Iteration-complexity of first-order penalty methods for convex programming (Q1949272): Difference between revisions

The problem under study consists in minimizing a \(C^{1,1}\) convex function \(f\) defined on a (sufficiently simple) compact convex set \(X\subseteq \mathbb{R}^{n}\), subject to the constraint \(\mathcal{A}(x)\in \mathcal{K}^{\ast }\); the mapping \(\mathcal{A}:\mathbb{R}^{n}\rightarrow \mathbb{R}^{m}\) is supposed to be affine, and \(\mathcal{K}^{\ast }\) denotes the dual cone of \(\mathcal{K}\), a closed convex cone in \(\mathbb{R}^{m}.\) For this problem, the authors consider a quadratic penalty approach, in which the penalty subproblems are solved by a Nesterov type method [\textit{Yu. E. Nesterov}, Sov. Math., Dokl. 27, 372--376 (1983); translation from Dokl. Akad. Nauk SSSR 269, 543--547 (1983; Zbl 0535.90071); Math. Program. 103, No. 1 (A), 127--152 (2005; Zbl 1079.90102)]; they present iteration-complexity bounds for two variants of the proposed method.