Convexity Conditions and Existence Theorems

in Nonlinear Elasticity
JOHN M . BALL

Communicated by S.S. ANTMAN& C. M. DAFERMOS

Table of Contents
0. I n t r o d u c t i o n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337
1. B o u n d a r y - v a l u e p r o b l e m s of n o n - l i n e a r hyperelasticity . . . . . . . . . . . . . . . . 342
2. Technical p r e l i m i n a r i e s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347
3. Q u a s i c o n v e x i t y and the L e g e n d r e - H a d a m a r d c o n d i t i o n . . . . . . . . . . . . . . . . 349
4. Sufficient c o n d i t i o n s for q u a s i c o n v e x i t y . . . . . . . . . . . . . . . . . . . . . . . 356
5. I s o t r o p i c convex and p o l y c o n v e x functions . . . . . . . . . . . . . . . . . . . . . 363
6. Sequential weak c o n t i n u i t y of m a p p i n g s on Orlicz-Sobolev spaces . . . . . . . . . . . 367
7. Existence t h e o r e m s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
8. A p p l i c a t i o n s to specific m o d e l s of elastic m a t e r i a l s . . . . . . . . . . . . . . . . . . 388
9. A n e x a m p l e of n o n - u n i q u e n e s s ; b u c k l i n g of a rod . . . . . . . . . . . . . . . . . . 392
10. C o n c l u d i n g r e m a r k s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 398
Acknowledgement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399

O. Introduction
The purpose of this article is to present existence theorems for various
equilibrium boundary-value problems of nonlinear elasticity in one, two and
three dimensions under realistic hypotheses on the material response. Although
some of the results may be extended to cover Cauchy elasticity, we shall restrict
our discussion to hyperelastic (Green elastic) materials, that is, to elastic mate-
rials possessing a stored-energy function. We ignore thermal effects. For such
materials a typical boundary-value problem takes the form of finding a vector
field Uo: f2 -+ ~ " making the integral

I (u, f2)= S f (x, u(x), V u(x)) dx (0.1)

f2

stationary in a suitable class of functions *. Here f2 is a non-empty, bounded, open

subset of ~", n = 1, 2, 3. The integrand f will usually have the form
f (x, u, Vu)= #~(x, Vu)+c~(x, u), (0.2)
* F o r t r a c t i o n b o u n d a r y - v a l u e p r o b l e m s there will also be a surface integral term.
338 J.M. BALL

where ~ is the stored-energy function and q5 is a body force potential. In this

introduction we assume that (0.2) holds and that n = 3.
We attack this problem by using the direct method of the calculus of variations
to establish the existence of minimizers for I(u, ~2) in the class considered. Such
a programme has been successfully carried out by ANTMAN[1-83 in an important
series of papers on the existence of equilibrium solutions in various problems
arising from theories of nonlinear elastic rods and axisymmetric shells (with or
without Cosserat structure). In these papers ANTMAN emphasizes the crucial
importance of choosing hypotheses on the material response that both ensure
the success of the analysis and are reasonable physically. In his work, and in mine,
the problem of existence is inextricably linked with that of finding satisfactory
constitutive inequalities for nonlinear elasticity (cf. TRUESDELL[1]).
As an illustration, consider the effect of imposing the constitutive requirement
that ~ be convex with respect to Vu. This mathematically simple hypothesis,
when augmented with suitable smoothness and growth assumptions, ensures the
existence of minimizers for (0.1), (0.2). Existence theorems under this assumption
have been given by several authors (e.g., BEJu [1, 23, ODEN [13). Unfortunately
these results are only of mathematical interest since convexity of ~U with respect
to Vu is unacceptable physically *. Firstly, as was shown by COLEMAN& NOLL
[1] (see also TRVESDELL & NOLL [1, p. 163]) such convexity conflicts with the
requirement that ~ be objective (cf. (1.12)). Secondly, consider, for example,
a mixed displacement, dead load traction boundary-value problem for such a
material. Any equilibrium solution for such a problem must necessarily be an
absolute minimizer for I(u, f2); in particular, if a strict absolute minimizer exists
then it is the only equilibrium solution*. This fact, which is an elementary
consequence of the theory of convex functions (cf MOREAU [1], EKELAND t~
T~MAM [13), and for the truth of which ~ need not be strictly convex, rules
out the nonuniqueness essential for the description of buckling* *. Some less
restrictive condition on ~/U is therefore required.
A suitable condition, termed quasiconvexity, was introduced by MORREY[13
in a fundamental paper in 1952. ~ is said to be quasiconvex if
S ~(Xo,~o+V~(x))cSx>= ~ ~(Xo,~o)dx=m(O)~r(Xo,Fo) (0.3)
D D

holds for each fixed Xoef2, for each constant 3 x 3 matrix Fo, for each bounded
open subset D_cN3, and for all ~e@(D). Here m denotes Lebesgue measure
and N(D) is the set of all infinitely differentiable functions with compact support
contained in D. We regard quasiconvexity as a constitutive restriction on ~/f*. It
may be interpreted as follows: For any homogeneous body made from the material
found at any point of f2, and for any displacement boundary-value problem
with zero body force for such a body that admits as a possible displacement a
* For hyperelastic rods and shells ANTMAN makes certain convexity hypotheses on the stored-
energy functions. Because of the coordinates employed these hypotheses are not subject to the ob-
jections made below.
* HILL [1] was the first to observe that strict convexity with respect to lzu implies uniqueness.
** The nonuniqueness established by ANTMAN arises from the presence of lower order terms.
* In the case when ~ is not everywhere defined the condition must be slightly modified. See
section 3.
 Existence Theorems in Non-Linear Elasticity 339

h o m o g e n e o u s strain, we require that this h o m o g e n e o u s strain be an absolute

minimizer for the total energy. N o t e that if in the above we admitted for considera-
tion i n h o m o g e n e o u s bodies, or if we considered mixed displacement traction
b o u n d a r y - v a l u e problems, then the condition would be unacceptable, as we
should expect certain buckled states to have lower total energy than the h o m o -
geneous strain. As stated, however, the condition has a certain plausibility.
MORREY showed that if f ( - , u," ) is quasiconvex for every u, and if certain
continuity and growth hypotheses are satisfied, then for various boundary-value
problems there exist minimizers for l(u, 0). Conversely, if u is a minimizer for
l(u, f2) a m o n g C1(f2) functions satisfying given Dirichlet b o u n d a r y conditions,
and if Xoe(2, F o = Vu(xo), then (0.3) holds. This fact m a y be used (see T h e o r e m 3.2)
to motivate quasiconvexity by showing that it is a necessary condition for the
existence of sufficiently regular minimizers for a class of displacement b o u n d a r y -
value problems. The degree of regularity required is, however, fairly severe.
Furthermore, if ~ is quasiconvex and twice continuously differentiable, then
~/U satisfies the Legendre-Hadamard or ellipticity condition*:

t 3 u l c3u~,p 2 i 2 i # , / ~ p > 0 for all 2 , / ~ 3 . (0.4)

(It is not k n o w n whether the converse holds.) Because we have chosen to impose
quasiconvexity as a constitutive restriction, we must therefore regard the Legendre-
H a d a m a r d condition also as a constitutive restriction**.
The statement above that quasiconvexity is sufficient for existence must n o w
be qualified. In fact, MORREY'S remarkable existence theorem fails to apply
directly to nonlinear elasticity. For compressible materials his growth conditions
are too stringent; in particular, they prohibit any singular behaviour of ~,, such
as the natural condition

~(x,F)~ as detF~0. (0.5)

Moreover, his work gives no indication of h o w to treat the unilateral constraint

d e t V u > 0 #. Incompressible materials require the constraint d e t V u = l , which
also poses problems.

t Throughout this article we employ the summation convention for repeated suffices.
** This contrasts with the views expressed by TRUESDELL& NOLLI-1,p. 275], who suggested that
the Legendre-Hadamard condition should be regarded not as a constitutive restriction, but as a
stability condition. They conjectured that violation of the Legendre-Hadamard condition at a point
would lead to wave motion tending to move an elastic body from an unstable to a stable equilibrium
configuration and that this process may help explain buckling. While the violation of the Legendre-
Hadamard condition at a point may well result in certain kinds of instabilities (cf. ERICKSENI-3]), it
is by no means necessary for buckling. Indeed, in Section 9 we show that buckling can occur when
the Legendre-Hadamard condition holds everywhere. Moreover, other kinds of instabilities, such as
necking, may well be compatible with the Legendre-Hadamard condition (cf ANTMAN[6]). Another
suggestion of TRUESDELL& NOLL [i, p. 129] concerning internal buckling of a rod would, if true,
directly contradict quasiconvexity, but the behaviour described by them as implausible seems typical
of buckling.
ANTMAN'Smaterial hypotheses for rods and shells are those appropriate under the assumption
that (0.4) holds in the three-dimensional theory.
The analogous problem for rods has been studied by ANTMAN[2-5, 7, 8].
340 J.M. BALL

To overcome these difficulties we investigate in Section 6 sequential weak

continuity properties of functions, defined on Orlicz-Sobolev spaces, having
the form
O: u~--~dp(Vu(.)), (0.6)

where q~ is a continuous real-valued function defined on the set of all 3 x 3

matrices. This map is sequentially continuous from W 1' ~176 with the weak 9
topology to/2(0) with the weak topology if and only if 0 has the form

0 (u) = A + B "i (Vu),i + C~(adj Vu)i~+ D det Vu, (0.7)

where A, B~, C~,D are constants and adj Vu is the transpose of the matrix of
cofactors of Vu. When the domain of 0 is a larger Orlicz-Sobolev space, the
problem is more delicate. In this case we give various theorems guaranteeing
sequential continuity or closure of 0 relative to various weak topologies.
We combine these results with standard techniques of the calculus of
variations to establish the existence of minimizers for I(u, Y2) in various classes
of functions when ~ has the form

~tr(x, F)=g(x, F, adjF, detF) (o.8)

with g(x, ", ", ') convex for each x. We call such functions ~U polyconvex. Note
that F, adjF and detF govern the deformations of line, surface and volume
elements respectively. If ~ is polyconvex, then ~ is quasiconvex; in fact
polyconvexity is equivalent to a sufficient condition for quasiconvexity
given by MORAY. However our existence theorems are valid under weaker
growth conditions than MORREY'S. Moreover, we can handle the pointwise
constraints on det Vu mentioned above by using our sequential weak continuity
results. Since there are few known examples of quasiconvex functions that are
not polyconvex, the restriction to polyconvex functions is not serious. It appears
that neither the quasiconvexity nor the polyconvexity condition has been con-
sidered previously in the context of elasticity.
A wide variety of realistic models of nonlinear elastic materials satisfy the
hypotheses of our existence theorems. In particular, these include the Mooney-
Rivlin material and certain stored-energy functions similar to, and for incom-
pressible materials identical to, those of OGDEN [2, 3]. That these stored-energy
functions are polyconvex follows from sufficient conditions for the polyconvexity
of isotropic functions given in Section 5, where some related results are also
discussed.
Our existence theorems apply to displacement, mixed displacement traction,
pure traction, mixed displacement pressure, and pure pressure boundary-value
problems. Similar methods work for more general classes of mixed boundary
conditions. For the most part we consider only polynomial growth hypotheses,
which result in a theory based on Sobolev spaces. For stored-energy functions
of slower growth an Orlicz-Sobolev space setting is required; for brevity we
treat only the displacement boundary-value problem for such functions. An
example is given of a stored-energy function requiring this more elaborate theory.
 Existence Theorems in Non-Linear Elasticity 341

Many of the results may be extended to give new existence theorems for non-
linear elliptic systems in higher dimensions; some results in this direction are
given in BALL [2].
In Section 9 I apply existence theorems to establish nonuniqueness for the
mixed displacement, zero traction boundary-value problem for a Mooney-
Rivlin rod under compression. My main result is that nonuniqueness occurs
for sufficiently long compressed rods of arbitrary uniform cross-section. Despite
the intuitively obvious nature of this result, such nonuniqueness has not pre-
viously been established for any mixed boundary-value problem of nonlinear
elasticity.
In Sections3 and 4 the conditions of quasiconvexity, polyconvexity and
ellipticity are examined in detail. In particular, necessary and sufficient conditions
are given for polyconvexity based upon the work of BUSEMANN, EWALD &
SHEPHARD. It would be interesting to have a statical interpretation of poly-
convexity. None of these three constitutive restrictions has at present a micro-
scopic or thermodynamic motivation, in contrast, for example, to the situation
pertaining to the Navier-Stokes equations, for which the Clausius-Duhem
inequality gives conditions closely related to those sufficient for existence.
The reader interested in the constitutive theory of elasticity, but not in the
details of existence theorems, can omit Sections 2, 6 and 7 without much loss.
With the exception of the existence theorems themselves, the parts of this article
which bear most directly on the relevance of the quasiconvexity and poly-
convexity conditions to realistic models of elastic materials are Theorems 3.1,
3.2, 3.4, 4.5, 5.1, 5.2 and the discussion in Section 8.
Existence theorems for linear elasticity (of. FICHERA [1]) have a different
character from those presented here on account of the geometrical approximation
made. In particular we have no need of an analogue of KORN'S inequality. The
existence theorems of STOPPELLI [1] and VANBUREN [1] for nonlinear elasticity
(cf. TRUESDELL& NOLL [1], WANG & TRUESDELL[1]) are based on those of the
linear theory. In these theorems the inverse function theorem is used to establish
the existence and uniqueness of small solutions to boundary-value problems
with small body forces and boundary data. The material response is assumed
to be such that existence, uniqueness and regularity theorems hold for the
equilibrium equations linearized about the zero data solution. In the case of the
pure traction boundary-value problem the degeneracy of the linearized problem
forces the authors to assume that the surface and body forces possess no axis
of equilibrium. This hypothesis is unnecessary under the assumptions of our
existence theorem for the pure traction problem. Note, however, that STOPPELLI
and VAN BUREN make assumptions about the material response only for strains
close to those of the zero data solution.
This article by no means exhausts the problem of the existence of equilibrium
solutions even for hypereleasticity. The most notable shortcoming of this work
is that in the two important cases of incompressible materials and compressible
materials satisfying (0.5) I have so far been unable to show that the minimizers
whose existence has been established are smooth enough even to satisfy a weak
form of the equilibrium equations. I hope to discuss this question in later work.
 342 J.M. BALL

1. Boundary-Value Problems of Nonlinear Hyperelasticity

This section begins with a brief presentation of the basic equations of nonlinear
elasticity. We then go on to consider the various boundary-value problems for
which we later prove existence theorems. The calculations here are purely formal,
in the sense that we assume that the various quantities appearing have sufficient
smoothness to justify any operations required (such as integration by parts).
The theory of nonlinear elasticity is discussed at length in the books by GREEN
t~ ZERNA [1], TRUESDELL • NOLL ['1] and W A N G & TRUESDELL [1], and the
reader is referred to these texts when clarification is necessary.
We consider a material body N whose particles are labelled by their positions
x = ( G ) = ( X l , x 2, x3) with respect to a rectangular Cartesian co-ordinate system
in a reference configuration (2 which is a bounded open subset of ~3. f2 need not
be homeomorphic to an open ball. In a given motion the position of the particle x
at time t is denoted u(x, t)*. The deformation gradient F is defined by

F = lZu; Fj= Ox" =u i,,. (1.1)

We suppose that u: f2 ~ ~f3 is orientation-preserving and locally invertible, so that

J = detF > 0. (1.2)
Consideration of the stronger requirement that u be globally one-to-one is
beyond the scope of this article,
The symmetric, positive-definite right and left stretch tensors U, V and the
right and left Cauchy-Green tensors C, B are defined by
C=U2=FTF, B = V 2 = F F T. (1.3)

The following relations hold:

F = RU = VR, V= RUR r, (1.4)
where R is the orthogonal rotation tensor. The eigenvalues vl, vz, v 3 of U and V
are positive and are termed the principal stretches of the deformation. The principal
invariants of B and C are given by

= iIc = + 1)I + 1) ,2 (1.5)

i i i n = i i i c = 1)12 1)2/)~
2 9
We suppose that at each particle x the material of the body is elastic, so that a
constitutive equation of the form
T~(x, t)= •R(F(x, t), x) (1.6)
holds, where TR denotes the first Piola-Kirchhoff stress tensor. TR is related to
the Cauchy stress tensor T by
Ta= JT(F-1) r. (1,7)
9 We choose the notation common in partial differential equations rather than that of continuum
mechanics where our x, u are customarily denoted X, x respectively.
 Existence Theoremsin Non-LinearElasticity 343

The surface traction tR measured per unit area in the reference configuration,
and the actual stress vector t measured per unit area of the deformed configura-
tion, are given by
t R = TR N ; t= Tn (1.8)

respectively, where N and n denote the unit outward normals to the boundaries
~?~2and 0u(~2, t) respectively.
The pointwise form of the balance laws of linear and angular momentum are
given by
Div TR + PR b = PR ii, (1.9)
T= T ~r, (1.10)
where
(Div TR)i %f OT~,(x)
~x ~ ,
pa(x) is the density in the reference configuration, and b is the body force per
unit mass.
Throughout this article we assume that the material is hyperelastic; i.e., there
exists a real-valued stored-energy function ~tU(x, F) such that

r, ~ - . (1.11)
R , - 0Fj
~r is objective if and only if
~(x, Qr)= ~(x, r) (1.12)

for all proper orthogonal matrices Q. If ~ is objective then

~(x, F)= ~(x, ~), (1.13)

and it follows from (1.11) that (1.10) is satisfied identically. ~ is isotropic if

and only if ~ is objective and
~r Q F Q r) = ~U(x, F) (1.14)

for all orthogonal matrices Q. In this case

YU(x, F) = 4'(x, v1, rE, v3), (1.15)

where 9 is symmetric in the vi.

Deformations of incompressible materials are restricted by the pointwise
constraint
J = d e t F = 1. (1.16)
For incompressible materials the above theory has to be modified by replacing T
by the extra stress
TE=T+pl, (1.17)

where p is an indeterminate hydrostatic pressure. The stored-energy function

~/U for an incompressible material need be defined only for F satisfying (1.16).
344 J.M. BALL

We shall be concerned only with equilibrium configurations of N. If ~ is

objective we see from (1.9) that u is an equilibrium configuration if and only if

A~f u~a + qi + p. bi=O, (1.18)

where
A~f (x, 11~)d~'ef~2 ~/'(X, F) /~2 ~/'(x, F)
aFi c3Fd ' qi= ax ~ c~F~ (1.19)

A. The mixed displacement-traction boundary-value problem

for a compressible material
In this problem we seek u satisfying (1.18) in O and satisfying the boundary
conditions
u(x)=~(x) for x~00 x, (1.20)
tR(X)=-iR(X) for x~002, (1.21)

where 0 0 = 0f21 w 002, 0f21 c~ 0f22 = ~b, and ~: 0~21~ ~3, -/R: 002 ~ ~3 are given
functions. The boundary condition (1.21) is a condition of dead loading, i.e.,
the loads acting on u(aO2) have fixed direction and fixed magnitude per unit
area of 0g22 . If 002 = ~b then we have a pure displacement boundary-value problem,
while if dr21 = ~b we have a traction boundary-value problem.
Suppose that the body force b is conservative, so that

b = - grad ~P, (1.22)

where ~ = ~(u) is a real-valued potential, and where

.def Ollr/
(grad ~)i = (~Ui '
Define fl (x, u, F) by

fx(x, u, F)= ~/U(x, F)+ PR(X) tP(u). (1.23)

Consider the functional

Jo (u)d-~-'eIf fl( x, u(x), F(x)) d x - ~ u(x)"tR(X) dS. (1.24)

12 ~I22
Let 0f214=~b. Then a standard formal calculation shows that Jo(uo) is
stationary with respect to u satisfying (1.20) if and only if the Euler-Lagrange
equations (1.18) and the natural boundary conditions (1.21) hold, i.e., if and
only if u 0 is a solution to the mixed boundary-value problem.
If 001 = q~ then in general solutions to the boundary value problem will not
exist, since a necessary condition for a function u o to render Jo stationary subject
to (1.21) is that
a =0 (1.25)
where
def 1
a = m~-(~ pRb(u~ +e~-iRdS)" (1.26)
 Existence T h e o r e m s in N o n - L i n e a r Elasticity 345

Condition (1.25) says that the total force on the body due to external loads is
zero (cf. TRUESDELL& NOLL [1, p. 127]). To describe the effect of this condition
we consider two situations corresponding to different types of existence theorems
proved in Section 7.
1. b(u) is not a constant vector. In this case, under suitable hypotheses on b
the set of functions u 0 satisfying (1.25) will be nonempty, so that under certain
conditions* it is likely that a function u 0 such as to render Jo stationary subject
to (1.21) exists. If so, then u 0 is a solution to the traction boundary-value problem.
2. b(u)=bo, b o constant. In this case a is independent of u o, so that (1.25)
is a condition on the data of the problem. It proves convenient to consider Jo(U)
as a functional defined on functions u satisfying the constraint
S u dx = e, (l.27)

where e is an arbitrary constant vector. The constraint (1.27) removes the

indeterminacy resulting from a possible rigid-body translation of u(Q). Jo is
stationary at u = u 0 subject to (1.27) if and only if (1.21) holds and

Div TR + PR bo = a. (1.28)
To prove this note first that if (1.21) and (1.28) hold, then

~ J~176 d~176+ r v) = - S a . v dx + (tR-~R) dS, (1.29)

9= 0 1"~ 0R
which is zero if ~ v dx =0. The converse statement is a direct application of the
a
multiplier rule for isoperimetric problems with a playing the r61e of the Lagrange
multiplier corresponding to the constraint (1.27). A rigorous proof may also
be constructed by using a result of SCHWARTZ [-1, p. 59].
If a----0 then u o is an equilibrium solution. If a ~e0 then
t2
u(x, t)de--~fUo(X) + ~ a (1.30)

is a solution to the dynamic traction boundary-value problem (1.9), (1.21).

Note that any equilibrium solution u must also satisfy the zero moment
condition
~uApRbodx+ ~ uAtRdS=O, (1.31)

This condition, unlike (1,25), depends explicitly on the unknown function u,

and so cannot be imposed a priori.

B. The mixed displacement pressure boundary-value problem

for a compressible material
In this problem we seek u satisfying (1.18) in ~ and satisfying the boundary
conditions
u(x)=~(x) for X E ~r'~l , (1.32)
(Tn)(x)=-prn for x~?Or ( r = 2 ..... M), (1.33)
t if ~ takes the form of a powerful potential well, for example.
346 J.M. BALL
M
where •f2= U ~ , , 0 f 2 ~ f 2 1 = ~ b (k+-l), ~: ~f21~o~3 is a given function, and

Pr(r=2 ..... M) are constant pressures. We assume that for r > 2 dr2, is either
a closed surface or is bounded by a closed curve lying in t?f21. Suppose also that
there exists a C1(0) function p: O--,N taking the value p, on ~3f2r for each
r = 2 , ...,M.
Consider the functional
J~(u)= S fz(x, u(x), F(x)) dx, (1.34)
12
where
f2 (x, u, F)~f fa (x, u, F) + p J + ~ p, reijkeea"F~F~ u k, (i .3 5)
and where J is defined in (1.16). By the divergence theorem
Jl( il)= S L dx-~- S 1~p eij k epqr u*, pU~,qukN, dS. (1.36)
f~ Of2
Suppose now that ~f2a4=q~. Then for v satisfying v--0 on ~f21 we obtain
d
J~(u+ev) I = - S(DiVTR+PRb).vdx+ ~ t , . v d S
de .~=o ~ m
+ ~ 89 pUj qvkNrdS
m (1.37)
- dS
Of2
- ~ ~(ep"rp, qN,)(eijkui,,ukv ~) dS.
~f2

The fourth integral in (1.37) is zero by Kelvin's theorem applied to each 0f2,,
and the last integral is zero since Vp/x N = 0 on ~f2r. Thus

d J l ( U + e V ) = o = - S(DivTR+PRb).vdx + ~ (t+pn).vdS. (1.38)

de ~ o,<o~
Thus Y~(Uo) is stationary if and only if u o is a solution to the mixed displacement
pressure boundary-value problem. The calculation above is a slightly simplified
version of that of SEWELL [-1]; see also BEATTY [1].
If Of2~= ~ then we proceed in a fashion similar to A.
The functionals that we study in later sections include functionals of the
form do and ./1. For the purposes of the existence theorems it is not necessary
to assume that ~ is objective. If, however, this assumption is not made, the
resulting minimizers, if smooth, will not necessarily satisfy (1.10).
For incompressible materials the admissible functions are restricted by the
additional constraint (1.16).
Finally in this section we discuss briefly the analogous problems of one and
two-dimensional hyperelasticity. These problems arise from special deformations
of three-dimensional elastic bodies. In two-dimensional plane strain we consider
deformations having the form

U = (t/l(Xl, X2) , U2 (X1, X2) , ,~X3) ). >0 constant. (1.39)

 Existence T h e o r e m s in N o n - L i n e a r Elasticity 347

For such a deformation

1 ,2
F = H 1
U2
,2 (1.40)
0
and the incompressibility condition is
2(u~1 U2,z-Ul, zu20= 1. (1.41)
In one-dimensional uniaxial strain we let
u = (u 1 (xl),/~ x2, )~x3), (1.42)
where # and 2 are positive constants. The corresponding deformation gradient is

F= ~t . (1.43)
0
We leave to the reader the routine task of formulating the relevant boundary-
value problems in these two cases.

2. Technical Preliminaries
For ease of reference we list here various well known spaces used in this
article. As general references we cite FRIEDMAN [-I], KRASNOSEL'SKII & RUTICKII
[1], SCHWARTZ[1] and for functional analytic aspects DUNFORD & SCHWARTZ[1].
Throughout this work (2 denotes a nonempty, bounded, open subset of ~m
with Lebesgue measure dx, where, except in Section 3, m takes the value 1, 2 or 3.
C~(f2) denotes the space of infinitely-differentiable real-valued functions defined
on f2. ~(f2) consists of those elements of C~(O) with compact support contained
in O. We give 9(f2) the strict inductive limit topology of SCHWARTZ. The dual
space of ~(12) is denoted ~'((2), and its elements are called distributions. Any
locally integrable function f defines a distribution Ty through the equation
Ty(qS)-- ~ fq5 dx, qS~(f2). A sequence T~~ T in Y(f2) if and only if T~(q~)~ T(q~)
f2
for all ~b~(f2). (Here and throughout this work we consider only convergence
~T
of sequences, rather than nets.) If T ~ ' ( f 2 ) , then we may define ~ - by

~x ~ (~)= - T ~

c~T
Then T~--~-~x
~ maps ~'(f2) into itself and is continuous.
C(f2), C1(~) denote respectively the spaces of continuous and continuously
differentiable real-valued functions defined on f2 with the usual supremum
norms. ~+ denotes the non-negative real numbers, ~ the extended real line
with the usual topology. If E is a subset of ~ then CoE denotes the convex
hull of E.
The spaces LP(f2), l<p__<oo, of (equivalence classes of) integrable real-
valued functions are defined in the standard way. The Sobolev space
W~'P(O), 1 <p<oo, consists of those functions u belonging to LP(f2) with weak
348 J . M . BALL

~U
derivatives ?x-~~- (1 < ~ < m) belonging to LP(f2). W I' v (f2) is a Banach space under
the norm
m 13
IlUllw~,~(~)= IlulIL~(~)+ ~ ~ 9 (2.1)
a=l t~ LP(~)

The closure of N(f2) in WI'P(f2) is denoted Wol'V(f2). W~o

1,1
~ (f2) denotes the space
~u
of functions u which together with their weak derivatives 0x~~- ( l < ~ < m ) are
locally integrable. When dealing with the spaces in this paragraph we assume
that p < ~ unless otherwise stated.
Weak and weak 9 convergence of sequences are denoted by ~ and *-,
respectively. In the case of the Banach space W ~' ~(f2) we define the weak 9 topology
to be that induced by the natural imbedding of W 1' ~(f2) in the product space
(L~(f2))I +m, where each factor has the weak 9 topology. Thus a sequence u,-*~u in
Ou, ~u
W 1' ~176 if and only if u,~-~u in L*(O) and ~x;x, *--~~x-g in L~176 (1 < c~< m).
If A is a real-valued, continuous, even, convex function of t ~ satisfying
A(t) A(t)
A(t)>0 for t > 0 , ~ 0 as t ~ 0 , ~ as t ~ then we call A an N-
t t
function. If A is an N-function, its conjugate function /1 is defined by A(t)=
sup{ts-A(s): s~t}. A is also an N-function and satisfies A = A . Furthermore
Young's inequality,
t s <=A(t) + .4(s), (2.2)
holds for all s, t ~ . If A, B are N-functions then we write A<B if and only if
there exist positive numbers t o and k such that
A (t) < B (k t) (2.3)

for all t>=to. We write A,,~B if and only ifA~,B and B'<A, and A <<B if and only
if
A(0
lim B - - ~ = u (2.4)

for every 2>0. If A is an N-function the Orlicz class haA(O) consists of all (equiv-
alence classes of) real-valued measurable functions u on f2 such that

S A(u(x))dx < ~ . (2.5)

12

The Orlicz space LA(f2) is the linear hull of LfA(12). LA(O) is a Banach space with
respect to the Luxemburg norm

1lull(a)= i n f { k > 0 ; S A(u/k)dx <=1}.

I2

If A <~B then LB(f2) ~_L A(f2), while if A<<B then LB(f2)~ L A(f2). The space EA(f2)
is defined as the closure of the bounded functions in the LA(f2) norm. We have that
Ea(fJ ) _ 5CA(f2) _ La(f2). The dual of EA(f2) can be identified by means of the scalar
product ~ uvdx with Lx(O ). The norm on Lx(O) dual to II [I(A)on EA(f2) is de-
12
 Existence Theorems in Non-Linear Elasticity 349

noted }1 I]x and is equivalent to II Ilta). HOlder's inequality is valid in the form
uvdx<= Ilu[[ta)I[v[lx (2.6)
f2
for all ueLa(~2), veLx(f2).
The Orlicz-Sobolev space W1LA(O) (W ~EA(O)) is defined as the set of functions
Ou
ueLA(a) (EA(O))such that the weak derivatives Ux;eLA(a)(EA(e))(1<e<m).
W 1LA(O ) is a Banach space under the norm

JlUllw'LA~)=IlUi}(A)+~=X
~ O~U~xU~CA) (2.7)

and similarly for W 1EA(O). The closure of ~(O) in W 1LA(f~) is written Wo~LA(I2).
In the special case when A(t).~ [t[e (p > 1) we have the equalities
LA(Q ) = EA(Y2) = LP(~), W 1LA(~2 ) = W 1 EA(~2 ) -~ W I , P ( Q ) ,

1 1
and A(t).-~ [tIp', where p + ~ - = i.
Throughout this article we shall be dealing with vector and matrix functions ~.
When we write ~ e X , where X is any one of the spaces introduced above, we mean
that each component of ~ belongs to X, and we define II~[]x to be ~ I[~l[x
where the sum is over all components of ~ of ~ .
Finally we define some regularity conditions for O.
(i) f2 has the segment property if there exists a locally finite open covering
{0~} of ~?O and corresponding vectors {y~} such that x + t y ~ f 2 for all
x~Oc~ 0i and for all re(0, 1).
(ii) O satisfies the cone condition if there exists a fixed cone kr2___~" such that
each point x~O0 is the vertex of a cone k~(x) that lies in O and is congruent
to k~.
(iii) t2 satisfies a strong Lipschitz condition if each x e 00 has a neighbourhood
~//~ such that in some co-ordinate system, with origin at x, O c~q/~ is re-
presented in ~ by ~,,<F(~'), ~ ' = ( ~ , - . . , ~ , , - 0 with F a Lipschitz con-
tinuous function.

3. Quasiconvexity and the Legendre-Hadamard Condition

We consider integrals of the form
I(u, t2)= S f(x, u(x), Vu(x))dx,
t2

where (x, u (x)) ~ ~2 x N?", (V u(x))~ - Ou'(x) u',(x) (1 < i < n, 1 < ~ < m), and where
~x~
the real-valued function f is defined and continuous on a given relatively open sub-
set S of ~2x ~ " x M" • m. Here M" • " denotes the space of real n x m matrices, with
the induced norm of ~"". We assume that for each xr there exist u e ~ " ,
F6M" • with (x, u, F)~S.
350 J.M. BALL

Definition 3.1 (cf. MORREY [1].) Let U be an open subset of M "• Let g: U ~
be continuous. Then g is said to be quasiconvex* at F o 6 U if and only if
g(F0 + V~(y))dy > g(Fo)m(D ) (3.1)
D
for every bounded open subset D___~"~ and for every ~@(D) which satisfies
Fo + V~(y)~ U for all y~D. g is quasiconvex on U if it is quasiconvex at each Foe U.
Note that ifg is quasiconvex at FoE U, and if(~ W~' ~(D) satisfies F0 + F~(y)~K
for almost all y~D and for some compact subset K of U, then (3.1) holds for ~.
In fact by the definition of W~'| there exists a sequence of functions ~ , ~ ( D )
that converges to ~ in W ~' o~(D). Since K is compact there exists an integer N and a
compact set K 1 with K c K 1 ~ U such that /70+ V~,(y)~K 1 for all y~D and all
r >__N. If r>__N then (3.1) holds for ~r. By the compactness of K1, the continuity of
g and the bounded convergence theorem we obtain (3.1) for ~.
Let A = { w e WIIo~
1 1([2): (x,w(x), Vw(x))ES for almost all xE[2, and I(w,[2)
exists and is finite}.
Theorem 3.1. Let u~A be such that
I (u, [2)_~ I (w, [2)
for all w~A with w - u ~ ( [ 2 ) and Nw-Ulqc~t~ sufficiently small. Let Xo~[2 and sup-
pose that u and Vu have representatives, again denoted by u and [7u, that are con-
tinuous at x o with (Xo, U(Xo),Vu(xo))~S. Let U={F:(xo,U(Xo),F)eS }. 7hen
f(x o, U(Xo), .) is quasiconvex at Vu(xo)e U.
Proof. Let u satisfy the hypotheses of the theorem, let D be a bounded open subset
of ~ " and let ~6~(D) satisfy (Xo, U(Xo),VU(Xo)+V~(y))ES for all y~D. For
e > 0 define u~: [ 2 ~ m by
u~(x)=u(x)+~ ( ~ ~ if x - x ~

= u(x) otherwise.
For ~ small enough the set xo+eD, on which u and u~ differ, is contained in [2,
and thus u~-u~@([2). Also for x ~ x o +eD we have

which by our continuity assumptions and our assumptions about ~ is bounded

uniformly above on the set Xo+eD for z small enough. Thus u~eA and I(u, [2)<
x-x o
I(u,, [2). Making the change of variables y = - we obtain

~f(Xo + e y, U(Xo + ey)+e;(y), V u(xo + ~y) + V ;(y))em d y

D
> ~f(Xo+ey, U(Xo+ey), V~(Xo+~y))~dy.
D
* This is MORREY'S original terminology. In his book [2] he calls such functions (for U = M" • =)
strongly quasiconvex, retaining the term quasiconvex for functions satisfying the weakened form of
the Legendre-Hadamard condition that we shall term rank 1 convexity. The reader is warned that
quasiconvexity has other meanings in the literature on convex analysis.
 Existence Theorems in Non-Linear Elasticity 351

Dividing by ~'~ and letting e ~ 0 we obtain (3.1) for F0= Vu(xo) by the bounded
convergence theorem. []
Corollary 3.1.1, (Cf MORREY [1, p. 43] and MEYERS [1, p. 128].) Let Uc_M "~m
be open, let g: U ~ ~ be continuous, let F o ~ U and let (3.1) hold for a given bounded
open subset Do ~_~8" and for every r which satisfies Fo + V((y)~U for all
y e D o. Then (3.1) holds for all such D,~, i.e., g is quasiconvex at F o.
Proof. Apply the theorem to the integrand g(F) with f2 = D o and ui(x) = (Fo)~ x ~. []
Theorem 3.1 is essentially the same as a result stated by S~LVeR~AN [1],
following earlier work of BUSEMANN & SHEPHARD [1, p. 31].
We next show that for integrands that are independent o f x and u the existence
of a sufficiently regular minimizer to certain Dirichlet problems implies that the
quasiconvexity condition holds.
Theorem 3.2. Let U~_ M" • be open and let g: U ~-~~ be continuous. Suppose
that either (i) n = 1, (~ is arbitrary', or (ii) (2 is a hypercube, (2= { x 6 ~ " : 0 < x ~ < 1,
1 -<~_<m}, say. Let Uo: (2--~ ~ " be defined by

ug( ) = x'+z

where F E U and z ~ ~" are constants. Let

J(u) = Ig(W(x))dx

and let A I = { u 6 C 1 ( ~ ) : Vu(x)eU for all x e ~ , J(u) exists and is finite, and u = u o
on OE2}. Suppose there exists r 6 A 1 such that

J(v)<=J(u) for all u6A~.

Then g is quasiconvex at F c U.
Proof. Let w = r - u o. Then w~Cl(~) and w = 0 on Of2. In Case (i) there exists
Xo~2 with Vw(xo)=O. Therefore Vv(xo)=F and the result follows from Theorem
3.1. In Case (ii) we have that Vw(xr)~ 0 as r ~ oo for any sequence {xr} _~g2 with
x r ~ 0 as r--+ oo. By Theorem 3.1, g is quasiconvex at F + Vw(xr)~ U. Taking the
limit r ~ oo in the quasiconvexity condition gives the result. []
Remarks. For n---1 or m--1 quasiconvexity is equivalent to convexity (see
MORRV.V [1, 2]). The analogue of Theorem 3.2 for n > 1, f2 arbitrary, is false. As
an example, let m---n=2, (2={(x 1,x2):x~+xz2<l}. Define g: M 2 • by
g(F)=p(r), where r = l F l = t r ( F F r ) ~ and where p: ~ + ~ is zero for r > l and
positive for 0 < r < 1. We show that for any Uoe C 1((2) there exists an absolute
minimizer for J ( u ) = ~g(Vu(x))dx among C1(f2) functions u satisfying u = u o on
~f2. Let ~ C 1 ( ~ ) satisfy ~--0 on c~f2 and let ]Vr for all xEQ; e.g., we
may take r where h 1 and h 2 are C1(f2) functions satis-
fying h 1 = h z --0 on Of2, Vhi(x)=Oif and only if x = a i (i= 1, 2) with a 1 4=a 2. For
large enough k > 0 and for all x ~ s we have

I V(uo+kr ~ k ~ - I I Vuollc<~)> 1,
352 J.M. BALL

so that J(uo+k~)=O and uo+k~ is an absolute minimizer. But g is not quasi-

convex, as may be seen by putting u o = F x in the above argument, where F E M 2 x 2
is constant with IFI < 1. With a little more work one can show that the absolute
minimizer u corresponding to u o may be chosen so that det 17u(x)>c>O for all
xeO.
Definition 3.2. Let U be an open subset of M" • m. A function g: U ~ ~ is rank 1
convex on U if it is convex on all closed line segments in U with end points dif-
fering by a matrix of rank 1, i.e.,
g(F+(1-2) a|174
for all F~U, 26[0, 1], a 6 ~ m, b ~ " , with F + # a | for all #~[0, 1]. Here
( a @ b )i, d=e fa i b, (l <i<n, l<c~<m).
Theorem 3.3. Let U be an open subset of M "xm and let g: U---, ~. The .following
conditions (i)-(iv) are equivalent:
(i) g is rank 1 convex on U;
(ii) for each fixed F ~ M "~', b ~ "~ the function a~--,g(F + a | is convex on all
closed line segments in the set {a: F + a | be U};
(iii) for each fixed F ~ M" • m, a ~ Yi~ the function b ~ g(F + a @ b) is convex on all
closed line segments in the set {b: F + a @ b 6 U } ;
(iv) the inequality
g(H)<2g(H +c| ( H - l-@-~ c | ) (3.2)

holds for all 2 6 [0, 1) and for all H ~ M" • ", c ~ ~t", d ~ Yl~ satisfying H + # c | de U
forall#e[2@,l ].
I f ge C(U), then (i)-(iv) are equivalent to
(v) for each F~ U there exists A ( F ) e M " • such that
g(F + a | b) > g(F) + A~(F) a i b~ (3.3)

whenever F + 2 a | b ~ U for all 2 e [0, 1].

I f ge C ~(U), then A~(F)= ~3g(F)
~vi"
l f ge Cz(U), then (i)-(v) are equivalent to
(vi) (Legendre-Hadamard condition)

02g(F) aiaJb~ba>O forall a ~ n, b ~ m, F6U.

~?Fj~?F~
Proof. The equivalence of (i), (ii) and (iii) is clear (cf SILVERMaN [1, Thin. 4]).
The equivalence of (i) and (iv) is proved by making the change of variables c =
( 2 - 1 ) a, d = b, H = F + (1 - 2 ) a @ b. Let g e C(U). That (v) implies (ii) follows by a
well known condition for convexity. To show that (ii) implies (v) one can use the
arguments of MORREY [1, p. 47] to establish (3.3) for a@b belonging to some
neighbourhood of zero in M "• m, and then deduce (v) from the convexity of the
function a~--~g(F+a| The remaining assertions of the theorem are obvious.
[]
 Existence Theorems in Non-Linear Elasticity 353

Condition (iv) was derived by GRAVES [1] and has recently been studied by
ERICKSEN [2]. Motivated by (iv) we say that g is rank 1 convex at He U if the
inequality (3.2) holds whenever the right-hand side is defined. In this case it is
easy to see that the Legendre-Hadamard condition holds at H. The prototype
for the following theorem was discovered by HADAMAm9 [1, 2], the first rigorous
proof being that of GRAVES [-1]. For other proofs and relevant literature see
DUHEM [1], MCSHANE [1], CATTANEO [1], VAN HOVE [1, 2], TRUESDELL & NOLL
[1, p. 253] and MORREY [2, p. 10]. The proof here is based on MORREY [1, p. 45]
and on Theorem 3.1.
Theorem 3.4. Let u, x o satisfy the hypotheses of Theorem 3.1. Let

u0 = {v: (Xo, .(Xo),V)es}

Then f(xo, U(Xo), ")is rank I convex at Vu(xo)eUo.
Proof. Define g: Uo ~ by g(F)=f(Xo, U(Xo),F). Let H=Vu(xo). Let Ae[0, 1),
2
let c, d be such that H + c | and H - - ( L ~ - c | belong to U0 and assume
without loss of generality that d # 0 , 24=0. Let p > 0 , gl=d/Idl, h=l/]dl and
1-2
k= ~ h. Choose vectors p~ (1 <fl <m) such that (PI .... ,/~m) is an orthonormal
set in ~ . Let D denote the rectangular parallelepiped
- k < y x <h, lYal<p (1 <fl__<m),
where yp=x.laa. Let F1-, F1+, F~, F~+ (l<fl_-<m) be the faces y l = - k , yl=h,
y p = - p , ya=0, respectively. Let rt~ (1 < f l < m ) be the pyramid with base F~
and with vertex at the origin. Let ~ be defined o n / ) to be continuous on/), zero
on ~D, and linear on each rc~ and rc~- with ~(0)= - c . Then
[ - k -1 c| on z i-
h -1C@~ 1 on rq+
on
[ p-~c| on n~.
Provided p is large enough Theorem 3.1 and the remark following Definition 3.1
imply that
S g(H + V~(x)) dx >=g(H) m(D) = g(H) 2 "-a pro- l(h + k).
D
Hence
1 [ 2h 2k
2m ,_~-k - g(H +h-I c| h ~ k g ( H - k-1 e|

+ ~ ( g ( H - p - t C| -1 c|
B=2
Letting p ~ c~ we obtain (3.2). []
The proof above shows that, roughly speaking, quasiconvexity implies the
Legendre-Hadamard condition. Whether or not the converse holds is an im-
354 J.M. BALL

portant open question (cf the comments of MORREY [2, p. 122]). The conditions
are known to be equivalent only in certain special cases, for example in the qua-
draticcase f(F)=a~f F~'i F~j with aT~ constant and m, n arbitrary (MORREY [1, 2],
VAN HOVE [1]), and for certain parametric integrands when n = m + 1 (MORREY
[1, 2]). In particular nothing interesting is known about the case m = n > 1, which
occurs in nonlinear elasticity.
To discuss this problem, consider a continuous integrand f(Vu), defined on
all of M "• m, and independent of x and u. Suppose that f is rank 1 convex on
M "• It is well known that if D is a bounded open set in ~ n then any function
~ ~(D) can be approximated in WoE ~ (D) by piecewise affine functions (EKELAND
& T~MAM [1, p. 286]). Thus a natural method of attack is to follow the lead of
the proof of Theorem 3.4 and to seek domains D with a partition into a finite
number of disjoint open sets D k and a set of measure zero, such that the quasi-
convexity condition
S f(Fo + 17~(x)) dx > f(Fo) re(D) (3.4)
D
holds for any F o ~ M "• and for any ~ W d ' ~ ( D ) that is affine on each D k (cf.
SILVER~IAN [1, Thm. 2]).
For ease of illustration we consider the case m = 2, n arbitrary; similar com-
ments apply for m > 3. First let D be the interior of a triangle in ~2 with vertices
al, aa, a3, and let e be an interior point of D. Let D1, DE, D 3 be the interiors of
the triangles a z ea3, a 3 ea~, ax ea: respectively. Let n~ be the unit outward normal
to t3D on the side a 2 a3, let ll=la2-a31 , and let n2,/13, 12, 13 be defined analo-
gously. Let ~ W o 1' ~ be affine on each D k with ~(e)=c. Then

lk
V~=--C| on D k,
2 m(Dk)
and (3.4) becomes
3( 2k f F o-t 2m(Dk) e |
)
k >f(Fo),
k=l
where 2k=m(Dk)/m(D ). But this inequality follows from rank 1 convexity o f f
because
}'k lk nk
k=l 2m(Dk)=0.
A similar argument shows that (3.4) holds for piecewise affine functions if D is
the interior of a convex polygon and the Dk'S are triangles formed by joining a
single interior point of D to adjacent vertices of the polygon.
A different situation arises if we introduce more interior nodes into the parti-
tion of D. For example let D be an equilateral triangle A 1 A 2 A 3 of side 1 parti-
tioned into 16 congruent equilateral subtriangles of side 88(see Fig. 1). Let n j, n 2,
n 3 be the unit outward normals shown, and let e~, e2, e 3 be the position vectors
of the three interior nodes B1, B 2 and B 3. Let Cl, cz, c 3 be given and let ~s W~' ~(D)
8
be affine on each subtriangle with ~(el)= ~ ci. The values of V~ in each sub-
V~
 Existence Theorems in Non-Linear Elasticity 355

A1

-clan 1
n3 / \ / \ 112

cl| \ ~/ cIQn2
-cl| / \ -cl|

+ C3~) n3 33
-c2|

c2| \ / c3|

A2 A3

T
I11
Fig. 1

triangle are shown in Fig. 1. The corresponding quasiconvexity inequality is

easy to write down, but does not seem to follow from rank 1 convexity of f This
suggests that the Legendre-Hadamard condition does not imply quasiconvexity.
Unfortunately the search for a counter-example is hampered by the fact that, as
we have mentioned, any such f cannot be quadratic in Yu. It is conceivable that
if m=n then the Legendre-Hadamard condition implies quasiconvexity for ob-
jective functions.
Finally we mention in passing an implication of the configuration in Figure 1
for finite element methods. Let D be as in Figure 1 and suppose that one wishes
to solve a boundary-value problem for u subject to the pointwise constraint
det 17u(x)= 1 for all xeD (the two-dimensional analogue of the incompressibility
constraint (1.16)) and, for example, the boundary conditions u(x)=x for all
x~OD. Then the only function (e W1' ~~ that is affine on each subtriangle with
det 17(x + ~(x))= 1 for all x e D is ~ = 0. Indeed, any such map x ~ x + ~(x) deforms
each subtriangle into another triangle with equal area. It follows that, for example,
B 1 can be displaced only along the line through B 1 parallel to A 1 A3, and also
along the line through B1 parallel to A 1 A2. Hence B~ is fixed by the map, and
356 J.M. BALL

similarly for B2, B 3. A similar argument applies when the number of congruent
subtriangles in the partition of D is increased. Thus either nonlinear interpolation
functions or nonconforming elements must be used. A related difficulty for in-
compressible fluids is discussed by T~MAM [1].

4. Sufficient Conditions for Quasiconvexity

The quasiconvexity condition is not a pointwise condition on the function f
and is therefore difficult to verify in particular cases. In this section we shall be
concerned with more accessible conditions that are sufficient for quasiconvexity.
These conditions apply to functions f for which the Legendre-Hadamard con-
dition is not known to be equivalent to quasiconvexity.
Throughout the rest of the article we assume, unless the contrary is stated, that
m = n = l , 2 or 3.
We first study those functions ~b(F) that belong to the null-space of the Euler-
Lagrange operator; i.e., those functions for which the corresponding Euler-
Lagrange equations are identically satisfied. For smooth ~b the following result
is a special case of ERICKSEN [1], EDELEN [1, 23 and RtJND [1, 2].
Theorem 4.1. Let c~: M"• ~ Y t be continuous and such that both c~ and -c~ are
rank 1 convex on M "• so that
q~(F + ( 1 - 2 ) a| - 2 ) (9(F+a| (4.1)
for all F s M "• a, b s ~ " , 26[0, 1]. Then c~has the form
dp(F)=a+bF /f n = l ,
~b(F) = a I +fl~ F~+ 7 det F if n=2, (4.2)
q~(F) = A + B~ F~ + C~(adj F)i, + D det F if n = 3,
where a, b, a 1, fl~, y, A, B~, C~, D are arbitrary constants, and where adj F denotes
the adjugate matrix o f F (i.e., the transpose of the matrix of cofactors).
Proof. We just treat the case n = 3; the cases n= 1, 2 are easier. Suppose first that
~b is C 2. Then by Theorem 3.3 (vi), (4.1) is equivalent to
A~f(F) a i a-i b~, btj = 0 for all a, b E ~ 3 and for all F E M 3 • 3,

wh?re Aij(u
,~ - -)d=e e02
- - -q~(F)
-. It follows that A~f=A~f(F) is alternating; i.e., A~f=
0 ;0 j
A~i = -Aj/~ and A~f = 0 if ~=fl or i=j. Since A~ =A2222=A333:0 it follows that
qS(f) is affine in each of F~, F2 and F33, so that
tp(r) = ~bo(g) F( F 2 F 3 + ~b~(if) F 2 F 3 + ~bz(ff) F 3 F? + q~3(/~) F• F2
+ 0~(F) F( +02(,~) r~ + 0~(g) F~ + Z(F),
where ff denotes the matrix of off-diagonal elements of F, and where the functions
c~i, Oi, Z are C z. Since A ~ =0, etc., we obtain the equations
0~b0 -- 0 4 2 : 0r -- 001
--=0,
0F2x 0F21 0F21 0F2a
 Existence Theorems in Non-Linear Elasticity 357

etc., so that 4'o, ~b,, ~b2, q~3 are constants, Oi=O,(F2;F3), 02=02(F 3, F1) and
03=03(F~, F2). Applying the conditions A~ 2= -AlE 2, etc., we can reduce ~b to
the form ~b(F)-- ~b(ff)+ C~(adj F)i, + D det F, where i j - 0 for all i, j, ~, ft.
The result follows. ~F~' ~F~
For a general continuous q~ we use a mollifier argument. Let p6@(M 3• 3)
satisfy p > 0 , p ( F ) = 0 if IFI>I, ~ p ( F ) d F = l . For e > 0 let p~(F)=e-9p(F/e).
M3•
j def
Then q)~ -- Pc * q~ clearly satisfies (4.1), is C ~, and thus

q~(F)=A(~)+ B~9 (e) Fj' + C~~(e)(adj" F)i, + D(e) det F.

But since ~b is continuous, ~b~---,~b uniformly on bounded subsets of M 3 • 3 as

e ~ 0 . For fixed i,~ let U = t , F~=0 i f j + i or fl#~t. Then the functions

g~ (t) = A(~) + ~ ( ~ ) t

converge uniformly on compact subsets of ~ to a function g(t), which is easily

shown to have the form g(t)=A+B~t. Thus A(e)--,A and B~(e)~B~ as e ~ 0 .
By choosing F so that (adj F)~=t, (adj F)~=0 ifj:#i or fl#~, we obtain similarly
that C~.(e)~ C~, D(e) ~ D as e ~ 0. The result for continuous q~ follows. []
Corollary 4.1.1. Let dp: M " • be continuous. Then both q~ and - ~ are quasi-
convex on m "• if and only if c~ has the form (4.2).
Proof. If q~ and -~b are quasiconvex then by Theorems 3.3 and 3.4 both ~b and
-~b are rank 1 convex on M" • and so by the theorem ~b has the form (4.2).
Conversely any q~ of the form (4.2) is such that, in the notation of the preceding
proof, A~f is alternating. Then for any F o e M" • ", for any bounded open set D _ ~t",
and for any ~e~(D) we have

and the result follows. []

We next recall some results of BUSEMANN,EWALO & SHEPrtARD [-1] concerning
convex functions defined on non-convex sets. Let s > 1 and let M _~~ be such that
the dimension of the convex hull Co M of M is s (i.e., the linear subspace spanned
by M is ~ ) . We do not assume that M is convex. Let ~-: M - - ~ . For variable r > 1
we denoted by A = {2~, ..., 2,} a variable set of non-negative real numbers 2i
r
with ~ 2~= 1.
i=l

Definition 4.1. ~- has a convex lower bound if and only if there exists a real-valued
convex function C(z) defined on Co M such that ~ ( z ) > C(z) in M. (Without loss
of generality C may be assumed to be affine.)
is said to be convex on M if it is the restriction to M of a real-valued convex
function (in the usual sense) defined on Co M; equivalently, ~ may be extended
to a convex function on Co M.
358 J.M. BALL

T h e o r e m 4.2 (BuSEMANN; EWALD • SHEPHARD[1]).

(i) ~ is convex on M if and only if it has a convex lower bound and the inequality
r

~(z~)<= y~ :(~(z,)
i=1

holds for all z I . . . . . z r and ZA= ~'~.iZl lying in M. A suitable convex extension
to Co M is given by i= 1

g ~ ( z ) = inf ~ 2i~(zi), zism , l<r<oo.

z=zA i=1

(ii) Let Co M be open. Then either of the following conditions is necessary and
sufficient for ~ to be convex on M:
(a) ~ h a s a convex lower bound and the inequality
s+l
~(z~) <= 2 ~,~(z,)
i=1

s+l
holds for all z 1. . . . . Zs+ 1 and z a = ~ z i lying in M.
i=1

(b) for each point zo e M there exist numbers ai(Zo) (i = 1. . . . . s) such that
s

~(z) > ~(Zo) + Y~a,(Zo)(Z~-- ~o),

i=1

for all z ~ M .
We now define finite-dimensional Euclidean spaces E and E 1 by

E=E x xN,
where
E 1 is e m p t y if n = 1,
E I = M 2• if n = 2 ,
El=M3•215 3• if n = 3 .

Thus E m a y be identified with Ns(,}, where s(1) = 1, s(2) = 5 and s(3) = 19.
Define the m a p T: M " • by

T(F) = F if n = l ,
T(F) = (F, det F) if n = 2 ,
T(F) = (F, adj F, det F) if n = 3 .

Let U ~ _ M "• By the t h e o r e m on the invariance of d o m a i n the set T(U)~_E is

open if and only if U is open. However, T(U) is not in general convex even if U
is convex (except in the case n = 1). The following result shows, in particular, that
in certain i m p o r t a n t cases when U is open, so is Co T(U).
Theorem 4.3. Let K ~_~ be nonempty and convex, and let U = {F~M"• det F ~ K } .
Then Co T ( U ) = E 1 x K.
 Existence Theoremsin Non-LinearElasticity 359

Proof. We give the proof for n=3, that for n = 2 being similar. For k ~ define
Vk_CEt by
Vk={F, adj F): F e M 3• d e t F = k } .

It suffices to show that Co Vk = E 1 for all k. Suppose not. Then there is a closed
half-space
n={(F,A)~EI: EiG~d-AiH
- i - - - ~ - - i ~ <~} ,

(G, tt)4:0, with Vk___n (RoCKAFELLAR [1, p. 99]). If RI,R2EM 3• are proper
orthogonal then

A~i H i = t r [ ( R 1 FR2)(R~ GR'~)+ (R2T AR~)(R~ HR2) ] .

Since adj(R1FRz)=R~(ad j F)R~, det (RIFRz)=detF, we may without loss of
generality suppose that H is diagonal. Suppose that H . 0 and assume without
loss of generality that H~ 4=0. Let
F = diag (k N - 1 sgn H~, N 89sgn H~, N~).
Then adj F = diag (N sgn H~, kN -~ sgn H~, kN -89 and det F = k. Hence (F, adj F)~
Vk, but for N > 0 large enough (F, adj F)$n. If H = 0 then we may assume that
G~4:0, let F = ( k N s g n G ~ , N - i s g n G ~ , N -~) and proceed similarly. Hence
Vk ~ n and this contradiction proves the result. []
For g: U--+~ we may define a function G: T ( U ) ~ t by G(T(F))=g(F), i.e.,
g=G if n = l ,
g(F)= G(F, det F) if n = 2 , (4.3)
g(F)=G(F, adjF, detF) if n=3.
Definition 4.2. A function g: U ~ is polyconvex if and only if G: T(U)---,R
defined by (4.3) is convex on T(U).
If n = 1 and U is convex, then polyconvexity of g is the same as convexity in the
usual sense. If n = 2 or n = 3 , polyconvexity is characterized by the following
Theorem 4.4. Let U be such that Co T(U) is open. Then g is polyconvex if and
only if one of the following three equivalent conditions holds:
(i) Ifn = 2:
there exists a convex function C(F, 3) on Co T(U) with
g(F)>C(F, detF) forall F6U,
and the inequality

g 2i F (0 2i g(F (1))
i
6
holds for all 2i>0 with ~ 2i= 1, and for all Fti~ U satisfying
i=1

~ 2i det F ~i)= det 9 F (i~ . (4.4)

i=1 i=
360 J.M. BALL

If n=3:
there exists a convex function C (F, A, 6) on Co T(U) with
g(F)>C(F, adjF, detF) forall F~U,
and the inequality
20 20

2O
holds for all 2 i > 0 with ~ 2i= 1, and for all F(~ U satisfying
i-1

2i adj F (i)= adj 21F (1)

i=1 i=l

and (4.5)
20 20 )
~ ).idetF(i)=det (~=lJ.iF(i) .
i=1 i

(ii) If n = 2:
for each F ~ U there exist numbers a~(F), a(F) such that
-' - ~') + a( F) (det F - det F)
g (F) > g (F) + a,~(F) (F2 (4.6)
for all F6 U.
If n=3:
for each F ~ U there exist numbers a~(F), b~(F), c(F) such that
g(F)>__g(F)+a~(F)(F~-F~)+b~(F)((adj F)i~-(adj F)~)+c(F)(detF-det F) (4.7)
for all ff 6 U.
(iii) If n = 2:
for each F6 U there exist numbers A~(F), a(F) such that
g (F + n) > g (F) + A~'(F) rt', + a (F) det zr (4.8)
for all F + ~ U .
If n=3:
for each F ~ U there exist numbers A~(F), B~(F), c(F) such that
g (F + n) > g (F) + A~ (F) zd, + B~'(F) (adj n)~ + c (F) det n (4.9)
for all F + rc6 U.
Proof. That polyconvexity of g is equivalent to (i) or (ii) follows immediately
from Theorem 4.2 (ii) and Theorem 4.3. That (ii) and (iii) are equivalent follows
by setting/~= F + n and rewriting the right-hand sides of (4.7) and (4.9). (c(F) has
the same value in both conditions.) []
Of course, if G is C z, then the coefficients on the right-hand sides of (4.6) and
(4.7) are given by the derivatives of G with respect to its arguments. Condition
(iii) is the form given by MORREV [2, p. 123], who proved the following theorem:
 ExistenceTheorems in Non-Linear Elasticity 361

Theorem 4.5. Let U be such that Co T(U) is open. If g is polyconvex, then g is

quasiconvex on U.
Proof. We give the proof for n-- 3. Since G is the restriction of a convex function
to the set T(U), it is continuous and hence so is g. Let D be a bounded open subset
o f ~ 3, let Fo~U and let ~ ( D ) satisfy Fo+ V~(y)~U for all y~D. By Corollary
4.1.1 we have
(i~ dy = S [(adj (Fo + V0)/, - (adj Fo)~] dy = ~ [det (Fo + V ( ) - det/7o] dy = 0,
D D D

and the quasiconvexity of g follows from (4.7). []

The converse to Theorem 4.5 is false if n=3. In fact when U=M 3• 3 and
g(F)= (4.10)
with a~f constant, it is easily seen that g is polyconvex if and only if there are
constants B~ with
g(F)-B~(adj F)~>O for all F. (4.11)
As was pointed out by MORREY [1, p. 26], an example of TERPSTRA [1] shows
that if n = 3 there exist constants a~f such that (4.11) is violated for any B~, even
though
d[jp2i2JFt~p~>O for all nonzero 2, p~.~ 3, (4.12)
so that g is quasiconvex. TERPSTRA also showed that for n = 2 any quasiconvex g
of the form (4.10) satisfies (4.11) for suitable BT, and hence is polyconvex. I know of
no counterexample to the converse of Theorem 4.5 if n=2. Conditions of the
type (4.11) were studied by CLEBSCH [1] and HADAMARD [-1, 3].
The conditions of quasiconvexity and polyconvexity may be contrasted as
follows. Considering for illustration the case in which n -- 3, let 2 i and F (i) be given
satisfying (4.5). There is then no reason to suppose that a domain D, and a partition
of D into 20 open subsets D~ of volume 2 tl) and a set of measure zero, can be found
such that there is a continuous piecewise affine function on D whose gradient takes
the value F tiJ on Di (cf the discussion at the end of Section 3). Only for such D, D i
can the corresponding polyconvexity inequality be deduced from the quasi-
convexity condition. I remark that Theorem 4.4 (i) offers a way of proving that a
given function is not polyconvex.
By sacrificing the pointwise nature of the condition imposed on g, we may
obtain a sufficient condition for quasiconvexity generalizing polyconvexity.
For brevity we discuss this new condition just when n--3.
We consider continuous functions G: Co T(U) ~ .~ with U ___M 3 • 3 arbitrary.
Roughly speaking we require that the integral of G, rather than G itself, be convex.
To be precise, let 7 > 1, ~ > 1, v > 1 and let u e W ~' ~(f2) satisfy Vu(x)~ U for almost
all x ~ 2 . Suppose further that adj VueLU(O)9, det VusL~((2). Thus if I;(u) denotes
(Vu, adj Vu, det Fu), then X(u)~M d----~f/?(O)9x Lu((2)9 x E(O). Let
R,,-- {2(u + 0 : ~e~((2)}
and let cg. denote the closed affine subspace of N spanned by R,. Let
W. = {ascg.: a ( x ) e C o T(U) almost everywhere in f2}.
362 J.M. BALL

Define
= dx.
O

Suppose that J. exists and is finite or +oo for all a e c f . . Let G be such that J. is
a convex function on the convex set o,~f..
Definition 4.3. If g: U - - , ~ then g is said to satisfy condition (PT,u,v) at u if and
only if there exists G = G.: Co T ( U ) ~ Yt with the above properties and such that
g(F) = G(F, adj F, det F) (4.13)
for all F e U .
Theorem 4.6. Let U be such that Co T(U) is open. Let 7, I~, v be arbitrary.
(i) Let g: U + ~l be polyconvex and bounded below, and let u be as above. Then
g satisfies P~,u,v at u.
(ii) Let u i ( x ) = F i~ x ' + z i with FoeU and z e ' ~ 3 both constant. Let g satisfy
P~,u.v at u. Then g is quasiconvex at FoeU.
Proof. (i) Since g is polyconvex there exists a convex function G: Co T ( U ) ~ Yl
satisfying (4.13). Since Co T(U) is open G is continuous, and hence G(a(" )) is
measurable for each a e Y . . By Theorem 4.2 (i) we may suppose that G is bounded
below on Co T(U). Thus J. exists and is finite or + oo for all #e~f., and J. is
clearly convex.
(ii) Suppose first that G is C 1 on Co T(U). Let ~e~(f2) satisfy Fo+ V~(x)eU
for all x ~ f2. Let ~ = IF(u), ~ = ,~(u + 0- Then a, 6 ~ ~,"~. A standard argument shows
that if
o (t) = j' + dx,
D

then OeCl([0, 1])with the obvious derivative. By the convexity of J. we have

that O(0)< O ( 1 ) - O'(0). Hence

a ~ ' e(adj F)~ (X(u))(adj 17~)~,

+ O(det F--)-(IF(u)) det 17~ dx,]

and so g is quasiconvex at Fo 9 U.
The result follows for general continuous G by a mollifier argument. []
Condition P~,u,~ does not in general imply polyconvexity. Indeed let g(F)
be given by (4.10) with a~fl satisfying (4.12) but not (4.11) for any B~', so that g is
not polyconvex. Let 7 = 2 with # and v arbitrary and let U = M 3 • 3. Define G: E ~
by G(F, A, 6)= g(F) for all (F, A, 6)eE. Let u be as above. For any ~ e W~' 2 (f2) we
have (VANHOVE [1], MORREY [2])
-U -,,~(#dx>=O"

It follows that 7. is convex on ~:., and hence g satisfies P~,~, ~ at u.

There are some grounds, connected with the results of Section 6, for believing
that condition P~,u, v at affine u and the quasiconvexity condition are equivalent,
but I have been unable to prove or disprove this.
 Existence Theorems in Non-LinearElasticity 363

To complete this section I remark that many of the results may be extended
without undue difficulty to arbitrary m and n; the polyconvexity condition is then
a requirement of convexity with respect to the basis elements of the null space of
the Euler-Lagrange operator. (See BALL[2].)

5. Isotropic Convex and Polyconvex Functions

The purpose of this section is to give a method for producing a wide variety
of nontrivial isotropic polyconvex functions. These functions will prove valuable
in Section 8 when we apply our existence theorems to certain models which have
been proposed for rubbers. We begin by discussing isotropic convex functions of
n x n matrices for arbitrary n > 1. We recall that the singular values of an n x n
matrix F are by definition the eigenvalues of the positive semidefinite symmetric
matrix F ] ~ . When F is the deformation gradient these eigenvalues are the
principal stretches of the deformation. When examining the results below the
reader should bear in mind equation (1.15).
Notation. Vectors in ~" are denoted by x = (x1 , ..., x,) and the inner product
of two vectors x, y e ~ " is written (x, y). N+ denotes the positive orthant
{x: xi>O for 1 < i n n }
of N". ~, denotes the permutation group on n symbols (an element P of ~, acts
on an n vector by permuting its entries).
We shall prove the following theorem:

Theorem 5.1. Let n > 1. Let q~(v 1. . . . . v,) be a symmetric real-valued function
defined on ~ + . For F ~ M "• " d e f n e

W ( F ) = ~(vl . . . . . v.), (5.1)

where vl .... , v, are the singular values o f F . Then

(i) W is convex on K = { V s M " • V is positive-semidefinite and symmetric}
if and only if ~ is convex.
(ii) W is convex on M "• if and only if q~ is convex and nondecreasing in each
variable v i.

Remarks: Part (i), which is probably known to matrix theorists, was stated
by HILL [3] for n = 3. HILL'S proof relies on a property of the trace of the product
of two symmetric matrices, a recent proof of which has been given by THEOBALD
[1]. HILL assumed that W is differentiable everywhere, an assumption which rules
out several simple and useful functions and which can be surprisingly tedious to
verify. Proofs using differentiability obscure the geometric nature of the result.
The harder implication in (ii) is due to THOMPSON8r FREEDE [3]. Our method
of proof is broadly similar in approach, but rather different in detail. The result
extends work of YON NEUMANN [1] on gauge-functions and matrix norms (see
especially his Remark 5). For further information and references on singular value
inequalities see AMIR-MOEZ [1], AMIR-MOEZ&HORN [1], THOMPSON [1],
THOMPSON & FREEDE[1-3].
364 J.M. BALL

L e m m a 5.1. (VONNEUMANN [1]; see also MIRSKY [1,2]). Let A , B E M "• have
singular values ~1 > 0~2 ~-~ " " >=(Xn ~> 0 and fll > fiE ~-~ " " :> fin >=O. T h e n

Itr (AB)I <(at,/~). (5.2)

L e m m a 5.2. (YON NEUMANN [1]). Under the hypotheses of Lemma 5.1
m a x tr (AQBR)--(at, fl), (5.3)
Q,/i
where the maximum is taken over all pairs Q, R of orthogonal matrices.
Proof. There exist o r t h o g o n a l matrices Q:, 02, R1, RE such that A = Qx diag(at) R 1,
B = O 2 diag(fl) RE. Choose Q = R ~ O 2r ' R = g 2r O1. r Then

tr (AQBR)= (at,fl).

But for any o r t h o g o n a l Q, R the matrices AQ and BR have singular values at and fl
respectively. H e n c e tr (AQBR)< (at, fl) by L e m m a 5.1. []

L e m m a 5.3. Let r I >=rE>= " " ~rn~0. Then (r, v) is a convex function of F, where
V 1 ~> V 2 ~-~"" ~ V n ~ 0 are the singular values ofF.
Proof. In L e m m a 5.2 put A = F, B = diag r. Then

(r, v)-- m a x tr (FQBR), (5.4)

Q,R
and each tr (FQBR) is a convex function of F. []
Remark. By letting r = ( 1 . . . . . 1,00...,0) in L e m m a 5 . 3 it follows that for
kth
k place
1 _< k _<n, ~ v/is a convex function of F.
/=1
L e m m a 5.4. Let q ~ c 2 )__... )_c,)=O. Define the sets
L={yG~l"+:(r, P y ) < ( r , c ) forall rl>=r2>...>r,>O andall PG~.},

L 1 = Co {0, V ( q , c 2 .... , c l, 0, ..., 0): P c 3~., 1 _-<l_-<n},

M 1 = C o {Pc: P ~ , } .
Then L = L: and M = M 1 .
Proof. L is a convex set containing 0 and the points P ( q , c 2 . . . . . c l, 0, ..., 0).
Thus L 1 ~ L . To show that L ~ L 1 we prove that any closed half-space in ,~"
containing La also contains the closed convex set L. Let such a half-space be
~ = { y ~ . ~ " : ( y , x ) < # } , where xG.~", / ~ . ~ are fixed. Let the coordinates of x in
some order be
XI~_~X2 >= "-=_~x k ~ O > Xk+l ~---... ~_~Xn,

where the symbols to the right of 0 are omitted if k = n. Let y e L and let

;1 ~>; 2 >= "'" ~ ; n ~ 0

 E x i s t e n c e T h e o r e m s in N o n - L i n e a r E l a s t i c i t y 365

be the coordinates of y in some order. Then

(y, x)<;~ ~ + ... + ~k'2k
C1 X1 "j- "'" -'[-Ck Xk
<=1~,

so that y e n . Hence L 1 =L.

M is a convex set containing the points P c. Thus M 1 ~ M. Let rc contain M1,
let y e M and let ~ >y2_>_ ... >y,_>_0 be the coordinates of y in some order. Then

(y, x)__<~ Yi~,

i=1

= ~ Yi xi + ci- i Yc,,
i=1 i i=1 /

= ~ y~(~- ~.) + ~. c~.

i=1 i=1

Choosing r i = x i - x n ( 1 < i < n), r. = 0 in the definition of L we obtain

n--1
(y,x)___ c,=
'=1 '=1 '=1

Thus y e ~ a n d M ~ = M . []
Proof of Theorem 5.1.
(i) That the convexity of W implies the convexity of 9 is immediate. Thus let
be convex and symmetric. Let U, V e K have singular values (eigenvalues)

Ul>U2>=...>=u~>O, vl>v2>...>v,>O.

Let 2e[0, 1] and let A ~f ) , U + ( 1 - 4 ) V have singular values a 1 __>a2 > ... > a , > 0 .
Let c = 2 u + ( 1 - 2 ) v . Note that for some orthogonal Q, R we have

ai= ~ ( 2 Q ~ . u ~ + ( 1 - 2 ) R ~ v i ) = ~ c,. (5.5)

i=1 i,j=l i=l

By (5.5) and Lemma 5.3 we have a e M . Hence by Lemma 5.4 a e M 1. Therefore

W(A) = ~P(a)= 9 4, P~c , (5.6)

i

where P / e ~ , 2,_>0, ~ 2i= 1. Thus

/=1

w(A)~ ~ ~,~(P,~)= ~ ~,~(c)=~(c)<=~r

'=1 '=1

where we have used the symmetry of ~. Hence.

w(A)_<,~w(u)+(1-4) w(v) (5.7)
as required.
366 J . M . BALL

(ii) Let W be convex on M "• Then clearly 9 is convex. Also for fixed non-
negative v 1 , ..., Vk_ 1 , Vk + 1 , ..., Vn

g(v) a"J W(diag (v1. . . . , Vk_l, V, V k + i , . . . , V,))

is convex in v. But g(v)---~b(h .... ,Vk_l,lV[,Vk+l, ...,V,) and any even convex
function of te.~ is nondecreasing for t>0. Hence ~b is nondecreasing in each
variable.
Let 9 be convex, symmetric and nondecreasing in each variable. Let 2el0, 1]
and let F, G e M " x ,, A = ;t F + (1 - 2) G have singular values

Ul ~ U2 ~-~ " " ' ~-~/2n ~m~0, l)l ~_~192 ~ "'" ~-~ Vn ~ 0 , al>=a2~...>=a,>=O
respectively. Let c = 2 u + ( 1 - 2 ) v . By Lemma 5.3 a e L . Thus a e L l , and since cb
is nondecreasing in each variable we have
W ( A ) = q ~ ( a ) < Cb(c)<,~ W ( F ) + ( 1 - 2 ) W(G). []

The geometrical basis for Theorem 5.1 is easily seen by considering the special
cases n = 2 and n = 3 (see Fig. 2).

2 A (v2,v1) v 3

S ~ (c2,c3,cl . . c3;c2,c1)
[ I ]%.\ \
! \
-~. ,

Vl v2
Fig. 2. (i) n = 2 Fig. 2. (ii) n = 3

Taking the case n = 2 (Fig. 2 (i)) first, we see that the proof of Theorem 5.1 (i) shows
that a lies on the line segment P Q ( n o t e also that the points splitting A C, B D in the
ratio 2 : 1 - ) . lie on PQ) and clearly cb is less than ~(c) on this segment by convexity.
In the case (ii) a lies in the shaded area L. If n = 3 (Fig. 2 (ii)) the set M is the hexa-
gonal area enclosed by the points P Q R S T U and lying on the plane
V1 -{- U2-~/93 = C 1 -~-C2-{-C3,
while L is the convex hull of M and its projections onto the coordinate planes, so
that it is part of the cube of side cl shown.
We now give some sufficient condition for polyconvexity. L e t n = 2 or 3. To
keep things simple we consider stored-energy functions W ( F ) defined on sets of
the form U = {FeM"• det F e K } , where K~_~+ is convex.
 Existence Theoremsin Non-LinearElasticity 367

Theorem 5.2. I f n = 2 let

W ( F ) = ~, (v1, v2, v1 v2), (5.8)
where vl, v 2 are the singular values o f F ~ U, and where ~" ~2+ • K - - . ~ is convex
and satisfies
(a) q/(x 1, x 2 , 6 ) = ~ ( x z, x 1, 6) f o r all x 1, x 2 ~ + , 6 ~ K ,
(b) ~(Xl, x2,6) is nondecreasing in xl, x2.
/ f n = 3 let
W(F)= I~t(Vl, v2, v3, v2 v3, v3/)1, Vl v2, Vl v2 v3), (5.9)
where vl, v 2 , v 3 are the singular values o f F e U, and where ~ : ~16 • K --. ~t is convex
and satisfies
(a) ~ , ( P x , P y , f ) - - ~ , ( x , y , 6) f o r all P, P e ~ 3 and all x , y ~ 1 3 , 6 ~ K ,
(b) q/(xl, x2, x3, Ya, Y2, Y3,6) is nondecreasing in each x i, y~.
Then W is polyconvex on U.
Proof. We give the proof for n=3. Define G: E 1 x K - - , ~ by
G(F, A, 6)= ff(v 1, ve, v3, al, a2, a3,0), (5.10)
where vi, ai are the singular values of F, A. G is well-defined by (a). Clearly
W ( F ) = G (F, adj F, det F) (5.11)

for all F e U. It remains to show that G is convex. Let F, H, A, B e M 3 • 3, 6, p ~ K ,

2 ~ [0, 1]. Let F, H, A, B, 2F + (1 - 2) H, )LA+ (1 - 2) B have singular values vl >
v 2 > v 3 , h 1> h 2 ~ h3, a 1 > a z > a 3 , b 1 > b 2 >=b3, U1 ~ U2 ~ U3, d I > d z > d 3, respective-
ly. Let w = 2 v + ( 1 - 2 ) h, e = 2 a + ( 1 - 2 ) b. Using Lemma 5.3, Lemma 5.4, (a) and
(b) we see that
G (2F + ( 1 - 2 ) H, 2A § 26§ ~ (u, d, 2 c5§ (1-2)/~)
<if(w, d, 2 ~ + ( 1 - 2 ) p)
__< (w, c, + (1 u)
< 2 ~(v, a, ~) § (1 - 2) ~(h, b, p)
= 2G(F, A, 6 ) + ( 1 - 2 ) G(H, B, p).
[]

A special case of a function ~ satisfying the hypotheses of Theorem 5.2 for n = 3 is

r a, r 2(a) + (5.12)
where the ffi are convex, and where ~01,if2 are symmetric and nondecreasing in
each variable. We use this example in Section 8,

6. Sequential Weak Continuity of Mappings on Orlicz-Sobolev Spaces

Definition 6.1. Let X and Y be real Banach spaces. A map f : X --+ Y is sequentially
weakly continuous if and only if X r ~ X in X implies f ( x , ) ~ f ( x ) in Y.
(In general a nonlinear sequentially weakly continuous map f : X ~ Y is not
continuous with respect to the weak topologies on X and Y (see BALL [1]).)
368 J.M. BALL

Sequentially weak 9 continuous maps are defined analogously. In this section

we study the case when X is an Orlicz-Sobolev space and Y= L1(f2). The reader
unfamiliar with the theory of Orlicz and Orlicz-Sobolev spaces can replace them
everywhere by the corresponding Lebesgue and Sobolev spaces, the results for
which are indicated in parentheses in some of the theorems which follow. There
will be no great loss of generality in doing so, at least as far as most of the examples
in Section 8 are concerned. Some of the results proved here in the framework of
Orlicz-Sobolev spaces are proved for ordinary Sobolev spaces in BALL [2].
We first obtain necessary conditions.

Theorem 6.1. (MORREY[1]). Let m and n be arbitrary. Let ~ : ~m • ~ , X M"• --} ~I

be continuous. Then
I (u, f2)= ~ q/(x, u(x), 17u(x)) dx
12

is sequentially weak 9 lower semicontinuous on WI'~(f2) (i.e., ur *-~u in Wl'~(f2)

implies l(u, f2)< li_m I(u,, f2)) if and only if O(x o, Uo, ") is quasiconvex on M "•
r ~o3

for each Xoe~l", UoeYl".

Corollary 6.1.1. Let n= 1, 2 or 3 and let Ca:M "• ~ be continuous. Then the
map u~-+J(u, f2)= ~ (a(Vu(x))dx is a sequentially weak * continuous map from
W 1'~176
(f2) to ~1 if and only if c~ has the form (4.2).

Proof. If the given map is sequentially weak 9 continuous then by the theorem
both ~b and -q~ are quasiconvex, so that by Corollary 4.1.1 ~b has the form (4.2).
Conversely, let ~b have the form (4.2) and let u, *, u in W 1' oo(f2). The sequence
q5(17ur (.)) is bounded in L~ (O), so that there exists a subsequence u, of u, such that
~)(Vu~(')) *-0 in U~ Let ~ : ~ - , ~ be continuous and define q51(x,F)=
+ qS(F) e(x) so that q~l is quasiconvex. By the theorem
q~(Vu~,(x)) c~(x) dx ---, ~ (a (Vu(x)) a(x) dx.
I2 12

The arbitrariness of a implies that 0 = q~(V u (')), and hence ~b(Vu,('))*--~ (~ (Vu ('))
in L~(f2). The results follows. []

Corollary 6.1.2. Let n = 1, 2 or 3 and let A be an N-function (cf. Section 2). Let
(a : M" • ~ ~1 be continuous and such that u ~--~(Vu(" )) is a sequentially continuous
map of W 1 La(f2) with the weak 9 topology into I~(f2) with the weak topology. Then
c~ has the form (4.2).

Proof. The hypotheses imply that u ~ J(u, f2) is a sequentially weak 9 continuous
map from W 1' ~ (f2) ~ ~. []

Remark: MORREY'S proof of Theorem 6.1 may be easily adapted to show that
if O is a bounded open subset of ~m (m arbitrary) and if q5: ~ -+ ~, then the map
0 ~ qS(0(.)) is sequentially weakly continuous from LP(Q)--} L1((2) if and only if
q5 is affine. For details see BALL [2].
 Existence Theorems in Non-Linear Elasticity 369

We turn now to sufficient conditions. Our results are based on the following
elementary identities for C 2 functions u:
n = 2 : det Vu =(U 1 U2 2),i --(U 1 U2,1),2 (6.1)
n = 3 : (adj ~,,~_t,,~+2,,~+1
" ' 1 i --~," ~
~ ,~+11,~+2 --( uI+2 u~+l ,a~+2J,~+l (6.2)
det Vu=[u~(adj Vu)i],j=[uX(u ~, 2 u~, 3 -u2 ,3 u , 2 ) ] , 1 (6.3)
+[-ul(u2,3 U3,1--U2,1 U3,3)],2+[ul(u2,1 U3,2--U2,2 U3,1)],3

(In (6.2) there is no implied summation, and the indices are to be taken modulo 3.)
Lemma 6.1.
(i) n= 2: I f u~ WL2(O) then det VueLX(O) and formula (6.1) holds in ~'(0).
(ii) n = 3: (a) I f u ~ W L 2(0) then adj V u ~ L1(0) and formula (6.2) holds in ~'(0).
(b) Let A,B be N-functions with A ~ t z, A~-B. I f u~W~ EA(O) and
adj Vu~En(f2) (e.g., u~WI'p(O), p>2, and adj ~zu~/e'(o)) then
det Vu~L ~(0) and formula (6.3) holds in ~'(I2).
Proof. (i) That det 17u~ L1(f2) is obvious. Formula (6.1) holds in ~ ' (0) if and only if

~(det Vu)c~dx= S(ulu2 1 ( a , 2 - u l u 2 2 c ~ , l ) d x for all q~E~(O). (6.4)

t2

But (6.4) holds trivially if u~ C~ and C~ is dense in W 1' 2(0) in its norm
topology. Since both sides of (6.4) are continuous functions of u e W ~' 2(O), (6.4)
holds for u e W 1' 2(0).
(ii) The proof of (a) is identical to that of (i). Let w be defined by wj = (adj Vu)~.
To prove (b) we first note that uliEEa(O), wJ~Ex(O) so that det Vu~LI(O). We
next show that d i v w = 0 in a weak sense; i.e.,

~wJ~,jdx=O for all ~be~(O). (6.5)

If u~C~(f2), then d i v w = 0 and (6.5) holds. Since C~~ is dense in Wl'2(O),

(6.5) holds for any u~ W 1Ea(O).
To show that (6.3) holds in ~'(O) it is thus sufficient to prove that

~uljwJ(adx= - ~ulwJc~,jdx for all ~ b ~ ( O ) (6.6)

(2 f2

whenever w~EB(O ) and satisfies (6.5).

By the results of DONALDSON• TRUDINGER[1, Thm. 22] there exists a sequence
Utk)eC~(~2) with Utk)~U in WIEa(O). Let p ~ ( ~ a ) , p__>0, S p ( x ) d x = l , and
~3
define pke~(~ a) by p k ( X ) = kp(kx). Extend w by zero outside O, so that w~EB(~la).
DONALDSON & TRUDINGER [1, Lemma2.1] show that the convolutions pk*W
are in EB(~ 3) and Pk * W ~ W ill EB(~ a) as k ~ ~ . Fix q~e~(O). Then if k is large
enough, (6.5) implies that
div (Pk * W)(X)--- ~ Pk, j ( X - y) wJ(y)d y = 0 (6.7)
~3
370 J . M . BALL

for all xesuppqS. Therefore if Sc2~ 3 is an open ball containing suppqS, then
u~L,j(Pk * WJ) 6 dx = idiv (U~k)(pk* W)6)dx - S U~)(pk * W~)q~,j dx
12 S f~

= - * wJ) 4oj d x .
f2

Since ,~I-<B we obtain (6.6) by letting k--. oe. []

Remark. Note that A > t 2 if and only if A ~ A . This may be proved directly
from the definition of ,4.
The functions det Vu (n=2), adj Vu and det Vu (n=3) can be given a meaning
as distributions under weaker conditions than those of Lemma 6.1. We thus
define the distributions Det Vu (n=2), Adj Vu and Det Vu (n=3) by

n=2: Det g u = ( u I U2,2),2 - - ( U 1/,/21),2 , (6.8)

n = 3 : (Adj 17"~-l''i+2''i+1
----]i - - ' , ~
~
~',~t+ll,e+
["i+2tti+l
2 --'~ ~
~
~,a+2],a+l ~
(6.9)
Det Vu = [u ~(Adj Vu){],j, (6.10)

when these distributions are meaningful. Obviously if u satisfies the hypotheses

of Lemma 6.1 then these distributions may be identified with the L~(f2) functions
det Vu (n=2), adj Vu and det Vu (n=3) respectively.
Let A be an N-function. Following DONALDSON & TRUDINGER [1], we let
A-l(t)
ga(t)- tl+i/,, t>O, (6.11)

where A-1 denotes the inverse function to A on [0, oe). If A satisfies

1 o9
~ga(t) d t < o o , ~gA(t)dt=oO, (6.12)
0 1

then we define the N-function A* by

It[
(A*) -1 (Itl)= j ga(s) ds. (6.13)
0

Note that for A (t) = It Ip, 1 < p < n, we have

gA(t)=tp n , A*(t)= . (6.14)

Lemma 6.2. Let f2 satisfy the cone condition.

0o
(i) n = 2 : Let A be an N-function satisfying either ~ gA(t)dt< ~ , or both (6.12)
1
and ,4<A*. I f u~ W1LA(Q) (e.g., if uE Wt'r then u1 u 2, 2 and u 1 u 2,1
belong to L1 (f2), so that Det Vu exists as an element of ~'(t2).
(ii) n = 3 : (a) Let A be as in (i). I f u~W1LA(~2) (e.g., if u~Wt'k(f2)), then
Ui+2Ui+x, ~ + 1 and ui+2u i+1,a~+2 belong to LI(O) so that (Adj Vu)~ exists
as an element of ~'(Q).
 Existence Theoremsin Non-LinearElasticity 371
oo

(b) Let A , B be N-functions with either A satisfying Sga(t) d t < oo,

i
or with A satisfying (6.12) and with B ~ A * , I f ueW1LA(Q) and
Adj Vu6L~(~) (e.g., if u~WI'P((2), Adj Vu6Lq(~) with p > 1, q > l and
1+1<,,
- - - - = g ) , then uX(Adj Vu){~LI((2), so that Det Vu exists as an ele-
P q
ment of ~'(g2).
oo

Proof. (i) If S gA(t)dt<ov then the imbedding theorem of DONALDSON & TRY-
1
DINGER [1, Thin. 3.2-] implies that ul~U~ while if A satisfies (6.12) the same
theorem implies that uI~LA+(f2). The result follows by Young's inequality.
(ii) This is proved similarly. []
Remark. If f2 is an arbitrary bounded open set then Lemma 6.2 holds with
L1(g2) replaced by L~o~(s
The main result of this section is the following:
Theorem 6.2. oo

(i) n = 2 : Let A be an N-function satisfying either ~ ga(t)dt < oo, or both (6.12)
1
and AT<<A*. I f u r ~ u in W~LA((2) (e.g., /f ur-~u in wa'p(Q), p>~),
then Det Vu, --+Det Vu in @'((2).
(ii) n = 3 : (a) Let A be as in (i). I f u r - ~ u in WI LA((2) (e.g., if u r ~ u in WI'p(~),
P > 3) then (Adj Vu,)~--+ (Adj Vu)~ in ~'((2).
oo

(b) Let A, B be N-functions with either A satisfying ~ gA(t)dr< oo, or

1
with A satisfying (6.12) and with /~<<A*. I f u~ * ~u in W1LA(O) and
Adj Vu, * - Adj Vu in LB((2) (e.g., if u , ~ u in W 1'P(~2), Adj Vur-~Adj Vu
1 1
in I3(f2) with p > l , q > l and - - + - - <45), then D e t V u , ~ D e t V u in
~'(~), P q

Proof. (i) Fix ~be~(~2) and let O' be an open set with Q = Q ' z s u p p ~ b and such
that the imbedding theorems of DONALDSON 86 TRUDINGER hold for O' ~. Then
since 11U~IIW~LA~e')is bounded and ur ~ u in Lt (~'), it follows (see KRASNOSEL'SKII
& RUTICKII [1, p. 132]) that u,-+u in L~~ ') or L,+(O'). Therefore the H61der
inequality and the boundedness of ]]u,[lw,L~(a,) imply that
I(ur r,2 2 --U U, 2)~ d x = I (U~--ul) U2 2 +dxq-- i ul (Ur,22 --U2,2)~) d x
.Q (a' .Q'
tends to zero as r--, oo. Hence u~l u~, 1 U~,1
2 z--+ u l u 2,2 in ~'(~2). Similarly u~ 2 _.+b/1 U21
in ~'(Q). The result follows. The proof of (ii) is similar. []
Corollary 6.2.1.
(i) n = 2 : Let A be as in Theorem6.2(i) and let u~W1LA(~"2). Then
Det Vu=(uZul,1),2-(u2ul, 2),a in ~'(Q). (6.15)
+ We may take for Q' a finitesubcover by open balls of the closure of supp ~b.
372 J.M. BALL

(ii) n = 3 : (a) Let A be as in Yheorem6.2(iia) and let u~W1LA((2). Then

(Adj [7U)~=(R i+1~/i+2 , a + 2 1] , ~ + 1 - - ( u~+I
" ,,i+2
v" , a + l l~, a + 2 in ~'(f2). (6.16)
(b) Let A, B be as in Theorem 6.2 (ii b) and let u ~ W 1LA(Q), A d j [7U~ L B ( ~ )-
Then
Det Vu=[uZ(Adj Vu) 89 17u~],~ in ~'(f2). (6.17)
Proof. (i) Let q~E~(f2) and let f2' be an open set with f2~f2'=supp~b and satis-
fying the segment property. Then by the results of Goss~z [1, Thin. 1.3] there
exists a sequence ur~ C~(f2 ') with ur*-~-~u in W 1LA(f2' ). Clearly (Det Vu,)(~b)=
2 1 2 1
[(u,u,.O,z-(U~U,,z),x](4~ ). Letting r ~ we obtain from the theorem that
(Det Vu)(~) =[(u 2 u 1 l ) , z - (uZH1,z),l](~b), and the result follows.
(ii) The proof of (a) is identical to that of (i). The proof of (b) is similar to
that of Lemma6.1(iib), the principal change being the use of Lemma 1.6 of
GOSSEZ [1] to show that if w~LB(~2), then Pk*W--*-~W in Ln(f2). We omit the
details. []
Corollary 6.2.2.*
(i) n = 2 : The map u~-+detVu: WI'P(f2)--+LP/Z(f2) is sequentially weakly con-
tinuous if p > 2.
(ii) n = 3 : (a) The map uF-+adj Vu: WI'P(f2)--t LP/Z(f2) is sequentially weakly con-
tinuous if p > 2.
(b) The map u w+det V u: W L P(f2)~ Lp/3 (f2) is sequentially weakly con-
tinuous if p > 3.
Proof. We just prove (iib). Let p > 3. It is clear from the H61der inequality that
if u ~ W a' p (f2), then det V u ~ Lp/3 (O). Let u ~ u in W t' p(f2). Then adj Vu, is bounded
in the reflexive space Lp/2`O"), a n d hence by the theorem (part (iia)) adj V u , ~ a d j 17u
in LP/2(~2). But det 17u, is bounded in the reflexive space Lp/3(f2) and thus by
part (iib) of the theorem det Vu~-~det Vu in LP/3(f2). []
Warning. The distributions det Vu (adj Vu) and Det Vu (Adj Vu) need not be
the same even if the former is a continuous function, as the following example
shows.
Example 6.1, n = 2 or 3.
Let r=[x[ and let R(r) be a smooth real-valued function on (0, 1].
Let f 2 = { I x l < l } and define u: f 2 ~ " by
R(r)
u(x)= x, r>O, u(O) arbitrary. (6.18)
r
Then for r > 0 we have rR'-R
u=
'iR~6~i+' r r3 3~exixP (6.19)
and
R,-1R,
det Vu = r,_~ (6.20)
Note added in proof This corollary follows from results of Y. G. RESnETNYAK [1, Thm. 4],
[2, Thin. 2]. Theorem 4 in BALL [2] is also essentially a consequence of RESHETNVAK'Swork, to which
I would have referred had I seen it in time.
 Existence Theorems in Non-Linear Elasticity 373

Let re(O, 1), B~={fxl<r} and let qSe~(f2). Then

c)u{=dx=-~ c~u'n~dS- I (~ u~dx" (6.21)

But
(auln~dS = ~ c~(rx)ui(rx)n,r "-1 dS.
OB~ Of2

Provided r "-1 R(r)--+ 0 as r--+ 0 and u, V u ~ L 1(O) we therefore obtain from (6.21)
that
I - I 4, dx,
O f~

where we have used the dominated convergence theorem. Thus under these
conditions Vu, as defined by (6.19), is the matrix of weak derivatives of u.
In particular, letting R(r)= 1 +r, we find that u~ Wa'P(O) for any p<n. Now
let n=2. Let ~be C Oo([0, 1]) take the values 1 and 0 in neighbourhoods of r = 0 and
r = 1 respectively. Then (~(x)d-~-ef@(lx[)belongs to ~(f2), and
1
~(det Vu)e~ dx = 2 rc ~(1 + r) ff (r)d r,
I2 0
1
(u~u2,~~b,2 - u ~u2.2 qS,1)dx = -7z ~ (1 + r) 2 if' (r)d r.
f2 0

Hence formula (6.1) does not hold in this case, so that det V u # D e t Vu. Note
also that if p < 2 there is no sequence of C~~ functions ur such that u,~-u in
Wt'P(Q) and det Vu,~-det Vu in LI(f2), since such a sequence would satisfy

det Vu, dx=4rc,

whereas
Sdet V u d x = 3 n .
~2

When n = 3 , a similar calculation shows that u e WI'p(~'~) for p < 3 , adj VueIY(O)
for q < 3~, but that (6.17) does not hold.
In the above example Det 17u has an atom at x = 0 . I do not known whether
det V u = D e t Vu if Det Vu is a function.

7. Existence Theorems
We have already stated the result of MORREY, Theorem 6.1, which shows
that for a continuous integrand quasiconvexity is necessary and sufficient for
sequential weak .lower semicontinuity in W 1"~(f2). MORREY 1-1,2] has also
given sufficient conditions for sequential weak lower semicontinuity in Wl"(f2),
s > 1. For purposes of comparison, and for future use, we given an extension of
his results due to MEYERS [1].
Theorem 7.1. Let f: f2 x ~n • Mn• ~ be continuous, and let f (x, u, ") be quasi-
convex for all x6f2, u ~ ~. Suppose there exist real constants Ki>O (i=1,2),
s > l , 0 < 7 < 1 and a function beU(f2) such that
374 J.M. BAtt

(i) f ( x , u, F)>b(x),
(ii) I f ( x , u + v , F + H ) - f ( x , u , F ) l < K~ {l +(luL+lvl+lFl+lHI)S-~'}([v[+lHI) ~',
(iii) If(x + y, u, F ) - f ( x , u, F)l-<_ K 2 { 1 + [FIq rt(lYl),
.for all values of the various arguments, where rl: ~t +--,~ + is a continuous in-
creasing Jimction with ~/(0)=0. (Here and elsewhere [F I denotes any fixed norm
on F E M" • ".) Then
I(u, f2)= ~f(x, u(x),, Vu(x)) dx
12

is sequentially weakly lower semicontinuous on WI'~(O).

Remarks.
1. Let XoS(2. Then conditions (ii) and (iii) imply in particular that

If(x,u,f)-f(x,O,O)l<~K~[l+(lul+lF[)S-~']{Lu[+lFIp' (7.1)
and
If(x, O, O) - f ( x o, 0, 0)[ < K 2 r/(I x - Xo]). (7.2)

Combining (7.1) and (7.2) we see that

(i)' if(x, u, F ) J < K o {1 +(lyl+]FI)~},
for all (x, u,F), where K o > 0 is a constant. Conditions (i)', (ii) and (iii) are
MEYERS' continuity and growth conditions for the function f - b , while (i)
implies a further hypothesis of his theorem.
2. If s < n then the growth conditions with respect to u may be weakened by
use of the Sobolev imbedding theorems; the reader is referred to MEVERS [1]
for details. An extension to an Orlicz-Sobolev space setting could also prob-
ably be made.
In order to prove existence theorems by use of Theorem 7.1, it is necessary
to make, in addition to conditions (i)-(iii), a coercivity assumption on f. Typically
we might assume that s > 1 and
(iv) f ( x , u , F ) > K 3 l F I S + b ( x ) , where K 3 > 0 .
The conditions (i)-(iv) are extremely restrictive with regard to applications to
nonlinear elasticity. Firstly, (ii) precludes any singular behaviour of f (for ex-
ample (0.5)). Secondly (i)' and (iv) together rule out integrands typified by the
example
f (F) = IFI~+ [det FL' (7.3)

with nr>s; we shall see that many such integrands belong to a physically inter-
esting class included in our existence theorems.
Definition7.1. Let D ~ Y t k be open. A map G I : D • ~ is said to be of
Carath~odory type if
(a) for almost all x~D, G1(x, ") is continuous on ~ , and
(b) for all a ~ v, G1 (', a) is measurable on D.
 Existence Theorems in Non-Linear Elasticity 375

We shall use the following lower semicontinuity theorem, which is a special case
of a result given by EKELAND & T~MAM [1, Thm. 2.1, p. 226]. For related results
see CESARI [1, 2, 3].
Theorem 7.2. Let GI: f2 x (~" • ~ / ~ ) ~ be of Carathkodory type (Definition 7.1
with v = n + a) and satisfy
G 1 (x, u, a) > ~b(ra l) (7.4)
for some N-function ~. Suppose that GI(X, u," ) is convex on ~l~ for all x~f2, u6~t".
Let a,-~a in L 1(f2) and let {u,} be a sequence of measurable functions with u, ~ u
almost everywhere in f2. Then
Ga (x, u(x), a(x))dx< lim S G1 (x, ur(x), a,(x))dx. (7.5)
I2 r~ Ft

We now describe the main ingredients of our existence theory, starting first
with those relevant for compressible materials.
For each x~f2 let W(x) be a nonempty convex open subset of E = ~ s("), such
that for each a~E the set W - l ( a ) ~ f { x ~ f 2 : ae W(x)} is measurable. We impose
as a local constraint on our variational problem that, in a sense to be made precise
later, T(Vu(x))~ W(x) almost everywhere in f2, where we are using the notation
of Section 4. In applications to nonlinear elasticity W(x) will often have the form

W(x) = {a ~ E: C1 (X, a) < K,a (x), Is2 ( x ) ,~ - - c 2 ( x , a ) } , (7.6)

where q , c 2 : ~ • E ~ ~ are of Carath6odory type and convex with respect to

a s E, and where tq, ~:2 : f2 ~ ~ are measurable. Examples of relevant choices of
ci, xi (i = 1, 2) are
1. c i is arbitrary, x a _-- + oo, x2 -- - oo (no constraint).
2. (n = 3) - c 2 (x, F, A, 6) = ~, ~c2 -- 0 (corresponding to the continuity condition
det Vu(x)>O and an additional unilateral constraint, absent if Ka = + oo, on the
measure of strain c a (x, 17u, adj Vu, det Vu)#).
We now make continuity, growth and polyconvexity hypotheses on the
integrand. Because of the nature of the growth conditions, and because we wish
to consider situations in which the distributions adj 17u and Adj Vu, det Vu and
Det lTu may be different, we make these hypotheses on the associated function
G(x, u, a) (cf. (4.3)).
Let 5 e = { ( x , u , a ) E f 2 x N " x E : asW(x)}. Let G: 5 e ~ N be such that **
(H x)**e G is continuous with respect to u, a on 5~,
(H2) for all u ~ " , aeE, G(.,u,a) ismeasurableon W-a(a),
(H3) for almost all xeO, G(x, u, a ) ~ +oo as a--,OW(x), the convergence
being uniform with respect to u in any bounded subset of N",
(//4) (Polyconvexity) for each xef2, ueYt", G(x, u, ") is convex on W(x),
(Hs) (Coercivity)
~* Such constraints are by no means unrealistic. A unilateral constraint at large strains might be
relevant, for example, for a mixture of elastic materials with one or more constituents possessing
limited extensibility (el NIEDERER [1]). See also the comments in Section 10.
# * For simplicity we suppose that G is defined for all u e,~".
9 * * Weaker conditions are possible (cf EKELAND&; TI~MAM[1]).
376 J . M . BALL

n=l: there exists an N-function A, and a function b 9 L1 (~) such that

G(x, u, F)>b(x)+A(F) for all (x, u, F) 9 (7.7)
oo
n = 2 : there exist N-functions A,B, with A satisfying either ~ gA(t)dt<~ or
1
both (6.12) and ~i<<A*, and a function b9 such that
G(x,u,F,f)>b(x)+A(IFI)+B(6) for all (x, u,F, f) 9 ; (7.8)
oo
n = 3 : there exist N-functions A, B, C satisfying either ~ ga (t) d t < oo or the con-
1
ditions (6.12), A-<<A* and/~<<A*, and a function b9 such that
G (x, u, F, H, fi)
(7.9)
~b(x)+A(IFI)+B(IH[)+C(6) for all (x,u,F,H, 6)9162
If we define G 1 : f2 • (~" x E) ~ ~ by
Gl(x,u,a)=G(x,u,a)-b(x ) if (x,u,a) 9
= + ~ otherwise,
then clearly G 1 is of Carath6odory type and satisfies (7.4) for some N-function cb.
We define the admissibility set st by
n = 1 : d = {u 9 W 1LA(12): Fu(x)E W(x) almost everywhere in ~2},
n = 2 : ~r {u 9 W a LA(f2): Det Vu9 (Vu(x), Det Vu(x))9 W(x) almost every-
where in f2},
n = 3 : d={u 9 Adj Vu9 Det Vu9 (Vu(x), (Adj Vu)(x),
Det 17u(x))9 W(x) almost everywhere in f2}.
The equivalence classes in d under the equivalence relation
u~v if and only if u-t~9
are termed the Dirichlet classes in d .
If u 9 d then it follows from results of E KELANDt~ Tf~MAM[1, Prop. 1.1, p. 218]
that d(u) exists and is finite or + oo, where

J(u) d=cf~ G(x, u(x), Vu(x)) dx if n = 1,

de~ I G(x, u(x), Vu(x), Det Vu(x))dx if n = 2 ,

~2

d~ ~ G(x,u(x), Vu(x),AdjVu(x),DetVu(x))dx if n = 3 .
f2

We are now in a position to present our first existence theorem, which in-
cludes as a special case the displacement boundary-value problem of nonlinear
hyperelasticity. For n = 1, of course, the result is well known.
Theorem 7.3. Let G satisfy (Ha)-(H5) above. Let ~ be a Dirichlet class in ~r and
suppose that there exists ul 9 with J ( u l ) < ~ . 7hen there exists uo 9 that mini-
mizes J(u) in c~.
 Existence Theorems in Non-Linear Elasticity 377

Proof. We just give the proof for n = 3, the other cases being easier. It is sufficient
to establish the existence of a minimizer in ~g for

J ( u ) ~ f ~ G 1 (x, u(x), Vu(x), Adj Vu(x), Det Vu(x))dx. (7.10)

f2

J(u) is bounded below on ~. Let u r be a minimizing sequence from ~. By (7.9)

the quantities
~A([Vu,(x)[)dx, i B([Adj Vu,(x)l)dx, ~ C(DetVu,(x))dx
I2 12 12

are bounded independently of r. The Poincar6 inequality for WILA(O) (GossEz

[1, p. 202]) implies (cf KRASNOSEL'SKII 8r RUTICKII [1, p. 131], EKELAND &
T~MAM [1, p. 223]) that for a subsequence {us} we have
uj *-*-'uo in W1LA(O), Vuj~VUo in D(Y2),
us ~ u o almost everywhere,
Adj Vuj*--*--H in Ln(f2), Adj V u s ~ H in L1(12),
and
DetVuj*--~3 in Lc(O), DetVu~3 in 12(0).
By Theorem 6.2(ii) H = A d j Vu o and 6 = Det Vu o. By Theorem 7.2,
J(uo) < lim J(u).
j ~ oo

Since Gl(x, u, a)= +oo if a6 W(x) it follows that (Vuo(x), Adj Vuo(X),
Det Vuo(X))e W(x) almost everywhere. Thus u o ~ and the result follows. []
We now give a modified version of Theorem 7.3 for the case in which n = 3
and G is independent of 6. The proof is similar and is omitted.
Theorem 7.4. Let n= 3. In the definitions of W(x) and 5P replace E by E 1 =
M 3 • 3 x M 3 • 3. Let G: 5~ ~ ~ satisfy' (HI)-(H4) and the following hypothesis:
oo

(H6) There exist N-functions A , B with A satisfying either S gA( t)dt < oO or both
(6.12) and/I<<A*, and a function beLl(f2) such that 1
G ( x , u , F , H ) > b ( x ) + A([FI)+B([It]) for all ( x , u , F , H ) e S C
Define
~r {ue W' LA(f2): Adj VueLB(Y2), (Vu(x), Adj Vu(x))e W(x)
almost everywhere in f2}.
Let
J(u) = ~ G (x, u (x), V u (x), Adj Vu (x)) d x. (7.11)
~2

Let cg be a Dirichlet class in d , and suppose that there exists u 1ecg with J ( u l ) < oQ.
Then there exists U o ~ that minimizes J(u) in cg.
Remark. Similar modifications to Theorem 7.3 can be made when n=2, and
when n = 3 and G is independent of both H and 3; the details are left to the reader.
378 J.M. BALL

Next we give an existence theorem for the displacement boundary-value

problem in three dimensions for an incompressible hyperelastic body.
Theorem 7.5. Theorem 7.4 remains valid f d is redefined by s~ = {u6 WI La([2):
AdjVuELB((2 ), (Vu(x),AdjVu(x))~W(x) almost everywhere in Q, Det Vu(x)=l
o~

almost everywhere in Y2}, provided that if ~ ga(t)dt = ~ , we make the extra assump-
tion/~ K< A*. 1

Proof. Let {ur} be a minimizing sequence from c~. Then {Det Vur} is bounded in
L~ independently of r. As in the proof of Theorem 7.3 we may extract a sub-
sequence {uj} ~_c~ with, among other properties,
DetVuj * ,6 in L~
Clearly 6(x)= 1 almost everywhere. By Theorem 6.2(ii), 6 = Det Vu, and the result
follows. []
Remark. A variety of 'weakly closed' constraints can be treated in this way.
Analogous results hold for n = 2.
For the remainder of this section we restrict our attention to the cases n = 2, 3
and we impose growth hypotheses that are partly of polynomial type. This will
enable us to work in Sobolev spaces, rather than in Orlicz-Sobolev spaces. It
would be possible to extend most of our results to an Orlicz-Sobolev space setting.
Such an extension would involve the use of trace theory for Orlicz-Sobolev
spaces (see DONALDSON 8r TRUDINGER [1], FOUGI~RES[1], LACROIX[1]).
For ease of reference we now restate hypotheses (Hs) and (H6) in modified
form. Later we shall put extra restrictions on the constants appearing in these
hypotheses.
(HT) n = 2 : there exists an N-function B, real constants K 1 >0, K2>0, 7> 1, s > 1,
and a function b~/2(f2) such that
G(x, u, F, 3)> b(x)+ Kl lFl~+ KzlulS + B(6)
for all (x, u, F, 3)~5t. (7.12)

n = 3 : there exists an N-function C, real constants K I > 0 , K2>0, 7>1,

/~> 1, s > 1, and a function bs/2(~2) such that
G(x, u, F, H, 6) > b(x) + KI (IF[~+ IH[u) + K 2 lu]s + C(O)
for all (x, u, F, H, 3)e~. (7.13)

(H8) n - 3 : there exist constants Ka >0, K2>0, 7> 1,/~> 1, s>_ 1, and a function
b~12(O) such that
G(x, u, F, H) > b(x) + KI (IFI~+ IHIu) + K 2 [u[s
(7.14)
for all (x, u, F, H)s 6e.

Mixed displacement traction boundary-value problems

Theorem 7.6 (cf. Section 1A). Let O satisfy a strong Lipschitz condition. Let
0~~= ~-~1 k-)~'~2, let 001 c~0~2 = ~a, and let 0~"~1 and 902 be measurable as subsets of
 Existence Theorems in Non-Linear Elasticity 379

Of2 with Of2t having positive measure. Let ~: Of21 --* ~t" be measurable and let
tR~//(Sf22) with cr> l. Let G: 5 / ~ satisfy hypotheses (H1)-(H+) and (H7). I f
t
4 Y
n=2, let V>~, a = ~ - / f ? < 2 , a > l / f ? = 2 , a = 1 / f ? > 2 , K 2 = 0 , and

~r = { u ~ W 1' ~(f2): Det Vu e L~ (f2), ( Vu (x), Det Vu (x)) ~ W(x) almost

everywhere in f2, u = ~ almost everywhere in 8f2~ }.
1+1_< 4 27'
I f n = 3 ' l e t T > ~ ' 7 # 3, ~ = 3 - /f? < 3, a > l / f T = 3, a = l / f ? >3, K z = 0 , and

~r = {u~ W 1' ~(f2): Adj Vu~LU(f2), Det Vu~Lc(f2 ), (Vu(x), Adj Vu(x), Det Vu(x))
W(x) almost everywhere in O, u = ~ almost everywhere in 8(21}.
Let
Jo ( u ) = J ( u ) - I u(x).tR(x)dS, (7.15)
0122

where the integral is defined in the sense of trace.

Suppose that there exists u16~' with Jo(ut)<oo. Then there exists U o ~
that minimizes Jo(u) in d .
Proof. We give the proof just for n = 3.
By the trace theorems (cf MORREY [2], NE(~AS [1]) ~eL~(Sf2~) and there
exists k > 0 such that
Jlullw,,~(~>=k PlUllL='(0~) for all ue WX'~(O). (7.16)
Since ~3f2~has positive measure, a result of MORREY [2, p. 82] implies that there
exists kt > 0 such that
~lulTdx<kx[~lVul~dx+(~I~IdS) r] for all u~Wa'r(f2). (7.17)
f2 I2 0121

By (HT), (7.16) and (7.17) we have for arbitrary u e s r

Jo(u)> ~ b(x)dx + K 1 ~ IVu]~dx + K~ ~ [Adj 17ulUdx+ ~ C(Det Vu)dx
I2 f2 g2 f~

- ~ u(x).L,(x)dS
~f22

+ f~-~ lul d x - ~ - (o~,II~ldS)'+&Vllull[o'(oo,

dy 1 I[i~tl~'~(ea~)+ ~ C(DetVu)dx,

for e > 0 and d>0. Choosing e and d small enough with d=<k(e?) ~/r we obtain
Jo(u)>-_c+eo Ilull~,,~,+K1 ~ IAdj Vul~dx+ ~ C(Det Vu)dx, (7.18)
12 f2
where c and c o > 0 are constants.
* In the sense oftrace.
380 J.M. BALL

Let {u,} be a minimizing sequence for Jo- It follows from (7.18) that a sub-
sequence {u j} satisfies

uj-~Uo in W~'r(f2), uj~Uo in L"(t?f2),

u j---, u o almost everywhere in f2 and t?f2,
A d j V u ~ A d j l 7 u o in /~(f2), DetVuj * ,DetVu 0 in Lc(f2 ).
For J given by (7.10) it follows that

J(uo) < lim ](uj), (7.19)

j~oo
while
Uo(X) .-iR(x)dS = lim ~ uj(x).tR(x)dS. (7.20)
dot j ~ oo t3121

Noting that u 0 = fi almost everywhere in ~?f21we see, as in the proof of Theorem 7.3,
that Uoed. The result follows. []
Remark. In Theorem 7.6, and in the results below, the hypotheses on fi are
concealed in the assumption that d is nonempty.
Theorem 7.7. If n=2 let 7>2, a > l /f7=2, a = l /f7>2, K2=0. If n=3 let y>2,
1 1 27'
- + - < 1, c r = ~ - / f 2 <7 <3, t r > l / f 7 =3, t r = l / f 7 > 3, K2=O.
7 //
Let the other hypotheses of Theorem 7.6 remain unchanged. Then Theorem 7.6
remains valid with Adj Vu, Det Vu replaced everywhere by adj Vu, det Vu respectively.
Proof. This is immediate from Lemma 6.1. []

Remark. In Theorem 7.7, and in those results below that concern the distri-
butions detVu, adj Vu it is only necessary for G to be defined on the set
{(x,u,a):x~f2, u ~ " , a ~ C o ( T ( M 3 • (see Section 4 and the remark
after Example 6.1).
Next we give the analogue of Theorems 7.6 and 7.7 for incompressible materials.
The proof is similar to that of Theorem 7.5 and is omitted. An analogue of Theorem
7.4 may also be simply proved.

Theorem 7.8. Let n = 3. Let I2, ~f21, 0f22, u, tR be as in Theorem 7.6. In the definitions
of W(x) and 6r replace E by E 1. Let G: b ~ --* ~t satisfy hypotheses (Ha)-(H4) and (Hs).
Either let 7, #, a, K z be as in Theorem 7.6 and let

~4 = {u~ W 1' r(~2); Adj Vu~LU(f2), (Vu(x), Adj Vu(x))E W(x) almost everywhere
in f2, u = ~ almost everywhere in t?f21, Det 17u(x)= 1 almost everywhere
in O}

or let ~, I~, tr, K 2 be as in Theorem 7.7 and let

d = {u e WI'~(O): adj Vu e Lu(f2), (leu(x), adj IZu(x)) ~ W(x) almost everywhere

in f2, u = ~ almost everywhere in ~f21, det 17u(x)= 1 almost everywhere in f2}.
 Existence Theorems in Non-Linear Elasticity 381

Let
Jo(u) = J ( u ) - ~ u(x). t~(x)dS, (7.21)

where Y is given by (7.11).

Suppose there exists ul6s~ with J o ( u t ) < ~ . Then there exists U o ~ that
minimizes Jo(u) in ~ .

Pure traction boundary-value problems

Theorem 7.9 (cf Section 1, A1). Let f2 satisfy a strong Lipschitz condition. Let
tR~L~ tr>=l. Let G: 5~--~t satisfy hypotheses (H1)-(H4) and (H7) with
K 2 > 0. Let K = rain (7, s).

K/
n=2: Let 7>5, a = ~ - /f l__<K<2, a > l /J" ~c=2, tr=l /f ~>2. l f s = l let
[[tR[[Loo(oa)< ko K2, where ko((2 ) > 0 is a certain constant. Let
d -- {u ~ WI' ~(O) 9 Det Vu ~ LB(g2), (Vu (x), Det Vu(x))~ W(x)
almost everywhere in ~}.

n=3" L e t T > ~'

3 ~1+ ~ -1< ~ '4 ~ = 52tr
-/f ~ <3, o > 1/fK=3, ~= 1/f ~ > 3.

l f s= 1 let IPtRJIL~(0a~<ko K 2, where ko(O)>0 is a certain constant. Let

,~r = {u e W t' ~(~2): Adj Vu ~ ~(~2), Det ~7u9 Lc(~),
(Vu(x), Adj Vu(x), Det Vu(x))e W(x) almost everywhere in f2}.
Let
Jo(u) = J ( u ) - ~ u(x). tR(X) dS. (7.22)
O~

Suppose that there exists u~es~ such that Jo(ua)<~. Then there exists u o 6 ~
that minimizes Jo(U) in d . If, in addition,

n=3:n=2:y>2 /
y>2, 1 + 1 _ < 1 , (7.23)
- 7 u - j

then the result holds with Adj Vu, Det Vu replaced everywhere by adj Vu, det Vu
respectively.
Proof. Let n=3. By using the hypothesis K 2 > 0 instead of (7.17) we obtain the
a priori bound
Jo(u)> c + eo IIVull~r + K~ f lAdj Vul" dx + ~ C(Det Vu) dx + c, I lu]~ dx,

1
where c1>0. If s = l then k 0 is such that Ilullw,,,(a~>~~ [lullL,(oa) for all
ue W1'l(s'2). By using the Poincar6 inequality (MORREV [2, p. 82-1) we can com-
plete the proof in the same way as for Theorems 6.6, 6.7. []
382 J.M. BALL

Theorem 7.10 (cf Section 1, A2). Let f2 satisfy a strong Lipschitz condition. Let
tn~IY (Of2). Let G: 5~ - - ~ satisfy hypotheses (H1)-(H4) and (H7) with K2-0.
7'
If n=2, let 7 >4~, a = ~ - if 7<2, a > 1 if 7=2, a = l if 7>2, e be a constant
vector and
~r = {u~ WX'r(f2): Det Vu~LB(f2),
(Vu(x), Det Vu(x))e W(x) almost everywhere in O, ~ u(x)dx=e}.
0

1 1 4 27'
If n=3, let 7> 3, T + ~ - < ~ , a = ~ - /f7<3, a > l /f7=3, a = l i f ? > 3 , e be a
constant vector, and

d = {u~ Wl'r(f2): Adj Vu~L~(f2), Det Vu~Lc(f2 ),

(Vu(x), Adj Vu(x), Det Vu(x))~ W(x) almost everywhere in 12, ~ u(x)dx = e}.
I2

Let p~EIA' (f2), let b o ~ l n and let Jo be given by

Jo(U) = d ( u ) - ~ pR(x) b o . u(x) dx - ~ u. tR dS. (7.24)

I2 O0

Suppose there exists u 1~ d with Jo(Ul)< oo. Then there exists U o ~ d that minimizes
Jo(U) in d .
If, in addition, we assume (7.23), then the result holds with Adj Vu, Det Vu
replaced everywhere by adj Vu, det Vu, respectively.
Proof. If u ~ d , then

Thus by a version of the Poincar~ inequality (MORREV [2, p. 83]) there are con-
stants k 3, k 4 > 0, such that

j"[ul' d x < k 3 +k4 ~lVul ~ dx (7.25)

f2 f2

for all u ~ d .
Applying (7.25) and the simple estimate

- ~nPRbo'udx>- [ ~ IlullL~.~+Td7
' 1 IIPRbol[~',,,~,] (7.26)

we again obtain the bound given for n---3 by (7.18). The rest of the proof follows
the usual pattern. []
Remark. We re-emphasise (cf. Section 1, A2 and Theorem 7.13) that for non-
linear elasticity, when G is independent of u, it is necessary to impose the extra
condition (1.31) in order to show that u o is in some sense a solution of the equilib-
rium equations.
 Existence Theorems in Non-Linear Elasticity 383

Mixed displacement pressure boundary-value problems

For these problems we restrict our attention to the case n = 3. The case n = 2
can be treated similarly.
Theorem 7.11 (ef Section 1B). Let n = 3. Let 12 satisfy a strong Lipschitz condition.
Let Or2= 0f2 t w S, Of2~c~X=c~, with ~f2 t having positive measure as a subset of
t?f2 *. Let ~ : Of21 _ ~ 3 be measurable, and let pc W 1' ~(0). Let G: 5e ~ satisfy
hypotheses (H1)-(H4) and (HT). Let K 2 =0. I f p(x) =-constant almost everywhere,
1 1 4 1 1
let 7>~, 7 - + ~ < ~ . I f p(x)~constant almost everywhere, let ~ > ~ , - - + - - <tt1 7

a n d / f t ~ + t ~ = l let []Vpl]L~t~ <ka, where k3=k3(K 1, K 2, 7, P, f2, 0f21) is a cer-

7 #
rain constant. Let

ezr = {u ~ W 1' ~(f2): Adj Vu ~ Lu(~2), Det Vu ~ Lc(f2 ),

(Vu(x), Adj Vu(x), Det Vu(x))~ W(x) almost everywhere in O,
u = ~ almost everywhere in Of2~}.
Let
J~ (u) = J (u) + P(u), (7.27)

where (cf 1.35))

P(u) ~f ~ [_p Det 17u +89 VU)~kUk] dx. (7.28)
f2

Suppose that there exists u t 6 d with J ~ ( u t ) < ~ . Then there exists U o ~ d that
minimizes J1 (u) in d .
If in addition
7>2, 1+1<1, (7.29)

then the result holds with Adj Vu, Det 17u replaced everywhere by adj Vu, det 17u
respectively.

Proof. It suffices to establish the bound

K 1
Jl(u)~C+Co rlull~v~,~,+~-~ IAdj Vul~'dx + 89~oC(Det Vu) dx (7.30)

for all u ~ d , where c, Co>0 are constants.

We use the Sobolev inequality

Ilutlw,,~(~j~k2 IlUllL~(~ for all u~ Wl'7(~e~), (7.31)

where k2>0 and v= 3 7 / 3 - y i f 7 < 3 , 1 < v < o o i f 7 = 3 , v=oo if ~>3.

* In applications we shall have X = dOr with the 3Q, as in Section 1 B.
r=2
384 J . M . BALL

For u e d , we can use (7.17), (7.31) to obtain

Jl (u)>=~b(x, dx+ (-~--e) ~ lVul' dx+ Kl ~ lAdj Vul" dx+ ~ C(Det lZu)dx
I2 f2

I2 k, r = l
1 ., ]

where e>0, d > 0 .

1 1
If p is constant, then clearly (7.30) follows. If p is not constant a n d - - + - - < 1,
7 #
then ~ >#' so that we obtain (7.30) by choosing e and d small enough. If p is not
1 1
constant and - - + - - = 1, then 7 = # ' and (7.30) follows similarly. []
7 /~
The above proof is valid if e is taken to be zero, but the value of k 3 so obtained
is then smaller. There is no difficulty in giving the analogous results to Theorem
7.11 for the pure pressure boundary-value problem and for incompressible
materials.

Solutions to the equilibrium equations

We now turn to the question of whether the minimizers whose existence we
have established satisfy the corresponding Euler-Lagrange equations. There is
at present no available regularity theory for our problems under acceptable
hypotheses, and we therefore confine our discussion to whether the minimizers
are weak solutions. Unfortunately there are technical problems associated with
the two most important cases, namely (i) when the material is compressible and
W(x) is given by
W(x) ={aeE: ~ > 0},
and (ii) when the material is incompressible.
We therefore consider the simpler situation when W(x)=E for all x, so that
there is no local constraint.
We replace (H7) by the following stronger hypotheses on G:
(//9) n = 2 : G is continuously differentiable in u, F, 6 for all xef2, and there
exist real constants K I > 0 , K2>0, BI>0, Ci>0, pi>0, l < v < 7 , z>0, s > l , and
a function b e/2 (O) such that
G(x, u, F,f)>b(x)+K 1 IFI~+K2 lul~+nl I~1v, (7.32)

Ou --Pl + C1 [[ul~+ lEt +161v], (7.33)

Oa~-~
G <P2 + C2Clul ' + IFI ~+ 1,57-1, (7.34)
O/"
 Existence Theorems in Non-Linear Elasticity 385

t3G < ~--

=P3 + C3 [fu(+ IFI ~-1 + 161 ~' ], (7.35)
27
for all values of the arguments. If 7 < 2, we assume further that T=
2-7
n = 3 : G is continuously differentiable in u,F,H, 6 for all xef2, and there
exist real constants K I > 0 , K2>0, BI>0, Cz>0, pz>0, 1 < v < / ~ < 7 , ~>0, s > l ,
and a function beLl(O) such that

G(x,u,F,H, 6)>_b(x)+KI(IFIr+IHIU)+K2 lull+B1 I~1v, (7.36)

~-uG < p l + Cl[lUl~+lFIr+ Ial~+l~lV], (7.37)

~G <
~ff- =P2 + C2[lu( + r F r + I a l u + 16[~], (7.38)

~G < ~ u
~--~ =P3 q- C3[]ul 7 + Ill r-1 + Inl ~' + I~1 <], (7.39)

_-<p, + C4[lul e' + Ill ~' + [HI"-~ + I~1~' ], (7.40)

for all values of the arguments. If ? < 3 we assume further that z = 37

3-?
Let f be given by

n = 2 : f ( x , u , F ) = G ( x , u , F , detF) ~ (7.41)
n = 3" f(x, u, F) = G(x, u, F, adj F, det F)]"
Two typical results are the following:
Theorem 6.12. In the hypotheses of Theorems 7.7, 7.9 replace (HT) by (H9). Suppose
also that (cf (7.23))
/f n = 2 : 7>2,
1 1 (7.42)
if n=3: y > 2 , --+--< 1.

Then the minimizing functions u = Uo satisfy the Euler-Lagrange equation

,743,
, 0/?z

for all veC~(Jl ") with v]om =0.

Proof. We give the proof for the case in which n = 3. Fix ve C ~ (~3) with vl0nl =O.
Since 1 < v < / ~ < 7 it follows that uo+eve~/for any e. Also we clearly have

d
d~ ~ (u~ ~ v'-iRdS" (7.44)
0122 01"12
386 J.M. BALL

It thus suffices to show that

l ( u ) ~ f ~f(x, u(x), Vu(x)) dx (7.45)

I2

is G~tteaux differentiable at u o, and that I'(uo)(V) is given by the left-hand side

of (7.43). By the dominated convergence theorem it is enough to establish the
estimate
~-~ f(x, u(x) + e v(x), Vu(x) + e Vv(x)) < O(x), (7.46)

for fixed u ~ d with I(u)< ~ , where O~/2(12) and is independent of e~(0, 1).
Carrying out the indicated differentiation in (7.46), we find from (7.41) that

ff--~-f(x, u+~v, [Tu+eVv)

r, o+ 1:Tu'] (7.47,
_ LI72. i IVul+ IVul+ ladj
for fixed u ~ r with I(u)<oo, and for all e~(0, 1). Here and below c denotes a
generic constant. The estimate (7.46) now follows from (7.37)-(7.40) and Hi$1der's
inequality. For brevity we display the calculation only for the last term of (7.47).
We have that

c ladj Pul<cp4 ladj lZul+eC4[lul r" +lVul"'

+ ladj Vul"-1 + Idet Vul"'] ladj Vul

<c[ladj Vul + lul ~' + l V u r + l a d j gul"+ldet Vul ~]

=<el1 + lul~ + l Vul~ + ladj Vul" + Idet Vulv]

for some OleLl(O), where we have used (7.36) and the facts that if 7>3 then
u~L~(f2), while i f 2 < 7 < 3 then u~E(O). []
Theorem 7.13 (cf Section 1, A2). In the hypotheses of Theorem 7.10 replace (HT)
by (Hg) , let (7.42) hold, and let f = ~IU(x,F) be independent of u so that
Jo(U)= ~ ~ ( x , 17u(x))d x - ~PR(X) b o 9u(x) d x - ~ u" tR dS. (7.48)
12 I2 ~I2

Suppose in addition that

tR dS + ~PR bo dx = 0. (7.49)

Then the minimizing function u = u o satisfies the Euler-Lagrange equations

~"t,ts" i 1

for all v~ C~(~").

 Existence Theorems in Non-Linear Elasticity 387

Proof. Proceeding as in Theorem 7.12 we obtain (7.50) for all v~ C ~ ( ~ ") satisfying
~vdx=O. (7.51)
f2

Let we C ~ ( ~ ") be arbitrary and set

1
v(x) = w ( x ) - m(f2~ ~ w(x) dx. (7.52)
Then

j[ pRbo'W--~li Wi
~u , ~t '
.] dx+ f W'tRdS
or2
1
- m(O) ~ w(y) dy. [~ PR bo dx + oa~ -jR dS] = O. []

Remarks. A similar result can be proved for the minimizer in Theorem 7.11.
We may also waive the assumption (7.42) at the expense of obtaining the Euler-
Lagrange equations only in terms of the distributions Adj Vu, Det Vu and with
derivatives of G replacing the derivatives of f. In Case (i) above an analogous
local result to Theorems 7.12, 7.13 may be proved under the a priori assumption
that det Vu > d > 0 locally. The details of these proofs are left to the reader.

Existence under other hypotheses

We next give a sample theorem under the hypothesis P~,u,~ for the displace-
ment boundary-value problem for a homogeneous material. This result almost
certainly has generalizations to integrands with x, u dependence.
Theorem 7.14. Let n=3. Let U _ ~ M 3• be such that Co T(U) is open. Let
Ul~Wl'~((2), adj Vul~L~((I), det VUl~L"(Q), where 7>2, # > 1 , l-I + 1 < 1 , v > l .
7 #
In the notation of Definition 4.3 let g: U---,~ satisfy P~,u,v at ul. Let the corre-
sponding function G satisfy
G(F,H, 6)>=b+Kl(IFl~+lHlU+13l ~) forall (F,H,f)6CoT(U), (7.53)
where b and K~ > 0 are constants, and
G(F, H, 6) ---, oo as (F, H, 6) ~ ~?(Co T(U)) (7.54)
Let
d = {u ~ Wl' ~(f2): u - u 1~ W0X'~(g?), adj Vu ~ LU(f2),
det Vuel2(O), 2;(u)eCo T(U) almost everywhere in 0}.
Define for u e ~ '
1(u)= ~g(Vu) dx. (7.55)
g?

Then if I ( u j < oo, there exists Uo~d that minimizes I(u) in d .

Proof. Let {u,} be a minimizing sequence. By (7.53), {2(u,)} is bounded in the
reflexive Banach space 9~=/2(f2)9 x LU(O)9 x/2(0). By the Poincar6 inequality
388 J.M. BALL

and our now standard arguments there exists a subsequence {uj} such that
u j ~ u o in Wl'~(f2), .F,(uj)~,F,(Uo) in ~r ~ ,
with Uoed. Define Gl: E ~ ~ by
GI(F, H, `5)= G(F, H, ,5) if (F, H, `5)eCo T(U)
= ~ otherwise.
Let
J(a) = ~ Gl(a(x)) dx. (7.56)

Then J: cg~ Jg, l ~ ~" Since g satisfies P~.u. v at u I it follows that J is convex.
Applying Fatou's lemma to the integrand
GI (F, H, 6) - b -- KI ([F]~ + [HI u + 1`51v)
we see that J is (strongly) lower semicontinuous. Hence (cf EKELAND & T/~MAM
[1, p. 33]) J is sequentially weakly lower semicontinuous. Thus
J(Z(Uo)) < lim J(Z(u,)),
j~oo
and the proof is complete. []

The most general integrands for which our methods establish the existence
of minimizers are given by the sum of polyconvex functions satisfying a suitable
subset of hypotheses (H1)-(H9), quasiconvex functions satisfying the hypotheses
of Theorem 6.1, and, where appropriate, functions satisfying condition P~,u. v as
in the above theorem. By suitably combining the growth conditions of each of
the terms in this sum various existence theorems may be given. At present both
the scarcity of examples of quasiconvex functions that are not polyconvex and
the abundance of physically useful polyconvex functions make these theorems of
little interest. We therefore leave the routine formulation of the results to the
reader.

8. Applications to Specific Models of Elastic Materials

Many forms of the stored-energy function have been proposed for nonlinear
elastic materials, particularly for various rubbers. An excellent review of the
literature can be found in the papers of OGDEN [2, 3]. We now examine the extent
to which these models satisfy the hypotheses of our existence theorems. By con-
centrating on a certain class of models below we do not mean to imply that other
models are inferior for empirical reasons. Neither should the omission of a model
from the discussion be construed as suggesting that it fails to satisfy our existence
hypotheses. Our purpose is simply to indicate the flexibility of the hypotheses to
serve for a variety of stored-energy functions and also to discuss the position
occupied with respect to these hypotheses by certain well known models.
We assume that n = 3 unless otherwise stated, and consider for simplicity
only isotropic materials, for which the stored-energy function has the form
(see (1.15))
~ ( x , F) = ~(x, v1, ~)2, V3)" (8.1)
 Existence Theorems in Non-Linear Elasticity 389

For compressible materials, in the case when the local constraint set W(x) has
the form
W(X)= E 1 • K ( x ) , (8.2)
with K ( x ) ~ nonempty, open and convex, it follows from Theorem 4.3 that
Co U(x)= W(x), where U(x)= {FEM 3• 3: det F~K(x)}. Necessary and sufficient
conditions for ~q/'(x, .) to be polyconvex on U(x) are given by Theorem 4.4. For
simplicity we shall assume in the compressible case that K(x) = {6 > 0} (continuity
condition), while in the incompressible case W(x)= E~ (no local constraint).
We consider a modification of a class of stored-energy functions introduced
by OGDEN [2, 31. For a > i let
~(~)=/)~-'~/)~-J-/)~-- 3, Z ( a ) = ( v 2 v3)a-~(v3 Vl)~+(1)l v2)a-- 3. (8.3)

Consider first incompressible materials, and let

M N
fg'(x, F) = B(x) + ~ a,(x) t~(ai)+ ~ ci(x) Z([3,), (8.4)
i=1 i=l

where a t > ... >aM > 1, fll > "'" >flu > 1, and where B, a~, ci are functions in/2(I2)
satisfying
ai(x)>k>O , ci(x)>k>O, for almost all x~f2 (8.5)
for some constant k.
By Theorem 5.2 #~(x, F) is polyconvex on U(x). Since v~ + v~ + v~ is a contin-
uous function of F ~ it follows that
v~ + v~ + v~_>_d(a)lel9 (8.6)
for some constant d(a) > O. Similarly

(/)2 /)3)a-~- (/)3 /)l)a-[- (/)1 /)2)~ e(a) ladj FI ", e(a)> 0. (8.7)

It follows from (8.4)-(8.7) that qCr considered as a function of x, F, adj F satisfies

hypotheses (H0-(H4), and (H8) with
= a t, ~t=fl t, K 2=0. (8.8)
Thus if
cq>_32, 1 1 4 (8.9)

we obtain from Theorems 7.8 and the analogues of Theorems 7.9-7.11 for in-
compressible materials the existence of minimizers for the various boundary-
value problems in terms of the distributions Vu, Adj gu. These minimizers satisfy
the incompressibility condition Det g u = l almost everywhere. Note that in
order to obtain existence for variable pressures p in the mixed displacement
pressure problem we require (Theorem 7.11) either
1 1
1 (8.10)

~* Becanse, for example, it is a finite-valued convex function.

390 J, M. BALL

or
1 1
~, + ~ - = 1, ]]VP]]L~(~,<k3(ai, cl, cti, fli, f2, 0t21). (8.11)
i. t 1

If 1 1
--+--<1, ~1>2, (8.12)
-- 0~I fll
then we have the stronger forms of the existence theorems with the incompressibil-
ity condition det Vu = 1 holding almost everywhere.
As a special case we consider the inhomogeneous Mooney-Rivlin material,
for which B - O , M = N = 1, ct1 =ill =2, so that
r F) = a 1(x) (I ~ - 3) + c 1(x) ( I I , - 3). (8.13)
Clearly (8.12) is satisfied so that the Mooney-Rivlin material is included in the
existence theory. Note that in the mixed displacement pressure problem the
critical case (8.11) applies.
The incompressible Neo-Hookean material
~W(x, F ) = cq (x)(I B - 3) (8.14)
is not covered by the theorems. To illustrate this point let us consider a single
term stored-energy function
~r(x, F)= ax(X) ~,(~) (8.15)
with a~ satisfying (8.5). To prove existence by our methods under the incom-
pressibility constraint Det 17u = 1 (det Vu = 1) it is necessary that if u, ~ u in
WI':(f2), then Det Vu r ~ Det Vu in ~'(f2) (det Vu r -* det Vu in @'(f2)). We must
thus have (cf the methods of Section 6) c(> 9 (~>3). Note, however, that we do
get existence theorems for the Neo-Hookean material in two dimensions.
OGDEN fitted a stored-energy function with three terms (M=2, N = 1) of the
form (8.4) to data of TRELOARfor homogeneous vulcanized rubber*. The values
of the various constants obtained were
~1 = 5.0, ~2 = 1.3, fit = 2,
a t =2.4 • 10 -3, a 2 -=4.8, c t ----0.05 kg cm -2, (8.16)
B=0.
Clearly (8.12) is satisfied. Furthermore, since a~>3 the minimizers u o of the
various boundary-value problems belong to C(I2); indeed by the results of
MORREY [2] (see also FRIEDMAN [1] and NE~AS [1]) we have in this case Uo6
C~ the space of H61der continuous functions on ~ with exponent 2. Note
that the imbedding theorems do not imply that Uo~ C(~) for the Mooney-Rivlin
material.
For compressible materials OGDEN [ 3 ] considered the effect of adding a
term F(det F) to (8.4)~. Suppose that
F(t)> C(t) for all t > 0 , where C is an N-function,'(
r (8.17)
F is convex on (0, oo), F ( t ) - . o o as t - . 0 + . )
* F o r r e l a t e d e x p e r i m e n t a l w o r k see TRELOAR [ 1 ] .
* OGDEN r e t a i n e d the values of the c o n s t a n t s B, cq, ~2, a t , a2, c1, b u t for i n c o m p r e s s i b l e m a t e r i a l s
r e p l a c e d the t e r m c 1 Z(2) b y c 1 [v I - 2 + v2- 2 + v 3 - 2 _ 3], T h e s e t e r m s are identical if v1 v 2 v3 = 1, a n d
since v~ v 2 v 3 is in p r a c t i c e very close to 1 the a l t e r a t i o n is i n s i g n i f i c a n t for e x p e r i m e n t a l c o r r e l a t i o n s .
 Existence Theorems in Non-Linear Elasticity 391

Then the modified stored-energy function satisfies (H1)-(H4) and (HT) with ~ = al,
/~--ill, K2 = 0, and thus satisfies the hypotheses of our existence theorems under
the conditions (8.9)-(8.12).
It is clear that a wide variety of stored-energy functions having the form (5.11)
(with x-dependence if necessary) can be treated by our theory in an way analogous
to that for the models discussed above. We end this section by exhibiting such
a stored-energy function, which satisfies the hypotheses of Theorem 7.3 but not
those of Theorem 7.6, etc. (for ~O z =~b), and thus requires the Orlicz-Sobolev
space apparatus. Our example is of a stored-energy function with slow growth. For
functions of very fast growth the Orlicz-Sobolev space setting would also be
necessary for any proof that the Euler-Lagrange equations are satisfied.* We
need two lemmas:
L e m m a 8.1. Let C, D be N-functions. Then

C - 1 (S)
---~0 as s---~ ~ (8.18)
D-I (s)
if and only if D<<C.
Proof. We just prove the 'only if' part, the 'if' part being easier. Set s = C(2t)
for 2 = 0 . Then by (8.18) 2t/D-l(C(2t))-*O as t ~ ~ . Hence t < D 1(C(2t)) for t
large enough. By the convexity of D we have for t large enough

D(t) < t 1 C-I(C(2t))

C ( 2 t ) = D - I ( C ( 2 t ) ) - 2 D-X(c(2t)) ~0 as t---~oo. []

L e m m a 8.2. Let g, k be non-negative continuous functions on ~+ such that

k(s)-~O as s--~oo and such that
co

g(t) d t= ~ .
0
Let
s
k(t) g(t) dt
0(s)= o
8

g(t) dt
0
Then O(s) ~ 0 as s ~ c~.
Proof. This follows immediately from L'HOpital's rule. []
Now let A be an N-function with principal part

A (t) = t ~ log t. (8.19)

Let
~g/'(F)=aA(Vl +V2+V3)+cA(v2v3+VsVl +VlV2)+F(VlV2V3) (8.20)
* In order to give a comprehensivedescription of anisotropy and inhomogeneity in rod theories
ANTMAN[8] uses more general Orlicz-Sobolev spaces than are used in this work. The methods de-
scribed here probably extend to these spaces, with a consequent broadening in the applicability of
the existence theorems presented here.
392 J.M. BALL

where F is as in (8.17) and a, c are positive constants. Since A ~ t ~+~ for any e > 0
Theorem 7.6 does not apply. To show that (Hs) is satisfied, so that Theorem 7.3
may be applied, we must prove that ,4<<A*. Let B(t)=t ~. Then B~,A so that
A-~B ~ t a n B*. It is therefore sufficient to prove that B*<<A*. We have that
A-~ (t) k (t)
gA(t) = - -t r - - - , t ~ (8.21)

where
k(t)~f 1 ~0 as t---, oo.
(logA- 1(t))~
t 1 o0
Also gB( ) = ~ - , and ~ gA(t)dt= ~ gB(t)dt=~. Therefore
0 0

k(t) g~(t) dt
A*- 1(s) o
$ (8.22)
B*-I(s) ~gB(t) dt
0

By Lemmas 8.1 and 8.2 we deduce B*<<A* as required.

9. An E x a m p l e of Nonuniqueness; Buckling of a Rod

In this section we establish nonuniqueness for the mixed displacement zero

traction boundary-value problem corresponding to buckling of a rod of a homo-
geneous, incompressible, Mooney-Rivlin material having a uniform cross-section.
We do this by exhibiting an admissible displacement field with total energy lower
than that of the trivial solution, and then applying Theorem 7.8 to ensure the
existence of a nontrivial minimizer for the total energy. Under suitable conditions
a similar but more complex analysis can be carried out for incompressible rods
consisting of material not of Mooney-Rivlin type. The extension to compressible
materials, however, is not so easy because an explicit trivial solution is not available.
In the stress-free reference configuration the rod occupies the region
f 2 = D x (0,/), />0, (9.1)
where the cross-section D is a nonempty bounded open set in N2 satisfying a
strong Lipschitz condition. We suppose that the density in the reference con-
figuration is a constant PR > 0, and that the plane x2 = 0 contains the line of centroids
of the rod in the reference configuration, so that

S x2 dS =0, dS = dx 1 dx 2. (9.2)
D

Let O [ 2 1 = b x {0,/}, af22=3Dx(O,l). Let A = SdS be the area of D. For 2 > 0

D
we consider the equilibrium mixed boundary-value problem with boundary
conditions
u= ~ on ~f21, (9.3)
tR=O on ~f22,
 Existence Theorems in Non-Linear Elasticity 393

where ~: ~,'~___,~3 is given by

~(X) = ( 2 - 89 2 - 89 X3). (9.4)

The stored-energy function has the form (cf. 8.11))
~ ( F ) = a ( I ~ - 3) + c(II n - 3), (9.5)
where a > 0 , c > 0 are constants. In the notation of Theorem 7.8 we set
y = # = 2, K 2 = 0, W(x) = E 1 = M 3 x 3 x M 3 • 3, (9.6)
d={ue W 1' 2(O): adj 17ueL2(f2), u = ~ almost everywhere in c301,
det 17u = 1 almost everywhere in g2},
,l (u) = ~ :tK (17u(x)) dx. (9.7)
f2
From (9.4) we obtain
17~i= diag (2- 89 2- ~, 2), det 17~i= 1, (9.8)
and
J(fi)= [a(2 2- 2 + 2 2 - 3)+ b(22 + 2- 2 - 3)] A l < oe. (9.9)
It is easily shown that ~ satisfies the equilibrium equations and boundary con-
ditions for a suitable hydrostatic pressure. By Theorem 7.8 (with u 1 =~) there
exists u 0 e d that minimizes J(u) in sl.
It thus remains to construct a function u s d with *
J(u) < J(fi). (9.10)
We perform this construction in a manner reminiscent of derivations of rod
theories in engineering. Let Yo, Oo be real valued functions with yoEC2([O, l]),
O0~ C2([0, l]) and

Yo(O)=Yo(l)=O, Oo(O)=Oo(1)=Oo(1)=O'o(1)=O. (9.11)

Let e>0, y = e y o , 0 = e 0 o, and define
Ue(X)=(X1 g ( X 2 , X3) , y(X3)+)C-89 2 COS0(X3) , 2X 3-2- 89 s i n 0 ( x 3 ) ), (9.12)
where g is a function to be chosen, u~ represents a deformation in which points
(0, 0, x3) are mapped to (0, Y(X3) ,/~X3) , and in which cross-sections normal to the
x3-axis in the reference configuration stay plane and are so inclined that their
normals remain parallel to the x2x 3 plane and make an angle 0(x3) with the
x3-axis (Fig. 3).
From (9.12) we obtain
Xlg,2 xlg,3 \
2-~cos0 y ' - 2 - ~ x 2 sin 0 90'~. (9.13)
- 2 -~ sin 0 2 - 2 - ~ x2 c o s 0 . 0 ' /
* Note that we do not have to satisfy the zero traction condition on c~f22 because our existence
theorem incorporates this as a natural boundary condition.
394 J.M. BALL

-x.2 -x.2
!

I
I
I

y (=a)
P 7"3 [ ~-3

(i) Reference configuration (ii) D e f o r m e d configuration

Fig. 3

Therefore
det V u, = g [2 89cos 0 + 2 - § y' sin 0 - 2-1 x 2 0']. (9.14)
Since 2 > 0 the expression in brackets in (9.14) is positive for all xEf2 if e is small
enough. Thus we m a y choose

g=[2~ cosO+ 2-~y' sinO-2-1x20'] -1, (9.15)

whence
det 17u~= 1 for all x~f2. (9.16)

It follows from (9.11)-(9.16) that u , ~ r for e small enough. Let

B = Vu~ Vu~r. (9.17)

A routine but tedious calculation shows that

B1 = )~-l + 2e2- lx20,o + e212-10g_22- 20oYo +(3x2 + xt))~2 -4, Oot2 ..~ ~ - 4 X12 X2002,,2"]
+ o (e2),
2 -1 2 t2
B2 = ~ + ~ [Yo - 2-1 00~] + o (d),
B33=22- 2e212 x2 0o + e 2 [2-102+2-1xZO'oZ]+o(e2), (9.18)
B1= B 21=~2 - 3 t
xlOo+o(~),
B~=B~=~,~ lxlx20;+o(~),
B z = B 3 = e [ 2 y o - 2 - t 0o ] + o(e),

and hence that

i . = 2~ -1 + ,~2+ 2~x2O,o(~-~ - ~ ) + ~:[~- ~O~o+ ((3x~ + x~),~-. + z- ~~)o,g

+ x12x222-40o,,Z 2)~-2y'oOo + Y'o2]+o(e2),
(9.19)
1i~=,~- ~ + 2,~ + 2~O'o(~- ~-,~- ~)+ ~[,~- ~o~ +((~ + ~),~- ~ + 3~,~- ~)og
+ xZxx~ 2- 5 0'~2 -- 22- 3 0oY'o + )~- l y'oZ] + o(e2),
where o(e) denotes a function of x such that o(e)/e--~0 uniformly in f2 as e--~ 0.
F r o m (9.2), (9.5), (9.7) we obtain (with the standard meaning for o(e2))
l
J(u,) = 3(~) + e2 ~ [d ~ 02 + dx 0,o2 + dE 0'~~ -- d 30oY,o + d4y,o~] dx 3 + o(e2), (9.20)
0
 Existence Theoremsin Non-LinearElasticity 395

where
do= A2- Z(a2 +c),
d 1 =a[(3k2+kl)2-4+k22-1]+c[(kl +k2)2-2 + 3k22- 53,
d 2 = k3,~- 5(a2 + c), (9.21)
d 3=2A2-3(a2+c),
d4=A2-1(a2+c),
and
kl=~ X2dS, k2=Ix22 dS, k3 =jfx2xZdS.1 a (9.22)
D D D

The Euler-Lagrange equations corresponding to the quadratic part of (9.20) are

2d 4 y o - d3 0o = ?, (9.23)
d3 ,
d2 0o" - dl 0o + do 0o - ~ - Yo = 0, (9.24)

where y is a constant. Setting 7 = 0 and combining (9.23), (9.24) we obtain

0~" - ~10g + ~2 0o = 0, (9.25)

where
dI d o -d~/4d 4 A(). 3 - 1)
~1 = - - ~ 0~2- (9.26)
d2 d2 k3
l
With the above as motivation we seek a solution, antisymmetric about x 3 = ~ - ,
to the equation
tt r O~ 13tt
0o - t C'o+ c~20o = 0 (9.27)
- -

subject to boundary conditions

0o=0 o = 0 at x 3 =0, l, (9.28)

for s o m e ~2 > ~2" If 0o is such a solution and

X3
Yo(X3) ~r ~ ! Oo(s)ds (9.29)

then Yo satisfies (9.11) and we have from (9.20), (9.26)-(9.29) that

l
J(I/e) ~---3(/i) --I-22 d 2 (~2 - ~2) ~ 02 d x 3 + 0 (/32), (9.30)
0
It follows that for e small enough J(u,)< J(~) as required.
To solve (9.27), (9.28) we first note that for all Re[0, 1] we have
3ck 2
cq > ka(a+c ) > 0 , CtEN0. (9.31)
Set
~ 2 = ~ 2 -'F'~, 0<T< 1 ( 3ck2 )2. (9.32)
k3(a+c)
396 J.M. BALL
I

IL

Fig. 4

The indicial equation for (9.27) is

m4 - ~ m2 + ~2 -----0 (9.33)
with roots m = +~, +_ip,where
x~_ 31 + ~ 4 3 2 ~2= ~ - 4 3 2 - ~ 1
2 ' 2 ' (9.34)

and x>/~>0. (x,/~ are real by (9.31), (9.32).) Hence

Oo(x3)=Al sinh ~c(x3 - 1 ) + A2 sin ll (X3- 1 ) (9.35)

will satisfy (9.27), (9.28) provided that

tan = tanh~-. (9.36)

l
Also 00 given1 by (9.35)1is antisymmetric about x 3 =~-. The graphs of the func-
tions ~=~---tant/, ~=~--tanht/ are sketched in Fig. 4. It is easy to prove the
indicated monotonicity properties.
Let ~co, ~:l be the values of-(-ex+ ~ 2 4~2)~- at 2 = 0, 1, respectively, so that

[3k 2 1/9k~ A\ ~ =[4kz+k,~ ~

/s -|~S~--
] \ "~-Ll/
V ~S-2--I--~
K
~ 3 K'3 ! ' /s ~] , ~ , 9 (9.37)
 Existence Theorems in Non-Linear Elasticity 397

It is not hard to show that x and/~ are continuous functions of 2 in [-0, 1], and
that x is not constant in [0, 1]. Therefore there exists an interval in the range of
K: [0, 1]---,~+ with length 6 > 0 . If K0#:x ~ then we may take

6 = I~:o- xll + r(z), (9.38)

where r(z) --* 0 as z ---, 0.
From Figure 4 it is thus clear that if

~>2n, (9.39)

then there exists 0 < 2 < 1 such that (9.36) is satisfied. The corresponding 0o is
the function required. To satisfy (9.39) one need only choose I large enough.
Thus sufficiently long rods of arbitrary cross-section will exhibit nonuniqueness for
some 2e(0, 1). (An obvious refinement of this argument shows that if 0 < 2o < 1
then for 1 large enough there will be nonuniqueness for some 2 with 2 o <2 < 1.)
Usually ~co 4=x I so that by choosing z >0 small enough we get nonuniqueness
for some 2e(0, 1) whenever
4n
I~o - K l l > T . (9.40)

This is a condition expressed entirely in terms of I and the cross-sectional parame-

ters kl, k2, k 3 and A.
Example. Let D be the disc x 2 + x 2 < a 2. Then
~za 4 7za 6
kl = k 2 = 4 ' k 3 - 24 ' A = n a2 (9.41)

so that

1r lye. (9.42)

Condition (9.40) therefore becomes

a
T<v, v-0.09. (9.43)

The condition (9.40) is somewhat crude; indeed it is possible that no such con-
dition is necessary. Improved estimates, and lower bounds on the supremum of
0 <2 < 1 for which nonuniqueness occurs, may be obtained by more detailed
calculations based on Figure 4. I have not included these results since they are
messy and since my method is severely limited in scope due to the type of trial
deformation considered in (9.12). Even in situations where we envisage non-
uniqueness occurring by Euler buckling, the deformation of cross-sections implied
by (9.12) is unrealistic.
There are numerous formal stability calculations in the literature for rods in
tension or compression, and for other problems in three dimensional nonlinear
elasticity. For the most part these calculations are based on the theory of small
deformations superposed upon large; the status of this theory with respect to
398 J.M. BALL

nonlinear elasticity has yet to be established. The reader is referred for details
and references to FOSDICK & SHIELD [1], HOLDEN [1], KNOPS & WILKES [1],
SENSENIG [1], WESOLOWSIO[1], WILKES [1]. Nonuniqueness for the pure traction
boundary-value problem of a rectangular block of Neo-Hookean material loaded
uniformly on each face has been established explicitly by RIVLIN [1, 2, 3]. For
various one and two-dimensional rod and shell theories rigorous proofs of non-
uniqueness have been given by ANTMAN [2, 3, 6].

10. Concluding Remarks

The main implication of this work for constitutive inequalities is that the
quasiconvexity condition (and in particular the Legendre-Hadamard condition)
is consistent with realistic models of hyperelastic solids. In the one-dimensional
case, when convexity and quasiconvexity of ~/r ") are the same, Theorem 3.2
shows that the existence of C 1(t-i) minimizers for various homogeneous displace-
ment boundary-value problems implies that ~ is quasiconvex, while the same
result holds in three dimensions if f2 is a cube. If "/U is not quasiconvex then
minimizers may exist that are not C 1. Some examples in one dimension are
discussed by ERICKSEN [3]. It should be noted that we have not proved that
C1(f2) minimizers exist in general for displacement boundary-value problems
when Of2 is suitably regular under any reasonable hypotheses on ~ .
The existence theorems proved in this article take the form that existence is
established for a given material for all suitable boundary data. In general such
unqualified existence is not to be expected for real materials, since rupture will
occur under extreme conditions of deformation. We may also not be interested
in solutions having at some points deformation gradients that lie outside the
range in which the material behaves elastically. One way of partially circum-
venting these difficulties is to choose the local constraint set W(x) so as to pro-
hibit such behaviour, and then to check a posteriori whether the minimizer u o
is such that Vuo(X)S8 W(x) for any x. One would then like a priori conditions
on the size of the boundary data to prevent this happening. The derivation of
any such conditions would require delicate estimates. The reader is referred to
the papers by ERICKSEN[4] and KNOWLES• STERNBERG[1] for further discussion
of some of these points.
In general weak lower semicontinuity will not hold if the quasiconvexity or
polyconvexity hypotheses are replaced by a hypothesis of convexity of the func-
tion ~U restricted to positive definite symmetric tensors U. It is nevertheless
instructive to see how an attempt to establish lower semicontinuity in this case
breaks down. The difficulty is that if u,-~u in the Sobolev space W 1' ~(g2), then
the weak limit in El(f2) of the sequence Ur=~/Vu f Vu,, will not necessarily be
1/Vu r Vu, and indeed may not arise from any displacement. This is because
~/Vu T Vu is not of the form (4.2) and hence not sequentially weakly continuous.
The difficulty is also connected with the nonlinearity of the Riemann-Christoffel
tensor based on C. Similar mathematical problems arise from attempts to estab-
lish existence under the COLEMAN & NOEL condition [1], or HILL'S inequalities
[2, 3]. These conditions do not imply the Legendre-Hadamard condition. The
COLEMAN & NOEL condition cannot apply to all hyperelastic materials because
 Existence Theorems in Non-Linear Elasticity 399

it is violated for nearly incompressible materials such as rubber (see HILL [2],
OGDEN [-1, 3], RIVLIN [2], SIDOROFF [1]).
For bodies that are not homeomorphic to an open ball the various minimizers
whose existence we have established may represent deformations topologically
isolated from those desired on physical grounds t. One might require, for example,
all admissible deformations to be accessible from a given deformation by a
homotopy of globally invertible configurations. In this article we have not studied
such global constraints (although they may be of a weakly closed type), but have
concentrated on local constraints such as the local invertibility condition
det Vu > 0. Local invertibility is a relatively weak requirement; indeed a hollow
sphere may be everted without violation Of the condition in any intermediate
deformation (SMALE [ 1 ] ) . #
Finally I remark on the implications of the results of Section 6 for theories
of elasticity incorporating pointwise constraints on the deformation gradient F.
These results suggest strongly that the only nontrivial homogeneous constraints
giving rise to a well posed theory have the form (see (4.2))
(a(F)=A + B~ F~+ C~(adj F)i,+ O det F=O, (10.1)
where A, B~, C~, D are constants. It is not hard to show that the only objective
constraints of this form (i.e., e~ satisfying q5(QF)= q5(F) for all orthogonal Q) are
those with BT= C7=0, so that detF is specified. In particular, as we have seen,
the incompressibility condition detF--1 gives rise to a well posed theory. Note,
however, that the constraint of inextensibility (TRUESDELL 8,:; NOLL [ l , p. 72]) is
not included. It seems possible, therefore, that solutions do not in general exist for
boundary-value problems of inextensible elasticity, and that a higher order
theory is required to make such constraints well behaved.

Acknowledgement
The research reported here was begun in the summer of 1974 when I held part of a Science
Research Council research fellowship at the Lefschetz Center for Dynamical Systems, Brown Uni-
versity. I am greatly indebted to CONSTANTINE DAFERMOS,without whose enthusiastic interest
and consistently excellent advice this article would not have been written. I would also like to thank
STUART ANTMAN for several useful suggestions, and for his careful criticism of the manuscript. Finally,
I am glad to acknowledge helpful discussions at various stages of the project with ZvI ARTSTEIN,
KEN BROWN, DAVID EDMUNDS, J.L. ERICKSEN, ROBIN KNOPS and A.C. PIPKIN.

References
A. R. AMIR-MoI~z [1] Extreme properties of eigenvalues of a Hermitian transformation
and singular values of sum and product of linear transformations,
Duke Math. J., 23 (1956), 463-476.
A. R. AM1R-Mo~z [1] Singular values of a matrix, Amer. Math. Monthly (1958), 742-748.
& A. HORN
S. S. ANTMAN [1] Equilibrium states of nonlinearly elastic rods, J. Math. Anal. Appl.
23 (1968), 459-470.
t For bodies that are homeomorphic to an open ball the same may occur for mixed problems; the
situation for displacement boundary-value problems is unclear.
* A visualization of the eversion due to SHAPIRO can be found in PHILLIPS [1].
400 J. M. BALL

I2] Existence of solutions of the equilibrium equations for nonlinearly

elastic rings and arches, Indiana Univ. Math. J. 20 (1970) 281-302.
1'3] Existence and nonuniqueness of axisymmetric equilibrium states of
nonlinearly elastic shells, Arch. Rational Mech. Anal. 40 (1971),
329-372.
[4] "The theory of rods", in Handbuch der Physik, Vol Via/2, ed.
C. TRUESDELL, Springer, Berlin, 1972.
[5] Monotonicity and invertibility conditions in one-dimensional non-
linear elasticity, in "Nonlinear Elasticity", ed. R.W. DICKEY, Aca-
demic Press, New York, 1973.
1'6] Nonuniqueness of equilibrium states for bars in tension, J. Math.
Anal. Appl. 44 (1973), 333-349.
[7] Ordinary differential equations of nonlinear elasticity I: Foundations
of the theories of nonlinearly elastic rods and shells, Arch. Rational
Mech. Anal. 61 (1976), 307-351.
1'8] Ordinary differential equations of nonlinear elasticity II: Existence
and regularity theory for conservative boundary value problems,
Arch. Rational Mech. Anal. 61 (1976), 353-393.
J. M. BALL 1'1] Weak continuity properties of mappings and semigroups, Proc. Roy.
Soc. Edin (A) 72 (1973/4), 275-280.
[2] On the calculus of variations and sequentially weakly continuous
maps, Proc. Dundee Conference on Ordinary and Partial Differential
Equations 1976, Springer Lecture Notes in Mathematics, to appear.
M. F. BEATTY [1] Stability of hyperelastic bodies subject to hydrostatic loading, Non-
linear Mech. 5 (1970), 367-383.
I. BEJU [1] Theorems on existence, uniqueness and stability of the solution of
the place boundary-value problem, in statics, for hyperelastic mate-
rials, Arch. Rational Mech. Anal., 42 (1971), 1-23.
[2] The place boundary-value problem in hyperelastostatics, I. Dif-
ferential properties of the operator of finite elastostatics, Bull. Math.
Soc. Sci. Math. R.S. Roumanie 16 (1972), 132-149, II. Existence,
uniqueness and stability of the solution, ibid. 283-313. r
H. BUSEMANN,G. EWALD 1'1] Convex bodies and convexity on Grassman cones, Parts I-IV, Math.
& G.C. SHEPHARD Ann., 151 (1963), 1-41.
H. BUSEMANN 1'1] Convexity on nonconvex sets, Proc. Coll. on Convexity, Copen-
G.C. SHEPHARD hagen, Univ. Math. Inst., Copenhagen, (1965), 20-33.
C. CATTANEO [1] Su un teorema fondamentale nella teoria delle onde di discontinuith,
Atti. Accad. Sci. Lincei Rend., C1. Sci. Fis. Mat. Nat. Ser 8, 1 (1946),
66-72.
L. CESARI 1'1] Closure theorems for orientor fields and weak convergence, Arch.
Rational Mech. Anal. 55 (1974), 332-356.
1'2] Lower semicontinuity and lower closure theorems without seminor-
reality conditions, Annali Mat. Pura. Appl. 98 (1974), 381-397.
1'3] A necessary and sufficient condition for lower semicontinuity, Bull.
Amer. Math. Soc. 80 (1974), 467-472.
A. CLEBSCH [1] Ober die zweite Variation vielfacher Integrale, J. Reine Angew.
Math., 56 (1859), 122-149.
B. D. COLEMAN I-1] On the thermostatics of continuous media, Arch. Rational Mech.
8Z W. NOLL Anal., 4 (1959), 97-128.
T.K. DONALDSON [1] Orlicz-Sobolev spaces and imbedding theorems, J. Funct. Anal.,
& N. S. TRUDINGER 8 (1971), 52-75.
P. DUHEM 1'1] Recherches sur l'61asticit6, troisi6me partie. La stabilit6 des milieux
61astiques, Ann. Ecole Norm., 22 (1905), 143-217. Reprinted Paris,
Gauthier-Villars 1906.
N. DUNFORD [[1] "Linear operators", Pt. 1., Interscience, New York 1958.
& J.T. SCHWARTZ
n . G . B . EDELEN [i] The null set of the Euler-Lagrange operator, Arch. Rational Mech.
Anal., 11 (1962), 117-121.
 Existence Theorems in Non-Linear Elasticity 401

[2] "Non local variations and local invariance of fields", Modern

analytic and computational methods in science and engineering
No. 19, Elsevier, New York, 1969.
I. EKELAND& R. Tt~MAM I-1] "Analyse convexe et probl~mes variationnels", Dunod, Gauthier-
Villars, Paris, 1974.
J.L. ERICKSEN I-1] Nilpotent energies in liquid crystal theory, Arch. Rational Mech.
Anal., 10 (1962), 189-196.
[23 Loading devices and stability of equilibrium, in "Nonlinear Elasti-
city", ed. R.W. DICKEY, Academic Press, New York 1973.
1-33 Equilibrium of bars, J. of Elasticity, 5 (1975), 191-201.
[43 Special topics in elastostatics, to appear.
G. FICHERA 1-13 Existence theorems in elasticity, in Handbuch der Physik, ed. C.
TRUESDELL, Vol. Via/2, Springer, Berlin, 1972.
R. L. FOSDICK 1-1] Small bending of a circular bar superposed on finite extension or
& R.T. SHIELD compression, Arch. Rational Mech. Anal., 12 (1963), 223-248.
A. FOUGI~RES 1-1] Thesis, Besanqon, 1972.
A. FRIEDMAN [13 "Partial differential equations", Holt Rinehart and Winston, New
York, 1969.
J-P. GossEz [13 Nonlinear elliptic boundary value problems for equations with
rapidly (or slowly) increasing coefficients, Trans. Amer. Math. Soc.,
190 (1974), 163-204.
L.M. GRAVES 1-13 The Weierstrass condition for multiple integral variation problems,
Duke Math. J., 5 (1939), 656-660.
A. E. GREEN (~ W. ZERNA 1-13 "Theoretical elasticity", 2nd edition, Oxford Univ. Press, 1968,
J. HADAMARD [1] Sur une question de calcul des variations, Bull. Soc. Math. France,
30 (1902), 253-256.
1-2] "Leqons sur la propagation des ondes", Paris, Hermann, 1903.
[33 Sur quelques questions de calcul des variations, Bull. Soc. Math de
France, 33 (1905), 73-80.
R. HILL [1] On uniqueness and stability in the theory of finite elastic strain, J.
Mech. Phys. Solids 5 (1957), 229-241.
I-2] On constitutive inequalities for simple materials, I, J. Mech. Phys.
Solids, 16 (1968), 229-242.
I-3] Constitutive inequalities for isotropic elastic solids under finite
strain. Proc. Roy. Soc. London A314 (1970), 457-472
J.T. HoLDEN 1-13 Estimation of critical loads in elastic stability theory, Arch. Rational
Mech. Anal., 17 (1964), 171-183.
R. J. KNOPS ~: [1] "Theory of elastic stability", in Handbuch der Physik, Vol. Via/3,
E. W. WILKES ed. C.TRUESDELL, Springer, Berlin, 1973.
J. K. KNOWLES [13 On the ellipticity of the equations of nonlinear elastostatics for a
E. STERNBERG special material, 3. of Elasticity, 5 (1975), 341-361.
M. A. KRASNOSEL'SKII 1-1] "Convex functions and Orlicz spaces", trans. L. F. BORON, Nordhoff,
& YA. B. RUTICKII Groningen, 1961.
M-T. LACROIX [13 Espaces de traces des espaces de Sobolev-Orlicz, J. de Math. Pures
Appl., 53 (1974), 439-458.
E.J. McSHANE 1-1] On the necessary condition of Weierstrass in the multiple integral
problem of the calculus of variations, Annals of Math. Series 2, 32
(1931), 578-590.
N. G. MEYERS I-1] Quasi-convexity and lower semicontinuity of multiple variational
integrals of any order, Trans. Amer. Math. Soc, 119 (1965), 125-149.
L. MIRSKY [1] On the trace of matrix products, Math. Nach. 20 (1959), 171-174.
1-21 A trace inequality of John von Neumann, Monat. ffir Math., 79
(1975), 303-306.
3. J. MOREAU 1-i] Fonctionnelles convexes, S6minaire sur les 6quations aux d6riv6es
partielles, Coll~ge de France, 1966-1967.
C. B. MORREY, Jr. 1-13 Quasi-convexity and the lower semicontinuity of multiple integrals,
Pacific J. Math. 2 (1952), 25-53.
402 J.M. BALL

[2] "Multiple Integrals in the Calculus of Variations", Springer, Berlin,

1966.
J. NECAS I-1] "Les m6thodes directes en th6orie des 6quations elliptiques", Masson,
Paris, 1967.
P. NIEDERER [1] A molecular study of the mechanical properties of arterial wall
vessels, Z.A.M.P., 25 (1974), 565-578.
J. T. ODEN [1] Approximations and numerical analysis of finite deformations of
elastic solids, in "Nonlinear Elasticity" ed. R.W.DICKEY, Academic
Press, New York, 1973.
R. W. OGDEN [1] Compressible isotropic elastic solids under finite strain-constitutive
inequalities, Quart, J. Mech. Appl. Math., 23 (1970), 457-468.
I-2] Large deformation isotropic elasticity- on the correlation of theory
and experiment for incompressible rubberlike solids, Proc. Roy. Soc.
London A326 (1972), 565-584.
[3] Large deformation isotropic elasticity: on the correlation of theory
and experiment for compressible rubberlike solids, Proc. Roy. Soc.
London A 328 (1972), 567-583.
A. PHILLIPS [-1] Turning a surface inside out, Scientific American, May 1966.
Y. G. RESHETNYAK [1] On the stability of conformal mappings in multidimensional spaces,
Sibirskii Math. 8 (1967), 91-114.
[2] Stability theorems for mappings with bounded excursion, Sibirskii
Math. 9 (1968), 667-684.
R. S. RIVLIN [1] Large elastic deformations of isotropic materials. II. Some uniqueness
theorems for pure homogeneous deformation, Phil. Trans. Roy. Soc.
London 240 (1948), 491-508.
[2] Some restrictions on constitutive equations, Proc. Int. Symp. on the
Foundations of Continuum Thermodynamics, Bussaco, 1973.
I3] Stability of pure homogeneous deformations of an elastic cube under
dead loading, Quart. Appl. Math. 32 (1974), 265-272.
R.T.RoCKAFELLAR [1] "Convex analysis", Princeton University Press, Princeton, New
Jersey, 1970.
H. RUND I-1] "The Hamilton-Jacobi theory in the calculus of variations", Van
Nostrand, London, 1966.
I-2] Integral formulae associated with the Euler-Lagrange operators of
multiple integral problems in the calculus of variations, Aequationes
Math., 11 (1974), 212-229.
L. SCHWARTZ [1] "Th6orie des distributions", Hermann, Paris, 1966.
C. B. SENSENIG [1] Instability of thick elastic solids, Comm. Pure Appl. Math., 17 (1964),
451-491.
M. J. SEWELL [1] On configuration-dependent loading, Arch. Rational Mech. Anal.,
23 (1967), 327-351.
F. SIDOROFF [1] Sur les restrictions ~timposer/t l'6nergie de d6formation d'un mat6riau
hyper61astique, C.R.Acad. Sc. Paris A, 279 (1974), 379-382.
E. SILVERMAN [1] Strong quasi-convexity, Pacific J. Math., 46 (1973), 549-55'4.
S. SMALE [1] A classification of immersions of the two-sphere, Trans. Amer. Math.
Soc., 90, (1959), 281-290.
F. STOPPELLI 1-1] Un teorema di esistenza e di unicifft relativo all6 equazioni dell'elasto-
statica isoterma per deformazioni finite, Ricerche Matematica, 3
(1954), 247-267.
R.TEMAM [1] On the theory and numerical analysis of the Navier-Stokes equations,
Lecture notes in Mathematics No. 9, University of Maryland.
F.J.TERPSTRA [1] Die Darstellung biquadratischer Formen als Summen von Quadraten
mit Anwendung auf die Variationsrechnung, Math. Ann. 116 (1938),
166-180.
C. M.THEOBALD 1-1] An inequality for the trace of the product of two symmetric matrices,
Math. Proc. Camb. Phil. Soc., 77 (1975), 265-268.
R. C. THOMPSON [1] Singular value inequalities for matrix sums and minors, Linear
Algebra and Appl., 11 (1975), 251-269.
 Existence Theorems in Non-Linear Elasticity 403

R. C. THOMPSON & [1] On the eigenvalues of sums of Hermitian matrices, Linear Algebra
L. J. FREEDE and Appl, 4 (1971), 369-376.
[2] On the eigenvalues of sums of Hermitian matrices II, Aequationes
Math., 5 (1970), 103-115.
[-3] Eigenvalues of sums of Hermitian matrices III, J.Research Nat. Bur.
Standards B, 75B (1971), 115-120.
L. R. G. TRELOAR [-1] "The physics of rubber elasticity", 3rd edition, Oxford Univ. Press,
Oxford, 1975.
C. TRUESDELL [-1] The main open problem in the finite theory of elasticity (1955),
reprinted in "Foundations of Elasticity Theory", Intl. Sci. Rev. Set.
New York: Gordon and Breach 1965.
C. TRUESDELL & [1] "The non-linear field theories of mechanics", in Handbuch der
W. NOEL Physik Vol. III/3, ed. S.FLOGGE, Springer, Berlin, 1965.
W. VAN BUREN [1] "On the existence and uniqueness of solutions to boundary value
problems in finite elasticity", Thesis, Department of Mathematics,
Carnegie-Mellon University, 1968. Research Report 68-ID7-
MEKMA-RI, Westinghouse Research Laboratories, Pittsburgh,
Pa. 1968.
L. VAN HOVE [1] Sur l'extension de la condition de Legendre du calcul des variations
aux int6grales multiples h plusieurs fonctions inconnues, Proc.
Koninkl. Ned. Akad. Wetenschap 50 (1947), 18-23.
[2] Sur le signe de la variation seconde des int6grales multiples h plusieurs
fonctions inconnues, Koninkl. Belg. Acad., Klasse der Wetenschappen,
Verhandelingen, 24 (1949).
J. VON NEUMANN [1] Some matrix-inequalities and metrization of matric-space, Tomsk
Univ. Rev. 1 (1937), 286-300. Reprinted in Collected Works Vol. IV
Pergamon, Oxford, 1962.
C.-C. WANG & [1] "Introduction to rational elasticity," Noordhoff, Groningen, 1973.
C. TRUESDELL
Z. WESOLOWSKI [1] "'Zagadnienia dynamiczne nieliniowej teorii sprezystosci", Polska
Akad. Nauk. IPPT, Warsaw, 1974.
E. W. WILKES [1] On the stability of a circular tube under end thrust, Quart. J. Mech.
Appl. Math. 8 (1955), 88-100.

Department of Mathematics
Heriot-Watt University
Edinburgh EH144AS
(Received August 15, 1976)