LOCAL LINEAR CONVERGENCE OF THE ALTERNATING DIRECTION METHOD OF MULTIPLIERS ON QUADRATIC OR LINEAR PROGRAMS

被引:105
作者
Boley, Daniel [1 ]
机构
[1] Univ Minnesota, Dept Comp Sci, Minneapolis, MN 55455 USA
基金
美国国家科学基金会;
关键词
ADMM; linear programming; quadratic programming; SPLITTING ALGORITHMS; SHRINKAGE; PENALTY; INVERSE;
D O I
10.1137/120878951
中图分类号
O29 [应用数学];
学科分类号
070104 ;
摘要
We introduce a novel matrix recurrence yielding a new spectral analysis of the local transient convergence behavior of the alternating direction method of multipliers (ADMM), for the particular case of a quadratic program or a linear program. We identify a particular combination of vector iterates whose convergence can be analyzed via a spectral analysis. The theory predicts that ADMM should go through up to four convergence regimes, such as constant step convergence or linear convergence, ending with the latter when close enough to the optimal solution if the optimal solution is unique and satisfies strict complementarity.
引用
收藏
页码:2183 / 2207
页数:25
相关论文
共 58 条
[1]  
[Anonymous], 2013, MATRIX COMPUTATIONS
[2]  
[Anonymous], 2011, CVX MATLAB SOFTWARE
[3]  
[Anonymous], 2012, Physical Communication, DOI DOI 10.1016/J.PHYCOM.2011.07.005
[4]  
[Anonymous], 1983, STUDIES MATH ITS APP
[5]  
[Anonymous], TR1214 CAAM RIC U
[6]  
[Anonymous], 2009, APPL LAGRANGIAN BASE
[7]  
[Anonymous], 2004, INTRO LECT CONVEX OP
[8]  
[Anonymous], 1959, The Theory of Matrices
[9]  
[Anonymous], 2007, GRADIENT METHODS MIN
[10]  
[Anonymous], 1998, Solution of large-scale eigenvalue problems with implicitly restarted Arnoldi methods, DOI DOI 10.1137/1.9780898719628