Fejér processes with diminishing disturbances (Q1048512)

From MaRDI portal
scientific article
Language Label Description Also known as
English
Fejér processes with diminishing disturbances
scientific article

    Statements

    Fejér processes with diminishing disturbances (English)
    0 references
    0 references
    12 January 2010
    0 references
    Consider a finite family \(\mathcal{F}\) of Fejér operators with respect to a compact subset \(V\) of a finite-dimensional Euclidean space \(E\), which are continuous on an open extension of \(V\). Given a point \(x^0 \in E\) and a sequence of points \(\{z^s\}\) converging to zero in \(E\), the author investigates Fejér processes with disturbances of type \[ x^{s+1}=F_s(x^s+z^s), \quad s \in\{0,1,2,\dots\}, \] where, for each \(s\), the operator \(F_s \in \mathcal{F}\) is assumed to be locally strongly Fejér at \(x^s\), i.e., there exists a neighborhood \(U\) of zero and an attraction coefficient \(\alpha <1\) such that \(\|F_s(y)-v\| \leq \alpha \|y-v\|\) for all \(v \in V\) and \(y \in x^s+U\). In particular, it is shown that, whenever the sequence \(\{x^s\}\) is bounded, its limit points belong to \(V\). Moreover, some special classes of disturbances \(\{z^s\}\) are presented for which one can ensure that all limit points of \(\{x^s\}\) belong to a certain subset \(Z\) of \(V\). These results are then applied to convex optimization problems of the type \(\min \{f(x) \mid x \in V\}\), where the feasible set can be expressed as an intersection of convex constraint sets, \(V=\bigcap_{i=1,N}V_i\). In this case, the operators \(F_s\) involved in the Fejér process actually are projectors onto certain constraint sets \(V_{i_s}\), while disturbances are defined as \(z_s = -\lambda_s g^s\), where \(g_s \in \partial f(x^s)\), \(\lambda_s \rightarrow +0\) and \(\sum_{s=0}^\infty \lambda_s = \infty\).
    0 references
    0 references
    0 references
    0 references
    0 references
    Fejér operator
    0 references
    localized attractant
    0 references
    convex optimization
    0 references
    gradient projection
    0 references