You are on page 1of 466

INTERNATIONAL CENTRE FOR MECHANICAL SCIENCES

COURSES AND LECTURES - No. 302

NONSMOOTH MECHANICS
AND APPLICATIONS

EDITED BY
J.J. MOREAU
UNIVERSITY OF MONTPELLIER

P.O. PANAGIOTOPOULOS
ARISTOTLE UNIVERSITY
AND
R.W.T.H.

SPRINGER-VERLAG WIEN GMBH


Le spese di stampa di questo volume sono in parte coperte da contributi
dei Consiglio Nazionale delle Ricerche.

This volume contains 77 illustrations.

This work is subject to copyright.

All rights are reserved,

whether the whole or part of the material is concerned

specifically those of translation, reprinting, re-use of illustrations,

broadcasting, reproduction by photocopying machine

or similar means, and storage in data banks.

Cl 1988 by Springer-Verlag Wien

Originally published by Springer Verlag Wien-New York in 1988

ISBN 978-3-211-82066-7 ISBN 978-3-7091-2624-0 (eBook)


DOI 10.1007/978-3-7091-2624-0
PREFACE

This book is devoted to problems of Mechanics involving nonsmooth relations.


In place of classical derivatives current concepts of Nonsmooth Analysis are used, such as
subdifferentials, generalized gradients,fans or quasidifferentials. The corresponding problems
may take the form of variational or hemivariational inequalities or of dif./erential and integral
inclusions. Other mechanical topics investigated in the present volume lead to the
determination of saddle points for non-differentiable functions.
In Dynamics, velocity is not supposed to be a smooth function of time, but only to have
locally bounded variation, so that acceleration is a vector-valued measure on the concerned
interval of time. Evolution is thus governed by measure differential equations or by measure
differential inclusions. The formalism used makes easy to discretize for numerical treatment.
Generally, the practice/ Nonsmooth Analysis requires the handling ofmultivalued mappings
(i.e. set-valued mappings, also called multifunctions). In Mechanics, this concept has for long
been implicit in such classical domains as Plasticity of Dry Friction. Its use in a formalizedway
provides a new insight into these subjects and concurrently proves to be operationally efficient.
The same is true for a number of mechanical or thermodynamical topics.
Chapter l by J.J. Moreau, concerns the Dynamics ofsystems with afinite number ofdegrees
of freedom, involving unilateral contact whereupon possible friction is assumed to obey
Coulomb's law. New formulations of the corresponding evolution problems are developed
which lead to effective numerical algorithms. These numerical techniques in particular prove
effective in the handling of the catastrophic events which may be manifested in the presence of
dry friction.
Chapter Il by P. D. Panagiotopoulos, deals with mechanicalproblems admitting variational
formulations in terms of Hemivariational Inequalities. The notions of substationarity and
quasidifferentiability are introduced and some new nonclassical variational principles in
inequality form are presented.
Chapter Ill by M. Fremond, develops the use of nonsmooth potentials in reversible or
irreversible Thermodynamics. This provides new approaches to subjects such as Phase
Changes, Heterogeneous Media, Soil Freeze, Adhesions of Solids or Shape-memory Alloys.
Chapter IV, by J. Has linger, is devoted to the Numerical Analysis of Unilateral Problems of
Deformable Bodies and Shape Optimization. The approach is based on the formulation of the
contact boundary conditions, with friction taken into account, through saddle point properties
or variational and quasi variational inequalities.
Chapter V, by P.M. Suquet, emphasizes first that discontinuous solutions should normally
be expected in the Yield Design (Limit Analysis) of Plastic Structures and in the quasi-static
evolution of Elastop/astic Bodies. The treatment of a bond between two deformable solids as
the limit, in a mathematical sense, of a thin layer with nonlinear behaviour is presented. The
chapter introduces the use of the concept of epi-convergence in homogenization, i.e. the
derivation of macroscopic constitutive laws for microscopically heterogeneous bodies.
Finally, in Chapter VI, J.J. Te/ega, applies duality methods to the nonlinear boundary value
problems one meets when studying the equilibrium or the quasistatic evolution of elastic and
inelastic bodies submitted to unilateral contacts. More elaborate models of friction than
Coulomb~ law are presented and also the homogenization of microfissured elastic solids and
plates, with unilaterality is considered.
In short, the book demonstrates how several currently developed mathematical tools of
Nonsmooth Analysis effectively work in Mechanics and Engineering Science, with definite
advantages from the standpoint of fundamental understanding as well as practical
computation. It gathers in one volume both some innovative ideas and a good deal of
unpublished material.

J.J. Moreau, P.D. Panagiotopou/os


CONTENTS

Page

Preface

Unilateral Contact and Dry Friction in Finite Freedom Dynamics


by J.J. Moreau ............................ ............................ ........ . I

Nonconvex Superpotentials and Hemivariational Inequalities. Quasi


Differentiability in Mechanics
by P. D. Panagiotopoulos ............................ .......................... 82

Contact with Adhesion


by M. Fremond ............................ ............................ ...... 177

Approximation of Contact Problems. Shape Optimization in Contact Problems


by J. Haslinger ............................ ............................ ....... 223

Discontinuities and Plasticity


by P.M. Suquet ........................... ........................... ...... 279

Topics on Unilateral Contact Problems of Elasticity and Inelasticity


by J.J. Telega ............................ ............................ ...... . 341
UNILATERAL CONTACT AND DRY FRICTION
IN FINITE FREEDOM DYNAMICS

J.J. Moreau
Universite des Sciences et Techniques du Languedoc
Montpellier, France

ABSTRACT
An approach to the dynamics of mechanical systems with a finite number of
degrees of freedom, involving unilateral constraints, Is developed. In the
n-dimensional linear spaces of forces and velocities, some classical con-
cepts of Convex Analysis are used, but no convexity assumption is made
concerning the constraint Inequalities. The velocity is not supposed to be a
differentiable function of time, but only to have locally bounded variation, so
the role of the acceleration is held by a n-dlmenslonal measure on the consi-
dered time interval. Dynamics is then governed by measure differential
inclusions, which treat possible velocity jumps on the same footing as
smooth motions. Possible collisions are described as soft, thus dissipative.
Friction is taken into account under a recently proposed expression of
Coulomb's law. These formulations have the advantage of generating
numerical algorithms of time-discretization, able to handle, In particular,
the nonsmooth effects arising from unllaterallty and from dry friction.
2 J.J. Moreau

1.1NTRODUCTION
Usual mechanisms consist of parts which, at the first level of
Investigation, are treated as perfectly rigid bodies. The mechanism operation
rests on the fact that some of these parts may come Into contact or get
loose from each other, but none of them can overlap. Similarly, the parts may
touch the external bodies which support the mechanism, but can never
encroach upon the region of space they occupy.
In terms of the parameters q1, ... , qn, making an element of IRn denoted In
the sequel by q, which are used to locate every position or the considered
system, the above Impenetrability properties may as a rule be expressed by a
set of Inequalities, say foc(q)<O, ocE{ I ,... ,y}. Each of these inequalities
corresponds to what is traditionally called a l/nilateral constraint
Naturally, the description of a constraint in Mechanics does not reduce to the
geometric restriction It Imposes to the system possible positions. Pre-
dicting the system behaviour always requires some additional information
about the forces of constraint or reactions needed by the system dyna-
'mlcs, for the geometric conditions to be satisfied at every Instant.
Constraints whose geometric effect Is expressed by equalities are, In
contrast, said bilateral They are commonly realized by the conjunction of
several unilateral constraints and, In practice, this may leave a residual
looseness whose Investigation has primary Importance in some applications.
However omnipresent unilateral constraints are In machines, the place
they receive In the books on Classical Mechanics Is very modest. Here Is the
traditional approach of the situation.
Starting with a position of the system In which some of the contacts
are effective (I.e. the corresponding Inequalities hold as equalities> and
Unilateral Contact and Dry Friction 3

with velocities compatible with the persistence of these contacts, the


subsequent motion is calculated under the tentative assumption that all the
said contacts do persist. The calculation is identical to what is more
familiarly done in the case of bilateral constraints, with friction possibly
taken into account. At every instant of the calculated motion, the respective
contact forces are evaluated. As long as the direction of each of these
vectors is found compatible with the unllaterality of the corresponding
contact, the calculated solution is accepted (rigorously, this does not
dispense from investigating also the posslbllty of contact breaking, since
the uniqueness of solution to an initial value problem of Dynamics is not
granted in genera D.
But If the above calculation yields, after an Instant t, a non feasible
value for some of the contact forces, the tentative assumption has to be
rejected from this instant onward, and other types of motions, in which
some of the contacts initially present get loose, are to be tested in the same
way. The number of combinations to be tried may be high, If many unilateral
contacts are involved. In practical situations, such Instants as t make a
finite set, but this cannot be asserted in general.
It was not before the first quarter of this century that E. Delassus
(cf.[ 1] for the frictionless case) observed that, contrary to what had been
formerly belleved, the contacts which get loose at time 1 are not
necessarily those for which the calculation, performed at t> 1, yields
contact forces of non feasible direction. Delassus papers seem today rather
intricate; a clearer account of his arguments may be found in [2] . More
recently [3][4]. the same question has been revisited, by the means of
elementary Convex Analysis and Quadratic Programming.
4 J.J. Moreau

The present lectures develop a novel approach to the dynamics of


systems Involving un11ateral constraints. Here are the dominent features.

1 The function t-.q(t>EIRn describing the Investigated motion on a time


Interval I, with Initial instant t 0, Is not a priori assumed differentiable
everywhere. Instead, one supposes that q equals the time Integral of a
velocity function t-.u<t)EIRn with locally bounded variation on the interval;
notation: uelbvO,IRn). Classically, such a function u may have discon-
tinuities but, for every t In the Interior of I , the existence of the
rlgnt-1/mtt u+<t> and of the left-limit u-<t> Is secured <see Sec.2 below
for a convention concerning the case t=t 0 >. These limits equal the respective
one-side derivatives of the function q: 1--.IRn at point t.

2 In view of these discontinuities, the existence of the acceleration q"=u


cannot be expected everywhere. But, with every uElbv(I,IR0 ), one classically
associates an R0 -valued measure [5][6] on the Interval I, called In the
sequel the differential measure [7] of u and denoted by du.
The function u is locallyabsolutelycontlnuousif and only If the vector
measure du possesses a density function, say uiEL \)Co, dt; R0 ), relative
to the Lebesgue measure on the interval I. We denote the latter measure by
dt ; this Is in fact the differential measure of the real function t-.t, which
evidently belongs to lbvCI,IR). A function u of this sort may constltute a
solution to a differential equation, In the classical sense of Caratheodory.
Here Is another special case: suppose that, for some tEint I, one has
u-(t);tU+(t). Then, the IR 0 -valued measure du possesses at point 1: an
atom with value u+Ct)-U-(t). This value Is an element of IR0 that we shall
Unilateral Contact and Dry Friction 5

call the jump of u at Instant 1 .

In general, a function uElbv(I,IRn) may be a solution to some measure


differential equation, a notion about which the reader could find some
information In [8].

3 Velocity functions with locally bounded variation make the setting in


which we develop the Nonsmooth Dynamics of mechanical systems w1th a
finite number of degrees of freedom. This is governed by an extensiOn of
Lagrange equations that we introduce in Sec.7. It Includes as a special case
the traditional equations of the Dynamics of Percussions. Concerning the
connection of this general formulation with the classical principles of
Dynamics, some details may be found In [9].

4 The set of inequalities f(X~O <with f(XEC 1and \lf(X;z:Q) defines in IRn the
feasible region, denoted by 4> and assumed in the sequel independent of time
If a motion t-+q(t) is described in the above terms and if Q(0E4> for every t,
one elementarily finds (see Sec.2) that u+(t) belongs to a certain polyhedral
conic convex subset of IR", denoted by V(q(t)). This is the tangent cone. to the
region 4> at point q(t), equal in particular to the whole of IR" when q(t) is
Interior to 4>. Actually, a cone denoted by V(q), and Its polar cone N(q) in
the sense of the standard scalar product of IR", will In the sequel be defined
even for qt4>. When QE4>, the cone N(q) Is nothing but the <outward) normal
cone to 4> at this point <reduced to {0} if q is interior to 4>).

so The mechanical formulation of unilateral constraints has to encompass


the geometric condition VtEI : Q(0E4>, together with some lnfomatlon about
6 J.J. Moreau

the associated forces of constraint In the framework of traditional


<smooth) Analytical Dynamics, this system of forces Is represented, for
every t, by its covariant components, say r 1,. .. , r 0 , relative to the
generalized coordinates in use. This makes an element of IRn that we shall
denote by r
The simplest case is that of frictionless contact. This classically
means that the force of constraint at every possible point of contact is
normal to the concerned bodies, with direction agreeing with unilaterality.
One elementarily finds <see Sec.S below> that, if all the considered unila-
teral constraints are of this sort, the whole Information about them lets
itself be summarized into the writing
\ftEI : q(t)EcJ? and -r(t)EN(q(t)). (I. 1)

(About the concept of a frictionless contact In the case of a less regular


feasible region than above, see [I 0].)
Starting from ( 1.1 ), a decisive observation is made in Sec.S, namely
that, for smooth motions, it implies the stronger assertion
-r<t> E a'~~v<q<m<u\t>>. <1.2)
According to the usual notations of Convex Analysts, the right-hand member
equals the normal cone at point u+(t) to the convex subset V(q(t)) of IR 0 .
In addition, it is established that, if the initial data satisfy q(t0 )E4>,
then ( 1.2), assumed to hold for (dt-almost) every t, secures that Q(t) will
remain In 4>.
The advantage of ( 1.2) over< 1.1 >lies in the following. First, as we shall
develop in Sec.S, this writing directly suggests some algorithms of tlme-
dlscretlzatlon for computing the solutions to initial value problems.
Secondly, by entering the velocity Into the contact law, it paves the way to
Unilateral Contact and Dry Friction 7

the consideration of friction. Furthermore, It Is easily generalized to


Nonsmooth Dynamics.

6 The function t-+r(t)EIRn which, in the traditional Lagrange equations,


represents the forces of constraints has, in Nonsmooth Dynamics, to be re-
placed by an IR0 -valued measure on the time-Interval I, called the contact
Impulsion and denoted by dR. For smooth motions, this measure admits the
above function as its density relative to Lebesgue measure. A priori, there
exists an Infinity of representations of a vector measure, such as dR, In the
form dR=Rv dbl , where dbl is a nonnegative real measure and R~ a vector-
valued locally dv-integrable density function. We shall admit, as the Jaw of
frictionless contact in Nonsmooth Dynamics, the following generalization of
( 1.2), to be satisfied for every t in I,
-R~(t) E a,V(q(t))(u+(t)). ( 1.3)

Because the right-hand side Is a cone, one shows that t!Jis condition Is
Indifferent to t!Je c/Joice of tile base measure dbJ. Furthennore, the existence
of a function R~ verifying< 1.3) implies that u+(t> belongs to V(q(t)). Through
Prop.2.4 below, this ensures q(t)E4l for every t, provided the Initial data
satisfy q<t0>eP,.
Assertion <I .3) about the contact impulsion makes the definition of the
class of unilateral constraints that we cal! frictionless and soft. When
transported Into the equality of IR0 -valued measures, which governs
Nonsmooth Dynamics, It generates a measure differential Inclusion. The
existence of solutions to the resulting Initial value problems has so far been
established only In some special cases [II][ 12][ 13] and Is currently under
investigation. The velocity Jumps possibly occurtng In such solutions are of
the sort the author has previously called "standard Inelastic shocks" [ 14]
8 J.J. Moreau

[IS]. These are dissipative, so the corresponding evolution problems are


essentially different from those one meets when the possible bounces are
assumed "elastic" [16][17][18][19]. A synthetic view may be gained from the
energy balance drawn In Sec.IO below. The replacement of u+ In <1.2) by
some weighted mean of u+ and u- results In the Introduction of a
"dissipation Index" 5, with zero value in an elastic bounce, while the
softness case corresponds to 5= I.

r Dry friction at a point of contact will be described by some extension of


Coulomb's Jaw to possibly anisotropic surfaces. The traditional formulation
of this law rests on the decomposition of a contact force into Its normal and
tangential components; the formulation then consists of two separate state-
ments respectively pertaining to zero and nonzero sliding velocity. In some
of the author's early papers [20)[21 )[22], It has been observed that, as soon
as the normal component Is treated as Known, these pair of statements Jets
Itself be synthetically expressed as a Jaw of resistance deriving from a
"pseudopotential". This In tum may be transcribed into a variational
Inequality [23], reflecting a "principle of maximal dissipation".
By a law of resistance, we mean a relation (In the present case,
nonsmooth and not expressible through a single-valued function) between the
contact force and the sliding velocity. Recall that significant mathematical
and numerical papers have, In recent years, been devoted to problems which
Instead Involve a "pseudo-friction" law. These problems are developed In the
frameworK of small deviations and the sliding velocity vector is replaced by
the tangential relative displacement of the contacting bodies. The status of
such a pseudo-friction, compared to proper friction, Is similar to that of
HencKy plasticity with respect to proper plastic flow rules.
Unilateral Contact and Dry Friction 9

The present lectures rest on a newer formulation of the possibly


anisotropic coulomb Jaw, avoiding the decomposition of the contact force
[24][25]. Similarly to what has been shown for the frictionless case, these
formulations suggest numerical algorithms of time discretization. Further-
more, the resulting relation being conic with regard to the contact force, it
admits an extension to Nonsmooth Dynamics, independent, as before, of the
choice of a base measure dv.

ao Here again, the possible nonsmooth motions are found to be governed by


some measure differential Inclusions. These differential Inclusions are
applied in Sec. IS to the dynamics of velocity jumps.
Singularities in the dynamics of systems involving Coulomb friction
used to be a matter of controversy during the first quarter of this century. It
was observed that some Initial value problems could admit several solutions
or no solution and also that the behaviour of the Investigated system
depended on its constants on a discontinuous way. At the time, these
findings were considered by such authors as P.Palnleve as contradicting the
very bases of Physics. In modern views, nothing looks paradoxical in that, so
there only remains of all these discussions the assertion, first made by
L.Lecornu [26]. that, in the presence of dry friction, velocity jumps are not
necessarily the consequences of collisions.
It is shown by an example that the numerical techniques we propose can
handle these frictional catastrop!Jes without difficulty.

go These lectures are restricted, for brevity, to time-independent cons-


traints. However, by changing the reference frame and introducing adequate
fictitious forces, one is able to apply the proposed methods to the motion of
10 J.J. Moreau

a small object lying on a vibrating table or, in the course of an earthquake,


on the ground surface. An example is displayed, exhibiting some unexpected
features.

1oo For better agreement with the behaviour of real systems, one is
commonly Jed to apply the traditional law of Coulomb with different values
of the friction coefficient, depending on whether the sliding velocity
vanishes or not. This distinction made between the "static" an "dynamic"
friction coefficients seems, at first glance, to destroy the unity brought into
the formulations by the use of Convex Analysis. Actually, It Is shown in
Sec.17 that, far from beeing a mere empirical alteration of these formu-
lations, such a distinction is inherently involved in the consequent deve-
lopments. The numerical techniques proposed in these lectures are able to
handle it without causing any computing problem. In fact, whether the
sliding velocity exactly vanishes or not at the end of a time-step IS
explicitely determined by the algorithms; so the friction coefficient for the
next step may be adjusted accordingly.

Let us close this Introduction by aknowledging that Coulomb's law can


provide only a rather crude approximation of the reality of dry friction (a
recent review of the subject may be found in [27]). Also, the collisions
affecting parts in real machines cannot be expected to fall exactly under one
of the categories respectively described as "soft" or "elastic". And it is
unlikely that any deflnlte value of the "dissipatlon index" could be identified
on a clear basis. But a fact of life is that, in most engineering situations, the
higher order information needed for more accurate description IS not
available. So one has to be content with some moderately precise
Unilateral Contact and Dry Friction 11

calculation, accounting at least for the main features of phenomena. In three


years of experiments, the approach we propose has proved to be very
workable. Because of their theoretical consistency and numerical stability,
the described algorithms seem to be "robust" enough for accepting in the
future various empirical alterations, aimed at improving their power of
prediction.
No allusion is made in these lectures to the contact between deforma-
ble bodies, currently a very active domain of research. The reader will find
references to this subject in other parts of this volume. In what concerns
computation, since the spatial discretization of a continuous medium, using
for instance a finite element scheme, generates a finite-dimensional space
of positions, the design of numerical procedures may take an inspiration
from the methods presented here <see e.g. [28] ). But some fundamental
differences between continuum dynamics and finite freedom dynamics have
to be kept in mind. Because, in continuous media, every contact particle has
zero mass, the concept of a soft contact, as opposed to an elastic contact,
becomes unsignlflcant (it only stays as an option in constructing numerical
algorithms). Possible dissipation reenters the scene through the constitutive
laws which govern the behaviour of the concerned bodies. In elastic bodies,
shock waves are expected to originate from boundary impacts. It Is only
when the time taken by these waves to travel the whole system is short,
with respect to some other typical time values, that the treatment of
deformable systems may be strictly conducted in the lines of finite freedom
dynamics. Actually, most papers on continuous systems so far are restricted
to quasistatic evolution problems, l.e. the terms Involving Inertia are
neglected.
12 J.J. Moreau

2.DIFFERENTIAL PROPERTIES
Let a mechanical system have a finite number n of degrees of freedom;
every possible position of It may be located through the value It Imparts to
Q=(Q 1, . ,qn ), an element of some open subset 0 of IRn . This holds at least
I n
locally; In other words, q ,... ,q are local coordinates In the manifold of the
system possible positions.
One defines a motion by making q depend on time. If the derivatives q' i
or the n functions t-+q1 exist at an Instant t, we shall refer to the element
q'=(q' 1,... ,q.n) of IRn as the velocity of the system at this instant.
Motions will be studied on some time interval I, containing its origin t 0
but nonnecessar11y closed nor bounded from the right. We shall not suppose
the function q:I-+IRn derivable everywhere. Instead, we assume the existence
of a velocity function u:I-+Rn such that
'v'tEI : q<t>=q<t0 > + Jt u('r) dt. (2. 1)
This makes sense as soon as u Is locally Lebesgue-integrable on I. More
specially, we shall suppose that the function u has locally bounded
variation on I, I.e. It has bounded variation on every compact subinterval of
I; notation: uEibvO,IRn). This secures that, at every t In the Interior or I, the
rlg!Jt-1/mlt u+(t) and the left-limit u-(t) exist.
By convention, for the Initial Instant t 0 , the left-limit u-<t 0 > is
understood as equal to u<t0>. This Is more than a notational trick; such a
writing actually reflects the significance we generally mean to give to
the Initial condition u<t0>=u 0 of an evolution problem. It Is Intimated that
Investigation begins at t 0, but that the mechanical system was already In
Unilateral Contact and Dry Friction 13

existence before. By u0 is Imparted some abridged Information about the


system history, precisely the left-limit u-(t0 ).
Symmetrically, if I possesses a right end, say tr, and contains it, the
writing u+<tr) = u<tr) will prove convenient.
From (2.1) 1t results that the function q possesses at every t>t 0 a
left-derivative q-c(), equal to u- Ct:) and, at every t different from the
possible right end of I, a rigflt-derivatlve q +(t), equal to u+(1).
In addition to the constraints which have permitted the q parametri-
zation, we assume that the system is submitted to some unilateral
constraints whose geometric effect is expressed by a finite set of
inequalities
f()((q)~O. <XE( 1,2, ... ,~}. (2.2)

The functions fiX :0-IR are supposed c 1 with respective gradients 'VfiX =
<afiX/aq 1,... ,afiX/aq0 ) different from zero, at least in a neighbourhood of the
corresponding hypersurface fIX =0.

Inequalities (2.2) define tile feasible region 4l ofO ; for brevity. we


assume tflat tile functions fiX do not depend on time, thereby leaving aside
the possibility of moving constraints.

Through the chain rule, the existence of one-side derivatives for the
functions t-qi(t) implies the same fort- fiX{q{t)). Consequently, If a motion
verifies f()({q{t))~O for every t, then at any instant t such that f()({Q(t))=O,
one readily finds u+(t).'VfiX{Q{t))~O and u-(t).'VfiX(Q{t))~O {the dot refers to
the usual scalar product of IR 0 ).
14 J.J. Moreau

Generally, let us put:


NOTATION2.1 Forevery qEO define
J(q) := {<XE{ 1,.... ~}: f<X(q)~O} (2.3)

and
V(q) := (vEIRn: \f<XEJ(q), v.\7f <X(q)~O} (2.4)

(observe tllat V(q) equals tile whole of IRn ;/J(q)=0 )

Using as above the one-side derivatives, one obtains:

PROPOSITION 2.2 If a motion t-+q(t) agrees wit/7 tile set of constraint


fnequalitles(2.3), ie. q(t)E4> forevery t, tllen
\ftEint I : u+(t)EV(q(t)) and u-(t)E- V(q(t)).

REMARK 2.3 In existential studies as well as in numerical algorithms, the


definition (2.4) of V(q) will commonly be invoked with q~4>. Then, the
following is useful:

PROPOSITION 2.4 Let tile function t-+q(t) be associated wit/7 some

uEL 1Joc(I,IRn) tllroug/7 (2.1 ). Suppose tllat q(t 0 )E4> and tllat, for Lebesgue-
almost every tEl , one /las u(t)EV(q(t)). Tllen q(t)E4> for every t.

PROOF. Let us suppose the existence of some LEI , with q(T)~<t> and look for

contradiction. There exists <XE{ 1,.... ~}such that f<X(q(r))>O. The set {tEl: t~T
and f<X(q(T))~O} Is nonempty Cit contains t 0); let a denote its l.u.b .. Due to the
continuity of f<X, one has f<X(q(O))=O. Since f<X lsc 1, the function t-+ f<X(q(t))
is absolutely continuous on [o,T] ; after expressing its derivative by the
chain rule, one may write
f(X(Q(T)) =I~ U(t).\7f(X(q(t)) dt.
Unilateral Contact and Dry Friction 15

In view of the definition (2.4) of V, the integrand should be ~0 for


Lebesgue-almost every t, hence fo:(q("t:))~O, which is a contradiction.

If u has locally bounded variation, it belongs to 11ocO,IR 0 ) and the set of


its discontinuity points is countable, hence Lebesgue-negligible. Thus, in
using the above Proposition, one may replace u in the assumption
u(t)EV(q(t)) by u+ or u- or any weighted mean of them.

REMARK 2.5 The subset V(q) of IR0 is a closed convex cone. In case qE4>, this
coincides with what is usually called the tangent cone to the region 4> at
point q (equal, in particular, to the whole of IR 0 if qEint 4>). On the contrary,
if qi4>, one commonly agrees to say that the tangent cone to 4> at this point
is empty; so is not V(q).
Some caution is needed when Interpreting the concept of a tangent cone.
Let q0E4> and vEIR 0 ; in view of Prop.2.2, for the existence of a mapping q:t-4>
such that q<t0)=Qo and q'+(t 0 >=v, it is necessary that vEV(q0 >; a
counter-example may be found in [ 15], showing that this is not sufficient.
However, if In addition one assumes lnt V(q 0 );z!.0, then existence Is secured
[29]. Through classical Convex Analysis, the latter assumption Is equivalent
to the polar cone of V<q0 > having a compact basis; this is the convex cone
generated in IR 0 by the elements \i'fo:<q 0), with <XEJ(q 0), so the assumption
amounts to assert the existence of a hyperplane in IR0 , not containing the
origin, which intersects all the half -lines generated by these elements. We
shall meet this cone again in Sec.S.

REMARK 2.6 A deeper insight into the situation could be gained by


16 J.J. Moreau

considering the differential mamfold P of the system positions, without


preference to any pecullar system of local coordinates. A motion is the
conceived as a mapping p:l-+1>. The (possibly one-side) velocity of the system
at some instant "C equals, by definition, the (possibly one-side) derivative of
this mapping, an element, say p'(t), of the tangent space Yp(t) toP at point
p(t). The real numbers q'i (t) considered In the foregoing equal the
components of p'(t), relative to the base Induced In P~t> by the local
coordinate system In use. lnequallties (2.2) are imparted a coordinate-free
meaning provided one understands the functions f(X as c 1 mappings of P to IR,
without reference to any choice of local coordinates. Then, by the gradient
\7fo:<P> is meant an element of p; ,the cotangent space to P at point p.
Also in this llnear space, the local coordinate system induces a base; the
partial derivatives precedingly invoked equal the components of \7fo:(p)
relative to this base. In Definition (2.4), the Euclldean scalar product of IR"
should then be replaced by the bilinear form <.,.> which puts the llnear
spaces P'P and p; in duality. Thereby, for every position p, a pair of mutually
polar convex cones Is defined In these dual linear spaces, without reference
to any choice of local coordinates.
The concept of a convex cone in Ppor In p; is meanlgful, In view of the
linear structure of these tangent spaces. Concerning, on the contrary, the
feasible region, the differentiable manifold P cannot in general support any
convexity assumption. If such an assumption is made, it only refers to some
peculiar coordinate system. However, let us mention the following special
case.
For the treatment of dynamical problems concerning a scleronomic
Unilateral Contact and Dry Friction 17

system, the expression ~AiJ(q)q'iq.J of the kinetic energy has to be


introduced. This is a positive definite quadratic form In q' and, classically,
by putting ds 2 = A1J(q) dqidqJ, one equips the differential manifold 1' with a
Riemannian metric independent of the coordinates in use. So this metric Is
intrinsically associated with the dynamical structure of the mechanical
system. Now. it may happen that some local coordinates exist, such that the
coefficients AiJ(q) are constant in q; so is the case, for Instance, If the
system consists of a single rigid body performing only motions parallel to a
fixed plane. Under such circumstances, the curvature of 1' Is zero; In other
words, this manifold is locally Euclidean. Then. at least In sufficiently small
regions, the concept of the convexity of a subset of P becomes mechanically
meaningful. The mathematical paper [ 16] was precisely based on the
convexity of the feasible region.
Anyway, the writing in (2.1) makes sense only as far as the functions
t-.q and t-.u take their values in a fixed 11near space, namely IRn for the
present. On the contrary, in the differential geometric setting, the velocity
at time t would be an element of the tangent space F~u, which depends on t
through the unknown mapping p: I-.F.

3.KINEMATICS
In all the sequel, each of the inequalities foc~O will be understood as
characterizing the system positions agreeing with the mutual impenetrabi-
Hty of a certain pair of rigid bodies. For instance, let us drop the subscript
ex and assume that condition f ~o expresses that some rigid part :B 1 of the
system does not overlap a given external obstacle :B0, fixed relative to the
18 ].].Moreau

reference frame in use. The impenetrability of two rigid bodies 13 1 and 13 2 ,


which both are constituents of the system, would finally result in the same
formalism (see [30], where the case of an external obstacle with prescribed
motion is also considered).
Equality f(q)=O means that, in the position q of the system, the part 13 1
touches 130 . We shall always assume that contact takes place through a
single particle of 13 1 , which in general depends on q , say ){1(q). The res-
pective boundaries of contacting bodies will be supposed to permit the defi-
nition of a common tangent plane at ){1(q) to these boundaries. This does not
preclude edges or vertices; one of the bodies may even reduce to a single
particle, provided the boundary of the other is a smooth surface.
Let 1{q denote the unit vector; normal to this tangent plane and di-
rected toward 13 1 .
As usual, the primitive constraints of the system, i.e. the constraints
which have permitted the parametrization through (q 1, ... ,Qn ), are assumed
smooth enough for the following to hold. Let a motion be described by giving
q as a function oft. For every t such that the (possibly one-side) derivatives
q 1,... ,Q.n exist, every particle, say ){, of the system possesses a (possibly
one-side) velocity vector, relative to the reference frame in use. Calculating
this vector yields an expression V()f, q, q'), affine with regard to q.
For brevity, we shall restrict the sequel to the scleronomic case, I.e.
the primitive constraints do not depend on time; then the above expression is
linear in q'. Let us apply this to the contact particle ){1(q) and put the
notation
(3.1)
Unilateral Contact and Dry Friction 19

For every q such that f(Q)=O, this introduces the mapping q- Gqq', I Inear of
IR 0 to the Euclidean linear space ~3 of the vectors of physical space.
In the case of a pair of bodies which both are parts of the system, a
linear mapping slm11ar to Gq would express from q' the relative velocity, at
a possible contact point, of one of this part with respect to the other.
The writing In (3.1) Is not restricted to motions agreeing with the lm-
penetrabi I ity constraint. We now are to take this constraint Into account. Let
a value of q correspond to contact, i.e. f(q)=O, and let veiR 0 . Let a motion
start from this position q at some time 1, with right-velocity q+ equal to v
at this instant. Evaluating f at all subsequent pos1tions, one obtains a
function of time whose right-derivative at 1 may be expressed through the
chain rule, namely v.\7f(q). A~sume v.\7f(Q)<0; then Instant 1 Is followed by a
nonzero time interval over which f <0, I.e. ~1 and ~0 break contact. This
implies that, at 1 , the right-velocity aqv of the contact particle )(1(q) of
~1 veri fie 1'{Q. Gqv ;;:O; otherwise the motion of this particle would require
of ~ 1 to overlap ~ 0 , so making f;;:O.
This shows that, for vEIR 0 and for any fixed q satisfying f(q)=O, one has
the implication
v.\7f(q)<0 ~ 1'{Q. Gqv ;;:0. (3.2)
Let us introduce now the 1inear mapping G~ : x3 - IR", the transpose of
Gq x
in the sense of the Euc11dean autodua11ties of 3 and IR" ; then 1'{q aqv =
*
v. Gq 1tq Recall that we have assumed 'Vf ';'!
.
0 ; through a unilateral version
of the Lagrange multiplier theorem, implication (3.2) yields:

PROPOSITION 3.1 For ever_y q venfying f(q):O, there exists A.;..O SfJCIJ

tllat
20 J.J. Moreau

G;nq = _,._ vHq>. (3.3)

REMARK 3.2 We shall later need t~tat the element G;nq of IRn differ frorn
zero, i.e. "A>O. It is a general fact that the kernel of G; equals the subspace
of ~3 orthogonal to the range Gq<IRn) of Gq. In particular, when Gq<IRn) is
the whole of ~3 , the kernel of G; reduces to {0} and this secures G;1{q;r0.
But, in some usual applications, Gq<IRn) will be a strict subspace of ~3 ; for
instance, if the primitive constraints allow ~1 to perform only motions
parallel to some fixed plane, then dim Gq<IRn) =2. What precedes shows that
G;nq =0 if and only if Gq<IRn) is contained in the two-dimensional subspace
of ~3 consisting of the vectors parallel to the common tangent plane to
contacting bodies.

REMARK 3.3 So far, Gq has been defined only for such q that f(q)=O. In
computation, as well as in existential studies, it will prove useful to extend
the definitions of Gq and 1tq, in a smooth arbitrary way, to the whole range 0
of the local coordinates in use, or at least to some neighbourhood of the
hypersurface f=O. This extension may additionally be required to preserve
the property (3.3).

REMARK 3.4 Let a motion comply with condition f~O at every time. Consider
an instant of contact, i.e. at which f(q)=O, and suppose that the two-sided
derivative q' exists at this instant. Since the latter equals the common value
of q'+ and q-, the observations made in Sec.2 imply q'.V'f(q)=O. In view of

Proposition 3.1, this yields 1tq.Gqq'=O, expressing that the velocity of the
particle :M1(q) of ~ 1 through which this body touches the fixed obstacle ~ 0 is
a vector parallel to the common tangent plane.
Unilateral Contact and Dry Friction 21

Similar reasoning would apply to a pair of moving parts of the system: If


contact holds at some instant and if, at this instant, the time-derivative q'
exists, in the ordinary two-side sense, then the relative velocity of one of
these bodies with respect to the other, at the contact point, is a vector
parallel to the common tangent plane. The same fact is classically
established in elementary Kinematics, under the stronger assumption of
permanent contact. One refers to this relative velocity as the sliding
velocity of the first body upon the second.

4. LAGRANGE EOUA TIONS


Let a motion of the system be described under the form (2. I). It will be
said smooth if the velocity function u is locally absolutely continuous, i.e.
absolutely continuous on every compact subinterval of I. This Implies the
existence of the derivative u'<t)=qu(t) for Lebesgue-almost every t. We shall
refer to the element qH of IRn as the acceleration of the system.
Such a motion agrees with Dynamics If and only If the function t-q(t)
verifies the Lagrange differential equation
&<a:E'cfaqi>- a:E/aq 1 =c1. (4. I)

Here :fc(q,q') denotes the expression of the kinetic energy; since we assume
the system scleronomlc, this Is
:fc(q,q') = ~ Aij(q) q' iq, j ' (4.2)

a positive definite quadratic form In q'.


By c1 are denoted the covariant components of the totality of the
forces acting on the system. These possibly comprise some given forces,
whose covariant components F1 are known functions of time, position and
22 J.J. Moreau

velocity, and also comprise the a priori unknown reactions or constraint


forces, involved In the constraints that the system experiences. As usual,
when Lagrange equations are applied, we shall assume that the primitive
constraints, i.e. those which have permitted the q parametrization, are
perfect, In the sense that the corresponding reactions have zero covariant
components. But we shall have to take into account the reactions of the
superimposed unilateral constraints.
The left-hand side of (4.1) may be developed into
A.. q.. j +
(A 'j .. - -21 A.... ) q.j q.1< ' (4.3)
l) l ..... Jr.. ,l

where Atj,k denotes the partial derivative of A11 with regard to qk: Therefore,
(4.1) takes on the form A11 q.. j = K1 + r 1 ; here the term r 1 refers to the
totality of the reactions of superimposed constraints, while K1 is a known
function of t, q, q, equal to Fi minus all the terms in (4.3) which Involve
q'Jq.k. Since the matrix A is invertible, we may finally write this down as
q" =A- 1K + A- 1r. (4.4)
It has been assumed that each Inequality foc<q>=O expresses the contact
between some pair of rigid bodies. Let us denote by r 1ex ,... ,r~ the covariant
components of the corresponding reaction, making an element of IR0 denoted
by roc. Formally, this term will be Introduced also when the said contact Is
not In effect, so we shall state
(4.5)
To fix the ideas, suppose, as in Sec.3, that the contact expressed by
foc(q) =0 takes place between some part, say :Bo:, of the system and an
unmoving external obstacle. Suppose that the contact action on :Boc results in
Unilateral Contact and Dry Friction 23

a single force 'Rex, applied to a particle of this body denoted by :Mcx(q) . By


definition, the covariant components of this force make the element roc. of IR 0
such that
(4.6)
On the left-hand side, the dot refers to the Euclidean scalar product of :3 ,
on the right-hand side to the standard scalar product of IR 0 . As before, due
to scleronomy, V(:Mcx(q), q, v) is a linear expression in v, say G~v. Then, by
introducing the transpose mapping ~: :3 -+ IR 0 , one equivalently wr1tes
down (4.6) as
(4.7)

5.SMOOTH FRICTIONLESS MOTIONS


In this Section, we shall assume that the possible unilateral contacts
are frictionless Under the preceding notations, this means that, for every <X
such that f (q) =0,
oc.
oc. oc.
3pEIR : 1{ = p 1(Q , (5. I )
where 1(~ denotes the common normal unit at 11fcx(Q) to the contacting
bodies, directed toward ~ex .
We shall assume In addition that the contact Is unilateral without
adhesive effect, I.e. p ~0.
Through Prop.3.1, conditions (4.7) and (5.1) imply
(5.2)

Note tnat tne repetition of a Greek index Will never be understood as


1mp!)1ing !J(J!l7!77at/on.
Provided that ~1(~ ;z! 0 (see Remark 3.2), it may more precisely be
24 J.J. Moreau

checked that (5.2) is equivalent to the existence of 'Rex agreeing with the
above assumptions. And, in view of the convention made in (4.5), this
equivalence remains valld for every QE<I> if one stipulates that Vcx =0 when
foc(Q)<O. Furthermore, in formulating our evolution problem under the
geometric condition QE<Il , it is Immaterial to state that (5.2) also holds for
non feasible q.
Therefore, a value of the total reaction term r = I, rcx is compatible
with the stated laws of contact if and only if 1t satisfies
r E - N(Q), (5.3)
where N(Q) denotes t/Je convex cone generated in IR" by the elements Vfcx(q),
<XEJ(q) (see Notation 2.1 ). According to an usual convention, if J(q)=0 this
cone consists of the zero of IR". In all cases V(Q), as defined in (2.4), and N(q)
make a pair of mutually polar or conju,.qate cones. When q belongs to the
feasible region, N(q) Is nothing else than the <outward) normal cone to 4> at
point Q, but what we are denoting here by N(q) also makes sense and is
nonempty for qt<ll.
Eliminating r between (5.3) and the Lagrange equations, as they are
displayed In Sec.4, one obtains that a smooth motion of the system agrees
with all the mechanical conditions stated, if and only If the differential
inclusion
-A(q) q" + K(t,q,q') E N(q) (5.4)

Is satisfied Lebesgue-a.e. in I, together with the geometric condition of the


un11ateral constraints,
VtEI : q(t) E ~. (5.5)

The Proposition below marks a turning point, regarding all our subject
Unilateral Contact and Dry Friction 25

matter. For every (closed, convex> subset of IR0 , say C, we denote by 'c Its
Indicator function, I.e. c<x>=O If xEC and +oo otherwise. The subdlfferen-
tlal O'i'c<x> is known to equal the normal cone to Cat point x <empty if xtC).

PROPOSITION 5.1 A smooth motion_ with Initial data q<t0>belonging to 4>,


Is a solution of (5.4), (5.5) If and only if the velocity function u asso-
oated with q through (2.1) satisfies Lebesgue-a. e. In 1 the differential
Inclusion
-A(q) u' + K<t,q,u) E OljiV(Q)(u). (5.6)

PROOF. For every t such that (5.6) holds, the right-hand side 1s nonempty,
hence u(t)EV(q(t)). Since, by assumption, u is a (locally absolutely) conti-
nuous function, (2. 1) entalls that u(t) equals, for every tE1nt I, the (two-
sided) derivative of t-.q(t) Through Proposition 2.4, one concludes that, 1f
(5.6) is verified Lebesgue-a.e., then (5.5) holds. Furthermore, (5.6) implles
(5.4) because a'i'V(q)<u> is essentially a subset of N(Q), the polar cone of V(Q).
Conversely, suppose that the function t-.q(t) satisfies (5.5). Since u Is
continuous, Proposition 2.2 shows that u(t), for every tEint I, belongs to
V(q(t))n-V(q(t)) wt1ich is the linear subspace of IR0 orthogonal to N(q(t)).
Therefore, if (5.4) holds Lebesgue-a.e., then for Lebesgue-almost every t, the
values of u and -Au'+ K are orthogonal and respectively belong to the pair of
mutually polar cones V(q(t)) and N(q(t)); consequently they are conJiJ..qate
pomts relative to this pair, i.e. (5.6) holds.
REMARK 5.2 At the present stage, where the motion smoothness, i.e. the
local absolute continuity of u, is a priori assumed, the same symmetry
between past and future is observed as in the classical case of b11aterally
26 J.J. Moreau

constrained frictionless systems. In particular, for the differential inclusion


(5.6) to take care automatically of condition qE, it is enough that q('()
belongs to at some HI, nonnecessarily the initial instant. Also an
equivalence similar to what is stated in the above Proposition may
symmetrically be established, with (5.6) replaced by
-A(q) u' + K(t,q,u) E- oljl_y(q)(u). (5.7)

Similarly to (5.6), this implies the orthogonality of the elements u and


-Au'+K of IR0 . From such an orthogonality, the same power equation may be
derived as In the traditional case of frictionless time-independent bilateral
constraints:
(5.8)

REMARK 5.3 As an introduction to forthcoming Sections, let us indicate how


the formulation (5.6) directly suggests a procedure of time discretization
for computing approximately the motion consequent to some initial data
q<t 0) = q0. given in <P (5.9)

u<t0> = u0, given in V(q 0). (5.10)

Let <t1,tF) be a time-step (here I is understood as referring to "initial"


and F to "final"), with length h=tF-t 1 and midpoint tM=t 1 +~ h . From the
approximate values q1, u1of the functions q and u at t 1, one has to compute qF
and uf, corresponding to tr.
Using <uF-u 1)/h as an approximant of u', one discretizes the differential
Inclusion (5.6) into
(5.11)

Here qM = q1+~ hu 1 Is a midpoint approximant of q; by li'M is denoted the


Unilateral Contact and Dry Friction 27

indicator function of V(qM). Inserting uF as an approximant of u In the


right-hand side tends to qualify this discretization scheme as "implicW.
However, the smoothness of the given function K allows one to replace in it u
by u1, so the procedure may be said "semi-implicit".
We shall come back later to more general algorithms of the same sort;
Jet us only show here how (5.11) uniquely determines llr Suppose, for
simplicity, that A(qM) reduces to the unit matrix. This actually entails no
loss of generallty: in the llne of Remark 2.6, it amounts to make of the
tangent space, at the point qM of the position manifold, a Euclldean linear
space, with scalar product defined through A(qM), and to take an orthonormal
base in this space <more detail on the practical use of this trick may be
found in [IS]). Then (5.1 1) becomes
-(UF-ul) + hK E aljiM(UF).

The multiplication of both members of (5.11) by the positive number h has


not altered the right-hand side, because a,McuF) is a cone. In view of the
classical characterization of the proximal point to u1+ hK in the closed
convex subset V(QM) of the Euclidean linear space, this is equivalent to
UF = prox ( u1+ hK, V(QM)). (5.12)

Observe that u1+ hK is nothing but the value that uF would take in the case
J(QM)=0, i.e. the value that the discretization of Lagrange equations would
yield in the absence of superimposed unilateral constraints.
After determining uF, one finishes the computation step by calculating
QF= qM+~h UF.
28 J.J. Moreau

6. PERCUSSIONS AND FRICTIONLESS SHOCKS


The preceding Section was restricted to motions a priori assumed
smooth. For such a motion, there may exist in particular a time interval
during which one or more of the contacts persist, say the contact expressed
by fiX =0. Call tc the end of this Interval and suppose that a nonzero interval
follows, over which fiX< 0. In other words, as soon as t exceeds tc, the set
J(q(t)) ceases to contain <X; thus the cone V(q(t)) suddenly Increases. This
Involves no contradiction with the assumed (locally absolute) continuity of
u, nor with the inclusion uEVn-v, resulting from Prop.2.2. Certain motions
of a unilateral pendulum provide familiar examples of this.
Computationally, there is no difficulty in approximating a motion
showing such a smooth contact break, through the algorithm of Remark 5.3. It
only happens that, from a certain time-step to the next, the dimension of
V(QM) suddenly increases, without producing any notable irregularity in the
sequence of the calculated values of u.
Imagine, on the contrary, that an Interval of smooth motion ends at some
Instant ts wlt!J t!Je occurence of new contacts, I.e. J(q) suddenly increases.
In view of Prop.2.2, the continuity of u at ts would require of the left-limit
u-(t5), an element of -V(q(t5)), to belong also to V(q(ts)) This would mean
that the new contacts are attained tangentially, an event which cannot be
expected In general. So, a discontinuity of u at t 5 has to be contemplated.
This Is called a s!Jock and, to deal with It, Classical Mechanics provides the
concept of percussion, that we are to review in a few words.
Assume that ts is followed by another Interval of smooth motion. It will
be understood that, because of slight deformabillty in the system parts, the
Unilateral Contact and Dry Friction 29

velocity change is not rigorously instantaneous, but takes place on a "very


short" time interval, say <t 5,t5+8), over which the differential equations of
smooth dynamics supposedly hold. In view of the steep velocity change, these
equations are expected to involve "very large" values of the contact forces.
By equalling the integrals on (t 5,t5+8) of both members of the differential
equatlons, one obtains the momentum c!Jange formula. This Is a balance
equation, through which the net velocity change is related to the time
integral, say n, of the function t-r<t)EIR 0 which, in tt-,e notations of Sec.4,
represents the contact forces. Compared to It, the term K yields a negligible
integral, because e Is "very small"; for the same reason, the variations of q
on the interval are neglected. Once obtained, the balance formula Is Inserted
into the former setting of instantaneous shock; the element n of IR" is then
said to make t!Je components of t!Je contact percussions. This procedure may
be seen as an early example of a multiple scaling; it permits to calculate
the after-shock velocity u+(t5), as far as sufficient Information Is available
about contact percussions.
Usually, the above reasoning Is applied under the assumption of
frictionless contacts. Then it seems natural to admit, similarly to (5.2),
that, for every <XEJ(q5), the contribution to n of the corresponding contact,
say n((' has the following form
3M(( ;J~O : ntx = - M(( 'Vf (( (q (t5)). (6. I)

The argument commonly proposed in support to this assertion about n is


that, in (5.2), the vector 'Vftx(q(t)) should remain nearly constant during the
very short time interval (t 5 ,t5+8), because the variations of q are very small;
thus Mtx would simply equal the integral of the nonnegative real function
30 J.J. Moreau

t-vtX<t). In the author's opinion, this conclusion cannot be accepted without


further discussion, though physical situations certainly exist in which (6.1)
accurately agrees with reality. In fact, the use of the equations of regular
Dynamics rests on the smoothing effect of a certain amount of deformation
in the system parts. The very meaning of the parametrization q may then be
questioned. Furthermore, the resulting alterations of the functions frx,
however small in amplitude and possibly concentrated in the vicinity of the
point q<ts), are prone to generate nonnegligible variations of the vectors
V'ftX(q(t)) in the course of the interval <ts,t 5+8). A similar discussion could
also be conducted on equations (4.7) and (5.1 ): the deformation of the contac-
ting bodies, however small, may appreciably alter the vector 1(~ in 1:3 as
well as the mapping G~*: 1:3-IRn
At all events, (6.1) does not bring enough information about n to
determine u+(ts) completely, even in the simplest case where ..J(q(ts))
consists of a single element. Classically, a shock (in a scleronomic system)
is said elastic If it preserves the total kinetic energy; this additional
assertion permits, in the case of a single contact, to determine u+<tJ .)

unambiguously. This may be given a geometric form by using in IR 0 the kinetic


metric, i.e. the Euclidean metric defined by the matrix A(q(t 5 )). Then, if
J(q(t5 ))={<X}, one finds that u+(t 5 ) equals the mirror image of u-Ct. 5 ) relative
to the tangent plane at the hypersurface f rx =0.
But one can hardly justify energy conservation by any physical argument
In fact the deformation induced by the impact is expected to propagate
dynamically all through the various parts of the system and possibly also
through the external connected bodies. Even if the <very slightly) deformable
Unilateral Contact and Dry Friction 31

materials of which the system is built may be asserted perfectly elastic,


the various parts usually remain, after a bounce, in a state of vibration
which, In the energy balance drawn at macroscopic level, amounts to
dissipation.
In short, predicting accurately the outcome of a shock requires some
higher order information, unavailable in usual situations. The pertinence of
the model of an elastic shock has to be discussed In each particular
application. The same is true for the other sort of shoe~ we shall present In
Sec.B, which however offers the advantage of be~ter formal consistency and
easier numerical handling.

7. NONSMOOTH DYNAMICS
We now intend to Insert the description of shocks Into a generalized
formulation of the dynamics of the Investigated system, which does not
require -the local absolute continuity of the velocity function u:I-R". This
function will only be assumed to have locally bounded variation., i.e. to have
bounded variation on every compact subinterval of I ; notation: uE lbvO,IR 0 ).
Classically, with such a function, an IR"-valued measure on the Interval I Is
associated, that we shall call the differential measure of u and denote by du.
A characteristic property of this measure Is that, for every compact sub-
Interval [0,1:] c intI, one has
ho;t] du = u+("t)- u-(O). (7.1)
In fact, the bounded variation assumption secures the existence of the
one-sided limits of u at any point of lnt I . Equality <7.1) remains valid for
o=to provided that, as already proposed in Sec.2, we agree on the convention
32 J.J. Moreau

u-(t 0>=u(t 0). The symmetrical convention may also be used for u+ at the
possible right end of I.
By making 0=1:, one sees in particular that, if u is discontinuous at
point 1: , then the measure du is expected to possess an atom at this point,
with value equal to the total jump of u.
It is clear on (7.1) that du depends on the function u only through u+
and u-. Tne values that u may- take at its discontinuity points are immate-
rial Neither nave these values any- effect on tne expression (2. 1) of Q
since tne set of tne discontinuity- points of u is countable,. hence
Lebes._que-negligible.
The reader may refer to [7] as a monograph on the lbv functions of a real
interval and their differential measures, with values in a Banach space X. For
a vast class of Banach spaces, in particular for X= IR 0 , there comes out that,
if u is locally absolutely continuous, the measure du possesses, relatively
to the Lebesgue measure on I, here denoted by dt, a densi!J~' tunct ion , say
ut E L \x:<l, dt; IR">. Of course, the latter is defined up to the addition or a
dt-negliglble function. This is commonly expressed by saying that the
IR"-valued measure du equals tne product of tne real measure dt by the
IR"- valued_ local!)/ dt- integrable, function ut; notation du = ut dt. Also du
is said local/}' dt- continuous. Conversely, if du has sucrl a form, Ule
function uElbv(I,IR 0 ), possibly after correcting the unessential values it
takes on a countable set, is locally absolutely continuous. Furthermore, for
Lebesgue-almost every t , the value ui<O of the density function makes the
(two-sided) derivative of u.
Throughout these lectures, we shall comply with the widespread usage
Unilateral Contact and Dry Friction 33

of affectmg the character L to non-Hausdorff spaces consisting of functions


defined everywhere, while L will refer to the corresponding Hausdorff
spaces of equivalence classes.

In this setting, let us come back to Lagrange equations. If u is locally


absolutely continuous, with q related to it through (2.1 ), the notation used
in Sec.4 becomes
A.. ( q) ut, j + (A. ~- ( q) - 1
2 A.~- .(q)) uj uK -c
- .. (7.2)
IJ I J " J11. ,I I

The right-hand side refers to the covariant components of the totality of the
forces acting on the system. including the reactions of possible contacts. For
these differential equations to make sense. then functions t-. ci have to be
elements of L 11oc(l, dt; IR) ; so each of then equations (7.2) may equivalently
be expressed as an equality of measures on the interval I,
. . k
A.. (q) du 1 +(A .. ~-<q) - -2
1 A." .(q)) u1u dt =c. dt. (7.3)
1J IJ ,11. J11. ,1 I

One readily checks that the functions of t, by which the measures dui or dt,
on the left-hand side, are multiplied, have the local integrability properties
required in order that the products make sense.
Now, this new writing keeps meaningful for general uElbv<I,IR\ and
allows one to replace the terms ci dt, on the right-hand side by some real
measures dCi , said to be the covariant components of the total impulsion
dC experienced by the system. These will equal the sum of the dt-continuous
measures Fi(t,q,u) dt, describing given forces, and of the covariant
components dRi of the contact impulsion dR, an IRn -valued measure on I. For
instance, in the case of an isolated collision occuring at instant t 5 , as
34 J.J. Moreau

investigated in Sec.6, the measure dR would involve an atom with mass n,


placed at point t 5 ; this should be added to the dt-continuous measure r dt,
expressing the contact actions in the course of possible episodes of smooth
motion with persistent contact.
As before, we shall denote by K the known 1Rn-valued function with

components
K.(t,Q,U)
1
=F1.(t,Q,U) - (A1}... k(Q) - -21 A.k} ,1.(Q)) uj uk .
T!Jen t!Je following equality or IRn- valued measures on t!Je Interval 1

will be adopted as governing t!Je dynamics ofpossibly nonsmoot!J motions:


A(Q) du- K(t,Q,U) dt = dR. (7.4)

The connection between such an extension of Lagrange equations and the


principles of Classical Dynamics is discussed with more precision in [9].

8. CONTACT SOFTNESS
Let us consider again the differential inclusion (5.6), which has been
found to govern the assumedly smooth motions of the system. This inclusion
equivalently means that the expression t-r<t>EIRn of the covariant
components of the contact forces makes an element of L \x:<l, dt ,IRn) which
vertftes, for Lebesgue-almost every t,
- r<t> E a11'v(Q(t))<u<t>>. (8.1)

In the context of Sec.S, q(t) belongs to <P for every t; therefore, due to Prop.
2.2 and to the continuity of u, one has u(t)EV(q(t)) for every t. Hence, the
right-hand side of (8.1) is nonempty for every t Ut contains at least the
Unilateral Contact and Dry Friction 35

zero of IRn). This enables one, by altering the function r on a Lebesgue-


negligible subset of I , to make (8.1) hold everrwhere in 1.
In the language of Sec.7, the IRn -valued measure dR = r dt constitutes,
for the above case, the total contact impulsMn. We now propose to adapt
(8.1) to the more general setting of Nonsmooth Dynamics.
The Lebesgue measure thus loses its preeminence and we shall
definitely give up using the expression "almost everywhere". As observed in
Sec.7, if u is discontinuous, only u+ and u- have relevance to the motion;
so. one has to make a choice about what Is to replace u In the right-hand
side of (8. 1).

DEFINITION 8.1 The set of superimposed constraints Is said frictionless and


soft If the total contact Impulsion admits a representation dR=R'vdV, where
dV denotes a nonnegatlve real measure on I and R'~,~ an element of
\~P ,dv;IRn) such that. tor every t in 1 .

(8.2)

This first implies that the right-hand side is nonempty, hence


u+(t)EV(q(t)) for every t. Therefore, if the initial data satisfy q<t0 )Ecp, (8.2)
will secure, thanks to Prop.2.4, that q(t)Ecp for every tEl .
The following Proposition shows that the concept introduced by
Definition 8.1 does not actually depend on the peculiar choice of dV .

PROPOSITION 8.2 Inclusion (8.2) holds for every tEl If and only If the same
Is true after replacing dV by another nonnegative real measure relative to
which dR possesses a density function.
36 J.J. Moreau

PROOF Suppose that (8.2) holds for every t in I and denote by dv another
nonnegative real measure such that R'\) exists. If we put do=d~+dv, the
Radon-Nikodym theorem ensures the existence of the density functions ~a

and v'0 , nonnegative elements of 00 (1, do; on Then dR=R'vt.~'0 do=R'v v'0 do ;
so the IR 0 -valued functions R'vt.~' 0 and R'v v'0 are equal, except possibly in
some do-negllgible (hence dv-negl1gible) subset S of I. The subset N={tEI :
v'0 (0=0} is dv-negllgible. Outside the union SUN , the above implies R'v=
R'vV'0 /v' 0 , with V'ufv'0 ~0; then the expected inclusion holds, since the right-
hand member of (8.2) Is a cone. For tESUN, this <closed, convex> cone, being
nonempty, contains at least the zero of IR 0 . After replacing by zero the
values that the function R'\) , as formerly defined, may take in SUN, one
obtains the asserted conclusion, with dlJ and dv playing symmetric roles.

If the superimposed constraints agree with Definition 8.1, the


elimination of dR between (8.2) and the equation (7.4) of Nonsmooth
Dynamics yields the following characterization of the possible motions
- A(q) uv + K(t,q,u) t'v E a,V(q(t))<u+(t)), (8.3)

required to hold for every tEl . Here dtJ may equivalently be replaced by any
nonnegative real measure. relative to which du and dt possess density
functions. The existence of such measures is a priori secured by the
Radon-Nikodym theorem; one may take, for mstance, d~ = ldul + dt . where
ldul denotes the nonnegative real measure modulus (or variation measure)
[6][7] of the vector measure du. Since K is continuous and since the
discontinuity set of u is dt-negligible, it does no matter to replace u, on the
left-hand side, by u+ or u-.
Unilateral Contact and Dry Friction 37

The indifference of (8.3) regarding the choice of dp (this could be


checked directly, through the same reasoning as in the proof of Prop. 8.2),
suggests to strip down the writing into
- A(q) du + K(t,q,u) dt E o~Y(q(t))(U+(t)). (8.4)
This may be called a measure differential inclusion. The existence of
solutions to Initial value problems governed by conditions of this sort and
their possible uniqueness have so far been studied only In some special cases
[11][12][13][31].
As before, one observes that, provided that Q(t 0)E, inclusion (8.3)
entails q(t)E for every tEl .

REMARK 8.3 Similarly to what precedes, the simple assumption of


no-friction, without reference to "softness", as It has been expressed for the
case of smooth motions in (5.3), may be adapted to Nonsmooth Dynamics. It
will merely consist in stating that the contact impulsion dR possesses,
relative to some dbf ~0, a density function which satisfies for every tEl
- R'V(t) E N(q(t)). (8.5)

Here again, the fact that the right-hand member is a cone entails that such
an assumption does not actually depend on the peculiar choice of dv.
Since the right-hand member of (8.2) Is conta.lned In N(q(t)), (8.5)
constitutes a weaker assumption than what expresses Definition 8.1, I.e. the
set of the superimposed constraints may be frictionless without being soft.

REMARK 8.4 Classically, a subdifferential relation such as (8.2) admits


some alternative formulations. Here, since the polar cone of V(q) equals N(q),
(8.2) is found equivalent to asserting, for every t,
38 ].].Moreau

(8.6)

9. SOFT COLLISIONS
In the setting of the above Section, the following characterizes the
possible velocity jumps

PROPOSITION 9.1 Let tEl , different from t!Je possible ng!Jt end of t!J!s
Interval. For any motion satisfying (8.3), one !Jas
u+(t) = prox <u-(t), V(Q(t)), (9. I)

where t!Je prox!mat!on Is understood m t!Je sense or t!Je kinetic metric,


/e. t!Je Euclidean metric defined In IRn by t!Je matrtr A(Q(t))

PROOF In view of (7. I),

u\t) - u-(t) = J{ '( }du = J{ '( }uvdf.J. = v'( uv(t),


where blt~O denotes the integral of dbJ over the singleton {t}. If motlon
agrees with (8.3), one has

- A(q(t)) u'.,/t) + K(t,Q(t),U(t)) t'V(t) E OIJIV(q(t))(U+(t)).

Now Vtt'v(t)=O, since the Lebesgue measure dt has no atom. Then, after
multiplying both members or the above by Vt' one obtains

- A(Q(t))(U+(t)- U-(t)) E OIJIV(Q(t))(U+(T)).

If one uses in IRn a base orthonormal relatively to the metric in view, A(Q(t))

becomes the unit matrix, reducing this to the classical characteristic

property of proximal points.

This Proposition, which, under the convention u-<t 0)=u<t 0), also holds
Unilateral Contact and Dry Friction 39

for t=t 0 , shows that u exhibits a nonzero jump at point t 1f and only if

u-(t)V(q(t)), Le. a nontangential impact occurs at this instant.


Assume in addition that t >to, so u-(t)E-V(q(1:)) in view of Prop.2.2.
Then, the condition for nonzero velocity jump becomes
u-(l:)E V(Q(l:))n- V(Q(l:)).
The right-hand member is a linear subspace of IRn. If J(Q(1))=0, i.e.
Q(l:)Eint <J?, this is the whole of IRn. If J(Q(l:)) consists of a singleton, say {ex},
the subspace is the vector hyperplane tangent at q(l:) to hypersurface
f ()( =const.. For larger J(q(l:)), the point q(l:) lies on what may be called, in the
wide sense, an edge and the said subspace (possibly reduced to {0}) is
declared tangent to this edge.
Also for 1 > -r 0 , one observes that u-(1), being an element of - V(q(t)),
cannot belong to the interior of V(q('r)); thus (9.1) yields that u+ actually
I ies on the boundary of this polyhedral cone.

REMARK 9.2 Here is another consequence of Prop.9.1. Suppose that, on some


open subinterval I' of I , the motion evolves in such a way that the set-valued
function t-+J(q(t)) never increases. In other words, during this time interval,
some of the contacts may get loose, but no collision occurs. Let 1EI' and let
cxEJ(Q(l:)). Then, for every tEl' such that t~1. one has f()((q(t))=O, an equality
which, through the chain rule, entails
u-(l:).Vlf()( (Q(l:)) = 0.
Hence u-(l:)EV(q(t)), so Prop.9.1 shows that no velocity jump occurs at
instant t.
In the traditional treatment of unilateral constraints, the latter Is
taken for granted: one accepts to enter percussions into the analysis only at
40 J.J. Moreau

instants where geometry makes them unavoidable. This agrees with the
heuristic maxim of the "minimal singularity", but does not result from any
explicit mechanical assumption. Here is a familiar example demonstrating
this method deficiency.
Suppose an object performing a sliding motion in the contact of a table
<or simply at rest on it>. If an operator hits the table with a hammer, the
object is commonly observed to jump. So the table has imparted an Impulse
to the contacting object, without itself exhibiting any motion at the
macroscopic observation scale. In contrast, the assumption that unilateral
constraints are frictionless and soft rules out such an active behaviour of
boundaries.
Incidentally, the replacement of softness, in frictionless unilateral
constraints, by the quite different assumption of energy conservation
would also permit a deductive treatment of the above situation. It will be
shown In Sec. I 0 that energy loss, in velocity jumps, should on the contrary
be expected when frictionless soft constraints are present.

REMARK 9.3 Equation (9.1) expresses that, In the considered motion, all
velocity jumps are of the sort that the author has previoulsy called stan-
dard inelastic shocks [ 14][ 15]. These were proposed as a generalization of
the shocks which, In the case of a system involving a single constraint
Inequality, say f(q),O, are traditionally called "inelastic" or also "soft". In
fact, if f(q(t 5))=0, the tangent cone to the feasible region of IR 0 at point
q5 =Q{t5) Is simply the half-space V(q 5) = {vEIR 0 : v.V'f{q 5),0}. Since the
lert-slde velocity u-(t5) must belong to -V(q 5) (at least If one supposes
t 5 >t 0>. equ.(9.1) yields In this special case that u<t5 > equals t/Je ort!?ogo-
Unilateral Contact and Dry Friction 41

nal project /on of u-( t 5 ) to t!Je vector !Jyperplane tangent at q5 to hyper-


surface f=O. More information on general standard Inelastic shocks may be
found in [ 15] ; incidentally, the contact percussion receives an extremal
characterization, dual to (9.1 ).

10. ENERGY BALANCE


PROPOSITION 10.1 For every motion satisfying (8.3) t!Je function t-+Xc
belongs to lbv( I,IR ). In t!Je sense of t!Je ordering of real measures on 1,
one !Jas
dXc" F. dq, (I 0.1)

wit!J equality if and only If u !Jas no jump /IJ I.

PROOF For the traditional case of smooth motions with frictionless


(time-independent) superimposed constraints, the powerequat/o!J
d j
dt Xc = F1Ct,q,u) u
is easily derived from Lagrange equations, with left-hand side developed In
the form (4.3). For a motion governed by (7.4), with the contact Impulsion
dR=R'vdbl satisfying (8.2), there is only to retrace the same calculation,
under the replacement of some steps, based on the rules of usual Differential
Calculus, by what follows.
1 If u1 and uj belong to lbv{I,IR), the same holds for the product u1uj and
its differential measure is given [7] [32] by
d(u 1uj) = ui+ duj + uj- du 1.
The products of measures by functions, which appear on the right-hand side,
make sense because the functions t-+ u1+(t) and t-+ uj- belong to lbv(IJIR); so
they are locally integrable relative to any real measure. Hence, in view of
the symmetry of A,
42 J.J. Moreau

Ai{q)d(uiuj) = Aij(q)( u1++ ui-) duJ


Furthermore, since t--+q(t) is locally absolutely continuous, with t-u<n as
derivative, the differential measure of t-AiJ(q(t)) equals
d Aij =Aq ,k uk dt
and one has
i j i j i j
d(Aqu u) = u u dAq + Aq d(u u ).

2 The real measure Aq(q)(ui+- ui-) duj is nonnegative; it vanishes if and


only if u is continuous on I. In fact, this measure cons1sts of a countable
and locally summable collection of point measures located at the Jump
instants of u . Let t5 denote one of these instants; under the notations
u\t5 )=u;, u-(t 5 )=u;, q(t5 )=q5 , the mass of the corresponding pomt measure
equals
i+ i- j+ j-
Aq<qs)(us -us ) (us - us ).
Since Aq(q) is, for every q, a positive definite matrix, thiS real number 1s
nonnegative; it vanishes if and only if u;- u; =0.

3 Equality u.r=O, a consequence of the no-friction assumption in the case


of smooth mot1ons, is replaced at present by u+ R'b'=O, a fact observed in
(8.6).

REMARK 10.2 A more general concept than frictionless soft constraints is


obtained by inserting into the right-hand side of (8.2), instead of u+. some
weighted mean
( 10.2)
where 6 is a chosen real number, here supposed independent of t, for
simplicity. Since u8(t)=u(t), except at the jump points of u, which make a
Unilateral Contact and Dry Friction 43

Lebesgue-negligible subset of I, the law of constraint


- R'v(t) E oljly(q(t))(u 8(t)) ( 10.3)
implies, exactly like (8.2), that condition q(t)EP is satisfied for every t as
soon as it holds for t=t 0.
For 6>0 (even larger than 1), the law of constraint ( 10.3) entails the
same as what has been stated in Prop.1 0.1. Choosing 6<0 would yield the
reverse mequality, physically unacceptable <unless the possible bounces are
artificially enhanced, as in some electric billiard games).
If 6=0, i.e. u6 equals the mtdpoint or u- and u+, one f1nds eqvallly
d~c=F.dq , expressing the same e.ner..qr conservation as in smooth motions.
In that sense, the constraint law ( 10.3) with 6=0 may be said "elastic".
We suggest to call othe dissipation index of the constraint law ( 10.3).

11. TIME DISCRETIZATION ALGORITHM


The principle of such an algorithm has been introduced, for smooth
motions, in Remark 5.3. The main observation we now have to make is that
the same numerical technique applies in the framework of Nonsmooth
Dynamics, provided that the superimposed frictionless unilateral constraints
are assumed soft.
The expression in (7.4) of the system dynamics, in term of measures,
directly suggests time discretization: the rule will be to equal some appro-
ximants of the respective integrals of both members over each subinterval
of 1determined by the discretization nodes.
Let <t 1,tF) be one of these intervals (here I refers to "initial" an F to
"final"), with length h=tr-t 1 possibly variable from one step to another. From
the approximants q1and u1of q and u at the beginning of the Interval, one has
44 J.J. Moreau

to compute some approximants qF and uF , assigned to instant tF and which,


1n turn, will be used as in1t1al values 1n the next step.

Then
-1
ul =U 1+ h AM KM
is the value that the discretized equations of Dynamics would yield for uF in
the absence of contact force (here one may read the subscript L as referring
to "loose").
If QME int4> (i.e. all f(X(qM) are strictly negative) or uL EV(qM), one consi-
ders that contact forces have no effect on the calculated step; so one rnakes
uF=uL and goes to Stage 3 (this decis10n is actually a trivial case of Stage 2
below).

Stage 2 On the left-hand side of (8.4), let us replace A(q) and K(t,q,u) by Ar1
and KM . Then, an approximant of the integral of this left-hand side over
Ct 1,tF) is -A.r1CuF-uL). Concerning the right-hand side, it will be considered
that the set V(q(t)) keeps, throughout the interval, the constant value V(qM),
whose indicator function will be denoted by ~M Furthermore, let us take uF
as an approximant -one may rather say a simulat1m- of u+ This yields as a
simulation of (8.4) on the said interval
-AM(UF-UL) E aljJM(UF)
If the matrix AM is used in order to define a Euclidean metric on IRn, this
characterizes uF as the proximal point to ul in the closed convex set V(qM).
Therefore, computing uF is a Quadratic Programming problem: to minirm~?e
Unilateral Contact and Dry Friction 45

on V(qM) tiN? real function x-<x-uL).AM(x-uL). Recall that V(qM) is a


polyhedral cone, the intersection of a collection of half-spaces determined
by the index set J(qM). These correspond to the values of <X such that, in the
test position qM, the inequality ftx~O holds as an equality or is violated.
If their number is not too large, the proximal point will be constructed
algebraically; otherwise some of the classical algorithms of Quadratic
Programming will have to be applied.

Stage 3 One terminates the computation step by


qF =qM + ~ hUF
REMARK 11.1 In [24] are reported some computer experiments with methods
of the above sort. These methods prove to be stable. The finer the time
discretization is, the better the computed motion complies with the Inequa-
lities ftx~O. In that respect, some improvement may be achieved by
evaluat~ng, in Stage 2, the cone V(q) at another point than qM. A good choice
appears to be the point q1+ hu 1.
If these methods are applied to the calculation of mechanisms, which in
reality always involve some imperfectly known friction, no great precision
can anyway be expected in predicting the motion. From that viewpoint, a
moderately fine time discretization will be enough. In contrast, the violation
of the constraint Inequalities must sharply be kept In check. A very effective
way of doing it consists in completing each time step by a stage of linear
correction of the poss;b!e violation. Let us explain it In the simple case
where the position qF, calculated in Stage 3, violates only one of the cons-
traint inequalities. In other words, the real number lj)=f<qr> is found strictly
positive, with f denoting one of the functions ftx. A plausible correction of
46 J.J. Moreau

this violation <naturally assumed "small") would be to replace qF by Its


proximal point, say qc , in the region f~O. Proxlm1ty here should be
understood In the sense of some Euclidean metric on IRn; the most justified
choice is to rely on the kinetic metric, defined by the matrix A(q) To save
computation, the latter will be evaluated at a pomt where 1t has already
been calculated in the current step. Using an affine approximation of tr1e
function f, one obtains O=f(qc)=tp+Cqc-qF).'Vf+. . Here U1e gradient \7f
should be evaluated at some neighbour point. This precisely must have been
done at the Stage 2 of the current step, where also U1e vector A- 1\7f has
been needed. Defining qc as the proximal point to qF in the region f::;O,
relative to the metr1c in view, means that the vector oc-QF is parallel to
A-1.\lf(qc) Since the latter is estimated to be close to H1e calculated value
of A-l'Vf, this finally yields the approximate formula
-1 -1 -1
qC = qF- lp ('Vf.A \7f) A 'Vf +
The same techniques of violation control applies to the numerical
methods presented In the sequel, for unilateral constraints with friction.

REMARK 11.2 Also in [24], a numerical procedure is developed for a system


with a frictionless unilateral constraint expressed by a single inequality
f~O. assuming that the dissipation index, as we have defined It in Sec.10,
equals 1. In other words, the possible collisions are elastic bounces. In
contrast with the excellent stability of the preceding method, a careful
check of the energy balance of each time-step here is needed, in order to
prevent divergent oscillations when the algorithm is applied to the
computation of a motion with assumedly persistent contact
Unilateral Contact and Dry Friction 47

12. COULOMB FRICTION


This lectures are meant to provide only an Introduction to the treatment
of friction. So we shall restrict ourselves to a system Involving a single
unilateral constraint with geometric condition f(Q)~O (there is presented
in [30] a rather usual case, where a system with multiple possible contacts
may be reduced, through decompositt~n, to this simple setting).
For every q such that f(q)~O, the cone V(q) equals the half-space {vEIR":
v.'Vf(q)~O}, with boundary
T(q) = {vEIR": v.'Vf(Q)=O};
the latter is the vector hyperplane tangent at point q to the hypersurface
f =const. drawn through this point.
Let a motion satisfy f(q(t))~O for every tEl. If the velocity function u
is continuous at some 1>t0 , then, In view of Prop.2.2, u('t) belongs to
V(q('r))n-v<q(t)); this set equals T(q('t)) if f(q(r))~O and otherwise the
whole of IR 11 . The same is true for 1=t0, as far as the initial data have the
meaning we agreed to give them in Sect.2, namely u0 equals the left-limit
u-<t 0 > in a motion taking place before t 0, wtth the unilateral constraint
already in effect.
To ftx the Ideas, suppose, as In Sec.3, that equality f(q)=O expresses
that in the position q, some part :B 1 of the system touches the unmoving
external obstacle 130. Then, under the previous notations,
U = 'V(:M1(q),q,u) = Gqu
is the velocity of 1l 1 relative to 110 at the contact point.
Denoting again by 1{q the common normal unit vector to the contacting
bodies, directed toward 11 1, we assume G~1{Q ~ 0 <see Remark 3.2). Then the
assertion uEHQ) Is equivalent to U belonging to T(q), the linear subspace of
48 J.J. Moreau

L'3 orthogonal to 1{Q , I.e. the common tangent vector plane at the contact
point
The above is the situation in which, traditionally, Coulomb's law of tiry
friction is formulated. This Jaw is a relation between trle sltding veloci(v
UET(q) and the contact force REI'3 experienced by ~1

For brevity, let us write 1{ for 1{Q and Tfor T<ql. The familiar formu-
lation rests on the decompositlOn of 1{ into
1{ =Rr + p 1{. with RrET and p~O. <12. 1)

and consists of two well known separate assertions concerning the res-
pective cases U=O and U;rO in T
In some of the author's early papers [20] [21], it has been observed that.
as far as the normal component p is treated as known, this pair of
assertions is equivalent to a relation. between tM elements RT ami U of
the linear space T, which derives from a "pseudo-potential". Furthermore,
this formulation readily extends to the description of poss1bly am~-:otropic

friction, as a relation of the form


- U E aljl1)(1{ T), ( 12.2)

with 1) =pD 1. By D 1 is denoted a given closed convex subset ofT, containing


the origin; in the traditional case of isotropic friction, D 1 equals the disk
centered at the origin, with rad1us equal to the lrictwn coe.tlicient, say 'l
We ::.'flail restrict ourselves in !IIese lectures to tile c"'ase or bounded
friction, le.. tile set D 1 is bounded To take D 1 unbounded would provide a
way of including in our approach the situation traditionally called a
nonllolonomic constraint. But this would cause some complications in
further statements.
Through the standard calculation rules of Convex Analysis, ( 12.2) may
Unilateral Contact and Dry Friction 49

equivalently be written as
( 12.3)
where the dissipation function tp equals the support function of the set
-1> 1; in particular, for isotropic friction tp=yll.ll.
Elementary applicatiOns, where p in fact is known, may be found in
(22]. Furthermore, having to treat p as known does not prevent using this
pseudo-potential formulation <or an equivalent variational inequality which
expresses a "principle of maximal dissipation") in the proof of existence of
solutions to dynamical problems. See [331, where the normal components of
the contact forces become the primary unknowns in some functional analytic
arguments
In the present lectures, we choose to formulate the same Jaw under a
comcal equivalent form [24], avoiding the decomposition ( 12.1 ). The fric-
tion cone C at the contact point is introduced, a closed convex conical
subset of the 1inear space ~3 (reca I 1 that, speaking of a cone in a I inear
space, one understands that it has vertex at the origin). In traditional iso-
tropic friction, this is a cone of revolution about 1{; generally, C equals
t!Je cone. generated In ~3 by t!Je. set 1> 1+ 1{. So C contains 1{ and lies
entirely on the corresponding side ofT.

PROPOSITION 12.1 T!Je pair of relations( 12.1 ), ( 12.2) Is equivalent to


- u E projT ao/c<1U <12.4>

PROOF Assume that <12.4) holds; hence -UET. Denoting by [1{] the linear
subspace generated in ~3 by 1{, one has
3VEotjiC(R), 3'WE[1{] : -U=V+'W. ( 12.5)

Let us decompose 1{ in the form ( 12.1 ); necessarily p )0, since ( 12.4) implies
50 J.J. Moreau

that a~J~c<10 is nonempty, thus 1{EC. By construction, 1{ belongs to the affine

plane :4 =T+ p1{. Therefore a, .:..<'V = [1{] and< 12.5) means that

-UE o~J~.:..<'V + o~J~J.R) c a'.:..ncCR.> <12.6)

In view of the definition of 1>, one has .:..nC=1>+p1{, thus. using a


translation in the evaluation of subdiffentials,
o!p .:..nJ.lV =olji1>(.R - p1{)= oljl1l(.R T).
Then ( 12.6) entails that ( 12.2) holds in the sense of the Euclidean autoduality
of ~3 . Since U and 1{T are elements of the linear subspace T, the same is
true relatively to the Euclidean autoduality of this subspace.
Conversely, let us assume that 'UETand that< 12.2) holds in the sense of
the Euclidean autodual ity of T, with 'Rr defined by <12.1) (observe that p~O

Is stated at this place). Then <12.2) Is true also in the sense of the auto-
duality of ~3 and, using translations as above, one concludes
- 'U E Oljl ~nJ.lV

Here we need to know whether the inclusion on the right side of ( 12.6)
actually holds as an equality of sets. If p>O, this equality results from a
known calculation rule for the subdlfferentlal of a sum of l.s.c. proper convex
functions in finite-dimensional spaces: in fact there exists a point in the
relative Interior or C=domljlc where ~~~~takes a finite value <see [34),

Theorem 23.8). In that case, by going from <12.6) backward to <12.5), one
estab11shes <12.4).
It Is only for p=O that the assumption of bounded friction. made once
for all In the preceding, has to be used. If p=O, the set 1> reduces to {0}, so
( 12.2) simply consists In the assertion:" U arbitrary in Tand 'Rr = o ";one has
Unilateral Contact and Dry Friction 51

to establish that such is also the meaning of (12.5) in this special situation.
For ~ = 0, the subd1fferent1al oi!Jc(.R) consists of the polar cone co of c.
Because the section of C by the plane Jl (constructed for instance with P= I)
is compact,co contains the vector -1( in its interior (see e.g.[35], parag.8.7).
Consequently, the proJection of co to T equals the whole of this subspace;
tt1is completes the proof.
In turn ( 12.4) may be transformed as follows [25]:

PROPOSITION 12.2 Define in :3 t!Je extended real function e


u- ecrn =~nun 2 +'r<U>. U2.7>
Then( 12.4) is equivalent to
( 12.8)

PROOF Since ~Ill is a smooth function, with gradient mapping equal to


identity, one has for every U in X3
aeCU)= U + oiJIT(U).

Now, OljiT(U) equals [1(] if 'UETand, otherwise, is empty. Therefore, (12.8) is

equivalent to
UET and OEo~!~c<~>+U+[1(],

which is precisely ( 12.4).

Incidentally, observe that resistance laws involving a pair of subdlf-


fential mappings, as in ( 12.8), have a wider interest than describing contact
friction in three-dimensional space. For instance, in Plasticity, this form
may be used in formulating constitutive laws for which the flow rule is not
52 J.J. Moreau

"associated" with the yield criterion.

Recall that T and C, in what precedes, depend on the position q of the


system, under the assumption fCQ)=O. Let us make the same writing
meaningful also for positions wh/c!J do not Involve. contact.
To this end, we shall agree that, when f(qkO, the cone C(q) reduces to
{0} and that T(q)=I"3. Then the relations< 12.4) or C12.8) simply express that
1{=0, with 'U arbitrary in IR 0 .
Furthermore, In what concerns evolution problems under condition f~O.

it Is Immaterial to choose any (adequately smooth) extension of the mul-


tifunction q-C(q)ci"3 to the case f(q)>O. Similarly, the linear mapping
Gq:IR 0-I"3 will be extended to such q, as well as the normal unit 1(Q. with
attention to preserving (3.2). The reason for such extensions lies in
numerical methods, where a certain amount of violation of the desired
inequality f~O has naturally to be faced.

The set of the values of 1{ EI"3 that ( 12.4) or ( 12.8) make correspond
with each 'UEI"3 <actually the empty set if 'UfT) is a cone, since the mul-
tiplication of 1{ by any strictly positive number leaves alllc<R) invariant.

Like in preceding Sections, this fact wi II now prove essential, as we come to


formulating Nonsmooth Dynamics in terms of measure differential
inclusions.
In the course of a smooth motion, the contact force 1{ is a function of
time that we may denote by 'Pi. This in fact Is the density, relative to the
Lebesgue measure dt, of the local contact Impulsion d'P, an I"3-valued
measure on the time interval I. For nonsmooth motions, d'P can no more be
Unilateral Contact and Dry Friction 53

expected to possess a density relative to dt, but in any case this measure
may be represented in the form 'Pvd~. where d~ is a nonnegative real measure
on I and 'Pv EL\.lc(l dbf; ~3 ) The 1Rn-valued contact impulsion dR, as
introduced in Sect.7 for insertion into the measure equation of Nonsmooth
Dynamics (74), has the form dR=R~d~, and. simi Jarly to (4.7),
Rv<U=G;w 'Pv<o <12.9>
rlOlcls for every t.
As far as Coulomb's law is accepted for the description of dry friction,
one naturally admits that, in possible nonsmooth motions, the density 'Pv<t>
of the local contact impulsion will be related, for every t, to the sliding
velocity 'U through the same relation as 1{ is in ( 12.4), or equivalently in
<12.8). This, at least, raises no discussion when t-.'U is continuous. At ins-
tants of velocity jumps, we decide tllat tile same relation will llold wltll 'U
replaced by Its rlgllt-llmlt, namely u+ =Gqu+, since the linear mapping
Gq: !Rn --.~3 continuously depends on q. This assumption entails that U\t)
belongs to T<q(t)) for every t (an immaterial assertion when f(q(t))<O, since
it has been agreed that T(q(t)) = :3 in this evenn In view of Proposition 3.1,
this is equivalent to u\t>ET<q(U), a property which, in the frictionless case
of Sect.9, has been identified as characterizing the "softness" of unilateral
constraints. We thus are induced to put the following definition.

DEFINITION 12.3 Tile unllaterc71 constraint Investigated above Is ~-aid soft


with Coulomb friction 11: In an,v mot/on wltll lb. v. veloci(Y lunct/on.. tile
contact Impulsion llas tile lorm d'P='Pvd~, witll 'Pv EL \oc<l, d~ ;:3) ve.ri-
(''1119 lor every t

- u+ <t> E pro ~<q< 0 >a"'c<q< t> ><'Pv <t>>. ( 12.1 0)

(recall tllat T(q)=~3 and C(q)={O} wllen f(qkO) or tile equivalent form
54 J.J. Moreau

given to a relation of this sort /Jy Pro,o 122

Because this relation is conical with regard to the element Tv (t), t11e
reasoning already used in the proof of Prop.8.2 shows that the choice of the
nonnegative real measured~ is immaterial, as long as d'Ppossesses, relative
to it, a density function.

REMARK 12.4 Since we have agreed to extend the definitions of T(q) and C(q)
to positions such that f(q)>O, <12.1 0) makes sense also in that case. But, as
previously observed, this relation implies that u\t) belongs to T(q(t)), hence
to V(q(t)). In view of Prop.2.4, this secures that, provided t!Je im!ial data
satts~:vf(q<t 0 ))~0, inequali"v f(q(t)h;O will hold t!Jrou_q!Jout I.

REMARK 12.5 The validity of ( 12. I 0) for t=to calls for some comments. In
beginning this section, we recalled the meaning given to the initial condition
u<t0>=u 0 of an evolution problem. It is understood that u0 equals the
left-limit u-<t 0) in some anterior motion, during which the unilateral
constraint was already in effect. In particular, one may have f(q 0)=0 and u0
interior to the half-space -V<q 0); this implies that a collision takes place at
instant t 0. Then softness, as expressed by ( 12. 10), makes that u+<t 0) belongs
to the linear space TCq0)=V<q 0>n-V(q0). But the case f(q0>=0 with u0
interior to V(q 0), i.e. Initial velocity implying contact break, Is excluded
from the present study. This will cause no great inconvenience in practice.

REMARK 12.6 Put the notation Gq<V(q))=V(q); this is the closed half -space
of :t3 lying on the same side as 1{Q with respect to T(q). The set a!c<P~,) is
contained in co, the polar cone of C, which in turn is contained in the half-
space -V(q). Then, In ( 12.1 0), the operation projT(q) might equivalently be
Unilateral Contact and Dry Friction 55

replaced by the proximation mapping to V(q). This would be numerically


inconvenient, but is llable to improve the consistency of some further
developments.

REMARK 12.7 In applications, describing dry friction through Coulomb's law


can only provide a rather crude approximation. However, as this law retains
the essential features ot' the phenomenon, its use at the stage of a first
study is extremely valuable in numerous situations. As soon as this law has
been adopted, under its traditional form, there is little doubt that its
generalization ( 12.1 0) can also be accepted for every motion in which the
velocity function is continuous, even in the absence of local absolute
continuity. We shall discuss later its use in the case of a veloa~yjump.

13. TWO-DIMENSIONAL CONTACT


In the same setting as in the above Section, we now make the following
additional assumption. However three-dimensional the contact may physi-
cally be, we sr1all suppose that, for every q such that f(q)=O (and also for
f(q)~O. after the extension we have agreed to make) tile range Gq<IRn) of Gq
reduces to a two-dimensti:mal su!J~;pace Wq of x3. Such is the case, for
instance, if the part 11 1 of the system is astrained, by the primitive
constraints, to only perform motions parallel to a fixed plane. The condition
found in Remark 3.2, in order that G;T{Q;rO will be supposed fullfilled, l.e.
WQand the tangent plane Tq are distinct: let us denote by l a unit vector of
their intersection.
Due to the expression (4.7) of the covariant components of the contact
force :R (or, in Nonsmooth Dynamics, the covariant components of the density
Pv(t) of contact impulsion). the dynamically significant information
56 J.J. Moreau

concerning this vector is entirely conveyed tt1rough its equivalence class


modulo ker G;. This kernel equals the subspace of T 3 orthogonal to 'Wq,
consequently, we may in the sequel replaceR by its orthogonal proJection to
'Wq, also called R, by abuse of notation.
We shall come back, in Sec. 17, to a dJscuss1on of what m general
becomes the three-dimensional law of friction under such a geometric
two-dimensional reduction. Let us restrict ourselves at present to tile usual
case where the result is simply the familiar two-dimensional version of
Coulomb's Jaw. The considerations of the preceding section might readily be
adapted to this case. Here we shall rather choose to express the two-
dimensional Coulomb law in the followmg alternatwe form. For more readi-
bility, the subscript q will momentarily be omitted.
Tnere ex1st in t!Je vector /}lane 'W two !Jail-lines 1> + and 1> - ,
emanating lrom t!Je origin and lyi17[l wit!J res/}ect to T, on t!Je same s1de
as 1{. The convex cone C now equals their convex hull. In smootr1 motions,
the sliding velocity is essentially an element of rnw, say U= sf with sEIR.
Coulomb's law col7sists of t!Je t!Jree im/}1/cat/ons
S>O :::::> RE 'D + ( 13. 1)
S<O :::::> RE 'D ( 13.2)
S=O :::::> REC ( 13.3)
The angles that'D+ and 'D _make with I have respective tangent equal to
-I/~+ and 1/~_,where the positive real numbers ~"'andy_ respectively are
the familiar friction coefficients correspondmg to positive and negative
sliding.
All these elements are defined for such q that f(q)=O; as before. we
shall imagine an extension of them to every q such that f(q)~O (at least m
Unilateral Contact and Dry Friction 57

a neighborhood of hypersurf ace f =0).


The vector plane W depends on q ; it will prove convenient to send it
..,
onto a fixed copy of IR"', called the calculation plane, equipped with the
usual base consisting of vectors 1 = (I ,0) and j = <0, I). To this end, a
regular linear mapping Tq:Wq-1R 2 will be constructed, in such a way that
Tql is a positive vector of the first axis. Denote by GQ: IR 0 -IR 2 the product
mapping 1QGq. Assume that the real function f is at least c2 ; then it will be
possible to choose Tq in order that Gq depend on q in a c 1 way.
Let us mean by Tq... the transpose of 1Q , in the sense of the natural
scalar products of IR 2 and W, and denote by D+ and D_ the images of 1l +and
-1
1'1 _under Tq
There would remain enough arbitrariness in the choice of Tq for making
these images equal two fixed half-lines of IR 2, e.g. the half-lines generated
by (-I, I) and (I, I) . The drawback is that the two friction coefficients f +
and f _ would then have to depend smoothly on q . So we shall not use this
trick in the sequel and only assume that t/Je convex /Jull, say C, of D+ and
D_ contains J In Its lnter;or.
Under these notations, the friction law ( 13.1) to ( 13.3) may equivalently
be formulated as a relation of the same form between R= Tq"'- 11{ and u=
:rqu = Gqu = oi. Due to the assumptions made, o is a real number of the same
sign as s or vanishmg with it. Henceforth, there onry is to replace s , 1'1 +'
1'1_, C respectively by o, D+, D_, C.
In order to express the dynamics of the considered system, one has to
bring together the above reaction law and the Lagrange equations. The
covariant components of the contact force make the element r of IRn, related
toR through
58 J.J. Moreau

( 13.4)

After having so restated the two-dimensional law of Coulomb, we now


propose to extend it to Nonsmooth Dynamics, in the same line as in Sec. 12.
Recall that, in such a context, velocity functions are significant only through
their one-sided limits. Instead of u, 71 and U, the rig!Jt-lirmts u+, 71+= Gqu+
and u+ = Gqu+ are introduced mto the above writing. The contact force 1{ is
replaced, for every t, by the value P~(t) of the density function of the
three-dimensional contact impulsion, or equivalently by the orthogonal
projection of Pt,<t> to 'W(q(U). When coming to the use of the calculation
plane, the function Pt,= 'Fq*- 1 pv is considered.
The dynamics of the system is now expressed by the Lagrange equation
In IRn
A(q) u~ - K(t,q,u) t~ = R~, ( 13.5)
to be joined with
R' = G* P' ( 13.6)
v Q v
3GEIR: U+= a1 (13.7)

0'>0 ~ P~E D+ ( 13.8)

0'<0 ~ P~E D_ ( 13.9)

0'=0 ~ PVE c' ( 13.1 0)


for every t such that f(q(t));;~:O.

On the contrary, when f(q(t))<O , then R~=O.

14. VELOCITY JUMPS IN FRICTIONAL DYNAMICS


The formulation given above, for the Nonsmooth Dynamics of a system
with single possible contact and two-dimensional Coulomb friction, will
Unilateral Contact and Dry Friction 59

now be used in discussing the event of a velocity jump.


This means that, at some instant t 5, the IR 0 -valued measure du
possesses an atom, the mass of which equals
u+(ts)- u-<ts) = uv<ts) lls.
Here, the positive number Vs is the mass of the atom that the measure dlJ
should possess at point t 5.
Recalling that the Lebesgue measure dt has no atom, one derives from
( 13.5) that, at t=t5 ,

u+- u- =Aq-1 Gq* P1:'' lls ( 14. 1)


If f(q(t 5))<0, i.e. no contact, one has Pt,<t 5 )=0, so no jump of u can occur.
We therefore shall assume f(q(t 5 ));1!:0 (equivalently f(q(t5))=0) and denote
simply by G the value of Gq at q=q<ts>- Then, for t=t5 , one has Rv=G"'Pt,. Put
Pv(t 5)V 5=P and apply the linear mapping G:IR 0 -+IR 2 to both members of
( 14.1 ); this yields
u+- u- = HP, U4.2>
where H denotes the s,ymmetric 2x2 positive defimte matrix GA-l G*.
This has to be joined with two-dimensional Coulomb law, expressed as
in Sec.13 by
3OEIR : u+ = o 1 ( 14.3)
O>O~PED+ ( 14.4)
0<0 ~ P ED - ( 14.5)
0=0 ~ P E C. ( 14.6)
We are to discuss how, starting with given u-, the system of conditions
( 14.1) to ( 14.6) allows one to determine u+.
Combining <14.2) and <14.3), one obtains
P =-H -1 U- + oH-1 I, (14.7)
60 J.J. Moreau

which expresses that P belongs to the line 6 of IR 2 drawn through the


point -W 1u-, with W 1t as directing vector. Easy calculation (for instance
by observing that 6 is orthogonal to Hj) yields that this line intersects
the second axis at point
S = -(J.U-)j/(j.HJ). (14.8)
Due to the way we have chosen the mapping T, the real number
J. u-=T''j.tr has the same sign as 1{:r.c =G;x. u- or vanishes with it.
Through Prop.3.1, one concludes that the coordinate of 5 on the j axis is
positive, negative or zero if and only if the same is true for u-.\7f(q(t 5)).

First case: impact.


This Is the event where f(q(t 5 ))=0, with u-<t5 ).\7f(q(t5 ))>0. Consequently
f(q(t))<O on some left-neighborood of t 5, I.e. t 5 is an instant of nontangen-
tial collision. The point S lies in the interior of C , so the line 6 inter-
sects C and does not pass through the origin. Observe that l.W 11>0, i.e. the
vector W 1I directs 6 from left to right. Therefore, the point P, as expres-
sed In <14.7), lies on the right of W 1u- If O>O and on the left In the
reverse case. By comparison with ( 14.4) to ( 14.6), one concludes that the
formulated set of conditions is satisfied if and only if P equals t/Je nearest
point to WI u- in t/Je intersection 6nC.
So the problem of determining u+ possesses a unique solution in that
case.

Second case: sliding.


We now assume that f(q(t 5 ))=0, with u-<t5 ).\7f(q(t5))=0 (equivalently
u-<t5).\7f(q(t5))(0, since the anterior motion Is assumed to agree with f(O).
Unilateral Contact and Dry Friction 61

Then 5=0 , i.e. the line 6 passes .through the origin.


If 6nc consists only of the origin, the formulated set of condition admits
P=O, hence Lt = u-, as unique solution: no velocity jump occurs.
If 6 intersects also the interior of C, one has to determine whether 1t is
possible for P to lie in this interior. In view of condtttons (14.4) to (14.6),
this requires U+=O.It is in fact a solution if -W 1U-EintC: then the sliding
suddenly sticks. Concurrently P=O, with u+ = u-, i.e. no velocity jump, is also
a solution in this case.
If 6 contains one of the half-lines D+ or D_, a value of P on this
half -line meets the requirements provided the resulting value of u+ =
u-+HP= oi agrees with conditions ( 14.4) to ( 14.6). For instance, imagine
D+c6: these conditions require o~O. This can happen only if -W 1U-ED+and
then every value of P belonging to t/Je line segment [-H-lu-, 0] is a
solution. The corresponding values of u+ cover the line segment [0, u-].

REMARK 14.1 The latter is the most interesting item of the discussion. It
shows that, contrary to the frictionless case, an episode of smooth motion
with persistent contact may end with a velocity jump without any collision
occurring This is a dynamical analogue to the locking effect, well known in
the statics of mechanisms with dry friction.
The earliest reference we know of, where such a possibility is asserted,
is a note by L. Lecornu [26]. At the time, a controversy has been opened by P.
Painleve [36], with the observation that, in systems involving Coulomb
friction, some initial value problems could have no solution, or also several
solutions. In addition, the behaviour of the system depended on its constants
in a discontinuous way. To Palnleve, and later to E. Delassus [37][38], these
findings seemed In contradiction with the very bases of Physics. In the
62 J.J. Moreau

subsequent years, different opinions were sustained by such authors as F.


Klein [39], R.v.Mises [40], G. Hamel [41] or L Prandtl [42]. Even after H.
Beghin [43][44] had clearly demonstrated that the incriminated findings
actually agreed with common observation, some suspicion remained in the
scientific community that Coulomb's law could be intrinsically Illogical.
Today, one Is accustomed to meet multiple solutions or the absence of
solution to physical problems, usually ascribing these facts to the nature of
the treated information, without opening any discussion about determinism
in Physics. One is also familiar with discontinuous behaviour.
Dynamic locking, that we propose to call a frictional catastrophe, Is
commonly observed in practice. The example of the chattering motion of a
piece of chalk driven at an angle against a blackboard, so that a dotted line
Is drawn, was already put forward by E. Delassus [37]. A model of this
phenomenon is presented In [30], with some drawings generated by a
computer using the time discretization procedure of Sect.15 below. This
displays an Instance of a "stick-slip" motion. Depending on the system
constants, frictional catastrophes and Intermittent contact breaking may
occur or not. No attempt so far has been made at comparing this model,
quantitatively, with experimental measurements.
A very simple example of frictional catastrophe is presented In Sect. I 5,
as a demonstration of the ability of our numerical technique to handle
nonsmooth solutions.

REMARK t 4.2 From the mathematical standpoint, we think that little has to
be retained of the early discussions on the subject. At the time, differential
equations were impltcttely understood in the sense of the elementary theory:
Unilateral Contact and Dry Friction 63

solutlons should possess some regulated functions as their derivatives of


the highest order involved. In other words, It was admitted that the
acceleration q" possessed a right-limit and a left-limit at every t. Certain
of these limits played an essential role in the discussions by Painleve and
his followers Unfortunately, m usual instances where an Interval of smooth
motion precedes the catastrophic instant t 5 , one finds that the norm of q"
actually tends to infinity when t tends to t 5 from the left. The concept of a
measure differential inclusion, on which the present lectures are based,
provides a more synthetic view, since it allows one to express Dynamics on
the whole interval I , Including t 5 , and does not rely on the existence of
one-side accelerations.

REMARK 14.3 There remains to discuss whether the velocity jumps agreeing
with the constraint law (12.10) are physically realistic. We have already
stressed that, even in the absence of friction, predicting safely the outcome
of a shock would require some high order of information, actually una-
vailable in engineering situations. Things naturally become worse if friction
is entered into account. What do we know about the Physics of high pressure
friction, during the "very short" interval of time on which the velocity
change takes place? Already for the frictionless case, we have in Sect.6 been
reluctant in accepting the invariance of the direction of \lf(q) during this
interval. The latter invariance, if admitted, makes of the no-friction
assumption a time-independent linear condition imposed on the contact
force at every instant; therefrom, the normality of the contact percussion is
inferred, by integration on the interval. On the contrary, Coulomb law
imposes on contact force a nonlinear condition. Even if one assumes this
condition independent of time, it cannot be expected in general to commute
64 J.J. Moreau

with integration. The contact percussion can safely be asserted to verify


( 12.1 0) only as far as the sllding velocity is sure to remain zero or to keep a
constant direction for the short duration of the investigated shock. It is of
course in the case of two-dimensional friction that the latter event proves
the easiest to discuss.
The reader may find In [2] an attempt at analysing frictional impact, in
the line formerly suggested by G. Darboux. This consists in investigating the
variation of the velocity as a function of some "micro-time", relatively to
which the system position remains a constant. Even so, the conclusion is
subject to some assumption about the shock end which seems difficult to
justify.
In conclusion, the concept of a frictional and soft contact, as involved in
Definition 12.3, only generates a special model of frictional shock, with the
advantage of good theoretical consistency. As we shall see in the next
Section, the corresponding motions are also very tractable numerically.
Physical situations to which this model is relevant very probably exist, but
experimentation is still needed to safely Identify them in practice.

15. ALGORITHM FOR TWO-DIMENSIONAL FRICTION


We now present a time-discretization procedure for computing a motion
under the conditions of Sec.13. The notations are the same as in the
frictionless case, exposed in Sec. II.
Stage !. Calculate tM=t 1 +~h, the midpoint approximants
QM=Q 1 +~hu I EIRn, AM=A(qM) EIRnxn , KM=K(tM,qM,ul) EIRn
ana the "loose velocity"
-1
ul =u 1+hAM KM.
Unilateral Contact and Dry Friction 65

As in Sec.13, a single constraint inequality, say f~O, is taken into account.


1f f(qM) < 0 or uL.\?f(qM) ~ 0, then make uF = uL and go to Stage 3.

Stage 2 If, on the contrary, f(qM) ;>- 0 and uL.\?f(qM > 0 , contact Is
estimated to have effect on the considered time-step. One constructs a
discrete analog to the measure differential inclusion of Dynamics
du = A- 1(q) KU,q,u) dt + A- 1(q) dR
by equalling some approximate values of the integrals on <t 1,tF) of the
respect 1ve members, namely
UF - u1 =hAM-1 KM + AM-1 R. ( 15.1 )
Let GM denote the value taken at q=qM by the linear mapping GQ: IRn -+IR 2
defined in Sec.13. Then an approximate version of ( 13.7) reads
R = G~ P,
where P, an element of the calculation plane, is an approximant of the total
contact impulsion on U1e interval <t 1,tF). Puttmg u1 = GMul and UF = GMuF ,one
derives from <15. 1) that
-1
UF - U1 = hGMAM KM + HMP. ( 15.2)
Here, similarly to what has been done in Sec. 13, one denotes by HM the
symmetric positive definite 2x2 matrix GMAr1 1 G~.
According to the decision made In Sec. 11, of considering uF as a
simulation of u+, the softness condition ( 13.7) will, in the present discre-
tization procedure, be transcribed as
30 E IR : UF =0 i. ( 15.3)
Discretizing the two-dimensional Coulomb law consists in relating P to UF
by the system of implications
0>0 :::> PED+
0<0 :::> PED
66 J.J. Moreau

CJ=O => PEC.


In the case where the friction coefficients depend on Q, the elements D+. D_
and C will be evaluated at q = QM.
The determination of P and uF from this set of conditions is similar to
what has been done in one of the cases of shock investigated in Sec.14. If one
puts Ul =GMul , ( 15.2) becomes
UF = Ul + HMP. ( 15.4)
By combination wlth <15.3), this yields
-1 -1
3CJEIR : P = -HM UL + crHM i .
This expresses that P belongs to the line 6 of IR2, drawn through the known
point -Ht1 1UL wlth known directing vector Ht1 11 . This llne intersects the
second axis of IR 2 at point
SM = -(j.UL)j/(j.HMJ).
A similar expression was discussed in ( 14.8). Since the present computation
stage Is developed under the assumption uL.Vf(qM)>O, one finds in the same
way as in Sec. 14 that J. UL <0, so SM Is sure to be interior to C. Observe in
addition that P lies In 6 on the right or on the left of -Ht1 1UL according to
the sign of a. One concludes to the existence of a unique solution P, charac-
terized as t!Je nearest point to -Ht1 1UL In the Intersection 6nc.
For computation it is more convenient to formulate the same as follows:
-1 -1
If -HM UL E C, then P = -HM UL .
-I
Otherwise -HM UL.1 < o => P = 6nD +
-I
-HMUL.i>O => P=6nD_.
After P is calculated, one derives uF from< 15.1 >.namely
UF =UL + AM-1 GM P.
Unilateral Contact and Dry Friction 67

Stage J'. The computation step finishes with qF = qM +~huF.

REMARK 15.1 A more intuitive description of the above discretization


procedure may be found in [24]. Instead of relying on the calculation plane, it
uses the image of C under G;-l, a two-dtmenstonal cone in IRn. Thts makes
the comparison with the frictionless case clearer, but numerically Is less
effective. The calculation plane is also useful at the stage of deriving the
inequalities r{eeded in the study of existence and regularity of solutions.

REMARK 15.2 The case -H~ 1 ULEC yields UF=O, i.e. zero sliding velocity <the
discuss JOn here is simpler than that of the simi Jar geometric construction
made in Sec. 14, because SM is certainly interior to C).
In that connection. the algorithm works very well to compute a motion
involving the event which, in Remark 14.1, we have call~Q a frictional
catastrophe. Now, we have just seen that each computation step is deter-
ministic, i.e. it yields a untque pair qF, uF. This contrasts wtth the conclu-
sions of Sec.14, showing multiple possible outcomes for such a cata-
strophe: all the points of a line segment are solutions in wM"t concerns the
contact percussion P and similarly in what concerns u\t5 ) or U+(t5 ) (the
latter may take any value between zero and u-(t 5 )).
In fact, the algorithm is able to approximate any of these solutions. As
soon as the successive discretization intervals are chosen, a sequence of
values of uF is unambiguously generated. This sequence is smooth, except for
a jump in one of the intervals, said catastrophic. Before this jump, the
computation of the motion, from given initial data q(t0), u(t 0), yields
consistent results, for arbitrarily fine discretizations. But the value of the
jump obtamed in a catastrophic interval depends on the ratio in which the
68 J.J. Moreau

exact instant divides this interval. So, calculations made with different
discretization meshes may yield different results after the catastrophe.

EXAMPLE 15.3 A round-tipped rigid body :B 1 performs motions parallel to a


vertical plane. It Is submitted to gravity and confined by a horizontal fixed
boundary :B0, with friction coefficient equal to 0.5. Initial conditions are
those of contact, with negative angular velocity and sl idmg velocity directed
to the right.
On Figure 15.1, the computer has drawn the profile of :B 1 for every third
step of the time-discretization <numbers refer to these steps). After an epi-

FIXED BOUNDARY

Figure 15. 1
sode of persistent contact with sliding of constant direction, a catastrophe
occurs. The horizontal component of the velocity of the lowest particle of :B 1
(this equals the sliding velocity in the case of contact) presents a sudden
Unilateral Contact and Dry Friction 69

drop. At the same instant, contact is broken, with zero normal velocity. In
the process, a percussion is imparted to ~1 from ~0 , making the negative
angular velocity increase in magnitude.
Though discretization is rather rough, the corrective procedure of the
possible constraint violation. described in Remark 11.1, has not been used.
Drawing is however found to comply very well with the unilateral constraint.
On Figure 15.2, the horizontal component of the velocity of the lowest
particle is plotted versus time. All curves correspond to the same initial
conditions as above, but are computed with finer discretization. In order to
display the multiple possible outcomes of the catastrophe, computation has

HORIZ. VELOCITY OF LOWEST POINT

TIME

Figure 15.2
been repeated, each time with uniform time-mesh, but successively using
different values of the step-length, namely
70 J.J. Moreau

h = 0.00046 - 0.0001 RND.


Here RND denotes the built-in random sequence of the computer, with values
in [0, 1[. The maximal catastrophe, i.e. sliding velocity dropping to zero, is
repeatedly obtained, more than two times out of three on an average, and
followed with a well defined contact-free motion. The other curves show
various sliding velocity drops of smaller amplitudes. More extensive expe-
rimentation has confirmed that each value of the sliding velocity drop
corresponds to a well defined consequent motion. Statistically, the values of
the drop are not uniformly distributed: frequency is found maximal in the
vicinity of zero drop.
As expected, consistent results are obtained regarding the motion prior
to the catastrophe. It is apparent that the rate of change of the sliding velo-
city tends to -oo on the left of the catastrophic instant The acceleration
qN: J...,.IRn thus beeing unbounded, cannot possess a left- limit at the cata-
strophic Instant.

16. AN EXAMPLE OF THREE-DIMENSIONAL FRICTION


We suppose in this section that the system consists of a single particle
of unit mass, confined in a region 4> of physical space by an unmoving
boundary L This is a model of a small object which, in the case of contact
with the boundary, Is assumed to slide on It, without rolling nor tumbling.
Then q 1, q2, q3 may be Identified with the particle coordinates, relative to
some inertial orthonormal Cartesian frame and A(q) consists, for every q, in
the unit matrix.
In addition to the possible frictional reaction of the boundary, the
Unilateral Contact and Dry Friction 71

particle is submitted to a force given as a smooth function F of t, q, u. The


case or a boundary with prescribed motion may be reduced to this one,
through changing the reference frame; there only is to include in F the
fictitious forces, thanks to which the new reference frame may be treated
as inertial.
The particle dynamics is expressed by this equality of 1:3-valued
measures on the time-interval I
du = dR + F(t, Q, U) dt, ( 16.1)
i.e. after representing vector measures by density functions relative to some
nonnegative real measure dV,
U~(t) = R~(t) +F(t, Q, U) t~(t),

an equality to be satisfied for every tEl.


In the present case, the mapping Gq reduces to identity for every q. Then,
using the law of frictional contact in the form ( 12.8), one obtains the
measure differential inclusion
0 Eaii'C(q)(UV- f(t, Q, u)tV) +aeq(U+). (16.2)
The feasible region q> of 1R 3 Is defined as before by a single Inequality f(q)(O.
For f(q)=O, C(q) denotes the friction cone at the point q of the boundary.
Again, let us agree to extend its definition, in a smooth arbitrary way, to the
values of q such that f(q))O. In addition, C(q) is Interpreted as reducing to
the zero of 1R 3 when f(q)<O. With every q such that f(q))O, the vector plane
Hq) also Is associated, orthogonal to ~Vf(q). For f(q)~O, we agree t1
understand T(q) as consisting of the whole of 1R 3. The extended-real function
eQ, as defined 1n (12.7), equals ~11.11 2 + 'i'Hq)
In particular, at every t such that f(q(t))<O, one has aeQ(u<x> = {x} for
every xeiR3, whtle the value of alJicCq<m<x> equals R3 for x=O .and. 0
72 J.J. Moreau

otherwise. Therefore ( 16.2) reduces in that case to the differential equation


ui- F(t,q,u) = 0 of contact-free motion.
Recall that a condition such as ( 16.2) implies that q(t)E<P for every t, as
soon as this is assumed to hold for t=t 0.
Some cases of existence, for the solutions to the coresponding imtial
value problem, are investigated in [ 13].
Here we shall only present a time-discretization rnetr10d for their
approximate computation [25]. With the same notations as in previous
sections, each time-step runs as follows.

Stage I. Calculate tM=t 1 +~h , the midpoint approximant qM=q 1 +~hu 1, the
force estimation FM= F<tM,qM,u 1) and the "loose velocity" uL =u 1 + h FM.

Stage2.
If f(qM) < 0 or UL.V'f(qM) ~ 0, then UF = UL .
Otherwise, uF is determined by a semi-implicit discretization of (16.2).
In view of the positive homogeneity of the multifunction a~c this is
oEaljlcCuF- uL) +aecuF),
whith the cone C and the function e evaluated at point qM Using the
definition of e, one gives to this inclusion the form OE alji 0CuF) +uF, where D
denotes the set T(qM)n<CCqM)+uL). Through elementary Convex Analysis, this
means that uF equals the proximal point to the origin in this set, with regard
to the usual Euclidean metric of 1R 3 . For the traditional, isotropic, Coulomb
law, Dis a disk, so the proximal point is specially easy to calculate.

REMARK 16.1 Even if one assumes isotropic friction, computing the motion
Unilateral Contact and Dry Friction 73

of systems involvmg three-dimensional friction is not in general as simple


as above, due to the role of the mapping Gq The determination of uF, at each
t1me-step, usually is a non convex problem, which may possess several
solutions.

EXAMPLE 16.2 The numer1cal technique presented here has been applied [24]
to the motion of a particle P submitted to gravity and confined by a plane
boundary with prescribed motion. This particle represents an object which,
in the case of contact. may sl1de on the boundary, without rolling nor
tumbling. The plane boundary may be the ground surface, in the course of an
earthquake, or also a vibrating table. Motions of the following sort are
common in industrial conveyors.

z z

0 ____ p
----IlL--

-- 0

Figure 16. 1
The vibrating table 1s assumed to have the motion of the shaft in a
crank and s!Jaft mechanism. Let orthonormal axes Oxyz be attached to the
table, with Oxy in its surface. Axes oy anct oz move In fixed plane OYZ. The
point 0 is guided along a segment of the line OY. The point A (0, a, 0) of the
table is astrained, by an eccentric, to describe, at constant velocity, a circle
in the plane OYZ, centered on OY. Therefore, the plane OXY is the mean
posit1on of the vibrating table.
74 J.J. Moreau

In order to produce a clear pattern of trajectories, the whole machine is


set at a slant: the plane OZX is vertical, but OZ is not in line with gravity.
Hence, OX is the direction of steepest descent in the plane OXY and
determines the general trend of the motion of P.
A computer program, using the numerical technique described above,
draws the projections to Oxy an to Oyz of the trajectories of P relative to
t!Je table. No experiment has so far been conducted for comparison with
reality.
At the initial instant, P is left on the table with zero relative velocity.
Subsequent trajectories are drawn for several choices of this initial
position, at various distances of Ox. Motions taking place sufficiently far
from Ox involve intermittent contact break; the loops then observed on the
Oyz projection correspond to the parabolic motion that P ~1ave, when referred
to fixed axes.
Here are the values of the system constants, understood as referring to
c.g.s. units. Gravity equals 981; OZ makes an angle of 13 with the upward
vertical direction. The eccentric has radius 0.5 and rotation speed 1o rps. The
length OA equals 50. Friction coefficient is taken equal to 0.4.
Wlth this values, it turns out that the table shake is strong enough for
leaving no place where P could rest without sliding. The whole pattern of
trajectories admits Ox as an axis of symmetry. Depending on the direction in
which the eccentric rotates, this line is a locus of attraction or of repulsion,
a fact which could be asserted from qualitative reasoning. More inexpected
is the existence of other lines of attraction or of repulsion (they exchange
their roles when rotation is reversed), parallel to this one. Such a "quantic"
effect appears to be connected with the number of flapplngs that the table
Unilateral Contact and Dry Friction 75

performs while the particle runs through each episode of contact-free


motion. Actually, the farther they lie from Ox, the more confuse these
attraction loci appear, due to the chaotic behaviour that trajectories then
have.
Figure 16.2 shows the trajectories of P, consequent to 15 initial
positions equally spaced from y=5 to y=320; the eccentric rotates In the
reverse direct ion to that indicated on Fig. 16. 1.
2

0 A

Figure 16.2
Fig.16.3 displays a larger scale drawing, corresponding to the same
direction of rotation as on Fig.16.1. The trajectories correspond to 7 lnfttal
positions equally spaced from y=5 toy= 150.
76 ].]. Moreau

ImlllrrfrllftM'iii2 LWJ.IJ
0 y_

Figure 16.3

17. STATIC AND DYNAMIC FRICTIONS


In many familiar stt'uations, friction appears h1gher when U1e contacting
bodies are to be set in motion from rest than during an episode of
established sliding. Such an effect of "tangent sticking" is traditionally
accounted for by introducing a larger value for the static friction coeffi-
cient, i.e. relative to zero sliding velocity, than for tr1e dynamic one.
The numerical techniques proposed in the foregoing sections handle this
refinement without difficulty. There only 1s, in each step of t1me
discretization, to make the cone C depend on the sliding status in the
antecedent step For instance, in the algorithm described m Sec.15, U1is
Unilateral Contact and Dry Friction 77

antecedent step Involves sliding or no sliding, according to -Hf1 1UL belonging


to cor not. This precisely makes one of the branching conditions one anyway
had to consider.
At the stage of the analytical formulation, one has to make the cone C
depend not only on q, but also on 'U or, if the possibility of a velocity jump is
considered, on 'U-. At first glance, this seems to reduce the advantage of the
formulatwns of friction presented in the foregoing. But the example below
tends to demonstrate that, far from being an heterogeneous addition to the
previous theorization, such a refinement actually proves Inherent In the
subject matter. In fact, this example shows that, even if one starts from a
law of friction with single coefficient, it may happen that the logical
derivation of consequences eventually makes some contact force appear to
obey a Coulomb law with coefficient depending on sliding velocity.
Le1 us consider again the sltuation of Sec.l3, namely, in a position q of
tile system, with f(Q)=O, the range Gq<IRn) is assumed to reduce to a
two-dimensional subspace 'W of I"3, different from the common tangent
plane T to the contacting bodies. In a motion with continuous velocity, it is
assumed that the contact force 1{ er3 Is related to 'U through Coulomb's law,
expressed as before in the form
(17.1)

As already observed, 1{ pertains to the equations of mechanics only through


its equivalence class modulo ker G;. a natural representative of which is
1{ = proJw~

Furthermore, since 'U essentially belongs to the tangent vector plane T, It


has the form 'U =sf, where I denotes a unit vector of the line rnw I and
sEIR. We are going to show that the resulting relation between sand ~may
78 J.J. Moreau

have a more general form than it has been assumed in Sect.13.


Let us consider only, for simplicity, the traditional case where cIs a
cone of revolution about 1(. Clearly, for all s>O, the set of the values of :R
that relation ( 17.1 > associates with 'U.=sf consists of a certain boundary
half-line, say 1l+, of the cone. With all s<O Is associated the half-line 1f-,
symmetric to 1f+relative to the cone axis. Finally, to s=O, correspond all the
points of C. The orthogonal projections of 1f+and 1f-to Ware two half-lines,
say 1'+ and :v-. lying on the same side of rnw. The expected relation
between sand~ is then expressed by the three implications
S>O => 1{*E 1)+
5<0 => ~ E 7)-

S=O => ~E projw c.

Figure 17.1
Depending on the span of the cone C and on the angle that W makes w1tt1
1(, the orthogonal projection of C toW may equal the whole of W or some
Unilateral Contact and Dry Friction 79

angular region c* of this two-dimensional space.


In the latter case, c*contains :D+ and :v- but has no reason 1n general to
possess them as its edges: this means that the relation found between U and
R* is equivalent to a two-dimensional Coulomb law with static coefficient
larger tnan the dynamic one. The equality of these coefficients, namely the
simple case studied In Sec.l3, is however achieved If W Is orthogonal toT.
If projwC=W, every value of R"' in W Is associated with U=O, possibly

making with 1{ an angle larger than n/2. In statical problems, this could be
described as a wedging effect.
The above discussion provides an example of the Interaction between the
constraints of the system and the frictional effects at possible contact
pomts. The treatment of systems involving several contacts with Coulomb
friction, a question left aside In these lectures, leads in general to similar
situations .

.
REFERENCES
I. Delassus, E.: Memoire sur Ia theorle des liaisons flnles unllaterales, Ann.
Sci. Ecole Norm. Sup., 34 <1917), 95-179.
2. Peres, J.: Mecanlque Generale, Masson, Paris 1953.
3. Moreau, J.J.: Les liaisons unllaterales et le prlnclpe de Gauss, C.R. Acad.
Sci. Paris, 256 ( 1963), 871-874.
4. Moreau, J.J.: Quadratic programming In mechanics: dynamics of one-sided
constraints, SIAM J. Control, 4 (1966), 153-158.
5. Dunford, N. and J.T. Schwartz: Linear Operators, Part 1: General Theory,
lnterscience Pub. Inc., New York 1957.
6. Dlnculeanu, N.: Vector Measures, Pergamon, London, New York 1967.
7. ~1oreau, J.J.: B.ounded variation in time, in: Topics in Nonsmooth Mechanics
(Ed. J.J. Moreau, P.O. Panaglotopoulos and G. Strang>, Blrl<hauser, to
appear.
8. Pandlt, S.G. and S.G. Deo: Differential Systems Involving Impulses,
80 J.J. Moreau

Lecture Notes In Math., vol. 954, Springer-Verlag, Berlin, Heidelberg, New


York 1982.
9. Moreau, J.J.: Une formulation de Ia dynamique classique, C.R. Acad. Scl.
Paris, Ser.ll, 304(1987), 191-194.
I 0. Panaglotopoulos, P.O.: Inequality Problems in Mechanics and Applications,
BlrkMuser, Boston, Basel, Stuttgart 1985.
II. Monteiro Marques, M.D.P.: Chocs lnelastiques standards: un resultat
d'exlstence, Travaux du Seminalre d'Analyse Convexe, Univ. des Sci. et
Techniques du Languedoc, vol.l5, expose n 4, Montpell ier, 1985.
12. Monteiro Marques, M.D.P.: Rafle par un convexe semi-continu
lnferleurement, d'lnterieur non vide, en dimension finie, C.R. Acad. Sci.
Paris, Ser.l, 299 <1984), 307-31 o.
13. Monteiro Marques, M.D.P.: lnclusoes Dlferenciais e Choques lnelasticos,
Doctoral Dissertation, Faculty of Sciences, University of Lisbon, 1988.
14. Moreau, J.J.: Liaisons unllaterales sans frottement et chocs inelastiques,
C.R. Acad. Sci. Paris, Ser.ll ,296 <1983 >. 1473-1476.
15. Moreau, J.J.: Standard inelastic shocks and the dynamics of unilateral
constraints, In: Unilateral Problems In Structural Analysis <Ed. G. Del
Piero and F. MacerD, CISM Courses and Lectures No.288, Springer-Verlag,
Wlen, New York 1985.
16. Schatzman, M.: A class of nonlinear differential equations of second
order in time, J. Nonlinear Analysis, Theory, Methods and Appl.,2
( 1978),355-373.
17. Buttazzo, G. and D. Perclvale: On the approximation of the elastic bounce
problem on Riemannian manifolds, J. Dlff. Equations, 47 <1983), 227-245.
18. Carriero, M. and E. Pascali: Uniqueness of the one-dimensional bounce
problem as a generic property In L1<[0,TJ;R), Bolletlno U.M.I.(6) 1-A
( 1982), 87-91.
I 9. Perclvale, D.: Uniqueness In the elastic bounce problem, J. Diff. Equations,
56( 1985), 206-215.
20. Moreau, J.J.: Surles lois de frottement, de plastlclte et de viscoslte, C.R.
Acad. Sci. Paris, Ser.A, 271 <1970), 608-611.
21. Moreau, J.J.: On unilateral constraints, friction and plasticity, In: New
Variational Techniques In Mathematical Physics <Ed. G. Caprlz and G.
Stampacchla), CJME 2" clclo 1973, Edlzlonl Cremonese, Roma, 1974,
173-322.
22. Moreau, J.J.: Application of convex analysis to some problems of dry
friction, In: Trends In Applications of Pure Mathematics to Mechanics,
Unilateral Contact and Dry Friction 81

vol.2 <Ed. H. ZorskO, Pitman Pub. Ltd., London 1979, 263-280.


23. Duvaut, G. and J.L. Lions: Les lnequations en Mecanique et en Physique,
Dunod, Paris 1972.
24. Moreau, J.J.: Dynamlque de systemes a liaisons unilaterales avec
frottement sec eventuel; essais numeriques, Note Technique 85-1, Lab.
de Mecanique Generate des Milieux Contlnus, Unlv. des Sci. et Techniques
du Languedoc, Montpellier, 1985.
25. Moreau, J.J.: Une formulation du contact a frottement sec; application au
cal cui numerique, C.R. Acad. Sci. Paris, Ser.ll,302 <1986), 799-80 I.
26. Lecornu, L.: Sur Ia Joi de Coulomb, C.R. Acad. Sci. Paris, 140 <1905),
847-848.
27. Oden, J.T. and J.A.C. Martins: Models and computational methods for
dynamic friction phenomena, Computer Methods in Appl. Mech. and Engng.,
52 ( 1985), 527-634.
28. Jean, M. and G. Touzot: Implementation of unilateral contact and dry
friction In computer codes dealing with large deformations problems, to
appear in: Numerical Methods in Mechanics of Contacts Involving Friction,
J. de Mecanique theor. et appl., Special issue, 1988.
29. Abadie, J.: On the Kuhn-Tucker theorem, in: Nonlinear Programming (Ed.
J.Abadie), North-Holland Pub. Co., Amsterdam 1967, 19-36.
30. Jean, M. and J.J. Moreau: Dynamics in the presence of unilateral contacts
and dry friction; a numerical approach, in: Unilateral Problems in
Structural Analysis 2" <Ed. G. Del Piero and F. MacerO, CISM Courses and
Lectures No 304, Springer-Verlag, Wien 1987, 151-196.
31. Moreau, J.J.: Evolution problem associated with a moving convex set in a
Hilbert space, J. Diff. Equations, 26 <1977), 347-374.
32. Moreau, J.J.: Surles mesures differentielles de fonctions vectorielles et
certains problemes d'evolution, C.R. Acad. Sci. Paris, Ser.A, 282 <1976),
837-840.
33. Jean, M. and E. Pratt: A system of rigid bodies with dry friction, Int. J.
Engng. Sci., 23 ( 1985), 497-513.
34. Rockafellar, R.T.: Convex Analysis, Princeton Univ. Press, Princeton 1970.
35. Moreau, J.J.: Fonctlonnelles Convexes, Lecture Notes, Seminaire sur les
Equations aux Derivees Partlelles, College de France, Paris 1967.
36. Painleve, P.: Sur les lois du frottement de gllssement, C.R. Acad. Sci.
Paris, 121 (1895> . 112-115. Same title, Ibid. 141 (1905), 401-405 and
141 ( 1905), 546-552.
37. Delassus, E.: Considerations sur le frottement de glissement, Nouv. Ann.
82 J.J. Moreau

de Math. (4eme serie), 20 <1920), 485-496.


38. Delassus, E.: Sur les lois du frottement de glissement, Bull. Soc. Math.
France, 51 <1923), 22-33.
39. Klein, F.: Zu Palnleves Kritik des Coulombschen Reibungsgesetze, Zeitsch.
Math. Phys., 58< 191 0), 186-191.
40. Mises, R.v.: Zur Kritik der Relbungsgesetze, ibid., 191-195.
41. Hamel, G.: Bemerkungen zu den vorstehenden Aufsatzen der Herren F. Klein
und R. v. Mises, ibid., 195-196.
42. Prandtl, L.: Bemerkungen zu den Aufsatzen der Herren F. Klein, R. v. Mises
und G. Hamel, ibid., 196-197.
43. Beghin, H.: Sur certains problemes de frottement, Nouv. Ann. de Math., 2
( 1923-24), 305-312.
44. Beghln, H.: Sur l'indetermlnatlon de certains problemes de frottement,
Nouv. Ann. de Math., 3 <1924-25), 343-347.
NONCONVEX SUPERPOTENTIALS AND HEMIVARIATIONAL INEQUALITIES.
QUASIDIFFERENTIABILITY IN MECHANICS

P.D. Panagiotopoulos
Aristotle University, Thessaloniki, Greece
R.W.T.H., Aachen, FRG

ABSTRACT
These lectures concern the study of mechanical problems involving non-
convex and possibly nondifferentiable energy functions, the superpo-
tentials. First superpotentials connected with the notion of generalized
gradient are considered, then the V-superpotentials are defined and fina~
ly the quasi differential. After the defintion and formulation of several
classes of mechanical problems involving superpotentials we study the
questions of existence and approximation of the solution of a variation~
hemivariational inequality resulting in the delamination theory of von
Karman laminated plates. Then a general semicoercive hemivariational in-
equality is studied and necessary and sufficient conditions are derived.
The last section is devoted to the new notion of quasidifferentiability
and its application to the study of mechanical problems.
84 P.D. Panagiotopoulos

1. INTRODUCTION
The present Chapter deals mainly with the study of mechanical
problems expressed in terms of nonconvex superpotentials and general-
ized gradients. These problems have as variational formulations hemi-
variational inequalities, expressions of the principle of virtual work
or power in inequality form. We mention here, for instance, problems
involving nonmonotone, possibly multivalued laws between stresses and
strains or reactions and displacements, or generally speaking, between
generalized forces and velocities. A further aim of the present Chapter
is the discussion of some other types of variational formulations which
result from V-superpotentials and from quasidifferentials. All the a-
forementioned problems are expressed in terms of nonconvex energy
functions and their treatment differs inherently from that of problems
having as variational formulations variational inequalities.
The study of variational inequalities started in 1963 with the works of
G. Fichera [1],[2] and of J.L. Lions with G. Stampacchia [3]. An im-
portant achievement of J.J. Moreau [4] in 1968 was the introduction of
the notion of (convex) superpotential and his connection with the theory
of variational inequalities. By means of Moreau's superpotential large
classes of complicated, yet unsolved problems in mechanics and engi-
neering, could now be formulated and solved correctly, which could not
be achieved with the methods of classical bilateral mechanics. Because
the variational inequalities are expressions of the principle of virtu-
al work or power in inequality form all these problems are also called
inequality problems.
Until 1981 all the inequality problems studied were expressed in terms
of variational inequalities and included convex energy functions (super-
potentials). But the convexity of an energy function implies the mono-
tonicity, e.g. of the underlying stress- strain relation. Thus, only
monotone, possibly multivalued conditions between generalized forces
and velocities could be considered. In order to overcome the constraint
of monotonicity the notion of nonconvex superpotential was introduced
by P.D. Panagiotopoulos [5],[6] by using the generalized gradient of
F.H. Clarke. Thus a new type of variational expressions in inequality
form was obtained, the hemivariational inequalities which are no longer
connected with monotonicity. With respect to static hemivariational in-
equalities we have substationarity "principles" for the potential and
the complementary energy and nonuniqueness of the solution, instead of
the minimum "principles" and the uniqueness of the solution as happens
in the case of convex superpotentials. For the formulation and treat-
ment of variational inequality problems we refer the reader to [2],[7],
[8],[9],[10],[11],[12] and for the hemivariational inequalities to [6],
[12] both for the mechanical and the mathematical aspects of the theory.
In this Chapter several types of hemivariational inequalities are
studied. The mechanical problems chosen are representative and form
the appropriate substratum upon which the mathematical theory is de-
Hemivariational Inequalities 85

veloped. In Sec. 2 we give some definitions and notations for the


reader's convenience and we derive formally for certain mechanical
problems some hemivariational and variational-hemivariational inequali-
ties. Also the numerical treatment of the arising problems is dis-
cussed, together with the V-superpotentials and the basic questions of
stability and unloading. Sec. 3 concerns the mathematical study of
laminated von Karman plates subjected to a nonmonotone interlaminar
law and to monotone boundary conditions, i.e. we deal with variational-
hemivariational inequalities. In Sec. 4 a general semicoercive hemi-
variational inequality is studied and necessary and sufficient con-
ditions for the solution of the problem have been derived. Finally in
Sec. 5 the notion of quasidifferentiability is introduced into mechanics
and certain applications are given. It is worth noting that all the
mathematical methods developed here for the study of the coercive and
semicoercive hemivariational inequalities which arise in concrete me-
chanical problems can be extended without many modifications to a more
abstract mathematical setting.
Closing this introduction we would like to point out once again the in-
herent differences between "Smooth Mechanics", based on the notion of
classical potentials, and "Nonsmooth Mechanics" which is concerned with
the study of the nonsmooth and/or nonfinite, convex or nonconvex super-
potentials and the related mechanical problems.

2. MATHEr1ATICAL PRELif~INARIES. FORMULATION OF MECHANICAL PROBLEMS


2.1 Mathematical Notions and Definitions. A Survey
Let X be a locally convex Hausdorff topological vector space, X' its
dual space and <X' ,x> the duality pairing. The reader who is not fa-
miliar with this terminology could perceive X as a classical Hilbert
space (then X=X' and <X',x> is the scalar product of the Hilbert space),
or more simply as the classical n-dimensional Euclidean space ffin (then
<x',X>= xixi i=1,2, ... ,n). A functional f:X--+(-=,+=]with fj!loo is called
proper. Moreover f is lower semicontinuous (1 .s.c) on X if and only if
the set epif= {(x,A.)If(x):A., A.ElR} is closed in XxlR. We also recall
that f is Lipschitzian (locally) at x if a neighborhood U of x exists
on which f is finite and
( 2. 1)

where c is a positive constant depending on U and p is a continuous


seminorm on X. If (2.1) holds at every xEACX then f is called (locally)
Lipschitzian on A. For instance f is Lipschitzian at x, if it is con-
86 P.D. Panagiotopoulos

tinuously differentiable at x, or convex (resp. concave) and finite at


x or a linear combination of Lipschitzian functions at x. We denote by
f 0 (x,y) the directional differential in the sense of F.H. Clarke at x
in the direction y which is defined for f Lipschitzian at x by the ex-
pression (see e.g. [13])
fo(x,y)= limsup f(x+h+Ayl-f(x+h) . (2.2)
A-+0+
h-+0
Then the generalized gradient off at x for f(x) finite is by definition
af(x)= {x'EX'If0 (x,x 1-x)<;: <x',x 1-x> vx 1EX} . (2.3)
It should be noted that if f is convex then af(x) coincides with the
subdifferential af(x) defined for f(x) finite by the expression
af(x)= {x'EX'If(x 1)-f(x)<;: <x',x 1-x> vx 1EX} . (2.4)
Iff is continuously differentiable at X then af(x)={gradf(x)}.
Note that in the mathematical literature the term subdifferential and
the symbol af are used both for the subdifferential (2.4) of convex a-
nalysis and the generalized gradient (2.3) for which here the notation
af is used. The generalized gradient can be defined for any functional
f:X-+[~,~]. In this case f 0 (x,y) should be replaced in (2.3) by the
more complicated notion of the upper subdifferential of R.T. Rockafel-
lar [14] ft(x,y). Then we have that
af(x)= {x'EX'il(x,x 1-x);;; <x',x 1-x> vx 1EX}. (2.5)
Moreover af(x)= ~ if ft(x,O)= ~.otherwise af(x)~ ~. Note that af(x)
is never empty, iff is Lipschitzian at x, or iff attains a local mini-
mum at x.
For further properties and other equivalent definitions of the general-
ized gradient we refer the reader, for a short presentation of the re-
sults to [12](Ch.4) and for further details and proofs to [13],[14].
We define here anly two other notions. The notion of substationarity
and the notion of a-regularity. A point x0 is called a substationarity
point off, if it is a solution of the multivalued equation
Hemivariational Inequalities 87

O{af(x) . (2.6)
Every local minimum and every saddle point is a substationarity point.
Also a local maximum x0 is a substationarity point iff is Lipschitzian
around x0 The substationarity has a more profound equivalent defi-
nition in terms of the directions of approximately uniform descent; see
in this respect [15] p.102, and [12].
A functional f:X~[~,~l is called a-regular if

ft (x,y) = t' (x,y) = 1 im f(x+A.yL-f(x) VyEX . (2. 7)


A.~O+

Here t'(x,y) is the one-sided directional Gateaux-differential off at x


in the direction y defined by the last equality in (2.7). For instance,
iff is a convex, or a maximum type function, then f is a-regular. The
property of a-regularity is important since it guarantees the additivity
of the generalized gradient.

2.2 Nonconvex Superpotentials. Generalities


Let~ be a mechanical system characterized by the generalized force

vector fEF and the generalized velocity vector uEU. Here F and U are
vector spaces placed in separating duality through the bilinear form
<U,f> expressing the work produced by f due to u. Let~:~[~,~] and
suppose that between f and u a relation of the form
-fEa<I>(u) in ~ (2.8)
holds. By definition (2.8) is equivalent to the relation
<I>t(u,v-u);;; <f,v-u> vvEU (2.9)
which is called a hemivariational inequality.
i) Suppose now that <I> is a maximum type function, i.e .. ~(u)=mqx{~.(u)},
1 1
i=1, ... ,m, u={u 1, ... ,~}and ~i() are smooth functions. Let us
introduce the sets Ai={uj~(u)=~i(u)}. Then (cf. also [15])
a<I>(u)= {gra~i(u)} if uEAi (2.10)
a<I>(u)= co{grad~.(u),grad~.(u)} if uEA.nA. (2.11)
1 J 1 J
a<I>(u) = co{ grad~.1 (u) ,grad~.(u) ,grad~k(u)} if uE:A.nA .nAk. (2.12)
J 1 J
88 P.O. Panagiotopoulos

Here "co" denotes the convex hull and thus in (2.11) (resp. (2.12))
a~(u) is the line segment (resp. the triangle) connecting the ends
of the two (resp. the three) gradients.
ii) Suppose that~ is the indicator Ic of a closed set CCU=ffin, i.e.
~(u)=Ic(u)={O if uEC, oo if u~C}, and that u={u 1 , ... ,un}. Then
3~(u)=Nc(u), where NC(u) d2notes the nor~al cone to Cat the
point u.
iii) If more specifically C={uEffin ~i(u)SO i=1, ... ,m} where the
functions ~i are smoothmthen it can be shown (cf.[12] p. 146) that
NC(u)= {fEF=ffin f= EA.. grad~.(u), A..;:;o,~.sO,A..~.=O} (2.13)
i=1
1

1 1 1 1 1 1

on the assumption that if uE boundary C a vector y exists such that


<y, grad~.(u)> < 0 ( 2. 13a)
1

for every i characterizing an active constraint (~.(u)=O)


1
at u.
Roughly speaking property (2.13) implies the validity of Lagrange
multiplier type formulas for the nonconvex problems.
We may make use of (2. 13) in the study of the equilibrium problem of a
material point which is constrained to remain in c. Then we can easily
verify that if R is the reaction force, then a constraint of the form
-RENC(u) holds and thus, if F0 is the external force acting on the ma-
terial point, we shall have
F= -RENC(u)= aiC(u), ( 2. 14)

where u denotes the displacement, or equivalently


I~(u,v-u);:; <F,v-u> WEU ( 2. 15)
t
But Ic(u,v)=IT (u)(v) [14], where Tc(u) denotes the tangent cone to C at
the point u ana thus (2. 15) is equivalent to
( 2. 16)

If the variations v-u are sufficiently small, as is assumed in the


classical formulation of the principle of virtual work and if at u the
admissible set C is tangentially regular (e.g., if C has not a reentrant
Hemivariational Inequalities 89

corner at u, cf. also [15] p.41.) we obtain the same inequality as in


(2.16) holding for every vEC, i.e. we find the inequality of the classi-
cal principle of virtual work [16]. However, (2.16) is the exact and
general formulation of this principle for the problem under consider-
ation.

2.3 Nonmonotone Unilateral Contact and Friction Conditions


We consider a deformable body Q and let r be its boundary. Q is re-
ferred to a fixed orthogonal Cartesian system ox 1x2x3 r is decomposed
into three nonoverlaping parts rF' ru and r 5. On rF the boundary forces
S={Si} are given, i.e.
S.=
1
F.,
1
F.=
1
F.(x)
1
, (2.17)
and on ru the displacements are prescribed
u.=
1
u.,
1
U.=
1
u.(x)
1
, (2.18)
where S.=o .. n. (i,j=1,2,3- summation convention), o={o 1.J.} is the stress
1 1J J
tensor and n={ni} is the outward unit normal vector tor. We assume
that on r 5 S (and u) are decomposed into normal and tangential com-
ponents SN and ST' and uN and uT respectively. The assumption of given
tangential displacements or forces on r 5 say
(2.19)

can be combined with a nonlinear relation between SN and uN. Thus we


may write on r 5 a relation of the form (Fig. 1a)
if UN< 0 then SN = 0 ( 220 )
if UN ;;: 0 then SN+k (UN) = 0 .'
where k() is a nonmonotone function. (2.20) describes the unilateral
contact with a granular support (rock, concrete, soil). Moreover the
graph of k may include some jumps which describe locking and local
crushing effects. If the contact surface exhibits a small resistance
in tension, e.g. due to an adhesive material, we can write the condition
(Fig. 1b)
90 P.D. Panagiotopoulos

if UN> UN then SN+k(uN) = 0


0
if UN= UN then 0;; SN ;; k( uN (2.21)
0 0
if UN< UN then sN= 0
0

The dotted line in Fig. 1b is more realistic, since we avoid the ideally
brittle behaviour. All the aforementioned laws can be put in the form
-SN{ajN(uN) , (2.22)
where jN() results by integration. For instance, in the case of Fig. 1a

UN+
jN(uN)= f k(0d1:, (2.23)
0

where uN+ = sup(O,uN).


In the tangential direction the previous normal laws may be combined
with a general tangential nonmonotone relation of the form
(2.24)
in order to describe frictional phenomena. For SN= CN' given, we depict
in Fig. 1c the friction law of Coulomb. In Fig. 1d to Fig. 1f some
other nonmonotone, possibly multivalued, friction laws are depicted.
All of them hold for QEm 2 ; in the case OEm 3 the additional assumption
that the vectors -STand uT are collinear has to be made. The law of
Coulomb can be written in the form (2.24) with jT(uT)= 11!CN !!uri; 11 is
the coefficient of friction. Due to the convexity of the superpotential
a can be replaced by the subdifferential a. The laws of Figs. 1d- 1f
are realistic generalizations of Coulomb's law. The graphs of (-ST,uT)
diagrams may include vertical jumps describing local cracking and crush-
ing of the contact surface asperities. Worthy of note is the existing
analogy between Scanlon's diagram for reinforced concrete in tension
(see e.g. [17] and Fig. 1g) and the sawtooth frictiondiagrams describing
the modification of the mechanical properties of the contact surface
during the frictional contact.
In Fig. 1h the friction diagram between reinforcement and concrete is
Hemivariational Inequalities 91

depicted. Here [uT] denotes the relative tangential displacement be-


tween the two structural elements in contact. All the above diagrams
and their twodimensional extensions can be put in the form (2.24) or if
SN is not prescribed in the form
(2.25)
The nonmonotone unilateral contact boundary condition coupled with non-
monotone friction has the form
-sNEajN <uN) and if SN = 0 then sT= o
(2.26)
if SN< 0 then -STEajT(uT;ST).

Relation (2.24) includes [18] as a special case the anisotropic friction.


Let us give further a method for the derivation of general nonmonotone
friction laws, which are not necessarily onedimensional. or which do not
assume the collinearity of-STand uT. Let K(SN) be a closed nonconvex
subset of the ST-space which may depend on SN. For the present we as-
sume no relation between SN and ST. Then we consider the law (gener-
alized gradient with respect to ST)
-uTEaiK(ST;SN)= NK(ST;SN) ( 2. 27)
which implies (cf. (2.14- 2.16)) that for STEK(SN)
(2.28)

This last inequality recalls Hill's "principle" in holonomic elasto-


plasticity; it coincides with this principle if K(SN) is tangentially
regular at STand the variation ST-ST is appropriately small. The tan-
gential regularity is guaranteed, e.g., if K(SN) does not have a reen-
trant corner & ST. Note also that (2.27) implies that <uT,ST> is sub-
stationary with respect to K(SN) at ST.
Suppose now that K(SN)= {STifi(ST;SN)~ 0 i=1, .. ,m} where fi are con-
tinuously differentiable functions of ST defining the boundary between
sliding friction and adhesive friction. Moreover, let a condition simi-
lar to (2.13a) be valid. Thus we may write the nonmonotone friction law
in the form
92 P.D. Panagiotopoulos

-SN -sN

SN>~

UN UN
A k(UNo)

(a) (b)

-ST -sr

ur ur

(c) (d)

-Sr -Sr

'
'
ur ur

(e) (f)

a -sr

(g) (h)

Fig. 1, Nonconvex Superpotential Laws


Hemivariational Inequalities 93

m af.(ST;SN)
-uT= i: 1A; 1 asT , A;~ 0, fi(ST;SN)~ 0, "-/i(ST;SN)= 0. (2.29)

With respect to the general friction law (2.25) we can adapt the idea
of nonlocal effects caused by the asperities of the contact surface [19].
In this case SN should be replaced by Q(SN), where Q is an appropriate
smoothing operator.
The previous static friction laws are of importance for static and
certain quasistatic problems. Due to the similarity of this theory to
the deformation theory of plasticity (holonomic theory of plasticity),
the existing criticism and comparisons between this theory and the flow
theory of plasticity applies also to all the static friction laws. For
dynamic problems we have to consider analogous relations between ST and
vT, where vT is the tangential velocity. In this case we may write the
same laws as before by replacing uT by vT. We can explain the noncon-
vexity of the boundary between the sliding and adhesive friction by
means of analogous arguments to those of Green and Naghdi [20] in plas-
ticity; thus a star-shaped friction boundary results. Moreover we can
justify a nonconvex superpotential law of the form
-vTEaW(ST;SN) (2.30)
if we assume that the elastic properties of the asperities change during
the tangential deformation. The justification we shall present is a-
nalogous to the one given in nonconvex plasticity [12]. Let ST be the
actual tangential stress vector assumed to be on the friction surface
(Fig.2a) and let Sf be on or inside the friction surface; we denote by
K the admissible set for the ST's. An external action changes the Sf
into ST and let an additional action produce an additional displacement
increment -dur Let dST be the tangential stress vector increment which
does not need to be tangential to the friction surface, if the latter
changes with frictional deformation (analogously to the hardening ef-
fects in plasticity). Further the external action releases dST and the
tangential vector returns to the initial value Sf. We assume that for
the stress path ABr6EA the work produced consists of one conservative
94 P.D. Panagiotopoulos

{a) {b)

Fig. 2, Nonconvex Friction Laws. Interfaces

part, which does not depend on the path, and a nonconservative part.
The first results from the elastic deformation of the asperities, where-
as the second from irreversible local fractures causing a change of the
elastic properties of the asperities. Let us assume that the nonrecove~
able work w is a function of ST,ST' of the path followed and of SN. Let
also w= 0 if ST= ST' We assume that the work produced over ABI't.EA is
positive, i.e.
(2.31)
Setting ST= ST, implies -<dST,dur>> 0 and by making this expression as
small as possible we get the inequality
-<ST-s:y,dur>+ w(ST,ST,path,SN) ~ 0 VSTEK . (2.32)
Now we assume the existence of a function W:R 3~[~.~J such that
Hemivariational Inequalities 95

(2.33)

and thus (2.32) implies that


(2.34)
or, if one replaces duT by the velocity vT, (2.30) is obtained.
r~any dissipation mechanisms giving rise to nonmonotone stress-strain

relations can be explained by the generalized (nonconvex) hypothesis of


normal dissipation after an appropriate choice of hidden variables.
This hypothesis, first developed in [12](p.157), is a direct generali-
zation of the hypothesis of normal dissipation of B. Halphen and
Nguyen Q. Son [21], see also [22], based on convexity and subdifferenti-
ability. By means of this new hypothesis we can extend the theory of
Rice of the viscoplastic potential, the convex theory of plasticity and
viscoplasticity to the case of lack of convexity. For nonconvex locking
materials we refer to [6].

2.4 Interface Laws, Adhesive Joints and Composite Materials


Let us assume that in a body n there is a surface A along which
debonding and slip of the material is possible. The surface A may be
a rock joint, a fault or an interface between different structural com-
ponents, e.g. between the adjacent laminae in laminated structures, or
between a fiber and the matrix i~ a fibre reinforced structure or be~
tween reinforcement and concrete etc. Much research work has been done
on the simulation of the behaviour of an interface and on the influence
of the interface on the equilibrium of the whole structure. Here we
propose a model based on nonconvex superpotentials. We assume that the
interface can be simulated by joint elements having two strain com-
ponents: the normal strain EN and the shearing strain ET (Fig. 2b).
Let oN and oT be the corresponding stresses. Then we may assume between
oN and EN' and betw~en oT and ET the same laws as those of Fig. 1, where,
now, -SN and -ST are replaced by oN and oT respectively (note the sign
difference between boundary tractions and (internal) stresses). This
96 P.D. Panagiotopoulos

treatment assumes A as a part of a given body Q with different material


behaviour. There is also another approach which is in use especially
for laminated structures. We consider that each side of the interface
belongs to a different body and then we assume on the interface contact
and friction laws between -SN (resp. -ST) and the relative normal (resp.
tangential) displacement of the two bodies [uN](resp. [uT]) expressed
in terms of nonconvex superpotentials as in (2.22) and in (2.25). Of
course uncoupling of the normal and the tangential action might be
questionned; more general superpotential laws in the form

(2.35)

could be considered. At this point we would like to note that diagrams


of the form depicted in Fig. 1b simulate in the normal and in the tan-
gential direction the delamination effect, as well as the interlaminar
slip effect in laminated structures with adhesive material between the
laminae and lead to the development of a technical theory for these
structures [23].

If we want to study the behaviour of a structure under a given loading


we may cons.ider stress-strain diagrams of each element defined in the
whole length of the strain axis (Fig. 3a) [24] ,[25]. Even, if only the
final values E 0 and E~ are known, we may render the respective law com-
plete by assuming a brittle (vertical lines) or a semibrittle behaviour
beyond E0 and E~. Accordingly, a complete law [25] has as a main ad-
vantage the consideration of several nonlinearities like cracking, crus~
ing, delamination, slip etc. using a static or a quasi-static method.
This approach may be applied to composite material structures, because
it provides an estimation of the behaviour of the structure through
purely static methods, as is the case in several engineering theories,
e.g., the calculation of reinforced concrete structures etc. But this
is possible only using superpotentials, since the sawtooth form is pre-
dominant in the stress-strain curves obtained experimentally in com-
posite materials. The diagram of Fig. 3b describes the progressive,
Hemivariational Inequalities 97

almost ideally brittle failure of the fibres in a fibre reinforced ma-


terial [26]. In the peaks of the diagram fibre failure happens; the
jumps (which are complete) correspond to the load transfer to the rest
of the fibres. The upward parts are due to the strength recovery after
the redistribution of the stresses between fibres and matrix. Slow
pull-off tests of reinforced specimens of laminated structures give the
diagram of Fig. 3c [27]. For other diagrams of this kind and their in-
terpretation the reader is referred to [28],(29] ,[30].
It is obvious that all the aforementioned onedimensional phenomenologi-
cal laws can be written through a nonconvex superpotential in the form

a a

(a) (b)

p 0

(c) (d)

Fig. 3, Complete Diagrams. Behaviour of Fibre Reinforced


Materials (Phenomenological Diagrams)
98 P.D. Panagiotopoulos

oEaw(E), where w denotes the area between the graph and the horizontal
axis on which the strains are measured.
Generalizations of the above onedimensional examples for threedimension-
al continua can be constructed in analogy to the ideally plastic and
locking materials. Let us e.g. construct a threedimensional analogon to
the onedimensional stress-strain diagram of Fig. 3d. One method would
be to consider four elastic energy structures which intersect each other
and form reentrant corners causing the stress jumps. Moreover the last
energy surface should guarantee that o=O after the last jump (cf.Fig.4a).

/
/

(a) (b)

/
/
,......a=O

0 cl

o,
a
a'1 c3

El E E'l E

(c) (d)

Fig. 4, Construction of Superpotentials for


Composite Materials for 3D-Bodies
P.D. Panagiotopoulos 99

However, a more general definition of the superpotential of a composite


material is the following: Let us define in the space m 6 of the sym-
metric strain tensors E={Eij} certain surfaces Fi(E)=O i=1,2, ,m.
l~e assume that the Fi 's are continuously differentiable. Moreover

K 1 ={EiF 1 (E)~O} is a closed set. If EEintK 1={EiF 1(E)<O} then we have a


linear elastic stress-strain behaviour governed by the elasticity tensor
c 1 ={C~jhk}. If E satisfies F1(E)=O then we have a brittle fracture
(vertical segment in Fig. 3d) and we assume that stress
1
cr= {crij}= {CijhkEhk}+A. 1gradF 1(E) A. 1E[-a 1,0] (2.36)
and a 1 ~o satisfies the relation that
1 2
{CijhkEhk}-a 1gradF 1(E)= {CijhkEhk} , (2.37)
where c 2 ={C~jhk} is the elasticity tensor of the subsequent linear e-
lastic region. This region lies within the admissible strain set
2
K 2 ={EiF 2 (E)~O}. For EE{EiF 1(E)>O F2(E)<O} we have that crij=CijhkEhk
and for E lying on the boundary of K2 we can again assume a law of the
form (2.36). We continue in this way until the last brittle fracture
surface, after which the stress becomes zero, i.e. the right-hand-side
of (2.37) becomes zero (Fig. 4b). The aforementioned procedure can be
modified to include the case of semibrittle fractures (Fig. 4c). We
note that the segment AB in Fig. 4c is equal to (E-PK 1 E)(tg~ 1 +tg~ 2 ),
where PK (E) denotes the projection of E onto K 1 ={EIE~E 1 }; in our case
PK (E)=E~. Therefore we may generalize the law of Fig. 4c in three di-
me~sions by writing that for F 1 (E)~O we have
1
cr= {cr lJ
.. }= {C.lJ"hkEhk}-1J1{E lJ
.. -(PK 1E) lJ
.. } 11 1>0 . (2.38)

Moreover for F1(E)<O we have a linear elastic behaviour governed by the


Hooke's tensor c1 Relation (2.38) holds for all {Eij} which satisfy
the inequality

{C~jhkEhk}-IJ{~ij-(PK1E)ij}~{C~jhkEhk} . (2.39)
1 .I
Let us denote their set by S . Let {e 1 .. } be the value for which (2.39)
lJ
is satisfied as equality. This procedure continues until the final
100 P.D. Panagiotopoulos

surface a=O is reached. Here we may assume convexity and closedness of


K1, K2 etc. in order to guarantee the uniqueness of the projection.

Note that in case of semibrittle fracture we may characterize the


fracture by means of closed (and for simplicity convex) sets defined in
the stress space by the inequalities r 1 (a)~O F 2 (a)~O etc. Indeed we
have from Fig. 4c that E=acotg 1+(PKa-a)(cotg 2+cotg 1) where PKa is
now the projection of a onto the set ~a 1 . The generalization is obvi-
ous (cf. also [12] p.97) and reads
1 (2.40)
if r,(a)< 0 then Eij = cijhkahk
if r,(a);:;; 0 then E. . = c ~ . hkah k-il [a . - ( P-K a) . . ] ~>0. (2.41)
lJ lJ "'1 lJ 1 lJ
Here c 1 is the inverse tensor of c1, K1 is the set {aiF 1 (a)~O}. Re-
lation (2.41) holds until the value {a 1ij} is attained satisfying the
relation
2 1 1 (P (2 42)
a1 hk 11 1 a1 ij- K1a1 ij
1
cijhka1 hk cijhk -- ( I I) )
=

This scheme continues until the final surface a=O is attained. It is


also worth noting that the linear elasticity laws can be replaced by
nonlinear elasticity laws of the form a=~: , or more generally by the
monotone law oEaw(E), where w is a convex, l.s.c and proper functional
and a denotes the subdifferential (Fig. 4d). All the aforementioned
threedimensional generalizations of the sawtooth onedimensional laws
for composite materials, due to their formulation, can be put in the
general forms
oEaw(E), or EEa'W(a) , (2.43)
where w or ware nonconvex superpotentials and a denotes the generalized
gradient. The form of w is given in the case of (2.38) by the ex-
pression

W(E)
Hemivariational Inequalities 101

In the expressions, which we have formulated, Fi and Fi are very general.


We can, e.g., assume for them analogous forms to the ones for ideally
locking and ideally plastic materials. In the case of brittle fractures
we may assume that the mutual positions of the hyperplanes, defined by
the elasticity tensors Ci' and of the gradients to the surfaces Fi(E)=O
always give bounded stresses for every strain; then the nonconvex super-
potential w is locally Lipschitz continuous. A deeper discussion would
impose further restrictions on the forms of Fi' F 1, c1 etc.; e.g. if one
wanted to assure the feasibility of the previous geometric constructions
(i.e. nonvoid intersections of hypersurfaces, finite ai in (2.37) etc.)
and that aw(E) (resp. aw(o)) iS nonempty and bounded for each E (resp.o).

All the above considerations hold for the case of loading. Obviously
the unloading along an elastic path could be included in the same model,
but then we have to work with stress and strain increments (cf. Sec.2.9).

2.5 Derivation of Hemivariational Inequalities and of Variational-Hemi-


variational Inequalities
In this Section we shall derive formally some hemivariational in-
equalities. We choose a mechanical problem, the problem of adhesive
contacts of elastic bodies and for a given external loading we ask for
the possible equilibrium positions. This is, for instance, the case of
laminated structures; we have the delamination problem between the
laminae, where a priori it is not known which parts between the laminae
remain in contact and which not. If the diagram of Fig. 1b holds with
uN replaced by [uN], we do not know a priori which parts of the adhesive
have (-SN,[uN]) on AB, on Ar (or Ar'), or on r~ (orr'~). Analogous is
the situation concerning the interlaminar slip, with the difference that
there is a coupling between delamination and slip which we shall discuss
in Sec. 2. 9.
Let Qm m=1,2, . ,1 be a set of deformable bodies possibly with different
elasticity properties, with the boundaries rm m=1,2, ,1 assumed to be
appropriately regular. Let x={xi} i=1,2,3 be a point of R3 and let
o(m)=o~~) and E(m)=E~~) i,j=1,2,3 be the stress and strain tensors of
lJ lJ
102 P.D. Panagiotopoulos

(m) (m) (m) (m)}


the m-body. He denote by f ={fi } and u ={u; the volume force
and the displacement vector in each body. If n(m)={n~m)} is the out-
ward unit normal vector to r(m), the boundary force o~ r(m) is s~m)=
=o~~)n~~)(summation convention). Let S~m) and Sfm) be the norma~ and
tangential components of it respectively. The corresponding displace-
ment components are u~m) and u~m). The boundary r(m) is divided into
three non-overlaping parts r~), r~m) and r~m) On r~m) the displace-
ments are given, i.e.,
u~m) = u~m) on ru(m) (2.45)
1 1

on r~m) the forces are prescribed, i.e.,


s~m)=F~m) on rF(m) (2.46)
1 1

and on r~m) - which corresponds to the interface of structure m with


other substructures - nonmonotone interface conditions hold describing
slip and delamination effects. We write in the general case Q(m)c~ 3
the interface conditions in the form
-s(m){aj (S(m) [u(m)J) (2.47)
N N(m) ' N
-sim)EajT(m)(S(m);[uim)]) (2.48)

in the normal and in the tangential direction to the interface. The


superpotentials jN and jT are assumed to be functions of the interlayer
gap [uN] and slip [uT] (locally Lipschitz continuous) respectivelyand of
the interface stresses S (e.g. in the case of Coulomb's friction), which
is also a function of u. Here, however, we assume that (2.47) and
(2.48) are uncoupled, i.e. that S(m) is considered as having a given
value, or that jN(m) and jT(m) do not depend on S(m). Then (2.47,48)
are equivalent to the inequalities
j0 ([u(m)] v-[u(m)])> -s(m)(v-[u(m)]) 'v'VER (2.49)
N(m) N ' N ~ N N

.o ( [uT(m) ] .v-[uT(m) ] ) > (m)( (m)]) i= 12. 3 . (2.50)


JT(m) ~-ST. vi-[UT
1

In the framework of small strains and linear elastic behaviour for


Hemivariational Inequalities 103


o.. (m) m= 1 , 2 , ... , 1 we can wr1 t e th ere 1a t 1ons
a~~). + f~m) = 0 , (2.51)
1J 'J 1

(2.52)

(2.53)

The comma denotes the differentiation and Hooke's tensor C(m)={C~J~k}


satisfies the well-known symmetrya11d ellipticity conditions. We write
the principle of virtual work for every body Q(m) in the form

'v'VEU~~), ( 2. 54)
where U~~) is the kinematically admissible set of Q(m), i.e.

u(md) = {v(m)lv(m) = v~m) ,v~m)EU(Q(m)),v~m)=U~m) on ru(m)} . (2.55)


a 1 1 1 1

Here U(Q(m)) denotes a space of functions defined on n(m). Adding with


respect tom all the expressions (2.54) and taking into account the in-
terconnection of the bodies yields a relation of the form

(2.56)

Q,
where u d= u u(m)
a m=O ad
104 P.D. Panagiotopoulos

In (2.56) we introduce the integrals along the joints r , q=1, ... ,k.
The new enumeration of the r~m)_boundaries has the adva~tage that final-
ly the energy of each joint appears. Care should be taken of the fact
that the variation of the energy of each constraint of the form (2.47)
and (2.48) must appear only once in the last terms of (2.56). Further
we introduce the elastic energy of the m-structure

a(u
(m)
,v
(m) f (m) ( (m)) 8 ( (m))
) = Q(m)cijhk 8 ij u hk v (2.57)
dQ

and by taking into account (2.49),(2.50) and (2.57) we obtain from


(2.56) the following hemivariational inequality:
Find uEUad such as to satisfy
2 k
E a(u(m) ,v(m)_u(m))+ E U [j 0 ([u(q)],[v(q)]- [u(q)]) +
m=1 q=1 r(q) N(q) N N N
2
+ j 0 ([u(q)] [v(q)]- [u(q)])]dr~ E U f~m)(v~m)_u~m))dQ +
T( q ) T ' T T - m= 1 Q( m) 1 1 1

+ f F~m)(v~m)_u~m))dr] (2.58)
r(m) 1 1 1
F

This hemivariational inequality is the expression of the principle of


virtual work in its inequality form for the structure under consider-
ation.
To check in which sense a solution of (2.58) satisfies (2.51), the
boundary conditions on r~m) m=1, .. ,1 and the interface relations
(2.47),(2.48) we must make the functional setting of the problem more
precise. So we assume that f~m)EL 2 (Q(m)), F(m)EL 2 (r~m)), c~j~kEL00 (Q(m)),
u~m) ,v~m)EH 1 (Q(m)) (classical Sobolev space).
1 1
Then u~m), ui~)EH 1 1 2 (r(m))and s~m), si~)EH- 1 1 2 (r(m)). We set in
(2.58) v~m)_u~m) = ~~m) where ~~m) belongs to the space r& infinitely
differen~iabl~ funct~ons with c~mpact support in Q(m), ~(Q(m)). Then
from (2 58) by setting v~m)
1
-u~m)
1
= ~~m)
1
for m=n and v~m)
1
-u~m)
1
= 0 for m,Fn
we obtain
Hemivariational Inequalities 105

(2.59)

since <P~n)=O on r(n). (2.59) implies that (2.51) holds on Q{n) in the
sence o~ distributions over Q(n). This procedure is repeated for
n=1,2, ... 1. Now applying the Green-Gauss theorem to each body we
obtain the equality

(2.60)

More correctly in (2.60) we should have instead of f(m) and f(m) the
duality pairing<.,.> between H112(r) and H- 1/ 2{r).rF rs
From (2.60) and (2.58) we derive the inequality

(2.61)

If in (2.61) we consider that 6n rF(m) ,v\m)_u~m)= r\m)EH 112(r(m)) for


1 1 1

m=n. and that v~m)_u;m)= 0 for mfn on r~m) and on rq for every q, we
obtain s;n)= F~n) as an equality in H- 112(r(n)); this can be shown for
every n. From (2.61) by setting [v~q)]-[u~q)] = r~q) on rq for q=n and
the same difference =0 for qfn. and setting [viq)]-[u+q)] = 0 on rq for
every q we obtain

(2.62)
106 P.D. Panagiotopoulos

which constitutes a "weak" formulation of (2.47) on H- 1/ 2(r)xH 112(r).


~nalogously we obtain from (2.61) a weak form of (2.48).
It is noteworthy that the interface superpotential does not need to be a
Lipschitz continuous function. Then Clarke's directional differentials
j~(.,.) and j~(.,.) must be replaced in the previous hemivariational in-
equalities by j~(.,.) and ji(.,.). Analogously we proceed in dynamic
problems on the assumption of small displacements. Then f~m) has to be
a2u~m) ,
replaced by f~m)- p(m) at~ , where p(m) is the density of the m-body,
and initial conditions for the displacements u~m) and the velocities
au~m);at have to be considered. The resulting 1 hemivariational inequali-
ty is analogous to (2.58) and expresses the d'Alembert's principle in
inequality form. In the dynamic case the holonomic interface relations
(2.47) and (2.48) may be replaced by the relations
au(m)
-s~m){ajN(m) (S(m); [--J---1), (2.63)

au(m)
-s(m)Eaj (S(m) [~Ttl). (2.64)
T T(m) ' a
In this case we write again (2.54) but by considering instead of dis-
placement variations, velocity variations. Thus we are led to a hemi-
variational inequality as the (2.58) but now having instead of v(())
-u(m) and [v(m)]-[u(m)] the variations v(m)_ au(m) and [v(m)]-[auNm]
N N at N at .
Suppose now that the substructures Q(m) m=1, .. ,1 obey a general non-
monotone law of the form o(m)EaW(m)(E), where W(m) is an extended real-
valued function nonconvex and noneverywhere differentiable (cf. e.g.
(2.43)). Then the variational expression of the problem is the same as
(2.58) but now the term m~ 1 a(u(m) ,v(m)_u(m)) has to be replaced by
j 1J(m)w{m)(E(u(m\E(v(m)_u(m)))dQ.
g
If the w(m) 's are convex super-
potentials, i.e. are convex, l.s.c and proper functionals, then we
define
Hemivariational Inequalities 107

(2.65)
otherwise
and we are led to a variational formulation analogous to (2.58); now
the elastic energy function a(.,.) is replaced by the difference

(E(v(m)))- Wm(E(u(m)))]. This variational form is called vari-
:E 1[vJ
m= m
ational-hemivariational inequality and is the expression of the princi-
ple of virtual work for the considered problem.
In all the above variational formulations we can assume that one or
more of the ni's are rigid. Thus we can incorporate into our model
boundary conditions of the form (2.22) and (2.24).

2.6 Substationarity of the Potential and of the Complementary Energy


Let us consider the following substationarity problem:
Find uEUad such that the potential energy of the structure
k
rr ( v) = _!_ :E a ( v( m) , v(m) ) + :E f[ j ( [ v(q ) ]) +j ( [ v( q ) ]) ] dr-
2 m= 1 q= 1r N{q) N T(q) T
q

(2.66)

is substationary at v=u, where vEUad" In other words uEUad is a so-


lution of the inclusion
OE 3rr(u) for VE Ugd . (2.67)
The following proposition holds.
Proposition 2.1: Suppose that the superpotentials ~~jN(m)(~) and
~~jT(m)(~) are locally Lipschitz and a-regular for m=1, ... ,1. Then
every solution of the substationarity problem is a solution of the hemi-
variational inequality (2.58) and conversely.
Proof: In the functional framework introduced in the previous section
we can write (2.67) in the equivalent form
108 P.D. Panagiotopoulos

(2.68)
where

(2.69)

and
rr(v)=rr 1(v)- (l,v). (2.70)
Now let us compute directly arr 1(u) by using the definition (2.2) of
'If(u,v-u). Note that A(u(m)) = 1!2a(u(m) ,u(m)) is a-regular and
A0 (u(m),v(m)) = a(u(m),v(m)) . (2.71)
Moreover let us denote by JN(q) the integral
JN(q)([u~q)]) = ~jN(q)([u~q)])dr, (2.72)

which may be either finite or infinite.


We shall show first that,
Jo ( [ ( q) ] [ (q ) ] ) - f .o ( [ (q) ] [ (q) ] )d ( 2. 73)
N(q) UN ' VN - JN(q) UN ' VN r
r
In this proof we omit N,(q) for the sake of simplicity.
Let us denote by gA,h the difference quotient
_ j(u+h+AV)- j(u+h)
gA,h ( u,v ) - A (2.74)

But ~j(E) is locally Lipschitz and thus


lgA,h(u,v)l;;;clvl, (2.75)
where c depends on the neighborhood of (u+h)(x). We note that u,vEL 2(r)
and that ~j(E) is continuous. Thus x~A,h(u(x),v(x)) is measurable.
Accordingly we apply Fatou's lemma - for not necessarily integrable
functions, see [31]p.152- and we have (changing the signs) that
flimsup(gA h- clvl)dr <:: limsupJ(gA h- clvl)dr. (2.76)
rA~o+ ' A~o+ r '
~0 ~0

Here c is a function of the neighborhood of (u+h)(x) but disappears


Hemivariationallnequalities 109

from both sides: Indeed we assume in (2.76) that ~0 taking values in


a denumberable dense subset of L2(r), since this space is separable.Let
us denote these functions by h.1 i=1, ... ,n ... and we have respective
constants ci depending on the neighborhood of (u+hi)(x). Defining
c>limsup c. leads to the existence of Msuch that n>M implies cn<c.
11-+oo 1
Thus we can put in (2.76) a c independent of h namely c= max{c,c 1 , ,
eM}. From (2.76) we obtain that
Jj 0 (u,v)dr ~ J0 (u,v) , (2.77)
r
where the integrals are either finite or infinite. We have also using
the definition of limsup,Fatou's lemma,the a-regularity of j,and (2.77),
that
J 0 (u,v)~ liminf J(u+A.vl- J(u)~ fliminf j(u+A.vl- j(u)dr=
>..~o+ r>..~o+

~ flim j(u+A.vl- j(u)dr= [J'(u,v)dr= Jj 0 (u,v)dr~ J0 (u,v).


r>..~o+ r r (2.78)
Accordingly we have shown (2.73). At this point it is worth noting
that if j satisfies an inequality of the form
lj(u)-j(v) I~ c(x) lu(x)-v(x) I (2.79)
with c(x)EL 2(r), then j(u())EL 1(r) and thus J is finite and Lipschitz-
ian. ( 1)
From (2.71),(2.72) and the same equality for J~(.,.) we obtain the
hemivariational inequality (2.58). Now we shall show the converse,
i.e. that any solution of (2.58) is a solution of the substationarity
problem (2.68). We show first that JN and JT are a-regular:
Indeed we have as in (2.78) due to Fatou's lemma( 2)

( 1) The finiteness of J may result from the properties of u oo r, e.g.


if uECO(r) as it happens for r regular in plate theory (HZ(g)cC 0 (o)
CC 0 (r)). Then x~j(u(x)) is Riemann integrable.
(2) For the proof of the first equality in (2.81) the dominated con-
vergence theorem is applied by some authors. The proof here seems to
be more general. Moreover in (2.76) is omitted in the literature the
proof that c is a function of a neighborhood of u and not of u+h.
110 P.D. Panagiotopoulos

J 0 (u,v);;: liminf J(u+A.vtJ(u)> fliminf j(u+A.vtj(u) =


A.-+0 + r A.-+0 +

= fJ'(u,v)dr= flimsup j(u+A.vl-j(u)> limsup .J(u+A.vl-J(u)>


r r A.-+0+ A.-+ o+

> .. f J(u+A.v)-J(u)
11m1n (2.80)
= A.
A.-+0 +

Thus
J(u,v)= fJ'(u,v)dr= fj 0 (u,v)dr= J 0 (u,v) ( 2. 81 )
r r
The a-regularity of the functions of rr 1 implies that (the proof follows
the same steps indicated in (2.78) and (2.80) but now for finite
sums ( 1) )

rr~(u,v-u) = rr' (u,v-u) ( 2.82)


and therefore (2.58) is equivalent, due to (2.82),(2.81) and the d-
regularity of A(u(m)), to
rr~(u,v-u);;: (l,v-u) VuEUad, (2.83)

which implies (2.68) q.e.d.


Note that if instead of (2.47) and (2.48) relations of the form
-[ (m)JEa* (S(m)) (2.84)
uN JN(m) N
-[u(m)]Ea/ (S(m)) (2.8S)
T T(m) T
hold, then (2.49) and (2.50) are replaced by the inequalities

( 2.86)

and
( 1) Instead of Fatou's lemma the inequality
limsup(f+g)(y):,; limsupf(y) + limsupg(y)
y-+x y-+x y-+x
is applied and the analogous one for liminf.
Hemivariational Inequalities 111

.*o ( (m) (m))> (m) ( (m)) ,- ._ (2.87)


JT(m) ST ,TT- ST ~-[uTi ] TTi- STi 'v'TTi':IR 1-1,2,3

For the problem defined by (2.45),(2.46),(2.51-53), and (2.84),(2.85) we


can formulate a hemivariational inequality by considering stress vari-
ations T\~)-a\~) and the statically admissible set
1J 1J

L; (md)=h(m) I T(m) = {T\~)}, T\~)EL;(Q(m)) ,T\~). + f\m) = 0 in n(m),


a 1J 1J 1J ,J 1
T\m)= F\m) on rF(m)} (2.88)
1 1

and by formulating the principle of complementary virtual work for the


m-body. We obtain for the whole structure through addition that

'v'TEl;ad. (2.89)
- ( m)
Here T.-- T.. n., TN and TT correspond to SN and ST, L: ad- m=O
u L: d , and
a
1 1J J
T= {c( 1 ) , ,T()}. Moreover let us introduce the elastic energy of
the m-structure expressed in terms of the stresses
(m) (m)) _ J (m) (m) (m)
A(a ,a - n(m)cijhkaij ahk dn , (2.90)

where c(m)={c~j~k} is the inverse tensor of C(m) From (2.86),(2.87),


(2.89) and (2.90) we obtain the following hemivariational inequality
which expresses the principle of complementary virtual work for the
whole structure: Find aEL;ad such as to satisfy
112 P.D. Panagiotopoulos

( 2. 91)

Let TI* be the complementary energy of the structure

Then the following substationarity problem is formulated:


Find oEEad such that
OEaiT* (T) for TEEad (2.93)

The relation between the solution of the hemivariational inequality


(2.91) and the solution of the substationarity problem (2.93) obeys a
proposition analogous to Prop. 2.1.
If each of the substructures Q(m) obeys a general nonmonotone law of
the form E(m)EaW~m)(o(m)) then the variational formulation of the
problem results from (2.91) by replacing the quadratic terms by
mt6(m)W~~)(o(m) ,T(m)_o(m))dn. If moreover, monotone laws o(m)Ea~m)(E(m))
hold, where the w(m) 's are convex, 1 .s.c and proper functionals, then
the conjugate functionals w(m) permit to write the above relation in
the form E(m)Eaw(m)(o(m)). Then we define

J( )we( ) (a)dQ if we() L1(n(m))


We (a) = { Qm m m (2.94)
(m) co otherwise
and we are led to a variational formulation analogous to (2.91) but
with the elastic energy replaced by m~ 1 [W(m)(T(m))-w(m)(o(m))], which
Hemivariational Inequalities 113

is a variational-hemivariational inequality.

2.7 Applications of the Nonconvex Superpotentials to Reinforced Concrete


and Cracked Structures
Until now we have given general examples showing the range of appli-
cability of nonconvex superpotentials. Here we shall focus our at-
tention on the reinforced concrete structures, where several nonmonotone
and sawtooth stress-strain relations can be experimentally verified.
i) It is known that in the behaviour of reinforced concrete structures
the phenomenon of concrete confinement plays an important role, es-
pecially in the case of large displacements and inelastic stress
states [32]. Various analytical models based on experiments have
been proposed to predict the concrete behaviour when it is confined
by rectangular ties. The confinement of the concrete is character-
ized by zones of tensile stresses in the direction of confining
forces. These zones appear both in the levels of ties and along the
longitudinal reinforcement. The cross-section inside this area is
the section of the effectively confined concrete (the shaded areas
in Fig. 5). The boundary ABC of the effectively confined concrete
is not a priori known and in the literature it is assumed to be a
parabola. Using the theory of hemivariational inequalities we can
determine this free boundary; the resulting free BVP has a great
similarity with analogous problems in hydraulics as, e.g. the de-
termination of the free surface in an earth dam, where the theory of
variational inequalities plays an important role [9]. In order to con-
struct an adequate model for the present problem it is necessary to
have a "complete" law for the structural element, i.e. a stress-
strain diagram which will be able to describe both the cracking and
the crushing of the concrete, the behaviour of the steel and the be-
haviour of the interface between steel and concrete with the arising
contact and sliding effects. Thus a diagram analogous to the one of
Fig. 3a can be adopted for the concrete, where the two jumps corre-
spond to the cracking and the crushing effects. The interface be-
114 P.D. Panagiotopoulos


-~

~
(a) (b)
A
(c)
c

(J

E E
(d) (e)

Fig. 5, On the Behaviour of a Reinforced Concrete Element

haviour can be described e.g. by the diagrams of Figs. 1b and 1e,


whereas for the steel an appropriate [33] inelastic stress-strain
diagram should be considered. From all the above it is apparent that
the global behaviour of a reinforced concrete structure is described
by a hemivariational inequality analogous to (2.58) or to (2.91),
with the quadratic energy terms replaced by terms involving upper
subderivatives (cf. Sees. 2.5, 2.6).
For the practical calculations concerning the concrete confinement
the experimentally verified nonmonotone law of Fig. 5d is adopted.
Obviously this law implies a nonmonotone moment-rotation diagram.
Let us denote further by rrR the potential energy of the reinforcement
and by rrR/C the interaction energy between reinforcement and concrete
(e.g. representing sliding, lack of adhesion etc). On the assumption
that rrR and rrR/C are convex the dynamic behaviour of the structure is
Hemivariational Inequalities 115

governed by the hemivariational-variational inequality

n o
i:lwi(Ei(u),Ei(v)-Ei(u))+TIR(v)-TIR(u)+TIR/C(v)-TIR/C(u) ~

;;: ( p-~1 ~~~- c ~~) T ( v-u) 'v'VEXd (2.95)

Here wi is the energy of the i-element of the structure i=1, . ,n,


M (resp. C) is the mass (the damping) matrix and Xd is the kinematic-
ally admissible set. If TIR and/or TIR/C are nonconvex then instead
of the differences TIR(v)-TIR(u) etc. the terms TI~(u,v-u) etc. appear.

i i) The theory of hemivariational inequalities combined with the "smeared-


fracture" concept [17] permits the calculation of structures at the
cracked stage and the determination of the regions where fracture
(cracking or crushing) has occured. The structure (Fig. 6a) is con-
sidered to be discretized through a finite element scheme. At an
integration point say P0 (Fig. 6b) the material of the proportionate
subdomain of the considered element is assumed to be either intact
or crushed or cracked with an assumed pattern of fracture (e.g.
equidistant parallel cracks in one or two given directions). The
action of the fractured area is described by means of one or two
fictitious springs at the integration points located normally to the
given directions of the cracks and simulating the action of the
cracks around the integration point under consideration (Fig. 6b).

c directions of cracks at P1
a

.. : ...
L ....l..~\, .....
~ . . . . . ..
. . . 1'-
... , .
a
.. .. :...
B E
a)
c
b) c)
Fig. 6, On the Determination of Cracked Zones in a Structure
116 P.D. Panagiotopoulos

Note that for each crack direction we may introduce both a normal
and a tangential spring in order to study the tangential interface
effects (e.g. friction) at the crack. The directions of the fic-
titious springs may be different at every integration point but
they indicate the directions of all the cracks in the area of the
proportionate subdomain of the element. Thus, if wp is the weight
coefficient of the integration point PP for the numerical integration
and F is the area of the considered element ABC then the cracked or
crushed area around Pp is wpF (Fig. 6b). The aforementioned springs
have a stress-strain law like the one depicted in Fig. 6c whose
values are determined by the ultimate strength in tension and com-
pression of the material. Further nonlinearities of the stress-
strain law of the material may be incorporated into the classical
finite element analysis of the structure. Due to the consideration
of the law of Fig. 6c we can determine the areas which are cracked
(o=O EE[e,f)), crushed (o=O EE(a,b] ), intact ((o,e:) is on cd) or
they are at an intermediate stage ((o,E) on de or de' and on cb or
ca' ). The problem is described by the hemivariational inequality
(cf. (2.95))

n o n' o ~o
}:; w.(Ei(u),E.(v-u))+}:; (j (u ,v -u )+j (u ,v -u ))+
i=1 1 1 p=1 p p p p p p p p

+ITR(v)-rrR(u)+rr~/C(u,v-u)~ pT(v-u) vvEXP (2.96)


~

where jp and jp denote the superpotentials of the fictitious springs


at the point PP and wi is the strain energy of the intact material
of the i-th finite element. Compared with the bilateral method of
[17] based also on a "smeared-fracture" concept, the method presented
here seems to describe the behaviour of the structure with greater
generality, based on a smaller number of assumptions, but no numeric-
al experience yet exists.
iii) The use of the nonconvex superpotentials permits the rational con-
sideration of the tension stiffening effect in reinforced concrete
Hemivariational Inequalities 117

structures, i.e. of the effect representing the capacity of the in-


tact concrete between adjacent cracks to sustain tensile stresses.
For the study of this effect stress-strain laws of having the form
of Fig. 1g (resp. Fig. 5e) are proposed for the concrete in tension
(resp. the tension steel) [17], which obviously can be written in
terms of nonconvex superpotentials (cf. e.g. (2.44)).

2.8 On the Properties of Solids Containing Voids or Cracks


The aim of the present section is to show the mode of inheritance of
certain ultimate strength properties from the microscopic to the macro-
scopic level. Let us consider in the framework of small strain theory a
nonlinear elastic solid characterized by the stress energy density
function ~w(o) assumed to be locally Lipschitz and a-regular. Also let
us consider a macroscopic volume element V of this solid which includes
certain plane microcracks and traction-free voids. We denote by VM
(resp. Vv) the region occupied by the material (resp. by the voids) and
let Ac be the planes of the cracks included into VM. We denote here by
E and E the macroscopic stress and strains; generally, capital letters
denote macroscopic quantities whereas minuscules denote local pointwise
(called here in an abuse of the language "microscopic") quantities. The
element V is loaded on its boundary r by a uniform traction Eijnj where
n={ni} is the local outward unit normal and E is independent of the po-
sition. This loading causes at a microscopic level the strains E, the
stresses a and the displacements u. According to [35] the macroscopic
strain is given by the formula

E.. = 21v J(u.n.+u.n.)dr ( 2. 97)


1J r 1 J J 1
Assume now that
(2.98)
If L:*-L: is a variation of L: we have, applying the principle of virtual
work, that
118 P.D. Panagiotopoulos

E.. (~~.-~ .. ) V-1 I u.n.(~~.-~ .. )dr = V-1 I u.n.(a~.-a . . )dr =


1J 1J 1J r 1 J 1J 1J r 1 J 1J 1J

v- 1I e: . . (a~ .-a . . )dV;;; v- 1I w0 (a,a*-a)dV (2.99)


v 1J 1J 1J v
M M
Let us introduce the "macroscopic" superpotential

~-+ W(~) = .!. Iw(a)dV (2.100)


vv.,
Due to the a-regularity of the locally Lipschitz microscopic superpo-
tential o-.w(a) we can show (see the proofs in Sec. 2.6) that

W 0 (~.~*-~) =~I w0 (a,a*-a)dV (2.101)


VM

Thus (2.99) and (2.101) imply that


EEaW(~) , (2.102)
where Wis again a-regular. Thus (2.98), (2.100) and (2.102) indicate
how the properties of the E-+cr relation are inherited in the macroscopic
E-~ relation. For instance, if e:-o has sawtooth parts simulating e.g.
the friction on the crack planes, then (2.100) "transmits" this proper-
ties to the macroscopic level. The same happens obviously with the ulti-
mate strength properties of the microscopic level: a e:-o law of the form
of Fig. 6c is inherited by theE-~ level in the sense of (2.100) (con-
sider e.g. a onedimensional homogeneous macroscopic element). Analogous-
ly we may argue for the relation between microscopic and macroscopic
yield frunctions. As shown already (cf. [35] and the references given
there), even if the local yield surface is independent of the mean stress,
the macroscopic yield surface for a material containing voids and cracks
will generally depend on ~ii' will be convex, and nonsmooth. Repeating
the proof of (2.30) in Sec. 2.2 we can show that the convexity is no
longer guaranteed if the state of voids and cracks changes with the e-
lastic deformation (cf. [12] p.154) thus causing a change of the elastic
Hemivariational Inequalities 119

properties of the macroscopic volume element. We leave this proof as an


exercise for the reader.

2.9 V-superpotentials, Unloading, Numerical Methods and Miscellanea


i) The subdifferential a and the generalized gradient a are the oper-
ators which give rise to variational and hemivariational inequalities
when acting on superpotentials. The arising mechanical laws are multi-
valued and with respect to (2.3) and (2.4) f(x 1)-f(x) and f 0 (x,x 1-x)
express the virtual power (or work) of the constraint (or the law) for a
generalized velocity x and for a virtual generalized velocity x1-x
Let us put ourselves in the framework of (2.8). On the mechanical system
~ a multivalued mechanical law of the form

- f~H ( u) ( 2. 103)
is introduced, which is by definition equivalent to the expression
W(u)= {fEFIG(u,v-u,{B},{8},path,history,time derivatives, ... )~
~ <f, v-u> VvEU}. (2.104)
G is an appropriately defined function expressing the virtual power (or
work) of the system. It may be of a nonlocal nature depending on the ob-
servable variables {B} and on the hidden variables {8}, on the "path" of
the variation v-u, on the history of the mechanical system etc. The law
(2.103) is called V-superpotential (virtual power superpotential) law
and may serve for the following "Hypothesis of Dissipation".
(H) for every real thermodynamic process a V-superpotential Wcan be de-
termined such that the "flux" uEU associated with the "force" fEF satis-
fies (2.103).
This hypothesis is a generalization of the hypothesis of normal dissi-
pation of [21]. We can easily verify that if
G(u,-u,{B},{8},path,history, ... ) ~ 0 (2.105)
then the inequality fiui~ 0 expressing the second principle of thermo-
dynamics is satisfied. If furthermore we assume that
G( u, v, ... ) > G( u, o, ... ) WEU, H 0 ( 2. 106)
120 P.D. Panagiotopoulos

then we can show that, if at u


G(u,o, ... ) ~ 0 (2.107)
then there corresponds to u at least one force f such as to satisfy
(2.103) and conversely. Indeed if (2.103) holds we obtain from
(2.104) for v=u (2.107). Conversely from (2.107) and (2.106) we ob-
tain that W(u)*0 since F and U are finite dimensional spaces. A
special case of the V-superpotentials are the superpotentials con-
nected with Ioffe's fans [36]. Through the use of V-superpotentials
we may define generalized viscoplastic and viscoelastic materials by
following the method of [12] p.158. Defining also "normal" cones
through the indicator we can adapt the ideas of [37] to this more gener-
al framework in order to obtain more realistic friction laws (cf.
also [18]).
ii) Until this point we have not considered the case of unloading in the
stress-strain relations which we have introduced. We recall that
all these relations are generally multivalued and nonmonotone. In
[38] a method was developed for the calculation of a structure for a
given unloading path and the method of the "macroincrements" was de-
veloped. Here we shall present a more general method which takes
into account all the possibilities of loading and/or unloading (not
only along a given unloading path), i.e. it describes completely the
phenomenon. This is possible by using appropriately defined multi-
functions involving E,cr,E and cr. Here the dot means the time
partial derivatives (assumptions of small strains and displacements).
Let us assume that the unloading is linear and that the modulus of
elasticity changes with the strain. In Fig. 7 the onedimensional
case is depicted or an equivalent stress-strain uniaxial law.
At each point along the softening branch AB we pose the question
whether E={Eij} is positive, i.e. that we remain on AB, or negative
and then the elastic unloading paths AC, A'C' etc. should be real-
ized. If the stress-strain of an element are on OAB at the end of
a load incremen~ then for the next load increment we can write that
Hemivariational Inequalities 121

ae: acp(e:) if e:<e: 0 , or if E~EO and E~0 ( 2. 108)


0

a= C(e:)e: if e;;;:e: 0 and e:< 0 (2.109)

If the stress and strain of an element are, e.g., on A'C' then


a= C(e:')e: (2.110)
where e:' denotes the strain at A'. We may easily conclude that
(2.108-110) can be put in the form
OEacp(e:)l!J(e:)+ [acp(e:)x+ + C(e:)(e:)xJq(e:,a)(1-l!J(e:))+
+ C(e:')(e:)(1-q(e:,a))(1-l!J(e:)) , (2.111)
where l!J(e:)={1 for e:<e: 0 , 0 for e:~e: 9 } and e: 0 is the strain for which
irreversible strains appear, x= *and x and x are the positive
1 I + -
and the negative parts of x respectively, q(e:,a)={1 if (e:,a)EE,
0 otherwise} with E={(e:,a)jaEacp(e:)}. Note that in (2.111) C(e:') is
assumed as known and corresponds to (e:,a)tE. The aforementioned
thoughts hold for any type of nonmonotone stress-strain law with un-
loading, i.e. uniaxial or multiaxial laws. Thus, for instance, if
(2.44) holds then the corresponding a,e: and E: satisfy an inclusion
similar to (2.111) with the difference that l!J(e:)={1 if e:Eelastic
region, 0 otherwise}. Note that a characterization of the elastic
and plastic regions in terms of e:, a or both is necessary for the
formulation of any loading-unloading law. The general form of such
a law is a differential inclusion of the form
OEF(e:,a,E:) , (2.112)
where F is an appropriately defined multifunction. For instance, the
classical elastoplastic law is a special case of (2.112). Let us
know a and e: for a given load p and let a new load Pot be imposed.
Now for the total load p+pot the stresses and strains become a~t
and e:+E:ot and we may write generally an incremental relation of the
form
(2.113)
122 P.D. Panagiotopoulos

Fig. 7, Nonmonotone Law with Unloading

For (2.112) the B.V.P will be defined by the following relations in


the framework of small strains and small displacements:
Find in Qx(O,T) a displacement field u which satisfies
1(u . . +u .. ) , OEF(E,o,)
pU .=oiJ" J""f., E.. - 2 (2.114)
1 ' 1 1J 1,J J,1
together With the initial conditions U=U , u=u 1, o=o and the bounda-
0 0
ry conditions e.g. given displacements on ru and given tractions on
rF (r=rFUruUr,mesr =0, rF and ru nonoverlapping). For (2. 113) the
corresponding B.V.P (e.g. the quasistatic one) reads:
Find in Qx(OT) a displacement increment u such as to satisfy
1
a .. . +p.=O, E: ~(u . . +u .. ),oEF 1(E,O,E) (2.115)
1J,J 1 1J L 1,J J,1
together with the initial conditions u=u 0 and o=o0 and the boundary
condition (e.g. as before). Both BVPs are still open; it seems until
now that some polynomial growth assumptions for some selections of
the multifunctions play an important role in the discussion of the
existence of the solution.

iii)Several methods have been proposed until now for the numerical treat-
ment of the hemivariational inequalities. Let us begin for the
moment with the monotone case and with a very special discrete
problem, the elastic cable structures for small displacements (see
e.g. [39],[40]):
The arising system of equalities and inequalities is equivalent to a
quadratic programming (Q.P.) problem (cf. also [12] p.350).
Hemivariational Inequalities 123

The initial complete system of equalities and inequalities- and


under the term "complete" we rrean that we have to consider both the
static and the kinematic inequalities as well as the complementarity
relations - are the Kuhn-Tucker conditions for the Q.P. problem,
which, if the matrix of the quadratic form is positive definite, are
necessary and sufficient for the existence of the solution. The nu-
merical treatment of such a cable-structure under a given loading
seems to be difficult, because we do not know a priori which cables
are slack and which not, and this thought might lead someone to the
use of a combinatorial-like method. However things are more simple!
Indeed the inequality constrained Q.P. problem when numerically
solved by an appropriate Q.P. algorithm "decides'' which cables are
slack and which have tensile forces. For a positive definite matrix
the solution is unique. Moreover any classical method (incremental-
iterative, trial and error etc.) can be applied; its solution will
be solution of the problem if it satisfies (cf. [41]) the complete
set of Kuhn-Tucker conditions i.e. all the statical and kinematical
equations and inequalities together with the complementarity con-
ditions (stating that a cable will be either slack or will have a
tensile force i.e. sivi=O, where si~O is the cable force and vi~O
is the negative slackness of it). Any attempts for obtaining the
solution through trial and error methods which do not control final-
ly all the Kuhn-Tucker conditions may lead to a wrong result (cf.
[12] Ch.10). Now let us assurre that the final value of the loading
is attained through a sequence of load increments p. Suppose then
that by using any classical solution scheme we find a displacement,
a strain and a stress field. Obviously we have to check whether this
"solution" satisfies the Kuhn-Tucker conditions and then we may con-
clude that it is the solution of the problem due to the uniqueness
of it. However the classical schemes of increrrental methods (cf.
[42]) may not be able to give a result due to the horizontal and
possibly vertical part of the graphs of the cable law (e.g. if the
law si~O, vi~O, sivi=O is considered).
124 P.D. Panagiotopoulos

Analogous questions arise in all problems expressed in terms of


variational inequalities (monotone case) i.e. involving convex super-
potentials. Again wemay apply aminimization procedure, or any other
c1ass ica 1 incrementa 1 i ntera ti ve method a1ways with the restrictions
cited before: the final result must satisfy all the static and kine-
matic relations of the problem and the incremental method may fail
to work for horizontal or vertical parts of the arising E-o material
laws. For these parts the use of Q.P. algorithms is advisable in
order to overcome this difficulty (cf. e.g. [40] ,[41]). Moreover,
in the case of large displacements or lack of uniqueness of the so-
l uti on [43], one must be careful in the application of an increment-
al iterative method. The increments must be small enough to avoid
the omission of equilibrium configurations during the incremental
procedure. Indeed, because of stability questions, there is an in-
herent risk of jumps from one solution path to another (cf. e.g.
the bifurcation mode of a simple cantilever beam). For nonconvex
superpotentials and hemivariational inequalities we have analogous
but more difficult questions (cf. also [44]). Let us consider first
the purely static problem: The total load p is given and we want to
determine all possible positions of equilibrium. The problem is
formulated as a substationarity problem and every local minimum of
the potential energy corresponds to an equilibrium configuration.
There are several numerical difficulties concerning the determi-
nation of a local minimum of a nonconvex generally nondifferentiable
function. We mention here as a possible tool the bundle method of
Lemarechal-Strodiot [45] which is now at the stage of experimen-
tation concerning its applicability to this problem. Analogously
to the case of cable structures, and, let us say, with respect to
the diagram of Fig. 1g, one could make some combinatorial attempts
to find on which part of Fig. 1g the solution can appear. But the
substationarity property of the solution gives the answer and no
need of combinatorial thoughts exists. Another method related
to the previous one is the regularization method whose convergence
Hemivariational Inequalities 125

has been already discussed by the author in several problems ([46] +


[48] ). Let us describe the method with respect to the hemivariation-
al inequality (2.58). We assume that superpotentials jN(q) and
jT(q) are approximated by the smooth functions jNE and jTE depending
on E (we omit (q)), such that, as E~O jNE (resp. jTE) tends to jN
(resp. jT). Any heuristic method for the regularization can be
used: For instance the graphs of the corresponding stress-strain
laws can be appropriately smoothened and vertical lines should be
replaced by inclined lines. Due to the smoothness of jNE and jTE'
the derivatives j~(E,z), j~(E,z) are replaced by
djN(E) djr(E)
dE z and dE z and thus we are led to consider the follow-
ing problem: Find u~m) m=1,2, . ~. such as to satisfy the vari-
ational equality
k d"
E a(u(m),v(m))+ E { f{[ JNE(q)([u(q)])[v(q)]+
m= 1 E q= 1 r duN NE N
q

(2.116)

By means of discretization, (2.116) is equivalent to a system of


semilinear algebraic equations depending on E. The solution of this
system "tends" as E~O to the solution of the initial problem. The
system of nonlinear algebraic equations has the general form
Ku+B E (Tu) = p (2.117)
where u (resp. p) is the total displacement (resp. load) vector and
K is the stiffness matrix. Here T is a transformation matrix such
that Tu gives the appropria~e relative displacements and BE() is
126 P.D. Panagiotopoulos

the vector of the nonlinear terms depending on E. Noteworthy is the


sparsity of the nonlinear terms which in some cases only slightly
influences the monotonicity of the problem. In this case the classi-
cal Newton-Raphson algorithm has given acceptable results. However,
it remains still an open question the testing of other types of
algorithms like the fixed point algorithms and the homotopy algo-
rithms. At this point let us make a parenthesis and let us examine
what happens if the superpotentials depend on S(q) which itself is
an unknown function of u. Then the foregoing methods cannot be used.
This happens, for instance in the case of real Coulomb's friction.
For such problems the following algorithm is proposed, which seems
to be reasonable from the standpoint of mechanics, on the assumption
that in (2.58) jN depends on STand jT depends on SN. We split the
problem into the following two subproblems: In the first (resp. the
second) we assume that S~q)= C~~),(resp.S~q)=C~~)),where C~~) (resp.
C~~)) is given and we solve (2.58) with j~(q)= 0 (resp. j~(q)= 0)
and with the terms I c(q~(v(g)_u(g))ctr (resp. I c(q)(v(q)_u(q))dr)
r T11 T1 T1 f' N2 N2 N2
q q
q=1, .. k, added in the right-hand side of it. Obviously, the first
problem is the normal contact problem and the second is the tan-
gential contact problem. From the first problem we obtain a value
for S~q), say C~~), which is used for the solution of the second
problem. From it we obtain a value of S~q), say c~j), which is used
for the solution of the first problem, and so on, until the differ-
ence of the displacements of two consecutive steps becomes appropri-
ately small. The proposed algorithm recalls the fixed point algo-
rithm of [49] proved in [50], and the decomposition techniques in
multilevel optimization [51].
Let us come back now to the numerical implementation of hemivari-
ational inequalities with strong nonmonotonicities. In this case
the microspring or sawtooth approximation of the decreasing branches
is proposed by Koltsakis [52]. Then the Newton-Raphson algorithm is
applied. This idea is roughly described in Fig. Sa. The dotted
Hemivariational Inequalities 127
a

:/

lZT
w
A c

E
a) ( b)
8'
( c)

Fig. B, Microspring Approximation of Decreasing Branches


and the Question of Stability

vertical parts of the graph denote local jumps and the decreasing
branch is "attained" by small monotone increasing "stress recovery"
parts AA' ,BB'. The physical explanation of this approximation is
obvious and is based on the idea of gradual failure of the material
on the decreasing branch. Small numerical examples solved by this
method are given in [23] and in [53]. The method is fully examined
in [52] and has certain advantages concerning its ability to describe
the localization of the failure. However there are considerable nu-
merical difficulties in obtaining numerically all the solutions of
the problem. At this point we would like to remark that a structure
may have a position of overall equilibrium in which one ore some ele-
ments of the structure are in unstable equilibrium. For instance
the framework of Fig. Bb with strong linear elastic vertical and
horizontal beams is always in equilibrium under the loading P even
if the diagon~l AB has a stress-strain law derived by the potential
function of Fig. Be. Note that a simple rod-element having this
strain energy would have as a position of stable equilibrium the
point Band as other possible position of equilibrium the point C
and all the points between rand~ (r~ is an horizontal line segment).
Indeed all these positions of equilibrium are substationarity points.
It is also noteworthy that a convexification of the energy would e-
liminate several positions of equilibrium (dotted line in Fig. Be).
Let us now assume that the final value of the loading is attained
128 P.D. Panagiotopoulos

through a sequence of load increments. Let us then apply a classi-


cal incremental-iterative procedure. Note that in analogy to the
numerical treatment of classical plasticity with or without work-
hardening the ultimate strength surface introduced in Sec. 2.4
will be used for the correction of the incremental paths. On this
approach the same comments as for the monotone case can be made,
whereas the danger of obtaining a "wrong" result is greater due to
the lack of monotonicity of the problem. It is well-known (see e.g.
[44]) from the numerical work on nonmonotone laws with the classical
incremental-iterative methods that severe numerical difficulties
appear especially if one wants to take into account all the causes
of structural failure in a finite element scheme, for instance com-
plete vertical jumps in a stress-strain diagram for which the
classical incremental-iterative schemes may have difficulties.
Analogously to the large displacement theory of cable structures
where an incremental-iterative procedure is combined with a Q.P.
problem [42],[43], we propose here combination of the incremental-
iterative procedures with a substationarity problem which can take
into consideration irregularities like jumps; then the numerical
solution within the increment in which the substationarity problem
is considered is obtained from a local optimization problem or from
a regularization procedure. Closing this Section we would like to
comment that the numerical efforts concerning the hemivariational
inequalities are still at the stage of experimentation.

3. STUDY OF VARIATIONAL HEMIVARIATIONAL INEQUALITIES


3.1 Delamination Effects of Laminated von Karman plates.
Formulation of the Problem
In this Section we show the general method for the proof of existence
and approximation of the solution of a problem connected with nonconvex
superpotentials. This is achieved by studying a mechanical problem: the
delamination effect for laminated plates undergoing large displacements
(von Karman plates) is studied. Recall that the delamination, which is
Hemivariational Inequalities 129

mainly due to the normal interlayer stresses [54], is one of the main
causes of strength-degradation of composite plates. Here the mechanical
behaviour of the interlayer binding material, together with the possi-
bility of local cracking and crushing effects, are described by a non-
monotone, possibly multivalued law between the interlaminar bonding
forces with the corresponding relative displacements. This law can be
derived from a nonconvex superpotential and leads to a hemivariational
inequality. Here in order to study a larger and more realistic class of
problems we assume at the boundary of the von Karman plate monotone
boundary conditions which include e.g. the Signorini-Fichera boundary
condition, the plastic hinge condition, etc. (see e.g. [7],[12]). These
conditions are derived through subdifferentiation from a convex generally
nondifferentiable and nonfinite functional, the convex superpotential of
Moreau.
Unilateral problems for a single plate have been studied for monotone
boundary conditions in [7] and [12] and for nonmonotone boundary con-
ditions in [47]. A laminated-plate theory concerning Kirchhoff plates
(small displacements - no stretching) has been presented in [55],[56]. We
refer also at this point the work of Naniewicz and Wozniak [57] ,[58],[59]
on the hemivariational inequalities and their applications to the de-
bonding processes in layered composites.
Let us consider a laminated plate consisting of two laminae and the
binding material between them (Fig. 9). In the undeformed state the
middle surface of each lamina, say the j one, occupies an open, bounded
and connected subset Qj of R2 Qi is referred to a fixed Cartesian
right-handed coordinate system Ox 1x2x3 . The boundary of the j-th lamina
rj is assumed to be a Lipschitz boundary C0 ' 1 The interlaminar binding
material occupies a.subset g' such that o'co 1no 2 and ono 1 =~. ono 2 =~.
Let us denote by 6(J)(x) the vertical deflection of the point xEQ. of
the j-th lamina, by f(j)=(O,o,fp)(x)) the distributed vertical l~ad on
the j-th lamina, and by u(j)={u~j) ,u~j)} the in-plane displacements of
the j-th lamina. The j-th lamina has constant thickness hj' while the
interlaminar binding layer has constant thickness h. For each lamina
130 P.D. Panagiotopoulos

the von Karman plate theory is adopted, i.e. each lamina is regarded as
a thin plate undergoing large deflections. Thus we have the following
system of differential equations:
K.M {;( j ) - h . (a ( j ) ~ ( j ) ) f (j ) in Q.
J J ae ,B ,a J
(3. 1)
a( j) =0 in Q.
ae.B J '
(3.2)
and a(j) = C(j) (E(j)(u(j))~(j)6(j)) in Qj (3.3)
ae aeyo yo 2 ,y ,a
The subscripts a,B,y,6 = 1,2 indicate the coordinate directions; {a~l},
{~)} and {C~~ 0 } are the stress, strain an~ elasticity tensors in the
plane of the j-lamina. The components of c(J) are assumed to be ele-
ments of L~(Q.) and they satisfy the well-known symmetry and ellipticity
properties ofJthe Hooke's tensor. Moreover, Kj= Eh~/12(1-v 2 ) is the
bending rigidity of the j-th plate withE Young's modulus and v Poisson's
ratio. Here we consider isotropic homogeneous plates of constant thick-
nesses. However, in place of (3.1-4) we could assume the appropriate
equations for orthotropic or anisotropic plates in bending. The mathe-
matical treatment is the same. In order to model the action of the in-
terlaminar normal stresses a 33 , f(j) is splitte~ ~nto f(j), which de-
scribes the interaction of the two plates, and f(J)EL 2(Qj), which re-
presents the external loading of the j-th plate:
f(j) = f(j)+ f(j) in Qj' j=1,2. (3.4)
If f denotes the stress in the interlaminar binding layer, we have that
f = 1( 1) = -1( 2 ) in Q' (3.5)
Then a phenomenological law connecting f with the relative deflection of
the plates (see Fig. 9b)
[{;]=6( 1)-6( 2) (3.6)
1\
is introduced. Let f be a multivalued nonmonotone funct1on B (cf.
Fig. 10a), of [~;] i.e.
1\
-fEB([{;]) in Q' (3.7)
Hemivariational lnequalities 131

r1

(a)

=(1)
+1

(b)

Fig. 9, The Laminated Plate Geometry


132 P.D. Panagiotopoulos

Note that cracking as well as crushing effects of brittle or semi-britee


nature can be described by this law. The impenetrability restriction
would imply a vertical branch AB in Fig. 10a. However, a lightly in-
clined branch AB' is considered here in order to take into account the
slight compres~ion of the lamina in the Ox 3-direction. We assume
further for f(J) that
1( j ) = 0 in Qj - Q' , j =1, 2 (3.8)
Now we shall express relation (3.7) in terms of a nonconvex super-
potential. Thus we assume that S: m~~ ffi) is defined in the following
way. Let 13ELi0 c( ffi) and let
]3 (E)= esssup 13(E 1) and 13 (E)= essinf 13(E 1) . (3.9)
s IE1-EI~E '- IE1-EI~E

These are increasing and decreasing functions of s, respectively, for


any s>O, EEIR; thus, their limits for E-+0 are well-defined; let these
limits be denoted by J3(E) and 13(E) respectively. The multivalued
1\ -
function 13: IR~ IR) is now defined by the interval-expressi on
1\ -
13(E) = [_(E), 13(~)] . (3.10)
If 13(~ 0 ) exists for every EE IR, then [60] a locally Lipschitz (non-
convex) function J: ffi-+ffi can be determined up to an additive constant,
such that
1\ -
13(E) = aJ(~) (3.11)
Thus (3.7) is now written in the form
-fE~( [~J) = aJ( [~J) (3.12)
0
Here the directional differential of F.H. Clarke, J (.,.) is

(3.13)

Moreover, we assume that the following boundary conditions hold on the


subset rj of the plate boundaries:
Hemivaria tional Inequalities 133

-f. p"

A
I
I
I
I

B ' B
a)

b - O,M

b)

Fig. 10, The Binding Behaviour of the Interlaminar Adhesive (a)


and Subdifferential Boundary Conditions (b)
134 P.D. Panagiotopoulos

( .) d (j)
M. (1: J )Eb. (-1:- ' j=1,2' (3. 14)
J J 3n

-Q. ((:(j) )Eb '. ((:(j) ' j=1,2 ' (3.15)


J J

and h:(j) = ~ = 0 on r.-r. ' j= 1 '2 (3.16)


an J J

Here bj' bj, j=1,2 are generally multivalued maximal monotone operators
from 1R into !P( lR) [12], i.e. complete nondecreasing curves [61] (see
also the graphs of Fig. 10b). Accordingly, convex, l.s.c, proper
functionals [62] ~j' ~j j=1,2 can be determined such that
a(:(j) 31: ( j)
bj(--ar;-) d~j (--ar;-)' j= 1,2 (3.17a)

and b~(h:(j)) a~~(h:(j))' j=1,2 . (3.17b)


J J
For the physical meaning of the boundary conditions (3.14-16), which in-
clude also all the classical boundary conditions, we refer to [7], [12].
Futher we introduce the Sobolev space H2(n.) for the deflections h:(j)
and we define the convex, l.s.c and properJ functionals on H2(nj)' j=1,2:

a (j) ()
J~.(-z-)dr-{~' .(z J )dr,
r. J an r. J
J J
d(j) 1 () 1
if ~j(~)EL (rj) and ~j(z J ) L (rj)

oo otherwise, j=1 ,2 . (3.18)


The in-plane displacements are assumed to fulfill the boundary conditioffi

(3.19)

For the derivation of the variational formulation of the problem we as-


sume first sufficiently regular functions. Then multiplying (3.1) by
z(j)_h:(j), integrating and applying the Green-Gauss theorem, implies
the expressions:
Hemivariational Inequalities 135

j=1 ,2, a,[3=1 ,2 . (3.20)


Here n(j) denotes the outward normal unit vector to rj'
a(~,z)= Kf[(1-v)~ ~~z ~~+v~~~z]dQ, a,[3=1,2, 0<V<0,5 , (3.21)
g o'-'4-'o'-'4-'

2 2
M(~) = -K[v~~+(1-v)(2n1n2~,12+n1~,11+n2~,22] ' (3.22)

- a~~ a 2 2
(3.23)
Q(O = -K[Tn+ ( 1-v) a1: [n1n2(~.22-~,11)+(n1-n2)~,12]]

Here 1: is the unit vector tangential to r, such that n,1: and the Ox 3-
axis form a right-handed system. In (3.22-23) Mand Q are the bending
moment and the total shearing force (Q= Q-aM/ac) respectively on the
boundary r. Analogously we get from (3.2) the expression

j=1 ,2, a,[3=1 ,2 (3.24)


The following notations are introduced:
R(m,k) = fC~~ 5m~~k 0 dQ, a,[3,y,5=1,2 (3.25)
g '-'4-'y '-'4-' y

and
(3.26)

where m= {mae} and k= {kae}, a,[3=1 ,2. We assume also that


u(j) ,v(j)E[H 1(n.)J 2 and that ~(j) ,z(j)EZ., where
J J
136 P.O. Panagiotopoulos

z.J {ziz H2(nJ.), z=O on r.-r.


J J

(3.27)

Taking into account (3.25, 3.26), (3.20) and (3.24), the boundary con-
ditions (3. 14-16) and (3. 19), the inequalities defining the subdiffer-
entials a in (3. 17a,b), and (3. 18), as well as the interface conditions
(3.5),(3.8),(3. 12) and the definition of the generalized gradient a we
obtain from (3.20) for j=1 and j=2 and addition and from (3.24) the
following problem~. in which due to the aforementioned addition
the relative deflection [~] appears.
Problem~: Find u(j)E[H 1(nj)] 2 and ~(j)EZj, j=1 ,2, such as to satisfy
the variational-hemivariational inequality

~ ~
a. . ( ( j ) , Z ( j )- ~ ( j ) ) + ~ h . R(E( u( j ) ) + _lp (~ ( j ) ) , P( ~ ( j ) , Z ( j ) -~ ( j ) ) ) +
j=1 J j=1 J 2

+ fJ 0 ([~],[z]-[~])dQ+ ~ {<I>.(z(j))-<I>.(~(j))};;;
Q' j =1 J J

;;; ~ ff(j)(z(j)_~(j))dQ, vz( 1)EZ 1, vz( 2)EZ 2 , (3.28)


j=1~

and the variational equalities

R(E(U(j))+~P(~(j)), E(V(j)_u(j))) = 0 ,
vv(j)E[H 1(nj)J 2, j=1,2 (3.29)

Inequality (3.28) is called a variational-hemivariational inequality,


because of the convex (resp. nonconvex) terms <I>(z)-<I>(~) (resp. J 0 (.,.))
as in the theory of variational (resp. hemivariational) inequalities,
and together with (3.29) is the expression of the principle of virtual
work.
It is obvious that the same method is applicable to the case of r-lami-
nae. Then in (3.28) the summation j~ 1 is replaced by j~ 1 , the term
Hemivariational Inequalities 137

m'
J J 0 (.,.)dn is replaced by the term J 1 ,J J~(.,.)dn, where m'=r-1 is
n' n
(m) ( ")
the total number of interfaces, and (3.28) must hold for very z J EZj'
j=1, ... ,r.
Several other types of boundary conditions may be considered. The only
condition which should be fulfilled is that the classical boundary con-
ditions must ensure the coerciveness of the quadratic form of bending
aj(~(j) .~(j)), i.e. they must prevent any rigid-plate deflection. This
is guaranteed, if, e.g., a plate is partially clamped or simply sup-
ported on a curved boundary part. Moreover the boundary conditions
(3.19) can be replaced by u1=u 2=0 or any other combination of classical
boundary conditions which, together with the boundary conditions in
bending, will guarantee the vanishing of the term
J h.o(j)~(j)n(j)(z(j)_~(j))dr in (3.20). For nonhomogeneous classical
r. J al3 ,13 a
J
boundary conditions we obtain through an appropriate translation analo-
gous variational problems as in the homogeneous case. In the general
"'
model problem P', which will be studied further, the classical boundary
conditions of each plate in bending define the kinematically admissible
sets z. which are assumed to be subspaces H2(n) on which the bilinear
form a~(~(j) .~(j)) are coercive. Our study is mainly based on [63].

Problem~: Find ~(j)EZj and u(j)E[H 1(nj)], j=1, . ,r, such as to


satisfy the variational-hemivariational inequality

~ a.(~(j),z(j)_~(j))+ ~ h.R(E(u(j))+l(~(j)),P(~(j),z(j)_~(j)))+
j=1 J j=1 J

+ ~ J J~([~] (m) ,[z] (m)_[~] (m))dl+ f {cll.(z(j)')- cll.(~(j))} ~


m=1n' j=1 J J
m
~ ff f(j)(z(j)_~(j))dn, YZ(j)EZ., j=1, . ,r. (3.30)
j=1n. J
J
and the variational equalities
138 P.D. Panagiotopoulos

R(E(u(j)l+fU:(j)), dv(j)_u(j))) = 0,
vv(j)E[H 1(Qj)J 2, j=1 , ... ,r. (3 .31)

3.2 On the Existence of the Solution. Convergence of the Galerkin Scheme


R(.,.) is a continuous, symmetric, coercive bilinear form on
[L (Q)] 4 , and P:[H 2(Q)] 2-.[L 2(Q)J 4 of (3.26), is a completely continuous
2
operator [64]. Thus (3.31) implies by the Lax-Milgram theorem that to
every deflection \':(j)EZ., j=1,2, ... ,r there corresponds a plc.ne displace-
( .) (") 2 J2
ment u J (\': J )E[H (Q.)] . Indeed by Korn's inequality [2] R(E(u),s(v))
is a bilinear coerciv~ form on the quotient space [H 1(Q)J 2/R, where R is
the space of in-plane rigid-plate displacements defined by
- -,- 1 2 - -
R= {r: rE[H (Q)] , r 1 = a 1+bx 2, r 2=a 2-bx 2 , a 1, a2 , bE lR}
Thus (3.31) implies that E(j)(u(j)(\':(j))):Zj->[L 2(Qj)J 4 is uniquely
determined and is a completely continuous quadratic function of \':(j),
j=1,2, .. ,r, since E(j)(u(j)(\':(j))) is a linear continuous function of
P(\';;(j)). Then the completely continuous, quadratic function G.:Z.->
-.[L 2(Qj)J 4 defined by: J J

\': ( j) ->G j ( \': ( j) ) = E( j ) ( U ( j ) (\': ( j ) ) ) + ~ ( \': ( j ) ) ( 3. 32)

is introduced, which satisfy the equations

(3.33)

Let us now introduce the operators A.: Z.->Z '. and C.: Z.-.z '. (where ZJ'. is
J J J J J J
the dual space to Z.) such that
J

a(\':(j) ,z(j)) = <A.(';;(j) ,z(j)> (3.34)


J
and
(3.35)
A. is a continuous linear monotone operator, C. is a completely continu-
J J
ous operator, and<.,.> denotes the duality pairing between Zj and Zj.
Hemivariational Inequalities 139

Moreover

<C.(~(j)),~(j)> = h.R(G.(~(j)),2G.(~(j))) ~ 0,
J J J J
'v'~(j )EZj, j=1 ,2, .. ,r (3.36)
From the coercivity of aj(~(j) .~(j)) on Zj we obtain

<(A . +C . ) (~ ( j ) ) '~ ( j )> = <T . ( ~ ( j ) '~ ( j )> ~ c II ~ (j ) II 2'


J J J
'v'~(j)EZ., j=1,2, ... ,r, cconst.> 0, (3.37)
J
where II Ii denotes the Z.-norm.
Thus Problem~ takes theJfollowing form (here (.,.) is the L2-product):
Problem P: Find ~(j)EZj' j=1,2, ... ,r such as they satisfy the vari-
ational-hemivariational inequality

~ <T.(~(j)),z(j)_~(j)>+ ~ fJo([~](m),[z](m)_[~](m))dQ +
j=1 J m=1~ m

+ ~ {~.(z(j))-~.(~(j))} ~ ~ (f(j) ,z(j)_~(j)),


j=1 J J j=1
'v'z(j)EZ .. (3.38)
J

The basic assumption for the existence and the approximation of the so-
lution of our problem, concerns the interface law: For some EE R

esssupf3(m)(E) :s essinff3(m)(E), m=1,2, .. ,m', (3.39)


( -oo,-E) (+E,-t<x:>)
i.e. that the nonmonotone f3 (m) increases "ultimately" to the right.
Obviously one can assume by performing an appropriate translation of
the coordinate axis, that for some EE R
ess sup f3(m) (E) :s 0 :s ess inf f3 (m) (E), m=1 ,2, ,m' ( 3 .39a)
(-oo,-E) (+E,-t<x:>)
Two cases will be separately treated: The "differentiable" and the
"nondifferentiable". In the first the grad~j exists everywhere for
140 P.D. Panagiotopoulos

every j, in the second this does not occur.


a) The "differentiable" case: Relation (3.38) of P is equivalently
written in the form

; <T . ( ~ ( j ) ) , / j ) -~ ( j )> + ~ JJ 0 ([ ~] ( m) , [ z l ( m) - [~ l ( m) ) dr.l +


j=1 J m=1r.l' m
m
+ ~ <grad~. (~ ( j) ) , z ( j) -(;; ( j) > ~ ~ ( f (j) , z ( j) -(;; ( j)) ,
j=1 J j=1
vz(j)EZ. . (3.40)
J
Indeed from (3.38) by setting z(j)=~(j)+A(z(j)_~(j)), A((0,1), A~O+ we
obtain (3.40). (Consider that ~~J 0 ([(;;] .~) is positively homogeneous and
the definition of gra~.). Conversely (3.40) implies (3.38) because of
J
the convexity inequality

( 3.41)
Now the regularized problem P is defined:
E
Let p be a mollifier (pEC;(-1,+1), p~ 0 with :s p(~)d~= 1), and let
(3.42)

where pE(E)= (1/E)p(E/E) and* denotes the convolution product.

Problem PE: Find (;;~j)EZj' j=1, ... ,r, such as to satisfy the vari-
ational equality

; <T.(~{j)),z(j)>+ ; < grad<!>.(~{j)),z{j)> +


j=1 J E j=1 J E

+ ~ J(3 ( m) ( [~ J (m) )[ zJ (m) dr.l = ~ ( f {j ) , z ( j ) ) ,


m=1 ~ E E j=1
vz{j )EZ .. (3.43)
J

Further a Galerkin basis of Zj, j=1, ... ,r is introduced. Let Zjn be the
corresponding n-dimensional subspace of Zj. The corresponding finite
dimensional Problem PEn reads:
Hemivariational Inequalities 141

Problem PEn: Find ~~~)EZjn' j=1,2, ... ,r such that

~ <T.(~(j)),z(j)>+ ~ <grad~.(~(j)),z(j)> +
j= 1 J En j= 1 J En

+ ~ Jl3(m)([~
, E En
J (m))[z] (m)dn = ~ (f(j) ,z(j)),
. 1
m= 1r;m J=

Vz(j)EZ., j=1, .. ,r (3.44)


Jn

We make now the simplifying assumption


grad~j(O) = 0 , j=1, ... ,r (3.45)

Proposition 3.1: Let assumptions (3.39a) and (3.45) hold. Then for
f(j)EL 2(aj)' j=1, ... ,r problem P has at least one solution.

Proof: (3.44) is written in the form

<A(tEn),z> = 0, for z = {z(j)}, vz(j)EZjn' j=1,2, . ,r (3.46)

We verify easily that an inequality analogous to (3.39a) can be written


for 13 (now only with inf and sup). As a result of it we may determine
for e~ery m, p~m)>O and p~m)>O such that 13~m)(E)~O if E>P~m) ,13~m)(E)~O
if E<-p~m) and l13~m)(E)I~P~m) if IEI~P~m) and thus

~ 0- p~m)P~m)mesQ~ (3.47)

The monotonicity of grad~j and (3.45) imply that(l)

<gra.(r(j)) -grad~.(O) r(j)_O >~ 0 vr (j) r::z. (3.48)


J l:.En J 'l:.En ' l:.En - Jn

( 1)(3.45) could be omitted. Then we shall use the fact that there is 6
such that grad~.(6 0 ) is finite. Again (3.48) with 6 instead of 0
zero would be u~ed. . 0
142 P .D. Panagiotopoulos

From (3.44) using (3.37),(3.47) and (3.48) we obtain that


r m'
~ I: c 1: ( j ) I 2 + 0- I: p (m) p ( m) me so '
j=1 J. ' 1
En 1
ZJ.
' m= 1 1 2 m

- ~
."
c'.' r (j) '!
J 1ISEn
1
IZ.' c., c'. const .> 0 ' j=1, ... ,r . (3.49)
J=1 J J J

Due to (3.49) we can apply Brouwer's fixed point theorem and thus (3.46)
has a sou ;,En w1 th 11II sEn
l t 10n 9'
0
T (j ) :, <
0
-1 2
1 ~cj' J-, , ... ,r.
Now we may de t erm1ne 0 0

subsequences, again denoted by {1:~~)}, j=1 , ... ,r, such that for E~O,n~

sEn ~ T(j)
T(j) s wea kl y 1. n Zj, J. = 1, ... , r , (3 . 50)
2 2
and because of the compact imbeddings H (oj)CL (oj), j=1 , ... ,r

C~~) ~ 1: ( j) strongly in L2 (0 j) , j =1, ... , r . ( 3. 51 )

Accordingly
c(j)~
En
1:(j) a.e. in QJ. ' j=1, ... ,r . (3.52)

Further we may omit the indices and (m) if no ambiguity occurs. Now we
prove that 13~m)([1:En](m)) is Heakly precompact in L 1 (o~), m=1, ... ,m'.
Equivalently, due to the Dunford-Pettis theorem (e.g. [62] p.239) we
must show that for each ~ m>0 a 6 m(~ m)>0 can be determined such that for

( 3. 53)

The inequality
(3.54)

implies that

f I13E( [1:En]) Ido;;


w
t- J 0 w
II3E( [1:En] )[1;;En] Idn +

(3.55)
Hemivariational Inequalities 143

But

= J I . . . I dQ - J i I dQ + 2f : I dQ :s
I u:En(x)] [>p I [~En(x)] [,;p1 I [~En(x)] l<;;p1

dQ + 2f I .. I dQ
: [~En(x)] i<;;p1

(3.56)

Now let us set z( 1)=z( 1)EZ 1n in (3.44) such that z( 1 )=z( 2 ) on Ql'

z( 2 )=z( 2 )EZzn such that z( 2 )=z( 3 ) on Q2 z(j- 1 )=z(j- 1)EZ(j- 1 )n such

that z(j- 1 )=~(j) on Q~ z(j)=~(j) on Q. z(j+1)=~(j+ 1 ) on Q.


En J -1 ' En J' En J+1 '
z(j+ 2 )=z(j+ 2 )EL. such that z(j+ 2 )=~(j+ 1 ) on Q'. 1 and finally
lJ+2)n En J+
z(r)=z(r)EZrn such that z(r)=z(r- ) on Q~_ Thus, using (3.56) and
1
1
(3.48) we obtain

+2 J 1!3 (m)([~ ](m))[~ ](m)ldQ;;;


I[~En(x)](m)l:sp\m) E En En

:s c+2p(m)P(m)mesQ'- ; <grad<I>.(~(j))-grad<I>.(O) ~(j)_O> :S


1 2 m j= 1 J En J ' En

< 2 (m) (m) ,.,,


= c+ P 1 P2 mes m, vm--1 , 2 , ,m ' ( 3. 57)
144 P.D. Panagiotopoulos

The definition of the mollifier and (3.42) imply easily the estimate
(3.58)

Choosing E0 , such that for all E and n

~ LI(3E([~En])p;:Enlldo'~ ~o(c+2p1p2meso') ~ ~' (3.59)

and 6 such that for meSW<6

(3.60)

we obtain that

f sup 1(3 ([~ J)ldo ~


w u~
En (x)] I~Eo E En

(3.61)

From (3.55),(3.57),(3.59) and (3.61),(3.53) results i.e. the weak pre-


compactness of (3~m)([k:En](m)), m=1,2, ... ,m' in L 1 (o~). Thus, as E-+0 and
n-+~, a subsequence again denoted by {(3(m)([k:
E En ](m))} can be determined
such that

(3.62)

For the present specific plate problem we have that H 2 (o)CC 0 (Q)CL~(g)
(imbeddings). Moreover (3 is a C~-function and thus (3.62) results im-
E
mediately without using the Dunford-Pettis theorem and the estimates
(3.53)7(3.61) based on [65]. However, this procedure is necessary in
most other types of nonmonotone nonlinearity and in any other type of
functional setting. Indeed in any hemivariational inequality formulated
in the H1-space the above procedure is necessary because H 1 (o)~C 0 (Q).
The same holds for any type of nonlinearity term (3() for which the
previous C0 -imbedding does not hold: For instance, let us consider in
Hemivariational Inequalities 145

a plate problem a nonmonotone boundary condition ME~(~~); then


~H 1 / 2 (r) but H 1 1 2 (r)~CA 0 (r). On the contrary, if in a plate problem
on
a boundary condition -QE0(~) is considered, we have the imbeddings
H2(n)cC 0 (Q)CC 0 (r)CL00 (r) and again we can omit the general procedure of
the estimates (3.53)~(3.61).
From (3.44) we find using special variations that

\<gradq>j(~~~)),z(j)>\ ~ c \\z(j) 1\zj , vz(j)EZj' j=1,2, ... ,r, (3.63)


i.e.
\\gradq>.(~(j)) liz'~ c, j=1,2, ... ,r. (3.64)
J En j

Thus, there is a subsequence again denoted by {gradq>.(~(j))}


J En
such that,
a s E.:-+0 ' n-+oo

graj(~~~))-+W(j) weakly in Zj , j=1,2, ... ,r. (3.65)

From (3.44),(3.62) and (3.65)we obtainpassinq tothe limit n-+oo, E:-+0 that

j=1,2, ... ,r, ( 3. 66)

where we have used the fact that

(3.67)

At this point it is worth noting that, if (3.67) did not hold, then we
would choose appropriately the Galerkin basis. For instance for an
analogous problem formulated in a V-space with V~C 0 (Q)CL00 (Q) we would
consider a Galerkin basis in the space VnL00 (Q). Then it is necessary
that VnL00 (Q) be dense in V for the V-norm (cf. Sec.4 and [12] p.272).
To complete the proof we will show that
w(j)= gradq>j(~(j)) in Zj , j=1,2, ... ,r, (3.68)
146 P.D. Panagiotopoulos

and that

(3 .69)

To prove (3.68) first the nonnegative expression

X = ~ <grad<1?.(1:(j))-grad<1?.(&.) r:(j)_&.> :; 0
n j= 1 J En J J ' En J '

(3.70)

is considered, which by means of (3.44) implies that

( 3. 71)

Furthermore as E~O, n~oo

To prove it we form the difference

(3.73)

Since H2(n') is compactly imbedded into L00 (Q' ), we have that limB= 0 due
to (3.62). Also it implies that
(3.74)

and thus
(3.75)
Hemivariational Inequalities 147

which yields limA= 0 . The proof of relations of the type (3.72) is


necessary only in the case of variational hemivariational inequalities
in order to treat the monotone nonlinearity, i.e. to show (3.68). Their
proof relies on the properties of the function space considered. Thus
for instance in a H1-space (H 1 (Q)~C 0 (g)) the proof of a relation (3.72)
is not possible. From (3.71), using (3.72), we obtain for E~O ~~

+ ~ (F(j).~(j))- ~ <ll'(j),8.>- ~ <gra<W.(8.),~(j)_8.> (3.76)


j=1 j=1 J j=1 J J J

(3.66) and (3.76) imply

r (") (") (")


:{<Ill J .~ J -8.>-<grad<l>.(8.),~ J -8.>}~ o, v8.EZ., j=1, ... ,r,
j=1 J J J J J J
(3.77)

which by Minty's well-known monotonicity argument (cf. e.g. [7] p.55)


yields that
m(j) = grad"'.(r(j))
T ""J \,
1"n Z'.,
J
J. 1 2
: ' '''''
r o
(3.78)

To show (3.69) we proceed as follows [66]: (3.52) implies by Egoroff's


theorem, that for everya>O, we can determine wj' j=1, ... ,r with meswj<a.
such that
~~~)~~(j) uniformly in Qj-wj , j=1,2, .. ,r , (3.79)

with ~(j)EL00 (Qj-wj). Due to this uniform convergence, for every a>O a
w with meSW<a. can be found such that for any ~>0 and for E<E 0 <~/2 and
n>n 0 >2/~

I [~En] - [~] I < ~ ' VxEQ' -w (3.80)


Owing to (3.42) we obtain easily the inequality
f3 (!;) = (p *S)(!;) = +EJS(!;-t)p (t)dt: esssup S(!;-t) , (3.81a)
E E -E E It I:E
148 P.D. Panagiotopoulos

and analogously
ess inf 13(!;-t) :S 13E(!;) (3.81b)
JtJ:St=:
From (3.80) and (3.81a,b) we have that

;;; ess sup 13(!;) = ]3 ( [r;J) (3.82)


J[r:J-!;J<~ ~
and
13 ([r;]) :S ess inf 13(!;);;; 13 ([r; ]) , (3.83)
-~ I [r;]-!;J<~ E t=:n
respectively. Due to (3.82),(3.83) we obtain by choosing e~O a.e. in
Q'-w with eEL (Q'-w) that
00

[ ~~([r;])edQ :Sf 13E([r;En])edQ;;; f ]3~([r;])edQ (3.84)


~w Q~ Q~

which as t=:~O and ~oo becomes

(3.85)

Now passing to the limit ~0 we have by Lebesque's theorem

f 13([r:J )e d Q ;;; f xed Q ;;; f ]3([r:J )e d Q , (3.86)


QLw- QLw QLw
and since e~O is arbitrary, (3.86) yields that
A
xEI3 ([ r: J ) a e . i n Q' ~

Now taking a as small as possible, implies (3.69) q.e.d.

b) The "nondifferen tiable" case: In this case graclci>j does not exist
everywhere for some j. Then we regularize the convex superpotentia ls
~j' by assuming that for every j=1,2, .. ,r a sequence of convex Gateaux-
Hemivariationallnequ alities 149

differentiable functionals {~. } exists, depending on the parameter p


JP
with the properties:
i) As p-+0, ~jp(z(j)) .... ~j(z(j)), Vz(j)EZj (3.87)

ii) grad~. (0)= 0, and (3.88)


JP
iii) if z(j) .... z(j)
p
weakly in z.J for p-+0, and if

~- (z(j))< c where cis a constant, then liminf<1). (z(j))~'l>.(z(j)),


JP p ' p-+0 JP P J
j=1,2, ... ,r. (3.89)
First the regularized-discr etized form of the problem is given:
Problem P~pn Find ~~~~EZjn' j=1,2, ... ,r such as to satisfy the vari-
ational equality

+ ~<grad~. (~(j)),z(j)> = ~ (f(j) ,z(j)),


j=1 JP Epn j=1

'v'Z(j)EZ. , j=1,2, .. ,r. (3.90)


Jn
For this case we prove the following results:
Proposition 3.2: Suppose that (3.39) (or (3.39a)), (3.87),(3.88) and
(3.89) hold. Then Problem P has at least one solution.

Proof: Following the same procedure as in Prop. 3.1 Brouwer's fixed


point theorem implies the existence of a solution of P' such that
ll~(j)
Epn
II~ c.,
J
j=1,2, ... ,r. Thus, as E:->0, n->oo Epn

~(j) .... ~(j) weakly in ZJ., j=1,2, ... ,r, (3.91)


E:pn P
and as p-+0
(3.92)
150 P.D. Panagiotopoulos

Now, we prove that (3.53) holds for {[3(m)([!;: ](m))} (cf. Prop. 3.1)
E Epn
and thus,

[3~m) ([r;Epn] (m) )-+ x~m) weakly in L1 (Q~), m=1, ... ,m' , (3.93)
and

x~m)-+X(m) weakly in L 1 (r.J~), m=1, ... ,m'. (3.94)


Moreover we get the estimate

(3.95)

where the constants do not depend on E,n,p. Therefore we have for E-+0,

gradcl>. (!;::(j))-+W(j) weakly in Z'. , j=1,2, ... ,r , ( 3. 96)


JP Epn p J

and for p-+0


1~ j)-+ 1 ( j) weak 1y in Z j, j =1 , 2, ... , r . (3.97)
From (3.90) we obtain for E-+0, n-+=

j=1,2, ... ,r . (3.98)

The relations

w0(j) -- gra dct>jp (!:(j))


P . Z'j'
1n -1 , 2 , . ,r,
J- (3.99)

are proved as in Prop. 3.1. From (3.98) and (3.99) and the convexity
inequalities
ct>j 0 (E(j))-ct>jp(z(j)) ~ <9radct>j 0 (z(j)),E(j)_z(j)>, VE(j)EZj we get

~ <T.(!:(j)),z(j)_!:(j)> + ~ {ci>. (z(j))-ct>. (!:(j))} +


j=1 J p p j=1 JP JP p
Hemivariational Inequalities 151

vz (j) Ezj , "-1 , 2 , ... , r .


J- (3 . 100)

We let z(j) in (3.100) be such that <I>.(z(j))<=, j=1,2, ... ,r. Then, due
to (3.87) there exist constants c.
J
su~h that <I>. (z(j))< c., j=1,2, ... ,r
JP J
and from (3. 100) it results that

"" (r(j))~_
"' s c'. , 1, 2 , ... ,r
J= (3.101)
JP p J

But (3. 101) and (3.91) imply (3.89) for each <I>jp' Now let us write
(3.100) as

~<I>. (z(j))+ ~ <T.(f: (j)),z(j)>+ ~' Jx(m)[z](m)dQ;;;


j=1 JP j=1 J 0 m=1 ~P

(3.102)

For o-+0 we obtain

uz(j)EZ.,
v J . 1 2
J=,, ,r (3.103)
152 P.D. Panagiotopoulos

It can be easily verified that due to the functional framework of the


problem

( 3. 104)

and that

Relations (3.103),(3.104),(3.105),(3.87) and (3.89) imply that

vz(j)EZj , j=1,2, ... ,r . (3.106)

The proof is completed by proving that

(3.107)

It can be shown exactly as in the differentiable case (relations (3.79)~


(3.86)) q.e.d.
Until now we have shown the weak convergence of the approximate solution
in the H2-space and due to the compact imbedding H2(n)cC 0 (o) the strong
convergence in the C0 -space. From the complete continuity of operator G
defined by (3.32) we deduce the strong convergence of the planar stresses
in L2 Further we give without proof (see e.g. [63]) a proposition which
under some additional assumptions guarantees the strong convergence of
the solution ' in the H2-space.
Proposition 3.3: Let the assumptions of Prop. 3.2 hold. Moreover,
suppose that p,q, q different for each m and j exist and constants c
such that
Hemivariational Inequalities 153

IB(m)(E)I ;; c(1+IEIP), VEElR , m=1, ... ,m' (3.108)


'
lb/E) I ;; c(1+IEI) VEElR , (3.109)
Ibjp(!;) I ;; c(1+IEI) VEElR , j=1, ... ,r, (3.110)
and
Ibj(E) I ;; c(1+1Eiq), VEElR , (3.111)
~

lbjp(E)I ;; c(1+1Eiq), VEElR , j=1, ... ,r. (3.112)

Then, for E->0, n->oo,


c(j tc (j) strongly in z.cH 2(Q.), j=1,2, ... ,r. (3.113)
En J J

In the next Section we deal with a semicoercive hemivariational in-


equality.

4. A SEMICOERCIVE HEMIVARIATIONAL INEQUALITY


The method developed in this Section for the study of hemivariational
inequalities is general and applies to a general class of hemivari-
ational inequalities. Let us consider for u,vEV where V is a real
Hilbert space, a symmetric continuous bilinear form a(.,.):VXV->lR.
Let V' be the dual space of V and assume that for Q open and bounded
subset of lRn
VCL 2(Q)CV I ( 4. 1)
where the injections are continuous. We denote by (.,.) the L2-product
and duality pairing, the norm of V by II II and the norm of L2(g) by
11 . We recall here that the linear form(.,.) extends uniquely from
VXL~(g) to vxv [67]. Further let us assume that
VcL 2(Q) is compact (4.2)
and that
VnL00 (Q) is dense in V for the 1111-norm. (4.3)
The bilinear form is assumed to have a nonzero kernel, i.e.
kera(u,u)= {qla(q,q)= 0} f {0}, and let
154 P.D. Panagiotopoulos

kera be finite dimensional . ( 4.4)


The norm JJvJJ on Vis assumed to be equivalent to lllvlll = p(v)+JqJ 2,
where v=v+q, qEkera, vEkera.l (i.e. (v,q)=O vqEkera), p(v) is a seminorm
on V such that p(v)=p(v+q) V'VEV, qEkera and let
a(v,v) ~ c(p(v)) 2 V'vEV c const> 0 . (4.5)
Now the following problem is formulated for fEL 2(n).
Problem P: Find uEV such that
a(u,v-u)+Jj 0 (u,v-u)dn~ (f,v-u) wr:.V . (4.6)
n'
Here Q'CQ and j results from @EL~0 c(IR) as in (3.9)-;-(3.13). Thus j is a
locally Lipschitz function such that
- 1\
aj(E) = ~(~) (4.7)
1\
where ~ is the multivalued function which results from~ by "filling in"
the discontinuities. We denote further by q and q the positive and
the negative parts of q respectively, i.e. q:=q+1qi and q = JqJ-q
The following proposition gives a necessary condition for the existence
of the solution. Further the notation
~(-co) limsup ~(E) and ~(co) = liminf ~(E) (4.8)
E-+ co
is used.
Proposition 4.1: Let
(4.9)
Then a necessary condition for the existence of a solution uEV of
Problem P is the inequality
J[~(~)q+-~(co)q_]dn ~ (f,q) ~ J[~(co)q+-~(~)q_]dn VqEkera. (4.10)
~ d
If (4.9) holds as a strict inequality (with < instead of~) the same is
the case for the necessary condition (4.10).
Hemivariational Inequalities 155

Proof: Let us set in (4.6) v-u= qEkera, qt 0. We obtain


Jj 0 (u,q)do;;; (f,q) VqEkera, qt 0. (4.11)
Q'

Then (4. 11) is written as


Jj 0 (u,q)do;; (f,q);; -Jj 0 (u,q)do VqEkera, qt 0 (4.12)
Q' Q'

because ~j 0 (u,q) is positively homogeneous. From the definition of j


and j 0 we obtain by means of (4.9) that
Jj 0 (u,q)do ~ J[6(=)q+-6(~)q_]dr.l VqEkera, qt 0 (4.13)
Q' Q'

and analogously for Jj 0 (u,-q)dQ. Thus (4. 10) is shown. The rest of the
Q'
proposition is obvious q.e.d. Further we give a sufficient condition.
Proposition 4.2: Let
6(-=)<6(=). (4.14)
Then if
J[6(-=)q -6(=)q ]dO< (f,q)< JW(=)q -6(-=)q ]do 'v'qEkera,q tO,
Q' + - Q' + -
(4.15)
problem P has a solution.
Proof: The proof has some parts similar to the proof of Prop. 3.1. We
point out here the new arguments. First a Galerkin basis of VnL=(o) is
introduced. This is possible due to (4.3). Let Vn be then-dimensional
subspace of VnL=(o). We regularize Problem P through the mollifier p .
E
Then the corresponding fin~te dimensional problem PEn is formulated.
Problem PEn Find uEn EV n such that
a ( u n, v) + f6 ( u ) vdQ = ( f, v) VVEVn . (4.16)
E Q' E En

Now the finite dimensional variational equality (4.16) is written in the


form
(4.17)
Because of (4.14) the estimate (3.47) is obtained fore. From (4.17),
E
156 P.D. Panagiotopoulos

this estimate, and (4.5) we find that

(A(u n),u );;; c[p(u n)] 2 - c: llu I I- o 1o2mesn' c const> 0. (4.18)


E En E En

Now we will apply Brouwer's theorem to show that (4. 16) has at least one
solution uEn and that { iluEn [[}is bounded. According to this theorem
we have to prove that r> 0 exists such that

(4.19)

Here it will be shown that a number M> 0 can be determined such that

(4.20)

Then we may take lluEnll= r>M. For the proof of (4.20) it is suf-
ficient to prove that

(4.21)

Due to (4.18) we deduce that, if (A(uEn),uEn);;;O, then a constant c>O


exists such that

(4.22)

where u =
En
u +q Accordingly it is sufficient to prove that
En En
(A(uEn),uEn);;; 0 implies JqEn[ 2 ;;; c, or equivalently, that a number R> 0
can be determined such that

(4.23)

This last relation will be shown now. From the definition of 8 we prove
E
that
+E
8 (ex>) = 1 im8 (f;) = 1im J8(f;-t)p ( t)dt~ 1 imess inf8(x) ;;;
E ~ E ~" E f;-+c4x-(;[;;;E

;;; 1 imess inf 8(x) = 1 im inf 8(x) = 8(ex>) (4.24)


f;~-E;;;X~ X-+ex>

Similarly we show that 8 (-oo);;; 8(-oo). Thus (4.15) implies that


E
Hemivariational Inequalities 157

f[8 (-a:>)q -8 (co)q ]dQ< (f,q)< f[;3 (co)q -8 (-oo)q ]dQ ,


Q'E +E- Q'E +E-
'v'qkera q f; 0 . (4.25)
Noting that from (4. 14) the inequality (3.39) results as a strict in-
equality, a numberM>o 1 (cf. (3.47)) can be chosen such that for any
function uEV with ju(x) I>M and sign u(x)= sign q(x) for almost every
xEQ', we have from (4.25) that for q>O, q=q+, u(x)>M and thus
f 8 (U (X) ) q (X) dQ- ( f, q) > 0 , (4.26)
{xlq(x)>O} E
-J 8 (-u(x))q(x)cil+ (f,q)> o. (4.27)
{xjq(x)>O} E
For q<O we have q=-q_, u(x)<-M and thus we obtain from (4.25)
J 8 (u(x))q(x)ctn- (f,q)> o, (4.28)
{xjq(x)<O} E
-f 8 ( -U (X ) ) q (X) dQ + ( f, q) > 0 (4.29)
{xlq(x)<O} E
By an appropriate choice of the numbers 6E(0,1], N>1, n>O, and a>O, and
by taking into account that 8E(u(x))u(x)~O and that sign u(x)= sign q(x)
these inequalities imply for every u as above, the relations
( 1 - ir) I
{x jq(x)j>fu}
J 8e: (u ( X) ) q( X) dQ - ( f. q) > n Iq I2 (4.30)

-(1-ir) I J 8~-u(x))q(x)dQ+(f,q)>nlql 2 (4.31)


{x jq(x)l>fu}
as is obvious by taking ~0 +. We write
_1 now_ uEn =uEn +q En and for N as
1 1
in (4.30),(4.31) and for a.>a.0 = M6 (1- N) we have

lq En (x)j>fu
158 P.D. Panagiotopoulos

;; f 1\:(uc:n)uc:ndr- o 1o 2mesQ' ;;;


r'iJc:n (xllf-n
1 N
I qc:n(x) I >fu

;;; ( 1 - _l)
N
f qc:n13c: ( uc:n ) dQ - p p me sQ '
1 2
(4.32)
00
luc:n (x) I<N

I qc:n ( x) I >fu

Indeed for luc:n(x)l<fu/N and lqc:n(x)l>fu we have that for a.>a


0

I uc:n I = u
I c:n + qc:n I > ( 1 - l loo> f.f
~~
(4.33)

and thus 13c:(uc:n)uc:n;;; 0, and 13c:(uc:n)qc:n;;; 0. Further it can be verified


that for q > fu
c:n
~ fu 1
uc:n(x)= uc:n(x)+qc:n(x)>-N+ qc:n(x)> qc:n(x)(1-N) (4.34)

and therefore
(4.35)

Similarly for q <-fu. Now we get the inequality


En
(A(uEn) ,uEn) ;;; ( 1 - ~) ~ f 00 qEni3E(uEn)dQ -
IuEn (x) 1<-N
lq (x)l>fu
En

- o 1o 2mesn'- (f,q
En
)- (f,u ) (4.36)
En

Using (4.30), we obtain for a>a 0 sufficiently large from (4.36), that

(A(uc:n),uc:n) > nlqc:n1 2-o 1o2mesn'- (f,uc:n) ;;;

( 4.37)

From (4.37) we obtain assuming that a>a0 and that (4.22) holds the final
Hemivariational Inequalities 159

est i rna te
(A(uEn),uEn)>n[qEnl 2 -c[qEnl~/ 2 -c', c,c' const>O. (4.38)

The right-hand side of (4.38) is positive if we choose R>O such that


[q En [2>R>6ao(mesn) 112 . Thus (4.23) has been proved. Thus by Brouwer's
fixed point theorem, problem PEn has a solution uEn with [[u 8 nll<c. The
rest of the proof is the same as the proof of prop. 3.1: From the
boundedness of uEn we deduce the weak convergence in V and the strong
convergence in L2(n) because of (4.2), and then using Dunford-Pettis
theorem we show as in (3.53)~(3.61) the weak convergence of 0E(uEn) to
x in L1(n'). Because of (4.3) we pass to the 1imit 11-+"" E-+0 in
(4. 16) and then by the same procedure as the one indicated by (3.79) ~
1\
(3.86) we show that xE0(u) a.e. inn'. q.e.d.
The sufficient condition is similar to the sufficient conditions de-
rived in the Landesrnan-Lazer theory [68] ,[69]. Concerning the number
of solutions both for the coercive and the semicoercive problem most
questions are still open. There are some results along the lines of
the theory of semilinear differential equations and more precisely in
the area of the contemporary development of the Landesman-Laser theory,
which might be e~tended or directly applied to the theory or hemivari-
ational inequalities. However, the problem of the multiplicity of so-
lutions should be considered as still open.

5. APPLICATION OF QUASIDIFFERENTIABILITY TO MECHANICS(l)


5.1 Generalities and Mathematical Preliminaries
The aim of the present Section is to introduce the notion of quasi-
differentiability into Mechanics. Until now we have used nonsmooth
energy functions, which, through the subdifferential in the convex case
and the generalized gradient in the nonconvex case, give rise to vari-
ational or to hemivariational inequalities respectively. Any research

( 1)The author is indebted to Professor V.F. Demyanov for introducing


him to the mathematical theory of quasidifferentiability.
This Section is based on an additional Appendix in the Russian
translation of the author's book [12].
160 P.D. Panagiotopoulos

effort on "nonsmooth mechanics" is closely connected with the careful ex-


amination of the generalizations of the classical derivative, especially
concerning the possibility of formulating and studying through them me-
chanical problems involving nonsmooth energy functions. But this is not
the only reason why the notion of quasidifferentiability is important
for the study of nonsmooth mechanical problems. Let us examine this
point more carefully: The subdifferential of convex analysis seems to
be a well-defined notion from the standpoint of mechanics; no logical
discrepancy appears. The same does not hold for the nonconvex case.
Indeed the comparison of the generalized gradient with the derivate con-
tainer [70] as well as with the quasidifferentiability [71]~[73] has
shown that the generalized gradient "does not contain all possible infor-
mation on the directional behaviour" of a function at a point. This is
caused mainly by the fact that the definition of the generalized gradient
af is based on the notions of the directional differentials ft(.,.) and
f 0 (.,.) which are roughly speaking "approximations" of the directional
(Gateaux) differential t'(.,.). On the other hand the "cumulative be-
haviour of a function" described by the generalized gradient of
F.H. Clarke enables us to formulate in Mechanics a wide class of vari-
ational formulations in inequality form, the hemivariational inequali-
ties, whenever a mechanical law is expressed in terms of generalized
gradients. The generalized gradient af leads to hemivariational in-
equalities because of its definition which is based on the definitions
of ft(.,.) and/or f 0 (.,.). These definitions have as a main task to
achieve for nonconvex nondifferentiable functionals the validity of a
relation analogous to the well-known relation
f' (x,h) = max{<x' ,h>lx'EClf(x)} VhEX ( 5. 1 )

holding for convex functionals. Indeed, it was proved that a similar


relation holds for the directional differential for a maximum type
function (see e.g. [73] ). Therefore, whenever af(x)t 0 we have for
every yEX the validity of
ft(x,y)= max{<y,x'>lx'Eaf(x)} , (5.2)
Hemivariational Inequalities 161

if ft(x,y) is finite for every y. The quasidifferential calculus is


appropriately developed with the task to describe the "directional
properties" of a function at a point accurately. It leads - in contrast
to the calculus of the generalized gradient- to equalities and not to
inclusions. Therefore the class of quasidifferentiable functions
is a linear space closed with respect to all algebraic operations. All
these properties of the quasidifferentiability make this notion very
attractive in the theory of optimization. But this notion is equally
important in Mechanics for two main reasons: Suppose that in the frame-
work of a mechanical theory a minimum problem is formulated involving
non-necessarily differentiable convex or nonconvex functions. For
instance, consider the problems of minimum potential energy for a
Hencky-body (or an elastic ideally locking body) in unilateral contact
with a rigid support, or also a minimum weight problem. In this case
it is important, from the numerical point of view, to accurately de-
termine for the function to be minimized the descent directions at given
points. This fact reveals the possible advantages of the quasidiffer-
ential (cf. [72] p.11). The resulting necessary and sufficient con-
ditions from the theory of quasidifferentiability, when interpreted in
terms of mechanical notions, give a more clear description of the be-
haviour of the mechanical system. Another, more profound reason for the
introduction of the quasidifferentiability into Mechanics is the follow-
ing: When we study a mechanical system we try actually to find relations
between the directional variations of certain orders, of the quantities
involved in the problem. This general idea already appears in the liter-
ature of mechanics in a hidden form concerning differentiable processes
(cf. e.g. the zero, first, second gradient mechanical theories [86]}.
Before entering into the main part of this Section we would like to point
out that especially in the applied mechanics it is believed that always
a nonsmooth problem can be approxi'mated by smooth problems. This ap-
proach which is very useful in proving existence and general approxi-
mation results may lead to erroneous results in the case of a purely
numerical procedure because many properties of the original nonsmooth
162 P.D. Panagiotopoulos

problem can be lost during the smoothening operation. To this context


we refer the reader for a very illustrative example to [74] p.7.
Let us define first the notion of quasidifferential introduced in 1979
by V.F. Demyanov, L.N. Polyakova and A.M. Bubinov [76] ,[77]. Let f:~ffi
where A is an open subset of ffin. The functional f is said to be quasi-
differentiable at xEA if it is directionally differentiable at x (i.e.
~ n
f'(x,h) exists for every hEffi (cf. (2.7)) and if two convex compact
sets ~' f( x )C ffin and a f ( x)Cffin can be determined such that
f' (x,h)= max{<x 1 ,h>lx 1 E~'f(x)}+ min{<x 2,h>lx 2Ea'f(x)}
x1 x2
'v'hEffi n . ( 5. 3 )
(5.3) is equivalently written in the form
f' (x,h) =min max {<x 1+x 2 ,h> lx 1q'f(x) ,x{a'f(x)}. (5.3a)
x2 x1
Here<.,.> denotes the ffin-inner product. The pair of sets
Df(x) = {~' f(x) ,a'f(x)} (5.4)
is called a quasidifferential of f at x. We call the sets ~'f(x) and
3'f(x) 3'-subdifferential and 3'-superdifferential off at X respective-
ly and we use the prime in order to indicate that no confusion should be
made with the subdifferential af(x) of convex analysis(l). From (5.3)
and (5.4) it results that Df(x) is not uniquely determined at x. This
is not considered a disadvantage. Indeed each form of Df(x) describes
equally "well" and "reliably" the directional properties off at x. The
above definition can be generalized by replacing mn by a locally convex
Hausdorff topological vector space X. Then ~'f(x) and a'(x) are E(X' ,X)-
compact subsets of the dual space X' and<.,.> denotes the duality pair-
ing between X and X'. For further generalizations see [71].

(l)In references [71~77] the prime is omitted as well as in 3'-sub-


differential and 3'-superdifferential the symbol a.
Hemivariational Inequalities 163

If f is continuously differentiable on A then f is quasidifferentiable


at every xE A and
Df(x)= {gradf(x),O} or Df(x)= {O,gradf(x)} . (5.5)
Thus f is both a'-subdifferentiable and a'-superdifferentiable at x.
If f is convex defined on a convex set A then (5.3) is written in the
form (5.1). Accordingly
Df(x) = {af(x) ,0} ( 5.6)
Similarly, if f is concave and A is convex then f 1 = -f is convex and we
obtain that
Df(x)= {O,af(x)} , (5. 7)

where

is the superdifferential of the concave function fat x [61]. Analo-


gously we may obtain for a a-regular function (see (2.7)) such that
ft(x,h)t co for every h, that
t
(5.9)
~
f'(x,h)= f (x,h)= max{<x 1,h>lx 1E3f(x)}
and thus
Df(x) = {af(x) ,0}, (5.10)
where af(x) is the generalized gradient. For instance, iff is a maxi-
mum type function, e.g. f= max{ 1, ... , m}, where.= .(x) i=1, ... ,m,
n 1 _ 1
xE ffi , are smooth functions, then we know that f is a-regular and thus
Df(x)= {co{gradi(x), ... ,gradk(x)},O} ( 5. 11 )

if xE{xli= f, ... ,k= f}, where co denotes the convex hull. This
formula can be shown independently of the theory of generalized gradient
[73]. Now we consider pairs of sets ri= {Ai,Bi}' i=1,2 , where AiCX'
and BiCX' and we define their addition by
(5.12)

and the multiplication by AE ~as


164 P.D. Panagiotopoulos

if ),.2; 0
( 5. 13)
if A.< 0

Using these definitions we can show that, if fi' i=1, ... ,n, are quasi-
differentiable functions at x then g= ~a.f. is also quasidifferentiable
1 1 1
at x(a 1.ElR ) and Dg =~a. Of .. Moreover it can be shown that if f 1 and f 2
1 1 1
are quasidifferentiable at x then g= f 1f 2 has the same property and

(5.14)

moreover for f 2(x)f 0, g= f 1;f 2 is quasidifferentiable at x and

Dg(x)= [f 2(x)Df 1(x)- f 1 (x)Df 2 (x)]/f~(x) . ( 5. 15)

The following properties are of importance: With i' i=1, ... ,n,
f= m~x{i} is quasidifferentiable at x and for l(x)= {ili(x)= f(x)}

a'f(x)= co{3'k(x)- I: 3'.(x)lk~I(x)} (5.16)


- - iEI(x) 1
ifk
3'f(x)= E 3' (x). (5.17)
k'E I (x) k

Analogous formula holds for a minimum type function f. In this case


3'f(x) (resp . ..'f(x)) is given by (5.16) (resp. (5.17)) by replacing in
the right hand side._' by 3' and 3' by._'. Iff= f 1+f 2 where f 1 (resp.
f 2) is convex (resp. concave) and f 1(x), f 2(x) are finite then f is
quasidifferentiable and we have that [78]
(5.18)

Due to (5.3a) for every x2E3'f(x) (resp. x1E..'f(x)) the function


max{<q 1 ,>iq 1 Ex 2 +~'f(x)}= rx 2 () (resp. min{<q 2,>jq 2Ex 1+3'f(x)}= rx 1())

is an upper convex (resp. a lower concave) approximation of f at x and


the set of functions {rx 2()1x 2E3'f(x)} (resp. {rx 1()1x 1E..'f(x)}) re-
presents an exhaustive family of upper convex (resp. lower concave) ap-
proximations off at x [74]. It can be easily shown [73] that iff is
Hemivariational Inequalities 165

quasidifferentiable on X a necessary condition for the existence of a


solution of the problem
f(x 0 ) = min{f(x) lxEX} (5.19)
resp.
f(x0 ) = max{f(x) lxEX} (5.20)
is
-a'f(x 0 )Cl'f(x 0 ) (5.21)
resp.
-l'f(x0 )ca'f(x0 ) (5.22)

A point x0 (resp. x) satisfying (5.21) (resp. (5.22)) is called an


inf-stationary (resp. a sup-stationary) point off on X. It is also
verified that if x0 does not satisfy (5.21) and X= lRn then the di-
rection

E=-
v +w I
vo+wo , maxmin{llv+wll wEa'f(x 0 ),vEl'f(x 0 )}= llv 0 +w0 ll
II o o II w v
( 5. 23)
is a direction of steepest descent (not unique) of f at x0 , and if x0
does not satisfy (5.22) then the direction

, maxmin {II v+wiiiWEa'f(x 0 ),vEl'f(x 0 )}= IIV'/w0 II


v w
(5.24)
is a direction of steepest ascent (not unique) of f at x0

5.2 Quasidifferentiability in Mechanics


With respect to a mechanical system E characterized by the triplet
E= {U,<U,f>, F} any ~rechanical law of the form (2.7), where \tl is a-regu-
lar can be written in the quasidifferential form
{ -f ,O}ED\tl(u) = (altl(u) ,0} (5.25)
This holds for all theories based. on nonconvex superpotentials which are
a-regular. This remark enables us to apply the results of quasidiffer-
ential calculus for the case of a-regularity, and therefore to obtain
sharper "optimality" conditions than with the calculus of generalized
gradients [79].
166 P.D. Panagiotopoulos

With respect to the same mechanical system ~~e may introduce a more
general law of the form
{-r,-f}E~(u)= {~'<ll(u),a'<ll(u)} (5.26)
where f,fEF and <ll is a quasidifferentiable function. The greater possi-
bilities offered by (5.26) in comparison with (5.25) are obvious. To
explain better (5.26) let us assume, for the sake of simplicity, that
U=lRn and F=lRn. For a bounded subset ACF we define the support function
sA(u) = sup{<u,f>lfEA} . (5.27)

Every support function is sublinear (i.e. it is subadditive:


sA(u 1 +u 2 )~ sA(u 1)+sA(u 2 ), and positively homogeneous), and every sub-
linear function h is conversely the support function of a convex compact
set which is uniquely determined by the formula
A = {fi<u,f> ~ h(u) VuEU} . (5.28)
Accordingly (5.26) can be put by means ofthe definition(5.3) in the form
(5.29)
Obviously
~'<ll(u) = {fl<f,v> ~ s~'<ll(u)(v) VvEU} (5.30)

and analogous inequality is valid also for a'<ll(u). Thus in (5.29) we


have a difference of sublinear functions. Let us denote by Q the class
of all convex, compact subsets of F. According to the previous results
with every function~ which is the difference of two sublinear functions
we may associate AEQ and BEQ such that
~(u) = max{<u,f>lfEA}+ min{<u,f>lfEB} = sA(u)-s 8 (u) . (5.31)

On QxQ we introduce an equivalence relation Jt is introduced such that


{A 1 ,B 1 }~{A 2 ,B 2 } if and only if A1-s 2 = A2-B 1. Further let Qbe the
quotient space (QxQ)/Jt and let us denote by [A,B] the equivalence class
with representant {A,B}. On Q a vector space structure is imposed by
~

means of (5.12),(5.13). Moreover the norm induced by the Hausdorff


Hemivariational Inequalities 167

distance d(A,-B) turns Q into a normed-space [80]. The vector space e


of functions which are the difference of sublinear functions is a normed
space for the sup-norm jjjcpjjj = sup{jcp(u)JiuEU,jjujj=1} and it has been
proved [76] that the correspondence [A,B]->sA(.)-s_ 8(.) is an isometry.
The above explain the meaning of (5.26). Several mechanical laws of the
same form may be added and we shall have for i=1 , ... ,n

1' .
{-L;T . , -L;
.1.1.1 1
a
-.1
a L;<D 1 ( u) }
}n: D<D . ( u ) = DL;<D ( u) ={ I L;<D ( u ) , I (5.32)
1 1 1 1 1 1

If some of the <lli S in (5.32) are maximum or minimum type functions, as


1

it happens in the case of generalized locking phenomena, then (5.16) and


(5. 17) permit to write the mechanical law in an interesting multiplier
form (due to the appearance of the convex hull). If in (5.32) L;<IJ.(u)
. 1
=
1
= cp 1 (u)~ 2 (u), where <P 1 (resp. cp2) is convex (resp. concave), then (5.18)
implies for cp 1(u), cp2(u) finite that

<P 1(v)-cp 1(u) ~ <-~fi,V-U> VVEU , (5.33)


1

(5.34)

If the position of equilibrium is characterized by ~(f 1 +f 2 )= 0 then the


1
solution with respect to u of the system of this last equation together
with the variational inequalities (5.33,34) supplies the possible po-
sitions of equilibrium. Analogously we operate in the general case
(5.32) if the position of equilibrium is characterized by a relation
(or relations) of the general form ~(airi+bifi)= 0, with ai,biE ffi.
Finally we can consider mechanical laws of the form
(5.35)
Now let us consider a deformable body ncm3 with a fixed boundary r.
The principle of virtual work implies (u.=O
1
on r i=1,2,3)

fa ..(u)E: 1J.. (v-u)ctn-Jf.(v.-u.)dQ=


g 1J g 1 1 1
0 WEU d ,
a .
(5.36)
168 P.D. Panagiotopoulos

where Uad is the kinematically admissible set of the displacements


(small strain theory). We assume material laws of the form

a .. E:~. = w'(E,E*) VE:*= {E:.~}clR 6 , (5.37)


1J 1J 1J
resp.
o .. E:~.;Sw'(E:,~::*) VE:*={E:.~}flR 6 (5.38)
1J 1J 1J
We are led to the following problem for f;fL 2(n).
Problem 1: Find uE[H 1(n)J 3 such as to satisfy the variational equality
(resp. the hemivariational inequality)

f(w'(E:(u),E:(v-u))-f.(v.-u.))dn= (resp.~) 0 VVE[H 1(n)J 3 . (5.39)


Q 1 1 1
(resp. (5.40))
The material law (5.37) describes hyperelastic materials. In the more
general law (5.38) the inequality means the appearance of certain dissi-
pation mechanisms. Suppose now that w is quasidifferentiable, i.e. we
can determine convex and compact sets A= ~'w(~::) and B= a'w(~::) such that

.. ~::.~11:=
w'(E,E*)= max{l: lJ {'r .. }EA}+ min{~ .. ~::.~1
lJ lJ 1J 1J

(5.41)

Using the notation l:ijEij= (1:,~::*) etc, we obtain the relation

w' (E,E:*) ~ (l:,E:*) + min{(~,E:*) I~Ea'w(~::)} vl:E~'W(E:), VE*EJR 6 (5.42)


and further
3TEa' w( E) such that w' (E: ,E:*) ~ (1: + ~ ,E:*) VE*E:JR6
and vl:E~'w(~::) . (5.43)
Therefore (5.39) has the following equivalent formulations.
Problem 2: Find uE[H 1(n)J 3 such as to satisfy
Hemivariational Inequalities 169

that Jf.(v.-u.)dQ ~ J(i+~,E(v-u))dQ


Q 1 1 1 Q

VTEl'W(E(u)) and VvE[H 1(Q)] 3 , (5.44)


or equivalently,
Find uE[H 1(Q)] 3 such as to satisfy the relation
0
Prob 1em 2':

f[minmax (i+~,E(v))- f.v.]dQ= 0


Q i-:Bl:'EA 1 1

VvE[H 1(Q)J 3 (5.45)


In the case of (5.40) the equality to zero in (5.45) is repl~ced by ~0.

Several materials obey the foregoing material law. For instance materi-
als having as a strain energy density w a finite function which is the
sum of a co~vex and a concave finite part i.e. w=w 1+w 2. Then ~w(E)=aw1 (E)
and a'w(E) =aw2 (E). Analogous is the case of a strain energy which with
respect to certain components E1 of the strain tensor is convex and with
respect to the remaining components ~ is concave. Let for instance
E-+W(E) = w(1 ,E 2) be finite on E{E 2ClR where E1ClRa, E2cm13 ,,a+t3= 6, are
open and convex. Then w(.,E 2) (resp. w(E 1,.)) is co~vex (resp. concave)
and the sets aE 1w(.,E 2) -the subdifferential -and aE 2w(E 1,.) -the
superdifferential - are nonempty, compact, convex sets. It can be
proved [ 73] p.232) that E-+W(E 1,E 2) is quasidifferentiable on E,xE 2,
and that at E= (E 1 ,E 2)

~'W(E 1 ,E 2 )= {aE W(E 1,E 2),0}ElRaxlRt3 and


1 (5.46)
a'w(E 1,E 2)= {0, aE 2W(E 1 ,E 2)}EJRax 1Rt3

Note that analogous formulations are obtained if we have an elastic body


and if the boundary condition
( 5 .47)

holds where j is quasidifferentiable.


170 P.D. Panagiotopoulos

To give a more concrete example we consider a "total" contact law of the


form
( 5. 48)

where uN~j(uT,uN) is concave, ore more generally d-regular, and finite


and uT~j(uT;uN) is convex and finite. For instance, we can take as
uT~j(uT,uN)= clurl -the classical friction potential -and as
uN~j(uT,uN)any a-regular function, e.g. the function resulting from the
delaminations law. We can show easily by modifying slightly the proof
of [73] p.232, that (uT,uN)~j(uTuN) is ~uasidifferentiable and that con-
ditions analogous to (5.46) hold, with a replaced by -a.

The "weakness" of the generalized gradient results roughly speaking from


the lack of distinction of the convex and the concave "parts" of the
function w under consideration. The influence of the concave (resp.
convex) "part" on the directional variations of'.v is described more accu-
rately by the quasidifferential calculus. For all the minimum problems
characterizing positions of equilibrium and containing nonconvex energy
functionals a common characteristic is that every local minimum corre-
sponds to a position of equilibrium but not conversely. Indeed in the
context of generalized gradients every local minimum is a substationarity
point and therefore corresponds to an equilibrium position. Analogous
is the situation in the large displacement elasticity [81],[82] ,[83].
The resulting minimum problems are generally nonconvex and nondiffer-
entiable in the presence of unilateral constraints. The local minima
can be checked by the sharp criteria developed by V.F. Demyanov (see e.g.
[84] ). We shall not elaborate on this idea here and we leave it as an
exercise for the reader. We note simply that the resulting interesting
expressions of the necessary and sufficient conditions reveal once
more the new possibilities offered by the quasidifferentiability. It is
true [77] that every closed cone C is associated with a function ~which
is the difference of two sublinear functions such that C={x8Rnl~(x)~ 0}.
Hemivariational Inequalities 171

According to (5.31) we may write that


C= {xE1Rn i(p(x)= sA(x)- s_ 8 (x)~ 0}. (5.49)

Then the set of all ~'s satisfying (5.49) is a closed convex cone n[C]
in the space e and the set of the corresponding elements [A,B] of Q
~

forms a convex closed cone N[C] in the space Q. Now let us consider a
n
closed set TC1R and x0 ET. Then a vector y is said to be tangent toT
at x0 if a sequence {xn}ET and a sequence {tn},tn>O, with tn~o+ can be
determined such that lim(xn-x 0 )/t= y as ~oo. All the tangent vectors
to T at x0 ~orm the tangent cone ~eT(x 0 ) which is closed. Suppose now
that [A,B]EQ. ~~e say that [A,B] is quasinormal toT at x0 if for C a
closed cone such that CC~T(x 0 ) we have that [A,B]EN[C]. Further let us
denote by nT(x 0 ) the set of all quasinormals toT at x0 . Suppose
further that T= {x[p(x) SO} where pis a quasidifferentiable function
and let x0 be such that p(x 0 )= 0. Let us denote by r(.) the function
(sa'p(x ) - s-a'p(x ))() and let the nondegeneracy assumption
- 0 0

{x[r(x)<O}= {x[r(x)SO} hold. Then

Dp(x 0 ) = {a'p(x ),a'p(x 0 )} (5.50)


- 0

is a quasinormal corresponding to zT(x 0 ). After this mathematical


introduction we define plasticity laws of the form

(5.51)

This general law includes as a special case the nonassociated plasticity


laws for convex yield surfaces.

We shall close this Section with the following interesting observation.


Rubinov and Yagubov have shown in [85], that, if f'(x:h) is continuous
in h then a pair of star-shaped sets {U,V}.exists such that

f'(x,h) = max{A.<O[hE-A.V}+ min{A.>O[hEA.U} (5.52)

We recall here that a closed set UC1Rn is called star-shaped if OEintU


172 P.D. Panagiotopoulos

and every ray {A.x(A.;;: 0, x~U} does not intersect in more than one point
the boundary of U. As it is known [20] ,[87] star-shaped surfaces
constitute a very general type of yield surfaces in the theory of
plasticity. The expected connection of (5.52) and the calculus de-
veloped in [85] with the theory of plasticity and the limit load analy-
sis is still an open problem.

REFERENCES:
[1] Fichera, G.: Problemi elastostatici con vincoli unilaterali: il
problema di Signorini con ambigue condizioni al contorno,
Mem. Accad. Naz. Lincei, VIII 7 (1964), 91-140.
[2] Fichera, G.: Existence Theorems in Elasticity. In: Encyclopedia of
Physics (ed. by S. FlUgge) Vol. VIa/2, Springer-Verlag, Berlin 1972.
[3] Lions, J.L. and G. Stampacchia: Variational Inequalities, Comm. pure
and applied Math., XX (1967), 493-519.
[4] Moreau, J.J.: La notion de sur-potentiel et les liaisons unilatera-
les en elastostatique, C.R. Acad. Sc. Paris, 267A (1968), 954-957.
[5] Panagiotopoulos, P.O.: Non-convex Superpotentials in the Sence of
F.H. Clarke and Applications, Mech. Res. Comm, 8 (1981),335-340.
[6] Panagiotopoulos, P.O.: Nonconvex Energy Functions. Hemivariational
Inequalities and Substationarity Principles, Acta Mechanica, 42
(1983), 160-183.
[7] Duvaut, G. and J.L. Lions: Les inequations en Mecanique et en
Physique, Dunod, Paris 1972.
[8] Baiocchi, C. and A. Capelo: Variational and Quasivariational In-
equalities, J. Wiley, New York 1984.
[9] Kinderlehrer, D. and G. Stampacchia: An Introduction to Variational
Inequalities and their Applications, Academic Press, New York 1980.
[10] Friedman, A.: Variational Principles and Free Boundary Problems,
J. Wiley, New York 1982.
[11] Rodriguez, J.F.: Obstacle Problems in Mathematical Physics,
North Holland, Amsterdam 1987.
[12] Panagiotopoulos, P.O.: Inequality Problems in Mechanics and Appli-
cations. Convex and Nonconvex Energy Functions, Birkhauser Verlag,
Boston-Basel 1985 (Rus. Transl. M/R Pub., Moscow 1987).
[13] Clarke, F.H.: Optimization and Nonsmooth Analysis, J. Wiley,
New York 1983.
[14] Rockafellar, R.T.: Generalized Directional Derivatives and Sub-
gradients of Non-convex Functions, Can. J. Math., XXXII (1980),
257-280.
[15] Rockafellar, R.T.: La theorie des sous-gradients et ses applications
a l 'optimization. Fonctions convexes et non-convexes, Les Presses
de l 'Universite de Montreal, Montreal 1979.
[16] Hamel, G.: Theoretische Mechanik, Springer-Verlag, Berlin 1967.
Hemivariaitonal Inequalities 173

[17] Floegl, H. and H.A. Mang: Tension Stiffening Concept Based on Bond
Slip, ASCE, ST 12, 108 (1982), 2681-2701.
[18] Panagiotopoulos, P.D.: Hemivariational Inequalities in Frictional
Contact Problems and Applications. In: Mech. of Material Interfaces
(ed. by A.P.S. Selvadurai, G.Z. Voyatzis), Elsevier Sc. Publ.,
Amsterdam 1986.
[19] Pires, E.B. and J.T. Oden: Analysis of Contact Problems with
Friction Under Oscillating Loads,Comp. Meth. Appl. Mech. Eng., 39
(1983), 337-362.
[20] Green, A.E. and P.M. Naghdi: A General Theory of an Elastic-Plastic
Continuum, Arch. Rat. Mech. Anal., 18 (1965), 251-281.
[21] Halphen, B. and N.Q. Son: Surles Materiaux Standards Generalises,
J. de Mecanique, 14 (1975), 39-63.
[22] Germain, P.: Cours de Mecanique des milieux continus I, Masson,
Paris 1973.
[23] Panagiotopoulos, P.D. and E.K. Koltsakis: Interlayer Slip and Dela-
mination Effect: A Hemivariational Inequality Approach, Trans. of
the C.S.M.E., 11 (1987), 43-52. -
[24] Baniotopoulos, C.: Analysis of Structures with Complete Stress-
strain Laws, Doct. Thesis, Aristotle Univ., Dept. Civil Eng.,
Thessaloniki 1985.
[25] Panagiotopoulos, P.D. and C. Baniotopoulos: A Hemivariational In-
equality and Substationarity Approach to the Interface Problem:
Theory and Prospects of Applications, Eng. Anal., 1 (1984), 20-31.
[26] Holister, G.S. and C. Thomas: Fibre Reinforced Materials, Elsevier,
London 1966.
[27] Green, A.K. and W.H. Bowyer: The Testing and Analysis of Novel Top-
Hat Stiffener Fabrication Methods for Use in GRP Ships, Proc. 1st
Int. Conf. on Composite Structures (ed. by I.H. Marshall), Applied
Science Publishers, London 1981.
[28] Williams, J.G. and M.D. Rhodes: Effect of Resin on Impact Damage
Tolerance of Graphite/Epoxy Laminates, Proc. 6th Int. Conf. on
Composite Materials, Testing and Design, ASTM STP 787 (ed. by
I.M. Daniel), ASTM, Philadelphia 1982.
[29] Schwartz, M.M.: Composite Materials Handbook, McGraw-Hill, New York
1984.
[30] Baniotopoulos, C.C. and P.D. Panagiotopoulos: A Hemivariational In-
equality Approach to the Analysis of Composite Material Structures,
Proc. Conf. Composite Structures Patras, Omega Press, London 1987.
[31] Dunford, N. and J. Schwartz: Linear Operators, Part I: General
Theory, Interscience Publ., New York 1966.
[32] Sheikh, A. and S.M. Uzumeri: Analytical model for concrete confine-
ment in tied columns, J. Struct. Div. ASCE, 108 (ST12) (1982),2703.
[33] Emori, K. and W.C. Schnobrick: Inelastic behavior of concrete
frame-wall structures, J.Struct. Div. ASCE, 107 (ST1) (1981), 145.
[34] Gilbert, J. and R.F. Warner: Tension stiffening in reinforced
concrete slabs, J. Struct. Div. ASCE, 104 (ST12) (1978), 1885.
[35] Hutchinson, J.W.: Micromechanics of Damage in Deformation and
Fracture, Edition of Solid Mech. Dept., T~ Denmark 1987.
174 P.D, Panagiotopoulos

[36] Panagiotopoulos, P.O.: Ioffe's Fans and Unilateral Problems: A ~~ew


Conjecture, Proc. 3nd Conf. Unil. Problems CISM, Springer-Verlag,
Wien 1987.
[37] Klarbring, A.: Contact Problems in Linear Elasticity, Linkoping
Studies in Science and Technology, Dissertation No. 133, Linkoping
1985.
[38] Panagiotopoulos, P.O.: Dynamic and Incremental Variational In-
equality Principles, Differential Inclusions and their Applications
to Co-Existent Phases Problems, Acta Mechanica, 40 (1981), 85-107.
[39] Nitsiotas, G.: Die Berechnung statisch unbestimmter Tragwerke mit
einseitigen Bindungen, Ing. Archiv, 41 (1971), 46-60.
[40] Panagiotopoulos, P.O.: A Variational Inequality Approach to the In-
elastic Stress-Unilateral Analysis of Cable-Structures, Comp. and
Struct., 6 (1976), 133-139.
[41] Mitsopoulou, E.: Unilateral Contact, Dynamic Analysis of Beams by
a Time-stepping Quadratic Programming Procedure, Meccanica, 18
(1983), 254-265.
[42] Doudoumis, I.N.: f~odell ing "infill finite elements" with unilateral
Interface conditions and general nonlinear constitutive laws, Doct.
Thesis, Aristotle Univ., Dept. Civil Eng., Thessaloniki 1988 (to
appear).
[43] Mitsopoulou, E.N. and I.N. Doudoumis: A Contribution to the Analy-
sis of Unilateral Contact Problems with Friction, S.M. Archives,
12 (1987), 165-186.
[44] Crisfield, M.A. and J. Wills: Solution Strategies and Softening
Materials, Comp. Meth. Appl. Mech. Eng., 66 (1988), 267-289.
[45] Strodiot, J.J. and V.H. Nguyen: On the Numerical Treatment of the
Inclusion OEaf(x). In: Topics in Nonsmooth Mechanics (ed. by
J.J. Moreau, P.O. Panagiotopoulos, G. Strang), Birkh~user Verlag,
Basel 1988.
[46] Panagiotopoulos, P.O.: Nonconvex Problems of Semipermeable Media
and Related Topics, ZAt~M, 65 (1985), 29-36.
[47] Panagiotopoulos, P.O.: Hemivariational Inequalities and Sub-
stationarity Principles in the Static Theory of von Karman Plates,
ZAMM, 65 (1985), 219-229.
[48] Panagiotopoulos, P.O. and G. Stavroulakis: A Variational-Hemivari-
a tiona 1 I nequa 1 i ty Approach to the Lamina ted Plate Theory under
Subdifferential Boundary Conditions, Q. of Appl. Math. (to appear).
[49] Panagiotopoulos, P.O.: A Nonlinear Programming Approach to the
Unilateral Contact- and Friction-Boundary Value Problem in the
Theory of Elasticity, Ing. Archiv, 44 (1975), 421-432.
[50] Necas, J., Jarusek, J. and J. Haslinger: On the Solution of the
Variational Inequality to the Signorini Problem with Small Friction,
Bulletino U.M.I., 17B (1980), 796-811.
[51] Panagiotopoulos, P.O.: Variational Inequalities and Multilevel
Optimization Techniques in the Theory of Plasticity, Comp. and
Struct., 8 (1978), 649-650.
[52] Koltsakis, E.K.: Doct. Thesis, Aristotle Univ., Dept. Civil Eng.,
Thessaloniki (to appear).
Hemivariational Inequalities 175

[53] Panagiotopoulos, P.J. and E.K. Koltsakis: Hemivariational Inequali-


ties for Linear and Nonlinear Elastic Materials, Meccanica, (1987),
(to appear) .
[54] Jones, R.: Mechanics of composite materials, McGraw Hill, New York
1975.
[55] Stavroulakis, G. and P.O. Panagiotopoulos: Laminated Orthotropic
Plates under Subdifferential Boundary Conditions. A Hemivariational
Inequality Approach, ZAMM, (to appear).
[56] Panagiotopoulos, P.D. and G. Stavroulakis: A Hemivariational In-
equality Approach to the Delamination Effect in the Theory of
Layered Plates, Archives of Mechanics, (to appear).
[57] Naniewicz, Z. and C.Z. Wozniak: On the quasi-stationary models of
debonding processes in layered composites, Ing. Archiv, (to appear).
[58] Naniewicz, Z.: On some nonmonotone subdifferential boundary con-
ditions in Elastostatics, J. Eng. Math., (to a~pear).
[59] Naniewicz, Z.: On some Nonconvex Variational Problems related to
Hemivariational Inequalities Nonlinear Analysis and Applications,
(to appear).
[60] Chang, K.C.: Variational Methods for Non-differentiable Functionals
and their Applications to Partial Differential Equations, J. Math.
Anal. Appl., 80 ( 1981), 102-129.
[61] Rockafellar, R.T.: Convex Analysis, Princeton Univ. Press, Prince-
ton 1970.
[62] Ekeland, I. and R. Temam: Convex Analysis and Variational Problems,
North-Holland, Amsterdam and American Elsevier, New York 1976.
[63] Panagiotopoulos, P.D. and G.E. Stavroulakis: The Delamination Effect
in Laminated v.Karman Plates under Unilateral Boundary Conditions.
A Variational-Hemivariational Inequality Approach, J. of Elasticity,
(to appear).
[64] Berger, M.S. and P.C. Fife: Von Karman's Equations and the Buckling
of a Thin Elastic Plate II. Plate with General Edge Conditions,
Comm. Pure Appl. ilath., XXI (1968), 227-241.
[65] Strauss, W.: On weak solutions of semilinear hyperbolic equations,
An. Acad. Brasil. Ci., 42 (1970), 645-651.
[66] Rauch, J.: Discontinuous semilinear differential equations and
multiple valued maps. Proc. A.M.S., 64 (1977), 277-282.
[67] Aubin, J.P.: Applied Functional Analysis, J. Wiley, New York 1979.
[68] Landesman, E.M. and A.C. Lazer: Nonlinear Perturbations of Linear
Elliptic Boundary Value Problems at Resonance, J. Math. Mech., 19
(1970), 609-623.
[69] McKenna, P.J. and R. Rauch: Strongly Nonlinear Perturbations of
Nonnegative Boundary Value Problems with Kernel, J. Diff. eq., 28
(1978), 253-265.
[70] Warqa, J.: Derivate Containers, Inverse Functions and Controll-
ability, In: Calculus of Variations and Control Theory (ed. by
D.L. Russell), Acad. Press, New York 1976.
[71] Demyanov, V.F. and A.M. Rubinov: On Quasidifferentiable Mappinqs,
Math. Operationsforsch. u.Statist. Ser. Optimization, 14 (1983),
3-21.
176 P.D. Panagiotopoulos

[72] Demyanov, V.F. and L.C.W. Dixon (eds): Quasidifferential Calculus,


Math. Progr. St., 29 (1986),
[73] Demyanov, V.F. and L.V. Vasil 'ev: Noncifferentiable Optimization,
Optimization Software Inc., New York 1985.
[74] Demyanov, V.F., Polyakova, L.N. and A.f~. Rubinov: Nonsmoothness and
Quasidifferentiability, Math. Progr. Study, 29 (1986), 1-19.
[75] Shapiro, A.: On Optimality conditions in quasidifferentiable opti-
mization, SIAM J. Control Opt., 22 (1984), 610-617.
[76] Demyanov, V.F. and A.M. Rubinov: On quasidifferentiable functionals,
Soviet Math. Doklady, 21 (1980), 13-17.
[77] Demyanov, V.F., Polyakova, L.N. and A.M. Rubinov: On one generali-
zation of the concept of subdifferential. In: All Union Conf. on
Dynamic Control, Abstracts and Reports, Sverdlovsk 1979.
[78] Polyakova, L.N.: On minimizing the sum of a convex function and a
concave function, Math. Progr. Study, 29 (1986), 69-73.
[79] Kiwiel, K.C.: A linearization method for minimizing certain quasi-
differentiable functions, Math. Progr. Study, 29 (1986), 85-94.
[80] Radstrom, H.: An embedding theorem for spaces on convex sets, Proc.
AMS, 3 (1952), 165-169.
[81] Ball, J.M.: Convexity Conditions and Existence Theorems in Nonlinear
Elasticity, Arch. Rational ~~ech. Anal., 63 ( 1977), 337-403.
[82] Ciarlet, P.G. and J. Necas: Unilateral Problems in Nonlinear Three-
Dimensional Elasticity, Arch. Rational Mech. Anal., 87 (1985), 319-
338.
[83] Ciarlet, P.G. and J. Necas: Injectivity and Self-Contact in Non-
linear Elasticity, Arch. Rational Mech. Anal., 97 (1987), 171-188.
[84] Demyanov, V.F.: Quasidifferentiable functions: Necessary conditions
and descent directions, Math. Progr. Study, 29 (1986), 20-43.
[85] Rubinov, A.M. and A.A. Yagubov: The space of star-shaped sets and
its applications in nonsmooth optimization, Math. Progr. Study, 29
(1986), 176-202.
[86] Germain, P.: La methode de puissances virtuelles en mecanique des
milieux continus, 1ere partie. Theorie du second gradient, J. de
Mecanique, 12 (1973), 235-274.
[87] Salencon, J. and A. Tristan-Lopez: Analyse de la stabilite des talus
en sofs coherents anisotropes, C.R. Acad. Sc. Paris, 2908 (1980),
493-496.
CONTACT WITH ADHESION

M. Fremond
Laboratoire Central des Ponts et Chaussees
Laboratoire de Modelisation des Materiaux et des Structures du Genie Civil
Paris, France

1. INTRODUCTION

Tearing adhesive paper glued onto a support requires a


notably greater effort than that needed to overcome its
weight. The additional effort is due to the adhesion. We give
a description of the phenomena based on continuum mechanics.

The main idea is that usual displacements fields are not


sufficient to describe the contact. Two solids in contact can
be either strictly bonded (adhesion is present) or simply
juxtaposed without any bounding (no adhesion is present). In
both situations the displacements are the same but the mecha-
nical situations are different. We therefore introduce a new
variable to describe the state of contact : the ratio ~ of
active bonds called the intensity of adhesion.
178 M. Fremond

2. PRINCIPLE OF VIRTUAL POWER

2.1. Kinematic quantities

Let there be a structure, occupying at time t a closed


region ~ c ~n (n = 2 or 3) Along a part r of its boundary the
structure is in contact with a support S which is assumed for
the sake of simplicity to be rigid (figure 1).

support S
Figure 1
The system formed by ~ and the part r of the support.

The mechanical system we consider is the closed domain


~. Along the part r of the boundary it comprises both the ma-
terial points of the support S and those of the structure.
The kinematic description on r is achieved through the velo-
city field u(x) of the points of the structure and the velo-
city field uextx) of the points of the support. In all other
parts of ~, the velocity u(x) describes the kinematic state,
(figure 1).

In our opinion these velocities are not sufficient. The


formation or destruction of bounds between ~ and s must be
accounted for. Even if the velocities u(x) and uext (x) are
the same at a point x, the state of bounds between 0 and s
may differ. Thus, a new kinematic variables to describe the
microscopic displacements that form or destroy bonds appears
to be necessary.
Contact with Adhesion 179

Figure 2
Links ensuring adhesion

To this end we define~, the intensity of adhesion,


being the proportion of the active bonds between the struc-
ture and the support. If ~ = O,all the bonds are broken, no
adhesion exists ; if ~ = 1, all the bonds are active, total
adhesion is present ; if 0 < ~ < 1, a part ~ of the bonds re-
mains active, the remaining bounds are broken, the adhesion
is partial. The intensity of adhesion ~(x,t) depends on the
time t and the point x. We assume this point x to be a mate-
rial point of the structure n. Other assumptions are possi-
ble. This one is sufficient for our purpose. The velocity
.
~ = ~. . . . .
dt 1s a macroscop1c representat1on of the m1croscop1c ve-
locities that form and destroy the bonds.

We choose the triplet (u, uext I ~) to describe the mo-


tion of n and of its support. one must note that the velocity
~ is an element of a linear space although ~ is not due to
the definition of ~ : 0 s ~ s 1.

We define the linear spaces v X vext of the velocities


of the material points of the system and the linear space B
of the velocities of the adhesion intensities.
180 M. Fremond

We assume that elements of V x Vext x E are smooth


enough for the following calculations to be coherent. An ar-
bitrary element of u = v X vext X E is denoted by
... ...
d = (VI VeX t I 'Y)

2.2. Virtual power of the internal forces.

Let us consider a movement such that u = 0, uext = o and


~ 7 0. There is no macroscopic velocity but in this movement
microscopic velocities create and destroy bonds. We think
that the power of these microscopic velocities must appear in
the expression of the power of the interior forces. The usual
expression of the power of the interior forces has the form

...
where u denotes the stresses, D(u) the strain rates
(oii (u) ~(ui ,i+ ui,i >) and R the internal reaction. It is
obvious that this expression does not take into account the
power of the microscopic velocities. Thus it should appear in
another term having the usual expression of a power : that is
a linear function of the velocity. The simpliest choice avai-
lable is

where the interior force F has the dimension of a surface


work.

The theory that results from this choice is a zero gra-


dient theory in ~ because no spatial derivatives of ~ ap-
pears. A first gradient theory would results from the choice

for the power of the microscopic velocities [3].


Contact with Adhesion 181

In order to define the virtual power ~; of the internal


forces to a 9losed domain D c n, which is a linear form on
the space U, we must fulfill the virtual power axiom [6]

~; = o, for every rigid body velocity field at time t.

To satisfy this axiom we need to define the subset of the


rigid body velocities. It is clear that D(v) = o,vext-v = o
on r for such a velocity field. For a rigid body velocity
field, the state of the bonds does not change : no bonds are
formed or destroyed. It therefore appears reasonable to assu-
me ~ = 0 for a rigid velocity field. Thus a rigid velocity
field is defined by

-
D(v) = 0 in D vextv = 0 and~= 0 in D n r.
The first two relations are classical. The last one is new.
One can point out that it finds strong support from existing
experience which emphasises the necessity of some deformation
of the structure in order to modify the adherence (that is to
say in order to have~~ 0) [9].

Having defined the linear subset of the rigid body velo-


cities, we retain as power of the internal forces to a closed
subdomain n c n, the expression

that satisfies obviously the virtuar power axiom.

2.3. Virtual power of the external forces.

Within the interior of a domain D there is the classical


volume external force f which is a remotely exerted force. On
the boundary D n r, one can assume, besides the classical
traction T, that the local action of the outside of n onto
n n r involves a surface work A. For instance A can represent
the work of an electric current which modifies the state of
the bonding of two plastic sheets in contact.
182 M. Fremond

The virtual power of the exernal forces is

The local external action T is applied to the support on the


part D n 1 because the material points of the support are
those which are in contact with the exterior of the system.

2.4. Principal of virtual power. Equation of equi-


librium.

We restrict ourselves to the quasi-static case for which


the principle reads [6)

VD c 0, Vd E U, ~i + ~e 0.

By choosing D such that D n 1 = 0, we can first obtain


the classical equilibrium equations,

a.1 J. . + f
J 1 = 0 in D, ( 1)

( 2)

where nj are the direction cosines of the ourward normal n to


D. Then by choosing D such that D n 1 7 0, we obtain the new
equations of equilibrium

F = A, ( 3)

( 4)

In the present zero gradient theory for~, the internal force


F is known as soon as the external force A is known. This si-
tuation can be compared to the isostatic situation for beams.
The relations (3) and (4) give also physical meaning to the
internal forces R and F : R is the reaction of the support
and F is the surface density of work provided to the system.
Contact with Adhesion 183

3. ENERGY BALANCE

The internal energy E within a closed domain D can be


decomposed into a volume density Pe (p is the volume density)
and two surface densities Pnen and Ps es, where Pn, ~n are
the surface density and the internal energy of the bonds
which remain with the structure; Ps, en are the surface
density and internal energy of the bonds which remain with
the support S. The energy Pn, en refers to the bonds with
macroscopic velocity u and the
- -
energy Ps, es to those with
macroscopic velocity uext.

In accordance with the quasi-static assumption, we neg-


lect the kinetic energy in the energy balance

dE
dt
~e + Q =- ~i + Q, (1)

where Q is the rate of heat supplied to D and ~e' ~i the ac-


tual powers of the external and internal forces.

We choose for Q

L~ cmq c-qn) d.Y,


where r is the volume rate of heat production inside n, the r
surface rate of heat production (for instance, the rate of
heat produced by a chemical reaction), q
is the volume heat
flux vector in n, q the surface heat flux vector in r (we as-
sume that heat can flow through the bonds), m'is the outward
normal to the domain D n r in r, w is the heat which is sup-
plied by the exterior to D through D n r.

Note. Mass conservation on r.


184 M. Fremond

The mass conservation of the bonds which remain with r


reads

o, ( 2)

where div1 (PnU) is the surface divergence.

The mass conservation of the bonds which remain with the


support reads

0, ( 3)

Assuming ! to be flat,

div1 f = foe., oc.

where the ~'s are the two indices of the orthogonal coordi-
nates in r.

By using relations (1) and (3), it is easy to prove that

~r
dtJmf'
{Pnen + p 5 e 5 }dO
= Lr{Pn
den
- - + Ps
dt
de 5
dt
}en.
Let us recall that the relation

-d
dt
J
D
pe dO= J pde
- dO,
D dt

is classical. The conservation of energy equation ( 1) gives


Contact with Adhesion 185

v:o c n,

By choosing :0 such that :0 nr = 0, we obtain

p
de
dt
+ div q= aD(;) + r, in :0, ( 4)

We then obtain

-
Pn --- + p5 --- + divrq F13 + R(u-uext) + r + (w+qn), (5)
dt dt

This is the energy conservation equation on n n r. It


can be considered as a boundary condition for the energy con-
servation equation (4) in n, because it gives the heat flux
- qn which is provided from the support to the structure. One
can also note that the quantity (w+qn) is the difference bet-
ween the heat w that flows into r from the exterior and the
heat - qn which is provided to the structure and leaves r.
The quantity divrq accounts for the diffusion of heat through
the bonds.

4. SECOND PRINCIPLE OF THERMODYNAMICS

We assume again th~t the entropy S of a closed domain :0


is the sum of volume and surface contributions
186 M. Frcmond

Denoting by T the absolute temperature, the second principle


of thermodynamics is

dS
dt
~
J:0 E
T
dO +

which must be satisfied for any domain :0 and any actual evo-
lution. We obtain easily that

ds q r
P + div ~ o, in n,
dt T T

dsn ds 5 r+w+qn
- - + Ps - - +
dt dt
T ~ o, in r.

Multiplying these equations by T and adding the energy con-


servation equations (3.4) and (3.5), we obtain

Pn
dsn
[T ---
dt
den
dt
l +
ds 5
P, [T ---
dt
de,
dt
l +

+ F~ + R(uext-u) - -
q gradr T
T
~ o,

where gradrT is the surfacic gradient.

- - - - -
Defining ~ = e-Ts, ~n = en-Tsn, ~s e 5 -Ts 5 and inte-
grating over a domain :0 we obtain the Clausius-Duhem
inequality,
Contact with Adhesion 187

v:n, r {p
J:n
d'l<
dt
+ Ps
dt
dT}dO + Jmr{en d.J..n
- - + P.n S.n
dt dt
dnT}
d'l<s
d'T}
- - + Ps ss - - d1 s
+ {" dt dt f:u{ao(\i) q grad
T
T}dn

L,{~
_ q gradrT}
+ + R(uext-u) - d1,
T

. . d.nT d5 T
where the der1vat1ve dt and dt are the material derivatives
with respect to the structure and the support

aT aT
+ grad T u, +grad T uext.
at at

The Clausius-Duhem inequality must be satisfied for any

--
domain :0 and by any actual evolution; i.e., for any veloci-
ties u, uext, ~. and any heat fluxes q and q. -
The free energies '1<, .J. 5 and '1< 5 depend on the
temperature T and other quantities xi . Let us define the
functions

(t,T) ----> ~(t,T) 'l<(xi (t) ,T),


(t,T) ----> ~.n (t,T) .J..n (Xi (t) IT) I

(t,T) ----> ~ 5 (t,T) '1< 5 (Xi (t), T),

and assume the Helmholtz postulate


188 M. Frcmond

The Clausius-Duhem inequality gives

I a1',}
Lr{Pn
a>11 a>1'n
p --dll + - - + Ps - dr
at at at
J:o
~ grad
s J:o{ao(u) - q T}dll ( 1)
r

Lr{ro
q gradrT}
+ R (ue x t -u) - dr,
+ T

Let us define the intrinsic dissipation

o, J:o{ao(u) - p -a>11} dll +


at

""} fJ
( 2)

J {ro
mr
+ RcU ex t -U> - Pn ---
a>1'n
at Ps - - dr
at '

and the thermal dissipation

-
02 =
I :0-
q~ grad
T
T dll + I -
mr
- q gradr T
T
dr, ( 3)

The Clausius-Duhem inequality takes the form

v:o, Vu, uext I ~I grad T, gradrT, o, + 52 ~ 0.

Following the usual practice, we replace the inequality


61 +62 ~ 0 by the two stronger ones

0 1 ~ 0 and 62 ~ o. ( 4)

For the thermal dissipation, we choose as usual the


Fourier's laws
Contact with Adhesion 189

q = - k grad T, q - k gradrT, (5)

-
where k and k are the thermal conductivities of the structure
and the bonds. Thus the second inequality of (4) is
satisfied.

Note.- Let us define the function

One can prove by using the mass conservation laws (3.1)


and (3.2) that

a>j;
I p --
at
D

a> o
The quantity at is the material derivative of - 0 where
the temperature T is kept to its value T(t) at time t.

Thus the inequality (4) can be written

(6)

o,

5. FREE ENERGY WITHIN THE SMALL DEFORMATIONS ASSUMPTION

From now on we suppose the deformations to be small. The


quantities u(x,t),uext (x,t), will further denote the
small displacements.

We assume the free energy to depend on the


=
- -
temperature T, on ~ and on the small deformations (~) D(~)
and uext-u.
190 M. Frcrnond

We assume that if the relative displacement u ext-u of


the support and the structure is not zero, all the bonds are
broken. This assumption describes the behaviour of the con-
tact and should be taken into account by the free energy. One
can choose an other behaviour allowing interactions at
distance in order to describe e.g. a glue film that ruptures
progressively as the crack between the support and the struc-
ture opens. The behaviour we have chosen implies

- -
f3 ( ue x t -u) = 0' ( 1)

and
- --
(uext-u)n?:: 0, ( 2)

besides
0 :::; f3 :::; 1, ( 3)

The last condition results from the definition of~.


The last but one is the non interpenetration condition
between the body and the support.

The actual evolutions of the system that satisfy the


Clausius-Duhem inequality at every time instant, satisfy also
the internal constraints (1), (2) and (3). We can therefore
define the free energy at our convenience if the internal
constraints are not satisfied, without affecting the
Clausius-Duhem inequality. Taking advantage of this, we can
include the internal constraints into the free energy by
assuming this free energy to take the value +oo if one of the
constraints is not satisfied. As we have said, we do think
that the internal constraints are actually parts of the
physical or mechanical properties of the system and that the
free energy must account for them.

We
support
now suppose,
is rigid - for the sake of simplicity, that the
u ext = 0 and that the structure is fixed
along a part 1 0 u = o on 1 0 (figure 3).
Contact with Adhesion 191

support S
Figure 3
The structure fixed on the part 1 0

Let us give some notations :

-
{v v E V : v = 0 on 1 0 } x B,

11:, (D, u, vI T) fD{A(T)Ekk (u)Ekk (v) + 21-L(T)E; j (U)E; j (;)}dll,

11:, (D,u,v,T) = JDk 2 (T)uv dr,

K = { (v, 'Y) I (v, 'Y) E 'U 0 ; o ~ 'Y ~ 1: 'Yv = o: - vn ~ o on r} ,

where A(T), 1-L(T) are the Lame elastic parameters and k 2 (T) a
positive elastic rigidity, (3A+21-L > o, 1-L > o, k 2 > 0).

We define the free energy of a domain D by

-
~(D,u,~,T) = 12 - -
11:1 (D,u,u,T) + i1 ~ (D,-u,-u,T)
- -

where w(T) is the Dupre's energy, ~ the thermal expansion


coefficient, T0 is the reference temperature at which we
192 M. Fremond

consider the system (A 0 = A(T 0 ), ~ 0 = ~(T 0 )), Cis the thermal


heat capacity of the structure, IK is the indicator function
of the non convex set K (IK (x) = 0 if X >:: K,

IK (X) =+~if X~ K).

The free energy just defined is smooth and concave with


respect to the temperature T as it should. Regarding the
other variables, it is convex on every convex set on which it
is finite. The difference from the usual situation is
that the function ~ is not convex with respect to B and u

In order to calculate the internal energy E let us make


the usual assumption that A(T), ~(T) and k 2 (T) are linear
function of T. Thus we obtain

E = ~- Ta~
- = J PCT + (3A 0 +~ 0 )cxT 0 kk (u)dO
aT D

+ Jmr (w(T) - T dw) (1-B)dr.


dT

We can note that the surface energy is equal to zero if


the bonds are unaltered(~= 1). In this situation the con-
tact zone is not distinguished from the structure. It is only
when B differs from 0 or when some the bonds are brocken that
the internal energy of the contact is distinguished from the
internal energy of the structure. Of course other assumptions
are possible : for instance, one can assume heat capacities
-
Cn and C5 for the bonds.

Note. The function ~0 (T,t) previously defined is

~a (T,t) = ~c:o,iict), B (t) ,T).

One can remark that, due to the small deformations as-


sumption, V does not depends on the time t.

As we have included the internal constraints in the free


energy, the function ~ 0 is not smooth. Thus we must focus on
a possible chain rule
Contact with Adhesion 193

.,..
~ + --- u.
au
We cannot use the subdifferential of a convex function
[10) because ~ is not convex. Thus we define a similar notion
the local subdifferential.

5.1. Local subdifferential of a non smooth function.

Let ~ be a mapping from a topological vector space X, in


duality with x*, into~ U {+oo}. Then~ is said to be locally
subdifferentiable at the point x E X if it is finite at x and
there exists a neighbourhood ~(x) of x and x* E x* such that

Vz E ~(x), ~(z) ~ ~(x) + < x* ,z-x >,

where< , >are the duality brackets.

Every x* satisfying this inequality is called a subgra-


dient of ~ at x. The set of all the subgradients is the sub-
differential of ~ at x denoted a~(x).

5.2. Properties of the free energy.

The Clausius-Duhem inequality must hold for any subdo-


a~o
main D c n. This imposes to at to have the same structure as
the first term of (4.6). This implies that the subgradient of
~0(0), that will be calculated, can be restricted to the
subdomains of n.

This requirement is natural because the behaviour of a


material does not depend on the shape of the sample
considered.

-
The elements d = (v,~) of U0 or the deformations and the
adhesion intensity are in duality with the internal forces
f = (a,R,F) by the bilinear form
194 M. Fremond

We shall show that the function ~(O,u,~,T) is locally


subdifferentiable with respect to tne variables u and B and
that among its local subgradients, there exists one whose
restriction to Dis a local subgradient of ~(D,u,~,T).

PROPOSITION. The function ~(O,u,~,T) is concave and


smooth with respect to T and convex with respect to the
...
variables u, ~ on the convex domains where it is finite. It
is locally subdifferentiable with respect to u and~. There
exists a subgradient of ~(O,u,~,T) whose restriction to D is
also a subgradient of ~(D,u,~,T), i.e.

VD c 0, Vd E 'W(d 1 ) , < f,d-d 1 >n 5 ~(D,d,T) - ~(D,d 1 ,T),


...
where 'W(d 1 ) is a neighbourhood of d 1 = (u,~).

Proof. The only point not obvious is the local subdif-


ferentiability of ~. The function ~ is the sum of a smooth
part wr and a non smooth part IK (W = Wr+IK). The regular
part is convex. We assume that the topology on U0 is such
that ~r is differentiable and that the differential is the
classical one defined by the volume density

>..(T)kk (u(x))n + 2~(T) (u(x)) - cx(3>.. 0 +2~ 0 ) (T-T 0 )n

and the surface densities,

Fr (x) =- w(T).
...
Let us define fr = (ar,Rr,Fr) and d1 (u,~). Because
wr is convex, we have
Contact with Adhesion 195

( 1)

for any d E U0 and any D c 0. Moreover, as IK ~ 0 we have

'll(D,d,T) ~ '~'r (D,d,t), (2)

For d 1 -
(u,~) E K, we have

(3)

Relations (1), (2), (3) give

Vd 1 (u,~) E K,
there exist fr such that

VD c 0, \fd E U0 , < fr ,d-d 1 >D S 'll(D,d,T) - 'll(D,d 1 ,T),

which proves that '1'(0) is locally subdifferential (for this


particular subgradient fr the neighbourhood of d 1 is the
whole space U0 ) and that the restriction of fr to Dis a
subgradient of 'll(D).

We denotes by o'll(d 1 ) the set of the all the subgradients


of '1'(0) having the restriction property just mentionned. It
is clear that a>(d 1 ) does not reduce to the element fr.

6. INTRINSIC DISSIPATION. CONSTITUTIVE LAWS.

a> o
Let us calculate which is an element of inequality
at
(4.6).

Let f e o>(d 1 Assuming


-
), d, (t)

(u(t),~(t)) to be continuous with respect tot, we have

d 1 (t+6t) E V(d 1 (t)),


196 M. Frcmond

for 6t small enough. Therefore, since ~0 (D,t,T) = ~(D,d 1 (t) ,T)

< f,d 1 (t+6t) - d 1 (t) >:o ~ ~o (:O,t+6t,T) - ~0 (t,T), (1)

for 6t small enough.

Let us divide relation (1) by 6t > 0 and let 6t tend to


zero, we obtain

( 2)

Let us divide relation (1) by 6t < 0 and let 6t tend to


zero, we obtain

dd, {'Jill 0
< f, - - >::;. (3)
dt at '

Relations (2), (3) give

Therefore we obtain an useful expression for the intrinsic


dissipation 51 such that the inequality (4.6) reads

51 = JD(a-T) (~)dO+ f:onr{(R-G 1 ) (-~) + (G 2 -F 2 )~}dr ::::> o, (4)


for any :0 and any actual velocities~ and~.

In order to satisfy the inequality (4), we assume that


there exists a pseudo-potential of dissipation [7], [11)

(5)

where <p and <p are convex, positive functions with


-
<p(O) = <p(O,O) = 0.

We say that firr ( Tirr , Girr


1 , Girr)
2 is an element of the
Contact with Adhesion 197

subdifferential o~(~ 1 6) if

VD 1 Vd E U0 1 < firr 1 d-d 1 > ~ ~(D 1 d) - ~(D 1 d 1 )

with d 1

We choose as constitutive laws

f E a> (d 1 ) I ( 6)

(7)

- I F) = f + f
(a I R 1r r (8)

The internal forces are the sum of reversible internal


forces and of dissipative (or irreversible) internal forces
which depend on the velocities. It is easy to check that re-
lation (4) is satisfied with constitutive laws (6) 1 (7) and
( 8)

7. PROPERTIES OF THE INTERNAL FORCES

The properties of the dissipative forces are classical.


We focus on the reversible internal forces and assume the
dissipative forces to be zero (~ 0). We then have =
(a 1 R F)1 E a>(u 1 f3).

or by denoting d 1 = (u -1 (3) 1

( 1)

+ famr <R<-<v-ti + ('Y-f3)F}d1 ~ 'lt(d~T> - 'lt(d, IT> I

Let us give some useful technical results.


198 M. Frcmond

7.1. Technical results

Since the functions ~1 and ~ are quadratic, the rela-


tion (1) gives

\ld E '\1( ( d, ) n K, v:o c n,


1
:A, cn,v-li,v-u) +-12 ~ cn,v-li,v-u> > I (rr-')E (;-~)dOl
I
2 ( 2)

-fDl! (R+ku> cv-u)d! + JmF(F+w) c~-e);,


where T'

Let U~ -
be the subspace of the (u,~) such that the sup-
port of~ is contain in that of~' and Uu be the subspace of
the (v,~) such that the support of v is contained in that of
u. It is worth noting that the function d ~ ~(d) is convex
on these subspaces. It can therefore be shown that the rela-
tion (1) or (2} is satisfied not only on a neighbourhood of
d 1 but on U~ n K and Uu n K as well.

Let us now examine the properties of the internal


forces.

7.2. Constitutive law for u.

Let D be a domain such that D n 1 = 0. We have

where U6 is the first space component of U0

As :A, is quadratic relation (3} is equivalent to

This relation gives div(u-T') 0 in D and (u-T')n 0


Contact with Adhesion 199

on aD (n is the outward normal to D) . The directionn n being


arbitrary because Dis (provided D n 1 = 0), we then have

which is the classical elastic constitutive law.


~

7.3. Constitutive laws for Rand F.


~

7.3.1. Contact (u = 0)

We shall show that where contact exists

F -w, if 1 0 < p < 1,


F 2>: -w, if, p 1, (5)
F ~ -w, if 1 p o.
will also show that the positive part of the normal
We
traction R~ and the tangential traction = RNn are not RT R-
too large (RN = Rn, n is the ouward normal to 0 on 1 ;
R~ = RN if RN 2': 0, R~ = 0 if RN ~ 0).

In order to prove these relations, one should first


perform variations of (v,~) that will not change the nature
of the contact and use the classical tests of the variational
methods. In this way the properties of F are obtained.
Variations of (v,~) that do change the nature of the contact
(contact replaced by separation) give the properties of R.

Let D be a domain such that = o on D n 1. Let us first u


choose (u,~) E U~ n K in relation (1). we have

This relation is satisfied by any ~ such that o ~ ~ ~ 1.


It gives immediately the relations (5).

Let ~be a domain of D u 1. We choose~ 0 on ~ and


v u outside (figure 4).
200 M. Frcmond

Figure 4
A domain D and a neighbourhood 0 of a point x of 1
~

We assume that ~(d 1 ) (d 1 = (u,~)) is large enough for


~

(v,~) E ~(d 1 ). Relation (2) gives

( 6)
~ JDl1 (w+F) (~-~)d1.
We let

c 2 (0) =
Inf{-Jnn 1 (w+F) (~+~)d1 I ~ 0 on 0 ;
~

~v
~

~u outside (J)}.
By relations
~ ~

(5) and v u outside 0, we obtain

Let us now note that

w= v on :0 n 1}

1 uv-ullfDr 1 uvu~,
2 2
Contact with Adhesion 201

-
is the square of a norm of the difference u-v on D n 1 which
differs from u only on C (it is a H1 ' 2 Sobolev norm).

Let us return to relation (6) which we can now write

(7)

is a tension (RN 0) .
-= o
Let us first suppose that ~ ~

Then there exist a ~ with ~n ~ o on C and ~ outside C


such that

-
where the norm of R is the dual norm of the norm of v. Let us
take v = ~~(~ > 0) and thus (7) gives
- -+

-
Taking the minimum of the right hand side with respect
to~ (~ can be bounded since d = (~~~~) must be an element of

'W(d 1 ) , we see that IIRIIc is bounded. Thus the traction R is


not too large if RN is a tension.

Let us now suppose that RN is a compression (RN ~ 0).


Relation (7) can be written

where vN = vn is the normal displacement (vN ~ 0 since


(v,~) E K) and Vr = v-vnn is the tangential displacement. It
follows that

Defining
202 M. Fremond

o on 0 } ( 8)

the preceding relation implies that

( 9)

-
where <p is the function which achieves the supremum in ( 8) .
Taking the infimum of the right hand side of (9) with respect
to Ct > 0 (et
-
can be bounded since d = (et<p J') must be an
I

element of 'W ( d, ) ) , we prove that RT is bounded.

-
Note.If c 2 (0) = 0, relations (7) and (9) show that RT = 0
and RN ~ 0 is arbitrary. This is the classical unilateral
constitutive law. This occurs for example if f3 and u are
simultanously zero on 0 or if f3 < 1 (since F+w = 0,
relation (5)). The quantity c 2 (0) is strictly positive only
if (3 = 1.

-
The properties of R comply well with what is expected
for contact behaviour : if is possible to press down without
any limit but it is not possible to pull too much.

If the external action A = 0, the equilibrium equation


(2.3) F =A= 0 and the constitutive law (5) showthat the
only possible value of f3 is 1. This corresponds to the
following experimental result : if there is contact, the
adhesion is total. This is the case for glass-polyurethane
contact [8].

7.3.2. No adhesion (f3 = 0)

We will show that where there is no adhesion the classi-


cal unilateral conditions are satisfied
Contact with Adhesion 203

o, (10)

We will further show that

w+F-H :<:; 0

where H is the energy release rate which is defined below.

In order to prove these properties, we follow the same


lines as in the previous paragraph ; first introduce varia-
tions of (u,~) - that do not alter the state of contact, then
introduce variations that affect the nature of the contact
state.

n nr
-
Let D be a domain of and ~ a subdomain of D where
~ = 0. Let us first take (v,~) E K with v- = u- ouside ~. By
virtue of (2), we have

It results easily from this relation that

which gives the relations '( 10) .

Let us choose(v, - ~) such that v 0 on ~ and ~ ~ out-


side ~ (figure 5).
204 M. Fremond

Figure 5
A neighbourhood ~ of a point x of r where there is not
....
contact and functions (v,~)

we suppose that (v,~) E ~(d 1 )n K. It results from (2)


that

Let us define

c2 (~)

rnf{~ A, c:o,v-u,v-u) + ~ ~ c:o,v-u,v-u) + f:onr<R+ku) cv-u)drc 11 )


I v = 0, on ~ ~ = o ; vN ~ o outside o},
By (10), this expression is positive. We therefore have

Let us suppose that ~ is a neighbourhood of some point


x of r. Dividing this relation by the measure of~, meas(~),
and defining
Contact with Adhesion 205

c 2 (0)
H(x) lim (12)
0--->x meas (0)'

we obtain

w+F ~ H.

Note. We have in fact, (w+F)+ ~ H, because H is


positive.

Let us call H the energy release rate. This term, the


energy release rate, is already used in cra~k theory : it is
the derivative of the potential energy with respect to the
length of the crack [2]. We prove in the next section that
our definition is almost the same : H is the derivative of
the potential energy with respect to the contact surface.

7.3.3. A property of the energy release rate.

Let us consider a domain D of the structure such that


n n r contains a part where the structure is fixed to r, i.e.
points where ~ > 0. Due to the constitutive equations and the
equilibrium equations, the displacement u is the actual disp-
lacement of D loaded by body forces diva= (a;j,j)and surfa-
ce forces aijnj on an-r, and submitted to unilateral boundary
conditions on the part of D n r where it is not fixed. The
stress field a= a(u,T), given by relation (4), is the actual
stress field. This property results also from (1) with ~ = ~
and vbeing fixed tor where u is, i.e., where~ > o.

Let us define

1 1
2~
~ ~ ~ ~ ~
Y[D,v,C] = 2 A, (D,v,v) + (D,v,v)

with v = o on the part C of r.


~
206 M. Frcmond

Let us go back to c 2 (Q) (relation (11)). We can write

F[D,v,Z(u) U Q)] - F[D,u,Z(u)]


... ...
because an= Ron D n 1, relation (2.4). The function u lS
zero on the part o(u) = {x I x E 1
B(x) > 0} and the func-
i
tion vis zero on the part o(u) u Q of L

We then have

c2 (Q) = Inf{B I v = 0 on Q 0, vN S 0 outside OJ

rnf{F[D,v,o(u) u Q 1 v

- F[D,u,o(u)J.

F[D,v,o(u)U Q] is the potential energy


The quantity
for the displacement v of the part D of the structure being
fixed on the part o(u) u Q of 1 and being in unilateral con-
tact with the support on the remaining part of 1. The infimum
with respect to v is the actual potential energy of the part
D of the structure being loaded by the volume forces a.. . ,
1 J 'J

the surface forces a; jnj on aD-1, being fixed on o(u) u 0 and


being in unilateral contact on the remaining part of 1. The
Contact with Adhesion 207

infimum is also F[D,u,D(u) u OJ where u is the actual


equilibrium displacement of the structure with these boundary
conditions.

It is obvious that F[D,u,D(u)) is the actual potential


energy of the part D of the structure being loaded by the
same forces but fixed only on the part D(u) and in unilateral
contact on the remaining part of ann r.

The energy release rate

F[D,u,D(u) U OJ - F[D,u,D(u)J
H(x) lim
0--+x meas(O)

is the derivative of the potential energy with respect to the


part on which the structure is fixed. It is the amount of
energy by unit area necessary to modify the state of contact
at the point x. This property justifies H to be called
energy release rate. The usual definition is the derivative
of the potential energy with respect to some crack length [2J.

It seems that the energy release rate depends on D. Let


us prove that this is not the case.

THEOREM 7.1. The energy release rate H does not depend


on the part D under consideration.

Proof. Let us consider two domains D and D' such that


D c D' (figure 6). Let there be f the interior boundary
between D and D'-D.
208 M. Fremond

Figure 6
The domains D and D' (D c D').
The interior boundary f between D and D'-D.

Let us define

~ ~
B(D,v-u) = 21 ~,
- - - -
(D,v-u,v-u) + 2
1 ~
- - - -
(D,v-u,v-u)

and

c 2 (O,D) = Inf{B(D,v-u) -
v 0 on i5 (u) u 0

It is the quantity c 2 (0) for the domain D. We have

c 2 (O,D')
(13)

We have also

c 2 (O,D) + Inf{B(D'-D,v-u) -
v
(14)
Contact with Adhesion 209

because the fonction w is a candidate for the minimum of


13(:01).

Let us define

L(:o~-:o~v<:O)) = rnf<E<:o~-:o~ v-u) v

It is obvious that

From (13) and (14) 1 we get

c2 ( () I :0 I ) ~ c2 ( () I :0) + L (:0 I -:0 I v(:0 I ) ) I

(15)
c2 (C 1 :0) + L(:0 -:0 1 v(:O)) ~ c 2 (C 1 :0 1
1 ).

We have

-
L(:D'-:0 1 v) 1
=- -
llv-ull 2 112 -
2 H <r >

where the norm is the Sobolev H1 ' 2 norm of v-u on f.

We assume that

cv (:O) -u) (x) = JcG (x-y) c-u (y)) dy


where G is a Green function. The function (v(:O)-u) and its
derivatives then are proportionnal to meas(C). Therefore it
can be proved that

L(:o~-:o~v<:O)) L (:0 I -:0 I v (:0 I ) )


lim lim 0.
()--4x me as() ()--4x measC

We then conclude from (15) that

c 2 (C I :D) c2 (C 1 :0 1 )
lim
meas(C) = lim meas(C) H.
()--4x ()--4x
210 M. Fremond

We have proved (at least for the situation of figure 6)


that H does not depend on D. We assume that this is true ln
the other situations.

The energy release rate can be also defined when there


is contact. Let us consider a subdomain ~ of D n r where
there is contact (u
= o on~). Let us define

v-u 0 outside ~}

This quantity is negative since v = u is a possible


choice. By relation (6) or (7), we have

Dividing by the measure of~. meas(~), and making~~ x we


obtain

(w+F) f3 ;?: H

where

c' 2 (~)
H(x) lim ;?: o,
~x meas(~)

is the energy release rate. When w+F > 0 (it is the case when
0 =A= F, relation (2.3)), we have~= 1 and

w+F ;?: H.

When w+F 0, we have

H 0.
Contact with Adhesion 211

In this situation we have 0 ~ ~ ~ 1 by relation (5).

We have therefore obtained the following relations

w+F ~ H, where ~ o,
w+F H = o, where 0 < ~ < 1, (16)
w+F ~ H, where ~ 1.

Let us remark that, if A 0, by the equilibrium equa-


tion F = A = o, we have

w ~ H, where ~ 0,

w ~ H, where ~ = 1,

which is the classical Griffith criterium of fracture mecha-


nics. Let us also again remark that ~ cannot have values
between 0 and 1 since w ~ 0 (relation (16), for 0 < B < 1).

8. ISOTHERMAL EQUILIBRIUM OF A NON DISSIPATIVE STRUCTURE.


VARIATIONAL FORMULATION.

Let us consider a structure 0 on which act volume forces


f, surface forces g on the part 1 1 and possibly the work A on
the part 1 where the structure is in contact with the support
1. The structure is fixed on the part 1 0 We assume that the
behaviour of the structure and the contact on 1 are non dis-
sipative or elastic (~ =
0). We assume that the temperature
is constant and known everywhere (T = T0 ) (figure 7).
212 M. Fremond

Figure 7
The structure loaded by surface forces on r, .
The equations that have to be satisfied by an equilibri-
um position d 1 = (u 1 ~) are the equilibrium equations (2.1) 1
(2. 2) 1 (2. 3) 1 (2. 4) and the constitutive law

f 1 = (aIR, F) E aIJI (d, ) .


The equilibrium equations are equivalent to the virtual work
theorem

THEOREM 8.1. The equilibrium equations (2.1) I (2.2)1


(2.3) 1 (2.4), are equivalent to

vn c n, Vd E Uo I ~e (Did) + ~i (Did) = 01

where the work of the internal (resp. external) forces


~i (D,d) (resp. ~e (D,d)) is equal to ~i (D,d) (resp. >e (D 1 d)).

Proof. It suffices to follow backwards the calculations


of paragraph 2.4.

The data of the problem are IJI, T0 , -f


1 g, A. The unknown
is d 1 = (u,~). Let us give a variational formulation of the
equations. We have

Vd E W(d 1 ) n K1 IJI(O,d) ~ .P(n,d 1 ) + <f 1 ,d-d 1 >n.


Contact with Adhesion 213

Moreover

By means of the virtual work theorem, we have

\ld E 'W(d 1 ) n K,

We then have

THEOREM 8.2. The equilibrium positions of the structure


are the local minimum on ~ 0 n K of the potential energy

1
=2
-+ -+
Y[d] ~(O,d) - be (O,d) ~ (O,v,v)

By appropriately defining the respective functional spa-


ces we can prove an existence result

THEOREM 8.3. Assuming the set n to be smooth, meas r0 > o,


meas r > o, ~ > o, 3A+2~ > o, k 2 ~ o,
-+
'U~ = {v

problem (1) has at least one solution. If w+A > 0 (this is


-+ .
the case if A =0), ~ = 1, where u = o. Il w+A < 0, ~ = o
everywhere on r.

If A = o, we find that as long as there is contact, the


adhesion is total (~ = 1). This point which conforms with ex-
periment, has already been mentionned. This remark leads to
the classical formulation of adherence problems.
214 M. Fremond

9. CLASSICAL ADHERENCE PROBLEMS

We have just shown that if A = 0 the adhesion


intensity f3 assumes only the values 0 and 1. Taking advantage
of this result we restrict ourselves to function ~ with va-
lues o or 1. Let us define

K -- {V I vE 'U 01

K(v) = meas{x 1 x E r
~

v(x) = 0},

where K(v) is the contact area corresponding to the displace-


ment v

Problem (8.1) is therefore equivalent to finding the lo-


cal minima of Y on K. That is to say, finding a u
E K and a
neighboorhood Wof u (in the sense of the topology induced by
that of 'U 0 ) such that

This relation is the basic relation introducing the


adhesion theory in physics [8], [9].

10. DISSIPATIVE BEHAVIOUR

Let us examine the structure of the pseudo potential of


dissipation (6.5). The dissipation effects that are of inte-
rest to us are those related to f3. We therefore limit oursel-
ves to the case where~= 0 and ~(~ 1 P) = ;(f3).
Contact with Adhesion 215

The constitutive equations are

(a,R,G 2 ) E o'l'(d 1 ), (F-G 2 ) E o<p (~) ( 1)

They imply that

+
w+G 2 -H :5: o, where ~ = 0
w+G 2 = H = O, where 0 < ~ < ( 2)
w+G 2 -H 2 o, where ~ = 1

(in relation (7.1) and (7.16) the non dissipative force G2 is


equal to the internal force F. Here F is equal to the sum of
G2 and of the dissipative force element of o~(~)).

The velocity field ~ should take into account eventual


jumps from~ = 0 to~ = 1. When e.g. tearing wall paper off,
~ passes from 1 to zero along the debonding line. The veloci-

ty ~ is therefore a Dirac function. We consequently assume ~


to be a measure on r. The irreversible force F-G2 , the pro-
duct of which with P
is a power, is then an element of a li-
near space in duality with the measures on r (for example,
F-G2 is a continuous function on r). Thus the subgradients
of o~(~)must be element of this linear space in dualy with
the measures on r (for example continuous functions). Let us
give three examples of such pseudo-potentials.

10.1. Linear viscous behaviour

Let us choose

1 ..
<~>,(f3)=
2
(g * ~)~,

where g * ~ is the convolution of the continuous function g


with p on the surface r

g * ~ (x) frg(x-y)~(y)dy.
216 M. Fremond

The pseudo-potential

cJ> 1 ( ~) = fr ( g * f3 ) (X) X f3 (X) dX

is convex, positive and zero at the origin, if g is the Feu-


rier transform of a positive, bounded measure (e.g.
g = e- I x I ' 1/ ( 1 +x2 ) ' 1, ... ). We therefore have

F-G2 = g * f3.

The behaviour is linearly viscous. The convolution imp-


lies a kinematical interaction between neighbouring points.
The merit of this choice, guided by the need of mathematical
consistency is to be in accordance with well established ex-
perimental results, as we shall see further. Let us further
remark that the unusual from of the dissipation is related to
the absence of reversible interaction between neighbouring
points due to the zero gradient theory for f3. The first gra-
dient theory allows fort classical quadratic dissipation with
respect to f3

10.2. Norton-Hoff viscous behaviour

Let p > 1 be a number and define the pseudo-potential

we then have

which gives a nonlinear viscous behaviour. If p 2, we have


again the linear viscous behaviour.

10.3. viscous behaviour with impossible bond restitution

In this situation only debonding is possible. From the


moment it takes place, it is not possible to restore the
Contact with Adhesion 217

bonds no matter how large the applied forces are. Such is the
case with glued wall paper.

Let~- be the set of negative measures on r. We choose

where I~- is the indicator function of the set ~- We there-


fore have, letting 0 to be the interior of the support of~,

F(x} - G2 (x} = (g * ~}(X}, if X E 0,

F(x} - G2 (x} ~ (g * ~} (x}, if x $0.

10.4. Evolution equation

Let us consider the same structure n (figure 7} which we


now suppose to be made of a dissipative material, loaded by
the same external actions, possibly time-dependant. The un-
knowns are d 1 (t} and T(t} which must satisfy the equilibrium
equations (2.1}, (2.2}, (2.3}, (2.4}, the constitutive laws

(a,R,F} E a>l'(d,}, F - G2 E oct>(~) ,

the energy balance equation, the Fourier's law, as well as


boundary and initial conditions.

10.5. consequences of the dissipation

As we have already pointed out that we can consider two


processes for the bonding and the debonding

a - Bonding and debonding along a line

The phenomenon takes place along a moving line which


separates the bonded from the debonded areas. Just like when
tearing wall paper off, the debonding happens along a line.
218 M. Fremond

b - Surface or block bonding and debonding

The phenomenon takes place over the whole of a surface.


When e.g. we glue a postage stamp by pressing it down, the
gluing takes place in block. On the other hand the debonding
of the stamp takes place along a line. This behaviour seems
to be general : bonding is mostly block bonding and debonding
is always line debonding.

Let us show that the dissipative constitutive laws ac-


count for these experimental facts. For the sake of simpli-
city, we assume it is a two dimensions problem and that the
temperature remains constant.

THEOREMS 10.1. Let us consider a structure with contact


being viscous with impossible bond restitution. If w+A > o
(in particular if A = 0) only line debonding is possible.

Proof. Let us assume the structure to debond in block on


a subdomain 0 of its support. Due to the equilibrium
equations and the constitutive law (1), (2) we have

u = O, 0 < ~ < 1, F-G2 = g * ~' G2 - w, F = A on 0.

This gives

w+A = g * ~. ( 3)

Since ~ ~ 0 everywhere on r, g * ~ ~ 0 which is impos-


sible because w+A > o. Thus block debonding is impossible.

Let us examine line debonding (figure 8) and denote by


s(t) the abscissa of the point of debonding.
Contact with Adhesion 219

,G =1
---..s(t) > o
~=0 u=O
s( t) X

Figure 8
Line debonding (s(t) > O).

We have ~(x 1 t) = H(x-s(t)) where H is the Heaviside's


function and
. .
~(X 1 t) = &(x-s(t))(-s(t)) 1

where E. is the Dirac's function and s(t) > 0 if debonding


occurs.

The equations (2) 1 the constitutive law F-G2 g * ~ and


the equilibrium equations imply

w+A-H(x 1 t) ~- g(x-s(t))s(t), if x < s(t), (~ = 0) 1 }

w+A-H(s(t) 1 t) - g(o>s<t> I if X = S (t) 1 I ( 4)


w+A-H(s(t) 1 t) ~- g(x-s(t))s(t) 1 if X > S (t) 1 (~ = 1)
It is possible for these equations to be satisfied if
H is large where debonding has already taken place and H is
small where the bonds are still intact. Thus line debonding
is possible.

It is also possible to prove the following theorem.

THEOREM 10.2. If w+A > 0 (in particular if A = 0) both


line and surface bonding are possible.
220 M. Fremond

Proof. Equation (3) can be satisfied with 0 ~ o. Equa-


tions (4) can be satisfied with s < 0.

10.6. An experimental and theoretical result

Experiments have shown that line debonding can have a


nonlinear viscous behaviour. The analogue of equation (3) for
a Norton-Hoff [5) behaviour at the point s(t) is

w-H(s(t),t) = - (g(O))P/ 2 (s(t))P 1 , (s>O).

This relation was experimentally established with


p 1.6 for glass-polyureth ane contact [1].

References

[1] M. BARQUINS.- sur le pelage spontane des elastomeres,


C.R. Acad. Sci., Paris, Serie II, 298, 1984, p. 725-730.

[2] M. D. BUI.- Mecanique de la rupture fragile. Masson, Pa-


ris, 1978.

[3) M. FREMOND.- Contact unilateral avec adherence. Procee-


dings on the Second Meeting on Unilateral Problems in
Structural Analysis, Ravello (Italie), September 22/24/1983,
Edited by G. del Piero, F. Maceri) Springer-Verla g, New Y-
ork, 1985.

[4] M. FREMOND. -Adherence des solides. Journal de Mecanique


Theorique et Appliquee, vol. 6, N3: 1987, p. 383-407.

[5) A. FRIAA.- La loi de Norton-Hoff generalisee en plastici-


te et viscoplasticit e. These d'Etat, Universite Pierre et
Marie curie, 1979.

[6] P. GERMAIN.- Mecanique des milieux continus, Masson, Pa-


ris, 1973.
Contact with Adhesion 221

[7] B. HALPHEN, C.S. NGUYEN.- Sur les lois de comportement


elastoviscoplastique a potentiel generalise. c.R. Acad.
sci., Paris, Serie A, 277, (1973), p. 319-321.

(8] D. MAUGIS, M. BARQUINS.- Fracture Mechanics and Adherence


of Viscoelastic solids, in Adhesion and Adsorption of
Polymers, Part A, L.H. Lee, ed., Plenum, New York, 1980.

(9] D. MAUGIS.- Adherence of Solids, in Macroscopic Aspects


of Adhesion and, Lubrification, J.M. Georges, ed. Elsevier,
Amsterdam, 1982.

(10] J.J. MOREAU.-Fonctionnelle Convexes. Seminaire sur les


equations aux derivees partielles. College de France,
Paris, 1966.

[11] J.J. MOREAU.- Fonctions de resistance et fonctions de


dissipation. Seminaire d'Analyse Convexe, Montpellier, 1971.
APPROXIMATION OF CONTACT PROBLEMS.
SHAPE OPTIMIZATION IN CONTACT PROBLEMS

J. Haslinger
KFK MFFUK, Praha, Czechoslovakia

ABSTRACT
The main part of this contribution deals with the approximation of variational in-
equalities, with special emphasize to contact problems. Finite element technique is
used, starting from different variational formulations (primal, mixed). More details,
concerning the approximation and numerical realization of contact problems can be
found in [9]. Second part of this contribution is devoted to the optimization of the
shape of contact zone of an elastic body, unilaterally supported by a rigid founda-
tion, in order to obtain an even distribution of normal forces along contact part (for
more details see [10]).
224 J. Haslingcr

1. PRIMAL, DUAL AND MIXED VARIATIONAL FORMULATION OF ELLIP-


TIC INEQUALITIES
Let V be a real Hilbert space equipped with the norm Jill and a scalar product.
denoted by ( , ). Let V' be the dual space over V with duality pairing denoted by
( , ). Let a: V x V-+ R1 be a bilinear form, which is
- continuou.! on V:

3M= const. > 0 Vu,v E V ia(u,v)l:::; Mllullllvll ;


- V- elliptic on V:

3a = const. > 0 Vv E V a(v, v) ~ aJivll 2 ;

- symmetric on V:

a(u,v) = a(v,u) Vu,v E V.

Finally let j : V -+ R1 be a convex, continuous and nonnegative functional and set

L(v) = J(v) + j(v) Vv E V,

where
1
J(v) = 2a(v,v)- (J,v), f E V'

is a quadratic functional.
Now, let K 'f= 0 be a closed, convex subset of V. We shall assume the following
minimization problem
find u E K such that
{ (P)
L(u):::; L(v) Vv E K.
Using classical arguments of calculus of variations and assumptions, concerning L,
one has
THEOREM 1.1. There exists a unique solution u of(P). Moreover, this solution can
be equivalently characterized through the relation

uEK: a(u,v-u)+j(v)-j(u)~(J,v-u) VvEK. (1.1)

REMARK 1.1. Let A: V 1-+ V' be the linear, continuous mapping defined by

a(u,v) = {Au,v) Vu,v E V.

Then u E K solves (1.1) iff f- Au E oj(u), where oj(u) is the subgradient of j at


u.
Approximation of Contact Problems 225

REMARK 1.2. In our next considerations, L will play the role of the total poten-
tial energy of the system, V the space of virtual displacements and ]{ the set of
kinematically admissible displacements.
Sometimes it becomes that formulation (P) is not suitable, sometimes from
theoretical, sometimes from practical point of view. In these cases, reformulation of
(P) is useful. Below we describe one way how to do it, making use of the saddle-point
approach. Using the form of L, one can rewrite (P) in the following form:

inf. L( v) = inf sup .C( v, J.L) , (1.2)


vEh vEA JLEB

A ~ V, B ~ Y with a suitable choice of A, B (Y being another Hilbert space) and


.C : A x B ~ R 1 . Here is a typical example of such a reformulation.
EXAMPLE: (releasing of constraints)
Let V X
<I> : A~ R 1 , where A is a convex cone with vertex at B (zero element of Y),
A~ Y and
<P(v,pp) = p<P(v,J.L) Vp ~ 0, V[v,J.L] E V x A

(i.e. <I> is positively !-homogeneous in the 2nd component).


Now, we shall suppose that the following characterization of]{ holds:

LEMMA 1.1. It holds

if v E ]{
I(v) =sup <P(v,J.L) ={ 0
JLEA +oo if v tt I<,
1.e. I is the indicator function of I<.
PROOF:
1 Ifv E I<, then <P(v,J.L) ~ 0 VJ.L E A and <P(v,B) = 0 so that I(v) = 0.
2 Let v tt I<. Then there exists j1 E A such that

<P(v,jl) = c > 0.

But
I(v) ~ <P(v,pfl) = p<P(v,Jl) = pc ~ +oo, p ~ oo.
D

It is readily seen that one can formally write

inf L(v) inf sup{L(v) + <P(v,J.L)},


= vEV (1.3)
vEK JLEA
226 J. Haslinger

i.e. C(v,p,) ~f L(v) + <I>(v,p,), A= V, B =A.


REMARK 1.3. The functional Lis possibly non-differentiable, due to the presence of
j. Similar approach, namely dualization of j can be used in order to achieve smooth
formulation of (P). This will be used in the case of contact problems with friction.
According to the definition of (P) and (1.2), the problem

inf sup ( v, J.L) = inf. L( v) ( 1.4)


vEA p.EB vEk

will be called primal, while


sup inf ( v, J.L) ( 1.5)
p.EB vEA

will be dual to (1.4).


Let
S(p,) = inf ( v, p,) .
. vEA

Then S : B --+ R 1 is called dual functional to L and J.L the dual variable to v.
An interesting question arises, namely: what relation holds between the primal
and the dual formulation. This relation can be established on the basis of the saddle-
point theory, some results of which we present below (for details see [1 ]).
Let 1{ : A x B --+ R 1 be a general functional.
DEFINITION 1.1. A point {u, A} E A X B is said to be a saddle point of 1{ on A x B
if and only if
1i(u,p,):::; 1i(u,A):::; 1i(v,>..) (1.6)
holds for any v E A, J.L E B.
An equivalent characterization of a saddle point is given by
THEOREM 1.2. {u, A} is a saddle point of1i on Ax B if and only if

1i(u,A) =minsup1i(v,p,) =max inf1i(v,p,). (1. 7)


vEA p.EB p.EB vEA

PROOF:
1 Let v E A, J.L E B be arbitrary. Then

inf 1i( v, J.L) :::; 1i( v, J.L) =::::}


vEA
sup inf 1i( v, J.L) :::; sup 1i( v, J.L) Vv E A ,
p.EB vEA p.EB

from which
sup inf 1i(v,p,):::; inf sup1i(v,p,). (1.8)
p.EB vEA vEA p.EB
Approximation of Contact Problems 227

Let {tt,).} E Ax B be a saddle point of 'H on A X B. Then


inf sup 'H( v, J-l) ::; sup 'H( tt, J-l) = 'H( u, ).)
vE.41lEB 11EB

= inf 'H(v,A) :<;sup inf 'H(v,J-l).


vEA llEBvEA

Comparing this with (1.8) we see that


'H(tt,A) =sup inf 'H(v,J-l) = inf sup'H(v,J-l). (1.9)
IEB vEA vEA llEB

But
'H( u, ).) = inf 'H( v, ).) = S().) = sup [inf 'H(v, J-l)] = sup S(J-l) . (1.10)
vEA llEB vEA llEB

Analogously
'H(tt,A) = sup'H(tt,J-l) = inf sup'H(v,J-l) :S sup'H(v,J-l).
11EB vEAilEB 11EB

This, together with (1.9) and (1.10), leads to (1.7).


2 Contrary, let {u, ).} be a point from Ax B, satisfying (1.7). Then
max inf 'H(v,J-l) = inf 'H(v, ).) :S 'H(u, ).)
11EB vEA vEA

and
minsup'H(v,J-l) = sup'H(u,J-l) ~ 'H(u,A)
vEA IEB 11EB

so that
'H( u, J-l) :<; sup 'H( u, J-l) = 'H( u, ). ) = inf 'H( v, ). ) :S 'H( v, ). )
11EB vEA

holds for any J-l E B, v EA. 0

CONSEQUENCE 1.1. Let us assume the situation given by Example. Then the first
component of a saddle point of C on V X A belongs to I< and it is a solution of the
constraint minimization problem:
u E I< : L( u) :<; L( v) \lv E I< ,
while ). E A is a solution of the dual problem:
S().) = inf C(v,A) =sup inf C(v,J-l) = supS(J-l).
vEV IlEA vEV IlEA

Next we prove some useful results concerning the properties of saddle points,
their existence, eventually uniqueness.
To this end we shall impose the following supplementary conditions, concerning
A, B and 'H: namely let A, B be closed, convex and
(j) VJ-l E B, v ~--t 'H(v,J-l) is convex and weakly lower semicontinuous;
(jj) Vv E A, J-ll--t 'H( v, J-l) is concave and weakly upper semicontinuous.
As far as the existence of a saddle point is concerned, we have
228 J. Haslinger

THEOREM 1.3. Let assumptions (j), (jj) concerning 'H be satisfied and let A, B be
moreover bounded. Then there exists at least one saddle paint of 'H on A x B.
If A and B are not bounded, one can use
THEOREM 1.4. Let (j), (jj) be satisfied and moreover let

ii) 3J.Lo E B such that lim 'H( v, J.Lo) = +oo


llvll--+oo
vEA
(coerciveness with respect to the first component)
~) 3vo E A such that lim 'H( vo, J.L) = -oo .
IIPII-oo
pEB

Then there exists at least one saddle point of'H on Ax B.


REMARK 1.4. If only one of the sets A, B is bounded, then the coercivness of 'H
with respect to unbounded set has to be required.
The most general result, which we shall need, is given in
THEOREM 1.5. Let (j), (jj) be satisfied. Moreover, let
- A be bounded or there exists J.Lo E B such that

lim 'H( v, J.Lo)


llvll--+oo
= oo
vEA

- B be bounded or

lim inf 'H(v,J.L)---+ -oo.


IIJLII--+oo vEA
pEB

Then there exists at least one saddle point of 'H on A x B.


In what follows, we present one very important application of the previous re-
sults. Let J : V ---+ R1 be a quadratic functional introduced at the beginning of
this chapter. We define the closed, convex set K ~ V as the kernel of linear and
continuous mapping B : V ~---+ Y (Y another Hilbert space), i.e.

K={vEV!Bv=O}.

Finally, let 'H : V X Y' ---+ R 1 be a functional given by

'H(v,J.L) = 21 a(v,v)- (f,v} + [J.L,Bv],


where J.L E Y' and [, ) denotes the duality pairing between Y' andY.
Approximation of Contact Problems 229

T-HEOREM 1.6. Let there exist a constant (3 > 0 such that

(1.11)

Tl1en there exists a unique saddle point of'H on V x Y'.


PROOF:
1 Uniqueness. Uniqueness of the first component follows from strict convexity of
the mapping v ~-+ 'H(v,Jt), v E V. Let {u,>.}, {u,>.} E V x Y' be two saddle points
of 'H on V x Y' or equivalently

a(u,v) + [>.,Bv] = {f,v}


a(u,v)+[>.,Bv]={f,v} VvEV.
Subtracting the second equation from the first one, we get

[>.- >., Bv] = 0 Vv E V


[>.- >., Bv]
-:::} ~~e II vii =0
===? >. = >.

as follows from ( 1.11).


2 Existence. We shall apply the results of Theorem 1.5. First of all

lim 'H( v, fl)


!lvll--+oo
= +oo (1.12)
vEV

holds for any fl E Y'. We have to prove that

lim inf 'H( v, fl) = -oo .


ll~tll--+oo vEV
pEY'

Let fl E Y' be fixed and let us solve the problem

find u = u(ft) E V such that


{
'H(u,ft) = vEV
inf 'H(v,ft) (1.13)

or
find u E V such that
{ (1.13')
a(u,v) + [ft,Bv] = {f,v} Vv E V.
Let A be the mapping V ~-+ V' defined by means of

a(u,v) = {Au,v}.
230 J. Haslinger

From Lax-Milgram theorem it follows that A maps V onto V' and

If B* denotes the adjoint mapping to B, the solution u of (1.13) can be written:

Let H:V ~----+ R1 be given by

H(v) = (f,v)- [J.L,Bv].

As a(u,u) = H(u), one has

1 1
'H(u,J.L) = - 2a(u,u) = - 2(Au,u)
= -~(f,A- 1 f)+ ~(f,A- 1 B*J.L) + ~(B*J.L,A- 1 f)- ~(B*J.L,A- 1 B*J.L) (1. 14 )

~ -c1IIB*J.LII~, + c2IIB*J.LIIv, + ca
with Ct, c2, ca > 0.
From (1.11) it follows that

sup -[J.L,Bv]
- - =sup [B*J.L,v]
II II = liB* J.L I V' 2: f3 II J.L I Y' .
vEV 11 V 11 V vEV V V

Finally, this and (1.14) yield

if IIJ.LIIY'
oo. All assumptions of Theorem 1.5 are satisfied.
-t D

Let : Ax B ~----+ R1 be a given functional, which is related to the problem (P)


by means of (1.2).
DEFINITION 1.2. The problem of finding a saddle point of on A X B will be called
a mixed formulation of (P).
Approximation of Contact Problems 231

2. APPROXI~fATION OF VARIATIONAL Il\'EQ\.."ALITIES

2.1. Approximation of primal formulation


Let {!1.- h}, h -+ 0+ be a family of comex, closed subsets of V, not necessarily
imbedded into E. Instead of (P) we shall assume a family of minimization problems:

Recall that under our assumptions (Ph) has a unique solution tth for any h > 0.
REUARK 2.1. As in continuous case. llh E J{h sohing (Ph) can be equiYalently
characterized through the relation:

A natural question arises: what is the relation between u and tth. First we prove
LEMMA 2.1. Let u, uh be solutions of(P), (Ph), respectively. Then

o:llu- tthll 2 + (f,uh- v)


-:5,(f,u- vh)
+ a(uh- u,vh- u) + a(u, v- tth) + a(u,vh- u) (2.2)
+ j(v)- j(u) + j(vh)- j(uh)
holds for any v E K, vh E I\h.
PROOF: Let v E I\, Vh E J{h Using l'-ellipticity of a we can write:

aiiu- uh 11 2 -:5: a( u- tth, u- uh)


=a( u, u) +a( tth, tth)- a( u, uh)- a( uh, u)
s; a(u, v) + (f, t t - v) + j(v)- j(u) + a(uh, vh)
+ (f,uh- vh) + j(vh)- j(uh)- a(u,uh)- a(uh,u)
= (f, U- Vh) + (f, Uh - v) +a( tt, V- Uh)

+ a(uh- u,vh- u) + a(u,vh- u)


+ j(v)- j(u) + j(vh)- j(uh).
0
REMARK 2.2. If J{h C J{ Vh, then we can set v = uh in (2.2) and we obtain:

aiJu- uhJJ 2-:5:(f,u- vh) + a(uh- u,vh- u) + a(u,vh- u) + j(vh)- j(u) (2.3)
Vvh E I<h .

REMARK 2.3. (2.2) and (2.3) will be used for estimating the discretization error
JJu- uhJJ.
Concerning the convergence of uh to u, the following result will be needed:
232 J. Haslingcr

THEORDI 2.1. Let the following conditions will be satisfied:

Vv E J( 3vh E ]{h such that Vh ___. v, h ___. 0+ ; (2.4)


Vh E J(h , Vh __.. v (weakly) implies v E J{. (2.3)

Then
llu- uhll ___. 0 , h ___. 0 + .
PROOF: From the definition of L and (2.4), the boundness of {uh} follows, i.e.

3c > 0 such that lluhll:::; c Vh > 0.

As V is a Hilbert space, one can find a subsequence {Uh'} C {uh} such that

Uh' __.. u* E J{

because of (2.5 ). Let us prove that u* is the solution of (P ). Let v E J{ be given.


Then there exists a sequence {vh}, vh E J( h such that

( 2.G)

(see (2.4)). From the definition of (Ph'):

L(uh'):::; L(vh') . (2.7)

Passing to the limit in (2. 7) and using weak lower semicontinuity of L, we arrive at

L(u*):::; L(v) Vv E K, (2.8)

i.e. u* solves (P). As u* is unique, the whole sequence {uh} tends weakly to 1t = u*.
In order to prove that Uh ___. u, h ___. 0+ in V, we use (2.2). Let fh E J(h be such
that
Vh ___. u , h ___. 0+ (2.0)
and substitute v := u. Using the fact that uh __.. u, h ___. 0+ as well as the weak
lower semicontinuity of j and (2.9), we arrive at the assertion. o

2.2. Approximation of mixed formulation of elliptic inequalities


Here we shall assume a special case of functional : V x Y ~---+ R 1 ,

(v,JL) = J(v) + b(v,JL)- [g,J.L], (2.10)

where J is a quadratic functional, given by a symmetric, V-elliptic and continuous


bilinear form a, b : V X Y ___. R 1 is a continuous bilinear form, g E Y' and [g, {t]
denotes the value of g at Jl The norm in Y will be denoted by I 1. Finally, let
J( ~ V, A ~ Y be two non-empty, closed convex subsets (for details see [2]). We
make the following assumptions, concerning]{ and A: A is either
(CC) a convex cone with vertex at B and]{= V
Approximation of Contact Problems 233

or
(BC) a bounded convex subset of Y.
Let {u, >.} be a saddle point of .C on ]{ x A, i.e .

.C(u,p,)::; .C(u,>.)::; .C(v,>.) Vv E K, Vp, EA.

In accordance with the theory presented in Chapter 1, the first component u E ]{


solves the problem
J (u) + j ( u) = inf {J (v) + j ( v)} ,
vEK

where j(v) = sup{b(v,p,)- [g,p,]}.


A
Now, let h, H-+ 0+ be two parameters, tending to zero. With any couple h, H
we associate finite dimensional subspaces Vh C V and YH C Y, respectively. Let J(h
and AH be closed, convex subsets of Vh and Y H, respectively. Similarly as in the
continuous case we make the following assumptions: AH is either
(CCH) a convex cone with vertex at(} and Kh = Vh
or
(BCH) a convex subset ofYH, bounded uniformly in Y, i.e. there exists a positive
constant c such that

By the approximation of (P) we mean the problem of finding a saddle-point


{uh, AH} of .Con Kh X AH:

find {uh, AH} E Kh X AH such that


{
.C(uh,P,H) :S .C(uh,AH) :S .C(vh,AH) Vvh E Kh, Vp,H E AH,

or equivalently

find {uh, AH} E Kh x AH such that


{ a( uh, Vh - uh) + b( Vh - uh, AH) ~ (!, Vh - uh)
b(uh,P,H- AH) :S [g,P,H- >.H] Vp,H E AH.

Analogously to the continuous case, the first component Uh E Kh minimizes the


functional J(vh) + jH(vh) over Kh, where iH(vh) = sup{b(vh,P,H)- [g,p,H]}.
AH
As far as the existence, eventually the uniqueness of {uh, AH} is concerned, we
can directly apply results of Chapter 1. We shall distinguish both cases (CCH) and
(BCH) separately.
234 J. Haslinger

THEOREM 2.2. If(CCH) and b(vh,J.iH) = 0 Vvh E Vh ===} J.iH = 0, then there
exists a unique solution of (PhH ).
If { AH} are uniformly bounded, situation is simpler. One has
THEOREM 2.3. Let (BCH) be satisfied. Then there exists a solution of (PhH ), the
first component of which is uniquely determined.
Our aim will be to establish relations between uh, u and AH, >-.. To this end we
give another, equivalent formulation of (f5).
Let 'H = V x Y be a Hilbert space, equipped with the norm

A : 'H X 'H r-t R 1 a bilinear form defined by

A(U, V) = a(u,v) + b(v,>-.)- b(u,p,), U = (u,>-.) E 'H,


V=(v,p,)E'H,

and F : 'H ---+ R 1 a linear functional

(F, V) = (f,v)- [g,p,], V = (v,p,) E 'H.

The definition of A immediately implies

A(V, V) = a(v,v) VV = (v,p,) E 'H; (2.11)


3M= const. > 0 : IA(U, V)l ~ MIIUIIHIIVIIH VU, V E 'H.
(2.12)

It is readily seen that (f5) is equivalent to

find U = ( u, ).. ) E JC = f{ x A such that


{
A(U, V- U) 2:: (F, V- U) VV E K .

Next, let JChH = f{h x AH be a closed, convex subset of 'H; JChH ct. JC, in general.
The problem

find uhH = { Uh' AH} E JChH such that


(:PhH)
A(UhH, vhH - uhH) 2:: (F, vhH - uhH) vvha E KhH

represents the approximation of (P), equivalent to (P~H) (or (PhH) ).


First we prove an auxiliary lemma.
Approximation of Contact Problems 235

LEMMA 2.2. Let {u,.X}, {uh,AH} besolutionsof(P) and(PhH), respectively. Then

llu- uhll 2 ~c[{llu- vhll 2 + I.X- JlHI 2} + A1(vh)


+ A2(v) + {b(u, AH- Jl)- [g, AH- Jl]} (2.13)
+ {b(u,.X- JlH)- [g,.X- JlH]} + I.X- AHI 2 )
holds for any Vh E I<h, v E K, JlH E AH, Jl E A, where
A1(vh) = a(u,vh- u) + b(vh- u,.X) + (f,u- vh)
A2(v) = a(u,v- uh) + b(v- uh,.X) + (f,uh- v)
and cis positive constant independent of h, H.
PROOF: By virtue of (2.11) and the definition of (P) and (PhH) we get:
allu- uhll 2 ~ A(U -UhH,U -UhH)
= A(U,U)- A(UhH,U)- A(U,UhH) + A(UhH,UhH)
~ {:F,U- V) + A(U, V) + (:F,UhH- VhH) + A(UhH, VhH)
- A(UhH,U)- A(U,UhH)
= (:F,U- VhH) + {:F,UhH- V) + A(U, V- UhH) (2.13')
+ A(UhH - U, VhH - U) + A(U, VhH - U)
= At(vh) + A2(v) + {b(u, AH- p)- [g, AH- p]}
+ {b(u, A- JlH)- [g, A- J.lH]}
+a( uh - u, Vh - u) + b( Vh - u, AH -.X) - b( Uh - u, J.lH- .X) .

The boundedness of a, b together with the inequality 2hf:::; !h 2 + c:P yield (c > 0)
c
allu- uhll2
~ At(vh) + A2(v) + {b(u, AH- Jl)- [g, AH- Jl]}

+ {b(u,.X- JlH)- [g,.X- J.LH] + Mtcllu- vhll 2 + &11u- vhll 2 (2 14)


c
+ Mc2 llvh- ull 2 + M2c:I.XH- .AI 2 + M2c:llu- uhll 2 + Mc2I.A- J.LHI 2
For c: > 0 sufficiently small, we arrive at (2.13). o
In order to prove the relation between .A and AH we shall need the following
assumption:
3p
> 0 independent on h, H > 0 such that
{ b(vh,J.tH) ~ (S)
s~,; llvhll ~ .81~-tHI holds for any J.lH E YH .
A direct consequence of Lemma 2.2 is
244 J. Haslinger

THEOREM 3.2. There exists a unique solution u of (P ). Moreover, the solution can
be equivalently characterized through the relation

uEK: (r(u),c:(v-u))+ { g(jvti-iuti)ds


irK (3.19)
~ (F;, v;- u;)o + (P;, v;- u;)o Vv E K .
REMARK 3.1. If g:::: 0 on rK, (3.18) reduces to

Tt(x) = 0 on rK' (3.18')


which corresponds to frictionless case.

3.4. Formulation of contact problems in terms of stresses


Let S be a space of 2 x 2 symmetric matrices: .

and set
N;j = C:ij(v). (3.20)
Then, using (3.20), we can formally write

with notation of Section 3.3. We may look at (3.20) as constraints, which will be
removed by introducing the following lagrangian:

H([N,v],-X) =1it(N,v)+ In A;j(t:;j(v)-N;j)dx,


{[N,vj,-X} E S x K x S.
It is easy to see that

if N = c:(v)
otherwise.
Thus
inf ,J( v) = inf sup H([N, v], ,\) . (3.21)
vEK vEK >.ES
NES

Dual problem to (3.21) now reads:

sup inf H([N, v], -X) . (3.22)


>.ES J!e~

Using a direct calculation, we obtain


Approximation of Contact Problems 245

LEMMA 3.1. There exists a unique saddle-point ([N, v], 5.) of 1{ on S x K x S.


Moreover
N = c:(u), v = u, 5. = r(u),
where u E K is a solution of primal formulation (P ).
Denote by
S(-\) = inf H([N, v], >.)
vEK
NES

the dual formulation to ,J. One can easy prove that

where aijkl are coefficients of inverse Hooke's law. Moreover

supS(-\)= sup S(>.),


>.es >.eKJ;,P,g(n)

with ICJ;,P,g(n) given by

Kj; Pg(n) ={A E s


' '
I r
~
AijC:ij(v)dx + j(v) 2: (f,v) Vv E K}.

Instead of S we introduceS= -S.


The explicit form of the dual problem to (P) now reads

find,\* E Kf,P,g(f!) : S(>.*) ~ S(>.) V>. E Kf,P,g(Q) . (P*)

As S is a quadratic functional, one immediately has:


THEOREM 3.3. There exists a unique solution,\* of (P* ). Moreover

,\* = r(u) ,

where u E K is the solution of the primal formulation (P ).


REMARK 3.2. Using integration by parts, we can show that ,\ E K"J;,P,g(f!) if and
only if
inn' i = 1,2;

-\;ini = P; on rp ' i = 1, 2; (3.23)


A;jnjni ~ 0 on rK;
i>.;jnjt;i ~ g on rK.
238 J. Haslinger

By virtueof(2.23), (2.24), u* E K, A* EA. Let us show that {u*,A*} is a solution of


(P). Let {v, J.L} E K X A be an arbitrary element. From (2.21 ), (2.22) the existence
{ vh}, {J.LH }, Vh E Kh, J.lh E AH such that
vh ---+ v in V , J.l H ---+ J.l in Y (2.26)

follows. {uh', AH'} being solution of (Ph' H') (or (Ph' H')) satisfies

a( uh', Uh' - Vh') + b( uh' - Vh', AH') ~ (!, uh' - Vh') Vvh' E Kh'
(2.27)
b(uh',J.lH' -AH') ~ [g,J.lH'- AH'] 'VJ.LH' E AH'. (2.28)
Passing to the limit with h', H'---+ 0+ in (2.27) and using (2.25), (2.26) we obtain
a( u * , u * - v) + lim inf b( u h' , AH' ) - b( v, A* ) ~ (!, u * - v)
h',H'-->0+ (2.29)
Vv E K.
The same procedure can be applied to (2.28):

b(u*,J.L)- [g,J.L- A*]~ liminf b(uh',AH') 'VJ.L EA. (2.30)


h',H'-->0+
Setting J.L =A* in (2.30), we obtain

b(u*,A*) ~ liminf b(uh',AH'). (2.31)


h',H'-->0+
Substitution of (2.31) into (2.29) yields:
a(u*,u*- v) + b(u*- v,A*) ~ (f,u*- v) Vv E K.
The choice v = u* in (2.29) implies

liminf b(uh',AH') ~ b(u*,A*).


h',H ->0+
1

From this and (2.30), we have

b(u*,J.L- A*)~ [g,J.L- A*] 'VJ.L EA.


Thus {u*, A*} is a solution of (P). By virtue of its uniqueness, the whole sequences
{ uh}, {AH} tend weakly to u, A, respectively. Let us show that uh ---+ u strongly in
V. Let {vh}, Vh E Kh, {PH}, PH E AH be such that

Applying (2.20) with v = u, J.l = A, Vh = vh, J.LH = flH and using the weak conver-
gence uh ____. u, AH ____. A, we obtain uh ---+ u in V. D
REMARK 2.6. If Kh C K, AH C A, the conditions (2.23), (2.24), respectively, are
satisfied.
Approximation of Contact Problems 239

THEOREM 2.7. Let (CC), (CCH) and (S) be satisfied. Let {u,.\} be the unique
solution of (P). Moreover, let us suppose that
Vv E v 3vh E vh : Vh ~ v in v ; (2.32)
Vp, E A P,H ~ p, in L ;
3p,H E AH : (2.33)
Ji.H E AH , Ji.H ~ p, in L implies p, E A ; (2.34)
there exist a real number d, a positive number c and a (2.35)
bounded sequence {ih}, Vh E Vh such that j H( Vh) ~ d
Vvh E Vh, Vh, HE (0, 1), ]H(vh) :S c Vh, HE (0, 1) .
Then Uh ~ u, AH ~ .\.
PROOF: We shall prove the boundedness of {uh} and {.\H} only. The rest of the
proof is analogous to that of Theorem 2.6. The convergence of AH to .\follows from
(2.16).
According to the interpretation of (PJ.H ), Uh E vh satisfies
a(tth,Vh- Uh) + JH(vh)- JH(uh) ~ (f,vh- Uh) Vvh E vh 0

Hence
a(tth,uh) + JH(uh) :S a(tth,vh) + JH(iih)- (f,vh- uh).
This and (2.35) imply the boundness of {uh} and by virtue of (2.16) we deduce the
boundedness of {.\H }. D
REMARK 2.7. If AH C A VH E (0, 1), (2.34) is automatically satisfied.
Condition (S), guaranteeing the convergence of AH to .\, is very restrictive. If we
are interested in the convergence of uh to u, condition (S) may be avoided. Indeed
let us suppose that the functions
j(v) = sup{b(v,p,)- (g,p,]}
A
JH(vh) = sup{b(vh,Ji.H)- [g,p,H]}
AH
take their values from the set {0, +oo }. We shall denote by
K={vEVIj(v)=O}
JChH = {vh E vh I JH(vh) = 0}
'
i.e. j and j H are the indicator functions of the closed convex sets JC and JChH,
respectively. Let {u,.\} E V x A and {uh,.\H} E Vh x AH be solutions of (P) and
(PhH ), respectively. Following the interpretation of these problems we already know
that u E K and Uh E JChH are solutions of the minimizing problems:
J(u) :S J(v) Vv E JC
and.

respectively. As far as the convergence of uh to u is concerned, we have


240 J. Haslinger

THEOREM 2.8. Let (CC), (CCH) be satisfied and there exist solutions {u,>.} and
{ uh, AH} of ('P) and (PhH ), respectively, the first components of which are uniquely
determined. Let

Vv E IC 3vh E IChH Vh -+ v in V ; (2.36)


Vh E IChH , Vh ___., v in V implies v E IC . (2.37)

Then uh-+ u in V.

PROOF is a direct consequence of Theorem 2.1.

3. PRIMAL, DUAL AND MIXED VARIATIONAL FORMULATION OF CON-


TACT PROBLEMS

3.1. Basic notations


We begin with some basic concepts of the theory of linear elacticity in two
dimensions (see [3]). Practical situations of three dimensions can often be reduced
into two-dimensional cases by symmetry, etc.
Let us consider a two-dimensional elastic body represented by a bounded domain
n c R 2 subjected to internal and external forces. These forces cause the body to
deformate so that the point x, x = (x1, x2) of the undeformed body becomes the
point x', x' = (xLx~) of the deformed body; x' can be written as x' = x + u(x),
where u = (u1,u2) denotes the vector of deformation.
The strain caused by displacements u is characterized by the so called strain
tensor 1
c = { Cij ( u )} 2i '1=1, Cij
(
u
)
= -2l(aui
-a
Xj
aui) .
+ -a
Xi
(3.1)

Similarly, the stresses can be characterized by the so called stress tensor (sym-
metric)
(3.2)

In the equilibrium state the sresses r are related to the volumic forces F =
(F 1 , F2) by the system of equilibrium equations

inn. (3.3)

1 Here we assume the linearized strain tensor.


Approximation of Contact Problems 241

In the case of linear elasticity the stress and strain tensors are connected by the
linearized Hooke's law 1
'Tij(u) = Cijk/Ck/(u)

=L
2
(3.4)
Cijk/t:kl(u),
k,l=l

where Cijkl are the elasticity coefficients satisfying symmetry conditions

Cijk/ = Cjikl = Ck/ij Vi,j,k,l=1,2 a.e.inn. (3.5)

In the case of homogenous and isotropic material Hooke's law reads

0) (11)
0 t:zz , (3.6)
J.1 12

where A and J.1 are Lame's coefficients.


By multiplying (3.3) with v and integrating both sides over n and applying
Green's formula we obtain for u, v E (H 1 (n))2

(3.7)

J J J
so that
'Tij(u)t:;j(v)dx- 'Tij(u)njv;ds = F;v;dx, (3.8)
n an n
making use of (3.3). Here n = (n 1 ,n 2 ) denotes the outward normal unit vector to
on.
To guarantee the uniqueness of the displacement field u, we impose the condition
u = 0 on a given non-empty part r u open in On the remaining part r P of the
on.
boundary the surface tractions P = (P1 , Pz) are given, i,e. T; r;ini = P;, i = 1, 2. =
We denote by V the space of admissible displacements:

v= {v = (vl,vz) E (H 1 (n)? I Vj = 0 on ru, i = 1,2}. (3.9)

For any v E V the equation (3.8) yields:

J
n
'Tij( u )t:ij( v) dx = J
n
F;v; dx + J
rp
P;v; ds =(J, v) . (3.10)

1 In the sequel this summation convention will be used.


242 J. Haslinger

In order to obtain the existence and uniqueness of the solution of (3.10), it is


necessary to prove the V-ellipticity of the bilinear form a(,), given by a(u,v) =
=(
JTij( u )t:;j(v) dx r( u ), c:( v )), i.e.
n
31 = const. > 0 : a(v,v) ~ 1llvll~,n \:/v E V, (3.11)

where llvi!I,n := llvtll~ n + llv21i~ n


' '
Assuming that the elasticity coefficients satisfy the algebraic ellipticity condition
3qo > 0 : Cijkl( X )eijekl ~ qoeijeij \;/ eij = eji E Rn (3.12)
a.e. inn, then (3.11) follows from Korn's inequality

3C > 0 : j C:;j(v)t:;j(v)dx ~ Cllvlltn \:/v E V. (3.13)


n

3.2. Classical formulation of contact problems


Let us assume a structure, composed of two deformable bodies in mutual con-
tact and submitted to a body force F and surface tractions P on a part of the
boundary. The aim is to determine the behaviour of this structure. The problem
is non-classical in the sense that the partition of the boundary into the contact and
non-contact zones is not known a-priori as well as reactive forces between bodies.
The deformation of each body depends not only on given applied forces but also on
contact tractions along contact surfaces. There are many applications of this prob-
lem in several branches of engineering, e.g. in soil and rock mechanics, in mechanical
engineering etc. To simplify the explanation, we restrict ourselves to the study of a
contact problem between a plane elastic body and a rigid foundation.
Let a body be represented in its non-deformable state by a bounded domain
n, the Lipschitz boundary of which is decomposed as follows: an = f' u U f' p u f' K,
where r Ul rp and rK are non-empty parts of an, open in an. By a classical
solution of a contact problem we mean a displacement field u = (Ut, u 2 ), which is in
the equilibrium state with F and P, i.e.:

(equilibrium equations):
or
a (u) + F; = 0
I) inn i = 1,2; (3.14)
Xj
and it satisfies the following set of boundary conditions:
T;j(u)nj = P; on rp ; (3.15)
u; = 0 on r u i = 1, 2; (3.16)
Tn(u) = Tij(u)n;nj ~ 0' Un = u. n ~ 0' Tn(u)un =0 on rK ;
(3.17)
ITt(x)l ~ g(x) }
ITt(x)l < g(x) ::::::} Ut(x) = u(x) t = 0 on r/( .
ITt(x)l = g(x) ::::::} 3A(x) ~ 0 such that Ut(x) = -A(x)Tt(x) (3.18)
Approximation of Contact Problems 243

Here T n ( x), Tt ( x) denotes the normal, tangential component of the stress vector
T(u) at a point x, tis the tangential vector to an, g(x) is bounded, measurable and
non-negative function defined on an. (3.17) is a condition of unilateral contact of n
with a rigid support. 1 Influence of friction is modeled by (3.18). We consider a very
simple model, namely the so called "model with a given friction". Later, we shall
briefly discuss the case of coulombien friction.

3.3. Primal variational formulation of contact problems


In order to give a variational form of (3.14)-(3.18), we introduce the following
notations:

v = {v E (H 1 (n)? I v; = 0 on ru, i = 1,2}


K = {v E v I Vn = v . n :::; 0 on r K}
..1(v)=~ [ T;j(v)e;j(v)dx- [ F;v;dx- [ P;v;ds+ [ glvtlds.
2 ln ln lrp lrK
In accordance with notations of Chapter 1,

a(u,v) = fo 'Tij(u)e;j(v)dx = (r(u),c:(v)),


=1n F;v; dx + lrp[ P;v; ds =(F;, v;)o + (P;, v;)o ,
(!, v)

j(v) = { glvtl ds ,
lrK
L(v) = ..1(v) .
K is the closed, convex subset of V of all admissible displacements, ..1 is a total
potential energy functional. Next, we shall assume that F = (F~, F 2 ) E (L 2 (0))2,
P = (P1,P2) E (L 2 (rp)) 2
By a variational8olution we mean a function u E K, satisfying

..1(u) :S ..1(v) VvEK. (P)


The relation between a classical and a variational formulation is given by

THEOREM 3.1. (P) and (3.14)-(3.18) are formally equivalent.

Applying Theorem 1.1, we immediately have

1 Here we assume that n is unilaterally supported by a rigid foundation along the whole r K, i.e.
dist(rK,rigid foundation) =
0 and rK cannot enlarge during deformation.
236 J. Haslinger

THEOREM 2.4. Let (CC), (CCH) and (S) be satisfied. Let there exist a solution
{u,-X} of(P). Then

llu- uhll 2 :=:; c[{llu- vhl[ 2 +I-X -t-tHI 2} (2.15)

+ {b(u,AH -t-t)- [g,AH -t-tl}+ {b(u,-\ -t-tH)- [g,-\ -t-tH]}] ,


I-X- AHI:::; c{llu- uhll +I-X -t-tHI} (2.16)
hold for any Vh E vh, 1-t E A, 1-LH E AH with a positive constant c.
PROOF: Since (CC) and (CCH) are satisfied, J{ = V, J{h = Vh, i.e. J{ and J(h are
linear sets. Therefore
A1(vh) = 0 Vvh E Vh . (2.17)
As J{ =V and Vh C V Vh E (0, 1), we can choose v = Uh in (2.13). Hence
A2(v) = 0. (2.17')
Let 1-LH E AH be arbitrary. Applying (S) we obtain
' b( v h' 1-t H - AH)
,BI-XH -t-tHI :::; s~f llvhll . (2.18)

Now we may write


b(vh,/-LH- AH) = b(vh,/-LH)- b(vh,AH)
= b(vh,/-LH) + a(uh,vh)- (f,vh)
= b(vh, 1-tH) + a(uh, vh)- a(u, vh)- b(vh, >.)
= b( vh, t-t H - ,\) + a( u h - u, vh)
:S c{lt-tH- -XI+ lluh- ull}llvhll
This, together with (2.18) lead to
lt-tH- AHI :=:; c{llu- uhll +I-X -t-tHI} Vt-tH E AH.
Using the triangle inequality
I-X- AHI:::; I-X -t-tHI + lt-tH- AHI Vt-tH E AH '
we obtain (2.16). Finally, replacing the term M2c-I-XH - -XI on the right hand side
of (2.14) by (2.16) and making use of (2.17), (2.17'), we obtain (2.15) for c; > 0
sufficiently small. o
REMARK 2.4. If AH C A for VH E (0, 1), we can insert t-t = AH into (2.15). There-
fore, (2.15) takes the following simpler form:

llu- uhll 2 :Sc[{llu- Vhll 2 + J>. -t-tHI 2 }


(2.19)
+ {b(u,-X -t-tH)- [g,-X -t-tHn] Vvh E vh, t-tH E AH.
Approximation of Contact Problems 237

THEOREM 2.5. Let (BC) and (BCH) be satisfied. Then

llu- uhll 2 ::;c[AI(vh) + Az(v) + {llu- vhll 2 + lA- JLHI 2 }


+ llu- vhll + {b(u,AH- JL)- [g,AH- JL]} (2.20)
+ {b( u, A - JL H) - [g, A - JL H]}]
holds for any Vh E Kh, v E K, JL E A, JLH E AH. Moreover, if K = v, Kh = vh and
(S) is satisfied, then (2.15) and (2.16) hold.
PROOF: We have to prove (2.20) only. As A, AH are bounded in Y,

Hence (2.20) follows due to (2.13'). o


REMARK 2.5. If Kh C K, AH C A, Vh, HE (0, 1), then setting v = uh, JL = AH we
obtain Az(v) = 0, b(u,AH- JL)- [g,AH- JL] = 0.
Next, let us suppose that the pair of real parameters h, H satisfies

h --+ 0+ {:=::=:? H --+ 0+ .

Relations (2.15), (2.16) and (2.20) can be used to estimate the rate of convergence
of Uh to u and AH to A, provided the exact solution is smooth enough. Other
applications are given by the following convergence theorems.
THEOREM 2.6. Let (BC), (BCH) be satisfied and, moreover, let

Vv E K 3vh E Kh : in V ;
Vh--+ v (2.21)
'VJL E A 3jLH E AH : JLH--+ JL in Y; (2.22)
Vh E Kh , Vh __. v (weakly) in V implies v E K; (2.23)
JLH E AH , JLH __. JL (weakly) in Y implies JL EA. (2.24)

Let the solution {u, A} E K x A of (P) be unique. Then

Uh--+ u in V , AH __.A (weakly) in Y.

PROOF: First, {uh}, PH} are bounded. For PH} this follows from (BCH ), for {uh}
from (2.21) and (P~H)z. Hence, there exist a subsequence {uh,AH} C {uh,AH}
and an element {u*, A*} E V x Y such that

(2.25)
246 J. Haslinger

3.5. Mixed formulation of contact problems


From numerical point of view there are two difficulties in treatment of (P),
namely
- (P) is a constrained problem;
- (P) is non-smooth.
To overcome these difficulties, duality approach can be applied. First we present
some auxiliary results, useful in what follows. For the sake of simplicity we assume
that rP = 0 and Hooke's law (3.4) will be written as

r = Ae, (3.4')

where A is the symmetric and positive definite mapping from the space S (see Section
3.4) into itself. By Ai we denote the square root of A. A-1, A-! are the inverse
mappings to A, A!, respectively.
Let V be endowed with the equivalent norm

lllvlll = (Ae(v),e(v)). (3.24)

The set Hi (r K) of all traces of functions, belonging to V, will be equipped with the
norm

The symbol !V denotes the trace of v E V. From this definition, it follows immedi-
ately that

where u E H 1 (D.) is the unique solution of

{ (Ae(u),e(v)) =0 \fv E V,
(3.25)
"{V = cp on r u.

Let 8: V--+ H!(rK) be a mapping defined by the relation 8v = (vn,vt) Finally,


by H( div, D.), we denote the space of symmetric tensors, the divergence of which is
square integrable in n:

.
H (dlv,n ) = { a= (aij )2i,j=l E S,dlV
. a= (8a1j
8xj ' axk 8a2k) E [L 2(n )Jz} .
We equip H( div, n) with the norm

p.
1
llaiiH(div,!l) = {(A-la, a)+ (diva, div a)o (3.26)
Approximation of Contact Problems 247

The closed subspace of H( div, 0) consisting of functions with zero divergence will
be denoted by H 0 (div,f2). Clearly, H 0 (div,f2) is a Hilbert space with norm

(3.27)

The following Green's formulas hold for any r E H( div, 0) and v E V:


(i) there exists a unique mapping T = (T1,T2) E .C(H(div,O),H-~(rK))
such that

(ii) there exists a unique mapping T = (Tn,Tt) E .C(H(div,n),H-~(rK))


such that

Now we describe certain properties ofT and T.


LEMMA 3.2. T maps Ho(div,O) onto H-~(rK)
PROOF. Let r.p* E H- ~ (r K) be given. Then the problem

{ find u E V such that (3.28)


(Ac:(u),c:(v)) = (r.p*,f'v), Vv E V,

has a unique solution u = u(r.p*). On setting a = Ac:(u) and using (i), we get
immediately that a E H 0 (div,f2) and T(a) = r.p*.
The symbolii'P*II-~,rK denotes the usual dual norm, namely

Next we derive some more useful expressions for II'P*II-~,rK


LEMMA 3.3. The following equalities,

hold, where u = u(r.p*) is the solution o(3.28) and a= Ac:(u).


PROOF. Let a E Ho(div,O) satisfy T(r) = r.p*. Then, for any r.p E H-~(rK), (i)
yields

(r.p*,r.p) = (T(r),r.p) = (r,c:(v)) = (A-h,A~c:(v)) Vv E V, f'V = 'P


248 J. Haslinger

Hence
(cp*,cp} ::=; llriiHo(div,n)lllvlll Vv E V, 'YV = cp,
so that
llcp*ll-i,rK ::=; llriiHo(div,n) Vr E Ho(div,n), T(r) = cp*. (3.29)
Let u = u(cp*) be the solution of (3.28) and let a= Ac(u) E Ho(div,n). Then

iiall1to(div,n) = (A- 1 a,a) = (c(u), Ac(u)) = lllu(cp*)lll 2 . (3.30)

By inserting v = u(cp*) into (3.28), using (3.30) and the definition of II lli,rK' we
obtain
(cp*,1u} = (Ac(u),c(u)) = lllu(cp*)IW = iiaiiHo(div,n)lllulll
;::: iiaiiHo(div,n)llrull~.rK
Hence
llcp*ll-i,rK ;::: iiaiiHo(div,n)
This inequality and (3.29) yield the following lemma.
LEMMA 3.4. The following expression holds:

II cp *II -i,rK = s~ (cp*,,v}


lllvlll '
\J
vcp E
* H- 1 (r )
2 K

PROOF. From the definition of II'P*II-1 rK as the dual norm, it follows that
2'

and so
(cp,,v} <II *II (3.31)
s~p lllvlll - cp -i,rK
On the other hand, let u = u( cp*) be the solution of (3.28). Then, for a = Ac( u ), we
get
(cp*,1u} = lllu(cp*)IW = lllu(cp*)lllllaiiHo(div,n)
= lllu(cp*)lllllcp*ll-i,rK'
by virtue of Lemma 3.3. From this, and from (3.31), the lemma follows.
In a similar way to the proof of Lemma 3.2, one can prove
LEMMA 3.5. T maps H0 (div,!l) onto H-i(rK)
Let cp* be an arbitrary element of H-i (r K ). By virtue of Lemma 3.2, there
exists r E Ho(div,n) such that T(r) = cp*. Let us set,.,.= T(r) E H-!(rK) and
write,.,.= (3*cp* with (3*: H-i(rK)--+ n-i(rK) It is readily seen that,.,. does
not depend on the choice ofT E Ho(div,!l) where T(r) = cp*. By comparing Green's
formulas (i), (ii), we obtain
Approximation of Contact Problems 249

(3.32)

In order to transform problem ('P) into an unconstrained and smooth one, we


introduce the following sets of Lagrange multipliers A1, A2:

A1 = H=~(rk) = {111 E H-~(rk)I(JLl,vn} 2::0 Vv E I<},


A2 = {112 E L 2(rK) 111121 :S g a.e. on rK}'

be two closed convex subsets of H- ~ (r K) of Lagrange multipliers and let : V x


A1 x A2 - t R 1 be a Lagrange function defined by means of the relation

where(,} is the duality pairing between H-~(rK) and H~(rK)


The mixed variational formulation of the Signorini problem with given friction
is defined as the problem of finding a saddle-point {w, >.1, >.2} E V X A1 X A2 of
on V x A1 x A2 satisfying

The relation between ('P) and (f>) is given by


THEOREM 3.4. There exists a unique solution ( w, >.1, .A 2) of (f> ). Moreover,

where u E I< is the solution of ('P) and Tn( u ), Tt( u) are the corresponding normal
and tangential components of the stresses on r K.
REMARK 3.3. If we wish to remove the non-differentiability of ('P) only, it is sufficient
to introduce the set A= A 2 and to define

C(v,JL) = ~(f\.c(v),c(v))- (F;,v;)o + { JLVt ds, (v,JL) E I< x A.


lrK
It is very easy to prove that posses a unique saddle-point ( v, 5.) on !{ x A. Moreover,

v= u, 5. = Tt(u),

where u E I< solves ('P).


250 J. Haslinger

3.6. Reciprocal formulation of contact problems. Case I.


In the mixed variational formulation (P), presented in the last section, nor-
mal and tangential tractions along rK play the role of Lagrange multipliers asso-
ciated with the constraint and the nondifferentiability of .:J. Elimination of the
displacement field v leads to a new variational problem (the so-called reciprocal)
for Tn( u ), Tt( u) only. The derivation of this formulation is the aim of the present
section.
Let {w,.A 1 ,.A2} E V X A1 X A2 be the solution of (P). Then:

(3.33)

Let us set
(3.34)

We derive the explicit form of S. For (f.l 1 , fl2) E A1 X A2 fixed, (3.34) is connected
with the solution of the following linear elliptic boundary value problem:

find z = z(fll, f.l2) E V such that


(3.35)
(Ac:(z),c:(v)) = (f.ll,vn) + (f.l2,vt) +(F;,v;)o, Vv E V.

In view of the linearity of (3.35), one can split z into z and z and write z = z + z,
where z, zE V are the unique solutions of

(Ac:(z),c:(v)) = (F;,v;)o Vv E V, (3.36)

(Ac:(z),c:(v)) = (f.ll,vn) + (f.l2,vt) Vv E V. (3.37)


In the sequel, we denote by G : V' ---+ V the Green's operator, corresponding to V
and the bilinear form (Ac:(z),c:(v)). For brevity, instead of (3.36), (3.37), we shall
write

Since z E V is a minimizer of the quadratic functional ( v, /ll, fl 2), we have

1 1 1 1 -
(z,{lt,fl2) =- 2(f.ll,zn)- 2(f.l2,zt)- 2(F;,z;)o = -2(/ll,zn)
(3.38)
- ~(f.l2,zt)- ~(f.li,zn)- ~(f.l2,it)- ~(F;,z;)o- ~(F;,z;)o.
On inserting v = z in (3.36), using the symmetry of (Ac:(z),c:(v)) and (3.37), we
obtain
Approximation of Contact Problems 251

Thus (3.38) can be written as follows:

(Z,f-LI,I/2) =- ~(f-LI,zn)- ~(f-L2,it)- (f-LI,in)- (f-L2,it)- ~(Fi,Zi)o


1 1
=: -2(J.ti,G(pi,f-L2)n )- 2(/12,G(f-LI,/12)t)- (f-LI,G(F)n) (3.39)

- (112, G(F)t)- ~(Fi, Zi )o.

f3(f..L,v) = (f..LI,G(vi,v2)n) + (J.t2,G(vl,v2)t),f-L = (f-LI,f-L2),v = (v1,v2) (3.40)

and let~: [H-!(r/\)]2 -t R1 be given by


~(f-L) = -(f-LI,G(F)n)- (f-L2,G(F)t).

By making use of this notation, we can rewriteS as follows:

Then the term ( F;, i;) does not depend on (Ill, f-L2) and, therefore, it can be omitted.
Finally, let us set

Now we are in a position to define the reciprocal variational formulation. It is the


problem of finding A = ( A1, A2) E A1 x A2 such that

The relation between (P) and (Pr) is given by


THEOREM 3.5. Tbere exists a unique solution (). 1 ,A2) of(Pr) Moreover,
(3.41)

wbere u E f( is tbe solution of (P).


By using result of Section 3.5, one can give an equivalent form of S:
1 1 1
S(f-L) =2f3(f..L,f-L)- ~(f-L) = 2(f-LI,G(f-LI,f-L2)n) + 2(f-L2,G(f-LI,f-L2)t)- ~(t-L)
1 1 1
=2(f-LI,Zn) + 2(f-L2,Zt)- ~(f-L) = 2(f-L,OZ)- ~(f-L)
1
=2lllz(t-L)IW- e(f-L), 11 = (f-LI,f-L2),
252 J. Haslinger

where z E Vis a solution of


(At:(z),t:(v)) = (J.l,8v) = (f3*- 1 J.l,"fV} Vv E V,
and /3*- 1 : n-! (rK) --+ n-!(rK) denotes the inverse of /3*. The results of Lemmas
3.3 and 3.6 make it possible to write

lllz(J.l)lll 2 = 11/3*- 11-lll:.!.rK = 111-lll:.!,rK


Hence Scan be written in the form

S(J.l) = ~111-lll:.!.rK- ~(J.l). (3.42)

3.7. Reciprocal formulation of contact problems. Case II.


Starting point for this case will be the formulation presented in Section 3.4. For
the sake of simplicity we assume that g = 0 on r K, i.e. frictionless case is considered.
Let us repeat notations:
S(r)=(A- 1 r,r), rES,
K'};,p(O.) = KJ;,P,o(O.) ={rES I (r,c(v)) ~ (f,v) Vv E K}.
Let

and
KF,P(O.) ={rES I (r,c(v)) = (f,v) Vv E Vo}.
It is easy to see that r E KF,P(n) if and only if
O
- +
8T;j
8xi
F,
;= inn' i = 1,2 j

T;jnj = P; on rp ' i = 1, 2 j
Tt(r) = T;jnjti =0 on rK.
Moreover

and
sup (-(Tn( T ), Vn}] (r E KF,P(f?.))
vEK
is the indicator function of Kj; p(O.). Thus
'
inf S(r) = inf sup [-21 (A- 1 r,r)- (Tn(r),vn)]
rEKf,p(O) rEKF,P(O) vEK

Denote by
.C(r,v) = ~(A- 1 r,r)- (Tn(r),vn}
the lagrangian, defined on KF,P(O.) x K. One has
Approximation of Contact Problems 253

THEOREM 3.6. There exists a unique saddle-point of ,C on KF,P(Q) x K. Moreover

a*= T(u) = Ac:(u),


u* = u on r/{ '
where u E J{ solves (P) (with g =0).
Using the fact that (a*,u*) is a saddle-point of ,Con KF,P(n) x ]{,we may
write:
,C(a*,u*) = ~nf supC(T,v) =sup ~nf C(T,v).
rEli.F,P (f!) vEl\ tE/\ rEI\F,P (f!)

For v E f{, we denote


ti>(v)= inf C(T,v) (3.43)
rEKF,P (f!)

the dual functional to S. Next we derive its explicit form.


Let a= a(v), v E K, be the solution of (3.43). Using well-known results, we
immediately obtain that
a= Ac:(w),
where the deplacement field w solves the following elasticity problem:

F
--+
oaij
OXj
i= 0 inn' i = 1, 2;
Wi =.0 on ru ' i = 1' 2 ;
Uijnj =Pi on r p ' i = 1' 2 ;
Tt(a) = a;jnjti =0 on rK;
on rK.

Let To E S be such that

(To,c:(v)) = (f,v) Vv E Vo.

Splitting KF,P(n) we may write

KF,P(n) =To+ Koo(n) ,

where
Koo(n) = {T E S I (T,c:(v)) = 0 Vv E V0 } .
Then
inf C(T,v) = inf
rEKoo(f!)
C(To + r,'v).
rEKF,P(f!)
254 J. Haslinger

A direct calculation yields:

C( To + T, v) = 'H( T) + term depending on To only,

with
1
'H(T) = 2(A- 1 T,T) + (A- 1 r,ro)- (Tn(r),vn),
T E Koo(O). Next we shall analyze problem

inf 'H( r) . (3.44)


rEKoo(Sl)

Relation between (3.43) and (3.44) is clear: iff solves (3.44), f +To solves (3.43). f
satisfies the following relation

'iT E Koo(O) .

Let uo be a displacement field, related to To:

To= Ac:(uo) .
Then
(A- 1To,T) = (uo,T(T))r., + (uon,Tn(T))rK,
because ofT E ]{00 (0). Next we shall assume that

Uo = 0 on f u
Uon = 0 on rK.

This choice leads to:


(A -ITo, T) =0
and
1
'H(f) = rEKoo(Sl)
inf 'H(r) = --2 (Tn(f),vn).

Again, classical result says that

f = Ac:(u) ,
where u is a displacement field solving the linear elasticity problem:

OTij
OXj
= 0 Z = 1, 2 1n Hr. ,

Tijnj = 0 i=1,2 onrp;


Ui =0 i = 1,2 on ru j
Approximation of Contact Problems 255

Let H : H~ (r I\) ~-----+ Koo(Sl) be a mapping defined by

where f is a solution of (3.44). Then

Going back to the problem

inf .C(r,v)= inf .C(r+ro,v)=.C(f+ro,v),


rEJ\F,P(n) rEJ\oo(n)

we see that
1
.C(f + To,vn) = -2(H(vn) n,vn)- (Tn(ro),vn)

(omitting the term (A - 1 r 0 , r 0 )). Thus

Denote by 1( Vn, vn) = ~ (H( Vn) n, Vn} the quadratic part of <J?, 8( Vn) =
-(Tn(ro),vn) its linear term, and

Now we show that the problem

{
find v~ E Hj(ri,) such t~at
(3.45)
<ll(v~):::; <ll(vn) Vvn E H~(rK)

has a unique solution. Here

To this end we show that

From this, the existence and the uniqueness of v~ immediately follows.

1 Definition of II ll.12' r K see below.


256 J. Haslinger

Let V be equipped with the norm


lllvlll = (As(v),c(v)),
which is equivalent to the usual norm in (H 1 (0)) 2 due to Korn's inequality. Let
t.p E H~(rK) be an arbitrary element. Then

clef
11'-PII.!2' rK = inf
vEV
lllvlll = lllulll (3.46)
Vn='i' on fK

defines the norm in H~(rK)


It is easy to see that u, realizing minimum in (3.46), is a solution of the linear
elasticity problem
i = 1, 2 inn;

r(u) = As(u) inn;


Ui = 0 i = 1,2 on ru; (3.47)
Tij(il)nj = 0 i = 1,2 on rp;

~~(~~) = 0} on rK.

Iff E K 00 (0) solves (3.44) then


1 1
'H(f) = -2(H(vn) n,vn) = - 2 (A- 1 f,f)
1
= - 2(s(u),r(u)),
where u solves (3.47) with t.p replaced by Vn Thus

!(Vn,vn) = ~(H(vn)n,vn) = ~IIJuiW = ~llvnll~,rK


From previous analysis the physical meaning of v~ follows: v~ is equal to the
normal component of the displacement field U on r K.
3.8. Variational formulation of contact problems with coulombien friction
Up to now, a function g appearing in (3.18) was given a priori. In classical
Coulomb's law the known function g is replaced by an unknown value FITn(u)l,
where F is the coefficient of Coulomb's friction. So, instead of (3.18) we have the
following conditions:
ITt(u)l ~ FITn(u)l a.e. on rK;
if ITt(u)(x)l < FITn(u)(x)l ===} Ut(x) = 0;
(3.18')
if ITt(u)(x)l = FITn(u)(x)l
===} 3.\ = .\(x) : Ut(x) = -.\(x)Tt(u)(x) .
Approximation of Contact Problems 257

In order to give the variational formulation of such a problem, we use the model
with "a given" friction as an auxiliary one. Nevertheless, some modification is neces-
1
sary: instead of g E L=(rK), we shall write :Fg, where g E H~ 2 (rK) = -A 1 . Then
term frK glvtJ ds will be replaced by (:Fg, JvtJ). Let u = u(g) be the solution of the
following Signorini problem with given friction equal to :Fg:

find u = u(g) E I< such that


{ (Ac:(u),c:(v- u)) + (:Fg, lvtl-lutl)
2': (Fi, Vi- ui)o +(Pi, Vi - ui)o Vv EK .
1
It is easy to prove that -Tn(u) E H~ 2 (rK) This last enables us to define the
1 1
mapping : H~ 2 (r K) --+ H~ 2 (r K) by means of the relation

(g)= -Tn(u).

A weak solution of a Signorini problem with friction obeying Coulomb's law is defined
1
as a fixed point of in H~ 2 (r K ). The existence of a fixed point of for small values
of :F and special geometry of n has been proved (see [4]).
Let S be a quadratic functional given by (3.42) and K:Fg by

K:F,g = {JJ = (J1t,J12) E (H-~(rK)) 2 1111:::; o,


(J12, Vt) + (:Fg, Jvti) 2': 0 Vv E V}
1
with g E H~ 2 (rK) Problem

find (.At(g), .\2(g)) E K:Fg such that


{ (3.48)
S(At(g),.\2(g)) :S S(J1t,J12) V(J1I,J12) E K:Fg

is a reciprocal variational formulation to the contact problem with a given friction


equal to :Fg (studied in Section 3.6). Again, one can prove that

Here (.\1,.-\2) depends on g, as K:Fg depends on g.


- 1 1
Introducing the mapping ci> : H~ 2 (r K) f-t H~ 2 (r K) by means of

we see that coulombien friction corresponds to the case, when g = -.\ 1 (g), i.e. when
g is a fixed point of 4>. As a result, we are led to the so called quasivariational
inequalityfor -Tn(u).
258 J. Haslinger

4. APPROXIMATION OF CONTACT PROBLEMS BY FINITE ELEMENTS


In this chapter we shall analyse the approximation of contact problems without
friction by finite element technique. Results of Chapter 2 will be employed. In the
first part we shall discuss the case of a quite general domain. So, curved elements for
the approximation will be used. In the second part we shall use mixed formulation
for the approximation of contact problems. In both cases error estimates will be
derived, provided the exact solution is smooth enough.
Notations are the same as in Section 3.3.

4.1. Approximation of contact problems - frictionless case. Primal for-


mulation.
We start with the construction of finite-dimensional approximations of K. For
the sake of simplicity we restrict ourselves to the case, when only r K is curved.
Let 'II be a continuous concave (it is not necessary) function defined on (a, b), the
graph of which is r K. We choose ( m + 1) points Al' ... 'Am+l on r K in such a
way that A1,Am+1 are boundary points of rK. Let A;,Ai+l E rK, Q E 0. By
a curved element T we call a closed set bounded by the streight lines QA;, QA;+ 1
and the arc XJii+l The minimal interior angle of the curved element T is called
the minimal angle of the curved element T. A triangulation Th of Q contains curved
elements along r K and internal triangular ones. By the symbols h and {) we denote
the maximal diameter and the minimal interior angle, respectively, of all elements
T E Th. We shall assume only the so called regular systems of triangulations: a
constant {) 0 > 0 exists, independent of h and such that
{) ~ fJo if h ---+ 0+ .
A family of triangulations will be called a-(3 regular, if fJo =a and
h
-<(3
hmin- '

where hmin is the minimal diameter of all T E Th. Define

and
Kh = {v E vh I v. n(A;) :::; 0, i = 1, ... 'm + 1} '
where H(T) denotes the set of linear polynomials, defined on T. It is easy to see
that Kh represents a finite dimensional approximation of K and Kh ct. K, in general.
An approximation of our problem is defined as the solution of the following
problem:
find Uh E Kh such that
{
:f(uh) ::=; :f(v) Vv E Kh
with :r defined in Section 3.3 (with g = 0 on r K ).
Now we shall need some auxiliary results:
Approximation of Contact Problems 259

LEMMA 4.1. Let n C R 2 be a bounded convex domain, the boundary of which is


twice continuously differentiable and let {Th} be a a-(3 regular system of triangula-
tions with a < 1r /8, f3 = 2. Then

iiu- UJiio,an :s; ch~ llull2,n


where UJ denotes the piecewise-linear Lagrange interpolate ofu, c > 0 is independent
of h > 0.
PROOF: See [5].
LEMMA 4.2. Let v E Pt(T), where Tis a dosed triangular element. Let Th be the
triangle generated by replacing the curved side by its chord. Then

ilvii~,6(T,Th) :s; chiivlltrh '


wl1ere !::,.(T, Th) = (T\ Th) U (Th \ T) and c > 0 is independent of h.
PROOF: See [6].
The main result of this section is
THEOREM 4.1. Let the solution u of('P) belong to (H 2 (n)) 2 , r(u) E H(div,n) and
T n (u) E L 2(r K). Let the system of triangulations {Th} satisfy the assumptions of
Lemma 4.1 and w, describing rK, be from C 3((a,b)). Then

llu- uhii :s; c(u)h~ .


PROOF: Using the definition of ('P), Green's formula and (2.2), we deduce

1
2a(u- uh,u- uh) :s; 1
2a(vh- u,uh- u) + 1
rK
Tn(vn- Ukn)ds

+ { Tn(Vhn- Un)ds 'Vv E K, Vh E Kh.


lrK
Let Vh = U[' where U[ is the piecewise linear Lagrange interpolate of u on n. It is
easy to see that u 1 E K h and

a(uh- u, vh- u) :s; Me:iiuh- ullin+


'
M iiu- u1ili n
'

= Me:iiuh- ulli,n + O(h 2 ) 'Ve > 0 (4.1)

[ Tn( UJ - u) n ds :s; ciiui- uiio,rK :s; ch ~ llull2,n ,


lrK
where the assertion of Lemma 4.1 and the regularity assumption concerning u have
been used. The most difficult is to estimate the term

{ Tn(Vn- Ukn) ds . (4.2)


lrK
260 J. Haslinger

In what follows we shall construct a function v E K such that (4.2) is small. We


identify the origin of coordinate system (x1,x2) with the point Ai. Let Ei be a
closed set bounded with the arc Di+l
= Si c rK and the chord AiAi+l Let
x E Ei. By the symbol P(x) and Q(x), respectively, we denote the intersection of
the perpendicular line through the point x with Si and AiAi+ 1, respectively. Let us
- m
define functions Uh, Uh on U Ei by means of the following relations:
i=l

Uh(x) = uh(x) n(P(x)) ,


lh(x) = Uh(Q(x)). n(P(x)) = uh(x). n(P(x)) '

where we set uh(x) = uh(Q(x)). Clearly

Uh(x) = f.h(x) , x E AiAi+l .

Let ~i(x), X E AiAi+l be the linear Lagrange interpolate of uh on AiAi+l and let
- m
us define ~ on U Ei as follows:
i=l

It is readily seen that ~ ~ 0 on r K. We shall estimate

We may write:

m m

IIUh- Uhll~,rK = L IIUh- Uhll~,s; ~ 2 L lluh- uhll~,s;. (4.3)


i=l i=l

Let q be the arc's parameter of the point P(x) = (P1(x),P2 (x)) and denote Q1(x) =
x1. Then for j = 1, 2 we have

Integrating and using Fubini's theorem we obtain

j = 1,2.
Approximation of Contact Problems 261

From this and Lemma 4.2 we have


m
IIUh- Uhll5,rK ~ ch 2 L luhli,E, ~ ch luhli,n
i=l
3 (4.4)

II<!!- Uhll~,rK = L II<!!- Uhll~,s;


i=l

i<l!(q)- Uh(q)l 2 ~ chi<P;- Uhli,A,A;+ 1


3 - 2
(4.5)
~ ch 1Uhi2,A;A;+ 1

As Uh(x) = uh(x) n(P(x)) and uh E P1(A;Ai+1), we may write

Thus, (4.5) and the inverse inequality between H 1(A;Ai+ 1) and H~(A;A;+I) yield:

(4.6)

Adding (4.6) for i = 1, ... , m we obtain:

(4.7)
m
where rh = U A;Ai+l is the polygonal approximation of rK. Using the trace
i=l
theorem we obtain:

lluhll!,rh ~ clluhlli,n,uE,. ~ clluhlli,n,


262 J. Haslinger

where c > 0 does not depend on h for h sufficiently small. Using these estimates
(4.3), (4.4) and (4.7) we deduce

\\<I>- Uh\\o,rK ~ ch~ 1\uh\\l.o ( 4.8)

Next, let v E V be such that

vn=<i> on rJ( .
Then v n ~ 0 on rK, consequently v E I<. Finally we may write

{ Tn(Vn-Uhn)ds= { Tn(<i>-Uh)ds
lrK lrK ( 4.9)
~ ch ~ 1\uh \\l,o .

Since the norms 1\uh\\l.o remain bounded, the assertion of Theorem now follows from
(4.1), (4.9).
In the above error estimates we needed strong regularity assumptions, concern-
ing the solution u. Unfortunately, there are no reasons to expect such a great smooth-
ness. This is why we are going to prove the convergence of uh to u without estimating
the rate of convergence, using no regularity assumptions. According to Theorem 2.1,
it is necessary to verify (2.4) and (2.5).
LEMMA 4.3. (verification of (2.4))
Let us suppose that f' u n f' K = 0 and there exists only a finite number of boundary
points f'p n f'K, f' u n f'p. Then the set

M =I< n (C 00 (0)) 2

is dense in I< in H 1 -norm.


PROOF: The proof for polygonal domains is given in [7], but its slight modification
gives the same density result also in our case.
LEMMA 4.4. The condition (2.5) holds.
PROOF: Let Vh E ]{h be such that

Vh ~ v in V , h ~ 0+ . (4.10)

It is sufficient to show that v n ~ 0 on rK or equivalently

[ v n <p ds ~0
lrK
Approximation of Contact Problems 263

for any i.p E CI((a,b)), i.p 2 0 on (a, b).


Since the trace mapping is completely continuous from V into ( L2 (r K) )2, we
have
(4.11)

hence
Vhn ---+ Vn in L 2 (r I\) , h ---+ 0 + .
Let 'l1 h be the piecewise linear function defined on (a, b), nodes of which are the
points AI, ... , Am+ I Then

is the linear approximation of r r\. Let us set

19(xi) = v(xi, 'l!(xi)) ,


lJh(XI) = Vh(:l"I, \lf(xl )) ,
lJhh(XI) = Vh(XI, \lfh(XI )) , XI E (a, b) .

By virtue of ( 4.11)

(4.12)

Let us prove also


(4.13)

We may write

(4.14)

From the definition of 19hh it follows that these ones are piecewise linear Lagrange
interpolates of 19h on (a, b). Corresponding division of (a, b) will be denoted by
a = tr' < t'2 < < t~+l = b. Using the approximative property of 19hh we have

l 19h -19hh llo,(a,b) :S ch! ll19hll!,(a,b)


:S ch! llvhiii,f! :S ch! ,

where c > 0 is independent of h for h sufficiently small and ( 4.10) has been used.
From this, (4.12) and (4.14), (4.13) follows. Now, let us prove that
264 J. Haslinger

for any t.p E C 1 ( (a, b)), t.p 2: 0 on (a, b). Using (4.13) we have

(4.15)

For the numerical computation of J: 1Jhh nt.p dx 1 we use the trapezoid formula:
1b 1Jhh nt.pdxl ~h((1Jhh n)(t~)t.p(t~) + 2(1Jhh n)(t2)t.p(t2)
+ + (1Jhh n)(t~+ 1 )t.p(t~+ 1 )) =[1Jhh n,t.p].
Since (1Jhh n)(tj)t.p(tj) = (vh n)(Aj)t.p(tj) ~ 0 Vj = 1, ... , m + 1 we have
Vh > 0.
The proof will be finished, if

[1Jhh n,t.p]-+1b 1J nt.pdx1, h--+ 0 + . (4.16)

We may write

1b 1J nt.pdx1- [1Jhh n,t.p]


( 4.17)
~ 1b 1J nt.p dx1 - 1b f)hh nt.p dx1 + 1b 1Jhh nt.p dx1 - [1Jhh n, t.pJ

By virtue of the inverse inequality between H~ ((a, b)) and H 1 ((a, b)):

1b 1Jhh nt.pdxl- [1Jhh n,t.p] ~ chi1Jhh nt.pb,(a,b)


~ c(n,t.p)hii1Jhhlb,(a,b) ~ c(n,t.p)h~II1Jhhllva,b)
~ c(n,t.p)h~llvhiil,l"h ~ c(n,t.p)h~,
where i"h is the polygonal domain bounded with r u, rp and rKh From this, (4.10),
(4.15) and (4.17) we obtain (4.16). Hence

('13 n)(xt) = (v n)(x1, \ll(xt)) ~ 0, x1 E {a, b).


THEOREM 4.2. Let the assumptions of Lemma 4.3 be satisfied. Then
Approximation of Contact Problems 265

PROOF: The assertion of the Theorem is an immediate consequence of Theorem 2.1.

4.2. Approximation of contact problems without friction by a mixed finite


element method
In this section we shall analyse the approximation of mixed variational formu-
lation of contact problems without friction. Let us start with the formulation of the
continuous case.
By C we denote the Lagrangian, given by

defined for (v, JL) E V x A, where

v ={v E (H 1 (0)) 2 I v
= 0 on ru}'
A= n_:;!(rK) = {JL E n-!(rK) I (JL,Vn} ~ 0 Vv E v, Vn ~ 0 on rK}.

Mixed formulation of the contact problem without friction is defined as the problem
of finding a saddle-point (u*, A) E V x A of C over V X A, i.e.:

C(u*,JL) ~ C(u*,A) ~ C(v,A) Vv E V, VJL E A

(see also Section 3.5). It is easy to prove that there exists a unique saddle-point
( u *,A) of C over V x A. Moreover, the following interpretation holds:

u* = u , A= -Tn(u), (4.18)

where u E K solves ('P).


Approximation of (P) will be based on the theory, presented in Section 2.2,
case (CC). For the sake of simplicity we assume that n is a polygonal domain. Let
{Th}, h --. 0+ be a regular family of triangulations of n satisfying usual require-
ments, concerning the mutual position of triangles. With any Th, the following finite
dimensional space will be associated:

vh = {vh - 21 Vh IT E (Pl(T)) 2 VT E Th, Vh


E (C(!l)) = 0 on ru}'
i.e. Vh contains all piecewise linear func~ions over a given triangulation, vanishing on
r U Let {TH} be a partition of the contact part rK, nodes of which will be denoted
by b1 , ... , bM. Let us emphasize that bi don't coincide with nodes of 7h lying on r K,
in general. With any TH the following approximation of A will be associated:
266 J. Haslinger

i.e. AH contains all non-negative and piecewise constant functions (over TH ). The
approximation of (P) is now defined as follows:
find ( uj;, AH) E vh X AH such that
{
(u'h,J-LH):::; (u'h,.AH):::; (vh,AH) Vvh E vh, VJ-LH E AH
or equivalently
find (ui.,.AH) such that
{ a(u'h,vh) + (.AH,Vhn) = (F;,Vih)o + (Pi,Vih)o
(ui.n,J-lH- AH):::; 0 VJ-LH E AH.
Interpretation of uj;. Let

KhH = {vh E vh I r__ Vhnds:::; 0 Vi= 1, ... ,M}'


}t;,bi+l
i.e. KhH contains all piecewise linear functions Vh, the mean value of which in non-
positive on b;bi+l Using the definition of KhH and uj;, it is easy to see that u;;
belongs to KhH and it minimizes the total potential energy functional .:J over KhH.
Next we shall try to estimate llu- ui;ll and IITn(u)- AHII-~,rK' making use of the
general results of Section 2.2.
Let us recall that in notations of Section 2.2:

b(vh,J-lH)=(vhn,J-lH)= { VhnJ-lHds,
lrK
g =0.
As already known, the crucial point is the verification of the stability conditions (S):

(4.19)

with S> 0 independent of h, H. Let us repeat some results, guaranteeing (4.19),


proofs of which can be found in [8].
We introduce an auxiliary elliptic problem:
find w E V such that
{ (4.20)
(c;j(w),c;j(v)) = (cp*,vn) Vv E V,
where <p* E n-!(rK) Let (4.20) be regular in the following sense: if <p* E
H-!+e(rK) (c > 0 a positive number), then the solution W of (4.20) belongs to
(Hl+~(n))2 and
(4.21)
holds with a constant c(c) > 0, depending on c > 0, in general. Now, the following
lemma can be proved:
Approximation of Contact Problems 267

LEMMA 4.5. If (4.20) is regular in above mentioned manner, then (4.19) is satisfied,
provided the ratio h/ H is sufliciently small.
The main result of this section is:
THEOREM 4.3. Let the solution of (P) belong to I< n (Hl+q(n))2 for some q > 0
and let Tn(u) E L 2(rK) Then

llu- u;;11 = O(Hii)


IITn(u)- AHII-~,rK = O(Hii), H-+ 0+,

where q =min{ q, :l- }.


PROOF: Using (2.15) and (2.16) we see that

llu- u;;ll 2~ C{llu- vhll 2+ ITn(u)- /lHI:_~,rK (4.22)

+(un,Tn(u)-!lH)}
IITn(u)- AHII-~,rK ~ C{llu- uhll + ITn(u)- llHI-~,rK} (4.23)

with a constant c > 0, which doesn't depend on h, H. Let Vh = U[ E vh be the


piecewise linear Lagrange interpolate of n and JlH the orthogonal L 2 -projection of
-Tn(u) on AH. Then

llu- uiii2 = O(h2q) '


ITn(u)- /lHI-~,rK = O(H~),

l(un, Tn(u)- !lH)I ~ llunll~,rKITn(u)- llHi-~,rK = O(H~),

making use of classical approximation results. This, together with (4.22) and (4.23)
leads to the assertion of Theorem 4.3.
Higher regularity assumptions make possible to improve the rate of convergence
(see [8]).

5. SHAPE OPTIMIZATION IN CONTACT PROBLEMS


Let us suppose that a part of an elastic body n is close to a rigid foundation.
Thus a part r K of n's boundary may get into contact with the foundation.
Let us assume that the boundary of n is decomposed as follows:

(5.1)

where different boundary conditions will be assumed, namely u = 0 on r u, Tij(u)ni =


Pi, i = 1,2 on rp, and rK includes the contact area. We suppose that ru and rK
are non-empty and open in an.
268 J. Haslinger

To concretize the situation we assume that the shape of n will be given by


Figure 5.1 and rK is described by a graph of a function a, the choice of which will
be specified later:

(5.2)

Let the rigid foundation be given by the set {(x1,x2) I x2 ~ 0}. Then we have
the following system of unilateral conditions on rK:

u2(x1,a(xl));::: -a(x!), (non-penetration of n into rigid foundation)


{ T2(x1,a(xl));::: 0,
(5.3)
(u2 + a)T2 = 0, (no c~nt~ct-no forces)
=
T1 0, (no fnct10n).

In correspondence with unilateral conditions (5.3), we define the closed, convex


set I< of kinematically admissible displacements,

I<= {v E VI v2(x1,a(xl));::: -a(x!) Vx1 E (a, b)}. (5.4)

The variational solution of a contact problem is then defined in usual way:

find u E I<: 3(u) ~ 3(v) Vv E I<. (5.5)

5.1. Setting of the optimal shape problem. Existence result.


Up to now we considered the situation when the shape of an elastic body is
given. Now we shall assume a family of contact problems. As a (control) parameter
we consider the function a, describing the shape of the contact part rK of the
boundary an.
Let C0 , C 1, C2 be given positive constants chosen in such a way that a set Uad
defined by

Uad ={a E C 01([a,b]) I 0 ~a~ Co, la'(xl)l ~ cl, measO(a) = C2}, (5.6)

is non-empty and define

Let n=(a, b) X (0,-y) (see Figure 5.1).


Let us note that n ::) O(a) \fa E Uad Next we shall assume that all physical
datas are defined in n.
Approximation of Contact Problems 269

ru
co -----

Figure 5.1.

20
30
40
50
60

Figure 5.2.
By V(a), K(a) we denote corresponding sets of functions, defined over fl(a)
and set
.7a(v) =2(r(v),e(v)) (a)- (Fi,vi)o,O(a)- (Pi,vi)o,rp
1
0

= ~1 Tij(v)ei;(v)dx-1 Fividx- [ Pivids.


2 O(a) O(a) lrP
For any a we consider a contact problem (P(a)):
find u(a) E K(a) such that
{ (P(a))
.7a(u(a)) ~ .7a(v) \fv E K(a).
A physical phenomenon closely connected with the contact problems is the be-
haviour of the contact stress T2( a). A typical configuration of the contact stresses
along the contact boundary rK(a) is given in Figure 5.2.
270 J. Haslinger

As the stress peaks are undesirable from the practical point of view it is natural
to ask whether it is possible, and how to findsuch a shape for the contact surface
that the normal stresses become evenly distributed, see the joint Figure 5.3 .

T2 for
20 final design
. . . . . . . . . . . . . r 2 for
60
inial design

Figure 5.3

This would seem to motivate the study of the problem of minimizing the maximal
stresses on the boundary, i.e.
min E(a),
where E(a) = max JT2 (a)(x)J. This approach however leads to some difficulties.
rK(a)
First of all, the functional E(a) may not be defined, in general. Secondly, if it were
possible to define E(a) (by using a-priori regularity assumptions, for example), then
E(a) would not be differentiable, in general.
So another choice of the cost functional is necessary. A good choice (frequently
used by engineers) is the total potential energy functional, evaluated in the equilib-
rium state u(a)
E(a) = ..1a(u(a)). (5.7)
For a suitable Uad it is possible to show that the minimizer of E( a) in Uad yields an
"even" distribution of contact forces. Moreover, E( a) is differentiable with respect
to a despite the fact that the mapping a ~---+ u( a) is non-differentiable, in general.
So our problem reads:

find a* E Uad such that


{ (P)
E(a*)::;E(a) VaEUad,

with E(a) given by (5.7) and Uad given by (5.6).


For the existence of the solution a* of (P) we have
THEOREM 5.1. There exists at least one solution of (P).
Before we prove the result, we shall need some auxiliary results.
Approximation of Contact Problems 271

LEMMA 5.1. Let an=:a in [a,b] and let <p E K(a), <p = (<pt,<f'2) be given. Then
there exist <pj E (H 1 (f2)) 2 and a subsequence {an(j)} C {an} such that <f'iln( 01 n<n) E
K(an(j)) and <pj - t (pin (H 1 (f2)) 2 , where (p = ((p 1 , (p 2 ) denotes the Calderon exten-
sion of <p from n( a) on n.
PROOF: Define a function

i.e. tfln( 01 ) E K(a), where '' = (7ft, tP2) with tPt = cp1. Let us split 'P as follows:
'P ='If + 1>. From the construction of 'If we see that

Applying the classical density result, one can find a sequence {ci> i}, ci> i = ( 1> 1j, 1> 2j) E
ccoo(n)) 2 with <>tj, 1>2j vanishing in a neighbourhood of I\, r u u rK(a), respec-
tively, and such that

Let us define now:

Then
llrfi- <~'illt,n = IIP- Pillt,n -4 o, i - oo.

Let us denote dj = dist{ supp 1>2j' r u u r K( a)} and let io be fixed. As a j::::a in
[a, b], then there exists no = n(jo) such that the graph of the function an 0 does not
meet supp <>2j 0 As

and
tP2(Xt,X2) ~ -X2 V(xt,x2) En'
(as follows from the definition of '1f2) we immediately get

Hence
272 J. Haslinger

LEMMA 5.2. Let an=! a in [a,b] and let Un E K(an) be the solution ofP(an), i.e. 1

(r(un),c:(z- Un))n(an)
(5.8)
~ (F, z- Un)o,r!(an) + (P, z- Un)o,rp Vz E K(an) .

Then there exists a subsequence of {Un}, denoted again by {Un} such that

Un _. U

for any integer m, with Gm = {(x1, x2) ER2 I x1 E (a, b), a( xi)+ ~ < X2 < r },
where u E K(a) is a solution of (P(a)).
PROOF: As F E (L 2(Q)) 2 , P E (L 2(rp )) 2 , from (5.8) and Korn's inequality it
follows that 2
Vn. (5.9)
Let m be fixed and let Gm be given as above. As an=!a in [a, b], there exists no( m)
such that
Gm c fl(an) Vn ~ no(m).
From this and (5.9) it follows

(5.10)

Hence, one may extract a subsequence {unJ C {un} such that

A function u<m) E (H 1 (Gm)) 2 is such that u<m) = 0 on run 8Gm. Similarly, as


Gm+l :) Gm, there exists no(m + 1) ~ no(m) such that

Vn ~ no(m + 1) .

Hence a subsequence {Un 2 } C { Un 1 } can be chosen in such a way that

in Gm.
1 We assume that FE [ 2 (0)]2, P E [ 2 (80)]2

(P;,vi)o,rp
and (F,v)o,O(a) = =
(F;,v;)o,O(ar) (P,v)o,rp

2 Here we use the fact that the constant in Korn's inequality


can be chosen independently of an E
Uad (see [11)).
Approximation of Contact Problems 273

Proceeding in this way for any integer k, one can construct a subsequence {unJ C
{unk_J such that

for any j < k + m - 1. Let {u!;} be a diagonal sequence determined by {Unk}. From
the above analysis it follows that

(5.11)

for any integer m, where


in Gm.

Clearly, u E (H 1(i1(a))) 2 , u = 0 on 8i1(a) n ru. Instead of {u!;}, {a!;}, we shall


next write simply {un}, {an} As Un E K(an), an=!a in [a,b], (5.9) and (5.11) hold,
a function u E K(a). It remains to verify that u = u(a) is a solution of (P(a)).
Let~ E K(a) be an arbitrary function. Let ~j E K(an;) be such that

(5.12)

({is the Calderon extension of~ on 0). The existence of such a sequence is ensured
by Lemma 5.1. Let Un; = u(an;) be the corresponding solution of (P(ani )):

(5.13)

In (5.13), a function ~j E K(an;) satisfying (5.12) is assumed. Now

(r(un;),t:(~j- Un;))!l(an.)
J

= (r(un;),c(~j -Un;)) 0 m + (r(un;),c(~j -Un;))!1(an.)\!1(a)


J

+ (r( Un; ), c( ~j - Un;)) (!1(a) \ Gm)n!1(an.)


J

~ (r(un;),c(~j- Un;)) 0 m + (r(un;),c(~j)) 0 (an.)\!1(a)


J

+ (r(t1n;),c(~;))!1(a)\Gm .
274 J. Haslinger

From this, (5.10), (5.11) and (5.12) we easily get 1

.lim sup (r(un.),t:(~j-


J-+00 J
unJ)"(
.u Onj
)
(5.14)
~ (r(u),t:(~- u))cm +C ll~llt,fl(<>)\G,.

Similarly,
(F,~j- unJo,fl(<>n)
J

= (F,~j- Uni)O,Gm + (F,~j- unJo,fl(<>nj)\fl(a)


+ (F,~j -unJo,(fl(<>)\Gm)nfl(<>n) J

From this, (5.10), (5.11) and (5.12) we get

lim inf (F, ~j- Uni ) 0 , fl( On; )


J--+00
(5.15)

Finally,

where the one-dimensional Lebesgue measure of


1 1
Mm = {(x1,x2) E R2 I X2 E (a(a),a(a) + -),
m
x2 E (a(b),a(b) + -)}
m

is 1/m (see Figure 5.1) (this consideration can be omitted if dist(fp, rK(a)) > 0).
Then
.lim inf (P, ~j- UnJ ) 0 ' r p
J--+00
(5.16)
2 (P,~- u)o,rp\Mm- C(IIPIIo,M"' + ll~llo,M"')
Combining (5.14), (5.15) and (5.16) we have

(r(u),t:(~- u))cm + Cll~lll,fl(a)\Gm


2 (F, ~- u)o,Gm + (P, ~- u)o,rp \ Mm
- C(IIFIIo,O(a) \ Gm + ll~llo,fl(a)\ Gm)- C(IIPIIo,Mm + ll~llo,Mm)

Letting m ---+ oo we obtain

(r(u),c:(e- u))o(a)
(5.17)
2 (F,~- u)o,O(a) + (P,~- u)o,rp V~ E I<(a),

1 Here we use the uniform extension property of {O(a)}, a E Uad


Approximation of Contact Problems 275

i.e. u is a solution of (P(a)).


PROOF OF THEOREM 5.1: Let us denote

q= inf E(a)
aEUAtl

and let an E Uad be a minimizing sequence, i.e.

q = lim E(an) .
n-+oo

Using the compactness of Uad, we can extract a s~bsequence {an;} C {an} such
that
in [a, b] . (5.18)
Clearly a*. E Uad Let Un;(an;) E K(an;) be solutions of (P(an; )). From Lemma
5.2 it follows immediately that there exists a subsequence of {Un; (an;)} (deno~ed
again by {u 11; (an;)}) such that

(5.19)

for any m, where

and u(a*) E K(a*) is the solution of (P(a*)). From this it follows that

E(a*);?: q . (5.20)

On the other hand, form fixed and nj sufficiently large,

where
1
Eam(a)(an;) =2 (r(un; ),e(Un; ))Gm(a)
- (F, unJo,Gm(a) - (P, Un; )o,rp \ Mm
and
1
En(an;)\Gm(a)(an;) = 2 (r(un;),e(Un;))n(an)\Gm(a)
- (F, unJo,O(an;) \ Gm(a) - (P, unJo,Mm

;?: - (F,un;)o,n(an;)\Gm(a)- (P,unJo,M.,. '


276 J. Haslinger

where the one-dimensional Lebesgue measure of Mm is 1/m (see Figure 5.1) (with
eventual modification if dist(rP,rK(a)) > 0). Then

.lim inf E( ani)


J-00

2: J-00
.lim inf Eam(a)(anJ +.lim inf En(an.)\Gm(a)(an j)
J-00 1

(5.21)
2: Eam(a)(a*) + i~~ inf (- (F, un;)0,!1(anj )\ Gm(a) - (P, Unj )O,Mm)
2: Eam(a)(a*)- C (11FIIo,!1(a)\ Gm(a) + IIPIIo,Mm)
holds for any m. Here

The lower weak semicontinuity of Eam(a) (5.19), and (5.10) have been used. Let-
ting m-+ oo, we have from (5.21):

q = .lim inf E(an)


J-00 1
2: E(a*) .

This, together with (5.20), gives the assertion of Theorem 5.1.

REFERENCES:
1. Ekeland, I. and R. Temam, "Analyse convexe et problemes variationnels,"
Dunod, Gauthier-Villars, Paris-Bruxelles-Montreal, 1974.
2. Haslinger, J ., Mixed formulation of variational inequalitie8 and it8 approxima-
tion, Apl. Mat. 26, 462-4 75.
3. Necas, J. and I. Hlavacek, "Mathematical theory of elastic and elasto-plastic
bodies: an introduction," Elsevier Scientific Publishing Company, Amsterdam-
Oxford-New York, 1981.
4. Necas, J., J. Jarusek and J. Haslinger, On the 8olution of the variational equality
to the Signorini problem with 8mall friction, Boll. Unione Mat. Ital. (5) 17-B,
796-811.
5. Nitzsche, J.A., Uber ein Variation8prinzip zur Lo8ung von Dirichlet-Problemen
bei Verwendung von Teilraumen, die keinen Randbedingungen unterworfen sind,
Abh. Math. Sem. Univ. Hamburg 36, 9-15.
6. Fix G. and G. Strang, "An analysis of the finite element method," Prentice-Hall,
Englewood Cliffs.
7. Hlavacek, I. and J. Lovi'sek, A finite element analy8i8 for the Signorini problem
in plane ela8to8tatic8, Apl. Mat. 22 (1977), 215-228.
8. Haslinger, J. and I. Hlavacek, Approximation of the Signorini problem with
friction by a mixed finite element method, JMAA, No 1 86 (1982), 99-121.
Approximation of Contact Problems 277

9. Hlavacek, I., J. Haslinger, J. Necas and J. Lovisek, "Solution of variational


inequalities in mechanics," to appear in Springer-Verlag, 1988.
10. Haslinger, J. and P. Neittaanma.ki, "Optimal shape design in systems governed
by elliptic variational inequalities," to appear in John Wiley & Sons.
11. Haslinger, J., P. Neittaanmaki and T. Tiihonen, Shape optimization of an
elastic body in contact based on penalization of the state inequality, Apl. Mat.
31 (1986), 54-77.
DISCONTINUITIES AND PLASTICITY

P.M. Suquet
Universite des Sciences et Techniques du Languedoc:,
Montepellier, France
280 P.M. Suquet

1. INTRODUCTION

The guide line of these lectures is that Plasticity and spatial dis-
continuities are two companion phenomena :
a) on the one hand displacements (or displacement rates) solutions
of Plasticity problems are naturally discontinuous. This conclusion is
easily drawn from the asymptotically linear growth of the functionals
arising in the variational formulation of Plasticity problems :

Inf J(u) = J j(E(u))dx- L(u) (I. 1)


u n
where E stands for the usual deformation operator (infinitesimal
strains), L is a linear form and j satisfies :

j(e) ~ k 0 iel- k 1 (1.2)

Accounting for (1.2) the natural space of definition for J contains


discontinuous vector fields. But this is only on~ of the difficulties
arising from the variational problems (1.1) . Another one is the non
lower semi continuity of functionals, which has for main consequence a
~elaxation p~oe~dun~ necessary to obtain well posed variational problems.
Our presentation of this subject follows and extends TEMAM & STRANG's [1]
original work.
b) on the other hand it is known from experimental observation that
Plasticity takes its source in microscopic slips along specific cristal-
lographic planes. To give a more mathematical basis to this assertion we
examine a model of rigid blocks slipping ones over others. The blocks
size is assumed to be a small parameter 1/n, and the problem can be
described by minimizing a functional Jn . In order to derive the homo-
g~niz~d ~obl~m (when 1/n tends to 0), we use the theory of ~p~
eonv~g~ne~ which allows to compute the limits of variational problems.
We shall point out a few singular aspects of epi-convergence in Plasti-
city mainly arising from (roughly speaking) the non eommutativ~ty o6 ~e
la~on and ~p~ eonv~genee.
c) located midway between these two aspects of "Plasticity and dis-
continuities" the problems of ~n p!Mile ta.yeJL6 will be studied in
these notes.
Our aim is to bring ~ention 06 ~ea.deJL6 on recent mathematical de-
velopments which could open significantly wide fields of investigation
for researchers interested in mathematical aspects of Mechanics. Therefore
Discontinuities and Plasticity 281

detailled proofs are often omitted, but conjectures and theorems are, as
far as possible, distinguished. The author's personal contribution in
theoretical developments is most of the time rather limited, but the
goal is to attract readers to a m~cha~L[cal app~oach of TEMAM's or
ATTOUCH' s books [2] [3] , to BOUCHITTE' s thesis [4l , or to recent works
of the Italian School. The background necessary to follow the text inclu-
des a basic knowledge of convex analysis, of measure theory and of clas-
sical functional analysis, and can be found, together with numerous other
subjects, in PANAGIOTOPOULOS [9] .

ACKNOWLEDGMENTS
Several discussions with Guy BOUCHITTE and numerous uses of his re-
sults are gratefully acknowledged.
282 P.M. Suquet

2. LIMIT ANALYSIS

2.1 STATICAL AND KINEMATICAL FORMULATIONS


In all further developments we consider that the material under con-
sideration occupies a domain ~ of ~3 and is submitted to body forces
Af and boundary forces Ag on a part r 1 of its boundary a~ . r ,
complementary of r I in an will be submitted to imposed displac~ments.

--
- Figure 2. I -

Throughout the text we shall assume that the following assumption is met

(I) is cl
ro and r I are closed in a~ ,
}
a~
r
o
n r
I
= 0 , ro u rl = a~ ( 2. 1)
( 2) f E 1 00 (~) 3 g E L""(rl)3

Assumption (I) can be weakened but we do not seak here for minimal hy-
pothesis.
In this section the only information used on the constitutive law
of ~ is its strength capacft) defined, at every particle x of ~ ,
through a set P(x) in ~ s9 + delimiting allowable stress tensors o(x)

( +) 9
~ 3 x 3 symmetric tensors
s
Discontinuities and Plasticity 283

o admissible < = > o(x) E P(x) a.e. x 1n S] (2. 2)

Throughout the following we shall assume that :

P(x) is a closed convex set of ~ 9 containing


s ' ( 2. 3)
0 1n its interio~ and symmetric with respect to 0

Although physically relevant examples of P(x) are given by Tresca or


Von Mises criterions we shall tacitly assume (unless explicitely speci-
fied) that P(x) 1s a ball of radius k(x)
9
P(x) = {o E ~ (2.4)
s
We emphasize that the knowledge of P(x) does not presume of the remai-
ning material properties (elasticity, viscosity ... ) since several diffe-
rent materials, exhibiting different behaviors, might have the same
strength capacity (cf Figure 2.2) and therefore should be considered as
identical in this section.

0 0

Ductile Brittle
Two ck66eJten;t ma.J:e!U._a.!.> (duc..tile and b!U.Ute) wah the .6ame .6-tfteng:th c.apa-
c.ily illM.:llta):ed by :the ultima-te .6tlteM o
0

Figure 2.2

The equilibrium equations, independent of the material properties, read as

div o + Af = 0 in Q o.n = Ag on r1 (2.5)

and the space of statically admissible stress fields 1s defined as :


2
S(H , Ag) = {o E IL (S]) , o satisfies (2.5)}, where n.l(S]) =Lp(S]) 9
s
284 P.M. Suquet

The central problem of imit a~aly~~(+) is to find the max~mum


value of the loading parameter A for which a stress field a satis-
fying both material constraints (2.2) and equilibrium equations (2.5)
can be found [sj

As = Sup{A E 1R jja E S(Af, Ag) , a(x) E P(x) a.e. x in ~}. (2.6)

As is the limit load, and (2.6) is its statical definition, since it


relies on stress fields a only. A dual definition of the limit load,
involving only kinematical fields is classically given [s] . Let
A < As , a a statically and plastically admissible stress field asso-
ciated with A by (2.6) , and let u be any kinematically admissible
displacement rate field

( 2. 7)

The principle of virtual work, elsewhere called Green's formula, yields

A L(u) J aE(u)dx (2.8)


~
where L(u) is related (multiplied by A) to the potential of given forces

L(u) =J fudx + fr guds (2.9)


~ I
We estimate the rig~t+~and side of (2.8) with the help of the support
function of P(x)

e E 1Rg ,TT(x,e) Sup (T .e) ,


s
T E P(x)

J TT(x,E(u))dx
~
i.e.
IT ( u)
Inf (2.10)
uEV jL(u) I
Noting that IT is positively homogeneous of degree I (consequence of the
same property valid for TT) and that IT is even (consequence of assump-
tion (2K3)) we obtain an alternative definition of the ~~emati~al imit
load A
Ak = 0 lnf{IT(u) , u E V , L(u) = 1}. (2.11)

The inequality As ~ Ak ~s a simple consequence of the derivation of Ak


(+)
also called yield de6ign when no reference to Plasticity is made
(cf [s]).
(++)
rx is the indicator function of the set X . fx is the conjugate of
f in the appropriate duality.
Discontinuities and Plasticity 285

(see (2.10)) . The equality As= Ak is a more difficult result, exten-


sively discussed in the mechanical literature (see for instance NAYROLES
[6]) and to which we devote the next section.
To close these preliminaries we sum up basic properties of the
functional II

II(cm) = [a[II(u) , ( 2. 12)

( 2. 13)

and under assumption (2.4)

k [du)[ 1 ~II(u)~k 1 [E(u)[ 1 (2 .14)


0 JL (rl) L (rl)
(2.12)(2.14) are immediate, (2.13) is a consequence of (2.11) and
of the convexity of II .
2.2 HOMOGENEOUS MATERIAL, DUALITY RESULT
The following theorem is due to TEMAM-STRANG [1] , in a more gene-
ral setting since assumption (2.4) simplifies several technical points
of [1] .
Theo~em 2.1 (TEMAM-STRANG). We ~~ume that ~~umptio~ (2.1)(2.3)
(2.4) ane met. Then
s k
A (Sup(2.6)) =A (Inf(2.11)) (2.15)

P~oo6. Theorem 2. I results from a general argument of Convex Analysis


[7] [8] [9] .
Abstract result. Let V and Y be two Banach spaces et A be a linear
~;ppi~g-;~;--v into Y We consider the following variational problem
(P) and its dual (Px)

(P) Inf [F(v) + G(Av)j


v E V

(P:t:) Sup [- Fx(Axp:t:) - G\-px)]


Px E Y:t:
where F(resp. G) are convex, lower semi continuous (l.s.c.) , proper
functions from V (resp. Y) into lR U {+ oo}. Assume that
i) Inf(P) E lR
ii) <] u0 E V) F(u () ) <+oo G continuous at Au 0
Then
iii) Inf(P) = Sup(P :t: ) and iv) (P:t:)
possesses at least one solution.
This abstract result is applied to limit analysis, specifying V , Y ,
A , F , G as follows : 2
V given by (2.7) , Y =L (rl) A= E
F(v) = 0 if L(v) = I , + oo otherwise,
286 P.M. Suquet

G(p) = fn
1T(p)dx .
With these notations (P) is exactly (2.11) . It can be checked that(+)
x
- A if 3 A E 1R such that -p E S(Af , !.g) ,
F*cr..xp*)
{ +"" otherwise,
x
p (x) E p a.e.
0 if X ~n Q'
Gx(p*) = { +"" otherwise.
Therefore (Px) reads as

- Sup{- A I jo(= -p*) E S(H, >.g) o(x) E P a.e. x ~n 0.}

which is exactly (2.6) . Since G is everywhere continuous and finite


on Y , since Inf P is obviously finite, general conclusions iii)
and iv) of the abstract result apply and complete the proof of theo-
rem 2. I
The kinematical approach (2.11) iskmore convenient for a numeri-
cal computation of the limit load As = A , since most of finite
elements approximations are based on displacements. For this purpose any
descent method will provide a minimizing sequence and one has to check
that this procedure eventually converges to a minimizer of (2.11) ,
the existence of which is not ensured by theorem 2.1 . A more thorough
study of (2.11) is therefore required, and we recall the most classi-
cal arguments used to obtain solutions of minimization problems in the
form

Inf <!>(x) (2.16)


xEK
where K is a convex subset of a topological vector space X . Usual
assumptions are
i) X is the dual of a Banach space Y (X ~s often assumed to be
reflexive).
ii) K is closed for the weak x topology of X
iii) <!> is convex, proper, l.s.c. for the weak x topology of X .
iv) K ~s bounded or

lim <l>(x) +oo (2.17)


xE K , Ixi ~ + oo
By experience we know that the choice of X is often governed by the
growth condition (2. 17) . The program of study for the variational
problem (2.16) in a specific situation is therefore the following :

(+)a similar reasoning will be detailed ~n section 2.4 .


Discontinuities and Plasticity 287

a) Find the optimal space X 1n order to ensure iv) : in the case of


limit analysis the answer 1s provided by BD(n) to which section 2.3
is devoted . Check i) .
b) Check ii) and iii) in the case of limit analysis K ~ not
cto~ed 6o~ the weak x topo!ogy o6 X , and we need to introduce the
l.s.c. hull of ~ + IK through a so-called ~eiaxed p~obtem.

2.3 VECTOR FIELDS WITH BOUNDED DEFORMATION : BD(n)


A first guess for X, suggested by the growth condition (2.14)
would be that it consists of veJtor fields V with a deformation E(V)
in L 1 (n) . However, since L (n) is not the dual of a Banach space
(and of course not the dual of L00 (n)) , such a first guess fails.
The correct functional framework, introduced independently by
G. STRANG [10] and the author [11] [12], is

BD(n) = {uE L 1 (n)N, E (u) E M1 (n), 1 ~ i , j ~ N} (+) (2.18)


1]

where M1 (n) stands for the space of bounded measures on n, dual of


C~(n) space of continuous functions on n vanishing on an [13] . One
could naturally ask whether BD(n) is equal to BV(~)N or not : the
answer is negative as proved by MAJTIES & al [14] who established
that Korn's inequality fails in W ' (~)N . Thts result motivated fur-
ther studies of BD(~) and we shall briefly(++ sum up a few results
on BD(~) which help to understand the (non)regularity of possible solu-
tions of Limit Analysis and Plasticity equations.
1) Structure endowed with the following norms is a Banach space
--------- BD(~)
N N
luiBDW) L: Iu.l 1 + L: IE .. (u) I 1
i=1 1 L (~) i' j = 1 1J M (~)

or equivalently [2]
N
luiBD(n) = r(u) + L: IE .. (u) I 1 (2.19)
i' j = 1 1J M (~)

where r(u) is any continuous semi-norm on BD(~) strictly positive on


rigid displacements: r(u) > 0 if u = a +b A x a,b E lR3, 1 u,f, 0 .
BD(n) ~ not a ~e6!exive Bana~h ~pa~e (similarly to M (n)) howe-
ver an important result due to MATTHIES & al [14] asserts that

BD(~) (2.20)

Therefore it can be endowed with a weak x topology for which bounded sets
are relatively compact.

(+) For practical applications N = 3 .


(++)For further details see [2] [9] [12]
288 P.M. Suquet

2) g~g~l~E!!z Followin~ the classical derivation of Sobolev's inequa-


lities one can prove [2J [I 2] that BD~) is continuously embedded into
Lp (n) N for I ~ p ~ N/ (N- I) (N=3 in 1R ) , with a compact embeddiyg for
I~ p < N/(N-I) . Moreover, under the assumption (2.I) (an is C)
one can state [2] the following result (of Deny-Lions type)

[u E V'(n)N, E .. (u) E MI(n)] => u E BD(n) .


1J .
Vector fields in BD(n) admit a trace on an as specified by the
following theorem [2] [I iJ
The.otLe.rn 2. 2 M.wme. tha;t ( 2 . I) ~ rne.t. The.n.
i) TheJte. e.wu a Une.aJt and c.ontinu.ow., opeJta;totL y mapping BD(S1)
ant~ ~I(an)N ~u.c.h tha;t 0on e.veJty u in BD(n) and e.ve.ny ~ in
ci(n)N
s
the. 0ottowing Gne.e.n'~ 0onrnu.ta hoi~ tnu.e. :
Jn u.~ . . . dx +
1 1J, J
Jn ~-. dE.. (u) =
1J 1J
Jan y(u) .~ .. n.ds .
1 1J J
(2.2I)

Whe.n u ~a ne.gu.taJt 6ie.id (e..g. C0 (IT)) y(u) c.oinc.id~ with ulan .


U J Let s be. an onie.ntabie. c I v~e.ty indu.de.d in n , ~ e.paJtrnng
n into n and n . TheJte. e.wu two Une.aJt and c.ontinu.ow., tnac.e. ope.-
nxto~ y- and r; mapping BD(n) o~to 2 LI(s)N , ~u.c.h tha;t 0on e.ve.ny
u in Bntn) and non e.veJty ~ in V(n)~ the. 6oltowing Gne.e.n'~ 6onrnu.-
fu hoi~ tnu.e.

Jn u.~ . . . dx + Jn ~ .. dE .. ( u) J +
y 5 (u). ~- .n.ds
(+)
1 1J .J 1J 1J s 1 1J J
(2.22)

f_n f_n J y +(u1_~ .. n.ds


s 8
u.~ . . . dx + ~-.dE .. (u)
1 1J , J 1J 1J 1J J
- -
and_ y;(u) Me. the. int0nal and e.x{Vtnal tnac.~ o 0 u on s .
C0 (n)N ve.c.ton 6ie.ld, y 5 (u) and y 5 (u) c.oinc.ide with uls
For the sake of simplicity we shall
denote
+ +
u = r;(u) u = r 5 (u) and

u an I = r ( u)

Figure 2.3 -

(+) n is oriented from n to n


+
Discontinuities and Plasticity 289

Let us note that Green's formula (2.21) plays an essential role in the
defiTiition of the trace operator y Indeed we shall point out that
V(S'i")"~ ~ n.o:t devtJ.>e in BD(Q) for the topology defined by (2.19) .
Therefore the usual linear trace operator defined on V(IT)N can be ex-
tended to BD(IG) in infinitely many operators. However only one of these
possible continuations satisfies (2.21) .
Another worth noting result is that elements of BD(IG) might admit
discontinuities on arbitrary surfaces included in ~ A simple formula
relates discontinuities of u and a mass of E(u) on a surface S .
By difference between the two members of (2.22) we obtain

Js '{1 1J dE 1J.. (u) = Js (u:1 -u~)'{l .. n.ds


1 1J J
= Js '{l .. ([u) ~
1J s
n) .. ds
1J
(2.23)

where the symmetry of '{I has been used in the last equality, and where

[ u] = u + - u- , ( q n) . . = I I 2 ( q. n. + q. n.) ( 2. 24)
. s 1J 1 J J 1
As a simple consequenc~ if u E C(~ ) 3 U C(~ ) 3 (with different traces
on S), then u E BD(IG) with - +

E(u) = E(u) In Xn (x) + E(u) In Xn (x) + [u] s n os , (2.25)


- - + +
xx(x) is the characteristic function of X , and os is the Dirac
measure on s .
3) Approximation We already mentionned the fact that V(IT)N ~ no:t
since its closure consists of vector fie!d~ u in
den~e-zn--BD(i'i")-,
Ll(n)N with a deformation E(u) in ~l(n) However V(n) is dense
in BD(n) for topologies which are weaker than the strong one (the
weak* topology for instance).
One of these convenient topologies is defined by the "total deforma-
tion" (term used by analogy with the total variation of an element of BV).
More specifically

(V u E BD(IG)) (j u E V(~)N) lim (u - u) 0 in L I (n) N


n n-++oo n ( 2. 26)

lim (u -u)j = 0 in L 1 (an)N, lim jfjE(u)j-fHu)I!=O


n-+ + 00 n an n-++oo Q n n
'V
4) Extension Let w :D n , u 1 E BD(n) and u2 E BD(w- ~) Then u
defi~~d-b~l~w belongs to BD(w)
'V 'V
u(x) = u 1 (x) if X E Q , u(x) = u 2 (x) if X E w-IT

2.4 HOMOGENEOUS MATERIAL, RELAXED PROBLEM


Coming back to the variational problem (2.11) in which we want to
recognize the structure of (2.16) with X= BD(n) , we check that the
growth condition (2.17) is likely to be satisfied by virtue of (2.14)
~ is a proper, l.s.c., convex functional on BD(n) However another
problem arises when examing the properties of K :
290 P.M. Suquet

K = K1 n K2 K1 = {u E BD(n) u 0 on r }
0

K2 = {u E BD(n) L(u) 1}

Indeed neither K1 nor K2 are closed sets for the weak x topology
since the trace map y , although continuous for the strong topologies
of BD(n) and L1(an) , is not. continuous for the weak x to~ologies of
these spaces. This is readily seen in dimension 1 n = 1 L, consi-. Jo,
der the sequence :
u (x) = 1 if x ~ 1/n u (x) = nx if 0 ~ x ~
n n n
Then u (0) = 0 and lu IB ~ 2 . From this last estimate we deduce that
u con~erges in BD(n) ~ea~ x to u = 1 the trace of which at x = 0
i~ 1 and not 0 . This example proves that K1 is not closed but it
also proves that the linear form
u E BD(n) guds
----+- ~
1
is not continuous on BD(n) endowed with the weak x topology. Therefore
K2 is not closed either. Consequently 4> +IK is not a l.s.c. functional
on X, and we have to compute its l.s.c. hull, also called~ cto~une
defined as the largest l.s.c. function minorizing 4> + IK ,

4> +IK(u) = lim inf (4> + I~un) (2.27)


u ~ u
where the topologY T is understood to be the weak x topology on BD(n)
Computing the above closure is a difficult mathematical problem, and
heuristic mechanical considerations greatly help to guess the result. For
this purpose we interpret more thoroughly the terms entering (2.11) :
1) IT(u) is the energy dissipated in a rigid plastic body by the mecha-
nism u We classically know that u can exhibit discontinuities
across the body (BD(n) accoun~ for this fact). However discontinuities
might also appear on the boundary as a result of its possible plastifica-
tion. Therefore we must abandon the commonly admitted idea that the mate-
rial body occupies an open ~et, and consider that the body oeeup~~ the
cio~ed ~et ~ , since the boundary may contribute to the dissipation.
In other terms ~he mechanism should be defined on ~ = n U an by a
couple <uln. u 1 an) and the total dissipation should read as

IT(u, u )
A +
=f 1r((u) )dx + J b +
1T (u - u)ds (2.28)
n an
where 1Tb(x,A) = 1T(A ~ n(x))(+) (2.29)
s

(+) It should be noted that, even if 1T does not depend on the space
variable x , 1Tb may depend on X through the normal unit vector
n(x) .
Discontinuities and Plasticity 291

nb is the dissipation due to a strain localization on the boundary


(u+ ~ u-) . Expressions (2.28) and (2.29) are consistent with (2.25).
2) Boundary conditions are to be satisfied by u+lan sin~e they are im-
posed by a "tool" external to the body n . Therefore u lr = 0 and a
displacement rate field is defined by (uln, u+jri) . 0

3) L(u) is (up to the factor A) the power of external forces. There-


fore the power of body forces f should be evaluated in the displacement
rate u1 0 , while the pow~r of surface forces g should be evaluated
in the <hsplacement rate u Ir I
- + +
L(u, u ) = J fudx + ~ gu ds
n I
Finally, heuristic mechanical arguments lead us to the following kinema-
tical definition of the limit load
-k - + - +
, (u,u+) E HI(n) 3 x HI/ 2 (ri) 3 } (2.30)
A = In {rr(u,u ) , L(u,u )

IT(u,u+) = J n(E(u))dx + J nb(-u)ds + ~ nb(u+ -u)ds ~


n ro I
This definition may also be recovered basing on mathematical developments.
Let A < As , a a,,ociated with A by (2.6) , u and u+ be elements
of HI(n)3 and H 2 (ri)3 .
+
Jr a. n u + ds = A ~ gu ds ,
I I
J ae-:(u)dx- ~ a.n u ds = A J fudx + ~ a.n u+ds .
n o n I
Taking the difference between the above equalitie~ noting that by virtue
of the symmetry of a we have a.n u = a(u n) , taking into account
the ~efinition of thl su~portiY~nctiSn n , w~ recover (2.30) where
(u,u) belong to H (Q) x H (ri) . But our initial goal is not yet
reached, and at least tw~ qu~~tions need an answer
i) Does the equality A = A hold ?
ii) Do we obtain a well posed variational problem (lower semi continuity
of the "relaxed" functional) ?
The answer to the first question is provided by theorem 2.3
Theanem 2.3 ~~ume that ao~umptian6 (2.I)(2.3)(2.4) ane met. Then

~k(Inf(2.30)) = Ak(Inf(2.II)) (2.3I)


Pnoafi. In order to prove (2.3I) it is sufficient to prove that (2.30)
and (2.II) have the same dual problem. Computation of the dual problem
of (2.30) relies on the abstract result used in theorem 2.I We set :

V HI(n)3 xHI/2(ri)3, v = (u,u+) ' Y =n.. 2(Q) xHI/2(ro)3 xHI/2(ri)3

A (AI x A2 x A3 )(u,u+)- (E(u),ujro, ,/ -ujri)


292 P.M. Suquet

+
=
A

F(v) 0 if L(u,u ) I , + oo otherwise

G(p) GI(pi) + G2(p2) + G3(p3)

GI (pi) =J 7r(pi)dx ' G2(p2) = ~ 1Tb(p2)ds


' G3(p3) = fr 1Tb(p3)ds
n o I
Problem (2.30) is now under the standard form :

Inf [F(v) + G(Av)]


vEV

Le.mma 2. 1 UndeJt a.6.6 ump:U.o M ( 2. I )

-
. PI:t + "f
d lV ' 0
:t :t
p3 = PIn )..g on ri
(2.32)
:t :t
PIn p2 on r 0
b) :t :t :t
F (J\ p ) = + oo

P~oo6. By definition

F:t(A:tp:t) Sup [(A:tp:t,v) - F(v)] = Sup (p:t,Av) A

vE V L(v) =I
Sup J :t r :t r:t+
p 1 E(u)dx + lr p 2 u ds + 1' p 3 (u - u)ds (2.33)
(u,u+) E V f2 o I
+
=
A

L(u,u )
+ +
Let (u ,u ) be an element of V such that L(u ,u ) I . For every
0 0
( u,u +) o.1n0 V we d e f'1ne
--+ + + +
(u,u) = (u,u) + (I- L(u,u ))(u ,u)
A

0 0

and we check that :


A--+ --+
L(u,u ) = I , (u,u )E V
--+
Thus (u,u ) can be used in (2.33) :

Su~ [J p~ E(u)dx + fr p~uds + ~ p;(u+ -u)ds


(u 'u ) n 0 I

- )..(1-L(u,u+)) A J (2.34)

where )..=-Jn
p:ti(u)dx-r p:t
o "r 2 uds-r
o
p:t (u+-u)ds
1' 1 3 o o
I~ a first step the sup in (~.34) is lim1ted to u in V(n) 3 and
u = 0 , and we obtain
Discontinuities and Plasticity 293

J (p~ E( u) + Afu) dx
~
The last supremum is +co except if

*
- di v pI + A f = 0 (2.35)

~erif~re F*(A *p*)


= +co except if (2. 35) holds, and we evaluate
F (A p ) for those p *
only. Coming back to (2.33) we obtain with
the help of Green's formula

+~ (p~.n-p;)uds]
I
+ .
u Ir ' u Ir ' u l r I can be choosen ~ndependently and the above supremum
is 0 + co elce~t ~f all the conditions (2.32) are satisfied, in which
case F*(A*p ) = - A This completes the proof of lemma I To end the
proof of theorem 3 we have to compute G*(p*) , but this computation
is standard (and rather technical for G~ and and we obtain : c;)
pI* (x) E P
{
0 if a.e. x ~n ~
* =
GI* (pI) (2.36)
+co otherwise,

and we use lemma 6.2 of TEMAM [2] to show that

under the same conditions (2.38)

The abstract result ( 2.2) together with lemma 2.1, (2.36),(2.37),


(2.38), completes the proof of theorem 2.3 .
A simplified form of the relaxed problem can be given. Acheving first
the infimum with respect to u+ we compute part of the relaxation term :

Inf{fr 1rb(u+-u)ds, L(u,u+) =I, u+E HI/ 2 (ri) 3 }


+ 1
u
= lll I -L(u) I (2.39)

where

\1 =Sup {aE 1R lag(x) E C(x) a.e. xE ri}

C(x) = dom(lf b (x,.) ) * (2 .40)


b
1T(x,.) is a positively homogeneous function of degree one and thus is
the support function of a closed convex set ~n 1R3 denoted by C(x)
Lemma 6.3 of TEMAM allows to characterize C(x) as the set of all
294 P.M. Suquet

stress vectors a .n, when a is plastically admissible and has a smooth


divergence. In other terms

l.l = Sup{a E lR I ja, a(x) E P a.e. x in n '


div a E 13 un 3 , a .n(x) =a g(x) a.e. x 1n (2.41)

The proof of (2.39) (left as an exercise to the reader) relies on the


abstract result recalled in section 2.2 .
We are now in a position to simplify (2.30) with the help of (2.39)

>..k = Inf{IIR(u) + l.l II- L(u) I


} (2.42)
IIR(u) = II(u) + ~ nb(-u)ds
0
(2.42) makes contact with TEMAM & STRANG's work who first introduced the
following "relaxed" problem
k 1 3
>.. = Inf{IIR(u) , u E H (!1) , L(u) = 1} (2.43)

Equality between the infimum in (2.42) and (2.43) results from equa-
lity of the dual problems. However it should be noted that the constraint
L(u) = 1 in (2.43) is not closed.
Re~k. (2.30), introduced in BOUCHITTE & SUQUET [20] seems to be
original (to the author's knowledge), although the heuristic argument
leading to (2.30) was already exposed in SUQUET [21] [22] . However
the simplified form (2.41) bears a strong resemblance with the relaxed
problem introduced by DEMENGEL [23] .
2.5 RELAXED PROBLEM II, FUNCTION OF A MEASURE
We argued in section 2.3 that displacement rates naturally live in
BD(!1) and it is readily seen that u+ should be taken el~ment of
M1(r 1)3 . Consequently we need to extend the definition of II to elements
of these spaces. The case of an homogeneous material satisfying (2.4)
offers no difficulty 'since

n(e) = klel ~(q) = klq ~s nl (2.44)

Every bounded measure on !1 admits an absolute value (BOURBAKI [13])


and this allows to define II(u) for u in BD(!1) . Assuming that !1
has a cl boundary (assumption (2.1)) we can define (q ~ n) .. as a
I s 11
bounded measure on r 1 as soon as q belongs to M (r 1) . Therefore
this particular case offers no difficulty. The general case (P(x) in a
general form andAdepending on x) is more difficult and the appropriate
tool to extend II to irregular displacement rates fields is the so cal-
led theory of 6unilion.6 o 6 meMU/l.e.-6.
This theory takes its roots in GOFFMAN & SERRIN's work (no dependen-
ce on the space variable). It has been improved by DEMENGEL-TEMAM [16] ,
GIAQUINTA .& al [17] , VALADIER [l8] , HADHRI [l9J , BOUCHITTE [4]
and applies now to non homogeneous materials. The following result is no~
Discontinuities and Plasticity 295

op~mal but contains main desirable informations for mechanical purposes


in the present context.
Function of a measure. An abstract result
-----------------------------------------
Consider a functional J defined on M1 (n) m by
1 m
I j(x,h)dx if ~ = h dx, h E L (n)
](~) = { n (2.45)
+oo otherwise ,

where j obeys a "linear-growth condition"

(J A
0
> O)(j bE L1 (n)) j(x,z) ~ A lzl - b(x).
0
(2.46)
We wish to extend the definition of ] to all ~ in 1 (n)m , through
the l.s.c. closure of J for the weak * topology of M (n)m . For this
purpose we need additional notations
i) j"" is the p!U..n.upal paM of j

n(x,z) lim ~ j(x,tz) (2.47)


t~+oo

n(x,.) is the support function of IP(x) where

P(x) = dom(j*(x,.)) (2.48)


ii)

I j*(x,.p(x))dx < +oo} (2.49)


n
and h(x,z) = Sup (.p(x) .z) . (2.50)
.p E E
Note that h can be different from 1T
iii) Given a bounded measure ~ on n , Radon-Nikodym' s theorem allows to
define its integrable and its singular parts with respect to the Lebesgue
measure dx
1
~ E L (Q) , ~
a s
Then we can state the following (not optimal but sufficient here) theorem.
Theo~em 2.4 A6~ume that j obey~ (2.46) an.d mo~eov~ that
.<.) j ,w 1-~.c.. on. n x lRm
U) 3 cE L (n) IHx,O) I ~ y<x) !i.e. X En
Then. the J's cl.M~e .<.n. M (n)m weak* A/.l
d~
}(~) In j(x,~(x))dx+I h(x,~)dl~sl
a Q u I "'s I
(2.51)

I6 mMeove~
296 P.M. Suquet

( z E IR.m) n(., z) ~ u.ppe!t .6em.i. c.on.tin.tWU-6 (u .6. c..) the_n. h = n ( 2. 52)

Comme.n..:t6. Proof of theorem 2.4 is beyond the scope of these notes.


However we can interpret a few of its conclusions :
a) If "the material is homogeneous", i.e. if j does not depend on x
we require only that j is l.s.c. with respect to z . } is given by
(2. 51) (2. 52). Following examples illuminates the role played by the prin-
cipal part of j in the definition of ] on singular measures with
respect to the Lebesgue measure. Let m = I ,
rl = ] -I , +I [ , and consider the usual
n
approximation of the Dirac measure o by
stepwise functions 0

I
;I/2n 0 if lxl :;: 2n
u (x) = { 1
n
n if lxl ~
2n
Then ](u ) is finite and
n
J ( u ) = _!_ j ( n) which tends to n(l)
n n
Therefore
J(o ) = n(l)
0
b) J is oytained as the hi-conjugate function of J (J = Jxx) in the
duality (M (rl)m, C0 (rl)m)
c
}(lJ) = Sup [<lJ,<P> - J (2.53)
rl
cpEc 0 (rl)m
This duality relaion explains the role played by c.ontin.uoU-6 functions
in the definition of h We give an example (taken from BOUCHITTE [ 4])
which illustrates this point :
j(x, z) = k(x) lzl = n(x,z) where k(x) E L00 (rl) , k :;: 0
J(u) J k(x)luldx u E L 1 (rl)
rl
Then [4]

J ( ]J) J k(x)di]JI (2.54)


rl
where k 1S the closure of "'k rl + IR+ U { + 00}

"'k(x) lim J
1
k(y)dy
p+O IB(x,p)l B(x,p)
(B(x, P) is the ball centered in X with radius p) Indeed

I<P(x) I ~ k(x) a.e. in rl} (2.55)

Averaging the inequality contained in (2.55) on balls yields easi-


ly
Discontinuities and Plasticity 297

and by lower semi continuity

l~(x) I~ k(x) for eve~y x in ~


Therefore an alternate (and more precise) definition of E (2.55) is

E = {~ E C0 (n) , l~(x)
c
I ~ k(x) for every x in ~} (2.56)

(2.54) follows easily from (2.56)


c) From its true definition, h is a l.s.c. function with respect to z .
Note that it is also a l.s.c. function with respect to x . Indeed,
taking JJ n = z o in ( 2. 51) with x converging to X00 in ~ ,
yields (by 1. s. c~of j) n

J(zox ) =h(x"', z) ~ lim inf J( z <'\ ) = h(xn , z) .


00 n++oo n
The following consequence is worth noting : we come back to the
preceeding example and assume that k(x) is piecewise constant :
k(x) = k 1 if x E ~~ , k(x) = k 2 if x E ~ 2 where ~~ , ~ 2 are open
subsets of ~-'- ~~ n ~ 2 = f/J ~~ U ~ 2 ="IT . k is not precisely defined
on S = ~~ n ~ 2 (whicn a set of zero measure) but it results from the
l.s.c. of h (or directly in this case from (2.54)) that if we want
to define k everywhere its correct value on S ~s inf(k 1 , k 2 ) Ex-
tending this remark to materials with a limited strength,contact is made
with the following HADHRI' s result [24] : consider a nonhomogeneous
material, the strength domain of which takes two values,_ P 1 _in ~~ ,
P 2 in ~ 2 , then the strength domain for x in S = ~~ n ~ 2 is
P 1 n P2
d1 Theorem 2.4 is devoted to closures of functionals defined on
M (~)m , but we are mainly interested in functionals defined on BD(~)
Although these notions are obviously related, there is a gap between them:
let J be defined on Ml(~) by (2.45) and set

J(u) = J(E(u)) , J ,(})closures of J,(J) in BD(~), (M 1 (~) 9s )


It is clear that J ~] but the equality is not obvious and requires
additional assumptions on the dependence of j with respect to the space
variable x (see for instance DAL MASO [25]) However in case of an
homogeneous material the following theorem due to BOUCHITTE [.26] provi-
des the desired closure.
Theoltem Z. 5 (BOUCHITTE [Z6] ) Let j : IR9 + IR u { + "'} be a c.onvex,
t.~.c.., p~op~ 6unc.tion ~~6ylng s
<1o., S, yE IR~) o.lzl- Y ~ j(z) ~ S(lzl +1) z E IR.9
s
and J de6lned on Ll(~)3 by

{ 1
j(E~))dx in
u E H 1 (~)) , u 0 on
J(u) = ~
+ otheJtW~e
Then the elo~~e o6 J in L 1 (~) 3 ~
298 P.M. Suquet

J(E:(u)) + J rr(-u 0
s
n)ds u E BD(rl)
J(u) Clrl
+"" o.:the/WJ~e.

Theorems 2.4 and 2.5 have two consequences, ~n case of an homogeneous


material submitted to body forces only
a) IT (u) can be extended to elements of BD(rl)
b) t~e obtained extension defines through (2.30) (where u lies
in BD(rl) , and u+ does not appear since r 1 = 0) the relaxed problem
associated with (2.11) (uniquely defined as the variational problem for-
mulated on the closure of the initial functional).
2.6 BIBLIOGRAPHICAL COMMENTS, OPEN PROBLEMS.
I) We have often assumed a very particular form (2.4) for P .
Examples of more appropriate strength domains are provided by Von Mises
or Tresca criterions in which only the deviatoric part of o is limited.
By a duality argument, kinematically admissible vector fields for the
limit analysis problem are ~n~omp~~~~ble. This constraint complicates
significantly the analysis, mainly in approximations by smooth functions
of kinematical fields. The reader is referred to TEMAM 12] for additio-
nal details and references.
2) To the author's knowledge (2.30) is a new form of the relaxed
problem of limit analysis. However, in case of a nonhomogene.oUD ma.:te~a
the author is not aware of minimal assumption to be imposed on material
variations in order to extend (2.30) with the help of functions of
measures, into a variational problem formulated on a l.s.c. functional.
3) Numerical resolution of limit analysis problems often require
~egulaniz~on, on which we shall comment in the context of elasto-
plasticity (see next sections). The reader is referred to MERCIER CnJ ,
FREMOND-FRIAA [2 ~ , JOHNSON [29] and their references which are
related to the approach developped here, although mostly based on conti-
nuous approximations. Other relevant references on limit analysis are
CHRISTIANSEN [30] , and NGUYEN DANG HUNG f)U .
4) Regularization is mostly used to obtain a d{&&e~en~able functio-
nal to minimize. However advantage can be taken of the linear growth of
the original functional, by using linear programming techniques. For addi-
tional information on this point of view, mostly developped by an Italian
school, the reader is referred for instance to GAVARINI [32]
Discontinuities and Plasticity 299

3. ELASTO-PLASTICITY

The previous section discussed the maximal value of an applied load


schematized by the loading parameter A The present section deals with
a question which naturally follows that of the limit load : can a load
A < Ak = As be supported by the material ?
The answer to this question is no in general, as, proved by the fol-
lowing example, but y~ in the context of elasto-plastic materials as
described below. Let us show however that limit analysis applied to brittle
materials overestimates the load capacity of a structure.
!2~2!!11!:
fixed a
//Ill/./

.
2

E2
Material Haterial 2

Figure 3.1

Considers two parallel unidimepsional bars of unit cross section, in


parallel, made from two brittle materials and loaded by a unique
,, force M
Ultimate stresses and strains of each phase are denoted by a 1:1 and E
1 1
Equilibrium reads as
a = constant (a.) 1n each phase ,
1
a 1 + a2 = A
Therefore the limit load defined (statically) by (2.6) is
s M M
A = a 1 + a2
300 P.M. Suquet

But (2.6) does not account for the limitation of strains ~n both phases
M M M
E:i = 2 ~ Inf(1 '2) = 1
Therefore the maximal stress which can be reached by phase 2 , (as long
as phase I ~s unbroken), is
M
a2 E2 I
and the resulting maximal load (two phases unbroken) is
M M M M
A= a 1 + E2 1 < a 1 + a 2
However a higher load could be reached if phase breaks
M
A= a 2 (phase broken : a 1 = 0)
In conclusion the maximal value of A which can be associated with a phy-
sical state of the material (both a and satisfying equilibrium equa-
tions and constitutive law) is
,Max M M M s
1\ = Sup(a 2 , a 1 + E2 1) < A
This example shows that the limit load determined in section 2 ove~
eo~mateo the strength capacity of a brittle structure. However in case of
elasto-plastic constituents the limit load is exact, in the sense that
with every A < As we can associate at least one physical state (a , E:)
of the structure satisfying all the requirements of equilibrium and cons-
titutive laws.
3,1 TWO MODELLINGS OF ELASTO-PLASTIC BEHAVIOR
The typical stress-strain curve of an elasto-plastic material is plot-
ted on Figure 2.2 and two constitutive laws modelling this behavior have
been proposed.
g~g~~y~~-~~~! The total strain splits into an elastic part and a
plastic part ,

The elastic part depends linearly on the stress tensor


e
E:ij = Aijkh akh in short) , (3. 2)

while the plastic part is given by

(3. 3)

P is the strength domain, its interior is the elasticity domain (E:p = 0


if a E Int P) , its boundary is the yield surface. Note that (3.3)
forces a to stay in P, otherwise oi would be empty. A = (AiJ.kh) and
P may depend on x when the material pis nonhomogeneous.
Prandtl-Reuss'model The total htnain-~ate splits into an elastic part
and-a-plastic-part-
Discontinuities and Plasticity 301

e p
(u) = + (3.4)
(an over dot stands for time derivative). The elastic part depends linear-
ly on the stress rate


e
A a (3.5)

while the plastic part is given by the flow law

(3.6)

The main difference between these two models is that Hencky's law is
a model of (physically) nonlinear elasticity rather than a model of plasti-
city, and therefore does not account for the irreversible character of
yielding, while Prandtl-Reuss' law really accounts for the difference of
behavior between loading or unloading paths.

-- --
a a
loading loading

K---------~

Hencky Prandtl-Reuss

Figure 3.2

Mathematically these two models are again different. Hencky's law


leads to minimization problems for convex functionals, whi~Prandtl
Reuss' equations lead to evolution problems. However they have several
common points : discontinuities of displacements (or displacement-rates)
arise in both models, and time-discretized Prandtl-Reuss equations are
close to Hencky's formulation. Therefore we study in a first step Hencky's
model.
Remank. Alternate writings of (3.3) or (3.6) can be proposed. The
more significant one takes the form of HILL's pninciple o6 maximum wo~k

-
(3. 3) ~ aE p ( T E P) (p '[ - a) ~ 0 (3. 7)

(3. 6) a E p ( T E P) (~P ''[ - a) ~ 0 (3.8)

3.2 HENCKY'S MODEL


Governing equations of the problem are summarized below
302 P.M. Suquet

div a + A.f = 0 in n

a.n A.g on r1 u = u on r (3.9)


0 0

E(u) = Aa + EP EP E arP(a)

Y~!i~!i~g~!-~~!~~!~!i~g-~~--i~~2l __ ig_!~!~~-~~-~!!~~~~~ [ ~
The stress analysis of (3.9) is classical (cf DUVAUT-LIONS 30J).
We assume that P(x) is constant on borelian subsets of n and we set :
ll' = {T E ILZ (Q) , T(x) E P(x) a.e. x E Q}
where throughout the text we denote as previously
ILP (n) = Lp(n): ,
and recalling the space of statically admissible stress fields
S(Af' A.g) ={TEIL 2 (n) 'div T +Af = 0 a.e. in n, T.n = Ag a.e r1}
we set
KA ll' n S(Af , A.g)
Choosing T in KA , we obtain by Green's formula

J E(u)(T-a)dx=Jr u(T-a)nds ( 3. 10)


n o o
and from the constitutive law (cf (3.7))
Jn E(u) (T- a)dx ~ Jn A a(T- a)dx
Therefore the variational formulation of (3.9) in terms of stresses is

F-ind a E KA J..uc.h thOvt } (3.11)


(:V T E KA) A(a , T- a) :;: R-(T- a)

where

A(a, T) = Jn AaTdx , R-(T) = Jro u


0
T.n ds (3. 12)

In order to formulate a very classical result of existence and uniqueness


for a we need additional assumptions

u E H 112 (r ) 3
0 0
A is coercive, symmetric, bounded :
(3 2
a.> 0) (:V E;, E JR 9 ) (a.e. xE Q) Aijkh (x) E;,ij E;,kh ~alE;, llR9
s
s (3. 13)
Aijkh = Ajikh = ~hij
Aijkh is measurable and
(j f3 > O)(a.e. X En) (:V ijkh) I (Aijkh(x) I ~ f3
Discontinuities and Plasticity 303

P~opo~~on 3. 1 A~~ume thaX (2.1)(2.3)(3.13) ane met, and thaX


A < As de6~ned by (2.6) Then theM eWU a uMque a Mlution o6
(3. II) . Mo~eove~ a ~ Mlution o6 the ~M~zation pMblem
I (3.14)
Inf 2 A(T,T) - ,(T)
T E KA
The assumption A < As ensures the non vacuity of KA The proof
of proposition (3.1) is classical (DUVAUT & LIONS [33] , TEMAM [2] ,
PANAGIOTOPOULOS [9] , NECAS- HLAVACEK [34]) . Denoting by (ae , ue)
the elastic response of the material to the same loading (Af , Ag , u ) ,
it is readily seen that a is the projection of ae on KA for thg
scalar product on n) (~) associated with the bilinear form A(.,.) (3. 12).
Y~Ei~!iQg~l_iQE~gl~!iQg_ig_!~~~-Qi_~i~El~~~~g!~_ii~~--I~ll
Let us set for ~ in 1R
s
jx(x,~) = ~ A(x)~ ~ + IP(x) (~)
and for e ~n JR9
s ( +)
j(x,e) = (j~(x.))x(e)

~ J~ j(x,E(u))dx
J(u) (3.15)
~ + 00 otherwise

It should be noted that j has a linear growth at infinity, s~nce


P = dom jx is bounded (cf (2.4))
(Jk 0 ,k 1 ,CEJR:) (a.e. xE~) k 0 lel-c,:::j(x,e) ,:::k 1 (1el +I)
Therefore the natural domain of definition of J ~s BD(~) provided
that we take J's closure for the weak x topology of BD(~) . This is
technical in case of a nonhomogeneous material and we shall therefore
assume that

A and P do not depend on x ,


and again to avoid technical difficulties, (3.16)
A ~ ~o~op~c.

The minimization problem ~n terms of displacements (the dual of which ~s


(3.14)) reads as
b
Inf (J(u) + j r 1T (u -u)ds- H(u)) (3.17)
0
uEBD(~) o

(+) j is the (nonquadratic) density of elastic energy for Hencky's model.


304 P.M. Suquet

where L(u) = frl fr


fu + gu ds
The following theorem i~
a compendium of TEMAM's results [i] , slightly
adapted to the present context :
Th~o~~m 3.1 A~~um~ that (2. 1)(2.3)(2.4)(3.13)(3.16) an~ m~t, and that
A < As . Th~n
..{.) ( 3 . 14) ~ th~ dual. o 6 (3 . I 7)
.{...{.) Th~ 6unctiona (3.17) ~ l.~.c. on BD(rl) w~ak x .
.{..{...{.) Th~ m.{.n.{.m.{.zation p~obl~m (3.17) adm.{.U at l~Mt on~ ~olution.

Comm~~. Proof of theorem 3.1 is beyond the scope of these lectures,


however we can comment its results. The main point is that, although L
is not l.s.c. or u.s.c. on BD(rl) weak x (3. 17) is l.s.c. under the
assumption A < As . Existence of a solution follows easily from this
result, which is highly technical and follows mainly KOHN-TEMAM [35]
Extension of these results to nonhomogeneous materials, basing on recent
results on functions of measures, is completed in HADHRI [36] .
Basing on the variational properties (3.14) or (3.17) , numerical
approximations of solutions for elastic plastic problems can be proposed.
The kinematical formulation is (as usual) the most convenient one, although
is should be performed with discontinuous shape functions. The reader is
referred to BEN DHIA [37] , HADHRI -BEN DHIA [38] and the bibliography
of [2] for additional information on this subject. But the (relative)
flexibility of the kinematical approach should not hide that the interes-
ting unknown from a mechanical point of view is the ~~~~ t~~o~ a ,
which yields the whole information on plastified zones. However the stress
formulation (3.11) is not appropriate to (internal) finite element
approximations since it contains a constraint in divergence form (equi-
librium equations). Probably motivated by this argument, JOHNSON [39]
introduced for the Prandtl-Reuss' material a mixed formulation (both a
and u are unknown) which we describe now in the present context.
Mixed variational formulation
-----neI~e-i-Bi~ach-~?ice--s by =

S = {T E IL 2 (rl) , div T E L3 (rJ) 3 }


and choose T in S n lP With the help of Green's formula and of the
constitutive law (we use (3.7)) we obtain, on the one hand
JrlE(u)(T-a)dx = -frlu(div T- div a)dx + J,r uo (T- a)nds + J,r u(T-a) .nds ,
0 1
and on the other hand
Jrl A a(T- a)dx ~ Jrl E(u) (T- a)dx ('V T E lP) .
Since we expect solutions u to be discontinuous, even on the boundary
arJ ' we introdyce a separate displacement u+ on rl ' which is expec-
ted to be in M Cr 1) 3 Therefore we have to require additional regulari-
ty for g and T.n

(3. 18)

and we search for a 1n the Banach space


Discontinuities and Plasticity 305

S
c = { T E TI.. 00 (rl) ,
d~v t E L (rl)
3 3 , T .n E C0 (f.) 3 }
I
The proposed mixed variational formulation is

F-Ln.d (a, u, u+) Esc n lP x BD(S'l) x M1(r 1) 3 l.lu.c.h tha-t

A(o 'T-o)+ b1(u,T-a) + b2(u+' T-o)::! (t-a)VtE sCn]p


(3. 19)

where div T dx b2(ll 't) -fr t.n d\.1


I
-J g d]J
ri
The.otte.m 3. 1 [4 0] AMume tha-t (2.1) (2.3) (2.4) (3.13) (3.18) Me me_t +
and that A < As . The.n. (3.19) adm-i;u a-t .te.Mt on.e. Miu.-ti_on. (a , u , u )
a ~ u.n.-Lq u.e. .
Pttoo6. The proof relies on a regularization procedure.

-- --~-(:~~;--= +
Regularization Define
I T - II ( ) T 1 2 9
ll ll Px 1R
s
V = {u E H 1 (rl) 3 u = uo on r }
ad 0

and consider the following problem

F-Ln.d (all , u ll) E S ( Af , Ag) x Vad .wc.h tha-t


(3.20)
Aa\.1 + d~ (oil) = E(u\.1)
\.1

(3.20) admits a unique solution which has the following varia-


tional property
all Arg Min {i A(t,t) +
\.1
(t)- (t) , t E S(Af,Ag))}

Arg Min{[i A(.,.) + (.)]*(E(v)) - AL(v) , v E V d}


ll a
~-EE~~E~-e~!i~~!~~ Note that inequality A < As impl~es that the follo-
wing l.la6e.1y Ml.lump:t-Lon. is met

(j X E S(Af,Ag) n IP)(J o> 0) l.luc.h tha-t


(3.21)
ltl"" <a ~x+tEIP
n. (rl)
To check (3.21) choose n such that A < n < As , and let a be an
0
element ofA S(nf,ng) nIP
Then X = - a satisfies all the requirements of (3.21) , with
n o
306 P.M. Suquet

o -- n->.la
2n o
I oo
(Recall that W contains 0 as an intericr point).
Then taking a~ - X as trial function in (3.20) , we obtain

(3. 22)

But monotonicity of a~ and noting that a~ (X) = 0 , we have


~ ~
In a~~(a]..l)(a]..l- x)dx ~ 0 ,
and by Green's formula

In e:(u].J)(a].J-x)dx =~ u0 (a].J-x)nds= Ine:<ue)(a].J-x)dx, (3.23)


0
where ue denotes an element of H1 (n) 3 with trace u on r There-
0
fore we deduce from (3.22) that 0

ala].JI 2 2 ~ cla].JI (+)


1L (n) 1L 2 (n)
i.e.

a]..l is bounded in 1L 2 (n) (3.24)

Coming back to (3.22) with the help of (3.24)(3.23) yields

0 ~ <I> (a].J) = <I> (a~) - <I> (x) ~ In()~ (a~) (a]..l- x)dx ~ c (3.25)
].J ].J ].J " ].J

Lemma 3. 1 AMume .tha;t


In Cl~/a].J)(a~ -x)dx ~ c
Then
~~ (a~)l 1 ~ c
PJtoo6. ].J 1L (n)

I
I a~ (a].J) I 1 =5 Sup In ~ (a~) .T du (3. 26)
~ (n) IT Ii" {n) ~ 0
~
1L

But

(3.27)

The first integral of the right hand side of (3.27) is non positive : by
the safety assumption x + T E W , by ~~' s definition ~ (x + T) = 0
and the result follows from the monotonicity of ()~ The cgnjunction
of (3.26)(3.27) yields ].J

(+) Throughout the following C denotes a (varying !) constant which


does not depend on ~
Discontinuities and Plasticity 307

la..P (aJ.J)I 1 ~-t Ina..pJ.J(aJ.J)(aJ.J-x)dx


J.J IL (n)
and completes the proof of lemma 3.1 .
We deduce from (3.24) and lemma 3.1 that

E(uJ.l) is bounded in M1(n) 9 .


s
From the boundary condition u = u on r0 and (2.19) (with
r(u) = Ir lulds) we obtain that 0
0
uJ.l is bounded in BD(n) (3.28)

~~~~~~8-~~-fh~-l~~~~--~-~-2 .
Fro~ (3.2f) and (3.28) we d~duce that there exists
(a , u , u ) E IL (n) x BD(n) x M1(r 1) such that (for a subsequence)
aJ.J - - 4 a in IL 2 (n) weak ,

uJ.l ~ u in BD(n) weak x


uJ.ll ~ u+ in M1(r 1) 3 weak x
r1
Moreover a belongs to S(Af , Ag) which is a closed convex subset of
IL2 (n) Therefore the two last assertions in (3.19) are satisfied.
Then we note from (3.25) that :

Jc, laJ.l -II aJ.ll 2dx = 2J.J q, (aJ.l) ~ 2 ].JC (3.29)


" p ].J

The lower semi continuity on IL 2 (n) of the functional F below, together


with (3. 29) , proves that a belongs to ll' ,
F(T) =In
IT- IIPTI 2 dx.
It remains to establish the first assertion of (3.19) For this purpose
we note that for every T in Sc n ll' we have

A(aJ.l, T- aJ.l) +In d'{)/aJ.J)(T -aJ.l)dx + b 1(uJ.l, T -aJ.l)

+ h 2 (uJ.JI , T -aJ.J) = R.(T -aJ.J) (3. 30)


r1
By l.s.c. we have

- A(a ,a) ~ lim sup [ - A(a ].J , a ].J ) J , (3.31)


].J + 0
and by monotonicity

(3.32)
Moreover
308 P.M. Suquet

Passing to the limit in (3.30) with the help of (3.31)(3.32) establishes


the first a~sertion of (3.19) The proof of theorem 3.2 ~s complete.
3.3 PRANDTL-REUSS MODEL
We summarize equations of the problem

div a + :U= 0 in rl

a.n f..g on rl ' u = u


0
on r0 (3. 33)
du) = Aa + ~p ~p E diP (a)

Note that (3.33) is now an ~vo~ution p~ob~~m for a and therefore re-
quires data of f..f, f..g , u on a time interval [O,Tj , and initial
conditions for a , while 0 u interfers in (3.33) by its time-derivative
u only, which is the kinematical unknown. For simplicity only,we assume
that ,>. is a function of t , while f and g are function of x only

,>..(.) E w1 "'co,T)
The initial stress state a is assumed to obey
0
(3.34)
a ES(,>..(O)f,f..(O)g)niP
0
Moreover
u E WI '"' ( 0, T ; HI / 2 ( r ) 3 ) , me s ( r ) > 0
0 0 0

Stress formulation
-----si~c~--A--Is-now a function of time, the set of plastically and
statically admissible stress fields is a moving closed convex set in ~ 2 (rl)
K(t) = S(f..(t)f, f..(t)g) n IP .
Following a parallel path to that which led to (3.11) we can formulate
(3.33) in terms of stresses only,

I
Find a(t) E K(t) -6 u.c.h .that &o~ a.~. t -i..n Jo,T[
( T E K(t)) A(a, T- a) ~ i(T- a) (3.35)
a (0) a
0

We assume that the following safe load assumption is met.


~f~_l~~-~~~~~E~i~
<:lx E w 1 "'co,r;~ "'en))) x(t) E S(A(t)f,f..(t)g) t E]o,T[_ (3. 36)

<3 0 > 0) IT I "' (rl)


< 0 => X( t) + 0 E IP t E Jo,T[
~

Th~o~~m 3.3 ~-6u.m~ .that (2. 1)(2.3)(2.4)(3.13)(3.36) ~~ m~.t. Th~n


(3.35) admLt6 a u.n-i..qu.~ M~ution
a E w1 2 (0,T; ~ 2 (n))
Discontinuities and Plasticity 309

Let us show how (3.35) can be given the form of a "sweeping problem".
Define on 11 2 (rl) on new scalar product

((a, T)) = A(a,T) . (3.37)

Then (3. 35) amounts to a differential -lvtc!UI.!-lovt in 11 2 (rl)


a+ :nK(t)(a) 3 o l
a (0) = a
0
J
which has been studied in the seventies under the name of "swe~ing
problem". The reader is referred to BREZIS [41] and MOREAU L42] for
further details. Theorem 3.3 is a consequence of these works.
~!~E!~~~~~!-!~!~~-f2!~~!~!!2~ A stress-rate for~ulation of (3.33)
can also be established : it is the HODGE & PRAGER's variational principle
(see for instance ANZELOTTI [43]). Its dual problem (termed GREENBERG's
principle) formulates (3.19) in terms of displacement rates only. However
rates principles, which can be useful in theoretical discussions such as
stability of elastic plastic bodies, must be abandonned as for the nume-
rical resolution of the problem.
Mixed variational formulation
-----JOHNSON--t)9J--pr~p~sed-amixed variational formulation of (3.19) ,
further extended by the author [40] By similar arguments to those
developped in section 3.2 we obtain
+ 00 T~ --+ S c n =x BD(") x M1 (r ) 3 .6 u.c.h .tha:t 6oJt
F-Lvtd (aU u')
a.. e. tEJO,T[ '
' ' " 1 ...
A(o,T-a) + b 1 (u,T-a) + b 2 (u+,T-a) ~ i(T -a)(YT Esc n:n>)
b 1 (v,a) A.L 1 (v) (Y v E BD(rl))
b 2 (~,a) A.L 2 (~) (Y~E M1 (r 1) 3 ) (3.38)

a(O) = a
0

where i(T) u T.n ds


0
0

TheoJtem 3.4 [37] A6.6u.me .tha:t (2.1) (2.3)(2.4)(3.1~)(3.36) a.Jte me.:t.


Thevt (3.38) a.drnd-6 a:t leM:t ovte Mlu.tiovt (a, u, u) . a -t-6 u.vt-lqu.e.
We briefly sketch the proof of theorem 3.4
B~g!!!.!!!:i~~!i2~ Following DUVAUT-LIONS [33] and JOHNSON [3~ we re~u
larize (3.19) with the help of a viscoplastic law due to PERZYNA [44j

AO~ + '?1P (a~) = a0


a~ E S(Af,A.g)
~

on r
0
} (3.39)

Existence and uniqueness of a solution to (3.39) are easy consequences


of the CAUCHY-LIPSCHITZ theorem. Indeed let (ae , ue) be the elastic
310 P.M. Suquet

-ll
response of ll to the loading ( H , /..g , u ) . Then cr 0 ll _ 0e ~s

solution of a differential equation in the Hilbert space


0 S(O,O) endowed
with the elastic scalar product ((.)) (cf (3.37))

;-ll + B( t , cr ll) h ( t) in S ( 0, 0)
}
=
(3. 40)
-ll e
(J (0) = (J - (J (0) '
0

where the Lipschitz operator B(t, .) on S(O,O) , and the second member
h in S(O,O) are defined, with the help of Riescz theorem, by
((B(t,cr),T)) = frl ~ ll (a+ cre(t))T dx
((h(t),T)) = -fn" Acre, dx + Jr u
o
T.n ds
0
(3.40) is a usual differential equation, with a Lip~chitz oferator, and
admits a unique solution. We define an element of L (O,T;~ (rl)) by
e(t) = A all
(crll) - E(ue) + ~
ll
which satisfies for almost every t in Jo,T[
Jrl e ( t) T dx = 0 V T E S(O,O)

Lemma 3.2 , the p2oof of whih is left to the reader, implies the ex~s-
tence of ull in L (O,T;HI(rl) ) such that
.!.ll ....:..ll
e = E(u ) u 0 on r
0
and we set ull = ~ll + lie (all, ull) satisfy ( 3. 39) .
Lemma 3. 2 Le:t e -<.n ~ 2 (rl) be Mc.h :that .Jr2 e T dv "' 0 6ofL e_ve_fLu T

-<.n S(O,O) The.n :th eJr..e. e_xi!.,V., a un,i.que v --LYt H 1 (rl) 3 ~.>uc.h :that ~

e = E(v) v = o on r 0
~-2EiEi-~~!i~!~~--l Multiplying (3.39) by crll - X (where X satis-
fies the safety assumption (3.36)) yields with the help of Grornwall
lemma
co 2
is bounded ~n L (O,T; ~ (rl)) (3.4I)
I
In <1P (crll) (crll -x)dx
ll
is bounded in L (O,T), and by lemma 3. I,

a..P (all) is bounded in (3.42)


l1
~-2EiEi-~~!i~!~~--!! We multiply (3.39) by all - X , and note that
(use the safe-load assumption) T
I IT
0
In a<P(l1 cr ll) (all - x) dx I ~ <l>
ll
( cr ll ( T)) + c I0 I aP
ll
( cr l1) I
~
I
(r~)
dt ,

Ifr~ ll
E( u ) (a
ll .
- X) dx I = IJnr ( ue ) ( "ll (J -
. I ll
X) dx I ~ c a - XI 2 '
~ (n)

and it follows readily from (3.4I)(3.42) that


Discontinuities and Plasticity 311

all is bounded in L2 (0,T;lL 2 (SG)) (3.43)

Multiplying (3.39) by all - x


again, and using (3.41) (3.43) we obtain
2
In C31P ll (all) (all - x) dx is bounded in L (O,T) ,
and by lemma 3. 1 ,
'diP (all) is bounded 1n L2 (0,T;lL 1 (n))
ll
i.e. by (3.39) and (3.43)
dull) is bounded in L2 (0,T;:rr... 1 (n))
From the boundary condition till = u0 we deduce
till is bounded in L2 (0,T;BD(n)) (3.44)

tilll is bounded in L2 (0,T;M 1(r 1) 3) (3.45)


r1

f~~~i~;-~~3~~hH~H)(~:~4~(J.45) we fxtract from (all, till, ) a ~lllr


subsequence converging in w1 2 (0,T;lL (n)) x L2 (0,T;BD(n))xL (O,'r;Ml(r ) 3)
weak;: to (a, ti, u+) (+) Moreover a(t) befongs to S(U"; Xg) and 1
satisfies a(O) = o . In addition we note that
0
T T
Io In lo- rrpollldxdt ~ 2llio In<d~P/oll),oll-x)dxdt ~ 2Cll
By l.s.c. o(t) belongs to IP for a.e. t in ]o,r[. It remains to
prove the first inequality of (3.38) . The proof is similar to that of
section 3.2 , with the help of the following remark

(3.46)

Since we already know that weaQ eonve~genee holds in (3.46) it is


sufficient to prove that
lim sup A(oll,oll) ~ A(o,o) for every t in ]o,r[
Multipl~f~~ (3.39) by oll- o integrating on n x ]O,t[ and using
the monotonicity of (3!pll (note that a~(o) = 0) ' yields
J,Ot J,n Aoll oll dxds J,t J, ll J,t r (.e) ( o ll -o ) dxds
~ O n Aa adxds + OJn u

We can pass to the limit in the above inequality and we obtain


lim sup I0tA(crll,olJ)ds ~JOt A(o,o)ds,
IJ-+0ll
and since o (0) = o(O) = o
0
lim sup A(oll(t),oll(t)) ~ A(o(t),o(t)) ,
ll -+ 0

(+) if X' is the topolo~ical dual of a Banach space X ~ then the dual
of 12(0, T ; X) is 1{(0, T ; X') which is equal to L (0, T;X') only
if X is reflexive. Applying this result to X' BD(n) and
X' M1(r 1)3 successively explains the previous functional spaces.
312 P.M. Suquet

which completes the proof of (3.46) . The remaining of the proof of


theorem 3.4 is easy. More details can be found in PANAGIOTOPOULOS ~]
or SUQUET [40j
),4 EXAMPLES OF DISCONTINUOUS SOLUTIONS IN ELASTO-PLASTICITY
Theorem 3.4 lets open the question of Jtegula.JU;ty artd urt-<..querteM for
u . The two following simple examples show that u is rtot urt-<..que, and
may be discontinuous even when the safe load assumption (3.36) is met.
a) Discontinuous solutions(+)
N = I Q=]-1,+1[ A z I p = [-I ,+1]
f(x) = (l+x) if x < 0, f(x) = -(1-x) if x > 0, /\(t) = t

The initial condition for the stress ~s a0 = 0 The statically admis-


sible stress fields have the form
2 2
X
t(x,t) = C1 (t) -t(x+x2 ) if x~O, C1 (t) +t(x-2) if x~O (3.47)

where C ~s a function of t only.


T
ER..aJ.Jtic. MR..utiort.
The elastic solution ae ~s a minimizer for the elastic energy among
statically admissible fields
I +I 2
2 f_ 1 It (x) I dx
We find C = - t/3 Therefore the body remains elastic up to t 3
where the point x =0 plastifies.

f
T

-1
cT

Fi~nre 3.3

(+)
This example is an academic one and is not relevant from a physical
point of view. However it is relevant in the discussion of the mathe-
matical equations of plasticity.
Discontinuities and Plasticity 313

It can be easily checked that the safe-load condition is satisfied


up to t = 4 . For t E ] 3, 4 [ we know by the general theorem that the
stress solution is unique. It reads as
2 x2
a(x,t) = -1-t(x + x2 ) if x ~ 0 -1 +t(x- 2 ) if x~O (3.48)

To prove (3.48) it 1s sufficient to check that


+I
1-I O(T- a)dx ~ 0
for all T1 S in the form (3.47) , which is an easy task. The only plas-
tified point is x = 0 and outside this point the body remains elastic :
(. du
u) = dx = a
.1.
for x r 0
i.e. by integration over]-1,0[ and ]0, I [
2 3
u(x, t) = Cl (t) - ~ + ~ if X< 0 , = (3.49)
+ 2 6

a:+- and a are functions of t only and are determined by requiring


that u 0 for x = I we find

+
(l 1/3 (l 1/3
However we see from (A.3) that
u-(O,t) = lim u(x, t) = 1/3 u+(o,t) = lim u(x, t) -1/3
x<O x>O
x-+0 X -+0

The displacement rate u , which ~ uniqu~ ~n ~ ~xampl~ since the


plastic zone reduces to one point, is discontinuous 1n x =0
b) ~2~-~~ig~~~~~~-2!_!~~-~i~El~~~~~!-E~!~
N = I , S1 =] 0, I [ , P = [- I , + 1] , A
The body is loaded by an imposed displacement
u(O) = 0 , u(l) = t f = 0
The initial condition for the stress is a 0 .
0
EiMtic Miution. u = tx a = t The body remains elastic up to t =I.
EiMtic piMtic J.>Oiution. It can be easily checked.that the safe load
condition is satisfied 6o~ ~v~~y t by X = 0 . For t ~ I the stress
remains constant a(x) = I a.e. x in S1 , and the body is entirely
plastified. The mixed formulation (3.38) can be shown to reduce to
I
J0 u(div T - div a)du ~ T(l) - a(l)
and since a = I
I . dT
J0 u dx dx ~ T( I) - I (3.50)

(3.50) admits an infinite number of solutions. For instance u = 0 ,


u= u = x are continuous solutions but discontinuous solutions can
I ,
also be proposed (any stepwise increasing function. The solution u = I
314 P.M. Suquet

gives an example of loss of boundary condition at x =0 .


3.5 BIBLIOGRAPHICAL COMMENTS. OPEN PROBLEMS
1. It results from theorem 3.1 that the closure of the strain energy
in a Hencky's model does not require to consider a displacement u+lr 1
different from ulr 1 This should also be the case for the mixed vatia-
tion~l formulation, and one should be able to prove that the solution
(u,u) given by theorem 3.2 can be chosen such that u+ = ulr 1 The
author was unable to prove this point.
2. Lemma 3. 1 1s a fundamental contribution of JOHNSON [39] .
3. Regularization of elasto-plasticity can be achieved by introduction
of viscous or hardening effects. A successful viscous law has been propo-
sed by FRIAA [45] : let j be the j auge function of P and the asso-
ciate potential ~
p
j(T) = Inf {k ~ 0 T E kP}

~ (T) = l (j(T))p
p p
Instead of (3.39) consider the elastic visco-plastic material
AoP + a~ (oP) = (liP) .
p
Evolution problems for such materials are studied in DJAOUA & SUQUET
[67] , and similar constitutive equations are considered in BLANCHARD &
LE TALLEC [68] .
Plasticity with hardening is dealt with in JOHNSON [46l and NGUYEN
QUOC SON [47] .
4. Numerical approximations of elasto-plastic problems by mixed finite
eleme~ts do not seem so popular. Let us mention JOHNSON [48] , NGUYEN
DANG HUNG [31] , and DJAOUA [49] (viscoplasticity). They should be more
developped since they provide a good approximation of a
Discontinuities and Plasticity 315

4. THIN LAYERS

In this section we examin the asymptotic behavior of a thin layer


of small width 8 , between two elastic bodies, when 8 goes to 0 .
Such a problem can be considered as a preliminary to the study of bonds
between two materials, the behavior of bonds being rather different (and
often more complex) than the behavior of the elastic bodies stuck together.
In the model problem considered below the layer is assumed to obey a non-
linear elastic power law of Norton's type. Considering a viscoplastic
layer (obeying FRIAA's constitutive equations for instance) would be more
realistic but would lead to additional technical difficulties (since both
u and u interfer in constitutive equations) which are not our main
point of interest.
4.1 THE THREE BODIES PROBLEM : o >0
Main notations are displayed on figure 4. I below
XN
~ {xE~ XN > 0}

~0 0
{xE ~ > -}
~ 2
~6 ~0 u ~0
+
Bo 0
{x E ~ ; lxNI <-}
2
Bo B0 n ~

so {x E ~ +i}
~ -2
s {x E 5I 0}
~

Pip:ure 4.1

For the sake of simplicity we have assumed that the layer develops
in the vicinity of the plane xN = 0 . Extension to curved interfaces
seems possible but technical arguments have not yet been completed by the
author.
316 P.M. Suquet

In the following we consider the body fixed on its boundary, and


submitted to body forces f E L 00 (~)N
0
u 0 on div a 0 + f =0 1n
The part ~0 = ~0 of the material is elastic
0 0 + a = A-I
a = 1n
a e(u )
while the layer B0 is occupied by a nonlinear elastic material the be-
havior of which is given by the model Norton's law
a0 k 0 JE(u 0 ) lp- 2 E(u 0 ) = k 0 ~'(E(u 0 )) 1n B0,
wh~re ~ lelp/p
(e) Exponent ~ takes its values in the range
=
] I, 2] . Pp describes an elastic material, while p = I gives a rigid
=2
plastic behavior (p =I. 5 for a superplastic material, p = I. 2 for a hot-
worked metal, in the more realistic model where u stands for the dis-
placement rate).
The coefficient k 0 depends on o 1n an essential manner
k0 = K oa 0 < K < + 00

The equations of the problem are summarized below


0 0 0 k ol E(uo) Ip-2 E( U 0 ) ln
}
a =a E(u) 1n ~0 a Bo
( 4. I)
0
div a + f = 0 in ~ u0 0 on 3~

Classica! va~iational arguments ensure existence and uniqueness of a solu-


tion (a , u ) , within the following functional framework
vo {u I ul~o = ul E Hl(~o)N, ul o u2 E wl,p(Bo)N
I 2 o B
u u on s uld~ 0}
0
The variationa~ for~ulation of (4.1) is to find u 1n such that
for every v 1n V

JO a E(u 0 )E(v)dx + k 0 J0 IE(u 0 )ip- 2 E(u 0 )E(v)dx = J fvdx (4.2)


~ B ~
which 1s equivalent to the following minimization problem

Inf 0 J 0 (v) (4.3)


v E V
o
J (v)
I
= 2 J aE(v)E(v)dx +
0
-J
k
oldv)ipdx- f fvdx (4.4)
~o p B ~
J0 is a strictly convex, continuous functional on V0 and existence and
uniqueness of a minimizer is ensured by classical theorems.
4, 2 THE ASYMPTOTIC PROBLEM o = 0
0 0
We now investigate the behavio~ of the solution (a , u ) when o
goes to 0 , i.e. when the layer B shrinks to the surface S In the
Discontinuities and Plasticity 317

limit a jump of u across S is to be expected for suitable values of


exponent a , and we recall notation for such discontinuous fields
+ + ~
[u] = u
~
- u u- (x) =lim+ u(x, xN)
o-
On S t h e norma 1 vector n ~s ~--:
. po~nt~ng from ~ to ~+
(.~n t h e part~-
.
cular case under consideration it coincides with- eN)
The.ofLe.m 4.1 A~gwne. tha-t I< p ~ 2, an.d tha-t M~wnptioM (2.1) Me.
me.t. The.n. (a 0 , u ) c.on.ve.!Lgu, ~n. a ~e.Me. w~c.h will be. ~pe.&Me.d ~n. the.
te.x;t, towM~ (a 0 , u 0 ) uMque. Mlution. ~n. n)(~) xHI(~ + u ~ - )N o6

I
a
0
= a du 0 ) ~n. ~
+
u ~
0 0
div a + f = 0 ~n. ~ ' u 0 on. d~
0
on s a > p-I (4.5)
a .n 0 ~6
a .n
0
K j' ([uoj) on. s ~6 a = P- I
p
The. pote.n:tlat jp ~ de.Mn.e.d on. s M

jp (,\) ='{) (,\ <8) n)


p s
an.d
aj
j'(,\). = ~(,\) = -1,\ n Ip- 2 ( ,\ . + ,\ n ./ +)
I
p (j,\. 2 s n ~
whe.!Le. ,\ = A..n.
~ ~ n J J
~

P!Loo6. We give the proof for a = p- I (if a < p- I take K = 0)

I~g!i2g~!_I~!i~ig~ri~~ The following functional space will be useful


~n the sequel
V = {u E H 1 (~ + U ~ - )N , u = 0 on a~}
V is a Hilbert space for the norm
lulv = lui 1 N + lui 1 N
H (~+) H (~_)
Let a be a stress field in s (f) ' then the following Green's formula
holds true for every v in v

J~ u ~ a ( v) dx + J~ f v dx = - JS a. n[v]ds (4.6)
+
0 0
We turn back to the study of (a , u ) .
0
~-EEiEi_~~!i~~!~~ Multiply ( 4. I) by u and integrate over ~ to
obtain

(+)
aj 2 A.n. + A.n.
_P(,\) 1 p- < ~ l l ~) since >... appears in
Note that l;x. n 2 nj
();\.. s ~
~

(,\ <9 n) .. and in (;\. n) ..


s ~J s J~
318 P.M. Suquet

fniS aE(uiS)E(uiS)dx + K ISp-l ~ISidu.~lpdx = fnfuiS. (4.7)

Then, taking advantage of the fact that UIS vanishes on an


lfn f uiSdxl ~ lfiLoo(n)N luiSILI(n)N ~ cldu 0 )1n)(n) (4.8)
By Holder's inequality we obtain

I E(u 8 ) 1n..I (n) ~ IISI 112 1 8 1 8 1


n E(u ) n,2 (no) + 1B 1 /p'l E(u 1S1
)'ILp(Bo)
(4.9)
~ clduiS) IL2(niS) + ISp-l/piE(uiS) ln,P(Bo)

Coercivity of the elasticity tensor, together with (4.7)(4.8)(4.9) yields

( 4. I 0)

implies :

I E(uiS) In,2 (Qo) ~ c (4.11)

p-1/p I IS I c (4 .12)
o E(u ) n,P(B ~ ~

We deduce from (4.9)(4.11)(4.12) and the boundary conditions that


0
u is bounded in BD(Q) (4. 13)
0 0
f~~Y~!8~~~_! __ i~-~-~-L On the one hand u 0 contains a subsequence
which converges in BD(rl) weak x to an element u0 of this space

(4.14)

On the other hand for ~very n > 0 we deduce from (4.11) and the
boundary conditions that u N is bounded in Hl(nn)N , and therefore
that u0 belongs to Hl(nn) for every n > 0 with

lu 0 IHI(nn)N ~ C (4.15)

Letting n tend to 0 in (4.15) shows that u0 belon~s to V . More-


over, denoting by XniS the characteristic function of n we obtain
IS o
XniS E(u ) ~ E(u ) in n, ~n u n ) weak (4.16)
+ -
The compact embedding of BD(rl) into L1(n)N shows that

u tS ~
__ . , o
u
1n L I ( Q)N strong (4.17)
IS
We deduce from (4.11) that a is a bounded sequence in n. 2 (n 0 ) , and
Discontinuities and Plasticity 319

from (4.12) that

J 0 Ia 0 jP I
dx =
I
fo Kp opldu )jPdx
0
~ c 0
B B
Since p 1 = _: 1 ~ 2 we obtain that a 0 is bounded in n. 2 (n), with a
weakly conveFg1ng subsequence (without loss of generality we can assume
that it converges 0 for the same subsequence of o than the subsequence
extracted from u ) , and we denote by a0 its limit. It is straight-
forward to check that a0 and u0 are related by

a0 E TI.. 2
+
(n) , div a0

n u n
-
- f = 0 1n n
}
and the only remaining point concerns the behavior o~ a 0 and u0 on S.
u0 may admit discontinuities on S , while a0 .n is uniquely defined
in H-1/2(s)N (by virtue of the equilibrium equations). For this purpose
an intermediate result 2 will be usefull : note that for a "smooth" stress
field T (in w 1 00 (n)~ for instance) we have

lim ~ 0 TE(u 0 )dx = J5 T.n[u'1dx (4.18)


0 -+ 0
Indeed

lim J T E(u 0)dx = lim CJn T E(u 0)dx - 10 Tdu 0 )dx]


o -+ o B0 o -+ o n
lim [-JndivT}dx- fn un x 0
0 Tdu )dx]
o-+O + _n

According to (4.16)(4.17) and (4.6) we obtain

= - fn di v T u0 dx - ~ U n T
0
E(u )dx J,
s
T .n[u~ ds
+
which proves (4.18) .
f~~!~!_1~~ It remains to prove that a0 .n and [u~ are related by
the contact law (4.5) . For this purpos~ according to the maximal mo-
notonicity of the operator j 1 (z) = lziP z , it is sufficient to prove
p

~(a 0 .n- Kj~([v])([u]- [v])ds ~ 0 (4.19)

for every v in V. (4.19) is derived from the inequality

l 0 (a 0 - k
0
I{) 1 ( d v 0) ) , d u 0) - 0
e:( v ) ) dx ~ 0 (4.20)
B p
0
applied to a suitable approximation v of v in V which we further
describe.
Let v be an element of V .N By a density argument we can clearly
assume that v is regular (C 00 (n+) x C00 (n-)N) ~n n and n and that
its support is compact in n . +
320 P.M. Suquet
0
v
v

o/2

Figure 4.2 Approximation v0

0
We approximate v by a more regular vector field v continuous
across S :

lxl > i2
0
v (x) =
V~(x)) + X: [v] 0 (x) if

x -- v <~x,-+ i)
0
wh ere v+o(~) 2 and [vi(x) = v 6 (x)
+ v~ (x) v exhibits
the following properties : N
i) v 0 tends to v in L2 (n) strong.
ii) Trace operator on a surface xN = o/2 ~s a continuous function of
the surface. Therefore

lim [v]0 (x) = [v] (x) in H112 (S)N(hence L2 (s)N) strong (4.21)
o+O
and consequently, for every r E [1, 2]

/:-~ o- 1 1[v]0 s n- [v]s nl~r(Bo) = 0 (4.22)

iii) Deformation associated with v0 reads as

i and j IN
o I o o ~;:( .. (v o)- .. (v o)) (4.23)
, .(v) =-2 ( .. (v )+ .. (v ))+-
~J ~J + ~J
2
- 0 u ~i + ~J 0- 0
i .1. N J. 0 I 0 0 I av+N <lv_N XN av+N av_N
r N iN(v) =u(v+i -v-i)+t;(~+----ax.-)+21(----ax.------ax.-)
~ ~ ~ ~

i = N j N NN(v 0 ) = t (v:N(x) - v~N(x))


Lemma 4.1 Let v be an element o6 Coo(n+ u n_)N , and v 0 defi~ned
a-6 above. Then

i) lim I k 0~'((v 0 ))(v 0 )dx = K I j'([v])[v]ds


o +0 B0 p S p

ii) lim I k 0~'((v 0 ))(u 0 )dx = K I j'([vJ)[u0]ds


o+ o B0 P s P
Discontinuities and Plasticity 321

iii) lim J 0 o 0 E(v 0 )dx = J o 0 .n[v]ds


0 +0 B S
iv) lim sup 0 o J
0 E(u 0 )dx ~ o 0 .n[u0 ]ds J
o+O B S
It is obvious from the lemma that (4. 19) is a mere consequence of (4.20).
Therefore the proof of lemma 4.1 will complete the proof of theorem 4. I.
Proof of lemma 4. I For every r E [1, 2]
------------~872 __ _
lim J dxN J I E(v 0 ) -~ [v] 0s nl rds ~
o+O -o/2 s
(where cl and c2 depend only on v) Therefore

lim ldv 0 ) -~[v] 0 0s nl~r(Bo) 0 (4.24)


0 +0
Combining (4.22) and (4.24) yields
p-1
lim 0 ldv 0 ) -@ 0 niP 0 (4.25)
0 +0 0 s n..P (Bo)

Consequently

lim Kop-IJ ~'(E(v 0 ))E(v 0 )dx


o+ o B0 P

KJI(v]~ n pds=Kjj'([v])[v]ds(4.26)
s s p

which proves point i) .


Moreover ~ JLP (B 0 ) JLP (B 0 )
1
is Holder continuous from into
(cf GLOWINSKI-MAROCCO [50])

Therefore

~cop-lld})-~[v] nlp-l 8 IE(u 8 )1 8


s 11 p ( B ) 11 P (B )

:s C(op-ll E(v 0 ) - _!_ [v] niP )p-l/p(op-l/p IE(}) I )


o s 11p (Bo) 11 p (Bo)
We deduce from estimate (4.12) and from (4.25) that

lim f 0k ~'(E(v 0 ))E(u 0 )dx = lim Kop-l f ~'(hl 0


n)E(u )dx
o+ 0 B p o+ 0 B0 P 0 s
322 P.M. Suquet

lim K f 0TE(})dx where


0 +0 B
according to (4.18) we obtain

= K Js ~'([v]@ n).n[u 0 ]ds


p s
which proves point ii)
Then we note that
0
lim J a 0 E{v 0 )dx = lim Ji:l a 0 E(v 0)dx- J a X 0 E(v)dx
0 +0 B0 o+ o i:l r2
0
(note that X 0 E(V ) = X 0 E(v))
iJ iJ
= lim (- J f v 0dx - J a0x 0 E(v)dx)
o+O i:l i:l i:l

= -J f v dx - J
iJ iJ UiJ
+
which proves iii)
By lower sem1 continuity we obtain

J a 0 E(u 0 )dx=J aE(u 0 )E(u0 )dx~liminfj a(x ..,du 0 ))(x 8 E(}))dx


iJ UiJ iJ UiJ o+ 0 iJ
+
UiJ r2 iJ
+ +
~ lim inf J a 0 E(u 0 )dx
a + o n 0 0
Using the equilibrium equations for both a and a yields

- J a 0 n[u 0 ]ds ~ lim inf [- J0 a 0 E(u 0 )dx]


s 0 + 0 B

(Note that JiJ f u 0 dx = JiJ a 0 E( u 0 ) dx


Thus E +0
J0 0 0
J a0 .n[u ]ds ~lim sup
0 a E(u )dx
s 0 + 0 B
which ends the proof of lemma 4.1 and of theorem 4.1 .
4.3 COMMENTS, OPEN PROBLEMS
1. As noted in the introduction a more physically relevant problem would
be that of a real viscoplastic layer. Equations of the problem read as

}
0
a0 = a E(u 0) in i:l 0 a 1n
0
div o 0 + f = 0 in iJ u 0 on ai:l
2. Again a more physically relevant problem should include incompressi-
bility of the layer
div u 0 = 0 in B0 a 8 =- p:U:d + k 0 ldu 0 )ip-Z E(})
It follows from this constraint that [u0].n = 0 on S , and the resul-
Discontinuities and Plasticity 323

ting constitutive law of the interface should still be given by (4.5)


with due account of the above equality. However the precise statement and
the proof of this result have not been completed by the author.
3. Theorem 4. 1 lets open the case a < p-1 . One could think that
the limit behavior is simply obtained by letting K tend to + oo in
(4.5) which yields
[u'] = 0

This result would be incomplete as proved by CAILLERIE ~1] for p = 2


For negative a the interface behaves like a membrane or like a plate.
The total energy of the system contains a contribution of S with first
or second derivatives of u 0 on S . The extension of CAILLERIE's re-
sults and the determination of critical exponents a in the present
context (1 < p ~ 2) is an open problem to the auth<>r's knowledge. When
a plate behavior of the interface is expected,BLANCHARD & PAULMIEKs ~2]
modelling of viscoplastic plates should be useful.
4. The case of elasto-plastic layers is treated in DEST~iNDER & NEVEU
~3] . It is proved that, assuming that the safety assumption is met
un6o~mty with ~eop~et to o , the interface has a limited strength cha-
racterized by a domain C which delimits the set of admissible stress
vectors
cr.n E C ([u],cr.n-T) ~ 0 V T E C
C is nothing else than the set of stress vectors defined in section 2 .
This result can be proved by a very simple use of the epi-convergence
theory (see next section) and of Mosca-convergence (see ATTOUCH [3]) .
An interesting Eroblem arises when discussing the limit load problem. Let
us denote by A the limit load for a structure n made from two ela~tic
parts bonded by a band of width o > 0 When o goes to 0 does A
tends to A, obtained by considering an interface with the limited
strength defined by C ? The answer is yeo if the structure is not loaded
on its boundary (but fixed), and no if the structure is loaded on its
boundary. The next section will provide an example of a similar phenomenon.
5. Similar problems to those dealt with in the present section, are
treated by BREZIS & al [54] , ACERBI & al [ss] .
324 P.M. Suquet

5, EPI-CONVERGENCE, DISCONTINUITIES AND PLASTICITY

5.1 ELEMENTS ON EPI-CONVERGENCE


'
Th e t h eory of ep~-conve~gence (+) . . .
appl1es typ1cally to the follow1ng
problem. Let (X,T) be a topological vector space (for simplicity T
1s assumed to be metrizable), and Jn a sequence of functionals mapping
X into JR. { + oo} Consider the variational problem

Inf
A ( 5. I)
n
x E X
We define the T epi-limits inf and sup of as follows

x E X lnf (5. 2)
T
x--+x n -+ + co
n

V x E X J(x) Inf lim sup (5. 3)


T
X--+ X n _.. + oo
n
Note that J ~ J . Jn is said to be T epi-convergent to J if
J = J = J A first important result significantly reduces the number
of possible T epi-limits.
Theo~em 5. 1 Wdh .the a.bove no.ta:UonJ.J J a.nd J Me T .towe~ .oem.l
continuou.o nunctiona..f.o.
We have already encountered epi-convergence (but we did not know it)
as proved by the important corollary.
Co~o.t.fMy 5.1 Ta.ke Jn = J no~ a..f.t n . Then Jn ep~ conve~ge.o .to
.the T U0.6~e On J
Therefore epi-convergence con.t~nJ.J the difficult problem of computing
the lower semi continuous hull of a given function.
The following abstract theorem specifies the behavior of the infima
A 1n (5.1).
n

( +) . .
a general and complete exposure on ep1-convergence, together w1th
numerous applications and references can be found in ATTOUCH [3]
Discontinuities and Plasticity 325

Theaftem 5.2 Let Jn be a uquenc.e a6 Sunc.tiana.6 T ep-i.c.anveftg-i.ng


;to J , and let x be ~uc.h that
n

Then 6oft evefty T c.anve.ftg-i.ng ~ ubuq uenc.e (UmUxl we have

J(x) = Min J(x) = lim J~(x )


xEX k-++oo nk

Provided that we can extract from the approximate minimizers a T con-


verging subsequence, the above theorem ensures convergence of the infima

A = Min
J(x) = lim A (5.4)
x EX x-++oo n
A very simple corollary of theorem 5.2 will be useful 1n the study of
limit analysis problems :
CaftailMy 5. 2 In add-i.tian to the. M~umptia~ a6 the.O!tem 5. 2 let L
be a c.antmua~ UneM 601tm an (X, T) and c.a~-i.deft the va!t-i.atianal
pftable.m
n
A = Inf{J (x), L(x) =I} (5.5)
n
n
and let x be ~uc.h that L(Xu) = I. J (x ) ~ A + 1/n Then 6oft
evefty T ganveftg-i.ng ~ub~equenc.e X ( UM-u: X) n We have
nk
J(x) = Inf{J(x),L(x) = I} lim A~= lim J~(x ) (5.6)
k -+ + 00 k-+ + 00 ~

In conclusion we recall a useful criterion in two steps in order to


prove an epi-convergence result
J ~ the T ep-i.-Umi;t a6 Jn -<-6 the twa 6ailaw-i.ng c.ond-i.tio~ Me.
met
i) (X EX)( X EX) lim X =X -i_n (X,T),J(x) ~liminf Jn(x) (5.7)
n
n-++oon n++oo n

ii) (xEX)(}x EX) limx =xln (X,T) andlimsupJn(x)~J(x) (5.8)


n n-+~n n-++oo n
When T is not metrizable, i) and ii) define the sequential epi-
limit of Jn Such is the case of weak% topologies ori Ml(n) or BD(Q).
For simplicity we shall omit the word nsequential", which will be implicit
as soon as a non metrizable topology is used.
5.2 A MODEL PROBLEM OF SLIPPING BLOCKS
The onset and the development of macroscopic plastic strains is phy-
sically related to dislocations, i.e. to defects of the atomic structure
at the microscopic scale. Motion of dislocations .is similar to a friction
phenomenon and the present section is devoted to a modelling of the micro-
326 P.M. Suquet

scopic situation. We consider a periodic assembly of parallelepipedic


blocks slipping at their interfaces (one could also think of a wall made
from bricks), and our goal is to derive the limit (or macroscopic) cons-
titutive law of this assembly when the parallepipeds' size goes to 0 .
This procedure has been popularized under the term homog~~z~on.
For simplicity we reduce our attention to 2-dim problems, and we
assume that the domain ~ is the assembly of identical cells stuck toge-
ther on their boundaries, the strength of which is limited

la .. n.(x)l ~ k i = I, 2 for all X E r (5.9)


lJ J
where r denotes the union of all cells boundaries while ~* denotes the
remaining part of the body ~* = ~ r -

,...y
' ..
lJ


center M.
lJ

cell Y .
lJ

Figure 5. I

Each open cell Y . is centered on the node M. of a periodic net-


work extending on lR2 l~ The vertical right part, lJ the horizontal upper
part of aY are denoted respectively r~. and r~. We note that
lJ lJ lJ
u (Y .. n ~) = ~* u cr~. u r~.) n ~ = r (5. I 0)
-oo<i,j<+oo lJ ' -oo < i' j < +oo lJ lJ
We shall only deal here with the limit analysis problem. Blocks can
support an arbitrary load but the interfaces have a limited strength
(both to shear and to normal stresses) described by (5.9) . For simpli-
city the body is loaded by body forces only (f = a~) The statical
definition of the limit load is 0

A.s = Sup{A. E lR I ]a E :n..Z(~) , diva + Af 0 a.e. 1n

lo.nl ~ k a.e. on r} (5.11)


Discontinuities and Plasticity 327

A kinematical definition can also be proposed. Let u be a displa-


cement rate field element of
v {u E H 1 cn*) 2 , u = 0 on an} C+)
and A < As a associated with A by (5.11) Then
J a E(u)dx l: J a E(u)dx
n* i,j Y . n n
q

A ~-.l:. f y .. n n f u dxjl + JaY . . n no. n u ds (5. 12)


_lJ lJ lJ
Adjacent cells have opposite normal vectors on their common boundaries,
and a rearrangment of the second integral in (5.12) yields (according
to (5. I 0))

A fn f udx = fr a.n [u] ds + Jn* a E(u)dx (5. 13)

Now we take the supremum on a satisfying (5.11) . Since we have no


information on a I the supremum is +co except if
n*
E(u) = 0 in every Y. . , i.e. u = p. . 1n Y .
lJ lJ 1]
where P.. 1s a rigid displacement on Y. . . For those displacement
rate fieids we use the limited strength c6~dition in the form

(5.14)

and obtain

Inf {IT(u) , L(u) I , u p.. on Y . } (5. 15)


u E V lJ lJ
where
IT(u) fr ns([u] )ds , L(u) = fn fu dx
Equality between As and Ak (left as an exercise to the reader) is a
consequence of the abstract result recalled in section 2.3 .
We turn now to our main point of interest. The limit behavior of the
assembly of cells with a size 1/n which goes to 0 . Cells number is
of the order of n 2 , cells boundaries r and limit loads Ak are now
indexed by n . Our problem of variationaY convergeyce ~an be ~et in the
general form (5.5) , (X,T) being identified to L (n) endowed with
its strong topology (BD(n) weak;: would lead to identical developments):

Jn(u)=l Jn kl[t.illlds
+cor oth erw1se
.
-;---.,..------~

(+) Note that if u belongs to H1 (n*) 2 it can admit discontinuities


along the cell boundaries.
328 P.M. Suquet

TheoJtem 5 3 Let P be de6~ned ao


p (5.17)

a.nd 11 fihe J.Ju.pp0!!.-t_6u.nc.tion o6. P: 11(e~ = k(!e 11 ! + 2le 12 ! + le 22 !),


then J (5. 16) ep~-~onve~tge-6 ~n L 1 (~) J.Jtftong to
J~ 11( E( u)) +
J{l~ b
1T ( -u) ds ;L
"'D u E BD(~) (+)

J(u) (5.18)

.
PJtoo 01 (++) Th e boundary term 1n is obviously a relaxation term ( 5. 18)
due to the boundary condition u = 0 in the definition of Jn This
term, which is not our main point of interest, complicates the analysis,
and we prove a weaker (but essential) form of theorem 5.3 which does
not account for the boundary conditions.
Let Jn be de6~ned ao
fnkl[u]lds ~tl u=p .. on Y .. ,uEH 1 (rJ*) 2
r lJ lJ

otheJWJM e + oo

Then Jn ep~-conve~tge-6 ~n L 1 (~) 2 J.Jtftong to


J(u) = f~ ll(E(u)) ~tl u E BD(~) + oo

The result is attained with the help of the above described criterion
(5.7)(5.8) for epi-convergence.
First step Let u be a sequence 1n L 1 (~) 2 converging to u and such
th~t--ii~- inf ]n(un) < + oo u 1s piecewise smooth and belongs to
BD(rl) (cf(2.25)) wlfth n

ldun)IMI(~)4 ~
s
fr ![uJ!ds ~ k Jn(u)
n n
< C (5.19)

Moreover (u) is bounded in L 1 (rl) 2 (since it is convergent) and with


(5. 19) (u )n is a bounded sequence in BD(rl) Therefore u belongs to
BD(rl) andn J(u) is finite.
Consider a 1n C0 (rl) 4 such that s
c
( X E rl) a(x) E P (5.20)

Then for u 1n the domain of Jn (i.e. piecewise C00 ) we have


according to (2~25) and understanding integrals in the sense of the

(+) J ll(E(u)) ig to be understood in the sense of functions of measure


( 2.5) , n is defined in (2.29)
( ++) . . d
Th1s proof 1s adapte from BOUCHITTE [56] who treated the case of
a scalar unknown u .
Discontinuities and Plasticity 329

duality C0 (rl) , M1 (n) :


c
f.," aE ( u ) ~ J k lll:u 11lds
n rn n"
E(u ) converges to E(u) in weak x and therefore
n

Sup 0 4 .b a E(u) ~ lim inf Jn(un) (5. 21)


a E Cc (fl) s n + + oo
a(x) E P
But the first member in (5.21) is the true definition of fn n(E(u))
according to section 2.5 . We have achieved the first step (5.7)
( u E L l (rl) 2 ) lim inf J(u) ~
n++oo
~~~~~-~!~E In order to prove (5.8) in our specific example~ we can
restrict attention to u in C (n) 2 which is dense in Ll(n)L 00

For u in C (n) 2 we define its vorticity by


00

a~ auJL
1/2 ( - - - )
axJI, a~
(~ote that w11 = w22 = 0, Define a rigid
d1splacement p on Y
1J
ij ij ij
pl (x) =u 1 (Mij) -w 21 (Mij)(x 2 -x 2 ), p 2 (x)
ij -
where ~ - ~(M .. ) . 1 2
Finally we pfJpose to approximate u in L (n) strong by
u (x) = pij(x) xE Y if
n 1J
The following (easy) inequality ensures the convergence of u towards
u in 1l (n) 2 strong n
1
1 u - u 1 00 2 ~ - I grad u I 00 4
A n L (Q) n L (Q)
Jn(u) is the sum of elementary integrals on r~. and r~. (see nota-
tionR on figure 5. 1) 1J 1J

kf <IPi+I,j- Pi,jl + (5. 22)


rx 1 1
ij
and a similar integral over r~ .. In order to compute the above integral
we use a Taylor development of 1Ju and w12 in the neighborhood of M..
1 au 1 I 1J
ui(M. I .) - ui(M .. ) =--a- (M .. ) + 0(- 2 )
1+ ,J 1J n xi 1J n
Iau2 I
u 2 (M. I . ) - u 2 (M .. ) = --a-(M .. ) + 0(- )
1+ ,J 1J n xi 1J n2
I I
= ~(E2I(u)(Mij) + w2I(Mij)) + 0(-z)

*
n

lw2l(Mi+l,j)- w2I(Mij) I ~ ID2uiLoo(n)8


330 P.M. Suquet

Therefore, recalling that on r:'. we have x - xij = 2n '


; +I J. I . I . ~J . . I . I .I 1
,J = x~J x~+ ,J = x~J + -
xl-xl
L
= - 2n ' and x~+
'
2 2 ' I I n
we obtain

IP~l+l,j(x)- p~lj(x)l = _!_l~ul (M .. )I + 0(_!_2)


n ax 1 ~J n

IP2i +I' j (x)-p2ij (x) I I u2(M. I .)-u2(M .. )-(w2l(M. I .)+w2l(H .. ))-2 I I


=
~ + 'J ~J ~ + 'J ~J n
I I I
= lu 2 (M. 1 .) -u 2 (H .. ) -- w, 1 CM .. )I +0(2) IE 21 Cu)(M .. )J +0(-z)
~+ , J ~J n z ~J n n ~J n
We also note that
JY .. IEkh(u(x)) ldx
~J
Therefore the integral over r~. amounts to
~J

k Jy .. ( I EII ( u) I + I E12 ( u)l )dx + 0(_!_)


3
(5.23)
~J n
A similar argument shows that the integral over r~. amounts to
~J

kfy .. CIE 12 Cu)l + IE 22 Cu)l)dx + O(~) (5.24)


~J n
Combining (5.23) and (5.24) , achieving the sum over indices ij
yields :
Jn(un) = Jrl k(IE 11 Cu)l + 2IE 12 Cu)l + IE 22 (u)l)dx + 0(*))
and therefore
Jn(u ) = J(u)
lim
n oo~+
n
Proof of theorem 5.3 is complete.
COfloilMlj 5.3 .
hm A
k
= A
k
whe/t.e
n~+oo n
Inf {Jn(u) , L(u) = I} (5.25)
u

Inf {J(u) , L(u) = I} (5.26)


u

To combine theorem 5.3 with corollary 5.2 it ~s sufficient to


pyove that approximate minimizers of (5.25) form a compact set in
L (rl)2 l or are bounded in BD(rl) (use the compact embedding of BD(rl)
into L (rl)2 cf 2.3). But this last estimate ~sa result of (5. 19)
and (2.19) with the following sem~-norm
r ( u) = farl Iu Ids
which vanishes on Jn's domain. The proof of corollary 5. 3 is complete.
(5.26) has an important mechanical interpretation
An MJ.J embR..y o 6 Jr)_gj_d bR..oc.IM, the )_n.tett6ac.M o6 whj_c_h have a umded
J.J.t!Leng.th, behaveJ.J mac.ttoJ.Jc.opj_c.aLty (i.e. when the blocks' size becomes
Discontinuities and Plasticity 331

so small that they cannot be distingued one from another) ~ a ~g~d


p{~tic_ body wah. a fl.t!tmg.th dom~n p dei~ned ~n (5. 17)
5.3 ASSEMBLY OF MATERIALS ~liTH LIMITED STRENGTH I, NO LOADS
ON THE BOUNDARY
This section deals with the homogenized properties of a periodic
assembly of materials with a limited strength.

,
L_f""i 7 !')~
,
, I', -
/J # I Enlargment
fl
' I

-~
I I I I
I! I I I I I
L, ..,..
, , I 'I
I I I
I I ""
~~
I! , I
,,7
~

, If
I I I
I I I I I / I
-Y

, , , ~17
I I I
, I''T
I I / width 1/n
~
,..._
I -I\

{:
I I
I I if y E y]
'-~..t...
- ,.....,.
P(y)
if y E y2
Figure 5.2

Main notations of the problem are displayed on figure 5.2 . Material pro-
perties ar~ defined on the unit cell Y = 0, I [3 and extended by periodi- J
city to JR. We assume for simplicity that P(y) is simply defined by
a ball of radius k(y) (assumption (2.4)) and that k(y) takes two (non
vanishing) values on two borelian sets Y1 and Y2

k(y) k(y) = k2 if y E Y2 (5.27)


thus ~(y,e) = k(y) lei k(y) is extended by periodicity to m. 3 and we
set

kn is periodic with a period 1/n ~n each direction.


The sequence of limit loads ~s defined by

,\
n
(5.28)

~ Jrl ~n(x,E(u))dx if ' u 0 on r


0
(5.29)

~ + oo otherwise

L(u) 1rl f udx


332 P.M. Suquet

(in this section we consider that r = an ' ri = 0) .


The homogenized strength ~ro~erties ~s given by theorem 5.4 due in its
~enerality to BOUCHITTE L26J , extending a partial result of the author
[5~ . We need further notations to formulate it clearly : < > denotes
the average symbol on the unit cell
<f> = -
I
!y f(y)dy .
We define a~~~nvex l.s.c. function, positively homogeneous of degree one,
on JR9
s
nhom(E) = Inf <n(y,E + (w))> (5.30)
w E H1 (Y) 3
I per
where H (Y) is the space of functions in HI(Y) which take equal
va 1ues onoer .
oppos1te s1. des o f Y . Basing on the abstract re~ult recalled
in section 2.2 , a duality argument allows to identify n om as the
conjugate function of I p h om
phom = dom((nhom)*) (5.3I)

{EEJR;i=JcrEn. 2 (Y), divcr= 0 , cr(y)EP(y) a.e. yE Y

cr.n opposite on opposite sides of Y , <cr> = E}


Then we have
n
The.o!te.m 5. 4 [2y] 3 M-6wne. thOvt a~J..U.mp.Uon ( 2. I) -i..-6 me..t. The.n TI e.p-i..-
c.onve.Jtge.-6 A..n L (n) -6-tJLong .to IT om

In TI
hom (E(u)) + Jr (n hom ) b (-u)ds ..i.6 uEBD(n)
IT hom(u) )
(5. 32)
+"" otheJWJ-i..-6e. o

Mo!te.ove.Jt
hom
lim A = A = Inf{IT (u) , L(u) = I} (5.33)
n
n-++oo
Comme.n.t.Theorem 5.4 states that when heterogeneities'size becomes
small (I/n-+ 0) , the limit load computed on the nonhomogeneous material
is well approximated by the limit load gf a fictitious homogenized mate-
rial the strength domain of which is P om (5.29) .
5.4 ASSEMBLY OF MATERIALS WITH LIMITED STRENGTH. LOADS ON THE
BOUNDARY
When the assumption r =an ' ri 0 is abandonned, new interesting
problems arise. Indeed the ~inear form L is no more continuous on BD(n)
(as we already noticed in section 2.5) , because of its boundary term

L(u) = fn fu du + Jr g u ds (5. 34)


I
Discontinuities and Plasticity 333

Therefore Corollary 5.2 no more applies. More specifically, canter-


examples to an extrapolation of (5.33) have been given in the recent
litterature (DE BUHAN ['58] , TURGEMAN-PASTOR [59] consider the
sequence A of limit loads for the nonhomogeneous materials, defined
by (5.28) ~ut where L is now defined by (5.34) Then (A ) doe6
n
not eonve~ge to the natural guess
An = Inf{!Ihom(u) , L(u) = I} !Ihom given by (5.32)) (5.35)

We describe an adaptation of DE BUHAN' s example (in which an was not c 1)


which illustrates this surprising negative result.
~~~~E1~ Consider in E 2 a stratified medium made from two homogeneous
constituents obeying (5.27)

Figure 5.3

Phom (given by (5.31)) can be determined. It is contained in the ball of


center 0 and radius k 2 , and l.ltJUc.tiy contains the ball of center 0
and radius k 1 (we assume k 2 > k 1) Therefore
(j E0 E phom)(jn E E 3 ), lnl =I , IE 0 .nl > k 1
(By continuity n can be chosen to be not no~mal to the stratification
plane). n is chosen with a smooth boundary containing a flat part r 2
normal to n (cf Figure 5.3) , and we set
0
g(x) = E .n(x) V x E f I f = 0
In the nonhomogeneous material we have for an admissible A < As
IAg(x) I ~ k(x) a.e. xE r 1
i.e. on

Therefore A ~ 8 < I But it can be checked that satisfies


n
334 P.M. Suquet

0
div E0 =0 , E .n(x) = g(x) :v x E r1
and therefore that An ~ I In this example lim sup A. and can-
n
not be equal.
The correct limit to the sequence (A. ) is provided by corollary 5.4
below. In order to understand it we recallnthat we introduced in section
2.5 an appropriate relaxed form of the limit analysis problem, for loaded
boundaries (cf(2.30)~
- + - +
A. Inf {!In(u,u ) , L(u,u )
n +
u,u

1T
nb ( -u) ds +Jr 1T
nb ( u + - u) ds
I
n n +
II ( u) + J ( u - u).
Although the epi-limit of a sum is not the sum of epi limits,it can be
useful to compute separately the epilimifis of nn (this isdone in
theorem 5.4), and Jn. We note that Tin (x,.) is the support function
of set cn(x) previously defined cf (2.40) , or in other terms that
Jn (I in the duality M1 (r ) 3 - C0 (r ) 3
en )x I c I
,n {~ E C0 (r 1) 3 , ~(x) E Cn(x) a.e. x E r 1}
By duality we have now to determine the limit of ~n It is given by
the following result due to G. BOUCHITTE [4J _._ LeA: U6 de.Mne.
nn1 = {x En Pn (x) = P1} An = r 1 n nn1
We. ~~urne. ~hat (2.3) is met and that(+)
= 0 , int(An) ~A, int(r 1 \A)~ B
H2 (aAn)
The.n I~n e.p~-eanv~g~ ~n C0 (r 1) 3 ~tnang ~a I~
0 3
~ = {~ E c (r) , ~ (x) E C (x) xEr 1}
G:(x) =PI n(x) ~6 x E A \B C(x) = P2 n(x) X E B \A
t(x) = P1 n(x) n P 2 n(x) ~6 x E An B
Let Tir (x,.) denote the support f~nction of t(x) which can be extended
in a fudction of measure on Ml(r 1) , and set
J(~) = fr Tir (x,~)
I I
Then a duality result (ATTOUCH [3]) asserts that Jn epi-converges
(sequentially) in MI(r )3 weak x to J . We are now in a position to
state a precise result ~BOUCHITTE & SUQUET [20J) .

(+) H2 is the 2-dim Hausdorff measure


Discontinuities and Plasticity 335

The011.em 5.5
weak x ;to
irhom<u,u+) = rrhom<u) +fr 1Thomb <-u)ds + Ir 1Tr <u+ -uds) (5.36)
0 1 1 3
Co~olt~y 5.4 A~~um~ ;tha.;t g belong~ ;to C0 (r 1) Then A conve~-
gu ;to Ah om whe~e : n
+ 1 3
i) Ahom = I n f + {Tlhom(u,u+) ' L-( u,u +) = 1 ' u,u EBD(rl) xM (r 1) }
u,u
ii) A = Inf (A,., , A ) (5. 37)
hom "
,&., given by
inf{a E lR I ag(x) E chom(x) () C(x) VxEr 1} (5.38)

whe~e chom(x) = phom n(x) .


Comment. Corollary 5.4 contains two statements
i) is a direct consequence of theorem 5.5 , since L ,U., con;t,[nuoU6 on
BD(n) x M1(r 1)3 weak x (therefore corollary 5.2 applies).
ii) shows tnat the homogenized limit analysis problem is twofold. A first
limit analysis problem set on an homogeneous n with strength domain
phom , is to be solved and yields A~ . A second limit analysis problem,
in which only the strength of r 1 is limited, is to be solved for A
The real limit load is the infimum of these two values. It should be noted
that A cannot be derived from the homogenized behavior in n (open
set) only : the homogenized behavior of r 1 is different. It should also
be noted that "relaxation and epi-convergence do not always commute", in
the following sense :
n
IIR(u) = IT n (u) + An I 1- L(u) I
which epi-converges to
i 1 - L(u) I
II~ 0 m(u) = rrhom(u) + A
However !I n .
ep~-converges to
rrhom , an d the relaxation term for
associated with the constraint L(u) = 1 is
rrhom(u) + Ahoml1- L(u) I
Ahom Sup {a E lR I ag(x) E chom(x)

(strict inequality can occur) .


5.5 BIBLIOGRAPHICAL COMMENTS, OPEN PROBLEMS
1. Epi-convergence and f-convergence have been extensively studied by
a very active Italian school (see BUTTAZO & al [60j , DE GIORGI & al
[61] , and additional references in ATTOUCH r_3J) .
2. The approach developped here is U~en;t,[a.lty k,lnema.;t,[ca.{ (epi-
convergence of functionals defined on displacement rates). By duality
336 P.M. Suquet

arguments we can obtain informations on homogenized strength capac1t1es.


However a direct study of statical problems (formulated in terms of
stresses only) would be interesting. For duality arguments we refer to
AZE [62] , and for convergence of variational inequalities with obstacles
to DAL MASO & al [63] ATTOUCH-PICARD [64] .
3. Theorem 5. 4 is due to G. BOUCHITTE [65] who extended a previous
(partial) result of the author [57] . This work ~ncludes results on
the elasto-plastic problem. DEMENGEL & TANG QUI [6~ extended BOUCHITTE's
work (by a different method) to the case of Von Mises criterion, for
which a coRstraint of incompressibility is to be added in definitions of
II n and II om . However the treatment of boundary loads in [66] does
not let appear clearly a different homogenized behavior on r . The
present exposure of "loaded boundaries" follows BOUCHITTE & SOQUET [20]
in which, to the author's knowledge, the necessity of two limit analysis
problems (inside n and on r 1) was first established.
4. We have explicitely made regularity assumption : an is c 1 g is
C0 However interesting problems arise when these assumptions are drop-
ped, as illustrated by TURGEMAN & PASTOR's example [?~ in which an
exhibits corners.
Discontinuities and Plasticity 337

REFERENCES

I. TEMAM R., STRANG, G. "Duality and relaxation in Plasticity",


JoWLVLal de Mec.ai'Uque., 19, 1980, p. 493-527.
2. TEMAM R. : Pttobemeh mathema.:t{_que-6 m Pl.Mtiede, Gauthier-Villars,
Paris, 1983.
3. ATTOUCH H. Va/Ua;t{_oVLaX C.OVLVe.ttgeVLc.e. 0ott oUVLc.tiOVL-6 aVLd ope_ttatoM,
Pitman, London, 1984.
4. BOUCHITTE G. "Calcul des variations en cadre non reflexif ... ",
Thesis. University of Perpignan. France. 1987.
5. SALENCON J. Calc.ul a l.a ttuptWLe. e.t aVLaXyhe ~mite., Presses ENPC,
Paris, 1983.
6. NAYROLES B. "Essai de theorie fonctionnelle des structures rigi-
des plastiques parfaites", J. de. Mec.ai'Uque_, 9, 1970, 491-506.
7. ROCKAFELLAR R.T. CoVLve.x AVLaXy-6~. Princeton Univ. Press. Prince-
ton. 1970.
8. EKELAND I., TEMAM R. CoVLve.x AVLaXy-6~ aVLd Va/Ua;t{_onai Pttobl.em-6,
North Holland, Amsterdam, 1976.
9. PANAGIOTOPOULOS P.D. IVLe.quat{ty pttobl.em-6 ~VL Mec.hai'Uc.-6 and App~-
c_a;t{_oVLfJ. Birkhauser. Boston. 1985.
10. STRANG G. "A family of model problems in Plasticity", Pttoc.. Symp.
Camp. Meth. ~VL Appl. Sc.. Ed. R. Glowinski and J.L. Lions, Lecture Notes
in Math. no 704, 1978, p. 292-308.
II. SUQUET P. "Sur un nouveau cadre fonctionnel pour les equations de
la Plasticite", C.R. Ac.ad. Su. P~, A, 286, 1978, p. 1129-1132.
12. SUQUET P. "Un espace fonctionnel pour les equations de la Plasti-
cite", Ann. Fa.c. Su. TouloU-!le, I, 1979, p. 77-87.
13. BOURBAKI N. IVLtegtta;t{_oVL. Hermann. Paris. 1954.
14. MATTHIES H., STRANG G. : "The saddle point of a differential program"
in Enettgy Method-6 ~ F~~e Element AVLaly-6~, Ed. Glowinski, Robin,
Zienckiewicz. John Wiley. New-York. 1979.
15. GOFFMAN C., SERRIN J. "Sublinear functions of measures and va-
riational integrals", Vulze. Math.]., 31, 1964, p. 159-178.
16. DEMENGEL F., TEMAM R. "Convex functions of a measure and appli-
cations", In~aa Ui'Uv. Math. ]., 33, 1984, p. 673-709.
17. GIAQUINTA M., MODICA G., SOUCEK J. "Functionals with linear
growth in the calculus of variations", Com. Math. UI'Uv. Catto~VLa, 20,
1979, p. 143-171.
18. VALADIER M. "Fonctions et operateurs sur les mesures", C.R. Acad.
Sc.. P~, I, 304, 1986, p. 135-138.
338 P.M. Suquet

I9. HADHRI T. "Fonction convexe d'une mesure", C.R. Ac.ad. Sc.. Pa!UJ.J,
I, 30I, I985, p. 687-690.
20. BOUCHITTE G., SUQUET P. "Charges limites, plasticite et homo-
geneisatiofl: le cas d'un bard charge", C.R. Ac.ad. Sc.. Pa!UJ.J, I, 305,
I 98 7, p. 44 I -444 .
21. SUQUET P. "Existence and regularity of solutions for Plasticity
problems" in va!Ua:tton.at MeA.:hod6 -<-n. :the Mec.haMCJ, o t] Soud6, Ed
S. NEMAT-NASSER, Pergamon Press, Oxford, I980, p. 304-309.
22. SUQUET P. "Sur les equations de la Plasticite : existence et re-
gularite des solutions". J. de ,IAec.aMque, 2d, 198I, p .. 3-39.
23. DEHENGEL F. "Espaces de fonctions a derivees mesures et applica-
tion a. l'analyse limite en pla.sticite", C.R. Ac.ad. Sc.. P~, I, 302,
1986, p. 179-182.
24. HADHRI T. "Sur ~e glissement a l'interface de deux milieux homo-
genes constituant un solide de Hencky", C.R. Ac.ad. Sc.. Pa!UJ.J, II, 302,
1986, p. 1181-1184. .
25. DAL MASO G. "Integral representation on BV(S"l) of r-1imit of
variational integrals", Man.w.,c.JUp:ta Ma:th., 30, 1980; p. 387-416.
26. BOUCHITTE G. ''Convergence-et relaxation de fonctionnelles du
calcul. des variations a croissance lineaire. Application a l'homogenei-
sation en plasticite", Ann.. Fac.. Sc.. Toulow.,e., 8, 1986-1987, -p. 7-36.
27. HERCIER B. "Une methode pour resoudre le probleme des charges
limites", J. Mec.a.,16, 1977.
28. FREMOND M., FRIAA A. .. "Les methodes statique et cinematique en
calcula la rupture et en analyse limite", J. Mec.a. Th. Appl., 1,
1982, p. 881-905.
29. JOHNSON C. "Error estimates for some finite element methods for
a model problem in perfect plasticity", Meeting "Problemi matematici
della meccanica dei continui", Trento, Italy, 1981.
30. CHRISTIANSEN E. "Computation of limit loads", In.:t. ]. Num. Me:th.
Eng., 17, 1981, p. 1547-1570.
31. NGUYEN DANG HUNG "Sur la plasticite et le calcul des etats limites
par elements finis", These de Doctorat Special. Liege, 1985.
32. GAVARINI C. "Plastic analysis of structures and duality ~n linear
programming", Me.c.c.aMc.a 1, 1966.
33. DUVAUT G., LIONS J.L. Lv., -<_n.equa:Uon.-6 e.n. Mec.aMque e:t en. Phy-6-<_-
que., Dunod, Paris, 1972.
34. NECAS J., HLAVACEK I. Ma:thema:Uc.at :the.o~y ot] elaJ.J;(_:_{_c. and
elaJ.J;(_:_{_c.o-plaJ.J;(_:_{_c. bo~v., :an. -<-~oduc.;t;_on., Elsevier, Amsterdam, 1981.
35. KOHN R., TEMAM R. "Dual spaces of stresses and strains with
applications to Hencky Plasticity", Appl. Ma:th. Op;t;_m., 10, 1983, p. 1-35.
36. HADHRI T. "Convex function of a measure and application to a
problem of nonhomogeneous plastic material", Preprint 140, Ecole Poly-
technique, Palaiseau,France, 1986.
37. BEN DHIA H., HADHRI T. "Existence result and discontinuous fini-
te element discretization for a plane stress Hencky problem", Preprint
146, Ecole Polytechnique, Palaiseau, France, 1986.
38. BEN DHIA H. "Numerical analysis of a bidimensional Hencky problem
approximated by a discontinuous finite element method", Preprint I61,
Ecole Polytechnique, Palaiseau, France, 1987.
Discontinuities and Plasticity 339

39. JOHNSON C. "Existence theorems for Plasticity problems", J. Math.


PutLe.o App.t., 55, 1976, p. 431-444.
40. SUQUET P. P.fcv.,tic..U:e e:t homogene~ilion, These d'Etat, Paris,
1982. See also "A few mathematical aspects of incremental plasticity",
Cours CIMPA, Monastir, 1986. To be published. Internal report 86-2,
Montpellier, France.
41. BREZIS H. "Un probleme d'evolution avec _contraintes unilaterales
dependant du temps~', C.R. Ac.a.d. Sc.. Pa..tU.J.,, A, 274, 1972, p. 310-312.
42. MOREAU J.J. "Evolution problem associated with a moving convex
set in a Hilbert space", J. Vi66 Eq., 26, 1977, p. 347-374.
43. ANZELOTTI "On the existence of the rates of stress and displace-
ment tor Prandtl Reuss Plasticity", QuaJt-t. AppL Ma.:th., 46, 1983,
p. 181-208. .
44. PERZYNA P. "Fundamental problems in viscoplasticity" in Adva.nc.u
in Applie.d Me.c.han-i.CJ.J, 9, 1966, p. 241-258.
45. FRIAA A. "Le materiau de Norton-Hoff gene-ralise et s~s applic<=~-
tions a 1 I analyse limhe", c. R. Ac.a.d. Sc.. Pa..tU.J.,' A, 286, 1978,
p. 953-956.
46. JOHNSON C. "On plasticity with hardening", J. Ma.:th. AnaL AppL,
62, 1978, p. 325-336.
47. NGUYEN QUOC SON "Materiaux plastiques ecrouissables. Distribution
de la contrainte dans une evolution quasi-statique", Attk. Me.c.h., 25,
1973, p. 695-702.
48. JOHNSON C. "A mixed finite element method for Plasticity with
hardening", SIAM]. Nume.Jt. AnaL, 14, 1977, p. 575-583.
49. DJAOUA M. "Analyse mathematique et numerique de quelques problemes
en mecanique de la rupture", These de Doctorat d'Etat, Paris, 1983.
50. GLOWINSKI R., MAROCCO A. "Sur l'approximation par elements finis
d'ordre un et la resolution par penalisation-dualite d'une classe de
problemes de Dirichlet non lineaires", RAIRO, 1975; p. 41-76.
51. CAILLERIE D. "The effect of a thin inclusion of high rigidity 1n
an elastic body", Ma:th. Me.:th. AppL Sc.., 2, 1980, p. 251-270.
52. BLANCHARD D., PAUMIER J.C. "Une justification de modeles de
plaques viscoplastiques", RAIRO AnalyJ.Je. Nume!tique., 18, 1984, p. 377-406.
53. DESTUYNDER P., NEVEU D. "Surles modeles de lignes plastiques
en Mecanique de la Rupture", Ma.:th. Mode.lling and Nume.Jt. Anal. , 20,
1986, p. 251-263.
54. BREZIS H., CAFFARELLI L., FRIEDMAN A. "Reinforcement problems
for elliptic equations and variational inequalities", Ann. Ma.:t. PU!La.
AppL, 123, 1980, p. 219-246.
55. ACERBI E., BUTTAZO G. "Reinforcement problems in the calculus
of variations", Sc.uo.fa No!tm. Sup. P~a, 1984.
56. BOUCHITTE G. Private communication, 1986.
57. SUQUET P. "Analyse limite et homogeneisation", C.R. Ac.ad. Sc..
Pa..tU.J.,, II, 296, 1983, p. 1355-1358.
58. DE BUHAN P. "Approche fondamentale du calcul a la rupture des
ouvrages en sols renforces", These d'Etat, Paris, 1986.
59. TURGEMAN S., PASTOR J. "Comparaison des charges limites d'une
structure heterogene et homogeneisee", J. Me~a. Th. App.f., 6, 1987,
121-143.
340 P.M. Suquet

60. BUTTAZO G., DAL MASO G. "r limit of integral functionals",


J. AnaL Math., 37, 1980, p. 145-185.
61. DE GIORGI E., AMBROSIO L., BUTTAZO G. "Integral representa-
tion and relaxation for functionals defined on measures", Scuola Norm.
Sup. Pisa, 1986.
62. AZE D. "Homogeneisation primale et duale par eoi-convergencP".
Publications AVAMAC. Universite de Perpignan, 1984.
&3. DAL MASO G., LONGO P. "r-limits of obstacles", Ann. Mat. Pu.tc..a
AppL, 128, 1981, p. 1-50.
64. ATTOUCH H., PICARD C. "Variational inequalities with varying
obstacles", J. Fu.nc;t, AnaL, 50, 1983, p. 329-386.
65. BOUCHITTE G. "B.omogeneisation sur BV(S'2) de fonctionnelles
incegrales a croissance lineaire", C.R. A~ad. S~. P~, I, 301, 1985,
p. 785-788.
56. DEMENGEL F., TANG QI "Homogeneisation en Plasticite", C.R. A~ad.
S~. P~, 303, I, 1986, p. 339-342.
67. DJAOUA H., SUQUET P. "Evolution quasi-statique des milieux
viscoplastiques de Maxwell-Norton", Math. Meth. Appl. s~., 6 ' 1984,
p. 192-205.
68. BLANCHARD D., LE TALLEC P. "Numerical Analysis of the equations
of small strains quasistatic elastoviscoplasticity", Nu.me~. Math., 50,
1986, p. 147-169.
TOPICS ON UNILATERAL CONTACT PROBLEMS
OF ELASTICITY AND INELASTICITY

J.J. Telega
Polish Academy of Sciences, Warsaw, Poland

ABS'IRACT
These lectures present selected unilateral boundary and
initial value problems of elasticity and inelasticity.
Particularly, for elastic Signorini's problem with general
subdifferential sliding rule t\~O dual problems are derived
as well as the bidual one. Some historical and modern
views on friction phenomenon are briefly presented. The
next chapter is concerned with unilateral frictionless and
frictional problems for viscoelastic, plastic and visco-
plastic bodies. In the last chapter homogenization of mi-
crofissured elastic solids and plates is studied. It is
assumed that microfissures behave unilaterally.
342 J.J. Telega

1 IN'IRODUCTION
These lectures are focussed on variational approach
to unilateral contact problems for elastic and inelastic
bodies. Non-smooth effects are imposed by unilateral con-
ditions at the boundary. In the case of perfect plasticity
/Chapter 3/ and fissured solids and plates /Chapter 4/
certain functions like displacements and velocities are al-
so non-smooth /discontinuous/.
In my opinion it is covenient to classify contact
problems of solid and structural mechanics as follows.
/ i / External and intern~l contact problems. In the first
case the body considered is in contact with another rigid
or deformable body /elastic or inelastic/, or with a sys-
tem of such bodies. Internal contact problems are posed by
fracture and damage mechanics, see Chapter 4 of these lec-
tures.
/ii/ Contact of bodies made of: inorganic materials, for
instance metal - metal contact;organic materials, for ex-
ample cartilage - cartilage contact in human and animal
joints, see Refs [1,9,10,22,26,27,36,37;38,40,52] and
Fig.1;

bone

synovial Fig.1, Diagrammatic


fluid synovial representation of a
membrune human joint

artic ular
cartilage

organic and inorganic materials /in inplants, for instance


bone-polymethyl methacrylate cement 1 bone-metal contact/,
of. Refs [2,6,12,13,16,17,28-31 ,48,49,57] and Fig.2.
/iii/ Dependence upon friction: frictionless contact, con-
tact with friction but without lubrication /dry contact/,
lubricated contact , see Refs [4,15,20,32,42,45,46,53,56].
/iv/ Unilateral or non-unilateral conditions.
/v/ Presence or absence of adhesion /see the lectures by
Fremond/ and wear.
Obviously, in real situations various combinations of
points / i / - /v/ occur.
Now I shall briefly present the subject of my lectu-
res. Since I largely deal with unilateral contact problems
with friction, therefore some historical and modern
Topics on Unilateral Contact Problems 343

Fig.2, Schematic illu-


stration of a Charnley to-
tal hip prosthesis /ace-
tabular and femoral com-
ponent/ fixated in the
bone with acrylic cement

cortical bo ne
[0J spongeous bone
EJ acrylic cement
lllliD polyethylene
D metol
fe mur

concepts of the friction phenomenon are first sucointly


presented. Various phenomenological friction laws or sli-
ding rules are next introduced. Both associated and non-
-associated laws are discussed. Of main concern in these
notes are subdifferential laws which are very convenient
in variational formulations. Having formulated the elastic
Signorini's contact problem with friction I pass to the
study of dual and bidual problems. Toward this end the the-
ory of duality in the sense of Mosco-Capuzzo Dolcetta-Mat-
zen /M-CD-M theory o duality/ is used. Existence problems
are also discussed .
Chapter 3 deals with unilateral contact problems for
inelastic solids: viscoelastic, rigid-perfectly plastic,
elasto-plastic and viscoplastic. Large deformations are
also mentioned.
In the last chapter homogenization or averaging of
periodically microfissured elastic solids and plates is
considered. I assume that microfissures behave unilateral-
ly. Therefore suoh problems leads naturally to homogeni-
zation of variational inequalities posed on domains depen-
ding on a small parameter After homogenization
we arrive at equations since microfissures are "smeared
out". In the case of frictional behaviour of microfissures
implicit variational inequality has to be homogenized.
Before ending this introductory chapter 1 would like
to comment briefly on some relevant references. The reader
interested in classical contact problems should refer to
344 ].]. Telega

the books [23,24,2S,32,4 7,5D. The classical, non-variat-


ional contact mechanics deals essentially with problems in-
volving no unilateral conditions at the boundary. The
founder of such approach was Hertz [54). The survey paper
L33) may serve as a good introduction to both the classic-
al and unilateral contact mechanics. Various interface
problems, including unilateral conditions, are investigat-
ed in [50). The books [14,18,34,44,46 ] present numerous
results on the variational approach to the unilateral con-
tact mechanics. The extensive paper [55] summarizes re-
sults obtained by using such approach to the solid and
structural mechanics in the case of both elastic and in-
elastic materials.
The books [3,5,7,11,19,20 ,21,35,41] and the lengthy
paper [8) are strongly advised to mathematically inclined
reader.

2. DUALITY FOR SIGNORlNI 'S PROBLEMS WITH FRICTION lN TilE


CASE OF LINEAR ELASTIC MATERIALS
2.1 Origins of friction as a science
Before passing to the presentation of phenomenologi c-
al friction conditions and laws it seems worthwile to know
something about the origins of friction as the science. In
this regard confusion exists in the current literature,
particularly in textbooks. An excellent historical account
of friction up to modern times presents the book by Dowson
[931 First important contributions Dowson [93) attributes
to Leonardo da Vinci /1452-1519/. However, Leonardo's in-
vestlgatloiia- of-frlotion remaine.d unfamiliar since, as
Dowson writes: "Valuable notebooks /of Leonardo da Vinci -
J.J.T./ containing records in the form of sketches and no-
tes of studies of bearings, friction and wear were found
in Madrid as recently as 1967 His experimental approach
was essentially the same as that employed by Charles-Augus-
tin de Coulomb some three centuries later ".
Guillaume Amontons /1663-1705/ Due to studies by Dow-
son we-now-kiiow-l bil-Imontons investigated not "dry" frict-
ion as is commonly believed but the frictional characteris-
tics of surfaces greased with old pork fat under the con-
ditions nowadays described as boundary lubrication. Amen-
tons' observations embodied thefirst and second laws of
friction, that is: 1. the friction force is proportional
to the applied load, 2. the friction force is independent
of the apparent area of contact. Amontons attributed fric-
tion to r~id and elastic asperities.
These two empirical laws of friction can in fact be
deduced from Leonardos studies.
Topics on Unilateral Contact Problems 345

Phillippe de la Hire /164o-1718/ introduced what is


now caiiia-tne-ooncept-o? permanent sur~ace de~ormation
and shearing o~ the inter~ace /asperities/. He pointed out
the possibility that resistance to motion would depend up-
on the number o~ asperities, and hence the size o~ the
sur~aces.
Gott~ried Wilhelm von Leibniz /1646-1716/ made a dis-
tinctlon-betwiin-iiia!Di-ana-roiilng ~riction.
John Theofhilus Des~liers /1683-1744/ initiated an
alternatlvi-view-o?-the-?rlctlon phenomena, namely he con-
sidered the ~luence o~ adhesion between contacting bo-
dies /and not cohesion, as Dowson misleadingly writes, see
[78)/.
Leonard Euler /1707-1783/. An importance o~ Euler's
contriSution-Ii-aue to the development o~ an analytical ap-
proach to ~riction, the introduction o~ the symbol ~ ~or
the "coe~~icient o~ ~riction" as well as the distinction
between static and kinetic ~riction.
Isaac Newton /1642-1727/ In Section IX o~ Book II o~
his Prlnclpla-Rathematica Philosophiae Naturalis /1687/,
which deals with "Circular Motion o~ Fluids", the term
"want o~ lubricity" was introduced. It corresponds to what
we now describe as internal ~riction or viscosity.
Charles Auustin Coulomb /1736-1806/. He studied ex-
perimentaiiy-iuch-matirlaii-as oak, green oak, guaiac wood,
elm and ~ir in various combinations in both dry and lubri-
cated conditions. Coulomb ~ound that in most cases ~riction
was almost proportional to load and independent o~ the size
o~ the contacting sur~aces. In the case o~ ~rictional re-
sistance to sliding o~ horizontal sur~aces the limiting
~riction ~orce was written as

F =
a + P~ ' /2.1/
where ~ is the inverse (!) o~ the coe~~icient o~ ~riction~
while a was attributed to adhesion. For Coulomb, however,
adhesion had only a very small ~luence upon ~riction.It
is woth noticing that he was the ~irst to use the two-term
expression /2.1/ ~or ~riction, where the second term is at-
tributed to de~ormation or ploughing action o~ asperities.
Coulomb investigated also kinetic and rolling ~riction.
2.2 On ~riction mechanisms and coe~~icient o~ ~riction
Nowadays it is commonly assumed that ~riction between
sliding sur~aces is caused by three phenomena: 1/ de~ormat
ion o~ sperities, 2/ ploughing by wear particles and hard
sur~aces asperities and, 3/ adhesion; see also Fig.3.
Thus the coe~icient o riction expresses as ~ollows
Jk = F/P = ~ (JLd}J-p}J-a), /2.2/
where ~d' flp and fla denote the asperity de~ormation
346 J.J. Telega

DEFORMED LAYERS

~~ ~
-- --- ---
--- --

A -PRIMARY ADHESION
d2 Fig.J, Typical asperity and
wear particle at a sliding
interface (1 .37]

B -PLOWING OF HARD ASPERITY


C -SECONDARY ADHESION
D -INTERLOCKING OF ASPERITY

component, the ploughing component and adhesion component,


respectively; F is the friction force while P denotes the
applied load. The relative contribution of' these components
depends on the condition of' the sliding interface, which
is affected by the history of' sliding, the surfaces topo-
graphy, the specific materials used and the enviroment.
Suh and Sin [16.3] affirm that the coefficient of' friction
is not an inherent material property.
The-static coefficient of' friction is determined primarily
by asperities deformation.
Figures below exemplify influences of' various factors
on the coefficient of' friction.

3 In air - Diamond-Pt
at room
z tempera lure
0
1-
w 2
0:
LL.

LL.
0
0
1-
z
LLJ
w +
"-_Cl..o-0-o--D- Diamond-( u
LL.
LL.
LLJ
-o-
0
w
500 1000
TEMPER A TURE o C

Fig.4, Friction of' diamond on metal after cleaning


in vacuum. High friction, especially with platinum,
shows that very strong interfacial adhesion can
occur [78)
Topics on Unilateral Contact Problems 347

-----TimED SURFACE
---SURFACE RUSHED WITH ALUMINA
t- --ELECTROPOLISHED SURFACE
z
w
u
u..
u.. I
w I
0
u 3 \ c
\ '' _.,------------------a
z
0 .2 \ ' ' .......
''---------~-------------b
_____________ , ,,
1-
u .1
0::
u..

30 IIJ so 60 70 00 90 10
SUCCESSIVE PASSES OVER FRICTION SURFACE

Fig.5, Effect of surface condition of aluminium


on friction coefficient [78)

AIR AT 1ATM SATURATED WITH


D VACUUM (10-lOtorrl

w
.....
0::
0
u..

z
0
1-
u
0::
u..

0 200 1.00 600 l(XXl 1200


LOAD , G

Fig.6 1 Friction force as function of load for glass


sliding on glass. Sliding velocity 30 em/min; tem-
perature 2j0C [78]

We observe that glass is one o the few materials ex-


hibiting higher friction in moist air than it does in a
348 ].]. Telega

vacuum environment in the olean state. In the olean state


the bonding oroes are weaker than they are when moisture
is present on the solid surface.

COMPLETE SEIZURE

......
u
0::
u...
u...
0

......
z
LLJ
u 1.5
u...
u...
LLJ
0
u ---IRON
-------~PERCENT
SILICON -IRON

10-10 10-s 10-6 10-4 10-2 10


0.0 '-:-:----'--::----'-:----'-:--~---'--::-----'--::----'-;---

OXYGEN EXPOSURE torr- sec


10 2 104

Fig.?, Coefficient o friction for iron and 3~ sili-


con-iron as function of oxygen exposure.
Sliding velocity, 0.001 cm/s; ambient temperature 20C;
ambient pressure, 10- 10 torr [78]

Though many authors emphasize the role of adhesion yet


Bikerman [75] attributes no significance to it. He writes
/p.13/: "As long as the term friction is employed in its
common meaning, adhesion does not cause friction and has no
influence on frictional phenomena"
Basing on some available experimental results Madakson
~37] proposed to treat the friction force F as a unction
of the applied normal load P, sliding velocity v, real area
of contact A, density 9 , hardness H, suraoe roughness
R, suraoe energy S, deformation energy E, the deformed
area Q, temperature T, the environment D and the experiment-
al system Y. Thus we oan write
F =F(P,v,A, g ,H,R,S,E,Q,T,D,Y). /2.J/
Dimensional analysis yields
Topics on Unilateral Contact Problems 349

/2.4/

where K is a constant depending on temperature, environment


and experimental system. The formula /2.4/ applies to two
identical materials. In the case o dissimilar materials ~
is the density of the softer one.
We observe that in Madakson's formula /2.4/ the temperature
does not enter explicitly. It seems, however, that in a ge-
neral formula the temperature should be explicitly present
according to experimental results, see (76,79,134] and Figs
8,9.

z 0
0
i=
u
0::
u...
u...
0
Sn-Cu
I-
z
UJ
L.J
u... 0
u...
UJ
0 h o S n-A l
u Tin crystal
trnnsformation 13(
;.. l l l
~150 -100 -so o so 100 1so
TEMPERATURE o C

Fig.8, Coefficient of friction for polycrystalline tin


and tin alloys at various temperatures. Sliding velo-
city, 0.7 mm/min; load 10 G; 8 2 10
pressure 1,JJx10- N/m /10- torr/[78]
For instance, Bereznyakov et al. [71] proposed the fol-
lowing formula for temperature dependence of the friction
force 1/
F =0,15[Pc(1+ocT /HB ] 4 P+5,9410- 2 _51 Lln(TwfT) /2.5/
0
where oe is a small parameter, oe ~ 0,01; HB denotes hard-
ness in the sense of Brinell at T = 29JK, pc0 is the contour
pressure while T and L stand for melting temperature and
m
specific heat, respectively. Hence we infer that JL also
depends on T.
350 J.J. Telega

0.5
~~.
<>~ [l....Q.....Q._D-c-o-c
~ <>~----<>
04
:::L

c:
03
:~~-
51 ( """ 'b
0 + (a l
:;::: 38( ......... "o

--
u 0 +
;: <> + 18 ( "'--.!'
02 4 - 8 (
0 0 -41 (
...... v-72(
c:
0.1

-
<U
;:;
;;::::
~
w
-4 -3 -2 -1 0
log v

0.4
~

c:
0 0.3 ( b)
......u

--
;:

0
0.2
......
c:
0.1

--
<U
u

<U
0
w
-4 -3 -2 -1 0
log v

Fig.9, Sliding riction o a hard steel slider over


,a polymer surace as a t.mction o speed and tempera-
ture. /a/ Nylon 6,6; /b/ Low-molecular weight branched
polyethene [135]
Speciic orm o /2.4/ is [137)
jJ-= F/P = KF1 ( 9Av
p
2
S ) F) PAR
) F2 ( Hv2 (!!i...) /2.6/

I the sliding process is in equilibrium then we can assume


Topics on Unilateral Contact Problems 351

that the functions F 2 and F 3 are equal to constants. Hence


2
jJ- = K1F 1 ( 9 ~v ) /2.7/

An approximate form of /2.7/ is


,Av2) c
}J- = K1 ( p /2.8/
where c is a constant.
If a loaded hemispherical slider rests on a flat sur-
face of an identical material then A:K 2 Pn, where K2 and n
are constants; n=1 for plastic deformation and n<1 for
elastic deformation.
Fig.10 represents experimental and theoretical results, the
latter being obtained from /2.8/ for c=0,12 and K 1:2,4 [13~
::::r
z: 1.00
Q
t::; .80
a:
u...
u..
0
.60 Fig.10, Theoretical/--/ and
1-
experimental / . / variations of
z: .40 friction coefficient with load
w
LJ for Duraluminium sliding on
u..
u...
.20 aluminium at 154 mm/s [137]
w
0
LJ 0. 00 t___.__.....___,__--'----'---'----'-__.._____J
0.00 I4J 00 80.00 mxl 1ffi.OO
20.00 ro.m rom 1m.oo 1so.oo
LOAD N

Further theoretical results derived from /2.8/ are presen-


ted in Fig.11
I a) I bl

z
0
"
>=
u
fE
~
0
'L
C"
>-
z
w
u --~-~
zo
~ ~~t::
~ 8
;JJOO 'IJOO 1!100 llJOO 10000 zoo 400 600 800 1000

AREA OF CONTACT I PERCENT I SLIDING VELOCITY I M/S I

Fig.11, Plots of friction coefficient against:/a/area


of contact,/b/sliding velocity for Duraluminium on aluminium
352 ].]. Telega

From the last figure we infer that with increasing loading


the coefficient of riction becomes a1most independent of
velocity and area of contact. Thus the classical observat-
ions by Leonardo da Vinci and Coulomb are verified.
Experimental results of the dependence of ~ _on the
sliding speed are presented in Figs 12 and 13 ~37J

(aI ( bI

45 45

40 40

0 606 N ~ 35
::1.
:z 3.0 :z
0
0
E 25 0 25

f
cr
~ 2 0 ..--- 1 52 N ~ LL

LL
20

~;~c~
~ 15~ t
!
i:J 10 212 N

200 300 II1J ~ rro m an 'm 1000


~ ~~ 200 :m
=+=.

II1J 'jX)
: ==t' '
'==:1,:K:ri:::4 ~
600 700 !lXl 900 1000
SPEED , RPM SPEED. RPM

Fig.12, The effect of applied load on the variation in


friction coefficient with speed for: /a/ silver steel
on aluminium, /b/ copper on copper /lubricant, white
spirit/

We observe that a rapid increase in load during sliding


leads to a sharp decrease in the coefficient of friction,
see also r103).
Earlter I have already mentioned about the importance
of the real area of contact. Deformatio~ of asperities is
obviously closely related to this area l70, 164, 165] Depen-
ding upon applied loads and specific materials used asperit-
ies deform elastically f72) or inelastically.
The study of plastic detormation of asperities have been
preferred, what seems natural [83,100,119,156] ; see also
(32,56,102,106,108,169].
To study the sliding contact between unlubricated me-
tals Bowden and Tabor [77] assumed that normal load is sup-
ported by minute junctions which are continuously being
formed and broken between the surfaces as they slide over
each other. By using this concept in conjunction with
Griffith theory of fracture and Rice J-integral De Celis
[90) estimated the size of junctions. Thus for brittle ma-
Topics on Unilateral Contact Problems 353

I al Ib l
RPM =rev/ mm RPM = rev I m1n
P =0,606 N P =2,12 N

45
40
::t
35 35
1-
;:, 3 0
21'1lRPM f+445RPM f-960RPM f+ ZIBRPM
u
E
w
25
8 20

~ 15
I-
!:! 1 0
a:
LL
05
oo~~~~~--~~~~ oo~~~~~~~~~~.
20 25 J) 35 40 20 25 30 35 1.0
TIME , MINS TIME, MINS

Fig.1J, The e~~ect o~ rapid change in sliding speed


on the ~riction coe~~icient ~or silver steel on
aluminium /lubricant, white spirit/
terials they are o~ the order o~ 10- 9 m while ~or ductile
materi~ls their size is about 10- 4 - 10-5m.
With sliding ~riction at low speeds is closely connec-
ted the complex but very important phenomenon o~ "stick-
-slip", see (60,67, 155]. The study o~ this process have
even raised some doubts about the validity o the concept
o~ static riction.
Apart ~rom metals o~ practical interest are rictional
characteristics o~ many other materials like wood [64, 16.5) ,
polymeric materials [.56,65,69,135,165], rubber f_68,165) and
ceramics (:16 5]
In my subsequent presentation o boundary value prob-
lems, generally anisotropic riction conditions and sliding
rules will be taken into account.
Directional e~ects occuring during sliding have already
been reported in the available literature [59,78,1 6,173],
see also the ~ollowing ~igure 14.
For a study o~ a structare-soil interaction ~rictional
characteristics play an important role. A detailed experi-
mental study ~or the determination o~ the ooe~~icient of
~riction between various sands and steel was carried out in
the papers b24,170] , see also the reerences cited there-
in and [118) The coe~ficient of ~riotion was ~ound to be
354 ].]. Telega

(a)

( b)

O.OJ ~
0.08
~
0.07 1~:----o
o.~ \ ,.._____A~------
~ 0.(15
~ --=-------------0------
~------ t:r-_

0.04

0.03

0.02

Q01

3060 125 250 500 1000


LOAD G

Fig.14, /a/ The four possible directions of sliding


for an oriented PTFE/polytetrafluor thylene/ slider on an
oriented PTFE surface, /b/ the variation of the coefficient
of friction between an oriented PTFE hemispherical slider
and an oriented PTFE surface with load. o, direction d 1 ;
x, direction d 2 ; A , directions dJ and d 4 [166]
strongly influenced by sand type and surface roughness of
steel while influences of normal stresses and mean grain
size were of little significance.
Last but not least are frictional characteristics of hu-
man and animal joints, see [22 ,94,112, 144] In this case
three types of joints must be distinguisned: normal, patho-
logical and artificial ones. Fig. 15, taken from [ 61] ,
exhibit that the coefficient of friction in normally funct-
ioning synovial joints is extremely low.
Topics on Unilateral Contact Problems 355

(a l
LOAD

(bl

z
0
I-
~ Q02
~

I-
~ 0.01
9u.
u.
w
~ 0.00 +------'.lL.:...._..,.---=:Jc:.__--+J
0.0 1.0 2.0(MN/m2)
NORMAL STRESS
Fig.15, /a/ Experimenta1 configuration, /b/ variation
o~ ~riction coe~~icient ~or articular cartilage with
load
Dawson et al. (94) reported the results o~ an experiment in
which a severely damaged rheumatoid hip was investigated
at loads up to 1500 N. The coe~~icient o~ ~riction was as
high as o.4, thus indicating a ~i~teen-~old increase over
the average value o~ 0,025 recorded ~or healthy hip joints.
2.3 Phenomenological ~riction conditions andsliding rules
Physical considerations like those pres~ted in the
preceding section are very important in our understanding
o~ ~riotion phenomenon yet are insu~~ioient ~or the study
o~ boundary and initial value problems. For instance, ~riot
ion laws or sliding conditions do not ollow ~rom such con-
siderations, so they must be postulated.
A broad class o~ sliding rules can be obtained from
the corresponding constitutive equations o~ the mechanics
356 ].]. Telega

o~ solids. Such point o~ view is now brie~ly presented,see


also R3a] Though I restrict mysel~, ~or the sake o~ simp-
licity, to isothermal case and small.displacements yet non-
-isothermal laws such as those resembling thermoplastic
constitutive equations could likewise be proposed. For in-
stance, one could use the papers 029,1541 to ~ormulate
thermoplastic sliding rules ~or both smafl and large slid-
ing displacements.
Felder [99] and Zmi trowicz b74] have proposed to mo-
del phenomena occurring in contact layers by introducing
o~ a material inter~ace. Such approach seems to be promis-
sing, yet the inter~ace plays the role o~ third, two-dimen-
sional body. Consequently the contact problem becomes more
complicated due to large number o~ independent physical
fields like inter~ace stress tensor; the latter may even
be non-symmetric 1'174] Signorini 's contact conditions,
which are o~ main "interest in these lectures, were not ta-
ken into account in ~9, 174].
Let us pass to isothermal sliding rules. Suppose that
a material particle M o~ a deformable /or rigid/ body ~
is acted upon by a ~orce s = (sN!!rr) , ME:-S, see Fig.16.
Here S is the plane domain o~ contact with rigid plane
Y.3=0, and !.T~T~)( ce=1 ,2)
is the tangential ~orce
vector or ~riction ~orce
acting on M while ~
stands for the normal
component o~ s. Let
!=~i~ be a stress ten-
sor, devi its deviatoric
part and p= (tr!,)/~ .Fur-
ther, let a
be the exter-
ior unit normal to y 3 :o.
and e= (e . ) the strain
- 1j/
tensor; i 1 j=1,2,3.

n.lS

Fig.16
Topics on Unilateral Contact Problems 357

--
Under the f'ollowing correspondences
e e + ep u e + up
. == . == -
-
e u

--
!.
e P
.!! + .!. u ue + uP
t = devt + P.! s = ~!!. + !.T ,
various sliding rules can readily be f'ormulated; the super-
scripts "e" and "p" mean "elastic", "plastic" /or generally
inelastic/, respectively. Here u is the displacement vector
of' M and u
its velocity. -
For suf'ficiently small loads the material element M
def'orms elastically, thus ~p =
o. Then we can assume
u;; = ~ uToe = E.,c j s 1
= ksN, T /2.10/
in the linear case. Here k> 0 , ! =! and

Ect: j see s.f )/ k 1sce see , k 1"> 0 '!r/!. = ( s~) /2.11/


Suppose now that M deforms plastically, see (58]. Obviousl.y
if the body considered is rigid then the term "plastic" has
no meaning yet plasticity type sliding rules stil.l apply.
Let a f'riction condition be given by

g(~, !.TJ ~ 0 /2.12/,


where g is not necessarily isotropic.
Assume that the set
/2.13/
is convex and closed. The associated subdifferential slid-
ing rule has the form

/2.14/
where ~ is the indicator function of K

~(!.) = ( ~ ' ,i!thir~i!e /2.15/

Such friction law is not very reasonabl.e physically since


it yields the normal. component of' the velocity ~*0 In
the case of' Coulomb's friction condition

I!.T I - fL I~ I ~ 0 ) I 2. 16/
we obtain U: = - A sgn sN = >.,~ )'/ 0 since ). )1 0 and
358 J.J. Telega

sN ~ 0 /non-physica1 situation/.
If instead of /2.14/ we take

(-U:: , -~) t d~(.:!.)' /2.17/


then we get ~ = - ~ JA-E:. 0 and M wou1d be forced to 1eave
the p1ane y =
o. The friction 1aws /2.14/ and /2.17/ cor-
respond to ~onstitutive equations of associated p1asticity
of compressib1e materia1s. Such 1aws cou1d probab1y be used
in the case of a deformab1e foundation.
Of practica1 interest are the s1iding ru1es correspon-
ding to non-associated incompressib1e p1asticity. Now the
friction condition is sti11 given by /2.12/ yet the assump-
tion concerning convexity may be weakened. We set [167]
K ( ~) = ti = (~ I g l~, i) ~ 0 ) ' I 2. 18/
and assume that for each s the set K{~) is c1osed and con-
vex. Thus the set K, defin~d by /2.13/, is not necessari1y
convex too.
As is known Cou1omb's friction 1aw expresses as fo11ows

>
p
l!rrl < p.l~l ~ =0 /2.19/
l!rrl = t- 1~1 ) P
~
= -).. ~T '

Cou1omb's friction /s1iding/ potentia1 h is thus given by


h(.:!.TJ =l!frl +P , /2.20/
where p is a parameter,since potentia1 is in fact a one pa-
rameter fami1y. The parameter p is defined from the condi t-
ion h(!.T) = g(~!rr). The potentia1 /2.20/ yie1ds

= - >. = -). !.T t. = 'A 1


~ ~

~ C) h/o sT , .!T
or
/2.21/
-~ E: o~(sN)(.:!,T) , where K(~)= ti\1.11- fA. I~!~ 0)
Hence the friction 1aw /2.19/ is not associated with
Cou1omb's friction condition /2.16/.
We observe that subdifferentia1 approach to contact ru1es
was initiated by Moreau, see (140).
I pass now to the genera1 case. The s1iding ru1e is as-
sumed in the form

-~ ~ ()~(~) (!.T) 1
/2.22/
which is equiva1ent to
Topics on Unilateral Contact Problems 359

. - !.TI. . ~
(! );. 0 , /2.23/
Denoting by t the conjugate variable, the support function
d (~,,!*) o:f-the convex set K(~) is given by

d(~,!i-) = s~p(.f ! -~(~) (!' =


-
=sup fl !) !~K(~)~. /2.24/
The :function d (~, ) is positively homogeneous o:f degree
one f157).
Duality o:f the indicator :function ~ {~) ( ) with the
support :function d(~,.) results in

!tr E "d 2 d(~,-~) , /2.25/

where ~ 2d( ~ ,-~) denotes the subdif:ferential- of the


:function d(~,) Hence we in:fer that~=-~. Either of
sliding rules /2.22/ or /2.25/ is equivalent to

~(~)( !.T) + d(~, -~) = !,T ( -~) , /2.26/


or, since !tr ~ K(~) ,

j
p) d:f (
( ~~ = P)
d ~-!:!:f = !tr (-~
P)
. /2.27/
The :function j is the friction dissipation density. We ob-
serve that due to the nonassociated character o:f the :fric-
tion law /2.22/ or /2.29/, the :friction dissifation defends
exnlioitlv on normal stresse&-a;a-not-only-on -;p_____ ----
--~------'-------------------
By using Rocka:fellar's Th.1'Je5 we get
~ .
j (sN'~) = ~).. g*( ~- )\- 1 ~) ,:for il;#O. /2.28/
Obviously, :for ~=0 we have J(~,o) = 0
I:f K(~) is closed but nonoonvex then the friction law
/2.28/ remains valid, provided that d ~(~) (!fr) stands
:for ~~! subdifferential in the sense o:f Clarke [85] or ra-
th~lrookafellar. Unfortunately, in the nonconvex case the
definition o:f the subdi:fferential is more complicated, par-
ticularly in the case of :functions which are not locally
Lipsohitzian, see also the lectures by Panagiotopoulos.
360 J.J. Telega

I think that even in such nonconvex case one can still


consider convex sliding rules by simple application of the
duality; that is we take /2.25/ as the subdifferential
friction law since the function d(~,.) is always convex
[97] /and lower semicontinuous/. Also the formulae /2.26/
and /2.27/ hold.
Example 2.1. Suppose that g is symmetric /isotropic/
in the following sense:
g(A1~'!2!.TJ = g (~ '!.T) /2.29/
where A1 is the orthogonal transformation which reverses
the sign of the normal component, whereas A2 is an orthogo-
nal transformation. Then we have, see f157~p,11~
L .J

/2.30/

Hence we obtain
/2.31/

In the specific case of Coulomb's friction the last relat-


ion immediately yields
/2.32/

Example 2.2 [1671. As a natural generalization of


Coulomb's friction condition the following anisotropic one
is proposed
/2.33/

The tensor N can be called anisotropic friction tensor. We


assume that- N = ! T . Now we have

K(~)= l!' =(ti)C)I f Noe J toe tJ - H(sN) ~ 0). /2.34/


The set K(~r) is convex if and only if ~ is positive semi-
-definite. ~he nonassociated friction law takes on the form

where
-~=-ANs

A=
- -T
(1>1 o( , u.r
P aC P p )
ll;f
"
i/2
")/ 0 ,

~~ =
N-1
/2.35/

/2.36/
(2H (sN)) 1/ 2
provided that N is invertible.
The friction dTssipation density is expressed as follows
Topics on Unilateral Contact Problems 361

.( . p)
J sN'.!:.i' /2.37/

Non-associated sliding rules corresponding to compres-


sible non-associated plastic flow laws can also be studied.
In the case of purely static considerations the ve-
locity should obviously be replaced by displacement, see
the next section. Then friction laws correspond to the
deformational theory of plasticity sometimes called Hencky
plasticity. li'or instance, the subdifferential sliding rule
takes the :form
- !! G alrc1 (sN) (!.T) /2.)8/

Let us pass to sliding rules corresponding to incom-


pressible non-associated plasticity accounting :for hardening
and/or softening, see also [140]. No'" the :friction condition
is given by g(~,!_T,g_) ~ 0 , '"here (!_,,S) are generalized
forces. lie set

/2.)9/

and assume that K 1 (sN) is convex.


The generalized sliding rule is given by

(- ~' -9_) t 0 ~ 1 (sN) ( !.Tt.S)' /2.40/

Thus 9. is the generalized velocityassociated with S The


friction dissipation density is nol.r given by

/2.41/

Example 2.). By using anisotropic yield conditions of


plasticity ldth hardening it seems natural to generalize
the friction condition studied in Example 2.2 as follows

/2.42/
where, for instance~
362 J.J. Telega
t
q 2 ( t) = q 2 (0) + Sf ~ ~ l dt , t - time ~
0
and a stands or adhesion.
The sliding rule /2.4o/ gives

~ = - ~ N~' sT , 41or: = 'A NeeJ Q1; , 42 = )t eH/"0 Q2 , >. )} o


provided that!= !T is positive semi-definite and H(~}
is concave.
Remark 2.1
Dynamical equations o discrete systems with Coulomb
friction and Signorini's type conditions have been investi-
gated in [114, 152] , see also the paper by Moreau in [14].
Moreau [141] has formulated a subdierential sli-
ding rule not requiring the decomposition o the contact
force into normal and tangential parts.
Contact conditions are also discussed in [50,89,116,
125' 126' 147' 148]
2.4. Signorini's contact problem with friction: kinematic
and stress approach
In the sequel o the present Chapter I shall study in
detail the static Signorini's contact problem or a linear
elastic body. In this case the sli~ing rule is o deformat-
ional plasticity type /2.38/ and ~ = 0 , ~ = 0 /on the
surface o cont~ct/.
Let .Q. C RJ be a bounded sufficiently regular domain.
Its boundary o~ , denoted by s, consists o three nonover-
lapping parts: s 0 ,s 1 ,s 2 and s = 0 s us
1US 2 , where s 0 denotes
the closure o 50 , etc. By~ :(ni) we denote the outer unit
normal to s. Let ~ =(uJ be a displacement vector and
eij(~) = (ui,j + uj,i)/2 the strain tensor. Stress tensor
is denoted by~ :(sij). Latin indices take values 1,2,3
while Greek indices 1,2.
Since S is in general a surface and not a plane doma-
in thereori the sliding rule /2.38/ have to be reconside-
red. Let w c R 2 be a two-dimensional domain and s 2 = !Cw)
where ~ is o class C1 (C3 ) The plane R 2 is ref' erred to co-
ordinates (yoC) while R3 to (xi). For x = ~ (y) we set
!., ~ = 'd!/'6yc" , or xi, = () ~ i/'d y.::C. Now we have, on S

ui = ~ni + 'U.ri , sijnj = s~i + sTi /2.44/


where ,
Topics on Unilateral Contact Problems 363

Let ! = ( Aia = J [!; J


~ t!! = [xi ,oe ,ni] Obviousl.y, the
mat.ri.x A is invertibl.e. Hence we can write
~oe = xi, oc uTi ' u.ra = Aiall.ri ' ( u.ra) =(u.roe 'u.r3)

ST.~ = X.~,()C ST .,_


-r = Ai a STa '

Obviously we have u.r 3 = niu.ri = 0 , sTj = 0 , sTiu.ri =


= SToC u.r~
Let at a point x = ~ (y) the continuous friction con-
dition be given by g0 (~,sT~ )~ 0 Then, al.ways at the
same point,
g(~,sTi) = go(sN'!-1!tr) ' !.T = (sTi J /2.45/

The function g (sN, ) remains convex if such is g 0 ( sN, )


We set
K0 (sN) = ti =(toe) j g0 (~i) ~ o) /2.46/

K(sN) = ti = {ti) I g(sN,E_) ~ 0) . /2.47/


Hence
~(sN) (~T! -
and
/2.48/

The subdif'f'erential sl.iding rule has the form

/2.49/

The foll.owing elementary theorem is very useful [157,pp.J~


and [225) n m
Theorem 2.1. Let A be a l.inear transformation from R toR
m
Then for each convex function f on R , the function fA defi-
ned by
(fA) (z) = f(Az) ) /2.50/
n
is convex on R Moreover, if f is a proper lower semiconti-
nuous /:l.s.c./ convex function and there exists a point in
the range of A at which f is finite and continuous then
o(fA')(z) =AT a f(Az), /2.51/
where AT is the transpose of A.
364 ].]. Telega

By using this theorem, from /2.48/ and /2.49/ lve


readi1y get
(-u.rce ,o J = (-AiauTiJ f cl ~0 ( sN) (sTaJ = "d IK (sN) \Aia 8 TaJ
Hence
-ATUm AT
- -... -
d IK . . f Sm \
lSN)\-.a.J
t _sm
.a.
= {\ ST.;... '1\
Since ! is invertib1e, so we fina11y obtain

In the above considerations the function g0 (sN,~TJ


was identified 'dth a certain function g 1 (~,(~T'O)).
After these pre1iminaries '"e can pass to the Signori-
ni 's contact prob1em 1dth friction. Its strong /displace-
ment/ formu1ation reads:
Prob1em P 1
Find a disp1acemen t fie1d u = ~ (x) , x E: Q such that

s ij, j + b i =0 , in .Q > /2.53/


sij c~) = aijk1 ek1 (~) /2.54/
u =0 on 5 0 /2.55/
s ij nj = Fi on s1 /2.56/
1.-1J ~ O, sN ~ O, SN1.-1J : 0 , on s2 > /2.57/

-~t G, d IK(sN) ( ~T) ' on 52 /2.58/

Here 2 and [ stand for the body forces and surface traction,
respective1y. The e1asticity tensor {aijk1 ) possesses the
usua1 symmetry properties: aijk1 = ak1 ij = ajikl J moreover

/2.59/

where c is a positive constant '"hile IM~ lR) denotes the spa-


ce of symmetric 3x3 rea1 matrices. The three conditions en-
tering /2.57/ are the Signorini's condition or complementa-
rity ones.
lie assume that
aijkl E. Lc:>O ( S2.) > ( 2r[) E v*, /2.60/
Topics on Unilateral Contact Problems 365

where
v = { y = (vi) 1 y = o /2.61/
ParticuJ.arJ.y, we can assume that
The choice of the space of dispJ.acements V is discussed in
Section 2. 7.
Let us put
Kd = t :!I ;y: <:: v , vN ~ 0 on s 2 ~ , /2.62/
a(~,y) = ~ aijkJ. eij(\!r.)ek1 (;y:)d.x ; ~,;y: E: V, /2.6'J/
;st.

/2.64/

/2.65/

Problem P 1 is now formulated in the variational., or weak


form as
Problem Pd
Find ~ c Kd such that

a(~,;y:-~) + J(~:!) - J(~,~) ~ L(y-~), v;y: ~ Kd. /2.66/


We observe that /2.66/ is vaJ.id regardl.ess of whether the
cJ.osed set K(sN) is convex or not.
The proof of /2.66/ is straightforward.
MuJ.tipJ.ying /2.53/ by ;y:-~, ;y: 6 Kd and performing the in-
tegration by parts we obtain

a(~,;y:-~) + SC-.!T)(;y:T-Sr)dS =
s2
+ 5 sN (vN-~)dS /2.67/
s2
The subdifferential friction J.aw /2.52/ is equivaJ.ent to

j (sN ,;y:T) - j ( sN !:!rr) ~ (-!.T) ( :Y;T-Sr) ~


Taking account of the Signorini's conditions /2.57/ and the
last inequality in /2.67/ we arrive at the implicit varia-
tional. inequaJ.ity =
I.V.I. /2.66/.
Let us study now the stress formuJ.ation of the same ri-
ctiona1 probJ.em. Toward this end we set b !!.-1 and =
366 J.J. Telega

where
Vs = t.i = (tijJ' HsJtij,j +bi=O, in 2 ;tijnj = F. ~

on s 1 ). /2.69/
The function spaces are introduced in Section 2.7.
Let us mu1tip1y the constitutive equation
bijkl.skl.- eij(:!;!) = 0 by (!-.!.), .!.! E: Ks\~) ,
and integrate over 51
We readil.y obtain

0 = b(,!,,i-,!) - 5(tij-sij) njui dS - 5 ~~-sN) 1.1J


so s2
/2.70/

.k
where
b \.!, 1 ,!) = b ijkl. s ij tkl. dx ; !.!. (: Hs /2. 71/
We know that u:O on S Taking account of /2.57/ we infer

I
-
that the integral. 0
over s 2 is non-negative. Thus we can
formul.ate

K:
Probl.em P
Find .!. & (.sN) such that
b (.!, 1 ,!-.!,) /; 0 ,i Ks (.sN). /2.72/
We observe that if ~ = ~ on S 0 then /2.70/ yiel.ds

b(!,,,i-,!)- J(tijnj-sijnj)u idS) 0, -,ifKslsN') /2.73/


0 1
so
The inequal.ity /2.72/ or /2.73/ is the principl.e of virtual.
stresses for the Signorini probl.em with friction.
It is worth not~c:i.ng that /2.72/ is the quasi-variational.
inequal.ity =
Q.V.I., since the set K (~)of statical.l.y ad-
missible stresses depends on the sol.ti~n .!.
2.,5. Mosco-Capuzzo Dol.cetta-~iatzeu theory of duality for
impl.ioit variational. probl.ems
Mechanicians are usuall.y convinced that for variation-
al. probl.ems l.ike Pd and Ps no dual. formul.ation is avail.abl.e.
Such statement is fal.se. To prove it, I shall. empl.oy the so
oal.l.ed Mosco-Capuzzo Dol.cetta-Matze~ theory of dual.ity for
I.V.Ps. For short, here I call. it M-CD-M theory of duality.
Topics on Unilateral Contact Problems 367

In the present section the general theory is presented,sin-


ce the original paper [82] avoids important details and is
restricted to reflexive Banach spaces.
Let V and v* be locally convex topological vector spa-
ces = LCTVS put in separating duality by a bilinear form
<, v* x V ~ R Thus for each v E> V, v#O, there exists
'> : such
v~~'e: ~ that {vt ,v> #O and :for each v* c;. v* , v* #O,
there exists v 6 V such that ( v* ,v) #O. Let Y be another
LCTVS and
(H1) I A:v---.r a linear and contli~uous operator.
The implicit variational problem to be studied invol-
ves functionals 'f and g such that
w-. -.f(Av,w) is, for every vE:V, a convex l.s.c.
:function on V not identically equal to +oo.
w-+ g (v, w) is, for every v E:- V , a real-valued convex
function which is continuous for w v =
w__.. g(v ,w) has, :for every v (:- V , the Gateaux deriva-
tive lll'ith respect to the second variable D2 g(v,w) at
(n4} w::v, such that for every w*'f:v* the set
tv ~ V I D2 g (v, v) = w* 13
Contains at most one element (D 2 g)- (w~).
I recall that the aforementioned Gateaux derivative is de:fi-
ned as :follOlvS
<n 2g(v,,v) ,z) = 1\_.o+
lim g(v,lt+).. ~ - g(v,w) ,z E:V, /2.74/

The primal I.V.P. reads:

(P-. J find u "V such that


) ~ (Au,u) + g(.u,u) ~'f'(Au,,_.) + g(u,w), we V, /2.75/

The F enchel conjugate of 'f (Av, ) is defined by

'f*(Av,w*) = supt<w*,,.,.)- f(Av,,_.)I,.,.E:v),w*E:V~ /2.76/


The functional g*(v,w*) is defined similarly.
The subdi:f:ferential of the convex function g(v,.) is

-a 2 g\v,w-) = \. w~ E: v* 1 g(v,z)- g(v,w)-q<w*,z-,... >,


z E: V ) I 2. 77I
Now we :formulate the dual problem II' as :follOlfS
find u E: V , u~ E: v* such that
(P) - u* E: (l 2 g(..u,u) ,
~it(Au,u*)- (u* ,u) ~ ~* (A.u,w*)- ( w* ,u), w* ~v*
The problemsP and P~ are interrelated by
368 J.J. Telega

Theorem 2.2. Let theasswr.tti.ons (Hl)-(H 1 ) be satisfied. If u


is a so~ution of P, then there ex sts u~v* such that
(u,u~) solves the dua~ problem P*. Conversely, if (u,u*)
solves P~ then u is a solution of the problem P. Moreover,
the extremality condition

-, \Au, u \1
.or + ~o*r
,- \Au 1 u't'i =,/ u * , u '1 = /2.78/

is satisfied.
Before proving this theorem I first recall some indis-
pensab~e notions of the convex analysis [97,131 1 157,158,22~.
By r (
V) we denote the set of _proper ~ower semi-con-
t~uous coftvex functions on v, _or 1' 0 (V): f'lf':V-+Ru +{..o), f
f"/i + o 0 , - convex and l.s.c.J. Now several useful lemmas
are formu~ated.
Lemma 2. 1 Let f' E: r
0 ( v) The following properties are
equiva~ent
/i/ f(u) + f*(u*) = (u*,u) ,
/ii/ u* E "a f'(.u) ,
/iii/ u E: j):f* (.u*)
Lemma 2.2. Let f 1 ,f 2 t r
0 (v). Suppose that there exists
a point v 0 V such that f 1 and f 2 take finite values at
this point and one of them is continuous at v 0
Then
'd(f' 1 +f 2 )(v) =
of 1 (v)+ df' 2 (v), vf;V. /2.79/

Lemma 2.3. Let f':V.-......R = R u(- co~ u(+ cO~


An element u ~V satisfies
f' lu) = min f'( v) , /with (u) finite/
Vf:.V
if' and only if' 0 E: o f(u)
Definition 2.1. Let f 1 and f' 2 be tlro unctions, foe : V.-.R.
The function
f(v):: inf'tf' 1 (v 1)+ f 2 (v 2)jv 1 +v 2 : v, V.,C ; V):

= infff 1 (v 1) + f 2 (v-v 1)) v 1 e:v) /2.80/

is ca~led the inf-convo~ution /inf'imal convo~ution/ and


will be denoted by f f V f =
Some properties of' ~f-cSnvolution are now ~isted.
/a/ If f 1 and f 2 are convex, then f 1V f' 2 is convex too.
/b/ ( f' 1 " f' 2)*' = *
f 1 + f' 2 ' *
provided that foe.= +~.
/c/ Let :f 1 ,:r 2 E. r0 (V) Suppose that there exists vit E: yit
such that f~ (v>) , f'~ l v*J are finite and one of' them is con-
tinuous at v*' Then f' 1V f 2 e r 0 (v) Moreover, the inf-con-
volution is exact.
Topics on Unilateral Contact Problems 369

Definition 2.2. lie say that ~ 1 v f 2 is exact if for each


v 6 V the infimum in the definition of inf-convolution is
attained, i.e. for each v E: V
(r 1 v r 2]
lv) =
min(f 1 lv 1) + r 2 (v) v 1+v 2 I = v}
Thus there exist v 1 , v 2 t: V, such that v 1 +V 2
I , .
= v and
(_f 1 v f 2) \..v) = f 1 ( v 1) + f 2 \.. v 2)
/d./ Let f 1 ,f 2 :V--+R, fo<>~ +cO If the inf-convolution
=
r 1 v f 2 is exact at v v 1 +v2 then one has
d(f 1 vf 2') (v) = df 1 (v~n af 2 (v 2~
/e/ Under the assumptions of Lemma 2.2 one has
(f1+f21\* = f1* v f2* ,
and the inf-convolution is exact.
Proof of Th.2.2. Suppose that u solves the primal problem P.
Then
f(Au,w) + g(u,w)- (f(Au,u)+ g(u,u~{o, w-u), w(:-V
Hence, by using (H 3) and Lemmas 2.2 and 2.3
0 f. () 2 ('-P(Au,u)+ g(u,u)) =d 2 'f(Au,u)+ ) 2 g(u,u),
where 0 '= v*. Lemma , 2.1 and the properties of the inf-convo-
lution result in
0 t d 2 (f +g) (Au,u)~ )u' "d 2 ( f +gJ (Au,o) =
= "0 2 ('-P*vg*")(Au,o) = o 2 'f*(Au,u*)n"d 2/(Au,-u*),
since 0 =
u~ + uZ , or u~ u~ -u~ = =
Thus we can write
u E 'd 2f*" (Au,u*) and u E. d 2 g"f. ( u,-u*).,
or, equivalently
u*' ~ d 2 'f (Au,u) and /2.81/
By using the property /i/ of Lemma 2.1 we infer

~ (Au,u) + f*(Au,~ =(uti ,u) , /2.82/


g(u,u) + g*(u,u*) =( -u* ,u) /2.83/
Further, we have,see [97,13D
'-f(Au,u) = =
f** (Au,u) supt(w*,u)-~*"(Au,w*)lw*~ v*).
The last relation and /2.82/ yield
370 J.J. Telega

"f*(Au,u*) - (uit ,u)'i*(Au,w*')- (w,.,u) v-w*f:V.


The 1ast inequa1ity combined with /2.81/ 2 exhibit that
(u,u*) so1ves the dua1 prob1em P*.
The second part o the thesis o theorem resu1ts by an
inversion o the above reasoning. Fina1ly, the extrema1ity
relation /2.78/ o1lows immediately from /2.82/ and /2.83/.
Thus the theorem has been proved.
To eliminate the primal variab1e u from the dual pro-
blem P we use the assumption (H 4). Consequently we arrive~~
Theorem 2. 3 Under the assumptions (H 1) -lH 4) an element
u E: V is a so1ution o the primal problem P if and on1y if
u* =-D 2g(u,u) solves the fo1lowing dual problem
ind u* E:: vit such that
(P*) f* (A[(n 2g) - 1 (-u*)] ,d) - ( u*, (n 2g) - 1 (-u*)) ~
~..P*(A[(n 2 g)- 1 (-u*)], ~- <w*,(D 2g)- 1 (-u*)), w*r:vtt /2.84/
Moreover, the extrema1ity re1ation /2.78/ ho1ds.
- **
Bidua1 prob1em P Treating the problem P~ as the
prima1 onethe dua1ity operation can be iterated. In this
manner we sha11 obtain the bidua1 prob1em pM~, provided that
the assumption (H4) remains va1id. Toward this end we set

fA(v* ,w~) = ~*(A[(n 2 g)- 1 (-v'*) ,w*), /2.85/

i(v* ,w*) = - <w'*, (n 2 g) - \ -v*)) . /2.86/


or al1 v~, ,,.~ E:: v*. The bidual problem reads
ind u E. V such that
(p~*) i! ([(n 2i)- 1 (-u)1,u)- ( (n 2 g)- 1 (-u) ,u) ~
~1~ ([(n 2 g')- 1 (-u) ,w)- ((D 2 g)- 1 (-u),w) ,Yw~V
The unctions g and g are interre1ated by
Lemma 2.4. The unction g satisfies (H 4), and for each
v,w E: V one has

(n 2i)- 1 (-w) = - D2 g(w,w) , /2.87/

:t:(f(n2ir 1 (-v)] ,w) = ~ (Av,w ). /2.88/


Proof. The re1ation /2.86/ gives
D 2 g(w*,w~) = -(n 2 g)- 1 (-w*).
Topics on Unilateral Contact Problems 371

Thus for any w~ V there exists a uniquely def~ed w~&V*


such that
D2 g(w*,w*) = -w ,

w* being obviously given by w* -D 2 g (w,w'). =


Hence /2.87/ is proved.
To prove /2.88/ we observe that /2.85/ and /2.87/
yield
'f: ([(n {n-
2
1 (-v)],w) =

= sup\..< w* ,,,)- f*~[(n g)- 1 (-n 2 g - 1 (-v)] ,w*)l w~ E.


2 v*) =
= sup t (w* ,w) - ftAv,w*)l w* E: v*} = flt"*~v,w) =
= 1 (Av,w)
Due to Lemma 2.4, the problem P** can be formulated
as follows
I find u E. V such that
(Pi\~) (Dg(u,u) ,w-u) )/ '-f (Av,u~- (Au,w), WE: V 'f
A simple consequenc e of the above considerati ons is
Theorem 2.4. Suppose that the assumption s {H 1) -(H 4 ) are
satisfiea. Then u tV is a solution of (P) if and only if
u is a solution of (P*~.
Proof. From Th.2.2 we infer that u is a solution of (P) if
ru1.d only if u~ = -Dg(u,u) solves (P*), hence if and only if
-Dg(u~ ,u*) = ,ug)- 1 (-u*) = u
is a solution of the Problem P~
Corollary 2.1. The variationa l problems P and plllf coincide
if and only if
g(v,,,r) = <(D 2g)-\-v) ,v-w) /2.89/

Proof. Suppose that (P) co~cides \dth (Pl(~). Then


g(v,w) -g ~,v) =
(D 2 g(v,v) ''~-v) =
<(n 2 g) - 1 (-v) ,v-w) ,
by virtue of /2.87/.
Now let us assume that /2.89/ holds. Then
g(v,w) - g(v,v) = ((D 2g)- 1 (-v) ,v-w) =
= (D,g(v,v) ,,y..v).
Thus we finally have (P) (P ~ =.. lf!4-)

Remark 2.2. In the original paper by Capuzzo Dolcetta and


Matzel.&. @2] , A =
identity and V is a reflexive Banach space.
We observe that A could even be a nonlinear operator.
Remark 2.J. For ~(v,,,r) independen t of v and g(v,w) =
= <. N (.v) ,w-v) , where N is an injective map, the dual varia-
tional inequality was introduced by Mosco [142], see also
372 ].]. Telega

f101 ,14Jl. In this case the problems P and P* have the fol-
lowing form, respectively
(P) I u E: DlN) : (N C..u) ,v-u) ); ~ (u)- ~ (v), v G V, /2.90/
(r}')j u* E:D(N'): <N'(u*) ,v*-u~)~~*(u*)-1* (_v*), V'vilf:-...l /2.91/
provided that D(N) is the domain of N and
N '= -N-1- :v*--.. -.A.-1 ( -v*) , D {N ') = range ( N), while '- is
convex and l.s.c.
Problem /2.91/ results directly from equivalences
(P)( )t
u (;- DlN) : -N (u) (: "d ~ (u) )( )\_ u*f: D(N ') : -N '(u) E:'d'fj-( u* )~
Mosco (142] has also generalized such theory of duali-
ty to the case when N is a multivalued mapping.
It is worth noticing that Mosco's theory of duality
and more generally, M-CD-M theory, apply to problems invol-
ving non-potential operators for l'fhich extremum principles
are not available. Problems described by potential opera-
tors are certainly easier, see for instance [136].
2.6. Application of l1-CD-H theory of duality to Signorini's
problem with friction
In the present section, by using H-CD-H theory of dua-
lity problems P~ and P: liill be derived. It will also be
shown that =
(P:) (PdJ and (P!*) (Pd). =
Let us pass to the derivation of the problem P~ .ife set
g(u,w) = a(~,~-~) - L(~-~) , /2.92/
~(A~,w) = J(~,~) + ~ (~) , /2.93/
d
where A:V~N(~); Vis defined 21;~2.61/.
Thus space Y coincides with H (S 2) in the general ca-
se or is equal to L 2 (s 2), see the next section.
The assumptions (H 1) -(H 4 ) being obviously verified we pass
to proving (H~ Let us first find conjugate functionals.
The following elementary lemma is useful for the derivation
of 'f*'(Av,~.
Lemma 2.5; If V is a reflexive Banach space and C c V a clo-
sedconvex cone such that 0 E: c, then

(rc)~ ( vlf) = Ic* ( v~) , /2.94/


where c* is the polar cone of c, that is

c* ={.v*l v*E: v*,v*~ o),


Here v* ~ 0 means that <v* ,v) ~ 0 , v fC.
Topics on Unilateral Contact Problems 373

f o, if __ ,
-w* = Bv-L
= a(:!,:!) L'v) + /2.105/
"- L +o() , otherwise
where !: =(b ,[).
The Giteaux derivative D2 g expresses as follows
(D 2g(:!t:!!)h) = a(:!t!:!) - L(h)= (By-!: 1!:!). /2.106/
lienee l'le have

and
D 2 g(~ 1 ~) = B~-!: , or B~ = !: + D 2 g(~,~), /2.107/

~ = GB~ = G(~,[) + GD 2 g(~,~) /2.108/


Comparing /2.104/ with /2.108/ we iner that
/2.109/

Hence we can write (n 2g) - 1 (-~*) = ~ = ~ = G(la_) + S, , and the


property (H 4) is satisfied.
Taking account of /2.84/ and /2.109/ l..re finally formulate
the dual problem of (Pd) as
Problem Pd:
-~)xc(sN) such that
Find.!.. =(sN,_!T)f= (
<!-~ 1 G(~ +~))1 0, ,!= (~!T}E:(-~)xC(~), /2.110/
or
<~-sN' [Gl,L)]N + ~~ +<!T-,!T'[G(~:)]T+ ~)1 0 /2.111/
-~f. -~ , -iT E- c(~).
The extremality condition /2.?8/ gives
J(~,~) = -(,!T'~T)- (sN,~l = -a~,~)+ L(~). /2.112/
Hence
J(~t~) = -<.!T~) ' /2.113/
and
a(~t~)- L(~)-(.!Tt~) -(~ 1 ~) = 0 /2.114/
Obviously, in /2.110/- /2.114/ the duality pairings(.,.)
are defined over the part S of the boundary. Further, it
is worth noticing that /2.1~0/, or equivalently /2.111/,can-
not be uncoupled into tlro inequalities for an independent
determination of the normal sN and tangential ~ stresses
over the contact surface s 2 The coupling is implied by
,!Tf: c(~), since /2.110/ is the quasi-variational in-
equality posed on s 2 only. Such Q.V.I. is useful for the
determination of contact pressure and tangential /frictio-
nal/ stresses provided that the Green's matrix [Gik(x,yU ,
or the Green's operator G is known. Unfortunately, this is
not generally the case and approximatiorl procedures are
most1y indispensable.
374 J.J. Telega

To apply Lemma 2.5 we observe that

where
~ = tz:vNjz~ o, z =(!.!)1 5 , ,!ev) /2.96/
2
and 'l is the trace opera tor; wN= N (.!.) = C't .!,) N t
We can assume that VN=H 1/2(5 21\ If sNE: H-1/2 , then (sN,vN)
is to be under~tood in the sense of duality. On the other
hand, if ~'= L (5 2} then

(sN,vN l = J sNvNdS '


52
in the sense of Lebesque.
Having introduced indispensable preliminaries, we find

~ (Av, w*) = sup~< w~ ,w) - J (:!.!!) -~ (.!)j:! E: V~


d
=
= su~<v{,"'N)- ~("'N)] + [ (.!!~!!T)-
- 5J j ( SN ( :! ) , .!!~ d5JI (t .!!) I5 2 = ( WN '.!!T) I 5 , .!! (: v)
2
2
Taking account of /2.64/, we conclude that
'f* (A:!,!!") = ~ (w;) + Ic (sN (:!)) ( ~) ' /2.97/
where
~ = t z* ~ V~ j (z* ,z) v ~v ~ 0, Vz~)
N N
while
C (sN(:!)) =liT .!.T l =\_ i :!!)T ' on 5 2 ( iT'.!!T) ~J(:!t.!!)}
/2.99/
The last relation follows directly from Corollary 13.2.1
of Rockafellar [157), which for l.s.c. functions reads:
"Let f be any positively homogeneous l.s.c. convex function
which is not identically +OO Then f is the support function
of a certain closed convex set c, namely
=
C {v!t J vv, (v, v) ~ f(v))"
Although Rockafellar considers the finite-dimensional case
only, yet the above corollary readi1y extends to infinite-
dimensional case, suchas studied in the preceding section.
The closed and convex set C \~) is the global one. If
Topics on Unilateral Contact Problems 375

sNf:: L 2 (s 2) then C (sN) ::x)= K(~ (x)) , almost everywhere on s 2


The relation /2.97/ yields
*
~ (A:!,1!*) =
l
0, if ~ E; ~ and !!f E- C ( ~ l v~ ,
~ + oo , otherwise
/.2.100/

To find an explicit il- form of g*(v,w*\\- I


, we introduce the
linear operator B,B:V-+V , defined as follows
<B:!t!!) = J
s ij (:!) eij (:!) dx = a(Y!!); Yt!! E: v. /2.101/
52
Hence
<Bv,v)= , c) 0 , /2.102/
provided that meas S 0 ~ 0 ; c is a constant.
As a norm on V we can take any of equivalent norms, for in-
stance [18, 43]
JSl eij (:!)eij (Y.) dx
The operator B is continuous, since
<B:!,:!!) ~ c 111 :!llvll:!!l~ ; Yt!! EV.
Hence we infer that an inverse operator G:B- 1 exists,
G:v*~v. This operator is obviously continuous and v* -el-
liptic, that is
( :!* ,G'!:.*) ~ c2U '!:.lfll V* ~ t'll yif > '

<'!:.* ,G'!:.~) )1
c 3u,!trv*
where c 2 and c 3 are positive constants. The linear operator
G is nothing else as the Green's operator for the mixed
boundary value problem of the elastostatics.
Let us put i =(si .n.}, on s 2 Now we can write, see
[146 ,p.1J5] J J

uk(x) = [G(~,:.,1.)1
k = ( bi(y)Gik(x,y)d S(Y) +
~ /2.10.3/
+J
51
Fi{Y)Gik(x,y) dS(y) + ( L i(Y)Gik(x,y)dS (Y).
2
Due to the linearity gf G we can write
u = G(~,E) + G(~) = ~ + G(~) , /2.104/
where "
u = G(~t[)
Taking account of /2.92/ and /2.101/ we obtain
t/(y,:!!*) = supt<!!* t:!!) - a(y,:!!-y)+ L(:!!-Y)) ~ f V = 1
= a(:!Y) - L(y) + sup t <~* :!!) - a(:!:,:!!) + L(~l :!! E: v)=
376 J.J. Telega

The proof that (P~~) = (P ~ is straightfor1..rard and re-


sults from Corollary 2.1 and Lemma 2.4. In our case we have
g(:!,_!!) = a(:!,!!-:!) - L( !!-Y) = By-L,!!-Y) = <
= (D 2 g(y,:!),!!-:!) = <(n 2g.)- 1 ( -y),y-_!!) ;simihrly P.:~~'~~'::~.
Proof that (P:) -;:: (Pa). T~ward ~his end we set
g 1 ( !.!) = b(!.,! -!.) ; !.! e{L (Q) 1 , 12.1151
f 1 ( SN , ! ) = IKs ( SN) ( 1) I 2 11 6I
Hence we obtain
g*1 (s,t*)=
,-- sup r<t~t)- b(s,t-s'' =
-t 't - - ---J)
b(!_,!.) + -;tp { r ( t1j-bijklsklJ tijdx
- b
= =

= ( b(.!_,.!_) , if t~j = bijkl skl ,

Thus t*
l+ oD otherwise
,
is the strain tensor.
12.1171

The Giteaux derivative is


<n2g1 (!.,.:) ,!!) = b(!.!!).
Hence
D2gl !.!) = ( bijkl sklJ ' 1 2 1181
and thus _1
(n 2 g 1) (-f) = ! , = - a s . 12.1191
Let us pass to the derivation of the conjugate functio-
nal f~ (sN,t*). 1ve may take i* = ~(!!), .!! E; V , where V is
defined by 12.611. Then one has
f~(sN'~(.!!)) = supt(.:,~(:!!)) - '{> 1 (sN'!)I ! f H$ =

= sup{f tijeijl:!!)dxl i~Ks(~)~ =


Q
=sup\_-) tij,jwidx + ~ tijnjwidsj ! f; Ksl~}) =
Sl
= Ll!!)+ sup ( ~wNdS+ sup { ( ,!T :!!TdS l.!rr (: C ( SNJ) =
tN~ 0 J J
--l L(~)
+ 0o
+ J
s2
(sN ,~)
-~
otherwise
, if'
s2
w...T ); 0
l'l
on s 2 > /1.110/

where
Topics on Unilateral Contact Problems 377

J(sN,:!!T) =sup t) iT!:TdSl!T G C(sNJJ /2.121/


82
In /2.121/ the roles of supremum and integration may be
intechanged provided that !T"!:T is a measurable function
belonging to a so called decomposable space, for instance
1, - ~ -
to L tS 2 ) L158 ,p.185J.
Taking account of Th.2.3, /2.119/ and /2.120/, the du-
al problem P~ reads
s
,_.., I
find u {: -Kd -
such that
\1" ) '
. s l L\u,
I , ... ) -c -- """,
+ J -~\U) ,~j +

+
Now we set u = -'U and v =
on the form- - -
find ~ ~ Kd such that
{!>!)
-L(~) + J(!!.t!!.) + a(!!.t!!.)t- Lt:Y)+ J(!!.t:Y)+
+ a(:y,!!.), V' y E- Kd , /2.123/
where J(!!_rY) = J(sNL!!.)r -!T) , see also /2.64/. Thus we
see that (P:) = (Pd)
The extremality condition /2.78/ takes now the form
-L(!!.) + J(~r~) = - (!.r!.(!!.)) = -b(!.t!.). /2.124/

where (!.r!.(~)) = J sijeij(~)dx


Summarizing, the~ain resu~ts o this section can be
represented by the following diagram

.( Ps}====[*=l====:;>>( f=d} [*l >( P:}


~
[*)

Fig.17
provided that(~) stands for the duality operation.
378 J.J. Telega

Remark 2.4. Suppose that in /2.111/ ~=!T=O, on s 2 Then


the Q.V.I. /2.111/ simpl.if'i.es to the following variational
inequality def'inea, as previously, on s 2
(P~) 1 f'~nd sN E: _- ~_such ;hat Y.

I'~-~ ,LG(sN)]N + ~)~0, ~f.; -KN /2.125/


The J.atter V.I. has already been derived by Kikuchi [120]
and cal.led the "reciprocal variational inequality". This
author does not employ the lwfosco's theory of' dual.ity, see
Remark 2.2. We observe that /2.125/ f'ol.lows directly from
/2.91/. To solve the problem P~ a numerical procedure has
been proposed in n2o], see also the l.ectures by Hasl.inger
and (91 ,121 1 123]. ~eciprocal f'ormul.ation of the Signorini's
problem with Coulomb's friction and prescribed contact pres-
sure together with an approximation by Ritz-Galerkin method
is studied in f110) Si.mil.ar problem in the dynamical case
i.s investigated iri [162).
Various numerical. methods for solving elastic unilate-
ral contact probl.ems with friction have been proposed in
~1,105,109,111,117,122,123,125,126,128,1.30,138,147,149,
151, 162)
An a alytical approach employing complex variable me-
thod to dynamical. frictional problems without Signorini's
conditions is used in [160).
Geometrically nonl.inear problems are studied in [98
150,159,161] ; the f'rictionl.ess case is considered in [84).
Mechani.cal.l.y important frictional. probl.ems for pl.ates
are investigated in ~2 1 63,113) The sl.iding rule empl.oyed
in (2 1 6.3) is motivated numerical.ly and not physical.ly.
R$Wark 2.5. M-CD-M theory of' dual.ity readily appl.ies to two
el.astic sol.ids S2.~ (ce =1 ,2), cl.amped al.ong ftC. Now we have
1 2 1 2 ] l1) (.2) 0 t1) (.2)
52=52=52 and .!T=.!T=-~ 'ff~ = ~ - '-'N ~o, sN=sN= -sN ,
sN [~B = 0 /on s 2/, provided that the unit normal.~ is ta-
ken as exterior to 52 1 The sliding rul.e takes the form

[ Sr] ~ ") IK"N) (.!T) /2.126/

Let us set.!! =(J1),:!_1;?>) , ~ =u1\~ftZ>), !. ={t1),~) and

a(.!!r:!) : r a(GC)~V,cl !,t~>), L(!.)


d;
= ac:.l L{or:}(!,(.ocJ), /2.127/

and,Cor inatance
:J1> (t>) = J
~1.1
Topics on Unilateral Contact Problems 379

The set of kinematically admissible displacements is


Kd = t :y,: =(t11 ,{l) (; Vl[vN] ~ 0 on s2) > /2.129/

where )
V = V Y.V :Uvl1 1 v~)l v{c()f:V_ v("')= 0 on s~\ /2.1-lO/
f 2 I.,_ - I I - <:4 J - 0 J J

With obvious identification s, the Signorini's problem


with the sliding rule /2.126/ takes on the same form as the
problem Pd,studied previously. I denote it by (Pd 2).
Now we have
;:_l1) = _~} + Gi!) ' 42) = u~)- G2(.6_) ' /2.131/
where
U,(~J
-
=G 0(
(b\.a)
-
t(oCI).
,_
, /no summint
'\;.
.1 /.

The dual problem P~2 reads:

find ~ E:- -~xc(sN) such that /2.132/

(,!-~, G 1( ~) + G/ ~) + g_t1) + fl ))10, t f -~xc(sN)

and

Of particular interest is the case where one of the


domains, say 51 2 , is unbow1ded. For instance, let Sl 2 be
a half-space:
S2. 2 =tx = (xi)ER 3 x 3 <_o)
Further, let S 1 be empty. Then the bilinear form a(1)(.,.)
is not coerciv~, yet the Green operator G2 still exists
D46] for that traction /second/ boundary value problem of
elasticity.
We observe that the numerical procedure proposed by
Kikuchi D20J for frictionless case extends to problems,
where s 1 is empty by introducing of an artificial part of
0
the boundary with prescribed displacements.
380 ].]. Telega

2.7. Existence problems


The existence of a solution for the problem Pd is sol-
ved only partially, even in the case of Coulomb friction
[86,92,95,115,123,130,145,1711. Particularly, Jarusek [115]
studies the case where jL f c 1 (s 2) and suppJL is compact.
Thus the coefficient of friction is equal to zero in the
vicinity of d s 2 , what is not very realistic.
Duvaut [95] introduced the notion of non-local frict-
ion which has been exploited in many papers by Oden and
hiS coworkers, see [55,92,123,138,147,148,149,151,168].
This notion proves also useful from the numerical point of
view.
Cocu and Radoslovescu [87) have proved the following
regularity theorem for the problem Pd in the case of non-
-local Coulomb friction.
Theorem 2.5. Let S2, be of class c3 in all x E s 2 Then for
every open set u, U c 51 C. s 2 , we have

.!:! E [ H 2 (u) ) 3
Existence problems for elastic bodies in the presence
of viscous sliding rules /parabolic problem/, but without
Signorini 's conditions have been studied in (96, 1 33]
Problem Pd 2 for both linear elastic and viscoelastic
bodies is investigated by Boucher [183,279) , see also [192,
220) However, :friction is negl.ected, but Q 2 is a hal.:f
space, see Remark 2.5. The space v 2 is a Barros-Neto space
[278) , equival.ently de:fined as :follows

v2 =b: =(vi) I vi b L6(~) ' vi, j f L 2 ( Q 2J )


Such space is larger then H 1(.2 2) and both spaces coincide
if .Q. 2 is bounded.
It can also be shown that there is a continuous imbedding
of the trace space K 11 2 (R 2 ) = K 1/ 2 ("d @.. 2} of functions of
V 2 into L 4 ( d .Q. 2) Thus we see that the linear form C2)( vi;!>)
makes sense if bit L 6 /5 (.Q. 2} , F i f L 41 3 (s~). Rigid body
motions of the body ~ 1 are also studied by Boucher [183,
279).
Oden and Martins [147) proposed new contact conditions
with a rough surface. I will now show that those conditions
correspond to associated rigid plasticity with hardening
and/or softening.
Topics on Unilateral Contact Problems 381

Let the initial gap beh1een the body and the foundation
r 4 .,
be g ); 0 along S 2 According to the paper L1 7) , the con-
tact conditions are:
/i/ normal interface response
sN = -~t~-g}+n
, 'm
on S
2 /2.136/
/ii/ friction conditions
'-'N ~ g - ) ~T =0
(I.:!TI ~ cT ( u,r-g) :
I o
=>~~I.!TI < cT(~-g)+l- -~~ = T,
.D1-
~)g /2.137/
m.r 3'
o
1.!TI = CT (~-g) + ~ ).qO, ~-T = - A .!T
o
where cn,cT,mn and~ are material parameters, and T is
the tangential velocity of the subgrade with which s 2 comes
in contact; (~-g)+ stands for the positive part of (~-g).
rie observe that Coulomb's friction law is recovered if
m 11=~; then ~=c~cn is the usual coefficient of friction.
Oden-1-Iartins formulae imply that the friction condit-
ion is given by
f(sN'.!T'u;) = l.!TI - fl(u;) I sNI 1 2 13 8/
where p. ( u;) mT m
= cT( ~-g)+ /en ( "'N-g) +n is the coefficient
of :friction, \which depends on the normal displacement ~
In the case considered
on the form
U:=o,
~=0 The sliding rule takes

(-~i' -~T) E: ()~(.,~) (sN,.!T) /2.139/

where

Yt-r = llT
o
- T , and

K( ~ ) = t! =(tN '!T) I f (~ '!T 'u;) ~ 0 )


the density of dissipation produced by normal and
l~ence
tangential /:frictional/ stresses is given by
j(u;,~,~) = supttN(-~)+tT(-~T)j(tN,!TJ E K(u;))=
m ~,
= cn(un-g)+n I~~ + cT(un-g)+ I\!.T-;1 /2.141/
The coefficient of friction fL(u:) may be seen as a speci-
382 J.J. Telega

f'ic case of' )1adakson 1 s f'ormul.a /2.6/.


Static case is studied in [15J] 1 where existence and
uniqueness of' variational /weak/ solutions are proved. H.e-
sul.ts are established either f'or arbitrary external forces
and suf'f'icientl.y smal.l coefficient of' friction or f'or arbi-
trary coef'f'icient of' friction and sufficiently small ex-
ternal. f'orces. For this purpose the implicit function the-
orem is used.
Now I pass to proving that a sol.ution of' the probl.em
Pd exists in the general. case of' a nonl.ocal sliding rule.
Such rules are introduced as f'oll.ows.
lie assume that
V = l!: = (vi) j vi~ H 1 ( &) 1 v = 0 on S 0 ) >
Hs =L~. = (tij) I tijE L2(Sl), tij,j f.L2(Q.)~.
Then the Green f'ormul.a shews that tijnj E- II-1/2,\S) 1 and par-
ticul.arl.y sN (!:!) (: H- 11 2 (s ) This is a source of' mathemati-
cal troubles with the protil.em Pd. One way of' overcoming h
them is to replace sN by more regular normal stresses sN.
For instance, we can take

5w h (1 X-YI SN (Y) dS (y)


s2
The last operation is call.ed mol.lif'ication. The function
~h(r) is of class CoO and has the follOl'iing properties

/i/ W h(r) ) 0 , r <h


/ii/ c.lh (r) =0 , r 9- h

/ii/ 5
f.\Jh(r)dy = S~h(r)dx = 1
r<;.h r<.h
Let us put
Jh(!!!:) = 5 J(s;(!!)YT)ds /2.144/
s2
Instead of' the probl.em Pd we consider the f'ol.l.owing one
f'ind :!;!_ f Kd such that
( phd) a(!!_,y-!!_) + Jh(!!_,y)-Jh(!:!.!:!.)~ L(!:-!!)' y E: Kd
Now let us recall a general existence result [14J], here
formulated f'or ref'l.exive spaces only.
Topics on Unilateral Contact Problems 383

Theorem 2.6. Let V be a reflexive Banach space and


g:C>tC-.R, with r:(y,;::_) ">"/ 0
/a/ C c V is a closed convex suoset.

v C. lie assume that

/b/ For each v (: C, g(v,.) is convex.


/ c/ ~or each ,.,. E: C, G( , w) is \veakly upper s em icon tinuous
/u.s.c./.
/d/ There exists a bounded subset B c V and an element
'''o E: B (\ C such that for all v ~ C\B t.re have g\v,t.r )
0 0. <
Then there exists at least one solution .!::. E C f\ B of the in-
equality
g(u,v) ); 0 v E: C /2.146/
To employ this theorem l:e shall also need the follot...--
ing one, proved in [66J, see al,so [80]
Theorem 2. 7. Let F :9'xR 5 ~ t - . R v { + o<:~) ( R u .(_- oo)) satisfy
the following properties :
/i/ F(.,z,v :e-+nv{+o a) ( nu{-oa)) is measurable for
ev-ery { z, v) E: Jl 5 ~ t ,

almost all X f. e ,
t
/ii/ F(x,.,.) :ll 5 xilt-+I~v +oe)(nu{-c:.o}) is continuous for
/iii/ F(x ,z,.): H t - H v { + oo) (R u L- ..o)) is convex /concave/
for almost all x E. 9 and all z E. H. s
Let z(n),z :R 5 __.,. n v t +CXl) (n v .{_-oo))be measurable function,.f
such that z(n)_. z D.lmost everyt.rhere and let vc.nhv in
[L 1(e)]t as n--.oe. Suppose further that there exists
f E. J}\Er) such that
l'(x,z(n)(x) ,vl.11\x)) ~ f(x) , F (x,z(x) ,v(x~~ f(x) ,
(} (x ,z (n)(x) ,vtn\xl) ~ f(X), F(x,z (_x) ,v \x)) ~ f(x),
for all n and almost all x ~ e-. Then

SF(x,z(x) ,v(_x)) dx ~ 1 iii!J


n-+o<>
1:' (x,z(n\x) ,v ln\x)) dx ,/2.147/
e Er

( )(f'(x,z(x),v lx))dx); lim


n- .,p
5 F(x,ztn)(x),J-1l(x))dx). /2.148/
6 e-
The uext lemma has been proved in 92l
L .J
r
Lemma 2. 6. Let u - u t.reakly in V, where V is defined by
/2.142/. Then -n -
/a/ sN(.!::.rr-- sN(!:.:J in H- 1/ 2 (s) '"eakly, that is
.(s}; (.:!n) , z) --.(sN (.:!), z) z f.: H1/2( 5 ).
384 ].]. Telega

/b/ in strongly.

Let us set C = Kd , and

g(:!:,~) = a(:!:,~-:!:) + Jh(:!:,1!) - Jh(:!:,:!:)- L(~-:!:) /2.149/

The function i' involved in Th.2. 7 is d. As ;;re irn.ow, the


function d(x,z,.) is convex. For a fixed :!:,d(x,z,:!:) is
trivially concave 1.dth respect to v.
Let us assume that d(x,. ,. ) is continuous lvhile d(. ,z,v)
measurab_le.The assumpion concerning continuity can be weak-
ened [soj. In the case of Coulomb's friction continuity is
obvious while measurability is satisfied if" )A- E. L00(s,,).
Our next assumption is /c-rolYth condition/ 4

2- E.
Jh ( :!:,~ ) ~ c (~)II :!: II v ' t '> 0 , :!: ~ v, /2.150/
provided that ~ t V is held f'ixcd. In the case of Coulomb
friction ~ = 1
The function g(y,. ) defined by /2.149/ is obviously
convex.
The functional g(. ,w) is weakly u.s.c. for each '' : Kd.
To corroborate this statement we observe that a(:!:,y) Is
weakly l.s.c., hence -a(y,y) is weakly u.s.c. Further, the
functionals a(.,y) and L(.) are linear and continuous on v.
To prove that Jh(y_,y) is weakly l.s.c., ;re emply Th.2.7
w1d Lemma 2.6. The latter liuplies that at least for a sub-
sequ~ncc, still denoted by{~~' s~(;;!.11)_. s~(!::_) almost eve-
ry,vhere. Since Jh(:!:,~) represents the 1.vork of the frictional
stresses, hence it is reasonable to assume that d(.x,z,y)71 f(X)
x E s 2 , or even that f (x):: o. Thus >ve see that Jh(;;!, 1 ;;!,) is
l.s c., and -Jh(;;!,,;;!,) u.s.c. Now we need the upper semicon-
tinuity of Jh(.,:!!), which results from our assumptions and
the "concavity part" of Th.2.7.
Thus the condition /c/ of Th.2.6 is satisfied. It is >vorth
noticing that, apart from other rather classical properties,
all we need to prove the condition /c/ is the sequentially
weak continuity of Jh(,~)
To prove the condition / d/, let us take :!!a E Kd and de-
fine B by ,. 1 1
B = \_:!:f;Kdl g(y,:!!o) ), 0 J
The set B is bounded. In order to prove i t let us suppose
that D is unbounded. Then a sequence { y C Kd 3
exists,
II ~II V--+ oO as n---. ao , such tha
Topics on Unilateral Contact Problems 385

/2.151/
for all n.
Consider the functional -g(v ,w). By using Korn's in-
-n -o
equality, /2.150/ ~~d knowing that Jh(~,~ is bounded from
below we arrive at
-g(~~) '>t c111 ~II; - c(~) ll Ynll ~-E - c2\l 'Ln II v + c3 '
where c,c 1 ,c 2 ~~d c 3 are const~~ts and c 1 ~o. Hence,
-g(v ,w ) ----+oo or g(v
-n-o ,w ) --. - oo ; a contradiction with
-n-o
/2.151/.
Thus Th. 2. 6 yields finally the existence of .!! E: Kd sol-
ving the nonlocal frictional problem P~.
Such a solution is unique under additional assumptions,
see [86,Th.J.1).
Let R be a real Hilbert space, C c H a non-empty closed
convex subset and let V:C-+H be an operator, such that
(B(u) - Btv) ,u-v) ~ m llu-v 112, u,v,w E:C /2.152/
(B lu) - B (v) ,w) ~ H II u-v 1111 l i Jl ,
where m and M are positive constants, (,. ,) stands for the
inner product on H and II 11 2 =(,., ),
In our case the operator B is defined by /2.101/ and
then /2.152/ certainly holds.
Theorem 2.8. Let us assume that /2.152/ is satisfied and let
J: CXC -+R u {. + oo} be a proper functional such that J( v,.) is
convex and l.s.c. on C. Let L be a linear and continuous
functional. on li and suppose that J sati.s:fi.es the :fol.l.owi.ng
inequal.ity
IJ(u 1 ,v1.)+J (u 2 ,v 1) -J (u 1 ,v 1) -J(u 2 ,v2 )1 ~ c1Ju 1-u 2 1111v 1-v 2 11
u 1 ,u 2 ,v 1 ,v 2 6 C, /2.153/
where c > 0 is a constant, m~ c.

(P )I
Then there exists a unique solution of the foll.owing problem

B
find u f C such that
(n(u) ,v-u)+J(u,v)-J(u,u)~ L(v-u) v(: C. /2.154/
h
To apply Th.2.8 to the Problem Pd, I assume that the functio-
nal Jh satisfies /2.153/ and the requirements of this theo-
h
rem. Then .!! E: Kd solving the problem P d is unique.
In the case of Coulomb's friction uniqueness is implied
by a 11 smallness" of p. [86 ,92), provided that JA-'= LoO(S 2 ) and
386 J.J. Telega

0 ~ /"'- (x :1 ~ )1. 1 Cocu ( 82] proves that in such a case


I Jh(~1 ~2) + Jh(~2 :!:1) - '\(~1 :!:1) - JhU!2 Y2) I~
~ 0 0 IIJA-11 Loe(s 2)1J ~ 1 -~2 1111 :!: 1-:!:2 11 /2.1.55/
<
Taking ,}J. 1 me 0 we conclude that /2.1.53/ and m-, c are sa-
tisfied if' c = c f- 1
0

J INELASTIC T.i'NILATERAL CONTACT PHOBLEl-lS


As one could expect, mostly studied are unilateral
elastic problems. For instance, mathematically rigorous
study of unilateral contact problems for plastic bodies ru1d
structures is at the very begining.
In this chapter I will summarize the variational ap-
proach to the study of unilateral proble~s of' viscoelasti-
city and plasticity. Some new results are also presented as
well as open problems.
3.1. Im act of a viscoelastic rod l-.rith an obstacle
In the papers [1 9,190 an impact of a viscoelastic
rod with rigid or deformable /linear/ obstacles has been
studied. Omitting the dissipation and shock l\"tt.Ves duri:n.g a
shock between the rod and the obstacle, the problem is in
fact considered as the dynamic Signorini's problem without
friction. In the case of a rigid obstacle the variational
:formula ti.on results i.n a vari.a tional. equation ~ a V.I. ,
which are coupled. It seems that this dynamic situation for
the rod is typical for many unilateral problems involving
the Signorini's conditions. Surprisingly, but this simple
conclusion seems to have been overlooked in the relevant
literature. Formally the same problem arises in the quasi-
-static case involving the Signorini's conditions and the
flow lal-.r type sliding rule /2.22/, see Section J. 2.
Let us pass to the study of the aforementioned rod pro-
blem. Suppose that in the undeformed configuration the rod
is identified with S2. =(0,1), 170, see Fig.18. The density
is denoted by 9 , while s(x,t) stands for the normal stress
in a section x at time t. Viscoelastic co:tstitutive equation
has the form
s a= + bE a") 0 , b"? 0 1 /J.1/
= ,x=
li'here E (u) u d u/() x, E = "d /?J t The case b:O cor-
responds to purely elastic material. Let us put
U(t) = u(O,t), R(t) =
-s(O,t), tf[O,T), T>O /J.2/
Thus U denotes the displacement of the lo,,.er section of the
rod liThile R is the reaction of the obstacle. The Sigaorini 's
conditions take on the form
Topics on Unilateral Contact Problems 387

l+u(l,t)

Fig.18

u ); 0 , R>--- 0 , /J.J/
ru1d are equivalent to
u )1 o , R (v-u)'>,o lf t K . /J.4/
The convex set K of kinematically admissible displace-
mcnts for the lower section is defined by
t ={w(t) = w(o,t)J lf(t)~ o , tE: (o,Tn.
Obviously, ii(t) stands for the virtual displacement of the
lo;ver section of the rod and l'lill be precised in the sequel.
From the vie;.-point of the existence theory and numerUc\1.
calculations it is convenient to replace the rigid obstacle
by a deformable one /penalty method/.
Elastic obstacle. Denoti~ b U+ ,u- the positive and nega-
gative parts of -u (u =
u+ -u-} ' l'l"e have
if U)/ 0
~ -kU,
0 , ,
R. = kU- <~) k = /J.6/
if U ~ 0 >
where k stands for the rigidity of the obstacle. The rigid
case is recovered in the limit as k - oO
Hore general obstacle behaviour can be described as
follows. For instance, we may take a nonlinear dependence
R = R(u-) 1 /J.7/
where I{ is a continuous and non-decreasing function, not
necessarily differentiable, see Fig.19. In such a case the
function
388 J.J. Telega
1.)-

r (u-) =jli~) dZ /3.8/


R
0
is a well defined
l.s.c. proper con-
vex f~~ction, see
[157,Th.24.2]. The
f'unction r can be
u defined on a 1 ,since
for U)1 0 r(U)=o
Then r is subdfffe-
rentiable on R
tl'ithin the frame-
l\'Ork of covexi ty
most general respon-
se of a subgrade is
Fig.19 described by maxi-
mally monotone graphs [2.57] or equivalently complete non-
-decreasing curves [1.57,p.2J2].
Viscoelastic obstacle. If the behaviour of the obstacle is
linear then
R = kU- + ~ u--,
where V is the coefficient of viscosity and U-= U
..!.._ /J.9/
Similarly as previously, both elastic and viscous be-
haviour can be nonlinear.
Let us pass to the variational formulation. The equat-
ion of motion is
2 2 ..
,x + f = g 'd u/'d t = ~ u ,
s
where f stands for density of body forces. Assume that~:= 1
while the initial conditions are
u(o) = u 0 , ~ (o) = u 1 /3.11/
Now a function v(x,t) is treated as a vector function
v: (o,T)- B, '"here B is a Banach space.
A virtual velocity vis such that v(T):O. Multiplying /J.10/
by v and integrating over (0,1) we get
(li,v)+(s,t(v)) =(f,v)+ R(t) v(t), v(t)= v(o,t), /3.12/
where 1
(v '''~) = ~ vwdx
0
Let us set
1
a(u,v) = 5 a E' ( u ) f. l v) dx ,
0
Topics on Unilateral Contact Problems 389

1
=Sb f: ( u) : l v) dx
0
Integrating /3.4/ an.d /3.12/ over the time interval, next
taking account of /J.1/ and /3.11/ ,.,.e readily obtain the
variational formulation of the impact problem considered.
Problem P
Find a function u(t), tE:.(O,T}, such that
T
~ (-(il\t),v(t) +a(u~t) ,v(t~ +b(~(t) ,v(t~) dt-(u 1 ,v(o)) =

Ib T . . .
= SHf(t),vtt)) +R(t)V\t))dt , Vv, vlT)=O /J.15/
0
U(t) = ulo,t) 'q 0 , /3.16/
T
~ Il ( t) ( w(t) - u ( t )) d t >/ 0 Vii E K ,
0
u(O) = u 0
As one can see, Problem P consists of the variational equat-
ion /J.15/ and the V.I. /3.17/. The latter involves only
displacements whereas the former both displacements and ve-
locities.
Substituting /39/ into /3.12/ we obtain the penalized prob-
lem, provided that k') 0 and ll )/ 0 are given.
Problem P 1
Find a function u(t), t f [o,T], such that
('U(.t) ,v) + a(u(t) ,v) + b(u\t) ,v) = (f (t) ,v) +
+ kU-(t)V + yU-(t)V, Vv,
and which satisfies /3.11/.
Let us pass to discussion of results concerning exis-
tence and uniqueness. The case )I =0 is studied.
Theorem J.1. Let k> o, T> o, u E: H 1(0,l), u 1 ~ L 2 (o,l)
2 ') 0
and f : L ( 0, T; L"'" ( 0, 1)) Then there exists a unique function
uk ,.,.hich satisfies /3.19/ ()I =0) for almost every t~(o,T)
and every v f H 1 ( 0, 1) and such that
ukE L<:oe(o,T;H\o,l )) , uk~ LoO(o,T;L 2 (o,l))nL 2 (o,T;H 1 (o,l))
/J.20/
390 J.J. Telega

Sketch of the proof. The notation uk for a fm<ction solving


/3.19/ lvith k$ 0 is used to stress that u depends on k..
To prove uniqueness we use Gronwall's lemma f 25 7 ,p. 1561:
1 1 1.. .1
"Let mE-L ( 0 ,T;R ) be a 1'unction such that \~~ 0. a. e. on
(O,T) and let a be a constru1t.} o. Let l~E:C([O,TJ)be a
function such that t 1

w(t) ~ a+ ~ m(s) l~(s) ds for each t E: Lo,T].


Then 0
t
~ m(s) ds
_,o
w(t) = ae II
/3.22/
Hore familar form of this lemma is for m being a. constant,
see [283]
It is known that for given ) > 0 , there exists o<: '). 0
(~ ') 0) such that [ 18]
a Cv, v) + f. I v 12 ~ oc\1 v 11 2 , v E: H 1 (. 0, 1)
/3.23/
(b(v,v) + Alv1 2 ~ Jllv11 2 , v~H 1 (0,l).
Here II II = II II H 1 ( 0 , 1 ) , I I= I I 1,2 (0 , 1 )
Uniqueness. Suppose that uk ( 1) and uk ( 2 ) are two solutions.
lie put u:uk{ 1) -uk( 2 ) for A') 0 given a.."ld fixed; we set
z"' ( t ) =
e -~t ~(ce) z l)
t =
e - ) t u (t ) , o( 1 , 2, =
For the sake of simplicity we take k=1
\fri ting /3.19/ lY :0) for uk t 1 ) and uk( 2 ), next subtracting
and taking v:z we arrive at

t :ttlz\ 2 + )1 2 lzl 2 +a(z)+~b(z))+2).1 zl 2 +btz)~IZI12:j,


since lz~ -z~lf: jz 1 -z 2 1; here a(z)= a(z,z), b(z)= b(z,z),
Z (t)= z(o,t). Knowing that z(O)= o, z
(0)=0, using/J.2V and
Hl:Slders inequality l~e finally obtain

oel\ z ll
2
~ c J0t \1 z II 2 ds ,
where c is a positive constant. Applyincr Gronwall's lemma
we conclude that z:O. Hence, the solution is unique.
Existence is proved by using the Faedo-Galerkin method anu
compacity arguments. For details, the reader should refer
to (190).
The results for the original problem P are the follow-
ing.
Topics on Unilateral Contact Problems 391

Theorem ].2. Let the data on T,u ,u 1 and t be as previously.


Further, let U =u.(o,o)> 0. Then, 0 as k-.oa we have, at
least for a suBsequence
uk__.. u '"eak star in Lo<l (O,T;H 1 (o,l)) /3.24/
U.k.---. u ;,eak star in L00 ( O,l';L2 (o,i'l) and

\veakly in L 2 (o,T;H 1 (o,l)}, /3.25/


c k-.u uniormly on [o,T], /J.26/
Ilk= kU~ - R in the iveak topology of H ( [0, T]). /3.27/
Moreover, the following relations are satisfied
,.........

) t-(~(t),vlt)) + a{u(t) ,v(t)) + b(u(t) ,v(t))) -(u 1 ,v(o)) =


0
T
= ) (f (t) ,v (t)) dt +(R,V), -v~ L 2 (o,T;H 1 (o,l)J\ ldth-
0 v:L 1(o,T;L 2 (o,l)), VtC 0 ((o,TJ),v(T}~O, /3.28/
u(O)= u 0 , /3.29/
u ~ o, on (o,T) , /'J.30/
<R,W-U)~ O, "VWEC 0 ([o,TJ),
W~O on (o,T). /J.31/
A detailed proof is given in [190] f'or )I :0. It starts from
taking v = etk (._t) in /3.19/, and next integration over ( 0 ,T)
is performed. A priori estimates result in /3.24/-/3.26/.
The second a priori estimates are deduced by taking in
/3.19/ f'or the virtual motion a rigid one, i.e. v(x)=V~o,
V being a constant. A:fter integration over (O,T) we arrive
at the estimation
T
5
kU~(t)dt~ const., or kU-k is bounded
0 in L 1 (o,T). /3.32/
llence /J.27/ results. In :fact at this moment Banach-rAlaoglu.
theorem is used. It states that if X is a Banach space then
the w1i t ball in X~ is compact in the topology (S (x* ,X).
To prove /3.31/ one proceeds in the following way. Let
Wt C0 ( [ 0, T]) The function
h fy) ly2/2 if y~ 0 , =
" 0 if Y') 0 t
is obviously convex. Subdifferentiability furnishes
tuw-) 2 -(u~) 2 ]~-u~(W-UkJ on[o,T].
1

Taking nOlv 'd ~ 0 on [ 0 ,T], tve obtain


392 J.J. Telega

Hence
-( W- ut~'\1 ~ 0 ,
Uk on ,
[ o, T .J ,
T
S kU~ (w-uk)ctt >; o
0
If k--+-oe, then by taking account of' /J. 32/~re arrive at
/3.31/.
Remark 3.1. Do [189] has formulated a theorem similar to
Th.J.2 above, yet for )I>/ 0. During the p<:o.ssage with k to
infinity Y is held fixed. Obviously, then u depends on v
i.e. u=uJ'. According to [189] such uv also satisfies /J.24/-
-/J.J1/. This result, given by Do ldthout proof, is not
clear.
\:'e also observe that u solving the problem P is not ne-
cessarily U.."'1.ique, or at least uniqueness remains, as far as
I ~"'1.ow, an open problem.
3.2. General variational formulations for a class of inelas-
tic bodies
The model problem P of the previous section can be ge-
neralized in such a way that the flow la\\' type sliding rule
is readily included.
Let g. c. n 3 be a sufficiently regular domain while Q
represents an undeformed configuration of a not necessarily
elastic body. The equations of motion are
2
o ui
9 (l t2 = sij,j + bi ' in 5). /J.JJ/
Now all functions may depends on xEQ. and t ~(o,T),t -time.
The initial conditions are
.!!(x,o) = !!o(x) , ~(x,o) =!:!. (:~). 13 341
1
In the quasi-static case the dynamic term in /J.JJ/ and
/3.34/ 2 must be omitted. Suppose that the boundary condit-
ions on s 2 are given by /2.22/, /2.55/-/2.57/. Due to /2.55/,
i!=O on s 2 Hul tiplying /J.JJ/ by :!-i (t) , where :! is a vir-
tual fieid, and integrating >'le arrive at

)~~(t) (:!-i!(t)) dx = - Jsij(t)eij(v-~(t)) dx

Sl ~
+ ) ,!T (t) (:!T-~T) dS +L(:!-i:_ (t)) , t t [ 0, T] , vf:V, /J.J5/
s2
where V is a space of virtual fields
=
V l:! = (vi) ( :! = 0 on S 0 "\ /J.J6/
It is assumed that all integrals make sense.
Topics on Unilateral Contact Problems 393

Taking account of /2.22/ in /J.J5/ we readily get


j[qli (_t') (:y-f! lt)) +Sijeij (:y-~ (t))] dx +J 1 ( SN ,:y)-J 1 ( SN ti"! (t)) -
3l - J sN(t) (vN-~(t) dS ~ L(:y-~ ( t)) ,t ~[o,ij,v~v, /3.37/
s2
where f
J 1 (sN'~(t)) = j d(sN,-~(t)) dS.
s2
The global form of the Signorini's condition is
~f~ ~ sN(t)(w-~(t))ds~o, tE:[o,T],w~~, /3.38/
s2
where ~ is a closed cone in a suitably chosen space of vec-
tor valued functions defined over s 2 ; for instance ~ =
= t'"E:-L 2 (0,T;H 1/ 2 (s 2 ) I w~ 0~ in the case of linear elasti-
city or viscoelasti~ity.
Thus we see that /J.36/ and /3.38/ constitute the coupled
system of two inequalities. The coupling is produced by the
Signorini's conditions and flow law type sliding rule.
We can formulate the following variational problem.
Problem P 2
Find .!::!,lt), t ~[o,T], such that /J.34/, /3.37/ and /3.38/ are
satisfied.
It should.be remembered that now,!=,!(~.(.!::!,\) , or.!.(.!<!!.)), or
.!~~) ,.!(.!::!.)) depending upon the body ~onsidered.
rt is worth noticing that such principle of virtual
fields holds irrespective of the material of the body con-
sidered.
Examples of constitutive re1ations a11owab1e by /).)7/:
/a/ .! E: oD l!:.t.!::!.)) or nonlinear elastic material with subdiffe-
rentiable strain potential.
/b/ Viscoelastic materials with long or short memory.
d t d
/ c/ s~j ( t) = eij l t)- ~ R 1 ( t+p (x) , 't +P (x)) eijl't:) d'(
to
J R2 (t+p(x) ,'t+p(x)) el'Cjd"(,
slt-(t1= e(t)-
0

where s = .! + 3 s! .2. = .2.d + 31 e! , and


d 1

*
s ij ( t) s ( t)
ij 5 =
2G(t+p(x))' s*(t) = E(t+p{x))
The functions p,G,E,R 1 and R2 are given.
Such constitutive equations describe the ~o called ageing
isotropic materials. Viscoelastic materials constitute a
394 J.J. Telega

particular class, where Roe. ( t+p, t(' +P) = Roe ( t- t) . Obvious-


ly, anisotropic ageing materials may also be considered
[~77]. The function p(x) characterizes nonhomogeneity of
the process of ageing.
/d/ Rigid-perfectly plastic materials, l>"here
.! " ~D (e(~))
For more details, see Section 3.5.
Remark 3.2. Quasi-static Signorini's problem ldthout frict-
ion for ageing materials is studied in [177 1 178, 181 , 182, 202]
see also [204,240]. Various problems related to unilateral 'J
contact problems for linear viscoelastic bodies are investi-
gated in (18,18),220 1 221 ,230]. All those problems are sim-
pler than the Signorini's problem with friction.

An alternative approach to the study of the Signorini


problem with prescribed normal contact stresses for a class
of inelastic materials has been proposed by Raous [2411 ,see
also ~39,240). I think that more general setting including
the sliding rule /2.J8/ is readily available. Unfortunately,
Raous' approach precludes the flow law type sliding rule
/2.22/. Let~ =(eij) denote the strain tensor and ein its
e in
inelastic part, i.e. ~ = ~ + ~ We can write

/J.J9/
Let us consider the quasi-static formulation. I recall that
time dependent functions are treated as vector functions
mapping the time interval into a Banach space. Therefore I
write ~(t) 1 I(t), _!(t), etc. The convex set Kd of kinemati-
cally admissible displacements is given by /2.62/.
The constitutive relation is assumed in the following form
~in= 1 (~,~in)
' ~in (o) = ~ ' /3.40/
where e is a given function of x ~ Q
-o
For instance, in the case of Maxwell model of viscoelastic
behaviour of isotropic materials we have [233]
iin= !,~ = ~ -1.! = "l. -1!:(~-~v) , 13 411
where .2; = aijkl stands for the isotropic tensor;aiikl=O
since div!V = 0.
In the case of elastic perfectly-plastic materials
/J.40/ is written in the form
~in= !p t t>~(.!)= d ~[.2:(~-~p>] ) /3.42/
where K denotes the so called elasticity convex determined
by the yield locus. The inelastic part ein of the'strain
tensor may also stand for viscoplastic deformation.
Topics on Unilateral Contact Problems 395

Hardening and/or softening effects can readily be taken in-


to account.
Let us pass to the variational, or rather mixed /varia-
tional-strong/, formulation of the quasi static evolution
problem for materials obeying /J.40/. The Signorini's con-
ditions /2.57/ and the sliding rule /2.38/ are present on
s 2 Z.Iultiplying /2.53/ by (:!-~ (t)) and taking account of
/3.J9/, /J.40/ as well as of the boundary conditions
/2.55/-/2.58/ one can formulate the following problem, whe-
re eij (~ tt)) =o
ui (t) / '?> t.

I
Problem P 2 .
Find ~ {t) (:. Kd and !_1 n (t~ , t f: (o ,T) such that

fa(~ (t) ,:!-~ tt)) - J aijkl e;~ (t) eij (:!-~ (t)) dx +
32.
+ J(~(t) ,:!)- J(~lt) ,~(t)) ~ L(:Y-~(t~ , \r(;Kd > /3.43/

~
in
=CL(
:r !_,~in) , /3.#4/

!.in (o) = !o /3.45/


Obviously, the problem P is not purely variational.
Existence of a solution of the problem P is not a sim-
ple matter and as such the problem remains opin. Nevertheless
in [240,24~ Raous presented implicit and semi-implicit me-
thods for numerical solving of some particular oases of the
problem P 2 Results obtained exhibit that a formulation
such as /3.43/-/3.45/ is not useless.
J.J. Steady-state rolling oontaotot rigid cylinder travers-
ing a viscoelastic half-space
Consider the viscoelastic half-space u~,z)lz~/ o},
(x=(~)), traversed by a rigid roller moving with a constant
velocity v, see Fig.20. This figure shows that relative to
the symmetric position x=O /occupied by the center of the
contact region for an elastic solid/, the contact region is
displaced in the direction of rolling. Thus, even in the ab-
sence of surface shear tractions /frictionless rolling/,the
net force on the cylinder possesses a nonvanishing component
opposing the motion. Such frictionless case is investigated
in this section. The model of a behaviour of a material of
the half-space is given by the generalized Voigt's model
lii th one additional spring, for which GD stands for the dy-
namic shear modulus, see Fig.21. For tn1s model the creep
function is given by
f (t) = G~ 1 [1+ 1, fr (1-e t/ty)] , /J.46/

where fr=G; 1 /G~ 1 , while 'rr (r=1, ,n) are relaxation ti-
396 J.J. Telega

Fig.20, Schematic diagram of rolling cylinder

11n
Fig.21

mes. Specifically, the creep-strain behaviour of the mater-


ial in shear) f(t), in response to a constant stress s,ap-
plied at time t:O is expressed by
f(t): s 'f(t)H (t) /3.47/
where H (t) is Heaviside 's unit step function.
Hunter found that in the coordinate system moving 1v-ith
the roller and having the direction of a motion as the po-
sitive x 1-axis, the normal surface displacement is [203),

5
see also [239] 010

l'l(x)= ue(x)+ ~ fr e-'A ue(x 1 +V'tr~ ,x 2)d). 1 /3.48/

0
where u (x) is the instantaneous elastic displacement. This
elasticedisplacement is given by the Boussinesq equation
( )
ue x
1 - )/
21t O:D
J Jx-yJ
1' ( y)
dy ' /3.49
R2
398 ].]. Telega

The inverse Fourier trans~orm F- 1 is expressed by

F- 1(t:) = ~(x)= 1 ( f(~) ei(x,n d' . /'J.52/


(2 'Jr) 2 ~2
A linear ~unctional ~ de~ined and continuous on s(R 2 )
is called a tempered distribution /in R2/. The totality o~
such tempered distributions is denoted by s'(R2).
The Fourier trans~orm o~ ~he tempered distribution
~ ~ S 'Gt2) ( ~-~un2tioi}&.l on S { R )) is de~ined as the genera-
lized ~unction ~ ~ S Q12) such that
f ti>=(f, f)= (2 'Jt) 2 (~. i), .y.'-f' s <n 2)
Below we shall use the ~ollowing relations
F( ,!. ) = 21( ,; I , /'J-5'3/
F(~(x-a)) = ei(a,~) 1(~), a= (a 1 ,a2) ~ R2 /'J.54/
Applying now ~ormally the Fourier trans~orm to /'3.49/
and taking account o~ /'35'3/ and /'J.54/, we obtain
ue (S!)
'}
= r
)
u
e
(x) ei (x,~) dx = 1-
2'JtGD
)I J p(y) ei (x,~) dxdy:
Jx-YI
R2 R2XR2
1- v
=~ 1
~ p 1
.... (~) I '355I
~ploy now the Fourier transorm to /3.48/j we have
00

w(~) = ue(~) + ~ ~r S ( Se-'l\ ue(x 1 +Vt[ rA ,x 2) d).) ei(x,~)d~ =


R2 0 c0

('-;)) e-( 1 + 'i1V'C


=lie+ i ~r~e ri)~d>..
r 0
Performing the integration in the last relation we arrive at
'W (, - (~) ! ue ( ~) /'3.56/
\~) = ue 'l + r ~r 1+ ~ 1 V'[ri
Eliminating u
(~) by applying the inverse Fourier trans~orm
to /'J.56/ we ecan write
w = Bp = 1 2 ( 'W~(,) e-i(x,~) d~ =
~1{) J2
= 1- I' _ LJ _L
R (1 + ~ ) +V [' ~ i
r
( ~)
r "> 1 'P(~) e-i x,., d~ . /'J. 5 ?/
4 1t 2G J;t I'i I 1+V 't r ~ 1 i
D R2
398 ].]. Telega

The inverse Fourier trans~orm F- 1 is expressed by

F- 1(t:) = ~(x)= 1 ( f(~) ei(x,n d' . /'J.52/


(2 'Jr) 2 ~2
A linear ~unctional ~ de~ined and continuous on s(R 2 )
is called a tempered distribution /in R2/. The totality o~
such tempered distributions is denoted by s'(R2).
The Fourier trans~orm o~ ~he tempered distribution
~ ~ S 'Gt2) ( ~-~un2tioi}&.l on S { R )) is de~ined as the genera-
lized ~unction ~ ~ S Q12) such that
f ti>=(f, f)= (2 'Jt) 2 (~. i), .y.'-f' s <n 2)
Below we shall use the ~ollowing relations
F( ,!. ) = 21( ,; I , /'J-5'3/
F(~(x-a)) = ei(a,~) 1(~), a= (a 1 ,a2) ~ R2 /'J.54/
Applying now ~ormally the Fourier trans~orm to /'3.49/
and taking account o~ /'35'3/ and /'J.54/, we obtain
ue (S!)
'}
= r
)
u
e
(x) ei (x,~) dx = 1-
2'JtGD
)I J p(y) ei (x,~) dxdy:
Jx-YI
R2 R2XR2
1- v
=~ 1
~ p 1
.... (~) I '355I
~ploy now the Fourier transorm to /3.48/j we have
00

w(~) = ue(~) + ~ ~r S ( Se-'l\ ue(x 1 +Vt[ rA ,x 2) d).) ei(x,~)d~ =


R2 0 c0

=lie+ ~ ~r~e ('-;)) e-(1+ 'S1V'C ri)~d>..


r 0
Performing the integration ~ the last relation we arrive at
'W (, - (~) ! ue ( ~) /'3.56/
\~) = ue 'l + r ~r 1+ ~ 1 V'[ri
Eliminating u
(~) by applying the inverse Fourier trans~orm
to /'J.56/ we ecan write
w = Bp = 1 2 ( 'W~(,) e-i(x,~) d~ =
~1{) J2
= 1- I' _ LJ _L
R (1 + ~ ) +V [' ~ i
r
( ~)
r "> 1 'P(~) e-i x,., d~ . /'J. 5 ?/
4 1t 2G J;t I'i I 1+V 't r ~ 1 i
D R2
Topics on Unilateral Contact Problems 399

The fundamental properties of the pseudodifferential


operator B are given below. Beforehand however, the functio-
nal setting is introduced. Hs (R2) (s E- R) is the usual Sobo-
lev' s space with the norm
! v
!lv !I = ) ( 1 + I~ l 2 ) sI l ~) 12 d ~
R2
/'].58/

Let D be some fixed, sufficiently regular domain of R 2


Fundamental for the study of the operator B above are the
follol'ling spaces:
i'Is(D)= tvE-Hs(R 2 )jsupp vcD~, /'3.59/
which is the closure of C 00 (D) in Hs(R 2 ); next is the spa-
ce Hs(D) or the space of 0 functions /distributions for
s <O/ in D admitting an extension on R2 belonging to Hs(R 2 ).
The norm of v E: Hs (D) is
llv lls,D= inffllLvll sl LvE:Hs(R 2 ) and LviD=v).
lil'e have H1/ 2 (D) c H1/ 2 (D) , and the inclusion is strict, so
that the norms II. II s and II II s, D are different.
The notion of a pseudodifferential operator(P.D.O) A
is introduced after Eskin [195].
Let A(~) ~ L 1 10 c(Rn) satisfy the following condition ..
lA(~) I ~ c (1+ l~l)oC.. /'].60/
The class of such functions is denoted by S~ Obviously,
in our case n:2.
Definition '31 The operator A is called pseudodifferential
if it is defined on S(Rn) according to the formula
Au= 1 n S A(~) u (~) e-i(x,f;) d~. /'].61/
(2'Ji.) Rn
The function A (') is called the symbol of the operator A.
If A(~) is a polynomial with respect to~ , i.e.
A ( ~) = ~~ m ak ~ k , then

Au= ( 1 n
2 'f()
r L
) I~ m
ak ~ku (~)e-i(x, ~) dtr = r
lk I ~
akDku (x).
m
Rn
Thus A is then the differential operator.
Comparing /'J-57/ with /'3.61/ we conclude that B is the
pseudodifferential operator.
The symbol B (~) of B is
400 J.J. Telega

/3.62/

Lemma 31 Let A(ts) t S0 Then the P.D.O. A with the sym-


bol A(~) sa tisf'ies, f'or each s ~ R 1 , the inequality
IIAu.lls-oe: ~ OS II u II s , - uE: s(Rn). /3.63/
Consequently, by continuity itJ;an be extended to the bound-
ed operator f'rom Hs(Rn) to H - (Rn).
Proof'. We have
(IAufl !-ce = J 0+ 1~1 '(s-ae)IA(~)u(i)l 2 d~ ~
Rn
~OS j (1+ l~lfs ru (~)1 2 d~ ~
Rn
OS II uu! 0

Properties of' the operator B


.1 B is positive, that is Bq~O f'or all. q ~coD (R 2) ,q~O.
~his property follows directly f'rom /3.48/. 0
.& IB(~)I~o l~t- 1 , where o=o(f'r,V't ~; see /3.62/.
l The previous property and Lemma 3.1 imply that the ope-
ra tor B is bounded f'rom Hs (D) to Hs+ 1 (D) f'or every s ~ R
{ce=-1) Hence
(Bp 1 q) =J2
(Bp) (x)q(x)dx p,q H-1/2( D) ,
0
/3.64/
R by the continuity f'or all. p,q H-
is well. defined 1/2 (D )
! The operator B is coercive on H:- 1/ 2 (n), that is
(Bq,q)); c II qll 2 _1/2 , 'V'q H:- 112 (n) /3.65/
where C is a positive constant.
To prove /3.65/ let us decompose B into the sum: B:B 1 +B 2 ,
which is obtained in the following way, We may write
Bq= 1-v L (
[(1+f'r)+V'(r'1i){1-V't'r's1i)ql~)e-i(x,~l~.
4'J 2GD r J2 ~~~ (1+Vtr ~ i)(1-vCCr~ 1 i) 1
Hence
Topics on Unilateral Contact Problems 401

1ie seoe that B 2 =-B~ 01 w4~re B~ is the adjoint of B 2 Thus


(B 2 q,q) :0 for q'= H-11 (D).
In physical situations we a~l'l'ays have Y < 1. ~en fro~ /
/].66/ l'i'e obtain (B 1 q,q))/CI\qll_ 112 , since 1~1- '>/(1+ l~l )- 1 ;
ru~
\B
l
q q)-
1- V
I J -
1 ~+frJ +
(
,v '[ r)
2 2
1l . . .
't)\ 2 d~'J
~ q'-.,
1 ' - 4 1[ 2 c0 r ~ 2 I~ I 1+ ( V '( r) 2 ~ ~
The latter relation follows from the Parseval's formula
~ f(x) g(x) dx = ~) r(~)g(~) d~ ,
R2 (2 '.i1) R2
l'l'here g (x) is a real fUnction. Thus the coerciveness proper-
ty /3.65/ of B has been proved.
Having introduced the P.D.O. B we can pass to the va-
riational formulation. In a steady motion the rigid rolier
immerses at a fixed depthA. Let GcR2 be the unknown do-
main of contact; therefore the rolling problem considered
belongs to so called free boundary problems /as for the
rest the unilateral problems studied previously/.
For z:O we can write
w(x)= A +h(x) , p(x) ); 0 , x (-G, /3.68/
w (x) ~A +h (x) , p (x) 0, xt G =
Taking account of /3-57/ we arrive at the following comple-
mentarity conditions
p(x))/0, Bp(x)~A +h(_x), p(x)Bp(x)=p(x) (A +h(x)), /3.69/
2 2
for all .x. ~ R , provided that h has been extended on R
Such an extension must preserve /3.69/ 2
A known and fixed domain D is now chosen, Gc D. For
instance, the projection of the roller on x-plane may be
suitable. We put
t I
H~ = q & H:- 1I 2 (D) q -q 0 and ( 1 , q) =Q ) /3 70/
The set H~ is closed and convex in H-
1/ 2 (D) and Q has the
meaning or the total pressure. We see that at least for
p f:- L 2 (D) /3.69/ is equivalent to
(Bp,q))1 (A +h,q), q ~H~ ,
/3.71/
(Bp,p) =(b +h,p).
Subtraction yields
(Bp,q-p) ) 1 (6 ,q-P)+ (h,q-p) = a(1 ,q-p) + (h,q-p).
But (1 ,q-p)=(1 ,q)- (1,p):O, since p,qEH~
In this manner we eventually obtain the fo.llowing V.I. t'or
the determination of a distribution ot' the contact pressure
402 J.J. Telega

/3.72/

To so1ve this prob1em,the weight Q of the roller must be


prescribed.
The next prob1em arises for A >O prescribed. Toward
this end let us define the closed and convex cone
H+ c ~- 1 1 2 (n) , i.e.
H+ ={qfH- 1/ 2 (n)lq>1 o'l. /3.73/
From /3.69/ one readily obtains the following variationa1
prob1em:
find p t H+ such that
(Bp,q)~ (A +h,q) , q ~ H+ /3-74/
and (Bp,p)= (A +h,p). /3.75/
The re1ations /3.74/ and /3.75/ are equivalent to the fol-
1owing variational ineqa1ity
I pE H+: (Bp,q-p) ~(A +h,q-p), 'q E- H+.
Now we can formu1ate [17~
Theorem 33 a/ The variationa1 ~?~blem P 3 possesses a
unique so1ution pQ for each hE: H (D) , Q q 0.
b/ The variational problem P 4 has a unique so1ution Pc. for
each hfH 112 (D), A fR+:(O,oo).
c/ p 6 =PQ if Q= (1 tPt:J and pQ=pA if A = (BpQ-h'pQ) /Q.
d/ The correspondence A...,__. Q= (1 ,p6} is a continuous,
strict1y increasing function from R+ onto R+. The srune ho1ds
for its inverse: Q,___..._A=(BpQ-h'pQ) /Q.
e/ Suppose that p 1 and p 2 are solutions of (P 3) with Q:Q 1 Fo
and Q:Q 2 fo, respeotively. Then
I Q1-Q21 ')/ ll Q1p2-Q2P1II-1/2
Points a/ and b/ resu1t from the general theory of varia-
tiona1 inequa1ities due to /3.65/ and the continuity of the
respective 1inear functionals [351 The remaining points
are proved in (179] Finite element approximations are also
studied in ~79). Elastic prob1em was ear1ier investigated
in DBo). .
Remark 33 It wou1d be interesting to apply ~fosco's theory
of dua1ity to variational problems P 3 and P 4
Remark ').4. L'vov [223) developed an incrementa1 procedure
for the study of a friction1ess contact of a shal1ow creep-
ing shell with a rigid stamp. Geometrically 1inear case was
investigated.
Topics on Unilateral Contact Problems 403

3.4. A finitely deforming viscoelastic cylinder in steady-


-state ro1lin& contact with a rough foundation
An inverse problem, in a certain sense, to that stud-
ied in the preceding section will be presented in this sec-
tion. Now a cylinder rolling with a constant velocity v is
finitely deformable while the foundation is rigid but 0
rough /rolling with friction/. The physical situation may be
describe in the follo,dng way: a viscoelastic cylinder un-
dergoing finite deformations is rolling on a rough foundat-
ion at a constant angular velocity ~ , see Fig.22 for an
example. Suoh
problem has been
intensively stud-
ied in (234,23.5] ,
see also [150).
By v we denote
-o
( a)
the velocity of
the foundation
/e.g. roadway or
belt/ relative to
the axis of the
cylinder /in the
absence of slip-
ping' v 0 ::R 0 lA:J
for rigid bodies/.
The cylinder can
be attached to a
rigid cylindrical
axle of radius
R . , and this ax-
J..
~e spins at a con-
stant angular ve-
locity G) An ob-
server riding on
the axle will see
the same geomet-
ry of the defor-
med cylinder at
all times t. The
~'ig. 22 ,/a/ Geometry of a rolling
situation is al-
cylinder. /b/ Rolling cylinder so equivalent to
under applied load a cylinder spin-
ning about a fix-
ed axis and. in
contact with a moving belt. A key consideration is the ki-
nematics: we compare the geometry of the deformed cylinder
in its current configuration C with that of a rigid cylin-
der spinning at the same angular velocity 6> , the latter
404 J.J. Telega

characterizing the reference configuration C Polar cylin-


drical coordinates r,e,z of a particle l'lith0 labels (rt, Q ,z)
at some arbitrary reference time t:O are defined by
r::R, 9= 9 + {.a)t , z:Z , for t '# 0. /3.77/
Alternatively, one can employ the Cartesian reference coor-
dinates
x 1:rcos9 , x 2 :rsin9 , x 3:z. /3.78/
Let us denote by (x.) (i=1 ,2,3) the spatial coordinates;
these are the Cartesian Coordinates of points in space rela-
tive to a moving ooordina te frame ld th origin fixed on the
axes of the moving cylinder: x 1 parallel to the foundation,
x noDmal to the foundation and x directed along the axis
ot the cylinder. The motion ~ of 3 the cylinder is defined
by an invertible, twice differentiable map ~ that takes
the configuration C into the configuration C; i.e.,
0
xi=~i(r,e,z)= li(xi), i,I=1,2,3. /3.79/
The deformation gradient tensor ! is given by
!,=Vi, or !(r,9,z)= ["d~i/'dr,'d~i/()9,"d~i/'dz]. /3.80/
Thus we conclude that time enters our description of motion
only implici ty as 9= +cut.
The right and left Cauchy-Green deformation tensors, C
and ~' and the Green strain tensor ~ are given, respectiveZ
ly,by C:FTF:U2 B:F FT:V 2 E= 1(c-1\ /3.81/
----,----, - 2--JJ
where, according to the polar decomposition theorem,
=
! = !!. :!! :Y !!. /J.82/
If r denotes the position vector of the particle l~ith
reference coordinates (r,e,z) at time t, the vector
~ = !,(r,e,z)-!., /3.83/
is the displacement vector of this particle.
The reference velocity vector is given by
o~i(r,e,z) "
vi = 'd t = c.> d xi/() t. /'J. 84/
The acceleration vector has the form
dVi 2 2-'\ 2
ai= ~ = G.> 'd xi/"a e /J.85/
We observe that since the reference coordinates are time de-
pendent , hence
()ui/'0 t= vi- ()ri/'d t: (,.) (d~i/'d 9- ari/o e), /3.86/
where ri are the Cartesian components of r.
-In our case we have
'dx' 11 t= ~ GC 1xoe , -ax 3; 'd t=O /3.87/
Topics on Unilateral Contact Problems 405

where Greek indices take values 1 and 2 while (~ 9 )is the


permutation symbol.
Further, /3. 80/ furnishes
; e ["d ~i/d r, 1 xi/de, d~i/"0 z] ="''d!/d e.
f.= c.> /3.88/
Taking account of /3.87/, from /J.88/ we obtain
- 1\
Foej = wX'~-.f,..t..#'~x~,>-t , FJj=FjJ=O,
l'lhere xoe.,
jt =
d x~.f 'd XJ d X t . Hence
co('' =c.> x,. t 1 >- (xf'-' )IcC x)"-, 1 +~JA-, '). 1 x,..,cot') , c j 3=o 1 J. 89/
Let s denote the Cauchy stress tensor and S the second
/symmetric/ Fiola-Kirchhoff stress tensor, related to one
another according to 1 T
.= (detf.)!.- !. f..- /J.90/
The first /unsynwetric/ Fiola-Kirchhoff stress tensor T is
expressed by
! = f. 2. /J.91/
A viscoelastic material of the rolling cylinder obeys
the follOl'lin.g constitutive equation
s:G U>(x,c)+ G t2 >(!,c), /3.92/
where G ( 1) and G ( 2) ~; re;p~nse-functions.
Inthe case considered, taking account of /3.89/, l.re
can l'lrite _2 ( 1)
s(x,vx,
vi,"')= G (x,~ ~J k) +
-- - - - -
..
1,J ,
+
Q:. (2) (X, WXt~f~
" ,..Xp.J ct: "X}J-') "
9 " \ I
1 x,_,~J J9?
I +x_r,,,
Thus rate dependence enters /J.9J/ as a depenaence of G(2
2
on V ~ and CAl
The equation of motion is expressed by
T
DiV! + q ~ = 9~, /3.94/
where Div stands for the divergence operator corresponding
to the reference configuration, b is the body force per
unit mass in the reference configuration and ,xi=ai , g=(2fi}
of. the formula /J.85/ .
LetS= {~,e,z) r=R.~. Ve assume that if the cylinder
is fixed 0 to the rigid 1 spinning axle, then
i(Ri,e,z)=(Ri,e,z) on 5 0 /J.9'S/
Since the fotmdation is rough, so contact and friction
conditions have to be ~posed on s 2 Here s 2 denotes the
candidate contact surface in the reference configuration
/5 2 could even be taken as the entire outside surface of the
oyiinder/. Let n denote a unit exterior normal to 5 2 while
n stands for a -o unit outl"~ard normal to the deformea sur-
1aoe, see Fig.23. The surface traction vector! at a point
on s 2 is expressed by
406 J.J. Telega

t
-
=
T n
- -o
T
13.96/
The vector t can be
decomposed Into the
normal and tangent-
ia~ parts in the fo~
~owing way

t
n=t
-- n =
=t J ~ Jini' 13-971
~
n, !m=
t =tn- - ...
= t-t -~

A good presentation
of .the kinematics of
finite deformations
as well as of stress
measures is given
in the book (285)
Fig.23

The s~ip velocity !T is defined by


:!!T =
:!-Yo ' on 82 '
where :! denote the particle velocity on s2.
!h~_!~2!~!!!_2~~!]!2U! take the form
~ 2 -H~o, tn~o, tn(~ 2 -H)=o,ons 2 , 13.991
where H is the distance from the axis of the cylinder to
the foundation , see Fig.23.
The friction lalt is assumed to be of the flow ~al'IT type,
cf.I2:227:-Now-we-hive
-!T E: d~ (tn) ( .t_r) , on s2 )
where
K(tn)= t!.lg(tn!.)~o) , on s 2
We note that in [234,235) such genera~ law is not consider-
ed.
The set of admissib~e motions is defined by
V= {~=i(r,e,z) ~~~B J ~is prescribed on S 0
and the energy is liell defined), I .3. 101 I
where B is in general a Banach space, for instance B~i 21 P(QJ3
~!r~gar!:i!:~ interior of the cylinder in the reference
The set K of admissible motions has the form
Topics on Unilateral Contact Problems 407

/3.102/
Let us also introduce a special velocity - motion space
vH = \.~ = vail!. tv) ' I 3. 10 3/
and
Kl-1 =t!~ vMI! =ve&, i(: K) /3.104/
where ve. =
("'-l"de~ 1 ,~ 2 , Ca.) () 9 x 3). The vector',!~ VM seems
to be artificial since its second component has the dimen-
sion of a displacement whereas the first and third of sped.
Therefore it lrould be better to present the problem at the
very begining in a non-dimensional form to avoid such am-
biguity.
It is convenient to introduce a collection of nonli-
near forms:
A(!,~)=~ tr[t .(X, Vi, v 2i, c..>) V"!,.ldSl. 0 ' /3.105/

/virtual power of the stresses/;


B(i,"!l=~ ~(-a~)de~d.Q.o' /3.106/

/the power developed by inertial forces/;


L( ~) =.d 9 !?. ~ d.Q.0 /3.107/
0
/the virtual power of body forces/;
J(g, ~_) d =J (tn
(V ,
v 2~, w) ,-_!!T) dS ,
so
/virtual. pol.rer of the frictional. forces/,
where tr stand for trace of tensors.
The variational formulation is obtained in the usual
way: we multiply the equations of motion /3.94/ by
(i7& .;., V ~), integrate over St , perform the integration by
parts and next use /3.97/ ~d /3.100/. Finally we arrive
at the following implicit variational inequality:

i- Vei)+ J(!, '!J -J(i, V e! )~/


l''ind i E: K such that
(P) A(i,
'>iw 2 B(i,'!-v 9 !_ +L(~-V 9~), 1E: ~I
/3.108/
IVI /3.108/ is not the usual one since !, (:K while '!! E ~
Moreover, the sets K and KM are not necessarily convex
and closed due to a depenaence of the set V of admissible
motions on the energy. Such an example is given by the con-
dition det:) 1.
408 J.J. Telega

lie note that formulation such as /3.108/ is available


for a very limited class of problems. It seems that even
for geometrically and kinematically the same problem but
in the case of transient process, where time enters expli-
citly, IVI similar to /3.108/ cannot be derived. As in
Section 3.2, however, we can always obtain a variational
formulation in the form of coupled inequalities. The gene-
ral procedure is similar to that employed in Section 3.2.
Let us apply this method to the same steady-state rolling
problem of finitely deforming cylinder.
For this purpose by Vad is denoted the space of suf-
ficiently regular virtual velocity fields; in general
Vad#V:z..t Multiplying /3.94/ by(~ -y_), Y.=!, 1. ~Vad' next
integratint- over ~0 , in the standard manner we obtain the
IVI /3.109/ below. The Signorini's conditions /3.99/ are
represented in the global form /3.110/ below where V(S~)
is a properly chosen space of functions defined on s 2 ;~
for instance v(s 2) = [1 2 - 1/p,p (s 2))t =W- 2 + 1/p,q (s 2) , 1/p+1/q=1
The variational formulation of the steady-state rol-
ling of the cylinder considered reads:
Find ~ ~ V such that
A(i, ;-Y.) + J(i, '1) -J(i,Y.)- ~ tn( ~ n-v~~ dS ~
(Pnat) 2 2

- -y_),
5-
and
~~ B(,tt-!)+ L(~
-
.:!f'i(:Vad , /3.109/

i 2 -H~ o : 8 tn(z-x 2)dS>,.o, z & Kn. /3.110/


2
/P at - natural formulation/. Obviously tn depends on V !,,
; and c..> ; moreover
Kn={ z E: v(s 2) I
z-n 2 ~ o) . /3.111/
As I have already mentioned, formulation such as (Pna~
can be used to non-stationary problems, yet the form
of inequalities /3.109/ and /3110/ will depend on the par-
ticular case considered.
A great number of numer~cal examples is given in [150,
234, 235) Particularly, in L234) to the hyperelas tic law
/see Gd) in /3.92// viscoelastic aspects of the response
of the oyl:inder are appended :in the form of the following
constitutive equation t

~(2)('1:)~=-oo= jg{!(t)- Jexp{-~- 't) /Tr~~:,('l:)d'C


t-r_()D /3.112/
Topics on Unilateral Contact Problems 409

where is the glassy modulus of the material and T is


~
g r
the relaxatio11 time. Such law can describe viscous behaviour
of elastomers.
To discretize the problem P, the non-smooth functional
J has to be regularized. In the case of Coulomb's friction
one can take
.!T =-
p.l tnl f E(ll!TJ) l!rr!'ll!T I J /J 11 J/
where ~~ is a sufficiently smooth function, see Fig.24. It
is a continuous, monotone, real-value& function of the non-
negative real numbers 1~\
depending on a parameter l"? 0
such that
/a/ o ~ 'f f (I:!T 1)~ 1 ;
/b/ (~ 'f E (I:!TI) = 1 ,-I:!Tl) O;
---------;----------wr /c/ lim f f. (ll!TI) =1, l)O
\l!T)-.oo

l
Examples of functions satis-
fying these properties are
1 , if I:!T I'> ~ ,
(a)
.J
'
(lw o- /J.114/
-T - jl!TI /(,,if ll!TI!-'
and
I AI f (ll!TI) = tanhQ~I j /3.115/
IBl___.>"''r, Instead of J we take the re-
gularized functional J~ de-
-f: f fined by
JE (i_, '_!) =
s (tn,fEQ~~sJ
= d t">O /J.116/
s2
provided that the density d
(b) depends on t and w"' As
n - ...
previously, t depends on
n
Fig.24, /a/ Classical Cou-
lomb's friction law;/b/ re- v!, .f! and cu
gularized Coulomb's friction
law

The rigid foundation is replaced by a deformable one,


characterized, for instance, by the functional, see the
formula /2.136/,
410 J.J. Telega

Regularization of the problem P signifies that for


each fixed ~ 0 we consider the problem:
find "
~f : V such that
O:>~) A(~p '!)- c.>2B(~ ,'!)+D(i(J'.~) +Jt(~i''!)=
= L t ~) - ]: (; VM
The variational equation /3.118/ is to be used as the
basis of discretizations Problem P~ can be discretized
by using Lagrangian finite elements. For this purpose 1Y'e
construct a partition of ~~ into a mesh of finite elements
~ over which the motion
~t is approximated by piece-
ldS~ polynomials in such a way
that a finite dimensional
subspace vh of v is formed, h ~ 0 being the fresh parameter.
Over each element ~ , the restriction of !.t
of !,L is defi-
ned as a linear comb!nation of shape functions N6 , A =
= 1, ,N ;N being the number of aodes of element ~,see
Fig.25. Tfiuse we have e
N
hh e h
2ft =b~1 Nt.(~ /l.) ~A

IA(iE'~)-i c.iB(i~,~)
The discrete form of the problem Pl is the following
problem: h _h
h) find ~ v-- such that
~E. +D (i~,~)+J~(i~,~) = /3.120/
= L(~) z E;V~
Formulations such as (P), (P~) and (p~) characterize
compressible materials. Incompressibility can be handled by
the method of Lagrange multipliers [234)
Numerical results presented in [234) concern the rol-
ling cylinder '\Y"hich is in a state of plane strain;R 1 =1 in.
and R =2 in. The elastic term G{1)in the constitutiv~ equat-
ion ig characterized by Mooney::"Rivlin law 1dth coefficients
of the stored elastic energy function of C 1 =80 psi and c 2 =
=20 psi. I recall that the Hooney-Rivlin strain energy
function W is expressed by /incompressible material/
w(I 1 ,I 2) = c 1 (I 1-3) + c 2 (I 2 -3),
where I 1 ,I 2 are the principal invariants of the tensor Cor
I 1 =tr. , I 2 = (tr.) 2/2 - (tr.2)/2 , 2
since for 4ncompressible materials I~= (det[) =1, cf.[285).
Further data are: c.> = 1 0 rad/ s , IJISP=R -H=O. 2 in, v =
0 0
:c.>H, c n =105, mn =1, r-
.u..:O.J /coefficient of friction/,
Topics on Unilateral Contact Problems 411

Fig.25
.j::>.
......
N

z z 1-0 t----r-------.-----..---.
< C1=80 C.a=20 DISP=.2 W=10Vo=18 ~ C,=~ C2.=~ DISP=.2 W=10 Vo=18
~ 1.0 VIS= 0 FRIC=.3
a: ~ VIS.: 0 FRI(=.3
" VIS= 0 FRIC =-0 & VIS= 0 FRIC=.O
+ V1S=100 FRIC=-3
0.5
VI VI +VIS =100 FRIC =.3
> >
+
.... +
u .!} o.o
+ o.o +
u u
:::: -0.5
VI
'
VI VI
w v1
a: w
~
VI -1.0
g: -1.0
Vl
u.. LL
u.. LL
0
::c ~ -1.5
::c
u ::c
u
a: a:
:..:: S2
-2.0 I
-2.0
< <(
_J
0 -'
0
0:: 0:: - 2 .5
D D
z z
N N

50.0 100.0 150.0 50.0 100.0 150.0


DEGREE DEGREE

<-...;
Radial stress near bonded Circumferential stress near ~
surface bonded surface >-3
("!)
Fig.26 (D
~
Q-3
]
"'0
::t
c::::
::t
t=:.
0.4 r-----,-----r---~---. P>
.,.
(I)
11= 80 (720 DISP=.2 W=10 Yo= 18 C1=1Kl Cz= 20 DISP= . 2 W=10 Vo=10 ....
z e:...
:5 VIS= 0 FRIC=.3 eVIS::.Q FRIC=.3
~ 2.0
A VIS= 0 FRIC=-0 .a. VIS= 0 FRI C-= - 0 g
.,.
+ V IS=100 FRIC=-3 + VIS=1Xl FRI C= .3
VI 0.3 .,.~
>+ ::,0
0
s 1.0
+ f
.s i3
0.2 "'
VI
VI
LLJ 0.0
IX
1-
VI
-z
c(

u.. Cl 0.1
u.. <(
0 Q;

c:::t: -1.0 -
IX
z
<(
~
I
IX
1-
c( Ill
....J
0
a: -2.0 z
LLJ
Cl .....
z
..... IX
1.::1
100.0 150D 50.0 100.0
DEGREE DEGREE

Shear stress near bonded Radial strain near bonded


sur:faoe sur:faoe
Rig. 27 .j::o.
1-'
\.>)
414 J.J. Telega

~ :0.25 in/s, VIS: "'=100 psi, T :0.1 s. Some of the nume-


rical results are ~e~resented inrFigs 26 and 27 after (234).
We conclude that the inclusion of viscoelastic effects pro-
duces a softer material: peak values of the dimensionless
radial stress, circumeferential stress, and contact pres-
sure are seen to be smaller for the viscous case tha.n {or the
same elastio stress. All strain components increase slight-
ly for the viscous case, as one could expect.
35 Unilateral contact problems of plasticity:perfectly
plastic bodies
In the present section the classical limit analysis
theory is extended to unilateral boundary conditions. The
frictionless case has been already studied in ~67,198,24~.
Extension of limit analysis theorem to the case with frict-
ion has been proposed in [167].
Let ~ be a rigid perfectly-plastic body and let ~= (U.iJ
and!.,= (si) , != (t .. ) denote a velocity field and stress
fields, r~spectivt!y. Signorini's unilateral boundary con-
ditions, in the absence ot friction, are given by
c:
!.T=O , ~ .... 0 ,
<.
sN.., 0 , sN~ =0 on s 2 /3.121/
By e we denote a convex of plasticity.
For instance, if f{!)=O stands for the yield surface, then
e =~! = (tij) 1f (!)~o) /3.122/
The support function de of e is given by
de(.!.) :sup{!_ i - I 0 (t)) :sup{!: .!.Ji E c\ , /3.123/
t
where !! :tiJeij and i is the strain rate tensor. The as-
sociated flo,., rule is assumed in the subdifferental form
! E. ai 0 (!) , or ! f. o de (.!.) . /3.124/
Each o' the relations /3.124/ is equivalent to, see Lemma
2.1
Ic(!) + de(.!) = t .a .
-- /3.125/
Since ! ~ e, hence
de(.!) =
!.! ) /3.126/
in physical situations. Thus de is the density of the plas-
tic dissipation. It is convex and l.s.c.
The total dissipation is given by
n(!) = J51.de{!(x}) dx. /3127/
The volume integral is assumed to include contributions
from ve1ooity discontinuities.
Physica11y, it is sufficient to assume that D is convex and
subdifferentiable. However, mathematically the problem is
muoh more subtle, see (251,280,28~ and the lectures by
Suquet.
Topics on Unilateral Contact Problems 415

For this purpose I recall the notion of convex functionals


of a measure ..
Let f be a convex function on Rn satisfying the follow-
ing property
3 c 0 ~c1 ;> O, '3 k 0
c 1 lz\ -k0 ~ ftz)~ c 0 (1+ \z\)
,
zc.Rn /.3.128/
The condition /.3.128/ implies continuity of f. Then K
= dom f* = { z lf"'tz) <. + oe) is a bounded convex set in Rn.
=
The support function of K is defined by (
. f 00 (z) = suplzz*l z* EK = lim >. f z/>a)
~-..o+
It is known that f~ is convex, positively homogeneous and
satisfies /,3.128//with k 0 =0/.
Let .Q. c Rn be a bounded domain and M1(.Q. ,Rn) the spa-
ce of bounde~ Radon measures on Q with values in Rn. The
norm of)A-f:. M (Q ,Rn) is denoted by 5S?.IfA-1 /the total variat-
ion/ and is expressed by
1\}lll= ~ IJ11 =sup\_l<,..,g)llgEC 0 (Sl,Rn);sup,g(x)l~1). /.3.129/
2 X
We have *
,Rn) =
H
1
0 (5'2. ,nn)] .
(52. [c
Definition 3.2. To every convex function satisfying /.3.129/
corresponds the functional F' denoted by f(f) defined on J
M1 l&2.
,nn) by Sl.
F{r-)= sup(~ vr- J.f'(v(x~dxlvE:oC 0 (.Q.,Rn)). /.3.1.30/
St ;n.
Notation )fl~> is justified by the following integ-
ral representatio~ theorem.
Theorem ,3.4.
/i/ F <;>= j f(f) is the weak lower semicontinuous regula-
rized functfonal /weak l.s.c. envelope/ in :t-1 1(Q ,nn) /in
the sense of duality (M
1(Q ,nn), C0 (Q ,Rn)) of the functio-
nal F defined by
,..., {{ f (h (x)) dx , if fA- =h (x) dx
Ff u.) = Jg
y-+ Qe , otherwise.
F is convex and strongly continuous in M1 (Sl,nn).
/ii/ If p. :hdx + p..8 ( h : L 1{51 ,nn)) represents the Lebesque
decomposition of~, then
416 J.J. Telega

/3.131/

The theorem just formulated is profound and some com-


ments are necessary.
By l.s.c. envelope ofF is meant the greatest lower
semicontinuous functional F, such that F!: F, see [286).
~ is the Radon-Nikodym derivative of u.. 5 1dth respect
di)A-sl " -
to the total variation ljLsl.
Let us continue the study of the limit analysis. In
this case we have f=de, k 0 =0 and de( .2,) ~ c 0 ( 1 + 1~I), provided
that e is bounded. The dissipation density is non-negative.
We assume that l.h.s. inequality in /3.128/ is also satis-
fied. Then dom !:e and f 00 is the bidual of f, that is
foc=3 =de
A proper space for velocity fields is the space BD
defined by
BD(.Q)={:!:! E [L 1 lSl)J 3 !eij(:!:!) ~ .!-1 1 (2.)~. /3.132/
Thus we have .2.(:!:!) '~1\&1, 1Ms(R 3)), where (f}fs (R 3)) is the spa-
ce of symmetric 3x3 real matrices /2x2 in the tl~-dimension
al case/.
If a velocity fields vE:BD(_Q,) 1 then the total dissi-
pation D~~~ has the precise sense resulting f:om /3.131/,
or D is the convex functional of the measure ,2.(!.).
Now I pass to the formulation of limit analysis theo-
rems in the presence of Signorini's boundary conditions
without friction.
A stress field t defined over ~ is said to be stati-
cally admissible if -
t. . .+b.
l.J ' J .
0 ,
l.
=
in Q. , /3.133/
.!(~)~ etx) in Q.' /3.134/
s 0
tijnj=m pi , on s1 , /3.135/
.!T = 0 , ~~ o , on s 2 . /3.136/
Here ms is the static loeamultiplier, whereas p 0 is pres-
cribed. The conditions /'3.136/ may be called "non-classical"
in the theory of limit analysis. We also note that ms does
not affect body forces. Such an approach seems to me reason-
able,sinoe the usual.J.y considered 11 body forces" ms!?_ are
physically anrealizable within the theory considered.
Topics on Unilateral Contact Problems 417

A velocity field v defined over Sl is said to be


kinematically admissible if
v = 0 on 8
0
/3~137/
vN~ 0 , on 8 2 /J.1J8/
5 P.o. dS ) 0 /3.139/
81
For incompressible materials such a field must ad.di tional-
ly satisfy the incompressibi lity condition
v.1, i
= 0 , in .Q..
The kinematically admissible load multiplier mk is defined
as follows D\e(:~:))-~~~dx
mk() =
5p_<!'J:..d8
81
If the load affects also body forces /usual approach/
then, instead of /3-141/, we have
D (.! (:t))
mk() =
) p_<?~d8 +) ~: ~d.x
81 ~
The difference between mk(:!) and mk {) is obvious.
By a complete solution of the limit analysis problem
we mean a triple {m,~,~) such that the str~ss fields ~ is
statically admissible, the velocity field u is kinematical-
ly admissible where~s m is the associated Yoad multiplier,
that is m:mS(.!!) = m (U.). Horeover ~and e(!) are interrela-
ted by the flow rule 73.124/ 1 or /3.124J;.
Lower bound theorem states that
m8 (.-~.) ~ m , /3 1 4 J/
where t is statically admissible. In other words the lower
bound theorem reads
find s
sup{ m (i)l! E Ks) J
where
K8 =t!= (tij) \ .:!:,(x) ~ c (x), x E: S2,, tij ,j+bi:o, S1;
s 0 ~ l
tijnj=m pi' 8 1 ; tN.::t o, .:!:t:r=O, 8 2 } '

is the set of statically admissible fields.


418 J.J. Telega

To prove the inequality / J. 1 4 3/ \ve take


( (ti.-s. )
. u.dx = 0 , tE.K. /3.145/
)S?. J
::1.. ~J
- s ,J
Integrating by parts and taking account of the Signorini's
condition sN~=O, \ve readily arrive at

o ~ ~ (tij-aij) eij(_0dx =(ms-m)) p~uidS+ JtNunds.


Sl s1 52
The inequalities tN ~ 0 and ~ ~ 0 imply tN~ )/ 0.
Hence we eventually obtain /j.143/.
Upper bound theorem states that
m~mk() , {:Kv, /3.146/
where K is the set of kinematically admissible velocity
fields,vi.e.
Kv={ =(.vi)j
:o, s 0 ; vN~ o , s 2) , /3.147/
or in the case of incompressible materials
Kv=L~:Jvi,i=o,2.; ~=O, s 0
; vN~o, s 2). /J.148/
To prove /3.146/ \ve assume that D is subdifferentiable; then
.D (.!(i~J) -D (.!(i:. ~ ~sijeij(-S!)cix , :! : Kv
According to /3.141/, ~substitute

mk( J.
.) E.0ydS+ ~ ydx , m 1?_o.~dS+ ~~ ~dx
5 1

s1 Q s2
for D(.!(!J) and D(.!(~~' respectively. Thus we readily arri-

J sp_~ydS JsN~NdS
ve at
mk l?.C?.dS ~ m +
s1 s1 s2
Since sNVN)..-: 0 and the relation /3.139/ holds, \ve finally
obtain 13.146/.
Upper and lower bound theorem remain valid if,instead
of the relation /J.141/, the following definition of the
kinematic load multiplier is assumed~
1(! .! (i':) -: )cix - ~ vNdS J
m~(:!)
- J
- 9. 2 /3.149/
p_'!y_dS
s1
lvhere! E. d de(.~()). The formula /3.149/ results directly
Topics on Unilateral Contact Problems 419

from the ~rinciple of virtual velocities. We observe that


mk(.f!) = m1 U!,), provided that !! enters a complete solution,
since then ~N~=O
Such theory of limit analysis where the load multiplier
affects only some components of loadings can readily be ex-
tended to perfectly plastic plates and shells. In the exis-
ting mathematically rigorous studies of plastic plates the
load multiplier always affects all components of an applied
loading [185, 188, 250)
A rigorous proof of the relation
m= sup ms(i'= inf mk1 ti) /3.150/
t ~K / v ~K \. 7
- s - v
is still not available.
However, one can prove that
- .....s. ,..k,.
m1 = sup m i =.inf m \~) /3.151/
t6K v~K
- s - v

/3.152/
For this purpose Christiansen's min-max theorem (184]
will be applied.
Theorem 35 Assume that X and Y are normed real veotor spa
ces with a continuous bilinear form a (. , ) on X1. Y. Let B c. X
a."'ld C c Y be convex sets such that B has nonempty interior,
and C is closed. Assume the following conditions:
/i/ If 'Y*E Y* satisf'ies sup{yif(Y) l y (:. C )<oD, then there exists
x G X such that yl'{y) = a (x ,y) f'or each y ~ Y.
/ii./ ~f' xli ~xJt satisf'ies 0
inf't~(x)l x E: B)) -~ and
a(x,y) =0 "fyE: Y=)x*{x)=O
then there exists y 0 E: Y0 such that
x~(x)= a(x,y) xEX.
Under these 0011.di tions
inf' sup a(x,y) = max inf a(x,y),
xf:ByE:C y(:.CxfB
provided that the left hand side is finite.
(3x0 EB : sup a(x0 ,Y)< +oo; we can take x 0 =0; B= interior
of' B) Y = c
The detailed proof' is given in [184].
To apply Th.3.5 we first define the bilinear form
a~.,.) and spaces X,Y. Let us confine to incompressible ma-
terials . In this case the yield locus is ~nbo~nded.
We set
420 J.J. Telega

~ = 1- ~ (tr1)! , trt = ti.


- 1
, I = (
-
b 1J
..) ,
and X= ! ,
! being the stress space. The yield set K is
assumed to be of the form P
Kp={!t-~1 ~ (x) E: C a.e. in /3.153/ .9.1,
where C cR6 is boWlded, closed and convex and 0 6
Let J <. p <. o.o /in the three-dimensional case/. The
c.
stress space ! consists of tensors t E IH (R) li'i th the pro-
perties, see [184] , - s
2
tijtL (5'2.), t~jteo(Sll tij,jE:LP(Q,), /3.154/

tijnj 6 lv 1 - ( 1/P) ,p (s 1) tN 6 eo(s 2) /3.155/


The norm of t 6 2 is clearly the sum of norms.
The space Y:U of displacement rates consists in pairs

/3.156/
which satisfy the condition that the map defined for
t ~ w 1 ,p (Q.) by
s
J
ij
! -.- )(v!) :tSl + (E; !}:Y: 1 /3.157/
~ s1
is continuous with respect to the m~~imum norm on ! I re-
call that BD (Sl) c (Lll11 (Q.))n , rt':n/ (n-1) in the general case,
and 11 1 ,p (.Q,)cC(.Q.) . .Horeover, the trace of a function
.:!;!, E BD(S't) belongs to [L
1 (s)] 3 Under our assumptions, /3.157/
is li'ell defined.
U is the normed space /though incomplete/ in the norm
Sl. s1
'1;!-llu = D~ llnn(S).) +II:!;! 11~ 1 - (1/r) ,p (s~*. /J.158/
To simplify notations we put .:!;!_=.:!;!,~, .:!;!,=~ a.."1.d ~=(.:!;!_, 1.:!;!,)
1
Thus u t U if and only if there is a symmetric tensor of mea-
sure~,#-:E'; M(~ ,JI>ts (R],
such that for all ! E: c 1 ,IMs (R)) (Sl
we have
~,!) = - ) ('Y.E).:!;!, + j (E; 1). 1.:!;!,
.2 s1
The principle of virtual velocities is defined in the follow-
ing way
a.~1!!)= ) tijeiJ(!!)- J~~= - ~ (v~.:!;!, + J(E;!)1.:!;!.=L(~) /J.159/
st. s2 ~ s1
Topics on Unilateral Contact Problems 421

here
L (f:!.) = (E.~ + 5 I2.0 1~ ' I 3. 160/
JS2. s1
and t E. ~ , ~ 6 U.
By definition, the limit load multiplier m1 is the ma-
ximal value of load multipliers for which there exists a
stress field !! E': K~=Kp II [ ! ~ L \~ '- 0 on S 2) , which is in
equilibrium with (m 1 ,m 1 p_0 ):

ml = suptm I 3!! (; K~ : a(!!,~)= 'illLl~) fj 6 eN) =

= sup 1 in~ a(,i,~)


t K L (u) =1
- s ~'Kv
The l~inematical formulation is obtained by changing the
sequence of inf and sup operations.
By applying Th. 3 .5 l'le conclude that
sup 1 inf a (i, u) = miz; sup a(!,~) ,
f:K L ~) =1 ,., L. t~) =1 t L K1
s )3 {: Kv ,!;! a~:v - ...- s
The proof of /3.162/ follows along the same lines as in
(187], and is omitted here.
To prove the existence of a maximum for /3o161/ we pro-
ceed in the follol'ling way. Let us define the subspace U0 of
U /since l'le deal with incompressibility/
U0 =t.J;!-Uitr{f-:)=0 in H(Q ,Ns(R~.
The convex of' plasticity or the yield set is
K =KP"t'!lf ~ L 2 <.Q)).
0
Thus K contains only deviatoric parts of stress tensors de-
fining0the yield locus.
Take now a sequence ,... ,..., J'"ikt)
mk~m 1 /from below/and let ~
satisfy k 1 k
t~ )EK0 1"\Ks , a(i ,Jl)=mkL(~) E:U0 /3.164/ f!
There exists a subsequence, still denotes by lik), such that
~he trace representatives are weak-*' convergent in L00 ( ~)
/Banach-Alaoglu theorem/. If s6K 1 is the limit, then it
satisfies the equilibrium equations- s and the boundary condit-
ion on s1.
Mathematical aspects of the theory of limit analysis
in the absence of unilateral boundary conditions are studied
in [184.,.188, 2.50, 2.51]
Let us pass to constructing of counterparts o the li-
mit analysis theorems in the case of Signorini''s conditions
422 ].]. Telega

with friction. The friction law is given by /2.22/, where


~=~ The counterparts mentioned l~ere proposed in [167].
It is worth noticing that i f the lower and upper bow~d the-
oremlJare applied in the usual manner to the boundary value
problems with friction then i t may happen that ms ';> ml~ /or
ms> mk: The approach which will nolv be presented excludes
inherently such a possibility.
Let us set
K (eN) = {!= ( t. J) \ t. . .+b. =O, ! ~ K , in S?.. ;
s J. l.JtJ J. /3.165/
tijnj:msp~ ,on S1; tN !S 0, !T ~ K (eN) ,on s 2)
where K is the global convex of plasticity and K (e~~') is de-
fined by /2.18/.
The static approach is the follo\ving one. Let ~N be de-
fined-over-~~;-~;~-~--and g(eN,!T)~ o makes sense. ~~en we
consider the convex maximizat1on problem
m~(eN,~) = sup ms lt). /3.166/
_:E{:Ks ( 9N)
A stress field s solving /3.166/ clearly depends on eN'
that is .!=.!(eN) -
For a tensor field ~ belonging to an appropriately cho-
sen space >ve define TN(.:t.) in the following way;

JTN~!) wNdS: ~ (tijeij (w) -biw~ dx , V;v such that


s2 5?.
~\S =0, !!TIs =0. /3.167/
2 2
From /3.167/ we conclude that i f .:E satisfies the equilibrium
equations /2.53/, then TN(!)=~, on s 2 In this manner a
nonlinear operator N:SN~TN(11 has been implicity defined.
Let.! be a solution of the problem /3.166/ for some eN.
Suppose that 8N=TN(!) Then the stress field ~ is called
a static solut1on of the limit analysis in the case of Si-
gnorini*s conditions with friction.
Such a definition of the static solution is correct
since then eN:TNl~)=N(eN). Hence we infer that eN:sN is a
fixed point of the operator N. lfe observe that if sN=N (sN)
then we have the following counterpart of the lower
bound theorem
m 8 (i)~m~(sN,,!J !EKs(sN), /3.168/
where ms {!) enters the definition of the convex set K {sJ,
see /3.165/. A field 1{:-Ks(sN) is ~_:!:~t!~~!li:_~~!~:!!~!-2
If m@J is a solution or the maximization problem
Topics on Unilateral Contact Problems 423

ms li) = sup ms (.i) ,


t~K
where - s
Ks=tt= (tij)l tij,j+bi=O, !GK, in 6l.j

tijnj=msp~, on s 1 ; ~~O,g (~,!T)~o,s 21, /,3.169/


. then obviously lie have m~ (sN 1 .) ~ ms (!).
Thus non-associated friction laws diminish limit loads in
comparison with corresponding associated laws.
If sN is such that sN=N(sN)' then we define a!!~!!!-
t!=a!!l-!~!!!i2!~-v~lo2!tz_f!~l~ as an arbitrary field
! (: K, provided that the kinematical load multiplier is de-
fine'X by
D(~(Y~ +J(sN 1
~
y)-
l;!ydx-S sNvNdS
5
lq
m \SN'~)=
0

spvdS
o.
1
/,3.170/
s1- -
where J(sN,y) stands for the total power of frictional
forceso
A counterpart of the upper bound theorem is formulated
a minimization problem
~nf mk (sN ,y) /.3.171/
VE:K
- v
Obviously, we have m~ (sN,.)~ mk.(sN,i:) ,y E-Kv
It seems reasonable to define a.limi t load multiplier m
in the general case of friction, as follows:
mk. (sN ,y).
m=m~ {sN ,~) = /,3.1 ?2/
where~ is a solution of /.3.171/.
As one can see, the kinematical approach is now coup-
led with the static approach. Such conclusion is not sur-
prising since the dissipation produced by friction forces
depends not only on the tangential component of a velocity
vector but also explicit~y on S~ From the viewpoint
of applications, the method of the construction of a static
field~ is not effective. However, an iterative procedure
overcomes f~s drawback.. For this purpose let us take '~~)
function t;i l defined over s 2 and such that ~)~ 0 and K ~ "N
makes sense. In the next step we solve the convex problem
sup m~ (:t.). /.3.17.3/
!EK8 (~l)
A solut;!.on J1) of /.3.17.3/ yields 4!-)
on s 2 and hence also on
s 2 In this manner we obtain a sequence ot: maximization
problems
424 J.J. Telega

n:1,2,

s 0
in 52. t .. n .=m p. ,
~J J n ~

on S 1 ; tN~ 0 , ,iT f K ( t~- 1)) , on S 2) , /3.175/


and
K(~-'-={!Tig(~- 1 >,.!T)~o, on s 2 ) . /3.176/

Hence we have a sequence t!n)) n= 1 , 2, In the limit, if it


exists, we obtain lim -1(~=~
n...,..oo
Mathematically rigorous study of the limit analysis in
the presence of friction remains open.
Remark 35 A large class of unilateral problems lvith frict-
ion can be studied in the form of an infinite sequence of
minimum, maximum or mini-max problems. For instance, prob-
lem /2.66/ furnishes the following sequence of minimization
problems min(Fn(Y)\:y:Kd)' /3.177/
lll'here 1
.Fnly) = 2 a(:y,y)+J(sN(~n-1) ,yT)-L(x),
and u 1 is a minimizer of the functional F n- 1 In comlect-
-n-
ion with such sequence of functionals /or saddle functionals
in the case of the limit analysis/ there arises a prob1em
r
of an application of the theory of -convergence' see (176,
252] The problem remains open.
Remark 3 6. Hill [2011 writes /p. 23/: "Idealized rigid/plas-
tic response, by contrast, has not been sho1m to admit dual
potentials of any kind". It seems that Hill rediscovers an
old problem solved already some fifteen years ago by J.Ioreau
and the French School of Hechanics /Debordes, Nayrole s,
Salen~on/.
3.6. Unilateral contact problems of plasticity: elastic-
-pl.astic bodies
For such bodies no mathematically rigorous studies of
the Signorini's contact problem with friction seems to exist
in the available literature.
In[200,205,211-217,23 1) frictionless problems are stud-
ied for materials obeying Henoky's law. In fact such laws
resemble small strain, nonlinear elasticity.
Duvaut ~9) studies frictional contact problems without
Signorini's conditions provided that normal stresses are
a priori prescribed on s2.
Kuz'menko (215-219] studies unilateral contact prob-
lems without and with friction for materials obeying the
Topics on Unilateral Contact Problems 425

fo~~owing constitutive re~ation

iij = Aijk~(.:) ek~ , /3.178/


where r is an interna~ parameter. He is interested in an
evo~utTon problem on a time interval, see Section 3.2 of
the present ~eotures. However, he ignores the evolution of
the parameter .!: and thus his variational approach seems
to be doubtfu~.
lIost popu~ar are numerically motivated incremental for-
mulations [194, 197,207,208,242,243, 248] In the case of the
geometrically ~inear theory such approach can be brief~y
presented in the fol~owing way.
Let us assume that the rate eqg,ations are given by

s.j
l.. ,

j +b.l.. o, in Sl , = /3.179/
eij (._Y.) = u (i, j), in S1 /3.180/
sij l!:!)= fij (:r,2(~~ , in 56 /3.181/
u
-
= o, on S
0
, /3.182/
sijnJ = Fi,on s1 , /3.183/
~Eli; T=O, sN(vN-~J~ O,on s 2 , /3.184/
'lihere st.isthe known domain occupied by the bo!ly considered
at~iven instant of time. In practice rate~ep~aced by in-
crements; for instance we take Au instead of is a con- 'u.K
vex set, possib~y depending on internal parameter and~.
lihich are kno,m. we observe that /3.181/ ino~udes also a
class of non-associated flow laws /non-standard materia~s/.
In the usual t~Y we arrive at the principle of virtua~
ve~ooities:

I ~sfind f! Eo K
ij (!!) e ij
such that
(y -~) dx~ L (y-~) y E: K ,

tihere 52.
li) = (1:, i vi dx j Fi vidS
-h
L + /3.186/
s1
Such an approach al~ows to inclu~e the fol~owing sli-
ding rule
.!T =
G (sN ~~~~SrJ ,
/3.187/
where q are internal pa~ameters describing the process of
friction. For instance, we can consider the sliding rule of
the form
!T = -G(sN,g_'!!T)~ /3.188/
The variationa~ formu~ation of the problem /3.179/-
-/3.184/ reads:
426 J.J. Telega

find f! E K such that


J J
sij (~) eij (y-u )dx- ~T(!1)(YT-s ~ L (i-.Y) y E: K /3.189/
~ 2
provided that the condition ~T=O /on s 2 /, is replaced by
/3.187/. one
To rate problems formulated abovM"an apply the H-CD-1'1
theory of duality provided that constitutive relation has
the form /3.178/, say, while the sliding rule is given by
/3.188/. Consider the frictionless case and set K=Kd, where
Kd is /2.62/; clearly now instead of displacements we take
velocity vectors. The bilinear form /2.63/ must be replaced

bya(u,y) = -kAijkl(r,~)
eij (B) ek1(v) dx /'].189/
Assuming that the following inequality ~s satisfied
Aijkl eijekl~ o 0 eijeij , ~= (eij) E: 1'1-Is (R) , /3.190/
we can construct the Green's operator G similar'ly as in
Section 2.6. The variational inequality holds provided that
instead of s}. ,tN'~r we take their rates.
In the ~ase o~ geometrically nonlinear problems vario-
us formulations are possible due to a flexibility in the
choice of stress and strain measures and inor~nental pro-
cedures, see [207,242,243,247,248]. Consider a two-body
frictionless unilateral contact problem [207]. One of them
may be rigid. The times at the starting and finishing poin-
ts in a load increment are denoted by t and t+ At, respec-
tively. iie assume that me~ssl 2)>0 while st1>=fl/ is not ex-
cluded. The motions of the boRies are oongidered in a fixed
Cartesian coordinate system in which all kinematic and sta-
tio variables are defined. Body forces are neglected. A
vector (qi) i=1,2,'J describes the unknown increment in the
rigid body displacement of 51 from time t to t+ At. A su-
perscript ~ is used to desiknate quantities associated
with the body~ {oc=1 ,2). Comma denotes differentiation
t
with respect to the space coordinates x . The coordinates
J t
of a point of the body at time t are ( x.) while at time
t+ At \ .1
t+ At we denote them by ( XiJ The J.norements in the
displacements from time t to t+ At are defined by
au = t+ A.tx-"t;x /3.191/
Obviously, the configuration of the bodies at time t+ At
is not known. Therefore the quantities are referred to a
known configuration at time t.
The Cartesian compo11ents of the Cauchy stress tensor
t+ At
at time t+ At are denoted by s and those of the se-
cond Piola-Kirohhoff stress tensor corresponding to the
configuration at time t+ At but measured in configuration
Topics on Unilateral Contact Problems 427

at time t between

/3.192/

/3.193/
t t
since .= ... Here A. is the /second/ Piola-Kirchhof 'f'
stress increment tensor.
The Green-Lagrru~ge strain tensor constructed f'rom Ay
has the :form 1J1 (

Eij (A.!:!)=2\.(Aui),j + Auj),i+ (Auk),i (A~),J]. /3.194/


The compressive cont~ct ppessure p at the current time
t+ At is de:fined on s 2 :s~> = sf>
by the equation
p = _t+Ats~1~ t+Atn(~) t+~tJ~) , ; 3 195;
~J ~ J

where t+Atn( 1) is the unit outward normal vector to the


boundary of the body 1 at time t+ At.
Since the body 1 may undergo rigid body displacements,
and may even be rigid, therefore the equilibrium o:f the ex-
ternal :forces acting on the boundary o:f the body 1 should
be satisfied. ife have

s
lD
Fihij t+AtdS: ~ poeij t+Atn~
~
t+btdS ' /3.196/
~1 52
\~here hij ar;d ceij represent rigid body displ.aoements o:f a
point on s} )
and s 2 , respectively, in the i-th coordinate
direction due to a unit displacement in the j-th rigid body
degree o:f :freedom. They are determined only by kinematic
relations o:f the body 1. F i are the components of' the sur-
:face traction at time t+ At and t+btdS ~tands :for a,surf'ace
element at time t+ 4t.
Internal equilibrium equations for each body are
(Asij), j + [(A ui) ,J t+hs jkJ,k = o /3.197/
t
since sij,j:O.
The f'ollowing constitutive relations are assumed f'or
each body
s
ij= c:j~1Ekl' I 3198/
where Ce-p is the elasto-plastic matrix at time t and is
ijrs
428 J.J. Telega

BALL

A 1-TPiR+J~-..-----, Fig.28, Finite ele-


B ment mesh or ball
indentati on problem.
Dimensio ns in mm

1~~--------------------------,
II SIMll.ATION
z
.X 0 EXPERIMENT

1.1:10
....,
1..1
a:
0
u.. Fig.29, Load vs ra-
dius of indentati on
~ 5 or ball indentati on
1-
z problem
~
z
Topics on Unilateral Contact Problems 429

8
SIMULATION
z "
-"' HERTZ 9JlJTION
6 (ELASTIC)
w
u
a:
0
u..
4
.. Fig.JO, Load vs ba11
l.J
z
disp1acement for ba11
I- " indentation prob1em
z
w 2
Cl "
~

0 "
0.0 0.05 0.10 0.15
BALL DISPLACEMENT (mm)

2500 r - - - - - - - - - - - - - - - - - - ,
o F= 100 N
~ F= SOON
F= 2000N
~ F= 6272N
::;)
Vl
Fig.J1, Contact pres-
Vl
w sure distribution
a: for ba11 indentation
a..
prob1em
1-
u
<(
1-
z
0
u
0 ~~~~~~~~~~~~~~--~
0 10 15 28
ANGLE FROM AXIS ( CIGREES)
430 J.J. Telega

r /R

Fig.J2, Enlarged
vie1.r of displacement
profiles for diffe-
rent load levels

w /R

Fig.JJ, Plastic
enclaves at various
loads

0.6
z/R
Topics on Unilateral Contact Problems 431

determined by the yield function, flo,., la\i and hardening ru-


le.
Boundary conditions:
6E,(ot.) = 0 on

t+~t t+bt ~) ~)
sij nj = Fi on s1 /J.200/
p .>,. 0 , ~ ~0 , ~p = 0 on S 2 , /J.201/

/3.202/

/3.203/
/3.204/

/3.205/

uhere dVt stands for a volume element at time t.


Using a discretization technique, say FEM, one can ob-
tain from /3.203/ and /3.204/ a nonlinear programming prob-
lem. Such approach has been applied in [207] to solve the
axisymmetric ball indentation problem provided that the ma-
terial obeys the Prandtl-Reuss equation, the von J.fises yield
criterion and the isotropic hardening rule. The ball itself
is assumed to be rigid. The elastic modulus and Poisson's
ratioused are 72e500 J.Wa and 0.33, respectively. In the ef-
fective stress-effective plastic strain relation which is
expressed by ~ =
t;y+ ~:;0 E~ ; coy' \7'0 and n are 392 MPa, 301
HPa and 0.283, respectively.
Results of calculations are represented in Figs 28-34.
Remark 37 The reader interested in inelastic unilateral
contact problems of structural mechanics should refer to
[199, 222-226, 247]
In {.244,245] a possibility of a generalization of the
shakedown theorems [21 0, 238) to frictional boundary condi t-
ions has been conjectured. However, no precise statement was
given and the problem remains open.
432 J.J. Telega

r/R

Fig.34, Effective plastic


strain distribution at load
6272 N

4. HOMOGENIZATION OF FISSURED ELASTIC SOLIDS AND PLAT!:;S


In the case of nonhomogeneous solids and structures,
like fissured elastic plates, it is desirable to know the
overall properties. To determine such properties the me-
thods of homogenization have proved to be very useful, see
[252,254,255,256,260.261,262,266,271,272-275]. Yet these
methods are most effective in the case of peri~icity or
non-uniform periodicity.
In this chapter I shall present the problem of homo-
genization of a geometrically linear, fissured elastic solid.
Homogenization of Kirchhoff and Reissner-like plates will
also be discussed.
4.1. Homogenization of a fissured elastic solid
Primarily the problem was posed and solved by Sanchez-
Palencia [272], see also [265] He employed the tl,ro-scale
method of asymptotic expansions. Such an approach is formal,
though very useful. Next Attouch and Hurat (253] delivered
a rigorous proof of convergence. Yet they considered only
the scalar case. In the sequel the vector case is conside-
red. The study follows the main lines of the approach pro-
posed in [253].
Let us consider an elastic solid l'feakened by periodi-
cally d~stributed microfissures, see Fig.35, wheretwo-di-
mensional case is schematically represented.
Topics on Unilateral Contact Problems 433

domain rl= Q \FE


x2

\ \ \ \
"\ \ \ \ \
\ \ \ \ \
\ \ \ \
\
\

./e assume that the body is c~amped at the boundary while


the microf'issures F'E.,i are distributed periodically and do
not intersect the boundary )~. Every cell Y~ . is homo-
~,:~.

thetic to the so called basic cell Y, see Fig.)6.


The latter is damaged by a fissure F. l!e assume that I<' is
of class c1 and F:l<'C Y. Horeover \~e assi.une that YF:Y\F is
connected /F may be a sum of disjoint fissures/. The para-
meter . is positive and homogenization signifies the pas-
sage to a limit wi th this parameter / -+0/. The homogenized
solid has no fissures,but elastic properties are different
from those of the original material.
The following notation is introduced for the sum of
microfissures such that the corresponding Y - cells are
contained in the domain ~
FE =UF /4.1/
iE:I() t' 1
434 J.J. Telega

y2
n
t fj 03 Fig.J6

~ ' ' ..... r2


t
n y2 n

't
\
,[
" ...... ....
o, n t r, 02 y

Further we set sl
= ,9. \F(. The space of Y-periodic funct-
ions 1 1
I
Hper (YF) = tY= (v;i) vi 6 H (YF) , v takes equal values
at the opposite sides of
is of fundamental i.mportf]ce.
/4.2/ YJ
The Sobolev space H (YF) may be defined in two equiva-
lent manners [263]
=
H 1 ( YF) tv E: L 2 (YF)I u . 6 L 2 ( YF) \
.,1
=
=tVb H 1 (Y 1)u H 1 (Yk on t\F~ > /4.3/ I 't1 lv)= t J_v)
where~ (~=1,2) is the trace operator for functions defi-
ned on~ Clearly, the choice o f t is arbitrary1 not ne-
cessarily such as represented in Fig.J6.
For v E H 1 (YF) we define the jump across F in the fol-
lOloJ'ing way
[v] = t 2 (v)- f 1 lv) , on F, /4.4/
lie assume that on FE the Signorini's conditions are
satisfied
f.
~T=O , [~~~ 0 , s~~ 0 , s~[~B =0 on Ff. , /4.5/
where
s~ = s~ I 1 on F'
Topics on Unilateral Contact Problems 435

'\".'e put
a ~(;:,y) = aijkl eij (B) ekl (Y) dx; !!,Y f [ H 1(Q~)] 3' /4.6/

where a. "kl : Loa lQ). One could also assume that the elasti-
cities !ijkl are Y-periodic.
~;e assume that there exists constants c 0 ,c 1 such that

.. e.j~
c 0 e ~J ~
a.jkle
~
.. ekl~ c 1 e.je.j
1J 1 ~
~e~lf-:1
- s (R).
Let us introduce the closed and convex sets of con-
straints
I
KE=he [H\S"t)] 3 y=O on ':)2, [vN]~ o on Jf~. /4.8/
Then the equilibrium problem of the fissured body SG may
be formulated as the minimization problem

mintt af"(y,y)-) E_y dxl ~f. K( \ ) /4.9/


sl-
or equivalently in the form of the VI:

(~) I
find gf. E K such that
a E (E ,y-,ld)'~, L(Y.-B) -y E: K E ' /4.10/
where
L(y)= (gydx, /4.11/
)52.~
Hence we infer that the problem ~ represents an internal
Signorini's problem.In the homogenization we are interested
in a passage >vi th t to zero. One can prove the following
results.
Theorem 4. 1. Let u'i solves (P) and s~j=aijkl ekl (.). Then we
have strongly in L 2(.Q. ,R3) ,
u-u
J-
~ weakly in L 2 (&1 ,Jlvi tR~ ,
where !!. is the minimizer of the total pot~ntial energy of
the homogenized solid

mint-1-a(:y,y)-) Qydx I y (: [H~(~)] 3 ) , /4.12/


and S?.
a(J:!tY.) = itijkleij(y)ekl(y)dx; ..!:!Y ~[H~($2.)] 3 /4.13/
The ~~~~~~~!~2~-~~!!~-!~-~r~~~~!~~~!~ and its potential
is given by
ift(~)= minftSaijkl(eij+eij(~~kl+ekl(~)l:!! E: KperJ' /4.14/
YF
436 J.J. Telega

Proof. Such an operator may be constructed as follOl;s. Let


~denote the space of rigid displacements. Then each
~~[H 1 (Y\F~~J can be decomposed according to
y =y1 + , /4.18/
where ~ ~ and y 1J.. ~ in [L 2 (Y\Fnz)] J
'l'his
To extend y on Y liTe extend y 1 continuous1)')6perator is deno-
ted by 1P Thus 1i'l
1P y = It' y 1 +
Let us prove /a/. 'h'e have
111P .YII 0 y=li1P Y 1 +.rll 0 y = 111P (.Y 1 +.)11 0 y=
' ' '
~ell Y,+11 o , Y\F'lt. = cllyllo ' Y\F.nz
The first inequality in /b/ has been proved by Lene
(266). The second one is obvious since Y\F"l c YF.

F = F c F-rz Y =Y, u Y2 uL Y, n Y2 = fi
F = F~ u F~ u ( L n F-rz )

F; = F"l n Y~ I ex = 1 2
I

Fig.J7
Topics on Unilateral Contact Problems 437

where 1 1
Kper :lwE:Ii
t- per lYF) _ 1
[wN.~]) 0onFJ, /4.15/
and : Ns (H)
Horeover
s = 'd 1.,11I -a 'i

J
and
a (u. ,.l)---. ~. h 'e (.1.1)) dx /4.17/
51 .w.h
The next theorem characterizes !h~ potential
Theorem 4.2. The elastic potential 11 is of class c1, posi-
tive, strictly convex and satisfies the following proper-
ties
/i/ c 2 e.je .. ~ 1:h(e)~ c 1 e ~J
~ ~J -
.. e .. ,
~J
~t IMslR),
for some c 2 ) O.
/ ii/ For each ~ E: ~~is lR) '\..re have

~ = d .rh/'0 ~
The proofs of the above theorems are lengthy. Therefo-
re I shall only sketch the main points. The scalar case is
studied in [253]. The proof of Th. 4. 2. is given in [2721.
The convergence in the vector case is implied by the resul-
ts obtained in [277] for Reissner-like plates /two-dimensio-
nal case/.
The proof of Th.4.1 is divided into several steps:
1/ the construction of an extension operator ~~ 2j bounded-
nes of the sequence tgE~ , 3/ localization, 4/ identificat-
ion of . and .ll
1fe observe that the domain St: depends on Accor-
ding to the approach of Attouch and 1-iurat [253) we must con-
struct an extension operator crf
such that (Q ~ E: [ H~($6)] 3.
For this purpose we "enlarge" the fissure F to a hole F"l'
see Fig.37.
The parameter f12.> 0 is kept fixed, nz = nz and F c F~
The boundary of F~ is sufficiently smooth. To 0 construct
the operator~' we proceed as follows.
\;e shall first construct the extension operator~.
Lenuna 4.1. There exists an extension operator
1P: [H 1 (Y\F"l))3-..[H 1 (Y)) 3 such that
I a/ ll1P Y II 0 y~ c II Y II 0 Y\ F
' ' oz.
/b/ i~jlleijtiJ> C~))llo,Y~ c l;jlleij(Y)II o,Y\Ftrz.~ cll~(i)ll o,YF
438 ].]. Telega

The operator CQ is defineias foll01>'S


Definition 4.1. The extension operator
Q : [H 1 \YF)] .3_,.[n 1 (Y)] J
is equal to
Q = 'JPo1R J /4.19/
where 1R : [H 1 ~l:"F) J 3 --+ [n 1 ( Y\Fttz.)1 3 is the restriction opera-
tor.
The operator Q is characterized by
Lemma 4.2. The operator Q has the follolving properties
/i/ tQ Y= y on Y\F"'Z.
/ii/ IIQ Yllo y~ cll.xllo Y\'"'' ~ Cll.YIIo YF= ciiYIIo y
' ' ~ ' '
/iii/ llg~ y)JI o, y ~c II g,(Y) II o, Y\F ~ c II e ( Y) II o YF

/iv/ Y-YII 1 YF ~ c II ~(Y) Uo YF o


lli
"l. '
The proof is given in [277]
,'
Having defined and examined the operator ~ we pass to
the extension operator ~~ acting on functions determined

on 5r ive know that Fr . c Y c for every i E. I (t) and Y.,C:.)~.
c:.,~ c;.,~

is the E.Y cell corresponding to "i" or


YE. i , = e,Y + ~.,,
-~,~
~
-~,c:.
.,E.RJ
E .
Next operators (Q, ,~ are constructed, see the scheme below.
The global operator ~E is derived from QE,i by the method
E.
of stitching of the operators~ '~. The global operator Q
is obtained in the following way
~ E.(H\52.)]3
! restriction
zf.,i E: [H 1 (Y .\F 3:[H 1 ( iYF+ 1i
- ''"
3 )1 ,.)1
j
- t,~ ,i \
translation and change of.scale
~ ~ [H 1 (YF)] .3 or ~ (Y) =~E.,~ ( EY+ ~
1 1 J,
l restriction and extension -
~2 = Qz1 ( [H1 (Y)] .3
J translation and change of scale
~J:::Q 'ili2 ~ [H 1 ( Yf.' ~1.3 or !3lx) = !2( x-
! s:itching with respect to iE:I(t)
~4= (Q ~
We observe that x= Y+ ~i,f implies Y= (x- !i,~.J / Accor-
ding to our earlier assumptions the operator Qt may be set
Topics on Unilateral Contact Problems 439

equal to the identity near the boundary S of 5i. , see Lemma


4.J below. Then
~ f: [n~ (d)] 3 i> ~E ~ E [n~ (Sl) J3
where 1
H1
~ ( 1 . :1 J
(Sl) =tY: f: [n (5}; )j
l y = o on s 1J
The basic properties of the operator Q are given by
Lemma 4. J. For each . ) 0 the opera tor
(t : [H 1(st')1 J _____.. [ H 1 (51) 1 J is linear and continuous. Hore-
over lve have
c._'i g =~ on

11'2~~ o~
0 , ,J[, c\1~\1 0 , S2.'
u~(~r{~) ll 0 ,Sb ~ ell~(~) ll 0 ,.9. '
IIQf. ~ - ~11 0 ,.9. ~ c E I\~(~)\\ 0 , S?.l ,

1\v(c{~-~)\1 0 c-.. ~ c\1~(.~)1\ 0 g.t .


,u~ '
By noting that ncl~-zl\1 ,@.l ~ (c1E. +c) II ~J~)II
formulate
we can o,st
Theorem 4.J. /Korn's inequality for 5Gt /. For each
~ f: [ H~ l S't)) 3 the Korn' s inequality

,52.l .S ~E +C~ 1\ ~(~)\1 O, ~t


~~~~~ 1 /4.20/
is satisfied. Here 0 < < and is held fixed.
Remark 4.1. For the scalar 0 case a gimilar role is played by
the Poincare inequality in which the parameter also enters
expJ.ici tJ.y [25 J)
Nolv we can formulate
Theorem 4. 4. For any sequence ~ E.-O satisfying t"t
sup II yt 11 1 Sl.t < ce there exists a sequence t(Q yf. ): 0
E) 0 ' ..,
bounded in [H 1 (~) 3 and such that
lltf.r _!f.ll o,~o as - + o.
By using /4 . 9/, /4,20/ and Th.4.4 we prove that

!!~ u strongly in
si~sij weakly in
f.
l'l'here sij= aijkl ekl (\!:!E.) Next we consider the inequality

~c,<)(aijkl u (i) -aijkl 0kl(~~ 0ij (u' -4) dx'9 /4. 2 1/


440 J.J. Telega

y2
b)
y2 N F
b
y1 l1

a y1

Fig.j8
Topics on Unilateral Contact Problems 441

where lf (;; J)+ (Q.} , ' ~ IMs lR) and


Y! = E.Y(x/f.) = .f 1 (x) +f. Ye (x/e.), /4.22/

Here :r l?') is a solution of the local problem /4.14/. Per-


forming the integration by parts in /4.21/ and localizing
we arrive at the inequality
<~ (x) - o ~le) , ~u (x __g)
R 3x R 3 )/ 0 Ji ~ IHs (RJ

The maximal monotonici ty of the subdifferential ~n(l implies,


see [258, p.22]
_:?. ~x) = ) yh (%. ~~ for almost every xe $1.
To prove fi~.17/ one must employ the methods of r -con-
vergence, see [2.5 3)
Remark 4. 2. As a by-product, in [2.53] the authors show that
H1 \YF) C BV(Y), where BV stands for the space of /scalar/ in-
tegrable fLmctions v such that v . GI-1 1 (i). Similarly it c~
demonstrated that [H 1 (YF)] 3 c. BD 'lY).
1

Remark 4.J. Obviously, if friction occurs on F~ then the


homogenization problem becomes much more difficult. Rigo-
rous results are not available, yet heuristic approach is
used in [265,276). For instance, in [276] the local problem
has the form of the IVI:
for e and oc. given find w E K
- - - per
such that
5aijkl (eij (.:!!) + ei~ ekl (~-~) dy +
rF
+ J ( "!!' y-~) - J (!! '~-cc.) ~ 0 vf: K
- per
,

where J is the total work of frictional forces. The macros-


copic potential is defined as follows

WsN(~,~J= 1~1 t j aijkl(ekl(~)+ ek~ (eij(~)+eijJdy +


1"\F (
+ I ~I j d(sN (~)' [~T] -~) ds'
F
and sN (~li, ~~ is treated as a parameter in the process
of differentiation.
4.2. Co1nments on homogenization of fissured elastic plates
So far, only fissured Kirchhoff plates in bending
[267,268] and Reissner-like plates [269,277] were studied.
In the case of the Kirchhoff plates in bending we have five
442 J.J. Telega

(a)

(b)

Fig.39
Topics on Unilateral Contact Problems 443

different sets of unilateral constraints


K~ = {v {: H~ U~l) [ v] ~ 0 , on F() ,
K~ = lv~H~(Si..t) [vD,o, [6v/'dg1l =O,on :rl),
K3 = lv"H~(g.l) [v]= o, [dv/d.uj~o, on Ft:~

K~ = tv t n~(.Q.t) I [:(tv/ 'dn, ]~o , on Ft ~ ,

K~ = l v<: H~(Q.i.) [v]~o , [dv/6a.D~o, on FE ,


\ll'here 2
H1 ( ~
f.)
=lv E: H 2lS2E.) \ t (v) = 0, 't ( "'~;
;) ) :~on S)
1
Obviously now the domain SG stands for the mid-plane of
the undamaged plate.
Hence after homogenization we obtain five hyperelastic pla-
tes 1dthout fissures. In [267,268] the method of two-scales
as}~ptotic expru~sions has been used. It is interesting to
note that it was necessary to postulate the following ex-
pansion for the transverse displacement wt. of the micro-
fissured plate
t (o) 2 (2) /:
\l (x) = w (x) + E w (x, Y) + , y: x '
where ,,. (o) & H~ (S?.) and ,,.l2) is defined on Qx YF
The results obtained were applied to the plate in ben-
ding lveakened by fissures parallel to the x -axis, see
Fig.J8. Clearly, then the assumption that flssures should
not meet the boundary is violated. Consequently, the strict
convexity of the homogenized potential is lost.
Still more complicated are fissured Reissner-like pla-
tes. In this case 211 modes of unilateral fissures are ad-
mitted by the model. To have a greater flexibility the mid-
plane is shifted by e, see Fig.J9
.re observe that both Kirchhoff and Reissner-like plates ad-
mit penetration zone/1dthin the model considered/.
To avoid interpretation zones more refined models of
plates are required (or ,,.e just have to solve three-dimensio-
nal problems!).
Acknolvlede;ement. I am indebted to i'rofessorsJ.J.:Moreau and
P.D. Panagiotopoulos for their invitation to deliver these
lectures at CISM.
444 J.J. Telega

REFERENCES
IN'IRODUCTION
Armstrong, G.C., V.C.Mow: Biomechanics of Normal and
Osteoarthrotic Articular Cartilage, in: Clinical Trends
in Orthopaedics /Ed. P.D.Wilson and L.R.Straub/,
Tbieme-Stratton, New York 1982, 189-197
2. Atkinson, J.R., D.Dowson, G.H.Isaac and B.M.Wroblewski:
Laboratory wear tests and clinical observations of the
penetration of femoral heads into acetabular cups in
total replacement hip joints. II: A microscopical stu-
dy of the surfaces of Charnley polyethylene acetabular
sockets. IIIt The measurement of internal volume chan-
ges in explanted Charnley sockets after 2-16 years in
vivo and determination of wear factors, Wear, 104/198.5/,
217-224, 22.5-244. .
Baiocchi, c. and A.Capelo: Variational and Quasivaria-
tional Inequalities: Applications to Free-Boundary
Problems, John Wiley and Sons, Chichester 1984.
4. Bayada, G. and M.Chambat: On the Various Aspects of
the Thin Film Equation in Hydrodynamic Lubrication
when the Proughness Occurs, in: Application of Multi-
ple Scaling in Mechanics /Ed. P.G.Ciarlet and E.San-
chez-Palencia/, Masson, Paris 1987.
Bensoussan, A. and J.L.Lionsi Application d~Inequat
ions Variationelles et Controle Stochastique, Dunod,
Paris 1978.
6. Bernadou, M., P.Cbristel and Y.M.Crolet: Simulation
Numerique des Contraintes aux Interfaces Prothese de
Hanbhe-Os, in: Computing Methods in Applied Soienoes
and Engineering VI /Ed. R.Glowinski and J .L.Lions/,
Elsevier Science Publishers, North Holland 1984, 381-
4oo.
Boesavit, A., A.Damlamian and M.Fremond /Eds/: Free
Boundary Problems: Applications and Theory, Vol.III,
and IV, Pitman, London 198,5.
s. Brezis, H.: Probl~es unilateraux, J.Math.Pures Appl.,
.51 /1972/, 1-168.
Brow.n, T.D. and A.M.Gigioia III: A contact-coupled
finite element analysis of the natural adult hip,
J.Biomechanics, 17/1984/, 437-448.
10. Chand, R., E.Haug and K.Rim: Stresses in the human
knee joint, J.Biomechanics, 9/1976/, 417-422.
11 Chipot, M.: Variational Inequalities and Flow in Porous
Media, Springer-Verlag, New York-Berlin 1984.
12 Clech, J.P., L.M.Keer and J.L.Lewis: A model of tension
and compression cracks with cohesive zone at a bone-
cement interface, J.Biomechanical Eng., 107/198.5/, 17.5-
182.
Topics on Unilateral Contact Problems 445

Crolet,J.-M.: L'ancrage du composant cotylo1dien dans


les protheses totales de hanche: simulation exploratoire
These de Doctorat d'Etat en Sciences, Universite de
Technologie de Compiegne, 198.5.
14. Del Piero, G. and F.Maceri /Eds/: Unilateral Problems
in Structural Analysis, Springer-Verlag, Wien 198.5.
1.5 Dowson, D. and G.R.Higgisona Elasto-Hydrodynamio Lu-
brication, Pergamon Press, Oxford 1977
16. Dowson, D. and N.C.Wallbridge: Laboratory wear tests
and clinical observations of the penetration of femoral
heads into acetabular cups in total replacement.hip
joints. I: Charnley prostheses with polytetraluoro~hy
lene acetabular cups, Wear, 104/198.5/, 203~21.5
Dowson, D., B.J.Gillis and J.R.Atkinsonr Penetration
of Metallic Femoral Components into Polymeric Tibial
Components Observed in a Knee Joint Simulator, ina ACS
Symposium Series No 287, Polymer Wear and Its Control,
/Ed. Liang-Huang Lee/, American Chemical Society,198S,
21.5-228.
18. Duvaut, G. and J.L.Lions: Les Inequations en M6canique
et en Physique, Dunod, Paris 1972.
Elliot, C.M. and J.R.Ockendon: Weak and Variational
Methods for Moving Boundary Problems, Pitman, London
1982.
20. Fasano, A. and M. Primiceno /Eds/: Free Boundary Prob-
lems: Theory and Applications, Vol.I and II, Pitman,
London 198).
Friedman, A.: Variational Principles and Free-Boundary
Problems, John Wiley and Sons, New York 1982.
22. Fung, Y.C.: Biomechanics: Mechanical Properties of Li-
ving Tissue, Springer-Verlag, New York-Heidelberg-Ber-
lin 1981.
Galin, L.A.: Development of the Theory of Contact Prob-
lems in USSR, Izd.Nauka, Moskva 1976./in Russian/
24. Galin, L.A.: Contact Problems of the Theory of Elasti-
city and Viscoelasticity, Izd.Nauka, Moskva 1980./in
Russian/
Gladwell, G.M.L.: Contact Problems in the Classical
Theory of Elasticity, Sijthoff and Noordhoff, Alphen
aan den Rijn 1980
26. Hayes, w.c., L.M.Keer, G.Herrmann and L.F.Mookrosr A
mathematical analysis for indentation test's of articu-
lar cartilage, J.Biomechanics, 5/1972/, .541-551
27. Hori, R.Y. and L.F.Mockros: Indentation testa of humall
articular cartilage, J.Biomechanics, 9/1976/,2.59-268.
28. Huiskes, R.: Some fundamental aspects of human joint
replacement, Acta Orthop.Scand., Supplementum, 1980,
1-208.
Huiskes, R.: Design Fixation and St~ess Analysis of
Permanent Orthopedic Implants: the Hip Joint, in:
446 J.J. Telega

Functiona~ Behavior of Orthopedic Biomaterials, Vol.II:


Applications /Eds P.Ducheyne and G.W.Hastings/, CRC
Press Inc., Boca Ration, F~orida 1984, 121-162.
jO. Huiskes, R.: Biomechanics of Bone-Implant Interactions,
in: Frontiers in Biomechanics /Eds G.W.Schmid-SchGn-
bein, S.L.-Y.Voo and B.w.zweifach/, Springer-Verlag,
New York-Berlin-Heidelberg-Tokyo 1986, 245-262.
Huiskes, R. and E.Y.S.Chayo: A survey of finite e~e
ment ana~ysis in orthopedic biomechanics: the first
decade, J.Biomechanics, 16/198:3/, :385-409.
Johnson, K.L.: Contact Mechanics, Cambridge University
Press, Cambridge 1985.
jj. Ka~ker, J.J.: A survey of the mechanics of contact
between solid bodies, ZAMM 57/1977/, T:l-T17
Kikuchi, N. and J.T.Oden: Contact Problems in Elastici-
ty, SIAM Studies in Applied Mathematics, Philadelphia
1986.
Kinderlehrer, D. and G.Stampacchia: An Introduction to
Variational Inequalities and Their Applications, Acade-
mic Press, New York 1980.
Mow, v.c. and v.M.Lai: Mechanics of animal joints, Ann.
Rev.Fluid Mech.,11/1979/, 247-288.
Mow, V.C. and W.M.Lai: Recent developments in synovial
joint biomechanics, SIAM Review 2:3/1980/, 275-:317
Mow, Y.c., M.H.Holmes and V.M.Lai: Fluid transport and
mechanical properties of articular cartilage: a review,
J.Biomechanics, 17/1984/, :377-:394.
Murase, K., R.D.Crow.ninshie~d, D.R.Pedersen and T.S.
Changa An analysis of tibial component design in tota~
knee arthoplasty, J.Biomechanics, 16/198:3/, 1:3-22.
4o. Myers, E.R. and V.C.Mow: Biomechanics of Cartilage and
its Response to Biomechanical Stimuli, in: CartUage,
vo1.1 /Ed. B.K.Ha~l/, Academic Press 198:}.
Naumann, J.: Einf8hrung in die Theorie parabolischer
Variationsungleichungen, Teubner-Texte zur Mathematik,
Bd. 64, Leipzig 1984.
Oh, K.P.: The formulation of the mixed lubrication
problem as a generalized nonlinear complementarity
problem, Trans.ASME, J.Tribology, 108/1986/, 598-604.
Neoas, J.N. and I.Hlavaoek: Mathematical Theory of
Elastic and Elastico-Plastic Bodies: An Introduction,
Elsevier, Amsterdam 1981.
44. Panagiotopoulos, P.D.: Inequality Problems in Mechanics
and Applica tiona Convex and Nonconvex Energy Functions.
Birkhauser Verlag Boston-Basel-Stuttgart 1985.
Pinkus, Oa Tbe Reynolds centennial: a brief history of
the theory of hydrodynamic lubrication, J.Tribology,
109/1987/, 2-20.
46. Rodrigues, J.-F.: Obstacle Problems in Mathematical
Physics, North-Holland Mathematics Studies, 1:34, Am-
Topics on Unilateral Contact Problems 447

sterdam 1987.
47. Rvachev, V.L. and V.S.Protzenko: Contact Problems of
the Theory of Elasticity for Non-Classical Domains,
Naukova Dumka, Kiev 1977. /in Russian/
48. Saho, s. and S.Pal: Improvement of mechanical proper-
ties of acrylic bone cement by fibre reinforcement, J.
Biomechanics, 17/1984/, 467-478.
49. Santare, M.H., L.M.Keer and J.L.Lewis: Cracks emanating
from a fluid filled void loaded in compression: appli-
cation to the bone-~plant interface, J.Biomech.Eng.
109/1987/, 55-59
50. Selvadurai, A.P.s. and G.Z.Voyiadjis /Eds/: Mechanics
of Material Interfaces, Elsevier, Amsterdam 1986.
51 Seymov, V.M.: Dynamic Contact Problems, Naukova Dumka,
Kiev 1976. /in Russian/
52. Stormont, T.J., K.N.An, B.F.Morrey and E.Y.Chao:Elbow
joint contact study: comparison of techniques, J.Bio-
mechanics,18/1985/, 329-336.
53 Strozzi, A.: An assessment of the numerical solution
of the elastohydrodynamic problem for soft contacts,
Wear 115/1987/, 53-61.
54. Sdsskind, c.: Heinrich Hertz: Man and Scientist, inl
Contact Mechanics and Wear of Rail / Wheel Systems,
Proc.of the Int.Symp. held at University of British
Columbia, July 6-9,1982,/Eds J.Kolousek, J.Dukkipat
and G.M.L.Gladwell/, University of Waterloo Press 1983,
149-157.
55 Telega, J.J.: Variational methods in contact problems
of mechanics, Uspekhi Mekhaniki Adv. in Mechanics,
in print /in Russian/.
56. Tribology - Friction, Lubrication and Wear. Fifty Years
On, Vol.I and II, International Conference, 1-3 July
1987, London, Published for the Institution of Mechani-
cal Engineers by Mechanical Engineering Publications
Ltd, London 1987.
57 Yang, R.J., K.K.Choi, R.D.Crowninshield and R.A.Brand:
Design sensitivity analysis: a new method for inplant
design and a comparison with parametric finite element
analysis, J.Biomechanics, 17/1984/, 849-854.
CHAPER 2
58. Akagaki, T. and K.Kato: Plastic flow process of surface
layers in flow wear under boundary lubricated conditions,
Wear, 117/1987/, 179-196.
59 Aleksandrovich, A.I., B.S.Vekshin and I.N.Potapov: Ten-
sor of friction coefficients of anisotropic surfaces,
Treii,Ye i Iznos, 6/1985/, 996-1004 /in Russian/.
6o. Antoniou, s.s., A.Cameron and C.R.Gentle: The friction-
-speed relation from stick-slip data, Wear, 36/1976/,
235-254.
448 ].]. Telega

61. Armstrong, e.G. and V.C.Mow: Friction, Lubrication and


Wear of Synovial Joints, in: Scientific Foundations of
Orthopaedics and Traumatology /Eds R.o,~en, J .lf.Good-
fellow and P.G.Bullough/, William Heineman, London
1980, 223-232.
62. Asoione, L. and D.Bruno: An analysis of the unilateral
contact problem with friction of beams and plates on
an elastiehalf-spaoe, University of Calabria, Depart-
ment of Structures, Report 77, 1985.
63. Asoione, L., D.Bruno and A.G.Grimaldi: Some Statio and
Dynamical Contact Problems Between a Mindlin Plate and
an Elastic Foundation, in: Euromech 219, Refined Dyna-
mical Theories of Beams, Plates and Shells and Their
Applications, September 23-26, 1986, Kassel; Springer-
Verlag, in print.
64. Atack, D. and D.Tabor: The friction of wood, Proc.Roy.
Soc.London A246/1958/, 539-555
65. Aybinder, S.B. and E.L.Tyunina: An Introduction to The-
ory of Friction of Polymers, Riga 1978 /in Russian/.
66. Ball, J.M., J.C.Currie and P.J.Olver: Null Lagrangians,
weak continuity and variational problems of arbitrary
order, J.Funct.Analysis, 41/1981/, 135-174.
67. Banerjee, A.K.: Influence of kinetic friction on the
critical velocity of stick-slip motion, Wear, 12/1986/,
107-n6.
68. Barquins, M.: Adherence, Friction and Wear of Rubber-
Like Materials, in: Tribology - Friction, Lubrication
and Wear. Fifty Years On, Vol.I Inst.Conf.,1-3 July
1987, London, Published for the Institution of Mechani-
cal Engineers by Mechanical Engineering Publications
Ltd, London 1987, 227-238.
69. Bartenev, G.M. and Yu.V.Zelenev: Physics and Mechanics
of Polymers, Vyschaya Sohola, Moskva 1983. /in Russian/
70. Bay, N. and T. Wanheim: Real area of contact and frict-
ion stress of high pressure sliding contact, Wear
38/1976/, 201-209.
71. Bereznyakov, A.I., S.V.Ventzel, V.N.Mamayev and s.v.
Merkulova: Temperature dependence of friction force,
Problemy Trenya i Iznaschyvanya, 27/1985/, 11-13 /in
Russian/.
72. Bartha, D. and Ph.Vergne: An elastic approach to rough
contact with asperity interactions, Wear, 117/1987/,
211-222.
73 Bielski, W.R. and J.J.Telega: A note on duality for
von Karman plates in the case of the obstacle problem,
Aroh.Meoh. 37/1985/, 135-141.
74. Bielski, W.R. and J.J.Telega: A contribution to oontaot
problems for a class of solids and struotures,Arch.Mech.
37/1985/, 303-320.
75 Bikerman, J.J.: Adhesion in friction, Wear, 39/1976/,
Topics on Unilateral Contact Problems 449

1-13
76. Blok, H.: Thermo-Tribology-Fifty Years On: in:Tribolo-
gy-Friction, Lubrication and Wear, Fifty Years On, Vol.
I, Int.Conference, 1-3 July 1987, Proc.of the Institut-
ion of Mechanical Engineers, Mechanical Engineering
Publications Ltd, London 1987, 1-8.
77 Bowden, E.P. and D.Tabor: Friction and Lubrication of
Sol~ds, Vol.I Clarendon, Oxford 1950, Vol.II. 1964.
78. Buckley, D.H.: Surface Effects in Adhesion, Friction,
Wear and Lubrication, Elsevier Scientific Publ.Co.,
Amsterdam 1981.
79. Burton, R.A.: Thermal deformation in frictionally heat-
ed contact, Wear, 59/1980/, 1-20.
80. Buttazzo, G.: Semicontinuity,relaxation and integral
representation problems in the calculus of variations,
Textos e Notas 34, CMAF Universidades de Lisboa 1986.
81. Campos, L.T., J.T.Oden and N.Kikuchi: A numerical ana-
lysis of a class of contact problems with friction in
elastostatics~ Comp.Meth.Appl.Mech.Eng. 34/19~2/,
821-845. .
82. Capuzzo Dolcetta, I. and M.Matzeu: Duality for implicit
variational problems and numerical applications, Uni-
versita degli Studi, Roma, Istituto Matematico "G.Cas-
telnuovo" 6 marzo 1980, 13-49.
83. Challen, J.M. and P.L.B.Oxley: An explanation of the
different regimes of friction and wear using asperity
deformation models, Wear, 53/1979/, 229-243.
84. Ciarlet, P.G. and J.Necas: Unilateral problems in non-
linear three-dimensional elasticity, ARMA 87/1985/,
319-338.
85. Clarke, F.H.: Optimization and Nonsmooth Analysis,
John Wiley and Sons, New York 1983.
86. Cocu, M.: Existence of solutions of Signorini problems
with friction, Int.J.Eng.Sci., 22/1984/, 567-575
87. Cocu, M. and A.R.Radoslovescu: Regularity properties
for the solutions of a class of variational inequalit-
ies, Nonlinear Analysis, Theory, Methods and Appl.
11/1987/, 221-230.
88. Curnier, A.: A theory of friction, Int.J.Solids Struct.,
20/1984/, 637-647.
89. Curnier, A.R. and R.L.Taylor: A thermomechanical formu-
lation and solution of lubricated contacts between de-
formable solids, Trans. ASME, J.Lubric. Technology
104/1982/ t 109-117
90. De Celis, B.: Theoretical analysis of dry friction in
brittle and ductile materials, Wear, 116/1987/,287-298.
91 Demkowicz, L.: On some results concerning the reciprocal
formulation for the Signorini's problem, Comp.and Math.
with Appl., 8/1982/, 57-74.
92. Demkowicz, L. and J.T.Oden: On some existence and
450 J.J. Telega

uniqueness results in contact problems with nonlocal


friction, Nonlinear Anal.Theory, Methods Applic~6/1982/
1075-109,3.
9.3. Dowson, D.: The History of Tribology, Longmans, London
1978.
94. Dowson, D., A.Unsworth, A.F.Cooke and D.Gvozdanovic:
Lubrication of Joints, in: An Introduction to the Bio-
mechanics of Joints and Joint Replacement, /Eds D.Dow-
son and V.Wright/, Mechanical Engineering, London 1981,
120-145.
95 Duvaut, G.: Equilibre d'un solide elastique avec con-
tact unilateral et frottement de Coulomb, C.R.Acad.Sci.
Paris, A290/1980/, 26.3-265.
96. Duvaut, G.: Elasticite avec frottement visqueux, c.R.
Acad.Sci., Paris, A291/1980/, 511-514.
97. Ekeland, I. and R.Tem~: Convex Analysis and Variation-
al Problems, North Holland, Amsterdam 1976.
98. Endo, T., J.T.Oden, E.B.Becker and T.Miller: A numeri-
cal analysis of contact and limit-point behavior in a
class of problems of finite elastic deformation, Comp.
and Structures, 18/1984/, 899-910.
99 Felder, E.F.: Formulation thermodynamique des inter-
actions superficielle entre deux corps, J.Mec.Theor.
Appl., 4/1985/, 283-JOJ.
100. Funabashi, K. and T.Nakamura: Microscopic normal dis-
placement of contacting bodies with tangential loads,
Wear, 114/1987/, JJ9-J54.
101. Fusciardi, A., U.Mosco, F.Scarpini and A.Schiaffino:
A dual method for the numerical solution of some varia-
tional inequalities, J.Math.Anal.Applic. 40/1972/,
471-49.3
102. Galanov, B.A.: On nonlinear boundary equations of con-
tact mechanics of rough surfaces, Prikl.Mat.Mekhanika,
50/1986/, 470-474.
10,3. Gaylord, E.W. and H.Shu: Coefficients of static frict-
ion under statically and dynamically applied loads,
Wear, 4/1961/, 401-412.
104. Giltrow, J.P. and J.K.Lancaster: Friction and wear
properties of carbon fibre-reinforced metals, lvear,
12/1968/, 91-105.
Goldstein, R.V., A.F.Zazovskij, A.A.Spektor and R.P.
Fedorenko: Solutions of three-dimensional rolling prob-
lems with slip and adhesion by variational methods,
Uspekhi Mekhaniki /Advances in ~fechanics/, 5/1982/,
61-102. /in Russian/
106. Greenwood, J.A., K.L.Johnson and E.Matsubara: A surface
roughness parameter in Hertz contact, Wear, 100/1984/,
47-57
107. Halaunbrenner, M.: Directional effects in friction,
Wear, J/1960/, 421-425.
Topics on Unilateral Contact Problems 451

108. Halling, J.: A contribution to the theory of friction,


Wear, 37/1976/, 169-184.
109. Haslinger, J. and !.Hlavacek: Approximation of the Si-
gnorini problem with friction by a mixed finite ele-
ment method, J.Math.Anal.Applic. 86/1982/, 99-122.
110. Haslinger, J. and P.D.Panagiotopoulos: The reciprocal
variational approach to the Signorini problem with
friction. Approximation results, Proc.Roy.Soc.Edinb.
98A/1984/, 365-383.
111. Haslinger, J. and M.T~rdy: Approximation and numerical
solution of contact problems with friction, Apl.Mate-
matiky, 28/1983/, 55-71
112. Higginson, G.R.: Elastohydrodynamic lubrication in hu-
man joints, Proc.Inst.Mech.Eng., 191/1977/, 217-223.
113. Hayer, M.w., E.C.Klang and D.E.Cooper: The effects of
pin elasticity clearance and friction on the stresses
in a pin loaded orthotropic plate, J.Comp.Materials,
21/1987/, 190-206.
114. Ivanov, A.P.: On well-posedness of the fundamental
problem in systems with friction, Prikl.Mat.Mekh.,
50/1986/, 712-716. /in Russian/
115. Jarusek, J.: Contact problems with bounded friction,
Coercive case, Czech.Math.J., 33/1983/, 237-261; Se-
micoercive case, ibid, 34/1984/, 619-629.
116. Kalker, J.J.: Mathematical models of friction for con-
tact problems in elasticity, Wear, 113/1986/, 61-77
117. Kalker, J.J.: The principle of virtual work and its
dual for contact problems, Ing.-Archiv., 56/1986/,
453-467.
118. Kanatani, K.-I.: A theory of contact force distribution
in grannular materials, Powder Technology, 28/1981/,
167-172.
119 Kayaba, T., K.Kato and K.Hokkirigawa: Theoretical ana-
lysis of the plastic yielding of a hard asperity sli-
ding on a soft flat surface, Wear, 87/1983/, 151-161.
120. Kikuchi, N.: A Class of Signorini's Problems by Reci-
procal Variational Inequalities, in: Computational
Techniques for Interface Problems, AMD, Vol.30 /Eds
K.C.Park and D.K.Gartling/, Am.Soc.Mech.Eng., New York
1978, 135-153
121. Kikuchi,N.: Beam bending problems on a Pasternak foun-
dation using reciprocal variational inequalities,
Quart.Appl.Math., 37/1980/, 91-108.
122. Kikuchi, N.: Friction Contact Problems by Using Penal-
ty Regularization Methods, in: Contact Mechanics and
Wear of Rail/Wheel Systems, /Eds J.Kolousek, R.V.Du-
kkipati and G.M.L.Gladwell/, University of Waterloo
Press, 1983, 37-59
123. Kikuchi, N. and J.T.Oden: Contact Problems in Elasti-
city, SIAM, Philadelphia 1986.
452 J.J. Telega

124. Kishida, H. and M.Uesugi: Tests o the interace bet-


ween sand and steel in the simple shear apparatus,
Geotechnique, 37/1987/, 45-52.
Klarbring, A.: Contact problems in linear elasticity.
Friction laws and mathematical programming applicat-
ions, Linkftping Studies in Science and Technology, Dis-
sertations No 133, 1985, Lin~ping University, Sweden.
126. Klarbring, A. : The Inluence o Slip Hardening and
Interace Compliance on Contact Stress Distributions.
A Mathematical Programming Approach, in: Mechanics o
Material Interfaces /Eds A.P.S.Selvadurai and G.Z.
Voyiadjis/, Elsevier Science Publishers, Amsterdam
1986, 43-59
127. Klarbring, A.: General contact boundary conditions and
the analysis of frictional systems, Int.J.Solids Struc-
tures, 22/1986/, 13771398.
128. Klarbring, A.: A mathematical programming approach to
three dimensional contact problems with friction,Comp.
Meth.Appl.Mech.Eng., 58/1986/, 175-200.
129. Kleiber, M. and B.Raniecki: Elastic-Plastic Materials
at Finite Strains, in: Plasticity Today. :Modelling,
Methods and Applications /Eds A.Sawczuk and G.Bianchi/
Elsevier Applied Science Publishers, London-New York
1985, 3-46.
oo. Kravchuk, A.S.: A contribution to the theory of contact
problems with taking into account o riction on the
surace of contact, Prikl.:Mat.Mekh., 44/1980/,122-129
/in Russian/.
Laurent, P.J.: Approximation et Optimisation, Herrmann,
Paris 1972.
1.32. Leszek, W. and W.Zwierzycki: Modern interpretation of
riction and wear notions - Part I: General description
of solid riction, Trybologia, 18/1987/, 4-7 /in Polish/.
133 Licht, c.: Un problema d'elasticite avec frottement
visqueux non lineaire, J.Mec.Theor.Appl. 4/1985/,15-26.
1.)4. Ling, F.F. and S.L.Pu: Probable interface temperatures
of solids in sliding contact, Wear, 7/1964/, 23-34.
135. Ludema, K.C. and D.Tabor: The riction and visco-elas-
tic properties of polymeric solids, Wear, 9/1966/,
329-348.
136. Mackie, A.G.: Complementary variational inequalities,
IMA J.Appl.Math., 36/1986/, 293-305
1.37 Madakson, P.B.: The rictional behaviour of materials,
Wear, 87/1983/, 191-206.
138. Martins, J.A.C. and J.T.Oden: A numerical analysis o
a class o problems in elastodynamics with friction,
Comp.Meth.Appl.Mech.Eng., 40/1983/, 327-360.
139 Micha1owski, R. and Z.Mr6z: Associated and non-associa-
ted sliding rules in contact riction problems, Arch.
Mech.Stos., 39/1981/, 259-276.
Topics on Unilateral Contact Problems 453

140. Moreau, J.J.: La Convexite en Statique, in: Analyse


Convexe et ses Applications /Ed. J.Aubin/, Lecture No-
tes in Economics and Mathematical Systems, Vol.102,
Springer-Verlag, Berlin 1974, 141-167.
141. Moreau, J.J.: Un formulation du contact a frottement
sec; application au calcul numerique, C.R.Acad.Sci.,
Paris, Serie II, 302/1986/, 799-801.
142. ~losco, u.: Dual variational inequalities, J .Math.Anal.
Appl., 40/1972/, 202-206.
143. Mosco, U.: Implicit Variational Problems and Quasi-
Variational Inequalities, in: Nonlinear Operators and
the Calculus of Variations, Lecture Notes in Math.,
Vol.543, Springer-Verlag, Berlin 1976, 83-156.
144. Mow, v.c. and A.F.Mak: Lubrication of Diarthroidal
Joints, in: The Bioengineering Handbook /Eds R.Skalak
and S.Chein/, McGraw Hill Publishers, New York 1984.
145. Neoas, J., J.Jarusek and J.Haslinger: On the solution
of the variational inequality to the Signorini problem
with small friction, Boll.Unione Mat.Ital.,17-B/1980/,
796-811.
146. Nowacki, w.: Theory of Elasticity, Panstwowe wydawnic-
two Nauko,,e, Warszawa 1970 /in Polish/.
147. Oden, J.T. and J.A.C.Martins: Model of computational
methods for dynamic friction phenomena, Comp.Meth.
Appl.Mech.Eng., 52/1985/, 527-634.
148. Oden, J.T. and E.B.Pires: Nonlocal and nonlinear frict-
ion laws and variational principles for contact problems
in elasticity, Trans.ASME, J.Appl.Mech., 50/1983/,
67-76.
149. Oden, J.T. and E.B.Pires: Algorithms and numerical re-
sults for finite element approximations of contact
problems with non-classical friction laws, Comp.Struct.,
19/1984/, 137-147.
150. Oden, J.T., E.B.Becker, T.L.Lin and L.Demkowicz: For-
mulation and Finite Element Analysis of a General Class
of Rolling Contact Problems with Finite Elastic Defor-
mations, in: Mathematics of Finite Elements with Appli-
cations /Ed. J.R.Whiteman/, Academic Press 1985.
151 Pires, E.B. and J.T.Oden: Analysis of contact problems
with friction under oscillating loads, Comp.Meth.Appl.
Mech.Eng., 39/1983/, 337-362.
152. Pozharytskii, G.K.: Extension of Gauss principle to
systems with dry friction, Prikl.Mat.Mekh., 25/1961/,
391-406. /in Russian/
153 Rabier, P., J.A.C.Martins, J.T.Oden and L.Campos: Exis-
tence and local uniquenes of solutions to contact pro-
blems in elasticity with nonlinear friction laws, Int.
J.Eng.Sci., 24/1986/, 1755-1768. ~nd.
154. Raniecki, B.: Thermodynamic Aspects of Cyclidi:Monotone
Plasticity, in: The Constitutive Law in Thermoplasti-
454 J.J. Telega

city /Ed. Th.Lehmann/, Springer-Verla g, Wien 1984,


251-321.
155 Richardson, R.S., Nolle H.: Surface friction under
time-dependent loads, liear, 37/1976/, 87-101.
156. Rigney, D.A. and J.P.Hirth: Plastic deformation and
sliding friction of metals, Wear, 53/1979/, 345-370.
157 Rockafellar, R.T.: Convex Analysis, Princeton Univer-
sity Press, Princeton 1970.
158. Rockafellar, R.T.: Integral Functionals, Normal Inte-
grand& and Measurable Selections, in: Nonlinear Opera-
tors and Calculus of Variations, Lecture Notes in Ma-
thematics, Vol.543, Springer-Varla g, Berlin 1976,
157-207.
159 Rothert, H., Heidelberger, W.Jacobi and L.Niemann: On
geometrically nonlinear contact problems with friction,
Comp.Meth.Ap pl.Mech.Eng.,5 1/1985/,1J9-155
160. Simonov,I.V.: On behaviour of solutions of dynamical
problems in the vicinity of boundary of contact zone
of elastic bodies, Prikl.Mat.Mek h., 51/1987/, 85-94.
/in Russian/
Simons, J.W. and P.G.Bergan: A finite element formu-
lation of three dimensional contact problems with slip
and friction, Comp.Mech., 1/1986/, 153-164.
Spector, A.A.: Variational methods in three dimensional
problems of transient interaction of elastic bodies
with friction, Prikl.Math.Mek h. 51/1987/,76-83 /in
Russian/.
Suh, N.P. and H.-C.Sin: The genesis of friction,Wear,
69/1981/, 91-114.
Tabor, D.: Friction- the present state of our under-
standing, Trans.ASME, J.Lubr.Techn., 10J/1981/,169- 179
Tabor, D.: Friction and Wear- Development over the
Last Fifty Years, in: Tribology - Friction,Lubr ication
and Wear, Fifty Years On, Vol.I, Int.Conf.1-3 July
1987, Proc.of the Institution of Mechanical Engineers,
Mechanical Engineering Publications Ltd, London 1987,
157-172.
166. Tabor, D. and D.E.Wynne Williams: The effect of orien-
tation of the friction of polyetrafluoro ethylene,
liear, 4/1961/, 391-400.
Telega, J.J.: Limit analysis theorems in the case of
Signorini 's boundary conditions and friction , Arch.
Mech., 37/1985/, 549-562.
Telega, J.J.: Variational Inequalities in Contact Pro-
blems, in: Mechanics of Contact of Solid Bodies, Osso-
lineum, in press /in Polish/.
Trip,J.H., L.G.Houpert, E.Ioannides and A.A.Lubrecht:
Dry and Lubricated Contact of Rough Surfaces, in: Tri-
bology-Frictio n, Lubrication and Wear. Fifty Years On,
Vol.I, Int.Conf.,1-J July, Proc.of the Institution of
Topics on Unilateral Contact Problems 455

170. Uesugi, M. and H.Kishida: Frictional resistance at


yield between dry sand and mild steel, Soil and Foun-
dations, 26/1986/, 139-149.
171. Vovkushevskii, A.V.: On variational formulation of the
Signorini's problem with friction, Mekh.Tv.Tela,
6/1984/, 73-78.
172. 1voo, K.L.and T.R.Thomas: Contact of rough surfaces: a
review of experimental work, Wear, 58/1980/,331-340.
173 Zmitrowicz,A: A theoretical model of anisotropic dry
friction, Wear, 73/1981/,9-39
174. Zmitrowicz, A: A thermodynamical model of contact
friction and wear: I. Governing equations, II. Consti-
tutive equations for materials and linearized theorie~
III. Constitutive equations for friction, wear and
frictional heat, Wear 114/1987/,135-168, 169-197,
199-221.
CHAPTER 3
175 Anzellotti, G. and s.Luckhaus: Dynamical evolution of
elasto-perfectly plastic bodies, Appl.Math.and Optim.,
15/1987/, 121-140.
176. Attouch, H. and R.J.-B. \vets: A convergence theory
for saddle functionals, Trans.Amer.Math.Soc.,1, 280
/1983/.
177 Arutyunyan, N.H. and V.B.Kolmanovskii: Theory of Creep
of Nonhomogeneous Bodies, Nauka, J.foskva 198 3 /in Rus-
sian/.
178. Arutyunyan, N.H. and B.A.Shoykhet: Asymptotic beha-
viour of solutions of boundary value problem of the
theory of creep of nonhomogeneous ageing bodies with
unilateral constraints, Mekh.Tv.Tela, 3/1981/, 31-48
/in Russian/.
179 Bogomolny, A.: Variational formulation of the roller
contact problem, Math.Meth.in the Appl.Sci., 6/1984/,
84-96.
180. Bogomolnii,A., G.Eskin and Zuchowizkii, Numerical so-
lution of the stamp problem, Comp.Meth.Appl.Mech.Eng.,
15/1978/, 149-159
181. Bouc, R., G.Geymonat, 1-I.Jean and B.Nayroles: Hilbertian
Unilateral Problems in Viscoelasticity, in:Applicat-
ions of Methods of Functional Analysis to Problems in
Mechanics, /Eds P.Germain and B.Nayroles/, Lect.Notes
in J.~thematics, vol.503,Springer-Verlag, Berlin 1975,
219-2.34.
182. Bouc, R., G.Geymonat, M.Jean and B.Nayroles: Cauchy
and periodic unilateral problems for ag i.ng linear
viscoelastic materials, J.Math.Anal.Appl., 61/1977/,
7-39.
183. Boucher, M.: Signorini's Problem in Viscoelasticity,
in: The Mechanics of the Contact Between Deformable
456 J.J. Telega

Bodies /Eds A.D.de Pater and J.J.Kalker/, Delft Univer-


sity Press 1975
184. Christiansen, E.: Limit analysis in plasticity as a ma-
thematical programming problem, Calcolo, 17/1980/,
41-65.
185. Christiansen, E.: Limit analysis for plastic plates,
SIAM J.Math.Anal., 11/1980/, 514-522.
186. Christiansen,E .: Examples of collapse solutions in li-
mit analysis, Utilitas Mathematica, 22/1982/,77-10 2.
187. Christiansen, E.: On the collapse solution in limit
analysis, ARMA, 91/1986/, 119-135.
188. Demengel, F.: Problemas variationneles en plasticite
parfaite des plaques, Numer.Funct.A nal.and Optim.,
6/1983/, 73-119
189. Do, c.: On the Dynamic Deformation of a Bar Against an
Obstacle, in: V~iational Methods.in the Mechanics of
Solids /Ed. S.Nemat-Nasse r/, Pergamon Press 1980,
237-241.
190. Do, c., Raupp, A. and R.A.Feijoo: The dynamics of a bar
in the presence of obstacles, Bol.Soc.Bras.M at.,11/1980/
55-78.
191. Duvaut, G.: ttude d'un problema dynamique en elasto-
viscoplasticit e et plasticite parfaite avec conditions
de frottement a la frontiere, Sc.Techn.Armem ent,
47/1973/, 219-227.
192. Duvaut, G.: Problemas de Contact Entre Corps Solides
Deformables, in: Applications of Methods of Functional
Ana~ysis to Problemas in Mechanics, /Eds P.Germain and
B.Nayroles/, Lect.Notes in Mathematics, Vol.503, Sprin-
ger-Verlag, Berlin 1976, 317-327.
193 Drozd, M.s., M.M.V~lich and Yu.I.Sidyakin: Engineer-
ing Calculations of Elastic-Plasti c Contact Deformation,
Mashinostroeny e, 1-foskva 1986 /in Russian/.
194. Endahl, N.: Elastoplastic identation problems, LinkO-
ping Studies in Science and Technology Dissertations ,
No 128, Link6ping University 1985.
195.Eskin, G.: Boundary Value Problems for Elliptic Pseudo-
differential Equations, Transl.of Math.Monograph s,AMS,
vo1.52, 1981.
196. Fredriksson, B.: Finite element solution of surface
nonlinearities in structural mechanics liith special
emphasis to contact and fracture mechanics problems,
Comp.Structur es, 6/1976/, 281-290.
197 Fredriksson, B.B.: Elastic Contact Problems in Fractu-
re Mechanics, in: Fracture 1977, vo1.3, ICF 4, lvater-
loo, Canada, June 19-24,1977, 427-435
198. Fremond, M.: Conditions unilaterales et non linearite
en calcul ala rupture, Mat.Aplic.Com p., 2/1983/,
237-256.
199 Ga~cki, A.: Elastic-plasti c beams and frames with uni-
Topics on Unilateral Contact Problems 457

lateral boundary conditions, J.Struct.Mech., 14/1986/,


5.3-76.
200. Haslinger, J. and Hlavacek I.: Contact between elastic
perfectly plastic bodies, Apl.Matematiky, 27/1982/,
27-45.
201. Hill, R.: Constitutive dual potentials in classical
plasticity, J.Mech.Phys.Solids, .35/1987/, 2.3-.3.3
202. Hludnev, A.M.: On solution of boundary value problems
for generalized equations of creep with a unilateral
condition at the boundary, Prikl.Mat.J.1ekh.,48/1984/,
44-49.
20,3. Hunter; S.C.: The rolling contact of a rigid cylinder
with a viscoelastic half space, Trans.ASME, J.Appl.
Mech., 28/1961/, 611-617.
204. Jean, l-1. and B.Nayroles: SignoriniProblem in Linear
Viscoelasticity, in: Duality and Complementarity in
Mechanics of Solids /Ed.A.Borkowski/, Ossolineum,
\vrocl.aw, 1979, 419-481.
205. Johnson, G.: An elasto-plastic contact problem, RAIRO
Anal.Numer., 12/1978/, 59-74.
206. Johnson, K.L.: Aspects of Contact Mechanics, in: T~i
bology - Friction, Lubrication and lfear, Fifty Years
On, Vol.II, Int.Conference, 1-.3 July, London, Publi-
shed for the Institution of Mechanical Engineers by
Mechanical Engineering Publications Ltd, London, 919-
9.32.
207. Joo, J.w. and B.M.Kwak: Analysis and applications of
elasto-plastic contact problems considering large de-
formation, Comp.Struct., 24/1984/, 95.3-961.
208. Kikuchi, N. and K.Skalski: An elasto-plastic rigid
punch problem using variational inequalities, Arch.
Mech., 33/1981/, 865-877
209. Kohn, R. and R.Temam: Dual spaces of stresses and
strains with applications to Henoky plasticity, Appl.
~Iath.Optim., 10/198.3/, 1-.35
210. K6nig, J.A.: Shakedown of Elastic-Plastic Structures,
Panstwowe Wydaw.niotwo Naukowe, Warszawa, Elsevier,
Amsterdam 1987.
211 Kravtchuk, A.S.: On Hertz problem for linear and non-
linear elastic bodies, Prikl.Mat.Mekh., 41/1977/,
.329-.3.37 /in Russian/..
212. Kravchuk, A.s.: Formulation of contact problem for se-
veral deformable bodies as a nonlinear programming
problem, Prikl.Mat.Mekh., 42/1978/, 466-474 /in Rus-
sian/.
21.3 Kravohuk, A.S.and V.A.Vasil'ev: Numerical methods of
solving the contact problem for linear and nonlinear
bounded elastic bodies, Prikl.Mekh., 16/1980/, 9-15,
/in Russian/.
214. Kuz'menko, V.I.: On a variational approach in the the-
458 J.J. Telega

ory of contact problems for nonlinear layered elastic


bodies, Prikl.Mat.Mekh., 4.3/1979/,89.3-901 /in Russian/
Kuz'menko, V.I.: On contact problems of the theory of
plasticity under complex loading, Prikl.Mat.Mekh.,
48/1984/, 47.3-481 /in Russian/.
216. Kuz'menko, V.I.: On unloading process in the case of
contact interaction, Prikl.Mat.Mekh., 49/198.5/, 44.5-
4.52 /in Russian/.
217. Kuz'menko, V.I.: Contact problems for an elastic-plas-
tic strip under complex loading, Mekh.Tv.Tela, 6/1985/
128-1.35 /in Russian/.
218. Kuz'menko, V.I.: On inverse contact problems of the
theory of plasticity, P~ikl.Mat.Mekh., 50/1986/, 475-
482 /in Russian/.
219 Kuz'menko, V.I.: Contact problems of plasticity with
taking account of contact friction, Trenye i Iznos,
8/1987/, 45-.52 /in Russiall/.
220. Lena, F. and G.Loppin: Sur quelques problemas de con-
tact entre corps solides deformables, Colloque Franco-
Brasilien sur les Methodes Numeriques de l'Ingenieur,
Aout 1976.
Liolios, A.A.: Upper and lower solution estimates in
unilateral viscoelastodynamics, Acta Mechanica,
66/1987/, 275-278.
222. L'vov, G.I.: Variational formulation of the contact
problem for linear elastic and physically nonlinear
shallow shells, Prikl.Mat.Mekh., 46/1982/, 841-846,
/in Russian/
22,3. L'vov, G.I.: Contact problems of creep of shallow
shells, Mekh.Tv.Tela, 5/1984, 116-124, /in Russian/.
224. Maier,G. and F.Andreuzzi: Elastic and elasto-plastic
analysis of submarine pipelines as unilateral contact
problems, Comp.Struct., 8/1978/, 421-4.31.
22!). Maier, G. and J.Munno: Mathematical programming appli-
cations to engineering plastic analysis, Appl.Mech.
Revs., .3.5/1982/, 16.31-164,3.
226. Maier, G., F.Andreuzzi, F.Giannessi, L.Jurina and
F.Taddei: Unilateral contact, elastoplasticity and
complementarity with reference to offshore pipeline
design, Comp.Meth.Appl.Mech. Eng., 17/18/1979/, 465-
495
227. McLinden, L.: A minimax theorem, preprint, Department
of Mathematics University of Illinois at Urbana -
Champaign, 1979
228. MOreau, J.J.: On Unilateral Constraints, Friction and
Plasticity, in: New Variational Techniques in ~mthema
tical Physics /Eds G.Capriz and G.Starnpacchia/, Edi-
zioni Cremonese, Roma 1974, 17.3-.322.
Nagaraj, H.S.: Elastoplastic contact of bodies with
friction under normal and tangential loading, Trans.
Topics on Unilateral Contact Problems 459

ASME, J.Tribology, 106/1984/, 519-526.


230. Naumann, J.: Periodic solutions to certain evolution
inequalities, Czech.Math.J., 27/1977/, 424-433
231. Necas, J. and I.Hlavacek: Solution of Signorini's con-
tact problem in the deformation theory of plasticity
by secant modules method, Apl.Matematiky, 28/1983/,
199-214.
232. Nguyen, Q.S.: Uniqueness, Stability and Bifurcation
of Standard Systems, in: Plasticity Today, Modelling,
Methods and Applications /Eds A.Sawczuk and G.Bianc~/t
Elsevier Applied Science Publishers, London and New
York 1985, 399-412.
233. Nowacki, w.: Theory of Creep, Arkady, liarszawa 1963
/in Polish/.
234. Oden, J.T. ru1d Lin T.L.: On the general rolling con-
tact problem for finite deformations of a viscoelastic
cylinder, Comp.Meth.Appl.Mech.Eng., 57/1986/,297-367.
235 Oden, J.T., E.B.Becker, T.L.Lin and K.T.Hsieh: Numeri-
cal Analysis of Some Problems Related to the Mechanics
of Pneumatic Tires: Finite Deformation/Rolling Contact
of a Viscoelastic Cylinder and Finite Deformation of
Cord-Reinforced Rubber Composites, in: Research in
Structures and Dynamics - 1984, NASA Conference Pub-
lication 2335, Symposium held of Washington, D.c.,
October 22-25, 1984, 297-307.
236. Panek, c. and J.J. Kalker: Three-dimensional contact
of rigid roller traversing a viscoelastic half spaoe,
J.Inst.Maths Applies, 26/1980/, 299-313.
237. Paymushin, B.N. and V.A.Firsov: Equation of nonlinear
theory of contact interaction of thin shells with de-
formable fundations of variable thickness, Mekh.Tv.
Tela, 3/1985, 119-128 /in Russian/.
238. Ponter, A.R.s., A.D.Hearle and K.L.Johnson: Applicat-
ion of the kinematical shakedown theorem to rolling
and sliding point contacts, J.Meoh.Phys.Solids,
33/1985/, 339-362.
239 Raous, M.: On Two Variational Inequalities Arising
from a Periodic Viscoelastic Unilateral Problem, in:
Variational Inequalities and Complementarity Problems.
Theory and Applications /Eds R.W.Cottle, F.Giannesi
and J.L.Lions/, John Wiley and Sons, Chichester- New
York 1980, 285-302.
240. Raous, M.: Comportement d'un solide fissure sous ohar-
~es alternatives en viscoelasticie non lineaire avec
eorouissage, J.Mec.Theor.Appl., Numero special, 1982,
125-146.
241. Raous, M.: Contacts Uni~ateraux avec Frottement en Vis-
coelastioite, in: Unilateral Problems in Structural
Analysis /Eds G.Del Piero and F.Maoeri/, Springer-
Verlag, Wien-New York 1985, 269-297.
460 J.J. Telega

242. Rolid.a, J.,


R.Bogacz and M.Brzozowski: Infinitesimal
and large strain in rolling contact problems, Ing.-
Archiv., .56/1986/, 241-2.53
Ronda, J., O.Mahrenholtz, R.Bogacz and M.Brzozowski:
The rolling contact problem for an elastic-plastic
strip and a rigid roller, Mech.Res.Comm., 13/1986/,
119-132.
244. Rydholm, G.: On inequalities and sha k .edown in contact
problems, Link.8ping Studies in Science and Technology,
Dissertations No 61, Link6ping University, Institute
of Technology, Link6ping 1981.
Rydholm, G. and B.Fredriksson: Shakedown Analysis in
Rolling Contact Problems, in: Conf.on Mechansms of De-
formation and Fracture, Sept.20-22, 1878, Lule&, Sweden.
246. Seregin, G.A.: Variational problems and evolution va-
riational inequalities in non-reflexive spaces with
applications to problems of geometry and plasticity,
Izv.Akad.Nauk SSSR, Ser.Mat., 48/1984/, 420-44.5 /in
Russian/.
Simo, J.c., P.Wriggers, K.H.Schweizerhof and R.L.Tay-
lor: Finite deformation post-buckling analysis invol-
ving inelasticity and contact constraints, Int.J.Numer.
Meth.Eng., 23/1986/, 779-800.
248. Taylor, L.M. and E.B.Beoker: Some computational as-
pects of large deformation, rate-dependent plasticity
problems, Comp.Meth.Appl.Meoh.Eng., 41/1983/,2.51-277
Telega, J.J.: Variational Methods and Extremum Prin-
ciples for Some Non-classical Problems of Plasticity&
Unilateral Boundary Conditions, Friction,Disoontinuit-
ies, in: Variational Method in Engineering /Ed. C.A.
Brebbia/, Springer-Verlag, Berlin 1985,/8-27/~/8-36/.
2!)0. Temam, R.: Mathematical Problems in Plasticity,
Gauthier-Villars, Bordas, Paris 198!).
2.51 Temam, R. and G.Strang: Duality and relaxation in the
variational problems of plasticity, J.Mec., 19/1980/,
493-.527.
CHAPTER 4
2!)2. Attouch, H.: Variational Convergence for Functions and
Operators, Pitman, Boston-London-Melbourne 1984.
Attouoh, H~ and F.Murat: Homogenization of fissured
elastic materials, Publications AVAMAC, No 85-03, Uni-
versite de Perpignan, 1985.
2!)4. Bakhvalov, N.s. and G.P.Panasenko: Averaging of Pro-
cesses in Periodic Media, Nauka, Moskva.1984, /in Rus-
sian/.
2.5.5 Bensoussan, A., J.-L.Lions and G.Papanicolaou: Asympto-
tic Analysis for Periodic Structures, North-Holland,
Amsterdam 1978.
2!)6. Bergman, D., J.L.Lions, G.Papanicolaou, F.Murat, L.Tar-
Topics on Unilateral Contact Problems 461

tar and E.Sanchez-Palencia: Les Methodes de l'Homoge-


neisation: Theorie et Applications en Physique, Edit-
ions Eyrolles, Paris 1985.
Brezis, H.: Operateur Maximaux Monotones, North-Hol-
land Publishing Company, Amsterdam 1973
Brezis, H.: Analyse Fonctionnelle, Masson, Paris 1983.
Bruch, J.C.:Coupled variational inequalities or low
from a non-symmetric ditch, Contemporary Mathematics,
11/1982/, 49-70.
260. Burrige, R., S.Childress and G.Papanicolaou /Eds/:
Macroscopic Properties o Disordered Media, Lecture
Notes in Physics, Vol.154, Springer-Verlag, Berlin
1982.
Ciarlet, P.G. and E.Sanchez-Palencia /Eds/:Applications
o Multiple Scaling in Mechanics, Masson, Paris 1987.
262. Eriksen, J.L., D.Kinderlehrer, R.Kohn and J.L.Lions
/Eds/: Homogenization and Eective Moduli of Mater-
ials and Media, Springer-Verlag, New York-Berlin-Hei-
delberg-Tokyo 1986.
263. Grisvard, P.: Elliptic Problems in Nonsmooth Domains,
Pitman, Boston-London-Melbourne 1985
264. Karlsson, T.: Wiener's criterion and obstacles proble-
ms or vector valued functions, Arkiv .Mat., 23/1986/,
315-325.
Leguillon, D. and E.Sanchez-Palencia: On the behaviour
of a cracked elastic body with /or without/ riotion,
, ', ' '
J.Mec.Theor.Appliquee, 1' I 1982
,, I , 195-209. ,
266. Lene, F.: Contribution a 1 etude des materiaux compo-
sites et de leur endommagement, These de Doctorat
d'Etat, Universite Pierre et Marie Curie, 1984.
Lewinski, T. and J.J.Te~ega: On homogenization o fis-
sured elastic plates, Mech.Res.Comm., 12/1985/, 271-
281.
268. Lewinski, T. and J.J.Telega: Asymptotic method of ho-
mogenization of fissured elastic plates, J.Elasticity,
1987.
Lewinski, T. and J.J.Telega: Homogenization of fissur-
ed Reissner-like plates, Part 1. Method of two-soa~e
asymptotic expansion; Part III. Some particular oases
and illustrative example, Aroh.Mech., 4o/1988/, in
print. Part 3 see Ref. 277
270. Lions, J.L.: Problemas aux Limites dans les Equations
aux D~rivees Partielles, Les Presses d l'Universite
de Montreal, 1962.
271 Lurie, K. and A.W.Cherkaiev: Effective characteristics
of composites and optimum structural design, Uspekhi
Mekhaniki /Advances in Mechanics/, 9/1986/, 3-81 /in
Russian/.
272. Sanchez-Palencia, E.: Non-Homogeneous Media and Vibra-
tion Theory, Lecture Notes in Physics, 127, Springer-
462 J.J. Telega

-Ver1ag 1980.
273. Sanchez-Pa1encia, E.: Homogenization Method for the
Study of Composite Media, in: Asymptotic Analysis,
Lecture Notes in Mathematics, Vo1.985, Springer-Ver1ag,
Ber1in 1983, 192-214.
274. Sanchez-Palencia, E. and A.Zaoni /Eds/: Homogenization
Techniques for Composite Media, Lecture Notes in Phy-
sics, 272, Springer-Verlag 1987.
275. Suquet, P.M.: Approach by Homogenization of Some Li-
near and Non1inear Prob1ems in Solid Mechanics, in:
P1astic Behaviour of Anisotropic So1ids /Ed.J.-P.
Boeh1er/, Editions du ~~s, Paris 1985, 77-117.
276. Te1ega, J.J.: Homogenization of fissured solids in
the presence of unilatera1 conditions and friction,
in preparation.
277. Te1ega, J.J. and T.Lewi.D.ski: Homogenization of fissur-
ed Reissner-like p1ates. Part 2. Convergence, Arch.
Mech. 40/1988/, in print. Part 1 and 3 see Ref. 269
SUPPLEMENT
278. Barros-Neto, J.: Inhomogeneous boundary va1ue prob1ems
in a ha1f space, Anna1i del1a Scuola Norma1e Superiore
di Pisa, Serie III, 19/1965/,331-365.
279 Boucher, H.: Prob1eme de Si~norini dans l'hypotheso
d'un support deformab1e, These J-eme cycle, Universi-
te de Paris VI, 1973
Bouchitte, G' Convergence et re1axation de fonction-
nelles du calcu1 des variations a croissance lineaire.
Application a 1'homogEmeisation en plasticit~ 1 Publi
cations AVAMAC, Universite de Perpignan 85-10.
281. Bouchitte, Ga: Representation integrale de fonctionne-
1les convexes sur un es~ace de mesures - I, Publica-
tions AVAMAC, Universite de Perpignan, 86-11.
282. Castaing, c. and Valadier 1-1.: Convex Ana1ysis and
Heasurable Mu1tifunctions, Springer-Ver1ag, Berlin-
Heidelberg-New York 1977
Harau:x:, A.: Non1inear Evolution Equations- Global
Behavior of So1utions, Springer-Verlag, Berlin-Hei-
de1berg-New York 1981.
284. Bouchitte, G.: Homogeneisation sur BV{.Q.) de fonction-
ne1les integrales a croissance 1ineaire. Ap~1ication
a un prob1eme d 1 ana1yse 1imite en plasticite, C.R.
Acad.Sc.Paris, Serie I, 301/1985/, 785-788.
285. Ogden, R. ~~. : Non-Linear Blastic Deformations, B11is
HorliOOd Limited Publishers,Chicheste r 198l~.
286. Da1 Haso, G. and L.Hodica: A genera1 theory of varia-
tional functionals, in:Topics in Fw1.ctional Analysis,
Scuola Normale Superiore di Pisa, Pisa 1982, 149-221.
K1arbring,A. ,Milce1ic A.and Shi11or M. :On friction pro-
b1ems with norma1 compliance,Nonlinear Anal.Theory
Meth.App1., submitted.

You might also like