MIXED FINITE ELEMENT METHODS FOR ELLIPTIC PROBLEMS* DOUGLAS N. ARNOLD† Abstract. This paper treats the basic ideas of mixed finite element methods at an introductory level. Although the viewpoint presented is that of a mathematician, the paper is aimed at practitioners and the mathematical prerequisites are kept to a minimum. A classification of variational principles and of the corresponding weak formulations and Galerkin methods—displacement, equilibrium, and mixed—is given and illustrated through four significant examples. The advantages and disadvantages of mixed methods are discussed. The concepts of convergence, approximability, and stability and their interrelations are developed, and a r´ esum´ e is given of the stability theory which governs the performance of mixed methods. The paper concludes with a survey of techniques that have been developed for the construction of stable mixed methods and numerous examples of such methods. Key words. mixed method, finite element, variational principle 1. Introduction. The term mixed method was first used in the 1960’s to describe finite element methods in which both stress and displacement fields are approximated as primary variables. We begin with the most classical example, the system of linear elasticity. The equations of linear elasticity consist of the constitutive equation AS = E (u) in Ω and the equilibrium equation div S = f in Ω. Here Ω denotes the region in three dimensional space, R 3 , occupied by the elastic body, u :Ω → R 3 denotes the displacement field, E (u) denotes the corresponding infinitesimal strain tensor, (i.e., the symmetric part of the gradient of u, ǫ ij (u)=(u i,j + u j,i )/2)), f denotes the imposed volume load, and S :Ω → R 3×3 s (the space of symmetric 3 × 3 tensors) denotes the stress field. The divergence of S , div S , is applied to each row of S , so that (div S ) i = ∑ j s ij,j . The material properties are determined by the compliance tensor A which is a positive definite symmetric operator from R 3×3 s to itself, 1 possibly depending on the point x ∈ Ω. The constitutive equations can equally well be written as S = C E (u) in Ω *This work was supported by NSF grant DMS-89-02433. †Department of Mathematics, The Pennsylvania State University, University Park, Pennsylvania 16827. 1 This means that the action of A can be written as (AS ) ij = ∑ kl a ijkl s kl with the components a ijkl satisfying the usual major symmetries a ijkl = a klij , minor symmetries a ijkl = a jikl , and positivity condition ∑ ijkl a ijkl s ij s kl ≥ γ ∑ ij s 2 ij , for all S , where γ> 0. Typeset by A M S-T E X