On the numerical solution of linear integral equation

Abstract

The numerical solution of linear integral equations of the types studied by Volterra has formed the subject of a recent memoir by E. T. Whittaker. Numerical methods are needed also for the solution of the linear integral equation studied by Fredholm and Hilbert and the object of this paper is to describe a method which may sometimes be useful. The linear integral equation of the second kind, f(s) = Ф(s) - λ ∫ 1 0 k(s, t) Ф(t)dt , may be solved for the unknown function Ф(t) teither directly or indirectly. In the direct methods of solution the function Ф(t) is expressed by means of infinite series of terms involving repeated integrals. In the so-called method of Neumann only one infinite series is used, while in the more complete method of Fredholm the function Ф(t) is expressed as the ratio of two infinite series which converge for all values of the parameter λ. The method of Neumann has been employed on many occasions to obtain numerical results and is generally more convenient to use than Fredholm’s method on account of the great complexity of the expressions occurring in Fredholm’s series. There are occasions, however, when the Neumann series fails to converge, or converges only slowly, and then some other method such as Fredholm’s must be used. It seems desirable if possible to devise simple approximate methods which possess some of the advantages of Fredholm’s method, because in many cases the evaluation of repeated integrals becomes very tedious and the use of the direct methods of solution becomes impracticable except for a rough approxi­mation.