Condition of boundary integral equations in which the sought-for function and the given right-hand side are defined on different domains; round-off errors in the numerical solutions (Q1122394)

Boundary integral equations in which the sought-for function and the given right-hand side are defined on different curves have the great advantage of regularity. Discretization of them, however, yields ill- conditioned systems of algebraic equations, i.e., systems susceptible to round-off errors. In this paper the round-off error is determined as a function of the number of nodes and as a function of the distance between the domains of definition of the solution and of the boundary values. The results corroborate and specify a theoretically deduced rule [the author, Comput. Struct. 8, 199-205 (1978; Zbl 0373.73024)] on how to carry out the numerical evaluation without spoiling the solution by round-off errors. In Sect. 3 the two integral equations of Oliveira [\textit{E. R. Oliveira}, Plane stress analysis by a general integral method. J. Eng. Mech. Div. ASCE 94 EM1, 79-101 (1968)] are introduced in detail and their limiting cases for \(c\to 0\) are also compiled. Section 4 comments on the results of other papers concerning our subject. At the beginning the eigenvalues of the integral operators of Oliveira's equations for the half plane and those of their limiting cases \(c\to 0\) are compiled. Starting from the spectra of the integral operators it is shown which condition the integral equations should have theoretically. It turns out to be desirable to determine the round-off error of numerical solutions of \(c\neq 0\) integral equations. In Sect. 5 is outlined the method for numerical treatment of the integral equation. Sect. 6 shows how to evaluate the round-off error. Norms are introduced in order to be able to represent the large quantity of data clearly. Sect. 7 presents characteristical numerical results. Plots of the round-off error depending on the distance c between the boundary S and the auxiliary curve \(\hat S\) and depending on the number J of nodes are given for three examples. In the first case S and \(\hat S\) are concentric circles, in the second example S and \(\hat S\) are equidistant oval-shaped curves and in the third case S is an oval and \(\hat S\) a circle. In Sect. 8 the results are interpreted and conclusions are drawn as to how to reasonably apply the \(c\neq 0\) method. It turns out, that the distance c should be adapted to the fineness of the discretization of the problem. For increasing numbers of nodes, c should be diminished. A formula is deduced which describes the circumstances only roughly but which has the advantage of being easily applicable.