Fast wavelet iterative solvers applied to the Neumann problem (Q4498296)

The authors study the Neumann problem \[ -\Delta u= f\quad\text{in }\Omega,\quad{\partial u\over\partial n}= 0\quad\text{on }\partial\Omega,\tag{1} \] where \(\Omega= ]0, 1[^d\). The solution of (1) is not unique. If (1) is discretized by a finite difference or finite element scheme to a linear system \[ Au= b,\tag{2} \] with \(b\in \mathbb{R}^N\), then the solution (2) also is non-unique, though \(b\) is in the range of \(A\). The efficiency of direct methods and performance of iterative methods cannot be guaranteed for \(d\geq 2\) because of the singularity of \(A\). If one projects (2) onto \((\text{Ker}(A))^\perp\), then the condition number grows as \(N^2\). It was recently shown that for elliptic boundary value problems (non-singular systems!) in wavelet bases the condition of the system is bounded independently of \(N\). The aim of the present paper is to extend the use of wavelet bases to singular problems to construct preconditioners of \(A\). The idea is to apply to (2) a discrete orthogonal wave transform \(Q\). The study is restricted to \(d= 1\) and two methods are proposed to deal with the singularity of the wavelet-transformed linear system. The algorithms are described explicitly and an extensive series of tests is given with a careful discussion of the complexity. The authors also point to related and future work.