LQP交替方向法是求解可分离结构型单调变分不等式问题的一种非常有效的方法.它不仅可以充分地利用目标函数的可分结构，将原问题分解为多个更易求解的子问题，还更适合求解大规模问题.对于带有三个可分离算子的单调变分不等式问题，结合增广拉格朗日算法和LQP交替方向法提出了一种部分并行分裂LQP交替方向法，构造了新算法的两个下降方向，结合这两个下降方向得到了一个新的下降方向，沿着这个新的下降方向给出了最优步长.并在较弱的假设条件下，证明了新算法的全局收敛性.
Logarithmic-quadratic proximal (LQP) alternating direction method of multipliers is a very effective method for solving monotone variational inequality with separable structure. It can make full use of the objective function of the separable structure, the original problem is decomposed into multiple sub-problems which is easier to be solved. Logarithmic-quadratic proximal (LQP) alternating direction method of multipliers are also suitable ones for solving large-scale problem. For monotone variational inequality problem with three separable operators, combining the augmented Lagrangian method with the LQP alternating direction method of multipliers, a partial parallel splitting LQP alternating direction method of multipliers is obtained. We construct two descent directions, the new direction is obtained by combined with the two descent directions, and an appropriate step size is derived along this new descent direction. And we prove the global convergence of the algorithm under a weaker assumption.
[1] Gabay D, Mercier B. A dual algorithm for the solution of nonlinear variational problems via finite-element approximations[J]. Computers Mathematics with Applications, 1976, 2:17-40.
[2] Gabay D. Chapter ix applications of the method of multipliers to variational inequalities[J]. Studies in Mathematics and its Applications, 1983, 15:299-331.
[3] Boyd S, Parikh N, Chu E, et al. Distributed optimization andstatistical learning via the alternating direction method of multipliers[J]. Foundations & Trends in Machine Learning, 2010, 3:1-122.
[4] He B S, Liao L Z. Improvements of some projection methods for monotone nonlinear variational inequalities[J]. Journal of Optimization Theory and Applications, 2002, 112:111-128.
[5] Xu M H, Wu T. A class of linearized proximal alternating direction methods[J]. Journal of Optimization Theory and Applications, 2011, 151:321-337.
[6] Fazel M, Pong T K, Sun D F, et al. Hankel matrix rank minimization with applications to system identification and realization[J]. SIAM Journal on Matrix Analysis & Applications, 2013, 34:946-977.
[7] Yuan X M, Li M. An LQP-based decomposition method for solving a class of variational inequalities[J]. SIAM Journal on Optimization, 2011, 21(4):1309-1318.
[8] Auslender A, Teboulle M, Ben-Tiba S. A logarithmic-quadratic proximal method for variational inqualities[J]. Computational Optimization and Application, 1999, 12(1-3):31-40.
[9] Chen C H, He B S, Ye Y Y, et al. The direct extension of ADMM for multi-block convex minimization problems is not necessarily convergent[DB-CD]. Mathematical Programming, 2016, 155(1-2):57-79.
[10] Hestenes M R. Multiplier and gradient methods[J]. Journal of Optimization Theory and Applications, 1969, 4:303-320.
[11] He B S. Parallel splitting augmented lagrangian methods for monotone structured variational inequalities[J]. Computational Optimization and Applications, 2009, 42:195-212.
[12] He B S, Tao M, Xu M H, et al. Alternating directions based contraction method for generally separable linearly constrained convex programming problems[J]. Optimization, 2013, 62:573-596.
[13] Zheng P, Wu D H. A partial parallel splitting augmented lagrangian method for solving constrained matrix optimization problems[J]. Computers and Mathematics with Applications, 2010, 60:1515-1524.
[14] Cao C X, Han D R, Xu L L. A new partial parallel splitting augmented lagrangian method for minimizing the sum of three convex functions[J]. Applied Mathematics and Computation, 2013, 219:5449-5457.
[15] 王宜举, 修乃华. 非线性最优化理论与方法[M]. 北京:科学出版社, 2012.
[16] Tao M, Yuan X M. On the O(1/t) convergence rate of alternating direction method with logarithmic-quadratic proximal regularization[J]. SIAM Journal on Optimization, 2012, 22(4):1431-1448.
