scientific article; zbMATH DE number 2042705

From MaRDI portal
Publication:4450552





Let \(T\) be a maximal monotone operator on a real Hilbert space. The inertial proximal method for finding a zero of \(T\) generates a sequence \(\{x_k\}\) as follows: given \(x_{k-1}\) and \(x_k \in H\) and two parameters \(\alpha_k \in [0,1)\) and \(\lambda_k >0\), find \(x_{k+1} \in H\) such that \[ \lambda_k T(x_{k+1}) +x_{k+1}-x_k - \alpha_k(x_k-x_{k-1})\ni 0. \] In order to obtain an implementable algorithm, the authors replace in the previous relation, \(T(x_{k+1})\) by an enlargement \(T^{\varepsilon}(x_{k+1})\). Then for the sequence generated by the resulting algorithm, they prove the weak convergence and the strong convergence to a zero of \(T\). In the second part, the particular case of convex minimization is examined. This time, \(T^{\varepsilon}\) is replaced by \(\partial_{\varepsilon}f\), the \(\varepsilon\)-subdifferential of \(f\), and, when \(f\) is nonsmooth, the function \(f\) itself is approximated by a monotone decreasing sequence \(\{f_k\}\) of proper closed convex functions such that \(f=\) cl\((\inf_k\,f_k)\). A weak convergence theorem is proven for the resulting algorithm.




Cited in
(52)








This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4450552)