Linearly Convergent First-Order Algorithms for Semidefinite Programming

Linearly Convergent First-Order Algorithms for Semidefinite Programming

Year:    2017

Author:    Cong D. Dang, Guanghui Lan, Zaiwen Wen

Journal of Computational Mathematics, Vol. 35 (2017), Iss. 4 : pp. 452–468

Abstract

In this paper, we consider two different formulations (one is smooth and the other one is nonsmooth) for solving linear matrix inequalities (LMIs), an important class of semidefinite programming (SDP), under a certain Slater constraint qualification assumption. We then propose two first-order methods, one based on subgradient method and the other based on Nesterov's optimal method, and show that they converge linearly for solving these formulations. Moreover, we introduce an accelerated prox-level method which converges linearly uniformly for both smooth and non-smooth problems without requiring the input of any problem parameters. Finally, we consider a special case of LMIs, i.e., linear system of inequalities, and show that a linearly convergent algorithm can be obtained under a much weaker assumption.

You do not have full access to this article.

Already a Subscriber? Sign in as an individual or via your institution

Journal Article Details

Publisher Name:    Global Science Press

Language:    English

DOI:    https://doi.org/10.4208/jcm.1612-m2016-0703

Journal of Computational Mathematics, Vol. 35 (2017), Iss. 4 : pp. 452–468

Published online:    2017-01

AMS Subject Headings:   

Copyright:    COPYRIGHT: © Global Science Press

Pages:    17

Keywords:    Semi-definite programming Linear matrix inequalities Error bounds Linear convergence.

Author Details

Cong D. Dang

Guanghui Lan

Zaiwen Wen