A wavelet algorithm for the solution of a singular integral equation over a smooth two-dimensional manifold (Q1286348): Difference between revisions

The subject of the present paper is to apply the wavelet technique to the piecewise bilinear collocation solution of singular integral equations over a smooth surface. In particular, the collocation for the singular boundary integral equation corresponding to the oblique derivative problem for Laplace's equation, or the similar equation for the Molodensky problem is included. If the underlying surface is smooth (continuously differentiable up to a certain order) and diffeomorphic to the torus, then it is clear that the wavelet algorithms admit high order compressions. For general smooth surfaces represented by a set of parametrizations, similar results hold if the wavelet functions are suitably chosen. Supposing that the parameter domains are squares, one can define the wavelets of the trial space as tensor products of the orthogonal wavelets and scaling functions over the interval. However, due to the orthogonality, these wavelets are not optimal. Indeed, to reduce the amount of work for the quadratures applied during the computation of the stiffness matrix, wavelets with smaller supports but with the same moment conditions seem to be preferable. Thus, the piecewise linear univariate biorthogonal wavelets are considered. These wavelets have the smallest support among all the piecewise linear wavelets with two vanishing moments. By reflection techniques boundary wavelets are defined and a stable wavelet system (Riesz basis) over the interval is obtained. Applying well-known tensor product techniques, a wavelet basis over the square is introduced, and using the parametrization mappings, the author ends up with continuous wavelet functions over the boundary manifold. For these wavelets, the Riesz basis property and the usual decay property for the coefficients of a smooth bilinear function are proved. A wavelet basis for the space of test functionals is defined. The wavelet algorithm which is based on the introduced bases in the test and trial spaces is described. It is shown that the \(n\times n\) stiffness matrix corresponding to the wavelet bases admits a compression up to a matrix with no more than \(O(n[\log n]^4)\) nonzero entries and that, replacing the full stiffness matrix by the compressed matrix, the same asymptotic convergence rate \(O(n^{-1})\) as for the conventional collocation solution is obtained. For this estimate, the second order moment condition for the wavelets along the common boundary of two subsurfaces defined by different parameter representations is not necessary. A quadrature algorithm for the computation of the compressed stiffness matrix with no more than \(O(n^{4/3}[\log n]^{4/3})\) operations is introduced. The corresponding error of the discretized collocation solution is less than \(O(n^{-1}\log n).\) The global analyticity of the parametrizations is not necessary. Local analyticity is sufficient. More exactly, if the thrice continuously differentiable surface is given by certain grid points and if this surface is replaced by a suitable interpolation, then the twice continuously differentiable and piecewise polynomial parametrizations may be supposed. For this situation, the complexity estimate \(O(n^{4/3}[\log n]^{4/3})\) remains true. Finally, it is indicated how an algorithm of complexity \(O(n)\) times a certain power of \(\log n\) can be obtained.