Adaptive control of stochastic calculating processes (Q1973274)

The authors propose adaptive control methods for Monte-Carlo procedures of multidimensional integration and numerical solution of Fredholm-type integral equations. Let, for example, the \(n\)-dimensional integral \( J=\int_D f(x) dx=E\{f(\xi)/p(\xi)\} \) over a bounded closed set \(D\subset\mathbb{R}^n\) be evaluated. Here \(\xi\) is a random variable with values in \(D\) distributed with a density \(p=p(x)>0\), \(x\in D\). Suppose that the estimate \(\widehat J_N\) of \(J\) is represented in the form \(\widehat J_N=\widehat\theta^T_N\int_D\psi(x)p(x) dx\), where the integral \(\int_D\psi(x)p(x) dx\) is considered to be known. The authors use a recurrent system of equations satisfied by \(\widehat\theta_N\) as a stochastic object of control with the control action represented by a function of unit distribution density of a random grid of integration. The criterion of evaluation accuracy is chosen to be the criterion of optimal functioning.