Computer Methods in Applied Mechanics and Engineering 98 (1992) 329-343

North-Holland

CMA 242

A drill rotation formulation for geometrically

exact shells
D.D. Fox and J.C. Simo
Division of Applied Mechanics, Department of Mechanical Engineering, Stanford University,
Stanford, California, USA

Received 23 November 1990

This paper considers the formulation of classical nonlinear shell models which incorporate an
additional rotation degree of freedom called the drill rotation. This class of models correspond to
convenient reformulations of classical shell theory, and not to higher order (Cosserat) theories. In the
present context, the additional local equation for the drill rotation is a constraint condition which
identifies the rotation of the mid-surface of the shell with the drill rotation. The remaining equations
are those of the classical model. Proper account of the geometry of the mid-surface in a finite
deformation setting makes the actual formulation of this additional equation nontrivial. A variational
formulation of the shell equations incorporating drill rotation is constructed by introducing an
additional Lagrange multiplier term. In sharp contrast with formulations based on the so-called Biot
stress, here the Lagrange multiplier enforcing the constraint vanishes at equilibrium, allowing for a
regularized variational principle which reproduces the proper set of shell balance equations for any
value of the regularization parameter.

1. Introduction

In [l] a formulation of finite elasticity is presented which incorporates a point-wise

independent rotation field. The introduction of this independent rotation field into the
kinematic description of the solid results in three additional rotational degrees of freedom, the
evolution of which are governed by a kinematic constraint that identifies the rotation field with
the rotation of the medium, as defined by the polar decomposition of the deformation
gradient. The analysis in this paper constitutes an extension of these results aimed at
incorporating drill rotations into nonlinear geometrically exact shell models. Formulations of
nonlinear shell theory incorporating drill rotations have important computational advantages.
In particular, problems involving the reinforcement of shells with other structural elements,
such as rods or beams, are easily resolved in a finite element context since all the elements
involved now possess an identical number of independent rotational degrees of freedom. No
matching of dissimilar rotational degrees of freedom is required.
Since only two independent rotations are required to orient a unit director field, a
distinctive feature of classical plate and shell theories is a rotation field defined in terms of
only rwo independent degrees of freedom. Rotations about the director itself - the so-called
drill rotations - are irrelevant and, for that matter, undefined in classical shell theory. In fact,
properly formulated classical shell models are invariant with respect to superposed drill
rotations (one speaks of an S’-symmetry of the model). Consequently, finite element

00457825/92/$05.00 0 1992 Elsevier Science Publishers B.V. All rights reserved

330 D.D. Fox, J.C. Simo, A drill rotation formulation

implementations of classical models must be constructed that maintain this property, so that
the underlying S’-symmetry is preserved. This is the case in the numerical implementation
described in [2,3] of the geometrically exact shell model considered in 141.
The objective of this paper is to propose an alternative formulation of classical geometrical-
ly exact shell models which incorporate at the outset drill rotations as a third rotational degree
of freedom. To accomplish this task, we adopt the same point of view as Toupin [5], and
regard classical shell theory as a generalized (Cosserat) model incorporating an independent
drill rotation which is constrained to coincide with the rotation of the medium as defined, in
the present context, by the rotation of the mid-surface. As in [l], the implementation of the
preceding idea is accomplished by considering an extended configuration space of the shell
enhanced to include a full three-dimensional orthogonal transformation which orients the
director field. The enforcement of the constraint identifying the drill rotation as the mid-
surface rotation in a finite deformation setting is nontrivial, and is shown to involve a
symmetry constraint on the director orienting orthogonal transformation. In the context of
linear shell theory, formulations of the shell equations incorporating drill rotations have been
explicitly given by Reissner [6], who appends balance of angular momentum as the con-
strained equation. This approach leads naturally to the Hellinger-Reissner type of variational
formulations best suited for linear theory.
This paper is divided into two parts. First, the structure of the underlying classical nonlinear
shell model is summarized in Section 2. The presentation in this section differs from our
previous work by the emphasis placed on the proper form of the shell hyper-elastic resultant
constitutive equations. The enhanced or extended kinematics of the shell model incorporating
drill rotations is introduced in Section 3. Here, the constraint on the drill rotation is
conveniently enforced as a constraint on the director orienting orthogonal transformation via a
symmetry condition based on certain polar decomposition. The balance equations governing
the shell model incorporating drill rotation are then obtained merely by augmenting the
balance equations summarized in Section 2 with this symmetry condition. Variational princi-
ples are then immediately constructed merely by regarding these balance equations as
Euler-Lagrange equations of a potential energy function. The resulting formulation has
several desirable features:
(1) The final variational principle consists of a Lagrange multiplier term added to the
elastic potential energy function of the classical shell model. Hence, from a computational
perspective, the incorporation of drill rotations into the classical model reduces merely to the
addition of a single term to finite element formulation of the classical model.
(2) In a finite element discretization, the director field at the nodes is now described by
three independent rotation; i.e., by a proper orthogonal transformation. Exact updates can be
immediately constructed by exploiting the closed form expression of the exponential mapping
in the rotation group, exactly as in rod and beam models; see [7]. Finally, we remark that
polar decompositions need not be performed at any stage of the implementation.
(3) The Lagrange multiplier enforcing the constraint on the director field is shown to
vanish at equilibrium. As a result, a regularization of the variational principle similar to that
presented in [8] for the linear theory, and in [l] for the nonlinear theory, can be constructed.
The computational advantage of this regularization lies in the fact that the balance equations
are recovered for any value of the regularization parameter.
We conclude the paper with a summary account of the matrix formulation of the weak form
of the momentum balance equations along with a brief derivation of the corresponding
tangent operator. These are the essential ingredients involved in a finite element implementa-
tion of present model.
 D.D. Fox, J.C. Simo, A drill rotation formulation 331

2. Classical geometrically exact shells

To motivate our subsequent developments, in the first part of this paper we summarize the
single director nonlinear shell theory following the recent presentations in [4,3]. In particular,
we outline the basic kinematics, the effective stress resultants, and the momentum balance
equations. Then hyper-elastic shell resultant constitutive equations are introduced. The
discussion of hyper-elastic constitutive equations is particularly important for constructing
variational principles for the drill rotation formulation covered in the second part of this
paper.

2.1. Summary of the shell momentum balance equations

In the geometrically exact shell theory, points x in any deformed configuration of the shell
are assumed to be defined by the following expression, often referred to as the single director
Cosserat surface assumption:

x= cp(5?7
t’> + 545’>s’> with (S’, t’)E & and 5 E 9 .

Here L&C R* defines the domain of the mid-surface parametrization and 4 = [h-, h’], where
h- < 0 and h’ > 0, defines the thickness of the shell. Note that d is not assumed to be unit;
hence, assumption (1) incorporates thickness change. It follows from (1) that configurations of
the shell are completely determined by pairs (cp, d). Therefore, the abstract configuration
manifold for the shell, denoted by %‘,is defined as

% = { @ = (cp, d) : sd+ R3 x lR31cplaod= 6 and dlddd = d”} , (2)

where a,.& and a,& are the parts of the boundary where the displacement and the director
field are specified, respectively. We shall denote by h = lldll the magnitude of the director field
and let t = 1 /Ad. At each point of the mid-surface, we define the convected or natural frame as

Following the standard convention, Roman indices run from 1 to 3 while Greek indices run
from 1 to 2. Capital letters typically refer to objects defined in the reference configuration. In
particular, {A 1, A,, A3} := { cppI, (pp2, D} is the natural basis in the reference configuration
which is defined by (cp”,0) E %. In addition, we define the surface-director deformation
gradient f and the director gradient g by the expressions

f=cp,,@Aa+dC3A3 and g=d,,@A”, (4)

where the dual basis vectors A’ are defined through the relationship A i * A’ = S i. Note that f is
the shell counterpart to the deformation gradient F in the three-dimensional theory. The
Jacobian determinant J : = det[f] is given by
--
j= j/j” , where 7’ = A, xA,*D and /=a, X a,-d. (5)

By analogy with the three-dimensional theory, we define shell right Cauchy-Green tensors
associated with f and g by the expressions
332 D. D. Fox, J. C. Simo, A drill rotation formulation

cr =f’f and cg =f’g . (6)

The component expressions for cr and cg relative to the reference dual basis {A’} are obtained
from (4) as

(7)
c, = /&Aa @A’ + &A3 @A” ,

where the mid-surface metric, the transverse shear components, the thickness parameter, the
director curvature and the couple shear components are defined, respectively, by

a A*=d.d,
aP = ‘p., . ‘p,p 7 9, = P,, - d 3
(8)
Kap = ‘p., *d., and &a = de d,, .

Let {na, 1, I;;“}, with (Y= 1,2, be the system of resultant forces acting on the mid-surface of
the shell, resolved in the natural frame according to the following component expressions:

na= nBaag+ q”d ,

1= lQa, + 13d, (9)

Here ltPa are the mid-surface membrane forces, qa are the transverse shear forces, and I* and
l3 are the components of the through-the-thickness stress resultant. Finally, A@’ and Ki33aare
the components of the bending couples. See Fig. 1 for a pictorial interpretation of these
resultants. In these expressions, the in-plane membrane resultants nPa are not symmetric.
We denote the surface density in the current surface by 6, given in units of mass per unit
area, and let pa be the surface density in the reference surface. Conservation of mass requires
that ji = 5”. Let Ztidenote the scalar bending inertia (typically IF = h*pll2). The momentum
balance equations in terms of the stress resultants vectors in (9) are

Fig. 1. The resultant stress, resultant stress couple and the resultant through-the-thickness stress, along with the
convected mid-surface basis.
 D.D. Fox, J.C. Simo, A drill rotation formulation 333

( 104

Wb)

n*xa,+rii”xd,,+lXd=O, ( 104

along with the boundary conditions

X = naVa ona,&, W)

i = Gal), on a,.d. ( 104

Here, a,,ti and a,,,& are the parts of the boundary a.& where tractions and couples are
prescribed, respectively, and v = VIA* is the outward normal field to the boundary of the
shell. Equations (lOa) correspond to balance of linear momentum, balance of director
momentum (equivalent to the first moment of the stress about the mid-surface) and balance of
angular momentum, respectively.

REMARK 1. By introducing vector and tensor identities, the balance of angular momentum
equation (10~) can be written equivalently as

skew[n”@aa, +hi”@d., +Z@d]=O. (11)

Define the effective (symmetric) stress resultant tensor n” as

ii:=n”C3aa, +l@d-d,,@iim”. (12)

Then the balance of angular momentum from (11) reduces to the symmetry condition
skew@] = 0. Alternatively, let

n-“:=~aa={‘“ap+~d,
(13)
i:= ;a3 = Pa, + 1”d.

Then, the condition skew@] = 0 is equivalent to the component symmetry conditions n”“’ =
cp* and p = 4”“.

2.2. Elastic shells: hyper-elastic constitutive relations

Motivated by Toupin [5] (see also [9]), and following [3], we characterize the elastic
response of the shell by means of a stored-energy function of the form w(f, g). Following a
standard practice, the dependence of w on the reference surface is implicit and is often
omitted from the argument list. Analogous to the three-dimensional first Piola-Kirchhoff
stress tensor, we introduce the following combination of the stress and stress couple resultants:

P=J[n”@AA,+l@D] and M=.f[h”@AII]. (1% b)

334 D.D. Fox, J.C. Simo, A drill rotation formulation

In terms of P and M, the hyper-elastic shell resultant constitutive relations are (see [lo])

p = Qcl-~ d and M = a,w(f, g) . (15)

Note that P is the nominal stress tensor and M can be regarded as the nominal couple stress
tensor. In fact, in the absence of the field g, one recovers from (15) classical finite elasticity
particularized to the shell mid-surface.
Frame invariance of the stored-energy function requires that

w(Rf, Rg) = w(f, g) for all R E SO(3) , (16)

and implies the existence of a function W such that w(f, g) = G(f’f, f’g). The following
proposition plays a central role in our subsequent developments.

PROPOSITION 1. The frame indifference condition (16) implies balance of angular momen-
tum, which can be written in the equivalent form

skew[Pf’ + Mg’] = 0. (17)

PROOF. In (16), choose R, = exp[sG] E SO(3) for some .Z> 0 and G E SO(~), and differen-
tiate with respect to E at E = 0, to obtain

0=2 e~$exWW~ expL+ld

=P,w(L s>.f’+ ~,w(.f, gk’l: ri . (18)

After introducing constitutive equations (15)) (18) becomes

0 = (Pf’ + Mg’): ij )
(19)
which holds for arbitrary ?/ E SO(~). Expression (19) implies that (17) holds and, since
Pf’ + Mg’ = j(n” @a, + 18 d + rk” @ d.,), also implies that balance of angular momentum in
(11) is satisfied. Cl

It proves convenient to express the elastic constitutive equations in terms of the effective
stress resultant tensor (12). To this end, the chain rule can be used to express constitutive
equations (15) in terms of the frame indifferent stored-energy function $cf, cg), to give

M = QCf’, g>=f %$cf> cg) .

Observe that

g[aCp@(cf, c,)]‘f’ = gM’ = jd,, @ iii* . (21)

From (14a) it follows that Pf’ = J[na 63 a, + I@ d]; multiplying (20) on the right by f’ and
using definition (12) gives the following hyper-elastic constitutive equations for the effective
resultants:
 D. D. Fox, J.C. Simo, A drill rotation formulation 335

where I%: = ii* C3Q, is the effective stress couple resultant. Observe that in the absence of the
gradient field g, constitutive equation (22a) for the effective stress resultant tensor fi would be
analogous to the constitutive equation for the Cauchy stress tensor in nonlinear elasticity.
Furthermore, as in the three-dimensional theory, symmetry of ii is equivalent to balance of
angular momentum.

REMARK 2. Hyper-elastic constjtutive expressions (22) lead to the following remarkably

simple form for the shell resultant total potential energy function:

n(@>= j-d
w(cr,c,)i” d5’ dt2 + n&D) , (23)

where @ = (cp, d) E %Yand the potential energy of the external loading is given by

A straightforward calculation shows that the Euler-Lagrange equations associated with

potential function (23) are the static balance of linear momentum and balance of director
momentum equations (lOa, b). Furthermore, from Proposition 1, the balance of angular
momentum equation (10~) is satisfied by using the properly invariant form of the stored-
energy function W.

3. Extended kinematics: drill rotations formulation

To motivate our subsequent developments, we introduce the notion of a director-orienting

orthogonal transformation. Suppose that a stress-free reference configuration of the shell is
specified by (cp”,0) E (8’.For convenience and without loss of generality, we assume that the
reference director is unit length and coincides with the normal to the mid-surface; i.e.,
D = A, = to. Consider an arbitrary configuration of the shell defined by (cp, d) E %. The
orientation of the unit director field t := d/h, with A := lldll, re 1a t ive to the reference director
t” can be defined by means of an orthogonal transformation

A : ti* SO(3) , such that t = At0 , (25)

which is referred to as a director-orienting rotation. It is clear, however, that within the realm
of the classical theory such a director-orienting transformation is not unique. In fact, consider
a one-parameter subgroup of rotations about to, defined by exp[ O?‘] and referred to as drill
rotations. If A is a director-orienting rotation satisfying (25), it follows that

t = et0 ) for Q = A exp[ 02’1 , (26)

i.e., Q is also a director-orienting rotation for arbitrary value of the drill angle 0. Further-
more, one can show [4], that the classical shell equations are invariant with respect to drill
rotations. This indeterminacy in the definition of the drill rotation can be resolved within the
framework of an extended kinematic description regarding the classical model as a constrained
theory.
336 D. D. Fox, J. C. Simo, A drill rotation formulation

3.1. Extended kinematics

Following the analysis of the three-dimensional theory presented in [l], we consider an

extended configuration space in which the unit director field is replaced by an orthogonal
transformation Q, which defines both the orientation of a director via (25) and a drill rotation;
i.e.,

e = (6 = (cp, Q, A) : ti+ R” X SO(3) x Iw, ] d = AQt”, with (cp, d) E %} . (27)

In the three-dimensional theory, the classical model is recovered by enforcing the constraint
that the independent rotation field coincides with the rotation tensor in the polar decomposi-
tion of the deformation gradient F. In shell theory, the counterpart of this constraint condition
for the shell deformation gradient f must be consistent with the director-orienting constraint
(25) and, therefore, can be enforced only in the tangent plane to the shell. Consequently, we
require that

The fact that this condition furnishes the appropriate constraint can be motivated as follows.
Suppose we require that Q be the rotation tensor in the polar decomposition off. Then Q’f
becomes the stretching tensor and (28) holds with (Yand p E {1,2,3}. Such a requirement,
however, is compatible with the condition t = Qt” only if to is an eigenvalue of the stretching
tensor, i.e. in the absence of shear deformation; clearly an unacceptable restriction. We also
note that from the definition of f in (4), fAa = a,, and the two-dimensional symmetry
condition (29) can be written in the equivalent form

3.2. Drill rotation constraint

Remarkably, the symmetry condition (29) along with the director-orienting condition (25)
implies that the independent rotation field Q in (27) is the rotation tensor of the following
modified deformation gradient. Let-q, E t’ be the projection of a, onto the plane perpen-
dicular to t. Define the linear map f by the relation

f”= q, @A” + d@A” ,

(30)
~=a,-xA y, = taa, .

Clearly det[f] > 0 since, by assumption, t. a, X a2 > 0. We have the following result.

PROPOSITION 2. The rotation tensor Q in the polar decomposition f” is uniquely determined

by the two conditions:

(1) t= Qt” and (2) (A,, Q’u,) = (A,, Q'u,). (31)

PROOF. Let f”= QU be the polar decomposition of j To show that Q satisfies conditions (1)
and (2) in (31) observe first that to is an eigenvector of U : = Q’f since

U= =f”‘f”= 7, . rjoAa @A’ + A2A3@A” + Ut” = ht” . (32)

 D. D. Fox, J. C. Simo, A drill rotation formulation 337

Consequently, since 3” = At, we have

Qt” +J -‘to = A-$’ = t . (33)

Hence, Q satisfies condition (1) in (31). To show that Q also satisfies condition (2), note that

UAa = Q’j’A, = Q’q, . (34)

Using the definition of q,, it follows that (A,, UAa ) = (AP , Q’q, ) = (A,, Q’u, ) , and
condition (2) follows from the symmetry of U.
It remains to show that_conditions (1) and (2) uniquely define the rotation tensor in the
polar decomposition of J Suppose that conditions (1) and (2) are satisfied by some
Q E SO(3). Define the tensor U by the relation

(35)
Then fiA, = At0 and CA, = &‘q,. Furthermore, we have

(A,, CA,) = (A,, UA,) = 0,

(36)
(A,, fiA,) = (A,, Q’v,> = (A,> Q’qJ = (4, fiA,s).

These relations imply that the tensor fi is symmetric. Hence, from definition (35) we conclude
that f”= Qfi with Q E SO(3) and fi symmetric and positive definite. Therefore, Q = Q and
U = U by the uniqueness of the polar decomposition. Cl

3.3. Constrained rotation balance equations

In accordance with our preceding discussion, the appropriate set of local governing
equations is obtained by augmenting the local governing equations of the classical model with
the constraint conditions (31) that define a unique director orienting rotation incorporating a
drill rotation. Using constitutive equations (20) along with the definitions of the nominal stress
and stress couple tensors in (14), the constrained rotation balance equations become

(a) Constitutive equations:

na = j-‘(2f a@ + &$‘)A” , (374

I = 7-‘(2f a,,6 + g[c3,i@]‘)A3 , WV
m"a = Pf aeglGAU. (374
(b) Momentum balance equations:
338 D.D. Fox, J.C. Simo, A drill rotation formulation

(c) Constraint equations:

d= hQtO and (A,, Q’aa) = (A,, Q’a,) , WY 8)

(d) Boundary conditions:

We show below that these equations arise as Euler-Lagrange equations of a constrained

variational principle in which the orienting condition equation (37f) is enforced point-wise on
the configurations and the drill condition (37g) is enforced via a Lagrange multiplier.

3.4. A constrained variational principle

To state the proposed variational principle, we f&t introduce the space of test functions
associated with the extended configuration space (6’. Let ‘5’”be the space of test functions
associated with a classical configuration (cp, d) E %’ parametrized by the triplet (cp, t, A) with
d = ht. Enforcing the linearized version of the constraint lltll = 1 on variations associated with
t yields

(38)

Here, d, ti, a,.& and a,& are the parts of the boundary a& where mid-surface displaceme_nts,
unit directo: displacements and stretches are prescribed, respectively. Now let @ =
(cp, Q, A) E % be an extended configuration. Since, by assumption, Q satisfies the director
orienting condition (37f), the associated space of test functions is given by

‘?:={(q,@p):A ~[W3X[W3XIWl(rl,8XC,~)E~}. (39)

With this notation in hand, we claim that the governing equations (37) arise as Euler-
Lagrange equations associated with stationarity of the following functional:

11(6,i) = Jd[w“( cf, cg) + i : Q'[u, @AA”]]j”’ d[’ dt2 + 17,&D) . (40)

Here II,,, is the potential of the external load defined by (24) and i plays the role of a
Lagrange multiplier with component expression

i = LeaPl$Aa @A, , (41)

where E@ is the permutation symbol. Note that L is characterized by a single scalar quantity
L E L*(d) and represents a two-dimensional skew-symmetric tensor. From a physical stand-
point, i is a constraint stress which enforces the condition that the drill rotation equals the
rotation of the mid-surface. Further, observe that the Lagrange multiplier term can be written
in the more convenient scalar form as

it : Q’<a, @A”) = fo ((A,, Q&j - (4, Q’q)). (42)

 D. D. Fox, J. C. Simo, A drill rotation formulation 339

To show that variational principle (40) delivers the constrained set of balance equations (37),
we first compute the directional derivative of the director field d : = hQt” as (see [3])

6d:=Dd+j,O,p)=pd+tlxd. (43)

Taking the first variation of constrained potential energy function fi and using (43) yields

tiff = ~ [(2f &,I,? + g[a, 61’ + Qi)Aa. q,, + (2f a,6 + g[r3, ;]‘)A’ *6d
J K K

+f dcx;A” *ad,, + 3 : Q’[u, @Aa] + Qid x a, *O]i” dtl dt2 + WI,,, , (44)

for arbitrary test functions ( 7, 8, CL)E ? and 3 = SE@/~I~~)A~

@A,, and S E L’(d). Equation
(44) can be expanded by computing the first variation of the potential energy of the external
loading from (24), using the divergence theorem, and introducing constitutive equation (37a),
to obtain, after collecting terms, the result

Now observe that given any arbitrary vector u, the two conditions u X d = 0 and u * d = 0
implies that u = 0. Following standard arguments, (45) implies the following Euler-Lagrange
equations:

-Z+rii 1 -j-‘QiAaxa,=O,

(46~)

(4, Q'qd = (A,, Q'a,> , (46d)

along with the boundary conditions

and the constitutive equations (37a-c). To show that (37a-c) and (46) reduce to (37), take the
inner product of (46b) with t to obtain
340 D.D. Fox, J.C. Simo, A drill rotation formulation

O=Qih*xa;t=LA*.Q’a,. (47)

Since A” * Q’arl = A*” (AB, Qfan) # 0, (47) implies that L = 0 or equivalently that i = 0.
Consequently, (46a) reduces to the balance of linear momentum equation (37d). Further-
more, the condition i = 0, along with (46b) and (46c), yields the result

1 TN0
pm ),,-z+m=o, (48)
which is the stress couple balance equation (37e). Recall that the balance of angular
momentum equation is satisfied by using the properly invariant stored-energy function.

3.5. Regularized variational principle

A regularized form of the variational principle in (40) is possible. This regularization is
similar to that described in [l]. Define the regularized functional

where y E (0, ~0). The Euler-Lagrange equations corresponding to the functional IIY produce
the appropriate momentum balance equations for any positive finite value of the parameter y
and not just in the limit as y-a. The associated Euler-Lagrange equations are the
constitutive equations (37a-c) along with the following balance equations:

w4

p?z ),U-z+m=O, ww
fi ((A,>Q'a,> - (4, Q'a,>> ; L=O, GOC)

L=O, GOdI

plus the boundary conditions (46e, f).

3.6. Matrix differential operator expressions

The weak form of the momentum balance equations corresponding to the variational
principle (40) ( i.e. Sfi = 0) are now written in matrix operator notation. Recall from Remark
2 that the first variation of II produces the weak form of the momentum balance equation for
the drill-free formulation; hence, define

G(@; a@) = 611, (51)

where @ = (cp, d) E % and a@’ = [q 8 x t p]. We refer to [3] for a detailed discussion of the
weak form of the drill-free formulation G(@; 6@). For convenience, define the following
vector of variation:

S&=[7j 8 p]. (52)

 D. D. Fox, J. C. Simo, A drill rotation formulation 341

Conceptually, the first variation of If can then be written as

Sn=G(@;@D)+j-- [$B,Sg+; ((A,, Q'u,) - (A,, Q'u,,,]? dtl d12,

(53)
where the matrix differential operator B, is defined as

l& = ]Bcm B,, 01, where B,, = (QA,>’ 3 - (QA,>’ $

and B,, = (QA, X a2 + a, X QA2)t. (54)

Note that the weak form of the momentum balance equations (53) is simply that of the
drill-free formulation plus terms arising from the Lagrange multiplier term.

REMARK 3. The Lagrange multiplier i can be eliminated from the variational principle in
(49) by satisfying (50~) point-wise, that is by setting j7)L = ?((A,, Q’a,) - (AZ, Q’a,)). The
resulting variational principle is

n,(@=n(@>+ ; j-&
$2[(A,, Q'a,) - (AZ, Q’q)]‘i” d[’ dt2 . (5%

For this variational principle, the weak form of the momentum balance equations becomes

MI7 = G(@; 6@) + y j-& [~((a,,Q'o,)-(A,,Qtu,))5~SC]PdSldC2. (56)

As with the regularized variational principle (49), the potential energy function (55) produces
the desired momentum balance equations for any finite positive value of the parameter y, not
just in the limit as y+ 03.

3.7. Consistent tangent operator

In this section, we briefly outline the second variation for the constrained variational
principle given by (40). The second variation contains two contributions, one part, denoted
DG(@; S@) * A@, which is identical to that from the drill-free formulation presented in [3],
and the other part rest+ from the con$rained rotation formulation. Denoting the second
arbitrary variation by A@ = (v, w, 6) E ‘V and for H E L’(d), the second variation of 17 can
then be written as

Here, Kd represents the linearization term that arises in the linearization of the drill-free
formulation. The matrix differential operator W and the matrix K, are defined as
342 D. D. Fox, J. C. Simo, A drill rotation formulation

0 0 C@$)
and Kc= 4 !k 2 -(QAA , (58)
J -(QA,> (QA,) kc I

where the 3 X 3 contribution k, is defined as

kc = QA, 8% - QA,@a, -[(A,, Q'a,) - (A*, Q'a,)]l.

The ‘drill’ part of the geometric tangent stiffness K, is defined as

Kd =
fi*‘d@
0
0
q,
0
0
riia2d@ q,
&“‘d@ q,
riia2dC3r),
k, 1 , (60
where the 3 X 3 contribution k, is defined as

(61)
where K,~ = a, * t.p.

REMARK 4. As was discussed in [ll], the appropriate definition of the Hessian operator
differs from the consistent linearization by an additional term. It can be shown (see the thesis
of Fox [lo]) that the appropriate tangent operator (the Hessian) is only the symmetric part of
the geometric tangent stiffness as given above.

4. Closure

The central idea in the formulation of classical nonlinear shells models incorporating drill
rotations is the same as that underling the development of nonlinear elasticity with in-
dependent rotations. The configuration space of the classical model is enhanced by introducing
a proper orthogonal transformation attached to each point of the medium. The classical model
is then recovered by introducing a kinematic constraint that identifies this independent
rotation field with the rotation of the medium. For nonlinear shells this kinematic constraint
identifies the drill rotation with the rotation of the mid-surface of the shell. In a computational
context, a three rotational degree of freedom formulation is obtained merely by enforcing this
constraint weakly via a Lagrange multiplier term appended to the variational principle of the
classical model. Since the corresponding Lagrange multiplier is shown to vanish at equilib-
rium, it is possible to construct a regularization of this constrained variational principle which
yields the equilibrium equations for any value of the regularization parameter. The present
approach is in contrast to the formulation proposed by Reissner [6] (and the essentially
identical approach given in [12]) which uses balance of angular momentum, expressed as a
symmetry condition on the stress resultants, as the constraint equation. Such a formulation
leads naturally to Hellinger-Reissner type of variational formulations which appear to be best
suited for linear analysis.
From a computational point of view, the present formulation offers several advantages.
First, as alluded to above, the regularized variational principle yields the exact equilibrium
 D. D. Fox, J. C. Simo, A drill rotation formulation 343

equations of the classical model for any value of the regularization parameter. Thus, in sharp
contrast with standard penalty methods, the accuracy of the method does not depend upon
adopting high values for the regularization parameter. Alternatively, mixed finite element
methods can be constructed based on two and three field variational formulations. Second, the
discrete finite element model possesses three independent rotational degrees of freedom at
each node. Thus, ad hoc matching procedures of dissimilar rotational degrees of freedom for
different types of structural elements are circumvented. Finally, exact configuration update
procedures for the rotational matrix associated with the three nodal degrees of freedom can be
immediately developed by using the closed form expression for the exponential mapping in the
rotation group and its linearization given in [7].

Acknowledgment

We are indebted to T. J.R. Hughes and R.A. Toupin for many helpful discussions. Support
for this research was provided by AFOSR grants 2-DJA-544 and 2-DJA-771 with Stanford
University. This support is gratefully acknowledged.

References

[l] J.C. Simo, D.D. Fox and T.J.R. Hughes, Formulations of finite elasticity with independent rotations,
Comput. Methods Appl. Mech. Engrg. 95 (1992) 277-288.
[2] J.C. Simo, D.D. Fox and M.S. Rifai, On a stress resultant geometrically exact shell model. Part III:
Computational aspects of the nonlinear theory, Comput. Methods Appl. Mech. Engrg. 79 (1990) 21-70.
(31 J.C. Simo, M.S. Rifai and D.D. Fox, On a stress resultant geometrically exact shell model. Part IV: Variable
thickness shells with through-the-thickness stretching, Comput. Methods Appl. Mech. Engrg. 81 (1990)
91-126.
[4] J.C. Simo and D.D. Fox, On a stress resultant geometrically exact shell model. Part I: Formulation and
optimal parametrization, Comput. Methods Appl. Mech. Engrg. 72 (1989) 267-304.
[5] R.A. Toupin, Theories of elasticity with couple-stress, Arch. Rational Mech. Anal. 17 (1964) 85-112.
[6] E. Reissner, On the form of variationally derived shell equations, J. Appl. Mech. 31 (1964) 233-238.
[7] J.C. Simo and L. Vu-Quoc, Three-dimensional finite-strain rod model. Part II: Computational aspects,
Comput. Methods Appl. Mech. Engrg. 58 (1986) 79-116.
[8] T.J.R. Hughes and F. Brezzi, On drill degrees of freedom, Comput. Methods Appl. Mech. Engrg. 72 (1989)
105-121.
[9] M.M. Carroll and P.M. Naghdi, The influence of the reference geometry on the response of elastic shells,
Arch. Rational Mech. Anal. 48 (1972) 302-318.
[lo] D.D. Fox, A geometrically exact shell theory, Ph.D. Dissertation, Applied Mechanics Division, Stanford
University, Stanford, California, 1990.
[ll] J.C. Simo, The (symmetric) Hessian for geometrically nonlinear models in solid mechanics: Intrinsic definition
and geometric interpretation, Comput. Methods Appl. Mech. Engrg. 96 (1992) 189-200.
[12] P.M. Naghdi, On a variational theorem in elasticity and its application to shell theory, J. Appl. Mech. 31
(1964) 647-653.