1  Introduction

The boundary element method (BEM) is the most potent tool for the solution of linear elliptic partial differential equations (PDEs) in an exterior domain. The standard BEM also serves as an attractive alternative to methods such as the finite element and finite difference methods for interior linear elliptic problems. The BEM has thus become an established computational method over recent years (Jaswon and Symm (1977), Brebbia (1978), Banerjee and Butterfield (1981)). However, apart from suffering the limitation of being inappropriate for non-linear problems, the BEM is unable to cope directly with discontinuities in the variables in the domain of the PDE. The discontinuity will be assumed to have the topology of a shell - an open surface in three-dimensional problems, a line in two dimensions. An illustration of the general domain is given in figure 1.


The traditional BEM is derived from a boundary integral equation (BIE) formulation of the PDE by dividing the boundary into boundary elements and applying an integral equation method (usually collocation) to obtain the method of solution. For domains in the form of figure 1, the traditional BEM can be applied by subdividing the domain into subdomains, as illustrated in figure 2. Boundary integral equation reformulations of the PDEs on each subdomain can now be obtained and, through coupling these equations across common boundaries, the solution throughout the domain can be obtained.


A similar method to the traditional BEM for the solution of a PDE in the infinite domain exterior to a shell discontinuity can be derived through recasting the PDE as an integral equation termed a shell integral equation (SIE). A numerical method can then be derived in a similar way to the BEM (see, for example, Ben Mariem and Hamdi (1987) or Warham (1988)). However, the main value of shell elements is in their use in conjunction with boundary elements. In this paper, it is shown how the PDE in the domain of figure 1 can be reformulated straightforwardly as an integral equation termed a boundary and shell integral equation (BSIE) and thus how a numerical method termed the boundary and shell element method (BSEM) can be derived.
Boundary element methods have traditionally fallen into two distinct classes, direct BEMs and indirect BEMs, based on direct and indirect integral equation formulations. In this paper direct and indirect boundary and shell integral equation formulations are given for the interior two-dimensional Laplace equation. The BSIEs are a hybrid of the corresponding direct or indirect boundary integral equation with the shell integral equation. Hence new integral equation based methods for the solution of linear elliptic PDEs with discontinuities in its domain are introduced in this paper. The methods are demonstrated on the two-dimensional interior Laplace problem
Ñ2 j(p) = 0  .
An illustration of the domain is given in figure 1. It consists of a region D with boundary S and with shell discontinuities G. In order to specify the problem fully a conditions for points on the boundary and on the shell must be stated. These are the boundary condition and the shell condition.
In Warham (1988), the shell integral equations are derived by first assuming that the shells have finite thickness and hence the standard boundary integral equation formulation is valid. The shell thickness is then allowed to approach zero. A similar limiting process can be used to derive the boundary and shell integral equation by assuming S to be fixed and taking the limit as the thickness of the shells approach zero. The integral equation formulations of the interior Laplace problem are stated in section 2.
In order to derive a particular method, the boundary and shell are divided into uniform elements and the functions defined on the boundary and shell are approximated by a constant on each element. The integral equation method is then derived through collocation. The methods are demonstrated through their application to a test problem where the domain is the unit square and a discontinuity lies between ([1/2],[1/2]) and ([1/2],1).