An FEM-CSM combined method for planar exterior Laplace problems (Q5944162)

This paper is concerned with the solution of the Dirichlet problem for the Poisson equation, \(-\Delta u=f\), in an infinite planar region \(\Omega\) exterior to a smooth curve \(C\). The solution \(u\) is assumed bounded at infinity and the charge density \(f\) is assumed to have compact support. The problem is solved by dividing the exterior region into a portion inside a disk of radius \(s\), \(D_a\), and a portion outside. Denote by \(\Omega_i\) the intersection \(\Omega\cap D_a\) and by \(\Omega_e\) the (infinite) remainder. If an appropriate boundary operator \(b(u,v)\) can be defined on the circle \(\partial D_a\), then the problem breaks into two problems: (1) solve the Poisson equation on \(\Omega_i\) with Dirichlet boundary conditions on \(C\); and, (2) find a bounded solution of the Laplace equation on \(\Omega_e\), each with boundary conditions on \(\partial D_a\) induced by the operator \(b\). An approximate solution on \(\Omega_i\) can be found using the finite element method (FEM) and an approximate solution on \(\Omega_e\) can be found in terms of Fourier series using the charge simulation method (CSM). The full problem, then, reduces to finding an approximate boundary operator \(b\). For two functions \(g,h\in H^{1/2}(\partial D_a)\), define the operator \[ b(g,h)=2\pi\sum_{n=\infty}^{\infty}|n|g_n\overline{h_n} \] where \(g_n\) and \(h_n\) are Fourier coefficients of \(g\) and \(h\), respectively. This operator is non-local, but it can easily be approximated as a finite Fourier series. If the circle \(\partial D_a\) is divided into \(N\) equally-sized arcs, then the Fourier series arising in the charge simulation method involves the same coefficients as appear in \(b\). When those arcs are also chosen as boundaries for some finite elements used to find an approximate solution in \(\Omega_i\), finding a numerical solution becomes practical. The main result in the paper is to show that the approximate method can be carried out and that the error in the approximation is \(O(1/N)\). The proof focusses on \textit{a priori} estimates and spectral properties of the operator \(b\) and its finite dimensional approximation.