Robust analysis and global minimization of a class of discontinuous functions. II (Q1178649)

[For part I see the author, ibid. 6, No. 3, 205-223 (1990; Zbl 0718.49025).] The paper deals with the problem of global minimization of a robust function \(f(x)\) over a compact robust set \(S\). A subset \(S\) of a topological space \(X\) is said to be robust iff \(\text{cl} S=\text{cl int} S\), a function \(f: X\to R^ 1\) is robust iff \(H_ c=\{x\in X\mid\;f(x)<c\}\) is a robust set for all \(c\). A number of necessary and sufficient conditions for \(\bar x\in S\) to be a global minimzer for \(f(x)\) over \(S\) are derived. One of them is: \[ (\mu(S\cap H_{\bar c}))^{-1}\int_{S\cap H_{\bar c}}f(x)d\mu=\bar c, \qquad \bar c=f(\bar x), \] where \(\mu\) is a measure on a \(\sigma\)-field of subsets of \(X\). On this basis two abstract algorithms for solving the problem are proposed. One of the algorithms starts from an arbitrary point \(x_ 0\in S\), generates estimates \(c_ k\) for \(f_ *=\min\{f(x)\mid\;x\in S\}\) and approximations \(S\cap H_{c_ k}\) for \(X_ *=\text{Arg}\min\{f(x)\mid\;x\in S\}\) such that \[ c_{k+1}=(\mu(S\cap H_{c_ k}))^{-1}\int_{S\cap H_{c_ k}}f(x)d\mu, \qquad c_ 0=f(x_ 0). \] Under appropriate assumptions it is stated that \(\lim_{k\to\infty}c_ k=f_ *\) and \(X_ *=\bigcap_{k=1}^ \infty(S\cap H_{c_ k})\). An implementable version of the algorithm using Monte Carlo techniques for numerical integration is described. Results of numerical tests with multiextremal discontinuous functions \(f(x)\), \(x\in R^ n\), \(n=5\div 50\) are also given.