owN Division of Engineering

BROWN UNIVERSITY
PROVIDENCE, R. I.

I) O SPERMUS '-4 00

ON NUMERICALLY ACCURATE
FINITE ELEMENT SOLUTIONS
IN THE FULLY PLASTIC RANGE ~ U

J. C. NAGTEGAAL, D. M. PARKS,
AND J. R. RICE

Reproduced by l r
NATIONAL TECHNICAL
INFORMATION SERVICE
US Department of Commerce v E tne
Springfield, VA. 22151 ; 0

Technical Report
NASA NGL 40-002-080/14 to the
National Aeronautics and Space Administration

March 1974
 NOTICE

THIS DOCUMENT HAS BEEN REPRODUCED FROM THE

BEST COPY FURNISHED US BY THE SPONSORING

AGENCY. ALTHOUGH IT IS RECOGNIZED THAT CER-

TAIN PORTIONS ARE ILLEGIBLE, IT IS BEING RE-

LEASED IN THE INTEREST OF MAKING AVAILABLE

AS MUCH INFORMATION AS POSSIBLE.

 On Numerically Accurate Finite Element Solutions in the Fully Plastic Range

by

J. C. Nagtegaal, D. M. Parks, and J. R. Rice

Division of Engineering, Brown University, Providence, R. I., USA

Abstract

It is often found that tangent-stiffness finite element solutions for

elastic-plastic materials exhibit much too stiff a response in the fully plastic

range. This is most striking for the perfectly plastic material idealization,

in which case a limit load exists within conventional small displacement gradient

assumptions. However, finite element solutions often exceed the limit load by

substantial amounts, and in some cases have no limit load at all. It is shown

that a cause of this inaccuracy is that incremental deformation fields of typi-

cal two and three-dimensional finite elements are highly constrained at or near

limit load. This is shown to enforce unreasonable kinematic constraints on the

modes of deformation which assemblages of elements are capable of exhibiting.

A general criterion for testing a mesh with topologically similar repeat units

is given, and the analysis shows that only a few conventional element types and

arrangements are,or can be made suitable for computations in the fully plastic

range. Further, a new variational principle, which can easily and simply be

incorporated into an existing finite element program, is presented. This allows

accurate computations to be made even for element designs that would not normally

be suitable. Numerical results are given for three plane strain problems, namely

pure bending of a beam, a thick-walled tube under pressure, and a deep double

edge cracked tensile specimen. These illustrate the effects of various element

designs and of the new variational procedure. An appendix extends the discussion

to elastic-plastic computation at finite strain.

t Currently 570 Ordnance Regiment, Royal Dutch Army

e
 -1-

1. Introduction

In recent years, the finite element method has been employed for analysis

of structures exhibiting elastic-plastic.material behavior [1], and many successful

forms of the
applications have been made (see [2] for a comparison of possible

required incremental analysis). However, with the application of the method,

at least in its tangent-stiffness form, to problems of plane strain, axisymmetric

and three-dimensional problems, inaccurate results are often obtained. The problem

is most clearly demonstrated for structures of ideally plastic material. Then a

limit load exists, which can sometimes be calculated exactly or bounded from above
exhibit
by the kinematical theorem, while the finite element solution often seems to

no limit load at all, but rather a steadily rising load-displacement curve attaining

values far in excess of the true limit load. Similarly, for strain-hardening

materials having typical values for the plastic tangent modulus of two or more

orders of magnitude less than the elastic modulus, the finite element solution

exhibits an artificially high terminal slope of the load-deflection curve in the

fully plastic range.

Later we shall see examples of this for a beam in pure bending and for a

punch problem. In the case of a plane strain extrusion problem [3], when a
-3
bilinear constitutive law with slight plastic work-hardening (da/de-- H = 4.4x10-3 E;
p
E , the Young's modulus) was incorporated, a finite element extrusion pressure of

1.5 times a slip-line solution (using the same yield stress as the linearly har-

dening model) was attained with overall billet displacements only of order six

times the displacement obtained at the slip line limit load. This indicates that

the computed stiffness in the fully plastic range far exceeds what would be ex-

pected for the small hardening modulus used.

In some cases, however, the correct limit load is obtained. Fig. 1

illustrates some of the peculiarities which have been encountered by the authors
 -2-

and co-workers in elastic-perfectly plastic analysis of plane strain bodies

containing cracks. The finite element meshes shown in fig. 2 were used to

solve the single-edge notched plane strain tensile specimen subject to both

uniform end displacement [4] (i.e., no end rotation) and the static equivalent

of a mid-ligament concentrated force. These two loading conditions have the same

analytical limit load, namely, twice the yield stress in shear on the net section.

As can be seen in fig. 1, the overall load-deflection curves of the two loading

conditions are quite different in the fully plastic regime. The solution for

the mid-ligament loading clearly indicates the correct limit load, while the

constant end displacement solution exceeds limit load, with the load continuing

to rise at a roughly constant rate.

A cause of these problems relates to the fact that the deformation state

of an elastic-perfectly plastic material is highly constrained at limit load;

for the usual material idealization, deformation increments at limit load will

be strictly incompressible. In the usual finite element formulation, in terms

of kinematically admissible displacement fields, the same condition will have

to be satisfied.

In particular, a tangent-stiffness finite element solution satisfies the

incremental virtual work principle

f dS
S elem.. elem. ij13ijdVelem

precisely, where .ij is the stress rate following from the prescribed con-

stitutive law in terms of the current stress a.. and strain rate s.. within

each element. This stress rate can be written as

 -3-

* * 1 (1.2)
=
1j ij 3 13 kk

where si is the deviatoric stress increment. Since the plastic deformation

1 •
will be assumed purely deviatoric, the hydrostatic stress increment 3 akk can

be expressed as

-akk = Cek (1.3)

3kk Kkk

where

K = E/3(1-2v) (1.4)

is the elastic bulk modulus and skk is the dilatational strain increment.

Substitution of (1.2) and (1.3) in (1.1) then furnishes

T.u.dS = +
K(kk)2] dV elem
s..e.. (1.5)
S elem. Velem.

where e.. is the deviatoric strain increment. In the vicinity of the limit

load, the term s..e.. will tend to vanish pointwise, but, as follows from

plastic normality, will never be negative. Hence, from (1.5), this implies that

the incremental finite element solution satisfies

T.u.dS f K( kk)2 dVelem. (1.6)

S elem. Velem.

The following result may thus be stated:

In order for a limit load to exist for the discretized finite element

model of an elastic-plastic problem, it is necessary that the elements

be capable of deforming so that kk = 0 pointwise throughout the ele-

 -4-

ments. Otherwise (1.6) requires that the load-deflection curve be

steadily rising - i.e., no limit load exists.

In the next section it will be shown that, except for plane stress

problems, the requirement skk = 0 severely constrains the class of de-

formations of which typical finite element grids are capable. When they are

forced to deform in such a constrained fashion, unreasonably high limit loads,

or no limit loads at all, will result. On the other hand, even if no limit

load is achieved in the finite element formulation, it is to be expected (and

is, indeed, observed) that the slope of the load-deflection curve is reduced

substantially from its initial elastic value at load levels near the theo-

retical limit load. This, however, does not allow an accurate inference of

the limit load.

Even if the material is linearly work-hardening with a small hardening

coefficient, the terminal slope of the load-deflection curve will not necessarily

be accurately determined (as was argued previously and will be demonstrated in

an example), since the material will deform in a nearly incompressible fashion

in the fully plastic range.

It is clear that the problems discussed here have a strong similarity to

problems encountered in the analysis of incompressible fluids and rubber-like

solids. The difference is that the material behaves incompressibly only as

limit conditions are approached because the effective shear modulus then tends

to vanish, whereas fluids and rubber-like solids behave incompressibly from the

start because the bulk modulus is very large. The essential problem is the
same in both cases, however, in the sense that the incompressibility require-

ment puts too severe constraints on the possible deformation modes.

 - 5-

2. Analysis

In the previous section, it has been observed that elements must be capable

of deforming without change of volume pointwise if a limit load is to be obtained.

It is therefore useful to investigate the constraints this enforces upon each

element, and the effect of these constraints on the behavior of an assemblage

of elements.

Consider, for instance, the grid of 4-node rectangular isoparametric ele-

ments shown in fig. 3. Within each element, the displacement increments are

of the form

} x = {a} + {b}x + {c)y + {d}xy , (2.1)

y

where the vectors {a), {b}, etc., are expressed in terms of the nodal veloci-

ties and coordinates. Now, the incompressibility constraint for plane strain has

the form

S +E _ x - : 0 (2.2)
xx yy axx y

and this requires that {d} = {0} as well as that b + c = 0 , a total of

x y
three constraints.

The fact that {d} = {O) means that at limit load each element has strain

increments which are constant throughout the element. It is then evident from

displacement continuity that all elements marked * in fig. 3 must have the

same value of Exx , whereas all elements marked t must have the same value

of E . Moreover, since xx = -E xx and E will be the same in all

yy x yy xx row
elements marked * or t . Since a similar argument can be set up for any row
 -6-

and column, E and must be the same in every element of the grid.

Clearly, this is a very unrealistic constraint.

It is at least possible for this mesh to have a non-uniform shear e xy ,

but the situation is even worse in case the grid consists of arbitrary quadri-

lateral isoparametric elements. Within an element the displacement increments

are of the form

{u} = {a} + {bln + {c} + {d}nE , (2.3)

where n and 5 are defined by

{ } = f{} + {B}n + {y}c + {6}nh . (2.4)

(See, for instance, [63). A lengthy but straightforward calculation furnishes

the three incompressibility constraints per element:

by- c8 - by + Cy8 = 0 ,
xy x y yx y x

b6 - d -b6 + dy = 0 , (2.5)
xy xy yx yx
dxy - Cxy - dyYx + Cy6 = 0

A solution to this system of equations in terms of three parameters A, B and C is

bx = ABx + B , b C8x - AS

cx Ayx + BY , c =Cyx - Ay (2.6)

dx = A x + B6 , d = C6x - A6 .

Substitution of (2.6) in (2.3) then makes it possible to eliminate n and E

from (2.3) and (2.4), which furnishes the displacement increments:

F
 -7-

U = D + Ax + By
(2.7)
u = E + Cx - Ay

where the constants D and E do not depend on A, B and C . Hence, if the

incompressibility constraint is to be satisfied, the strain increments will be

constant throughout the element.

This has catastrophic effects for an arbitrary grid. Consider, for instance,

the grid in fig. 4, and specify the displacement increments of the three nodes

A, B and C . Because the displacement increments of nodes A and B define

the extensional strain along AB in element I , and because the element must

deform with a constant strain of zero dilatation, we must then specify the com-

ponent of displacement increment of node E in the direction normal to AB .

Similarly, because the displacement increments of nodes B and C determine

the extensional strain along BC in element II , we must also specify the dis-

placement component of node E in the direction normal to BC . The incompress-

ibility constraints of elements I and II then specify the components of displace-

ment increment of node E in two linearly independent (though not orthogonal)

directions. Thus the displacement increment vector of E is fully determined,

and, in fact, it corresponds to that increment which causes identical strain

increments in elements I and II . The displacement increments of nodes D and

F are then also determined. Continuation of the argument then furnishes, that

the strain increment will be constant throughout the grid. It should be noted

that this argument holds only if the element boundaries do not form a straight

line through the body; over such a line the shear strain increment can be dis-

continuous.
 Examples of the unreasonable constraints enforced by pointwise incompress-

ibility upon displacement increment fields of other two and three-dimensional

mesh configurations are given in Appendix I.

We are now in a position to explain the results of fig. 1, bearing in mind

a result of perfectly-plastic limit analysis: namely, that while the limit load

is unique, the deformation field at limit load need not be unique. An acceptable

limit field for both loading conditions, giving the correct limit load, is that

of concentrated deformation on 450 lines from the crack tip to the surface, as

shown in fig. 5a. An alternative limit field for the mid-ligament loading is

shown in fig. 5b. This field consists of constant strain increments within the

region bounded by the two 450 lines from the crack tip. The rest of the body is

rigid, so the overall effect is that of a rotation of the two rigid regions about

the crack tip. In fact, this was exactly the velocity field obtained at limit

conditions in the finite element solution of the mid-ligament loading problem.

Since the deforming regions do so uniformly, and the 450 lines coincide with

element boundaries, the mesh of focused isoparametric quadrilaterals could

readily accommodate the incompressibility constraints, hence leading to a very

accurate finite element prediction of the limit load.

In contrast, however, the rotation produced by this field is incompatible

with the constant end-displacement boundary condition of the other loading. The
same constant strain pattern cannot develop, and an ever-worsening overestimate

of the limit load results.

Although the problem has now been analyzed sufficiently for four-noded iso-

parametric quadrilateral elements, a more systematic approach is needed to find

a solution to it. It is therefore useful to consider the matter of convergence

of the solution if the mesh is refined. Refinement of the mesh will have two
 -9-

opposing effects:

1) It will increase the number of nodes, and since each node represents

a specific number of degrees of freedom, it will increase the total number

of degrees of freedom.

2) On the other hand, it will increase the number of elements, and since

each element has a certain number of incompressibility constraints enforced

upon itself, it will increase the total number of constraints.

It is clear that convergence will occur only if the total number of degrees of

freedom increases faster than the total number of constraints. This furnishes a

relatively simple criterion to check whether a mesh of a certain type will be

adequate to furnish accurate limit loads and flow fields. It should be noted

that the total number of constraints is not always equal to the number of ele-

ments times the number of constraints per element. Sometimes elements can be

arranged such that the constraints are no longer independent, an important

example of which will be discussed in the next section. For the moment, however,

these special cases will not be considered.

It is thus important to determine the ratio of the total number of nodes

to the total number of elements in the grid when it is refined. Consider a body,

loaded in plane strain, the cross-section of which is subdivided into a mesh of

elements of identical type. Define the nodal angles e of an element a as

the inner angle formed at the node by the two adjacent element boundaries. For

each type of element, the sum of these nodal angles is a given number. Consider,

for instance, the six-noded triangle of fig. 6. Each of the nodal angles at the

midpoints of the sides is equal to 7 radians, whereas the sum of the angles

8a 82 and 83 is also equal to r radians. Hence the sum of all nodal angles

is equal to 4w radians. Similarly simple calculations can be made for other

types of elements. For generality, let us assume that the sum of the
 -10-

nodal angles of a particular element type is equal to nn radians. If the grid

consists of p elements, then the sum of all nodal angles of all elements is

equal to

e = pe . (2.8)

The sum of all nodal angles can also be calculated in a different way.

Suppose the mesh has k nodes inside the body and Z nodes on the boundary.

The sum of the nodal angles around each interior node is equal to 2n radians,

whereas the sum of the nodal angles at a boundary node is equal to 7-y radians,

where y is a small angle due to the curvature of the boundary. In this way,

the sum is readily calculated as

ea = (2k + Z + 2c - 4)n (2.9)

a i

where c is the degree of connectivity of the body (c = 1 if simply connected,

etc.). From (2.8) and (2.9) then follows the equality

pn = 2k + Z + 2c - , (2.10)

or, alternatively,

P 2 £ 2c-4
k n nk nk (2.11)

Now if the mesh is refined, p, k and t will all increase; hence the last

term will vanish. Moreover, if the mesh is refined uniformly, k (the number
of interior nodes) will increase approximately as the square of £ (the number

of boundary nodes). Hence

lim ( )
2
k+
k4m k n (2.12)
 -11-

Consider what this means for a grid of isoparametric quadrilateral elements.

The sum of the nodal angles in one element is equal to 2w radians, and hence

n = 2 . Substitution in (2.12) then furnishes that in the limit the number of

nodes will become equal to the number of elements. Since each node represents

two degrees of freedom, and three incompressibility constraints are enforced

upon each element, the total number of degrees of freedom is only 2/3 of the

total number of constraints. Hence convergence will not in general occur, which

reaffirms the previously obtained result. Similar derivations can be made for

other planar elements as well as for axisymmetric elements, since (2.12) holds

as well for this class. The results are displayed in Table I. The range of value

of constraints per element from 6 to 8 given for the 8-noded quadrilateral depends

on whether the sides of the element are linear or quadratic, respectively.

For three-dimensional elements, however, the ratio of nodes to elements

is not unique, since the sum of the solid angles of a polyhedron is not a con-

stant. Therefore one has to determine this ratio for a specific arrangement of

elements. For some common regular grids, and thus of any grid which is equi-

valent, the results are displayed in Table II. It may be assumed that these

results are fairly representative for most common types of arrangements.

The results in Tables I and II indicate that none of the usual finite

elements, except the 6-noded plane strain triangle and, to a much lesser degree,

the 10-noded tetrahedron, is adequate to analyze (approximately) incompressible,

and thus limit load, behavior. Before going into a more systematic approach to

solve the problem, let us consider what can be obtained by a special arrange-

ment of the elements.

 -12-

3. Special Arrangements of Elements

As has been mentioned previously, the total number of constraints is

not
necessarily equal to the product of the number of elements and the
number of
constraints per element. For special arrangements of elements, the total number
of constraints may be less than this product. The obtained improvement, however,
will be small, since it is necessary to combine a number of elements to eliminate

one constraint. Thus this method will work only for elements for which the ratio

of degrees of freedom to constraints in Tables I orII is

close or equal to one.

Consideration of Tables I and II with this in mind indicates

that only a few
types of elements could possibly be used for this purpose:
for plane strain, the
3-noded triangular element and perhaps the 8-noded quadrilateral
element, and for
three-dimensional problems, the 10-noded tetrahedron. None of the axisymmetric
elements could possibly be used. As yet the only successful arrangement that has
been discovered is the combination of four 3-noded triangles
as shown in fig. 7.
Note that the triangular element boundaries form a quadrilateral
"element" and
its diagonals. This combination has the special property that if the
incompress-
ibility constraint is satisfied in three of the elements,
the constraint in the
fourth element is automatically satisfied. This brings the ratio of degrees of
freedom to constraints from 1 to 4/3, which makes the arrangement
suitable for
analysis near limit load. The proof of this special property is given below.

Let the coordinates of the node i be x. , and let the node have a dis-
placement increment . The area A of element I before the displacement
increment has taken place is then equal to

A )
x-( -x)] k , (3.1)
 -13-

where x and * denote the usual vector and scalar products respectively, and

k is the unit vector perpendicular to the plane of the elements. Similarly one

finds that after the displacement increment has taken place,

AA - [(x2+-x -)x (x+u -x -u )] k . (3.2)

2 '2 2 no o1 - -o0
o

Subtraction of (3.1) from (3.2) furnishes

1[(x -x )x( -u -x )x(u - )+(u -u )x(u -ui)] * k . (3.3)

2 2 -o 1 o 1 -o2 -o -~ o - -o

When one neglects the second order terms, the incompressibility constraint

takes the form

(x -x ) x (u -u = (x 1 -x ) x (u -u ) (3.4)
2 -o 1 ~o 1 -o 2 -o

Similarly, for the elements II and III , one obtains

(x -x ) x (u -u ) (x 2 -x ) x (u -u ) , (3.5)
3 ~o -2 -o ~o 3 -o

(x -x ) x (u u ) (X -x x (u -u ) (3.6)

Now multiply (3.4) by Ix 3 -xo/x/Ix -x , furnishing

x -x-
- ( )
X-xxx ( - ) x (u -ux (3.7)
- -o 3 O -2 -o
Ik7oI -2 'o
 -14-

since x - x and x - x° have opposite directions. Similarly, one obtains

for (3.6)

Ix -x
-(x -x )x(u-u ) (x -x )x(u -u (3.8)
2 o 3 o 3 o ~4 o

Substitution of (3.7) and (3.8) in (3.5) then furnishes

x xx -x
2
(3-
X-x )X( - ) =- (x'-x)x(u4 -uo) ' (3.9)
-l 1'4-ool

or, alternatively,

(x
-2-x
-oo)x( ~ -u~o ) = (x -X)x( -u ) . (3.10)
3 ~o 4 "o
1 2 -xo II x,3
3-x
,o
I

With the same argument as was used for (3.7) and (3.8) this yields

(xx )x( ) -x )x(-u ) , (3.11)

4 o 1 Po -o 4--o

which proves that the incompressibility constraint in element IV is satisfied.

It should be noted that this approach requires no modification of existing

finite element programs. The groups of four elements have all the generality in

application of the usual quadrilateral elements, and the calculations can be

carried out without changes to an already existing program. -The principal short-

coming of the method is that it can be applied only to problems of plane strain,

for which another alternative (the 6-noded triangular element) already exists.

Even this alternative can be slightly improved by arranging the 6-noded elements

similar to the 3-noded ones. The ratio of degrees of freedom to constraints then
 -25-

improves from 4/3 to 16/11

It should also be noted that the degrees of freedom available need not

actually be incorporated into the finite element solution. For example the

degrees of freedom corresponding to the interior node "0" of fig. 7 may be

"condensed out" of the master stiffness matrix and recovered later. (see, for

example, [7])

As yet no method has been devised to arrange the lO0-noded tetrahedra such

that an acceptable ratio of degrees of freedom to constraints is obtained.

Therefore a different, more general approach is needed to solve problems other

than those of plane strain.

 -16-

4. A Modified Variational Principle

It has been discussed before that an absolute requirement for the existence

of a limit load is that the dilatation increment ;kk vanishes pointwise. It

has also been demonstrated that only a few of the conventional element types are,

or can be made, useful to analyze problems under this constraint. Therefore it

is desired that different elements be constructed, for which fewer constraints

per element are sufficient to satisfy the incompressibility requirement. This

goal can be obtained by taking care that the dilatation is governed by fewer pa-

rameters than in the conventional elements. Clearly, this procedure may be applied

only to higher order elements for which the conventional number of constraints per

element exceeds unity. Theoretically it is possible to construct interpolation

functions that have this property; in fact, if the arrangement in fig. 7 is taken

as a single element, and the displacement of node 0 is chosen such that the di-

latation in all subelements I through IV is the same, one has obtained such a

function. This special field must, however, be considered as a "lucky guess", and

a general approach to construct similarly simple fields for other elements does

not seem to exist.

An alternative way to solve the problem is to create a variational

principle in which the dilatational strain increment and the displacement incre-

ments are present as independent variables. We then have complete freedom to

characterize the dilatation by as many (or as few) parameters as we choose. Such

a variational principle is akin to the Hellinger-Reissner principle [8], which

admits the displacements and the stresses as independent variables. The validity

of the present principle will, however, be proved here independently of the H-R

principle.
 -17-

Consider the functional

( + K)] dV - udS (4.1)

_ V, ) K(k,k 2 S
where

e. - ; - 6.uk is the deviatoric strain increment;

13 2 1i, 3,i 3 ij k,k
W'(eij) is the deviatoric rate potential [ W' = s..e.. ; 6W' = s..6e .
1 2 i3ij 3 13
is the (independent) dilatational strain increment;

K is the (instantaneous) bulk modulus.

This functional is well defined if a finite bulk modulus K exists.

That this functional corresponds to a valid variational principle is readily

proved in the usual way. By taking the first variations in u. and * one finds

6 V . 6uk,k)d dS + f(,k- )6$dV (4.2)

,T

and since

a + O
s ij K 6 ,j (4.3)

this can be written

61 = ij dv- T 6dST + VK(,k-$)6dV (4.4)

The first two terms express the virtual work principle, and
hence imply that
aij satisfy the continuing equilibrium equations in V and force rate boundary
conditions on ST , in the usual manner. The last term provides the additional
result that * = ;,k
 -18-

It is appropriate to compare the present principle with the principle

created by Herrmann [5] for the analysis of incompressible, linear elastic

materi-
als. The functional used by Herrmann would have the incremental form

H[u,=hJ [ + 2
vhekk - v(l-2v)h JdV - j Ti dS ' (4.5)
T

where p is the shear modulus, v is Poisson's ratio, and where the "mean
pressure function: is given by

akk
2p(l+v) (4.6)

Apart from its specialization to linear elasticity, the fact that ;kk enters
Herrmann's principle quadratically is an essential difference. Indeed, this does
not allow for it the same simple procedures that can be based on the
present
principle, and that allow the easy adaption of existing, conventional
stiffness
programs to it.

Before going to the application of the principle, let us determine how

many
constraints per element would be desired, in order to get an as close
as possible
approximation to continuum behavior. In the continuum, each material point repre-
sents two (in problems of plane strain or axisymmetric problems) or
three (in
three-dimensional problems) degrees of freedom. For incompressibility, one con-
straint is valid at each material point. Hence the ratio of degrees of freedom
to constraints in the continuum is equal to two or three, depending
on the type
of problem.

It seems plausible that a finite element solution to an elastic-plastic

prob-
lem would give the best overall approximation if the ratio degrees of freedom
to
constraints would be the same as in the continuum. Hence a "two" would be desired
 -19-

in the last column of Table I and a "three" in the last column of Table II. This

readily leads to a desired number of constraints per element. For instance, for

the 4-noded quadrilateral elements in Table I, the desired number would

be one, and for the 8-noded quadrilateral elements, it would be three. Similarly,

for the 8-noded cube in Table II, one constraint would be desired. Similar desired

numbers can be obtained for the other elements in the tables.

Now let us, for the moment, restrict the discussion to those elements for

which the desired number of constraints is equal to one. As has been discussed

before, one then needs a dilatation increment governed by a single parameter.

It is, on the other hand, necessary for convergence that at least a constant

dilatational strain increment be obtainable within an element (see, for instance,

[7). Hence, it is necessary to choose in an element a

S= = constant . (4.7)

Substitution of (4.7) in (4.1) then furnishes

I = + K 1 2
I
a=1 V
[W'(e) dV-
S T.u.dS (4.8)

Variation of ~ then yields the relation

= Kuk ikdV KdV (4.9)

In practically all cases, the bulk modulus r will be constant within an element;

in problems of non-dilatational plasticity, for instance, K = Kel is a constant,

and may be brought outside the integral in (4.9) so that

 -20-

V- 1= ,kdV . (4 10)
ta a ukka
aV

Substitution of (4.10) in (4.8) then furnishes the modified functional

W'e ) +
t (V- u kkdVa )2 dV - u dS . (4.11)
a=1 Vk a S
a a T

This functional can be simplified by defining the modified strain increment

i - e..ij +6

- (u. +u. .) + 6..VI u dV - ki (4.12)

2 (1,)juj,i) -3 ij. V ,k ,k

so that the functional (4.11) can be written as

p
SW(j..)dV I T.u.dS , (4.13)
U=1 V ST
a T

where W represents the total rate potential (i.e., 6W(C) = ij..6..) , but

evaluated for the modified strain increment i.. . It should be observed that the

functional (4.13) may be expressed purely in terms of nodal displacement increments;

hence no additional variables (i.e., mean pressure) have to be used. It should also

be noted that the functional is identical to the usual rate potential functional,

with the exception of the relation between strain and displacement increments. As

was mentioned before, this makes it extremely simple to adapt an already existing

program for this method. One need only rewrite the matrix -convertingnodal param-

eters to strains in accord with (4.12). An equally simple formulation cannot be

obtained with Herrmann's variational principle.

It is interesting to note that the frequently employed procedure of replacing

the radially-varying strains in axisymmetric triangular elements by their values

at the element centroid could be viewed as being based on a variational principle

similar to the present one, in both cases implemented by changing the strain-
 -21-

displacement relationship.

For element types for which the desired numer of constraints is larger than

one, the situation is only slightly more complicated. Consider for instance the
8-noded quadrilateral plane strain element. For this element type, the desired

number of constraints is equal to three. A possible choice for ;a is then

= * + (x-x ) 4x yy+(- (4.14)

a a 0 a o a

where 0 , 4 and ;) are constants, and x , y is the centroid of the

a a a 0 0
element. Following a procedure similar to the one followed before, one readily

derives the relations

0 = V-1 n ,kdV

a a V

= (y-y) ,kdVa (.15)

where Ix a and I are the moments of inertia around the x and yo axes.
A similar procedure can be followed for the 8-noded quadrilateral axisymmetric

element and the 20-noded cube. In each case c.. is defined analogously to (4.12)

in terms of nodal displacements, and the incremental stiffness equations are

derived in the usual way from (4.13).

Finally it should be noted that only the quadrilateral or cubic elements

have a desired number of constraints which corresponds to a simple and symmetric

choice of the dilatational strain increments. For the triangular or tetrahedral

elements the desired number of constraints does not make such a simple choice

possible. If we did not already know that the six-noded plane strain triangle
 -22-

was suitable for incompressible analysis, we would conclude, for example, that

the desired number of constraints is two, which does not correspond to a complete,

symmetric development of the dilatational strain increments.

Because the most significant errors in computed overall load-deflection curves

occur in the fully plastic range, where the deformation gradients may no longer be

infinitesimal, a rigorous finite strain finite element formulation may in some cases

be required. An implementation of the variational principle presented here for such

a formulation is given in Appendix II. It may also be noted that the present varia-

tional principle, eq. (4.1), may be put in a mean stress form by letting K4 = q ,
say, be the variable, and this is suitable for strictly incompressible materials

(K w)
 -23-

5. Numerical Examples

In order to examine the results of the analyses of the preceding sections

in specific numerical solutions, plane strain beam bending, the thick-walled plane

strain tube under internal pressure, and the tensile analogue of the Prandtl punch

problem were solved.

For the first type of problem, the kinematic requirement of pure bending that

plane sections remain plane was enforced, so it was necessary to examine only one

row of elements across the beam thickness. Ten square 4-noded isoparametric ele-

ments were placed across the thickness. Displacement boundary conditions were

imposed, and the total moment was computed from the calculated nodal tractions.

The material was modeled as elastic-perfectly plastic with a Poisson's ratio of

0.3. The problem was also solved by applying identical boundary conditions to

a mesh of the constant dilatation isoparametrics developed in Section 4, and to

a mesh composed of 10 squares, each of which was composed of 4 diagonally crossed

triangles, as presented in Section 3.

The results are shown in fig. 8. The limit moment for plane strain, assuming

a Mises yield condition, is simply

limit 2 (5.1)

where a is the tensile yield stress and h is the beam thickness. The beam

curvature X , is normalized with respect to X , the curvature necessary to

bring the outer beam fibers to yield.

As can be seen, the ordinary isoparametric element does not find a limit load

at all; all elements across the thickness reach yield, and the moment-curvature

curve continues to rise. The steepest curve shown in fig. 8 is the case for which
 -24-

one stress state per element, calculated at the element centroid, is used to

represent the stress state throughout the element. The next steepest curve

results from calculating and storing the stresses at two thickness-direction

Gaussian integration stations per element.

The third steepest curve is the result of using the constant dilatation

quadrilateral element with one stress state per element. Finally, the two

essentially coincident curves which actually find the correct limit load are

the results of the crossed triangles and the constant dilatation quadrilaterals,

calculating and storing the stress state within each element at each Gaussian

integration station.

We may interpret these results by recalling that point-wise incompressibility

is a necessary but not sufficient condition for a limit load to exist in the

finite element model. In fact, it is also necessary that, at limit load, all

deviatoric strain increments be normal (pointwise) to the yield surface. How-

ever, it is not, in general, possible for the constant dilatation quadrilateral

to deform in a manner so that the deviatoric strain increment, which is permitted

to vary within the element, be pointwise normal to a single stress state on the

yield surface. When the stress state was calculated and stored at the Gaussian

integration stations, however, the computed deviatoric stress increments tended to

vanish in the fully plastic range for both the regular and constant dilatation iso-

parametrics. In fact, the difference between the terminal slopes of the moment-

curvature relations for one vs. four stress measures per element was identical for

the regular and constant dilatation isoparametrics.

Thus, for this particular problem, the relative magnitudes of the errors in

the terminal load-deflection curve associated with failure to meet the necessary
 -25-

of deviatoric
conditions of pointwise incompressibility and pointwise normality

strain increments in isoparametric quadrilaterals are given, respectively, by the

terminal slopes of the second and third steepest curves of fig. 8.

The same problem was also solved for the case of linearly hardening material
In this
behavior, with a hardening modulus of one-hundredth the elastic modulus.

problem, a constant terminal slope to the moment-curvature relation is readily

calculable. The results are shown in fig. 9, where it is seen that the ordinary
fully
isoparametric element formulation exhibits much too stiff a response in the

plastic range, a result discussed previously. The essentially coincident terminal

dilatation
slopes of the moment-curvature curves exhibited by both the constant
in error from
quadrilaterals and the crossed triangles were less than one percent

the analytical terminal slope.

The beam-bending problems were also solved with ten irregularly-shaped con-

stant dilatation quadrilaterals across the thickness, and the results were es-

sentially unaffected.

The thick-walled plain strain tube under internal pressure proved a less

dramatic example of the analysis of the preceding sections. A 10-degree sector

of a tube with outer diameter to inner diameter ratio of two was modeled with

five quadrilateral elements, the nodes having equal radial spacing. The problem

was also solved with the same quadrilaterals formed from four triangles with the

additional node lying at the intersection of the diagonals of the quadrilateral.

Radial displacement increments were prescribed at the two nodes on the inner dia-

meter, and all boundary nodal displacements were constrained to be radial in

direction. Pressures were inferred in the virtual work sense from the nodal forces

at the two inner-diameter nodes.

The trend in load-displacement curves obtained was the same as in the beam-
 -26-

bending problem. However, the quantitative differences between loads at a

given deformation were greatly reduced. In fact, the maximum difference in any

computed load at any point in the deformation was less than four percent. The

load obtained when "steady state" conditions were reached was within a few percent

of the known limit load, and the pressure-expansion curves agreed closely with

that obtained by Hodge and White [9] using finite differences and presented

graphically in [10].

Finally, the problem of the plane strain deep, double-edge-notched (DEN)

tensile specimen, shown in fig. 10, was solved using a very fine mesh of first

regular then constant dilatation isoparametrics. This problem is the tensile

analogue of Prandtl's punch problem (see, for example, [101). In this case, the

limit load in terms of the net stress on the ligament is given by

alim = (2+)ao//3 = 2.97 ao (5.2)

for the von Mises yield criterion, where ao is again the uniaxial tension

yield strength.

By symmetry, only one-fourth of the DEN specimen was analyzed. The specimen
width to ligament width ratio was W/b = 10 . The loading was accomplished by

imposing increments of constant end displacement at the upper end of the specimen.

The load was computed both from the nodal tractions along the top row of nodes

and from the average stresses.in the top row of elements. Both methods gave
essentially the same results.

Figure 10 shows the load-displacement curves as computed from the two

finite element formulations. As can be seen, the ordinary isoparametric elements

fail to find the correct limit load, and, in fact, the load-deflection curve con-

tinues to rise at roughly constant rate, exceeding the limit value by approximately
 -27-

25% at the point for which computation was stopped. Indeed, this large error

has accumulated at an end displacement of only about 5 times the displacement

given by extrapolating the linear elastic loading line to the limit load.

The load-deflection curve obtained from the mesh of constant dilatation

quadrilateral elements is indistinguishable from that of the regular isopara-

metric elements up to roughly 60% of the limit load, at which point it begins

to decrease in slope more rapidly. As the deformation proceeds into the fully

plastic region the difference between the two load-deflection curves increases

rapidly. At the point at which computation was stopped, the load was approxi-

mately 3% less than the analytical limit load, and the tangent modulus load-

deflection slope was roughly 0.1% of its elastic value, versus an apparently

not decreasing 5% for the regular isoparametric elements.

Acknowledgement

The authors are grateful for financial support of this work by the U. S.

National Aeronautics and Space Administration, under Grant NGL 40-002-080

to
Brown University.
 -28-

References

El] J. H. Argyris, Continua and Discontinua, Proc. ist Conf. Matrix Methods
Struct. Mech., Wright-Patterson Air Force Base, Ohio, October (1965).

P. V. Marcal and I. P. King, Elastic-Plastic Analysis of Two-Dimensional

Stress Systems by the Finite Element Method, Int. J. Mech. Sci., 9, pp. 143-
155 (1967)

0. C. Zienkiewicz, S. Vallipian, and I. P. King, Elasto-Plastic Solutions

of Engineering Problems, Initial-stress, Finite Element Approach, Int. J.
Num. Meth. in Eng., 1, pp. 75-100, (1969).

[2] J. H. Argyris and D. W. Scharpf, Methods of Elastoplastic Analysis,

ZAMP, 23, pp. 517-552, (1972).

[3) K. Iwata, K. Osakada, and S. Fujino, Analysis of Hydrostatic Extrusion by the

Finite Element Method, Trans. ASME, Ser. B, J. of Eng. for Ind., 94, pp. 697-
703 (1972).

[4] D. M. Tracey, On the Fracture Mechanics Analysis of Elastic-Plastic Materials

Using the Finite Element Method, Ph.D. Thesis, Brown University, Providence,
R. I., (1973).

[5) L. R. Herrmann, Elasticity Equations for Incompressible and Nearly Incompress-

ible Materials by a Variational Theorem, AIAA J., 3, pp. 1896-1900, (1965).

[6] 0. C. Zienkiewicz, The Finite Element Method in Engineering Science, McGraw-

Hill, London, (19717.

[7] C. S. Desai, and J. F. Abel, Introduction to the Finite Element Method,

Van Nostrand Reinhold, New York, (1972).

[8) E. Reissner, On a Variational Theorem in Elasticity, J. Math. Phys., 24,

pp. 90-95 (1950).

[9] P. G. Hodge, Jr., and G. N. White, Jr., A Quantitative Comparison of Flow

and Deformation Theories of Plasticity, J. Appl. Mech. 17, pp. 180-184 (1950).

[10) W. Prager, and P. G. Hodge, Jr., Theory of Perfectly Plastic Solids, Dover,
New York, (1968).

[11] R. Hill, Some Basic Principles in the Mechanics of Solids Without a Natural
Time, J. Mech. Phys. Solids, 7, pp. 209-225, (1959).

[12] R. M. McMeeking and J. R. Rice, Finite Element Formulations for Problems of

Large Elastic-Plastic Deformation, in preparation.

[13] R. M. McMeeking, An Eulerian Finite Element Formulation for Problems of Large

Displacement Gradients, Sc. M. Thesis, Brown University, Providence, R. I.,
(1974).

[14] R. Hill, on Constitutive Inequalities for Simple Materials, I and II, J. Mech.
Phys. Solids, 16, pp. 229-242 and 315-332, (1968).
 -29-

Ratio Ratio
Constraints Nodes Dea.Freedom
Element Type Element Elements Constraints

constant strain 1 1/2 1

triangle

3 1S2-node 2/3
quadrilateral

linear strain
3 2 4/3
triangle

8-node 6 to 8 3 1 to 3/4
quadrilateral

3 1/2 1/3

5 1 2/5
4-

x 6 2 2/3

S> 9 3 . 2/3

Table I. The effect of mesh refinement on the ratio of total degrees

of freedom to total number of incompressibility constraints for some
common two-dimensional finite elements.
 Ratio Ratio
Element Type and Constraints Nodes Deg.Freedom,
Arrangement Element Elements Constraints

5 constant strain

cubes in regular
lattice

8-node
So risoparametric
cubes in 7 1 3/7
oregular lattice

5 linear strain
tetrahedra in 4 7/5 21/20
cube; cubes in
regular lattice

20-node
isoparametric > 16 4
cubes in 16 3/4
r~'-0 regular
lattice

Table II. The effect of mesh refinement on the ratio of total

degrees of freedom to total number of incompressibility constraints
for some common arrangements of three-dimensional finite elements.
 -31-

Figure Titles

1) Load-displacement curves obtained from finite element solutions of plane

strain single edge cracked tensile specimens subjected to mid-ligament
loading and constant end displacement boundary conditions.

2) Finite element meshes used to solve the mid-ligament loading and constant
end displacement single edge cracked tensile specimens.

3) Propagation of incompressibility constraints in a mesh of rectangular four-node

isoparametric plane strain finite elements.

4) Mesh of arbitrarily skewed four-node isoparametric plane strain finite ele-

ments.

5a) Acceptable limit mechanism for plane strain single edge notched tensile speci-
men subject to either mid-ligament loading or constant end displacement.

b) Alternative limit mechanism for mid-ligament loading.

6) Nodal angles for a linear strain triangular finite element.

7) Four constant strain triangular finite elements arranged to form a quadrilat-

eral and its diagonals.

8) Moment-curvature curves obtained from finite element solutions of perfectly

plastic plane strain beam bending.
 -32-

9) Moment-curvature curves obtained from finite element solutions of linearly

hardening plane strain beam bending.

10) Load-displacement curves obtained from finite element solutions of a deep

double edge notched tensile specimen.

11) Propagation of incompressibility constraints in a rectangular mesh of singly

skewed constant strain triangular finite elements.

finite
12) Three-dimensional block of rectangular eight-node isoparametric brick
elements.
 1.4

LIMIT LOAD - -
1.2 -

ID-LIGAMENT
1.0 CONSTANT END
N 1.0 DISPLACEMENT
DLOADING
NET
F/ LOAD POINT END LOAD POINT
F,u
00 0.8

h h

0.4 - - h - h_
2 2

0.2

0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0

Figure 1.
EU
u.
E LOAD POINT/. 0 h
 UEND

4h

3h

h h
Figure 2.
 -35-

yX

Figure 3.

B
Figure 4.
 -36-

I RIGID

450 450

- - = -yy= CONST.
xx yy

II I I I RIGID

F
(a) (b)
Figure 5.
 -37-

-'i-

ea

7.
7r4

a
2

Figure 6.

Figure 7.
 REGULAR ISOPARAMETRICS.
1.4 - ONE STRESS/ELEMENT
STRESS AT EACH INT. STA.

1.2 - LIMIT LOAD

CONSTANT DILATATION ISOPARAMETRICS
1.0 ONE STRESS/ELEMENT
CROSSED TRIANGLES
AND
0.8 CONSTANT DILATATION ISOPARAMETRICS
STRESS AT EACH INTEGRATION STATION
o.
0.6
PERFECTLY PLASTIC M
0.4 PLANE STRAIN
BEAM BENDING hI
0.2
M
0.0
0 2 4 6 8 10 12 14 16 18 20
Figure 8. O
1.4- REGULAR
ISOPARA METRICS
1.2- CORRRECT
TERMINAL
SLOPE

I CROSSED TRIANGLES
AND
0.8. - CONSTANT DILATATION ISOPARAMETRICS

0.6

0.4- PLANE STRAIN

BEAM BENDING, E
BI-LINEAR HARDENING .E -. 01
-- E
0.2 E -
0.0 --J I
0 2 4 6 8 10 12 14 16 18 20
Figure 9.
 3.5-
-LIMIT LOAD
3.0

S 2.5.5 CONSTANT DILATATION

'NET ISOPARAMETRICS
0 2.0
REGULAR ISOPARAMETRICS

1.5
H b
1.0 H/W=3
W/b= 10
0.5

0.1 I 1
I I 1 I I I
0 I 2 3 4 5 6 7 8 9 10
Figure
Figure 10.
 -41-

Figure l1.

Figure 12.
 -42-

Appendix I: Incompressibility Effects in Finite Element Meshes

Lkk = 0 , pointwise, has been shown in the text to enforce

The condition

the unrealistic constraint on rectangular plain strain 4-node isoparametric ele-

ments that c and c must be the same for every element of the grid, and,
xx yy
if the 4-node isoparametrics are arbitrarily skewed quadrilaterials, then the

shear strain increment rate E must also be the same in every element.
xy

In this appendix, the consequences of Ekk = 0 pointwise will be examined

for some other typical mesh configurations in both two and three dimensions to

demonstrate the unrealistic constraints which are enforced upon admissible dis-

placement rate fields.

Consider the array of triangular constant strain elements shown in fig. 11,

and generated by single skewing of a rectangular grid. Since exx + C = 0

yy
one can go from element to element along the band of elements marked * and

show that x and E are the same in every element of such a band.
xx yy

Consider a rectangular array of 8-noded, three dimensional isoparametric

elements, shown in fig. 12. The displacement rate field within such an element

can be expressed as

u
x
u = {a}) + {b) x + {c) y + {d) z + {e) xy + {f} yz

{g) zx + {h} xyz . (AI.1)

uz

The incompressibility constraint, namely

x+x +yy + zz " +-a- + = 0 , (AI.2)

 -43-

requires that {h) = 0 and that the other terms be constrained such that the

direct strains vary linearly within the element according to

xx = P - ay + YZ

y = q - ax - yz (AI.3)

ZZ = -(p+q) + ax + ay

where p , q , a , , and y are constants within an element and are expressible

in terms of nodal coordinates and displacement increment rates of a kind compat-

ible with the incompressibility constraint.

Now consider any inter-element interface of the kind z = constant. The

tangential strains Exx and eyy must be continuous across the interface and,
because
zz
= -(E xx + yy ) always, so must zz
zz be continuous across the inter-
face. But from (AI.3), since czz does not vary with z within an element, and
since czz is continuous across element boundaries, we therefore have the con-

straint that zz has the same value at all points along a line extending in the

z-direction. That is, Czz is independent of z , and, by similar reasoning,

xx is independent of x and yy is independent of y .

In fact, by further considerations of continuity, it may be shown that the

normal strain rates throughout the entire array are given by equations of
the
kind
=
ax B(y) - C(z)

= C(z) - A(x) (AI.4)

yy
zz = A(x) - B(y)

where A(x) , B(y) , and C(z) are each continuous functions of their respective

arguments, varying linearly within elements.

 -44-

To see how severely this constrains admissible incremental deformation

fields, suppose that the plane z = 0 corresponds to a fixed boundary on which

all displacements must be zero. Then cxx and cyy vanish on this plane and,

by the argument advanced earlier, so does E Hence, from (AI.4),

zz

A(x) - B(y) = 0

C(O) - A(x) = 0 (AI.5)

B(y) - c(0) = 0

which requires that A(x) = B(y) = C(O) , and hence the most general admissible

deformation field for the entire array of elements is

S = - = C(z) - C(0)
y XX (AI.6)
E; = 0 .
zz

If we further suppose that the plane x = 0 also coincides with a fixed

boundary, then

= = = 0 (AI.7)
xx yy zz

throughout.
 -45-

Appendix II: Generalization to Finite Deformations

For the class of elastic-plastic materials deformed at finite strain, which

admit the existence of a constitutive rate potential, Hill [ll] has shown that

the solution of boundary value problems can be obtained from the following varia-

tional principle, and an Eulerian finite-element formulation for problems of

large plastic flow has recently been based upon it [12,13]:

S U(D .)dV - dV - vd 0 .
(AII.1)

Here V is the current (deformed) volume, and ST is the portion of the current

boundary on which nominal traction rates T.1 , based on the current state as the

reference state, are prescribed. The rate of deformation tensor is D.. , a..
av. ij 13
is the Cauchy (true) stress tensor, and - is the velocity gradient with respect

to current coordinates. The function U(Dij) is homogeneous of degree 2 in D.i

and has the property that

V au
Ti aD . - ijkg D * (AII.2)
1]

Here Tijk£ are the instantaneous moduli appropriate to the adopted measure of

stress rate, and are possibly dependent on the direction of D.. . The stress rate
1)
V
itself, rij , is the Jaumann, or co-rotational, rate of Kirchhoff stress, based on

a reference state that has been chosen to coincide instantaneously with the current

state, and Kirchhoff stress t.. is defined as a.. times the ratio of reference

state to current material density, so that Tij = oij instantaneously. Note that
the existence of U requires that moduli have the symmetry
 -46-

and this ensures a symmetric incremental stiffness matrix for the corresponding

finite element equations. The class of materials which admit rate potentials

include all elastic materials for which a strain energy function exists, and also

all elastic-plastic materials which satisfy the normality rule when phrased in

terms of work conjugate stress and finite strain measures [14].

A generalization of the Prandtl-Reuss equations for finite deformation [12,13]

which admits the rate potential U is obtained by relating the Jaumann rate of

Kirchhoff stress to the rate of deformation tensor in the usual Prandtl-Reuss

form:

e l+v V v V
ij E 1T E ij kk
, , V (AII.4)
D?. ij k£ kk
13 4hr-2

and

D.. De. + D'.

1 1ij 1)

-2 3
for continued plastic loading. Here T = Tr .. and h is the slope of the
2 ijij
stress-logarithmic plastic strain curve for simple tension. Note that in this

case, and for other plastically incompressible materials, .. differs from

13
V
a only by the term (1-2v)a ikk/E , and this is negligible by comparison for
kijk
almost all. technological applications of the theory to metals.

Thus equations (AII.4) can be inverted, and the resulting Jaumann rates of

Kirchhoff stress separated into hydrostatic and deviatoric parts to give

 -47-

j l+V j -a2 2h 'E

3 l+v
V E (AI 5)
vkk = 2 Dkk = 3KDkk (AII.5)

1 if at yield and rktDkt > 0

0 otherwise .

Thus the deviatoric and hydrostatic Jaumann rates of Kirchhoff stress have

functional dependence only on the deviatoric and hydrostatic parts of the rate

of deformation tensor, respectively.

From these relations, the volume integral of the rate potential

U= - TijD..i in (AII.1) can be separated into deviatoric and hydrostatic parts to

obtain

Si.DijdV = !.D!. + I KD kdV . (AII.6)

V 2 13 -
2 ij I 2 kk A

Because this integrand has the same functional form as the conventional

small-strain rate potential for Prandtl-Reuss materials, the variational principle

1 2
of Section 4 can be readily generalized by replacing the term I KDkk in (AII.6)
2 kk
with (Dkk 12) , as in (4.1). As in the text, independent variation of the

velocities and of 4 furnishes = Dkk , the dilatation rate.

The finite element implementation of the finite strain version of the varia-

tional principle of Section 4 then follows straightforwardly, with the total

element stiffness matrix being the sum of the stiffness corresponding to the modi-

fied rate potential and the "initial stress" stiffness, which corresponds to the

second volume integral of (AII.1).