Construction of Preconditioners for Wiener-Hopf Equations by Operator Splitting.pdf

Construction of Preconditioners for Wiener-HopfEquations by Operator SplittingMichael K. Ngy, Fu-Rong Linz and Raymond H. ChanxNovember 18, 1993AbstractIn this paper, we propose a new type of preconditioners for solving nite sectionWiener-Hopf integral equations (I +A )x = g by the preconditioned conjugategradient algorithm. We show that for an integer u 1, the operator I +A canbe decomposed into a sum of operators I + P(u;v) for 0  v u. Here P(u;v) aref!vg-circulant integral operators that are the continuous analog of f!vg-circulantmatrices. For u  1, our preconditioners are dened as (1=u)Pv(I + P(u;v) )1.Thus the way the preconditioners are constructed is very similar to the approachused in the additive Schwarz method for elliptic problems. As for the conver-gence rate, we prove that the spectra of the resulting preconditioned operators[(1=u)Pv(I + P(u;v) )1][I +A ] are clustered around 1 and thus the algorithmconverges suciently fast. Finally, we discretize the resulting preconditioned equa-tions by rectangular rule. Numerical results show that our methods converges fasterthan those preconditioned by using circulant integral operators.Abbreviated Title: Splitting of Wiener-Hopf Integral Operators.Key Words. Wiener-Hopf integral operator, projection method, preconditioned con-jugate gradient method, f!vg-circulant integral operator, Fourier transform, rectangularrule.AMS(MOS) Subject Classications. 45E10, 45L10, 65R20, 65J10.Research supported in part by HKRGC grant no. CUHK 178/93E.yDepartment of Mathematics, The Chinese University of Hong Kong, Shatin, Hong Kong.zDepartment of Mathematics, University of HongKong, Pokfulam Road, Hong Kong.xDepartment of Mathematics, The Chinese University of Hong Kong, Shatin, Hong Kong.1 1 IntroductionWiener-Hopf integral equations arise in a variety of practical applications in mathematicsand engineering especially in the solutions of inverse problems. Typical examples arelinear prediction problems for