1973, Koiter, Foundations of Shell Theory

Foundations of shell theory
W. T. Koiter J. G. Simmonds
Delft, The Netherlands Charlottesville, VA, USA
1. Introduction
Shell theory attempts the impossible: to provide a two-dimensional representation of an
intrinsically three-dimensional phenomenon. A shell occupies a volume in space; indeed its
relative thickness is the crucial parameter governing its response to external loads. Yet any
theory of shells, virtually by definition, deals with variables defined only on a reference surface,
normally the middle surface. The expectation, based on intuition and experience, is of course
that the thinner the shell, the more accurately the actual three-dimensional stress and dis-
placement fields can be inferred from a two-dimensional solution (except possibly near edges
or regions of highly concentrated loading). The need of a two-dimensional theory is obvious
because, even in the present computer age, such a theory is from the mathematical point of
view immensely more tractable than a three-dimensional one.
To establish the foundations of shell theory in the strictest sense, it would seem that it
must be shown that, given a boundary value problem in three-dimensional elasticity theory
for a "thin-walled" region, there exists a set of two-dimensional shell equations with appro-
priate boundary conditions which predict the actual stress and displacement fields to within
errors of a specified order of magnitude approaching zero with the shell thickness. To give an
inkling of the enormous complexity of such a task, we note that it was only in 1970-nearly
a century after the appearance of Aron's pioneering paper [I)-that the foundation question
was answered affirmatively for the simplest non-trivial case imaginable. We refer here to the
point-wise precise error estimates in the analysis by Ho and Knowles [2] ·of the linear axi-
symmetric torsion problem for an elastically isotropic shell of revolution of constant thickness,
subject only to static end loads. The success of this analysis hinges on the fact that the exact
elasticity formulation of the problem in question is actually two-dimensional, be it of an entirely
different type than two-dimensional shell theory, and the associated shell problem ill therefore
only one-dimensional. While it seems plausible that this work can be extended to cases in which
symmetry or separation of variables permits the three-dimensional elasticity problem to be
decomposed into a system or sequence of two-dimensional problems, it is clear that more general
situations demand new mathematical tools, if not an entirely different approach.
Perhaps the most serious obstacle in deriving a two-dimensional theory of shells is the
circumstance that the stress distribution in an edge zone, of depth of the order of the shell
thickness h, is nearly always essentially three-dimensional. This difficulty is even aggravated
by the fact that we rarely know more about the applied edge stresses than the resulting
force and moment per unit edge length on the middle surface. This state of affairs suggests
that a solution of shell theory ought to be compared with a class of three-dimensional
elasticity solutions. An essential part of the foundation problem thus becomes the identification
of these "equivalent" classes of elasticity probiems and the quantitative proof of Saint-Venant's
principle for the difference between two solutions in the same class.
Unfortunately, a complete answer to even this relaxed form of the foundation question
seems out of reach at present. For example, in nonlinear problems virtually nothing is known
E. Becker et al. (eds.), Theoretical and Applied Mechanics

© Springer-Verlag, Berlin · Heidelberg 1973
Foundations of shell theory 151
concerning bounds on the errors of shell solutions. The best results to date, obtained by John
[3, 4, 5], consist of effective error estimates tor the two-dimensional differential equations in the
interior domain in the absence of surface loads, but they say nothing about the equally import-
ant question of boundary conditions. In the linear theory of plates Morgenstern [6, 7] has shown
that the en·ors of plate theory approach zero with the plate thickness. A similar technique has
been applied by Nordgren [8], Danielson [9] and the authors [10, 11, 12] to obtain various root
mean square (L 2 ) solution error estimates, but the only point-wise error estimates available are
those of Ho and Knowles [2] for the special case of torsion mentioned before. The results of
Sensenig [13] occupy some intermediate position.
The problem of establishing a firm foundation of shell theory, as we have posed it above,
is the problem of showing that two-dimensional shell theory is consistent with three-dimensional
elasticity theory, and the criteria employed for assessing the validity of shell theory are error
estimates, either for the differential equations (John's approach [3, 4, 5]) or for the solutions.
Many writers, e.g., Green [14], Gol'denweizer and Kolos [15-24], Rutten [25, 26], favour a
different approach for the same purpose, viz. the asymptotic integration of the three-dimensional
equations in terms of the shell thickness h, divided by some characteristic length a, as a small
parameter l . A distinct advantage of the asymptotic approach is its entirely systematic character
which develops shell theory as a logical consequence of three-dimensional theory. At the same
time, however, it is an unfortunate aspect of this systematic character (shared by John's
equally systematic analysis) that the analysis is extremely complicated and tedious, much more
so than is needed in the applications (0£. in particular [26J). Moreover, the asymptotic method
does not enable us so far to assess the accuracy achieved. In spite of these disadvantages the
asymptotic approach has contributed enormously to a deepening of our insight into the shell
problem.
Finally, we mention a fundamentally different approach to shell theory in which the ques-
tion of its foundation is almost irrelevant. We refer to the socalled direct approach in which it is
postulated, ab initio, that the shell is a two-dimensional continuum subject to certain physical
principles analogous to those of three-dimensional continuum mechanics. Various kinematic
and static quantities are represented by three-dimensional Cartesian vectors, whose components
are functions of position on the reference surface only. If a quadratic strain energy per unit area
is postulated, the elastic constants (there are more than in conventional shell theory) are to be
determined by laboratory tests, just as is necessary in three-dimensional elasticity. The notion
of constructing a two-dimensional model of a shell, completely independent of three-dimensional
considerations, originated over half a century ago with the Cosserat brothers [27]. They
characterized the deformation of a shell by a mid-surface displacement vector plus an additional,
usually independent "director" vector. This work was ignored for a long time, and it was only
followed up by many writers in the past decade. We refer to a recent exhaustive exposition by
N aghdi [28J in the "Handbuch der Physik", where a complete list of references may be found.
The elegance, economy and logical consistency of the direct approach, in its simplest form,
are admittedly appealing. Even if the transverse shearing strains and the normal component of
the stress couple vector are initially assumed to be non-zero, the assumption of elastic isotropy
and homogeneity and an argument along the lines suggested by Niordson [29] lead to the field
equations and boundary conditions of classical shell theory. It still seems to us, however, that
the lack of a comparison, or even a connection, with three-dimensional theory is so serious a
weakness that the direct approach, in spite of its considerable merits as a model of shell behaviour,
can hardly contribute to a strengthening of the foundations of shell theory.
In this paper we shall review some recent efforts toward reinforcing the foundations of shell
theory, most of them achieved in the last decade. Wherever possible we have attempted to
simplify and unify or extend known results, and where it seemed appropriate to us we have
drawn attention to open problems. Lack of space prevents us from discussing the significant
1 Reference should also be made here to an early important paper by Goodier [73J, overlooked by most
later writers.
152 W. T. KoiterjJ. G. Simmonds
developments over a longer period of time and it even compels us to a selection in the review
of recent advances. We ask for the reader's indulgence in connection with the personal bias
involved. In addition to the references at the end of the paper we refer to some text books
[30-37,84,85], two review articles [38, 39], and in particular to Love's famous treatise [40],
for more historical information.
To keep the essential ideas in focus, we shall restrict our attention to static deformations
of elastic shells which in their un deformed state are stress-free and of constant thickness h.
We adopt Reissner's point of view [41, 42] that, often, the most convenient path connecting
shell thcory with the three-dimensional theory of elasticity is an ascending one which starts
from shell theory as a base point. We are indeed fully confident that the equations of classical
(linear or nonlinear) first-approximation shell theory, as formulated, for example, in [43-51],
are entirely adequate, and this conjecture will shape our entire approach to establishing the
relationship between shell theory and the three-dimensional elasticity theory.
The body of our paper thus starts with a review of the equations of classical nonlinear shell
theory. There are many ways leading to Rome-or to Moscow, for that matter-and as we have
Eeen, classical shell theory is no exception to that rule. In the main we shall follow the lines of
the discussion in [47], but we shall now incorporate significant improvements due to Danielson
[49]. Lack of space prevents even a brief discussion of the alternative approach by means of a
finite rotation vector [50, 51]. This generalisation of Reissner's highly successful idea in the
case of axisymmetric deformations of shells of revolution [52] holds much promise for future
developments.
In Section 3 we shall exploit John's results [3] in 'Jrder to obtain precise error estimates
for our (approximate) constitutive equations and for the derivatives of stress resultants and
bending strains along the middle surface. This discussion will enable us in Section 4 to derive a
comparatively simple form of intrinsic nonlinear shell equations with rigorous error estimates,
which we venture to call the canonical/orm of these equations. The most striking simplification
in comparison with John's refined equations [4, 5] is the absence of the cubic coefficients of
the strain energy density, and it is largely due to the fact that in John's lowest order interior
equations [3] only the equivalent of the Mainardi-Codazzi equations and the tangential equili-
brium equations are actually in need of a refinement. Additional reasons for the simplified form
of the equations are the proper choice of the tensors of stress resultants and changes of curvature
as dependent variables, due to Danielson [49], and the appropriate use of the lowest order equa-
tions (with error estimates) to simplify the refined equations, as suggested already by John
himself [5]. We note also that the term (1 - v) AKn~ in Danielson's Equation [49, (3.15)] is
missing in our equations, and the omission of this term can apparently onlj be justified by
an appeal to John's error estimates. Finally, we reduce the equations to their canonical form by
introducing the modified tensor of changes of curvature e"'fJ [43, 44,47], defined here by (4.13).
The linearized version of the equations then takes the form of the "best" linear theory in the
sense of Budiansky and Sanders [44].
Whereas John's error estimates for the equations of shell theory represent a significant
advance in the past decade, it would be even more important to have estimates for the errors
of the solutions of the approximate equations, as compared with the solutions of the exact
three-dimensional equations [39]. Some steps in this direction have been made in the case of
the linear theory. We shall make a distinction in Section 5 between intrinsic error estimates,
which provide a rigorous basis for establishing the equivalence of different forms of the linear
shell equations [53-55], and the more fundamental extrinsic error estimates, which provide
bounds on the difference between the three-dimensional stress and displacement fields implied
by linear shell theory and those predicted by classical three-dimensional elasticity theory
[8-12]. An extension of these error estimates to the case of the nonlinear theory is not imme-
diately obvious. In fact, the possible occurrence of limit points or snap loads in the nonlinear
case may imply finite errors of shell theory in the neighbourhood of such a snap load. In this
case it may be more appropriate to compare shell theury and the three-dimensional theory at
slightly different load levels. At present, however, we are not aware of any concrete results
in this direction. On the other hand, the few available results for bifurcation buckling according
to three-dimensional theory fully support our confidence in the predictions of shell theory
[56-59].
For a closed shell, or for a shell on whose edges the tractions are distributed over the thick-
ness in accordance with the requirements implied by shell theory, the root mean square error
+
of the three-dimensional stress field predicted by lineal' shell theory is of order 8 2 = h21L2 hlR,
where R is the smallest principal radius of curvature of the midsurface and L is the smallest
local wave length of the deformat~on pattern [9, 11]. In case of a more "irregular" distribution
of the edge tractions the resulting root mean square errors as well as the local errors in the
interior domain are, in general, much larger, of order c:* = hjL + hlR, whereas the local errors
in the edge zone are even of order unity. The latter fact was already noted by Friedrichs [60],
but the significant effect of the distribution of the edge loads on the accuracy of shell theory in
the interior domain seems to us to have been stressed first by Gol'denweizer (e.g. [17, 22]). In
practice we hardly ever know anything about the actual distribution of edge loads over the
thickness, except in the case of a free edge to be discussed presently. In most cases we shall have
to rest content with errors of order c:*, and the significant effect of the edge load distribution
also detracts much of the value of the many attempts to refine shell theory beyond a "first
approximation", to be mentioned later.
In the special case of a shell with a free edge, discussed in Section 6, it seems again possible
to approximate the actual three-dimensional stress and displacement fields to within a root
mean square error of order c: 2, simply by imposing suitably modified Kirchhoff-type boundary
condi~ions [61]. These modified boundary conditions are obtained by subtracting from the
classical strain energy expression for shells a line integral along the free edge line on the middle
surface, involving the square of the edge twist, thus accounting for the loss in torsional energy
in a zone of depth O(h) near a free edge. The modified boundary conditions may also be visualized
by means of a fictitious edge beam with a negative torsional rigidity along the free edge, and
they agree in the case of a flat plate with the boundary conditions proposed earlier by Gol'den-
weizer [17]. A rigorous proof of a root mean square accuracy of order c: 2 , along the lines of [11],
is somewhat tedious and is still under examination.
Even in the case of a free edge it is not equally easy to obtain an equal accuracy of order c: 2
in the entire edge zone. As pointed out by Green [14], and in particular in a sequence of meti-
culous asymptotic analyses by Gol'denweizer and Kolos [15-24], this would require the
(complicated) solution of a plane strain problem for a semi-infinite strip of width h equal to
the shell thickness, in addition to the (simple) solution of the torsion problem for this strip.
Fortunately, the most important circumferential stresses in the edge zone, namely those at
the outer and inner faces of the shell at the edge itself, may be calculated without solving the
plane strain problem, again with an error of order c: 2 • In the bending case of a flat plate with a
cylindrical hole this analysis has been carried out by van der Heijden [62], and his results
are confirmed by the earlier exact three-dimensional analysis of this problem by Alblas [63].
The formidable problems of a consistent refinement of shell theory beyond its first appro-
ximation, which may be needed for not-so-thin shells, are briefly discussed in Section 7. The
paper ends with mention of some open problems in our topic which in the authors' opinion
merit further research.
2. Classical nonlinear shell theory

Let r(x"'), IX = 1, 2, denote the radius vector from a fixed origin in space to a generic point
on the middle surface of the undeformed shell as a vector-valued function of the pair of Gaussian
surface coordinates. The tangential base vectors are ac< = r,co where the comma preceding the
subscript IX denotes partial differentiation with respect to the coordinate xC<. The covariant
and contravariant metric tensors of the middle surface are defined by aap = a",. ap and by
a",paP). =t:5!. These tensors are employed in lowering and raising indices of all surface tensors,
and the reciprocal base vectors are defined by aC< = a{XPap. The determinant of the covariant
154 W. T. Koiter/J. G. Simmonds
metric tensor is denoted by a, the altenating tensor by elX{J or e"'fJ. The normal to the middle
1
surface is defined by the unit vector n = 'ie"'Pa", X a{J' The Christoffel symbols ofthe first and
1
second kind for the undeformed middle surface are given by r)."'{J = 'i(a).",,{J + a).{J,1X - a"'{J,).)
and r:{J = a").rA"'{J' Covariant surface differentiation with respect to a coordinate XIX is denoted
by an additional subscript (X preceded by a vertical stroke. The second fundamental tensor of
the undeformed middle surface is specified by b"'{J = n . r,"'{J' ,All derivatives in the analysis are
assumed to be continuous. The equations of Gau88 and M ainardi-Codazzi are written in the form
[47, Eqs. (2.20), (2.22)]
e",{Je)."[r"'{J",A + r~"rl<{JA + b",,,b{J).] = 0, (2.1)
e"'fJeA"b{J).I" = elX{Je)."[b{J).,,, - rp"b,,).] = o. (2.2)
A point in undeformed shell space is identified by its distance z to the middle surface and
by the surface coordinates of its projection on the middle surface. The shell faces z = ±hj2,
where h is the constant shell thickness, are surfaces parallel to the middle surface. The coordi-
nate z is orthonormal to the surface coordinates, and the components y",p of the spatial metric
tensor Yij (with determinant y) reduce to a",p at the middle surface. The edge of the undeformed
shell is assumed to be a ruled surface formed by the normals to the middle surface along an
edge curve on this surface. Let v denote the unit vector in the tangent plane, normal to the
edge curve and positive outwards. The positive sense on the edge curve is defined by the tan-
gential unit vector t = n X v.
In a deformation of the shell the original (material) middle surface is described by a vector-
valued function 'r(x"') = r(x"') + u(x"'), where u is the displacement vector. This ne~ confi-
guration of the original middle surface will be called the deformed middle surface, but it
should be borne in mind that it does not in general coincide with the actual "middle" surface
of the deformed shell. The deformed base vectors and metric tensors are defined by 'a", = 'r,,,,,
'a",p = 'a",. 'afJ and 'a"'fJ"a{JA = b!, where the double prime at the contravariant tensor indicates
that indices have not been raised in accordance with our previous convention. The reciprocal
base vectors are defined by "a'" = "a",{J'ap. The determinant of the covariant deformed metric
tensor is denoted by 'a, the alternating tensor by 'e"{J = V'a!ae",p, "e"'fJ = Va!'ae"'p, where the
double prime indicates again that the indices have not been raised by means of a"'p. The unit
1
normal vector to the deformed middle surface is given by 'n = 'i'e"'{Ja", , X 'a{J' and the Christoffel
1
symbols are 'r)'lX{J = 2 ('a).""p + 'a).fJ,'" - 'alXp,A) and "r:p = "a").'r).""B' Covariant surface different-
iation with respect to the deformed metric is denoted by a semi-colon (;) preceding the subscript
cif the coordinate with respect to which the differentiation is performed. The second fundamen-
tal tensor of the deformed middle surface is specified by 'b"'{J = 'n • 'r,,,,p' The equations of
Gauss and Mainardi-Codazzi are the counterpart of (2.1), (2.2) with the primed quantities
(double primes wherever superscripts occur).
The deformation of the middle surface, considered as a surface in space, is described
1
completely by the pair of symmetric tensors Y"'{J = 2 ('a",{J - a",{J) and '{!",p = 'b",{J - b"p, which
are called the middle surface strain tensor and the tensor of changes of curvature 1. Complete
expressions for these strain measures in terms of displacement derivatives are given in [47]
but will not be needed in the sequel. We do need, however, their equations of compatibility,
resulting from the fact that the equations of Gaus8 and Mainardi.Codazzi must hold in both
the undeformed and the deformed configuration of the middle surface. We introduce the
abbreviation
(2.3)
1 We have retained here the primed notation of [47] for the latter tensor in order to keep the bare symbol
available for the "best" modified tensor of changes of curvature for the linear theory [43, 44].
and we note the relations ([47], Section 5)
'r"",p = 'a A" r"",p + ,AtxP' "r"txp -- r"txp + "a"A,}."'P·
11 11 (2.4)
The equations of compatibility, corresponding respectively with Gauss's equation and the
Mainardi-Codazzi equations, may now be written in the form
1 1
Ky~ + e"'PeAP[y",I'!P}. + btxl"ep}. +"2 'e"'I"ep}. + 2 "a"vY ""'I'YvpA] = 0, (2.5)
e",peAI'['ePA!1' - "a"V(b"A + 'e",,) YVPI'] = 0, (2.6)

1
where K = 2 etx"ePPb",pb}", is the Gaussian curvature of the undeformed middle surface.
The classical equations of equilibrium for thin shells, as formulated by Love [40], are six
in number and involve ten unknowns, viz. four tangential stress resultants, two transverse
shear forces and four stress couples. These six equations in ten unknowns may be reduced to a
system of three equations involving no more than six unknowns, viz. a symmetric tensor of
tangential stress resultants and a symmetric tensor of stress couples. This remarkable reduction
is due to Lurie [64] and Gol'denweizer [65]. Its physical explanation is to be found in the
principle of virtual work [47, 66, 67], and for this reason we select from the various methods
available fur establishing the desired equations and boundary conditions this basic principle.
A virtual displacement of the deformed middle surface is described by a continuously
differentiable infinitesimal (additional) displacement field, which may be decomposed with
respect to the deformed reciprocal base vectors and the deformed unit normal vector
(2.7)
The associated virtual changes in the strain measures of the middle surface are given by
[47, Eqs. (3.42), (3.43)]
(2.8)
(2.9)
where "b"IX = "a"P'b tx" and 'cap = "b"'b
IX "p'
We restrict our attention to spatial virtual displacement fields which obey the Kirchhoff-Love
assumptions with respect to the deformed middle surface, and we emphasize that this involves
no approximation but only a restriction of our considerations. The associated virtual strains
at any point in the deformed shell space x"', 'z, where x'" are the surface coordinates of its
projection on the deformed middle surface and 'z is the distance to this deformed middle sur-
face, are then homogeneous linear expressions in by",p and b'e",p at this point x'" of the middle
surface. It follows that the internal virtual work b'V per unit area of the deformed middle
surface is a linear form in the virtual strain measures whose coefficients are necessarily sym-
metric tensors of stress resultants and stress couples
b'V = 'naP (jYap + 'maP (j'e",p. (2.10)
The external loads on the shell faces and the body forces are reduced to statically equivalent
loads acting in the deformed middle surface, in accordance with our restricted Kirchhoff-Love
type spatial virtual displacement field. This leduction is discussed in detail by Naghdi [38].
For the sake uf brevity we shall assume that the reduction of loads does not introduce surface
couples on the deformed middle surface. The reduced loads per unit area of the deformed middle
surface may be decomposed with respect to the deformed base vectors and unit normal vector
(2.11)
In connection with our restricted Kirchhoff-Love type virtual displacement fields we replace,
for the time being, the shell edge by a ruled surface in the deformed configuration, formed by
the normals to the deformed middle surface along the (deformed) edge curve on this surface.
This involves no approximation in the linear theory, where no distinction need be made between
the undeformed and deformed geometries in deriving the equations of equilibrium and the
dynamic boundary conditions, and the justification of this approximation in the nonlinear
first-approximation theory will be discussed in Section 3. The loads on the shell edge are now
likewise reduced to statically equivalent line loads along the (deformed) edge curve on the
deformed middle surface, a force 'N and a couple 'M, both per unit arc length in the deformed
state, where the couple vector lies in the (deformed) tangent plane. Let 'v = 'v,,"a" denote the
unit vector in the deformed tangent plane, normal to the edge curve and positive outwards.
The positive sense on the edge curved is defined by the (deformed) tangential unit vector
't = 't", "a'" = 'n X 'v. We may decompose the edge loads in several ways, for example
(2.12)
(2.13)
The principle of virtual work now yields by standard methods the equations of equilibrium
([47], Eqs.(6.3), (6.4))
(2.14)
'mc</3 ;a/3 _ 'c a/3 'm a/3 - 'b a/3 'n a /3 - 'p = 0' (2.15)
as well as the dynamic boundary conditions ([47], Eqs. (6.10)-(6.12))
('n/3a + 2"b~'m/3"') 'P/3 = 'Na + "b~'M><, (2.16)
"m"/3 ;" '.,'/3 + ('m a/3,., 't)
/3 ,'8 -- 'M
'IX
'Q (v),'8 - , (2.17)
'm cx /3,.,'ex '.,'/3 -- - 'M (t), (2.18)
where the subscript's preceded by a comma denotes partial differentiation with respect to
arc length along the deformed edge curve. We emphasize that Eqs. (2.14) and (2.15) are both
necessary and sufficient to ensure the equilibrium of any shell element of finite (deformed)
shell thickness 'h(x a ), bounded by the shell faces and by the ruled surface formed by the normals
to the deformed middle surface through any finite or infinitesimal smooth simple closed curve
on his surface.
Danielson and one of the authors have proposed the introduction of modified symmetric
tensors of stress resultants and stress couples [49, 86]
n"/3 = V-'a/a['n /3 + 21 ("b~'rnx/3 + IIb~'mxc<)],

cx m lX /3 = V-
'a/a'm /3,
lX (2.19)
with the purpose of simplifying the equations after introduction of the (approximate) stress-
strain relations. We note that their tensor of stress resultants agrees in the linear theory with
the tensor employed in [43]. Employing the identity, holdlng for any differentiable symmetric
tensor T a /3
(2.20)
one first obtains from (2.14) and (2.15) the modified exact equations of equilibrium [49, Eqs.
(2.12), (2.13)]
+ IIb~mxfJl/3 + "b"'''(2Y''illfJ - YilfJl,,) m ilfJ + V'a/a'p'" = 0, (2.21)

mafJ/afJ + ["a"''''(2YxillfJ - YAfJl") mAfJJla - 'b"fJ ncxfJ - V'a/a'p = o. (2.22)
An obvious and distinct advantage of the exact equations in this form is already that all
covariant differentiations, like those in the compatibility conditions (2.5), (2.6), are now per-
formed with reference to the undeformed metric which is known a priori. We refer to ([49],
Eqs.(4.6)-(4.8)) for the similar modification of the exact dynamic boundary conditions. We
postpone the further (approximate) simplificatio'n of the equations and boundary conditions
to Section 4 where this approximation may be argued more securely on the basis of the results
achieved by John [3].
We emphasize once again that the equations of compatibility (2.5), (2.6) and the equations
of equilibrium, (2.14), (2.15) or (2.21), (2.22), are all fully exact. A minor approximation has
only been made in the dynamic boundary conditions (2.16)-(2.18) in the case of the nonlinear
theory, and this approximation will be justified in Section 3. The foregoing theory, however,
involves six equations in twelve unknowns, the symmetric tensors of middle surface strains,
changes of curvature, stress resultants and stress couples. It has to be supplemented by
constitutive equations connecting the stress resultants and stress couples on the one hand to the
strain measures on the other hand. Such constitutive equations necessarily always involve
certain approximations, both in the linear and nonlinear theories (unless in a direct theory all
relations with three-dimensional theory are ignored). As we have mentioned before, this
approximate character of all conceivable two-dimensional shell theories is an inevitable
consequence of our desire to describe the behaviour of an actually always three-dimensional
structure in terms of two-dimensional quantities.
We shall restrict our present discussion of the constitutive equations to the simplest possible
case, based on the following assumptions:
(a) the material of the shell is elastic, homogeneous and isotropic;
(b) the strains are small everywhere in the shell;
(c) the state of stress is approximately plane and parallel to the middle surface.
We note from John's analysis in [3] that the last assumption is actually a mathematical
consequence of assumptions (a) and (b) in the absence of surface loads and at a sufficient
distance from an edge. In the presence of slowly varying surface loads assumption (c) is easily
made plausible by a physical argument, and the solution error estimates [8-12], discussed
in Section 5, confirm this argument. In the neighbourhood of a shell edge, however, the last
assumption, in general, fails to hold true in a zone of depth O(h) near the edge, and the required
modification of the theory will be discussed in Section 6 for the case of a free edge.
Let EiXfJA" denote the tensor of elastic moduli for plane stress in the middle surface
(2.23)
where G is the shear modulus and v is Poisson's ratio. Let V denote the (approximate) expres-
sion for the strain energy per unit area of the undeformed middle surface, given by Love's
formula [40, 43, 47, 29]
V = ~ hEafJ A" lYafJYAll + 112 h2 '(!"fJ'(!A" 1· (2.24)
The associated constitutive equations are by (2.10)

1 -
'nlXfJ=-Val'a (8V - +8YfJIX
-,
8V) 1 -
'mcxfJ = -Val'a (8V
__ + -8V) . (2.25 )
2 8YIXfJ 2 8'(!lXfJ 8'(!fJrx
The relative error in (2.24) has been estimated (qualitatively) in [43] to be of order of magni-
tude 1], hlR or h2 1L2, whichever of these numbers may be critical, where 1] is the maximum
absolute principal strain in the shell, R is the minimum principal radius of curvature of the
middle surface, and L is the minimum "wave length" of the deformation pattern on the middle
surface. It follows that the associated constitutive equations, (2.25), involve similar errors,
and we may therefore replace the square root by unity without any essential additional error.
Moreover, we may also take the modified stress resultants and stress couples (2.19) to be speci-
fied by the same expressions in terms of the strain measures, viz.
mlXfJ = .
Eh 3
12(1 - v2 )
[(1 - v) ,nlXfJ
<.:
+ vacxfJ'n"]
<':" ,
(2.26)
where E = 2(1 + v) G denotes Young's modulus. The inversion of these (approximate) consti-
tutive equations reads
_ 1 ,,12
YrxfJ - Eh [(1 + v) nap - vaafJn ,,], '(!afJ = Eh 3 [(1 + v) mcxfJ - va",pm~]. (2.27)
The (approximate) stress-strain relations (2.26) or (2.27) complement the exact equations
of compatibility (2.5), (2.6) and the equally exact equations of equilibrium (2.21), (2.22) to a
158 w. T. KoiterjJ. G. Simmonds
complete eight-order set of twelve equations in twelve unknowns. From a mathematical point of
view it would seem immaterial whether the constitutive Eqs. (2.26) or (2.27) are employed in
order to eliminate either the stress resultants and stress couples or the strain measures. From
a physical standpoint, however, we shall explain in Section 3 our definite preference for re-
taining the tensor of stress resultants nrx{J and the tensor of changes of curvature '(Lrx{J as our
basic dependent variables, eliminating the stress couples and the middle surface strains by
means of the second Eq. (2.26) and the first Eq. (2.27). We also postpone a discussion of the
considerable simplifications which may be applied to the basic exact Eqs. (2.5), (2.6), (2.21)
and (2.22) in view of the approximate nature of the stress-strain relations.
3. Reappraisal of John's results

[3,4,5]
John's greatest contribution to shell theory, in our opinion, was to derive rigorous and
concrete pointwise estimates for all stresses and all their derivatives in the interior domain
and in the absence of surface loads [3]. He accomplished this by first using the equilibrium and
compatibility conditions of three-dimensional theory for a nonlinear elastic isotropic solid to
derive inequalities for the L 2 -norms of the derivatives, and then invoking Sobolev's lemma to
infer pointwise inequalities. It will be convenient here and in the sequel to assume, without any
loss in generality, that the components of the metric tensor all> a22 and its determinant are all
of order unity. The "generalized" minimum principal radius of curvature R of the middle surface
of the undeformed shell is defined in John's analysis by estimates on the order of magnitude
of the second fundamental tensor brx{J of the undeformed middle surface and all its derivatives l
brx{J = O(R-l), b"'{J[><l"'><n = O(R-l-n). (3.1)
Further important geometric parameters are the (constant) shell thickness h and the distance
d to the shell edge. The magnitude of the strains in the shell is characterized by the maximum
absolute principal extension r;. Basic in John's analysis are then the dimensionless parameter
(3.2)
and the length parameter

A = h/() = min (d, VhR, h/V;J). (3.3)
In his three-dimensional analysis John employs the spatial stress tensor tii, defined by [3]
"= V-
t'J
8W
g/'g-, (3.4)
8Yii
where W is the strain energy density function of the isotropic elastic solid, written symmetri-
cally with respect to Yii and Yii, and 'g is the determinant of the spatial metric tensor in the
deformed body. With the aid of his pointwise estimates, John was able to show that
trx{J(x", z) = trx{J(x") - z(Jrxp(x") + O(Er;()2), (3.5)
t",3(X", z) = O(Er;()), t 33 (X", z) = O(Er;()2) , (3.6)
1
YcxP = E [(1 + v) t",p - va",pt~] + O(r;()2), (3.7)
1
'erxp =E [(1 + v) (Ja{J - vaapa~] + O(r;()2/h) , (3.8)
tii["l ... ,,)X"', z) = o(Er;A -n), (3.9)

where Ya{J and '(La{J are again the middle surface strain tensor and the tensor of changes of
curvature ('b ap - b",p). In his first paper [3] John also derived his "lowest order interior shell
1 From a physical point of view it would be worthwhile to try and relax the required estimate for the
derivatives to O(R-l-n/2 h- n/2 ) or O(R-IJ.-n), where J. is defined by (3.3).
equations" as a consequence of his various estimates
- ~ -r~I~ + SaPIXA!, (ba!"ePA + -} 'eIX!"ePA) = 0('Y)(j2j).2) , (3.10)
SIXPSA!"epAI!' = 0('Y)(j2jh)') , (3.11)

Eh2
12(1 _ p2) 'e~l~ - (bIXp + 'elXp) -r IXP = 0(Eh'Y)(j2jJ.2) , (3.12)
-rPIXlp = 0(E'Yj(j2jJ.) , (3.13)

written as far as possible in our notation.
It was a considerable disappointment to note [47] that John's lowest order interior shell
f<lquations are evidently inadequate to cover all possible situations since they fail already to
describe, for example, the simple cases of nearly inextensional bending of the linear theory. In
an attempt to eliminate these shortcomings, John presented a refined analysis in [4, 5]1. If
his first offspring, Eqs. (3.10)-(3.13), was a bit on the lean side, his second one was depressingly
ponderous.
Complexity a::;ide, the most disturbing feature of John's refined interior shell equations,
which he expresses in terms of Yap and 'eap, is that the three coefficients of the cubic terms in
the strain energy density function W appear sprinkled throughout. Physically, this is quite
unsatisfactury, if nut unacceptable. Many engineering materials such as steel or aluminium
have very low yield strains, less than 0.005, making an experimental determination of the cubic
elastic coefficients virtually impossible. Moreover, the variation of the constants of linear
elasticity from one specimen to another can be expected to mask completely any contributions
coming from higher order terms in the strain energy density. Fortunately, all can be set aright
by a judicious combination of John's results in [3] and the exact equations of classical nonlinear
shell theory (2.5), (2.6), (2.21), (2.22). A crucial point in this combination is the proper selection
of the dependent variables in the problem, and it is also essential to restrict the refinement to
those equations which are actually inadequate in John's lowest order theory.
Perhaps the most compelling reason for the elimination of the middle surface strain tensor
Yap is to be found in the linear membrane theory of shells in the form of the simplified equations
of equilibrium
(3.14)
where the loads are now referred to the undeformed middle surface and its base vectors and
unit normal vector. For shells of positive Gaussian curvature, subject to smooth surface loads
and restrained at the edges so as to prevent (nearly) inextensional deformations, the linear
membrane equations often yield an accurate description of the actual stresses away from narrow
edge layers. The general solution of the three Eqs. (3.14) in the three unknowns of the symmetric
tensor n"P is independent of the material properties, and this feature is of course lost, if the stress
resultants are eliminated in favour of the middle surface strains.
Nearly equally powerful is the argument at the other end of the scale, that the equations
of the in extensional bending theory likewise ought to fall out as a simple special case of the
general theory. When this theory applies, the stress resultants may be obtained, in terms of
the transverse shear forces, again from the equations of equilibrium. It is obviously unreasonable
to attempt their calculation from the (small) middle surface strains. In fact, in many cases of
(nearly) inextensional bending it is convenient to introduce zero middle surface strains as a
constraint, and the only means available to determine the stress resultants (as internal "reac-
tions") is then the consideration of equilibrium. A similar situation is found in the theory of
curved beams, where the axial force is always obtained from equilibrium considerations, even
if extension of the center line are permitted.
Inspection of the equations of compatibility (2.5), (2.6) in the case of nearly inextensional
bending shows immediately that it would be equally unwise to eliminate the tensor of changes
of curvature 'el'<p in favour of the tensor of stress couples. Our argument to retain the pair naP
1 The equations in [4] contain some clerical errors which have been cQrrecbed in [5]. Cf. also [68].
and 'elXfJ as the proper dependent variables is also supported by the well-known static-geo-
metric analogy of linear shell theory [64, 65], which holds for a suitably defined modified tensor
of changes of curvature [38, 44].
We shall now examine the consequence for the accuracy of our approximate stress-strain
relations (2.26), (2.27) which may be drawn from John's analysis in [3]. It should be remember-
ed here that the original stress resultants ,nlXfJ and stress couples 'm"'fJ have been defined in (2.10)
by means of virtual work in spatial virtual displacement fields which obey the Kirchhoff-Lo\'e
assumptions with respect to the deformed middle surfl;tce. A point in deformed shell space
may be identified by the surface coordinates of its projection on the deformed middle surface
and by its distance 'z to this surface. The virtual strains in a surface 'z = const. parallel to
the deformed middle surface are in this new coordinate system (cf. [10], Section 4)
1
~'Y"fJ('z) = ~Y"fJ - 'z ~'e"fJ + 2' 'z2("b~ ~'e"fJ + "bp.~'e"", - 2"b~"b~ ~Y"A)' (3.15)
Let ,tii(,z) denote the stress tensor in deformed shell space, referred to our new coordinate
system, in a point at a distance 'z to the deformed middle surface. The internal virtual work
per unit deformed volume is given by ,tii(,z) ~'Yii('z). Integrating this expression over the
deformed shell thickness, say from 'z = -c1 to 'z = c2 , we obtain for the internal virtual work
per unit area of the deformed middle surface
c, __
J 't"fJ(,z) ~'Y "fJ('z) V"gj'a d'z,
-c,
(3.16)
where "g is the determinant of the spatial metric tensor in the new coordinate system in a point
at a distance 'z to the deformed middle surface. Equating the result of (3.14), (3.15) to (2.10),
we obtain (exact) expressions for the original stress resultants and stress couples in terms of
the stress distribut,ion over the shell thickness
(3.17)
-c,
(3.18)
It should be borne in mind, and our presenL notation should also make this abundantly
clear, that the stress tensor ,tii is not identical to John's stress tensor tii. The material points
x", 'z in the new coordinate system are also distinct from a material point x", z (except at the
middle surface where z = 'z = 0), and even c1 and c2 are (slightly) different from hj2. All
differences, however, are small by John's estimates, and the same holds true for surface
derivatives. In fact, let 'x", 'z denote the coordinates in the new system of the same material
point as described by x", z in John's coordinates. We have then the estimates 'x" = x" + O(Ol'jz);
'z = z + O(l'jz); cl> c2 = hj2 + O(l'jh), and we obtain from these and John's estimates for the
derivatives of the stresses 't"fJ(,x", 'z) = t"fJ(x", z) + 0(El'j02). Likewise, we have for the
differences between the oliginal stress resultants (3.17), (3.18) on the one hand, and the
modified quantities n"fJ, m"fJ in (3.19) on the other hand, the estimates Ehl'j02 and Eh21'j02
respectively. Our result for erwr estimates in the stress-strain relations (2.26), (2.27) is now
a consequence of (3.7), (3.8), viz.
=
Y"fJ A[(l + v) n"fJ - va",fJn=] + 0(1702), (3.19)
m"'fJ = D[(l - v) 'e"fJ + va"'fJ'e=J + 0(El'jh20Z) , (3.20)
where A = (Eh)-l is the extensional flexibility and D = Eh3 j12(1 - v2) is the flexural rigidity.
In each (surface) differentiation of these relations the error term is divided by A.. We also note
the sharper results
(3.21)
corresponding to John's Eqs. (3.11), (3.13). The verification of the second of these equations is
accomplished by substitution from (3.19), (3.20) and their derivatives into the exact equation of
equilibrium, (2.21), for zero surface loads.
The foregoing results based on J olm's analysis hold, of course, only for shells loaded by
edge tractions, in the absence of surface loads. The qualitative physical argument in [43] for
the validity of the stress-strain relations (3.19), (3.20), however, remains applicable in the pre-
sence of sufficiently smooth surface loads, provided that the parameter 0(3.2) is now replaced by
o= max (hIL, hid, Vh/R, V;:J), (3.22)
where L is the minimum "wave length" of the deformation pattern on the middle surface in
the sense of [43]. Our complete confidence in this physical argument is confirmed, at least in
the case of the linear theory, by the rigorous solution error estimates in [8-12], cf. also Sec-
tion 5.
On the other hand, the breakdown of the error estimates in the stress-strain relations near
edges, where 0 grows indefinitely, seems to be irreparable, except in the case of a special distri-
bution of edge tractions conforming to the requirements of the two-dimensional theory [10, 11].
The minor approximation made in Section 2 in the discussion of boundary conditions for the
non-linear theory, involving the replacement of the actual deformed shell edge by the ruled
surface of normals to the deformed middle surface along the edge curve in the middle surface,
implies relative errors of at most 0(0 2 ), and these may certainly be neglected in comparison
with the far more serious errors in the stress-strain relations in the edge zone. We cannot
escape concurrence with Gol'denweizer's view (e.g. [22]) that the interaction between the three-
dimensional boundary layer in the edge zone and the approximately two-dimensional interior
state of stress cannot be ignored in a thin shell. It is only a poor consolation that the question
is largely of academic interest because the actual distribution of the edge tractions is rarely
known with any precision. The exception is, of course, the case of a free edge to be discussed
in more detail in Section 6.
4. Canonical form of intrinsic equations for first· approximation nonlinear shell theory
For the sake of brevity we restrict our (approximate) simplification of the exact basic
Eqs. (2.5), (2.6), (2.21), (2.22) to the case of zero sur-face loads where John's results, discussed
in the previous section, provide rigorous error estimates for the reduced equations.
For the first term in (2.5) we have immediately the estimate
Ky~ = 0(17/R2) = 0(17()2/).2) , (4.1)
and for the last term we have equally easily the same estimate
"a"·Yx(XI'Y.fJi. = 0(17 2/).2) = 0(17 02 /).2). (4.2)
For the second term in (2.5) we obtain from (3.19) and the second Eq. (3.21) after some simple
algebra
(4.3)
and the final simplified equation (4.9) corresponds completely to John's lowest order Eq. (3.10).
The reduction of (2.6) requires somewhat more algebra, virtually spelt out in detail by
Danielson [49]. Replacement of "a"· by a"· involves an error term of order 17()4/hA, and errors
of the same order are accepted in. the reduction of the other terms. Use is made in particular
of both estimates (3.21) and the resulting relation
Yp/" = -vAn~.fJ + 0(17()2/).) , (4.4)
and the final simplified Eq.(2.6) is presented in (4.10).
In the equation of equilibrium (2.22) we drop the second term, and we accept therefore an
error term of order Eh2 17()2/).2. The first term in (2.22) is reduced by means of (3.21) to
m"fJl"fJ = Dle~l~ + 0(Eh217()2/).2) , (4.5)
and our final reduced equation is (4.11), again in complete correspondence to John's lowest
order Eq. (.312).
The tangential equilibrium equations, (2.21), require most algebra in their reduction [49].
In the second term we replace "a"" by a"", accepting an error term of order Eh17()4/X By means
of the stress-strain relations and the second Equation of (3.21), the second term in (2.21) is
now reduced to
(2y11P - YAp\") nAP = A [2nAn Ap+ 2v(nAn AP - n"Pn1)]IP
- 1j2 A [(1 + v) nApnAP - vn1n~]\" + 0(Ehrl)4/). ). (4.6)
For any continuously differentiable symmetric surface tensor T"P we have the identity [49]
(T1TAP - T"PT1)IP = 1/2 (TAPTAP - T1T~W. (4.7)
The second term in (2.21) is thus reduced to
"a""(2Y"AIP - YAPI") nAP = 2A(n1n.l.P)IP - 1/2 A [(1 - v) nApn.l.P + vn1n~]\" + 0(Ehr,,;4j). ). (4.8)
The last term but one in (2.21) is of the same order as the error term already accepted, and it
may therefore be omitted. In the remaining terms we replace "b~ by 'b~, with again the same
error term, and we reduce these terms by means of the stress-strain relations and the first
equation of (3.21). The resulting reduced equation is presented in (4.12), and we have now
arrived at the complete set of simplified equations
-An~\~ + b:8'e~ - b~'e~ + 1/2 'e:8'e~ - 1/2 'e~'e~ = 0(r]f)2j). 2) , (4.9)
'e~lp - 'e~l" + (1 + v) A [(b~ + 'e~) nl,p + (bq + 'e~) n~I"] - vA(b1 + '(1) nL = 0(1]04jhA) , (4.10)
D'e~l~ - (b:8 + 'e:8) n~ = 0(Eh21]02p. 2), (4.11)
n~lp + 2A (n~n~)IP - 1/2 At5~[(l - v) n~n~ + vn=n1]IP
+ 1/2 (1 - v) D(b~'e~ - b~'e!)IP + D(b~ + 'e~) 'el,p = 0(Eh1]()4jA). (4.12)
We note first of all that our Eqs.(4.9)-(4.12) agree with those of Danielson [49], derived
by a similar, if perhaps less conclusive argument, with the single exception that our derivation
on the basis of John's analysis in [3] has enabled us to omit the term (4.1) in Eq. (4.9). This
represents, of course, only a small further simplification, but it is not devoid of significance
since our Eqs.(4.9) and (4.11) are now formally identical to John's corresponding lowest order
Eqs. (3.10) and (3.12). After all, this is not too much of a surprise. The occasional failure of
John's lowest order equations, e.g., in the case of (nearly) inextensional bending, is indeed due
to inadequate accuracy of his other Eqs. (3.11) or (3.13), in particular the latter equations in
the inextensional bending case. It is no wonder therefore that a refinement only of these in-
adequate equations, as presented in (4.10) and (4.12), is actually sufficient in the context of a
first-approximation theory.
A further argument to absbain from the refinement of Eqs. (3.10) and (3.12) in the form
presented by John in [4, 5], at least in the context of a first-approximation theory, is to be
found in a method of discussion solutions of the equations which is often convenient, in parti-
cular in the linear theory. Eqs. (4.10) and (4.12) are then employed to express (or at least regard-
ed as expressing) the deviatoric parts of 'ep and np in terms of the invariants Ie: and n:. In the
case of a slow variation of the solution over the middle surface, say with a wave length of order
R (and such solutions occur in the membrane theory of shells as well as in theinextensional
bending case) the errors in the deviatoric parts of 'ep and np may be of order 1]04R/h). and
Eh'YJ()4 Rj). respectively. Substitution of this solution into Eqs. (4.9) and (4.11) then leads to
error terms of order 'YJ()3j).2 and Eh2'YJ()3j).2, one factor () more than in the inherent errors in these
equations, but one factor () less than in John's refined version in [4,5]. It is therefore questionab-
le whether John's refinement of his Eqs.(3.10), (3.12) in [4, 5], reducing the error estimates by
a factor ()2 and resulting in an extreme complexity, is even physically meaningful in the
absence of a further refinement of his Eqs. (3.11) and (3.13) beyond the already highly compli-
cated versions in [4, 5].
It may be worthwhile to note, and readers who are familiar with the complexities of nonlinear
shell theory will readily agree, that our equations are relatively simple. They are hardly more
complicated than the (less accurate) simplified equations in ([47], Section 9), where the simpli-
fication had (wittingly) been carried too far to cover all possible situations. We attach parti-
cular significance to the fact that the cubic coefficients in John's strain energy density function
W do not occur explicitly in our equations, although we have not made any assumption with
regard to a linear stress-strain law.
A minor weakness of our Eqs. (4.9)-(4.12) is that their linearized version does not reduce
to the equations of the "best" linear first-approximation shell theory in the sense of Budiansky
and Sanders [44], and for this reason Danielson [49] has appended a note to his paper that his
equations are not intended for use in the linear theory. This weakness is easily rectified, how-
ever, if the modified tensor of changes of curvature elXfJ [43, 47], defined by
'eafJ = ecxfJ + 1/2 (b~Y"fJ + bpY"cx), (4.13)
is introduced into our equations to replace 'Q,xfJ' Eliminating the middle surface strain tensor
again by means of (3.19), and applying again (3.21) where appropriate, we arrive at our final
set of equations with error estimates, which we venture to call the canonical form of intrinsic
equations in the first-approximation nonlinear theory of shells in the absence of surface loads
-An~l~ + b'Je~ - b~d+ 1/2 e'Je~ - 1/2 e~e~ = 0(r;()2j),2), (4.14)
e~lfJ - e~l" - 1/2(1 + v) A(b~n1 - b1n~)lfJ + Ab~n1.fJ
+ (1 + v) A(e~n1.fJ + e1n~I") - vAe1n~,,, = o(r;()4jh)') , (4.15 )
De~l~ - (b'J + e'J) n~ = 0(Eh2r;()2/),2) , (4.16)
n~lfJ + 1/2(1 - v) D(b~e1 - b1e~)lfJ + Db~e1.fJ + D[e~e1 - 1/2 t5~e1e~]lfJ
+ 2A(n~n1)lfJ - 1/2At5~[(1 - v) n1n~ + vn:n1]lfJ = 0(Ehr;()4p.). ( 4.17)
Within John's error estimates for the equations, it is evident from the canonical set (4.14)-
(4.17) that only Eq. (4.10) is affected in its form by our change in the measure for the changes of
curvature. Our canonical Eqs.(4.14)-(4.17) are equally simple (or complicated, if one likes)
as Danielson's equations in our form (4.9)-(4.12), and they reduce to the "best" linear theory
upon linearization. The well-known static-geometric analogy, for example, is easily recognized
in the linear terms of (4.14)-(4.17).
Even if our canonical intrinsic Eqs. (4.14)- (4.17) represent perhaps the simplest possible
formulation 1 of first-approximation nonlinear shell theory, they do not yield all answers to
possible questions. Their actual solution poses a formidable problem, and such a direct solution
seems only possible in those boundary value problems where the boundary conditions may be
expressed in terms of the stress resultants nafJ and the changes of curvature e",fJ (or 'eafJ)' This is
actually the case when dynamic boundary conditions are specified along the entire shell edge.
The reduced form of these boundary conditions is given by Danielson in terms of n cxfJ and 'e",!],
and since John's analysis is inherently incapable of yielding better error estimates for the
edge conditions, suffice it here to refer to [49]. Whenever kinematic boundary conditions are
specified along a part of the edge, we need an evaluation of the displacements from the solution
of our intrinsic equations. A serious problem arises here, in principle, that our equations ensure
only an approximate verification of the compatibility conditions. This means that the metric
tensor 'a"'fJ = arxfJ + 2Y"fJ and the second fundamental tensor Ib afJ = b"fJ + 'eafJ of the deformed
middle surface, as obtained from the solution of our intrinsic equations, do not correspond to a
surface which can actually exist in Euclidean space.
We venture to suggest a possible way out of this difficulty, even if no experience is available
to show that it is practicable. The violation of the equations of Gauss and Mainardi-Codazzi
for the deformed middle surface occurs in error terms in the right-hand members of order
r;()2/),2 and r;()4/h)' (0£. (4.14), (4.15)). We try to reduce these errors to zero by small corrections
LleiJ to e";, of order r;()2/h. It would seem that the three unknown corrections available should be
sufficient to satisfy the equations which were violated initially. If difficulties occur, we also
apply corrections LIn"; to n";, of order Ehr;()2, subject to the restriction Lln~lfJ = 0, and the asso-
ciated corrections to the middle surface strains give additional flexibility in meeting the
requirements of Gauss and Codazzi. It is easily verified that the error terms in (4.16), (4.17)
are not affected by these corrections, and the modified solution eiJ + LIe";, n"; + An"; is an equi-
valent solution which now admits the construction of an associated displacement field of the
middle surface.
1 An alternate epuivalent formulation may be given in terms of the finite rotation and stress function
vectors in [51].
5. Error estimates for solutions

In the introduction we have already stressed the importance of estimabes for the errors of
the solutions of the approximate, two-dimensional shell equations, as compared with the
solutions of the exact three-dimensional equations of elasticity theory. Ideally, one would like
to have pointwise plecise error estimates for all displacements and stresses, but this ideal has
only been achieved in the simple axisymmetric torsion problem for shells of revolution in linear
elasticity theory [2]. More general results have only been obtained in a more global form, as
root mean square or L 2 -norm error estimates, again restricted to the linear case [8-12]. It
would seem that nothing is known about solution errors in the nonlinear theory. The inherent
difficulties of this problem in the nonlinear case will be discussed briefly at the end of this
section.
In view of the wide variety in form in wich the equations of shell theory appear in the
literature, nearly every writer favouring his own equations, it is equally important to have
estimates for the differences between the solutions of the equations associated with different,
presumably equivalent shell theories. This simpler problem of so-called intrinsic error estimates,
involving a comparison of solutions of the same physical dimension, has, of course, a much wider
bearing. Our confidence in the predicbions of the exact classical theory of elasticity for applica-
tions in engineering is based on our firm belief, that these predictions remain (approximately)
valid for actual, (slightly) inhomogeneous, anisotropic and inelastic materials. A sound foun-
dation for this belief, however, is only to be found in error estimates for the solutions of classical
elasticity theory as an approximation to the solutions for actual engineering materials. In fact,
intrinsic error estimates form a necessary complement to classical elasticity theory. Here again,
concrete results have only be-en achieved in the linear theory, and we shall begin with a review
of these [53-55]1.
Let v(x) denote an arbitrary kinematically admissible displacement field of a one-, two-,
or three-dimensional elastic system. The elastic energy stored in the system is in the linear
theory a positive homogeneous quadratic functional P2[V] of the field v. This functional is
positive definite in the sense of Mikhlin [69], if the body is properly supported such that rigid-
body displacements are excluded. The external loads in the linear theory of elasticity have a
potential energy which is a linear functional P 1 [v] of the displacement field. Let u[x] denote
the solution of the elasticity problem. It may be identified with the minimizing displacement
field of the varia.tional problem
(5.1)
We assume the existence of a solution u(x) [0£. e.g. [69]]. Let w(x) denote an arbitrary con-
tinuously differentiable displacement field satisfying homogeneous boundary conditions w = 0
at every point where a kinematically admissible displacement field takes the zero or non-zero
value of a specified displacement vector u o' Introducing the obvious and well-known notation
P 2 [v + w] = P 2 [v] + P l1 [v, w] + P 2 [w] , (5.2)
the variational equation, associated with the minimum problem and satisfied by the solution u,
reads
(5.3)
holding for all admissible fields w.
Some problems are simplified considerably, if the elastic energy functional P 2 [v] is replaced
by an approximation pnv] which is also a positive homogeneous quadratic functional of the
kinematically admissible displacement field v. The approximation of the original functional is
expressed by the inequality
(5.4)
holding for all continuously differentiable displacement fields y(x) without any kinematic
restrictions at the boundary, and for some fixed positive number 8 2 < 1, normally 8 2 <{ 1.
1 The discussion in [53] is restrict.ed tacitly to homogeneous displacement boundary conditions, and we
remove here this unnecessary restriction.
In three-dimensional elasticity theory we have the example of the replacement of the actual,
(slightly) inhomogeneous and anisotropic elastic body by a homogeneous elastic body, in shell
theory, for example, the comparison between the theory of FlUgge and the Sanders-Koiter
theory. The solution u*(x) of the modified variational problem
pnv] + Pl[V] = Min., (5.5)
whose existence is again assumed, is now regarded as an approximation to the actual solution
u(x) of the original problem. The variational equation, satisfied by the approximate solution u*,
Pil[U*, w] + P 1 [w] = 0 (5.6)
holds again for all admissible fields w.
The basic solution error estimate obtained in [53] has a mean square character for the
strains and the stresses, and it is expressed by the inequality
P 2[u - u*] < ge 4p 2[u*]. (5.7)
If the elastic system is properly supported, we have the inequality exprcssing the positive
definite character of the energy functional [69]
/I vl12 < C2P 2[V], (5.8)
where I v /I is the L 2 -norm or root mean square of the kinematically admissible displacement
field, and C some positive constant. In this case we have the solution error estimate
Ilu - u*11 < 3Ce2{P2[u*J}1/2. (5.9)
In some shell problems it appears that a modification of the strain energy functional does
not yet provide the flexibility required in order to obtain the desired simplification of the field
equations. In some cases, however, our purpose may be achieved by a further (small) modifica-
tion of the stress-strain relations in such a way that they are no longer derivable.from a strain
energy function [54]. The linear boundary value problem in these cases is not self-adjoint, and
it cannot be cast into the form of a minimum problem, but it may still be formulated in terms
of a variational equation for the solution u*(x)
Ril[U, w] + P 1 [w] = 0, (5.10)
where Ril is now an asymmetric bilinear functional. The approximate character of the modified
Eq. (5.10) is now expressed by the inequality
IRit[y, z] - Pu[y, z] I < 2e 2{ P 2 [yW/ 2 {P 2[ZW/2, (5.11)
holding for all pairs y, z of continuously differentiable displacement fields, and again for some
positive number e < 1. The resulting error estimates for the approximate solution u*(x) are
identical in form to (5.7) and (5.9) except that the numerical factors 9 and 3 are replaced by 16
and 4 respectively. We refer to [54] for an application to shells whose middle surfaces have a
constant value of the mean curvature H.
The foregoing intrinsic error estimates all apply in the (root) mean square sense. Pointwise
estimates are now also available for the normal displacement component [55], and they are
even uniform for the normal deflection under a normal point load. At present it is still an open
question whether pointwise error estimates would also be possible for the tangential displacement
components or, even more important, for the stresses in the shell. The Prager-Synge hypercircle
method [70] may conceivably be applied to settle this question.
The problem of extrins'ic error 6stimates, which apply to the comparison of the predictions of
two-dimensional linear shell theory with the exact displacement and stress fields of three-dimen-
sionallinear elasticity theory, is more difficult on account of the difference in dimension of the
two problems. It has only been solved rigorously in the mean square sense, and under the
restriction to dynamic boundary conditions along the entire edge of the shell, with a distribution
of edge tractions conforming to the requirements of shell theory [8-12]. The basic tool in the
analysis is essentially a theorem of Prager and Synge [70], and we shall follow the presentation
in [11] in our review. The basic theorem is formulated in terms of the complementary elastic
energy of a three-dimensional elastic body 02[T], essentially the elastic strain energy expressed
in the stress tensor T. Let 0'* denote any statically admissible stress distribution, and 0'0 any
kinematically admissible stress distribution (i.e., a stress distribution obtained, by means of
Hooke's law, and the linear strain-displacement relations, from a kinematically admissible
displacement field). Finally, let 0' denote the actual solution for the stresses of the boundary
value problem of three-dimensional theory. The Prager-Synge theorem is now expressed by the
equation
(5.12)
holding for all kinematically admissible stress fields 0'0 and all statically admissible stress fields
0'*. This equation enables us to estimate the L 2-norm or the root mean square of the error of
an "approximate" solution (0'0 + 0'*)/2 in terms of the L 2 -norm of the stress difference
(0'0 -- 0'*).
The shell theory employed in [11] is the linearized version of the equations and boundary
conditions (2.14)-(2.18). Following in the footsteps of Trefltz [71], it is shown ([11], Section 4)
that a statically admissible stress distribution in the interior of the shell may be obtained from
V
a linear distribution of pseudo-stresses s",p(z) = yja a"'p(z) over the thickness. The restriction
"in the interior of the shell" in the previous sentence may now be dropped, if the edge tractions
in the three-dimensional problem are specified in accordance with the internal stress distribution
just obtained. From the middle surface displacement field we now construct a suitable kinemati-
cally admissible three-dimensional displacement fieid l . In the absence of kinematic boundary
conditions this is always easily possible [11, Section 5]. The stress difference (0''1 - 0'*) may now
be calculated exactly from the shell solution, and we have a precise L 2-norm error estimate
for (0'0 - 0'*)/2 as an approximate solution of the three-dimensional problem. In a thin shell
under sufficiently smooth surface loads, the error is always small, of order 8 2 = h2jL2 + hjR,
where L is again the minimum wave length of the deformation pattern on the middle surface.
A further improvement of this L2 estimate seems to be impossible since the constitutive equa-
tions of first-approximation shell theory are known to involve errors of order 8 2 • It would be
of great interest, however, to obtain similar pointwise error estimates.
We emphasize again that the error estimate of order 8 2 = h2jL2 + hjR applies only in the
case where the edge tractions are specified according to the demands of shell theory. As we
have mentioned before, unfortunately we have rarely accurate information about the distribution
1£ edge loads over the thickness (except in the case of a free edge to be discussed in Section 6).
on the absence of this information it is evidently impossible to obtain precise error estimates for
the solutions of shell theory. The more qualitative argument in [10], however, suggests that
we should expect a larger error, of order 8* = hjL + hjR, in the case of more "irregular" edge
load distributions, provided that the edge tractions do not exceed, in order of magnitude, the
maximum value of the primary stresses in the interior of the shell. This argument is supported
by the results of the next section, and a similar argument was developed in [10] for the case of
kinematic boundary conditions. The somewhat depressing conclusion for most shell problems is,
similar to the earlier conclusion of Gol'denweizer (e.g. [17, 22]), that no better accuracy of the
solutions can be expected than of order 8* = hjL + hjR, even if the equations of first-approxi-
mation shell theory would permit, in principle, an accuracy of order 8 2 = h2jL2 + hjR. This
conclusion also explains why we shall not attempt a very detailed review of more refined interior
shell theories in Section 7.
In the introduction we have already referred to the formidable difficulties encountered in
attempts to extend error estimates for the solutions to nonlinear shell theory. The possible
occurrence of snap loads seems even to require a different concept of solution errors, if an
effective attempt at their discussion is to be made. In fact, at loads in the vicinity of a snap
load for either theory, the exact or the approximate one, the other theory may not have a
solution at all in the neighbourhood of the first one. This situation may occur in the comparison
1 Substantial improvements in [11], in comparison with [10], are due to Danielson who graciously made
his results available already before their independent publication [9].
of two different "equivalent" nonlinear shell theories as well as in the case of a comparison
between shell theory and three-dimensional theory. We are not aware of any effective attempt
to solve the problem.
6. Modified boundary conditions at a free edge

It will be convenient to review first the general form of the classical dynamic boundary
conditions in the linear theory of shells. All quantities are now referred to the undeformed
middle surface with its base and normal vectors, as well as to the undeformed edge curve with
arc length s and its tangent vector t and outward normal vector v in the tangent plane. We
recall our sign convention t = n X v. We introduce the normal curvature u(v), and the geodetic
torsion and curvature u(t) and u(n) of the edge curve by the relations
v,s = u(n)t - u(/)n, t,s = u(v)n - u(n)v, n,s = U(/)V - u(v)t; (6.1)
u(v) = btxpt"'t{J, u(t) = -blX{Jt"''J!{J. (6.2)
The edge loads per unit arc length are decomposed with respect to the unit vectors
N = N(v)v + N(t)t + Qn, M = M(v)v + M(t)t. (6.3)
We employ the modified tensor of changes of curvature (!IX{J' defined by (4.13), and the inter-
nal virtual work per unit area of the middle surface is then specified by nlX{J r5YIX{J mlX{J r5(!IX{J. +
We make use of the expression [4, 7]
'(!tx{J = cf>txl{J - c",{Jw + bilu"I"" (6.4)
where cf>", specifies the projection of the (linear) rotation vector f/J on the tangent plane by the
relation cf>tx = B{J",f/J • a{J. The virtual work of the edge loads per unit length is given by
N . r5u + M . r5f/J, and the boundary conditions of Kirchhoff type follow by the standard
procedure of the calculus of variations in the form
[n{J<X + : b~m{J" - ~ b~m"''']v",v{J = N(v) - u(t)M(v), (6.5)
[ n {Jo< + 2"""
3 b'""m (J" - 1 b{J"m "'><] t ",v{J
2""" -- N (I) + u(v) M (V), (6.6)
m"{JI"v{J + (m {Jv",t{J),8 =
lX
M(v),s - Q, (6.7)
mc<{Jv",v{J = -M(t), (6.8)
in view of (2.19) in agreement with the linearized version of (2.16)-(2.18), and also in agree-
ment with [67] where a slightly different notation is employed. We emphasize again, perhaps
unnecessarily, that these edge conditions are fully exact as they stand. The approximate
character enters the picture when the stress-strain relations are introduced. The need of modi-
fied boundary conditions in shell theory does not arise from a defect in the Kirchhoff-type
boundary conditions as such, but from a failure of the stress-strain relations to hold, even
approximately, near an edge.
In the case of a free edge the right-hand members of (6.5)-(6.8) are all specified to be zeru.
This does not imply, however, that the edge tractions would also be zero. On the contrary,
the non-vanishing internal edge twisting moment m",{Jv(Xt{J implies tangential edge tractions of
the same order of magnitude as occur in the interior shell domain. These tangential edge trac-
tions are only absent in problems where the edge twisting moment happens to be zero. If we
wish to consider the case of a really traction-free edge, if perhaps again only approximately,
we have to investigate at least the effect of a removal of edge tractions of the same order as
the internal stresses. This situation is perhaps the main reason why a number of authors, in
particular Gol'denweizer, have devoted a major effort to analyse, by a, careful examination of
three-dimensional effects, the interaction between a boundary layer in the edge zone and the
interior ,state of stress.
As we shall show, it is relatively easy to achieve the more modest purpose of a modification
of the boundary conditions at a free edge such that at least the interior state of stress in the shell
is obtained with errors of the same order 6 2 = h2 1L2 +

hlR as are inherent in the first-approxi-
mation interior equations. To this end we observe that we have evidently overestimated the
elastic energy in the shell by integrating (2.24) over the entire middle surface, and thus ignoring
that the twisting energy decreases rapidly in an edge zone to zero at the edge itself. As happens
more often, this diagnosis readily suggests a cure. From the theory of torsion for a bar with a
narrow rectangular cross-section (width b and height h ~ b) we borrow the formula for the
torsional rigidity [e.g. 72]
St=! Gbh S(l-B: +0(: e-"b /2h)] , (6.9)
where the numerical constant B is given by!

192 00
B = -5 1: (2k + 1)-5 = 0.630 ... (6.10)
11: k=O
If classical plate theory were entirely applicable to this case, the torsional rigidity would be
given simply by the first term Gbh3/3. Ignoring the exponential term in (6.9), each of the two
zones near a free edge of the plate is responsible for a reduction in elastic energy per unit length
of the plate amounting to BGh40)2/12, where (/) is the specific twist. Bearing in lnind that shell
theory is only meaningful for relatively large wave lengths L of the deformation pattern,
hlL ~ 1, and consequently restricting our attention also to smooth edge curves, we generalize
this result to plates and shells with curved free edges by subtracting from the classical energy
expression a line integral along the free edges
J 112 BGh 4 (elXpv lXt p)2ds. (6.11)
It is now a simple matter, at least in principle, to derive the associated modified boundary
conditions. A difficulty, however, occurs in the circumstance that the variation (JelXpv"'t P con-
tains normal derivatives of the tangential virtual displacement components which can obviously
not be removed by partial integration along the edge curve. This difficulty may be overcome
by replacing the edge twist in (6.11) by
(6.12)
where 'e",p = 'b",p - b",p is the original tensor of changes of curvature, in which revised expres-
sion the terms with a factor u"I"'v'" may be shown tv cancel. Shell theory permits such a replace-
ment even in the calculation of the surface integral of elastic energy, because the accuracy of
first-approximation theory is not affected. This argument holds a fortiori for the line integral
(6.11).
Let rn* denote the twisting couple at the edge, expressed in (/)* by the classical relation of
shell theory rn* = Gh3(/)* 16. T.he line integral of internal virtual work, to be subtracted from
the usual surface integral of internal virtual work (extended over the entire middle surface),
is f Bhrn* (j(/)* ds. We refer to [61] for details of the fairly lengthy algebra, and we present the
result in the form of the modified boundary conditions at a free edge
[n PeX + : b~rnP" - ~ b~rn"'''] v"vp - [u(I,)Bhrn*J,s - u(Ou(n)Bhrn* = 0, (6.13)
[nPe< +~
2 b"m
P
" " - ..!:...bPrn lX
2" "] t'"vP = 0 , (6.14)
rn"Pj"vp + (rn"-Pv",tp),s + [u(n)Bhrn*J,s - u(v)u(t)Bhrn* = 0, (6.15)

rnlXiJv",vp + (Bhrn*),8 = O. (6.16)
It will be observed from a comparison with (6.5)-(6.8) that only one of the four Kirchhoff
conditions remains unchanged, namely (6.6) for the tangential edge loads. We also note that
1 Since we have employed the symbol A for the extensional flexibility (Eh)-l, we deviate from the
custom in the Soviet literature to denote this number by lJ2A.
the additional terms involving the couple Bhm*, after removal to the right-hand member, may
be interpreted as fictitious edge loads, applied to the shell in the conventional theory with
Kirchhoff boundary conditions, namely a force N* = N(.)v + Q*n and a couple M* = M~)t,
both per unit edge length, where the components of the fictitious edge loads are specified by
N Cv ) = [u(v)Bhm*J,8 + u(t)u(n)Bhm* , (6.17)
Q* = [u(n)Bhm*],s - u(v)u(t)Bhm*, (6.18)
M(t) = (Bhm*),8' (6.19)
It is easily verified that the fictitious loads, acting along a smooth closed free edge curve, are
self-equilibrated.
It is not easily possible to compare our results with those of Gol'denweizer in [22] because his
formulation of modified edge conditions is achieved in a completely different, in our view much
more complicated form. On the other hand, in the special case of a fla,t plate with a free edge
our results do indeed reduce to Gol'denweizer's in [17].
The simple form of our modified boundary conditions in terms of the fictitious edge loads is
appealing, and it suggests that they may perhaps be obtained in a simpler way. The clue to this
alternative, more elementary, if perhaps less conclusive approach is to be found in Lamb's
ingenious interpretation of the Kirchhoff conditions [78]. The starting-point is here again the
classical theory of torsion of a bar with a narrow rectangular cross-section (Fig. 1). The exact
stress distribution in the cross-section is statically equivalent to a uniform distribution of tor-
sional moments m* per unit width (over the entire width b) plus a pair of concentrated forces
m* at the ends, and a pair of concentrated couples Bhm* at the ends of opposite sign to the
uniformlydistriubtedcouples (Fig. 1). The forces m* and couples Bhm* are actually due to
z
m~per unit width
- -- "Edge beom" - - _---l

r- - b - - -- ---1
m'
Fig.!. Internal loads in cross· section statically equivalent to exact stress distribution in torsion.
narrow zones of width O(h) near the ends. We shall neglect the width of these zones, and we shall
regard them as fictitious "edge beams" of height h and negligible width, located at the ends of
the cross-section, which carry the shear force m* and the torsional couple Bhm*. The remainder
of the bar will be called the "plate" (of width b), and we shall restrict our attention to the right-
hand edge beam. We take the y-axis in the longitudinal direction, perpendicular to the plane of
Fig. 1, and such that xyz is a right-handed system.
The torsional moments in the plate exert uniformly distributed moments m* per unit
length on the edge beam whose vectors have the negative x-direction (right-hand screw rule I).
Equilibrium of moments for the edge beam is ensured by the presence of the transverse shear
force m* indicated in Fig. 1. Even if the twist, and hence also the torsional moments m* per
unit width in the plate, are functions of the axial coordinate y, the equilibrium of moments
around the x-axis is not disturbed. Equilibrium of forces in the z-direction, however, now requires
a, downward load m~ per unit length on the edge beam, and this load can be exerted only by
the plate. The reaction on the plate is an upward load m~ per unit length, and we have arrived
at Lamb's interpretation of the reduced shear force condition at a free edge of a plate.
So far it has been ignored, however, that equilibrium of moments around the y-axis is also
disturbed if m* depends on y. In order to restore this equilibrium the plate must exert on the
edge beam a moment around the negative y-axis of amount (Bhm*),y' Its reaction on the plate
is an "external" bending moment (Bhm*),y around the positive y-axis, in agreement with
(6.19). It also agrees with (6.17)-(6.19) that the torsional moment Bhrn* in the edge beam does
not result in other "external" edge loads on the plate in the case of a straight edge.
The generalisation of the fictitious edge beam concept to a curved edge of a shell is depicted
in Fig. 2. On the left we have drawn an edge element of the shell with the loads exerted by the
inner side of the edge beam, and on the right an element of the edge beam itself with the internal
loads in one cross-section. In order to avoid a confused sketch we have omitted the reactions of
Internal loads in
"edge beam"
Shell with loads (reactions from
exerted by loads on shell
"edge beam" omitted)
m*.Fl
Fig.2. Internal loads in edge beam and "external" loads exerted on shell edge.
the loads on the shell as well as the internal edge beam loads in the second cross-section. The
outer side of the edge beam is, of course, free from loads in the case of a free edge. The transverse
shear force rn* and the torsional moment Bhrn* are the predominant internal loads in a slightly
curved edge beam, but we have to allow now for additional (small) internal loads, forces F I ,
F 2 , F3 and couples Mv M 2, M3 in Fig. 2. The equations of equilibrium of an edge beam element
are
M 1 ,8 - + M 2 ) + X(t}M3 + F3 = 0,
x(n}(Bhrn* (6.20)
(Bhrn* + M 2 ),s - X(v}M3 + x(n}M l - M~) = 0, (6.21)
M 3,s - X(t}Ml + x(v}(Bhm* + M 2 ) - Fl = 0, (6.22)
}Ill,s - x(n}F 2 + x(t}(rn* + F 3 ) - N~} = 0, (6.23)
F 2,s - x(v}(rn* + F 3) + x(n)F l - N~) = 0, (6.24)
(rn* + F 3),8 - X(t}Fl + X(v}F2 - Q* = O. (6.25)
Conclusions may only be drawn from these equations of equilibrium by taking proper
account of the assumptions that the shell edge is only slightly curved, and that the stress
distribution in the shell has a relatively large minimum wave length. Let e* denote the larger
of the numbers hlL, h 1x(v) I, h !xU} I, h 1x(n} I· Our assumption that the shear force rn* and
the torsional moment Bhm * in the edge beam are the predominant internal loads is now
expressed by the (qualitative) estimates
Fv F 2 , F3 = O(e*m*), 1111> M 3, M3 = O(e*hm*). (6.26)
These estimates are also confirmed by a (qualitative) examination of the stress distribution in
the edge zone of depth O(h) in a cross-section of the shell perpendicular to the edge curve.
Inspection of Eqs. (6.20)-(6.25) shows that we may now ignore Mv M2 and M 3. From Eqs.
(6.20) and (6.22) we have then Fl = x(I.}Bhm* and F3 = x(n) Bhm*. The other equations may
now be written in the form
N~) = X(t)m * + [x(v)Bhm *],8 + x(t)x(n)Bhrn * - x(n)F2, (6.27)

N~) = -x(v}m* + F 2 ,8> (6.28)
Q* = m~ + [x(n)Bhrn*J.s - X(v}x(t)Bhm* + X(v)F2' (6.29)
M~) = (Bhrn*),8' (6.30)
If all terms in the right-hand members are ignored, except the dominating first terms in
(6.27)-(6.29), we arrive at the classical Kirchhoff conditions for a free edge. The correction
expressed by (6.16) or (6.19) is confirmed by (6.30). In order to verify the identification of
(6.17), (6.18) with (6.27), (6.29), and to confirm (6.14) by (6.28), we need an additional argument
to sharpen the estimate (6.26) in the case of F 2 • To this end we observe that the normal force
F 2 in the edge beam arises as a resultant force of normal stresses on a cross-section perpendicular
to the edge due to the presence, and in particular the variation of m* along the edge. The order
of magnitude of these stresses is thus e*m*jh2, and they penetrate only to a depth of order h
into the shell. Their distribution over the thickness is exactly skew-symmetric in the case of
a flat plate, and approximately so, with a relative error of order e*, in the case of a shell. Hence
we have indeed the sharper estimate F2 = 0(e*2m*), and we may omit the terms involving F2
in (6.27), (6.28) and (6.29), thus confirming completely the previous results based on energy
consider a tions.
The reader will have noticed that our reasoning, leading to the modified boundary condi-
tions at a free edge, is based on qualitative error estimates, of a similar character as in [43].
Even if we have full confidence in this physical argument, in particular in the energy approach,
a rigorous proof seems to require more concrete, quantitative error estimates for the solutions
of the shell equations with the modified boundary conditions. We have no doubt that the
Prager-Synge theorem will enable us again to derive such estimates for the L 2-norm of the
errors, but the required analysis will undoubtedly be laborious, and the problem is still under
investigation.
It is a more difficult problem to analyse in detail, and with the same accuracy
e2 = h2jL2 + hjR, the stress distribution in the dege zone of depth O(h). Apparently we cannot
evade here the solution of one of Gol'denweizer's plane strain problems (cf. e.g. [17]). Fortuna-
tely, however, the stresses of most interest may be calculated without solving the plane strain
problem. We refer here to the normal stresses on a cross-section perpendicular to the edge at
the so-called" corner points", the intersections of the faces with the free cdge. At these points,
always in a state of (approximately) plane stress, even if the edge tractions do not vanish, the
stress contribution of the plane strain problem, which is formulated in terms of edge loads to
be removed, may be calculated directly from the normal edge tractions at these points. This
contribution has to be added to the membrane and bending stresses of the solution of the shell
problem under the modified boundary conditions (6.13)-(6.16), and the (uniaxial) warping
stress associated with non-uniform torsion. We refer to van der Heijden's report [62] for
details of the analysis, and we note here only the general formula in which the results may be
presented
(Jtt ( ed ge, -
Z -
± ~2 h) -_ nit 6mtt
h =F h2
± [(1
+ v) 84,(3)
n3
_
v
1116'(5)] m~
n5 h ' (6.31)
1: (2k + 1)-11 is Riemann's zeta function

00
where ~(n) = 2n(2t1 - 1)-1 (we note in passing that

k=o
B = (186jn 5 ) '(5)). The first two terms in (6.31) represent the stresses of classical shell theory
for the modified boundary conditions (6.13) - (6.16). The first term between the br ackets represents
the uni-axial stress arising from constrained warping in non-uniform torsion, and the last term
within the brackets is due to the plane strain problem.
Our remarks about the need of a rigorous proof of the accuracy achieved in the interior
stresses with the aid of the modified boundary conditions also applies, of course, to the edge
stresses. As a partial check, van der Heijden [62] has applied his approximate solution to the
case of an infinite flat plate, weakened by a cylindrical hole of radius a, loaded in uniform torsion
at infinity. Alblas had much earlier obtained the exact three-dimensional solution of this prob-
lem, as well as the first terms, those of order (hja)O and (hja), in the asymptotic expansion for
the corner stress [63], and these exact results fully confirm van der Heijden's approximation.
Gol'denweizer has also proposed modified boundary conditions for simply-supported and
clamped edges of a plate or shell (e.g. [17]). The conditions imposed, however, are unlikely to
be more representative of the actual situation in engineering practice than the classical Kirch-
hoff conditions. We recall our earlier remarks that our lack of knowledge about the actual
edge conditions is the most serious obstacle to an improved accuracy in shell theory.
7. Refined shell theories

Classical shell theory is the basic tool in the design and analysis of shell structures, and the
error estimates for the nonlinear equations in section 4, and for the solutions of the linear
equations in Section 5, serve only to strengthen our belief in classical theory as an adequate
tool, if the basic assumptions of the first-approximation theory are satisfied, viz.
(a) the material is elastic, homogeneous and isotropic;
(b) the shell is of constant thickness h;
(c) the shell is thin, hjR ~ 1;
(d) the loads are smooth, hjL ~ 1;
(e) the strains are small, 1] ~ l.
Some of these assumptions are more essential than others. For example, the theory can
easily be generalized to the case of a slowly varying thickness h [10], and to anisotropic materials
for which the tangent planes of surfaces parallel to the middle surface are planes of elastic
symmetry [38]. The type of inhomogeneity present in sandwich or multi-layered shells requires
a modification of the theory, but appropriate ad-hoc assumptions in the case of sandwich shells
lead to a reasonably simple formulation with an even better accuracy than the first-approxi-
mation theory of homogeneous shells of the same thickness [76, 77, 79]. Inelastic behaviour
of the material poses quite a different problem, outside the scope of our review. Nor shall we
discuss in any detail large elastic deformations of thin shells or membranes, as may occur in
rubber-like materials. The interested reader is referred here to some monographs on nonlinear
elasticity theory [80, 81]. The distinction between two independent strain parameters, one for
the extensional strains and one for the flexural strains, helpful already in the theory of small
strains in the classification and further simplification of shell problems [66, 47], may here
become essential to obtain a theory of manageable complexity. Sensenig's recent report [82]
makes a first attempt in this direction in the case of flat plates.
Suffice it here to make a few comments on refined theories of shells in the more narrow
sense, i.e. for shells which are homogeneous and isotropic, but not-so-thin, say hj R in the range
0.1-0.2. It is a distinct advantage of the asymptotic approach that it is capable, in principle,
of systematic successive refinements. In view of the labour involved already in the derivation
and justification of first-approximation theory by means of the asymptotic method, it seesm
questionable to us, however, whether the effort required by a systematic refinement would be
justified by the results which may conceivably be achieved. It should also be borne in mind
here that the rapid development of numerical techniques in three-dimensional problems, in
particular the finite element technique, may obviate the need of a (complicated) refined shell
theory in the near future.
It seems to us that less systematic, more or less ad-hoc, but simpler methods, based on
physical intuition, may hold more promise for future developments for not-so-thin shells, where
a first refinement beyond classical theory becomes meaningful. In fact, classical theory itself
is an example of this approach, and the most significant contribution, due to the more sophisti-
cated analyses by Gal' denweizer, John and many other writers, is a justification a posteriori,
and occasionally a correction, of previously insecure results of the classical theory, rather than
the independent development of the theory as a whole.
Reissner's theory of flat plates, including the effect of transverse shear deformation [74],
extended to shells by Naghdi [75], is both a good and pioneering example of the type of theory
we have in mind, as well as a typical illustration of the difficulties which have still to be over-
come. Reissner's theory is a definite improvement in the interior domain, away from the edges,
provided that the surface loads are still sufficiently smooth. It.s primary aims, however, to allow
for a better approximation of the actual boundary conditions and for a more accurate description
of the state of stress near edges, have not yet been fully achieved. This somewhat unexpected
shortcoming of Reissner's theory is evident· from the three-dimensional analysis by Alblas for
a plate with a traction-free circular cylindrical hole [63], and its explanation is to be found in
the later more accurate asymptotic analyses of edge effects, referred to previously.
In connection with our advocacy for ad-hoc methods for the refinement of shell theory, which
are already quite popular in the literature, we cannot evade making again some cautionary
remarks. In the development of an approximate theory, based on apparently plausible physical
assumptions, it is essential to bear in mind the physical interpretation of intermediate results
at every stage in the analysis, and to apply appropriate corrections to the initial assumptions
where this is required by the physics of the problem. It. is indeed quite dangerous to derive a
physical theory by a systematic and rigorous mathematical development of initial (approxi-
mate) assumptions unless due account is taken of the physical consequences at every step in
the analysis. To physicists and engineers these remarks will look like the forcing of an open door,
but experience with quite a few papers on shell theory published in the last five years shows
the need of a repetition of such cautionary remarks, made on an earlier occasion [39]. We shall
mention no names, but we shall cite two examples, one of them leading to absurdities, the
other one to serious errors.
In the first example the usual kinematic assumption of Kirchhoff and Love (normals ofthe
undeformed middle surface moving to normals of the deformed middle surface without any
change in length) is adhered to so rigorously that the transverse shear stresses, and hence also
the transverse shear forces, are concluded to be zero. This "theory" has actually found its way
to print in a reputable scientific journal. The second example occurs in a paper, aiming at a
more precise theory of shells by allowance for transverse shear deformations. It is assumed
here that normals to the un deformed middle surface remain straight and retain their original
length, although they are not required to remain normal to the deformed middle surface.
Application of Hooke's law for the stresses parallel to the middle surface is equivalent to the
assumption of plane strain (for these stresses), resulting in an overestimate of the extensional
and flexural rigidities. In spite of the discrepancy with classical plate theory, noted in this
paper, the results are presented as those of a new "more precise" plate bending theory.
We end our cursory comments on refinements in shell theury with the suggestion that the
development of appropriate simplifying physical assumptions may perhaps benefit from exact
three-dimensional solutions of relatively simple problems. For example, the three-dimensional
analysis by Alblas, referred to previously, supports our approximate discussion of stresses near
a free edge. In the theory of flat plates Lur'e's three-dimensional analysis for sinusoidal loading
in both directions on the middle surface of an infinite plate [83] may be exploited to investigate
the detailed stress distribution in a rectangular plate with some sort of simply-supported edges
(zero deflection and vanishing normal stresses on the edge). ",Ve have already ascertained, in
unpublished notes by one of the authors and G. Hommel, that Lur'e's theory confirms the vali-
dity of Reissner's theory in the interior domain. Likewise, Leibenson's effective use of three-
dimensional theory in his solution of the buckling problem for a spherical shell under external
pressure [56] suggests spherical shells as a second convenient test case. Finally, the cylindrical
shell offers similar possibilities (cf. [59]).
8. Concluding remarks
We conclude our review with mention of a number of open problems is shell theory which,
in our opinion, are in particular need of further research. Most fundamental among these is,
perhaps, the question of error estimates for the solutions. It would be of great value, also for the
applications, to have (accurate) pointwise error estimates, both for displacements and stresses.
As mentioned before, the method employed by Ho and Knowles [2] may possibly be extended
to more general loading conditions in the linear theory for shells of revolution. At present we
have no clear idea, however, of the road to be followed for shells of arbitrary shape, not even in
the linear theory, and we recall our earlier remark that the problem of solution error estimates
in the case of the nonlinear theory is entirely open.
In the context of the nonlinear theory it seems desirable to compare the predictions of
Danielson's Eqs.(4.9)-(4.12) and of our canonical Eqs.(4.14)-(4.17) with the results of
simpler equations for special configurations, in particular Reissner's theory for axisymmetric
deformations of shells of revolution [52]. In the case of rubberlike materials it will also be
worthwhile to try and obtain reasonably simple and accurate stress-strain relations for finite
stretching (and small bending).
As we have already emphasized repeatedly, the most critical uncertainties in shell theory
arise from our ignorance of the exact nature of the actual boundary conditions, except in the
case of a free edge. This problem cannot be solved by shell theory alone since it refers to the
complete structure of which the shell is only an essential part. It is a regrettable consequence
of this situation that the predictions of shell theory have no better accuracy thane* = hjL hR +
for structural applications, and refinements of shell theory as such are of no avail. If a solution
of this problem is to be achieved at all, it will have to consider in detail the supporting
structure as well as the shell itself. It would seem that a penetrating analysis of this interaction
problem between the 8hell and its 8upporting 8tructure is a more rewarding theme for future
research than a detailed refinement of shell theory.
References
1. Aron, H.: J. Reine u. Angew. Math. 78 (1874) 17. Gol'denweizer, A. L.: Proc. 11th Int. Congr.
136-173. Appl. Mech. Mlinchen 1964. Berlin/Heidelberg/
2. Ho, C. L., Knowles, J. K.: Z. A. M. P. 21 (1970) New York: Springer 1966, 306-311.
352-377. 18. Kolos, A. V.: Prikl. Mat. Mekh. 28 (1964) 582-
3. John, F.: Comm. Pure and Appl. Math. 18 589. English translation 718-726.
(1965) 235-267. 19. Kolos, A. V.: Prikl. Mat. Mekh. 29 (1965) 771-
4. John, F.: Proc. 2nd IUTAM Symp. Theory of 781. English translation 914-925.
Thin Shells, Copenhagen 1967. Berlin/Heidel- 20. Gol'denweizer, A. L., Kolos, A. V.: Prikl. Mat.
berg/New York: Springer 1969,1-14. Mekh. 29 (1965) 141-155. English translation
5. John, F.: Comm. Pure and Appl. Math. 24 151-166.
(1971) 583-615. 21. Gol'denweizer, A. L.: Prikl. Mat. Mekh. 32
6. Morgenstern, D.: Arch. Rat. Mech. and Anal. 4 (1968) 684-695. English translation 704-718.
(1959) 145-152. 22. Gol'denweizer, A. L.: Prikl. Mat. Mekh. 33
7. Morgenstern, D., Szabo, 1.: Vorlesungen liber . (1969) 996-1028. English translation 971-
Theoretische Mechanik, § 9.5, Berlin/Giittingen/ 1001.
Heidelberg: Springer 1961.
23. Gol'denweizer, A. L.: Proc. 2nd IUTAM Symp.
8. Nordgren, R. P.: Quart. Appl. Math. 28 (1971)
Theory of Thin Shells, Copenhagen 1967. Berlin/
587-595.
Heidelberg/New York: Springer 1969,31-38.
9. Danielson, D. A.: Proc. Kon. Ned. Ak. Wet.
24. Gol'denweizer, A. 1,.: On two-dimensional equa-
B 74 (1971) 294-300.
tions of the general linear theory of thin elastic
10. Koiter, W. T.: Proc. Kon. Ned. Ak. Wet B 73
shells. Problems of Hydrodynamics and Con-
(1970) 169-195.
tinuum Mechanics (L. 1. Sedov anniversary
11. Koiter, W. T.: Proc. Int. Congr. Math. Nice
volume), Soc. Ind. Appl. Math., Philadelphia
1970. Paris: Gauthier-Villars. vol. 3 (1971)
123-130. (1969) 334-352.
12. Simmonds, J. G.: Quart. Appl. Math. 29 (1971) 25. Rutten, H. S.: Proc. 2nd IUTAM Symp. on the
439-447. Theory of Thin Shells, Copenhagen 1967. Berlin/
13. Sensenig, C. B.: Int. J. Engng. Sci. 6 (1968) Heidelberg/New York: Springer 1969,115-134.
435-464. 26. Rutten, H. S.: Asymptotic approximation in the
14. Green, A. E.: Proc. Roy. Soc. London A 269 three-dimensional theory of thin and thick
(1963) 481-491. elastic shells. Thesis Delft 1971, published by
15. Gol'denweizer, A. L.: Prikl. Mat. Mekh. 26 Netherlandse Boekdruk Industrie N.V., 's Her-
(1962) 668-686. English translation 1000- togenbosch.
1025. 27. Cosserat, E., Cosserat, F.: Theorie des corps
16. Gol'denweizer, A. L.: Prikl. Mat. Mekh. 27 deformables. Paris: Herrmanr, 1909.
(1963) 593-608. English translation 903-924. 28. Naghdi, P. M.: The theory of shells and plates.
Fliigge's Handbuch der Physik, 2nd edition, 51. Simmonds, J. G., Danielson, D. A.: J. Appl.
vol. VI-2. Berlin/Heidelberg/New York: Sprin- Mech. 39 (1972) 1085-1090.
ger 1972. 52. Reissner, E.: Proc. 3rd Symp. Appl. Math.,
29. Niordson, F. I.: Int. J. Solids and Structures 7 Ann Arbor 1949. New York: McGraw-Hill 1950,
(1971) 1573-1579. 27-52.
30. Fliigge, W.: Statik und Dynamik del' Schalen. 53. Koiter, W. T.: ZAMP 21 (1970) 534-538.
Berlin: Springer 1934. 54. Simmonds, J. G.: ZAMP 22 (1971) 339-345.
31. Fliigge, W.: Stresses in shells. Berlin/Gottingen/ 55. Simmonds, J. G.: To appear in ZAMP 23 (1972)-
Heidelberg: Springer 1960. 56. Leibenson, L. S.: Matematishkii Svornik 40
32. Wlassow, W. S.: Allgemeine Schalentheorie und (1933) 429-442.
ihre Anwendung in der Technik. Deutsche Aus- 57. Neut, A. van der: De elastische stabiliteit van
gabe der russischen Auflage (1949) unter Redak- den dunwandigen bol. Thesis Delft, published
tion von A. Kromm. Akademie-Verlag, Berlin by H. J. Paris, Amsterdam 1932.
(1958). 58. Koiter, W. T.: Proc. Kon. Ned. Ak. Wet. B 72
33. Novozhilov, V. V.: The theory of thin shells. (1969) 40-123.
Translated from the Russsian edition (1951) by 59. Newman, J. B.: To appear in J. Appl. Mech. 39
P. G. Lowe. Groningen: Noordhoff 1959. (1972).
34. Gol'denweizer, A. L.: Theory of elastic thin 60. Friedrichs, K. 0.: Proc. 3rd Symp. Appl. Math.,
shells. Translation from the Russian edition Ann Arbor 1949. New York: McGraw-Hill 1950,
(1953) ed. G. Herrmann, Oxford: Pergamon 117-124.
Press (1961). 61. Koiter, W. T., Heijden, A. van der: Modified
35. Girkmann, K.: Fliichentragwerke, 4.Aufl. Ox- boundary conditions at a free edge of a thin
ford: Springer 1956. plate or shell. To be published in Proc. Kon. Ned.
36. Mushtari, Kh. M., Galimov, K. Z.: Nonlinear Ak. Wet.
theory of thin elastic shells. Translated from the 62. Heijden, A. van der: Benaderingstheorie voor
Russian edition (1957) by J. Morgenstern and het bepalen van spanningen langs de vrije schaal-
J. J. Schorr-Kon. NASA-TT-F62 (1961). rand. Report 456, Laboratory of Engineering
37. Kraus, H.: Thin elastic shells. New York: Wiley Mechanics, Dept. of Mechanical Engineering,
1967. University of Technology, Delft 1971.
38. Naghdi, P. M.: Foundations of elastic shell 63. Alblas, J. B.: Theorie van de driedimensionale
theory. Progress in Solid Mechanics, ed. I. N. spanningstoestand in een doorboorde plaat.
Sneddon and R. Hill. Amsterdam: North-Hel· Thesis Delft, published by H. J. Paris, Amster-
land 1963, 1-90. dam 1957.
39. Koiter, W. T.: Proc. 2nd IUTAM Symp. on the 64. Lur'e, A. I.: Prikl. Mat. Mekh. 4, Nr. 2 (1940).
Theory of Thin Shells, Copenhagen 1967. Berlin/ 65. Gol'denweizer, A. L.: Prikl. Mat. Mekh. 4, Nr. 2
Heidelberg/New York: Springer 1969,93-105. (1940).
40. Love, A. E. H.: Mathematical theory of elasti- 66. Koiter, W. T.: Proc. Kon. Ned. Ak. Wet. B 64
city, 4th edition. Cambridge University Press (1961) 612-619.
1927. 67. Koiter, W. T.: Proc. Kon. Ned. Ak. Wet. B 67
41. Reissner, E.: J. Math. and Phys. 42 (1963) (1964) 117 -126.
263-277. 68. Koiter, W. T.: ZAMP 20 (1969) 642-653.
42. Reissner, E.: Proc. 11th Int. Congr. Appl. 69. Mikhlin, S. G.: The problem of the minimum
Mech., Miinchen 1964. Berlin/Heidelberg/New of a quadratic functional. Translated from the
York: Springer 1966, 20-30. Russian edition (1952) by A. Feinstein. San
43. Koiter, W. T.: Proc. IUTAM Symp. on the Francisco: Holden .Day 1965.
Theory of Thin Elastic Shells, Delft 1959. Am- 70. Prager, W., Synge, J. L.: Quart. Appl. Math. 5
sterdam: North-Holland 1960,12-33. (1947) 241-269. Cf. also J. L. Synge: The
44. Budiansky, B., Sanders, J. L.: On the "best" hypercircle in mathematical physics. Cambridge
first-order linear shell theory. Progress in App'- University Press 1957.
lied Mechanics (Prager Anniversary Volume), 71. Trefftz, E.: Z. angew. Math. und Mech. 15
New York: Macmillan 1963, 129-140. (1935) 101-108.
45. Leonard, R. W.: Nonlinear first approximation 72. Sokolnikoff, I. S. : Mathematical theory of elasti-
thin shell and membrane theory. Thesis Virginia city, 2nd edition, New York: McGraw Hill 1956.
Polytechnic Institute 1961. 73. Goodier, J. N.: Trans. Roy. Soc. Canada 32
46. Sanders, J. L.: Quart. Appl. Math. 21 (1963) (1938) 65-88.
21-36. 74. Reissner, E.: J. Appl. Mech.12 (1945) 69-77.
47. Koiter, W. T.: Proc. Kon. Ned. Ak. Wet. B 69 75. Naghdi, P. M.: Quart. Appl. Math. 14 (1957)
(1966) 1-54. 369-380.
48. Budiansky, B.: J. Appl. Mech. 35 (1968) 393- 76. Reissner, E.: Small bending and stretching of
401. sandwich.type shells. NACA Report 975 (1950).
49. Danielson, D. A.: Int. J. Engng. Sci. 8 (1970) 77. Plantema, F. J.: Sandwich construction. New
251-259. York: Wiley 1966.
50. Simmonds, J. G., Danielson, D. A.: Proc. Kon. 78. Lamb, H.: Proc. Lond. Math. Soc. 21 (1891)
Ned. Ak. Wet. B 73 (1970) 460-478. 119.
79. Grigoliouk, E. I., Chulkov, P. P.: On the theory 83. Lur'e, A. I.: Raumliche Probleme der Elastizi-
of multilayer shells. Contributions to the Theory tatstheorie. Deutsche Augsabe der russischen
of Aircraft Structures, Delft University Press Auflage (1955) von G. Landgraf. Berlin: Aka-
1972, 171-183. demie-Verlag 1963.
80. Green, A. E., Adkins, J. E.: Large elastic defor- 84. Chernykh, K. F.: Linear theory of shells, Parts I
mations. Oxford: Clarendon Press 1970. and II, NASA TT-F 11,562. Translated from
81. Stoker,J.J.: Nonlinear elasticity. New York: the Russian edition, Leningrad Univ. Publ.
Gordon and Breach 1968. House, Part I (1962), Part II (1964).
82. Sensenig, C. B.: Plate estimates involving two 85. Cicala, P.: Systematic approximation approach
strain parameters. Report IMM388. Courant to linear shell theory. Torino: Levrotto and
Institute of Mathematical Sciences, New York Bella (1965).
University 1971. 86. Danielson, D. A., Simmonds, J. G.: Int. J.
Engng. Sci. 7 (1969) 459-468.

1973, Koiter, Foundations of Shell Theory

Uploaded by

Copyright:

Available Formats

1973, Koiter, Foundations of Shell Theory

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

1973, Koiter, Foundations of Shell Theory

Uploaded by

Copyright:

Available Formats

Foundations of shell theory

E. Becker et al. (eds.), Theoretical and Applied Mechanics

2. Classical nonlinear shell theory

e",peAI'['ePA!1' - "a"V(b"A + 'e",,) YVPI'] = 0, (2.6)

n"/3 = V-'a/a['n /3 + 21 ("b~'rnx/3 + IIb~'mxc<)],

+ IIb~mxfJl/3 + "b"'''(2Y''illfJ - YilfJl,,) m ilfJ + V'a/a'p'" = 0, (2.21)

V = ~ hEafJ A" lYafJYAll + 112 h2 '(!"fJ'(!A" 1· (2.24)

The associated constitutive equations are by (2.10)

3. Reappraisal of John's results

and the length parameter

tii["l ... ,,)X"', z) = o(Er;A -n), (3.9)

- ~ -r~I~ + SaPIXA!, (ba!"ePA + -} 'eIX!"ePA) = 0('Y)(j2j).2) , (3.10)

SIXPSA!"epAI!' = 0('Y)(j2jh)') , (3.11)

-rPIXlp = 0(E'Yj(j2jJ.) , (3.13)

5. Error estimates for solutions

6. Modified boundary conditions at a free edge

[n{J<X + : b~m{J" - ~ b~m"''']v",v{J = N(v) - u(t)M(v), (6.5)

is obtained with errors of the same order 6 2 = h2 1L2 +

St=! Gbh S(l-B: +0(: e-"b /2h)] , (6.9)

where the numerical constant B is given by!

J 112 BGh 4 (elXpv lXt p)2ds. (6.11)

[n PeX + : b~rnP" - ~ b~rn"'''] v"vp - [u(I,)Bhrn*J,s - u(Ou(n)Bhrn* = 0, (6.13)

rn"Pj"vp + (rn"-Pv",tp),s + [u(n)Bhrn*J,s - u(v)u(t)Bhrn* = 0, (6.15)

- -- "Edge beom" - - _---l

N~) = X(t)m * + [x(v)Bhm *],8 + x(t)x(n)Bhrn * - x(n)F2, (6.27)

1: (2k + 1)-11 is Riemann's zeta function

where ~(n) = 2n(2t1 - 1)-1 (we note in passing that

7. Refined shell theories

You might also like

[n PeX + : b~rnP" - ~ b~rn"'''] v"vp - [u(I,)BhrnJ,s - u(Ou(n)Bhrn = 0, (6.13)

rn"Pj"vp + (rn"-Pv",tp),s + [u(n)BhrnJ,s - u(v)u(t)Bhrn = 0, (6.15)

N~) = X(t)m * + [x(v)Bhm ],8 + x(t)x(n)Bhrn - x(n)F2, (6.27)