Restriction matrices for numerically exploiting symmetry (Q2479575)

The authors develop a technique for exploiting symmetry in the numerical treatment of boundary value problems (BVPs) and eigenvalue problems. This technique exploits a number of existing ideas from other approaches but also contains new features. The starting point is a finite group \({\mathcal G}\) of congruences of \(\mathbb R^m\) under which the domain of definition of the BVP or eigenvalue problem is invariant. Both Abelian and non-Abelian groups are considered. Then the restriction matrices are constructed, starting from a system of irreducible representations of \({\mathcal G.}\) Using these matrices, the authors can decompose the invariant discretized problem into independent problems with reduced dimension. The paper gives a good survey of the existing literature and contains extensive numerical tests on a variety of 2D and 3D problems. However, except in one case there is no explicit comparison with other methods, with regard to either speed or accuracy.