A proximal point algorithm for finding a common zero of a finite family of maximal monotone operators in the presence of computational errors (Q448514)

The author applies the proximal point algorithm in order to obtain a good approximation of a point which is a common zero of a finite family of maximal monotone operators and a common fixed point of a finite family of nonexpansive operators. For the finite set \(\mathcal{L}_1\) of maximal monotone operators \(T: X \rightarrow 2^X\) and for a finite set \(\mathcal{L}_2\) of mappings \(T: X \rightarrow X\) (here it is supposed that the set \(\mathcal{L}_1 \cup \mathcal{L}_2\) is nonempty and that one of the sets \(\mathcal{L}_1\) or \(\mathcal{L}_2\) may be empty) the author supposes that \(F(T) \neq 0\) for any \( T \in \mathcal{L}_1 (F(T)= \{z\in X: 0 \in T(z)\})\). The set of all mappings \(S: \{0,1,2,\dots\} \rightarrow \mathcal{L}_2 \cup \{P_{C,T}: T \in \mathcal{L}_1 . c \in [\overline{\lambda}, \infty)\} \) with the properties: for each integer \(p \geq 0\) and each \(T \in \mathcal{L}_2\) there is \(i \in \{p \dots p+l-1\}\) for which \(S(i)=T\), and for each integer \(p \geq 0 \) and each \(T \in \mathcal{L}_1\) there exists \(i \in \{p \dots p+l-1\}\) and \(c \geq \overline{\lambda}\) such that \(S(i)=P_{C.T}\), is introduced. Main result: The author studies the behaviour of the sequences generated by \(S \in \mathcal{R}\) taking into account computational errors which are always present in practice. Namely, in practice the algorithm associated with \(S \in \mathcal{R}\) generates a sequence \(\{x_K\}^\infty_{K=0}\) such that for each integer \(K \geq 0\) \( \|x_{K+1}- [s(K)](x_K)\| \leq \delta\) with a constant \(\delta > 0\) which depends only on the computer system. The goal of this paper is to understand what subset of \(x\) attracts all sequences \(\{x_K\}^\infty_{K=0}\) generated by algorithms associated with \(S \in \mathcal{R}\). Further, the main goal is also, for a given \(\varepsilon >0\), to find a point \( x \in \widetilde{F}_\varepsilon\) considered as an \(\varepsilon\)-approximate solution associated with the family of operators \(\mathcal{L}_1 \cup \mathcal{L}_2\). The author shows that an \(\varepsilon\)-approximate solution can be obtained after \(l(n_0-1)\) iterations of the algorithm associated with \(S \in \mathcal{R}\) and under the presence of computational errors bounded from above by a constant \(\delta\), where \(\delta\) and \(n_0\) are constants depending on \(\varepsilon\).