Boundary approximation methods for solving elliptic problems on unbounded domains (Q917243)

The authors apply the boundary approximation method (BAM) to the model problem \(-\Delta u+u=0\) in \(\Omega =\{(x,y):\) \(x>0\), \(y>0\}\) under the boundary conditions \(u=1\) for \(x=0\) and \(0\leq y\leq 1\), \(\partial u/\partial x=0\) for \(x=0\) and \(y>1\), and \(u=1\) for \(y=0\) and \(x\geq 0\). This boundary value problem contains three kinds of singularities: 1) a ``crack'' singularity at (0,1); 2) an ``infinity'' singularity; 3) a ``mild'' singularity at (0,0). The BAM is derived from the well-known Trefftz method, but uses different expansions of the solutions in different subdomains into which the original domain is decomposed. For the model problem, \(\Omega\) is decomposed into 3 subdomains each of them contains one kind of singularities. Because the expansions satisfy the partial differential equation in the subdomains and the boundary conditions on \(\partial \Omega\), it remains to choose the free coefficients in the expansions such that the interface conditions on the interfaces between the subdomains are satisfied in a least-squares sense. The numerical experiments performed show that highly accurate solutions can be obtained very efficiently by means of a few expansion terms only.