COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING 73 (1989) 53-92

NORTH-HOLLAND

ON A STRESS RESULTANT GEOMETRICALLY EXACT SHELL MODEL.

PART II: THE LINEAR THEORY; COMPUTATIONAL ASPECTS

J.C. SIMO, D.D. FOX and M.S. RIFAI

Division of Applied Mechanics, Department of Mechanical Engineering, Stanford University,
Stanford, CA, U.S.A.

Received 2 October 1987

Revised manuscript received 7 June 1988

Computational aspects of a linear stress resultant (classical) shell theory, obtained by systematic
linearization of the geometrically exact nonlinear theory, considered in Part I of this work, are
examined in detail. In particular, finite element interpolations for the reference director field and the
linearized rotation field are constructed such that the underlying geometric structure of the continuum
theory is preserved exactly by the discrete approximation. A discrete canonical, singularity-free
mapping between the five and the six degree of freedom formulation is constructed by exploiting the
geometric connection between the orthogonal group (SO(3)) and the unit sphere ($2). The proposed-
numerical treatment of the membrane and bending fields, based o n a mixed Hellinger-Reissner
formulation, provides excellent results for the 4-node bilinear isoparametric element. As an example,
convergent results are obtained for rather coarse meshes in fairly demanding, singularity-dominated,
problems such as the classical rhombic plate test. The proposed theory and finite element implementa-
tion are evaluated through an extensive set of benchmark problems. The results obtained with the
present approach exactly match previous solutions obtained with state-of-the-art implementations
based on the so-called degenerated solid approach.

1. Introduction

In this paper, we address in detail numerical aspects involved in the finite element
implementation of a classical linear stress-resultant shell theory. The proposed formulation is
obtained by a systematic linearization about the refence state of the geometrically exact fully
nonlinear theory developed in Part I of this work [32]. ~ The resulting theory is termed
"classical" in the following sense:
(i) The momentum equations, formulated directly in terms of resultants, are ,exact within
the framework of a directed media or Cosserat continuum. These equations involve stress
resultants whose definitions agree with those of classical shell theory, but are at variance with
definitions often employed in the degenerated solid approach which dominates the current
computational literature. An example of this variance is the definition of the membrane stress
resultants. It is a well-known classical result, see [25, 26], that these stress resultants can never
be symmetric due to the initial curvature of the shell: It is also a classical result, which carries

1The term ~geometrically exact" reflects the fact that no additional kinematic assumptions are made beyond the
one-director a~sumption. In particular, approAi~ations of the type sin O - O - O3/6 are entirely avoided.

0045-7825/89/$3.50 © 1989, Elsevier Science Publishers B.C. (North-Holland)

54 1. C. Simo et al., Stress resultant geometrically exact shell model. Part H

over to shell models allowing for shear deformation, that only a certain combination of the
resultant membrane stresses and the resultant bending stress couples, known as the effective
stress resultants appears in the internal stress power. Remarkably, this is the only relevant
object in a weak formulation of the field equations.
(ii) The constitutive equations are formulated directly in terms of resultant stress and stress
couples. Classically, the constitutive coefficients entering in these equations can be determined
a priori by several standard procedures; i.e., asymptotic expansion of the three-dimensional
theory, or invariance arguments along with matching of exact simple three-dimensional
solutions. An important fact, demonstrated numerically in the following sections, is that these
simple constitutive models, which do not involve the costly pre-integration through the
thickness (a main feature of the degenerated solid approach), reproduce the exact solutions of
standard benchmark problems. We also note that three-dimensional constitutive models can
be pre-integrated numerically, if so desired, to obtain stress resultant models that fit within the
classical approach.
(iii) An important difference between the present formulation and classical shell theories,
however, concerns the parametrization and format in which the field equations are cast. From
the summary of the linear theory in Table 1, it is apparent that objects such as Christoffel
symbols, the second fundamental form, and covariant derivatives, which are not readily
accessible in a computational framework, never appear explicitly. A further difference
concerns the central role played by the underlying geometric structure of the theory. As an
example, by exploiting the relation between the orthogonal group and the unit sphere, we
construct a unique orthogonal transformation that maps the inertial frame onto an orthonor-
real direct ~ • frame without "drill" about the director.
The objective of this paper is to demonstrate that classical shell theory, rephrased in the
fo~'mat of Table 1, is amenable to a straightforward finite element implementation. In
particular, because of our parametrization of the field equations, the implementation is closely
related to numerical treatments of the classical Reissner-Mindlin plate model. From a
computational standpoint, novel aspects present in our approach are the following:
(iv) Construction of the interpolations for the reference director and the linearized director
field that exactly preserve the kinematic structure of the theory; i.e., the discretized reference
director has unit length, and the incremental rotation field is orthogonal to the director
throughout the element.
(v) Transformation from six to five degrees of freedom formulation by means of a unique
orthogonal transformation that maps the inertial frame onto an orthonormal director frame
without drill. (Geometrically this is accomplished by reducing SO(3) by $1.)
Our numerical implementation is carried out with reference to the four-node bilinear
isoparametric element. Within the context of linear Mindlin-Reissner plate theory, several
interpolation schemes have been proposed to ameliorate the well-known shear locking effect
which typically arises as the thickness of the plate goes to zero. In the present context, we
accommodate a scheme that essentially goes back: to MacNeal [20], subsequently extended and
reformulated in Hughes and Tezduyar [16], and MacNeal [21], and revisited in Bathe and
Dvorkin [5]. The only available mathematical analysis of this scheme, to some extent
incomplete, is due to Bathe and Brezzi [34]. To our knowledge, a complete mathematical
analysis comparable to the recent work of Brezzi and Fortin [9] and Arnold and Falk [1] for a
triangular element, for which these authors prove rigorous uniform convergence estimates for
 J.C. Simo et al., Stress resultant geometrically exact shell model. Pan H 55

Table 1
T h e linear theory. Summary
• Resultant stress couple vectors:

f'llL] lq I} fJl~ ll I
L~12J ~2 , /l~ LII~ 12J
• O r t h o g o n a l transformation: [such that A°(E × t °) = (E × t°)]

t oo t ~"
A°£ and to " "
A°£ . '

A ° = [t, I2 t] - ( E . / ° ) t + + [E × t °] + ] + E- t ° (E x t ° ) ® (£ × t °)

• Incremental displacement and rotation: [At. t o = &T- E = 0]

u= u2 ,
U3
At=,~°AT, • AT= AT2j and [
"Art
° A°2]
j.°:=tt, t2]= A°t A°2/
.A °, A°2.1 '

• Matrix differential operators:

ot O
~0.i +++,
ot 0
Bin----- ~02 2
o' 8 _ o' 0

F,o,+] B'b ~ ex3 A-z ox 2

B,., ~. [ to' ~J,-
t,0~ a o' $

ot 0 -o
Bbm 02 Bbb = 5o,2 0~2 A3x2
o' 0 ~:o' 0 I o' 0 _ o~ $
t,,"~ + t,2~'~J3x3
• Bilinear form B: T.oT x T.oT--*R.

S -- f.~ {[Bml)]tt~ "I- [BsmU "1"~sb ST]t# -I- [~bm v "I-Bbb T]'n~ } dlm°

any thickness, is still lacking. The empaasis in this work, however, is not placed on the
development of (yet) another shell element, but rather on the implementation and numerical
assessment of a mechanically sound methodology that finds its roots in classical shell theory.
Nevertheless, some interesting conclusions from the purely computational standpoint emanate
from the present work. In particular:
56 J.C. Simo et al., Stress resultant geometrically exact shell model. Part H

(vi) The proposed mixed interpolation for the bending field, which emanates from a
Hellinger-Reissner variational formulation, leads to an element which possesses proper rank,
and yields excellent results in standard benchmark problems. For the classical rhombic plate
test (a singularity-dominated problem) for instance, the performance of the element is clearly
superior to existing four-node and even nine-node elements; including T1 [16], the equivalent
element of Bathe and Dvorkin [5], and the four- and nine-node ANS formulations of Stanley
[38] and Parks and Stanley [27].
(vii) The interpolation for the membrane field, inspired in the treatment of Pian and
Sumihara [29] for the plane stress problem, also appears to yield optimal results (for bilinear
interpolations) in numerical experiments involving membrane dominated problems; such as
Cook's membrane test, and the Scordellis-Lo [31] test.
Although the finite element interpolation satisfies exactly all patch test requirments, (this is
at variance with some recently proposed assumed strain interpolations [18] 2) a satisfactory
mathematical analysis is still lacking.

2. Summary of the governing equations

In this section, we summarize the basic field equations governing the proposed linear shell
model. These equations emanate from the nonlinear theory examined in detail in Part I of the
work. We begin with a review of the nonlinear theory.

2.1. The nonlinear theory. Summary

We limit our presentation on the nonlinear theory to a brief overview of the basic
kinematics, field equations, weak form of the balance equations, and constitutive equations.
For further details, we refer to [32, 33].

2.1. I. The basic kinematics

Let S 2'= { t E R 31 [Itl[ = 1} be the unit sphere and T,S 2 := { to • R31t. to = 0} be the tangent
space at t E S 2. The configuration space of the shell, denoted by ~, is defined

¢'~ :~"~{ 0 "----(~, t): ~ C R 2'''> R 3 X S2}. (2.1)

Here, z~ C R 2 is compact, with boundary 0M. We denote by a , M C az~ and Ot,~ (= ~,~ the
regions of 0 d with prescribed displacements and rotations, respectively.
The basic kinematic assumption can be formulated as follows. Let Y denote any configura-
tion of the shell. It is assumed that

b~:={xER3[x=~o+ ~t where (~o,t)E ~ and ~ : E [ h - , h + ] } , (2.2)

where h = h ÷ - h- is the thickness of the shell, ~o: ~ C R 2--> R 3 defines the mid-surface of the
shell, and t( ~:~, ~:2) E S 2, for ( ~ l, ~2) E ~t.

2Our nun:ericai experiments appear to indicate that this nine-node element does not pass constant membrane
and bending patch tests.
 J.C. Simo et ai., Stress resultant geometrically exact shell model. Part H 57

2.1.2. Momentum balance equations

The (exact) local momentum equations are expressed in stress resultants and stress couples
as

1
_ (.l,n).~ - [ + m = ipt', (2.3)
/

where n ~= n#%p,a + q~t and n3~ = m ~ # ~ -~o~ + r~ 3at are the resultant stress and the director
c_oul31e resultants, ti and m are the applied loads, f = I1,,., × ~0.211 is the surface Jacobian, and
I = At + l, where I is the across-the-thickness stress resultant.
In addition to balance equations (2.3), there is a further constraint that balance of angular
momentum places on the admissible form of the constitutive equations. This constitutive
restriction (emanating from the symmetry condition ¢r = o't) is given by

~p,~ x n ~ + t,~ x ~ ~ + t x l = O . (2.4)

By systematically exploiting this restriction (see Part I of this work for details), we obtain
effective membrane and effective shear resultant forces, denoted by t~ = r7~#¢,~ ®~o# and
= t ~ o , , , respectively, and defined by the relations

II$ )

"= - A~,m , (2.5)

l = At + ( ~ + A~,r~3~')~0o,

where t,o = Aa# ~o,0+ A3ot.Note that K = K t,but n ~# # n #°.

2.1.3. Weak form o f the momentum balance equations

Recall that the tangent space of kinematicaily admissible variations is given by

T@~={(8~o, St) =: 8 0 ",~---> R3 x T, S2]8~o]o,~=O and 8tla,.~ = 0 } , (2.6)

where the condition 8t E TtS 2 implies t. 8t = 0.

By taking the dot product of the momentum equations with an arbitrary variation
8 ~ E T . ~ and integrating by parts, one obtains the weak form of the momentum balance
equations as

Gdyn(CJ~,SO)"= G(-~:~,8~[~)+ f~t[/~ "8~o + ioi:.8t] d#. (2.7)

Here, d/~ =)d~ :~d~:2 is the current surface measure and G(O, 80) is the staticweak form
defined as

G(O, 80) := fa [n~ •8~,.~+ t~" •8t,a +/.St] d/~ - Gex,(8O), (2.8)
58 J . C . Simo et al., Stress resultant geometrically exact shell mo d e l . Part H

where Gext(~i@): T O c~__>R is the virtual work of the external loading given by

G~t(~@)= fa[~.~p+m.~t]dl~ + fo ~ . 8 ~ p f d s + fa m.Btjds. (2.9)

hal m~

In this expression, n and m are the prescribed resultant force and the prescribed director
couple on the boundary 0,M and 0mM, respectively, where
o,,s no, s =O and OmM n O,M = O. (2.10)
Introducing the strain measures 3

e~a • = ½(~p,~ . ¢,~ - ~pO . ¢ ,0, ) , (2.11a)

3%:= (¢,,~" t - ~p,o. to), (2.11b)
p,~, := (¢.~ " t a - ¢o.~ . toa) , (2.11c)

it can be shown that only the symmetric effective membrane stress resultant appears in the
static weak form (2.8) which now takes the form

G(@, 8q~)= fa [~"~e,~a + n~'~aSt.-:t~ + q'~83,,~]d~t~--. G~,,t(8@ ) . (2.12)

Note that the components in expression (212) of the static weak form are given relative to
the surface convected (Gavssian) frame {¢,~, ¢.2, t}. An alternative expression involvi!ag
components relative to a local cartesian system is considered in Section 3.

2.1.4. Constitutive equations

The simplest example of a properly invariant elastic isotropic constitutive model for the
stress resultants in terms of the conjugate strain measures (2.11) is given by

~= Eh H~.~~
1- v2 e~ , ~ = K G Sha°'*ytj ,
(2.13)
rn (~,~ ) = Eh 3 Mal3v~ _
12(1- v 2) "-0 t't-,8),

where E is Young's modulus, v is Poisson's ratio, K is the shear reduction coefficient, G s is the
0 0 0
shear modulus, a,~ := ¢ . . ¢.~ is the reference surface metric, and
Ho ~ = { va°~"a °'8 + ½(1 - v)(a°°'a °~8 + a°°~a°"')}. (2.14)

R E M A R K 2.1. The constitutive equations for the resultant dilector couple is given in terms
of the symmetric part of n~~. Without loss of generality, it is assumed here and in the
following developments that the skew-symmetric part of n~"~ is zero, i.e.

aln Part I, the shear strain (2.11b) was denoted 8,~.

 J.C. Simo et al., Stress resultant geometrically exact shell model. Part H 59

t n [,~#I = 0 .
(2.15)

Asymptotic expansions of the three-dimensional theory reveal that rh[O#] = O(h3/RL2), where
h is the thickness, R is the radius of curvature of the shell, and L is a suitable in-plane length
scale of order O(1). Hence, the skew-symmetric part r~ [~#] can be neglected within the order
of approximation of the linear theory. [:]

2.2. Linearization. The model problem

We now proceed to linearize the nonlinear theory about the reference state (~0° t o) E ~. To
this end, let (u, At): ~--~ R 3 x T,0S2 be the displacement field for the reference state. Making
use of the exponential map in S2, consider the curve e ~(~,o, t o) E ~ defined as

e ~ ~po = cpo + e u ,
(2.16)
sinle Atl
o = cosle A t l t ° +
E~-> L6 EAt= t o ~" E A t d - O(E2).
leAtl

From the discussion in Part I of this work, we recall that At E T,oS2, i.e. t o. At = 0, and that
there exists a unique A ° satisfying A ° - ' = [A°] t and A°O = • with O ' E =0, such that
t o = A°E. We thus define At E T,oS2 by the relation At = A ° AT, where AT. E = 0.
If one selects E - E 3 , then AT is given in components relative to the standard basis as

AT = A TIEI + A T2E2 . (2.17)

Therefore, by expressing the incremental spatial rotation yector as A t - A ° AT, where AT is

given by (2.17), the constraint condition A t . t ° = 0 is identically satisfied since A T . E =
A t . t o= O. This observation is crucial to the numerical implementation considered below.
Use of construction (2.16) yields the following expression for the strain measures defined by
(2.11):
e.# • = ½ ~(~oo. u ,0 + u ,.., 0.#) + o ( : ),
~ -= e(~,,° • At + u,~" t°) + O(~), (2.18)

Pa#'~--"E(~?a"At,# "~"~[.a"I[?
8) + O(2).
By making use of the directional derivative formula and setting 8(,)=(d/de)l,~=o(O), we
obtain the linearized strain measures about the reference state as
• 0
~,~ = ;:,~# = ½(~.° . u , + u ,. . ~, ~ ) ,

~ . = ~/~ = ( ~ o . At+ ~ o . t°) , (2.19)

~p(~) .= ~ # ) = ~ (~, ~o .At.# + ,p,# at ~ + u ~ . t°# + u.# . to,OL ) '

where only the symmetric part of p,,# is considered.

60 .I.C. Simo et al., Stress resultant geometrically exact shell mqdel. Part II

Table 2
Example: Elastic isotropic constitutive equations
Lowest-order constitutive model
@ Reference surface metric:

o Elastic isotrcGic constitutive equations:

022
va”l’a + ao"ao 12

1 _ v)ao’zao’2

a022a 022 ao22ao 12

1 -2 v aollao22 +

a022a0 12
(1 + v) $2 012
-a a
,' 2 1

Higher-order constitutive model

e Second fundamental form:

0 Constitutive tensors:

(1 - 4
ff UPV~
0 =
I va
O~flaOY~+
-a
2 (
oay oD&+ ao~*aolJ)
a
I, 9
H@VE = v(ao~yP~ + uo’6~o~@)
+ (1 ; 4 (ao~ybo@+ aOq)oav + ao=~~opy+ a0e,oL6)} ,
1
I
j.pM = v+P~~o~~+ (1 ; 4 (bo”~p~ + b”.)p)}
2
I

@vS = v(a”aDdOy’+ aoy6do”“) , f1 ; ‘) (a~)aydOBs

+ a()Padf,or+ aoa6&Pr + a”pvdoo*)
3
I

@ Membrane and bending stress resultants:

n-4 -_ & ( hHP;Pv6& - ; H~pvB~~v(j,+ ; H;PvGEy8+ ; H”;P’6E,,) ,

 J.C. Simo et al., Stress resultant geometrically exact shell model. Part H 61

R E M A R K S 2.2. (1) In classical developments of linear shell theory, the bending strain
measure/~(~) is often replaced by ~(~) plus a linear combination involving products of E~ and
0 .
the second fundamental form of the surface b~o n 0~ . ~p.,,.
0
This is the case in the "best linear
theory" of Budiansky and Sanders [10], where the bending strain measure ~ - = ~ ( ~ ) -
1 [b0~
~t = + b~
,~ %~ 0~ ev~],
- first proposed by Sanders [30], is employed. The two motivations for this
choice are: (a) the resulting theory reduces to Love's first approximation for spherical
geometry, and (b) / ~ vanishes identically for constant displacement fields normal to the
mid-surface. Our numerical experiments indicate that no significant difference is obtained
using either strain measure.
(2) Strictly speaking, changing the strain measure while holding fixed the structure of the
constitutive equation, yields a different constitutive model. However, as noted by several
authors, see e.g., Naghdi [25, p. 606], or Niordson [26], the constitutive models differ in terms
involving products of the first and second fundamental form, which can be argued to be of
higher order. In this regard, our numerical experiments yield indistinguishable results for both
constitutive models in Table 2. These two models differ only in terms involving products of the
first and second fundamental forms.
(3) The remark above suggests that numerical integration through the thickness of simple
three-dimensional elastic constitutive equations, a standard paractice in the degenerated solid
approach, is a costly procedure which leads to no improved accuracy for most thin, or
moderately thick, shell problems. E]

2.3. Linearized static weak form. Matrix expressions

The static weak form of the momentum equations, linearized about the reference state, is
obtained in the standard manner by means of the Gateaux (directional) derivative formula.
This leads to the bilinear form B: T~0 ~ x T~0 ~--~ R defined as

d I,=0
B -- DG(OO; (v, St)). (u, At) := ~ee O( V °'' (v, St)) ' (2.20)

where • ° = (~o°, t d) is constructed as in (2.16) with variations (u, A t ) E T , 0 ~ . Sinc,e AtE

T,0S2, the constraint condition t o. At= 0 is identically satisfied by setting At= A ° AT, with
AT ET~S 2 A summary of the linear theory, along with matrix expressions relative ,to the
standard basis (E--E3), is given in Table 1. Table 2 contains matrix expressions fi~r the
simplest example of an elastic isotropic constitutive model, as well as a higher-order
constitutive model derived from an asymptotic expansion of the stored energy function i!or the
three-dimensional linear theory.

3. M i x e d v a r i a t i o n a l f o r m u l a t i o n

As noted in Section 2.1 and in Part I of this work, the standard Galerkin formulation and,
consequently, a displacement finite element implementation can be carried out entirely in
curvilinear coordinates relative to the surface convected (Gaussian) frame {~.l,~.2, t}.
However, the mixed formulation set forth below requires the use of a local Cartesian frame in
order to satisfy exact convergence requirements under constant strain fields. A simple
procedure to construct this local Cartesian cocrdinate system is considered next.
62 J . C . S i m o et a l . , S t r e s s r e s u l t a n t g e o m e t r i c a l l y exact shell m o d e l . P a r t H

3.1. Local Cartesian system in Tn0S2

Let n ° = ~po x ~p.°E/ll~p°1 x ~P.21I be the normal field to the reference surface ~po: ~t__>R3.
o
Define a local Cartesian system {x ~, x}, a = 1, 2, with base vectors {n d, n2, n d} by means of
the orthogonal transformation

A°n - ( E 3 • n°)13 -l-[E 3 × n °] -l- 1 0 ( E 3 × n ° ) t ~ ) ( E 3 × ,ne) (3.1)

1 +Ea.n
where
o o
na "= A~E~ (a = l, 2) and n°=A°E3 . (3.2)

Observe that :l ° maps E 3 • 5 2 s...> n ° . = A°nE3 E 8 2 without drill, such that ~p,o. n° = 0. Since
n~ •nbo=sab, a n d n o • n oo = 0, by construction, we set
o

oxa . 0 0 0 Oxa 0
(3.3a)

The transformation for partial differentiation then takes the standard form

0 0 Ox °
O~" - cgx" 0~:" " (3.3b)

We shall often make use of matrix notation• Accordingly, we set

Jo =
[0nn" ~.no nO:~P.2o]
o o o o • (3.4)
n2"~.l n2 ~ , 2

Here and in following sections, derivatives relative to the coordinate functions of the local
cartesian system {n~, n~, n °} are denoted by (e)o For subsequent use, we note that the
resultant strain measures in the Cartesian system take the form

- ½ (n',l'u ,, + n .u
K,,b
"= ~t ( n oo ' A t b + n bo ' A t o + u , o . t o.b + U b ' t oo) , (3.5)
~ / ~ ' = n o ' A0t + u . o • tO .

Finally, making use of standard tensor transformations involving (3.3b), the constitutive
equations for the effective stress resultants relative to the local cartesian system take the form

~ab _ Eh , ,abed - "ab = Eh 3 • ,obcd -

1 - 1/2 /'/o ecd, m 1 2 ( 1 - v 2) n o red' ~ a = tcGshSab~b '
(3.6)
where
Habcd
0
= v~°b~cd + -1- -T v [~ .~6b d + ~ ad~jcb]'
(3.7)

and6°1' = 6ob = 6bo is the Kronecker delta.

 1. C. Simo et al., Stress resultant geometrically exact shell model. Part H 63

3.2. Matrix notation: m e m b r a n e and bending

The matrix formulation in the local Cartesian system is identical to that a!ready considered
for the curvilinear system. For completeness, we record below the basic results. Define:

u l }
u.= u2 ,
U3
[ (11 1
~'=/~,2 ~',
12~,~j
Then, define Bin: [ H ' ( ~ ) ] 3-->[L2(,.~)] 3 as
f K-11]
'~'=/~_= i"
12,q~j
AT={AATI}
T2 •
(3.8)

o,O !

n,ot oX
0~ i
B m "-- ' n2 O F " (3.9)
o' 0 o' 0
n, -~x 2 + n 2 "~x l

In addition, define B b = [~]~bmBbb]: [HI(,~)] 5--->[L2(~)]3 by the expressions

o, O o, O
t,! ¢)X--"T n 1 tgx l

Bbm = t2o, 0X
o2 ' Bb b = n2o, a Ao (3.10)
__ o' a o' 0 c3x2o' t3
0t c) + t , 2 nl _ _ + n 2
t. 10X 2 ~X l OX2 ~X l
where
I" 0 0
All A12]
A°:=[t~ A°, (3.11)
3 °, a 2J
Finally set
1 v 0 ]
v 1 0
(3.12)
o o l-v "

With this notation at hand, we have

Eh Eh 3
= 1-v 2H~' n~
.--e_

1 2 ( 1 - v 2) H~,

= BmU , = [IBbmBt, b (3.13)

where we have set ~ := [~1~22~12], and • := [n~ ~lr~22n~~2]t.

64 J.C. Simo et ai., Stress resultant geometrically exact shell model. Part H

3.3. Matrix notation transverse shear

Because the finite element treatment of the transverse shear strains employs an assumed
strain method, it is more convenient to formulate the transverse shear strains relative to the
surface convected basis {~p°l.~o.°,t}. As is defined in Table 1, we set IBs=--[IBsmlB~b]"
[ H ' ( ~ ) ] s--> [L2(,~)] 2 as

Bsm := t o' and B~b'= ~Po1, fi,". (3.14)

L
In addition, we set
[a o~1 a o~2]
C~ = KGsh aO12 0221 , (3.15)
a j
so that the transverse shear strains become:
I/
"/2
and
q : = { ~ 2 } =- C~'F • (3.17)

Next we consider the mixed variational formulation employed in the proposed linear shell
model.

3.4. The mixed variational formulation

With the matrix notation introduced in the preceding sections, the linearized weak form of
the momentum balance equations takes the form

Gdy n "= n d- f~ [ p l l . v q- i0 )~t ° ~t] d//, 0 - Gext, (3.18)

where 8 ~ ° = (v, ~t)E T,a0~ is an arbitrary kinematical variation, and d/z° =]o d~:~ d~2 is the
reference surface measure,

Goxt:=f~["'v+m'8t]dl2+fo,~. vfda° + f0 n~d mS~

m'Stfda ° (3.19)

is the virtual work of the external loading, and B is the linearized static part of the weak form
of momentum balance, which takes the form

B.=f {tamv]'. + [[l~bmBbb](~)]'~-1-[[[~smBsb](

10 q/d~°"
~r}J (3.20)
 J.C. Simo et al., Stress resultant geometrically exact shell model. Part H 65

Here, we recall that spatial and material variations are connected through the relation
8t = A 0 ST, where t o = A°E.
In addition to the weak form (3.20), we enforce the constitutive equations for the
membrane and bending resultants through the variational forms:

Eh )-' n + Bmu ] d/~° = 0 ,

1 - v2//
(3.21)

where 8t~ E [L 2(~)]3, and

~3
o
v.,
with 8~fi • [L2(~)] s. The variational equations (3.20)-(3.22) constitute the basis for the finite
element formulation discussed in detail in the following sections.

4. Finite element interpolations

The finite element interpolation of the reference surface q,0:,~_,R3 and the surface
displacement field follows a standard procedure.
Let

I-1.= {(~, ~1)e [ - 1 , II x [-1, 11} (4.1)

denote the biunit square, as shown in Fig. 1. According to the standard isoparametric concept,
we define

~ : E]~R3 h UltRa (4.2)

4 ... O 3 ~ ~ ~
J ~

A C "-~ ~

+ • NodalPointsI.1,2,3,4
:
0 AssumedStrainPoints:M-A,B,C,D
1 B 2 + 2x2QuadratJrePoints
Fig. l. Standard isoparametricbiunit square, mapped into the shell element configuration.
66 J.C. Simo et ai., Stress resultant geometrically exact shell model. Part H

by setting
4
h
tpe "= E NI(~, r/)dl, (4.3a)
1=1
4
u h := E NI(~:, r/)u/, (4.3b)
!=1

where d ! E [~3 and u~ E R 3, I = 1 , . . . , 4, and NI(~:, 7)) are the standard bilinear isoparametric
shape functions defined as

N'(~:, r/):= }[1 + ¢'~:t][1 + ~ , ] , i = 1 , . . . , 4, (4.4)

where ( ~ , r/1)~ { ( - 1 , - 1 ) ; ( 1 , - 1 ) ; (1, 1); ( - 1 , 1)}.
For further details on the isoparametric concept, we refer to standard references [17, 45].
The distinctive feature of the present approach concerns the interpolation of the reference
director field t o. ~--->S 2 and the linearized director field At: ~-->TtS 2. Crucial to the
development of this interpolation is the geometric structure underlying the present shell
model, which is considered in detail in Part I of this work. It will be shown below that the
interpolation of the linearized director field is automatically defined and follows from the
interpolation for the reference director field.

4.1. Reference director field

Let t 0t E S 2 denote the directors (unit vectors) at the nodal points. Note that tt=
0
~..... 4 need
not be normal to the reference mid-surface. Since t °' ,~t--,S 2, we consider the following
interpolation:
4

E N't;
t h ~. i=1
(4.5)
Jf,
This interpolation satisfies the geometric requirement that t eh ~ s For convenience, we
introduce the notation

h i 71
7' "= E N', ° t,- [[~[[ • (4.6)
!=1

To simplify the notation, in what follows, the subindex "e" is often omitted.

4.1.1. Derivatives of the director field

In order to obtain the interpolation for the bending strain measures, one needs to calculate
the derivatives of the reference director field. This is accomplished as follows.
 J.C. Simo et al., Stress resulumt geometrically exact shell model. Part H 67

Let Pt: R3--> Tt $2 be the orthogonal projection defined as

Pt'=13-t®t. (4.7)

The fact that (4.7) defines an orthogonal projection follows from the properties:
(i) P, = p t ,
(ii) P, oP, = P, ::>P~ - P, VNEN + ,
(iii) P,t = 0 ,

IIP:II
(iv) IIP, II'= maxx,,0 ~ Ilxll
= 1 .

We now have the following

P R O P O S I T I O N 4.1. The derivative of t h" ~---> S 2 with respect to the surface coordinates is
given by

,,=1
" ]l~ ~th '=' N~.t, .
] (4.8)

PROOF. By the chain rule

oll?'ll _ o V?'.?

1 • t~ol = t" (4.9)

Therefore, from (4.6) we have

th., = ~ ~1~ [.~',, -- th(th ' ~,.)], (4.10)

and the result follows from definition (4.7). El

4.2. Interpolation o f the linearized director field

To obtain the interpolation for the linearized director field consistent with (4.6), we use the
property that At h" M--, TtS 2 must take value~ in T,S~ Consequently, we proceed as in the
continuum problem (for further details, see Section 6 of Part I) and consider the curves
4
- ~ "= X N' Cxp,o[, At,], (4.11)
!=1
where
0 sinll e At, ll
exPto[E At,] "= cosll* At, lit, + e At~ (4.12)
II~ at, II
68 J.C. Simo et al., Stress resultant geometrically exact shell model. Part H

is the discrete exponential mapping at nodal points. Then, to obtain a curve e ~->t~h • S 2, we set

e ~-> t~h := ~ t~ E S2
. (4.13)

Construction (4.11)-(4.13) must satisfy two required properties:

h
(i) t~[~=o = t h E S 2,
(ii) (d/de)l,=0 th =" At h ET,hS 2.
Property (i) follows immediately from (4.11), (4.12), and (4.13). For property (ii) we have

d I~=0 t~h= ~ 1
Ath : = d--e ~])th A t "h , (4.14)
where
4
A t h "= ~ N' At I . (4.15)
i=1

Therefore, by properties (i), (iii), and (iv) of Pth,

t h. P:, At h = pt, hth" At h ~ 0. (4.16)

Thus, it follows that At h as defined by (4.14), (4.15) is in Tt0S2. To summarize the results, we
have the following.

PROPOSITION 4.2. The interpolation for the linearized director field consistent with (4.6) is
given by
4
At"= 1 A~' where A~' "= ~ N I At I . (4.17)
!=i

4.2.1. Derivative of the linearized director field

We conclude this section by recording the counterpart of (4.8) for the director field. We
have the following.

PROPOSITION 4.3. The derivative of the linearized director field is given by

4

Ath~ = E Tt~ Atl , (4.18)

!=1
where
• .,9)

PROOF. The result follows immediately from (4.17) and by the chain rule with a calculation
similar to that in Proposition 4.1. D

R E M A R K S 4.4. (1) Interpolation (4.5), (4.6) is only one among several possible approxima-
tion schemes. Alternatively, one may define
 J.C. Simo et al., Stress resultant geometrically exact shell model. Part H 69
4 ."h
= ,=1
~ N'exp[,f,], teh= t~ (4.20)

where ,~t E T ES2; that is, ~ - E = O. Observe that ;t't = E x ,~! is the rotation vector of the
director t~0 ~ S 2 El
0
(2) Interpolation (4.20) requires the calculation of the rotation vector from E - E 3 to t t.
This extraction requires a careful use of quaternions to avoid singularities [34, 35, 36]. For this
reason, we advocate scheme (4.3), (4.6). []
(3) W e rcca|l that the transformation from two rotational degrees o f freedom ATt E TES 2 to
a three rotational degrees o f freedom At ! E TIoS e is accomplished through the (3 × 2) matrix
o This matrix constitutes the first two columns of

1 o ( E x t o~ ) ® ( E x t ° ) , (4.21)
A ° = r E • t°)13 + [E
X--~tll +
l+E't~
where t o - A°E. [2
A summary of the interpolation procedure is contained in Table 3.

Table 3
Interpolation of the director field
1. Reference director field:

~:-,__,Ntt~ =>t.--~,
h te

A ° = tB. :)1, + + 1 o ( E x t,)®(E×t

o °)
1+E.tt

2. Derivatives (at Gauss points):

P/, : = 13 - t* ® t h ,
4
/'~ : - E N' to
I--I

1 Pi"
th,a ---- I[~ th ,a

3. Linearized director field: ~o = [to t21,

,o

Air =/t° AT,,

4
Ath= ~ Pt* I=!
~ NI At1,

T~ := [[~1 {P,,N~, -- [ t ~ ~ t * + t h a t *. ."F .( th. " ~h )P,./II?IllN'),

4

a¢o- E r'. I--I

70 J.C. Simo et al., Stress resultant geometrically exact shell model. Part H

5. Treatment of membrane and bending strains

It is by now a well-established fact that strict adherance to the isoparametric concept for the
membrane strain field, within the context of displacement-type approximations, results in poor
performance of the finite element approximation. Stolarski and Belytschko [40, 41] coined the
term membrane locking to characterize the inability of (fully integrated) displacement finite
element interpolations to reproduce inextensional states of bending stress. Several schemes
have been proposed to ameliorate this effect. These range from selective and uniform reduced
integration techniques [13,28, 46] to assumed strain methods [11, 18,21, 27], and projection
methods [6]. These methodologies appear to fall within the framework of mixed methods
based on the Hu-Washizu principle [37]. Two observations concerning this class of methods
should be made:
(i) A unified, sound, mathematical basis appears to be lacking, even within the context of
plate theory. Recent attempts in this direction are made in [3, 4] among others. An
exception to this unsatisfactory state of affairs is the recent work by Brezzi and Fortin
[9] and Arnold and Falk [1].
(ii) In lieu of a more rigorous analysis, patch-test requirements [43] should be enforced. In
this regard, we note that several variants of recently proposed assumed strain methods
do not pass standard patch tests [18].
With these observations in mind, in this section we derive a treatment of the membrane and
bending fields characterized by the following features:
(1) The treatment is consistent with a Hellinger-Reissner formulation. For flat plates, the
interpolation procedure is closely related to the mixed formulation for plane stress
proposed by Pian and Sumihara [29] (which appears to be optimal). The present
formulation extends these ideas to shell elements with initial curvature.
(2) Correct rank: the element is free of spurious zero energy modes. In addition, exact
satisfaction of patch test requirements is obtained.
(3) Excellent performance in standard numerical tests is achieved. The element appears to
be free of membrane locking, and is among the top performers in the four-node
element class.

5.1. Approximation for the membrane stresses

Let t7"~ be the convected components of the effective membrane stresses. Accordingly:
= ~,~ _0 0 ~,b 0 0
~,~ ®~p.~ = n n , ® n b . (5.1)

The interpolation for the Cartesian components, $.b, of the effective membrane resultant is
constructed as follows:
(i) Assume interpolations, discontinuous over elements, of the form

I'L""]"= 0 /~;(, _ ~) , (5.2)

w h e r e / ~ = [~1,e J[32, ~;, /34,/~;]

-e t ~ R5 and ~, ~ E R are constants to be specified below.
 J.C. Simo et al., Stress resultant geometrically exact shell model. Part H 71

(ii) Let Jo "= Jol~--,--o be the Jacobian transformation from the basis {ti °, ti °} C T~oS2 to
{6,o , ~P.2}
-o c T~oS2. By (3.4), we have

i ol ~o,1
o nlo I cp 1
J0-L 0 (5.3a)

where ~0.o
-° ~,.o o
I,--~--o and n-~o = Ilao i ~='~ffiO"
Since the formulation is to be carried out in local Cartesian coordinates, the derivatives of
the shape functions need to be transformed. This is accomplished by setting

2Vt, l = tJo]-t{ N~e } (5.3b)

t/~f2J Nf, "

(iii) The approximation for ffab is based on (5.2) and the coordinate transformation:
fi,h= yo.Yoat~a~, (5.4)
where Joa are the components of J0.
This leads to the following finite-dimensional approximating subspace, Sh, for the effective
membrane field

where
s. := { I
,~ ~ [ L ~ ( ~ ) ] ~ ,~ := ,~'2J~ s(g, ,7)#,, t3, ~ R ~
=

t,
(5.5)

s(~, ~) = [131 (~ - fi)gr,(~ - ~ ) # ~ l , (5.6a)

with

~ =/(jo2 ,)2~ and jr =~(j202)'~. (5.6b)

L&g,J 14,:o~J
(iv) Select ~, ~ E R by the expressions

~=~ ~dg ° , (5.7)

~=~ee v/d/z

where d/.t° = )-o d~ dr/, and Me = f~ d/t°

R E M A R K S 5.1. (1) It is essential to perform the transformation (5.4) at the center of the
element, otherwise patch test requirement~ are violated.
(2) The motivation for selecting (~, ~) as in (5.7) lies in the simplicity of the resulting
computational architecture. For plane stress, the method reduces to a one-point quadrature
l~lus a rank-one stabilization, as in [7, 8]. The formulation in [29] corresponds to setting
~=~=o. E3
A summary of the mixed interpolation parameters is contained in Table 4.
72 J.C. Simo et al., Stress resultant geometrically exact shell model. Part H

Table 4
Mixed formulation parameters
• Central Cartesian coordinate frame:
-0 -0
~0= ¢. IX¢,2
IIC, x ~,o:11 '
Ao = [~o-o
n2 ti °]
1
= (E~. a°)13 + [E,
x~~ ]+ + ~io
(E, x ,i °) ® (E~ x ,i °)
t~ E 3 •

• Jacobian transformation:
-0 -0 -0 ~0
[n,.¢., n,. .1
]O=L-O -o
n2 ~).,
-o -oj
n2 ~,2

• Cartesian shape function derivatives:

• Mixed interpolation:

s(~, ,~) = [~., I (,~ - ,~)~,(~ - ~)~1,

! 2
f<J+,);) f< o,, 1
/,,,],,, L]~,,]~}
1 fa d." I f d/z"

5.1.1. Matrix notation

On the basis of the weak form (3.20), the contribution of the membrane stresses to the
stiffness matrix is evaluated in a straightforward fashion as follows. Set

u , . := f~s'ci.'s d~ ° . (5.8)

Note that because of the choice (5.7), H m takes the simple form

[~eCm ' °3×2]

H m := _ , (5.9)
O2x3 H e

where/te is (2 x 2) matrix with the following structure. Let ~, = ~:, ~2 = r/, and

h,~ = f~ ( ~. - ~,~)(~:. - ~.) dp, ° . (5.10)

 J.C. Simo et al., Stress resultant geometrically exact shell model. Part H 73

Then
ffleal3= ~gaCm ~ljhal3 ,
-t -1 -
(5.11)
where no sum on a,/3 is implied. Next compute

Gim:= f~ S,[Bm]
t ! d/ff, (5.12)

where [B~m]is the discrete displacement approximation to the membrane strain-displacement

operator (3.9) defined as

I N,lnl
ffi%l 0t 1
n~m = /~t,2n o'
2 (5.13)
~1 0t t%l 0t
/v,ln 2 +/N,2n I
"1
Here, N.,, denote the derivatives of the shape functions relative to the local Cartesian system,
which are computed using (5.3b).
The contribution of the membrane field to the stiffness matrix associated with nodes (I, J)
is then given by
K/mJ_. [Gm]
! t~g/m
-1 ~ mJ " (5.14)

REMARKS 5.2.
(1) The inverse of H= can be computed in closed form, because oi the simple structure of
equation (5.9).
(2) The product [Glm]tltmlGSm can be expanded and rephrased as a one-point quadrature
plus stabilization; the one-poitit qut,trature part being given by

Bltff , Rj '
0 ~"m~0 B°I .= 1 f~ Sm
~e ! d/~° ' (5.15)

which would be integrated using the central quadrature point. []

A summary of the membrane stiffness formulation is found in Table 5.

5.2. Approximation of the bending stresses

We start by recording the expressions for the discrete bending strain-displacement operator.
By substituting interpolation (4.3) into (3.10), we obtain

,1 ,1
1 = ~/~t
i tt o'
ot 1 (5.16)
Sbm / ,2 ,2 '
"I 0t + ~"I 0t
L N.It.2 N.2t.l
74 J.C. Simo et al., Stress resultantgeometrically exact shell model. Part H

Table 5
Membrane stiffness c a l c u l a t i o n

• I n v e r s e c o n s t i t u t i v e matrix:

c~
'I= -v

• S t r a i n - d i s p l a c e m e n t matrix:
1
0
0
°1
2(1 + v)

r 1
/~,t ot -- ~",! or/
L lV.ln2 -t- l V 2 n I J

M i x e d form:

t ,¢'-, - 1 ,.-. 7 0
Hm t= S ~m ~]L ,
L=-I

(J - P t s

G'm = Y~ ~ n' "m' I , .-"

~FL
L=I

• Membrane stiffness:

K,7, = G,,,H~,'G~
"

nluh = 0t
n2 T2
!
"3"°1, (5.17)
,, 0t,,,, i 0t,,,, !
tn 2 t l + n I l 2)

for I = 1 , . . . , 4, where TX~,is defined by (4.19). The interpolation of the curvatures then takes
the form

"e
,r,r
= I "~ t
4
= E [s~,,.Is',,,,]
{ "' 1 (5.18)
12ff12j e I=! AT! "

At this point, two finite element approximation schemes are considered.

5.2.1. Displacementformulation
One simply enforces the bending constitutive equations strongly (or point-wise) through the
relation
"h -h
me = CblCe • (5.19)
 J.C. Simo et al., Stress resultant geometrically exact shell model. Part II 75

From (3.13), it follows that the contribution of the bending stress resultants to the stiffness
matrix, associated with nodes (I, J) and denoted by K~J, is given by

B/tb Cb[nbmBbb] d~ ./,0. (5.20)

5.2.2. Mixed formulation

Alternatively, one may enforce the constitutive equation (5.19) weakly, through variational
equation (3.22). In particular, we consider the approximating subspace, Shin, of functions
discontinuous between elements:

S~--- ( ~ E [L2(~)]31 ~h£ "--S(~, 7/)a~"$ ot~ ER 5} , (5.21)

wheze S(~:, ~/) is given by (5.6a). The implementation of this approach is entirely analogous to
that considered in detail in Section 5.1 for the membrane stresses. The details are omitted,
and a summary of the matrix formulation is presented in Table 6.

Table 6
Bending stiffness calculation (mixed form)
• Inverse constitutive matrix:

= -v 1 0
~-~ 0 0 2(1+u)

• Strain-displacement matrix:
I ~]l to t not,m/~" 0 ]
,1 ,! 1 !11][I
B Ib ._ _,Oral "i 0
F~',2t °'.2 ~2 12111
" I 0t " I' 0 t / O t ~ l 0t / - 0
N i t 2 + N,2t,! ~n 2 1 1 "~- n I T2)At
• Mixed form:

H;' = [ G~Pts

L=I
S'Cb'S~ ~r.
]-i

G- Pts
t,"~= ~, S'e ':°
bl L ~t/'l_
L=!

• Bending stiffness:
!$ It
Xb --GbHb'G~
76 J.C. Simo et al., Stress resultant geometrically exact shell model. Part H

6. Treatment of transverse shear strains

To avoid the shear locking phenomenon that dominates the performance of displacement
finite element formulations in the thin shell limit, we employ an assumed strain method based
on the Hu-Washizu principle. The construction of the assumed covariant shear strain
interpolation is closely related to procedures advocated by MacNeal [21], Hughes and
Tezduyar [16], Bathe and Dvorkin [5]. The only available mathematical analysis of this
scheme is due to Bathe and Brezzi [3, 4].
Hece, we cast the approximation scheme in the context of a Hu-Washizu type of variational
principle following ideas of Simo and Hughes [37].

6.1. "Construction of the assumed strain field

Consider a typical isoparametric finite element, as depicted in Fig. 2, and denote by
M = A, B, C, D, the set of mid-points of the element boundaries. Following MacNeal [21] and
Bathe and Dvorkin [5], wc consider the following assumed transverse shear strain field:

~h = ½[(i -- ¢/)y B1 + (I + Z/)y ~1 (6.1a)

-h
V:, = ½[(1- ~)),2A + (1 + ~)'),c I , (6.1b)
where
+ a:. (6.2a)
~ = t B • U 8,~+At ~ . ¢ , ~B,
(6.2b)
yc=tc u c.2 + Atc" ¢,~.,
c
• (6.2c)
y t 1) = t ° - U o,I -I- A t D- ¢ 0.,! (6.20)

are the covariant transverse shear strains collocated at the midpoints of the element boun-

xl

t
D 3

, C ---,,.. ~

+
• Nodal Points: 1=1,2,3,4
O AssumedStrain Points: M,:A,B,,C,D
B 2 -I- 2x2 Quadrature Points

Fig. 2. Construction of the assumed strain field on the standard isoparametric biunit square.
 .J. C. Simo et al., Stress resultant geometrically exact shebi model. Part II 77

daries. Making use of interpolation (4.3), it follows that

Similar expressions are obtained for the surface displacement field u%: ClI+ R”.
Next, from the assumed interpolation of the director field given by (4.5)) we obtain

ti + t”P
M=A,B,C,D,
p = a = 1,2,3,4, (6.4)
llt8 + t”,ll ’ p = 4,1,2,3.

Finally, we note that

(6.5)

where rlc,= (l/ 11

P Il)P&)~. Recalling the properties of the projection operator
IF+: 1W3+T,S2,and making use of (4.17), the second term in (6.2a) now becomes:

(6.6a)

In this expression, *r&is defined as

1 9: - (rol: l tAPA
(6.6b)
a& := 7 P,Ap$ =
II II Illo! + Oil ’
Similar expressions are readily derived for the remaining terms in @2).
By making use of the relations (6.2)-(6.6), the assumed covariant transverse shear field
may be expressed as follows. Set

(6.7)

Then
(6 8)
I(

where
-(l - q)P (1 - q)tB’ (1 + rl)cDL -(l + 7j)tDt
B,, := d (6.9)
-(I - l)tAt -(l + 5)t”l (1+ 6)t”l (1 - 5)tA’ I
and
(6.10a)
78 J.C. Stmo et al., Stress resultant geometrically exact shell model. Part H

Here,/]to is a (12 x 8) transformation matrix defined as

A° -o ]
HO = A2 fi, o . (6.10b)
3 A0 J

This completes the development of the assumed strain field .~h. The discrete variational
formulation is constructed on the basis of the following mixed principle.

6.2. Mixed HU-Washizu formulation

Let
f o
//s" J~ ['-"
._
:'~ .Cs, Y
-~ ..~ qh(,}lh -'2~ )] d , (6.11)

be the potential energy associated with the transverse shear strain, expressed in a Hu-Washizu
form. Here ~h denotes the assumed transverse shear strain field, and yh is the shear strain
field emanating from the isoparametric interpolation; accordingly:

7eh __ BsmU e + Bs, b AT e (6.12)

where Bsm and B~b are obtained from (3.14) and by interpolations (4.3), (4.6), and (4.17). The
first variation of (6.1!) relative to (Vh,Sth)ER3× TpS 2 then yields the contribution of the
transverse shear ~train field of the weak form (3.20) as
Vh
(6.13)

This cxpression is identical to that in (3.20) with [BsmBsb] replaced by the assumed strain
operator [B~mBsb]. In addition to (6.13), from (6.11) one obtains the following two variational
equations:

f~(~h)t([BsmBsb ]_ [Bsm'~sb]) d ll, o{ AT

, h h } _. 0 (6.14)
and
=0, (6.15)

for arbitrary ~i~h E [L2(,~/)]: and (v,,, 8Te)E IRa x TES". It can be shown (see [.37] that ~h can
always be chosen such that (6.14) and (6.15) are satisfied, and (6.13) takes the form

c~ [v' ~r']
•
=
fo{'}
sb
/~t { /
B~ c d & ~ d d~° u~
~ AT e • (6.16)

In other words; the approximation procedure can be formulated as a ,~-method as proposed in

 J.C. Simo et al., Stress resultant geometrically exact shell model. Part H 79

Table 7
Shear stiffness calculation
• Constitutive matrix:
l- 011! a 012 ]

• Strain-displacement matrix (see (6.2)-(6.6)):

[ Nl,1 t tB I N2, ~.~ " fI, [[ N21t

n s "- [ .1 B
[ N 2.l 'i~.|" • A2
N It t ~4 ,4 ' t l Net t N3 c
L .2A N,2~.2 .2C .2~,2 A2

,1 D N.l~.l" ,l o N.l~.l A4
H
II N3.2 t tc 3 C
N.211,2 A3 N4,2 t 'A 4 A A4
-~,2'1~.2

• Shear stiffness:
G - Pts

KisS = 2~ltt'~'~J':'O
B s q.,*SBSJ L °~L
L=I

[12] We note that the interpolation scheme advocated in [16] falls within this class of
methods Consequently, the contribution of t~ae transverse shear strains to the stiffness matrix
becomes

Bsb

A summary of the malrix formulation of the assumed-strain shear stiffness is found in

Table 7.

7. N u m e r i c a l simulations

The performance of the proposed shell element is evaluated with several discriminating
problems selected from the literature. Patch tests for the constant states of stress of the
element are first listed, then some of the more challenging problems suggested by MacNeal
and Harder [22], Liu et al. [19], and Stanley [38] are performed. Problems examining
sensitivity of the element to mesh distortion are also examined. The convergence of the results
is compared to other well-known formulations ranging from four-node discrete Kirchoff
approximations to biquadratic degenerated solid formulations. A listing of these shell ele-
ments, and the abbreviations used to identify them henceforth, is contained in Table 8.

7.1. The patch test

A square plate is modeled using the single element shown in Fig. 3(a) and the skewed mesh
shown in Fig. 3(b), and subjected to constant states of tension and betiding. The stresses and
80 J.C. Simo et al., Stress resultant geometrically exact shell model. Part II

Table 8
Listing of some standard shell elements
Name Description
4-ANS bilinear assumed natural strain element, with full
integration Stanley [38], Parks and Stanley [27]
4-CBS bilinear continuum based shell element of Bathe and Brezzi [3]
4-DKQ dis,.,ete Kirchoff quadrilateral of Taylor [42]
4-RSDS bilinear resultant-stress degenerated-shell element, with
uniform reduced integration and stabilization Liu et al. [19]
4-SRI I-'linear degenerated shell element, with selective reduced
ih~egration Hughes and Liu [15]
4-URI bilinear uniformly reduced Lagrangian degenerated shell
9-ANS biquadratic assumed natural strain element, with full
integration Stanley [38], Parks and Stanley [27]
9-GAMMA biquadratic degenerated shell element, with uniform reduced
integration and ~/stabilization Belytchko et ai. [6]
9-SRI biquadratic degenerated shell element, with selective
reduced integration
T1 bilinear plate bending element of Hughes and Tezduyar [16]

Present formulations:
MIXED bilinear shell described in the previous sections with the
mixed formulation used for the membrane and bending
stresses and full 2 x 2 quadrature
DISP bilinear shell described in the previous sections with the
displacement formulatiov used for the membrane and bending
stresses and full 2 × 2 quadrature

Fig. 3. Constant stress patch tests: Tension, benuing, and in-plane twist
 J.C. Simo et al., Stress resultant geometrically exact shell model. Part 11 81

1.1

i.0
. . ' ~ .... ."2~.'--':--
I
,jg'ff ,.," ...... 9-ANS
~ 0.9- df ,s ° - - - 0 - - - - 4-ANS
F s" - - .-- .-- 4-URI
• .-------- Present
•,~ 0,8-
t - .
.......jt
E 16
• w,,,t

' 51 I
I
~ 0.7- I
I l
Z I /
# f

0,6- #
' l oI
0.5-~ ! !

0 5 10 115 20,

Number of Node,s per Side

Fig. 4. Description and results of Cook's membrane problem. Present results are for the mixed formulation.

displacements obtained with both meshes are exactly identical to the analytical answers. The
material and geometric properties are: E = 10 × 10 6 v = 0.3, and 1-= 10.

7.2. Cook's membrane problem

A trapezoidal plate is clamped on one end and subjecte,~ to a distributed in-plane bending
load on the other end, as show,l in Fig. 4. This problem has a considerable amount of shear
deformation, and is an excellent test of an element's abiili,y to model membrane dominated
situations with skewed mesi~es. A finite element converged solution of 23.91 is used to
normalize the results, which are shown in Fig. 4 and listed x, Table 9. The material properties
are E - 1 . 0 , v--0.33, and , ~ - 1.0. The proposed eler~ent's performance surpasses the
bi-linear shell elements and is i,.mong the ranks of the biqu~dratic elements. 70% of the final
solution is captured with one single element.

Table 9
Results of Cook's membrane problem
Number of Displacements
Elements (inches) (%)
1 16.743 70.0
4 21.124 88.3
16 23.018 96.3
64 23.685 99.1
256 23.878 99.9
82 J.C. Simo el al., Stres,~ ~esultant geometrically exact shell model. Part H

7. ~. Bending o f a rhombic plate

A simply supported rhombic plate of side l = 100, material properties E = 10 x 106, v = 0.3,
and thickness h = 1.0 is uniformly pressure loaded. The centerpoint deflection is normalized
with the analyucal solution of 0.04455, obtained by Morely [23, 24].
The difficulty in this problem arises from the singularity of the solution at the obtuse
vertices, where the stresses tend to infinity, and thus the usefulness of this problem has been
questioned. To alleviate some of the numerical difficulties, several different mesh configura-
tions have been used. Hughes [17] uses a uniform mesh on a quadrant of the problem with
smaller elements at the base, which yields elements of trapezoidal shape. Vu-Quoc and Mora
[44] use a uniform nine-node element mesh, but move the interior nodal points closer to the
obtuse vertex. To assess the convergence of shell elements of different configurations, we use
the uniform mesh shown in Fig. 5. This mesh is a more demanding configuration due to the
rhombic shape of the elements and the lack of mesh refinement at the singularity points.
Again, the proposed element proves to be superior to other bilinear formulations, and in
this case even to the biquadratic elements° Stanley [38] reports that the 4-SRI element initially
gives good answers, however it diverges from the correct answer as the mesh is refined. The
results are listed in Table 10 and shown in Fig. 5. The superior performance of the element is
attributed to the mixed formulation used for the bending stresses. When the displacement
formulation is used for the bending, the results are identical to the 4-CBS and T1 elements.

7.4. Shearing o f a twisted ribbon

This problem, a beam twisted 90 °, was suggested by MacNeal and Harder [22] as a test to
assess the effect of warping on the performance of a shell element. A more demanding version

1.2

I,O-
* * 0 • • • .d~oannm~ ob--',,,"
gaaEpq~"

• 0 5 P P 6 --" a e ~ m ' e n
.mq
*"
¢J 0.8- • • o~ I~ ALTCq

~* ~--,,-- 4-ANS
ji F ---- 4-CBS,TI
..~ 0.6 Present
P4
. ,,.,.j

Simply Supports! . , ~
0.4
z ~.,~-~-~'~......
- ~¢.'.~ 3o

0.2

{) !
o 1o 3'o 40

Number of I,~odes per Side

Fig. 5. Description and results of the rhombic plate under uniform presure loading. Present results are for the
mixed formulation.
 J.C. Simo et al., Stress resultant geometrically exact shell model. Part H 83

Table 10 Table 11
Results of the rhombic plate Results of the twisted ribbon
Center displacement Mesh Load case (a) Load Case (b)
Nodes per
side (inches) (%) size (inches) (%) (inches) (%)
5 0.04282 96.1 1x 6 1.378 99.1 0.3261 95.1
9 0.04264 95.7 2 x 12 1.387 99.8 0.3383 98.6
17 0.04387 98.5 4 x 24 1.389 99.9 0.3421 99.7
33 0.04496 100.9 8 ><48 1.390 100.0 0.3431 100.0

of the same problem (used here) is obtained for the case of a twist:d thin shell. A thin shell of
length l = 12, width w = 1.1, thickness t = 0 . 0 5 , and materiai properties E = 2 9 . 0 x 106
u = 0.22, twisted by 90 ° is cantilevered and subjected to two cases of end loading:
(a) Unit shear load in the width direction.
(b) Unit shear load in the thickness direction.
The results are listed in Table 11 and shown in Fig. 6. The proposed element encounters no
difficulties in modeling this problem effectively even with coarse meshes.

7.5. Pinched hemispherical shell

A pinched hemispherical shell, with two inward and two outward forces 90 ° apart is
modeled using symmetry boundary conditions on one quadrant. This problem is a good test of
the inextensionai bending behavior of an element, and an excellent test for the ability of an
element to model rigid-body motions.

1.5
Load Case (a)

].2 "

x
•~ 0.9-

F = 1.0 (a) J
'~ 0.6-

o.i]i- "z., ,

....... 4-DKQ

~. PreseNt
Load Case (b)

400
Number of Elements

Fig. 6. Description and results of the twisted ribbon under two cases of shear loading: (a) in the width direction,
(b) in the thickness direction. Present results are for the mixed formulation.
84 J.C. Simo et al., Stress resultant geometrically ex,~'cl shell model. :Part H

Table 12
Normalized results of the pinched
hemisphere with 18° hole
No6~:s Mixed 4-DKQ
per side (%) (%)
3 91.9 66.3
5 100.4 92.8
9 99.8 101.0
17 99.9 100.3

Two separate problems have traditionally been used to evaluate the performance of
elements: the pinched hemisphere with an 18° hole at the top, shown in Fig. 7, and the full
pinched hemisphere, shown in Fig. 8. Both problems share the same material properties:
E = 6.825 x 107 and v = 0.3, radius R = 10, and thickness h = 0.04.

7.5.1. H e m i s p h e r i c a l shell with 18 ° hole

A value of 0.094 has been published [22] for the displacement at the points of application of
the forces, however, analytical solutions based on asymptotic expansions [39] appear to yield
an answer of 0.093, which is used for the normalization. Table 12 lists the results of this
problem and Fig. 7 compares the proposed element's performance to the 4-DKQ, 9-SRI, and
9 - G A M M A elements.
The performance of the proposed element in this problem is quite remarkable. A mere
2 x 2 mesh produces 92% of the numerical ~nswer, and the results are practically converged
with the 4 x 4 mesh. No shear or membrane locking is detected.

1't4 ~~
1.2
......
----~-
4-DKQ
9-GAMMA
•"-- - " ~ 9-SR!
°~.~.p~..+. _~ Present
0
iw,I
,o L:..."" ............

0.8

'0
0.6

Z 0.4
F:?0
~pdp~e
0.2 // .,o
~O
0
~go~
0 I; i'5 20

Number of Nodes per Side

Fig. 7. Description and results of the pinched hemisphere with an 18° hole. Symmetry is used and only one
quadrant is modeled. Present results are for the mixed formulation.
 J.C. Simo et al., Stress resultant geometrically exact shell model. Fart H 85

Table 13
Normalized results of the full pinched hemisphere
Nodes Present formulations Other formulations
per side MIXED DISP 4-SRI 4-RSDS 4-DKQ
5 65.1 46.8 41.2 96.5 75.3
9 96.8 93.3 92.7 97.1 98.5
17 99.3 98.8 98.4 98.9 99.8

7.5.2. Full hemispherical shell

The mesh used for this problem is highly skewed for coarse meshes, and leads to less
elements for the same number of nodes l~er side than the previous mesh; for example, a 2 x 2
mdsh produces four elements in (7.5.1) but only three elements in the present situation.
The results of this problem are listed in Table 13 and Fig. 8. An analytical answer of 0.0924
is used to normalize the displacement at the points of application of the forces.
The performance of the proposed element matches that of some of the best bilinear
elements reported in the literature.

7.6. Scordellis-Lo barrel vault

The Scordellis-Lo barrel vault, a short cylindrical section loaded by gravity forces, is a
membrane-dominated problem, which serves to test the ability of the element to model
complex states of membrane stresses. The length of the cylinder is L = 50, the radius is
R = 25, the thickness is t = 0.25, and the span angle of the section is tp = 80*. The material

i.2

|.0"

°~f-~P'v~'~ ° ~,~o'i~o 4.DKQ

0.8- 4.SRI
4-RSDS
,n=l =. ~seat
~ o.6-

• w,,,,i

~ 0.4-
/
F= l.O
0,2a • X

0 ~ 1~ 75 2o
Number of Elements per Side
Fig. 8. Description and results of the full pinched hemisphere. Symmetry is used and only one quadrant is
modeled Present results are for the mixed formulation.
86 J.C. Simo et al., Stress revdtam geometrically exact shell model. Part H

Table 14
Normalized results of the ScordeUis-Lo barrel vault
Nodes Present formulations Other formulations
per side MIXED DISP 4-SRI 4-RSDS 4-DKQ
3 145.0 ! 19.8 126.3 190.4 139.1
5 108.3 94.4 96.4 120.1 104.8
9 101.5 97.3 98.4 104.6 100.5
17 100.0 99.2 99.9 i01.0 99.6

properties are E = 4.32 x 107 and v = 0. The solution to this problem is obtained numerically
by Scordellis and Lo [31]. These authors report a value of 0.3024. The results are listed in
Table 14 and shown in Fig. 9.
Most of the elements considered converge reasonably, and the proposed element conw~rges
to 100.C% of tile answer with the 16 x 16 mesh.

7. 7. Pinched cylinder with end diaphragms

A short cylinder, with two pinching vertical forces at the middle sectl¢.n, and two rigid
diaphragms at the end, is modeled using one octant and applying the appropriate symmetry
boundary conditions. The length of the cylinder is L = 600, the radius is R - 300, and the
thickness is t = 3. The material properties are E = 3 x 106 and u = 0.3. -
The numerical results are presented in Table 15 and Fig. 10, nermalized against the
analytical solution of 1.82488 x 10 '5.
This problem, by far, is the most demanding of all cases considered. It is a severe test of the
inextensional bending and complex membrane states of stress. Most forx-node shell elements
2,0

1.8,
st

• g,,l,t
| ~ , RigidDiaphragm
!.4 o,~-|
:%1
. '~ | ...... 4-DKQ
, I z
~ I I I I 4-SRi
1.2
= \ ' . V ". ~, - - - 4-RSDS
Z "~'~ %~% -""~- Present
.........
1.0

0,8
0

Number of Elements per 3ide

Fig. 9. Description and results of the Scordellis-Lo roof. Symmetry is used and only one quadrature is modeled.
Present results arc for the mixed formulation.
 J.C. Simo et al., Stress resultant geometrically exact shell model. Part 11 87

Table 15
Normalized results of the pinched cylinder
Present formulations Other formulations
Nodes
per side MIXED DISP 4-SRI 4-RSDS 4-DKQ
5 39.9 37.5 37.3 46.9 63.6
9 76.3 75.6 74.7 79.1 95.1
17 93.5 92.7 93.5 94.6 101.6

do not converge efficiently in this problem, except the discrete Kirchhoff formulations. The
present formulation is among the best performers of the bilinear elements which account for
shear deformation.

Z8. Effects o f mesh distortion

To study the effect of mesh distortion on the performance of the proposed element, two
problems are examined: a coarse mesh modeling a simply supported plate and a cantilevered
beam modeled with one row of elements.

7.8.1. Simply supported plate

A simply suppOrted plate under a central point load is modeled with a coarse 2 x 2 element
mesh (Fig. 11). Beginning with a square mesh, the mesh is distorted by moving the middle
node along the diagonal of one of the elements (a distance A in both the x- and y-directions).
The results are compared with the T1 plate bending element [16], and normalized with the

1.2-

1.0-
~'.'--- 4-RSDS

o 0.8 Present Yl
g

0.6

.i..,,

0.4
0
Z
0.2

0 20
Number of Elements per Side
Fig. 10. Description and results of the pinched cylinder with end diaphragms. Symmetry is used and only one
eighth of the cylinder is modeled. Present results are for the mixed formulation.
88 J.C. Simo et al., Stress resultant geometrically exact shell model. Part H

10-

g-

O Present

6-

r'-4 1
4

S~rnply-SupponN
E-4

0 I !
0 0.2 0.4 0.6

Distortion of Middle Node

Fig. 11. Effect of mesh distortion on the results of a simply supported plate, subjected to a central bending load.
Due to symmetry, only one quadrant is modeled. Present results are for the mixed formulation.

converged finite element solution in Table 16. The performance exhibited by the present
element is excellent. Even for the ease in which one of the elements exhibits a negative
Jacobian (clearly an unacceptable situation), 88% of the value obtained for the undistorted
~rnesh is retained.

7.8.2. Cantilevered beam

A cantileve'red beam under three different loading conditions, is modeled with a row of ten
elements, which are arbitrarily distorted as shown i,r~ Fig. 12. The results of the mixed
formulation for both meshes are compared to the analytical answers, see Table 17.

Table 16
Results of a simply supported plate
Present(MIXED) T1 plate
Distance
A , (10 -4 in) (%) (10 =4 in) (%)
0.0 ~ 8.457 100.2 8.357 99.1
0.1 8.418 99.8 8.316 98.6
0.2 8.333 98.8 8.234 97.6
0.3 8.197 97.2 8.104 96.1
0.4 8.004 94.9 7.918 93.9
0.5 h 7.746 91.8 7.654 90.7
0.6 c 7.412 87.9 7.241 85.8
aSquare mesh.
bjacobian in zero.
~Jacobian is negative.
 l.C. Simo et al., Stress resultant geometrically exact shell model. Part l l 89

F = 1.0 M 1 -- 1.0 M2 -- 1.0

Clamped Free

Fig. 12. The distorted and nondistorted meshes and three different loading conditions used on a cantilevered beam
to test the proposed element's sensitivity to mesh distortion. Present results are for the mixed formulation.

Table 17
Results of a cantilevered beam
Bending load Bending moment Twisting moment
Load type
displacement (10 -3 in) (%) (in) (%) (!0 .2 rad) (%)
Square mesh 8.0371 100.5 .012 100.0 2.9578 103.9
Skewed mesh 8.0337 100.4 .012 100.0 2.7559 96.9
Beam theory 8.0000 100.0 .012 100.0 2.8449 100.0

As can be seen from the results, the element exhibits a robust performance highly
insensitive to mesh distortion.

8. Concluding remarks

A main thrust of the present work it; to demonstrate that shell theories based upon the
classical notion of directed media of Cosserat continuum are amenable to an efficient
implementation within the framework of the finite element method. We emphasize that the
continuum formulation of the shell equations employed here is canonical as is the formulation
of the plate equations of the Mindlin-Reissner model. Although one could argue on the form
or suitability of a particular constitutive model, the momentum equations in resultant form,
and its weak formulation summarized in Table 1, are exact within the realm of the
one-director kinematic approximation.
We have shown that the present ~tress resultant model leads to numerical results in
complete agreement with the more cumbersome three-dimensional approach based on the
degenerated solid concept. An effort has been made to provide a detailed account of the
computational implementation of the model to facilitate and stimulate further comparison and
assessment. We have evaluated the proposed finite element formulation through a rather
complete set of benchmark problems reported in the literature. Concerning these numerical
results, we make the following two observations.
90 J.C. Simo et al.. Stress resultant geometrically exact shell model. Part H

(i) The mixed formulation for the membrane stresses, described in Section 5.1, was initially
constructed to alleviate the effects of the so-called membrane locking phenomenon. Interest-
ingly enough, our numerical simulations indicate that the full displacement formulation for the
membrane a_n.d bending fields does not appear to exhibit severe membrane-bending locking
problems. Yet, the performance of the mixed formulation is superior to that of the displace-
ment model in ait the benchmark cases considered. In particular, a dramatic improvement in
the response of the element is obtained with the proposed n~ixed interpolation for bending
stresses in the singularity-dominated rhombic plate test. Similar comments apply to the mixed
interpolation for the membrane field in membrane dominated problems, such as Cook's
membrane test.
(ii) For the pinched cylinder with end diaphragms (Section 7.7), the mixed element
considered in this work exhibited similar convergence behavior as the rest of the continuum
based formulations tested. By contrast, formulations employing the discrete Kirchoff concept
exhibit a clearly superior performance. This behavior seems to indicate that the problem may
be related to some form of shear locking that is not entirely alleviated by the assumed strain
method employed. As further numerical evidence in support of this conclusion, we note that if
the value of the shear modulus is decreased by a factor of ten, a dramatic improvement in the
results is obtained. Since for this thin-shell problem the shear coefficient is, effectively, a
penalty parameter d~,at enforce,,; the constraint of zero shear deformation, it would appear that
the source of difficulty in this problem should be attributed to the interpolation of shear
strains. Nevertheless, our numerical results indicate t h a t the overall performance of the
present stress-resultant based formulation is comparable, and in some instances superior, to the
best performers in the bilinear isoparametric class of degenerated solid-based elements.

Acknowledgment

This work was performed under the auspices of the Air Force Office of Scientific Research.
Support was provided by grant nos. 2-DJA-544 and 2-DJA-771 with Stanford University. This
support is gratefully acknowledged. We wish to thank Professors S. Antman, T. Hughes, P.
Krishnaprassad, J. Marsden, C. Steele and R. Taylor for many helpful discussions.

References

[1] D.N. Arnold and R.S. Falk, A uniformly accurate finite element method for the Mindlin-Reissner plate,
SIAM J. Numer. Anal. (1988) (to appear).
[2] K.J. Bathe, Finite Element Procedures in Engineering Analysis (Prentice-Hall, Englewood Cliffs, NJ, 1982).
[3] K.J. Bathe and F. Brezzi, On the convergence of a four-node plate bending element based on Mindlin-
Reissner plate theory and a mixed interpolation, in: J.R. Whiteman, ed., Proceedings Conference on
Mathematics of Finite Elements and Applications V (Academic Press, New York, 1985) 99.491-503.
[4] K.J. Bathe and F. Brezzi, A simplified analysis of two plate bending elements--The MITC4 and MITC9
elements, in: Proceedings Conference NUMETA 87, University College of Swansea, Wales, 1987.
[5] K.J. Bathe and E.N. Dvorkin, A continuum mechanics based four-node shell element for general non-linear
analysis, |nternat. J. Comput. Aided Engrg. Software 1 (1984).
[6] T. Belytschko, H. Stolarski, W.K. Liu, N. Carpenter and J. S.-J. Ong, Stress projection for membrane and
shear locking in shell finite elements, Comput. Meths. AplA. Mech. Engrg. 51 (1985) 221-258.
 J.C. Simo ¢t ai., Stress resultant geometrically exact shell model. Part H 91

[71 T. Belytschko and C.S. Tsay, A stabilization procedure for the quadrilateral plate element with one-point
quadrature, Internat. J. Numer. Meths. Engrg. 19 (1983) 405-420.
[81 T. Belytschko, J.S.-J. Ong, W.K. Liu and J.M. Kennedy, Hourglass control in linear and nonlinear problems,
Comput. Meths. Appl. Mech. Engrg. 43 (1984) 251-276.
[91 F. Brezzi and M. Fortin, Numerical approximation of Mindlin-Reissner plates, Math. Comput. 42 (175)
(1985) 151-158.
[10] B. Budiansky and J.L. Sanders Jr, On the "best" first-order linear shell theory, Prog. Appl. Mech. 20 (1963)
129-140.
[111 H.C. Huang and E. Hinton, A nine-node Lagrangian plate element with enhanced shear interpolation, Engrg.
Comput. 1 (1984) 369-379.
[12] T.J.R. Hughes, Generalization of selective integgration procedures to anisotropic and nonlinear media,
Internat. J. Numer. Meths. Engrg. 15 (9) (1980) 1413-1418.
[13] T.J.R. Hughes, M. Cohen and M. Haroun, Reduced and selective integration techniques in the finite element
analysis of plates, Nucl. Engrg. Design 46 (1978) 203-222.
[14] T.J.R. Hughes and W.K. Liu, Nonlinear finite element analysis of shells: Part I. Three-dimensional shells,
Comput. Meths. Appl. Mech. Engrg. 26 (1981) 331-362.
[151 T.J.R. Hughes and W.K. Liu, Nonlinear finite element analysis of shells: Part If. Two-dimensional shells,
Comput. Meths. Appl. Mech. Engrg. 27 (1981) 167-182.
[16] T.J.R. Hughes and T.E. Tezduyar, Finite elements based upon Mindlin plate theory with particular reference
to the four-node bilinear isoparametric element, J. Appl. Mech. (1981) 587-596.
[17] T.J.R. Hughes, The Finite Element Method (Prentice-Hall, Englewood-Cliffs, NJ, 1987).
[18] J.H. Jang and P. Pinsky, A nine-node assumed covariant strain shell element, Internat. J. Numer. Meths.
Engrg.
[1 1 K.K. Liu, E.S. Law, D. Lain and T. Belytschko, Resultant-stress degenerated-shell element, Comput. Meths.
Appl. Mech. Engrg. 55 (1986) 259-300.
[2)] R.H. MacNeal, A simple quadrilateral shell element, Comput. & Structures 8 (1978) 175-183.
[21] R.H. MacNeal, Derivation of element stiffness matrices by assumed strain distribution, Nucl. Engrg. Design
70 (1982) 3-12.
[22] R.H. MacNeal and R.L. Harder, A proposed standard set of problems to test finite element accuracy, in:
Proceedings of AIAA Conference on Structures and Structural Dynamics, Palm Springs, CA, 1984.
[23] L.S.D. Morely, Skew Plates and Structures, International Series of Monographs in Aeronautics and
Astronautics (MacMillan, New York, 1963).
[24] L.S.D. Morely and A.J. Morris, Conflict between finite elements and shell theory, Rept., Royal Aircraft
Establishment, London, 1978.
[25] EM. Naghdi, The theory of shells, in C. Truesdell, ed., Handbuch der Physik, Vol Via/2, Mechanics of Solids
II (Springer, Berlin, 1972).
[26] F.I. Niordson, Shell theory, North-Holland Series in Applied Mathematics and Mechanics (North-Holland,
Amsterdam, 1985).
[27] K.C. Parks and G.M. Stanley, A curved C ° shell element based on assumed natural-coordinate strains, J.
Appl. Mech. 53 (2) (1986) 278-290.
[281 H. Parisch, A critical survey of the 9-node degenerated shell element with special emphasis on thin shell
application and reduced integration, Comput. Meths. Appl. Mech. Engrg. 20 (1979) 323-350.
[29] T.H.H. Pian and K. Sumihara, Rational approach for assumed stress finite elements, Internat. J. Numer.
Meths. Engrg. 20 (1984) 1685-1695.
[30] J.L. Sanders Jr, An improved first-approximation theory for thin shells, Tech. Rept. R-24, 11, NASA, 1959.
[31] A.C. Scordellis and K.S. Lo, Computer analysis of cylindrical shells, J. Amer. Concr. Inst. 61 (1969) 539-561.
[32] J.C. Simo and D.D. Fox, On a stress resultant geometrically exact shell model. Part I: formulation and
optimal parametrization, Comput. Meths. Appl. Mech. Engrg. (1988).
[33] J.C. Simo, J.E. Marsden and ES. Krishnaprassad, The Hamiltonian structure of elasticity. The convective
representation of solids, rods and plates, Arch. Rat. Mech. Anal. (1987) (to appear).
[34] J.C. Simo and L.V. Quoc, A three-dimensional finite strain rod model. Part II: Geometric and computational
aspects, Comput. Meths. Appl. Mech. Engrg. 58 (1986) 79-116.
[35] J.C. Simo and L. Vu-Quoc, A beam model including shear and torsional warping distortions based on an exact
geometric description of nonlinear deformations, Internat. J. Solids and Structures (1987) (to appear).
92 J.C. Simo et ai., Stress resultant geometrically exact shell model. Part II

[36] J.C. Simo and L. Vu-Quoc, On the dynamics in space of rods undergoing large motions--a geometrically exact
approach, Comput. Meths. Appl. Mech. Engrg. (1987) (to appear).
[37] J.C. Simo and T.J.R. Hughes, On the variational foundations of assumed strain methods, J. Appl. Mech. 53
(1) (1986) 51-54.
[38] G. Stanley, Continuum-based shell elements, Ph.D. Dis~el~ation, Applied Mechanics Division, Stanford
University, Stanfc~rd, CA, 1985.
[39] C.R. Steele, Private Communication, 1987.
[40] H. Stolarski and T. Belytschko, Membrane locking and reduced integration for curved elements, J. Appi.
Mech. 49 (1982) 172-176.
[41] H. Stolarski and T. Belytschko, Shear and membrane locking in curved C o elements, Comput. Meths. Appl.
Mech. Engrg. 41 (1983) 279-296.
[42] R.L. Taylor, Finite element analysis of linear shell problems, in: S.R. Whiteman, ed., Proceedings of the
Mathematics of Finite Elements and Applications (MAFELAP 1987).
[43] R.L. Taylor, J.C. Simo, O.C. Zienkiewicz and A.C. Chan, The patch test: A condition for assessing finite
element convergence, Internat. J. Numer. Meths. Engrg. 22 (1) (1986) 39-62.
[44] L. Vu-Quoc and J.A. Mora, A class of simple and efficient degenerated shell elements, Rept. No.
UCB/SEMM-87/05, Department of Civil Engineering, University of California, 1987.
[45] O.C. Zienkiewicz, The Finite Element Method (McGraw-Hill, New York, 1978).
[46] O.C. Zienkiewic7 R.L. Ta~lor and J.M. Too, Reduced integration technique in general analysis of plates and
shells, Internat. J. Numer. Meths. Engrg. 3 (1971) 275-290.