This article was downloaded by: [130.132.123.

28]
On: 29 December 2014, At: 06:38
Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,
37-41 Mortimer Street, London W1T 3JH, UK

Journal of Structural Mechanics: An International

Journal
Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/lmbd18

Theoretical Formulation of Finite-Element Methods in

Linear-Elastic Analysis of General Shells∗
a a
SATYANADHAM ATLURI & THEODORE H. H. PIAN
a
MASSACHUSETTS INSTITUTE OF TECHNOLOGY , CAMBRIDGE, MASS
Published online: 17 Apr 2007.

To cite this article: SATYANADHAM ATLURI & THEODORE H. H. PIAN (1972) Theoretical Formulation of Finite-Element Methods
in Linear-Elastic Analysis of General Shells∗, Journal of Structural Mechanics: An International Journal, 1:1, 1-41, DOI:
10.1080/03601217208905331

To link to this article: http://dx.doi.org/10.1080/03601217208905331

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained
in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no
representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the
Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and
are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and
should be independently verified with primary sources of information. Taylor and Francis shall not be liable for
any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever
or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of
the Content.

This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
 J . Struct. Mech., l ( 1 ) 1-41 (1972)

Theoretical Formulation
o f Finite-Element Methods
in Linear-Elastic Analysis
o f General Shells"
Downloaded by [] at 06:38 29 December 2014

S A T V A N A D H A M ATLURI" and THEODORE H. H. PlAN

ABSTRACT

A systematic classification of the variational functionals whose stationarity

condiiions (Euler equations) can be used alternately to solve for the various
unknowns in a boundary-value problem in linear-shell theory is made. T h e
application of these alternate variational principles t o a finite-element as-
sembly of a shell and thus, the development of the properties of a n individual
discreie element are studied in detail. A classification of the finite-element
methods, formulated from the variational principles by systematically relaxing
the continuity requirements at the interelement boundaries o f adjoining dis-
crete elements is made.

*This work was supported by the US. Air Force under Contract F04694-67-C-0039.
CCN P008.
'*Presently with the Department of Aeronautics and Astronautics, University of Washing-
ton, Seaule.

Copyright 0 1972 by Marcel Dekker. Inc. NO P A R T o f rhis work moy be reprod!tced or

urilized in any form or by any meons, electronic or mechanical, including xerography, pho-
tocopying, microfilm, and recording, or by any information storage and retrieval system.
without the written permission of the publisher.
 S. Atluri and T. H. H. Pian

I. INTRODUCTION

With the widespread use of thin shells as load-carrying members in various

structures, the need for a reliable analysis has been greatly felt over the past
decade. Exact analytical' solutions in linear-shell theory have been possible
only in very limited cases of shells with simple geometry (such as cylinders,
cones, and spheres), shells whose boundaries coincide with the coordinate
curves used to define the shell surface, and shells without discontinuities
(cutouts), or without any attachments (stringers, etc.). T o overcome these
shortcomings and the additional complexities caused by variable thickness
and variable anisotropic properties, the use of an approximate method of
Downloaded by [] at 06:38 29 December 2014

analysis becomes necessary.

One such approximate method is the "finite-element method." A review
and a systematic classification of the finite-element methods as formulated
from the variational principles in solid mechanics are given by Pian and Tong
[I], who also discussed the application of such methods in plate-bending
analysis. The problem of a shell is more complicated because, in the general
bending behavior, the variables defining the displacement field are, in general,
coupled.
In the finite-element analysis of shells, Green, Strome, and Weikel [2] used
a n assemblage of quadrilateral or triangular flat membrane elements. Johnson
[3] compared several schemes of flat plate elements in the solution of cylin-
drical panels. In this approach, in addition to the structural idealization
errors inherent in the finite-element solution, geometrical idealization errors
are also present. For shells of revolution, the use of conical frusta as finite
elements has been attempted by Grafton and Strome [4]. The extension of the
conical element for unsymmetric deformations by means of Fourier series
was made by Percy, Pian, Klein, and Navaratna [5]. For thick shells of revolu-
tion, a ring element of triangular cross section was used, for symmetrical
deformations, by Clough and Rashid [6], and later extended for the asym-
metrical case by Wilson 171. All the abovementioned finite elements were
derived from a "compatible displacement model" which uses displacements
that are compatible both in the interior of the element as well as along the
interelement boundaries.
For curved shells with cutouts, sections of shells, and curved shells that are
not surfaces of revolution, there is a necessity for using doubly curved quadri-
lateral or triangular elements. Quadrilateral circular cylindrical thin shell
elements were developed by Cantin and Clough [8] who use five rigid body
modes and 15 straining modes of displacements for each element, by Bogner,
 Theorerical Formulation of Finite-Element Methods 3

Fox, and Schmit (91 who use 48 straining modes of displacement for each
element, and by Olson and Lindberg [lo] who use 28 modes of displacement
for each element. All these authors used a Kirchhoff deformation hypothesis
for shell deformation which reduces the displacement variables to three, viz.,
the two displacements parallel and the one displacement normal to the refer-
ence surface of the shell. Key [I I] developed a quadrilateral doubly curved
shell element which incorporates the effect of transverse shear deformation in
the shell. All the above elements were based on a "displacement model" (to be
discussed later in this paper). As has been discussed by Atluri 1121, however,
all the above elements violate, to various degrees, the criteria for the conver-
gence of the displacement solution, as discussed by Tong and Pian 1131, viz.,
I ) compatibility in displacements and slopes at interelement boundaries, 2)
choice of displacements such that constant strain states are represented, and
3) the inclusion o f a complete set of rigid-body modes in the element interior
Downloaded by [] at 06:38 29 December 2014

in the assumed displacement field.

The purpose of this paper is to systematically consider the various varia-
tional functionals whose stationary conditions can be used alternately to
solve for various unknowns in linear-shell theory. In applying these varia-
tional principles to the finite-element assembly of a shell, various "modified
variational principles" can be generated by systematically relaxing the
continuity (compatibility) requirements along the interelement boundary.
Several finite-element schemes can be generated using these modified varia-
tional principles, the unknowns in the final algebraic set of equations being
different in each case. The properties of a discrete element in each of the
different finite-element schemes are discussed. Presented are I) a "compatible
displacement model," 2) two "hybrid displacement models," 3) two "equilib-
rium stress models," 4) a "hybrid stress model" and several "mixed models."
An attempt is made to keep the formulations within the framework of a non-
Kirchhoff-type linear-shell theory, as proposed by Reissner 1141, which
includes both transverse shear stresses and couple-stress stress couples.
Specializations to a Kirchhoff-type shell theory are also indicated.

II. GENERAL THEORY OF T H I N SHELLS W I T H COUPLE

STRESSES

Referring to Fig. I, the geometry of the midsurface of a general shell is

represented by a radius vector R which is a function of two arbitrsry curvi-
linear coordinates and 5'. The base vectors a. on the midsurface and the
unit vector o, normal to the midsurface are given by
 p -
1
i

R'E',(E2,

b
+
I
S. AIluri and T.H. H. Pion

SHELL
MIDSURFACE
Downloaded by [] at 06:38 29 December 2014

Fig. I Nomenclafure for shell midsurface,

a, = dR/a? and n = a , x a,/&

where

a = det la,,l

and the metric tensors are given by

The contravariant metric tensors am6and the contravariant base vectors a"are
defined by

where

and
 Theoretical Formularion of Finite-Elemenl Methods 5

For further use, the Gauss-Weingarten relations for a surface are represented
by

where

and
Downloaded by [] at 06:38 29 December 2014

As shown in Fig. 2a, the vector of stress resultants N",and o f stress couples
M", acting on a side 5" = const can be written as

N" = Ji(NmPaP+ qmn) (6)

and
M" = Ji{n x (MmPas) P'n) + (7)

Fig. 2a Stress resultants and stress couples acting on a shell element

 S. Arluri and T. H. H. Pian
Downloaded by [] at 06:38 29 December 2014

Fig. 2b Forces acting on a differentialelement of a general shell in static equilibrium.

where NmPare the "membrane stress resultants," q" are the transverse shear
resultants, MEPare the stress couples, and Pa are the couple-stress stress
couples which were first introduced by Gunther [15].
The vectors of externally applied forces and moments per unit midsurface
area can be written, respectively, as

F = Fa, + F30
and
m = n x (mmaJ + m3n

The two vector equations of static equilibrium for the shell element depicted
in Fig. 2b become

-
E+FJ(a=O
at-
and

which are equivalent to six scalar equations that relate the 12 scalar com-
ponents of N" and M".
Consistent with the state of description of stress, the strain state of the
 Theoretical Formulation of Finite-Element Methods 7

shell is represented by membrane strains E = ~ , transverse shear strains y,,

curvature strains K , ~ , and couple-strain strain couples I,. These strains will,
in turn, depend on an appropriately chosen description of the state of defor-
mation of the shell midsurface. Thus, the midsurface strain vector E, and the
curvature strain vector K~ can be difined as

The internal energy of the shell per unit midsurface area can be then ex-
pressed consistently as
Downloaded by [] at 06:38 29 December 2014

The translational displacement vector u and the rotation vector 4 are then
introduced in order to be able to write the virtual work per unit midsurface
area for the external surface loads and moments as

Making use of Eqs. (10) and (I I), one can show following Reissner [14], that
for the virtual work principle to be satisfied, the following relations must
hold :

and

Hence, within the realm of linear theory,

and
 8 S. A ~ I u r and
i T. H. H. Pian

The individual components of strain are derived from

Since the four strain vectors and K~ are derived from only two vectors u and
#, the two "compatibility relations" for strain measures can be immediately
observed to be
Downloaded by [] at 06:38 29 December 2014

and

Noting the form of Eqs. (10) and ( I I) and Eq. (22) and (23), it can be ob-
served that if the mathematical analogy

N" = S"#K# and M" = Foep (25)

is observed, the equilibrium equations in the absence of external loads and

moments (lo) and (I I ) and the compatibility conditions (22) and (23) can be
derived from one another. This analogy is generally referred to as the "static
geometric analogy."
Just as the strains were derived from two deformation vectors u and #, the
stresses can be derived from analogous stress functions F and H such that

and
 Theoretical Formulation of Fiuire-Element Merhods

where the further analogy

is identified. Hence, one can write

u = u'a, + wn
4 = o x (pa,) + Qn
H = H"a, + Kn
F = P ( a , x n) + Jo
Downloaded by [] at 06:38 29 December 2014

The expressions for E.@ etc., in terms of u and 4 and for Napetc., in terms of F
and H are given in the Appendix.
In a Kirchhofl-type shell theory the transverse shear strains y, and the
stress couples P' are assumed to be zero, which leads to the assumptions

and

From Eqs. (33) and (34) one obtains relations between 4 and u, and F
and H, respectively. Thus u and H become the only independent displace-
ment and stress function vectors, respectively. The relations between
and u, and N a p and H, respectively, for a Kirchhoff-type shell theory are
+
listed in the Appendix where the usual definitions E , , = E , , = f ( c , , E,,)
and N1' = N" = f ( N L Z+ N Z 1 )are employed.

Ill. A GENERAL VARIATIONAL PRINCIPLE I N SHELL

THEORY

Neither the two vector equilibrium equations f o r the four vector stress
resultants nor the two vector compatibility conditions for the four vector
strain resultants can be solved independently. These equations need to be
 10 S. Atluri and T. H. H. Pian

supplemented by introducing the constitutive relations. In order to accom-

plish this, it is convenient to postulate the existence of the stress potential A
and the strain potential B such that

A = A" + A", A;; B = B" + B," + B; (35)

where the subscript M refers to "membrane energy," B to "bending energy,"
and the superscripts m and s refer to the parts of bending energy contributed
by moments and transverse shear, respectively. The stress and strain poten-
tials are defined by
Downloaded by [] at 06:38 29 December 2014

and

and
 Theoretical Formulalio~rof Fillire-Element Methods 11

With these definitions, one can now write the shell theory analog of the
most general functional n, as was done for three-dimensional elasticity by
Washizu [16, 181, Hu [17], and Reissner [19], where

where C, is the midsurface edge curve o n which stress vectors N and m are
prescribed, and C, is the midsurface edge curve where deformation vectors u
Downloaded by [] at 06:38 29 December 2014

and 4 are prescribed. By setting the variation 672, with respect to each of
the quantities N", M", e,, K ~ ,U, and 4 equal t o zero, one obtains as Euler
differential equations: I) the equlibrium equations, viz., Eqs. (10) and (I I),
2) the constitutive relations, viz.,Eq. (44), 3) the strain-displacement relations,
viz., Eqs. (20) and (21), and 4) the boundary conditions

N = N;M = M on C, and u = ii; 4 = 5 on C, (45)

I V . T H E S T A T I C G E O M E T R I C ANALOG O F T H E GENERAL
V A R I A T I O N A L PRINCIPLE

From the static geometric analogy, described earlier, a variational func-

tional that uses the equations for stresses in terms of stress functions, and the
complementary strain energy density B can be written immediately as

In the above functional, e., K., F, H, N', and M' all can be varied independ-
ently; however, N; and M; are the particular parts of the solution corre-
 12 S. Arluri and T. H. H. Pion

sponding to given external loads and hence cannot be subjected t o variation.

Thus, setting the variation Sn*,equal to zero leads in addition t o the Euler
equations: 1) the compatibility conditions (22) and (23), 2) the constitutive
relations (38), 3) the stress-stress function relations (26) and (27), and 4) the
boundary conditions

In using the variational functionals n, and nf in connection with a

Kirchhof-type shell theory, the constraint relations between 4 andu(33). the
relation between F and H (34), respectively, should be observed.
In the following, several subclasses of the general variational principles
hn, = 0 and an*, = 0. and their applications to finite-element theory are
Downloaded by [] at 06:38 29 December 2014

indicated.

V. THE F I N I T E - E L E M E N T V A R I A T I O N A L T E C H N I Q U E

A. The General Approach

It should be noted that in the functional n, given by (44), the quantities u,

4 , E,, K,, N", and MEcan be varied independently, and that in the functional
nf given by (46), the quantities F, H. N", ME, E,, and K, can be varied
independently. In practice, for a given material of a shell, the constitutive
relations are known; hence, one can find N" and M" if E, and K, are known,
and vice versa. Several subclasses of the two general variational functionals
can be constructed by assuming apriori some of the shell theory relations that
would otherwise follow from the general functionals. Thus, for instance, by
assuming a priori the stress-strain relations and the strain displacement
relations, the functional n, can be used t o find, through a variational tech-
nique, the displacement field generated in the shell in equilibrium under the
action of applied loads. By making proper assumptions a priori in the two
functionals n, and n:, the solution of any boundary-value problem of a
shell can be reduced to finding, from the stationary conditions of the spe-
cialized functionals, any vector pair of independent variables out of the
possible pairs: (u; 4 ) ; (e,; K,); (N", M"); (F, H); (u,. Ma), and (+,, N") subject
to the given boundary conditions. However, a special form of the variational
principle as indicated later, can be used to solve for the pairs (u, 4 ) and
(N", hl") simultaneously. Once these independent variables are solved for, the
 Theoretical Formrrlation of Finite-Eleme~rfMethods 13

other shell variables can be deduced from the relations that were considered
a prior;.
In the finite-element procedure, the shell domain is divided into a finite
number of regions called discrete elements: which are, in practice, well-
conditioned triangles o r quadrilaterals in shape for general shells o r meridio-
nally curved frusta for shells of revolution. Mathematically, the finite-element
technique consists of solving for the alternate sets of independent variables a s
indicated earlier, a t a finite number of control points called "nodes" of each
discrete element; these nodes are generally located at the boundaries of each
element, but may also be in the interior of each element. The values of the
above mentioned basic unknown variables in the interior of each element are
interpolated, in a predetermined fashion, in terms of their respective nodal
values. However, in choosing these interpolation functions in the interior of
Downloaded by [] at 06:38 29 December 2014

each element, there are several criteria regarding the distribution of a given
unknown quantity a t the interelement boundary s o as t o maintain "com-
patibility" (in a broad sense) with the neighboring element, so that the
respective variational functional can be defined for the finite-element assem-
blage of the shell a s the sum (extending over the number of elements) of the
value of the functional for each individual element. However. such interele-
ment "compatibility" o r "continuity" conditions can be relaxed by "modi-
fying" the variational functionals by using a Lagrangian multiplier technique
a s discussed by Pian and i'ong [I] and t o be discussed further in this paper.
Additional criteria t o insure the monotonic convergence of the solution, a s
the size of the discrete element is progressively reduced, have t o be taken into
account.
In the following subsections, several alternate schemes of finding the basic
unknown variables, a s indicated earlier, a t the nodes of a finite-element
assembly of a shell are examined.

6 . Displacement Models

Here, the finite-element models that can be constructed through the initial
assumption of a continuous displacement field in the interior of each element
are studied. Since the description of the state of deformation in a KirchhotT-
type theory differs from that in a Reissner-type theory, the basic differences
arising, therefore, in the finite-element technique using the above theories are
pointed out.
If in the functional n, of Eq. (44), the strain-displacement relations Eqs.
(18) and (19) and the stress-strain relations Eq. (39) are assumed a p r i o r i , then
the functional n, reduces t o
 14 S. A h r i and T. H. H. Pion

In the above C, and C,,are the edge curves of the shell where the stresses and
displacements are actually prescribed. However, it should be remarked that in
extending the function n, to an assemblage of finite elements such as ABCD
(see Fig. 3 ) u can be recognized as the displacement field in the interior of
each element, C,,refers to the interelement boundary, i.e., line ABCDA, ii is the
interelement boundary displacement, and the quantitiesm and Ff are the line
loads that are present along ABCDA, especially when ABCDA o r a part of it is
aligned with the edge of a shell.
Downloaded by [] at 06:38 29 December 2014

Fig. 3 A n isolated discrete element ABCD.

1. Compatible Displacmeot Model (CDM). If the displacement field in the

interior is assumed in such a way that interelement boundary displacements
are compatible, the line integral on C, in Eq. (48) vanishes for each element.
Such a model is termed a "compatible displacement model."
Since the deformation state in a Reissner-type theory is described by two
independent vectors u (translation) and ( (rotation), and since the deforma-
tion state in a Kirchhoff-type theory is defined b y u alone, the appropriate
generalized coordinates at each node A, B, C, and D (see Fig. 3 ) are:
 Theorelical Formularion of Finite-•’It-men1 Merhods

Reissner-type theory

4=n x (pa,) + Rn
KirchhoK-type theory
Downloaded by [] at 06:38 29 December 2014

Thus, in a Reissner-type theory, the interpolation function that is used t o

describe the interior displacement field (u, 4)is such that u and 4 are linear
along any edge, say, BC in Fig. 3 , such that they could be determined uniquely
from their respective values a t nodes B and C. This automatically insures
interelement displacement field continuity, whereas, in a Kirchhoff-type
theory, the displacement field should be so chosen that not only u%nd w a r e
continuous across the interelement boundary, but also the first derivatives of
w should be continuous so that the rotations 4, would be continuous across
the interelement boundary.
If the displacement fields are so chosen, the total potential energy for the
finite-element assemblage of the shell, using discrete elements of the CDM
type, is denoted by (n,),,, and can be written as

where a, refers to the area of the tnth discrete element, C , " refers to that
portion of the edge curve of the mth element where stresses are prescribed,
and the summation extends over the N elements of the system.
Henceforth, let v represent the displacement field in a general sense, to
mean either u, 4 in a Reissner-type theory o r u in a KirchhoE-type theory. Let
the matrix of the assumed displacement field for each element be
 16 S. Arluri and T. H. H. Pian

where [U,] is a matrix of appropriately assumed functions in coordinates 5'

and (a} are undetermined parameters. As pointed out by Melosh [20], and
Tong and Pian (131, although it is not necessary from a convergence point of
view, it is desirable to include a complete set of rigid-body displacement
modes in the set of functions U , . However, in general, it is extremely difficult
to include these rigid-body modes and yet satisfy the interelement com-
patibility. The inclusion of such rigid-body modes will be considered in the
subsequent discussion on the hybrid displacement models.
The corresponding strain distribution for each elements is
Downloaded by [] at 06:38 29 December 2014

Using Eqs. (52) and (53) in Eq. (51), one obtains

where

and [C] is the elastic constant matrix.

The values of the nodal generalized displacements defined in terms of the
local coordinate system for each element can he expressed as

If one takes the same number of displacement parameters as the nodal

coordinates, the transformation matrix [ B ] is square. Hence,

Using Eq. (59) in Eq. (54), one can write

 Theoretical Formulation of Finite-Element Merhods

where, from usual definitions,

[K,] = stiffness matrix for the mth element

= [JIT [ H I ][Jl (61)

and
{Q.] = element nodal load matrix for the mth element
= [JIT {FI + St} (62)

The element nodal displacements { q ) for different elements, which are not
independent, can be related to a column of independent global displacements
( q * } thus:
Downloaded by [] at 06:38 29 December 2014

where [J*] includes the effect of transforming from local coordinates for in-
dividual elements to the global coordinates for the element assembly. Thus,
Eq. (60) can be written as

where

and

If some of the global coordinates are prescribed, Eq. (64) can be rewritten as

where { q : ] is unknown, {q:} is prescribed, ( F : } is given, and {F:] is un-

known. Then the stationarity condition of the functional in Eq. (67) leads to
 18 S. Atluri and T. H. H. Pian

which can be solved for (9:). Then one obtains

A doubly curved quadrilateral element for a general shell of revolution

based on the above model was developed by Atluri [I21 and successfully
tested in the program STACUSS by Kotanchik [21].

2. Hybrid Displacement Model-: (HDM-I). The condition of interelement

compatibility on the choice of displacement field for each element can be
relaxed by choosing an arbitrary displac~mentfield in the interior of each
element and including the Lagrangian multiplier terms
Downloaded by [] at 06:38 29 December 2014

for each element in the functional of Eq. (51); where u,(4J denotes the
displacement (rotation) on the boundary which results from the displacement
(rotation) field assumed for the interior of the element, ii(& is the interelement
displacement (rotation) which can be prescribed independently of u,(+.)
such that it is determinable uniquely in terms of the generalized nodal
displacements and the stress resultants N, and M, along the interelement
boundary of element m playing the role of Lagrange multipliers. Adding the
term in Eq. (70) to Eq. (51). the modified functional can be written as

It should be noted in the above equation that the independent variables in

each element are u, 4, N,, and M,, whereas the prescribed interelement
displacement fields ii and 6 are not independent for each element.
Since the variations Su, 6 4 , 6NL, and 6ML are arbitrary and independent
for each individual element, it can be shown that the Euler equations for the
functional in Eq. (71) are

aNZ
-+F&=O o n a,,,
JY
 Theoretical Formulation of Finite-Element Merhods 19

JM"
-
Jr+ a. x N' +mJi = o on a,
-
u, = ii; 4, = 4 on C, (74)
N, = N,; M, = M, on C, (75)
N, = A; M. =R on C, (76)

Equations (72) and (73) state that the stress developed by the assumed interior
displacement field satisfy the static equilibrium for each element. The condi-
tion that the interelement displacements generated by the assumed interior
displacement field match the prescribed interelement element displacements
along C, is represented by Eq. (74). The fact that the interelement stresses
Downloaded by [] at 06:38 29 December 2014

generated by the interior assumed displacement field match the prescribed

Lagrangian stresses is represented by Eq. (75). Finally, for edges CSmalong
which there are prescribed (actually present) stress resultants, Eq. (76) states
that these prescribed stresses are matched by the edge tractions which result
from the assumed internal displacement field. The fact that the Lagran-
gian stresses are assumed independently for each element indicates that stress
discontinuities are still present across the interelement boundaries in this
formulation.
Since there is considerable latitude in the selection of a displacement field in
the interior, one selects: I) a six-mode rigid-body displacement field and 2) a
general linearly independent n mode straining displacement field. Thus

where [U,] and [U2] represent the straining and the six rigid-body modes,
respectively. The strains can be represented as

since the rigid-body modes generate zero strains. Likewise the Lagrangian
multiplier terms at the boundary of each element can be chosen in two parts:
a set corresponding to the homogeneous solution, and a set corresponding to
the particular solution, respectively, of the equilibrium equation. Thus, for
sach element, one can write
 20 S. Atluri and T. H. H . Pion

where [ R , ] and [ R 2 ] correspond to homogeneous and particular solutions,

respectively. As will be shown later, it is better to have the dimensions of
{ a ) and ( a ) in Eq. (79) equal to those of { P } and 1 8 ) . respectively, as in Eq.
(77).
\ ,
Further, the prescribed edge displacement field is represented in terms of a
finite number of nodal displacements { q } as

Substituting Eqs. (77). (79), and (80) in Eq. (71), one can show that

-
Downloaded by [] at 06:38 29 December 2014

( X P ) H D M -=
I [ f l P i [ H ~ l { B ) lBJ{Fl} - L B J { F-
~ LPIIPll{a}
m

- uw2iw- IBI[P,I{@I + L ~ I [ C ~ +
I { L~~}I [ G , I { ~ )(81)
)

where

J co ( N L , ~ +. ML.O.) ds = I B J [ ~ ~ +
I {m~P) 2 i { a ) + L I ) I I P ~ I W } (84)

and

In the above [C] is the elastic constant matrix, and the matrices [ P I ] and
[ P 3 ]are square because of the equality of the dimensions of { a ) and {/I], and
{ a ) and { p ) , respectively. Since { a ) , { P I , { a ) , and {/?I.are independent for
each element, taking the variations with respect to them leads to the equa-
tions
 Theoretical Formulation of Finite-Element Methods 21

{B), {a), and { d } can each be expressed in

Using Eqs. (86) through (89). {/I},
terms of {q} only. Thus,

Hence, one can identify the element stiffness and load matrices as

and
Downloaded by [] at 06:38 29 December 2014

Thus, the unknowns in the final set of matrix equations are the generalized
nodal displacements.
The above model was used by Tong [22] to solve plate-bending problems. A
doubly curved quadrilateral element for a shell of revolution wad developed
by Atluri [I21 and successfully tested in the program STACUSS by Kotanchik
[2 1I.

3. Hybrid Displacement Model-2 (HDM-2). The requirement of inter-

element displacement compatibility can also be relaxed, as suggested by
Yamamoto [23], by introducing Lagrangian multiplier terms, such that the
generalized displacement coordinates of the system are the integrated
averages of the edge displacements rather than the nodal displacement as in
the earlier two models. The functional for this model can be written as

where N, and M, are the Lagrangian multiplier terms, which can be inter-
preted physically as the edge tractions for each element, and are chosen for
each discrete element as

The parameters { A } can be interpreted as the generalized nodal forces for

 22 S. Arluri and T.H. H. Pian

each element and the functions [@]are such that the edge traction { T ) of any
particular kind for a particular element could be determined uniquely in
terms of the relevant quantities {I)at the nodes pertaining to the edge of the
element in question. The element displacement field which need not satisfy
the interelement compatibility can be chosen arbitrarily for each element as

where [ U , ] and [U,] refer to straining and rigid-body modes, respectively.

Neglecting, for the present, the possibility of any actual edge loadings A and
R for each element, one can write Eq. (93) as
Downloaded by [] at 06:38 29 December 2014

where, as before,

and

From Eq. (97), by analogy, the generalized nodal displacements correspond-

ing to generalized nodal forces {A) can be defined as

and thus are the weighted integrals of the boundary displacements. Hence,
maintaining the compatibility of these generalized displacements does not
guarantee compatibility along the entire boundary.
Since { p } and {B) are the only independent variables for each element, the
stationarity conditions for the potential in Eq. (93) give
 Theoretical Formulation of Finite-Element Methods 23

and

It can be observed that Eq. (100) represents six equations of static equilibrium
for the generalized forces around the element. Thus, Eq. (100) can be re-
written as

and thus,
Downloaded by [] at 06:38 29 December 2014

Thus, the independent generalized nodal forces for each element are (1,) only.
Using Eqs. (99) and (100) one can express Eq. (93) entirely in terms of the
unknown forces (1,). Thus using the generalized forces (1:) for the finite-
element assembly, and since these can be subjected to independent variation,
one obtains the final set of equations of the form

Thus, even though the model has been constructed with an initial assumption
of element displacements, the unknowns in the final set of equations are the
redundant generalized nodal forces {A:}. Since this is a matrix force method,
considerable care should be exercised in the choice of the basic unknown
forces {I:} such that the matrix equations are not ill-conditioned. It should be
noted that in this model, even though the interelement displacement com-
patibility is satisfied only on a n average, interelement stress continuity is
maintained. To the authors' knowledge, no shell element has been developed
so far based on this model.

C. Stress Models

In this subsection, finite-element models that could be constructed through

the initial assumption of equilibrium stress field within each element are
studied. First, a stress model that can be constructed by applying the static
geometric analogy to the "compatible displacement model" is presented. In
this approach, the unknowns to be solved from the final set of equations are
the nodal values of the stress functions. Two other models, involving the
 24 S. Alluri and T. H. H. Pion

initial assumption of a stress field are also presented: in both of the models,
the unknowns to be solved in the final set of equations obtained from the
governing variational statement for the assembled structure are the gener-
alized nodal displacements.

1. Equilibrium Stress Model-1: (ESM-I). As discussed by Pian [24], this

model can be constricted directly from the compatible displacement model
through the static geometric analogy as explained in Eqs. (25) through (28).
where it has already been shown that there exists a direct analogy between the
displacement vector u and the stress function vector H, between the rotation
vector @ and the stress function vector F, between the stress-resultant vector
N' and the curvature strain vector Kg,and between the stress-couple vector
M" and the stretching strain vector eu. Thus. just as the C D M (where dis-
Downloaded by [] at 06:38 29 December 2014

placements are assumed) is used to solve for a displacement boundary-value

problem in shells, an analogous model using stress functions can be used to
solve a stress boundary-value problem. If the analogy between displacement
variables and the stress function variables is observed, i t then follows that for
similar types of functions, as are used in the C D M , the stress functions would
be continuous across the interelement boundary. Even though displacement
compatibility is maintained in the C D M , strains, however, are not necessarily
continuous across the interelement boundary. By direct analogy, it then
follows that stresses will not be continuous, but nevertheless stress equilibrium
conditions are satisfied at the interelement boundary in ESM-I, if the stress
functions are analogous to the displacement functions as in C D M . Also, the
unknowns in the finite-element formulation of ESM-I, are by analogy, the
nodal values of stress functions. When once these are solved, the stress and
strain distribution must be obtained from the derivatives of stress functions,
while the displacements can only be calculated by integrating the stress
displacement relations. In view of the approximate character of the stresses
provided by the finite-element analysis, the deflection obtained by integration
is, in general, dependent upon the chosen integration path; hence it is not a
unique solution.
There are other drawbacks to this equilibrium model. In general, it is a
difficult task to include the externally applied loads and moments on the shell
surface in this formulation. Also it is often obscure to interpret physically the
stress boundary conditions at the edges of the shell in terms of stress func-
tions.

2. Equilibrium Stress Model (ESM)-2. If in the general variational func-

tional of Eq. (49, the equilibrium equiation for the shell, Eqs. (10) and (I I),
are assumed apriori, the functional can be shown to reduce to
 Theoretical Formulation of Finite-Element Methods

In the case of a finite-element assembly of a shell, Eq. (103) can be written

as
Downloaded by [] at 06:38 29 December 2014

m
In Eq. (104), can be interpreted as the stress at the interelement boundary
and ii as the interelement boundary displacement. If the stress field in the
interior of each element is selected such that it generates interelement bound-
ary tractions that are compatible with the neighboring element, the integral
over Csmin Eq. (104) vanishes. Hence

Thus, the stress field that satisfies not only equilibrium, but also interelement
compatibility can be chosen for each element as

where the set of functions [ R , ] can be derived from stress functions (to
represent the homogeneous solution of equilibrium equations) with unknown
parameters (a) and [R,] represents any stress field that is statically equivalent
to the applied loads and thus is considered as known. The corresponding edge
tractions are

For stress continuity along the interelement boundary, these edge tractions
should be representable uniquely from a set of generalized loads a t a finite
number of control points along the boundary of the element in question.
 26 S. Atluri and T. H. H. Pian

Thus, (7')should be expressible as

where {A} are the generalized nodal loads. From Eqs. (107) and (108) there
exists a unique relation

-
{A) = [$l-'[Rl.lb)
[Gll{al
+ [91-'{R21
+ {G2} (109)

The matrix C, obviously is a square matrix. The corresponding generalized

edge displacements {q} are defined from the equation
Downloaded by [] at 06:38 29 December 2014

and thus represent weighted integral averages of the boundary displacements.

Using Eqs. (106) and (1 lo), one can rewrite Eq. (105) as

where D is the matrix of elastic constants. Carrying on the integration, one

can then write

Taking variations with respect to {a}, which are independent for each element,
leads to an equation for a's in terms of q's. Thus, the complementary energy
of the entire can be expressed in terms of the system coordinates {q*}.
Setting bn, = 0 then leads to the usual matrix equations in {q*}.
T o the authors' knowledge no shell element has been developed so far for
this model.
 Theoretical Formulation of Finite-Element Methods 27

3. Hybrid Stress Model-1 (HSM-I). This hybrid stress formulation is an

analog of the hybrid displacement HDM-2 model. Pian [25] used this HSM-I
formulation first in plane stress and plate-bending problems.
In this method, an equilibrium stress field is assumed in the interior of each
element and stress equilibrium along the interelement boundary need not be
satisfied. Stress continuity along the interelement boundary is satisfied only
in an average sense by introducing Lagrangian multipliers along the interele-
ment boundary. These Lagrangian multipliers, which can be physically
interpreted as the interelement boundary displacements, are chosen such that
they are compatible with the neighboring elements. The functional for the
HSM-I model can be written as
Downloaded by [] at 06:38 29 December 2014

where u,, 9, (Lagrangian multipliers) can be chosen as

where [L] is a matrix of interpolation functions such that the displacement on

any particular edge of the element is interpolated uniquely in terms of the
nodal values {q} along that edge; these generalized nodal displacements are to
be regarded as global, a "master" in the sense that they are common to and
apply to all elements which share any given node. For the assumed stress field
in the interior of each element. one may write

where [ R , ] represents the homogeneous solution with unknown parameters

( a } and { R , ) any particular solution. The boundary tractions generated by
this stress field can be written as

( T ) = [R1,1{4 + {R2J (116)

The generalized nodal forces ( r } can be identified from the expression
 and thus are equal to the integrated averages of the edge fractions. Thus, the
functional for the finite-element assemblage becomes

which can be written as

Downloaded by [] at 06:38 29 December 2014

Taking variations with respect to (a}, one gets the equation

[ J l l I b l - [ G x 9 } - {J22} = 0 (121)

Solving for {a} from Eq. (121) and substituting in Eq. (120) one obtains

where

[K] = stiffness matrix for the mth element

= [Gll [J;,'lT [GtIT (123)
and
{Q} = nodal loads for the mth element
= [Gll [J;:I {J22} + {G22} ( 1 24)

Here, a comment about the particular solution (R,) in Eq. (115) is necessary.
Tong and Pian [26] have shown, for plate bending, that when stress modes are
so chosen that the polynomials in (R,) are of the same degree as in [R,], then
 Theoretical Formrdarion of Finite-Element Methods 29

the finite-element solutions will be independent of the particular solution.

Tanaka [27] developed a triangular shell element, which is part of a cone,
based on the above model, and which was tested successfully in the program
STACUSS by Kotanchik [21].

D. Mixed Models

Thus far, finite-element models t h ~ s e don either the assumptio n of

a displacement field or the assumption of a stress field within each element
have been considered. In shell theory, it is possible t o construct finite-
element models bas:d on a simultaneous assumption of a suitable combi-
nation of displacement and stress fields which are mutually independent.
Downloaded by [] at 06:38 29 December 2014

These possibilities are explored in this subsection.

1. Models Based on Reissner's Variational Principle. If in the variational

functional given by Eq. (49, one assumes a priori that the stress-strain
relations (39) are fulfilled, one obtains the Reissner functional

in which u, 4, N", and M' can be varied independently. For reasons of clarity
and convenience, the Reissner functional is written in its expanded form for
Kirchhoff-type shell theory in orthogonal curvilinear coordinates as follows
 S. Alluri and T. H. H. Pion

In the integrals on C, and C,, the superscript (or subscript) n is the direction
normal to the local tangent of the boundary, and similarly the super (sub)
script s denotes the direction along the tangent to the boundary. Also, it
should be noted that the KirchhoR-reduced boundary tractions are used in
Downloaded by [] at 06:38 29 December 2014

the integral on C,.

In the finite-element analysis, stresses and displacements can be assumed
separately for each individual element. Several different possibilities arise
depending on the continuity requirements for 1) stresses and 2) displacements
at the interelement boundaries so that the function n, can be defined for the
finite-element assemblage. Pian and Tong [I] studied this type of problem in
detail for plates.
Based on different requirements, Eq. (126) for n, can be written in
different forms for finite-elcment analysis. Thus, one may write

where 2, is a constraint term of various types to allow for the various dis-
continuities of stresses and displacements a t the interelement boundaries. For
example, it can be seen from Eq. (126) that, if the displacements ( u , , u,, and w)
and the slope awlan along the normal to the boundary of the element are
discontinuous, the integral over the element area leads to a delta function.
But the functional n, can be written in the form of Eq. (127), if the stress
resultants Napand M"%re continuous across the interface, by including a
proper I, term. On the other hand, if the displacements and normal slopes
are continuous across the interface, the integral exists even if the stress
resultants Napand Mapare discontinuous.
One can verify in a similar manner that the sufficient conditions for the
 Theorericol Formolofion o/ Finite-Element Merhods 31

functional n, to be defined for the discrete element assemblage are given by

each of the following sets of conditions, and the corresponding A. terms are
indicated.
I) Continuity of us, u., and w , and Jw/Jn across the interelement boundary
even though NSPand Map may be discontinuous:

2) Continuity of N"", N'", M"", M"", and the normal derivative of M"" across
the interelement boundary, even though the quantities us,u,, w, and Jw/Jn are
discontinuous:
Downloaded by [] at 06:38 29 December 2014

3) Continuity of u , , u,, w, and M"", even though JwlJn, N " , M"" are
discontinuous:

A," = I(.""E) ds

4) Continuity of us, u,, M"", M"', and M""/dn, while w, awlan, and W Pare
discontinuous across the interelernent boundary:

5) Continuity of w, JwlJn, N"", and Nu, while u, u,, M"", and M"' are
discontinuous across the interelement boundary:

6) Continuity of w, N"", N"',and M"", while u,, u,, and Jw/Jn are discon-
tinuous across the interelement boundary:

A," =
I [Nn"u, + N"'u, + Mn"@,]ds (133)

Several other combinations are possible, but they appear to be irrelevant to

shell analysis. Again, it should be pointed out that several variations on the
lines of hybrid models presented earlier are possible in all of the six cases
presented here by relaxing the requirements on the continuity of the needed
quantities in each case through a Lagrangian multiple technique.
 32 S. Ahri and T. H. H. Pion

The Reissner functional can be further modified in shell theory as shown in

the next two subsections.

2. Mixed Model 1 (MMI). In the general functional n, of Eq. (44) consider

a priori I) the moment equilibrium equation [Eq. (1 I)], 2) the membrane
strain-displacement relations [Eq. (18)], and 3) the constitutive relations

It can be shown that if the usual definition N I L N z l = +(N1' + NIz) is

made, the moment equilibrium equations [Eq. (1 I)] contain only M u and q"
terms. Likewise if the definition el, = E,, = +(E,, + c,,) is made, the
Downloaded by [] at 06:38 29 December 2014

membrane strains depend on u alone. Under the above apriori assump-

tions, the general functional ri, can be shown to reduce to

- ICr
{N.u - (M - M ) , + ) ds -
CD
[(u - ii1.N - $.MI ds (135)

in which u and M" can be subjected to variation and whose Euler equations
are I) the stress-resultant equilibrium equation [Eq. (lo)], 2) the curvature
strain-rotation relations [Eq. (19)], and 3) the boundary conditions u = 8 ;
q5 = 4 on C , and N = N, M = Kl on C,.Thus, in the finite-element method,
the displacement vector u and the moment resultants Maa (and through them
q3 are assumed independently for each element. Note from the first two terms
in the integral of nu, that it is defined if the displacements are discontinuous,
even if the stress q" are discontinuous across the interelement boundary. O n
the other hand, if w is discontinuous across the interelement boundary, the
integral is defined if only the stresses are continuous. Thus, for a finite-
element assemblage,

where 1, = 0 if u is continuous across the interelement boundary, and

 Theorericol Formulation of Finite-Element Merhods 33

I
A, = q' w,ds if u, is discontinuous but q' is continuous across the interele-
ment boundary.
Several variations of the method on the lines of the hybrid methods
presented earlier are possible by relaxing the various continuity conditions
through a Lagrange multiplier technique. A model for a triangular shallow
shell element, with linear variations of u and MZBin each element, appears to
be firs: presented by Prato [28].

3. Mixed Model 2 (MMZ). If, in the general functional n, of Eq. (44)one

assumes a p r i o r i I) the stress-resultant equilibrium equation [Eq. (lo)], 2) the
curvature strain-displacement relation [Eq. (19)], and 3) the constitutive rela-
tions
Downloaded by [] at 06:38 29 December 2014

then the functional n, can be shown to reduce to

- I, {(N - N).u + M.+}ds - {(4 - $).M - ii.N)} ds (138)

in which 4 and N' are the independent variables, whose Euler equations
are I) the equilibrium equations for moments [Eq. ( I I)], 2) membrane strain-
displacement relations [Eq. (18)], and 3) the relevant boundary conditions. If,
further, the usual assumption N1' = N Z ' is made, then one can find that

In the finite-element analysis, the rotation vector 4 and the stress resultant
vector N' are assumed independently for each element. The stress resultants
NaB and q" are chosen in two parts: one to satisfy the homogeneous equilib-
rium equations (using stress functions as indicated in the Appendix) and the
other, any particular solution.
If a Reissner-type theory is used, the vector 4 is independent of u, and
since no differentiations of q' and 4 are involved in the functional nFb. the
integral and hence the functional n,+ are defined even if q" o r 4 o r both are
 Downloaded by [] at 06:38 29 December 2014

TABLE 1
Classification of Finite-Element Methods i n Linear Shell Theory

Variational Assumed Inside Along Interelement Unknowns i n

Model Princi~le Each Element Boundarv Final Eauations

Compatible displace- Minimum potential Smooth displacement Continuous displace- Nodal dis~lacemcnts
ment energy field (u. $5) ments and rotations and rotations
Hybrid displacement Modified potential Smooth displacement Assumed boundary Nodal displacements
I energy field (u. 4) tractions. assumed and rotations
\ compatible displace-
ments and rotations
Hybrid displacement Modified potential Smooth displacemont Assumed equilibrating Boundary redundant
I1 energy field (u. $5) boundary tractions forces
Equilibrium stress Complementary Smooth stress func- Continuous stress Nodal values of
I energy lion vectors F. H functions F. H stress funclions
Equilibrium stress Modified comple- Equilibrating Equilibrium bound- Generalized nodal
11 mentary energy Stress resultants ary lractions displacements and
and moments rotations
Hybrid stress 1 Modified comple- Equilibrating Compatible displace- Nodal displacements
mentary energy stress resultants ments and rotations and rotations
and moments
Mixed Reissner's varia- Smooth displacements. Several alternatives
tional principle rotations. stress are possible. See Eqs
resultants, and couples (128) through (131)
Modified varia- Smooth displacements. Several alternatives
tional principle I stress couples are possible. See Eq. (136) .q
%
Modified varia- Smooth rotations, Several alternatives >
tional orinci~le
. . 2 stress resultants possible. See Eq. (140)
2
3
 Theorerical Formrrlarion of Finire-Element Methods 35

discontinuous at the element interface. If a Kirchhoff-type theory is used, the

first term in the integral in Eq. (139) is

Thus, as has been discussed before, the functional n,,., using a Kirchhoff-type
theory can be written as
Downloaded by [] at 06:38 29 December 2014

where A, = 0 if w is continuous across the element interface and 2, = jqSw,ds

if w is discontinuous but q' is continuous across the interface. To the authors'
knowledge no such element has been developed for shells.

SUMMARY

A general first-order shell theory including tranverse shear strains and

couple stress-strain couples, a general variational principle and its static
geometric analog are discussed in tensor form. A systematic classification of
the various possible finite-element models, depending on an initial assump-
tion of the element displacement field (displacement models) or the stress
field (stress models), or a proper combination of both (mixed models) is made,
and is summarized in Table I for convenience.

APPENDIX

The expressions for individual physical components of strain in terms of

the physical displacement components and for the individual physical com-
ponents of stress and moment resultants in terms of physical stress function
components are given for both Reissner- and Kirchhoff-type shell theories.
 36 S. Atluri and T. H. H. Pion

Orthogonal curvilinear coordinates (not necessarily line o f curvature coordi-

nates) are considered. In what follows, the notations

are employed.

A. Strain Displacement Relations of a Reissner-Type Theory

The relations given by Eqs. (18) and (19) can be expanded as

Downloaded by [] at 06:38 29 December 2014

One could make the usual hypothesis z , , = &, = +(E,, + E,,) and get

from which

Likewise,
 Theoretical Formulation of Finite-Element Methods 37

Thus, in the modified Reissner-type theory, when the strain tensor is made
symmetric, the only independent displacement measures are u', w, $,, and $,
the third rotation component R being expressed in terms of u" and w.

B. Strain Displacement Relations for Kirchhoff-Type Theory

Under the constraint condition the transverse shear strains y = 0,it follows
from Eq. (33) that
Downloaded by [] at 06:38 29 December 2014

Also, if the tensor E , ~is made symmetric as before,

Hence the only independent displacement measures are u" and w. The strain
displacement relations then become

= -I- du
! + - I +u - .aa w
, e l , = - TI+ au,
-- U , aa, +- w
a , 85' a,a,a€' R,, a2at a l a , atL R,,
 38 S. Afhri and T.H. H. Pion

It is to be observed that even when the Kirchhofl'assumption is made, the

couple strains 1, do not vanish identically. Thus,
Downloaded by [] at 06:38 29 December 2014

But by appropriate modifications in the constitutive relations, the effect of

couple strains can be ignored.

C. Stress-Stress Function Relations for a Reissner-Type

Theory

From the static geometric analogy as given by Eqs. (26) and (27), we can
write, in component form,

If one makes the usual hypothesis that M , , = M , , = + ( M I , + M , , ) and

 Theoretical Formularion of Finite-Elemenr Merhods

N 1 2= N , , = + ( N , , + N,,), one can write

I dH, Hz da, I dH2
M,,=M2, = --
a,dtl a,a2dt1 a2dt2

from which

and
Downloaded by [] at 06:38 29 December 2014

la, a
N , , = N,, =
2 ~at1
, (3+--- (J (it,
I N , ~ F,
2 a 2 at2
- -J
R'J
(A.13)

Thus, in a modified Reissner-type theory when the stress tensor is made

symmetric, the only independent stress functions are H,, K, and F,, with the
function J being expressed in terms of (H,,K) as above.

D. Stress-Stress Function Relations for the Kirchhoff-Type

Theory

Here, under the constraint that the couple-stress stress couples P. vanish, it
follows from Eq. (34) that

Again if Ma# is made symmetric, one sees, as above, that

Thus, the only independent stress functions are H, and K. Thus,

 S . Atluri and T. H . H. Pian
Downloaded by [] at 06:38 29 December 2014

(A. 16)

REFERENCES

I. T. H. H. Pian and Pin Tong, Basis of finite element methods for solid continua, lnl.
J. Numer. Melh. Eng. 1: 3-28 (1969).
2. B. E. Green, D. R. Strome. and R. C. Weikel. A ~ d i c a l i o nof the stiffness method t o
the an3lyrls u l shell itructurcs. ASML: Pawr No. 61, AV-58. Mar. 1961.
3. C. P. Johnson. I hc anal,sir of thin shells hy a tinltc clement procedure. Ph. D. Thesis.
Dep. Civil Eng., Univ. California, Berkeley, 1967.
4. P. E. Grafton, and D. R. Strome, Analysis ofaxisymmetrical shells by the direct stiff-
ness method, AIAA J. 1 (10): 2342-2347 (1963).
5. J. H. Percy, T. H. H. Pian, D. R. Navaratna, and S. Klein, Application of the matrix
displacement method t o the linear elastic analysis of shells of revolution, Massa-
chusetts Institute of Technology, Tech. Rep. ASRL T R 121-7, Jan. 1965.
6. R. W. Clough, and Y. Rashid, Finite element analysis ofanisymmetric solids, ASCE
Eng. Mech. Div. 3.91 (EMI): 71-85 (1965).
7. E. L. Wilson, Structural analysis of axisymmetric solid, AIAA Paper No. 65-163,
Jan. 1965.
8. G. Cantin and R. W. Clough, A curved cylindrical-shell finite element, AIAA 3. 6
(6): 1057-1062 (1968).
9. F. K. Bogner, R. L. Fox, and L. A. Schmidt, A cylindrical shell discrete element,
AIAA J. 5 (4): 745-750 (1967).
10. M. D. Olson and G. M. Lindberg. Vibration analysis of cantilevered curved plates
using a new cylindrical shell finite&ment, presented at the 2nd Conf. Matrix ~ e t h o d s
in Structural Mechanics, Wright-Patterson AFB, Ohio, Oct. 1968.
11. S. W. Key and 2.E. ~eisinger,The analysis of thin shells with transverse shear strains
by the finite element method, presented at the 2nd Conf. Matrix Methods in Structural
Mechanics, Wright-Patterson AFB, Ohio, Oct. 1968.
12. S. Atluri. Static analysis of shells of revolution using doublysurved quadrilateral
elements derived from alternate variational models. Tech. Rep. SAMSO TR 69-394.
Norton Air Force Base, June 1969.
 Theoretical Formulation of Finite-Element Melhods 41

13. Pin Tong and T. H. H. Pian, The convergence of finite-element method in solving
linear elastic problems, Int. J. Solids Srruet. 3 : 865-879 (1967).
14. E. Reissner, On the foundations of the theory of elastic shells, Proc. 11th lnt. Congr.
Appl. Mech. (1964), pp. 20-30 (1966).
15. W. Gunther, Analoge Systeme von Schalen-Gleichungen, Iwenieur-Arehiv. 30:
160-186.
16. K. Washizu, Voriotional Methods in Elasticity and Plasticity, Pergamon Press, New
York, 1968.
17. H. C. Hu, On some variational principles in the theory of elasticity and plasticity,
Scintio Sinico 4 (I): 33-54 (1955).
18. K. Washizu, On the variational principle of elasticity and plasticity, Massachusetts
Institute of Technology, Aeroelastic and Structures Res. Lab., Tech. Rep. ASRL
TR 25-18, Mar. 1955.
19. E. Reissner, On some variational theorems in elasticity, in Problems of Conrinuum
Mechanics (Mushkelishvilli Anniversary Volume), 1961, p. 370 (Soc. Ind. Appl.
Math.).
Downloaded by [] at 06:38 29 December 2014

20. R. J. Melash, Basis for derivation of matrices for the direct stiffness method, AIAA I.
1: 1631-1637 (1963).
21. J. J. Kotanchik and B. A. Berg, STACUSS I : A discrete element program far the
static analysis of single-layer curved stiffened shells subjected to mechanical and
thermal loads, Massachusetts Institute of Technology, Aeroelastic and Structures Res.
Lab., Tech. Rep. ASRL TR 1469.
22. Pin Tong, New displacement hybrid finite element model for solid continua, Inr. I.
Numer. Meth. Em. 2 : 78-83.
23. Y. Yamamoto, formulation of matrix displacement method, Dep. Aeronautics and
Astronautics Rep., Massachusetts lnstitute of Technology, 1966.
24. T. H. H. Pian, in Recenr Advances in Matrix Methods of Srrucrurol Analysis andDesign,
I970 (Proc. US-Japan Seminar on Matrix Methods in Structural Analysis and
Design, Tokyo, 1969, R. H. Gallagher, Y. Yamada, and J. T. Oden, eds.); also in For-
mulorions ofFinife Elemenr Merhodsfor Solid Conrinun, University of Alabama Press,
Tusculoosa (to be published).
25. T. H. H. Pian, Element stiffness matricesfor boundarycompatibility and for prescribed
boundarv stresses. Proe. Conf.Matrix Methods in Structural Mechanics. AFFDL T R
66-80, pp. 427-479.
26. Pin Tone -
- and T. H. H. Pian. The convereence of finite element method in solvinn-
linear elastic problems, Inr. J . Solidr S f r u d . 3 : 865-879 (1967).
27. M. Tanaka. Develooment and evaluation of a trianaular - thin shell element based uoon
the hybrid assumed stress model, Ph. D. Thesis, Dep. Aeronautics and Astronautics,
Massachusetts Institute of Technoloav. . . June 1970.
.
28. C. Prato, A mixed finite element method for thin shell analysis, Ph. D. Thesis, Dep.
Civil Eng., Massachusetts lnstitute of Technology, 1968.

Received M a r c h 7 , I971