Distributed Semidefinite Programming With Application to Large-Scale System Analysis

摘要: Distributed algorithms for solving coupled semidefinite programs commonly require many iterations to converge. They also put high computational demand on the agents. In this paper, we show that in case problem has an inherent tree structure, it is possible devise efficient distributed algorithm such problems. The proposed relies predictor–corrector primal-dual interior-point methods, where use a message-passing compute search directions distributedly. Message passing here closely related dynamic programming over trees. This allows us exact finite number of steps. because computing requires recursion structure and, hence, terminates after upward and downward pass through tree. Furthermore, can be computed priori only depends coupling problem. We analyzing robustness large-scale uncertain systems test performance using numerical examples.