Numer.

Math 26, 79---84 (t976)

9 by Springer-Verlag t976

On Finite Element Methods for Plasticity Problems

Claes Johnson
Received August 8, 1975

Summary. We prove an error estimate for an incremental finite element method

for obtaining approximations to the stresses in an elastic-perfectly plastic body. We
also comment on the limit load problem.

Introduction
In this note we shall prove an error estimate for an incremental finite-element
(Galerkin) m e t h o d for obtaining approximate solutions of the following problem:
Find the stresses in an elastic-perfectly plastic body, obeying the Prandtl-Reuss
flow rule, under the action of a time dependent load. This problem can be formula-
ted as a variational inequality with a time dependent constraint. Existence of
an exact solution of this variational inequality was proved in [5]. We shall con-
sider a finite-element method where the equilibrium equations and the yield
condition are satisfied exactly ("internal" method), c.f. (t.3) below.
We shall also comment briefly on the limit load problem (see e.g. [l], [7]).

1. The Plasticity Problem

Let us introduce some notation: Let ~ be a bounded domain in R a = { x =
(xt, x,, x3) : xi ER} and let I = [0, T], T > 0, be a time interval B y a (7 and Z)
we shall denote (stress) vectors in R e depending on (x, t) E ~ • I with components
a#, i, j-----t, 2, 3, such t h a t aii ~ aj ~. Below we shall use the convention t h a t repeat-
ed indices indicates s u m m a t i o n from t to 3. Derivatives of non-smooth functions
are to be interpreted in the distribution sence. Define
H = [L2(~)]6,
(x,z)= f xiiziidx, *:,xEH,
D

a (7, ~r = fA~/kh,~/Zkhd x,
D

where the A~ikh denote elasticity constants such that A~ikh=Aiikh=Akh~i, and
for some positive constant r, a (7, 7 ) > fl lit ]]2, ~EH.
Further, let #": R 6-+ R be a given continous convex function such that 0 EintB,
where B = { T E R 6 : # ' ( 7 ) ~ 0 } . The function #" defines the yield condition and
80 C. Johnson

the set P defined by

P - - { z E H : ~(x)EB a.e. on ~2},
denotes the set of stress vectors satisfying this condition. Finally, let V =
[L~(~)] 3, l e t / : I---~V be a given function w i t h / ( 0 ) = 0 and define
E(t)-----{~EH: D r = / ( t ) } ,
K (0 = E (0 :n P.
We shall assume that the load ] has the form
l(t, x)=g(0 f(x),
where F E V and g EC2(I). A common choice in practice is J(t, x)= t. F (x).
We shall seek approximate solutions of the following
Plasticity Problem. Find a(t)EK(t), tel, such that a.e on I,
a(a'(t), ~--a(t)) >0, ,EK(t), a(0)=0, (t.t)
where a'=da[dt.
This is the problem of finding the stresses in an elastic-perfectly plastic body,
obeying the Prandtl-Reuss flow rule, under the load ] in the case when the dynami-
cal forces are considered to be negligible (quasi-static case). No boundary condi-
tions are imposed on a. This means that we consider the case when the displace-
ments of the body are prescribed on its boundary. For a discussion and motiva-
tion of the formulation (t.l), see [4], [5].
A necessary condition for existence of a solution of (t .t) is clearly that K(t) 4=0
for t EI. Existence and uniqueneess of a solution a of (t.t) satisfying a EL ~~([0, T] ; H)
and a'EL2([O, T ] ; H ) was proved in [5] under the following somewhat stronger
condition:
There exists ~:EK(~), where ] g ( t ) ] = max ]g(t)],
such that for some positive constants C and 6, (t.2)
]~(x)]~C for xE,Q and 4-(t+6)~EP.
Note that if B is bounded in R 6 and g(t)=t, then a solution of (tA) exists on
[0, T] if K ( T + 6 ) :4=0 for some 6 > 0 .
For the purpose of finding approximate solutions of the Plasticity problem,
let E h (0, 0 < h < t, be a finite dimensional affine space satisfying E h (t) ( E (t), t EI.
The space E h (0 could be constructed for instance by using piecewise polynomial
functions defined on some finite element subdivision of ~Q (see Example t below).
The parameter h could then be the maximal diameter of the finite elements.
Furthermore, let the convex set B h be an approximation of the set B such that
B h ( B and define P h = { z E H : z(x)EB h a.e. in~}. In practice B h is often a piece-
wise linear approximation of B, i.e,,
B , = { r E R e : Fi,h(~ ) ~ 0 , i = ~ . . . . . Mh},
where the Fi,~ are certain linear functions. Defining K h(t) = E h (t) c~ Ph, we then
have
K~(t) <K(t), t~l. 0.3)
 Finite Element Methods for Plasticity Problems 8t

We shall use a finite difference discretization in time. Let k = T/N, N a posi-

tive integer, t * = n k, k = 0 . . . . . N, ~ = r (tn), and introduce the difference quotient
~" = (,'- r"-l)/k.
Let us now pose the following discrete problem: Find a~kEKh(t.), n=O . . . . . N,
such that for n = 1 . . . . . N,

a ( e ~ , ~-aL) ~o, ~K~(t,), ~L=o. (t.4)

We shall assume that the following analogue of condition (t .2) is satisfied:
There exists Zh E Kh (7) such that for some positive
constants C and ~ independent of h, [Xh(x)[ __C (t.5)
for xEQ and : t : ( I + 6 ) Z h E Ph-
Existence and uniqueness of a solution of (1.4) follows from the fact t h a t for
n = 1 . . . . . N, a ~ is alternatively characterized as the function which minimizes
the strictly convex functional
89 ,. )--a(o~;fl. ),
over the closed convex set Kh(tn). Thus, in order to find ahk one has to solve a
convex minimization problem at each time level tn. For simplicity we assume that
this problem is solved exactly. In practice the minimization problem is solved
only approximately usually using some iterative method.
It will be convenient to assume also that
g(t)E[g(t~_x),g(~n) l for tEI~, (t.6)
where I , = [t, -x, t,]. To ensure (1.6) it m a y be necessary to work with a non-
uniform partition {to. . . . tN} of I. The argument in this situation is, except for
obvious changes, the same as that given for a uniform partition. Without loss
of generality we m a y therefore restrict ourselves for simplicity to the case of a
uniform partition.
The following a priori estimate is crucial for the error analysis below. The proof
is the same as t h a t of L e m m a 2 in [5] and is not repeated here. We use the nota-
tion

L e m m a 1. There is a constant C such t h a t for k sufficiently small, 0 < h < 1,

II c.
Define
e (h, k ) = i n f []a-- ~][,,(n),
where the infimum is taken over all , such that ~'~EKh(t,), n = 0 . . . . . N. The
q u a n t i t y e (h, k) is a measure of how well the exact solution a (t) can be approxi-
mated b y functions in K h (t). The magnitude of e (h, k) is essentially ~letermined
by the approximation properties of E h (t) and the smoothness of the exact solution
a. Unfortunately, the information we have about the latter is too weak to make
it possible to obtain a n y estimates of e (h, k).
The result of this note is the following error estimate.
6 Numer. Math., Bd. 26
 82 C. Johnson

T h e o r e m 1. Assume t h a t (t.2) and (1.5) hold and let a and a,h be the solutions
of (tA) and (1.4) respectively. Then for k sufficiently small, O < h < t,
m .a x IId ' - - a l , II--- c(V~R~,k) + #)
Proo] o] Theorem 1. We have
a(a', •:--a) >0, xs a.e. on I, (1.7)
a(ea**, ~-~'~,) >0, TeKh(t.), n = t . . . . . N. (t.8)

L e t us extend a** to I b y setting

a**(t)=h(t) ~'~+O--h(t)) ~,, te_r.,
where
h(t)= g(t)-g(t.) for teI. if g(t.)#:g(t~_~),
g (tn_x) - g (tn)
h(t)=(t.--t)/k for tel. if g(t.)--g(t._x).
Then ahk(t)EE(t), tel, and since b y (t.6), O ~ h ( t ) ~ t and P is convex, we also
have t h a t a hk (t) E P, t E I, so t h a t ah k (t) EK (t), t E I. T a k i n g now z = a hk in (t.7), we get
a(a',~hk--a)~O a.e. on/.
Integrating this inequality over I s and dividing b y k, we obtain
1
a(Or r T f a(a', a'~, -- a** (t) + a (t) -- o~) dr. (t .9)
in
L e t now TleK,(t,,), n=O . . . . . N, be such t h a t Ila-~,ll,,~.)~2e(h, k). T a k i n g
,-----vn in (t.8) and adding the inequality to (t.9), we see t h a t with e=a--a**,

a (~e", e")<a(or ,I-r + Ir.I,

where r, denotes the right-hand side of (t .9). Multiplication b y k and s u m m a t i o n
over n shows t h a t
37

max 11e" I1' --~ c 11~ <',- B,,<-~ II~,- <'It,.<,,, + c k Y, I r. I.

n k=l
Since

Ir.I ~- ll'(Olll:hllO4~ll+V~ Ik"ll'as '] at

1.

c ,,.(f II<,'II'a O's' IiooI~II+ c i.f I1='II'a,

sc f ll<,'ll,at+cnoe,li, k,
we therefore conclude, using L e m m a I and the fact t h a t a' EL2( [0, T] ; H), t h a t
max Ile'll < C (V~Ch,k)+ ~).
This completes the proof of Theorem t.
 Finite Element Methods for Plasticity Problems 83

Example 1. Let us now indicate one w a y of constructing a finite element

space K,(t) satistying Kh(t ) (K(t) which is based on the use of so-called stress-
functions. For simplicity we shall consider a two-dimensional problem. We shall
then use the same notation as above with the natural changes from three to
two dimensions. Let us also assume t h a t Ph-----P.
Let 14~ be a piecewise polynomial space based on a regular triangulation of $2
such that Wh ( C 1(/2). Let h be the m a x i m a l diameter of the triangles of the triangu-
lation. To construct such a space Wh we can use for instance the Clough-Tocher
element (see [3]). Define

It is then easy to check t h a t Dr----0 for z = (rn, zx~, r2~)EE ~ Let n o w e E H be a

particular solution of the equilibrium equations, i.e.,
De=/ on ~9•

Defining E h (t) = {e (t) + , : r E E~ we then clearly have E h (t) ( E (t), so t h a t with

Kh(t)=Eh(t)c~P, we obtain Kh(t)(K(t ). It is easy to see that if a and ff are
sufficiently smooth o n / 2 • I, then e (h, k) < Ch (if the Clough-Tocher element is
used).
Let us also mention that in the case of elastic-perfectly plastic plates, which
can be analyzed in a completely analogous way, one can obtain finite element
spaces satisfying the analogue of (1.3) b y using the H e l l a n - H e r r m a n n element or
the Visser element (see ~2], [6~).

2. A Note on the L i m i t Load P r o b l e m

Let us assume in this section t h a t / ( t , x)=tF, FEV, and for simplicity that
~-(v) = IT [ - - t . The limit load problem is to find
-=- sup {t: K (t) # 0}.
It follows from the results of [4] quoted above that the elastic-plastic b o d y can
resist loads of magnitude t. F for t < t', whereas this is not possible for t > t.
In order to find an approximation of t it is natural to consider the discrete
problem: Find
~,= sup{t: g,(0 #~}.
Since K h (t) ( K (t), we clearly have t'h < t'. In order to get an estimate of a qualitative
nature of t" from above, we argue as follows: Let T E K ( 7 - - e ) , e > 0 ; we can for
instance choose ~ = a ( t - - e ) , where a is the solution of the Plasticity problem.
Let now ~hEE h (~ -- e) and set

Then since 9 E P, we have

t ~,EPnE,(~).

6*
84 C. Johnson

t--8
T a k i n g here K h (t) = E h (t) n P , we thus have K h ~ ~= O, so t h a t

Thus, the error ~ - - t'h depends on how well one can a p p r o x i m a t e a s u i t a b l y chosen
function x ~ K (t) in the m a x i m u m n o r m for t close to [ with functions in E h (t). As
above, the more d e t a i l e d knowledge of the r e g u l a r i t y of such a , n e e d e d to o b t a i n
e s t i m a t e s for [ - - th is, as far as we know, n o t available.

References
t. Anderheggen, E., Kn6pfel, H. : Finite element limit analysis using linear program-
ming. Int. J. of Solids and Structures 8, t4t3-143t (1972)
2. Bitcklund, J. : Mixed finite element analytis of elasto-plastic plates in bending.
Chalmers Inst. of Technology, G6teborg, 1972
3. Ciarlet, P. G.: Sur l'6Mment de Clough et Tocher. R. A. I. R. O. R-2 (t974), t9-24
4. Duvaut, G., Lions, J. L. : Les in6quations en m6chanique et en physique. Paris:
Dunod t 972
5. Johnson, C. : Existence theorems for plasticity problems. Submitted to J. Math.
Pures et Appl.
6. Johnson, C. : On the convergence of a mixed finite element method for platebending
problems. Numer. Math. 21, 43-62 (t973)
7. Strang, G. : The finite element method--linear and nonlinear applications, Proceed-
ings of the International Congress of Mathematicians, Vancouver, Canada, t974

Claes Johnson
Dept. of Mathematics
Univ. of Chicago
Chicago, Ill. 60637
USA