A method of approximating stiffness matrices for some pseudodifferential equations (Q1105336)

The author presents a method for computing some entries in a stiffness matrix in terms of the others. The stiffness matrix can arise from either a partial differential or a pseudodifferential operator discretized over a uniform square mesh. The underlying idea is to relate values of the basis functions in mesh squares with sides of length h to those on mesh squares with sides of length 2h. This relation plus homogeneity of the operator can be used to express the stiffness matrix in terms of a finite subset of set of integrals which do not depend on h. The author presents applications to the pseudodifferential operator \(A(v)={\mathcal F}^{-1}(| \cdot |^{\alpha}{\mathcal F}v)\) where \({\mathcal F}\) denotes the Fourier transform and \(\alpha =\pm 1\), corresponding to the so-called crack and stamp problems. In those cases, certain of the stiffness matrix entries depend on singular integrals and others depend only on more easily approximated integrals. It is the entries depending on singular integrals which he expresses in terms of the others, thereby gaining accuracy in the numerical approximations. Numerical examples are provided.