A Schur complement based semi-proximal ADMM for convex quadratic conic programming and extensions (Q5962721)

The paper is concerned with the development and convergence proof of a Schur complement based first order semi-proximal alternating direction method of multipliers (ADMM) for the solution of convex optimization problems which have an objective function that is formed by the sum of two proper closed convex functions and an arbitrary number of convex quadratic or linear functions and which include a coupling linear equality constraint. The proposed method is particularly suitable for the solution of convex quadratic semidefinite programming (QSDP) problems with constraints consisting of linear equalities, a positive semidefinite cone, and a simple convex polyhedral set, and it can especially be used to generate good starting points for fast locally convergent methods like semismooth Newton-CG methods for the solution of linear SDP problems. The numerical performance of the new algorithm is demonstrated by its application to various problems including QSDP problems and shows that this algorithm is superior to other recent ADMM type methods.