Title: Path-Following Gradient-Based Decomposition Algorithms For Separable Convex Optimization

Abstract: A new distributed optimization algorithm, called path-following gradient-based decomposition, is proposed to solve separable convex optimization problems. This algorithm is a combination of three techniques: smoothing, Lagrangian decomposition and path-following gradient method. By applying a smoothing technique via barrier functions, we show that the dual problem possesses smoothness properties which make it suitable for applying a first order method of multiplier and lead to a decomposition scheme. We prove the global convergence of the algorithm and analyze its local convergence rate. Then, we modify the proposed algorithm by applying Nesterov's accelerating scheme to get a new variant which has a better local convergence rate. Finally, we present preliminary numerical tests that confirm the theory development.