摘要
指出直接推广的经典乘子交替方向法对三个算子的问题不能保证收敛的原因,并且给出将其改造成收敛算法的相应策略.同时,在一个统一框架下,证明了修正的乘子交替方向法的收敛性和遍历意义下具有0(1/t)收敛速率.
In this paper, we indicate the reason of divergence, and illustrate the strategies which modify the alternating direction method of multipliers (ADMM) to a convergent one for the linearly constrained separable convex optimization with three individual functions. Finally, using a uniform framework, we give the simple proofs for the convergence and O(1/t) convergence rate in the ergodic sense of the ADMM-like methods.
出处
《运筹学学报》
CSCD
北大核心
2015年第3期57-70,共14页
Operations Research Transactions
基金
国家自然科学基金(No.11471156)
关键词
凸优化
分裂收缩算法
变分不等式
统一框架
收敛速率
convex optimization, splitting contraction methods, variational inequal-ity, uniform framework, convergence rate