Online Distributed Constrained Optimization Over General Unbalanced Digraphs

This paper studies the distributed online constrained optimization problem over a multiagent network, where each agent exchanges local information with neighbors on an unbalanced digraph with row-stochastic matrix. For solving such problem, an distributed online projection gradient algorithm is presented and the dynamic regret is adopted to measure its performance. With proper assumptions, we establish a dynamic-regret bound for each agent, which grows sublinearly as long as the increasing rate of the minimizer sequence' deviation lies in a certain range. Finally, a simulation example is provided to show the effectiveness of the proposed approach.