Stochastic approximation - A powerful method for solving deterministic numerical problems (Q1116288)

In general, this article is devoted to a consideration of the stochastic approximation method only in numerical frameworks. The stochastic approximation method of \textit{H. Robbins} and \textit{S. Monro} [Ann. Math. Statist. 22, 400-407 (1951; Zbl 0054.059)] after some modifications due to \textit{H. Kesten} [ibid. 29, 41-59 (1958; Zbl 0087.134)] is considered. Both algebraic equations and two-point boundary value problems in ordinary differential equations are solved on numerical examples. The second one is to find \(y'(0)\) for the problem \(y''=g(y',y,x)\) and \(y(0)=c\), \(y(1)=b\). Basing on numerical investigations it is shown that the results are surprisingly accurate. A deterministic problem with multiple roots can be solved whereas the original Robbins-Monro method is restricted to a monotone increasing regression function with a single root. It is shown that all solutions are obtained by different initial points. But there are no comments about the choice of these initial iterations.