Arch. Rational Mech. Anal. 129 (1995) 11-45.

9 Springer-Verlag 1995

On the Rigidity of Certain Surfaces with Folds

and Applications to Shell Theory
G I U S E P P E G E Y M O N A T & E N R I Q U E SANCHEZ-PALENCIA

Communicated by H. BREZIS

Abstract

In the asymptotic theory of thin elastic shells the rigidity of the mid-surface with
kinematic boundary conditions plays an important role. Rigidity is understood in
the sense of infinitesimal (linearized) rigidity, i.e., the displacements vanish provided
the variation of the first fundamental form vanishes. In this case the surface is also
called "stiff", as it cannot undergo pure bendings. A stiff surface is imperfectly stiff
or perfectly stiff when the origin respectively does or does not belong to the
essential spectrum of the boundary-value problem. These questions are investi-
gated in the framework of Douglis-Nirenberg elliptic systems, with boundary
conditions and transmission conditions at the folds. The index properties ensures
quasi-stiffness, i.e. stiffness up to a finite number of degrees of freedom. The concept
of perfect stiffness is linked with estimates for the rigidity system at an appropriate
level of regularity for the data and the solution. It is proved that surfaces with folds
are never perfectly stiff. It is also shown that the transmission conditions at
the folds contain more conditions than those satisfying the Shapiro-Lopatinskii
property. This leads to certain rigidity properties of the folds. Some examples are
given.

Contents

0. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1. Preliminary Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2. Regularity Lemmas in the Subspace G of Pure Bendings and Equivalent
Definitions of Perfect and Imperfect Stiffness . . . . . . . . . . . . . . . . . . . . . . . 19
3. Quasi-Rigidity Properties of Certain Elliptic Surfaces with Folds . . . . . . . . . . . 20
4. New Smoothness Properties of Elements of G . . . . . . . . . . . . . . . . . . . . . . 25
5. Examples of Perfectly and Imperfectly Stiff Surfaces . . . . . . . . . . . . . . . . . . . 27
6. Shells with Folds are not Perfectly Stiff . . . . . . . . . . . . . . . . . . . . . . . . . . 29
7. The Subspace G in Cartesian Coordinates. Considerations on the
Formal Cauchy Problem in the Characteristic and Non-Characteristic Cases . . . . 32
12 G. GEYMONAT& E. SANCHEZ-PALENCIA

8. Further Study of Folds in the Subspace G. Characteristic and

Non-Characteristic Cases. Rigidification . . . . . . . . . . . . . . . . . . . . . . . . . . 36
9. New Examples of Perfectly and Imperfectly Stiff Surfaces . . . . . . . . . . . . . . . . 40
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

O. Introduction

In the framework of linear thin elastic shells [G62, N85], two very different
kinds of asymptotic behavior appear when the thickness of the shell tends to zero.
These two asymptotic processes depend on whether the mid-surface S (subjected to
the corresponding kinematic boundary conditions) admits pure bendings, i.e.,
displacements keeping invariant the intrinsic metric of S (also called inextensional
displacements). Since the local rigidity of a thin shell is very much larger with
respect to extensions than to bendings, the limit behavior is described by pure
bendings provided they are allowed by the surface. On the other hand, if such pure
bendings are not allowed, the limit behavior is mainly described by the membrane
approximation (along with boundary layers in certain cases). The surfaces which
admit pure bendings are called non-stiff(or non-rigid in the geometric terminology,
but we prefer non-stiff, as rigidity has a different meaning in mechanics). The
surfaces which do not admit pure bendings are called stiff(or rigid). Of course the
rigidity considered here in a linear context is infinitesimal rigidity. These different
asymptotic processes were studied in [SP89a, SP89b, SP92]. It is also possible to
obtain these limit behaviors by starting from three-dimensional elasticity instead of
classical thin elastic shell theory [SP90], but here we adhere to the first point of
view, which is mathematically more rigorous and leads to a better understanding of
the corresponding boundary conditions.
The fundamental monograph of VEKUA [-V62] contains many results on rigidity
of surfaces including surfaces with folds. Some of the results are analogous to ours;
however VEKUA'S proofs are based on generalized analytic functions and are
therefore independent from ours.
Our functional framework allows us to introduce the more subtle concept of
perfect stiffness, which is interesting both from the mechanical and the numerical
point of view. Indeed, the operator corresponding to the membrane approximation
does not enjoy compactness properties. Consequently an essential spectrum aess is
involved [GLT79, GG77, SPV92]. It turns out that this essential spectrum may
contain the origin, which then is not an eigenvalue. If this occurs, the surface is
called imperfectly stiff; on the other hand, if 2 = 0 (where 2 is the spectral para-
meter) is a point of the resolvent set, the surface is called perfectly stiff. The
motivation for this terminology, introduced in [SP89bJ and [SP92], is the analogy
with HADAMARD'Sdefinition of a well-posed problem. Indeed, when the surface is
not perfectly stiff, the static problem (for 2 = 0) is not solvable for certain systems of
given forces (which appear on the right-hand side of the operator equation) because
the range of the associated operator is not closed. More precisely, the surface
admits pseudo-bendings (in the terminology of [GLT79]), which are displacements
(or more exactly, sequences of displacements) with norms equal to 1 such that the
 Rigidity of Surfaces with Folds 13

corresponding variations of the first fundamental form of the surface tend to zero.
The existence of these Weyl sequences say that 2 = 0 belongs to the essential spectrum
of the operator [GG77, SP92]. The physical meaning is that the surface, which is of
course stiff (rigid), "staggers"; conversely, in the perfectly stiff case, it is "firm".
In the fundamental book [GLT791 only smooth surfaces with simple boundary
conditions were studied. For them the essential spectrum can be completely
characterized in terms of the failure of the classical Agmon-Douglis-Nirenberg
a priori estimates.
In this paper we consider properties of stiffness and perfect stiffness of certain
classes of surfaces, mainly in the case when the surfaces have folds (in their natural
configurations and therefore in their deformed configurations. We do not consider
folds generated by deformations). In this case, the transmission conditions across
the fold involve the Shapiro-Lopatinskii condition with non-classical values of the
indices. The Agmon-Douglis-Nirenberg a priori estimates are not suitable for our
problem, and we explicitly construct ad hoc Weyl sequences in order to prove the
imperfect stiffness.
When we apply the adjective smooth to functions, curves, and surfaces, they are
understood to have enough smoothness for our analytic methods to make sense.
For example, in some places we require the smoothness necessary to justify the use
of Korn's inequality, and in other places we require the more stringent smoothness
required for the a priori estimates of AGMON, DOU~LIS & N~RENBER~.
We point out that folds are responsible for important rigidity properties (or
restrictions on the subspace of pure bendings) as our everyday experience shows
(think of a folded paper). But the specific properties of rigidification depend on the
position of the fold with respect to the asymptotic curves of the adjacent surfaces.
These rigidity properties imply that, generally speaking, the limit behavior of
a smooth surface is very different from that of a polyhedral one; this fact has
important consequences in connection with the so-called membrane locking, which
appears in numerical computations, and which amounts to a discrepancy between
the finite-element discrete subspace and the subspace of pure bendings of the
smooth surface [ASP92]. Some of the results of the present paper were announced
in [ G S P g l ] and [SP89b].

Notation. Vectors in the sense of entities of the physical space with three compo-
nents are in boldface, for instance, u = (Ul, u2, u3), and correspondingly, a similar
notation is often used for vector (functional) spaces, for instance, u e L 2 - (L2) 3.
The usual convention of summation of repeated indices is used; Latin and
Greek indices are summed from 1 to 3 and from 1 to 2, respectively.
We use the classical notation II-IIo, II" II~ . . . . . for the norms in the Sobolev
spaces L 2, H ~. . . . , when the domain of definition is irrelevant.

1. Preliminary Results

In this paper we consider linear theory of thin shells in the framework of KO1TER
[K70] and more exactly the mathematical framework of this theory, including an
14 G. GEYMONAT& E. SANCHEZ-PALENCIA

existence and uniqueness theorem by BERNADOU, CIm~LET, and MIARA [BC76,

CM91]. We now give the notation and general features of this theory; a more
specific description of fold and boundary conditions will be given later.
Let IE3 be the Euclidean space referred to the orthonormal frame (0, el, e2, e3),
and let ~ be a bounded open set of IR2 with boundary F. The mid-surface of the
shell S is the image of f~ by the map
~k:(y~, y 2 ) e f i --~ ~k(y)elE a.

(Here we only consider one local map, since the general case of an atlas may be
handled in the same way.)
At each point of S we consider the two tangent vectors

a~-O,~--tVy~, ~ = 1,2,

and the unit normal vector

a 1 A a 2

a3 -- l al A.2~"
Here /x denotes the vector product in IEa.
Let u = u ( y 1, y2) be the displacement vector of S. The linear theory of shells is
described in terms of the deformation tensor of the mid-surface

and of the change-of-curvature tensor

In these expressions a~f and b~a denote the coefficients of the first and second
fundamental forms of S before deformation, and ~i~f and b'~f denote these coeffi-
cients after deformation, i.e.,

a~f = a~ 9 af, b~f = -- a 3 9 a3, f = "3 9 a~, f = a 3 9 .at,

where ,~ denotes differentiation with respect to y~, and either D~ or ]~ is used for
covariant differentiation. The expressions for 8~a and b'~f in the linearized small-
displacement theory are obtained from these expressions by replacing ~ by ~k + u
and only keeping the terms linear in u:
(1.1) ~f(u) = 89 + u~l~) - b~u~,
). 2 2 2
(1.2) p~f(u) = ual~fl + bBl=U z + blsU21 ~ + b.u21fl - b. bxfu3
where covariant and contravariant components are used (see, for instance, EN85,
K70, $69]). Specifically
,~ u
u~l fl = u~,fl -- F~fl ~,
(1.3) b}l~ = b},~ + r =2~ b f v - F. fv= b 2~,
2 u
U3 l o:fl = U3, o:fl - - 1V'~zfl .z
 Rigidity of Surfaces with Folds 15

and the F's are the Christoffel symbols of the surface:

F~ = F ~~ = a ~. aT, ~ = a ~ . a , , ~ .

It proves useful to introduce the notations

(1.4) 6~,(u) = 89 + u~l~) = e=,(u) - FAux,

(1.5) e=p(u) = 89 + ua,~) = e=,(t~)

where t~ = (ul, u2) is the tangential displacement.
The bilinear forms for membrane deformation and bending are respectively

E
(1.6) ao(u, v) - 1 - f2 ! [(1 -- 92)7~(u)yg(v) + fTg(u)Tff(v)] dS,

E
(1.7) al(u, v) = 12(1 - 02)! [(1 - ~a)p~(u)p~(v) + fp~(u)p~(v)] dS

where E and f denote the Young modulus and the Poisson ratio of the material.
The shell is supposed to be clamped on a part Bo of the boundary, fixed on
another part B1 and free of the rest. The kinematic boundary conditions to be
prescribed are

u = O, ~u3/& = O on Bo,
(1.8)
u = 0 on B1

where v denotes the normal to the boundary (other boundary conditions will be
considered later, in particular in Section 3). Let V be the space of kinematically
admissible displacements, defined by
(1.9) V = {E v~EHl(f2), v3 eH2(f~), v satisfies (1.8)}

where H 1 and H 2 are the classical Sobolev spaces.

We also define the space
(1.10) H = (L2(S)) 3

associated with the kinetic energy and the subspace G of V of the pure bendings:
G = {v v; = 0} = v; ao(v, v) = 0}

(1.11) ={v~V;ao(V,w)=0 VweV}.

These three definitions of G are equivalent by virtue of classical positivity
properties of the elastic coefficients appearing in (1.6); this same equivalence holds
even in more general cases with anisotropic elastic coefficients. Under the same
assumptions one can prove the coerciveness property [BC76, CM91]:

Lemma 1.1. There exists a constant c > 0 such that

a o ( V , v ) + a l ( v , v ) > cllv[l~ Vv~r.

16 G. GEYMONAT8r E. SANCHEZ-PALENCIA

Remark 1.2. Clearly Lemma 1.1 holds in the (standard) case when the boundary
conditions (1.8) prevent rigid displacements. In special cases, Lemma 1.1 only holds
on the quotient space with respect to these rigid (irrelevant) displacements. The
quotient space will be used without explicit mention.

The (static) problem of deformation of the shell is

(1.12) Find u~ E V such that

ao(u~, v) + eal(u~, v) = ( f V)H V v e V
where e is a small parameter proportional to the square of the thickness of the shell,
and f e H is the given force field.

Definition 1.3. The surface S, (along with the boundary conditions (1.8)) is said to
be stiff(or rigid) if the subspaces of pure bendings have dimension 0 (i.e., G = {0}),
and non-stiff (or non-rigid) otherwise. The surface S is said to be quasi-stiff (or
quasi-rigid) if the subspace G is of finite dimension.

Remark 1.2 applies here, and this sometimes introduces an irrelevant confusion
between the definitions of rigid and quasi-rigid. Correspondingly, a part P of the
surface S is stiff (or rigid) if and only v]p is a rigid displacement when v ~ G (here [p
denotes the restriction to P).
We now introduce the concept of a perfectly stiff surface.

Definition 1.4. Any stiff surface S is said to be perfectly stiffif there exists a constant
c > 0 such that

(1.13) ao(V, v) > c[lv][~ VvE V.

Correspondingly, a part P of S is said to be perfectly stiff if it is stiff and if there
exists a constant C > 0 such that
(1.14) IlVlp[lL2(p) < Cao(V,v) VveV
where we emphasize that the left side of (1.14) involves only the restriction ofv to P,
but the right side involves all of v.

Without giving the details, which may be found in [-SP92J, we recall that the
asymptotic behavior of u, in (1.12) as e N 0 is given by

(1.15) v~ = eu~,
v ~ v o weakly in V,
where v o is the unique solution of

Vo E G,
(1.16)
al (Vo, w) = (l;, w ) . Vw~G
provided G ~ {0}, i.e., in the non-stiff case.
 Rigidity of Surfaces with Folds 17

In the stiff case, a~ V) 1/2 is a norm on V, and (1.12) is a classical singular

perturbation problem. If V denotes the completion of V in that norm, the limit
behavior is given by

(1.17) u~ ~ Uo strongly in IV

where Uo satisfies

U 0 (E IV,
(1.18) ao(uo, w) = (f, w)n V w e IV.

Clearly, this problem h a s a unique solution provided the right-hand side defines
a continuous functional on V, i.e., i f f e IV'. Sincefis any element of H, this is true if
17 c H. Conversely, if 17 c H, then (1.18) is solvable only for s o m e f The property
that f r H is equivalent to (1.13), i.e., to the perfect stiffness of the shell.
In order to understand this special behavior in terms of spectral properties, it
proves useful to define a new space 17 as the completion of V in the norm

(1.19) Ilvll~ = [-ao( v, v) + Ilvll~] ~/2.

The spectral properties of the limit problem (1.18) are classically considered in
the framework of the spaces

(1.20) 17 a H = H ' c I7';

the corresponding operator equation is

(1.21) (Ao - 2)u = f

where Ao is the self-adjoint operator on H defined as the restriction to H of

(Aou, V)~,r = ao(u,v) Vu, v ~ 9 .

Then, we have

Proposition 1.5. The surface S is stiff if and only if2 = 0 is not an eigenvalue of Ao.
More precisely, S is perfectly stiff when )o = 0 belongs to the resolvent set of Ao and is
imperfectly stiff when ). = 0 belongs to the essential spectrum of Ao. Equivalently, S is
imperfectly stiff when the range of Ao is dense in H, but does not fill it, the inverse Ao a
is unbounded from H to H and it is only defined for f belonging to the range of Ao.

We now consider a surface S with folds F. A fold F is a curve in f~ of class C 1

across which the tangent planes to the surface are discontinuous. The angle
0 between the tangent planes to the surfaces on each side of the fold is supposed to
be a function of class C 1 different from 0 and re.
We consider two different types of transmission conditions for u across F:
A fold is f x e d if

(1.22) u Iv1 = u tr2

18 G. GEYMONAT& E. SANCHEZ-PALENCIA

where Ir . denotes the trace on F when considered from the side ~ of the fold. A fold
is clamped if

(1.23) u Iv1 = u IF2, c50 = 0

where 60 denotes the variation (in the linearized sense) of the angle 0 produced by
the displacement u. The explicit expression for 60 may be written by using the
rotation field which will be defined later (Proposition 7.2). It must be continuous
across F. Alternatively, it may be expressed in terms of the variation of the unit
normal vector produced by u (cf. [N85, p. 57]):

(1.24) (~ a 3 = -- [b~pup + u3, ~] eL

The two kinds of folds correspond to different kinds of physical situations.

The case (1.22) is that of two different pieces pasted together with F of class C t,
whereas (1.23) is rather the case of a unique piece with the shape of the folded
surface.

Remark 1.6. One may wonder if it is natural to prescribe 60 = 0 or, conversely, if

the theory of elasticity may explain a variation of 0. We must consider the fact that
shell theory is an approximate description of elasticity when the thickness h of the
shell tends to zero. It is easy to see that 60 ~=0 amounts to a deformation tensor of
order O(1), as is u. This is consistent with the asymptotic structure of the deforma-
tion field in the membrane approximation, but not in the case of pure flexure,
where the deformation tensor is O(h). Thus in the study of the subspace G of pure
bendings and in the definition of stiffness, it is natural to prescribe 60 = O.
A rigorous proof of this fact in the case of plates may be found in [LD89].
Conversely, in the study of the membrane approximation, (1.22) is more appropri-
ate. We shall make precise the specific boundary conditions taken into account.

For technical reasons in Section 3, we also consider relaxed transmission

conditions, where a discontinuity of the tangential component of u is allowed in
(1.22).

Remark 1.7. For surfaces with folds, Remark 1.2 must be understood in the sense
that there may appear a rigid displacement for each smooth component of S. For
the kinetic conditions on the folds, this usually implies that there is only a rigid
displacement for the whole surface, but this may fail to be true in certain special
geometric situations, which are not considered here.

Finally, let us recall Korn's inequalities (cf. [-DL72], for instance), which will be
applied to plane vector fields (with components u~) defined on the bounded domain
f / o f the reference plane:

(1.25) Zlle~(v)llo2 + Ilvll~ ~ c[Ivll 2 VvEHI(~)

 Rigidity of Surfaces with Folds 19

or even

(1.26) ~ [[e=e(v) Iioz __> c II v 1112

on the quotient space in which plane rigid displacement fields are identified.
We classify the points of a surface S as elliptic, parabolic and hyperbolic
according as the second fundametal form is definite, degenerate and indefinite.
Surfaces parabolic at every point are developable. Moreover, the surfaces and their
boundaries are assumed smooth, and the hypotheses of ellipticity or hyperbolicity
are supposed to be satisfied uniformly except in certain cases when the properties
under consideration are clearly local.

2. Regularity Lemmas in the Subspace G of Pure Bendings and

Equivalent Definitions of Perfect and Imperfect Stiffness

According to the definition of V and G given by (1.9) and (1.11), G is a dosed

subspace of H 1 • H 1 • H 2 (piecewise, if folds are present). We also have

Lemma 2.1. I f u e G and if P is a smooth part of S, then u~IPEH2(p), ~ = 1, 2.

Proof. Let us consider the restrictions ofui to P. Using the notations (1.4) and (1.5),
we write 7,~(u) = 0 in the form

e11(~) - F~lu~ - bllu3 = O,

(2.1) e/2(t~) -- ]7~2u~ - b22u3 = 0,

e12(~) -- F~2u a -- bl2u 3 = 0

and we see that the tangent displacement field tl = (ul,uz) is such that
e,a(fi) e H1 (p). By differentiating we have
e~,p(v~)eL2(p), v~ = (ua,e, u2, e),

and using KORN'S inequality (1.25) we see that the first-order derivatives of ul and
u2 belong to Hi(P). []

Corollary 2.2. An equivalent definition of G is

(2.2) G = {v eH2(S); 7~t~(v)----O, v satisfies (1.8)}.
(Of course, this is understood piecewise and with the corresponding transmission
conditions in the case with folds.)

The definition (2.2) is more symmetric than (1.9) with respect to the different
components and will be used later in the Cartesian components.
Let us prove now some equivalent definitions and properties of perfect-stiffness
which will be used in the sequel.
20 G. GEYMONAT& E. SANCHEZ-PALENCIA

Since 17 is the completion of V in the norm

Ilvl[~ = ao(V, v) + Ilvll~,

we have

Proposition 2.3. Inequality (1.13) of Definition 1.4 holds for all v E V or for all v ~ V.

Now we show that perfect stiffness implies H 1 regularity for the tangential
components. More precisely,

Proposition 2.4. Definition 1.4 may be modified in the following way: S is said to be
perfectly stiff if it is stiff and if there exists a constant c > 0 such that
(2.3) a~ +llvzl[~ +llv3l[~) VveV (orevenVvsV).
Of course, if there are folds, the norms at the right side of (2.3) are understood
piecewise.

Proof. Let S be stiff. From

~e(v) = e~(v) - F~pv~ -- b~pv 3

we have
Ile~a(v)llo _-< c(llrllo + Ilvllo) VveV
and using Korn's inequality (1.25) for the field (vl, v2), we get

IIv~lh < c(ll~llo + Ilvllo)

and even
(2.4) }[villi2 + IIv2[l~ + Ilv~llo~ <_-c(ll~'llo~ + Ilvll~) VveV.
Now, let S be perfectly stiff; from (2.4) and Definition 1.4, we obtain (2.3), and
conversely. []

We now give the (elementary and classical) form of G for a plane surface (e.g.,
a plate or a part of a surface which is plane).

Proposition 2.5. Let P be a connected planar part of the surface S. Then the
restriction of u ~ G to P is such that (ul, u2) is a rigid displacement in the plane of
P and u3 is an arbitrary function of H2(p).

3. Quasi-Rigidity Properties of Certain Elliptic Surfaces with Folds

In this section we take advantage of the fact that, for surfaces with elliptic
points, the system 7 ~ ( u ) = 0 is elliptic in the sense of DOUGLIS & N~RENBERG
[DN58] (hereafter denoted D.N.), then the index properties of [ADN65] furnish
quasi-stiffness of such surfaces.
 Rigidity of Surfaces with Folds 21

Let us write the system y~e(u) = 0 in the explicit form

~,~(u) - a,u, - rhua - V2,u2 - b**u3 = ~,~,

(3.1) 722(u) ~ O2U2 - - r122ul - VZ2u2 - b22u3 = ~/22,

~12(U) ~ 1(6~2U 1 71- 631U2) - - F I 2 U 1 - - C22u2 - - b 1 2 u 3 ~-- I//12

where of course the right sides O,p are taken to be zero.

Proposition 3.1. A t elliptic points of the surface S, system (3.1) is D.N. elliptic with
equation indices Sl = s2 = s3 = 0 and unknown indices tt = t 2 = 1, t 3 = O.

Proof. With this choice of indices, the principal part of (3.1) is

0 - bli
(3.2) 82 -- b22

bl2J
and the determinant of the principal symbol is

(3.3) 1 2
--g[b11~l + b22~-2b12~l~2],

which is a positive-definite form when the second fundamental form of the surface
is. []

For example, suppose that S is an ovaloid, i.e., a uniformly elliptic surface

without boundary. According to classical properties of D.N. systems, the solutions
of (3.1) with 0,~ = 0 are C ~ functions forming a finite-dimensional space; this
means that an ovaloid is quasi-stiff. Moreover, it is a classical result that it is stiff
(even under smoothness hypotheses on S much weaker than here [P73, V62]). G is
then the space of the rigid displacements. Moreover, the classical estimate for
solutions of (3.1) is

(3.4) Ilu1112, + Ilu2llU+, + Ilu3llU ~ cll~llU

for any l; here of course we took the quotient space with respect to the rigid
displacements. For l = 0, (3.4) gives (2.3) and we have

Proposition 3.2. A n ovaloid is perfectly stiff.

Let us now consider surfaces S with boundary conditions. Let S be a uniformly

elliptic surface. The differential operator 7 of (3.1) has total order 2. To have
a well-posed elliptic boundary value problem i.e., one with index in the natural
Sobolev spaces, we must adjoin one scalar boundary condition u ~ Bu Ir satisfying
the complementing condition (also known as the Shapiro-Lopatinskii condition
[ADN65, LM68]). We have
22 G. GEYMONAT8~; E. SANCHEZ-PALENCIA

Proposition 3.3.Each o f the following boundary conditions u --* Bu = g satisfies the

Shapiro-Lopatinskii condition:

(a) u t = 9 , (b) u ~ = 9 , (c) u 3 = 9 , (d) a~ut=9, (e) O~u~=9, (f) ~ u 3 = 9

where ~ denotes the derivative normal to the boundary and tangent to the surface, and
ut and u~ are the components o f the displacement tangent to the boundary F and
normal to it (in the tangent plane to S), respectively.

Proof. Taking local maps in the neighborhood of a point P of the boundary F, we

take axes z 1, z 2 normal (to ~) and tangent to F. The operator B is of order o- + tj
with respect to the unknown u j, with o- = - 1 in the cases (a) and (b), ~r = 0 in (c),
(d), (e) and o- = 1 in (f). Let us verify the Shapiro-Lopatinskii condition in case (a)
(the others are analogous). Taking the principal terms in the system (3.1) in the new
z variables, we have
(3.5) ~1ul - bllu3 = O,

~2U2 - - b22u 3 = 0,

1(~2U 1 "~- ~1U2) - - b12u 3 = 0

in the half-plane z~ > 0; here b~ is the second fundamental form in the new
variables at the considered point. Moreover, we prescribe the boundary condition
(the principal part of which is itself)
(3.6) u2=0 forz 1=0,
and we seek solutions of the form
(3.7) u(z 1, z 2) = U e i~z2 + ~zl

with 42 e R\{0}. By eltipticity, ~ is not purely imaginary. System (3.5) becomes

(3.8) ~U1 - b l l U 3 = O,

i~2U 2 -- b 2 2 U 3 = O,

1 ( i ~ 2 U 1 + ~ U 2 ) - - b12U3 = O,

and (3.6) is merely U2 = 0. The Shapiro-Lopatinskii condition is satisfied if there is

no solution with Re( < 0. This is easily checked from (3.8) with Us = 0, since, by
ellipticity, b22 4 0, and then the second equation of (3.8) gives U3 = 0. Finally, the
last equation yields U1 = 0. []

Thus, system (3.1) with any of the boundary conditions of Proposition 3.3 is an
operator with index
(3.9) H p + 1 • H p + 1 • H p ~ (u 1, u2, u3) = u ~ (V11, V22, 712, Bu) ~ (H p)3 x H p + 1/2

for any p _>_o- + 1, where o- was defined in the various cases in the proof of
Proposition 3.3. We point out that the admissible p are those for which the
boundary conditions make sense as traces. Consequently, the kernel of the oper-
ator has finite dimension, and we have
 Rigidity of Surfaces with Folds 23

Proposition 3.4. An elliptic surface S with one of the boundary conditions (a), (b), (c),
(d), (e), specified in Proposition 3.3 is quasi-stiff. With the boundary conditions (f) it is
quasi-stiff in the space Ha x Ha x H E, which does not correspond to (2.2). Then
7forms a D.N. elliptic system in the spaces specified in (3.9) with p > 0 in the cases (a)
and (b) p > 1 in (c), (d) and (e), and p > 2 in case (f). The following estimate then holds:

(3.1o) 2
Ilu~ll~+~ + Ilu=l[~+~ + Ilu~ll~ ~ c(ll~ll~ + Ilgll~+~/~)
for the specified values of p. When the kernel of the operator has dimension > O, then
(3.10) is understood for the equivalence classes of u obtained by taking the kernel as
quotient.

Clearly, Proposition 3.4 furnishes quasi-stiffness with boundary conditions

which are considerably weaker than (1.8). These kinds of boundary conditions were
also considered by VEI<UA [V62] under the name of "bush boundary conditions".
We now consider piecewise elliptic surfaces with folds F. From Proposition 3.1
it is clear that the number of transmission conditions to be prescribed at F is two.
But in (1.22) we have three boundary conditions (or even four in (1.23) when the
condition of constant angle is imposed). Then, as before, for the boundary condi-
tions, we consider here relaxed transmission conditions at the edges, allowing
a mutual tangential displacement of the two parts of the surface adjacent to F.
In order to write down these conditions explicitly, we use local coordinate systems
on each piece of surface S § S adjacent to F. The local orthonormal basis e~
is such that e~- = es is tangent to F; the local coordinates are denoted z ~ -+.
We then have

Definition 3.5. Let F be an edge with angle 0. The transmission conditions (see
Fig. 3.1)

ul- - cos 0 u [ + sin 0 u~- = 0,

(3.11)
u~ - sin 0 ui ~ - cos 0 u~ = 0

+ S+

U+

S_
e~
u;

Figure 3.1
24 G. GEYMONAT ~; E. SANCHEZ-PALENCIA

are called relaxed transmission conditions. Conditions (3.11) together with

(3.12) u~ = u f
are called the strict transmission conditions for a "fixed fold", i.e., for (1.22).

Remark 3.6. Since the system 7~ = 0 implies that the length of the curves remains
constant (in the linearized sense), the relaxed transmission condition allows the
possibility of a tangential sliding of the adjacent paths S + and S - along F.

Proposition 3.7. The relaxed transmission conditions (3.11)for 0 ~=0 and 0 ~- n

satisfy the Shapiro-Lapotinskii condition (with indices rl = rE = 0).

Proof. We must take the principal terms in (3.11). This means that we only
consider terms of order tj for unknown uj, i.e., we only keep the terms in u3:
(3.13) sin0u + = 0,
(3.14) us - cos0u~ = 0,
which are equivalent to u~ = 0 and u~ = 0. Then the proof follows from Proposi-
tion 3.3 in case (c). []

Denoting by B~u = g~, ~ = 1, 2, the two transmission conditions (3.11) (there

written with g~ = 0), in the case of a relaxed fold we have an elliptic operator
analogous to (3.9):
H p+I x H p+I x HP~(ul,u2, u2)
(3.15)
--+ (])11, 722, ])12, Bt u B2u)e(HP) 3 x (H p+ 1/2)2

with p > 1, We also have the analog of Proposition 3.4:

Proposition 3.8. A piecewise smooth surface S with folds (and no boundaries) subject
to the relaxed boundary conditions (3,11) is quasi-stiff 7 is then a D.N. elliptic
operator in the spaces (3.15) with p >-_t. The estimate analogous to (3.10) is
(3.16) [lulll~+l+llu211~+l+llu3ll~_-<c(ll~ll~+llgl]2+l/2), p=>l
where g = (gl, g2) is a possible right side of(3.11). I f the kernel has dimension > O,
this inequality is understood to hold in the quotient space in which the elements of the
kernel are identified.
When S has folds and boundary conditions of the kind of Proposition 3.3,
quasi-stiffness also holds, and the operator works in the natural forms of (3.9) and
(3.15) for p > 1 (unless there is a boundary condition of the type (f) of Proposition 3.3,
in which case p > 2).

Remark 3.9. It is clear that the quasi-stiffness result of Proposition 3.8 holds if the
strict transmission condition holds (Definition 3.5) or even if the condition 30 = 0
for a clamped edge is prescribed. In that case, the estimate (3.16) holds for the
corresponding solutions (if any, as isomorphism properties are no longer true). In
 Rigidity of Surfaces with Folds 25

any case, the right side of (3.16) does not contain the corresponding term for the
supplementary conditions.

Remark 3.10. According to Proposition 2.4, perfect stiffness amounts to (3.16) with
p = 0, which is not allowed. We shall see later that a surface with a fold is never
perfectly stiff.

Remark 3.11. When a surface "lacks" a boundary condition necessary to satisfy the
Shapiro-Lopatinskii condition, the surface is neither stiff not quasi-stiff because the
kernel of the corresponding operator has infinite dimension. This is an obvious
consequence of the isomorphism properties of D.N. elliptic operators [ADN65].
For certain surfaces there is an excess of boundary conditions (with respect to the
Shapiro-Lopatinskii condition) on a part of the boundary and no (kinematic)
conditions on another part. In this case, stiffness may hold (we shall see examples of
it), but the surface is never perfectly stiff ([GLT79, Sec. 5.10] and [SP89b]).

Remark 3.12. Since the "level of regularity" p = 0 is not allowed, the present
treatment is not to be confused with the Lions-Magenes theory [LM68], in which
Sobolev spaces H are replaced by special spaces 3 at that level of regularity.

4. New Smoothness Properties of Elements of G

In Section 2 we studied some general smoothness properties of the elements of

G. We now consider local smoothness properties of the elements of G under
additional hypotheses on the surface S. In this section 7~p(u) is taken to be zero, i.e.,
the right sides of (3.1) as well as g in Proposition 3.3 vanish locally (in a neighbor-
hood of the region under consideration).
Let F (see Figure 4.1) be a fold with angle 0 # 0 and # ~. The adjacent portions
of the surface are denoted by S § S-; they, as well as F, are smooth.
We shall see later that certain kinematic properties of u on F are expressed by
the traces of the second derivatives of u on F. For the time being, u is (piecewise)

S- r

s +

Figure 4.1
26 G. GEYMONAT • E. SANCHEZ-PALENCIA

of class H 2 (Corollary 2.2) and traces of second-order derivatives do not make sense
in general. Moreover, Proposition 2.5 (in the case of the plane) shows that
sometimes such traces do not exist at all. We then introduce new hypotheses (not
satisfied in the case of the plane, of course).

Proposition 4.1. Let S + and S - both be locally elliptic. The transmission conditions
on F describe a fixed fold (1.22) or a clamped fold (1.23) (or are even relaxed (3.11)).
Then u ~ G implies that u Is+ and u Is- are locally smooth (their smoothness depends on
that of S + and S - ). In particular, if the surface is piecewise in C ~~ then u Is- and u Is+
are locally in C ~. The same holds of course in the case of the boundary conditions of
Proposition 3.3, provided F is smooth.

Proof. Our problem is locally in the framework of Section 3. Relaxing constraints

(u2 Ir + = uz Iv- and possibly constant angle) we have an elliptic system with trans-
mission or boundary conditions satisfying the Shapiro-Lopatinskii condition.
Since the local data vanish, we have local regularity. []

Theorem 4.2. Under the general hypotheses of this section, let at least one orS +, S -
be hyperbolic in the region (S +, say) under consideration and let F not be tangent to
the asymptotic curves orS +. Then u Is+ is such that u and its derivatives of order < 2
make sense in LE(F).

Proof. Let u e V. By Corollary 2.2, u Is+ e H 2. Let us change from the tangent and
normal components ul, u2, u3, to Cartesian components in a fixed frame
O, x l , x2, z. Let z = q)(xl, x2) be S + and let ux, uy, u~ be the components of the
displacements. It is known that in this case the system is not of Douglis-Nirenberg
type; it is a non-Kovalevskian system (see Section 7 and [J29]). Eliminating ul, u2
we get a second-order equation for u~ [D1896, SP89b]:

(4.1) r 22uz, 11 - 2(o 12Uz, 12 q- ~0 llUz, 22 = 0.

Let us change to curvilinear coordinates Yl, Y2 in the (xl, Xz)-plane such that
the characteristics of (4.1) (which correspond to the asymptotic curves of S +
according to Proposition 7.2 below) are Yl = const., Y2 = const. Then (4.1) be-
comes

(4.2) uz, 12 = first-order terms in Hi(S+).

Let F' be the projection of F on the (xl, Xz)-plane and let F " be its image in the
(YI, ye)-plane. (Figures 4.2 and 4.3).
Let us prove that uz Ir is locally of class H2(F"). To this end it is sufficient to
prove that ~3uz/~yl and ~uz/Oye have traces on F " belonging to H 1(F") (since in this
case first-order derivatives in any direction have traces e H j (F")). We write (4.2) in
the form

(4.3) .
 Rigidity of Surfaces with Folds 27

r
Z

S§

r'

2
x1

Figure 4.2

' r II

Figure 4.3

By hypothesis, F is not tangent to any asymptotic curve of S +, so that F' is not

tangent to the characteristics, and F " is not parallel to the coordinate axes (Figure
4.3). Then, considering Ouz/Oy2 in (4.3) as a function of Yl with values in H 1(F") we
see that Ouz/c~y2 is a continuous function of Yl with values in H 1(F") (this implies
a classical change of coordinates, see, for instance, [$64, Sec. 113]). In the same
way, Ou~/~?y~ is a continuous function of Y2 with values in H~(F"). Thus
uz Ir ~ HI(F"). Now the Cartesian axis z is (almost) arbitrary, and we may certainly
take it in three linearly independent directions. Then u [re H2(F'). []

5. Examples of Perfectly and Imperfectly Stiff Surfaces

In Proposition 3.2 we gave a result of perfect stiffness for elliptic surfaces

without boundary. We consider here surfaces with boundaries (but without
folds).
28 G. GEYMONAT & E. SANCHEZ-PALENCIA

Theorem 5.1. Let S be a uniformly elliptic stiff surface, satisfying at least one of the
boundary conditions (in the notation of Proposition 3.3)
(5.1) u~ = 0, u~ -- 0
all along the boundary. Then S is perfectly stiff.

Remark 5.2. In Theorem 5.1, the surface may be subjected to additional boundary
conditions. It is stiff perhaps with the help of these additional boundary conditions.
Without them, S is quasi-stiff (Proposition 3.4) but is not stiff in general (think of
a hemisphere with u~ = 0 which may turn around its axis of symmetry in the
linearized sense).

Proof of Theorem 5.1. Since the conditions (a) or (b) of Proposition 3.4 hold, the
inequality (3.10) with p = 0 gives
(5.2) [[ulll~ + Ilu2[I2 + Iluall 2 _-<cllr]l 2,
which is the perfect stiffness condition by Proposition 2.4. []

We now give an example where stiffness is obtained with the help of the
Holmgren uniqueness theorem and analytic continuation. In any case, the analyti-
city hypotheses on S may be relaxed by using other unique continuation theorems
(see, for instance, [T81, Chap. 14]).

Theorem 5.3. Let S be a uniformly elliptic analytic surface with boundary F satisfying
one at least of the conditions (5.1) everywhere on F. Moreover, let F have a part where
both conditions (5.1) are satisfied. Then S is perfectly stiff.

Proof. Only stiffness is to be proved since perfect stiffness then follows from
Theorem 5.1. Since S is elliptic, b~ 1 and b22 do not vanish. We consider the system
(3.1) with ~ = 0. If we eliminate u3 from the first equation, this system is
equivalent to
(5.3) u3 = b{11 [OlUl -- F~lUct],

(5.4)

u +5 u2+( r l-rr2 u =0
The system (5.4) is clearly an elliptic first-order system with the unknowns
Ul, u 2. The boundary conditions u~ = u~ = 0 on a part of F are Cauchy conditions
there; from the Holmgren uniqueness theorem (see [CH62], for instance) ul, u2
vanish in a neighborhood of this portion of the boundary. Moreover, ul, u2 are
analytic on f~ according to [MN57], and by analytic continuation they vanish
everywhere. Then (5.3) also gives u3 = 0. []
 Rigidity of Surfaces with Folds 29

We now give a result of imperfect stiffness for boundary conditions (c) or (f) of
Proposition 3.3. The essential difference from Theorem 5.1 is the difference of the
level of regularity for (3.10): p > 0 in the case of Theorem 5.1, p > 1 for (c) and p > 2
for (f).

T h e o r e m 5.4. Let S be a stiff surface. Let 7 be a part of the boundary where the only
boundary condition is

(5.5) u3 = 0.

Then S is imperfectly stiff. The same conclusion holds when (5.5) is replaced with

(5.6) c~u3 = O.

Proof. We use Proposition 2.3 in order to work in 17 instead of V (see (1.19)

and Definition 1.4). We must prove that there exists a sequence u"~17
such that

(5.7)
I[~(u")llo
- - ~ 0 asn~ co.
Ilu"llo
In the completion process of passing from V to 17, the boundary conditions (5.5)
and (5.6) are lost and the elements of l? are locally (in a neighborhood of 7)
arbitrary elements o f H a x H a x L 2 (this is classical, since H I x H 1 x C~ is dense in
this space). We see that the proof of (5.7) is the same as in the case when there is no
boundary condition at all. Then, the result is classical. The sequences (5.7) are the
"pseudobendings" ([GLT79, Sec. 5.10] and [SP89b]). The explicit proof is not
given here, since we give an analogous proof in a more complicated situation later
(Theorem 6.1). In any case, we point out that (5.7) amounts to giving a Weyl
sequence for the rigidity system (3.1) without boundary conditions. We note that
there is no ellipticity hypothesis in Theorem 5.4. In fact, if there is a non-elliptic
point, then the surface is never perfectly stiff [GLT79, Sect. 5.6]. Then we need only
consider the elliptic case. []

6. Shells with Folds are not Perfectly Stiff

In this section we prove that a shell with a fold, with or without a constant angle
(so that (1.23) or (1.22) hold), is never perfectly stiff. The corresponding theorem is
analogous to Theorem 5.4.
We consider a fold F as in Fig. 4.1. We define local variables y~ -+ slightly
different from those of Section 3, in order to have more symmetric relations
between S § and S-. The local basis e + consists of vectors of unit length with
e~ = e2 along the tangent to F, but e[ and ei- are defined so that they coincide if
the angle of deviation (denoted by 20) vanishes (see Fig. 6.1, which is in the plane
normal to F).
30 G. GEYMONAT8Z E. SANCHEZ-PALENCIA

le~" / /
/f

S§

Figure 6.1

An equivalent form of conditions (3.11), (3.12) is obtained by taking projections

on the two bisectors of the normal section and gives

(6.1) u + - u2 = 0,

(6.2) (u~ - ui-)cos~p + (u + + us = 0,

(6.3) (u + + ui-)sinO - (u~- - us ~, = 0.

Moreover, if the angle of the edge is fixed (so that the two sides are clamped to each
other), we must add the condition that the 2-components of the rotation of S + and
S - at the point under consideration are equal. In the orthogonal frame being used
this component of the rotation is equal to - (6u3)1. Using (1.24) we have

(6.4) [bl,u, + u3, t] + - [bl~u, + u3, 1]- = 0.

Theorem6.1. L e t S be a stiff surface containing a fold F. L e t the angle 2~ of deviation

(Fig. 6.1) be different from 0 and n in the neighborhood o f some point 0 o f F . The
transmission conditions at the fold are either (6.1)-(6.3) ( corresponding to a fold with
a free angle) or (6.1) (6.4) (corresponding to a fold with a f i x e d angle). Under the
assumption that the surfaces S +- are o f class C 3 in a neighbourhood o f O, S is
imperfectly stiff.

Proof. We may assume that the surfaces S + are elliptic in a neighborhood of 0;

otherwise, the surface is certainly imperfectly stiff (cf. [GLT79, Sec. 5.6]). As in the
proof of Theorem 5.4, we pass to the space 17 (see (1.19) and Definition 1.4). It is
clear that condition (6.1) is satisfied by the elements of 17. Conversely, (6.2)-(6.4)
disappear, as is evident by writing them in the form

u + = 89 + + u [ ) -- ctgO(u + -- ui-)],

u; = ~[ - tgO.(u~ + u;) - ctg~(u~ - u;)],

+ + --
u3, 1 - u~, 1 = b ~ u 2 - bl~U~
 Rigidity of Surfaces with Folds 31

As in the p r o o f of T h e o r e m 5.4, we then see that in a n e i g h b o r h o o d of 0,

(6.5) 17_~ {v = ( v + , v - ) e ( H 1 • ~ x L 2 ) 2, v f Iv = v2 Iv}.
We must show that there exist a sequence u" belonging to 17 satisfying

(6.6)
II~'(u")llo *0 as n ~ ~ .
][unllo

In fact, we take u" to vanish outside of a n e i g h b o r h o o d of O, belonging to the

space on the right-hand side of (6.5). O n each of the surfaces S -+, let us define u" of
the form
(6.7) u n+
~ - = - IV+- einr + "YO(nl/2y), u~-+ = v~e me+""YO(nl/Zy)
n
where ~ -+ = const., v + = const, to be determined later, and 0 is some function of
@(IR2), with and integral different from zero on each half plane YI > 0, Yl < 0. An
easy calculation gives (the indices __+ are not written):

(6.8) = ~1 + bar t-ein~ y

f2~1/2 /~V n - l v ~l~in~'y

We n o w choose v and ~ such that the first term of the right-hand side of (6.8)
(which is its principal part as n ~ oe) vanishes at the point 0:

i~lvl -- bllV3 = O,
(6.9) i ~2/)2 -- bz2v 3 =- 0,

~2v~ i
q- ~ 1 / ) 2 -- b12v 3 = 0

for b o t h + and - . The vanishing of the determinant of (6.9) gives

(6.10) b12Ct~2 - g(bll~2
1 2 + b22~ 2) = 0.
Since the surfaces are elliptic, this form is definite and we m a y take ~ 2-+ = 1,
~1 = any of two possible conjugate complex numbers. We shall choose ~ and ~i-
with positive and negative i m a g i n a r y part, respectively. O f course, each satisfies
(6.10) with the corresponding b y . W h e n ~ is known, the system (6.9) gives v up to
a constant factor, for each of the systems + and - . We observe that if v - 0, then
/)2 z# 0. Then we choose v+ = / ) 2 = 1, and v -+ are well determined. The condition
in (6.5) is automatically satisfied because of the choice of/)+ and ~ + . Let us verify
(6.6). F r o m (6.7) we immediately have
(6.11) Ilu"llo = O(F/-1/2)

(the leading term is u~, and we pass to z, = n~/Zy~). M o r e o v e r , since the first line of
the right-hand side of (6.8) vanishes at O, we have
(6.12) 7~a(u") = - [b~p(y) - b~(O)]v30e i'~r + second line of (6.8).
32 G. GEYMONAT 8,:: E. SANCHEZ-PALENCIA

Since the surface S is of class C 3, the coefficients b~a of the second fundamental form
are of class C a and the brackets in (6.12) are of order O ( n - z / z ) in modulus. It then
follows that 7~a is of order O ( n - 1 / 2 ) . Then on account of the smallness of the
support as in (6.11), we have

(6.13) [I 7~p(u")Iio = O(n- ~).

Thus (6.6) follows from (6.11) and (6.13). []

7. The Subspace G in Cartesian Coordinates. Considerations on the Formal

Cauchy Problem in the Characteristic and Non-Characteristic Cases

In this and the following section we obtain "rigidity" properties of the folds with
fixed angle (for which the two sides are clamped to each other). T h e transmission
conditions (6.1)-(6.4) overdetermine the system of rigidity (3.1), which is of total
order 2; thus, the number of classical transmission conditions for such a system is 2.
We already used this property in Section 3 when "relaxing" one of the conditions.
We may expect that conditions (6.1) (6.4) furnish supplementary constraints which
imply rigidification properties of the folds. This is a familiar fact of everyday life,
which we see, for example, when we fold a paper to construct a nearly rigid
structure. This property is highly dependent on whether the fold is a characteristic
of a certain equation which is roughly equivalent to the system (3.1). These
properties will be studied by nearly explicit computations, for which it proves
useful to write (3.1) in Cartesian coordinates. We recall that this description was
already used in the proof of Theorem 4.2.
Let us choose (local) Cartesian coordinates x, y, z such that S may be written in
the form
(7.1) z = ~o(x, y).
The displacement vector u is given in Cartesian components as

(7.2) u = Uxex + uyey + uzez

where ex, etc., are the unit vectors of the axes. The rigidity system in the framework
(7.1), (7.2) takes the classical form (cf., for instance, [D1896])

Ux, x + cp, xU~, x = O,

(7.3) uy, y + ~p,yuz,y = 0,
ux, y + uy, x + q~,xuz, y + (P,yuz, x = O.

According to Corollary 2.2, the subspace G is formed by functions of class H 2

in the components 1, 2, 3. Since the surface is assumed to be sufficiently smooth,
the Cartesian components of the elements of G are of Class H 2. Then we have
locally
(7.4) G ~ {U E H 2 satisfying (7.3) and the kinematic boundary and
transmission conditions}
 Rigidity of Surfaces with Folds 33

R e m a r k 7.1. The rigidity system in normal coordinates is a Douglis-Nirenberg

elliptic system (provided the surface is elliptic), but (7.3) is not. It is a classical
feature of Douglis-Nirenberg systems that their form is not preserved under
a change of dependent variables [DN58].

It will prove useful to define the rotation field e) associated with a solution u of
(7.3):

Proposition 7.2. L e t u be a solution of(7.3) o f class H 2 in s o m e region. T h e r e exists

an associated rotation field o~ defined by

(7.5) cox = u:,y; coy = - u:,x; coz = Uy, x + ~o,xU:,y -- - ux, y -- (p,yuz,~

which is o f class H 1. T h e fields u and e~ are then related by

(7.6) du = ~ /x dx.

The proof of this proposition is merely a verification.

Relation (7.6) has a simple geometric interpretation which explains the term
"rotation field" (Fig. 7.1):

Y surface

d(r+u)
undeforrned surfoce

Figure 7.1
34 G. GEYMONAT• E. SANCHEZ-PALENCIA

The elements of the tangent plane to the surface S remain "rigid" because of the
invariance of the first fundamental form; they are changed by a translation u and
a rotation o), as is apparent from the following relation, which has to be considered
in the linear framework:
d(r + u) = dr + du = dr + o)/x dr.
The elimination of u~, ur in (7.3) leads to a second-order equation for uz:
(7.7) q),yyu. . . . - 2qo, xyuz, xr + q),xxuz, yr = O.
Conversely, if a solution of (7.7) of class H 2 is substituted in (7.3), then u....
ur, r and ux, r + ur, x belong to H 1. Then u~, u r are determined in H 2 up to a rigid
displacement in the plane x, y.
The proof of the following proposition is immediate:

Proposition 7.3. Equation (7.7) is elliptic, parabolic and hyperbolic at elliptic, para-
bolic and hyperbolic points o f S, respectively. Moreover, the characteristics o f (7.7)
are the projections on the (x, y)-pIane of the asymptotic curves of S.

Remark 7.4. System (7.3) consists of three equations of first order with three
unknowns, but it is not Kovalevskian, i.e., every curve is a characteristic. This fact
is easily checked, but it may be proved without computation. Indeed, the Cauchy
problem for a first-order system consists in giving the values of u on a curve; but
because of the geometric interpretation of (7.3) (conservation of the lengths), the
Cauchy data must satisfy the condition of conservation of lengths. Then the
Cauchy data cannot be arbitrary, i.e., any curve is characteristic. This fact is also
linked with the orders of (7.3) and (7.7), which are not equal. See [J29] for other
examples of this fact.

The last remark shows that it is easier to handle equation (7.7) than system (7.3).
In any case, the relation between the Cauchy data for the system and the equation
is not classical. We shall designate a curve as "characteristic" or "non-character-
istic" according to equation (7.7), and not according to system (7.3).
In the remainder of this section we briefly discuss the formal Cauchy problem
for system (7.3) on a plane curve F, locally in the neighborhood of some point P.
Let F be a curve of S contained in a plane which is not tangent to S (in
a neighborhood of P). Taking the plane of the curve as the (y, z)-plane, we may
express the surface (locally) in the form (7.1). The Cauchy problem for system (7.3)
consists in defining u(x, y) in a neighborhood of the origin when u(0, y) is given
(Fig. 7.2). The formal Cauchy problem consists merely in defining all the derivatives
of u on F', which is the projection of F on the (x, y)-plane of the independent
variables. As an interval of the axis y, F' is a characteristic of equation (7.7) when
~o yy(0, y) vanishes and is not a characteristic when ~o.ry(0, y) does not vanish. This
amounts to saying that F' is or is not a characteristic of(7.7) provided that locally it
is or is not a straight line.
Let us consider at first the non-characteristic case, in which q) ry(0, y) 4= 0 in
a neighborhood of the origin. Let u(0, y) be given. Then the second equation of (7.3)
 Rigidity of Surfaces with Folds 35

I"
Y

J Figure 7.2

is a c o m p a t i b i l i t y condition for the Cauchy data (cf. Remark 7.4):

(7.8) ur, y(0, y) + q),r(0, y ) u z , r(O, y) = O.

Let us suppose that this condition is satisfied. Then differentiating the

second equation of (7.3) with respect to x and the third with respect to y, we
obtain

(7.9) q),yrUz, x + ux, yy + q),:,uz, yy = O.

Since q), yy + 0 in a neighborhood of the origin, we solve for uz, x from (7.9) and
substitute it into the first and the third equations (7.3) to get

-1 2 u
Ux, x = qo,yy[(p, xUx, yy q'- (p . . . . yy],
(7.10) bty,x = ( p,yr
- 1 [(p, yUx,yy + r yUz, ry'] - - Ux, y - - (p,xUz, y,

We then see that all the derivatives of u are defined on (0, y) when u(0, y) is known.
We also may define the Cauchy data for (7.7) when u(0, y) is known; these data
are uz(0, y), uz, x(0, y); the first is known and the second is obtained from the third
equation of (7.10). We note that the first-order derivative u,x is defined by the
second-order derivatives with respect to y. This is a usual feature of non-Kovalev-
skian systems [-J29].
Let us briefly consider the characteristic case, when ~0 rr(0, y) = 0, i.e., when F is
a straight line. Without loss of generality, we may take the axis 0y coincident with
F itself, i.e.,

(7.11) ~o(0, y) = O.
36 G. GEYMONAT& E. SANCHEZ-PALENCIA

Then if u (0, y) is known, the second equation of (7.3) and (7.9) are compatibility
conditions, which become

(7.12) uy, r(O, y) - O,

(7.13) u~,y,(0, y) + q),x(0, y)uz, yy(0, y) = 0.

When these conditions are satisfied, uz, x(0, y) may be defined arbitrarily, and
the first and the third equations of (7.3) define Ux,~(O,y) and uy, x(0, y). All the
successive derivatives are then defined on (0, y).
We summarize this discussion:

Proposition 7.5. Let u(O, y) be the data for the formal Cauchy problem for the system
(7.3).
I f ~p,ry(O, y) #p O, then the Cauchy problem for equation (7.7) is not characteristic.
(The Cauchy problem is always characteristic for system (7.3).) The Cauchy data
must satisfy the compatibility condition (7.8). 1fit is satisfied, then u .... ux. ~ and uy,x
are given by (7.9) and the first and third equations of(7.10), respectively.
I f ~o yy(O,y) - O, then the Cauchy problem for (7.7) is characteristic. The data are
equivalent to q)(O, y) = O, to within a rotation. The Cauchy data must satisfy the two
compatibility conditions (7.12), (7.13). Then uz, ~(0, y) may be chosen arbitrarily and
u~,,x(O, y), ur, x(O, y) are 9iven by the first and the third equations of (7.3).

The difference between the non-characteristic and characteristic cases may be

illustrated by the case when the Cauchy data u(0, y) vanish. In the non-character-
istic case, all the derivatives of u on (0, y) vanish. In the characteristic case, uz, x(0, y)
is an arbitrary function. This is a very familiar fact for a surface S fixed to a plane
along a curve F (Fig. 7.2): the fixing is effective only if F is curved (i.e., is
non-characteristic).

Remark 7.6. In the framework of this section, let S be either elliptic or hyperbolic in
the considered region. Then, provided the data are sufficiently smooth, the traces of
u make sense in Ha(F'), according to Proposition 4.1 and Theorem 4.2.

8. Further Study of Folds in the Subspace G. Characteristic and

Non-Characteristic Cases. Rigidification

In this section we consider the transmission conditions across a fold F for

solutions u in the subspace G. The fold F is supposed to be a plane curve, and the
conditions are (1.23), i.e., the fold is considered to have a constant angle (so that the
adjacent parts S § S- are clamped to each other). We know (from Section 3 and the
beginning of Section 7) that such conditions imply extra conditions with respect to
the transmission conditions for the system (3.1): These extra conditions produce
rigidification phenomena. In the first step we perform formal explicit computations
 Rigidity of Surfaces with Folds 37

~o

/
Figure 8.1

analogous to those of the previous section. Then, the smoothness properties of

Sec. 4 allow us to obtain rigorous conditions for such folds.
Let F be a plane fold. We take O y z to be the plane of F, with the axis 0z (locally)
non-tangent to F.
We first consider the case w h e n F is not an a s y m p t o t i c curve o f the adjacent
surfaces S § S - . Thus the plane Oyz, which is the osculating plane to F, is not
(locally) tangent to S § and S-. Then S + and S may be (locally) given by

(8.1) z = ~o + (x, y).

The condition that F is not an asymptotic curve implies that the axis 0y is
not a characteristic of the corresponding equations (7.7) with ~o-+. This implies
that

(8.2) q, y~(0,
+ y) + 0.
The geometric disposition is shown in Fig. 8.1. Of course, S + and S - may be
located on the same side of the plane (e.g., in the region x > 0). We also suppose
that the angle of deviation 2 0 (see Figs. 6.1 and 8.1) is different from 0 and from n.
Let u -+ (x, y) be the displacement vector. Then, u + (0, y) = u - (0, y) is the dis-
placement vector on F. Let us consider it as known for the time being. Then
considering each one of the surfaces, we may use (7.10) to compute the derivatives
of u -+ with respect to x. Moreover, we may compute the rotation vectors co -+ at
points of F, which are defined by (7.5). This gives

gO+ =U z-+
, y,

(8.3) co+ = - + = (~o,~,)

U~Vx + - 1 [Ux, y, + + uz,~],
~0,~
co~ = -- Ux, y -- ~o,~u z,~
+- = -- u~,y + ~p+" + ' - * [ u x , , y + cp,~+ Uz, yv].
,-~ tcp,-~y)
38 G. GEYMONAT&; E. SANCHEZ-PALENCIA

That the vector u takes the same value on both sides of F implies automatically
that cox takes the same value. Moreover, if the angle of the fold is fixed, the vectors
to § and to- must coincide on F. Thus, we must prescribe
(8.4) [~toy~ = ~toz~ = 0
where the symbol [[ ~ denotes "jump":
(8.5) - 9 + - 9 -

We note that, since the surfaces coincide on F and this curve has a non-
vanishing curvature, ~0,rr
+ and q)~y are equal and different from zero. On the other
hand, q),+~and 9,-x are different, since the angle of the surfaces does not vanish. Of
course, Ux,yr(0, y) also takes the same value on both sides. We then have
o = = ,,

which implies that

(8.6) uz, ,y(0, y) - 0.
The condition that [btoz~vanishes gives us again (8.6). On the other hand, u(O, y)
must satisfy the compatibility condition (7.8) (the inextensibility of F), and we have
(8.7) uz(O, y) = a + by.
Moreover, ur, r(0, y) + bqL y(0, y) = 0 implies that
(8.8) u,(0, y) = c - bq~(0, y)
and, since q~ is nothing else than the z coordinate, equations (8.7), (8.8) show that
the projection of the displacement on the yz plane is a rigid (linearized) motion. On
the other hand, ux(0, y) is arbitrary. Thus we have proved

Proposition 8.1. Let F be a fold with fixed angle, which is a plane curve with non-zero
curvature. Let the tangent planes to S +, S- be different from the plane o f F and let
their angle 2~ be different from 0 and ~z. Let u be an inextensible displacement (i.e.,
u ~ G). Then, u Ir reduces (apart from a rigid motion) to an arbitrary displacementfield
normal to the plane of F.

Remark 8.2. This proposition expresses the very familiar and intuitive behavior of
some "elongated curvilinear dihedra with fixed angle" (see Fig. 8.2). The curve
F behaves rigidly in its plane because a modification of its curvature implies,
according to the inextensibility of the neighboring curves F § and F - , a modifica-
tion of the angle of the dihedron.

Remark 8.3. The computations leading to Proposition 8.1 use the existence of
traces of u on F of class Ha(F) (see (7.10) and (8.3), which contains u, yy).Such traces
certainly exist in two cases. The first occurs when S § and S- are locally elliptic (cf.
Proposition 4.1). The second occurs when one of the surfaces, S § for instance, is
locally hyperbolic. By Theorem 4.2 the trace of u on F is of class H 2, and it is
common to S § and S-. This does not cover (apart from degenerate cases) the cases
 Rigidity of Surfaces with Folds 39

Figure 8.2

when S + and S are both parabolic (i.e., developable surfaces) or one is parabolic
and the other elliptic. In such cases, Proposition 8.1 is only formal or, more exactly,
only holds for elements u of G of class H r with r > ~.
There are many cases concerning the types of surfaces S +, S- in the vicinity of
F. We only consider here the two most common cases, concerning a plane edge
which is characteristic for both adjacent surfaces (Proposition 8.4) and character-
istic for one of the surfaces (Remark 8.6).
Let us consider a straight fold F in the vicinity of a point where S + and S- form
an angle 20 different from 0 and 7r. The straight line is of course an asymptotic
curve of S + and S-, which consequently are not elliptic surfaces. We consider the
two cases when the surfaces are simply fixed and when they have a fixed angle (i.e.,
they are clamped to each other).
Let us take the fold F as axis y and the plane yz in such a direction that the two
adjacent surfaces may be described in the form (8.1). Let us suppose for the time
being that the displacement field u(0, y) on F is known. The considerations of
Section 7 in the characteristic case are valid. We then have the analogue of (7.12)
and (7.13):

(8.9) uy+y(0,y) = O implies ud- (0, y) = c,

(8.1o) u+.(0, y) + x(0, y)uz+.(0, y) = 0.

Then, taking the difference of both (8.10) with indices + and - , we get

(8.11) ~qo,x~ uz, y,(0, y) = 0

since ~o x~ does not vanish because the angle of deviation 20 ~ 0, re. We obtain

(8.12) uz(O, y) = a + by,

and (8.10) gives Ux,yy(0, y) = 0, which also implies that

(8.13) u~(O, y) = d + ey.

It follows from (8.9), (8.12), (8.13) that the displacement field u(0, y) on F is rigid.
Otherwise, the field of rotations around F, coy(0,y) is arbitrary (cf. Proposition 7.4,
characteristic case). If the fold has a fixed angle, then o)1(0, y) takes the same value
on the S + and S- sides.
40 G. GEYMONAT~:; E. SANCHEZ-PALENCIA

Proposition 8.4. A fold F which is a straight line such that the angle of S + and S- is
different from 0 and ~z behaves like a rigid segment (see (8.9), (8.12), (8.13)). The
rotation field around it is arbitrary, on each side S +, S - if the surfaces are simplyfixed
and arbitrary along F but equal on both sides if the fold has a fixed angle.

Remark 8.5. The proof of Proposition 8.4 involves traces of the second-order
derivatives of u. Cases when elements of G certainly have the necessary smoothness
are: a) S + and S - are both hyperbolic as in Remark 8.3. b) S + and S - are two
planes. In these cases the required regularity follows immediately from Proposition
2.5. In these two cases Proposition 8.4 is rigorously proved for elements of G. In
other cases (for instance elliptic surfaces degenerating into parabolic ones on F),
Proposition 8.4 should be considered as formal.

Remark 8. 6. The frequently discussed case of the intersection of a plane with any
type of surface is elementary, since the displacement field on the plane surface is
almost explicitly known by Proposition 2.5. The fold F has a displacement field u Ir
which is, up to a rigid motion, a field in the direction of the normal to the plane, and
otherwise arbitrary. This behavior obviously holds for a simply fixed fold and for
a fold with a fixed angle.

Remark 8.7. The comparison of Proposition 8.1 and Remark 8.6 gives a very
interesting property, which will be used in the examples of the next section: Under
the hypotheses of Proposition 8.1, the fold F behaves exactly as if, in addition to S §
and S-, there is a plane containing F.

9. New Examples of Perfectly and Imperfectly Stiff Surfaces

We first recall a general result [P73] on stiffness of surfaces, which will be used
in the sequel. It concerns the stiffness of a very large class of ovaloids, i.e., convex
closed and bounded surfaces. Here a closed surface is understood to be a surface
without boundary that encloses a simply-connected region. The ovaloid may have
folds which are considered "simply fixed". Then, the stiffness results hold afortiori
for a fold with fixed angle. The surface may contain parts which are planes. On the
other hand, stiffness holds for a very large class of functions which contains the
piecewise H 2 functions considered here. We have

Theorem 9.1. Let S be an ovaloid according to the previous definition. Then S is stiff
up to motions of the plane parts, which are of the form described in Proposition 2.5.

Let us comment briefly on this result. If there are no plane parts, the theorem
states that there is the usual stiffness (up to a rigid displacement). However, for
a surface containing plane parts, stiffness holds for the "curved parts", whereas the
plane parts may of course have displacements in the framework of Proposition 2.5.
It is important to note that if the plane parts are removed, the curved part is no
longer stiff: The plane parts play an important role prescribing displacements of the
 Rigidity of Surfaces with Folds 41

form of Proposition 2.5 on their boundaries with the curved parts, but they are not
themselves stiff. We shall see an elementary example of this situation in Remark 9.8.
Familiar examples of this situation occur for convex polyhedra made out of paper.
The structure formed by the folds is stiff, but the plane faces may have displace-
ments as in Proposition 2.5, i.e., infinitesimal flexions.
Another very general result which will be used is the local uniqueness of the
Cauchy problem for an equation with analytic coefficients on a non-characteristic
curve. This is the classical Holmgren theorem [H63, Theorem 5.3.1-1, which holds
in the vicinity of a non-characteristic curve of class C 1. The uniqueness holds for
solutions in the distribution sense, and then in the sense o f H 2 considered here. This
theorem was used in the case of a first-order system in the proof of Theorem 5.3.
Here, we have

Proposition 9.2. Consider the Cauchy problem of Proposition 7.4 (i.e., for a plane
non-characteristic curve F). Let the surface S be (locally) analytic and (locally) either
elliptic of hyperbolic. Let u(O, y) = 0 on F, in the vicinity of some point O. Then any
u ~ G is such that u(x, y) = 0 in a neighborhood of this point.

Proof. Since the surface is either elliptic or hyperbolic, traces in H 2 make sense
(Remark 8.3.). uz satisfies a Cauchy problem with vanishing data and Holmgren's
theorem gives uz = 0 locally. The conclusion follows. []

Remark 9.3. The local uniqueness may be extended to other regions according to
the type of equation. In the elliptic case with analytic coefficients, the solution is
analytic, and global uniqueness follows from analytic continuation. Of course,
analyticity hypotheses may be relaxed by using other unique continuation the-
orems (see IT81, Chapter 14], for instance). In the hyperbolic case uniqueness holds
on the whole domain of determinacy formed by the characteristics (which are of
course the asymptotic curves of S).
We now give an example exhibiting these properties. Let us consider a piece-
wise elliptic surface with edges F which are plane curves, as in Section 3, and with
a free boundary F0, i.e., a boundary of the surface without kinematic boundary
conditions. To fix ideas we may consider the case of Fig. 9.1, i.e., a surface of
revolution with a plane edge F and a free boundary Fo. We denote by S o and S 1 the
parts of the surface indicated in Fig. 9.1. We also assume that S o is analytic. Then
we have

Proposition 9.4. Let S be the piecewise elliptic surface described above (Fig. 9.1).
(a) I f the edge F is simply fixed, then S is not stiff. The space G has infinite dimension.
(b) If the edge F has a fixed angle, then S is stiff.

Proof. Part (a) follows immediately from the considerations of Section 3. Indeed,
relaxing the tangential displacement we are in the situation of Remark 3.11 and the
corresponding kernel is infinite-dimensional. Prescribing again that the jump of the
tangential component vanishes, we remove one degree of freedom (Remark 3.6) and
the dimension of the kernel is again infinite.
42 G. GEYMONAT• E. SANCHEZ-PALENCIA

r 0

Figure 9.1

Let us consider case (b). Since F has a fixed angle, we may use Remark 8.7. The
displacements of S are the same as in the case when the plane disk of F (denoted by
D) is joined to S. But, when considering this case, we may apply Proposition 9.1 to
the convex surface formed by S 1 and th disk D, which is then stiff. As a conseqence,
the curve F itself is rigid (up to a rigid displacement, of course) and we may apply
Theorem 9.2 to the surface S ~ u then vanishes in a neighborhood of Fo, and thus
everywhere on S o by analytic continuation, as in Remark 4.9. []

Remark 9.5. The hypothesis that S o is elliptic in Proposition 9.4 may be replaced
by hyperbolicity provided that the two characteristics (asymptotic curves) starting
from any point of S o encounter F; this is a broad hypothesis.

Remark 9.6. The problem of Fig. 9.1 admits very many variants. Let us consider,
for instance, a portion S 1 of a surface bounded by two (or several) plane edges with
fixed angle (Fig. 9.2). This portion of the surface is stiff, as in the proof of
Proposition 9.4. The uniqueness may then be extended to the adjacent parts of the
surface according to their properties, as in Proposition 9.4 and Remark 9.5.
Let us now give an example of the perfect stiffness of a part of a surface (see
Definitions 1.3 and 1.4).
Let S be a surface formed by a portion of an elliptic surface, denoted by S , an
edge of which is in contact with a plane annulus S § as shown in Fig. 9.3. We may
also consider a complete disk instead of an annulus; the properties are the same.
The fold F may either be simply fixed or with fixed angle (i.e., S § and S - are
clamped to each other).

Theorem 9.7. The surface S described in Fig. 9.3 is such that the part S - is perfectly
stiff.

Remark 9.8. The stiffness of the part S - is known [V62] and is proved by using
Theorem 9.1. That the annulus may be replaced by a disk follows from the
structure of the elements of G for a plane (Proposition 2.5). We remark that the
 Rigidity of Surfaces with Folds 43

Figure 9.2

- -

Figure 9.3

presence of the plane part stiffens S - , but the plane part is not stiff itself. This is
a remarkable fact, widely known by people wearing hats.

Proof of Theorem 9.7. According to Remark 9.8, only the perfect stiffness of S - is
to be proved, i.e., estimate (1.14), which becomes in our case

(9.1) Ilu-Iio 5 c[l~(u)l[o

with an obvious notation. Let us consider the part S +. Since the b~a and the
F~a vanish there, Korn's inequality (1.26) gives

IlUl~ II2 + Ilu~-II 2 _-_ cllr(u)llo2.

Then taking the trace of the tangent component to F (denoted u2), we have

(9.2) Ilu2 Ir112/2 _-< cllr(u)ll 2.

Of course, u2 has the same trace from S § and S-. We now consider the part S -
and we write the estimate for system (3.1) with boundary condition u2 on F ((3.10)
of Proposition 3.4 in case (a), with p = 0):

(9.3) Ilul II2 + Ilu~ I1~ + Ilu; I102<_- c(llr(u)l[ 2 + Ilu21r 112/2);
using (9.2) we obtain (9.1), and even more. []
44 G. GEYMONAT~; E. SANCHEZ-PALENCIA

Remark 9.9. In the situation of Theorem 9.7, only one of the two sides of F is
perfectly stiff; S + is not stiff at all. There is no contradiction with Theorem 6.1.

Acknowledgment. We are indebted to Professors A. L. GOLDENVEIZER and D. G. VAS-

SILIEV for valuable discussions and comments. This work is part of the Project
"Junctions in Elastic Multi-Structures" of the p r o g r a m S.C.I.E.N.C.E. of the
Commission of the European Communities (contract n~

References

[ADN65] S. AGMON, A. DOUGLIS & L. NIRENBERG, Estimates near the boundary for
solutions of elliptic partial differential equations satisfying general boundary
conditions - - Part II, Comm. Pure Appl. Math., 17 (1965) 35 92.
[ASP92] 3. L. AKIAN • E. SANCHEZ-PALENCIA,Approximation de coques ~lastiques
minces par facettes planes. PhOnom~nes de blocage membranaire, Comp. Rend.
Acad. Sci. Paris. S6r. I, 315 (1992) 363-369.
[BC76] M. BERNADOUd~ P. G. CIARLET,Sur l'ellipticitk du moddle lindaire des coques de
W. T. Koiter in Computing Methods in Sciences and Engineering, R.
GLOWINSKI, J. L. LIONS, editors, 89-136, Lecture Notes in Economics and
Math. Systems, Springer, 134 (1976).
[CM91] P. G. CIARLET& B. MIARA, Une ddmonstration simple de l' ellipticitk des moddles
de coques de W. T. Koiter et de P. M. Naghdi, Comp. Rend. Acad. Sci. Paris,
S6r, I, 312 (1991) 411 415.
[CH62] R. COURANT& O. HILBERT,Methods of Mathematical Physics, Vol. 2, Intersci-
ence, New York (1962).
[D1896] G. DARABOUX, Thdorie 9dnkrale des surfaces, Vol. 4, Gauthier-Villars, Paris
(1896).
[DN58] A. DOUGLIS& L. NIRENBERG,Interior estimates for elliptic systems of partial
differential equations, Comm. Pure Appl. Math., 8 (1958) 503-538.
[DL72] G. DUVAUT& J. L. LIONS,Les indquations en mdcanique et en physique, Dunod,
Paris (1972).
[G62] A. L. GOLDENVEIZER, Theory of elastic thin shells, Pergamon, New York
(1962).
[GG77] G. GRUBB & G. GEYMONAT,The essential spectrum of elliptic systems of mixed
order, Math. Ann., 227 (1977) 247-276.
[GLT79] A. L. GOLDENVEIZER,V. B. LIDSKI & P. E. TOVSTIK,Free oscillations of thin
elastic shells (in Russian), Nauka, Moscow (1979).
[GSP91] G. GEYMONAT & E. SANCHEZ-PALENCIA,Remarques sur la rigiditd infini-
tbsimale de certaines surfaces elliptiques non r~gulidres, non convexes et applica-
tions, Comp. Rend. Acad. Sci. Paris, S6r, I, 313 (1991) 645-651.
[H63] L. HORMANDER, Linear partial differential operators, Springer, Berlin (1963).
[3291 M. JANET,Leqons sur les systbmes d'dquations aux dkrivbes partielles, Gauthier-
Villars, Paris (1929).
[K70] W. T. KOITER, On the foundations of the linear theory of thin elastic shells, Proc.
Kon. Ned. Akad. Wetensch, B73 (1970) 169 195.
[LD89] H. LE DRET, Folded plates revisited, Comput. Mech., 5 (1989) 345 365.
[LM68] J. L. LIONS & E. MAGENES, Problbmes aux limites non homogknes et applica-
tions, Vol. I. Dunod, Paris (1968).
 Rigidity of Surfaces with Folds 45

[MN57] C. B. MORREY 8r L. NIRENBERG, On the analyticity of the solution of linear

elliptic systems of partial differential equations, Comm. Pure Appl. Math., 10
(1957) 271 280.
[N853 F. NIORDSON, Shell theory, North-Holland, Amsterdam (1985).
1-P73] A. V. POGORELOV, Extrinsic geometry of convex surfaces, Amer. Math. Soc.,
Providence (1973).
[$64] V. 1. SMIRNOV,Course of higher mathematics, Vol. 5, Pergamon, Oxford (1964).
[$69] J. J. STOKER,Differential geometry, Wiley, New York (1969).
[S75] M. SPIVAK,A comprehensive introduction to differential geometry, Publish or
Perish, Houston (1975).
[SP89a] E. SANCHEZ-PALENCIA,Statique et dynamique des coques minces. I. Cas de
flexion pure non inhibbe, Comp. Rend. Acad. Sci. Paris, S6r. I, 309 (1989)
411 417.
[SP89b] E. SANCHEZ-PALENCIA,Statique et dynamique des coques minces. II. Cas de
flexion pure inhibde, approximation membranaire, Comp. Rend. Acad. Sci. Paris,
S6r. I, 309 (1989) 531-537.
[SP90] E. SANCHEZ-PALENCIA,Passage h la limite de lu tridimensionnelle h la
th~orie asymptotique des coques minces, Comp. Rend. Acad. Sci. Paris, Sbr. II,
311 (1990) 909-916.
[SP92] E. SANCHEZ-PALENCIA,Asymptotic and spectral properties of a class of singu-
lar-stiffproblems, Jour. Math. Pure Appl., 71 (1992) 379 406.
[SPV92] E. SANCHEZ-PALENCIA8r D. G. VASSILIEV,Remarks on vibration of thin elastic
shells and their numerical computation, Comp. Rend. Acad. Sci. Paris, S6r. II, 314
(1992) 445 452.
IT81] M. E. TAYLOR, Pseudodifferential operators, Princeton Univ. Press, Princeton
(1981).
[V62] I. N. VEKUA,Generalized analytic functions, Pergamon Press, Oxford (1962).

Laboratoire de M6canique et Technologie

ENS de Cachan
61 avenue du Pr6sident Wilson
94235 Cachan C6dex
and
Laboratoire de Mod61isation en M6canique
Universit6 Paris 6
4 place Jussieu
75252 Paris Cedex 05

(Accepted May 12, 1994)