1: \begin{abstract}

2: We apply the Min-Sum message-passing protocol to solve the consensus problem in distributed optimization. We show that while the ordinary Min-Sum algorithm does not converge, a modified version of it known as Splitting yields convergence to the problem solution. We prove that a proper choice of the tuning parameters allows Min-Sum Splitting to yield subdiffusive accelerated convergence rates, matching the rates obtained by shift-register methods. The acceleration scheme embodied by Min-Sum Splitting for the consensus problem bears similarities with lifted Markov chains techniques and with multi-step first order methods in convex optimization.

3: %yielding convergence times that match the diameter lower bound in some graph topologies of interest.

4: %with the advantage that no construction of a lifted space is required as Min-Sum is supported on the topology of the directed edges of the original communication graph.

5: %which is based on a reparametrization of the original problem, 

6: %matching the rates obtained by shift-registers algorithms. 

7: \end{abstract}

8: