|
Teoriya Veroyatnostei i ee Primeneniya, 1974, Volume 19, Issue 4, Pages 817–824
(Mi tvp3981)
|
|
|
|
This article is cited in 3 scientific papers (total in 3 papers)
Short Communications
On convergence of a random search method in convex minimization problems
V. G. Karmanov M. V. Lomonosov Moscow State University
Abstract:
In the present paper, the minimization problem is considered for a convex function $\varphi(x)$ on a convex and closed set $X$ of the $n$-dimensional Euclidean space $E_n$, and a method is proposed for constructing a recurrent sequence $x^0,x^1,\dots,\in X$ by the formula $x^{k+1}=x^k+\beta_ks^k$, where $s^k$ is a random vector, and $\beta_k$ is determined so as to minimize $\varphi(x)$ on the straight line $x^k+\beta s^k$ $(|\beta|<\infty)$.
Under sufficiently general assumptions, it is proved that
$$
\mathbf P\{\varphi(x^m)\to\min\varphi(x)\quad(x\in X,\quad m\to\infty)\}=1.
$$
In case $X=E_n$, it is proved that
$$
\lim_{m\to\infty}\mathbf P\biggl\{\varphi(x^m)-\min\varphi(x)\le\frac cm\biggr\}=1,
$$
where $c=\mathrm{const}>0$.
Received: 20.12.1973
Citation:
V. G. Karmanov, “On convergence of a random search method in convex minimization problems”, Teor. Veroyatnost. i Primenen., 19:4 (1974), 817–824; Theory Probab. Appl., 19:4 (1975), 788–794
Linking options:
https://www.mathnet.ru/eng/tvp3981 https://www.mathnet.ru/eng/tvp/v19/i4/p817
|
|