On hypersingular boundary integral equations for certain problems in mechanics (Q1263453)

Improper or weakly singular integrals, once thought to be a handicap in computations, are now accepted as a source of effectiveness and stability in the numerical solution of many problems in mechanics. Such integrals naturally arise, for example, in the Boundary Integral Equation (BIE) method, and the literature contains many examples of successful computations based on Boundary Element-type (BEM) solutions of the BIEs. For some vector problems, the BIE/BEM process gives rise to a stronger type of singular integral which exists in the sense of the Cauchy Principal Value (CPV). This integral also has been treated numerically with success. Less commonly, but with growing frequency it seems, the gradient or normal derivative of such boundary integrals is taken, especially in the formulation of mechanics problems involving cracks. Then integrals more (hyper) singular than the CPV can explicitly arise. Usually, however, rather than confront such hypersingular integrals directly, a process of regularization [e.g.: \textit{A. J. Burton} and \textit{G. F. Miller}, Proc. R. Soc. London, Ser. A 323, 201-210 (1971; Zbl 0235.65080); \textit{H. D. Bui}, J. Mech. Phys. Solids 25, 29-39 (1977; Zbl 0355.73074)] is employed to lower the singularity of the integrands. Such regularization usually carries a formulational complexity and computational cost if it is even possible. However, the alternatives to regularization seem to be divergent integrals or numerical computation with integrals more singular than the CPV with, perhaps, even questionable definition. The purpose of this brief paper is to examine a reasonable alternative to the mentioned regularization. Indeed, we consider the essential analytical issues surrounding the occurrence of hypersingular boundary integrals in the context of a crack problem. Then we suggest certain computational strategies. The relevant integrals are contrasted with truly divergent integrals and conditions under which they exist in the finite-part sense of Hadamard are stated. The relation of such integrals to both one-sided and two-sided CPV integrals is examined as well. The critical role of smoothness of relevant functions in the integrands for existence of these integrals is emphasized. Our goal is to encourage understanding and use of finite-part integrals in mechanics. We wish to discourage the practice of going to considerable lengths to avoid them as seems presently the case. How to compute them, analytically or numerically, is a rich area of research. Indeed, we recommend the some works which are probably not sufficiently familiar to the mechanics community. While we confine attention here to an easy-to-describe (scalar) sound-scattering crackline problem in two dimensions, extension of the present ideas to (vector) problems for static loading or wave scattering by crack surfaces in three dimensions is relatively clear.