MATHEMATICS OF COMPUTATION
VOLUME 41. NUMBER 164
OCTOBER 19R3. PAGES 399-423
Analysis of Some Mixed Finite Element Methods
for Plane Elasticity Equations
By J. Pitkäranta and R. Stenberg
Abstract. We analyze some mixed finite element methods, based on rectangular elements, for
solving the two-dimensional elasticity equations. We prove error estimates for a method
proposed by Taylor and Zienkiewicz and for some new variants of the known equilibrium
methods. A numerical example is given demonstrating the performance of the various
algorithms considered.
1. Introduction. In the numerical solution of problems of continuum mechanics,
the stresses are normally of primary interest in the elastic region. It is therefore
natural to design the numerical algorithms so that the stresses can be obtained
directly without first computing the displacements. Such methods can be derived
from the dual variational formulation of the elasticity problem. The corresponding
finite element algorithms are usually formulated as mixed methods where both the
displacements and the stresses are first approximated, and the displacements are
then eliminated from the discrete equations. In many cases the elimination can be
rather effectively done using penalty/perturbation techniques or their iterative
variants; cf. [3],[11],[12].
The best known finite element methods of the above type are the so-called
equilibrium methods, first proposed by Fraejis de Veubeke [17] (cf. also [14], [16],
[18]) and analyzed theoretically by Johnson and Mercier [9] (cf. also [8]). In these
methods, one uses specific composite elements which allow the equilibrium condi-
tion between the stresses and the volume load to be satisfied exactly in the case
where the volume load is zero.
The main drawback of the equilibrium methods proposed so far is the relatively
high number of free parameters as compared with displacement methods of the same
order of accuracy. For example, if the composite quadrilateral element of [17], [9] is
used on a regular rectangular grid, one has eight degrees of freedom per each interior
node of the grid (after the local condensation of three extra degrees of freedom per
node, cf. [9]) and the convergence rate 0(h2) for the stresses in L2 [9]. On the other
hand, using the displacement method with reduced biquadratic elements (cf. [6]), one
has the same convergence rate with six parameters per node, so the displacement
method seems superior.
It is clear from the above example that the mixed or equilibrium methods should
be further developed if they are desired to be competitive with displacement
Received August 18, 1982.
1980 Mathematics Subject Classification. Primary 65N30.
©1983 American Mathematical Society
0025-5718/83 $1.00 + $.25 per page
399
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use