 )
+ =0.
∂x22 ∂x21
The function (x1 , x2 ) is not smooth and its second derivatives must be
interpreted in a generalized sense. The 3d graph of (x1 , x2 ) depicted in
172 M. Angelillo
a b
c d
Figure 41. 3d plot of the elastic uniaxial strain (x1 , x2 ): (a). In (b): first derivative
of (x1 , x2 ) with respect to x1 : piecewise constant field. In (c): second derivative; a
constant line Dirac delta with support on the center line Γ. In (d) the anelastic strain
λ(x1 , x2 ) restoring compatibility: a parabolic line Dirac delta with support on the center
line Γ
Figure 41a, can help to visualize the first and second derivatives of (x1 , x2 )
with respect to x1 reported in Figures 41b,c.
A function λ(x1 , x2 ) that solves the compatibility equation and is non
negative is
γ
λ(x1 , x2 ) = (L − x22 ) δ(Γ) ,
E
where δ(Γ) is the line Dirac delta defined on the line Γ of equation x1 = 1
(see the graph depicted in Figure 41d).
The displacement u = uê1 + vê2 is then obtained integrating the total
deformation. In components

2
2E (x2 − 1)
γ
, x1 < 1 ,
u= 2
2E (1 − x2 )
γ
, x1 > 1 ,

γ
2E x2 (x2 − x1 ) , x1 < 1 ,
v= γ
2E x2 (x2 + x1 − 2) , x1 > 1 ,
p
a
3a
2a
2

2
2
1
a b c
Figure 42. Ring under external and internal pressures: (a). Fundamental partition
of the domain corresponding to the solution for α = 25/48: (b). Fundamental partition
corresponding to α = 1/2: (c).
The deformation corresponding to this discontinuous displacement field,

singular on the center line Γ, gives a concentrated fracture varying parabol-
ically with x2 .
Exact solution 4: Ring under internal and external pressures.

The fourth problem I consider concerns a ring of internal radius a and
external radius 2a, composed of NENT material and subjected to an internal
pressure p and an external pressures q = α p (see Figure 42).
The real number α is a load parameter varying in the interval [1/2, 1].
Notice that the case α = 1 corresponds to the elastic solution of uniform
pressure.
First Case: supersafe load. First I consider the case in which 1/2 <
α < 1. To fix the ideas I consider the special case α = 2548 .
In this case the load is supersafe since there exists a strictly s.a. stress
field, uniformly bounded from above by a uniform compression of value
−p/18. Such s.a. stress field, in physical components in the polar coordinate
system {r, ϑ} with origin in the center of the ring, takes the form

a r 2 − a2
Trr = −p + ,
r 36r2
r 2 + a2
Tϑϑ = −p ,
36r2
Trϑ = 0 .
174 M. Angelillo
The existence of a purely compressive s.a. stress field does not imply
that the solution of the BVP is necessarily of pure biaxial compression. In
the case at hand we shall see that both uniaxial and biaxial stress states,
and fractures open up in the inner part of the domain.
With reference to the linear elastic solution for an isotropic pressur-
ized ring, recalling the formula for the stress (see Timoshenko and Goodier
(1951)), one easily finds that: in the region Ω1 = {{r, ϑ} s.t. 32 a < r < 2a},
the stress coincides with the classical elastic solution; in the region Ω2 =
{{r, ϑ} s.t. a < r < 32 a}, the stress is uniaxial and radial (see Figure 42b).
Namely:
pa
− r , a ≤ r ≤ 32 a ,
Trr = 2 2
p(9a +4r )
− 12r2 , 23 a ≤ r ≤ 2a ,

0 , a ≤ r ≤ 32 a ,
Tϑϑ = 2 2
p(4r −9a )
− 12r2 , 32 a ≤ r ≤ 2a ,
Trϑ = 0 .
We leave to the reader to verify that this stress field verifies the bal-
ance equations with zero body forces, matches the given pressures at the
inner and outer boundary, and is compressive. In the region Ω1 the stress
field being coincident with the elastic solution, gives compatible strains and
the physical components{ur , 0} of the displacement u, can be easily found
through the relations:
ur
Eϑϑ = Tϑϑ − νTrr , Eϑϑ = .
r
In Ω2 , ur can be found, modulo a constant, through the equations
∂ur
Err = Trr − νTϑϑ , Err = .
∂r
The constant is finally determined by imposing the continuity condition

u r r − = ur r + ,
at r = 32 a. Then the displacement ur takes the following form
p a

3a
E log 2r + ν , a ≤ r ≤ 32 a ,
ur = p(4r2 (−1+ν)+9a2 (1+ν))
12Er , 32 a ≤ r ≤ 2a ,
e
In Ω2 the elastic deformations Err e
= Trr , Eϑϑ = −νTrr , left alone, are
not compatible. To restore compatibility I must add the fracture field
∂ur
a
Eϑϑ = − Eϑϑ
e
,
∂r
that is

3a
a
pa
Er log 2r , a ≤ r ≤ 32 a ,
Eϑϑ =
0 , 32 a ≤ r ≤ 2a .
Then the fracture field is described by a regular deformation consisting
into a diffuse uniaxial circumferential strain taking place in the internal ring
Ω2 , of outer radius 32 a, and becoming vanishingly small at the boundary
between Ω1 and Ω2 (as shown pictorially in Figure 42b).
Second case: Limit load. Consider now the case in which α = 12 (see
Figure 42c).
In this case the load is limit since there exists at the same time, a
stress field belonging to H, that is a statically admissible (but not strictly
admissible) stress field, and also a non-zero displacement field belonging to
K ◦ for which the load does zero work.
This s.a. stress field, in physical components in the polar coordinate
system {r, ϑ} with origin in the center of the ring, is
a
Trr = −p ,
r
Tϑϑ = 0 ,
Trϑ = 0 .
Notice that, in this case, this is also the unique statically admissible
solution, that is the set H of s.a. stress fields is a singleton (with the
language of Structural Mechanics one may say that the structure, with this
kind of load, is statically determined).
The given loads do zero work for the mechanism described by the fol-
lowing physical components of displacement:
ur = u , uϑ = 0 ,
where u is an arbitrary positive constant.
The solution of the BVP in terms of displacements reads

pa 2a
ur = u + log + ν , uϑ = 0 ,
E r
and the corresponding fracture strain is
176 M. Angelillo

a u pa 2a
Eϑϑ = + log .
r Er r
We point out that, being u an unknown positive parameter, the solution,
in terms of strains and displacements, is not unique.
3.8 Masonry-like panels under flexure, shear and compression:

Mansfield-Fortunato semi-analytical solutions
The problem of equilibrium for normal elastic no-tension materials can
be formulated as a minimum problem for the complementary energy func-
tional over the set of statically admissible stress fields H.
Masonry facades are often formed by an assembly of units in the form of
rectangular panels; an example of a wall with openings in which rectangular
panels can be devised in the masonry, is shown in Figure 43. A special
class of mixed boundary value problem for such units can be solved by
minimizing the complementary energy over the subset H̃ of H defined by the
admissible stress fields that are of rank one. These minimal uniaxial stress
states represent often the exact solution of such BVP’s in some parts of the
domain. This can be verified by deriving the corresponding fracture strains
through the compatibility equations and checking the boundary conditions
for the associated displacements.
These approximate solutions for rectangular panels have some relevance
both for masonry structures as well as for elastic membranes due to simpli-
fied models which impose unilateral constraints on the normal stress, namely
the No-Tension theory (NT-T) for masonry (see Fortunato (2010))and the
Tension Field Theory (TF-T) for membranes (see Mansfield (1969) and
Steigmann (1990) and references therein). Solutions of similar problems,
consisting of uniaxial stress fields directed along a one parameter family of
rays15 , can be found in the technical literature dating back to sixties.
The first to analyze the shearing problem, under the ad hoc assumptions
of TF-T, in mathematically rigorous terms, was Mansfield (Mansfield, 1969).
In Mansfield (1989) he considers more general boundary conditions. For NT
materials, the shearing case was also studied with a variational approach
by Angelillo and Olivito (1995) in the general case of unilateral, anisotropic
elastic materials.
In a recent paper, Fortunato (2010), a comprehensive study, valid for any
relative rigid displacements of the bases (including rotations), is presented;
15
We recall that (see Remark 3, Remark 9 and Figure 6), in the regions of uniaxial stress
and under vanishing body forces, a classical result of TF-T is that the lines of principal
traction (compression for NT-T) form a family of straight lines.
Figure 43. Structural scheme of a masonry wall with openings (in a simplified model
dashed parts can be considered as rigid).
on introducing a curvilinear coordinate system convected with the compres-

sion rays, the problem is reduced to an ordinary differential equation for the
ray distribution (that is the Euler equation of the minimum problem for the
complementary energy over the set H̃), without the need of any ad hoc as-
sumptions besides energy minimization. The basic tension field assumption
(that is the restriction to H̃), in the present context, reduces the compres-
sion ray solution to an approximate solution of the minimum problem for
the complementary energy, and gives a lower bound for the stiffness of the
panel. A definition of the partition of the domain Ω into uniaxial stress
part Ω2 and slack part Ω3 , explicit in terms of the relative displacement
parameters, is given. Simple ordinary differential equations are obtained
from the compatibility condition, from which the anelastic strain (that is
the fracture strain for NT-T and the wrinkling strain for TF-T) can be
computed through integration.
Unilateral models for masonry and wrinkling. The last result has
significant relevance for the problem of wrinkling of thin membranes. When
a thin panel is forced into the post-buckling range by loads applied to its
edges, load is transmitted primarily along one of the principal axes of stress,
while bending effects remain secondary. The panel deforms into a wavy or
wrinkly surface, the crests of the waves coinciding approximately with the
trajectories of the tensile stresses (see Figure 44b).
178 M. Angelillo
a b
Figure 44. Wrinkling of a rectangular sheet under relative shearing of the bases and
free on the lateral sides. Boundary conditions: (a) . In (b) wrinkles developing in an
alluminium sheet subject to the b.c. described in (a).
TF-T defines approximately the magnitude and direction of the stress

in the wrinkled zone. The anelastic strains, in TF-T, are a measure of
wrinkling, in the sense that they may be viewed as limits of sequences of
ordinary deformations whose gradients oscillate finely on any interval. The
definition of the wave length and the amplitude of the wrinkles is of vital
importance in many applications (such as sails or space solar panels). In
recent works, some authors (Epstein (1999), Wong and Pellegrino (2006))
propose approximate formulas for estimating the amplitude as well as the
wave length of wrinkles, based on the knowledge of the TFT stress field and
the corresponding anelastic strain. Then the knowledge of the anelastic
strain can be used to test the adherence of such formulas to the description
of known experimental results on panels.
Basic Boundary Value Problem for a Masonry Panel In this Sec-

tion I summarize the main results of the analysis given by Fortunato in
(Fortunato, 2010). Consider a masonry-like rectangular panel, traction free
on the lateral sides and subject to zero body forces and prescribed rigid
body displacements of the top and bottom bases. A Cartesian frame of
reference {O; x1 = x, x2 = y} (Figure 45a) is introduced, with associated
unit base vectors (e1 , e2 ); let us define {uA , vA , ϕA } as the translation and
rotation parameters of the block RA relative to the pole A◦ , {uB , vB , ϕB }
as the translation and rotation parameters of the block RB relative to the
pole B ◦ , {U, V, Φ} as the relative rigid displacement parameters between
the top and bottom bases.
y y
U B
Φ 0.6
V 0.4
α
0.2
H g 0.0
x x
-0.2
-0.4
V
U Φ -0.6
A -0.4 -0.2 0.0 0.2 0.4
B a b c
Figure 45. Masonry panel undergoing rigid relative displacements of the bases: a.
Slope g of a compression ray: (b). Typical minimizing slope g as a function of ϑ1 = x:
(c).
Modulo an ineffective rigid body displacement, it is straightforward to

relate the displacement components at the top and bottom bases a, b with
the relative (rigid) displacement parameters, written
ua = ua e1 + va e2 = U e1 + (V + Φx)e2 ,
ub = ub e1 + vb e2 = −U e1 − (V + Φx)e2 .
Based on the complementary energy principle stated in Subsection 3.5,
the equilibrium solution can be searched by minimizing the complementary
energy over the set H defined in (45). In keeping with the spirit of TF-T,
an approximate solution is looked for in the restricted set H̃ obtained by
considering stress fields T ∈ H such that T is of rank-one.
Remark 11. As shown by some of the elementary solutions previously
discussed, the displacement solution, in the closure of the regions Ω2 ∪
Ω3 , can exhibit singularities affecting the latent strain. These singularities
correspond to discontinuities in the displacement through lines that can be
interpreted as fracture lines. The normality assumptions (34), (36), imply
that displacement discontinuities be orthogonal to the discontinuity line.
Remark 12. On the interface between Ω2 and Ω3 the stress must be
continuous in order to avoid shear discontinuities (violating the normality
rule). As a consequence the stress along the interface between the regions
Ω2 and Ω3 must be zero, the interface itself a straight unextended line.
The minimizer T◦ of the complementary energy over the restricted set
H̃ is generally not the exact solution of the problem, rather an approximate
solution.
180 M. Angelillo
For any T ∈ H̃ one has Ω = Ω2 ∪ Ω3 , that is Ω1 = ∅. Once the free

boundary between Ω2 and Ω3 is identified, the solution of the equilibrium
problem is reduced to the search of the stress field solution in Ω2 . As already
remarked, in Ω2 one of the two families of principal stress curves (i.e., the
integral curves of the stress eigenvectors) is made of straight lines, that are
called compression rays. The compression rays carry the non zero stress
and do not overlap. Since the lateral sides of the panel are stress free the
compression rays intersect the boundary along the bases. Therefore the
definition of the behaviour of a single panel, under the above assumptions,
is reduced to finding in Ω2 the optimal compression ray distribution, with
the optimal choice being determined by energy convenience.
In what follows, I summarize the results of the analysis given by Fortu-
nato in (Fortunato, 2010), omitting all the proofs and referring to the cited
paper for the detailed derivations. The main results contained in Fortunato
(2010) concern: 1. the explicit definition of the partition Ω = Ω2 ∪ Ω3 , in
terms of the data {U, V, Φ}. 2. The explicit definition of the stress and
complementary energy to be minimized, in terms of the slope of the rays
g. 3. The formalization of the Euler equations and boundary conditions
for all the possible data. 4. The explicit definition of the anelastic strains
(fractures) in terms of the slope g.
Partition of the panel and free boundary between Ω2 and Ω3 Con-

sider the rectangular panel of base B◦ and height H◦ . In the sequel a rect-
angular normalized panel is considered, which is a rectangular panel whose
base is of unit length and whose height is H = H◦ /B◦ . The slope function
g of a compression ray is introduced; the compression ray intersects the
bottom and top bases at the abscissae xA , xB and the horizontal axis at
ϑ1 = xA +x
2
B
, and the slope is defined as a function of ϑ1
[−1/2, 1/2] ! ϑ1 → g(ϑ1 ) := tan(α(ϑ1 )

where α is the angle between such ray and the y axis, as shown in Figure 45b,
to which I refer for notations.
In order that the rays belong entirely to Ω, the geometrical constraints
on g
1 − 2ϑ1 1 − 2ϑ1
− ≤ g(ϑ1 ) ≤ s.t. ϑ1 > 0 , (49)
H H
1 + 2ϑ1 1 + 2ϑ1
− ≤ g(ϑ1 ) ≤ s.t. ϑ1 ≤ 0 , (50)
H H
must be satisfied.
On observing that, for compatibility (that is in order that the strain

satisfies the elasticity relations for shortening and the normality conditions
for lengthening), the interface between the regions Ω2 , Ω3 , if any, must be
an unextended ray, the following equation definining such class of rays is
derived:
V Φ
g(ϑ1 ) = − − ϑ1 . (51)
U U
The free boundary must be chosen in this class.
It is easy to show that all the rays satisfying the constraint (51) pass
through a common centre C whose coordinates {x◦ , y◦ ) depend only on the
parameters {U, V, Φ}. The expressions for the coordinates of the centre C
are
V U
x◦ = − , y◦ = .
Φ Φ
The partition of the panel into the disjoint regions Ω2 and Ω3 , and in
particular their free boundary (necessarily made by rays that satisfy the
constraint (51) ), can be obtained by the position of the center C, and is
independent of the size of the rigid displacement parameters, provided the
ratios U V
Φ and Φ stay constant.
If U > 0 the inequality
V Φ
g(ϑ1 ) < − − ϑ1 . (52)
U U
defines the ray passing through the points P and Q on the top and bot-
tom bases which are shortened for the given rigid displacements parameters
{U, V, Φ} of the bases.
This is a kinematic constraint on the slope g for the existence of the
compression rays, that is for the existence of the region Ω2 .
In the case U = 0, g(ϑ1 ) is not restricted and the rays are shortened, as
long as ϑ1 > − VΦ if Φ > 0 and ϑ1 < − VΦ if Φ < 0. In the case U < 0 then
Φ 1
g(ϑ1 ) > − VΦ − U ϑ .
Summing up, the restrictions on g(ϑ1 ) that have been introduced can be
reformulated as follows:
Given the rigid boundary displacement parameters {U, V, Φ}, find the
pair {ϑ1 , g(ϑ1 )}, so that the geometrical constraints (49), (50) and the kine-
matical constraint (52) hold.
These five inequalities define a feasible region for the pair ({ϑ1 , g(ϑ1 )}
that can be easily visualized with a graph through which the interface be-
tween the Ω2 and Ω3 regions can be located.
182 M. Angelillo
Depending on the values of the triplet {U, V, Φ}, the feasible region for g
can be either empty or not empty. In the latter case the region is a polygon
with three to five sides whose extrema, with respect to the ϑ1 component,
1
define the boundary of the Ω2 region. These extrema are called ϑ1 and ϑ .
Based on the values of U, V, Φ, there are essentially five representative
cases to be considered:
(i) Shearing, shortening, flexure: U = 0 , V > 0, Φ = 0 ;
(ii) Shearing, elongation, flexure: U = 0 , V < 0, Φ = 0 ;
(iii) Shortening: U = 0 , V > 0, Φ = 0 ;
(iv) Shearing: U = 0 , V = 0, Φ = 0 ;
(v) Shortening, flexure: U = 0 , V < 0, Φ = 0 .
Figure 46 shows the admissible repertoire of g(ϑ1 ) in the range [−1/2, 1/2]
for one of these cases.
1min , 1max 4 C
0.6 3
0.4
0.2 2
0.0 1
0.2 1
0.4
0.6
0.40.20.00.20.4 0.51.0
Figure 46. Definition of the taut region for the case of shearing, elongation, flexure.
Restrictions on g and limit slopes: (a). Position of the center C: (b). Taut region: (c).
Curvilinear Coordinates in Ω2 In Ω2 it is convenient to introduce a

system of curvilinear coordinates {ϑ1 , ϑ2 } with one of the curvilinear lines,
say ϑ2 , convected to the compression rays. In terms of the Cartesian coor-
dinates x1 = x, x2 = y, with associated base vectors e1 , e2 , the curvilinear
coordinates {ϑ1 , ϑ2 } are defined as follows
x1 = ϑ1 + gϑ2 , x2 = ϑ2 ,
g being the slope of the rays, a function of ϑ1 alone.

A physical reference system, that is a variable orthonormal base {e1 , e2 },
with e2 tangent to the compression rays at any point of Ω2 , is also intro-
duced.
Strain and Stress in Ω2 Calling uα , the covariant components of the

displacement field u in the curvilinear system {ϑ1 , ϑ2 } , the local infinitesia-
mal strain E has the following covariant components in the same curvilinear
system
g ϑ2
E11 = u1,1 − u1 ,
1 + g ϑ2
E22 = u2,2 ,
1 g
E12 = (u1,2 − u2,1 ) − u1 .
2 1 + g ϑ2
The physical component of strain in the ray direction, that is the strain
component in the direction of e2 ⊗e2 , is related to the covariant component
E22 through the relation
1
E(2 2 ) = E22 .
1 + g2
Notice that the total strain E, in the variable hortonormal base {e1 , e2 },
is described by the matrix

λ(ϑ1 , ϑ2 ) − ν(ϑ1 , ϑ2 ) 0
{E} = .
0 (ϑ1 , ϑ2 )
where ν is the Poisson ratio, = E(2 2 ) is the elastic strain component and
λ is the unknown fracture field.
The decomposition of the total strain into its anelastic and elastic parts
is then

λ(ϑ1 , ϑ2 ) 0
{Ea } = ,
0 0

−ν(x1 , x2 ) 0
{Ee } = .
0 (x1 , x2 )
The uniaxial stress T, in order to satisfy the equilibrium with zero body
forces, must take the form (for a pictorial description of equilibrium along
a compression ray see Remark 3 and Figure 6)
T = σe2 ⊗ e2 ,
with
184 M. Angelillo
f (1 + g 2 )
σ= .
1 + g ϑ2
The elastic part of the deformation, is easily obtained, through the
Hooke’s law, in the form
σ
, =
E
E being the Young modulus of the material.
Energy The complementary energy (44) can be written in the form
1 1 H
ϑ
1 ϑ 2 f 2 (1 + g 2 )2 1 2
Ec = − 2f (U g +V +Φϑ1 )dϑ1 + dϑ dϑ . (53)
ϑ1 2 ϑ1 −H
2
E 1 + g ϑ2
The complementary energy, depending parametrically on {U, V, Φ} is

a functional of f , g, that is of two unknown numeric functions of ϑ1 . The
function f can be determined explicitly by pre-minimizing Ec with rewspect
to f . Such a minimizer is
Eg 2(U g + V + Φϑ1
f= .
(1 + g 2 )2 log(1 − g H2 ) − log(1 − g H2 )
By substituting the previous expression into the stress, one has
1 Eg 2(U g + V + Φϑ1
σ= ,
1 + g ϑ2 1 + g 2 log(1 − g H2 ) − log(1 − g H2 )
and substituting back into the energy, after some algebra, the following
reduced expression of Ec is obtained
1
ϑ
(U g + V + Φϑ1 )2 g 2
Ec = −2E dϑ1 . (54)
1+g H
ϑ1 (1 + g 2 )2 log 2
1−g H
2
This is a functional of the sole unknown function g, to be minimized for

g with the boundary conditions
1
g(ϑ1 ) = g , g(ϑ ) = g .
The solution, that is called g◦ , can be determined by finding the zeroes
of the derivative (the Euler equation) associated to the minimum problem
for (54) (see equation (58), p.101 of Fortunato (2010)). The Euler equation
1min , 1max
g1
0.6 0.10
0.4
0.05
0.2
0.0 1 0.00 1
0.2 0.05
0.4
0.10
0.6 0.2 0.0 0.2 0.4
0.40.2 0.0 0.2 0.4
1min , 1max
g1
0.6 0.0
0.4 0.1
0.2 0.2
0.0 1
0.3
0.2
0.4 0.4
0.6 0.5
0.4 0.2 0.0 0.2 0.4
0.40.2 0.0 0.2 0.4
1min , 1max g1

0.6 0.5
0.4 0.4
0.2 0.3
0.0 1
0.2
0.2
0.4 0.1
0.6 0.0
0.40.2 0.0 0.2 0.4 0.2 0.1 0.0 0.1 0.2 0.3
a b
Figure 47. Numerical solution of three special cases. In column a: fundamental
partition and minimizing slope g. In column b: compression rays corresponding to the
minimizing g and level curves of the principal stress. First row: shortening and flexure.
Second row: pure relative shearing. Third row: shearing, elongation and flexure.
is a second order non-linear differential equation for g that, due to the

precence of log terms in g , results rather stiff. The equation is integrated
by Fortunato in (Fortunato, 2010) for some special cases, by employing a
multi-shooting technique and a Gauss-Kronrod quadrature formula.
Compatibility, Anelastic Deformation The stress field T◦ correspond-

ing to the minimizer g◦ , that is the uniaxial stress field directed as the
compression rays for which the principal stress component is
186 M. Angelillo
0.6
0.4
0.2
0.0
0.2
0.4
0.6
0.4 0.2 0.0 0.2 0.4
Figure 48. Level curves for of the fracture strain for the case of pure relative shearing.
1 Eg◦ 2(U g◦ + V + Φϑ1

σ= 2 2
,
1 + g◦ ϑ 1 + g◦ log(1 − g◦ H2 ) − log(1 − g◦ H2 )
can be either an approximation to the exact solution or the solution of the
BVP. In order to check that T◦ is the exact solution of the BVP, compati-
bility of the strain E, and compatibility of the corresponding displacements
on the part ∂ΩD of the boundary, must be verified. Actually a continuous
displacement field u of which E is the symmetric part of its gradient exists
in Ω2 if and only if
E11/22 + E22/11 − 2E12/12 = 0 , (55)
where / followed by indices denotes covariant differentiation with respect to

those indices.
Recalling that the total strain E admits the representation
E = (λ − ν)e1 ⊗ e1 + e2 ⊗ e2 ,

and that is the elastic stretch determined by the known stress T◦ , the
unknown fracture strain field λ can be found by solving the differential
equation (55).
In Fortunato (2010) the author finds the explicit solution of this differ-
ential equation in terms of g◦ . The solution is
z(1 + g◦2 )
λ= + ν + g◦2
1 + g◦ ϑ2
where z is the solution of the differential equation

p
z,22 =
1 + g ◦ ϑ2
p being a specified function of f , g, that is of g◦ (see (73), p.103 Fortunato
(2010)).
Rectangular panels under elongation, flexure and shearing: ex-

amples In this section the solution of some peculiar cases is presented. A
rectangular panel of normalized lengths and Poisson ratio ν = 0, subject to
three different combinations of the given displacements {U, V, Φ} is consid-
ered. The Euler equation is solved numerically for each of the three cases
with a multi-shooting technique. In Figure 47 the fundamental partition,
the minimizer g(ϑ1 ), the compression rays and the principal stress levels
are shown. In particular for the example of pure shearing the contour plot
of the fracture λ, that is the only non vanishing component of the anelastic
strain in Ω2 , is shown in Figure 48. Since the solution of the Euler equation
is obtained numerically the optimal g, and the corresponding σ, are affected
by numerical errors depending essentially on the integration scheme chosen.
In the examples, an adaptive algorithm has been adopted: the function g is
1
integrated over the interval [ϑ1 , ϑ ], by using a Gauss-Kronrod quadrature
formula, from which an integral value (usually overestimated) and an error
estimate are obtained. If the estimation is too big, the interval is divided
in half and the integration is performed over each of the halves using the
quadrature formula. If the total error is still too big, the interval with the
biggest error is again bisected. The process is repeated until the desired pre-
cision is reached. For the integration scheme adopted the anelastic strain
is always non negative all over Ω2 . This outcome gives an indication that
the compression ray solution is the exact solution for the cases considered
(ν = 0).
4 Model two (ML)

In this section the main ingredients of the theory concerning model two for
masonry materials, namely the Masonry-Like (ML) material, that is the
n > 1 version of the 1d model introduced in Chapter 1, are presented.
The constitutive assumptions, the balance equations and the boundary and
initial conditions are given and the evolutionary boundary value problem
for ML materials is formulated. A number of illustrative exact solutions
and examples are discussed and a new technique for handling numerically
the solution of specific ML problems, is introduced.
188 M. Angelillo
4.1 The equilibrium problem for ML materials

The crushing behaviour of masonry, that is its limited strength in com-
pression, is modeled as perfectly plastic, then the crushing behavior of the
material is assumed to be represented within the classical framework of a
convex elastic domain coupled with the normality law, the yield surface
being fixed in the stress space (no hardening or softening).
The initial-boundary value problem describing the quasi-static evolution
of a NT elasto-plastic body occupying a bounded domain Ω with boundary
∂Ω is then considered.
The first to propose and analyse this model for masonry materials were
Lucchesi and Zani (1996).
Plastic behaviour is described in terms of strain rates and the problem
is not merely a BVP but rather an evolutionary problem. The evolution is
assumed to be quasi-static, that is, to occur so slowly that inertial effects
may be ignored.
The initial-boundary value problem in two dimensions. Again I

assume small strains and restrict to 2d problems.
Time dependent data are considered, such as the fields b(x, t) (body
forces per unit volume), s(x, t) (surface tractions per unit area), and u(x, t)
(surface displacements):
b : (x, t) ∈ Ω × [0, t) → b(x, t) ∈ V 2 ,
s : (x, t) ∈ ∂ΩN × [0, t) → s(x, t) ∈ V 2 ,
u : (x, t) ∈ ∂ΩD × [0, t) → u(x, t) ∈ V 2 ,

with T = [0, t) the time interval in which the evolution is considered and t
the final instant of the evolution. Usually it is assumed that
b(x, 0) = 0 , s(x, 0) = 0 , u(x, 0) = 0 .

We consider that the body Ω is composed of NT material, that is the
stress T is negative semidefinite (T ∈ Sym− ).
We further assume that the stress cannot be arbitrarily large but is
confined to belong to a bounded convex set K of Sym containing the origin.
Then the interior of K ∩Sym− is the elastic region:
T ∈ K ∩ Sym− , (56)
T2
Ω’2 Ω2 λ Ω3
T1
Ω1
Ω2
K∩NSym
Ω’1
NSym Ω’2
Figure 49. Section of the Elastic Domain in the stress space Sym with the plane
T12 = 0.
while its boundary is the yield surface. Notice that K ∩Sym− is convex but
need not to be smooth. The boundary of K may be represented by a level
set of a function f , called the crushing function, so that
K = {T ∈ Sym : f (T) ≤ 0} .
For simplicity I consider
σ◦2
f (T) = Φ̃(T) − , (57)
2E
where Φ̃ is the strain energy density of the isotropic NENT material, ex-
pressed as a function of the stress T, and σ◦ is the crushing stress in uni-axial
compression; that is the boundary of K is a level set of the free energy of
the NENT material. In Figure 49 the intersection of the elastic domain with
the plane T12 = 0, is depicted.
Under the small strain hypothesis, the total deformation is again de-
scribed by the tensor E, and an additive decomposition of the total strain
can be considered in the form
E = E∗ + Ep , (58)
where Ep is the plastic strain. Notice that now E∗ represents the reversible
part of the deformation, in turn composed itself of two parts:
190 M. Angelillo
E ∗ = Ee + E a , (59)
the elastic part being linearly related to the stress T:
Ee = A [T] . (60)
For the latent anelastic part (a measure for fracture) of the reversible
strain, again normality to the cone Sym− is assumed
(T − T) · Ea ≤ 0 , ∀T ∈ Sym− . (61)

p
For the time rate Ė of the plastic strain the associative flow rule
p
(T − T) · Ė ≤ 0 , ∀T ∈ K , (62)
is considered.
Notice that the reversible part of the strain E∗ can still be derived by
an energy density:
∂Φ
T= .
∂E∗
In the isotropic case the form of the energy density Φ can be constructed
explicitly in terms of the eigenvalues e∗1 , e∗2 of E∗ (here e∗1 < e∗2 is assumed).
In the case of generalized plane stress this form is
⎧
⎪
⎨0 , e∗1 ≥ 0 and e∗2 ≥ 0 ,
Φ = 12 E(e∗1 )2 , e∗ < 0 and e∗2 ≥ −νe∗1 , (63)
⎪
⎩1 E
∗ 2 ∗ 2 ∗ ∗
1∗
2 1−ν 2 (e1 ) + (e2 ) + 2νe1 e2 e1 < 0 and e∗2 < −νe∗1 ,
E, ν being the Young modulus and the Poisson ratio. Notice that the stress
T derived from Φ satisfies identically the unilateral restriction (30), that is
there is no need to impose it as a constraint.
Energy formulation, internal variables. The problem is described in

terms of the total strain and of the recorded history of mechanical be-
haviour, by introducing as an internal variable the total plastic strain Ep .
If I consider that the form of Φ(E − Ep ) is prescribed and that T = ∂E ∂Φ
∗,
the instantaneous values of T are known if E is given and the entire process
t p p
of plastic strain is known. Obviously Ep = 0 Ė dt, and Ė is described by
the flow rule (62).
Notice that, based on the ML model, fracture strains are reversible and
are perfectly recoiled upon load inversion. Crushing strains, by contrast,
cannot be healed and, being totally irreversible, can either stay or grow.
In other words, smeared fractures cannot cancel crushing strains; the two
mechanisms being completely independent.
4.2 Numerical minimization strategy

The numerical method I adopt to solve approximately the BVP for
NENT materials is to search for the minimum of the potential energy (26)
and is based on the direct minimization of such a functional through a de-
scent method. The search is carried out in the subset of the set of statically
admissible displacements K defined by the C ◦ displacement fields obtained
by employing a standard finite element approximation based on a triangular
finite element discretization Πh (Ω) of the domain Ω, where h denotes the
mesh size. This kind of discretization excludes discontinuities in u, that is,
real cracks. The reason for considering such a simplification is twofold:
Firstly, I believe that fractures in NENT materials will appear smeared
within the domain if the loads are safe (that is they are not collapse loads)
in the sense specified in Subsection 2.9.
Secondly, in limit cases in which the loads approach the collapse limit the
fracture strain may accumulate in narrow bands indicating the occurrence
of real cracks in the limit.
Descent methods for NENT materials. The convenience of descent

methods, favoured in recent years by the widespread availability of compu-
tational power, is recovered in the case of unilateral and non-smooth energy
shapes. In the specific case of NENT materials the method is particularly
indicated since the problem becomes unconstrained, the potential energy is
a convex function of its arguments and the method of descent is insensitive
to zero-energy modes (source of major troubles with standard FEM based
on updated stiffness matrices, see Alfano et al. (2000)).
Several numerical tests performed on simple problems, for which the ex-
act solutions are known, show the competitiveness of the descent approach
with respect to more classical techniques. Some of these benchmark prob-
lems are reported in what follows, a larger number of examples can be found
in the recent paper by Angelillo et al. (2010). Comparisons with numerical
solutions obtained by other developing codes (see Lucchesi and Zani (2008))
and commercial programs (Abaqus, 6.12) indicate that the descent method
192 M. Angelillo
seems to be the right choice to overcome the difficulties which are inherent
to the NT constraint.
) h }) of
The energy functional E is approximated with the function E({u
the nodal displacements {uh }:

) h }) = −
E({u Lr s(xr ) · ur − fm · u m − An b(xn ) · un − Aq Φ(xq )
r m n q
(64)
where ur is the displacement at the midpoint xr of the r-th edge of length Lr
on ∂ΩN , um the displacement of the m-th mesh node where the concentrated
force fm is applied, un the displacement at the Gauss point xn of the n-th
mesh triangle with area An , and xq the Gauss point of the q-th mesh triangle
of area Aq , where the strain energy density Φ is evaluated for integration. In
the discretized version (64) of the potential energy, all the displacements uj ,
as well as the strain energy density Φ(xq ) of the q-th triangle, are clearly
explicit functions of the nodal displacements {uh }, via the linear shape
functions of a standard triangular mesh.
The iterative procedure adopted to minimize the function (64) is based
on a step-by-step minimization method. Let us denote {uh }j the nodal
displacements at the j-th minimization step. The force acting on the mesh
nodes is given by the negative gradient of the energy fj = −∇j E. ) The
descent method implemented computes the current velocity pj employing
the nodal forces at the current and previous step as
pj = ηj pj−1 + fj ,
where the scalar ηj is

fj · (fj − fj−1 )
ηj = M ax ,0 ,
fj−1 · fj−1
in the Polak-Ribiere version of the conjugate gradient method, and
fj · fj
ηj = ,
fj−1 · fj−1
if the Fletcher-Reeves variant of the method is employed (the Polak-Ribiere
method is usually adopted in the applications reported herein).
If the nodes of the mesh are constrained, the velocity pj is projected
onto the tangent space of the constraint equations to obtain the compat-
ible velocity p∗j . The velocity p∗j gives the direction for the minimization
motion while obeying all the constraints imposed on the nodes. The nodal
displacement {uh }j is computed as
{uh }j = {uh }j + κj p∗j ,
where κj is the amplitude of the minimization step in the direction of p∗j and
is computed via a line search method16 to minimize the energy E({u ) h }) in
∗
the direction of the velocity pj . The iteration process stops when a suitable
norm of the energy gradient becomes sufficiently small (for the decrease
conditions see for example (Kelley, 1999)).
Descent methods for ML (dissipative) materials. When consider-

ing ML materials, that is, adding to the NENT model a crushing strength
criterion (of elasto-plastic associative type), it seems that the proposed nu-
merical technique, based on energy minimization (extremely efficient for
unilateral materials) should be abandoned. The elastoplastic behavior is
indeed inherently path-dependent: the stress state at time t depends, in
general, on the whole strain history in the interval T = [0, t) rather than on
the strain at time t. Then the equilibrium problem for such a material is es-
sentially a dissipative evolution problem whose solution cannot be obtained
by simply minimizing an energy functional. The proposed technique can
still be applied to this evolutive problem considering the exact trajectory as
the limit of a sequence of minimum problems. This is done by discretizing
the time interval into steps and updating the energy in a suitable way. The
evolutive problem is then approximated as a sequence of a discrete number
of minimizing movements. The evolutive solution is obtained as the limit
of the discrete evolution by letting the time step go to zero (see De Giorgi
(1996) for the general formulation, Mielke and Ortiz (2008) for the conver-
gence proofs in the general case of rate independent materials and Dal Maso
et al. (2004) in the specific case of perfect elastoplasticity).
The time interval T = [0; t) is discretized into k subintervals by means of
the instants 0 = t◦ ≤ t1 ≤ .. ≤ ti ≤ .. ≤ tk = t. The idea is approximate the
exact trajectory (u(t), T(t), Ea (t), Ep (t)) (which solves the initial-boundary
value problem defined above), with a sequence of states obtained by solving,
at each time step ti , the minimum problem for a suitably defined, updated
energy functional, characteristic of an evolving hyper-elastic NT material.
To model perfect plasticity in compression, I assume a linearly-growing
extension of the strain energy function defined in (63), beyond the yield
surface defined in (57) (see Figure 50).
16
The line search method calculates the energy for several values of the scale factor κj
(doubling or halving each time) until the minimum energy is passed. The optimum
scale is then calculated by quadratic interpolation.
194 M. Angelillo
φ~i
s 0 ( e)
i 2
( ie)10
Figure 50. Linear prolongation of the strain energy beyond the crushing limit.
Namely, at any time step ti , the functional form implemented in the

finite element code is
⎧
⎪
⎪ 0, (e∗1 )i ≥ 0 , (e∗2 )i ≥ 0 ,
⎪
⎪
⎪
⎪ 1
E(e∗ )2 , (e∗1 )i < 0 , (e∗2 )i ≥ −ν(e∗1 )i , f (E∗i ) ≤ 0 ,
⎪
⎨2* 1 i
) i (E∗i ) = α 12 E(e∗1 )2i + β ,
Φ (e∗1 )i < 0 , (e∗2 )i ≥ −ν(e∗1 )i , f (E∗i ) > 0 ,
⎪
⎪
⎪
⎪ ϕ((e∗j )i ) (e∗1 )i < 0 , (e∗2 )i < −ν(e∗1 )i , f (E∗i ) ≤ 0 ,
⎪
⎪ *
⎪
⎩α ϕ((e∗ )i ) + β
j (e∗1 )i < 0 , (e∗2 )i < −ν(e∗1 )i , f (E∗i ) > 0 ,
where
1 E
∗ 2
ϕ((e∗j )i ) = 2
(e1 )i + (e∗2 )2i + 2ν(e∗1 )i (e∗2 )i ,
21−ν
(e∗1 )i , (e∗2 )i ((e∗1 )i < (e∗2 )i ) are the principal values of E∗i , and the elastic
strain at time ti is given by the difference between the total strain Ei at
the same time step and the plastic strain Epi−1 inherited from the previous
solution step, that is E∗i = Ei − Epi−1 ; by using this relation Φ ) i (E∗ ) can be
i
expressed as a function of Ei and becomes Φ & i (Ei ). The constants
1
2 σ2
α= σ◦ , β = − ◦
E 2E
are introduced to preserve the C 1 regularity of Φ ) i (E∗ ). A representation of
i
) ∗
Φi (Ei ) in the space of principal elastic strains is depicted in Figure 50.
The descent procedure finds the minimum of the total potential energy
at time ti , defined as

Ei (ui ) = − s · ui − b · ui + & i (ui )) ,
Φ(E (65)
∂ΩN Ω Ω
via a finite element discretization of the domain and descent minimization.
The solution at the previous loading step is used as the initial condition for
the minimization of the function

& h }) = −
E({u Lr s(xr ) · ur − fm · u m − An b(xn ) · un − & q)
Aq Φ(x
r m n q
(66)
representing the finite element approximation of the total potential energy
(65), as in the case of NENT materials (see (64)). At each step i, the
& h }) is performed via the descent method
minimization of the function E({u
described previously for NENT materials.
A plastic strain update is then performed at each Gauss point. The yield
condition f (E∗i ) = 0 defines, in the space of principal reversible strains
E∗ , a curve whose position vector is y, of coordinates {yj }. It is useful
to give a parametric description y(γ) of the yielding curve in the space
of principal elastic strain, γ being the parameter. The return mapping
algorithm, according to the principle of minimum dissipation imposed by
the assumption of associated plasticity (Ortiz and Simo, 1986), consists in
finding the value of the parameter giving the minimum distance (in the
energy norm) of the current elastic strain E∗i from the curve y(γ):
minγ Φ(y(γ) − E∗i ) .

This minimum problem can be easily formulated as
d
Φ(y(γ) − E∗i ) = 0 ,
dγ
that can be solved for γ, at each Gauss point, via the Newton–Raphson
p
method. The tensor of plastic strain rate Ė at time ti is coaxial to the
∗
reversible strain tensor Ei , therefore the principal components (Δepj )i (with
p
j=1,2) of the plastic strain increment ΔEpi (i.e., the discrete version of (Ė )i )
are simply computed as
(Δepj )i = (e∗j )i − yj (γi◦ ) .

Once the plastic strain has been updated in the global reference frame,
i.e. Epi = Epi−1 + ΔEpi (a backward Euler finite difference scheme), the
196 M. Angelillo
energy density dissipation at the given Gauss point, at time step ti , can be
computed as
& i , Ep ) − Φ(E
Di = Φ(E & i , Ep ) − 1 Ti · A[Ti ]
& i , Ep ) = Φ(E
i−1 i i−1
2
from which one sees that the plastic strain increments produce energy loss.
4.3 Numerical examples

The ability of descent methods to approximate the solution of boundary
value problems for NENT and ML materials is tested in the paper (Angelillo
et al., 2010) in two ways: the numerical solutions are compared first with
some simple exact solutions, then with some experimental results; finally
the numerical solutions obtained with our code for more complex boundary
value problems concerning masonry facades are presented. Here I report
some of those results.
Example 1: Simple Flexure. In Angelillo et al. (2010), we solved
numerically the problem of Flexure (whose exact solution is described in
Subsection 3.7) both for the isotropic NENT material and for the isotropic
ML material, by putting ν = 0, H = B = 2 m.
For the first case (NENT material), we assumed for the Young modulus
E = 660 MPa; the value of the rotation was set to Φ = 0.001 radiants. To
test convergence a sequence of structured discretizations of decreasing mesh
size h was considered. By introducing the normalized
√ mesh size h/D, D be-
ing the diameter of the domain, the four values 2{1/16, 1/32, 1/64, 1/128}
were considered. In Figure 51a, the contour plot of the maximum principal
fracture strain obtained with the finest mesh, is reported. Fractures are
non-zero in the region indicated as Ω3 in Figure 38b, and their distribution
suggests that the numerical solution is close to the rigid-block displacement
represented in Figure 51b, which can be thought of as an energetically-
equivalent alternative to the two analytical solutions described in Subsec-
tion 3.7 (see also Figure 39).
The numerical experiment was repeated for a ML panel by assuming for
the maximum compressive stress in uniaxial compression, the value σ◦ =
19.8 MPa. The value of the given rotation Φ = 0.006 radiants is selected in
such a way that, for the exact solution, the strip y > B/4 is forced into the
yielding regime. For this value of Φ the exact stress solution is
⎧
⎪
⎨−σ◦ e1 ⊗ e1 , y ≥ B/4 ,
T◦ = −4σ◦ B y
e1 ⊗ e1 , B/4 > y ≥ 0 ,
⎪
⎩
0, 0>y,
x
2
10.0
1.0 2.04.0
0.5
1.5 0.1
x 1 0.1
0.5
y
0
1 0.5 0 −0.5 −1
y a b
Figure 51. Numerical solution of the flexure problem for an isotropic NENT material.
In (a): maximum fracture strain [×10−3 . ]. Deformation on the right and rigid-block
displacement on the right, close to the solution obtained numerically through descent.
Again the same sequence of discretizations of decreasing mesh size is

considered. Figure 52a shows the computed isostatic curves for the finest
mesh. They are very close to being vertical lines as expected. In Figure 52b
the contour plot of the computed maximum compressive stress (solid lines)
is comapared with the exact solution (dash-dotted lines). The numerical
solution shows, graphically, good qualitative accuracy. The maximum frac-
ture strain is reported in Figure 52c, and it can be noticed that the fracture
distribution in the strip y < 0 (region Ω3 in Figure 38b) is similar to the one
observed in the NENT material (see Figure 51a). The distribution of frac-
ture strain in the Ω2 region resembles closely the crushing vertical fractures
that appear in the experiment shown in Figure 52d.
For this example,both for the NENT and ML materials, we performed
a numerical convergence study on the stress, by considering the sequence
of discretizations of decreasing mesh size h described above. In Figure 53 a
plot of the approximation error for the stress
Th − Tcirc L2

eh = ,
Th L2
versus the dimensionless mesh size h/D (T0 , Th , being the exact stress
solution snd the stress field computed for the mesh of size h) is reported. A
linear convergence of the method is obtained in both cases.
Example 2: Pure relative shearing. The simple rectangular panel
depicted in Figure 54 subject to a given relative horizontal translation of the
bases (of value 2U ) and vanishing tractions on the lateral sides is considered.
198 M. Angelillo
2 2
−18
1.5 1.5
−14
x 1 x 1 −10
−6
0.5 0.5
−2
0 0
1 0.5 0 −0.5 −1 1 0.5 0 −0.5 −1
y a y b
2
2 4
1
0.5
1.5 0.5
1
x 1 0.1
0.5
0.1
0
1 0.5 0 −0.5 −1
y c d
Figure 52. Numerical solution of the flexure problem for an isotropic ML material.
In (a), (b): isostatic lines of compression and contour plot of the minimum principal
stress (the other principal stress is almost zero). In (c): contour plot of the maximum
anelastic strain component (fracture strain). Flexure test on a masonry panel (courtesy
G. Castellano): (d).
0.14 0.14
0.12 0.12
0.1 0.1
0.08 0.08
eh
eh
0.06 0.06
0.04 0.04
0.02 0.02
0 0 0 0 0.02 0.04 0.06 0.08 0.1
0.02 0.04 0.06 0.08 0.1
h/D h/D
a b
Figure 53. Flexure: convergence diagram. In (a): NENT material; in (b): ML
material.
U −0.5
2
−1.0
1.5 −2.0
x 1 −3.0
0.5
−4.0
0 −5.0
U −8.0
1 0.5 0 −0.5 −1
y a b
Figure 54. Pure relative shearing of a NENT panel. In (a): boundary conditions and
mesh used in the numerical simulation. In (b): Comparison of the minimum principal
compressive stress levels corresponding to the numerical (solid lines) and exact solution
(dotted lines).
NENT material. The material parameters and the geometry are the
same adopted in the previous Section. The value U = 1mm is considered.
In Figures 54 and 55 the numerical solution is compared with the semi-
analytical solution described in Subsection 3.8.
The distribution of the maximum compressive stress, the fracture strain,
and the form of the isostatic lines computed numerically for the NENT
panel are in good agreement with the results of the semi-analytical method
of Fortunato (2010), as summarized in Figures 54 and 55.
ML material. The numerical experiment was repeated for a ML panel, by
assuming the same material parameters adopted for the NENT material and
putting for the limit stress: σ◦ = 19.8 MPa. The value of the displacement
at the boundary, U = 1 mm, previously considered, is small enough to give
very limited yielding (mainly located in the vicinity of the corners, and is
used as the first step in the discretized loading of the ML panel, whose
evolution and crushing spreading is followed approximately, as the relative
shearing U is gradually increased, by discretizing the real trajectory into
steps.
In Figure 56 three stages of the evolving solution are reported. As the
boundary displacement increases, a diagonal band, uniaxially and uniformly
compressed at the limit stress of 19.8 MPa, forms progressively, and the
isostatic lines of compression (compression rays) become more and more
parallel.
The contour plots in Figure 57 represent the maximum fracture strain,
which concentrates on two sub-diagonal lines, and the maximum plastic
200 M. Angelillo
0.5
1.0
2.0
3.0
a1 a2
b1 b2
Figure 55. Pure relative shearing of a NENT panel. Fracture strain for the numerical:
(a) and for the exact: (b) solutions. Compression rays for the numerical: (c) and the
exact: (d) solutions.
strain, which concentrates at the constrained boundary, near two opposite

corners. Concentration of plastic strain is expected, due to the assumed
perfect plasticity of the material.
In Figure 58, a plot of the energy as a function of the step-wise increment
of the boundary displacement U is reported, where: black circles denote the
energy level of the computed solution at the beginning of the time step
i, open circles denote the energy level due to the loading increment, and
diamond marks denote the energy level reached after convergence.
Therefore the upward jump in energy represents the effect of updating
the given disploacement from the value at time i to the one at time i + 1;
the jump back of the energy is the effect of the numerical minimization of
the updated energy, and the last smaller drop is the effect of the numerical
2 2
−1
−3
1.5 −5 1.5
−7
−9
x 1 x 1
0.5 −11 0.5
0 −14 0
1 0.5 0 −0.5 −1 1 0.5 0 −0.5 −1
y a1 y b1
2 2
−10
−7
−13 −4
1.5 −1 1.5
−16
x 1 x 1
−19
0.5 0.5
0 0
1 0.5 0 −0.5 −1 1 0.5 0 −0.5 −1
y a2 y b2
2 2
−19
−16
1.5 −13 1.5
−10
−7
x 1 −4 x 1
−1
0.5 0.5
0 0
1 0.5 0 −0.5 −1 1 0.5 0 −0.5 −1
y a3 y b3
Figure 56. Pure shearing of a ML panel. Evolution of the contour plot of the minimum
principal compressive stress and of the compression rays at three steps of the loading
process.
202 M. Angelillo
2 2
−2.8
5 −5.6
1.5 1.5 −8.4
10
−14.0 −11.2
15
x 1 20 x 1
25
0.5 0.5
0 0
1 0.5 0 −0.5 −1 1 0.5 0 −0.5 −1
y a y b
Figure 57. Pure shearing of a ML panel. Contour plot of the maximum principal
fracture strain: (a), and minimum plastic strain: (b), for the final value of the relative
displacement.
5 5
horizontal force at the base [105N]
4
total potential energy [kJ]
3 3
2 2
1 1
0 0
0 5 10 15 0 5 10 15
U [mm] U [mm]
Figure 58. Pure shearing of a ML panel. Energetics.
update of the plastic strain that gives the approximate solution at time i+1.
This energy loss represents energy dissipation due to plastic work, and
gives evidence of spreading of plastic deformation as the load is increased.
The dashed line (that is the envelope of the solution points) represents the
numerical approximation to the exact time history of the total potential
energy of the body. In Figure 58b the push-over plot, that is the evolution
of the horizontal component of the computed reaction at the base (horizontal
force) as the displacement U increases, is depicted. The shear force plateaus,
as expected for a perfectly plastic structure approaching collapse.
Validation against experimental tests. Here the ML material model
is validated against independent sets of experimental results from Benedetti
U N0
Figure 59. Loading scheme in the experiments by Benedetti and Steli (2008) and
Eucentre (2010)
and Steli (2008) and Eucentre (2010), performed on different types of ma-
sonry. Figure 59 reports the constraints and the load scheme used in the
numerical analysis to simulate the experimental setup.
A masonry panel of width B, height H, and thickness D is fixed to the
ground at the bottom and to a steel beam at the top (the gray strip in Fig-
ure 59). A uniform load is distributed at the top part of the steel beam and
the horizontal load is applied in incremental steps by imposing the horizon-
tal displacement U of the left edge of the steel beam. Neither Benedetti and
Steli (2008) and Eucentre (2010) report measurements of the Poisson’s ratio,
and because of that ν = 0 is assumed in the simulations, since, paramet-
ric studies, not reported here, show that the simulated force-displacement
curves, under shear, manifest very low sensitivity to the Poisson’s ratio. The
graphs in the top row of Figure 60 compare the numerical simulations and
the experimental results for specimens (1A-08,2C-03) of the experiments
by Benedetti and Steli (2008). In these tests, the masonry is composed of
crushed stones and injected crushed stones. Our model reproduces quanti-
tatively the substantial features of the measured force-displacement curves,
with a slight overestimation of the force for higher levels of the horizontal
displacement U .
The graphs of Figure 60c,d report the comparison between experimental
data and numerical simulations for the specimens CS00, CS02, and CT01
of (Eucentre, 2010), all composed of stone masonry. Specimen CS00 differs
from the other two because the mortar has been reinforced with 20% sand
in mass fraction. This might explain the flatness of the force-displacement
curve for experiment CS00 which is very well captured by the model. The
204 M. Angelillo
1A−08 2C−03
250 300
200 250
horizontal force [kN]

200
150
150
100
100
50 50
experiment
experiment experiment
simulation
simulation simulation
0 0
0 1 2 3 4 5 6 0 2 4 6 8 10
U [mm] a U [mm] b
CS00 − CS02 CT01
50 250

40 200
30 150
20 100
10 experiment CS02 50 experiment

experiment CS00 simulation
simulation
0 0
0 5 10 15 20 0 2 4 6 8 10 12
U [mm] c U [mm] d
Figure 60. Comparison of numerical simulations with the experimental results of

Benedetti and Steli (2008) (1A-08,2C-03) and Eucentre (2010) (CS00-CS02, CT01)
marked reduction of the horizontal force at higher levels of the displacement

U might be due to softening effects induced in the masonry by the unrein-
forced mortar, an effect that is not captured by the ML model as presented
here.
Example 3: Masonry walls with regular openings. In this section
I apply the ML material model to simulate the response of a simple two-
storey facade to vertical loads, seismic loads, and differential foundation
subsiding. The geometry of the facade, the applied loads, and the boundary
conditions are summarized in Figure 61a.
The facade is provided, above the openings, with 25 cm-thick wood
beams and is assumed to be made of tuff and mortar of good quality. Wood
is modeled as elastic, with Young’s modulus E = 11 GPa, Poisson’s ra-
tio ν = 0.35, and density ρ = 800 kg/m3 . For the tuff wall we assumed
a Young’s modulus E = 0.66 GPa, Poisson’s ratio ν = 0.2, density 1800
kg/m3 , and compression limit σ ◦ = 1.98 MPa. The whole structure is
assumed to be 0.5 m thick.
Working loads. Working loads are represented by the weight of masonry
25kN/m
1m Q
39kN/m
0.25m
3m
25kN/m
seismic load
1m
1800kg/m3
weight
3m
subsiding
6 cm
1.5m 1.5m 2.5m 1.5m 1.5m

a b
c d
Figure 61. Masonry wall with openings. Loading schemes: (a). Numerical results for
vertical loads: b,c,d. Contour plot of maximum principal fracture strain: (b). Minimum
principal stress: (c). Isostatic lines: (d)
and the force transmitted by the floors (25 kN/m); the results of the sim-
ulation are reported in Figure 61b,c,d. The structure sustains the working
loads without detectable crushing. The value of the maximum stress at the
base of the wall is about 0.35 MPa. The partition of the domain can be
inferred from Figure 61d: in Ω1 both families of isostatic lines are depicted;
in Omega2 the family of isostatic lines corresponding to zero stress are not
reported.
Horizontal loads. We simulated the response of the facade to a uniform
horizontal force per unit length of 39 kN/m, distributed on the right side of
the structure and superimposed to the structure subject to working loads
(see Figure 61a). The total horizontal load is equivalent to 70% of the
weight of the structure and is applied in ten steps. This kind of loading
can be adopted to simulate seismic loads if horizontal ties or connections
are present. Crushing strain accumulates in very localized regions near the
206 M. Angelillo
a b
c d
Figure 62. Numerical results for vertical and horizontal loads. Contour plot of max-
imum principal plastic strain: (a). Contour plot of maximum principal fracture strain:
(b). Minimum principal stress: (c). Isostatic lines: (d)
corners (see Figure 62a). The formation of a compressed diagonal truss

element, in the panels in between the openings, is evident from Figures 62c
and 62d.
The force-displacement curve, depicted in Figure 64, shows that the
structure is approaching collapse for the maximum horizontal load applied.
Differential foundation subsiding. A 6 cm subsiding of the base of the
central wall is imposed in fifteen steps.This action is superimposed to the
effect of working loads. The structure shows a peculiar kinematical effect,
consisting in a vertical displacement of the central wall and an outward
rotation of the lateral walls around the corner points of the bases, where
the plastic strain concentrates (see Figure 63).
The computed vertical component of the reaction of the central wall
drops from about 390 kN to 240 kN after the subsiding. This computa-
tion indicates that the vertical loads migrate from the central panel to the
lateral ones. Such a stress redistribution is also highlighted from the com-
a b
c d
Figure 63. Masonry walls with openings under the effect of vertical loads and a vertical
settlement of the central pier: mumerical results. Contour plot of maximum principal
plastic strain: (a). Contour plot of maximum principal fracture strain: (b). Minimum
principal stress: (c). Isostatic lines: (d)
parison of the isostatic lines depicted in Figure 63d with the ones reported
in Figure 61d.
Bibliography
Dassault Systemes Abaqus, ver. 6.12 http://www.3ds.com/fileadmin/
PRODUCTS/SIMULIA
G. Alfano, L. Rosati and N. Valoroso. A numerical strategy for finite element
analysis of no-tension materials. Int. J. Numer. Methods Eng., 48 (3):
317–350, 2000.
L. Ambrosio, N. Fusco and D. Pallara . Functions of bounded variation and
free discontinuity problems, Clarendon Press. 2000.
M. Angelillo and A. Fortunato. Compatibility of loads and distortions for
unilateral materials. In preparation.
208 M. Angelillo
3.5
3.0
horizontal force [105N]

2.5
2.0
1.5
1.0
0.5
0.0
0 5 10 15 20 25
horizontal displacement [mm]
Figure 64. Masonry walls with openings under vertical and horizontal loads: shearing
force versus roof displacement
M. Angelillo and L. Giliberti. Statica delle strutture murarie. Giornale del

genio Civile, 1988 (in Italian).
M. Angelillo and R.S. Olivito. Experimental analysis of masonry walls
loaded horizontally in plane. Masonry International, 8 (3):91–100, 1995.
M. Angelillo and F. Rosso. On statically admissible stress fields for a plane
masonry-like structure. Quarterly Of Applied Mathematics, 53 (4):731–
751, 1995.
M. Angelillo, L. Cardamone, and A. Fortunato. A numerical model for
masonry-like structures. Journal of Mechanics of Materials and Struc-
tures, 5:583–615, 2010.
M. Angelillo, E. Babilio, and A. Fortunato. Singular stress fields for
masonry-like vaults. Continuum Mechanics And Thermodynamics, 2012.
M. Angelillo, A. Fortunato, M. Lippiello, and A. Montanino. Singular stress
fields and the equilibrium of masonry walls. Meccanica, under revision,
2013.
A. Benedetti and E. Steli. Analytical models for shear-displacement curves
of unreinforced and frp reinforced masonry panels. Constr. Build. Mater.,
22 (3):175–185, 2008.
E. Benvenuto. An Introduction on the History of Structural Mechanics Part
II: Vaulted Structures and Elastic Systems, Springer-Verlag. Springer
Verlag, 1991.
G. Dal Maso, A. De Simone, and M.G. Mora. Quasistatic evolution problems
for linearly elastic perfectly plastic materials. Arch. Rat. Mech. Anal.,
2004.
E. De Giorgi. Congetture riguardanti alcuni problemi di evoluzione. Duke
Math. J., 81(2):255–268, 1996.
G. Del Piero. Constitutive equation and compatibility of the external loads
for linear elastic masonry–like materials. Meccanica, 24:150–162, 1989.
G. Del Piero. Limit analysis and no–tension materials. Int. J. Plasticity,

14:259–271, 1998.
F. Derand. L’architecture des voutes, Cramoisy. 1643.
C. L. Dym and I. H. Shames. Solid Mechanics: a variational approach, Mc
Graw Hill. 1973.
M. Epstein. On the wrinkling of anisotropic elastic membranes. J. Elast.,
55:99–108, 1999.
Eucentre. Prove murature. 2010. URL
http://www.eucentre.it/provemurature.
A. Fortunato. Elastic solutions for masonry-like panels. J. Elas., 98:87–110,
2010.
M. Giaquinta and E. Giusti. Researches on the equilibrium of masonry
structures. Arch. Rational Mech. Analysis, 88:359–392, 1985. ISSN 0950-
2289.
E. M. Gurtin. The linear theory of elasticity, in Handbuch der Physik, band
VIa/2, Springer-Verlag. 1972.
J. Heyman. The stone skeleton: structural engineering of masonry archi-
tecture. Cambridge University Press, 1995.
S. Huerta. Arcos, bovedas y cupulas. geometria y equilibrio en el calculo
tradicional de estructuras de fabrica. Report: Instituto Juan de Herrera,
2004 (in Spanish).
S. Huerta. The analysis of masonry architecture: a historical approach.
Arch. Sc. Review, 51(4):297–328, 2008.
C. T. Kelley. Iterative Methods for Optimization”, Frontiers in Applied
Mathematics 18, SIAM. 1999.
E. Kreyszig. Introductory Functional Analysis with Applications, John Wi-
ley. 1989.
C. Padovani M. Lucchesi and N. Zani. Masonry-like solids with bounded
compressive stress. Int. J. Solids Struct., 33 (14):1961–1964, 1996.
M. Šilhavý M. Lucchesi and N. Zani. Singular equilibrated stress fields
for no-tension panels. In Lecture Notes in Applied and Computational
Mechanics, 23, Springer, pages 255–265, 2005.
G. Pasquinelli M. Lucchesi, C. Padovani and N. Zani. Masonry construc-
tions: mechanical models and numerical applications, Lecture Notes in
Applied and Computational Mechanics 39, Springer. 2008.
E. H. Mansfield. Tension field theory. In Proc. 12th Int. Cong. App. Mech.,
M. Hetenyi and W. G. Vincenti (eds.), Springer, 1969.
E. H. Mansfield. The bending and stretching of plates, Cambridge University
Press. 1989.
E. Mery. Memoire sur l’equilibre des voutes en berceau. Annales des pontes
et chausees, 1 (2):50–57, 1840.
210 M. Angelillo
A. Mielke and M. Ortiz. A class of minimum principles for characterizing the

trajectories and the relaxation of dissipative systems. ESAIM Control
Optim. Calc. Var., 14 (3):494–516, 2008.
M. Ortiz and J. C. Simo. An analysis of a new class of integration algorithms
for elastoplastic constitutive relations. Int. J. Numer. Methods Eng., 23
(3):353–366, 1986.
D. J. Steigmann. Tension–field theory. Proc. R. Soc. Lond. A, 429:141–173,
1990.
R. Temam and G. Strang. Functions of bounded deformation. Arch. Rat.
Mech. Anal., 75 (1):57–73, 1994. ISSN 1980.
S. Timoshenko and J. N. Goodier. Theory of elasticity, Mc Graw Hill. 1951.
Y. W. Wong and S. Pellegrino. Wrinkled membranes ii: analytical models.
J. Mech. Mater. Struct., 1:27–60, 2006.
II Refined Models
Modeling the interfaces in masonry structures
Frédéric Lebon∗
Aix-Marseille University, LMA, UPR7051, CNRS, Centrale Marseille, 13402
Marseille, France
[email protected]
Abstract This chapter deals with some models for interfaces in

the case of masonry structures. Some experimental studies are re-
called in the first part. In the second part, four interface models
are presented.
1 Introduction
The aim of the chapter is to present some ideas for modeling interfaces in
masonry structures. Devising means of modeling interfaces between solids
in structural assemblies is obviously now of great importance in the fields
of mechanical technology and civil engineering. These interfaces contribute
crucially to the strength of many structures, such as optics lenses, airplanes,
asphalt pavements and masonry, for example. It is therefore necessary to
develop rather fine models. One of the main problems which often arises
in this context is that of developing a unified theory: from the tribological
point of view, the contact is often unique, in the sense that it depends on the
materials, roughness, wear, etc., and especially on the mechanical system
involved. The problem of cracking in pavements is obviously quite different
from that of the cornering of an airplane tire. Another problem is due to the
smallness of the interface in comparison with the size of the structure, as well
as the possibly weak mechanical characteristics (in the case of old mortar,
for example). A large number of studies have been devoted to the behavior
of interfaces. Two main modeling approaches used for this purpose are phe-
nomenological modeling and deductive modeling. In the first approach, the
thickness of the interface is taken to be zero and the mechanical properties
are obtained from physical considerations and experiments (see for exam-
ple Frémond (1987); Point and Sacco (1996); Freddi and Frémond (2006);
Raous (2011); Bonetti and Frémond (2011) and references therein). The
∗
The author thanks F. Fouchal, C. Pelissou, A. Rekik, R. Rizzoni and I. Titeux for their
contribution in this work
M. Angelillo (Ed.), Mechanics of Masonry Structures, CISM International Centre for Me-
chanical Sciences, DOI 10.1007/ 978-3-7091-1774-3_5, © CISM, Udine 2014
214 F. Lebon
second approach consists in focusing on the thin layers of material at the

micro-mechanical level, which are usually called the interphase. The me-
chanical parameters of the interface model, characterized by zero thickness,
are identified on the basis of the parameters of the material constituting
this interphase. These boundary conditions have been extensively used to
model imperfect interface properties. One of the method commonly used
to model interface conditions is based on the use of asymptotic techniques
(see for example Klarbring (1991); Licht and Michaille (1997); Lebon et al.
(1997, 2004); Lebon and Ronel (2007); Lebon and Rizzoni (2008) and refer-
ences therein), in order to include microscopic considerations in the interface
model.
Figure 1. Irregular type of masonry a) b) d) Typical Lozerian structures,

France c) Miramas-Le-Vieux, France.
Obviously, one of the problem which arises when modeling interfaces in

masonry is due to the the irregularity of the structures (see figure 1). In
this chapter, we will deal only with fairly regular structures (see figure 2).
Another problem arises with the constitutive equation for the blocks because
stones are quite rigid blocks (see Lucchesi et al. (2008); Como (2010)). In
Modeling the Interfaces in Masonry Structures 215
Figure 2. Examples of regular masonry a)Lozere, France b) Bologne, Italy

c) Albi, France d) Ferrara, Italy.
what follows, we will deal only with deformable elastic bricks.

The last main problem depends on whether or not mortar is present in the
structure and if so, how to write the constitutive equation for the mortar.
If there is no mortar, the contact between blocks can be assimilated to dry
friction. In the presence of mortar, it is necessary to model this component.
The thickness of the mortar is also an important modeling parameter. If
the mortar is thin, an interface law can be used directly for this purpose
(see figure 3). If the thickness of the mortar is not negligible, it is neces-
sary to introduce some additional considerations in order to account for the
interactions between the bricks and the mortar (see figure 4).
In this chapter, it is proposed to present some interface laws. In the
first part, the classical phenomenological law of unilateral contact with dry
friction is recalled. In the second part, a phenomenological law of adhe-
sion is presented and modeled based on the adhesion variable introduced
by Fremond (Frémond (1987)). Two deductive interface laws are presented:
the first one is linear and the second one is non linear and takes the dam-
age which occurs at interface of this kind into account. The last modeling
216 F. Lebon
Figure 3. A wall without mortar: a study at local level.
approach presented here includes mechanical processes responsible for the

failure, such as the unilateral contact, the friction at the microscopic scale
and the normal and tangential damage. The four models discussed in this
chapter were developed in the context of multibody mechanics.
This paper is composed of three parts. In the first part, experimental

results are recalled. The second part is devoted to the modeling of friction
between blocks with no mortar. In the third part of the paper, the phe-
nomenological model of adhesion is presented in the form of two imperfect
interface models based on the use of asymptotic techniques. Some numerical
examples are provided.
2 Some comments on experimental results

2.1 Brick-brick interactions: dry friction
Michel Jean (Jean and Moreau (1994)) conducted experiments on a small
wall (figure 5) consisting of 105 rigid blocks. each of which was 49 mm high,
and 124 mm wide (62 mm in the case of half bricks). The wall was set on
Figure 4. A wall with mortar: a study at local level.
a rigid foundation. The left half of the rigid foundation could be moved
down. The total possible displacement was 6 cm.
Figure 5 shows the displacement of the blocks. Note that the interactions
between the blocks are limited to contact (there is no penetration) and dry
friction. The displacement of the blocks is strongly heterogeneous, especially
along the diagonal.
2.2 Brick-mortar interactions

This section deals with the behavior of deformable blocks. The local
behavior of the interfaces between full and hollow bricks and mortar joints,
which are typical quasi-brittle interfaces (see Gabor (2002); Gabor et al.
(2006)), has been studied by various authors. We will attempt here to
summarize the results obtained.
The experimental device developed by Fouchal (2006); Fouchal et al. (2009)
(figures. 6 and 7) was designed to study on the local scale the shear behavior
of a simple assembly consisting of two and three full or hollow bricks (210 ×
50 mm) connected by a mortar joint 10 mm thick. The samples were
subjected to a monotonous increasing load up to failure.
218 F. Lebon
Figure 5. Experimental device involving 105 blocks (Jean and Moreau

(1994))
Figure 6. Experimental device involving two bricks
The following findings were obtained (see also figure 8):

• rigid elastic behavior up to the failure, followed by friction sliding
behavior;
• the behavior of full bricks was fragile beyond the limit strength;
• the behavior of hollow bricks was quasi-fragile beyond the limit strength;
• hollow brick samples showed great dispersion, mainly due to the non
uniform distribution of the mortar spikes and local defects in the com-
Figure 7. Experimental device involving three bricks
ponents of the bricks;

• samples consisting of two and three bricks showed similar behavior.
The choice of basic cell therefore has no effect on the local scale.
3 Masonry structures without mortar: dry friction

modeling (model 1)
This section focuses on structures devoid of mortar, or including mortar
with a very low stiffness. In a first approach, the contact is assumed to
involve dry friction with no penetration between the blocks.
For the sake of simplicity, we have adopted the framework of contact
between two deformable solids (see figure 9). The contact can be defined
by a punctual correspondence between two surfaces in contact Γ1c and Γ2c
belonging to the domains Ω1 and Ω2 of d (d = 2, 3), respectively. We
assume that initially Γc = Γ1c = Γ2c . The relative displacement between
two points located on the two surfaces in contact is denoted by [u] with
[u] = u1 − u2 . Let F be the density of the contact forces. We take n1 and
n2 to denote the external unit normal vectors to the boundaries of the two
domains. The decomposition into normal and tangential parts is written:
[u] = [uN ]n1 + [uT ] with [uN ] = [u].n1 (1)

1 1
F = FN n + FT with FN = F.n (2)
220 F. Lebon
Figure 8. Typical behavior of the mortar/brick interface (with full and

hollow bricks)
3.1 Unilateral contact

The contact conditions (Signorini conditions) are written:
FN ≥ 0; [uN ] ≥ 0; FN [uN ] = 0 (3)
3.2 Dry Friction

The friction conditions (Coulomb’s law) are written:
||FT || ≤ μ|FN |
||FT || < μ|FN | ⇒ [u˙T ] = 0 (4)
||FT || = μ|FN | ⇒ ∃λ ≥ 0, [u˙T ] = λFT
where μ is the friction coefficient.
3.3 Formulations
The bodies are assumed to be elastic and the deformations are assumed
to be small. In this case, formulation of three kinds can be used: primal (the
unknowns are the displacements), dual (the unknowns are the stresses) and
mixed formulations (the unknowns are the displacements and the stresses).
Figure 9. Dry contact between two blocks
In this chapter, two kinds of formulations are presented. Note that there
exists a large class of methods for solving problems of this kind (see Raous
et al. (1988); Lebon and Raous (1992); Chabrand et al. (1998); Lebon (2003);
Fortin et al. (2002); Alart and Curnier (1991); Alart and Lebon (1995);
Wriggers (2006) and references therein). Two corresponding algorithms are
presented below.
Primal formulation (see Cocu et al. (1995))

Let us take H 1 (Ω) to denote the order 1 Sobolev space, and H and H0
to denote the subspaces of (H 1 (Ω))2 defined by
H = {v ∈ (H 1 (Ω))2 , v = U on Γd },
(5)
H0 = {v ∈ (H 1 (Ω))2 , v = 0 on Γd },
We define K as the convex of contact of kinematically admissible fields

K = {v ∈ H, [vN ] ≥ 0 on Γc }. u0 is taken to denote the initial displacements
222 F. Lebon
fields.
⎧
⎪
⎪ u0 ∈ H be given, find u ∈ V = H 1 (0, T ; K) such that u(0) = u0 in Ω
⎪
⎪
⎨ a(u, v − u̇) + j(u, v) − j(u, u̇) ≤ l(v − u̇), ∀v ∈ H0

⎪
⎪ σN (u)([zN ] − [uN ])dl ≤ 0, ∀z ∈ K
⎪
⎪
⎩
Γc
(6)
with
j(u, v) = − μσN (u)||[v̇T ]||dl,
Γc

l(v) = F vdx + f vdl (7)
Ω Γf

a(u, v) = e(u)Ae(v)dx.
Ω
k+1 k
Δuk
Using a finite difference, u̇(tk+1 ) " u Δt−u = Δt , with uk = u(tk ), at
time tk+1 , we have

Find uk+1 = uk + Δuk ∈ K such that
a(uk+1 , v − Δuk ) + j(uk+1 , v) − j(uk+1 , Δuk ) ≤ l(v − Δuk ), ∀v ∈ K.
(8)
Note that the results obtained on the existence, unicity and regularity of the
solution for this problem have been discussed in Cocu (1984). An example
of non uniqueness is given in Hild (2003).
Fixed point Method FPM (see Raous et al. (1988); Lebon and Raous
(1992)) In the following sections, the indices k (time) will be omitted (this
is a static formulation). The above problem is expressed as a fixed point
problem on the sliding limit (Duvaut and Lions (1976); Cocu (1984))
λ −→ −μσN (u(λ)), (9)
u is the solution of a variational inequation similar to the above one, where
j(., .) is replaced by j(.) defined by

j(v) = λ|vT |dl. (10)
Γc
The latter problem is equivalent to the minimization of L(v) = 21 a(v, v) −

l(v) + j(v) in K. It is possible to regularize the non-differentiable term
in the minimization problem, by replacing the√ absolute value by a smooth

function. For example, a suitable value of x2 + 2 or Ln(ch(x/)) can
be chosen (Lebon (1995)). This regularization procedure does not affect
numerical structures. The problem is usually discretized by a linear inter-
polation (Finite Element Method). Let h be the discretization step. The
displacement gives to a vector of dimension Nh . If Ih is the set of degrees of
freedom involved in the friction, the fixed point iteration consists in finding
the vector of dimension Ih , λh , which is the fixed point of the previously
discretized problem. The algorithm, which is convergent, is written:
0 Initialization u0 be given
1 Fixed Point (Iteration k) λk −→ −μσN (u(λk−1 ))
2 Minimization (Relaxation)
A Initialization uk,0 = uk
B Resolution (iteration n, (aij ) is the stiffness matrix)
Case 1 With a normal component of a contact node
k,n+ 21 1
%
j=i−1 %h
j=N
ui = aii (fi − aij uk,n
j − aij uk,n+1
j )
j=1 j=i+1 (11)
k,n+ 21
If ui ≥ 0 then uk,n+1
i = 0.
Case 2 With a tangential component of a contact node
1
j=i−1 h
j=N
uk,n+1
i = (fi − k,n
aij uj − aij uk,n+1
j
aii j=1 j=i+1
+λki (uk,n+1
i ))
If x ≥ 0, (x) = 1, if x ≤ 0, (x) = −1, if x = 0, (x) = 0.
(12)
Case 3 With a component of a free node (not involved in the
contact)
1
j=i−1 h
j=N
uk,n+1
i = (fi − k,n
aij uj − aij uk,n+1
j ). (13)
aii j=1 j=i+1
C Convergence test : Yes : uk+1 = uk,n+1 , Go to 3; No : Go to B

3 Convergence Test Yes = End; No = Go to 1
At each iteration k, the energy present in the discretized convex Kh has
to be minimized. This step is performed using the relaxation procedure
presented above. Since the functional is strictly convex, the minimization
procedure has a unique solution.
224 F. Lebon
Mixed formulation (see Alart and Curnier (1991); Alart and Lebon
(1995))
The idea underlying this method is to write the problem as an equilib-
rium equation,
⎧
⎨ F ind u and λ such that
F int (u) + F ext + F([u], F ) = 0, (14)
⎩
a(F − F([u], F )) = 0,
where a is a given coefficient, F ext are the given external forces, F int are
internal forces which depend on the constitutive equation and the kinemat-
ics, and F([u], λ) is the friction map. In the case of a node in contact with a
rigid obstacle (i.e. [u] = u), in elasto-statics, we have (see Alart and Curnier
(1991); Alart and Lebon (1995))
F(u, F ) = projR− (τN )n + projC(projR− (τN )) (τT ), (15)
proj is the projection operator, C(FN ) is the Coulomb’s cone and r is a

parameter (usually equal to −r−1 ). τN et τT are given by
τN = FN + ruN , τT = FT + ruT . (16)
The non-linear problem is then written
D(x) + U (x) = 0 (17)
where / 0 / 0
Au + f ext F(u, F )
D(x) = and U(x) = . (18)
aF −aF(u, F )
This system is solved using a Generalized Newton method
xi+1 = xi − (#D(xi ) + ∂U(xi ))−1 (D(xi ) + U (xi )) (19)
∂U (xi ) is a point in the Jacobian set of matrices. The linearized system (19)
is solved by a solver dedicated to non symmetric systems. The algorithm
can be summed up as follows:
0 Initialization x0 = (u0 , λ0 ) be given
1 Resolution (Non-symmetric linear system)
y i+1 = (#D(xi ) + ∂U(xi ))−1 (D(xi ) + U (xi ))
2 Updating xi+1 = xi − y i+1
3 Convergence Test Yes = End; No = Go to Step 1
Concluding comment The numerical problem can be solved using the

open computer code LMGC90 (http://www.lmgc.univ-montp2.fr/˜dubois-
/LMGC90/). This code is a numerical platform dedicated to the modeling
and simulation of dynamic multibody problems. Problems are approached
in the general framework of dynamics (see Moreau (1988); Jean (1999)).
The discretized equations involved in the problem are written:
M q̈ = F (q, q̇) + P (t) + r,

+ interface conditions (20)
+ initial conditions and boundary conditions
where q is a parametrization of the system (degrees of freedom), M is the

mass matrix, F (q, q̇) + P (t) are the internal and external loading vectors,
and r is the vector of contact forces.
The problem of the 105 blocks presented in the previous section can be
solved using this technique. The results are presented in figure 10. These
results were obtained by J. J. Moreau (Jean and Moreau (1994)). The
correspondence observed seems to be perfect.
Figure 10. Comparison between experimental and numerical findings on

the example of 105 blocks (Jean and Moreau (1994))
226 F. Lebon
4 Masonry blocks with mortar

4.1 A phenomenological model taking the adhesion into account
(model 2)
In this section, the RCCM model is presented (Raous et al. (1999);
Monerie and Raous (2000)). This adhesion model was used to model mor-
tar/brick interfaces in Fouchal et al. (2009). In this model, the unilateral
contact conditions (non penetration between the mortar and the brick, as
in the previous section) are combined with dry friction and adhesion be-
tween bricks and mortar. The local constitutive equations required for this
model are deduced from thermodynamic considerations and based on a ma-
terial surface hypothesis on the contact zone. The basic idea underlying this
model is to introduce a new state variable describing the contact state. This
adhesion intensity variable, denoted b, was initially introduced by Frémond
(1987). This variable gives the relative proportion of the active links be-
tween two bodies in contact. This variable is chosen so that:
• b = 1, total adhesion
• 0 < b < 1, partial adhesion
• b = 0, no adhesion
As in the previous section, we are working in the framework of the con-
tact between two deformable solids, and the same notations are adopted
here. w is taken to denote the Dupré energy, and CN (resp. CT ) to de-
note the initial normal (resp. tangential) stiffness of the interface. In what
follows, p, q and r are three given numbers. The constitutive equations of
the interface are given by the following equations, based on state laws and
complementarity laws:
Unilateral contact with adhesion
RN − CN [uN ]bp ≥ 0; [uN ] ≥ 0; (RN − CN [uN ]bp )[uN ] = 0 (21)
Friction with adhesion

||RT − CT [uT ]bq || ≤ μ|RN − CN [uN ]bp |
||RT − CT [uT ]bq || < μ|RN − CN [uN ]bp | ⇒ [u˙T ] = 0
(22)
||RT − CT [uT ]bq || = μ|RN − CN [uN ]bp | ⇒ ∃λ ≥ 0,
[u˙T ] = λ(RT − CT [uT ]bq )
Evolution of the intensity of adhesion
0 = −(w − ( 21 CN [uT ]2 + 12 CT |[uT ]|2 )br )− if b ∈ [0, 1[ (23)

Note that if there is no adhesion (b = 0), this model involves the classical
Signorini-Coulomb problem.
A graphic interpretation of the normal part of this RCCM model given in
figure 11 (for (p, q, r) = (2, 2, 1)) shows the changes with time in the normal
forces depending on the normal displacement jump. The changes in b lead
to irreversible effects. If b decreases, the adhesive forces will decrease and
eventually disappear. In the case of pure traction ([uN ] > 0), the adhesion
resistance (RN = CN [uN ]p ) is activated (elasticity without damage). b
decreases when the displacement becomes sufficiently large for the elastic
energy to become larger than the adhesion limit w.
Figure 11. Normalized normal forces versus the normalized normal jump
in the displacement
A numerical example (see Fouchal et al. (2009))

Problems are usually approached using θ-methods and the Non-Smooth
Contact Dynamics (NSCD) method (Moreau (1988); Jean (1999)). Due to
the contact conditions, a fairly small time-step is chosen and the problem is
condensed in the local frame associated with the contact nodes. The local
problem is solved using a non-linear Gauss-Seidel method.
The interface is governed by the RCCM law presented above. Contact
between bodies is defined by contact nodes. Contact nodes are located
between two nodes in the mesh of an element in contact at distances of 0.2
and 0.8, respectively, along each of the segments in contact.
The bodies (bricks and mortar) are modeled using Q4 quadrangular finite
elements. The numerical tests are performed with a constant time step
equal to Δt = 0.5 ∗ 10−3 s. The computations require 2000 increments in
228 F. Lebon
order to reach values resembling the experimental data, and θ (in the time
integration method) is fixed and taken to be equal to 0.55.
The example presented shows the evolution of a triplet of full bricks and
comparisons are made in figures 12 and 13 between two experiments and
the numerical results.
Figure 12. Damage evolution in a triplet of full-bricks
4.2 Deductive models: linear (and non linear) multi-scale models

Generalities on asymptotic methods There exist a large class of asymp-
totic methods, such as matched asymptotic expansions (Eckhaus (1979);
Sanchez-Hubert and Sanchez-Palencia (1992)). The general idea is to dilate
(blow up) the interphase from the thickness η to 1 (see Figure 14). We have
two expansions of the displacement uη , the strain e(uη ) and the stress σ η
in the powers of η, that is, an external one in the adherents and an internal
in the joint. We have to connect these two expansions along the interface.
In what follows, we study a problem in 2 dimensions in order to simplify
the computations. The relations obtained in the internal expansions will be
expressed using values that intervene in the external expansions.
a) External expansions The external expansion is a classical expansion
Figure 13. Comparison between experimental and numerical results ob-

tained on full brick triplets
Figure 14. Blow-up process

230 F. Lebon
in powers of η
uη (x1 , x2 ) = u0 (x1 , x2 ) + ηu1 (x1 , x2 ) + ...,

eij (uη )(x1 , x2 ) = e0ij + ηe1ij + ...,
1 ∂ui
l ∂ulj (24)
elij = 2 ( ∂xj + ∂xi ),
η 0 1
σij (x1 , x2 ) = σij (x1 , x2 ) + eσij (x1 , x2 ) + ...
b) Internal expansions In the internal expansion, we perform a blow-up of

the second variable. Let y2 = xe2 . The internal expansion gives
uη (x1 , x2 ) = v 0 (x1 , y2 ) + ηv 1 (x1 , y2 ) + ...,

εij (uη )(x1 , y2 ) = η −1 e−1 0 1
ij + eij + eεij + ...,
∂v1l
el11 = ∂x1 ,
∂v2l+1
el22 = ∂y2 ,
1 ∂v2
l
∂v1l+1
el12 = 2 ( ∂x1 + ∂y2 ),
−1
η
σij (x1 , y2 ) = η −1 τij (x1 , y2 ) + τij0 (x1 , y2 ) + ητij1 (x1 , y2 ) + ...,
η % l ∂τi1
∞ l
∂τ l+1
σij,j = η ( ∂x1 + ∂yi22 ).
l=−2
(25)
We use the convention
v l = 0, l < 0, τ l = 0, l < −1. (26)
c) Continuity conditions The third step in the method consists in the con-
necting of the two expansions. If the interface between the mortar and the
brick is perfect, we have continuity of the displacement and of the stress
tensor along this interface. This gives:
(i) v 0 (x1 , ±1/2) = u0 (x1 , ±1/2),

(ii) τ −1 (x1 , ± ± 1/2) = 0, (27)
(ii) τ 0 (x1 , ±1/2).e2 = σ 0 (x1 , ±1/2).e2 .
A linear multi-scale model (model 3) The equilibrium equation be-

comes:
−1
(η −1 τij + τij0 + ητij1 + ...),j = 0 (28)
We obtain
τij−1 = 0
0 (29)
τi2,2 = 0
that is
+ 0
,
τi2 .e2 = 0 (30)
The interphase is assumed to be linearly elastic, and the above theory is
applied to this kind of material. We have
−1

η −1 τij + τij0 + ητij1 + ... = Cijkl η −1 e−1 0 1
kl + ekl + ηekl + (31)
where C denotes the stiffness tensor.

We obtain:
0 0
τ12 = C1212 v1,2
0 0 (32)
τ22 = C2222 v2,2
that is
0
+ ,
τ12 = C1212 +v10 ,
0 (33)
τ22 = C2222 v20
Using the continuity conditions, we obtain
0
σij = 0 in the brick and the mortar
+ , (34)
0
σ .e2 & u0 along the interface
= C
& is a matrix consisting of C1212 and C2222 . Note that this technique
where C
was used in Rekik and Lebon (2010) and Rekik and Lebon (2012).
An example of a non linear multi-scale model including micro-

cracks (model 4) (see Pelissou and Lebon (2009))
General considerations and notations
The model described in this section is an extension of the bulk model
introduced in Gambarotta and Lagomarsino (1997), which takes the damage
to the mortar joint into account. The interface modeling procedure (figure
15) consists of three steps:
• Let us take a macroscopic bulk model for quasi-fragile materials;
• The structure is assumed to consist of three phases: material 1 (brick,
for example), material 2 (mortar, for example) and a thin interphase
between the two materials, consisting of the material described in the
first step;
• Since the interphase is thin, an interface model is developed by per-
forming an asymptotic analysis, as described above (Geymonat and
Krasucki (1997); Lebon and Ronel-Idrissi (2004)) (the thickness of the
interface tends to zero).
232 F. Lebon
Figure 15. Interface modeling in three steps
The model obtained memorizes some of the geometrical and mechanical

characteristics of the interphase, such as the thickness, elastic coefficients,
normal and tangential stress, damage variable, etc.
For the sake of simplicity, the structure is taken to occupy an open
bounded set Ω of 2 with a smooth boundary ∂Ω. The two dimensional
space is referred to the orthonormal frame (O, x1 , x2 ).
First step: The bulk model
In this section, the bulk model introduced by Gambarotta and Lago-
marsino (1997) is briefly described. In this model dedicated to masonry
structures, the masonry is regarded as a ”material” showing nonlinear dam-
age behavior. The macroscopic behavior is accounted for by applying av-
eraging process based on microscopic considerations. In the line with the
classical procedure, the strain tensor is decomposed in its linear and non-
linear (anelastic) parts:
e = eel + ean (35)
where
σ = Ceel and ean = Sσ σ (36)
At the local level (in an elementary volume, see figure 16), the stress vector
is decomposed into its normal σN and tangential σT parts, and the normal
and tangential components of ean are expressed by (37) :

ean
N = hαH(σN )σN
(37)
T = kα(σT − f )
ean
where f = σT if σT ∈ I = ] − μσN , +μσN [and f = ±μσN if σT ∈ / I (figure

17), H is the Heaviside function of the unilateral response of the joint and
α is the damage variable. h and k are positive coefficients standing for the
opening and sliding compliances of the mortar joint and μ is the internal
friction coefficient.
Figure 16. Normal and tangential stress vector components
Figure 17. Relation between f and σT (friction threshold)
We can write
ean
t = kαχI (σT )(σT ± μσN ) (38)
234 F. Lebon
10
R
9
0
0 2 4 6 8 10 12
damage
Figure 18. Evolution of R(α)
where χI is the characteristic function of the tangential damage under-

gone by the set I, such that:
- If σT ∈ I, χI (τ ) = 1, there is tangential damage.
- If σT ∈
/ I, χI (σT ) = 0, there is no tangential damage.
Therefore Sσ is given locally in matrix form by:

hα H(σN ) 0
Sσ = (39)
± kα χI (σT ) μ kα χI (σT )
The damage is governed by a yield condition Φ(α) ≤ 0, where
Φ(α) = Y − R(α)
1 2 1 (40)
Y = hH(σN )σN + kχI (σT )σT2
2 2
where R is the toughness of the material defined by (see figure 18)
R(α) = R0 α, if 0 ≤ α ≤ 1
(41)
R(α) = R0 /α, if α ≥ 1
Second step: The interphase

In this section, the structure is taken to consist of two materials sepa-
rated by a thin interphase which is a ”mixture” of the other two materials.
The interphase is assumed to be parallel to the x1 -axis and the thickness,
which is constant, is denoted by η. In what follows, the constitutive equa-
tions are those given in the previous section, i.e., the interface consists of a
quasi-brittle material. The elastic part is assumed to be isotropic, and in

view of the external normal vector x2 , the constitutive equation is
−1
σ = Id + CSσ Ce = Cσ ε (42)
Upon introducing the Lamé’s coefficients, λ and G, we can write:

⎛ λ2 hαH λ(λ+2G)hαH
⎞
λ + 2G − 1+(λ+2G)hαH λ − 1+(λ+2G)hαH 0
⎜ λ λ+2G ⎟
Cσ = ⎝ 1+(λ+2G)hαH 1+(λ+2G)hαH 0 ⎠
±μλGkαχI ±μ(λ+2G)GkαχI
− (1+(λ+2G)hαH)(1+Gkαχ I)
− (1+(λ+2G)hαH)(1+GkαχI )
G
1+GkαχI
(43)
Interface behavior
Using this asymptotical approach (see the previous section or Lebon and
Ronel-Idrissi (2004) in another context), and substituting the asymptotic
expansions into the constitutive equations and the equilibrium equations, we
obtain expressions linking the stress vector to the jump in the displacement
denoted [u]. This gives:
0
0 λ+2G ∂v2
τ22 = lim
e→0 1+(λ+2G)hαH ∂y2 (44)
0 ±μ(λ+2G)GkαχI ∂v20 G ∂v10
τ12 = lim (1+(λ+2G)hαH)(1+Gkαχ )
I ∂y2
+ lim 1+Gkαχ I ∂y2
e→0 e→0
By integration, the A-GL (Asymptotic Gambarotta-Lagomarsino) model is

therefore given by the following system (in the terms of the normal and
tangential components):
⎧ CN
⎪
⎪
⎨ σN = 1 + C h̄α H(σ ) [uN ]
N N
⎪
⎪ ± C CT μ k̄α χI CT
⎩ σT = N
[uN ] + [uT ]
( 1 + CN h̄α H(σN ) ) ( 1 + CT k̄αχI ) 1 + CT k̄α χI
(45)
where
CN = (λ + 2G)/e
CT = G/e
(46)
h̄ = he
k̄ = ke
The resulting matrix is not diagonal, contrary to the classical case: a non
symmetric coupling term occurs between the normal compliance term CN
and the tangential compliance term CT . It can be noted that the configu-
ration of traction when f = 0 is combined in this general expression with
236 F. Lebon
that of the compression when f = σT and when the friction threshold is

reached at f = ± μ σN .
The damage is governed by a yield condition Φ̄(α) = Ȳ − R̄(α) ≤ 0,
where
⎧
⎪ 1 2 1
⎪
⎪
⎪ Ȳ = h̄ H(σN ) σN + k̄ χI (σT ) (σT − ±μσN )2
⎨ 2 ⎧ 2
⎨R̄0 α if 0 < α < 1 (47)
⎪
⎪
⎪
⎪ R̄ (α) = lim R e =
⎩ e→0 ⎩ R̄0 if α > 1
α
A numerical example: shear test on a simple brick Let us perform

an academic shear test on a single rectangular piece. The piece is bonded at
the bottom and subjected to horizontal forces on the left. Details of this test
are presented in figure 19. This horizontal force which is equal to 20 kN , is
applied progressively. The load is applied on the right and left sides. The
(academic) coefficients are CN = CT = 500 kN cm−3 , μ = 0.3, h̄ = k̄ =
0.04kN −1 cm3 , R̄0 = 10kN cm−1 .
Figure 19. Shear test
Due to the geometry and the loading conditions, the behavior of the
contact zone is very complex. Figure 20 shows the evolution of the dam-
age along the contact zone, which is strongly non linear. In particular,
the damage increases strongly in an intermediate zone, which is subjected

to both shear and traction forces. Note that the tangential displacement
shows a linear pattern of evolution in the first phase and a non linear pat-
tern corresponding to the failure, in the second phase, as was to be expected.
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
0 2 4 6 8 10 12 14 16 18
Figure 20. Damage evolution for four typical elements (shear test).
5 Conclusion
In this chapter, four of the many existing interface models are presented. It
is quite difficult to solve large problems using this approach. In Rekik and
Lebon (2010, 2012), we established that it can be possible to solve larger
problems (small walls), however. It will obviously be necessary in future
studies to review all the existing models (a simple look at Science DirectT M
with the keywords ”masonry” and ”interface” gives more than 2500 papers),
in order to choose the most suitable one and use efficient solvers (DDM, MG,
etc.) before it will be possible to solve real physical problems.
Bibliography
P. Alart and A. Curnier. A mixed formulation for frictional contact problems
prone to Newton like solution methods. Computer Methods in Applied
Mechanics and Engineering, 92:353–375, 1991.
P. Alart and F. Lebon. Solution of frictional contact problems using ILU
and coarse/fine preconditioners. Computational Mechanics, 16:98–105,
1995.
238 F. Lebon
E. Bonetti and M. Frémond. Analytical results on a model for damaging

in domains and interfaces. ESAIM-Control Optimisation and calculus of
Variations, 17:955–974, 2011.
P. Chabrand, F. Dubois, and M. Raous. Various numerical methods for solv-
ing unilateral contact problems with friction. Mathematical Computer
and Modeling, 28(4-8):97–108, 1998.
M. Cocu. Existence of solutions of Signorini problems with friction. Inter-
national Journal of Engineering Science, 22:567–575, 1984.
M. Cocu, E. Pratt, and M. Raous. Formulation and approximation of
quasistatic frictional contact. Comptes Rendus Académie des Sciences,
Paris, Série I, 320:1413–1420, 1995.
M. Como. Statica delle costruzioni storiche in muratura. Aracne, 2010.
G. Duvaut and J. L Lions. Inequalities in mechanics and physics. Springer
Verlag, 1976.
W. Eckhaus. Asymptotic analysis of singular perturbations. North-Holland,
1979.
J. Fortin, M. Hjiaj, and G. De Saxcé. An improved discrete element method
based on a variational formulation of the frictional contact law. Com-
puters and Geotechnics, 29:609–640, 2002.
F. Fouchal. Contribution à la modélisation numérique des interfaces dans
les structures maçonnées. PhD thesis, Université de Reims Champagne
Ardennes, 2006.
F. Fouchal, F. Lebon, and I. Titeux. Contribution to the modeling of in-
terfaces in masonry construction. Construction and Building Materials,
23:2428–2441, 2009.
F. Freddi and M. Frémond. Damage in domains and interfaces: A coupled
predictive theory. Journal of Mechanics of Materials and Structures, 1:
1205–1233, 2006.
M. Frémond. Adhesion of solids. Journal de Mécanique Théorique et Ap-
pliquée, 6:383–407, 1987.
A. Gabor. Contribution à la caractérisation et à la modélisation des
maçonneries non renforcées et renforcées par materiaux composites. PhD
thesis, Université Claude-Bernard Lyon I, 2002.
A. Gabor, A. Bennani, E. Jacquelin, and F. Lebon. Modelling approaches
of the in-plane shear behaviour of unreinforced and frp strengthened
masonry panels. Composites Structures, 74:277–288, 2006.
L. Gambarotta and S. Lagomarsino. Damage models for the seismic response
of brick masonry walls. part i : The continuum model and its application.
Earthquake Engineering and Structural Dynamics, 26:441–462, 1997.
G. Geymonat and F. Krasucki. Analyse asymptotique du comportement en
flexion de deux plaques collées. Compte Rendu Académie des Sciences
Série I, 325:307–314, 1997.
P. Hild. An example of nonuniqueness for the continuous static unilateral

contact model with coulomb friction. Comptes Rendus Académie des
Sciences, Paris, Série I, 337:685–688, 2003.
M. Jean. The non-smooth contact dynamics method. Computer Methods
in Applied Mechanics and Engineering, 177:235–257, 1999.
M. Jean and J. J. Moreau. Monuments under seismic action. Technical
report, Commission of the European Communities, November 1994. En-
vironment Programme, Contract n◦ EV5V CT93 0300.
A. Klarbring. Derivation of the adhesively bonded joints by the asymptotic
expansion method. International Journal of Engineering Science, 29:
493–512, 1991.
F. Lebon. Two-grid method for regularized frictional elastostatics problems.
Engineering Computations, 12:657–664, 1995.
F. Lebon. Contact problems with friction: Models and simulations. Simu-
lation, Modelling, Practice and Theory, 11:449–464, 2003.
F. Lebon and M. Raous. Friction modelling of a bolted junction under
internal pressure loading. Computers and Structures, 43:925–933, 1992.
F. Lebon and R. Rizzoni. Asymptotic study of soft thin layer: the non
convex case. Mechanics of Advanced Materials and Structures, 15:12–20,
2008.
F. Lebon and S. Ronel. First order numerical analysis of linear thin layers.
ASME Journal of Applied Mechanics, 74:824–828, 2007.
F. Lebon and S. Ronel-Idrissi. Asymptotical analysis of Mohr-Coulomb
and Drucker-Prager soft thin layers. Steel and Composite Structures, 4:
133–147, 2004.
F. Lebon, A. Ould-Khaoua, and C. Licht. Numerical study of soft adhesively
bonded joints in finite elasticity. Computational Mechanics, 21:134–140,
1997.
F. Lebon, R. Rizzoni, and S. Ronel. Analysis of non-linear soft thin inter-
faces. Computers and Structures, 82:1929–1938, 2004.
C. Licht and G. Michaille. A modeling of elastic adhesive bonded joints.
Advances in Mathematical Sciences and Applications, 7:711–740, 1997.
M. Lucchesi, C. Padovani, G. Pasquinelli, and N. Zani. Masonry construc-
tions: mechanical models and numerical applications. Springer, 2008.
Y. Monerie and M. Raous. A model coupling adhesion to friction for the
intercation between a crack and a fiber/matrix interface. ZAMM, 80:
205–209, 2000.
J. J. Moreau. Unilateral contact and dry friction in finite freedom dynamics,
volume 302 of CISM - Courses and Lectures, pages 1–82. Springer, 1988.
C. Pelissou and F. Lebon. Asymptotic modeling of quasi-brittle interfaces.
Computers and Structures, 87:1216–1223, 2009.
240 F. Lebon
N. Point and E. Sacco. A delamination model for laminated composite.

International Journal of Solids and Structures, 33:483–509, 1996.
M. Raous. Interface models coupling adhesion and friction. Comptes Rendus
Mécanique, 339:491–501, 2011.
M. Raous, P. Chabrand, and F. Lebon. Numerical methods for solving uni-
lateral contact problem with friction. Journal of Theoretical and Applied
Mechanics, 7:111–128, 1988.
M. Raous, L. Cangémi, and M. Cocou. A consistent model coupling ad-
hesion, friction and unilateral contact. Compututer Methods in Applied
Mechanics and Engineering, 177:383–399, 1999.
A. Rekik and F. Lebon. Identification of the representative crack length
evolution in a multi-level interface model for quasi-brittle masonry. In-
ternational Journal of Solids and Structures, 47:3011–3021, 2010.
A. Rekik and F. Lebon. Homogenization methods for interface modeling in
damaged masonry. Advances in Engineering Software, 46:35–42, 2012.
J. Sanchez-Hubert and E. Sanchez-Palencia. Introduction aux méthodes
asymptotiques et à l’homogénisation. Masson, 1992.
P. Wriggers. Computational Contact Mechanics. Springer, 2006.
Micro, Multiscale and Macro Models for
Masonry Structures
Elio Sacco
Dipartimento di Ingegneria Civile e Meccanica
Università di Cassino e del Lazio Meridionale
1 Introduction
The development of adequate stress analyses for masonry structures rep-
resents an important task not only for verifying the stability of masonry
constructions, as old buildings, historical town and monumental structures,
but also to properly design effective strengthening and repairing interven-
tions. The analysis of masonry structures is not simple at least for two
reasons: the masonry material presents a strong nonlinear behavior, so
that linear elastic analyses generally cannot be considered as adequate; the
structural schemes, which can be adopted for the masonry structural anal-
yses, are more complex than the ones adopted for concrete or steel framed
structures, as masonry elements require often to be modeled by two- or
three-dimensional elements. As a consequence, the behavior and the analy-
sis of masonry structures still represents one of the most important research
field in civil engineering, receiving great attention from the scientific and
professional community.
Several numerical techniques have been developed to investigate and to
predict the behavior of masonry structures. In fact, in the last decades,
the scientific community has demonstrated great interest in the develop-
ment of sophisticated numerical tools as an opposition to the tradition of
rules-of-thumb or empirical and geometrical formulas adopted to evaluate
the safety of masonry buildings. In particular, nonlinear models imple-
mented in suitable finite elements formulations currently represent the most
common advanced strategy to simulate the structural behavior of masonry
structures. The main problem in the development of accurate stress analy-
ses for masonry structures is the definition and the use of suitable material
constitutive laws.

242 E. Sacco
Taking into account the heterogeneity of the masonry material, which

results from the composition of blocks connected together by mortar joints,
several modeling approaches have been proposed in literature.
Micro-models consider the units and the mortar joints separately, char-
acterized by different constitutive laws; thus, the structural analysis is per-
formed considering each constituent of the masonry material. The mechan-
ical properties that characterize the models adopted for units and mortar
joints, are obtained through experimental tests conducted on the single ma-
terial components (compressive test, tensile test, bending test, etc.). This
approach leads to structural analyses characterized by great computational
effort; in fact, in a finite element formulation framework, both the unit
blocks and the mortar beds have to be discretized, obtaining a problem
with a high number of nodal unknowns.
Micro-macro, i.e. multiscale, models consider different constitutive laws
for the units and the mortar joints; then, a homogenization procedure is
performed obtaining a macro-model for masonry which is used to develop
the structural analysis. Also in this case, the mechanical properties of units
and mortar joints are obtained through experimental tests. The micro-
macro models appear very appealing, as they allow to derive in a rational
way the stress-strain relationship of the masonry, accounting in a suitable
manner for the mechanical properties of each material component. More-
over, this approach can lead to effective models.
Macro-models, or macroscopic models, are based on the use of phe-
nomenological constitutive laws for the masonry material; i.e. the stress-
strain relationships adopted for the structural analysis are derived perform-
ing tests on masonry, without distinguishing the blocks and the mortar
behavior. A macroscopic model could be unable to describe in a detailed
manner some micromechanisms occurring in the damage evolution of ma-
sonry, but it is very effective from a computational point of view when
structural analyses are performed.
In the following, the micromechanical, the multiscale and the macro-
mechanical approaches are presented, giving some details for specific models
proposed in the last years. In particular, the discussion is limited to three
models (micro, micro-macro and macro) framed in the 2D small strain and
displacement approach.
2 Micromechanical modeling
The micromechanical analysis of masonry elements is performed considering
different constitutive laws for the bricks and for the mortar. Moreover, the
adhesion between the mortar and the brick can play a fundamental role
Micro, Multiscale and Macro Models for Masonry Structures 243
Figure 1. (a) the interface models the mortar and the adhesion mortar-
brick; (b) the interface models only the adhesion mortar-brick.
in the overall response of the masonry texture. Thus, the modeling of the
response of the mortar-brick interface can be necessary in order to reproduce
the possible decohesion of the mortar from the brick.
In the micromechanical modeling of the masonry, at least two possible
approaches can be distinguished, schematically illustrated in figure 1:
• the mortar and the adhesion surfaces between the mortar and the
brick are modeled as a unique interface, characterized by a mechanical
response which accounts for the behavior of the mortar and of the
mortar-brick adhesion; in this case, as the interface has zero thickness,
the units are expanded in both the directions by the mortar thickness
(37);
• the mortar is modeled as a continuum material, eventually charac-
terized by nonlinear response, and the adhesion surface between the
mortar and the brick is modeled by a specific interface (59).
In both cases, the interface models are characterized by constitutive laws
relating the stress acting at the interface with the displacement discontinu-
ity. Several interface models have been presented in literature; in particular
some of them have been developed to reproduce the gradual process of crack
opening, in which the incipient separation of the two edges of the crack is
constrained by cohesive stresses due to interaction and friction between
aggregate or bridging phenomena. It is the salient feature of the quasi-
brittle materials like clay brick, mortar, ceramics, rock or concrete, which
244 E. Sacco
fail because of a process of progressive internal crack growth. The idea of

developing a cohesive model interface was introduced by Dugdale (20) and
Barenblatt (7), who proposed several distributions of the cohesive stresses.
In order to model the behavior of masonry mortar joints, Lofti and Shing
(36) proposed a dilatant interface constitutive model capable of simulating
the initiation and the propagation of interface fracture under combined nor-
mal and shear stress in both tension-shear and compression-shear regions.
Giambanco and Di Gati (29) formulated a simple cohesive interface model
adopting a yield surface expressed by the classical bi-linear Coulomb con-
dition with a tension cut-off and with a non-associated flow law. Gam-
barotta and Lagomarsino (27) postulated the constitutive equation of in-
terface model in term of two internal variables representing the frictional
sliding and the damage occurring in the mortar joint because of cyclic load.
Lourenço and Rots (38) and, then, Oliveira and Lourenço (51) implemented
a constitutive interface model, based on an incremental formulation of plas-
ticity theory, able to simulate the cyclic behavior of the cohesive zone by
reproducing the nonlinear response during unloading. Giambanco and Mroz
(31) proposed a model of interface for which the contact stresses and strains
interact with the internal stresses or strains within the joint, separated by
two interfaces from the adjacent material. Alfano and Sacco (5) and Alfano
et al. (4) introduced a new method to combine interface damage and fric-
tion in a cohesive-zone model on the basis of a simplified micromechanical
formulation. They showed the effectiveness of the proposed model to pre-
dict the failure mechanisms and the response of a masonry structure. In
particular, the experimental behavior of a masonry wall subjected to a com-
pression and shear load was reproduced (5). The developed interface model
was able to simulate the progressive damage and failure of the mortar layers
as well as the possible fracture initiation and propagation within the bricks.
Then, this model has been properly modified by Sacco and Toti (59) and
applied to describe the brick-mortar interface response. The interface model
was utilized by Fouchal et al. (25) to simulate the experimental behavior
of the mortar-brick adhesion zones of the masonry structures, remarking
that the decohesion between the constituents of masonry is the main re-
sponsible of its nonlinear response. Numerical analyses of the masonry unit
cell and of a simple masonry structure have been recently developed by Gi-
ambanco et al. (30), making use of a new interphase model implemented
in a finite element framework. An innovative deductive approach based on
a micromechanical analysis and on a homogenization procedure has been
proposed recently by Sacco and Lebon (58). A homogenization technique
has been presented and applied to derive the interface model considering
the brick–mortar interaction.
Figure 2. Micromechanical scheme of the interface.
In the following the model proposed in (5) and (4) and reviewed in (59)
is presented.
2.1 Interface model

The interface cohesive model combining damage and friction is described
in this section. The interface model, developed on the basis of mechanical
model proposed in (5; 4; 59), adopts a micromechanical approach. With
reference to a typical interface zone, as represented in figure 2, a micro-
mechanical analysis of the damaging process is performed. Three different
states can be recognized at the interface: point A, the mortar-brick connec-
tion is absolutely undamaged; point B, partial decohesion between the two
contact surfaces occurred; point C, the decohesion phenomenon is complete.
At each point of the interface a representative element area (REA) is
introduced. At the point A, the interface does not present any detachment
and the associated REA results undamaged. The REA corresponding to
point B contains partial decohesion due to the presence of micro-cracks,
so that the representative area can be schematically split in two parts: an
undamaged part and a damaged one. In the REA corresponding to point C
the coalescence of micro-cracks occurred and a total decohesion is present,
so that a macro-crack appears into the representative element which results
completely damaged. Summarizing, the total area A of the REA can be
246 E. Sacco
Figure 3. Scheme of the REA at a typical point of the interface.
decomposed into a undamaged part Au and a completely damaged part

Ad , such that A = Au + Ad . Following standard arguments of continuum
damage mechanics, the damage parameter D is introduced as ratio between
the damaged part and the whole area:
Ad
D= (1)
A
the two parts of the total area A, i.e. Au and Ad can be recovered as:
Au = (1 − D)A, Ad = A D (2)
According to this definition, D can vary in the range [0, 1], where the case
D = 0 corresponds to the virgin material state and the case D = 1 to the
total damaged state.
The relative displacement vector at the typical point of the interface
is denoted by s. Thus, the vector s represents in the REA the relative
displacement of the two surfaces in contact, as schematically illustrated in
figure 2. Denoting with the subscripts N and T the components in the
normal and tangential direction to the interface, respectively, according to
the local coordinate system illustrated in figure 1 and in figure 2, the relative
T
displacement can be written as s = {sN sT } .
As schematically illustrated in figure 3, the overall behavior of the REA
can be obtained as the superposition of three schemes: the first scheme
considers the REA subjected to a relative displacement se of the interface,
assuming the crack mouths opening equal to zero; in the second scheme,
a relative displacement c, corresponding to a crack opening, is prescribed,
which leads to the overall relative displacement sc of the interface; in the
third scheme, the REA is subjected to a relative displacement p at the
crack mouths, due to the frictional sliding, which induces an overall relative
displacement sp . In such a way, the overall relative displacement is obtained
as:
s = se + s c + s p (3)
The overall constitutive response of the REA is determined developing

a simplified micro-mechanical analysis. In fact, the behavior of the contact
zone is modeled considering the different mechanical responses of the un-
damaged and damaged parts of the REA, associated to the typical point of
the interface.
On both parts of the REA, i.e. on the undamaged part Au and on
the damaged part Ad , the interface stress is supposed to be constant. In
particular, on Au the stress is a linear function of the relative displacement,
while on Ad inelastic effects are accounted for:
τ ue = K se τ uc = K sc τ up = K sp
(4)
τ de = K se τ dc = K (sc − c) τ dp = K (sp − p)
where the inelastic displacement vectors c and p account for the unilateral
nature of the contact and for the friction phenomenon, while K is the di-
agonal matrix collecting the stiffness values in the normal and tangential
directions to interface: / 0
KN 0
K= (5)
0 KT
Then, the stress vectors in the undamaged and damaged parts of the
REA are obtained as:
τu = τ ue + τ uc + τ up
= K se + K sc + K sp = K s
(6)
τd = τ de + τ dc + τ dp
= K se + K (sc − c) + K (sp − p) = K [s − (c + p)]
respectively. The overall interface stress vector on the REA is denoted by
τ and it is obtained by suitably weighting the two stresses τ u and τ d given
by equations (6):
τ = (1 − D) τ u + Dτ d (7)
The inelastic relative displacements c, due to the unilateral contact, is
defined by the relationship:

sN
c = H (sN ) (8)
0
where H (sN ) denotes the Heaviside function, i.e. H (sN ) = 0 if sN ≤ 0 and
H (sN ) = 1 if sN > 0.
The evolution of the inelastic slip relative displacement p, occurring on
the damaged part of the REA, is assumed to be governed by the classical
Coulomb yield function:

d
φ τ d = μ τN −
+ τTd = μτN
d
+ τTd (9)
248 E. Sacco
d
where μ is the friction coefficient and τN −
denotes the negative part of the
d
normal stress τN . The evolution law of p is governed by the relationships:
⎧ ⎫ ⎧ ⎫
⎨ 0 ⎬ ⎨ 0 ⎬
ṗ = λ̇ dφ = λ̇ τd
⎩ ⎭ ⎩ Td ⎭ (10)
dτTd

d τT

λ̇ ≥ 0, φ τ ≤ 0, λ̇φ τ d = 0
with λ the plastic multiplier.

About the evolution of the damage parameter, a model which accounts
for the coupling of mode I of mode II of fracture is considered. In fact, the
two quantities ηN and ηT , defined as the ratios between the first cracking
relative displacements s0N and s0T and the full damage relative displacements
sfN and sfT , respectively, are introduced:
s0N s0N τN
0
s0T s0T τT0
ηN = = , ηT = = (11)
sfN 2 GcN sfT 2 GcT
0
where τN and τT0 are the peak stresses corresponding to the first cracking
relative displacements and GcN and GcT are the specific fracture energies
in mode I and mode II, respectively. Then, the parameter η, which relates
the two modes of fracture, is defined as follows:
2
sN + s2T
η= 2 ηN + 2 ηT (12)
s̃ s̃
T
where s̃ = {sN sT } , with the McCauley bracket . selecting the positive
part of a number. The equivalent relative displacement ratio is introduced
as:
* sN + sT
Y = YN2 + YT2 with YN = , YT = 0 (13)
s0N sT
The damage parameter is assumed to be function of η and of the history of
the equivalent relative displacement Y as:

Y −1
D = max 0, min 1, (14)
history Y (1 − η)
The damage evolution law given by equation (14) allows to obtain a lin-
ear stress - relative displacement softening when the ratio sN /sT is assigned
(59).
Figure 4. Geometry of the arch.
2.2 Numerical applications

The proposed interface model is implemented in a research version of the
finite element code FEAP (66). In particular, a four-node interface element
is developed, integrating the evolutive equations through a step by step al-
gorithm. The time step is solved at a typical point of the interface, adopting
a predictor-corrector algorithm through a displacement-based formulation.
In order to ensure the quadratic rate of asymptotic convergence that char-
acterizes the use of the Newton’s iterative method at the global level in
the FEAP code, the consistent tangent interface stiffness matrix is com-
puted and implemented in the program, accordingly to the local integration
algorithmic scheme.
In particular, two structural applications are performed; the first appli-
cation concerns the study of a masonry arch, while the second one deals
with the analysis of a masonry panel.
Analysis of a masonry arch In this section, numerical analyses concern-

ing a masonry arch are developed. The numerical results are compared with
the experimental ones, recovered from the experimental campaign carried
out by Cancelliere et al. (11).
The geometrical data of the round arch, schematically reported in figure
4, are the following: internal radius r = 456 mm, width w = 255 mm,
250 E. Sacco
Figure 5. Position of the four hinges for the arch.
thickness t = 120 mm, height f = 510 mm, abutment angle Φ = 8◦ and

span da = 900 mm. The arch was built in the experimental laboratory
using 23 standard clay bricks joined by 22 mixed mortars; the conventional
numeration of bricks is clearly shown in figure 4. During the experimental
tests, the arch in addition to its weight was subjected to a point-wise load
applied with an eccentricity df = 140 mm with respect to the keystone;
in particular, the load was applied on the 14-th brick by the action of
a hydraulic jack. The arch was restrained at the springs by rigid steel
elements, in order to avoid horizontal sliding.
The experimental test of the arch showed that (figure 5): the first hinge
arose on the extrados between the mortar number 13 and the clay brick
number 14, the second hinge formed on the intrados between the mortar
number 7 and the clay brick number 8, the third hinge formed between on
the extrados the clay brick number 1 and between the mortar number 1 and
the fourth hinges formed on the intrados between the clay brick number 19
and the mortar number 19.
Finite element analyses are developed in order to reproduce the mechan-
ical response of the arch. In particular, the clay bricks and the mortar joints
are modeled by four nodes quadrilateral elements, while the mortar-brick in-
terface and the brick-support interface by the developed four node interface
elements. Experimental evidences showed that the decohesion phenomenon
between a mortar joint and two adjacent bricks is activated mainly on only
one of the two interfaces, as illustrated in figure 5. For this reason and in
order to simplify the finite element model of the arch, only one nonlinear in-
terface is considered for each mortar bed. Figure 6 illustrates the numerical
model of the arch, emphasizing that the interface elements are positioned
on only one of the two surfaces joining the mortar with the bricks, while a
perfect mortar-brick adhesion is assumed for the other surface.
In the numerical simulations the geometrical cross-section of the arch
was reduced of 6 mm in height and width with respect to the nominal size
Figure 6. Finite element mesh and constraints adopted for the analyses.
of the cross-section; this choice is justified by the presence of defects and

irregularities of construction of the beds mortar, which make smaller the
contact surface between brick and mortar joint.
Linear elastic constitutive laws are adopted for both mortar and bricks
to simulate the behavior of the arch. This choice is due to the observation
that the collapse mechanism for the arch is mainly governed by the unilat-
eral and frictional behavior of the mortar-brick interface, while no material
failure due to compressive stresses has been noted during the experimen-
tal tests, as it was reasonably expected. The mechanical properties of the
brick and mortar materials, adopted for the computations, are determined
by experimental tests described in (11):
Eb = 16000 MPa, νb = 0.2

Em = 1500 MPa, νm = 0.2
where Eb , νb and Em , νm are the elastic modulus and the Poisson ratio of
the brick and the mortar, respectively. The following values of the mechan-
ical properties, derived by the experimental evidences, are adopted for the
interface elements:
252 E. Sacco
Figure 7. Limit load, experimental results and numerical results obtained

considering different meshes for the arch.
0 3
τN = 0.3 MPa, GcN = 0.3 N/mm, KN = 1500 N/mm ,
3
τT0 = 3.0 MPa, GcT = 0.3 N/mm, KT = 750 N/mm ,
μ = 0.5
Three different discretizations are considered for the arch; they are ob-
tained introducing the three discretization parameters nh , nb end nm , rep-
resenting the number of subdivisions along the height of the brick, along
the base of the brick and in the thickness of mortar joint, respectively, as
illustrated in figure 6. In particular, the computations are performed set-
ting nh = 2 and nm = 1, while different values for the parameter nb are
considered, nb = 5, 10 and 20. In the numerical simulations, the arc-length
technique is adopted in order to be able to capture the possible softening
branch of the equilibrium path. In particular, the incremental negative dis-
placement u2 along the x2 −direction (see figure 4) of the loaded node is
chosen as control parameter.
In the graph of figure 7 the results of the numerical simulations, plotted
in term of nodal force F versus the negative displacement u2 of the loaded
node, are compared with the experimental data provided in (11). In same
figure the value of the failure load (650 N) deduced by applying the kine-
Figure 8. Equilibrium configuration for v = −0.1 mm.
matic theorem of the limit analysis is also reported. It can be remarked

that:
• the three numerical models satisfactory approximate the experimental
behavior of the arch and the theoretical limit load;
• increasing the discretization along the base of the clay brick, the nu-
merical result tends to the experimental one;
• the analyses are stable;
• the collapse load deduced by the limit analysis represents the upper
bound of the ultimate strength of the arch.
It can be remarked that finer meshes lead to better evaluations of the
limit load, because of the improvement in the reproduction of the stress
profile at mortar-brick interface. Indeed, the elastic deformation of both
bricks and mortar does not play any role in the determination of the limit
load of the arch, but it can influence only the first part of the response of
the structure.
In figure 9, the finest discretization adopted for the computations is
reported with the corresponding deformed configuration when the vertical
displacement of the loaded node reaches the value u2 = −0.1 mm. It can
be remarked that the numerical simulation is able to predict the collapse
mechanism of the arch.
Raijmakers-Vermeltfoort panel The second analyzed micromechani-

cal problem is the masonry wall loaded in compression and shear studied
by Lourenço (37) on the basis of the experimental tests performed by Raij-
makers and Vermeltfoort (55).
The geometry and loading are shown in figure 9. The initial compression
is obtained by prescribing the vertical displacement of the top of the wall
where a steel beam is positioned until the total sum of the vertical reactions
is equal to Q = 30 kN; such vertical displacements are then kept constant
254 E. Sacco
Figure 9. Masonry panel loaded in compression and shear: geometry and

loading (dimensions in mm).
during the rest of the analysis, in which the horizontal displacement of the
top-right corner, u1 , is incremented left-ward.
Computations are performed modeling the mortar and the adhesion-
contact surface mortar-brick by means of the interface element. Thus, the
numerical model is constituted by bricks, suitably enlarged, directly joined
by interfaces.
Interface elements have even been placed in the middle of each brick
to take into account the possible failure of a brick by allowing a crack to
initiate and propagate vertically on its middle. A regular finite-element
discretization is adopted for the computations. In particular, each half
brick is discretized with 2 × 2 4-noded, plane stress elements with enhanced
strains.
The material properties of the brick, adopted for the computations, are:
Eb = 17600 MPa, νb = 0.15
while the following values of the mechanical properties are adopted for the
interface elements:
• mortar joint
0 3
τN = 0.25 MPa, GcN = 0.018 N/mm, KN = 82 N/mm ,
3
τT0 = 0.35 MPa, GcT = 0.125 N/mm, KT = 36 N/mm ,
μ = 0.75
• brick-brick interface
0 3
τN = 2.00 MPa, GcN = 0.080 N/mm, KN = 1E6 N/mm ,
3
τT0 = 4.00 MPa, GcT = 0.200 N/mm, KT = 1E6 N/mm ,
μ = 0.75
In figure 10, the numerically computed horizontal reaction F is plotted

versus the horizontal prescribed displacement u1 of the top-right corner,
and compared with the experimental data. The numerical and experimental
results appear in good agreement. Computations are performed combining
the local-control arc-length method with the displacement control. The
special numerical procedure adopted, described in details in (5), is able to
determine the complex load-displacement equilibrium curve.
Figure 11(a) reports the contour plot of the shear stress component on
the deformed configuration, when the prescribed displacement is equal to
4.24 mm, with a magnification factor of 20, showing the corresponding crack
pattern. Figure 11(b) shows the experimentally found crack pattern in two
tested walls (37); it can be remarked the good agreement of the crack path
determined numerically with the ones obtained from experimental tests.
3 Multiscale modeling
The masonry is a composite, material; for this reason, in the last twenty
years, several models derived from homogenization approaches have been
proposed in the literature. Different assumptions have been considered in
the development of the models.
The differences among the models concern: the arrangement of the ma-
sonry, i.e. regular or irregular texture; the constitutive model adopted for
the brick, i.e. rigid, deformable with linear or nonlinear response, including
damage and/or plasticity effects; the model adopted for the mortar, i.e. in-
terface or continuum material characterized by linear or nonlinear response;
the macroscopic model obtained by homogenization, i.e. Cauchy, Cosserat
or higher order continua.
Moreover, various homogenization techniques have been proposed and
applied for the masonry material. Thus, selecting among the different pos-
sible mechanical assumptions and adopting different homogenization tech-
256 E. Sacco
Figure 10. Load-displacement curve: comparison between the numerical

and the experimental results reported in Reference (37).
niques, many masonry models and numerical procedures have been proposed
in literature.
Among the others, Cauchy models have been derived for the partic-
ular microstructural geometry of masonry material, developing simplified
homogenization techniques by Kralj et al. (33), by Pietruszczak and Niu
(54) and by Gambarotta and Lagomarsino (27; 28), considering the nonlin-
ear behavior of the constituents. Periodic homogenization techniques have
been proposed by Anthoine (6) and by Luciano and Sacco (41), considering
the elastic behavior of both brick and mortar. Pegon and Anthoine (52) de-
rived the in-plane macroscopic nonlinear behavior of masonry considering a
damage model for the brick and the mortar in the framework of the plane
stress and generalized plane strain states. The effect of the head joints in
the overall nonlinear response of the masonry has been remarked. Luciano
and Sacco (40) derived a damage model for masonry material character-
ized by periodic micro-structure. As for old masonry the strength of the
mortar is much lower than the strength of the bricks, it has been assumed
that damage can develop only in the mortar material. Uva and Salerno
Figure 11. Contour-plot and deformed shape at u1 = 4.24 mm (a) and

crack pattern reported in Reference (37) (b).
(67) presented the homogenized response of the masonry accounting for the
damage development within the bed joints and the frictional dissipation
over cracks’ faces. Multiscale FE (finite element) analyses of large-scale
panels have been shown. Massart et al. (46) proposed an enhanced mul-
tiscale model using nonlocal implicit gradient isotropic damage models for
both the constituents, describing the damage preferential orientations and
employing at the macroscopic scale an embedded band model. Zucchini and
Lourenço (70) proposed an improved micromechanical model for masonry
homogenization in the nonlinear domain, incorporating suitably chosen de-
formation mechanisms coupled with damage and plasticity models. Wei
and Hao (68) developed a continuum damage model for masonry account-
258 E. Sacco
ing for the strain rate effect, using a homogenization theory implemented
in a numerical algorithm. Sacco (60) presented a nonlinear homogeniza-
tion procedure for the masonry, considering a linear elastic behavior for the
blocks and a nonlinear constitutive law for the mortar joints. In particu-
lar, the mortar constitutive law accounted for the coupling of the damage
and friction phenomena occurring in the mortar joints during the loading
history.
In the framework of the Cosserat continuum models, Masiani et al. (44)
and Masiani and Trovalusci (45) studied the case of two-dimensional peri-
odic rigid block assemblies joined by elastic mortar, deducing the macro-
scopic characterization of the equivalent medium by equating the virtual
power of the coarse model with the virtual power of the internal actions
of the discrete fine model. Casolo (12) considered isotropic linear elastic
models both for the brick and the mortar and used a computational ap-
proach to identify the homogenized elastic tensor of the equivalent Cosserat
medium. Salerno and de Felice (61) investigated the accuracy of various
identification schemes for Cauchy and Cosserat continua, showing that in
the case of non-periodic deformation states micropolar continuum better
reproduces the discrete solutions, due to its capability to take scale effects
into account. Addessi et al. (3) developed a two scales model, characterized
by the Cosserat micropolar continuum at the macro level and the classical
Cauchy medium at the micro level. Nonlinear response of the mortar joints
was considered by using a cohesive-friction constitutive model. The nonlin-
ear behavior for both brick and mortar has been considered by De Bellis
and Addessi (17). A full multiscale model for the structural analysis of
the in-plane response of masonry panels has been presented by Addessi and
Sacco (2).
Micro-macro, i.e. multiscale, analysis based on nonlocal Cauchy-Cauchy
modeling has been presented by Marfia and Sacco (42), where an effective
homogenization procedure is developed and implemented at Gauss point
level, in 2D plane state finite element. A nonlocal integral model is adopted
in order to overcome problems due to the localization of strain and damage.
An interesting numerical procedure, based on a multilevel computational
approach, has been proposed by Brasile et al. (9), (10) for the static and
dynamic multiscale analysis of masonry walls.
Most of the existing models for masonry concern periodic microstruc-
tures. A non-periodic masonry, typical of historical buildings, has been
analyzed in (14) by means of a perturbation approach, while the evaluation
of the strength domain for non-periodic masonry using a random media
micromechanical approach is discussed in (13).
The out-of-plane analysis of masonry panels is a very important and
interesting issue. In fact, recent earthquakes show that the out-of-plane

failure of masonry walls is the main responsible of the loss of human lives.
Mercatoris and Massart (47) presented a multiscale framework for the failure
of periodic quasi-brittle thin planar shells, using a shear-enhanced element
with the Reissner - Mindlin description and employing it for the failure of
out-of-plane loaded masonry walls.
When softening behavior is considered for the masonry constituents,
the classical Cauchy model leads to the well-known localization and mesh-
dependency problems. On the contrary, the Cosserat continuum model
accounts for a material length scale into the constitutive description, pro-
viding a natural way to obtain a dependence of the overall response of the
composite material on the absolute size of the constituents and to achieve
a realistic description of a micro-structurally triggered macroscopic local-
ization. In fact, the Cosserat model is able to overcome the localization
problems for shear loaded panels.
In this section, the two scales model, characterized by the Cosserat mi-
cropolar continuum at the macro level and the classical Cauchy medium at
the micro level (3; 2), is reviewed and illustrated. The model is presented
together with numerical applications.
In the following, quantities identified in the macro and in the micro scales
are denoted with capital and small letters, respectively.
3.1 Macro-level model

At the macro-level, where the Cosserat continuum model is employed,
T
the displacement vector U = {U1 U2 Φ} contains three independent kine-
matic fields, representing the translations U1 and U2 and the rotation Φ,
T
at each point X = {X1 X2 } of the body volume. The compatibility equa-
tions, relating the deformation components to the displacement fields, are
introduced in the compact form of Voigt notation as:
⎡ ∂
⎤
∂X1 0 0
⎢ 0 ∂
0 ⎥
⎢ ∂X2 ⎥
⎢ ∂ ∂
0 ⎥
⎢ ∂X2 ⎥
E = DU with D = ⎢ ∂X1
⎥ (15)
⎢ − ∂X∂ ∂
−2 ⎥
⎢ 2 ∂X1 ⎥
⎣ 0 0 ∂
∂X1
⎦
∂
0 0 ∂X2
T
where E = E1 E2 Γ12 Θ K1 K2 is the strain vector and D
the compatibility operator. The first three components of E are the in-
plane Cauchy extensional and symmetric shear strains, Θ = 2(W − Φ) is
the rotational deformation, representing two times the difference between
260 E. Sacco
the rigid rotation

1 ∂U2 ∂U1
W = −
2 ∂X1 ∂X2
and the Cosserat rotation field Φ, and K1 and K2 are the curvatures.
Accordingly, the stress vector is expressed as
T
Σ= Σ1 Σ2 Σ12 Z M1 M2 ,
where Σ1 and Σ2 are the normal stresses, Σ12 is the symmetric shear stress,
Z is the stress component associated to the rotational deformation Θ and
M1 and M2 are the couples.
In order to derive the constitutive response in each macroscopic material
point a homogenization procedure is applied by means of the analysis of the
corresponding RVE (representative volume element). In particular, for each
RVE a boundary value problem is solved and the constitutive relationship
is determined, deriving the macroscopic elastic 6 × 6 constitutive matrix C.
3.2 Micro-level model

At the micromechanical level, the compatibility, equilibrium and consti-
tutive equations governing the behavior of the masonry RVE are introduced
in the 2D plane strain framework. The analysis is limited to the case of
periodic masonry made of regular textures of bricks and mortar joints.
The displacement vector u = {u1 u2 }T at each point x = {x1 x2 }T of the
RVE occupying the volume ω is introduced in a representation form as the
superposition of a prescribed field ū, defined as function the macroscopic
deformation E, and an unknown periodic fluctuation ũ, satisfying proper
periodicity conditions on the RVE boundary (65; 41), namely:
u = ū (x) + ũ(x) in ω (16)
The strain ε = {ε1 ε2 γ12 }T is obtained from the compatibility equation as:
⎡ ∂ ⎤
∂x1 0
ε = du in ω with d=⎣ 0 ∂
∂x2
⎦ (17)
∂ ∂
∂x2 ∂x1
and according to equation (16) ε is expressed as:
ε = ε̄ (x) + ε̃ (x) (18)
being ε̄ and ε̃ the strain fields compatible with ū and ũ, respectively.
Linear elastic stress-strain relationship is adopted for the brick:
σ B = CB ε (19)

B T
where CB is the elastic matrix and σ B = σ1B , σ2B , τ12 is the stress
vector in the brick.
Regarding the mortar material a coupled damage-plasticity constitutive
model is adopted. Indeed, the stress-strain relationship can be considered
as the extension to the case of continuum material of the constitutive law
presented in the previous section for the interface. It is able to conjugate
the damage due to the evolution of the micro-cracks and the plasticity due
to the unilateral and friction effects. Herein, the model is enriched intro-
ducing a limit compressive stress in order to take into account the crushing
mechanism of the masonry. Indeed, the crushing of the masonry panel is
generally due to the failure of the bricks; in order to work with a simple
model, leaving a linear elastic stress-strain relationship for the bricks, the
compressive crushing is introduced in the mortar constitutive law.
The proposed nonlinear model for the masonry, described in detail in
(58; 2), can be obtained by a suitable modification of the interface model
presented in section 2. An additive decomposition of the stress vector σ M
at a typical point of the mortar is assumed as follows:
σ M = (1 − D) σ u + Dσ d (20)
with D the scalar damage parameter. The two stress vectors σ u and σ d are
given by the relationships:
σ u = CM (ε − εκ ) , σ d = CM (ε − εp − εκ ) (21)
where: ⎡ ⎤
CTMT M
CN T 0
CM = ⎣ CN
M
T
M
CN N 0 ⎦ (22)
0 0 GM
represents the elasticity matrix of the mortar, εp is the inelastic strain vector
accounting for the possible unilateral opening effect and for the friction
sliding, while εκ is the plastic strain due to the crushing.
Taking into account the constitutive equations (21), the relationship (20)
becomes:
σ M = CM (ε − π) (23)
where π is the vector collecting all the inelastic strains:
⎧ ⎫ ⎧ ⎫ ⎧ ⎫
⎨ πT ⎬ ⎨ 0 ⎬ ⎨ H (εN − εκN ) εT ⎬
π= πN = εκN +D H (εN − εκN ) (εN − εκN ) (24)
⎩ ⎭ ⎩ ⎭ ⎩ p ⎭
πN T 0 γN T
262 E. Sacco
accounting for the crushing, εκN , the damage, D, the unilateral contact by
means of the component H (εN − εκN ) (εN − εκN ) and the slip by means of
p
T ; H (εN − εN ) is the Heaviside function, which assumes
κ
the component γN
the following values: H (εN − εN ) = 0 if (εN − εκN ) ≤ 0 and H (εN − εκN ) =
κ
1 if (εN − εκN ) > 0. Because of the simplified form of the inelastic strain (24),
the constitutive law (21) is able to provide zero normal stress in transversal
d
direction, σN = 0, as well as in longitudinal direction, σTd = 0, when opening
of the mortar joint occurs.
The crushing and the friction effects are modeled as classical plasticity
problems. The evolution law of the crushing inelastic strain component εκN
is stated as:
ε̇κN = −λ̇κ (25)
and it is ruled by the Kuhn-Tucker conditions, being λκ the inelastic mul-
tiplier:
λ̇κ ≥ 0 ψ (σ) ≤ 0, λ̇κ ψ (σ) = 0 (26)
The yield function is assumed as:
ψ (σ) = − (σN + σy + κξ) (27)
where σy is the compressive yield threshold, κ ≤ 0 is the softening parameter

and ξ is the accumulated plastic strain evaluated as:
t
ξ= |ε̇κN | dτ (28)
0
As concerning the friction mechanism, the evolution of the inelastic slip

p
strain component γN T is governed by the Coulomb yield function:

d
ϕ σ d = μ σNd
+ τN T
(29)
where μ is the friction parameter. A non-associated flow rule is considered

as:
τd
p
γ̇N NT
T = λ̇μ d (30)
τ NT
with the following loading-unloading Kuhn-Tucker conditions:

λ̇μ ≥ 0 ϕ σ d ≤ 0, λ̇μ ϕ σ d = 0 (31)
where λμ is the inelastic multiplier.

A model which accounts for the coupling of mode I and mode II of
fracture is considered for the evolution of the damage parameter D. The
two quantities ηN and ηN T , which depend on the first cracking strains ε0N
0 0 0
and γN T , on the peak values of the stresses σN and τN T and on the fracture
energies GcN and GcN T , respectively, are introduced in the form:
ε0N σN
0 0
γN 0
T τN T
ηN = , ηN T = (32)
2 GcN 2 GcN T
The equivalent strain measures YN and YN T are defined as:
2 2
YN = (εN − εκN ) , YN T = (γN T ) (33)
where the bracket operator • gives the positive part of the quantity •.
Then, the strain ratios are determined as:
(
1 YN YN T
η = 2 (YN ηN + YN T ηN T ) , β = 2 + 2 , α = YN + YN T
α εN,0 γN T,0
(34)
Finally, the damage is evaluated according to the following law:

β−1
D = max 0, min 1, (35)
history (1 − η) β
3.3 Homogenization technique

Different homogenization procedures have been proposed in the litera-
ture in order to evaluate the nonlinear behavior of composites. An interest-
ing approach is the Transformation Field Analysis (TFA) proposed initially
by Dvorak (21) and adopted and extended to plasticity and viscoplasticity
problems by Fish and Shek (23). According to TFA approach, the inelas-
tic strain, i.e. the transformation strain, is assumed to be uniform within
each individual phase in the composite. Dvorak et al. (22) and Chaboche
et al. (15) improved the TFA for deriving the nonlinear behavior of dam-
aging composites, subdividing each phase into subdomains, at the expense
of increasing the complexity of the model. Michel and Suquet (48) pro-
posed a nonuniform TFA, where the inelastic strain field is considered as
nonuniform and described as the superposition of functions, called inelas-
tic modes and determined numerically by simulating the response of the
composite along monotone loading paths. The technique was then modified
and implemented for the analysis of the morphological anisotropy of micro-
heterogeneous materials with particle reinforcement by Fritzen and Böhlke
(26). A new approach for the nonuniform TFA has been recently proposed
by Sepe et al. (62).
The use of the TFA requires the computation of localization and transfor-
mation tensors. To this end, numerical techniques based on generalizations
264 E. Sacco
Figure 12. Selected RVE for the running bond masonry texture (a); num-
bering of the mortar joints (b).
of the Eshelby method (50) can be quite successfully adopted for compos-
ites characterized by random microstructure. When the geometry of the
microstructure is complex, as in the case of masonry material, numerical
techniques, most often based on the assumption of microstructure periodic-
ity, can be adopted. In fact, finite element methods or fast Fourier transform
solutions are able to describe accurately local stress and strain fields, so that
the correct nonlinear behavior of the phases can be determined.
A RVE is selected; it is characterized by rectangular shape with di-
mensions 2a1 and 2a2 , parallel to the coordinate axes x1 and x2 , mortar
thickness denoted by s and brick sizes by b and h, as shown in figure 12(a).
The RVE accounts for all the geometrical and constitutive properties of the
masonry components.
Akinematic map linking the macro-and micro-level is established. Herein,
following the methodology proposed by Forest and Sab (24), third order
polynomial expansions are assumed for the assigned part of the microscopic
displacement ū(x), which allows to take into account all the macroscopic
Cosserat deformation components. In the case of a rectangular cell, the fol-
lowing form of the assigned displacement ū is adopted in compact notation:
ū = A (x) E (36)
where
/ 1
0
x1 0 2 x2 −a(x32 − 3ρ2 x21 x2 ) −x1 x2 − 12 x22
A= 1 1 2 (37)
0 x2 2 x1 −ρ2 a(ρ2 x31 − 3x1 x22 ) 2 x1 x1 x2
with
5 a21 + a22 a2
a= ρ= (38)
4 a41 a1
In order to activate the Cauchy deformation modes independently from the
Cosserat ones, in equation (36) the fourth component of the Cosserat strain
vector E is redefined as:
1 ρ2 − 1
Θ̂ = Θ + Γ12 (39)
2 ρ2 + 1
The stress variable conjugated to Θ̂ is denoted with Ẑ.

The strain vector at the micro-level, equation (18), can now be expressed
as:
ε = B(x)E + ε̃(x) (40)
where ε(x) is the periodic strain, satisfying null average condition in the
RVE volume ω, and:
⎡ ⎤
1 0 0 6 a ρ2 x1 x2 −x2 0
B=⎣ 0 1 0 2

2 6 a ρ
x12x2 2 2 0 x1 ⎦ (41)
0 0 1 3 a ρ − 1 x 2 − ρ x1 0 0
The in-plane periodicity and continuity conditions lead to the following

boundary conditions:
ũ(a1 , x2 ) = ũ(−a1 , x2 ) ∀x2 ∈ [−a2 , a2 ]

(42)
ũ(x1 , a2 ) = ũ(x1 , −a2 ) ∀x1 ∈ [−a1 , a1 ]
The masonry RVE is subjected to the macroscopic Cosserat strain E

applied to the whole RVE and to the inelastic strain π i , with i = 1, . . . , m,
applied to each of the m mortar joints illustrated in figure 12(b). Note that
the inelastic strain is assumed to be uniform in each mortar joint.
The micromechanical problem of the RVE subjected to the prescribed
value of the six components of E and to the three components of the inelastic
strains π i in the m mortar joints, have to be solved. The micromechanical
strain field, resulting after solving the boundary value problem for the RVE
subjected to E, can be written in the following representation form:
e = Re (x) E (43)
266 E. Sacco
where the localization matrix Re (x) is introduced. Consequently, the av-

erage strain in the brick and in each mortar joint M i results as:
i i
ēB = R̄B
e E, ēM = R̄M
e E (44)
B Mi
where R̄e and R̄e denote the average localization matrices in the brick
volume and in the mortar joint volume, respectively.
The homogenized Cosserat stress in the whole RVE volume ω is obtained
by applying the generalized Hill-Mandel principle, resulting:
Σe = C E (45)
where the overall elastic constitutive matrix is defined as (3):

⎡ ⎤
m
1 j
C = ⎣ RTe CB e dω + RTe CM e dω ⎦ (46)
ω j=1
B Mj
Moreover, the average stress in the mortar joint M i may be evaluated as

i i i
σMe = CM ēM = CM R̄M e E, as well as in the brick σ̄ e = C ē
B B B
=
B B
C R̄e E.
Similarly, after solving the micromechanical problem of the RVE sub-
jected to an inelastic strain π i prescribed in the mortar joint M i , the re-
sulting local strain field is expressed in the form:
qi = Riπ (x) π i (47)
being Riπ (x) the associated localization matrix. Note that the local strain
field qi is characterized by null average. The elastic strain in the mortar
joint M j results:
j j
j

η i,M = qi,M − δij π i = RM π i − δ ij I πi (no sum) (48)
j j
where qi,M and RM j
π i are the restriction to the mortar M of the fields q
i
and Rπi , respectively. The elastic strain η i,B

in the brick coincides with
the total strain.
The corresponding overall Cosserat stress can be obtained again by ap-
plying the generalized Hill-Mandel principle in the form:
⎡ ⎤
m
1 ⎣
B T B B j T j j
⎦ i
Σπ i = Rπi C Rπi dω + RMπi CM R M π i − δij I dω π
ω j=1
B Mj
(49)
When the RVE is subjected to the overall elastic strain E and to the
inelastic strains π i , i = 1, 2, . . . , m, the superposition of the effects can be
performed. In such a way, it is possible to compute the overall stress:

m
m
Σ = Σe + Σπi = CE + Σπ i (50)
i=1 i=1
and the average stress in the m mortar joints:

j
j
σ M = CM ε M − π j (51)
The nonlinear behavior of the RVE is assumed to depend on the average

stresses and strains evaluated in each of the m mortar joints.

The FE method is used to solve the macro and micromechanical prob-
lems. In particular, the multiscale solution procedure is implemented in a
research version of the finite element code FEAP (69).
The proposed homogenization procedure initially requires the evalua-
tion of the overall constitutive matrix C and of the localization matrices
j j
R̄M
e and R̄M
π i . They are computed by solving 6 + 3m FE linear elastic
micromechanical problems at the RVE level.
A standard Cauchy 4-node quadrilateral FE is adopted to model the
RVE with two translational degrees of freedom at each node. At the macro-
level a 2D 4-node quadrilateral FE is implemented to discretize the Cosserat
medium, with three degrees of freedom at each node, two translational and
one rotational. A step-by-step solution technique based on the classical
backward-Euler algorithm (64) is adopted for the time integration of the
governing equations. The nonlinear problem at each time step is solved by
means of the Newton-Raphson technique.
The multiscale procedure is employed to analyze the structural response
of two masonry panels for which experimental results are available.
Masonry panel under vertical load The masonry panel acting as a

deep beam shown in figure 13 was experimentally investigated by Milani et
al. (49).
The dimensions of the panel are: width B = 290 mm, eight H = 270 mm,
thickness T = 30 mm. In the experimental tests the panel was restrained
at the bottom side with two steel rollers and the vertical external load was
applied through steel plates located on the top side of the panel as shown
in figure 13. The geometrical parameters of bricks and mortar are the
268 E. Sacco
Figure 13. Masonry panel subjected to a vertical load: geometry and

boundary conditions.
28
Figure 14. Force-displacement equilibrium path for the masonry panel

subjected to a vertical load.
following: size of the brick b = 56 mm, h = 15 mm; thickness of the mortar

joints s = 2 mm. Furthermore, the material mechanical parameters are:
• brick
E = 1850 MPa ν = 0.15
• mortar
E = 233 MPa ν = 0.15 εN,0 = 0.0015 γN T,0 = 0.007
GcI = 0.0013 MPa GcII = 0.0013 MPa σy = 5 MPa κ=0
μ = 0.1
where E and ν are Young’s and Poisson moduli, respectively. Due to the
symmetry of the problem, only one half of the panel is analyzed and a mesh
with 108 FEs is adopted. The three nodes of the right-top side, close to
the symmetry axis, are subjected to the same vertical displacement. The
support steel rollers are discretized by means of 4-node quadrilateral FEs,
assuming a linear elastic constitutive law with Young’s modulus equal to
210000 MPa and Poisson modulus 0.3.
In figure 14 the global response curve of the panel is shown, i.e. the
global vertical reaction computed at the restrained node of the bottom side
versus the top applied displacement is depicted. Three curves are reported;
they refer to the numerically obtained results, obtained with two different
270 E. Sacco
Figure 15. Load-displacement curve: comparison between the numerical

and the experimental results reported in Reference (37).
meshes (solid line and dashed line), and to the experimental ones (line with
diamond symbols). As it can be noted the numerical curves are in a very
good agreement with the experimental one.
Raijmakers-Vermeltfoort panel The masonry wall shown in figure 9,

yet analyzed adopting a micromechanical model, is studied performing the
multiscale procedure. The material mechanical parameters adopted for the
computations are:
• brick
E = 16700 MPa ν = 0.15
• mortar
E = 798 MPa ν = 0.11 εN,0 = 0.0003 γN T,0 = 0.001
GcI = 0.00179 MPa GcII = 0.0126 MPa σy = 5 MPa κ=0
μ = 0.75
Three different meshes are considered with 15 × 15, 20 × 20 and 30 × 30 FEs.
In figure 15 the global response curve of the wall is reported. In par-
ticular, the global horizontal reaction computed at the bottom side versus
the top applied displacement is depicted. Four different curves are shown
Figure 16. Damage distributions in horizontal (first line) and vertical joints
(second line) evaluated at different values of the applied horizontal displace-
ment.
referring to the numerically obtained results and the experimental ones (line
with diamond symbols). It is evident that the numerical curves match very
well the experimental one. Furthermore, the dashed line curve referring to
the 15 × 15 mesh is not a perfectly converged solution, while the solid and
dash-dot line curves referring to the finer mesh are indistinguishable, so
proving the capability of the Cosserat model adopted at the macro-level to
lead to mesh-independent FE results. After the initial linear elastic behav-
ior, the nonlinear mechanisms are activated and the global response curve
reaches a peak load of about 48 N, a little lower than the experimental one,
at the applied displacement value of 2.5 mm. Then, the global response
curve shows a softening trend.
272 E. Sacco
In figure 16 the map of the macroscopic damage variable is reported for

the horizontal bed joints (first line) and the vertical head joints (second line),
respectively, at two values of the top applied displacement, U1 = 1 mm and
U1 = 3 mm. By observing the first line, it can be seen that damage starts
in the bed joints located at the bottom left and top right of the wall, where
it experiences the most severe bending deformation state. After, damage
spreads involving also the bed joints located near the center of the wall,
where the shear mechanisms are predominant. As for the damage in the
vertical joints, it is mainly concentrated into two vertical bands. The map of
the damage distributions can be compared with the crack pattern obtained
during the experimental tests and illustrated in figure 11(b); the satisfactory
agreement between the fracture path obtained from the experiments and the
damage map recovered using the multiscale masonry model can be noted.
4 Macromechanical modeling
Masonry is a cohesive material characterized by softening response. Indeed,
his nonlinear behavior is due to the damage and plastic micromechanical
processes. From a microscopic point of view the damage is linked to the
growth and coalescence of microcracks, leading to the formation of macro-
cracks which can induce the collapse of the structure. The plasticity can
simulate the presence of irreversible displacements due, for instance, to the
effects of the friction. Various phenomenological nonlinear models have
been proposed in the literature to describe the softening response of struc-
tural elements made of masonry material. The available models adopted for
structural computations are mainly based on macromechanical approaches
using damage mechanics (34) and plasticity theory (39).
Among the others, Lotfi and Shing (35) developed smeared crack finite
element analyses of masonry structures in order to assess the capability of
this approach in capturing the strength and various failure mechanisms of
masonry shear walls. They compared the numerical results with experimen-
tal data, investigating the objectivity of numerical results with respect to
mesh size. Lourenço (37) presented an anisotropic continuum model based
on multisurface plasticity, considering a Rankine type yield surface for ten-
sion and a Hill type yield surface for compression. The proposed model is
completed with a computational algorithm which is used to perform compar-
isons between numerical and experimental results, available in the literature.
Berto et al. (8) proposed an orthotropic damage model for the analysis of
brittle masonry subjected to in-plane loading. The material model is gov-
erned by four independent internal damage parameters, one in compression
and one in tension for each of the two natural axes of the masonry, and
it is able to account for the stiffness recovery at crack closure. Compari-

son between numerical results and experimental data are reported both for
masonry shear panels and for a large-scale masonry wall. Dhanasekar and
Haider (19) presented an explicit finite element analysis of masonry shear
walls containing reinforcements. Ramalho et al. (56) developed a nonlin-
ear finite element model to simulate the nonlinear response of a masonry
construction, considering a nonlinear constitutive law governed by a dam-
age tensor, which allows the damage-induced anisotropy accompanying the
cracking process. The results of the numerical computations concerning
the analysis of a three-leaf masonry specimens was put in comparison with
the experimental data, remarking the effectiveness of the approach. Reyes
et al. (57) presented a numerical procedure for fracture of brickwork ma-
sonry based on the strong discontinuity approach. The model, which takes
into account the anisotropy of the material, has been implemented into a
commercial code by means of a user subroutine. Comparison between nu-
merical and experimental results have been provided, remarking that the
proposed numerical procedure is able to accurately predict the experimental
mixed-mode fracture for different orientations of the brick layers on masonry
panels. Marques and Lourenço (43) presented a simple design tool based on
structural component models for investigating the seismic assessment of a
two-storey masonry building using pushover analysis. They demonstrated
that macro-modeling can be adequate approaches for the seismic design of
unreinforced masonry buildings, as the tool requires very low computational
resources, allowing easy interpretation of results and provides satisfactory
accuracy. Grande et al. (32) developed a simple beam finite element for
the nonlinear analysis of masonry structures in the context of the “equiv-
alent frame model”. Numerical applications have been presented to assess
the reliability of the proposed approach, performing comparisons between
experimental results available in literature and the numerical predictions.
Pelá et al. (53) presented a numerical model for nonlinear analysis of ma-
sonry structural elements based on orthotropic continuum damage mechan-
ics. The capabilities of the model for representing the behavior of different
types of brickwork masonry has been illustrated by means of simulations of
experimental tests.
In the following, a phenomenological model for studying the mechan-
ical response of masonry structural elements is presented. In particular,
the plastic nonlocal damage model, resulting effective and computationally
efficient, proposed in (1) is presented.
In order to take into account the mechanical properties degradation, a
damage scalar variable is introduced in the constitutive law of the material.
The damage is isotropic, i.e. the damage variable has the same influence on
274 E. Sacco
all the stiffness matrix components.

As remarked in section 3, the softening constitutive relationships for co-
hesive materials lead to strain localization and produce mesh dependent
results in finite element analyses. The nonlocal models are based on the
introduction of a characteristic length that determines the size of the neigh-
borhood of a typical point of the body which affects the nonlocal constitu-
tive equation. A quite wide review of the nonlocal approaches can be found
in (18). They can be based on integral models, gradient models and rate
dependent damage models.
Herein, a regularization technique, based on the introduction of the dam-
age Laplacian in the damage limit function, is adopted in order to account
for the nonlocal effects in the material response.
4.1 Plastic-nonlocal damage model

Let the displacement, the strain and the stress fields defined in the two-
T T
dimensional body Ω, be denoted as u = {u1 u2 } , ε = {ε1 ε2 γ12 } and
T
σ = {σ1 σ2 τ12 } , respectively.
Next, a plastic-damage model, based on the following assumptions, is
proposed:
• the damage evolution process is controlled by the elastic strain εe
in tension, since experimental tests demonstrate that plastic response
under tensile loading is negligible for cohesive materials, like masonry;
• the damage evolution process is controlled by the total strain ε in com-
pression; in fact, plasticity plays a fundamental role in the material
behavior in compression;
• the plastic deformation flow is governed by the effective stress σ ) , de-
fined in the following.
The adopted stress-strain law is:
2
σ = (1 − D) C εe (52)
2
= (1 − D) C (ε − ε ) p
where εp is the plastic deformation, C is the elastic isotropic constitutive

matrix and D is the damage variable.
Damage model The damage variable D is defined as the amount of mi-

crostructural defects in a typical section of the representative volume ele-
ment and hence it represents a measure of the material degradation. For the
isotropic damage model, as considered in the following, the variable D is a
function only of the material point and it does not depend on the particular
direction of the normal section.
From the relation between the stress and the strain (52), the effective
) deduced by the complementary energy equivalence
constitutive matrix C,
principle proposed by Sideroff (63), is defined as:
) = (1 − D)2 C
C (53)
It can be pointed out that, since the damage variable D has an equal in-
fluence on all the components of the isotropic matrix C, the relation (53)
define an isotropic constitutive matrix during the whole damage process.
The effective stress is defined as:
σ
)=
σ 2 (54)
(1 − D)
The proposed model, able to describe the damage processes both in
tension and in compression by means of a unique damage scalar variable
D, is derived introducing a proper definition of the equivalent tensile and
compressive deformations Yt and Yc as:

Yt = < ee1 >2 + < ee2 >2 Yc = < −e1 >2 + < −e2 >2 (55)
The equivalent elastic and total strains eei and ei are defined as:
eei = (1 − 2ν)ε̂ei + ν(ε̂e1 + ε̂e2 ) ei = (1 − 2ν)ε̂i + ν(ε̂1 + ε̂2 ) (56)
with ε̂ei and ε̂i the elastic and total principal strains.
Then, the variable Ỹ is defined as:
Yt Yc
Ỹ = + 0 (57)
Yt0 Yc
where Yt0 > 0 and Yc0 > 0 are the initial damage thresholds in tension and in
compression, respectively. Hence, the quantity Ỹ , defined by formula (57),
represents an equivalent deformation, function of the principal positive and
negative strains.
A suitable choice for the damage limit function is proposed:

F = Ỹ − 1 − ãỸ + K̃ D + h∗ ∇2 D (58)
where ∇2 (•) = ∂ 2 (•)/∂x21 +∂ 2 (•)/∂x22 is the Laplacian operator. The model

constants K̃ and ã are defined on the basis of the material characteristic
parameters Kt , Kc , at and ac as:
ã = αt at + α c ac
(59)
K̃ = αt Kt /Y0t + αc Kc /Y0c
276 E. Sacco
Figure 17. Evolution of the damage yield function in the strain (a) and
stress plane (b).
where Kt/c ≥ 0 and 0 ≤ at/c ≤ 1 control the damage rate growth and the
softening branch slope, respectively. The quantity αt/c is defined as:
ηt ηc
αt = , αc = ⇒ αt + α c = 1 (60)
ηt + ηc ηt + ηc
with
Yt Yce
ηt = , ηc = (61)
Y0t + (at Yt + Kt ) D Y0c + (ac Yce + Kc ) D
where
Yce = < −ee1 >2 + < −ee2 >2 (62)
In the figure 17(a) the damage function is schematically represented in
the principal strain plane ε̂1 − ε̂2 and in the principal stress plane σ̂1 − σ̂2 ,
for different values of the damage variable. Figure 17(b) shows the stress
hardening and softening phases occurring during the damage evolution.
The nonlocal parameter h∗ , linked to the characteristic length of the
material, controls the size of the localization region. The determinations of
the dimension of the internal length is an important and interesting issue.
A procedure for the definition of h∗ as function of is discussed in (1).
In the proposed model, the value of h∗ is set as:

αt αc
h∗ = h + (63)
Y0t Y0c
where h is the constant nonlocal parameter used for the pure tensile and
compressive states. Thus, h∗ depends on the equivalent deformation Y . By
substituting equations (59) and (63) into relation (58), the limit function
takes the explicit form:

F = Y t + Yc − 1
Y0t Y0c
− (αt at + αc ac ) YYt + YYc + αt Y K t + α Kc
cY (64)
0t 0c 0t 0c
α α
+ h Y t + Y c ] ∇2 D
0t 0c
The damage evolution is ruled by the consistency condition:

Ḟ Ḋ = 0 (65)
together with the classical Kuhn-Tucker conditions:
Ḋ ≥ 0 F ≤ 0 F Ḋ = 0 (66)
Equations (66) translates in mathematical form the mechanical condition
of the irreversibility of the damage process. According to formulas (65)
and (66) the points inside F represent the material states without damage
evolution (Ḋ = 0), while the points on the boundary of F represent pos-
sible damage evolution states (Ḋ > 0). The damage parameter rate Ḋ is
explicitly determined using the consistency condition (65) in the form:
Ḟ = 0 when Ḋ > 0 (67)
Applying the consistency condition (67) to the nonlocal limit function
(64), the following partial differential equation is obtained:
∂F ∂F ∂F
Ḋ + Ẏ + ∇2 Ḋ = 0 with Ḋ > 0 (68)
∂D ∂Y ∂∇2 D
where, it is:
∂F ∂F ∂Yt ∂F ∂Yc ∂F ∂F
= + = Y0t + Y0c (69)
∂Y ∂Yt ∂Y ∂Yc ∂Y ∂Yt ∂Yc
It can be remarked that the equation (69) contains the term ∇D∂h∗ /∂Y ,
i.e the adopted nonlocal regularization leads to a damage evolution process
governed by the differential equation (68). Hence, the evaluation of the
damage increment requires a resolution of a differential equation.
Note that the proposed damage model is based on a single damage pa-
rameter and, as consequence, crack closure effects cannot accounted for. In
order to take into account the crack closure a two damage variables model,
one in tension and one in compression, has to be formulated.
278 E. Sacco
4.2 Plastic model

A plastic model with isotropic hardening, which accounts for the different
material strength in tension and in compression is introduced. A plastic
limit function FP = FP () ) defined
σ , q), which depends on the effective stress σ
in equation (54) and on the thermodynamic force q, is considered. The force
q is associated to the internal variable α by the relation:
q = −χ α (70)
where χ is the hardening parameter.

The evolution laws are ruled by the equations:
∂FP
ε̇P = λ̇P (71)
∂)
σ
∂FP
α̇ = λ̇P (72)
∂q
FP ≤ 0 λ̇P ≥ 0 FP λ̇P = 0 (73)
where λ̇P is the plastic multiplier rate.

The following yield function FP is considered:
σ , q) = 3J2 + (σc − σt ) I1 − σc σt + q
FP () (74)
where σc and σt are the compressive and tensile yield stresses, respectively,
I1 is the first stress tensor invariant and J2 is the second deviatoric stress
invariant, defined as:
I1 = )1 + σ
σ )2 (75)
1 2
J2 = σ1 − σ
() )2 ) + σ )22 + 6)
)12 + σ 2
τ12
6
4.3 Plastic-damage rate problem

According to the proposed plastic-damage model, the rate problem is
governed by the plastic and damage evolution laws:
2

σ̇= (1 − D) C ε̇−ε̇P − 2 (1 − D)Ḋ C ε − εP (76)
/ 0
∂FP ∂I1 ∂J2
P
ε̇ = λ̇P = λ̇P (σc − σt ) +3
∂)
σ ∂)
σ ∂)
σ
∂FP (77)
α̇ = λ̇P = λ̇P
∂q
FP ≤ 0 λ̇P ≥ 0 FP λ̇P = 0
F ≤ 0 Ḋ ≥ 0
F Ḋ = 0 (78)
∗

2 ∂h
Ḟ = −(aY + K)Ḋ + 1 − aD + ∇ D Ẏ
∂Y
+h∗ ∇2 Ḋ = 0 with Ḋ > 0
As already underlined, a differential equation has to be solved to evaluate

Ḋ, because of the presence of Laplacian in the limit function F . Moreover, it
can be pointed out that the plastic deformation increment can be evaluated
locally, also for the nonlocal damage model.
The rate damage law (78) can be rewritten in weak form (16), resulting
as:
0= Ḟ δ Ḋ dΩ with Ḋ > 0 (79)
ΩD
where δ Ḋ represents the virtual variation of the damage variable and ΩD

represents the part of the whole domain Ω where damage evolution is oc-
curring at the current time, i.e.:

ΩD = X ∈ Ω / Ḋ (X) > 0
with X typical point of the material body Ω.

Integrating by parts equation (79), it results:
/ 0
∂h∗
0 = −(aY + K)Ḋ + 1 − aD + ∇2 D Ẏ δ Ḋ dV (80)
ΩD ∂Y

− h∗ ∇Ḋ · δ∇Ḋ dV + ∇Ḋ · n δ Ḋ dA
ΩD ∂ΩD
The following boundary conditions arise from equation (80):

∇Ḋ · n Ḋ = 0 on ∂ΩD (81)
leading to:
if X ∈ ∂ΩD and X ∈
/ ∂Ω then Ḋ = 0
(82)
if X ∈ ∂ΩD and X ∈ ∂Ω then ∇Ḋ · n = 0
The damage evolution equation (79) is formulated as the stationary con-

dition of a constrained minimization problem of a convex functional (16; 1).
280 E. Sacco
In fact, it can be proved that equation (80) is equivalent to the minimization

problem:
min L Ḋ / Ḋ ≥ 0 (83)
where the functional L is defined as:
: ;
Ḋ2 ∗
h∗
2 ∂h
L Ḋ = (aY + K) − 1 − aD + ∇ D ḊẎ + ∇Ḋ · ∇Ḋ dΩ
Ω 2 ∂Y 2
(84)
Note that, the problem (83) is completed with the further constraint:
0≤D≤1 in Ω (85)
Finally, the solution of the evolution elasto-plastic nonlocal damage
problem is determined solving the plasticity evolution problem (77) and
the minimization problem (83) under the constraint (82) and (85).
In order to solve the evolutionary equations (76), (77) and the mini-
mization problem (83), a predictor-corrector solution procedure based on
the splitting method within a time discretization algorithm, is adopted.
In the predictor phase, the elasto-plastic problem is solved, taking the
damage evolution frozen. On the other hand, in the corrector phase, the
damage field increment is evaluated solving the minimization constrained
problem and considering the total and the plastic deformation as constant.
Hence, the solution algorithm consists in the following two steps:
• an elasto-plastic predictor phase, with a plastic corrector; in this
phase, the consistent elasto-plastic tangent operator is computed,
• a damage evaluation phase; in this phase, the total and plastic strains
are assumed as frosen.
The integration of the evolutionary equations (76), (77) and (83) is per-
formed developing a backward-Euler implicit integration procedure (64).

Some numerical applications are developed to analyze the mechanical
response of two-dimensional structural elements made of masonry material.
The proposed elasto-plastic nonlocal damage model is adopted to re-
produce the constitutive behavior of the masonry material. The following
values for the model parameters are assumed:
• elastic parameters:
E = 1.0 · 104 MPa ν = 0.1
where E and ν are the Young’s modulus and the Poisson ratio, re-
spectively
• damage parameters:
Y0t = 1.0 · 10−5 Y0c = 2.0 · 10−4

Kt = 1.0 · 10−4 Kc = 8.0 · 10−3
at = 0.9 ac = 0.9
• plastic parameters:
σt = 10.0 MPa σc = 10.0 MPa

χ = 1.0 · 106 MPa
The value of the damage parameter h is related to the internal length of the
material. For masonry walls it can be set = 300 mm, which corresponds
about to h = 100 mm2 .
Three masonry panels, schematically reported in figure 18, subjected to
pure shear loading are analyzed. They are characterized by the following
geometrical parameters:
wall1 H = 4000 mm B = 6000 mm

wall2 H = 4000 mm B = 4000 mm
wall3 H = 6000 mm B = 4000 mm
The adopted material properties of the masonry are reported above. Meshes
obtained adopting 4-node elements are considered for the computations. In
particular, wall1 is discretized by 15×20 elements, wall2 by 15×15 elements
and wall3 by 20 × 15 elements.
In figures 19, 20 and 21, the damage and the minimum principal stress
maps for the prescribed displacement u1 = 6 mm are plotted for wall1,
wall2 and wall3, respectively (figure 18). It can be noted that the failure
mechanism is characterized by the formation, growth and propagation of
inclined damage bands, as it typically occurs in masonry panels subjected
to horizontal forces, e.g. seismic loading.
Because of the different geometrical properties of the three walls, the
damaging process in wall1 and wall2 is concentrated in a single band, while
in the widest panel, denoted as wall3, two damage bands appear. Then, as
it can be seen in figures 19(b), 20(b) and 21(b), the resistant part of the
walls consists in compressive inclined trusses carrying the external applied
load. These compressive zones are separated by the damage bands.
In figure 22, the load versus the prescribed displacement is plotted for
wall1 adopting three different discretizations, consisting in 15 × 20 = 300,
20 × 30 = 600 and 30 × 40 = 1200 quadrilateral elements. According to the
adopted discretizations, the minimum and maximum element dimensions
282 E. Sacco
Figure 18. Masonry walls subjected to shear action.
Figure 19. Damage (a) and minimal normal stress distribution (b) for the
wall1.
wall2.
wall3.
are 133.33 mm and 266.67 mm, respectively; since the masonry damage lo-
calization size is taken 600 mm, it results that the mesh size is lower than
the damage length.
It can be noted that the mechanical response of wall1 subjected to shear
loading is characterized by:
• an initial elastic response;
• a damage phase characterized by hardening response;
• a first steep softening branch due to the damage propagation, concen-
trated where the maximum tensile strains occur;
• a hardening phase during which the plastic evolution process becomes
more significant than the damage one;
• a softening branch due to the formation and growth of the damage
band.
284 E. Sacco
Figure 22. Mechanical response of wall1 in terms of base force vs top

displacement for three different discretizations.
It can be remarked that the regularization technique, herein adopted, leads

to numerical results characterized by a satisfactory mesh independence.
In figure 23 the load versus the prescribed displacement is plotted for
wall1, wall2 and wall3. It can be noted that the different geometrical prop-
erties of the three walls lead to different shear strength and compliance.
In particular, wall3 is characterized by a shear strength significantly higher
than the other two walls, while wall1 is the most deformable. Finally, it
can be pointed out that the overall response of wall2 and wall3 is similar to
that of wall1 described above.
5 Conclusions
The modeling of masonry structures is an interesting and important problem
related both to the safety of historical and monumental constructions and
to the safeguard of the human lives of many people who live in masonry
buildings.
Even limiting the interest only to the computational modeling of ma-
Figure 23. Mechanical response of the three walls in terms of base force
vs top displacement.
sonry structures, the approaches illustrated in this chapter are absolutely

not exhaustive. In fact, other approaches can be found in literature; for
instance, one of the most adopted for design intents is the so-called macro-
element approach, which has not been treated herein.
Moreover, for each considered approach only one (specific) model is pre-
sented, while a great variety of models are available within each approach.
For instance, in the framework of the macro-models it is neither discussed
nor presented the no-tension model which can lead to very interesting results
and can be considered very effective in several cases; on the other hand, the
no-tension model is object of other chapters in this book.
The main conclusion of the discussion on the different modeling ap-
proaches for masonry constructions is that all the approaches can satisfac-
tory reproduce the response of specific masonry elements, there is no one
that can be considered in absolute better than the others. On the contrary,
each approach can be more appropriate for specific typologies of masonry
structures. For instance, the micromechanical approach can be satisfactory
adopted to reproduce the response of small size (laboratory) elements or of
286 E. Sacco
structures characterized by big blocks, in which the size of the blocks has
the same order of magnitude of the size of the structural element. On the
contrary, it is less suitable for the analysis of large structures because of
the required great computational effort and for the difficulty in adequately
describing the specific geometry of the whole masonry texture, i.e. the
position of each brick or block in the structure.
As consequence, the most suitable approach for the stress analysis de-
pends on the typology and on the characteristics of masonry elements; the
choice of the most appropriate approach and, then, of the most suitable
model depends on masonry construction, on the specific masonry behavior
and on the knowledge of the material parameters of the material.
Acknowledgments
This chapter contains the results of the researches developed together with
Daniela Addessi, Giulio Alfano, Sonia Marfia and Jessica Toti, who are
deeply and kindly thanked for their nice and fruitful collaboration. Without
their work this chapter would be not written.
Bibliography
[1] D. Addessi, S. Marfia, and E. Sacco. A plastic nonlocal damage model.
Computer Methods in Applied Mechanics and Engineering, 191:1291–
1310, 2002.
[2] D. Addessi and E. Sacco. A multi-scale enriched model for the analysis
of masonry panels. International Journal of Solids and Structures,
49:865–880, 2012.
[3] D. Addessi, E. Sacco, and A. Paolone. Cosserat model for periodic
masonry deduced by nonlinear homogenization. European Journal of
Mechanics - A/Solids, 29:724–737, 2010.
[4] G. Alfano, S. Marfia, and E. Sacco. A cohesive damage-friction interface
model accounting for water pressure on crack propagation. Computer
Methods in Applied Mechanics and Engineering, 196:192–209, 2006.
[5] G. Alfano and E. Sacco. Combining interface damage and friction in a
cohesive-zone model. International Journal for Numerical Methods in
Engineering, 68(5):542–582, 2006.
[6] A. Anthoine. Derivation of the in-plane elastic characteristics of ma-
sonry through homogenization theory. International Journal of Solids
and Structures, 32(2):137–163, 1995.
[7] G. I. Barenblatt. The mathematical theory of equilibrium cracks in the
brittle fracture. Advances in Applied Mechanics, 7:55–129, 1962.
[8] L. Berto, A. Saetta, R. Scotta, and R. Vitaliani. An orthotropic damage

model for masonry structures. International Journal for Numerical
Methods in Engineering, 55:127–157, 2002.
[9] S. Brasile, R. Casciaro, and G. Formica. Multilevel approach for brick
masonry walls part i: A numerical strategy for the nonlinear analysis.
Computer Methods in Applied Mechanics and Engineering, 196:4934–
4951, 2007.
[10] S. Brasile, R. Casciaro, and G. Formica. Multilevel approach for brick
masonry walls part ii: On the use of equivalent continua. Computer
Methods in Applied Mechanics and Engineering, 196:4801–4810, 2007.
[11] I. Cancelliere, M. Imbimbo, and E. Sacco. Numerical and experimental
study of masonry arches. Engineering Structures, 32:776–792, 2010.
[12] S. Casolo. Macroscopic modelling of structured materials: Relationship
between orthotropic Cosserat continuum and rigid elements. Interna-
tional Journal of Solids and Structures, 43(3-4):475–496, 2006.
[13] N. Cavalagli, F. Cluni, and V. Gusella. Strength domain of non-periodic
masonry by homogenization in generalized plane state. European Jour-
nal of Mechanics A/Solids, 30:113–126, 2011.
[14] Antonella Cecchi and Karam Sab. Discrete and continuous models
for in plane loaded random elastic brickwork. European Journal of
Mechanics - A/Solids, 28:610–625, 2009.
[15] J.L. Chaboche, S. Kruch, J.F. Maire, and T. Pottier. Towards a mi-
cromechanics based inelastic and damage modeling of composites. In-
ternational Journal of Plasticity, 17:411–439, 2001.
[16] C. Comi. Computational modelling of gradient-enhanced damage in
quasi-brittle materials. Mech Coh.-Frictional Mater., 4:17–36, 1999.
[17] M.L. De Bellis and D. Addessi. A Cosserat based multi-scale model
for masonry structures. International Journal of Computational Engi-
neering, 9:543–563, 2011.
[18] Francesco Marotti de Sciarra. A nonlocal model with strain-based dam-
age. International Journal of Solids and Structures, 46(22-23):4107 –
4122, 2009.
[19] M. Dhanasekar and W. Haider. Explicit finite element analysis of
lightly reinforced masonry shear walls. Computers and Structures,
86:15–26, 2008.
[20] D.S. Dugdale. Yielding of steel sheets containing slits. J. Mech. Phys.
Solids, 8:100–104, 1960.
[21] G.J. Dvorak. Transformation field analysis of inelastic composite ma-
terials. Proceedings of the Royal Society of London A, 437:311–327,
1992.
288 E. Sacco
[22] G.J. Dvorak and A. Bahei-El-Din. Inelastic composite materials: trans-

formation field analysis and experiments. In P. Sequet, editor, Con-
tinuum Micromechanics, CISM Course and Lecture 377, pages 1 – 59.
Springer, Berlin, 2005.
[23] J. Fish and K. Shek. Multiscale analysis of composite materials and
structures. Composites Science and Technology, 60:2547–2556, 2000.
[24] S. Forest and K. Sab. Cosserat overall modelling of heterogeneous
materials. Mechanics Research Communications in Numerical Methods
in Engineering, 4:449–454, 1998.
[25] F. Fouchal, F. Lebon, and I. Titeux. Contribution to the modelling of
interfaces in masonry construction. Construction and Building Mate-
rials, 23:2428–2441, 2009.
[26] F. Fritzen and T. Bhlke. Nonuniform transformation field analysis
of materials with morphological anisotropy. Composites Science and
Technology, 71:433–442, 2011.
[27] L. Gambarotta and S. Lagomarsino. Damage models for the seismic
response of brick masonry shear walls part i: the mortar joint model
and its application. Earthquake Engineering and Structural Dynamics,
26:423–439, 1997.
[28] L. Gambarotta and S. Lagomarsino. Damage models for the seismic
response of brick masonry shear walls part ii: the continuum model
and its application. Earthquake Engineering and Structural Dynamics,
26:441–462., 1997.
[29] G. Giambanco and L. Di Gati. A cohesive interface model for the
structural mechanics of block masonry. Mechanics Research Commu-
nications, 24(5):503–512, 1997.
[30] G. Giambanco, G. Fileccia Scimemi, and A. Spada. The interphase
finite element. Computational Mechanics, 50:353–366, 2012.
[31] G. Giambanco and Z. Mroz. The interphase model for the analysis of
joints in rock masses and masonry structures. Meccanica, 36:111–130,
2001.
[32] E. Grande, M. Imbimbo, and E. Sacco. A beam finite element for non-
linear analysis of masonry elements with or without fiber-reinforced
plastic (frp) reinforcements. International Journal of Architectural Her-
itage, 5:693–716, 2011.
[33] B. Kralj, G.K. Pande, and J. Middleton. On the mechanics of frost
damage to brick masonry. Computers and Structures, 41:53–66, 1991.
[34] J. Lemaitre and J.L. Chaboche. Mechanics of Solid Materials. Cam-
bridge University Press, 1990.
[35] H.R Lofti and B.P. Shing. An appraisal of smeared crack models for
masonry shear wall analysis. Computers and Structures, 41:413–425,
1991.
[36] H.R Lofti and B.P. Shing. Interface model applied to fracture of ma-
sonry structures. Journal of Structural Engineering, 120:63–80, 1994.
[37] P. B. Lourenço. Computational Strategies for Masonry Structures. PhD
thesis, Delft University of Technology, 1996.
[38] P. B. Lourenço and J. G. Rots. Multisurface interface model for analysis
of masonry structures. Journal of Engineering Mechanics, 123(7):660–
668, 1997.
[39] J. Lubliner. Plasticity Theory. Macmillan Publishing Company, New
York, 1990.
[40] R. Luciano and E. Sacco. Homogenization technique and damage model
for old masonry material. International Journal of Solids and Struc-
tures, 32(24):3191–3208, 1997.
[41] R. Luciano and E. Sacco. Variational methods for the homogenization
of periodic heterogeneous media. European Journal of Mechanics -
A/Solids, 17(4):599–617, 1998.
[42] S. Marfia and E. Sacco. Multiscale damage contact-friction model for
periodic masonry walls. Computer Methods in Applied Mechanics and
Engineering, 205-208:189–203, 2012.
[43] R. Marques and P.B. Lourenço. Possibilities and comparison of struc-
tural component models for the seismic assessment of modern unre-
inforced masonry buildings. Computers and Structures, 89:2079–2091,
2011.
[44] R. Masiani, R. Rizzi, and P. Trovalusci. Masonry as structured contin-
uum. Meccanica, 30(6):673–683, 1995.
[45] R. Masiani and P. Trovalusci. Cauchy and Cosserat materials as con-
tinuum models of brick masonry. Meccanica, 31(4):421432., 1996.
[46] T. J. Massart, R. H. J. Peerlings, and M. G. D. Geers. An enhanced
multi-scale approach for masonry wall computations with localization
of damage. International Journal for Numerical Methods in Engineer-
ing, 69(5):1022–1059, 2007.
[47] B.C.N. Mercatoris and T.J. Massart. A coupled two-scale computa-
tional scheme for the failure of periodic quasi-brittle thin planar shells
and its application to masonry. International Journal for Numerical
Methods in Engineering, 85:1177–1206, 2011.
[48] J.C. Michel and P. Suquet. Nonuniform transformation field analysis.
International Journal for Solids and Structures, 40:6937–6955, 2003.
[49] G. Milani, P. B. Lourenço, and Tralli. Homogenization approach for
the limit analysis of out-of-plane loaded masonry walls. Journal of
Structural engineering, 132:1650–1663, 2006.
[50] T. Mura. Micromechanics of defects in solid. Martinus, 1987.
290 E. Sacco
[51] D.V. Oliveira and Lourenço P.B. Implementation and validation of a

constitutive model for the cyclic behaviour of interface elements. Com-
puters & Structures, 82:1451–1461, 2004.
[52] P. Pegon and A. Anthoine. Numerical strategies for solving continuum
damage problems with softening: application to the homogenization of
masonry. Computers & Structures, 64:623–642, 1997.
[53] L. Pel, M. Cervera, and P. Roca. An embedded cohe-
sive crack model for finite element analysis of brickwork
masonry fracture. Construction and Building Materials,
http://dx.doi.org/10.1016/j.conbuildmat.2012.07.014, 2012.
[54] S. Pietruszczak and X. Niu. A mathematical description of macro-
scopic behaviour of brick masonry. International Journal of Solids and
Structures, 29(5):531–546, 1992.
[55] T. M. J. Raijmakers and A. T. Vermeltfoort. Deformation controlled
tests in masonry shear walls, In: Report B-92-1156. TNO-Bouw, Delft,
The Netherlands, 1992.
[56] M.A Ramalho, A. Taliercio, A. Anzani, L. Binda, and E. Papa. A
numerical model for the description of the nonlinear behaviour of multi-
leaf masonry walls. Advances in Engineering Software, 39:249–257,
2008.
[57] E. Reyes, J.C. Galvez, M.J. Casati, D.A. Cendon, J.M. Sancho, and
J. Planas. An embedded cohesive crack model for finite element anal-
ysis of brickwork masonry fracture. Engineering Fracture Mechanics,
76:1930–1944, 2009.
[58] E. Sacco and F. Lebon. A damage-friction interface model derived
from micromechanical approach. International Journal of Solids and
Structures, 49:3666–3680, 2012.
[59] E. Sacco and J. Toti. Interface elements for the analysis of masonry
structures. International Journal for Computational Methods in Engi-
neering Science and Mechanics, 11:354–373, 2010.
[60] Elio Sacco. A nonlinear homogenization procedure for periodic ma-
sonry. European Journal of Mechanics - A/Solids, 28(2):209–222, 2009.
[61] G. Salerno and G. de Felice. Continuum modeling of periodic brick-
work. International Journal of Solids and Structures, 46(5):1251–1267,
2009.
[62] V. Sepe, S. Marfia, and E. Sacco. A nonuniform tfa homogenization
technique based on piecewise interpolation functions of the inelastic
field. International Journal of Solids and Structures, 50:725–742, 2013.
[63] F. Sideroff. Description of anisotropic damage application to elasticity.
In Proceeding of IUTAM Colloquium on Physical Nonlinearrities in
Structural Analysis, pages 237–244. Springer. Berlin, 1981.
[64] J.C. Simo and T. J. R. Hughes. Computational Inelasticity. New Yok:

Springer, 1998.
[65] P. Suquet. Elements of homogenization for inelastic solid mechanics.
In Homogenization Techniques for Composite Media. Springer-Verlag,
Berlin., 1987.
[66] R.L. Taylor. FEAP-A finite element analysis program, Version 8.3. De-
partment of Civil and Environmental Engineering, University of Cali-
fornia at Berkeley, California, 2011.
[67] G. Uva and G. Salerno. Towards a multiscale analysis of periodic ma-
sonry brickwork: A fem algorithm with damage and friction. Interna-
tional Journal of Solids and Structures, 43:3739–3769, 2006.
[68] X. Wei and H. Hao. Numerical derivation of homogenized dynamic ma-
sonry material properties with strain rate effects. International Journal
of Impact Engineering, 36:522–536, 2009.
[69] O. C. Zienkiewicz and R. L. Taylor. The finite element method. Fourth
edition, volume 1. McGraw-Hill Publ., New York, 1991.
[70] A. Zucchini and P.B. Lourenço. A micromechanical homogenisation
model for masonry: Application to shear walls. International Journal
of Solids and Structures, 46(3-4):871 – 886, 2009.
Homogenization and Seismic Assessment:
Review and Recent Trends
* †
Paulo B. Lourenço and Gabriele Milani
*
Institute for Sustainability and Innovation in Structural Engineering (ISISE),
Department of Civil Engineering, University of Minho, Portugal
†
Department of Architecture, Built environment and Construction engineering
(A.B.C.), Politecnico di Milano, Italy
Abstract The mechanics of masonry structures has been for long

underdeveloped in comparison with other fields of knowledge. Pres-
ently, non-linear analysis is a popular field in masonry research and
advanced computer codes are available for researchers and practi-
tioners. The chapter presents a discussion of masonry behaviour and
clarifies how to obtain the non-linear data required by the compu-
tations. The chapter also addresses different homogenisation tech-
niques available in the literature in the linear and rigid-plastic case,
aiming at defining a catalogue and at discussing the advantages and
disadvantages of the different approaches. Special attention is given
to stress assumed models based either on a polynomial expansion
of the micro-stress field or in the discretization of the unit cell by
means of a few constant stress finite elements CST with joints re-
duced to interfaces. Finally, the aspects of seismic assessment are
presented and case studies involving the use of macro-block anal-
ysis, static (pushover) analysis and time integration analysis are
discussed.
Keywords: masonry, non-linear data, homogenisation, limit anal-

ysis, Finite Elements, seismic assessment
1 Homogenization Theory, Basic Assumptions

This section briefly recalls the basis of the theory of homogenization applied
to masonry structures, with particular emphasis on running bond texture.
Consider a masonry wall Ω, constituted by the periodic arrangements of
masonry units and mortar joints as shown in Figure 1. The periodicity

294 P. B. Lourenço and G. Milani
allows to regard Ω as the repetition of a representative element of volume

Y (REV or elementary cell).
x2
Block
y2
Mortar
y1
y3
Y
l
y2
x1
Elementary cell Y y1
b
y3
y2 + -
Y3 Y3
t y2
h a/2
e
y1 a
eh
a/2 y3 y1
b/2 b/2 y3
ev
Figure 1. Representative Element of Volume (REV) for running bond ma-

sonry
Let x = [x1 , x2 ] be a frame of reference for the global description of Ω

(macroscopic scale) and y = [y1/, y2 , y3 ]0 a frame of reference for Y . The Y
t t
module is defined as Y = ω × − , , where Y ∈ R3 is the elementary
2 2
cell and ω ∈ R2 represents the middle plane of the plate (Caillerie, 1984).
The ∂Y boundary surface of the elementary cell is shown also in Figure 1.
The basic idea of homogenization consists in introducing averaged quan-
tities representing the macroscopic stress and strain tensors (denoted here
respectively as E and Σ), as follows:

1 1
E = = (u) dY Σ = σ = σ dY (1)
V Y V Y
where V stands for the volume of the elementary cell, and σ stand for
Homogenization and Seismic Assessment:… 295
the local quantities (stresses and strains respectively) and ∗ is the average
operator.
According to Anthoine (1995) and Cecchi et al. (2005) the homogeniza-
tion problem in the linear elastic range in presence of coupled membranal
and flexural loads, under the assumption of the Kirchhoff-Love plate theory,
can be written as follows:
⎫
div σ = 0 (a) ⎪
⎪
⎪
σ = a(y) (b) ⎪
⎪
⎪
⎪
⎪
= E + y3 χ + sym (grad uper ) (c) ⎬
(2)
σe3 = 0 on ∂Y3+ and ∂Y3− (d) ⎪⎪
⎪
⎪
σn antiperiodic on ∂Yl (e) ⎪⎪
⎪
⎪
⎭
uper periodic on ∂Yl (f )
where σ is the microscopic stress tensor (micro-stress), uper is a ω-periodic
displacement field, E is the macroscopic in-plane strain tensor, χ is the out-
of-plane strain tensor (curvature tensor), a(y) represents a ω-periodic linear
elastic constitutive law for the components (masonry units and mortar), as
given in equation (2b). Equation (2a) represents the micro-equilibrium for
the elementary cell with zero body forces, usually neglected in the framework
of homogenization.
Furthermore, in equation (2c), the micro-strain tensor is obtained as
a linear combination among macroscopic E and χ tensors and a periodic
strain field. E and χ tensors are related to a(y) represents a ω-periodic lin-
ear elastic constitutive law for the components (masonry units and mortar),
as given in equation (2b). Equation (2a) represents the micro-equilibrium
for the elementary cell with zero body forces, usually neglected in the
framework of homogenization. The macroscopic displacement field compo-
nents U1 (x1 , x2 ), U2 (x1 , x2 ) and U3 (x1 , x2 ) by means of the classic relations
1
Eαβ = (Uα,β + Uβ,α ) , with Ei3 = 0, and χαβ = −U3,αβ with χi3 = 0,
2
α, β = 1, 2 and i = 1, 2, 3.
Macroscopic homogenized membrane and bending constants can be ob-
tained solving the elastostatic problem (2) and making use of the classic
relations: ∗
N = σ = AE + Aχ
∗ (3)
M = y3 σ = BT E + Dχ
where A, B, and D are the constitutive homogenized plate tensors. Usu-
ally, the elementary cell has a central symmetry, hence B = 0. As a rule, a
solution for the problem given by equation (2) can be obtained using stan-
dard FE packages, as suggested for the in-plane case by Anthoine (1995).
The governing equations in the non-linear case are formally identical to

equation (2) provided that a non-linear stress-strain law for the constituent
materials is assumed. Extensions to the non-linear case have been provided
by e.g. Luciano and Sacco (1997); Pegon and Anthoine (1997); Massart
(2003); Massart et al. (2004); Zucchini and Lourenço (2004), etc.
1.1 Closed-form and simplified solutions in the linear

elastic range
This section briefly recalls some of the most popular simplified approach-
es that appeared in the past in the technical literature for obtaining homog-
enized elastic moduli for masonry. Since the elastostatic problem given in
equation (2) cannot be solved in closed form for running bond masonry, sev-
eral simplifications were assumed in order to obtain “easily” homogenized
elastic moduli.
Two-step approaches. One of the first ideas presented, Pande et al.

(1989) and Maier et al. (1991), was to substitute the complex geometry of
the basic cell with a simplified geometry, so that a closed-form solution for
the homogenization problem was possible.
In particular, Pande et al. (1989) presented a model in which a two-step
stacked system with alternative isotropic layers was considered (Figure 2).
In this way, a so called “two-step homogenization” was obtained. In the first
step, a single row of masonry units and vertical mortar joints were taken into
consideration and homogenized as a layered system. In the second step, the
“intermediate” homogenized material was further homogenized with hori-
zontal joints in order to obtain the final material. In this manner, a very
simple mechanical system constituted by elastic springs was obtained and
explicit formulas based on classical elasticity concepts were presented.
Obviously, this simplification leads:
1. To underestimate the horizontal stiffness of the homogenized mate-
rial, since no information on the texture (running bond, stack bond,
Flemish bond, etc.) is considered. Furthermore, the inability of the
model to consider the regular offset of vertical mortar joints belonging
to two consecutive layered unit courses results in significant errors in
the case of non-linear analysis;
2. To obtain a homogenized material different if the steps of homogeniza-
tion are inverted (i.e. if bed joints and masonry units are homogenized
in the first step).
Following the idea of a multi-step approach, many other models involving
different approximations and ingenious assumptions have been sought, with
an increasing large number of papers in the recent years (Pietruszczak and
(a) (b)
First step
Second step
Homogenized material
Figure 2. Two-step homogenization

Niu, 1992), where a two-stage homogenization procedure was employed with

the head joints considered as uniformly dispersed elastic inclusions and the
bed joints assumed to represent a set of continuous weakness.
Reduction of joints to interfaces. A different approach, proposed by

de Felice (1995) and Cecchi and Sab (2002), is based on the reduction of
joints to interfaces. This idea arose from the observation that masonry units
are generally much stiffer than mortar and joints show a small thickness if
compared with the size of masonry units.
These studies resulted in the definition of the homogenized masonry
constitutive function by means of the introduction of several parameters
that measure smallness:
3. (scale parameter) → 0 , which represents the ratio between the size
of the cell and the dimension of the overall panel;
4. ξ , representing the ratio between the Young’s modulus of the mortar
Em
Em and the Young’s modulus of the masonry unit Eb ξ = ;
Eb
5. ϕ, representing the ratio between the thickness of the joints e and the
e
size of the characteristic module a ϕ = .
a
ξ and ϕ are parameters that take into account the effects of joint sizes
and the ratio of deformability of mortar and block. For fixed elastic tensors
of both block and mortar and for fixed geometric parameters a, b, t (defined
in 1), the so-called “asymptotic case” is obtained when ξ → 0 and ϕ → 0. If ϕ
tends to zero, the joint becomes and interface, whereas if ξ tends to zero the
mortar becomes infinitely deformable with respect to the unit. Therefore,
the typology of asymptotic problem depends on how ξ and ϕ tend to zero.
A first simplification usually introduced adopts ξ = ξ(ϕ) and addition-
ally, limϕ→0 ξ(ϕ)ϕ−1 = ρ = 0. Such asymptotic problem shows cohesive
zero thickness interfaces between the masonry units with possible jump of
the displacement field. Hence, the field problem may be formulated with
reference only to the aa elastic tensor of the masonry unit with disconti-
nuity at the interfaces, where the constitutive function is a linear relation
between the stresses on the unit surfaces and the jump of the displacement
field. Both in de Felice (1995) and Cecchi and Sab (2002), elastic springs
with diagonal constitutive tensor K for the joints are used, so introducing a
simplification related to the fact that Poisson effect of the joint is neglected.
In particular K takes the following explicit form:
1
M

K= μ I + μM + λM (n ⊗ n) (4)
e
where e is the thickness of the joint, n is the normal to the interface, and
μM , λM are the Lamé constants of mortar.
(a) (b)
y 2
y 2
Y : boundary of
elementary cell Y : elementary cell Y Y ev
y 1
eh y 1
a
b b
y 2
y 2
Yl Yl
e
2
Y3- Y3-
t y t y
1
e
1
1
y 3 y 3
e
3
Y3+ Y3+
Figure 3. unit cell utilized in de Felice (1995) and Cecchi and Sab (2002).
Left: joints reduced to interfaces. Right: actual thickness of the joints.
De Felice (1995) assumed also rigid masonry units, in order to reduce

further the complexity of the problem. In this way, see Figure 3-a, this
author showed that the problem given by equation (2) can be solved in
closed form for running bond masonry and permits to obtain analytical
formulas for the homogenized elastic constants, which depend only on the
geometry of the elementary cell and on the mechanical properties of joints.
Following this idea, Cecchi and Sab (2002) proposed a multi-parameter
homogenization study for the 2D and the 3D in-plane case, removing the
hypothesis of rigid masonry units (Figure 3-b). The finite thickness of the
joints was considered in an approximated way only in the constitutive re-
lation of the interfaces. A symbolic FE procedure was adopted, in which
the elementary cell was discretized by means of a coarse triangular mesh.
Here the term symbolic is use to indicate that the homogenization problem
was handled in symbolic form, using a mathematical software. In this way,
these authors were able to find “quasi-analytical” formulas.
The disadvantages of this approach are the following:
1. The reduction of joints to interfaces, may strongly reduce the accuracy
of the results in presence of thick mortar joints and in presence of ξ
ratios tending to zero Cecchi et al. (2005).
2. The introduction of elastic masonry units leads to formulas derived
from symbolic FE procedures and does not allow solving analytically
the homogenization problem.
3. A possible development of the method in the non-linear range can
result in non-negligible errors with respect to finite element approaches
and experimental evidences, since the role of joint thickness is lost in

the simplifications assumed.
FE procedures. Anthoine (1995) was the first to suggest the utilization

of standard FE codes for solving the homogenization problem given by equa-
tion (2) in the case of both stacked and running bond masonry. Anthoine
(1995, 1997) and Lourenço (1997) also underlined that homogenized moduli
depend on the order of the steps and 3D effects are always present.
Cecchi et al. (2005) applied FE procedures to masonry out-of-plane
loaded and stressed that:
1. Flexural moduli may significantly differ from membrane moduli, es-
pecially in the presence of weak mortar joints. As a consequence,
Kirchhoff-Love orthotropic homogenized coefficients cannot be ob-
tained simply by integration of membrane moduli.
2. Cohesive interface closed-form solutions give unreliable results when
the ξ ratio is small.
The classical assumptions adopted in the FE method applied to homog-
enization are the following:
1. Perfect continuity between units and mortar;
2. The periodic displacement that has to be imposed fulfils the con-
stant/linear assumption, at the boundary of the cell, of the macro-
scopic kinematic descriptors E and χ.
In this way, considering only a macroscopic strain tensor E acting, suit-
able boundary conditions (Figure 4) for uper periodic and σn anti-periodic
on ∂Yl (which represents the boundary of the module orthogonal to the
middle plane) are imposed, meaning that the elastostatic problem can be
formulated only on Y . It is worth noting that several engineering ap-
proaches recently presented in the technical literature do not satisfy exactly
this hypothesis (Lopez et al., 1999; Zucchini and Lourenço, 2002). In this
case, the symmetry of the cell allows to simplify the numerical model and
permits to discretize only 1/4 of the elementary cell.
The advantages of the adoption of a FE technique include:
1. The FE solution approximates the actual solution for a suitable refined
mesh;
2. Mortar joints thickness is taken into account for the evaluation of the
homogenized moduli. This leads to estimate numerically homogenized
moduli that can differ from interface moduli;
3. The influence on the homogenized horizontal Young modulus, due to
the masonry units staggering, is caught by the model, especially in
presence of mortar joints with poor mechanical properties or non-
linear behaviour.
I11
E
I12
E
u1-u1 =E11 y1
per
u1-u1 =E11 y1
per
y2 u1-u1 =E12 y2
per
y2
u2-u2 =E12 y1
per
u2-u2 =E12 y1
per
y1 y1
fiber u1-u1 =E12 y2

per
per per
u2 =0 u1 =0
per
u =0 per
u1 =0 u2 =0per per
u2 =0
per per
u2 =0 u1 =0
Figure 4. Applied displacement boundary conditions on the elementary
cell.
On the other hand, the most severe limitation of the approach is that
the computational cost of a FE procedure does not compete favourably
with macroscopic approaches when non-linear problems are treated, since
the homogenization field problem has to be solved for each Gauss point
of each loading step. This leads to handle continuously a “two-size” FE
problem (macroscopic and cell level), where the averaged results obtained
at a cell level are utilized at a structural level (in the framework of non-linear
numerical procedure).
1.2 Simplified stress assumed FE discretization

Simplified models based on a stress assumed FE discretization in the
elastic range have been recently proposed by Milani et al. (2006a) and Milani
(2011). Essentially, the approach relies into a rough FE discretization of
the unit cell, where inside each element a polynomial interpolation of the
stress field is a-priori given. Equilibrium inside each element and at the
interface between contiguous elements and anti-periodicity conditions are
imposed. The solution in the elastic range is simply achieved minimizing
the total complementary energy stored into the unit cell, subjected to a
certain number of equality constraints representing equilibrium and anti-
periodicity.
As shown in Figure 5, one-fourth of the REV is sub-divided into nine
Mortar Brick 12
1 1 15 11
2 ev b 2 ev 10
18 14 3
13 2
a 3 2 1 17
30 6 1
eh 6 5 4 H 16 5
29 9
a 33 8 4
9 8 7 28
32 7
L 36 21
20
35 31
23
24 19
34
26
27 22
25
Figure 5. Subdivision in sub domains adopted: Left: subdivision and ge-

ometrical characteristics of one-fourth of the elementary cell, Right: subdi-
vision into 36 sub-domains for the entire cell.
geometrical elementary entities (sub-domains), so that all the cell is sub-

divided into 36 sub-domains.
For each sub-domain, polynomial distributions of degree (m) are a priori
assumed for the stress components. The generic ij-th component can be
written as follows:
(k)
σij = X(y)STij y ∈ Y K (5)
where:
%1
– X(y) = y1 y2 y12 y1 y2 y22 . . .&;
(1) (2) (3) (4)
(6) (5)
– Sij = Sij Sij Sij SijSij. . . is a vector of length
Sij
2

))
(N N) = m + 3m + 1 = (m + 1)(m + 2) representing the un-
2 2 2
known stress parameters;
– Y k represents the k th sub-domain.
The imposition of equilibrium with zero body forces inside every sub-do-
main, the continuity of the stress vector on interfaces and anti-periodicity
of σn permit to strongly reduce the total number of independent stress
parameters.
In particular, equilibrium has to be imposed everywhere inside each sub-
domain, i.e. σij,j (x, ) = 0, i = 1, 2∀(x, y) ∈ sub-domain. Since σij (x, Y ) is a
polynomial expression of degree (m), a linear combination of its derivatives

(div σ) is a polynomial of degree (m − 1). Therefore equilibrium inside each
sub-domain leads to write (2N ) linear independent equations in the stress
m2 3m (m + 1)(m + 2)
coefficients, where N = + +1= .
2 2 2
(a) (b)
y2(j) Quadratic interpolation for (or )
j-th sub-domain: frame of reference Local frame of reference
O (j) y1(j) (k)
(k-r) interface
y2(r) y2(k) y2
(r)
H O(r) y1(r) O(k) y1(k)
(k)
H (k) y2(q) (r)
(r) L y1(q) (k) 1 2 3
L O(q)
(q) (q)
H L (q) y1
(q-k) interface (q)
Equilibrium to check in 3 points
(c) (d)
1 2
elementary cell
(m)
n1 1 2
n2
n1 3
4
(n)
n2 3
4
Figure 6. Contiguous sub-domains. -a: geometry and frame of reference

of the sub-domains; -b: vertical/horizontal interfaces between adjacent sub-
domains; -c: anti-periodicity conditions on the unit cell; -d: linear depen-
dence of some equilibrium equations.
A further reduction of the total unknowns is obtained a priori imposing

(k) (r)
the continuity of the stress vector on internal interfaces (σij nj + σij nj =
0, i = 1, 2) for every (k) and (r) contiguous sub-domains with a common
interface of normal n (6). Being Eqs. (5) polynomial expressions of degree
(m) in the abscissa s of the interface, other 2N equations (where N =
m + 1) in Ŝ(k) and Ŝ(r) for each (k ↔ r) interface can be written from (5),
see Figure 6-b.
Finally anti-periodicity of σn on ∂V requires 2N additional equations
per pair of external faces (m)(n) (Figure 6-c), i.e. it should be imposed that
stress vectors σn are opposite on opposite sides of ∂V :

(m) (n)
X̂ij (y) Ŝ(m) n1 = −X̂ij (y) Ŝ(n) n2 (6)
Where n(m) and n(n) are oriented tensors of the external faces of the paired
sub-domains (m)(n).
To conclude, some elementary assemblage operations on the local vari-
ables (handled automatically) lead to write the stress vector inside every
sub-domain as follows:
σ ) (k) (y) S
) (k) = X ) k = 1, . . . , k max (7)
where:
) (k) is the vector of membrane actions inside the k th sub-domain;
– σ
– X) (k) is a 3 × Nun matrix which contains only geometrical coefficients;
its elements are polynomial forms in the microscopic coordinate y;
) is the vector (of length Nun ) of the total stress parameters unknown.
– S
The equations written in order to satisfy internal equilibrium, equilib-
rium on interfaces and anti-periodicity of the stress vector lead to a system
of equations AS = 0 , where S is the vector of total stress parameters. Nev-
ertheless, not all the rows of this system are linearly independent. This can
be easily shown if four generic rectangular elements with four common in-
terfaces and subjected only to constant non zero shear stress are considered,
as reported in Figure 6-d. Internal equilibrium is a priori satisfied, whereas
four equations for ensuring equilibrium on interfaces have to be written.
Nevertheless, only three of these four equations are linear independent.
Finally, four different models of increasing accuracy (P0 P2 P3 P4 )
have been obtained increasing the degree of the polynomial expansion.
1.3 Linear elastic case

A preliminary study in the linear elastic range may be done consider-
ing the quadratic functional Π of the complementary energy evaluated in
the REV. With the stress-assumed discretization previously discussed, an
approximation of Π can be written as:
max
k
∗ 1 ) T ) (k)T ) (k) S
) dYk − ) (k)T ū dSj (8)
)T X
Π = S X (y) Cb,m X S
Yk 2 j Sj
k=1
Where Cb,m is the compliance matrix of units or mortar joints and ū is the
displacement imposed on the boundary ∂Y of the elementary cell, repre-
senting a given macroscopic strain tensor E.
The minimization of the approximated expression of Π leads to the fol-

lowing expression:
-kmax .

∇Π =∗ )
X (k)T b,m
(y) C X ) dYk S
(k) )
k=1 Yk
(9)
− ) (k)T (y) ū dSj
X
j Sj
) − Ū = 0
= Chom S
which enables to find both S,) by factorization of the matrix Chom and Σ,
from integration of the local stress field.
(a) (b)
0 0
10 10
P4 P4
P3 P3
P2 P2
P0 P0
FEM 2D FEM 2D
A1122/A 1122
A1111/A 1111
b
b
-1
10
0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 90
E b/E m E b/E m
(c) (d)
0 0
10 10
P4 P4
P3 P3
P2 P2
P0 P0
FEM 2D FEM 2D
A1212/A 1212
A2222/A 2222
b
-1
10
-1
10
0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 90
E b/E m E b/E m
Figure 7. Homogenised in-plane moduli. -a: A1111 modulus; -b: A1212

modulus; -c: A1122 modulus; -d: A2222 modulus.
A comparison between the elastic moduli provided by the model pro-

posed and a standard numerical 2D FEM procedure, Anthoine (1995), is
Table 1. Initial mechanical properties assumed for the elastic simulations

reported in Figure 7.
E[ N/mm2 ] ν
Young modulus Poisson ratio
unit 11000 0.2
mortar 2200 0.25
presented in Figure 7. The dimensions of the units are assumed to be

250 × 120 × 55 mm3 (brick UNI5628/65) and the thickness of mortar joints
is equal to 10 mm. The initial mechanical characteristics of materials are
reported in Table 1; the simulations are handled keeping the brick Young
modulus Eb constant and progressively reducing the mortar Young modulus
Em , so assuming a wide range of Eb /Em ratios (from 5 to 90), in order to
simulate also historical brickwork. Homogenized Aijhk membrane moduli
are represented varying Eb /Em ratio. The homogenized moduli are normal-
ized versus the corresponding moduli of the brick. As it can be noticed, the
provided moduli are reliable in a large range of Eb /Em ratios, even for the
simplest model with constant stresses tensor (P 0). Nevertheless, Figure 7-a
shows that a progressively reduced accuracy of the P 0 model can be noticed
for the A1111 module, due to the presence of shear stresses in the bed joint.
1.4 Constant Stress Triangular elements discretization of the

unit cell
In Milani (2011), an alternative static model relying into the subdivision
of the unit cell into 24 constant stress triangular elements and joints reduced
to interfaces is presented. Due to the very limited number of optimization
variables involved, the model can be handled also without the assistance of
a computer.
Joints are reduced to interfaces with zero thickness and blocks are dis-
cretized by means of a coarse mesh constituted by three-noded plane-stress
elements, Figure 8. The choice of meshing 1/4 of the brick through at least
3 triangular elements is due to the need of reproducing the presence of shear
stress in the bed joint (element 2 in Figure 8) in horizontal stretching. All
the non-linearity in the RVE is concentrated exclusively on interfaces be-
tween adjoining elements both on brick and joint. Brick-brick interfaces
allow, at least in principle, the reproduction of blocks failure. The six CST
elements used for the discretization of the upper-right 1/4 of the REV are
indicated in Figure 14 as 1, 2, 3, 1’, 2’, 3’.
In the model, the non-dimensional geometrical coefficient ρ indicates the
y2
1. reduction of joint to interfaces
t/2 2. subdivision of the REV into triangular FEs
t/2 6: Masonry middle plane 3. non linarity concentrated on brick-brick and
mortar interfaces
e2 Anti-periodicity of micro-stress and periodicity of displacement
y2
y3 periodic fluctuation
e3 Y n
e1
y1 P2
P1
n y1
Y per
per
u (P2 )
u (P1 ) P1
P2
REV 1 stress acting on elements and equilibrium on

4 REV (2) (1)
y2 yy (2)
interfaces yy (1)
(6) (4) (3) (1) (3)

(2) (1)
(5) (2) (1) (2)
(2)
xx
n (1) (1)
xx
(5') (4') (2') (1')

(6') (3') a (2) (1)
n (1)
(12) (9) (2') yy
(11) (10) (8) (7) (3') (1') (2) n
(1)
(1)
y1 (2)
yy xx
(11') (10') (8') (7')

(12') (9')
b Brick element Brick interface Mortar interface
Figure 8. The micro-mechanical model proposed. Subdivision of the

REV into 24 CST triangular elements (and 1/4 into 6 elements) and anti-
periodicity of the micro-stress field.
ratio between brick semi-length and height, i.e. ρ = b/2a and the super-
script (n), with n positive integer, a stress component belonging to the n-th
element. In this way, assuming a plane-stress condition, the Cauchy stress
tensor inside the n-th CST element σ (n) is constituted by the components
(n) (n)
σxx (horizontal stress), σyy (vertical stress) and τ (n) (shear stress).
Within the static approach of limit analysis, equilibrium inside each
element is a priori satisfied, div σ = 0. On the contrary, two equality con-
straints involving Cauchy stress tensor components of triangular elements
must be imposed for each internal interface between adjoining elements.
It can be shown that the imposition of equilibrium on interfaces is rep-
resented by a set of 10 equations (being 5 the interfaces for 1/4 of the REV)
in 18 unknowns (three stress components for each triangular element).
Having in mind to analyze masonry macroscopic behaviour under com-
bined states of stress acting on its middle plane, all the REV must be con-
sidered, as depicted schematically in Figure 8. Anti-periodicity constrains
for the stress vector field are written on couples of triangles 1–6, 1’–6’, 7–12,
7’–12’, 1–7’, 3–9’, 4–10’, 6–12’.
When the whole cell problem is considered, independent variables are
represented by stresses acting on the triangular elements (3 unknowns per

element, i.e. 72 unknowns per REV) and the three homogenized stresses
Σxx , Σyy and Σxy (75 total optimization variables).
Σxx , Σyy and Σxy are linked with local stress variables by means of the
following set of three equations:
tr
1 (i) (i)
N
Σxx = A σ
AC i=1 tr xx
tr
1 (i) (i)
N
Σyy = A σ (10)
AC i=1 tr yy
tr
1 (i) (i)
N
Σxy = A τ
AC i=1 tr
(i)
Where, apart quantities already introduced, Atr is the area of the i-th
CST element, N tr is the total number of elements in the unit cell and AC
is the total area of the unit cell.
From equilibrium equations, anti-periodicity and (10) a set of linear
equations in both the elastic and rigid-plastic problem is obtained.
When dealing with the elastic case, membrane elastic homogenized mod-
uli may be obtained in the same way followed for the polynomial expansion
shown in the previous case, i.e. minimizing the complementary energy in
the unit cell. In this case, the complementary energy assumes the following
quadratic form:
N tr
: ;
(i)2 (i)2
1 (i) σxx
(i) (i) 2
∗ σxx σyy σyy τ (i)
Π = A + 2νb + +
2 i=1 tr Eb Eb Eb Gb
NI
: ; (11)
(i)2
1 (i)2 σn
2
τ (i)
+ A + − AC Σij Eij
2 i=1 I Em Gm
(i)
Where N I is the total number of mortar interfaces, AI is the area of the
i-th mortar interface and Σij Eij is a summation saturating indices i and j
(assuming either value x or y) and Eij is a prescribed macroscopic strain
component.
Analogously to the previous case, the determination of the membrane
elastic moduli may be obtained by a constrained minimization of the com-
plementary energy, which is a quadratic form on the 72 independent micro-
stress variables of the elements and the three variables Σij representing the
homogenized stress. Thanks to the very limited number of optimization

variables involved, a standard large scale quadratic programming routine is
utilized to solve the elastic problem on the unit cell, varying Eb /Em ratio
in a wide range and assuming as initial values for the simulations those
reported in Table 1.
Results of the simulations are summarized in Figure 9, where the ana-
lytic solution by Lourenço (1996) is also reported, to compare with A2 222
modulus. Essentially, such approach consists in homogenizing a masonry
pillar constituted by two half bricks (height h) and a joint with thickness e.
For such a structural system it can be proved that the vertical membrane
h+e
elastic modulus is A2222 = .
h e
(1 − νb2 ) 2 )
+ (1 − νm
Eb Em
1
A1111
A2222
0.8
A1122
A1212
0.6 A2222 analytic
Aijhk/A1111
b
0.4
0.2
0
0 10 20 30 40 50 60 70 80
Eb/Em
Figure 9. Homogenized in-plane moduli. -a: A1111 modulus; -b: A1212

modulus; -c: A1122 modulus; -d: A2222 modulus.
1.5 Homogenized failure surfaces

Both stress assumed approaches previously discussed may be extended
in the case of materials exhibiting rigid plastic behaviour, infinite ductility
and associated flow rules. This is a classic homogenization problem in the
rigid plastic case, for which upper and lower bound theorems have been
provided some decades ago by Suquet (1983) in a general framework. Basing

on such theorems, limit analysis combined with homogenization has been
extensively applied to masonry structures by e.g. de Buhan and de Felice
(1997); Milani et al. (2006a,c); Milani (2011), etc.
De Buhan and de Felice (1997) were the first to apply the kinematic theo-
rem of limit analysis in the framework of masonry homogenization, assuming
joints reduced to interfaces with a classic Mohr-Coulomb failure criterion
and masonry units infinitely resistant. Milani et al. (2006a,c) and Milani
(2011) adopted a static approach, in which either polynomial equilibrated
and admissible stress fields were a priori assumed in a finite number of sub-
domains or a rough discretization with stress constant triangular elements
was utilized. In this way, both compressive failure and actual thickness of
the joints as well as units crushing may be considered.
These approaches have the following advantages:
1. Masonry homogenized failure surfaces can be recovered making use of
well-known linear programming routines and requiring a very limited
computational effort;
2. The homogenized failure surfaces so obtained can be implemented in
FE limit analysis codes for collapse analysis, without limitations and
not requiring to solve a cell problem in each Gauss point at a structural
level;
3. They can compete favourably with macroscopic approaches and give
relevant information at failure.
On the other hand, some limitations are worth noting:
1. Limit analysis is incapable to give an information on displacements at
collapse;
2. As experimental evidences show, frictional behaviour is typically non
associated and, at present, mathematical theorems concerning non
associated limit analysis applied to homogenization are not available;
3. Masonry behaviour can be quasi-brittle. As a consequence, the as-
sumption of infinite ductility for the constituent materials can be in-
adequate and preclude the models to be predictive.
Stress assumed approach and polynomial expansion on rectangu-

lar sub-domains. The polynomial expansion of the stress field in rect-
angular sub-domains automatically represents an equilibrated stress field
inside the unit cell. To be statically admissible, each point of each sub-
domain should be subjected to a stress state not violating the material
yield. In this framework, the static theorem of limit analysis assures that a
point of the failure surface may be found solving the following optimisation
problem:
⎫
max {λ̂} ⎪
⎪
⎧ ⎪
⎪
4kmax ⎪
⎪
⎪
⎪ 1 ⎪
⎪
⎪
⎪ λ̂nΣ = X ) dY
) (k) (y) S (a) ⎪
⎪
⎪
⎬
⎪
⎪ Y
⎨ k=1 Y
(12)
such that y ∈Yk (b) ⎪
⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ )
) (k) (y) S ⎪
⎪
⎪ ) =X
σ (c) ⎪
⎪
⎪
⎪
⎩ ⎪
⎭
) (y) ∈ S k k = 1, . . . , 4k max
σ (d)
where + ,T
– nΣ = α11 α22 α12 is a tensor in the homogenized stress space
Σ11 Σ22 Σ12
– λ̂nΣ represents a macroscopic stress state on the homogenized failure
surface S hom , belonging to a straight line from the origin of direction
nΣ
– S k stands for the failure surface of the component (unit or mortar)
belonging to the ith sub-domain.
– Y is the area of the elementary cell.
The optimisation problem given by Eqs. (12) is generally non-linear as
a consequence of the (possible) non-linearity of the strength functions of
the components. In addition, condition Eqs. (12) (d) has to be checked in
every point of the domain Y . Nevertheless, as suggested in a classical paper
by Belytschko and Hodge (1970), the check could be avoided imposing the
material admissibility only where the stress status is the maximum. This
is feasible only for the P 0 and P 1 models; alternatively, the discretisation
proposed here consists in enforcing, in every sub-domain, the admissibility
condition in a regular grid of “nodal points” with step r × q (quasi-lower
bound approach).
Within this assumption, the optimisation problem reduces to:
⎫
max {λ̂} ⎪
⎪
⎧ ⎪
⎪
1 ⎪
⎪
⎪
⎪ ) (k) ) ⎪
⎪
⎪
⎪
λ̂n Σ = X (y) S dY ⎪
⎪
⎪
⎪ Y Y ⎪
⎪
⎪
⎪
k ⎬
⎪
⎨yj ≡ nodal point (13)
such that
j ⎪
⎪
⎪σ) =X
j ) (k) ) ⎪
⎪
⎪
⎪ y S ⎪
⎪
⎪
⎪ j ⎪
⎪
⎪
⎪ )
σ ∈ S j
j = 1, . . . , rq ⎪
⎪
⎪
⎪ ⎪
⎪
⎩ max ⎭
k = 1, . . . , 4k
Optimisation problem (13) generally remains non-linear. In order to use

linear pro-gramming algorithms, each of the non-linear inequalities of Eqs.
(13) could be approximated by a set of linear inequalities (as proposed in
the past, for instance, by Anderheggen and Knöpfel (1972) or Olsen (1998),
by replacing the yield surfaces with inscribed hyper polyhedrons. Finally,
the convergence of the solution obtained should be checked progressively
increasing the number of planes of the approximation, see Sloan (1988);
Sloan and Kleeman (1995) and Olsen (1998) and many others.
Alternatively and more efficiently, an iterative procedure may be adopt-
ed, taking advantage of the fact that the simplex method proceeds from
basic solution to basic solution towards an optimal basic solution, i.e. on
the vertices of the hyper polyhedron.
The basic idea of the iterative procedure adopted is the following: in the
starting step, a coarse linear approximation of the non-linear failure surfaces
of the components is adopted, as shown in Figure 10-a. The application of
the simplex method in the optimisation at the i-th step leads to an optimal
solution in a corner of the domain.
From the iterative i-th solution point a new tangent plane is added in P
as shown in Figure 10-b, so restarting an (i + 1)th optimisation procedure.
The iterations continue until a fixed tolerance in the error between the ith
and (i + 1)th solution is reached.
(a) (b)
Si: solution step i 2 2
P1 P5 P1
P2 P2
P'
1 1
O O
P4 P6 P4
P3 P3
step i step i+1
Figure 10. Iterative procedure utilized in the optimization problem.
In Figure 11, the strength domain obtained increasing the degree of the
polynomial expansion is represented in the macroscopic stress space with
Σ12 = 0 ; the results are compared with a full finite element limit analysis
on the REV. Mechanical characteristics of the constituent materials are

summarized in Table 2; a Mohr-Coulomb failure criterion in plane stress is
chosen for mortar joints, while units are supposed infinitely resistant. Units
dimensions are 52.5×17.5 mm(length×height) and mortar joints are 10 mm
thick.
Table 2. Mechanical properties assumed for mortar joints (plane stress,

units infinitely resistant).
Mortar
Frictional angle (Φ) Cohesion (c)
37◦ 1.0028 MPa
2c cos(Φ) 2c cos(Φ)
σt = 1+sin(Φ) σc = 1−sin(Φ)
As Figure 11 shows, the model with constant stress tensor (P 0) is unable

to reproduce the typical anisotropic behaviour of masonry at failure (Mi-
lani et al., 2006a), while the refined models give a progressively increased
accuracy of the results (especially P 3 and P 4) in comparison with the FE
analysis. Therefore, a cubic interpolation P 3 is at least recommended.
tension-tension range 12 =0
1.25
=81°
=72°
6°
°
4°
5°
=3
=63
=4
=5
°
P0 Model =27
P2 Model
1.00
P3 Model
P4 Model
F.E. Model
0.75
=18°
V t
v
0.50
=9°
0.25
0
0 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00 2.20
h
V t
Figure 11. Failure surface in the tension-tension range for the models
proposed without shear actions.
Stress assumed approach and CST elements. When dealing with

the discretization of the unit cell by means of CST elements and in presence
of finite resistance of both interfaces and triangular elements the homoge-
nization problem may be re-written as:
max λ
⎧
⎪
⎪ %
24
(i)
⎪
⎪ σxx Ai
⎪
⎪
⎪
⎪ λα = i=1
⎪
⎪ 2ab
⎪
⎪
⎪
⎪ %24
⎪
⎪ (i)
⎪
⎪
σyy Ai
⎪
⎪ i=1
⎪
⎪ λβ =
⎪
⎪ 2ab
⎪
⎪
⎨ %24
(14)
subject to τ (i) Ai
⎪
⎪ λγ = i=1
⎪
⎪
⎪
⎪
2ab
⎪
⎪ I I
⎪
⎪ A eq X = b eq
⎪
⎪
⎪
⎪ A ap
X = b ap
⎪
⎪ eq eq
⎪
⎪
⎪
⎪ i (i) (i) (i)
≤ 0 ∀i = 1, . . . , 24
⎪ E xx yy
⎪ f σ , σ , τ
⎪
⎪
⎪
⎪
⎩fIi σ (i) , τ (i) ≤ 0 ∀i = 1, . . . , 32
I I
where symbols used in equation (14) have the following meaning:

– α, β and γ indicate the components of the unitary vector nΣ , see
Figure 12, in the homogenized membrane stress space. The solution of
the optimization problem (14) allows the determination of a point on
the homogenized failure surface having coordinates Σxx = αλ, Σyy =
βλ and Σxy = γλ. Note that independent variables pass from 75 to 73
introducing the failure multiplier. Usually, masonry in-plane failure
surface sections are represented assuming a fixed angle ϑ between the
bed joint and the macroscopic horizontal action Σ11 . Such sections are
obtained keeping ϑ fixed and varying point by point ψ angle, defined
as ψ = tan−1 (Σ22 /Σ11 ) , where Σ22 is the macroscopic vertical action.
In this framework, vector nΣ has the following form:
1
nΣ (1) = (cos(ψ) (1 + cos(2ϑ)) + sin(ψ) (1 − cos(2ϑ)))
2
1
nΣ (2) = (cos(ψ) (1 − cos(2ϑ)) + sin(ψ) (1 + cos(2ϑ))) (15)
2
1
nΣ (3) = (cos(ψ) cos(2ϑ) − sin(ψ) cos(2ϑ)) tan(2ϑ)
2
– Ai is the area of the i-ith element (ab/8 or ab/16);

– X is a 73 × 1 row vector and collects all the optimization unknowns
(elements stress components and collapse multiplier);
– AIeq = bIeq is a set of linear equations collecting equilibrium constraints
on all interfaces. Since 32 interfaces are present in the discretized REV
and 2 equality constraints has to be written for each interface (it has
to be ensured that both normal and shear stresses are continuous on
the interface), AIeq is a 64 × 73 matrix and bIeq is a 64 × 1 vector of
all zeros.
– Aap ap
eq X = beq collects anti-periodicity conditions and it is therefore a
set of 16 equations. Thus Aap eq is a 16 × 73 matrix and beq is a 16 × 1
ap
of all zeros.
vector
(i) (i)
– fEi σxx , σyy , τ (i) ≤ 0 is a set of non linear inequalities constraints
representing the failure surface adopted for the i-th element. Within
a linear programming scheme, such failure surfaces are normally lin-
earized. The linearization is usually a lower bound one when a static
approach is used, to ensure that a strict lower bound estimation of the
collapse load is obtained. Such a lower bound approximation is easily
obtained
bymeans of a Delaunay tessellation.
(i) (i)
– fIi σI , τI ≤ 0 ∀i = 1, . . . , 32 plays the role of fEi for the interfaces.
Two typologies of interfaces are present in the model, namely brick-
brick interfaces and mortar joints reduce to interfaces. When deal-
ing with the numerical applications reported hereafter, a linearized
Lourenço and Rots (1997) failure criterion is adopted for joints re-
duced to interfaces and a classic Mohr-Coulomb failure criterion is
used for brick interfaces. While in the first case a rough approxima-
tion of the elliptic cap is assumed (in agreement with Sutcliffe et al.
(2001), in the second the constraint is already linear.
(i) (i)
– σI and τI indicate respectively the normal and shear stress acting
on interface i.
(14) is a standard linear programming problem and the reader is referred
to e.g Anderheggen and Knöpfel (1972) for a critical discussion of efficient
(classic) linear programming tools suited for solving (14). On the other
hand, it is worth noting that recent trends in limit analysis have demon-
strated that the linearization of the strength domain can be circumvented
using conic/semidefinite programming (e.g. Krabbenhoft et al. (2005)).
In-plane homogenized failure surfaces. The brickwork considered by

Raijmakers and Vermeltfoort (1992) for performing some experimental tests
on shear walls is examined. Brick dimensions are 210 × 52 × 100 mm3 ,
xy
macroscopic strength domain
yy
xx
Figure 12. General in-plane load. Meaning of multiplier in the homoge-

nized stress space (Σxx = nΣ (1) , Σyy = nΣ (2) and Σxy = nΣ (3))
whereas the thickness of the mortar joints is 10 mm. Such shear walls have
been examined through numerical simulations and a micro-mechanical ap-
proach by many authors, e.g. Lourenço and Rots (1997); Milani et al.
(2006a); Sutcliffe et al. (2001), etc.
It is worth noting that a comparison with a kinematic formulation is
possible for joints reduced to interfaces and bricks infinitely resistant. The
kinematic formulation, again solved using linear programming, is the fol-
lowing: ⎫
[[v]] σ ds⎪
1
χ = min ⎪
⎪
v Γ Γ ⎪
⎪
⎪
⎬
0
Σ :D=1 (16)
⎪
⎪
n
⎪
⎪
[[v]] = λ̇i ∇σ f (i) ⎪
⎪
⎭
i=1
where χ is the kinematic limit multiplier of the assigned macroscopic stress

Σ0 and λ̇i is the plastic multiplier associated with the (linear) inequality
constraint f (i) ≤ 0 which determines the admissible stress state.
Two models (A and B) for joints are critically examined, see Table (3).
They differ only for the compressive cap, which is vertical in model A and
with a very prominent shape in model B.
The homogenized surfaces —at different orientations ϑ of the biaxial
load with respect to material axes— in the compression-compression re-
Table 3. Raijmakers and Vermeltfoort experimental data. Mechanical

properties adopted for mortar joints reduced to interfaces and bricks.
Mortar joints reduced to interfaces
(Mohr-Coulomb failure criterion with tension cutoff)
and linearized compressive cap)
Model A Model B
cohesion [ MPa ] c 1.4ft
tensile strength [ MPa ] ft 0.25
compressive strength [ MPa ] fc 10.5
friction angle [Deg] Φ 37
shape of the linearized compressive
Ψ 30 90
cap [Deg]
Brick interfaces and triangular brick elements
(Mohr-Coulomb failure criterion)
cohesion [ MPa ] cb 2
friction angle [Deg] Fb 45
gion obtained using the presented static models (results of both models in
practice coincide for polynomial expansions with degrees higher than 5) are
reported in Figure 13 (Model A) and Figure 14 (Model B), whereas the ho-
mogenized behaviour in the tension-tension region is depicted in Figure 15.
In this latter case, obviously Model A and B provide the same result. Along
with static failure surfaces, the corresponding kinematic strength domains
obtained solving (16) are represented. For graphical convenience, kinematic
failure surfaces are slightly shifted. As it can be noted, the agreement
between the here revised static approaches and the kinematic procedure
is almost perfect for all the points inspected. The results show that the
homogenized surface depends on the geometrical and mechanical character-
istics assumed for the components.
-2
T =0° present model
Homogenized stress6 22 [MPa]
T =22.5°
-4 T =45°
T =0° kinematic model
T =22.5°
-6
T =45°
-8
-10
-12
-20 -18 -16 -14 -12 -10 -8 -6 -4 -2 0
Homogenized stress 6 11 [MPa]
Figure 13. Model A. Compression-compression region. Biaxial strength

domain at different orientation of the horizontal action with respect to bed
joint direction. Comparison between static approaches and kinematic so-
lution. For graphical convenience, kinematic failure surfaces are slightly
shifted from their actual position.
0
-2
-4
-6
-8
-10
-12 T =0° present model

T =22.5°
-14 T =45°
T=0° kinematic model
-16
T =22.5°
-18 T =45°
-25 -20 -15 -10 -5 0

Figure 14. Model B. Compression-compression region. Biaxial strength

domain at different orientation of the horizontal action with respect to bed
joint direction. Comparison between static approaches and kinematic so-
lution. For graphical convenience, kinematic failure surfaces are slightly
shifted from their actual position.
0.5 T =0° present model

T =22.5°
0.4 T =45°
T=0° kinematic model
T =22.5°
0.3 T =45°
0.2
0.1
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Figure 15. Tension-tension region. Biaxial strength domain at different ori-

entation of the horizontal action with respect to bed joint direction. Com-
parison between static approaches and kinematic solution. For graphical
convenience, kinematic failure surfaces are slightly shifted from their actual
position.
subdivision of masonry thickness in layers
Li =t/n layer thickness

y2
t/2 L1
t/2
L2
Ln
y3
y1
i-th layer discretization
(6) (4) (3) (1)
(5) (2)
(5') (4') (2') (1')

(6') (3') a
(12) (9)
(11) (10) (8) (7)
(11') (10') (8') (7')

(12') (9')
b
Figure 16. The micro-mechanical model proposed for out-of-plane actions.
Subdivision in layers along the thickness and discretization of each layer
into triangular equilibrated elements.
Generalization to the out-of-plane case. The generalization of the

second static model, presented in this Chapter, to out-of-plane actions
(Kirchhoff-Love hypothesis), if performed within the utilization of standard
linear programming routines, requires a subdivision (nL ) of the wall thick-
ness t into several layers (Figure 16-a), with a priori fixed constant thickness
δL = t/nl for each layer. Hence, to estimate a point of the failure surface
in the bending moment-torque space it is necessary to solve the following
linear programming problem:
max{λ}
⎧ %
⎪
⎪ %
nL
t+ΔLi
24
(i,j)
⎪
⎪ ΔL i − jΔLi σxx Ai
⎪
⎪
2
⎪
⎪ λα =
j=1 i=1
⎪
⎪ 2ab
⎪
⎪
⎪
⎪ %
nL %24
⎪
⎪ t+ΔLi
−
(i,j)
⎪
⎪ Δ L i 2 jΔ L i σyy Ai
⎪
⎪
⎪
⎪ λβ =
j=1 i=1
⎪
⎪ 2ab
⎪
⎪ %
⎪
⎪ %
nL 24
⎪
⎪ ΔL i
t+ΔLi
− jΔ τ (i,j) Ai
⎪
⎪ 2 L i
⎪
⎪ j=1 i=1
⎪
⎪ λγ =
⎪
⎪ 2ab
⎪
⎪
⎪
⎪ %
nL %
24
⎪
⎪ ΔLi
(i,j)
σxx Ai
⎪
⎪
⎪
⎨Nxx = j=1 i=1
= Σxx t = 0 (17)
s.t. 2ab
⎪
⎪ %
nL %24
⎪
⎪ (i,j)
⎪
⎪ ΔL i σyy Ai
⎪
⎪ j=1 i=1
⎪Nyy =
⎪ = Σyy t = 0
⎪
⎪ 2ab
⎪
⎪
⎪
⎪ %
nL %
24
⎪
⎪ ΔLi τ (i,j) Ai
⎪
⎪
⎪
⎪ j=1 i=1
⎪
⎪ Nxy = = Σxy t = 0
⎪
⎪ 2ab
⎪
⎪
⎪
⎪ AI X = bIeq
⎪
⎪ eq
⎪
⎪
⎪
⎪ Aap eq X = beq
ap
⎪
⎪
⎪
⎪
⎪
⎪ f i,j
σ (i,j)
, σ (i,j) (i,j)
, τ ≤ 0 ∀i = 1, . . . , 24 ∀j = 1, . . . , nL
⎪
⎪ E xx yy
⎪
⎪
⎪
⎩f i,j σ (i,j) , τ (i,j) ≤ 0 ∀i = 1, . . . , 32 ∀j = 1, . . . , nL
I I I
where all the symbols have been already introduced for the in-plane case.
With respect to the in-plane case, the following key issues are worth
noting:
– λ is the value of the failure strength in the Mxx − Myy − Mxy space;
– α, β and γ indicate the components of the unitary vector nΣ , see
Figure 12, in the homogenized Mxx − Myy − Mxy space, in analogy to
what stated for the in-plane case;
– AIeq X = bIeq collects equilibrium constraints of all interfaces of each
layer. Since between contiguous layers no shear stresses are exchanged,
such constraints are the same of the in-plane case, one set written for
each layer. AIeq is a 64nL × (72nL + 1) matrix and bIeq is a 64nL × 1
vector of all zeros. Analogous considerations can be repeated for the
equations set Aap ap
eq X = beq , which collects anti-periodicity conditions
for each layer. For the out-of-plane case, Aap eq is a 16nL × (72nL + 1)
matrix and bap eq is a 16nL × 1 vector of all zeros;
– differently to the in-plane case, three additional equality constraints
have to be imposed, corresponding to require that homogenized mem-
brane actions Nxx − Nyy − Nxy are equal to zero; - vector X collects
all the unknown stresses of each FE of each layer). Therefore, X is a
vector of length 3 × 24 × nL .
Finally, it is worth noting that membrane actions are kept, for the sake
of simplicity, constant and independent from load multiplier. Consequently,
in-plane actions effect optimization only in the evaluation of Mxx , Myy , Mxy
strength domains. Generally, this assumption is technically acceptable for
experimental tests where vertical pre compression Nyy is constant and ap-
plied before out-of-plane actions.
1.6 Out-of-plane homogenized failure surfaces

In the first example, the ultimate masonry horizontal bending, torsion
and vertical bending (i.e. Mxx , Mxy and Myy ) are evaluated in absence of
pre-compression.
Table 4. Mechanical properties adopted for the out-of-plane numerical sim-

ulations in absence of vertical pre-compression (UNI bricks).
Mortar joints reduced to interfaces
(Mohr-Coulomb failure criterion)
friction angle [Deg] Φ 27
cohesion [ MPa ] c 0.132
Standard Italian UNI bricks of dimensions 55 × 120 × 250 mm3 (height ×

thickness × length) and mortar joints reduced to interfaces with a Clas-
sic Mohr-coulomb failure criterion are considered. The same simulations
have been performed in Cecchi et al. (2007) using a kinematic approach.
Mechanical characteristics adopted for joints are summarized in Table 4.
In Figure 17 and Figure 18 respectively, sections Mxx − Myy and Mxx −

Mxy of the macroscopic failure surface are reported, at 4 increasing values
of nL , compared with Cecchi et al. (2007) results. The resultant failure
surfaces well approximate the upper bound reference surface for nL > 10.
hence very coarse discretizations along the thickness may be used. Results
show again a dependence on the geometrical and mechanical characteristics
assumed for the components.
nL=4
[kN*mm/mm]
3 nL=10
nL=20
2
nL=80
22
Homogenized bending moment M
1 kinematic model
-1
-2
-3
-6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6
Homogenized bending moment M11 [kN*mm/mm]
Figure 17. Mxx − Myy Failure surface obtained at increasing refinement

of the subdivision along the thickness.
2 Structural Level
The homogenized failure surface obtained with the above approaches may be
easily coupled with finite element limit analysis codes. Both upper and lower
bound approaches have been developed, for in- and out-of-plane loaded ma-
sonry walls (Milani et al., 2006b,c) with the aim of providing a full set of
numerical data for the design and/or the structural assessment of complex
structures. For in-plane loads, the finite element lower bound analysis is
based on the equilibrated triangular element bySloan (1988), while the up-
per bound is based on a modified version of the triangular element with
discontinuities of the velocity field in the interfaces by Sloan and Kleeman
(1995). The modification takes into account the actual shape of the yield
surface for the homogenized material in the interfaces.
nL=4
2
nL=10
Homogenized torque M12 [kN*mm/mm]
1.5 nL=20
nL=80
1
kinematic model
0.5
-0.5
-1
-1.5
-2
-4 -3 -2 -1 0 1 2 3 4
Homogenized bending moment M11 [kN*mm/mm]
Figure 18. Mxx − Mxy Failure surface obtained at increasing refinement

of the subdivision along the thickness.
When dealing with out-of-plane loads, the triangular plate bending el-
ement proposed independently by Hellan (1967) and Herrmann (1967) has
been adopted for lower bound calculations, whereas the triangular element
proposed by Munro and da Fonseca (1978) has been implemented and em-
ployed for the upper bound analyses.
2.1 In-plane loads: shear wall with central opening

The first structural example discussed consists on a windowed masonry
shear wall. A set of experimental tests are available from Raijmakers and
Vermeltfoort (1992). Two series were tested, corresponding to specimens
with and without openings. Here only windowed panels are considered
for the sake of conciseness. Two identical walls were tested, labeled as
J2G and J3G. The width/height ratio (L/H) of the shear walls is 990/1000
([mm]/[mm]); the panels were built up with 18 courses of bricks, from which
16 courses were active and 2 were clamped in stiff steel beams, Figure 19.
Brick dimensions are 210 × 52 × 100 mm3 and the mortar joints are 10 mm
thick. A vertical pre-compression of 0.3 N/mm2 was applied on the top and
its resultant was kept constant during the complete loading procedure. The
stiff steel beam did not allow rotations of the top and was subsequently
pushed with an increasing horizontal force.
The central opening defines two small relatively weak piers and forces the
compressive strut that develops under horizontal loading to spread around

both sides of the opening. The experimental crack pattern as reported by
Raijmakers and Vermeltfoort is illustrated in Figure 19-a.
Rotation
p precluded p
Steel Beams
L L
Panel J2G Panel J3G
Figure 19. Windowed masonry shear wall by Raijmakers and Vermeltfoort.
For the numerical analysis, a homogenized mesh constituted by 448 el-

ements is utilized. Mechanical properties assumed for the constituent ma-
terials are summarized in Table 3 and are taken in agreement with both
experimental data provided by Raijmakers and Vermeltfoort (where avail-
able) and Lourenço (1996), where the same numerical analyses have been
presented within a heterogeneous approach.
In Figure 20, a comparison among collapse load provided by the ho-
mogenized limit analysis model proposed in Milani (2011) and here re-
called, the heterogeneous approach by Lourenço (1996) and experimental
load-displacement curves (2 replicates) is presented. Furthermore, Figure
21 illustrates the limit analysis resultant deformed shape at failure (upper
bound solution) and the map of normalized plastic dissipation (representing
damage in the framework of limit analysis).
As it is possible to notice, the response provided by limit analysis seems
in good agreement with existing literature, both in terms of collapse load
predictions and crack pattern.
2.2 Out-of-plane loads: windowed panel in two-way bending

An out-of-plane validation of the homogenization models presented pre-
viously is available in Milani et al. (2006c) and Milani (2011).
The panels re-analyzed here consist of solid clay brick masonry. The
tests were carried out by Chong et al. (1994) and are denoted by SB. Four
different configurations were tested, built in stretcher bond between two stiff
abutments with the vertical edges simply supported (allowance for in-plane
50
40
External load P [kN]
30
20
Raijmakers and Vermeltfoort 1992 experimental
Raijmakers and Vermeltfoort 1992 experimental
10 Lourenço and Rots numerical 1997
Present limit analysis
0
0 5 10 15
Max displacement G [mm]

Comparison among collapse load provided by homogenized limit analysis,
a numerical heterogeneous approach and experimental load-displacement
curves.

Left: deformed shape at collapse. Right: normalized plastic dissipation
patch.
displacements was provided) and the top edge free. A completely restrained
support was provided at the base because of practical difficulties in provid-
ing a simple support. The panels were loaded by air-bags until failure with
increasing out-of-plane uniform pressure p. The reader is referred to Mi-
lani et al. (2006c) for a detailed description of geometric dimensions, loads
application, structural FE implementation and discussion of results.
Figure 22 shows typical comparisons between experimental pressure-
displacement curves by Chong et al. (1994), numerical pressure displace-
ment curves obtained by means of an orthotropic elasto-plastic macro-model
(Lourenço, 2000) and the homogenized limit analysis results (Milani et al.,
2006c). In addition, Figure 23 shows typical results of the numerical analysis
in terms of principal moment distribution and mechanisms at failure. The
agreement with experimental results is worth noting in all cases analysed.
Free Simply Clamped Panel SB02

edge supported edge edge
Panel SB01
3.0 2.5
2.34 U.B.
2.60 U.B. 2.25 L.B.
2.31 L.B. 2.0
2.0
p [kN/m2 ]
p [kN/m2 ]
1.5
University of Plymouth data University of Plymouth data

Lourenço model (1997) 1.0 Lourenço model (1997)
1.0 Proposed homogenized model Proposed homogenized model
Panel SB01/SB05
0.5
Panel SB02
0.0 5.0 10.0 15.0 20.0 25.0 0.0 5.0 10.0 15.0 20.0 25.0
Max displacement [mm] Max displacement [mm]
Figure 22. Stress field polynomial expansion approach (out-of-plane

model). Comparison between experimental and numerical results obtained,
University of Plymouth experimental tests
3 Dynamics and Seismic Behaviour

The seismic response of buildings is particularly difficult to characterize due
to its nature, the low number of strong events in a given location, the site
effects, the attenuation laws, the non-linear response of the structure, the
relevance of execution defects, and many other factors. The seismic action
is usually defined in codes via elastic response spectra, which are graphical
representations of the maximum value of the response for a single-degree-
of-freedom system as a function of the period. From the elastic response
spectrum synthetic accelerograms can be generated, which provide the time
history of accelerations at the foundations. Recorded accelerograms from
0.3
0.2
2
0.1
1.5
0
0 1
1
2 0.5
3
4
5 0
Figure 23. Typical numerical results (Panel SB02) obtained by means

of the stress field polynomial expansion approach (out-of-plane model).
(a) lower bound results (principal moments at collapse); (b) upper bound
results (deformed mesh at collapse).
real earthquakes also exist, being their most relevant characteristics the
amplitude, the frequency contents and the duration.
In case of seismic loading on unreinforced masonry buildings, it is certain
that non-linear behaviour is triggered at early stages of loading and linear
elastic analysis is not an option. Alternative options seem to be push-over
methods, as recommended in most codes for earthquake safety assessment,
or non-linear time integration methods, which provide complex and time
consuming tools hardly available for practitioners. Another much relevant
property in case of seismic loading is the presence of floors that provide
diaphragmatic action and the so-called “box-behaviour”. This possible fea-
ture is not usually present in ancient masonry buildings while being present
in modern unreinforced masonry buildings, requiring different models of
analysis, as addressed next.
3.1 Masonry structures with box behaviour

Modern masonry buildings usually adopt solutions for the slabs that
provide considerable in-plane stiffness. This is done by using monolithic
solutions for the floors, in concrete and steel, and also by establishing an ef-
fective connection between slabs and walls. Moreover, many existing build-
ings originally constructed with timber floors are capable of providing some
diaphragmatic action or have been rehabilitated by stiffening the floors and
by providing adequate connections.
The effect of floor diaphragms combined with the in-plane response of
structural walls provides box behaviour to the buildings, which usually leads
to good performance when subjected to earthquakes. The first assessment

method for seismic analysis of masonry buildings was developed under this
simple hypothesis. As result of research studies in former Yugoslavia and
the 1976 Friuli earthquake, the POR method was introduced in the Italian
region of Friuli-Venezia Giulia, to assess the seismic performance of existing
masonry buildings, see Marques and Lourenço (2011). Despite its simplicity,
this method was used until very recently in the analysis of existing buildings
and in the design of new buildings, in Italy. Its limitations, namely the
consideration of an independent storey mechanism, and the 2002 Molise
earthquake, clearly stressed the need of methods that consider the overall
response of the masonry structures.
Modern regulations, namely Eurocode 8 (CEN, 2004), seriously penalize
structural masonry with a low behaviour factor that renders unreinforced
masonry impossible, in opposition with experimental findings. Methods
inspired in the POR method and based on macro-elements have been de-
veloped, particularly in Italy, e.g. Gambarotta and Lagomarsino (1996)
and Magenes and Fontana (1998). These methods seem the most appropri-
ate for design and assessment of masonry buildings, given their widespread
in commercial software, the simplicity of modelling, the straightforward
interpretation of results and the accuracy demonstrated in different vali-
dations. The methods are incorporated in the 3Muri [www.stadata.com]
and ANDILWall/SAM II [www.crsoft.it/andilwall] computer codes, respec-
tively. While 3Muri formulation is based on the kinematic equilibrium of the
macro-elements according to the panel degrees of freedom, SAM II creates
an equivalent frame idealization for a global analysis.
The 3Muri and SAM II computer codes perform the safety verification
by a nonlinear static (pushover) analysis that simulates the evolution of the
structural condition during the earthquake, through application of incre-
mental horizontal forces until collapse. The behaviour of the structure is
represented by the so-called “capacity curve”, which represents the value of
the base shear versus the displacement of a control point (usually the mass
centroid of the roof slab). In Marques and Lourenço (2011), a benchmark
analysis was carried out demonstrating good agreement of the results for a
pushover analysis on two buildings.
To demonstrate the possibilities of construction with unreinforced ma-
sonry and the inadequacy of European codes, the seismic safety of buildings
with one up to three storeys, based on a pushover analysis carried out in
the 3Muri computer code, is considered next. The building configurations
studied are shown in Figure 24, namely a one-storey module, and two- and
three-storey buildings for semi-detached houses.
Figure 25 illustrates the ultimate response in terms of deformed configu-
Figure 24. Building configurations studied.
ration and damage of the three buildings, where it can be observed that the
collapse mechanisms are essentially induced by flexure, while plastic mech-
anisms by shear are only found for the three-storey building in spandrels
adjacent to the first slab.
Figure 25. Assessment of ultimate damage and deformed configurations.
Based on the requirements for earthquake resistance imposed by the

Italian code OPCM 3431 (2005), and assuming the seismic parameters de-
fined in the Portuguese Annex to Eurocode 8 (CEN, 2004), the possibility

to construct the studied buildings in Portugal was evaluated using 3Muri,
see Figure 26. Performing an elastic analysis adopting a behaviour factor
of 1.5, as recommended by Eurocode 8, provides a safety over conservative
assessment, as shown in Figure 27a. A better correspondence between the
pushover and linear analysis is achieved by assuming the behaviour factor
values proposed by OPCM 3431, as shown in Figure 27b. In the case of the
regular building configuration adopted, behaviour factors of 4.0, 3.0 and
3.5, respectively for the one-to-three storey buildings allow a good match
between the linear and nonlinear analysis. This example clearly demon-
strates the need of adopting adequate analysis techniques for the seismic
assessment of unreinforced masonry structures.
agR
Zone Soil (m/s2) S
1.1 A 2.50 1.00
B 1.20
agR
1.2 A 2.00 1.00 Zone Soil (m/s2)
B 1.20 S
A 1.50 1.00 A 1.70 1.00
2.3
1.3 B 1.35
B 1.20
A 1.00 1.00 A 1.10 1.00
2.4
1.4 B 1.33
B 1.30
A 0.50 1.00 A 0.80 1.00
2.5
1.5 B 1.35
B 1.30
(a) Seismic zones (b) 1-to-3 storeys

Unsafe Safe in soil type A Safe in soil types A and B
Figure 26. Safety mapping in Portuguese seismic zones by a pushover

analysis.
3.2 Masonry structures without box behaviour

Differently from the structures considered in the previous section, unre-
inforced masonry structures without box behaviour have shown poor per-
formance in many past earthquakes. The reasons for the poor performance
are the inherent brittleness, lack of tensile strength, lack of ductility, flexible
floor diaphragms and lack of connection between the structural elements.
Next, illustrative examples of different analysis tools for the seismic analysis
of these structures are presented, see (Lourenço et al., 2011) for additional
details.
The first example is the S. Torcato church in Portugal, where limit anal-
ysis using macro-blocks was carried out for the seismic performance assess-
ment, as the church exhibits significant damage and requires strengthening.
(a) Portuguese Annex to EC8 (b) OPCM 3431

Unsafe Safe in soil type A Safe in soil types A and B
Figure 27. Safety mapping in Portuguese seismic zones by a linear elastic

analysis.
In existing masonry buildings partial collapses often occur due to seismic

action, generally, with the loss of equilibrium of rigid bodies. In this case
study, four mechanisms were defined, based on the inspection and structural
analysis of the structure. The mechanisms were partially marked by existing
cracks. Figure 28 shows the mechanisms considered in the limit kinematic
analysis. According to the limit analysis, the church is safe and the lowest
safety factor is equal to 1.69 (with overturning of the tympanum).
This analysis method is conceptually simple and an abacus of possible
mechanisms is available at MIBAC 2007 (2007). In the present case, the
method is easy to apply as the collapse mechanisms are also partially de-
fined by existing cracking. It is believed that the benefits of using collapse
mechanism analysis are the following: (a) the method is intuitive and does
not requires advanced knowledge of physics or mechanics, being therefore at
reach of most practitioners; (b) the abacus of possible collapse mechanisms
and the observation of previous collapses under earthquake action provide
are the basis of the inductive approach; (c) the method is conceptually and
analytically correct, if the proper collapse mechanisms are selected. It is
believed also the method possesses a strong drawback: if wrong collapse
mechanisms are selected, the seismic assessment (and related strengthening
measures, if applicable) is meaningless. Therefore, practitioners must en-
sure correct selection of collapse mechanism, either by a detailed inspection
of the structure being studied or by adopting more sophisticated analysis
methods. This is certainly the case of complex or unusual structures, for
which the mechanisms might not be obvious. Another example would be to
use pushover analysis or time integration analysis to get more confidence on
Figure 28. Possible collapse mechanisms: (a) overturning of the left

tower; (b) overturning of the right tower; (c) overturning of the façade;
(d) overturning of the tympanum. (FS is the safety factor).
collapse mechanisms and then adopt the correct collapse mechanism analy-
sis to calculate the required strengthening.
A second example of application is the Qutb Minar, in New Deli, India.
To evaluate its seismic performance different techniques of structural anal-
yses were used, namely non-linear dynamic analysis and non-linear static
analysis (pushover analysis). In the analyses different numerical models
were considered. Two models were prepared using the Finite Element
Method (FEM), both are three-dimensional models but one uses 3D solid el-
ements (Solid Model) while the other one was performed with 3D composite
beams (Beam Model). A simplified in-plane model of the minaret based on
the Rigid Element Method was also developed. The Rigid Element Method
idealizes the masonry structure as a mechanism made of rigid elements and
springs (Casolo and Peña, 2007). The numerical models were updated from
dynamic identification tests.
In the FEM models, the physical non-linear behaviour of the masonry
was simulated using the Total Strain Crack Model detailed in DIANA
(2005), with non-linear behaviour given by a parabolic law in compression

and an exponential law in tension (fixed crack model with variable shear
retention). In the rigid body and spring model (RBSM), the constitutive
law for axial springs is parabolic in compression and bi-linear in tension
with softening. A Mohr-Coulomb law was considered for shear springs in
order to relate the shear stresses with the axial stresses.
The dynamic analyses were carried out using five artificial accelerograms
compatible with the elastic response spectrum for Delhi (PGA = 0.20g).
Pushover analyses were carried out considering a uniform acceleration dis-
tribution. The load was applied with increasing acceleration in the hori-
zontal direction and a control point at the top of the tower was considered.
Figure 29a shows the capacity curves (lateral displacement — seismic co-
efficient at the base level). Similar behaviour was found with the different
models. It can be observed that the average seismic factor is 0.20 and the
minaret collapses by overturning at the base.
Figure 29. Results of the pushover analyses: (a) capacity curves of the
pushover analyses proportional to the mass; (b) comparison between the
drifts obtained through modal pushover analysis and trough dynamic anal-
yses of the Beam and RBSM Models.
In order to study the influence of the distribution of the lateral load in

the pushover analysis, additional non-linear static analyses were performed.
Four different configurations of lateral loads were considered: (a) linear dis-
tribution of the displacement along the height; (b) loads proportional to
the first modal shape; (c) adaptive pushover analysis, changing the load
distribution according to the changes in the first modal shape during the
analysis; (d) modal pushover analysis (Chintanapakdee and Chopra, 2003).
The results of the pushover analyses, namely the failure mode and dis-
placements’ distribution along the height, are not in agreement with the
non-linear dynamic analysis. Even modal pushover analysis, in which the
responses of the first seven modes were combined, is not able to simulate the
amplification of the response at higher levels (Figure 29b). This example
demonstrates again the need of adopting adequate analysis techniques for
the seismic assessment of unreinforced masonry structures.
4 Conclusions
Constraints to be considered in the use of advanced modelling are the cost,
the need of an experienced user/engineer, the level of accuracy required, the
availability of input data, the need for validation and the use of the results.
As a rule, advanced modelling is a necessary means for understanding
the behaviour and damage of (complex) historical masonry constructions
and examples have been addressed here. For this purpose, it is necessary
to have reliable information on material data, and recommendations are
provided in this chapter.
Micro-modelling techniques for masonry structures allow a deep under-
standing of the mechanical phenomena involved. For large scale applica-
tions, macro-block approaches or average continuum mechanics must be
adopted and homogenization techniques represent a popular and active field
in masonry research. Homogenization techniques represent a popular and
active field in masonry research. Several approaches have been recently in-
troduced by different authors and a first attempt to catalogue them and to
discuss pros and cons are carried out in this Chapter. Even if it impossible
to predict the future of masonry research, this Chapter addresses in detail
two different static approaches considered particularly relevant. The first
approach is based on a polynomial expansion of the stress field coupled with
limit finite elements analysis, whereas the second relies into a discretization
of the unit cell by means of a few constant stress finite elements (CST) with
joints reduced to interfaces.
Finally, the possibilities of assessment unreinforced masonry structures
subjected to seismic loading is addressed using different techniques. It is ad-
vocated that linear elastic analysis can hardly be used, as masonry features
low tensile strength, and different models must be used in the presence or
absence of adequately connected floors, the so-called box behaviour.
In case of box behaviour the available methods have been briefly re-
viewed. Their performance is good and the knowledge is sound, with some
corrections needed in the recent European regulations (Eurocode 8). When
box behaviour cannot be guaranteed, the analysis of masonry structures
becomes rather complex. The use of macro-models and limit analysis seems
the current trend but difficulties arise in the practical use, namely with re-
spect to validation of the hypothesis of the user and the risk of selecting
inadequate failure mechanisms. The non-linear static analysis could be a
good and easily understood approach, because it is based on the simple
evaluation of the requested deformation with respect to the displacement
capacity of the building. This approach is in agreement with the modern
provisions for structural assessment. Still, the results obtained from the
non-linear static and dynamic analyses indicate quite different response of
these structures to earthquakes. It is therefore concluded that non-linear
pushover analysis does not simulate correctly the failure mode of masonry
structures without box behaviour, even if higher modes are considered via
modal pushover analysis.
Bibliography
E. Anderheggen and H. Knöpfel. Finite element limit analysis using linear
programming. International Journal of Solids and Structures, 8(12):
1413–1431, 1972.
A. Anthoine. Derivation of the in-plane elastic characteristics of masonry
through homogenization theory. International Journal of Solids and
Structures, 32(2):137–163, 1995.
A. Anthoine. Homogenization of periodic masonry: plane stress, generalized
plane strain or 3d modelling? Communications in numerical methods in
engineering, 13(5):319–326, 1997.
T. Belytschko and P.G. Hodge. Plane stress limit analysis by finite elements.
Journal of the Engineering Mechanics Division, 96(6):931–944, 1970.
D. Caillerie. Thin elastic and periodic plates. Mathematical Methods in the
Applied Sciences, 6(1):159–191, 1984.
S. Casolo and F. Peña. Rigid element model for in-plane dynamics of ma-
sonry walls considering hysteretic behaviour and damage. Earthquake
engineering & structural dynamics, 36(8):1029–1048, 2007.
CEB-FIP. Model Code 90. Thomas Telford Ltd., UK, 1993.
A. Cecchi and K. Sab. A multi-parameter homogenization study for mod-
elling elastic masonry. European Journal of Mechanics-A/Solids, 21(2):
249–268, 2002.
A. Cecchi, G. Milani, and A. Tralli. Validation of analytical multiparameter
homogenization models for out-of-plane loaded masonry walls by means
of the finite element method. Journal of engineering mechanics, 131(2):
185–198, 2005.
A. Cecchi, G. Milani, and A. Tralli. A Reissner–Mindlin limit analysis

model for out-of-plane loaded running bond masonry walls. International
journal of solids and structures, 44(5):1438–1460, 2007.
CEN. Eurocode 8: Design of structures for earthquake resistance, Part 1:
General rules, seismic action and rules for buildings. European Com-
mittee for Standardization, Belgium, 2004.
CEN. Eurocode 6: Design of masonry structures, Part 1–1: General rules
for reinforced and unreinforced masonry structures. European Commit-
tee for Standardization, Belgium, 2005.
C. Chintanapakdee and A.K. Chopra. Evaluation of modal pushover analy-
sis using generic frames. Earthquake engineering & structural dynamics,
32(3):417–442, 2003.
V.L. Chong, C. Southcombe, and I.M. May. The behaviour of laterally
loaded masonry panels with openings. In 3th Int. Masonry Conf. Proc.
Brit. Mas. Soc., pages 178–182, London, 1994.
P. de Buhan and G. de Felice. A homogenization approach to the ultimate
strength of brick masonry. Journal of the Mechanics and Physics of
Solids, 45(7):1085–1104, 1997.
G. de Felice. Metodi di omogeneizzazione per sistemi regolari di corpi rigidi.
In Proceedings of the XII AIMETA Congress. Naples, Italy, pages 3–6,
1995.
DIANA. Displacement method ANAlyser, 2005. release 9.1, CD-ROM,
Netherlands.
L. Gambarotta and S. Lagomarsino. On the dynamic response of masonry
walls (in Italian). In Proc. National Congress “La Meccanica delle Mu-
rature tra Teoria e Progetto”, Messina, 1996.
K. Hellan. Analysis of elastic plates in flexure by a simplified finite element
method. Number 46 in Acta polytechnica Scandinavica. Norges tekniske
vitenskapsakademi, 1967.
L.R. Herrmann. Finite-element bending analysis for plates. Journal Engi-
neering Mechanics Division, 93:13–26, 1967.
H.K. Hilsdorf. Investigation into the failure mechanism of brick masonry
loaded in axial compression. In F.H. Johnson, editor, Designing, en-
gineering and constructing with masonry products, pages 34–41. Gulf
Publishing Company, Houston, Texas, 1969.
G. Hoffman and P. Schubert. Compressive strength of masonry parallel
to the bed joints. In N.G. Shrive and A. Huizer, editors, Proc. 10th
Int. Brick and Block Masonry Conf., pages 1453–1462, Calgary, Alberta,
1994.
K. Krabbenhoft, A.V. Lyamin, M. Hjiaj, and S.W. Sloan. A new discontin-
uous upper bound limit analysis formulation. International Journal for
Numerical Methods in Engineering, 63(7):1069–1088, 2005.
J. Lopez, S. Oller, E. Onate, and J. Lubliner. A homogeneous constitu-

tive model for masonry. International journal for numerical methods in
engineering, 46(10):1651–1671, 1999.
P.B. Lourenço. Computational strategies for masonry structures. PhD the-
sis, Delft University of Technology, the Netherlands, 1996. Available from
www.civil.uminho.pt/masonry.
P.B. Lourenço. On the use of homogenisation techniques for the analysis of
masonry structures. Masonry International, 11(1):26–32, 1997.
P.B. Lourenço. Experimental and numerical issues in the modelling of the
mechanical behaviour of masonry. In P. Roca and et al., editors, Proc.
Structural analysis of historical constructions II, CIMNE, pages 57–91,
Barcelona, 1998a.
P.B. Lourenço. Sensitivity analysis of masonry structures. In Proc. 8th
Canadian Masonry Symp., pages 563–574, Jasper, Canada, 1998b.
P.B. Lourenço. Anisotropic softening model for masonry plates and shells.
Journal of Structural Engineering, 126(9):1008–1016, 2000.
P.B. Lourenço. Computations on historic masonry structures. Progress in
Structural Engineering and Materials, 4(3):301–319, 2002.
P.B. Lourenço and J.L. Pina-Henriques. Validation of analytical and con-
tinuum numerical methods for estimating the compressive strength of
P.B. Lourenço and L.F. Ramos. Characterization of cyclic behavior of dry
masonry joints. Journal of Structural Engineering, 130(5):779–786, 2004.
P.B. Lourenço and J.G. Rots. A multi-surface interface model for the anal-
ysis of masonry structures. Journal of engineering mechanics, 123(7):
660–668, 1997.
P.B. Lourenço, J.O. Barros, and J.T. Oliveira. Shear testing of stack bonded
masonry. Construction and Building Materials, 18(2):125–132, 2004.
P.B. Lourenço, J.C. Almeida, and J.A. Barros. Experimental investigation
of bricks under uniaxial tensile testing. Masonry International, 18(1):
11–20, 2005.
P.B. Lourenço, G. Milani, A. Tralli, and A. Zucchini. Analysis of masonry
structures: review of and recent trends in homogenization techniques
this article is one of a selection of papers published in this special issue
on masonry. Canadian Journal of Civil Engineering, 34(11):1443–1457,
2007.
P.B. Lourenço, N. Mendes, L.F. Ramos, and D.V. Oliveira. Analysis of
masonry structures without box behavior. International Journal of Ar-
chitectural Heritage, 5(4-5):369–382, 2011.
R. Luciano and E. Sacco. Homogenization technique and damage model for
old masonry material. International Journal of Solids and Structures, 34
(24):3191–3208, 1997.
G. Magenes and A.D. Fontana. Simplified non-linear seismic analysis of

masonry buildings. In Proc. British Masonry Society, number 8, pages
190–195, 1998.
G. Maier, E. Papa, and A. Nappi. On damage and failure of unit masonry.
In Proc. Experimental and numerical methods in earthquake engineering,
pages 223–245, Brussels and Luxembourg, 1991. Balkema Editions.
R. Marques and P.B. Lourenço. Possibilities and comparison of structural
component models for the seismic assessment of modern unreinforced
masonry buildings. Computers & Structures, 89(21):2079–2091, 2011.
T.J. Massart. Multi-scale modeling of damage in masonry structures. PhD
thesis, University of Bruxelles, Belgium, 2003.
T.J. Massart, R.H.J. Peerlings, and M.G.D. Geers. Mesoscopic modeling
of failure and damage-induced anisotropy in brick masonry. European
Journal of Mechanics-A/Solids, 23(5):719–735, 2004.
N. Mendes. Seismic assessment of ancient masonry buildings: Shaking table
tests and numerical analysis. PhD thesis, University of Minho, Portugal,
2012.
MIBAC 2007. Guidelines for the evaluation and mitigation of seismic risk to
cultural heritage. Ministry for Cultural Heritage and Activities, Gangemi
Editore, Rome, Italy, 2007.
G. Milani. Simple lower bound limit analysis homogenization model for
in-and out-of-plane loaded masonry walls. Construction and Building
Materials, 25(12):4426–4443, 2011.
G. Milani, P.B. Lourenço, and A. Tralli. Homogenised limit analysis of
masonry walls, part I: Failure surfaces. Computers & structures, 84(3–
4):166–180, 2006a.
G. Milani, P.B. Lourenço, and A. Tralli. Homogenised limit analysis of
masonry walls, part II: Structural examples. Computers & structures, 84
(3–4):181–195, 2006b.
G. Milani, P.B. Lourenço, and A. Tralli. Homogenization approach for the
limit analysis of out-of-plane loaded masonry walls. Journal of structural
engineering, 132(10):1650–1663, 2006c.
J. Munro and A.M.A. da Fonseca. Yield line method by finite elements and
linear programming. Structural Engineer, 56(2), 1978.
P.C. Olsen. The influence of the linearisation of the yield surface on the
load-bearing capacity of reinforced concrete slabs. Computer methods in
applied mechanics and engineering, 162(1):351–358, 1998.
OPCM 3431 . Technical regulations for the design, assessment and seis-
mic adaptation of buildings. Gazzetta Ufficiale Serie Generale n.107 del
10/05/2005, Italy, 2005.
G.N. Pande, J.X. Liang, and J. Middleton. Equivalent elastic moduli for
brick masonry. Computers and Geotechnics, 8(3):243–265, 1989.
M. Paulo-Pereira. Assessment of the seismic performance of building en-

closures (in Portuguese). PhD thesis, University of Minho, Portugal,
2012.
P. Pegon and A. Anthoine. Numerical strategies for solving continuum
damage problems with softening: application to the homogenization of
PIET-70. P.I.E.T. 70 Masonry work. Prescriptions from Instituto Ed-
uardo Torroja. Consejo Superior de Investigaciones Científicas, Madrid
(in Spanish), 1971.
S. Pietruszczak and X. Niu. A mathematical description of macroscopic be-
haviour of brick masonry. International journal of solids and structures,
29(5):531–546, 1992.
T.M.J. Raijmakers and A.T. Vermeltfoort. Deformation controlled tests in
masonry shear walls. Report B-92, 1156, 1992. TNO-Bouw,Delft, The
Netherlands (in Dutch).
J.G. Rots, editor. Structural masonry: An experimental/numerical basis for
practical design rules. Balkema, Rotterdam, the Netherlands, 1997.
P. Schubert. The influence of mortar on the strength of masonry. In J.W.
de Courcy, editor, Proc. 8th Int. Brick and Block Masonry Conf., pages
162–174, London, 1988. Elsevier Applied Science.
S.W. Sloan. Lower bound limit analysis using finite elements and linear pro-
gramming. International Journal for Numerical and Analytical Methods
in Geomechanics, 12(1):61–77, 1988.
S.W. Sloan and P.W. Kleeman. Upper bound limit analysis using discon-
tinuous velocity fields. Computer Methods in Applied Mechanics and
Engineering, 127(1):293–314, 1995.
P. Suquet. Analyse limite et homogeneisation. Comptes Rendus de
l’Academie des Sciences-Series IIB–Mechanics, 296:1355–1358, 1983.
D.J. Sutcliffe, H.S. Yu, and A.W. Page. Lower bound limit analysis of
unreinforced masonry shear walls. Computers & Structures, 79(14):1295–
1312, 2001.
M. Tomazevic. Earthquake-resistant design of masonry buildings. Imperial
College Press, London, 1999.
R. van der Pluijm. Out-of-plane bending of masonry: Behavior and Strenght.
PhD thesis, Eindhoven University of Technology, The Netherlands, 1999.
G. Vasconcelos and P.B. Lourenço. In-plane experimental behavior of stone
masonry walls under cyclic loading. Journal of structural engineering,
135(10):1269–1277, 2009.
G. Vasconcelos, P.B. Lourenço, C.A.S. Alves, and J. Pamplona. Experimen-
tal characterization of the tensile behaviour of granites. International
journal of rock mechanics and mining sciences, 45(2):268–277, 2008.
A. Zucchini and P.B. Lourenço. A micro-mechanical model for the ho-

mogenisation of masonry. International Journal of Solids and Structures,
39(12):3233–3255, 2002.
A. Zucchini and P.B. Lourenço. A coupled homogenisation–damage model
for masonry cracking. Computers & structures, 82(11):917–929, 2004.

Angelillo Mechanics of Masonry Structures PDF

Uploaded by

Copyright:

Available Formats

Angelillo Mechanics of Masonry Structures PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Angelillo Mechanics of Masonry Structures PDF

Uploaded by

Copyright:

Available Formats

What is the book about?

What is the book about?

What topics are covered in the book?

What topics are covered in the book?

CISM International Centre for Mechanical Sciences 551

Courses and Lectures

The Secretary General

The series presents lecture notes, monographs, edited works and

For further volumes:

© CISM, Udine 2014

All contributions have been typeset by the authors

Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)

It is maybe a trivial remark saying that the vast majority of masonry

1. Masonry behaviour and modelling

Basic behaviour of masonry and simpliﬁed unilateral

Masonry behavior of regularly arranged masonry struc-

Vector valued measures 38

The existence of equilibrium states 46

Families of measures and the weak compatibility of

3. A numerical method for solving BVP of masonry-like

4. Practical applications of unilateral models to Ma-

Basic tools 109

Model zero (RNT) 118

Model one (NENT) 157

Model two (ML) 187

II Reﬁned Models 211

Some comments on experimental results 216

Masonry structures without mortar: dry friction mod-

Masonry blocks with mortar 226

Micromechanical modeling 242

Multiscale modeling 255

Macromechanical modeling 272

7. Homogenization and Seismic Assessment: Review

Homogenization Theory, Basic Assumptions 293

Structural Level 323

Dynamics and Seismic Behaviour 327

Maurizio Angelillo‡ , Paulo B. Lourenço *

Abstract In this Chapter we present the basic experimental facts

M. Angelillo (Ed.), Mechanics of Masonry Structures, CISM International Centre for

2 Basic behaviour of masonry and simpliﬁed

2.1 Local failure modes

the “good” ones, that is those contributing to the accommodation and

Figure 1. Fracture of detachment in brick walls at diﬀerent scales

2. The second one is a kind of mixed mode in which fractures of de-

Figure 2. Detachment and sliding due to combined compression and shear.

2.2 Structural failure mechanisms

Figure 3. Crushing due to compression.

Figure 4. Out of plane rocking.

Figure 5. In plane failure due to compression and shear.

2.3 Experimental observations: results of typical tests

of masonry behaviour that can be captured by the simple models.

Masonry Type σc ( daN

be represented by the graph of Figure 9a.

2.4 Simpliﬁed uniaxial models

in incompressible materials). This model requires no material parameters

E, since strength in compression is assumed to be inﬁnite, whilst strength

3 Masonry behavior of regularly arranged masonry

The diﬃculties in performing advanced testing of this type of structures