We consider a family of linearly elastic shells indexed by their half-
thickness epsilon, all having the same middle surface S = rho(omega),
with rho:omega subset of R(2) --> R(3), and clamped along a portion of
their lateral face whose trace on S is rho rho(gamma(0)), where gamma
(0) is a fixed portion of partial derivative omega with length gamma(0
) > 0. Let (gamma alpha beta(eta)) be the linearized strain tensor of
S. We make an essential geometric and kinematic assumption, according
to which the semi-norm \.\(omega)M = {Sigma(alpha,beta)\\gamma(alpha b
eta)(eta)\\(L2(omega))(2)}(1/2) is a norm over the space V(omega) = {e
ta is an element of H-1(omega); eta = 0 on gamma(0)}, excluding howeve
r the already analyzed 'membrane' shells, where gamma(0) = partial der
ivative omega and S is elliptic. This new assumption is satisfied for
instance if gamma(0) not equal partial derivative omega and S is ellip
tic, or if S is a portion of a hyperboloid of revolution. We then show
that, as epsilon --> 0, the averages 1/2 epsilon integral(-epsilon)(e
psilon) u(i)(epsilon)dx(3)(epsilon) across the thickness of the shell
of the covariant components u(i)(epsilon) of the displacement of the p
oints of the shell strongly converge in the completion V-m(#)(omega) o
f V(omega) with respect to the norm \.\(M)(omega), toward the solution
of a 'generalized membrane' shell problem. This convergence result al
so justifies the recent formal asymptotic approach of D. Caillerie and
E. Sanchez-palencia. The limit problem found in this fashion is 'sens
itive', according to the terminology recently introduced by J.L. Lions
and E. Sanchez-Palencia, in the sense that it possesses two unusual f
eatures: it is posed in a space that is not necessarily contained in a
space of distributions, and its solution is 'highly sensitive' to arb
itrarily small smooth perturbations of the data. Under the same assump
tion, we also show that the average 1/2 epsilon integral(-epsilon)(eps
ilon)udx(3)(epsilon) = (u(i)(epsilon)), and the solution xi epsilon is
an element of V-K(omega) of Koiter's equations have the same principa
l part as epsilon --> 0 in the same space V-M(#)(omega) as above. For
such 'generalized membrane' shells, the two-dimensional shell model of
W.T. Koiter is thus likewise justified.